diff options
Diffstat (limited to 'pkg/sentry/kernel/kernel.go')
| -rw-r--r-- | pkg/sentry/kernel/kernel.go | 103 |
1 files changed, 71 insertions, 32 deletions
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 402aa1718..d6c21adb7 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -220,13 +220,18 @@ type Kernel struct { // danglingEndpoints is used to save / restore tcpip.DanglingEndpoints. danglingEndpoints struct{} `state:".([]tcpip.Endpoint)"` - // sockets is the list of all network sockets the system. Protected by - // extMu. + // sockets is the list of all network sockets in the system. + // Protected by extMu. + // TODO(gvisor.dev/issue/1624): Only used by VFS1. sockets socketList - // nextSocketEntry is the next entry number to use in sockets. Protected + // socketsVFS2 records all network sockets in the system. Protected by + // extMu. + socketsVFS2 map[*vfs.FileDescription]*SocketRecord + + // nextSocketRecord is the next entry number to use in sockets. Protected // by extMu. - nextSocketEntry uint64 + nextSocketRecord uint64 // deviceRegistry is used to save/restore device.SimpleDevices. deviceRegistry struct{} `state:".(*device.Registry)"` @@ -414,6 +419,8 @@ func (k *Kernel) Init(args InitKernelArgs) error { return fmt.Errorf("failed to create sockfs mount: %v", err) } k.socketMount = socketMount + + k.socketsVFS2 = make(map[*vfs.FileDescription]*SocketRecord) } return nil @@ -507,6 +514,10 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // flushMountSourceRefs flushes the MountSources for all mounted filesystems // and open FDs. func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { + if VFS2Enabled { + return nil // Not relevant. + } + // Flush all mount sources for currently mounted filesystems in each task. flushed := make(map[*fs.MountNamespace]struct{}) k.tasks.mu.RLock() @@ -533,11 +544,6 @@ func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { // // Precondition: Must be called with the kernel paused. func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.FileDescription) error) (err error) { - // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - if VFS2Enabled { - return nil - } - ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { @@ -556,6 +562,10 @@ func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.Fi func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if VFS2Enabled { + return nil + } + return ts.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil @@ -888,17 +898,18 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, opener fsbridge.Lookup fsContext *FSContext mntns *fs.MountNamespace + mntnsVFS2 *vfs.MountNamespace ) if VFS2Enabled { - mntnsVFS2 := args.MountNamespaceVFS2 + mntnsVFS2 = args.MountNamespaceVFS2 if mntnsVFS2 == nil { // MountNamespaceVFS2 adds a reference to the namespace, which is // transferred to the new process. mntnsVFS2 = k.globalInit.Leader().MountNamespaceVFS2() } // Get the root directory from the MountNamespace. - root := args.MountNamespaceVFS2.Root() + root := mntnsVFS2.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. defer root.DecRef(ctx) @@ -1008,7 +1019,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, UTSNamespace: args.UTSNamespace, IPCNamespace: args.IPCNamespace, AbstractSocketNamespace: args.AbstractSocketNamespace, - MountNamespaceVFS2: args.MountNamespaceVFS2, + MountNamespaceVFS2: mntnsVFS2, ContainerID: args.ContainerID, } t, err := k.tasks.NewTask(config) @@ -1508,20 +1519,27 @@ func (k *Kernel) SupervisorContext() context.Context { } } -// SocketEntry represents a socket recorded in Kernel.sockets. It implements +// SocketRecord represents a socket recorded in Kernel.socketsVFS2. +// +// +stateify savable +type SocketRecord struct { + k *Kernel + Sock *refs.WeakRef // TODO(gvisor.dev/issue/1624): Only used by VFS1. + SockVFS2 *vfs.FileDescription // Only used by VFS2. + ID uint64 // Socket table entry number. +} + +// SocketRecordVFS1 represents a socket recorded in Kernel.sockets. It implements // refs.WeakRefUser for sockets stored in the socket table. // // +stateify savable -type SocketEntry struct { +type SocketRecordVFS1 struct { socketEntry - k *Kernel - Sock *refs.WeakRef - SockVFS2 *vfs.FileDescription - ID uint64 // Socket table entry number. + SocketRecord } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (s *SocketEntry) WeakRefGone(context.Context) { +func (s *SocketRecordVFS1) WeakRefGone(context.Context) { s.k.extMu.Lock() s.k.sockets.Remove(s) s.k.extMu.Unlock() @@ -1532,9 +1550,14 @@ func (s *SocketEntry) WeakRefGone(context.Context) { // Precondition: Caller must hold a reference to sock. func (k *Kernel) RecordSocket(sock *fs.File) { k.extMu.Lock() - id := k.nextSocketEntry - k.nextSocketEntry++ - s := &SocketEntry{k: k, ID: id} + id := k.nextSocketRecord + k.nextSocketRecord++ + s := &SocketRecordVFS1{ + SocketRecord: SocketRecord{ + k: k, + ID: id, + }, + } s.Sock = refs.NewWeakRef(sock, s) k.sockets.PushBack(s) k.extMu.Unlock() @@ -1546,29 +1569,45 @@ func (k *Kernel) RecordSocket(sock *fs.File) { // Precondition: Caller must hold a reference to sock. // // Note that the socket table will not hold a reference on the -// vfs.FileDescription, because we do not support weak refs on VFS2 files. +// vfs.FileDescription. func (k *Kernel) RecordSocketVFS2(sock *vfs.FileDescription) { k.extMu.Lock() - id := k.nextSocketEntry - k.nextSocketEntry++ - s := &SocketEntry{ + if _, ok := k.socketsVFS2[sock]; ok { + panic(fmt.Sprintf("Socket %p added twice", sock)) + } + id := k.nextSocketRecord + k.nextSocketRecord++ + s := &SocketRecord{ k: k, ID: id, SockVFS2: sock, } - k.sockets.PushBack(s) + k.socketsVFS2[sock] = s + k.extMu.Unlock() +} + +// DeleteSocketVFS2 removes a VFS2 socket from the system-wide socket table. +func (k *Kernel) DeleteSocketVFS2(sock *vfs.FileDescription) { + k.extMu.Lock() + delete(k.socketsVFS2, sock) k.extMu.Unlock() } // ListSockets returns a snapshot of all sockets. // -// Callers of ListSockets() in VFS2 should use SocketEntry.SockVFS2.TryIncRef() +// Callers of ListSockets() in VFS2 should use SocketRecord.SockVFS2.TryIncRef() // to get a reference on a socket in the table. -func (k *Kernel) ListSockets() []*SocketEntry { +func (k *Kernel) ListSockets() []*SocketRecord { k.extMu.Lock() - var socks []*SocketEntry - for s := k.sockets.Front(); s != nil; s = s.Next() { - socks = append(socks, s) + var socks []*SocketRecord + if VFS2Enabled { + for _, s := range k.socketsVFS2 { + socks = append(socks, s) + } + } else { + for s := k.sockets.Front(); s != nil; s = s.Next() { + socks = append(socks, &s.SocketRecord) + } } k.extMu.Unlock() return socks |