diff options
author | gVisor bot <gvisor-bot@google.com> | 2020-09-29 18:54:47 +0000 |
---|---|---|
committer | gVisor bot <gvisor-bot@google.com> | 2020-09-29 18:54:47 +0000 |
commit | fbaf56af454c281979112ea542784155f77ede96 (patch) | |
tree | bad5780ddb0582a60e4b5e234e0e793863e51761 /pkg/sentry/kernel | |
parent | 44f50b440fa6241105a90f17501417ed0b34f172 (diff) | |
parent | 7d64bc1fdf538869de5964475110a69580b481ad (diff) |
Merge release-20200921.0-66-g7d64bc1fd (automated)
Diffstat (limited to 'pkg/sentry/kernel')
-rw-r--r-- | pkg/sentry/kernel/kernel.go | 83 | ||||
-rw-r--r-- | pkg/sentry/kernel/kernel_state_autogen.go | 103 | ||||
-rw-r--r-- | pkg/sentry/kernel/socket_list.go | 32 |
3 files changed, 140 insertions, 78 deletions
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 08bb5bd12..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 @@ -1512,20 +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() @@ -1536,9 +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() @@ -1550,29 +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 diff --git a/pkg/sentry/kernel/kernel_state_autogen.go b/pkg/sentry/kernel/kernel_state_autogen.go index d0ff135d7..f20800960 100644 --- a/pkg/sentry/kernel/kernel_state_autogen.go +++ b/pkg/sentry/kernel/kernel_state_autogen.go @@ -297,7 +297,8 @@ func (x *Kernel) StateFields() []string { "netlinkPorts", "danglingEndpoints", "sockets", - "nextSocketEntry", + "socketsVFS2", + "nextSocketRecord", "deviceRegistry", "DirentCacheLimiter", "SpecialOpts", @@ -317,7 +318,7 @@ func (x *Kernel) StateSave(m state.Sink) { var danglingEndpoints []tcpip.Endpoint = x.saveDanglingEndpoints() m.SaveValue(24, danglingEndpoints) var deviceRegistry *device.Registry = x.saveDeviceRegistry() - m.SaveValue(27, deviceRegistry) + m.SaveValue(28, deviceRegistry) m.Save(0, &x.featureSet) m.Save(1, &x.timekeeper) m.Save(2, &x.tasks) @@ -343,15 +344,16 @@ func (x *Kernel) StateSave(m state.Sink) { m.Save(22, &x.nextInotifyCookie) m.Save(23, &x.netlinkPorts) m.Save(25, &x.sockets) - m.Save(26, &x.nextSocketEntry) - m.Save(28, &x.DirentCacheLimiter) - m.Save(29, &x.SpecialOpts) - m.Save(30, &x.vfs) - m.Save(31, &x.hostMount) - m.Save(32, &x.pipeMount) - m.Save(33, &x.shmMount) - m.Save(34, &x.socketMount) - m.Save(35, &x.SleepForAddressSpaceActivation) + m.Save(26, &x.socketsVFS2) + m.Save(27, &x.nextSocketRecord) + m.Save(29, &x.DirentCacheLimiter) + m.Save(30, &x.SpecialOpts) + m.Save(31, &x.vfs) + m.Save(32, &x.hostMount) + m.Save(33, &x.pipeMount) + m.Save(34, &x.shmMount) + m.Save(35, &x.socketMount) + m.Save(36, &x.SleepForAddressSpaceActivation) } func (x *Kernel) afterLoad() {} @@ -382,26 +384,26 @@ func (x *Kernel) StateLoad(m state.Source) { m.Load(22, &x.nextInotifyCookie) m.Load(23, &x.netlinkPorts) m.Load(25, &x.sockets) - m.Load(26, &x.nextSocketEntry) - m.Load(28, &x.DirentCacheLimiter) - m.Load(29, &x.SpecialOpts) - m.Load(30, &x.vfs) - m.Load(31, &x.hostMount) - m.Load(32, &x.pipeMount) - m.Load(33, &x.shmMount) - m.Load(34, &x.socketMount) - m.Load(35, &x.SleepForAddressSpaceActivation) + m.Load(26, &x.socketsVFS2) + m.Load(27, &x.nextSocketRecord) + m.Load(29, &x.DirentCacheLimiter) + m.Load(30, &x.SpecialOpts) + m.Load(31, &x.vfs) + m.Load(32, &x.hostMount) + m.Load(33, &x.pipeMount) + m.Load(34, &x.shmMount) + m.Load(35, &x.socketMount) + m.Load(36, &x.SleepForAddressSpaceActivation) m.LoadValue(24, new([]tcpip.Endpoint), func(y interface{}) { x.loadDanglingEndpoints(y.([]tcpip.Endpoint)) }) - m.LoadValue(27, new(*device.Registry), func(y interface{}) { x.loadDeviceRegistry(y.(*device.Registry)) }) + m.LoadValue(28, new(*device.Registry), func(y interface{}) { x.loadDeviceRegistry(y.(*device.Registry)) }) } -func (x *SocketEntry) StateTypeName() string { - return "pkg/sentry/kernel.SocketEntry" +func (x *SocketRecord) StateTypeName() string { + return "pkg/sentry/kernel.SocketRecord" } -func (x *SocketEntry) StateFields() []string { +func (x *SocketRecord) StateFields() []string { return []string{ - "socketEntry", "k", "Sock", "SockVFS2", @@ -409,25 +411,49 @@ func (x *SocketEntry) StateFields() []string { } } -func (x *SocketEntry) beforeSave() {} +func (x *SocketRecord) beforeSave() {} -func (x *SocketEntry) StateSave(m state.Sink) { +func (x *SocketRecord) StateSave(m state.Sink) { + x.beforeSave() + m.Save(0, &x.k) + m.Save(1, &x.Sock) + m.Save(2, &x.SockVFS2) + m.Save(3, &x.ID) +} + +func (x *SocketRecord) afterLoad() {} + +func (x *SocketRecord) StateLoad(m state.Source) { + m.Load(0, &x.k) + m.Load(1, &x.Sock) + m.Load(2, &x.SockVFS2) + m.Load(3, &x.ID) +} + +func (x *SocketRecordVFS1) StateTypeName() string { + return "pkg/sentry/kernel.SocketRecordVFS1" +} + +func (x *SocketRecordVFS1) StateFields() []string { + return []string{ + "socketEntry", + "SocketRecord", + } +} + +func (x *SocketRecordVFS1) beforeSave() {} + +func (x *SocketRecordVFS1) StateSave(m state.Sink) { x.beforeSave() m.Save(0, &x.socketEntry) - m.Save(1, &x.k) - m.Save(2, &x.Sock) - m.Save(3, &x.SockVFS2) - m.Save(4, &x.ID) + m.Save(1, &x.SocketRecord) } -func (x *SocketEntry) afterLoad() {} +func (x *SocketRecordVFS1) afterLoad() {} -func (x *SocketEntry) StateLoad(m state.Source) { +func (x *SocketRecordVFS1) StateLoad(m state.Source) { m.Load(0, &x.socketEntry) - m.Load(1, &x.k) - m.Load(2, &x.Sock) - m.Load(3, &x.SockVFS2) - m.Load(4, &x.ID) + m.Load(1, &x.SocketRecord) } func (x *pendingSignals) StateTypeName() string { @@ -2264,7 +2290,8 @@ func init() { state.Register((*FSContextRefs)(nil)) state.Register((*IPCNamespace)(nil)) state.Register((*Kernel)(nil)) - state.Register((*SocketEntry)(nil)) + state.Register((*SocketRecord)(nil)) + state.Register((*SocketRecordVFS1)(nil)) state.Register((*pendingSignals)(nil)) state.Register((*pendingSignalQueue)(nil)) state.Register((*pendingSignal)(nil)) diff --git a/pkg/sentry/kernel/socket_list.go b/pkg/sentry/kernel/socket_list.go index d2d4307a1..246fba405 100644 --- a/pkg/sentry/kernel/socket_list.go +++ b/pkg/sentry/kernel/socket_list.go @@ -13,7 +13,7 @@ type socketElementMapper struct{} // This default implementation should be inlined. // //go:nosplit -func (socketElementMapper) linkerFor(elem *SocketEntry) *SocketEntry { return elem } +func (socketElementMapper) linkerFor(elem *SocketRecordVFS1) *SocketRecordVFS1 { return elem } // List is an intrusive list. Entries can be added to or removed from the list // in O(1) time and with no additional memory allocations. @@ -27,8 +27,8 @@ func (socketElementMapper) linkerFor(elem *SocketEntry) *SocketEntry { return el // // +stateify savable type socketList struct { - head *SocketEntry - tail *SocketEntry + head *SocketRecordVFS1 + tail *SocketRecordVFS1 } // Reset resets list l to the empty state. @@ -43,12 +43,12 @@ func (l *socketList) Empty() bool { } // Front returns the first element of list l or nil. -func (l *socketList) Front() *SocketEntry { +func (l *socketList) Front() *SocketRecordVFS1 { return l.head } // Back returns the last element of list l or nil. -func (l *socketList) Back() *SocketEntry { +func (l *socketList) Back() *SocketRecordVFS1 { return l.tail } @@ -63,7 +63,7 @@ func (l *socketList) Len() (count int) { } // PushFront inserts the element e at the front of list l. -func (l *socketList) PushFront(e *SocketEntry) { +func (l *socketList) PushFront(e *SocketRecordVFS1) { linker := socketElementMapper{}.linkerFor(e) linker.SetNext(l.head) linker.SetPrev(nil) @@ -77,7 +77,7 @@ func (l *socketList) PushFront(e *SocketEntry) { } // PushBack inserts the element e at the back of list l. -func (l *socketList) PushBack(e *SocketEntry) { +func (l *socketList) PushBack(e *SocketRecordVFS1) { linker := socketElementMapper{}.linkerFor(e) linker.SetNext(nil) linker.SetPrev(l.tail) @@ -106,7 +106,7 @@ func (l *socketList) PushBackList(m *socketList) { } // InsertAfter inserts e after b. -func (l *socketList) InsertAfter(b, e *SocketEntry) { +func (l *socketList) InsertAfter(b, e *SocketRecordVFS1) { bLinker := socketElementMapper{}.linkerFor(b) eLinker := socketElementMapper{}.linkerFor(e) @@ -124,7 +124,7 @@ func (l *socketList) InsertAfter(b, e *SocketEntry) { } // InsertBefore inserts e before a. -func (l *socketList) InsertBefore(a, e *SocketEntry) { +func (l *socketList) InsertBefore(a, e *SocketRecordVFS1) { aLinker := socketElementMapper{}.linkerFor(a) eLinker := socketElementMapper{}.linkerFor(e) @@ -141,7 +141,7 @@ func (l *socketList) InsertBefore(a, e *SocketEntry) { } // Remove removes e from l. -func (l *socketList) Remove(e *SocketEntry) { +func (l *socketList) Remove(e *SocketRecordVFS1) { linker := socketElementMapper{}.linkerFor(e) prev := linker.Prev() next := linker.Next() @@ -168,26 +168,26 @@ func (l *socketList) Remove(e *SocketEntry) { // // +stateify savable type socketEntry struct { - next *SocketEntry - prev *SocketEntry + next *SocketRecordVFS1 + prev *SocketRecordVFS1 } // Next returns the entry that follows e in the list. -func (e *socketEntry) Next() *SocketEntry { +func (e *socketEntry) Next() *SocketRecordVFS1 { return e.next } // Prev returns the entry that precedes e in the list. -func (e *socketEntry) Prev() *SocketEntry { +func (e *socketEntry) Prev() *SocketRecordVFS1 { return e.prev } // SetNext assigns 'entry' as the entry that follows e in the list. -func (e *socketEntry) SetNext(elem *SocketEntry) { +func (e *socketEntry) SetNext(elem *SocketRecordVFS1) { e.next = elem } // SetPrev assigns 'entry' as the entry that precedes e in the list. -func (e *socketEntry) SetPrev(elem *SocketEntry) { +func (e *socketEntry) SetPrev(elem *SocketRecordVFS1) { e.prev = elem } |