diff options
180 files changed, 6349 insertions, 3907 deletions
diff --git a/g3doc/architecture_guide/performance.md b/g3doc/architecture_guide/performance.md index 39dbb0045..b981f0c01 100644 --- a/g3doc/architecture_guide/performance.md +++ b/g3doc/architecture_guide/performance.md @@ -30,7 +30,7 @@ is distinct from **structural costs**. Improvements here are ongoing and driven by the workloads that matter to gVisor users and contributors. This page provides a guide for understanding baseline performance, and calls out -distint **structural costs** and **implementation costs**, highlighting where +distinct **structural costs** and **implementation costs**, highlighting where improvements are possible and not possible. While we include a variety of workloads here, it’s worth emphasizing that gVisor @@ -211,7 +211,7 @@ url="/performance/applications.csv" title="perf.py http.(node|ruby) The above figure shows the result of simple `node` and `ruby` web services that render a template upon receiving a request. Because these synthetic benchmarks -do minimal work per request, must like the `redis` case, they suffer from high +do minimal work per request, much like the `redis` case, they suffer from high overheads. In practice, the more work an application does the smaller the impact of **structural costs** become. diff --git a/g3doc/architecture_guide/resources.md b/g3doc/architecture_guide/resources.md index 1dec37bd1..fc997d40c 100644 --- a/g3doc/architecture_guide/resources.md +++ b/g3doc/architecture_guide/resources.md @@ -19,12 +19,12 @@ sandboxed process: Much like a Virtual Machine (VM), a gVisor sandbox appears as an opaque process on the system. Processes within the sandbox do not manifest as processes on the -host system, and process-level interactions within the sandbox requires entering +host system, and process-level interactions within the sandbox require entering the sandbox (e.g. via a [Docker exec][exec]). ## Networking -The sandbox attaches a network endpoint to the system, but runs it's own network +The sandbox attaches a network endpoint to the system, but runs its own network stack. All network resources, other than packets in flight on the host, exist only inside the sandbox, bound by relevant resource limits. diff --git a/images/benchmarks/nginx/Dockerfile b/images/benchmarks/nginx/Dockerfile index 2444d04b1..c8e3330d0 100644 --- a/images/benchmarks/nginx/Dockerfile +++ b/images/benchmarks/nginx/Dockerfile @@ -9,3 +9,4 @@ RUN mkdir -p /local && \ RUN touch /local/index.html COPY ./nginx.conf /etc/nginx/nginx.conf +COPY ./nginx_gofer.conf /etc/nginx/nginx_gofer.conf diff --git a/images/benchmarks/nginx/nginx_gofer.conf b/images/benchmarks/nginx/nginx_gofer.conf new file mode 100644 index 000000000..dbba2a575 --- /dev/null +++ b/images/benchmarks/nginx/nginx_gofer.conf @@ -0,0 +1,19 @@ +user nginx; +worker_processes 1; +daemon off; + +error_log /var/log/nginx/error.log warn; +pid /var/run/nginx.pid; + +events { + worker_connections 1024; +} + + +http { + server { + location / { + root /local; + } + } +} diff --git a/pkg/abi/linux/ioctl.go b/pkg/abi/linux/ioctl.go index 3356a2b4a..dc9ac7e7c 100644 --- a/pkg/abi/linux/ioctl.go +++ b/pkg/abi/linux/ioctl.go @@ -115,7 +115,7 @@ const ( // Constants from uapi/linux/fs.h. const ( - FS_IOC_GETFLAGS = 2147771905 + FS_IOC_GETFLAGS = 2148034049 FS_VERITY_FL = 1048576 ) diff --git a/pkg/coverage/coverage.go b/pkg/coverage/coverage.go index 6831adcce..a4f4b2c5e 100644 --- a/pkg/coverage/coverage.go +++ b/pkg/coverage/coverage.go @@ -100,12 +100,9 @@ var coveragePool = sync.Pool{ // instrumentation_filter. // // Note that we "consume", i.e. clear, coverdata when this function is run, to -// ensure that each event is only reported once. -// -// TODO(b/160639712): evaluate whether it is ok to reset the global coverage -// data every time this function is run. We could technically have each thread -// store a local snapshot against which we compare the most recent coverdata so -// that separate threads do not affect each other's view of the data. +// ensure that each event is only reported once. Due to the limitations of Go +// coverage tools, we reset the global coverage data every time this function is +// run. func ConsumeCoverageData(w io.Writer) int { once.Do(initCoverageData) @@ -117,23 +114,23 @@ func ConsumeCoverageData(w io.Writer) int { for fileIndex, file := range globalData.files { counters := coverdata.Cover.Counters[file] for index := 0; index < len(counters); index++ { - val := atomic.SwapUint32(&counters[index], 0) - if val != 0 { - // Calculate the synthetic PC. - pc := globalData.syntheticPCs[fileIndex][index] - - usermem.ByteOrder.PutUint64(pcBuffer[:], pc) - n, err := w.Write(pcBuffer[:]) - if err != nil { - if err == io.EOF { - // Simply stop writing if we encounter EOF; it's ok if we attempted to - // write more than we can hold. - return total + n - } - panic(fmt.Sprintf("Internal error writing PCs to kcov area: %v", err)) + if atomic.LoadUint32(&counters[index]) == 0 { + continue + } + // Non-zero coverage data found; consume it and report as a PC. + atomic.StoreUint32(&counters[index], 0) + pc := globalData.syntheticPCs[fileIndex][index] + usermem.ByteOrder.PutUint64(pcBuffer[:], pc) + n, err := w.Write(pcBuffer[:]) + if err != nil { + if err == io.EOF { + // Simply stop writing if we encounter EOF; it's ok if we attempted to + // write more than we can hold. + return total + n } - total += n + panic(fmt.Sprintf("Internal error writing PCs to kcov area: %v", err)) } + total += n } } diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go index 511179e31..fece3e762 100644 --- a/pkg/sentry/devices/memdev/full.go +++ b/pkg/sentry/devices/memdev/full.go @@ -24,6 +24,8 @@ import ( const fullDevMinor = 7 // fullDevice implements vfs.Device for /dev/full. +// +// +stateify savable type fullDevice struct{} // Open implements vfs.Device.Open. @@ -38,6 +40,8 @@ func (fullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // fullFD implements vfs.FileDescriptionImpl for /dev/full. +// +// +stateify savable type fullFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go index 4918dbeeb..ff5837747 100644 --- a/pkg/sentry/devices/memdev/null.go +++ b/pkg/sentry/devices/memdev/null.go @@ -25,6 +25,8 @@ import ( const nullDevMinor = 3 // nullDevice implements vfs.Device for /dev/null. +// +// +stateify savable type nullDevice struct{} // Open implements vfs.Device.Open. @@ -39,6 +41,8 @@ func (nullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // nullFD implements vfs.FileDescriptionImpl for /dev/null. +// +// +stateify savable type nullFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go index 5e7fe0280..ac943e3ba 100644 --- a/pkg/sentry/devices/memdev/random.go +++ b/pkg/sentry/devices/memdev/random.go @@ -30,6 +30,8 @@ const ( ) // randomDevice implements vfs.Device for /dev/random and /dev/urandom. +// +// +stateify savable type randomDevice struct{} // Open implements vfs.Device.Open. @@ -44,6 +46,8 @@ func (randomDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, } // randomFD implements vfs.FileDescriptionImpl for /dev/random. +// +// +stateify savable type randomFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go index 60cfea888..1929e41cd 100644 --- a/pkg/sentry/devices/memdev/zero.go +++ b/pkg/sentry/devices/memdev/zero.go @@ -27,6 +27,8 @@ import ( const zeroDevMinor = 5 // zeroDevice implements vfs.Device for /dev/zero. +// +// +stateify savable type zeroDevice struct{} // Open implements vfs.Device.Open. @@ -41,6 +43,8 @@ func (zeroDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // zeroFD implements vfs.FileDescriptionImpl for /dev/zero. +// +// +stateify savable type zeroFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/ttydev/ttydev.go b/pkg/sentry/devices/ttydev/ttydev.go index 664e54498..a287c65ca 100644 --- a/pkg/sentry/devices/ttydev/ttydev.go +++ b/pkg/sentry/devices/ttydev/ttydev.go @@ -30,6 +30,8 @@ const ( ) // ttyDevice implements vfs.Device for /dev/tty. +// +// +stateify savable type ttyDevice struct{} // Open implements vfs.Device.Open. diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 77d1f493d..903135fae 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -35,6 +35,8 @@ import ( const Name = "devpts" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // Name implements vfs.FilesystemType.Name. @@ -58,6 +60,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.Filesystem.VFSFilesystem(), root.VFSDentry(), nil } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -110,6 +113,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // rootInode is the root directory inode for the devpts mounts. +// +// +stateify savable type rootInode struct { implStatFS kernfs.AlwaysValid @@ -131,7 +136,7 @@ type rootInode struct { root *rootInode // mu protects the fields below. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // replicas maps pty ids to replica inodes. replicas map[uint32]*replicaInode @@ -242,6 +247,7 @@ func (i *rootInode) DecRef(context.Context) { i.rootInodeRefs.DecRef(i.Destroy) } +// +stateify savable type implStatFS struct{} // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 98d4ffb22..69c2fe951 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -31,6 +31,8 @@ import ( ) // masterInode is the inode for the master end of the Terminal. +// +// +stateify savable type masterInode struct { implStatFS kernfs.InodeAttrs @@ -56,14 +58,12 @@ func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernf return nil, err } - mi.IncRef() fd := &masterFileDescription{ inode: mi, t: t, } fd.LockFD.Init(&mi.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { - mi.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -89,6 +89,7 @@ func (mi *masterInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds return mi.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) } +// +stateify savable type masterFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -103,7 +104,6 @@ var _ vfs.FileDescriptionImpl = (*masterFileDescription)(nil) // Release implements vfs.FileDescriptionImpl.Release. func (mfd *masterFileDescription) Release(ctx context.Context) { mfd.inode.root.masterClose(mfd.t) - mfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/devpts/replica.go b/pkg/sentry/fsimpl/devpts/replica.go index 816bac80c..6515c5536 100644 --- a/pkg/sentry/fsimpl/devpts/replica.go +++ b/pkg/sentry/fsimpl/devpts/replica.go @@ -30,6 +30,8 @@ import ( ) // replicaInode is the inode for the replica end of the Terminal. +// +// +stateify savable type replicaInode struct { implStatFS kernfs.InodeAttrs @@ -52,14 +54,12 @@ type replicaInode struct { var _ kernfs.Inode = (*replicaInode)(nil) // Open implements kernfs.Inode.Open. -func (si *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - si.IncRef() +func (ri *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &replicaFileDescription{ - inode: si, + inode: ri, } - fd.LockFD.Init(&si.locks) + fd.LockFD.Init(&ri.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { - si.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -67,34 +67,35 @@ func (si *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kern } // Valid implements kernfs.Inode.Valid. -func (si *replicaInode) Valid(context.Context) bool { +func (ri *replicaInode) Valid(context.Context) bool { // Return valid if the replica still exists. - si.root.mu.Lock() - defer si.root.mu.Unlock() - _, ok := si.root.replicas[si.t.n] + ri.root.mu.Lock() + defer ri.root.mu.Unlock() + _, ok := ri.root.replicas[ri.t.n] return ok } // Stat implements kernfs.Inode.Stat. -func (si *replicaInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - statx, err := si.InodeAttrs.Stat(ctx, vfsfs, opts) +func (ri *replicaInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := ri.InodeAttrs.Stat(ctx, vfsfs, opts) if err != nil { return linux.Statx{}, err } statx.Blksize = 1024 statx.RdevMajor = linux.UNIX98_PTY_REPLICA_MAJOR - statx.RdevMinor = si.t.n + statx.RdevMinor = ri.t.n return statx, nil } // SetStat implements kernfs.Inode.SetStat -func (si *replicaInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { +func (ri *replicaInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { if opts.Stat.Mask&linux.STATX_SIZE != 0 { return syserror.EINVAL } - return si.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) + return ri.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) } +// +stateify savable type replicaFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -106,37 +107,35 @@ type replicaFileDescription struct { var _ vfs.FileDescriptionImpl = (*replicaFileDescription)(nil) // Release implements fs.FileOperations.Release. -func (sfd *replicaFileDescription) Release(ctx context.Context) { - sfd.inode.DecRef(ctx) -} +func (rfd *replicaFileDescription) Release(ctx context.Context) {} // EventRegister implements waiter.Waitable.EventRegister. -func (sfd *replicaFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - sfd.inode.t.ld.replicaWaiter.EventRegister(e, mask) +func (rfd *replicaFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + rfd.inode.t.ld.replicaWaiter.EventRegister(e, mask) } // EventUnregister implements waiter.Waitable.EventUnregister. -func (sfd *replicaFileDescription) EventUnregister(e *waiter.Entry) { - sfd.inode.t.ld.replicaWaiter.EventUnregister(e) +func (rfd *replicaFileDescription) EventUnregister(e *waiter.Entry) { + rfd.inode.t.ld.replicaWaiter.EventUnregister(e) } // Readiness implements waiter.Waitable.Readiness. -func (sfd *replicaFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { - return sfd.inode.t.ld.replicaReadiness() +func (rfd *replicaFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return rfd.inode.t.ld.replicaReadiness() } // Read implements vfs.FileDescriptionImpl.Read. -func (sfd *replicaFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { - return sfd.inode.t.ld.inputQueueRead(ctx, dst) +func (rfd *replicaFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + return rfd.inode.t.ld.inputQueueRead(ctx, dst) } // Write implements vfs.FileDescriptionImpl.Write. -func (sfd *replicaFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { - return sfd.inode.t.ld.outputQueueWrite(ctx, src) +func (rfd *replicaFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + return rfd.inode.t.ld.outputQueueWrite(ctx, src) } // Ioctl implements vfs.FileDescriptionImpl.Ioctl. -func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (rfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { t := kernel.TaskFromContext(ctx) if t == nil { // ioctl(2) may only be called from a task goroutine. @@ -146,35 +145,35 @@ func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, arg switch cmd := args[1].Uint(); cmd { case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ // Get the number of bytes in the input queue read buffer. - return 0, sfd.inode.t.ld.inputQueueReadSize(t, io, args) + return 0, rfd.inode.t.ld.inputQueueReadSize(t, io, args) case linux.TCGETS: - return sfd.inode.t.ld.getTermios(t, args) + return rfd.inode.t.ld.getTermios(t, args) case linux.TCSETS: - return sfd.inode.t.ld.setTermios(t, args) + return rfd.inode.t.ld.setTermios(t, args) case linux.TCSETSW: // TODO(b/29356795): This should drain the output queue first. - return sfd.inode.t.ld.setTermios(t, args) + return rfd.inode.t.ld.setTermios(t, args) case linux.TIOCGPTN: - nP := primitive.Uint32(sfd.inode.t.n) + nP := primitive.Uint32(rfd.inode.t.n) _, err := nP.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCGWINSZ: - return 0, sfd.inode.t.ld.windowSize(t, args) + return 0, rfd.inode.t.ld.windowSize(t, args) case linux.TIOCSWINSZ: - return 0, sfd.inode.t.ld.setWindowSize(t, args) + return 0, rfd.inode.t.ld.setWindowSize(t, args) case linux.TIOCSCTTY: // Make the given terminal the controlling terminal of the // calling process. - return 0, sfd.inode.t.setControllingTTY(ctx, args, false /* isMaster */) + return 0, rfd.inode.t.setControllingTTY(ctx, args, false /* isMaster */) case linux.TIOCNOTTY: // Release this process's controlling terminal. - return 0, sfd.inode.t.releaseControllingTTY(ctx, args, false /* isMaster */) + return 0, rfd.inode.t.releaseControllingTTY(ctx, args, false /* isMaster */) case linux.TIOCGPGRP: // Get the foreground process group. - return sfd.inode.t.foregroundProcessGroup(ctx, args, false /* isMaster */) + return rfd.inode.t.foregroundProcessGroup(ctx, args, false /* isMaster */) case linux.TIOCSPGRP: // Set the foreground process group. - return sfd.inode.t.setForegroundProcessGroup(ctx, args, false /* isMaster */) + return rfd.inode.t.setForegroundProcessGroup(ctx, args, false /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY @@ -182,24 +181,24 @@ func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, arg } // SetStat implements vfs.FileDescriptionImpl.SetStat. -func (sfd *replicaFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { +func (rfd *replicaFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() - return sfd.inode.SetStat(ctx, fs, creds, opts) + fs := rfd.vfsfd.VirtualDentry().Mount().Filesystem() + return rfd.inode.SetStat(ctx, fs, creds, opts) } // Stat implements vfs.FileDescriptionImpl.Stat. -func (sfd *replicaFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() - return sfd.inode.Stat(ctx, fs, opts) +func (rfd *replicaFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := rfd.vfsfd.VirtualDentry().Mount().Filesystem() + return rfd.inode.Stat(ctx, fs, opts) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (sfd *replicaFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { - return sfd.Locks().LockPOSIX(ctx, &sfd.vfsfd, uid, t, start, length, whence, block) +func (rfd *replicaFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return rfd.Locks().LockPOSIX(ctx, &rfd.vfsfd, uid, t, start, length, whence, block) } // UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (sfd *replicaFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { - return sfd.Locks().UnlockPOSIX(ctx, &sfd.vfsfd, uid, start, length, whence) +func (rfd *replicaFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return rfd.Locks().UnlockPOSIX(ctx, &rfd.vfsfd, uid, start, length, whence) } diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index a23094e54..6d1753080 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -33,8 +33,10 @@ import ( const Name = "devtmpfs" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct { - initOnce sync.Once + initOnce sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1664): not yet supported. initErr error // fs is the tmpfs filesystem that backs all mounts of this FilesystemType. diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go index bb0bf3a07..1c27ad700 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd.go @@ -33,6 +33,8 @@ import ( // EventFileDescription implements vfs.FileDescriptionImpl for file-based event // notification (eventfd). Eventfds are usually internal to the Sentry but in // certain situations they may be converted into a host-backed eventfd. +// +// +stateify savable type EventFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 7a1b4219f..9bfed883a 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -20,6 +20,8 @@ import ( ) // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 0fc01668d..452450d82 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -28,6 +28,8 @@ import ( ) // directory represents a directory inode. It holds the childList in memory. +// +// +stateify savable type directory struct { inode inode @@ -39,7 +41,7 @@ type directory struct { // Lock Order (outermost locks must be taken first): // directory.mu // filesystem.mu - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // childList is a list containing (1) child dirents and (2) fake dirents // (with diskDirent == nil) that represent the iteration position of @@ -120,6 +122,8 @@ func (i *inode) isDir() bool { } // dirent is the directory.childList node. +// +// +stateify savable type dirent struct { diskDirent disklayout.Dirent @@ -129,6 +133,8 @@ type dirent struct { // directoryFD represents a directory file description. It implements // vfs.FileDescriptionImpl. +// +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index 08ffc2834..aca258d40 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -34,6 +34,8 @@ import ( const Name = "ext" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // Compiles only if FilesystemType implements vfs.FilesystemType. diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go index c36225a7c..04917d762 100644 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ b/pkg/sentry/fsimpl/ext/extent_file.go @@ -24,6 +24,8 @@ import ( ) // extentFile is a type of regular file which uses extents to store file data. +// +// +stateify savable type extentFile struct { regFile regularFile diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index a4a6d8c55..917f1873d 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -38,11 +38,13 @@ var ( ) // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem // mu serializes changes to the Dentry tree. - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` // dev represents the underlying fs device. It does not require protection // because io.ReaderAt permits concurrent read calls to it. It translates to diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 30636cf66..9009ba3c7 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -37,6 +37,8 @@ import ( // |-- regular-- // |-- extent file // |-- block map file +// +// +stateify savable type inode struct { // refs is a reference count. refs is accessed using atomic memory operations. refs int64 diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index e73e740d6..4a5539b37 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -31,6 +31,8 @@ import ( // regularFile represents a regular file's inode. This too follows the // inheritance pattern prevelant in the vfs layer described in // pkg/sentry/vfs/README.md. +// +// +stateify savable type regularFile struct { inode inode @@ -67,6 +69,8 @@ func (in *inode) isRegular() bool { // directoryFD represents a directory file description. It implements // vfs.FileDescriptionImpl. +// +// +stateify savable type regularFileFD struct { fileDescription vfs.LockFD @@ -75,7 +79,7 @@ type regularFileFD struct { off int64 // offMu serializes operations that may mutate off. - offMu sync.Mutex + offMu sync.Mutex `state:"nosave"` } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index f33592d59..5e2bcc837 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -23,6 +23,8 @@ import ( ) // symlink represents a symlink inode. +// +// +stateify savable type symlink struct { inode inode target string // immutable @@ -64,6 +66,8 @@ func (in *inode) isSymlink() bool { // symlinkFD represents a symlink file description and implements // vfs.FileDescriptionImpl. which may only be used if open options contains // O_PATH. For this reason most of the functions return EBADF. +// +// +stateify savable type symlinkFD struct { fileDescription vfs.NoLockFD diff --git a/pkg/sentry/fsimpl/fuse/connection.go b/pkg/sentry/fsimpl/fuse/connection.go index dbc5e1954..8ccda1264 100644 --- a/pkg/sentry/fsimpl/fuse/connection.go +++ b/pkg/sentry/fsimpl/fuse/connection.go @@ -39,15 +39,18 @@ const ( ) // connection is the struct by which the sentry communicates with the FUSE server daemon. +// // Lock order: // - conn.fd.mu // - conn.mu // - conn.asyncMu +// +// +stateify savable type connection struct { fd *DeviceFD // mu protects access to struct memebers. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // attributeVersion is the version of connection's attributes. attributeVersion uint64 @@ -75,7 +78,7 @@ type connection struct { initialized int32 // initializedChan is used to block requests before initialization. - initializedChan chan struct{} + initializedChan chan struct{} `state:".(bool)"` // connected (connection established) when a new FUSE file system is created. // Set to false when: @@ -113,7 +116,7 @@ type connection struct { // i.e. `!request.noReply` // asyncMu protects the async request fields. - asyncMu sync.Mutex + asyncMu sync.Mutex `state:"nosave"` // asyncNum is the number of async requests. // Protected by asyncMu. @@ -174,6 +177,22 @@ type connection struct { noOpen bool } +func (conn *connection) saveInitializedChan() bool { + select { + case <-conn.initializedChan: + return true // Closed. + default: + return false // Not closed. + } +} + +func (conn *connection) loadInitializedChan(closed bool) { + conn.initializedChan = make(chan struct{}, 1) + if closed { + close(conn.initializedChan) + } +} + // newFUSEConnection creates a FUSE connection to fd. func newFUSEConnection(_ context.Context, fd *vfs.FileDescription, opts *filesystemOptions) (*connection, error) { // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index f690ef5ad..1b86a4b4c 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -31,6 +31,8 @@ import ( const fuseDevMinor = 229 // fuseDevice implements vfs.Device for /dev/fuse. +// +// +stateify savable type fuseDevice struct{} // Open implements vfs.Device.Open. @@ -49,6 +51,8 @@ func (fuseDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // DeviceFD implements vfs.FileDescriptionImpl for /dev/fuse. +// +// +stateify savable type DeviceFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -79,7 +83,7 @@ type DeviceFD struct { writeCursorFR *futureResponse // mu protects all the queues, maps, buffers and cursors and nextOpID. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // waitQueue is used to notify interested parties when the device becomes // readable or writable. @@ -88,12 +92,20 @@ type DeviceFD struct { // fullQueueCh is a channel used to synchronize the readers with the writers. // Writers (inbound requests to the filesystem) block if there are too many // unprocessed in-flight requests. - fullQueueCh chan struct{} + fullQueueCh chan struct{} `state:".(int)"` // fs is the FUSE filesystem that this FD is being used for. fs *filesystem } +func (fd *DeviceFD) saveFullQueueCh() int { + return cap(fd.fullQueueCh) +} + +func (fd *DeviceFD) loadFullQueueCh(capacity int) { + fd.fullQueueCh = make(chan struct{}, capacity) +} + // Release implements vfs.FileDescriptionImpl.Release. func (fd *DeviceFD) Release(ctx context.Context) { if fd.fs != nil { diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 1d42a51f4..65786e42a 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -41,8 +41,11 @@ const Name = "fuse" const maxActiveRequestsDefault = 10000 // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} +// +stateify savable type filesystemOptions struct { // userID specifies the numeric uid of the mount owner. // This option should not be specified by the filesystem owner. @@ -73,6 +76,8 @@ type filesystemOptions struct { } // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { kernfs.Filesystem devMinor uint32 @@ -240,6 +245,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { inodeRefs kernfs.InodeAttrs @@ -419,7 +426,7 @@ func (*inode) Valid(ctx context.Context) bool { } // NewFile implements kernfs.Inode.NewFile. -func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { +func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*kernfs.Dentry, error) { kernelTask := kernel.TaskFromContext(ctx) if kernelTask == nil { log.Warningf("fusefs.Inode.NewFile: couldn't get kernel task from context", i.nodeID) @@ -433,15 +440,11 @@ func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) }, Name: name, } - d, err := i.newEntry(ctx, name, linux.S_IFREG, linux.FUSE_CREATE, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, linux.S_IFREG, linux.FUSE_CREATE, &in) } // NewNode implements kernfs.Inode.NewNode. -func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) { +func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*kernfs.Dentry, error) { in := linux.FUSEMknodIn{ MknodMeta: linux.FUSEMknodMeta{ Mode: uint32(opts.Mode), @@ -450,28 +453,20 @@ func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) }, Name: name, } - d, err := i.newEntry(ctx, name, opts.Mode.FileType(), linux.FUSE_MKNOD, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, opts.Mode.FileType(), linux.FUSE_MKNOD, &in) } // NewSymlink implements kernfs.Inode.NewSymlink. -func (i *inode) NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) { +func (i *inode) NewSymlink(ctx context.Context, name, target string) (*kernfs.Dentry, error) { in := linux.FUSESymLinkIn{ Name: name, Target: target, } - d, err := i.newEntry(ctx, name, linux.S_IFLNK, linux.FUSE_SYMLINK, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, linux.S_IFLNK, linux.FUSE_SYMLINK, &in) } // Unlink implements kernfs.Inode.Unlink. -func (i *inode) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { +func (i *inode) Unlink(ctx context.Context, name string, child *kernfs.Dentry) error { kernelTask := kernel.TaskFromContext(ctx) if kernelTask == nil { log.Warningf("fusefs.Inode.newEntry: couldn't get kernel task from context", i.nodeID) @@ -494,7 +489,7 @@ func (i *inode) Unlink(ctx context.Context, name string, child *vfs.Dentry) erro } // NewDir implements kernfs.Inode.NewDir. -func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { +func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*kernfs.Dentry, error) { in := linux.FUSEMkdirIn{ MkdirMeta: linux.FUSEMkdirMeta{ Mode: uint32(opts.Mode), @@ -502,15 +497,11 @@ func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) }, Name: name, } - d, err := i.newEntry(ctx, name, linux.S_IFDIR, linux.FUSE_MKDIR, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, linux.S_IFDIR, linux.FUSE_MKDIR, &in) } // RmDir implements kernfs.Inode.RmDir. -func (i *inode) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { +func (i *inode) RmDir(ctx context.Context, name string, child *kernfs.Dentry) error { fusefs := i.fs task, creds := kernel.TaskFromContext(ctx), auth.CredentialsFromContext(ctx) @@ -528,12 +519,7 @@ func (i *inode) RmDir(ctx context.Context, name string, child *vfs.Dentry) error return err } - // TODO(Before merging): When creating new nodes, should we add nodes to the ordered children? - // If so we'll probably need to call this. We will also need to add them with the writable flag when - // appropriate. - // return i.OrderedChildren.RmDir(ctx, name, child) - - return nil + return i.dentry.RemoveChildLocked(name, child) } // newEntry calls FUSE server for entry creation and allocates corresponding entry according to response. @@ -563,11 +549,6 @@ func (i *inode) newEntry(ctx context.Context, name string, fileType linux.FileMo return nil, syserror.EIO } child := i.fs.newInode(out.NodeID, out.Attr) - if opcode == linux.FUSE_LOOKUP { - i.dentry.InsertChildLocked(name, child) - } else { - i.dentry.InsertChild(name, child) - } return child, nil } diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 91d2ae199..18c884b59 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -117,11 +117,12 @@ func (d *dentry) createSyntheticChildLocked(opts *createSyntheticOpts) { d.syntheticChildren++ } +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 dirents []vfs.Dirent } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 97b9165cc..94d96261b 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1416,11 +1416,11 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - if err := d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()); err != nil { - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + err = d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + if err != nil { return err } - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) @@ -1556,11 +1556,11 @@ func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - if err := d.setXattr(ctx, rp.Credentials(), &opts); err != nil { - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + err = d.setXattr(ctx, rp.Credentials(), &opts) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + if err != nil { return err } - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil @@ -1575,11 +1575,11 @@ func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - if err := d.removeXattr(ctx, rp.Credentials(), name); err != nil { - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + err = d.removeXattr(ctx, rp.Credentials(), name) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + if err != nil { return err } - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index aaad9c0d9..8608471f8 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -62,9 +62,13 @@ import ( const Name = "9p" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -77,7 +81,7 @@ type filesystem struct { iopts InternalFilesystemOptions // client is the client used by this filesystem. client is immutable. - client *p9.Client + client *p9.Client `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // clock is a realtime clock used to set timestamps in file operations. clock ktime.Clock @@ -95,7 +99,7 @@ type filesystem struct { // reference count (such that it is usable as vfs.ResolvingPath.Start() or // is reachable from its children), or if it is a child dentry (such that // it is reachable from its parent). - renameMu sync.RWMutex + renameMu sync.RWMutex `state:"nosave"` // cachedDentries contains all dentries with 0 references. (Due to race // conditions, it may also contain dentries with non-zero references.) @@ -107,7 +111,7 @@ type filesystem struct { // syncableDentries contains all dentries in this filesystem for which // !dentry.file.isNil(). specialFileFDs contains all open specialFileFDs. // These fields are protected by syncMu. - syncMu sync.Mutex + syncMu sync.Mutex `state:"nosave"` syncableDentries map[*dentry]struct{} specialFileFDs map[*specialFileFD]struct{} @@ -120,6 +124,8 @@ type filesystem struct { // dentries, it comes from QID.Path from the 9P server. Synthetic dentries // have have their inodeNumber generated sequentially, with the MSB reserved to // prevent conflicts with regular dentries. +// +// +stateify savable type inodeNumber uint64 // Reserve MSB for synthetic mounts. @@ -132,6 +138,7 @@ func inoFromPath(path uint64) inodeNumber { return inodeNumber(path &^ syntheticInoMask) } +// +stateify savable type filesystemOptions struct { // "Standard" 9P options. fd int @@ -177,6 +184,8 @@ type filesystemOptions struct { // InteropMode controls the client's interaction with other remote filesystem // users. +// +// +stateify savable type InteropMode uint32 const ( @@ -235,6 +244,8 @@ const ( // InternalFilesystemOptions may be passed as // vfs.GetFilesystemOptions.InternalData to FilesystemType.GetFilesystem. +// +// +stateify savable type InternalFilesystemOptions struct { // If LeakConnection is true, do not close the connection to the server // when the Filesystem is released. This is necessary for deployments in @@ -534,6 +545,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -563,7 +576,7 @@ type dentry struct { // If file.isNil(), this dentry represents a synthetic file, i.e. a file // that does not exist on the remote filesystem. As of this writing, the // only files that can be synthetic are sockets, pipes, and directories. - file p9file + file p9file `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // If deleted is non-zero, the file represented by this dentry has been // deleted. deleted is accessed using atomic memory operations. @@ -575,7 +588,7 @@ type dentry struct { cached bool dentryEntry - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` // If this dentry represents a directory, children contains: // @@ -607,7 +620,7 @@ type dentry struct { // To mutate: // - Lock metadataMu and use atomic operations to update because we might // have atomic readers that don't hold the lock. - metadataMu sync.Mutex + metadataMu sync.Mutex `state:"nosave"` ino inodeNumber // immutable mode uint32 // type is immutable, perms are mutable uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic @@ -638,7 +651,7 @@ type dentry struct { // other metadata fields. nlink uint32 - mapsMu sync.Mutex + mapsMu sync.Mutex `state:"nosave"` // If this dentry represents a regular file, mappings tracks mappings of // the file into memmap.MappingSpaces. mappings is protected by mapsMu. @@ -662,12 +675,12 @@ type dentry struct { // either p9.File transitions from closed (isNil() == true) to open // (isNil() == false), it may be mutated with handleMu locked, but cannot // be closed until the dentry is destroyed. - handleMu sync.RWMutex - readFile p9file - writeFile p9file + handleMu sync.RWMutex `state:"nosave"` + readFile p9file `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. + writeFile p9file `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. hostFD int32 - dataMu sync.RWMutex + dataMu sync.RWMutex `state:"nosave"` // If this dentry represents a regular file that is client-cached, cache // maps offsets into the cached file to offsets into @@ -1627,12 +1640,14 @@ func (d *dentry) decLinks() { // fileDescription is embedded by gofer implementations of // vfs.FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.LockFD - lockLogging sync.Once + lockLogging sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. } func (fd *fileDescription) filesystem() *filesystem { diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go index 104157512..a9ebe1206 100644 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -25,6 +25,8 @@ import ( // handle represents a remote "open file descriptor", consisting of an opened // fid (p9.File) and optionally a host file descriptor. +// +// These are explicitly not savable. type handle struct { file p9file fd int32 // -1 if unavailable diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 24f03ee94..eeaf6e444 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -39,11 +39,12 @@ func (d *dentry) isRegularFile() bool { return d.fileType() == linux.S_IFREG } +// +stateify savable type regularFileFD struct { fileDescription // off is the file offset. off is protected by mu. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 } @@ -898,6 +899,8 @@ func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { // dentryPlatformFile is only used when a host FD representing the remote file // is available (i.e. dentry.hostFD >= 0), and that FD is used for application // memory mappings (i.e. !filesystem.opts.forcePageCache). +// +// +stateify savable type dentryPlatformFile struct { *dentry @@ -910,7 +913,7 @@ type dentryPlatformFile struct { hostFileMapper fsutil.HostFileMapper // hostFileMapperInitOnce is used to lazily initialize hostFileMapper. - hostFileMapperInitOnce sync.Once + hostFileMapperInitOnce sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. } // IncRef implements memmap.File.IncRef. diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go index 85d2bee72..326b940a7 100644 --- a/pkg/sentry/fsimpl/gofer/socket.go +++ b/pkg/sentry/fsimpl/gofer/socket.go @@ -36,12 +36,14 @@ func (d *dentry) isSocket() bool { // An endpoint's lifetime is the time between when filesystem.BoundEndpointAt() // is called and either BoundEndpoint.BidirectionalConnect or // BoundEndpoint.UnidirectionalConnect is called. +// +// +stateify savable type endpoint struct { // dentry is the filesystem dentry which produced this endpoint. dentry *dentry // file is the p9 file that contains a single unopened fid. - file p9.File + file p9.File `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // path is the sentry path where this endpoint is bound. path string diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 576c57491..71581736c 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -34,11 +34,13 @@ import ( // special files, and (when filesystemOptions.regularFilesUseSpecialFileFD is // in effect) regular files. specialFileFD differs from regularFileFD by using // per-FD handles instead of shared per-dentry handles, and never buffering I/O. +// +// +stateify savable type specialFileFD struct { fileDescription // handle is used for file I/O. handle is immutable. - handle handle + handle handle `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // isRegularFile is true if this FD represents a regular file which is only // possible when filesystemOptions.regularFilesUseSpecialFileFD is in @@ -56,7 +58,7 @@ type specialFileFD struct { queue waiter.Queue // If seekable is true, off is the file offset. off is protected by mu. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 } diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index c0cef3453..ffe4ddb32 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -137,6 +137,8 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs } // filesystemType implements vfs.FilesystemType. +// +// +stateify savable type filesystemType struct{} // GetFilesystem implements vfs.FilesystemType.GetFilesystem. @@ -166,6 +168,8 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { } // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -185,6 +189,8 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { kernfs.InodeNoStatFS kernfs.InodeNotDirectory @@ -233,7 +239,7 @@ type inode struct { canMap bool // mapsMu protects mappings. - mapsMu sync.Mutex + mapsMu sync.Mutex `state:"nosave"` // If canMap is true, mappings tracks mappings of hostFD into // memmap.MappingSpaces. @@ -511,6 +517,8 @@ func (i *inode) open(ctx context.Context, d *kernfs.Dentry, mnt *vfs.Mount, flag } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -525,7 +533,7 @@ type fileDescription struct { inode *inode // offsetMu protects offset. - offsetMu sync.Mutex + offsetMu sync.Mutex `state:"nosave"` // offset specifies the current file offset. It is only meaningful when // inode.seekable is true. diff --git a/pkg/sentry/fsimpl/host/mmap.go b/pkg/sentry/fsimpl/host/mmap.go index 65d3af38c..b51a17bed 100644 --- a/pkg/sentry/fsimpl/host/mmap.go +++ b/pkg/sentry/fsimpl/host/mmap.go @@ -27,11 +27,13 @@ import ( // cannot implement both kernfs.Inode.IncRef and memmap.File.IncRef. // // inodePlatformFile should only be used if inode.canMap is true. +// +// +stateify savable type inodePlatformFile struct { *inode // fdRefsMu protects fdRefs. - fdRefsMu sync.Mutex + fdRefsMu sync.Mutex `state:"nosave"` // fdRefs counts references on memmap.File offsets. It is used solely for // memory accounting. @@ -41,7 +43,7 @@ type inodePlatformFile struct { fileMapper fsutil.HostFileMapper // fileMapperInitOnce is used to lazily initialize fileMapper. - fileMapperInitOnce sync.Once + fileMapperInitOnce sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. } // IncRef implements memmap.File.IncRef. diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index e02b9b8f6..f5c596fec 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -30,6 +30,8 @@ import ( // TTYFileDescription implements vfs.FileDescriptionImpl for a host file // descriptor that wraps a TTY FD. +// +// +stateify savable type TTYFileDescription struct { fileDescription diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 7d040481f..b929118b1 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -58,7 +58,7 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, devMajor, devMinor uint // Open implements Inode.Open. func (f *DynamicBytesFile) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &DynamicBytesFD{} - if err := fd.Init(rp.Mount(), d.VFSDentry(), f.data, &f.locks, opts.Flags); err != nil { + if err := fd.Init(rp.Mount(), d, f.data, &f.locks, opts.Flags); err != nil { return nil, err } return &fd.vfsfd, nil @@ -87,12 +87,12 @@ type DynamicBytesFD struct { } // Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *vfs.FileLocks, flags uint32) error { +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *Dentry, data vfs.DynamicBytesSource, locks *vfs.FileLocks, flags uint32) error { fd.LockFD.Init(locks) - if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, flags, m, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return err } - fd.inode = d.Impl().(*Dentry).inode + fd.inode = d.inode fd.SetDataSource(data) return nil } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 2b4294228..0a4cd4057 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -29,6 +29,8 @@ import ( ) // SeekEndConfig describes the SEEK_END behaviour for FDs. +// +// +stateify savable type SeekEndConfig int // Constants related to SEEK_END behaviour for FDs. @@ -41,6 +43,8 @@ const ( ) // GenericDirectoryFDOptions contains configuration for a GenericDirectoryFD. +// +// +stateify savable type GenericDirectoryFDOptions struct { SeekEnd SeekEndConfig } @@ -56,6 +60,8 @@ type GenericDirectoryFDOptions struct { // Must be initialize with Init before first use. // // Lock ordering: mu => children.mu. +// +// +stateify savable type GenericDirectoryFD struct { vfs.FileDescriptionDefaultImpl vfs.DirectoryFileDescriptionDefaultImpl @@ -68,7 +74,7 @@ type GenericDirectoryFD struct { children *OrderedChildren // mu protects the fields below. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // off is the current directory offset. Protected by "mu". off int64 @@ -195,8 +201,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent // these. childIdx := fd.off - 2 for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { - inode := it.Dentry.Impl().(*Dentry).inode - stat, err := inode.Stat(ctx, fd.filesystem(), opts) + stat, err := it.Dentry.inode.Stat(ctx, fd.filesystem(), opts) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index daad1e04a..5cc1c4281 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -37,8 +37,7 @@ import ( // * !rp.Done(). // // Postcondition: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, mayFollowSymlinks bool) (*vfs.Dentry, error) { - d := vfsd.Impl().(*Dentry) +func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, mayFollowSymlinks bool) (*Dentry, error) { if !d.isDir() { return nil, syserror.ENOTDIR } @@ -55,20 +54,20 @@ afterSymlink: // calls d_revalidate(), but walk_component() => handle_dots() does not. if name == "." { rp.Advance() - return vfsd, nil + return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(ctx, vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, d.VFSDentry()); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() - return vfsd, nil + return d, nil } - if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, d.parent.VFSDentry()); err != nil { return nil, err } rp.Advance() - return &d.parent.vfsd, nil + return d.parent, nil } if len(name) > linux.NAME_MAX { return nil, syserror.ENAMETOOLONG @@ -79,7 +78,7 @@ afterSymlink: if err != nil { return nil, err } - if err := rp.CheckMount(ctx, &next.vfsd); err != nil { + if err := rp.CheckMount(ctx, next.VFSDentry()); err != nil { return nil, err } // Resolve any symlink at current path component. @@ -102,7 +101,7 @@ afterSymlink: goto afterSymlink } rp.Advance() - return &next.vfsd, nil + return next, nil } // revalidateChildLocked must be called after a call to parent.vfsd.Child(name) @@ -122,7 +121,7 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir if !child.inode.Valid(ctx) { delete(parent.children, name) vfsObj.InvalidateDentry(ctx, &child.vfsd) - fs.deferDecRef(&child.vfsd) // Reference from Lookup. + fs.deferDecRef(child) // Reference from Lookup. child = nil } } @@ -133,7 +132,8 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir if err != nil { return nil, err } - // Reference on childVFSD dropped by a corresponding Valid. + // Reference on c (provided by Lookup) will be dropped when the dentry + // fails validation. parent.InsertChildLocked(name, c) child = c } @@ -149,15 +149,14 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // // Postconditions: Caller must call fs.processDeferredDecRefs*. func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*Dentry, error) { - vfsd := rp.Start() + d := rp.Start().Impl().(*Dentry) for !rp.Done() { var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd, true /* mayFollowSymlinks */) + d, err = fs.stepExistingLocked(ctx, rp, d, true /* mayFollowSymlinks */) if err != nil { return nil, err } } - d := vfsd.Impl().(*Dentry) if rp.MustBeDir() && !d.isDir() { return nil, syserror.ENOTDIR } @@ -176,20 +175,19 @@ func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingP // * !rp.Done(). // // Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { - vfsd := rp.Start() +func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*Dentry, error) { + d := rp.Start().Impl().(*Dentry) for !rp.Final() { var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd, true /* mayFollowSymlinks */) + d, err = fs.stepExistingLocked(ctx, rp, d, true /* mayFollowSymlinks */) if err != nil { - return nil, nil, err + return nil, err } } - d := vfsd.Impl().(*Dentry) if !d.isDir() { - return nil, nil, syserror.ENOTDIR + return nil, syserror.ENOTDIR } - return vfsd, d.inode, nil + return d, nil } // checkCreateLocked checks that a file named rp.Component() may be created in @@ -197,10 +195,9 @@ func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving // // Preconditions: // * Filesystem.mu must be locked for at least reading. -// * parentInode == parentVFSD.Impl().(*Dentry).Inode. // * isDir(parentInode) == true. -func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { +func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parent *Dentry) (string, error) { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return "", err } pc := rp.Component() @@ -210,11 +207,10 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v if len(pc) > linux.NAME_MAX { return "", syserror.ENAMETOOLONG } - // FIXME(gvisor.dev/issue/1193): Data race due to not holding dirMu. - if _, ok := parentVFSD.Impl().(*Dentry).children[pc]; ok { + if _, ok := parent.children[pc]; ok { return "", syserror.EEXIST } - if parentVFSD.IsDead() { + if parent.VFSDentry().IsDead() { return "", syserror.ENOENT } return pc, nil @@ -288,12 +284,12 @@ func (fs *Filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa fs.mu.RLock() defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() - vfsd, _, err := fs.walkParentDirLocked(ctx, rp) + d, err := fs.walkParentDirLocked(ctx, rp) if err != nil { return nil, err } - vfsd.IncRef() // Ownership transferred to caller. - return vfsd, nil + d.IncRef() // Ownership transferred to caller. + return d.VFSDentry(), nil } // LinkAt implements vfs.FilesystemImpl.LinkAt. @@ -303,12 +299,15 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -325,11 +324,11 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EPERM } - childVFSD, err := parentInode.NewLink(ctx, pc, d.inode) + child, err := parent.inode.NewLink(ctx, pc, d.inode) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, child) return nil } @@ -340,12 +339,15 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -353,14 +355,14 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - childVFSD, err := parentInode.NewDir(ctx, pc, opts) + child, err := parent.inode.NewDir(ctx, pc, opts) if err != nil { if !opts.ForSyntheticMountpoint || err == syserror.EEXIST { return err } - childVFSD = newSyntheticDirectory(rp.Credentials(), opts.Mode) + child = newSyntheticDirectory(rp.Credentials(), opts.Mode) } - parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, child) return nil } @@ -371,12 +373,15 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -384,11 +389,11 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - newVFSD, err := parentInode.NewNode(ctx, pc, opts) + newD, err := parent.inode.NewNode(ctx, pc, opts) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, newVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, newD) return nil } @@ -450,13 +455,13 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf return d.inode.Open(ctx, rp, d, opts) } afterTrailingSymlink: - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return nil, err } // Check for search permission in the parent directory. - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. @@ -471,10 +476,10 @@ afterTrailingSymlink: return nil, syserror.ENAMETOOLONG } // Determine whether or not we need to create a file. - childVFSD, err := fs.stepExistingLocked(ctx, rp, parentVFSD, false /* mayFollowSymlinks */) + child, err := fs.stepExistingLocked(ctx, rp, parent, false /* mayFollowSymlinks */) if err == syserror.ENOENT { // Already checked for searchability above; now check for writability. - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -482,12 +487,14 @@ afterTrailingSymlink: } defer rp.Mount().EndWrite() // Create and open the child. - childVFSD, err = parentInode.NewFile(ctx, pc, opts) + child, err := parent.inode.NewFile(ctx, pc, opts) if err != nil { return nil, err } - child := childVFSD.Impl().(*Dentry) - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + // FIXME(gvisor.dev/issue/1193): Race between checking existence with + // fs.stepExistingLocked and parent.InsertChild. If possible, we should hold + // dirMu from one to the other. + parent.InsertChild(pc, child) child.inode.IncRef() defer child.inode.DecRef(ctx) unlock() @@ -500,7 +507,6 @@ afterTrailingSymlink: if mustCreate { return nil, syserror.EEXIST } - child := childVFSD.Impl().(*Dentry) if rp.ShouldFollowSymlink() && child.isSymlink() { targetVD, targetPathname, err := child.inode.Getlink(ctx, rp.Mount()) if err != nil { @@ -559,11 +565,10 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Resolve the destination directory first to verify that it's on this // Mount. - dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) + dstDir, err := fs.walkParentDirLocked(ctx, rp) if err != nil { return err } - dstDir := dstDirVFSD.Impl().(*Dentry) mnt := rp.Mount() if mnt != oldParentVD.Mount() { return syserror.EXDEV @@ -589,7 +594,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Can we create the dst dentry? var dst *Dentry - pc, err := checkCreateLocked(ctx, rp, dstDirVFSD, dstDirInode) + pc, err := checkCreateLocked(ctx, rp, dstDir) switch err { case nil: // Ok, continue with rename as replacement. @@ -600,14 +605,14 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } dst = dstDir.children[pc] if dst == nil { - panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDirVFSD)) + panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDir)) } default: return err } var dstVFSD *vfs.Dentry if dst != nil { - dstVFSD = &dst.vfsd + dstVFSD = dst.VFSDentry() } mntns := vfs.MountNamespaceFromContext(ctx) @@ -627,14 +632,14 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if err := virtfs.PrepareRenameDentry(mntns, srcVFSD, dstVFSD); err != nil { return err } - replaced, err := srcDir.inode.Rename(ctx, src.name, pc, srcVFSD, dstDirVFSD) + replaced, err := srcDir.inode.Rename(ctx, src.name, pc, src, dstDir) if err != nil { virtfs.AbortRenameDentry(srcVFSD, dstVFSD) return err } delete(srcDir.children, src.name) if srcDir != dstDir { - fs.deferDecRef(srcDirVFSD) + fs.deferDecRef(srcDir) dstDir.IncRef() } src.parent = dstDir @@ -643,7 +648,11 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa dstDir.children = make(map[string]*Dentry) } dstDir.children[pc] = src - virtfs.CommitRenameReplaceDentry(ctx, srcVFSD, replaced) + var replaceVFSD *vfs.Dentry + if replaced != nil { + replaceVFSD = replaced.VFSDentry() + } + virtfs.CommitRenameReplaceDentry(ctx, srcVFSD, replaceVFSD) return nil } @@ -682,7 +691,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } - if err := parentDentry.inode.RmDir(ctx, d.name, d.VFSDentry()); err != nil { + if err := parentDentry.inode.RmDir(ctx, d.name, d); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } @@ -736,12 +745,15 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -749,11 +761,11 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ return err } defer rp.Mount().EndWrite() - childVFSD, err := parentInode.NewSymlink(ctx, pc, target) + child, err := parent.inode.NewSymlink(ctx, pc, target) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, child) return nil } @@ -787,7 +799,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.Unlink(ctx, d.name, d.VFSDentry()); err != nil { + if err := parentDentry.inode.Unlink(ctx, d.name, d); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 5cc196980..49210e748 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -31,6 +31,8 @@ import ( // count for inodes, performing no extra actions when references are obtained or // released. This is suitable for simple file inodes that don't reference any // resources. +// +// +stateify savable type InodeNoopRefCount struct { } @@ -50,30 +52,32 @@ func (InodeNoopRefCount) TryIncRef() bool { // InodeDirectoryNoNewChildren partially implements the Inode interface. // InodeDirectoryNoNewChildren represents a directory inode which does not // support creation of new children. +// +// +stateify savable type InodeDirectoryNoNewChildren struct{} // NewFile implements Inode.NewFile. -func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*Dentry, error) { return nil, syserror.EPERM } // NewDir implements Inode.NewDir. -func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*Dentry, error) { return nil, syserror.EPERM } // NewLink implements Inode.NewLink. -func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*Dentry, error) { return nil, syserror.EPERM } // NewSymlink implements Inode.NewSymlink. -func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*Dentry, error) { return nil, syserror.EPERM } // NewNode implements Inode.NewNode. -func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*Dentry, error) { return nil, syserror.EPERM } @@ -81,6 +85,8 @@ func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOpt // inodeDirectory and inodeDynamicDirectory sub interfaces. Inodes that do not // represent directories can embed this to provide no-op implementations for // directory-related functions. +// +// +stateify savable type InodeNotDirectory struct { } @@ -90,42 +96,42 @@ func (InodeNotDirectory) HasChildren() bool { } // NewFile implements Inode.NewFile. -func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*Dentry, error) { panic("NewFile called on non-directory inode") } // NewDir implements Inode.NewDir. -func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*Dentry, error) { panic("NewDir called on non-directory inode") } // NewLink implements Inode.NewLinkink. -func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*Dentry, error) { panic("NewLink called on non-directory inode") } // NewSymlink implements Inode.NewSymlink. -func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*Dentry, error) { panic("NewSymlink called on non-directory inode") } // NewNode implements Inode.NewNode. -func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*Dentry, error) { panic("NewNode called on non-directory inode") } // Unlink implements Inode.Unlink. -func (InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { +func (InodeNotDirectory) Unlink(context.Context, string, *Dentry) error { panic("Unlink called on non-directory inode") } // RmDir implements Inode.RmDir. -func (InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { +func (InodeNotDirectory) RmDir(context.Context, string, *Dentry) error { panic("RmDir called on non-directory inode") } // Rename implements Inode.Rename. -func (InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { +func (InodeNotDirectory) Rename(context.Context, string, string, *Dentry, *Dentry) (*Dentry, error) { panic("Rename called on non-directory inode") } @@ -149,6 +155,8 @@ func (InodeNotDirectory) Valid(context.Context) bool { // dymanic entries (i.e. entries that are not "hashed" into the // vfs.Dentry.children) can embed this to provide no-op implementations for // functions related to dynamic entries. +// +// +stateify savable type InodeNoDynamicLookup struct{} // Lookup implements Inode.Lookup. @@ -169,6 +177,8 @@ func (InodeNoDynamicLookup) Valid(ctx context.Context) bool { // InodeNotSymlink partially implements the Inode interface, specifically the // inodeSymlink sub interface. All inodes that are not symlinks may embed this // to return the appropriate errors from symlink-related functions. +// +// +stateify savable type InodeNotSymlink struct{} // Readlink implements Inode.Readlink. @@ -186,6 +196,8 @@ func (InodeNotSymlink) Getlink(context.Context, *vfs.Mount) (vfs.VirtualDentry, // inode attributes. // // Must be initialized by Init prior to first use. +// +// +stateify savable type InodeAttrs struct { devMajor uint32 devMinor uint32 @@ -330,13 +342,16 @@ func (a *InodeAttrs) DecLinks() { } } +// +stateify savable type slot struct { Name string - Dentry *vfs.Dentry + Dentry *Dentry slotEntry } // OrderedChildrenOptions contains initialization options for OrderedChildren. +// +// +stateify savable type OrderedChildrenOptions struct { // Writable indicates whether vfs.FilesystemImpl methods implemented by // OrderedChildren may modify the tracked children. This applies to @@ -352,12 +367,14 @@ type OrderedChildrenOptions struct { // directories. // // Must be initialize with Init before first use. +// +// +stateify savable type OrderedChildren struct { // Can children be modified by user syscalls? It set to false, interface // methods that would modify the children return EPERM. Immutable. writable bool - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` order slotList set map[string]*slot } @@ -390,7 +407,7 @@ func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint3 if child.isDir() { links++ } - if err := o.Insert(name, child.VFSDentry()); err != nil { + if err := o.Insert(name, child); err != nil { panic(fmt.Sprintf("Collision when attempting to insert child %q (%+v) into %+v", name, child, d)) } d.InsertChild(name, child) @@ -407,7 +424,7 @@ func (o *OrderedChildren) HasChildren() bool { // Insert inserts child into o. This ignores the writability of o, as this is // not part of the vfs.FilesystemImpl interface, and is a lower-level operation. -func (o *OrderedChildren) Insert(name string, child *vfs.Dentry) error { +func (o *OrderedChildren) Insert(name string, child *Dentry) error { o.mu.Lock() defer o.mu.Unlock() if _, ok := o.set[name]; ok { @@ -431,10 +448,10 @@ func (o *OrderedChildren) removeLocked(name string) { } // Precondition: caller must hold o.mu for writing. -func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs.Dentry { +func (o *OrderedChildren) replaceChildLocked(name string, new *Dentry) *Dentry { if s, ok := o.set[name]; ok { // Existing slot with given name, simply replace the dentry. - var old *vfs.Dentry + var old *Dentry old, s.Dentry = s.Dentry, new return old } @@ -450,7 +467,7 @@ func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs. } // Precondition: caller must hold o.mu for reading or writing. -func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) error { +func (o *OrderedChildren) checkExistingLocked(name string, child *Dentry) error { s, ok := o.set[name] if !ok { return syserror.ENOENT @@ -462,7 +479,7 @@ func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) er } // Unlink implements Inode.Unlink. -func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { +func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *Dentry) error { if !o.writable { return syserror.EPERM } @@ -478,12 +495,13 @@ func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.De } // Rmdir implements Inode.Rmdir. -func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { +func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *Dentry) error { // We're not responsible for checking that child is a directory, that it's // empty, or updating any link counts; so this is the same as unlink. return o.Unlink(ctx, name, child) } +// +stateify savable type renameAcrossDifferentImplementationsError struct{} func (renameAcrossDifferentImplementationsError) Error() string { @@ -499,8 +517,8 @@ func (renameAcrossDifferentImplementationsError) Error() string { // that will support Rename. // // Postcondition: reference on any replaced dentry transferred to caller. -func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (*vfs.Dentry, error) { - dst, ok := dstDir.Impl().(*Dentry).inode.(interface{}).(*OrderedChildren) +func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *Dentry) (*Dentry, error) { + dst, ok := dstDir.inode.(interface{}).(*OrderedChildren) if !ok { return nil, renameAcrossDifferentImplementationsError{} } @@ -542,6 +560,8 @@ func (o *OrderedChildren) nthLocked(i int64) *slot { } // InodeSymlink partially implements Inode interface for symlinks. +// +// +stateify savable type InodeSymlink struct { InodeNotDirectory } @@ -615,6 +635,8 @@ func (s *StaticDirectory) DecRef(context.Context) { } // AlwaysValid partially implements kernfs.inodeDynamicLookup. +// +// +stateify savable type AlwaysValid struct{} // Valid implements kernfs.inodeDynamicLookup.Valid. @@ -624,6 +646,8 @@ func (*AlwaysValid) Valid(context.Context) bool { // InodeNoStatFS partially implements the Inode interface, where the client // filesystem doesn't support statfs(2). +// +// +stateify savable type InodeNoStatFS struct{} // StatFS implements Inode.StatFS. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 14bf43ede..6d3d79333 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -29,7 +29,7 @@ // // Reference Model: // -// Kernfs dentries represents named pointers to inodes. Dentries and inode have +// Kernfs dentries represents named pointers to inodes. Dentries and inodes have // independent lifetimes and reference counts. A child dentry unconditionally // holds a reference on its parent directory's dentry. A dentry also holds a // reference on the inode it points to. Multiple dentries can point to the same @@ -66,15 +66,17 @@ import ( // Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory // filesystem. Concrete implementations are expected to embed this in their own // Filesystem type. +// +// +stateify savable type Filesystem struct { vfsfs vfs.Filesystem - droppedDentriesMu sync.Mutex + droppedDentriesMu sync.Mutex `state:"nosave"` // droppedDentries is a list of dentries waiting to be DecRef()ed. This is // used to defer dentry destruction until mu can be acquired for // writing. Protected by droppedDentriesMu. - droppedDentries []*vfs.Dentry + droppedDentries []*Dentry // mu synchronizes the lifetime of Dentries on this filesystem. Holding it // for reading guarantees continued existence of any resolved dentries, but @@ -97,7 +99,7 @@ type Filesystem struct { // defer fs.mu.RUnlock() // ... // fs.deferDecRef(dentry) - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` // nextInoMinusOne is used to to allocate inode numbers on this // filesystem. Must be accessed by atomic operations. @@ -108,7 +110,7 @@ type Filesystem struct { // processDeferredDecRefs{,Locked}. See comment on Filesystem.mu. // // Precondition: d must not already be pending destruction. -func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { +func (fs *Filesystem) deferDecRef(d *Dentry) { fs.droppedDentriesMu.Lock() fs.droppedDentries = append(fs.droppedDentries, d) fs.droppedDentriesMu.Unlock() @@ -160,6 +162,8 @@ const ( // to, and child dentries hold a reference on their parent. // // Must be initialized by Init prior to first use. +// +// +stateify savable type Dentry struct { DentryRefs @@ -173,7 +177,11 @@ type Dentry struct { name string // dirMu protects children and the names of child Dentries. - dirMu sync.Mutex + // + // Note that holding fs.mu for writing is not sufficient; + // revalidateChildLocked(), which is a very hot path, may modify children with + // fs.mu acquired for reading only. + dirMu sync.Mutex `state:"nosave"` children map[string]*Dentry inode Inode @@ -240,9 +248,8 @@ func (d *Dentry) Watches() *vfs.Watches { func (d *Dentry) OnZeroWatches(context.Context) {} // InsertChild inserts child into the vfs dentry cache with the given name under -// this dentry. This does not update the directory inode, so calling this on -// its own isn't sufficient to insert a child into a directory. InsertChild -// updates the link count on d if required. +// this dentry. This does not update the directory inode, so calling this on its +// own isn't sufficient to insert a child into a directory. // // Precondition: d must represent a directory inode. func (d *Dentry) InsertChild(name string, child *Dentry) { @@ -254,10 +261,12 @@ func (d *Dentry) InsertChild(name string, child *Dentry) { // InsertChildLocked is equivalent to InsertChild, with additional // preconditions. // -// Precondition: d.dirMu must be locked. +// Preconditions: +// * d must represent a directory inode. +// * d.dirMu must be locked. func (d *Dentry) InsertChildLocked(name string, child *Dentry) { if !d.isDir() { - panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) + panic(fmt.Sprintf("InsertChildLocked called on non-directory Dentry: %+v.", d)) } d.IncRef() // DecRef in child's Dentry.destroy. child.parent = d @@ -273,7 +282,7 @@ func (d *Dentry) InsertChildLocked(name string, child *Dentry) { // isn't sufficient to remove a child from a directory. // // Precondition: d must represent a directory inode. -func (d *Dentry) RemoveChild(name string, child *vfs.Dentry) error { +func (d *Dentry) RemoveChild(name string, child *Dentry) error { d.dirMu.Lock() defer d.dirMu.Unlock() return d.RemoveChildLocked(name, child) @@ -283,7 +292,7 @@ func (d *Dentry) RemoveChild(name string, child *vfs.Dentry) error { // preconditions. // // Precondition: d.dirMu must be locked. -func (d *Dentry) RemoveChildLocked(name string, child *vfs.Dentry) error { +func (d *Dentry) RemoveChildLocked(name string, child *Dentry) error { if !d.isDir() { panic(fmt.Sprintf("RemoveChild called on non-directory Dentry: %+v.", d)) } @@ -291,7 +300,7 @@ func (d *Dentry) RemoveChildLocked(name string, child *vfs.Dentry) error { if !ok { return syserror.ENOENT } - if &c.vfsd != child { + if c != child { panic(fmt.Sprintf("Dentry hashed into inode doesn't match what vfs thinks! Child: %+v, vfs: %+v", c, child)) } delete(d.children, name) @@ -318,7 +327,6 @@ func (d *Dentry) Inode() Inode { // // - Checking that dentries passed to methods are of the appropriate file type. // - Checking permissions. -// - Updating link and reference counts. // // Specific responsibilities of implementations are documented below. type Inode interface { @@ -328,7 +336,8 @@ type Inode interface { inodeRefs // Methods related to node metadata. A generic implementation is provided by - // InodeAttrs. + // InodeAttrs. Note that a concrete filesystem using kernfs is responsible for + // managing link counts. inodeMetadata // Method for inodes that represent symlink. InodeNotSymlink provides a @@ -346,7 +355,7 @@ type Inode interface { // Open creates a file description for the filesystem object represented by // this inode. The returned file description should hold a reference on the - // inode for its lifetime. + // dentry for its lifetime. // // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing // the inode on which Open() is being called. @@ -400,30 +409,30 @@ type inodeDirectory interface { HasChildren() bool // NewFile creates a new regular file inode. - NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) + NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*Dentry, error) // NewDir creates a new directory inode. - NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) + NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*Dentry, error) // NewLink creates a new hardlink to a specified inode in this // directory. Implementations should create a new kernfs Dentry pointing to // target, and update target's link count. - NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) + NewLink(ctx context.Context, name string, target Inode) (*Dentry, error) // NewSymlink creates a new symbolic link inode. - NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) + NewSymlink(ctx context.Context, name, target string) (*Dentry, error) // NewNode creates a new filesystem node for a mknod syscall. - NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) + NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*Dentry, error) // Unlink removes a child dentry from this directory inode. - Unlink(ctx context.Context, name string, child *vfs.Dentry) error + Unlink(ctx context.Context, name string, child *Dentry) error // RmDir removes an empty child directory from this directory // inode. Implementations must update the parent directory's link count, // if required. Implementations are not responsible for checking that child // is a directory, checking for an empty directory. - RmDir(ctx context.Context, name string, child *vfs.Dentry) error + RmDir(ctx context.Context, name string, child *Dentry) error // Rename is called on the source directory containing an inode being // renamed. child should point to the resolved child in the source @@ -431,7 +440,7 @@ type inodeDirectory interface { // should return the replaced dentry or nil otherwise. // // Precondition: Caller must serialize concurrent calls to Rename. - Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (replaced *vfs.Dentry, err error) + Rename(ctx context.Context, oldname, newname string, child, dstDir *Dentry) (replaced *Dentry, err error) } type inodeDynamicLookup interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index c9f81d734..e413242dc 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -176,38 +176,36 @@ func (d *dir) DecRef(context.Context) { d.dirRefs.DecRef(d.Destroy) } -func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { +func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*kernfs.Dentry, error) { creds := auth.CredentialsFromContext(ctx) dir := d.fs.newDir(creds, opts.Mode, nil) - dirVFSD := dir.VFSDentry() - if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { + if err := d.OrderedChildren.Insert(name, dir); err != nil { dir.DecRef(ctx) return nil, err } d.IncLinks(1) - return dirVFSD, nil + return dir, nil } -func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { +func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*kernfs.Dentry, error) { creds := auth.CredentialsFromContext(ctx) f := d.fs.newFile(creds, "") - fVFSD := f.VFSDentry() - if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { + if err := d.OrderedChildren.Insert(name, f); err != nil { f.DecRef(ctx) return nil, err } - return fVFSD, nil + return f, nil } -func (*dir) NewLink(context.Context, string, kernfs.Inode) (*vfs.Dentry, error) { +func (*dir) NewLink(context.Context, string, kernfs.Inode) (*kernfs.Dentry, error) { return nil, syserror.EPERM } -func (*dir) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (*dir) NewSymlink(context.Context, string, string) (*kernfs.Dentry, error) { return nil, syserror.EPERM } -func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*kernfs.Dentry, error) { return nil, syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 443121c99..58a93eaac 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -24,6 +24,8 @@ import ( // StaticSymlink provides an Inode implementation for symlinks that point to // a immutable target. +// +// +stateify savable type StaticSymlink struct { InodeAttrs InodeNoopRefCount diff --git a/pkg/sentry/fsimpl/kernfs/synthetic_directory.go b/pkg/sentry/fsimpl/kernfs/synthetic_directory.go index 845de6c1d..ea7f073eb 100644 --- a/pkg/sentry/fsimpl/kernfs/synthetic_directory.go +++ b/pkg/sentry/fsimpl/kernfs/synthetic_directory.go @@ -41,12 +41,12 @@ type syntheticDirectory struct { var _ Inode = (*syntheticDirectory)(nil) -func newSyntheticDirectory(creds *auth.Credentials, perm linux.FileMode) *vfs.Dentry { +func newSyntheticDirectory(creds *auth.Credentials, perm linux.FileMode) *Dentry { inode := &syntheticDirectory{} inode.Init(creds, 0 /* devMajor */, 0 /* devMinor */, 0 /* ino */, perm) d := &Dentry{} d.Init(inode) - return &d.vfsd + return d } func (dir *syntheticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode) { @@ -69,12 +69,12 @@ func (dir *syntheticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, } // NewFile implements Inode.NewFile. -func (dir *syntheticDirectory) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*Dentry, error) { return nil, syserror.EPERM } // NewDir implements Inode.NewDir. -func (dir *syntheticDirectory) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*Dentry, error) { if !opts.ForSyntheticMountpoint { return nil, syserror.EPERM } @@ -87,16 +87,16 @@ func (dir *syntheticDirectory) NewDir(ctx context.Context, name string, opts vfs } // NewLink implements Inode.NewLink. -func (dir *syntheticDirectory) NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewLink(ctx context.Context, name string, target Inode) (*Dentry, error) { return nil, syserror.EPERM } // NewSymlink implements Inode.NewSymlink. -func (dir *syntheticDirectory) NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewSymlink(ctx context.Context, name, target string) (*Dentry, error) { return nil, syserror.EPERM } // NewNode implements Inode.NewNode. -func (dir *syntheticDirectory) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*Dentry, error) { return nil, syserror.EPERM } diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index 7ab42e71e..df4492346 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -100,12 +100,13 @@ func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string return whiteouts, readdirErr } +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 dirents []vfs.Dirent } @@ -116,10 +117,12 @@ func (fd *directoryFD) Release(ctx context.Context) { // IterDirents implements vfs.FileDescriptionImpl.IterDirents. func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + d := fd.dentry() + defer d.InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent) + fd.mu.Lock() defer fd.mu.Unlock() - d := fd.dentry() if fd.dirents == nil { ds, err := d.getDirents(ctx) if err != nil { diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index e9ce4bde1..bd11372d5 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -499,7 +499,13 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if err := create(parent, name, childLayer == lookupLayerUpperWhiteout); err != nil { return err } + parent.dirents = nil + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parent.watches.Notify(ctx, name, uint32(ev), 0 /* cookie */, vfs.InodeEvent, false /* unlinked */) return nil } @@ -631,6 +637,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. } return err } + old.watches.Notify(ctx, "", linux.IN_ATTRIB, 0 /* cookie */, vfs.InodeEvent, false /* unlinked */) return nil }) } @@ -975,6 +982,7 @@ func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.Resolving // just can't open it anymore for some reason. return nil, err } + parent.watches.Notify(ctx, childName, linux.IN_CREATE, 0 /* cookie */, vfs.PathEvent, false /* unlinked */) return &fd.vfsfd, nil } @@ -1236,6 +1244,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } } + vfs.InotifyRename(ctx, &renamed.watches, &oldParent.watches, &newParent.watches, oldName, newName, renamed.isDir()) return nil } @@ -1352,6 +1361,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error delete(parent.children, name) ds = appendDentry(ds, child) parent.dirents = nil + parent.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0 /* cookie */, vfs.InodeEvent, true /* unlinked */) return nil } @@ -1359,12 +1369,25 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + return err + } + err = d.setStatLocked(ctx, rp, opts) + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if err != nil { return err } + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ctx, ev, 0 /* cookie */, vfs.InodeEvent) + } + return nil +} + +// Precondition: d.fs.renameMu must be held for reading. +func (d *dentry) setStatLocked(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { mode := linux.FileMode(atomic.LoadUint32(&d.mode)) if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err @@ -1555,11 +1578,14 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error panic(fmt.Sprintf("unrecoverable overlayfs inconsistency: failed to create whiteout during UnlinkAt: %v", err)) } + var cw *vfs.Watches if child != nil { vfsObj.CommitDeleteDentry(ctx, &child.vfsd) delete(parent.children, name) ds = appendDentry(ds, child) + cw = &child.watches } + vfs.InotifyRemoveChild(ctx, cw, &parent.watches, name) parent.dirents = nil return nil } @@ -1636,13 +1662,20 @@ func (fs *filesystem) getXattr(ctx context.Context, d *dentry, creds *auth.Crede func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) return err } - return fs.setXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), &opts) + err = fs.setXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), &opts) + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0 /* cookie */, vfs.InodeEvent) + return nil } // Precondition: fs.renameMu must be locked. @@ -1673,13 +1706,20 @@ func (fs *filesystem) setXattrLocked(ctx context.Context, d *dentry, mnt *vfs.Mo func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) return err } - return fs.removeXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), name) + err = fs.removeXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), name) + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0 /* cookie */, vfs.InodeEvent) + return nil } // Precondition: fs.renameMu must be locked. diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index 6e04705c7..853aee951 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -39,6 +39,7 @@ func (d *dentry) readlink(ctx context.Context) (string, error) { }) } +// +stateify savable type nonDirectoryFD struct { fileDescription @@ -47,7 +48,7 @@ type nonDirectoryFD struct { // fileDescription.dentry().upperVD. cachedFlags is the last known value of // cachedFD.StatusFlags(). copiedUp, cachedFD, and cachedFlags are // protected by mu. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` copiedUp bool cachedFD *vfs.FileDescription cachedFlags uint32 @@ -183,6 +184,9 @@ func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) return err } d.updateAfterSetStatLocked(&opts) + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) + } return nil } diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index d0d26185e..dfbccd05f 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -51,6 +51,8 @@ import ( const Name = "overlay" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // Name implements vfs.FilesystemType.Name. @@ -60,6 +62,8 @@ func (FilesystemType) Name() string { // FilesystemOptions may be passed as vfs.GetFilesystemOptions.InternalData to // FilesystemType.GetFilesystem. +// +// +stateify savable type FilesystemOptions struct { // Callers passing FilesystemOptions to // overlay.FilesystemType.GetFilesystem() are responsible for ensuring that @@ -76,6 +80,8 @@ type FilesystemOptions struct { } // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -98,7 +104,7 @@ type filesystem struct { // renameMu synchronizes renaming with non-renaming operations in order to // ensure consistent lock ordering between dentry.dirMu in different // dentries. - renameMu sync.RWMutex + renameMu sync.RWMutex `state:"nosave"` // lastDirIno is the last inode number assigned to a directory. lastDirIno // is accessed using atomic memory operations. @@ -367,6 +373,8 @@ func (fs *filesystem) newDirIno() uint64 { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -399,7 +407,7 @@ type dentry struct { // and dirents (if not nil) is a cache of dirents as returned by // directoryFDs representing this directory. children is protected by // dirMu. - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` children map[string]*dentry dirents []vfs.Dirent @@ -409,7 +417,7 @@ type dentry struct { // If !upperVD.Ok(), it can transition to a valid vfs.VirtualDentry (i.e. // be copied up) with copyMu locked for writing; otherwise, it is // immutable. lowerVDs is always immutable. - copyMu sync.RWMutex + copyMu sync.RWMutex `state:"nosave"` upperVD vfs.VirtualDentry lowerVDs []vfs.VirtualDentry @@ -454,6 +462,13 @@ type dentry struct { isMappable uint32 locks vfs.FileLocks + + // watches is the set of inotify watches on the file repesented by this dentry. + // + // Note that hard links to the same file will not share the same set of + // watches, due to the fact that we do not have inode structures in this + // overlay implementation. + watches vfs.Watches } // newDentry creates a new dentry. The dentry initially has no references; it @@ -513,6 +528,14 @@ func (d *dentry) checkDropLocked(ctx context.Context) { if atomic.LoadInt64(&d.refs) != 0 { return } + + // Make sure that we do not lose watches on dentries that have not been + // deleted. Note that overlayfs never calls VFS.InvalidateDentry(), so + // d.vfsd.IsDead() indicates that d was deleted. + if !d.vfsd.IsDead() && d.watches.Size() > 0 { + return + } + // Refs is still zero; destroy it. d.destroyLocked(ctx) return @@ -541,6 +564,8 @@ func (d *dentry) destroyLocked(ctx context.Context) { lowerVD.DecRef(ctx) } + d.watches.HandleDeletion(ctx) + if d.parent != nil { d.parent.dirMu.Lock() if !d.vfsd.IsDead() { @@ -559,19 +584,36 @@ func (d *dentry) destroyLocked(ctx context.Context) { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. func (d *dentry) InotifyWithParent(ctx context.Context, events uint32, cookie uint32, et vfs.EventType) { - // TODO(gvisor.dev/issue/1479): Implement inotify. + if d.isDir() { + events |= linux.IN_ISDIR + } + + // overlayfs never calls VFS.InvalidateDentry(), so d.vfsd.IsDead() indicates + // that d was deleted. + deleted := d.vfsd.IsDead() + + d.fs.renameMu.RLock() + // The ordering below is important, Linux always notifies the parent first. + if d.parent != nil { + d.parent.watches.Notify(ctx, d.name, events, cookie, et, deleted) + } + d.watches.Notify(ctx, "", events, cookie, et, deleted) + d.fs.renameMu.RUnlock() } // Watches implements vfs.DentryImpl.Watches. func (d *dentry) Watches() *vfs.Watches { - // TODO(gvisor.dev/issue/1479): Implement inotify. - return nil + return &d.watches } // OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) OnZeroWatches(context.Context) {} +func (d *dentry) OnZeroWatches(ctx context.Context) { + if atomic.LoadInt64(&d.refs) == 0 { + d.fs.renameMu.Lock() + d.checkDropLocked(ctx) + d.fs.renameMu.Unlock() + } +} // iterLayers invokes yield on each layer comprising d, from top to bottom. If // any call to yield returns false, iterLayer stops iteration. @@ -652,6 +694,8 @@ func (d *dentry) updateAfterSetStatLocked(opts *vfs.SetStatOptions) { // fileDescription is embedded by overlay implementations of // vfs.FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -679,17 +723,33 @@ func (fd *fileDescription) GetXattr(ctx context.Context, opts vfs.GetXattrOption // SetXattr implements vfs.FileDescriptionImpl.SetXattr. func (fd *fileDescription) SetXattr(ctx context.Context, opts vfs.SetXattrOptions) error { fs := fd.filesystem() + d := fd.dentry() + fs.renameMu.RLock() - defer fs.renameMu.RUnlock() - return fs.setXattrLocked(ctx, fd.dentry(), fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), &opts) + err := fs.setXattrLocked(ctx, d, fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), &opts) + fs.renameMu.RUnlock() + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // RemoveXattr implements vfs.FileDescriptionImpl.RemoveXattr. func (fd *fileDescription) RemoveXattr(ctx context.Context, name string) error { fs := fd.filesystem() + d := fd.dentry() + fs.renameMu.RLock() - defer fs.renameMu.RUnlock() - return fs.removeXattrLocked(ctx, fd.dentry(), fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), name) + err := fs.removeXattrLocked(ctx, d, fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), name) + fs.renameMu.RUnlock() + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 33e247578..4e2da4810 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type filesystemType struct{} // Name implements vfs.FilesystemType.Name. @@ -43,6 +44,7 @@ func (filesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile panic("pipefs.filesystemType.GetFilesystem should never be called") } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -76,6 +78,8 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 03b5941b9..05d7948ea 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -41,6 +41,7 @@ func (FilesystemType) Name() string { return Name } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -84,6 +85,8 @@ func (fs *filesystem) Release(ctx context.Context) { // dynamicInode is an overfitted interface for common Inodes with // dynamicByteSource types used in procfs. +// +// +stateify savable type dynamicInode interface { kernfs.Inode vfs.DynamicBytesSource @@ -99,6 +102,7 @@ func (fs *filesystem) newDentry(creds *auth.Credentials, ino uint64, perm linux. return d } +// +stateify savable type staticFile struct { kernfs.DynamicBytesFile vfs.StaticData @@ -118,10 +122,13 @@ func newStaticDir(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64 // InternalData contains internal data passed in to the procfs mount via // vfs.GetFilesystemOptions.InternalData. +// +// +stateify savable type InternalData struct { Cgroups map[string]string } +// +stateify savable type implStatFS struct{} // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 47dc0ac9a..47ecd941c 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -116,6 +116,7 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb return offset, nil } +// +stateify savable type subtasksFD struct { kernfs.GenericDirectoryFD diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index ae461bb48..1f99183eb 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -129,6 +129,8 @@ func (i *taskInode) DecRef(context.Context) { // taskOwnedInode implements kernfs.Inode and overrides inode owner with task // effective user and group. +// +// +stateify savable type taskOwnedInode struct { kernfs.Inode diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index c7104458f..0866cea2b 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -51,6 +51,7 @@ func taskFDExists(ctx context.Context, t *kernel.Task, fd int32) bool { return true } +// +stateify savable type fdDir struct { locks vfs.FileLocks diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index feed5bc3f..b81c8279e 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -785,6 +785,7 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } +// +stateify savable type namespaceSymlink struct { kernfs.StaticSymlink @@ -832,6 +833,8 @@ func (s *namespaceSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.Vir // namespaceInode is a synthetic inode created to represent a namespace in // /proc/[pid]/ns/*. +// +// +stateify savable type namespaceInode struct { implStatFS kernfs.InodeAttrs @@ -865,6 +868,8 @@ func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *ker // namespace FD is a synthetic file that represents a namespace in // /proc/[pid]/ns/*. +// +// +stateify savable type namespaceFD struct { vfs.FileDescriptionDefaultImpl vfs.LockFD diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 1607eac19..e7f748655 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -616,6 +616,7 @@ type netSnmpData struct { var _ dynamicInode = (*netSnmpData)(nil) +// +stateify savable type snmpLine struct { prefix string header string diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index ad08c3626..d8f5dd509 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -236,6 +236,8 @@ func (i *tasksInode) DecRef(context.Context) { // staticFileSetStat implements a special static file that allows inode // attributes to be set. This is to support /proc files that are readonly, but // allow attributes to be set. +// +// +stateify savable type staticFileSetStat struct { dynamicBytesFileSetAttr vfs.StaticData diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 459a8e52e..f268c59b0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type selfSymlink struct { implStatFS kernfs.InodeAttrs @@ -74,6 +75,7 @@ func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, return syserror.EPERM } +// +stateify savable type threadSelfSymlink struct { implStatFS kernfs.InodeAttrs @@ -121,6 +123,8 @@ func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Creden // dynamicBytesFileSetAttr implements a special file that allows inode // attributes to be set. This is to support /proc files that are readonly, but // allow attributes to be set. +// +// +stateify savable type dynamicBytesFileSetAttr struct { kernfs.DynamicBytesFile } @@ -131,6 +135,8 @@ func (d *dynamicBytesFileSetAttr) SetStat(ctx context.Context, fs *vfs.Filesyste } // cpuStats contains the breakdown of CPU time for /proc/stat. +// +// +stateify savable type cpuStats struct { // user is time spent in userspace tasks with non-positive niceness. user uint64 diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index a3ffbb15e..3312b0418 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type tcpMemDir int const ( diff --git a/pkg/sentry/fsimpl/signalfd/signalfd.go b/pkg/sentry/fsimpl/signalfd/signalfd.go index 3c02af8c9..bf11b425a 100644 --- a/pkg/sentry/fsimpl/signalfd/signalfd.go +++ b/pkg/sentry/fsimpl/signalfd/signalfd.go @@ -27,6 +27,8 @@ import ( ) // SignalFileDescription implements vfs.FileDescriptionImpl for signal fds. +// +// +stateify savable type SignalFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -43,7 +45,7 @@ type SignalFileDescription struct { target *kernel.Task // mu protects mask. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // mask is the signal mask. Protected by mu. mask linux.SignalSet diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index f1e75e277..29e5371d6 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -28,6 +28,8 @@ import ( ) // filesystemType implements vfs.FilesystemType. +// +// +stateify savable type filesystemType struct{} // GetFilesystem implements vfs.FilesystemType.GetFilesystem. @@ -44,6 +46,7 @@ func (filesystemType) Name() string { return "sockfs" } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -80,6 +83,8 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { kernfs.InodeAttrs kernfs.InodeNoopRefCount diff --git a/pkg/sentry/fsimpl/sys/kcov.go b/pkg/sentry/fsimpl/sys/kcov.go index 252f75d26..b75d70ae6 100644 --- a/pkg/sentry/fsimpl/sys/kcov.go +++ b/pkg/sentry/fsimpl/sys/kcov.go @@ -36,6 +36,8 @@ func (fs *filesystem) newKcovFile(ctx context.Context, creds *auth.Credentials) } // kcovInode implements kernfs.Inode. +// +// +stateify savable type kcovInode struct { kernfs.InodeAttrs kernfs.InodeNoopRefCount @@ -63,6 +65,7 @@ func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.D return &fd.vfsfd, nil } +// +stateify savable type kcovFD struct { vfs.FileDescriptionDefaultImpl vfs.NoLockFD diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 8bcefc103..1568c581f 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -34,9 +34,13 @@ const Name = "sysfs" const defaultSysDirMode = linux.FileMode(0755) // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -117,6 +121,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dir implements kernfs.Inode. +// +// +stateify savable type dir struct { dirRefs kernfs.InodeAttrs @@ -169,6 +175,8 @@ func (d *dir) StatFS(ctx context.Context, fs *vfs.Filesystem) (linux.Statfs, err } // cpuFile implements kernfs.Inode. +// +// +stateify savable type cpuFile struct { implStatFS kernfs.DynamicBytesFile @@ -190,6 +198,7 @@ func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode li return d } +// +stateify savable type implStatFS struct{} // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go index ac8a4e3bb..8853c8ad2 100644 --- a/pkg/sentry/fsimpl/timerfd/timerfd.go +++ b/pkg/sentry/fsimpl/timerfd/timerfd.go @@ -28,6 +28,8 @@ import ( // TimerFileDescription implements vfs.FileDescriptionImpl for timer fds. It also // implements ktime.TimerListener. +// +// +stateify savable type TimerFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go index ac54d420d..9129d35b7 100644 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ b/pkg/sentry/fsimpl/tmpfs/device_file.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" ) +// +stateify savable type deviceFile struct { inode inode kind vfs.DeviceKind diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 070c75e68..e90669cf0 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// +stateify savable type directory struct { // Since directories can't be hard-linked, each directory can only be // associated with a single dentry, which we can store in the directory @@ -44,7 +45,7 @@ type directory struct { // (with inode == nil) that represent the iteration position of // directoryFDs. childList is used to support directoryFD.IterDirents() // efficiently. childList is protected by iterMu. - iterMu sync.Mutex + iterMu sync.Mutex `state:"nosave"` childList dentryList } @@ -86,6 +87,7 @@ func (dir *directory) mayDelete(creds *auth.Credentials, child *dentry) error { return vfs.CheckDeleteSticky(creds, linux.FileMode(atomic.LoadUint32(&dir.inode.mode)), auth.KUID(atomic.LoadUint32(&child.inode.uid))) } +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 1362c1602..e39cd305b 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -673,11 +673,11 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RUnlock() return err } - if err := d.inode.setStat(ctx, rp.Credentials(), &opts); err != nil { - fs.mu.RUnlock() + err = d.inode.setStat(ctx, rp.Credentials(), &opts) + fs.mu.RUnlock() + if err != nil { return err } - fs.mu.RUnlock() if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) @@ -822,11 +822,11 @@ func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.mu.RUnlock() return err } - if err := d.inode.setXattr(rp.Credentials(), &opts); err != nil { - fs.mu.RUnlock() + err = d.inode.setXattr(rp.Credentials(), &opts) + fs.mu.RUnlock() + if err != nil { return err } - fs.mu.RUnlock() d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil @@ -840,11 +840,11 @@ func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.mu.RUnlock() return err } - if err := d.inode.removeXattr(rp.Credentials(), name); err != nil { - fs.mu.RUnlock() + err = d.inode.removeXattr(rp.Credentials(), name) + fs.mu.RUnlock() + if err != nil { return err } - fs.mu.RUnlock() d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 5b0471ff4..d772db9e9 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type namedPipe struct { inode inode diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index b8699d064..a199eb33d 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -36,6 +36,8 @@ import ( ) // regularFile is a regular (=S_IFREG) tmpfs file. +// +// +stateify savable type regularFile struct { inode inode @@ -66,7 +68,7 @@ type regularFile struct { writableMappingPages uint64 // dataMu protects the fields below. - dataMu sync.RWMutex + dataMu sync.RWMutex `state:"nosave"` // data maps offsets into the file to offsets into memFile that store // the file's data. @@ -325,13 +327,14 @@ func (*regularFile) InvalidateUnsavable(context.Context) error { return nil } +// +stateify savable type regularFileFD struct { fileDescription // off is the file offset. off is accessed using atomic memory operations. // offMu serializes operations that may mutate off. off int64 - offMu sync.Mutex + offMu sync.Mutex `state:"nosave"` } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go index 3ed650474..5699d5975 100644 --- a/pkg/sentry/fsimpl/tmpfs/socket_file.go +++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go @@ -21,6 +21,8 @@ import ( ) // socketFile is a socket (=S_IFSOCK) tmpfs file. +// +// +stateify savable type socketFile struct { inode inode ep transport.BoundEndpoint diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go index b0de5fabe..a102a2ee2 100644 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ b/pkg/sentry/fsimpl/tmpfs/symlink.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) +// +stateify savable type symlink struct { inode inode target string // immutable diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 4658e1533..cefec8fde 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -51,9 +51,13 @@ import ( const Name = "tmpfs" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -67,7 +71,7 @@ type filesystem struct { devMinor uint32 // mu serializes changes to the Dentry tree. - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` nextInoMinusOne uint64 // accessed using atomic memory operations } @@ -78,6 +82,8 @@ func (FilesystemType) Name() string { } // FilesystemOpts is used to pass configuration data to tmpfs. +// +// +stateify savable type FilesystemOpts struct { // RootFileType is the FileType of the filesystem root. Valid values // are: S_IFDIR, S_IFREG, and S_IFLNK. Defaults to S_IFDIR. @@ -221,6 +227,8 @@ var globalStatfs = linux.Statfs{ } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -300,6 +308,8 @@ func (d *dentry) Watches() *vfs.Watches { func (d *dentry) OnZeroWatches(context.Context) {} // inode represents a filesystem object. +// +// +stateify savable type inode struct { // fs is the owning filesystem. fs is immutable. fs *filesystem @@ -316,12 +326,12 @@ type inode struct { // Inode metadata. Writing multiple fields atomically requires holding // mu, othewise atomic operations can be used. - mu sync.Mutex - mode uint32 // file type and mode - nlink uint32 // protected by filesystem.mu instead of inode.mu - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - ino uint64 // immutable + mu sync.Mutex `state:"nosave"` + mode uint32 // file type and mode + nlink uint32 // protected by filesystem.mu instead of inode.mu + uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic + gid uint32 // auth.KGID, but ... + ino uint64 // immutable // Linux's tmpfs has no concept of btime. atime int64 // nanoseconds @@ -668,6 +678,8 @@ func (i *inode) checkXattrPermissions(creds *auth.Credentials, name string, ats // fileDescription is embedded by tmpfs implementations of // vfs.FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/verity/verity.go b/pkg/sentry/fsimpl/verity/verity.go index a81e7f714..3129f290d 100644 --- a/pkg/sentry/fsimpl/verity/verity.go +++ b/pkg/sentry/fsimpl/verity/verity.go @@ -74,9 +74,13 @@ var noCrashOnVerificationFailure bool var verityMu sync.RWMutex // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -101,11 +105,13 @@ type filesystem struct { // renameMu synchronizes renaming with non-renaming operations in order // to ensure consistent lock ordering between dentry.dirMu in different // dentries. - renameMu sync.RWMutex + renameMu sync.RWMutex `state:"nosave"` } // InternalFilesystemOptions may be passed as // vfs.GetFilesystemOptions.InternalData to FilesystemType.GetFilesystem. +// +// +stateify savable type InternalFilesystemOptions struct { // RootMerkleFileName is the name of the verity root Merkle tree file. RootMerkleFileName string @@ -259,6 +265,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -285,7 +293,7 @@ type dentry struct { // and dirents (if not nil) is a cache of dirents as returned by // directoryFDs representing this directory. children is protected by // dirMu. - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` children map[string]*dentry // lowerVD is the VirtualDentry in the underlying file system. @@ -429,6 +437,8 @@ func (d *dentry) readlink(ctx context.Context) (string, error) { // FileDescription is a wrapper of the underlying lowerFD, with support to build // Merkle trees through the Linux fs-verity API to verify contents read from // lowerFD. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/kernel/kcov.go b/pkg/sentry/kernel/kcov.go index aad63aa99..d3e76ca7b 100644 --- a/pkg/sentry/kernel/kcov.go +++ b/pkg/sentry/kernel/kcov.go @@ -89,6 +89,10 @@ func (kcov *Kcov) TaskWork(t *Task) { kcov.mu.Lock() defer kcov.mu.Unlock() + if kcov.mode != linux.KCOV_TRACE_PC { + return + } + rw := &kcovReadWriter{ mf: kcov.mfp.MemoryFile(), fr: kcov.mappable.FileRange(), diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index d9c62ff91..d6c21adb7 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -514,6 +514,10 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // flushMountSourceRefs flushes the MountSources for all mounted filesystems // and open FDs. func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { + if VFS2Enabled { + return nil // Not relevant. + } + // Flush all mount sources for currently mounted filesystems in each task. flushed := make(map[*fs.MountNamespace]struct{}) k.tasks.mu.RLock() @@ -540,11 +544,6 @@ func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { // // Precondition: Must be called with the kernel paused. func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.FileDescription) error) (err error) { - // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - if VFS2Enabled { - return nil - } - ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { @@ -563,6 +562,10 @@ func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.Fi func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if VFS2Enabled { + return nil + } + return ts.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index e5acbac50..163af329b 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -52,6 +52,7 @@ var _ = socket.SocketVFS2(&socketVFS2{}) func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol int, fd int, flags uint32) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) s := &socketVFS2{ socketOpsCommon: socketOpsCommon{ diff --git a/pkg/sentry/socket/netlink/provider_vfs2.go b/pkg/sentry/socket/netlink/provider_vfs2.go index bb205be0d..e8930f031 100644 --- a/pkg/sentry/socket/netlink/provider_vfs2.go +++ b/pkg/sentry/socket/netlink/provider_vfs2.go @@ -52,6 +52,7 @@ func (*socketProviderVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol vfsfd := &s.vfsfd mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 59f6fe44d..3345124cc 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -37,6 +37,8 @@ import ( // SocketVFS2 implements socket.SocketVFS2 (and by extension, // vfs.FileDescriptionImpl) for Unix sockets. +// +// +stateify savable type SocketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -54,6 +56,7 @@ var _ = socket.SocketVFS2(&SocketVFS2{}) func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &vfs.FileLocks{}) if err != nil { diff --git a/pkg/sentry/state/state.go b/pkg/sentry/state/state.go index a06c9b8ab..245d2c5cf 100644 --- a/pkg/sentry/state/state.go +++ b/pkg/sentry/state/state.go @@ -61,8 +61,10 @@ func (opts SaveOpts) Save(k *kernel.Kernel, w *watchdog.Watchdog) error { log.Infof("Sandbox save started, pausing all tasks.") k.Pause() k.ReceiveTaskStates() - defer k.Unpause() - defer log.Infof("Tasks resumed after save.") + defer func() { + k.Unpause() + log.Infof("Tasks resumed after save.") + }() w.Stop() defer w.Start() diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 9c4db3047..bdfd3ca8f 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -52,6 +52,8 @@ const ( ) // anonFilesystemType implements FilesystemType. +// +// +stateify savable type anonFilesystemType struct{} // GetFilesystem implements FilesystemType.GetFilesystem. @@ -69,12 +71,15 @@ func (anonFilesystemType) Name() string { // // Since all Dentries in anonFilesystem are non-directories, all FilesystemImpl // methods that would require an anonDentry to be a directory return ENOTDIR. +// +// +stateify savable type anonFilesystem struct { vfsfs Filesystem devMinor uint32 } +// +stateify savable type anonDentry struct { vfsd Dentry diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index a69a5b2f1..320ab7ce1 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -89,6 +89,8 @@ func (d *Dentry) Impl() DentryImpl { // DentryImpl contains implementation details for a Dentry. Implementations of // DentryImpl should contain their associated Dentry by value as their first // field. +// +// +stateify savable type DentryImpl interface { // IncRef increments the Dentry's reference count. A Dentry with a non-zero // reference count must remain coherent with the state of the filesystem. diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index 1e9dffc8f..dde2ad79b 100644 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -22,6 +22,8 @@ import ( ) // DeviceKind indicates whether a device is a block or character device. +// +// +stateify savable type DeviceKind uint32 const ( @@ -44,6 +46,7 @@ func (kind DeviceKind) String() string { } } +// +stateify savable type devTuple struct { kind DeviceKind major uint32 diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 754e76aec..8f36c3e3b 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -27,6 +27,8 @@ import ( var epollCycleMu sync.Mutex // EpollInstance represents an epoll instance, as described by epoll(7). +// +// +stateify savable type EpollInstance struct { vfsfd FileDescription FileDescriptionDefaultImpl @@ -38,11 +40,11 @@ type EpollInstance struct { // interest is the set of file descriptors that are registered with the // EpollInstance for monitoring. interest is protected by interestMu. - interestMu sync.Mutex + interestMu sync.Mutex `state:"nosave"` interest map[epollInterestKey]*epollInterest // mu protects fields in registered epollInterests. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // ready is the set of file descriptors that may be "ready" for I/O. Note // that this must be an ordered list, not a map: "If more than maxevents @@ -55,6 +57,7 @@ type EpollInstance struct { ready epollInterestList } +// +stateify savable type epollInterestKey struct { // file is the registered FileDescription. No reference is held on file; // instead, when the last reference is dropped, FileDescription.DecRef() @@ -67,6 +70,8 @@ type epollInterestKey struct { } // epollInterest represents an EpollInstance's interest in a file descriptor. +// +// +stateify savable type epollInterest struct { // epoll is the owning EpollInstance. epoll is immutable. epoll *EpollInstance diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 73bb36d3e..1eba0270f 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -37,11 +37,13 @@ import ( // FileDescription methods require that a reference is held. // // FileDescription is analogous to Linux's struct file. +// +// +stateify savable type FileDescription struct { FileDescriptionRefs // flagsMu protects statusFlags and asyncHandler below. - flagsMu sync.Mutex + flagsMu sync.Mutex `state:"nosave"` // statusFlags contains status flags, "initialized by open(2) and possibly // modified by fcntl()" - fcntl(2). statusFlags can be read using atomic @@ -56,7 +58,7 @@ type FileDescription struct { // epolls is the set of epollInterests registered for this FileDescription. // epolls is protected by epollMu. - epollMu sync.Mutex + epollMu sync.Mutex `state:"nosave"` epolls map[*epollInterest]struct{} // vd is the filesystem location at which this FileDescription was opened. @@ -88,6 +90,8 @@ type FileDescription struct { } // FileDescriptionOptions contains options to FileDescription.Init(). +// +// +stateify savable type FileDescriptionOptions struct { // If AllowDirectIO is true, allow O_DIRECT to be set on the file. AllowDirectIO bool @@ -451,6 +455,8 @@ type FileDescriptionImpl interface { } // Dirent holds the information contained in struct linux_dirent64. +// +// +stateify savable type Dirent struct { // Name is the filename. Name string diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 78da16bac..48ca9de44 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -42,6 +42,8 @@ import ( // FileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl to obtain implementations of many FileDescriptionImpl // methods with default behavior analogous to Linux's. +// +// +stateify savable type FileDescriptionDefaultImpl struct{} // OnClose implements FileDescriptionImpl.OnClose analogously to @@ -166,6 +168,8 @@ func (FileDescriptionDefaultImpl) RemoveXattr(ctx context.Context, name string) // DirectoryFileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl that always represent directories to obtain // implementations of non-directory I/O methods that return EISDIR. +// +// +stateify savable type DirectoryFileDescriptionDefaultImpl struct{} // Allocate implements DirectoryFileDescriptionDefaultImpl.Allocate. @@ -196,6 +200,8 @@ func (DirectoryFileDescriptionDefaultImpl) Write(ctx context.Context, src userme // DentryMetadataFileDescriptionImpl may be embedded by implementations of // FileDescriptionImpl for which FileDescriptionOptions.UseDentryMetadata is // true to obtain implementations of Stat and SetStat that panic. +// +// +stateify savable type DentryMetadataFileDescriptionImpl struct{} // Stat implements FileDescriptionImpl.Stat. @@ -210,12 +216,16 @@ func (DentryMetadataFileDescriptionImpl) SetStat(ctx context.Context, opts SetSt // DynamicBytesSource represents a data source for a // DynamicBytesFileDescriptionImpl. +// +// +stateify savable type DynamicBytesSource interface { // Generate writes the file's contents to buf. Generate(ctx context.Context, buf *bytes.Buffer) error } // StaticData implements DynamicBytesSource over a static string. +// +// +stateify savable type StaticData struct { Data string } @@ -242,14 +252,24 @@ type WritableDynamicBytesSource interface { // // DynamicBytesFileDescriptionImpl.SetDataSource() must be called before first // use. +// +// +stateify savable type DynamicBytesFileDescriptionImpl struct { data DynamicBytesSource // immutable - mu sync.Mutex // protects the following fields - buf bytes.Buffer + mu sync.Mutex `state:"nosave"` // protects the following fields + buf bytes.Buffer `state:".([]byte)"` off int64 lastRead int64 // offset at which the last Read, PRead, or Seek ended } +func (fd *DynamicBytesFileDescriptionImpl) saveBuf() []byte { + return fd.buf.Bytes() +} + +func (fd *DynamicBytesFileDescriptionImpl) loadBuf(p []byte) { + fd.buf.Write(p) +} + // SetDataSource must be called exactly once on fd before first use. func (fd *DynamicBytesFileDescriptionImpl) SetDataSource(data DynamicBytesSource) { fd.data = data @@ -382,6 +402,8 @@ func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.M // LockFD may be used by most implementations of FileDescriptionImpl.Lock* // functions. Caller must call Init(). +// +// +stateify savable type LockFD struct { locks *FileLocks } @@ -409,6 +431,8 @@ func (fd *LockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { // NoLockFD implements Lock*/Unlock* portion of FileDescriptionImpl interface // returning ENOLCK. +// +// +stateify savable type NoLockFD struct{} // LockBSD implements vfs.FileDescriptionImpl.LockBSD. diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 7dae4e7e8..c93d94634 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -506,6 +506,8 @@ type FilesystemImpl interface { // PrependPathAtVFSRootError is returned by implementations of // FilesystemImpl.PrependPath() when they encounter the contextual VFS root. +// +// +stateify savable type PrependPathAtVFSRootError struct{} // Error implements error.Error. @@ -516,6 +518,8 @@ func (PrependPathAtVFSRootError) Error() string { // PrependPathAtNonMountRootError is returned by implementations of // FilesystemImpl.PrependPath() when they encounter an independent ancestor // Dentry that is not the Mount root. +// +// +stateify savable type PrependPathAtNonMountRootError struct{} // Error implements error.Error. @@ -526,6 +530,8 @@ func (PrependPathAtNonMountRootError) Error() string { // PrependPathSyntheticError is returned by implementations of // FilesystemImpl.PrependPath() for which prepended names do not represent real // paths. +// +// +stateify savable type PrependPathSyntheticError struct{} // Error implements error.Error. diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index f2298f7f6..bc19db1d5 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -55,10 +55,13 @@ type registeredFilesystemType struct { // RegisterFilesystemTypeOptions contains options to // VirtualFilesystem.RegisterFilesystem(). +// +// +stateify savable type RegisterFilesystemTypeOptions struct { - // If AllowUserMount is true, allow calls to VirtualFilesystem.MountAt() - // for which MountOptions.InternalMount == false to use this filesystem - // type. + // AllowUserMount determines whether users are allowed to mount a file system + // of this type, i.e. through mount(2). If AllowUserMount is true, allow calls + // to VirtualFilesystem.MountAt() for which MountOptions.InternalMount == false + // to use this filesystem type. AllowUserMount bool // If AllowUserList is true, make this filesystem type visible in diff --git a/pkg/sentry/vfs/genericfstree/genericfstree.go b/pkg/sentry/vfs/genericfstree/genericfstree.go index 8882fa84a..2d27d9d35 100644 --- a/pkg/sentry/vfs/genericfstree/genericfstree.go +++ b/pkg/sentry/vfs/genericfstree/genericfstree.go @@ -27,6 +27,8 @@ import ( ) // Dentry is a required type parameter that is a struct with the given fields. +// +// +stateify savable type Dentry struct { // vfsd is the embedded vfs.Dentry corresponding to this vfs.DentryImpl. vfsd vfs.Dentry diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index aff220a61..3f0b8f45b 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -37,6 +37,8 @@ const inotifyEventBaseSize = 16 // // The way events are labelled appears somewhat arbitrary, but they must match // Linux so that IN_EXCL_UNLINK behaves as it does in Linux. +// +// +stateify savable type EventType uint8 // PathEvent and InodeEvent correspond to FSNOTIFY_EVENT_PATH and diff --git a/pkg/sentry/vfs/lock.go b/pkg/sentry/vfs/lock.go index 42666eebf..55783d4eb 100644 --- a/pkg/sentry/vfs/lock.go +++ b/pkg/sentry/vfs/lock.go @@ -33,6 +33,8 @@ import ( // Note that in Linux these two types of locks are _not_ cooperative, because // race and deadlock conditions make merging them prohibitive. We do the same // and keep them oblivious to each other. +// +// +stateify savable type FileLocks struct { // bsd is a set of BSD-style advisory file wide locks, see flock(2). bsd fslock.Locks diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 9da09d4c1..dfc3ae6c0 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -65,7 +65,7 @@ type Mount struct { // // Invariant: key.parent != nil iff key.point != nil. key.point belongs to // key.parent.fs. - key mountKey + key mountKey `state:".(VirtualDentry)"` // ns is the namespace in which this Mount was mounted. ns is protected by // VirtualFilesystem.mountMu. @@ -345,6 +345,7 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti return nil } +// +stateify savable type umountRecursiveOptions struct { // If eager is true, ensure that future calls to Mount.tryIncMountedRef() // on umounted mounts fail. @@ -414,7 +415,7 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns } } mnt.IncRef() // dropped by callers of umountRecursiveLocked - mnt.storeKey(vd) + mnt.setKey(vd) if vd.mount.children == nil { vd.mount.children = make(map[*Mount]struct{}) } @@ -439,13 +440,13 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns // * vfs.mounts.seq must be in a writer critical section. // * mnt.parent() != nil. func (vfs *VirtualFilesystem) disconnectLocked(mnt *Mount) VirtualDentry { - vd := mnt.loadKey() + vd := mnt.getKey() if checkInvariants { if vd.mount != nil { panic("VFS.disconnectLocked called on disconnected mount") } } - mnt.storeKey(VirtualDentry{}) + mnt.loadKey(VirtualDentry{}) delete(vd.mount.children, mnt) atomic.AddUint32(&vd.dentry.mounts, math.MaxUint32) // -1 mnt.ns.mountpoints[vd.dentry]-- diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go index 3335e4057..cb8c56bd3 100644 --- a/pkg/sentry/vfs/mount_test.go +++ b/pkg/sentry/vfs/mount_test.go @@ -38,7 +38,7 @@ func TestMountTableInsertLookup(t *testing.T) { mt.Init() mount := &Mount{} - mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) + mount.setKey(VirtualDentry{&Mount{}, &Dentry{}}) mt.Insert(mount) if m := mt.Lookup(mount.parent(), mount.point()); m != mount { @@ -79,7 +79,7 @@ const enableComparativeBenchmarks = false func newBenchMount() *Mount { mount := &Mount{} - mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) + mount.loadKey(VirtualDentry{&Mount{}, &Dentry{}}) return mount } @@ -94,7 +94,7 @@ func BenchmarkMountTableParallelLookup(b *testing.B) { for i := 0; i < numMounts; i++ { mount := newBenchMount() mt.Insert(mount) - keys = append(keys, mount.loadKey()) + keys = append(keys, mount.saveKey()) } var ready sync.WaitGroup @@ -146,7 +146,7 @@ func BenchmarkMountMapParallelLookup(b *testing.B) { keys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { mount := newBenchMount() - key := mount.loadKey() + key := mount.saveKey() ms[key] = mount keys = append(keys, key) } @@ -201,7 +201,7 @@ func BenchmarkMountSyncMapParallelLookup(b *testing.B) { keys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { mount := newBenchMount() - key := mount.loadKey() + key := mount.getKey() ms.Store(key, mount) keys = append(keys, key) } @@ -283,7 +283,7 @@ func BenchmarkMountMapNegativeLookup(b *testing.B) { ms := make(map[VirtualDentry]*Mount) for i := 0; i < numMounts; i++ { mount := newBenchMount() - ms[mount.loadKey()] = mount + ms[mount.getKey()] = mount } negkeys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { @@ -318,7 +318,7 @@ func BenchmarkMountSyncMapNegativeLookup(b *testing.B) { var ms sync.Map for i := 0; i < numMounts; i++ { mount := newBenchMount() - ms.Store(mount.loadKey(), mount) + ms.Store(mount.saveKey(), mount) } negkeys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { @@ -372,7 +372,7 @@ func BenchmarkMountMapInsert(b *testing.B) { b.ResetTimer() for i := range mounts { mount := mounts[i] - ms[mount.loadKey()] = mount + ms[mount.saveKey()] = mount } } @@ -392,7 +392,7 @@ func BenchmarkMountSyncMapInsert(b *testing.B) { b.ResetTimer() for i := range mounts { mount := mounts[i] - ms.Store(mount.loadKey(), mount) + ms.Store(mount.saveKey(), mount) } } @@ -425,13 +425,13 @@ func BenchmarkMountMapRemove(b *testing.B) { ms := make(map[VirtualDentry]*Mount) for i := range mounts { mount := mounts[i] - ms[mount.loadKey()] = mount + ms[mount.saveKey()] = mount } b.ResetTimer() for i := range mounts { mount := mounts[i] - delete(ms, mount.loadKey()) + delete(ms, mount.saveKey()) } } @@ -447,12 +447,12 @@ func BenchmarkMountSyncMapRemove(b *testing.B) { var ms sync.Map for i := range mounts { mount := mounts[i] - ms.Store(mount.loadKey(), mount) + ms.Store(mount.saveKey(), mount) } b.ResetTimer() for i := range mounts { mount := mounts[i] - ms.Delete(mount.loadKey()) + ms.Delete(mount.saveKey()) } } diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index da2a2e9c4..b7d122d22 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -34,6 +34,8 @@ import ( // structurally identical to VirtualDentry, but stores its fields as // unsafe.Pointer since mutators synchronize with VFS path traversal using // seqcounts. +// +// This is explicitly not savable. type mountKey struct { parent unsafe.Pointer // *Mount point unsafe.Pointer // *Dentry @@ -47,19 +49,23 @@ func (mnt *Mount) point() *Dentry { return (*Dentry)(atomic.LoadPointer(&mnt.key.point)) } -func (mnt *Mount) loadKey() VirtualDentry { +func (mnt *Mount) getKey() VirtualDentry { return VirtualDentry{ mount: mnt.parent(), dentry: mnt.point(), } } +func (mnt *Mount) saveKey() VirtualDentry { return mnt.getKey() } + // Invariant: mnt.key.parent == nil. vd.Ok(). -func (mnt *Mount) storeKey(vd VirtualDentry) { +func (mnt *Mount) setKey(vd VirtualDentry) { atomic.StorePointer(&mnt.key.parent, unsafe.Pointer(vd.mount)) atomic.StorePointer(&mnt.key.point, unsafe.Pointer(vd.dentry)) } +func (mnt *Mount) loadKey(vd VirtualDentry) { mnt.setKey(vd) } + // mountTable maps (mount parent, mount point) pairs to mounts. It supports // efficient concurrent lookup, even in the presence of concurrent mutators // (provided mutation is sufficiently uncommon). @@ -92,6 +98,7 @@ type mountTable struct { // length and cap in separate uint32s) for ~free. size uint64 + // FIXME(gvisor.dev/issue/1663): Slots need to be saved. slots unsafe.Pointer `state:"nosave"` // []mountSlot; never nil after Init } diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index b33d36cb1..bc79e5ecc 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -21,6 +21,8 @@ import ( // GetDentryOptions contains options to VirtualFilesystem.GetDentryAt() and // FilesystemImpl.GetDentryAt(). +// +// +stateify savable type GetDentryOptions struct { // If CheckSearchable is true, FilesystemImpl.GetDentryAt() must check that // the returned Dentry is a directory for which creds has search @@ -30,6 +32,8 @@ type GetDentryOptions struct { // MkdirOptions contains options to VirtualFilesystem.MkdirAt() and // FilesystemImpl.MkdirAt(). +// +// +stateify savable type MkdirOptions struct { // Mode is the file mode bits for the created directory. Mode linux.FileMode @@ -56,6 +60,8 @@ type MkdirOptions struct { // MknodOptions contains options to VirtualFilesystem.MknodAt() and // FilesystemImpl.MknodAt(). +// +// +stateify savable type MknodOptions struct { // Mode is the file type and mode bits for the created file. Mode linux.FileMode @@ -72,6 +78,8 @@ type MknodOptions struct { // MountFlags contains flags as specified for mount(2), e.g. MS_NOEXEC. // MS_RDONLY is not part of MountFlags because it's tracked in Mount.writers. +// +// +stateify savable type MountFlags struct { // NoExec is equivalent to MS_NOEXEC. NoExec bool @@ -93,6 +101,8 @@ type MountFlags struct { } // MountOptions contains options to VirtualFilesystem.MountAt(). +// +// +stateify savable type MountOptions struct { // Flags contains flags as specified for mount(2), e.g. MS_NOEXEC. Flags MountFlags @@ -103,13 +113,17 @@ type MountOptions struct { // GetFilesystemOptions contains options to FilesystemType.GetFilesystem(). GetFilesystemOptions GetFilesystemOptions - // If InternalMount is true, allow the use of filesystem types for which - // RegisterFilesystemTypeOptions.AllowUserMount == false. + // InternalMount indicates whether the mount operation is coming from the + // application, i.e. through mount(2). If InternalMount is true, allow the use + // of filesystem types for which RegisterFilesystemTypeOptions.AllowUserMount + // == false. InternalMount bool } // OpenOptions contains options to VirtualFilesystem.OpenAt() and // FilesystemImpl.OpenAt(). +// +// +stateify savable type OpenOptions struct { // Flags contains access mode and flags as specified for open(2). // @@ -135,6 +149,8 @@ type OpenOptions struct { // ReadOptions contains options to FileDescription.PRead(), // FileDescriptionImpl.PRead(), FileDescription.Read(), and // FileDescriptionImpl.Read(). +// +// +stateify savable type ReadOptions struct { // Flags contains flags as specified for preadv2(2). Flags uint32 @@ -142,6 +158,8 @@ type ReadOptions struct { // RenameOptions contains options to VirtualFilesystem.RenameAt() and // FilesystemImpl.RenameAt(). +// +// +stateify savable type RenameOptions struct { // Flags contains flags as specified for renameat2(2). Flags uint32 @@ -153,6 +171,8 @@ type RenameOptions struct { // SetStatOptions contains options to VirtualFilesystem.SetStatAt(), // FilesystemImpl.SetStatAt(), FileDescription.SetStat(), and // FileDescriptionImpl.SetStat(). +// +// +stateify savable type SetStatOptions struct { // Stat is the metadata that should be set. Only fields indicated by // Stat.Mask should be set. @@ -174,6 +194,8 @@ type SetStatOptions struct { // BoundEndpointOptions contains options to VirtualFilesystem.BoundEndpointAt() // and FilesystemImpl.BoundEndpointAt(). +// +// +stateify savable type BoundEndpointOptions struct { // Addr is the path of the file whose socket endpoint is being retrieved. // It is generally irrelevant: most endpoints are stored at a dentry that @@ -193,6 +215,8 @@ type BoundEndpointOptions struct { // GetXattrOptions contains options to VirtualFilesystem.GetXattrAt(), // FilesystemImpl.GetXattrAt(), FileDescription.GetXattr(), and // FileDescriptionImpl.GetXattr(). +// +// +stateify savable type GetXattrOptions struct { // Name is the name of the extended attribute to retrieve. Name string @@ -207,6 +231,8 @@ type GetXattrOptions struct { // SetXattrOptions contains options to VirtualFilesystem.SetXattrAt(), // FilesystemImpl.SetXattrAt(), FileDescription.SetXattr(), and // FileDescriptionImpl.SetXattr(). +// +// +stateify savable type SetXattrOptions struct { // Name is the name of the extended attribute being mutated. Name string @@ -221,6 +247,8 @@ type SetXattrOptions struct { // StatOptions contains options to VirtualFilesystem.StatAt(), // FilesystemImpl.StatAt(), FileDescription.Stat(), and // FileDescriptionImpl.Stat(). +// +// +stateify savable type StatOptions struct { // Mask is the set of fields in the returned Statx that the FilesystemImpl // or FileDescriptionImpl should provide. Bits are as in linux.Statx.Mask. @@ -238,6 +266,8 @@ type StatOptions struct { } // UmountOptions contains options to VirtualFilesystem.UmountAt(). +// +// +stateify savable type UmountOptions struct { // Flags contains flags as specified for umount2(2). Flags uint32 @@ -246,6 +276,8 @@ type UmountOptions struct { // WriteOptions contains options to FileDescription.PWrite(), // FileDescriptionImpl.PWrite(), FileDescription.Write(), and // FileDescriptionImpl.Write(). +// +// +stateify savable type WriteOptions struct { // Flags contains flags as specified for pwritev2(2). Flags uint32 diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index 00eeb8842..d48520d58 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -26,6 +26,8 @@ import ( ) // AccessTypes is a bitmask of Unix file permissions. +// +// +stateify savable type AccessTypes uint16 // Bits in AccessTypes. diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 3304372d9..e4fd55012 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -35,6 +35,8 @@ import ( // FilesystemImpl methods. // // ResolvingPath is loosely analogous to Linux's struct nameidata. +// +// +stateify savable type ResolvingPath struct { vfs *VirtualFilesystem root VirtualDentry // refs borrowed from PathOperation @@ -88,6 +90,7 @@ func init() { // so error "constants" are really mutable vars, necessitating somewhat // expensive interface object comparisons. +// +stateify savable type resolveMountRootOrJumpError struct{} // Error implements error.Error. @@ -95,6 +98,7 @@ func (resolveMountRootOrJumpError) Error() string { return "resolving mount root or jump" } +// +stateify savable type resolveMountPointError struct{} // Error implements error.Error. @@ -102,6 +106,7 @@ func (resolveMountPointError) Error() string { return "resolving mount point" } +// +stateify savable type resolveAbsSymlinkError struct{} // Error implements error.Error. diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 1ebf355ef..5bd756ea5 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -163,6 +163,8 @@ func (vfs *VirtualFilesystem) Init(ctx context.Context) error { // PathOperation is passed to VFS methods by pointer to reduce memory copying: // it's somewhat large and should never escape. (Options structs are passed by // pointer to VFS and FileDescription methods for the same reason.) +// +// +stateify savable type PathOperation struct { // Root is the VFS root. References on Root are borrowed from the provider // of the PathOperation. diff --git a/pkg/state/types.go b/pkg/state/types.go index 215ef80f8..84aed8732 100644 --- a/pkg/state/types.go +++ b/pkg/state/types.go @@ -107,6 +107,14 @@ func lookupNameFields(typ reflect.Type) (string, []string, bool) { } return name, nil, true } + // Sanity check the type. + if raceEnabled { + if _, ok := reverseTypeDatabase[typ]; !ok { + // The type was not registered? Must be an embedded + // structure or something else. + return "", nil, false + } + } // Extract the name from the object. name := t.StateTypeName() fields := t.StateFields() @@ -313,6 +321,9 @@ var primitiveTypeDatabase = func() map[string]reflect.Type { // globalTypeDatabase is used for dispatching interfaces on decode. var globalTypeDatabase = map[string]reflect.Type{} +// reverseTypeDatabase is a reverse mapping. +var reverseTypeDatabase = map[reflect.Type]string{} + // Register registers a type. // // This must be called on init and only done once. @@ -358,4 +369,7 @@ func Register(t Type) { Failf("conflicting name for %T: matches interfaceType", t) } globalTypeDatabase[name] = typ + if raceEnabled { + reverseTypeDatabase[typ] = name + } } diff --git a/pkg/tcpip/adapters/gonet/gonet_test.go b/pkg/tcpip/adapters/gonet/gonet_test.go index c975ad9cf..12b061def 100644 --- a/pkg/tcpip/adapters/gonet/gonet_test.go +++ b/pkg/tcpip/adapters/gonet/gonet_test.go @@ -61,8 +61,8 @@ func TestTimeouts(t *testing.T) { func newLoopbackStack() (*stack.Stack, *tcpip.Error) { // Create the stack and add a NIC. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol}, }) if err := s.CreateNIC(NICID, loopback.New()); err != nil { diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index b769094dc..19627fa9b 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -339,7 +339,7 @@ func NoChecksum(noChecksum bool) TransportChecker { udp, ok := h.(header.UDP) if !ok { - return + t.Fatalf("UDP header not found in h: %T", h) } if b := udp.Checksum() == 0; b != noChecksum { @@ -348,14 +348,14 @@ func NoChecksum(noChecksum bool) TransportChecker { } } -// SeqNum creates a checker that checks the sequence number. -func SeqNum(seq uint32) TransportChecker { +// TCPSeqNum creates a checker that checks the sequence number. +func TCPSeqNum(seq uint32) TransportChecker { return func(t *testing.T, h header.Transport) { t.Helper() tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if s := tcp.SequenceNumber(); s != seq { @@ -364,14 +364,14 @@ func SeqNum(seq uint32) TransportChecker { } } -// AckNum creates a checker that checks the ack number. -func AckNum(seq uint32) TransportChecker { +// TCPAckNum creates a checker that checks the ack number. +func TCPAckNum(seq uint32) TransportChecker { return func(t *testing.T, h header.Transport) { t.Helper() tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if s := tcp.AckNumber(); s != seq { @@ -380,18 +380,52 @@ func AckNum(seq uint32) TransportChecker { } } -// Window creates a checker that checks the tcp window. -func Window(window uint16) TransportChecker { +// TCPWindow creates a checker that checks the tcp window. +func TCPWindow(window uint16) TransportChecker { return func(t *testing.T, h header.Transport) { t.Helper() tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in hdr : %T", h) } if w := tcp.WindowSize(); w != window { - t.Errorf("Bad window, got 0x%x, want 0x%x", w, window) + t.Errorf("Bad window, got %d, want %d", w, window) + } + } +} + +// TCPWindowGreaterThanEq creates a checker that checks that the TCP window +// is greater than or equal to the provided value. +func TCPWindowGreaterThanEq(window uint16) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + tcp, ok := h.(header.TCP) + if !ok { + t.Fatalf("TCP header not found in h: %T", h) + } + + if w := tcp.WindowSize(); w < window { + t.Errorf("Bad window, got %d, want > %d", w, window) + } + } +} + +// TCPWindowLessThanEq creates a checker that checks that the tcp window +// is less than or equal to the provided value. +func TCPWindowLessThanEq(window uint16) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + tcp, ok := h.(header.TCP) + if !ok { + t.Fatalf("TCP header not found in h: %T", h) + } + + if w := tcp.WindowSize(); w > window { + t.Errorf("Bad window, got %d, want < %d", w, window) } } } @@ -403,7 +437,7 @@ func TCPFlags(flags uint8) TransportChecker { tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if f := tcp.Flags(); f != flags { @@ -420,7 +454,7 @@ func TCPFlagsMatch(flags, mask uint8) TransportChecker { tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if f := tcp.Flags(); (f & mask) != (flags & mask) { diff --git a/pkg/tcpip/faketime/faketime.go b/pkg/tcpip/faketime/faketime.go index 1193f1d7d..f7a4fbde1 100644 --- a/pkg/tcpip/faketime/faketime.go +++ b/pkg/tcpip/faketime/faketime.go @@ -24,6 +24,26 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" ) +// NullClock implements a clock that never advances. +type NullClock struct{} + +var _ tcpip.Clock = (*NullClock)(nil) + +// NowNanoseconds implements tcpip.Clock.NowNanoseconds. +func (*NullClock) NowNanoseconds() int64 { + return 0 +} + +// NowMonotonic implements tcpip.Clock.NowMonotonic. +func (*NullClock) NowMonotonic() int64 { + return 0 +} + +// AfterFunc implements tcpip.Clock.AfterFunc. +func (*NullClock) AfterFunc(time.Duration, func()) tcpip.Timer { + return nil +} + // ManualClock implements tcpip.Clock and only advances manually with Advance // method. type ManualClock struct { diff --git a/pkg/tcpip/network/BUILD b/pkg/tcpip/network/BUILD index 46083925c..376583f3c 100644 --- a/pkg/tcpip/network/BUILD +++ b/pkg/tcpip/network/BUILD @@ -9,6 +9,7 @@ go_test( "ip_test.go", ], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 81e286e80..b47a7be51 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -15,17 +15,11 @@ // Package arp implements the ARP network protocol. It is used to resolve // IPv4 addresses into link-local MAC addresses, and advertises IPv4 // addresses of its stack with the local network. -// -// To use it in the networking stack, pass arp.NewProtocol() as one of the -// network protocols when calling stack.New. Then add an "arp" address to every -// NIC on the stack that should respond to ARP requests. That is: -// -// if err := s.AddAddress(1, arp.ProtocolNumber, "arp"); err != nil { -// // handle err -// } package arp import ( + "sync/atomic" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -41,15 +35,57 @@ const ( ProtocolAddress = tcpip.Address("arp") ) -// endpoint implements stack.NetworkEndpoint. +var _ stack.AddressableEndpoint = (*endpoint)(nil) +var _ stack.NetworkEndpoint = (*endpoint)(nil) + type endpoint struct { - protocol *protocol - nicID tcpip.NICID + stack.AddressableEndpointState + + protocol *protocol + + // enabled is set to 1 when the NIC is enabled and 0 when it is disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + + nic stack.NetworkInterface linkEP stack.LinkEndpoint linkAddrCache stack.LinkAddressCache nud stack.NUDHandler } +func (e *endpoint) Enable() *tcpip.Error { + if !e.nic.Enabled() { + return tcpip.ErrNotPermitted + } + + e.setEnabled(true) + return nil +} + +func (e *endpoint) Enabled() bool { + return e.nic.Enabled() && e.isEnabled() +} + +// isEnabled returns true if the endpoint is enabled, regardless of the +// enabled status of the NIC. +func (e *endpoint) isEnabled() bool { + return atomic.LoadUint32(&e.enabled) == 1 +} + +// setEnabled sets the enabled status for the endpoint. +func (e *endpoint) setEnabled(v bool) { + if v { + atomic.StoreUint32(&e.enabled, 1) + } else { + atomic.StoreUint32(&e.enabled, 0) + } +} + +func (e *endpoint) Disable() { + e.setEnabled(false) +} + // DefaultTTL is unused for ARP. It implements stack.NetworkEndpoint. func (e *endpoint) DefaultTTL() uint8 { return 0 @@ -60,19 +96,13 @@ func (e *endpoint) MTU() uint32 { return lmtu - uint32(e.MaxHeaderLength()) } -func (e *endpoint) NICID() tcpip.NICID { - return e.nicID -} - -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.linkEP.Capabilities() -} - func (e *endpoint) MaxHeaderLength() uint16 { return e.linkEP.MaxHeaderLength() + header.ARPSize } -func (e *endpoint) Close() {} +func (e *endpoint) Close() { + e.AddressableEndpointState.Cleanup() +} func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, *stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported @@ -93,6 +123,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu } func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { + if !e.isEnabled() { + return + } + h := header.ARP(pkt.NetworkHeader().View()) if !h.IsValid() { return @@ -103,15 +137,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { localAddr := tcpip.Address(h.ProtocolAddressTarget()) if e.nud == nil { - if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { + if e.linkAddrCache.CheckLocalAddress(e.nic.ID(), header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } addr := tcpip.Address(h.ProtocolAddressSender()) linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) - e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), addr, linkAddr) } else { - if r.Stack().CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { + if r.Stack().CheckLocalAddress(e.nic.ID(), header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } @@ -137,7 +171,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) if e.nud == nil { - e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), addr, linkAddr) return } @@ -169,14 +203,16 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { return tcpip.Address(h.ProtocolAddressSender()), ProtocolAddress } -func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { - return &endpoint{ +func (p *protocol) NewEndpoint(nic stack.NetworkInterface, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &endpoint{ protocol: p, - nicID: nicID, - linkEP: sender, + nic: nic, + linkEP: nic.LinkEndpoint(), linkAddrCache: linkAddrCache, nud: nud, } + e.AddressableEndpointState.Init(e) + return e } // LinkAddressProtocol implements stack.LinkAddressResolver.LinkAddressProtocol. @@ -238,13 +274,11 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu return 0, false, parse.ARP(pkt) } -// ReturnError implements stack.TransportProtocol.ReturnError. -func (*protocol) ReturnError(*stack.Route, tcpip.ICMPReason, *stack.PacketBuffer) *tcpip.Error { - // In ARP, there is no such response so do nothing. - return nil -} - // NewProtocol returns an ARP network protocol. -func NewProtocol() stack.NetworkProtocol { +// +// Note, to make sure that the ARP endpoint receives ARP packets, the "arp" +// address must be added to every NIC that should respond to ARP requests. See +// ProtocolAddress for more details. +func NewProtocol(*stack.Stack) stack.NetworkProtocol { return &protocol{} } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 9c9a859e3..626af975a 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -176,8 +176,8 @@ func newTestContext(t *testing.T, useNeighborCache bool) *testContext { } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, arp.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol4}, NUDConfigs: c, NUDDisp: &d, UseNeighborCache: useNeighborCache, @@ -442,7 +442,7 @@ func TestLinkAddressRequest(t *testing.T) { } for _, test := range tests { - p := arp.NewProtocol() + p := arp.NewProtocol(nil) linkRes, ok := p.(stack.LinkAddressResolver) if !ok { t.Fatal("expected ARP protocol to implement stack.LinkAddressResolver") diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index 96c5f42f8..e247f06a4 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -43,5 +43,6 @@ go_test( library = ":fragmentation", deps = [ "//pkg/tcpip/buffer", + "//pkg/tcpip/faketime", ], ) diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 6a4843f92..e1909fab0 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -81,6 +81,8 @@ type Fragmentation struct { size int timeout time.Duration blockSize uint16 + clock tcpip.Clock + releaseJob *tcpip.Job } // NewFragmentation creates a new Fragmentation. @@ -97,7 +99,7 @@ type Fragmentation struct { // reassemblingTimeout specifies the maximum time allowed to reassemble a packet. // Fragments are lazily evicted only when a new a packet with an // already existing fragmentation-id arrives after the timeout. -func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation { +func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration, clock tcpip.Clock) *Fragmentation { if lowMemoryLimit >= highMemoryLimit { lowMemoryLimit = highMemoryLimit } @@ -110,13 +112,17 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea blockSize = minBlockSize } - return &Fragmentation{ + f := &Fragmentation{ reassemblers: make(map[FragmentID]*reassembler), highLimit: highMemoryLimit, lowLimit: lowMemoryLimit, timeout: reassemblingTimeout, blockSize: blockSize, + clock: clock, } + f.releaseJob = tcpip.NewJob(f.clock, &f.mu, f.releaseReassemblersLocked) + + return f } // Process processes an incoming fragment belonging to an ID and returns a @@ -155,15 +161,17 @@ func (f *Fragmentation) Process( f.mu.Lock() r, ok := f.reassemblers[id] - if ok && r.tooOld(f.timeout) { - // This is very likely to be an id-collision or someone performing a slow-rate attack. - f.release(r) - ok = false - } if !ok { - r = newReassembler(id) + r = newReassembler(id, f.clock) f.reassemblers[id] = r + wasEmpty := f.rList.Empty() f.rList.PushFront(r) + if wasEmpty { + // If we have just pushed a first reassembler into an empty list, we + // should kickstart the release job. The release job will keep + // rescheduling itself until the list becomes empty. + f.releaseReassemblersLocked() + } } f.mu.Unlock() @@ -211,3 +219,27 @@ func (f *Fragmentation) release(r *reassembler) { f.size = 0 } } + +// releaseReassemblersLocked releases already-expired reassemblers, then +// schedules the job to call back itself for the remaining reassemblers if +// any. This function must be called with f.mu locked. +func (f *Fragmentation) releaseReassemblersLocked() { + now := f.clock.NowMonotonic() + for { + // The reassembler at the end of the list is the oldest. + r := f.rList.Back() + if r == nil { + // The list is empty. + break + } + elapsed := time.Duration(now-r.creationTime) * time.Nanosecond + if f.timeout > elapsed { + // If the oldest reassembler has not expired, schedule the release + // job so that this function is called back when it has expired. + f.releaseJob.Schedule(f.timeout - elapsed) + break + } + // If the oldest reassembler has already expired, release it. + f.release(r) + } +} diff --git a/pkg/tcpip/network/fragmentation/fragmentation_test.go b/pkg/tcpip/network/fragmentation/fragmentation_test.go index 416604659..189b223c5 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation_test.go +++ b/pkg/tcpip/network/fragmentation/fragmentation_test.go @@ -21,6 +21,7 @@ import ( "time" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/faketime" ) // vv is a helper to build VectorisedView from different strings. @@ -95,7 +96,7 @@ var processTestCases = []struct { func TestFragmentationProcess(t *testing.T) { for _, c := range processTestCases { t.Run(c.comment, func(t *testing.T) { - f := NewFragmentation(minBlockSize, 1024, 512, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 1024, 512, DefaultReassembleTimeout, &faketime.NullClock{}) firstFragmentProto := c.in[0].proto for i, in := range c.in { vv, proto, done, err := f.Process(in.id, in.first, in.last, in.more, in.proto, in.vv) @@ -131,25 +132,126 @@ func TestFragmentationProcess(t *testing.T) { } func TestReassemblingTimeout(t *testing.T) { - timeout := time.Millisecond - f := NewFragmentation(minBlockSize, 1024, 512, timeout) - // Send first fragment with id = 0, first = 0, last = 0, and more = true. - f.Process(FragmentID{}, 0, 0, true, 0xFF, vv(1, "0")) - // Sleep more than the timeout. - time.Sleep(2 * timeout) - // Send another fragment that completes a packet. - // However, no packet should be reassembled because the fragment arrived after the timeout. - _, _, done, err := f.Process(FragmentID{}, 1, 1, false, 0xFF, vv(1, "1")) - if err != nil { - t.Fatalf("f.Process(0, 1, 1, false, 0xFF, vv(1, \"1\")) failed: %v", err) + const ( + reassemblyTimeout = time.Millisecond + protocol = 0xff + ) + + type fragment struct { + first uint16 + last uint16 + more bool + data string } - if done { - t.Errorf("Fragmentation does not respect the reassembling timeout.") + + type event struct { + // name is a nickname of this event. + name string + + // clockAdvance is a duration to advance the clock. The clock advances + // before a fragment specified in the fragment field is processed. + clockAdvance time.Duration + + // fragment is a fragment to process. This can be nil if there is no + // fragment to process. + fragment *fragment + + // expectDone is true if the fragmentation instance should report the + // reassembly is done after the fragment is processd. + expectDone bool + + // sizeAfterEvent is the expected size of the fragmentation instance after + // the event. + sizeAfterEvent int + } + + half1 := &fragment{first: 0, last: 0, more: true, data: "0"} + half2 := &fragment{first: 1, last: 1, more: false, data: "1"} + + tests := []struct { + name string + events []event + }{ + { + name: "half1 and half2 are reassembled successfully", + events: []event{ + { + name: "half1", + fragment: half1, + expectDone: false, + sizeAfterEvent: 1, + }, + { + name: "half2", + fragment: half2, + expectDone: true, + sizeAfterEvent: 0, + }, + }, + }, + { + name: "half1 timeout, half2 timeout", + events: []event{ + { + name: "half1", + fragment: half1, + expectDone: false, + sizeAfterEvent: 1, + }, + { + name: "half1 just before reassembly timeout", + clockAdvance: reassemblyTimeout - 1, + sizeAfterEvent: 1, + }, + { + name: "half1 reassembly timeout", + clockAdvance: 1, + sizeAfterEvent: 0, + }, + { + name: "half2", + fragment: half2, + expectDone: false, + sizeAfterEvent: 1, + }, + { + name: "half2 just before reassembly timeout", + clockAdvance: reassemblyTimeout - 1, + sizeAfterEvent: 1, + }, + { + name: "half2 reassembly timeout", + clockAdvance: 1, + sizeAfterEvent: 0, + }, + }, + }, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + clock := faketime.NewManualClock() + f := NewFragmentation(minBlockSize, HighFragThreshold, LowFragThreshold, reassemblyTimeout, clock) + for _, event := range test.events { + clock.Advance(event.clockAdvance) + if frag := event.fragment; frag != nil { + _, _, done, err := f.Process(FragmentID{}, frag.first, frag.last, frag.more, protocol, vv(len(frag.data), frag.data)) + if err != nil { + t.Fatalf("%s: f.Process failed: %s", event.name, err) + } + if done != event.expectDone { + t.Fatalf("%s: got done = %t, want = %t", event.name, done, event.expectDone) + } + } + if got, want := f.size, event.sizeAfterEvent; got != want { + t.Errorf("%s: got f.size = %d, want = %d", event.name, got, want) + } + } + }) } } func TestMemoryLimits(t *testing.T) { - f := NewFragmentation(minBlockSize, 3, 1, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 3, 1, DefaultReassembleTimeout, &faketime.NullClock{}) // Send first fragment with id = 0. f.Process(FragmentID{ID: 0}, 0, 0, true, 0xFF, vv(1, "0")) // Send first fragment with id = 1. @@ -173,7 +275,7 @@ func TestMemoryLimits(t *testing.T) { } func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { - f := NewFragmentation(minBlockSize, 1, 0, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 1, 0, DefaultReassembleTimeout, &faketime.NullClock{}) // Send first fragment with id = 0. f.Process(FragmentID{}, 0, 0, true, 0xFF, vv(1, "0")) // Send the same packet again. @@ -268,7 +370,7 @@ func TestErrors(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, DefaultReassembleTimeout) + f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, DefaultReassembleTimeout, &faketime.NullClock{}) _, _, done, err := f.Process(FragmentID{}, test.first, test.last, test.more, 0, vv(len(test.data), test.data)) if !errors.Is(err, test.err) { t.Errorf("got Process(_, %d, %d, %t, _, %q) = (_, _, _, %v), want = (_, _, _, %v)", test.first, test.last, test.more, test.data, err, test.err) diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index f044867dc..9bb051a30 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -18,9 +18,9 @@ import ( "container/heap" "fmt" "math" - "time" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) @@ -40,15 +40,15 @@ type reassembler struct { deleted int heap fragHeap done bool - creationTime time.Time + creationTime int64 } -func newReassembler(id FragmentID) *reassembler { +func newReassembler(id FragmentID, clock tcpip.Clock) *reassembler { r := &reassembler{ id: id, holes: make([]hole, 0, 16), heap: make(fragHeap, 0, 8), - creationTime: time.Now(), + creationTime: clock.NowMonotonic(), } r.holes = append(r.holes, hole{ first: 0, @@ -116,10 +116,6 @@ func (r *reassembler) process(first, last uint16, more bool, proto uint8, vv buf return res, r.proto, true, consumed, nil } -func (r *reassembler) tooOld(timeout time.Duration) bool { - return time.Now().Sub(r.creationTime) > timeout -} - func (r *reassembler) checkDoneOrMark() bool { r.mu.Lock() prev := r.done diff --git a/pkg/tcpip/network/fragmentation/reassembler_test.go b/pkg/tcpip/network/fragmentation/reassembler_test.go index dff7c9dcb..a0a04a027 100644 --- a/pkg/tcpip/network/fragmentation/reassembler_test.go +++ b/pkg/tcpip/network/fragmentation/reassembler_test.go @@ -18,6 +18,8 @@ import ( "math" "reflect" "testing" + + "gvisor.dev/gvisor/pkg/tcpip/faketime" ) type updateHolesInput struct { @@ -94,7 +96,7 @@ var holesTestCases = []struct { func TestUpdateHoles(t *testing.T) { for _, c := range holesTestCases { - r := newReassembler(FragmentID{}) + r := newReassembler(FragmentID{}, &faketime.NullClock{}) for _, i := range c.in { r.updateHoles(i.first, i.last, i.more) } diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index e45dd17f8..56a56362e 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -17,6 +17,7 @@ package ip_test import ( "testing" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -98,9 +99,10 @@ func (t *testObject) checkValues(protocol tcpip.TransportProtocolNumber, vv buff // DeliverTransportPacket is called by network endpoints after parsing incoming // packets. This is used by the test object to verify that the results of the // parsing are expected. -func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt *stack.PacketBuffer) { +func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt *stack.PacketBuffer) stack.TransportPacketDisposition { t.checkValues(protocol, pkt.Data, r.RemoteAddress, r.LocalAddress) t.dataCalls++ + return stack.TransportPacketHandled } // DeliverTransportControlPacket is called by network endpoints after parsing @@ -194,8 +196,8 @@ func (*testObject) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, }) s.CreateNIC(nicID, loopback.New()) s.AddAddress(nicID, ipv4.ProtocolNumber, local) @@ -210,8 +212,8 @@ func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { func buildIPv6Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, }) s.CreateNIC(nicID, loopback.New()) s.AddAddress(nicID, ipv6.ProtocolNumber, local) @@ -228,8 +230,8 @@ func buildDummyStack(t *testing.T) *stack.Stack { t.Helper() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, }) e := channel.New(0, 1280, "") if err := s.CreateNIC(nicID, e); err != nil { @@ -247,10 +249,137 @@ func buildDummyStack(t *testing.T) *stack.Stack { return s } +var _ stack.NetworkInterface = (*testInterface)(nil) + +type testInterface struct { + tester testObject + + mu struct { + sync.RWMutex + disabled bool + } +} + +func (*testInterface) ID() tcpip.NICID { + return nicID +} + +func (*testInterface) IsLoopback() bool { + return false +} + +func (*testInterface) Name() string { + return "" +} + +func (t *testInterface) Enabled() bool { + t.mu.RLock() + defer t.mu.RUnlock() + return !t.mu.disabled +} + +func (t *testInterface) setEnabled(v bool) { + t.mu.Lock() + defer t.mu.Unlock() + t.mu.disabled = !v +} + +func (t *testInterface) LinkEndpoint() stack.LinkEndpoint { + return &t.tester +} + +func TestEnableWhenNICDisabled(t *testing.T) { + tests := []struct { + name string + protocolFactory stack.NetworkProtocolFactory + protoNum tcpip.NetworkProtocolNumber + }{ + { + name: "IPv4", + protocolFactory: ipv4.NewProtocol, + protoNum: ipv4.ProtocolNumber, + }, + { + name: "IPv6", + protocolFactory: ipv6.NewProtocol, + protoNum: ipv6.ProtocolNumber, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var nic testInterface + nic.setEnabled(false) + + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{test.protocolFactory}, + }) + p := s.NetworkProtocolInstance(test.protoNum) + + // We pass nil for all parameters except the NetworkInterface and Stack + // since Enable only depends on these. + ep := p.NewEndpoint(&nic, nil, nil, nil) + + // The endpoint should initially be disabled, regardless the NIC's enabled + // status. + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + nic.setEnabled(true) + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + + // Attempting to enable the endpoint while the NIC is disabled should + // fail. + nic.setEnabled(false) + if err := ep.Enable(); err != tcpip.ErrNotPermitted { + t.Fatalf("got ep.Enable() = %s, want = %s", err, tcpip.ErrNotPermitted) + } + // ep should consider the NIC's enabled status when determining its own + // enabled status so we "enable" the NIC to read just the endpoint's + // enabled status. + nic.setEnabled(true) + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + + // Enabling the interface after the NIC has been enabled should succeed. + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + if !ep.Enabled() { + t.Fatal("got ep.Enabled() = false, want = true") + } + + // ep should consider the NIC's enabled status when determining its own + // enabled status. + nic.setEnabled(false) + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + + // Disabling the endpoint when the NIC is enabled should make the endpoint + // disabled. + nic.setEnabled(true) + ep.Disable() + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + }) + } +} + func TestIPv4Send(t *testing.T) { - o := testObject{t: t, v4: true} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, nil, &o, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + v4: true, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, nil) defer ep.Close() // Allocate and initialize the payload view. @@ -266,10 +395,10 @@ func TestIPv4Send(t *testing.T) { }) // Issue the write. - o.protocol = 123 - o.srcAddr = localIpv4Addr - o.dstAddr = remoteIpv4Addr - o.contents = payload + nic.tester.protocol = 123 + nic.tester.srcAddr = localIpv4Addr + nic.tester.dstAddr = remoteIpv4Addr + nic.tester.contents = payload r, err := buildIPv4Route(localIpv4Addr, remoteIpv4Addr) if err != nil { @@ -285,11 +414,21 @@ func TestIPv4Send(t *testing.T) { } func TestIPv4Receive(t *testing.T) { - o := testObject{t: t, v4: true} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + v4: true, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.tester) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + totalLen := header.IPv4MinimumSize + 30 view := buffer.NewView(totalLen) ip := header.IPv4(view) @@ -308,10 +447,10 @@ func TestIPv4Receive(t *testing.T) { } // Give packet to ipv4 endpoint, dispatcher will validate that it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv4Addr - o.dstAddr = localIpv4Addr - o.contents = view[header.IPv4MinimumSize:totalLen] + nic.tester.protocol = 10 + nic.tester.srcAddr = remoteIpv4Addr + nic.tester.dstAddr = localIpv4Addr + nic.tester.contents = view[header.IPv4MinimumSize:totalLen] r, err := buildIPv4Route(localIpv4Addr, remoteIpv4Addr) if err != nil { @@ -324,8 +463,8 @@ func TestIPv4Receive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 1 { - t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) + if nic.tester.dataCalls != 1 { + t.Fatalf("Bad number of data calls: got %x, want 1", nic.tester.dataCalls) } } @@ -355,11 +494,20 @@ func TestIPv4ReceiveControl(t *testing.T) { } for _, c := range cases { t.Run(c.name, func(t *testing.T) { - o := testObject{t: t} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.tester) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + const dataOffset = header.IPv4MinimumSize*2 + header.ICMPv4MinimumSize view := buffer.NewView(dataOffset + 8) @@ -400,27 +548,37 @@ func TestIPv4ReceiveControl(t *testing.T) { // Give packet to IPv4 endpoint, dispatcher will validate that // it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv4Addr - o.dstAddr = localIpv4Addr - o.contents = view[dataOffset:] - o.typ = c.expectedTyp - o.extra = c.expectedExtra + nic.tester.protocol = 10 + nic.tester.srcAddr = remoteIpv4Addr + nic.tester.dstAddr = localIpv4Addr + nic.tester.contents = view[dataOffset:] + nic.tester.typ = c.expectedTyp + nic.tester.extra = c.expectedExtra ep.HandlePacket(&r, truncatedPacket(view, c.trunc, header.IPv4MinimumSize)) - if want := c.expectedCount; o.controlCalls != want { - t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) + if want := c.expectedCount; nic.tester.controlCalls != want { + t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, nic.tester.controlCalls, want) } }) } } func TestIPv4FragmentationReceive(t *testing.T) { - o := testObject{t: t, v4: true} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + v4: true, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.tester) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + totalLen := header.IPv4MinimumSize + 24 frag1 := buffer.NewView(totalLen) @@ -457,10 +615,10 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Give packet to ipv4 endpoint, dispatcher will validate that it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv4Addr - o.dstAddr = localIpv4Addr - o.contents = append(frag1[header.IPv4MinimumSize:totalLen], frag2[header.IPv4MinimumSize:totalLen]...) + nic.tester.protocol = 10 + nic.tester.srcAddr = remoteIpv4Addr + nic.tester.dstAddr = localIpv4Addr + nic.tester.contents = append(frag1[header.IPv4MinimumSize:totalLen], frag2[header.IPv4MinimumSize:totalLen]...) r, err := buildIPv4Route(localIpv4Addr, remoteIpv4Addr) if err != nil { @@ -475,8 +633,8 @@ func TestIPv4FragmentationReceive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 0 { - t.Fatalf("Bad number of data calls: got %x, want 0", o.dataCalls) + if nic.tester.dataCalls != 0 { + t.Fatalf("Bad number of data calls: got %x, want 0", nic.tester.dataCalls) } // Send second segment. @@ -487,17 +645,26 @@ func TestIPv4FragmentationReceive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 1 { - t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) + if nic.tester.dataCalls != 1 { + t.Fatalf("Bad number of data calls: got %x, want 1", nic.tester.dataCalls) } } func TestIPv6Send(t *testing.T) { - o := testObject{t: t} - proto := ipv6.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, channel.New(0, 1280, ""), buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv6.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, nil) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + // Allocate and initialize the payload view. payload := buffer.NewView(100) for i := 0; i < len(payload); i++ { @@ -511,10 +678,10 @@ func TestIPv6Send(t *testing.T) { }) // Issue the write. - o.protocol = 123 - o.srcAddr = localIpv6Addr - o.dstAddr = remoteIpv6Addr - o.contents = payload + nic.tester.protocol = 123 + nic.tester.srcAddr = localIpv6Addr + nic.tester.dstAddr = remoteIpv6Addr + nic.tester.contents = payload r, err := buildIPv6Route(localIpv6Addr, remoteIpv6Addr) if err != nil { @@ -530,11 +697,20 @@ func TestIPv6Send(t *testing.T) { } func TestIPv6Receive(t *testing.T) { - o := testObject{t: t} - proto := ipv6.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv6.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.tester) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + totalLen := header.IPv6MinimumSize + 30 view := buffer.NewView(totalLen) ip := header.IPv6(view) @@ -552,10 +728,10 @@ func TestIPv6Receive(t *testing.T) { } // Give packet to ipv6 endpoint, dispatcher will validate that it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv6Addr - o.dstAddr = localIpv6Addr - o.contents = view[header.IPv6MinimumSize:totalLen] + nic.tester.protocol = 10 + nic.tester.srcAddr = remoteIpv6Addr + nic.tester.dstAddr = localIpv6Addr + nic.tester.contents = view[header.IPv6MinimumSize:totalLen] r, err := buildIPv6Route(localIpv6Addr, remoteIpv6Addr) if err != nil { @@ -569,8 +745,8 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 1 { - t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) + if nic.tester.dataCalls != 1 { + t.Fatalf("Bad number of data calls: got %x, want 1", nic.tester.dataCalls) } } @@ -609,11 +785,20 @@ func TestIPv6ReceiveControl(t *testing.T) { } for _, c := range cases { t.Run(c.name, func(t *testing.T) { - o := testObject{t: t} - proto := ipv6.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv6.ProtocolNumber) + nic := testInterface{ + tester: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.tester) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + dataOffset := header.IPv6MinimumSize*2 + header.ICMPv6MinimumSize if c.fragmentOffset != nil { dataOffset += header.IPv6FragmentHeaderSize @@ -666,19 +851,19 @@ func TestIPv6ReceiveControl(t *testing.T) { // Give packet to IPv6 endpoint, dispatcher will validate that // it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv6Addr - o.dstAddr = localIpv6Addr - o.contents = view[dataOffset:] - o.typ = c.expectedTyp - o.extra = c.expectedExtra + nic.tester.protocol = 10 + nic.tester.srcAddr = remoteIpv6Addr + nic.tester.dstAddr = localIpv6Addr + nic.tester.contents = view[dataOffset:] + nic.tester.typ = c.expectedTyp + nic.tester.extra = c.expectedExtra // Set ICMPv6 checksum. icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) ep.HandlePacket(&r, truncatedPacket(view, c.trunc, header.IPv6MinimumSize)) - if want := c.expectedCount; o.controlCalls != want { - t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) + if want := c.expectedCount; nic.tester.controlCalls != want { + t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, nic.tester.controlCalls, want) } }) } diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index f9c2aa980..0a7e98ed1 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -10,6 +10,7 @@ go_library( ], visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 5fe73315f..5c4f715d7 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -40,7 +40,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match an address we own. src := hdr.SourceAddress() - if e.stack.CheckLocalAddress(e.NICID(), ProtocolNumber, src) == 0 { + if e.protocol.stack.CheckLocalAddress(e.nic.ID(), ProtocolNumber, src) == 0 { return } @@ -110,7 +110,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { localAddr = "" } - r, err := r.Stack().FindRoute(e.NICID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) + r, err := r.Stack().FindRoute(e.nic.ID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) if err != nil { // If we cannot find a route to the destination, silently drop the packet. return @@ -200,16 +200,6 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { // ======= ICMP Error packet generation ========= -// ReturnError implements stack.TransportProtocol.ReturnError. -func (p *protocol) ReturnError(r *stack.Route, reason tcpip.ICMPReason, pkt *stack.PacketBuffer) *tcpip.Error { - switch reason.(type) { - case *tcpip.ICMPReasonPortUnreachable: - return returnError(r, &icmpReasonPortUnreachable{}, pkt) - default: - return tcpip.ErrNotSupported - } -} - // icmpReason is a marker interface for IPv4 specific ICMP errors. type icmpReason interface { isICMPReason() diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 135444222..cf0a76274 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -12,17 +12,14 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package ipv4 contains the implementation of the ipv4 network protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv4.NewProtocol() as one of the network -// protocols when calling stack.New(). Then endpoints can be created by passing -// ipv4.ProtocolNumber as the network protocol number when calling -// Stack.NewEndpoint(). +// Package ipv4 contains the implementation of the ipv4 network protocol. package ipv4 import ( + "fmt" "sync/atomic" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -51,22 +48,115 @@ const ( fragmentblockSize = 8 ) +var ipv4BroadcastAddr = header.IPv4Broadcast.WithPrefix() + +var _ stack.GroupAddressableEndpoint = (*endpoint)(nil) +var _ stack.AddressableEndpoint = (*endpoint)(nil) +var _ stack.NetworkEndpoint = (*endpoint)(nil) + type endpoint struct { - nicID tcpip.NICID + nic stack.NetworkInterface linkEP stack.LinkEndpoint dispatcher stack.TransportDispatcher protocol *protocol - stack *stack.Stack + + // enabled is set to 1 when the enpoint is enabled and 0 when it is + // disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + + mu struct { + sync.RWMutex + + addressableEndpointState stack.AddressableEndpointState + } } // NewEndpoint creates a new ipv4 endpoint. -func (p *protocol) NewEndpoint(nicID tcpip.NICID, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { - return &endpoint{ - nicID: nicID, - linkEP: linkEP, +func (p *protocol) NewEndpoint(nic stack.NetworkInterface, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &endpoint{ + nic: nic, + linkEP: nic.LinkEndpoint(), dispatcher: dispatcher, protocol: p, - stack: st, + } + e.mu.addressableEndpointState.Init(e) + return e +} + +// Enable implements stack.NetworkEndpoint. +func (e *endpoint) Enable() *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + // If the NIC is not enabled, the endpoint can't do anything meaningful so + // don't enable the endpoint. + if !e.nic.Enabled() { + return tcpip.ErrNotPermitted + } + + // If the endpoint is already enabled, there is nothing for it to do. + if !e.setEnabled(true) { + return nil + } + + // Create an endpoint to receive broadcast packets on this interface. + ep, err := e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(ipv4BroadcastAddr, stack.NeverPrimaryEndpoint, stack.AddressConfigStatic, false /* deprecated */) + if err != nil { + return err + } + // We have no need for the address endpoint. + ep.DecRef() + + // As per RFC 1122 section 3.3.7, all hosts should join the all-hosts + // multicast group. Note, the IANA calls the all-hosts multicast group the + // all-systems multicast group. + _, err = e.mu.addressableEndpointState.JoinGroup(header.IPv4AllSystems) + return err +} + +// Enabled implements stack.NetworkEndpoint. +func (e *endpoint) Enabled() bool { + return e.nic.Enabled() && e.isEnabled() +} + +// isEnabled returns true if the endpoint is enabled, regardless of the +// enabled status of the NIC. +func (e *endpoint) isEnabled() bool { + return atomic.LoadUint32(&e.enabled) == 1 +} + +// setEnabled sets the enabled status for the endpoint. +// +// Returns true if the enabled status was updated. +func (e *endpoint) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&e.enabled, 1) == 0 + } + return atomic.SwapUint32(&e.enabled, 0) == 1 +} + +// Disable implements stack.NetworkEndpoint. +func (e *endpoint) Disable() { + e.mu.Lock() + defer e.mu.Unlock() + e.disableLocked() +} + +func (e *endpoint) disableLocked() { + if !e.setEnabled(false) { + return + } + + // The endpoint may have already left the multicast group. + if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv4AllSystems); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv4AllSystems, err)) + } + + // The address may have already been removed. + if err := e.mu.addressableEndpointState.RemovePermanentAddress(ipv4BroadcastAddr.Address); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error when removing address = %s: %s", ipv4BroadcastAddr.Address, err)) } } @@ -81,16 +171,6 @@ func (e *endpoint) MTU() uint32 { return calculateMTU(e.linkEP.MTU()) } -// Capabilities implements stack.NetworkEndpoint.Capabilities. -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.linkEP.Capabilities() -} - -// NICID returns the ID of the NIC this endpoint belongs to. -func (e *endpoint) NICID() tcpip.NICID { - return e.nicID -} - // MaxHeaderLength returns the maximum length needed by ipv4 headers (and // underlying protocols). func (e *endpoint) MaxHeaderLength() uint16 { @@ -232,8 +312,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // iptables filtering. All packets that reach here are locally // generated. - nicName := e.stack.FindNICNameFromID(e.NICID()) - ipt := e.stack.IPTables() + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Output, pkt, gso, r, "", nicName); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesOutputDropped.Increment() @@ -249,7 +329,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // short circuits broadcasts before they are sent out to other hosts. if pkt.NatDone { netHeader := header.IPv4(pkt.NetworkHeader().View()) - ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) + ep, err := e.protocol.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) if err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) ep.HandlePacket(&route, pkt) @@ -289,10 +369,10 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe pkt = pkt.Next() } - nicName := e.stack.FindNICNameFromID(e.NICID()) + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) // iptables filtering. All packets that reach here are locally // generated. - ipt := e.stack.IPTables() + ipt := e.protocol.stack.IPTables() dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r, nicName) if len(dropped) == 0 && len(natPkts) == 0 { // Fast path: If no packets are to be dropped then we can just invoke the @@ -312,7 +392,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } if _, ok := natPkts[pkt]; ok { netHeader := header.IPv4(pkt.NetworkHeader().View()) - if ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil { + if ep, err := e.protocol.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() route := r.ReverseRoute(src, dst) @@ -389,6 +469,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { + if !e.isEnabled() { + return + } + h := header.IPv4(pkt.NetworkHeader().View()) if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() @@ -397,7 +481,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. - ipt := e.stack.IPTables() + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Input, pkt, nil, nil, "", ""); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesInputDropped.Increment() @@ -463,21 +547,141 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { return } r.Stats().IP.PacketsDelivered.Increment() - e.dispatcher.DeliverTransportPacket(r, p, pkt) + + switch res := e.dispatcher.DeliverTransportPacket(r, p, pkt); res { + case stack.TransportPacketHandled: + case stack.TransportPacketDestinationPortUnreachable: + // As per RFC: 1122 Section 3.2.2.1 A host SHOULD generate Destination + // Unreachable messages with code: + // 3 (Port Unreachable), when the designated transport protocol + // (e.g., UDP) is unable to demultiplex the datagram but has no + // protocol mechanism to inform the sender. + _ = returnError(r, &icmpReasonPortUnreachable{}, pkt) + default: + panic(fmt.Sprintf("unrecognized result from DeliverTransportPacket = %d", res)) + } } // Close cleans up resources associated with the endpoint. -func (e *endpoint) Close() {} +func (e *endpoint) Close() { + e.mu.Lock() + defer e.mu.Unlock() + + e.disableLocked() + e.mu.addressableEndpointState.Cleanup() +} + +// AddAndAcquirePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(addr, peb, configType, deprecated) +} + +// RemovePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) RemovePermanentAddress(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.RemovePermanentAddress(addr) +} + +// AcquireAssignedAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB stack.PrimaryEndpointBehavior) stack.AddressEndpoint { + e.mu.Lock() + defer e.mu.Unlock() + + loopback := e.nic.IsLoopback() + addressEndpoint := e.mu.addressableEndpointState.ReadOnly().AddrOrMatching(localAddr, allowTemp, func(addressEndpoint stack.AddressEndpoint) bool { + subnet := addressEndpoint.AddressWithPrefix().Subnet() + // IPv4 has a notion of a subnet broadcast address and considers the + // loopback interface bound to an address's whole subnet (on linux). + return subnet.IsBroadcast(localAddr) || (loopback && subnet.Contains(localAddr)) + }) + if addressEndpoint != nil { + return addressEndpoint + } + + if !allowTemp { + return nil + } + + addr := localAddr.WithPrefix() + addressEndpoint, err := e.mu.addressableEndpointState.AddAndAcquireTemporaryAddress(addr, tempPEB) + if err != nil { + // AddAddress only returns an error if the address is already assigned, + // but we just checked above if the address exists so we expect no error. + panic(fmt.Sprintf("e.mu.addressableEndpointState.AddAndAcquireTemporaryAddress(%s, %d): %s", addr, tempPEB, err)) + } + return addressEndpoint +} + +// AcquirePrimaryAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquirePrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.AcquirePrimaryAddress(remoteAddr, allowExpired) +} + +// PrimaryAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PrimaryAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PrimaryAddresses() +} + +// PermanentAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PermanentAddresses() +} + +// JoinGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) { + if !header.IsV4MulticastAddress(addr) { + return false, tcpip.ErrBadAddress + } + + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.JoinGroup(addr) +} + +// LeaveGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.LeaveGroup(addr) +} + +// IsInGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) IsInGroup(addr tcpip.Address) bool { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.IsInGroup(addr) +} + +var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) +var _ stack.NetworkProtocol = (*protocol)(nil) type protocol struct { - ids []uint32 - hashIV uint32 + stack *stack.Stack // defaultTTL is the current default TTL for the protocol. Only the - // uint8 portion of it is meaningful and it must be accessed - // atomically. + // uint8 portion of it is meaningful. + // + // Must be accessed using atomic operations. defaultTTL uint32 + // forwarding is set to 1 when the protocol has forwarding enabled and 0 + // when it is disabled. + // + // Must be accessed using atomic operations. + forwarding uint32 + + ids []uint32 + hashIV uint32 + fragmentation *fragmentation.Fragmentation } @@ -550,6 +754,20 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu return ipHdr.TransportProtocol(), !ipHdr.More() && ipHdr.FragmentOffset() == 0, true } +// Forwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) Forwarding() bool { + return uint8(atomic.LoadUint32(&p.forwarding)) == 1 +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) SetForwarding(v bool) { + if v { + atomic.StoreUint32(&p.forwarding, 1) + } else { + atomic.StoreUint32(&p.forwarding, 0) + } +} + // calculateMTU calculates the network-layer payload MTU based on the link-layer // payload mtu. func calculateMTU(mtu uint32) uint32 { @@ -571,7 +789,7 @@ func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber, hashIV ui } // NewProtocol returns an IPv4 network protocol. -func NewProtocol() stack.NetworkProtocol { +func NewProtocol(s *stack.Stack) stack.NetworkProtocol { ids := make([]uint32, buckets) // Randomly initialize hashIV and the ids. @@ -582,9 +800,10 @@ func NewProtocol() stack.NetworkProtocol { hashIV := r[buckets] return &protocol{ + stack: s, ids: ids, hashIV: hashIV, defaultTTL: DefaultTTL, - fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), + fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout, s.Clock()), } } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 86187aba8..0b3ed9483 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -36,8 +36,8 @@ import ( func TestExcludeBroadcast(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) const defaultMTU = 65536 @@ -517,8 +517,8 @@ func TestInvalidFragments(t *testing.T) { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ - ipv4.NewProtocol(), + NetworkProtocols: []stack.NetworkProtocolFactory{ + ipv4.NewProtocol, }, }) e := channel.New(0, 1500, linkAddr) @@ -929,8 +929,8 @@ func TestReceiveFragments(t *testing.T) { t.Run(test.name, func(t *testing.T) { // Setup a stack and endpoint. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, 1280, tcpip.LinkAddress("\xf0\x00")) if err := s.CreateNIC(nicID, e); err != nil { @@ -1140,7 +1140,7 @@ func TestWriteStats(t *testing.T) { func buildRoute(t *testing.T, ep stack.LinkEndpoint) stack.Route { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) if err := s.CreateNIC(1, ep); err != nil { t.Fatalf("CreateNIC(1, _) failed: %s", err) diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index 8bd8f5c52..97adbcbd4 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -5,11 +5,14 @@ package(licenses = ["notice"]) go_library( name = "ipv6", srcs = [ + "dhcpv6configurationfromndpra_string.go", "icmp.go", "ipv6.go", + "ndp.go", ], visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go b/pkg/tcpip/network/ipv6/dhcpv6configurationfromndpra_string.go index d199ded6a..09ba133b1 100644 --- a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go +++ b/pkg/tcpip/network/ipv6/dhcpv6configurationfromndpra_string.go @@ -14,7 +14,7 @@ // Code generated by "stringer -type DHCPv6ConfigurationFromNDPRA"; DO NOT EDIT. -package stack +package ipv6 import "strconv" diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 072c8ccd7..4b4b483cc 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -15,6 +15,8 @@ package ipv6 import ( + "fmt" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -39,7 +41,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match an address we own. src := hdr.SourceAddress() - if e.stack.CheckLocalAddress(e.NICID(), ProtocolNumber, src) == 0 { + if e.protocol.stack.CheckLocalAddress(e.nic.ID(), ProtocolNumber, src) == 0 { return } @@ -207,14 +209,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme return } - s := r.Stack() - if isTentative, err := s.IsAddrTentative(e.nicID, targetAddr); err != nil { - // We will only get an error if the NIC is unrecognized, which should not - // happen. For now, drop this packet. - // - // TODO(b/141002840): Handle this better? - return - } else if isTentative { + if e.hasTentativeAddr(targetAddr) { // If the target address is tentative and the source of the packet is a // unicast (specified) address, then the source of the packet is // attempting to perform address resolution on the target. In this case, @@ -227,7 +222,20 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // stack know so it can handle such a scenario and do nothing further with // the NS. if r.RemoteAddress == header.IPv6Any { - s.DupTentativeAddrDetected(e.nicID, targetAddr) + // We would get an error if the address no longer exists or the address + // is no longer tentative (DAD resolved between the call to + // hasTentativeAddr and this point). Both of these are valid scenarios: + // 1) An address may be removed at any time. + // 2) As per RFC 4862 section 5.4, DAD is not a perfect: + // "Note that the method for detecting duplicates + // is not completely reliable, and it is possible that duplicate + // addresses will still exist" + // + // TODO(gvisor.dev/issue/4046): Handle the scenario when a duplicate + // address is detected for an assigned address. + if err := e.dupTentativeAddrDetected(targetAddr); err != nil && err != tcpip.ErrBadAddress && err != tcpip.ErrInvalidEndpointState { + panic(fmt.Sprintf("unexpected error handling duplicate tentative address: %s", err)) + } } // Do not handle neighbor solicitations targeted to an address that is @@ -240,7 +248,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // section 5.4.3. // Is the NS targeting us? - if s.CheckLocalAddress(e.nicID, ProtocolNumber, targetAddr) == 0 { + if r.Stack().CheckLocalAddress(e.nic.ID(), ProtocolNumber, targetAddr) == 0 { return } @@ -275,7 +283,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme } else if e.nud != nil { e.nud.HandleProbe(r.RemoteAddress, r.LocalAddress, header.IPv6ProtocolNumber, sourceLinkAddr, e.protocol) } else { - e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, sourceLinkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), r.RemoteAddress, sourceLinkAddr) } // ICMPv6 Neighbor Solicit messages are always sent to @@ -353,20 +361,26 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // NDP datagrams are very small and ToView() will not incur allocations. na := header.NDPNeighborAdvert(payload.ToView()) targetAddr := na.TargetAddress() - s := r.Stack() - - if isTentative, err := s.IsAddrTentative(e.nicID, targetAddr); err != nil { - // We will only get an error if the NIC is unrecognized, which should not - // happen. For now short-circuit this packet. - // - // TODO(b/141002840): Handle this better? - return - } else if isTentative { + if e.hasTentativeAddr(targetAddr) { // We just got an NA from a node that owns an address we are performing // DAD on, implying the address is not unique. In this case we let the // stack know so it can handle such a scenario and do nothing furthur with // the NDP NA. - s.DupTentativeAddrDetected(e.nicID, targetAddr) + // + // We would get an error if the address no longer exists or the address + // is no longer tentative (DAD resolved between the call to + // hasTentativeAddr and this point). Both of these are valid scenarios: + // 1) An address may be removed at any time. + // 2) As per RFC 4862 section 5.4, DAD is not a perfect: + // "Note that the method for detecting duplicates + // is not completely reliable, and it is possible that duplicate + // addresses will still exist" + // + // TODO(gvisor.dev/issue/4046): Handle the scenario when a duplicate + // address is detected for an assigned address. + if err := e.dupTentativeAddrDetected(targetAddr); err != nil && err != tcpip.ErrBadAddress && err != tcpip.ErrInvalidEndpointState { + panic(fmt.Sprintf("unexpected error handling duplicate tentative address: %s", err)) + } return } @@ -396,7 +410,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // address cache with the link address for the target of the message. if len(targetLinkAddr) != 0 { if e.nud == nil { - e.linkAddrCache.AddLinkAddress(e.nicID, targetAddr, targetLinkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), targetAddr, targetLinkAddr) return } @@ -424,7 +438,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme localAddr = "" } - r, err := r.Stack().FindRoute(e.NICID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) + r, err := r.Stack().FindRoute(e.nic.ID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) if err != nil { // If we cannot find a route to the destination, silently drop the packet. return @@ -568,9 +582,9 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme e.nud.HandleProbe(routerAddr, r.LocalAddress, header.IPv6ProtocolNumber, sourceLinkAddr, e.protocol) } - // Tell the NIC to handle the RA. - stack := r.Stack() - stack.HandleNDPRA(e.nicID, routerAddr, ra) + e.mu.Lock() + e.mu.ndp.handleRA(routerAddr, ra) + e.mu.Unlock() case header.ICMPv6RedirectMsg: // TODO(gvisor.dev/issue/2285): Call `e.nud.HandleProbe` after validating @@ -671,16 +685,6 @@ func (*protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bo // ======= ICMP Error packet generation ========= -// ReturnError implements stack.TransportProtocol.ReturnError. -func (p *protocol) ReturnError(r *stack.Route, reason tcpip.ICMPReason, pkt *stack.PacketBuffer) *tcpip.Error { - switch reason.(type) { - case *tcpip.ICMPReasonPortUnreachable: - return returnError(r, &icmpReasonPortUnreachable{}, pkt) - default: - return tcpip.ErrNotSupported - } -} - // icmpReason is a marker interface for IPv6 specific ICMP errors. type icmpReason interface { isICMPReason() diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 0f50bfb8e..31370c1d4 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -75,7 +75,8 @@ type stubDispatcher struct { stack.TransportDispatcher } -func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, *stack.PacketBuffer) { +func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, *stack.PacketBuffer) stack.TransportPacketDisposition { + return stack.TransportPacketHandled } type stubLinkAddressCache struct { @@ -102,6 +103,30 @@ func (*stubNUDHandler) HandleConfirmation(addr tcpip.Address, linkAddr tcpip.Lin func (*stubNUDHandler) HandleUpperLevelConfirmation(addr tcpip.Address) { } +var _ stack.NetworkInterface = (*testInterface)(nil) + +type testInterface struct{} + +func (*testInterface) ID() tcpip.NICID { + return 0 +} + +func (*testInterface) IsLoopback() bool { + return false +} + +func (*testInterface) Name() string { + return "" +} + +func (*testInterface) Enabled() bool { + return true +} + +func (*testInterface) LinkEndpoint() stack.LinkEndpoint { + return nil +} + func TestICMPCounts(t *testing.T) { tests := []struct { name string @@ -120,8 +145,8 @@ func TestICMPCounts(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, UseNeighborCache: test.useNeighborCache, }) { @@ -149,9 +174,13 @@ func TestICMPCounts(t *testing.T) { if netProto == nil { t.Fatalf("cannot find protocol instance for network protocol %d", ProtocolNumber) } - ep := netProto.NewEndpoint(0, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}, nil, s) + ep := netProto.NewEndpoint(&testInterface{}, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + r, err := s.FindRoute(nicID, lladdr0, lladdr1, ProtocolNumber, false /* multicastLoop */) if err != nil { t.Fatalf("FindRoute(%d, %s, %s, _, false) = (_, %s), want = (_, nil)", nicID, lladdr0, lladdr1, err) @@ -258,8 +287,8 @@ func TestICMPCounts(t *testing.T) { func TestICMPCountsWithNeighborCache(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, UseNeighborCache: true, }) { @@ -287,9 +316,13 @@ func TestICMPCountsWithNeighborCache(t *testing.T) { if netProto == nil { t.Fatalf("cannot find protocol instance for network protocol %d", ProtocolNumber) } - ep := netProto.NewEndpoint(0, nil, &stubNUDHandler{}, &stubDispatcher{}, nil, s) + ep := netProto.NewEndpoint(&testInterface{}, nil, &stubNUDHandler{}, &stubDispatcher{}) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + r, err := s.FindRoute(nicID, lladdr0, lladdr1, ProtocolNumber, false /* multicastLoop */) if err != nil { t.Fatalf("FindRoute(%d, %s, %s, _, false) = (_, %s), want = (_, nil)", nicID, lladdr0, lladdr1, err) @@ -423,12 +456,12 @@ func (e endpointWithResolutionCapability) Capabilities() stack.LinkEndpointCapab func newTestContext(t *testing.T) *testContext { c := &testContext{ s0: stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, }), s1: stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, }), } @@ -723,7 +756,7 @@ func TestICMPChecksumValidationSimple(t *testing.T) { e.LinkEPCapabilities |= stack.CapabilityResolutionRequired s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: test.useNeighborCache, }) if isRouter { @@ -919,7 +952,7 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { t.Run(typ.name, func(t *testing.T) { e := channel.New(10, 1280, linkAddr0) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(_, _) = %s", err) @@ -1097,7 +1130,7 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { t.Run(typ.name, func(t *testing.T) { e := channel.New(10, 1280, linkAddr0) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -1203,7 +1236,10 @@ func TestLinkAddressRequest(t *testing.T) { } for _, test := range tests { - p := NewProtocol() + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + }) + p := s.NetworkProtocolInstance(ProtocolNumber) linkRes, ok := p.(stack.LinkAddressResolver) if !ok { t.Fatalf("expected IPv6 protocol to implement stack.LinkAddressResolver") diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 5b1cca180..990e67210 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -12,18 +12,15 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package ipv6 contains the implementation of the ipv6 network protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv6.NewProtocol() as one of the network -// protocols when calling stack.New(). Then endpoints can be created by passing -// ipv6.ProtocolNumber as the network protocol number when calling -// Stack.NewEndpoint(). +// Package ipv6 contains the implementation of the ipv6 network protocol. package ipv6 import ( "fmt" + "sort" "sync/atomic" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -45,14 +42,302 @@ const ( DefaultTTL = 64 ) +var _ stack.GroupAddressableEndpoint = (*endpoint)(nil) +var _ stack.AddressableEndpoint = (*endpoint)(nil) +var _ stack.NetworkEndpoint = (*endpoint)(nil) +var _ stack.NDPEndpoint = (*endpoint)(nil) +var _ NDPEndpoint = (*endpoint)(nil) + type endpoint struct { - nicID tcpip.NICID + nic stack.NetworkInterface linkEP stack.LinkEndpoint linkAddrCache stack.LinkAddressCache nud stack.NUDHandler dispatcher stack.TransportDispatcher protocol *protocol stack *stack.Stack + + // enabled is set to 1 when the endpoint is enabled and 0 when it is + // disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + + mu struct { + sync.RWMutex + + addressableEndpointState stack.AddressableEndpointState + ndp ndpState + } +} + +// NICNameFromID is a function that returns a stable name for the specified NIC, +// even if different NIC IDs are used to refer to the same NIC in different +// program runs. It is used when generating opaque interface identifiers (IIDs). +// If the NIC was created with a name, it is passed to NICNameFromID. +// +// NICNameFromID SHOULD return unique NIC names so unique opaque IIDs are +// generated for the same prefix on differnt NICs. +type NICNameFromID func(tcpip.NICID, string) string + +// OpaqueInterfaceIdentifierOptions holds the options related to the generation +// of opaque interface indentifiers (IIDs) as defined by RFC 7217. +type OpaqueInterfaceIdentifierOptions struct { + // NICNameFromID is a function that returns a stable name for a specified NIC, + // even if the NIC ID changes over time. + // + // Must be specified to generate the opaque IID. + NICNameFromID NICNameFromID + + // SecretKey is a pseudo-random number used as the secret key when generating + // opaque IIDs as defined by RFC 7217. The key SHOULD be at least + // header.OpaqueIIDSecretKeyMinBytes bytes and MUST follow minimum randomness + // requirements for security as outlined by RFC 4086. SecretKey MUST NOT + // change between program runs, unless explicitly changed. + // + // OpaqueInterfaceIdentifierOptions takes ownership of SecretKey. SecretKey + // MUST NOT be modified after Stack is created. + // + // May be nil, but a nil value is highly discouraged to maintain + // some level of randomness between nodes. + SecretKey []byte +} + +// InvalidateDefaultRouter implements stack.NDPEndpoint. +func (e *endpoint) InvalidateDefaultRouter(rtr tcpip.Address) { + e.mu.Lock() + defer e.mu.Unlock() + e.mu.ndp.invalidateDefaultRouter(rtr) +} + +// SetNDPConfigurations implements NDPEndpoint. +func (e *endpoint) SetNDPConfigurations(c NDPConfigurations) { + c.validate() + e.mu.Lock() + defer e.mu.Unlock() + e.mu.ndp.configs = c +} + +// hasTentativeAddr returns true if addr is tentative on e. +func (e *endpoint) hasTentativeAddr(addr tcpip.Address) bool { + e.mu.RLock() + addressEndpoint := e.getAddressRLocked(addr) + e.mu.RUnlock() + return addressEndpoint != nil && addressEndpoint.GetKind() == stack.PermanentTentative +} + +// dupTentativeAddrDetected attempts to inform e that a tentative addr is a +// duplicate on a link. +// +// dupTentativeAddrDetected removes the tentative address if it exists. If the +// address was generated via SLAAC, an attempt is made to generate a new +// address. +func (e *endpoint) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + addressEndpoint := e.getAddressRLocked(addr) + if addressEndpoint == nil { + return tcpip.ErrBadAddress + } + + if addressEndpoint.GetKind() != stack.PermanentTentative { + return tcpip.ErrInvalidEndpointState + } + + // If the address is a SLAAC address, do not invalidate its SLAAC prefix as an + // attempt will be made to generate a new address for it. + if err := e.removePermanentEndpointLocked(addressEndpoint, false /* allowSLAACInvalidation */); err != nil { + return err + } + + prefix := addressEndpoint.AddressWithPrefix().Subnet() + + switch t := addressEndpoint.ConfigType(); t { + case stack.AddressConfigStatic: + case stack.AddressConfigSlaac: + e.mu.ndp.regenerateSLAACAddr(prefix) + case stack.AddressConfigSlaacTemp: + // Do not reset the generation attempts counter for the prefix as the + // temporary address is being regenerated in response to a DAD conflict. + e.mu.ndp.regenerateTempSLAACAddr(prefix, false /* resetGenAttempts */) + default: + panic(fmt.Sprintf("unrecognized address config type = %d", t)) + } + + return nil +} + +// transitionForwarding transitions the endpoint's forwarding status to +// forwarding. +// +// Must only be called when the forwarding status changes. +func (e *endpoint) transitionForwarding(forwarding bool) { + e.mu.Lock() + defer e.mu.Unlock() + + if !e.Enabled() { + return + } + + if forwarding { + // When transitioning into an IPv6 router, host-only state (NDP discovered + // routers, discovered on-link prefixes, and auto-generated addresses) is + // cleaned up/invalidated and NDP router solicitations are stopped. + e.mu.ndp.stopSolicitingRouters() + e.mu.ndp.cleanupState(true /* hostOnly */) + } else { + // When transitioning into an IPv6 host, NDP router solicitations are + // started. + e.mu.ndp.startSolicitingRouters() + } +} + +// Enable implements stack.NetworkEndpoint. +func (e *endpoint) Enable() *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + // If the NIC is not enabled, the endpoint can't do anything meaningful so + // don't enable the endpoint. + if !e.nic.Enabled() { + return tcpip.ErrNotPermitted + } + + // If the endpoint is already enabled, there is nothing for it to do. + if !e.setEnabled(true) { + return nil + } + + // Join the IPv6 All-Nodes Multicast group if the stack is configured to + // use IPv6. This is required to ensure that this node properly receives + // and responds to the various NDP messages that are destined to the + // all-nodes multicast address. An example is the Neighbor Advertisement + // when we perform Duplicate Address Detection, or Router Advertisement + // when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861 + // section 4.2 for more information. + // + // Also auto-generate an IPv6 link-local address based on the endpoint's + // link address if it is configured to do so. Note, each interface is + // required to have IPv6 link-local unicast address, as per RFC 4291 + // section 2.1. + + // Join the All-Nodes multicast group before starting DAD as responses to DAD + // (NDP NS) messages may be sent to the All-Nodes multicast group if the + // source address of the NDP NS is the unspecified address, as per RFC 4861 + // section 7.2.4. + if _, err := e.mu.addressableEndpointState.JoinGroup(header.IPv6AllNodesMulticastAddress); err != nil { + return err + } + + // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent + // state. + // + // Addresses may have aleady completed DAD but in the time since the endpoint + // was last enabled, other devices may have acquired the same addresses. + var err *tcpip.Error + e.mu.addressableEndpointState.ReadOnly().ForEach(func(addressEndpoint stack.AddressEndpoint) bool { + addr := addressEndpoint.AddressWithPrefix().Address + if !header.IsV6UnicastAddress(addr) { + return true + } + + switch addressEndpoint.GetKind() { + case stack.Permanent: + addressEndpoint.SetKind(stack.PermanentTentative) + fallthrough + case stack.PermanentTentative: + err = e.mu.ndp.startDuplicateAddressDetection(addr, addressEndpoint) + return err == nil + default: + return true + } + }) + if err != nil { + return err + } + + // Do not auto-generate an IPv6 link-local address for loopback devices. + if e.protocol.autoGenIPv6LinkLocal { + // The valid and preferred lifetime is infinite for the auto-generated + // link-local address. + e.mu.ndp.doSLAAC(header.IPv6LinkLocalPrefix.Subnet(), header.NDPInfiniteLifetime, header.NDPInfiniteLifetime) + } + + // If we are operating as a router, then do not solicit routers since we + // won't process the RAs anyway. + // + // Routers do not process Router Advertisements (RA) the same way a host + // does. That is, routers do not learn from RAs (e.g. on-link prefixes + // and default routers). Therefore, soliciting RAs from other routers on + // a link is unnecessary for routers. + if !e.protocol.Forwarding() { + e.mu.ndp.startSolicitingRouters() + } + + return nil +} + +// Enabled implements stack.NetworkEndpoint. +func (e *endpoint) Enabled() bool { + return e.nic.Enabled() && e.isEnabled() +} + +// isEnabled returns true if the endpoint is enabled, regardless of the +// enabled status of the NIC. +func (e *endpoint) isEnabled() bool { + return atomic.LoadUint32(&e.enabled) == 1 +} + +// setEnabled sets the enabled status for the endpoint. +// +// Returns true if the enabled status was updated. +func (e *endpoint) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&e.enabled, 1) == 0 + } + return atomic.SwapUint32(&e.enabled, 0) == 1 +} + +// Disable implements stack.NetworkEndpoint. +func (e *endpoint) Disable() { + e.mu.Lock() + defer e.mu.Unlock() + e.disableLocked() +} + +func (e *endpoint) disableLocked() { + if !e.setEnabled(false) { + return + } + + e.mu.ndp.stopSolicitingRouters() + e.mu.ndp.cleanupState(false /* hostOnly */) + e.stopDADForPermanentAddressesLocked() + + // The endpoint may have already left the multicast group. + if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv6AllNodesMulticastAddress); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv6AllNodesMulticastAddress, err)) + } +} + +// stopDADForPermanentAddressesLocked stops DAD for all permaneent addresses. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) stopDADForPermanentAddressesLocked() { + // Stop DAD for all the tentative unicast addresses. + e.mu.addressableEndpointState.ReadOnly().ForEach(func(addressEndpoint stack.AddressEndpoint) bool { + if addressEndpoint.GetKind() != stack.PermanentTentative { + return true + } + + addr := addressEndpoint.AddressWithPrefix().Address + if header.IsV6UnicastAddress(addr) { + e.mu.ndp.stopDuplicateAddressDetection(addr) + } + + return true + }) } // DefaultTTL is the default hop limit for this endpoint. @@ -66,16 +351,6 @@ func (e *endpoint) MTU() uint32 { return calculateMTU(e.linkEP.MTU()) } -// NICID returns the ID of the NIC this endpoint belongs to. -func (e *endpoint) NICID() tcpip.NICID { - return e.nicID -} - -// Capabilities implements stack.NetworkEndpoint.Capabilities. -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.linkEP.Capabilities() -} - // MaxHeaderLength returns the maximum length needed by ipv6 headers (and // underlying protocols). func (e *endpoint) MaxHeaderLength() uint16 { @@ -110,8 +385,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // iptables filtering. All packets that reach here are locally // generated. - nicName := e.stack.FindNICNameFromID(e.NICID()) - ipt := e.stack.IPTables() + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Output, pkt, gso, r, "", nicName); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesOutputDropped.Increment() @@ -127,7 +402,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // short circuits broadcasts before they are sent out to other hosts. if pkt.NatDone { netHeader := header.IPv6(pkt.NetworkHeader().View()) - if ep, err := e.stack.FindNetworkEndpoint(header.IPv6ProtocolNumber, netHeader.DestinationAddress()); err == nil { + if ep, err := e.protocol.stack.FindNetworkEndpoint(header.IPv6ProtocolNumber, netHeader.DestinationAddress()); err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) ep.HandlePacket(&route, pkt) return nil @@ -170,8 +445,8 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe // iptables filtering. All packets that reach here are locally // generated. - nicName := e.stack.FindNICNameFromID(e.NICID()) - ipt := e.stack.IPTables() + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) + ipt := e.protocol.stack.IPTables() dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r, nicName) if len(dropped) == 0 && len(natPkts) == 0 { // Fast path: If no packets are to be dropped then we can just invoke the @@ -191,7 +466,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } if _, ok := natPkts[pkt]; ok { netHeader := header.IPv6(pkt.NetworkHeader().View()) - if ep, err := e.stack.FindNetworkEndpoint(header.IPv6ProtocolNumber, netHeader.DestinationAddress()); err == nil { + if ep, err := e.protocol.stack.FindNetworkEndpoint(header.IPv6ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() route := r.ReverseRoute(src, dst) @@ -224,6 +499,10 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuff // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { + if !e.isEnabled() { + return + } + h := header.IPv6(pkt.NetworkHeader().View()) if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() @@ -241,8 +520,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { hasFragmentHeader := false // iptables filtering. All packets that reach here are intended for - // this machine and will not be forwarded. - ipt := e.stack.IPTables() + // this machine and need not be forwarded. + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Input, pkt, nil, nil, "", ""); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesInputDropped.Increment() @@ -482,7 +761,18 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { r.Stats().IP.PacketsDelivered.Increment() // TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 error // in response to unrecognized next header values. - e.dispatcher.DeliverTransportPacket(r, p, pkt) + switch res := e.dispatcher.DeliverTransportPacket(r, p, pkt); res { + case stack.TransportPacketHandled: + case stack.TransportPacketDestinationPortUnreachable: + // As per RFC 4443 section 3.1: + // A destination node SHOULD originate a Destination Unreachable + // message with Code 4 in response to a packet for which the + // transport protocol (e.g., UDP) has no listener, if that transport + // protocol has no alternative means to inform the sender. + _ = returnError(r, &icmpReasonPortUnreachable{}, pkt) + default: + panic(fmt.Sprintf("unrecognized result from DeliverTransportPacket = %d", res)) + } } default: @@ -498,19 +788,333 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } // Close cleans up resources associated with the endpoint. -func (*endpoint) Close() {} +func (e *endpoint) Close() { + e.mu.Lock() + e.disableLocked() + e.mu.ndp.removeSLAACAddresses(false /* keepLinkLocal */) + e.stopDADForPermanentAddressesLocked() + e.mu.addressableEndpointState.Cleanup() + e.mu.Unlock() + + e.protocol.forgetEndpoint(e) +} // NetworkProtocolNumber implements stack.NetworkEndpoint.NetworkProtocolNumber. func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return e.protocol.Number() } +// AddAndAcquirePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { + // TODO(b/169350103): add checks here after making sure we no longer receive + // an empty address. + e.mu.Lock() + defer e.mu.Unlock() + return e.addAndAcquirePermanentAddressLocked(addr, peb, configType, deprecated) +} + +// addAndAcquirePermanentAddressLocked is like AddAndAcquirePermanentAddress but +// with locking requirements. +// +// addAndAcquirePermanentAddressLocked also joins the passed address's +// solicited-node multicast group and start duplicate address detection. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) addAndAcquirePermanentAddressLocked(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { + addressEndpoint, err := e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(addr, peb, configType, deprecated) + if err != nil { + return nil, err + } + + if !header.IsV6UnicastAddress(addr.Address) { + return addressEndpoint, nil + } + + snmc := header.SolicitedNodeAddr(addr.Address) + if _, err := e.mu.addressableEndpointState.JoinGroup(snmc); err != nil { + return nil, err + } + + addressEndpoint.SetKind(stack.PermanentTentative) + + if e.Enabled() { + if err := e.mu.ndp.startDuplicateAddressDetection(addr.Address, addressEndpoint); err != nil { + return nil, err + } + } + + return addressEndpoint, nil +} + +// RemovePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) RemovePermanentAddress(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + addressEndpoint := e.getAddressRLocked(addr) + if addressEndpoint == nil || !addressEndpoint.GetKind().IsPermanent() { + return tcpip.ErrBadLocalAddress + } + + return e.removePermanentEndpointLocked(addressEndpoint, true) +} + +// removePermanentEndpointLocked is like removePermanentAddressLocked except +// it works with a stack.AddressEndpoint. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) removePermanentEndpointLocked(addressEndpoint stack.AddressEndpoint, allowSLAACInvalidation bool) *tcpip.Error { + addr := addressEndpoint.AddressWithPrefix() + unicast := header.IsV6UnicastAddress(addr.Address) + if unicast { + e.mu.ndp.stopDuplicateAddressDetection(addr.Address) + + // If we are removing an address generated via SLAAC, cleanup + // its SLAAC resources and notify the integrator. + switch addressEndpoint.ConfigType() { + case stack.AddressConfigSlaac: + e.mu.ndp.cleanupSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) + case stack.AddressConfigSlaacTemp: + e.mu.ndp.cleanupTempSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) + } + } + + if err := e.mu.addressableEndpointState.RemovePermanentEndpoint(addressEndpoint); err != nil { + return err + } + + if !unicast { + return nil + } + + snmc := header.SolicitedNodeAddr(addr.Address) + if _, err := e.mu.addressableEndpointState.LeaveGroup(snmc); err != nil && err != tcpip.ErrBadLocalAddress { + return err + } + + return nil +} + +// hasPermanentAddressLocked returns true if the endpoint has a permanent +// address equal to the passed address. +// +// Precondition: e.mu must be read or write locked. +func (e *endpoint) hasPermanentAddressRLocked(addr tcpip.Address) bool { + addressEndpoint := e.getAddressRLocked(addr) + if addressEndpoint == nil { + return false + } + return addressEndpoint.GetKind().IsPermanent() +} + +// getAddressRLocked returns the endpoint for the passed address. +// +// Precondition: e.mu must be read or write locked. +func (e *endpoint) getAddressRLocked(localAddr tcpip.Address) stack.AddressEndpoint { + return e.mu.addressableEndpointState.ReadOnly().Lookup(localAddr) +} + +// AcquireAssignedAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB stack.PrimaryEndpointBehavior) stack.AddressEndpoint { + e.mu.Lock() + defer e.mu.Unlock() + return e.acquireAddressOrCreateTempLocked(localAddr, allowTemp, tempPEB) +} + +// acquireAddressOrCreateTempLocked is like AcquireAssignedAddress but with +// locking requirements. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) acquireAddressOrCreateTempLocked(localAddr tcpip.Address, allowTemp bool, tempPEB stack.PrimaryEndpointBehavior) stack.AddressEndpoint { + return e.mu.addressableEndpointState.AcquireAssignedAddress(localAddr, allowTemp, tempPEB) +} + +// AcquirePrimaryAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquirePrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { + e.mu.RLock() + defer e.mu.RUnlock() + return e.acquirePrimaryAddressRLocked(remoteAddr, allowExpired) +} + +// acquirePrimaryAddressRLocked is like AcquirePrimaryAddress but with locking +// requirements. +// +// Precondition: e.mu must be read locked. +func (e *endpoint) acquirePrimaryAddressRLocked(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { + // addrCandidate is a candidate for Source Address Selection, as per + // RFC 6724 section 5. + type addrCandidate struct { + addressEndpoint stack.AddressEndpoint + scope header.IPv6AddressScope + } + + if len(remoteAddr) == 0 { + return e.mu.addressableEndpointState.AcquirePrimaryAddress(remoteAddr, allowExpired) + } + + // Create a candidate set of available addresses we can potentially use as a + // source address. + var cs []addrCandidate + e.mu.addressableEndpointState.ReadOnly().ForEachPrimaryEndpoint(func(addressEndpoint stack.AddressEndpoint) { + // If r is not valid for outgoing connections, it is not a valid endpoint. + if !addressEndpoint.IsAssigned(allowExpired) { + return + } + + addr := addressEndpoint.AddressWithPrefix().Address + scope, err := header.ScopeForIPv6Address(addr) + if err != nil { + // Should never happen as we got r from the primary IPv6 endpoint list and + // ScopeForIPv6Address only returns an error if addr is not an IPv6 + // address. + panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", addr, err)) + } + + cs = append(cs, addrCandidate{ + addressEndpoint: addressEndpoint, + scope: scope, + }) + }) + + remoteScope, err := header.ScopeForIPv6Address(remoteAddr) + if err != nil { + // primaryIPv6Endpoint should never be called with an invalid IPv6 address. + panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err)) + } + + // Sort the addresses as per RFC 6724 section 5 rules 1-3. + // + // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5. + sort.Slice(cs, func(i, j int) bool { + sa := cs[i] + sb := cs[j] + + // Prefer same address as per RFC 6724 section 5 rule 1. + if sa.addressEndpoint.AddressWithPrefix().Address == remoteAddr { + return true + } + if sb.addressEndpoint.AddressWithPrefix().Address == remoteAddr { + return false + } + + // Prefer appropriate scope as per RFC 6724 section 5 rule 2. + if sa.scope < sb.scope { + return sa.scope >= remoteScope + } else if sb.scope < sa.scope { + return sb.scope < remoteScope + } + + // Avoid deprecated addresses as per RFC 6724 section 5 rule 3. + if saDep, sbDep := sa.addressEndpoint.Deprecated(), sb.addressEndpoint.Deprecated(); saDep != sbDep { + // If sa is not deprecated, it is preferred over sb. + return sbDep + } + + // Prefer temporary addresses as per RFC 6724 section 5 rule 7. + if saTemp, sbTemp := sa.addressEndpoint.ConfigType() == stack.AddressConfigSlaacTemp, sb.addressEndpoint.ConfigType() == stack.AddressConfigSlaacTemp; saTemp != sbTemp { + return saTemp + } + + // sa and sb are equal, return the endpoint that is closest to the front of + // the primary endpoint list. + return i < j + }) + + // Return the most preferred address that can have its reference count + // incremented. + for _, c := range cs { + if c.addressEndpoint.IncRef() { + return c.addressEndpoint + } + } + + return nil +} + +// PrimaryAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PrimaryAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PrimaryAddresses() +} + +// PermanentAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PermanentAddresses() +} + +// JoinGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) { + if !header.IsV6MulticastAddress(addr) { + return false, tcpip.ErrBadAddress + } + + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.JoinGroup(addr) +} + +// LeaveGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.LeaveGroup(addr) +} + +// IsInGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) IsInGroup(addr tcpip.Address) bool { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.IsInGroup(addr) +} + +var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) +var _ stack.NetworkProtocol = (*protocol)(nil) + type protocol struct { + stack *stack.Stack + + mu struct { + sync.RWMutex + + eps map[*endpoint]struct{} + } + // defaultTTL is the current default TTL for the protocol. Only the - // uint8 portion of it is meaningful and it must be accessed - // atomically. - defaultTTL uint32 + // uint8 portion of it is meaningful. + // + // Must be accessed using atomic operations. + defaultTTL uint32 + + // forwarding is set to 1 when the protocol has forwarding enabled and 0 + // when it is disabled. + // + // Must be accessed using atomic operations. + forwarding uint32 + fragmentation *fragmentation.Fragmentation + + // ndpDisp is the NDP event dispatcher that is used to send the netstack + // integrator NDP related events. + ndpDisp NDPDispatcher + + // ndpConfigs is the default NDP configurations used by an IPv6 endpoint. + ndpConfigs NDPConfigurations + + // opaqueIIDOpts hold the options for generating opaque interface identifiers + // (IIDs) as outlined by RFC 7217. + opaqueIIDOpts OpaqueInterfaceIdentifierOptions + + // tempIIDSeed is used to seed the initial temporary interface identifier + // history value used to generate IIDs for temporary SLAAC addresses. + tempIIDSeed []byte + + // autoGenIPv6LinkLocal determines whether or not the stack attempts to + // auto-generate an IPv6 link-local address for newly enabled non-loopback + // NICs. See the AutoGenIPv6LinkLocal field of Options for more details. + autoGenIPv6LinkLocal bool } // Number returns the ipv6 protocol number. @@ -535,16 +1139,36 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { } // NewEndpoint creates a new ipv6 endpoint. -func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { - return &endpoint{ - nicID: nicID, - linkEP: linkEP, +func (p *protocol) NewEndpoint(nic stack.NetworkInterface, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &endpoint{ + nic: nic, + linkEP: nic.LinkEndpoint(), linkAddrCache: linkAddrCache, nud: nud, dispatcher: dispatcher, protocol: p, - stack: st, } + e.mu.addressableEndpointState.Init(e) + e.mu.ndp = ndpState{ + ep: e, + configs: p.ndpConfigs, + dad: make(map[tcpip.Address]dadState), + defaultRouters: make(map[tcpip.Address]defaultRouterState), + onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), + slaacPrefixes: make(map[tcpip.Subnet]slaacPrefixState), + } + e.mu.ndp.initializeTempAddrState() + + p.mu.Lock() + defer p.mu.Unlock() + p.mu.eps[e] = struct{}{} + return e +} + +func (p *protocol) forgetEndpoint(e *endpoint) { + p.mu.Lock() + defer p.mu.Unlock() + delete(p.mu.eps, e) } // SetOption implements NetworkProtocol.SetOption. @@ -595,6 +1219,35 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu return proto, !fragMore && fragOffset == 0, true } +// Forwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) Forwarding() bool { + return uint8(atomic.LoadUint32(&p.forwarding)) == 1 +} + +// setForwarding sets the forwarding status for the protocol. +// +// Returns true if the forwarding status was updated. +func (p *protocol) setForwarding(v bool) bool { + if v { + return atomic.SwapUint32(&p.forwarding, 1) == 0 + } + return atomic.SwapUint32(&p.forwarding, 0) == 1 +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) SetForwarding(v bool) { + p.mu.Lock() + defer p.mu.Unlock() + + if !p.setForwarding(v) { + return + } + + for ep := range p.mu.eps { + ep.transitionForwarding(v) + } +} + // calculateMTU calculates the network-layer payload MTU based on the link-layer // payload mtu. func calculateMTU(mtu uint32) uint32 { @@ -605,10 +1258,69 @@ func calculateMTU(mtu uint32) uint32 { return maxPayloadSize } -// NewProtocol returns an IPv6 network protocol. -func NewProtocol() stack.NetworkProtocol { - return &protocol{ - defaultTTL: DefaultTTL, - fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), +// Options holds options to configure a new protocol. +type Options struct { + // NDPConfigs is the default NDP configurations used by interfaces. + NDPConfigs NDPConfigurations + + // AutoGenIPv6LinkLocal determines whether or not the stack attempts to + // auto-generate an IPv6 link-local address for newly enabled non-loopback + // NICs. + // + // Note, setting this to true does not mean that a link-local address is + // assigned right away, or at all. If Duplicate Address Detection is enabled, + // an address is only assigned if it successfully resolves. If it fails, no + // further attempts are made to auto-generate an IPv6 link-local adddress. + // + // The generated link-local address follows RFC 4291 Appendix A guidelines. + AutoGenIPv6LinkLocal bool + + // NDPDisp is the NDP event dispatcher that an integrator can provide to + // receive NDP related events. + NDPDisp NDPDispatcher + + // OpaqueIIDOpts hold the options for generating opaque interface + // identifiers (IIDs) as outlined by RFC 7217. + OpaqueIIDOpts OpaqueInterfaceIdentifierOptions + + // TempIIDSeed is used to seed the initial temporary interface identifier + // history value used to generate IIDs for temporary SLAAC addresses. + // + // Temporary SLAAC adresses are short-lived addresses which are unpredictable + // and random from the perspective of other nodes on the network. It is + // recommended that the seed be a random byte buffer of at least + // header.IIDSize bytes to make sure that temporary SLAAC addresses are + // sufficiently random. It should follow minimum randomness requirements for + // security as outlined by RFC 4086. + // + // Note: using a nil value, the same seed across netstack program runs, or a + // seed that is too small would reduce randomness and increase predictability, + // defeating the purpose of temporary SLAAC addresses. + TempIIDSeed []byte +} + +// NewProtocolWithOptions returns an IPv6 network protocol. +func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory { + opts.NDPConfigs.validate() + + return func(s *stack.Stack) stack.NetworkProtocol { + p := &protocol{ + stack: s, + fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout, s.Clock()), + + ndpDisp: opts.NDPDisp, + ndpConfigs: opts.NDPConfigs, + opaqueIIDOpts: opts.OpaqueIIDOpts, + tempIIDSeed: opts.TempIIDSeed, + autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, + } + p.mu.eps = make(map[*endpoint]struct{}) + p.SetDefaultTTL(DefaultTTL) + return p } } + +// NewProtocol is equivalent to NewProtocolWithOptions with an empty Options. +func NewProtocol(s *stack.Stack) stack.NetworkProtocol { + return NewProtocolWithOptions(Options{})(s) +} diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 7d138dadb..d85b5c00f 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -141,18 +141,18 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { tests := []struct { name string - protocolFactory stack.TransportProtocol + protocolFactory stack.TransportProtocolFactory rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, - {"UDP", udp.NewProtocol(), testReceiveUDP}, + {"ICMP", icmp.NewProtocol6, testReceiveICMP}, + {"UDP", udp.NewProtocol, testReceiveUDP}, } for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{test.protocolFactory}, }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { @@ -174,11 +174,11 @@ func TestReceiveOnSolicitedNodeAddr(t *testing.T) { tests := []struct { name string - protocolFactory stack.TransportProtocol + protocolFactory stack.TransportProtocolFactory rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, - {"UDP", udp.NewProtocol(), testReceiveUDP}, + {"ICMP", icmp.NewProtocol6, testReceiveICMP}, + {"UDP", udp.NewProtocol, testReceiveUDP}, } snmc := header.SolicitedNodeAddr(addr2) @@ -186,8 +186,8 @@ func TestReceiveOnSolicitedNodeAddr(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{test.protocolFactory}, }) e := channel.New(1, 1280, linkAddr1) if err := s.CreateNIC(nicID, e); err != nil { @@ -273,7 +273,7 @@ func TestAddIpv6Address(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -579,8 +579,8 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, 1280, linkAddr1) if err := s.CreateNIC(nicID, e); err != nil { @@ -1549,8 +1549,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, 1280, linkAddr1) if err := s.CreateNIC(nicID, e); err != nil { @@ -1668,8 +1668,8 @@ func TestInvalidIPv6Fragments(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ - NewProtocol(), + NetworkProtocols: []stack.NetworkProtocolFactory{ + NewProtocol, }, }) e := channel.New(0, 1500, linkAddr1) @@ -1847,7 +1847,7 @@ func TestWriteStats(t *testing.T) { func buildRoute(t *testing.T, ep stack.LinkEndpoint) stack.Route { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(1, ep); err != nil { t.Fatalf("CreateNIC(1, _) failed: %s", err) @@ -1895,3 +1895,30 @@ func (lm *limitedMatcher) Match(stack.Hook, *stack.PacketBuffer, string) (bool, lm.limit-- return false, false } + +func TestClearEndpointFromProtocolOnClose(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + }) + proto := s.NetworkProtocolInstance(ProtocolNumber).(*protocol) + ep := proto.NewEndpoint(&testInterface{}, nil, nil, nil).(*endpoint) + { + proto.mu.Lock() + _, hasEP := proto.mu.eps[ep] + proto.mu.Unlock() + if !hasEP { + t.Fatalf("expected protocol to have ep = %p in set of endpoints", ep) + } + } + + ep.Close() + + { + proto.mu.Lock() + _, hasEP := proto.mu.eps[ep] + proto.mu.Unlock() + if hasEP { + t.Fatalf("unexpectedly found ep = %p in set of protocol's endpoints", ep) + } + } +} diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/network/ipv6/ndp.go index 97ca00d16..84c082852 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/network/ipv6/ndp.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package stack +package ipv6 import ( "fmt" @@ -23,9 +23,27 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) const ( + // defaultRetransmitTimer is the default amount of time to wait between + // sending reachability probes. + // + // Default taken from RETRANS_TIMER of RFC 4861 section 10. + defaultRetransmitTimer = time.Second + + // minimumRetransmitTimer is the minimum amount of time to wait between + // sending reachability probes. + // + // Note, RFC 4861 does not impose a minimum Retransmit Timer, but we do here + // to make sure the messages are not sent all at once. We also come to this + // value because in the RetransmitTimer field of a Router Advertisement, a + // value of 0 means unspecified, so the smallest valid value is 1. Note, the + // unit of the RetransmitTimer field in the Router Advertisement is + // milliseconds. + minimumRetransmitTimer = time.Millisecond + // defaultDupAddrDetectTransmits is the default number of NDP Neighbor // Solicitation messages to send when doing Duplicate Address Detection // for a tentative address. @@ -34,7 +52,7 @@ const ( defaultDupAddrDetectTransmits = 1 // defaultMaxRtrSolicitations is the default number of Router - // Solicitation messages to send when a NIC becomes enabled. + // Solicitation messages to send when an IPv6 endpoint becomes enabled. // // Default = 3 (from RFC 4861 section 10). defaultMaxRtrSolicitations = 3 @@ -131,7 +149,7 @@ const ( minRegenAdvanceDuration = time.Duration(0) // maxSLAACAddrLocalRegenAttempts is the maximum number of times to attempt - // SLAAC address regenerations in response to a NIC-local conflict. + // SLAAC address regenerations in response to an IPv6 endpoint-local conflict. maxSLAACAddrLocalRegenAttempts = 10 ) @@ -163,7 +181,7 @@ var ( // This is exported as a variable (instead of a constant) so tests // can update it to a smaller value. // - // This value guarantees that a temporary address will be preferred for at + // This value guarantees that a temporary address is preferred for at // least 1hr if the SLAAC prefix is valid for at least that time. MinMaxTempAddrPreferredLifetime = defaultRegenAdvanceDuration + MaxDesyncFactor + time.Hour @@ -173,11 +191,17 @@ var ( // This is exported as a variable (instead of a constant) so tests // can update it to a smaller value. // - // This value guarantees that a temporary address will be valid for at least + // This value guarantees that a temporary address is valid for at least // 2hrs if the SLAAC prefix is valid for at least that time. MinMaxTempAddrValidLifetime = 2 * time.Hour ) +// NDPEndpoint is an endpoint that supports NDP. +type NDPEndpoint interface { + // SetNDPConfigurations sets the NDP configurations. + SetNDPConfigurations(NDPConfigurations) +} + // DHCPv6ConfigurationFromNDPRA is a configuration available via DHCPv6 that an // NDP Router Advertisement informed the Stack about. type DHCPv6ConfigurationFromNDPRA int @@ -192,7 +216,7 @@ const ( // DHCPv6ManagedAddress indicates that addresses are available via DHCPv6. // // DHCPv6ManagedAddress also implies DHCPv6OtherConfigurations because DHCPv6 - // will return all available configuration information. + // returns all available configuration information when serving addresses. DHCPv6ManagedAddress // DHCPv6OtherConfigurations indicates that other configuration information is @@ -207,19 +231,18 @@ const ( // NDPDispatcher is the interface integrators of netstack must implement to // receive and handle NDP related events. type NDPDispatcher interface { - // OnDuplicateAddressDetectionStatus will be called when the DAD process - // for an address (addr) on a NIC (with ID nicID) completes. resolved - // will be set to true if DAD completed successfully (no duplicate addr - // detected); false otherwise (addr was detected to be a duplicate on - // the link the NIC is a part of, or it was stopped for some other - // reason, such as the address being removed). If an error occured - // during DAD, err will be set and resolved must be ignored. + // OnDuplicateAddressDetectionStatus is called when the DAD process for an + // address (addr) on a NIC (with ID nicID) completes. resolved is set to true + // if DAD completed successfully (no duplicate addr detected); false otherwise + // (addr was detected to be a duplicate on the link the NIC is a part of, or + // it was stopped for some other reason, such as the address being removed). + // If an error occured during DAD, err is set and resolved must be ignored. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) - // OnDefaultRouterDiscovered will be called when a new default router is + // OnDefaultRouterDiscovered is called when a new default router is // discovered. Implementations must return true if the newly discovered // router should be remembered. // @@ -227,56 +250,55 @@ type NDPDispatcher interface { // is also not permitted to call into the stack. OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) bool - // OnDefaultRouterInvalidated will be called when a discovered default - // router that was remembered is invalidated. + // OnDefaultRouterInvalidated is called when a discovered default router that + // was remembered is invalidated. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) - // OnOnLinkPrefixDiscovered will be called when a new on-link prefix is - // discovered. Implementations must return true if the newly discovered - // on-link prefix should be remembered. + // OnOnLinkPrefixDiscovered is called when a new on-link prefix is discovered. + // Implementations must return true if the newly discovered on-link prefix + // should be remembered. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool - // OnOnLinkPrefixInvalidated will be called when a discovered on-link - // prefix that was remembered is invalidated. + // OnOnLinkPrefixInvalidated is called when a discovered on-link prefix that + // was remembered is invalidated. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) - // OnAutoGenAddress will be called when a new prefix with its - // autonomous address-configuration flag set has been received and SLAAC - // has been performed. Implementations may prevent the stack from - // assigning the address to the NIC by returning false. + // OnAutoGenAddress is called when a new prefix with its autonomous address- + // configuration flag set is received and SLAAC was performed. Implementations + // may prevent the stack from assigning the address to the NIC by returning + // false. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool - // OnAutoGenAddressDeprecated will be called when an auto-generated - // address (as part of SLAAC) has been deprecated, but is still - // considered valid. Note, if an address is invalidated at the same - // time it is deprecated, the deprecation event MAY be omitted. + // OnAutoGenAddressDeprecated is called when an auto-generated address (SLAAC) + // is deprecated, but is still considered valid. Note, if an address is + // invalidated at the same ime it is deprecated, the deprecation event may not + // be received. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. OnAutoGenAddressDeprecated(tcpip.NICID, tcpip.AddressWithPrefix) - // OnAutoGenAddressInvalidated will be called when an auto-generated - // address (as part of SLAAC) has been invalidated. + // OnAutoGenAddressInvalidated is called when an auto-generated address + // (SLAAC) is invalidated. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) - // OnRecursiveDNSServerOption will be called when an NDP option with - // recursive DNS servers has been received. Note, addrs may contain - // link-local addresses. + // OnRecursiveDNSServerOption is called when the stack learns of DNS servers + // through NDP. Note, the addresses may contain link-local addresses. // // It is up to the caller to use the DNS Servers only for their valid // lifetime. OnRecursiveDNSServerOption may be called for new or @@ -288,8 +310,8 @@ type NDPDispatcher interface { // call functions on the stack itself. OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) - // OnDNSSearchListOption will be called when an NDP option with a DNS - // search list has been received. + // OnDNSSearchListOption is called when the stack learns of DNS search lists + // through NDP. // // It is up to the caller to use the domain names in the search list // for only their valid lifetime. OnDNSSearchListOption may be called @@ -298,8 +320,8 @@ type NDPDispatcher interface { // be increased, decreased or completely invalidated when lifetime = 0. OnDNSSearchListOption(nicID tcpip.NICID, domainNames []string, lifetime time.Duration) - // OnDHCPv6Configuration will be called with an updated configuration that is - // available via DHCPv6 for a specified NIC. + // OnDHCPv6Configuration is called with an updated configuration that is + // available via DHCPv6 for the passed NIC. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. @@ -320,7 +342,7 @@ type NDPConfigurations struct { // Must be greater than or equal to 1ms. RetransmitTimer time.Duration - // The number of Router Solicitation messages to send when the NIC + // The number of Router Solicitation messages to send when the IPv6 endpoint // becomes enabled. MaxRtrSolicitations uint8 @@ -335,24 +357,22 @@ type NDPConfigurations struct { // Must be greater than or equal to 0s. MaxRtrSolicitationDelay time.Duration - // HandleRAs determines whether or not Router Advertisements will be - // processed. + // HandleRAs determines whether or not Router Advertisements are processed. HandleRAs bool - // DiscoverDefaultRouters determines whether or not default routers will - // be discovered from Router Advertisements. This configuration is - // ignored if HandleRAs is false. + // DiscoverDefaultRouters determines whether or not default routers are + // discovered from Router Advertisements, as per RFC 4861 section 6. This + // configuration is ignored if HandleRAs is false. DiscoverDefaultRouters bool - // DiscoverOnLinkPrefixes determines whether or not on-link prefixes - // will be discovered from Router Advertisements' Prefix Information - // option. This configuration is ignored if HandleRAs is false. + // DiscoverOnLinkPrefixes determines whether or not on-link prefixes are + // discovered from Router Advertisements' Prefix Information option, as per + // RFC 4861 section 6. This configuration is ignored if HandleRAs is false. DiscoverOnLinkPrefixes bool - // AutoGenGlobalAddresses determines whether or not global IPv6 - // addresses will be generated for a NIC in response to receiving a new - // Prefix Information option with its Autonomous Address - // AutoConfiguration flag set, as a host, as per RFC 4862 (SLAAC). + // AutoGenGlobalAddresses determines whether or not an IPv6 endpoint performs + // SLAAC to auto-generate global SLAAC addresses in response to Prefix + // Information options, as per RFC 4862. // // Note, if an address was already generated for some unique prefix, as // part of SLAAC, this option does not affect whether or not the @@ -366,12 +386,12 @@ type NDPConfigurations struct { // // If the method used to generate the address does not support creating // alternative addresses (e.g. IIDs based on the modified EUI64 of a NIC's - // MAC address), then no attempt will be made to resolve the conflict. + // MAC address), then no attempt is made to resolve the conflict. AutoGenAddressConflictRetries uint8 // AutoGenTempGlobalAddresses determines whether or not temporary SLAAC - // addresses will be generated for a NIC as part of SLAAC privacy extensions, - // RFC 4941. + // addresses are generated for an IPv6 endpoint as part of SLAAC privacy + // extensions, as per RFC 4941. // // Ignored if AutoGenGlobalAddresses is false. AutoGenTempGlobalAddresses bool @@ -410,7 +430,7 @@ func DefaultNDPConfigurations() NDPConfigurations { } // validate modifies an NDPConfigurations with valid values. If invalid values -// are present in c, the corresponding default values will be used instead. +// are present in c, the corresponding default values are used instead. func (c *NDPConfigurations) validate() { if c.RetransmitTimer < minimumRetransmitTimer { c.RetransmitTimer = defaultRetransmitTimer @@ -439,8 +459,8 @@ func (c *NDPConfigurations) validate() { // ndpState is the per-interface NDP state. type ndpState struct { - // The NIC this ndpState is for. - nic *NIC + // The IPv6 endpoint this ndpState is for. + ep *endpoint // configs is the per-interface NDP configurations. configs NDPConfigurations @@ -458,8 +478,8 @@ type ndpState struct { // Used to let the Router Solicitation timer know that it has been stopped. // // Must only be read from or written to while protected by the lock of - // the NIC this ndpState is associated with. MUST be set when the timer is - // set. + // the IPv6 endpoint this ndpState is associated with. MUST be set when the + // timer is set. done *bool } @@ -492,7 +512,7 @@ type dadState struct { // Used to let the DAD timer know that it has been stopped. // // Must only be read from or written to while protected by the lock of - // the NIC this dadState is associated with. + // the IPv6 endpoint this dadState is associated with. done *bool } @@ -537,7 +557,7 @@ type tempSLAACAddrState struct { // The address's endpoint. // // Must not be nil. - ref *referencedNetworkEndpoint + addressEndpoint stack.AddressEndpoint // Has a new temporary SLAAC address already been regenerated? regenerated bool @@ -567,10 +587,10 @@ type slaacPrefixState struct { // // May only be nil when the address is being (re-)generated. Otherwise, // must not be nil as all SLAAC prefixes must have a stable address. - ref *referencedNetworkEndpoint + addressEndpoint stack.AddressEndpoint - // The number of times an address has been generated locally where the NIC - // already had the generated address. + // The number of times an address has been generated locally where the IPv6 + // endpoint already had the generated address. localGenerationFailures uint8 } @@ -578,11 +598,12 @@ type slaacPrefixState struct { tempAddrs map[tcpip.Address]tempSLAACAddrState // The next two fields are used by both stable and temporary addresses - // generated for a SLAAC prefix. This is safe as only 1 address will be - // in the generation and DAD process at any time. That is, no two addresses - // will be generated at the same time for a given SLAAC prefix. + // generated for a SLAAC prefix. This is safe as only 1 address is in the + // generation and DAD process at any time. That is, no two addresses are + // generated at the same time for a given SLAAC prefix. - // The number of times an address has been generated and added to the NIC. + // The number of times an address has been generated and added to the IPv6 + // endpoint. // // Addresses may be regenerated in reseponse to a DAD conflicts. generationAttempts uint8 @@ -597,16 +618,16 @@ type slaacPrefixState struct { // This function must only be called by IPv6 addresses that are currently // tentative. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, addressEndpoint stack.AddressEndpoint) *tcpip.Error { // addr must be a valid unicast IPv6 address. if !header.IsV6UnicastAddress(addr) { return tcpip.ErrAddressFamilyNotSupported } - if ref.getKind() != permanentTentative { + if addressEndpoint.GetKind() != stack.PermanentTentative { // The endpoint should be marked as tentative since we are starting DAD. - panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.nic.ID())) + panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.ep.nic.ID())) } // Should not attempt to perform DAD on an address that is currently in the @@ -617,18 +638,18 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // existed, we would get an error since we attempted to add a duplicate // address, or its reference count would have been increased without doing // the work that would have been done for an address that was brand new. - // See NIC.addAddressLocked. - panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.nic.ID())) + // See endpoint.addAddressLocked. + panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.ep.nic.ID())) } remaining := ndp.configs.DupAddrDetectTransmits if remaining == 0 { - ref.setKind(permanent) + addressEndpoint.SetKind(stack.Permanent) // Consider DAD to have resolved even if no DAD messages were actually // transmitted. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, true, nil) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, true, nil) } return nil @@ -637,25 +658,25 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref var done bool var timer tcpip.Timer // We initially start a timer to fire immediately because some of the DAD work - // cannot be done while holding the NIC's lock. This is effectively the same - // as starting a goroutine but we use a timer that fires immediately so we can - // reset it for the next DAD iteration. - timer = ndp.nic.stack.Clock().AfterFunc(0, func() { - ndp.nic.mu.Lock() - defer ndp.nic.mu.Unlock() + // cannot be done while holding the IPv6 endpoint's lock. This is effectively + // the same as starting a goroutine but we use a timer that fires immediately + // so we can reset it for the next DAD iteration. + timer = ndp.ep.protocol.stack.Clock().AfterFunc(0, func() { + ndp.ep.mu.Lock() + defer ndp.ep.mu.Unlock() if done { // If we reach this point, it means that the DAD timer fired after - // another goroutine already obtained the NIC lock and stopped DAD - // before this function obtained the NIC lock. Simply return here and do - // nothing further. + // another goroutine already obtained the IPv6 endpoint lock and stopped + // DAD before this function obtained the NIC lock. Simply return here and + // do nothing further. return } - if ref.getKind() != permanentTentative { + if addressEndpoint.GetKind() != stack.PermanentTentative { // The endpoint should still be marked as tentative since we are still // performing DAD on it. - panic(fmt.Sprintf("ndpdad: addr %s is no longer tentative on NIC(%d)", addr, ndp.nic.ID())) + panic(fmt.Sprintf("ndpdad: addr %s is no longer tentative on NIC(%d)", addr, ndp.ep.nic.ID())) } dadDone := remaining == 0 @@ -663,33 +684,34 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref var err *tcpip.Error if !dadDone { // Use the unspecified address as the source address when performing DAD. - ref := ndp.nic.getRefOrCreateTempLocked(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint) + addressEndpoint := ndp.ep.acquireAddressOrCreateTempLocked(header.IPv6Any, true /* createTemp */, stack.NeverPrimaryEndpoint) // Do not hold the lock when sending packets which may be a long running // task or may block link address resolution. We know this is safe // because immediately after obtaining the lock again, we check if DAD - // has been stopped before doing any work with the NIC. Note, DAD would be - // stopped if the NIC was disabled or removed, or if the address was - // removed. - ndp.nic.mu.Unlock() - err = ndp.sendDADPacket(addr, ref) - ndp.nic.mu.Lock() + // has been stopped before doing any work with the IPv6 endpoint. Note, + // DAD would be stopped if the IPv6 endpoint was disabled or closed, or if + // the address was removed. + ndp.ep.mu.Unlock() + err = ndp.sendDADPacket(addr, addressEndpoint) + ndp.ep.mu.Lock() + addressEndpoint.DecRef() } if done { // If we reach this point, it means that DAD was stopped after we released - // the NIC's read lock and before we obtained the write lock. + // the IPv6 endpoint's read lock and before we obtained the write lock. return } if dadDone { // DAD has resolved. - ref.setKind(permanent) + addressEndpoint.SetKind(stack.Permanent) } else if err == nil { // DAD is not done and we had no errors when sending the last NDP NS, // schedule the next DAD timer. remaining-- - timer.Reset(ndp.nic.stack.ndpConfigs.RetransmitTimer) + timer.Reset(ndp.configs.RetransmitTimer) return } @@ -698,16 +720,16 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // integrator know DAD has completed. delete(ndp.dad, addr) - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, dadDone, err) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, dadDone, err) } // If DAD resolved for a stable SLAAC address, attempt generation of a // temporary SLAAC address. - if dadDone && ref.configType == slaac { + if dadDone && addressEndpoint.ConfigType() == stack.AddressConfigSlaac { // Reset the generation attempts counter as we are starting the generation // of a new address for the SLAAC prefix. - ndp.regenerateTempSLAACAddr(ref.addrWithPrefix().Subnet(), true /* resetGenAttempts */) + ndp.regenerateTempSLAACAddr(addressEndpoint.AddressWithPrefix().Subnet(), true /* resetGenAttempts */) } }) @@ -722,28 +744,31 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // sendDADPacket sends a NS message to see if any nodes on ndp's NIC's link owns // addr. // -// addr must be a tentative IPv6 address on ndp's NIC. +// addr must be a tentative IPv6 address on ndp's IPv6 endpoint. // -// The NIC ndp belongs to MUST NOT be locked. -func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error { +// The IPv6 endpoint that ndp belongs to MUST NOT be locked. +func (ndp *ndpState) sendDADPacket(addr tcpip.Address, addressEndpoint stack.AddressEndpoint) *tcpip.Error { snmc := header.SolicitedNodeAddr(addr) - r := makeRoute(header.IPv6ProtocolNumber, ref.address(), snmc, ndp.nic.linkEP.LinkAddress(), ref, false, false) + r, err := ndp.ep.protocol.stack.FindRoute(ndp.ep.nic.ID(), header.IPv6Any, snmc, ProtocolNumber, false /* multicastLoop */) + if err != nil { + return err + } defer r.Release() // Route should resolve immediately since snmc is a multicast address so a // remote link address can be calculated without a resolution process. if c, err := r.Resolve(nil); err != nil { // Do not consider the NIC being unknown or disabled as a fatal error. - // Since this method is required to be called when the NIC is not locked, - // the NIC could have been disabled or removed by another goroutine. + // Since this method is required to be called when the IPv6 endpoint is not + // locked, the NIC could have been disabled or removed by another goroutine. if err == tcpip.ErrUnknownNICID || err != tcpip.ErrInvalidEndpointState { return err } - panic(fmt.Sprintf("ndp: error when resolving route to send NDP NS for DAD (%s -> %s on NIC(%d)): %s", header.IPv6Any, snmc, ndp.nic.ID(), err)) + panic(fmt.Sprintf("ndp: error when resolving route to send NDP NS for DAD (%s -> %s on NIC(%d)): %s", header.IPv6Any, snmc, ndp.ep.nic.ID(), err)) } else if c != nil { - panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.nic.ID())) + panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.ep.nic.ID())) } icmpData := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize)) @@ -752,17 +777,16 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd ns.SetTargetAddress(addr) icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - pkt := NewPacketBuffer(PacketBufferOptions{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), Data: buffer.View(icmpData).ToVectorisedView(), }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, - NetworkHeaderParams{ + stack.NetworkHeaderParams{ Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, - TOS: DefaultTOS, }, pkt, ); err != nil { sent.Dropped.Increment() @@ -778,11 +802,9 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd // such a state forever, unless some other external event resolves the DAD // process (receiving an NA from the true owner of addr, or an NS for addr // (implying another node is attempting to use addr)). It is up to the caller -// of this function to handle such a scenario. Normally, addr will be removed -// from n right after this function returns or the address successfully -// resolved. +// of this function to handle such a scenario. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { dad, ok := ndp.dad[addr] if !ok { @@ -801,30 +823,30 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { delete(ndp.dad, addr) // Let the integrator know DAD did not resolve. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, false, nil) } } // handleRA handles a Router Advertisement message that arrived on the NIC // this ndp is for. Does nothing if the NIC is configured to not handle RAs. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { - // Is the NIC configured to handle RAs at all? + // Is the IPv6 endpoint configured to handle RAs at all? // // Currently, the stack does not determine router interface status on a - // per-interface basis; it is a stack-wide configuration, so we check - // stack's forwarding flag to determine if the NIC is a routing - // interface. - if !ndp.configs.HandleRAs || ndp.nic.stack.Forwarding(header.IPv6ProtocolNumber) { + // per-interface basis; it is a protocol-wide configuration, so we check the + // protocol's forwarding flag to determine if the IPv6 endpoint is forwarding + // packets. + if !ndp.configs.HandleRAs || ndp.ep.protocol.Forwarding() { return } // Only worry about the DHCPv6 configuration if we have an NDPDispatcher as we // only inform the dispatcher on configuration changes. We do nothing else // with the information. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { var configuration DHCPv6ConfigurationFromNDPRA switch { case ra.ManagedAddrConfFlag(): @@ -839,11 +861,11 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { if ndp.dhcpv6Configuration != configuration { ndp.dhcpv6Configuration = configuration - ndpDisp.OnDHCPv6Configuration(ndp.nic.ID(), configuration) + ndpDisp.OnDHCPv6Configuration(ndp.ep.nic.ID(), configuration) } } - // Is the NIC configured to discover default routers? + // Is the IPv6 endpoint configured to discover default routers? if ndp.configs.DiscoverDefaultRouters { rtr, ok := ndp.defaultRouters[ip] rl := ra.RouterLifetime() @@ -881,20 +903,20 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { for opt, done, _ := it.Next(); !done; opt, done, _ = it.Next() { switch opt := opt.(type) { case header.NDPRecursiveDNSServer: - if ndp.nic.stack.ndpDisp == nil { + if ndp.ep.protocol.ndpDisp == nil { continue } addrs, _ := opt.Addresses() - ndp.nic.stack.ndpDisp.OnRecursiveDNSServerOption(ndp.nic.ID(), addrs, opt.Lifetime()) + ndp.ep.protocol.ndpDisp.OnRecursiveDNSServerOption(ndp.ep.nic.ID(), addrs, opt.Lifetime()) case header.NDPDNSSearchList: - if ndp.nic.stack.ndpDisp == nil { + if ndp.ep.protocol.ndpDisp == nil { continue } domainNames, _ := opt.DomainNames() - ndp.nic.stack.ndpDisp.OnDNSSearchListOption(ndp.nic.ID(), domainNames, opt.Lifetime()) + ndp.ep.protocol.ndpDisp.OnDNSSearchListOption(ndp.ep.nic.ID(), domainNames, opt.Lifetime()) case header.NDPPrefixInformation: prefix := opt.Subnet() @@ -928,7 +950,7 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { // invalidateDefaultRouter invalidates a discovered default router. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { rtr, ok := ndp.defaultRouters[ip] @@ -942,32 +964,32 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDefaultRouterInvalidated(ndp.nic.ID(), ip) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDefaultRouterInvalidated(ndp.ep.nic.ID(), ip) } } // rememberDefaultRouter remembers a newly discovered default router with IPv6 // link-local address ip with lifetime rl. // -// The router identified by ip MUST NOT already be known by the NIC. +// The router identified by ip MUST NOT already be known by the IPv6 endpoint. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { - ndpDisp := ndp.nic.stack.ndpDisp + ndpDisp := ndp.ep.protocol.ndpDisp if ndpDisp == nil { return } // Inform the integrator when we discovered a default router. - if !ndpDisp.OnDefaultRouterDiscovered(ndp.nic.ID(), ip) { + if !ndpDisp.OnDefaultRouterDiscovered(ndp.ep.nic.ID(), ip) { // Informed by the integrator to not remember the router, do // nothing further. return } state := defaultRouterState{ - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { ndp.invalidateDefaultRouter(ip) }), } @@ -982,22 +1004,22 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { // // The prefix identified by prefix MUST NOT already be known. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) { - ndpDisp := ndp.nic.stack.ndpDisp + ndpDisp := ndp.ep.protocol.ndpDisp if ndpDisp == nil { return } // Inform the integrator when we discovered an on-link prefix. - if !ndpDisp.OnOnLinkPrefixDiscovered(ndp.nic.ID(), prefix) { + if !ndpDisp.OnOnLinkPrefixDiscovered(ndp.ep.nic.ID(), prefix) { // Informed by the integrator to not remember the prefix, do // nothing further. return } state := onLinkPrefixState{ - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { ndp.invalidateOnLinkPrefix(prefix) }), } @@ -1011,7 +1033,7 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) // invalidateOnLinkPrefix invalidates a discovered on-link prefix. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { s, ok := ndp.onLinkPrefixes[prefix] @@ -1025,8 +1047,8 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { delete(ndp.onLinkPrefixes, prefix) // Let the integrator know a discovered on-link prefix is invalidated. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnOnLinkPrefixInvalidated(ndp.nic.ID(), prefix) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnOnLinkPrefixInvalidated(ndp.ep.nic.ID(), prefix) } } @@ -1036,7 +1058,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { // handleOnLinkPrefixInformation assumes that the prefix this pi is for is // not the link-local prefix and the on-link flag is set. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformation) { prefix := pi.Subnet() prefixState, ok := ndp.onLinkPrefixes[prefix] @@ -1089,7 +1111,7 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio // handleAutonomousPrefixInformation assumes that the prefix this pi is for is // not the link-local prefix and the autonomous flag is set. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInformation) { vl := pi.ValidLifetime() pl := pi.PreferredLifetime() @@ -1125,7 +1147,7 @@ func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInform // // pl is the new preferred lifetime. vl is the new valid lifetime. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { // If we do not already have an address for this prefix and the valid // lifetime is 0, no need to do anything further, as per RFC 4862 @@ -1142,15 +1164,15 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } state := slaacPrefixState{ - deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + deprecationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix)) } - ndp.deprecateSLAACAddress(state.stableAddr.ref) + ndp.deprecateSLAACAddress(state.stableAddr.addressEndpoint) }), - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix)) @@ -1189,7 +1211,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } // If the address is assigned (DAD resolved), generate a temporary address. - if state.stableAddr.ref.getKind() == permanent { + if state.stableAddr.addressEndpoint.GetKind() == stack.Permanent { // Reset the generation attempts counter as we are starting the generation // of a new address for the SLAAC prefix. ndp.generateTempSLAACAddr(prefix, &state, true /* resetGenAttempts */) @@ -1198,32 +1220,27 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { ndp.slaacPrefixes[prefix] = state } -// addSLAACAddr adds a SLAAC address to the NIC. +// addAndAcquireSLAACAddr adds a SLAAC address to the IPv6 endpoint. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) addSLAACAddr(addr tcpip.AddressWithPrefix, configType networkEndpointConfigType, deprecated bool) *referencedNetworkEndpoint { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) addAndAcquireSLAACAddr(addr tcpip.AddressWithPrefix, configType stack.AddressConfigType, deprecated bool) stack.AddressEndpoint { // Inform the integrator that we have a new SLAAC address. - ndpDisp := ndp.nic.stack.ndpDisp + ndpDisp := ndp.ep.protocol.ndpDisp if ndpDisp == nil { return nil } - if !ndpDisp.OnAutoGenAddress(ndp.nic.ID(), addr) { + if !ndpDisp.OnAutoGenAddress(ndp.ep.nic.ID(), addr) { // Informed by the integrator not to add the address. return nil } - protocolAddr := tcpip.ProtocolAddress{ - Protocol: header.IPv6ProtocolNumber, - AddressWithPrefix: addr, - } - - ref, err := ndp.nic.addAddressLocked(protocolAddr, FirstPrimaryEndpoint, permanent, configType, deprecated) + addressEndpoint, err := ndp.ep.addAndAcquirePermanentAddressLocked(addr, stack.FirstPrimaryEndpoint, configType, deprecated) if err != nil { - panic(fmt.Sprintf("ndp: error when adding SLAAC address %+v: %s", protocolAddr, err)) + panic(fmt.Sprintf("ndp: error when adding SLAAC address %+v: %s", addr, err)) } - return ref + return addressEndpoint } // generateSLAACAddr generates a SLAAC address for prefix. @@ -1232,10 +1249,10 @@ func (ndp *ndpState) addSLAACAddr(addr tcpip.AddressWithPrefix, configType netwo // // Panics if the prefix is not a SLAAC prefix or it already has an address. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixState) bool { - if r := state.stableAddr.ref; r != nil { - panic(fmt.Sprintf("ndp: SLAAC prefix %s already has a permenant address %s", prefix, r.addrWithPrefix())) + if addressEndpoint := state.stableAddr.addressEndpoint; addressEndpoint != nil { + panic(fmt.Sprintf("ndp: SLAAC prefix %s already has a permenant address %s", prefix, addressEndpoint.AddressWithPrefix())) } // If we have already reached the maximum address generation attempts for the @@ -1255,11 +1272,11 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt } dadCounter := state.generationAttempts + state.stableAddr.localGenerationFailures - if oIID := ndp.nic.stack.opaqueIIDOpts; oIID.NICNameFromID != nil { + if oIID := ndp.ep.protocol.opaqueIIDOpts; oIID.NICNameFromID != nil { addrBytes = header.AppendOpaqueInterfaceIdentifier( addrBytes[:header.IIDOffsetInIPv6Address], prefix, - oIID.NICNameFromID(ndp.nic.ID(), ndp.nic.name), + oIID.NICNameFromID(ndp.ep.nic.ID(), ndp.ep.nic.Name()), dadCounter, oIID.SecretKey, ) @@ -1272,7 +1289,7 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt // // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by // LinkEndpoint.LinkAddress) before reaching this point. - linkAddr := ndp.nic.linkEP.LinkAddress() + linkAddr := ndp.ep.linkEP.LinkAddress() if !header.IsValidUnicastEthernetAddress(linkAddr) { return false } @@ -1291,15 +1308,15 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt PrefixLen: validPrefixLenForAutoGen, } - if !ndp.nic.hasPermanentAddrLocked(generatedAddr.Address) { + if !ndp.ep.hasPermanentAddressRLocked(generatedAddr.Address) { break } state.stableAddr.localGenerationFailures++ } - if ref := ndp.addSLAACAddr(generatedAddr, slaac, time.Since(state.preferredUntil) >= 0 /* deprecated */); ref != nil { - state.stableAddr.ref = ref + if addressEndpoint := ndp.addAndAcquireSLAACAddr(generatedAddr, stack.AddressConfigSlaac, time.Since(state.preferredUntil) >= 0 /* deprecated */); addressEndpoint != nil { + state.stableAddr.addressEndpoint = addressEndpoint state.generationAttempts++ return true } @@ -1309,10 +1326,9 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt // regenerateSLAACAddr regenerates an address for a SLAAC prefix. // -// If generating a new address for the prefix fails, the prefix will be -// invalidated. +// If generating a new address for the prefix fails, the prefix is invalidated. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) regenerateSLAACAddr(prefix tcpip.Subnet) { state, ok := ndp.slaacPrefixes[prefix] if !ok { @@ -1332,7 +1348,7 @@ func (ndp *ndpState) regenerateSLAACAddr(prefix tcpip.Subnet) { // generateTempSLAACAddr generates a new temporary SLAAC address. // -// If resetGenAttempts is true, the prefix's generation counter will be reset. +// If resetGenAttempts is true, the prefix's generation counter is reset. // // Returns true if a new address was generated. func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *slaacPrefixState, resetGenAttempts bool) bool { @@ -1353,7 +1369,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla return false } - stableAddr := prefixState.stableAddr.ref.address() + stableAddr := prefixState.stableAddr.addressEndpoint.AddressWithPrefix().Address now := time.Now() // As per RFC 4941 section 3.3 step 4, the valid lifetime of a temporary @@ -1392,7 +1408,8 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla return false } - // Attempt to generate a new address that is not already assigned to the NIC. + // Attempt to generate a new address that is not already assigned to the IPv6 + // endpoint. var generatedAddr tcpip.AddressWithPrefix for i := 0; ; i++ { // If we were unable to generate an address after the maximum SLAAC address @@ -1402,7 +1419,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla } generatedAddr = header.GenerateTempIPv6SLAACAddr(ndp.temporaryIIDHistory[:], stableAddr) - if !ndp.nic.hasPermanentAddrLocked(generatedAddr.Address) { + if !ndp.ep.hasPermanentAddressRLocked(generatedAddr.Address) { break } } @@ -1410,13 +1427,13 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla // As per RFC RFC 4941 section 3.3 step 5, we MUST NOT create a temporary // address with a zero preferred lifetime. The checks above ensure this // so we know the address is not deprecated. - ref := ndp.addSLAACAddr(generatedAddr, slaacTemp, false /* deprecated */) - if ref == nil { + addressEndpoint := ndp.addAndAcquireSLAACAddr(generatedAddr, stack.AddressConfigSlaacTemp, false /* deprecated */) + if addressEndpoint == nil { return false } state := tempSLAACAddrState{ - deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + deprecationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr)) @@ -1427,9 +1444,9 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla panic(fmt.Sprintf("ndp: must have a tempAddr entry to deprecate temporary address %s", generatedAddr)) } - ndp.deprecateSLAACAddress(tempAddrState.ref) + ndp.deprecateSLAACAddress(tempAddrState.addressEndpoint) }), - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr)) @@ -1442,7 +1459,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState) }), - regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + regenJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr)) @@ -1465,8 +1482,8 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla prefixState.tempAddrs[generatedAddr.Address] = tempAddrState ndp.slaacPrefixes[prefix] = prefixState }), - createdAt: now, - ref: ref, + createdAt: now, + addressEndpoint: addressEndpoint, } state.deprecationJob.Schedule(pl) @@ -1481,7 +1498,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla // regenerateTempSLAACAddr regenerates a temporary address for a SLAAC prefix. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) regenerateTempSLAACAddr(prefix tcpip.Subnet, resetGenAttempts bool) { state, ok := ndp.slaacPrefixes[prefix] if !ok { @@ -1496,14 +1513,14 @@ func (ndp *ndpState) regenerateTempSLAACAddr(prefix tcpip.Subnet, resetGenAttemp // // pl is the new preferred lifetime. vl is the new valid lifetime. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixState *slaacPrefixState, pl, vl time.Duration) { // If the preferred lifetime is zero, then the prefix should be deprecated. deprecated := pl == 0 if deprecated { - ndp.deprecateSLAACAddress(prefixState.stableAddr.ref) + ndp.deprecateSLAACAddress(prefixState.stableAddr.addressEndpoint) } else { - prefixState.stableAddr.ref.deprecated = false + prefixState.stableAddr.addressEndpoint.SetDeprecated(false) } // If prefix was preferred for some finite lifetime before, cancel the @@ -1565,7 +1582,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // If DAD is not yet complete on the stable address, there is no need to do // work with temporary addresses. - if prefixState.stableAddr.ref.getKind() != permanent { + if prefixState.stableAddr.addressEndpoint.GetKind() != stack.Permanent { return } @@ -1608,9 +1625,9 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat newPreferredLifetime := preferredUntil.Sub(now) tempAddrState.deprecationJob.Cancel() if newPreferredLifetime <= 0 { - ndp.deprecateSLAACAddress(tempAddrState.ref) + ndp.deprecateSLAACAddress(tempAddrState.addressEndpoint) } else { - tempAddrState.ref.deprecated = false + tempAddrState.addressEndpoint.SetDeprecated(false) tempAddrState.deprecationJob.Schedule(newPreferredLifetime) } @@ -1635,8 +1652,8 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // due to an update in preferred lifetime. // // If each temporay address has already been regenerated, no new temporary - // address will be generated. To ensure continuation of temporary SLAAC - // addresses, we manually try to regenerate an address here. + // address is generated. To ensure continuation of temporary SLAAC addresses, + // we manually try to regenerate an address here. if len(regenForAddr) != 0 || allAddressesRegenerated { // Reset the generation attempts counter as we are starting the generation // of a new address for the SLAAC prefix. @@ -1647,57 +1664,58 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } } -// deprecateSLAACAddress marks ref as deprecated and notifies the stack's NDP -// dispatcher that ref has been deprecated. +// deprecateSLAACAddress marks the address as deprecated and notifies the NDP +// dispatcher that address has been deprecated. // -// deprecateSLAACAddress does nothing if ref is already deprecated. +// deprecateSLAACAddress does nothing if the address is already deprecated. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) deprecateSLAACAddress(ref *referencedNetworkEndpoint) { - if ref.deprecated { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) deprecateSLAACAddress(addressEndpoint stack.AddressEndpoint) { + if addressEndpoint.Deprecated() { return } - ref.deprecated = true - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnAutoGenAddressDeprecated(ndp.nic.ID(), ref.addrWithPrefix()) + addressEndpoint.SetDeprecated(true) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressDeprecated(ndp.ep.nic.ID(), addressEndpoint.AddressWithPrefix()) } } // invalidateSLAACPrefix invalidates a SLAAC prefix. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateSLAACPrefix(prefix tcpip.Subnet, state slaacPrefixState) { - if r := state.stableAddr.ref; r != nil { + ndp.cleanupSLAACPrefixResources(prefix, state) + + if addressEndpoint := state.stableAddr.addressEndpoint; addressEndpoint != nil { // Since we are already invalidating the prefix, do not invalidate the // prefix when removing the address. - if err := ndp.nic.removePermanentIPv6EndpointLocked(r, false /* allowSLAACInvalidation */); err != nil { - panic(fmt.Sprintf("ndp: error removing stable SLAAC address %s: %s", r.addrWithPrefix(), err)) + if err := ndp.ep.removePermanentEndpointLocked(addressEndpoint, false /* allowSLAACInvalidation */); err != nil { + panic(fmt.Sprintf("ndp: error removing stable SLAAC address %s: %s", addressEndpoint.AddressWithPrefix(), err)) } } - - ndp.cleanupSLAACPrefixResources(prefix, state) } // cleanupSLAACAddrResourcesAndNotify cleans up an invalidated SLAAC address's // resources. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidatePrefix bool) { - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), addr) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressInvalidated(ndp.ep.nic.ID(), addr) } prefix := addr.Subnet() state, ok := ndp.slaacPrefixes[prefix] - if !ok || state.stableAddr.ref == nil || addr.Address != state.stableAddr.ref.address() { + if !ok || state.stableAddr.addressEndpoint == nil || addr.Address != state.stableAddr.addressEndpoint.AddressWithPrefix().Address { return } if !invalidatePrefix { // If the prefix is not being invalidated, disassociate the address from the // prefix and do nothing further. - state.stableAddr.ref = nil + state.stableAddr.addressEndpoint.DecRef() + state.stableAddr.addressEndpoint = nil ndp.slaacPrefixes[prefix] = state return } @@ -1709,14 +1727,17 @@ func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPr // // Panics if the SLAAC prefix is not known. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaacPrefixState) { // Invalidate all temporary addresses. for tempAddr, tempAddrState := range state.tempAddrs { ndp.invalidateTempSLAACAddr(state.tempAddrs, tempAddr, tempAddrState) } - state.stableAddr.ref = nil + if state.stableAddr.addressEndpoint != nil { + state.stableAddr.addressEndpoint.DecRef() + state.stableAddr.addressEndpoint = nil + } state.deprecationJob.Cancel() state.invalidationJob.Cancel() delete(ndp.slaacPrefixes, prefix) @@ -1724,12 +1745,12 @@ func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaa // invalidateTempSLAACAddr invalidates a temporary SLAAC address. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateTempSLAACAddr(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { // Since we are already invalidating the address, do not invalidate the // address when removing the address. - if err := ndp.nic.removePermanentIPv6EndpointLocked(tempAddrState.ref, false /* allowSLAACInvalidation */); err != nil { - panic(fmt.Sprintf("error removing temporary SLAAC address %s: %s", tempAddrState.ref.addrWithPrefix(), err)) + if err := ndp.ep.removePermanentEndpointLocked(tempAddrState.addressEndpoint, false /* allowSLAACInvalidation */); err != nil { + panic(fmt.Sprintf("error removing temporary SLAAC address %s: %s", tempAddrState.addressEndpoint.AddressWithPrefix(), err)) } ndp.cleanupTempSLAACAddrResources(tempAddrs, tempAddr, tempAddrState) @@ -1738,10 +1759,10 @@ func (ndp *ndpState) invalidateTempSLAACAddr(tempAddrs map[tcpip.Address]tempSLA // cleanupTempSLAACAddrResourcesAndNotify cleans up an invalidated temporary // SLAAC address's resources from ndp. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidateAddr bool) { - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), addr) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressInvalidated(ndp.ep.nic.ID(), addr) } if !invalidateAddr { @@ -1765,35 +1786,29 @@ func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWi // cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's // jobs and entry. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { + tempAddrState.addressEndpoint.DecRef() + tempAddrState.addressEndpoint = nil tempAddrState.deprecationJob.Cancel() tempAddrState.invalidationJob.Cancel() tempAddrState.regenJob.Cancel() delete(tempAddrs, tempAddr) } -// cleanupState cleans up ndp's state. -// -// If hostOnly is true, then only host-specific state will be cleaned up. +// removeSLAACAddresses removes all SLAAC addresses. // -// cleanupState MUST be called with hostOnly set to true when ndp's NIC is -// transitioning from a host to a router. This function will invalidate all -// discovered on-link prefixes, discovered routers, and auto-generated -// addresses. -// -// If hostOnly is true, then the link-local auto-generated address will not be -// invalidated as routers are also expected to generate a link-local address. +// If keepLinkLocal is false, the SLAAC generated link-local address is removed. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) cleanupState(hostOnly bool) { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) removeSLAACAddresses(keepLinkLocal bool) { linkLocalSubnet := header.IPv6LinkLocalPrefix.Subnet() - linkLocalPrefixes := 0 + var linkLocalPrefixes int for prefix, state := range ndp.slaacPrefixes { // RFC 4862 section 5 states that routers are also expected to generate a // link-local address so we do not invalidate them if we are cleaning up // host-only state. - if hostOnly && prefix == linkLocalSubnet { + if keepLinkLocal && prefix == linkLocalSubnet { linkLocalPrefixes++ continue } @@ -1804,6 +1819,21 @@ func (ndp *ndpState) cleanupState(hostOnly bool) { if got := len(ndp.slaacPrefixes); got != linkLocalPrefixes { panic(fmt.Sprintf("ndp: still have non-linklocal SLAAC prefixes after cleaning up; found = %d prefixes, of which %d are link-local", got, linkLocalPrefixes)) } +} + +// cleanupState cleans up ndp's state. +// +// If hostOnly is true, then only host-specific state is cleaned up. +// +// This function invalidates all discovered on-link prefixes, discovered +// routers, and auto-generated addresses. +// +// If hostOnly is true, then the link-local auto-generated address aren't +// invalidated as routers are also expected to generate a link-local address. +// +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) cleanupState(hostOnly bool) { + ndp.removeSLAACAddresses(hostOnly /* keepLinkLocal */) for prefix := range ndp.onLinkPrefixes { ndp.invalidateOnLinkPrefix(prefix) @@ -1827,7 +1857,7 @@ func (ndp *ndpState) cleanupState(hostOnly bool) { // startSolicitingRouters starts soliciting routers, as per RFC 4861 section // 6.3.7. If routers are already being solicited, this function does nothing. // -// The NIC ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) startSolicitingRouters() { if ndp.rtrSolicit.timer != nil { // We are already soliciting routers. @@ -1848,27 +1878,37 @@ func (ndp *ndpState) startSolicitingRouters() { var done bool ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() { - ndp.nic.mu.Lock() + ndp.rtrSolicit.timer = ndp.ep.protocol.stack.Clock().AfterFunc(delay, func() { + ndp.ep.mu.Lock() if done { // If we reach this point, it means that the RS timer fired after another - // goroutine already obtained the NIC lock and stopped solicitations. - // Simply return here and do nothing further. - ndp.nic.mu.Unlock() + // goroutine already obtained the IPv6 endpoint lock and stopped + // solicitations. Simply return here and do nothing further. + ndp.ep.mu.Unlock() return } // As per RFC 4861 section 4.1, the source of the RS is an address assigned // to the sending interface, or the unspecified address if no address is // assigned to the sending interface. - ref := ndp.nic.primaryIPv6EndpointRLocked(header.IPv6AllRoutersMulticastAddress) - if ref == nil { - ref = ndp.nic.getRefOrCreateTempLocked(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint) + addressEndpoint := ndp.ep.acquirePrimaryAddressRLocked(header.IPv6AllRoutersMulticastAddress, false) + if addressEndpoint == nil { + // Incase this ends up creating a new temporary address, we need to hold + // onto the endpoint until a route is obtained. If we decrement the + // reference count before obtaing a route, the address's resources would + // be released and attempting to obtain a route after would fail. Once a + // route is obtainted, it is safe to decrement the reference count since + // obtaining a route increments the address's reference count. + addressEndpoint = ndp.ep.acquireAddressOrCreateTempLocked(header.IPv6Any, true /* createTemp */, stack.NeverPrimaryEndpoint) } - ndp.nic.mu.Unlock() + ndp.ep.mu.Unlock() - localAddr := ref.address() - r := makeRoute(header.IPv6ProtocolNumber, localAddr, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false) + localAddr := addressEndpoint.AddressWithPrefix().Address + r, err := ndp.ep.protocol.stack.FindRoute(ndp.ep.nic.ID(), localAddr, header.IPv6AllRoutersMulticastAddress, ProtocolNumber, false /* multicastLoop */) + addressEndpoint.DecRef() + if err != nil { + return + } defer r.Release() // Route should resolve immediately since @@ -1876,15 +1916,16 @@ func (ndp *ndpState) startSolicitingRouters() { // remote link address can be calculated without a resolution process. if c, err := r.Resolve(nil); err != nil { // Do not consider the NIC being unknown or disabled as a fatal error. - // Since this method is required to be called when the NIC is not locked, - // the NIC could have been disabled or removed by another goroutine. + // Since this method is required to be called when the IPv6 endpoint is + // not locked, the IPv6 endpoint could have been disabled or removed by + // another goroutine. if err == tcpip.ErrUnknownNICID || err == tcpip.ErrInvalidEndpointState { return } - panic(fmt.Sprintf("ndp: error when resolving route to send NDP RS (%s -> %s on NIC(%d)): %s", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID(), err)) + panic(fmt.Sprintf("ndp: error when resolving route to send NDP RS (%s -> %s on NIC(%d)): %s", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.ep.nic.ID(), err)) } else if c != nil { - panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP RS (%s -> %s on NIC(%d))", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID())) + panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP RS (%s -> %s on NIC(%d))", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.ep.nic.ID())) } // As per RFC 4861 section 4.1, an NDP RS SHOULD include the source @@ -1907,21 +1948,20 @@ func (ndp *ndpState) startSolicitingRouters() { rs.Options().Serialize(optsSerializer) icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - pkt := NewPacketBuffer(PacketBufferOptions{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), Data: buffer.View(icmpData).ToVectorisedView(), }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, - NetworkHeaderParams{ + stack.NetworkHeaderParams{ Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, - TOS: DefaultTOS, }, pkt, ); err != nil { sent.Dropped.Increment() - log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) + log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.ep.nic.ID(), err) // Don't send any more messages if we had an error. remaining = 0 } else { @@ -1929,19 +1969,19 @@ func (ndp *ndpState) startSolicitingRouters() { remaining-- } - ndp.nic.mu.Lock() + ndp.ep.mu.Lock() if done || remaining == 0 { ndp.rtrSolicit.timer = nil ndp.rtrSolicit.done = nil } else if ndp.rtrSolicit.timer != nil { // Note, we need to explicitly check to make sure that // the timer field is not nil because if it was nil but - // we still reached this point, then we know the NIC + // we still reached this point, then we know the IPv6 endpoint // was requested to stop soliciting routers so we don't // need to send the next Router Solicitation message. ndp.rtrSolicit.timer.Reset(ndp.configs.RtrSolicitationInterval) } - ndp.nic.mu.Unlock() + ndp.ep.mu.Unlock() }) } @@ -1949,7 +1989,7 @@ func (ndp *ndpState) startSolicitingRouters() { // stopSolicitingRouters stops soliciting routers. If routers are not currently // being solicited, this function does nothing. // -// The NIC ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) stopSolicitingRouters() { if ndp.rtrSolicit.timer == nil { // Nothing to do. @@ -1965,7 +2005,7 @@ func (ndp *ndpState) stopSolicitingRouters() { // initializeTempAddrState initializes state related to temporary SLAAC // addresses. func (ndp *ndpState) initializeTempAddrState() { - header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.nic.stack.tempIIDSeed, ndp.nic.ID()) + header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.ep.protocol.tempIIDSeed, ndp.ep.nic.ID()) if MaxDesyncFactor != 0 { ndp.temporaryAddressDesyncFactor = time.Duration(rand.Int63n(int64(MaxDesyncFactor))) diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 7434df4a1..25464a03a 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -17,6 +17,7 @@ package ipv6 import ( "strings" "testing" + "time" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" @@ -35,8 +36,8 @@ func setupStackAndEndpoint(t *testing.T, llladdr, rlladdr tcpip.Address, useNeig t.Helper() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, UseNeighborCache: useNeighborCache, }) @@ -65,10 +66,94 @@ func setupStackAndEndpoint(t *testing.T, llladdr, rlladdr tcpip.Address, useNeig t.Fatalf("cannot find protocol instance for network protocol %d", ProtocolNumber) } - ep := netProto.NewEndpoint(0, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}, nil, s) + ep := netProto.NewEndpoint(&testInterface{}, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}) + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + t.Cleanup(ep.Close) + return s, ep } +var _ NDPDispatcher = (*testNDPDispatcher)(nil) + +// testNDPDispatcher is an NDPDispatcher only allows default router discovery. +type testNDPDispatcher struct { + addr tcpip.Address +} + +func (*testNDPDispatcher) OnDuplicateAddressDetectionStatus(tcpip.NICID, tcpip.Address, bool, *tcpip.Error) { +} + +func (t *testNDPDispatcher) OnDefaultRouterDiscovered(_ tcpip.NICID, addr tcpip.Address) bool { + t.addr = addr + return true +} + +func (t *testNDPDispatcher) OnDefaultRouterInvalidated(_ tcpip.NICID, addr tcpip.Address) { + t.addr = addr +} + +func (*testNDPDispatcher) OnOnLinkPrefixDiscovered(tcpip.NICID, tcpip.Subnet) bool { + return false +} + +func (*testNDPDispatcher) OnOnLinkPrefixInvalidated(tcpip.NICID, tcpip.Subnet) { +} + +func (*testNDPDispatcher) OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool { + return false +} + +func (*testNDPDispatcher) OnAutoGenAddressDeprecated(tcpip.NICID, tcpip.AddressWithPrefix) { +} + +func (*testNDPDispatcher) OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) { +} + +func (*testNDPDispatcher) OnRecursiveDNSServerOption(tcpip.NICID, []tcpip.Address, time.Duration) { +} + +func (*testNDPDispatcher) OnDNSSearchListOption(tcpip.NICID, []string, time.Duration) { +} + +func (*testNDPDispatcher) OnDHCPv6Configuration(tcpip.NICID, DHCPv6ConfigurationFromNDPRA) { +} + +func TestStackNDPEndpointInvalidateDefaultRouter(t *testing.T) { + var ndpDisp testNDPDispatcher + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocolWithOptions(Options{ + NDPDisp: &ndpDisp, + })}, + }) + + if err := s.CreateNIC(nicID, &stubLinkEndpoint{}); err != nil { + t.Fatalf("s.CreateNIC(%d, _): %s", nicID, err) + } + + ep, err := s.GetNetworkEndpoint(nicID, ProtocolNumber) + if err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, ProtocolNumber, err) + } + + ipv6EP := ep.(*endpoint) + ipv6EP.mu.Lock() + ipv6EP.mu.ndp.rememberDefaultRouter(lladdr1, time.Hour) + ipv6EP.mu.Unlock() + + if ndpDisp.addr != lladdr1 { + t.Fatalf("got ndpDisp.addr = %s, want = %s", ndpDisp.addr, lladdr1) + } + + ndpDisp.addr = "" + ndpEP := ep.(stack.NDPEndpoint) + ndpEP.InvalidateDefaultRouter(lladdr1) + if ndpDisp.addr != lladdr1 { + t.Fatalf("got ndpDisp.addr = %s, want = %s", ndpDisp.addr, lladdr1) + } +} + // TestNeighorSolicitationWithSourceLinkLayerOption tests that receiving a // valid NDP NS message with the Source Link Layer Address option results in a // new entry in the link address cache for the sender of the message. @@ -98,7 +183,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) e := channel.New(0, 1280, linkAddr0) if err := s.CreateNIC(nicID, e); err != nil { @@ -202,7 +287,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOptionUsingNeighborCache(t *testi for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: true, }) e := channel.New(0, 1280, linkAddr0) @@ -475,7 +560,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: stackTyp.useNeighborCache, }) e := channel.New(1, 1280, nicLinkAddr) @@ -596,7 +681,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) e := channel.New(0, 1280, linkAddr0) if err := s.CreateNIC(nicID, e); err != nil { @@ -707,7 +792,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOptionUsingNeighborCache(t *test for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: true, }) e := channel.New(0, 1280, linkAddr0) @@ -1172,7 +1257,7 @@ func TestRouterAdvertValidation(t *testing.T) { e := channel.New(10, 1280, linkAddr1) e.LinkEPCapabilities |= stack.CapabilityResolutionRequired s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: stackTyp.useNeighborCache, }) diff --git a/pkg/tcpip/sample/tun_tcp_connect/main.go b/pkg/tcpip/sample/tun_tcp_connect/main.go index 91fc26722..51d428049 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/main.go +++ b/pkg/tcpip/sample/tun_tcp_connect/main.go @@ -127,8 +127,8 @@ func main() { // Create the stack with ipv4 and tcp protocols, then add a tun-based // NIC and ipv4 address. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) mtu, err := rawfile.GetMTU(tunName) diff --git a/pkg/tcpip/sample/tun_tcp_echo/main.go b/pkg/tcpip/sample/tun_tcp_echo/main.go index 3f58a15ea..8e0ee1cd7 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/main.go +++ b/pkg/tcpip/sample/tun_tcp_echo/main.go @@ -112,8 +112,8 @@ func main() { // Create the stack with ip and tcp protocols, then add a tun-based // NIC and address. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol, arp.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) mtu, err := rawfile.GetMTU(tunName) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 7f1d79115..2eaeab779 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -54,8 +54,8 @@ go_template_instance( go_library( name = "stack", srcs = [ + "addressable_endpoint_state.go", "conntrack.go", - "dhcpv6configurationfromndpra_string.go", "forwarder.go", "headertype_string.go", "icmp_rate_limit.go", @@ -65,7 +65,6 @@ go_library( "iptables_types.go", "linkaddrcache.go", "linkaddrentry_list.go", - "ndp.go", "neighbor_cache.go", "neighbor_entry.go", "neighbor_entry_list.go", @@ -106,6 +105,7 @@ go_test( name = "stack_x_test", size = "medium", srcs = [ + "addressable_endpoint_state_test.go", "ndp_test.go", "nud_test.go", "stack_test.go", @@ -116,6 +116,7 @@ go_test( deps = [ ":stack", "//pkg/rand", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/checker", diff --git a/pkg/tcpip/stack/addressable_endpoint_state.go b/pkg/tcpip/stack/addressable_endpoint_state.go new file mode 100644 index 000000000..270ac4977 --- /dev/null +++ b/pkg/tcpip/stack/addressable_endpoint_state.go @@ -0,0 +1,717 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" +) + +var _ GroupAddressableEndpoint = (*AddressableEndpointState)(nil) +var _ AddressableEndpoint = (*AddressableEndpointState)(nil) + +// AddressableEndpointState is an implementation of an AddressableEndpoint. +type AddressableEndpointState struct { + networkEndpoint NetworkEndpoint + + // Lock ordering (from outer to inner lock ordering): + // + // AddressableEndpointState.mu + // addressState.mu + mu struct { + sync.RWMutex + + endpoints map[tcpip.Address]*addressState + primary []*addressState + + // groups holds the mapping between group addresses and the number of times + // they have been joined. + groups map[tcpip.Address]uint32 + } +} + +// Init initializes the AddressableEndpointState with networkEndpoint. +// +// Must be called before calling any other function on m. +func (a *AddressableEndpointState) Init(networkEndpoint NetworkEndpoint) { + a.networkEndpoint = networkEndpoint + + a.mu.Lock() + defer a.mu.Unlock() + a.mu.endpoints = make(map[tcpip.Address]*addressState) + a.mu.groups = make(map[tcpip.Address]uint32) +} + +// ReadOnlyAddressableEndpointState provides read-only access to an +// AddressableEndpointState. +type ReadOnlyAddressableEndpointState struct { + inner *AddressableEndpointState +} + +// AddrOrMatching returns an endpoint for the passed address that is consisdered +// bound to the wrapped AddressableEndpointState. +// +// If addr is an exact match with an existing address, that address is returned. +// Otherwise, f is called with each address and the address that f returns true +// for is returned. +// +// Returns nil of no address matches. +func (m ReadOnlyAddressableEndpointState) AddrOrMatching(addr tcpip.Address, spoofingOrPrimiscuous bool, f func(AddressEndpoint) bool) AddressEndpoint { + m.inner.mu.RLock() + defer m.inner.mu.RUnlock() + + if ep, ok := m.inner.mu.endpoints[addr]; ok { + if ep.IsAssigned(spoofingOrPrimiscuous) && ep.IncRef() { + return ep + } + } + + for _, ep := range m.inner.mu.endpoints { + if ep.IsAssigned(spoofingOrPrimiscuous) && f(ep) && ep.IncRef() { + return ep + } + } + + return nil +} + +// Lookup returns the AddressEndpoint for the passed address. +// +// Returns nil if the passed address is not associated with the +// AddressableEndpointState. +func (m ReadOnlyAddressableEndpointState) Lookup(addr tcpip.Address) AddressEndpoint { + m.inner.mu.RLock() + defer m.inner.mu.RUnlock() + + ep, ok := m.inner.mu.endpoints[addr] + if !ok { + return nil + } + return ep +} + +// ForEach calls f for each address pair. +// +// If f returns false, f is no longer be called. +func (m ReadOnlyAddressableEndpointState) ForEach(f func(AddressEndpoint) bool) { + m.inner.mu.RLock() + defer m.inner.mu.RUnlock() + + for _, ep := range m.inner.mu.endpoints { + if !f(ep) { + return + } + } +} + +// ForEachPrimaryEndpoint calls f for each primary address. +// +// If f returns false, f is no longer be called. +func (m ReadOnlyAddressableEndpointState) ForEachPrimaryEndpoint(f func(AddressEndpoint)) { + m.inner.mu.RLock() + defer m.inner.mu.RUnlock() + for _, ep := range m.inner.mu.primary { + f(ep) + } +} + +// ReadOnly returns a readonly reference to a. +func (a *AddressableEndpointState) ReadOnly() ReadOnlyAddressableEndpointState { + return ReadOnlyAddressableEndpointState{inner: a} +} + +func (a *AddressableEndpointState) releaseAddressState(addrState *addressState) { + a.mu.Lock() + defer a.mu.Unlock() + a.releaseAddressStateLocked(addrState) +} + +// releaseAddressState removes addrState from s's address state (primary and endpoints list). +// +// Preconditions: a.mu must be write locked. +func (a *AddressableEndpointState) releaseAddressStateLocked(addrState *addressState) { + oldPrimary := a.mu.primary + for i, s := range a.mu.primary { + if s == addrState { + a.mu.primary = append(a.mu.primary[:i], a.mu.primary[i+1:]...) + oldPrimary[len(oldPrimary)-1] = nil + break + } + } + delete(a.mu.endpoints, addrState.addr.Address) +} + +// AddAndAcquirePermanentAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior, configType AddressConfigType, deprecated bool) (AddressEndpoint, *tcpip.Error) { + a.mu.Lock() + defer a.mu.Unlock() + ep, err := a.addAndAcquireAddressLocked(addr, peb, configType, deprecated, true /* permanent */) + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since addAndAcquireAddressLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil, err + } + return ep, err +} + +// AddAndAcquireTemporaryAddress adds a temporary address. +// +// Returns tcpip.ErrDuplicateAddress if the address exists. +// +// The temporary address's endpoint is acquired and returned. +func (a *AddressableEndpointState) AddAndAcquireTemporaryAddress(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior) (AddressEndpoint, *tcpip.Error) { + a.mu.Lock() + defer a.mu.Unlock() + ep, err := a.addAndAcquireAddressLocked(addr, peb, AddressConfigStatic, false /* deprecated */, false /* permanent */) + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since addAndAcquireAddressLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil, err + } + return ep, err +} + +// addAndAcquireAddressLocked adds, acquires and returns a permanent or +// temporary address. +// +// If the addressable endpoint already has the address in a non-permanent state, +// and addAndAcquireAddressLocked is adding a permanent address, that address is +// promoted in place and its properties set to the properties provided. If the +// address already exists in any other state, then tcpip.ErrDuplicateAddress is +// returned, regardless the kind of address that is being added. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) addAndAcquireAddressLocked(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior, configType AddressConfigType, deprecated, permanent bool) (*addressState, *tcpip.Error) { + // attemptAddToPrimary is false when the address is already in the primary + // address list. + attemptAddToPrimary := true + addrState, ok := a.mu.endpoints[addr.Address] + if ok { + if !permanent { + // We are adding a non-permanent address but the address exists. No need + // to go any further since we can only promote existing temporary/expired + // addresses to permanent. + return nil, tcpip.ErrDuplicateAddress + } + + addrState.mu.Lock() + if addrState.mu.kind.IsPermanent() { + addrState.mu.Unlock() + // We are adding a permanent address but a permanent address already + // exists. + return nil, tcpip.ErrDuplicateAddress + } + + if addrState.mu.refs == 0 { + panic(fmt.Sprintf("found an address that should have been released (ref count == 0); address = %s", addrState.addr)) + } + + // We now promote the address. + for i, s := range a.mu.primary { + if s == addrState { + switch peb { + case CanBePrimaryEndpoint: + // The address is already in the primary address list. + attemptAddToPrimary = false + case FirstPrimaryEndpoint: + if i == 0 { + // The address is already first in the primary address list. + attemptAddToPrimary = false + } else { + a.mu.primary = append(a.mu.primary[:i], a.mu.primary[i+1:]...) + } + case NeverPrimaryEndpoint: + a.mu.primary = append(a.mu.primary[:i], a.mu.primary[i+1:]...) + default: + panic(fmt.Sprintf("unrecognized primary endpoint behaviour = %d", peb)) + } + break + } + } + } + + if addrState == nil { + addrState = &addressState{ + addressableEndpointState: a, + addr: addr, + } + a.mu.endpoints[addr.Address] = addrState + addrState.mu.Lock() + // We never promote an address to temporary - it can only be added as such. + // If we are actaully adding a permanent address, it is promoted below. + addrState.mu.kind = Temporary + } + + // At this point we have an address we are either promoting from an expired or + // temporary address to permanent, promoting an expired address to temporary, + // or we are adding a new temporary or permanent address. + // + // The address MUST be write locked at this point. + defer addrState.mu.Unlock() + + if permanent { + if addrState.mu.kind.IsPermanent() { + panic(fmt.Sprintf("only non-permanent addresses should be promoted to permanent; address = %s", addrState.addr)) + } + + // Primary addresses are biased by 1. + addrState.mu.refs++ + addrState.mu.kind = Permanent + } + // Acquire the address before returning it. + addrState.mu.refs++ + addrState.mu.deprecated = deprecated + addrState.mu.configType = configType + + if attemptAddToPrimary { + switch peb { + case NeverPrimaryEndpoint: + case CanBePrimaryEndpoint: + a.mu.primary = append(a.mu.primary, addrState) + case FirstPrimaryEndpoint: + if cap(a.mu.primary) == len(a.mu.primary) { + a.mu.primary = append([]*addressState{addrState}, a.mu.primary...) + } else { + // Shift all the endpoints by 1 to make room for the new address at the + // front. We could have just created a new slice but this saves + // allocations when the slice has capacity for the new address. + primaryCount := len(a.mu.primary) + a.mu.primary = append(a.mu.primary, nil) + if n := copy(a.mu.primary[1:], a.mu.primary); n != primaryCount { + panic(fmt.Sprintf("copied %d elements; expected = %d elements", n, primaryCount)) + } + a.mu.primary[0] = addrState + } + default: + panic(fmt.Sprintf("unrecognized primary endpoint behaviour = %d", peb)) + } + } + + return addrState, nil +} + +// RemovePermanentAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) RemovePermanentAddress(addr tcpip.Address) *tcpip.Error { + a.mu.Lock() + defer a.mu.Unlock() + + if _, ok := a.mu.groups[addr]; ok { + panic(fmt.Sprintf("group address = %s must be removed with LeaveGroup", addr)) + } + + return a.removePermanentAddressLocked(addr) +} + +// removePermanentAddressLocked is like RemovePermanentAddress but with locking +// requirements. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { + addrState, ok := a.mu.endpoints[addr] + if !ok { + return tcpip.ErrBadLocalAddress + } + + return a.removePermanentEndpointLocked(addrState) +} + +// RemovePermanentEndpoint removes the passed endpoint if it is associated with +// a and permanent. +func (a *AddressableEndpointState) RemovePermanentEndpoint(ep AddressEndpoint) *tcpip.Error { + addrState, ok := ep.(*addressState) + if !ok || addrState.addressableEndpointState != a { + return tcpip.ErrInvalidEndpointState + } + + return a.removePermanentEndpointLocked(addrState) +} + +// removePermanentAddressLocked is like RemovePermanentAddress but with locking +// requirements. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) removePermanentEndpointLocked(addrState *addressState) *tcpip.Error { + if !addrState.GetKind().IsPermanent() { + return tcpip.ErrBadLocalAddress + } + + addrState.SetKind(PermanentExpired) + a.decAddressRefLocked(addrState) + return nil +} + +// decAddressRef decrements the address's reference count and releases it once +// the reference count hits 0. +func (a *AddressableEndpointState) decAddressRef(addrState *addressState) { + a.mu.Lock() + defer a.mu.Unlock() + a.decAddressRefLocked(addrState) +} + +// decAddressRefLocked is like decAddressRef but with locking requirements. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) decAddressRefLocked(addrState *addressState) { + addrState.mu.Lock() + defer addrState.mu.Unlock() + + if addrState.mu.refs == 0 { + panic(fmt.Sprintf("attempted to decrease ref count for AddressEndpoint w/ addr = %s when it is already released", addrState.addr)) + } + + addrState.mu.refs-- + + if addrState.mu.refs != 0 { + return + } + + // A non-expired permanent address must not have its reference count dropped + // to 0. + if addrState.mu.kind.IsPermanent() { + panic(fmt.Sprintf("permanent addresses should be removed through the AddressableEndpoint: addr = %s, kind = %d", addrState.addr, addrState.mu.kind)) + } + + a.releaseAddressStateLocked(addrState) +} + +// AcquireAssignedAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint { + a.mu.Lock() + defer a.mu.Unlock() + + if addrState, ok := a.mu.endpoints[localAddr]; ok { + if !addrState.IsAssigned(allowTemp) { + return nil + } + + if !addrState.IncRef() { + panic(fmt.Sprintf("failed to increase the reference count for address = %s", addrState.addr)) + } + + return addrState + } + + if !allowTemp { + return nil + } + + addr := localAddr.WithPrefix() + ep, err := a.addAndAcquireAddressLocked(addr, tempPEB, AddressConfigStatic, false /* deprecated */, false /* permanent */) + if err != nil { + // addAndAcquireAddressLocked only returns an error if the address is + // already assigned but we just checked above if the address exists so we + // expect no error. + panic(fmt.Sprintf("a.addAndAcquireAddressLocked(%s, %d, %d, false, false): %s", addr, tempPEB, AddressConfigStatic, err)) + } + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since addAndAcquireAddressLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil + } + return ep +} + +// AcquirePrimaryAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AcquirePrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) AddressEndpoint { + a.mu.RLock() + defer a.mu.RUnlock() + + var deprecatedEndpoint *addressState + for _, ep := range a.mu.primary { + if !ep.IsAssigned(allowExpired) { + continue + } + + if !ep.Deprecated() { + if ep.IncRef() { + // ep is not deprecated, so return it immediately. + // + // If we kept track of a deprecated endpoint, decrement its reference + // count since it was incremented when we decided to keep track of it. + if deprecatedEndpoint != nil { + a.decAddressRefLocked(deprecatedEndpoint) + deprecatedEndpoint = nil + } + + return ep + } + } else if deprecatedEndpoint == nil && ep.IncRef() { + // We prefer an endpoint that is not deprecated, but we keep track of + // ep in case a doesn't have any non-deprecated endpoints. + // + // If we end up finding a more preferred endpoint, ep's reference count + // will be decremented. + deprecatedEndpoint = ep + } + } + + // a doesn't have any valid non-deprecated endpoints, so return + // deprecatedEndpoint (which may be nil if a doesn't have any valid deprecated + // endpoints either). + if deprecatedEndpoint == nil { + return nil + } + return deprecatedEndpoint +} + +// PrimaryAddresses implements AddressableEndpoint. +func (a *AddressableEndpointState) PrimaryAddresses() []tcpip.AddressWithPrefix { + a.mu.RLock() + defer a.mu.RUnlock() + + var addrs []tcpip.AddressWithPrefix + for _, ep := range a.mu.primary { + // Don't include tentative, expired or temporary endpoints + // to avoid confusion and prevent the caller from using + // those. + switch ep.GetKind() { + case PermanentTentative, PermanentExpired, Temporary: + continue + } + + addrs = append(addrs, ep.AddressWithPrefix()) + } + + return addrs +} + +// PermanentAddresses implements AddressableEndpoint. +func (a *AddressableEndpointState) PermanentAddresses() []tcpip.AddressWithPrefix { + a.mu.RLock() + defer a.mu.RUnlock() + + var addrs []tcpip.AddressWithPrefix + for _, ep := range a.mu.endpoints { + if !ep.GetKind().IsPermanent() { + continue + } + + addrs = append(addrs, ep.AddressWithPrefix()) + } + + return addrs +} + +// JoinGroup implements GroupAddressableEndpoint. +func (a *AddressableEndpointState) JoinGroup(group tcpip.Address) (bool, *tcpip.Error) { + a.mu.Lock() + defer a.mu.Unlock() + + joins, ok := a.mu.groups[group] + if !ok { + ep, err := a.addAndAcquireAddressLocked(group.WithPrefix(), NeverPrimaryEndpoint, AddressConfigStatic, false /* deprecated */, true /* permanent */) + if err != nil { + return false, err + } + // We have no need for the address endpoint. + a.decAddressRefLocked(ep) + } + + a.mu.groups[group] = joins + 1 + return !ok, nil +} + +// LeaveGroup implements GroupAddressableEndpoint. +func (a *AddressableEndpointState) LeaveGroup(group tcpip.Address) (bool, *tcpip.Error) { + a.mu.Lock() + defer a.mu.Unlock() + + joins, ok := a.mu.groups[group] + if !ok { + return false, tcpip.ErrBadLocalAddress + } + + if joins == 1 { + a.removeGroupAddressLocked(group) + delete(a.mu.groups, group) + return true, nil + } + + a.mu.groups[group] = joins - 1 + return false, nil +} + +// IsInGroup implements GroupAddressableEndpoint. +func (a *AddressableEndpointState) IsInGroup(group tcpip.Address) bool { + a.mu.RLock() + defer a.mu.RUnlock() + _, ok := a.mu.groups[group] + return ok +} + +func (a *AddressableEndpointState) removeGroupAddressLocked(group tcpip.Address) { + if err := a.removePermanentAddressLocked(group); err != nil { + // removePermanentEndpointLocked would only return an error if group is + // not bound to the addressable endpoint, but we know it MUST be assigned + // since we have group in our map of groups. + panic(fmt.Sprintf("error removing group address = %s: %s", group, err)) + } +} + +// Cleanup forcefully leaves all groups and removes all permanent addresses. +func (a *AddressableEndpointState) Cleanup() { + a.mu.Lock() + defer a.mu.Unlock() + + for group := range a.mu.groups { + a.removeGroupAddressLocked(group) + } + a.mu.groups = make(map[tcpip.Address]uint32) + + for _, ep := range a.mu.endpoints { + // removePermanentEndpointLocked returns tcpip.ErrBadLocalAddress if ep is + // not a permanent address. + if err := a.removePermanentEndpointLocked(ep); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error from removePermanentEndpointLocked(%s): %s", ep.addr, err)) + } + } +} + +var _ AddressEndpoint = (*addressState)(nil) + +// addressState holds state for an address. +type addressState struct { + addressableEndpointState *AddressableEndpointState + addr tcpip.AddressWithPrefix + + // Lock ordering (from outer to inner lock ordering): + // + // AddressableEndpointState.mu + // addressState.mu + mu struct { + sync.RWMutex + + refs uint32 + kind AddressKind + configType AddressConfigType + deprecated bool + } +} + +// NetworkEndpoint implements AddressEndpoint. +func (a *addressState) NetworkEndpoint() NetworkEndpoint { + return a.addressableEndpointState.networkEndpoint +} + +// AddressWithPrefix implements AddressEndpoint. +func (a *addressState) AddressWithPrefix() tcpip.AddressWithPrefix { + return a.addr +} + +// GetKind implements AddressEndpoint. +func (a *addressState) GetKind() AddressKind { + a.mu.RLock() + defer a.mu.RUnlock() + return a.mu.kind +} + +// SetKind implements AddressEndpoint. +func (a *addressState) SetKind(kind AddressKind) { + a.mu.Lock() + defer a.mu.Unlock() + a.mu.kind = kind +} + +// IsAssigned implements AddressEndpoint. +func (a *addressState) IsAssigned(allowExpired bool) bool { + if !a.addressableEndpointState.networkEndpoint.Enabled() { + return false + } + + switch a.GetKind() { + case PermanentTentative: + return false + case PermanentExpired: + return allowExpired + default: + return true + } +} + +// IncRef implements AddressEndpoint. +func (a *addressState) IncRef() bool { + a.mu.Lock() + defer a.mu.Unlock() + if a.mu.refs == 0 { + return false + } + + a.mu.refs++ + return true +} + +// DecRef implements AddressEndpoint. +func (a *addressState) DecRef() { + a.addressableEndpointState.decAddressRef(a) +} + +// ConfigType implements AddressEndpoint. +func (a *addressState) ConfigType() AddressConfigType { + a.mu.RLock() + defer a.mu.RUnlock() + return a.mu.configType +} + +// SetDeprecated implements AddressEndpoint. +func (a *addressState) SetDeprecated(d bool) { + a.mu.Lock() + defer a.mu.Unlock() + a.mu.deprecated = d +} + +// Deprecated implements AddressEndpoint. +func (a *addressState) Deprecated() bool { + a.mu.RLock() + defer a.mu.RUnlock() + return a.mu.deprecated +} diff --git a/pkg/tcpip/stack/addressable_endpoint_state_test.go b/pkg/tcpip/stack/addressable_endpoint_state_test.go new file mode 100644 index 000000000..de4e0d7b1 --- /dev/null +++ b/pkg/tcpip/stack/addressable_endpoint_state_test.go @@ -0,0 +1,72 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +// TestAddressableEndpointStateCleanup tests that cleaning up an addressable +// endpoint state removes permanent addresses and leaves groups. +func TestAddressableEndpointStateCleanup(t *testing.T) { + var s stack.AddressableEndpointState + s.Init(&fakeNetworkEndpoint{}) + + addr := tcpip.AddressWithPrefix{ + Address: "\x01", + PrefixLen: 8, + } + + { + ep, err := s.AddAndAcquirePermanentAddress(addr, stack.NeverPrimaryEndpoint, stack.AddressConfigStatic, false /* deprecated */) + if err != nil { + t.Fatalf("s.AddAndAcquirePermanentAddress(%s, %d, %d, false): %s", addr, stack.NeverPrimaryEndpoint, stack.AddressConfigStatic, err) + } + // We don't need the address endpoint. + ep.DecRef() + } + { + ep := s.AcquireAssignedAddress(addr.Address, false /* allowTemp */, stack.NeverPrimaryEndpoint) + if ep == nil { + t.Fatalf("got s.AcquireAssignedAddress(%s) = nil, want = non-nil", addr.Address) + } + ep.DecRef() + } + + group := tcpip.Address("\x02") + if added, err := s.JoinGroup(group); err != nil { + t.Fatalf("s.JoinGroup(%s): %s", group, err) + } else if !added { + t.Fatalf("got s.JoinGroup(%s) = false, want = true", group) + } + if !s.IsInGroup(group) { + t.Fatalf("got s.IsInGroup(%s) = false, want = true", group) + } + + s.Cleanup() + { + ep := s.AcquireAssignedAddress(addr.Address, false /* allowTemp */, stack.NeverPrimaryEndpoint) + if ep != nil { + ep.DecRef() + t.Fatalf("got s.AcquireAssignedAddress(%s) = %s, want = nil", addr.Address, ep.AddressWithPrefix()) + } + } + if s.IsInGroup(group) { + t.Fatalf("got s.IsInGroup(%s) = true, want = false", group) + } +} diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index e30927821..4e4b00a92 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -20,6 +20,7 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -45,6 +46,8 @@ const ( // use the first three: destination address, source address, and transport // protocol. They're all one byte fields to simplify parsing. type fwdTestNetworkEndpoint struct { + AddressableEndpointState + nicID tcpip.NICID proto *fwdTestNetworkProtocol dispatcher TransportDispatcher @@ -53,12 +56,18 @@ type fwdTestNetworkEndpoint struct { var _ NetworkEndpoint = (*fwdTestNetworkEndpoint)(nil) -func (f *fwdTestNetworkEndpoint) MTU() uint32 { - return f.ep.MTU() - uint32(f.MaxHeaderLength()) +func (*fwdTestNetworkEndpoint) Enable() *tcpip.Error { + return nil } -func (f *fwdTestNetworkEndpoint) NICID() tcpip.NICID { - return f.nicID +func (*fwdTestNetworkEndpoint) Enabled() bool { + return true +} + +func (*fwdTestNetworkEndpoint) Disable() {} + +func (f *fwdTestNetworkEndpoint) MTU() uint32 { + return f.ep.MTU() - uint32(f.MaxHeaderLength()) } func (*fwdTestNetworkEndpoint) DefaultTTL() uint8 { @@ -78,10 +87,6 @@ func (f *fwdTestNetworkEndpoint) PseudoHeaderChecksum(protocol tcpip.TransportPr return 0 } -func (f *fwdTestNetworkEndpoint) Capabilities() LinkEndpointCapabilities { - return f.ep.Capabilities() -} - func (f *fwdTestNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return f.proto.Number() } @@ -106,7 +111,9 @@ func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt *PacketBu return tcpip.ErrNotSupported } -func (*fwdTestNetworkEndpoint) Close() {} +func (f *fwdTestNetworkEndpoint) Close() { + f.AddressableEndpointState.Cleanup() +} // fwdTestNetworkProtocol is a network-layer protocol that implements Address // resolution. @@ -116,6 +123,11 @@ type fwdTestNetworkProtocol struct { addrResolveDelay time.Duration onLinkAddressResolved func(cache *linkAddrCache, neigh *neighborCache, addr tcpip.Address, _ tcpip.LinkAddress) onResolveStaticAddress func(tcpip.Address) (tcpip.LinkAddress, bool) + + mu struct { + sync.RWMutex + forwarding bool + } } var _ NetworkProtocol = (*fwdTestNetworkProtocol)(nil) @@ -145,17 +157,15 @@ func (*fwdTestNetworkProtocol) Parse(pkt *PacketBuffer) (tcpip.TransportProtocol return tcpip.TransportProtocolNumber(netHeader[protocolNumberOffset]), true, true } -func (*fwdTestNetworkProtocol) ReturnError(*Route, tcpip.ICMPReason, *PacketBuffer) *tcpip.Error { - return nil -} - -func (f *fwdTestNetworkProtocol) NewEndpoint(nicID tcpip.NICID, _ LinkAddressCache, _ NUDHandler, dispatcher TransportDispatcher, ep LinkEndpoint, _ *Stack) NetworkEndpoint { - return &fwdTestNetworkEndpoint{ - nicID: nicID, +func (f *fwdTestNetworkProtocol) NewEndpoint(nic NetworkInterface, _ LinkAddressCache, _ NUDHandler, dispatcher TransportDispatcher) NetworkEndpoint { + e := &fwdTestNetworkEndpoint{ + nicID: nic.ID(), proto: f, dispatcher: dispatcher, - ep: ep, + ep: nic.LinkEndpoint(), } + e.AddressableEndpointState.Init(e) + return e } func (*fwdTestNetworkProtocol) SetOption(tcpip.SettableNetworkProtocolOption) *tcpip.Error { @@ -190,6 +200,21 @@ func (*fwdTestNetworkProtocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber return fwdTestNetNumber } +// Forwarding implements stack.ForwardingNetworkProtocol. +func (f *fwdTestNetworkProtocol) Forwarding() bool { + f.mu.RLock() + defer f.mu.RUnlock() + return f.mu.forwarding + +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (f *fwdTestNetworkProtocol) SetForwarding(v bool) { + f.mu.Lock() + defer f.mu.Unlock() + f.mu.forwarding = v +} + // fwdTestPacketInfo holds all the information about an outbound packet. type fwdTestPacketInfo struct { RemoteLinkAddress tcpip.LinkAddress @@ -311,7 +336,7 @@ func (e *fwdTestLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protoco func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol, useNeighborCache bool) (ep1, ep2 *fwdTestLinkEndpoint) { // Create a stack with the network protocol and two NICs. s := New(Options{ - NetworkProtocols: []NetworkProtocol{proto}, + NetworkProtocols: []NetworkProtocolFactory{func(*Stack) NetworkProtocol { return proto }}, UseNeighborCache: useNeighborCache, }) diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 5e43a9b0b..73a01c2dd 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -150,10 +150,10 @@ type ndpDNSSLEvent struct { type ndpDHCPv6Event struct { nicID tcpip.NICID - configuration stack.DHCPv6ConfigurationFromNDPRA + configuration ipv6.DHCPv6ConfigurationFromNDPRA } -var _ stack.NDPDispatcher = (*ndpDispatcher)(nil) +var _ ipv6.NDPDispatcher = (*ndpDispatcher)(nil) // ndpDispatcher implements NDPDispatcher so tests can know when various NDP // related events happen for test purposes. @@ -170,7 +170,7 @@ type ndpDispatcher struct { dhcpv6ConfigurationC chan ndpDHCPv6Event } -// Implements stack.NDPDispatcher.OnDuplicateAddressDetectionStatus. +// Implements ipv6.NDPDispatcher.OnDuplicateAddressDetectionStatus. func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) { if n.dadC != nil { n.dadC <- ndpDADEvent{ @@ -182,7 +182,7 @@ func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, add } } -// Implements stack.NDPDispatcher.OnDefaultRouterDiscovered. +// Implements ipv6.NDPDispatcher.OnDefaultRouterDiscovered. func (n *ndpDispatcher) OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) bool { if c := n.routerC; c != nil { c <- ndpRouterEvent{ @@ -195,7 +195,7 @@ func (n *ndpDispatcher) OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip. return n.rememberRouter } -// Implements stack.NDPDispatcher.OnDefaultRouterInvalidated. +// Implements ipv6.NDPDispatcher.OnDefaultRouterInvalidated. func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) { if c := n.routerC; c != nil { c <- ndpRouterEvent{ @@ -206,7 +206,7 @@ func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip } } -// Implements stack.NDPDispatcher.OnOnLinkPrefixDiscovered. +// Implements ipv6.NDPDispatcher.OnOnLinkPrefixDiscovered. func (n *ndpDispatcher) OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool { if c := n.prefixC; c != nil { c <- ndpPrefixEvent{ @@ -219,7 +219,7 @@ func (n *ndpDispatcher) OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip return n.rememberPrefix } -// Implements stack.NDPDispatcher.OnOnLinkPrefixInvalidated. +// Implements ipv6.NDPDispatcher.OnOnLinkPrefixInvalidated. func (n *ndpDispatcher) OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) { if c := n.prefixC; c != nil { c <- ndpPrefixEvent{ @@ -261,7 +261,7 @@ func (n *ndpDispatcher) OnAutoGenAddressInvalidated(nicID tcpip.NICID, addr tcpi } } -// Implements stack.NDPDispatcher.OnRecursiveDNSServerOption. +// Implements ipv6.NDPDispatcher.OnRecursiveDNSServerOption. func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) { if c := n.rdnssC; c != nil { c <- ndpRDNSSEvent{ @@ -274,7 +274,7 @@ func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tc } } -// Implements stack.NDPDispatcher.OnDNSSearchListOption. +// Implements ipv6.NDPDispatcher.OnDNSSearchListOption. func (n *ndpDispatcher) OnDNSSearchListOption(nicID tcpip.NICID, domainNames []string, lifetime time.Duration) { if n.dnsslC != nil { n.dnsslC <- ndpDNSSLEvent{ @@ -285,8 +285,8 @@ func (n *ndpDispatcher) OnDNSSearchListOption(nicID tcpip.NICID, domainNames []s } } -// Implements stack.NDPDispatcher.OnDHCPv6Configuration. -func (n *ndpDispatcher) OnDHCPv6Configuration(nicID tcpip.NICID, configuration stack.DHCPv6ConfigurationFromNDPRA) { +// Implements ipv6.NDPDispatcher.OnDHCPv6Configuration. +func (n *ndpDispatcher) OnDHCPv6Configuration(nicID tcpip.NICID, configuration ipv6.DHCPv6ConfigurationFromNDPRA) { if c := n.dhcpv6ConfigurationC; c != nil { c <- ndpDHCPv6Event{ nicID, @@ -319,13 +319,12 @@ func TestDADDisabled(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent, 1), } - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, - } - e := channel.New(0, 1280, linkAddr1) - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + })}, + }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -413,19 +412,21 @@ func TestDADResolve(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent), } - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, - } - opts.NDPConfigs.RetransmitTimer = test.retransTimer - opts.NDPConfigs.DupAddrDetectTransmits = test.dupAddrDetectTransmits e := channelLinkWithHeaderLength{ Endpoint: channel.New(int(test.dupAddrDetectTransmits), 1280, linkAddr1), headerLength: test.linkHeaderLen, } e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + NDPConfigs: ipv6.NDPConfigurations{ + RetransmitTimer: test.retransTimer, + DupAddrDetectTransmits: test.dupAddrDetectTransmits, + }, + })}, + }) if err := s.CreateNIC(nicID, &e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -558,6 +559,26 @@ func TestDADResolve(t *testing.T) { } } +func rxNDPSolicit(e *channel.Endpoint, tgt tcpip.Address) { + hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborSolicitMinimumSize) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) + pkt.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns.SetTargetAddress(tgt) + snmc := header.SolicitedNodeAddr(tgt) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, header.IPv6Any, snmc, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: 255, + SrcAddr: header.IPv6Any, + DstAddr: snmc, + }) + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{Data: hdr.View().ToVectorisedView()})) +} + // TestDADFail tests to make sure that the DAD process fails if another node is // detected to be performing DAD on the same address (receive an NS message from // a node doing DAD for the same address), or if another node is detected to own @@ -567,39 +588,19 @@ func TestDADFail(t *testing.T) { tests := []struct { name string - makeBuf func(tgt tcpip.Address) buffer.Prependable + rxPkt func(e *channel.Endpoint, tgt tcpip.Address) getStat func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter }{ { - "RxSolicit", - func(tgt tcpip.Address) buffer.Prependable { - hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborSolicitMinimumSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) - pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) - ns.SetTargetAddress(tgt) - snmc := header.SolicitedNodeAddr(tgt) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, header.IPv6Any, snmc, buffer.VectorisedView{})) - payloadLength := hdr.UsedLength() - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) - ip.Encode(&header.IPv6Fields{ - PayloadLength: uint16(payloadLength), - NextHeader: uint8(icmp.ProtocolNumber6), - HopLimit: 255, - SrcAddr: header.IPv6Any, - DstAddr: snmc, - }) - - return hdr - - }, - func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + name: "RxSolicit", + rxPkt: rxNDPSolicit, + getStat: func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { return s.NeighborSolicit }, }, { - "RxAdvert", - func(tgt tcpip.Address) buffer.Prependable { + name: "RxAdvert", + rxPkt: func(e *channel.Endpoint, tgt tcpip.Address) { naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize hdr := buffer.NewPrependable(header.IPv6MinimumSize + naSize) pkt := header.ICMPv6(hdr.Prepend(naSize)) @@ -621,11 +622,9 @@ func TestDADFail(t *testing.T) { SrcAddr: tgt, DstAddr: header.IPv6AllNodesMulticastAddress, }) - - return hdr - + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{Data: hdr.View().ToVectorisedView()})) }, - func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + getStat: func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { return s.NeighborAdvert }, }, @@ -636,16 +635,16 @@ func TestDADFail(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent, 1), } - ndpConfigs := stack.DefaultNDPConfigurations() - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, - } - opts.NDPConfigs.RetransmitTimer = time.Second * 2 + ndpConfigs := ipv6.DefaultNDPConfigurations() + ndpConfigs.RetransmitTimer = time.Second * 2 e := channel.New(0, 1280, linkAddr1) - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + NDPConfigs: ndpConfigs, + })}, + }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -664,13 +663,8 @@ func TestDADFail(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want) } - // Receive a packet to simulate multiple nodes owning or - // attempting to own the same address. - hdr := test.makeBuf(addr1) - pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - Data: hdr.View().ToVectorisedView(), - }) - e.InjectInbound(header.IPv6ProtocolNumber, pkt) + // Receive a packet to simulate an address conflict. + test.rxPkt(e, addr1) stat := test.getStat(s.Stats().ICMP.V6PacketsReceived) if got := stat.Value(); got != 1 { @@ -754,18 +748,19 @@ func TestDADStop(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent, 1), } - ndpConfigs := stack.NDPConfigurations{ + + ndpConfigs := ipv6.NDPConfigurations{ RetransmitTimer: time.Second, DupAddrDetectTransmits: 2, } - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, - NDPConfigs: ndpConfigs, - } e := channel.New(0, 1280, linkAddr1) - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + NDPConfigs: ndpConfigs, + })}, + }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) } @@ -815,19 +810,6 @@ func TestDADStop(t *testing.T) { } } -// TestSetNDPConfigurationFailsForBadNICID tests to make sure we get an error if -// we attempt to update NDP configurations using an invalid NICID. -func TestSetNDPConfigurationFailsForBadNICID(t *testing.T) { - s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - }) - - // No NIC with ID 1 yet. - if got := s.SetNDPConfigurations(1, stack.NDPConfigurations{}); got != tcpip.ErrUnknownNICID { - t.Fatalf("got s.SetNDPConfigurations = %v, want = %s", got, tcpip.ErrUnknownNICID) - } -} - // TestSetNDPConfigurations tests that we can update and use per-interface NDP // configurations without affecting the default NDP configurations or other // interfaces' configurations. @@ -863,8 +845,9 @@ func TestSetNDPConfigurations(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + })}, }) expectDADEvent := func(nicID tcpip.NICID, addr tcpip.Address) { @@ -892,12 +875,15 @@ func TestSetNDPConfigurations(t *testing.T) { } // Update the NDP configurations on NIC(1) to use DAD. - configs := stack.NDPConfigurations{ + configs := ipv6.NDPConfigurations{ DupAddrDetectTransmits: test.dupAddrDetectTransmits, RetransmitTimer: test.retransmitTimer, } - if err := s.SetNDPConfigurations(nicID1, configs); err != nil { - t.Fatalf("got SetNDPConfigurations(%d, _) = %s", nicID1, err) + if ipv6Ep, err := s.GetNetworkEndpoint(nicID1, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID1, header.IPv6ProtocolNumber, err) + } else { + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(configs) } // Created after updating NIC(1)'s NDP configurations @@ -1113,12 +1099,13 @@ func TestNoRouterDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: handle, - DiscoverDefaultRouters: discover, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: handle, + DiscoverDefaultRouters: discover, + }, + NDPDisp: &ndpDisp, + })}, }) s.SetForwarding(ipv6.ProtocolNumber, forwarding) @@ -1151,12 +1138,13 @@ func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1192,12 +1180,13 @@ func TestRouterDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + })}, }) expectRouterEvent := func(addr tcpip.Address, discovered bool) { @@ -1285,7 +1274,7 @@ func TestRouterDiscovery(t *testing.T) { } // TestRouterDiscoveryMaxRouters tests that only -// stack.MaxDiscoveredDefaultRouters discovered routers are remembered. +// ipv6.MaxDiscoveredDefaultRouters discovered routers are remembered. func TestRouterDiscoveryMaxRouters(t *testing.T) { ndpDisp := ndpDispatcher{ routerC: make(chan ndpRouterEvent, 1), @@ -1293,12 +1282,13 @@ func TestRouterDiscoveryMaxRouters(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1306,14 +1296,14 @@ func TestRouterDiscoveryMaxRouters(t *testing.T) { } // Receive an RA from 2 more than the max number of discovered routers. - for i := 1; i <= stack.MaxDiscoveredDefaultRouters+2; i++ { + for i := 1; i <= ipv6.MaxDiscoveredDefaultRouters+2; i++ { linkAddr := []byte{2, 2, 3, 4, 5, 0} linkAddr[5] = byte(i) llAddr := header.LinkLocalAddr(tcpip.LinkAddress(linkAddr)) e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr, 5)) - if i <= stack.MaxDiscoveredDefaultRouters { + if i <= ipv6.MaxDiscoveredDefaultRouters { select { case e := <-ndpDisp.routerC: if diff := checkRouterEvent(e, llAddr, true); diff != "" { @@ -1358,12 +1348,13 @@ func TestNoPrefixDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: handle, - DiscoverOnLinkPrefixes: discover, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: handle, + DiscoverOnLinkPrefixes: discover, + }, + NDPDisp: &ndpDisp, + })}, }) s.SetForwarding(ipv6.ProtocolNumber, forwarding) @@ -1399,13 +1390,14 @@ func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: false, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: false, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1445,12 +1437,13 @@ func TestPrefixDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1545,12 +1538,13 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1621,33 +1615,34 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { } // TestPrefixDiscoveryMaxRouters tests that only -// stack.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered. +// ipv6.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered. func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { ndpDisp := ndpDispatcher{ - prefixC: make(chan ndpPrefixEvent, stack.MaxDiscoveredOnLinkPrefixes+3), + prefixC: make(chan ndpPrefixEvent, ipv6.MaxDiscoveredOnLinkPrefixes+3), rememberPrefix: true, } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: false, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: false, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(1) = %s", err) } - optSer := make(header.NDPOptionsSerializer, stack.MaxDiscoveredOnLinkPrefixes+2) - prefixes := [stack.MaxDiscoveredOnLinkPrefixes + 2]tcpip.Subnet{} + optSer := make(header.NDPOptionsSerializer, ipv6.MaxDiscoveredOnLinkPrefixes+2) + prefixes := [ipv6.MaxDiscoveredOnLinkPrefixes + 2]tcpip.Subnet{} // Receive an RA with 2 more than the max number of discovered on-link // prefixes. - for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ { + for i := 0; i < ipv6.MaxDiscoveredOnLinkPrefixes+2; i++ { prefixAddr := [16]byte{1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 0} prefixAddr[7] = byte(i) prefix := tcpip.AddressWithPrefix{ @@ -1665,8 +1660,8 @@ func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { } e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, optSer)) - for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ { - if i < stack.MaxDiscoveredOnLinkPrefixes { + for i := 0; i < ipv6.MaxDiscoveredOnLinkPrefixes+2; i++ { + if i < ipv6.MaxDiscoveredOnLinkPrefixes { select { case e := <-ndpDisp.prefixC: if diff := checkPrefixEvent(e, prefixes[i], true); diff != "" { @@ -1716,12 +1711,13 @@ func TestNoAutoGenAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: handle, - AutoGenGlobalAddresses: autogen, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: handle, + AutoGenGlobalAddresses: autogen, + }, + NDPDisp: &ndpDisp, + })}, }) s.SetForwarding(ipv6.ProtocolNumber, forwarding) @@ -1749,14 +1745,14 @@ func checkAutoGenAddrEvent(e ndpAutoGenAddrEvent, addr tcpip.AddressWithPrefix, // TestAutoGenAddr tests that an address is properly generated and invalidated // when configured to do so. -func TestAutoGenAddr(t *testing.T) { +func TestAutoGenAddr2(t *testing.T) { const newMinVL = 2 newMinVLDuration := newMinVL * time.Second - saved := stack.MinPrefixInformationValidLifetimeForUpdate + saved := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = saved + ipv6.MinPrefixInformationValidLifetimeForUpdate = saved }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) @@ -1766,12 +1762,13 @@ func TestAutoGenAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1876,14 +1873,14 @@ func TestAutoGenTempAddr(t *testing.T) { newMinVLDuration = newMinVL * time.Second ) - savedMinPrefixInformationValidLifetimeForUpdate := stack.MinPrefixInformationValidLifetimeForUpdate - savedMaxDesync := stack.MaxDesyncFactor + savedMinPrefixInformationValidLifetimeForUpdate := ipv6.MinPrefixInformationValidLifetimeForUpdate + savedMaxDesync := ipv6.MaxDesyncFactor defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = savedMinPrefixInformationValidLifetimeForUpdate - stack.MaxDesyncFactor = savedMaxDesync + ipv6.MinPrefixInformationValidLifetimeForUpdate = savedMinPrefixInformationValidLifetimeForUpdate + ipv6.MaxDesyncFactor = savedMaxDesync }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration - stack.MaxDesyncFactor = time.Nanosecond + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + ipv6.MaxDesyncFactor = time.Nanosecond prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) @@ -1931,16 +1928,17 @@ func TestAutoGenTempAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: test.dupAddrTransmits, - RetransmitTimer: test.retransmitTimer, - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, - TempIIDSeed: seed, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: test.dupAddrTransmits, + RetransmitTimer: test.retransmitTimer, + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + TempIIDSeed: seed, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2119,11 +2117,11 @@ func TestAutoGenTempAddr(t *testing.T) { func TestNoAutoGenTempAddrForLinkLocal(t *testing.T) { const nicID = 1 - savedMaxDesyncFactor := stack.MaxDesyncFactor + savedMaxDesyncFactor := ipv6.MaxDesyncFactor defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MaxDesyncFactor = savedMaxDesyncFactor }() - stack.MaxDesyncFactor = time.Nanosecond + ipv6.MaxDesyncFactor = time.Nanosecond tests := []struct { name string @@ -2160,12 +2158,13 @@ func TestNoAutoGenTempAddrForLinkLocal(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, - AutoGenIPv6LinkLocal: true, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + AutoGenIPv6LinkLocal: true, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2211,11 +2210,11 @@ func TestNoAutoGenTempAddrWithoutStableAddr(t *testing.T) { retransmitTimer = 2 * time.Second ) - savedMaxDesyncFactor := stack.MaxDesyncFactor + savedMaxDesyncFactor := ipv6.MaxDesyncFactor defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MaxDesyncFactor = savedMaxDesyncFactor }() - stack.MaxDesyncFactor = 0 + ipv6.MaxDesyncFactor = 0 prefix, _, addr := prefixSubnetAddr(0, linkAddr1) var tempIIDHistory [header.IIDSize]byte @@ -2228,15 +2227,16 @@ func TestNoAutoGenTempAddrWithoutStableAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: dadTransmits, - RetransmitTimer: retransmitTimer, - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: dadTransmits, + RetransmitTimer: retransmitTimer, + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2294,17 +2294,17 @@ func TestAutoGenTempAddrRegen(t *testing.T) { newMinVLDuration = newMinVL * time.Second ) - savedMaxDesyncFactor := stack.MaxDesyncFactor - savedMinMaxTempAddrPreferredLifetime := stack.MinMaxTempAddrPreferredLifetime - savedMinMaxTempAddrValidLifetime := stack.MinMaxTempAddrValidLifetime + savedMaxDesyncFactor := ipv6.MaxDesyncFactor + savedMinMaxTempAddrPreferredLifetime := ipv6.MinMaxTempAddrPreferredLifetime + savedMinMaxTempAddrValidLifetime := ipv6.MinMaxTempAddrValidLifetime defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor - stack.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime - stack.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime + ipv6.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime + ipv6.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime }() - stack.MaxDesyncFactor = 0 - stack.MinMaxTempAddrPreferredLifetime = newMinVLDuration - stack.MinMaxTempAddrValidLifetime = newMinVLDuration + ipv6.MaxDesyncFactor = 0 + ipv6.MinMaxTempAddrPreferredLifetime = newMinVLDuration + ipv6.MinMaxTempAddrValidLifetime = newMinVLDuration prefix, _, addr := prefixSubnetAddr(0, linkAddr1) var tempIIDHistory [header.IIDSize]byte @@ -2317,16 +2317,17 @@ func TestAutoGenTempAddrRegen(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2), } e := channel.New(0, 1280, linkAddr1) - ndpConfigs := stack.NDPConfigurations{ + ndpConfigs := ipv6.NDPConfigurations{ HandleRAs: true, AutoGenGlobalAddresses: true, AutoGenTempGlobalAddresses: true, RegenAdvanceDuration: newMinVLDuration - regenAfter, } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2382,8 +2383,11 @@ func TestAutoGenTempAddrRegen(t *testing.T) { // Stop generating temporary addresses ndpConfigs.AutoGenTempGlobalAddresses = false - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) + if ipv6Ep, err := s.GetNetworkEndpoint(nicID, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) + } else { + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(ndpConfigs) } // Wait for all the temporary addresses to get invalidated. @@ -2439,17 +2443,17 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { newMinVLDuration = newMinVL * time.Second ) - savedMaxDesyncFactor := stack.MaxDesyncFactor - savedMinMaxTempAddrPreferredLifetime := stack.MinMaxTempAddrPreferredLifetime - savedMinMaxTempAddrValidLifetime := stack.MinMaxTempAddrValidLifetime + savedMaxDesyncFactor := ipv6.MaxDesyncFactor + savedMinMaxTempAddrPreferredLifetime := ipv6.MinMaxTempAddrPreferredLifetime + savedMinMaxTempAddrValidLifetime := ipv6.MinMaxTempAddrValidLifetime defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor - stack.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime - stack.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime + ipv6.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime + ipv6.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime }() - stack.MaxDesyncFactor = 0 - stack.MinMaxTempAddrPreferredLifetime = newMinVLDuration - stack.MinMaxTempAddrValidLifetime = newMinVLDuration + ipv6.MaxDesyncFactor = 0 + ipv6.MinMaxTempAddrPreferredLifetime = newMinVLDuration + ipv6.MinMaxTempAddrValidLifetime = newMinVLDuration prefix, _, addr := prefixSubnetAddr(0, linkAddr1) var tempIIDHistory [header.IIDSize]byte @@ -2462,16 +2466,17 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2), } e := channel.New(0, 1280, linkAddr1) - ndpConfigs := stack.NDPConfigurations{ + ndpConfigs := ipv6.NDPConfigurations{ HandleRAs: true, AutoGenGlobalAddresses: true, AutoGenTempGlobalAddresses: true, RegenAdvanceDuration: newMinVLDuration - regenAfter, } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2545,9 +2550,12 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { // as paased. ndpConfigs.MaxTempAddrValidLifetime = 100 * time.Second ndpConfigs.MaxTempAddrPreferredLifetime = 100 * time.Second - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) + ipv6Ep, err := s.GetNetworkEndpoint(nicID, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) } + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(ndpConfigs) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) select { case e := <-ndpDisp.autoGenAddrC: @@ -2565,9 +2573,7 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout ndpConfigs.MaxTempAddrValidLifetime = newLifetimes ndpConfigs.MaxTempAddrPreferredLifetime = newLifetimes - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) - } + ndpEP.SetNDPConfigurations(ndpConfigs) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) expectAutoGenAddrEventAsync(tempAddr3, newAddr, regenAfter+defaultAsyncPositiveEventTimeout) } @@ -2655,20 +2661,21 @@ func TestMixedSLAACAddrConflictRegen(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2), } e := channel.New(0, 1280, linkAddr1) - ndpConfigs := stack.NDPConfigurations{ + ndpConfigs := ipv6.NDPConfigurations{ HandleRAs: true, AutoGenGlobalAddresses: true, AutoGenTempGlobalAddresses: test.tempAddrs, AutoGenAddressConflictRetries: 1, } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: test.nicNameFromID, - }, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: test.nicNameFromID, + }, + })}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) s.SetRouteTable([]tcpip.Route{{ @@ -2739,8 +2746,11 @@ func TestMixedSLAACAddrConflictRegen(t *testing.T) { ndpDisp.dadC = make(chan ndpDADEvent, 2) ndpConfigs.DupAddrDetectTransmits = dupAddrTransmits ndpConfigs.RetransmitTimer = retransmitTimer - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) + if ipv6Ep, err := s.GetNetworkEndpoint(nicID, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) + } else { + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(ndpConfigs) } // Do SLAAC for prefix. @@ -2754,9 +2764,7 @@ func TestMixedSLAACAddrConflictRegen(t *testing.T) { // DAD failure to restart the local generation process. addr := test.addrs[maxSLAACAddrLocalRegenAttempts-1] expectAutoGenAddrAsyncEvent(addr, newAddr) - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) select { case e := <-ndpDisp.dadC: if diff := checkDADEvent(e, nicID, addr.Address, false, nil); diff != "" { @@ -2794,14 +2802,15 @@ func stackAndNdpDispatcherWithDefaultRoute(t *testing.T, nicID tcpip.NICID, useN } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: ndpDisp, - UseNeighborCache: useNeighborCache, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: ndpDisp, + })}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, + UseNeighborCache: useNeighborCache, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -3036,11 +3045,11 @@ func TestAutoGenAddrJobDeprecation(t *testing.T) { for _, stackTyp := range stacks { t.Run(stackTyp.name, func(t *testing.T) { - saved := stack.MinPrefixInformationValidLifetimeForUpdate + saved := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = saved + ipv6.MinPrefixInformationValidLifetimeForUpdate = saved }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) @@ -3258,12 +3267,12 @@ func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) { const infiniteVLSeconds = 2 const minVLSeconds = 1 savedIL := header.NDPInfiniteLifetime - savedMinVL := stack.MinPrefixInformationValidLifetimeForUpdate + savedMinVL := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = savedMinVL + ipv6.MinPrefixInformationValidLifetimeForUpdate = savedMinVL header.NDPInfiniteLifetime = savedIL }() - stack.MinPrefixInformationValidLifetimeForUpdate = minVLSeconds * time.Second + ipv6.MinPrefixInformationValidLifetimeForUpdate = minVLSeconds * time.Second header.NDPInfiniteLifetime = infiniteVLSeconds * time.Second prefix, _, addr := prefixSubnetAddr(0, linkAddr1) @@ -3307,12 +3316,13 @@ func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3357,11 +3367,11 @@ func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) { func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) { const infiniteVL = 4294967295 const newMinVL = 4 - saved := stack.MinPrefixInformationValidLifetimeForUpdate + saved := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = saved + ipv6.MinPrefixInformationValidLifetimeForUpdate = saved }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second prefix, _, addr := prefixSubnetAddr(0, linkAddr1) @@ -3449,12 +3459,13 @@ func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) { } e := channel.New(10, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3515,12 +3526,13 @@ func TestAutoGenAddrRemoval(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3700,12 +3712,13 @@ func TestAutoGenAddrStaticConflict(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3781,18 +3794,19 @@ func TestAutoGenAddrWithOpaqueIID(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: func(_ tcpip.NICID, nicName string) string { - return nicName + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, }, - SecretKey: secretKey, - }, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: func(_ tcpip.NICID, nicName string) string { + return nicName + }, + SecretKey: secretKey, + }, + })}, }) opts := stack.NICOptions{Name: nicName} if err := s.CreateNICWithOptions(nicID, e, opts); err != nil { @@ -3856,11 +3870,11 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { const lifetimeSeconds = 10 // Needed for the temporary address sub test. - savedMaxDesync := stack.MaxDesyncFactor + savedMaxDesync := ipv6.MaxDesyncFactor defer func() { - stack.MaxDesyncFactor = savedMaxDesync + ipv6.MaxDesyncFactor = savedMaxDesync }() - stack.MaxDesyncFactor = time.Nanosecond + ipv6.MaxDesyncFactor = time.Nanosecond var secretKeyBuf [header.OpaqueIIDSecretKeyMinBytes]byte secretKey := secretKeyBuf[:] @@ -3938,14 +3952,14 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { addrTypes := []struct { name string - ndpConfigs stack.NDPConfigurations + ndpConfigs ipv6.NDPConfigurations autoGenLinkLocal bool prepareFn func(t *testing.T, ndpDisp *ndpDispatcher, e *channel.Endpoint, tempIIDHistory []byte) []tcpip.AddressWithPrefix addrGenFn func(dadCounter uint8, tempIIDHistory []byte) tcpip.AddressWithPrefix }{ { name: "Global address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, HandleRAs: true, @@ -3963,7 +3977,7 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { }, { name: "LinkLocal address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, }, @@ -3977,7 +3991,7 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { }, { name: "Temporary address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, HandleRAs: true, @@ -4029,16 +4043,17 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { ndpConfigs := addrType.ndpConfigs ndpConfigs.AutoGenAddressConflictRetries = maxRetries s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: func(_ tcpip.NICID, nicName string) string { - return nicName + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: func(_ tcpip.NICID, nicName string) string { + return nicName + }, + SecretKey: secretKey, }, - SecretKey: secretKey, - }, + })}, }) opts := stack.NICOptions{Name: nicName} if err := s.CreateNICWithOptions(nicID, e, opts); err != nil { @@ -4059,9 +4074,7 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { } // Simulate a DAD conflict. - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) expectAutoGenAddrEvent(t, &ndpDisp, addr, invalidatedAddr) expectDADEvent(t, &ndpDisp, addr.Address, false) @@ -4119,14 +4132,14 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { addrTypes := []struct { name string - ndpConfigs stack.NDPConfigurations + ndpConfigs ipv6.NDPConfigurations autoGenLinkLocal bool subnet tcpip.Subnet triggerSLAACFn func(e *channel.Endpoint) }{ { name: "Global address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, HandleRAs: true, @@ -4142,7 +4155,7 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { }, { name: "LinkLocal address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, AutoGenAddressConflictRetries: maxRetries, @@ -4165,10 +4178,11 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, - NDPConfigs: addrType.ndpConfigs, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, + NDPConfigs: addrType.ndpConfigs, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -4198,9 +4212,7 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { expectAutoGenAddrEvent(addr, newAddr) // Simulate a DAD conflict. - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) expectAutoGenAddrEvent(addr, invalidatedAddr) select { case e := <-ndpDisp.dadC: @@ -4250,21 +4262,22 @@ func TestAutoGenAddrContinuesLifetimesAfterRetry(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: dadTransmits, - RetransmitTimer: retransmitTimer, - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenAddressConflictRetries: maxRetries, - }, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: func(_ tcpip.NICID, nicName string) string { - return nicName + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: dadTransmits, + RetransmitTimer: retransmitTimer, + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenAddressConflictRetries: maxRetries, }, - SecretKey: secretKey, - }, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: func(_ tcpip.NICID, nicName string) string { + return nicName + }, + SecretKey: secretKey, + }, + })}, }) opts := stack.NICOptions{Name: nicName} if err := s.CreateNICWithOptions(nicID, e, opts); err != nil { @@ -4296,9 +4309,7 @@ func TestAutoGenAddrContinuesLifetimesAfterRetry(t *testing.T) { // Simulate a DAD conflict after some time has passed. time.Sleep(failureTimer) - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) expectAutoGenAddrEvent(addr, invalidatedAddr) select { case e := <-ndpDisp.dadC: @@ -4459,11 +4470,12 @@ func TestNDPRecursiveDNSServerDispatch(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(1) = %s", err) @@ -4509,11 +4521,12 @@ func TestNDPDNSSearchListDispatch(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -4694,15 +4707,16 @@ func TestCleanupNDPState(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, test.maxAutoGenAddrEvents), } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: true, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - DiscoverOnLinkPrefixes: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: true, + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + DiscoverOnLinkPrefixes: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) expectRouterEvent := func() (bool, ndpRouterEvent) { @@ -4967,18 +4981,19 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } - expectDHCPv6Event := func(configuration stack.DHCPv6ConfigurationFromNDPRA) { + expectDHCPv6Event := func(configuration ipv6.DHCPv6ConfigurationFromNDPRA) { t.Helper() select { case e := <-ndpDisp.dhcpv6ConfigurationC: @@ -5002,7 +5017,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Even if the first RA reports no DHCPv6 configurations are available, the // dispatcher should get an event. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) - expectDHCPv6Event(stack.DHCPv6NoConfiguration) + expectDHCPv6Event(ipv6.DHCPv6NoConfiguration) // Receiving the same update again should not result in an event to the // dispatcher. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) @@ -5011,19 +5026,19 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Receive an RA that updates the DHCPv6 configuration to Other // Configurations. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) - expectDHCPv6Event(stack.DHCPv6OtherConfigurations) + expectDHCPv6Event(ipv6.DHCPv6OtherConfigurations) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() // Receive an RA that updates the DHCPv6 configuration to Managed Address. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false)) - expectDHCPv6Event(stack.DHCPv6ManagedAddress) + expectDHCPv6Event(ipv6.DHCPv6ManagedAddress) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false)) expectNoDHCPv6Event() // Receive an RA that updates the DHCPv6 configuration to none. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) - expectDHCPv6Event(stack.DHCPv6NoConfiguration) + expectDHCPv6Event(ipv6.DHCPv6NoConfiguration) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) expectNoDHCPv6Event() @@ -5031,7 +5046,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // // Note, when the M flag is set, the O flag is redundant. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true)) - expectDHCPv6Event(stack.DHCPv6ManagedAddress) + expectDHCPv6Event(ipv6.DHCPv6ManagedAddress) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true)) expectNoDHCPv6Event() // Even though the DHCPv6 flags are different, the effective configuration is @@ -5044,7 +5059,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Receive an RA that updates the DHCPv6 configuration to Other // Configurations. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) - expectDHCPv6Event(stack.DHCPv6OtherConfigurations) + expectDHCPv6Event(ipv6.DHCPv6OtherConfigurations) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() @@ -5059,7 +5074,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Receive an RA that updates the DHCPv6 configuration to Other // Configurations. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) - expectDHCPv6Event(stack.DHCPv6OtherConfigurations) + expectDHCPv6Event(ipv6.DHCPv6OtherConfigurations) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() } @@ -5217,12 +5232,13 @@ func TestRouterSolicitation(t *testing.T) { } } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - MaxRtrSolicitations: test.maxRtrSolicit, - RtrSolicitationInterval: test.rtrSolicitInt, - MaxRtrSolicitationDelay: test.maxRtrSolicitDelay, - }, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + MaxRtrSolicitations: test.maxRtrSolicit, + RtrSolicitationInterval: test.rtrSolicitInt, + MaxRtrSolicitationDelay: test.maxRtrSolicitDelay, + }, + })}, }) if err := s.CreateNIC(nicID, &e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -5357,12 +5373,13 @@ func TestStopStartSolicitingRouters(t *testing.T) { checker.NDPRS()) } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - MaxRtrSolicitations: maxRtrSolicitations, - RtrSolicitationInterval: interval, - MaxRtrSolicitationDelay: delay, - }, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + MaxRtrSolicitations: maxRtrSolicitations, + RtrSolicitationInterval: interval, + MaxRtrSolicitationDelay: delay, + }, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) diff --git a/pkg/tcpip/stack/neighbor_entry.go b/pkg/tcpip/stack/neighbor_entry.go index 213646160..9a72bec79 100644 --- a/pkg/tcpip/stack/neighbor_entry.go +++ b/pkg/tcpip/stack/neighbor_entry.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" ) // NeighborEntry describes a neighboring device in the local network. @@ -439,7 +440,7 @@ func (e *neighborEntry) handleConfirmationLocked(linkAddr tcpip.LinkAddress, fla e.notifyWakersLocked() } - if e.isRouter && !flags.IsRouter { + if e.isRouter && !flags.IsRouter && header.IsV6UnicastAddress(e.neigh.Addr) { // "In those cases where the IsRouter flag changes from TRUE to FALSE as // a result of this update, the node MUST remove that router from the // Default Router List and update the Destination Cache entries for all @@ -447,9 +448,17 @@ func (e *neighborEntry) handleConfirmationLocked(linkAddr tcpip.LinkAddress, fla // 7.3.3. This is needed to detect when a node that is used as a router // stops forwarding packets due to being configured as a host." // - RFC 4861 section 7.2.5 - e.nic.mu.Lock() - e.nic.mu.ndp.invalidateDefaultRouter(e.neigh.Addr) - e.nic.mu.Unlock() + // + // TODO(gvisor.dev/issue/4085): Remove the special casing we do for IPv6 + // here. + ep, ok := e.nic.networkEndpoints[header.IPv6ProtocolNumber] + if !ok { + panic(fmt.Sprintf("have a neighbor entry for an IPv6 router but no IPv6 network endpoint")) + } + + if ndpEP, ok := ep.(NDPEndpoint); ok { + ndpEP.InvalidateDefaultRouter(e.neigh.Addr) + } } e.isRouter = flags.IsRouter diff --git a/pkg/tcpip/stack/neighbor_entry_test.go b/pkg/tcpip/stack/neighbor_entry_test.go index e530ec7ea..a265fff0a 100644 --- a/pkg/tcpip/stack/neighbor_entry_test.go +++ b/pkg/tcpip/stack/neighbor_entry_test.go @@ -28,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/faketime" + "gvisor.dev/gvisor/pkg/tcpip/header" ) const ( @@ -233,18 +234,15 @@ func entryTestSetup(c NUDConfigurations) (*neighborEntry, *testNUDDispatcher, *e nudDisp: &disp, }, } + nic.networkEndpoints = map[tcpip.NetworkProtocolNumber]NetworkEndpoint{ + header.IPv6ProtocolNumber: (&testIPv6Protocol{}).NewEndpoint(&nic, nil, nil, nil), + } rng := rand.New(rand.NewSource(time.Now().UnixNano())) nudState := NewNUDState(c, rng) linkRes := entryTestLinkResolver{} entry := newNeighborEntry(&nic, entryTestAddr1 /* remoteAddr */, entryTestAddr2 /* localAddr */, nudState, &linkRes) - // Stub out ndpState to verify modification of default routers. - nic.mu.ndp = ndpState{ - nic: &nic, - defaultRouters: make(map[tcpip.Address]defaultRouterState), - } - // Stub out the neighbor cache to verify deletion from the cache. nic.neigh = &neighborCache{ nic: &nic, @@ -817,6 +815,8 @@ func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { c := DefaultNUDConfigurations() e, nudDisp, linkRes, _ := entryTestSetup(c) + ipv6EP := e.nic.networkEndpoints[header.IPv6ProtocolNumber].(*testIPv6Endpoint) + e.mu.Lock() e.handlePacketQueuedLocked() e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ @@ -830,9 +830,7 @@ func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { if got, want := e.isRouter, true; got != want { t.Errorf("got e.isRouter = %t, want = %t", got, want) } - e.nic.mu.ndp.defaultRouters[entryTestAddr1] = defaultRouterState{ - invalidationJob: e.nic.stack.newJob(&testLocker{}, func() {}), - } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ Solicited: false, Override: false, @@ -841,8 +839,8 @@ func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { if got, want := e.isRouter, false; got != want { t.Errorf("got e.isRouter = %t, want = %t", got, want) } - if _, ok := e.nic.mu.ndp.defaultRouters[entryTestAddr1]; ok { - t.Errorf("unexpected defaultRouter for %s", entryTestAddr1) + if ipv6EP.invalidatedRtr != e.neigh.Addr { + t.Errorf("got ipv6EP.invalidatedRtr = %s, want = %s", ipv6EP.invalidatedRtr, e.neigh.Addr) } e.mu.Unlock() diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 06d70dd1c..212c6edae 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -18,7 +18,6 @@ import ( "fmt" "math/rand" "reflect" - "sort" "sync/atomic" "gvisor.dev/gvisor/pkg/sleep" @@ -28,13 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) -var ipv4BroadcastAddr = tcpip.ProtocolAddress{ - Protocol: header.IPv4ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: header.IPv4Broadcast, - PrefixLen: 8 * header.IPv4AddressSize, - }, -} +var _ NetworkInterface = (*NIC)(nil) // NIC represents a "network interface card" to which the networking stack is // attached. @@ -49,18 +42,18 @@ type NIC struct { neigh *neighborCache networkEndpoints map[tcpip.NetworkProtocolNumber]NetworkEndpoint + // enabled is set to 1 when the NIC is enabled and 0 when it is disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + mu struct { sync.RWMutex - enabled bool spoofing bool promiscuous bool - primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint - endpoints map[NetworkEndpointID]*referencedNetworkEndpoint - mcastJoins map[NetworkEndpointID]uint32 // packetEPs is protected by mu, but the contained PacketEndpoint // values are not. packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint - ndp ndpState } } @@ -84,25 +77,6 @@ type DirectionStats struct { Bytes *tcpip.StatCounter } -// PrimaryEndpointBehavior is an enumeration of an endpoint's primacy behavior. -type PrimaryEndpointBehavior int - -const ( - // CanBePrimaryEndpoint indicates the endpoint can be used as a primary - // endpoint for new connections with no local address. This is the - // default when calling NIC.AddAddress. - CanBePrimaryEndpoint PrimaryEndpointBehavior = iota - - // FirstPrimaryEndpoint indicates the endpoint should be the first - // primary endpoint considered. If there are multiple endpoints with - // this behavior, the most recently-added one will be first. - FirstPrimaryEndpoint - - // NeverPrimaryEndpoint indicates the endpoint should never be a - // primary endpoint. - NeverPrimaryEndpoint -) - // newNIC returns a new NIC using the default NDP configurations from stack. func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICContext) *NIC { // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For @@ -122,19 +96,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC stats: makeNICStats(), networkEndpoints: make(map[tcpip.NetworkProtocolNumber]NetworkEndpoint), } - nic.mu.primary = make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint) - nic.mu.endpoints = make(map[NetworkEndpointID]*referencedNetworkEndpoint) - nic.mu.mcastJoins = make(map[NetworkEndpointID]uint32) nic.mu.packetEPs = make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint) - nic.mu.ndp = ndpState{ - nic: nic, - configs: stack.ndpConfigs, - dad: make(map[tcpip.Address]dadState), - defaultRouters: make(map[tcpip.Address]defaultRouterState), - onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), - slaacPrefixes: make(map[tcpip.Subnet]slaacPrefixState), - } - nic.mu.ndp.initializeTempAddrState() // Check for Neighbor Unreachability Detection support. var nud NUDHandler @@ -162,7 +124,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC for _, netProto := range stack.networkProtocols { netNum := netProto.Number() nic.mu.packetEPs[netNum] = nil - nic.networkEndpoints[netNum] = netProto.NewEndpoint(id, stack, nud, nic, ep, stack) + nic.networkEndpoints[netNum] = netProto.NewEndpoint(nic, stack, nud, nic) } nic.linkEP.Attach(nic) @@ -170,29 +132,28 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC return nic } -// enabled returns true if n is enabled. -func (n *NIC) enabled() bool { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - return enabled +// Enabled implements NetworkInterface. +func (n *NIC) Enabled() bool { + return atomic.LoadUint32(&n.enabled) == 1 } -// disable disables n. +// setEnabled sets the enabled status for the NIC. // -// It undoes the work done by enable. -func (n *NIC) disable() *tcpip.Error { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - if !enabled { - return nil +// Returns true if the enabled status was updated. +func (n *NIC) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&n.enabled, 1) == 0 } + return atomic.SwapUint32(&n.enabled, 0) == 1 +} +// disable disables n. +// +// It undoes the work done by enable. +func (n *NIC) disable() { n.mu.Lock() - err := n.disableLocked() + n.disableLocked() n.mu.Unlock() - return err } // disableLocked disables n. @@ -200,9 +161,9 @@ func (n *NIC) disable() *tcpip.Error { // It undoes the work done by enable. // // n MUST be locked. -func (n *NIC) disableLocked() *tcpip.Error { - if !n.mu.enabled { - return nil +func (n *NIC) disableLocked() { + if !n.setEnabled(false) { + return } // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be @@ -210,38 +171,9 @@ func (n *NIC) disableLocked() *tcpip.Error { // again, and applications may not know that the underlying NIC was ever // disabled. - if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok { - n.mu.ndp.stopSolicitingRouters() - n.mu.ndp.cleanupState(false /* hostOnly */) - - // Stop DAD for all the unicast IPv6 endpoints that are in the - // permanentTentative state. - for _, r := range n.mu.endpoints { - if addr := r.address(); r.getKind() == permanentTentative && header.IsV6UnicastAddress(addr) { - n.mu.ndp.stopDuplicateAddressDetection(addr) - } - } - - // The NIC may have already left the multicast group. - if err := n.leaveGroupLocked(header.IPv6AllNodesMulticastAddress, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { - return err - } - } - - if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { - // The NIC may have already left the multicast group. - if err := n.leaveGroupLocked(header.IPv4AllSystems, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { - return err - } - - // The address may have already been removed. - if err := n.removePermanentAddressLocked(ipv4BroadcastAddr.AddressWithPrefix.Address); err != nil && err != tcpip.ErrBadLocalAddress { - return err - } + for _, ep := range n.networkEndpoints { + ep.Disable() } - - n.mu.enabled = false - return nil } // enable enables n. @@ -251,162 +183,39 @@ func (n *NIC) disableLocked() *tcpip.Error { // routers if the stack is not operating as a router. If the stack is also // configured to auto-generate a link-local address, one will be generated. func (n *NIC) enable() *tcpip.Error { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - if enabled { - return nil - } - n.mu.Lock() defer n.mu.Unlock() - if n.mu.enabled { + if !n.setEnabled(true) { return nil } - n.mu.enabled = true - - // Create an endpoint to receive broadcast packets on this interface. - if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { - if _, err := n.addAddressLocked(ipv4BroadcastAddr, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { - return err - } - - // As per RFC 1122 section 3.3.7, all hosts should join the all-hosts - // multicast group. Note, the IANA calls the all-hosts multicast group the - // all-systems multicast group. - if err := n.joinGroupLocked(header.IPv4ProtocolNumber, header.IPv4AllSystems); err != nil { - return err - } - } - - // Join the IPv6 All-Nodes Multicast group if the stack is configured to - // use IPv6. This is required to ensure that this node properly receives - // and responds to the various NDP messages that are destined to the - // all-nodes multicast address. An example is the Neighbor Advertisement - // when we perform Duplicate Address Detection, or Router Advertisement - // when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861 - // section 4.2 for more information. - // - // Also auto-generate an IPv6 link-local address based on the NIC's - // link address if it is configured to do so. Note, each interface is - // required to have IPv6 link-local unicast address, as per RFC 4291 - // section 2.1. - _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber] - if !ok { - return nil - } - - // Join the All-Nodes multicast group before starting DAD as responses to DAD - // (NDP NS) messages may be sent to the All-Nodes multicast group if the - // source address of the NDP NS is the unspecified address, as per RFC 4861 - // section 7.2.4. - if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil { - return err - } - - // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent - // state. - // - // Addresses may have aleady completed DAD but in the time since the NIC was - // last enabled, other devices may have acquired the same addresses. - for _, r := range n.mu.endpoints { - addr := r.address() - if k := r.getKind(); (k != permanent && k != permanentTentative) || !header.IsV6UnicastAddress(addr) { - continue - } - - r.setKind(permanentTentative) - if err := n.mu.ndp.startDuplicateAddressDetection(addr, r); err != nil { + for _, ep := range n.networkEndpoints { + if err := ep.Enable(); err != nil { return err } } - // Do not auto-generate an IPv6 link-local address for loopback devices. - if n.stack.autoGenIPv6LinkLocal && !n.isLoopback() { - // The valid and preferred lifetime is infinite for the auto-generated - // link-local address. - n.mu.ndp.doSLAAC(header.IPv6LinkLocalPrefix.Subnet(), header.NDPInfiniteLifetime, header.NDPInfiniteLifetime) - } - - // If we are operating as a router, then do not solicit routers since we - // won't process the RAs anyways. - // - // Routers do not process Router Advertisements (RA) the same way a host - // does. That is, routers do not learn from RAs (e.g. on-link prefixes - // and default routers). Therefore, soliciting RAs from other routers on - // a link is unnecessary for routers. - if !n.stack.Forwarding(header.IPv6ProtocolNumber) { - n.mu.ndp.startSolicitingRouters() - } - return nil } -// remove detaches NIC from the link endpoint, and marks existing referenced -// network endpoints expired. This guarantees no packets between this NIC and -// the network stack. +// remove detaches NIC from the link endpoint and releases network endpoint +// resources. This guarantees no packets between this NIC and the network +// stack. func (n *NIC) remove() *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() n.disableLocked() - // TODO(b/151378115): come up with a better way to pick an error than the - // first one. - var err *tcpip.Error - - // Forcefully leave multicast groups. - for nid := range n.mu.mcastJoins { - if tempErr := n.leaveGroupLocked(nid.LocalAddress, true /* force */); tempErr != nil && err == nil { - err = tempErr - } - } - - // Remove permanent and permanentTentative addresses, so no packet goes out. - for nid, ref := range n.mu.endpoints { - switch ref.getKind() { - case permanentTentative, permanent: - if tempErr := n.removePermanentAddressLocked(nid.LocalAddress); tempErr != nil && err == nil { - err = tempErr - } - } - } - - // Release any resources the network endpoint may hold. for _, ep := range n.networkEndpoints { ep.Close() } + n.networkEndpoints = nil // Detach from link endpoint, so no packet comes in. n.linkEP.Attach(nil) - - return err -} - -// becomeIPv6Router transitions n into an IPv6 router. -// -// When transitioning into an IPv6 router, host-only state (NDP discovered -// routers, discovered on-link prefixes, and auto-generated addresses) will -// be cleaned up/invalidated and NDP router solicitations will be stopped. -func (n *NIC) becomeIPv6Router() { - n.mu.Lock() - defer n.mu.Unlock() - - n.mu.ndp.cleanupState(true /* hostOnly */) - n.mu.ndp.stopSolicitingRouters() -} - -// becomeIPv6Host transitions n into an IPv6 host. -// -// When transitioning into an IPv6 host, NDP router solicitations will be -// started. -func (n *NIC) becomeIPv6Host() { - n.mu.Lock() - defer n.mu.Unlock() - - n.mu.ndp.startSolicitingRouters() + return nil } // setPromiscuousMode enables or disables promiscuous mode. @@ -423,7 +232,8 @@ func (n *NIC) isPromiscuousMode() bool { return rv } -func (n *NIC) isLoopback() bool { +// IsLoopback implements NetworkInterface. +func (n *NIC) IsLoopback() bool { return n.linkEP.Capabilities()&CapabilityLoopback != 0 } @@ -440,200 +250,41 @@ func (n *NIC) setSpoofing(enable bool) { // // If an IPv6 primary endpoint is requested, Source Address Selection (as // defined by RFC 6724 section 5) will be performed. -func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) *referencedNetworkEndpoint { - if protocol == header.IPv6ProtocolNumber && len(remoteAddr) != 0 { - return n.primaryIPv6Endpoint(remoteAddr) - } - +func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) AssignableAddressEndpoint { n.mu.RLock() defer n.mu.RUnlock() - var deprecatedEndpoint *referencedNetworkEndpoint - for _, r := range n.mu.primary[protocol] { - if !r.isValidForOutgoingRLocked() { - continue - } - - if !r.deprecated { - if r.tryIncRef() { - // r is not deprecated, so return it immediately. - // - // If we kept track of a deprecated endpoint, decrement its reference - // count since it was incremented when we decided to keep track of it. - if deprecatedEndpoint != nil { - deprecatedEndpoint.decRefLocked() - deprecatedEndpoint = nil - } - - return r - } - } else if deprecatedEndpoint == nil && r.tryIncRef() { - // We prefer an endpoint that is not deprecated, but we keep track of r in - // case n doesn't have any non-deprecated endpoints. - // - // If we end up finding a more preferred endpoint, r's reference count - // will be decremented when such an endpoint is found. - deprecatedEndpoint = r - } - } - - // n doesn't have any valid non-deprecated endpoints, so return - // deprecatedEndpoint (which may be nil if n doesn't have any valid deprecated - // endpoints either). - return deprecatedEndpoint -} - -// ipv6AddrCandidate is an IPv6 candidate for Source Address Selection (RFC -// 6724 section 5). -type ipv6AddrCandidate struct { - ref *referencedNetworkEndpoint - scope header.IPv6AddressScope -} - -// primaryIPv6Endpoint returns an IPv6 endpoint following Source Address -// Selection (RFC 6724 section 5). -// -// Note, only rules 1-3 and 7 are followed. -// -// remoteAddr must be a valid IPv6 address. -func (n *NIC) primaryIPv6Endpoint(remoteAddr tcpip.Address) *referencedNetworkEndpoint { - n.mu.RLock() - ref := n.primaryIPv6EndpointRLocked(remoteAddr) - n.mu.RUnlock() - return ref -} - -// primaryIPv6EndpointLocked returns an IPv6 endpoint following Source Address -// Selection (RFC 6724 section 5). -// -// Note, only rules 1-3 and 7 are followed. -// -// remoteAddr must be a valid IPv6 address. -// -// n.mu MUST be read locked. -func (n *NIC) primaryIPv6EndpointRLocked(remoteAddr tcpip.Address) *referencedNetworkEndpoint { - primaryAddrs := n.mu.primary[header.IPv6ProtocolNumber] - - if len(primaryAddrs) == 0 { - return nil - } - - // Create a candidate set of available addresses we can potentially use as a - // source address. - cs := make([]ipv6AddrCandidate, 0, len(primaryAddrs)) - for _, r := range primaryAddrs { - // If r is not valid for outgoing connections, it is not a valid endpoint. - if !r.isValidForOutgoingRLocked() { - continue - } - - addr := r.address() - scope, err := header.ScopeForIPv6Address(addr) - if err != nil { - // Should never happen as we got r from the primary IPv6 endpoint list and - // ScopeForIPv6Address only returns an error if addr is not an IPv6 - // address. - panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", addr, err)) - } - - cs = append(cs, ipv6AddrCandidate{ - ref: r, - scope: scope, - }) - } - - remoteScope, err := header.ScopeForIPv6Address(remoteAddr) - if err != nil { - // primaryIPv6Endpoint should never be called with an invalid IPv6 address. - panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err)) - } - - // Sort the addresses as per RFC 6724 section 5 rules 1-3. - // - // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5. - sort.Slice(cs, func(i, j int) bool { - sa := cs[i] - sb := cs[j] - - // Prefer same address as per RFC 6724 section 5 rule 1. - if sa.ref.address() == remoteAddr { - return true - } - if sb.ref.address() == remoteAddr { - return false - } - - // Prefer appropriate scope as per RFC 6724 section 5 rule 2. - if sa.scope < sb.scope { - return sa.scope >= remoteScope - } else if sb.scope < sa.scope { - return sb.scope < remoteScope - } - - // Avoid deprecated addresses as per RFC 6724 section 5 rule 3. - if saDep, sbDep := sa.ref.deprecated, sb.ref.deprecated; saDep != sbDep { - // If sa is not deprecated, it is preferred over sb. - return sbDep - } - - // Prefer temporary addresses as per RFC 6724 section 5 rule 7. - if saTemp, sbTemp := sa.ref.configType == slaacTemp, sb.ref.configType == slaacTemp; saTemp != sbTemp { - return saTemp - } - - // sa and sb are equal, return the endpoint that is closest to the front of - // the primary endpoint list. - return i < j - }) - - // Return the most preferred address that can have its reference count - // incremented. - for _, c := range cs { - if r := c.ref; r.tryIncRef() { - return r - } - } - - return nil -} - -// hasPermanentAddrLocked returns true if n has a permanent (including currently -// tentative) address, addr. -func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool { - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] - + ep, ok := n.networkEndpoints[protocol] if !ok { - return false + return nil } - kind := ref.getKind() - - return kind == permanent || kind == permanentTentative + return ep.AcquirePrimaryAddress(remoteAddr, n.mu.spoofing) } -type getRefBehaviour int +type getAddressBehaviour int const ( // spoofing indicates that the NIC's spoofing flag should be observed when - // getting a NIC's referenced network endpoint. - spoofing getRefBehaviour = iota + // getting a NIC's address endpoint. + spoofing getAddressBehaviour = iota // promiscuous indicates that the NIC's promiscuous flag should be observed - // when getting a NIC's referenced network endpoint. + // when getting a NIC's address endpoint. promiscuous ) -func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous) +func (n *NIC) getAddress(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) AssignableAddressEndpoint { + return n.getAddressOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous) } // findEndpoint finds the endpoint, if any, with the given address. -func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, address, peb, spoofing) +func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) AssignableAddressEndpoint { + return n.getAddressOrCreateTemp(protocol, address, peb, spoofing) } -// getRefEpOrCreateTemp returns the referenced network endpoint for the given -// protocol and address. +// getAddressEpOrCreateTemp returns the address endpoint for the given protocol +// and address. // // If none exists a temporary one may be created if we are in promiscuous mode // or spoofing. Promiscuous mode will only be checked if promiscuous is true. @@ -641,9 +292,8 @@ func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.A // // If the address is the IPv4 broadcast address for an endpoint's network, that // endpoint will be returned. -func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getRefBehaviour) *referencedNetworkEndpoint { +func (n *NIC) getAddressOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getAddressBehaviour) AssignableAddressEndpoint { n.mu.RLock() - var spoofingOrPromiscuous bool switch tempRef { case spoofing: @@ -651,274 +301,54 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t case promiscuous: spoofingOrPromiscuous = n.mu.promiscuous } - - if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok { - // An endpoint with this id exists, check if it can be used and return it. - if !ref.isAssignedRLocked(spoofingOrPromiscuous) { - n.mu.RUnlock() - return nil - } - - if ref.tryIncRef() { - n.mu.RUnlock() - return ref - } - } - - if protocol == header.IPv4ProtocolNumber { - if ref := n.getIPv4RefForBroadcastOrLoopbackRLocked(address); ref != nil { - n.mu.RUnlock() - return ref - } - } n.mu.RUnlock() - - if !spoofingOrPromiscuous { - return nil - } - - // Try again with the lock in exclusive mode. If we still can't get the - // endpoint, create a new "temporary" endpoint. It will only exist while - // there's a route through it. - n.mu.Lock() - ref := n.getRefOrCreateTempLocked(protocol, address, peb) - n.mu.Unlock() - return ref + return n.getAddressOrCreateTempInner(protocol, address, spoofingOrPromiscuous, peb) } -// getRefForBroadcastOrLoopbackRLocked returns an endpoint whose address is the -// broadcast address for the endpoint's network or an address in the endpoint's -// subnet if the NIC is a loopback interface. This matches linux behaviour. -// -// n.mu MUST be read or write locked. -func (n *NIC) getIPv4RefForBroadcastOrLoopbackRLocked(address tcpip.Address) *referencedNetworkEndpoint { - for _, ref := range n.mu.endpoints { - // Only IPv4 has a notion of broadcast addresses or considers the loopback - // interface bound to an address's whole subnet (on linux). - if ref.protocol != header.IPv4ProtocolNumber { - continue - } - - subnet := ref.addrWithPrefix().Subnet() - if (subnet.IsBroadcast(address) || (n.isLoopback() && subnet.Contains(address))) && ref.isValidForOutgoingRLocked() && ref.tryIncRef() { - return ref - } +// getAddressOrCreateTempInner is like getAddressEpOrCreateTemp except a boolean +// is passed to indicate whether or not we should generate temporary endpoints. +func (n *NIC) getAddressOrCreateTempInner(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, createTemp bool, peb PrimaryEndpointBehavior) AssignableAddressEndpoint { + if ep, ok := n.networkEndpoints[protocol]; ok { + return ep.AcquireAssignedAddress(address, createTemp, peb) } return nil } -/// getRefOrCreateTempLocked returns an existing endpoint for address or creates -/// and returns a temporary endpoint. -// -// If the address is the IPv4 broadcast address for an endpoint's network, that -// endpoint will be returned. -// -// n.mu must be write locked. -func (n *NIC) getRefOrCreateTempLocked(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint { - if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok { - // No need to check the type as we are ok with expired endpoints at this - // point. - if ref.tryIncRef() { - return ref - } - // tryIncRef failing means the endpoint is scheduled to be removed once the - // lock is released. Remove it here so we can create a new (temporary) one. - // The removal logic waiting for the lock handles this case. - n.removeEndpointLocked(ref) - } - - if protocol == header.IPv4ProtocolNumber { - if ref := n.getIPv4RefForBroadcastOrLoopbackRLocked(address); ref != nil { - return ref - } - } - - // Add a new temporary endpoint. - netProto, ok := n.stack.networkProtocols[protocol] - if !ok { - return nil - } - ref, _ := n.addAddressLocked(tcpip.ProtocolAddress{ - Protocol: protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: address, - PrefixLen: netProto.DefaultPrefixLen(), - }, - }, peb, temporary, static, false) - return ref -} - -// addAddressLocked adds a new protocolAddress to n. -// -// If n already has the address in a non-permanent state, and the kind given is -// permanent, that address will be promoted in place and its properties set to -// the properties provided. Otherwise, it returns tcpip.ErrDuplicateAddress. -func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType, deprecated bool) (*referencedNetworkEndpoint, *tcpip.Error) { - // TODO(b/141022673): Validate IP addresses before adding them. - - // Sanity check. - id := NetworkEndpointID{LocalAddress: protocolAddress.AddressWithPrefix.Address} - if ref, ok := n.mu.endpoints[id]; ok { - // Endpoint already exists. - if kind != permanent { - return nil, tcpip.ErrDuplicateAddress - } - switch ref.getKind() { - case permanentTentative, permanent: - // The NIC already have a permanent endpoint with that address. - return nil, tcpip.ErrDuplicateAddress - case permanentExpired, temporary: - // Promote the endpoint to become permanent and respect the new peb, - // configType and deprecated status. - if ref.tryIncRef() { - // TODO(b/147748385): Perform Duplicate Address Detection when promoting - // an IPv6 endpoint to permanent. - ref.setKind(permanent) - ref.deprecated = deprecated - ref.configType = configType - - refs := n.mu.primary[ref.protocol] - for i, r := range refs { - if r == ref { - switch peb { - case CanBePrimaryEndpoint: - return ref, nil - case FirstPrimaryEndpoint: - if i == 0 { - return ref, nil - } - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - case NeverPrimaryEndpoint: - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - return ref, nil - } - } - } - - n.insertPrimaryEndpointLocked(ref, peb) - - return ref, nil - } - // tryIncRef failing means the endpoint is scheduled to be removed once - // the lock is released. Remove it here so we can create a new - // (permanent) one. The removal logic waiting for the lock handles this - // case. - n.removeEndpointLocked(ref) - } - } - +// addAddress adds a new address to n, so that it starts accepting packets +// targeted at the given address (and network protocol). +func (n *NIC) addAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { ep, ok := n.networkEndpoints[protocolAddress.Protocol] if !ok { - return nil, tcpip.ErrUnknownProtocol + return tcpip.ErrUnknownProtocol } - isIPv6Unicast := protocolAddress.Protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(protocolAddress.AddressWithPrefix.Address) - - // If the address is an IPv6 address and it is a permanent address, - // mark it as tentative so it goes through the DAD process if the NIC is - // enabled. If the NIC is not enabled, DAD will be started when the NIC is - // enabled. - if isIPv6Unicast && kind == permanent { - kind = permanentTentative - } - - ref := &referencedNetworkEndpoint{ - refs: 1, - addr: protocolAddress.AddressWithPrefix, - ep: ep, - nic: n, - protocol: protocolAddress.Protocol, - kind: kind, - configType: configType, - deprecated: deprecated, - } - - // Set up resolver if link address resolution exists for this protocol. - if n.linkEP.Capabilities()&CapabilityResolutionRequired != 0 { - if linkRes, ok := n.stack.linkAddrResolvers[protocolAddress.Protocol]; ok { - ref.linkCache = n.stack - ref.linkRes = linkRes - } - } - - // If we are adding an IPv6 unicast address, join the solicited-node - // multicast address. - if isIPv6Unicast { - snmc := header.SolicitedNodeAddr(protocolAddress.AddressWithPrefix.Address) - if err := n.joinGroupLocked(protocolAddress.Protocol, snmc); err != nil { - return nil, err - } - } - - n.mu.endpoints[id] = ref - - n.insertPrimaryEndpointLocked(ref, peb) - - // If we are adding a tentative IPv6 address, start DAD if the NIC is enabled. - if isIPv6Unicast && kind == permanentTentative && n.mu.enabled { - if err := n.mu.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil { - return nil, err - } + addressEndpoint, err := ep.AddAndAcquirePermanentAddress(protocolAddress.AddressWithPrefix, peb, AddressConfigStatic, false /* deprecated */) + if err == nil { + // We have no need for the address endpoint. + addressEndpoint.DecRef() } - - return ref, nil -} - -// AddAddress adds a new address to n, so that it starts accepting packets -// targeted at the given address (and network protocol). -func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { - // Add the endpoint. - n.mu.Lock() - _, err := n.addAddressLocked(protocolAddress, peb, permanent, static, false /* deprecated */) - n.mu.Unlock() - return err } -// AllAddresses returns all addresses (primary and non-primary) associated with +// allPermanentAddresses returns all permanent addresses associated with // this NIC. -func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { - n.mu.RLock() - defer n.mu.RUnlock() - - addrs := make([]tcpip.ProtocolAddress, 0, len(n.mu.endpoints)) - for _, ref := range n.mu.endpoints { - // Don't include tentative, expired or temporary endpoints to - // avoid confusion and prevent the caller from using those. - switch ref.getKind() { - case permanentExpired, temporary: - continue +func (n *NIC) allPermanentAddresses() []tcpip.ProtocolAddress { + var addrs []tcpip.ProtocolAddress + for p, ep := range n.networkEndpoints { + for _, a := range ep.PermanentAddresses() { + addrs = append(addrs, tcpip.ProtocolAddress{Protocol: p, AddressWithPrefix: a}) } - - addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: ref.protocol, - AddressWithPrefix: ref.addrWithPrefix(), - }) } return addrs } -// PrimaryAddresses returns the primary addresses associated with this NIC. -func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { - n.mu.RLock() - defer n.mu.RUnlock() - +// primaryAddresses returns the primary addresses associated with this NIC. +func (n *NIC) primaryAddresses() []tcpip.ProtocolAddress { var addrs []tcpip.ProtocolAddress - for proto, list := range n.mu.primary { - for _, ref := range list { - // Don't include tentative, expired or tempory endpoints - // to avoid confusion and prevent the caller from using - // those. - switch ref.getKind() { - case permanentTentative, permanentExpired, temporary: - continue - } - - addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: proto, - AddressWithPrefix: ref.addrWithPrefix(), - }) + for p, ep := range n.networkEndpoints { + for _, a := range ep.PrimaryAddresses() { + addrs = append(addrs, tcpip.ProtocolAddress{Protocol: p, AddressWithPrefix: a}) } } return addrs @@ -930,147 +360,26 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { // address exists. If no non-deprecated address exists, the first deprecated // address will be returned. func (n *NIC) primaryAddress(proto tcpip.NetworkProtocolNumber) tcpip.AddressWithPrefix { - n.mu.RLock() - defer n.mu.RUnlock() - - list, ok := n.mu.primary[proto] - if !ok { + addressEndpoint := n.primaryEndpoint(proto, "") + if addressEndpoint == nil { return tcpip.AddressWithPrefix{} } - - var deprecatedEndpoint *referencedNetworkEndpoint - for _, ref := range list { - // Don't include tentative, expired or tempory endpoints to avoid confusion - // and prevent the caller from using those. - switch ref.getKind() { - case permanentTentative, permanentExpired, temporary: - continue - } - - if !ref.deprecated { - return ref.addrWithPrefix() - } - - if deprecatedEndpoint == nil { - deprecatedEndpoint = ref - } - } - - if deprecatedEndpoint != nil { - return deprecatedEndpoint.addrWithPrefix() - } - - return tcpip.AddressWithPrefix{} -} - -// insertPrimaryEndpointLocked adds r to n's primary endpoint list as required -// by peb. -// -// n MUST be locked. -func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) { - switch peb { - case CanBePrimaryEndpoint: - n.mu.primary[r.protocol] = append(n.mu.primary[r.protocol], r) - case FirstPrimaryEndpoint: - n.mu.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.mu.primary[r.protocol]...) - } + addr := addressEndpoint.AddressWithPrefix() + addressEndpoint.DecRef() + return addr } -func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { - id := NetworkEndpointID{LocalAddress: r.address()} - - // Nothing to do if the reference has already been replaced with a different - // one. This happens in the case where 1) this endpoint's ref count hit zero - // and was waiting (on the lock) to be removed and 2) the same address was - // re-added in the meantime by removing this endpoint from the list and - // adding a new one. - if n.mu.endpoints[id] != r { - return - } - - if r.getKind() == permanent { - panic("Reference count dropped to zero before being removed") - } - - delete(n.mu.endpoints, id) - refs := n.mu.primary[r.protocol] - for i, ref := range refs { - if ref == r { - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - refs[len(refs)-1] = nil - break - } - } -} - -func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) { - n.mu.Lock() - n.removeEndpointLocked(r) - n.mu.Unlock() -} - -func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { - r, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return tcpip.ErrBadLocalAddress - } - - kind := r.getKind() - if kind != permanent && kind != permanentTentative { - return tcpip.ErrBadLocalAddress - } - - switch r.protocol { - case header.IPv6ProtocolNumber: - return n.removePermanentIPv6EndpointLocked(r, true /* allowSLAACInvalidation */) - default: - r.expireLocked() - return nil - } -} - -func (n *NIC) removePermanentIPv6EndpointLocked(r *referencedNetworkEndpoint, allowSLAACInvalidation bool) *tcpip.Error { - addr := r.addrWithPrefix() - - isIPv6Unicast := header.IsV6UnicastAddress(addr.Address) - - if isIPv6Unicast { - n.mu.ndp.stopDuplicateAddressDetection(addr.Address) - - // If we are removing an address generated via SLAAC, cleanup - // its SLAAC resources and notify the integrator. - switch r.configType { - case slaac: - n.mu.ndp.cleanupSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) - case slaacTemp: - n.mu.ndp.cleanupTempSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) - } - } - - r.expireLocked() - - // At this point the endpoint is deleted. - - // If we are removing an IPv6 unicast address, leave the solicited-node - // multicast address. - // - // We ignore the tcpip.ErrBadLocalAddress error because the solicited-node - // multicast group may be left by user action. - if isIPv6Unicast { - snmc := header.SolicitedNodeAddr(addr.Address) - if err := n.leaveGroupLocked(snmc, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { +// removeAddress removes an address from n. +func (n *NIC) removeAddress(addr tcpip.Address) *tcpip.Error { + for _, ep := range n.networkEndpoints { + if err := ep.RemovePermanentAddress(addr); err == tcpip.ErrBadLocalAddress { + continue + } else { return err } } - return nil -} - -// RemoveAddress removes an address from n. -func (n *NIC) RemoveAddress(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - return n.removePermanentAddressLocked(addr) + return tcpip.ErrBadLocalAddress } func (n *NIC) neighbors() ([]NeighborEntry, *tcpip.Error) { @@ -1121,91 +430,66 @@ func (n *NIC) clearNeighbors() *tcpip.Error { // joinGroup adds a new endpoint for the given multicast address, if none // exists yet. Otherwise it just increments its count. func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - return n.joinGroupLocked(protocol, addr) -} - -// joinGroupLocked adds a new endpoint for the given multicast address, if none -// exists yet. Otherwise it just increments its count. n MUST be locked before -// joinGroupLocked is called. -func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { // TODO(b/143102137): When implementing MLD, make sure MLD packets are // not sent unless a valid link-local address is available for use on n // as an MLD packet's source address must be a link-local address as // outlined in RFC 3810 section 5. - id := NetworkEndpointID{addr} - joins := n.mu.mcastJoins[id] - if joins == 0 { - netProto, ok := n.stack.networkProtocols[protocol] - if !ok { - return tcpip.ErrUnknownProtocol - } - if _, err := n.addAddressLocked(tcpip.ProtocolAddress{ - Protocol: protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: addr, - PrefixLen: netProto.DefaultPrefixLen(), - }, - }, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { - return err - } + ep, ok := n.networkEndpoints[protocol] + if !ok { + return tcpip.ErrNotSupported } - n.mu.mcastJoins[id] = joins + 1 - return nil + + gep, ok := ep.(GroupAddressableEndpoint) + if !ok { + return tcpip.ErrNotSupported + } + + _, err := gep.JoinGroup(addr) + return err } // leaveGroup decrements the count for the given multicast address, and when it // reaches zero removes the endpoint for this address. -func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - return n.leaveGroupLocked(addr, false /* force */) -} +func (n *NIC) leaveGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { + ep, ok := n.networkEndpoints[protocol] + if !ok { + return tcpip.ErrNotSupported + } -// leaveGroupLocked decrements the count for the given multicast address, and -// when it reaches zero removes the endpoint for this address. n MUST be locked -// before leaveGroupLocked is called. -// -// If force is true, then the count for the multicast addres is ignored and the -// endpoint will be removed immediately. -func (n *NIC) leaveGroupLocked(addr tcpip.Address, force bool) *tcpip.Error { - id := NetworkEndpointID{addr} - joins, ok := n.mu.mcastJoins[id] + gep, ok := ep.(GroupAddressableEndpoint) if !ok { - // There are no joins with this address on this NIC. - return tcpip.ErrBadLocalAddress + return tcpip.ErrNotSupported } - joins-- - if force || joins == 0 { - // There are no outstanding joins or we are forced to leave, clean up. - delete(n.mu.mcastJoins, id) - return n.removePermanentAddressLocked(addr) + if _, err := gep.LeaveGroup(addr); err != nil { + return err } - n.mu.mcastJoins[id] = joins return nil } // isInGroup returns true if n has joined the multicast group addr. func (n *NIC) isInGroup(addr tcpip.Address) bool { - n.mu.RLock() - joins := n.mu.mcastJoins[NetworkEndpointID{addr}] - n.mu.RUnlock() + for _, ep := range n.networkEndpoints { + gep, ok := ep.(GroupAddressableEndpoint) + if !ok { + continue + } - return joins != 0 + if gep.IsInGroup(addr) { + return true + } + } + + return false } -func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt *PacketBuffer) { - r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) +func (n *NIC) handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, remotelinkAddr tcpip.LinkAddress, addressEndpoint AssignableAddressEndpoint, pkt *PacketBuffer) { + r := makeRoute(protocol, dst, src, n, addressEndpoint, false /* handleLocal */, false /* multicastLoop */) r.RemoteLinkAddress = remotelinkAddr - - ref.ep.HandlePacket(&r, pkt) - ref.decRef() + addressEndpoint.NetworkEndpoint().HandlePacket(&r, pkt) + addressEndpoint.DecRef() } // DeliverNetworkPacket finds the appropriate network protocol endpoint and @@ -1216,7 +500,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // the ownership of the items is not retained by the caller. func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { n.mu.RLock() - enabled := n.mu.enabled + enabled := n.Enabled() // If the NIC is not yet enabled, don't receive any packets. if !enabled { n.mu.RUnlock() @@ -1274,17 +558,21 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp src, dst := netProto.ParseAddresses(pkt.NetworkHeader().View()) - if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { - // The source address is one of our own, so we never should have gotten a - // packet like this unless handleLocal is false. Loopback also calls this - // function even though the packets didn't come from the physical interface - // so don't drop those. - n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() - return + if n.stack.handleLocal && !n.IsLoopback() { + if r := n.getAddress(protocol, src); r != nil { + r.DecRef() + + // The source address is one of our own, so we never should have gotten a + // packet like this unless handleLocal is false. Loopback also calls this + // function even though the packets didn't come from the physical interface + // so don't drop those. + n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() + return + } } // Loopback traffic skips the prerouting chain. - if !n.isLoopback() { + if !n.IsLoopback() { // iptables filtering. ipt := n.stack.IPTables() address := n.primaryAddress(protocol) @@ -1295,8 +583,8 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } - if ref := n.getRef(protocol, dst); ref != nil { - handlePacket(protocol, dst, src, n.linkEP.LinkAddress(), remote, ref, pkt) + if addressEndpoint := n.getAddress(protocol, dst); addressEndpoint != nil { + n.handlePacket(protocol, dst, src, remote, addressEndpoint, pkt) return } @@ -1312,20 +600,20 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } // Found a NIC. - n := r.ref.nic - n.mu.RLock() - ref, ok := n.mu.endpoints[NetworkEndpointID{dst}] - ok = ok && ref.isValidForOutgoingRLocked() && ref.tryIncRef() - n.mu.RUnlock() - if ok { - r.LocalLinkAddress = n.linkEP.LinkAddress() - r.RemoteLinkAddress = remote - r.RemoteAddress = src - // TODO(b/123449044): Update the source NIC as well. - ref.ep.HandlePacket(&r, pkt) - ref.decRef() - r.Release() - return + n := r.nic + if addressEndpoint := n.getAddressOrCreateTempInner(protocol, dst, false, NeverPrimaryEndpoint); addressEndpoint != nil { + if n.isValidForOutgoing(addressEndpoint) { + r.LocalLinkAddress = n.linkEP.LinkAddress() + r.RemoteLinkAddress = remote + r.RemoteAddress = src + // TODO(b/123449044): Update the source NIC as well. + addressEndpoint.NetworkEndpoint().HandlePacket(&r, pkt) + addressEndpoint.DecRef() + r.Release() + return + } + + addressEndpoint.DecRef() } // n doesn't have a destination endpoint. @@ -1398,11 +686,13 @@ func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) TransportPacketDisposition { state, ok := n.stack.transportProtocols[protocol] if !ok { + // TODO(gvisor.dev/issue/4365): Let the caller know that the transport + // protocol is unrecognized. n.stack.stats.UnknownProtocolRcvdPackets.Increment() - return + return TransportPacketHandled } transProto := state.proto @@ -1423,59 +713,47 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // we parse it using the minimum size. if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok { n.stack.stats.MalformedRcvdPackets.Increment() - return + // We consider a malformed transport packet handled because there is + // nothing the caller can do. + return TransportPacketHandled } - } else { - // This is either a bad packet or was re-assembled from fragments. - transProto.Parse(pkt) + } else if !transProto.Parse(pkt) { + n.stack.stats.MalformedRcvdPackets.Increment() + return TransportPacketHandled } } - if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() { - n.stack.stats.MalformedRcvdPackets.Increment() - return - } - srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() - return + return TransportPacketHandled } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} if n.stack.demux.deliverPacket(r, protocol, pkt, id) { - return + return TransportPacketHandled } // Try to deliver to per-stack default handler. if state.defaultHandler != nil { if state.defaultHandler(r, id, pkt) { - return + return TransportPacketHandled } } // We could not find an appropriate destination for this packet so // give the protocol specific error handler a chance to handle it. // If it doesn't handle it then we should do so. - switch transProto.HandleUnknownDestinationPacket(r, id, pkt) { + switch res := transProto.HandleUnknownDestinationPacket(r, id, pkt); res { case UnknownDestinationPacketMalformed: n.stack.stats.MalformedRcvdPackets.Increment() + return TransportPacketHandled case UnknownDestinationPacketUnhandled: - // As per RFC: 1122 Section 3.2.2.1 A host SHOULD generate Destination - // Unreachable messages with code: - // 3 (Port Unreachable), when the designated transport protocol - // (e.g., UDP) is unable to demultiplex the datagram but has no - // protocol mechanism to inform the sender. - np, ok := n.stack.networkProtocols[r.NetProto] - if !ok { - // For this to happen stack.makeRoute() must have been called with the - // incorrect protocol number. Since we have successfully completed - // network layer processing this should be impossible. - panic(fmt.Sprintf("expected stack to have a NetworkProtocol for proto = %d", r.NetProto)) - } - - _ = np.ReturnError(r, &tcpip.ICMPReasonPortUnreachable{}, pkt) + return TransportPacketDestinationPortUnreachable case UnknownDestinationPacketHandled: + return TransportPacketHandled + default: + panic(fmt.Sprintf("unrecognized result from HandleUnknownDestinationPacket = %d", res)) } } @@ -1508,96 +786,23 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp } } -// ID returns the identifier of n. +// ID implements NetworkInterface. func (n *NIC) ID() tcpip.NICID { return n.id } -// Name returns the name of n. +// Name implements NetworkInterface. func (n *NIC) Name() string { return n.name } -// Stack returns the instance of the Stack that owns this NIC. -func (n *NIC) Stack() *Stack { - return n.stack -} - -// LinkEndpoint returns the link endpoint of n. +// LinkEndpoint implements NetworkInterface. func (n *NIC) LinkEndpoint() LinkEndpoint { return n.linkEP } -// isAddrTentative returns true if addr is tentative on n. -// -// Note that if addr is not associated with n, then this function will return -// false. It will only return true if the address is associated with the NIC -// AND it is tentative. -func (n *NIC) isAddrTentative(addr tcpip.Address) bool { - n.mu.RLock() - defer n.mu.RUnlock() - - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return false - } - - return ref.getKind() == permanentTentative -} - -// dupTentativeAddrDetected attempts to inform n that a tentative addr is a -// duplicate on a link. -// -// dupTentativeAddrDetected will remove the tentative address if it exists. If -// the address was generated via SLAAC, an attempt will be made to generate a -// new address. -func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return tcpip.ErrBadAddress - } - - if ref.getKind() != permanentTentative { - return tcpip.ErrInvalidEndpointState - } - - // If the address is a SLAAC address, do not invalidate its SLAAC prefix as a - // new address will be generated for it. - if err := n.removePermanentIPv6EndpointLocked(ref, false /* allowSLAACInvalidation */); err != nil { - return err - } - - prefix := ref.addrWithPrefix().Subnet() - - switch ref.configType { - case slaac: - n.mu.ndp.regenerateSLAACAddr(prefix) - case slaacTemp: - // Do not reset the generation attempts counter for the prefix as the - // temporary address is being regenerated in response to a DAD conflict. - n.mu.ndp.regenerateTempSLAACAddr(prefix, false /* resetGenAttempts */) - } - - return nil -} - -// setNDPConfigs sets the NDP configurations for n. -// -// Note, if c contains invalid NDP configuration values, it will be fixed to -// use default values for the erroneous values. -func (n *NIC) setNDPConfigs(c NDPConfigurations) { - c.validate() - - n.mu.Lock() - n.mu.ndp.configs = c - n.mu.Unlock() -} - -// NUDConfigs gets the NUD configurations for n. -func (n *NIC) NUDConfigs() (NUDConfigurations, *tcpip.Error) { +// nudConfigs gets the NUD configurations for n. +func (n *NIC) nudConfigs() (NUDConfigurations, *tcpip.Error) { if n.neigh == nil { return NUDConfigurations{}, tcpip.ErrNotSupported } @@ -1617,49 +822,6 @@ func (n *NIC) setNUDConfigs(c NUDConfigurations) *tcpip.Error { return nil } -// handleNDPRA handles an NDP Router Advertisement message that arrived on n. -func (n *NIC) handleNDPRA(ip tcpip.Address, ra header.NDPRouterAdvert) { - n.mu.Lock() - defer n.mu.Unlock() - - n.mu.ndp.handleRA(ip, ra) -} - -type networkEndpointKind int32 - -const ( - // A permanentTentative endpoint is a permanent address that is not yet - // considered to be fully bound to an interface in the traditional - // sense. That is, the address is associated with a NIC, but packets - // destined to the address MUST NOT be accepted and MUST be silently - // dropped, and the address MUST NOT be used as a source address for - // outgoing packets. For IPv6, addresses will be of this kind until - // NDP's Duplicate Address Detection has resolved, or be deleted if - // the process results in detecting a duplicate address. - permanentTentative networkEndpointKind = iota - - // A permanent endpoint is created by adding a permanent address (vs. a - // temporary one) to the NIC. Its reference count is biased by 1 to avoid - // removal when no route holds a reference to it. It is removed by explicitly - // removing the permanent address from the NIC. - permanent - - // An expired permanent endpoint is a permanent endpoint that had its address - // removed from the NIC, and it is waiting to be removed once no more routes - // hold a reference to it. This is achieved by decreasing its reference count - // by 1. If its address is re-added before the endpoint is removed, its type - // changes back to permanent and its reference count increases by 1 again. - permanentExpired - - // A temporary endpoint is created for spoofing outgoing packets, or when in - // promiscuous mode and accepting incoming packets that don't match any - // permanent endpoint. Its reference count is not biased by 1 and the - // endpoint is removed immediately when no more route holds a reference to - // it. A temporary endpoint can be promoted to permanent if its address - // is added permanently. - temporary -) - func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() @@ -1690,153 +852,12 @@ func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep } } -type networkEndpointConfigType int32 - -const ( - // A statically configured endpoint is an address that was added by - // some user-specified action (adding an explicit address, joining a - // multicast group). - static networkEndpointConfigType = iota - - // A SLAAC configured endpoint is an IPv6 endpoint that was added by - // SLAAC as per RFC 4862 section 5.5.3. - slaac - - // A temporary SLAAC configured endpoint is an IPv6 endpoint that was added by - // SLAAC as per RFC 4941. Temporary SLAAC addresses are short-lived and are - // not expected to be valid (or preferred) forever; hence the term temporary. - slaacTemp -) - -type referencedNetworkEndpoint struct { - ep NetworkEndpoint - addr tcpip.AddressWithPrefix - nic *NIC - protocol tcpip.NetworkProtocolNumber - - // linkCache is set if link address resolution is enabled for this - // protocol. Set to nil otherwise. - linkCache LinkAddressCache - - // linkRes is set if link address resolution is enabled for this protocol. - // Set to nil otherwise. - linkRes LinkAddressResolver - - // refs is counting references held for this endpoint. When refs hits zero it - // triggers the automatic removal of the endpoint from the NIC. - refs int32 - - // networkEndpointKind must only be accessed using {get,set}Kind(). - kind networkEndpointKind - - // configType is the method that was used to configure this endpoint. - // This must never change except during endpoint creation and promotion to - // permanent. - configType networkEndpointConfigType - - // deprecated indicates whether or not the endpoint should be considered - // deprecated. That is, when deprecated is true, other endpoints that are not - // deprecated should be preferred. - deprecated bool -} - -func (r *referencedNetworkEndpoint) address() tcpip.Address { - return r.addr.Address -} - -func (r *referencedNetworkEndpoint) addrWithPrefix() tcpip.AddressWithPrefix { - return r.addr -} - -func (r *referencedNetworkEndpoint) getKind() networkEndpointKind { - return networkEndpointKind(atomic.LoadInt32((*int32)(&r.kind))) -} - -func (r *referencedNetworkEndpoint) setKind(kind networkEndpointKind) { - atomic.StoreInt32((*int32)(&r.kind), int32(kind)) -} - // isValidForOutgoing returns true if the endpoint can be used to send out a // packet. It requires the endpoint to not be marked expired (i.e., its address) // has been removed) unless the NIC is in spoofing mode, or temporary. -func (r *referencedNetworkEndpoint) isValidForOutgoing() bool { - r.nic.mu.RLock() - defer r.nic.mu.RUnlock() - - return r.isValidForOutgoingRLocked() -} - -// isValidForOutgoingRLocked is the same as isValidForOutgoing but requires -// r.nic.mu to be read locked. -func (r *referencedNetworkEndpoint) isValidForOutgoingRLocked() bool { - if !r.nic.mu.enabled { - return false - } - - return r.isAssignedRLocked(r.nic.mu.spoofing) -} - -// isAssignedRLocked returns true if r is considered to be assigned to the NIC. -// -// r.nic.mu must be read locked. -func (r *referencedNetworkEndpoint) isAssignedRLocked(spoofingOrPromiscuous bool) bool { - switch r.getKind() { - case permanentTentative: - return false - case permanentExpired: - return spoofingOrPromiscuous - default: - return true - } -} - -// expireLocked decrements the reference count and marks the permanent endpoint -// as expired. -func (r *referencedNetworkEndpoint) expireLocked() { - r.setKind(permanentExpired) - r.decRefLocked() -} - -// decRef decrements the ref count and cleans up the endpoint once it reaches -// zero. -func (r *referencedNetworkEndpoint) decRef() { - if atomic.AddInt32(&r.refs, -1) == 0 { - r.nic.removeEndpoint(r) - } -} - -// decRefLocked is the same as decRef but assumes that the NIC.mu mutex is -// locked. -func (r *referencedNetworkEndpoint) decRefLocked() { - if atomic.AddInt32(&r.refs, -1) == 0 { - r.nic.removeEndpointLocked(r) - } -} - -// incRef increments the ref count. It must only be called when the caller is -// known to be holding a reference to the endpoint, otherwise tryIncRef should -// be used. -func (r *referencedNetworkEndpoint) incRef() { - atomic.AddInt32(&r.refs, 1) -} - -// tryIncRef attempts to increment the ref count from n to n+1, but only if n is -// not zero. That is, it will increment the count if the endpoint is still -// alive, and do nothing if it has already been clean up. -func (r *referencedNetworkEndpoint) tryIncRef() bool { - for { - v := atomic.LoadInt32(&r.refs) - if v == 0 { - return false - } - - if atomic.CompareAndSwapInt32(&r.refs, v, v+1) { - return true - } - } -} - -// stack returns the Stack instance that owns the underlying endpoint. -func (r *referencedNetworkEndpoint) stack() *Stack { - return r.nic.stack +func (n *NIC) isValidForOutgoing(ep AssignableAddressEndpoint) bool { + n.mu.RLock() + spoofing := n.mu.spoofing + n.mu.RUnlock() + return n.Enabled() && ep.IsAssigned(spoofing) } diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index ef6e63b3e..fdd49b77f 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -15,96 +15,40 @@ package stack import ( - "math" "testing" - "time" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" ) -var _ LinkEndpoint = (*testLinkEndpoint)(nil) +var _ AddressableEndpoint = (*testIPv6Endpoint)(nil) +var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) +var _ NDPEndpoint = (*testIPv6Endpoint)(nil) -// A LinkEndpoint that throws away outgoing packets. +// An IPv6 NetworkEndpoint that throws away outgoing packets. // -// We use this instead of the channel endpoint as the channel package depends on +// We use this instead of ipv6.endpoint because the ipv6 package depends on // the stack package which this test lives in, causing a cyclic dependency. -type testLinkEndpoint struct { - dispatcher NetworkDispatcher -} - -// Attach implements LinkEndpoint.Attach. -func (e *testLinkEndpoint) Attach(dispatcher NetworkDispatcher) { - e.dispatcher = dispatcher -} - -// IsAttached implements LinkEndpoint.IsAttached. -func (e *testLinkEndpoint) IsAttached() bool { - return e.dispatcher != nil -} - -// MTU implements LinkEndpoint.MTU. -func (*testLinkEndpoint) MTU() uint32 { - return math.MaxUint16 -} - -// Capabilities implements LinkEndpoint.Capabilities. -func (*testLinkEndpoint) Capabilities() LinkEndpointCapabilities { - return CapabilityResolutionRequired -} +type testIPv6Endpoint struct { + AddressableEndpointState -// MaxHeaderLength implements LinkEndpoint.MaxHeaderLength. -func (*testLinkEndpoint) MaxHeaderLength() uint16 { - return 0 -} + nicID tcpip.NICID + linkEP LinkEndpoint + protocol *testIPv6Protocol -// LinkAddress returns the link address of this endpoint. -func (*testLinkEndpoint) LinkAddress() tcpip.LinkAddress { - return "" + invalidatedRtr tcpip.Address } -// Wait implements LinkEndpoint.Wait. -func (*testLinkEndpoint) Wait() {} - -// WritePacket implements LinkEndpoint.WritePacket. -func (e *testLinkEndpoint) WritePacket(*Route, *GSO, tcpip.NetworkProtocolNumber, *PacketBuffer) *tcpip.Error { +func (*testIPv6Endpoint) Enable() *tcpip.Error { return nil } -// WritePackets implements LinkEndpoint.WritePackets. -func (e *testLinkEndpoint) WritePackets(*Route, *GSO, PacketBufferList, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - // Our tests don't use this so we don't support it. - return 0, tcpip.ErrNotSupported -} - -// WriteRawPacket implements LinkEndpoint.WriteRawPacket. -func (e *testLinkEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { - // Our tests don't use this so we don't support it. - return tcpip.ErrNotSupported +func (*testIPv6Endpoint) Enabled() bool { + return true } -// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. -func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType { - panic("not implemented") -} - -// AddHeader implements stack.LinkEndpoint.AddHeader. -func (e *testLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { - panic("not implemented") -} - -var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) - -// An IPv6 NetworkEndpoint that throws away outgoing packets. -// -// We use this instead of ipv6.endpoint because the ipv6 package depends on -// the stack package which this test lives in, causing a cyclic dependency. -type testIPv6Endpoint struct { - nicID tcpip.NICID - linkEP LinkEndpoint - protocol *testIPv6Protocol -} +func (*testIPv6Endpoint) Disable() {} // DefaultTTL implements NetworkEndpoint.DefaultTTL. func (*testIPv6Endpoint) DefaultTTL() uint8 { @@ -116,11 +60,6 @@ func (e *testIPv6Endpoint) MTU() uint32 { return e.linkEP.MTU() - header.IPv6MinimumSize } -// Capabilities implements NetworkEndpoint.Capabilities. -func (e *testIPv6Endpoint) Capabilities() LinkEndpointCapabilities { - return e.linkEP.Capabilities() -} - // MaxHeaderLength implements NetworkEndpoint.MaxHeaderLength. func (e *testIPv6Endpoint) MaxHeaderLength() uint16 { return e.linkEP.MaxHeaderLength() + header.IPv6MinimumSize @@ -144,23 +83,24 @@ func (*testIPv6Endpoint) WriteHeaderIncludedPacket(*Route, *PacketBuffer) *tcpip return tcpip.ErrNotSupported } -// NICID implements NetworkEndpoint.NICID. -func (e *testIPv6Endpoint) NICID() tcpip.NICID { - return e.nicID -} - // HandlePacket implements NetworkEndpoint.HandlePacket. func (*testIPv6Endpoint) HandlePacket(*Route, *PacketBuffer) { } // Close implements NetworkEndpoint.Close. -func (*testIPv6Endpoint) Close() {} +func (e *testIPv6Endpoint) Close() { + e.AddressableEndpointState.Cleanup() +} // NetworkProtocolNumber implements NetworkEndpoint.NetworkProtocolNumber. func (*testIPv6Endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return header.IPv6ProtocolNumber } +func (e *testIPv6Endpoint) InvalidateDefaultRouter(rtr tcpip.Address) { + e.invalidatedRtr = rtr +} + var _ NetworkProtocol = (*testIPv6Protocol)(nil) // An IPv6 NetworkProtocol that supports the bare minimum to make a stack @@ -192,12 +132,14 @@ func (*testIPv6Protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) } // NewEndpoint implements NetworkProtocol.NewEndpoint. -func (p *testIPv6Protocol) NewEndpoint(nicID tcpip.NICID, _ LinkAddressCache, _ NUDHandler, _ TransportDispatcher, linkEP LinkEndpoint, _ *Stack) NetworkEndpoint { - return &testIPv6Endpoint{ - nicID: nicID, - linkEP: linkEP, +func (p *testIPv6Protocol) NewEndpoint(nic NetworkInterface, _ LinkAddressCache, _ NUDHandler, _ TransportDispatcher) NetworkEndpoint { + e := &testIPv6Endpoint{ + nicID: nic.ID(), + linkEP: nic.LinkEndpoint(), protocol: p, } + e.AddressableEndpointState.Init(e) + return e } // SetOption implements NetworkProtocol.SetOption. @@ -221,11 +163,6 @@ func (*testIPv6Protocol) Parse(*PacketBuffer) (tcpip.TransportProtocolNumber, bo return 0, false, false } -// ReturnError implements NetworkProtocol.ReturnError. -func (*testIPv6Protocol) ReturnError(*Route, tcpip.ICMPReason, *PacketBuffer) *tcpip.Error { - return nil -} - var _ LinkAddressResolver = (*testIPv6Protocol)(nil) // LinkAddressProtocol implements LinkAddressResolver. @@ -246,38 +183,6 @@ func (*testIPv6Protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAdd return "", false } -// Test the race condition where a NIC is removed and an RS timer fires at the -// same time. -func TestRemoveNICWhileHandlingRSTimer(t *testing.T) { - const ( - nicID = 1 - - maxRtrSolicitations = 5 - ) - - e := testLinkEndpoint{} - s := New(Options{ - NetworkProtocols: []NetworkProtocol{&testIPv6Protocol{}}, - NDPConfigs: NDPConfigurations{ - MaxRtrSolicitations: maxRtrSolicitations, - RtrSolicitationInterval: minimumRtrSolicitationInterval, - }, - }) - - if err := s.CreateNIC(nicID, &e); err != nil { - t.Fatalf("s.CreateNIC(%d, _) = %s", nicID, err) - } - - s.mu.Lock() - // Wait for the router solicitation timer to fire and block trying to obtain - // the stack lock when doing link address resolution. - time.Sleep(minimumRtrSolicitationInterval * 2) - if err := s.removeNICLocked(nicID); err != nil { - t.Fatalf("s.removeNICLocked(%d) = %s", nicID, err) - } - s.mu.Unlock() -} - func TestDisabledRxStatsWhenNICDisabled(t *testing.T) { // When the NIC is disabled, the only field that matters is the stats field. // This test is limited to stats counter checks. diff --git a/pkg/tcpip/stack/nud_test.go b/pkg/tcpip/stack/nud_test.go index 2b97e5972..8cffb9fc6 100644 --- a/pkg/tcpip/stack/nud_test.go +++ b/pkg/tcpip/stack/nud_test.go @@ -60,7 +60,7 @@ func TestSetNUDConfigurationFailsForBadNICID(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The networking // stack will only allocate neighbor caches if a protocol providing link // address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, UseNeighborCache: true, }) @@ -137,7 +137,7 @@ func TestDefaultNUDConfigurations(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The networking // stack will only allocate neighbor caches if a protocol providing link // address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: stack.DefaultNUDConfigurations(), UseNeighborCache: true, }) @@ -192,7 +192,7 @@ func TestNUDConfigurationsBaseReachableTime(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -249,7 +249,7 @@ func TestNUDConfigurationsMinRandomFactor(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -329,7 +329,7 @@ func TestNUDConfigurationsMaxRandomFactor(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -391,7 +391,7 @@ func TestNUDConfigurationsRetransmitTimer(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -443,7 +443,7 @@ func TestNUDConfigurationsDelayFirstProbeTime(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -495,7 +495,7 @@ func TestNUDConfigurationsMaxMulticastProbes(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -547,7 +547,7 @@ func TestNUDConfigurationsMaxUnicastProbes(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -599,7 +599,7 @@ func TestNUDConfigurationsUnreachableTime(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 77640cd8a..567e1904e 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -15,6 +15,8 @@ package stack import ( + "fmt" + "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -152,10 +154,10 @@ type TransportProtocol interface { Number() tcpip.TransportProtocolNumber // NewEndpoint creates a new endpoint of the transport protocol. - NewEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) + NewEndpoint(netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) // NewRawEndpoint creates a new raw endpoint of the transport protocol. - NewRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) + NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) // MinimumPacketSize returns the minimum valid packet size of this // transport protocol. The stack automatically drops any packets smaller @@ -197,6 +199,21 @@ type TransportProtocol interface { Parse(pkt *PacketBuffer) (ok bool) } +// TransportPacketDisposition is the result from attempting to deliver a packet +// to the transport layer. +type TransportPacketDisposition int + +const ( + // TransportPacketHandled indicates that a transport packet was handled by the + // transport layer and callers need not take any further action. + TransportPacketHandled TransportPacketDisposition = iota + + // TransportPacketDestinationPortUnreachable indicates that there weren't any + // listeners interested in the packet and the transport protocol has no means + // to notify the sender. + TransportPacketDestinationPortUnreachable +) + // TransportDispatcher contains the methods used by the network stack to deliver // packets to the appropriate transport endpoint after it has been handled by // the network layer. @@ -207,7 +224,7 @@ type TransportDispatcher interface { // pkt.NetworkHeader must be set before calling DeliverTransportPacket. // // DeliverTransportPacket takes ownership of pkt. - DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) + DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) TransportPacketDisposition // DeliverTransportControlPacket delivers control packets to the // appropriate transport protocol endpoint. @@ -244,9 +261,254 @@ type NetworkHeaderParams struct { TOS uint8 } +// GroupAddressableEndpoint is an endpoint that supports group addressing. +// +// An endpoint is considered to support group addressing when one or more +// endpoints may associate themselves with the same identifier (group address). +type GroupAddressableEndpoint interface { + // JoinGroup joins the spcified group. + // + // Returns true if the group was newly joined. + JoinGroup(group tcpip.Address) (bool, *tcpip.Error) + + // LeaveGroup attempts to leave the specified group. + // + // Returns tcpip.ErrBadLocalAddress if the endpoint has not joined the group. + LeaveGroup(group tcpip.Address) (bool, *tcpip.Error) + + // IsInGroup returns true if the endpoint is a member of the specified group. + IsInGroup(group tcpip.Address) bool +} + +// PrimaryEndpointBehavior is an enumeration of an AddressEndpoint's primary +// behavior. +type PrimaryEndpointBehavior int + +const ( + // CanBePrimaryEndpoint indicates the endpoint can be used as a primary + // endpoint for new connections with no local address. This is the + // default when calling NIC.AddAddress. + CanBePrimaryEndpoint PrimaryEndpointBehavior = iota + + // FirstPrimaryEndpoint indicates the endpoint should be the first + // primary endpoint considered. If there are multiple endpoints with + // this behavior, they are ordered by recency. + FirstPrimaryEndpoint + + // NeverPrimaryEndpoint indicates the endpoint should never be a + // primary endpoint. + NeverPrimaryEndpoint +) + +// AddressConfigType is the method used to add an address. +type AddressConfigType int + +const ( + // AddressConfigStatic is a statically configured address endpoint that was + // added by some user-specified action (adding an explicit address, joining a + // multicast group). + AddressConfigStatic AddressConfigType = iota + + // AddressConfigSlaac is an address endpoint added by SLAAC, as per RFC 4862 + // section 5.5.3. + AddressConfigSlaac + + // AddressConfigSlaacTemp is a temporary address endpoint added by SLAAC as + // per RFC 4941. Temporary SLAAC addresses are short-lived and are not + // to be valid (or preferred) forever; hence the term temporary. + AddressConfigSlaacTemp +) + +// AssignableAddressEndpoint is a reference counted address endpoint that may be +// assigned to a NetworkEndpoint. +type AssignableAddressEndpoint interface { + // NetworkEndpoint returns the NetworkEndpoint the receiver is associated + // with. + NetworkEndpoint() NetworkEndpoint + + // AddressWithPrefix returns the endpoint's address. + AddressWithPrefix() tcpip.AddressWithPrefix + + // IsAssigned returns whether or not the endpoint is considered bound + // to its NetworkEndpoint. + IsAssigned(allowExpired bool) bool + + // IncRef increments this endpoint's reference count. + // + // Returns true if it was successfully incremented. If it returns false, then + // the endpoint is considered expired and should no longer be used. + IncRef() bool + + // DecRef decrements this endpoint's reference count. + DecRef() +} + +// AddressEndpoint is an endpoint representing an address assigned to an +// AddressableEndpoint. +type AddressEndpoint interface { + AssignableAddressEndpoint + + // GetKind returns the address kind for this endpoint. + GetKind() AddressKind + + // SetKind sets the address kind for this endpoint. + SetKind(AddressKind) + + // ConfigType returns the method used to add the address. + ConfigType() AddressConfigType + + // Deprecated returns whether or not this endpoint is deprecated. + Deprecated() bool + + // SetDeprecated sets this endpoint's deprecated status. + SetDeprecated(bool) +} + +// AddressKind is the kind of of an address. +// +// See the values of AddressKind for more details. +type AddressKind int + +const ( + // PermanentTentative is a permanent address endpoint that is not yet + // considered to be fully bound to an interface in the traditional + // sense. That is, the address is associated with a NIC, but packets + // destined to the address MUST NOT be accepted and MUST be silently + // dropped, and the address MUST NOT be used as a source address for + // outgoing packets. For IPv6, addresses are of this kind until NDP's + // Duplicate Address Detection (DAD) resolves. If DAD fails, the address + // is removed. + PermanentTentative AddressKind = iota + + // Permanent is a permanent endpoint (vs. a temporary one) assigned to the + // NIC. Its reference count is biased by 1 to avoid removal when no route + // holds a reference to it. It is removed by explicitly removing the address + // from the NIC. + Permanent + + // PermanentExpired is a permanent endpoint that had its address removed from + // the NIC, and it is waiting to be removed once no references to it are held. + // + // If the address is re-added before the endpoint is removed, its type + // changes back to Permanent. + PermanentExpired + + // Temporary is an endpoint, created on a one-off basis to temporarily + // consider the NIC bound an an address that it is not explictiy bound to + // (such as a permanent address). Its reference count must not be biased by 1 + // so that the address is removed immediately when references to it are no + // longer held. + // + // A temporary endpoint may be promoted to permanent if the address is added + // permanently. + Temporary +) + +// IsPermanent returns true if the AddressKind represents a permanent address. +func (k AddressKind) IsPermanent() bool { + switch k { + case Permanent, PermanentTentative: + return true + case Temporary, PermanentExpired: + return false + default: + panic(fmt.Sprintf("unrecognized address kind = %d", k)) + } +} + +// AddressableEndpoint is an endpoint that supports addressing. +// +// An endpoint is considered to support addressing when the endpoint may +// associate itself with an identifier (address). +type AddressableEndpoint interface { + // AddAndAcquirePermanentAddress adds the passed permanent address. + // + // Returns tcpip.ErrDuplicateAddress if the address exists. + // + // Acquires and returns the AddressEndpoint for the added address. + AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior, configType AddressConfigType, deprecated bool) (AddressEndpoint, *tcpip.Error) + + // RemovePermanentAddress removes the passed address if it is a permanent + // address. + // + // Returns tcpip.ErrBadLocalAddress if the endpoint does not have the passed + // permanent address. + RemovePermanentAddress(addr tcpip.Address) *tcpip.Error + + // AcquireAssignedAddress returns an AddressEndpoint for the passed address + // that is considered bound to the endpoint, optionally creating a temporary + // endpoint if requested and no existing address exists. + // + // The returned endpoint's reference count is incremented. + // + // Returns nil if the specified address is not local to this endpoint. + AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint + + // AcquirePrimaryAddress returns a primary endpoint to use when communicating + // with the passed remote address. + // + // If allowExpired is true, expired addresses may be returned. + // + // The returned endpoint's reference count is incremented. + // + // Returns nil if a primary endpoint is not available. + AcquirePrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) AddressEndpoint + + // PrimaryAddresses returns the primary addresses. + PrimaryAddresses() []tcpip.AddressWithPrefix + + // PermanentAddresses returns all the permanent addresses. + PermanentAddresses() []tcpip.AddressWithPrefix +} + +// NDPEndpoint is a network endpoint that supports NDP. +type NDPEndpoint interface { + NetworkEndpoint + + // InvalidateDefaultRouter invalidates a default router discovered through + // NDP. + InvalidateDefaultRouter(tcpip.Address) +} + +// NetworkInterface is a network interface. +type NetworkInterface interface { + // ID returns the interface's ID. + ID() tcpip.NICID + + // IsLoopback returns true if the interface is a loopback interface. + IsLoopback() bool + + // Name returns the name of the interface. + // + // May return an empty string if the interface is not configured with a name. + Name() string + + // Enabled returns true if the interface is enabled. + Enabled() bool + + // LinkEndpoint returns the link endpoint backing the interface. + LinkEndpoint() LinkEndpoint +} + // NetworkEndpoint is the interface that needs to be implemented by endpoints // of network layer protocols (e.g., ipv4, ipv6). type NetworkEndpoint interface { + AddressableEndpoint + + // Enable enables the endpoint. + // + // Must only be called when the stack is in a state that allows the endpoint + // to send and receive packets. + // + // Returns tcpip.ErrNotPermitted if the endpoint cannot be enabled. + Enable() *tcpip.Error + + // Enabled returns true if the endpoint is enabled. + Enabled() bool + + // Disable disables the endpoint. + Disable() + // DefaultTTL is the default time-to-live value (or hop limit, in ipv6) // for this endpoint. DefaultTTL() uint8 @@ -256,10 +518,6 @@ type NetworkEndpoint interface { // minus the network endpoint max header length. MTU() uint32 - // Capabilities returns the set of capabilities supported by the - // underlying link-layer endpoint. - Capabilities() LinkEndpointCapabilities - // MaxHeaderLength returns the maximum size the network (and lower // level layers combined) headers can have. Higher levels use this // information to reserve space in the front of the packets they're @@ -280,9 +538,6 @@ type NetworkEndpoint interface { // header to the given destination address. It takes ownership of pkt. WriteHeaderIncludedPacket(r *Route, pkt *PacketBuffer) *tcpip.Error - // NICID returns the id of the NIC this endpoint belongs to. - NICID() tcpip.NICID - // HandlePacket is called by the link layer when new packets arrive to // this network endpoint. It sets pkt.NetworkHeader. // @@ -297,6 +552,17 @@ type NetworkEndpoint interface { NetworkProtocolNumber() tcpip.NetworkProtocolNumber } +// ForwardingNetworkProtocol is a NetworkProtocol that may forward packets. +type ForwardingNetworkProtocol interface { + NetworkProtocol + + // Forwarding returns the forwarding configuration. + Forwarding() bool + + // SetForwarding sets the forwarding configuration. + SetForwarding(bool) +} + // NetworkProtocol is the interface that needs to be implemented by network // protocols (e.g., ipv4, ipv6) that want to be part of the networking stack. type NetworkProtocol interface { @@ -316,7 +582,7 @@ type NetworkProtocol interface { ParseAddresses(v buffer.View) (src, dst tcpip.Address) // NewEndpoint creates a new endpoint of this protocol. - NewEndpoint(nicID tcpip.NICID, linkAddrCache LinkAddressCache, nud NUDHandler, dispatcher TransportDispatcher, sender LinkEndpoint, st *Stack) NetworkEndpoint + NewEndpoint(nic NetworkInterface, linkAddrCache LinkAddressCache, nud NUDHandler, dispatcher TransportDispatcher) NetworkEndpoint // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -342,19 +608,6 @@ type NetworkProtocol interface { // does not encapsulate anything). // - Whether pkt.Data was large enough to parse and set pkt.NetworkHeader. Parse(pkt *PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) - - // ReturnError attempts to send a suitable error message to the sender - // of a received packet. - // - pkt holds the problematic packet. - // - reason indicates what the reason for wanting a message is. - // - route is the routing information for the received packet - // ReturnError returns an error if the send failed and nil on success. - // Note that deciding to deliberately send no message is a success. - // - // TODO(gvisor.dev/issues/3871): This method should be removed or simplified - // after all (or all but one) of the ICMP error dispatch occurs through the - // protocol specific modules. May become SendPortNotFound(r, pkt). - ReturnError(r *Route, reason tcpip.ICMPReason, pkt *PacketBuffer) *tcpip.Error } // NetworkDispatcher contains the methods used by the network stack to deliver @@ -458,8 +711,8 @@ type LinkEndpoint interface { // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. // - // Attach will be called with a nil dispatcher if the receiver's associated - // NIC is being removed. + // Attach is called with a nil dispatcher when the endpoint's NIC is being + // removed. Attach(dispatcher NetworkDispatcher) // IsAttached returns whether a NetworkDispatcher is attached to the diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 2cbbf0de8..5ade3c832 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -42,17 +42,27 @@ type Route struct { // NetProto is the network-layer protocol. NetProto tcpip.NetworkProtocolNumber - // ref a reference to the network endpoint through which the route - // starts. - ref *referencedNetworkEndpoint - // Loop controls where WritePacket should send packets. Loop PacketLooping + + // nic is the NIC the route goes through. + nic *NIC + + // addressEndpoint is the local address this route is associated with. + addressEndpoint AssignableAddressEndpoint + + // linkCache is set if link address resolution is enabled for this protocol on + // the route's NIC. + linkCache LinkAddressCache + + // linkRes is set if link address resolution is enabled for this protocol on + // the route's NIC. + linkRes LinkAddressResolver } // makeRoute initializes a new route. It takes ownership of the provided -// reference to a network endpoint. -func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, localLinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, handleLocal, multicastLoop bool) Route { +// AssignableAddressEndpoint. +func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, nic *NIC, addressEndpoint AssignableAddressEndpoint, handleLocal, multicastLoop bool) Route { loop := PacketOut if handleLocal && localAddr != "" && remoteAddr == localAddr { loop = PacketLoop @@ -62,29 +72,40 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip loop |= PacketLoop } - return Route{ + linkEP := nic.LinkEndpoint() + r := Route{ NetProto: netProto, LocalAddress: localAddr, - LocalLinkAddress: localLinkAddr, + LocalLinkAddress: linkEP.LinkAddress(), RemoteAddress: remoteAddr, - ref: ref, + addressEndpoint: addressEndpoint, + nic: nic, Loop: loop, } + + if nic := r.nic; linkEP.Capabilities()&CapabilityResolutionRequired != 0 { + if linkRes, ok := nic.stack.linkAddrResolvers[r.NetProto]; ok { + r.linkRes = linkRes + r.linkCache = nic.stack + } + } + + return r } // NICID returns the id of the NIC from which this route originates. func (r *Route) NICID() tcpip.NICID { - return r.ref.ep.NICID() + return r.nic.ID() } // MaxHeaderLength forwards the call to the network endpoint's implementation. func (r *Route) MaxHeaderLength() uint16 { - return r.ref.ep.MaxHeaderLength() + return r.addressEndpoint.NetworkEndpoint().MaxHeaderLength() } // Stats returns a mutable copy of current stats. func (r *Route) Stats() tcpip.Stats { - return r.ref.nic.stack.Stats() + return r.nic.stack.Stats() } // PseudoHeaderChecksum forwards the call to the network endpoint's @@ -95,12 +116,12 @@ func (r *Route) PseudoHeaderChecksum(protocol tcpip.TransportProtocolNumber, tot // Capabilities returns the link-layer capabilities of the route. func (r *Route) Capabilities() LinkEndpointCapabilities { - return r.ref.ep.Capabilities() + return r.nic.LinkEndpoint().Capabilities() } // GSOMaxSize returns the maximum GSO packet size. func (r *Route) GSOMaxSize() uint32 { - if gso, ok := r.ref.ep.(GSOEndpoint); ok { + if gso, ok := r.addressEndpoint.NetworkEndpoint().(GSOEndpoint); ok { return gso.GSOMaxSize() } return 0 @@ -138,8 +159,8 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { nextAddr = r.RemoteAddress } - if r.ref.nic.neigh != nil { - entry, ch, err := r.ref.nic.neigh.entry(nextAddr, r.LocalAddress, r.ref.linkRes, waker) + if neigh := r.nic.neigh; neigh != nil { + entry, ch, err := neigh.entry(nextAddr, r.LocalAddress, r.linkRes, waker) if err != nil { return ch, err } @@ -147,7 +168,7 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { return nil, nil } - linkAddr, ch, err := r.ref.linkCache.GetLinkAddress(r.ref.nic.ID(), nextAddr, r.LocalAddress, r.NetProto, waker) + linkAddr, ch, err := r.linkCache.GetLinkAddress(r.nic.ID(), nextAddr, r.LocalAddress, r.NetProto, waker) if err != nil { return ch, err } @@ -162,12 +183,12 @@ func (r *Route) RemoveWaker(waker *sleep.Waker) { nextAddr = r.RemoteAddress } - if r.ref.nic.neigh != nil { - r.ref.nic.neigh.removeWaker(nextAddr, waker) + if neigh := r.nic.neigh; neigh != nil { + neigh.removeWaker(nextAddr, waker) return } - r.ref.linkCache.RemoveWaker(r.ref.nic.ID(), nextAddr, waker) + r.linkCache.RemoveWaker(r.nic.ID(), nextAddr, waker) } // IsResolutionRequired returns true if Resolve() must be called to resolve @@ -175,27 +196,27 @@ func (r *Route) RemoveWaker(waker *sleep.Waker) { // // The NIC r uses must not be locked. func (r *Route) IsResolutionRequired() bool { - if r.ref.nic.neigh != nil { - return r.ref.isValidForOutgoing() && r.ref.linkRes != nil && r.RemoteLinkAddress == "" + if r.nic.neigh != nil { + return r.nic.isValidForOutgoing(r.addressEndpoint) && r.linkRes != nil && r.RemoteLinkAddress == "" } - return r.ref.isValidForOutgoing() && r.ref.linkCache != nil && r.RemoteLinkAddress == "" + return r.nic.isValidForOutgoing(r.addressEndpoint) && r.linkCache != nil && r.RemoteLinkAddress == "" } // WritePacket writes the packet through the given route. func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error { - if !r.ref.isValidForOutgoing() { + if !r.nic.isValidForOutgoing(r.addressEndpoint) { return tcpip.ErrInvalidEndpointState } // WritePacket takes ownership of pkt, calculate numBytes first. numBytes := pkt.Size() - err := r.ref.ep.WritePacket(r, gso, params, pkt) + err := r.addressEndpoint.NetworkEndpoint().WritePacket(r, gso, params, pkt) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { - r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) + r.nic.stats.Tx.Packets.Increment() + r.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) } return err } @@ -203,76 +224,75 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuf // WritePackets writes a list of n packets through the given route and returns // the number of packets written. func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (int, *tcpip.Error) { - if !r.ref.isValidForOutgoing() { + if !r.nic.isValidForOutgoing(r.addressEndpoint) { return 0, tcpip.ErrInvalidEndpointState } // WritePackets takes ownership of pkt, calculate length first. numPkts := pkts.Len() - n, err := r.ref.ep.WritePackets(r, gso, pkts, params) + n, err := r.addressEndpoint.NetworkEndpoint().WritePackets(r, gso, pkts, params) if err != nil { r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(numPkts - n)) } - r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) + r.nic.stats.Tx.Packets.IncrementBy(uint64(n)) writtenBytes := 0 for i, pb := 0, pkts.Front(); i < n && pb != nil; i, pb = i+1, pb.Next() { writtenBytes += pb.Size() } - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) + r.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) return n, err } // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (r *Route) WriteHeaderIncludedPacket(pkt *PacketBuffer) *tcpip.Error { - if !r.ref.isValidForOutgoing() { + if !r.nic.isValidForOutgoing(r.addressEndpoint) { return tcpip.ErrInvalidEndpointState } // WriteHeaderIncludedPacket takes ownership of pkt, calculate numBytes first. numBytes := pkt.Data.Size() - if err := r.ref.ep.WriteHeaderIncludedPacket(r, pkt); err != nil { + if err := r.addressEndpoint.NetworkEndpoint().WriteHeaderIncludedPacket(r, pkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return err } - r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) + r.nic.stats.Tx.Packets.Increment() + r.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) return nil } // DefaultTTL returns the default TTL of the underlying network endpoint. func (r *Route) DefaultTTL() uint8 { - return r.ref.ep.DefaultTTL() + return r.addressEndpoint.NetworkEndpoint().DefaultTTL() } // MTU returns the MTU of the underlying network endpoint. func (r *Route) MTU() uint32 { - return r.ref.ep.MTU() + return r.addressEndpoint.NetworkEndpoint().MTU() } // NetworkProtocolNumber returns the NetworkProtocolNumber of the underlying // network endpoint. func (r *Route) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { - return r.ref.ep.NetworkProtocolNumber() + return r.addressEndpoint.NetworkEndpoint().NetworkProtocolNumber() } // Release frees all resources associated with the route. func (r *Route) Release() { - if r.ref != nil { - r.ref.decRef() - r.ref = nil + if r.addressEndpoint != nil { + r.addressEndpoint.DecRef() + r.addressEndpoint = nil } } -// Clone Clone a route such that the original one can be released and the new -// one will remain valid. +// Clone clones the route. func (r *Route) Clone() Route { - if r.ref != nil { - r.ref.incRef() + if r.addressEndpoint != nil { + _ = r.addressEndpoint.IncRef() } return *r } @@ -296,7 +316,7 @@ func (r *Route) MakeLoopedRoute() Route { // Stack returns the instance of the Stack that owns this route. func (r *Route) Stack() *Stack { - return r.ref.stack() + return r.nic.stack } func (r *Route) isV4Broadcast(addr tcpip.Address) bool { @@ -304,7 +324,7 @@ func (r *Route) isV4Broadcast(addr tcpip.Address) bool { return true } - subnet := r.ref.addrWithPrefix().Subnet() + subnet := r.addressEndpoint.AddressWithPrefix().Subnet() return subnet.IsBroadcast(addr) } @@ -330,7 +350,10 @@ func (r *Route) ReverseRoute(src tcpip.Address, dst tcpip.Address) Route { LocalLinkAddress: r.RemoteLinkAddress, RemoteAddress: src, RemoteLinkAddress: r.LocalLinkAddress, - ref: r.ref, Loop: r.Loop, + addressEndpoint: r.addressEndpoint, + nic: r.nic, + linkCache: r.linkCache, + linkRes: r.linkRes, } } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 68cf77de2..57d8e79e0 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -144,10 +144,7 @@ type TCPReceiverState struct { // PendingBufUsed is the number of bytes pending in the receive // queue. - PendingBufUsed seqnum.Size - - // PendingBufSize is the size of the socket receive buffer. - PendingBufSize seqnum.Size + PendingBufUsed int } // TCPSenderState holds a copy of the internal state of the sender for @@ -366,38 +363,6 @@ func (u *uniqueIDGenerator) UniqueID() uint64 { return atomic.AddUint64((*uint64)(u), 1) } -// NICNameFromID is a function that returns a stable name for the specified NIC, -// even if different NIC IDs are used to refer to the same NIC in different -// program runs. It is used when generating opaque interface identifiers (IIDs). -// If the NIC was created with a name, it will be passed to NICNameFromID. -// -// NICNameFromID SHOULD return unique NIC names so unique opaque IIDs are -// generated for the same prefix on differnt NICs. -type NICNameFromID func(tcpip.NICID, string) string - -// OpaqueInterfaceIdentifierOptions holds the options related to the generation -// of opaque interface indentifiers (IIDs) as defined by RFC 7217. -type OpaqueInterfaceIdentifierOptions struct { - // NICNameFromID is a function that returns a stable name for a specified NIC, - // even if the NIC ID changes over time. - // - // Must be specified to generate the opaque IID. - NICNameFromID NICNameFromID - - // SecretKey is a pseudo-random number used as the secret key when generating - // opaque IIDs as defined by RFC 7217. The key SHOULD be at least - // header.OpaqueIIDSecretKeyMinBytes bytes and MUST follow minimum randomness - // requirements for security as outlined by RFC 4086. SecretKey MUST NOT - // change between program runs, unless explicitly changed. - // - // OpaqueInterfaceIdentifierOptions takes ownership of SecretKey. SecretKey - // MUST NOT be modified after Stack is created. - // - // May be nil, but a nil value is highly discouraged to maintain - // some level of randomness between nodes. - SecretKey []byte -} - // Stack is a networking stack, with all supported protocols, NICs, and route // table. type Stack struct { @@ -405,13 +370,6 @@ type Stack struct { networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver - // forwarding contains the whether packet forwarding is enabled or not for - // different network protocols. - forwarding struct { - sync.RWMutex - protocols map[tcpip.NetworkProtocolNumber]bool - } - // rawFactory creates raw endpoints. If nil, raw endpoints are // disabled. It is set during Stack creation and is immutable. rawFactory RawFactory @@ -464,9 +422,6 @@ type Stack struct { // TODO(gvisor.dev/issue/940): S/R this field. seed uint32 - // ndpConfigs is the default NDP configurations used by interfaces. - ndpConfigs NDPConfigurations - // nudConfigs is the default NUD configurations used by interfaces. nudConfigs NUDConfigurations @@ -474,15 +429,6 @@ type Stack struct { // by the NIC's neighborCache instead of linkAddrCache. useNeighborCache bool - // autoGenIPv6LinkLocal determines whether or not the stack will attempt - // to auto-generate an IPv6 link-local address for newly enabled non-loopback - // NICs. See the AutoGenIPv6LinkLocal field of Options for more details. - autoGenIPv6LinkLocal bool - - // ndpDisp is the NDP event dispatcher that is used to send the netstack - // integrator NDP related events. - ndpDisp NDPDispatcher - // nudDisp is the NUD event dispatcher that is used to send the netstack // integrator NUD related events. nudDisp NUDDispatcher @@ -490,14 +436,6 @@ type Stack struct { // uniqueIDGenerator is a generator of unique identifiers. uniqueIDGenerator UniqueID - // opaqueIIDOpts hold the options for generating opaque interface identifiers - // (IIDs) as outlined by RFC 7217. - opaqueIIDOpts OpaqueInterfaceIdentifierOptions - - // tempIIDSeed is used to seed the initial temporary interface identifier - // history value used to generate IIDs for temporary SLAAC addresses. - tempIIDSeed []byte - // forwarder holds the packets that wait for their link-address resolutions // to complete, and forwards them when each resolution is done. forwarder *forwardQueue @@ -520,13 +458,25 @@ type UniqueID interface { UniqueID() uint64 } +// NetworkProtocolFactory instantiates a network protocol. +// +// NetworkProtocolFactory must not attempt to modify the stack, it may only +// query the stack. +type NetworkProtocolFactory func(*Stack) NetworkProtocol + +// TransportProtocolFactory instantiates a transport protocol. +// +// TransportProtocolFactory must not attempt to modify the stack, it may only +// query the stack. +type TransportProtocolFactory func(*Stack) TransportProtocol + // Options contains optional Stack configuration. type Options struct { // NetworkProtocols lists the network protocols to enable. - NetworkProtocols []NetworkProtocol + NetworkProtocols []NetworkProtocolFactory // TransportProtocols lists the transport protocols to enable. - TransportProtocols []TransportProtocol + TransportProtocols []TransportProtocolFactory // Clock is an optional clock source used for timestampping packets. // @@ -544,13 +494,6 @@ type Options struct { // UniqueID is an optional generator of unique identifiers. UniqueID UniqueID - // NDPConfigs is the default NDP configurations used by interfaces. - // - // By default, NDPConfigs will have a zero value for its - // DupAddrDetectTransmits field, implying that DAD will not be performed - // before assigning an address to a NIC. - NDPConfigs NDPConfigurations - // NUDConfigs is the default NUD configurations used by interfaces. NUDConfigs NUDConfigurations @@ -561,24 +504,6 @@ type Options struct { // and ClearNeighbors. UseNeighborCache bool - // AutoGenIPv6LinkLocal determines whether or not the stack will attempt to - // auto-generate an IPv6 link-local address for newly enabled non-loopback - // NICs. - // - // Note, setting this to true does not mean that a link-local address - // will be assigned right away, or at all. If Duplicate Address Detection - // is enabled, an address will only be assigned if it successfully resolves. - // If it fails, no further attempt will be made to auto-generate an IPv6 - // link-local address. - // - // The generated link-local address will follow RFC 4291 Appendix A - // guidelines. - AutoGenIPv6LinkLocal bool - - // NDPDisp is the NDP event dispatcher that an integrator can provide to - // receive NDP related events. - NDPDisp NDPDispatcher - // NUDDisp is the NUD event dispatcher that an integrator can provide to // receive NUD related events. NUDDisp NUDDispatcher @@ -587,31 +512,12 @@ type Options struct { // this is non-nil. RawFactory RawFactory - // OpaqueIIDOpts hold the options for generating opaque interface - // identifiers (IIDs) as outlined by RFC 7217. - OpaqueIIDOpts OpaqueInterfaceIdentifierOptions - // RandSource is an optional source to use to generate random // numbers. If omitted it defaults to a Source seeded by the data // returned by rand.Read(). // // RandSource must be thread-safe. RandSource mathrand.Source - - // TempIIDSeed is used to seed the initial temporary interface identifier - // history value used to generate IIDs for temporary SLAAC addresses. - // - // Temporary SLAAC adresses are short-lived addresses which are unpredictable - // and random from the perspective of other nodes on the network. It is - // recommended that the seed be a random byte buffer of at least - // header.IIDSize bytes to make sure that temporary SLAAC addresses are - // sufficiently random. It should follow minimum randomness requirements for - // security as outlined by RFC 4086. - // - // Note: using a nil value, the same seed across netstack program runs, or a - // seed that is too small would reduce randomness and increase predictability, - // defeating the purpose of temporary SLAAC addresses. - TempIIDSeed []byte } // TransportEndpointInfo holds useful information about a transport endpoint @@ -714,36 +620,28 @@ func New(opts Options) *Stack { randSrc = &lockedRandomSource{src: mathrand.NewSource(generateRandInt64())} } - // Make sure opts.NDPConfigs contains valid values only. - opts.NDPConfigs.validate() - opts.NUDConfigs.resetInvalidFields() s := &Stack{ - transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), - networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), - linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), - nics: make(map[tcpip.NICID]*NIC), - cleanupEndpoints: make(map[TransportEndpoint]struct{}), - linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), - PortManager: ports.NewPortManager(), - clock: clock, - stats: opts.Stats.FillIn(), - handleLocal: opts.HandleLocal, - tables: DefaultTables(), - icmpRateLimiter: NewICMPRateLimiter(), - seed: generateRandUint32(), - ndpConfigs: opts.NDPConfigs, - nudConfigs: opts.NUDConfigs, - useNeighborCache: opts.UseNeighborCache, - autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, - uniqueIDGenerator: opts.UniqueID, - ndpDisp: opts.NDPDisp, - nudDisp: opts.NUDDisp, - opaqueIIDOpts: opts.OpaqueIIDOpts, - tempIIDSeed: opts.TempIIDSeed, - forwarder: newForwardQueue(), - randomGenerator: mathrand.New(randSrc), + transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), + networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), + linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), + nics: make(map[tcpip.NICID]*NIC), + cleanupEndpoints: make(map[TransportEndpoint]struct{}), + linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), + PortManager: ports.NewPortManager(), + clock: clock, + stats: opts.Stats.FillIn(), + handleLocal: opts.HandleLocal, + tables: DefaultTables(), + icmpRateLimiter: NewICMPRateLimiter(), + seed: generateRandUint32(), + nudConfigs: opts.NUDConfigs, + useNeighborCache: opts.UseNeighborCache, + uniqueIDGenerator: opts.UniqueID, + nudDisp: opts.NUDDisp, + forwarder: newForwardQueue(), + randomGenerator: mathrand.New(randSrc), sendBufferSize: SendBufferSizeOption{ Min: MinBufferSize, Default: DefaultBufferSize, @@ -755,10 +653,10 @@ func New(opts Options) *Stack { Max: DefaultMaxBufferSize, }, } - s.forwarding.protocols = make(map[tcpip.NetworkProtocolNumber]bool) // Add specified network protocols. - for _, netProto := range opts.NetworkProtocols { + for _, netProtoFactory := range opts.NetworkProtocols { + netProto := netProtoFactory(s) s.networkProtocols[netProto.Number()] = netProto if r, ok := netProto.(LinkAddressResolver); ok { s.linkAddrResolvers[r.LinkAddressProtocol()] = r @@ -766,7 +664,8 @@ func New(opts Options) *Stack { } // Add specified transport protocols. - for _, transProto := range opts.TransportProtocols { + for _, transProtoFactory := range opts.TransportProtocols { + transProto := transProtoFactory(s) s.transportProtocols[transProto.Number()] = &transportProtocolState{ proto: transProto, } @@ -873,42 +772,37 @@ func (s *Stack) Stats() tcpip.Stats { return s.stats } -// SetForwarding enables or disables packet forwarding between NICs. -func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) { - s.forwarding.Lock() - defer s.forwarding.Unlock() - - // If this stack does not support the protocol, do nothing. - if _, ok := s.networkProtocols[protocol]; !ok { - return +// SetForwarding enables or disables packet forwarding between NICs for the +// passed protocol. +func (s *Stack) SetForwarding(protocolNum tcpip.NetworkProtocolNumber, enable bool) *tcpip.Error { + protocol, ok := s.networkProtocols[protocolNum] + if !ok { + return tcpip.ErrUnknownProtocol } - // If the forwarding value for this protocol hasn't changed then do - // nothing. - if forwarding := s.forwarding.protocols[protocol]; forwarding == enable { - return + forwardingProtocol, ok := protocol.(ForwardingNetworkProtocol) + if !ok { + return tcpip.ErrNotSupported } - s.forwarding.protocols[protocol] = enable + forwardingProtocol.SetForwarding(enable) + return nil +} - if protocol == header.IPv6ProtocolNumber { - if enable { - for _, nic := range s.nics { - nic.becomeIPv6Router() - } - } else { - for _, nic := range s.nics { - nic.becomeIPv6Host() - } - } +// Forwarding returns true if packet forwarding between NICs is enabled for the +// passed protocol. +func (s *Stack) Forwarding(protocolNum tcpip.NetworkProtocolNumber) bool { + protocol, ok := s.networkProtocols[protocolNum] + if !ok { + return false + } + + forwardingProtocol, ok := protocol.(ForwardingNetworkProtocol) + if !ok { + return false } -} -// Forwarding returns if packet forwarding between NICs is enabled. -func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { - s.forwarding.RLock() - defer s.forwarding.RUnlock() - return s.forwarding.protocols[protocol] + return forwardingProtocol.Forwarding() } // SetRouteTable assigns the route table to be used by this stack. It @@ -943,7 +837,7 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp return nil, tcpip.ErrUnknownProtocol } - return t.proto.NewEndpoint(s, network, waiterQueue) + return t.proto.NewEndpoint(network, waiterQueue) } // NewRawEndpoint creates a new raw transport layer endpoint of the given @@ -963,7 +857,7 @@ func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network return nil, tcpip.ErrUnknownProtocol } - return t.proto.NewRawEndpoint(s, network, waiterQueue) + return t.proto.NewRawEndpoint(network, waiterQueue) } // NewPacketEndpoint creates a new packet endpoint listening for the given @@ -1070,7 +964,8 @@ func (s *Stack) DisableNIC(id tcpip.NICID) *tcpip.Error { return tcpip.ErrUnknownNICID } - return nic.disable() + nic.disable() + return nil } // CheckNIC checks if a NIC is usable. @@ -1083,7 +978,7 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return false } - return nic.enabled() + return nic.Enabled() } // RemoveNIC removes NIC and all related routes from the network stack. @@ -1161,14 +1056,14 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { for id, nic := range s.nics { flags := NICStateFlags{ Up: true, // Netstack interfaces are always up. - Running: nic.enabled(), + Running: nic.Enabled(), Promiscuous: nic.isPromiscuousMode(), - Loopback: nic.isLoopback(), + Loopback: nic.IsLoopback(), } nics[id] = NICInfo{ Name: nic.name, LinkAddress: nic.linkEP.LinkAddress(), - ProtocolAddresses: nic.PrimaryAddresses(), + ProtocolAddresses: nic.primaryAddresses(), Flags: flags, MTU: nic.linkEP.MTU(), Stats: nic.stats, @@ -1232,7 +1127,7 @@ func (s *Stack) AddProtocolAddressWithOptions(id tcpip.NICID, protocolAddress tc return tcpip.ErrUnknownNICID } - return nic.AddAddress(protocolAddress, peb) + return nic.addAddress(protocolAddress, peb) } // RemoveAddress removes an existing network-layer address from the specified @@ -1242,7 +1137,7 @@ func (s *Stack) RemoveAddress(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { defer s.mu.RUnlock() if nic, ok := s.nics[id]; ok { - return nic.RemoveAddress(addr) + return nic.removeAddress(addr) } return tcpip.ErrUnknownNICID @@ -1256,7 +1151,7 @@ func (s *Stack) AllAddresses() map[tcpip.NICID][]tcpip.ProtocolAddress { nics := make(map[tcpip.NICID][]tcpip.ProtocolAddress) for id, nic := range s.nics { - nics[id] = nic.AllAddresses() + nics[id] = nic.allPermanentAddresses() } return nics } @@ -1278,7 +1173,7 @@ func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocol return nic.primaryAddress(protocol), nil } -func (s *Stack) getRefEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) { +func (s *Stack) getAddressEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) AssignableAddressEndpoint { if len(localAddr) == 0 { return nic.primaryEndpoint(netProto, remoteAddr) } @@ -1295,9 +1190,9 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n isMulticast := header.IsV4MulticastAddress(remoteAddr) || header.IsV6MulticastAddress(remoteAddr) needRoute := !(isLocalBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr)) if id != 0 && !needRoute { - if nic, ok := s.nics[id]; ok && nic.enabled() { - if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { - return makeRoute(netProto, ref.address(), remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()), nil + if nic, ok := s.nics[id]; ok && nic.Enabled() { + if addressEndpoint := s.getAddressEP(nic, localAddr, remoteAddr, netProto); addressEndpoint != nil { + return makeRoute(netProto, addressEndpoint.AddressWithPrefix().Address, remoteAddr, nic, addressEndpoint, s.handleLocal && !nic.IsLoopback(), multicastLoop && !nic.IsLoopback()), nil } } } else { @@ -1305,20 +1200,20 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !route.Destination.Contains(remoteAddr)) { continue } - if nic, ok := s.nics[route.NIC]; ok && nic.enabled() { - if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { + if nic, ok := s.nics[route.NIC]; ok && nic.Enabled() { + if addressEndpoint := s.getAddressEP(nic, localAddr, remoteAddr, netProto); addressEndpoint != nil { if len(remoteAddr) == 0 { // If no remote address was provided, then the route // provided will refer to the link local address. - remoteAddr = ref.address() + remoteAddr = addressEndpoint.AddressWithPrefix().Address } - r := makeRoute(netProto, ref.address(), remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()) + r := makeRoute(netProto, addressEndpoint.AddressWithPrefix().Address, remoteAddr, nic, addressEndpoint, s.handleLocal && !nic.IsLoopback(), multicastLoop && !nic.IsLoopback()) if len(route.Gateway) > 0 { if needRoute { r.NextHop = route.Gateway } - } else if subnet := ref.addrWithPrefix().Subnet(); subnet.IsBroadcast(remoteAddr) { + } else if subnet := addressEndpoint.AddressWithPrefix().Subnet(); subnet.IsBroadcast(remoteAddr) { r.RemoteLinkAddress = header.EthernetBroadcastAddress } @@ -1356,21 +1251,20 @@ func (s *Stack) CheckLocalAddress(nicID tcpip.NICID, protocol tcpip.NetworkProto return 0 } - ref := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint) - if ref == nil { + addressEndpoint := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint) + if addressEndpoint == nil { return 0 } - ref.decRef() + addressEndpoint.DecRef() return nic.id } // Go through all the NICs. for _, nic := range s.nics { - ref := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint) - if ref != nil { - ref.decRef() + if addressEndpoint := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint); addressEndpoint != nil { + addressEndpoint.DecRef() return nic.id } } @@ -1839,7 +1733,7 @@ func (s *Stack) LeaveGroup(protocol tcpip.NetworkProtocolNumber, nicID tcpip.NIC defer s.mu.RUnlock() if nic, ok := s.nics[nicID]; ok { - return nic.leaveGroup(multicastAddr) + return nic.leaveGroup(protocol, multicastAddr) } return tcpip.ErrUnknownNICID } @@ -1891,53 +1785,18 @@ func (s *Stack) AllowICMPMessage() bool { return s.icmpRateLimiter.Allow() } -// IsAddrTentative returns true if addr is tentative on the NIC with ID id. -// -// Note that if addr is not associated with a NIC with id ID, then this -// function will return false. It will only return true if the address is -// associated with the NIC AND it is tentative. -func (s *Stack) IsAddrTentative(id tcpip.NICID, addr tcpip.Address) (bool, *tcpip.Error) { - s.mu.RLock() - defer s.mu.RUnlock() - - nic, ok := s.nics[id] - if !ok { - return false, tcpip.ErrUnknownNICID - } - - return nic.isAddrTentative(addr), nil -} - -// DupTentativeAddrDetected attempts to inform the NIC with ID id that a -// tentative addr on it is a duplicate on a link. -func (s *Stack) DupTentativeAddrDetected(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { - s.mu.Lock() - defer s.mu.Unlock() - - nic, ok := s.nics[id] - if !ok { - return tcpip.ErrUnknownNICID - } - - return nic.dupTentativeAddrDetected(addr) -} - -// SetNDPConfigurations sets the per-interface NDP configurations on the NIC -// with ID id to c. -// -// Note, if c contains invalid NDP configuration values, it will be fixed to -// use default values for the erroneous values. -func (s *Stack) SetNDPConfigurations(id tcpip.NICID, c NDPConfigurations) *tcpip.Error { +// GetNetworkEndpoint returns the NetworkEndpoint with the specified protocol +// number installed on the specified NIC. +func (s *Stack) GetNetworkEndpoint(nicID tcpip.NICID, proto tcpip.NetworkProtocolNumber) (NetworkEndpoint, *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() - nic, ok := s.nics[id] + nic, ok := s.nics[nicID] if !ok { - return tcpip.ErrUnknownNICID + return nil, tcpip.ErrUnknownNICID } - nic.setNDPConfigs(c) - return nil + return nic.networkEndpoints[proto], nil } // NUDConfigurations gets the per-interface NUD configurations. @@ -1950,7 +1809,7 @@ func (s *Stack) NUDConfigurations(id tcpip.NICID) (NUDConfigurations, *tcpip.Err return NUDConfigurations{}, tcpip.ErrUnknownNICID } - return nic.NUDConfigs() + return nic.nudConfigs() } // SetNUDConfigurations sets the per-interface NUD configurations. @@ -1969,22 +1828,6 @@ func (s *Stack) SetNUDConfigurations(id tcpip.NICID, c NUDConfigurations) *tcpip return nic.setNUDConfigs(c) } -// HandleNDPRA provides a NIC with ID id a validated NDP Router Advertisement -// message that it needs to handle. -func (s *Stack) HandleNDPRA(id tcpip.NICID, ip tcpip.Address, ra header.NDPRouterAdvert) *tcpip.Error { - s.mu.Lock() - defer s.mu.Unlock() - - nic, ok := s.nics[id] - if !ok { - return tcpip.ErrUnknownNICID - } - - nic.handleNDPRA(ip, ra) - - return nil -} - // Seed returns a 32 bit value that can be used as a seed value for port // picking, ISN generation etc. // @@ -2026,16 +1869,14 @@ func (s *Stack) FindNetworkEndpoint(netProto tcpip.NetworkProtocolNumber, addres defer s.mu.RUnlock() for _, nic := range s.nics { - id := NetworkEndpointID{address} - - if ref, ok := nic.mu.endpoints[id]; ok { - nic.mu.RLock() - defer nic.mu.RUnlock() - - // An endpoint with this id exists, check if it can be - // used and return it. - return ref.ep, nil + addressEndpoint := nic.getAddressOrCreateTempInner(netProto, address, false /* createTemp */, NeverPrimaryEndpoint) + if addressEndpoint == nil { + continue } + + ep := addressEndpoint.NetworkEndpoint() + addressEndpoint.DecRef() + return ep, nil } return nil, tcpip.ErrBadAddress } @@ -2052,3 +1893,8 @@ func (s *Stack) FindNICNameFromID(id tcpip.NICID) string { return nic.Name() } + +// NewJob returns a new tcpip.Job using the stack's clock. +func (s *Stack) NewJob(l sync.Locker, f func()) *tcpip.Job { + return tcpip.NewJob(s.clock, l, f) +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 9ef6787c6..fda22c550 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -29,6 +29,7 @@ import ( "github.com/google/go-cmp/cmp" "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/rand" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -68,18 +69,26 @@ const ( // use the first three: destination address, source address, and transport // protocol. They're all one byte fields to simplify parsing. type fakeNetworkEndpoint struct { + stack.AddressableEndpointState + nicID tcpip.NICID proto *fakeNetworkProtocol dispatcher stack.TransportDispatcher ep stack.LinkEndpoint } -func (f *fakeNetworkEndpoint) MTU() uint32 { - return f.ep.MTU() - uint32(f.MaxHeaderLength()) +func (*fakeNetworkEndpoint) Enable() *tcpip.Error { + return nil } -func (f *fakeNetworkEndpoint) NICID() tcpip.NICID { - return f.nicID +func (*fakeNetworkEndpoint) Enabled() bool { + return true +} + +func (*fakeNetworkEndpoint) Disable() {} + +func (f *fakeNetworkEndpoint) MTU() uint32 { + return f.ep.MTU() - uint32(f.MaxHeaderLength()) } func (*fakeNetworkEndpoint) DefaultTTL() uint8 { @@ -118,10 +127,6 @@ func (f *fakeNetworkEndpoint) PseudoHeaderChecksum(protocol tcpip.TransportProto return 0 } -func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities { - return f.ep.Capabilities() -} - func (f *fakeNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return f.proto.Number() } @@ -156,7 +161,9 @@ func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack return tcpip.ErrNotSupported } -func (*fakeNetworkEndpoint) Close() {} +func (f *fakeNetworkEndpoint) Close() { + f.AddressableEndpointState.Cleanup() +} // fakeNetworkProtocol is a network-layer protocol descriptor. It aggregates the // number of packets sent and received via endpoints of this protocol. The index @@ -165,6 +172,11 @@ type fakeNetworkProtocol struct { packetCount [10]int sendPacketCount [10]int defaultTTL uint8 + + mu struct { + sync.RWMutex + forwarding bool + } } func (f *fakeNetworkProtocol) Number() tcpip.NetworkProtocolNumber { @@ -187,13 +199,15 @@ func (*fakeNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Addres return tcpip.Address(v[srcAddrOffset : srcAddrOffset+1]), tcpip.Address(v[dstAddrOffset : dstAddrOffset+1]) } -func (f *fakeNetworkProtocol) NewEndpoint(nicID tcpip.NICID, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint, _ *stack.Stack) stack.NetworkEndpoint { - return &fakeNetworkEndpoint{ - nicID: nicID, +func (f *fakeNetworkProtocol) NewEndpoint(nic stack.NetworkInterface, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &fakeNetworkEndpoint{ + nicID: nic.ID(), proto: f, dispatcher: dispatcher, - ep: ep, + ep: nic.LinkEndpoint(), } + e.AddressableEndpointState.Init(e) + return e } func (f *fakeNetworkProtocol) SetOption(option tcpip.SettableNetworkProtocolOption) *tcpip.Error { @@ -216,11 +230,6 @@ func (f *fakeNetworkProtocol) Option(option tcpip.GettableNetworkProtocolOption) } } -// ReturnError implements NetworkProtocol.ReturnError -func (*fakeNetworkProtocol) ReturnError(*stack.Route, tcpip.ICMPReason, *stack.PacketBuffer) *tcpip.Error { - return nil -} - // Close implements NetworkProtocol.Close. func (*fakeNetworkProtocol) Close() {} @@ -236,7 +245,21 @@ func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProto return tcpip.TransportProtocolNumber(hdr[protocolNumberOffset]), true, true } -func fakeNetFactory() stack.NetworkProtocol { +// Forwarding implements stack.ForwardingNetworkProtocol. +func (f *fakeNetworkProtocol) Forwarding() bool { + f.mu.RLock() + defer f.mu.RUnlock() + return f.mu.forwarding +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (f *fakeNetworkProtocol) SetForwarding(v bool) { + f.mu.Lock() + defer f.mu.Unlock() + f.mu.forwarding = v +} + +func fakeNetFactory(*stack.Stack) stack.NetworkProtocol { return &fakeNetworkProtocol{} } @@ -273,7 +296,7 @@ func TestNetworkReceive(t *testing.T) { // addresses attached to it: 1 & 2. ep := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -433,7 +456,7 @@ func TestNetworkSend(t *testing.T) { // existing nic. ep := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("NewNIC failed:", err) @@ -460,7 +483,7 @@ func TestNetworkSendMultiRoute(t *testing.T) { // addresses per nic, the first nic has odd address, the second one has // even addresses. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") @@ -560,7 +583,7 @@ func TestAttachToLinkEndpointImmediately(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) e := linkEPWithMockedAttach{ @@ -579,7 +602,7 @@ func TestAttachToLinkEndpointImmediately(t *testing.T) { func TestDisableUnknownNIC(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.DisableNIC(1); err != tcpip.ErrUnknownNICID { @@ -591,7 +614,7 @@ func TestDisabledNICsNICInfoAndCheckNIC(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) e := loopback.New() @@ -638,7 +661,7 @@ func TestDisabledNICsNICInfoAndCheckNIC(t *testing.T) { func TestRemoveUnknownNIC(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.RemoveNIC(1); err != tcpip.ErrUnknownNICID { @@ -650,7 +673,7 @@ func TestRemoveNIC(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) e := linkEPWithMockedAttach{ @@ -711,7 +734,7 @@ func TestRouteWithDownNIC(t *testing.T) { setup := func(t *testing.T) (*stack.Stack, *channel.Endpoint, *channel.Endpoint) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(1, defaultMTU, "") @@ -877,7 +900,7 @@ func TestRoutes(t *testing.T) { // addresses per nic, the first nic has odd address, the second one has // even addresses. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") @@ -957,7 +980,7 @@ func TestAddressRemoval(t *testing.T) { remoteAddr := tcpip.Address("\x02") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1004,7 +1027,7 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) { remoteAddr := tcpip.Address("\x02") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1095,7 +1118,7 @@ func TestEndpointExpiration(t *testing.T) { for _, spoofing := range []bool{true, false} { t.Run(fmt.Sprintf("promiscuous=%t spoofing=%t", promiscuous, spoofing), func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1253,7 +1276,7 @@ func TestEndpointExpiration(t *testing.T) { func TestPromiscuousMode(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1305,7 +1328,7 @@ func TestSpoofingWithAddress(t *testing.T) { dstAddr := tcpip.Address("\x03") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1371,7 +1394,7 @@ func TestSpoofingNoAddress(t *testing.T) { dstAddr := tcpip.Address("\x02") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1434,7 +1457,7 @@ func verifyRoute(gotRoute, wantRoute stack.Route) error { func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1477,7 +1500,7 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { // Create a new stack with two NICs. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1578,7 +1601,7 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { } { t.Run(tc.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1635,8 +1658,8 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { func TestNetworkOption(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{}, }) opt := tcpip.DefaultTTLOption(5) @@ -1662,7 +1685,7 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { for never := 0; never < 3; never++ { t.Run(fmt.Sprintf("never=%d", never), func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1729,7 +1752,7 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { func TestGetMainNICAddressAddRemove(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1814,7 +1837,7 @@ func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.Proto func TestAddAddress(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1841,7 +1864,7 @@ func TestAddAddress(t *testing.T) { func TestAddProtocolAddress(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1875,7 +1898,7 @@ func TestAddProtocolAddress(t *testing.T) { func TestAddAddressWithOptions(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1906,7 +1929,7 @@ func TestAddAddressWithOptions(t *testing.T) { func TestAddProtocolAddressWithOptions(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -2027,7 +2050,7 @@ func TestCreateNICWithOptions(t *testing.T) { func TestNICStats(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -2094,7 +2117,7 @@ func TestNICForwarding(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) s.SetForwarding(fakeNetNumber, true) @@ -2218,7 +2241,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName string autoGen bool linkAddr tcpip.LinkAddress - iidOpts stack.OpaqueInterfaceIdentifierOptions + iidOpts ipv6.OpaqueInterfaceIdentifierOptions shouldGen bool expectedAddr tcpip.Address }{ @@ -2234,7 +2257,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "nic1", autoGen: false, linkAddr: linkAddr1, - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:], }, @@ -2279,7 +2302,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "nic1", autoGen: true, linkAddr: linkAddr1, - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:], }, @@ -2291,7 +2314,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { { name: "OIID Empty MAC and empty nicName", autoGen: true, - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:1], }, @@ -2303,7 +2326,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "test", autoGen: true, linkAddr: "\x01\x02\x03", - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:2], }, @@ -2315,7 +2338,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "test2", autoGen: true, linkAddr: "\x01\x02\x03\x04\x05\x06", - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:3], }, @@ -2327,7 +2350,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "test3", autoGen: true, linkAddr: "\x00\x00\x00\x00\x00\x00", - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, }, shouldGen: true, @@ -2341,10 +2364,11 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1), } opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: test.autoGen, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: test.iidOpts, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: test.autoGen, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: test.iidOpts, + })}, } e := channel.New(0, 1280, test.linkAddr) @@ -2416,15 +2440,15 @@ func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) { tests := []struct { name string - opaqueIIDOpts stack.OpaqueInterfaceIdentifierOptions + opaqueIIDOpts ipv6.OpaqueInterfaceIdentifierOptions }{ { name: "IID From MAC", - opaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{}, + opaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{}, }, { name: "Opaque IID", - opaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ + opaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: func(_ tcpip.NICID, nicName string) string { return nicName }, @@ -2435,9 +2459,10 @@ func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: true, - OpaqueIIDOpts: test.opaqueIIDOpts, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: true, + OpaqueIIDOpts: test.opaqueIIDOpts, + })}, } e := loopback.New() @@ -2466,12 +2491,13 @@ func TestNICAutoGenAddrDoesDAD(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent), } - ndpConfigs := stack.DefaultNDPConfigurations() + ndpConfigs := ipv6.DefaultNDPConfigurations() opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - AutoGenIPv6LinkLocal: true, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + AutoGenIPv6LinkLocal: true, + NDPDisp: &ndpDisp, + })}, } e := channel.New(int(ndpConfigs.DupAddrDetectTransmits), 1280, linkAddr1) @@ -2527,7 +2553,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { for _, ps := range pebs { t.Run(fmt.Sprintf("%d-to-%d", pi, ps), func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -2818,14 +2844,15 @@ func TestIPv6SourceAddressSelectionScopeAndSameAddress(t *testing.T) { t.Run(test.name, func(t *testing.T) { e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDispatcher{}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDispatcher{}, + })}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -2874,7 +2901,7 @@ func TestAddRemoveIPv4BroadcastAddressOnNICEnableDisable(t *testing.T) { e := loopback.New() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) nicOpts := stack.NICOptions{Disabled: true} if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil { @@ -2926,7 +2953,7 @@ func TestLeaveIPv6SolicitedNodeAddrBeforeAddrRemoval(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { @@ -2987,7 +3014,7 @@ func TestJoinLeaveMulticastOnNICEnableDisable(t *testing.T) { t.Run(test.name, func(t *testing.T) { e := loopback.New() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, }) nicOpts := stack.NICOptions{Disabled: true} if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil { @@ -3064,12 +3091,13 @@ func TestDoDADWhenNICEnabled(t *testing.T) { dadC: make(chan ndpDADEvent), } opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: dadTransmits, - RetransmitTimer: retransmitTimer, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: dadTransmits, + RetransmitTimer: retransmitTimer, + }, + NDPDisp: &ndpDisp, + })}, } e := channel.New(dadTransmits, 1280, linkAddr1) @@ -3428,7 +3456,7 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, }) ep := channel.New(0, defaultMTU, "") if err := s.CreateNIC(nicID1, ep); err != nil { @@ -3466,7 +3494,7 @@ func TestResolveWith(t *testing.T) { ) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, arp.NewProtocol}, }) ep := channel.New(0, defaultMTU, "") ep.LinkEPCapabilities |= stack.CapabilityResolutionRequired @@ -3504,3 +3532,91 @@ func TestResolveWith(t *testing.T) { t.Fatal("got r.IsResolutionRequired() = true, want = false") } } + +// TestRouteReleaseAfterAddrRemoval tests that releasing a Route after its +// associated address is removed should not cause a panic. +func TestRouteReleaseAfterAddrRemoval(t *testing.T) { + const ( + nicID = 1 + localAddr = tcpip.Address("\x01") + remoteAddr = tcpip.Address("\x02") + ) + + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + }) + + ep := channel.New(0, defaultMTU, "") + if err := s.CreateNIC(nicID, ep); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil { + t.Fatalf("s.AddAddress(%d, %d, %s): %s", nicID, fakeNetNumber, localAddr, err) + } + { + subnet, err := tcpip.NewSubnet("\x00", "\x00") + if err != nil { + t.Fatal(err) + } + s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}}) + } + + r, err := s.FindRoute(nicID, localAddr, remoteAddr, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("s.FindRoute(%d, %s, %s, %d, false): %s", nicID, localAddr, remoteAddr, fakeNetNumber, err) + } + // Should not panic. + defer r.Release() + + // Check that removing the same address fails. + if err := s.RemoveAddress(nicID, localAddr); err != nil { + t.Fatalf("s.RemoveAddress(%d, %s): %s", nicID, localAddr, err) + } +} + +func TestGetNetworkEndpoint(t *testing.T) { + const nicID = 1 + + tests := []struct { + name string + protoFactory stack.NetworkProtocolFactory + protoNum tcpip.NetworkProtocolNumber + }{ + { + name: "IPv4", + protoFactory: ipv4.NewProtocol, + protoNum: ipv4.ProtocolNumber, + }, + { + name: "IPv6", + protoFactory: ipv6.NewProtocol, + protoNum: ipv6.ProtocolNumber, + }, + } + + factories := make([]stack.NetworkProtocolFactory, 0, len(tests)) + for _, test := range tests { + factories = append(factories, test.protoFactory) + } + + s := stack.New(stack.Options{ + NetworkProtocols: factories, + }) + + if err := s.CreateNIC(nicID, channel.New(0, defaultMTU, "")); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + ep, err := s.GetNetworkEndpoint(nicID, test.protoNum) + if err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, test.protoNum, err) + } + + if got := ep.NetworkProtocolNumber(); got != test.protoNum { + t.Fatalf("got ep.NetworkProtocolNumber() = %d, want = %d", got, test.protoNum) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 0774b5382..4ebc30ca8 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -155,7 +155,7 @@ func (epsByNIC *endpointsByNIC) transportEndpoints() []TransportEndpoint { func (epsByNIC *endpointsByNIC) handlePacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) { epsByNIC.mu.RLock() - mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()] + mpep, ok := epsByNIC.endpoints[r.nic.ID()] if !ok { if mpep, ok = epsByNIC.endpoints[0]; !ok { epsByNIC.mu.RUnlock() // Don't use defer for performance reasons. @@ -626,7 +626,7 @@ func (d *transportDemuxer) findTransportEndpoint(netProto tcpip.NetworkProtocolN epsByNIC.mu.RLock() eps.mu.RUnlock() - mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()] + mpep, ok := epsByNIC.endpoints[r.nic.ID()] if !ok { if mpep, ok = epsByNIC.endpoints[0]; !ok { epsByNIC.mu.RUnlock() // Don't use defer for performance reasons. diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 4d6d62eec..698c8609e 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -51,8 +51,8 @@ type testContext struct { // newDualTestContextMultiNIC creates the testing context and also linkEpIDs NICs. func newDualTestContextMultiNIC(t *testing.T, mtu uint32, linkEpIDs []tcpip.NICID) *testContext { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) linkEps := make(map[tcpip.NICID]*channel.Endpoint) for _, linkEpID := range linkEpIDs { @@ -182,8 +182,8 @@ func TestTransportDemuxerRegister(t *testing.T) { } { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) var wq waiter.Queue ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &wq) diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index cbb34d224..62ab6d92f 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -39,7 +39,7 @@ const ( // use it. type fakeTransportEndpoint struct { stack.TransportEndpointInfo - stack *stack.Stack + proto *fakeTransportProtocol peerAddr tcpip.Address route stack.Route @@ -59,8 +59,8 @@ func (*fakeTransportEndpoint) Stats() tcpip.EndpointStats { func (*fakeTransportEndpoint) SetOwner(owner tcpip.PacketOwner) {} -func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { - return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} +func newFakeTransportEndpoint(proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { + return &fakeTransportEndpoint{TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} } func (f *fakeTransportEndpoint) Abort() { @@ -143,7 +143,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { f.peerAddr = addr.Addr // Find the route. - r, err := f.stack.FindRoute(addr.NIC, "", addr.Addr, fakeNetNumber, false /* multicastLoop */) + r, err := f.proto.stack.FindRoute(addr.NIC, "", addr.Addr, fakeNetNumber, false /* multicastLoop */) if err != nil { return tcpip.ErrNoRoute } @@ -151,7 +151,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Try to register so that we can start receiving packets. f.ID.RemoteAddress = addr.Addr - err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, ports.Flags{}, 0 /* bindToDevice */) + err = f.proto.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, ports.Flags{}, 0 /* bindToDevice */) if err != nil { return err } @@ -190,7 +190,7 @@ func (f *fakeTransportEndpoint) Accept(*tcpip.FullAddress) (tcpip.Endpoint, *wai } func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { - if err := f.stack.RegisterTransportEndpoint( + if err := f.proto.stack.RegisterTransportEndpoint( a.NIC, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, @@ -218,7 +218,6 @@ func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportE f.proto.packetCount++ if f.acceptQueue != nil { f.acceptQueue = append(f.acceptQueue, fakeTransportEndpoint{ - stack: f.stack, TransportEndpointInfo: stack.TransportEndpointInfo{ ID: f.ID, NetProto: f.NetProto, @@ -262,6 +261,8 @@ type fakeTransportProtocolOptions struct { // fakeTransportProtocol is a transport-layer protocol descriptor. It // aggregates the number of packets received via endpoints of this protocol. type fakeTransportProtocol struct { + stack *stack.Stack + packetCount int controlCount int opts fakeTransportProtocolOptions @@ -271,11 +272,11 @@ func (*fakeTransportProtocol) Number() tcpip.TransportProtocolNumber { return fakeTransNumber } -func (f *fakeTransportProtocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newFakeTransportEndpoint(stack, f, netProto, stack.UniqueID()), nil +func (f *fakeTransportProtocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return newFakeTransportEndpoint(f, netProto, f.stack.UniqueID()), nil } -func (*fakeTransportProtocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (*fakeTransportProtocol) NewRawEndpoint(tcpip.NetworkProtocolNumber, *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return nil, tcpip.ErrUnknownProtocol } @@ -326,15 +327,15 @@ func (*fakeTransportProtocol) Parse(pkt *stack.PacketBuffer) bool { return ok } -func fakeTransFactory() stack.TransportProtocol { - return &fakeTransportProtocol{} +func fakeTransFactory(s *stack.Stack) stack.TransportProtocol { + return &fakeTransportProtocol{stack: s} } func TestTransportReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -404,8 +405,8 @@ func TestTransportReceive(t *testing.T) { func TestTransportControlReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -481,8 +482,8 @@ func TestTransportControlReceive(t *testing.T) { func TestTransportSend(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -527,8 +528,8 @@ func TestTransportSend(t *testing.T) { func TestTransportOptions(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) v := tcpip.TCPModerateReceiveBufferOption(true) @@ -546,8 +547,8 @@ func TestTransportOptions(t *testing.T) { func TestTransportForwarding(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) s.SetForwarding(fakeNetNumber, true) diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index fa73cfa47..0d4a803ff 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -237,6 +237,14 @@ type Timer interface { // network node. Or, in the case of unix endpoints, it may represent a path. type Address string +// WithPrefix returns the address with a prefix that represents a point subnet. +func (a Address) WithPrefix() AddressWithPrefix { + return AddressWithPrefix{ + Address: a, + PrefixLen: len(a) * 8, + } +} + // AddressMask is a bitmask for an address. type AddressMask string @@ -1987,14 +1995,3 @@ func DeleteDanglingEndpoint(e Endpoint) { // AsyncLoading is the global barrier for asynchronous endpoint loading // activities. var AsyncLoading sync.WaitGroup - -// ICMPReason is a marker interface for network protocol agnostic ICMP errors. -type ICMPReason interface { - isICMP() -} - -// ICMPReasonPortUnreachable is an error where the transport protocol has no -// listener and no alternative means to inform the sender. -type ICMPReasonPortUnreachable struct{} - -func (*ICMPReasonPortUnreachable) isICMP() {} diff --git a/pkg/tcpip/tests/integration/loopback_test.go b/pkg/tcpip/tests/integration/loopback_test.go index fecbe7ba7..f35dcc084 100644 --- a/pkg/tcpip/tests/integration/loopback_test.go +++ b/pkg/tcpip/tests/integration/loopback_test.go @@ -120,8 +120,8 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) if err := s.CreateNIC(nicID, loopback.New()); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) @@ -203,7 +203,7 @@ func TestLoopbackSubnetLifetimeBoundToAddr(t *testing.T) { otherAddr := tcpip.Address(addrBytes) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) if err := s.CreateNIC(nicID, loopback.New()); err != nil { t.Fatalf("s.CreateNIC(%d, _): %s", nicID, err) diff --git a/pkg/tcpip/tests/integration/multicast_broadcast_test.go b/pkg/tcpip/tests/integration/multicast_broadcast_test.go index 659acbc7a..72d86b5ab 100644 --- a/pkg/tcpip/tests/integration/multicast_broadcast_test.go +++ b/pkg/tcpip/tests/integration/multicast_broadcast_test.go @@ -140,11 +140,9 @@ func TestPingMulticastBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - ipv4Proto := ipv4.NewProtocol() - ipv6Proto := ipv6.NewProtocol() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4Proto, ipv6Proto}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4(), icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol4, icmp.NewProtocol6}, }) // We only expect a single packet in response to our ICMP Echo Request. e := channel.New(1, defaultMTU, "") @@ -176,18 +174,18 @@ func TestPingMulticastBroadcast(t *testing.T) { var rxICMP func(*channel.Endpoint, tcpip.Address) var expectedSrc tcpip.Address var expectedDst tcpip.Address - var proto stack.NetworkProtocol + var protoNum tcpip.NetworkProtocolNumber switch l := len(test.dstAddr); l { case header.IPv4AddressSize: rxICMP = rxIPv4ICMP expectedSrc = ipv4Addr.Address expectedDst = remoteIPv4Addr - proto = ipv4Proto + protoNum = header.IPv4ProtocolNumber case header.IPv6AddressSize: rxICMP = rxIPv6ICMP expectedSrc = ipv6Addr.Address expectedDst = remoteIPv6Addr - proto = ipv6Proto + protoNum = header.IPv6ProtocolNumber default: t.Fatalf("got unexpected address length = %d bytes", l) } @@ -205,7 +203,7 @@ func TestPingMulticastBroadcast(t *testing.T) { t.Errorf("got pkt.Route.RemoteAddress = %s, want = %s", pkt.Route.RemoteAddress, expectedDst) } - src, dst := proto.ParseAddresses(pkt.Pkt.NetworkHeader().View()) + src, dst := s.NetworkProtocolInstance(protoNum).ParseAddresses(pkt.Pkt.NetworkHeader().View()) if src != expectedSrc { t.Errorf("got pkt source = %s, want = %s", src, expectedSrc) } @@ -380,8 +378,8 @@ func TestIncomingMulticastAndBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, defaultMTU, "") if err := s.CreateNIC(nicID, e); err != nil { @@ -466,8 +464,8 @@ func TestReuseAddrAndBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) if err := s.CreateNIC(nicID, loopback.New()); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index 941c3c08d..87d510f96 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -13,12 +13,7 @@ // limitations under the License. // Package icmp contains the implementation of the ICMP and IPv6-ICMP transport -// protocols for use in ping. To use it in the networking stack, this package -// must be added to the project, and activated on the stack by passing -// icmp.NewProtocol4() and/or icmp.NewProtocol6() as one of the transport -// protocols when calling stack.New(). Then endpoints can be created by passing -// icmp.ProtocolNumber or icmp.ProtocolNumber6 as the transport protocol number -// when calling Stack.NewEndpoint(). +// protocols for use in ping. package icmp import ( @@ -42,6 +37,8 @@ const ( // protocol implements stack.TransportProtocol. type protocol struct { + stack *stack.Stack + number tcpip.TransportProtocolNumber } @@ -62,20 +59,20 @@ func (p *protocol) netProto() tcpip.NetworkProtocolNumber { // NewEndpoint creates a new icmp endpoint. It implements // stack.TransportProtocol.NewEndpoint. -func (p *protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { if netProto != p.netProto() { return nil, tcpip.ErrUnknownProtocol } - return newEndpoint(stack, netProto, p.number, waiterQueue) + return newEndpoint(p.stack, netProto, p.number, waiterQueue) } // NewRawEndpoint creates a new raw icmp endpoint. It implements // stack.TransportProtocol.NewRawEndpoint. -func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (p *protocol) NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { if netProto != p.netProto() { return nil, tcpip.ErrUnknownProtocol } - return raw.NewEndpoint(stack, netProto, p.number, waiterQueue) + return raw.NewEndpoint(p.stack, netProto, p.number, waiterQueue) } // MinimumPacketSize returns the minimum valid icmp packet size. @@ -135,11 +132,11 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { } // NewProtocol4 returns an ICMPv4 transport protocol. -func NewProtocol4() stack.TransportProtocol { - return &protocol{ProtocolNumber4} +func NewProtocol4(s *stack.Stack) stack.TransportProtocol { + return &protocol{stack: s, number: ProtocolNumber4} } // NewProtocol6 returns an ICMPv6 transport protocol. -func NewProtocol6() stack.TransportProtocol { - return &protocol{ProtocolNumber6} +func NewProtocol6(s *stack.Stack) stack.TransportProtocol { + return &protocol{stack: s, number: ProtocolNumber6} } diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 4778e7b1c..518449602 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -94,6 +94,7 @@ go_test( shard_count = 10, deps = [ ":tcp", + "//pkg/rand", "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 881752371..6891fd245 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -898,7 +898,7 @@ func (e *endpoint) makeOptions(sackBlocks []header.SACKBlock) []byte { // sendRaw sends a TCP segment to the endpoint's peer. func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size) *tcpip.Error { var sackBlocks []header.SACKBlock - if e.EndpointState() == StateEstablished && e.rcv.pendingBufSize > 0 && (flags&header.TCPFlagAck != 0) { + if e.EndpointState() == StateEstablished && e.rcv.pendingRcvdSegments.Len() > 0 && (flags&header.TCPFlagAck != 0) { sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) @@ -1003,9 +1003,8 @@ func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { // (indicated by a negative send window scale). e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) - rcvBufSize := seqnum.Size(e.receiveBufferSize()) e.rcvListMu.Lock() - e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) + e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale()) // Bootstrap the auto tuning algorithm. Starting at zero will // result in a really large receive window after the first auto // tuning adjustment. @@ -1136,12 +1135,11 @@ func (e *endpoint) handleSegments(fastPath bool) *tcpip.Error { } cont, err := e.handleSegment(s) + s.decRef() if err != nil { - s.decRef() return err } if !cont { - s.decRef() return nil } } @@ -1221,6 +1219,12 @@ func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { return true, nil } + // Increase counter if after processing the segment we would potentially + // advertise a zero window. + if crossed, above := e.windowCrossedACKThresholdLocked(-s.segMemSize()); crossed && !above { + e.stats.ReceiveErrors.ZeroRcvWindowState.Increment() + } + // Now check if the received segment has caused us to transition // to a CLOSED state, if yes then terminate processing and do // not invoke the sender. @@ -1233,7 +1237,6 @@ func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { // or a notification from the protocolMainLoop (caller goroutine). // This means that with this return, the segment dequeue below can // never occur on a closed endpoint. - s.decRef() return false, nil } @@ -1425,10 +1428,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ e.rcv.nonZeroWindow() } - if n¬ifyReceiveWindowChanged != 0 { - e.rcv.pendingBufSize = seqnum.Size(e.receiveBufferSize()) - } - if n¬ifyMTUChanged != 0 { e.sndBufMu.Lock() count := e.packetTooBigCount diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go index 94207c141..560b4904c 100644 --- a/pkg/tcpip/transport/tcp/dual_stack_test.go +++ b/pkg/tcpip/transport/tcp/dual_stack_test.go @@ -78,8 +78,8 @@ func testV4Connect(t *testing.T, c *context.Context, checkers ...checker.Network ackCheckers := append(checkers, checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), )) checker.IPv4(t, c.GetPacket(), ackCheckers...) @@ -185,8 +185,8 @@ func testV6Connect(t *testing.T, c *context.Context, checkers ...checker.Network ackCheckers := append(checkers, checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), )) checker.IPv6(t, c.GetV6Packet(), ackCheckers...) @@ -283,7 +283,7 @@ func TestV4RefuseOnV6Only(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) } @@ -319,7 +319,7 @@ func TestV6RefuseOnBoundToV4Mapped(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) } @@ -352,7 +352,7 @@ func testV4Accept(t *testing.T, c *context.Context) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) @@ -492,7 +492,7 @@ func TestV6AcceptOnV6(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 120483838..7ad894840 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -63,6 +63,17 @@ const ( StateClosing ) +const ( + // rcvAdvWndScale is used to split the available socket buffer into + // application buffer and the window to be advertised to the peer. This is + // currently hard coded to split the available space equally. + rcvAdvWndScale = 1 + + // SegOverheadFactor is used to multiply the value provided by the + // user on a SetSockOpt for setting the socket send/receive buffer sizes. + SegOverheadFactor = 2 +) + // connected returns true when s is one of the states representing an // endpoint connected to a peer. func (s EndpointState) connected() bool { @@ -149,7 +160,6 @@ func (s EndpointState) String() string { // Reasons for notifying the protocol goroutine. const ( notifyNonZeroReceiveWindow = 1 << iota - notifyReceiveWindowChanged notifyClose notifyMTUChanged notifyDrain @@ -384,13 +394,26 @@ type endpoint struct { // to indicate to users that no more data is coming. // // rcvListMu can be taken after the endpoint mu below. - rcvListMu sync.Mutex `state:"nosave"` - rcvList segmentList `state:"wait"` - rcvClosed bool - rcvBufSize int + rcvListMu sync.Mutex `state:"nosave"` + rcvList segmentList `state:"wait"` + rcvClosed bool + // rcvBufSize is the total size of the receive buffer. + rcvBufSize int + // rcvBufUsed is the actual number of payload bytes held in the receive buffer + // not counting any overheads of the segments itself. NOTE: This will always + // be strictly <= rcvMemUsed below. rcvBufUsed int rcvAutoParams rcvBufAutoTuneParams + // rcvMemUsed tracks the total amount of memory in use by received segments + // held in rcvList, pendingRcvdSegments and the segment queue. This is used to + // compute the window and the actual available buffer space. This is distinct + // from rcvBufUsed above which is the actual number of payload bytes held in + // the buffer not including any segment overheads. + // + // rcvMemUsed must be accessed atomically. + rcvMemUsed int32 + // mu protects all endpoint fields unless documented otherwise. mu must // be acquired before interacting with the endpoint fields. mu sync.Mutex `state:"nosave"` @@ -891,7 +914,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.probe = p } - e.segmentQueue.setLimit(MaxUnprocessedSegments) + e.segmentQueue.ep = e e.tsOffset = timeStampOffset() e.acceptCond = sync.NewCond(&e.acceptMu) @@ -904,7 +927,12 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { result := waiter.EventMask(0) switch e.EndpointState() { - case StateInitial, StateBound, StateConnecting, StateSynSent, StateSynRecv: + case StateInitial, StateBound: + // This prevents blocking of new sockets which are not + // connected when SO_LINGER is set. + result |= waiter.EventHUp + + case StateConnecting, StateSynSent, StateSynRecv: // Ready for nothing. case StateClose, StateError, StateTimeWait: @@ -1075,6 +1103,8 @@ func (e *endpoint) closeNoShutdownLocked() { e.notifyProtocolGoroutine(notifyClose) } else { e.transitionToStateCloseLocked() + // Notify that the endpoint is closed. + e.waiterQueue.Notify(waiter.EventHUp) } } @@ -1129,10 +1159,16 @@ func (e *endpoint) cleanupLocked() { tcpip.DeleteDanglingEndpoint(e) } +// wndFromSpace returns the window that we can advertise based on the available +// receive buffer space. +func wndFromSpace(space int) int { + return space / (1 << rcvAdvWndScale) +} + // initialReceiveWindow returns the initial receive window to advertise in the // SYN/SYN-ACK. func (e *endpoint) initialReceiveWindow() int { - rcvWnd := e.receiveBufferAvailable() + rcvWnd := wndFromSpace(e.receiveBufferAvailable()) if rcvWnd > math.MaxUint16 { rcvWnd = math.MaxUint16 } @@ -1209,14 +1245,12 @@ func (e *endpoint) ModerateRecvBuf(copied int) { // reject valid data that might already be in flight as the // acceptable window will shrink. if rcvWnd > e.rcvBufSize { - availBefore := e.receiveBufferAvailableLocked() + availBefore := wndFromSpace(e.receiveBufferAvailableLocked()) e.rcvBufSize = rcvWnd - availAfter := e.receiveBufferAvailableLocked() - mask := uint32(notifyReceiveWindowChanged) + availAfter := wndFromSpace(e.receiveBufferAvailableLocked()) if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above { - mask |= notifyNonZeroReceiveWindow + e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow) } - e.notifyProtocolGoroutine(mask) } // We only update prevCopied when we grow the buffer because in cases @@ -1293,18 +1327,22 @@ func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) { v := views[s.viewToDeliver] s.viewToDeliver++ + var delta int if s.viewToDeliver >= len(views) { e.rcvList.Remove(s) + // We only free up receive buffer space when the segment is released as the + // segment is still holding on to the views even though some views have been + // read out to the user. + delta = s.segMemSize() s.decRef() } e.rcvBufUsed -= len(v) - // If the window was small before this read and if the read freed up // enough buffer space, to either fit an aMSS or half a receive buffer // (whichever smaller), then notify the protocol goroutine to send a // window update. - if crossed, above := e.windowCrossedACKThresholdLocked(len(v)); crossed && above { + if crossed, above := e.windowCrossedACKThresholdLocked(delta); crossed && above { e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow) } @@ -1481,11 +1519,11 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro } // windowCrossedACKThresholdLocked checks if the receive window to be announced -// now would be under aMSS or under half receive buffer, whichever smaller. This -// is useful as a receive side silly window syndrome prevention mechanism. If -// window grows to reasonable value, we should send ACK to the sender to inform -// the rx space is now large. We also want ensure a series of small read()'s -// won't trigger a flood of spurious tiny ACK's. +// would be under aMSS or under the window derived from half receive buffer, +// whichever smaller. This is useful as a receive side silly window syndrome +// prevention mechanism. If window grows to reasonable value, we should send ACK +// to the sender to inform the rx space is now large. We also want ensure a +// series of small read()'s won't trigger a flood of spurious tiny ACK's. // // For large receive buffers, the threshold is aMSS - once reader reads more // than aMSS we'll send ACK. For tiny receive buffers, the threshold is half of @@ -1496,17 +1534,18 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro // // Precondition: e.mu and e.rcvListMu must be held. func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed bool, above bool) { - newAvail := e.receiveBufferAvailableLocked() + newAvail := wndFromSpace(e.receiveBufferAvailableLocked()) oldAvail := newAvail - deltaBefore if oldAvail < 0 { oldAvail = 0 } - threshold := int(e.amss) - if threshold > e.rcvBufSize/2 { - threshold = e.rcvBufSize / 2 + // rcvBufFraction is the inverse of the fraction of receive buffer size that + // is used to decide if the available buffer space is now above it. + const rcvBufFraction = 2 + if wndThreshold := wndFromSpace(e.rcvBufSize / rcvBufFraction); threshold > wndThreshold { + threshold = wndThreshold } - switch { case oldAvail < threshold && newAvail >= threshold: return true, true @@ -1636,17 +1675,23 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { // Make sure the receive buffer size is within the min and max // allowed. var rs tcpip.TCPReceiveBufferSizeRangeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %#v) = %s", ProtocolNumber, &rs, err)) + } + + if v > rs.Max { + v = rs.Max + } + + if v < math.MaxInt32/SegOverheadFactor { + v *= SegOverheadFactor if v < rs.Min { v = rs.Min } - if v > rs.Max { - v = rs.Max - } + } else { + v = math.MaxInt32 } - mask := uint32(notifyReceiveWindowChanged) - e.LockUser() e.rcvListMu.Lock() @@ -1660,14 +1705,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { v = 1 << scale } - // Make sure 2*size doesn't overflow. - if v > math.MaxInt32/2 { - v = math.MaxInt32 / 2 - } - - availBefore := e.receiveBufferAvailableLocked() + availBefore := wndFromSpace(e.receiveBufferAvailableLocked()) e.rcvBufSize = v - availAfter := e.receiveBufferAvailableLocked() + availAfter := wndFromSpace(e.receiveBufferAvailableLocked()) e.rcvAutoParams.disabled = true @@ -1675,24 +1715,31 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { // syndrome prevetion, when our available space grows above aMSS // or half receive buffer, whichever smaller. if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above { - mask |= notifyNonZeroReceiveWindow + e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow) } e.rcvListMu.Unlock() e.UnlockUser() - e.notifyProtocolGoroutine(mask) case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. var ss tcpip.TCPSendBufferSizeRangeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %#v) = %s", ProtocolNumber, &ss, err)) + } + + if v > ss.Max { + v = ss.Max + } + + if v < math.MaxInt32/SegOverheadFactor { + v *= SegOverheadFactor if v < ss.Min { v = ss.Min } - if v > ss.Max { - v = ss.Max - } + } else { + v = math.MaxInt32 } e.sndBufMu.Lock() @@ -2699,13 +2746,8 @@ func (e *endpoint) updateSndBufferUsage(v int) { func (e *endpoint) readyToRead(s *segment) { e.rcvListMu.Lock() if s != nil { + e.rcvBufUsed += s.payloadSize() s.incRef() - e.rcvBufUsed += s.data.Size() - // Increase counter if the receive window falls down below MSS - // or half receive buffer size, whichever smaller. - if crossed, above := e.windowCrossedACKThresholdLocked(-s.data.Size()); crossed && !above { - e.stats.ReceiveErrors.ZeroRcvWindowState.Increment() - } e.rcvList.PushBack(s) } else { e.rcvClosed = true @@ -2720,15 +2762,17 @@ func (e *endpoint) readyToRead(s *segment) { func (e *endpoint) receiveBufferAvailableLocked() int { // We may use more bytes than the buffer size when the receive buffer // shrinks. - if e.rcvBufUsed >= e.rcvBufSize { + memUsed := e.receiveMemUsed() + if memUsed >= e.rcvBufSize { return 0 } - return e.rcvBufSize - e.rcvBufUsed + return e.rcvBufSize - memUsed } // receiveBufferAvailable calculates how many bytes are still available in the -// receive buffer. +// receive buffer based on the actual memory used by all segments held in +// receive buffer/pending and segment queue. func (e *endpoint) receiveBufferAvailable() int { e.rcvListMu.Lock() available := e.receiveBufferAvailableLocked() @@ -2736,14 +2780,35 @@ func (e *endpoint) receiveBufferAvailable() int { return available } +// receiveBufferUsed returns the amount of in-use receive buffer. +func (e *endpoint) receiveBufferUsed() int { + e.rcvListMu.Lock() + used := e.rcvBufUsed + e.rcvListMu.Unlock() + return used +} + +// receiveBufferSize returns the current size of the receive buffer. func (e *endpoint) receiveBufferSize() int { e.rcvListMu.Lock() size := e.rcvBufSize e.rcvListMu.Unlock() - return size } +// receiveMemUsed returns the total memory in use by segments held by this +// endpoint. +func (e *endpoint) receiveMemUsed() int { + return int(atomic.LoadInt32(&e.rcvMemUsed)) +} + +// updateReceiveMemUsed adds the provided delta to e.rcvMemUsed. +func (e *endpoint) updateReceiveMemUsed(delta int) { + atomic.AddInt32(&e.rcvMemUsed, int32(delta)) +} + +// maxReceiveBufferSize returns the stack wide maximum receive buffer size for +// an endpoint. func (e *endpoint) maxReceiveBufferSize() int { var rs tcpip.TCPReceiveBufferSizeRangeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { @@ -2894,7 +2959,6 @@ func (e *endpoint) completeState() stack.TCPEndpointState { RcvAcc: e.rcv.rcvAcc, RcvWndScale: e.rcv.rcvWndScale, PendingBufUsed: e.rcv.pendingBufUsed, - PendingBufSize: e.rcv.pendingBufSize, } // Copy sender state. diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 41d0050f3..b25431467 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -44,7 +44,7 @@ func (e *endpoint) drainSegmentLocked() { // beforeSave is invoked by stateify. func (e *endpoint) beforeSave() { // Stop incoming packets. - e.segmentQueue.setLimit(0) + e.segmentQueue.freeze() e.mu.Lock() defer e.mu.Unlock() @@ -178,7 +178,7 @@ func (e *endpoint) afterLoad() { // Resume implements tcpip.ResumableEndpoint.Resume. func (e *endpoint) Resume(s *stack.Stack) { e.stack = s - e.segmentQueue.setLimit(MaxUnprocessedSegments) + e.segmentQueue.thaw() epState := e.origEndpointState switch epState { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 371067048..5bce73605 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -12,12 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package tcp contains the implementation of the TCP transport protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing tcp.NewProtocol() as one of the -// transport protocols when calling stack.New(). Then endpoints can be created -// by passing tcp.ProtocolNumber as the transport protocol number when calling -// Stack.NewEndpoint(). +// Package tcp contains the implementation of the TCP transport protocol. package tcp import ( @@ -138,6 +133,8 @@ func (s *synRcvdCounter) Threshold() uint64 { } type protocol struct { + stack *stack.Stack + mu sync.RWMutex sackEnabled bool recovery tcpip.TCPRecovery @@ -164,14 +161,14 @@ func (*protocol) Number() tcpip.TransportProtocolNumber { } // NewEndpoint creates a new tcp endpoint. -func (p *protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newEndpoint(stack, netProto, waiterQueue), nil +func (p *protocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return newEndpoint(p.stack, netProto, waiterQueue), nil } // NewRawEndpoint creates a new raw TCP endpoint. Raw TCP sockets are currently // unsupported. It implements stack.TransportProtocol.NewRawEndpoint. -func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return raw.NewEndpoint(stack, netProto, header.TCPProtocolNumber, waiterQueue) +func (p *protocol) NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return raw.NewEndpoint(p.stack, netProto, header.TCPProtocolNumber, waiterQueue) } // MinimumPacketSize returns the minimum valid tcp packet size. @@ -510,8 +507,9 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { } // NewProtocol returns a TCP transport protocol. -func NewProtocol() stack.TransportProtocol { +func NewProtocol(s *stack.Stack) stack.TransportProtocol { p := protocol{ + stack: s, sendBufferSize: tcpip.TCPSendBufferSizeRangeOption{ Min: MinBufferSize, Default: DefaultSendBufferSize, diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index cfd43b5e3..48bf196d8 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -47,22 +47,24 @@ type receiver struct { closed bool + // pendingRcvdSegments is bounded by the receive buffer size of the + // endpoint. pendingRcvdSegments segmentHeap - pendingBufUsed seqnum.Size - pendingBufSize seqnum.Size + // pendingBufUsed tracks the total number of bytes (including segment + // overhead) currently queued in pendingRcvdSegments. + pendingBufUsed int // Time when the last ack was received. lastRcvdAckTime time.Time `state:".(unixTime)"` } -func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8, pendingBufSize seqnum.Size) *receiver { +func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8) *receiver { return &receiver{ ep: ep, rcvNxt: irs + 1, rcvAcc: irs.Add(rcvWnd + 1), rcvWnd: rcvWnd, rcvWndScale: rcvWndScale, - pendingBufSize: pendingBufSize, lastRcvdAckTime: time.Now(), } } @@ -85,15 +87,30 @@ func (r *receiver) acceptable(segSeq seqnum.Value, segLen seqnum.Size) bool { // getSendParams returns the parameters needed by the sender when building // segments to send. func (r *receiver) getSendParams() (rcvNxt seqnum.Value, rcvWnd seqnum.Size) { - // Calculate the window size based on the available buffer space. - receiveBufferAvailable := r.ep.receiveBufferAvailable() - acc := r.rcvNxt.Add(seqnum.Size(receiveBufferAvailable)) - if r.rcvAcc.LessThan(acc) { - r.rcvAcc = acc + avail := wndFromSpace(r.ep.receiveBufferAvailable()) + if avail == 0 { + // We have no space available to accept any data, move to zero window + // state. + r.rcvWnd = 0 + return r.rcvNxt, 0 + } + + acc := r.rcvNxt.Add(seqnum.Size(avail)) + newWnd := r.rcvNxt.Size(acc) + curWnd := r.rcvNxt.Size(r.rcvAcc) + + // Update rcvAcc only if new window is > previously advertised window. We + // should never shrink the acceptable sequence space once it has been + // advertised the peer. If we shrink the acceptable sequence space then we + // would end up dropping bytes that might already be in flight. + if newWnd > curWnd { + r.rcvAcc = r.rcvNxt.Add(newWnd) + } else { + newWnd = curWnd } // Stash away the non-scaled receive window as we use it for measuring // receiver's estimated RTT. - r.rcvWnd = r.rcvNxt.Size(r.rcvAcc) + r.rcvWnd = newWnd return r.rcvNxt, r.rcvWnd >> r.rcvWndScale } @@ -195,7 +212,9 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum } for i := first; i < len(r.pendingRcvdSegments); i++ { + r.pendingBufUsed -= r.pendingRcvdSegments[i].segMemSize() r.pendingRcvdSegments[i].decRef() + // Note that slice truncation does not allow garbage collection of // truncated items, thus truncated items must be set to nil to avoid // memory leaks. @@ -384,10 +403,16 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { // Defer segment processing if it can't be consumed now. if !r.consumeSegment(s, segSeq, segLen) { if segLen > 0 || s.flagIsSet(header.TCPFlagFin) { - // We only store the segment if it's within our buffer - // size limit. - if r.pendingBufUsed < r.pendingBufSize { - r.pendingBufUsed += seqnum.Size(s.segMemSize()) + // We only store the segment if it's within our buffer size limit. + // + // Only use 75% of the receive buffer queue for out-of-order + // segments. This ensures that we always leave some space for the inorder + // segments to arrive allowing pending segments to be processed and + // delivered to the user. + if r.ep.receiveBufferAvailable() > 0 && r.pendingBufUsed < r.ep.receiveBufferSize()>>2 { + r.ep.rcvListMu.Lock() + r.pendingBufUsed += s.segMemSize() + r.ep.rcvListMu.Unlock() s.incRef() heap.Push(&r.pendingRcvdSegments, s) UpdateSACKBlocks(&r.ep.sack, segSeq, segSeq.Add(segLen), r.rcvNxt) @@ -421,7 +446,9 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { } heap.Pop(&r.pendingRcvdSegments) - r.pendingBufUsed -= seqnum.Size(s.segMemSize()) + r.ep.rcvListMu.Lock() + r.pendingBufUsed -= s.segMemSize() + r.ep.rcvListMu.Unlock() s.decRef() } return false, nil diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 94307d31a..13acaf753 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -15,6 +15,7 @@ package tcp import ( + "fmt" "sync/atomic" "time" @@ -24,6 +25,15 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/stack" ) +// queueFlags are used to indicate which queue of an endpoint a particular segment +// belongs to. This is used to track memory accounting correctly. +type queueFlags uint8 + +const ( + recvQ queueFlags = 1 << iota + sendQ +) + // segment represents a TCP segment. It holds the payload and parsed TCP segment // information, and can be added to intrusive lists. // segment is mostly immutable, the only field allowed to change is viewToDeliver. @@ -32,6 +42,8 @@ import ( type segment struct { segmentEntry refCnt int32 + ep *endpoint + qFlags queueFlags id stack.TransportEndpointID `state:"manual"` route stack.Route `state:"manual"` data buffer.VectorisedView `state:".(buffer.VectorisedView)"` @@ -100,6 +112,8 @@ func (s *segment) clone() *segment { rcvdTime: s.rcvdTime, xmitTime: s.xmitTime, xmitCount: s.xmitCount, + ep: s.ep, + qFlags: s.qFlags, } t.data = s.data.Clone(t.views[:]) return t @@ -115,8 +129,34 @@ func (s *segment) flagsAreSet(flags uint8) bool { return s.flags&flags == flags } +// setOwner sets the owning endpoint for this segment. Its required +// to be called to ensure memory accounting for receive/send buffer +// queues is done properly. +func (s *segment) setOwner(ep *endpoint, qFlags queueFlags) { + switch qFlags { + case recvQ: + ep.updateReceiveMemUsed(s.segMemSize()) + case sendQ: + // no memory account for sendQ yet. + default: + panic(fmt.Sprintf("unexpected queue flag %b", qFlags)) + } + s.ep = ep + s.qFlags = qFlags +} + func (s *segment) decRef() { if atomic.AddInt32(&s.refCnt, -1) == 0 { + if s.ep != nil { + switch s.qFlags { + case recvQ: + s.ep.updateReceiveMemUsed(-s.segMemSize()) + case sendQ: + // no memory accounting for sendQ yet. + default: + panic(fmt.Sprintf("unexpected queue flag %b set for segment", s.qFlags)) + } + } s.route.Release() } } @@ -138,6 +178,11 @@ func (s *segment) logicalLen() seqnum.Size { return l } +// payloadSize is the size of s.data. +func (s *segment) payloadSize() int { + return s.data.Size() +} + // segMemSize is the amount of memory used to hold the segment data and // the associated metadata. func (s *segment) segMemSize() int { diff --git a/pkg/tcpip/transport/tcp/segment_queue.go b/pkg/tcpip/transport/tcp/segment_queue.go index 48a257137..54545a1b1 100644 --- a/pkg/tcpip/transport/tcp/segment_queue.go +++ b/pkg/tcpip/transport/tcp/segment_queue.go @@ -22,16 +22,16 @@ import ( // // +stateify savable type segmentQueue struct { - mu sync.Mutex `state:"nosave"` - list segmentList `state:"wait"` - limit int - used int + mu sync.Mutex `state:"nosave"` + list segmentList `state:"wait"` + ep *endpoint + frozen bool } // emptyLocked determines if the queue is empty. // Preconditions: q.mu must be held. func (q *segmentQueue) emptyLocked() bool { - return q.used == 0 + return q.list.Empty() } // empty determines if the queue is empty. @@ -43,14 +43,6 @@ func (q *segmentQueue) empty() bool { return r } -// setLimit updates the limit. No segments are immediately dropped in case the -// queue becomes full due to the new limit. -func (q *segmentQueue) setLimit(limit int) { - q.mu.Lock() - q.limit = limit - q.mu.Unlock() -} - // enqueue adds the given segment to the queue. // // Returns true when the segment is successfully added to the queue, in which @@ -58,15 +50,23 @@ func (q *segmentQueue) setLimit(limit int) { // false if the queue is full, in which case ownership is retained by the // caller. func (q *segmentQueue) enqueue(s *segment) bool { + // q.ep.receiveBufferParams() must be called without holding q.mu to + // avoid lock order inversion. + bufSz := q.ep.receiveBufferSize() + used := q.ep.receiveMemUsed() q.mu.Lock() - r := q.used < q.limit - if r { + // Allow zero sized segments (ACK/FIN/RSTs etc even if the segment queue + // is currently full). + allow := (used <= bufSz || s.payloadSize() == 0) && !q.frozen + + if allow { q.list.PushBack(s) - q.used++ + // Set the owner now that the endpoint owns the segment. + s.setOwner(q.ep, recvQ) } q.mu.Unlock() - return r + return allow } // dequeue removes and returns the next segment from queue, if one exists. @@ -77,9 +77,25 @@ func (q *segmentQueue) dequeue() *segment { s := q.list.Front() if s != nil { q.list.Remove(s) - q.used-- } q.mu.Unlock() return s } + +// freeze prevents any more segments from being added to the queue. i.e all +// future segmentQueue.enqueue will return false and not add the segment to the +// queue till the queue is unfroze with a corresponding segmentQueue.thaw call. +func (q *segmentQueue) freeze() { + q.mu.Lock() + q.frozen = true + q.mu.Unlock() +} + +// thaw unfreezes a previously frozen queue using segmentQueue.freeze() and +// allows new segments to be queued again. +func (q *segmentQueue) thaw() { + q.mu.Lock() + q.frozen = false + q.mu.Unlock() +} diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 8b2217a98..5b504d0d1 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -21,6 +21,7 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -349,8 +350,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ SrcPort: context.TestPort, @@ -380,8 +381,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(0), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst))) } @@ -479,8 +480,8 @@ func TestConnectResetAfterClose(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -521,8 +522,8 @@ func TestConnectResetAfterClose(t *testing.T) { // RST is always generated with sndNxt which if the FIN // has been sent will be 1 higher than the sequence number // of the FIN itself. - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(0), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -561,8 +562,8 @@ func TestCurrentConnectedIncrement(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -597,8 +598,8 @@ func TestCurrentConnectedIncrement(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -644,8 +645,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -665,8 +666,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -725,8 +726,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(0), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -777,8 +778,8 @@ func TestSimpleReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1030,7 +1031,7 @@ func TestSendRstOnListenerRxSynAckV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } func TestSendRstOnListenerRxSynAckV6(t *testing.T) { @@ -1058,7 +1059,7 @@ func TestSendRstOnListenerRxSynAckV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } // TestTCPAckBeforeAcceptV4 tests that once the 3-way handshake is complete, @@ -1095,8 +1096,8 @@ func TestTCPAckBeforeAcceptV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } // TestTCPAckBeforeAcceptV6 tests that once the 3-way handshake is complete, @@ -1133,8 +1134,8 @@ func TestTCPAckBeforeAcceptV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } func TestSendRstOnListenerRxAckV4(t *testing.T) { @@ -1162,7 +1163,7 @@ func TestSendRstOnListenerRxAckV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } func TestSendRstOnListenerRxAckV6(t *testing.T) { @@ -1190,7 +1191,7 @@ func TestSendRstOnListenerRxAckV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } // TestListenShutdown tests for the listening endpoint replying with RST @@ -1306,8 +1307,8 @@ func TestTOSV4(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), // Acknum is initial sequence number + 1 + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), // Acknum is initial sequence number + 1 checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), checker.TOS(tos, 0), @@ -1355,8 +1356,8 @@ func TestTrafficClassV6(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), checker.TOS(tos, 0), @@ -1546,8 +1547,8 @@ func TestOutOfOrderReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1597,8 +1598,8 @@ func TestOutOfOrderReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1608,8 +1609,8 @@ func TestOutOfOrderFlood(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - // Create a new connection with initial window size of 10. - c.CreateConnected(789, 30000, 10) + rcvBufSz := math.MaxUint16 + c.CreateConnected(789, 30000, rcvBufSz) if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrWouldBlock) @@ -1630,8 +1631,8 @@ func TestOutOfOrderFlood(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1651,8 +1652,8 @@ func TestOutOfOrderFlood(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1671,8 +1672,8 @@ func TestOutOfOrderFlood(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(793), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(793), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1713,8 +1714,8 @@ func TestRstOnCloseWithUnreadData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1728,7 +1729,7 @@ func TestRstOnCloseWithUnreadData(t *testing.T) { checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), // We shouldn't consume a sequence number on RST. - checker.SeqNum(uint32(c.IRS)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), )) // The RST puts the endpoint into an error state. if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { @@ -1782,8 +1783,8 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1796,7 +1797,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - checker.SeqNum(uint32(c.IRS)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), )) if got, want := tcp.EndpointState(c.EP.State()), tcp.StateFinWait1; got != want { @@ -1815,7 +1816,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { // RST is always generated with sndNxt which if the FIN // has been sent will be 1 higher than the sequence // number of the FIN itself. - checker.SeqNum(uint32(c.IRS)+2), + checker.TCPSeqNum(uint32(c.IRS)+2), )) // The RST puts the endpoint into an error state. if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { @@ -1861,7 +1862,8 @@ func TestFullWindowReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, 10) + const rcvBufSz = 10 + c.CreateConnected(789, 30000, rcvBufSz) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -1872,8 +1874,13 @@ func TestFullWindowReceive(t *testing.T) { t.Fatalf("Read failed: %s", err) } - // Fill up the window. - data := []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10} + // Fill up the window w/ tcp.SegOverheadFactor*rcvBufSz as netstack multiplies + // the provided buffer value by tcp.SegOverheadFactor to calculate the actual + // receive buffer size. + data := make([]byte, tcp.SegOverheadFactor*rcvBufSz) + for i := range data { + data[i] = byte(i % 255) + } c.SendPacket(data, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, @@ -1894,10 +1901,10 @@ func TestFullWindowReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), - checker.Window(0), + checker.TCPWindow(0), ), ) @@ -1920,10 +1927,10 @@ func TestFullWindowReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), - checker.Window(10), + checker.TCPWindow(10), ), ) } @@ -1932,12 +1939,15 @@ func TestNoWindowShrinking(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - // Start off with a window size of 10, then shrink it to 5. - c.CreateConnected(789, 30000, 10) - - if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 5); err != nil { - t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 5) failed: %s", err) - } + // Start off with a certain receive buffer then cut it in half and verify that + // the right edge of the window does not shrink. + // NOTE: Netstack doubles the value specified here. + rcvBufSize := 65536 + iss := seqnum.Value(789) + // Enable window scaling with a scale of zero from our end. + c.CreateConnectedWithRawOptions(iss, 30000, rcvBufSize, []byte{ + header.TCPOptionWS, 3, 0, header.TCPOptionNOP, + }) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -1946,14 +1956,15 @@ func TestNoWindowShrinking(t *testing.T) { if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrWouldBlock) } - - // Send 3 bytes, check that the peer acknowledges them. - data := []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10} - c.SendPacket(data[:3], &context.Headers{ + // Send a 1 byte payload so that we can record the current receive window. + // Send a payload of half the size of rcvBufSize. + seqNum := iss.Add(1) + payload := []byte{1} + c.SendPacket(payload, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, Flags: header.TCPFlagAck, - SeqNum: 790, + SeqNum: seqNum, AckNum: c.IRS.Add(1), RcvWnd: 30000, }) @@ -1965,46 +1976,93 @@ func TestNoWindowShrinking(t *testing.T) { t.Fatalf("Timed out waiting for data to arrive") } - // Check that data is acknowledged, and that window doesn't go to zero - // just yet because it was previously set to 10. It must go to 7 now. - checker.IPv4(t, c.GetPacket(), + // Read the 1 byte payload we just sent. + v, _, err := c.EP.Read(nil) + if err != nil { + t.Fatalf("Read failed: %s", err) + } + if got, want := payload, v; !bytes.Equal(got, want) { + t.Fatalf("got data: %v, want: %v", got, want) + } + + seqNum = seqNum.Add(1) + // Verify that the ACK does not shrink the window. + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(793), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), checker.TCPFlags(header.TCPFlagAck), - checker.Window(7), ), ) + // Stash the initial window. + initialWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() << c.RcvdWindowScale + initialLastAcceptableSeq := seqNum.Add(seqnum.Size(initialWnd)) + // Now shrink the receive buffer to half its original size. + if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufSize/2); err != nil { + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 5) failed: %s", err) + } - // Send 7 more bytes, check that the window fills up. - c.SendPacket(data[3:], &context.Headers{ + data := generateRandomPayload(t, rcvBufSize) + // Send a payload of half the size of rcvBufSize. + c.SendPacket(data[:rcvBufSize/2], &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, Flags: header.TCPFlagAck, - SeqNum: 793, + SeqNum: seqNum, AckNum: c.IRS.Add(1), RcvWnd: 30000, }) + seqNum = seqNum.Add(seqnum.Size(rcvBufSize / 2)) - select { - case <-ch: - case <-time.After(5 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") + // Verify that the ACK does not shrink the window. + pkt = c.GetPacket() + checker.IPv4(t, pkt, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + newWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() << c.RcvdWindowScale + newLastAcceptableSeq := seqNum.Add(seqnum.Size(newWnd)) + if newLastAcceptableSeq.LessThan(initialLastAcceptableSeq) { + t.Fatalf("receive window shrunk unexpectedly got: %d, want >= %d", newLastAcceptableSeq, initialLastAcceptableSeq) } + // Send another payload of half the size of rcvBufSize. This should fill up the + // socket receive buffer and we should see a zero window. + c.SendPacket(data[rcvBufSize/2:], &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: seqNum, + AckNum: c.IRS.Add(1), + RcvWnd: 30000, + }) + seqNum = seqNum.Add(seqnum.Size(rcvBufSize / 2)) + checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), checker.TCPFlags(header.TCPFlagAck), - checker.Window(0), + checker.TCPWindow(0), ), ) + // Wait for receive to be notified. + select { + case <-ch: + case <-time.After(5 * time.Second): + t.Fatalf("Timed out waiting for data to arrive") + } + // Receive data and check it. - read := make([]byte, 0, 10) + read := make([]byte, 0, rcvBufSize) for len(read) < len(data) { v, _, err := c.EP.Read(nil) if err != nil { @@ -2018,15 +2076,15 @@ func TestNoWindowShrinking(t *testing.T) { t.Fatalf("got data = %v, want = %v", read, data) } - // Check that we get an ACK for the newly non-zero window, which is the - // new size. + // Check that we get an ACK for the newly non-zero window, which is the new + // receive buffer size we set after the connection was established. checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), checker.TCPFlags(header.TCPFlagAck), - checker.Window(5), + checker.TCPWindow(uint16(rcvBufSize/2)>>c.RcvdWindowScale), ), ) } @@ -2051,8 +2109,8 @@ func TestSimpleSend(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2093,8 +2151,8 @@ func TestZeroWindowSend(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2115,8 +2173,8 @@ func TestZeroWindowSend(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2155,16 +2213,16 @@ func TestScaledWindowConnect(t *testing.T) { t.Fatalf("Write failed: %s", err) } - // Check that data is received, and that advertised window is 0xbfff, + // Check that data is received, and that advertised window is 0x5fff, // that is, that it is scaled. b := c.GetPacket() checker.IPv4(t, b, checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xbfff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0x5fff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2194,9 +2252,9 @@ func TestNonScaledWindowConnect(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xffff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0xffff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2230,7 +2288,8 @@ func TestScaledWindowAccept(t *testing.T) { } // Do 3-way handshake. - c.PassiveConnectWithOptions(100, 2, header.TCPSynOptions{MSS: defaultIPv4MSS}) + // wndScale expected is 3 as 65535 * 3 * 2 < 65535 * 2^3 but > 65535 *2 *2 + c.PassiveConnectWithOptions(100, 3 /* wndScale */, header.TCPSynOptions{MSS: defaultIPv4MSS}) // Try to accept the connection. we, ch := waiter.NewChannelEntry(nil) @@ -2260,16 +2319,16 @@ func TestScaledWindowAccept(t *testing.T) { t.Fatalf("Write failed: %s", err) } - // Check that data is received, and that advertised window is 0xbfff, + // Check that data is received, and that advertised window is 0x5fff, // that is, that it is scaled. b := c.GetPacket() checker.IPv4(t, b, checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xbfff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0x5fff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2341,9 +2400,9 @@ func TestNonScaledWindowAccept(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xffff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0xffff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2356,18 +2415,19 @@ func TestZeroScaledWindowReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - // Set the window size such that a window scale of 4 will be used. - const wnd = 65535 * 10 - const ws = uint32(4) - c.CreateConnectedWithRawOptions(789, 30000, wnd, []byte{ + // Set the buffer size such that a window scale of 5 will be used. + const bufSz = 65535 * 10 + const ws = uint32(5) + c.CreateConnectedWithRawOptions(789, 30000, bufSz, []byte{ header.TCPOptionWS, 3, 0, header.TCPOptionNOP, }) // Write chunks of 50000 bytes. - remain := wnd + remain := 0 sent := 0 data := make([]byte, 50000) - for remain > len(data) { + // Keep writing till the window drops below len(data). + for { c.SendPacket(data, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, @@ -2377,21 +2437,25 @@ func TestZeroScaledWindowReceive(t *testing.T) { RcvWnd: 30000, }) sent += len(data) - remain -= len(data) - checker.IPv4(t, c.GetPacket(), + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(remain>>ws)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), checker.TCPFlags(header.TCPFlagAck), ), ) + // Don't reduce window to zero here. + if wnd := int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()); wnd<<ws < len(data) { + remain = wnd << ws + break + } } // Make the window non-zero, but the scaled window zero. - if remain >= 16 { + for remain >= 16 { data = data[:remain-15] c.SendPacket(data, &context.Headers{ SrcPort: context.TestPort, @@ -2402,22 +2466,35 @@ func TestZeroScaledWindowReceive(t *testing.T) { RcvWnd: 30000, }) sent += len(data) - remain -= len(data) - checker.IPv4(t, c.GetPacket(), + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(0), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), checker.TCPFlags(header.TCPFlagAck), ), ) + // Since the receive buffer is split between window advertisement and + // application data buffer the window does not always reflect the space + // available and actual space available can be a bit more than what is + // advertised in the window. + wnd := int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()) + if wnd == 0 { + break + } + remain = wnd << ws } - // Read at least 1MSS of data. An ack should be sent in response to that. + // Read at least 2MSS of data. An ack should be sent in response to that. + // Since buffer space is now split in half between window and application + // data we need to read more than 1 MSS(65536) of data for a non-zero window + // update to be sent. For 1MSS worth of window to be available we need to + // read at least 128KB. Since our segments above were 50KB each it means + // we need to read at 3 packets. sz := 0 - for sz < defaultMTU { + for sz < defaultMTU*2 { v, _, err := c.EP.Read(nil) if err != nil { t.Fatalf("Read failed: %s", err) @@ -2429,9 +2506,9 @@ func TestZeroScaledWindowReceive(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(sz>>ws)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindowGreaterThanEq(uint16(defaultMTU>>ws)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -2498,8 +2575,8 @@ func TestSegmentMerging(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize+1), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+uint32(i)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+uint32(i)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2521,8 +2598,8 @@ func TestSegmentMerging(t *testing.T) { checker.PayloadLen(len(allData)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+11), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+11), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2569,8 +2646,8 @@ func TestDelay(t *testing.T) { checker.PayloadLen(len(want)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2616,8 +2693,8 @@ func TestUndelay(t *testing.T) { checker.PayloadLen(len(allData[0])+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2639,8 +2716,8 @@ func TestUndelay(t *testing.T) { checker.PayloadLen(len(allData[1])+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2701,8 +2778,8 @@ func TestMSSNotDelayed(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2753,8 +2830,8 @@ func testBrokenUpWrite(t *testing.T, c *context.Context, maxPayload int) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1+uint32(bytesReceived)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1+uint32(bytesReceived)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2996,7 +3073,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { // Set the buffer size to a deterministic size so that we can check the // window scaling option. const rcvBufferSize = 0x20000 - const wndScale = 2 + const wndScale = 3 if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, %d) failed failed: %s", rcvBufferSize, err) } @@ -3031,7 +3108,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagSyn), checker.SrcPort(tcpHdr.SourcePort()), - checker.SeqNum(tcpHdr.SequenceNumber()), + checker.TCPSeqNum(tcpHdr.SequenceNumber()), checker.TCPSynOptions(header.TCPSynOptions{MSS: mss, WS: wndScale}), ), ) @@ -3052,8 +3129,8 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), ), ) @@ -3346,8 +3423,8 @@ func TestFinImmediately(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3367,8 +3444,8 @@ func TestFinImmediately(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3389,8 +3466,8 @@ func TestFinRetransmit(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3400,8 +3477,8 @@ func TestFinRetransmit(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3421,8 +3498,8 @@ func TestFinRetransmit(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3445,8 +3522,8 @@ func TestFinWithNoPendingData(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3470,8 +3547,8 @@ func TestFinWithNoPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3492,8 +3569,8 @@ func TestFinWithNoPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3520,8 +3597,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3539,8 +3616,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3559,8 +3636,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3580,8 +3657,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3604,8 +3681,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3629,8 +3706,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3645,8 +3722,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3666,8 +3743,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3691,8 +3768,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3712,8 +3789,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3727,8 +3804,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3743,8 +3820,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3835,8 +3912,8 @@ func scaledSendWindow(t *testing.T, scale uint8) { checker.PayloadLen((1<<scale)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3974,7 +4051,7 @@ func TestReceivedSegmentQueuing(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -4025,8 +4102,8 @@ func TestReadAfterClosedState(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -4050,8 +4127,8 @@ func TestReadAfterClosedState(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(uint32(791+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(uint32(791+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -4223,8 +4300,8 @@ func checkSendBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { func TestDefaultBufferSizes(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) // Check the default values. @@ -4286,8 +4363,8 @@ func TestDefaultBufferSizes(t *testing.T) { func TestMinMaxBufferSizes(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) // Check the default values. @@ -4312,14 +4389,14 @@ func TestMinMaxBufferSizes(t *testing.T) { } } - // Set values below the min. - if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 199); err != nil { + // Set values below the min/2. + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 99); err != nil { t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 199) failed: %s", err) } checkRecvBufferSize(t, ep, 200) - if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 299); err != nil { + if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 149); err != nil { t.Fatalf("SetSockOptInt(SendBufferSizeOption, 299) failed: %s", err) } @@ -4330,19 +4407,21 @@ func TestMinMaxBufferSizes(t *testing.T) { t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } - checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20) + // Values above max are capped at max and then doubled. + checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20*2) if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 1+tcp.DefaultSendBufferSize*30); err != nil { t.Fatalf("SetSockOptInt(SendBufferSizeOption) failed: %s", err) } - checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) + // Values above max are capped at max and then doubled. + checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30*2) } func TestBindToDeviceOption(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}}) + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}}) ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) if err != nil { @@ -4391,11 +4470,11 @@ func TestBindToDeviceOption(t *testing.T) { func makeStack() (*stack.Stack, *tcpip.Error) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ - ipv4.NewProtocol(), - ipv6.NewProtocol(), + NetworkProtocols: []stack.NetworkProtocolFactory{ + ipv4.NewProtocol, + ipv6.NewProtocol, }, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) id := loopback.New() @@ -4678,8 +4757,8 @@ func TestPathMTUDiscovery(t *testing.T) { checker.PayloadLen(size+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(seqNum), - checker.AckNum(790), + checker.TCPSeqNum(seqNum), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -4930,8 +5009,8 @@ func TestKeepalive(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(uint32(790)), + checker.TCPSeqNum(uint32(c.IRS)), + checker.TCPAckNum(uint32(790)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -4964,8 +5043,8 @@ func TestKeepalive(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -4976,8 +5055,8 @@ func TestKeepalive(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagPsh), ), ) @@ -5002,8 +5081,8 @@ func TestKeepalive(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next-1)), - checker.AckNum(uint32(790)), + checker.TCPSeqNum(uint32(next-1)), + checker.TCPAckNum(uint32(790)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -5029,8 +5108,8 @@ func TestKeepalive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next)), - checker.AckNum(uint32(0)), + checker.TCPSeqNum(uint32(next)), + checker.TCPAckNum(uint32(0)), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -5070,7 +5149,7 @@ func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCooki checker.SrcPort(context.StackPort), checker.DstPort(srcPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } if synCookieInUse { @@ -5114,7 +5193,7 @@ func executeV6Handshake(t *testing.T, c *context.Context, srcPort uint16, synCoo checker.SrcPort(context.StackPort), checker.DstPort(srcPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } if synCookieInUse { @@ -5348,7 +5427,7 @@ func TestListenNoAcceptNonUnicastV4(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1))) + checker.TCPAckNum(uint32(irs)+1))) }) } } @@ -5356,8 +5435,8 @@ func TestListenNoAcceptNonUnicastV4(t *testing.T) { // TestListenNoAcceptMulticastBroadcastV6 makes sure that TCP segments with a // non unicast IPv6 address are not accepted. func TestListenNoAcceptNonUnicastV6(t *testing.T) { - multicastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x01") - otherMulticastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x02") + multicastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x01") + otherMulticastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x02") tests := []struct { name string @@ -5448,7 +5527,7 @@ func TestListenNoAcceptNonUnicastV6(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1))) + checker.TCPAckNum(uint32(irs)+1))) }) } } @@ -5496,7 +5575,7 @@ func TestListenSynRcvdQueueFull(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } checker.IPv4(t, b, checker.TCP(tcpCheckers...)) @@ -5674,7 +5753,7 @@ func TestSynRcvdBadSeqNumber(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } checker.IPv4(t, b, checker.TCP(tcpCheckers...)) @@ -5695,8 +5774,8 @@ func TestSynRcvdBadSeqNumber(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.AckNum(uint32(irs) + 1), - checker.SeqNum(uint32(iss + 1)), + checker.TCPAckNum(uint32(irs) + 1), + checker.TCPSeqNum(uint32(iss + 1)), } checker.IPv4(t, b, checker.TCP(tcpCheckers...)) @@ -5994,16 +6073,14 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { time.Sleep(latency) rawEP.SendPacketWithTS([]byte{1}, tsVal) - // Verify that the ACK has the expected window. - wantRcvWnd := receiveBufferSize - wantRcvWnd = (wantRcvWnd >> uint32(c.WindowScale)) - rawEP.VerifyACKRcvWnd(uint16(wantRcvWnd - 1)) + pkt := rawEP.VerifyAndReturnACKWithTS(tsVal) + rcvWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() time.Sleep(25 * time.Millisecond) // Allocate a large enough payload for the test. - b := make([]byte, int(receiveBufferSize)*2) - offset := 0 - payloadSize := receiveBufferSize - 1 + payloadSize := receiveBufferSize * 2 + b := make([]byte, int(payloadSize)) + worker := (c.EP).(interface { StopWork() ResumeWork() @@ -6012,11 +6089,15 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // Stop the worker goroutine. worker.StopWork() - start := offset - end := offset + payloadSize + start := 0 + end := payloadSize / 2 packetsSent := 0 for ; start < end; start += mss { - rawEP.SendPacketWithTS(b[start:start+mss], tsVal) + packetEnd := start + mss + if start+mss > end { + packetEnd = end + } + rawEP.SendPacketWithTS(b[start:packetEnd], tsVal) packetsSent++ } @@ -6024,29 +6105,20 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // are waiting to be read. worker.ResumeWork() - // Since we read no bytes the window should goto zero till the - // application reads some of the data. - // Discard all intermediate acks except the last one. - if packetsSent > 100 { - for i := 0; i < (packetsSent / 100); i++ { - _ = c.GetPacket() - } + // Since we sent almost the full receive buffer worth of data (some may have + // been dropped due to segment overheads), we should get a zero window back. + pkt = c.GetPacket() + tcpHdr := header.TCP(header.IPv4(pkt).Payload()) + gotRcvWnd := tcpHdr.WindowSize() + wantAckNum := tcpHdr.AckNumber() + if got, want := int(gotRcvWnd), 0; got != want { + t.Fatalf("got rcvWnd: %d, want: %d", got, want) } - rawEP.VerifyACKRcvWnd(0) time.Sleep(25 * time.Millisecond) - // Verify that sending more data when window is closed is dropped and - // not acked. + // Verify that sending more data when receiveBuffer is exhausted. rawEP.SendPacketWithTS(b[start:start+mss], tsVal) - // Verify that the stack sends us back an ACK with the sequence number - // of the last packet sent indicating it was dropped. - p := c.GetPacket() - checker.IPv4(t, p, checker.TCP( - checker.AckNum(uint32(rawEP.NextSeqNum)-uint32(mss)), - checker.Window(0), - )) - // Now read all the data from the endpoint and verify that advertised // window increases to the full available buffer size. for { @@ -6059,23 +6131,26 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // Verify that we receive a non-zero window update ACK. When running // under thread santizer this test can end up sending more than 1 // ack, 1 for the non-zero window - p = c.GetPacket() + p := c.GetPacket() checker.IPv4(t, p, checker.TCP( - checker.AckNum(uint32(rawEP.NextSeqNum)-uint32(mss)), + checker.TCPAckNum(uint32(wantAckNum)), func(t *testing.T, h header.Transport) { tcp, ok := h.(header.TCP) if !ok { return } - if w := tcp.WindowSize(); w == 0 || w > uint16(wantRcvWnd) { - t.Errorf("expected a non-zero window: got %d, want <= wantRcvWnd", w) + // We use 10% here as the error margin upwards as the initial window we + // got was afer 1 segment was already in the receive buffer queue. + tolerance := 1.1 + if w := tcp.WindowSize(); w == 0 || w > uint16(float64(rcvWnd)*tolerance) { + t.Errorf("expected a non-zero window: got %d, want <= %d", w, uint16(float64(rcvWnd)*tolerance)) } }, )) } -// This test verifies that the auto tuning does not grow the receive buffer if -// the application is not reading the data actively. +// This test verifies that the advertised window is auto-tuned up as the +// application is reading the data that is being received. func TestReceiveBufferAutoTuning(t *testing.T) { const mtu = 1500 const mss = mtu - header.IPv4MinimumSize - header.TCPMinimumSize @@ -6085,9 +6160,6 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // Enable Auto-tuning. stk := c.Stack() - // Set lower limits for auto-tuning tests. This is required because the - // test stops the worker which can cause packets to be dropped because - // the segment queue holding unprocessed packets is limited to 300. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 { @@ -6109,8 +6181,12 @@ func TestReceiveBufferAutoTuning(t *testing.T) { c.WindowScale = uint8(tcp.FindWndScale(maxReceiveBufferSize)) rawEP := c.CreateConnectedWithOptions(header.TCPSynOptions{TS: true, WS: 4}) - - wantRcvWnd := receiveBufferSize + tsVal := uint32(rawEP.TSVal) + rawEP.NextSeqNum-- + rawEP.SendPacketWithTS(nil, tsVal) + rawEP.NextSeqNum++ + pkt := rawEP.VerifyAndReturnACKWithTS(tsVal) + curRcvWnd := int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()) << c.WindowScale scaleRcvWnd := func(rcvWnd int) uint16 { return uint16(rcvWnd >> uint16(c.WindowScale)) } @@ -6127,14 +6203,8 @@ func TestReceiveBufferAutoTuning(t *testing.T) { StopWork() ResumeWork() }) - tsVal := rawEP.TSVal - // We are going to do our own computation of what the moderated receive - // buffer should be based on sent/copied data per RTT and verify that - // the advertised window by the stack matches our calculations. - prevCopied := 0 - done := false latency := 1 * time.Millisecond - for i := 0; !done; i++ { + for i := 0; i < 5; i++ { tsVal++ // Stop the worker goroutine. @@ -6156,15 +6226,20 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // Give 1ms for the worker to process the packets. time.Sleep(1 * time.Millisecond) - // Verify that the advertised window on the ACK is reduced by - // the total bytes sent. - expectedWnd := wantRcvWnd - totalSent - if packetsSent > 100 { - for i := 0; i < (packetsSent / 100); i++ { - _ = c.GetPacket() + lastACK := c.GetPacket() + // Discard any intermediate ACKs and only check the last ACK we get in a + // short time period of few ms. + for { + time.Sleep(1 * time.Millisecond) + pkt := c.GetPacketNonBlocking() + if pkt == nil { + break } + lastACK = pkt + } + if got, want := int(header.TCP(header.IPv4(lastACK).Payload()).WindowSize()), int(scaleRcvWnd(curRcvWnd)); got > want { + t.Fatalf("advertised window got: %d, want <= %d", got, want) } - rawEP.VerifyACKRcvWnd(scaleRcvWnd(expectedWnd)) // Now read all the data from the endpoint and invoke the // moderation API to allow for receive buffer auto-tuning @@ -6189,35 +6264,20 @@ func TestReceiveBufferAutoTuning(t *testing.T) { rawEP.NextSeqNum-- rawEP.SendPacketWithTS(nil, tsVal) rawEP.NextSeqNum++ - if i == 0 { // In the first iteration the receiver based RTT is not // yet known as a result the moderation code should not // increase the advertised window. - rawEP.VerifyACKRcvWnd(scaleRcvWnd(wantRcvWnd)) - prevCopied = totalCopied + rawEP.VerifyACKRcvWnd(scaleRcvWnd(curRcvWnd)) } else { - rttCopied := totalCopied - if i == 1 { - // The moderation code accumulates copied bytes till - // RTT is established. So add in the bytes sent in - // the first iteration to the total bytes for this - // RTT. - rttCopied += prevCopied - // Now reset it to the initial value used by the - // auto tuning logic. - prevCopied = tcp.InitialCwnd * mss * 2 - } - newWnd := rttCopied<<1 + 16*mss - grow := (newWnd * (rttCopied - prevCopied)) / prevCopied - newWnd += (grow << 1) - if newWnd > maxReceiveBufferSize { - newWnd = maxReceiveBufferSize - done = true + pkt := c.GetPacket() + curRcvWnd = int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()) << c.WindowScale + // If thew new current window is close maxReceiveBufferSize then terminate + // the loop. This can happen before all iterations are done due to timing + // differences when running the test. + if int(float64(curRcvWnd)*1.1) > maxReceiveBufferSize/2 { + break } - rawEP.VerifyACKRcvWnd(scaleRcvWnd(newWnd)) - wantRcvWnd = newWnd - prevCopied = rttCopied // Increase the latency after first two iterations to // establish a low RTT value in the receiver since it // only tracks the lowest value. This ensures that when @@ -6230,6 +6290,12 @@ func TestReceiveBufferAutoTuning(t *testing.T) { offset += payloadSize payloadSize *= 2 } + // Check that at the end of our iterations the receive window grew close to the maximum + // permissible size of maxReceiveBufferSize/2 + if got, want := int(float64(curRcvWnd)*1.1), maxReceiveBufferSize/2; got < want { + t.Fatalf("unexpected rcvWnd got: %d, want > %d", got, want) + } + } func TestDelayEnabled(t *testing.T) { @@ -6381,8 +6447,8 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6399,8 +6465,8 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Now send a RST and this should be ignored and not @@ -6428,8 +6494,8 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) } @@ -6500,8 +6566,8 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6518,8 +6584,8 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Out of order ACK should generate an immediate ACK in @@ -6535,8 +6601,8 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) } @@ -6607,8 +6673,8 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6625,8 +6691,8 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Send a SYN request w/ sequence number lower than @@ -6764,8 +6830,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6782,8 +6848,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) time.Sleep(2 * time.Second) @@ -6797,8 +6863,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Sleep for 4 seconds so at this point we are 1 second past the @@ -6826,8 +6892,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(0), + checker.TCPSeqNum(uint32(ackHeaders.AckNum)), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst))) if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { @@ -6926,8 +6992,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Now write a few bytes and then close the endpoint. @@ -6945,8 +7011,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+2), // Acknum is initial sequence number + 1 + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+2), // Acknum is initial sequence number + 1 checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -6960,8 +7026,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)+uint32(len(data))), - checker.AckNum(uint32(iss+2)), + checker.TCPSeqNum(uint32(c.IRS+1)+uint32(len(data))), + checker.TCPAckNum(uint32(iss+2)), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) // First send a partial ACK. @@ -7006,8 +7072,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(0), + checker.TCPSeqNum(uint32(ackHeaders.AckNum)), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst))) } @@ -7043,8 +7109,8 @@ func TestTCPUserTimeout(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -7078,8 +7144,8 @@ func TestTCPUserTimeout(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next)), - checker.AckNum(uint32(0)), + checker.TCPSeqNum(uint32(next)), + checker.TCPAckNum(uint32(0)), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -7140,8 +7206,8 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(uint32(790)), + checker.TCPSeqNum(uint32(c.IRS)), + checker.TCPAckNum(uint32(790)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -7166,8 +7232,8 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(0)), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(0)), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -7183,9 +7249,9 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { } } -func TestIncreaseWindowOnReceive(t *testing.T) { +func TestIncreaseWindowOnRead(t *testing.T) { // This test ensures that the endpoint sends an ack, - // after recv() when the window grows to more than 1 MSS. + // after read() when the window grows by more than 1 MSS. c := context.New(t, defaultMTU) defer c.Cleanup() @@ -7194,10 +7260,9 @@ func TestIncreaseWindowOnReceive(t *testing.T) { // Write chunks of ~30000 bytes. It's important that two // payloads make it equal or longer than MSS. - remain := rcvBuf + remain := rcvBuf * 2 sent := 0 data := make([]byte, defaultMTU/2) - lastWnd := uint16(0) for remain > len(data) { c.SendPacket(data, &context.Headers{ @@ -7210,46 +7275,43 @@ func TestIncreaseWindowOnReceive(t *testing.T) { }) sent += len(data) remain -= len(data) - - lastWnd = uint16(remain) - if remain > 0xffff { - lastWnd = 0xffff - } - checker.IPv4(t, c.GetPacket(), + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(lastWnd), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), checker.TCPFlags(header.TCPFlagAck), ), ) + // Break once the window drops below defaultMTU/2 + if wnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize(); wnd < defaultMTU/2 { + break + } } - if lastWnd == 0xffff || lastWnd == 0 { - t.Fatalf("expected small, non-zero window: %d", lastWnd) - } - - // We now have < 1 MSS in the buffer space. Read the data! An - // ack should be sent in response to that. The window was not - // zero, but it grew to larger than MSS. - if _, _, err := c.EP.Read(nil); err != nil { - t.Fatalf("Read failed: %s", err) - } - - if _, _, err := c.EP.Read(nil); err != nil { - t.Fatalf("Read failed: %s", err) + // We now have < 1 MSS in the buffer space. Read at least > 2 MSS + // worth of data as receive buffer space + read := 0 + // defaultMTU is a good enough estimate for the MSS used for this + // connection. + for read < defaultMTU*2 { + v, _, err := c.EP.Read(nil) + if err != nil { + t.Fatalf("Read failed: %s", err) + } + read += len(v) } - // After reading two packets, we surely crossed MSS. See the ack: + // After reading > MSS worth of data, we surely crossed MSS. See the ack: checker.IPv4(t, c.GetPacket(), checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(0xffff)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindow(uint16(0xffff)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -7266,10 +7328,9 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) { // Write chunks of ~30000 bytes. It's important that two // payloads make it equal or longer than MSS. - remain := rcvBuf + remain := rcvBuf * 2 sent := 0 data := make([]byte, defaultMTU/2) - lastWnd := uint16(0) for remain > len(data) { c.SendPacket(data, &context.Headers{ @@ -7283,38 +7344,29 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) { sent += len(data) remain -= len(data) - lastWnd = uint16(remain) - if remain > 0xffff { - lastWnd = 0xffff - } checker.IPv4(t, c.GetPacket(), checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(lastWnd), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindowLessThanEq(0xffff), checker.TCPFlags(header.TCPFlagAck), ), ) } - if lastWnd == 0xffff || lastWnd == 0 { - t.Fatalf("expected small, non-zero window: %d", lastWnd) - } - // Increasing the buffer from should generate an ACK, // since window grew from small value to larger equal MSS c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBuf*2) - // After reading two packets, we surely crossed MSS. See the ack: checker.IPv4(t, c.GetPacket(), checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(0xffff)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindow(uint16(0xffff)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -7359,8 +7411,8 @@ func TestTCPDeferAccept(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) // Give a bit of time for the socket to be delivered to the accept queue. time.Sleep(50 * time.Millisecond) @@ -7374,8 +7426,8 @@ func TestTCPDeferAccept(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } func TestTCPDeferAcceptTimeout(t *testing.T) { @@ -7412,7 +7464,7 @@ func TestTCPDeferAcceptTimeout(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1))) + checker.TCPAckNum(uint32(irs)+1))) // Send data. This should result in an acceptable endpoint. c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ @@ -7428,8 +7480,8 @@ func TestTCPDeferAcceptTimeout(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) // Give sometime for the endpoint to be delivered to the accept queue. time.Sleep(50 * time.Millisecond) @@ -7444,8 +7496,8 @@ func TestTCPDeferAcceptTimeout(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } func TestResetDuringClose(t *testing.T) { @@ -7470,8 +7522,8 @@ func TestResetDuringClose(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(irs.Add(1))), - checker.AckNum(uint32(iss.Add(5))))) + checker.TCPSeqNum(uint32(irs.Add(1))), + checker.TCPAckNum(uint32(iss.Add(5))))) // Close in a separate goroutine so that we can trigger // a race with the RST we send below. This should not @@ -7552,3 +7604,14 @@ func TestSetStackTimeWaitReuse(t *testing.T) { } } } + +// generateRandomPayload generates a random byte slice of the specified length +// causing a fatal test failure if it is unable to do so. +func generateRandomPayload(t *testing.T, n int) []byte { + t.Helper() + buf := make([]byte, n) + if _, err := rand.Read(buf); err != nil { + t.Fatalf("rand.Read(buf) failed: %s", err) + } + return buf +} diff --git a/pkg/tcpip/transport/tcp/tcp_timestamp_test.go b/pkg/tcpip/transport/tcp/tcp_timestamp_test.go index 44593ed98..0f9ed06cd 100644 --- a/pkg/tcpip/transport/tcp/tcp_timestamp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_timestamp_test.go @@ -159,9 +159,9 @@ func timeStampEnabledAccept(t *testing.T, cookieEnabled bool, wndScale int, wndS checker.PayloadLen(len(data)+header.TCPMinimumSize+12), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(wndSize), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(wndSize), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), checker.TCPTimestampChecker(true, 0, tsVal+1), ), @@ -181,7 +181,8 @@ func TestTimeStampEnabledAccept(t *testing.T) { wndSize uint16 }{ {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // DefaultReceiveBufferSize is 1MB >> 5. + // DefaultReceiveBufferSize is 1MB >> 5. Advertised window will be 1/2 of that. + {false, 5, 0x4000}, } for _, tc := range testCases { timeStampEnabledAccept(t, tc.cookieEnabled, tc.wndScale, tc.wndSize) @@ -219,9 +220,9 @@ func timeStampDisabledAccept(t *testing.T, cookieEnabled bool, wndScale int, wnd checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(wndSize), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(wndSize), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), checker.TCPTimestampChecker(false, 0, 0), ), @@ -237,7 +238,9 @@ func TestTimeStampDisabledAccept(t *testing.T) { wndSize uint16 }{ {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // DefaultReceiveBufferSize is 1MB >> 5. + // DefaultReceiveBufferSize is 1MB >> 5. Advertised window will be half of + // that. + {false, 5, 0x4000}, } for _, tc := range testCases { timeStampDisabledAccept(t, tc.cookieEnabled, tc.wndScale, tc.wndSize) diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 059c13821..faf51ef95 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -145,14 +145,18 @@ type Context struct { // WindowScale is the expected window scale in SYN packets sent by // the stack. WindowScale uint8 + + // RcvdWindowScale is the actual window scale sent by the stack in + // SYN/SYN-ACK. + RcvdWindowScale uint8 } // New allocates and initializes a test context containing a new // stack and a link-layer endpoint. func New(t *testing.T, mtu uint32) *Context { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) const sendBufferSize = 1 << 20 // 1 MiB @@ -261,18 +265,17 @@ func (c *Context) CheckNoPacket(errMsg string) { c.CheckNoPacketTimeout(errMsg, 1*time.Second) } -// GetPacket reads a packet from the link layer endpoint and verifies +// GetPacketWithTimeout reads a packet from the link layer endpoint and verifies // that it is an IPv4 packet with the expected source and destination -// addresses. It will fail with an error if no packet is received for -// 2 seconds. -func (c *Context) GetPacket() []byte { +// addresses. If no packet is received in the specified timeout it will return +// nil. +func (c *Context) GetPacketWithTimeout(timeout time.Duration) []byte { c.t.Helper() - ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second) + ctx, cancel := context.WithTimeout(context.Background(), timeout) defer cancel() p, ok := c.linkEP.ReadContext(ctx) if !ok { - c.t.Fatalf("Packet wasn't written out") return nil } @@ -299,6 +302,21 @@ func (c *Context) GetPacket() []byte { return b } +// GetPacket reads a packet from the link layer endpoint and verifies +// that it is an IPv4 packet with the expected source and destination +// addresses. +func (c *Context) GetPacket() []byte { + c.t.Helper() + + p := c.GetPacketWithTimeout(5 * time.Second) + if p == nil { + c.t.Fatalf("Packet wasn't written out") + return nil + } + + return p +} + // GetPacketNonBlocking reads a packet from the link layer endpoint // and verifies that it is an IPv4 packet with the expected source // and destination address. If no packet is available it will return @@ -486,8 +504,8 @@ func (c *Context) ReceiveAndCheckPacketWithOptions(data []byte, offset, size, op checker.PayloadLen(size+header.TCPMinimumSize+optlen), checker.TCP( checker.DstPort(TestPort), - checker.SeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), - checker.AckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), + checker.TCPSeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), + checker.TCPAckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -513,8 +531,8 @@ func (c *Context) ReceiveNonBlockingAndCheckPacket(data []byte, offset, size int checker.PayloadLen(size+header.TCPMinimumSize), checker.TCP( checker.DstPort(TestPort), - checker.SeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), - checker.AckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), + checker.TCPSeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), + checker.TCPAckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -652,6 +670,7 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) } tcpHdr := header.TCP(header.IPv4(b).Payload()) + synOpts := header.ParseSynOptions(tcpHdr.Options(), false /* isAck */) c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) c.SendPacket(nil, &Headers{ @@ -669,8 +688,8 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) checker.TCP( checker.DstPort(TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), ), ) @@ -687,6 +706,7 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) } + c.RcvdWindowScale = uint8(synOpts.WS) c.Port = tcpHdr.SourcePort() } @@ -758,17 +778,18 @@ func (r *RawEndpoint) SendPacket(payload []byte, opts []byte) { r.NextSeqNum = r.NextSeqNum.Add(seqnum.Size(len(payload))) } -// VerifyACKWithTS verifies that the tsEcr field in the ack matches the provided -// tsVal. -func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { +// VerifyAndReturnACKWithTS verifies that the tsEcr field int he ACK matches +// the provided tsVal as well as returns the original packet. +func (r *RawEndpoint) VerifyAndReturnACKWithTS(tsVal uint32) []byte { + r.C.t.Helper() // Read ACK and verify that tsEcr of ACK packet is [1,2,3,4] ackPacket := r.C.GetPacket() checker.IPv4(r.C.t, ackPacket, checker.TCP( checker.DstPort(r.SrcPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), + checker.TCPSeqNum(uint32(r.AckNum)), + checker.TCPAckNum(uint32(r.NextSeqNum)), checker.TCPTimestampChecker(true, 0, tsVal), ), ) @@ -776,19 +797,28 @@ func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { tcpSeg := header.TCP(header.IPv4(ackPacket).Payload()) opts := tcpSeg.ParsedOptions() r.RecentTS = opts.TSVal + return ackPacket +} + +// VerifyACKWithTS verifies that the tsEcr field in the ack matches the provided +// tsVal. +func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { + r.C.t.Helper() + _ = r.VerifyAndReturnACKWithTS(tsVal) } // VerifyACKRcvWnd verifies that the window advertised by the incoming ACK // matches the provided rcvWnd. func (r *RawEndpoint) VerifyACKRcvWnd(rcvWnd uint16) { + r.C.t.Helper() ackPacket := r.C.GetPacket() checker.IPv4(r.C.t, ackPacket, checker.TCP( checker.DstPort(r.SrcPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), - checker.Window(rcvWnd), + checker.TCPSeqNum(uint32(r.AckNum)), + checker.TCPAckNum(uint32(r.NextSeqNum)), + checker.TCPWindow(rcvWnd), ), ) } @@ -807,8 +837,8 @@ func (r *RawEndpoint) VerifyACKHasSACK(sackBlocks []header.SACKBlock) { checker.TCP( checker.DstPort(r.SrcPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), + checker.TCPSeqNum(uint32(r.AckNum)), + checker.TCPAckNum(uint32(r.NextSeqNum)), checker.TCPSACKBlockChecker(sackBlocks), ), ) @@ -900,8 +930,8 @@ func (c *Context) CreateConnectedWithOptions(wantOptions header.TCPSynOptions) * tcpCheckers := []checker.TransportChecker{ checker.DstPort(TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS) + 1), - checker.AckNum(uint32(iss) + 1), + checker.TCPSeqNum(uint32(c.IRS) + 1), + checker.TCPAckNum(uint32(iss) + 1), } // Verify that tsEcr of ACK packet is wantOptions.TSVal if the @@ -936,7 +966,7 @@ func (c *Context) CreateConnectedWithOptions(wantOptions header.TCPSynOptions) * // Mark in context that timestamp option is enabled for this endpoint. c.TimeStampEnabled = true - + c.RcvdWindowScale = uint8(synOptions.WS) return &RawEndpoint{ C: c, SrcPort: tcpSeg.DestinationPort(), @@ -1029,6 +1059,7 @@ func (c *Context) PassiveConnect(maxPayload, wndScale int, synOptions header.TCP // value of the window scaling option to be sent in the SYN. If synOptions.WS > // 0 then we send the WindowScale option. func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions header.TCPSynOptions) *RawEndpoint { + c.t.Helper() opts := make([]byte, header.TCPOptionsMaximumSize) offset := 0 offset += header.EncodeMSSOption(uint32(maxPayload), opts) @@ -1067,13 +1098,14 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // are present. b := c.GetPacket() tcp := header.TCP(header.IPv4(b).Payload()) + rcvdSynOptions := header.ParseSynOptions(tcp.Options(), true /* isAck */) c.IRS = seqnum.Value(tcp.SequenceNumber()) tcpCheckers := []checker.TransportChecker{ checker.SrcPort(StackPort), checker.DstPort(TestPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(iss) + 1), + checker.TCPAckNum(uint32(iss) + 1), checker.TCPSynOptions(header.TCPSynOptions{MSS: synOptions.MSS, WS: wndScale, SACKPermitted: synOptions.SACKPermitted && c.SACKEnabled()}), } @@ -1116,6 +1148,7 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // Send ACK. c.SendPacket(nil, ackHeaders) + c.RcvdWindowScale = uint8(rcvdSynOptions.WS) c.Port = StackPort return &RawEndpoint{ diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index a1d0f49d9..da5b1deb2 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -12,12 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package udp contains the implementation of the UDP transport protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing udp.NewProtocol() as one of the -// transport protocols when calling stack.New(). Then endpoints can be created -// by passing udp.ProtocolNumber as the transport protocol number when calling -// Stack.NewEndpoint(). +// Package udp contains the implementation of the UDP transport protocol. package udp import ( @@ -50,6 +45,7 @@ const ( ) type protocol struct { + stack *stack.Stack } // Number returns the udp protocol number. @@ -58,14 +54,14 @@ func (*protocol) Number() tcpip.TransportProtocolNumber { } // NewEndpoint creates a new udp endpoint. -func (*protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newEndpoint(stack, netProto, waiterQueue), nil +func (p *protocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return newEndpoint(p.stack, netProto, waiterQueue), nil } // NewRawEndpoint creates a new raw UDP endpoint. It implements // stack.TransportProtocol.NewRawEndpoint. -func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return raw.NewEndpoint(stack, netProto, header.UDPProtocolNumber, waiterQueue) +func (p *protocol) NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return raw.NewEndpoint(p.stack, netProto, header.UDPProtocolNumber, waiterQueue) } // MinimumPacketSize returns the minimum valid udp packet size. @@ -119,6 +115,6 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { } // NewProtocol returns a UDP transport protocol. -func NewProtocol() stack.TransportProtocol { - return &protocol{} +func NewProtocol(s *stack.Stack) stack.TransportProtocol { + return &protocol{stack: s} } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 64a5fc696..bac084acf 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -294,8 +294,8 @@ type testContext struct { func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() return newDualTestContextWithOptions(t, mtu, stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) } @@ -532,8 +532,8 @@ func newMinPayload(minSize int) []byte { func TestBindToDeviceOption(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}}) ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) if err != nil { @@ -807,8 +807,8 @@ func TestV4ReadSelfSource(t *testing.T) { } { t.Run(tt.name, func(t *testing.T) { c := newDualTestContextWithOptions(t, defaultMTU, stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, HandleLocal: tt.handleLocal, }) defer c.cleanup() @@ -1466,6 +1466,30 @@ func TestNoChecksum(t *testing.T) { } } +var _ stack.NetworkInterface = (*testInterface)(nil) + +type testInterface struct{} + +func (*testInterface) ID() tcpip.NICID { + return 0 +} + +func (*testInterface) IsLoopback() bool { + return false +} + +func (*testInterface) Name() string { + return "" +} + +func (*testInterface) Enabled() bool { + return true +} + +func (*testInterface) LinkEndpoint() stack.LinkEndpoint { + return nil +} + func TestTTL(t *testing.T) { for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -1483,16 +1507,19 @@ func TestTTL(t *testing.T) { if flow.isMulticast() { wantTTL = multicastTTL } else { - var p stack.NetworkProtocol + var p stack.NetworkProtocolFactory + var n tcpip.NetworkProtocolNumber if flow.isV4() { - p = ipv4.NewProtocol() + p = ipv4.NewProtocol + n = ipv4.ProtocolNumber } else { - p = ipv6.NewProtocol() + p = ipv6.NewProtocol + n = ipv6.ProtocolNumber } - ep := p.NewEndpoint(0, nil, nil, nil, nil, stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - })) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{p}, + }) + ep := s.NetworkProtocolInstance(n).NewEndpoint(&testInterface{}, nil, nil, nil) wantTTL = ep.DefaultTTL() ep.Close() } @@ -1516,18 +1543,6 @@ func TestSetTTL(t *testing.T) { c.t.Fatalf("SetSockOptInt(TTLOption, %d) failed: %s", wantTTL, err) } - var p stack.NetworkProtocol - if flow.isV4() { - p = ipv4.NewProtocol() - } else { - p = ipv6.NewProtocol() - } - ep := p.NewEndpoint(0, nil, nil, nil, nil, stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - })) - ep.Close() - testWrite(c, flow, checker.TTL(wantTTL)) }) } @@ -2357,9 +2372,8 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, defaultMTU, "") if err := s.CreateNIC(nicID1, e); err != nil { diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 4940ea96a..2e652ddad 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1059,8 +1059,8 @@ func newRootNetworkNamespace(conf *config.Config, clock tcpip.Clock, uniqueID st } func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (inet.Stack, error) { - netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()} - transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()} + netProtos := []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol, arp.NewProtocol} + transProtos := []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol, icmp.NewProtocol4} s := netstack.Stack{stack.New(stack.Options{ NetworkProtocols: netProtos, TransportProtocols: transProtos, diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index ff0e60283..548c68087 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -1843,8 +1843,9 @@ func TestUserLog(t *testing.T) { t.Fatal("error finding test_app:", err) } - // sched_rr_get_interval = 148 - not implemented in gvisor. - spec := testutil.NewSpecWithArgs(app, "syscall", "--syscall=148") + // sched_rr_get_interval - not implemented in gvisor. + num := strconv.Itoa(syscall.SYS_SCHED_RR_GET_INTERVAL) + spec := testutil.NewSpecWithArgs(app, "syscall", "--syscall="+num) conf := testutil.TestConfig(t) _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { diff --git a/runsc/fsgofer/filter/extra_filters_race.go b/runsc/fsgofer/filter/extra_filters_race.go index 885c92f7a..20a0732be 100644 --- a/runsc/fsgofer/filter/extra_filters_race.go +++ b/runsc/fsgofer/filter/extra_filters_race.go @@ -35,6 +35,7 @@ func instrumentationFilters() seccomp.SyscallRules { syscall.SYS_MUNLOCK: {}, syscall.SYS_NANOSLEEP: {}, syscall.SYS_OPEN: {}, + syscall.SYS_OPENAT: {}, syscall.SYS_SET_ROBUST_LIST: {}, // Used within glibc's malloc. syscall.SYS_TIME: {}, diff --git a/runsc/specutils/seccomp/seccomp_test.go b/runsc/specutils/seccomp/seccomp_test.go index 2079cd2e9..850c237ba 100644 --- a/runsc/specutils/seccomp/seccomp_test.go +++ b/runsc/specutils/seccomp/seccomp_test.go @@ -169,7 +169,7 @@ var ( }, }, }, - input: testInput(nativeArchAuditNo, "open", nil), + input: testInput(nativeArchAuditNo, "openat", nil), expected: uint32(allowAction), }, { diff --git a/test/benchmarks/base/size_test.go b/test/benchmarks/base/size_test.go index 3c1364faf..7d3877459 100644 --- a/test/benchmarks/base/size_test.go +++ b/test/benchmarks/base/size_test.go @@ -105,6 +105,7 @@ func BenchmarkSizeNginx(b *testing.B) { machine: machine, port: port, runOpts: runOpts, + cmd: []string{"nginx", "-c", "/etc/nginx/nginx_gofer.conf"}, }) defer cleanUpContainers(ctx, servers) diff --git a/test/benchmarks/base/startup_test.go b/test/benchmarks/base/startup_test.go index 4628a0a41..c36a544db 100644 --- a/test/benchmarks/base/startup_test.go +++ b/test/benchmarks/base/startup_test.go @@ -64,6 +64,7 @@ func BenchmarkStartupNginx(b *testing.B) { machine: machine, runOpts: runOpts, port: 80, + cmd: []string{"nginx", "-c", "/etc/nginx/nginx_gofer.conf"}, }) } @@ -123,8 +124,6 @@ func redisInstance(ctx context.Context, b *testing.B, machine harness.Machine) ( // runServerWorkload runs a server workload defined by 'runOpts' and 'cmd'. // 'clientMachine' is used to connect to the server on 'serverMachine'. func runServerWorkload(ctx context.Context, b *testing.B, args serverArgs) { - b.Helper() - b.ResetTimer() for i := 0; i < b.N; i++ { if err := func() error { diff --git a/test/benchmarks/network/nginx_test.go b/test/benchmarks/network/nginx_test.go index 036fd666f..9ec70369b 100644 --- a/test/benchmarks/network/nginx_test.go +++ b/test/benchmarks/network/nginx_test.go @@ -36,50 +36,63 @@ var nginxDocs = map[string]string{ func BenchmarkNginxConcurrency(b *testing.B) { concurrency := []int{1, 25, 100, 1000} for _, c := range concurrency { - b.Run(fmt.Sprintf("%d", c), func(b *testing.B) { - hey := &tools.Hey{ - Requests: c * b.N, - Concurrency: c, - Doc: nginxDocs["10kb"], // see Dockerfile '//images/benchmarks/nginx' and httpd_test. + for _, tmpfs := range []bool{true, false} { + fs := "Gofer" + if tmpfs { + fs = "Tmpfs" } - runNginx(b, hey, false /* reverse */) - }) + name := fmt.Sprintf("%d_%s", c, fs) + b.Run(name, func(b *testing.B) { + hey := &tools.Hey{ + Requests: c * b.N, + Concurrency: c, + Doc: nginxDocs["10kb"], // see Dockerfile '//images/benchmarks/nginx' and httpd_test. + } + runNginx(b, hey, false /* reverse */, tmpfs /* tmpfs */) + }) + } + } } // BenchmarkNginxDocSize iterates over different sized payloads, testing how // well the runtime handles sending different payload sizes. func BenchmarkNginxDocSize(b *testing.B) { - benchmarkHttpdDocSize(b, false /* reverse */) + benchmarkNginxDocSize(b, false /* reverse */, true /* tmpfs */) + benchmarkNginxDocSize(b, false /* reverse */, false /* tmpfs */) } // BenchmarkReverseNginxDocSize iterates over different sized payloads, testing // how well the runtime handles receiving different payload sizes. func BenchmarkReverseNginxDocSize(b *testing.B) { - benchmarkHttpdDocSize(b, true /* reverse */) + benchmarkNginxDocSize(b, true /* reverse */, true /* tmpfs */) } // benchmarkNginxDocSize iterates through all doc sizes, running subbenchmarks // for each size. -func benchmarkNginxDocSize(b *testing.B, reverse bool) { - b.Helper() +func benchmarkNginxDocSize(b *testing.B, reverse, tmpfs bool) { for name, filename := range nginxDocs { concurrency := []int{1, 25, 50, 100, 1000} for _, c := range concurrency { - b.Run(fmt.Sprintf("%s_%d", name, c), func(b *testing.B) { + fs := "Gofer" + if tmpfs { + fs = "Tmpfs" + } + benchName := fmt.Sprintf("%s_%d_%s", name, c, fs) + b.Run(benchName, func(b *testing.B) { hey := &tools.Hey{ Requests: c * b.N, Concurrency: c, Doc: filename, } - runNginx(b, hey, reverse) + runNginx(b, hey, reverse, tmpfs) }) } } } // runNginx configures the static serving methods to run httpd. -func runNginx(b *testing.B, hey *tools.Hey, reverse bool) { +func runNginx(b *testing.B, hey *tools.Hey, reverse, tmpfs bool) { // nginx runs on port 80. port := 80 nginxRunOpts := dockerutil.RunOpts{ @@ -87,7 +100,11 @@ func runNginx(b *testing.B, hey *tools.Hey, reverse bool) { Ports: []int{port}, } + nginxCmd := []string{"nginx", "-c", "/etc/nginx/nginx_gofer.conf"} + if tmpfs { + nginxCmd = []string{"sh", "-c", "mkdir -p /tmp/html && cp -a /local/* /tmp/html && nginx -c /etc/nginx/nginx.conf"} + } + // Command copies nginxDocs to tmpfs serving directory and runs nginx. - nginxCmd := []string{"sh", "-c", "mkdir -p /tmp/html && cp -a /local/* /tmp/html && nginx"} runStaticServer(b, nginxRunOpts, nginxCmd, port, hey, reverse) } diff --git a/test/benchmarks/network/static_server.go b/test/benchmarks/network/static_server.go index 3ef62a71f..e747a1395 100644 --- a/test/benchmarks/network/static_server.go +++ b/test/benchmarks/network/static_server.go @@ -25,7 +25,6 @@ import ( // runStaticServer runs static serving workloads (httpd, nginx). func runStaticServer(b *testing.B, serverOpts dockerutil.RunOpts, serverCmd []string, port int, hey *tools.Hey, reverse bool) { - b.Helper() ctx := context.Background() // Get two machines: a client and server. diff --git a/test/benchmarks/tcp/tcp_proxy.go b/test/benchmarks/tcp/tcp_proxy.go index 6cabfb451..5afe10f69 100644 --- a/test/benchmarks/tcp/tcp_proxy.go +++ b/test/benchmarks/tcp/tcp_proxy.go @@ -174,8 +174,8 @@ func newNetstackImpl(mode string) (impl, error) { } // Create a new network stack. - netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()} - transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol()} + netProtos := []stack.NetworkProtocolFactory{ipv4.NewProtocol, arp.NewProtocol} + transProtos := []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol} s := stack.New(stack.Options{ NetworkProtocols: netProtos, TransportProtocols: transProtos, diff --git a/test/packetimpact/tests/BUILD b/test/packetimpact/tests/BUILD index fbfea61e1..94731c64b 100644 --- a/test/packetimpact/tests/BUILD +++ b/test/packetimpact/tests/BUILD @@ -340,3 +340,13 @@ packetimpact_go_test( "@org_golang_x_sys//unix:go_default_library", ], ) + +packetimpact_go_test( + name = "tcp_rcv_buf_space", + srcs = ["tcp_rcv_buf_space_test.go"], + deps = [ + "//pkg/tcpip/header", + "//test/packetimpact/testbench", + "@org_golang_x_sys//unix:go_default_library", + ], +) diff --git a/test/packetimpact/tests/tcp_linger_test.go b/test/packetimpact/tests/tcp_linger_test.go index 913e49e06..b9a0409aa 100644 --- a/test/packetimpact/tests/tcp_linger_test.go +++ b/test/packetimpact/tests/tcp_linger_test.go @@ -251,3 +251,20 @@ func TestTCPLingerShutdownSendNonZeroTimeout(t *testing.T) { }) } } + +func TestTCPLingerNonEstablished(t *testing.T) { + dut := testbench.NewDUT(t) + newFD := dut.Socket(t, unix.AF_INET, unix.SOCK_STREAM, unix.IPPROTO_TCP) + dut.SetSockLingerOption(t, newFD, lingerDuration, true) + + // As the socket is in the initial state, Close() should not linger + // and return immediately. + start := time.Now() + dut.CloseWithErrno(context.Background(), t, newFD) + diff := time.Since(start) + + if diff > lingerDuration { + t.Errorf("expected close to return within %s, but returned after %s", lingerDuration, diff) + } + dut.TearDown() +} diff --git a/test/packetimpact/tests/tcp_rcv_buf_space_test.go b/test/packetimpact/tests/tcp_rcv_buf_space_test.go new file mode 100644 index 000000000..cfbba1e8e --- /dev/null +++ b/test/packetimpact/tests/tcp_rcv_buf_space_test.go @@ -0,0 +1,80 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tcp_rcv_buf_space_test + +import ( + "context" + "flag" + "syscall" + "testing" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/test/packetimpact/testbench" +) + +func init() { + testbench.RegisterFlags(flag.CommandLine) +} + +// TestReduceRecvBuf tests that a packet within window is still dropped +// if the available buffer space drops below the size of the incoming +// segment. +func TestReduceRecvBuf(t *testing.T) { + dut := testbench.NewDUT(t) + defer dut.TearDown() + listenFd, remotePort := dut.CreateListener(t, unix.SOCK_STREAM, unix.IPPROTO_TCP, 1) + defer dut.Close(t, listenFd) + conn := testbench.NewTCPIPv4(t, testbench.TCP{DstPort: &remotePort}, testbench.TCP{SrcPort: &remotePort}) + defer conn.Close(t) + + conn.Connect(t) + acceptFd, _ := dut.Accept(t, listenFd) + defer dut.Close(t, acceptFd) + + // Set a small receive buffer for the test. + const rcvBufSz = 4096 + dut.SetSockOptInt(t, acceptFd, unix.SOL_SOCKET, unix.SO_RCVBUF, rcvBufSz) + + // Retrieve the actual buffer. + bufSz := dut.GetSockOptInt(t, acceptFd, unix.SOL_SOCKET, unix.SO_RCVBUF) + + // Generate a payload of 1 more than the actual buffer size used by the + // DUT. + sampleData := testbench.GenerateRandomPayload(t, int(bufSz)+1) + // Send and receive sample data to the dut. + const pktSize = 1400 + for payload := sampleData; len(payload) != 0; { + payloadBytes := pktSize + if l := len(payload); l < payloadBytes { + payloadBytes = l + } + + conn.Send(t, testbench.TCP{Flags: testbench.Uint8(header.TCPFlagAck)}, []testbench.Layer{&testbench.Payload{Bytes: payload[:payloadBytes]}}...) + payload = payload[payloadBytes:] + } + + // First read should read < len(sampleData) + if ret, _, err := dut.RecvWithErrno(context.Background(), t, acceptFd, int32(len(sampleData)), 0); ret == -1 || int(ret) == len(sampleData) { + t.Fatalf("dut.RecvWithErrno(ctx, t, %d, %d, 0) = %d,_, %s", acceptFd, int32(len(sampleData)), ret, err) + } + + // Second read should return EAGAIN as the last segment should have been + // dropped due to it exceeding the receive buffer space available in the + // socket. + if ret, got, err := dut.RecvWithErrno(context.Background(), t, acceptFd, int32(len(sampleData)), syscall.MSG_DONTWAIT); got != nil || ret != -1 || err != syscall.EAGAIN { + t.Fatalf("expected no packets but got: %s", got) + } +} diff --git a/test/runtimes/exclude/java11.csv b/test/runtimes/exclude/java11.csv index 997a29cad..f779df8d5 100644 --- a/test/runtimes/exclude/java11.csv +++ b/test/runtimes/exclude/java11.csv @@ -1,9 +1,11 @@ test name,bug id,comment com/sun/crypto/provider/Cipher/PBE/PKCS12Cipher.java,,Fails in Docker +com/sun/jdi/InvokeHangTest.java,https://bugs.openjdk.java.net/browse/JDK-8218463, com/sun/jdi/NashornPopFrameTest.java,, com/sun/jdi/ProcessAttachTest.java,, com/sun/management/HotSpotDiagnosticMXBean/CheckOrigin.java,,Fails in Docker com/sun/management/OperatingSystemMXBean/GetCommittedVirtualMemorySize.java,, +com/sun/management/ThreadMXBean/ThreadCpuTimeArray.java,,Test assumes high CPU clock precision com/sun/management/UnixOperatingSystemMXBean/GetMaxFileDescriptorCount.sh,, com/sun/tools/attach/AttachSelf.java,, com/sun/tools/attach/BasicTests.java,, diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index f949bc0e3..96a775456 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -238,7 +238,7 @@ syscall_test( syscall_test( size = "medium", - add_overlay = False, # TODO(gvisor.dev/issue/317): enable when fixed. + add_overlay = True, test = "//test/syscalls/linux:inotify_test", ) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index c775a6d75..d9dbe2267 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1667,6 +1667,7 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/container:node_hash_set", "@com_google_absl//absl/strings", "@com_google_absl//absl/synchronization", "@com_google_absl//absl/time", diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index a5c421118..e4392a450 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -465,7 +465,9 @@ TEST(Inotify, ConcurrentFileDeletionAndWatchRemoval) { for (int i = 0; i < 100; ++i) { FileDescriptor file_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(filename, O_CREAT, S_IRUSR | S_IWUSR)); - file_fd.reset(); // Close before unlinking (although save is disabled). + // Close before unlinking (although S/R is disabled). Some filesystems + // cannot restore an open fd on an unlinked file. + file_fd.reset(); EXPECT_THAT(unlink(filename.c_str()), SyscallSucceeds()); } }; @@ -1256,10 +1258,7 @@ TEST(Inotify, MknodGeneratesCreateEvent) { InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); const TempPath file1(root.path() + "/file1"); - const int rc = mknod(file1.path().c_str(), S_IFREG, 0); - // mknod(2) is only supported on tmpfs in the sandbox. - SKIP_IF(IsRunningOnGvisor() && rc != 0); - ASSERT_THAT(rc, SyscallSucceeds()); + ASSERT_THAT(mknod(file1.path().c_str(), S_IFREG, 0), SyscallSucceeds()); const std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); @@ -1289,6 +1288,10 @@ TEST(Inotify, SymlinkGeneratesCreateEvent) { } TEST(Inotify, LinkGeneratesAttribAndCreateEvents) { + // Inotify does not work properly with hard links in gofer and overlay fs. + SKIP_IF(IsRunningOnGvisor() && + !ASSERT_NO_ERRNO_AND_VALUE(IsTmpfs(GetAbsoluteTestTmpdir()))); + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const TempPath file1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); @@ -1301,11 +1304,8 @@ TEST(Inotify, LinkGeneratesAttribAndCreateEvents) { const int file1_wd = ASSERT_NO_ERRNO_AND_VALUE( InotifyAddWatch(fd.get(), file1.path(), IN_ALL_EVENTS)); - const int rc = link(file1.path().c_str(), link1.path().c_str()); - // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. - SKIP_IF(IsRunningOnGvisor() && rc != 0 && - (errno == EPERM || errno == ENOENT)); - ASSERT_THAT(rc, SyscallSucceeds()); + ASSERT_THAT(link(file1.path().c_str(), link1.path().c_str()), + SyscallSucceeds()); const std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); @@ -1334,68 +1334,70 @@ TEST(Inotify, UtimesGeneratesAttribEvent) { } TEST(Inotify, HardlinksReuseSameWatch) { + // Inotify does not work properly with hard links in gofer and overlay fs. + SKIP_IF(IsRunningOnGvisor() && + !ASSERT_NO_ERRNO_AND_VALUE(IsTmpfs(GetAbsoluteTestTmpdir()))); + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - TempPath file1 = + TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); - TempPath link1(root.path() + "/link1"); - const int rc = link(file1.path().c_str(), link1.path().c_str()); - // link(2) is only supported on tmpfs in the sandbox. - SKIP_IF(IsRunningOnGvisor() && rc != 0 && - (errno == EPERM || errno == ENOENT)); - ASSERT_THAT(rc, SyscallSucceeds()); + + TempPath file2(root.path() + "/file2"); + ASSERT_THAT(link(file.path().c_str(), file2.path().c_str()), + SyscallSucceeds()); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); const int root_wd = ASSERT_NO_ERRNO_AND_VALUE( InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); - const int file1_wd = ASSERT_NO_ERRNO_AND_VALUE( - InotifyAddWatch(fd.get(), file1.path(), IN_ALL_EVENTS)); - const int link1_wd = ASSERT_NO_ERRNO_AND_VALUE( - InotifyAddWatch(fd.get(), link1.path(), IN_ALL_EVENTS)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), file.path(), IN_ALL_EVENTS)); + const int file2_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), file2.path(), IN_ALL_EVENTS)); // The watch descriptors for watches on different links to the same file // should be identical. - EXPECT_NE(root_wd, file1_wd); - EXPECT_EQ(file1_wd, link1_wd); + EXPECT_NE(root_wd, file_wd); + EXPECT_EQ(file_wd, file2_wd); - FileDescriptor file1_fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_WRONLY)); + FileDescriptor file_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_WRONLY)); std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); ASSERT_THAT(events, - AreUnordered({Event(IN_OPEN, root_wd, Basename(file1.path())), - Event(IN_OPEN, file1_wd)})); + AreUnordered({Event(IN_OPEN, root_wd, Basename(file.path())), + Event(IN_OPEN, file_wd)})); // For the next step, we want to ensure all fds to the file are closed. Do // that now and drain the resulting events. - file1_fd.reset(); + file_fd.reset(); events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); ASSERT_THAT( events, - AreUnordered({Event(IN_CLOSE_WRITE, root_wd, Basename(file1.path())), - Event(IN_CLOSE_WRITE, file1_wd)})); + AreUnordered({Event(IN_CLOSE_WRITE, root_wd, Basename(file.path())), + Event(IN_CLOSE_WRITE, file_wd)})); // Try removing the link and let's see what events show up. Note that after // this, we still have a link to the file so the watch shouldn't be // automatically removed. - const std::string link1_path = link1.reset(); + const std::string file2_path = file2.reset(); events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); ASSERT_THAT(events, - AreUnordered({Event(IN_ATTRIB, link1_wd), - Event(IN_DELETE, root_wd, Basename(link1_path))})); + AreUnordered({Event(IN_ATTRIB, file2_wd), + Event(IN_DELETE, root_wd, Basename(file2_path))})); // Now remove the other link. Since this is the last link to the file, the // watch should be automatically removed. - const std::string file1_path = file1.reset(); + const std::string file_path = file.reset(); events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); ASSERT_THAT( events, - AreUnordered({Event(IN_ATTRIB, file1_wd), Event(IN_DELETE_SELF, file1_wd), - Event(IN_IGNORED, file1_wd), - Event(IN_DELETE, root_wd, Basename(file1_path))})); + AreUnordered({Event(IN_ATTRIB, file_wd), Event(IN_DELETE_SELF, file_wd), + Event(IN_IGNORED, file_wd), + Event(IN_DELETE, root_wd, Basename(file_path))})); } // Calling mkdir within "parent/child" should generate an event for child, but @@ -1806,17 +1808,17 @@ TEST(Inotify, SpliceOnInotifyFD) { // Watches on a parent should not be triggered by actions on a hard link to one // of its children that has a different parent. TEST(Inotify, LinkOnOtherParent) { + // Inotify does not work properly with hard links in gofer and overlay fs. + SKIP_IF(IsRunningOnGvisor() && + !ASSERT_NO_ERRNO_AND_VALUE(IsTmpfs(GetAbsoluteTestTmpdir()))); + const TempPath dir1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const TempPath dir2 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir1.path())); std::string link_path = NewTempAbsPathInDir(dir2.path()); - const int rc = link(file.path().c_str(), link_path.c_str()); - // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. - SKIP_IF(IsRunningOnGvisor() && rc != 0 && - (errno == EPERM || errno == ENOENT)); - ASSERT_THAT(rc, SyscallSucceeds()); + ASSERT_THAT(link(file.path().c_str(), link_path.c_str()), SyscallSucceeds()); const FileDescriptor inotify_fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); @@ -1825,13 +1827,18 @@ TEST(Inotify, LinkOnOtherParent) { // Perform various actions on the link outside of dir1, which should trigger // no inotify events. - const FileDescriptor fd = + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(link_path.c_str(), O_RDWR)); int val = 0; ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); ASSERT_THAT(ftruncate(fd.get(), 12345), SyscallSucceeds()); + + // Close before unlinking; some filesystems cannot restore an open fd on an + // unlinked file. + fd.reset(); ASSERT_THAT(unlink(link_path.c_str()), SyscallSucceeds()); + const std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); EXPECT_THAT(events, Are({})); @@ -2055,21 +2062,21 @@ TEST(Inotify, ExcludeUnlinkDirectory_NoRandomSave) { // We need to disable S/R because there are filesystems where we cannot re-open // fds to an unlinked file across S/R, e.g. gofer-backed filesytems. TEST(Inotify, ExcludeUnlinkMultipleChildren_NoRandomSave) { - const DisableSave ds; + // Inotify does not work properly with hard links in gofer and overlay fs. + SKIP_IF(IsRunningOnGvisor() && + !ASSERT_NO_ERRNO_AND_VALUE(IsTmpfs(GetAbsoluteTestTmpdir()))); // TODO(gvisor.dev/issue/1624): This test fails on VFS1. SKIP_IF(IsRunningWithVFS1()); + const DisableSave ds; + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); std::string path1 = file.path(); std::string path2 = NewTempAbsPathInDir(dir.path()); + ASSERT_THAT(link(path1.c_str(), path2.c_str()), SyscallSucceeds()); - const int rc = link(path1.c_str(), path2.c_str()); - // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. - SKIP_IF(IsRunningOnGvisor() && rc != 0 && - (errno == EPERM || errno == ENOENT)); - ASSERT_THAT(rc, SyscallSucceeds()); const FileDescriptor fd1 = ASSERT_NO_ERRNO_AND_VALUE(Open(path1.c_str(), O_RDWR)); const FileDescriptor fd2 = @@ -2101,6 +2108,15 @@ TEST(Inotify, ExcludeUnlinkMultipleChildren_NoRandomSave) { // We need to disable S/R because there are filesystems where we cannot re-open // fds to an unlinked file across S/R, e.g. gofer-backed filesytems. TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { + // TODO(gvisor.dev/issue/1624): Fails on VFS1. + SKIP_IF(IsRunningWithVFS1()); + + // NOTE(gvisor.dev/issue/3654): In the gofer filesystem, we do not allow + // setting attributes through an fd if the file at the open path has been + // deleted. + SKIP_IF(IsRunningOnGvisor() && + !ASSERT_NO_ERRNO_AND_VALUE(IsTmpfs(GetAbsoluteTestTmpdir()))); + const DisableSave ds; const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); @@ -2110,18 +2126,6 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path().c_str(), O_RDWR)); - // NOTE(b/157163751): Create another link before unlinking. This is needed for - // the gofer filesystem in gVisor, where open fds will not work once the link - // count hits zero. In VFS2, we end up skipping the gofer test anyway, because - // hard links are not supported for gofer fs. - if (IsRunningOnGvisor()) { - std::string link_path = NewTempAbsPath(); - const int rc = link(file.path().c_str(), link_path.c_str()); - // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. - SKIP_IF(rc != 0 && (errno == EPERM || errno == ENOENT)); - ASSERT_THAT(rc, SyscallSucceeds()); - } - const FileDescriptor inotify_fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( diff --git a/test/syscalls/linux/kcov.cc b/test/syscalls/linux/kcov.cc index f3c30444e..6afcb4e75 100644 --- a/test/syscalls/linux/kcov.cc +++ b/test/syscalls/linux/kcov.cc @@ -36,12 +36,13 @@ TEST(KcovTest, Kcov) { constexpr int kSize = 4096; constexpr int KCOV_INIT_TRACE = 0x80086301; constexpr int KCOV_ENABLE = 0x6364; + constexpr int KCOV_DISABLE = 0x6365; int fd; ASSERT_THAT(fd = open("/sys/kernel/debug/kcov", O_RDWR), AnyOf(SyscallSucceeds(), SyscallFailsWithErrno(ENOENT))); - // Kcov not enabled. + // Kcov not available. SKIP_IF(errno == ENOENT); ASSERT_THAT(ioctl(fd, KCOV_INIT_TRACE, kSize), SyscallSucceeds()); @@ -62,6 +63,8 @@ TEST(KcovTest, Kcov) { // Verify that PCs are in the standard kernel range. EXPECT_GT(area[i], 0xffffffff7fffffffL); } + + ASSERT_THAT(ioctl(fd, KCOV_DISABLE, 0), SyscallSucceeds()); } } // namespace diff --git a/test/syscalls/linux/mknod.cc b/test/syscalls/linux/mknod.cc index 89e4564e8..ae65d366b 100644 --- a/test/syscalls/linux/mknod.cc +++ b/test/syscalls/linux/mknod.cc @@ -105,11 +105,13 @@ TEST(MknodTest, UnimplementedTypesReturnError) { } TEST(MknodTest, Socket) { + SKIP_IF(IsRunningOnGvisor() && IsRunningWithVFS1()); + ASSERT_THAT(chdir(GetAbsoluteTestTmpdir().c_str()), SyscallSucceeds()); - SKIP_IF(IsRunningOnGvisor() && IsRunningWithVFS1()); + auto filename = NewTempRelPath(); - ASSERT_THAT(mknod("./file0", S_IFSOCK | S_IRUSR | S_IWUSR, 0), + ASSERT_THAT(mknod(filename.c_str(), S_IFSOCK | S_IRUSR | S_IWUSR, 0), SyscallSucceeds()); int sk; @@ -117,9 +119,10 @@ TEST(MknodTest, Socket) { FileDescriptor fd(sk); struct sockaddr_un addr = {.sun_family = AF_UNIX}; - absl::SNPrintF(addr.sun_path, sizeof(addr.sun_path), "./file0"); + absl::SNPrintF(addr.sun_path, sizeof(addr.sun_path), "%s", filename.c_str()); ASSERT_THAT(connect(sk, (struct sockaddr *)&addr, sizeof(addr)), SyscallFailsWithErrno(ECONNREFUSED)); + ASSERT_THAT(unlink(filename.c_str()), SyscallSucceeds()); } TEST(MknodTest, Fifo) { diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index c1488b06b..6e3c536c0 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -47,6 +47,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/container/node_hash_set.h" #include "absl/strings/ascii.h" #include "absl/strings/match.h" #include "absl/strings/numbers.h" @@ -721,8 +722,8 @@ static void CheckFdDirGetdentsDuplicates(const std::string& path) { EXPECT_GE(newfd, 1024); auto fd_closer = Cleanup([newfd]() { close(newfd); }); auto fd_files = ASSERT_NO_ERRNO_AND_VALUE(ListDir(path.c_str(), false)); - std::unordered_set<std::string> fd_files_dedup(fd_files.begin(), - fd_files.end()); + absl::node_hash_set<std::string> fd_files_dedup(fd_files.begin(), + fd_files.end()); EXPECT_EQ(fd_files.size(), fd_files_dedup.size()); } diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index ab731db1d..e0981e28a 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1643,6 +1643,36 @@ TEST_P(SimpleTcpSocketTest, GetSocketDetachFilter) { SyscallFailsWithErrno(ENOPROTOOPT)); } +TEST_P(SimpleTcpSocketTest, CloseNonConnectedLingerOption) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + constexpr int kLingerTimeout = 10; // Seconds. + + // Set the SO_LINGER option. + struct linger sl = { + .l_onoff = 1, + .l_linger = kLingerTimeout, + }; + ASSERT_THAT(setsockopt(s.get(), SOL_SOCKET, SO_LINGER, &sl, sizeof(sl)), + SyscallSucceeds()); + + struct pollfd poll_fd = { + .fd = s.get(), + .events = POLLHUP, + }; + constexpr int kPollTimeoutMs = 0; + ASSERT_THAT(RetryEINTR(poll)(&poll_fd, 1, kPollTimeoutMs), + SyscallSucceedsWithValue(1)); + + auto const start_time = absl::Now(); + EXPECT_THAT(close(s.release()), SyscallSucceeds()); + auto const end_time = absl::Now(); + + // Close() should not linger and return immediately. + ASSERT_LT((end_time - start_time), absl::Seconds(kLingerTimeout)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); diff --git a/tools/bazel.mk b/tools/bazel.mk index 5cc1cdea2..4235c36ca 100644 --- a/tools/bazel.mk +++ b/tools/bazel.mk @@ -131,7 +131,7 @@ bazel-server-start: bazel-image ## Starts the bazel server. --workdir "$(CURDIR)" \ $(FULL_DOCKER_RUN_OPTIONS) \ $(BUILDER_IMAGE) \ - sh -c "tail -f --pid=\$$($(BAZEL) info server_pid)" + sh -c "tail -f --pid=\$$($(BAZEL) info server_pid) /dev/null" .PHONY: bazel-server-start bazel-shutdown: ## Shuts down a running bazel server. |