summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fsimpl/tmpfs
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/fsimpl/tmpfs')
-rw-r--r--pkg/sentry/fsimpl/tmpfs/BUILD13
-rw-r--r--pkg/sentry/fsimpl/tmpfs/benchmark_test.go56
-rw-r--r--pkg/sentry/fsimpl/tmpfs/device_file.go1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/directory.go13
-rw-r--r--pkg/sentry/fsimpl/tmpfs/filesystem.go218
-rw-r--r--pkg/sentry/fsimpl/tmpfs/named_pipe.go5
-rw-r--r--pkg/sentry/fsimpl/tmpfs/pipe_test.go22
-rw-r--r--pkg/sentry/fsimpl/tmpfs/regular_file.go117
-rw-r--r--pkg/sentry/fsimpl/tmpfs/socket_file.go2
-rw-r--r--pkg/sentry/fsimpl/tmpfs/symlink.go1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs.go242
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs_test.go8
12 files changed, 406 insertions, 292 deletions
diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD
index e73732a6b..5cd428d64 100644
--- a/pkg/sentry/fsimpl/tmpfs/BUILD
+++ b/pkg/sentry/fsimpl/tmpfs/BUILD
@@ -26,6 +26,17 @@ go_template_instance(
},
)
+go_template_instance(
+ name = "inode_refs",
+ out = "inode_refs.go",
+ package = "tmpfs",
+ prefix = "inode",
+ template = "//pkg/refs_vfs2:refs_template",
+ types = {
+ "T": "inode",
+ },
+)
+
go_library(
name = "tmpfs",
srcs = [
@@ -34,6 +45,7 @@ go_library(
"directory.go",
"filesystem.go",
"fstree.go",
+ "inode_refs.go",
"named_pipe.go",
"regular_file.go",
"socket_file.go",
@@ -47,6 +59,7 @@ go_library(
"//pkg/context",
"//pkg/fspath",
"//pkg/log",
+ "//pkg/refs",
"//pkg/safemem",
"//pkg/sentry/arch",
"//pkg/sentry/fs",
diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go
index 2fb5c4d84..5209a17af 100644
--- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go
@@ -83,7 +83,7 @@ func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent
}
err = fn(root, d)
- d.DecRef()
+ d.DecRef(ctx)
return err
}
@@ -105,17 +105,17 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to create mount namespace: %v", err)
}
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
var filePathBuilder strings.Builder
filePathBuilder.WriteByte('/')
// Create nested directories with given depth.
root := mntns.Root()
- defer root.DecRef()
+ defer root.DecRef(ctx)
d := root
d.IncRef()
- defer d.DecRef()
+ defer d.DecRef(ctx)
for i := depth; i > 0; i-- {
name := fmt.Sprintf("%d", i)
if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil {
@@ -125,7 +125,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to directory %q: %v", name, err)
}
- d.DecRef()
+ d.DecRef(ctx)
d = next
filePathBuilder.WriteString(name)
filePathBuilder.WriteByte('/')
@@ -136,7 +136,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to create file %q: %v", filename, err)
}
- file.DecRef()
+ file.DecRef(ctx)
filePathBuilder.WriteString(filename)
filePath := filePathBuilder.String()
@@ -176,24 +176,24 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) {
// Create VFS.
vfsObj := vfs.VirtualFilesystem{}
- if err := vfsObj.Init(); err != nil {
+ if err := vfsObj.Init(ctx); err != nil {
b.Fatalf("VFS init: %v", err)
}
vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
b.Fatalf("failed to create tmpfs root mount: %v", err)
}
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
var filePathBuilder strings.Builder
filePathBuilder.WriteByte('/')
// Create nested directories with given depth.
root := mntns.Root()
- defer root.DecRef()
+ defer root.DecRef(ctx)
vd := root
vd.IncRef()
for i := depth; i > 0; i-- {
@@ -212,7 +212,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to directory %q: %v", name, err)
}
- vd.DecRef()
+ vd.DecRef(ctx)
vd = nextVD
filePathBuilder.WriteString(name)
filePathBuilder.WriteByte('/')
@@ -228,12 +228,12 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) {
Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL,
Mode: 0644,
})
- vd.DecRef()
+ vd.DecRef(ctx)
vd = vfs.VirtualDentry{}
if err != nil {
b.Fatalf("failed to create file %q: %v", filename, err)
}
- defer fd.DecRef()
+ defer fd.DecRef(ctx)
filePathBuilder.WriteString(filename)
filePath := filePathBuilder.String()
@@ -278,14 +278,14 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to create mount namespace: %v", err)
}
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
var filePathBuilder strings.Builder
filePathBuilder.WriteByte('/')
// Create and mount the submount.
root := mntns.Root()
- defer root.DecRef()
+ defer root.DecRef(ctx)
if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil {
b.Fatalf("failed to create mount point: %v", err)
}
@@ -293,7 +293,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to mount point: %v", err)
}
- defer mountPoint.DecRef()
+ defer mountPoint.DecRef(ctx)
submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil)
if err != nil {
b.Fatalf("failed to create tmpfs submount: %v", err)
@@ -309,7 +309,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to mount root: %v", err)
}
- defer d.DecRef()
+ defer d.DecRef(ctx)
for i := depth; i > 0; i-- {
name := fmt.Sprintf("%d", i)
if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil {
@@ -319,7 +319,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to directory %q: %v", name, err)
}
- d.DecRef()
+ d.DecRef(ctx)
d = next
filePathBuilder.WriteString(name)
filePathBuilder.WriteByte('/')
@@ -330,7 +330,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to create file %q: %v", filename, err)
}
- file.DecRef()
+ file.DecRef(ctx)
filePathBuilder.WriteString(filename)
filePath := filePathBuilder.String()
@@ -370,24 +370,24 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
// Create VFS.
vfsObj := vfs.VirtualFilesystem{}
- if err := vfsObj.Init(); err != nil {
+ if err := vfsObj.Init(ctx); err != nil {
b.Fatalf("VFS init: %v", err)
}
vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
b.Fatalf("failed to create tmpfs root mount: %v", err)
}
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
var filePathBuilder strings.Builder
filePathBuilder.WriteByte('/')
// Create the mount point.
root := mntns.Root()
- defer root.DecRef()
+ defer root.DecRef(ctx)
pop := vfs.PathOperation{
Root: root,
Start: root,
@@ -403,9 +403,9 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to mount point: %v", err)
}
- defer mountPoint.DecRef()
+ defer mountPoint.DecRef(ctx)
// Create and mount the submount.
- if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil {
+ if _, err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil {
b.Fatalf("failed to mount tmpfs submount: %v", err)
}
filePathBuilder.WriteString(mountPointName)
@@ -432,7 +432,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
if err != nil {
b.Fatalf("failed to walk to directory %q: %v", name, err)
}
- vd.DecRef()
+ vd.DecRef(ctx)
vd = nextVD
filePathBuilder.WriteString(name)
filePathBuilder.WriteByte('/')
@@ -448,11 +448,11 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) {
Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL,
Mode: 0644,
})
- vd.DecRef()
+ vd.DecRef(ctx)
if err != nil {
b.Fatalf("failed to create file %q: %v", filename, err)
}
- fd.DecRef()
+ fd.DecRef(ctx)
filePathBuilder.WriteString(filename)
filePath := filePathBuilder.String()
diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go
index ac54d420d..9129d35b7 100644
--- a/pkg/sentry/fsimpl/tmpfs/device_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/device_file.go
@@ -22,6 +22,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/vfs"
)
+// +stateify savable
type deviceFile struct {
inode inode
kind vfs.DeviceKind
diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go
index 0a1ad4765..e90669cf0 100644
--- a/pkg/sentry/fsimpl/tmpfs/directory.go
+++ b/pkg/sentry/fsimpl/tmpfs/directory.go
@@ -25,6 +25,7 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
)
+// +stateify savable
type directory struct {
// Since directories can't be hard-linked, each directory can only be
// associated with a single dentry, which we can store in the directory
@@ -44,7 +45,7 @@ type directory struct {
// (with inode == nil) that represent the iteration position of
// directoryFDs. childList is used to support directoryFD.IterDirents()
// efficiently. childList is protected by iterMu.
- iterMu sync.Mutex
+ iterMu sync.Mutex `state:"nosave"`
childList dentryList
}
@@ -57,8 +58,9 @@ func (fs *filesystem) newDirectory(kuid auth.KUID, kgid auth.KGID, mode linux.Fi
return dir
}
-// Preconditions: filesystem.mu must be locked for writing. dir must not
-// already contain a child with the given name.
+// Preconditions:
+// * filesystem.mu must be locked for writing.
+// * dir must not already contain a child with the given name.
func (dir *directory) insertChildLocked(child *dentry, name string) {
child.parent = &dir.dentry
child.name = name
@@ -85,6 +87,7 @@ func (dir *directory) mayDelete(creds *auth.Credentials, child *dentry) error {
return vfs.CheckDeleteSticky(creds, linux.FileMode(atomic.LoadUint32(&dir.inode.mode)), auth.KUID(atomic.LoadUint32(&child.inode.uid)))
}
+// +stateify savable
type directoryFD struct {
fileDescription
vfs.DirectoryFileDescriptionDefaultImpl
@@ -95,7 +98,7 @@ type directoryFD struct {
}
// Release implements vfs.FileDescriptionImpl.Release.
-func (fd *directoryFD) Release() {
+func (fd *directoryFD) Release(ctx context.Context) {
if fd.iter != nil {
dir := fd.inode().impl.(*directory)
dir.iterMu.Lock()
@@ -110,7 +113,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba
fs := fd.filesystem()
dir := fd.inode().impl.(*directory)
- defer fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ defer fd.dentry().InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent)
// fs.mu is required to read d.parent and dentry.name.
fs.mu.RLock()
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index ed40f6b52..e39cd305b 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -25,7 +25,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
- "gvisor.dev/gvisor/pkg/usermem"
)
// Sync implements vfs.FilesystemImpl.Sync.
@@ -39,8 +38,10 @@ func (fs *filesystem) Sync(ctx context.Context) error {
//
// stepLocked is loosely analogous to fs/namei.c:walk_component().
//
-// Preconditions: filesystem.mu must be locked. !rp.Done().
-func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) {
+// Preconditions:
+// * filesystem.mu must be locked.
+// * !rp.Done().
+func stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*dentry, error) {
dir, ok := d.inode.impl.(*directory)
if !ok {
return nil, syserror.ENOTDIR
@@ -55,13 +56,13 @@ afterSymlink:
return d, nil
}
if name == ".." {
- if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil {
+ if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil {
return nil, err
} else if isRoot || d.parent == nil {
rp.Advance()
return d, nil
}
- if err := rp.CheckMount(&d.parent.vfsd); err != nil {
+ if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil {
return nil, err
}
rp.Advance()
@@ -74,7 +75,7 @@ afterSymlink:
if !ok {
return nil, syserror.ENOENT
}
- if err := rp.CheckMount(&child.vfsd); err != nil {
+ if err := rp.CheckMount(ctx, &child.vfsd); err != nil {
return nil, err
}
if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() {
@@ -97,10 +98,12 @@ afterSymlink:
// walkParentDirLocked is loosely analogous to Linux's
// fs/namei.c:path_parentat().
//
-// Preconditions: filesystem.mu must be locked. !rp.Done().
-func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) {
+// Preconditions:
+// * filesystem.mu must be locked.
+// * !rp.Done().
+func walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*directory, error) {
for !rp.Final() {
- next, err := stepLocked(rp, d)
+ next, err := stepLocked(ctx, rp, d)
if err != nil {
return nil, err
}
@@ -118,10 +121,10 @@ func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) {
// resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat().
//
// Preconditions: filesystem.mu must be locked.
-func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) {
+func resolveLocked(ctx context.Context, rp *vfs.ResolvingPath) (*dentry, error) {
d := rp.Start().Impl().(*dentry)
for !rp.Done() {
- next, err := stepLocked(rp, d)
+ next, err := stepLocked(ctx, rp, d)
if err != nil {
return nil, err
}
@@ -139,12 +142,13 @@ func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) {
// doCreateAt is loosely analogous to a conjunction of Linux's
// fs/namei.c:filename_create() and done_path_create().
//
-// Preconditions: !rp.Done(). For the final path component in rp,
-// !rp.ShouldFollowSymlink().
-func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error {
+// Preconditions:
+// * !rp.Done().
+// * For the final path component in rp, !rp.ShouldFollowSymlink().
+func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error {
fs.mu.Lock()
defer fs.mu.Unlock()
- parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry))
+ parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry))
if err != nil {
return err
}
@@ -182,7 +186,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa
if dir {
ev |= linux.IN_ISDIR
}
- parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */)
+ parentDir.inode.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */)
parentDir.inode.touchCMtime()
return nil
}
@@ -191,7 +195,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa
func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return err
}
@@ -202,7 +206,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds
func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return nil, err
}
@@ -222,7 +226,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op
func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- dir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry))
+ dir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry))
if err != nil {
return nil, err
}
@@ -232,7 +236,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa
// LinkAt implements vfs.FilesystemImpl.LinkAt.
func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error {
- return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error {
+ return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error {
if rp.Mount() != vd.Mount() {
return syserror.EXDEV
}
@@ -251,7 +255,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.
return syserror.EMLINK
}
i.incLinksLocked()
- i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */)
+ i.watches.Notify(ctx, "", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */)
parentDir.insertChildLocked(fs.newDentry(i), name)
return nil
})
@@ -259,7 +263,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.
// MkdirAt implements vfs.FilesystemImpl.MkdirAt.
func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error {
- return fs.doCreateAt(rp, true /* dir */, func(parentDir *directory, name string) error {
+ return fs.doCreateAt(ctx, rp, true /* dir */, func(parentDir *directory, name string) error {
creds := rp.Credentials()
if parentDir.inode.nlink == maxLinks {
return syserror.EMLINK
@@ -273,11 +277,11 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v
// MknodAt implements vfs.FilesystemImpl.MknodAt.
func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error {
- return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error {
+ return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error {
creds := rp.Credentials()
var childInode *inode
switch opts.Mode.FileType() {
- case 0, linux.S_IFREG:
+ case linux.S_IFREG:
childInode = fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)
case linux.S_IFIFO:
childInode = fs.newNamedPipe(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)
@@ -307,30 +311,43 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf
// don't need fs.mu for writing.
if opts.Flags&linux.O_CREAT == 0 {
fs.mu.RLock()
- defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
+ fs.mu.RUnlock()
return nil, err
}
+ d.IncRef()
+ defer d.DecRef(ctx)
+ fs.mu.RUnlock()
return d.open(ctx, rp, &opts, false /* afterCreate */)
}
mustCreate := opts.Flags&linux.O_EXCL != 0
start := rp.Start().Impl().(*dentry)
fs.mu.Lock()
- defer fs.mu.Unlock()
+ unlocked := false
+ unlock := func() {
+ if !unlocked {
+ fs.mu.Unlock()
+ unlocked = true
+ }
+ }
+ defer unlock()
if rp.Done() {
- // Reject attempts to open directories with O_CREAT.
+ // Reject attempts to open mount root directory with O_CREAT.
if rp.MustBeDir() {
return nil, syserror.EISDIR
}
if mustCreate {
return nil, syserror.EEXIST
}
+ start.IncRef()
+ defer start.DecRef(ctx)
+ unlock()
return start.open(ctx, rp, &opts, false /* afterCreate */)
}
afterTrailingSymlink:
- parentDir, err := walkParentDirLocked(rp, start)
+ parentDir, err := walkParentDirLocked(ctx, rp, start)
if err != nil {
return nil, err
}
@@ -364,11 +381,12 @@ afterTrailingSymlink:
creds := rp.Credentials()
child := fs.newDentry(fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode))
parentDir.insertChildLocked(child, name)
+ unlock()
fd, err := child.open(ctx, rp, &opts, true)
if err != nil {
return nil, err
}
- parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */)
+ parentDir.inode.watches.Notify(ctx, name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */)
parentDir.inode.touchCMtime()
return fd, nil
}
@@ -376,7 +394,7 @@ afterTrailingSymlink:
return nil, syserror.EEXIST
}
// Is the file mounted over?
- if err := rp.CheckMount(&child.vfsd); err != nil {
+ if err := rp.CheckMount(ctx, &child.vfsd); err != nil {
return nil, err
}
// Do we need to resolve a trailing symlink?
@@ -389,13 +407,17 @@ afterTrailingSymlink:
start = &parentDir.dentry
goto afterTrailingSymlink
}
- // Open existing file.
- if mustCreate {
- return nil, syserror.EEXIST
+ if rp.MustBeDir() && !child.inode.isDir() {
+ return nil, syserror.ENOTDIR
}
+ child.IncRef()
+ defer child.DecRef(ctx)
+ unlock()
return child.open(ctx, rp, &opts, false)
}
+// Preconditions: The caller must hold no locks (since opening pipes may block
+// indefinitely).
func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, afterCreate bool) (*vfs.FileDescription, error) {
ats := vfs.AccessTypesForOpenFlags(opts)
if !afterCreate {
@@ -445,7 +467,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open
func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return "", err
}
@@ -467,7 +489,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
// Resolve newParent first to verify that it's on this Mount.
fs.mu.Lock()
defer fs.mu.Unlock()
- newParentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry))
+ newParentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry))
if err != nil {
return err
}
@@ -555,7 +577,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
}
vfsObj := rp.VirtualFilesystem()
mntns := vfs.MountNamespaceFromContext(ctx)
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
var replacedVFSD *vfs.Dentry
if replaced != nil {
replacedVFSD = &replaced.vfsd
@@ -566,17 +588,19 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
if replaced != nil {
newParentDir.removeChildLocked(replaced)
if replaced.inode.isDir() {
- newParentDir.inode.decLinksLocked() // from replaced's ".."
+ // Remove links for replaced/. and replaced/..
+ replaced.inode.decLinksLocked(ctx)
+ newParentDir.inode.decLinksLocked(ctx)
}
- replaced.inode.decLinksLocked()
+ replaced.inode.decLinksLocked(ctx)
}
oldParentDir.removeChildLocked(renamed)
newParentDir.insertChildLocked(renamed, newName)
- vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD)
+ vfsObj.CommitRenameReplaceDentry(ctx, &renamed.vfsd, replacedVFSD)
oldParentDir.inode.touchCMtime()
if oldParentDir != newParentDir {
if renamed.inode.isDir() {
- oldParentDir.inode.decLinksLocked()
+ oldParentDir.inode.decLinksLocked(ctx)
newParentDir.inode.incLinksLocked()
}
newParentDir.inode.touchCMtime()
@@ -591,7 +615,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error {
fs.mu.Lock()
defer fs.mu.Unlock()
- parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry))
+ parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry))
if err != nil {
return err
}
@@ -626,17 +650,17 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
defer mnt.EndWrite()
vfsObj := rp.VirtualFilesystem()
mntns := vfs.MountNamespaceFromContext(ctx)
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil {
return err
}
parentDir.removeChildLocked(child)
- parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */)
+ parentDir.inode.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */)
// Remove links for child, child/., and child/..
- child.inode.decLinksLocked()
- child.inode.decLinksLocked()
- parentDir.inode.decLinksLocked()
- vfsObj.CommitDeleteDentry(&child.vfsd)
+ child.inode.decLinksLocked(ctx)
+ child.inode.decLinksLocked(ctx)
+ parentDir.inode.decLinksLocked(ctx)
+ vfsObj.CommitDeleteDentry(ctx, &child.vfsd)
parentDir.inode.touchCMtime()
return nil
}
@@ -644,19 +668,19 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
// SetStatAt implements vfs.FilesystemImpl.SetStatAt.
func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error {
fs.mu.RLock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
fs.mu.RUnlock()
return err
}
- if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil {
- fs.mu.RUnlock()
+ err = d.inode.setStat(ctx, rp.Credentials(), &opts)
+ fs.mu.RUnlock()
+ if err != nil {
return err
}
- fs.mu.RUnlock()
if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
- d.InotifyWithParent(ev, 0, vfs.InodeEvent)
+ d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent)
}
return nil
}
@@ -665,7 +689,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts
func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return linux.Statx{}, err
}
@@ -678,24 +702,15 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf
func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- if _, err := resolveLocked(rp); err != nil {
+ if _, err := resolveLocked(ctx, rp); err != nil {
return linux.Statfs{}, err
}
- statfs := linux.Statfs{
- Type: linux.TMPFS_MAGIC,
- BlockSize: usermem.PageSize,
- FragmentSize: usermem.PageSize,
- NameLength: linux.NAME_MAX,
- // TODO(b/29637826): Allow configuring a tmpfs size and enforce it.
- Blocks: 0,
- BlocksFree: 0,
- }
- return statfs, nil
+ return globalStatfs, nil
}
// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt.
func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error {
- return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error {
+ return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error {
creds := rp.Credentials()
child := fs.newDentry(fs.newSymlink(creds.EffectiveKUID, creds.EffectiveKGID, 0777, target))
parentDir.insertChildLocked(child, name)
@@ -707,7 +722,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ
func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error {
fs.mu.Lock()
defer fs.mu.Unlock()
- parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry))
+ parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry))
if err != nil {
return err
}
@@ -738,7 +753,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
defer mnt.EndWrite()
vfsObj := rp.VirtualFilesystem()
mntns := vfs.MountNamespaceFromContext(ctx)
- defer mntns.DecRef()
+ defer mntns.DecRef(ctx)
if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil {
return err
}
@@ -746,20 +761,20 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
// Generate inotify events. Note that this must take place before the link
// count of the child is decremented, or else the watches may be dropped
// before these events are added.
- vfs.InotifyRemoveChild(&child.inode.watches, &parentDir.inode.watches, name)
+ vfs.InotifyRemoveChild(ctx, &child.inode.watches, &parentDir.inode.watches, name)
parentDir.removeChildLocked(child)
- child.inode.decLinksLocked()
- vfsObj.CommitDeleteDentry(&child.vfsd)
+ child.inode.decLinksLocked(ctx)
+ vfsObj.CommitDeleteDentry(ctx, &child.vfsd)
parentDir.inode.touchCMtime()
return nil
}
-// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+// BoundEndpointAt implements vfs.FilesystemImpl.BoundEndpointAt.
func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return nil, err
}
@@ -768,67 +783,70 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath
}
switch impl := d.inode.impl.(type) {
case *socketFile:
+ if impl.ep == nil {
+ return nil, syserror.ECONNREFUSED
+ }
return impl.ep, nil
default:
return nil, syserror.ECONNREFUSED
}
}
-// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
-func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
+// ListXattrAt implements vfs.FilesystemImpl.ListXattrAt.
+func (fs *filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return nil, err
}
- return d.inode.listxattr(size)
+ return d.inode.listXattr(size)
}
-// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt.
-func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) {
+// GetXattrAt implements vfs.FilesystemImpl.GetXattrAt.
+func (fs *filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetXattrOptions) (string, error) {
fs.mu.RLock()
defer fs.mu.RUnlock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
return "", err
}
- return d.inode.getxattr(rp.Credentials(), &opts)
+ 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 {
+// SetXattrAt implements vfs.FilesystemImpl.SetXattrAt.
+func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error {
fs.mu.RLock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
fs.mu.RUnlock()
return err
}
- if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil {
- fs.mu.RUnlock()
+ err = d.inode.setXattr(rp.Credentials(), &opts)
+ fs.mu.RUnlock()
+ if err != nil {
return err
}
- fs.mu.RUnlock()
- d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent)
+ d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
}
-// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt.
-func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error {
+// RemoveXattrAt implements vfs.FilesystemImpl.RemoveXattrAt.
+func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error {
fs.mu.RLock()
- d, err := resolveLocked(rp)
+ d, err := resolveLocked(ctx, rp)
if err != nil {
fs.mu.RUnlock()
return err
}
- if err := d.inode.removexattr(rp.Credentials(), name); err != nil {
- fs.mu.RUnlock()
+ err = d.inode.removeXattr(rp.Credentials(), name)
+ fs.mu.RUnlock()
+ if err != nil {
return err
}
- fs.mu.RUnlock()
- d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent)
+ d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
}
@@ -847,8 +865,16 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe
}
if d.parent == nil {
if d.name != "" {
- // This must be an anonymous memfd file.
+ // This file must have been created by
+ // newUnlinkedRegularFileDescription(). In Linux,
+ // mm/shmem.c:__shmem_file_setup() =>
+ // fs/file_table.c:alloc_file_pseudo() sets the created
+ // dentry's dentry_operations to anon_ops, for which d_dname ==
+ // simple_dname. fs/d_path.c:simple_dname() defines the
+ // dentry's pathname to be its name, prefixed with "/" and
+ // suffixed with " (deleted)".
b.PrependComponent("/" + d.name)
+ b.AppendString(" (deleted)")
return vfs.PrependPathSyntheticError{}
}
return vfs.PrependPathAtNonMountRootError{}
diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go
index 739350cf0..d772db9e9 100644
--- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go
+++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go
@@ -21,6 +21,7 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
+// +stateify savable
type namedPipe struct {
inode inode
@@ -28,8 +29,8 @@ type namedPipe struct {
}
// Preconditions:
-// * fs.mu must be locked.
-// * rp.Mount().CheckBeginWrite() has been called successfully.
+// * fs.mu must be locked.
+// * rp.Mount().CheckBeginWrite() has been called successfully.
func (fs *filesystem) newNamedPipe(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode {
file := &namedPipe{pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize)}
file.inode.init(file, fs, kuid, kgid, linux.S_IFIFO|mode)
diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go
index 1614f2c39..be29a2363 100644
--- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go
@@ -32,7 +32,7 @@ const fileName = "mypipe"
func TestSeparateFDs(t *testing.T) {
ctx, creds, vfsObj, root := setup(t)
- defer root.DecRef()
+ defer root.DecRef(ctx)
// Open the read side. This is done in a concurrently because opening
// One end the pipe blocks until the other end is opened.
@@ -55,13 +55,13 @@ func TestSeparateFDs(t *testing.T) {
if err != nil {
t.Fatalf("failed to open pipe for writing %q: %v", fileName, err)
}
- defer wfd.DecRef()
+ defer wfd.DecRef(ctx)
rfd, ok := <-rfdchan
if !ok {
t.Fatalf("failed to open pipe for reading %q", fileName)
}
- defer rfd.DecRef()
+ defer rfd.DecRef(ctx)
const msg = "vamos azul"
checkEmpty(ctx, t, rfd)
@@ -71,7 +71,7 @@ func TestSeparateFDs(t *testing.T) {
func TestNonblockingRead(t *testing.T) {
ctx, creds, vfsObj, root := setup(t)
- defer root.DecRef()
+ defer root.DecRef(ctx)
// Open the read side as nonblocking.
pop := vfs.PathOperation{
@@ -85,7 +85,7 @@ func TestNonblockingRead(t *testing.T) {
if err != nil {
t.Fatalf("failed to open pipe for reading %q: %v", fileName, err)
}
- defer rfd.DecRef()
+ defer rfd.DecRef(ctx)
// Open the write side.
openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY}
@@ -93,7 +93,7 @@ func TestNonblockingRead(t *testing.T) {
if err != nil {
t.Fatalf("failed to open pipe for writing %q: %v", fileName, err)
}
- defer wfd.DecRef()
+ defer wfd.DecRef(ctx)
const msg = "geh blau"
checkEmpty(ctx, t, rfd)
@@ -103,7 +103,7 @@ func TestNonblockingRead(t *testing.T) {
func TestNonblockingWriteError(t *testing.T) {
ctx, creds, vfsObj, root := setup(t)
- defer root.DecRef()
+ defer root.DecRef(ctx)
// Open the write side as nonblocking, which should return ENXIO.
pop := vfs.PathOperation{
@@ -121,7 +121,7 @@ func TestNonblockingWriteError(t *testing.T) {
func TestSingleFD(t *testing.T) {
ctx, creds, vfsObj, root := setup(t)
- defer root.DecRef()
+ defer root.DecRef(ctx)
// Open the pipe as readable and writable.
pop := vfs.PathOperation{
@@ -135,7 +135,7 @@ func TestSingleFD(t *testing.T) {
if err != nil {
t.Fatalf("failed to open pipe for writing %q: %v", fileName, err)
}
- defer fd.DecRef()
+ defer fd.DecRef(ctx)
const msg = "forza blu"
checkEmpty(ctx, t, fd)
@@ -152,13 +152,13 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy
// Create VFS.
vfsObj := &vfs.VirtualFilesystem{}
- if err := vfsObj.Init(); err != nil {
+ if err := vfsObj.Init(ctx); err != nil {
t.Fatalf("VFS init: %v", err)
}
vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
t.Fatalf("failed to create tmpfs root mount: %v", err)
}
diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go
index 1cdb46e6f..a199eb33d 100644
--- a/pkg/sentry/fsimpl/tmpfs/regular_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go
@@ -36,12 +36,18 @@ import (
)
// regularFile is a regular (=S_IFREG) tmpfs file.
+//
+// +stateify savable
type regularFile struct {
inode inode
// memFile is a platform.File used to allocate pages to this regularFile.
memFile *pgalloc.MemoryFile
+ // memoryUsageKind is the memory accounting category under which pages backing
+ // this regularFile's contents are accounted.
+ memoryUsageKind usage.MemoryKind
+
// mapsMu protects mappings.
mapsMu sync.Mutex `state:"nosave"`
@@ -62,7 +68,7 @@ type regularFile struct {
writableMappingPages uint64
// dataMu protects the fields below.
- dataMu sync.RWMutex
+ dataMu sync.RWMutex `state:"nosave"`
// data maps offsets into the file to offsets into memFile that store
// the file's data.
@@ -86,14 +92,75 @@ type regularFile struct {
func (fs *filesystem) newRegularFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode {
file := &regularFile{
- memFile: fs.memFile,
- seals: linux.F_SEAL_SEAL,
+ memFile: fs.memFile,
+ memoryUsageKind: usage.Tmpfs,
+ seals: linux.F_SEAL_SEAL,
}
file.inode.init(file, fs, kuid, kgid, linux.S_IFREG|mode)
file.inode.nlink = 1 // from parent directory
return &file.inode
}
+// newUnlinkedRegularFileDescription creates a regular file on the tmpfs
+// filesystem represented by mount and returns an FD representing that file.
+// The new file is not reachable by path traversal from any other file.
+//
+// newUnlinkedRegularFileDescription is analogous to Linux's
+// mm/shmem.c:__shmem_file_setup().
+//
+// Preconditions: mount must be a tmpfs mount.
+func newUnlinkedRegularFileDescription(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, name string) (*regularFileFD, error) {
+ fs, ok := mount.Filesystem().Impl().(*filesystem)
+ if !ok {
+ panic("tmpfs.newUnlinkedRegularFileDescription() called with non-tmpfs mount")
+ }
+
+ inode := fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, 0777)
+ d := fs.newDentry(inode)
+ defer d.DecRef(ctx)
+ d.name = name
+
+ fd := &regularFileFD{}
+ fd.Init(&inode.locks)
+ flags := uint32(linux.O_RDWR)
+ if err := fd.vfsfd.Init(fd, flags, mount, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+ return nil, err
+ }
+ return fd, nil
+}
+
+// NewZeroFile creates a new regular file and file description as for
+// mmap(MAP_SHARED | MAP_ANONYMOUS). The file has the given size and is
+// initially (implicitly) filled with zeroes.
+//
+// Preconditions: mount must be a tmpfs mount.
+func NewZeroFile(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, size uint64) (*vfs.FileDescription, error) {
+ // Compare mm/shmem.c:shmem_zero_setup().
+ fd, err := newUnlinkedRegularFileDescription(ctx, creds, mount, "dev/zero")
+ if err != nil {
+ return nil, err
+ }
+ rf := fd.inode().impl.(*regularFile)
+ rf.memoryUsageKind = usage.Anonymous
+ rf.size = size
+ return &fd.vfsfd, err
+}
+
+// NewMemfd creates a new regular file and file description as for
+// memfd_create.
+//
+// Preconditions: mount must be a tmpfs mount.
+func NewMemfd(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, allowSeals bool, name string) (*vfs.FileDescription, error) {
+ fd, err := newUnlinkedRegularFileDescription(ctx, creds, mount, name)
+ if err != nil {
+ return nil, err
+ }
+ if allowSeals {
+ fd.inode().impl.(*regularFile).seals = 0
+ }
+ return &fd.vfsfd, nil
+}
+
// truncate grows or shrinks the file to the given size. It returns true if the
// file size was updated.
func (rf *regularFile) truncate(newSize uint64) (bool, error) {
@@ -226,7 +293,7 @@ func (rf *regularFile) Translate(ctx context.Context, required, optional memmap.
optional.End = pgend
}
- cerr := rf.data.Fill(ctx, required, optional, rf.memFile, usage.Tmpfs, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) {
+ cerr := rf.data.Fill(ctx, required, optional, rf.memFile, rf.memoryUsageKind, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) {
// Newly-allocated pages are zeroed, so we don't need to do anything.
return dsts.NumBytes(), nil
})
@@ -260,17 +327,18 @@ func (*regularFile) InvalidateUnsavable(context.Context) error {
return nil
}
+// +stateify savable
type regularFileFD struct {
fileDescription
// off is the file offset. off is accessed using atomic memory operations.
// offMu serializes operations that may mutate off.
off int64
- offMu sync.Mutex
+ offMu sync.Mutex `state:"nosave"`
}
// Release implements vfs.FileDescriptionImpl.Release.
-func (fd *regularFileFD) Release() {
+func (fd *regularFileFD) Release(context.Context) {
// noop
}
@@ -325,8 +393,15 @@ func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts
// PWrite implements vfs.FileDescriptionImpl.PWrite.
func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+ n, _, err := fd.pwrite(ctx, src, offset, opts)
+ return n, err
+}
+
+// pwrite returns the number of bytes written, final offset and error. The
+// final offset should be ignored by PWrite.
+func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) {
if offset < 0 {
- return 0, syserror.EINVAL
+ return 0, offset, syserror.EINVAL
}
// Check that flags are supported. RWF_DSYNC/RWF_SYNC can be ignored since
@@ -334,40 +409,44 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
//
// TODO(gvisor.dev/issue/2601): Support select preadv2 flags.
if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 {
- return 0, syserror.EOPNOTSUPP
+ return 0, offset, syserror.EOPNOTSUPP
}
srclen := src.NumBytes()
if srclen == 0 {
- return 0, nil
+ return 0, offset, nil
}
f := fd.inode().impl.(*regularFile)
+ f.inode.mu.Lock()
+ defer f.inode.mu.Unlock()
+ // If the file is opened with O_APPEND, update offset to file size.
+ if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 {
+ // Locking f.inode.mu is sufficient for reading f.size.
+ offset = int64(f.size)
+ }
if end := offset + srclen; end < offset {
// Overflow.
- return 0, syserror.EINVAL
+ return 0, offset, syserror.EINVAL
}
- var err error
srclen, err = vfs.CheckLimit(ctx, offset, srclen)
if err != nil {
- return 0, err
+ return 0, offset, err
}
src = src.TakeFirst64(srclen)
- f.inode.mu.Lock()
rw := getRegularFileReadWriter(f, offset)
n, err := src.CopyInTo(ctx, rw)
- fd.inode().touchCMtimeLocked()
- f.inode.mu.Unlock()
+ f.inode.touchCMtimeLocked()
putRegularFileReadWriter(rw)
- return n, err
+ return n, n + offset, err
}
// Write implements vfs.FileDescriptionImpl.Write.
func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
fd.offMu.Lock()
- n, err := fd.PWrite(ctx, src, fd.off, opts)
- fd.off += n
+ n, off, err := fd.pwrite(ctx, src, fd.off, opts)
+ fd.off = off
fd.offMu.Unlock()
return n, err
}
@@ -564,7 +643,7 @@ func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64,
case gap.Ok():
// Allocate memory for the write.
gapMR := gap.Range().Intersect(pgMR)
- fr, err := rw.file.memFile.Allocate(gapMR.Length(), usage.Tmpfs)
+ fr, err := rw.file.memFile.Allocate(gapMR.Length(), rw.file.memoryUsageKind)
if err != nil {
retErr = err
goto exitLoop
diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go
index 3ed650474..5699d5975 100644
--- a/pkg/sentry/fsimpl/tmpfs/socket_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go
@@ -21,6 +21,8 @@ import (
)
// socketFile is a socket (=S_IFSOCK) tmpfs file.
+//
+// +stateify savable
type socketFile struct {
inode inode
ep transport.BoundEndpoint
diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go
index b0de5fabe..a102a2ee2 100644
--- a/pkg/sentry/fsimpl/tmpfs/symlink.go
+++ b/pkg/sentry/fsimpl/tmpfs/symlink.go
@@ -19,6 +19,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
)
+// +stateify savable
type symlink struct {
inode inode
target string // immutable
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index d7f4f0779..cefec8fde 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -51,9 +51,13 @@ import (
const Name = "tmpfs"
// FilesystemType implements vfs.FilesystemType.
+//
+// +stateify savable
type FilesystemType struct{}
// filesystem implements vfs.FilesystemImpl.
+//
+// +stateify savable
type filesystem struct {
vfsfs vfs.Filesystem
@@ -67,7 +71,7 @@ type filesystem struct {
devMinor uint32
// mu serializes changes to the Dentry tree.
- mu sync.RWMutex
+ mu sync.RWMutex `state:"nosave"`
nextInoMinusOne uint64 // accessed using atomic memory operations
}
@@ -78,6 +82,8 @@ func (FilesystemType) Name() string {
}
// FilesystemOpts is used to pass configuration data to tmpfs.
+//
+// +stateify savable
type FilesystemOpts struct {
// RootFileType is the FileType of the filesystem root. Valid values
// are: S_IFDIR, S_IFREG, and S_IFLNK. Defaults to S_IFDIR.
@@ -185,7 +191,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
case linux.S_IFDIR:
root = &fs.newDirectory(rootKUID, rootKGID, rootMode).dentry
default:
- fs.vfsfs.DecRef()
+ fs.vfsfs.DecRef(ctx)
return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType)
}
return &fs.vfsfs, &root.vfsd, nil
@@ -197,11 +203,32 @@ func NewFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *au
}
// Release implements vfs.FilesystemImpl.Release.
-func (fs *filesystem) Release() {
+func (fs *filesystem) Release(ctx context.Context) {
fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor)
}
+// immutable
+var globalStatfs = linux.Statfs{
+ Type: linux.TMPFS_MAGIC,
+ BlockSize: usermem.PageSize,
+ FragmentSize: usermem.PageSize,
+ NameLength: linux.NAME_MAX,
+
+ // tmpfs currently does not support configurable size limits. In Linux,
+ // such a tmpfs mount will return f_blocks == f_bfree == f_bavail == 0 from
+ // statfs(2). However, many applications treat this as having a size limit
+ // of 0. To work around this, claim to have a very large but non-zero size,
+ // chosen to ensure that BlockSize * Blocks does not overflow int64 (which
+ // applications may also handle incorrectly).
+ // TODO(b/29637826): allow configuring a tmpfs size and enforce it.
+ Blocks: math.MaxInt64 / usermem.PageSize,
+ BlocksFree: math.MaxInt64 / usermem.PageSize,
+ BlocksAvailable: math.MaxInt64 / usermem.PageSize,
+}
+
// dentry implements vfs.DentryImpl.
+//
+// +stateify savable
type dentry struct {
vfsd vfs.Dentry
@@ -249,12 +276,12 @@ func (d *dentry) TryIncRef() bool {
}
// DecRef implements vfs.DentryImpl.DecRef.
-func (d *dentry) DecRef() {
- d.inode.decRef()
+func (d *dentry) DecRef(ctx context.Context) {
+ d.inode.decRef(ctx)
}
// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent.
-func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {
+func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) {
if d.inode.isDir() {
events |= linux.IN_ISDIR
}
@@ -266,9 +293,9 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {
d.inode.fs.mu.RLock()
// The ordering below is important, Linux always notifies the parent first.
if d.parent != nil {
- d.parent.inode.watches.Notify(d.name, events, cookie, et, deleted)
+ d.parent.inode.watches.Notify(ctx, d.name, events, cookie, et, deleted)
}
- d.inode.watches.Notify("", events, cookie, et, deleted)
+ d.inode.watches.Notify(ctx, "", events, cookie, et, deleted)
d.inode.fs.mu.RUnlock()
}
@@ -278,20 +305,19 @@ func (d *dentry) Watches() *vfs.Watches {
}
// OnZeroWatches implements vfs.Dentry.OnZeroWatches.
-func (d *dentry) OnZeroWatches() {}
+func (d *dentry) OnZeroWatches(context.Context) {}
// inode represents a filesystem object.
+//
+// +stateify savable
type inode struct {
// fs is the owning filesystem. fs is immutable.
fs *filesystem
- // refs is a reference count. refs is accessed using atomic memory
- // operations.
- //
// A reference is held on all inodes as long as they are reachable in the
// filesystem tree, i.e. nlink is nonzero. This reference is dropped when
// nlink reaches 0.
- refs int64
+ refs inodeRefs
// xattrs implements extended attributes.
//
@@ -300,12 +326,12 @@ type inode struct {
// Inode metadata. Writing multiple fields atomically requires holding
// mu, othewise atomic operations can be used.
- mu sync.Mutex
- mode uint32 // file type and mode
- nlink uint32 // protected by filesystem.mu instead of inode.mu
- uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic
- gid uint32 // auth.KGID, but ...
- ino uint64 // immutable
+ mu sync.Mutex `state:"nosave"`
+ mode uint32 // file type and mode
+ nlink uint32 // protected by filesystem.mu instead of inode.mu
+ uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic
+ gid uint32 // auth.KGID, but ...
+ ino uint64 // immutable
// Linux's tmpfs has no concept of btime.
atime int64 // nanoseconds
@@ -327,7 +353,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth
panic("file type is required in FileMode")
}
i.fs = fs
- i.refs = 1
i.mode = uint32(mode)
i.uid = uint32(kuid)
i.gid = uint32(kgid)
@@ -339,12 +364,15 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth
i.mtime = now
// i.nlink initialized by caller
i.impl = impl
+ i.refs.EnableLeakCheck()
}
// incLinksLocked increments i's link count.
//
-// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0.
-// i.nlink < maxLinks.
+// Preconditions:
+// * filesystem.mu must be locked for writing.
+// * i.nlink != 0.
+// * i.nlink < maxLinks.
func (i *inode) incLinksLocked() {
if i.nlink == 0 {
panic("tmpfs.inode.incLinksLocked() called with no existing links")
@@ -358,46 +386,36 @@ func (i *inode) incLinksLocked() {
// decLinksLocked decrements i's link count. If the link count reaches 0, we
// remove a reference on i as well.
//
-// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0.
-func (i *inode) decLinksLocked() {
+// Preconditions:
+// * filesystem.mu must be locked for writing.
+// * i.nlink != 0.
+func (i *inode) decLinksLocked(ctx context.Context) {
if i.nlink == 0 {
panic("tmpfs.inode.decLinksLocked() called with no existing links")
}
if atomic.AddUint32(&i.nlink, ^uint32(0)) == 0 {
- i.decRef()
+ i.decRef(ctx)
}
}
func (i *inode) incRef() {
- if atomic.AddInt64(&i.refs, 1) <= 1 {
- panic("tmpfs.inode.incRef() called without holding a reference")
- }
+ i.refs.IncRef()
}
func (i *inode) tryIncRef() bool {
- for {
- refs := atomic.LoadInt64(&i.refs)
- if refs == 0 {
- return false
- }
- if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) {
- return true
- }
- }
+ return i.refs.TryIncRef()
}
-func (i *inode) decRef() {
- if refs := atomic.AddInt64(&i.refs, -1); refs == 0 {
- i.watches.HandleDeletion()
+func (i *inode) decRef(ctx context.Context) {
+ i.refs.DecRef(func() {
+ i.watches.HandleDeletion(ctx)
if regFile, ok := i.impl.(*regularFile); ok {
// Release memory used by regFile to store data. Since regFile is
// no longer usable, we don't need to grab any locks or update any
// metadata.
regFile.data.DropAll(regFile.memFile)
}
- } else if refs < 0 {
- panic("tmpfs.inode.decRef() called without holding a reference")
- }
+ })
}
func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error {
@@ -452,7 +470,8 @@ func (i *inode) statTo(stat *linux.Statx) {
}
}
-func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx) error {
+func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs.SetStatOptions) error {
+ stat := &opts.Stat
if stat.Mask == 0 {
return nil
}
@@ -460,7 +479,7 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu
return syserror.EPERM
}
mode := linux.FileMode(atomic.LoadUint32(&i.mode))
- if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil {
+ if err := vfs.CheckSetStat(ctx, creds, opts, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil {
return err
}
i.mu.Lock()
@@ -557,6 +576,8 @@ func (i *inode) direntType() uint8 {
return linux.DT_LNK
case *socketFile:
return linux.DT_SOCK
+ case *namedPipe:
+ return linux.DT_FIFO
case *deviceFile:
switch impl.kind {
case vfs.BlockDevice:
@@ -606,66 +627,59 @@ func (i *inode) touchCMtime() {
i.mu.Unlock()
}
-// Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds
-// inode.mu.
+// Preconditions:
+// * The caller has called vfs.Mount.CheckBeginWrite().
+// * inode.mu must be locked.
func (i *inode) touchCMtimeLocked() {
now := i.fs.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) 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 {
+func (i *inode) getXattr(creds *auth.Credentials, opts *vfs.GetXattrOptions) (string, error) {
+ if err := i.checkXattrPermissions(creds, opts.Name, 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)
+ 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 {
+func (i *inode) setXattr(creds *auth.Credentials, opts *vfs.SetXattrOptions) error {
+ if err := i.checkXattrPermissions(creds, opts.Name, 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)
+ return i.xattrs.SetXattr(opts)
}
-func (i *inode) removexattr(creds *auth.Credentials, name string) error {
- if err := i.checkPermissions(creds, vfs.MayWrite); err != nil {
+func (i *inode) removeXattr(creds *auth.Credentials, name string) error {
+ if err := i.checkXattrPermissions(creds, name, vfs.MayWrite); err != nil {
return err
}
- if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) {
+ return i.xattrs.RemoveXattr(name)
+}
+
+func (i *inode) checkXattrPermissions(creds *auth.Credentials, name string, ats vfs.AccessTypes) error {
+ // We currently only support extended attributes in the user.* and
+ // trusted.* namespaces. See b/148380782.
+ if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) && !strings.HasPrefix(name, linux.XATTR_TRUSTED_PREFIX) {
return syserror.EOPNOTSUPP
}
- if !i.userXattrSupported() {
- return syserror.EPERM
+ mode := linux.FileMode(atomic.LoadUint32(&i.mode))
+ kuid := auth.KUID(atomic.LoadUint32(&i.uid))
+ kgid := auth.KGID(atomic.LoadUint32(&i.gid))
+ if err := vfs.GenericCheckPermissions(creds, ats, mode, kuid, kgid); err != nil {
+ return err
}
- 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
+ return vfs.CheckXattrPermissions(creds, ats, mode, kuid, name)
}
// fileDescription is embedded by tmpfs implementations of
// vfs.FileDescriptionImpl.
+//
+// +stateify savable
type fileDescription struct {
vfsfd vfs.FileDescription
vfs.FileDescriptionDefaultImpl
@@ -695,81 +709,55 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu
func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error {
creds := auth.CredentialsFromContext(ctx)
d := fd.dentry()
- if err := d.inode.setStat(ctx, creds, &opts.Stat); err != nil {
+ if err := d.inode.setStat(ctx, creds, &opts); err != nil {
return err
}
if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
- d.InotifyWithParent(ev, 0, vfs.InodeEvent)
+ d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent)
}
return nil
}
-// Listxattr implements vfs.FileDescriptionImpl.Listxattr.
-func (fd *fileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) {
- return fd.inode().listxattr(size)
+// StatFS implements vfs.FileDescriptionImpl.StatFS.
+func (fd *fileDescription) StatFS(ctx context.Context) (linux.Statfs, error) {
+ return globalStatfs, nil
}
-// 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)
+// ListXattr implements vfs.FileDescriptionImpl.ListXattr.
+func (fd *fileDescription) ListXattr(ctx context.Context, size uint64) ([]string, error) {
+ return fd.inode().listXattr(size)
}
-// Setxattr implements vfs.FileDescriptionImpl.Setxattr.
-func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error {
+// 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 {
d := fd.dentry()
- if err := d.inode.setxattr(auth.CredentialsFromContext(ctx), &opts); err != nil {
+ if err := d.inode.setXattr(auth.CredentialsFromContext(ctx), &opts); err != nil {
return err
}
// Generate inotify events.
- d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent)
+ d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
}
-// Removexattr implements vfs.FileDescriptionImpl.Removexattr.
-func (fd *fileDescription) Removexattr(ctx context.Context, name string) error {
+// RemoveXattr implements vfs.FileDescriptionImpl.RemoveXattr.
+func (fd *fileDescription) RemoveXattr(ctx context.Context, name string) error {
d := fd.dentry()
- if err := d.inode.removexattr(auth.CredentialsFromContext(ctx), name); err != nil {
+ if err := d.inode.removeXattr(auth.CredentialsFromContext(ctx), name); err != nil {
return err
}
// Generate inotify events.
- d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent)
+ d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
}
-// NewMemfd creates a new tmpfs regular file and file description that can back
-// an anonymous fd created by memfd_create.
-func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name string) (*vfs.FileDescription, error) {
- fs, ok := mount.Filesystem().Impl().(*filesystem)
- if !ok {
- panic("NewMemfd() called with non-tmpfs mount")
- }
-
- // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd inodes are set up with
- // S_IRWXUGO.
- inode := fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, 0777)
- rf := inode.impl.(*regularFile)
- if allowSeals {
- rf.seals = 0
- }
-
- d := fs.newDentry(inode)
- defer d.DecRef()
- d.name = name
-
- // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd files are set up with
- // FMODE_READ | FMODE_WRITE.
- var fd regularFileFD
- fd.Init(&inode.locks)
- flags := uint32(linux.O_RDWR)
- if err := fd.vfsfd.Init(&fd, flags, mount, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
- return nil, err
- }
- return &fd.vfsfd, nil
-}
-
// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block)
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go
index a240fb276..99c8e3c0f 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go
@@ -34,21 +34,21 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr
creds := auth.CredentialsFromContext(ctx)
vfsObj := &vfs.VirtualFilesystem{}
- if err := vfsObj.Init(); err != nil {
+ if err := vfsObj.Init(ctx); err != nil {
return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err)
}
vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
})
- mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
+ mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{})
if err != nil {
return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err)
}
root := mntns.Root()
return vfsObj, root, func() {
- root.DecRef()
- mntns.DecRef()
+ root.DecRef(ctx)
+ mntns.DecRef(ctx)
}, nil
}
generated by cgit v1.2.3 (git 2.39.1) at 2025-02-06 15:33:49 +0000