diff options
Diffstat (limited to 'pkg/sentry/kernel')
33 files changed, 293 insertions, 189 deletions
diff --git a/pkg/sentry/kernel/abstract_socket_namespace.go b/pkg/sentry/kernel/abstract_socket_namespace.go index 920fe4329..52ed5cea2 100644 --- a/pkg/sentry/kernel/abstract_socket_namespace.go +++ b/pkg/sentry/kernel/abstract_socket_namespace.go @@ -17,6 +17,7 @@ package kernel import ( "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" @@ -31,7 +32,7 @@ type abstractEndpoint struct { } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (e *abstractEndpoint) WeakRefGone() { +func (e *abstractEndpoint) WeakRefGone(context.Context) { e.ns.mu.Lock() if e.ns.endpoints[e.name].ep == e.ep { delete(e.ns.endpoints, e.name) @@ -64,9 +65,9 @@ type boundEndpoint struct { } // Release implements transport.BoundEndpoint.Release. -func (e *boundEndpoint) Release() { - e.rc.DecRef() - e.BoundEndpoint.Release() +func (e *boundEndpoint) Release(ctx context.Context) { + e.rc.DecRef(ctx) + e.BoundEndpoint.Release(ctx) } // BoundEndpoint retrieves the endpoint bound to the given name. The return @@ -93,13 +94,13 @@ func (a *AbstractSocketNamespace) BoundEndpoint(name string) transport.BoundEndp // // When the last reference managed by rc is dropped, ep may be removed from the // namespace. -func (a *AbstractSocketNamespace) Bind(name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { +func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { a.mu.Lock() defer a.mu.Unlock() if ep, ok := a.endpoints[name]; ok { if rc := ep.wr.Get(); rc != nil { - rc.DecRef() + rc.DecRef(ctx) return syscall.EADDRINUSE } } diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 4c0f1e41f..15519f0df 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -76,8 +76,8 @@ type pollEntry struct { // WeakRefGone implements refs.WeakRefUser.WeakRefGone. // weakReferenceGone is called when the file in the weak reference is destroyed. // The poll entry is removed in response to this. -func (p *pollEntry) WeakRefGone() { - p.epoll.RemoveEntry(p.id) +func (p *pollEntry) WeakRefGone(ctx context.Context) { + p.epoll.RemoveEntry(ctx, p.id) } // EventPoll holds all the state associated with an event poll object, that is, @@ -144,14 +144,14 @@ func NewEventPoll(ctx context.Context) *fs.File { // name matches fs/eventpoll.c:epoll_create1. dirent := fs.NewDirent(ctx, anon.NewInode(ctx), fmt.Sprintf("anon_inode:[eventpoll]")) // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{}, &EventPoll{ files: make(map[FileIdentifier]*pollEntry), }) } // Release implements fs.FileOperations.Release. -func (e *EventPoll) Release() { +func (e *EventPoll) Release(ctx context.Context) { // We need to take the lock now because files may be attempting to // remove entries in parallel if they get destroyed. e.mu.Lock() @@ -160,7 +160,7 @@ func (e *EventPoll) Release() { // Go through all entries and clean up. for _, entry := range e.files { entry.id.File.EventUnregister(&entry.waiter) - entry.file.Drop() + entry.file.Drop(ctx) } e.files = nil } @@ -423,7 +423,7 @@ func (e *EventPoll) UpdateEntry(id FileIdentifier, flags EntryFlags, mask waiter } // RemoveEntry a files from the collection of observed files. -func (e *EventPoll) RemoveEntry(id FileIdentifier) error { +func (e *EventPoll) RemoveEntry(ctx context.Context, id FileIdentifier) error { e.mu.Lock() defer e.mu.Unlock() @@ -445,7 +445,7 @@ func (e *EventPoll) RemoveEntry(id FileIdentifier) error { // Remove file from map, and drop weak reference. delete(e.files, id) - entry.file.Drop() + entry.file.Drop(ctx) return nil } diff --git a/pkg/sentry/kernel/epoll/epoll_test.go b/pkg/sentry/kernel/epoll/epoll_test.go index 22630e9c5..55b505593 100644 --- a/pkg/sentry/kernel/epoll/epoll_test.go +++ b/pkg/sentry/kernel/epoll/epoll_test.go @@ -26,7 +26,8 @@ func TestFileDestroyed(t *testing.T) { f := filetest.NewTestFile(t) id := FileIdentifier{f, 12} - efile := NewEventPoll(contexttest.Context(t)) + ctx := contexttest.Context(t) + efile := NewEventPoll(ctx) e := efile.FileOperations.(*EventPoll) if err := e.AddEntry(id, 0, waiter.EventIn, [2]int32{}); err != nil { t.Fatalf("addEntry failed: %v", err) @@ -44,7 +45,7 @@ func TestFileDestroyed(t *testing.T) { } // Destroy the file. Check that we get no more events. - f.DecRef() + f.DecRef(ctx) evt = e.ReadEvents(1) if len(evt) != 0 { diff --git a/pkg/sentry/kernel/eventfd/eventfd.go b/pkg/sentry/kernel/eventfd/eventfd.go index 87951adeb..bbf568dfc 100644 --- a/pkg/sentry/kernel/eventfd/eventfd.go +++ b/pkg/sentry/kernel/eventfd/eventfd.go @@ -70,7 +70,7 @@ func New(ctx context.Context, initVal uint64, semMode bool) *fs.File { // name matches fs/eventfd.c:eventfd_file_create. dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[eventfd]") // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &EventOperations{ val: initVal, semMode: semMode, @@ -106,7 +106,7 @@ func (e *EventOperations) HostFD() (int, error) { } // Release implements fs.FileOperations.Release. -func (e *EventOperations) Release() { +func (e *EventOperations) Release(context.Context) { e.mu.Lock() defer e.mu.Unlock() if e.hostfd >= 0 { diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 4b7d234a4..ce53af69b 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -98,7 +98,7 @@ type FDTable struct { func (f *FDTable) saveDescriptorTable() map[int32]descriptor { m := make(map[int32]descriptor) - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(context.Background(), func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { m[fd] = descriptor{ file: file, fileVFS2: fileVFS2, @@ -109,6 +109,7 @@ func (f *FDTable) saveDescriptorTable() map[int32]descriptor { } func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { + ctx := context.Background() f.init() // Initialize table. for fd, d := range m { f.setAll(fd, d.file, d.fileVFS2, d.flags) @@ -118,9 +119,9 @@ func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { // reference taken by set above. switch { case d.file != nil: - d.file.DecRef() + d.file.DecRef(ctx) case d.fileVFS2 != nil: - d.fileVFS2.DecRef() + d.fileVFS2.DecRef(ctx) } } } @@ -144,14 +145,15 @@ func (f *FDTable) drop(file *fs.File) { d.InotifyEvent(ev, 0) // Drop the table reference. - file.DecRef() + file.DecRef(context.Background()) } // dropVFS2 drops the table reference. func (f *FDTable) dropVFS2(file *vfs.FileDescription) { // Release any POSIX lock possibly held by the FDTable. Range {0, 0} means the // entire file. - err := file.UnlockPOSIX(context.Background(), f, 0, 0, linux.SEEK_SET) + ctx := context.Background() + err := file.UnlockPOSIX(ctx, f, 0, 0, linux.SEEK_SET) if err != nil && err != syserror.ENOLCK { panic(fmt.Sprintf("UnlockPOSIX failed: %v", err)) } @@ -161,10 +163,10 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { if file.IsWritable() { ev = linux.IN_CLOSE_WRITE } - file.Dentry().InotifyWithParent(ev, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(ctx, ev, 0, vfs.PathEvent) // Drop the table's reference. - file.DecRef() + file.DecRef(ctx) } // NewFDTable allocates a new FDTable that may be used by tasks in k. @@ -175,15 +177,15 @@ func (k *Kernel) NewFDTable() *FDTable { } // destroy removes all of the file descriptors from the map. -func (f *FDTable) destroy() { - f.RemoveIf(func(*fs.File, *vfs.FileDescription, FDFlags) bool { +func (f *FDTable) destroy(ctx context.Context) { + f.RemoveIf(ctx, func(*fs.File, *vfs.FileDescription, FDFlags) bool { return true }) } // DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FDTable) DecRef() { - f.DecRefWithDestructor(f.destroy) +func (f *FDTable) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, f.destroy) } // Size returns the number of file descriptor slots currently allocated. @@ -195,7 +197,7 @@ func (f *FDTable) Size() int { // forEach iterates over all non-nil files in sorted order. // // It is the caller's responsibility to acquire an appropriate lock. -func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { +func (f *FDTable) forEach(ctx context.Context, fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { // retries tracks the number of failed TryIncRef attempts for the same FD. retries := 0 fd := int32(0) @@ -214,7 +216,7 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes continue // Race caught. } fn(fd, file, nil, flags) - file.DecRef() + file.DecRef(ctx) case fileVFS2 != nil: if !fileVFS2.TryIncRef() { retries++ @@ -224,7 +226,7 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes continue // Race caught. } fn(fd, nil, fileVFS2, flags) - fileVFS2.DecRef() + fileVFS2.DecRef(ctx) } retries = 0 fd++ @@ -234,7 +236,8 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes // String is a stringer for FDTable. func (f *FDTable) String() string { var buf strings.Builder - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + ctx := context.Background() + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { switch { case file != nil: n, _ := file.Dirent.FullName(nil /* root */) @@ -242,7 +245,7 @@ func (f *FDTable) String() string { case fileVFS2 != nil: vfsObj := fileVFS2.Mount().Filesystem().VirtualFilesystem() - name, err := vfsObj.PathnameWithDeleted(context.Background(), vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) + name, err := vfsObj.PathnameWithDeleted(ctx, vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) if err != nil { fmt.Fprintf(&buf, "<err: %v>\n", err) return @@ -541,9 +544,9 @@ func (f *FDTable) GetVFS2(fd int32) (*vfs.FileDescription, FDFlags) { // // Precondition: The caller must be running on the task goroutine, or Task.mu // must be locked. -func (f *FDTable) GetFDs() []int32 { +func (f *FDTable) GetFDs(ctx context.Context) []int32 { fds := make([]int32, 0, int(atomic.LoadInt32(&f.used))) - f.forEach(func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { fds = append(fds, fd) }) return fds @@ -552,9 +555,9 @@ func (f *FDTable) GetFDs() []int32 { // GetRefs returns a stable slice of references to all files and bumps the // reference count on each. The caller must use DecRef on each reference when // they're done using the slice. -func (f *FDTable) GetRefs() []*fs.File { +func (f *FDTable) GetRefs(ctx context.Context) []*fs.File { files := make([]*fs.File, 0, f.Size()) - f.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -564,9 +567,9 @@ func (f *FDTable) GetRefs() []*fs.File { // GetRefsVFS2 returns a stable slice of references to all files and bumps the // reference count on each. The caller must use DecRef on each reference when // they're done using the slice. -func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { +func (f *FDTable) GetRefsVFS2(ctx context.Context) []*vfs.FileDescription { files := make([]*vfs.FileDescription, 0, f.Size()) - f.forEach(func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -574,10 +577,10 @@ func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { } // Fork returns an independent FDTable. -func (f *FDTable) Fork() *FDTable { +func (f *FDTable) Fork(ctx context.Context) *FDTable { clone := f.k.NewFDTable() - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { // The set function here will acquire an appropriate table // reference for the clone. We don't need anything else. switch { @@ -622,11 +625,11 @@ func (f *FDTable) Remove(fd int32) (*fs.File, *vfs.FileDescription) { } // RemoveIf removes all FDs where cond is true. -func (f *FDTable) RemoveIf(cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { +func (f *FDTable) RemoveIf(ctx context.Context, cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { f.mu.Lock() defer f.mu.Unlock() - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { if cond(file, fileVFS2, flags) { f.set(fd, nil, FDFlags{}) // Clear from table. // Update current available position. diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index 29f95a2c4..e3f30ba2a 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -154,7 +154,7 @@ func TestFDTable(t *testing.T) { if ref == nil { t.Fatalf("fdTable.Remove(1) for an existing FD: failed, want success") } - ref.DecRef() + ref.DecRef(ctx) if ref, _ := fdTable.Remove(1); ref != nil { t.Fatalf("r.Remove(1) for a removed FD: got success, want failure") @@ -191,7 +191,7 @@ func BenchmarkFDLookupAndDecRef(b *testing.B) { b.StartTimer() // Benchmark. for i := 0; i < b.N; i++ { tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() + tf.DecRef(ctx) } }) } @@ -219,7 +219,7 @@ func BenchmarkFDLookupAndDecRefConcurrent(b *testing.B) { defer wg.Done() for i := 0; i < each; i++ { tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() + tf.DecRef(ctx) } }() } diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index 47f78df9a..8f2d36d5a 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -17,6 +17,7 @@ package kernel import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -89,28 +90,28 @@ func NewFSContextVFS2(root, cwd vfs.VirtualDentry, umask uint) *FSContext { // Note that there may still be calls to WorkingDirectory() or RootDirectory() // (that return nil). This is because valid references may still be held via // proc files or other mechanisms. -func (f *FSContext) destroy() { +func (f *FSContext) destroy(ctx context.Context) { // Hold f.mu so that we don't race with RootDirectory() and // WorkingDirectory(). f.mu.Lock() defer f.mu.Unlock() if VFS2Enabled { - f.rootVFS2.DecRef() + f.rootVFS2.DecRef(ctx) f.rootVFS2 = vfs.VirtualDentry{} - f.cwdVFS2.DecRef() + f.cwdVFS2.DecRef(ctx) f.cwdVFS2 = vfs.VirtualDentry{} } else { - f.root.DecRef() + f.root.DecRef(ctx) f.root = nil - f.cwd.DecRef() + f.cwd.DecRef(ctx) f.cwd = nil } } // DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FSContext) DecRef() { - f.DecRefWithDestructor(f.destroy) +func (f *FSContext) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, f.destroy) } // Fork forks this FSContext. @@ -165,7 +166,7 @@ func (f *FSContext) WorkingDirectoryVFS2() vfs.VirtualDentry { // This will take an extra reference on the Dirent. // // This is not a valid call after destroy. -func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { +func (f *FSContext) SetWorkingDirectory(ctx context.Context, d *fs.Dirent) { if d == nil { panic("FSContext.SetWorkingDirectory called with nil dirent") } @@ -180,21 +181,21 @@ func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { old := f.cwd f.cwd = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // SetWorkingDirectoryVFS2 sets the current working directory. // This will take an extra reference on the VirtualDentry. // // This is not a valid call after destroy. -func (f *FSContext) SetWorkingDirectoryVFS2(d vfs.VirtualDentry) { +func (f *FSContext) SetWorkingDirectoryVFS2(ctx context.Context, d vfs.VirtualDentry) { f.mu.Lock() defer f.mu.Unlock() old := f.cwdVFS2 f.cwdVFS2 = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // RootDirectory returns the current filesystem root. @@ -226,7 +227,7 @@ func (f *FSContext) RootDirectoryVFS2() vfs.VirtualDentry { // This will take an extra reference on the Dirent. // // This is not a valid call after free. -func (f *FSContext) SetRootDirectory(d *fs.Dirent) { +func (f *FSContext) SetRootDirectory(ctx context.Context, d *fs.Dirent) { if d == nil { panic("FSContext.SetRootDirectory called with nil dirent") } @@ -241,13 +242,13 @@ func (f *FSContext) SetRootDirectory(d *fs.Dirent) { old := f.root f.root = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // SetRootDirectoryVFS2 sets the root directory. It takes a reference on vd. // // This is not a valid call after free. -func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { +func (f *FSContext) SetRootDirectoryVFS2(ctx context.Context, vd vfs.VirtualDentry) { if !vd.Ok() { panic("FSContext.SetRootDirectoryVFS2 called with zero-value VirtualDentry") } @@ -263,7 +264,7 @@ func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { vd.IncRef() f.rootVFS2 = vd f.mu.Unlock() - old.DecRef() + old.DecRef(ctx) } // Umask returns the current umask. diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index c5021f2db..daa2dae76 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -51,6 +51,7 @@ go_test( srcs = ["futex_test.go"], library = ":futex", deps = [ + "//pkg/context", "//pkg/sync", "//pkg/usermem", ], diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index bcc1b29a8..e4dcc4d40 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -19,6 +19,7 @@ package futex import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -66,9 +67,9 @@ type Key struct { Offset uint64 } -func (k *Key) release() { +func (k *Key) release(t Target) { if k.MappingIdentity != nil { - k.MappingIdentity.DecRef() + k.MappingIdentity.DecRef(t) } k.Mappable = nil k.MappingIdentity = nil @@ -94,6 +95,8 @@ func (k *Key) matches(k2 *Key) bool { // Target abstracts memory accesses and keys. type Target interface { + context.Context + // SwapUint32 gives access to usermem.IO.SwapUint32. SwapUint32(addr usermem.Addr, new uint32) (uint32, error) @@ -296,7 +299,7 @@ func (b *bucket) wakeWaiterLocked(w *Waiter) { // bucket "to". // // Preconditions: b and to must be locked. -func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int { +func (b *bucket) requeueLocked(t Target, to *bucket, key, nkey *Key, n int) int { done := 0 for w := b.waiters.Front(); done < n && w != nil; { if !w.key.matches(key) { @@ -308,7 +311,7 @@ func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int { requeued := w w = w.Next() // Next iteration. b.waiters.Remove(requeued) - requeued.key.release() + requeued.key.release(t) requeued.key = nkey.clone() to.waiters.PushBack(requeued) requeued.bucket.Store(to) @@ -456,7 +459,7 @@ func (m *Manager) Wake(t Target, addr usermem.Addr, private bool, bitmask uint32 r := b.wakeLocked(&k, bitmask, n) b.mu.Unlock() - k.release() + k.release(t) return r, nil } @@ -465,12 +468,12 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch if err != nil { return 0, err } - defer k1.release() + defer k1.release(t) k2, err := getKey(t, naddr, private) if err != nil { return 0, err } - defer k2.release() + defer k2.release(t) b1, b2 := m.lockBuckets(&k1, &k2) defer b1.mu.Unlock() @@ -488,7 +491,7 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch done := b1.wakeLocked(&k1, ^uint32(0), nwake) // Requeue the number required. - b1.requeueLocked(b2, &k1, &k2, nreq) + b1.requeueLocked(t, b2, &k1, &k2, nreq) return done, nil } @@ -515,12 +518,12 @@ func (m *Manager) WakeOp(t Target, addr1, addr2 usermem.Addr, private bool, nwak if err != nil { return 0, err } - defer k1.release() + defer k1.release(t) k2, err := getKey(t, addr2, private) if err != nil { return 0, err } - defer k2.release() + defer k2.release(t) b1, b2 := m.lockBuckets(&k1, &k2) defer b1.mu.Unlock() @@ -571,7 +574,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo // Perform our atomic check. if err := check(t, addr, val); err != nil { b.mu.Unlock() - w.key.release() + w.key.release(t) return err } @@ -585,7 +588,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo // WaitComplete must be called when a Waiter previously added by WaitPrepare is // no longer eligible to be woken. -func (m *Manager) WaitComplete(w *Waiter) { +func (m *Manager) WaitComplete(w *Waiter, t Target) { // Remove w from the bucket it's in. for { b := w.bucket.Load() @@ -617,7 +620,7 @@ func (m *Manager) WaitComplete(w *Waiter) { } // Release references held by the waiter. - w.key.release() + w.key.release(t) } // LockPI attempts to lock the futex following the Priority-inheritance futex @@ -648,13 +651,13 @@ func (m *Manager) LockPI(w *Waiter, t Target, addr usermem.Addr, tid uint32, pri success, err := m.lockPILocked(w, t, addr, tid, b, try) if err != nil { - w.key.release() + w.key.release(t) b.mu.Unlock() return false, err } if success || try { // Release waiter if it's not going to be a wait. - w.key.release() + w.key.release(t) } b.mu.Unlock() return success, nil @@ -730,7 +733,7 @@ func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool err = m.unlockPILocked(t, addr, tid, b, &k) - k.release() + k.release(t) b.mu.Unlock() return err } diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go index 7c5c7665b..d0128c548 100644 --- a/pkg/sentry/kernel/futex/futex_test.go +++ b/pkg/sentry/kernel/futex/futex_test.go @@ -22,6 +22,7 @@ import ( "testing" "unsafe" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" ) @@ -29,28 +30,33 @@ import ( // testData implements the Target interface, and allows us to // treat the address passed for futex operations as an index in // a byte slice for testing simplicity. -type testData []byte +type testData struct { + context.Context + data []byte +} const sizeofInt32 = 4 func newTestData(size uint) testData { - return make([]byte, size) + return testData{ + data: make([]byte, size), + } } func (t testData) SwapUint32(addr usermem.Addr, new uint32) (uint32, error) { - val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t[addr])), new) + val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), new) return val, nil } func (t testData) CompareAndSwapUint32(addr usermem.Addr, old, new uint32) (uint32, error) { - if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t[addr])), old, new) { + if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), old, new) { return old, nil } - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil } func (t testData) LoadUint32(addr usermem.Addr) (uint32, error) { - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil } func (t testData) GetSharedKey(addr usermem.Addr) (Key, error) { @@ -83,7 +89,7 @@ func TestFutexWake(t *testing.T) { // Start waiting for wakeup. w := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w) + defer m.WaitComplete(w, d) // Perform a wakeup. if n, err := m.Wake(d, 0, private, ^uint32(0), 1); err != nil || n != 1 { @@ -106,7 +112,7 @@ func TestFutexWakeBitmask(t *testing.T) { // Start waiting for wakeup. w := newPreparedTestWaiter(t, m, d, 0, private, 0, 0x0000ffff) - defer m.WaitComplete(w) + defer m.WaitComplete(w, d) // Perform a wakeup using the wrong bitmask. if n, err := m.Wake(d, 0, private, 0xffff0000, 1); err != nil || n != 0 { @@ -141,7 +147,7 @@ func TestFutexWakeTwo(t *testing.T) { var ws [3]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform two wakeups. @@ -174,9 +180,9 @@ func TestFutexWakeUnrelated(t *testing.T) { // Start two waiters waiting for wakeup on different addresses. w1 := newPreparedTestWaiter(t, m, d, 0*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 1*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform two wakeups on the second address. if n, err := m.Wake(d, 1*sizeofInt32, private, ^uint32(0), 2); err != nil || n != 1 { @@ -216,9 +222,9 @@ func TestWakeOpFirstNonEmpty(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address 0. if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 0, 0); err != nil || n != 2 { @@ -244,9 +250,9 @@ func TestWakeOpSecondNonEmpty(t *testing.T) { // Add two waiters on address sizeofInt32. w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address sizeofInt32 (contingent on // d.Op(0), which should succeed). @@ -273,9 +279,9 @@ func TestWakeOpSecondNonEmptyFailingOp(t *testing.T) { // Add two waiters on address sizeofInt32. w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address sizeofInt32 (contingent on // d.Op(1), which should fail). @@ -302,15 +308,15 @@ func TestWakeOpAllNonEmpty(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Add two waiters on address sizeofInt32. w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) + defer m.WaitComplete(w3, d) w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) + defer m.WaitComplete(w4, d) // Perform 10 wakeups on address 0 (unconditionally), and 10 // wakeups on address sizeofInt32 (contingent on d.Op(0), which @@ -344,15 +350,15 @@ func TestWakeOpAllNonEmptyFailingOp(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Add two waiters on address sizeofInt32. w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) + defer m.WaitComplete(w3, d) w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) + defer m.WaitComplete(w4, d) // Perform 10 wakeups on address 0 (unconditionally), and 10 // wakeups on address sizeofInt32 (contingent on d.Op(1), which @@ -388,7 +394,7 @@ func TestWakeOpSameAddress(t *testing.T) { var ws [4]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup @@ -422,7 +428,7 @@ func TestWakeOpSameAddressFailingOp(t *testing.T) { var ws [4]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup @@ -472,7 +478,7 @@ func (t *testMutex) Lock() { for { // Attempt to grab the lock. if atomic.CompareAndSwapUint32( - (*uint32)(unsafe.Pointer(&t.d[t.a])), + (*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked, testMutexLocked) { // Lock held. @@ -490,7 +496,7 @@ func (t *testMutex) Lock() { panic("WaitPrepare returned unexpected error: " + err.Error()) } <-w.C - t.m.WaitComplete(w) + t.m.WaitComplete(w, t.d) } } @@ -498,7 +504,7 @@ func (t *testMutex) Lock() { // This will notify any waiters via the futex manager. func (t *testMutex) Unlock() { // Unlock. - atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d[t.a])), testMutexUnlocked) + atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked) // Notify all waiters. t.m.Wake(t.d, t.a, true, ^uint32(0), math.MaxInt32) diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 15dae0f5b..1028d13c6 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -376,7 +376,8 @@ func (k *Kernel) Init(args InitKernelArgs) error { k.netlinkPorts = port.New() if VFS2Enabled { - if err := k.vfs.Init(); err != nil { + ctx := k.SupervisorContext() + if err := k.vfs.Init(ctx); err != nil { return fmt.Errorf("failed to initialize VFS: %v", err) } @@ -384,19 +385,19 @@ func (k *Kernel) Init(args InitKernelArgs) error { if err != nil { return fmt.Errorf("failed to create pipefs filesystem: %v", err) } - defer pipeFilesystem.DecRef() + defer pipeFilesystem.DecRef(ctx) pipeMount, err := k.vfs.NewDisconnectedMount(pipeFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create pipefs mount: %v", err) } k.pipeMount = pipeMount - tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(k.SupervisorContext(), &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) + tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(ctx, &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) if err != nil { return fmt.Errorf("failed to create tmpfs filesystem: %v", err) } - defer tmpfsFilesystem.DecRef() - defer tmpfsRoot.DecRef() + defer tmpfsFilesystem.DecRef(ctx) + defer tmpfsRoot.DecRef(ctx) shmMount, err := k.vfs.NewDisconnectedMount(tmpfsFilesystem, tmpfsRoot, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create tmpfs mount: %v", err) @@ -407,7 +408,7 @@ func (k *Kernel) Init(args InitKernelArgs) error { if err != nil { return fmt.Errorf("failed to create sockfs filesystem: %v", err) } - defer socketFilesystem.DecRef() + defer socketFilesystem.DecRef(ctx) socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create sockfs mount: %v", err) @@ -430,8 +431,8 @@ func (k *Kernel) SaveTo(w wire.Writer) error { defer k.extMu.Unlock() // Stop time. - k.pauseTimeLocked() - defer k.resumeTimeLocked() + k.pauseTimeLocked(ctx) + defer k.resumeTimeLocked(ctx) // Evict all evictable MemoryFile allocations. k.mf.StartEvictions() @@ -447,12 +448,12 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // Remove all epoll waiter objects from underlying wait queues. // NOTE: for programs to resume execution in future snapshot scenarios, // we will need to re-establish these waiter objects after saving. - k.tasks.unregisterEpollWaiters() + k.tasks.unregisterEpollWaiters(ctx) // Clear the dirent cache before saving because Dirents must be Loaded in a // particular order (parents before children), and Loading dirents from a cache // breaks that order. - if err := k.flushMountSourceRefs(); err != nil { + if err := k.flushMountSourceRefs(ctx); err != nil { return err } @@ -505,7 +506,7 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // flushMountSourceRefs flushes the MountSources for all mounted filesystems // and open FDs. -func (k *Kernel) flushMountSourceRefs() error { +func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { // Flush all mount sources for currently mounted filesystems in each task. flushed := make(map[*fs.MountNamespace]struct{}) k.tasks.mu.RLock() @@ -521,7 +522,7 @@ func (k *Kernel) flushMountSourceRefs() error { // There may be some open FDs whose filesystems have been unmounted. We // must flush those as well. - return k.tasks.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { + return k.tasks.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { file.Dirent.Inode.MountSource.FlushDirentRefs() return nil }) @@ -531,7 +532,7 @@ func (k *Kernel) flushMountSourceRefs() error { // each task. // // Precondition: Must be called with the kernel paused. -func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) (err error) { +func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.FileDescription) error) (err error) { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. if VFS2Enabled { return nil @@ -544,7 +545,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) if t.fdTable == nil { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { if lastErr := f(file, fileVFS2); lastErr != nil && err == nil { err = lastErr } @@ -555,7 +556,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - return ts.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { + return ts.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil } @@ -602,7 +603,7 @@ func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error { return nil } -func (ts *TaskSet) unregisterEpollWaiters() { +func (ts *TaskSet) unregisterEpollWaiters(ctx context.Context) { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. if VFS2Enabled { return @@ -623,7 +624,7 @@ func (ts *TaskSet) unregisterEpollWaiters() { if _, ok := processed[t.fdTable]; ok { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { if e, ok := file.FileOperations.(*epoll.EventPoll); ok { e.UnregisterEpollWaiters() } @@ -900,7 +901,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, root := args.MountNamespaceVFS2.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. - defer root.DecRef() + defer root.DecRef(ctx) // Grab the working directory. wd := root // Default. @@ -918,7 +919,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if err != nil { return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) } - defer wd.DecRef() + defer wd.DecRef(ctx) } opener = fsbridge.NewVFSLookup(mntnsVFS2, root, wd) fsContext = NewFSContextVFS2(root, wd, args.Umask) @@ -933,7 +934,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, root := mntns.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. - defer root.DecRef() + defer root.DecRef(ctx) // Grab the working directory. remainingTraversals := args.MaxSymlinkTraversals @@ -944,7 +945,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if err != nil { return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) } - defer wd.DecRef() + defer wd.DecRef(ctx) } opener = fsbridge.NewFSLookup(mntns, root, wd) fsContext = newFSContext(root, wd, args.Umask) @@ -1054,7 +1055,7 @@ func (k *Kernel) Start() error { // If k was created by LoadKernelFrom, timers were stopped during // Kernel.SaveTo and need to be resumed. If k was created by NewKernel, // this is a no-op. - k.resumeTimeLocked() + k.resumeTimeLocked(k.SupervisorContext()) // Start task goroutines. k.tasks.mu.RLock() defer k.tasks.mu.RUnlock() @@ -1068,7 +1069,7 @@ func (k *Kernel) Start() error { // // Preconditions: Any task goroutines running in k must be stopped. k.extMu // must be locked. -func (k *Kernel) pauseTimeLocked() { +func (k *Kernel) pauseTimeLocked(ctx context.Context) { // k.cpuClockTicker may be nil since Kernel.SaveTo() may be called before // Kernel.Start(). if k.cpuClockTicker != nil { @@ -1090,7 +1091,7 @@ func (k *Kernel) pauseTimeLocked() { // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.PauseTimer() @@ -1112,7 +1113,7 @@ func (k *Kernel) pauseTimeLocked() { // // Preconditions: Any task goroutines running in k must be stopped. k.extMu // must be locked. -func (k *Kernel) resumeTimeLocked() { +func (k *Kernel) resumeTimeLocked(ctx context.Context) { if k.cpuClockTicker != nil { k.cpuClockTicker.Resume() } @@ -1126,7 +1127,7 @@ func (k *Kernel) resumeTimeLocked() { } } if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.ResumeTimer() @@ -1262,6 +1263,13 @@ func (k *Kernel) Pause() { k.tasks.aioGoroutines.Wait() } +// ReceiveTaskStates receives full states for all tasks. +func (k *Kernel) ReceiveTaskStates() { + k.extMu.Lock() + k.tasks.PullFullState() + k.extMu.Unlock() +} + // Unpause ends the effect of a previous call to Pause. If Unpause is called // without a matching preceding call to Pause, Unpause may panic. func (k *Kernel) Unpause() { @@ -1511,7 +1519,7 @@ type SocketEntry struct { } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (s *SocketEntry) WeakRefGone() { +func (s *SocketEntry) WeakRefGone(context.Context) { s.k.extMu.Lock() s.k.sockets.Remove(s) s.k.extMu.Unlock() @@ -1600,7 +1608,7 @@ func (ctx supervisorContext) Value(key interface{}) interface{} { return vfs.VirtualDentry{} } mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2() - defer mntns.DecRef() + defer mntns.DecRef(ctx) // Root() takes a reference on the root dirent for us. return mntns.Root() case vfs.CtxMountNamespace: diff --git a/pkg/sentry/kernel/pipe/node.go b/pkg/sentry/kernel/pipe/node.go index 4b688c627..6497dc4ba 100644 --- a/pkg/sentry/kernel/pipe/node.go +++ b/pkg/sentry/kernel/pipe/node.go @@ -93,7 +93,7 @@ func (i *inodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.Fi if i.p.isNamed && !flags.NonBlocking && !i.p.HasWriters() { if !waitFor(&i.mu, &i.wWakeup, ctx) { - r.DecRef() + r.DecRef(ctx) return nil, syserror.ErrInterrupted } } @@ -111,12 +111,12 @@ func (i *inodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.Fi // On a nonblocking, write-only open, the open fails with ENXIO if the // read side isn't open yet. if flags.NonBlocking { - w.DecRef() + w.DecRef(ctx) return nil, syserror.ENXIO } if !waitFor(&i.mu, &i.rWakeup, ctx) { - w.DecRef() + w.DecRef(ctx) return nil, syserror.ErrInterrupted } } diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go index ab75a87ff..ce0db5583 100644 --- a/pkg/sentry/kernel/pipe/node_test.go +++ b/pkg/sentry/kernel/pipe/node_test.go @@ -167,7 +167,7 @@ func TestClosedReaderBlocksWriteOpen(t *testing.T) { f := NewInodeOperations(ctx, perms, newNamedPipe(t)) rFile, _ := testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, NonBlocking: true}, nil) - rFile.DecRef() + rFile.DecRef(ctx) wDone := make(chan struct{}) // This open for write should block because the reader is now gone. diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 79645d7d2..297e8f28f 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -152,7 +152,7 @@ func NewConnectedPipe(ctx context.Context, sizeBytes, atomicIOBytes int64) (*fs. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, ms, sattr), fmt.Sprintf("pipe:[%d]", ino)) // The p.Open calls below will each take a reference on the Dirent. We // must drop the one we already have. - defer d.DecRef() + defer d.DecRef(ctx) return p.Open(ctx, d, fs.FileFlags{Read: true}), p.Open(ctx, d, fs.FileFlags{Write: true}) } diff --git a/pkg/sentry/kernel/pipe/pipe_test.go b/pkg/sentry/kernel/pipe/pipe_test.go index bda739dbe..fe97e9800 100644 --- a/pkg/sentry/kernel/pipe/pipe_test.go +++ b/pkg/sentry/kernel/pipe/pipe_test.go @@ -27,8 +27,8 @@ import ( func TestPipeRW(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := []byte("here's some bytes") wantN := int64(len(msg)) @@ -47,8 +47,8 @@ func TestPipeRW(t *testing.T) { func TestPipeReadBlock(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) n, err := r.Readv(ctx, usermem.BytesIOSequence(make([]byte, 1))) if n != 0 || err != syserror.ErrWouldBlock { @@ -62,8 +62,8 @@ func TestPipeWriteBlock(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, capacity, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := make([]byte, capacity+1) n, err := w.Writev(ctx, usermem.BytesIOSequence(msg)) @@ -77,8 +77,8 @@ func TestPipeWriteUntilEnd(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, atomicIOBytes, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := []byte("here's some bytes") diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index aacf28da2..6d58b682f 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -33,7 +33,7 @@ import ( // the old fs architecture. // Release cleans up the pipe's state. -func (p *Pipe) Release() { +func (p *Pipe) Release(context.Context) { p.rClose() p.wClose() diff --git a/pkg/sentry/kernel/pipe/reader.go b/pkg/sentry/kernel/pipe/reader.go index 7724b4452..ac18785c0 100644 --- a/pkg/sentry/kernel/pipe/reader.go +++ b/pkg/sentry/kernel/pipe/reader.go @@ -15,6 +15,7 @@ package pipe import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/waiter" ) @@ -29,7 +30,7 @@ type Reader struct { // Release implements fs.FileOperations.Release. // // This overrides ReaderWriter.Release. -func (r *Reader) Release() { +func (r *Reader) Release(context.Context) { r.Pipe.rClose() // Wake up writers. diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 45d4c5fc1..28f998e45 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -101,7 +101,7 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s // If this pipe is being opened as blocking and there's no // writer, we have to wait for a writer to open the other end. if vp.pipe.isNamed && statusFlags&linux.O_NONBLOCK == 0 && !vp.pipe.HasWriters() && !waitFor(&vp.mu, &vp.wWakeup, ctx) { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EINTR } @@ -112,12 +112,12 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s // Non-blocking, write-only opens fail with ENXIO when the read // side isn't open yet. if statusFlags&linux.O_NONBLOCK != 0 { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.ENXIO } // Wait for a reader to open the other end. if !waitFor(&vp.mu, &vp.rWakeup, ctx) { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EINTR } } @@ -169,7 +169,7 @@ type VFSPipeFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *VFSPipeFD) Release() { +func (fd *VFSPipeFD) Release(context.Context) { var event waiter.EventMask if fd.vfsfd.IsReadable() { fd.pipe.rClose() diff --git a/pkg/sentry/kernel/pipe/writer.go b/pkg/sentry/kernel/pipe/writer.go index 5bc6aa931..ef4b70ca3 100644 --- a/pkg/sentry/kernel/pipe/writer.go +++ b/pkg/sentry/kernel/pipe/writer.go @@ -15,6 +15,7 @@ package pipe import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/waiter" ) @@ -29,7 +30,7 @@ type Writer struct { // Release implements fs.FileOperations.Release. // // This overrides ReaderWriter.Release. -func (w *Writer) Release() { +func (w *Writer) Release(context.Context) { w.Pipe.wClose() // Wake up readers. diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index e23e796ef..6c03d9041 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -1018,6 +1018,9 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + ar := ars.Head() n, err := target.Arch().PtraceGetRegSet(uintptr(addr), &usermem.IOReadWriter{ Ctx: t, @@ -1044,10 +1047,14 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + + mm := t.MemoryManager() + t.p.PullFullState(mm.AddressSpace(), t.Arch()) + ar := ars.Head() n, err := target.Arch().PtraceSetRegSet(uintptr(addr), &usermem.IOReadWriter{ Ctx: t, - IO: t.MemoryManager(), + IO: mm, Addr: ar.Start, Opts: usermem.IOOpts{ AddressSpaceActive: true, @@ -1056,6 +1063,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + t.p.FloatingPointStateChanged() ar.End -= usermem.Addr(n) return t.CopyOutIovecs(data, usermem.AddrRangeSeqOf(ar)) diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index 0e19286de..5c4c622c2 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -16,6 +16,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/syserror" @@ -70,7 +71,7 @@ func (s *Session) incRef() { // // Precondition: callers must hold TaskSet.mu for writing. func (s *Session) decRef() { - s.refs.DecRefWithDestructor(func() { + s.refs.DecRefWithDestructor(nil, func(context.Context) { // Remove translations from the leader. for ns := s.leader.pidns; ns != nil; ns = ns.parent { id := ns.sids[s] @@ -162,7 +163,7 @@ func (pg *ProcessGroup) decRefWithParent(parentPG *ProcessGroup) { } alive := true - pg.refs.DecRefWithDestructor(func() { + pg.refs.DecRefWithDestructor(nil, func(context.Context) { alive = false // don't bother with handleOrphan. // Remove translations from the originator. diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 55b4c2cdb..13ec7afe0 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -431,8 +431,8 @@ func (s *Shm) InodeID() uint64 { // DecRef overrides refs.RefCount.DecRef with a destructor. // // Precondition: Caller must not hold s.mu. -func (s *Shm) DecRef() { - s.DecRefWithDestructor(s.destroy) +func (s *Shm) DecRef(ctx context.Context) { + s.DecRefWithDestructor(ctx, s.destroy) } // Msync implements memmap.MappingIdentity.Msync. Msync is a no-op for shm @@ -642,7 +642,7 @@ func (s *Shm) Set(ctx context.Context, ds *linux.ShmidDS) error { return nil } -func (s *Shm) destroy() { +func (s *Shm) destroy(context.Context) { s.mfp.MemoryFile().DecRef(s.fr) s.registry.remove(s) } @@ -651,7 +651,7 @@ func (s *Shm) destroy() { // destroyed once it has no references. MarkDestroyed may be called multiple // times, and is safe to call after a segment has already been destroyed. See // shmctl(IPC_RMID). -func (s *Shm) MarkDestroyed() { +func (s *Shm) MarkDestroyed(ctx context.Context) { s.registry.dissociateKey(s) s.mu.Lock() @@ -663,7 +663,7 @@ func (s *Shm) MarkDestroyed() { // // N.B. This cannot be the final DecRef, as the caller also // holds a reference. - s.DecRef() + s.DecRef(ctx) return } } diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index 8243bb93e..b07e1c1bd 100644 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -76,7 +76,7 @@ func New(ctx context.Context, mask linux.SignalSet) (*fs.File, error) { } // Release implements fs.FileOperations.Release. -func (s *SignalOperations) Release() {} +func (s *SignalOperations) Release(context.Context) {} // Mask returns the signal mask. func (s *SignalOperations) Mask() linux.SignalSet { diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index c4db05bd8..5aee699e7 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -730,17 +730,17 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock { func (t *Task) IsChrooted() bool { if VFS2Enabled { realRoot := t.mountNamespaceVFS2.Root() - defer realRoot.DecRef() + defer realRoot.DecRef(t) root := t.fsContext.RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) return root != realRoot } realRoot := t.tg.mounts.Root() - defer realRoot.DecRef() + defer realRoot.DecRef(t) root := t.fsContext.RootDirectory() if root != nil { - defer root.DecRef() + defer root.DecRef(t) } return root != realRoot } diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index e1ecca99e..9d7a9128f 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -161,6 +161,10 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { return 0, nil, syserror.EINVAL } + // Pull task registers and FPU state, a cloned task will inherit the + // state of the current task. + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + // "If CLONE_NEWUSER is specified along with other CLONE_NEW* flags in a // single clone(2) or unshare(2) call, the user namespace is guaranteed to // be created first, giving the child (clone(2)) or caller (unshare(2)) @@ -237,7 +241,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { var fdTable *FDTable if opts.NewFiles { - fdTable = t.fdTable.Fork() + fdTable = t.fdTable.Fork(t) } else { fdTable = t.fdTable fdTable.IncRef() @@ -294,7 +298,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { nt, err := t.tg.pidns.owner.NewTask(cfg) if err != nil { if opts.NewThreadGroup { - tg.release() + tg.release(t) } return 0, nil, err } @@ -510,7 +514,7 @@ func (t *Task) Unshare(opts *SharingOptions) error { var oldFDTable *FDTable if opts.NewFiles { oldFDTable = t.fdTable - t.fdTable = oldFDTable.Fork() + t.fdTable = oldFDTable.Fork(t) } var oldFSContext *FSContext if opts.NewFSContext { @@ -519,10 +523,10 @@ func (t *Task) Unshare(opts *SharingOptions) error { } t.mu.Unlock() if oldFDTable != nil { - oldFDTable.DecRef() + oldFDTable.DecRef(t) } if oldFSContext != nil { - oldFSContext.DecRef() + oldFSContext.DecRef(t) } return nil } diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 7803b98d0..47c28b8ff 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -199,11 +199,11 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.Unlock() oldFDTable := t.fdTable - t.fdTable = t.fdTable.Fork() - oldFDTable.DecRef() + t.fdTable = t.fdTable.Fork(t) + oldFDTable.DecRef(t) // Remove FDs with the CloseOnExec flag set. - t.fdTable.RemoveIf(func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { + t.fdTable.RemoveIf(t, func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { return flags.CloseOnExec }) diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index 231ac548a..c165d6cb1 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -269,12 +269,12 @@ func (*runExitMain) execute(t *Task) taskRunState { // Releasing the MM unblocks a blocked CLONE_VFORK parent. t.unstopVforkParent() - t.fsContext.DecRef() - t.fdTable.DecRef() + t.fsContext.DecRef(t) + t.fdTable.DecRef(t) t.mu.Lock() if t.mountNamespaceVFS2 != nil { - t.mountNamespaceVFS2.DecRef() + t.mountNamespaceVFS2.DecRef(t) t.mountNamespaceVFS2 = nil } t.mu.Unlock() @@ -282,7 +282,7 @@ func (*runExitMain) execute(t *Task) taskRunState { // If this is the last task to exit from the thread group, release the // thread group's resources. if lastExiter { - t.tg.release() + t.tg.release(t) } // Detach tracees. diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index eeccaa197..d23cea802 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -27,6 +27,9 @@ const ( // maxStackDebugBytes is the maximum number of user stack bytes that may be // printed by debugDumpStack. maxStackDebugBytes = 1024 + // maxCodeDebugBytes is the maximum number of user code bytes that may be + // printed by debugDumpCode. + maxCodeDebugBytes = 128 ) // Infof logs an formatted info message by calling log.Infof. @@ -61,6 +64,7 @@ func (t *Task) IsLogging(level log.Level) bool { func (t *Task) DebugDumpState() { t.debugDumpRegisters() t.debugDumpStack() + t.debugDumpCode() if mm := t.MemoryManager(); mm != nil { t.Debugf("Mappings:\n%s", mm) } @@ -128,6 +132,45 @@ func (t *Task) debugDumpStack() { } } +// debugDumpCode logs user code contents at log level debug. +// +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) debugDumpCode() { + if !t.IsLogging(log.Debug) { + return + } + m := t.MemoryManager() + if m == nil { + t.Debugf("Memory manager for task is gone, skipping application code dump.") + return + } + t.Debugf("Code:") + // Print code on both sides of the instruction register. + start := usermem.Addr(t.Arch().IP()) - maxCodeDebugBytes/2 + // Round addr down to a 16-byte boundary. + start &= ^usermem.Addr(15) + // Print 16 bytes per line, one byte at a time. + for offset := uint64(0); offset < maxCodeDebugBytes; offset += 16 { + addr, ok := start.AddLength(offset) + if !ok { + break + } + var data [16]byte + n, err := m.CopyIn(t, addr, data[:], usermem.IOOpts{ + IgnorePermissions: true, + }) + // Print as much of the line as we can, even if an error was + // encountered. + if n > 0 { + t.Debugf("%x: % x", addr, data[:n]) + } + if err != nil { + t.Debugf("Error reading stack at address %x: %v", addr+usermem.Addr(n), err) + break + } + } +} + // trace definitions. // // Note that all region names are prefixed by ':' in order to ensure that they @@ -203,6 +246,6 @@ func (t *Task) traceExecEvent(tc *TaskContext) { trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>") return } - defer file.DecRef() + defer file.DecRef(t) trace.Logf(t.traceContext, traceCategory, "exec: %s", file.PathnameWithDeleted(t)) } diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 7d4f44caf..abaf29216 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -260,7 +260,7 @@ func (app *runApp) execute(t *Task) taskRunState { region := trace.StartRegion(t.traceContext, runRegion) t.accountTaskGoroutineEnter(TaskGoroutineRunningApp) - info, at, err := t.p.Switch(t.MemoryManager().AddressSpace(), t.Arch(), t.rseqCPU) + info, at, err := t.p.Switch(t, t.MemoryManager(), t.Arch(), t.rseqCPU) t.accountTaskGoroutineLeave(TaskGoroutineRunningApp) region.End() diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 79766cafe..2180fd27d 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -255,10 +255,11 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) } } + mm := t.MemoryManager() // Set up the signal handler. If we have a saved signal mask, the signal // handler should run with the current mask, but sigreturn should restore // the saved one. - st := &arch.Stack{t.Arch(), t.MemoryManager(), sp} + st := &arch.Stack{t.Arch(), mm, sp} mask := t.signalMask if t.haveSavedSignalMask { mask = t.savedSignalMask @@ -273,12 +274,13 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) // Please see the linux code as reference: // linux/arch/arm64/kernel/signal.c:setup_return() if act.Flags&linux.SA_RESTORER == 0 { - act.Restorer = t.MemoryManager().VDSOSigReturn() + act.Restorer = mm.VDSOSigReturn() } if err := t.Arch().SignalSetup(st, &act, info, &alt, mask); err != nil { return err } + t.p.FloatingPointStateChanged() t.haveSavedSignalMask = false // Add our signal mask. @@ -310,6 +312,7 @@ func (t *Task) SignalReturn(rt bool) (*SyscallControl, error) { // Restore our signal mask. SIGKILL and SIGSTOP should not be blocked. t.SetSignalMask(sigset &^ UnblockableSignals) + t.p.FloatingPointStateChanged() return ctrlResume, nil } @@ -636,6 +639,7 @@ func (t *Task) SetSavedSignalMask(mask linux.SignalSet) { // SignalStack returns the task-private signal stack. func (t *Task) SignalStack() arch.SignalStack { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) alt := t.signalStack if t.onSignalStack(alt) { alt.Flags |= arch.SignalStackFlagOnStack @@ -1050,6 +1054,8 @@ func (*runInterrupt) execute(t *Task) taskRunState { // Are there signals pending? if info := t.dequeueSignalLocked(t.signalMask); info != nil { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + if linux.SignalSetOf(linux.Signal(info.Signo))&StopSignals != 0 { // Indicate that we've dequeued a stop signal before unlocking the // signal mutex; initiateGroupStop will check for races with diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 8485fb4b6..64c1e120a 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -102,10 +102,10 @@ func (ts *TaskSet) NewTask(cfg *TaskConfig) (*Task, error) { t, err := ts.newTask(cfg) if err != nil { cfg.TaskContext.release() - cfg.FSContext.DecRef() - cfg.FDTable.DecRef() + cfg.FSContext.DecRef(t) + cfg.FDTable.DecRef(t) if cfg.MountNamespaceVFS2 != nil { - cfg.MountNamespaceVFS2.DecRef() + cfg.MountNamespaceVFS2.DecRef(t) } return nil, err } diff --git a/pkg/sentry/kernel/task_stop.go b/pkg/sentry/kernel/task_stop.go index 10c6e455c..296735d32 100644 --- a/pkg/sentry/kernel/task_stop.go +++ b/pkg/sentry/kernel/task_stop.go @@ -205,6 +205,22 @@ func (ts *TaskSet) BeginExternalStop() { } } +// PullFullState receives full states for all tasks. +func (ts *TaskSet) PullFullState() { + ts.mu.Lock() + defer ts.mu.Unlock() + if ts.Root == nil { + return + } + for t := range ts.Root.tids { + t.Activate() + if mm := t.MemoryManager(); mm != nil { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + } + t.Deactivate() + } +} + // EndExternalStop indicates the end of an external stop started by a previous // call to TaskSet.BeginExternalStop. EndExternalStop does not wait for task // goroutines to resume. diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 4dfd2c990..0b34c0099 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -308,7 +308,7 @@ func (tg *ThreadGroup) Limits() *limits.LimitSet { } // release releases the thread group's resources. -func (tg *ThreadGroup) release() { +func (tg *ThreadGroup) release(t *Task) { // Timers must be destroyed without holding the TaskSet or signal mutexes // since timers send signals with Timer.mu locked. tg.itimerRealTimer.Destroy() @@ -325,7 +325,7 @@ func (tg *ThreadGroup) release() { it.DestroyTimer() } if tg.mounts != nil { - tg.mounts.DecRef() + tg.mounts.DecRef(t) } } |