summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/vfs
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/vfs')
-rw-r--r--pkg/sentry/vfs/BUILD1
-rw-r--r--pkg/sentry/vfs/README.md4
-rw-r--r--pkg/sentry/vfs/epoll.go1
-rw-r--r--pkg/sentry/vfs/file_description.go25
-rw-r--r--pkg/sentry/vfs/file_description_impl_util.go80
-rw-r--r--pkg/sentry/vfs/file_description_impl_util_test.go1
-rw-r--r--pkg/sentry/vfs/genericfstree/genericfstree.go3
-rw-r--r--pkg/sentry/vfs/inotify.go1
-rw-r--r--pkg/sentry/vfs/mount.go50
-rw-r--r--pkg/sentry/vfs/options.go4
-rw-r--r--pkg/sentry/vfs/vfs.go2
11 files changed, 128 insertions, 44 deletions
diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD
index 774cc66cc..16d9f3a28 100644
--- a/pkg/sentry/vfs/BUILD
+++ b/pkg/sentry/vfs/BUILD
@@ -72,6 +72,7 @@ go_library(
"//pkg/sentry/memmap",
"//pkg/sentry/socket/unix/transport",
"//pkg/sentry/uniqueid",
+ "//pkg/sentry/vfs/lock",
"//pkg/sync",
"//pkg/syserror",
"//pkg/usermem",
diff --git a/pkg/sentry/vfs/README.md b/pkg/sentry/vfs/README.md
index 9aa133bcb..66f3105bd 100644
--- a/pkg/sentry/vfs/README.md
+++ b/pkg/sentry/vfs/README.md
@@ -39,8 +39,8 @@ Mount references are held by:
- Mount: Each referenced Mount holds a reference on its parent, which is the
mount containing its mount point.
-- VirtualFilesystem: A reference is held on each Mount that has not been
- umounted.
+- VirtualFilesystem: A reference is held on each Mount that has been connected
+ to a mount point, but not yet umounted.
MountNamespace and FileDescription references are held by users of VFS. The
expectation is that each `kernel.Task` holds a reference on its corresponding
diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go
index 8297f964b..599c3131c 100644
--- a/pkg/sentry/vfs/epoll.go
+++ b/pkg/sentry/vfs/epoll.go
@@ -31,6 +31,7 @@ type EpollInstance struct {
vfsfd FileDescription
FileDescriptionDefaultImpl
DentryMetadataFileDescriptionImpl
+ NoLockFD
// q holds waiters on this EpollInstance.
q waiter.Queue
diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go
index bb294563d..97b9b18d7 100644
--- a/pkg/sentry/vfs/file_description.go
+++ b/pkg/sentry/vfs/file_description.go
@@ -73,6 +73,8 @@ type FileDescription struct {
// writable is analogous to Linux's FMODE_WRITE.
writable bool
+ usedLockBSD uint32
+
// impl is the FileDescriptionImpl associated with this Filesystem. impl is
// immutable. This should be the last field in FileDescription.
impl FileDescriptionImpl
@@ -175,6 +177,12 @@ func (fd *FileDescription) DecRef() {
}
ep.interestMu.Unlock()
}
+
+ // If BSD locks were used, release any lock that it may have acquired.
+ if atomic.LoadUint32(&fd.usedLockBSD) != 0 {
+ fd.impl.UnlockBSD(context.Background(), fd)
+ }
+
// Release implementation resources.
fd.impl.Release()
if fd.writable {
@@ -420,13 +428,9 @@ type FileDescriptionImpl interface {
Removexattr(ctx context.Context, name string) error
// LockBSD tries to acquire a BSD-style advisory file lock.
- //
- // TODO(gvisor.dev/issue/1480): BSD-style file locking
LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error
- // LockBSD releases a BSD-style advisory file lock.
- //
- // TODO(gvisor.dev/issue/1480): BSD-style file locking
+ // UnlockBSD releases a BSD-style advisory file lock.
UnlockBSD(ctx context.Context, uid lock.UniqueID) error
// LockPOSIX tries to acquire a POSIX-style advisory file lock.
@@ -736,3 +740,14 @@ func (fd *FileDescription) InodeID() uint64 {
func (fd *FileDescription) Msync(ctx context.Context, mr memmap.MappableRange) error {
return fd.Sync(ctx)
}
+
+// LockBSD tries to acquire a BSD-style advisory file lock.
+func (fd *FileDescription) LockBSD(ctx context.Context, lockType lock.LockType, blocker lock.Blocker) error {
+ atomic.StoreUint32(&fd.usedLockBSD, 1)
+ return fd.impl.LockBSD(ctx, fd, lockType, blocker)
+}
+
+// UnlockBSD releases a BSD-style advisory file lock.
+func (fd *FileDescription) UnlockBSD(ctx context.Context) error {
+ return fd.impl.UnlockBSD(ctx, fd)
+}
diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go
index f4c111926..af7213dfd 100644
--- a/pkg/sentry/vfs/file_description_impl_util.go
+++ b/pkg/sentry/vfs/file_description_impl_util.go
@@ -21,8 +21,9 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/arch"
- "gvisor.dev/gvisor/pkg/sentry/fs/lock"
+ fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock"
"gvisor.dev/gvisor/pkg/sentry/memmap"
+ "gvisor.dev/gvisor/pkg/sentry/vfs/lock"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -153,26 +154,6 @@ func (FileDescriptionDefaultImpl) Removexattr(ctx context.Context, name string)
return syserror.ENOTSUP
}
-// LockBSD implements FileDescriptionImpl.LockBSD.
-func (FileDescriptionDefaultImpl) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error {
- return syserror.EBADF
-}
-
-// UnlockBSD implements FileDescriptionImpl.UnlockBSD.
-func (FileDescriptionDefaultImpl) UnlockBSD(ctx context.Context, uid lock.UniqueID) error {
- return syserror.EBADF
-}
-
-// LockPOSIX implements FileDescriptionImpl.LockPOSIX.
-func (FileDescriptionDefaultImpl) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error {
- return syserror.EBADF
-}
-
-// UnlockPOSIX implements FileDescriptionImpl.UnlockPOSIX.
-func (FileDescriptionDefaultImpl) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error {
- return syserror.EBADF
-}
-
// DirectoryFileDescriptionDefaultImpl may be embedded by implementations of
// FileDescriptionImpl that always represent directories to obtain
// implementations of non-directory I/O methods that return EISDIR.
@@ -384,3 +365,60 @@ func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.M
fd.IncRef()
return nil
}
+
+// LockFD may be used by most implementations of FileDescriptionImpl.Lock*
+// functions. Caller must call Init().
+type LockFD struct {
+ locks *lock.FileLocks
+}
+
+// Init initializes fd with FileLocks to use.
+func (fd *LockFD) Init(locks *lock.FileLocks) {
+ fd.locks = locks
+}
+
+// LockBSD implements vfs.FileDescriptionImpl.LockBSD.
+func (fd *LockFD) LockBSD(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error {
+ return fd.locks.LockBSD(uid, t, block)
+}
+
+// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD.
+func (fd *LockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error {
+ fd.locks.UnlockBSD(uid)
+ return nil
+}
+
+// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
+func (fd *LockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error {
+ return fd.locks.LockPOSIX(uid, t, rng, block)
+}
+
+// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX.
+func (fd *LockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, rng fslock.LockRange) error {
+ fd.locks.UnlockPOSIX(uid, rng)
+ return nil
+}
+
+// NoLockFD implements Lock*/Unlock* portion of FileDescriptionImpl interface
+// returning ENOLCK.
+type NoLockFD struct{}
+
+// LockBSD implements vfs.FileDescriptionImpl.LockBSD.
+func (NoLockFD) LockBSD(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error {
+ return syserror.ENOLCK
+}
+
+// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD.
+func (NoLockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error {
+ return syserror.ENOLCK
+}
+
+// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
+func (NoLockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error {
+ return syserror.ENOLCK
+}
+
+// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX.
+func (NoLockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, rng fslock.LockRange) error {
+ return syserror.ENOLCK
+}
diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go
index 3a75d4d62..5061f6ac9 100644
--- a/pkg/sentry/vfs/file_description_impl_util_test.go
+++ b/pkg/sentry/vfs/file_description_impl_util_test.go
@@ -33,6 +33,7 @@ import (
type fileDescription struct {
vfsfd FileDescription
FileDescriptionDefaultImpl
+ NoLockFD
}
// genCount contains the number of times its DynamicBytesSource.Generate()
diff --git a/pkg/sentry/vfs/genericfstree/genericfstree.go b/pkg/sentry/vfs/genericfstree/genericfstree.go
index 286510195..8882fa84a 100644
--- a/pkg/sentry/vfs/genericfstree/genericfstree.go
+++ b/pkg/sentry/vfs/genericfstree/genericfstree.go
@@ -43,7 +43,7 @@ type Dentry struct {
// IsAncestorDentry returns true if d is an ancestor of d2; that is, d is
// either d2's parent or an ancestor of d2's parent.
func IsAncestorDentry(d, d2 *Dentry) bool {
- for {
+ for d2 != nil { // Stop at root, where d2.parent == nil.
if d2.parent == d {
return true
}
@@ -52,6 +52,7 @@ func IsAncestorDentry(d, d2 *Dentry) bool {
}
d2 = d2.parent
}
+ return false
}
// ParentOrSelf returns d.parent. If d.parent is nil, ParentOrSelf returns d.
diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go
index 05a3051a4..7fa7d2d0c 100644
--- a/pkg/sentry/vfs/inotify.go
+++ b/pkg/sentry/vfs/inotify.go
@@ -57,6 +57,7 @@ type Inotify struct {
vfsfd FileDescription
FileDescriptionDefaultImpl
DentryMetadataFileDescriptionImpl
+ NoLockFD
// Unique identifier for this inotify instance. We don't just reuse the
// inotify fd because fds can be duped. These should not be exposed to the
diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go
index e4ac6524b..32f901bd8 100644
--- a/pkg/sentry/vfs/mount.go
+++ b/pkg/sentry/vfs/mount.go
@@ -55,6 +55,10 @@ type Mount struct {
// ID is the immutable mount ID.
ID uint64
+ // Flags contains settings as specified for mount(2), e.g. MS_NOEXEC, except
+ // for MS_RDONLY which is tracked in "writers". Immutable.
+ Flags MountFlags
+
// key is protected by VirtualFilesystem.mountMu and
// VirtualFilesystem.mounts.seq, and may be nil. References are held on
// key.parent and key.point if they are not nil.
@@ -81,10 +85,6 @@ type Mount struct {
// umounted is true. umounted is protected by VirtualFilesystem.mountMu.
umounted bool
- // flags contains settings as specified for mount(2), e.g. MS_NOEXEC, except
- // for MS_RDONLY which is tracked in "writers".
- flags MountFlags
-
// The lower 63 bits of writers is the number of calls to
// Mount.CheckBeginWrite() that have not yet been paired with a call to
// Mount.EndWrite(). The MSB of writers is set if MS_RDONLY is in effect.
@@ -95,10 +95,10 @@ type Mount struct {
func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount {
mnt := &Mount{
ID: atomic.AddUint64(&vfs.lastMountID, 1),
+ Flags: opts.Flags,
vfs: vfs,
fs: fs,
root: root,
- flags: opts.Flags,
ns: mntns,
refs: 1,
}
@@ -113,13 +113,12 @@ func (mnt *Mount) Options() MountOptions {
mnt.vfs.mountMu.Lock()
defer mnt.vfs.mountMu.Unlock()
return MountOptions{
- Flags: mnt.flags,
+ Flags: mnt.Flags,
ReadOnly: mnt.readOnly(),
}
}
-// A MountNamespace is a collection of Mounts.
-//
+// A MountNamespace is a collection of Mounts.//
// MountNamespaces are reference-counted. Unless otherwise specified, all
// MountNamespace methods require that a reference is held.
//
@@ -127,6 +126,9 @@ func (mnt *Mount) Options() MountOptions {
//
// +stateify savable
type MountNamespace struct {
+ // Owner is the usernamespace that owns this mount namespace.
+ Owner *auth.UserNamespace
+
// root is the MountNamespace's root mount. root is immutable.
root *Mount
@@ -163,6 +165,7 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth
return nil, err
}
mntns := &MountNamespace{
+ Owner: creds.UserNamespace,
refs: 1,
mountpoints: make(map[*Dentry]uint32),
}
@@ -265,8 +268,8 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia
if err != nil {
return err
}
+ defer mnt.DecRef()
if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil {
- mnt.DecRef()
return err
}
return nil
@@ -279,6 +282,9 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti
}
// MNT_FORCE is currently unimplemented except for the permission check.
+ // Force unmounting specifically requires CAP_SYS_ADMIN in the root user
+ // namespace, and not in the owner user namespace for the target mount. See
+ // fs/namespace.c:SYSCALL_DEFINE2(umount, ...)
if opts.Flags&linux.MNT_FORCE != 0 && creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, creds.UserNamespace.Root()) {
return syserror.EPERM
}
@@ -394,8 +400,15 @@ func (vfs *VirtualFilesystem) umountRecursiveLocked(mnt *Mount, opts *umountRecu
// references held by vd.
//
// Preconditions: vfs.mountMu must be locked. vfs.mounts.seq must be in a
-// writer critical section. d.mu must be locked. mnt.parent() == nil.
+// writer critical section. d.mu must be locked. mnt.parent() == nil, i.e. mnt
+// must not already be connected.
func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns *MountNamespace) {
+ if checkInvariants {
+ if mnt.parent() != nil {
+ panic("VFS.connectLocked called on connected mount")
+ }
+ }
+ mnt.IncRef() // dropped by callers of umountRecursiveLocked
mnt.storeKey(vd)
if vd.mount.children == nil {
vd.mount.children = make(map[*Mount]struct{})
@@ -420,6 +433,11 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns
// writer critical section. mnt.parent() != nil.
func (vfs *VirtualFilesystem) disconnectLocked(mnt *Mount) VirtualDentry {
vd := mnt.loadKey()
+ if checkInvariants {
+ if vd.mount != nil {
+ panic("VFS.disconnectLocked called on disconnected mount")
+ }
+ }
mnt.storeKey(VirtualDentry{})
delete(vd.mount.children, mnt)
atomic.AddUint32(&vd.dentry.mounts, math.MaxUint32) // -1
@@ -741,7 +759,10 @@ func (vfs *VirtualFilesystem) GenerateProcMounts(ctx context.Context, taskRootDi
if mnt.readOnly() {
opts = "ro"
}
- if mnt.flags.NoExec {
+ if mnt.Flags.NoATime {
+ opts = ",noatime"
+ }
+ if mnt.Flags.NoExec {
opts += ",noexec"
}
@@ -826,11 +847,12 @@ func (vfs *VirtualFilesystem) GenerateProcMountInfo(ctx context.Context, taskRoo
if mnt.readOnly() {
opts = "ro"
}
- if mnt.flags.NoExec {
+ if mnt.Flags.NoATime {
+ opts = ",noatime"
+ }
+ if mnt.Flags.NoExec {
opts += ",noexec"
}
- // TODO(gvisor.dev/issue/1193): Add "noatime" if MS_NOATIME is
- // set.
fmt.Fprintf(buf, "%s ", opts)
// (7) Optional fields: zero or more fields of the form "tag[:value]".
diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go
index 53d364c5c..f223aeda8 100644
--- a/pkg/sentry/vfs/options.go
+++ b/pkg/sentry/vfs/options.go
@@ -75,6 +75,10 @@ type MknodOptions struct {
type MountFlags struct {
// NoExec is equivalent to MS_NOEXEC.
NoExec bool
+
+ // NoATime is equivalent to MS_NOATIME and indicates that the
+ // filesystem should not update access time in-place.
+ NoATime bool
}
// MountOptions contains options to VirtualFilesystem.MountAt().
diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go
index 52643a7c5..9acca8bc7 100644
--- a/pkg/sentry/vfs/vfs.go
+++ b/pkg/sentry/vfs/vfs.go
@@ -405,7 +405,7 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential
vfs.putResolvingPath(rp)
if opts.FileExec {
- if fd.Mount().flags.NoExec {
+ if fd.Mount().Flags.NoExec {
fd.DecRef()
return nil, syserror.EACCES
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-18 20:46:08 +0000