diff options
Diffstat (limited to 'pkg/sentry')
74 files changed, 625 insertions, 362 deletions
diff --git a/pkg/sentry/control/logging.go b/pkg/sentry/control/logging.go index 811f24324..8a500a515 100644 --- a/pkg/sentry/control/logging.go +++ b/pkg/sentry/control/logging.go @@ -70,8 +70,8 @@ type LoggingArgs struct { type Logging struct{} // Change will change the log level and strace arguments. Although -// this functions signature requires an error it never acctually -// return san error. It's required by the URPC interface. +// this functions signature requires an error it never actually +// returns an error. It's required by the URPC interface. // Additionally, it may look odd that this is the only method // attached to an empty struct but this is also part of how // URPC dispatches. diff --git a/pkg/sentry/device/device.go b/pkg/sentry/device/device.go index 69e71e322..f45b2bd2b 100644 --- a/pkg/sentry/device/device.go +++ b/pkg/sentry/device/device.go @@ -188,6 +188,9 @@ type MultiDevice struct { // String stringifies MultiDevice. func (m *MultiDevice) String() string { + m.mu.Lock() + defer m.mu.Unlock() + buf := bytes.NewBuffer(nil) buf.WriteString("cache{") for k, v := range m.cache { diff --git a/pkg/sentry/fs/filesystems.go b/pkg/sentry/fs/filesystems.go index 084da2a8d..d41f30bbb 100644 --- a/pkg/sentry/fs/filesystems.go +++ b/pkg/sentry/fs/filesystems.go @@ -87,20 +87,6 @@ func RegisterFilesystem(f Filesystem) { filesystems.registered[f.Name()] = f } -// UnregisterFilesystem removes a file system from the global set. To keep the -// file system set compatible with save/restore, UnregisterFilesystem must be -// called before save/restore methods. -// -// For instance, packages may unregister their file system after it is mounted. -// This makes sense for pseudo file systems that should not be visible or -// mountable. See whitelistfs in fs/host/fs.go for one example. -func UnregisterFilesystem(name string) { - filesystems.mu.Lock() - defer filesystems.mu.Unlock() - - delete(filesystems.registered, name) -} - // FindFilesystem returns a Filesystem registered at name or (nil, false) if name // is not a file system type that can be found in /proc/filesystems. func FindFilesystem(name string) (Filesystem, bool) { diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index cb91355ab..82a02fcb2 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -308,9 +308,9 @@ func (t *TTYFileOperations) checkChange(ctx context.Context, sig linux.Signal) e task := kernel.TaskFromContext(ctx) if task == nil { // No task? Linux does not have an analog for this case, but - // tty_check_change is more of a blacklist of cases than a - // whitelist, and is surprisingly permissive. Allowing the - // change seems most appropriate. + // tty_check_change only blocks specific cases and is + // surprisingly permissive. Allowing the change seems + // appropriate. return nil } diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD index cf440dce8..93512c9b6 100644 --- a/pkg/sentry/fsimpl/devpts/BUILD +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -18,12 +18,12 @@ go_library( "//pkg/context", "//pkg/safemem", "//pkg/sentry/arch", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 9b0e0cca2..e6fda2b4f 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -28,7 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -117,7 +116,7 @@ type rootInode struct { kernfs.InodeNotSymlink kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks // Keep a reference to this inode's dentry. dentry kernfs.Dentry diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 1d22adbe3..69879498a 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -18,11 +18,11 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -35,7 +35,7 @@ type masterInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -189,6 +189,16 @@ func (mfd *masterFileDescription) Stat(ctx context.Context, opts vfs.StatOptions return mfd.inode.Stat(fs, opts) } +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (mfd *masterFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return mfd.Locks().LockPOSIX(ctx, &mfd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (mfd *masterFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return mfd.Locks().UnlockPOSIX(ctx, &mfd.vfsfd, uid, start, length, whence) +} + // maybeEmitUnimplementedEvent emits unimplemented event if cmd is valid. func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { switch cmd { diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index 7fe475080..cf1a0f0ac 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -18,10 +18,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -34,7 +34,7 @@ type slaveInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -185,3 +185,13 @@ func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() return sfd.inode.Stat(fs, opts) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (sfd *slaveFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return sfd.Locks().LockPOSIX(ctx, &sfd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (sfd *slaveFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return sfd.Locks().UnlockPOSIX(ctx, &sfd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 973fa0def..ef24f8159 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -54,13 +54,13 @@ go_library( "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/ext/disklayout", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls/linux", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 43be6928a..357512c7e 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -305,3 +306,13 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in fd.off = offset return offset, nil } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *directoryFD) 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *directoryFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 5caaf14ed..30636cf66 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -55,7 +54,7 @@ type inode struct { // diskInode gives us access to the inode struct on disk. Immutable. diskInode disklayout.Inode - locks lock.FileLocks + locks vfs.FileLocks // This is immutable. The first field of the implementations must have inode // as the first field to ensure temporality. diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index 152036b2e..66d14bb95 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -149,3 +150,13 @@ func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpt // TODO(b/134676337): Implement mmap(2). return syserror.ENODEV } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *regularFileFD) 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index acb28d85b..62efd4095 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -66,6 +66,7 @@ func (in *inode) isSymlink() bool { // O_PATH. For this reason most of the functions return EBADF. type symlinkFD struct { fileDescription + vfs.NoLockFD } // Compiles only if symlinkFD implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 5cdeeaeb5..4a800dcf9 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -69,7 +69,6 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/unet", diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index b98218753..480510b2a 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -299,3 +299,8 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in return 0, syserror.EINVAL } } + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *directoryFD) Sync(ctx context.Context) error { + return fd.dentry().handle.sync(ctx) +} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 3c467e313..f065c4bad 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -374,13 +374,16 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir return nil } if fs.opts.interop == InteropModeShared { - // The existence of a dentry at name would be inconclusive because the - // file it represents may have been deleted from the remote filesystem, - // so we would need to make an RPC to revalidate the dentry. Just - // attempt the file creation RPC instead. If a file does exist, the RPC - // will fail with EEXIST like we would have. If the RPC succeeds, and a - // stale dentry exists, the dentry will fail revalidation next time - // it's used. + if child := parent.children[name]; child != nil && child.isSynthetic() { + return syserror.EEXIST + } + // The existence of a non-synthetic dentry at name would be inconclusive + // because the file it represents may have been deleted from the remote + // filesystem, so we would need to make an RPC to revalidate the dentry. + // Just attempt the file creation RPC instead. If a file does exist, the + // RPC will fail with EEXIST like we would have. If the RPC succeeds, and a + // stale dentry exists, the dentry will fail revalidation next time it's + // used. return createInRemoteDir(parent, name) } if child := parent.children[name]; child != nil { @@ -518,7 +521,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b if child == nil { return syserror.ENOENT } - } else { + } else if child == nil || !child.isSynthetic() { err = parent.file.unlinkAt(ctx, name, flags) if err != nil { if child != nil { @@ -764,15 +767,17 @@ afterTrailingSymlink: parent.dirMu.Unlock() return fd, err } + parent.dirMu.Unlock() if err != nil { - parent.dirMu.Unlock() return nil, err } - // Open existing child or follow symlink. - parent.dirMu.Unlock() if mustCreate { return nil, syserror.EEXIST } + if !child.isDir() && rp.MustBeDir() { + return nil, syserror.ENOTDIR + } + // Open existing child or follow symlink. if child.isSymlink() && rp.ShouldFollowSymlink() { target, err := child.readlink(ctx, rp.Mount()) if err != nil { @@ -1193,7 +1198,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if newParent.cachedMetadataAuthoritative() { newParent.dirents = nil newParent.touchCMtime() - if renamed.isDir() { + if renamed.isDir() && (replaced == nil || !replaced.isDir()) { + // Increase the link count if we did not replace another directory. newParent.incLinks() } } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index ac051b3a7..43c8153a4 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -53,7 +53,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/usermem" @@ -665,7 +664,7 @@ type dentry struct { // endpoint bound to this file. pipe *pipe.VFSPipe - locks lock.FileLocks + locks vfs.FileLocks } // dentryAttrMask returns a p9.AttrMask enabling all attributes used by the @@ -1207,7 +1206,7 @@ func (d *dentry) setDeleted() { // We only support xattrs prefixed with "user." (see b/148380782). Currently, // there is no need to expose any other xattrs through a gofer. func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size uint64) ([]string, error) { - if d.file.isNil() { + if d.file.isNil() || !d.userXattrSupported() { return nil, nil } xattrMap, err := d.file.listXattr(ctx, size) @@ -1233,6 +1232,9 @@ func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vf if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { return "", syserror.EOPNOTSUPP } + if !d.userXattrSupported() { + return "", syserror.ENODATA + } return d.file.getXattr(ctx, opts.Name, opts.Size) } @@ -1246,6 +1248,9 @@ func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vf if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { return syserror.EOPNOTSUPP } + if !d.userXattrSupported() { + return syserror.EPERM + } return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) } @@ -1259,9 +1264,19 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { return syserror.EOPNOTSUPP } + if !d.userXattrSupported() { + return syserror.EPERM + } return d.file.removeXattr(ctx, name) } +// Extended attributes in the user.* namespace are only supported for regular +// files and directories. +func (d *dentry) userXattrSupported() bool { + filetype := linux.S_IFMT & atomic.LoadUint32(&d.mode) + return filetype == linux.S_IFREG || filetype == linux.S_IFDIR +} + // Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDirectory(). func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { // O_TRUNC unconditionally requires us to obtain a new handle (opened with @@ -1439,9 +1454,14 @@ func (fd *fileDescription) LockBSD(ctx context.Context, uid fslock.UniqueID, t f } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { +func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { fd.lockLogging.Do(func() { log.Infof("Range lock using gofer file handled internally.") }) - return fd.LockFD.LockPOSIX(ctx, uid, t, rng, block) + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) } diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 0d10cf7ac..404a452c4 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -72,7 +72,9 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs if offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -123,9 +125,12 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } + limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) if err != nil { return 0, err @@ -489,15 +494,24 @@ func (d *dentry) writeback(ctx context.Context, offset, size int64) error { func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { fd.mu.Lock() defer fd.mu.Unlock() + newOffset, err := regularFileSeekLocked(ctx, fd.dentry(), fd.off, offset, whence) + if err != nil { + return 0, err + } + fd.off = newOffset + return newOffset, nil +} + +// Calculate the new offset for a seek operation on a regular file. +func regularFileSeekLocked(ctx context.Context, d *dentry, fdOffset, offset int64, whence int32) (int64, error) { switch whence { case linux.SEEK_SET: // Use offset as specified. case linux.SEEK_CUR: - offset += fd.off + offset += fdOffset case linux.SEEK_END, linux.SEEK_DATA, linux.SEEK_HOLE: // Ensure file size is up to date. - d := fd.dentry() - if fd.filesystem().opts.interop == InteropModeShared { + if !d.cachedMetadataAuthoritative() { if err := d.updateFromGetattr(ctx); err != nil { return 0, err } @@ -525,7 +539,6 @@ func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) ( if offset < 0 { return 0, syserror.EINVAL } - fd.off = offset return offset, nil } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 289efdd25..a016cbae1 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -52,7 +51,7 @@ type specialFileFD struct { off int64 } -func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *lock.FileLocks, flags uint32) (*specialFileFD, error) { +func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, flags uint32) (*specialFileFD, error) { ftype := d.fileType() seekable := ftype == linux.S_IFREG mayBlock := ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK @@ -130,7 +129,9 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs if fd.seekable && offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -174,7 +175,9 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if fd.seekable && offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -222,21 +225,12 @@ func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) ( } fd.mu.Lock() defer fd.mu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - // SEEK_END, SEEK_DATA, and SEEK_HOLE aren't supported since it's not - // clear that file size is even meaningful for these files. - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL + newOffset, err := regularFileSeekLocked(ctx, fd.dentry(), fd.off, offset, whence) + if err != nil { + return 0, err } - fd.off = offset - return offset, nil + fd.off = newOffset + return newOffset, nil } // Sync implements vfs.FileDescriptionImpl.Sync. diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 54f16ad63..44a09d87a 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -27,6 +27,7 @@ go_library( "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/hostfd", "//pkg/sentry/kernel", @@ -39,7 +40,6 @@ go_library( "//pkg/sentry/unimpl", "//pkg/sentry/uniqueid", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 5ec5100b8..a65543028 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -28,13 +28,13 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" unixsocket "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -91,7 +91,9 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) isTTY: opts.IsTTY, wouldBlock: wouldBlock(uint32(fileType)), seekable: seekable, - canMap: canMap(uint32(fileType)), + // NOTE(b/38213152): Technically, some obscure char devices can be memory + // mapped, but we only allow regular files. + canMap: fileType == linux.S_IFREG, } i.pf.inode = i @@ -183,7 +185,7 @@ type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks // When the reference count reaches zero, the host fd is closed. refs.AtomicRefCount @@ -460,7 +462,8 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND - if fileType == syscall.S_IFSOCK { + switch fileType { + case syscall.S_IFSOCK: if i.isTTY { log.Warningf("cannot use host socket fd %d as TTY", i.hostFD) return nil, syserror.ENOTTY @@ -472,30 +475,33 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } // Currently, we only allow Unix sockets to be imported. return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d, &i.locks) - } - // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that - // we don't allow importing arbitrary file types without proper support. - if i.isTTY { - fd := &TTYFileDescription{ - fileDescription: fileDescription{inode: i}, - termios: linux.DefaultSlaveTermios, + case syscall.S_IFREG, syscall.S_IFIFO, syscall.S_IFCHR: + if i.isTTY { + fd := &TTYFileDescription{ + fileDescription: fileDescription{inode: i}, + termios: linux.DefaultSlaveTermios, + } + fd.LockFD.Init(&i.locks) + vfsfd := &fd.vfsfd + if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return vfsfd, nil } + + fd := &fileDescription{inode: i} fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil - } - fd := &fileDescription{inode: i} - fd.LockFD.Init(&i.locks) - vfsfd := &fd.vfsfd - if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err + default: + log.Warningf("cannot import host fd %d with file type %o", i.hostFD, fileType) + return nil, syserror.EPERM } - return vfsfd, nil } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. @@ -688,7 +694,8 @@ func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (i // Sync implements FileDescriptionImpl. func (f *fileDescription) Sync(context.Context) error { - // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. + // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData + // optimization, so we always sync everything. return unix.Fsync(f.inode.hostFD) } @@ -718,3 +725,13 @@ func (f *fileDescription) EventUnregister(e *waiter.Entry) { func (f *fileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { return fdnotifier.NonBlockingPoll(int32(f.inode.hostFD), mask) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (f *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return f.Locks().LockPOSIX(ctx, &f.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (f *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return f.Locks().UnlockPOSIX(ctx, &f.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index 68af6e5af..4ee9270cc 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -325,9 +326,9 @@ func (t *TTYFileDescription) checkChange(ctx context.Context, sig linux.Signal) task := kernel.TaskFromContext(ctx) if task == nil { // No task? Linux does not have an analog for this case, but - // tty_check_change is more of a blacklist of cases than a - // whitelist, and is surprisingly permissive. Allowing the - // change seems most appropriate. + // tty_check_change only blocks specific cases and is + // surprisingly permissive. Allowing the change seems + // appropriate. return nil } @@ -377,3 +378,13 @@ func (t *TTYFileDescription) checkChange(ctx context.Context, sig linux.Signal) _ = pg.SendSignal(kernel.SignalInfoPriv(sig)) return kernel.ERESTARTSYS } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (t *TTYFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, typ fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return t.Locks().LockPOSIX(ctx, &t.vfsfd, uid, typ, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (t *TTYFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return t.Locks().UnlockPOSIX(ctx, &t.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go index 2bc757b1a..412bdb2eb 100644 --- a/pkg/sentry/fsimpl/host/util.go +++ b/pkg/sentry/fsimpl/host/util.go @@ -49,16 +49,6 @@ func wouldBlock(fileType uint32) bool { return fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK } -// canMap returns true if a file with fileType is allowed to be memory mapped. -// This is ported over from VFS1, but it's probably not the best way for us -// to check if a file can be memory mapped. -func canMap(fileType uint32) bool { - // TODO(gvisor.dev/issue/1672): Also allow "special files" to be mapped (see fs/host:canMap()). - // - // TODO(b/38213152): Some obscure character devices can be mapped. - return fileType == syscall.S_IFREG -} - // isBlockError checks if an error is EAGAIN or EWOULDBLOCK. // If so, they can be transformed into syserror.ErrWouldBlock. func isBlockError(err error) bool { diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 0299dbde9..179df6c1e 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -45,11 +45,11 @@ go_library( "//pkg/fspath", "//pkg/log", "//pkg/refs", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", @@ -68,7 +68,6 @@ go_test( "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/usermem", "@com_github_google_go-cmp//cmp:go_default_library", diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 6418de0a3..c1215b70a 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -19,9 +19,9 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -39,7 +39,7 @@ type DynamicBytesFile struct { InodeNotDirectory InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks data vfs.DynamicBytesSource } @@ -86,7 +86,7 @@ type DynamicBytesFD struct { } // Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *lock.FileLocks, flags uint32) error { +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *vfs.FileLocks, flags uint32) error { fd.LockFD.Init(locks) if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { return err @@ -135,3 +135,13 @@ func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error { // DynamicBytesFiles are immutable. return syserror.EPERM } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *DynamicBytesFD) 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *DynamicBytesFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 33a5968ca..5f7853a2a 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -19,10 +19,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -57,7 +57,7 @@ type GenericDirectoryFD struct { // NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its // dentry. -func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *lock.FileLocks, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { +func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { fd := &GenericDirectoryFD{} if err := fd.Init(children, locks, opts); err != nil { return nil, err @@ -71,7 +71,7 @@ func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildre // Init initializes a GenericDirectoryFD. Use it when overriding // GenericDirectoryFD. Caller must call fd.VFSFileDescription.Init() with the // correct implementation. -func (fd *GenericDirectoryFD) Init(children *OrderedChildren, locks *lock.FileLocks, opts *vfs.OpenOptions) error { +func (fd *GenericDirectoryFD) Init(children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions) error { if vfs.AccessTypesForOpenFlags(opts)&vfs.MayWrite != 0 { // Can't open directories for writing. return syserror.EISDIR @@ -235,3 +235,13 @@ func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptio inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode return inode.SetStat(ctx, fd.filesystem(), creds, opts) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *GenericDirectoryFD) 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *GenericDirectoryFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 0e4927215..650bd7b88 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -557,7 +556,7 @@ type StaticDirectory struct { InodeNoDynamicLookup OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks } var _ Inode = (*StaticDirectory)(nil) diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index bbee8ccda..07a9dc830 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -425,10 +425,10 @@ type inodeDynamicLookup interface { // IterDirents is used to iterate over dynamically created entries. It invokes // cb on each entry in the directory represented by the FileDescription. // 'offset' is the offset for the entire IterDirents call, which may include - // results from the caller. 'relOffset' is the offset inside the entries - // returned by this IterDirents invocation. In other words, - // 'offset+relOffset+1' is the value that should be set in vfs.Dirent.NextOff, - // while 'relOffset' is the place where iteration should start from. + // results from the caller (e.g. "." and ".."). 'relOffset' is the offset + // inside the entries returned by this IterDirents invocation. In other words, + // 'offset' should be used to calculate each vfs.Dirent.NextOff as well as + // the return value, while 'relOffset' is the place to start iteration. IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 6749facf7..dc407eb1d 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -103,7 +102,7 @@ type readonlyDir struct { kernfs.InodeDirectoryNoNewChildren kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks dentry kernfs.Dentry } @@ -133,7 +132,7 @@ type dir struct { kernfs.InodeNoDynamicLookup kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem dentry kernfs.Dentry diff --git a/pkg/sentry/fsimpl/overlay/BUILD b/pkg/sentry/fsimpl/overlay/BUILD index f9413bbdd..8cf5b35d3 100644 --- a/pkg/sentry/fsimpl/overlay/BUILD +++ b/pkg/sentry/fsimpl/overlay/BUILD @@ -29,11 +29,11 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index 6f47167d3..f5c2462a5 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -263,3 +263,25 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in return 0, syserror.EINVAL } } + +// Sync implements vfs.FileDescriptionImpl.Sync. Forwards sync to the upper +// layer, if there is one. The lower layer doesn't need to sync because it +// never changes. +func (fd *directoryFD) Sync(ctx context.Context) error { + d := fd.dentry() + if !d.isCopiedUp() { + return nil + } + vfsObj := d.fs.vfsfs.VirtualFilesystem() + pop := vfs.PathOperation{ + Root: d.upperVD, + Start: d.upperVD, + } + upperFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &pop, &vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_DIRECTORY}) + if err != nil { + return err + } + err = upperFD.Sync(ctx) + upperFD.DecRef() + return err +} diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index e660d0e2c..e11a3ff19 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -35,9 +35,9 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -415,7 +415,7 @@ type dentry struct { devMinor uint32 ino uint64 - locks lock.FileLocks + locks vfs.FileLocks } // newDentry creates a new dentry. The dentry initially has no references; it @@ -610,3 +610,13 @@ func (fd *fileDescription) filesystem() *filesystem { func (fd *fileDescription) dentry() *dentry { return fd.vfsfd.Dentry().Impl().(*dentry) } + +// 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/pipefs/BUILD b/pkg/sentry/fsimpl/pipefs/BUILD index c618dbe6c..5950a2d59 100644 --- a/pkg/sentry/fsimpl/pipefs/BUILD +++ b/pkg/sentry/fsimpl/pipefs/BUILD @@ -15,7 +15,6 @@ go_library( "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/usermem", ], diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index e4dabaa33..dd7eaf4a8 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -82,7 +81,7 @@ type inode struct { kernfs.InodeNotSymlink kernfs.InodeNoopRefCount - locks lock.FileLocks + locks vfs.FileLocks pipe *pipe.VFSPipe ino uint64 diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 351ba4ee9..6014138ff 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -22,6 +22,7 @@ go_library( "//pkg/log", "//pkg/refs", "//pkg/safemem", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", @@ -35,7 +36,6 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/tcpip/header", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index e2cdb7ee9..36a89540c 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -24,7 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -38,7 +37,7 @@ type subtasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem task *kernel.Task diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 44078a765..8bb2b0ce1 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -39,7 +38,7 @@ type taskInode struct { kernfs.InodeAttrs kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks task *kernel.Task } diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index ef6c1d04f..fea29e5f0 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -54,7 +53,7 @@ func taskFDExists(t *kernel.Task, fd int32) bool { } type fdDir struct { - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem task *kernel.Task @@ -65,7 +64,7 @@ type fdDir struct { } // IterDirents implements kernfs.inodeDynamicLookup. -func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, absOffset, relOffset int64) (int64, error) { +func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { var fds []int32 i.task.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { @@ -73,7 +72,6 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, abs } }) - offset := absOffset + relOffset typ := uint8(linux.DT_REG) if i.produceSymlink { typ = linux.DT_LNK diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index e5eaa91cd..ba4405026 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -30,7 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -777,7 +777,7 @@ type namespaceInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks } var _ kernfs.Inode = (*namespaceInode)(nil) @@ -830,3 +830,13 @@ func (fd *namespaceFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) err func (fd *namespaceFD) Release() { fd.inode.DecRef() } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *namespaceFD) 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *namespaceFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 58c8b9d05..2f214d0c2 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -44,7 +43,7 @@ type tasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem pidns *kernel.PIDNamespace diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index 237f17def..a741e2bb6 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -15,7 +15,6 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", ], ) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index b84463d3a..fe02f7ee9 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -101,7 +100,7 @@ type dir struct { kernfs.InodeDirectoryNoNewChildren kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks dentry kernfs.Dentry } diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 062321cbc..e73732a6b 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -62,7 +62,6 @@ go_library( "//pkg/sentry/uniqueid", "//pkg/sentry/usage", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sentry/vfs/memxattr", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 72399b321..637d84e04 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -79,7 +79,7 @@ afterSymlink: } if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { // Symlink traversal updates access time. - atomic.StoreInt64(&d.inode.atime, d.inode.fs.clock.Now().Nanoseconds()) + child.inode.touchAtime(rp.Mount()) if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } @@ -237,18 +237,22 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EXDEV } d := vd.Dentry().Impl().(*dentry) - if d.inode.isDir() { + i := d.inode + if i.isDir() { return syserror.EPERM } - if d.inode.nlink == 0 { + if err := vfs.MayLink(auth.CredentialsFromContext(ctx), linux.FileMode(atomic.LoadUint32(&i.mode)), auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { + return err + } + if i.nlink == 0 { return syserror.ENOENT } - if d.inode.nlink == maxLinks { + if i.nlink == maxLinks { return syserror.EMLINK } - d.inode.incLinksLocked() - d.inode.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) - parentDir.insertChildLocked(fs.newDentry(d.inode), name) + i.incLinksLocked() + i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) + parentDir.insertChildLocked(fs.newDentry(i), name) return nil }) } @@ -368,6 +372,9 @@ afterTrailingSymlink: parentDir.inode.touchCMtime() return fd, nil } + if mustCreate { + return nil, syserror.EEXIST + } // Is the file mounted over? if err := rp.CheckMount(&child.vfsd); err != nil { return nil, err @@ -375,7 +382,7 @@ afterTrailingSymlink: // Do we need to resolve a trailing symlink? if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { // Symlink traversal updates access time. - atomic.StoreInt64(&child.inode.atime, child.inode.fs.clock.Now().Nanoseconds()) + child.inode.touchAtime(rp.Mount()) if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 77447b32c..b805aadd0 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -279,6 +279,12 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs if offset < 0 { return 0, syserror.EINVAL } + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { + return 0, syserror.EOPNOTSUPP + } + if dst.NumBytes() == 0 { return 0, nil } @@ -304,6 +310,12 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if offset < 0 { return 0, syserror.EINVAL } + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { + return 0, syserror.EOPNOTSUPP + } + srclen := src.NumBytes() if srclen == 0 { return 0, nil @@ -360,11 +372,6 @@ func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) ( return offset, nil } -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return nil -} - // ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { file := fd.inode().impl.(*regularFile) diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 64e1c40ad..146c7fdfe 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -138,48 +138,37 @@ func TestLocks(t *testing.T) { } defer cleanup() - var ( - uid1 lock.UniqueID - uid2 lock.UniqueID - // Non-blocking. - block lock.Blocker - ) - - uid1 = 123 - uid2 = 456 - - if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, block); err != nil { + uid1 := 123 + uid2 := 456 + if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, nil); err != nil { t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) } - if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, block); err != nil { + if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, nil); err != nil { t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) } - if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block), syserror.ErrWouldBlock; got != want { + if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, nil), syserror.ErrWouldBlock; got != want { t.Fatalf("fd.Impl().LockBSD failed: got = %v, want = %v", got, want) } if err := fd.Impl().UnlockBSD(ctx, uid1); err != nil { t.Fatalf("fd.Impl().UnlockBSD failed: err = %v", err) } - if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block); err != nil { + if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, nil); err != nil { t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) } - rng1 := lock.LockRange{0, 1} - rng2 := lock.LockRange{1, 2} - - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, rng1, block); err != nil { + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, 0, 1, linux.SEEK_SET, nil); err != nil { t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) } - if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng2, block); err != nil { + if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, 1, 2, linux.SEEK_SET, nil); err != nil { t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) } - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, rng1, block); err != nil { + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, 0, 1, linux.SEEK_SET, nil); err != nil { t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) } - if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng1, block), syserror.ErrWouldBlock; got != want { + if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, 0, 1, linux.SEEK_SET, nil), syserror.ErrWouldBlock; got != want { t.Fatalf("fd.Impl().LockPOSIX failed: got = %v, want = %v", got, want) } - if err := fd.Impl().UnlockPOSIX(ctx, uid1, rng1); err != nil { + if err := fd.Impl().UnlockPOSIX(ctx, uid1, 0, 1, linux.SEEK_SET); err != nil { t.Fatalf("fd.Impl().UnlockPOSIX failed: err = %v", err) } } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 71a7522af..a94333ee0 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -36,11 +36,11 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sentry/vfs/memxattr" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -310,7 +310,7 @@ type inode struct { ctime int64 // nanoseconds mtime int64 // nanoseconds - locks lock.FileLocks + locks vfs.FileLocks // Inotify watches for this inode. watches vfs.Watches @@ -761,9 +761,26 @@ func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name s // 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} + +// Sync implements vfs.FileDescriptionImpl.Sync. It does nothing because all +// filesystem state is in-memory. +func (*fileDescription) Sync(context.Context) error { + return nil +} diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 48911240f..4b7d234a4 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -29,6 +29,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" ) // FDFlags define flags for an individual descriptor. @@ -148,7 +149,12 @@ func (f *FDTable) drop(file *fs.File) { // dropVFS2 drops the table reference. func (f *FDTable) dropVFS2(file *vfs.FileDescription) { - // TODO(gvisor.dev/issue/1480): Release locks. + // Release any POSIX lock possibly held by the FDTable. Range {0, 0} means the + // entire file. + err := file.UnlockPOSIX(context.Background(), f, 0, 0, linux.SEEK_SET) + if err != nil && err != syserror.ENOLCK { + panic(fmt.Sprintf("UnlockPOSIX failed: %v", err)) + } // Generate inotify events. ev := uint32(linux.IN_CLOSE_NOWRITE) @@ -157,7 +163,7 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { } file.Dentry().InotifyWithParent(ev, 0, vfs.PathEvent) - // Drop the table reference. + // Drop the table's reference. file.DecRef() } diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 0db546b98..449643118 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -26,8 +26,8 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index c0e9ee1f4..a4519363f 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -63,12 +63,12 @@ func NewVFSPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *VFSPipe { // Preconditions: statusFlags should not contain an open access mode. func (vp *VFSPipe) ReaderWriterPair(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) (*vfs.FileDescription, *vfs.FileDescription) { // Connected pipes share the same locks. - locks := &lock.FileLocks{} + locks := &vfs.FileLocks{} return vp.newFD(mnt, vfsd, linux.O_RDONLY|statusFlags, locks), vp.newFD(mnt, vfsd, linux.O_WRONLY|statusFlags, locks) } // Open opens the pipe represented by vp. -func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *lock.FileLocks) (*vfs.FileDescription, error) { +func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *vfs.FileLocks) (*vfs.FileDescription, error) { vp.mu.Lock() defer vp.mu.Unlock() @@ -130,7 +130,7 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s } // Preconditions: vp.mu must be held. -func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *lock.FileLocks) *vfs.FileDescription { +func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *vfs.FileLocks) *vfs.FileDescription { fd := &VFSPipeFD{ pipe: &vp.pipe, } @@ -451,3 +451,13 @@ func spliceOrTee(ctx context.Context, dst, src *VFSPipeFD, count int64, removeFr } return n, err } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *VFSPipeFD) 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) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *VFSPipeFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 60c9896fc..ff81ea6e6 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -26,6 +26,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/hostfd", "//pkg/sentry/inet", @@ -34,7 +35,6 @@ go_library( "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip/stack", diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index c11e82c10..a92aed2c9 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -324,7 +324,7 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr return nil, syserr.ErrInvalidArgument } - // Whitelist options and constrain option length. + // Only allow known and safe options. optlen := getSockOptLen(t, level, name) switch level { case linux.SOL_IP: @@ -369,7 +369,7 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr // SetSockOpt implements socket.Socket.SetSockOpt. func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { - // Whitelist options and constrain option length. + // Only allow known and safe options. optlen := setSockOptLen(t, level, name) switch level { case linux.SOL_IP: @@ -415,7 +415,7 @@ func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt [] // RecvMsg implements socket.Socket.RecvMsg. func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { - // Whitelist flags. + // Only allow known and safe flags. // // FIXME(jamieliu): We can't support MSG_ERRQUEUE because it uses ancillary // messages that gvisor/pkg/tcpip/transport/unix doesn't understand. Kill the @@ -537,7 +537,7 @@ func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // SendMsg implements socket.Socket.SendMsg. func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { - // Whitelist flags. + // Only allow known and safe flags. if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { return 0, syserr.ErrInvalidArgument } diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index 027add1fd..ad5f64799 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -21,12 +21,12 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -61,7 +61,7 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in fd: fd, }, } - s.LockFD.Init(&lock.FileLocks{}) + s.LockFD.Init(&vfs.FileLocks{}) if err := fdnotifier.AddFD(int32(fd), &s.queue); err != nil { return nil, syserr.FromError(err) } @@ -134,6 +134,16 @@ func (s *socketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs return int64(n), err } +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *socketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *socketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} + type socketProviderVFS2 struct { family int } diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 66015e2bc..f7abe77d3 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -41,19 +41,6 @@ const errorTargetName = "ERROR" // change the destination port/destination IP for packets. const redirectTargetName = "REDIRECT" -// Metadata is used to verify that we are correctly serializing and -// deserializing iptables into structs consumable by the iptables tool. We save -// a metadata struct when the tables are written, and when they are read out we -// verify that certain fields are the same. -// -// metadata is used by this serialization/deserializing code, not netstack. -type metadata struct { - HookEntry [linux.NF_INET_NUMHOOKS]uint32 - Underflow [linux.NF_INET_NUMHOOKS]uint32 - NumEntries uint32 - Size uint32 -} - // enableLogging controls whether to log the (de)serialization of netfilter // structs between userspace and netstack. These logs are useful when // developing iptables, but can pollute sentry logs otherwise. @@ -83,29 +70,13 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT return linux.IPTGetinfo{}, syserr.FromError(err) } - // Find the appropriate table. - table, err := findTable(stack, info.Name) + _, info, err := convertNetstackToBinary(stack, info.Name) if err != nil { - nflog("%v", err) + nflog("couldn't convert iptables: %v", err) return linux.IPTGetinfo{}, syserr.ErrInvalidArgument } - // Get the hooks that apply to this table. - info.ValidHooks = table.ValidHooks() - - // Grab the metadata struct, which is used to store information (e.g. - // the number of entries) that applies to the user's encoding of - // iptables, but not netstack's. - metadata := table.Metadata().(metadata) - - // Set values from metadata. - info.HookEntry = metadata.HookEntry - info.Underflow = metadata.Underflow - info.NumEntries = metadata.NumEntries - info.Size = metadata.Size - nflog("returning info: %+v", info) - return info, nil } @@ -118,23 +89,13 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen return linux.KernelIPTGetEntries{}, syserr.FromError(err) } - // Find the appropriate table. - table, err := findTable(stack, userEntries.Name) - if err != nil { - nflog("%v", err) - return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument - } - // Convert netstack's iptables rules to something that the iptables // tool can understand. - entries, meta, err := convertNetstackToBinary(userEntries.Name.String(), table) + entries, _, err := convertNetstackToBinary(stack, userEntries.Name) if err != nil { nflog("couldn't read entries: %v", err) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } - if meta != table.Metadata().(metadata) { - panic(fmt.Sprintf("Table %q metadata changed between writing and reading. Was saved as %+v, but is now %+v", userEntries.Name.String(), table.Metadata().(metadata), meta)) - } if binary.Size(entries) > uintptr(outLen) { nflog("insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument @@ -143,44 +104,26 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen return entries, nil } -func findTable(stk *stack.Stack, tablename linux.TableName) (stack.Table, error) { - table, ok := stk.IPTables().GetTable(tablename.String()) - if !ok { - return stack.Table{}, fmt.Errorf("couldn't find table %q", tablename) - } - return table, nil -} - -// FillIPTablesMetadata populates stack's IPTables with metadata. -func FillIPTablesMetadata(stk *stack.Stack) { - stk.IPTables().ModifyTables(func(tables map[string]stack.Table) { - // In order to fill in the metadata, we have to translate ipt from its - // netstack format to Linux's giant-binary-blob format. - for name, table := range tables { - _, metadata, err := convertNetstackToBinary(name, table) - if err != nil { - panic(fmt.Errorf("Unable to set default IP tables: %v", err)) - } - table.SetMetadata(metadata) - tables[name] = table - } - }) -} - // convertNetstackToBinary converts the iptables as stored in netstack to the // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, // jumping to those offsets, parsing again, etc. -func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelIPTGetEntries, metadata, error) { - // Return values. +func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (linux.KernelIPTGetEntries, linux.IPTGetinfo, error) { + table, ok := stack.IPTables().GetTable(tablename.String()) + if !ok { + return linux.KernelIPTGetEntries{}, linux.IPTGetinfo{}, fmt.Errorf("couldn't find table %q", tablename) + } + var entries linux.KernelIPTGetEntries - var meta metadata + var info linux.IPTGetinfo + info.ValidHooks = table.ValidHooks() // The table name has to fit in the struct. if linux.XT_TABLE_MAXNAMELEN < len(tablename) { - return linux.KernelIPTGetEntries{}, metadata{}, fmt.Errorf("table name %q too long.", tablename) + return linux.KernelIPTGetEntries{}, linux.IPTGetinfo{}, fmt.Errorf("table name %q too long", tablename) } - copy(entries.Name[:], tablename) + copy(info.Name[:], tablename[:]) + copy(entries.Name[:], tablename[:]) for ruleIdx, rule := range table.Rules { nflog("convert to binary: current offset: %d", entries.Size) @@ -189,14 +132,14 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI for hook, hookRuleIdx := range table.BuiltinChains { if hookRuleIdx == ruleIdx { nflog("convert to binary: found hook %d at offset %d", hook, entries.Size) - meta.HookEntry[hook] = entries.Size + info.HookEntry[hook] = entries.Size } } // Is this a chain underflow point? for underflow, underflowRuleIdx := range table.Underflows { if underflowRuleIdx == ruleIdx { nflog("convert to binary: found underflow %d at offset %d", underflow, entries.Size) - meta.Underflow[underflow] = entries.Size + info.Underflow[underflow] = entries.Size } } @@ -251,12 +194,12 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI entries.Size += uint32(entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) - meta.NumEntries++ + info.NumEntries++ } - nflog("convert to binary: finished with an marshalled size of %d", meta.Size) - meta.Size = entries.Size - return entries, meta, nil + nflog("convert to binary: finished with an marshalled size of %d", info.Size) + info.Size = entries.Size + return entries, info, nil } func marshalTarget(target stack.Target) []byte { @@ -569,12 +512,6 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { // - There are no chains without an unconditional final rule. // - There are no chains without an unconditional underflow rule. - table.SetMetadata(metadata{ - HookEntry: replace.HookEntry, - Underflow: replace.Underflow, - NumEntries: replace.NumEntries, - Size: replace.Size, - }) stk.IPTables().ReplaceTable(replace.Name.String(), table) return nil diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 420e573c9..d5ca3ac56 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -20,6 +20,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", @@ -29,7 +30,6 @@ go_library( "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index 8bfee5193..dbcd8b49a 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -18,12 +18,12 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -78,7 +78,7 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV sendBufferSize: defaultSendBufferSize, }, } - fd.LockFD.Init(&lock.FileLocks{}) + fd.LockFD.Init(&vfs.FileLocks{}) return fd, nil } @@ -140,3 +140,13 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs n, err := s.sendMsg(ctx, src, nil, 0, socket.ControlMessages{}) return int64(n), err.ToError() } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index 0f592ecc3..ea6ebd0e2 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -28,6 +28,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", @@ -37,7 +38,6 @@ go_library( "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 738277391..c0b63a803 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -191,6 +191,7 @@ var Metrics = tcpip.Stats{ MalformedPacketsReceived: mustCreateMetric("/netstack/udp/malformed_packets_received", "Number of incoming UDP datagrams dropped due to the UDP header being in a malformed state."), PacketsSent: mustCreateMetric("/netstack/udp/packets_sent", "Number of UDP datagrams sent."), PacketSendErrors: mustCreateMetric("/netstack/udp/packet_send_errors", "Number of UDP datagrams failed to be sent."), + ChecksumErrors: mustCreateMetric("/netstack/udp/checksum_errors", "Number of UDP datagrams dropped due to bad checksums."), }, } diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index 1412a4810..d65a89316 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -19,13 +19,13 @@ import ( "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -66,7 +66,7 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu protocol: protocol, }, } - s.LockFD.Init(&lock.FileLocks{}) + s.LockFD.Init(&vfs.FileLocks{}) vfsfd := &s.vfsfd if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, @@ -318,3 +318,13 @@ func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []by return SetSockOpt(t, s, s.Endpoint, level, name, optVal) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 9b44c2b89..d2fb655ea 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -144,7 +143,7 @@ func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - var rs tcp.ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &rs) return inet.TCPBufferSize{ Min: rs.Min, @@ -155,7 +154,7 @@ func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { // SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - rs := tcp.ReceiveBufferSizeOption{ + rs := tcpip.StackReceiveBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -165,7 +164,7 @@ func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { // TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - var ss tcp.SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &ss) return inet.TCPBufferSize{ Min: ss.Min, @@ -176,7 +175,7 @@ func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { // SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - ss := tcp.SendBufferSizeOption{ + ss := tcpip.StackSendBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -186,14 +185,14 @@ func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { // TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. func (s *Stack) TCPSACKEnabled() (bool, error) { - var sack tcp.SACKEnabled + var sack tcpip.StackSACKEnabled err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &sack) return bool(sack), syserr.TranslateNetstackError(err).ToError() } // SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enabled))).ToError() + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enabled))).ToError() } // Statistics implements inet.Stack.Statistics. @@ -314,7 +313,7 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { udp.PacketsSent.Value(), // OutDatagrams. udp.ReceiveBufferErrors.Value(), // RcvbufErrors. 0, // Udp/SndbufErrors. - 0, // Udp/InCsumErrors. + udp.ChecksumErrors.Value(), // Udp/InCsumErrors. 0, // Udp/IgnoredMulti. } default: @@ -366,11 +365,6 @@ func (s *Stack) IPTables() (*stack.IPTables, error) { return s.Stack.IPTables(), nil } -// FillIPTablesMetadata populates stack's IPTables with metadata. -func (s *Stack) FillIPTablesMetadata() { - netfilter.FillIPTablesMetadata(s.Stack) -} - // Resume implements inet.Stack.Resume. func (s *Stack) Resume() { s.Stack.Resume() diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index 7d4cc80fe..cca5e70f1 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -21,6 +21,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", @@ -29,7 +30,6 @@ go_library( "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index 09c6d3b27..a1e49cc57 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -476,6 +476,9 @@ func (e *connectionedEndpoint) Readiness(mask waiter.EventMask) waiter.EventMask // State implements socket.Socket.State. func (e *connectionedEndpoint) State() uint32 { + e.Lock() + defer e.Unlock() + if e.Connected() { return linux.SS_CONNECTED } diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 8c32371a2..ff2149250 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -26,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -53,7 +53,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) - fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &lock.FileLocks{}) + fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &vfs.FileLocks{}) if err != nil { return nil, syserr.FromError(err) } @@ -62,7 +62,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) // NewFileDescription creates and returns a socket file description // corresponding to the given mount and dentry. -func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *lock.FileLocks) (*vfs.FileDescription, error) { +func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *vfs.FileLocks) (*vfs.FileDescription, error) { // You can create AF_UNIX, SOCK_RAW sockets. They're the same as // SOCK_DGRAM and don't require CAP_NET_RAW. if stype == linux.SOCK_RAW { @@ -300,6 +300,16 @@ func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []by return netstack.SetSockOpt(t, s, s.ep, level, name, optVal) } +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} + // providerVFS2 is a unix domain socket provider for VFS2. type providerVFS2 struct{} diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 696e1c8d3..35eba20c5 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -1111,17 +1111,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } -// LINT.ThenChange(vfs2/fd.go) - -const ( - _FADV_NORMAL = 0 - _FADV_RANDOM = 1 - _FADV_SEQUENTIAL = 2 - _FADV_WILLNEED = 3 - _FADV_DONTNEED = 4 - _FADV_NOREUSE = 5 -) - // Fadvise64 implements linux syscall fadvise64(2). // This implementation currently ignores the provided advice. func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { @@ -1146,12 +1135,12 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } switch advice { - case _FADV_NORMAL: - case _FADV_RANDOM: - case _FADV_SEQUENTIAL: - case _FADV_WILLNEED: - case _FADV_DONTNEED: - case _FADV_NOREUSE: + case linux.POSIX_FADV_NORMAL: + case linux.POSIX_FADV_RANDOM: + case linux.POSIX_FADV_SEQUENTIAL: + case linux.POSIX_FADV_WILLNEED: + case linux.POSIX_FADV_DONTNEED: + case linux.POSIX_FADV_NOREUSE: default: return 0, nil, syserror.EINVAL } @@ -1160,8 +1149,6 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, nil } -// LINT.IfChange - func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index f9ccb303c..e68b20bed 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -17,10 +17,12 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -167,8 +169,82 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } err := tmpfs.AddSeals(file, args[2].Uint()) return 0, nil, err + case linux.F_SETLK, linux.F_SETLKW: + return 0, nil, posixLock(t, args, file, cmd) default: // TODO(gvisor.dev/issue/2920): Everything else is not yet supported. return 0, nil, syserror.EINVAL } } + +func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescription, cmd int32) error { + // Copy in the lock request. + flockAddr := args[2].Pointer() + var flock linux.Flock + if _, err := t.CopyIn(flockAddr, &flock); err != nil { + return err + } + + var blocker lock.Blocker + if cmd == linux.F_SETLKW { + blocker = t + } + + switch flock.Type { + case linux.F_RDLCK: + if !file.IsReadable() { + return syserror.EBADF + } + return file.LockPOSIX(t, t.FDTable(), lock.ReadLock, uint64(flock.Start), uint64(flock.Len), flock.Whence, blocker) + + case linux.F_WRLCK: + if !file.IsWritable() { + return syserror.EBADF + } + return file.LockPOSIX(t, t.FDTable(), lock.WriteLock, uint64(flock.Start), uint64(flock.Len), flock.Whence, blocker) + + case linux.F_UNLCK: + return file.UnlockPOSIX(t, t.FDTable(), uint64(flock.Start), uint64(flock.Len), flock.Whence) + + default: + return syserror.EINVAL + } +} + +// Fadvise64 implements fadvise64(2). +// This implementation currently ignores the provided advice. +func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + length := args[2].Int64() + advice := args[3].Int() + + // Note: offset is allowed to be negative. + if length < 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // If the FD refers to a pipe or FIFO, return error. + if _, isPipe := file.Impl().(*pipe.VFSPipeFD); isPipe { + return 0, nil, syserror.ESPIPE + } + + switch advice { + case linux.POSIX_FADV_NORMAL: + case linux.POSIX_FADV_RANDOM: + case linux.POSIX_FADV_SEQUENTIAL: + case linux.POSIX_FADV_WILLNEED: + case linux.POSIX_FADV_DONTNEED: + case linux.POSIX_FADV_NOREUSE: + default: + return 0, nil, syserror.EINVAL + } + + // Sure, whatever. + return 0, nil, nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index 46d3e189c..5dac77e4d 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -313,6 +313,9 @@ func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpath if err != nil { return err } + if len(target) == 0 { + return syserror.ENOENT + } linkpath, err := copyInPath(t, linkpathAddr) if err != nil { return err diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index caa6a98ff..b463edf2a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -108,7 +108,7 @@ func Override() { s.Table[209] = syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}) s.Table[213] = syscalls.Supported("epoll_create", EpollCreate) s.Table[217] = syscalls.Supported("getdents64", Getdents64) - delete(s.Table, 221) // fdavise64 + s.Table[221] = syscalls.PartiallySupported("fadvise64", Fadvise64, "The syscall is 'supported', but ignores all provided advice.", nil) s.Table[232] = syscalls.Supported("epoll_wait", EpollWait) s.Table[233] = syscalls.Supported("epoll_ctl", EpollCtl) s.Table[235] = syscalls.Supported("utimes", Utimes) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 16d9f3a28..642769e7c 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -44,6 +44,7 @@ go_library( "filesystem_impl_util.go", "filesystem_type.go", "inotify.go", + "lock.go", "mount.go", "mount_unsafe.go", "options.go", @@ -72,7 +73,6 @@ 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 66f3105bd..4b9faf2ea 100644 --- a/pkg/sentry/vfs/README.md +++ b/pkg/sentry/vfs/README.md @@ -169,8 +169,6 @@ This construction, which is essentially a type-safe analogue to Linux's - binder, which is similarly far too incomplete to use. - - whitelistfs, which we are already actively attempting to remove. - - Save/restore. For instance, it is unclear if the current implementation of the `state` package supports the inheritance pattern described above. diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 13c48824e..e0538ea53 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -438,14 +438,10 @@ type FileDescriptionImpl interface { UnlockBSD(ctx context.Context, uid lock.UniqueID) error // LockPOSIX tries to acquire a POSIX-style advisory file lock. - // - // TODO(gvisor.dev/issue/1480): POSIX-style file locking - LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error + LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, start, length uint64, whence int16, block lock.Blocker) error // UnlockPOSIX releases a POSIX-style advisory file lock. - // - // TODO(gvisor.dev/issue/1480): POSIX-style file locking - UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error + UnlockPOSIX(ctx context.Context, uid lock.UniqueID, start, length uint64, whence int16) error } // Dirent holds the information contained in struct linux_dirent64. @@ -764,3 +760,13 @@ func (fd *FileDescription) LockBSD(ctx context.Context, lockType lock.LockType, func (fd *FileDescription) UnlockBSD(ctx context.Context) error { return fd.impl.UnlockBSD(ctx, fd) } + +// LockPOSIX locks a POSIX-style file range lock. +func (fd *FileDescription) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, start, end uint64, whence int16, block lock.Blocker) error { + return fd.impl.LockPOSIX(ctx, uid, t, start, end, whence, block) +} + +// UnlockPOSIX unlocks a POSIX-style file range lock. +func (fd *FileDescription) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, start, end uint64, whence int16) error { + return fd.impl.UnlockPOSIX(ctx, uid, start, end, whence) +} diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index af7213dfd..1e66997ce 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" 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" @@ -369,14 +368,19 @@ func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.M // LockFD may be used by most implementations of FileDescriptionImpl.Lock* // functions. Caller must call Init(). type LockFD struct { - locks *lock.FileLocks + locks *FileLocks } // Init initializes fd with FileLocks to use. -func (fd *LockFD) Init(locks *lock.FileLocks) { +func (fd *LockFD) Init(locks *FileLocks) { fd.locks = locks } +// Locks returns the locks associated with this file. +func (fd *LockFD) Locks() *FileLocks { + return fd.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) @@ -388,17 +392,6 @@ func (fd *LockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { 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{} @@ -414,11 +407,11 @@ func (NoLockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (NoLockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { +func (NoLockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, 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 { +func (NoLockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { return syserror.ENOLCK } diff --git a/pkg/sentry/vfs/lock/lock.go b/pkg/sentry/vfs/lock.go index 724dfe743..6c7583a81 100644 --- a/pkg/sentry/vfs/lock/lock.go +++ b/pkg/sentry/vfs/lock.go @@ -17,9 +17,11 @@ // // The actual implementations can be found in the lock package under // sentry/fs/lock. -package lock +package vfs import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -56,7 +58,11 @@ func (fl *FileLocks) UnlockBSD(uid fslock.UniqueID) { } // LockPOSIX tries to acquire a POSIX-style lock on a file region. -func (fl *FileLocks) LockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { +func (fl *FileLocks) LockPOSIX(ctx context.Context, fd *FileDescription, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + rng, err := computeRange(ctx, fd, start, length, whence) + if err != nil { + return err + } if fl.posix.LockRegion(uid, t, rng, block) { return nil } @@ -67,6 +73,37 @@ func (fl *FileLocks) LockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fsloc // // This operation is always successful, even if there did not exist a lock on // the requested region held by uid in the first place. -func (fl *FileLocks) UnlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) { +func (fl *FileLocks) UnlockPOSIX(ctx context.Context, fd *FileDescription, uid fslock.UniqueID, start, length uint64, whence int16) error { + rng, err := computeRange(ctx, fd, start, length, whence) + if err != nil { + return err + } fl.posix.UnlockRegion(uid, rng) + return nil +} + +func computeRange(ctx context.Context, fd *FileDescription, start uint64, length uint64, whence int16) (fslock.LockRange, error) { + var off int64 + switch whence { + case linux.SEEK_SET: + off = 0 + case linux.SEEK_CUR: + // Note that Linux does not hold any mutexes while retrieving the file + // offset, see fs/locks.c:flock_to_posix_lock and fs/locks.c:fcntl_setlk. + curOff, err := fd.Seek(ctx, 0, linux.SEEK_CUR) + if err != nil { + return fslock.LockRange{}, err + } + off = curOff + case linux.SEEK_END: + stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_SIZE}) + if err != nil { + return fslock.LockRange{}, err + } + off = int64(stat.Size) + default: + return fslock.LockRange{}, syserror.EINVAL + } + + return fslock.ComputeRange(int64(start), int64(length), off) } diff --git a/pkg/sentry/vfs/lock/BUILD b/pkg/sentry/vfs/lock/BUILD deleted file mode 100644 index d9ab063b7..000000000 --- a/pkg/sentry/vfs/lock/BUILD +++ /dev/null @@ -1,13 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "lock", - srcs = ["lock.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/fs/lock", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f9647f90e..afe2be8d7 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -94,6 +94,37 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, mode linu return syserror.EACCES } +// MayLink determines whether creating a hard link to a file with the given +// mode, kuid, and kgid is permitted. +// +// This corresponds to Linux's fs/namei.c:may_linkat. +func MayLink(creds *auth.Credentials, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { + // Source inode owner can hardlink all they like; otherwise, it must be a + // safe source. + if CanActAsOwner(creds, kuid) { + return nil + } + + // Only regular files can be hard linked. + if mode.FileType() != linux.S_IFREG { + return syserror.EPERM + } + + // Setuid files should not get pinned to the filesystem. + if mode&linux.S_ISUID != 0 { + return syserror.EPERM + } + + // Executable setgid files should not get pinned to the filesystem, but we + // don't support S_IXGRP anyway. + + // Hardlinking to unreadable or unwritable sources is dangerous. + if err := GenericCheckPermissions(creds, MayRead|MayWrite, mode, kuid, kgid); err != nil { + return syserror.EPERM + } + return nil +} + // AccessTypesForOpenFlags returns the access types required to open a file // with the given OpenOptions.Flags. Note that this is NOT the same thing as // the set of accesses permitted for the opened file: |