summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry')
-rw-r--r--pkg/sentry/arch/stack.go3
-rw-r--r--pkg/sentry/contexttest/contexttest.go4
-rw-r--r--pkg/sentry/fs/dirent.go6
-rw-r--r--pkg/sentry/fs/fdpipe/pipe_test.go4
-rw-r--r--pkg/sentry/fs/gofer/file_state.go1
-rw-r--r--pkg/sentry/fs/gofer/handles.go1
-rw-r--r--pkg/sentry/fs/gofer/inode.go5
-rw-r--r--pkg/sentry/fs/gofer/inode_state.go1
-rw-r--r--pkg/sentry/fs/gofer/session_state.go1
-rw-r--r--pkg/sentry/fs/gofer/util.go16
-rw-r--r--pkg/sentry/fs/host/inode.go3
-rw-r--r--pkg/sentry/fs/host/socket_test.go6
-rw-r--r--pkg/sentry/fs/inode.go3
-rw-r--r--pkg/sentry/fs/proc/mounts.go3
-rw-r--r--pkg/sentry/fs/proc/sys_net.go4
-rw-r--r--pkg/sentry/fs/proc/task.go28
-rw-r--r--pkg/sentry/fs/tmpfs/fs.go3
-rw-r--r--pkg/sentry/fsbridge/vfs.go28
-rw-r--r--pkg/sentry/fsimpl/ext/filesystem.go4
-rw-r--r--pkg/sentry/fsimpl/gofer/BUILD12
-rw-r--r--pkg/sentry/fsimpl/gofer/directory.go7
-rw-r--r--pkg/sentry/fsimpl/gofer/filesystem.go30
-rw-r--r--pkg/sentry/fsimpl/gofer/gofer.go118
-rw-r--r--pkg/sentry/fsimpl/gofer/gofer_test.go64
-rw-r--r--pkg/sentry/fsimpl/gofer/p9file.go14
-rw-r--r--pkg/sentry/fsimpl/gofer/regular_file.go7
-rw-r--r--pkg/sentry/fsimpl/gofer/special_file.go4
-rw-r--r--pkg/sentry/fsimpl/gofer/symlink.go2
-rw-r--r--pkg/sentry/fsimpl/gofer/time.go39
-rw-r--r--pkg/sentry/fsimpl/host/host.go42
-rw-r--r--pkg/sentry/fsimpl/kernfs/filesystem.go51
-rw-r--r--pkg/sentry/fsimpl/kernfs/inode_impl_util.go5
-rw-r--r--pkg/sentry/fsimpl/kernfs/kernfs.go16
-rw-r--r--pkg/sentry/fsimpl/kernfs/symlink.go5
-rw-r--r--pkg/sentry/fsimpl/proc/BUILD1
-rw-r--r--pkg/sentry/fsimpl/proc/task.go16
-rw-r--r--pkg/sentry/fsimpl/proc/task_fds.go125
-rw-r--r--pkg/sentry/fsimpl/proc/task_files.go73
-rw-r--r--pkg/sentry/fsimpl/proc/task_net.go92
-rw-r--r--pkg/sentry/fsimpl/proc/tasks_files.go10
-rw-r--r--pkg/sentry/fsimpl/sockfs/BUILD1
-rw-r--r--pkg/sentry/fsimpl/sockfs/sockfs.go29
-rw-r--r--pkg/sentry/fsimpl/tmpfs/BUILD2
-rw-r--r--pkg/sentry/fsimpl/tmpfs/directory.go2
-rw-r--r--pkg/sentry/fsimpl/tmpfs/filesystem.go75
-rw-r--r--pkg/sentry/fsimpl/tmpfs/regular_file.go4
-rw-r--r--pkg/sentry/fsimpl/tmpfs/socket_file.go34
-rw-r--r--pkg/sentry/fsimpl/tmpfs/stat_test.go12
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs.go140
-rw-r--r--pkg/sentry/kernel/BUILD1
-rw-r--r--pkg/sentry/kernel/fd_table.go55
-rw-r--r--pkg/sentry/kernel/kernel.go80
-rw-r--r--pkg/sentry/kernel/pipe/pipe.go13
-rw-r--r--pkg/sentry/kernel/ptrace.go1
-rw-r--r--pkg/sentry/kernel/shm/shm.go2
-rw-r--r--pkg/sentry/kernel/syscalls.go36
-rw-r--r--pkg/sentry/kernel/syscalls_state.go36
-rw-r--r--pkg/sentry/kernel/task.go9
-rw-r--r--pkg/sentry/kernel/task_context.go3
-rw-r--r--pkg/sentry/kernel/task_identity.go2
-rw-r--r--pkg/sentry/kernel/task_signals.go2
-rw-r--r--pkg/sentry/kernel/time/time.go10
-rw-r--r--pkg/sentry/mm/address_space.go6
-rw-r--r--pkg/sentry/mm/aio_context.go101
-rw-r--r--pkg/sentry/mm/aio_context_state.go2
-rw-r--r--pkg/sentry/platform/ring0/BUILD2
-rw-r--r--pkg/sentry/socket/netstack/netstack.go169
-rw-r--r--pkg/sentry/socket/netstack/provider.go8
-rw-r--r--pkg/sentry/socket/socket.go6
-rw-r--r--pkg/sentry/socket/unix/BUILD4
-rw-r--r--pkg/sentry/socket/unix/transport/BUILD1
-rw-r--r--pkg/sentry/socket/unix/transport/unix.go50
-rw-r--r--pkg/sentry/socket/unix/unix.go89
-rw-r--r--pkg/sentry/socket/unix/unix_vfs2.go348
-rw-r--r--pkg/sentry/strace/strace.go3
-rw-r--r--pkg/sentry/syscalls/linux/sys_aio.go34
-rw-r--r--pkg/sentry/syscalls/linux/sys_prctl.go4
-rw-r--r--pkg/sentry/syscalls/linux/sys_socket.go9
-rw-r--r--pkg/sentry/syscalls/linux/sys_splice.go4
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/BUILD7
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/getdents.go20
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go46
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/socket.go1138
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go123
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/xattr.go13
-rw-r--r--pkg/sentry/vfs/BUILD2
-rw-r--r--pkg/sentry/vfs/anonfs.go4
-rw-r--r--pkg/sentry/vfs/epoll.go2
-rw-r--r--pkg/sentry/vfs/file_description.go60
-rw-r--r--pkg/sentry/vfs/file_description_impl_util.go4
-rw-r--r--pkg/sentry/vfs/filesystem.go26
-rw-r--r--pkg/sentry/vfs/memxattr/BUILD15
-rw-r--r--pkg/sentry/vfs/memxattr/xattr.go102
-rw-r--r--pkg/sentry/vfs/mount.go15
-rw-r--r--pkg/sentry/vfs/mount_test.go2
-rw-r--r--pkg/sentry/vfs/options.go19
-rw-r--r--pkg/sentry/vfs/timerfd.go142
-rw-r--r--pkg/sentry/vfs/vfs.go8
98 files changed, 3359 insertions, 635 deletions
diff --git a/pkg/sentry/arch/stack.go b/pkg/sentry/arch/stack.go
index 09bceabc9..1108fa0bd 100644
--- a/pkg/sentry/arch/stack.go
+++ b/pkg/sentry/arch/stack.go
@@ -97,7 +97,6 @@ func (s *Stack) Push(vals ...interface{}) (usermem.Addr, error) {
if c < 0 {
return 0, fmt.Errorf("bad binary.Size for %T", v)
}
- // TODO(b/38173783): Use a real context.Context.
n, err := usermem.CopyObjectOut(context.Background(), s.IO, s.Bottom-usermem.Addr(c), norm, usermem.IOOpts{})
if err != nil || c != n {
return 0, err
@@ -121,11 +120,9 @@ func (s *Stack) Pop(vals ...interface{}) (usermem.Addr, error) {
var err error
if isVaddr {
value := s.Arch.Native(uintptr(0))
- // TODO(b/38173783): Use a real context.Context.
n, err = usermem.CopyObjectIn(context.Background(), s.IO, s.Bottom, value, usermem.IOOpts{})
*vaddr = usermem.Addr(s.Arch.Value(value))
} else {
- // TODO(b/38173783): Use a real context.Context.
n, err = usermem.CopyObjectIn(context.Background(), s.IO, s.Bottom, v, usermem.IOOpts{})
}
if err != nil {
diff --git a/pkg/sentry/contexttest/contexttest.go b/pkg/sentry/contexttest/contexttest.go
index 031fc64ec..8e5658c7a 100644
--- a/pkg/sentry/contexttest/contexttest.go
+++ b/pkg/sentry/contexttest/contexttest.go
@@ -97,7 +97,7 @@ type hostClock struct {
}
// Now implements ktime.Clock.Now.
-func (hostClock) Now() ktime.Time {
+func (*hostClock) Now() ktime.Time {
return ktime.FromNanoseconds(time.Now().UnixNano())
}
@@ -127,7 +127,7 @@ func (t *TestContext) Value(key interface{}) interface{} {
case uniqueid.CtxInotifyCookie:
return atomic.AddUint32(&lastInotifyCookie, 1)
case ktime.CtxRealtimeClock:
- return hostClock{}
+ return &hostClock{}
default:
if val, ok := t.otherValues[key]; ok {
return val
diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go
index 0266a5287..65be12175 100644
--- a/pkg/sentry/fs/dirent.go
+++ b/pkg/sentry/fs/dirent.go
@@ -312,9 +312,9 @@ func (d *Dirent) SyncAll(ctx context.Context) {
// There is nothing to sync for a read-only filesystem.
if !d.Inode.MountSource.Flags.ReadOnly {
- // FIXME(b/34856369): This should be a mount traversal, not a
- // Dirent traversal, because some Inodes that need to be synced
- // may no longer be reachable by name (after sys_unlink).
+ // NOTE(b/34856369): This should be a mount traversal, not a Dirent
+ // traversal, because some Inodes that need to be synced may no longer
+ // be reachable by name (after sys_unlink).
//
// Write out metadata, dirty page cached pages, and sync disk/remote
// caches.
diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go
index 5aff0cc95..a0082ecca 100644
--- a/pkg/sentry/fs/fdpipe/pipe_test.go
+++ b/pkg/sentry/fs/fdpipe/pipe_test.go
@@ -119,7 +119,7 @@ func TestNewPipe(t *testing.T) {
continue
}
if flags := p.flags; test.flags != flags {
- t.Errorf("%s: got file flags %s, want %s", test.desc, flags, test.flags)
+ t.Errorf("%s: got file flags %v, want %v", test.desc, flags, test.flags)
continue
}
if len(test.readAheadBuffer) != len(p.readAheadBuffer) {
@@ -136,7 +136,7 @@ func TestNewPipe(t *testing.T) {
continue
}
if !fdnotifier.HasFD(int32(f.FD())) {
- t.Errorf("%s: pipe fd %d is not registered for events", test.desc, f.FD)
+ t.Errorf("%s: pipe fd %d is not registered for events", test.desc, f.FD())
}
}
}
diff --git a/pkg/sentry/fs/gofer/file_state.go b/pkg/sentry/fs/gofer/file_state.go
index ff96b28ba..edd6576aa 100644
--- a/pkg/sentry/fs/gofer/file_state.go
+++ b/pkg/sentry/fs/gofer/file_state.go
@@ -34,7 +34,6 @@ func (f *fileOperations) afterLoad() {
flags := f.flags
flags.Truncate = false
- // TODO(b/38173783): Context is not plumbed to save/restore.
f.handles, err = f.inodeOperations.fileState.getHandles(context.Background(), flags, f.inodeOperations.cachingInodeOps)
if err != nil {
return fmt.Errorf("failed to re-open handle: %v", err)
diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go
index 9f7c3e89f..fc14249be 100644
--- a/pkg/sentry/fs/gofer/handles.go
+++ b/pkg/sentry/fs/gofer/handles.go
@@ -57,7 +57,6 @@ func (h *handles) DecRef() {
}
}
}
- // FIXME(b/38173783): Context is not plumbed here.
if err := h.File.close(context.Background()); err != nil {
log.Warningf("error closing p9 file: %v", err)
}
diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go
index 1c934981b..a016c896e 100644
--- a/pkg/sentry/fs/gofer/inode.go
+++ b/pkg/sentry/fs/gofer/inode.go
@@ -273,7 +273,7 @@ func (i *inodeFileState) recreateReadHandles(ctx context.Context, writer *handle
// operations on the old will see the new data. Then, make the new handle take
// ownereship of the old FD and mark the old readHandle to not close the FD
// when done.
- if err := syscall.Dup3(h.Host.FD(), i.readHandles.Host.FD(), 0); err != nil {
+ if err := syscall.Dup3(h.Host.FD(), i.readHandles.Host.FD(), syscall.O_CLOEXEC); err != nil {
return err
}
@@ -710,13 +710,10 @@ func init() {
}
// AddLink implements InodeOperations.AddLink, but is currently a noop.
-// FIXME(b/63117438): Remove this from InodeOperations altogether.
func (*inodeOperations) AddLink() {}
// DropLink implements InodeOperations.DropLink, but is currently a noop.
-// FIXME(b/63117438): Remove this from InodeOperations altogether.
func (*inodeOperations) DropLink() {}
// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange.
-// FIXME(b/63117438): Remove this from InodeOperations altogether.
func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {}
diff --git a/pkg/sentry/fs/gofer/inode_state.go b/pkg/sentry/fs/gofer/inode_state.go
index 238f7804c..a3402e343 100644
--- a/pkg/sentry/fs/gofer/inode_state.go
+++ b/pkg/sentry/fs/gofer/inode_state.go
@@ -123,7 +123,6 @@ func (i *inodeFileState) afterLoad() {
// beforeSave.
return fmt.Errorf("failed to find path for inode number %d. Device %s contains %s", i.sattr.InodeID, i.s.connID, fs.InodeMappings(i.s.inodeMappings))
}
- // TODO(b/38173783): Context is not plumbed to save/restore.
ctx := &dummyClockContext{context.Background()}
_, i.file, err = i.s.attach.walk(ctx, splitAbsolutePath(name))
diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go
index 111da59f9..2d398b753 100644
--- a/pkg/sentry/fs/gofer/session_state.go
+++ b/pkg/sentry/fs/gofer/session_state.go
@@ -104,7 +104,6 @@ func (s *session) afterLoad() {
// If private unix sockets are enabled, create and fill the session's endpoint
// maps.
if opts.privateunixsocket {
- // TODO(b/38173783): Context is not plumbed to save/restore.
ctx := &dummyClockContext{context.Background()}
if err = s.restoreEndpointMaps(ctx); err != nil {
diff --git a/pkg/sentry/fs/gofer/util.go b/pkg/sentry/fs/gofer/util.go
index 2d8d3a2ea..47a6c69bf 100644
--- a/pkg/sentry/fs/gofer/util.go
+++ b/pkg/sentry/fs/gofer/util.go
@@ -20,17 +20,29 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/p9"
"gvisor.dev/gvisor/pkg/sentry/fs"
+ ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
)
func utimes(ctx context.Context, file contextFile, ts fs.TimeSpec) error {
if ts.ATimeOmit && ts.MTimeOmit {
return nil
}
+
+ // Replace requests to use the "system time" with the current time to
+ // ensure that timestamps remain consistent with the remote
+ // filesystem.
+ now := ktime.NowFromContext(ctx)
+ if ts.ATimeSetSystemTime {
+ ts.ATime = now
+ }
+ if ts.MTimeSetSystemTime {
+ ts.MTime = now
+ }
mask := p9.SetAttrMask{
ATime: !ts.ATimeOmit,
- ATimeNotSystemTime: !ts.ATimeSetSystemTime,
+ ATimeNotSystemTime: true,
MTime: !ts.MTimeOmit,
- MTimeNotSystemTime: !ts.MTimeSetSystemTime,
+ MTimeNotSystemTime: true,
}
as, ans := ts.ATime.Unix()
ms, mns := ts.MTime.Unix()
diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go
index 1da3c0a17..62f1246aa 100644
--- a/pkg/sentry/fs/host/inode.go
+++ b/pkg/sentry/fs/host/inode.go
@@ -397,15 +397,12 @@ func (i *inodeOperations) StatFS(context.Context) (fs.Info, error) {
}
// AddLink implements fs.InodeOperations.AddLink.
-// FIXME(b/63117438): Remove this from InodeOperations altogether.
func (i *inodeOperations) AddLink() {}
// DropLink implements fs.InodeOperations.DropLink.
-// FIXME(b/63117438): Remove this from InodeOperations altogether.
func (i *inodeOperations) DropLink() {}
// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange.
-// FIXME(b/63117438): Remove this from InodeOperations altogether.
func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {}
// readdirAll returns all of the directory entries in i.
diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go
index eb4afe520..affdbcacb 100644
--- a/pkg/sentry/fs/host/socket_test.go
+++ b/pkg/sentry/fs/host/socket_test.go
@@ -199,14 +199,14 @@ func TestListen(t *testing.T) {
}
func TestPasscred(t *testing.T) {
- e := ConnectedEndpoint{}
+ e := &ConnectedEndpoint{}
if got, want := e.Passcred(), false; got != want {
t.Errorf("Got %#v.Passcred() = %t, want = %t", e, got, want)
}
}
func TestGetLocalAddress(t *testing.T) {
- e := ConnectedEndpoint{path: "foo"}
+ e := &ConnectedEndpoint{path: "foo"}
want := tcpip.FullAddress{Addr: tcpip.Address("foo")}
if got, err := e.GetLocalAddress(); err != nil || got != want {
t.Errorf("Got %#v.GetLocalAddress() = %#v, %v, want = %#v, %v", e, got, err, want, nil)
@@ -214,7 +214,7 @@ func TestGetLocalAddress(t *testing.T) {
}
func TestQueuedSize(t *testing.T) {
- e := ConnectedEndpoint{}
+ e := &ConnectedEndpoint{}
tests := []struct {
name string
f func() int64
diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go
index 55fb71c16..a34fbc946 100644
--- a/pkg/sentry/fs/inode.go
+++ b/pkg/sentry/fs/inode.go
@@ -102,7 +102,6 @@ func (i *Inode) DecRef() {
// destroy releases the Inode and releases the msrc reference taken.
func (i *Inode) destroy() {
- // FIXME(b/38173783): Context is not plumbed here.
ctx := context.Background()
if err := i.WriteOut(ctx); err != nil {
// FIXME(b/65209558): Mark as warning again once noatime is
@@ -397,8 +396,6 @@ func (i *Inode) Getlink(ctx context.Context) (*Dirent, error) {
// AddLink calls i.InodeOperations.AddLink.
func (i *Inode) AddLink() {
if i.overlay != nil {
- // FIXME(b/63117438): Remove this from InodeOperations altogether.
- //
// This interface is only used by ramfs to update metadata of
// children. These filesystems should _never_ have overlay
// Inodes cached as children. So explicitly disallow this
diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go
index 94deb553b..1fc9c703c 100644
--- a/pkg/sentry/fs/proc/mounts.go
+++ b/pkg/sentry/fs/proc/mounts.go
@@ -170,7 +170,8 @@ func superBlockOpts(mountPath string, msrc *fs.MountSource) string {
// NOTE(b/147673608): If the mount is a cgroup, we also need to include
// the cgroup name in the options. For now we just read that from the
// path.
- // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we
+ //
+ // TODO(gvisor.dev/issue/190): Once gVisor has full cgroup support, we
// should get this value from the cgroup itself, and not rely on the
// path.
if msrc.FilesystemType == "cgroup" {
diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go
index d4c4b533d..702fdd392 100644
--- a/pkg/sentry/fs/proc/sys_net.go
+++ b/pkg/sentry/fs/proc/sys_net.go
@@ -80,7 +80,7 @@ func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, dir
}
// Truncate implements fs.InodeOperations.Truncate.
-func (tcpMemInode) Truncate(context.Context, *fs.Inode, int64) error {
+func (*tcpMemInode) Truncate(context.Context, *fs.Inode, int64) error {
return nil
}
@@ -196,7 +196,7 @@ func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *f
}
// Truncate implements fs.InodeOperations.Truncate.
-func (tcpSack) Truncate(context.Context, *fs.Inode, int64) error {
+func (*tcpSack) Truncate(context.Context, *fs.Inode, int64) error {
return nil
}
diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go
index d6c5dd2c1..4d42eac83 100644
--- a/pkg/sentry/fs/proc/task.go
+++ b/pkg/sentry/fs/proc/task.go
@@ -57,6 +57,16 @@ func getTaskMM(t *kernel.Task) (*mm.MemoryManager, error) {
return m, nil
}
+func checkTaskState(t *kernel.Task) error {
+ switch t.ExitState() {
+ case kernel.TaskExitZombie:
+ return syserror.EACCES
+ case kernel.TaskExitDead:
+ return syserror.ESRCH
+ }
+ return nil
+}
+
// taskDir represents a task-level directory.
//
// +stateify savable
@@ -254,11 +264,12 @@ func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
}
func (e *exe) executable() (file fsbridge.File, err error) {
+ if err := checkTaskState(e.t); err != nil {
+ return nil, err
+ }
e.t.WithMuLocked(func(t *kernel.Task) {
mm := t.MemoryManager()
if mm == nil {
- // TODO(b/34851096): Check shouldn't allow Readlink once the
- // Task is zombied.
err = syserror.EACCES
return
}
@@ -268,7 +279,7 @@ func (e *exe) executable() (file fsbridge.File, err error) {
// (with locks held).
file = mm.Executable()
if file == nil {
- err = syserror.ENOENT
+ err = syserror.ESRCH
}
})
return
@@ -313,11 +324,22 @@ func newNamespaceSymlink(t *kernel.Task, msrc *fs.MountSource, name string) *fs.
return newProcInode(t, n, msrc, fs.Symlink, t)
}
+// Readlink reads the symlink value.
+func (n *namespaceSymlink) Readlink(ctx context.Context, inode *fs.Inode) (string, error) {
+ if err := checkTaskState(n.t); err != nil {
+ return "", err
+ }
+ return n.Symlink.Readlink(ctx, inode)
+}
+
// Getlink implements fs.InodeOperations.Getlink.
func (n *namespaceSymlink) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) {
if !kernel.ContextCanTrace(ctx, n.t, false) {
return nil, syserror.EACCES
}
+ if err := checkTaskState(n.t); err != nil {
+ return nil, err
+ }
// Create a new regular file to fake the namespace file.
iops := fsutil.NewNoReadWriteFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0777), linux.PROC_SUPER_MAGIC)
diff --git a/pkg/sentry/fs/tmpfs/fs.go b/pkg/sentry/fs/tmpfs/fs.go
index d5be56c3f..bc117ca6a 100644
--- a/pkg/sentry/fs/tmpfs/fs.go
+++ b/pkg/sentry/fs/tmpfs/fs.go
@@ -44,9 +44,6 @@ const (
// lookup.
cacheRevalidate = "revalidate"
- // TODO(edahlgren/mpratt): support a tmpfs size limit.
- // size = "size"
-
// Permissions that exceed modeMask will be rejected.
modeMask = 01777
diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go
index 79b808359..89168220a 100644
--- a/pkg/sentry/fsbridge/vfs.go
+++ b/pkg/sentry/fsbridge/vfs.go
@@ -26,22 +26,22 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
-// fsFile implements File interface over vfs.FileDescription.
+// VFSFile implements File interface over vfs.FileDescription.
//
// +stateify savable
-type vfsFile struct {
+type VFSFile struct {
file *vfs.FileDescription
}
-var _ File = (*vfsFile)(nil)
+var _ File = (*VFSFile)(nil)
// NewVFSFile creates a new File over fs.File.
func NewVFSFile(file *vfs.FileDescription) File {
- return &vfsFile{file: file}
+ return &VFSFile{file: file}
}
// PathnameWithDeleted implements File.
-func (f *vfsFile) PathnameWithDeleted(ctx context.Context) string {
+func (f *VFSFile) PathnameWithDeleted(ctx context.Context) string {
root := vfs.RootFromContext(ctx)
defer root.DecRef()
@@ -51,7 +51,7 @@ func (f *vfsFile) PathnameWithDeleted(ctx context.Context) string {
}
// ReadFull implements File.
-func (f *vfsFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) {
+func (f *VFSFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) {
var total int64
for dst.NumBytes() > 0 {
n, err := f.file.PRead(ctx, dst, offset+total, vfs.ReadOptions{})
@@ -67,12 +67,12 @@ func (f *vfsFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset i
}
// ConfigureMMap implements File.
-func (f *vfsFile) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error {
+func (f *VFSFile) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error {
return f.file.ConfigureMMap(ctx, opts)
}
// Type implements File.
-func (f *vfsFile) Type(ctx context.Context) (linux.FileMode, error) {
+func (f *VFSFile) Type(ctx context.Context) (linux.FileMode, error) {
stat, err := f.file.Stat(ctx, vfs.StatOptions{})
if err != nil {
return 0, err
@@ -81,15 +81,21 @@ func (f *vfsFile) Type(ctx context.Context) (linux.FileMode, error) {
}
// IncRef implements File.
-func (f *vfsFile) IncRef() {
+func (f *VFSFile) IncRef() {
f.file.IncRef()
}
// DecRef implements File.
-func (f *vfsFile) DecRef() {
+func (f *VFSFile) DecRef() {
f.file.DecRef()
}
+// FileDescription returns the FileDescription represented by f. It does not
+// take an additional reference on the returned FileDescription.
+func (f *VFSFile) FileDescription() *vfs.FileDescription {
+ return f.file
+}
+
// fsLookup implements Lookup interface using fs.File.
//
// +stateify savable
@@ -132,5 +138,5 @@ func (l *vfsLookup) OpenPath(ctx context.Context, pathname string, opts vfs.Open
if err != nil {
return nil, err
}
- return &vfsFile{file: fd}, nil
+ return &VFSFile{file: fd}, nil
}
diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go
index 48eaccdbc..afea58f65 100644
--- a/pkg/sentry/fsimpl/ext/filesystem.go
+++ b/pkg/sentry/fsimpl/ext/filesystem.go
@@ -476,7 +476,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath
}
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
-func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
+func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
_, _, err := fs.walk(rp, false)
if err != nil {
return nil, err
@@ -485,7 +485,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([
}
// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt.
-func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) {
+func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) {
_, _, err := fs.walk(rp, false)
if err != nil {
return "", err
diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD
index d15a36709..99d1e3f8f 100644
--- a/pkg/sentry/fsimpl/gofer/BUILD
+++ b/pkg/sentry/fsimpl/gofer/BUILD
@@ -1,4 +1,4 @@
-load("//tools:defs.bzl", "go_library")
+load("//tools:defs.bzl", "go_library", "go_test")
load("//tools/go_generics:defs.bzl", "go_template_instance")
licenses(["notice"])
@@ -54,3 +54,13 @@ go_library(
"//pkg/usermem",
],
)
+
+go_test(
+ name = "gofer_test",
+ srcs = ["gofer_test.go"],
+ library = ":gofer",
+ deps = [
+ "//pkg/p9",
+ "//pkg/sentry/contexttest",
+ ],
+)
diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go
index 5dbfc6250..49d9f859b 100644
--- a/pkg/sentry/fsimpl/gofer/directory.go
+++ b/pkg/sentry/fsimpl/gofer/directory.go
@@ -56,14 +56,19 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba
fd.mu.Lock()
defer fd.mu.Unlock()
+ d := fd.dentry()
if fd.dirents == nil {
- ds, err := fd.dentry().getDirents(ctx)
+ ds, err := d.getDirents(ctx)
if err != nil {
return err
}
fd.dirents = ds
}
+ if d.fs.opts.interop != InteropModeShared {
+ d.touchAtime(fd.vfsfd.Mount())
+ }
+
for fd.off < int64(len(fd.dirents)) {
if err := cb.Handle(fd.dirents[fd.off]); err != nil {
return err
diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go
index 269624362..cd744bf5e 100644
--- a/pkg/sentry/fsimpl/gofer/filesystem.go
+++ b/pkg/sentry/fsimpl/gofer/filesystem.go
@@ -356,7 +356,9 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir
if err := create(parent, name); err != nil {
return err
}
- parent.touchCMtime(ctx)
+ if fs.opts.interop != InteropModeShared {
+ parent.touchCMtime()
+ }
delete(parent.negativeChildren, name)
parent.dirents = nil
return nil
@@ -435,14 +437,19 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b
flags := uint32(0)
if dir {
if child != nil && !child.isDir() {
+ vfsObj.AbortDeleteDentry(childVFSD)
return syserror.ENOTDIR
}
flags = linux.AT_REMOVEDIR
} else {
if child != nil && child.isDir() {
+ vfsObj.AbortDeleteDentry(childVFSD)
return syserror.EISDIR
}
if rp.MustBeDir() {
+ if childVFSD != nil {
+ vfsObj.AbortDeleteDentry(childVFSD)
+ }
return syserror.ENOTDIR
}
}
@@ -454,7 +461,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b
return err
}
if fs.opts.interop != InteropModeShared {
- parent.touchCMtime(ctx)
+ parent.touchCMtime()
if dir {
parent.decLinks()
}
@@ -802,7 +809,6 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving
d.IncRef() // reference held by child on its parent d
d.vfsd.InsertChild(&child.vfsd, name)
if d.fs.opts.interop != InteropModeShared {
- d.touchCMtime(ctx)
delete(d.negativeChildren, name)
d.dirents = nil
}
@@ -834,6 +840,9 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving
}
childVFSFD = &fd.vfsfd
}
+ if d.fs.opts.interop != InteropModeShared {
+ d.touchCMtime()
+ }
return childVFSFD, nil
}
@@ -975,6 +984,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
oldParent.decLinks()
newParent.incLinks()
}
+ oldParent.touchCMtime()
+ newParent.touchCMtime()
+ renamed.touchCtime()
}
vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD)
return nil
@@ -1068,7 +1080,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath
}
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
-func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
+func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
var ds *[]*dentry
fs.renameMu.RLock()
defer fs.renameMuRUnlockAndCheckCaching(&ds)
@@ -1076,11 +1088,11 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([
if err != nil {
return nil, err
}
- return d.listxattr(ctx)
+ return d.listxattr(ctx, rp.Credentials(), size)
}
// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt.
-func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) {
+func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) {
var ds *[]*dentry
fs.renameMu.RLock()
defer fs.renameMuRUnlockAndCheckCaching(&ds)
@@ -1088,7 +1100,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, nam
if err != nil {
return "", err
}
- return d.getxattr(ctx, name)
+ return d.getxattr(ctx, rp.Credentials(), &opts)
}
// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt.
@@ -1100,7 +1112,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt
if err != nil {
return err
}
- return d.setxattr(ctx, &opts)
+ return d.setxattr(ctx, rp.Credentials(), &opts)
}
// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt.
@@ -1112,7 +1124,7 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath,
if err != nil {
return err
}
- return d.removexattr(ctx, name)
+ return d.removexattr(ctx, rp.Credentials(), name)
}
// PrependPath implements vfs.FilesystemImpl.PrependPath.
diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go
index 8e41b6b1c..2485cdb53 100644
--- a/pkg/sentry/fsimpl/gofer/gofer.go
+++ b/pkg/sentry/fsimpl/gofer/gofer.go
@@ -34,6 +34,7 @@ package gofer
import (
"fmt"
"strconv"
+ "strings"
"sync"
"sync/atomic"
"syscall"
@@ -44,6 +45,7 @@ import (
"gvisor.dev/gvisor/pkg/p9"
"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
"gvisor.dev/gvisor/pkg/sentry/memmap"
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
"gvisor.dev/gvisor/pkg/sentry/vfs"
@@ -72,6 +74,9 @@ type filesystem struct {
// client is the client used by this filesystem. client is immutable.
client *p9.Client
+ // clock is a realtime clock used to set timestamps in file operations.
+ clock ktime.Clock
+
// uid and gid are the effective KUID and KGID of the filesystem's creator,
// and are used as the owner and group for files that don't specify one.
// uid and gid are immutable.
@@ -376,6 +381,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
uid: creds.EffectiveKUID,
gid: creds.EffectiveKGID,
client: client,
+ clock: ktime.RealtimeClockFromContext(ctx),
dentries: make(map[*dentry]struct{}),
specialFileFDs: make(map[*specialFileFD]struct{}),
}
@@ -439,7 +445,8 @@ type dentry struct {
// refs is the reference count. Each dentry holds a reference on its
// parent, even if disowned. refs is accessed using atomic memory
- // operations.
+ // operations. When refs reaches 0, the dentry may be added to the cache or
+ // destroyed. If refs==-1 the dentry has already been destroyed.
refs int64
// fs is the owning filesystem. fs is immutable.
@@ -779,10 +786,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin
// data, so there's no cache to truncate either.)
return nil
}
- now, haveNow := nowFromContext(ctx)
- if !haveNow {
- ctx.Warningf("gofer.dentry.setStat: current time not available")
- }
+ now := d.fs.clock.Now().Nanoseconds()
if stat.Mask&linux.STATX_MODE != 0 {
atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode))
}
@@ -794,25 +798,19 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin
}
if setLocalAtime {
if stat.Atime.Nsec == linux.UTIME_NOW {
- if haveNow {
- atomic.StoreInt64(&d.atime, now)
- }
+ atomic.StoreInt64(&d.atime, now)
} else {
atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime))
}
}
if setLocalMtime {
if stat.Mtime.Nsec == linux.UTIME_NOW {
- if haveNow {
- atomic.StoreInt64(&d.mtime, now)
- }
+ atomic.StoreInt64(&d.mtime, now)
} else {
atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime))
}
}
- if haveNow {
- atomic.StoreInt64(&d.ctime, now)
- }
+ atomic.StoreInt64(&d.ctime, now)
if stat.Mask&linux.STATX_SIZE != 0 {
d.dataMu.Lock()
oldSize := d.size
@@ -864,7 +862,7 @@ func (d *dentry) IncRef() {
func (d *dentry) TryIncRef() bool {
for {
refs := atomic.LoadInt64(&d.refs)
- if refs == 0 {
+ if refs <= 0 {
return false
}
if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) {
@@ -887,13 +885,20 @@ func (d *dentry) DecRef() {
// checkCachingLocked should be called after d's reference count becomes 0 or it
// becomes disowned.
//
+// It may be called on a destroyed dentry. For example,
+// renameMu[R]UnlockAndCheckCaching may call checkCachingLocked multiple times
+// for the same dentry when the dentry is visited more than once in the same
+// operation. One of the calls may destroy the dentry, so subsequent calls will
+// do nothing.
+//
// Preconditions: d.fs.renameMu must be locked for writing.
func (d *dentry) checkCachingLocked() {
// Dentries with a non-zero reference count must be retained. (The only way
// to obtain a reference on a dentry with zero references is via path
// resolution, which requires renameMu, so if d.refs is zero then it will
// remain zero while we hold renameMu for writing.)
- if atomic.LoadInt64(&d.refs) != 0 {
+ refs := atomic.LoadInt64(&d.refs)
+ if refs > 0 {
if d.cached {
d.fs.cachedDentries.Remove(d)
d.fs.cachedDentriesLen--
@@ -901,6 +906,10 @@ func (d *dentry) checkCachingLocked() {
}
return
}
+ if refs == -1 {
+ // Dentry has already been destroyed.
+ return
+ }
// Non-child dentries with zero references are no longer reachable by path
// resolution and should be dropped immediately.
if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() {
@@ -953,9 +962,22 @@ func (d *dentry) checkCachingLocked() {
}
}
+// destroyLocked destroys the dentry. It may flushes dirty pages from cache,
+// close p9 file and remove reference on parent dentry.
+//
// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is
// not a child dentry.
func (d *dentry) destroyLocked() {
+ switch atomic.LoadInt64(&d.refs) {
+ case 0:
+ // Mark the dentry destroyed.
+ atomic.StoreInt64(&d.refs, -1)
+ case -1:
+ panic("dentry.destroyLocked() called on already destroyed dentry")
+ default:
+ panic("dentry.destroyLocked() called with references on the dentry")
+ }
+
ctx := context.Background()
d.handleMu.Lock()
if !d.handle.file.isNil() {
@@ -975,7 +997,10 @@ func (d *dentry) destroyLocked() {
d.handle.close(ctx)
}
d.handleMu.Unlock()
- d.file.close(ctx)
+ if !d.file.isNil() {
+ d.file.close(ctx)
+ d.file = p9file{}
+ }
// Remove d from the set of all dentries.
d.fs.syncMu.Lock()
delete(d.fs.dentries, d)
@@ -1000,21 +1025,50 @@ func (d *dentry) setDeleted() {
atomic.StoreUint32(&d.deleted, 1)
}
-func (d *dentry) listxattr(ctx context.Context) ([]string, error) {
- return nil, syserror.ENOTSUP
+// We only support xattrs prefixed with "user." (see b/148380782). Currently,
+// there is no need to expose any other xattrs through a gofer.
+func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size uint64) ([]string, error) {
+ xattrMap, err := d.file.listXattr(ctx, size)
+ if err != nil {
+ return nil, err
+ }
+ xattrs := make([]string, 0, len(xattrMap))
+ for x := range xattrMap {
+ if strings.HasPrefix(x, linux.XATTR_USER_PREFIX) {
+ xattrs = append(xattrs, x)
+ }
+ }
+ return xattrs, nil
}
-func (d *dentry) getxattr(ctx context.Context, name string) (string, error) {
- // TODO(jamieliu): add vfs.GetxattrOptions.Size
- return d.file.getXattr(ctx, name, linux.XATTR_SIZE_MAX)
+func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) {
+ if err := d.checkPermissions(creds, vfs.MayRead); err != nil {
+ return "", err
+ }
+ if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) {
+ return "", syserror.EOPNOTSUPP
+ }
+ return d.file.getXattr(ctx, opts.Name, opts.Size)
}
-func (d *dentry) setxattr(ctx context.Context, opts *vfs.SetxattrOptions) error {
+func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.SetxattrOptions) error {
+ if err := d.checkPermissions(creds, vfs.MayWrite); err != nil {
+ return err
+ }
+ if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) {
+ return syserror.EOPNOTSUPP
+ }
return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags)
}
-func (d *dentry) removexattr(ctx context.Context, name string) error {
- return syserror.ENOTSUP
+func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name string) error {
+ if err := d.checkPermissions(creds, vfs.MayWrite); err != nil {
+ return err
+ }
+ if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) {
+ return syserror.EOPNOTSUPP
+ }
+ return d.file.removeXattr(ctx, name)
}
// Preconditions: d.isRegularFile() || d.isDirectory().
@@ -1065,7 +1119,7 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool
// description, but this doesn't matter since they refer to the
// same file (unless d.fs.opts.overlayfsStaleRead is true,
// which we handle separately).
- if err := syscall.Dup3(int(h.fd), int(d.handle.fd), 0); err != nil {
+ if err := syscall.Dup3(int(h.fd), int(d.handle.fd), syscall.O_CLOEXEC); err != nil {
d.handleMu.Unlock()
ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err)
h.close(ctx)
@@ -1165,21 +1219,21 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions)
}
// Listxattr implements vfs.FileDescriptionImpl.Listxattr.
-func (fd *fileDescription) Listxattr(ctx context.Context) ([]string, error) {
- return fd.dentry().listxattr(ctx)
+func (fd *fileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) {
+ return fd.dentry().listxattr(ctx, auth.CredentialsFromContext(ctx), size)
}
// Getxattr implements vfs.FileDescriptionImpl.Getxattr.
-func (fd *fileDescription) Getxattr(ctx context.Context, name string) (string, error) {
- return fd.dentry().getxattr(ctx, name)
+func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOptions) (string, error) {
+ return fd.dentry().getxattr(ctx, auth.CredentialsFromContext(ctx), &opts)
}
// Setxattr implements vfs.FileDescriptionImpl.Setxattr.
func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error {
- return fd.dentry().setxattr(ctx, &opts)
+ return fd.dentry().setxattr(ctx, auth.CredentialsFromContext(ctx), &opts)
}
// Removexattr implements vfs.FileDescriptionImpl.Removexattr.
func (fd *fileDescription) Removexattr(ctx context.Context, name string) error {
- return fd.dentry().removexattr(ctx, name)
+ return fd.dentry().removexattr(ctx, auth.CredentialsFromContext(ctx), name)
}
diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go
new file mode 100644
index 000000000..82bc239db
--- /dev/null
+++ b/pkg/sentry/fsimpl/gofer/gofer_test.go
@@ -0,0 +1,64 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package gofer
+
+import (
+ "sync/atomic"
+ "testing"
+
+ "gvisor.dev/gvisor/pkg/p9"
+ "gvisor.dev/gvisor/pkg/sentry/contexttest"
+)
+
+func TestDestroyIdempotent(t *testing.T) {
+ fs := filesystem{
+ dentries: make(map[*dentry]struct{}),
+ opts: filesystemOptions{
+ // Test relies on no dentry being held in the cache.
+ maxCachedDentries: 0,
+ },
+ }
+
+ ctx := contexttest.Context(t)
+ attr := &p9.Attr{
+ Mode: p9.ModeRegular,
+ }
+ mask := p9.AttrMask{
+ Mode: true,
+ Size: true,
+ }
+ parent, err := fs.newDentry(ctx, p9file{}, p9.QID{}, mask, attr)
+ if err != nil {
+ t.Fatalf("fs.newDentry(): %v", err)
+ }
+
+ child, err := fs.newDentry(ctx, p9file{}, p9.QID{}, mask, attr)
+ if err != nil {
+ t.Fatalf("fs.newDentry(): %v", err)
+ }
+ parent.IncRef() // reference held by child on its parent.
+ parent.vfsd.InsertChild(&child.vfsd, "child")
+
+ child.checkCachingLocked()
+ if got := atomic.LoadInt64(&child.refs); got != -1 {
+ t.Fatalf("child.refs=%d, want: -1", got)
+ }
+ // Parent will also be destroyed when child reference is removed.
+ if got := atomic.LoadInt64(&parent.refs); got != -1 {
+ t.Fatalf("parent.refs=%d, want: -1", got)
+ }
+ child.checkCachingLocked()
+ child.checkCachingLocked()
+}
diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go
index 755ac2985..87f0b877f 100644
--- a/pkg/sentry/fsimpl/gofer/p9file.go
+++ b/pkg/sentry/fsimpl/gofer/p9file.go
@@ -85,6 +85,13 @@ func (f p9file) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAt
return err
}
+func (f p9file) listXattr(ctx context.Context, size uint64) (map[string]struct{}, error) {
+ ctx.UninterruptibleSleepStart(false)
+ xattrs, err := f.file.ListXattr(size)
+ ctx.UninterruptibleSleepFinish(false)
+ return xattrs, err
+}
+
func (f p9file) getXattr(ctx context.Context, name string, size uint64) (string, error) {
ctx.UninterruptibleSleepStart(false)
val, err := f.file.GetXattr(name, size)
@@ -99,6 +106,13 @@ func (f p9file) setXattr(ctx context.Context, name, value string, flags uint32)
return err
}
+func (f p9file) removeXattr(ctx context.Context, name string) error {
+ ctx.UninterruptibleSleepStart(false)
+ err := f.file.RemoveXattr(name)
+ ctx.UninterruptibleSleepFinish(false)
+ return err
+}
+
func (f p9file) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error {
ctx.UninterruptibleSleepStart(false)
err := f.file.Allocate(mode, offset, length)
diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go
index 3593eb1d5..857f7c74e 100644
--- a/pkg/sentry/fsimpl/gofer/regular_file.go
+++ b/pkg/sentry/fsimpl/gofer/regular_file.go
@@ -104,7 +104,7 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs
putDentryReadWriter(rw)
if d.fs.opts.interop != InteropModeShared {
// Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed().
- d.touchAtime(ctx, fd.vfsfd.Mount())
+ d.touchAtime(fd.vfsfd.Mount())
}
return n, err
}
@@ -139,10 +139,7 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
// Compare Linux's mm/filemap.c:__generic_file_write_iter() =>
// file_update_time(). This is d.touchCMtime(), but without locking
// d.metadataMu (recursively).
- if now, ok := nowFromContext(ctx); ok {
- atomic.StoreInt64(&d.mtime, now)
- atomic.StoreInt64(&d.ctime, now)
- }
+ d.touchCMtimeLocked()
}
if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 {
// Write dirty cached pages that will be touched by the write back to
diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go
index 274f7346f..507e0e276 100644
--- a/pkg/sentry/fsimpl/gofer/special_file.go
+++ b/pkg/sentry/fsimpl/gofer/special_file.go
@@ -76,7 +76,7 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs
// hold here since specialFileFD doesn't client-cache data. Just buffer the
// read instead.
if d := fd.dentry(); d.fs.opts.interop != InteropModeShared {
- d.touchAtime(ctx, fd.vfsfd.Mount())
+ d.touchAtime(fd.vfsfd.Mount())
}
buf := make([]byte, dst.NumBytes())
n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset))
@@ -117,7 +117,7 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
// Do a buffered write. See rationale in PRead.
if d := fd.dentry(); d.fs.opts.interop != InteropModeShared {
- d.touchCMtime(ctx)
+ d.touchCMtime()
}
buf := make([]byte, src.NumBytes())
// Don't do partial writes if we get a partial read from src.
diff --git a/pkg/sentry/fsimpl/gofer/symlink.go b/pkg/sentry/fsimpl/gofer/symlink.go
index adf43be60..2ec819f86 100644
--- a/pkg/sentry/fsimpl/gofer/symlink.go
+++ b/pkg/sentry/fsimpl/gofer/symlink.go
@@ -27,7 +27,7 @@ func (d *dentry) isSymlink() bool {
// Precondition: d.isSymlink().
func (d *dentry) readlink(ctx context.Context, mnt *vfs.Mount) (string, error) {
if d.fs.opts.interop != InteropModeShared {
- d.touchAtime(ctx, mnt)
+ d.touchAtime(mnt)
d.dataMu.Lock()
if d.haveTarget {
target := d.target
diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go
index 7598ec6a8..2608e7e1d 100644
--- a/pkg/sentry/fsimpl/gofer/time.go
+++ b/pkg/sentry/fsimpl/gofer/time.go
@@ -18,8 +18,6 @@ import (
"sync/atomic"
"gvisor.dev/gvisor/pkg/abi/linux"
- "gvisor.dev/gvisor/pkg/context"
- ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
"gvisor.dev/gvisor/pkg/sentry/vfs"
)
@@ -38,23 +36,12 @@ func statxTimestampFromDentry(ns int64) linux.StatxTimestamp {
}
}
-func nowFromContext(ctx context.Context) (int64, bool) {
- if clock := ktime.RealtimeClockFromContext(ctx); clock != nil {
- return clock.Now().Nanoseconds(), true
- }
- return 0, false
-}
-
// Preconditions: fs.interop != InteropModeShared.
-func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) {
+func (d *dentry) touchAtime(mnt *vfs.Mount) {
if err := mnt.CheckBeginWrite(); err != nil {
return
}
- now, ok := nowFromContext(ctx)
- if !ok {
- mnt.EndWrite()
- return
- }
+ now := d.fs.clock.Now().Nanoseconds()
d.metadataMu.Lock()
atomic.StoreInt64(&d.atime, now)
d.metadataMu.Unlock()
@@ -63,13 +50,25 @@ func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) {
// Preconditions: fs.interop != InteropModeShared. The caller has successfully
// called vfs.Mount.CheckBeginWrite().
-func (d *dentry) touchCMtime(ctx context.Context) {
- now, ok := nowFromContext(ctx)
- if !ok {
- return
- }
+func (d *dentry) touchCtime() {
+ now := d.fs.clock.Now().Nanoseconds()
+ d.metadataMu.Lock()
+ atomic.StoreInt64(&d.ctime, now)
+ d.metadataMu.Unlock()
+}
+
+// Preconditions: fs.interop != InteropModeShared. The caller has successfully
+// called vfs.Mount.CheckBeginWrite().
+func (d *dentry) touchCMtime() {
+ now := d.fs.clock.Now().Nanoseconds()
d.metadataMu.Lock()
atomic.StoreInt64(&d.mtime, now)
atomic.StoreInt64(&d.ctime, now)
d.metadataMu.Unlock()
}
+
+func (d *dentry) touchCMtimeLocked() {
+ now := d.fs.clock.Now().Nanoseconds()
+ atomic.StoreInt64(&d.mtime, now)
+ atomic.StoreInt64(&d.ctime, now)
+}
diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go
index 7d9dcd4c9..97fa7f7ab 100644
--- a/pkg/sentry/fsimpl/host/host.go
+++ b/pkg/sentry/fsimpl/host/host.go
@@ -74,31 +74,34 @@ func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, err
}
// Retrieve metadata.
- var s syscall.Stat_t
- if err := syscall.Fstat(hostFD, &s); err != nil {
+ var s unix.Stat_t
+ if err := unix.Fstat(hostFD, &s); err != nil {
return nil, err
}
fileMode := linux.FileMode(s.Mode)
fileType := fileMode.FileType()
- // Pipes, character devices, and sockets.
- isStream := fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK
+
+ // Determine if hostFD is seekable. If not, this syscall will return ESPIPE
+ // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character
+ // devices.
+ _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR)
+ seekable := err != syserror.ESPIPE
i := &inode{
hostFD: hostFD,
- isStream: isStream,
+ seekable: seekable,
isTTY: isTTY,
canMap: canMap(uint32(fileType)),
ino: fs.NextIno(),
mode: fileMode,
- // For simplicity, set offset to 0. Technically, we should
- // only set to 0 on files that are not seekable (sockets, pipes, etc.),
- // and use the offset from the host fd otherwise.
+ // For simplicity, set offset to 0. Technically, we should use the existing
+ // offset on the host if the file is seekable.
offset: 0,
}
- // These files can't be memory mapped, assert this.
- if i.isStream && i.canMap {
+ // Non-seekable files can't be memory mapped, assert this.
+ if !i.seekable && i.canMap {
panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped")
}
@@ -124,12 +127,12 @@ type inode struct {
// This field is initialized at creation time and is immutable.
hostFD int
- // isStream is true if the host fd points to a file representing a stream,
+ // seekable is false if the host fd points to a file representing a stream,
// e.g. a socket or a pipe. Such files are not seekable and can return
// EWOULDBLOCK for I/O operations.
//
// This field is initialized at creation time and is immutable.
- isStream bool
+ seekable bool
// isTTY is true if this file represents a TTY.
//
@@ -481,8 +484,7 @@ func (f *fileDescription) Release() {
// PRead implements FileDescriptionImpl.
func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
i := f.inode
- // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
- if i.isStream {
+ if !i.seekable {
return 0, syserror.ESPIPE
}
@@ -492,8 +494,7 @@ func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, off
// Read implements FileDescriptionImpl.
func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
i := f.inode
- // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
- if i.isStream {
+ if !i.seekable {
n, err := readFromHostFD(ctx, i.hostFD, dst, -1, opts.Flags)
if isBlockError(err) {
// If we got any data at all, return it as a "completed" partial read
@@ -538,8 +539,7 @@ func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, off
// PWrite implements FileDescriptionImpl.
func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
i := f.inode
- // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
- if i.isStream {
+ if !i.seekable {
return 0, syserror.ESPIPE
}
@@ -549,8 +549,7 @@ func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, of
// Write implements FileDescriptionImpl.
func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
i := f.inode
- // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
- if i.isStream {
+ if !i.seekable {
n, err := writeToHostFD(ctx, i.hostFD, src, -1, opts.Flags)
if isBlockError(err) {
err = syserror.ErrWouldBlock
@@ -593,8 +592,7 @@ func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offs
// allow directory fds to be imported at all.
func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (int64, error) {
i := f.inode
- // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null.
- if i.isStream {
+ if !i.seekable {
return 0, syserror.ESPIPE
}
diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go
index a429fa23d..baf81b4db 100644
--- a/pkg/sentry/fsimpl/kernfs/filesystem.go
+++ b/pkg/sentry/fsimpl/kernfs/filesystem.go
@@ -63,6 +63,9 @@ afterSymlink:
rp.Advance()
return nextVFSD, nil
}
+ if len(name) > linux.NAME_MAX {
+ return nil, syserror.ENAMETOOLONG
+ }
d.dirMu.Lock()
nextVFSD, err := rp.ResolveChild(vfsd, name)
if err != nil {
@@ -76,16 +79,22 @@ afterSymlink:
}
// Resolve any symlink at current path component.
if rp.ShouldFollowSymlink() && next.isSymlink() {
- // TODO: VFS2 needs something extra for /proc/[pid]/fd/ "magic symlinks".
- target, err := next.inode.Readlink(ctx)
+ targetVD, targetPathname, err := next.inode.Getlink(ctx)
if err != nil {
return nil, err
}
- if err := rp.HandleSymlink(target); err != nil {
- return nil, err
+ if targetVD.Ok() {
+ err := rp.HandleJump(targetVD)
+ targetVD.DecRef()
+ if err != nil {
+ return nil, err
+ }
+ } else {
+ if err := rp.HandleSymlink(targetPathname); err != nil {
+ return nil, err
+ }
}
goto afterSymlink
-
}
rp.Advance()
return &next.vfsd, nil
@@ -191,6 +200,9 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v
if pc == "." || pc == ".." {
return "", syserror.EEXIST
}
+ if len(pc) > linux.NAME_MAX {
+ return "", syserror.ENAMETOOLONG
+ }
childVFSD, err := rp.ResolveChild(parentVFSD, pc)
if err != nil {
return "", err
@@ -433,6 +445,9 @@ afterTrailingSymlink:
if pc == "." || pc == ".." {
return nil, syserror.EISDIR
}
+ if len(pc) > linux.NAME_MAX {
+ return nil, syserror.ENAMETOOLONG
+ }
// Determine whether or not we need to create a file.
childVFSD, err := rp.ResolveChild(parentVFSD, pc)
if err != nil {
@@ -461,19 +476,25 @@ afterTrailingSymlink:
}
childDentry := childVFSD.Impl().(*Dentry)
childInode := childDentry.inode
- if rp.ShouldFollowSymlink() {
- if childDentry.isSymlink() {
- target, err := childInode.Readlink(ctx)
+ if rp.ShouldFollowSymlink() && childDentry.isSymlink() {
+ targetVD, targetPathname, err := childInode.Getlink(ctx)
+ if err != nil {
+ return nil, err
+ }
+ if targetVD.Ok() {
+ err := rp.HandleJump(targetVD)
+ targetVD.DecRef()
if err != nil {
return nil, err
}
- if err := rp.HandleSymlink(target); err != nil {
+ } else {
+ if err := rp.HandleSymlink(targetPathname); err != nil {
return nil, err
}
- // rp.Final() may no longer be true since we now need to resolve the
- // symlink target.
- goto afterTrailingSymlink
}
+ // rp.Final() may no longer be true since we now need to resolve the
+ // symlink target.
+ goto afterTrailingSymlink
}
if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil {
return nil, err
@@ -661,7 +682,7 @@ func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu
if err != nil {
return linux.Statfs{}, err
}
- // TODO: actually implement statfs
+ // TODO(gvisor.dev/issue/1193): actually implement statfs.
return linux.Statfs{}, syserror.ENOSYS
}
@@ -742,7 +763,7 @@ func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath
}
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
-func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
+func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
fs.mu.RLock()
_, _, err := fs.walkExistingLocked(ctx, rp)
fs.mu.RUnlock()
@@ -755,7 +776,7 @@ func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([
}
// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt.
-func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) {
+func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) {
fs.mu.RLock()
_, _, err := fs.walkExistingLocked(ctx, rp)
fs.mu.RUnlock()
diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go
index 5c84b10c9..65f09af5d 100644
--- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go
+++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go
@@ -181,6 +181,11 @@ func (InodeNotSymlink) Readlink(context.Context) (string, error) {
return "", syserror.EINVAL
}
+// Getlink implements Inode.Getlink.
+func (InodeNotSymlink) Getlink(context.Context) (vfs.VirtualDentry, string, error) {
+ return vfs.VirtualDentry{}, "", syserror.EINVAL
+}
+
// InodeAttrs partially implements the Inode interface, specifically the
// inodeMetadata sub interface. InodeAttrs provides functionality related to
// inode attributes.
diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go
index 2cefef020..ad76b9f64 100644
--- a/pkg/sentry/fsimpl/kernfs/kernfs.go
+++ b/pkg/sentry/fsimpl/kernfs/kernfs.go
@@ -414,7 +414,21 @@ type inodeDynamicLookup interface {
}
type inodeSymlink interface {
- // Readlink resolves the target of a symbolic link. If an inode is not a
+ // Readlink returns the target of a symbolic link. If an inode is not a
// symlink, the implementation should return EINVAL.
Readlink(ctx context.Context) (string, error)
+
+ // Getlink returns the target of a symbolic link, as used by path
+ // resolution:
+ //
+ // - If the inode is a "magic link" (a link whose target is most accurately
+ // represented as a VirtualDentry), Getlink returns (ok VirtualDentry, "",
+ // nil). A reference is taken on the returned VirtualDentry.
+ //
+ // - If the inode is an ordinary symlink, Getlink returns (zero-value
+ // VirtualDentry, symlink target, nil).
+ //
+ // - If the inode is not a symlink, Getlink returns (zero-value
+ // VirtualDentry, "", EINVAL).
+ Getlink(ctx context.Context) (vfs.VirtualDentry, string, error)
}
diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go
index 5918d3309..018aa503c 100644
--- a/pkg/sentry/fsimpl/kernfs/symlink.go
+++ b/pkg/sentry/fsimpl/kernfs/symlink.go
@@ -55,6 +55,11 @@ func (s *StaticSymlink) Readlink(_ context.Context) (string, error) {
return s.target, nil
}
+// Getlink implements Inode.Getlink.
+func (s *StaticSymlink) Getlink(_ context.Context) (vfs.VirtualDentry, string, error) {
+ return vfs.VirtualDentry{}, s.target, nil
+}
+
// SetStat implements Inode.SetStat not allowing inode attributes to be changed.
func (*StaticSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error {
return syserror.EPERM
diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD
index 8156984eb..17c1342b5 100644
--- a/pkg/sentry/fsimpl/proc/BUILD
+++ b/pkg/sentry/fsimpl/proc/BUILD
@@ -22,7 +22,6 @@ go_library(
"//pkg/log",
"//pkg/refs",
"//pkg/safemem",
- "//pkg/sentry/fs",
"//pkg/sentry/fsbridge",
"//pkg/sentry/fsimpl/kernfs",
"//pkg/sentry/inet",
diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go
index aee2a4392..888afc0fd 100644
--- a/pkg/sentry/fsimpl/proc/task.go
+++ b/pkg/sentry/fsimpl/proc/task.go
@@ -214,22 +214,6 @@ func newIO(t *kernel.Task, isThreadGroup bool) *ioData {
return &ioData{ioUsage: t}
}
-func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry {
- // Namespace symlinks should contain the namespace name and the inode number
- // for the namespace instance, so for example user:[123456]. We currently fake
- // the inode number by sticking the symlink inode in its place.
- target := fmt.Sprintf("%s:[%d]", ns, ino)
-
- inode := &kernfs.StaticSymlink{}
- // Note: credentials are overridden by taskOwnedInode.
- inode.Init(task.Credentials(), ino, target)
-
- taskInode := &taskOwnedInode{Inode: inode, owner: task}
- d := &kernfs.Dentry{}
- d.Init(taskInode)
- return d
-}
-
// newCgroupData creates inode that shows cgroup information.
// From man 7 cgroups: "For each cgroup hierarchy of which the process is a
// member, there is one entry containing three colon-separated fields:
diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go
index 76bfc5307..046265eca 100644
--- a/pkg/sentry/fsimpl/proc/task_fds.go
+++ b/pkg/sentry/fsimpl/proc/task_fds.go
@@ -30,34 +30,35 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
)
-type fdDir struct {
- inoGen InoGenerator
- task *kernel.Task
-
- // When produceSymlinks is set, dirents produces for the FDs are reported
- // as symlink. Otherwise, they are reported as regular files.
- produceSymlink bool
-}
-
-func (i *fdDir) lookup(name string) (*vfs.FileDescription, kernel.FDFlags, error) {
- fd, err := strconv.ParseUint(name, 10, 64)
- if err != nil {
- return nil, kernel.FDFlags{}, syserror.ENOENT
- }
-
+func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) {
var (
file *vfs.FileDescription
flags kernel.FDFlags
)
- i.task.WithMuLocked(func(t *kernel.Task) {
- if fdTable := t.FDTable(); fdTable != nil {
- file, flags = fdTable.GetVFS2(int32(fd))
+ t.WithMuLocked(func(t *kernel.Task) {
+ if fdt := t.FDTable(); fdt != nil {
+ file, flags = fdt.GetVFS2(fd)
}
})
+ return file, flags
+}
+
+func taskFDExists(t *kernel.Task, fd int32) bool {
+ file, _ := getTaskFD(t, fd)
if file == nil {
- return nil, kernel.FDFlags{}, syserror.ENOENT
+ return false
}
- return file, flags, nil
+ file.DecRef()
+ return true
+}
+
+type fdDir struct {
+ inoGen InoGenerator
+ task *kernel.Task
+
+ // When produceSymlinks is set, dirents produces for the FDs are reported
+ // as symlink. Otherwise, they are reported as regular files.
+ produceSymlink bool
}
// IterDirents implements kernfs.inodeDynamicLookup.
@@ -128,11 +129,15 @@ func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry {
// Lookup implements kernfs.inodeDynamicLookup.
func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) {
- file, _, err := i.lookup(name)
+ fdInt, err := strconv.ParseInt(name, 10, 32)
if err != nil {
- return nil, err
+ return nil, syserror.ENOENT
+ }
+ fd := int32(fdInt)
+ if !taskFDExists(i.task, fd) {
+ return nil, syserror.ENOENT
}
- taskDentry := newFDSymlink(i.task.Credentials(), file, i.inoGen.NextIno())
+ taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno())
return taskDentry.VFSDentry(), nil
}
@@ -169,19 +174,22 @@ func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentia
//
// +stateify savable
type fdSymlink struct {
- refs.AtomicRefCount
kernfs.InodeAttrs
+ kernfs.InodeNoopRefCount
kernfs.InodeSymlink
- file *vfs.FileDescription
+ task *kernel.Task
+ fd int32
}
var _ kernfs.Inode = (*fdSymlink)(nil)
-func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64) *kernfs.Dentry {
- file.IncRef()
- inode := &fdSymlink{file: file}
- inode.Init(creds, ino, linux.ModeSymlink|0777)
+func newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry {
+ inode := &fdSymlink{
+ task: task,
+ fd: fd,
+ }
+ inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777)
d := &kernfs.Dentry{}
d.Init(inode)
@@ -189,21 +197,25 @@ func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64
}
func (s *fdSymlink) Readlink(ctx context.Context) (string, error) {
+ file, _ := getTaskFD(s.task, s.fd)
+ if file == nil {
+ return "", syserror.ENOENT
+ }
+ defer file.DecRef()
root := vfs.RootFromContext(ctx)
defer root.DecRef()
-
- vfsObj := s.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem()
- return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry())
-}
-
-func (s *fdSymlink) DecRef() {
- s.AtomicRefCount.DecRefWithDestructor(func() {
- s.Destroy()
- })
+ return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry())
}
-func (s *fdSymlink) Destroy() {
- s.file.DecRef()
+func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) {
+ file, _ := getTaskFD(s.task, s.fd)
+ if file == nil {
+ return vfs.VirtualDentry{}, "", syserror.ENOENT
+ }
+ defer file.DecRef()
+ vd := file.VirtualDentry()
+ vd.IncRef()
+ return vd, "", nil
}
// fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory.
@@ -238,12 +250,18 @@ func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry {
// Lookup implements kernfs.inodeDynamicLookup.
func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) {
- file, flags, err := i.lookup(name)
+ fdInt, err := strconv.ParseInt(name, 10, 32)
if err != nil {
- return nil, err
+ return nil, syserror.ENOENT
+ }
+ fd := int32(fdInt)
+ if !taskFDExists(i.task, fd) {
+ return nil, syserror.ENOENT
+ }
+ data := &fdInfoData{
+ task: i.task,
+ fd: fd,
}
-
- data := &fdInfoData{file: file, flags: flags}
dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data)
return dentry.VFSDentry(), nil
}
@@ -262,26 +280,23 @@ type fdInfoData struct {
kernfs.DynamicBytesFile
refs.AtomicRefCount
- file *vfs.FileDescription
- flags kernel.FDFlags
+ task *kernel.Task
+ fd int32
}
var _ dynamicInode = (*fdInfoData)(nil)
-func (d *fdInfoData) DecRef() {
- d.AtomicRefCount.DecRefWithDestructor(d.destroy)
-}
-
-func (d *fdInfoData) destroy() {
- d.file.DecRef()
-}
-
// Generate implements vfs.DynamicBytesSource.Generate.
func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error {
+ file, descriptorFlags := getTaskFD(d.task, d.fd)
+ if file == nil {
+ return syserror.ENOENT
+ }
+ defer file.DecRef()
// TODO(b/121266871): Include pos, locks, and other data. For now we only
// have flags.
// See https://www.kernel.org/doc/Documentation/filesystems/proc.txt
- flags := uint(d.file.StatusFlags()) | d.flags.ToLinuxFileFlags()
+ flags := uint(file.StatusFlags()) | descriptorFlags.ToLinuxFileFlags()
fmt.Fprintf(buf, "flags:\t0%o\n", flags)
return nil
}
diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go
index df0d1bcc5..2c6f8bdfc 100644
--- a/pkg/sentry/fsimpl/proc/task_files.go
+++ b/pkg/sentry/fsimpl/proc/task_files.go
@@ -64,6 +64,16 @@ func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) {
return m, nil
}
+func checkTaskState(t *kernel.Task) error {
+ switch t.ExitState() {
+ case kernel.TaskExitZombie:
+ return syserror.EACCES
+ case kernel.TaskExitDead:
+ return syserror.ESRCH
+ }
+ return nil
+}
+
type bufferWriter struct {
buf *bytes.Buffer
}
@@ -610,12 +620,31 @@ func (s *exeSymlink) Readlink(ctx context.Context) (string, error) {
return exec.PathnameWithDeleted(ctx), nil
}
+// Getlink implements kernfs.Inode.Getlink.
+func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) {
+ if !kernel.ContextCanTrace(ctx, s.task, false) {
+ return vfs.VirtualDentry{}, "", syserror.EACCES
+ }
+
+ exec, err := s.executable()
+ if err != nil {
+ return vfs.VirtualDentry{}, "", err
+ }
+ defer exec.DecRef()
+
+ vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry()
+ vd.IncRef()
+ return vd, "", nil
+}
+
func (s *exeSymlink) executable() (file fsbridge.File, err error) {
+ if err := checkTaskState(s.task); err != nil {
+ return nil, err
+ }
+
s.task.WithMuLocked(func(t *kernel.Task) {
mm := t.MemoryManager()
if mm == nil {
- // TODO(b/34851096): Check shouldn't allow Readlink once the
- // Task is zombied.
err = syserror.EACCES
return
}
@@ -625,7 +654,7 @@ func (s *exeSymlink) executable() (file fsbridge.File, err error) {
// (with locks held).
file = mm.Executable()
if file == nil {
- err = syserror.ENOENT
+ err = syserror.ESRCH
}
})
return
@@ -692,3 +721,41 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error {
i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf)
return nil
}
+
+type namespaceSymlink struct {
+ kernfs.StaticSymlink
+
+ task *kernel.Task
+}
+
+func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry {
+ // Namespace symlinks should contain the namespace name and the inode number
+ // for the namespace instance, so for example user:[123456]. We currently fake
+ // the inode number by sticking the symlink inode in its place.
+ target := fmt.Sprintf("%s:[%d]", ns, ino)
+
+ inode := &namespaceSymlink{task: task}
+ // Note: credentials are overridden by taskOwnedInode.
+ inode.Init(task.Credentials(), ino, target)
+
+ taskInode := &taskOwnedInode{Inode: inode, owner: task}
+ d := &kernfs.Dentry{}
+ d.Init(taskInode)
+ return d
+}
+
+// Readlink implements Inode.
+func (s *namespaceSymlink) Readlink(ctx context.Context) (string, error) {
+ if err := checkTaskState(s.task); err != nil {
+ return "", err
+ }
+ return s.StaticSymlink.Readlink(ctx)
+}
+
+// Getlink implements Inode.Getlink.
+func (s *namespaceSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) {
+ if err := checkTaskState(s.task); err != nil {
+ return vfs.VirtualDentry{}, "", err
+ }
+ return s.StaticSymlink.Getlink(ctx)
+}
diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go
index 373a7b17d..6595fcee6 100644
--- a/pkg/sentry/fsimpl/proc/task_net.go
+++ b/pkg/sentry/fsimpl/proc/task_net.go
@@ -24,7 +24,6 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/log"
- "gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
"gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/sentry/kernel"
@@ -32,6 +31,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/socket"
"gvisor.dev/gvisor/pkg/sentry/socket/unix"
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/usermem"
@@ -206,22 +206,21 @@ var _ dynamicInode = (*netUnixData)(nil)
func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error {
buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n")
for _, se := range n.kernel.ListSockets() {
- s := se.Sock.Get()
- if s == nil {
- log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock)
+ s := se.SockVFS2
+ if !s.TryIncRef() {
+ log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s)
continue
}
- sfile := s.(*fs.File)
- if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX {
+ if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX {
s.DecRef()
// Not a unix socket.
continue
}
- sops := sfile.FileOperations.(*unix.SocketOperations)
+ sops := s.Impl().(*unix.SocketVFS2)
addr, err := sops.Endpoint().GetLocalAddress()
if err != nil {
- log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err)
+ log.Warningf("Failed to retrieve socket name from %+v: %v", s, err)
addr.Addr = "<unknown>"
}
@@ -234,6 +233,15 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error {
}
}
+ // Get inode number.
+ var ino uint64
+ stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_INO})
+ if statErr != nil || stat.Mask&linux.STATX_INO == 0 {
+ log.Warningf("Failed to retrieve ino for socket file: %v", statErr)
+ } else {
+ ino = stat.Ino
+ }
+
// In the socket entry below, the value for the 'Num' field requires
// some consideration. Linux prints the address to the struct
// unix_sock representing a socket in the kernel, but may redact the
@@ -252,14 +260,14 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error {
// the definition of this struct changes over time.
//
// For now, we always redact this pointer.
- fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d",
+ fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %8d",
(*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct.
- sfile.ReadRefs()-1, // RefCount, don't count our own ref.
+ s.Refs()-1, // RefCount, don't count our own ref.
0, // Protocol, always 0 for UDS.
sockFlags, // Flags.
sops.Endpoint().Type(), // Type.
sops.State(), // State.
- sfile.InodeID(), // Inode.
+ ino, // Inode.
)
// Path
@@ -341,15 +349,14 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel,
t := kernel.TaskFromContext(ctx)
for _, se := range k.ListSockets() {
- s := se.Sock.Get()
- if s == nil {
- log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID)
+ s := se.SockVFS2
+ if !s.TryIncRef() {
+ log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s)
continue
}
- sfile := s.(*fs.File)
- sops, ok := sfile.FileOperations.(socket.Socket)
+ sops, ok := s.Impl().(socket.SocketVFS2)
if !ok {
- panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile))
+ panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s))
}
if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) {
s.DecRef()
@@ -398,14 +405,15 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel,
// Unimplemented.
fmt.Fprintf(buf, "%08X ", 0)
+ stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO})
+
// Field: uid.
- uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx)
- if err != nil {
- log.Warningf("Failed to retrieve unstable attr for socket file: %v", err)
+ if statErr != nil || stat.Mask&linux.STATX_UID == 0 {
+ log.Warningf("Failed to retrieve uid for socket file: %v", statErr)
fmt.Fprintf(buf, "%5d ", 0)
} else {
creds := auth.CredentialsFromContext(ctx)
- fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow()))
+ fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow()))
}
// Field: timeout; number of unanswered 0-window probes.
@@ -413,11 +421,16 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel,
fmt.Fprintf(buf, "%8d ", 0)
// Field: inode.
- fmt.Fprintf(buf, "%8d ", sfile.InodeID())
+ if statErr != nil || stat.Mask&linux.STATX_INO == 0 {
+ log.Warningf("Failed to retrieve inode for socket file: %v", statErr)
+ fmt.Fprintf(buf, "%8d ", 0)
+ } else {
+ fmt.Fprintf(buf, "%8d ", stat.Ino)
+ }
// Field: refcount. Don't count the ref we obtain while deferencing
// the weakref to this socket.
- fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1)
+ fmt.Fprintf(buf, "%d ", s.Refs()-1)
// Field: Socket struct address. Redacted due to the same reason as
// the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData.
@@ -499,15 +512,14 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error {
t := kernel.TaskFromContext(ctx)
for _, se := range d.kernel.ListSockets() {
- s := se.Sock.Get()
- if s == nil {
- log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID)
+ s := se.SockVFS2
+ if !s.TryIncRef() {
+ log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s)
continue
}
- sfile := s.(*fs.File)
- sops, ok := sfile.FileOperations.(socket.Socket)
+ sops, ok := s.Impl().(socket.SocketVFS2)
if !ok {
- panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile))
+ panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s))
}
if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM {
s.DecRef()
@@ -551,25 +563,31 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error {
// Field: retrnsmt. Always 0 for UDP.
fmt.Fprintf(buf, "%08X ", 0)
+ stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO})
+
// Field: uid.
- uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx)
- if err != nil {
- log.Warningf("Failed to retrieve unstable attr for socket file: %v", err)
+ if statErr != nil || stat.Mask&linux.STATX_UID == 0 {
+ log.Warningf("Failed to retrieve uid for socket file: %v", statErr)
fmt.Fprintf(buf, "%5d ", 0)
} else {
creds := auth.CredentialsFromContext(ctx)
- fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow()))
+ fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow()))
}
// Field: timeout. Always 0 for UDP.
fmt.Fprintf(buf, "%8d ", 0)
// Field: inode.
- fmt.Fprintf(buf, "%8d ", sfile.InodeID())
+ if statErr != nil || stat.Mask&linux.STATX_INO == 0 {
+ log.Warningf("Failed to retrieve inode for socket file: %v", statErr)
+ fmt.Fprintf(buf, "%8d ", 0)
+ } else {
+ fmt.Fprintf(buf, "%8d ", stat.Ino)
+ }
// Field: ref; reference count on the socket inode. Don't count the ref
// we obtain while deferencing the weakref to this socket.
- fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1)
+ fmt.Fprintf(buf, "%d ", s.Refs()-1)
// Field: Socket struct address. Redacted due to the same reason as
// the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData.
@@ -670,9 +688,9 @@ func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error {
if line.prefix == "Tcp" {
tcp := stat.(*inet.StatSNMPTCP)
// "Tcp" needs special processing because MaxConn is signed. RFC 2012.
- fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:]))
+ fmt.Fprintf(buf, "%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:]))
} else {
- fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat)))
+ fmt.Fprintf(buf, "%s: %s\n", line.prefix, sprintSlice(toSlice(stat)))
}
}
return nil
diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go
index 882c1981e..4621e2de0 100644
--- a/pkg/sentry/fsimpl/proc/tasks_files.go
+++ b/pkg/sentry/fsimpl/proc/tasks_files.go
@@ -63,6 +63,11 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) {
return strconv.FormatUint(uint64(tgid), 10), nil
}
+func (s *selfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) {
+ target, err := s.Readlink(ctx)
+ return vfs.VirtualDentry{}, target, err
+}
+
// SetStat implements Inode.SetStat not allowing inode attributes to be changed.
func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error {
return syserror.EPERM
@@ -101,6 +106,11 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) {
return fmt.Sprintf("%d/task/%d", tgid, tid), nil
}
+func (s *threadSelfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) {
+ target, err := s.Readlink(ctx)
+ return vfs.VirtualDentry{}, target, err
+}
+
// SetStat implements Inode.SetStat not allowing inode attributes to be changed.
func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error {
return syserror.EPERM
diff --git a/pkg/sentry/fsimpl/sockfs/BUILD b/pkg/sentry/fsimpl/sockfs/BUILD
index 790d50e65..52084ddb5 100644
--- a/pkg/sentry/fsimpl/sockfs/BUILD
+++ b/pkg/sentry/fsimpl/sockfs/BUILD
@@ -7,6 +7,7 @@ go_library(
srcs = ["sockfs.go"],
visibility = ["//pkg/sentry:internal"],
deps = [
+ "//pkg/abi/linux",
"//pkg/context",
"//pkg/sentry/fsimpl/kernfs",
"//pkg/sentry/kernel/auth",
diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go
index c13511de2..3f7ad1d65 100644
--- a/pkg/sentry/fsimpl/sockfs/sockfs.go
+++ b/pkg/sentry/fsimpl/sockfs/sockfs.go
@@ -16,6 +16,7 @@
package sockfs
import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
@@ -60,6 +61,10 @@ type filesystem struct {
}
// inode implements kernfs.Inode.
+//
+// TODO(gvisor.dev/issue/1476): Add device numbers to this inode (which are
+// not included in InodeAttrs) to store the numbers of the appropriate
+// socket device. Override InodeAttrs.Stat() accordingly.
type inode struct {
kernfs.InodeNotDirectory
kernfs.InodeNotSymlink
@@ -71,3 +76,27 @@ type inode struct {
func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
return nil, syserror.ENXIO
}
+
+// InitSocket initializes a socket FileDescription, with a corresponding
+// Dentry in mnt.
+//
+// fd should be the FileDescription associated with socketImpl, i.e. its first
+// field. mnt should be the global socket mount, Kernel.socketMount.
+func InitSocket(socketImpl vfs.FileDescriptionImpl, fd *vfs.FileDescription, mnt *vfs.Mount, creds *auth.Credentials) error {
+ fsimpl := mnt.Filesystem().Impl()
+ fs := fsimpl.(*kernfs.Filesystem)
+
+ // File mode matches net/socket.c:sock_alloc.
+ filemode := linux.FileMode(linux.S_IFSOCK | 0600)
+ i := &inode{}
+ i.Init(creds, fs.NextIno(), filemode)
+
+ d := &kernfs.Dentry{}
+ d.Init(i)
+
+ opts := &vfs.FileDescriptionOptions{UseDentryMetadata: true}
+ if err := fd.Init(socketImpl, linux.O_RDWR, mnt, d.VFSDentry(), opts); err != nil {
+ return err
+ }
+ return nil
+}
diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD
index 6ea35affb..4e6cd3491 100644
--- a/pkg/sentry/fsimpl/tmpfs/BUILD
+++ b/pkg/sentry/fsimpl/tmpfs/BUILD
@@ -24,6 +24,7 @@ go_library(
"filesystem.go",
"named_pipe.go",
"regular_file.go",
+ "socket_file.go",
"symlink.go",
"tmpfs.go",
],
@@ -50,6 +51,7 @@ go_library(
"//pkg/sentry/usage",
"//pkg/sentry/vfs",
"//pkg/sentry/vfs/lock",
+ "//pkg/sentry/vfs/memxattr",
"//pkg/sync",
"//pkg/syserror",
"//pkg/usermem",
diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go
index 37c75ab64..45712c9b9 100644
--- a/pkg/sentry/fsimpl/tmpfs/directory.go
+++ b/pkg/sentry/fsimpl/tmpfs/directory.go
@@ -68,6 +68,8 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba
fs.mu.Lock()
defer fs.mu.Unlock()
+ fd.inode().touchAtime(fd.vfsfd.Mount())
+
if fd.off == 0 {
if err := cb.Handle(vfs.Dirent{
Name: ".",
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index e678ecc37..f4d50d64f 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -46,6 +46,9 @@ func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) {
return nil, err
}
afterSymlink:
+ if len(rp.Component()) > linux.NAME_MAX {
+ return nil, syserror.ENAMETOOLONG
+ }
nextVFSD, err := rp.ResolveComponent(&d.vfsd)
if err != nil {
return nil, err
@@ -57,7 +60,7 @@ afterSymlink:
}
next := nextVFSD.Impl().(*dentry)
if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() {
- // TODO(gvisor.dev/issues/1197): Symlink traversals updates
+ // TODO(gvisor.dev/issue/1197): Symlink traversals updates
// access time.
if err := rp.HandleSymlink(symlink.target); err != nil {
return nil, err
@@ -133,6 +136,9 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa
if name == "." || name == ".." {
return syserror.EEXIST
}
+ if len(name) > linux.NAME_MAX {
+ return syserror.ENAMETOOLONG
+ }
// Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(),
// because if the child exists we want to return EEXIST immediately instead
// of attempting symlink/mount traversal.
@@ -153,7 +159,11 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa
return err
}
defer mnt.EndWrite()
- return create(parent, name)
+ if err := create(parent, name); err != nil {
+ return err
+ }
+ parent.inode.touchCMtime()
+ return nil
}
// AccessAt implements vfs.Filesystem.Impl.AccessAt.
@@ -251,8 +261,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
case linux.S_IFCHR:
childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor)
case linux.S_IFSOCK:
- // Not yet supported.
- return syserror.EPERM
+ childInode = fs.newSocketFile(rp.Credentials(), opts.Mode, opts.Endpoint)
default:
return syserror.EINVAL
}
@@ -328,7 +337,12 @@ afterTrailingSymlink:
child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode))
parent.vfsd.InsertChild(&child.vfsd, name)
parent.inode.impl.(*directory).childList.PushBack(child)
- return child.open(ctx, rp, &opts, true)
+ fd, err := child.open(ctx, rp, &opts, true)
+ if err != nil {
+ return nil, err
+ }
+ parent.inode.touchCMtime()
+ return fd, nil
}
if err != nil {
return nil, err
@@ -381,6 +395,8 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open
return newNamedPipeFD(ctx, impl, rp, &d.vfsd, opts.Flags)
case *deviceFile:
return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts)
+ case *socketFile:
+ return nil, syserror.ENXIO
default:
panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl))
}
@@ -398,6 +414,7 @@ func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st
if !ok {
return "", syserror.EINVAL
}
+ symlink.inode.touchAtime(rp.Mount())
return symlink.target, nil
}
@@ -515,7 +532,10 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
oldParent.inode.decLinksLocked()
newParent.inode.incLinksLocked()
}
- // TODO(gvisor.dev/issues/1197): Update timestamps and parent directory
+ oldParent.inode.touchCMtime()
+ newParent.inode.touchCMtime()
+ renamed.inode.touchCtime()
+ // TODO(gvisor.dev/issue/1197): Update timestamps and parent directory
// sizes.
vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD)
return nil
@@ -565,6 +585,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
parent.inode.decLinksLocked() // from child's ".."
child.inode.decLinksLocked()
vfsObj.CommitDeleteDentry(childVFSD)
+ parent.inode.touchCMtime()
return nil
}
@@ -600,7 +621,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu
if err != nil {
return linux.Statfs{}, err
}
- // TODO(gvisor.dev/issues/1197): Actually implement statfs.
+ // TODO(gvisor.dev/issue/1197): Actually implement statfs.
return linux.Statfs{}, syserror.ENOSYS
}
@@ -654,62 +675,68 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
parent.inode.impl.(*directory).childList.Remove(child)
child.inode.decLinksLocked()
vfsObj.CommitDeleteDentry(childVFSD)
+ parent.inode.touchCMtime()
return nil
}
// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
-//
-// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt.
func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
- return nil, syserror.ECONNREFUSED
+ fs.mu.RLock()
+ defer fs.mu.RUnlock()
+ d, err := resolveLocked(rp)
+ if err != nil {
+ return nil, err
+ }
+ switch impl := d.inode.impl.(type) {
+ case *socketFile:
+ return impl.ep, nil
+ default:
+ return nil, syserror.ECONNREFUSED
+ }
}
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
-func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
+func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- _, err := resolveLocked(rp)
+ d, err := resolveLocked(rp)
if err != nil {
return nil, err
}
- // TODO(b/127675828): support extended attributes
- return nil, syserror.ENOTSUP
+ return d.inode.listxattr(size)
}
// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt.
-func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) {
+func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- _, err := resolveLocked(rp)
+ d, err := resolveLocked(rp)
if err != nil {
return "", err
}
- // TODO(b/127675828): support extended attributes
- return "", syserror.ENOTSUP
+ return d.inode.getxattr(rp.Credentials(), &opts)
}
// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt.
func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error {
fs.mu.RLock()
defer fs.mu.RUnlock()
- _, err := resolveLocked(rp)
+ d, err := resolveLocked(rp)
if err != nil {
return err
}
- // TODO(b/127675828): support extended attributes
- return syserror.ENOTSUP
+ return d.inode.setxattr(rp.Credentials(), &opts)
}
// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt.
func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error {
fs.mu.RLock()
defer fs.mu.RUnlock()
- _, err := resolveLocked(rp)
+ d, err := resolveLocked(rp)
if err != nil {
return err
}
- // TODO(b/127675828): support extended attributes
- return syserror.ENOTSUP
+ return d.inode.removexattr(rp.Credentials(), name)
}
// PrependPath implements vfs.FilesystemImpl.PrependPath.
diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go
index 26cd65605..57e5e28ec 100644
--- a/pkg/sentry/fsimpl/tmpfs/regular_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go
@@ -286,7 +286,8 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs
rw := getRegularFileReadWriter(f, offset)
n, err := dst.CopyOutFrom(ctx, rw)
putRegularFileReadWriter(rw)
- return int64(n), err
+ fd.inode().touchAtime(fd.vfsfd.Mount())
+ return n, err
}
// Read implements vfs.FileDescriptionImpl.Read.
@@ -323,6 +324,7 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
f.inode.mu.Lock()
rw := getRegularFileReadWriter(f, offset)
n, err := src.CopyInTo(ctx, rw)
+ fd.inode().touchCMtimeLocked()
f.inode.mu.Unlock()
putRegularFileReadWriter(rw)
return n, err
diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go
new file mode 100644
index 000000000..25c2321af
--- /dev/null
+++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go
@@ -0,0 +1,34 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package tmpfs
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+)
+
+// socketFile is a socket (=S_IFSOCK) tmpfs file.
+type socketFile struct {
+ inode inode
+ ep transport.BoundEndpoint
+}
+
+func (fs *filesystem) newSocketFile(creds *auth.Credentials, mode linux.FileMode, ep transport.BoundEndpoint) *inode {
+ file := &socketFile{ep: ep}
+ file.inode.init(file, fs, creds, mode)
+ file.inode.nlink = 1 // from parent directory
+ return &file.inode
+}
diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go
index ebe035dee..d4f59ee5b 100644
--- a/pkg/sentry/fsimpl/tmpfs/stat_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go
@@ -29,7 +29,7 @@ func TestStatAfterCreate(t *testing.T) {
mode := linux.FileMode(0644)
// Run with different file types.
- // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets.
+ // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets.
for _, typ := range []string{"file", "dir", "pipe"} {
t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) {
var (
@@ -140,7 +140,7 @@ func TestSetStatAtime(t *testing.T) {
Mask: 0,
Atime: linux.NsecToStatxTimestamp(100),
}}); err != nil {
- t.Errorf("SetStat atime without mask failed: %v")
+ t.Errorf("SetStat atime without mask failed: %v", err)
}
// Atime should be unchanged.
if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil {
@@ -155,7 +155,7 @@ func TestSetStatAtime(t *testing.T) {
Atime: linux.NsecToStatxTimestamp(100),
}
if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil {
- t.Errorf("SetStat atime with mask failed: %v")
+ t.Errorf("SetStat atime with mask failed: %v", err)
}
if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil {
t.Errorf("Stat got error: %v", err)
@@ -169,7 +169,7 @@ func TestSetStat(t *testing.T) {
mode := linux.FileMode(0644)
// Run with different file types.
- // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets.
+ // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets.
for _, typ := range []string{"file", "dir", "pipe"} {
t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) {
var (
@@ -205,7 +205,7 @@ func TestSetStat(t *testing.T) {
Mask: 0,
Atime: linux.NsecToStatxTimestamp(100),
}}); err != nil {
- t.Errorf("SetStat atime without mask failed: %v")
+ t.Errorf("SetStat atime without mask failed: %v", err)
}
// Atime should be unchanged.
if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil {
@@ -220,7 +220,7 @@ func TestSetStat(t *testing.T) {
Atime: linux.NsecToStatxTimestamp(100),
}
if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil {
- t.Errorf("SetStat atime with mask failed: %v")
+ t.Errorf("SetStat atime with mask failed: %v", err)
}
if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil {
t.Errorf("Stat got error: %v", err)
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index 8bc8818c0..9fa8637d5 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -27,6 +27,7 @@ package tmpfs
import (
"fmt"
"math"
+ "strings"
"sync/atomic"
"gvisor.dev/gvisor/pkg/abi/linux"
@@ -37,6 +38,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/sentry/vfs/lock"
+ "gvisor.dev/gvisor/pkg/sentry/vfs/memxattr"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -186,6 +188,11 @@ type inode struct {
// filesystem.RmdirAt() drops the reference.
refs int64
+ // xattrs implements extended attributes.
+ //
+ // TODO(b/148380782): Support xattrs other than user.*
+ xattrs memxattr.SimpleExtendedAttributes
+
// Inode metadata. Writing multiple fields atomically requires holding
// mu, othewise atomic operations can be used.
mu sync.Mutex
@@ -315,7 +322,7 @@ func (i *inode) statTo(stat *linux.Statx) {
stat.Atime = linux.NsecToStatxTimestamp(i.atime)
stat.Ctime = linux.NsecToStatxTimestamp(i.ctime)
stat.Mtime = linux.NsecToStatxTimestamp(i.mtime)
- // TODO(gvisor.dev/issues/1197): Device number.
+ // TODO(gvisor.dev/issue/1197): Device number.
switch impl := i.impl.(type) {
case *regularFile:
stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS
@@ -331,7 +338,7 @@ func (i *inode) statTo(stat *linux.Statx) {
case *deviceFile:
stat.RdevMajor = impl.major
stat.RdevMinor = impl.minor
- case *directory, *namedPipe:
+ case *socketFile, *directory, *namedPipe:
// Nothing to do.
default:
panic(fmt.Sprintf("unknown inode type: %T", i.impl))
@@ -385,28 +392,41 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu
return syserror.EINVAL
}
}
+ now := i.clock.Now().Nanoseconds()
if mask&linux.STATX_ATIME != 0 {
- atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped())
+ if stat.Atime.Nsec == linux.UTIME_NOW {
+ atomic.StoreInt64(&i.atime, now)
+ } else {
+ atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped())
+ }
needsCtimeBump = true
}
if mask&linux.STATX_MTIME != 0 {
- atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped())
+ if stat.Mtime.Nsec == linux.UTIME_NOW {
+ atomic.StoreInt64(&i.mtime, now)
+ } else {
+ atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped())
+ }
needsCtimeBump = true
// Ignore the mtime bump, since we just set it ourselves.
needsMtimeBump = false
}
if mask&linux.STATX_CTIME != 0 {
- atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped())
+ if stat.Ctime.Nsec == linux.UTIME_NOW {
+ atomic.StoreInt64(&i.ctime, now)
+ } else {
+ atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped())
+ }
// Ignore the ctime bump, since we just set it ourselves.
needsCtimeBump = false
}
- now := i.clock.Now().Nanoseconds()
if needsMtimeBump {
atomic.StoreInt64(&i.mtime, now)
}
if needsCtimeBump {
atomic.StoreInt64(&i.ctime, now)
}
+
i.mu.Unlock()
return nil
}
@@ -466,6 +486,8 @@ func (i *inode) direntType() uint8 {
return linux.DT_DIR
case *symlink:
return linux.DT_LNK
+ case *socketFile:
+ return linux.DT_SOCK
case *deviceFile:
switch impl.kind {
case vfs.BlockDevice:
@@ -484,6 +506,92 @@ func (i *inode) isDir() bool {
return linux.FileMode(i.mode).FileType() == linux.S_IFDIR
}
+func (i *inode) touchAtime(mnt *vfs.Mount) {
+ if err := mnt.CheckBeginWrite(); err != nil {
+ return
+ }
+ now := i.clock.Now().Nanoseconds()
+ i.mu.Lock()
+ atomic.StoreInt64(&i.atime, now)
+ i.mu.Unlock()
+ mnt.EndWrite()
+}
+
+// Preconditions: The caller has called vfs.Mount.CheckBeginWrite().
+func (i *inode) touchCtime() {
+ now := i.clock.Now().Nanoseconds()
+ i.mu.Lock()
+ atomic.StoreInt64(&i.ctime, now)
+ i.mu.Unlock()
+}
+
+// Preconditions: The caller has called vfs.Mount.CheckBeginWrite().
+func (i *inode) touchCMtime() {
+ now := i.clock.Now().Nanoseconds()
+ i.mu.Lock()
+ atomic.StoreInt64(&i.mtime, now)
+ atomic.StoreInt64(&i.ctime, now)
+ i.mu.Unlock()
+}
+
+// Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds
+// inode.mu.
+func (i *inode) touchCMtimeLocked() {
+ now := i.clock.Now().Nanoseconds()
+ atomic.StoreInt64(&i.mtime, now)
+ atomic.StoreInt64(&i.ctime, now)
+}
+
+func (i *inode) listxattr(size uint64) ([]string, error) {
+ return i.xattrs.Listxattr(size)
+}
+
+func (i *inode) getxattr(creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) {
+ if err := i.checkPermissions(creds, vfs.MayRead); err != nil {
+ return "", err
+ }
+ if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) {
+ return "", syserror.EOPNOTSUPP
+ }
+ if !i.userXattrSupported() {
+ return "", syserror.ENODATA
+ }
+ return i.xattrs.Getxattr(opts)
+}
+
+func (i *inode) setxattr(creds *auth.Credentials, opts *vfs.SetxattrOptions) error {
+ if err := i.checkPermissions(creds, vfs.MayWrite); err != nil {
+ return err
+ }
+ if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) {
+ return syserror.EOPNOTSUPP
+ }
+ if !i.userXattrSupported() {
+ return syserror.EPERM
+ }
+ return i.xattrs.Setxattr(opts)
+}
+
+func (i *inode) removexattr(creds *auth.Credentials, name string) error {
+ if err := i.checkPermissions(creds, vfs.MayWrite); err != nil {
+ return err
+ }
+ if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) {
+ return syserror.EOPNOTSUPP
+ }
+ if !i.userXattrSupported() {
+ return syserror.EPERM
+ }
+ return i.xattrs.Removexattr(name)
+}
+
+// Extended attributes in the user.* namespace are only supported for regular
+// files and directories.
+func (i *inode) userXattrSupported() bool {
+ filetype := linux.S_IFMT & atomic.LoadUint32(&i.mode)
+ return filetype == linux.S_IFREG || filetype == linux.S_IFDIR
+}
+
// fileDescription is embedded by tmpfs implementations of
// vfs.FileDescriptionImpl.
type fileDescription struct {
@@ -511,3 +619,23 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions)
creds := auth.CredentialsFromContext(ctx)
return fd.inode().setStat(ctx, creds, &opts.Stat)
}
+
+// Listxattr implements vfs.FileDescriptionImpl.Listxattr.
+func (fd *fileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) {
+ return fd.inode().listxattr(size)
+}
+
+// Getxattr implements vfs.FileDescriptionImpl.Getxattr.
+func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOptions) (string, error) {
+ return fd.inode().getxattr(auth.CredentialsFromContext(ctx), &opts)
+}
+
+// Setxattr implements vfs.FileDescriptionImpl.Setxattr.
+func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error {
+ return fd.inode().setxattr(auth.CredentialsFromContext(ctx), &opts)
+}
+
+// Removexattr implements vfs.FileDescriptionImpl.Removexattr.
+func (fd *fileDescription) Removexattr(ctx context.Context, name string) error {
+ return fd.inode().removexattr(auth.CredentialsFromContext(ctx), name)
+}
diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index bb7e3cbc3..e0ff58d8c 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -169,6 +169,7 @@ go_library(
"//pkg/sentry/fs/lock",
"//pkg/sentry/fs/timerfd",
"//pkg/sentry/fsbridge",
+ "//pkg/sentry/fsimpl/kernfs",
"//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/hostcpu",
"//pkg/sentry/inet",
diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go
index d09d97825..ed40b5303 100644
--- a/pkg/sentry/kernel/fd_table.go
+++ b/pkg/sentry/kernel/fd_table.go
@@ -307,6 +307,61 @@ func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags
return fds, nil
}
+// NewFDsVFS2 allocates new FDs guaranteed to be the lowest number available
+// greater than or equal to the fd parameter. All files will share the set
+// flags. Success is guaranteed to be all or none.
+func (f *FDTable) NewFDsVFS2(ctx context.Context, fd int32, files []*vfs.FileDescription, flags FDFlags) (fds []int32, err error) {
+ if fd < 0 {
+ // Don't accept negative FDs.
+ return nil, syscall.EINVAL
+ }
+
+ // Default limit.
+ end := int32(math.MaxInt32)
+
+ // Ensure we don't get past the provided limit.
+ if limitSet := limits.FromContext(ctx); limitSet != nil {
+ lim := limitSet.Get(limits.NumberOfFiles)
+ if lim.Cur != limits.Infinity {
+ end = int32(lim.Cur)
+ }
+ if fd >= end {
+ return nil, syscall.EMFILE
+ }
+ }
+
+ f.mu.Lock()
+ defer f.mu.Unlock()
+
+ // From f.next to find available fd.
+ if fd < f.next {
+ fd = f.next
+ }
+
+ // Install all entries.
+ for i := fd; i < end && len(fds) < len(files); i++ {
+ if d, _, _ := f.getVFS2(i); d == nil {
+ f.setVFS2(i, files[len(fds)], flags) // Set the descriptor.
+ fds = append(fds, i) // Record the file descriptor.
+ }
+ }
+
+ // Failure? Unwind existing FDs.
+ if len(fds) < len(files) {
+ for _, i := range fds {
+ f.setVFS2(i, nil, FDFlags{}) // Zap entry.
+ }
+ return nil, syscall.EMFILE
+ }
+
+ if fd == f.next {
+ // Update next search start position.
+ f.next = fds[len(fds)-1] + 1
+ }
+
+ return fds, nil
+}
+
// NewFDVFS2 allocates a file descriptor greater than or equal to minfd for
// the given file description. If it succeeds, it takes a reference on file.
func (f *FDTable) NewFDVFS2(ctx context.Context, minfd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) {
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 0a448b57c..de8a95854 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -564,15 +564,25 @@ func (ts *TaskSet) unregisterEpollWaiters() {
ts.mu.RLock()
defer ts.mu.RUnlock()
+
+ // Tasks that belong to the same process could potentially point to the
+ // same FDTable. So we retain a map of processed ones to avoid
+ // processing the same FDTable multiple times.
+ processed := make(map[*FDTable]struct{})
for t := range ts.Root.tids {
// We can skip locking Task.mu here since the kernel is paused.
- if t.fdTable != nil {
- t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) {
- if e, ok := file.FileOperations.(*epoll.EventPoll); ok {
- e.UnregisterEpollWaiters()
- }
- })
+ if t.fdTable == nil {
+ continue
+ }
+ if _, ok := processed[t.fdTable]; ok {
+ continue
}
+ t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) {
+ if e, ok := file.FileOperations.(*epoll.EventPoll); ok {
+ e.UnregisterEpollWaiters()
+ }
+ })
+ processed[t.fdTable] = struct{}{}
}
}
@@ -1034,14 +1044,17 @@ 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 {
- // TODO(gvisor.dev/issue/1663): Add save support for VFS2.
- if !VFS2Enabled {
- t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) {
+ t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) {
+ if VFS2Enabled {
+ if tfd, ok := fd.Impl().(*vfs.TimerFileDescription); ok {
+ tfd.PauseTimer()
+ }
+ } else {
if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok {
tfd.PauseTimer()
}
- })
- }
+ }
+ })
}
}
k.timekeeper.PauseUpdates()
@@ -1066,15 +1079,18 @@ func (k *Kernel) resumeTimeLocked() {
it.ResumeTimer()
}
}
- // TODO(gvisor.dev/issue/1663): Add save support for VFS2.
- if !VFS2Enabled {
- if t.fdTable != nil {
- t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) {
+ if t.fdTable != nil {
+ t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) {
+ if VFS2Enabled {
+ if tfd, ok := fd.Impl().(*vfs.TimerFileDescription); ok {
+ tfd.ResumeTimer()
+ }
+ } else {
if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok {
tfd.ResumeTimer()
}
- })
- }
+ }
+ })
}
}
}
@@ -1435,9 +1451,10 @@ func (k *Kernel) SupervisorContext() context.Context {
// +stateify savable
type SocketEntry struct {
socketEntry
- k *Kernel
- Sock *refs.WeakRef
- ID uint64 // Socket table entry number.
+ k *Kernel
+ Sock *refs.WeakRef
+ SockVFS2 *vfs.FileDescription
+ ID uint64 // Socket table entry number.
}
// WeakRefGone implements refs.WeakRefUser.WeakRefGone.
@@ -1460,7 +1477,30 @@ func (k *Kernel) RecordSocket(sock *fs.File) {
k.extMu.Unlock()
}
+// RecordSocketVFS2 adds a VFS2 socket to the system-wide socket table for
+// tracking.
+//
+// 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.
+func (k *Kernel) RecordSocketVFS2(sock *vfs.FileDescription) {
+ k.extMu.Lock()
+ id := k.nextSocketEntry
+ k.nextSocketEntry++
+ s := &SocketEntry{
+ k: k,
+ ID: id,
+ SockVFS2: sock,
+ }
+ k.sockets.PushBack(s)
+ k.extMu.Unlock()
+}
+
// ListSockets returns a snapshot of all sockets.
+//
+// Callers of ListSockets() in VFS2 should use SocketEntry.SockVFS2.TryIncRef()
+// to get a reference on a socket in the table.
func (k *Kernel) ListSockets() []*SocketEntry {
k.extMu.Lock()
var socks []*SocketEntry
diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go
index 725e9db7d..62c8691f1 100644
--- a/pkg/sentry/kernel/pipe/pipe.go
+++ b/pkg/sentry/kernel/pipe/pipe.go
@@ -255,7 +255,8 @@ func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) {
// POSIX requires that a write smaller than atomicIOBytes (PIPE_BUF) be
// atomic, but requires no atomicity for writes larger than this.
wanted := ops.left()
- if avail := p.max - p.view.Size(); wanted > avail {
+ avail := p.max - p.view.Size()
+ if wanted > avail {
if wanted <= p.atomicIOBytes {
return 0, syserror.ErrWouldBlock
}
@@ -268,8 +269,14 @@ func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) {
return done, err
}
- if wanted > done {
- // Partial write due to full pipe.
+ if done < avail {
+ // Non-failure, but short write.
+ return done, nil
+ }
+ if done < wanted {
+ // Partial write due to full pipe. Note that this could also be
+ // the short write case above, we would expect a second call
+ // and the write to return zero bytes in this case.
return done, syserror.ErrWouldBlock
}
diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go
index 35ad97d5d..e23e796ef 100644
--- a/pkg/sentry/kernel/ptrace.go
+++ b/pkg/sentry/kernel/ptrace.go
@@ -184,7 +184,6 @@ func (t *Task) CanTrace(target *Task, attach bool) bool {
if targetCreds.PermittedCaps&^callerCreds.PermittedCaps != 0 {
return false
}
- // TODO: Yama LSM
return true
}
diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go
index 208569057..f66cfcc7f 100644
--- a/pkg/sentry/kernel/shm/shm.go
+++ b/pkg/sentry/kernel/shm/shm.go
@@ -461,7 +461,7 @@ func (s *Shm) AddMapping(ctx context.Context, _ memmap.MappingSpace, _ usermem.A
func (s *Shm) RemoveMapping(ctx context.Context, _ memmap.MappingSpace, _ usermem.AddrRange, _ uint64, _ bool) {
s.mu.Lock()
defer s.mu.Unlock()
- // TODO(b/38173783): RemoveMapping may be called during task exit, when ctx
+ // RemoveMapping may be called during task exit, when ctx
// is context.Background. Gracefully handle missing clocks. Failing to
// update the detach time in these cases is ok, since no one can observe the
// omission.
diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go
index 93c4fe969..2e3565747 100644
--- a/pkg/sentry/kernel/syscalls.go
+++ b/pkg/sentry/kernel/syscalls.go
@@ -209,65 +209,61 @@ type Stracer interface {
// SyscallEnter is called on syscall entry.
//
// The returned private data is passed to SyscallExit.
- //
- // TODO(gvisor.dev/issue/155): remove kernel imports from the strace
- // package so that the type can be used directly.
SyscallEnter(t *Task, sysno uintptr, args arch.SyscallArguments, flags uint32) interface{}
// SyscallExit is called on syscall exit.
SyscallExit(context interface{}, t *Task, sysno, rval uintptr, err error)
}
-// SyscallTable is a lookup table of system calls. Critically, a SyscallTable
-// is *immutable*. In order to make supporting suspend and resume sane, they
-// must be uniquely registered and may not change during operation.
+// SyscallTable is a lookup table of system calls.
//
-// +stateify savable
+// Note that a SyscallTable is not savable directly. Instead, they are saved as
+// an OS/Arch pair and lookup happens again on restore.
type SyscallTable struct {
// OS is the operating system that this syscall table implements.
- OS abi.OS `state:"wait"`
+ OS abi.OS
// Arch is the architecture that this syscall table targets.
- Arch arch.Arch `state:"wait"`
+ Arch arch.Arch
// The OS version that this syscall table implements.
- Version Version `state:"manual"`
+ Version Version
// AuditNumber is a numeric constant that represents the syscall table. If
// non-zero, auditNumber must be one of the AUDIT_ARCH_* values defined by
// linux/audit.h.
- AuditNumber uint32 `state:"manual"`
+ AuditNumber uint32
// Table is the collection of functions.
- Table map[uintptr]Syscall `state:"manual"`
+ Table map[uintptr]Syscall
// lookup is a fixed-size array that holds the syscalls (indexed by
// their numbers). It is used for fast look ups.
- lookup []SyscallFn `state:"manual"`
+ lookup []SyscallFn
// Emulate is a collection of instruction addresses to emulate. The
// keys are addresses, and the values are system call numbers.
- Emulate map[usermem.Addr]uintptr `state:"manual"`
+ Emulate map[usermem.Addr]uintptr
// The function to call in case of a missing system call.
- Missing MissingFn `state:"manual"`
+ Missing MissingFn
// Stracer traces this syscall table.
- Stracer Stracer `state:"manual"`
+ Stracer Stracer
// External is used to handle an external callback.
- External func(*Kernel) `state:"manual"`
+ External func(*Kernel)
// ExternalFilterBefore is called before External is called before the syscall is executed.
// External is not called if it returns false.
- ExternalFilterBefore func(*Task, uintptr, arch.SyscallArguments) bool `state:"manual"`
+ ExternalFilterBefore func(*Task, uintptr, arch.SyscallArguments) bool
// ExternalFilterAfter is called before External is called after the syscall is executed.
// External is not called if it returns false.
- ExternalFilterAfter func(*Task, uintptr, arch.SyscallArguments) bool `state:"manual"`
+ ExternalFilterAfter func(*Task, uintptr, arch.SyscallArguments) bool
// FeatureEnable stores the strace and one-shot enable bits.
- FeatureEnable SyscallFlagsTable `state:"manual"`
+ FeatureEnable SyscallFlagsTable
}
// allSyscallTables contains all known tables.
diff --git a/pkg/sentry/kernel/syscalls_state.go b/pkg/sentry/kernel/syscalls_state.go
index 00358326b..90f890495 100644
--- a/pkg/sentry/kernel/syscalls_state.go
+++ b/pkg/sentry/kernel/syscalls_state.go
@@ -14,16 +14,34 @@
package kernel
-import "fmt"
+import (
+ "fmt"
-// afterLoad is invoked by stateify.
-func (s *SyscallTable) afterLoad() {
- otherTable, ok := LookupSyscallTable(s.OS, s.Arch)
- if !ok {
- // Couldn't find a reference?
- panic(fmt.Sprintf("syscall table not found for OS %v Arch %v", s.OS, s.Arch))
+ "gvisor.dev/gvisor/pkg/abi"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+)
+
+// syscallTableInfo is used to reload the SyscallTable.
+//
+// +stateify savable
+type syscallTableInfo struct {
+ OS abi.OS
+ Arch arch.Arch
+}
+
+// saveSt saves the SyscallTable.
+func (tc *TaskContext) saveSt() syscallTableInfo {
+ return syscallTableInfo{
+ OS: tc.st.OS,
+ Arch: tc.st.Arch,
}
+}
- // Copy the table.
- *s = *otherTable
+// loadSt loads the SyscallTable.
+func (tc *TaskContext) loadSt(sti syscallTableInfo) {
+ st, ok := LookupSyscallTable(sti.OS, sti.Arch)
+ if !ok {
+ panic(fmt.Sprintf("syscall table not found for OS %v, Arch %v", sti.OS, sti.Arch))
+ }
+ tc.st = st // Save the table reference.
}
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index d6546735e..e5d133d6c 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -777,6 +777,15 @@ func (t *Task) NewFDs(fd int32, files []*fs.File, flags FDFlags) ([]int32, error
return t.fdTable.NewFDs(t, fd, files, flags)
}
+// NewFDsVFS2 is a convenience wrapper for t.FDTable().NewFDsVFS2.
+//
+// This automatically passes the task as the context.
+//
+// Precondition: same as FDTable.
+func (t *Task) NewFDsVFS2(fd int32, files []*vfs.FileDescription, flags FDFlags) ([]int32, error) {
+ return t.fdTable.NewFDsVFS2(t, fd, files, flags)
+}
+
// NewFDFrom is a convenience wrapper for t.FDTable().NewFDs with a single file.
//
// This automatically passes the task as the context.
diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go
index 0158b1788..9fa528384 100644
--- a/pkg/sentry/kernel/task_context.go
+++ b/pkg/sentry/kernel/task_context.go
@@ -49,7 +49,7 @@ type TaskContext struct {
fu *futex.Manager
// st is the task's syscall table.
- st *SyscallTable
+ st *SyscallTable `state:".(syscallTableInfo)"`
}
// release releases all resources held by the TaskContext. release is called by
@@ -58,7 +58,6 @@ func (tc *TaskContext) release() {
// Nil out pointers so that if the task is saved after release, it doesn't
// follow the pointers to possibly now-invalid objects.
if tc.MemoryManager != nil {
- // TODO(b/38173783)
tc.MemoryManager.DecUsers(context.Background())
tc.MemoryManager = nil
}
diff --git a/pkg/sentry/kernel/task_identity.go b/pkg/sentry/kernel/task_identity.go
index ce3e6ef28..0325967e4 100644
--- a/pkg/sentry/kernel/task_identity.go
+++ b/pkg/sentry/kernel/task_identity.go
@@ -455,7 +455,7 @@ func (t *Task) SetKeepCaps(k bool) {
t.creds.Store(creds)
}
-// updateCredsForExec updates t.creds to reflect an execve().
+// updateCredsForExecLocked updates t.creds to reflect an execve().
//
// NOTE(b/30815691): We currently do not implement privileged executables
// (set-user/group-ID bits and file capabilities). This allows us to make a lot
diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go
index 8802db142..6aa798346 100644
--- a/pkg/sentry/kernel/task_signals.go
+++ b/pkg/sentry/kernel/task_signals.go
@@ -513,8 +513,6 @@ func (t *Task) canReceiveSignalLocked(sig linux.Signal) bool {
if t.stop != nil {
return false
}
- // - TODO(b/38173783): No special case for when t is also the sending task,
- // because the identity of the sender is unknown.
// - Do not choose tasks that have already been interrupted, as they may be
// busy handling another signal.
if len(t.interruptChan) != 0 {
diff --git a/pkg/sentry/kernel/time/time.go b/pkg/sentry/kernel/time/time.go
index 706de83ef..e959700f2 100644
--- a/pkg/sentry/kernel/time/time.go
+++ b/pkg/sentry/kernel/time/time.go
@@ -245,7 +245,7 @@ type Clock interface {
type WallRateClock struct{}
// WallTimeUntil implements Clock.WallTimeUntil.
-func (WallRateClock) WallTimeUntil(t, now Time) time.Duration {
+func (*WallRateClock) WallTimeUntil(t, now Time) time.Duration {
return t.Sub(now)
}
@@ -254,16 +254,16 @@ func (WallRateClock) WallTimeUntil(t, now Time) time.Duration {
type NoClockEvents struct{}
// Readiness implements waiter.Waitable.Readiness.
-func (NoClockEvents) Readiness(mask waiter.EventMask) waiter.EventMask {
+func (*NoClockEvents) Readiness(mask waiter.EventMask) waiter.EventMask {
return 0
}
// EventRegister implements waiter.Waitable.EventRegister.
-func (NoClockEvents) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+func (*NoClockEvents) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
}
// EventUnregister implements waiter.Waitable.EventUnregister.
-func (NoClockEvents) EventUnregister(e *waiter.Entry) {
+func (*NoClockEvents) EventUnregister(e *waiter.Entry) {
}
// ClockEventsQueue implements waiter.Waitable by wrapping waiter.Queue and
@@ -273,7 +273,7 @@ type ClockEventsQueue struct {
}
// Readiness implements waiter.Waitable.Readiness.
-func (ClockEventsQueue) Readiness(mask waiter.EventMask) waiter.EventMask {
+func (*ClockEventsQueue) Readiness(mask waiter.EventMask) waiter.EventMask {
return 0
}
diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go
index 0332fc71c..5c667117c 100644
--- a/pkg/sentry/mm/address_space.go
+++ b/pkg/sentry/mm/address_space.go
@@ -201,8 +201,10 @@ func (mm *MemoryManager) mapASLocked(pseg pmaIterator, ar usermem.AddrRange, pre
if pma.needCOW {
perms.Write = false
}
- if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil {
- return err
+ if perms.Any() { // MapFile precondition
+ if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil {
+ return err
+ }
}
pseg = pseg.NextSegment()
}
diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go
index cb29d94b0..379148903 100644
--- a/pkg/sentry/mm/aio_context.go
+++ b/pkg/sentry/mm/aio_context.go
@@ -59,25 +59,27 @@ func (a *aioManager) newAIOContext(events uint32, id uint64) bool {
}
a.contexts[id] = &AIOContext{
- done: make(chan struct{}, 1),
+ requestReady: make(chan struct{}, 1),
maxOutstanding: events,
}
return true
}
-// destroyAIOContext destroys an asynchronous I/O context.
+// destroyAIOContext destroys an asynchronous I/O context. It doesn't wait for
+// for pending requests to complete. Returns the destroyed AIOContext so it can
+// be drained.
//
-// False is returned if the context does not exist.
-func (a *aioManager) destroyAIOContext(id uint64) bool {
+// Nil is returned if the context does not exist.
+func (a *aioManager) destroyAIOContext(id uint64) *AIOContext {
a.mu.Lock()
defer a.mu.Unlock()
ctx, ok := a.contexts[id]
if !ok {
- return false
+ return nil
}
delete(a.contexts, id)
ctx.destroy()
- return true
+ return ctx
}
// lookupAIOContext looks up the given context.
@@ -102,8 +104,8 @@ type ioResult struct {
//
// +stateify savable
type AIOContext struct {
- // done is the notification channel used for all requests.
- done chan struct{} `state:"nosave"`
+ // requestReady is the notification channel used for all requests.
+ requestReady chan struct{} `state:"nosave"`
// mu protects below.
mu sync.Mutex `state:"nosave"`
@@ -129,8 +131,14 @@ func (ctx *AIOContext) destroy() {
ctx.mu.Lock()
defer ctx.mu.Unlock()
ctx.dead = true
- if ctx.outstanding == 0 {
- close(ctx.done)
+ ctx.checkForDone()
+}
+
+// Preconditions: ctx.mu must be held by caller.
+func (ctx *AIOContext) checkForDone() {
+ if ctx.dead && ctx.outstanding == 0 {
+ close(ctx.requestReady)
+ ctx.requestReady = nil
}
}
@@ -154,11 +162,12 @@ func (ctx *AIOContext) PopRequest() (interface{}, bool) {
// Is there anything ready?
if e := ctx.results.Front(); e != nil {
- ctx.results.Remove(e)
- ctx.outstanding--
- if ctx.outstanding == 0 && ctx.dead {
- close(ctx.done)
+ if ctx.outstanding == 0 {
+ panic("AIOContext outstanding is going negative")
}
+ ctx.outstanding--
+ ctx.results.Remove(e)
+ ctx.checkForDone()
return e.data, true
}
return nil, false
@@ -172,26 +181,58 @@ func (ctx *AIOContext) FinishRequest(data interface{}) {
// Push to the list and notify opportunistically. The channel notify
// here is guaranteed to be safe because outstanding must be non-zero.
- // The done channel is only closed when outstanding reaches zero.
+ // The requestReady channel is only closed when outstanding reaches zero.
ctx.results.PushBack(&ioResult{data: data})
select {
- case ctx.done <- struct{}{}:
+ case ctx.requestReady <- struct{}{}:
default:
}
}
// WaitChannel returns a channel that is notified when an AIO request is
-// completed.
-//
-// The boolean return value indicates whether or not the context is active.
-func (ctx *AIOContext) WaitChannel() (chan struct{}, bool) {
+// completed. Returns nil if the context is destroyed and there are no more
+// outstanding requests.
+func (ctx *AIOContext) WaitChannel() chan struct{} {
ctx.mu.Lock()
defer ctx.mu.Unlock()
- if ctx.outstanding == 0 && ctx.dead {
- return nil, false
+ return ctx.requestReady
+}
+
+// Dead returns true if the context has been destroyed.
+func (ctx *AIOContext) Dead() bool {
+ ctx.mu.Lock()
+ defer ctx.mu.Unlock()
+ return ctx.dead
+}
+
+// CancelPendingRequest forgets about a request that hasn't yet completed.
+func (ctx *AIOContext) CancelPendingRequest() {
+ ctx.mu.Lock()
+ defer ctx.mu.Unlock()
+
+ if ctx.outstanding == 0 {
+ panic("AIOContext outstanding is going negative")
}
- return ctx.done, true
+ ctx.outstanding--
+ ctx.checkForDone()
+}
+
+// Drain drops all completed requests. Pending requests remain untouched.
+func (ctx *AIOContext) Drain() {
+ ctx.mu.Lock()
+ defer ctx.mu.Unlock()
+
+ if ctx.outstanding == 0 {
+ return
+ }
+ size := uint32(ctx.results.Len())
+ if ctx.outstanding < size {
+ panic("AIOContext outstanding is going negative")
+ }
+ ctx.outstanding -= size
+ ctx.results.Reset()
+ ctx.checkForDone()
}
// aioMappable implements memmap.MappingIdentity and memmap.Mappable for AIO
@@ -332,9 +373,9 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint
Length: aioRingBufferSize,
MappingIdentity: m,
Mappable: m,
- // TODO(fvoznika): Linux does "do_mmap_pgoff(..., PROT_READ |
- // PROT_WRITE, ...)" in fs/aio.c:aio_setup_ring(); why do we make this
- // mapping read-only?
+ // Linux uses "do_mmap_pgoff(..., PROT_READ | PROT_WRITE, ...)" in
+ // fs/aio.c:aio_setup_ring(). Since we don't implement AIO_RING_MAGIC,
+ // user mode should not write to this page.
Perms: usermem.Read,
MaxPerms: usermem.Read,
})
@@ -349,11 +390,11 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint
return id, nil
}
-// DestroyAIOContext destroys an asynchronous I/O context. It returns false if
-// the context does not exist.
-func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) bool {
+// DestroyAIOContext destroys an asynchronous I/O context. It returns the
+// destroyed context. nil if the context does not exist.
+func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) *AIOContext {
if _, ok := mm.LookupAIOContext(ctx, id); !ok {
- return false
+ return nil
}
// Only unmaps after it assured that the address is a valid aio context to
diff --git a/pkg/sentry/mm/aio_context_state.go b/pkg/sentry/mm/aio_context_state.go
index c37fc9f7b..3dabac1af 100644
--- a/pkg/sentry/mm/aio_context_state.go
+++ b/pkg/sentry/mm/aio_context_state.go
@@ -16,5 +16,5 @@ package mm
// afterLoad is invoked by stateify.
func (a *AIOContext) afterLoad() {
- a.done = make(chan struct{}, 1)
+ a.requestReady = make(chan struct{}, 1)
}
diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD
index cdcad0fdc..b69520030 100644
--- a/pkg/sentry/platform/ring0/BUILD
+++ b/pkg/sentry/platform/ring0/BUILD
@@ -71,8 +71,8 @@ go_library(
"lib_amd64.go",
"lib_amd64.s",
"lib_arm64.go",
- "lib_arm64_unsafe.go",
"lib_arm64.s",
+ "lib_arm64_unsafe.go",
"ring0.go",
],
visibility = ["//pkg/sentry:internal"],
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index f14c336b9..7ac38764d 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -300,7 +300,7 @@ type SocketOperations struct {
// New creates a new endpoint socket.
func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*fs.File, *syserr.Error) {
if skType == linux.SOCK_STREAM {
- if err := endpoint.SetSockOptInt(tcpip.DelayOption, 1); err != nil {
+ if err := endpoint.SetSockOptBool(tcpip.DelayOption, true); err != nil {
return nil, syserr.TranslateNetstackError(err)
}
}
@@ -535,7 +535,7 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO
}
if resCh != nil {
- t := ctx.(*kernel.Task)
+ t := kernel.TaskFromContext(ctx)
if err := t.Block(resCh); err != nil {
return 0, syserr.FromError(err).ToError()
}
@@ -608,7 +608,7 @@ func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader
}
if resCh != nil {
- t := ctx.(*kernel.Task)
+ t := kernel.TaskFromContext(ctx)
if err := t.Block(resCh); err != nil {
return 0, syserr.FromError(err).ToError()
}
@@ -663,7 +663,7 @@ func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error
// This is a hack to work around the fact that both IPv4 and IPv6 ANY are
// represented by the empty string.
//
-// TODO(gvisor.dev/issues/1556): remove this function.
+// TODO(gvisor.dev/issue/1556): remove this function.
func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress {
if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET {
addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00"
@@ -940,7 +940,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
// GetSockOpt can be used to implement the linux syscall getsockopt(2) for
// sockets backed by a commonEndpoint.
-func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) {
+func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) {
switch level {
case linux.SOL_SOCKET:
return getSockOptSocket(t, s, ep, family, skType, name, outLen)
@@ -965,8 +965,15 @@ func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int,
return nil, syserr.ErrProtocolNotAvailable
}
+func boolToInt32(v bool) int32 {
+ if v {
+ return 1
+ }
+ return 0
+}
+
// getSockOptSocket implements GetSockOpt when level is SOL_SOCKET.
-func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) {
+func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) {
// TODO(b/124056281): Stop rejecting short optLen values in getsockopt.
switch name {
case linux.SO_ERROR:
@@ -998,12 +1005,11 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.PasscredOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.PasscredOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.SO_SNDBUF:
if outLen < sizeOfInt32 {
@@ -1042,24 +1048,22 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.ReuseAddressOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.ReuseAddressOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.SO_REUSEPORT:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.ReusePortOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.ReusePortOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.SO_BINDTODEVICE:
var v tcpip.BindToDeviceOption
@@ -1089,24 +1093,22 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.BroadcastOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.BroadcastOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.SO_KEEPALIVE:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.KeepaliveEnabledOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.KeepaliveEnabledOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.SO_LINGER:
if outLen < linux.SizeOfLinger {
@@ -1156,47 +1158,41 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
return nil, syserr.ErrInvalidArgument
}
- v, err := ep.GetSockOptInt(tcpip.DelayOption)
+ v, err := ep.GetSockOptBool(tcpip.DelayOption)
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- if v == 0 {
- return int32(1), nil
- }
- return int32(0), nil
+ return boolToInt32(!v), nil
case linux.TCP_CORK:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.CorkOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.CorkOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.TCP_QUICKACK:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.QuickAckOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.QuickAckOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ return boolToInt32(v), nil
case linux.TCP_MAXSEG:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.MaxSegOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptInt(tcpip.MaxSegOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
@@ -1328,11 +1324,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- var o int32
- if v {
- o = 1
- }
- return o, nil
+ return boolToInt32(v), nil
case linux.IPV6_PATHMTU:
t.Kernel().EmitUnimplementedEvent(t)
@@ -1342,8 +1334,8 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
if outLen == 0 {
return make([]byte, 0), nil
}
- var v tcpip.IPv6TrafficClassOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptInt(tcpip.IPv6TrafficClassOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
@@ -1365,12 +1357,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- var o int32
- if v {
- o = 1
- }
- return o, nil
+ return boolToInt32(v), nil
default:
emitUnimplementedEventIPv6(t, name)
@@ -1386,8 +1373,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.TTLOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptInt(tcpip.TTLOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
@@ -1403,8 +1390,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.MulticastTTLOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptInt(tcpip.MulticastTTLOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
@@ -1429,23 +1416,19 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
return nil, syserr.ErrInvalidArgument
}
- var v tcpip.MulticastLoopOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptBool(tcpip.MulticastLoopOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- if v {
- return int32(1), nil
- }
- return int32(0), nil
+ return boolToInt32(v), nil
case linux.IP_TOS:
// Length handling for parity with Linux.
if outLen == 0 {
return []byte(nil), nil
}
- var v tcpip.IPv4TOSOption
- if err := ep.GetSockOpt(&v); err != nil {
+ v, err := ep.GetSockOptInt(tcpip.IPv4TOSOption)
+ if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
if outLen < sizeOfInt32 {
@@ -1462,11 +1445,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- var o int32
- if v {
- o = 1
- }
- return o, nil
+ return boolToInt32(v), nil
case linux.IP_PKTINFO:
if outLen < sizeOfInt32 {
@@ -1477,11 +1456,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- var o int32
- if v {
- o = 1
- }
- return o, nil
+ return boolToInt32(v), nil
default:
emitUnimplementedEventIP(t, name)
@@ -1541,7 +1516,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa
// SetSockOpt can be used to implement the linux syscall setsockopt(2) for
// sockets backed by a commonEndpoint.
-func SetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, level int, name int, optVal []byte) *syserr.Error {
+func SetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, level int, name int, optVal []byte) *syserr.Error {
switch level {
case linux.SOL_SOCKET:
return setSockOptSocket(t, s, ep, name, optVal)
@@ -1568,7 +1543,7 @@ func SetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, level int, n
}
// setSockOptSocket implements SetSockOpt when level is SOL_SOCKET.
-func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name int, optVal []byte) *syserr.Error {
+func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error {
switch name {
case linux.SO_SNDBUF:
if len(optVal) < sizeOfInt32 {
@@ -1592,7 +1567,7 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReuseAddressOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReuseAddressOption, v != 0))
case linux.SO_REUSEPORT:
if len(optVal) < sizeOfInt32 {
@@ -1600,7 +1575,7 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReusePortOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReusePortOption, v != 0))
case linux.SO_BINDTODEVICE:
n := bytes.IndexByte(optVal, 0)
@@ -1628,7 +1603,7 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BroadcastOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.BroadcastOption, v != 0))
case linux.SO_PASSCRED:
if len(optVal) < sizeOfInt32 {
@@ -1636,7 +1611,7 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.PasscredOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.PasscredOption, v != 0))
case linux.SO_KEEPALIVE:
if len(optVal) < sizeOfInt32 {
@@ -1644,7 +1619,7 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveEnabledOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.KeepaliveEnabledOption, v != 0))
case linux.SO_SNDTIMEO:
if len(optVal) < linux.SizeOfTimeval {
@@ -1716,11 +1691,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *
}
v := usermem.ByteOrder.Uint32(optVal)
- var o int
- if v == 0 {
- o = 1
- }
- return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.DelayOption, o))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.DelayOption, v == 0))
case linux.TCP_CORK:
if len(optVal) < sizeOfInt32 {
@@ -1728,7 +1699,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.CorkOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.CorkOption, v != 0))
case linux.TCP_QUICKACK:
if len(optVal) < sizeOfInt32 {
@@ -1736,7 +1707,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.QuickAckOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.QuickAckOption, v != 0))
case linux.TCP_MAXSEG:
if len(optVal) < sizeOfInt32 {
@@ -1744,7 +1715,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *
}
v := usermem.ByteOrder.Uint32(optVal)
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.MaxSegOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.MaxSegOption, int(v)))
case linux.TCP_KEEPIDLE:
if len(optVal) < sizeOfInt32 {
@@ -1855,7 +1826,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte)
if v == -1 {
v = 0
}
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv6TrafficClassOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.IPv6TrafficClassOption, int(v)))
case linux.IPV6_RECVTCLASS:
v, err := parseIntOrChar(optVal)
@@ -1940,7 +1911,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
if v < 0 || v > 255 {
return syserr.ErrInvalidArgument
}
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.MulticastTTLOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.MulticastTTLOption, int(v)))
case linux.IP_ADD_MEMBERSHIP:
req, err := copyInMulticastRequest(optVal, false /* allowAddr */)
@@ -1987,9 +1958,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
return err
}
- return syserr.TranslateNetstackError(ep.SetSockOpt(
- tcpip.MulticastLoopOption(v != 0),
- ))
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.MulticastLoopOption, v != 0))
case linux.MCAST_JOIN_GROUP:
// FIXME(b/124219304): Implement MCAST_JOIN_GROUP.
@@ -2008,7 +1977,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
} else if v < 1 || v > 255 {
return syserr.ErrInvalidArgument
}
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TTLOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.TTLOption, int(v)))
case linux.IP_TOS:
if len(optVal) == 0 {
@@ -2018,7 +1987,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
if err != nil {
return err
}
- return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v)))
+ return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.IPv4TOSOption, int(v)))
case linux.IP_RECVTOS:
v, err := parseIntOrChar(optVal)
diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go
index eb090e79b..c3f04b613 100644
--- a/pkg/sentry/socket/netstack/provider.go
+++ b/pkg/sentry/socket/netstack/provider.go
@@ -62,10 +62,6 @@ func getTransportProtocol(ctx context.Context, stype linux.SockType, protocol in
}
case linux.SOCK_RAW:
- // TODO(b/142504697): "In order to create a raw socket, a
- // process must have the CAP_NET_RAW capability in the user
- // namespace that governs its network namespace." - raw(7)
-
// Raw sockets require CAP_NET_RAW.
creds := auth.CredentialsFromContext(ctx)
if !creds.HasCapability(linux.CAP_NET_RAW) {
@@ -141,10 +137,6 @@ func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*
}
func packetSocket(t *kernel.Task, epStack *Stack, stype linux.SockType, protocol int) (*fs.File, *syserr.Error) {
- // TODO(b/142504697): "In order to create a packet socket, a process
- // must have the CAP_NET_RAW capability in the user namespace that
- // governs its network namespace." - packet(7)
-
// Packet sockets require CAP_NET_RAW.
creds := auth.CredentialsFromContext(t)
if !creds.HasCapability(linux.CAP_NET_RAW) {
diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go
index b5ba4a56b..6580bd6e9 100644
--- a/pkg/sentry/socket/socket.go
+++ b/pkg/sentry/socket/socket.go
@@ -269,7 +269,7 @@ func NewVFS2(t *kernel.Task, family int, stype linux.SockType, protocol int) (*v
return nil, err
}
if s != nil {
- // TODO: Add vfs2 sockets to global socket table.
+ t.Kernel().RecordSocketVFS2(s)
return s, nil
}
}
@@ -291,7 +291,9 @@ func PairVFS2(t *kernel.Task, family int, stype linux.SockType, protocol int) (*
return nil, nil, err
}
if s1 != nil && s2 != nil {
- // TODO: Add vfs2 sockets to global socket table.
+ k := t.Kernel()
+ k.RecordSocketVFS2(s1)
+ k.RecordSocketVFS2(s2)
return s1, s2, nil
}
}
diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD
index 08743deba..de2cc4bdf 100644
--- a/pkg/sentry/socket/unix/BUILD
+++ b/pkg/sentry/socket/unix/BUILD
@@ -8,23 +8,27 @@ go_library(
"device.go",
"io.go",
"unix.go",
+ "unix_vfs2.go",
],
visibility = ["//pkg/sentry:internal"],
deps = [
"//pkg/abi/linux",
"//pkg/context",
+ "//pkg/fspath",
"//pkg/refs",
"//pkg/safemem",
"//pkg/sentry/arch",
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/fs/fsutil",
+ "//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/time",
"//pkg/sentry/socket",
"//pkg/sentry/socket/control",
"//pkg/sentry/socket/netstack",
"//pkg/sentry/socket/unix/transport",
+ "//pkg/sentry/vfs",
"//pkg/syserr",
"//pkg/syserror",
"//pkg/tcpip",
diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD
index 74bcd6300..c708b6030 100644
--- a/pkg/sentry/socket/unix/transport/BUILD
+++ b/pkg/sentry/socket/unix/transport/BUILD
@@ -30,6 +30,7 @@ go_library(
"//pkg/abi/linux",
"//pkg/context",
"//pkg/ilist",
+ "//pkg/log",
"//pkg/refs",
"//pkg/sync",
"//pkg/syserr",
diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go
index 2ef654235..1f3880cc5 100644
--- a/pkg/sentry/socket/unix/transport/unix.go
+++ b/pkg/sentry/socket/unix/transport/unix.go
@@ -20,6 +20,7 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/tcpip"
@@ -838,24 +839,45 @@ func (e *baseEndpoint) SendMsg(ctx context.Context, data [][]byte, c ControlMess
// SetSockOpt sets a socket option. Currently not supported.
func (e *baseEndpoint) SetSockOpt(opt interface{}) *tcpip.Error {
- switch v := opt.(type) {
- case tcpip.PasscredOption:
- e.setPasscred(v != 0)
- return nil
- }
return nil
}
func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error {
+ switch opt {
+ case tcpip.BroadcastOption:
+ case tcpip.PasscredOption:
+ e.setPasscred(v)
+ case tcpip.ReuseAddressOption:
+ default:
+ log.Warningf("Unsupported socket option: %d", opt)
+ return tcpip.ErrUnknownProtocolOption
+ }
return nil
}
func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
+ switch opt {
+ case tcpip.SendBufferSizeOption:
+ case tcpip.ReceiveBufferSizeOption:
+ default:
+ log.Warningf("Unsupported socket option: %d", opt)
+ return tcpip.ErrUnknownProtocolOption
+ }
return nil
}
func (e *baseEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) {
- return false, tcpip.ErrUnknownProtocolOption
+ switch opt {
+ case tcpip.KeepaliveEnabledOption:
+ return false, nil
+
+ case tcpip.PasscredOption:
+ return e.Passcred(), nil
+
+ default:
+ log.Warningf("Unsupported socket option: %d", opt)
+ return false, tcpip.ErrUnknownProtocolOption
+ }
}
func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
@@ -914,29 +936,19 @@ func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) {
return int(v), nil
default:
+ log.Warningf("Unsupported socket option: %d", opt)
return -1, tcpip.ErrUnknownProtocolOption
}
}
// GetSockOpt implements tcpip.Endpoint.GetSockOpt.
func (e *baseEndpoint) GetSockOpt(opt interface{}) *tcpip.Error {
- switch o := opt.(type) {
+ switch opt.(type) {
case tcpip.ErrorOption:
return nil
- case *tcpip.PasscredOption:
- if e.Passcred() {
- *o = tcpip.PasscredOption(1)
- } else {
- *o = tcpip.PasscredOption(0)
- }
- return nil
-
- case *tcpip.KeepaliveEnabledOption:
- *o = 0
- return nil
-
default:
+ log.Warningf("Unsupported socket option: %T", opt)
return tcpip.ErrUnknownProtocolOption
}
}
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 4d30aa714..7c64f30fa 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -23,6 +23,7 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/fs"
@@ -33,6 +34,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/socket/control"
"gvisor.dev/gvisor/pkg/sentry/socket/netstack"
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip"
@@ -52,11 +54,8 @@ type SocketOperations struct {
fsutil.FileNoSplice `state:"nosave"`
fsutil.FileNoopFlush `state:"nosave"`
fsutil.FileUseInodeUnstableAttr `state:"nosave"`
- refs.AtomicRefCount
- socket.SendReceiveTimeout
- ep transport.Endpoint
- stype linux.SockType
+ socketOpsCommon
}
// New creates a new unix socket.
@@ -75,16 +74,29 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty
}
s := SocketOperations{
- ep: ep,
- stype: stype,
+ socketOpsCommon: socketOpsCommon{
+ ep: ep,
+ stype: stype,
+ },
}
s.EnableLeakCheck("unix.SocketOperations")
return fs.NewFile(ctx, d, flags, &s)
}
+// socketOpsCommon contains the socket operations common to VFS1 and VFS2.
+//
+// +stateify savable
+type socketOpsCommon struct {
+ refs.AtomicRefCount
+ socket.SendReceiveTimeout
+
+ ep transport.Endpoint
+ stype linux.SockType
+}
+
// DecRef implements RefCounter.DecRef.
-func (s *SocketOperations) DecRef() {
+func (s *socketOpsCommon) DecRef() {
s.DecRefWithDestructor(func() {
s.ep.Close()
})
@@ -97,7 +109,7 @@ func (s *SocketOperations) Release() {
s.DecRef()
}
-func (s *SocketOperations) isPacket() bool {
+func (s *socketOpsCommon) isPacket() bool {
switch s.stype {
case linux.SOCK_DGRAM, linux.SOCK_SEQPACKET:
return true
@@ -110,7 +122,7 @@ func (s *SocketOperations) isPacket() bool {
}
// Endpoint extracts the transport.Endpoint.
-func (s *SocketOperations) Endpoint() transport.Endpoint {
+func (s *socketOpsCommon) Endpoint() transport.Endpoint {
return s.ep
}
@@ -143,7 +155,7 @@ func extractPath(sockaddr []byte) (string, *syserr.Error) {
// GetPeerName implements the linux syscall getpeername(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
+func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
addr, err := s.ep.GetRemoteAddress()
if err != nil {
return nil, 0, syserr.TranslateNetstackError(err)
@@ -155,7 +167,7 @@ func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32,
// GetSockName implements the linux syscall getsockname(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
+func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) {
addr, err := s.ep.GetLocalAddress()
if err != nil {
return nil, 0, syserr.TranslateNetstackError(err)
@@ -178,7 +190,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
// Listen implements the linux syscall listen(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error {
+func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error {
return s.ep.Listen(backlog)
}
@@ -310,6 +322,8 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error {
}
// Create the socket.
+ //
+ // TODO(gvisor.dev/issue/2324): Correctly set file permissions.
childDir, err := d.Bind(t, t.FSContext().RootDirectory(), name, bep, fs.FilePermissions{User: fs.PermMask{Read: true}})
if err != nil {
return syserr.ErrPortInUse
@@ -345,6 +359,31 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint,
return ep, nil
}
+ if kernel.VFS2Enabled {
+ p := fspath.Parse(path)
+ root := t.FSContext().RootDirectoryVFS2()
+ start := root
+ relPath := !p.Absolute
+ if relPath {
+ start = t.FSContext().WorkingDirectoryVFS2()
+ }
+ pop := vfs.PathOperation{
+ Root: root,
+ Start: start,
+ Path: p,
+ FollowFinalSymlink: true,
+ }
+ ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop)
+ root.DecRef()
+ if relPath {
+ start.DecRef()
+ }
+ if e != nil {
+ return nil, syserr.FromError(e)
+ }
+ return ep, nil
+ }
+
// Find the node in the filesystem.
root := t.FSContext().RootDirectory()
cwd := t.FSContext().WorkingDirectory()
@@ -363,12 +402,11 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint,
// No socket!
return nil, syserr.ErrConnectionRefused
}
-
return ep, nil
}
// Connect implements the linux syscall connect(2) for unix sockets.
-func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error {
+func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error {
ep, err := extractEndpoint(t, sockaddr)
if err != nil {
return err
@@ -379,7 +417,7 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo
return s.ep.Connect(t, ep)
}
-// Writev implements fs.FileOperations.Write.
+// Write implements fs.FileOperations.Write.
func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) {
t := kernel.TaskFromContext(ctx)
ctrl := control.New(t, s.ep, nil)
@@ -399,7 +437,7 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO
// SendMsg implements the linux syscall sendmsg(2) for unix sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) {
+func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) {
w := EndpointWriter{
Ctx: t,
Endpoint: s.ep,
@@ -453,27 +491,27 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []
}
// Passcred implements transport.Credentialer.Passcred.
-func (s *SocketOperations) Passcred() bool {
+func (s *socketOpsCommon) Passcred() bool {
return s.ep.Passcred()
}
// ConnectedPasscred implements transport.Credentialer.ConnectedPasscred.
-func (s *SocketOperations) ConnectedPasscred() bool {
+func (s *socketOpsCommon) ConnectedPasscred() bool {
return s.ep.ConnectedPasscred()
}
// Readiness implements waiter.Waitable.Readiness.
-func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask {
+func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask {
return s.ep.Readiness(mask)
}
// EventRegister implements waiter.Waitable.EventRegister.
-func (s *SocketOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+func (s *socketOpsCommon) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
s.ep.EventRegister(e, mask)
}
// EventUnregister implements waiter.Waitable.EventUnregister.
-func (s *SocketOperations) EventUnregister(e *waiter.Entry) {
+func (s *socketOpsCommon) EventUnregister(e *waiter.Entry) {
s.ep.EventUnregister(e)
}
@@ -485,7 +523,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa
// Shutdown implements the linux syscall shutdown(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error {
+func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error {
f, err := netstack.ConvertShutdown(how)
if err != nil {
return err
@@ -511,7 +549,7 @@ func (s *SocketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS
// RecvMsg implements the linux syscall recvmsg(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) {
+func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) {
trunc := flags&linux.MSG_TRUNC != 0
peek := flags&linux.MSG_PEEK != 0
dontWait := flags&linux.MSG_DONTWAIT != 0
@@ -648,12 +686,12 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags
}
// State implements socket.Socket.State.
-func (s *SocketOperations) State() uint32 {
+func (s *socketOpsCommon) State() uint32 {
return s.ep.State()
}
// Type implements socket.Socket.Type.
-func (s *SocketOperations) Type() (family int, skType linux.SockType, protocol int) {
+func (s *socketOpsCommon) Type() (family int, skType linux.SockType, protocol int) {
// Unix domain sockets always have a protocol of 0.
return linux.AF_UNIX, s.stype, 0
}
@@ -706,4 +744,5 @@ func (*provider) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*fs.F
func init() {
socket.RegisterProvider(linux.AF_UNIX, &provider{})
+ socket.RegisterProviderVFS2(linux.AF_UNIX, &providerVFS2{})
}
diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go
new file mode 100644
index 000000000..3e54d49c4
--- /dev/null
+++ b/pkg/sentry/socket/unix/unix_vfs2.go
@@ -0,0 +1,348 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package unix
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/fspath"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/socket/control"
+ "gvisor.dev/gvisor/pkg/sentry/socket/netstack"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserr"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/tcpip"
+ "gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/pkg/waiter"
+)
+
+// SocketVFS2 implements socket.SocketVFS2 (and by extension,
+// vfs.FileDescriptionImpl) for Unix sockets.
+type SocketVFS2 struct {
+ vfsfd vfs.FileDescription
+ vfs.FileDescriptionDefaultImpl
+ vfs.DentryMetadataFileDescriptionImpl
+
+ socketOpsCommon
+}
+
+// NewVFS2File creates and returns a new vfs.FileDescription for a unix socket.
+func NewVFS2File(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) {
+ sock := NewFDImpl(ep, stype)
+ vfsfd := &sock.vfsfd
+ if err := sockfs.InitSocket(sock, vfsfd, t.Kernel().SocketMount(), t.Credentials()); err != nil {
+ return nil, syserr.FromError(err)
+ }
+ return vfsfd, nil
+}
+
+// NewFDImpl creates and returns a new SocketVFS2.
+func NewFDImpl(ep transport.Endpoint, stype linux.SockType) *SocketVFS2 {
+ // You can create AF_UNIX, SOCK_RAW sockets. They're the same as
+ // SOCK_DGRAM and don't require CAP_NET_RAW.
+ if stype == linux.SOCK_RAW {
+ stype = linux.SOCK_DGRAM
+ }
+
+ return &SocketVFS2{
+ socketOpsCommon: socketOpsCommon{
+ ep: ep,
+ stype: stype,
+ },
+ }
+}
+
+// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
+// a transport.Endpoint.
+func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+ return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen)
+}
+
+// blockingAccept implements a blocking version of accept(2), that is, if no
+// connections are ready to be accept, it will block until one becomes ready.
+func (s *SocketVFS2) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr.Error) {
+ // Register for notifications.
+ e, ch := waiter.NewChannelEntry(nil)
+ s.socketOpsCommon.EventRegister(&e, waiter.EventIn)
+ defer s.socketOpsCommon.EventUnregister(&e)
+
+ // Try to accept the connection; if it fails, then wait until we get a
+ // notification.
+ for {
+ if ep, err := s.ep.Accept(); err != syserr.ErrWouldBlock {
+ return ep, err
+ }
+
+ if err := t.Block(ch); err != nil {
+ return nil, syserr.FromError(err)
+ }
+ }
+}
+
+// Accept implements the linux syscall accept(2) for sockets backed by
+// a transport.Endpoint.
+func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) {
+ // Issue the accept request to get the new endpoint.
+ ep, err := s.ep.Accept()
+ if err != nil {
+ if err != syserr.ErrWouldBlock || !blocking {
+ return 0, nil, 0, err
+ }
+
+ var err *syserr.Error
+ ep, err = s.blockingAccept(t)
+ if err != nil {
+ return 0, nil, 0, err
+ }
+ }
+
+ // We expect this to be a FileDescription here.
+ ns, err := NewVFS2File(t, ep, s.stype)
+ if err != nil {
+ return 0, nil, 0, err
+ }
+ defer ns.DecRef()
+
+ if flags&linux.SOCK_NONBLOCK != 0 {
+ ns.SetStatusFlags(t, t.Credentials(), linux.SOCK_NONBLOCK)
+ }
+
+ var addr linux.SockAddr
+ var addrLen uint32
+ if peerRequested {
+ // Get address of the peer.
+ var err *syserr.Error
+ addr, addrLen, err = ns.Impl().(*SocketVFS2).GetPeerName(t)
+ if err != nil {
+ return 0, nil, 0, err
+ }
+ }
+
+ fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{
+ CloseOnExec: flags&linux.SOCK_CLOEXEC != 0,
+ })
+ if e != nil {
+ return 0, nil, 0, syserr.FromError(e)
+ }
+
+ t.Kernel().RecordSocketVFS2(ns)
+ return fd, addr, addrLen, nil
+}
+
+// Bind implements the linux syscall bind(2) for unix sockets.
+func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error {
+ p, e := extractPath(sockaddr)
+ if e != nil {
+ return e
+ }
+
+ bep, ok := s.ep.(transport.BoundEndpoint)
+ if !ok {
+ // This socket can't be bound.
+ return syserr.ErrInvalidArgument
+ }
+
+ return s.ep.Bind(tcpip.FullAddress{Addr: tcpip.Address(p)}, func() *syserr.Error {
+ // Is it abstract?
+ if p[0] == 0 {
+ if t.IsNetworkNamespaced() {
+ return syserr.ErrInvalidEndpointState
+ }
+ if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil {
+ // syserr.ErrPortInUse corresponds to EADDRINUSE.
+ return syserr.ErrPortInUse
+ }
+ } else {
+ path := fspath.Parse(p)
+ root := t.FSContext().RootDirectoryVFS2()
+ defer root.DecRef()
+ start := root
+ relPath := !path.Absolute
+ if relPath {
+ start = t.FSContext().WorkingDirectoryVFS2()
+ defer start.DecRef()
+ }
+ pop := vfs.PathOperation{
+ Root: root,
+ Start: start,
+ Path: path,
+ }
+ err := t.Kernel().VFS().MknodAt(t, t.Credentials(), &pop, &vfs.MknodOptions{
+ // TODO(gvisor.dev/issue/2324): The file permissions should be taken
+ // from s and t.FSContext().Umask() (see net/unix/af_unix.c:unix_bind),
+ // but VFS1 just always uses 0400. Resolve this inconsistency.
+ Mode: linux.S_IFSOCK | 0400,
+ Endpoint: bep,
+ })
+ if err == syserror.EEXIST {
+ return syserr.ErrAddressInUse
+ }
+ return syserr.FromError(err)
+ }
+
+ return nil
+ })
+}
+
+// Ioctl implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+ return netstack.Ioctl(ctx, s.ep, uio, args)
+}
+
+// PRead implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
+ return 0, syserror.ESPIPE
+}
+
+// Read implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+ // All flags other than RWF_NOWAIT should be ignored.
+ // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT.
+ if opts.Flags != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
+ if dst.NumBytes() == 0 {
+ return 0, nil
+ }
+ return dst.CopyOutFrom(ctx, &EndpointReader{
+ Ctx: ctx,
+ Endpoint: s.ep,
+ NumRights: 0,
+ Peek: false,
+ From: nil,
+ })
+}
+
+// PWrite implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+ return 0, syserror.ESPIPE
+}
+
+// Write implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
+ // All flags other than RWF_NOWAIT should be ignored.
+ // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT.
+ if opts.Flags != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
+ t := kernel.TaskFromContext(ctx)
+ ctrl := control.New(t, s.ep, nil)
+
+ if src.NumBytes() == 0 {
+ nInt, err := s.ep.SendMsg(ctx, [][]byte{}, ctrl, nil)
+ return int64(nInt), err.ToError()
+ }
+
+ return src.CopyInTo(ctx, &EndpointWriter{
+ Ctx: ctx,
+ Endpoint: s.ep,
+ Control: ctrl,
+ To: nil,
+ })
+}
+
+// Release implements vfs.FileDescriptionImpl.
+func (s *SocketVFS2) Release() {
+ // Release only decrements a reference on s because s may be referenced in
+ // the abstract socket namespace.
+ s.DecRef()
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask {
+ return s.socketOpsCommon.Readiness(mask)
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (s *SocketVFS2) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+ s.socketOpsCommon.EventRegister(e, mask)
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (s *SocketVFS2) EventUnregister(e *waiter.Entry) {
+ s.socketOpsCommon.EventUnregister(e)
+}
+
+// SetSockOpt implements the linux syscall setsockopt(2) for sockets backed by
+// a transport.Endpoint.
+func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []byte) *syserr.Error {
+ return netstack.SetSockOpt(t, s, s.ep, level, name, optVal)
+}
+
+// providerVFS2 is a unix domain socket provider for VFS2.
+type providerVFS2 struct{}
+
+func (*providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) {
+ // Check arguments.
+ if protocol != 0 && protocol != linux.AF_UNIX /* PF_UNIX */ {
+ return nil, syserr.ErrProtocolNotSupported
+ }
+
+ // Create the endpoint and socket.
+ var ep transport.Endpoint
+ switch stype {
+ case linux.SOCK_DGRAM, linux.SOCK_RAW:
+ ep = transport.NewConnectionless(t)
+ case linux.SOCK_SEQPACKET, linux.SOCK_STREAM:
+ ep = transport.NewConnectioned(t, stype, t.Kernel())
+ default:
+ return nil, syserr.ErrInvalidArgument
+ }
+
+ f, err := NewVFS2File(t, ep, stype)
+ if err != nil {
+ ep.Close()
+ return nil, err
+ }
+ return f, nil
+}
+
+// Pair creates a new pair of AF_UNIX connected sockets.
+func (*providerVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) {
+ // Check arguments.
+ if protocol != 0 && protocol != linux.AF_UNIX /* PF_UNIX */ {
+ return nil, nil, syserr.ErrProtocolNotSupported
+ }
+
+ switch stype {
+ case linux.SOCK_STREAM, linux.SOCK_DGRAM, linux.SOCK_SEQPACKET, linux.SOCK_RAW:
+ // Ok
+ default:
+ return nil, nil, syserr.ErrInvalidArgument
+ }
+
+ // Create the endpoints and sockets.
+ ep1, ep2 := transport.NewPair(t, stype, t.Kernel())
+ s1, err := NewVFS2File(t, ep1, stype)
+ if err != nil {
+ ep1.Close()
+ ep2.Close()
+ return nil, nil, err
+ }
+ s2, err := NewVFS2File(t, ep2, stype)
+ if err != nil {
+ s1.DecRef()
+ ep2.Close()
+ return nil, nil, err
+ }
+
+ return s1, s2, nil
+}
diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go
index 77655558e..b94c4fbf5 100644
--- a/pkg/sentry/strace/strace.go
+++ b/pkg/sentry/strace/strace.go
@@ -778,9 +778,6 @@ func (s SyscallMap) Name(sysno uintptr) string {
//
// N.B. This is not in an init function because we can't be sure all syscall
// tables are registered with the kernel when init runs.
-//
-// TODO(gvisor.dev/issue/155): remove kernel package dependencies from this
-// package and have the kernel package self-initialize all syscall tables.
func Initialize() {
for _, table := range kernel.SyscallTables() {
// Is this known?
diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go
index b401978db..38cbeba5a 100644
--- a/pkg/sentry/syscalls/linux/sys_aio.go
+++ b/pkg/sentry/syscalls/linux/sys_aio.go
@@ -114,14 +114,28 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca
func IoDestroy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
id := args[0].Uint64()
- // Destroy the given context.
- if !t.MemoryManager().DestroyAIOContext(t, id) {
+ ctx := t.MemoryManager().DestroyAIOContext(t, id)
+ if ctx == nil {
// Does not exist.
return 0, nil, syserror.EINVAL
}
- // FIXME(fvoznika): Linux blocks until all AIO to the destroyed context is
- // done.
- return 0, nil, nil
+
+ // Drain completed requests amd wait for pending requests until there are no
+ // more.
+ for {
+ ctx.Drain()
+
+ ch := ctx.WaitChannel()
+ if ch == nil {
+ // No more requests, we're done.
+ return 0, nil, nil
+ }
+ // The task cannot be interrupted during the wait. Equivalent to
+ // TASK_UNINTERRUPTIBLE in Linux.
+ t.UninterruptibleSleepStart(true /* deactivate */)
+ <-ch
+ t.UninterruptibleSleepFinish(true /* activate */)
+ }
}
// IoGetevents implements linux syscall io_getevents(2).
@@ -200,13 +214,13 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S
func waitForRequest(ctx *mm.AIOContext, t *kernel.Task, haveDeadline bool, deadline ktime.Time) (interface{}, error) {
for {
if v, ok := ctx.PopRequest(); ok {
- // Request was readly available. Just return it.
+ // Request was readily available. Just return it.
return v, nil
}
// Need to wait for request completion.
- done, active := ctx.WaitChannel()
- if !active {
+ done := ctx.WaitChannel()
+ if done == nil {
// Context has been destroyed.
return nil, syserror.EINVAL
}
@@ -248,6 +262,10 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) {
}
func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioCallback, ioseq usermem.IOSequence, ctx *mm.AIOContext, eventFile *fs.File) {
+ if ctx.Dead() {
+ ctx.CancelPendingRequest()
+ return
+ }
ev := &ioEvent{
Data: cb.Data,
Obj: uint64(cbAddr),
diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go
index 9c6728530..f92bf8096 100644
--- a/pkg/sentry/syscalls/linux/sys_prctl.go
+++ b/pkg/sentry/syscalls/linux/sys_prctl.go
@@ -161,8 +161,8 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
if args[1].Int() != 1 || args[2].Int() != 0 || args[3].Int() != 0 || args[4].Int() != 0 {
return 0, nil, syserror.EINVAL
}
- // no_new_privs is assumed to always be set. See
- // kernel.Task.updateCredsForExec.
+ // PR_SET_NO_NEW_PRIVS is assumed to always be set.
+ // See kernel.Task.updateCredsForExecLocked.
return 0, nil, nil
case linux.PR_GET_NO_NEW_PRIVS:
diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go
index 2919228d0..61b2576ac 100644
--- a/pkg/sentry/syscalls/linux/sys_socket.go
+++ b/pkg/sentry/syscalls/linux/sys_socket.go
@@ -31,6 +31,8 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
+// LINT.IfChange
+
// minListenBacklog is the minimum reasonable backlog for listening sockets.
const minListenBacklog = 8
@@ -244,7 +246,10 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
// Copy the file descriptors out.
if _, err := t.CopyOut(socks, fds); err != nil {
- // Note that we don't close files here; see pipe(2) also.
+ for _, fd := range fds {
+ _, file := t.FDTable().Remove(fd)
+ file.DecRef()
+ }
return 0, nil, err
}
@@ -1128,3 +1133,5 @@ func SendTo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal
n, err := sendTo(t, fd, bufPtr, bufLen, flags, namePtr, nameLen)
return n, nil, err
}
+
+// LINT.ThenChange(./vfs2/socket.go)
diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go
index fd642834b..fbc6cf15f 100644
--- a/pkg/sentry/syscalls/linux/sys_splice.go
+++ b/pkg/sentry/syscalls/linux/sys_splice.go
@@ -29,6 +29,10 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB
return 0, syserror.EINVAL
}
+ if opts.Length > int64(kernel.MAX_RW_COUNT) {
+ opts.Length = int64(kernel.MAX_RW_COUNT)
+ }
+
var (
total int64
n int64
diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD
index 2eb210014..b32abfe59 100644
--- a/pkg/sentry/syscalls/linux/vfs2/BUILD
+++ b/pkg/sentry/syscalls/linux/vfs2/BUILD
@@ -21,16 +21,19 @@ go_library(
"poll.go",
"read_write.go",
"setstat.go",
+ "socket.go",
"stat.go",
"stat_amd64.go",
"stat_arm64.go",
"sync.go",
+ "sys_timerfd.go",
"xattr.go",
],
marshal = True,
visibility = ["//:sandbox"],
deps = [
"//pkg/abi/linux",
+ "//pkg/binary",
"//pkg/bits",
"//pkg/fspath",
"//pkg/gohacks",
@@ -42,10 +45,14 @@ go_library(
"//pkg/sentry/limits",
"//pkg/sentry/loader",
"//pkg/sentry/memmap",
+ "//pkg/sentry/socket",
+ "//pkg/sentry/socket/control",
+ "//pkg/sentry/socket/unix/transport",
"//pkg/sentry/syscalls",
"//pkg/sentry/syscalls/linux",
"//pkg/sentry/vfs",
"//pkg/sync",
+ "//pkg/syserr",
"//pkg/syserror",
"//pkg/usermem",
"//pkg/waiter",
diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go
index a61cc5059..62e98817d 100644
--- a/pkg/sentry/syscalls/linux/vfs2/getdents.go
+++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go
@@ -97,6 +97,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error {
// char d_name[]; /* Filename (null-terminated) */
// };
size := 8 + 8 + 2 + 1 + 1 + len(dirent.Name)
+ size = (size + 7) &^ 7 // round up to multiple of 8
if size > cb.remaining {
return syserror.EINVAL
}
@@ -106,7 +107,12 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error {
usermem.ByteOrder.PutUint16(buf[16:18], uint16(size))
buf[18] = dirent.Type
copy(buf[19:], dirent.Name)
- buf[size-1] = 0 // NUL terminator
+ // Zero out all remaining bytes in buf, including the NUL terminator
+ // after dirent.Name.
+ bufTail := buf[19+len(dirent.Name):]
+ for i := range bufTail {
+ bufTail[i] = 0
+ }
} else {
// struct linux_dirent {
// unsigned long d_ino; /* Inode number */
@@ -125,6 +131,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error {
panic(fmt.Sprintf("unsupported sizeof(unsigned long): %d", cb.t.Arch().Width()))
}
size := 8 + 8 + 2 + 1 + 1 + 1 + len(dirent.Name)
+ size = (size + 7) &^ 7 // round up to multiple of sizeof(long)
if size > cb.remaining {
return syserror.EINVAL
}
@@ -133,9 +140,14 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error {
usermem.ByteOrder.PutUint64(buf[8:16], uint64(dirent.NextOff))
usermem.ByteOrder.PutUint16(buf[16:18], uint16(size))
copy(buf[18:], dirent.Name)
- buf[size-3] = 0 // NUL terminator
- buf[size-2] = 0 // zero padding byte
- buf[size-1] = dirent.Type
+ // Zero out all remaining bytes in buf, including the NUL terminator
+ // after dirent.Name and the zero padding byte between the name and
+ // dirent type.
+ bufTail := buf[18+len(dirent.Name):]
+ for i := range bufTail {
+ bufTail[i] = 0
+ }
+ bufTail[2] = dirent.Type
}
n, err := cb.t.CopyOutBytes(cb.addr, buf)
if err != nil {
diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go
index 7d220bc20..645e0bcb8 100644
--- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go
+++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go
@@ -44,21 +44,22 @@ func Override(table map[uintptr]kernel.Syscall) {
table[32] = syscalls.Supported("dup", Dup)
table[33] = syscalls.Supported("dup2", Dup2)
delete(table, 40) // sendfile
- delete(table, 41) // socket
- delete(table, 42) // connect
- delete(table, 43) // accept
- delete(table, 44) // sendto
- delete(table, 45) // recvfrom
- delete(table, 46) // sendmsg
- delete(table, 47) // recvmsg
- delete(table, 48) // shutdown
- delete(table, 49) // bind
- delete(table, 50) // listen
- delete(table, 51) // getsockname
- delete(table, 52) // getpeername
- delete(table, 53) // socketpair
- delete(table, 54) // setsockopt
- delete(table, 55) // getsockopt
+ // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2.
+ table[41] = syscalls.PartiallySupported("socket", Socket, "In process of porting socket syscalls to VFS2.", nil)
+ table[42] = syscalls.PartiallySupported("connect", Connect, "In process of porting socket syscalls to VFS2.", nil)
+ table[43] = syscalls.PartiallySupported("accept", Accept, "In process of porting socket syscalls to VFS2.", nil)
+ table[44] = syscalls.PartiallySupported("sendto", SendTo, "In process of porting socket syscalls to VFS2.", nil)
+ table[45] = syscalls.PartiallySupported("recvfrom", RecvFrom, "In process of porting socket syscalls to VFS2.", nil)
+ table[46] = syscalls.PartiallySupported("sendmsg", SendMsg, "In process of porting socket syscalls to VFS2.", nil)
+ table[47] = syscalls.PartiallySupported("recvmsg", RecvMsg, "In process of porting socket syscalls to VFS2.", nil)
+ table[48] = syscalls.PartiallySupported("shutdown", Shutdown, "In process of porting socket syscalls to VFS2.", nil)
+ table[49] = syscalls.PartiallySupported("bind", Bind, "In process of porting socket syscalls to VFS2.", nil)
+ table[50] = syscalls.PartiallySupported("listen", Listen, "In process of porting socket syscalls to VFS2.", nil)
+ table[51] = syscalls.PartiallySupported("getsockname", GetSockName, "In process of porting socket syscalls to VFS2.", nil)
+ table[52] = syscalls.PartiallySupported("getpeername", GetPeerName, "In process of porting socket syscalls to VFS2.", nil)
+ table[53] = syscalls.PartiallySupported("socketpair", SocketPair, "In process of porting socket syscalls to VFS2.", nil)
+ table[54] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil)
+ table[55] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil)
table[59] = syscalls.Supported("execve", Execve)
table[72] = syscalls.Supported("fcntl", Fcntl)
delete(table, 73) // flock
@@ -139,12 +140,13 @@ func Override(table map[uintptr]kernel.Syscall) {
table[280] = syscalls.Supported("utimensat", Utimensat)
table[281] = syscalls.Supported("epoll_pwait", EpollPwait)
delete(table, 282) // signalfd
- delete(table, 283) // timerfd_create
+ table[283] = syscalls.Supported("timerfd_create", TimerfdCreate)
delete(table, 284) // eventfd
delete(table, 285) // fallocate
- delete(table, 286) // timerfd_settime
- delete(table, 287) // timerfd_gettime
- delete(table, 288) // accept4
+ table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime)
+ table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime)
+ // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2.
+ table[288] = syscalls.PartiallySupported("accept4", Accept4, "In process of porting socket syscalls to VFS2.", nil)
delete(table, 289) // signalfd4
delete(table, 290) // eventfd2
table[291] = syscalls.Supported("epoll_create1", EpollCreate1)
@@ -153,9 +155,11 @@ func Override(table map[uintptr]kernel.Syscall) {
delete(table, 294) // inotify_init1
table[295] = syscalls.Supported("preadv", Preadv)
table[296] = syscalls.Supported("pwritev", Pwritev)
- delete(table, 299) // recvmmsg
+ // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2.
+ table[299] = syscalls.PartiallySupported("recvmmsg", RecvMMsg, "In process of porting socket syscalls to VFS2.", nil)
table[306] = syscalls.Supported("syncfs", Syncfs)
- delete(table, 307) // sendmmsg
+ // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2.
+ table[307] = syscalls.PartiallySupported("sendmmsg", SendMMsg, "In process of porting socket syscalls to VFS2.", nil)
table[316] = syscalls.Supported("renameat2", Renameat2)
delete(table, 319) // memfd_create
table[322] = syscalls.Supported("execveat", Execveat)
diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go
new file mode 100644
index 000000000..79a4a7ada
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/socket.go
@@ -0,0 +1,1138 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs2
+
+import (
+ "time"
+
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/binary"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+ "gvisor.dev/gvisor/pkg/sentry/socket"
+ "gvisor.dev/gvisor/pkg/sentry/socket/control"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
+ slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserr"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
+)
+
+// minListenBacklog is the minimum reasonable backlog for listening sockets.
+const minListenBacklog = 8
+
+// maxListenBacklog is the maximum allowed backlog for listening sockets.
+const maxListenBacklog = 1024
+
+// maxAddrLen is the maximum socket address length we're willing to accept.
+const maxAddrLen = 200
+
+// maxOptLen is the maximum sockopt parameter length we're willing to accept.
+const maxOptLen = 1024 * 8
+
+// maxControlLen is the maximum length of the msghdr.msg_control buffer we're
+// willing to accept. Note that this limit is smaller than Linux, which allows
+// buffers upto INT_MAX.
+const maxControlLen = 10 * 1024 * 1024
+
+// nameLenOffset is the offset from the start of the MessageHeader64 struct to
+// the NameLen field.
+const nameLenOffset = 8
+
+// controlLenOffset is the offset form the start of the MessageHeader64 struct
+// to the ControlLen field.
+const controlLenOffset = 40
+
+// flagsOffset is the offset form the start of the MessageHeader64 struct
+// to the Flags field.
+const flagsOffset = 48
+
+const sizeOfInt32 = 4
+
+// messageHeader64Len is the length of a MessageHeader64 struct.
+var messageHeader64Len = uint64(binary.Size(MessageHeader64{}))
+
+// multipleMessageHeader64Len is the length of a multipeMessageHeader64 struct.
+var multipleMessageHeader64Len = uint64(binary.Size(multipleMessageHeader64{}))
+
+// baseRecvFlags are the flags that are accepted across recvmsg(2),
+// recvmmsg(2), and recvfrom(2).
+const baseRecvFlags = linux.MSG_OOB | linux.MSG_DONTROUTE | linux.MSG_DONTWAIT | linux.MSG_NOSIGNAL | linux.MSG_WAITALL | linux.MSG_TRUNC | linux.MSG_CTRUNC
+
+// MessageHeader64 is the 64-bit representation of the msghdr struct used in
+// the recvmsg and sendmsg syscalls.
+type MessageHeader64 struct {
+ // Name is the optional pointer to a network address buffer.
+ Name uint64
+
+ // NameLen is the length of the buffer pointed to by Name.
+ NameLen uint32
+ _ uint32
+
+ // Iov is a pointer to an array of io vectors that describe the memory
+ // locations involved in the io operation.
+ Iov uint64
+
+ // IovLen is the length of the array pointed to by Iov.
+ IovLen uint64
+
+ // Control is the optional pointer to ancillary control data.
+ Control uint64
+
+ // ControlLen is the length of the data pointed to by Control.
+ ControlLen uint64
+
+ // Flags on the sent/received message.
+ Flags int32
+ _ int32
+}
+
+// multipleMessageHeader64 is the 64-bit representation of the mmsghdr struct used in
+// the recvmmsg and sendmmsg syscalls.
+type multipleMessageHeader64 struct {
+ msgHdr MessageHeader64
+ msgLen uint32
+ _ int32
+}
+
+// CopyInMessageHeader64 copies a message header from user to kernel memory.
+func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader64) error {
+ b := t.CopyScratchBuffer(52)
+ if _, err := t.CopyInBytes(addr, b); err != nil {
+ return err
+ }
+
+ msg.Name = usermem.ByteOrder.Uint64(b[0:])
+ msg.NameLen = usermem.ByteOrder.Uint32(b[8:])
+ msg.Iov = usermem.ByteOrder.Uint64(b[16:])
+ msg.IovLen = usermem.ByteOrder.Uint64(b[24:])
+ msg.Control = usermem.ByteOrder.Uint64(b[32:])
+ msg.ControlLen = usermem.ByteOrder.Uint64(b[40:])
+ msg.Flags = int32(usermem.ByteOrder.Uint32(b[48:]))
+
+ return nil
+}
+
+// CaptureAddress allocates memory for and copies a socket address structure
+// from the untrusted address space range.
+func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) {
+ if addrlen > maxAddrLen {
+ return nil, syserror.EINVAL
+ }
+
+ addrBuf := make([]byte, addrlen)
+ if _, err := t.CopyInBytes(addr, addrBuf); err != nil {
+ return nil, err
+ }
+
+ return addrBuf, nil
+}
+
+// writeAddress writes a sockaddr structure and its length to an output buffer
+// in the unstrusted address space range. If the address is bigger than the
+// buffer, it is truncated.
+func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error {
+ // Get the buffer length.
+ var bufLen uint32
+ if _, err := t.CopyIn(addrLenPtr, &bufLen); err != nil {
+ return err
+ }
+
+ if int32(bufLen) < 0 {
+ return syserror.EINVAL
+ }
+
+ // Write the length unconditionally.
+ if _, err := t.CopyOut(addrLenPtr, addrLen); err != nil {
+ return err
+ }
+
+ if addr == nil {
+ return nil
+ }
+
+ if bufLen > addrLen {
+ bufLen = addrLen
+ }
+
+ // Copy as much of the address as will fit in the buffer.
+ encodedAddr := binary.Marshal(nil, usermem.ByteOrder, addr)
+ if bufLen > uint32(len(encodedAddr)) {
+ bufLen = uint32(len(encodedAddr))
+ }
+ _, err := t.CopyOutBytes(addrPtr, encodedAddr[:int(bufLen)])
+ return err
+}
+
+// Socket implements the linux syscall socket(2).
+func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ domain := int(args[0].Int())
+ stype := args[1].Int()
+ protocol := int(args[2].Int())
+
+ // Check and initialize the flags.
+ if stype & ^(0xf|linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Create the new socket.
+ s, e := socket.NewVFS2(t, domain, linux.SockType(stype&0xf), protocol)
+ if e != nil {
+ return 0, nil, e.ToError()
+ }
+ defer s.DecRef()
+
+ if err := s.SetStatusFlags(t, t.Credentials(), uint32(stype&linux.SOCK_NONBLOCK)); err != nil {
+ return 0, nil, err
+ }
+
+ fd, err := t.NewFDFromVFS2(0, s, kernel.FDFlags{
+ CloseOnExec: stype&linux.SOCK_CLOEXEC != 0,
+ })
+ if err != nil {
+ return 0, nil, err
+ }
+
+ return uintptr(fd), nil, nil
+}
+
+// SocketPair implements the linux syscall socketpair(2).
+func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ domain := int(args[0].Int())
+ stype := args[1].Int()
+ protocol := int(args[2].Int())
+ addr := args[3].Pointer()
+
+ // Check and initialize the flags.
+ if stype & ^(0xf|linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Create the socket pair.
+ s1, s2, e := socket.PairVFS2(t, domain, linux.SockType(stype&0xf), protocol)
+ if e != nil {
+ return 0, nil, e.ToError()
+ }
+ // Adding to the FD table will cause an extra reference to be acquired.
+ defer s1.DecRef()
+ defer s2.DecRef()
+
+ nonblocking := uint32(stype & linux.SOCK_NONBLOCK)
+ if err := s1.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil {
+ return 0, nil, err
+ }
+ if err := s2.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil {
+ return 0, nil, err
+ }
+
+ // Create the FDs for the sockets.
+ flags := kernel.FDFlags{
+ CloseOnExec: stype&linux.SOCK_CLOEXEC != 0,
+ }
+ fds, err := t.NewFDsVFS2(0, []*vfs.FileDescription{s1, s2}, flags)
+ if err != nil {
+ return 0, nil, err
+ }
+
+ if _, err := t.CopyOut(addr, fds); err != nil {
+ for _, fd := range fds {
+ _, file := t.FDTable().Remove(fd)
+ file.DecRef()
+ }
+ return 0, nil, err
+ }
+
+ return 0, nil, nil
+}
+
+// Connect implements the linux syscall connect(2).
+func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ addrlen := args[2].Uint()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Capture address and call syscall implementation.
+ a, err := CaptureAddress(t, addr, addrlen)
+ if err != nil {
+ return 0, nil, err
+ }
+
+ blocking := (file.StatusFlags() & linux.SOCK_NONBLOCK) == 0
+ return 0, nil, syserror.ConvertIntr(s.Connect(t, a, blocking).ToError(), kernel.ERESTARTSYS)
+}
+
+// accept is the implementation of the accept syscall. It is called by accept
+// and accept4 syscall handlers.
+func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, flags int) (uintptr, error) {
+ // Check that no unsupported flags are passed in.
+ if flags & ^(linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 {
+ return 0, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, syserror.ENOTSOCK
+ }
+
+ // Call the syscall implementation for this socket, then copy the
+ // output address if one is specified.
+ blocking := (file.StatusFlags() & linux.SOCK_NONBLOCK) == 0
+
+ peerRequested := addrLen != 0
+ nfd, peer, peerLen, e := s.Accept(t, peerRequested, flags, blocking)
+ if e != nil {
+ return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS)
+ }
+ if peerRequested {
+ // NOTE(magi): Linux does not give you an error if it can't
+ // write the data back out so neither do we.
+ if err := writeAddress(t, peer, peerLen, addr, addrLen); err == syserror.EINVAL {
+ return 0, err
+ }
+ }
+ return uintptr(nfd), nil
+}
+
+// Accept4 implements the linux syscall accept4(2).
+func Accept4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ addrlen := args[2].Pointer()
+ flags := int(args[3].Int())
+
+ n, err := accept(t, fd, addr, addrlen, flags)
+ return n, nil, err
+}
+
+// Accept implements the linux syscall accept(2).
+func Accept(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ addrlen := args[2].Pointer()
+
+ n, err := accept(t, fd, addr, addrlen, 0)
+ return n, nil, err
+}
+
+// Bind implements the linux syscall bind(2).
+func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ addrlen := args[2].Uint()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Capture address and call syscall implementation.
+ a, err := CaptureAddress(t, addr, addrlen)
+ if err != nil {
+ return 0, nil, err
+ }
+
+ return 0, nil, s.Bind(t, a).ToError()
+}
+
+// Listen implements the linux syscall listen(2).
+func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ backlog := args[1].Int()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Per Linux, the backlog is silently capped to reasonable values.
+ if backlog <= 0 {
+ backlog = minListenBacklog
+ }
+ if backlog > maxListenBacklog {
+ backlog = maxListenBacklog
+ }
+
+ return 0, nil, s.Listen(t, int(backlog)).ToError()
+}
+
+// Shutdown implements the linux syscall shutdown(2).
+func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ how := args[1].Int()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Validate how, then call syscall implementation.
+ switch how {
+ case linux.SHUT_RD, linux.SHUT_WR, linux.SHUT_RDWR:
+ default:
+ return 0, nil, syserror.EINVAL
+ }
+
+ return 0, nil, s.Shutdown(t, int(how)).ToError()
+}
+
+// GetSockOpt implements the linux syscall getsockopt(2).
+func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ level := args[1].Int()
+ name := args[2].Int()
+ optValAddr := args[3].Pointer()
+ optLenAddr := args[4].Pointer()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Read the length. Reject negative values.
+ optLen := int32(0)
+ if _, err := t.CopyIn(optLenAddr, &optLen); err != nil {
+ return 0, nil, err
+ }
+ if optLen < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Call syscall implementation then copy both value and value len out.
+ v, e := getSockOpt(t, s, int(level), int(name), optValAddr, int(optLen))
+ if e != nil {
+ return 0, nil, e.ToError()
+ }
+
+ vLen := int32(binary.Size(v))
+ if _, err := t.CopyOut(optLenAddr, vLen); err != nil {
+ return 0, nil, err
+ }
+
+ if v != nil {
+ if _, err := t.CopyOut(optValAddr, v); err != nil {
+ return 0, nil, err
+ }
+ }
+
+ return 0, nil, nil
+}
+
+// getSockOpt tries to handle common socket options, or dispatches to a specific
+// socket implementation.
+func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) {
+ if level == linux.SOL_SOCKET {
+ switch name {
+ case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL:
+ if len < sizeOfInt32 {
+ return nil, syserr.ErrInvalidArgument
+ }
+ }
+
+ switch name {
+ case linux.SO_TYPE:
+ _, skType, _ := s.Type()
+ return int32(skType), nil
+ case linux.SO_DOMAIN:
+ family, _, _ := s.Type()
+ return int32(family), nil
+ case linux.SO_PROTOCOL:
+ _, _, protocol := s.Type()
+ return int32(protocol), nil
+ }
+ }
+
+ return s.GetSockOpt(t, level, name, optValAddr, len)
+}
+
+// SetSockOpt implements the linux syscall setsockopt(2).
+//
+// Note that unlike Linux, enabling SO_PASSCRED does not autobind the socket.
+func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ level := args[1].Int()
+ name := args[2].Int()
+ optValAddr := args[3].Pointer()
+ optLen := args[4].Int()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ if optLen < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+ if optLen > maxOptLen {
+ return 0, nil, syserror.EINVAL
+ }
+ buf := t.CopyScratchBuffer(int(optLen))
+ if _, err := t.CopyIn(optValAddr, &buf); err != nil {
+ return 0, nil, err
+ }
+
+ // Call syscall implementation.
+ if err := s.SetSockOpt(t, int(level), int(name), buf); err != nil {
+ return 0, nil, err.ToError()
+ }
+
+ return 0, nil, nil
+}
+
+// GetSockName implements the linux syscall getsockname(2).
+func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ addrlen := args[2].Pointer()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Get the socket name and copy it to the caller.
+ v, vl, err := s.GetSockName(t)
+ if err != nil {
+ return 0, nil, err.ToError()
+ }
+
+ return 0, nil, writeAddress(t, v, vl, addr, addrlen)
+}
+
+// GetPeerName implements the linux syscall getpeername(2).
+func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ addrlen := args[2].Pointer()
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Get the socket peer name and copy it to the caller.
+ v, vl, err := s.GetPeerName(t)
+ if err != nil {
+ return 0, nil, err.ToError()
+ }
+
+ return 0, nil, writeAddress(t, v, vl, addr, addrlen)
+}
+
+// RecvMsg implements the linux syscall recvmsg(2).
+func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ msgPtr := args[1].Pointer()
+ flags := args[2].Int()
+
+ if t.Arch().Width() != 8 {
+ // We only handle 64-bit for now.
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Reject flags that we don't handle yet.
+ if flags & ^(baseRecvFlags|linux.MSG_PEEK|linux.MSG_CMSG_CLOEXEC|linux.MSG_ERRQUEUE) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ var haveDeadline bool
+ var deadline ktime.Time
+ if dl := s.RecvTimeout(); dl > 0 {
+ deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond)
+ haveDeadline = true
+ } else if dl < 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ n, err := recvSingleMsg(t, s, msgPtr, flags, haveDeadline, deadline)
+ return n, nil, err
+}
+
+// RecvMMsg implements the linux syscall recvmmsg(2).
+func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ msgPtr := args[1].Pointer()
+ vlen := args[2].Uint()
+ flags := args[3].Int()
+ toPtr := args[4].Pointer()
+
+ if t.Arch().Width() != 8 {
+ // We only handle 64-bit for now.
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Reject flags that we don't handle yet.
+ if flags & ^(baseRecvFlags|linux.MSG_CMSG_CLOEXEC|linux.MSG_ERRQUEUE) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ var haveDeadline bool
+ var deadline ktime.Time
+ if toPtr != 0 {
+ var ts linux.Timespec
+ if _, err := ts.CopyIn(t, toPtr); err != nil {
+ return 0, nil, err
+ }
+ if !ts.Valid() {
+ return 0, nil, syserror.EINVAL
+ }
+ deadline = t.Kernel().MonotonicClock().Now().Add(ts.ToDuration())
+ haveDeadline = true
+ }
+
+ if !haveDeadline {
+ if dl := s.RecvTimeout(); dl > 0 {
+ deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond)
+ haveDeadline = true
+ } else if dl < 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+ }
+
+ var count uint32
+ var err error
+ for i := uint64(0); i < uint64(vlen); i++ {
+ mp, ok := msgPtr.AddLength(i * multipleMessageHeader64Len)
+ if !ok {
+ return 0, nil, syserror.EFAULT
+ }
+ var n uintptr
+ if n, err = recvSingleMsg(t, s, mp, flags, haveDeadline, deadline); err != nil {
+ break
+ }
+
+ // Copy the received length to the caller.
+ lp, ok := mp.AddLength(messageHeader64Len)
+ if !ok {
+ return 0, nil, syserror.EFAULT
+ }
+ if _, err = t.CopyOut(lp, uint32(n)); err != nil {
+ break
+ }
+ count++
+ }
+
+ if count == 0 {
+ return 0, nil, err
+ }
+ return uintptr(count), nil, nil
+}
+
+func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, flags int32, haveDeadline bool, deadline ktime.Time) (uintptr, error) {
+ // Capture the message header and io vectors.
+ var msg MessageHeader64
+ if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil {
+ return 0, err
+ }
+
+ if msg.IovLen > linux.UIO_MAXIOV {
+ return 0, syserror.EMSGSIZE
+ }
+ dst, err := t.IovecsIOSequence(usermem.Addr(msg.Iov), int(msg.IovLen), usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ if err != nil {
+ return 0, err
+ }
+
+ // FIXME(b/63594852): Pretend we have an empty error queue.
+ if flags&linux.MSG_ERRQUEUE != 0 {
+ return 0, syserror.EAGAIN
+ }
+
+ // Fast path when no control message nor name buffers are provided.
+ if msg.ControlLen == 0 && msg.NameLen == 0 {
+ n, mflags, _, _, cms, err := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, false, 0)
+ if err != nil {
+ return 0, syserror.ConvertIntr(err.ToError(), kernel.ERESTARTSYS)
+ }
+ if !cms.Unix.Empty() {
+ mflags |= linux.MSG_CTRUNC
+ cms.Release()
+ }
+
+ if int(msg.Flags) != mflags {
+ // Copy out the flags to the caller.
+ if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil {
+ return 0, err
+ }
+ }
+
+ return uintptr(n), nil
+ }
+
+ if msg.ControlLen > maxControlLen {
+ return 0, syserror.ENOBUFS
+ }
+ n, mflags, sender, senderLen, cms, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, msg.NameLen != 0, msg.ControlLen)
+ if e != nil {
+ return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS)
+ }
+ defer cms.Release()
+
+ controlData := make([]byte, 0, msg.ControlLen)
+ controlData = control.PackControlMessages(t, cms, controlData)
+
+ if cr, ok := s.(transport.Credentialer); ok && cr.Passcred() {
+ creds, _ := cms.Unix.Credentials.(control.SCMCredentials)
+ controlData, mflags = control.PackCredentials(t, creds, controlData, mflags)
+ }
+
+ if cms.Unix.Rights != nil {
+ controlData, mflags = control.PackRights(t, cms.Unix.Rights.(control.SCMRights), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags)
+ }
+
+ // Copy the address to the caller.
+ if msg.NameLen != 0 {
+ if err := writeAddress(t, sender, senderLen, usermem.Addr(msg.Name), usermem.Addr(msgPtr+nameLenOffset)); err != nil {
+ return 0, err
+ }
+ }
+
+ // Copy the control data to the caller.
+ if _, err := t.CopyOut(msgPtr+controlLenOffset, uint64(len(controlData))); err != nil {
+ return 0, err
+ }
+ if len(controlData) > 0 {
+ if _, err := t.CopyOut(usermem.Addr(msg.Control), controlData); err != nil {
+ return 0, err
+ }
+ }
+
+ // Copy out the flags to the caller.
+ if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil {
+ return 0, err
+ }
+
+ return uintptr(n), nil
+}
+
+// recvFrom is the implementation of the recvfrom syscall. It is called by
+// recvfrom and recv syscall handlers.
+func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLenPtr usermem.Addr) (uintptr, error) {
+ if int(bufLen) < 0 {
+ return 0, syserror.EINVAL
+ }
+
+ // Reject flags that we don't handle yet.
+ if flags & ^(baseRecvFlags|linux.MSG_PEEK|linux.MSG_CONFIRM) != 0 {
+ return 0, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, syserror.ENOTSOCK
+ }
+
+ if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ dst, err := t.SingleIOSequence(bufPtr, int(bufLen), usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ if err != nil {
+ return 0, err
+ }
+
+ var haveDeadline bool
+ var deadline ktime.Time
+ if dl := s.RecvTimeout(); dl > 0 {
+ deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond)
+ haveDeadline = true
+ } else if dl < 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0)
+ cm.Release()
+ if e != nil {
+ return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS)
+ }
+
+ // Copy the address to the caller.
+ if nameLenPtr != 0 {
+ if err := writeAddress(t, sender, senderLen, namePtr, nameLenPtr); err != nil {
+ return 0, err
+ }
+ }
+
+ return uintptr(n), nil
+}
+
+// RecvFrom implements the linux syscall recvfrom(2).
+func RecvFrom(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ bufPtr := args[1].Pointer()
+ bufLen := args[2].Uint64()
+ flags := args[3].Int()
+ namePtr := args[4].Pointer()
+ nameLenPtr := args[5].Pointer()
+
+ n, err := recvFrom(t, fd, bufPtr, bufLen, flags, namePtr, nameLenPtr)
+ return n, nil, err
+}
+
+// SendMsg implements the linux syscall sendmsg(2).
+func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ msgPtr := args[1].Pointer()
+ flags := args[2].Int()
+
+ if t.Arch().Width() != 8 {
+ // We only handle 64-bit for now.
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Reject flags that we don't handle yet.
+ if flags & ^(linux.MSG_DONTWAIT|linux.MSG_EOR|linux.MSG_MORE|linux.MSG_NOSIGNAL) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ n, err := sendSingleMsg(t, s, file, msgPtr, flags)
+ return n, nil, err
+}
+
+// SendMMsg implements the linux syscall sendmmsg(2).
+func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ msgPtr := args[1].Pointer()
+ vlen := args[2].Uint()
+ flags := args[3].Int()
+
+ if t.Arch().Width() != 8 {
+ // We only handle 64-bit for now.
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, nil, syserror.ENOTSOCK
+ }
+
+ // Reject flags that we don't handle yet.
+ if flags & ^(linux.MSG_DONTWAIT|linux.MSG_EOR|linux.MSG_MORE|linux.MSG_NOSIGNAL) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ var count uint32
+ var err error
+ for i := uint64(0); i < uint64(vlen); i++ {
+ mp, ok := msgPtr.AddLength(i * multipleMessageHeader64Len)
+ if !ok {
+ return 0, nil, syserror.EFAULT
+ }
+ var n uintptr
+ if n, err = sendSingleMsg(t, s, file, mp, flags); err != nil {
+ break
+ }
+
+ // Copy the received length to the caller.
+ lp, ok := mp.AddLength(messageHeader64Len)
+ if !ok {
+ return 0, nil, syserror.EFAULT
+ }
+ if _, err = t.CopyOut(lp, uint32(n)); err != nil {
+ break
+ }
+ count++
+ }
+
+ if count == 0 {
+ return 0, nil, err
+ }
+ return uintptr(count), nil, nil
+}
+
+func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescription, msgPtr usermem.Addr, flags int32) (uintptr, error) {
+ // Capture the message header.
+ var msg MessageHeader64
+ if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil {
+ return 0, err
+ }
+
+ var controlData []byte
+ if msg.ControlLen > 0 {
+ // Put an upper bound to prevent large allocations.
+ if msg.ControlLen > maxControlLen {
+ return 0, syserror.ENOBUFS
+ }
+ controlData = make([]byte, msg.ControlLen)
+ if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil {
+ return 0, err
+ }
+ }
+
+ // Read the destination address if one is specified.
+ var to []byte
+ if msg.NameLen != 0 {
+ var err error
+ to, err = CaptureAddress(t, usermem.Addr(msg.Name), msg.NameLen)
+ if err != nil {
+ return 0, err
+ }
+ }
+
+ // Read data then call the sendmsg implementation.
+ if msg.IovLen > linux.UIO_MAXIOV {
+ return 0, syserror.EMSGSIZE
+ }
+ src, err := t.IovecsIOSequence(usermem.Addr(msg.Iov), int(msg.IovLen), usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ if err != nil {
+ return 0, err
+ }
+
+ controlMessages, err := control.Parse(t, s, controlData)
+ if err != nil {
+ return 0, err
+ }
+
+ var haveDeadline bool
+ var deadline ktime.Time
+ if dl := s.SendTimeout(); dl > 0 {
+ deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond)
+ haveDeadline = true
+ } else if dl < 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ // Call the syscall implementation.
+ n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages)
+ err = slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file)
+ if err != nil {
+ controlMessages.Release()
+ }
+ return uintptr(n), err
+}
+
+// sendTo is the implementation of the sendto syscall. It is called by sendto
+// and send syscall handlers.
+func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLen uint32) (uintptr, error) {
+ bl := int(bufLen)
+ if bl < 0 {
+ return 0, syserror.EINVAL
+ }
+
+ // Get socket from the file descriptor.
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Extract the socket.
+ s, ok := file.Impl().(socket.SocketVFS2)
+ if !ok {
+ return 0, syserror.ENOTSOCK
+ }
+
+ if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ // Read the destination address if one is specified.
+ var to []byte
+ var err error
+ if namePtr != 0 {
+ to, err = CaptureAddress(t, namePtr, nameLen)
+ if err != nil {
+ return 0, err
+ }
+ }
+
+ src, err := t.SingleIOSequence(bufPtr, bl, usermem.IOOpts{
+ AddressSpaceActive: true,
+ })
+ if err != nil {
+ return 0, err
+ }
+
+ var haveDeadline bool
+ var deadline ktime.Time
+ if dl := s.SendTimeout(); dl > 0 {
+ deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond)
+ haveDeadline = true
+ } else if dl < 0 {
+ flags |= linux.MSG_DONTWAIT
+ }
+
+ // Call the syscall implementation.
+ n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, socket.ControlMessages{Unix: control.New(t, s, nil)})
+ return uintptr(n), slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendto", file)
+}
+
+// SendTo implements the linux syscall sendto(2).
+func SendTo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ bufPtr := args[1].Pointer()
+ bufLen := args[2].Uint64()
+ flags := args[3].Int()
+ namePtr := args[4].Pointer()
+ nameLen := args[5].Uint()
+
+ n, err := sendTo(t, fd, bufPtr, bufLen, flags, namePtr, nameLen)
+ return n, nil, err
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go b/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go
new file mode 100644
index 000000000..7938a5249
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go
@@ -0,0 +1,123 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs2
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+// TimerfdCreate implements Linux syscall timerfd_create(2).
+func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ clockID := args[0].Int()
+ flags := args[1].Int()
+
+ if flags&^(linux.TFD_CLOEXEC|linux.TFD_NONBLOCK) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ var fileFlags uint32
+ if flags&linux.TFD_NONBLOCK != 0 {
+ fileFlags = linux.O_NONBLOCK
+ }
+
+ var clock ktime.Clock
+ switch clockID {
+ case linux.CLOCK_REALTIME:
+ clock = t.Kernel().RealtimeClock()
+ case linux.CLOCK_MONOTONIC, linux.CLOCK_BOOTTIME:
+ clock = t.Kernel().MonotonicClock()
+ default:
+ return 0, nil, syserror.EINVAL
+ }
+ file, err := t.Kernel().VFS().NewTimerFD(clock, fileFlags)
+ if err != nil {
+ return 0, nil, err
+ }
+ defer file.DecRef()
+ fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{
+ CloseOnExec: flags&linux.TFD_CLOEXEC != 0,
+ })
+ if err != nil {
+ return 0, nil, err
+ }
+ return uintptr(fd), nil, nil
+}
+
+// TimerfdSettime implements Linux syscall timerfd_settime(2).
+func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ flags := args[1].Int()
+ newValAddr := args[2].Pointer()
+ oldValAddr := args[3].Pointer()
+
+ if flags&^(linux.TFD_TIMER_ABSTIME) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ tfd, ok := file.Impl().(*vfs.TimerFileDescription)
+ if !ok {
+ return 0, nil, syserror.EINVAL
+ }
+
+ var newVal linux.Itimerspec
+ if _, err := t.CopyIn(newValAddr, &newVal); err != nil {
+ return 0, nil, err
+ }
+ newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tfd.Clock())
+ if err != nil {
+ return 0, nil, err
+ }
+ tm, oldS := tfd.SetTime(newS)
+ if oldValAddr != 0 {
+ oldVal := ktime.ItimerspecFromSetting(tm, oldS)
+ if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil {
+ return 0, nil, err
+ }
+ }
+ return 0, nil, nil
+}
+
+// TimerfdGettime implements Linux syscall timerfd_gettime(2).
+func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ curValAddr := args[1].Pointer()
+
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ tfd, ok := file.Impl().(*vfs.TimerFileDescription)
+ if !ok {
+ return 0, nil, syserror.EINVAL
+ }
+
+ tm, s := tfd.GetTime()
+ curVal := ktime.ItimerspecFromSetting(tm, s)
+ _, err := t.CopyOut(curValAddr, &curVal)
+ return 0, nil, err
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go
index 89e9ff4d7..af455d5c1 100644
--- a/pkg/sentry/syscalls/linux/vfs2/xattr.go
+++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go
@@ -51,7 +51,7 @@ func listxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSyml
}
defer tpop.Release()
- names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop)
+ names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop, uint64(size))
if err != nil {
return 0, nil, err
}
@@ -74,7 +74,7 @@ func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
}
defer file.DecRef()
- names, err := file.Listxattr(t)
+ names, err := file.Listxattr(t, uint64(size))
if err != nil {
return 0, nil, err
}
@@ -116,7 +116,10 @@ func getxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli
return 0, nil, err
}
- value, err := t.Kernel().VFS().GetxattrAt(t, t.Credentials(), &tpop.pop, name)
+ value, err := t.Kernel().VFS().GetxattrAt(t, t.Credentials(), &tpop.pop, &vfs.GetxattrOptions{
+ Name: name,
+ Size: uint64(size),
+ })
if err != nil {
return 0, nil, err
}
@@ -145,7 +148,7 @@ func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
return 0, nil, err
}
- value, err := file.Getxattr(t, name)
+ value, err := file.Getxattr(t, &vfs.GetxattrOptions{Name: name, Size: uint64(size)})
if err != nil {
return 0, nil, err
}
@@ -230,7 +233,7 @@ func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
return 0, nil, err
}
- return 0, nil, file.Setxattr(t, vfs.SetxattrOptions{
+ return 0, nil, file.Setxattr(t, &vfs.SetxattrOptions{
Name: name,
Value: value,
Flags: uint32(flags),
diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD
index bf4d27c7d..9aeb83fb0 100644
--- a/pkg/sentry/vfs/BUILD
+++ b/pkg/sentry/vfs/BUILD
@@ -36,6 +36,7 @@ go_library(
"pathname.go",
"permissions.go",
"resolving_path.go",
+ "timerfd.go",
"vfs.go",
],
visibility = ["//pkg/sentry:internal"],
@@ -51,6 +52,7 @@ go_library(
"//pkg/sentry/fs",
"//pkg/sentry/fs/lock",
"//pkg/sentry/kernel/auth",
+ "//pkg/sentry/kernel/time",
"//pkg/sentry/limits",
"//pkg/sentry/memmap",
"//pkg/sentry/socket/unix/transport",
diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go
index d1f6dfb45..a64d86122 100644
--- a/pkg/sentry/vfs/anonfs.go
+++ b/pkg/sentry/vfs/anonfs.go
@@ -245,7 +245,7 @@ func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath
}
// ListxattrAt implements FilesystemImpl.ListxattrAt.
-func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) {
+func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath, size uint64) ([]string, error) {
if !rp.Done() {
return nil, syserror.ENOTDIR
}
@@ -253,7 +253,7 @@ func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([
}
// GetxattrAt implements FilesystemImpl.GetxattrAt.
-func (fs *anonFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) {
+func (fs *anonFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, opts GetxattrOptions) (string, error) {
if !rp.Done() {
return "", syserror.ENOTDIR
}
diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go
index 3da45d744..8e0b40841 100644
--- a/pkg/sentry/vfs/epoll.go
+++ b/pkg/sentry/vfs/epoll.go
@@ -99,6 +99,8 @@ func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) {
interest: make(map[epollInterestKey]*epollInterest),
}
if err := ep.vfsfd.Init(ep, linux.O_RDWR, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{
+ DenyPRead: true,
+ DenyPWrite: true,
UseDentryMetadata: true,
}); err != nil {
return nil, err
diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go
index 8ee549dc2..4fb9aea87 100644
--- a/pkg/sentry/vfs/file_description.go
+++ b/pkg/sentry/vfs/file_description.go
@@ -84,6 +84,17 @@ type FileDescriptionOptions struct {
// usually only the case if O_DIRECT would actually have an effect.
AllowDirectIO bool
+ // If DenyPRead is true, calls to FileDescription.PRead() return ESPIPE.
+ DenyPRead bool
+
+ // If DenyPWrite is true, calls to FileDescription.PWrite() return
+ // ESPIPE.
+ DenyPWrite bool
+
+ // if InvalidWrite is true, calls to FileDescription.Write() return
+ // EINVAL.
+ InvalidWrite bool
+
// If UseDentryMetadata is true, calls to FileDescription methods that
// interact with file and filesystem metadata (Stat, SetStat, StatFS,
// Listxattr, Getxattr, Setxattr, Removexattr) are implemented by calling
@@ -175,6 +186,12 @@ func (fd *FileDescription) DecRef() {
}
}
+// Refs returns the current number of references. The returned count
+// is inherently racy and is unsafe to use without external synchronization.
+func (fd *FileDescription) Refs() int64 {
+ return atomic.LoadInt64(&fd.refs)
+}
+
// Mount returns the mount on which fd was opened. It does not take a reference
// on the returned Mount.
func (fd *FileDescription) Mount() *Mount {
@@ -306,6 +323,7 @@ type FileDescriptionImpl interface {
// - If opts.Flags specifies unsupported options, PRead returns EOPNOTSUPP.
//
// Preconditions: The FileDescription was opened for reading.
+ // FileDescriptionOptions.DenyPRead == false.
PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error)
// Read is similar to PRead, but does not specify an offset.
@@ -337,6 +355,7 @@ type FileDescriptionImpl interface {
// EOPNOTSUPP.
//
// Preconditions: The FileDescription was opened for writing.
+ // FileDescriptionOptions.DenyPWrite == false.
PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error)
// Write is similar to PWrite, but does not specify an offset, which is
@@ -382,11 +401,11 @@ type FileDescriptionImpl interface {
Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error)
// Listxattr returns all extended attribute names for the file.
- Listxattr(ctx context.Context) ([]string, error)
+ Listxattr(ctx context.Context, size uint64) ([]string, error)
// Getxattr returns the value associated with the given extended attribute
// for the file.
- Getxattr(ctx context.Context, name string) (string, error)
+ Getxattr(ctx context.Context, opts GetxattrOptions) (string, error)
// Setxattr changes the value associated with the given extended attribute
// for the file.
@@ -515,6 +534,9 @@ func (fd *FileDescription) EventUnregister(e *waiter.Entry) {
// offset, and returns the number of bytes read. PRead is permitted to return
// partial reads with a nil error.
func (fd *FileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) {
+ if fd.opts.DenyPRead {
+ return 0, syserror.ESPIPE
+ }
if !fd.readable {
return 0, syserror.EBADF
}
@@ -533,6 +555,9 @@ func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opt
// offset, and returns the number of bytes written. PWrite is permitted to
// return partial writes with a nil error.
func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) {
+ if fd.opts.DenyPWrite {
+ return 0, syserror.ESPIPE
+ }
if !fd.writable {
return 0, syserror.EBADF
}
@@ -541,6 +566,9 @@ func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, o
// Write is similar to PWrite, but does not specify an offset.
func (fd *FileDescription) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) {
+ if fd.opts.InvalidWrite {
+ return 0, syserror.EINVAL
+ }
if !fd.writable {
return 0, syserror.EBADF
}
@@ -577,18 +605,23 @@ func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch.
// Listxattr returns all extended attribute names for the file represented by
// fd.
-func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) {
+//
+// If the size of the list (including a NUL terminating byte after every entry)
+// would exceed size, ERANGE may be returned. Note that implementations
+// are free to ignore size entirely and return without error). In all cases,
+// if size is 0, the list should be returned without error, regardless of size.
+func (fd *FileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) {
if fd.opts.UseDentryMetadata {
vfsObj := fd.vd.mount.vfs
rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{
Root: fd.vd,
Start: fd.vd,
})
- names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp)
+ names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp, size)
vfsObj.putResolvingPath(rp)
return names, err
}
- names, err := fd.impl.Listxattr(ctx)
+ names, err := fd.impl.Listxattr(ctx, size)
if err == syserror.ENOTSUP {
// Linux doesn't actually return ENOTSUP in this case; instead,
// fs/xattr.c:vfs_listxattr() falls back to allowing the security
@@ -601,34 +634,39 @@ func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) {
// Getxattr returns the value associated with the given extended attribute for
// the file represented by fd.
-func (fd *FileDescription) Getxattr(ctx context.Context, name string) (string, error) {
+//
+// If the size of the return value exceeds opts.Size, ERANGE may be returned
+// (note that implementations are free to ignore opts.Size entirely and return
+// without error). In all cases, if opts.Size is 0, the value should be
+// returned without error, regardless of size.
+func (fd *FileDescription) Getxattr(ctx context.Context, opts *GetxattrOptions) (string, error) {
if fd.opts.UseDentryMetadata {
vfsObj := fd.vd.mount.vfs
rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{
Root: fd.vd,
Start: fd.vd,
})
- val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, name)
+ val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, *opts)
vfsObj.putResolvingPath(rp)
return val, err
}
- return fd.impl.Getxattr(ctx, name)
+ return fd.impl.Getxattr(ctx, *opts)
}
// Setxattr changes the value associated with the given extended attribute for
// the file represented by fd.
-func (fd *FileDescription) Setxattr(ctx context.Context, opts SetxattrOptions) error {
+func (fd *FileDescription) Setxattr(ctx context.Context, opts *SetxattrOptions) error {
if fd.opts.UseDentryMetadata {
vfsObj := fd.vd.mount.vfs
rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{
Root: fd.vd,
Start: fd.vd,
})
- err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, opts)
+ err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, *opts)
vfsObj.putResolvingPath(rp)
return err
}
- return fd.impl.Setxattr(ctx, opts)
+ return fd.impl.Setxattr(ctx, *opts)
}
// Removexattr removes the given extended attribute from the file represented
diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go
index d45e602ce..f4c111926 100644
--- a/pkg/sentry/vfs/file_description_impl_util.go
+++ b/pkg/sentry/vfs/file_description_impl_util.go
@@ -130,14 +130,14 @@ func (FileDescriptionDefaultImpl) Ioctl(ctx context.Context, uio usermem.IO, arg
// Listxattr implements FileDescriptionImpl.Listxattr analogously to
// inode_operations::listxattr == NULL in Linux.
-func (FileDescriptionDefaultImpl) Listxattr(ctx context.Context) ([]string, error) {
+func (FileDescriptionDefaultImpl) Listxattr(ctx context.Context, size uint64) ([]string, error) {
// This isn't exactly accurate; see FileDescription.Listxattr.
return nil, syserror.ENOTSUP
}
// Getxattr implements FileDescriptionImpl.Getxattr analogously to
// inode::i_opflags & IOP_XATTR == 0 in Linux.
-func (FileDescriptionDefaultImpl) Getxattr(ctx context.Context, name string) (string, error) {
+func (FileDescriptionDefaultImpl) Getxattr(ctx context.Context, opts GetxattrOptions) (string, error) {
return "", syserror.ENOTSUP
}
diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go
index cd34782ff..a537a29d1 100644
--- a/pkg/sentry/vfs/filesystem.go
+++ b/pkg/sentry/vfs/filesystem.go
@@ -442,7 +442,13 @@ type FilesystemImpl interface {
// - If extended attributes are not supported by the filesystem,
// ListxattrAt returns nil. (See FileDescription.Listxattr for an
// explanation.)
- ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error)
+ //
+ // - If the size of the list (including a NUL terminating byte after every
+ // entry) would exceed size, ERANGE may be returned. Note that
+ // implementations are free to ignore size entirely and return without
+ // error). In all cases, if size is 0, the list should be returned without
+ // error, regardless of size.
+ ListxattrAt(ctx context.Context, rp *ResolvingPath, size uint64) ([]string, error)
// GetxattrAt returns the value associated with the given extended
// attribute for the file at rp.
@@ -451,7 +457,15 @@ type FilesystemImpl interface {
//
// - If extended attributes are not supported by the filesystem, GetxattrAt
// returns ENOTSUP.
- GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error)
+ //
+ // - If an extended attribute named opts.Name does not exist, ENODATA is
+ // returned.
+ //
+ // - If the size of the return value exceeds opts.Size, ERANGE may be
+ // returned (note that implementations are free to ignore opts.Size entirely
+ // and return without error). In all cases, if opts.Size is 0, the value
+ // should be returned without error, regardless of size.
+ GetxattrAt(ctx context.Context, rp *ResolvingPath, opts GetxattrOptions) (string, error)
// SetxattrAt changes the value associated with the given extended
// attribute for the file at rp.
@@ -460,6 +474,10 @@ type FilesystemImpl interface {
//
// - If extended attributes are not supported by the filesystem, SetxattrAt
// returns ENOTSUP.
+ //
+ // - If XATTR_CREATE is set in opts.Flag and opts.Name already exists,
+ // EEXIST is returned. If XATTR_REPLACE is set and opts.Name does not exist,
+ // ENODATA is returned.
SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error
// RemovexattrAt removes the given extended attribute from the file at rp.
@@ -468,6 +486,8 @@ type FilesystemImpl interface {
//
// - If extended attributes are not supported by the filesystem,
// RemovexattrAt returns ENOTSUP.
+ //
+ // - If name does not exist, ENODATA is returned.
RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error
// BoundEndpointAt returns the Unix socket endpoint bound at the path rp.
@@ -497,7 +517,7 @@ type FilesystemImpl interface {
// Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl.
PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error
- // TODO: inotify_add_watch()
+ // TODO(gvisor.dev/issue/1479): inotify_add_watch()
}
// PrependPathAtVFSRootError is returned by implementations of
diff --git a/pkg/sentry/vfs/memxattr/BUILD b/pkg/sentry/vfs/memxattr/BUILD
new file mode 100644
index 000000000..d8c4d27b9
--- /dev/null
+++ b/pkg/sentry/vfs/memxattr/BUILD
@@ -0,0 +1,15 @@
+load("//tools:defs.bzl", "go_library")
+
+package(licenses = ["notice"])
+
+go_library(
+ name = "memxattr",
+ srcs = ["xattr.go"],
+ visibility = ["//pkg/sentry:internal"],
+ deps = [
+ "//pkg/abi/linux",
+ "//pkg/sentry/vfs",
+ "//pkg/sync",
+ "//pkg/syserror",
+ ],
+)
diff --git a/pkg/sentry/vfs/memxattr/xattr.go b/pkg/sentry/vfs/memxattr/xattr.go
new file mode 100644
index 000000000..cc1e7d764
--- /dev/null
+++ b/pkg/sentry/vfs/memxattr/xattr.go
@@ -0,0 +1,102 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package memxattr provides a default, in-memory extended attribute
+// implementation.
+package memxattr
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+// SimpleExtendedAttributes implements extended attributes using a map of
+// names to values.
+//
+// +stateify savable
+type SimpleExtendedAttributes struct {
+ // mu protects the below fields.
+ mu sync.RWMutex `state:"nosave"`
+ xattrs map[string]string
+}
+
+// Getxattr returns the value at 'name'.
+func (x *SimpleExtendedAttributes) Getxattr(opts *vfs.GetxattrOptions) (string, error) {
+ x.mu.RLock()
+ value, ok := x.xattrs[opts.Name]
+ x.mu.RUnlock()
+ if !ok {
+ return "", syserror.ENODATA
+ }
+ // Check that the size of the buffer provided in getxattr(2) is large enough
+ // to contain the value.
+ if opts.Size != 0 && uint64(len(value)) > opts.Size {
+ return "", syserror.ERANGE
+ }
+ return value, nil
+}
+
+// Setxattr sets 'value' at 'name'.
+func (x *SimpleExtendedAttributes) Setxattr(opts *vfs.SetxattrOptions) error {
+ x.mu.Lock()
+ defer x.mu.Unlock()
+ if x.xattrs == nil {
+ if opts.Flags&linux.XATTR_REPLACE != 0 {
+ return syserror.ENODATA
+ }
+ x.xattrs = make(map[string]string)
+ }
+
+ _, ok := x.xattrs[opts.Name]
+ if ok && opts.Flags&linux.XATTR_CREATE != 0 {
+ return syserror.EEXIST
+ }
+ if !ok && opts.Flags&linux.XATTR_REPLACE != 0 {
+ return syserror.ENODATA
+ }
+
+ x.xattrs[opts.Name] = opts.Value
+ return nil
+}
+
+// Listxattr returns all names in xattrs.
+func (x *SimpleExtendedAttributes) Listxattr(size uint64) ([]string, error) {
+ // Keep track of the size of the buffer needed in listxattr(2) for the list.
+ listSize := 0
+ x.mu.RLock()
+ names := make([]string, 0, len(x.xattrs))
+ for n := range x.xattrs {
+ names = append(names, n)
+ // Add one byte per null terminator.
+ listSize += len(n) + 1
+ }
+ x.mu.RUnlock()
+ if size != 0 && uint64(listSize) > size {
+ return nil, syserror.ERANGE
+ }
+ return names, nil
+}
+
+// Removexattr removes the xattr at 'name'.
+func (x *SimpleExtendedAttributes) Removexattr(name string) error {
+ x.mu.Lock()
+ defer x.mu.Unlock()
+ if _, ok := x.xattrs[name]; !ok {
+ return syserror.ENODATA
+ }
+ delete(x.xattrs, name)
+ return nil
+}
diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go
index 7792eb1a0..f06946103 100644
--- a/pkg/sentry/vfs/mount.go
+++ b/pkg/sentry/vfs/mount.go
@@ -233,9 +233,9 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia
}
vd.dentry.mu.Lock()
}
- // TODO: Linux requires that either both the mount point and the mount root
- // are directories, or neither are, and returns ENOTDIR if this is not the
- // case.
+ // TODO(gvisor.dev/issue/1035): Linux requires that either both the mount
+ // point and the mount root are directories, or neither are, and returns
+ // ENOTDIR if this is not the case.
mntns := vd.mount.ns
mnt := newMount(vfs, fs, root, mntns, opts)
vfs.mounts.seq.BeginWrite()
@@ -274,9 +274,9 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti
}
}
- // TODO(jamieliu): Linux special-cases umount of the caller's root, which
- // we don't implement yet (we'll just fail it since the caller holds a
- // reference on it).
+ // TODO(gvisor.dev/issue/1035): Linux special-cases umount of the caller's
+ // root, which we don't implement yet (we'll just fail it since the caller
+ // holds a reference on it).
vfs.mounts.seq.BeginWrite()
if opts.Flags&linux.MNT_DETACH == 0 {
@@ -835,7 +835,8 @@ func superBlockOpts(mountPath string, mnt *Mount) string {
// NOTE(b/147673608): If the mount is a cgroup, we also need to include
// the cgroup name in the options. For now we just read that from the
// path.
- // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we
+ //
+ // TODO(gvisor.dev/issue/190): Once gVisor has full cgroup support, we
// should get this value from the cgroup itself, and not rely on the
// path.
if mnt.fs.FilesystemType().Name() == "cgroup" {
diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go
index 3b933468d..3335e4057 100644
--- a/pkg/sentry/vfs/mount_test.go
+++ b/pkg/sentry/vfs/mount_test.go
@@ -55,7 +55,7 @@ func TestMountTableInsertLookup(t *testing.T) {
}
}
-// TODO: concurrent lookup/insertion/removal
+// TODO(gvisor.dev/issue/1035): concurrent lookup/insertion/removal.
// must be powers of 2
var benchNumMounts = []int{1 << 2, 1 << 5, 1 << 8}
diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go
index 3e90dc4ed..534528ce6 100644
--- a/pkg/sentry/vfs/options.go
+++ b/pkg/sentry/vfs/options.go
@@ -16,6 +16,7 @@ package vfs
import (
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
)
// GetDentryOptions contains options to VirtualFilesystem.GetDentryAt() and
@@ -44,6 +45,10 @@ type MknodOptions struct {
// DevMinor are the major and minor device numbers for the created device.
DevMajor uint32
DevMinor uint32
+
+ // Endpoint is the endpoint to bind to the created file, if a socket file is
+ // being created for bind(2) on a Unix domain socket.
+ Endpoint transport.BoundEndpoint
}
// MountFlags contains flags as specified for mount(2), e.g. MS_NOEXEC.
@@ -127,6 +132,20 @@ type SetStatOptions struct {
Stat linux.Statx
}
+// GetxattrOptions contains options to VirtualFilesystem.GetxattrAt(),
+// FilesystemImpl.GetxattrAt(), FileDescription.Getxattr(), and
+// FileDescriptionImpl.Getxattr().
+type GetxattrOptions struct {
+ // Name is the name of the extended attribute to retrieve.
+ Name string
+
+ // Size is the maximum value size that the caller will tolerate. If the value
+ // is larger than size, getxattr methods may return ERANGE, but they are also
+ // free to ignore the hint entirely (i.e. the value returned may be larger
+ // than size). All size checking is done independently at the syscall layer.
+ Size uint64
+}
+
// SetxattrOptions contains options to VirtualFilesystem.SetxattrAt(),
// FilesystemImpl.SetxattrAt(), FileDescription.Setxattr(), and
// FileDescriptionImpl.Setxattr().
diff --git a/pkg/sentry/vfs/timerfd.go b/pkg/sentry/vfs/timerfd.go
new file mode 100644
index 000000000..42b880656
--- /dev/null
+++ b/pkg/sentry/vfs/timerfd.go
@@ -0,0 +1,142 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs
+
+import (
+ "sync/atomic"
+
+ "gvisor.dev/gvisor/pkg/context"
+ ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/pkg/waiter"
+)
+
+// TimerFileDescription implements FileDescriptionImpl for timer fds. It also
+// implements ktime.TimerListener.
+type TimerFileDescription struct {
+ vfsfd FileDescription
+ FileDescriptionDefaultImpl
+ DentryMetadataFileDescriptionImpl
+
+ events waiter.Queue
+ timer *ktime.Timer
+
+ // val is the number of timer expirations since the last successful
+ // call to PRead, or SetTime. val must be accessed using atomic memory
+ // operations.
+ val uint64
+}
+
+var _ FileDescriptionImpl = (*TimerFileDescription)(nil)
+var _ ktime.TimerListener = (*TimerFileDescription)(nil)
+
+// NewTimerFD returns a new timer fd.
+func (vfs *VirtualFilesystem) NewTimerFD(clock ktime.Clock, flags uint32) (*FileDescription, error) {
+ vd := vfs.NewAnonVirtualDentry("[timerfd]")
+ defer vd.DecRef()
+ tfd := &TimerFileDescription{}
+ tfd.timer = ktime.NewTimer(clock, tfd)
+ if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{
+ UseDentryMetadata: true,
+ DenyPRead: true,
+ DenyPWrite: true,
+ InvalidWrite: true,
+ }); err != nil {
+ return nil, err
+ }
+ return &tfd.vfsfd, nil
+}
+
+// Read implements FileDescriptionImpl.Read.
+func (tfd *TimerFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) {
+ const sizeofUint64 = 8
+ if dst.NumBytes() < sizeofUint64 {
+ return 0, syserror.EINVAL
+ }
+ if val := atomic.SwapUint64(&tfd.val, 0); val != 0 {
+ var buf [sizeofUint64]byte
+ usermem.ByteOrder.PutUint64(buf[:], val)
+ if _, err := dst.CopyOut(ctx, buf[:]); err != nil {
+ // Linux does not undo consuming the number of
+ // expirations even if writing to userspace fails.
+ return 0, err
+ }
+ return sizeofUint64, nil
+ }
+ return 0, syserror.ErrWouldBlock
+}
+
+// Clock returns the timer fd's Clock.
+func (tfd *TimerFileDescription) Clock() ktime.Clock {
+ return tfd.timer.Clock()
+}
+
+// GetTime returns the associated Timer's setting and the time at which it was
+// observed.
+func (tfd *TimerFileDescription) GetTime() (ktime.Time, ktime.Setting) {
+ return tfd.timer.Get()
+}
+
+// SetTime atomically changes the associated Timer's setting, resets the number
+// of expirations to 0, and returns the previous setting and the time at which
+// it was observed.
+func (tfd *TimerFileDescription) SetTime(s ktime.Setting) (ktime.Time, ktime.Setting) {
+ return tfd.timer.SwapAnd(s, func() { atomic.StoreUint64(&tfd.val, 0) })
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (tfd *TimerFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask {
+ var ready waiter.EventMask
+ if atomic.LoadUint64(&tfd.val) != 0 {
+ ready |= waiter.EventIn
+ }
+ return ready
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (tfd *TimerFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+ tfd.events.EventRegister(e, mask)
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (tfd *TimerFileDescription) EventUnregister(e *waiter.Entry) {
+ tfd.events.EventUnregister(e)
+}
+
+// PauseTimer pauses the associated Timer.
+func (tfd *TimerFileDescription) PauseTimer() {
+ tfd.timer.Pause()
+}
+
+// ResumeTimer resumes the associated Timer.
+func (tfd *TimerFileDescription) ResumeTimer() {
+ tfd.timer.Resume()
+}
+
+// Release implements FileDescriptionImpl.Release()
+func (tfd *TimerFileDescription) Release() {
+ tfd.timer.Destroy()
+}
+
+// Notify implements ktime.TimerListener.Notify.
+func (tfd *TimerFileDescription) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) {
+ atomic.AddUint64(&tfd.val, exp)
+ tfd.events.Notify(waiter.EventIn)
+ return ktime.Setting{}, false
+}
+
+// Destroy implements ktime.TimerListener.Destroy.
+func (tfd *TimerFileDescription) Destroy() {}
diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go
index 720b90d8f..f592913d5 100644
--- a/pkg/sentry/vfs/vfs.go
+++ b/pkg/sentry/vfs/vfs.go
@@ -680,10 +680,10 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti
// ListxattrAt returns all extended attribute names for the file at the given
// path.
-func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) ([]string, error) {
+func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, size uint64) ([]string, error) {
rp := vfs.getResolvingPath(creds, pop)
for {
- names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp)
+ names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp, size)
if err == nil {
vfs.putResolvingPath(rp)
return names, nil
@@ -705,10 +705,10 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede
// GetxattrAt returns the value associated with the given extended attribute
// for the file at the given path.
-func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, name string) (string, error) {
+func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetxattrOptions) (string, error) {
rp := vfs.getResolvingPath(creds, pop)
for {
- val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, name)
+ val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, *opts)
if err == nil {
vfs.putResolvingPath(rp)
return val, nil