summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fs/tty
diff options
context:
space:
mode:
authorgVisor bot <gvisor-bot@google.com>2019-06-02 06:44:55 +0000
committergVisor bot <gvisor-bot@google.com>2019-06-02 06:44:55 +0000
commitceb0d792f328d1fc0692197d8856a43c3936a571 (patch)
tree83155f302eff44a78bcc30a3a08f4efe59a79379 /pkg/sentry/fs/tty
parentdeb7ecf1e46862d54f4b102f2d163cfbcfc37f3b (diff)
parent216da0b733dbed9aad9b2ab92ac75bcb906fd7ee (diff)
Merge 216da0b7 (automated)
Diffstat (limited to 'pkg/sentry/fs/tty')
-rw-r--r--pkg/sentry/fs/tty/dir.go339
-rw-r--r--pkg/sentry/fs/tty/fs.go104
-rw-r--r--pkg/sentry/fs/tty/line_discipline.go443
-rw-r--r--pkg/sentry/fs/tty/master.go220
-rw-r--r--pkg/sentry/fs/tty/queue.go244
-rw-r--r--pkg/sentry/fs/tty/slave.go162
-rw-r--r--pkg/sentry/fs/tty/terminal.go46
-rwxr-xr-xpkg/sentry/fs/tty/tty_state_autogen.go202
8 files changed, 1760 insertions, 0 deletions
diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go
new file mode 100644
index 000000000..2603354c4
--- /dev/null
+++ b/pkg/sentry/fs/tty/dir.go
@@ -0,0 +1,339 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package tty provide pseudoterminals via a devpts filesystem.
+package tty
+
+import (
+ "fmt"
+ "math"
+ "strconv"
+ "sync"
+
+ "gvisor.googlesource.com/gvisor/pkg/abi/linux"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
+ "gvisor.googlesource.com/gvisor/pkg/syserror"
+ "gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+// dirInodeOperations is the root of a devpts mount.
+//
+// This indirectly manages all terminals within the mount.
+//
+// New Terminals are created by masterInodeOperations.GetFile, which registers
+// the slave Inode in the this directory for discovery via Lookup/Readdir. The
+// slave inode is unregistered when the master file is Released, as the slave
+// is no longer discoverable at that point.
+//
+// References on the underlying Terminal are held by masterFileOperations and
+// slaveInodeOperations.
+//
+// masterInodeOperations and slaveInodeOperations hold a pointer to
+// dirInodeOperations, which is reference counted by the refcount their
+// corresponding Dirents hold on their parent (this directory).
+//
+// dirInodeOperations implements fs.InodeOperations.
+//
+// +stateify savable
+type dirInodeOperations struct {
+ fsutil.InodeGenericChecker `state:"nosave"`
+ fsutil.InodeIsDirAllocate `state:"nosave"`
+ fsutil.InodeIsDirTruncate `state:"nosave"`
+ fsutil.InodeNoExtendedAttributes `state:"nosave"`
+ fsutil.InodeNoopWriteOut `state:"nosave"`
+ fsutil.InodeNotMappable `state:"nosave"`
+ fsutil.InodeNotRenameable `state:"nosave"`
+ fsutil.InodeNotSocket `state:"nosave"`
+ fsutil.InodeNotSymlink `state:"nosave"`
+ fsutil.InodeVirtual `state:"nosave"`
+
+ fsutil.InodeSimpleAttributes
+
+ // msrc is the super block this directory is on.
+ //
+ // TODO(chrisko): Plumb this through instead of storing it here.
+ msrc *fs.MountSource
+
+ // mu protects the fields below.
+ mu sync.Mutex `state:"nosave"`
+
+ // master is the master PTY inode.
+ master *fs.Inode
+
+ // slaves contains the slave inodes reachable from the directory.
+ //
+ // A new slave is added by allocateTerminal and is removed by
+ // masterFileOperations.Release.
+ //
+ // A reference is held on every slave in the map.
+ slaves map[uint32]*fs.Inode
+
+ // dentryMap is a SortedDentryMap used to implement Readdir containing
+ // the master and all entries in slaves.
+ dentryMap *fs.SortedDentryMap
+
+ // next is the next pty index to use.
+ //
+ // TODO(b/29356795): reuse indices when ptys are closed.
+ next uint32
+}
+
+var _ fs.InodeOperations = (*dirInodeOperations)(nil)
+
+// newDir creates a new dir with a ptmx file and no terminals.
+func newDir(ctx context.Context, m *fs.MountSource) *fs.Inode {
+ d := &dirInodeOperations{
+ InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0555), linux.DEVPTS_SUPER_MAGIC),
+ msrc: m,
+ slaves: make(map[uint32]*fs.Inode),
+ dentryMap: fs.NewSortedDentryMap(nil),
+ }
+ // Linux devpts uses a default mode of 0000 for ptmx which can be
+ // changed with the ptmxmode mount option. However, that default is not
+ // useful here (since we'd *always* need the mount option, so it is
+ // accessible by default).
+ d.master = newMasterInode(ctx, d, fs.RootOwner, fs.FilePermsFromMode(0666))
+ d.dentryMap.Add("ptmx", fs.DentAttr{
+ Type: d.master.StableAttr.Type,
+ InodeID: d.master.StableAttr.InodeID,
+ })
+
+ return fs.NewInode(d, m, fs.StableAttr{
+ DeviceID: ptsDevice.DeviceID(),
+ // N.B. Linux always uses inode id 1 for the directory. See
+ // fs/devpts/inode.c:devpts_fill_super.
+ //
+ // TODO(b/75267214): Since ptsDevice must be shared between
+ // different mounts, we must not assign fixed numbers.
+ InodeID: ptsDevice.NextIno(),
+ BlockSize: usermem.PageSize,
+ Type: fs.Directory,
+ })
+}
+
+// Release implements fs.InodeOperations.Release.
+func (d *dirInodeOperations) Release(ctx context.Context) {
+ d.master.DecRef()
+ if len(d.slaves) != 0 {
+ panic(fmt.Sprintf("devpts directory still contains active terminals: %+v", d))
+ }
+}
+
+// Lookup implements fs.InodeOperations.Lookup.
+func (d *dirInodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dirent, error) {
+ d.mu.Lock()
+ defer d.mu.Unlock()
+
+ // Master?
+ if name == "ptmx" {
+ d.master.IncRef()
+ return fs.NewDirent(d.master, name), nil
+ }
+
+ // Slave number?
+ n, err := strconv.ParseUint(name, 10, 32)
+ if err != nil {
+ // Not found.
+ return nil, syserror.ENOENT
+ }
+
+ s, ok := d.slaves[uint32(n)]
+ if !ok {
+ return nil, syserror.ENOENT
+ }
+
+ s.IncRef()
+ return fs.NewDirent(s, name), nil
+}
+
+// Create implements fs.InodeOperations.Create.
+//
+// Creation is never allowed.
+func (d *dirInodeOperations) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.FileFlags, perm fs.FilePermissions) (*fs.File, error) {
+ return nil, syserror.EACCES
+}
+
+// CreateDirectory implements fs.InodeOperations.CreateDirectory.
+//
+// Creation is never allowed.
+func (d *dirInodeOperations) CreateDirectory(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions) error {
+ return syserror.EACCES
+}
+
+// CreateLink implements fs.InodeOperations.CreateLink.
+//
+// Creation is never allowed.
+func (d *dirInodeOperations) CreateLink(ctx context.Context, dir *fs.Inode, oldname, newname string) error {
+ return syserror.EACCES
+}
+
+// CreateHardLink implements fs.InodeOperations.CreateHardLink.
+//
+// Creation is never allowed.
+func (d *dirInodeOperations) CreateHardLink(ctx context.Context, dir *fs.Inode, target *fs.Inode, name string) error {
+ return syserror.EACCES
+}
+
+// CreateFifo implements fs.InodeOperations.CreateFifo.
+//
+// Creation is never allowed.
+func (d *dirInodeOperations) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions) error {
+ return syserror.EACCES
+}
+
+// Remove implements fs.InodeOperations.Remove.
+//
+// Removal is never allowed.
+func (d *dirInodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string) error {
+ return syserror.EPERM
+}
+
+// RemoveDirectory implements fs.InodeOperations.RemoveDirectory.
+//
+// Removal is never allowed.
+func (d *dirInodeOperations) RemoveDirectory(ctx context.Context, dir *fs.Inode, name string) error {
+ return syserror.EPERM
+}
+
+// Bind implements fs.InodeOperations.Bind.
+func (d *dirInodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, data transport.BoundEndpoint, perm fs.FilePermissions) (*fs.Dirent, error) {
+ return nil, syserror.EPERM
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+func (d *dirInodeOperations) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+ return fs.NewFile(ctx, dirent, flags, &dirFileOperations{di: d}), nil
+}
+
+// allocateTerminal creates a new Terminal and installs a pts node for it.
+//
+// The caller must call DecRef when done with the returned Terminal.
+func (d *dirInodeOperations) allocateTerminal(ctx context.Context) (*Terminal, error) {
+ d.mu.Lock()
+ defer d.mu.Unlock()
+
+ n := d.next
+ if n == math.MaxUint32 {
+ return nil, syserror.ENOMEM
+ }
+
+ if _, ok := d.slaves[n]; ok {
+ panic(fmt.Sprintf("pty index collision; index %d already exists", n))
+ }
+
+ t := newTerminal(ctx, d, n)
+ d.next++
+
+ // The reference returned by newTerminal is returned to the caller.
+ // Take another for the slave inode.
+ t.IncRef()
+
+ // Create a pts node. The owner is based on the context that opens
+ // ptmx.
+ creds := auth.CredentialsFromContext(ctx)
+ uid, gid := creds.EffectiveKUID, creds.EffectiveKGID
+ slave := newSlaveInode(ctx, d, t, fs.FileOwner{uid, gid}, fs.FilePermsFromMode(0666))
+
+ d.slaves[n] = slave
+ d.dentryMap.Add(strconv.FormatUint(uint64(n), 10), fs.DentAttr{
+ Type: slave.StableAttr.Type,
+ InodeID: slave.StableAttr.InodeID,
+ })
+
+ return t, nil
+}
+
+// masterClose is called when the master end of t is closed.
+func (d *dirInodeOperations) masterClose(t *Terminal) {
+ d.mu.Lock()
+ defer d.mu.Unlock()
+
+ // The slave end disappears from the directory when the master end is
+ // closed, even if the slave end is open elsewhere.
+ //
+ // N.B. since we're using a backdoor method to remove a directory entry
+ // we won't properly fire inotify events like Linux would.
+ s, ok := d.slaves[t.n]
+ if !ok {
+ panic(fmt.Sprintf("Terminal %+v doesn't exist in %+v?", t, d))
+ }
+
+ s.DecRef()
+ delete(d.slaves, t.n)
+ d.dentryMap.Remove(strconv.FormatUint(uint64(t.n), 10))
+}
+
+// dirFileOperations are the fs.FileOperations for the directory.
+//
+// This is nearly identical to fsutil.DirFileOperations, except that it takes
+// df.di.mu in IterateDir.
+//
+// +stateify savable
+type dirFileOperations struct {
+ fsutil.FileNoopRelease `state:"nosave"`
+ fsutil.FileGenericSeek `state:"nosave"`
+ fsutil.FileNoFsync `state:"nosave"`
+ fsutil.FileNoopFlush `state:"nosave"`
+ fsutil.FileNoMMap `state:"nosave"`
+ fsutil.FileNoIoctl `state:"nosave"`
+ fsutil.FileNoSplice `state:"nosave"`
+ fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+ waiter.AlwaysReady `state:"nosave"`
+
+ // di is the inode operations.
+ di *dirInodeOperations
+
+ // dirCursor contains the name of the last directory entry that was
+ // serialized.
+ dirCursor string
+}
+
+var _ fs.FileOperations = (*dirFileOperations)(nil)
+
+// IterateDir implements DirIterator.IterateDir.
+func (df *dirFileOperations) IterateDir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) {
+ df.di.mu.Lock()
+ defer df.di.mu.Unlock()
+
+ n, err := fs.GenericReaddir(dirCtx, df.di.dentryMap)
+ return offset + n, err
+}
+
+// Readdir implements FileOperations.Readdir.
+func (df *dirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) {
+ root := fs.RootFromContext(ctx)
+ if root != nil {
+ defer root.DecRef()
+ }
+ dirCtx := &fs.DirCtx{
+ Serializer: serializer,
+ DirCursor: &df.dirCursor,
+ }
+ return fs.DirentReaddir(ctx, file.Dirent, df, root, dirCtx, file.Offset())
+}
+
+// Read implements FileOperations.Read
+func (df *dirFileOperations) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) {
+ return 0, syserror.EISDIR
+}
+
+// Write implements FileOperations.Write.
+func (df *dirFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) {
+ return 0, syserror.EISDIR
+}
diff --git a/pkg/sentry/fs/tty/fs.go b/pkg/sentry/fs/tty/fs.go
new file mode 100644
index 000000000..701b2f7d9
--- /dev/null
+++ b/pkg/sentry/fs/tty/fs.go
@@ -0,0 +1,104 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tty
+
+import (
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/device"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs"
+ "gvisor.googlesource.com/gvisor/pkg/syserror"
+)
+
+// ptsDevice is the pseudo-filesystem device.
+var ptsDevice = device.NewAnonDevice()
+
+// filesystem is a devpts filesystem.
+//
+// This devpts is always in the new "multi-instance" mode. i.e., it contains a
+// ptmx device tied to this mount.
+//
+// +stateify savable
+type filesystem struct{}
+
+func init() {
+ fs.RegisterFilesystem(&filesystem{})
+}
+
+// Name matches drivers/devpts/indoe.c:devpts_fs_type.name.
+func (*filesystem) Name() string {
+ return "devpts"
+}
+
+// AllowUserMount allows users to mount(2) this file system.
+func (*filesystem) AllowUserMount() bool {
+ // TODO(b/29356795): Users may mount this once the terminals are in a
+ // usable state.
+ return false
+}
+
+// AllowUserList allows this filesystem to be listed in /proc/filesystems.
+func (*filesystem) AllowUserList() bool {
+ return true
+}
+
+// Flags returns that there is nothing special about this file system.
+func (*filesystem) Flags() fs.FilesystemFlags {
+ return 0
+}
+
+// MountSource returns a devpts root that can be positioned in the vfs.
+func (f *filesystem) Mount(ctx context.Context, device string, flags fs.MountSourceFlags, data string, _ interface{}) (*fs.Inode, error) {
+ // device is always ignored.
+
+ // No options are supported.
+ if data != "" {
+ return nil, syserror.EINVAL
+ }
+
+ return newDir(ctx, fs.NewMountSource(&superOperations{}, f, flags)), nil
+}
+
+// superOperations implements fs.MountSourceOperations, preventing caching.
+//
+// +stateify savable
+type superOperations struct{}
+
+// Revalidate implements fs.DirentOperations.Revalidate.
+//
+// It always returns true, forcing a Lookup for all entries.
+//
+// Slave entries are dropped from dir when their master is closed, so an
+// existing slave Dirent in the tree is not sufficient to guarantee that it
+// still exists on the filesystem.
+func (superOperations) Revalidate(context.Context, string, *fs.Inode, *fs.Inode) bool {
+ return true
+}
+
+// Keep implements fs.DirentOperations.Keep.
+//
+// Keep returns false because Revalidate would force a lookup on cached entries
+// anyways.
+func (superOperations) Keep(*fs.Dirent) bool {
+ return false
+}
+
+// ResetInodeMappings implements MountSourceOperations.ResetInodeMappings.
+func (superOperations) ResetInodeMappings() {}
+
+// SaveInodeMapping implements MountSourceOperations.SaveInodeMapping.
+func (superOperations) SaveInodeMapping(*fs.Inode, string) {}
+
+// Destroy implements MountSourceOperations.Destroy.
+func (superOperations) Destroy() {}
diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go
new file mode 100644
index 000000000..20d29d130
--- /dev/null
+++ b/pkg/sentry/fs/tty/line_discipline.go
@@ -0,0 +1,443 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tty
+
+import (
+ "bytes"
+ "sync"
+ "unicode/utf8"
+
+ "gvisor.googlesource.com/gvisor/pkg/abi/linux"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/arch"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
+ "gvisor.googlesource.com/gvisor/pkg/syserror"
+ "gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+const (
+ // canonMaxBytes is the number of bytes that fit into a single line of
+ // terminal input in canonical mode. This corresponds to N_TTY_BUF_SIZE
+ // in include/linux/tty.h.
+ canonMaxBytes = 4096
+
+ // nonCanonMaxBytes is the maximum number of bytes that can be read at
+ // a time in noncanonical mode.
+ nonCanonMaxBytes = canonMaxBytes - 1
+
+ spacesPerTab = 8
+)
+
+// lineDiscipline dictates how input and output are handled between the
+// pseudoterminal (pty) master and slave. 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:
+//
+// * termios(3)
+// * tty_ioctl(4)
+//
+// This file corresponds most closely to drivers/tty/n_tty.c.
+//
+// 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
+// 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:
+//
+// input from terminal +-------------+ input to process (e.g. bash)
+// +------------------------>| input queue |---------------------------+
+// | (inputQueueWrite) +-------------+ (inputQueueRead) |
+// | |
+// | v
+// masterFD slaveFD
+// ^ |
+// | |
+// | output to terminal +--------------+ output from process |
+// +------------------------| output queue |<--------------------------+
+// (outputQueueRead) +--------------+ (outputQueueWrite)
+//
+// Lock order:
+// termiosMu
+// inQueue.mu
+// outQueue.mu
+//
+// +stateify savable
+type lineDiscipline struct {
+ // sizeMu protects size.
+ sizeMu sync.Mutex `state:"nosave"`
+
+ // size is the terminal size (width and height).
+ size linux.WindowSize
+
+ // inQueue is the input queue of the terminal.
+ inQueue queue
+
+ // outQueue is the output queue of the terminal.
+ outQueue queue
+
+ // termiosMu protects termios.
+ termiosMu sync.RWMutex `state:"nosave"`
+
+ // termios is the terminal configuration used by the lineDiscipline.
+ termios linux.KernelTermios
+
+ // column is the location in a row of the cursor. This is important for
+ // handling certain special characters like backspace.
+ column int
+
+ // 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"`
+}
+
+func newLineDiscipline(termios linux.KernelTermios) *lineDiscipline {
+ ld := lineDiscipline{termios: termios}
+ ld.inQueue.transformer = &inputQueueTransformer{}
+ ld.outQueue.transformer = &outputQueueTransformer{}
+ return &ld
+}
+
+// getTermios gets the linux.Termios for the tty.
+func (l *lineDiscipline) getTermios(ctx context.Context, io usermem.IO, 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,
+ })
+ 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) {
+ 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,
+ })
+ l.termios.FromTermios(t)
+
+ // If canonical mode is turned off, move bytes from inQueue's wait
+ // buffer to its read buffer. Anything already in the read buffer is
+ // now readable.
+ if oldCanonEnabled && !l.termios.LEnabled(linux.ICANON) {
+ l.inQueue.pushWaitBuf(l)
+ l.inQueue.readable = true
+ l.slaveWaiter.Notify(waiter.EventIn)
+ }
+
+ return 0, err
+}
+
+func (l *lineDiscipline) windowSize(ctx context.Context, io usermem.IO, 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,
+ })
+ return err
+}
+
+func (l *lineDiscipline) setWindowSize(ctx context.Context, io usermem.IO, 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,
+ })
+ return err
+}
+
+func (l *lineDiscipline) masterReadiness() waiter.EventMask {
+ // We don't have to lock a termios because the default master termios
+ // is immutable.
+ return l.inQueue.writeReadiness(&linux.MasterTermios) | l.outQueue.readReadiness(&linux.MasterTermios)
+}
+
+func (l *lineDiscipline) slaveReadiness() 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) inputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) {
+ l.termiosMu.RLock()
+ defer l.termiosMu.RUnlock()
+ n, pushed, err := l.inQueue.read(ctx, dst, l)
+ if err != nil {
+ return 0, err
+ }
+ if n > 0 {
+ l.masterWaiter.Notify(waiter.EventOut)
+ if pushed {
+ l.slaveWaiter.Notify(waiter.EventIn)
+ }
+ return n, nil
+ }
+ return 0, syserror.ErrWouldBlock
+}
+
+func (l *lineDiscipline) inputQueueWrite(ctx context.Context, src usermem.IOSequence) (int64, error) {
+ l.termiosMu.RLock()
+ defer l.termiosMu.RUnlock()
+ n, err := l.inQueue.write(ctx, src, l)
+ if err != nil {
+ return 0, err
+ }
+ if n > 0 {
+ l.slaveWaiter.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) outputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) {
+ l.termiosMu.RLock()
+ defer l.termiosMu.RUnlock()
+ n, pushed, err := l.outQueue.read(ctx, dst, l)
+ if err != nil {
+ return 0, err
+ }
+ if n > 0 {
+ l.slaveWaiter.Notify(waiter.EventOut)
+ if pushed {
+ l.masterWaiter.Notify(waiter.EventIn)
+ }
+ return n, nil
+ }
+ return 0, syserror.ErrWouldBlock
+}
+
+func (l *lineDiscipline) outputQueueWrite(ctx context.Context, src usermem.IOSequence) (int64, error) {
+ l.termiosMu.RLock()
+ defer l.termiosMu.RUnlock()
+ n, err := l.outQueue.write(ctx, src, l)
+ if err != nil {
+ return 0, err
+ }
+ if n > 0 {
+ l.masterWaiter.Notify(waiter.EventIn)
+ return n, nil
+ }
+ return 0, syserror.ErrWouldBlock
+}
+
+// transformer is a helper interface to make it easier to stateify queue.
+type transformer interface {
+ // transform functions require queue's mutex to be held.
+ transform(*lineDiscipline, *queue, []byte) int
+}
+
+// outputQueueTransformer implements transformer. It performs line discipline
+// transformations on the output queue.
+//
+// +stateify savable
+type outputQueueTransformer struct{}
+
+// transform does output processing for one end of the pty. See
+// drivers/tty/n_tty.c:do_output_char for an analogous kernel function.
+//
+// Preconditions:
+// * l.termiosMu must be held for reading.
+// * q.mu must be held.
+func (*outputQueueTransformer) transform(l *lineDiscipline, q *queue, buf []byte) int {
+ // transformOutput is effectively always in noncanonical mode, as the
+ // master termios never has ICANON set.
+
+ if !l.termios.OEnabled(linux.OPOST) {
+ q.readBuf = append(q.readBuf, buf...)
+ if len(q.readBuf) > 0 {
+ q.readable = true
+ }
+ return len(buf)
+ }
+
+ var ret int
+ for len(buf) > 0 {
+ size := l.peek(buf)
+ cBytes := append([]byte{}, buf[:size]...)
+ ret += size
+ buf = buf[size:]
+ // We're guaranteed that cBytes has at least one element.
+ switch cBytes[0] {
+ case '\n':
+ if l.termios.OEnabled(linux.ONLRET) {
+ l.column = 0
+ }
+ if l.termios.OEnabled(linux.ONLCR) {
+ q.readBuf = append(q.readBuf, '\r', '\n')
+ continue
+ }
+ case '\r':
+ if l.termios.OEnabled(linux.ONOCR) && l.column == 0 {
+ continue
+ }
+ if l.termios.OEnabled(linux.OCRNL) {
+ cBytes[0] = '\n'
+ if l.termios.OEnabled(linux.ONLRET) {
+ l.column = 0
+ }
+ break
+ }
+ l.column = 0
+ case '\t':
+ spaces := spacesPerTab - l.column%spacesPerTab
+ if l.termios.OutputFlags&linux.TABDLY == linux.XTABS {
+ l.column += spaces
+ q.readBuf = append(q.readBuf, bytes.Repeat([]byte{' '}, spacesPerTab)...)
+ continue
+ }
+ l.column += spaces
+ case '\b':
+ if l.column > 0 {
+ l.column--
+ }
+ default:
+ l.column++
+ }
+ q.readBuf = append(q.readBuf, cBytes...)
+ }
+ if len(q.readBuf) > 0 {
+ q.readable = true
+ }
+ return ret
+}
+
+// inputQueueTransformer implements transformer. It performs line discipline
+// transformations on the input queue.
+//
+// +stateify savable
+type inputQueueTransformer struct{}
+
+// transform does input processing for one end of the pty. Characters read are
+// transformed according to flags set in the termios struct. See
+// drivers/tty/n_tty.c:n_tty_receive_char_special for an analogous kernel
+// function.
+//
+// Preconditions:
+// * l.termiosMu must be held for reading.
+// * q.mu must be held.
+func (*inputQueueTransformer) transform(l *lineDiscipline, q *queue, buf []byte) int {
+ // If there's a line waiting to be read in canonical mode, don't write
+ // anything else to the read buffer.
+ if l.termios.LEnabled(linux.ICANON) && q.readable {
+ return 0
+ }
+
+ maxBytes := nonCanonMaxBytes
+ if l.termios.LEnabled(linux.ICANON) {
+ maxBytes = canonMaxBytes
+ }
+
+ var ret int
+ for len(buf) > 0 && len(q.readBuf) < canonMaxBytes {
+ size := l.peek(buf)
+ cBytes := append([]byte{}, buf[:size]...)
+ // We're guaranteed that cBytes has at least one element.
+ switch cBytes[0] {
+ case '\r':
+ if l.termios.IEnabled(linux.IGNCR) {
+ buf = buf[size:]
+ ret += size
+ continue
+ }
+ if l.termios.IEnabled(linux.ICRNL) {
+ cBytes[0] = '\n'
+ }
+ case '\n':
+ if l.termios.IEnabled(linux.INLCR) {
+ cBytes[0] = '\r'
+ }
+ }
+
+ // In canonical mode, we discard non-terminating characters
+ // after the first 4095.
+ if l.shouldDiscard(q, cBytes) {
+ buf = buf[size:]
+ ret += size
+ continue
+ }
+
+ // Stop if the buffer would be overfilled.
+ if len(q.readBuf)+size > maxBytes {
+ break
+ }
+ buf = buf[size:]
+ ret += size
+
+ // If we get EOF, make the buffer available for reading.
+ if l.termios.LEnabled(linux.ICANON) && l.termios.IsEOF(cBytes[0]) {
+ q.readable = true
+ break
+ }
+
+ q.readBuf = append(q.readBuf, cBytes...)
+
+ // Anything written to the readBuf will have to be echoed.
+ if l.termios.LEnabled(linux.ECHO) {
+ l.outQueue.writeBytes(cBytes, l)
+ l.masterWaiter.Notify(waiter.EventIn)
+ }
+
+ // If we finish a line, make it available for reading.
+ if l.termios.LEnabled(linux.ICANON) && l.termios.IsTerminating(cBytes) {
+ q.readable = true
+ break
+ }
+ }
+
+ // In noncanonical mode, everything is readable.
+ if !l.termios.LEnabled(linux.ICANON) && len(q.readBuf) > 0 {
+ q.readable = true
+ }
+
+ return ret
+}
+
+// shouldDiscard returns whether c should be discarded. In canonical mode, if
+// too many bytes are enqueued, we keep reading input and discarding it until
+// we find a terminating character. Signal/echo processing still occurs.
+//
+// Precondition:
+// * l.termiosMu must be held for reading.
+// * q.mu must be held.
+func (l *lineDiscipline) shouldDiscard(q *queue, cBytes []byte) bool {
+ return l.termios.LEnabled(linux.ICANON) && len(q.readBuf)+len(cBytes) >= canonMaxBytes && !l.termios.IsTerminating(cBytes)
+}
+
+// peek returns the size in bytes of the next character to process. As long as
+// b isn't empty, peek returns a value of at least 1.
+func (l *lineDiscipline) peek(b []byte) int {
+ size := 1
+ // If UTF-8 support is enabled, runes might be multiple bytes.
+ if l.termios.IEnabled(linux.IUTF8) {
+ _, size = utf8.DecodeRune(b)
+ }
+ return size
+}
diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go
new file mode 100644
index 000000000..afdf44cd1
--- /dev/null
+++ b/pkg/sentry/fs/tty/master.go
@@ -0,0 +1,220 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tty
+
+import (
+ "gvisor.googlesource.com/gvisor/pkg/abi/linux"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/arch"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/unimpl"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
+ "gvisor.googlesource.com/gvisor/pkg/syserror"
+ "gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+// masterInodeOperations are the fs.InodeOperations for the master end of the
+// Terminal (ptmx file).
+//
+// +stateify savable
+type masterInodeOperations struct {
+ fsutil.SimpleFileInode
+
+ // d is the containing dir.
+ d *dirInodeOperations
+}
+
+var _ fs.InodeOperations = (*masterInodeOperations)(nil)
+
+// newMasterInode creates an Inode for the master end of a terminal.
+func newMasterInode(ctx context.Context, d *dirInodeOperations, owner fs.FileOwner, p fs.FilePermissions) *fs.Inode {
+ iops := &masterInodeOperations{
+ SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, owner, p, linux.DEVPTS_SUPER_MAGIC),
+ d: d,
+ }
+
+ return fs.NewInode(iops, d.msrc, fs.StableAttr{
+ DeviceID: ptsDevice.DeviceID(),
+ // N.B. Linux always uses inode id 2 for ptmx. See
+ // fs/devpts/inode.c:mknod_ptmx.
+ //
+ // TODO(b/75267214): Since ptsDevice must be shared between
+ // different mounts, we must not assign fixed numbers.
+ InodeID: ptsDevice.NextIno(),
+ Type: fs.CharacterDevice,
+ // See fs/devpts/inode.c:devpts_fill_super.
+ BlockSize: 1024,
+ // The PTY master effectively has two different major/minor
+ // device numbers.
+ //
+ // This one is returned by stat for both opened and unopened
+ // instances of this inode.
+ //
+ // When the inode is opened (GetFile), a new device number is
+ // allocated based on major UNIX98_PTY_MASTER_MAJOR and the tty
+ // index as minor number. However, this device number is only
+ // accessible via ioctl(TIOCGDEV) and /proc/TID/stat.
+ DeviceFileMajor: linux.TTYAUX_MAJOR,
+ DeviceFileMinor: linux.PTMX_MINOR,
+ })
+}
+
+// Release implements fs.InodeOperations.Release.
+func (mi *masterInodeOperations) Release(ctx context.Context) {
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+//
+// It allocates a new terminal.
+func (mi *masterInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+ t, err := mi.d.allocateTerminal(ctx)
+ if err != nil {
+ return nil, err
+ }
+
+ return fs.NewFile(ctx, d, flags, &masterFileOperations{
+ d: mi.d,
+ t: t,
+ }), nil
+}
+
+// masterFileOperations are the fs.FileOperations for the master end of a terminal.
+//
+// +stateify savable
+type masterFileOperations struct {
+ fsutil.FilePipeSeek `state:"nosave"`
+ fsutil.FileNotDirReaddir `state:"nosave"`
+ fsutil.FileNoFsync `state:"nosave"`
+ fsutil.FileNoMMap `state:"nosave"`
+ fsutil.FileNoSplice `state:"nosave"`
+ fsutil.FileNoopFlush `state:"nosave"`
+ fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+
+ // d is the containing dir.
+ d *dirInodeOperations
+
+ // t is the connected Terminal.
+ t *Terminal
+}
+
+var _ fs.FileOperations = (*masterFileOperations)(nil)
+
+// Release implements fs.FileOperations.Release.
+func (mf *masterFileOperations) Release() {
+ mf.d.masterClose(mf.t)
+ mf.t.DecRef()
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (mf *masterFileOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+ mf.t.ld.masterWaiter.EventRegister(e, mask)
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (mf *masterFileOperations) EventUnregister(e *waiter.Entry) {
+ mf.t.ld.masterWaiter.EventUnregister(e)
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (mf *masterFileOperations) Readiness(mask waiter.EventMask) waiter.EventMask {
+ return mf.t.ld.masterReadiness()
+}
+
+// Read implements fs.FileOperations.Read.
+func (mf *masterFileOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) {
+ return mf.t.ld.outputQueueRead(ctx, dst)
+}
+
+// Write implements fs.FileOperations.Write.
+func (mf *masterFileOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) {
+ return mf.t.ld.inputQueueWrite(ctx, src)
+}
+
+// Ioctl implements fs.FileOperations.Ioctl.
+func (mf *masterFileOperations) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+ 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, mf.t.ld.outputQueueReadSize(ctx, io, args)
+ case linux.TCGETS:
+ // N.B. TCGETS on the master actually returns the configuration
+ // of the slave end.
+ return mf.t.ld.getTermios(ctx, io, args)
+ case linux.TCSETS:
+ // N.B. TCSETS on the master actually affects the configuration
+ // of the slave end.
+ return mf.t.ld.setTermios(ctx, io, args)
+ case linux.TCSETSW:
+ // TODO(b/29356795): This should drain the output queue first.
+ return mf.t.ld.setTermios(ctx, io, args)
+ case linux.TIOCGPTN:
+ _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(mf.t.n), usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ 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, mf.t.ld.windowSize(ctx, io, args)
+ case linux.TIOCSWINSZ:
+ return 0, mf.t.ld.setWindowSize(ctx, io, args)
+ default:
+ maybeEmitUnimplementedEvent(ctx, cmd)
+ return 0, syserror.ENOTTY
+ }
+}
+
+// maybeEmitUnimplementedEvent emits unimplemented event if cmd is valid.
+func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) {
+ switch cmd {
+ case linux.TCGETS,
+ linux.TCSETS,
+ linux.TCSETSW,
+ linux.TCSETSF,
+ linux.TIOCGPGRP,
+ linux.TIOCSPGRP,
+ linux.TIOCGWINSZ,
+ linux.TIOCSWINSZ,
+ linux.TIOCSETD,
+ linux.TIOCSBRK,
+ linux.TIOCCBRK,
+ linux.TCSBRK,
+ linux.TCSBRKP,
+ linux.TIOCSTI,
+ linux.TIOCCONS,
+ linux.FIONBIO,
+ linux.TIOCEXCL,
+ linux.TIOCNXCL,
+ linux.TIOCGEXCL,
+ linux.TIOCNOTTY,
+ linux.TIOCSCTTY,
+ linux.TIOCGSID,
+ linux.TIOCGETD,
+ linux.TIOCVHANGUP,
+ linux.TIOCGDEV,
+ linux.TIOCMGET,
+ linux.TIOCMSET,
+ linux.TIOCMBIC,
+ linux.TIOCMBIS,
+ linux.TIOCGICOUNT,
+ linux.TCFLSH,
+ linux.TIOCSSERIAL,
+ linux.TIOCGPTPEER:
+
+ unimpl.EmitUnimplementedEvent(ctx)
+ }
+}
diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go
new file mode 100644
index 000000000..11fb92be3
--- /dev/null
+++ b/pkg/sentry/fs/tty/queue.go
@@ -0,0 +1,244 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tty
+
+import (
+ "sync"
+
+ "gvisor.googlesource.com/gvisor/pkg/abi/linux"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/arch"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/safemem"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
+ "gvisor.googlesource.com/gvisor/pkg/syserror"
+ "gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+// waitBufMaxBytes is the maximum size of a wait buffer. It is based on
+// TTYB_DEFAULT_MEM_LIMIT.
+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
+// 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
+// readable is true.
+//
+// +stateify savable
+type queue struct {
+ // mu protects everything in queue.
+ mu sync.Mutex `state:"nosave"`
+
+ // readBuf is buffer of data ready to be read when readable is true.
+ // This data has been processed.
+ readBuf []byte
+
+ // waitBuf contains data that can't fit into readBuf. It is put here
+ // until it can be loaded into the read buffer. waitBuf contains data
+ // that hasn't been processed.
+ waitBuf [][]byte
+ waitBufLen uint64
+
+ // readable indicates whether the read buffer can be read from. In
+ // canonical mode, there can be an unterminated line in the read buffer,
+ // so readable must be checked.
+ readable bool
+
+ // transform is the the queue's function for transforming bytes
+ // entering the queue. For example, transform might convert all '\r's
+ // entering the queue to '\n's.
+ transformer
+}
+
+// readReadiness returns whether q is ready to be read from.
+func (q *queue) readReadiness(t *linux.KernelTermios) waiter.EventMask {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+ if len(q.readBuf) > 0 && q.readable {
+ return waiter.EventIn
+ }
+ return waiter.EventMask(0)
+}
+
+// writeReadiness returns whether q is ready to be written to.
+func (q *queue) writeReadiness(t *linux.KernelTermios) waiter.EventMask {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+ if q.waitBufLen < waitBufMaxBytes {
+ return waiter.EventOut
+ }
+ return waiter.EventMask(0)
+}
+
+// readableSize writes the number of readable bytes to userspace.
+func (q *queue) readableSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+ var size int32
+ if q.readable {
+ size = int32(len(q.readBuf))
+ }
+
+ _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), size, usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ return err
+
+}
+
+// read reads from q to userspace. It returns the number of bytes read as well
+// as whether the read caused more readable data to become available (whether
+// data was pushed from the wait buffer to the read buffer).
+//
+// Preconditions:
+// * l.termiosMu must be held for reading.
+func (q *queue) read(ctx context.Context, dst usermem.IOSequence, l *lineDiscipline) (int64, bool, error) {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+
+ if !q.readable {
+ return 0, false, syserror.ErrWouldBlock
+ }
+
+ if dst.NumBytes() > canonMaxBytes {
+ dst = dst.TakeFirst(canonMaxBytes)
+ }
+
+ n, err := dst.CopyOutFrom(ctx, safemem.ReaderFunc(func(dst safemem.BlockSeq) (uint64, error) {
+ src := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(q.readBuf))
+ n, err := safemem.CopySeq(dst, src)
+ if err != nil {
+ return 0, err
+ }
+ q.readBuf = q.readBuf[n:]
+
+ // If we read everything, this queue is no longer readable.
+ if len(q.readBuf) == 0 {
+ q.readable = false
+ }
+
+ return n, nil
+ }))
+ if err != nil {
+ return 0, false, err
+ }
+
+ // Move data from the queue's wait buffer to its read buffer.
+ nPushed := q.pushWaitBufLocked(l)
+
+ return int64(n), nPushed > 0, nil
+}
+
+// write writes to q from userspace.
+//
+// Preconditions:
+// * l.termiosMu must be held for reading.
+func (q *queue) write(ctx context.Context, src usermem.IOSequence, l *lineDiscipline) (int64, error) {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+
+ // Copy data into the wait buffer.
+ n, err := src.CopyInTo(ctx, safemem.WriterFunc(func(src safemem.BlockSeq) (uint64, error) {
+ copyLen := src.NumBytes()
+ room := waitBufMaxBytes - q.waitBufLen
+ // If out of room, return EAGAIN.
+ if room == 0 && copyLen > 0 {
+ return 0, syserror.ErrWouldBlock
+ }
+ // Cap the size of the wait buffer.
+ if copyLen > room {
+ copyLen = room
+ src = src.TakeFirst64(room)
+ }
+ buf := make([]byte, copyLen)
+
+ // Copy the data into the wait buffer.
+ dst := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf))
+ n, err := safemem.CopySeq(dst, src)
+ if err != nil {
+ return 0, err
+ }
+ q.waitBufAppend(buf)
+
+ return n, nil
+ }))
+ if err != nil {
+ return 0, err
+ }
+
+ // Push data from the wait to the read buffer.
+ q.pushWaitBufLocked(l)
+
+ return n, nil
+}
+
+// writeBytes writes to q from b.
+//
+// Preconditions:
+// * l.termiosMu must be held for reading.
+func (q *queue) writeBytes(b []byte, l *lineDiscipline) {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+
+ // Write to the wait buffer.
+ q.waitBufAppend(b)
+ q.pushWaitBufLocked(l)
+}
+
+// pushWaitBuf fills the queue's read buffer with data from the wait buffer.
+//
+// Preconditions:
+// * l.termiosMu must be held for reading.
+func (q *queue) pushWaitBuf(l *lineDiscipline) int {
+ q.mu.Lock()
+ defer q.mu.Unlock()
+ return q.pushWaitBufLocked(l)
+}
+
+// Preconditions:
+// * l.termiosMu must be held for reading.
+// * q.mu must be locked.
+func (q *queue) pushWaitBufLocked(l *lineDiscipline) int {
+ if q.waitBufLen == 0 {
+ return 0
+ }
+
+ // Move data from the wait to the read buffer.
+ var total int
+ var i int
+ for i = 0; i < len(q.waitBuf); i++ {
+ n := q.transform(l, q, q.waitBuf[i])
+ total += n
+ if n != len(q.waitBuf[i]) {
+ // The read buffer filled up without consuming the
+ // entire buffer.
+ q.waitBuf[i] = q.waitBuf[i][n:]
+ break
+ }
+ }
+
+ // Update wait buffer based on consumed data.
+ q.waitBuf = q.waitBuf[i:]
+ q.waitBufLen -= uint64(total)
+
+ return total
+}
+
+// Precondition: q.mu must be locked.
+func (q *queue) waitBufAppend(b []byte) {
+ q.waitBuf = append(q.waitBuf, b)
+ q.waitBufLen += uint64(len(b))
+}
diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go
new file mode 100644
index 000000000..2abf32e57
--- /dev/null
+++ b/pkg/sentry/fs/tty/slave.go
@@ -0,0 +1,162 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tty
+
+import (
+ "gvisor.googlesource.com/gvisor/pkg/abi/linux"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/arch"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/usermem"
+ "gvisor.googlesource.com/gvisor/pkg/syserror"
+ "gvisor.googlesource.com/gvisor/pkg/waiter"
+)
+
+// slaveInodeOperations are the fs.InodeOperations for the slave end of the
+// Terminal (pts file).
+//
+// +stateify savable
+type slaveInodeOperations struct {
+ fsutil.SimpleFileInode
+
+ // d is the containing dir.
+ d *dirInodeOperations
+
+ // t is the connected Terminal.
+ t *Terminal
+}
+
+var _ fs.InodeOperations = (*slaveInodeOperations)(nil)
+
+// newSlaveInode creates an fs.Inode for the slave end of a terminal.
+//
+// newSlaveInode takes ownership of t.
+func newSlaveInode(ctx context.Context, d *dirInodeOperations, t *Terminal, owner fs.FileOwner, p fs.FilePermissions) *fs.Inode {
+ iops := &slaveInodeOperations{
+ SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, owner, p, linux.DEVPTS_SUPER_MAGIC),
+ d: d,
+ t: t,
+ }
+
+ return fs.NewInode(iops, d.msrc, fs.StableAttr{
+ DeviceID: ptsDevice.DeviceID(),
+ // N.B. Linux always uses inode id = tty index + 3. See
+ // fs/devpts/inode.c:devpts_pty_new.
+ //
+ // TODO(b/75267214): Since ptsDevice must be shared between
+ // different mounts, we must not assign fixed numbers.
+ InodeID: ptsDevice.NextIno(),
+ Type: fs.CharacterDevice,
+ // See fs/devpts/inode.c:devpts_fill_super.
+ BlockSize: 1024,
+ DeviceFileMajor: linux.UNIX98_PTY_SLAVE_MAJOR,
+ DeviceFileMinor: t.n,
+ })
+}
+
+// Release implements fs.InodeOperations.Release.
+func (si *slaveInodeOperations) Release(ctx context.Context) {
+ si.t.DecRef()
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+//
+// This may race with destruction of the terminal. If the terminal is gone, it
+// returns ENOENT.
+func (si *slaveInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+ return fs.NewFile(ctx, d, flags, &slaveFileOperations{si: si}), nil
+}
+
+// slaveFileOperations are the fs.FileOperations for the slave end of a terminal.
+//
+// +stateify savable
+type slaveFileOperations struct {
+ fsutil.FilePipeSeek `state:"nosave"`
+ fsutil.FileNotDirReaddir `state:"nosave"`
+ fsutil.FileNoFsync `state:"nosave"`
+ fsutil.FileNoMMap `state:"nosave"`
+ fsutil.FileNoSplice `state:"nosave"`
+ fsutil.FileNoopFlush `state:"nosave"`
+ fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+
+ // si is the inode operations.
+ si *slaveInodeOperations
+}
+
+var _ fs.FileOperations = (*slaveFileOperations)(nil)
+
+// Release implements fs.FileOperations.Release.
+func (sf *slaveFileOperations) Release() {
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (sf *slaveFileOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+ sf.si.t.ld.slaveWaiter.EventRegister(e, mask)
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (sf *slaveFileOperations) EventUnregister(e *waiter.Entry) {
+ sf.si.t.ld.slaveWaiter.EventUnregister(e)
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (sf *slaveFileOperations) Readiness(mask waiter.EventMask) waiter.EventMask {
+ return sf.si.t.ld.slaveReadiness()
+}
+
+// Read implements fs.FileOperations.Read.
+func (sf *slaveFileOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) {
+ return sf.si.t.ld.inputQueueRead(ctx, dst)
+}
+
+// Write implements fs.FileOperations.Write.
+func (sf *slaveFileOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) {
+ return sf.si.t.ld.outputQueueWrite(ctx, src)
+}
+
+// Ioctl implements fs.FileOperations.Ioctl.
+func (sf *slaveFileOperations) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+ 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, sf.si.t.ld.inputQueueReadSize(ctx, io, args)
+ case linux.TCGETS:
+ return sf.si.t.ld.getTermios(ctx, io, args)
+ case linux.TCSETS:
+ return sf.si.t.ld.setTermios(ctx, io, args)
+ case linux.TCSETSW:
+ // TODO(b/29356795): This should drain the output queue first.
+ return sf.si.t.ld.setTermios(ctx, io, args)
+ case linux.TIOCGPTN:
+ _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(sf.si.t.n), usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ return 0, err
+ case linux.TIOCGWINSZ:
+ return 0, sf.si.t.ld.windowSize(ctx, io, args)
+ case linux.TIOCSWINSZ:
+ return 0, sf.si.t.ld.setWindowSize(ctx, io, args)
+ case linux.TIOCSCTTY:
+ // Make the given terminal the controlling terminal of the
+ // calling process.
+ // TODO(b/129283598): Implement once we have support for job
+ // control.
+ return 0, nil
+ default:
+ maybeEmitUnimplementedEvent(ctx, cmd)
+ return 0, syserror.ENOTTY
+ }
+}
diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go
new file mode 100644
index 000000000..2b4160ba5
--- /dev/null
+++ b/pkg/sentry/fs/tty/terminal.go
@@ -0,0 +1,46 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tty
+
+import (
+ "gvisor.googlesource.com/gvisor/pkg/abi/linux"
+ "gvisor.googlesource.com/gvisor/pkg/refs"
+ "gvisor.googlesource.com/gvisor/pkg/sentry/context"
+)
+
+// Terminal is a pseudoterminal.
+//
+// +stateify savable
+type Terminal struct {
+ refs.AtomicRefCount
+
+ // n is the terminal index.
+ n uint32
+
+ // d is the containing directory.
+ d *dirInodeOperations
+
+ // ld is the line discipline of the terminal.
+ ld *lineDiscipline
+}
+
+func newTerminal(ctx context.Context, d *dirInodeOperations, n uint32) *Terminal {
+ termios := linux.DefaultSlaveTermios
+ return &Terminal{
+ d: d,
+ n: n,
+ ld: newLineDiscipline(termios),
+ }
+}
diff --git a/pkg/sentry/fs/tty/tty_state_autogen.go b/pkg/sentry/fs/tty/tty_state_autogen.go
new file mode 100755
index 000000000..303e2eef2
--- /dev/null
+++ b/pkg/sentry/fs/tty/tty_state_autogen.go
@@ -0,0 +1,202 @@
+// automatically generated by stateify.
+
+package tty
+
+import (
+ "gvisor.googlesource.com/gvisor/pkg/state"
+)
+
+func (x *dirInodeOperations) beforeSave() {}
+func (x *dirInodeOperations) save(m state.Map) {
+ x.beforeSave()
+ m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes)
+ m.Save("msrc", &x.msrc)
+ m.Save("master", &x.master)
+ m.Save("slaves", &x.slaves)
+ m.Save("dentryMap", &x.dentryMap)
+ m.Save("next", &x.next)
+}
+
+func (x *dirInodeOperations) afterLoad() {}
+func (x *dirInodeOperations) load(m state.Map) {
+ m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes)
+ m.Load("msrc", &x.msrc)
+ m.Load("master", &x.master)
+ m.Load("slaves", &x.slaves)
+ m.Load("dentryMap", &x.dentryMap)
+ m.Load("next", &x.next)
+}
+
+func (x *dirFileOperations) beforeSave() {}
+func (x *dirFileOperations) save(m state.Map) {
+ x.beforeSave()
+ m.Save("di", &x.di)
+ m.Save("dirCursor", &x.dirCursor)
+}
+
+func (x *dirFileOperations) afterLoad() {}
+func (x *dirFileOperations) load(m state.Map) {
+ m.Load("di", &x.di)
+ m.Load("dirCursor", &x.dirCursor)
+}
+
+func (x *filesystem) beforeSave() {}
+func (x *filesystem) save(m state.Map) {
+ x.beforeSave()
+}
+
+func (x *filesystem) afterLoad() {}
+func (x *filesystem) load(m state.Map) {
+}
+
+func (x *superOperations) beforeSave() {}
+func (x *superOperations) save(m state.Map) {
+ x.beforeSave()
+}
+
+func (x *superOperations) afterLoad() {}
+func (x *superOperations) load(m state.Map) {
+}
+
+func (x *lineDiscipline) beforeSave() {}
+func (x *lineDiscipline) save(m state.Map) {
+ x.beforeSave()
+ if !state.IsZeroValue(x.masterWaiter) { m.Failf("masterWaiter is %v, expected zero", x.masterWaiter) }
+ if !state.IsZeroValue(x.slaveWaiter) { m.Failf("slaveWaiter is %v, expected zero", x.slaveWaiter) }
+ m.Save("size", &x.size)
+ m.Save("inQueue", &x.inQueue)
+ m.Save("outQueue", &x.outQueue)
+ m.Save("termios", &x.termios)
+ m.Save("column", &x.column)
+}
+
+func (x *lineDiscipline) afterLoad() {}
+func (x *lineDiscipline) load(m state.Map) {
+ m.Load("size", &x.size)
+ m.Load("inQueue", &x.inQueue)
+ m.Load("outQueue", &x.outQueue)
+ m.Load("termios", &x.termios)
+ m.Load("column", &x.column)
+}
+
+func (x *outputQueueTransformer) beforeSave() {}
+func (x *outputQueueTransformer) save(m state.Map) {
+ x.beforeSave()
+}
+
+func (x *outputQueueTransformer) afterLoad() {}
+func (x *outputQueueTransformer) load(m state.Map) {
+}
+
+func (x *inputQueueTransformer) beforeSave() {}
+func (x *inputQueueTransformer) save(m state.Map) {
+ x.beforeSave()
+}
+
+func (x *inputQueueTransformer) afterLoad() {}
+func (x *inputQueueTransformer) load(m state.Map) {
+}
+
+func (x *masterInodeOperations) beforeSave() {}
+func (x *masterInodeOperations) save(m state.Map) {
+ x.beforeSave()
+ m.Save("SimpleFileInode", &x.SimpleFileInode)
+ m.Save("d", &x.d)
+}
+
+func (x *masterInodeOperations) afterLoad() {}
+func (x *masterInodeOperations) load(m state.Map) {
+ m.Load("SimpleFileInode", &x.SimpleFileInode)
+ m.Load("d", &x.d)
+}
+
+func (x *masterFileOperations) beforeSave() {}
+func (x *masterFileOperations) save(m state.Map) {
+ x.beforeSave()
+ m.Save("d", &x.d)
+ m.Save("t", &x.t)
+}
+
+func (x *masterFileOperations) afterLoad() {}
+func (x *masterFileOperations) load(m state.Map) {
+ m.Load("d", &x.d)
+ m.Load("t", &x.t)
+}
+
+func (x *queue) beforeSave() {}
+func (x *queue) save(m state.Map) {
+ x.beforeSave()
+ m.Save("readBuf", &x.readBuf)
+ m.Save("waitBuf", &x.waitBuf)
+ m.Save("waitBufLen", &x.waitBufLen)
+ m.Save("readable", &x.readable)
+ m.Save("transformer", &x.transformer)
+}
+
+func (x *queue) afterLoad() {}
+func (x *queue) load(m state.Map) {
+ m.Load("readBuf", &x.readBuf)
+ m.Load("waitBuf", &x.waitBuf)
+ m.Load("waitBufLen", &x.waitBufLen)
+ m.Load("readable", &x.readable)
+ m.Load("transformer", &x.transformer)
+}
+
+func (x *slaveInodeOperations) beforeSave() {}
+func (x *slaveInodeOperations) save(m state.Map) {
+ x.beforeSave()
+ m.Save("SimpleFileInode", &x.SimpleFileInode)
+ m.Save("d", &x.d)
+ m.Save("t", &x.t)
+}
+
+func (x *slaveInodeOperations) afterLoad() {}
+func (x *slaveInodeOperations) load(m state.Map) {
+ m.Load("SimpleFileInode", &x.SimpleFileInode)
+ m.Load("d", &x.d)
+ m.Load("t", &x.t)
+}
+
+func (x *slaveFileOperations) beforeSave() {}
+func (x *slaveFileOperations) save(m state.Map) {
+ x.beforeSave()
+ m.Save("si", &x.si)
+}
+
+func (x *slaveFileOperations) afterLoad() {}
+func (x *slaveFileOperations) load(m state.Map) {
+ m.Load("si", &x.si)
+}
+
+func (x *Terminal) beforeSave() {}
+func (x *Terminal) save(m state.Map) {
+ x.beforeSave()
+ m.Save("AtomicRefCount", &x.AtomicRefCount)
+ m.Save("n", &x.n)
+ m.Save("d", &x.d)
+ m.Save("ld", &x.ld)
+}
+
+func (x *Terminal) afterLoad() {}
+func (x *Terminal) load(m state.Map) {
+ m.Load("AtomicRefCount", &x.AtomicRefCount)
+ m.Load("n", &x.n)
+ m.Load("d", &x.d)
+ m.Load("ld", &x.ld)
+}
+
+func init() {
+ state.Register("tty.dirInodeOperations", (*dirInodeOperations)(nil), state.Fns{Save: (*dirInodeOperations).save, Load: (*dirInodeOperations).load})
+ state.Register("tty.dirFileOperations", (*dirFileOperations)(nil), state.Fns{Save: (*dirFileOperations).save, Load: (*dirFileOperations).load})
+ state.Register("tty.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load})
+ state.Register("tty.superOperations", (*superOperations)(nil), state.Fns{Save: (*superOperations).save, Load: (*superOperations).load})
+ state.Register("tty.lineDiscipline", (*lineDiscipline)(nil), state.Fns{Save: (*lineDiscipline).save, Load: (*lineDiscipline).load})
+ state.Register("tty.outputQueueTransformer", (*outputQueueTransformer)(nil), state.Fns{Save: (*outputQueueTransformer).save, Load: (*outputQueueTransformer).load})
+ state.Register("tty.inputQueueTransformer", (*inputQueueTransformer)(nil), state.Fns{Save: (*inputQueueTransformer).save, Load: (*inputQueueTransformer).load})
+ state.Register("tty.masterInodeOperations", (*masterInodeOperations)(nil), state.Fns{Save: (*masterInodeOperations).save, Load: (*masterInodeOperations).load})
+ state.Register("tty.masterFileOperations", (*masterFileOperations)(nil), state.Fns{Save: (*masterFileOperations).save, Load: (*masterFileOperations).load})
+ state.Register("tty.queue", (*queue)(nil), state.Fns{Save: (*queue).save, Load: (*queue).load})
+ state.Register("tty.slaveInodeOperations", (*slaveInodeOperations)(nil), state.Fns{Save: (*slaveInodeOperations).save, Load: (*slaveInodeOperations).load})
+ state.Register("tty.slaveFileOperations", (*slaveFileOperations)(nil), state.Fns{Save: (*slaveFileOperations).save, Load: (*slaveFileOperations).load})
+ state.Register("tty.Terminal", (*Terminal)(nil), state.Fns{Save: (*Terminal).save, Load: (*Terminal).load})
+}