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/device_file.go1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/directory.go4
-rw-r--r--pkg/sentry/fsimpl/tmpfs/filesystem.go30
-rw-r--r--pkg/sentry/fsimpl/tmpfs/named_pipe.go1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/regular_file.go80
-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.go57
8 files changed, 120 insertions, 56 deletions
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 070c75e68..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
}
@@ -86,6 +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
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index 61d925a73..e39cd305b 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -673,11 +673,11 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts
fs.mu.RUnlock()
return err
}
- if err := d.inode.setStat(ctx, rp.Credentials(), &opts); 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(ctx, ev, 0, vfs.InodeEvent)
@@ -770,7 +770,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
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()
@@ -822,11 +822,11 @@ func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt
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(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
@@ -840,11 +840,11 @@ func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath,
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(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent)
return nil
@@ -865,8 +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 5b0471ff4..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
diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go
index 0710b65db..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
})
@@ -260,13 +327,14 @@ 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.
@@ -575,7 +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 4871e55d3..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.
@@ -221,6 +227,8 @@ var globalStatfs = linux.Statfs{
}
// dentry implements vfs.DentryImpl.
+//
+// +stateify savable
type dentry struct {
vfsd vfs.Dentry
@@ -300,6 +308,8 @@ func (d *dentry) Watches() *vfs.Watches {
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
@@ -316,12 +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
@@ -668,6 +678,8 @@ func (i *inode) checkXattrPermissions(creds *auth.Credentials, name string, ats
// fileDescription is embedded by tmpfs implementations of
// vfs.FileDescriptionImpl.
+//
+// +stateify savable
type fileDescription struct {
vfsfd vfs.FileDescription
vfs.FileDescriptionDefaultImpl
@@ -746,37 +758,6 @@ func (fd *fileDescription) RemoveXattr(ctx context.Context, name string) error {
return nil
}
-// NewMemfd creates a new tmpfs regular file and file description that can back
-// an anonymous fd created by memfd_create.
-func NewMemfd(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, 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(ctx)
- 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)
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-25 21:58:59 +0000