diff options
Diffstat (limited to 'pkg/sentry')
99 files changed, 1190 insertions, 420 deletions
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/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index 77c2c5c0e..b8b2281a8 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -50,6 +50,7 @@ go_library( "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/header", + "//pkg/tcpip/network/ipv4", "//pkg/usermem", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index 8615b60f0..e555672ad 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -54,7 +55,7 @@ type tcpMemInode struct { // size stores the tcp buffer size during save, and sets the buffer // size in netstack in restore. We must save/restore this here, since - // netstack itself is stateless. + // a netstack instance is created on restore. size inet.TCPBufferSize // mu protects against concurrent reads/writes to files based on this @@ -258,6 +259,9 @@ func (f *tcpSackFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSeque if src.NumBytes() == 0 { return 0, nil } + + // Only consider size of one memory page for input for performance reasons. + // We are only reading if it's zero or not anyway. src = src.TakeFirst(usermem.PageSize - 1) var v int32 @@ -383,11 +387,125 @@ func (p *proc) newSysNetCore(ctx context.Context, msrc *fs.MountSource, s inet.S return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) } +// ipForwarding implements fs.InodeOperations. +// +// ipForwarding is used to enable/disable packet forwarding of netstack. +// +// +stateify savable +type ipForwarding struct { + fsutil.SimpleFileInode + + stack inet.Stack `state:"wait"` + + // enabled stores the IPv4 forwarding state on save. + // We must save/restore this here, since a netstack instance + // is created on restore. + enabled *bool +} + +func newIPForwardingInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { + ipf := &ipForwarding{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + stack: s, + } + sattr := fs.StableAttr{ + DeviceID: device.ProcDevice.DeviceID(), + InodeID: device.ProcDevice.NextIno(), + BlockSize: usermem.PageSize, + Type: fs.SpecialFile, + } + return fs.NewInode(ctx, ipf, msrc, sattr) +} + +// Truncate implements fs.InodeOperations.Truncate. Truncate is called when +// O_TRUNC is specified for any kind of existing Dirent but is not called via +// (f)truncate for proc files. +func (*ipForwarding) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + +// +stateify savable +type ipForwardingFile struct { + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSplice `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileUseInodeUnstableAttr `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + + ipf *ipForwarding + + stack inet.Stack `state:"wait"` +} + +// GetFile implements fs.InodeOperations.GetFile. +func (ipf *ipForwarding) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true + return fs.NewFile(ctx, dirent, flags, &ipForwardingFile{ + stack: ipf.stack, + ipf: ipf, + }), nil +} + +// Read implements fs.FileOperations.Read. +func (f *ipForwardingFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + if offset != 0 { + return 0, io.EOF + } + + if f.ipf.enabled == nil { + enabled := f.stack.Forwarding(ipv4.ProtocolNumber) + f.ipf.enabled = &enabled + } + + val := "0\n" + if *f.ipf.enabled { + // Technically, this is not quite compatible with Linux. Linux + // stores these as an integer, so if you write "2" into + // ip_forward, you should get 2 back. + val = "1\n" + } + n, err := dst.CopyOut(ctx, []byte(val)) + return int64(n), err +} + +// Write implements fs.FileOperations.Write. +// +// Offset is ignored, multiple writes are not supported. +func (f *ipForwardingFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { + if src.NumBytes() == 0 { + return 0, nil + } + + // Only consider size of one memory page for input for performance reasons. + // We are only reading if it's zero or not anyway. + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return n, err + } + if f.ipf.enabled == nil { + f.ipf.enabled = new(bool) + } + *f.ipf.enabled = v != 0 + return n, f.stack.SetForwarding(ipv4.ProtocolNumber, *f.ipf.enabled) +} + func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { contents := map[string]*fs.Inode{ // Add tcp_sack. "tcp_sack": newTCPSackInode(ctx, msrc, s), + // Add ip_forward. + "ip_forward": newIPForwardingInode(ctx, msrc, s), + // The following files are simple stubs until they are // implemented in netstack, most of these files are // configuration related. We use the value closest to the diff --git a/pkg/sentry/fs/proc/sys_net_state.go b/pkg/sentry/fs/proc/sys_net_state.go index 6eba709c6..4cb4741af 100644 --- a/pkg/sentry/fs/proc/sys_net_state.go +++ b/pkg/sentry/fs/proc/sys_net_state.go @@ -14,7 +14,11 @@ package proc -import "fmt" +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" +) // beforeSave is invoked by stateify. func (t *tcpMemInode) beforeSave() { @@ -40,3 +44,12 @@ func (s *tcpSack) afterLoad() { } } } + +// afterLoad is invoked by stateify. +func (ipf *ipForwarding) afterLoad() { + if ipf.enabled != nil { + if err := ipf.stack.SetForwarding(ipv4.ProtocolNumber, *ipf.enabled); err != nil { + panic(fmt.Sprintf("failed to set IPv4 forwarding [%v]: %v", *ipf.enabled, err)) + } + } +} diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go index 355e83d47..6ef5738e7 100644 --- a/pkg/sentry/fs/proc/sys_net_test.go +++ b/pkg/sentry/fs/proc/sys_net_test.go @@ -123,3 +123,76 @@ func TestConfigureRecvBufferSize(t *testing.T) { } } } + +// TestIPForwarding tests the implementation of +// /proc/sys/net/ipv4/ip_forwarding +func TestIPForwarding(t *testing.T) { + ctx := context.Background() + s := inet.NewTestStack() + + var cases = []struct { + comment string + initial bool + str string + final bool + }{ + { + comment: `Forwarding is disabled; write 1 and enable forwarding`, + initial: false, + str: "1", + final: true, + }, + { + comment: `Forwarding is disabled; write 0 and disable forwarding`, + initial: false, + str: "0", + final: false, + }, + { + comment: `Forwarding is enabled; write 1 and enable forwarding`, + initial: true, + str: "1", + final: true, + }, + { + comment: `Forwarding is enabled; write 0 and disable forwarding`, + initial: true, + str: "0", + final: false, + }, + { + comment: `Forwarding is disabled; write 2404 and enable forwarding`, + initial: false, + str: "2404", + final: true, + }, + { + comment: `Forwarding is enabled; write 2404 and enable forwarding`, + initial: true, + str: "2404", + final: true, + }, + } + for _, c := range cases { + t.Run(c.comment, func(t *testing.T) { + s.IPForwarding = c.initial + ipf := &ipForwarding{stack: s} + file := &ipForwardingFile{ + stack: s, + ipf: ipf, + } + + // Write the values. + src := usermem.BytesIOSequence([]byte(c.str)) + if n, err := file.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { + t.Errorf("file.Write(ctx, nil, %q, 0) = (%d, %v); want (%d, nil)", c.str, n, err, len(c.str)) + } + + // Read the values from the stack and check them. + if got, want := s.IPForwarding, c.final; got != want { + t.Errorf("s.IPForwarding incorrect; got: %v, want: %v", got, want) + } + + }) + } +} diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index f0f2e0be7..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 @@ -187,8 +192,8 @@ func (i *rootInode) masterClose(t *Terminal) { } // Open implements kernfs.Inode.Open. -func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -198,7 +203,7 @@ func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.D } // Lookup implements kernfs.Inode.Lookup. -func (i *rootInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (i *rootInode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, error) { idx, err := strconv.ParseUint(name, 10, 32) if err != nil { return nil, syserror.ENOENT @@ -207,7 +212,7 @@ func (i *rootInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error defer i.mu.Unlock() if si, ok := i.replicas[uint32(idx)]; ok { si.dentry.IncRef() - return si.dentry.VFSDentry(), nil + return &si.dentry, nil } return nil, syserror.ENOENT @@ -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 83d790b38..bfcaf60bd 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 @@ -50,7 +52,7 @@ type masterInode struct { var _ kernfs.Inode = (*masterInode)(nil) // Open implements kernfs.Inode.Open. -func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { t, err := mi.root.allocateTerminal(rp.Credentials()) if err != nil { return nil, err @@ -62,7 +64,7 @@ func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vf t: t, } fd.LockFD.Init(&mi.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { mi.DecRef(ctx) return nil, err } @@ -89,6 +91,7 @@ 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 diff --git a/pkg/sentry/fsimpl/devpts/replica.go b/pkg/sentry/fsimpl/devpts/replica.go index 58f6c1d3a..9638eb6c5 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 @@ -52,13 +54,13 @@ type replicaInode struct { var _ kernfs.Inode = (*replicaInode)(nil) // Open implements kernfs.Inode.Open. -func (si *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (si *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { si.IncRef() fd := &replicaFileDescription{ inode: si, } fd.LockFD.Init(&si.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { si.DecRef(ctx) return nil, err } @@ -95,6 +97,7 @@ func (si *replicaInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, cred return si.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) } +// +stateify savable type replicaFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl 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 b3573f80d..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 @@ -280,6 +287,7 @@ func (fs *filesystem) newRootInode(creds *auth.Credentials, mode linux.FileMode) i := &inode{fs: fs} i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, 1, linux.ModeDirectory|0755) i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + i.EnableLeakCheck() i.dentry.Init(i) i.nodeID = 1 @@ -299,7 +307,7 @@ func (fs *filesystem) newInode(nodeID uint64, attr linux.FUSEAttr) *kernfs.Dentr } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { isDir := i.InodeAttrs.Mode().IsDir() // return error if specified to open directory but inode is not a directory. if !isDir && opts.Mode.IsDir() { @@ -395,14 +403,14 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr i.attributeTime = 0 } - if err := fd.vfsfd.Init(fdImpl, opts.Flags, rp.Mount(), vfsd, fdOptions); err != nil { + if err := fd.vfsfd.Init(fdImpl, opts.Flags, rp.Mount(), d.VFSDentry(), fdOptions); err != nil { return nil, err } return &fd.vfsfd, nil } // Lookup implements kernfs.Inode.Lookup. -func (i *inode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (i *inode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, error) { in := linux.FUSELookupIn{Name: name} return i.newEntry(ctx, name, 0, linux.FUSE_LOOKUP, &in) } @@ -418,7 +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) @@ -436,7 +444,7 @@ func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) } // 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), @@ -449,7 +457,7 @@ func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) } // 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, @@ -458,7 +466,7 @@ func (i *inode) NewSymlink(ctx context.Context, name, target string) (*vfs.Dentr } // 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) @@ -481,7 +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), @@ -493,7 +501,7 @@ func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) } // 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) @@ -511,17 +519,12 @@ 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. // Shared by FUSE_MKNOD, FUSE_MKDIR, FUSE_SYMLINK, FUSE_LINK and FUSE_LOOKUP. -func (i *inode) newEntry(ctx context.Context, name string, fileType linux.FileMode, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*vfs.Dentry, error) { +func (i *inode) newEntry(ctx context.Context, name string, fileType linux.FileMode, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*kernfs.Dentry, error) { kernelTask := kernel.TaskFromContext(ctx) if kernelTask == nil { log.Warningf("fusefs.Inode.newEntry: couldn't get kernel task from context", i.nodeID) @@ -546,12 +549,7 @@ 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.VFSDentry(), nil + return child, nil } // Getlink implements kernfs.Inode.Getlink. 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/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 dc960e5bf..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 } @@ -246,11 +248,12 @@ func (fd *specialFileFD) pwrite(ctx context.Context, src usermem.IOSequence, off d.touchCMtime() } buf := make([]byte, src.NumBytes()) - // Don't do partial writes if we get a partial read from src. - if _, err := src.CopyIn(ctx, buf); err != nil { - return 0, offset, err + copied, copyErr := src.CopyIn(ctx, buf) + if copied == 0 && copyErr != nil { + // Only return the error if we didn't get any data. + return 0, offset, copyErr } - n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf[:copied])), uint64(offset)) if err == syserror.EAGAIN { err = syserror.ErrWouldBlock } @@ -267,7 +270,10 @@ func (fd *specialFileFD) pwrite(ctx context.Context, src usermem.IOSequence, off atomic.StoreUint64(&d.size, uint64(offset)) } } - return int64(n), offset, err + if err != nil { + return int64(n), offset, err + } + return int64(n), offset, copyErr } // Write implements vfs.FileDescriptionImpl.Write. diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index db8536f26..ffe4ddb32 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -58,7 +58,7 @@ func newInode(fs *filesystem, hostFD int, fileType linux.FileMode, isTTY bool) ( canMap: fileType == linux.S_IFREG, } i.pf.inode = i - i.refs.EnableLeakCheck() + i.EnableLeakCheck() // Non-seekable files can't be memory mapped, assert this. if !i.seekable && i.canMap { @@ -126,7 +126,7 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) // For simplicity, fileDescription.offset is set to 0. Technically, we // should only set to 0 on files that are not seekable (sockets, pipes, // etc.), and use the offset from the host fd otherwise when importing. - return i.open(ctx, d.VFSDentry(), mnt, flags) + return i.open(ctx, d, mnt, flags) } // ImportFD sets up and returns a vfs.FileDescription from a donated fd. @@ -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 @@ -193,7 +199,7 @@ type inode struct { locks vfs.FileLocks // When the reference count reaches zero, the host fd is closed. - refs inodeRefs + inodeRefs // hostFD contains the host fd that this file was originally created from, // which must be available at time of restore. @@ -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. @@ -435,19 +441,9 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre return nil } -// IncRef implements kernfs.Inode.IncRef. -func (i *inode) IncRef() { - i.refs.IncRef() -} - -// TryIncRef implements kernfs.Inode.TryIncRef. -func (i *inode) TryIncRef() bool { - return i.refs.TryIncRef() -} - // DecRef implements kernfs.Inode.DecRef. func (i *inode) DecRef(ctx context.Context) { - i.refs.DecRef(func() { + i.inodeRefs.DecRef(func() { if i.wouldBlock { fdnotifier.RemoveFD(int32(i.hostFD)) } @@ -458,15 +454,15 @@ func (i *inode) DecRef(ctx context.Context) { } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { // Once created, we cannot re-open a socket fd through /proc/[pid]/fd/. if i.Mode().FileType() == linux.S_IFSOCK { return nil, syserror.ENXIO } - return i.open(ctx, vfsd, rp.Mount(), opts.Flags) + return i.open(ctx, d, rp.Mount(), opts.Flags) } -func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, error) { +func (i *inode) open(ctx context.Context, d *kernfs.Dentry, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, error) { var s syscall.Stat_t if err := syscall.Fstat(i.hostFD, &s); err != nil { return nil, err @@ -490,7 +486,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u return nil, err } // Currently, we only allow Unix sockets to be imported. - return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d, &i.locks) + return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d.VFSDentry(), &i.locks) case syscall.S_IFREG, syscall.S_IFIFO, syscall.S_IFCHR: if i.isTTY { @@ -500,7 +496,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd - if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := vfsfd.Init(fd, flags, mnt, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil @@ -509,7 +505,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u fd := &fileDescription{inode: i} fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd - if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := vfsfd.Init(fd, flags, mnt, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil @@ -521,6 +517,8 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -535,7 +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 1ee089620..b929118b1 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -56,9 +56,9 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, devMajor, devMinor uint } // Open implements Inode.Open. -func (f *DynamicBytesFile) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +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(), vfsd, 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 6518ff5cd..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 @@ -76,12 +82,12 @@ type GenericDirectoryFD struct { // NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its // dentry. -func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions, fdOpts GenericDirectoryFDOptions) (*GenericDirectoryFD, error) { +func NewGenericDirectoryFD(m *vfs.Mount, d *Dentry, children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions, fdOpts GenericDirectoryFDOptions) (*GenericDirectoryFD, error) { fd := &GenericDirectoryFD{} if err := fd.Init(children, locks, opts, fdOpts); err != nil { return nil, err } - if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, m, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return fd, nil @@ -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 89ed265dc..c2c62341d 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) @@ -122,25 +121,20 @@ 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 } } if child == nil { - // Dentry isn't cached; it either doesn't exist or failed - // revalidation. Attempt to resolve it via Lookup. - // - // FIXME(gvisor.dev/issue/1193): Inode.Lookup() should return - // *(kernfs.)Dentry, not *vfs.Dentry, since (kernfs.)Filesystem assumes - // that all dentries in the filesystem are (kernfs.)Dentry and performs - // vfs.DentryImpl casts accordingly. - childVFSD, err := parent.inode.Lookup(ctx, name) + // Dentry isn't cached; it either doesn't exist or failed revalidation. + // Attempt to resolve it via Lookup. + c, err := parent.inode.Lookup(ctx, name) if err != nil { return nil, err } // Reference on childVFSD dropped by a corresponding Valid. - child = childVFSD.Impl().(*Dentry) - parent.InsertChildLocked(name, child) + parent.InsertChildLocked(name, c) + child = c } return child, nil } @@ -153,20 +147,19 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // Preconditions: Filesystem.mu must be locked for at least reading. // // Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { - vfsd := rp.Start() +func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*Dentry, error) { + 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, nil, err + return nil, err } } - d := vfsd.Impl().(*Dentry) if rp.MustBeDir() && !d.isDir() { - return nil, nil, syserror.ENOTDIR + return nil, syserror.ENOTDIR } - return vfsd, d.inode, nil + return d, nil } // walkParentDirLocked resolves all but the last path component of rp to an @@ -181,20 +174,19 @@ 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 @@ -202,10 +194,9 @@ 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() @@ -216,10 +207,10 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v 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 @@ -228,8 +219,8 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v // checkDeleteLocked checks that the file represented by vfsd may be deleted. // // Preconditions: Filesystem.mu must be locked for at least reading. -func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { - parent := vfsd.Impl().(*Dentry).parent +func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry) error { + parent := d.parent if parent == nil { return syserror.EBUSY } @@ -258,11 +249,11 @@ func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) if err != nil { return err } - return inode.CheckPermissions(ctx, creds, ats) + return d.inode.CheckPermissions(ctx, creds, ats) } // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. @@ -270,20 +261,20 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op fs.mu.RLock() defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) if err != nil { return nil, err } if opts.CheckSearchable { - d := vfsd.Impl().(*Dentry) if !d.isDir() { return nil, syserror.ENOTDIR } - if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } } + vfsd := d.VFSDentry() vfsd.IncRef() // Ownership transferred to caller. return vfsd, nil } @@ -293,12 +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. @@ -308,12 +299,12 @@ 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) + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -330,11 +321,11 @@ 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.InsertChild(pc, child) return nil } @@ -345,12 +336,12 @@ 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) + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -358,14 +349,14 @@ 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.InsertChild(pc, child) return nil } @@ -376,12 +367,12 @@ 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) + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -389,11 +380,11 @@ 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.InsertChild(pc, newD) return nil } @@ -409,28 +400,27 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // Do not create new file. if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) if err != nil { fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) return nil, err } - if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) return nil, err } - inode.IncRef() - defer inode.DecRef(ctx) + d.inode.IncRef() + defer d.inode.DecRef(ctx) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) - return inode.Open(ctx, rp, vfsd, opts) + return d.inode.Open(ctx, rp, d, opts) } // May create new file. mustCreate := opts.Flags&linux.O_EXCL != 0 - vfsd := rp.Start() - inode := vfsd.Impl().(*Dentry).inode + d := rp.Start().Impl().(*Dentry) fs.mu.Lock() unlocked := false unlock := func() { @@ -447,22 +437,22 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if mustCreate { return nil, syserror.EEXIST } - if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - inode.IncRef() - defer inode.DecRef(ctx) + d.inode.IncRef() + defer d.inode.DecRef(ctx) unlock() - return inode.Open(ctx, rp, vfsd, opts) + 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. @@ -477,10 +467,10 @@ 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 { @@ -488,16 +478,15 @@ 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) + parent.InsertChild(pc, child) child.inode.IncRef() defer child.inode.DecRef(ctx) unlock() - return child.inode.Open(ctx, rp, childVFSD, opts) + return child.inode.Open(ctx, rp, child, opts) } if err != nil { return nil, err @@ -506,7 +495,6 @@ 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 { @@ -533,22 +521,22 @@ afterTrailingSymlink: child.inode.IncRef() defer child.inode.DecRef(ctx) unlock() - return child.inode.Open(ctx, rp, &child.vfsd, opts) + return child.inode.Open(ctx, rp, child, opts) } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { fs.mu.RLock() - d, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return "", err } - if !d.Impl().(*Dentry).isSymlink() { + if !d.isSymlink() { return "", syserror.EINVAL } - return inode.Readlink(ctx, rp.Mount()) + return d.inode.Readlink(ctx, rp.Mount()) } // RenameAt implements vfs.FilesystemImpl.RenameAt. @@ -565,11 +553,10 @@ 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 @@ -587,16 +574,15 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if err != nil { return err } - srcVFSD := &src.vfsd // Can we remove the src dentry? - if err := checkDeleteLocked(ctx, rp, srcVFSD); err != nil { + if err := checkDeleteLocked(ctx, rp, src); err != nil { return err } // 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. @@ -607,14 +593,14 @@ 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) @@ -630,17 +616,18 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa defer dstDir.dirMu.Unlock() } + srcVFSD := src.VFSDentry() 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 @@ -649,7 +636,11 @@ 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 } @@ -658,7 +649,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err @@ -667,14 +658,13 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, d); err != nil { return err } - d := vfsd.Impl().(*Dentry) if !d.isDir() { return syserror.ENOTDIR } - if inode.HasChildren() { + if d.inode.HasChildren() { return syserror.ENOTEMPTY } virtfs := rp.VirtualFilesystem() @@ -684,11 +674,12 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef(ctx) + vfsd := d.VFSDentry() if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.RmDir(ctx, d.name, vfsd); err != nil { + if err := parentDentry.inode.RmDir(ctx, d.name, d); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } @@ -699,7 +690,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -708,31 +699,31 @@ func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if opts.Stat.Mask == 0 { return nil } - return inode.SetStat(ctx, fs.VFSFilesystem(), rp.Credentials(), opts) + return d.inode.SetStat(ctx, fs.VFSFilesystem(), rp.Credentials(), opts) } // StatAt implements vfs.FilesystemImpl.StatAt. func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statx{}, err } - return inode.Stat(ctx, fs.VFSFilesystem(), opts) + return d.inode.Stat(ctx, fs.VFSFilesystem(), opts) } // StatFSAt implements vfs.FilesystemImpl.StatFSAt. func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statfs{}, err } - return inode.StatFS(ctx, fs.VFSFilesystem()) + return d.inode.StatFS(ctx, fs.VFSFilesystem()) } // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. @@ -742,12 +733,12 @@ 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) + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -755,11 +746,11 @@ 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.InsertChild(pc, child) return nil } @@ -768,7 +759,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() - vfsd, _, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err @@ -777,10 +768,9 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, d); err != nil { return err } - d := vfsd.Impl().(*Dentry) if d.isDir() { return syserror.EISDIR } @@ -790,10 +780,11 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef(ctx) + vfsd := d.VFSDentry() if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.Unlink(ctx, d.name, vfsd); err != nil { + if err := parentDentry.inode.Unlink(ctx, d.name, d); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } @@ -804,13 +795,13 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // BoundEndpointAt implements vfs.FilesystemImpl.BoundEndpointAt. func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return nil, err } - if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } return nil, syserror.ECONNREFUSED @@ -819,7 +810,7 @@ func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath // ListXattrAt implements vfs.FilesystemImpl.ListXattrAt. func (fs *Filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -832,7 +823,7 @@ func (fs *Filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, si // GetXattrAt implements vfs.FilesystemImpl.GetXattrAt. func (fs *Filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetXattrOptions) (string, error) { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -845,7 +836,7 @@ func (fs *Filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetXattrAt implements vfs.FilesystemImpl.SetXattrAt. func (fs *Filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -858,7 +849,7 @@ func (fs *Filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // RemoveXattrAt implements vfs.FilesystemImpl.RemoveXattrAt. func (fs *Filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 6ee353ace..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 { } @@ -90,47 +96,47 @@ 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") } // Lookup implements Inode.Lookup. -func (InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (InodeNotDirectory) Lookup(ctx context.Context, name string) (*Dentry, error) { panic("Lookup called on non-directory inode") } @@ -149,10 +155,12 @@ 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. -func (InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*Dentry, error) { return nil, syserror.ENOENT } @@ -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{} } @@ -542,12 +560,14 @@ func (o *OrderedChildren) nthLocked(i int64) *slot { } // InodeSymlink partially implements Inode interface for symlinks. +// +// +stateify savable type InodeSymlink struct { InodeNotDirectory } // Open implements Inode.Open. -func (InodeSymlink) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (InodeSymlink) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ELOOP } @@ -574,6 +594,7 @@ var _ Inode = (*StaticDirectory)(nil) func NewStaticDir(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode, children map[string]*Dentry, fdOpts GenericDirectoryFDOptions) *Dentry { inode := &StaticDirectory{} inode.Init(creds, devMajor, devMinor, ino, perm, fdOpts) + inode.EnableLeakCheck() dentry := &Dentry{} dentry.Init(inode) @@ -595,8 +616,8 @@ func (s *StaticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint3 } // Open implements kernfs.Inode.Open. -func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &s.OrderedChildren, &s.locks, &opts, s.fdOpts) +func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := NewGenericDirectoryFD(rp.Mount(), d, &s.OrderedChildren, &s.locks, &opts, s.fdOpts) if err != nil { return nil, err } @@ -614,6 +635,8 @@ func (s *StaticDirectory) DecRef(context.Context) { } // AlwaysValid partially implements kernfs.inodeDynamicLookup. +// +// +stateify savable type AlwaysValid struct{} // Valid implements kernfs.inodeDynamicLookup.Valid. @@ -623,6 +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 163f26ceb..c4e914530 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -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 +177,7 @@ type Dentry struct { name string // dirMu protects children and the names of child Dentries. - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` children map[string]*Dentry inode Inode @@ -273,7 +277,7 @@ 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 +287,7 @@ 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 +295,7 @@ 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) @@ -350,7 +354,7 @@ type Inode interface { // // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing // the inode on which Open() is being called. - Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) + Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) // StatFS returns filesystem statistics for the client filesystem. This // corresponds to vfs.FilesystemImpl.StatFSAt. If the client filesystem @@ -400,30 +404,30 @@ 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 +435,7 @@ 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 { @@ -449,7 +453,7 @@ type inodeDynamicLookup interface { // // Lookup returns the child with an extra reference and the caller owns this // reference. - Lookup(ctx context.Context, name string) (*vfs.Dentry, error) + Lookup(ctx context.Context, name string) (*Dentry, error) // Valid should return true if this inode is still valid, or needs to // be resolved again by a call to Lookup. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 09806a3f2..e413242dc 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -121,8 +121,8 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod return &dir.dentry } -func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, kd *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), kd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -162,8 +162,8 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte return &dir.dentry } -func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, kd *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), kd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -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 01ba72fa8..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) { @@ -60,8 +60,8 @@ func (dir *syntheticDirectory) Init(creds *auth.Credentials, devMajor, devMinor } // Open implements Inode.Open. -func (dir *syntheticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &dir.OrderedChildren, &dir.locks, &opts, GenericDirectoryFDOptions{}) +func (dir *syntheticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := NewGenericDirectoryFD(rp.Mount(), d, &dir.OrderedChildren, &dir.locks, &opts, GenericDirectoryFDOptions{}) if err != nil { return nil, err } @@ -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..aea599460 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 } diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index 6e04705c7..50b4a50f1 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 diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index d0d26185e..b03fa9121 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 @@ -652,6 +660,8 @@ 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 diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 7053ad6db..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 @@ -144,8 +148,8 @@ func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth. } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags, &i.locks) +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + return i.pipe.Open(ctx, rp.Mount(), d.VFSDentry(), opts.Flags, &i.locks) } // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index a45b44440..2e086e34c 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -100,6 +100,7 @@ go_library( "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/header", + "//pkg/tcpip/network/ipv4", "//pkg/usermem", ], ) 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 57f026040..47ecd941c 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -69,7 +69,7 @@ func (fs *filesystem) newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, } // Lookup implements kernfs.inodeDynamicLookup.Lookup. -func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (i *subtasksInode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, error) { tid, err := strconv.ParseUint(name, 10, 32) if err != nil { return nil, syserror.ENOENT @@ -82,9 +82,7 @@ func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, e if subTask.ThreadGroup() != i.task.ThreadGroup() { return nil, syserror.ENOENT } - - subTaskDentry := i.fs.newTaskInode(subTask, i.pidns, false, i.cgroupControllers) - return subTaskDentry.VFSDentry(), nil + return i.fs.newTaskInode(subTask, i.pidns, false, i.cgroupControllers), nil } // IterDirents implements kernfs.inodeDynamicLookup.IterDirents. @@ -118,6 +116,7 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb return offset, nil } +// +stateify savable type subtasksFD struct { kernfs.GenericDirectoryFD @@ -156,14 +155,14 @@ func (fd *subtasksFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) erro } // Open implements kernfs.Inode.Open. -func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &subtasksFD{task: i.task} if err := fd.Init(&i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }); err != nil { return nil, err } - if err := fd.VFSFileDescription().Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.VFSFileDescription().Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return fd.VFSFileDescription(), nil diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index e24c8a031..1f99183eb 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -107,8 +107,8 @@ func (i *taskInode) Valid(ctx context.Context) bool { } // Open implements kernfs.Inode.Open. -func (i *taskInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *taskInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { @@ -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 c492bcfa7..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 @@ -136,7 +137,7 @@ func (fs *filesystem) newFDDirInode(task *kernel.Task) *kernfs.Dentry { } // Lookup implements kernfs.inodeDynamicLookup.Lookup. -func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (i *fdDirInode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, error) { fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { return nil, syserror.ENOENT @@ -145,13 +146,12 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro if !taskFDExists(ctx, i.task, fd) { return nil, syserror.ENOENT } - taskDentry := i.fs.newFDSymlink(i.task, fd, i.fs.NextIno()) - return taskDentry.VFSDentry(), nil + return i.fs.newFDSymlink(i.task, fd, i.fs.NextIno()), nil } // Open implements kernfs.Inode.Open. -func (i *fdDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *fdDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { @@ -270,7 +270,7 @@ func (fs *filesystem) newFDInfoDirInode(task *kernel.Task) *kernfs.Dentry { } // Lookup implements kernfs.inodeDynamicLookup.Lookup. -func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, error) { fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { return nil, syserror.ENOENT @@ -283,13 +283,12 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, task: i.task, fd: fd, } - dentry := i.fs.newTaskOwnedFile(i.task, i.fs.NextIno(), 0444, data) - return dentry.VFSDentry(), nil + return i.fs.newTaskOwnedFile(i.task, i.fs.NextIno(), 0444, data), nil } // Open implements kernfs.Inode.Open. -func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 8f7e9b801..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 @@ -853,11 +856,11 @@ func (i *namespaceInode) Init(creds *auth.Credentials, devMajor, devMinor uint32 } // Open implements kernfs.Inode.Open. -func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &namespaceFD{inode: i} i.IncRef() fd.LockFD.Init(&i.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -865,6 +868,8 @@ func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd * // 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 6d60acc30..d8f5dd509 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -52,8 +52,8 @@ type tasksInode struct { // '/proc/self' and '/proc/thread-self' have custom directory offsets in // Linux. So handle them outside of OrderedChildren. - selfSymlink *vfs.Dentry - threadSelfSymlink *vfs.Dentry + selfSymlink *kernfs.Dentry + threadSelfSymlink *kernfs.Dentry // cgroupControllers is a map of controller name to directory in the // cgroup hierarchy. These controllers are immutable and will be listed @@ -81,8 +81,8 @@ func (fs *filesystem) newTasksInode(k *kernel.Kernel, pidns *kernel.PIDNamespace inode := &tasksInode{ pidns: pidns, fs: fs, - selfSymlink: fs.newSelfSymlink(root, fs.NextIno(), pidns).VFSDentry(), - threadSelfSymlink: fs.newThreadSelfSymlink(root, fs.NextIno(), pidns).VFSDentry(), + selfSymlink: fs.newSelfSymlink(root, fs.NextIno(), pidns), + threadSelfSymlink: fs.newThreadSelfSymlink(root, fs.NextIno(), pidns), cgroupControllers: cgroupControllers, } inode.InodeAttrs.Init(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) @@ -99,7 +99,7 @@ func (fs *filesystem) newTasksInode(k *kernel.Kernel, pidns *kernel.PIDNamespace } // Lookup implements kernfs.inodeDynamicLookup.Lookup. -func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (i *tasksInode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, error) { // Try to lookup a corresponding task. tid, err := strconv.ParseUint(name, 10, 64) if err != nil { @@ -118,8 +118,7 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } - taskDentry := i.fs.newTaskInode(task, i.pidns, true, i.cgroupControllers) - return taskDentry.VFSDentry(), nil + return i.fs.newTaskInode(task, i.pidns, true, i.cgroupControllers), nil } // IterDirents implements kernfs.inodeDynamicLookup.IterDirents. @@ -201,8 +200,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback } // Open implements kernfs.Inode.Open. -func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { @@ -237,6 +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 9e0966efe..3312b0418 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -27,9 +27,11 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type tcpMemDir int const ( @@ -67,6 +69,7 @@ func (fs *filesystem) newSysNetDir(root *auth.Credentials, k *kernel.Kernel) *ke "tcp_rmem": fs.newDentry(root, fs.NextIno(), 0644, &tcpMemData{stack: stack, dir: tcpRMem}), "tcp_sack": fs.newDentry(root, fs.NextIno(), 0644, &tcpSackData{stack: stack}), "tcp_wmem": fs.newDentry(root, fs.NextIno(), 0644, &tcpMemData{stack: stack, dir: tcpWMem}), + "ip_forward": fs.newDentry(root, fs.NextIno(), 0444, &ipForwarding{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the @@ -354,3 +357,63 @@ func (d *tcpMemData) writeSizeLocked(size inet.TCPBufferSize) error { panic(fmt.Sprintf("unknown tcpMemFile type: %v", d.dir)) } } + +// ipForwarding implements vfs.WritableDynamicBytesSource for +// /proc/sys/net/ipv4/ip_forwarding. +// +// +stateify savable +type ipForwarding struct { + kernfs.DynamicBytesFile + + stack inet.Stack `state:"wait"` + enabled *bool +} + +var _ vfs.WritableDynamicBytesSource = (*ipForwarding)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (ipf *ipForwarding) Generate(ctx context.Context, buf *bytes.Buffer) error { + if ipf.enabled == nil { + enabled := ipf.stack.Forwarding(ipv4.ProtocolNumber) + ipf.enabled = &enabled + } + + val := "0\n" + if *ipf.enabled { + // Technically, this is not quite compatible with Linux. Linux stores these + // as an integer, so if you write "2" into tcp_sack, you should get 2 back. + // Tough luck. + val = "1\n" + } + buf.WriteString(val) + + return nil +} + +// Write implements vfs.WritableDynamicBytesSource.Write. +func (ipf *ipForwarding) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { + if offset != 0 { + // No need to handle partial writes thus far. + return 0, syserror.EINVAL + } + if src.NumBytes() == 0 { + return 0, nil + } + + // Limit input size so as not to impact performance if input size is large. + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return 0, err + } + if ipf.enabled == nil { + ipf.enabled = new(bool) + } + *ipf.enabled = v != 0 + if err := ipf.stack.SetForwarding(ipv4.ProtocolNumber, *ipf.enabled); err != nil { + return 0, err + } + return n, nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys_test.go b/pkg/sentry/fsimpl/proc/tasks_sys_test.go index be54897bb..6cee22823 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys_test.go @@ -20,8 +20,10 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/inet" + "gvisor.dev/gvisor/pkg/usermem" ) func newIPv6TestStack() *inet.TestStack { @@ -76,3 +78,72 @@ func TestIfinet6(t *testing.T) { t.Errorf("Got n.contents() = %v, want = %v", got, want) } } + +// TestIPForwarding tests the implementation of +// /proc/sys/net/ipv4/ip_forwarding +func TestConfigureIPForwarding(t *testing.T) { + ctx := context.Background() + s := inet.NewTestStack() + + var cases = []struct { + comment string + initial bool + str string + final bool + }{ + { + comment: `Forwarding is disabled; write 1 and enable forwarding`, + initial: false, + str: "1", + final: true, + }, + { + comment: `Forwarding is disabled; write 0 and disable forwarding`, + initial: false, + str: "0", + final: false, + }, + { + comment: `Forwarding is enabled; write 1 and enable forwarding`, + initial: true, + str: "1", + final: true, + }, + { + comment: `Forwarding is enabled; write 0 and disable forwarding`, + initial: true, + str: "0", + final: false, + }, + { + comment: `Forwarding is disabled; write 2404 and enable forwarding`, + initial: false, + str: "2404", + final: true, + }, + { + comment: `Forwarding is enabled; write 2404 and enable forwarding`, + initial: true, + str: "2404", + final: true, + }, + } + for _, c := range cases { + t.Run(c.comment, func(t *testing.T) { + s.IPForwarding = c.initial + + file := &ipForwarding{stack: s, enabled: &c.initial} + + // Write the values. + src := usermem.BytesIOSequence([]byte(c.str)) + if n, err := file.Write(ctx, src, 0); n != int64(len(c.str)) || err != nil { + t.Errorf("file.Write(ctx, nil, %q, 0) = (%d, %v); want (%d, nil)", c.str, n, err, len(c.str)) + } + + // Read the values from the stack and check them. + if got, want := s.IPForwarding, c.final; got != want { + t.Errorf("s.IPForwarding incorrect; got: %v, want: %v", got, want) + } + }) + } +} 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 80b41aa9e..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 @@ -88,7 +93,7 @@ type inode struct { } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ENXIO } diff --git a/pkg/sentry/fsimpl/sys/kcov.go b/pkg/sentry/fsimpl/sys/kcov.go index 73f3d3309..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 @@ -44,7 +46,7 @@ type kcovInode struct { implStatFS } -func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { k := kernel.KernelFromContext(ctx) if k == nil { panic("KernelFromContext returned nil") @@ -54,7 +56,7 @@ func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.D kcov: k.NewKcov(), } - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{ + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, }); err != nil { @@ -63,6 +65,7 @@ func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.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 39952d2d0..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 @@ -148,8 +154,8 @@ func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.Set } // Open implements kernfs.Inode.Open. -func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, kd *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), kd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -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/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 9182df317..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 @@ -159,6 +165,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // verity, and should not be exposed or connected. mopts := &vfs.MountOptions{ GetFilesystemOptions: iopts.LowerGetFSOptions, + InternalMount: true, } mnt, err := vfsObj.MountDisconnected(ctx, creds, "", iopts.LowerName, mopts) if err != nil { @@ -258,6 +265,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -284,7 +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. @@ -428,6 +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 @@ -676,10 +687,10 @@ func (fd *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, of // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { - return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) + return fd.lowerFD.LockPOSIX(ctx, uid, t, start, length, whence, block) } // UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { - return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) + return fd.lowerFD.UnlockPOSIX(ctx, uid, start, length, whence) } diff --git a/pkg/sentry/inet/BUILD b/pkg/sentry/inet/BUILD index 07bf39fed..5bba9de0b 100644 --- a/pkg/sentry/inet/BUILD +++ b/pkg/sentry/inet/BUILD @@ -15,6 +15,7 @@ go_library( ], deps = [ "//pkg/context", + "//pkg/tcpip", "//pkg/tcpip/stack", ], ) diff --git a/pkg/sentry/inet/inet.go b/pkg/sentry/inet/inet.go index c0b4831d1..fbe6d6aa6 100644 --- a/pkg/sentry/inet/inet.go +++ b/pkg/sentry/inet/inet.go @@ -15,7 +15,10 @@ // Package inet defines semantics for IP stacks. package inet -import "gvisor.dev/gvisor/pkg/tcpip/stack" +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) // Stack represents a TCP/IP stack. type Stack interface { @@ -80,6 +83,12 @@ type Stack interface { // RestoreCleanupEndpoints adds endpoints to cleanup tracking. This is useful // for restoring a stack after a save. RestoreCleanupEndpoints([]stack.TransportEndpoint) + + // Forwarding returns if packet forwarding between NICs is enabled. + Forwarding(protocol tcpip.NetworkProtocolNumber) bool + + // SetForwarding enables or disables packet forwarding between NICs. + SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) error } // Interface contains information about a network interface. diff --git a/pkg/sentry/inet/test_stack.go b/pkg/sentry/inet/test_stack.go index 9771f01fc..1779cc6f3 100644 --- a/pkg/sentry/inet/test_stack.go +++ b/pkg/sentry/inet/test_stack.go @@ -14,7 +14,10 @@ package inet -import "gvisor.dev/gvisor/pkg/tcpip/stack" +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) // TestStack is a dummy implementation of Stack for tests. type TestStack struct { @@ -26,6 +29,7 @@ type TestStack struct { TCPSendBufSize TCPBufferSize TCPSACKFlag bool Recovery TCPLossRecovery + IPForwarding bool } // NewTestStack returns a TestStack with no network interfaces. The value of @@ -128,3 +132,14 @@ func (s *TestStack) CleanupEndpoints() []stack.TransportEndpoint { // RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. func (s *TestStack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} + +// Forwarding implements inet.Stack.Forwarding. +func (s *TestStack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { + return s.IPForwarding +} + +// SetForwarding implements inet.Stack.SetForwarding. +func (s *TestStack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) error { + s.IPForwarding = enable + return nil +} diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 22f9bb006..08bb5bd12 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -507,6 +507,10 @@ 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() @@ -533,11 +537,6 @@ 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 { @@ -556,6 +555,10 @@ 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/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index c410c96aa..67beb0ad6 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -17,6 +17,7 @@ package pipe import ( "fmt" + "io" "sync/atomic" "syscall" @@ -215,7 +216,7 @@ func (p *Pipe) readLocked(ctx context.Context, ops readOps) (int64, error) { if p.view.Size() == 0 { if !p.HasWriters() { // There are no writers, return EOF. - return 0, nil + return 0, io.EOF } return 0, syserror.ErrWouldBlock } diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 632e33452..b6ebe29d6 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -39,6 +39,9 @@ go_library( "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/syserror", + "//pkg/tcpip", + "//pkg/tcpip/network/ipv4", + "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", "//pkg/usermem", "//pkg/waiter", diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index fda3dcb35..faa61160e 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -30,6 +30,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -59,6 +62,8 @@ type Stack struct { tcpSACKEnabled bool netDevFile *os.File netSNMPFile *os.File + ipv4Forwarding bool + ipv6Forwarding bool } // NewStack returns an empty Stack containing no configuration. @@ -118,6 +123,13 @@ func (s *Stack) Configure() error { s.netSNMPFile = f } + s.ipv6Forwarding = false + if ipForwarding, err := ioutil.ReadFile("/proc/sys/net/ipv6/conf/all/forwarding"); err == nil { + s.ipv6Forwarding = strings.TrimSpace(string(ipForwarding)) != "0" + } else { + log.Warningf("Failed to read if ipv6 forwarding is enabled, setting to false") + } + return nil } @@ -468,3 +480,21 @@ func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { return nil } // RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. func (s *Stack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} + +// Forwarding implements inet.Stack.Forwarding. +func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { + switch protocol { + case ipv4.ProtocolNumber: + return s.ipv4Forwarding + case ipv6.ProtocolNumber: + return s.ipv6Forwarding + default: + log.Warningf("Forwarding(%v) failed: unsupported protocol", protocol) + return false + } +} + +// SetForwarding implements inet.Stack.SetForwarding. +func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) error { + return syserror.EACCES +} diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 36144e1eb..1028d2a6e 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -412,3 +412,24 @@ func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { func (s *Stack) RestoreCleanupEndpoints(es []stack.TransportEndpoint) { s.Stack.RestoreCleanupEndpoints(es) } + +// Forwarding implements inet.Stack.Forwarding. +func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { + switch protocol { + case ipv4.ProtocolNumber, ipv6.ProtocolNumber: + return s.Stack.Forwarding(protocol) + default: + panic(fmt.Sprintf("Forwarding(%v) failed: unsupported protocol", protocol)) + } +} + +// SetForwarding implements inet.Stack.SetForwarding. +func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) error { + switch protocol { + case ipv4.ProtocolNumber, ipv6.ProtocolNumber: + s.Stack.SetForwarding(protocol, enable) + default: + panic(fmt.Sprintf("SetForwarding(%v) failed: unsupported protocol", protocol)) + } + return nil +} diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 3688f22d2..b76c0e6fb 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 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/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index f55d74cd2..bf5c1171f 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -146,11 +147,6 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal panic("at least one end of splice must be a pipe") } - if n == 0 && err == io.EOF { - // We reached the end of the file. Eat the error and exit the loop. - err = nil - break - } if n != 0 || err != syserror.ErrWouldBlock || nonBlock { break } @@ -171,15 +167,16 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } } - if n == 0 { - return 0, nil, err + if n != 0 { + // On Linux, inotify behavior is not very consistent with splice(2). We try + // our best to emulate Linux for very basic calls to splice, where for some + // reason, events are generated for output files, but not input files. + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } - // On Linux, inotify behavior is not very consistent with splice(2). We try - // our best to emulate Linux for very basic calls to splice, where for some - // reason, events are generated for output files, but not input files. - outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + // We can only pass a single file to handleIOError, so pick inFile arbitrarily. + // This is used only for debugging purposes. + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "splice", outFile) } // Tee implements Linux syscall tee(2). @@ -251,11 +248,20 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo break } } - if n == 0 { - return 0, nil, err + + if n != 0 { + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + + // If a partial write is completed, the error is dropped. Log it here. + if err != nil && err != io.EOF && err != syserror.ErrWouldBlock { + log.Debugf("tee completed a partial write with error: %v", err) + err = nil + } } - outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + + // We can only pass a single file to handleIOError, so pick inFile arbitrarily. + // This is used only for debugging purposes. + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "tee", inFile) } // Sendfile implements linux system call sendfile(2). @@ -348,11 +354,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc for n < count { var spliceN int64 spliceN, err = outPipeFD.SpliceFromNonPipe(t, inFile, offset, count) - if spliceN == 0 && err == io.EOF { - // We reached the end of the file. Eat the error and exit the loop. - err = nil - break - } if offset != -1 { offset += spliceN } @@ -375,13 +376,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } else { readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) } - if readN == 0 && err != nil { - if err == io.EOF { - // We reached the end of the file. Eat the error before exiting the loop. - err = nil - } - break - } n += readN // Write all of the bytes that we read. This may need @@ -432,13 +426,20 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } } - if n == 0 { - return 0, nil, err + if n != 0 { + inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + + if err != nil && err != io.EOF && err != syserror.ErrWouldBlock { + // If a partial write is completed, the error is dropped. Log it here. + log.Debugf("sendfile completed a partial write with error: %v", err) + err = nil + } } - inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) - outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + // We can only pass a single file to handleIOError, so pick inFile arbitrarily. + // This is used only for debugging purposes. + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "sendfile", inFile) } // dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not 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. |