summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/kernel')
-rw-r--r--pkg/sentry/kernel/BUILD12
-rw-r--r--pkg/sentry/kernel/abstract_socket_namespace.go10
-rw-r--r--pkg/sentry/kernel/fd_table.go6
-rw-r--r--pkg/sentry/kernel/fd_table_unsafe.go13
-rw-r--r--pkg/sentry/kernel/fs_context.go24
-rw-r--r--pkg/sentry/kernel/ipc_namespace.go2
-rw-r--r--pkg/sentry/kernel/kernel.go176
-rw-r--r--pkg/sentry/kernel/pipe/node_test.go8
-rw-r--r--pkg/sentry/kernel/pipe/pipe.go50
-rw-r--r--pkg/sentry/kernel/pipe/pipe_test.go11
-rw-r--r--pkg/sentry/kernel/pipe/vfs.go8
-rw-r--r--pkg/sentry/kernel/ptrace.go4
-rw-r--r--pkg/sentry/kernel/semaphore/semaphore.go75
-rw-r--r--pkg/sentry/kernel/sessions.go32
-rw-r--r--pkg/sentry/kernel/shm/BUILD4
-rw-r--r--pkg/sentry/kernel/task_clone.go2
-rw-r--r--pkg/sentry/kernel/task_usermem.go95
-rw-r--r--pkg/sentry/kernel/vdso.go2
18 files changed, 346 insertions, 188 deletions
diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index c0de72eef..90dd4a047 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -79,7 +79,7 @@ go_template_instance(
out = "fd_table_refs.go",
package = "kernel",
prefix = "FDTable",
- template = "//pkg/refs_vfs2:refs_template",
+ template = "//pkg/refsvfs2:refs_template",
types = {
"T": "FDTable",
},
@@ -90,7 +90,7 @@ go_template_instance(
out = "fs_context_refs.go",
package = "kernel",
prefix = "FSContext",
- template = "//pkg/refs_vfs2:refs_template",
+ template = "//pkg/refsvfs2:refs_template",
types = {
"T": "FSContext",
},
@@ -101,7 +101,7 @@ go_template_instance(
out = "ipc_namespace_refs.go",
package = "kernel",
prefix = "IPCNamespace",
- template = "//pkg/refs_vfs2:refs_template",
+ template = "//pkg/refsvfs2:refs_template",
types = {
"T": "IPCNamespace",
},
@@ -112,7 +112,7 @@ go_template_instance(
out = "process_group_refs.go",
package = "kernel",
prefix = "ProcessGroup",
- template = "//pkg/refs_vfs2:refs_template",
+ template = "//pkg/refsvfs2:refs_template",
types = {
"T": "ProcessGroup",
},
@@ -123,7 +123,7 @@ go_template_instance(
out = "session_refs.go",
package = "kernel",
prefix = "Session",
- template = "//pkg/refs_vfs2:refs_template",
+ template = "//pkg/refsvfs2:refs_template",
types = {
"T": "Session",
},
@@ -229,7 +229,7 @@ go_library(
"//pkg/marshal/primitive",
"//pkg/metric",
"//pkg/refs",
- "//pkg/refs_vfs2",
+ "//pkg/refsvfs2",
"//pkg/safemem",
"//pkg/secio",
"//pkg/sentry/arch",
diff --git a/pkg/sentry/kernel/abstract_socket_namespace.go b/pkg/sentry/kernel/abstract_socket_namespace.go
index 1b9721534..0ddbe5ff6 100644
--- a/pkg/sentry/kernel/abstract_socket_namespace.go
+++ b/pkg/sentry/kernel/abstract_socket_namespace.go
@@ -19,7 +19,7 @@ import (
"syscall"
"gvisor.dev/gvisor/pkg/context"
- "gvisor.dev/gvisor/pkg/refs_vfs2"
+ "gvisor.dev/gvisor/pkg/refsvfs2"
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sync"
)
@@ -27,7 +27,7 @@ import (
// +stateify savable
type abstractEndpoint struct {
ep transport.BoundEndpoint
- socket refs_vfs2.RefCounter
+ socket refsvfs2.RefCounter
name string
ns *AbstractSocketNamespace
}
@@ -57,7 +57,7 @@ func NewAbstractSocketNamespace() *AbstractSocketNamespace {
// its backing socket.
type boundEndpoint struct {
transport.BoundEndpoint
- socket refs_vfs2.RefCounter
+ socket refsvfs2.RefCounter
}
// Release implements transport.BoundEndpoint.Release.
@@ -89,7 +89,7 @@ func (a *AbstractSocketNamespace) BoundEndpoint(name string) transport.BoundEndp
//
// When the last reference managed by socket is dropped, ep may be removed from the
// namespace.
-func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, socket refs_vfs2.RefCounter) error {
+func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, socket refsvfs2.RefCounter) error {
a.mu.Lock()
defer a.mu.Unlock()
@@ -109,7 +109,7 @@ func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep tran
// Remove removes the specified socket at name from the abstract socket
// namespace, if it has not yet been replaced.
-func (a *AbstractSocketNamespace) Remove(name string, socket refs_vfs2.RefCounter) {
+func (a *AbstractSocketNamespace) Remove(name string, socket refsvfs2.RefCounter) {
a.mu.Lock()
defer a.mu.Unlock()
diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go
index 0ec7344cd..7aba31587 100644
--- a/pkg/sentry/kernel/fd_table.go
+++ b/pkg/sentry/kernel/fd_table.go
@@ -110,7 +110,7 @@ func (f *FDTable) saveDescriptorTable() map[int32]descriptor {
func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) {
ctx := context.Background()
- f.init() // Initialize table.
+ f.initNoLeakCheck() // Initialize table.
f.used = 0
for fd, d := range m {
if file, fileVFS2 := f.setAll(ctx, fd, d.file, d.fileVFS2, d.flags); file != nil || fileVFS2 != nil {
@@ -240,6 +240,10 @@ func (f *FDTable) String() string {
case fileVFS2 != nil:
vfsObj := fileVFS2.Mount().Filesystem().VirtualFilesystem()
+ vd := fileVFS2.VirtualDentry()
+ if vd.Dentry() == nil {
+ panic(fmt.Sprintf("fd %d (type %T) has nil dentry: %#v", fd, fileVFS2.Impl(), fileVFS2))
+ }
name, err := vfsObj.PathnameWithDeleted(ctx, vfs.VirtualDentry{}, fileVFS2.VirtualDentry())
if err != nil {
fmt.Fprintf(&buf, "<err: %v>\n", err)
diff --git a/pkg/sentry/kernel/fd_table_unsafe.go b/pkg/sentry/kernel/fd_table_unsafe.go
index da79e6627..3476551f3 100644
--- a/pkg/sentry/kernel/fd_table_unsafe.go
+++ b/pkg/sentry/kernel/fd_table_unsafe.go
@@ -31,14 +31,21 @@ type descriptorTable struct {
slice unsafe.Pointer `state:".(map[int32]*descriptor)"`
}
-// init initializes the table.
+// initNoLeakCheck initializes the table without enabling leak checking.
//
-// TODO(gvisor.dev/1486): Enable leak check for FDTable.
-func (f *FDTable) init() {
+// This is used when loading an FDTable after S/R, during which the ref count
+// object itself will enable leak checking if necessary.
+func (f *FDTable) initNoLeakCheck() {
var slice []unsafe.Pointer // Empty slice.
atomic.StorePointer(&f.slice, unsafe.Pointer(&slice))
}
+// init initializes the table with leak checking.
+func (f *FDTable) init() {
+ f.initNoLeakCheck()
+ f.EnableLeakCheck()
+}
+
// get gets a file entry.
//
// The boolean indicates whether this was in range.
diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go
index d46d1e1c1..41fb2a784 100644
--- a/pkg/sentry/kernel/fs_context.go
+++ b/pkg/sentry/kernel/fs_context.go
@@ -130,13 +130,15 @@ func (f *FSContext) Fork() *FSContext {
f.root.IncRef()
}
- return &FSContext{
+ ctx := &FSContext{
cwd: f.cwd,
root: f.root,
cwdVFS2: f.cwdVFS2,
rootVFS2: f.rootVFS2,
umask: f.umask,
}
+ ctx.EnableLeakCheck()
+ return ctx
}
// WorkingDirectory returns the current working directory.
@@ -147,19 +149,23 @@ func (f *FSContext) WorkingDirectory() *fs.Dirent {
f.mu.Lock()
defer f.mu.Unlock()
- f.cwd.IncRef()
+ if f.cwd != nil {
+ f.cwd.IncRef()
+ }
return f.cwd
}
// WorkingDirectoryVFS2 returns the current working directory.
//
-// This will return nil if called after f is destroyed, otherwise it will return
-// a Dirent with a reference taken.
+// This will return an empty vfs.VirtualDentry if called after f is
+// destroyed, otherwise it will return a Dirent with a reference taken.
func (f *FSContext) WorkingDirectoryVFS2() vfs.VirtualDentry {
f.mu.Lock()
defer f.mu.Unlock()
- f.cwdVFS2.IncRef()
+ if f.cwdVFS2.Ok() {
+ f.cwdVFS2.IncRef()
+ }
return f.cwdVFS2
}
@@ -218,13 +224,15 @@ func (f *FSContext) RootDirectory() *fs.Dirent {
// RootDirectoryVFS2 returns the current filesystem root.
//
-// This will return nil if called after f is destroyed, otherwise it will return
-// a Dirent with a reference taken.
+// This will return an empty vfs.VirtualDentry if called after f is
+// destroyed, otherwise it will return a Dirent with a reference taken.
func (f *FSContext) RootDirectoryVFS2() vfs.VirtualDentry {
f.mu.Lock()
defer f.mu.Unlock()
- f.rootVFS2.IncRef()
+ if f.rootVFS2.Ok() {
+ f.rootVFS2.IncRef()
+ }
return f.rootVFS2
}
diff --git a/pkg/sentry/kernel/ipc_namespace.go b/pkg/sentry/kernel/ipc_namespace.go
index 3f34ee0db..b87e40dd1 100644
--- a/pkg/sentry/kernel/ipc_namespace.go
+++ b/pkg/sentry/kernel/ipc_namespace.go
@@ -55,7 +55,7 @@ func (i *IPCNamespace) ShmRegistry() *shm.Registry {
return i.shms
}
-// DecRef implements refs_vfs2.RefCounter.DecRef.
+// DecRef implements refsvfs2.RefCounter.DecRef.
func (i *IPCNamespace) DecRef(ctx context.Context) {
i.IPCNamespaceRefs.DecRef(func() {
i.shms.Release(ctx)
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 0eb2bf7bd..9b2be44d4 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -430,9 +430,8 @@ func (k *Kernel) Init(args InitKernelArgs) error {
// SaveTo saves the state of k to w.
//
// Preconditions: The kernel must be paused throughout the call to SaveTo.
-func (k *Kernel) SaveTo(w wire.Writer) error {
+func (k *Kernel) SaveTo(ctx context.Context, w wire.Writer) error {
saveStart := time.Now()
- ctx := k.SupervisorContext()
// Do not allow other Kernel methods to affect it while it's being saved.
k.extMu.Lock()
@@ -446,38 +445,55 @@ func (k *Kernel) SaveTo(w wire.Writer) error {
k.mf.StartEvictions()
k.mf.WaitForEvictions()
- // Flush write operations on open files so data reaches backing storage.
- // This must come after MemoryFile eviction since eviction may cause file
- // writes.
- if err := k.tasks.flushWritesToFiles(ctx); err != nil {
- return err
- }
+ if VFS2Enabled {
+ // Discard unsavable mappings, such as those for host file descriptors.
+ if err := k.invalidateUnsavableMappings(ctx); err != nil {
+ return fmt.Errorf("failed to invalidate unsavable mappings: %v", err)
+ }
- // 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(ctx)
+ // Prepare filesystems for saving. This must be done after
+ // invalidateUnsavableMappings(), since dropping memory mappings may
+ // affect filesystem state (e.g. page cache reference counts).
+ if err := k.vfs.PrepareSave(ctx); err != nil {
+ return err
+ }
+ } else {
+ // Flush cached file writes to backing storage. This must come after
+ // MemoryFile eviction since eviction may cause file writes.
+ if err := k.flushWritesToFiles(ctx); err != nil {
+ return err
+ }
- // 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(ctx); err != nil {
- return err
- }
+ // 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(ctx)
- // Ensure that all inode and mount release operations have completed.
- fs.AsyncBarrier()
+ // 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(ctx); err != nil {
+ return err
+ }
- // Once all fs work has completed (flushed references have all been released),
- // reset mount mappings. This allows individual mounts to save how inodes map
- // to filesystem resources. Without this, fs.Inodes cannot be restored.
- fs.SaveInodeMappings()
+ // Ensure that all inode and mount release operations have completed.
+ fs.AsyncBarrier()
- // Discard unsavable mappings, such as those for host file descriptors.
- // This must be done after waiting for "asynchronous fs work", which
- // includes async I/O that may touch application memory.
- if err := k.invalidateUnsavableMappings(ctx); err != nil {
- return fmt.Errorf("failed to invalidate unsavable mappings: %v", err)
+ // Once all fs work has completed (flushed references have all been released),
+ // reset mount mappings. This allows individual mounts to save how inodes map
+ // to filesystem resources. Without this, fs.Inodes cannot be restored.
+ fs.SaveInodeMappings()
+
+ // Discard unsavable mappings, such as those for host file descriptors.
+ // This must be done after waiting for "asynchronous fs work", which
+ // includes async I/O that may touch application memory.
+ //
+ // TODO(gvisor.dev/issue/1624): This rationale is believed to be
+ // obsolete since AIO callbacks are now waited-for by Kernel.Pause(),
+ // but this order is conservatively retained for VFS1.
+ if err := k.invalidateUnsavableMappings(ctx); err != nil {
+ return fmt.Errorf("failed to invalidate unsavable mappings: %v", err)
+ }
}
// Save the CPUID FeatureSet before the rest of the kernel so we can
@@ -486,14 +502,14 @@ func (k *Kernel) SaveTo(w wire.Writer) error {
//
// N.B. This will also be saved along with the full kernel save below.
cpuidStart := time.Now()
- if _, err := state.Save(k.SupervisorContext(), w, k.FeatureSet()); err != nil {
+ if _, err := state.Save(ctx, w, k.FeatureSet()); err != nil {
return err
}
log.Infof("CPUID save took [%s].", time.Since(cpuidStart))
// Save the kernel state.
kernelStart := time.Now()
- stats, err := state.Save(k.SupervisorContext(), w, k)
+ stats, err := state.Save(ctx, w, k)
if err != nil {
return err
}
@@ -502,7 +518,7 @@ func (k *Kernel) SaveTo(w wire.Writer) error {
// Save the memory file's state.
memoryStart := time.Now()
- if err := k.mf.SaveTo(k.SupervisorContext(), w); err != nil {
+ if err := k.mf.SaveTo(ctx, w); err != nil {
return err
}
log.Infof("Memory save took [%s].", time.Since(memoryStart))
@@ -514,11 +530,9 @@ func (k *Kernel) SaveTo(w wire.Writer) error {
// flushMountSourceRefs flushes the MountSources for all mounted filesystems
// and open FDs.
+//
+// Preconditions: !VFS2Enabled.
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()
@@ -561,13 +575,9 @@ func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.Fi
return err
}
-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 {
+// Preconditions: !VFS2Enabled.
+func (k *Kernel) flushWritesToFiles(ctx context.Context) error {
+ return k.tasks.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error {
if flags := file.Flags(); !flags.Write {
return nil
}
@@ -589,37 +599,8 @@ func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error {
})
}
-// Preconditions: The kernel must be paused.
-func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error {
- invalidated := make(map[*mm.MemoryManager]struct{})
- k.tasks.mu.RLock()
- defer k.tasks.mu.RUnlock()
- for t := range k.tasks.Root.tids {
- // We can skip locking Task.mu here since the kernel is paused.
- if mm := t.tc.MemoryManager; mm != nil {
- if _, ok := invalidated[mm]; !ok {
- if err := mm.InvalidateUnsavable(ctx); err != nil {
- return err
- }
- invalidated[mm] = struct{}{}
- }
- }
- // I really wish we just had a sync.Map of all MMs...
- if r, ok := t.runState.(*runSyscallAfterExecStop); ok {
- if err := r.tc.MemoryManager.InvalidateUnsavable(ctx); err != nil {
- return err
- }
- }
- }
- return nil
-}
-
+// Preconditions: !VFS2Enabled.
func (ts *TaskSet) unregisterEpollWaiters(ctx context.Context) {
- // TODO(gvisor.dev/issue/1663): Add save support for VFS2.
- if VFS2Enabled {
- return
- }
-
ts.mu.RLock()
defer ts.mu.RUnlock()
@@ -644,8 +625,33 @@ func (ts *TaskSet) unregisterEpollWaiters(ctx context.Context) {
}
}
+// Preconditions: The kernel must be paused.
+func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error {
+ invalidated := make(map[*mm.MemoryManager]struct{})
+ k.tasks.mu.RLock()
+ defer k.tasks.mu.RUnlock()
+ for t := range k.tasks.Root.tids {
+ // We can skip locking Task.mu here since the kernel is paused.
+ if mm := t.tc.MemoryManager; mm != nil {
+ if _, ok := invalidated[mm]; !ok {
+ if err := mm.InvalidateUnsavable(ctx); err != nil {
+ return err
+ }
+ invalidated[mm] = struct{}{}
+ }
+ }
+ // I really wish we just had a sync.Map of all MMs...
+ if r, ok := t.runState.(*runSyscallAfterExecStop); ok {
+ if err := r.tc.MemoryManager.InvalidateUnsavable(ctx); err != nil {
+ return err
+ }
+ }
+ }
+ return nil
+}
+
// LoadFrom returns a new Kernel loaded from args.
-func (k *Kernel) LoadFrom(r wire.Reader, net inet.Stack, clocks sentrytime.Clocks) error {
+func (k *Kernel) LoadFrom(ctx context.Context, r wire.Reader, net inet.Stack, clocks sentrytime.Clocks, vfsOpts *vfs.CompleteRestoreOptions) error {
loadStart := time.Now()
initAppCores := k.applicationCores
@@ -656,7 +662,7 @@ func (k *Kernel) LoadFrom(r wire.Reader, net inet.Stack, clocks sentrytime.Clock
// don't need to explicitly install it in the Kernel.
cpuidStart := time.Now()
var features cpuid.FeatureSet
- if _, err := state.Load(k.SupervisorContext(), r, &features); err != nil {
+ if _, err := state.Load(ctx, r, &features); err != nil {
return err
}
log.Infof("CPUID load took [%s].", time.Since(cpuidStart))
@@ -671,7 +677,7 @@ func (k *Kernel) LoadFrom(r wire.Reader, net inet.Stack, clocks sentrytime.Clock
// Load the kernel state.
kernelStart := time.Now()
- stats, err := state.Load(k.SupervisorContext(), r, k)
+ stats, err := state.Load(ctx, r, k)
if err != nil {
return err
}
@@ -684,7 +690,7 @@ func (k *Kernel) LoadFrom(r wire.Reader, net inet.Stack, clocks sentrytime.Clock
// Load the memory file's state.
memoryStart := time.Now()
- if err := k.mf.LoadFrom(k.SupervisorContext(), r); err != nil {
+ if err := k.mf.LoadFrom(ctx, r); err != nil {
return err
}
log.Infof("Memory load took [%s].", time.Since(memoryStart))
@@ -696,11 +702,17 @@ func (k *Kernel) LoadFrom(r wire.Reader, net inet.Stack, clocks sentrytime.Clock
net.Resume()
}
- // Ensure that all pending asynchronous work is complete:
- // - namedpipe opening
- // - inode file opening
- if err := fs.AsyncErrorBarrier(); err != nil {
- return err
+ if VFS2Enabled {
+ if err := k.vfs.CompleteRestore(ctx, vfsOpts); err != nil {
+ return err
+ }
+ } else {
+ // Ensure that all pending asynchronous work is complete:
+ // - namedpipe opening
+ // - inode file opening
+ if err := fs.AsyncErrorBarrier(); err != nil {
+ return err
+ }
}
tcpip.AsyncLoading.Wait()
diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go
index ce0db5583..d6fb0fdb8 100644
--- a/pkg/sentry/kernel/pipe/node_test.go
+++ b/pkg/sentry/kernel/pipe/node_test.go
@@ -22,7 +22,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/contexttest"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/syserror"
- "gvisor.dev/gvisor/pkg/usermem"
)
type sleeper struct {
@@ -66,7 +65,8 @@ func testOpenOrDie(ctx context.Context, t *testing.T, n fs.InodeOperations, flag
d := fs.NewDirent(ctx, inode, "pipe")
file, err := n.GetFile(ctx, d, flags)
if err != nil {
- t.Fatalf("open with flags %+v failed: %v", flags, err)
+ t.Errorf("open with flags %+v failed: %v", flags, err)
+ return nil, err
}
if doneChan != nil {
doneChan <- struct{}{}
@@ -85,11 +85,11 @@ func testOpen(ctx context.Context, t *testing.T, n fs.InodeOperations, flags fs.
}
func newNamedPipe(t *testing.T) *Pipe {
- return NewPipe(true, DefaultPipeSize, usermem.PageSize)
+ return NewPipe(true, DefaultPipeSize)
}
func newAnonPipe(t *testing.T) *Pipe {
- return NewPipe(false, DefaultPipeSize, usermem.PageSize)
+ return NewPipe(false, DefaultPipeSize)
}
// assertRecvBlocks ensures that a recv attempt on c blocks for at least
diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go
index 67beb0ad6..b989e14c7 100644
--- a/pkg/sentry/kernel/pipe/pipe.go
+++ b/pkg/sentry/kernel/pipe/pipe.go
@@ -26,18 +26,27 @@ import (
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
)
const (
// MinimumPipeSize is a hard limit of the minimum size of a pipe.
- MinimumPipeSize = 64 << 10
+ // It corresponds to fs/pipe.c:pipe_min_size.
+ MinimumPipeSize = usermem.PageSize
+
+ // MaximumPipeSize is a hard limit on the maximum size of a pipe.
+ // It corresponds to fs/pipe.c:pipe_max_size.
+ MaximumPipeSize = 1048576
// DefaultPipeSize is the system-wide default size of a pipe in bytes.
- DefaultPipeSize = MinimumPipeSize
+ // It corresponds to pipe_fs_i.h:PIPE_DEF_BUFFERS.
+ DefaultPipeSize = 16 * usermem.PageSize
- // MaximumPipeSize is a hard limit on the maximum size of a pipe.
- MaximumPipeSize = 8 << 20
+ // atomicIOBytes is the maximum number of bytes that the pipe will
+ // guarantee atomic reads or writes atomically.
+ // It corresponds to limits.h:PIPE_BUF.
+ atomicIOBytes = 4096
)
// Pipe is an encapsulation of a platform-independent pipe.
@@ -53,12 +62,6 @@ type Pipe struct {
// This value is immutable.
isNamed bool
- // atomicIOBytes is the maximum number of bytes that the pipe will
- // guarantee atomic reads or writes atomically.
- //
- // This value is immutable.
- atomicIOBytes int64
-
// The number of active readers for this pipe.
//
// Access atomically.
@@ -94,47 +97,34 @@ type Pipe struct {
// NewPipe initializes and returns a pipe.
//
-// N.B. The size and atomicIOBytes will be bounded.
-func NewPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *Pipe {
+// N.B. The size will be bounded.
+func NewPipe(isNamed bool, sizeBytes int64) *Pipe {
if sizeBytes < MinimumPipeSize {
sizeBytes = MinimumPipeSize
}
if sizeBytes > MaximumPipeSize {
sizeBytes = MaximumPipeSize
}
- if atomicIOBytes <= 0 {
- atomicIOBytes = 1
- }
- if atomicIOBytes > sizeBytes {
- atomicIOBytes = sizeBytes
- }
var p Pipe
- initPipe(&p, isNamed, sizeBytes, atomicIOBytes)
+ initPipe(&p, isNamed, sizeBytes)
return &p
}
-func initPipe(pipe *Pipe, isNamed bool, sizeBytes, atomicIOBytes int64) {
+func initPipe(pipe *Pipe, isNamed bool, sizeBytes int64) {
if sizeBytes < MinimumPipeSize {
sizeBytes = MinimumPipeSize
}
if sizeBytes > MaximumPipeSize {
sizeBytes = MaximumPipeSize
}
- if atomicIOBytes <= 0 {
- atomicIOBytes = 1
- }
- if atomicIOBytes > sizeBytes {
- atomicIOBytes = sizeBytes
- }
pipe.isNamed = isNamed
pipe.max = sizeBytes
- pipe.atomicIOBytes = atomicIOBytes
}
// NewConnectedPipe initializes a pipe and returns a pair of objects
// representing the read and write ends of the pipe.
-func NewConnectedPipe(ctx context.Context, sizeBytes, atomicIOBytes int64) (*fs.File, *fs.File) {
- p := NewPipe(false /* isNamed */, sizeBytes, atomicIOBytes)
+func NewConnectedPipe(ctx context.Context, sizeBytes int64) (*fs.File, *fs.File) {
+ p := NewPipe(false /* isNamed */, sizeBytes)
// Build an fs.Dirent for the pipe which will be shared by both
// returned files.
@@ -264,7 +254,7 @@ func (p *Pipe) writeLocked(ctx context.Context, ops writeOps) (int64, error) {
wanted := ops.left()
avail := p.max - p.view.Size()
if wanted > avail {
- if wanted <= p.atomicIOBytes {
+ if wanted <= atomicIOBytes {
return 0, syserror.ErrWouldBlock
}
ops.limit(avail)
diff --git a/pkg/sentry/kernel/pipe/pipe_test.go b/pkg/sentry/kernel/pipe/pipe_test.go
index fe97e9800..3dd739080 100644
--- a/pkg/sentry/kernel/pipe/pipe_test.go
+++ b/pkg/sentry/kernel/pipe/pipe_test.go
@@ -26,7 +26,7 @@ import (
func TestPipeRW(t *testing.T) {
ctx := contexttest.Context(t)
- r, w := NewConnectedPipe(ctx, 65536, 4096)
+ r, w := NewConnectedPipe(ctx, 65536)
defer r.DecRef(ctx)
defer w.DecRef(ctx)
@@ -46,7 +46,7 @@ func TestPipeRW(t *testing.T) {
func TestPipeReadBlock(t *testing.T) {
ctx := contexttest.Context(t)
- r, w := NewConnectedPipe(ctx, 65536, 4096)
+ r, w := NewConnectedPipe(ctx, 65536)
defer r.DecRef(ctx)
defer w.DecRef(ctx)
@@ -61,7 +61,7 @@ func TestPipeWriteBlock(t *testing.T) {
const capacity = MinimumPipeSize
ctx := contexttest.Context(t)
- r, w := NewConnectedPipe(ctx, capacity, atomicIOBytes)
+ r, w := NewConnectedPipe(ctx, capacity)
defer r.DecRef(ctx)
defer w.DecRef(ctx)
@@ -76,7 +76,7 @@ func TestPipeWriteUntilEnd(t *testing.T) {
const atomicIOBytes = 2
ctx := contexttest.Context(t)
- r, w := NewConnectedPipe(ctx, atomicIOBytes, atomicIOBytes)
+ r, w := NewConnectedPipe(ctx, atomicIOBytes)
defer r.DecRef(ctx)
defer w.DecRef(ctx)
@@ -116,7 +116,8 @@ func TestPipeWriteUntilEnd(t *testing.T) {
}
}
if err != nil {
- t.Fatalf("Readv: got unexpected error %v", err)
+ t.Errorf("Readv: got unexpected error %v", err)
+ return
}
}
}()
diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go
index 1a152142b..7b23cbe86 100644
--- a/pkg/sentry/kernel/pipe/vfs.go
+++ b/pkg/sentry/kernel/pipe/vfs.go
@@ -33,6 +33,8 @@ import (
// VFSPipe represents the actual pipe, analagous to an inode. VFSPipes should
// not be copied.
+//
+// +stateify savable
type VFSPipe struct {
// mu protects the fields below.
mu sync.Mutex `state:"nosave"`
@@ -52,9 +54,9 @@ type VFSPipe struct {
}
// NewVFSPipe returns an initialized VFSPipe.
-func NewVFSPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *VFSPipe {
+func NewVFSPipe(isNamed bool, sizeBytes int64) *VFSPipe {
var vp VFSPipe
- initPipe(&vp.pipe, isNamed, sizeBytes, atomicIOBytes)
+ initPipe(&vp.pipe, isNamed, sizeBytes)
return &vp
}
@@ -164,6 +166,8 @@ func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, l
// VFSPipeFD implements vfs.FileDescriptionImpl for pipes. It also implements
// non-atomic usermem.IO methods, allowing it to be passed as usermem.IO to
// other FileDescriptions for splice(2) and tee(2).
+//
+// +stateify savable
type VFSPipeFD struct {
vfsfd vfs.FileDescription
vfs.FileDescriptionDefaultImpl
diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go
index 1145faf13..1abfe2201 100644
--- a/pkg/sentry/kernel/ptrace.go
+++ b/pkg/sentry/kernel/ptrace.go
@@ -1000,7 +1000,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error {
// at the address specified by the data parameter, and the return value
// is the error flag." - ptrace(2)
word := t.Arch().Native(0)
- if _, err := word.CopyIn(target.AsCopyContext(usermem.IOOpts{IgnorePermissions: true}), addr); err != nil {
+ if _, err := word.CopyIn(target.CopyContext(t, usermem.IOOpts{IgnorePermissions: true}), addr); err != nil {
return err
}
_, err := word.CopyOut(t, data)
@@ -1008,7 +1008,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error {
case linux.PTRACE_POKETEXT, linux.PTRACE_POKEDATA:
word := t.Arch().Native(uintptr(data))
- _, err := word.CopyOut(target.AsCopyContext(usermem.IOOpts{IgnorePermissions: true}), addr)
+ _, err := word.CopyOut(target.CopyContext(t, usermem.IOOpts{IgnorePermissions: true}), addr)
return err
case linux.PTRACE_GETREGSET:
diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go
index c00fa1138..b99c0bffa 100644
--- a/pkg/sentry/kernel/semaphore/semaphore.go
+++ b/pkg/sentry/kernel/semaphore/semaphore.go
@@ -103,6 +103,7 @@ type waiter struct {
waiterEntry
// value represents how much resource the waiter needs to wake up.
+ // The value is either 0 or negative.
value int16
ch chan struct{}
}
@@ -283,6 +284,33 @@ func (s *Set) Change(ctx context.Context, creds *auth.Credentials, owner fs.File
return nil
}
+// GetStat extracts semid_ds information from the set.
+func (s *Set) GetStat(creds *auth.Credentials) (*linux.SemidDS, error) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ // "The calling process must have read permission on the semaphore set."
+ if !s.checkPerms(creds, fs.PermMask{Read: true}) {
+ return nil, syserror.EACCES
+ }
+
+ ds := &linux.SemidDS{
+ SemPerm: linux.IPCPerm{
+ Key: uint32(s.key),
+ UID: uint32(creds.UserNamespace.MapFromKUID(s.owner.UID)),
+ GID: uint32(creds.UserNamespace.MapFromKGID(s.owner.GID)),
+ CUID: uint32(creds.UserNamespace.MapFromKUID(s.creator.UID)),
+ CGID: uint32(creds.UserNamespace.MapFromKGID(s.creator.GID)),
+ Mode: uint16(s.perms.LinuxMode()),
+ Seq: 0, // IPC sequence not supported.
+ },
+ SemOTime: s.opTime.TimeT(),
+ SemCTime: s.changeTime.TimeT(),
+ SemNSems: uint64(s.Size()),
+ }
+ return ds, nil
+}
+
// SetVal overrides a semaphore value, waking up waiters as needed.
func (s *Set) SetVal(ctx context.Context, num int32, val int16, creds *auth.Credentials, pid int32) error {
if val < 0 || val > valueMax {
@@ -320,7 +348,7 @@ func (s *Set) SetValAll(ctx context.Context, vals []uint16, creds *auth.Credenti
}
for _, val := range vals {
- if val < 0 || val > valueMax {
+ if val > valueMax {
return syserror.ERANGE
}
}
@@ -396,6 +424,42 @@ func (s *Set) GetPID(num int32, creds *auth.Credentials) (int32, error) {
return sem.pid, nil
}
+func (s *Set) countWaiters(num int32, creds *auth.Credentials, pred func(w *waiter) bool) (uint16, error) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ // The calling process must have read permission on the semaphore set.
+ if !s.checkPerms(creds, fs.PermMask{Read: true}) {
+ return 0, syserror.EACCES
+ }
+
+ sem := s.findSem(num)
+ if sem == nil {
+ return 0, syserror.ERANGE
+ }
+ var cnt uint16
+ for w := sem.waiters.Front(); w != nil; w = w.Next() {
+ if pred(w) {
+ cnt++
+ }
+ }
+ return cnt, nil
+}
+
+// CountZeroWaiters returns number of waiters waiting for the sem's value to increase.
+func (s *Set) CountZeroWaiters(num int32, creds *auth.Credentials) (uint16, error) {
+ return s.countWaiters(num, creds, func(w *waiter) bool {
+ return w.value == 0
+ })
+}
+
+// CountNegativeWaiters returns number of waiters waiting for the sem to go to zero.
+func (s *Set) CountNegativeWaiters(num int32, creds *auth.Credentials) (uint16, error) {
+ return s.countWaiters(num, creds, func(w *waiter) bool {
+ return w.value < 0
+ })
+}
+
// ExecuteOps attempts to execute a list of operations to the set. It only
// succeeds when all operations can be applied. No changes are made if it fails.
//
@@ -548,11 +612,18 @@ func (s *Set) destroy() {
}
}
+func abs(val int16) int16 {
+ if val < 0 {
+ return -val
+ }
+ return val
+}
+
// wakeWaiters goes over all waiters and checks which of them can be notified.
func (s *sem) wakeWaiters() {
// Note that this will release all waiters waiting for 0 too.
for w := s.waiters.Front(); w != nil; {
- if s.value < w.value {
+ if s.value < abs(w.value) {
// Still blocked, skip it.
w = w.Next()
continue
diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go
index df5c8421b..5bddb0a36 100644
--- a/pkg/sentry/kernel/sessions.go
+++ b/pkg/sentry/kernel/sessions.go
@@ -477,20 +477,20 @@ func (tg *ThreadGroup) Session() *Session {
//
// If this group isn't visible in this namespace, zero will be returned. It is
// the callers responsibility to check that before using this function.
-func (pidns *PIDNamespace) IDOfSession(s *Session) SessionID {
- pidns.owner.mu.RLock()
- defer pidns.owner.mu.RUnlock()
- return pidns.sids[s]
+func (ns *PIDNamespace) IDOfSession(s *Session) SessionID {
+ ns.owner.mu.RLock()
+ defer ns.owner.mu.RUnlock()
+ return ns.sids[s]
}
// SessionWithID returns the Session with the given ID in the PID namespace ns,
// or nil if that given ID is not defined in this namespace.
//
// A reference is not taken on the session.
-func (pidns *PIDNamespace) SessionWithID(id SessionID) *Session {
- pidns.owner.mu.RLock()
- defer pidns.owner.mu.RUnlock()
- return pidns.sessions[id]
+func (ns *PIDNamespace) SessionWithID(id SessionID) *Session {
+ ns.owner.mu.RLock()
+ defer ns.owner.mu.RUnlock()
+ return ns.sessions[id]
}
// ProcessGroup returns the ThreadGroup's ProcessGroup.
@@ -505,18 +505,18 @@ func (tg *ThreadGroup) ProcessGroup() *ProcessGroup {
// IDOfProcessGroup returns the process group assigned to pg in PID namespace ns.
//
// The same constraints apply as IDOfSession.
-func (pidns *PIDNamespace) IDOfProcessGroup(pg *ProcessGroup) ProcessGroupID {
- pidns.owner.mu.RLock()
- defer pidns.owner.mu.RUnlock()
- return pidns.pgids[pg]
+func (ns *PIDNamespace) IDOfProcessGroup(pg *ProcessGroup) ProcessGroupID {
+ ns.owner.mu.RLock()
+ defer ns.owner.mu.RUnlock()
+ return ns.pgids[pg]
}
// ProcessGroupWithID returns the ProcessGroup with the given ID in the PID
// namespace ns, or nil if that given ID is not defined in this namespace.
//
// A reference is not taken on the process group.
-func (pidns *PIDNamespace) ProcessGroupWithID(id ProcessGroupID) *ProcessGroup {
- pidns.owner.mu.RLock()
- defer pidns.owner.mu.RUnlock()
- return pidns.processGroups[id]
+func (ns *PIDNamespace) ProcessGroupWithID(id ProcessGroupID) *ProcessGroup {
+ ns.owner.mu.RLock()
+ defer ns.owner.mu.RUnlock()
+ return ns.processGroups[id]
}
diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD
index f8a382fd8..80a592c8f 100644
--- a/pkg/sentry/kernel/shm/BUILD
+++ b/pkg/sentry/kernel/shm/BUILD
@@ -8,7 +8,7 @@ go_template_instance(
out = "shm_refs.go",
package = "shm",
prefix = "Shm",
- template = "//pkg/refs_vfs2:refs_template",
+ template = "//pkg/refsvfs2:refs_template",
types = {
"T": "Shm",
},
@@ -27,7 +27,7 @@ go_library(
"//pkg/context",
"//pkg/log",
"//pkg/refs",
- "//pkg/refs_vfs2",
+ "//pkg/refsvfs2",
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/kernel/auth",
diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go
index 682080c14..527344162 100644
--- a/pkg/sentry/kernel/task_clone.go
+++ b/pkg/sentry/kernel/task_clone.go
@@ -355,7 +355,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
}
if opts.ChildSetTID {
ctid := nt.ThreadID()
- ctid.CopyOut(nt.AsCopyContext(usermem.IOOpts{AddressSpaceActive: false}), opts.ChildTID)
+ ctid.CopyOut(nt.CopyContext(t, usermem.IOOpts{AddressSpaceActive: false}), opts.ChildTID)
}
ntid := t.tg.pidns.IDOfTask(nt)
if opts.ParentSetTID {
diff --git a/pkg/sentry/kernel/task_usermem.go b/pkg/sentry/kernel/task_usermem.go
index ce134bf54..94dabbcd8 100644
--- a/pkg/sentry/kernel/task_usermem.go
+++ b/pkg/sentry/kernel/task_usermem.go
@@ -18,7 +18,8 @@ import (
"math"
"gvisor.dev/gvisor/pkg/abi/linux"
- "gvisor.dev/gvisor/pkg/marshal"
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/sentry/mm"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
)
@@ -281,29 +282,89 @@ func (t *Task) IovecsIOSequence(addr usermem.Addr, iovcnt int, opts usermem.IOOp
}, nil
}
-// copyContext implements marshal.CopyContext. It wraps a task to allow copying
-// memory to and from the task memory with custom usermem.IOOpts.
-type copyContext struct {
- *Task
+type taskCopyContext struct {
+ ctx context.Context
+ t *Task
opts usermem.IOOpts
}
-// AsCopyContext wraps the task and returns it as CopyContext.
-func (t *Task) AsCopyContext(opts usermem.IOOpts) marshal.CopyContext {
- return &copyContext{t, opts}
+// CopyContext returns a marshal.CopyContext that copies to/from t's address
+// space using opts.
+func (t *Task) CopyContext(ctx context.Context, opts usermem.IOOpts) *taskCopyContext {
+ return &taskCopyContext{
+ ctx: ctx,
+ t: t,
+ opts: opts,
+ }
+}
+
+// CopyScratchBuffer implements marshal.CopyContext.CopyScratchBuffer.
+func (cc *taskCopyContext) CopyScratchBuffer(size int) []byte {
+ if ctxTask, ok := cc.ctx.(*Task); ok {
+ return ctxTask.CopyScratchBuffer(size)
+ }
+ return make([]byte, size)
+}
+
+func (cc *taskCopyContext) getMemoryManager() (*mm.MemoryManager, error) {
+ cc.t.mu.Lock()
+ tmm := cc.t.MemoryManager()
+ cc.t.mu.Unlock()
+ if !tmm.IncUsers() {
+ return nil, syserror.EFAULT
+ }
+ return tmm, nil
+}
+
+// CopyInBytes implements marshal.CopyContext.CopyInBytes.
+func (cc *taskCopyContext) CopyInBytes(addr usermem.Addr, dst []byte) (int, error) {
+ tmm, err := cc.getMemoryManager()
+ if err != nil {
+ return 0, err
+ }
+ defer tmm.DecUsers(cc.ctx)
+ return tmm.CopyIn(cc.ctx, addr, dst, cc.opts)
+}
+
+// CopyOutBytes implements marshal.CopyContext.CopyOutBytes.
+func (cc *taskCopyContext) CopyOutBytes(addr usermem.Addr, src []byte) (int, error) {
+ tmm, err := cc.getMemoryManager()
+ if err != nil {
+ return 0, err
+ }
+ defer tmm.DecUsers(cc.ctx)
+ return tmm.CopyOut(cc.ctx, addr, src, cc.opts)
+}
+
+type ownTaskCopyContext struct {
+ t *Task
+ opts usermem.IOOpts
+}
+
+// OwnCopyContext returns a marshal.CopyContext that copies to/from t's address
+// space using opts. The returned CopyContext may only be used by t's task
+// goroutine.
+//
+// Since t already implements marshal.CopyContext, this is only needed to
+// override the usermem.IOOpts used for the copy.
+func (t *Task) OwnCopyContext(opts usermem.IOOpts) *ownTaskCopyContext {
+ return &ownTaskCopyContext{
+ t: t,
+ opts: opts,
+ }
}
-// CopyInString copies a string in from the task's memory.
-func (t *copyContext) CopyInString(addr usermem.Addr, maxLen int) (string, error) {
- return usermem.CopyStringIn(t, t.MemoryManager(), addr, maxLen, t.opts)
+// CopyScratchBuffer implements marshal.CopyContext.CopyScratchBuffer.
+func (cc *ownTaskCopyContext) CopyScratchBuffer(size int) []byte {
+ return cc.t.CopyScratchBuffer(size)
}
-// CopyInBytes copies task memory into dst from an IO context.
-func (t *copyContext) CopyInBytes(addr usermem.Addr, dst []byte) (int, error) {
- return t.MemoryManager().CopyIn(t, addr, dst, t.opts)
+// CopyInBytes implements marshal.CopyContext.CopyInBytes.
+func (cc *ownTaskCopyContext) CopyInBytes(addr usermem.Addr, dst []byte) (int, error) {
+ return cc.t.MemoryManager().CopyIn(cc.t, addr, dst, cc.opts)
}
-// CopyOutBytes copies src into task memoryfrom an IO context.
-func (t *copyContext) CopyOutBytes(addr usermem.Addr, src []byte) (int, error) {
- return t.MemoryManager().CopyOut(t, addr, src, t.opts)
+// CopyOutBytes implements marshal.CopyContext.CopyOutBytes.
+func (cc *ownTaskCopyContext) CopyOutBytes(addr usermem.Addr, src []byte) (int, error) {
+ return cc.t.MemoryManager().CopyOut(cc.t, addr, src, cc.opts)
}
diff --git a/pkg/sentry/kernel/vdso.go b/pkg/sentry/kernel/vdso.go
index 9bc452e67..9e5c2d26f 100644
--- a/pkg/sentry/kernel/vdso.go
+++ b/pkg/sentry/kernel/vdso.go
@@ -115,7 +115,7 @@ func (v *VDSOParamPage) incrementSeq(paramPage safemem.Block) error {
}
if old != v.seq {
- return fmt.Errorf("unexpected VDSOParamPage seq value: got %d expected %d. Application may hang or get incorrect time from the VDSO.", old, v.seq)
+ return fmt.Errorf("unexpected VDSOParamPage seq value: got %d expected %d; application may hang or get incorrect time from the VDSO", old, v.seq)
}
v.seq = next
generated by cgit v1.2.3 (git 2.39.1) at 2025-03-08 12:42:40 +0000