diff options
Diffstat (limited to 'pkg/sentry/fsimpl/proc')
-rw-r--r-- | pkg/sentry/fsimpl/proc/BUILD | 1 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task.go | 16 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task_fds.go | 125 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task_files.go | 256 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task_net.go | 92 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/tasks_files.go | 10 |
6 files changed, 233 insertions, 267 deletions
diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 8156984eb..17c1342b5 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -22,7 +22,6 @@ go_library( "//pkg/log", "//pkg/refs", "//pkg/safemem", - "//pkg/sentry/fs", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index aee2a4392..888afc0fd 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -214,22 +214,6 @@ func newIO(t *kernel.Task, isThreadGroup bool) *ioData { return &ioData{ioUsage: t} } -func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { - // Namespace symlinks should contain the namespace name and the inode number - // for the namespace instance, so for example user:[123456]. We currently fake - // the inode number by sticking the symlink inode in its place. - target := fmt.Sprintf("%s:[%d]", ns, ino) - - inode := &kernfs.StaticSymlink{} - // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, target) - - taskInode := &taskOwnedInode{Inode: inode, owner: task} - d := &kernfs.Dentry{} - d.Init(taskInode) - return d -} - // newCgroupData creates inode that shows cgroup information. // From man 7 cgroups: "For each cgroup hierarchy of which the process is a // member, there is one entry containing three colon-separated fields: diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 76bfc5307..046265eca 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -30,34 +30,35 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -type fdDir struct { - inoGen InoGenerator - task *kernel.Task - - // When produceSymlinks is set, dirents produces for the FDs are reported - // as symlink. Otherwise, they are reported as regular files. - produceSymlink bool -} - -func (i *fdDir) lookup(name string) (*vfs.FileDescription, kernel.FDFlags, error) { - fd, err := strconv.ParseUint(name, 10, 64) - if err != nil { - return nil, kernel.FDFlags{}, syserror.ENOENT - } - +func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) { var ( file *vfs.FileDescription flags kernel.FDFlags ) - i.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - file, flags = fdTable.GetVFS2(int32(fd)) + t.WithMuLocked(func(t *kernel.Task) { + if fdt := t.FDTable(); fdt != nil { + file, flags = fdt.GetVFS2(fd) } }) + return file, flags +} + +func taskFDExists(t *kernel.Task, fd int32) bool { + file, _ := getTaskFD(t, fd) if file == nil { - return nil, kernel.FDFlags{}, syserror.ENOENT + return false } - return file, flags, nil + file.DecRef() + return true +} + +type fdDir struct { + inoGen InoGenerator + task *kernel.Task + + // When produceSymlinks is set, dirents produces for the FDs are reported + // as symlink. Otherwise, they are reported as regular files. + produceSymlink bool } // IterDirents implements kernfs.inodeDynamicLookup. @@ -128,11 +129,15 @@ func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // Lookup implements kernfs.inodeDynamicLookup. func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, _, err := i.lookup(name) + fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { - return nil, err + return nil, syserror.ENOENT + } + fd := int32(fdInt) + if !taskFDExists(i.task, fd) { + return nil, syserror.ENOENT } - taskDentry := newFDSymlink(i.task.Credentials(), file, i.inoGen.NextIno()) + taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno()) return taskDentry.VFSDentry(), nil } @@ -169,19 +174,22 @@ func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentia // // +stateify savable type fdSymlink struct { - refs.AtomicRefCount kernfs.InodeAttrs + kernfs.InodeNoopRefCount kernfs.InodeSymlink - file *vfs.FileDescription + task *kernel.Task + fd int32 } var _ kernfs.Inode = (*fdSymlink)(nil) -func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64) *kernfs.Dentry { - file.IncRef() - inode := &fdSymlink{file: file} - inode.Init(creds, ino, linux.ModeSymlink|0777) +func newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { + inode := &fdSymlink{ + task: task, + fd: fd, + } + inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -189,21 +197,25 @@ func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64 } func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { + file, _ := getTaskFD(s.task, s.fd) + if file == nil { + return "", syserror.ENOENT + } + defer file.DecRef() root := vfs.RootFromContext(ctx) defer root.DecRef() - - vfsObj := s.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() - return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry()) -} - -func (s *fdSymlink) DecRef() { - s.AtomicRefCount.DecRefWithDestructor(func() { - s.Destroy() - }) + return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) } -func (s *fdSymlink) Destroy() { - s.file.DecRef() +func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + file, _ := getTaskFD(s.task, s.fd) + if file == nil { + return vfs.VirtualDentry{}, "", syserror.ENOENT + } + defer file.DecRef() + vd := file.VirtualDentry() + vd.IncRef() + return vd, "", nil } // fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory. @@ -238,12 +250,18 @@ func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // Lookup implements kernfs.inodeDynamicLookup. func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, flags, err := i.lookup(name) + fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { - return nil, err + return nil, syserror.ENOENT + } + fd := int32(fdInt) + if !taskFDExists(i.task, fd) { + return nil, syserror.ENOENT + } + data := &fdInfoData{ + task: i.task, + fd: fd, } - - data := &fdInfoData{file: file, flags: flags} dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) return dentry.VFSDentry(), nil } @@ -262,26 +280,23 @@ type fdInfoData struct { kernfs.DynamicBytesFile refs.AtomicRefCount - file *vfs.FileDescription - flags kernel.FDFlags + task *kernel.Task + fd int32 } var _ dynamicInode = (*fdInfoData)(nil) -func (d *fdInfoData) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) -} - -func (d *fdInfoData) destroy() { - d.file.DecRef() -} - // Generate implements vfs.DynamicBytesSource.Generate. func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + file, descriptorFlags := getTaskFD(d.task, d.fd) + if file == nil { + return syserror.ENOENT + } + defer file.DecRef() // TODO(b/121266871): Include pos, locks, and other data. For now we only // have flags. // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt - flags := uint(d.file.StatusFlags()) | d.flags.ToLinuxFileFlags() + flags := uint(file.StatusFlags()) | descriptorFlags.ToLinuxFileFlags() fmt.Fprintf(buf, "flags:\t0%o\n", flags) return nil } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 8c743df8d..2c6f8bdfc 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -18,13 +18,10 @@ import ( "bytes" "fmt" "io" - "sort" - "strings" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -67,6 +64,16 @@ func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { return m, nil } +func checkTaskState(t *kernel.Task) error { + switch t.ExitState() { + case kernel.TaskExitZombie: + return syserror.EACCES + case kernel.TaskExitDead: + return syserror.ESRCH + } + return nil +} + type bufferWriter struct { buf *bytes.Buffer } @@ -613,12 +620,31 @@ func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { return exec.PathnameWithDeleted(ctx), nil } +// Getlink implements kernfs.Inode.Getlink. +func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + if !kernel.ContextCanTrace(ctx, s.task, false) { + return vfs.VirtualDentry{}, "", syserror.EACCES + } + + exec, err := s.executable() + if err != nil { + return vfs.VirtualDentry{}, "", err + } + defer exec.DecRef() + + vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() + vd.IncRef() + return vd, "", nil +} + func (s *exeSymlink) executable() (file fsbridge.File, err error) { + if err := checkTaskState(s.task); err != nil { + return nil, err + } + s.task.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() if mm == nil { - // TODO(b/34851096): Check shouldn't allow Readlink once the - // Task is zombied. err = syserror.EACCES return } @@ -628,57 +654,12 @@ func (s *exeSymlink) executable() (file fsbridge.File, err error) { // (with locks held). file = mm.Executable() if file == nil { - err = syserror.ENOENT + err = syserror.ESRCH } }) return } -// forEachMountSource runs f for the process root mount and each mount that is -// a descendant of the root. -func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { - var fsctx *kernel.FSContext - t.WithMuLocked(func(t *kernel.Task) { - fsctx = t.FSContext() - }) - if fsctx == nil { - // The task has been destroyed. Nothing to show here. - return - } - - // All mount points must be relative to the rootDir, and mounts outside - // will be excluded. - rootDir := fsctx.RootDirectory() - if rootDir == nil { - // The task has been destroyed. Nothing to show here. - return - } - defer rootDir.DecRef() - - mnt := t.MountNamespace().FindMount(rootDir) - if mnt == nil { - // Has it just been unmounted? - return - } - ms := t.MountNamespace().AllMountsUnder(mnt) - sort.Slice(ms, func(i, j int) bool { - return ms[i].ID < ms[j].ID - }) - for _, m := range ms { - mroot := m.Root() - if mroot == nil { - continue // No longer valid. - } - mountPath, desc := mroot.FullName(rootDir) - mroot.DecRef() - if !desc { - // MountSources that are not descendants of the chroot jail are ignored. - continue - } - fn(mountPath, m) - } -} - // mountInfoData is used to implement /proc/[pid]/mountinfo. // // +stateify savable @@ -692,92 +673,22 @@ var _ dynamicInode = (*mountInfoData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - forEachMount(i.task, func(mountPath string, m *fs.Mount) { - mroot := m.Root() - if mroot == nil { - return // No longer valid. - } - defer mroot.DecRef() - - // Format: - // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue - // (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11) - - // (1) MountSource ID. - fmt.Fprintf(buf, "%d ", m.ID) - - // (2) Parent ID (or this ID if there is no parent). - pID := m.ID - if !m.IsRoot() && !m.IsUndo() { - pID = m.ParentID - } - fmt.Fprintf(buf, "%d ", pID) - - // (3) Major:Minor device ID. We don't have a superblock, so we - // just use the root inode device number. - sa := mroot.Inode.StableAttr - fmt.Fprintf(buf, "%d:%d ", sa.DeviceFileMajor, sa.DeviceFileMinor) - - // (4) Root: the pathname of the directory in the filesystem - // which forms the root of this mount. - // - // NOTE(b/78135857): This will always be "/" until we implement - // bind mounts. - fmt.Fprintf(buf, "/ ") - - // (5) Mount point (relative to process root). - fmt.Fprintf(buf, "%s ", mountPath) - - // (6) Mount options. - flags := mroot.Inode.MountSource.Flags - opts := "rw" - if flags.ReadOnly { - opts = "ro" - } - if flags.NoAtime { - opts += ",noatime" - } - if flags.NoExec { - opts += ",noexec" - } - fmt.Fprintf(buf, "%s ", opts) - - // (7) Optional fields: zero or more fields of the form "tag[:value]". - // (8) Separator: the end of the optional fields is marked by a single hyphen. - fmt.Fprintf(buf, "- ") - - // (9) Filesystem type. - fmt.Fprintf(buf, "%s ", mroot.Inode.MountSource.FilesystemType) - - // (10) Mount source: filesystem-specific information or "none". - fmt.Fprintf(buf, "none ") - - // (11) Superblock options, and final newline. - fmt.Fprintf(buf, "%s\n", superBlockOpts(mountPath, mroot.Inode.MountSource)) + var fsctx *kernel.FSContext + i.task.WithMuLocked(func(t *kernel.Task) { + fsctx = t.FSContext() }) - return nil -} - -func superBlockOpts(mountPath string, msrc *fs.MountSource) string { - // gVisor doesn't (yet) have a concept of super block options, so we - // use the ro/rw bit from the mount flag. - opts := "rw" - if msrc.Flags.ReadOnly { - opts = "ro" + if fsctx == nil { + // The task has been destroyed. Nothing to show here. + return nil } - - // NOTE(b/147673608): If the mount is a cgroup, we also need to include - // the cgroup name in the options. For now we just read that from the - // path. - // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we - // should get this value from the cgroup itself, and not rely on the - // path. - if msrc.FilesystemType == "cgroup" { - splitPath := strings.Split(mountPath, "/") - cgroupType := splitPath[len(splitPath)-1] - opts += "," + cgroupType + rootDir := fsctx.RootDirectoryVFS2() + if !rootDir.Ok() { + // Root has been destroyed. Don't try to read mounts. + return nil } - return opts + defer rootDir.DecRef() + i.task.Kernel().VFS().GenerateProcMountInfo(ctx, rootDir, buf) + return nil } // mountsData is used to implement /proc/[pid]/mounts. @@ -789,33 +700,62 @@ type mountsData struct { task *kernel.Task } -var _ dynamicInode = (*mountInfoData)(nil) +var _ dynamicInode = (*mountsData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - forEachMount(i.task, func(mountPath string, m *fs.Mount) { - // Format: - // <special device or remote filesystem> <mount point> <filesystem type> <mount options> <needs dump> <fsck order> - // - // We use the filesystem name as the first field, since there - // is no real block device we can point to, and we also should - // not expose anything about the remote filesystem. - // - // Only ro/rw option is supported for now. - // - // The "needs dump"and fsck flags are always 0, which is allowed. - root := m.Root() - if root == nil { - return // No longer valid. - } - defer root.DecRef() - - flags := root.Inode.MountSource.Flags - opts := "rw" - if flags.ReadOnly { - opts = "ro" - } - fmt.Fprintf(buf, "%s %s %s %s %d %d\n", "none", mountPath, root.Inode.MountSource.FilesystemType, opts, 0, 0) + var fsctx *kernel.FSContext + i.task.WithMuLocked(func(t *kernel.Task) { + fsctx = t.FSContext() }) + if fsctx == nil { + // The task has been destroyed. Nothing to show here. + return nil + } + rootDir := fsctx.RootDirectoryVFS2() + if !rootDir.Ok() { + // Root has been destroyed. Don't try to read mounts. + return nil + } + defer rootDir.DecRef() + i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf) return nil } + +type namespaceSymlink struct { + kernfs.StaticSymlink + + task *kernel.Task +} + +func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { + // Namespace symlinks should contain the namespace name and the inode number + // for the namespace instance, so for example user:[123456]. We currently fake + // the inode number by sticking the symlink inode in its place. + target := fmt.Sprintf("%s:[%d]", ns, ino) + + inode := &namespaceSymlink{task: task} + // Note: credentials are overridden by taskOwnedInode. + inode.Init(task.Credentials(), ino, target) + + taskInode := &taskOwnedInode{Inode: inode, owner: task} + d := &kernfs.Dentry{} + d.Init(taskInode) + return d +} + +// Readlink implements Inode. +func (s *namespaceSymlink) Readlink(ctx context.Context) (string, error) { + if err := checkTaskState(s.task); err != nil { + return "", err + } + return s.StaticSymlink.Readlink(ctx) +} + +// Getlink implements Inode.Getlink. +func (s *namespaceSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + if err := checkTaskState(s.task); err != nil { + return vfs.VirtualDentry{}, "", err + } + return s.StaticSymlink.Getlink(ctx) +} diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 373a7b17d..6595fcee6 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -24,7 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -32,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/usermem" @@ -206,22 +206,21 @@ var _ dynamicInode = (*netUnixData)(nil) func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") for _, se := range n.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) + s := se.SockVFS2 + if !s.TryIncRef() { + log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) continue } - sfile := s.(*fs.File) - if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { + if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX { s.DecRef() // Not a unix socket. continue } - sops := sfile.FileOperations.(*unix.SocketOperations) + sops := s.Impl().(*unix.SocketVFS2) addr, err := sops.Endpoint().GetLocalAddress() if err != nil { - log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) + log.Warningf("Failed to retrieve socket name from %+v: %v", s, err) addr.Addr = "<unknown>" } @@ -234,6 +233,15 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { } } + // Get inode number. + var ino uint64 + stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_INO}) + if statErr != nil || stat.Mask&linux.STATX_INO == 0 { + log.Warningf("Failed to retrieve ino for socket file: %v", statErr) + } else { + ino = stat.Ino + } + // In the socket entry below, the value for the 'Num' field requires // some consideration. Linux prints the address to the struct // unix_sock representing a socket in the kernel, but may redact the @@ -252,14 +260,14 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { // the definition of this struct changes over time. // // For now, we always redact this pointer. - fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", + fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %8d", (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - sfile.ReadRefs()-1, // RefCount, don't count our own ref. + s.Refs()-1, // RefCount, don't count our own ref. 0, // Protocol, always 0 for UDS. sockFlags, // Flags. sops.Endpoint().Type(), // Type. sops.State(), // State. - sfile.InodeID(), // Inode. + ino, // Inode. ) // Path @@ -341,15 +349,14 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, t := kernel.TaskFromContext(ctx) for _, se := range k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + s := se.SockVFS2 + if !s.TryIncRef() { + log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) continue } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) + sops, ok := s.Impl().(socket.SocketVFS2) if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { s.DecRef() @@ -398,14 +405,15 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, // Unimplemented. fmt.Fprintf(buf, "%08X ", 0) + stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO}) + // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + if statErr != nil || stat.Mask&linux.STATX_UID == 0 { + log.Warningf("Failed to retrieve uid for socket file: %v", statErr) fmt.Fprintf(buf, "%5d ", 0) } else { creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow())) } // Field: timeout; number of unanswered 0-window probes. @@ -413,11 +421,16 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, fmt.Fprintf(buf, "%8d ", 0) // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + if statErr != nil || stat.Mask&linux.STATX_INO == 0 { + log.Warningf("Failed to retrieve inode for socket file: %v", statErr) + fmt.Fprintf(buf, "%8d ", 0) + } else { + fmt.Fprintf(buf, "%8d ", stat.Ino) + } // Field: refcount. Don't count the ref we obtain while deferencing // the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + fmt.Fprintf(buf, "%d ", s.Refs()-1) // Field: Socket struct address. Redacted due to the same reason as // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. @@ -499,15 +512,14 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { t := kernel.TaskFromContext(ctx) for _, se := range d.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + s := se.SockVFS2 + if !s.TryIncRef() { + log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) continue } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) + sops, ok := s.Impl().(socket.SocketVFS2) if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { s.DecRef() @@ -551,25 +563,31 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Field: retrnsmt. Always 0 for UDP. fmt.Fprintf(buf, "%08X ", 0) + stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO}) + // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + if statErr != nil || stat.Mask&linux.STATX_UID == 0 { + log.Warningf("Failed to retrieve uid for socket file: %v", statErr) fmt.Fprintf(buf, "%5d ", 0) } else { creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow())) } // Field: timeout. Always 0 for UDP. fmt.Fprintf(buf, "%8d ", 0) // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + if statErr != nil || stat.Mask&linux.STATX_INO == 0 { + log.Warningf("Failed to retrieve inode for socket file: %v", statErr) + fmt.Fprintf(buf, "%8d ", 0) + } else { + fmt.Fprintf(buf, "%8d ", stat.Ino) + } // Field: ref; reference count on the socket inode. Don't count the ref // we obtain while deferencing the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + fmt.Fprintf(buf, "%d ", s.Refs()-1) // Field: Socket struct address. Redacted due to the same reason as // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. @@ -670,9 +688,9 @@ func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { if line.prefix == "Tcp" { tcp := stat.(*inet.StatSNMPTCP) // "Tcp" needs special processing because MaxConn is signed. RFC 2012. - fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) + fmt.Fprintf(buf, "%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) } else { - fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) + fmt.Fprintf(buf, "%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) } } return nil diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 882c1981e..4621e2de0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -63,6 +63,11 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { return strconv.FormatUint(uint64(tgid), 10), nil } +func (s *selfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + target, err := s.Readlink(ctx) + return vfs.VirtualDentry{}, target, err +} + // SetStat implements Inode.SetStat not allowing inode attributes to be changed. func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM @@ -101,6 +106,11 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { return fmt.Sprintf("%d/task/%d", tgid, tid), nil } +func (s *threadSelfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + target, err := s.Readlink(ctx) + return vfs.VirtualDentry{}, target, err +} + // SetStat implements Inode.SetStat not allowing inode attributes to be changed. func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM |