diff options
Diffstat (limited to 'pkg/sentry/fsimpl')
-rw-r--r-- | pkg/sentry/fsimpl/devpts/BUILD | 4 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/devpts.go | 34 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/devpts_test.go | 4 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/line_discipline.go | 55 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/master.go | 37 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/queue.go | 14 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/replica.go (renamed from pkg/sentry/fsimpl/devpts/slave.go) | 89 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/terminal.go | 37 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/gofer/gofer.go | 40 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/gofer/p9file.go | 7 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/host/host.go | 2 |
11 files changed, 171 insertions, 152 deletions
diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD index 3f64fab3a..ac48ab34b 100644 --- a/pkg/sentry/fsimpl/devpts/BUILD +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -21,8 +21,8 @@ go_library( "line_discipline.go", "master.go", "queue.go", + "replica.go", "root_inode_refs.go", - "slave.go", "terminal.go", ], visibility = ["//pkg/sentry:internal"], @@ -43,6 +43,8 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 57580f4d4..dcf1ee25b 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -79,7 +79,7 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds // Construct the root directory. This is always inode id 1. root := &rootInode{ - slaves: make(map[uint32]*slaveInode), + replicas: make(map[uint32]*replicaInode), } root.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, devMinor, 1, linux.ModeDirectory|0555) root.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) @@ -133,8 +133,8 @@ type rootInode struct { // mu protects the fields below. mu sync.Mutex - // slaves maps pty ids to slave inodes. - slaves map[uint32]*slaveInode + // replicas maps pty ids to replica inodes. + replicas map[uint32]*replicaInode // nextIdx is the next pty index to use. Must be accessed atomically. // @@ -154,22 +154,22 @@ func (i *rootInode) allocateTerminal(creds *auth.Credentials) (*Terminal, error) idx := i.nextIdx i.nextIdx++ - // Sanity check that slave with idx does not exist. - if _, ok := i.slaves[idx]; ok { + // Sanity check that replica with idx does not exist. + if _, ok := i.replicas[idx]; ok { panic(fmt.Sprintf("pty index collision; index %d already exists", idx)) } - // Create the new terminal and slave. + // Create the new terminal and replica. t := newTerminal(idx) - slave := &slaveInode{ + replica := &replicaInode{ root: i, t: t, } // Linux always uses pty index + 3 as the inode id. See // fs/devpts/inode.c:devpts_pty_new(). - slave.InodeAttrs.Init(creds, i.InodeAttrs.DevMajor(), i.InodeAttrs.DevMinor(), uint64(idx+3), linux.ModeCharacterDevice|0600) - slave.dentry.Init(slave) - i.slaves[idx] = slave + replica.InodeAttrs.Init(creds, i.InodeAttrs.DevMajor(), i.InodeAttrs.DevMinor(), uint64(idx+3), linux.ModeCharacterDevice|0600) + replica.dentry.Init(replica) + i.replicas[idx] = replica return t, nil } @@ -179,11 +179,11 @@ func (i *rootInode) masterClose(t *Terminal) { i.mu.Lock() defer i.mu.Unlock() - // Sanity check that slave with idx exists. - if _, ok := i.slaves[t.n]; !ok { + // Sanity check that replica with idx exists. + if _, ok := i.replicas[t.n]; !ok { panic(fmt.Sprintf("pty with index %d does not exist", t.n)) } - delete(i.slaves, t.n) + delete(i.replicas, t.n) } // Open implements kernfs.Inode.Open. @@ -205,7 +205,7 @@ func (i *rootInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error } i.mu.Lock() defer i.mu.Unlock() - if si, ok := i.slaves[uint32(idx)]; ok { + if si, ok := i.replicas[uint32(idx)]; ok { si.dentry.IncRef() return si.dentry.VFSDentry(), nil @@ -217,8 +217,8 @@ func (i *rootInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error func (i *rootInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { i.mu.Lock() defer i.mu.Unlock() - ids := make([]int, 0, len(i.slaves)) - for id := range i.slaves { + ids := make([]int, 0, len(i.replicas)) + for id := range i.replicas { ids = append(ids, int(id)) } sort.Ints(ids) @@ -226,7 +226,7 @@ func (i *rootInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(id), 10), Type: linux.DT_CHR, - Ino: i.slaves[uint32(id)].InodeAttrs.Ino(), + Ino: i.replicas[uint32(id)].InodeAttrs.Ino(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { diff --git a/pkg/sentry/fsimpl/devpts/devpts_test.go b/pkg/sentry/fsimpl/devpts/devpts_test.go index b7c149047..448390cfe 100644 --- a/pkg/sentry/fsimpl/devpts/devpts_test.go +++ b/pkg/sentry/fsimpl/devpts/devpts_test.go @@ -22,8 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -func TestSimpleMasterToSlave(t *testing.T) { - ld := newLineDiscipline(linux.DefaultSlaveTermios) +func TestSimpleMasterToReplica(t *testing.T) { + ld := newLineDiscipline(linux.DefaultReplicaTermios) ctx := contexttest.Context(t) inBytes := []byte("hello, tty\n") src := usermem.BytesIOSequence(inBytes) diff --git a/pkg/sentry/fsimpl/devpts/line_discipline.go b/pkg/sentry/fsimpl/devpts/line_discipline.go index f7bc325d1..e6b0e81cf 100644 --- a/pkg/sentry/fsimpl/devpts/line_discipline.go +++ b/pkg/sentry/fsimpl/devpts/line_discipline.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -41,7 +42,7 @@ const ( ) // lineDiscipline dictates how input and output are handled between the -// pseudoterminal (pty) master and slave. It can be configured to alter I/O, +// pseudoterminal (pty) master and replica. It can be configured to alter I/O, // modify control characters (e.g. Ctrl-C for SIGINT), etc. The following man // pages are good resources for how to affect the line discipline: // @@ -52,8 +53,8 @@ const ( // // lineDiscipline has a simple structure but supports a multitude of options // (see the above man pages). It consists of two queues of bytes: one from the -// terminal master to slave (the input queue) and one from slave to master (the -// output queue). When bytes are written to one end of the pty, the line +// terminal master to replica (the input queue) and one from replica to master +// (the output queue). When bytes are written to one end of the pty, the line // discipline reads the bytes, modifies them or takes special action if // required, and enqueues them to be read by the other end of the pty: // @@ -62,7 +63,7 @@ const ( // | (inputQueueWrite) +-------------+ (inputQueueRead) | // | | // | v -// masterFD slaveFD +// masterFD replicaFD // ^ | // | | // | output to terminal +--------------+ output from process | @@ -101,8 +102,8 @@ type lineDiscipline struct { // masterWaiter is used to wait on the master end of the TTY. masterWaiter waiter.Queue `state:"zerovalue"` - // slaveWaiter is used to wait on the slave end of the TTY. - slaveWaiter waiter.Queue `state:"zerovalue"` + // replicaWaiter is used to wait on the replica end of the TTY. + replicaWaiter waiter.Queue `state:"zerovalue"` } func newLineDiscipline(termios linux.KernelTermios) *lineDiscipline { @@ -113,27 +114,23 @@ func newLineDiscipline(termios linux.KernelTermios) *lineDiscipline { } // getTermios gets the linux.Termios for the tty. -func (l *lineDiscipline) getTermios(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (l *lineDiscipline) getTermios(task *kernel.Task, args arch.SyscallArguments) (uintptr, error) { l.termiosMu.RLock() defer l.termiosMu.RUnlock() // We must copy a Termios struct, not KernelTermios. t := l.termios.ToTermios() - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), t, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := t.CopyOut(task, args[2].Pointer()) return 0, err } // setTermios sets a linux.Termios for the tty. -func (l *lineDiscipline) setTermios(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (l *lineDiscipline) setTermios(task *kernel.Task, args arch.SyscallArguments) (uintptr, error) { l.termiosMu.Lock() defer l.termiosMu.Unlock() oldCanonEnabled := l.termios.LEnabled(linux.ICANON) // We must copy a Termios struct, not KernelTermios. var t linux.Termios - _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &t, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := t.CopyIn(task, args[2].Pointer()) l.termios.FromTermios(t) // If canonical mode is turned off, move bytes from inQueue's wait @@ -144,27 +141,23 @@ func (l *lineDiscipline) setTermios(ctx context.Context, io usermem.IO, args arc l.inQueue.pushWaitBufLocked(l) l.inQueue.readable = true l.inQueue.mu.Unlock() - l.slaveWaiter.Notify(waiter.EventIn) + l.replicaWaiter.Notify(waiter.EventIn) } return 0, err } -func (l *lineDiscipline) windowSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { +func (l *lineDiscipline) windowSize(t *kernel.Task, args arch.SyscallArguments) error { l.sizeMu.Lock() defer l.sizeMu.Unlock() - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), l.size, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := l.size.CopyOut(t, args[2].Pointer()) return err } -func (l *lineDiscipline) setWindowSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { +func (l *lineDiscipline) setWindowSize(t *kernel.Task, args arch.SyscallArguments) error { l.sizeMu.Lock() defer l.sizeMu.Unlock() - _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &l.size, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := l.size.CopyIn(t, args[2].Pointer()) return err } @@ -174,14 +167,14 @@ func (l *lineDiscipline) masterReadiness() waiter.EventMask { return l.inQueue.writeReadiness(&linux.MasterTermios) | l.outQueue.readReadiness(&linux.MasterTermios) } -func (l *lineDiscipline) slaveReadiness() waiter.EventMask { +func (l *lineDiscipline) replicaReadiness() waiter.EventMask { l.termiosMu.RLock() defer l.termiosMu.RUnlock() return l.outQueue.writeReadiness(&l.termios) | l.inQueue.readReadiness(&l.termios) } -func (l *lineDiscipline) inputQueueReadSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { - return l.inQueue.readableSize(ctx, io, args) +func (l *lineDiscipline) inputQueueReadSize(t *kernel.Task, io usermem.IO, args arch.SyscallArguments) error { + return l.inQueue.readableSize(t, io, args) } func (l *lineDiscipline) inputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) { @@ -194,7 +187,7 @@ func (l *lineDiscipline) inputQueueRead(ctx context.Context, dst usermem.IOSeque if n > 0 { l.masterWaiter.Notify(waiter.EventOut) if pushed { - l.slaveWaiter.Notify(waiter.EventIn) + l.replicaWaiter.Notify(waiter.EventIn) } return n, nil } @@ -209,14 +202,14 @@ func (l *lineDiscipline) inputQueueWrite(ctx context.Context, src usermem.IOSequ return 0, err } if n > 0 { - l.slaveWaiter.Notify(waiter.EventIn) + l.replicaWaiter.Notify(waiter.EventIn) return n, nil } return 0, syserror.ErrWouldBlock } -func (l *lineDiscipline) outputQueueReadSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { - return l.outQueue.readableSize(ctx, io, args) +func (l *lineDiscipline) outputQueueReadSize(t *kernel.Task, io usermem.IO, args arch.SyscallArguments) error { + return l.outQueue.readableSize(t, io, args) } func (l *lineDiscipline) outputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) { @@ -227,7 +220,7 @@ func (l *lineDiscipline) outputQueueRead(ctx context.Context, dst usermem.IOSequ return 0, err } if n > 0 { - l.slaveWaiter.Notify(waiter.EventOut) + l.replicaWaiter.Notify(waiter.EventOut) if pushed { l.masterWaiter.Notify(waiter.EventIn) } diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 60feb1993..d07e1ded8 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -20,12 +20,14 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // masterInode is the inode for the master end of the Terminal. @@ -131,46 +133,51 @@ func (mfd *masterFileDescription) Write(ctx context.Context, src usermem.IOSeque // Ioctl implements vfs.FileDescriptionImpl.Ioctl. func (mfd *masterFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + // ioctl(2) may only be called from a task goroutine. + return 0, syserror.ENOTTY + } + switch cmd := args[1].Uint(); cmd { case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ // Get the number of bytes in the output queue read buffer. - return 0, mfd.t.ld.outputQueueReadSize(ctx, io, args) + return 0, mfd.t.ld.outputQueueReadSize(t, io, args) case linux.TCGETS: // N.B. TCGETS on the master actually returns the configuration - // of the slave end. - return mfd.t.ld.getTermios(ctx, io, args) + // of the replica end. + return mfd.t.ld.getTermios(t, args) case linux.TCSETS: // N.B. TCSETS on the master actually affects the configuration - // of the slave end. - return mfd.t.ld.setTermios(ctx, io, args) + // of the replica end. + return mfd.t.ld.setTermios(t, args) case linux.TCSETSW: // TODO(b/29356795): This should drain the output queue first. - return mfd.t.ld.setTermios(ctx, io, args) + return mfd.t.ld.setTermios(t, args) case linux.TIOCGPTN: - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(mfd.t.n), usermem.IOOpts{ - AddressSpaceActive: true, - }) + nP := primitive.Uint32(mfd.t.n) + _, err := nP.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCSPTLCK: // TODO(b/29356795): Implement pty locking. For now just pretend we do. return 0, nil case linux.TIOCGWINSZ: - return 0, mfd.t.ld.windowSize(ctx, io, args) + return 0, mfd.t.ld.windowSize(t, args) case linux.TIOCSWINSZ: - return 0, mfd.t.ld.setWindowSize(ctx, io, args) + return 0, mfd.t.ld.setWindowSize(t, args) case linux.TIOCSCTTY: // Make the given terminal the controlling terminal of the // calling process. - return 0, mfd.t.setControllingTTY(ctx, io, args, true /* isMaster */) + return 0, mfd.t.setControllingTTY(ctx, args, true /* isMaster */) case linux.TIOCNOTTY: // Release this process's controlling terminal. - return 0, mfd.t.releaseControllingTTY(ctx, io, args, true /* isMaster */) + return 0, mfd.t.releaseControllingTTY(ctx, args, true /* isMaster */) case linux.TIOCGPGRP: // Get the foreground process group. - return mfd.t.foregroundProcessGroup(ctx, io, args, true /* isMaster */) + return mfd.t.foregroundProcessGroup(ctx, args, true /* isMaster */) case linux.TIOCSPGRP: // Set the foreground process group. - return mfd.t.setForegroundProcessGroup(ctx, io, args, true /* isMaster */) + return mfd.t.setForegroundProcessGroup(ctx, args, true /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY diff --git a/pkg/sentry/fsimpl/devpts/queue.go b/pkg/sentry/fsimpl/devpts/queue.go index 331c13997..ca36b66e9 100644 --- a/pkg/sentry/fsimpl/devpts/queue.go +++ b/pkg/sentry/fsimpl/devpts/queue.go @@ -19,10 +19,12 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // waitBufMaxBytes is the maximum size of a wait buffer. It is based on @@ -30,7 +32,7 @@ import ( const waitBufMaxBytes = 131072 // queue represents one of the input or output queues between a pty master and -// slave. Bytes written to a queue are added to the read buffer until it is +// replica. Bytes written to a queue are added to the read buffer until it is // full, at which point they are written to the wait buffer. Bytes are // processed (i.e. undergo termios transformations) as they are added to the // read buffer. The read buffer is readable when its length is nonzero and @@ -83,17 +85,15 @@ func (q *queue) writeReadiness(t *linux.KernelTermios) waiter.EventMask { } // readableSize writes the number of readable bytes to userspace. -func (q *queue) readableSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { +func (q *queue) readableSize(t *kernel.Task, io usermem.IO, args arch.SyscallArguments) error { q.mu.Lock() defer q.mu.Unlock() - var size int32 + size := primitive.Int32(0) if q.readable { - size = int32(len(q.readBuf)) + size = primitive.Int32(len(q.readBuf)) } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), size, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := size.CopyOut(t, args[2].Pointer()) return err } diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/replica.go index a9da7af64..1f99f4b4d 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/replica.go @@ -20,15 +20,17 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) -// slaveInode is the inode for the slave end of the Terminal. -type slaveInode struct { +// replicaInode is the inode for the replica end of the Terminal. +type replicaInode struct { implStatFS kernfs.InodeAttrs kernfs.InodeNoopRefCount @@ -47,12 +49,12 @@ type slaveInode struct { t *Terminal } -var _ kernfs.Inode = (*slaveInode)(nil) +var _ kernfs.Inode = (*replicaInode)(nil) // Open implements kernfs.Inode.Open. -func (si *slaveInode) 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, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { si.IncRef() - fd := &slaveFileDescription{ + fd := &replicaFileDescription{ inode: si, } fd.LockFD.Init(&si.locks) @@ -65,109 +67,114 @@ func (si *slaveInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs } // Valid implements kernfs.Inode.Valid. -func (si *slaveInode) Valid(context.Context) bool { - // Return valid if the slave still exists. +func (si *replicaInode) Valid(context.Context) bool { + // Return valid if the replica still exists. si.root.mu.Lock() defer si.root.mu.Unlock() - _, ok := si.root.slaves[si.t.n] + _, ok := si.root.replicas[si.t.n] return ok } // Stat implements kernfs.Inode.Stat. -func (si *slaveInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { +func (si *replicaInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { statx, err := si.InodeAttrs.Stat(ctx, vfsfs, opts) if err != nil { return linux.Statx{}, err } statx.Blksize = 1024 - statx.RdevMajor = linux.UNIX98_PTY_SLAVE_MAJOR + statx.RdevMajor = linux.UNIX98_PTY_REPLICA_MAJOR statx.RdevMinor = si.t.n return statx, nil } // SetStat implements kernfs.Inode.SetStat -func (si *slaveInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { +func (si *replicaInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { if opts.Stat.Mask&linux.STATX_SIZE != 0 { return syserror.EINVAL } return si.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) } -type slaveFileDescription struct { +type replicaFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.LockFD - inode *slaveInode + inode *replicaInode } -var _ vfs.FileDescriptionImpl = (*slaveFileDescription)(nil) +var _ vfs.FileDescriptionImpl = (*replicaFileDescription)(nil) // Release implements fs.FileOperations.Release. -func (sfd *slaveFileDescription) Release(ctx context.Context) { +func (sfd *replicaFileDescription) Release(ctx context.Context) { sfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. -func (sfd *slaveFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - sfd.inode.t.ld.slaveWaiter.EventRegister(e, mask) +func (sfd *replicaFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + sfd.inode.t.ld.replicaWaiter.EventRegister(e, mask) } // EventUnregister implements waiter.Waitable.EventUnregister. -func (sfd *slaveFileDescription) EventUnregister(e *waiter.Entry) { - sfd.inode.t.ld.slaveWaiter.EventUnregister(e) +func (sfd *replicaFileDescription) EventUnregister(e *waiter.Entry) { + sfd.inode.t.ld.replicaWaiter.EventUnregister(e) } // Readiness implements waiter.Waitable.Readiness. -func (sfd *slaveFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { - return sfd.inode.t.ld.slaveReadiness() +func (sfd *replicaFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return sfd.inode.t.ld.replicaReadiness() } // Read implements vfs.FileDescriptionImpl.Read. -func (sfd *slaveFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { +func (sfd *replicaFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { return sfd.inode.t.ld.inputQueueRead(ctx, dst) } // Write implements vfs.FileDescriptionImpl.Write. -func (sfd *slaveFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { +func (sfd *replicaFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { return sfd.inode.t.ld.outputQueueWrite(ctx, src) } // Ioctl implements vfs.FileDescriptionImpl.Ioctl. -func (sfd *slaveFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + // ioctl(2) may only be called from a task goroutine. + return 0, syserror.ENOTTY + } + switch cmd := args[1].Uint(); cmd { case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ // Get the number of bytes in the input queue read buffer. - return 0, sfd.inode.t.ld.inputQueueReadSize(ctx, io, args) + return 0, sfd.inode.t.ld.inputQueueReadSize(t, io, args) case linux.TCGETS: - return sfd.inode.t.ld.getTermios(ctx, io, args) + return sfd.inode.t.ld.getTermios(t, args) case linux.TCSETS: - return sfd.inode.t.ld.setTermios(ctx, io, args) + return sfd.inode.t.ld.setTermios(t, args) case linux.TCSETSW: // TODO(b/29356795): This should drain the output queue first. - return sfd.inode.t.ld.setTermios(ctx, io, args) + return sfd.inode.t.ld.setTermios(t, args) case linux.TIOCGPTN: - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(sfd.inode.t.n), usermem.IOOpts{ - AddressSpaceActive: true, - }) + nP := primitive.Uint32(sfd.inode.t.n) + _, err := nP.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCGWINSZ: - return 0, sfd.inode.t.ld.windowSize(ctx, io, args) + return 0, sfd.inode.t.ld.windowSize(t, args) case linux.TIOCSWINSZ: - return 0, sfd.inode.t.ld.setWindowSize(ctx, io, args) + return 0, sfd.inode.t.ld.setWindowSize(t, args) case linux.TIOCSCTTY: // Make the given terminal the controlling terminal of the // calling process. - return 0, sfd.inode.t.setControllingTTY(ctx, io, args, false /* isMaster */) + return 0, sfd.inode.t.setControllingTTY(ctx, args, false /* isMaster */) case linux.TIOCNOTTY: // Release this process's controlling terminal. - return 0, sfd.inode.t.releaseControllingTTY(ctx, io, args, false /* isMaster */) + return 0, sfd.inode.t.releaseControllingTTY(ctx, args, false /* isMaster */) case linux.TIOCGPGRP: // Get the foreground process group. - return sfd.inode.t.foregroundProcessGroup(ctx, io, args, false /* isMaster */) + return sfd.inode.t.foregroundProcessGroup(ctx, args, false /* isMaster */) case linux.TIOCSPGRP: // Set the foreground process group. - return sfd.inode.t.setForegroundProcessGroup(ctx, io, args, false /* isMaster */) + return sfd.inode.t.setForegroundProcessGroup(ctx, args, false /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY @@ -175,24 +182,24 @@ func (sfd *slaveFileDescription) Ioctl(ctx context.Context, io usermem.IO, args } // SetStat implements vfs.FileDescriptionImpl.SetStat. -func (sfd *slaveFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { +func (sfd *replicaFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() return sfd.inode.SetStat(ctx, fs, creds, opts) } // Stat implements vfs.FileDescriptionImpl.Stat. -func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { +func (sfd *replicaFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() return sfd.inode.Stat(ctx, fs, opts) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (sfd *slaveFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { +func (sfd *replicaFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { return sfd.Locks().LockPOSIX(ctx, &sfd.vfsfd, uid, t, start, length, whence, block) } // UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (sfd *slaveFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { +func (sfd *replicaFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { return sfd.Locks().UnlockPOSIX(ctx, &sfd.vfsfd, uid, start, length, whence) } diff --git a/pkg/sentry/fsimpl/devpts/terminal.go b/pkg/sentry/fsimpl/devpts/terminal.go index 7d2781c54..731955d62 100644 --- a/pkg/sentry/fsimpl/devpts/terminal.go +++ b/pkg/sentry/fsimpl/devpts/terminal.go @@ -19,7 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // Terminal is a pseudoterminal. @@ -36,25 +36,25 @@ type Terminal struct { // this terminal. This field is immutable. masterKTTY *kernel.TTY - // slaveKTTY contains the controlling process of the slave end of this + // replicaKTTY contains the controlling process of the replica end of this // terminal. This field is immutable. - slaveKTTY *kernel.TTY + replicaKTTY *kernel.TTY } func newTerminal(n uint32) *Terminal { - termios := linux.DefaultSlaveTermios + termios := linux.DefaultReplicaTermios t := Terminal{ - n: n, - ld: newLineDiscipline(termios), - masterKTTY: &kernel.TTY{Index: n}, - slaveKTTY: &kernel.TTY{Index: n}, + n: n, + ld: newLineDiscipline(termios), + masterKTTY: &kernel.TTY{Index: n}, + replicaKTTY: &kernel.TTY{Index: n}, } return &t } // setControllingTTY makes tm the controlling terminal of the calling thread // group. -func (tm *Terminal) setControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { +func (tm *Terminal) setControllingTTY(ctx context.Context, args arch.SyscallArguments, isMaster bool) error { task := kernel.TaskFromContext(ctx) if task == nil { panic("setControllingTTY must be called from a task context") @@ -65,7 +65,7 @@ func (tm *Terminal) setControllingTTY(ctx context.Context, io usermem.IO, args a // releaseControllingTTY removes tm as the controlling terminal of the calling // thread group. -func (tm *Terminal) releaseControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { +func (tm *Terminal) releaseControllingTTY(ctx context.Context, args arch.SyscallArguments, isMaster bool) error { task := kernel.TaskFromContext(ctx) if task == nil { panic("releaseControllingTTY must be called from a task context") @@ -75,7 +75,7 @@ func (tm *Terminal) releaseControllingTTY(ctx context.Context, io usermem.IO, ar } // foregroundProcessGroup gets the process group ID of tm's foreground process. -func (tm *Terminal) foregroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { +func (tm *Terminal) foregroundProcessGroup(ctx context.Context, args arch.SyscallArguments, isMaster bool) (uintptr, error) { task := kernel.TaskFromContext(ctx) if task == nil { panic("foregroundProcessGroup must be called from a task context") @@ -87,24 +87,21 @@ func (tm *Terminal) foregroundProcessGroup(ctx context.Context, io usermem.IO, a } // Write it out to *arg. - _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(ret), usermem.IOOpts{ - AddressSpaceActive: true, - }) + retP := primitive.Int32(ret) + _, err = retP.CopyOut(task, args[2].Pointer()) return 0, err } // foregroundProcessGroup sets tm's foreground process. -func (tm *Terminal) setForegroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { +func (tm *Terminal) setForegroundProcessGroup(ctx context.Context, args arch.SyscallArguments, isMaster bool) (uintptr, error) { task := kernel.TaskFromContext(ctx) if task == nil { panic("setForegroundProcessGroup must be called from a task context") } // Read in the process group ID. - var pgid int32 - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgid, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + var pgid primitive.Int32 + if _, err := pgid.CopyIn(task, args[2].Pointer()); err != nil { return 0, err } @@ -116,5 +113,5 @@ func (tm *Terminal) tty(isMaster bool) *kernel.TTY { if isMaster { return tm.masterKTTY } - return tm.slaveKTTY + return tm.replicaKTTY } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 57bff1789..73d9e772d 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1300,30 +1300,36 @@ func (d *dentry) destroyLocked(ctx context.Context) { d.handleMu.Unlock() if !d.file.isNil() { + valid := p9.SetAttrMask{} + attr := p9.SetAttr{} if !d.isDeleted() { // Write dirty timestamps back to the remote filesystem. - atimeDirty := atomic.LoadUint32(&d.atimeDirty) != 0 - mtimeDirty := atomic.LoadUint32(&d.mtimeDirty) != 0 - if atimeDirty || mtimeDirty { + if atomic.LoadUint32(&d.atimeDirty) != 0 { + valid.ATime = true + valid.ATimeNotSystemTime = true atime := atomic.LoadInt64(&d.atime) + attr.ATimeSeconds = uint64(atime / 1e9) + attr.ATimeNanoSeconds = uint64(atime % 1e9) + } + if atomic.LoadUint32(&d.mtimeDirty) != 0 { + valid.MTime = true + valid.MTimeNotSystemTime = true mtime := atomic.LoadInt64(&d.mtime) - if err := d.file.setAttr(ctx, p9.SetAttrMask{ - ATime: atimeDirty, - ATimeNotSystemTime: atimeDirty, - MTime: mtimeDirty, - MTimeNotSystemTime: mtimeDirty, - }, p9.SetAttr{ - ATimeSeconds: uint64(atime / 1e9), - ATimeNanoSeconds: uint64(atime % 1e9), - MTimeSeconds: uint64(mtime / 1e9), - MTimeNanoSeconds: uint64(mtime % 1e9), - }); err != nil { - log.Warningf("gofer.dentry.destroyLocked: failed to write dirty timestamps back: %v", err) - } + attr.MTimeSeconds = uint64(mtime / 1e9) + attr.MTimeNanoSeconds = uint64(mtime % 1e9) + } + } + + // Check if attributes need to be changed before closing the file. + if valid.ATime || valid.MTime { + if err := d.file.setAttrClose(ctx, valid, attr); err != nil { + log.Warningf("gofer.dentry.destroyLocked: failed to close file with write dirty timestamps: %v", err) } + } else if err := d.file.close(ctx); err != nil { + log.Warningf("gofer.dentry.destroyLocked: failed to close file: %v", err) } - d.file.close(ctx) d.file = p9file{} + // Remove d from the set of syncable dentries. d.fs.syncMu.Lock() delete(d.fs.syncableDentries, d) diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go index 87f0b877f..21b4a96fe 100644 --- a/pkg/sentry/fsimpl/gofer/p9file.go +++ b/pkg/sentry/fsimpl/gofer/p9file.go @@ -127,6 +127,13 @@ func (f p9file) close(ctx context.Context) error { return err } +func (f p9file) setAttrClose(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAttr) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.SetAttrClose(valid, attr) + ctx.UninterruptibleSleepFinish(false) + return err +} + func (f p9file) open(ctx context.Context, flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { ctx.UninterruptibleSleepStart(false) fdobj, qid, iounit, err := f.file.Open(flags) diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 7561f821c..1bd0e4ee8 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -496,7 +496,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u if i.isTTY { fd := &TTYFileDescription{ fileDescription: fileDescription{inode: i}, - termios: linux.DefaultSlaveTermios, + termios: linux.DefaultReplicaTermios, } fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd |