summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fsimpl
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/fsimpl')
-rw-r--r--pkg/sentry/fsimpl/cgroupfs/cgroupfs.go5
-rw-r--r--pkg/sentry/fsimpl/gofer/gofer.go28
-rw-r--r--pkg/sentry/fsimpl/proc/task.go2
-rw-r--r--pkg/sentry/fsimpl/proc/task_files.go111
4 files changed, 129 insertions, 17 deletions
diff --git a/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go b/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go
index 24e28a51f..22c8b7fda 100644
--- a/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go
+++ b/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go
@@ -383,11 +383,6 @@ func (d *dir) DecRef(ctx context.Context) {
d.dirRefs.DecRef(func() { d.Destroy(ctx) })
}
-// StatFS implements kernfs.Inode.StatFS.
-func (d *dir) StatFS(ctx context.Context, fs *vfs.Filesystem) (linux.Statfs, error) {
- return vfs.GenericStatFS(linux.CGROUP_SUPER_MAGIC), nil
-}
-
// controllerFile represents a generic control file that appears within a cgroup
// directory.
//
diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go
index ec8d58cc9..25d2e39d6 100644
--- a/pkg/sentry/fsimpl/gofer/gofer.go
+++ b/pkg/sentry/fsimpl/gofer/gofer.go
@@ -1161,6 +1161,13 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs
if !d.isSynthetic() {
if stat.Mask != 0 {
+ if stat.Mask&linux.STATX_SIZE != 0 {
+ // d.dataMu must be held around the update to both the remote
+ // file's size and d.size to serialize with writeback (which
+ // might otherwise write data back up to the old d.size after
+ // the remote file has been truncated).
+ d.dataMu.Lock()
+ }
if err := d.file.setAttr(ctx, p9.SetAttrMask{
Permissions: stat.Mask&linux.STATX_MODE != 0,
UID: stat.Mask&linux.STATX_UID != 0,
@@ -1180,13 +1187,16 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs
MTimeSeconds: uint64(stat.Mtime.Sec),
MTimeNanoSeconds: uint64(stat.Mtime.Nsec),
}); err != nil {
+ if stat.Mask&linux.STATX_SIZE != 0 {
+ d.dataMu.Unlock() // +checklocksforce: locked conditionally above
+ }
return err
}
if stat.Mask&linux.STATX_SIZE != 0 {
// d.size should be kept up to date, and privatized
// copy-on-write mappings of truncated pages need to be
// invalidated, even if InteropModeShared is in effect.
- d.updateSizeLocked(stat.Size)
+ d.updateSizeAndUnlockDataMuLocked(stat.Size) // +checklocksforce: locked conditionally above
}
}
if d.fs.opts.interop == InteropModeShared {
@@ -1249,6 +1259,14 @@ func (d *dentry) doAllocate(ctx context.Context, offset, length uint64, allocate
// Preconditions: d.metadataMu must be locked.
func (d *dentry) updateSizeLocked(newSize uint64) {
d.dataMu.Lock()
+ d.updateSizeAndUnlockDataMuLocked(newSize)
+}
+
+// Preconditions: d.metadataMu and d.dataMu must be locked.
+//
+// Postconditions: d.dataMu is unlocked.
+// +checklocksrelease:d.dataMu
+func (d *dentry) updateSizeAndUnlockDataMuLocked(newSize uint64) {
oldSize := d.size
atomic.StoreUint64(&d.size, newSize)
// d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings
@@ -1257,9 +1275,9 @@ func (d *dentry) updateSizeLocked(newSize uint64) {
// contents beyond the new d.size. (We are still holding d.metadataMu,
// so we can't race with Write or another truncate.)
d.dataMu.Unlock()
- if d.size < oldSize {
+ if newSize < oldSize {
oldpgend, _ := hostarch.PageRoundUp(oldSize)
- newpgend, _ := hostarch.PageRoundUp(d.size)
+ newpgend, _ := hostarch.PageRoundUp(newSize)
if oldpgend != newpgend {
d.mapsMu.Lock()
d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{
@@ -1275,8 +1293,8 @@ func (d *dentry) updateSizeLocked(newSize uint64) {
// truncated pages have been removed from the remote file, they
// should be dropped without being written back.
d.dataMu.Lock()
- d.cache.Truncate(d.size, d.fs.mfp.MemoryFile())
- d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend})
+ d.cache.Truncate(newSize, d.fs.mfp.MemoryFile())
+ d.dirty.KeepClean(memmap.MappableRange{newSize, oldpgend})
d.dataMu.Unlock()
}
}
diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go
index cbbc0935a..f54811edf 100644
--- a/pkg/sentry/fsimpl/proc/task.go
+++ b/pkg/sentry/fsimpl/proc/task.go
@@ -78,7 +78,7 @@ func (fs *filesystem) newTaskInode(ctx context.Context, task *kernel.Task, pidns
"smaps": fs.newTaskOwnedInode(ctx, task, fs.NextIno(), 0444, &smapsData{task: task}),
"stat": fs.newTaskOwnedInode(ctx, task, fs.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}),
"statm": fs.newTaskOwnedInode(ctx, task, fs.NextIno(), 0444, &statmData{task: task}),
- "status": fs.newTaskOwnedInode(ctx, task, fs.NextIno(), 0444, &statusData{task: task, pidns: pidns}),
+ "status": fs.newStatusInode(ctx, task, pidns, fs.NextIno(), 0444),
"uid_map": fs.newTaskOwnedInode(ctx, task, fs.NextIno(), 0644, &idMapData{task: task, gids: false}),
}
if isThreadGroup {
diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go
index 5bb6bc372..0ce3ed797 100644
--- a/pkg/sentry/fsimpl/proc/task_files.go
+++ b/pkg/sentry/fsimpl/proc/task_files.go
@@ -661,34 +661,119 @@ func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error {
return nil
}
-// statusData implements vfs.DynamicBytesSource for /proc/[pid]/status.
+// statusInode implements kernfs.Inode for /proc/[pid]/status.
//
// +stateify savable
-type statusData struct {
- kernfs.DynamicBytesFile
+type statusInode struct {
+ kernfs.InodeAttrs
+ kernfs.InodeNoStatFS
+ kernfs.InodeNoopRefCount
+ kernfs.InodeNotDirectory
+ kernfs.InodeNotSymlink
task *kernel.Task
pidns *kernel.PIDNamespace
+ locks vfs.FileLocks
}
-var _ dynamicInode = (*statusData)(nil)
+// statusFD implements vfs.FileDescriptionImpl and vfs.DynamicByteSource for
+// /proc/[pid]/status.
+//
+// +stateify savable
+type statusFD struct {
+ statusFDLowerBase
+ vfs.DynamicBytesFileDescriptionImpl
+ vfs.LockFD
+
+ vfsfd vfs.FileDescription
+
+ inode *statusInode
+ task *kernel.Task
+ pidns *kernel.PIDNamespace
+ userns *auth.UserNamespace // equivalent to struct file::f_cred::user_ns
+}
+
+// statusFDLowerBase is a dumb hack to ensure that statusFD prefers
+// vfs.DynamicBytesFileDescriptionImpl methods to vfs.FileDescriptinDefaultImpl
+// methods.
+//
+// +stateify savable
+type statusFDLowerBase struct {
+ vfs.FileDescriptionDefaultImpl
+}
+
+func (fs *filesystem) newStatusInode(ctx context.Context, task *kernel.Task, pidns *kernel.PIDNamespace, ino uint64, perm linux.FileMode) kernfs.Inode {
+ // Note: credentials are overridden by taskOwnedInode.
+ inode := &statusInode{
+ task: task,
+ pidns: pidns,
+ }
+ inode.InodeAttrs.Init(ctx, task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeRegular|perm)
+ return &taskOwnedInode{Inode: inode, owner: task}
+}
+
+// Open implements kernfs.Inode.Open.
+func (s *statusInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
+ fd := &statusFD{
+ inode: s,
+ task: s.task,
+ pidns: s.pidns,
+ userns: rp.Credentials().UserNamespace,
+ }
+ fd.LockFD.Init(&s.locks)
+ if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil {
+ return nil, err
+ }
+ fd.SetDataSource(fd)
+ return &fd.vfsfd, nil
+}
+
+// SetStat implements kernfs.Inode.SetStat.
+func (*statusInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error {
+ return linuxerr.EPERM
+}
+
+// Release implements vfs.FileDescriptionImpl.Release.
+func (s *statusFD) Release(ctx context.Context) {
+}
+
+// Stat implements vfs.FileDescriptionImpl.Stat.
+func (s *statusFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) {
+ fs := s.vfsfd.VirtualDentry().Mount().Filesystem()
+ return s.inode.Stat(ctx, fs, opts)
+}
+
+// SetStat implements vfs.FileDescriptionImpl.SetStat.
+func (s *statusFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error {
+ return linuxerr.EPERM
+}
// Generate implements vfs.DynamicBytesSource.Generate.
-func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error {
+func (s *statusFD) Generate(ctx context.Context, buf *bytes.Buffer) error {
fmt.Fprintf(buf, "Name:\t%s\n", s.task.Name())
fmt.Fprintf(buf, "State:\t%s\n", s.task.StateStatus())
fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.task.ThreadGroup()))
fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.task))
+
ppid := kernel.ThreadID(0)
if parent := s.task.Parent(); parent != nil {
ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup())
}
fmt.Fprintf(buf, "PPid:\t%d\n", ppid)
+
tpid := kernel.ThreadID(0)
if tracer := s.task.Tracer(); tracer != nil {
tpid = s.pidns.IDOfTask(tracer)
}
fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid)
+
+ creds := s.task.Credentials()
+ ruid := creds.RealKUID.In(s.userns).OrOverflow()
+ euid := creds.EffectiveKUID.In(s.userns).OrOverflow()
+ suid := creds.SavedKUID.In(s.userns).OrOverflow()
+ rgid := creds.RealKGID.In(s.userns).OrOverflow()
+ egid := creds.EffectiveKGID.In(s.userns).OrOverflow()
+ sgid := creds.SavedKGID.In(s.userns).OrOverflow()
var fds int
var vss, rss, data uint64
s.task.WithMuLocked(func(t *kernel.Task) {
@@ -701,12 +786,26 @@ func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error {
data = mm.VirtualDataSize()
}
})
+ // Filesystem user/group IDs aren't implemented; effective UID/GID are used
+ // instead.
+ fmt.Fprintf(buf, "Uid:\t%d\t%d\t%d\t%d\n", ruid, euid, suid, euid)
+ fmt.Fprintf(buf, "Gid:\t%d\t%d\t%d\t%d\n", rgid, egid, sgid, egid)
fmt.Fprintf(buf, "FDSize:\t%d\n", fds)
+ buf.WriteString("Groups:\t ")
+ // There is a space between each pair of supplemental GIDs, as well as an
+ // unconditional trailing space that some applications actually depend on.
+ var sep string
+ for _, kgid := range creds.ExtraKGIDs {
+ fmt.Fprintf(buf, "%s%d", sep, kgid.In(s.userns).OrOverflow())
+ sep = " "
+ }
+ buf.WriteString(" \n")
+
fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10)
fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10)
fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10)
+
fmt.Fprintf(buf, "Threads:\t%d\n", s.task.ThreadGroup().Count())
- creds := s.task.Credentials()
fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps)
fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps)
fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps)
generated by cgit v1.2.3 (git 2.39.1) at 2025-03-25 21:43:15 +0000