diff options
30 files changed, 496 insertions, 344 deletions
diff --git a/pkg/abi/linux/dev.go b/pkg/abi/linux/dev.go index 89f9a793f..fa3ae5f18 100644 --- a/pkg/abi/linux/dev.go +++ b/pkg/abi/linux/dev.go @@ -36,6 +36,10 @@ func DecodeDeviceID(rdev uint32) (uint16, uint32) { // // See Documentations/devices.txt and uapi/linux/major.h. const ( + // UNNAMED_MAJOR is the major device number for "unnamed" devices, whose + // minor numbers are dynamically allocated by the kernel. + UNNAMED_MAJOR = 0 + // MEM_MAJOR is the major device number for "memory" character devices. MEM_MAJOR = 1 diff --git a/pkg/abi/linux/linux_abi_autogen_unsafe.go b/pkg/abi/linux/linux_abi_autogen_unsafe.go index ba13aef11..e9ce6e6a7 100755 --- a/pkg/abi/linux/linux_abi_autogen_unsafe.go +++ b/pkg/abi/linux/linux_abi_autogen_unsafe.go @@ -124,12 +124,12 @@ func (s *Statx) UnmarshalBytes(src []byte) { // Packed implements marshal.Marshallable.Packed. //go:nosplit func (s *Statx) Packed() bool { - return s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() + return s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() } // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. func (s *Statx) MarshalUnsafe(dst []byte) { - if s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() { + if s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() { safecopy.CopyIn(dst, unsafe.Pointer(s)) } else { s.MarshalBytes(dst) @@ -138,7 +138,7 @@ func (s *Statx) MarshalUnsafe(dst []byte) { // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. func (s *Statx) UnmarshalUnsafe(src []byte) { - if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { + if s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() { safecopy.CopyOut(unsafe.Pointer(s), src) } else { s.UnmarshalBytes(src) @@ -148,7 +148,7 @@ func (s *Statx) UnmarshalUnsafe(src []byte) { // CopyOutN implements marshal.Marshallable.CopyOutN. //go:nosplit func (s *Statx) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { - if !s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() { + if !s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() { // Type Statx doesn't have a packed layout in memory, fall back to MarshalBytes. buf := task.CopyScratchBuffer(s.SizeBytes()) // escapes: okay. s.MarshalBytes(buf) // escapes: fallback. diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 94db8fe5c..c03c65445 100755 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -51,21 +51,37 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return nil, nil, syserror.EINVAL } - fs, root := fstype.newFilesystem(vfsObj, creds) - return fs.VFSFilesystem(), root.VFSDentry(), nil + fs, root, err := fstype.newFilesystem(vfsObj, creds) + if err != nil { + return nil, nil, err + } + return fs.Filesystem.VFSFilesystem(), root.VFSDentry(), nil +} + +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 } // newFilesystem creates a new devpts filesystem with root directory and ptmx // master inode. It returns the filesystem and root Dentry. -func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*kernfs.Filesystem, *kernfs.Dentry) { - fs := &kernfs.Filesystem{} - fs.VFSFilesystem().Init(vfsObj, fstype, fs) +func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*filesystem, *kernfs.Dentry, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + + fs := &filesystem{ + devMinor: devMinor, + } + fs.Filesystem.VFSFilesystem().Init(vfsObj, fstype, fs) // Construct the root directory. This is always inode id 1. root := &rootInode{ slaves: make(map[uint32]*slaveInode), } - root.InodeAttrs.Init(creds, 1, linux.ModeDirectory|0555) + root.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, devMinor, 1, linux.ModeDirectory|0555) root.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) root.dentry.Init(root) @@ -74,7 +90,7 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds master := &masterInode{ root: root, } - master.InodeAttrs.Init(creds, 2, linux.ModeCharacterDevice|0666) + master.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, devMinor, 2, linux.ModeCharacterDevice|0666) master.dentry.Init(master) // Add the master as a child of the root. @@ -83,7 +99,13 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds }) root.IncLinks(links) - return fs, &root.dentry + return fs, &root.dentry, nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() } // rootInode is the root directory inode for the devpts mounts. @@ -140,7 +162,7 @@ func (i *rootInode) allocateTerminal(creds *auth.Credentials) (*Terminal, error) } // Linux always uses pty index + 3 as the inode id. See // fs/devpts/inode.c:devpts_pty_new(). - slave.InodeAttrs.Init(creds, uint64(idx+3), linux.ModeCharacterDevice|0600) + slave.InodeAttrs.Init(creds, i.InodeAttrs.DevMajor(), i.InodeAttrs.DevMinor(), uint64(idx+3), linux.ModeCharacterDevice|0600) slave.dentry.Init(slave) i.slaves[idx] = slave diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 9ab8fdc65..e68e37ebc 100755 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -81,6 +81,9 @@ type filesystem struct { // clock is a realtime clock used to set timestamps in file operations. clock ktime.Clock + // devMinor is the filesystem's minor device number. devMinor is immutable. + devMinor uint32 + // uid and gid are the effective KUID and KGID of the filesystem's creator, // and are used as the owner and group for files that don't specify one. // uid and gid are immutable. @@ -399,14 +402,21 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } // Construct the filesystem object. + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + attachFile.close(ctx) + client.Close() + return nil, nil, err + } fs := &filesystem{ mfp: mfp, opts: fsopts, iopts: iopts, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, client: client, clock: ktime.RealtimeClockFromContext(ctx), + devMinor: devMinor, + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, syncableDentries: make(map[*dentry]struct{}), specialFileFDs: make(map[*specialFileFD]struct{}), } @@ -464,6 +474,8 @@ func (fs *filesystem) Release() { // Close the connection to the server. This implicitly clunks all fids. fs.client.Close() } + + fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } // dentry implements vfs.DentryImpl. @@ -796,7 +808,8 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) - // TODO(gvisor.dev/issue/1198): device number + stat.DevMajor = linux.UNNAMED_MAJOR + stat.DevMinor = d.fs.devMinor } func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 144e04905..55de9c438 100755 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -115,15 +115,28 @@ func (filesystemType) Name() string { // // Note that there should only ever be one instance of host.filesystem, // a global mount for host fds. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &filesystem{} +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, err + } + fs := &filesystem{ + devMinor: devMinor, + } fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) - return fs.VFSFilesystem() + return fs.VFSFilesystem(), nil } // filesystem implements vfs.FilesystemImpl. type filesystem struct { kernfs.Filesystem + + devMinor uint32 +} + +func (fs *filesystem) Release() { + fs.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() } func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { @@ -219,7 +232,7 @@ func (i *inode) Mode() linux.FileMode { } // Stat implements kernfs.Inode. -func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { +func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { if opts.Mask&linux.STATX__RESERVED != 0 { return linux.Statx{}, syserror.EINVAL } @@ -227,73 +240,73 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro return linux.Statx{}, syserror.EINVAL } + fs := vfsfs.Impl().(*filesystem) + // Limit our host call only to known flags. mask := opts.Mask & linux.STATX_ALL var s unix.Statx_t err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(mask), &s) - // Fallback to fstat(2), if statx(2) is not supported on the host. - // - // TODO(b/151263641): Remove fallback. if err == syserror.ENOSYS { - return i.fstat(opts) - } else if err != nil { + // Fallback to fstat(2), if statx(2) is not supported on the host. + // + // TODO(b/151263641): Remove fallback. + return i.fstat(fs) + } + if err != nil { return linux.Statx{}, err } - ls := linux.Statx{Mask: mask} - // Unconditionally fill blksize, attributes, and device numbers, as indicated - // by /include/uapi/linux/stat.h. - // - // RdevMajor/RdevMinor are left as zero, so as not to expose host device - // numbers. - // - // TODO(gvisor.dev/issue/1672): Use kernfs-specific, internally defined - // device numbers. If we use the device number from the host, it may collide - // with another sentry-internal device number. We handle device/inode - // numbers without relying on the host to prevent collisions. - ls.Blksize = s.Blksize - ls.Attributes = s.Attributes - ls.AttributesMask = s.Attributes_mask - - if mask&linux.STATX_TYPE != 0 { + // Unconditionally fill blksize, attributes, and device numbers, as + // indicated by /include/uapi/linux/stat.h. Inode number is always + // available, since we use our own rather than the host's. + ls := linux.Statx{ + Mask: linux.STATX_INO, + Blksize: s.Blksize, + Attributes: s.Attributes, + Ino: i.ino, + AttributesMask: s.Attributes_mask, + DevMajor: linux.UNNAMED_MAJOR, + DevMinor: fs.devMinor, + } + + // Copy other fields that were returned by the host. RdevMajor/RdevMinor + // are never copied (and therefore left as zero), so as not to expose host + // device numbers. + ls.Mask |= s.Mask & linux.STATX_ALL + if s.Mask&linux.STATX_TYPE != 0 { ls.Mode |= s.Mode & linux.S_IFMT } - if mask&linux.STATX_MODE != 0 { + if s.Mask&linux.STATX_MODE != 0 { ls.Mode |= s.Mode &^ linux.S_IFMT } - if mask&linux.STATX_NLINK != 0 { + if s.Mask&linux.STATX_NLINK != 0 { ls.Nlink = s.Nlink } - if mask&linux.STATX_UID != 0 { + if s.Mask&linux.STATX_UID != 0 { ls.UID = s.Uid } - if mask&linux.STATX_GID != 0 { + if s.Mask&linux.STATX_GID != 0 { ls.GID = s.Gid } - if mask&linux.STATX_ATIME != 0 { + if s.Mask&linux.STATX_ATIME != 0 { ls.Atime = unixToLinuxStatxTimestamp(s.Atime) } - if mask&linux.STATX_BTIME != 0 { + if s.Mask&linux.STATX_BTIME != 0 { ls.Btime = unixToLinuxStatxTimestamp(s.Btime) } - if mask&linux.STATX_CTIME != 0 { + if s.Mask&linux.STATX_CTIME != 0 { ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime) } - if mask&linux.STATX_MTIME != 0 { + if s.Mask&linux.STATX_MTIME != 0 { ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime) } - if mask&linux.STATX_SIZE != 0 { + if s.Mask&linux.STATX_SIZE != 0 { ls.Size = s.Size } - if mask&linux.STATX_BLOCKS != 0 { + if s.Mask&linux.STATX_BLOCKS != 0 { ls.Blocks = s.Blocks } - // Use our own internal inode number. - if mask&linux.STATX_INO != 0 { - ls.Ino = i.ino - } - return ls, nil } @@ -305,36 +318,30 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro // of a mask or sync flags. fstat(2) does not provide any metadata // equivalent to Statx.Attributes, Statx.AttributesMask, or Statx.Btime, so // those fields remain empty. -func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { +func (i *inode) fstat(fs *filesystem) (linux.Statx, error) { var s unix.Stat_t if err := unix.Fstat(i.hostFD, &s); err != nil { return linux.Statx{}, err } - // Note that rdev numbers are left as 0; do not expose host device numbers. - ls := linux.Statx{ - Mask: linux.STATX_BASIC_STATS, - Blksize: uint32(s.Blksize), - Nlink: uint32(s.Nlink), - UID: s.Uid, - GID: s.Gid, - Mode: uint16(s.Mode), - Size: uint64(s.Size), - Blocks: uint64(s.Blocks), - Atime: timespecToStatxTimestamp(s.Atim), - Ctime: timespecToStatxTimestamp(s.Ctim), - Mtime: timespecToStatxTimestamp(s.Mtim), - } - - // Use our own internal inode number. - // - // TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well. - // If we use the device number from the host, it may collide with another - // sentry-internal device number. We handle device/inode numbers without - // relying on the host to prevent collisions. - ls.Ino = i.ino - - return ls, nil + // As with inode.Stat(), we always use internal device and inode numbers, + // and never expose the host's represented device numbers. + return linux.Statx{ + Mask: linux.STATX_BASIC_STATS, + Blksize: uint32(s.Blksize), + Nlink: uint32(s.Nlink), + UID: s.Uid, + GID: s.Gid, + Mode: uint16(s.Mode), + Ino: i.ino, + Size: uint64(s.Size), + Blocks: uint64(s.Blocks), + Atime: timespecToStatxTimestamp(s.Atim), + Ctime: timespecToStatxTimestamp(s.Ctim), + Mtime: timespecToStatxTimestamp(s.Mtim), + DevMajor: linux.UNNAMED_MAJOR, + DevMinor: fs.devMinor, + }, nil } // SetStat implements kernfs.Inode. @@ -453,8 +460,6 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.F } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. -// -// TODO(gvisor.dev/issue/1672): Implement Waitable interface. type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -471,12 +476,12 @@ type fileDescription struct { // SetStat implements vfs.FileDescriptionImpl. func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - return f.inode.SetStat(ctx, nil, creds, opts) + return f.inode.SetStat(ctx, f.vfsfd.Mount().Filesystem(), creds, opts) } // Stat implements vfs.FileDescriptionImpl. func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) { - return f.inode.Stat(nil, opts) + return f.inode.Stat(f.vfsfd.Mount().Filesystem(), opts) } // Release implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c7779fc11..1568a9d49 100755 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -44,11 +44,11 @@ type DynamicBytesFile struct { var _ Inode = (*DynamicBytesFile)(nil) // Init initializes a dynamic bytes file. -func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) { +func (f *DynamicBytesFile) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) { if perm&^linux.PermissionsMask != 0 { panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) } - f.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) + f.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeRegular|perm) f.data = data } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 615592d5f..982daa2e6 100755 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -192,15 +192,17 @@ func (InodeNotSymlink) Getlink(context.Context, *vfs.Mount) (vfs.VirtualDentry, // // Must be initialized by Init prior to first use. type InodeAttrs struct { - ino uint64 - mode uint32 - uid uint32 - gid uint32 - nlink uint32 + devMajor uint32 + devMinor uint32 + ino uint64 + mode uint32 + uid uint32 + gid uint32 + nlink uint32 } // Init initializes this InodeAttrs. -func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMode) { +func (a *InodeAttrs) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, mode linux.FileMode) { if mode.FileType() == 0 { panic(fmt.Sprintf("No file type specified in 'mode' for InodeAttrs.Init(): mode=0%o", mode)) } @@ -209,6 +211,8 @@ func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMo if mode.FileType() == linux.ModeDirectory { nlink = 2 } + a.devMajor = devMajor + a.devMinor = devMinor atomic.StoreUint64(&a.ino, ino) atomic.StoreUint32(&a.mode, uint32(mode)) atomic.StoreUint32(&a.uid, uint32(creds.EffectiveKUID)) @@ -216,6 +220,16 @@ func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMo atomic.StoreUint32(&a.nlink, nlink) } +// DevMajor returns the device major number. +func (a *InodeAttrs) DevMajor() uint32 { + return a.devMajor +} + +// DevMinor returns the device minor number. +func (a *InodeAttrs) DevMinor() uint32 { + return a.devMinor +} + // Ino returns the inode id. func (a *InodeAttrs) Ino() uint64 { return atomic.LoadUint64(&a.ino) @@ -232,6 +246,8 @@ func (a *InodeAttrs) Mode() linux.FileMode { func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) { var stat linux.Statx stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK + stat.DevMajor = a.devMajor + stat.DevMinor = a.devMinor stat.Ino = atomic.LoadUint64(&a.ino) stat.Mode = uint16(a.Mode()) stat.UID = atomic.LoadUint32(&a.uid) @@ -544,9 +560,9 @@ type StaticDirectory struct { var _ Inode = (*StaticDirectory)(nil) // NewStaticDir creates a new static directory and returns its dentry. -func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { +func NewStaticDir(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { inode := &StaticDirectory{} - inode.Init(creds, ino, perm) + inode.Init(creds, devMajor, devMinor, ino, perm) dentry := &Dentry{} dentry.Init(inode) @@ -559,11 +575,11 @@ func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, chil } // Init initializes StaticDirectory. -func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { +func (s *StaticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode) { if perm&^linux.PermissionsMask != 0 { panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) } - s.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) + s.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeDirectory|perm) } // Open implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 0aa6dc979..2ab3f53fd 100755 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -35,9 +35,9 @@ type StaticSymlink struct { var _ Inode = (*StaticSymlink)(nil) // NewStaticSymlink creates a new symlink file pointing to 'target'. -func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentry { +func NewStaticSymlink(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, target string) *Dentry { inode := &StaticSymlink{} - inode.Init(creds, ino, target) + inode.Init(creds, devMajor, devMinor, ino, target) d := &Dentry{} d.Init(inode) @@ -45,9 +45,9 @@ func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentr } // Init initializes the instance. -func (s *StaticSymlink) Init(creds *auth.Credentials, ino uint64, target string) { +func (s *StaticSymlink) Init(creds *auth.Credentials, devMajor uint32, devMinor uint32, ino uint64, target string) { s.target = target - s.InodeAttrs.Init(creds, ino, linux.ModeSymlink|0777) + s.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeSymlink|0777) } // Readlink implements Inode. diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 5375e5e75..cab771211 100755 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -17,8 +17,11 @@ package pipefs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" @@ -40,20 +43,36 @@ func (filesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile panic("pipefs.filesystemType.GetFilesystem should never be called") } -// TODO(gvisor.dev/issue/1193): -// -// - kernfs does not provide a way to implement statfs, from which we -// should indicate PIPEFS_MAGIC. -// -// - kernfs does not provide a way to override names for -// vfs.FilesystemImpl.PrependPath(); pipefs inodes should use synthetic -// name fmt.Sprintf("pipe:[%d]", inode.ino). +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 +} // NewFilesystem sets up and returns a new vfs.Filesystem implemented by pipefs. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &kernfs.Filesystem{} - fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) - return fs.VFSFilesystem() +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, err + } + fs := &filesystem{ + devMinor: devMinor, + } + fs.Filesystem.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) + return fs.Filesystem.VFSFilesystem(), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + inode := vd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode) + b.PrependComponent(fmt.Sprintf("pipe:[%d]", inode.ino)) + return vfs.PrependPathSyntheticError{} } // inode implements kernfs.Inode. @@ -71,11 +90,11 @@ type inode struct { ctime ktime.Time } -func newInode(ctx context.Context, fs *kernfs.Filesystem) *inode { +func newInode(ctx context.Context, fs *filesystem) *inode { creds := auth.CredentialsFromContext(ctx) return &inode{ pipe: pipe.NewVFSPipe(false /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize), - ino: fs.NextIno(), + ino: fs.Filesystem.NextIno(), uid: creds.EffectiveKUID, gid: creds.EffectiveKGID, ctime: ktime.NowFromContext(ctx), @@ -98,19 +117,20 @@ func (i *inode) Mode() linux.FileMode { func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { ts := linux.NsecToStatxTimestamp(i.ctime.Nanoseconds()) return linux.Statx{ - Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, - Blksize: usermem.PageSize, - Nlink: 1, - UID: uint32(i.uid), - GID: uint32(i.gid), - Mode: pipeMode, - Ino: i.ino, - Size: 0, - Blocks: 0, - Atime: ts, - Ctime: ts, - Mtime: ts, - // TODO(gvisor.dev/issue/1197): Device number. + Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, + Blksize: usermem.PageSize, + Nlink: 1, + UID: uint32(i.uid), + GID: uint32(i.gid), + Mode: pipeMode, + Ino: i.ino, + Size: 0, + Blocks: 0, + Atime: ts, + Ctime: ts, + Mtime: ts, + DevMajor: linux.UNNAMED_MAJOR, + DevMinor: vfsfs.Impl().(*filesystem).devMinor, }, nil } @@ -122,6 +142,9 @@ func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth. return syserror.EPERM } +// TODO(gvisor.dev/issue/1193): kernfs does not provide a way to implement +// statfs, from which we should indicate PIPEFS_MAGIC. + // Open implements kernfs.Inode.Open. func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags) @@ -132,7 +155,7 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr // // Preconditions: mnt.Filesystem() must have been returned by NewFilesystem(). func NewConnectedPipeFDs(ctx context.Context, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, *vfs.FileDescription) { - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + fs := mnt.Filesystem().Impl().(*filesystem) inode := newInode(ctx, fs) var d kernfs.Dentry d.Init(inode) diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 104fc9030..609210253 100755 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -41,6 +41,12 @@ func (FilesystemType) Name() string { return Name } +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { k := kernel.KernelFromContext(ctx) @@ -51,8 +57,13 @@ func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualF if pidns == nil { return nil, nil, fmt.Errorf("procfs requires a PID namespace") } - - procfs := &kernfs.Filesystem{} + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + procfs := &filesystem{ + devMinor: devMinor, + } procfs.VFSFilesystem().Init(vfsObj, &ft, procfs) var cgroups map[string]string @@ -61,21 +72,27 @@ func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualF cgroups = data.Cgroups } - _, dentry := newTasksInode(procfs, k, pidns, cgroups) + _, dentry := procfs.newTasksInode(k, pidns, cgroups) return procfs.VFSFilesystem(), dentry.VFSDentry(), nil } +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + // dynamicInode is an overfitted interface for common Inodes with // dynamicByteSource types used in procfs. type dynamicInode interface { kernfs.Inode vfs.DynamicBytesSource - Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) + Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) } -func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { - inode.Init(creds, ino, inode, perm) +func (fs *filesystem) newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { + inode.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, ino, inode, perm) d := &kernfs.Dentry{} d.Init(inode) diff --git a/pkg/sentry/fsimpl/proc/proc_state_autogen.go b/pkg/sentry/fsimpl/proc/proc_state_autogen.go index 26c7165d3..168e61390 100755 --- a/pkg/sentry/fsimpl/proc/proc_state_autogen.go +++ b/pkg/sentry/fsimpl/proc/proc_state_autogen.go @@ -23,9 +23,9 @@ func (x *subtasksInode) save(m state.Map) { m.Save("InodeAttrs", &x.InodeAttrs) m.Save("OrderedChildren", &x.OrderedChildren) m.Save("AlwaysValid", &x.AlwaysValid) + m.Save("fs", &x.fs) m.Save("task", &x.task) m.Save("pidns", &x.pidns) - m.Save("inoGen", &x.inoGen) m.Save("cgroupControllers", &x.cgroupControllers) } @@ -36,9 +36,9 @@ func (x *subtasksInode) load(m state.Map) { m.Load("InodeAttrs", &x.InodeAttrs) m.Load("OrderedChildren", &x.OrderedChildren) m.Load("AlwaysValid", &x.AlwaysValid) + m.Load("fs", &x.fs) m.Load("task", &x.task) m.Load("pidns", &x.pidns) - m.Load("inoGen", &x.inoGen) m.Load("cgroupControllers", &x.cgroupControllers) } @@ -475,7 +475,7 @@ func (x *tasksInode) save(m state.Map) { m.Save("InodeAttrs", &x.InodeAttrs) m.Save("OrderedChildren", &x.OrderedChildren) m.Save("AlwaysValid", &x.AlwaysValid) - m.Save("inoGen", &x.inoGen) + m.Save("fs", &x.fs) m.Save("pidns", &x.pidns) m.Save("selfSymlink", &x.selfSymlink) m.Save("threadSelfSymlink", &x.threadSelfSymlink) @@ -489,7 +489,7 @@ func (x *tasksInode) load(m state.Map) { m.Load("InodeAttrs", &x.InodeAttrs) m.Load("OrderedChildren", &x.OrderedChildren) m.Load("AlwaysValid", &x.AlwaysValid) - m.Load("inoGen", &x.inoGen) + m.Load("fs", &x.fs) m.Load("pidns", &x.pidns) m.Load("selfSymlink", &x.selfSymlink) m.Load("threadSelfSymlink", &x.threadSelfSymlink) diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index a5cfa8333..36a911db4 100755 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -37,23 +37,23 @@ type subtasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid + fs *filesystem task *kernel.Task pidns *kernel.PIDNamespace - inoGen InoGenerator cgroupControllers map[string]string } var _ kernfs.Inode = (*subtasksInode)(nil) -func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator, cgroupControllers map[string]string) *kernfs.Dentry { +func (fs *filesystem) newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) *kernfs.Dentry { subInode := &subtasksInode{ + fs: fs, task: task, pidns: pidns, - inoGen: inoGen, cgroupControllers: cgroupControllers, } // Note: credentials are overridden by taskOwnedInode. - subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + subInode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) inode := &taskOwnedInode{Inode: subInode, owner: task} @@ -78,7 +78,7 @@ func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, e return nil, syserror.ENOENT } - subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false, i.cgroupControllers) + subTaskDentry := i.fs.newTaskInode(subTask, i.pidns, false, i.cgroupControllers) return subTaskDentry.VFSDentry(), nil } @@ -102,7 +102,7 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(tid), 10), Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 66419d91b..482055db1 100755 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -43,45 +43,45 @@ type taskInode struct { var _ kernfs.Inode = (*taskInode)(nil) -func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { +func (fs *filesystem) newTaskInode(task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { // TODO(gvisor.dev/issue/164): Fail with ESRCH if task exited. contents := map[string]*kernfs.Dentry{ - "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), - "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), - "comm": newComm(task, inoGen.NextIno(), 0444), - "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), - "exe": newExeSymlink(task, inoGen.NextIno()), - "fd": newFDDirInode(task, inoGen), - "fdinfo": newFDInfoDirInode(task, inoGen), - "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), - "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), - "mountinfo": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountInfoData{task: task}), - "mounts": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountsData{task: task}), - "net": newTaskNetDir(task, inoGen), - "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ - "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), - "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), - "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), + "auxv": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &auxvData{task: task}), + "cmdline": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), + "comm": fs.newComm(task, fs.NextIno(), 0444), + "environ": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), + "exe": fs.newExeSymlink(task, fs.NextIno()), + "fd": fs.newFDDirInode(task), + "fdinfo": fs.newFDInfoDirInode(task), + "gid_map": fs.newTaskOwnedFile(task, fs.NextIno(), 0644, &idMapData{task: task, gids: true}), + "io": fs.newTaskOwnedFile(task, fs.NextIno(), 0400, newIO(task, isThreadGroup)), + "maps": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &mapsData{task: task}), + "mountinfo": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &mountInfoData{task: task}), + "mounts": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &mountsData{task: task}), + "net": fs.newTaskNetDir(task), + "ns": fs.newTaskOwnedDir(task, fs.NextIno(), 0511, map[string]*kernfs.Dentry{ + "net": fs.newNamespaceSymlink(task, fs.NextIno(), "net"), + "pid": fs.newNamespaceSymlink(task, fs.NextIno(), "pid"), + "user": fs.newNamespaceSymlink(task, fs.NextIno(), "user"), }), - "oom_score": newTaskOwnedFile(task, inoGen.NextIno(), 0444, newStaticFile("0\n")), - "oom_score_adj": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &oomScoreAdj{task: task}), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), - "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), + "oom_score": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, newStaticFile("0\n")), + "oom_score_adj": fs.newTaskOwnedFile(task, fs.NextIno(), 0644, &oomScoreAdj{task: task}), + "smaps": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &smapsData{task: task}), + "stat": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &statmData{task: task}), + "status": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &statusData{task: task, pidns: pidns}), + "uid_map": fs.newTaskOwnedFile(task, fs.NextIno(), 0644, &idMapData{task: task, gids: false}), } if isThreadGroup { - contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) + contents["task"] = fs.newSubtasks(task, pidns, cgroupControllers) } if len(cgroupControllers) > 0 { - contents["cgroup"] = newTaskOwnedFile(task, inoGen.NextIno(), 0444, newCgroupData(cgroupControllers)) + contents["cgroup"] = fs.newTaskOwnedFile(task, fs.NextIno(), 0444, newCgroupData(cgroupControllers)) } taskInode := &taskInode{task: task} // Note: credentials are overridden by taskOwnedInode. - taskInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + taskInode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) inode := &taskOwnedInode{Inode: taskInode, owner: task} dentry := &kernfs.Dentry{} @@ -126,9 +126,9 @@ type taskOwnedInode struct { var _ kernfs.Inode = (*taskOwnedInode)(nil) -func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { +func (fs *filesystem) newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, inode, perm) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, inode, perm) taskInode := &taskOwnedInode{Inode: inode, owner: task} d := &kernfs.Dentry{} @@ -136,11 +136,11 @@ func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode return d } -func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { +func (fs *filesystem) newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { dir := &kernfs.StaticDirectory{} // Note: credentials are overridden by taskOwnedInode. - dir.Init(task.Credentials(), ino, perm) + dir.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, perm) inode := &taskOwnedInode{Inode: dir, owner: task} d := &kernfs.Dentry{} diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 8ad976073..44ccc9e4a 100755 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -53,8 +53,8 @@ func taskFDExists(t *kernel.Task, fd int32) bool { } type fdDir struct { - inoGen InoGenerator - task *kernel.Task + fs *filesystem + task *kernel.Task // When produceSymlinks is set, dirents produces for the FDs are reported // as symlink. Otherwise, they are reported as regular files. @@ -85,7 +85,7 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, abs dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(fd), 10), Type: typ, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { @@ -110,15 +110,15 @@ type fdDirInode struct { var _ kernfs.Inode = (*fdDirInode)(nil) -func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { +func (fs *filesystem) newFDDirInode(task *kernel.Task) *kernfs.Dentry { inode := &fdDirInode{ fdDir: fdDir{ - inoGen: inoGen, + fs: fs, task: task, produceSymlink: true, }, } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + inode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -137,7 +137,7 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro if !taskFDExists(i.task, fd) { return nil, syserror.ENOENT } - taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno()) + taskDentry := i.fs.newFDSymlink(i.task, fd, i.fs.NextIno()) return taskDentry.VFSDentry(), nil } @@ -186,12 +186,12 @@ type fdSymlink struct { var _ kernfs.Inode = (*fdSymlink)(nil) -func newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { +func (fs *filesystem) newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { inode := &fdSymlink{ task: task, fd: fd, } - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -234,14 +234,14 @@ type fdInfoDirInode struct { var _ kernfs.Inode = (*fdInfoDirInode)(nil) -func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { +func (fs *filesystem) newFDInfoDirInode(task *kernel.Task) *kernfs.Dentry { inode := &fdInfoDirInode{ fdDir: fdDir{ - inoGen: inoGen, - task: task, + fs: fs, + task: task, }, } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + inode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -264,7 +264,7 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, task: i.task, fd: fd, } - dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) + dentry := i.fs.newTaskOwnedFile(i.task, i.fs.NextIno(), 0444, data) return dentry.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 515f25327..2f297e48a 100755 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -241,9 +241,9 @@ type commInode struct { task *kernel.Task } -func newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { +func (fs *filesystem) newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { inode := &commInode{task: task} - inode.DynamicBytesFile.Init(task.Credentials(), ino, &commData{task: task}, perm) + inode.DynamicBytesFile.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, &commData{task: task}, perm) d := &kernfs.Dentry{} d.Init(inode) @@ -596,9 +596,9 @@ type exeSymlink struct { var _ kernfs.Inode = (*exeSymlink)(nil) -func newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { +func (fs *filesystem) newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { inode := &exeSymlink{task: task} - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -729,7 +729,7 @@ type namespaceSymlink struct { task *kernel.Task } -func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { +func (fs *filesystem) newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { // Namespace symlinks should contain the namespace name and the inode number // for the namespace instance, so for example user:[123456]. We currently fake // the inode number by sticking the symlink inode in its place. @@ -737,7 +737,7 @@ func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentr inode := &namespaceSymlink{task: task} // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, target) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, target) taskInode := &taskOwnedInode{Inode: inode, owner: task} d := &kernfs.Dentry{} @@ -780,11 +780,11 @@ type namespaceInode struct { var _ kernfs.Inode = (*namespaceInode)(nil) // Init initializes a namespace inode. -func (i *namespaceInode) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { +func (i *namespaceInode) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode) { if perm&^linux.PermissionsMask != 0 { panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) } - i.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) + i.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeRegular|perm) } // Open implements Inode.Open. diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 9c329341a..6bde27376 100755 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -37,7 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { +func (fs *filesystem) newTaskNetDir(task *kernel.Task) *kernfs.Dentry { k := task.Kernel() pidns := task.PIDNamespace() root := auth.NewRootCredentials(pidns.UserNamespace()) @@ -57,37 +57,37 @@ func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task // network namespace. contents = map[string]*kernfs.Dentry{ - "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), - "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), + "dev": fs.newDentry(root, fs.NextIno(), 0444, &netDevData{stack: stack}), + "snmp": fs.newDentry(root, fs.NextIno(), 0444, &netSnmpData{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, if the file contains a header the stub is just the header // otherwise it is an empty file. - "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), - "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), - "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), - "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), - "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), + "arp": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(arp)), + "netlink": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(netlink)), + "netstat": fs.newDentry(root, fs.NextIno(), 0444, &netStatData{}), + "packet": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(packet)), + "protocols": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(protocols)), // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, // high res timer ticks per sec (ClockGetres returns 1ns resolution). - "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), - "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), - "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), - "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), - "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), - "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), + "psched": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(psched)), + "ptype": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(ptype)), + "route": fs.newDentry(root, fs.NextIno(), 0444, &netRouteData{stack: stack}), + "tcp": fs.newDentry(root, fs.NextIno(), 0444, &netTCPData{kernel: k}), + "udp": fs.newDentry(root, fs.NextIno(), 0444, &netUDPData{kernel: k}), + "unix": fs.newDentry(root, fs.NextIno(), 0444, &netUnixData{kernel: k}), } if stack.SupportsIPv6() { - contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) - contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) - contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) - contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) + contents["if_inet6"] = fs.newDentry(root, fs.NextIno(), 0444, &ifinet6{stack: stack}) + contents["ipv6_route"] = fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")) + contents["tcp6"] = fs.newDentry(root, fs.NextIno(), 0444, &netTCP6Data{kernel: k}) + contents["udp6"] = fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(upd6)) } } - return newTaskOwnedDir(task, inoGen.NextIno(), 0555, contents) + return fs.newTaskOwnedDir(task, fs.NextIno(), 0555, contents) } // ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 5aeda8c9b..b51d43954 100755 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -33,11 +33,6 @@ const ( threadSelfName = "thread-self" ) -// InoGenerator generates unique inode numbers for a given filesystem. -type InoGenerator interface { - NextIno() uint64 -} - // tasksInode represents the inode for /proc/ directory. // // +stateify savable @@ -48,8 +43,8 @@ type tasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid - inoGen InoGenerator - pidns *kernel.PIDNamespace + fs *filesystem + pidns *kernel.PIDNamespace // '/proc/self' and '/proc/thread-self' have custom directory offsets in // Linux. So handle them outside of OrderedChildren. @@ -64,29 +59,29 @@ type tasksInode struct { var _ kernfs.Inode = (*tasksInode)(nil) -func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { +func (fs *filesystem) newTasksInode(k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ - "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), - "filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), - "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), - "sys": newSysDir(root, inoGen, k), - "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), - "net": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), - "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), + "cpuinfo": fs.newDentry(root, fs.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), + "filesystems": fs.newDentry(root, fs.NextIno(), 0444, &filesystemsData{}), + "loadavg": fs.newDentry(root, fs.NextIno(), 0444, &loadavgData{}), + "sys": fs.newSysDir(root, k), + "meminfo": fs.newDentry(root, fs.NextIno(), 0444, &meminfoData{}), + "mounts": kernfs.NewStaticSymlink(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), "self/mounts"), + "net": kernfs.NewStaticSymlink(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), "self/net"), + "stat": fs.newDentry(root, fs.NextIno(), 0444, &statData{}), + "uptime": fs.newDentry(root, fs.NextIno(), 0444, &uptimeData{}), + "version": fs.newDentry(root, fs.NextIno(), 0444, &versionData{}), } inode := &tasksInode{ pidns: pidns, - inoGen: inoGen, - selfSymlink: newSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), - threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), + fs: fs, + selfSymlink: fs.newSelfSymlink(root, fs.NextIno(), pidns).VFSDentry(), + threadSelfSymlink: fs.newThreadSelfSymlink(root, fs.NextIno(), pidns).VFSDentry(), cgroupControllers: cgroupControllers, } - inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) + inode.InodeAttrs.Init(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -118,7 +113,7 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } - taskDentry := newTaskInode(i.inoGen, task, i.pidns, true, i.cgroupControllers) + taskDentry := i.fs.newTaskInode(task, i.pidns, true, i.cgroupControllers) return taskDentry.VFSDentry(), nil } @@ -144,7 +139,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback dirent := vfs.Dirent{ Name: selfName, Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { @@ -156,7 +151,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback dirent := vfs.Dirent{ Name: threadSelfName, Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { @@ -189,7 +184,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(tid), 10), Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, } if err := cb.Handle(dirent); err != nil { diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index e5f13b69e..7d8983aa5 100755 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -41,9 +41,9 @@ type selfSymlink struct { var _ kernfs.Inode = (*selfSymlink)(nil) -func newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { +func (fs *filesystem) newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { inode := &selfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) + inode.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -83,9 +83,9 @@ type threadSelfSymlink struct { var _ kernfs.Inode = (*threadSelfSymlink)(nil) -func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { +func (fs *filesystem) newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { inode := &threadSelfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) + inode.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index 0e90e02fe..6dac2afa4 100755 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -30,89 +30,89 @@ import ( ) // newSysDir returns the dentry corresponding to /proc/sys directory. -func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), - "shmall": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMALL)), - "shmmax": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMAX)), - "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), +func (fs *filesystem) newSysDir(root *auth.Credentials, k *kernel.Kernel) *kernfs.Dentry { + return kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "kernel": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "hostname": fs.newDentry(root, fs.NextIno(), 0444, &hostnameData{}), + "shmall": fs.newDentry(root, fs.NextIno(), 0444, shmData(linux.SHMALL)), + "shmmax": fs.newDentry(root, fs.NextIno(), 0444, shmData(linux.SHMMAX)), + "shmmni": fs.newDentry(root, fs.NextIno(), 0444, shmData(linux.SHMMNI)), }), - "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{k: k}), - "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), + "vm": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "mmap_min_addr": fs.newDentry(root, fs.NextIno(), 0444, &mmapMinAddrData{k: k}), + "overcommit_memory": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0\n")), }), - "net": newSysNetDir(root, inoGen, k), + "net": fs.newSysNetDir(root, k), }) } // newSysNetDir returns the dentry corresponding to /proc/sys/net directory. -func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { +func (fs *filesystem) newSysNetDir(root *auth.Credentials, k *kernel.Kernel) *kernfs.Dentry { var contents map[string]*kernfs.Dentry // TODO(gvisor.dev/issue/1833): Support for using the network stack in the // network namespace of the calling process. if stack := k.RootNetworkNamespace().Stack(); stack != nil { contents = map[string]*kernfs.Dentry{ - "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), + "ipv4": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "tcp_sack": fs.newDentry(root, fs.NextIno(), 0644, &tcpSackData{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the // value closest to the actual netstack behavior or any empty file, all // of these files will have mode 0444 (read-only for all users). - "ip_local_port_range": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("16000 65535")), - "ip_local_reserved_ports": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "ipfrag_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("30")), - "ip_nonlocal_bind": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "ip_no_pmtu_disc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "ip_local_port_range": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("16000 65535")), + "ip_local_reserved_ports": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")), + "ipfrag_time": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("30")), + "ip_nonlocal_bind": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "ip_no_pmtu_disc": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), // tcp_allowed_congestion_control tell the user what they are able to // do as an unprivledged process so we leave it empty. - "tcp_allowed_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_available_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), - "tcp_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), + "tcp_allowed_congestion_control": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")), + "tcp_available_congestion_control": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("reno")), + "tcp_congestion_control": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("reno")), // Many of the following stub files are features netstack doesn't // support. The unsupported features return "0" to indicate they are // disabled. - "tcp_base_mss": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1280")), - "tcp_dsack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_early_retrans": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen_key": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_invalid_ratelimit": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_intvl": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_probes": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("7200")), - "tcp_mtu_probing": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_no_metrics_save": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_probe_interval": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_probe_threshold": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_retries1": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_retries2": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("15")), - "tcp_rfc1337": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_slow_start_after_idle": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_synack_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "tcp_syn_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_timestamps": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_base_mss": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1280")), + "tcp_dsack": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_early_retrans": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_fack": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen_key": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")), + "tcp_invalid_ratelimit": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_intvl": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_probes": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_time": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("7200")), + "tcp_mtu_probing": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_no_metrics_save": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), + "tcp_probe_interval": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_probe_threshold": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_retries1": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("3")), + "tcp_retries2": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("15")), + "tcp_rfc1337": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), + "tcp_slow_start_after_idle": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), + "tcp_synack_retries": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("5")), + "tcp_syn_retries": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("3")), + "tcp_timestamps": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), }), - "core": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "default_qdisc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("pfifo_fast")), - "message_burst": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("10")), - "message_cost": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "optmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "rmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "rmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "somaxconn": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("128")), - "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "core": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "default_qdisc": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("pfifo_fast")), + "message_burst": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("10")), + "message_cost": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("5")), + "optmem_max": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "rmem_default": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), + "rmem_max": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), + "somaxconn": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("128")), + "wmem_default": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), + "wmem_max": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), }), } } - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents) + return kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, contents) } // mmapMinAddrData implements vfs.DynamicBytesSource for diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 239a9f4b4..ee0828a15 100755 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -16,8 +16,11 @@ package sockfs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -41,19 +44,42 @@ func (filesystemType) Name() string { return "sockfs" } +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 +} + // NewFilesystem sets up and returns a new sockfs filesystem. // // Note that there should only ever be one instance of sockfs.Filesystem, // backing a global socket mount. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &kernfs.Filesystem{} - fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) - return fs.VFSFilesystem() +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, err + } + fs := &filesystem{ + devMinor: devMinor, + } + fs.Filesystem.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) + return fs.Filesystem.VFSFilesystem(), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + inode := vd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode) + b.PrependComponent(fmt.Sprintf("socket:[%d]", inode.InodeAttrs.Ino())) + return vfs.PrependPathSyntheticError{} } // inode implements kernfs.Inode. -// -// TODO(gvisor.dev/issue/1193): Device numbers. type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink @@ -67,11 +93,15 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } // NewDentry constructs and returns a sockfs dentry. -func NewDentry(creds *auth.Credentials, ino uint64) *vfs.Dentry { +// +// Preconditions: mnt.Filesystem() must have been returned by NewFilesystem(). +func NewDentry(creds *auth.Credentials, mnt *vfs.Mount) *vfs.Dentry { + fs := mnt.Filesystem().Impl().(*filesystem) + // File mode matches net/socket.c:sock_alloc. filemode := linux.FileMode(linux.S_IFSOCK | 0600) i := &inode{} - i.Init(creds, ino, filemode) + i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.Filesystem.NextIno(), filemode) d := &kernfs.Dentry{} d.Init(i) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 00f7d6214..0af373604 100755 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -37,6 +37,8 @@ type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. type filesystem struct { kernfs.Filesystem + + devMinor uint32 } // Name implements vfs.FilesystemType.Name. @@ -46,7 +48,14 @@ func (FilesystemType) Name() string { // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + + fs := &filesystem{ + devMinor: devMinor, + } fs.VFSFilesystem().Init(vfsObj, &fsType, fs) k := kernel.KernelFromContext(ctx) maxCPUCores := k.ApplicationCores() @@ -77,6 +86,12 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.VFSFilesystem(), root.VFSDentry(), nil } +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + // dir implements kernfs.Inode. type dir struct { kernfs.InodeAttrs @@ -90,7 +105,7 @@ type dir struct { func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { d := &dir{} - d.InodeAttrs.Init(creds, fs.NextIno(), linux.ModeDirectory|0755) + d.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) d.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) d.dentry.Init(d) @@ -127,7 +142,7 @@ func (c *cpuFile) Generate(ctx context.Context, buf *bytes.Buffer) error { func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode linux.FileMode) *kernfs.Dentry { c := &cpuFile{maxCores: maxCores} - c.DynamicBytesFile.Init(creds, fs.NextIno(), c, mode) + c.DynamicBytesFile.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), c, mode) d := &kernfs.Dentry{} d.Init(c) return d diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index efc931468..405928bd0 100755 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -63,6 +63,9 @@ type filesystem struct { // clock is a realtime clock used to set timestamps in file operations. clock time.Clock + // devMinor is the filesystem's minor device number. devMinor is immutable. + devMinor uint32 + // mu serializes changes to the Dentry tree. mu sync.RWMutex @@ -96,11 +99,6 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt if memFileProvider == nil { panic("MemoryFileProviderFromContext returned nil") } - clock := time.RealtimeClockFromContext(ctx) - fs := filesystem{ - memFile: memFileProvider.MemoryFile(), - clock: clock, - } rootFileType := uint16(linux.S_IFDIR) newFSType := vfs.FilesystemType(&fstype) @@ -114,6 +112,16 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } } + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + clock := time.RealtimeClockFromContext(ctx) + fs := filesystem{ + memFile: memFileProvider.MemoryFile(), + clock: clock, + devMinor: devMinor, + } fs.vfsfs.Init(vfsObj, newFSType, &fs) var root *dentry @@ -125,6 +133,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt case linux.S_IFDIR: root = &fs.newDirectory(creds, 01777).dentry default: + fs.vfsfs.DecRef() return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) } return &fs.vfsfs, &root.vfsd, nil @@ -132,6 +141,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // Release implements vfs.FilesystemImpl.Release. func (fs *filesystem) Release() { + fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } // dentry implements vfs.DentryImpl. @@ -188,8 +198,8 @@ func (d *dentry) DecRef() { // inode represents a filesystem object. type inode struct { - // clock is a realtime clock used to set timestamps in file operations. - clock time.Clock + // fs is the owning filesystem. fs is immutable. + fs *filesystem // refs is a reference count. refs is accessed using atomic memory // operations. @@ -232,7 +242,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, if mode.FileType() == 0 { panic("file type is required in FileMode") } - i.clock = fs.clock + i.fs = fs i.refs = 1 i.mode = uint32(mode) i.uid = uint32(creds.EffectiveKUID) @@ -327,7 +337,8 @@ func (i *inode) statTo(stat *linux.Statx) { stat.Atime = linux.NsecToStatxTimestamp(i.atime) stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) - // TODO(gvisor.dev/issue/1197): Device number. + stat.DevMajor = linux.UNNAMED_MAJOR + stat.DevMinor = i.fs.devMinor switch impl := i.impl.(type) { case *regularFile: stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS @@ -401,7 +412,7 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return syserror.EINVAL } } - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() if mask&linux.STATX_ATIME != 0 { if stat.Atime.Nsec == linux.UTIME_NOW { atomic.StoreInt64(&i.atime, now) @@ -518,7 +529,7 @@ func (i *inode) touchAtime(mnt *vfs.Mount) { if err := mnt.CheckBeginWrite(); err != nil { return } - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() i.mu.Lock() atomic.StoreInt64(&i.atime, now) i.mu.Unlock() @@ -527,7 +538,7 @@ func (i *inode) touchAtime(mnt *vfs.Mount) { // Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). func (i *inode) touchCtime() { - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() i.mu.Lock() atomic.StoreInt64(&i.ctime, now) i.mu.Unlock() @@ -535,7 +546,7 @@ func (i *inode) touchCtime() { // Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). func (i *inode) touchCMtime() { - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() i.mu.Lock() atomic.StoreInt64(&i.mtime, now) atomic.StoreInt64(&i.ctime, now) @@ -545,7 +556,7 @@ func (i *inode) touchCMtime() { // Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds // inode.mu. func (i *inode) touchCMtimeLocked() { - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() atomic.StoreInt64(&i.mtime, now) atomic.StoreInt64(&i.ctime, now) } diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 271ea5faf..3617da8c6 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -373,7 +373,10 @@ func (k *Kernel) Init(args InitKernelArgs) error { return fmt.Errorf("failed to initialize VFS: %v", err) } - pipeFilesystem := pipefs.NewFilesystem(&k.vfs) + pipeFilesystem, err := pipefs.NewFilesystem(&k.vfs) + if err != nil { + return fmt.Errorf("failed to create pipefs filesystem: %v", err) + } defer pipeFilesystem.DecRef() pipeMount, err := k.vfs.NewDisconnectedMount(pipeFilesystem, nil, &vfs.MountOptions{}) if err != nil { @@ -381,7 +384,10 @@ func (k *Kernel) Init(args InitKernelArgs) error { } k.pipeMount = pipeMount - socketFilesystem := sockfs.NewFilesystem(&k.vfs) + socketFilesystem, err := sockfs.NewFilesystem(&k.vfs) + if err != nil { + return fmt.Errorf("failed to create sockfs filesystem: %v", err) + } defer socketFilesystem.DecRef() socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index a8278bffc..677743113 100755 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -50,8 +49,7 @@ var _ = socket.SocketVFS2(&socketVFS2{}) func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol int, fd int, flags uint32) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) s := &socketVFS2{ socketOpsCommon: socketOpsCommon{ diff --git a/pkg/sentry/socket/netlink/provider_vfs2.go b/pkg/sentry/socket/netlink/provider_vfs2.go index dcd92b5cd..bb205be0d 100755 --- a/pkg/sentry/socket/netlink/provider_vfs2.go +++ b/pkg/sentry/socket/netlink/provider_vfs2.go @@ -16,7 +16,6 @@ package netlink import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -52,8 +51,7 @@ func (*socketProviderVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol vfsfd := &s.vfsfd mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index f7d9b2ff4..191970d41 100755 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -53,8 +52,7 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu } mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) s := &SocketVFS2{ socketOpsCommon: socketOpsCommon{ diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 06d838868..45e109361 100755 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -50,8 +49,7 @@ var _ = socket.SocketVFS2(&SocketVFS2{}) // returns a corresponding file description. func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d) if err != nil { diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index adebaeefb..caf770fd5 100755 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -202,7 +202,7 @@ func (fs *anonFilesystem) StatAt(ctx context.Context, rp *ResolvingPath, opts St Ino: 1, Size: 0, Blocks: 0, - DevMajor: 0, + DevMajor: linux.UNNAMED_MAJOR, DevMinor: fs.devMinor, }, nil } diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index bda5576fa..1e9dffc8f 100755 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -103,7 +103,7 @@ func (vfs *VirtualFilesystem) OpenDeviceSpecialFile(ctx context.Context, mnt *Mo } // GetAnonBlockDevMinor allocates and returns an unused minor device number for -// an "anonymous" block device with major number 0. +// an "anonymous" block device with major number UNNAMED_MAJOR. func (vfs *VirtualFilesystem) GetAnonBlockDevMinor() (uint32, error) { vfs.anonBlockDevMinorMu.Lock() defer vfs.anonBlockDevMinorMu.Unlock() diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 8c8bad11c..f802bc9fb 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -334,7 +334,10 @@ func New(args Args) (*Loader, error) { if kernel.VFS2Enabled { // Set up host mount that will be used for imported fds. - hostFilesystem := hostvfs2.NewFilesystem(k.VFS()) + hostFilesystem, err := hostvfs2.NewFilesystem(k.VFS()) + if err != nil { + return nil, fmt.Errorf("failed to create hostfs filesystem: %v", err) + } defer hostFilesystem.DecRef() hostMount, err := k.VFS().NewDisconnectedMount(hostFilesystem, nil, &vfs.MountOptions{}) if err != nil { |