summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/kernel')
-rw-r--r--pkg/sentry/kernel/BUILD1
-rw-r--r--pkg/sentry/kernel/kernel.go157
-rw-r--r--pkg/sentry/kernel/sessions.go12
-rw-r--r--pkg/sentry/kernel/task_block.go12
-rw-r--r--pkg/sentry/kernel/task_context.go11
-rw-r--r--pkg/sentry/kernel/task_start.go3
-rw-r--r--pkg/sentry/kernel/thread_group.go179
-rw-r--r--pkg/sentry/kernel/tty.go28
8 files changed, 312 insertions, 91 deletions
diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index e61d39c82..41bee9a22 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -144,6 +144,7 @@ go_library(
"threads.go",
"timekeeper.go",
"timekeeper_state.go",
+ "tty.go",
"uts_namespace.go",
"vdso.go",
"version.go",
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 4c2d48e65..8c1f79ab5 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -112,11 +112,6 @@ type Kernel struct {
rootIPCNamespace *IPCNamespace
rootAbstractSocketNamespace *AbstractSocketNamespace
- // mounts holds the state of the virtual filesystem. mounts is initially
- // nil, and must be set by calling Kernel.SetRootMountNamespace before
- // Kernel.CreateProcess can succeed.
- mounts *fs.MountNamespace
-
// futexes is the "root" futex.Manager, from which all others are forked.
// This is necessary to ensure that shared futexes are coherent across all
// tasks, including those created by CreateProcess.
@@ -197,6 +192,15 @@ type Kernel struct {
// caches. Not all caches use it, only the caches that use host resources use
// the limiter. It may be nil if disabled.
DirentCacheLimiter *fs.DirentCacheLimiter
+
+ // unimplementedSyscallEmitterOnce is used in the initialization of
+ // unimplementedSyscallEmitter.
+ unimplementedSyscallEmitterOnce sync.Once `state:"nosave"`
+
+ // unimplementedSyscallEmitter is used to emit unimplemented syscall
+ // events. This is initialized lazily on the first unimplemented
+ // syscall.
+ unimplementedSyscallEmitter eventchannel.Emitter `state:"nosave"`
}
// InitKernelArgs holds arguments to Init.
@@ -290,7 +294,6 @@ func (k *Kernel) Init(args InitKernelArgs) error {
k.monotonicClock = &timekeeperClock{tk: args.Timekeeper, c: sentrytime.Monotonic}
k.futexes = futex.NewManager()
k.netlinkPorts = port.New()
-
return nil
}
@@ -384,11 +387,7 @@ func (k *Kernel) SaveTo(w io.Writer) error {
// flushMountSourceRefs flushes the MountSources for all mounted filesystems
// and open FDs.
func (k *Kernel) flushMountSourceRefs() error {
- // Flush all mount sources for currently mounted filesystems in the
- // root mount namespace.
- k.mounts.FlushMountSourceRefs()
-
- // Some tasks may have other mount namespaces; flush those as well.
+ // Flush all mount sources for currently mounted filesystems in each task.
flushed := make(map[*fs.MountNamespace]struct{})
k.tasks.mu.RLock()
k.tasks.forEachThreadGroupLocked(func(tg *ThreadGroup) {
@@ -497,7 +496,7 @@ func (ts *TaskSet) unregisterEpollWaiters() {
}
// LoadFrom returns a new Kernel loaded from args.
-func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack) error {
+func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) error {
loadStart := time.Now()
k.networkStack = net
@@ -541,6 +540,11 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack) error {
log.Infof("Overall load took [%s]", time.Since(loadStart))
+ k.Timekeeper().SetClocks(clocks)
+ if net != nil {
+ net.Resume()
+ }
+
// Ensure that all pending asynchronous work is complete:
// - namedpipe opening
// - inode file opening
@@ -550,7 +554,7 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack) error {
tcpip.AsyncLoading.Wait()
- log.Infof("Overall load took [%s]", time.Since(loadStart))
+ log.Infof("Overall load took [%s] after async work", time.Since(loadStart))
// Applications may size per-cpu structures based on k.applicationCores, so
// it can't change across save/restore. When we are virtualizing CPU
@@ -565,16 +569,6 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack) error {
return nil
}
-// Destroy releases resources owned by k.
-//
-// Preconditions: There must be no task goroutines running in k.
-func (k *Kernel) Destroy() {
- if k.mounts != nil {
- k.mounts.DecRef()
- k.mounts = nil
- }
-}
-
// UniqueID returns a unique identifier.
func (k *Kernel) UniqueID() uint64 {
id := atomic.AddUint64(&k.uniqueID, 1)
@@ -586,11 +580,17 @@ func (k *Kernel) UniqueID() uint64 {
// CreateProcessArgs holds arguments to kernel.CreateProcess.
type CreateProcessArgs struct {
- // Filename is the filename to load.
+ // Filename is the filename to load as the init binary.
//
- // If this is provided as "", then the file will be guessed via Argv[0].
+ // If this is provided as "", File will be checked, then the file will be
+ // guessed via Argv[0].
Filename string
+ // File is a passed host FD pointing to a file to load as the init binary.
+ //
+ // This is checked if and only if Filename is "".
+ File *fs.File
+
// Argvv is a list of arguments.
Argv []string
@@ -632,19 +632,12 @@ type CreateProcessArgs struct {
AbstractSocketNamespace *AbstractSocketNamespace
// MountNamespace optionally contains the mount namespace for this
- // process. If nil, the kernel's mount namespace is used.
+ // process. If nil, the init process's mount namespace is used.
//
// Anyone setting MountNamespace must donate a reference (i.e.
// increment it).
MountNamespace *fs.MountNamespace
- // Root optionally contains the dirent that serves as the root for the
- // process. If nil, the mount namespace's root is used as the process'
- // root.
- //
- // Anyone setting Root must donate a reference (i.e. increment it).
- Root *fs.Dirent
-
// ContainerID is the container that the process belongs to.
ContainerID string
}
@@ -682,16 +675,10 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} {
case auth.CtxCredentials:
return ctx.args.Credentials
case fs.CtxRoot:
- if ctx.args.Root != nil {
- // Take a reference on the root dirent that will be
- // given to the caller.
- ctx.args.Root.IncRef()
- return ctx.args.Root
- }
- if ctx.k.mounts != nil {
- // MountNamespace.Root() will take a reference on the
- // root dirent for us.
- return ctx.k.mounts.Root()
+ if ctx.args.MountNamespace != nil {
+ // MountNamespace.Root() will take a reference on the root
+ // dirent for us.
+ return ctx.args.MountNamespace.Root()
}
return nil
case fs.CtxDirentCacheLimiter:
@@ -735,30 +722,18 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
defer k.extMu.Unlock()
log.Infof("EXEC: %v", args.Argv)
- if k.mounts == nil {
- return nil, 0, fmt.Errorf("no kernel MountNamespace")
- }
-
// Grab the mount namespace.
mounts := args.MountNamespace
if mounts == nil {
- // If no MountNamespace was configured, then use the kernel's
- // root mount namespace, with an extra reference that will be
- // donated to the task.
- mounts = k.mounts
+ mounts = k.GlobalInit().Leader().MountNamespace()
mounts.IncRef()
}
tg := k.newThreadGroup(mounts, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits, k.monotonicClock)
ctx := args.NewContext(k)
- // Grab the root directory.
- root := args.Root
- if root == nil {
- // If no Root was configured, then get it from the
- // MountNamespace.
- root = mounts.Root()
- }
+ // Get the root directory from the MountNamespace.
+ root := mounts.Root()
// The call to newFSContext below will take a reference on root, so we
// don't need to hold this one.
defer root.DecRef()
@@ -768,15 +743,23 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
wd := root // Default.
if args.WorkingDirectory != "" {
var err error
- wd, err = k.mounts.FindInode(ctx, root, nil, args.WorkingDirectory, &remainingTraversals)
+ wd, err = mounts.FindInode(ctx, root, nil, args.WorkingDirectory, &remainingTraversals)
if err != nil {
return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err)
}
defer wd.DecRef()
}
- if args.Filename == "" {
- // Was anything provided?
+ // Check which file to start from.
+ switch {
+ case args.Filename != "":
+ // If a filename is given, take that.
+ // Set File to nil so we resolve the path in LoadTaskImage.
+ args.File = nil
+ case args.File != nil:
+ // If File is set, take the File provided directly.
+ default:
+ // Otherwise look at Argv and see if the first argument is a valid path.
if len(args.Argv) == 0 {
return nil, 0, fmt.Errorf("no filename or command provided")
}
@@ -788,7 +771,8 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
// Create a fresh task context.
remainingTraversals = uint(args.MaxSymlinkTraversals)
- tc, se := k.LoadTaskImage(ctx, k.mounts, root, wd, &remainingTraversals, args.Filename, args.Argv, args.Envv, k.featureSet)
+
+ tc, se := k.LoadTaskImage(ctx, mounts, root, wd, &remainingTraversals, args.Filename, args.File, args.Argv, args.Envv, k.featureSet)
if se != nil {
return nil, 0, errors.New(se.String())
}
@@ -1032,20 +1016,6 @@ func (k *Kernel) RootAbstractSocketNamespace() *AbstractSocketNamespace {
return k.rootAbstractSocketNamespace
}
-// RootMountNamespace returns the MountNamespace.
-func (k *Kernel) RootMountNamespace() *fs.MountNamespace {
- k.extMu.Lock()
- defer k.extMu.Unlock()
- return k.mounts
-}
-
-// SetRootMountNamespace sets the MountNamespace.
-func (k *Kernel) SetRootMountNamespace(mounts *fs.MountNamespace) {
- k.extMu.Lock()
- defer k.extMu.Unlock()
- k.mounts = mounts
-}
-
// NetworkStack returns the network stack. NetworkStack may return nil if no
// network stack is available.
func (k *Kernel) NetworkStack() inet.Stack {
@@ -1168,16 +1138,6 @@ func (k *Kernel) SupervisorContext() context.Context {
}
}
-// EmitUnimplementedEvent emits an UnimplementedSyscall event via the event
-// channel.
-func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) {
- t := TaskFromContext(ctx)
- eventchannel.Emit(&uspb.UnimplementedSyscall{
- Tid: int32(t.ThreadID()),
- Registers: t.Arch().StateData().Proto(),
- })
-}
-
// SocketEntry represents a socket recorded in Kernel.sockets. It implements
// refs.WeakRefUser for sockets stored in the socket table.
//
@@ -1246,7 +1206,10 @@ func (ctx supervisorContext) Value(key interface{}) interface{} {
// The supervisor context is global root.
return auth.NewRootCredentials(ctx.k.rootUserNamespace)
case fs.CtxRoot:
- return ctx.k.mounts.Root()
+ if ctx.k.globalInit != nil {
+ return ctx.k.globalInit.mounts.Root()
+ }
+ return nil
case fs.CtxDirentCacheLimiter:
return ctx.k.DirentCacheLimiter
case ktime.CtxRealtimeClock:
@@ -1272,3 +1235,23 @@ func (ctx supervisorContext) Value(key interface{}) interface{} {
return nil
}
}
+
+// Rate limits for the number of unimplemented syscall events.
+const (
+ unimplementedSyscallsMaxRate = 100 // events per second
+ unimplementedSyscallBurst = 1000 // events
+)
+
+// EmitUnimplementedEvent emits an UnimplementedSyscall event via the event
+// channel.
+func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) {
+ k.unimplementedSyscallEmitterOnce.Do(func() {
+ k.unimplementedSyscallEmitter = eventchannel.RateLimitedEmitterFrom(eventchannel.DefaultEmitter, unimplementedSyscallsMaxRate, unimplementedSyscallBurst)
+ })
+
+ t := TaskFromContext(ctx)
+ k.unimplementedSyscallEmitter.Emit(&uspb.UnimplementedSyscall{
+ Tid: int32(t.ThreadID()),
+ Registers: t.Arch().StateData().Proto(),
+ })
+}
diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go
index 81fcd8258..e5f297478 100644
--- a/pkg/sentry/kernel/sessions.go
+++ b/pkg/sentry/kernel/sessions.go
@@ -47,6 +47,11 @@ type Session struct {
// The id is immutable.
id SessionID
+ // foreground is the foreground process group.
+ //
+ // This is protected by TaskSet.mu.
+ foreground *ProcessGroup
+
// ProcessGroups is a list of process groups in this Session. This is
// protected by TaskSet.mu.
processGroups processGroupList
@@ -260,12 +265,14 @@ func (pg *ProcessGroup) SendSignal(info *arch.SignalInfo) error {
func (tg *ThreadGroup) CreateSession() error {
tg.pidns.owner.mu.Lock()
defer tg.pidns.owner.mu.Unlock()
+ tg.signalHandlers.mu.Lock()
+ defer tg.signalHandlers.mu.Unlock()
return tg.createSession()
}
// createSession creates a new session for a threadgroup.
//
-// Precondition: callers must hold TaskSet.mu for writing.
+// Precondition: callers must hold TaskSet.mu and the signal mutex for writing.
func (tg *ThreadGroup) createSession() error {
// Get the ID for this thread in the current namespace.
id := tg.pidns.tgids[tg]
@@ -346,6 +353,9 @@ func (tg *ThreadGroup) createSession() error {
ns.processGroups[ProcessGroupID(local)] = pg
}
+ // Disconnect from the controlling terminal.
+ tg.tty = nil
+
return nil
}
diff --git a/pkg/sentry/kernel/task_block.go b/pkg/sentry/kernel/task_block.go
index 2a2e6f662..dd69939f9 100644
--- a/pkg/sentry/kernel/task_block.go
+++ b/pkg/sentry/kernel/task_block.go
@@ -15,6 +15,7 @@
package kernel
import (
+ "runtime"
"time"
ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
@@ -121,6 +122,17 @@ func (t *Task) block(C <-chan struct{}, timerChan <-chan struct{}) error {
// Deactive our address space, we don't need it.
interrupt := t.SleepStart()
+ // If the request is not completed, but the timer has already expired,
+ // then ensure that we run through a scheduler cycle. This is because
+ // we may see applications relying on timer slack to yield the thread.
+ // For example, they may attempt to sleep for some number of nanoseconds,
+ // and expect that this will actually yield the CPU and sleep for at
+ // least microseconds, e.g.:
+ // https://github.com/LMAX-Exchange/disruptor/commit/6ca210f2bcd23f703c479804d583718e16f43c07
+ if len(timerChan) > 0 {
+ runtime.Gosched()
+ }
+
select {
case <-C:
t.SleepFinish(true)
diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go
index 54b1676b0..8639d379f 100644
--- a/pkg/sentry/kernel/task_context.go
+++ b/pkg/sentry/kernel/task_context.go
@@ -140,15 +140,22 @@ func (t *Task) Stack() *arch.Stack {
// * wd: Working directory to lookup filename under
// * maxTraversals: maximum number of symlinks to follow
// * filename: path to binary to load
+// * file: an open fs.File object of the binary to load. If set,
+// file will be loaded and not filename.
// * argv: Binary argv
// * envv: Binary envv
// * fs: Binary FeatureSet
-func (k *Kernel) LoadTaskImage(ctx context.Context, mounts *fs.MountNamespace, root, wd *fs.Dirent, maxTraversals *uint, filename string, argv, envv []string, fs *cpuid.FeatureSet) (*TaskContext, *syserr.Error) {
+func (k *Kernel) LoadTaskImage(ctx context.Context, mounts *fs.MountNamespace, root, wd *fs.Dirent, maxTraversals *uint, filename string, file *fs.File, argv, envv []string, fs *cpuid.FeatureSet) (*TaskContext, *syserr.Error) {
+ // If File is not nil, we should load that instead of resolving filename.
+ if file != nil {
+ filename = file.MappedName(ctx)
+ }
+
// Prepare a new user address space to load into.
m := mm.NewMemoryManager(k, k)
defer m.DecUsers(ctx)
- os, ac, name, err := loader.Load(ctx, m, mounts, root, wd, maxTraversals, fs, filename, argv, envv, k.extraAuxv, k.vdso)
+ os, ac, name, err := loader.Load(ctx, m, mounts, root, wd, maxTraversals, fs, filename, file, argv, envv, k.extraAuxv, k.vdso)
if err != nil {
return nil, err
}
diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go
index d60cd62c7..ae6fc4025 100644
--- a/pkg/sentry/kernel/task_start.go
+++ b/pkg/sentry/kernel/task_start.go
@@ -172,9 +172,10 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) {
if parentPG := tg.parentPG(); parentPG == nil {
tg.createSession()
} else {
- // Inherit the process group.
+ // Inherit the process group and terminal.
parentPG.incRefWithParent(parentPG)
tg.processGroup = parentPG
+ tg.tty = t.parent.tg.tty
}
}
tg.tasks.PushBack(t)
diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go
index 2a97e3e8e..0eef24bfb 100644
--- a/pkg/sentry/kernel/thread_group.go
+++ b/pkg/sentry/kernel/thread_group.go
@@ -19,10 +19,13 @@ import (
"sync/atomic"
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/fs"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
"gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/usage"
+ "gvisor.dev/gvisor/pkg/syserror"
)
// A ThreadGroup is a logical grouping of tasks that has widespread
@@ -245,6 +248,12 @@ type ThreadGroup struct {
//
// mounts is immutable.
mounts *fs.MountNamespace
+
+ // tty is the thread group's controlling terminal. If nil, there is no
+ // controlling terminal.
+ //
+ // tty is protected by the signal mutex.
+ tty *TTY
}
// newThreadGroup returns a new, empty thread group in PID namespace ns. The
@@ -324,6 +333,176 @@ func (tg *ThreadGroup) forEachChildThreadGroupLocked(fn func(*ThreadGroup)) {
}
}
+// SetControllingTTY sets tty as the controlling terminal of tg.
+func (tg *ThreadGroup) SetControllingTTY(tty *TTY, arg int32) error {
+ tty.mu.Lock()
+ defer tty.mu.Unlock()
+
+ // We might be asked to set the controlling terminal of multiple
+ // processes, so we lock both the TaskSet and SignalHandlers.
+ tg.pidns.owner.mu.Lock()
+ defer tg.pidns.owner.mu.Unlock()
+ tg.signalHandlers.mu.Lock()
+ defer tg.signalHandlers.mu.Unlock()
+
+ // "The calling process must be a session leader and not have a
+ // controlling terminal already." - tty_ioctl(4)
+ if tg.processGroup.session.leader != tg || tg.tty != nil {
+ return syserror.EINVAL
+ }
+
+ // "If this terminal is already the controlling terminal of a different
+ // session group, then the ioctl fails with EPERM, unless the caller
+ // has the CAP_SYS_ADMIN capability and arg equals 1, in which case the
+ // terminal is stolen, and all processes that had it as controlling
+ // terminal lose it." - tty_ioctl(4)
+ if tty.tg != nil && tg.processGroup.session != tty.tg.processGroup.session {
+ if !auth.CredentialsFromContext(tg.leader).HasCapability(linux.CAP_SYS_ADMIN) || arg != 1 {
+ return syserror.EPERM
+ }
+ // Steal the TTY away. Unlike TIOCNOTTY, don't send signals.
+ for othertg := range tg.pidns.owner.Root.tgids {
+ // This won't deadlock by locking tg.signalHandlers
+ // because at this point:
+ // - We only lock signalHandlers if it's in the same
+ // session as the tty's controlling thread group.
+ // - We know that the calling thread group is not in
+ // the same session as the tty's controlling thread
+ // group.
+ if othertg.processGroup.session == tty.tg.processGroup.session {
+ othertg.signalHandlers.mu.Lock()
+ othertg.tty = nil
+ othertg.signalHandlers.mu.Unlock()
+ }
+ }
+ }
+
+ // Set the controlling terminal and foreground process group.
+ tg.tty = tty
+ tg.processGroup.session.foreground = tg.processGroup
+ // Set this as the controlling process of the terminal.
+ tty.tg = tg
+
+ return nil
+}
+
+// ReleaseControllingTTY gives up tty as the controlling tty of tg.
+func (tg *ThreadGroup) ReleaseControllingTTY(tty *TTY) error {
+ tty.mu.Lock()
+ defer tty.mu.Unlock()
+
+ // We might be asked to set the controlling terminal of multiple
+ // processes, so we lock both the TaskSet and SignalHandlers.
+ tg.pidns.owner.mu.RLock()
+ defer tg.pidns.owner.mu.RUnlock()
+
+ // Just below, we may re-lock signalHandlers in order to send signals.
+ // Thus we can't defer Unlock here.
+ tg.signalHandlers.mu.Lock()
+
+ if tg.tty == nil || tg.tty != tty {
+ tg.signalHandlers.mu.Unlock()
+ return syserror.ENOTTY
+ }
+
+ // "If the process was session leader, then send SIGHUP and SIGCONT to
+ // the foreground process group and all processes in the current
+ // session lose their controlling terminal." - tty_ioctl(4)
+ // Remove tty as the controlling tty for each process in the session,
+ // then send them SIGHUP and SIGCONT.
+
+ // If we're not the session leader, we don't have to do much.
+ if tty.tg != tg {
+ tg.tty = nil
+ tg.signalHandlers.mu.Unlock()
+ return nil
+ }
+
+ tg.signalHandlers.mu.Unlock()
+
+ // We're the session leader. SIGHUP and SIGCONT the foreground process
+ // group and remove all controlling terminals in the session.
+ var lastErr error
+ for othertg := range tg.pidns.owner.Root.tgids {
+ if othertg.processGroup.session == tg.processGroup.session {
+ othertg.signalHandlers.mu.Lock()
+ othertg.tty = nil
+ if othertg.processGroup == tg.processGroup.session.foreground {
+ if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGHUP)}, true /* group */); err != nil {
+ lastErr = err
+ }
+ if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGCONT)}, true /* group */); err != nil {
+ lastErr = err
+ }
+ }
+ othertg.signalHandlers.mu.Unlock()
+ }
+ }
+
+ return lastErr
+}
+
+// ForegroundProcessGroup returns the process group ID of the foreground
+// process group.
+func (tg *ThreadGroup) ForegroundProcessGroup(tty *TTY) (int32, error) {
+ tty.mu.Lock()
+ defer tty.mu.Unlock()
+
+ tg.pidns.owner.mu.Lock()
+ defer tg.pidns.owner.mu.Unlock()
+ tg.signalHandlers.mu.Lock()
+ defer tg.signalHandlers.mu.Unlock()
+
+ // "When fd does not refer to the controlling terminal of the calling
+ // process, -1 is returned" - tcgetpgrp(3)
+ if tg.tty != tty {
+ return -1, syserror.ENOTTY
+ }
+
+ return int32(tg.processGroup.session.foreground.id), nil
+}
+
+// SetForegroundProcessGroup sets the foreground process group of tty to pgid.
+func (tg *ThreadGroup) SetForegroundProcessGroup(tty *TTY, pgid ProcessGroupID) (int32, error) {
+ tty.mu.Lock()
+ defer tty.mu.Unlock()
+
+ tg.pidns.owner.mu.Lock()
+ defer tg.pidns.owner.mu.Unlock()
+ tg.signalHandlers.mu.Lock()
+ defer tg.signalHandlers.mu.Unlock()
+
+ // TODO(b/129283598): "If tcsetpgrp() is called by a member of a
+ // background process group in its session, and the calling process is
+ // not blocking or ignoring SIGTTOU, a SIGTTOU signal is sent to all
+ // members of this background process group."
+
+ // tty must be the controlling terminal.
+ if tg.tty != tty {
+ return -1, syserror.ENOTTY
+ }
+
+ // pgid must be positive.
+ if pgid < 0 {
+ return -1, syserror.EINVAL
+ }
+
+ // pg must not be empty. Empty process groups are removed from their
+ // pid namespaces.
+ pg, ok := tg.pidns.processGroups[pgid]
+ if !ok {
+ return -1, syserror.ESRCH
+ }
+
+ // pg must be part of this process's session.
+ if tg.processGroup.session != pg.session {
+ return -1, syserror.EPERM
+ }
+
+ tg.processGroup.session.foreground.id = pgid
+ return 0, nil
+}
+
// itimerRealListener implements ktime.Listener for ITIMER_REAL expirations.
//
// +stateify savable
diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go
new file mode 100644
index 000000000..34f84487a
--- /dev/null
+++ b/pkg/sentry/kernel/tty.go
@@ -0,0 +1,28 @@
+// 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 kernel
+
+import "sync"
+
+// TTY defines the relationship between a thread group and its controlling
+// terminal.
+//
+// +stateify savable
+type TTY struct {
+ mu sync.Mutex `state:"nosave"`
+
+ // tg is protected by mu.
+ tg *ThreadGroup
+}
generated by cgit v1.2.3 (git 2.39.1) at 2025-02-16 03:59:35 +0000