diff options
Diffstat (limited to 'pkg/sentry/fs/proc')
| -rw-r--r-- | pkg/sentry/fs/proc/BUILD | 4 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/cpuinfo.go | 41 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/exec_args.go | 57 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/fds.go | 138 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/file.go | 58 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/inode.go | 96 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/net.go | 55 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/proc.go | 152 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/rpcinet_proc.go | 246 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/seqfile/BUILD | 7 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/seqfile/seqfile.go | 133 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/seqfile/seqfile_test.go | 45 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/sys.go | 116 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/sys_net.go | 325 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/sys_net_state.go | 17 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/sys_net_test.go | 28 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/task.go | 264 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/uid_gid_map.go | 17 | ||||
| -rw-r--r-- | pkg/sentry/fs/proc/uptime.go | 40 |
19 files changed, 1107 insertions, 732 deletions
diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index aff3c3c01..74954f213 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -8,9 +8,9 @@ go_library( "cpuinfo.go", "exec_args.go", "fds.go", - "file.go", "filesystems.go", "fs.go", + "inode.go", "loadavg.go", "meminfo.go", "mounts.go", @@ -32,6 +32,7 @@ go_library( "//pkg/abi/linux", "//pkg/sentry/context", "//pkg/sentry/fs", + "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/proc/device", "//pkg/sentry/fs/proc/seqfile", "//pkg/sentry/fs/ramfs", @@ -45,6 +46,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/proc/cpuinfo.go b/pkg/sentry/fs/proc/cpuinfo.go index f8be06dc3..f756c45bf 100644 --- a/pkg/sentry/fs/proc/cpuinfo.go +++ b/pkg/sentry/fs/proc/cpuinfo.go @@ -15,52 +15,21 @@ package proc import ( - "io" - "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" ) -// cpuinfo is a file describing the CPU capabilities. -// -// Presently cpuinfo never changes, so it doesn't need to be a SeqFile. -// -// +stateify savable -type cpuinfo struct { - ramfs.Entry - - // k is the system kernel. - k *kernel.Kernel -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (c *cpuinfo) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - features := c.k.FeatureSet() +func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { + k := kernel.KernelFromContext(ctx) + features := k.FeatureSet() if features == nil { // Kernel is always initialized with a FeatureSet. panic("cpuinfo read with nil FeatureSet") } - contents := make([]byte, 0, 1024) - for i, max := uint(0), c.k.ApplicationCores(); i < max; i++ { + for i, max := uint(0), k.ApplicationCores(); i < max; i++ { contents = append(contents, []byte(features.CPUInfo(i))...) } - if offset >= int64(len(contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, contents[offset:]) - return int64(n), err -} - -func (p *proc) newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - f := &cpuinfo{ - k: p.k, - } - f.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - - return newFile(f, msrc, fs.SpecialFile, nil) + return newStaticProcInode(ctx, msrc, contents) } diff --git a/pkg/sentry/fs/proc/exec_args.go b/pkg/sentry/fs/proc/exec_args.go index b4896053f..ddda67f54 100644 --- a/pkg/sentry/fs/proc/exec_args.go +++ b/pkg/sentry/fs/proc/exec_args.go @@ -18,12 +18,14 @@ import ( "fmt" "io" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // execArgType enumerates the types of exec arguments that are exposed through @@ -35,12 +37,12 @@ const ( environExecArg ) -// execArgFile is a file containing the exec args (either cmdline or environ) +// execArgInode is a inode containing the exec args (either cmdline or environ) // for a given task. // // +stateify savable -type execArgFile struct { - ramfs.Entry +type execArgInode struct { + fsutil.SimpleFileInode // arg is the type of exec argument this file contains. arg execArgType @@ -49,21 +51,52 @@ type execArgFile struct { t *kernel.Task } +var _ fs.InodeOperations = (*execArgInode)(nil) + // newExecArgFile creates a file containing the exec args of the given type. -func newExecArgFile(t *kernel.Task, msrc *fs.MountSource, arg execArgType) *fs.Inode { +func newExecArgInode(t *kernel.Task, msrc *fs.MountSource, arg execArgType) *fs.Inode { if arg != cmdlineExecArg && arg != environExecArg { panic(fmt.Sprintf("unknown exec arg type %v", arg)) } - f := &execArgFile{ - arg: arg, - t: t, + f := &execArgInode{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + arg: arg, + t: t, } - f.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(f, msrc, fs.SpecialFile, t) + return newProcInode(f, msrc, fs.SpecialFile, t) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (i *execArgInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &execArgFile{ + arg: i.arg, + t: i.t, + }), nil } -// DeprecatedPreadv reads the exec arg from the process's address space.. -func (f *execArgFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// +stateify savable +type execArgFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopWrite `state:"nosave"` + + // arg is the type of exec argument this file contains. + arg execArgType + + // t is the Task to read the exec arg line from. + t *kernel.Task +} + +var _ fs.FileOperations = (*execArgFile)(nil) + +// Read reads the exec arg from the process's address space.. +func (f *execArgFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index 5acbce75e..b8a0a5eff 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -21,11 +21,11 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/kdefs" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -55,7 +55,7 @@ func walkDescriptors(t *kernel.Task, p string, toInode func(*fs.File, kernel.FDF // readDescriptors reads fds in the task starting at offset, and calls the // toDentAttr callback for each to get a DentAttr, which it then emits. This is // a helper for implementing fs.InodeOperations.Readdir. -func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int, toDentAttr func(int) fs.DentAttr) (int, error) { +func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int64, toDentAttr func(int) fs.DentAttr) (int64, error) { var fds kernel.FDs t.WithMuLocked(func(t *kernel.Task) { if fdm := t.FDMap(); fdm != nil { @@ -69,7 +69,7 @@ func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int, toDentAttr func(i } // Find the fd to start at. - idx := sort.SearchInts(fdInts, offset) + idx := sort.SearchInts(fdInts, int(offset)) if idx == len(fdInts) { return offset, nil } @@ -80,28 +80,32 @@ func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int, toDentAttr func(i name := strconv.FormatUint(uint64(fd), 10) if err := c.DirEmit(name, toDentAttr(fd)); err != nil { // Returned offset is the next fd to serialize. - return fd, err + return int64(fd), err } } // We serialized them all. Next offset should be higher than last // serialized fd. - return fd + 1, nil + return int64(fd + 1), nil } -// fd is a single file in /proc/TID/fd/. +// fd implements fs.InodeOperations for a file in /proc/TID/fd/. type fd struct { ramfs.Symlink *fs.File } +var _ fs.InodeOperations = (*fd)(nil) + // newFd returns a new fd based on an existing file. // // This inherits one reference to the file. func newFd(t *kernel.Task, f *fs.File, msrc *fs.MountSource) *fs.Inode { - fd := &fd{File: f} - // RootOwner by default, is overridden in UnstableAttr() - fd.InitSymlink(t, fs.RootOwner, "") - return newFile(fd, msrc, fs.Symlink, t) + fd := &fd{ + // RootOwner overridden by taskOwnedInodeOps.UnstableAttrs(). + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, ""), + File: f, + } + return newProcInode(fd, msrc, fs.Symlink, t) } // GetFile returns the fs.File backing this fd. The dirent and flags @@ -142,7 +146,7 @@ func (f *fd) Close() error { return nil } -// fdDir implements /proc/TID/fd. +// fdDir is an InodeOperations for /proc/TID/fd. // // +stateify savable type fdDir struct { @@ -154,11 +158,15 @@ type fdDir struct { t *kernel.Task } +var _ fs.InodeOperations = (*fdDir)(nil) + // newFdDir creates a new fdDir. func newFdDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - f := &fdDir{t: t} - f.InitDir(t, nil, fs.RootOwner, fs.FilePermissions{User: fs.PermMask{Read: true, Execute: true}}) - return newFile(f, msrc, fs.SpecialDirectory, t) + f := &fdDir{ + Dir: *ramfs.NewDir(t, nil, fs.RootOwner, fs.FilePermissions{User: fs.PermMask{Read: true, Execute: true}}), + t: t, + } + return newProcInode(f, msrc, fs.SpecialDirectory, t) } // Check implements InodeOperations.Check. @@ -191,49 +199,55 @@ func (f *fdDir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent return fs.NewDirent(n, p), nil } -// DeprecatedReaddir lists fds in /proc/TID/fd. -func (f *fdDir) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - return readDescriptors(f.t, dirCtx, offset, func(fd int) fs.DentAttr { - return fs.GenericDentAttr(fs.Symlink, device.ProcDevice) +// GetFile implements fs.FileOperations.GetFile. +func (f *fdDir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + fops := &fdDirFile{ + isInfoFile: false, + t: f.t, + } + return fs.NewFile(ctx, dirent, flags, fops), nil +} + +// +stateify savable +type fdDirFile struct { + fsutil.DirFileOperations `state:"nosave"` + + isInfoFile bool + + t *kernel.Task +} + +var _ fs.FileOperations = (*fdDirFile)(nil) + +// Readdir implements fs.FileOperations.Readdir. +func (f *fdDirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + dirCtx := &fs.DirCtx{ + Serializer: ser, + } + typ := fs.RegularFile + if f.isInfoFile { + typ = fs.Symlink + } + return readDescriptors(f.t, dirCtx, file.Offset(), func(fd int) fs.DentAttr { + return fs.GenericDentAttr(typ, device.ProcDevice) }) } -// fdInfo is a single file in /proc/TID/fdinfo/. +// fdInfoInode is a single file in /proc/TID/fdinfo/. // // +stateify savable -type fdInfo struct { - ramfs.File +type fdInfoInode struct { + staticFileInodeOps file *fs.File flags fs.FileFlags fdFlags kernel.FDFlags } -// newFdInfo returns a new fdInfo based on an existing file. -func newFdInfo(t *kernel.Task, file *fs.File, fdFlags kernel.FDFlags, msrc *fs.MountSource) *fs.Inode { - fdi := &fdInfo{file: file, flags: file.Flags(), fdFlags: fdFlags} - fdi.InitFile(t, fs.RootOwner, fs.FilePermissions{User: fs.PermMask{Read: true}}) - // TODO: Get pos, locks, and other data. For now we only - // have flags. - // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt - - flags := file.Flags().ToLinux() | fdFlags.ToLinuxFileFlags() - fdi.Append([]byte(fmt.Sprintf("flags:\t0%o\n", flags))) - return newFile(fdi, msrc, fs.SpecialFile, t) -} - -// DeprecatedPwritev implements fs.HandleOperations.DeprecatedPwritev. -func (*fdInfo) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - return 0, ramfs.ErrInvalidOp -} - -// Truncate implements fs.InodeOperations.Truncate. -func (*fdInfo) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { - return ramfs.ErrInvalidOp -} +var _ fs.InodeOperations = (*fdInfoInode)(nil) -func (f *fdInfo) Release(ctx context.Context) { - f.File.Release(ctx) +// Release implements fs.InodeOperations.Release. +func (f *fdInfoInode) Release(ctx context.Context) { f.file.DecRef() } @@ -249,25 +263,37 @@ type fdInfoDir struct { // newFdInfoDir creates a new fdInfoDir. func newFdInfoDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - fdid := &fdInfoDir{t: t} - fdid.InitDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0500)) - return newFile(fdid, msrc, fs.SpecialDirectory, t) + fdid := &fdInfoDir{ + Dir: *ramfs.NewDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0500)), + t: t, + } + return newProcInode(fdid, msrc, fs.SpecialDirectory, t) } // Lookup loads an fd in /proc/TID/fdinfo into a Dirent. func (fdid *fdInfoDir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { - n, err := walkDescriptors(fdid.t, p, func(file *fs.File, fdFlags kernel.FDFlags) *fs.Inode { - return newFdInfo(fdid.t, file, fdFlags, dir.MountSource) + inode, err := walkDescriptors(fdid.t, p, func(file *fs.File, fdFlags kernel.FDFlags) *fs.Inode { + // TODO: Using a static inode here means that the + // data can be out-of-date if, for instance, the flags on the + // FD change before we read this file. We should switch to + // generating the data on Read(). Also, we should include pos, + // locks, and other data. For now we only have flags. + // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt + flags := file.Flags().ToLinux() | fdFlags.ToLinuxFileFlags() + contents := []byte(fmt.Sprintf("flags:\t0%o\n", flags)) + return newStaticProcInode(ctx, dir.MountSource, contents) }) if err != nil { return nil, err } - return fs.NewDirent(n, p), nil + return fs.NewDirent(inode, p), nil } -// DeprecatedReaddir lists fds in /proc/TID/fdinfo. -func (fdid *fdInfoDir) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - return readDescriptors(fdid.t, dirCtx, offset, func(fd int) fs.DentAttr { - return fs.GenericDentAttr(fs.RegularFile, device.ProcDevice) - }) +// GetFile implements fs.FileOperations.GetFile. +func (fdid *fdInfoDir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + fops := &fdDirFile{ + isInfoFile: true, + t: fdid.t, + } + return fs.NewFile(ctx, dirent, flags, fops), nil } diff --git a/pkg/sentry/fs/proc/file.go b/pkg/sentry/fs/proc/file.go deleted file mode 100644 index f659e590a..000000000 --- a/pkg/sentry/fs/proc/file.go +++ /dev/null @@ -1,58 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" - "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" -) - -// +stateify savable -type file struct { - fs.InodeOperations - - // nodeType is the file type of this file. - nodeType fs.InodeType - - // t is the associated kernel task that owns this file. - t *kernel.Task -} - -func newFile(node fs.InodeOperations, msrc *fs.MountSource, nodeType fs.InodeType, t *kernel.Task) *fs.Inode { - iops := &file{node, nodeType, t} - sattr := fs.StableAttr{ - DeviceID: device.ProcDevice.DeviceID(), - InodeID: device.ProcDevice.NextIno(), - BlockSize: usermem.PageSize, - Type: nodeType, - } - return fs.NewInode(iops, msrc, sattr) -} - -// UnstableAttr returns all attributes of this file. -func (f *file) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - uattr, err := f.InodeOperations.UnstableAttr(ctx, inode) - if err != nil { - return fs.UnstableAttr{}, err - } - if f.t != nil { - creds := f.t.Credentials() - uattr.Owner = fs.FileOwner{creds.EffectiveKUID, creds.EffectiveKGID} - } - return uattr, nil -} diff --git a/pkg/sentry/fs/proc/inode.go b/pkg/sentry/fs/proc/inode.go new file mode 100644 index 000000000..3c36af5ea --- /dev/null +++ b/pkg/sentry/fs/proc/inode.go @@ -0,0 +1,96 @@ +// Copyright 2018 Google LLC +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "gvisor.googlesource.com/gvisor/pkg/abi/linux" + "gvisor.googlesource.com/gvisor/pkg/sentry/context" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" + "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" + "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" +) + +// taskOwnedInodeOps wraps an fs.InodeOperations and overrides the UnstableAttr +// method to return the task as the owner. +// +// +stateify savable +type taskOwnedInodeOps struct { + fs.InodeOperations + + // t is the task that owns this file. + t *kernel.Task +} + +// UnstableAttr implement fs.InodeOperations.UnstableAttr. +func (i *taskOwnedInodeOps) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { + uattr, err := i.InodeOperations.UnstableAttr(ctx, inode) + if err != nil { + return fs.UnstableAttr{}, err + } + // Set the task owner as the file owner. + creds := i.t.Credentials() + uattr.Owner = fs.FileOwner{creds.EffectiveKUID, creds.EffectiveKGID} + return uattr, nil +} + +// staticFileInodeOps is an InodeOperations implementation that can be used to +// return file contents which are constant. This file is not writable and will +// always have mode 0444. +// +// +stateify savable +type staticFileInodeOps struct { + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes + fsutil.InodeStaticFileGetter +} + +var _ fs.InodeOperations = (*staticFileInodeOps)(nil) + +// newStaticFileInode returns a procfs InodeOperations with static contents. +func newStaticProcInode(ctx context.Context, msrc *fs.MountSource, contents []byte) *fs.Inode { + iops := &staticFileInodeOps{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + InodeStaticFileGetter: fsutil.InodeStaticFileGetter{ + Contents: contents, + }, + } + return newProcInode(iops, msrc, fs.SpecialFile, nil) +} + +// newProcInode creates a new inode from the given inode operations. +func newProcInode(iops fs.InodeOperations, msrc *fs.MountSource, typ fs.InodeType, t *kernel.Task) *fs.Inode { + sattr := fs.StableAttr{ + DeviceID: device.ProcDevice.DeviceID(), + InodeID: device.ProcDevice.NextIno(), + BlockSize: usermem.PageSize, + Type: typ, + } + if t != nil { + iops = &taskOwnedInodeOps{iops, t} + } + return fs.NewInode(iops, msrc, sattr) +} diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 2806d6035..3ed85a538 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -28,35 +28,36 @@ import ( // newNet creates a new proc net entry. func (p *proc) newNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) + var contents map[string]*fs.Inode if s := p.k.NetworkStack(); s != nil && s.SupportsIPv6() { - d.AddChild(ctx, "dev", seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc)) - d.AddChild(ctx, "if_inet6", seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc)) - - // 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. - d.AddChild(ctx, "arp", p.newStubProcFSFile(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device"))) - d.AddChild(ctx, "ipv6_route", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "netlink", p.newStubProcFSFile(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode"))) - d.AddChild(ctx, "netstat", p.newStubProcFSFile(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess"))) - d.AddChild(ctx, "packet", p.newStubProcFSFile(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode"))) - d.AddChild(ctx, "protocols", p.newStubProcFSFile(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em"))) - - // Linux sets these values to: nsec per usec, psched tick in ns, 1000000, - // high res timer ticks per sec (ClockGetres returns 1ns resolution). - psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) - d.AddChild(ctx, "psched", p.newStubProcFSFile(ctx, msrc, []byte(psched))) - - d.AddChild(ctx, "ptype", p.newStubProcFSFile(ctx, msrc, []byte("Type Device Function"))) - d.AddChild(ctx, "route", p.newStubProcFSFile(ctx, msrc, []byte("Iface Destination Gateway Flags RefCnt Use Metric Mask MTU Window IRTT"))) - d.AddChild(ctx, "tcp", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode"))) - d.AddChild(ctx, "tcp6", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode"))) - d.AddChild(ctx, "udp", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops"))) - d.AddChild(ctx, "udp6", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode"))) + contents = map[string]*fs.Inode{ + "dev": seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc), + "if_inet6": seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc), + + // 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": newStaticProcInode(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device")), + "ipv6_route": newStaticProcInode(ctx, msrc, []byte("")), + "netlink": newStaticProcInode(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode")), + "netstat": newStaticProcInode(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess")), + "packet": newStaticProcInode(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode")), + "protocols": newStaticProcInode(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em")), + // Linux sets psched values to: nsec per usec, psched + // tick in ns, 1000000, high res timer ticks per sec + // (ClockGetres returns 1ns resolution). + "psched": newStaticProcInode(ctx, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))), + "ptype": newStaticProcInode(ctx, msrc, []byte("Type Device Function")), + "route": newStaticProcInode(ctx, msrc, []byte("Iface Destination Gateway Flags RefCnt Use Metric Mask MTU Window IRTT")), + "tcp": newStaticProcInode(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")), + "tcp6": newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")), + "udp": newStaticProcInode(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops")), + "udp6": newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")), + } } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // ifinet6 implements seqfile.SeqSource for /proc/net/if_inet6. diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index 70e549c31..d1c699418 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -17,18 +17,17 @@ package proc import ( "fmt" - "io" "sort" "strconv" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/rpcinet" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -46,32 +45,6 @@ type proc struct { pidns *kernel.PIDNamespace } -// stubProcFSFile is a file type that can be used to return file contents -// which are constant. This file is not writable and will always have mode -// 0444. -// -// +stateify savable -type stubProcFSFile struct { - ramfs.Entry - - // contents are the immutable file contents that will always be returned. - contents []byte -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (s *stubProcFSFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - - if offset >= int64(len(s.contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, s.contents[offset:]) - return int64(n), err -} - // New returns the root node of a partial simple procfs. func New(ctx context.Context, msrc *fs.MountSource) (*fs.Inode, error) { k := kernel.KernelFromContext(ctx) @@ -83,29 +56,39 @@ func New(ctx context.Context, msrc *fs.MountSource) (*fs.Inode, error) { return nil, fmt.Errorf("procfs requires a PID namespace") } - p := &proc{k: k, pidns: pidns} - p.InitDir(ctx, map[string]*fs.Inode{ + // Note that these are just the static members. There are dynamic + // members populated in Readdir and Lookup below. + contents := map[string]*fs.Inode{ + "cpuinfo": newCPUInfo(ctx, msrc), "filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), "loadavg": seqfile.NewSeqFileInode(ctx, &loadavgData{}, msrc), "meminfo": seqfile.NewSeqFileInode(ctx, &meminfoData{k}, msrc), - "mounts": newMountsSymlink(ctx, msrc), + "mounts": newProcInode(ramfs.NewSymlink(ctx, fs.RootOwner, "self/mounts"), msrc, fs.Symlink, nil), + "self": newSelf(ctx, pidns, msrc), "stat": seqfile.NewSeqFileInode(ctx, &statData{k}, msrc), + "thread-self": newThreadSelf(ctx, pidns, msrc), + "uptime": newUptime(ctx, msrc), "version": seqfile.NewSeqFileInode(ctx, &versionData{k}, msrc), - }, fs.RootOwner, fs.FilePermsFromMode(0555)) + } + + // Construct the proc InodeOperations. + p := &proc{ + Dir: *ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)), + k: k, + pidns: pidns, + } + + // Add more contents that need proc to be initialized. + p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) - p.AddChild(ctx, "cpuinfo", p.newCPUInfo(ctx, msrc)) // If we're using rpcinet we will let it manage /proc/net. if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - p.AddChild(ctx, "net", newRPCInetProcNet(ctx, msrc)) + contents["net"] = newRPCInetProcNet(ctx, msrc) } else { - p.AddChild(ctx, "net", p.newNetDir(ctx, msrc)) + contents["net"] = p.newNetDir(ctx, msrc) } - p.AddChild(ctx, "self", p.newSelf(ctx, msrc)) - p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) - p.AddChild(ctx, "thread-self", p.newThreadSelf(ctx, msrc)) - p.AddChild(ctx, "uptime", p.newUptime(ctx, msrc)) - return newFile(p, msrc, fs.SpecialDirectory, nil), nil + return newProcInode(p, msrc, fs.SpecialDirectory, nil), nil } // self is a magical link. @@ -118,26 +101,21 @@ type self struct { } // newSelf returns a new "self" node. -func (p *proc) newSelf(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - s := &self{pidns: p.pidns} - s.InitSymlink(ctx, fs.RootOwner, "") - return newFile(s, msrc, fs.Symlink, nil) +func newSelf(ctx context.Context, pidns *kernel.PIDNamespace, msrc *fs.MountSource) *fs.Inode { + s := &self{ + Symlink: *ramfs.NewSymlink(ctx, fs.RootOwner, ""), + pidns: pidns, + } + return newProcInode(s, msrc, fs.Symlink, nil) } // newThreadSelf returns a new "threadSelf" node. -func (p *proc) newThreadSelf(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - s := &threadSelf{pidns: p.pidns} - s.InitSymlink(ctx, fs.RootOwner, "") - return newFile(s, msrc, fs.Symlink, nil) -} - -// newStubProcFsFile returns a procfs file with constant contents. -func (p *proc) newStubProcFSFile(ctx context.Context, msrc *fs.MountSource, c []byte) *fs.Inode { - u := &stubProcFSFile{ - contents: c, +func newThreadSelf(ctx context.Context, pidns *kernel.PIDNamespace, msrc *fs.MountSource) *fs.Inode { + s := &threadSelf{ + Symlink: *ramfs.NewSymlink(ctx, fs.RootOwner, ""), + pidns: pidns, } - u.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(u, msrc, fs.SpecialFile, nil) + return newProcInode(s, msrc, fs.Symlink, nil) } // Readlink implements fs.InodeOperations.Readlink. @@ -145,13 +123,13 @@ func (s *self) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { if t := kernel.TaskFromContext(ctx); t != nil { tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) if tgid == 0 { - return "", ramfs.ErrNotFound + return "", syserror.ENOENT } return strconv.FormatUint(uint64(tgid), 10), nil } // Who is reading this link? - return "", ramfs.ErrInvalidOp + return "", syserror.EINVAL } // threadSelf is more magical than "self" link. @@ -169,13 +147,13 @@ func (s *threadSelf) Readlink(ctx context.Context, inode *fs.Inode) (string, err tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) tid := s.pidns.IDOfTask(t) if tid == 0 || tgid == 0 { - return "", ramfs.ErrNotFound + return "", syserror.ENOENT } return fmt.Sprintf("%d/task/%d", tgid, tid), nil } // Who is reading this link? - return "", ramfs.ErrInvalidOp + return "", syserror.EINVAL } // Lookup loads an Inode at name into a Dirent. @@ -204,25 +182,44 @@ func (p *proc) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dire return fs.NewDirent(td, name), nil } -// Readdir synthesizes proc contents. -func (p *proc) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - // Serialize normal contents. - _, err := p.Dir.DeprecatedReaddir(ctx, dirCtx, offset) - if err != nil { - return offset, err +// GetFile implements fs.InodeOperations. +func (p *proc) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &rootProcFile{iops: p}), nil +} + +// rootProcFile implements fs.FileOperations for the proc directory. +// +// +stateify savable +type rootProcFile struct { + fsutil.DirFileOperations `state:"nosave"` + + iops *proc +} + +var _ fs.FileOperations = (*rootProcFile)(nil) + +// Readdir implements fs.FileOperations.Readdir. +func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + offset := file.Offset() + dirCtx := &fs.DirCtx{ + Serializer: ser, } - m := make(map[string]fs.DentAttr) - var names []string + // Get normal directory contents from ramfs dir. + names, m := rpf.iops.Dir.Children() - // Add special files. - m["sys"] = fs.GenericDentAttr(fs.SpecialFile, device.ProcDevice) - names = append(names, "sys") + // Add dot and dotdot. + root := fs.RootFromContext(ctx) + defer root.DecRef() + dot, dotdot := file.Dirent.GetDotAttrs(root) + names = append(names, ".", "..") + m["."] = dot + m[".."] = dotdot // Collect tasks. // Per linux we only include it in directory listings if it's the leader. // But for whatever crazy reason, you can still walk to the given node. - for _, tg := range p.pidns.ThreadGroups() { + for _, tg := range rpf.iops.pidns.ThreadGroups() { if leader := tg.Leader(); leader != nil { name := strconv.FormatUint(uint64(tg.ID()), 10) m[name] = fs.GenericDentAttr(fs.SpecialDirectory, device.ProcDevice) @@ -230,7 +227,7 @@ func (p *proc) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset } } - if offset >= len(m) { + if offset >= int64(len(m)) { return offset, nil } sort.Strings(names) @@ -241,12 +238,5 @@ func (p *proc) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset } offset++ } - return offset, err -} - -// newMountsSymlink returns a symlink to "self/mounts". -func newMountsSymlink(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - s := &ramfs.Symlink{} - s.InitSymlink(ctx, fs.RootOwner, "self/mounts") - return newFile(s, msrc, fs.Symlink, nil) + return offset, nil } diff --git a/pkg/sentry/fs/proc/rpcinet_proc.go b/pkg/sentry/fs/proc/rpcinet_proc.go index d025069df..65faa21f2 100644 --- a/pkg/sentry/fs/proc/rpcinet_proc.go +++ b/pkg/sentry/fs/proc/rpcinet_proc.go @@ -20,32 +20,72 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/rpcinet" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) -// rpcinetFile implments fs.InodeOperations as RPCs. -type rpcinetFile struct { - ramfs.Entry +// rpcInetInode implments fs.InodeOperations. +type rpcInetInode struct { + fsutil.SimpleFileInode - // filepath is the full path of this rpcinetFile. + // filepath is the full path of this rpcInetInode. filepath string k *kernel.Kernel } -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -// This method can panic if an rpcinetFile was created without an rpcinet +func newRPCInetInode(ctx context.Context, msrc *fs.MountSource, filepath string, mode linux.FileMode) *fs.Inode { + f := &rpcInetInode{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(mode), linux.PROC_SUPER_MAGIC), + filepath: filepath, + k: kernel.KernelFromContext(ctx), + } + return newProcInode(f, msrc, fs.SpecialFile, nil) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (i *rpcInetInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true + fops := &rpcInetFile{ + inode: i, + } + return fs.NewFile(ctx, dirent, flags, fops), nil +} + +// rpcInetFile implements fs.FileOperations as RPCs. +type rpcInetFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + inode *rpcInetInode +} + +// Read implements fs.FileOperations.Read. +// +// This method can panic if an rpcInetInode was created without an rpcinet // stack. -func (r rpcinetFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - s, ok := r.k.NetworkStack().(*rpcinet.Stack) +func (f *rpcInetFile) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) if !ok { panic("Network stack is not a rpcinet.") } - contents, se := s.RPCReadFile(r.filepath) + contents, se := s.RPCReadFile(f.inode.filepath) if se != nil || offset >= int64(len(contents)) { return 0, io.EOF } @@ -54,16 +94,12 @@ func (r rpcinetFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequenc return int64(n), err } -// Truncate implements fs.InodeOperations.Truncate. -func (r rpcinetFile) Truncate(context.Context, *fs.Inode, int64) error { - return nil -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -// This method can panic if an rpcinetFile was created without an rpcinet +// Write implements fs.FileOperations.Write. +// +// This method can panic if an rpcInetInode was created without an rpcInet // stack. -func (r rpcinetFile) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - s, ok := r.k.NetworkStack().(*rpcinet.Stack) +func (f *rpcInetFile) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { + s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) if !ok { panic("Network stack is not a rpcinet.") } @@ -78,116 +114,102 @@ func (r rpcinetFile) DeprecatedPwritev(ctx context.Context, src usermem.IOSequen return int64(n), err } - written, se := s.RPCWriteFile(r.filepath, b) + written, se := s.RPCWriteFile(f.inode.filepath, b) return int64(written), se.ToError() } -func newRPCProcFSFile(ctx context.Context, msrc *fs.MountSource, filepath string, mode linux.FileMode) *fs.Inode { - f := &rpcinetFile{ - filepath: filepath, - k: kernel.KernelFromContext(ctx), - } - f.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(mode)) - - fi := newFile(f, msrc, fs.SpecialFile, nil) - return fi -} - // newRPCInetProcNet will build an inode for /proc/net. func newRPCInetProcNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - - // Add all the files we want to forward for /proc/net. - d.AddChild(ctx, "arp", newRPCProcFSFile(ctx, msrc, "/proc/net/arp", 0444)) - d.AddChild(ctx, "dev", newRPCProcFSFile(ctx, msrc, "/proc/net/dev", 0444)) - d.AddChild(ctx, "if_inet6", newRPCProcFSFile(ctx, msrc, "/proc/net/if_inet6", 0444)) - d.AddChild(ctx, "ipv6_route", newRPCProcFSFile(ctx, msrc, "/proc/net/ipv6_route", 0444)) - d.AddChild(ctx, "netlink", newRPCProcFSFile(ctx, msrc, "/proc/net/netlink", 0444)) - d.AddChild(ctx, "netstat", newRPCProcFSFile(ctx, msrc, "/proc/net/netstat", 0444)) - d.AddChild(ctx, "packet", newRPCProcFSFile(ctx, msrc, "/proc/net/packet", 0444)) - d.AddChild(ctx, "protocols", newRPCProcFSFile(ctx, msrc, "/proc/net/protocols", 0444)) - d.AddChild(ctx, "psched", newRPCProcFSFile(ctx, msrc, "/proc/net/psched", 0444)) - d.AddChild(ctx, "ptype", newRPCProcFSFile(ctx, msrc, "/proc/net/ptype", 0444)) - d.AddChild(ctx, "route", newRPCProcFSFile(ctx, msrc, "/proc/net/route", 0444)) - d.AddChild(ctx, "tcp", newRPCProcFSFile(ctx, msrc, "/proc/net/tcp", 0444)) - d.AddChild(ctx, "tcp6", newRPCProcFSFile(ctx, msrc, "/proc/net/tcp6", 0444)) - d.AddChild(ctx, "udp", newRPCProcFSFile(ctx, msrc, "/proc/net/udp", 0444)) - d.AddChild(ctx, "udp6", newRPCProcFSFile(ctx, msrc, "/proc/net/udp6", 0444)) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "arp": newRPCInetInode(ctx, msrc, "/proc/net/arp", 0444), + "dev": newRPCInetInode(ctx, msrc, "/proc/net/dev", 0444), + "if_inet6": newRPCInetInode(ctx, msrc, "/proc/net/if_inet6", 0444), + "ipv6_route": newRPCInetInode(ctx, msrc, "/proc/net/ipv6_route", 0444), + "netlink": newRPCInetInode(ctx, msrc, "/proc/net/netlink", 0444), + "netstat": newRPCInetInode(ctx, msrc, "/proc/net/netstat", 0444), + "packet": newRPCInetInode(ctx, msrc, "/proc/net/packet", 0444), + "protocols": newRPCInetInode(ctx, msrc, "/proc/net/protocols", 0444), + "psched": newRPCInetInode(ctx, msrc, "/proc/net/psched", 0444), + "ptype": newRPCInetInode(ctx, msrc, "/proc/net/ptype", 0444), + "route": newRPCInetInode(ctx, msrc, "/proc/net/route", 0444), + "tcp": newRPCInetInode(ctx, msrc, "/proc/net/tcp", 0444), + "tcp6": newRPCInetInode(ctx, msrc, "/proc/net/tcp6", 0444), + "udp": newRPCInetInode(ctx, msrc, "/proc/net/udp", 0444), + "udp6": newRPCInetInode(ctx, msrc, "/proc/net/udp6", 0444), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // newRPCInetProcSysNet will build an inode for /proc/sys/net. func newRPCInetProcSysNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "ipv4", newRPCInetSysNetIPv4Dir(ctx, msrc)) - d.AddChild(ctx, "core", newRPCInetSysNetCore(ctx, msrc)) + contents := map[string]*fs.Inode{ + "ipv4": newRPCInetSysNetIPv4Dir(ctx, msrc), + "core": newRPCInetSysNetCore(ctx, msrc), + } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // newRPCInetSysNetCore builds the /proc/sys/net/core directory. func newRPCInetSysNetCore(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - - // Add all the files we want to forward over RPC for /proc/sys/net/core - d.AddChild(ctx, "default_qdisc", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/default_qdisc", 0444)) - d.AddChild(ctx, "message_burst", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/message_burst", 0444)) - d.AddChild(ctx, "message_cost", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/message_cost", 0444)) - d.AddChild(ctx, "optmem_max", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/optmem_max", 0444)) - d.AddChild(ctx, "rmem_default", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/rmem_default", 0444)) - d.AddChild(ctx, "rmem_max", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/rmem_max", 0444)) - d.AddChild(ctx, "somaxconn", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/somaxconn", 0444)) - d.AddChild(ctx, "wmem_default", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/wmem_default", 0444)) - d.AddChild(ctx, "wmem_max", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/wmem_max", 0444)) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "default_qdisc": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/default_qdisc", 0444), + "message_burst": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_burst", 0444), + "message_cost": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_cost", 0444), + "optmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/optmem_max", 0444), + "rmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_default", 0444), + "rmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_max", 0444), + "somaxconn": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/somaxconn", 0444), + "wmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_default", 0444), + "wmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_max", 0444), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // newRPCInetSysNetIPv4Dir builds the /proc/sys/net/ipv4 directory. func newRPCInetSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - - // Add all the files we want to forward over RPC for /proc/sys/net/ipv4. - d.AddChild(ctx, "ip_local_port_range", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_local_port_range", 0444)) - d.AddChild(ctx, "ip_local_reserved_ports", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_local_reserved_ports", 0444)) - d.AddChild(ctx, "ipfrag_time", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ipfrag_time", 0444)) - d.AddChild(ctx, "ip_nonlocal_bind", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_nonlocal_bind", 0444)) - d.AddChild(ctx, "ip_no_pmtu_disc", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_no_pmtu_disc", 0444)) - - d.AddChild(ctx, "tcp_allowed_congestion_control", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_allowed_congestion_control", 0444)) - d.AddChild(ctx, "tcp_available_congestion_control", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_available_congestion_control", 0444)) - d.AddChild(ctx, "tcp_base_mss", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_base_mss", 0444)) - d.AddChild(ctx, "tcp_congestion_control", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_congestion_control", 0644)) - d.AddChild(ctx, "tcp_dsack", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_dsack", 0644)) - d.AddChild(ctx, "tcp_early_retrans", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_early_retrans", 0644)) - d.AddChild(ctx, "tcp_fack", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fack", 0644)) - d.AddChild(ctx, "tcp_fastopen", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen", 0644)) - d.AddChild(ctx, "tcp_fastopen_key", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen_key", 0444)) - d.AddChild(ctx, "tcp_fin_timeout", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fin_timeout", 0644)) - d.AddChild(ctx, "tcp_invalid_ratelimit", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_invalid_ratelimit", 0444)) - d.AddChild(ctx, "tcp_keepalive_intvl", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_intvl", 0644)) - d.AddChild(ctx, "tcp_keepalive_probes", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_probes", 0644)) - d.AddChild(ctx, "tcp_keepalive_time", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_time", 0644)) - d.AddChild(ctx, "tcp_mem", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_mem", 0444)) - d.AddChild(ctx, "tcp_mtu_probing", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_mtu_probing", 0644)) - d.AddChild(ctx, "tcp_no_metrics_save", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_no_metrics_save", 0444)) - d.AddChild(ctx, "tcp_probe_interval", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_interval", 0444)) - d.AddChild(ctx, "tcp_probe_threshold", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_threshold", 0444)) - d.AddChild(ctx, "tcp_retries1", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries1", 0644)) - d.AddChild(ctx, "tcp_retries2", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries2", 0644)) - d.AddChild(ctx, "tcp_rfc1337", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_rfc1337", 0444)) - d.AddChild(ctx, "tcp_rmem", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_rmem", 0444)) - d.AddChild(ctx, "tcp_sack", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_sack", 0644)) - d.AddChild(ctx, "tcp_slow_start_after_idle", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_slow_start_after_idle", 0644)) - d.AddChild(ctx, "tcp_synack_retries", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_synack_retries", 0644)) - d.AddChild(ctx, "tcp_syn_retries", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_syn_retries", 0644)) - d.AddChild(ctx, "tcp_timestamps", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_timestamps", 0644)) - d.AddChild(ctx, "tcp_wmem", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_wmem", 0444)) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "ip_local_port_range": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_port_range", 0444), + "ip_local_reserved_ports": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_reserved_ports", 0444), + "ipfrag_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ipfrag_time", 0444), + "ip_nonlocal_bind": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_nonlocal_bind", 0444), + "ip_no_pmtu_disc": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_no_pmtu_disc", 0444), + "tcp_allowed_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_allowed_congestion_control", 0444), + "tcp_available_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_available_congestion_control", 0444), + "tcp_base_mss": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_base_mss", 0444), + "tcp_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_congestion_control", 0644), + "tcp_dsack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_dsack", 0644), + "tcp_early_retrans": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_early_retrans", 0644), + "tcp_fack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fack", 0644), + "tcp_fastopen": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen", 0644), + "tcp_fastopen_key": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen_key", 0444), + "tcp_fin_timeout": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fin_timeout", 0644), + "tcp_invalid_ratelimit": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_invalid_ratelimit", 0444), + "tcp_keepalive_intvl": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_intvl", 0644), + "tcp_keepalive_probes": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_probes", 0644), + "tcp_keepalive_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_time", 0644), + "tcp_mem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mem", 0444), + "tcp_mtu_probing": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mtu_probing", 0644), + "tcp_no_metrics_save": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_no_metrics_save", 0444), + "tcp_probe_interval": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_interval", 0444), + "tcp_probe_threshold": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_threshold", 0444), + "tcp_retries1": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries1", 0644), + "tcp_retries2": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries2", 0644), + "tcp_rfc1337": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rfc1337", 0444), + "tcp_rmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rmem", 0444), + "tcp_sack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_sack", 0644), + "tcp_slow_start_after_idle": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_slow_start_after_idle", 0644), + "tcp_synack_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_synack_retries", 0644), + "tcp_syn_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_syn_retries", 0644), + "tcp_timestamps": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_timestamps", 0644), + "tcp_wmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_wmem", 0444), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD index 53c475652..b4ba64e10 100644 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ b/pkg/sentry/fs/proc/seqfile/BUILD @@ -8,12 +8,15 @@ go_library( importpath = "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile", visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/sentry/context", "//pkg/sentry/fs", + "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/proc/device", - "//pkg/sentry/fs/ramfs", "//pkg/sentry/kernel/time", "//pkg/sentry/usermem", + "//pkg/syserror", + "//pkg/waiter", ], ) @@ -26,7 +29,7 @@ go_test( "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/fs", - "//pkg/sentry/fs/ramfs/test", + "//pkg/sentry/fs/ramfs", "//pkg/sentry/usermem", ], ) diff --git a/pkg/sentry/fs/proc/seqfile/seqfile.go b/pkg/sentry/fs/proc/seqfile/seqfile.go index 0499ba65b..16fc6789e 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile.go @@ -18,12 +18,15 @@ import ( "io" "sync" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // SeqHandle is a helper handle to seek in the file. @@ -87,7 +90,18 @@ func (s *SeqGenerationCounter) IsCurrent(generation int64) bool { // // +stateify savable type SeqFile struct { - ramfs.Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleExtendedAttributes + fsutil.InodeSimpleAttributes // mu protects the fields below. mu sync.Mutex `state:"nosave"` @@ -99,11 +113,14 @@ type SeqFile struct { lastRead int64 } +var _ fs.InodeOperations = (*SeqFile)(nil) + // NewSeqFile returns a seqfile suitable for use by external consumers. func NewSeqFile(ctx context.Context, source SeqSource) *SeqFile { - s := &SeqFile{SeqSource: source} - s.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return s + return &SeqFile{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + SeqSource: source, + } } // NewSeqFileInode returns an Inode with SeqFile InodeOperations. @@ -120,11 +137,19 @@ func NewSeqFileInode(ctx context.Context, source SeqSource, msrc *fs.MountSource // UnstableAttr returns unstable attributes of the SeqFile. func (s *SeqFile) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - uattr, _ := s.Entry.UnstableAttr(ctx, inode) + uattr, err := s.InodeSimpleAttributes.UnstableAttr(ctx, inode) + if err != nil { + return fs.UnstableAttr{}, err + } uattr.ModificationTime = ktime.NowFromContext(ctx) return uattr, nil } +// GetFile implements fs.InodeOperations.GetFile. +func (s *SeqFile) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &seqFileOperations{seqFile: s}), nil +} + // findIndexAndOffset finds the unit that corresponds to a certain offset. // Returns the unit and the offset within the unit. If there are not enough // units len(data) and leftover offset is returned. @@ -139,36 +164,74 @@ func findIndexAndOffset(data []SeqData, offset int64) (int, int64) { return len(data), offset } -// DeprecatedPreadv reads from the file at the given offset. -func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - s.mu.Lock() - defer s.mu.Unlock() +// updateSourceLocked requires that s.mu is held. +func (s *SeqFile) updateSourceLocked(ctx context.Context, record int) { + var h SeqHandle + if record == 0 { + h = nil + } else { + h = s.source[record-1].Handle + } + // Save what we have previously read. + s.source = s.source[:record] + var newSource []SeqData + newSource, s.generation = s.SeqSource.ReadSeqFileData(ctx, h) + s.source = append(s.source, newSource...) +} + +// seqFileOperations implements fs.FileOperations. +// +// +stateify savable +type seqFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + seqFile *SeqFile +} + +var _ fs.FileOperations = (*seqFileOperations)(nil) + +// Write implements fs.FileOperations.Write. +func (*seqFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.EACCES +} + +// Read implements fs.FileOperations.Read. +func (sfo *seqFileOperations) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + sfo.seqFile.mu.Lock() + defer sfo.seqFile.mu.Unlock() - s.Entry.NotifyAccess(ctx) - defer func() { s.lastRead = offset }() + sfo.seqFile.NotifyAccess(ctx) + defer func() { sfo.seqFile.lastRead = offset }() updated := false // Try to find where we should start reading this file. - i, recordOffset := findIndexAndOffset(s.source, offset) - if i == len(s.source) { + i, recordOffset := findIndexAndOffset(sfo.seqFile.source, offset) + if i == len(sfo.seqFile.source) { // Ok, we're at EOF. Let's first check to see if there might be // more data available to us. If there is more data, add it to // the end and try reading again. - if !s.SeqSource.NeedsUpdate(s.generation) { + if !sfo.seqFile.SeqSource.NeedsUpdate(sfo.seqFile.generation) { return 0, io.EOF } - oldLen := len(s.source) - s.updateSourceLocked(ctx, len(s.source)) + oldLen := len(sfo.seqFile.source) + sfo.seqFile.updateSourceLocked(ctx, len(sfo.seqFile.source)) updated = true // We know that we had consumed everything up until this point // so we search in the new slice instead of starting over. - i, recordOffset = findIndexAndOffset(s.source[oldLen:], recordOffset) + i, recordOffset = findIndexAndOffset(sfo.seqFile.source[oldLen:], recordOffset) i += oldLen // i is at most the length of the slice which is - // len(s.source) - oldLen. So at most i will be equal to - // len(s.source). - if i == len(s.source) { + // len(sfo.seqFile.source) - oldLen. So at most i will be equal to + // len(sfo.seqFile.source). + if i == len(sfo.seqFile.source) { return 0, io.EOF } } @@ -178,7 +241,7 @@ func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, // before continuing on to the next. We don't refresh our data source // before this record is completed. if recordOffset != 0 { - n, err := dst.CopyOut(ctx, s.source[i].Buf[recordOffset:]) + n, err := dst.CopyOut(ctx, sfo.seqFile.source[i].Buf[recordOffset:]) done += int64(n) dst = dst.DropFirst(n) if dst.NumBytes() == 0 || err != nil { @@ -190,15 +253,15 @@ func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, // Next/New unit, update the source file if necessary. Make an extra // check to see if we've seeked backwards and if so always update our // data source. - if !updated && (s.SeqSource.NeedsUpdate(s.generation) || s.lastRead > offset) { - s.updateSourceLocked(ctx, i) + if !updated && (sfo.seqFile.SeqSource.NeedsUpdate(sfo.seqFile.generation) || sfo.seqFile.lastRead > offset) { + sfo.seqFile.updateSourceLocked(ctx, i) // recordOffset is 0 here and we won't update records behind the // current one so recordOffset is still 0 even though source // just got updated. Just read the next record. } // Finish by reading all the available data. - for _, buf := range s.source[i:] { + for _, buf := range sfo.seqFile.source[i:] { n, err := dst.CopyOut(ctx, buf.Buf) done += int64(n) dst = dst.DropFirst(n) @@ -214,23 +277,3 @@ func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, } return done, nil } - -// updateSourceLocked requires that s.mu is held. -func (s *SeqFile) updateSourceLocked(ctx context.Context, record int) { - var h SeqHandle - if record == 0 { - h = nil - } else { - h = s.source[record-1].Handle - } - // Save what we have previously read. - s.source = s.source[:record] - var newSource []SeqData - newSource, s.generation = s.SeqSource.ReadSeqFileData(ctx, h) - s.source = append(s.source, newSource...) -} - -// DeprecatedPwritev is always denied. -func (*SeqFile) DeprecatedPwritev(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, ramfs.ErrDenied -} diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_test.go b/pkg/sentry/fs/proc/seqfile/seqfile_test.go index f9a2ca38e..35403ab7f 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile_test.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile_test.go @@ -23,7 +23,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/context/contexttest" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - ramfstest "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" ) @@ -91,10 +91,15 @@ type testTable struct { expectedError error } -func runTableTests(ctx context.Context, table []testTable, n fs.InodeOperations) error { +func runTableTests(ctx context.Context, table []testTable, dirent *fs.Dirent) error { for _, tt := range table { + file, err := dirent.Inode.InodeOperations.GetFile(ctx, dirent, fs.FileFlags{Read: true}) + if err != nil { + return fmt.Errorf("GetFile returned error: %v", err) + } + data := make([]byte, tt.readBufferSize) - resultLen, err := n.DeprecatedPreadv(ctx, usermem.BytesIOSequence(data), tt.offset) + resultLen, err := file.Preadv(ctx, usermem.BytesIOSequence(data), tt.offset) if err != tt.expectedError { return fmt.Errorf("t.Preadv(len: %v, offset: %v) (error) => %v expected %v", tt.readBufferSize, tt.offset, err, tt.expectedError) } @@ -115,12 +120,12 @@ func TestSeqFile(t *testing.T) { testSource.Init() // Create a file that can be R/W. - m := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) + m := fs.NewPseudoMountSource() ctx := contexttest.Context(t) contents := map[string]*fs.Inode{ "foo": NewSeqFileInode(ctx, testSource, m), } - root := ramfstest.NewDir(ctx, contents, fs.FilePermsFromMode(0777)) + root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) // How about opening it? inode := fs.NewInode(root, m, fs.StableAttr{Type: fs.Directory}) @@ -129,9 +134,13 @@ func TestSeqFile(t *testing.T) { t.Fatalf("failed to walk to foo for n2: %v", err) } n2 := dirent2.Inode.InodeOperations + file2, err := n2.GetFile(ctx, dirent2, fs.FileFlags{Read: true, Write: true}) + if err != nil { + t.Fatalf("GetFile returned error: %v", err) + } // Writing? - if _, err := n2.DeprecatedPwritev(nil, usermem.BytesIOSequence([]byte("test")), 0); err == nil { + if _, err := file2.Writev(ctx, usermem.BytesIOSequence([]byte("test"))); err == nil { t.Fatalf("managed to write to n2: %v", err) } @@ -141,7 +150,6 @@ func TestSeqFile(t *testing.T) { t.Fatalf("failed to walk to foo: %v", err) } n3 := dirent3.Inode.InodeOperations - if n2 != n3 { t.Error("got n2 != n3, want same") } @@ -170,13 +178,13 @@ func TestSeqFile(t *testing.T) { // Read the last 3 bytes. {97, 10, testSource.actual[9].Buf[7:], nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed with testSource.update = %v : %v", testSource.update, err) } // Disable updates and do it again. testSource.update = false - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed with testSource.update = %v: %v", testSource.update, err) } } @@ -188,25 +196,24 @@ func TestSeqFileFileUpdated(t *testing.T) { testSource.update = true // Create a file that can be R/W. - m := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) + m := fs.NewPseudoMountSource() ctx := contexttest.Context(t) contents := map[string]*fs.Inode{ "foo": NewSeqFileInode(ctx, testSource, m), } - root := ramfstest.NewDir(ctx, contents, fs.FilePermsFromMode(0777)) + root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) // How about opening it? inode := fs.NewInode(root, m, fs.StableAttr{Type: fs.Directory}) dirent2, err := root.Lookup(ctx, inode, "foo") if err != nil { - t.Fatalf("failed to walk to foo for n2: %v", err) + t.Fatalf("failed to walk to foo for dirent2: %v", err) } - n2 := dirent2.Inode.InodeOperations table := []testTable{ {0, 16, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf[:6]), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed: %v", err) } // Delete the first entry. @@ -224,7 +231,7 @@ func TestSeqFileFileUpdated(t *testing.T) { // Read the following two lines. {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after removing first entry: %v", err) } @@ -238,7 +245,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {50, 20, flatten(testSource.actual[4].Buf, testSource.actual[5].Buf), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after adding middle entry: %v", err) } // This will be used in a later test. @@ -249,7 +256,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {20, 20, []byte{}, io.EOF}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after removing all entries: %v", err) } // Restore some of the data. @@ -257,7 +264,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {6, 20, testSource.actual[0].Buf[6:], nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after adding first entry back: %v", err) } @@ -266,7 +273,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after extending testSource: %v", err) } } diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index 54562508d..ee6b9f262 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -22,39 +22,15 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/rpcinet" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) -// hostname is a file containing the system hostname. -// -// +stateify savable -type hostname struct { - ramfs.Entry -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (hostname) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - utsns := kernel.UTSNamespaceFromContext(ctx) - contents := []byte(utsns.HostName() + "\n") - - if offset >= int64(len(contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, contents[offset:]) - return int64(n), err -} - -func (p *proc) newHostname(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - h := &hostname{} - h.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(h, msrc, fs.SpecialFile, nil) -} - // mmapMinAddrData backs /proc/sys/vm/mmap_min_addr. // // +stateify savable @@ -101,36 +77,84 @@ func (*overcommitMemory) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandl } func (p *proc) newKernelDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "hostname", p.newHostname(ctx, msrc)) - - d.AddChild(ctx, "shmmax", p.newStubProcFSFile(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMAX, 10)))) - d.AddChild(ctx, "shmall", p.newStubProcFSFile(ctx, msrc, []byte(strconv.FormatUint(linux.SHMALL, 10)))) - d.AddChild(ctx, "shmmni", p.newStubProcFSFile(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMNI, 10)))) - return newFile(d, msrc, fs.SpecialDirectory, nil) + h := hostname{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + } + + children := map[string]*fs.Inode{ + "hostname": newProcInode(&h, msrc, fs.SpecialFile, nil), + "shmall": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMALL, 10))), + "shmmax": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMAX, 10))), + "shmmni": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMNI, 10))), + } + + d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newVMDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "mmap_min_addr", seqfile.NewSeqFileInode(ctx, &mmapMinAddrData{p.k}, msrc)) - d.AddChild(ctx, "overcommit_memory", seqfile.NewSeqFileInode(ctx, &overcommitMemory{}, msrc)) - return newFile(d, msrc, fs.SpecialDirectory, nil) + children := map[string]*fs.Inode{ + "mmap_min_addr": seqfile.NewSeqFileInode(ctx, &mmapMinAddrData{p.k}, msrc), + "overcommit_memory": seqfile.NewSeqFileInode(ctx, &overcommitMemory{}, msrc), + } + d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newSysDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "kernel", p.newKernelDir(ctx, msrc)) - d.AddChild(ctx, "vm", p.newVMDir(ctx, msrc)) + children := map[string]*fs.Inode{ + "kernel": p.newKernelDir(ctx, msrc), + "vm": p.newVMDir(ctx, msrc), + } // If we're using rpcinet we will let it manage /proc/sys/net. if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - d.AddChild(ctx, "net", newRPCInetProcSysNet(ctx, msrc)) + children["net"] = newRPCInetProcSysNet(ctx, msrc) } else { - d.AddChild(ctx, "net", p.newSysNetDir(ctx, msrc)) + children["net"] = p.newSysNetDir(ctx, msrc) } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) +} + +// hostname is the inode for a file containing the system hostname. +// +// +stateify savable +type hostname struct { + fsutil.SimpleFileInode +} + +// GetFile implements fs.InodeOperations.GetFile. +func (h *hostname) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, d, flags, &hostnameFile{}), nil +} + +var _ fs.InodeOperations = (*hostname)(nil) + +// +stateify savable +type hostnameFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSeek `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` } + +// Read implements fs.FileOperations.Read. +func (hf *hostnameFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + utsns := kernel.UTSNamespaceFromContext(ctx) + contents := []byte(utsns.HostName() + "\n") + if offset >= int64(len(contents)) { + return 0, io.EOF + } + n, err := dst.CopyOut(ctx, contents[offset:]) + return int64(n), err + +} + +var _ fs.FileOperations = (*hostnameFile)(nil) diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index b50d43d70..42e9bc47f 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -17,13 +17,17 @@ package proc import ( "fmt" "io" + "sync" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/inet" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) type tcpMemDir int @@ -33,21 +37,37 @@ const ( tcpWMem ) +// tcpMemInode is used to read/write the size of netstack tcp buffers. +// +// TODO: If we have multiple proc mounts, concurrent writes can +// leave netstack and the proc files in an inconsistent state. Since we set the +// buffer size from these proc files on restore, we may also race and end up in +// an inconsistent state on restore. +// // +stateify savable -type tcpMem struct { - ramfs.Entry - s inet.Stack `state:"wait"` +type tcpMemInode struct { + fsutil.SimpleFileInode + dir tcpMemDir + s inet.Stack `state:"wait"` + + // size stores the tcp buffer size during save, and sets the buffer + // size in netstack in restore. We must save/restore this here, since + // netstack itself is stateless. size inet.TCPBufferSize - dir tcpMemDir -} -func newTCPMem(s inet.Stack, size inet.TCPBufferSize, dir tcpMemDir) *tcpMem { - return &tcpMem{s: s, size: size, dir: dir} + // mu protects against concurrent reads/writes to files based on this + // inode. + mu sync.Mutex `state:"nosave"` } -func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, size inet.TCPBufferSize, dir tcpMemDir) *fs.Inode { - tm := newTCPMem(s, size, dir) - tm.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0644)) +var _ fs.InodeOperations = (*tcpMemInode)(nil) + +func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, dir tcpMemDir) *fs.Inode { + tm := &tcpMemInode{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + s: s, + dir: dir, + } sattr := fs.StableAttr{ DeviceID: device.ProcDevice.DeviceID(), InodeID: device.ProcDevice.NextIno(), @@ -57,62 +77,105 @@ func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, siz return fs.NewInode(tm, msrc, sattr) } -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (m *tcpMem) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// GetFile implements fs.InodeOperations.GetFile. +func (m *tcpMemInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + return fs.NewFile(ctx, dirent, flags, &tcpMemFile{tcpMemInode: m}), nil +} + +// +stateify savable +type tcpMemFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + tcpMemInode *tcpMemInode +} + +var _ fs.FileOperations = (*tcpMemFile)(nil) + +// Read implements fs.FileOperations.Read. +func (f *tcpMemFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset != 0 { return 0, io.EOF } - s := fmt.Sprintf("%d\t%d\t%d\n", m.size.Min, m.size.Default, m.size.Max) + f.tcpMemInode.mu.Lock() + defer f.tcpMemInode.mu.Unlock() + + size, err := readSize(f.tcpMemInode.dir, f.tcpMemInode.s) + if err != nil { + return 0, err + } + s := fmt.Sprintf("%d\t%d\t%d\n", size.Min, size.Default, size.Max) n, err := dst.CopyOut(ctx, []byte(s)) return int64(n), err } -// Truncate implements fs.InodeOperations.Truncate. -func (*tcpMem) Truncate(context.Context, *fs.Inode, int64) error { - return nil -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (m *tcpMem) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { +// Write implements fs.FileOperations.Write. +func (f *tcpMemFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { if src.NumBytes() == 0 { return 0, nil } - src = src.TakeFirst(usermem.PageSize - 1) + f.tcpMemInode.mu.Lock() + defer f.tcpMemInode.mu.Unlock() - buf := []int32{int32(m.size.Min), int32(m.size.Default), int32(m.size.Max)} + src = src.TakeFirst(usermem.PageSize - 1) + size, err := readSize(f.tcpMemInode.dir, f.tcpMemInode.s) + if err != nil { + return 0, err + } + buf := []int32{int32(size.Min), int32(size.Default), int32(size.Max)} n, cperr := usermem.CopyInt32StringsInVec(ctx, src.IO, src.Addrs, buf, src.Opts) - m.size = inet.TCPBufferSize{ + newSize := inet.TCPBufferSize{ Min: int(buf[0]), Default: int(buf[1]), Max: int(buf[2]), } - if err := m.writeSize(); err != nil { + if err := writeSize(f.tcpMemInode.dir, f.tcpMemInode.s, newSize); err != nil { return n, err } return n, cperr } -func (m *tcpMem) writeSize() error { - switch m.dir { +func readSize(dirType tcpMemDir, s inet.Stack) (inet.TCPBufferSize, error) { + switch dirType { + case tcpRMem: + return s.TCPReceiveBufferSize() + case tcpWMem: + return s.TCPSendBufferSize() + default: + panic(fmt.Sprintf("unknown tcpMemFile type: %v", dirType)) + } +} + +func writeSize(dirType tcpMemDir, s inet.Stack, size inet.TCPBufferSize) error { + switch dirType { case tcpRMem: - return m.s.SetTCPReceiveBufferSize(m.size) + return s.SetTCPReceiveBufferSize(size) case tcpWMem: - return m.s.SetTCPSendBufferSize(m.size) + return s.SetTCPSendBufferSize(size) default: - panic(fmt.Sprintf("unknown tcpMem.dir: %v", m.dir)) + panic(fmt.Sprintf("unknown tcpMemFile type: %v", dirType)) } } // +stateify savable type tcpSack struct { - ramfs.Entry - s inet.Stack `state:"wait"` + stack inet.Stack `state:"wait"` enabled *bool + fsutil.SimpleFileInode } func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { - ts := &tcpSack{s: s} - ts.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0644)) + ts := &tcpSack{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + stack: s, + } sattr := fs.StableAttr{ DeviceID: device.ProcDevice.DeviceID(), InodeID: device.ProcDevice.NextIno(), @@ -122,21 +185,48 @@ func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *f return fs.NewInode(ts, msrc, sattr) } -func (s *tcpSack) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// GetFile implements fs.InodeOperations.GetFile. +func (s *tcpSack) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true + return fs.NewFile(ctx, dirent, flags, &tcpSackFile{ + tcpSack: s, + stack: s.stack, + }), nil +} + +// +stateify savable +type tcpSackFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + tcpSack *tcpSack + + stack inet.Stack `state:"wait"` +} + +// Read implements fs.FileOperations.Read. +func (f *tcpSackFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset != 0 { return 0, io.EOF } - if s.enabled == nil { - sack, err := s.s.TCPSACKEnabled() + if f.tcpSack.enabled == nil { + sack, err := f.stack.TCPSACKEnabled() if err != nil { return 0, err } - s.enabled = &sack + f.tcpSack.enabled = &sack } val := "0\n" - if *s.enabled { + if *f.tcpSack.enabled { // Technically, this is not quite compatible with Linux. Linux // stores these as an integer, so if you write "2" into // tcp_sack, you should get 2 back. Tough luck. @@ -146,13 +236,8 @@ func (s *tcpSack) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, return int64(n), err } -// Truncate implements fs.InodeOperations.Truncate. -func (*tcpSack) Truncate(context.Context, *fs.Inode, int64) error { - return nil -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (s *tcpSack) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { +// Write implements fs.FileOperations.Write. +func (f *tcpSackFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { if src.NumBytes() == 0 { return 0, nil } @@ -163,100 +248,104 @@ func (s *tcpSack) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, if err != nil { return n, err } - if s.enabled == nil { - s.enabled = new(bool) + if f.tcpSack.enabled == nil { + f.tcpSack.enabled = new(bool) } - *s.enabled = v != 0 - return n, s.s.SetTCPSACKEnabled(*s.enabled) + *f.tcpSack.enabled = v != 0 + return n, f.tcpSack.stack.SetTCPSACKEnabled(*f.tcpSack.enabled) } func (p *proc) newSysNetCore(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - // 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). - d.AddChild(ctx, "default_qdisc", p.newStubProcFSFile(ctx, msrc, []byte("pfifo_fast"))) - d.AddChild(ctx, "message_burst", p.newStubProcFSFile(ctx, msrc, []byte("10"))) - d.AddChild(ctx, "message_cost", p.newStubProcFSFile(ctx, msrc, []byte("5"))) - d.AddChild(ctx, "optmem_max", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "rmem_default", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - d.AddChild(ctx, "rmem_max", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - d.AddChild(ctx, "somaxconn", p.newStubProcFSFile(ctx, msrc, []byte("128"))) - d.AddChild(ctx, "wmem_default", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - d.AddChild(ctx, "wmem_max", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "default_qdisc": newStaticProcInode(ctx, msrc, []byte("pfifo_fast")), + "message_burst": newStaticProcInode(ctx, msrc, []byte("10")), + "message_cost": newStaticProcInode(ctx, msrc, []byte("5")), + "optmem_max": newStaticProcInode(ctx, msrc, []byte("0")), + "rmem_default": newStaticProcInode(ctx, msrc, []byte("212992")), + "rmem_max": newStaticProcInode(ctx, msrc, []byte("212992")), + "somaxconn": newStaticProcInode(ctx, msrc, []byte("128")), + "wmem_default": newStaticProcInode(ctx, msrc, []byte("212992")), + "wmem_max": newStaticProcInode(ctx, msrc, []byte("212992")), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) + contents := map[string]*fs.Inode{ + // Add tcp_sack. + "tcp_sack": newTCPSackInode(ctx, msrc, s), + + // 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": newStaticProcInode(ctx, msrc, []byte("16000 65535")), + "ip_local_reserved_ports": newStaticProcInode(ctx, msrc, []byte("")), + "ipfrag_time": newStaticProcInode(ctx, msrc, []byte("30")), + "ip_nonlocal_bind": newStaticProcInode(ctx, msrc, []byte("0")), + "ip_no_pmtu_disc": newStaticProcInode(ctx, msrc, []byte("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": newStaticProcInode(ctx, msrc, []byte("")), + "tcp_available_congestion_control": newStaticProcInode(ctx, msrc, []byte("reno")), + "tcp_congestion_control": newStaticProcInode(ctx, msrc, []byte("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": newStaticProcInode(ctx, msrc, []byte("1280")), + "tcp_dsack": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_early_retrans": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_fack": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_fastopen": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_fastopen_key": newStaticProcInode(ctx, msrc, []byte("")), + "tcp_invalid_ratelimit": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_keepalive_intvl": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_keepalive_probes": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_keepalive_time": newStaticProcInode(ctx, msrc, []byte("7200")), + "tcp_mtu_probing": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_no_metrics_save": newStaticProcInode(ctx, msrc, []byte("1")), + "tcp_probe_interval": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_probe_threshold": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_retries1": newStaticProcInode(ctx, msrc, []byte("3")), + "tcp_retries2": newStaticProcInode(ctx, msrc, []byte("15")), + "tcp_rfc1337": newStaticProcInode(ctx, msrc, []byte("1")), + "tcp_slow_start_after_idle": newStaticProcInode(ctx, msrc, []byte("1")), + "tcp_synack_retries": newStaticProcInode(ctx, msrc, []byte("5")), + "tcp_syn_retries": newStaticProcInode(ctx, msrc, []byte("3")), + "tcp_timestamps": newStaticProcInode(ctx, msrc, []byte("1")), + } // Add tcp_rmem. - if rs, err := s.TCPReceiveBufferSize(); err == nil { - d.AddChild(ctx, "tcp_rmem", newTCPMemInode(ctx, msrc, s, rs, tcpRMem)) + if _, err := s.TCPReceiveBufferSize(); err == nil { + contents["tcp_rmem"] = newTCPMemInode(ctx, msrc, s, tcpRMem) } // Add tcp_wmem. - if ss, err := s.TCPSendBufferSize(); err == nil { - d.AddChild(ctx, "tcp_wmem", newTCPMemInode(ctx, msrc, s, ss, tcpWMem)) + if _, err := s.TCPSendBufferSize(); err == nil { + contents["tcp_wmem"] = newTCPMemInode(ctx, msrc, s, tcpWMem) } - // Add tcp_sack. - d.AddChild(ctx, "tcp_sack", newTCPSackInode(ctx, msrc, s)) - - // 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). - d.AddChild(ctx, "ip_local_port_range", p.newStubProcFSFile(ctx, msrc, []byte("16000 65535"))) - d.AddChild(ctx, "ip_local_reserved_ports", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "ipfrag_time", p.newStubProcFSFile(ctx, msrc, []byte("30"))) - d.AddChild(ctx, "ip_nonlocal_bind", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "ip_no_pmtu_disc", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - - // tcp_allowed_congestion_control tell the user what they are able to do as an - // unprivledged process so we leave it empty. - d.AddChild(ctx, "tcp_allowed_congestion_control", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "tcp_available_congestion_control", p.newStubProcFSFile(ctx, msrc, []byte("reno"))) - d.AddChild(ctx, "tcp_congestion_control", p.newStubProcFSFile(ctx, msrc, []byte("reno"))) - - // Many of the following stub files are features netstack doesn't support - // and are therefore "0" for disabled. - d.AddChild(ctx, "tcp_base_mss", p.newStubProcFSFile(ctx, msrc, []byte("1280"))) - d.AddChild(ctx, "tcp_dsack", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_early_retrans", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_fack", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_fastopen", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_fastopen_key", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "tcp_invalid_ratelimit", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_keepalive_intvl", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_keepalive_probes", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_keepalive_time", p.newStubProcFSFile(ctx, msrc, []byte("7200"))) - d.AddChild(ctx, "tcp_mtu_probing", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_no_metrics_save", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - d.AddChild(ctx, "tcp_probe_interval", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_probe_threshold", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_retries1", p.newStubProcFSFile(ctx, msrc, []byte("3"))) - d.AddChild(ctx, "tcp_retries2", p.newStubProcFSFile(ctx, msrc, []byte("15"))) - d.AddChild(ctx, "tcp_rfc1337", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - d.AddChild(ctx, "tcp_slow_start_after_idle", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - d.AddChild(ctx, "tcp_synack_retries", p.newStubProcFSFile(ctx, msrc, []byte("5"))) - d.AddChild(ctx, "tcp_syn_retries", p.newStubProcFSFile(ctx, msrc, []byte("3"))) - d.AddChild(ctx, "tcp_timestamps", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newSysNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) + var contents map[string]*fs.Inode if s := p.k.NetworkStack(); s != nil { - d.AddChild(ctx, "ipv4", p.newSysNetIPv4Dir(ctx, msrc, s)) - d.AddChild(ctx, "core", p.newSysNetCore(ctx, msrc, s)) + contents = map[string]*fs.Inode{ + "ipv4": p.newSysNetIPv4Dir(ctx, msrc, s), + "core": p.newSysNetCore(ctx, msrc, s), + } } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } diff --git a/pkg/sentry/fs/proc/sys_net_state.go b/pkg/sentry/fs/proc/sys_net_state.go index 7f46776c0..5f481a1cf 100644 --- a/pkg/sentry/fs/proc/sys_net_state.go +++ b/pkg/sentry/fs/proc/sys_net_state.go @@ -16,17 +16,26 @@ package proc import "fmt" +// beforeSave is invoked by stateify. +func (t *tcpMemInode) beforeSave() { + size, err := readSize(t.dir, t.s) + if err != nil { + panic(fmt.Sprintf("failed to read TCP send / receive buffer sizes: %v", err)) + } + t.size = size +} + // afterLoad is invoked by stateify. -func (m *tcpMem) afterLoad() { - if err := m.writeSize(); err != nil { - panic(fmt.Sprintf("failed to write previous TCP send / receive buffer sizes [%v]: %v", m.size, err)) +func (t *tcpMemInode) afterLoad() { + if err := writeSize(t.dir, t.s, t.size); err != nil { + panic(fmt.Sprintf("failed to write previous TCP send / receive buffer sizes [%v]: %v", t.size, err)) } } // afterLoad is invoked by stateify. func (s *tcpSack) afterLoad() { if s.enabled != nil { - if err := s.s.SetTCPSACKEnabled(*s.enabled); err != nil { + if err := s.stack.SetTCPSACKEnabled(*s.enabled); err != nil { panic(fmt.Sprintf("failed to set previous TCP sack configuration [%v]: %v", *s.enabled, err)) } } diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go index 0ce9d30f1..ea0d94fce 100644 --- a/pkg/sentry/fs/proc/sys_net_test.go +++ b/pkg/sentry/fs/proc/sys_net_test.go @@ -26,13 +26,14 @@ func TestQuerySendBufferSize(t *testing.T) { ctx := context.Background() s := inet.NewTestStack() s.TCPSendBufSize = inet.TCPBufferSize{100, 200, 300} - tm := newTCPMem(s, s.TCPSendBufSize, tcpWMem) + tmi := &tcpMemInode{s: s, dir: tcpWMem} + tmf := &tcpMemFile{tcpMemInode: tmi} buf := make([]byte, 100) dst := usermem.BytesIOSequence(buf) - n, err := tm.DeprecatedPreadv(ctx, dst, 0) + n, err := tmf.Read(ctx, nil, dst, 0) if err != nil { - t.Fatalf("DeprecatedPreadv failed: %v", err) + t.Fatalf("Read failed: %v", err) } if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { @@ -44,13 +45,14 @@ func TestQueryRecvBufferSize(t *testing.T) { ctx := context.Background() s := inet.NewTestStack() s.TCPRecvBufSize = inet.TCPBufferSize{100, 200, 300} - tm := newTCPMem(s, s.TCPRecvBufSize, tcpRMem) + tmi := &tcpMemInode{s: s, dir: tcpRMem} + tmf := &tcpMemFile{tcpMemInode: tmi} buf := make([]byte, 100) dst := usermem.BytesIOSequence(buf) - n, err := tm.DeprecatedPreadv(ctx, dst, 0) + n, err := tmf.Read(ctx, nil, dst, 0) if err != nil { - t.Fatalf("DeprecatedPreadv failed: %v", err) + t.Fatalf("Read failed: %v", err) } if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { @@ -85,12 +87,13 @@ func TestConfigureSendBufferSize(t *testing.T) { s := inet.NewTestStack() for _, c := range cases { s.TCPSendBufSize = c.initial - tm := newTCPMem(s, c.initial, tcpWMem) + tmi := &tcpMemInode{s: s, dir: tcpWMem} + tmf := &tcpMemFile{tcpMemInode: tmi} // Write the values. src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tm.DeprecatedPwritev(ctx, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("DeprecatedPwritev, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) + if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { + t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) } // Read the values from the stack and check them. @@ -105,12 +108,13 @@ func TestConfigureRecvBufferSize(t *testing.T) { s := inet.NewTestStack() for _, c := range cases { s.TCPRecvBufSize = c.initial - tm := newTCPMem(s, c.initial, tcpRMem) + tmi := &tcpMemInode{s: s, dir: tcpRMem} + tmf := &tcpMemFile{tcpMemInode: tmi} // Write the values. src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tm.DeprecatedPwritev(ctx, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("DeprecatedPwritev, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) + if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { + t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) } // Read the values from the stack and check them. diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 91bda8a95..41981a973 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -24,6 +24,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" @@ -32,6 +33,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/usage" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // getTaskMM returns t's MemoryManager. If getTaskMM succeeds, the MemoryManager's @@ -57,19 +59,19 @@ func getTaskMM(t *kernel.Task) (*mm.MemoryManager, error) { type taskDir struct { ramfs.Dir - // t is the associated kernel task that owns this file. - t *kernel.Task + t *kernel.Task + pidns *kernel.PIDNamespace } +var _ fs.InodeOperations = (*taskDir)(nil) + // newTaskDir creates a new proc task entry. func newTaskDir(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace, showSubtasks bool) *fs.Inode { - d := &taskDir{t: t} - // TODO: Set EUID/EGID based on dumpability. - d.InitDir(t, map[string]*fs.Inode{ + contents := map[string]*fs.Inode{ "auxv": newAuxvec(t, msrc), - "cmdline": newExecArgFile(t, msrc, cmdlineExecArg), + "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), "comm": newComm(t, msrc), - "environ": newExecArgFile(t, msrc, environExecArg), + "environ": newExecArgInode(t, msrc, environExecArg), "exe": newExe(t, msrc), "fd": newFdDir(t, msrc), "fdinfo": newFdInfoDir(t, msrc), @@ -87,11 +89,18 @@ func newTaskDir(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace "statm": newStatm(t, msrc), "status": newStatus(t, msrc, pidns), "uid_map": newUIDMap(t, msrc), - }, fs.RootOwner, fs.FilePermsFromMode(0555)) + } if showSubtasks { - d.AddChild(t, "task", newSubtasks(t, msrc, pidns)) + contents["task"] = newSubtasks(t, msrc, pidns) } - return newFile(d, msrc, fs.SpecialDirectory, t) + + // TODO: Set EUID/EGID based on dumpability. + d := &taskDir{ + Dir: *ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0555)), + t: t, + pidns: pidns, + } + return newProcInode(d, msrc, fs.SpecialDirectory, t) } // subtasks represents a /proc/TID/task directory. @@ -100,15 +109,19 @@ func newTaskDir(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace type subtasks struct { ramfs.Dir - t *kernel.Task - + t *kernel.Task pidns *kernel.PIDNamespace } +var _ fs.InodeOperations = (*subtasks)(nil) + func newSubtasks(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace) *fs.Inode { - s := &subtasks{t: t, pidns: pidns} - s.InitDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newFile(s, msrc, fs.SpecialDirectory, t) + s := &subtasks{ + Dir: *ramfs.NewDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0555)), + t: t, + pidns: pidns, + } + return newProcInode(s, msrc, fs.SpecialDirectory, t) } // UnstableAttr returns unstable attributes of the subtasks. @@ -123,35 +136,52 @@ func (s *subtasks) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.Unstab return uattr, nil } -// Lookup loads an Inode in a task's subtask directory into a Dirent. -func (s *subtasks) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { - tid, err := strconv.ParseUint(p, 10, 32) - if err != nil { - return nil, syserror.ENOENT - } +// GetFile implements fs.InodeOperations.GetFile. +func (s *subtasks) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &subtasksFile{t: s.t, pidns: s.pidns}), nil +} - task := s.pidns.TaskWithID(kernel.ThreadID(tid)) - if task == nil { - return nil, syserror.ENOENT - } - if task.ThreadGroup() != s.t.ThreadGroup() { - return nil, syserror.ENOENT - } +// +stateify savable +type subtasksFile struct { + fsutil.DirFileOperations `state:"nosave"` - td := newTaskDir(task, dir.MountSource, s.pidns, false) - return fs.NewDirent(td, p), nil + t *kernel.Task + pidns *kernel.PIDNamespace } -// DeprecatedReaddir lists a task's subtask directory. -func (s *subtasks) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - tasks := s.t.ThreadGroup().MemberIDs(s.pidns) +// Readdir implements fs.FileOperations.Readdir. +func (f *subtasksFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + dirCtx := fs.DirCtx{ + Serializer: ser, + } + + // Note that unlike most Readdir implementations, the offset here is + // not an index into the subtasks, but rather the TID of the next + // subtask to emit. + offset := file.Offset() + + if offset == 0 { + // Serialize "." and "..". + root := fs.RootFromContext(ctx) + defer root.DecRef() + dot, dotdot := file.Dirent.GetDotAttrs(root) + if err := dirCtx.DirEmit(".", dot); err != nil { + return offset, err + } + if err := dirCtx.DirEmit("..", dotdot); err != nil { + return offset, err + } + } + + // Serialize tasks. + tasks := f.t.ThreadGroup().MemberIDs(f.pidns) taskInts := make([]int, 0, len(tasks)) for _, tid := range tasks { taskInts = append(taskInts, int(tid)) } // Find the task to start at. - idx := sort.SearchInts(taskInts, offset) + idx := sort.SearchInts(taskInts, int(offset)) if idx == len(taskInts) { return offset, nil } @@ -163,12 +193,33 @@ func (s *subtasks) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, off attr := fs.GenericDentAttr(fs.SpecialDirectory, device.ProcDevice) if err := dirCtx.DirEmit(name, attr); err != nil { // Returned offset is next tid to serialize. - return tid, err + return int64(tid), err } } // We serialized them all. Next offset should be higher than last // serialized tid. - return tid + 1, nil + return int64(tid) + 1, nil +} + +var _ fs.FileOperations = (*subtasksFile)(nil) + +// Lookup loads an Inode in a task's subtask directory into a Dirent. +func (s *subtasks) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { + tid, err := strconv.ParseUint(p, 10, 32) + if err != nil { + return nil, syserror.ENOENT + } + + task := s.pidns.TaskWithID(kernel.ThreadID(tid)) + if task == nil { + return nil, syserror.ENOENT + } + if task.ThreadGroup() != s.t.ThreadGroup() { + return nil, syserror.ENOENT + } + + td := newTaskDir(task, dir.MountSource, s.pidns, false) + return fs.NewDirent(td, p), nil } // exe is an fs.InodeOperations symlink for the /proc/PID/exe file. @@ -181,9 +232,11 @@ type exe struct { } func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - exeSymlink := &exe{t: t} - exeSymlink.InitSymlink(t, fs.RootOwner, "") - return newFile(exeSymlink, msrc, fs.Symlink, t) + exeSymlink := &exe{ + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, ""), + t: t, + } + return newProcInode(exeSymlink, msrc, fs.Symlink, t) } func (e *exe) executable() (d *fs.Dirent, err error) { @@ -231,55 +284,48 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { return n, nil } -// namespaceFile represents a file in the namespacefs, such as the files in -// /proc/<pid>/ns. +// namespaceSymlink represents a symlink in the namespacefs, such as the files +// in /proc/<pid>/ns. // // +stateify savable -type namespaceFile struct { +type namespaceSymlink struct { ramfs.Symlink t *kernel.Task } -func newNamespaceFile(t *kernel.Task, msrc *fs.MountSource, name string) *fs.Inode { - n := &namespaceFile{t: t} - n.InitSymlink(t, fs.RootOwner, "") - +func newNamespaceSymlink(t *kernel.Task, msrc *fs.MountSource, name string) *fs.Inode { // TODO: 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. - n.Target = fmt.Sprintf("%s:[%d]", name, device.ProcDevice.NextIno()) - - return newFile(n, msrc, fs.Symlink, t) + target := fmt.Sprintf("%s:[%d]", name, device.ProcDevice.NextIno()) + n := &namespaceSymlink{ + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, target), + t: t, + } + return newProcInode(n, msrc, fs.Symlink, t) } // Getlink implements fs.InodeOperations.Getlink. -func (n *namespaceFile) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) { +func (n *namespaceSymlink) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) { if !kernel.ContextCanTrace(ctx, n.t, false) { return nil, syserror.EACCES } // Create a new regular file to fake the namespace file. - node := &ramfs.Entry{} - node.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0777)) - sattr := fs.StableAttr{ - DeviceID: device.ProcDevice.DeviceID(), - InodeID: device.ProcDevice.NextIno(), - BlockSize: usermem.PageSize, - Type: fs.RegularFile, - } - return fs.NewDirent(fs.NewInode(node, inode.MountSource, sattr), n.Symlink.Target), nil + iops := fsutil.NewNoReadWriteFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0777), linux.PROC_SUPER_MAGIC) + return fs.NewDirent(newProcInode(iops, inode.MountSource, fs.RegularFile, nil), n.Symlink.Target), nil } func newNamespaceDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(t, map[string]*fs.Inode{ - "net": newNamespaceFile(t, msrc, "net"), - "pid": newNamespaceFile(t, msrc, "pid"), - "user": newNamespaceFile(t, msrc, "user"), - }, fs.RootOwner, fs.FilePermsFromMode(0511)) - return newFile(d, msrc, fs.SpecialDirectory, t) + contents := map[string]*fs.Inode{ + "net": newNamespaceSymlink(t, msrc, "net"), + "pid": newNamespaceSymlink(t, msrc, "pid"), + "user": newNamespaceSymlink(t, msrc, "user"), + } + d := ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0511)) + return newProcInode(d, msrc, fs.SpecialDirectory, t) } // mapsData implements seqfile.SeqSource for /proc/[pid]/maps. @@ -290,7 +336,7 @@ type mapsData struct { } func newMaps(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &mapsData{t}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &mapsData{t}), msrc, fs.SpecialFile, t) } func (md *mapsData) mm() *mm.MemoryManager { @@ -330,7 +376,7 @@ type smapsData struct { } func newSmaps(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &smapsData{t}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &smapsData{t}), msrc, fs.SpecialFile, t) } func (sd *smapsData) mm() *mm.MemoryManager { @@ -376,7 +422,7 @@ type taskStatData struct { } func newTaskStat(t *kernel.Task, msrc *fs.MountSource, showSubtasks bool, pidns *kernel.PIDNamespace) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &taskStatData{t, showSubtasks /* tgstats */, pidns}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &taskStatData{t, showSubtasks /* tgstats */, pidns}), msrc, fs.SpecialFile, t) } // NeedsUpdate returns whether the generation is old or not. @@ -450,7 +496,7 @@ type statmData struct { } func newStatm(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &statmData{t}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &statmData{t}), msrc, fs.SpecialFile, t) } // NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. @@ -487,7 +533,7 @@ type statusData struct { } func newStatus(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &statusData{t, pidns}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &statusData{t, pidns}), msrc, fs.SpecialFile, t) } // NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. @@ -552,7 +598,7 @@ type ioData struct { } func newIO(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) } // NeedsUpdate returns whether the generation is old or not. @@ -590,25 +636,49 @@ func (i *ioData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se // // +stateify savable type comm struct { - ramfs.Entry + fsutil.SimpleFileInode t *kernel.Task } // newComm returns a new comm file. func newComm(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - c := &comm{t: t} - c.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(c, msrc, fs.SpecialFile, t) + c := &comm{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + t: t, + } + return newProcInode(c, msrc, fs.SpecialFile, t) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (c *comm) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &commFile{t: c.t}), nil +} + +// +stateify savable +type commFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + + t *kernel.Task } -// DeprecatedPreadv reads the current command name. -func (c *comm) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +var _ fs.FileOperations = (*commFile)(nil) + +// Read implements fs.FileOperations.Read. +func (f *commFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } - buf := []byte(c.t.Name() + "\n") + buf := []byte(f.t.Name() + "\n") if offset >= int64(len(buf)) { return 0, io.EOF } @@ -621,25 +691,47 @@ func (c *comm) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, off // // +stateify savable type auxvec struct { - ramfs.Entry + fsutil.SimpleFileInode t *kernel.Task } // newAuxvec returns a new auxvec file. func newAuxvec(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - a := &auxvec{t: t} - a.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0400)) - return newFile(a, msrc, fs.SpecialFile, t) + a := &auxvec{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + t: t, + } + return newProcInode(a, msrc, fs.SpecialFile, t) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (a *auxvec) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &auxvecFile{t: a.t}), nil +} + +// +stateify savable +type auxvecFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + + t *kernel.Task } -// DeprecatedPreadv reads the current auxiliary vector. -func (a *auxvec) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// Read implements fs.FileOperations.Read. +func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } - m, err := getTaskMM(a.t) + m, err := getTaskMM(f.t) if err != nil { return 0, err } diff --git a/pkg/sentry/fs/proc/uid_gid_map.go b/pkg/sentry/fs/proc/uid_gid_map.go index f70399686..815c40b7f 100644 --- a/pkg/sentry/fs/proc/uid_gid_map.go +++ b/pkg/sentry/fs/proc/uid_gid_map.go @@ -90,12 +90,13 @@ func newGIDMap(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { } func newIDMap(t *kernel.Task, msrc *fs.MountSource, gids bool) *fs.Inode { - imsf := &idMapSeqFile{seqfile.SeqFile{SeqSource: &idMapSeqSource{ - t: t, - gids: gids, - }}} - imsf.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0644)) - return newFile(imsf, msrc, fs.SpecialFile, t) + imsf := &idMapSeqFile{ + *seqfile.NewSeqFile(t, &idMapSeqSource{ + t: t, + gids: gids, + }), + } + return newProcInode(imsf, msrc, fs.SpecialFile, t) } func (imsf *idMapSeqFile) source() *idMapSeqSource { @@ -106,8 +107,8 @@ func (imsf *idMapSeqFile) source() *idMapSeqSource { // Linux 3.18, the limit is five lines." - user_namespaces(7) const maxIDMapLines = 5 -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (imsf *idMapSeqFile) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { +// Write implements fs.FileOperations.Write. +func (imsf *idMapSeqFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { // "In addition, the number of bytes written to the file must be less than // the system page size, and the write must be performed at the start of // the file ..." - user_namespaces(7) diff --git a/pkg/sentry/fs/proc/uptime.go b/pkg/sentry/fs/proc/uptime.go index 80c7ce0b4..40d0fd1fd 100644 --- a/pkg/sentry/fs/proc/uptime.go +++ b/pkg/sentry/fs/proc/uptime.go @@ -18,42 +18,64 @@ import ( "fmt" "io" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // uptime is a file containing the system uptime. // // +stateify savable type uptime struct { - ramfs.Entry + fsutil.SimpleFileInode // The "start time" of the sandbox. startTime ktime.Time } // newUptime returns a new uptime file. -func (p *proc) newUptime(ctx context.Context, msrc *fs.MountSource) *fs.Inode { +func newUptime(ctx context.Context, msrc *fs.MountSource) *fs.Inode { u := &uptime{ - startTime: ktime.NowFromContext(ctx), + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + startTime: ktime.NowFromContext(ctx), } - u.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(u, msrc, fs.SpecialFile, nil) + return newProcInode(u, msrc, fs.SpecialFile, nil) } -// DeprecatedPreadv reads the current uptime. -func (u *uptime) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// GetFile implements fs.InodeOperations.GetFile. +func (u *uptime) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &uptimeFile{startTime: u.startTime}), nil +} + +// +stateify savable +type uptimeFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + + startTime ktime.Time +} + +// Read implements fs.FileOperations.Read. +func (f *uptimeFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } now := ktime.NowFromContext(ctx) // Pretend that we've spent zero time sleeping (second number). - s := []byte(fmt.Sprintf("%.2f 0.00\n", now.Sub(u.startTime).Seconds())) + s := []byte(fmt.Sprintf("%.2f 0.00\n", now.Sub(f.startTime).Seconds())) if offset >= int64(len(s)) { return 0, io.EOF } |