summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fs/proc
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/fs/proc')
-rw-r--r--pkg/sentry/fs/proc/BUILD1
-rw-r--r--pkg/sentry/fs/proc/README.md4
-rw-r--r--pkg/sentry/fs/proc/mounts.go16
-rw-r--r--pkg/sentry/fs/proc/net.go5
-rw-r--r--pkg/sentry/fs/proc/sys_net.go4
-rw-r--r--pkg/sentry/fs/proc/task.go143
6 files changed, 137 insertions, 36 deletions
diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD
index 280093c5e..77c2c5c0e 100644
--- a/pkg/sentry/fs/proc/BUILD
+++ b/pkg/sentry/fs/proc/BUILD
@@ -36,6 +36,7 @@ go_library(
"//pkg/sentry/fs/proc/device",
"//pkg/sentry/fs/proc/seqfile",
"//pkg/sentry/fs/ramfs",
+ "//pkg/sentry/fsbridge",
"//pkg/sentry/inet",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/auth",
diff --git a/pkg/sentry/fs/proc/README.md b/pkg/sentry/fs/proc/README.md
index 5d4ec6c7b..6667a0916 100644
--- a/pkg/sentry/fs/proc/README.md
+++ b/pkg/sentry/fs/proc/README.md
@@ -11,6 +11,8 @@ inconsistency, please file a bug.
The following files are implemented:
+<!-- mdformat off(don't wrap the table) -->
+
| File /proc/ | Content |
| :------------------------ | :---------------------------------------------------- |
| [cpuinfo](#cpuinfo) | Info about the CPU |
@@ -22,6 +24,8 @@ The following files are implemented:
| [uptime](#uptime) | Wall clock since boot, combined idle time of all cpus |
| [version](#version) | Kernel version |
+<!-- mdformat on -->
+
### cpuinfo
```bash
diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go
index c10888100..94deb553b 100644
--- a/pkg/sentry/fs/proc/mounts.go
+++ b/pkg/sentry/fs/proc/mounts.go
@@ -60,13 +60,15 @@ func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) {
})
for _, m := range ms {
mroot := m.Root()
+ if mroot == nil {
+ continue // No longer valid.
+ }
mountPath, desc := mroot.FullName(rootDir)
mroot.DecRef()
if !desc {
// MountSources that are not descendants of the chroot jail are ignored.
continue
}
-
fn(mountPath, m)
}
}
@@ -91,6 +93,12 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se
var buf bytes.Buffer
forEachMount(mif.t, func(mountPath string, m *fs.Mount) {
+ mroot := m.Root()
+ if mroot == nil {
+ return // No longer valid.
+ }
+ defer mroot.DecRef()
+
// Format:
// 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue
// (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11)
@@ -107,9 +115,6 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se
// (3) Major:Minor device ID. We don't have a superblock, so we
// just use the root inode device number.
- mroot := m.Root()
- defer mroot.DecRef()
-
sa := mroot.Inode.StableAttr
fmt.Fprintf(&buf, "%d:%d ", sa.DeviceFileMajor, sa.DeviceFileMinor)
@@ -207,6 +212,9 @@ func (mf *mountsFile) ReadSeqFileData(ctx context.Context, handle seqfile.SeqHan
//
// The "needs dump"and fsck flags are always 0, which is allowed.
root := m.Root()
+ if root == nil {
+ return // No longer valid.
+ }
defer root.DecRef()
flags := root.Inode.MountSource.Flags
diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go
index 6f2775344..95d5817ff 100644
--- a/pkg/sentry/fs/proc/net.go
+++ b/pkg/sentry/fs/proc/net.go
@@ -43,7 +43,10 @@ import (
// newNet creates a new proc net entry.
func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode {
var contents map[string]*fs.Inode
- if s := p.k.NetworkStack(); s != nil {
+ // TODO(gvisor.dev/issue/1833): Support for using the network stack in the
+ // network namespace of the calling process. We should make this per-process,
+ // a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net.
+ if s := p.k.RootNetworkNamespace().Stack(); s != nil {
contents = map[string]*fs.Inode{
"dev": seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc),
"snmp": seqfile.NewSeqFileInode(ctx, &netSnmp{s: s}, msrc),
diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go
index 0772d4ae4..d4c4b533d 100644
--- a/pkg/sentry/fs/proc/sys_net.go
+++ b/pkg/sentry/fs/proc/sys_net.go
@@ -357,7 +357,9 @@ func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s ine
func (p *proc) newSysNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode {
var contents map[string]*fs.Inode
- if s := p.k.NetworkStack(); s != nil {
+ // TODO(gvisor.dev/issue/1833): Support for using the network stack in the
+ // network namespace of the calling process.
+ if s := p.k.RootNetworkNamespace().Stack(); s != nil {
contents = map[string]*fs.Inode{
"ipv4": p.newSysNetIPv4Dir(ctx, msrc, s),
"core": p.newSysNetCore(ctx, msrc, s),
diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go
index ca020e11e..4e9b0fc00 100644
--- a/pkg/sentry/fs/proc/task.go
+++ b/pkg/sentry/fs/proc/task.go
@@ -28,6 +28,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/fs/proc/device"
"gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile"
"gvisor.dev/gvisor/pkg/sentry/fs/ramfs"
+ "gvisor.dev/gvisor/pkg/sentry/fsbridge"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/mm"
@@ -71,24 +72,26 @@ var _ fs.InodeOperations = (*taskDir)(nil)
// newTaskDir creates a new proc task entry.
func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode {
contents := map[string]*fs.Inode{
- "auxv": newAuxvec(t, msrc),
- "cmdline": newExecArgInode(t, msrc, cmdlineExecArg),
- "comm": newComm(t, msrc),
- "environ": newExecArgInode(t, msrc, environExecArg),
- "exe": newExe(t, msrc),
- "fd": newFdDir(t, msrc),
- "fdinfo": newFdInfoDir(t, msrc),
- "gid_map": newGIDMap(t, msrc),
- "io": newIO(t, msrc, isThreadGroup),
- "maps": newMaps(t, msrc),
- "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
- "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
- "ns": newNamespaceDir(t, msrc),
- "smaps": newSmaps(t, msrc),
- "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns),
- "statm": newStatm(t, msrc),
- "status": newStatus(t, msrc, p.pidns),
- "uid_map": newUIDMap(t, msrc),
+ "auxv": newAuxvec(t, msrc),
+ "cmdline": newExecArgInode(t, msrc, cmdlineExecArg),
+ "comm": newComm(t, msrc),
+ "environ": newExecArgInode(t, msrc, environExecArg),
+ "exe": newExe(t, msrc),
+ "fd": newFdDir(t, msrc),
+ "fdinfo": newFdInfoDir(t, msrc),
+ "gid_map": newGIDMap(t, msrc),
+ "io": newIO(t, msrc, isThreadGroup),
+ "maps": newMaps(t, msrc),
+ "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
+ "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
+ "ns": newNamespaceDir(t, msrc),
+ "oom_score": newOOMScore(t, msrc),
+ "oom_score_adj": newOOMScoreAdj(t, msrc),
+ "smaps": newSmaps(t, msrc),
+ "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns),
+ "statm": newStatm(t, msrc),
+ "status": newStatus(t, msrc, p.pidns),
+ "uid_map": newUIDMap(t, msrc),
}
if isThreadGroup {
contents["task"] = p.newSubtasks(t, msrc)
@@ -249,7 +252,7 @@ func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
return newProcInode(t, exeSymlink, msrc, fs.Symlink, t)
}
-func (e *exe) executable() (d *fs.Dirent, err error) {
+func (e *exe) executable() (file fsbridge.File, err error) {
e.t.WithMuLocked(func(t *kernel.Task) {
mm := t.MemoryManager()
if mm == nil {
@@ -262,8 +265,8 @@ func (e *exe) executable() (d *fs.Dirent, err error) {
// The MemoryManager may be destroyed, in which case
// MemoryManager.destroy will simply set the executable to nil
// (with locks held).
- d = mm.Executable()
- if d == nil {
+ file = mm.Executable()
+ if file == nil {
err = syserror.ENOENT
}
})
@@ -283,15 +286,7 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) {
}
defer exec.DecRef()
- root := fs.RootFromContext(ctx)
- if root == nil {
- // This doesn't correspond to anything in Linux because the vfs is
- // global there.
- return "", syserror.EINVAL
- }
- defer root.DecRef()
- n, _ := exec.FullName(root)
- return n, nil
+ return exec.PathnameWithDeleted(ctx), nil
}
// namespaceSymlink represents a symlink in the namespacefs, such as the files
@@ -803,4 +798,92 @@ func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc
return int64(n), err
}
+// newOOMScore returns a oom_score file. It is a stub that always returns 0.
+// TODO(gvisor.dev/issue/1967)
+func newOOMScore(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ return newStaticProcInode(t, msrc, []byte("0\n"))
+}
+
+// oomScoreAdj is a file containing the oom_score adjustment for a task.
+//
+// +stateify savable
+type oomScoreAdj struct {
+ fsutil.SimpleFileInode
+
+ t *kernel.Task
+}
+
+// +stateify savable
+type oomScoreAdjFile struct {
+ fsutil.FileGenericSeek `state:"nosave"`
+ fsutil.FileNoIoctl `state:"nosave"`
+ fsutil.FileNoMMap `state:"nosave"`
+ fsutil.FileNoSplice `state:"nosave"`
+ fsutil.FileNoopFlush `state:"nosave"`
+ fsutil.FileNoopFsync `state:"nosave"`
+ fsutil.FileNoopRelease `state:"nosave"`
+ fsutil.FileNotDirReaddir `state:"nosave"`
+ fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+ waiter.AlwaysReady `state:"nosave"`
+
+ t *kernel.Task
+}
+
+// newOOMScoreAdj returns a oom_score_adj file.
+func newOOMScoreAdj(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ i := &oomScoreAdj{
+ SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC),
+ t: t,
+ }
+ return newProcInode(t, i, msrc, fs.SpecialFile, t)
+}
+
+// Truncate implements fs.InodeOperations.Truncate. Truncate is called when
+// O_TRUNC is specified for any kind of existing Dirent but is not called via
+// (f)truncate for proc files.
+func (*oomScoreAdj) Truncate(context.Context, *fs.Inode, int64) error {
+ return nil
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+func (o *oomScoreAdj) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+ return fs.NewFile(ctx, dirent, flags, &oomScoreAdjFile{t: o.t}), nil
+}
+
+// Read implements fs.FileOperations.Read.
+func (f *oomScoreAdjFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) {
+ if offset != 0 {
+ return 0, io.EOF
+ }
+ adj, err := f.t.OOMScoreAdj()
+ if err != nil {
+ return 0, err
+ }
+ adjBytes := []byte(strconv.FormatInt(int64(adj), 10) + "\n")
+ n, err := dst.CopyOut(ctx, adjBytes)
+ return int64(n), err
+}
+
+// Write implements fs.FileOperations.Write.
+func (f *oomScoreAdjFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) {
+ if src.NumBytes() == 0 {
+ return 0, nil
+ }
+
+ // Limit input size so as not to impact performance if input size is large.
+ src = src.TakeFirst(usermem.PageSize - 1)
+
+ var v int32
+ n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts)
+ if err != nil {
+ return 0, err
+ }
+
+ if err := f.t.SetOOMScoreAdj(v); err != nil {
+ return 0, err
+ }
+
+ return n, nil
+}
+
// LINT.ThenChange(../../fsimpl/proc/task.go|../../fsimpl/proc/task_files.go)
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-19 20:51:26 +0000