diff options
Diffstat (limited to 'pkg/sentry/fsimpl/proc/tasks.go')
| -rw-r--r-- | pkg/sentry/fsimpl/proc/tasks.go | 102 |
1 files changed, 79 insertions, 23 deletions
diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 50b2a832f..d8f92d52f 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -27,7 +27,11 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -const defaultPermission = 0444 +const ( + defaultPermission = 0444 + selfName = "self" + threadSelfName = "thread-self" +) // InoGenerator generates unique inode numbers for a given filesystem. type InoGenerator interface { @@ -45,6 +49,11 @@ type tasksInode struct { inoGen InoGenerator pidns *kernel.PIDNamespace + + // '/proc/self' and '/proc/thread-self' have custom directory offsets in + // Linux. So handle them outside of OrderedChildren. + selfSymlink *vfs.Dentry + threadSelfSymlink *vfs.Dentry } var _ kernfs.Inode = (*tasksInode)(nil) @@ -54,20 +63,20 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames contents := map[string]*kernfs.Dentry{ //"cpuinfo": newCPUInfo(ctx, msrc), //"filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), - "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), - "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), - "self": newSelfSymlink(root, inoGen.NextIno(), defaultPermission, pidns), - "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), - "thread-self": newThreadSelfSymlink(root, inoGen.NextIno(), defaultPermission, pidns), + "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), + "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), + "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), //"uptime": newUptime(ctx, msrc), //"version": newVersionData(root, inoGen.NextIno(), k), "version": newDentry(root, inoGen.NextIno(), defaultPermission, &versionData{k: k}), } inode := &tasksInode{ - pidns: pidns, - inoGen: inoGen, + pidns: pidns, + inoGen: inoGen, + selfSymlink: newSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), + threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), } inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) @@ -86,6 +95,13 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro // Try to lookup a corresponding task. tid, err := strconv.ParseUint(name, 10, 64) if err != nil { + // If it failed to parse, check if it's one of the special handled files. + switch name { + case selfName: + return i.selfSymlink, nil + case threadSelfName: + return i.threadSelfSymlink, nil + } return nil, syserror.ENOENT } @@ -104,41 +120,81 @@ func (i *tasksInode) Valid(ctx context.Context) bool { } // IterDirents implements kernfs.inodeDynamicLookup. -// -// TODO(gvisor.dev/issue/1195): Use tgid N offset = TGID_OFFSET + N. -func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { - var tids []int +func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, _ int64) (int64, error) { + // fs/proc/internal.h: #define FIRST_PROCESS_ENTRY 256 + const FIRST_PROCESS_ENTRY = 256 + + // Use maxTaskID to shortcut searches that will result in 0 entries. + const maxTaskID = kernel.TasksLimit + 1 + if offset >= maxTaskID { + return offset, nil + } + + // According to Linux (fs/proc/base.c:proc_pid_readdir()), process directories + // start at offset FIRST_PROCESS_ENTRY with '/proc/self', followed by + // '/proc/thread-self' and then '/proc/[pid]'. + if offset < FIRST_PROCESS_ENTRY { + offset = FIRST_PROCESS_ENTRY + } + + if offset == FIRST_PROCESS_ENTRY { + dirent := vfs.Dirent{ + Name: selfName, + Type: linux.DT_LNK, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } + if offset == FIRST_PROCESS_ENTRY+1 { + dirent := vfs.Dirent{ + Name: threadSelfName, + Type: linux.DT_LNK, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } - // Collect all 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. + // Collect all tasks that TGIDs are greater than the offset specified. Per + // Linux we only include in directory listings if it's the leader. But for + // whatever crazy reason, you can still walk to the given node. + var tids []int + startTid := offset - FIRST_PROCESS_ENTRY - 2 for _, tg := range i.pidns.ThreadGroups() { + tid := i.pidns.IDOfThreadGroup(tg) + if int64(tid) < startTid { + continue + } if leader := tg.Leader(); leader != nil { - tids = append(tids, int(i.pidns.IDOfThreadGroup(tg))) + tids = append(tids, int(tid)) } } if len(tids) == 0 { return offset, nil } - if relOffset >= int64(len(tids)) { - return offset, nil - } sort.Ints(tids) - for _, tid := range tids[relOffset:] { + for _, tid := range tids { dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(tid), 10), Type: linux.DT_DIR, Ino: i.inoGen.NextIno(), - NextOff: offset + 1, + NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, } if !cb.Handle(dirent) { return offset, nil } offset++ } - return offset, nil + return maxTaskID, nil } // Open implements kernfs.Inode. |