diff options
Diffstat (limited to 'pkg/sentry/fsimpl')
-rw-r--r-- | pkg/sentry/fsimpl/devpts/BUILD | 15 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/devpts/devpts.go | 7 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/fuse/BUILD | 13 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/fuse/fusefs.go | 7 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/kernfs/BUILD | 54 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 27 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/kernfs/kernfs.go | 24 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/kernfs/kernfs_test.go | 12 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/overlay/filesystem.go | 2 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/overlay/non_directory.go | 1 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/BUILD | 61 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/subtasks.go | 7 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task.go | 8 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task_fds.go | 16 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/task_net.go | 6 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/proc/tasks.go | 7 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/sys/BUILD | 15 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/sys/sys.go | 9 | ||||
-rw-r--r-- | pkg/sentry/fsimpl/tmpfs/tmpfs.go | 68 |
19 files changed, 299 insertions, 60 deletions
diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD index 93512c9b6..3f64fab3a 100644 --- a/pkg/sentry/fsimpl/devpts/BUILD +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -1,7 +1,19 @@ load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "root_inode_refs", + out = "root_inode_refs.go", + package = "devpts", + prefix = "rootInode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "rootInode", + }, +) + go_library( name = "devpts", srcs = [ @@ -9,6 +21,7 @@ go_library( "line_discipline.go", "master.go", "queue.go", + "root_inode_refs.go", "slave.go", "terminal.go", ], @@ -16,6 +29,8 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/log", + "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs/lock", diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 3f3a099bd..0eaff9087 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -83,6 +83,7 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds } root.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, devMinor, 1, linux.ModeDirectory|0555) root.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + root.EnableLeakCheck() root.dentry.Init(root) // Construct the pts master inode and dentry. Linux always uses inode @@ -110,6 +111,7 @@ func (fs *filesystem) Release(ctx context.Context) { // rootInode is the root directory inode for the devpts mounts. type rootInode struct { + rootInodeRefs kernfs.AlwaysValid kernfs.InodeAttrs kernfs.InodeDirectoryNoNewChildren @@ -233,3 +235,8 @@ func (i *rootInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, } return offset, nil } + +// DecRef implements kernfs.Inode. +func (i *rootInode) DecRef(context.Context) { + i.rootInodeRefs.DecRef(i.Destroy) +} diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 999111deb..53a4f3012 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -15,6 +15,17 @@ go_template_instance( }, ) +go_template_instance( + name = "inode_refs", + out = "inode_refs.go", + package = "fuse", + prefix = "inode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "inode", + }, +) + go_library( name = "fuse", srcs = [ @@ -22,6 +33,7 @@ go_library( "dev.go", "fusefs.go", "init.go", + "inode_refs.go", "register.go", "request_list.go", ], @@ -30,6 +42,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/log", + "//pkg/refs", "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 44021ee4b..9717c0e15 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -198,6 +198,7 @@ func (fs *filesystem) Release(ctx context.Context) { // inode implements kernfs.Inode. type inode struct { + inodeRefs kernfs.InodeAttrs kernfs.InodeNoDynamicLookup kernfs.InodeNotSymlink @@ -213,6 +214,7 @@ func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *ke i := &inode{} i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + i.EnableLeakCheck() i.dentry.Init(i) return &i.dentry @@ -324,3 +326,8 @@ func (i *inode) Stat(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptio return statFromFUSEAttr(out.Attr, opts.Mask, fusefs.devMinor), nil } + +// DecRef implements kernfs.Inode. +func (i *inode) DecRef(context.Context) { + i.inodeRefs.DecRef(i.Destroy) +} diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 3835557fe..637dca70c 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -26,9 +26,54 @@ go_template_instance( }, ) +go_template_instance( + name = "dentry_refs", + out = "dentry_refs.go", + package = "kernfs", + prefix = "Dentry", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "Dentry", + }, +) + +go_template_instance( + name = "static_directory_refs", + out = "static_directory_refs.go", + package = "kernfs", + prefix = "StaticDirectory", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "StaticDirectory", + }, +) + +go_template_instance( + name = "dir_refs", + out = "dir_refs.go", + package = "kernfs_test", + prefix = "dir", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "dir", + }, +) + +go_template_instance( + name = "readonly_dir_refs", + out = "readonly_dir_refs.go", + package = "kernfs_test", + prefix = "readonlyDir", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "readonlyDir", + }, +) + go_library( name = "kernfs", srcs = [ + "dentry_refs.go", "dynamic_bytes_file.go", "fd_impl_util.go", "filesystem.go", @@ -36,6 +81,7 @@ go_library( "inode_impl_util.go", "kernfs.go", "slot_list.go", + "static_directory_refs.go", "symlink.go", ], visibility = ["//pkg/sentry:internal"], @@ -59,11 +105,17 @@ go_library( go_test( name = "kernfs_test", size = "small", - srcs = ["kernfs_test.go"], + srcs = [ + "dir_refs.go", + "kernfs_test.go", + "readonly_dir_refs.go", + ], deps = [ ":kernfs", "//pkg/abi/linux", "//pkg/context", + "//pkg/log", + "//pkg/refs", "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 885856868..f442a5606 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -20,7 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -344,8 +343,6 @@ type OrderedChildrenOptions struct { // // Must be initialize with Init before first use. type OrderedChildren struct { - refs.AtomicRefCount - // Can children be modified by user syscalls? It set to false, interface // methods that would modify the children return EPERM. Immutable. writable bool @@ -361,14 +358,14 @@ func (o *OrderedChildren) Init(opts OrderedChildrenOptions) { o.set = make(map[string]*slot) } -// DecRef implements Inode.DecRef. -func (o *OrderedChildren) DecRef(ctx context.Context) { - o.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { - o.mu.Lock() - defer o.mu.Unlock() - o.order.Reset() - o.set = nil - }) +// Destroy clears the children stored in o. It should be called by structs +// embedding OrderedChildren upon destruction, i.e. when their reference count +// reaches zero. +func (o *OrderedChildren) Destroy() { + o.mu.Lock() + defer o.mu.Unlock() + o.order.Reset() + o.set = nil } // Populate inserts children into this OrderedChildren, and d's dentry @@ -549,6 +546,7 @@ func (InodeSymlink) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.D // // +stateify savable type StaticDirectory struct { + StaticDirectoryRefs InodeNotSymlink InodeDirectoryNoNewChildren InodeAttrs @@ -594,11 +592,16 @@ func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd return fd.VFSFileDescription(), nil } -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +// SetStat implements kernfs.Inode.SetStat not allowing inode attributes to be changed. func (*StaticDirectory) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } +// DecRef implements kernfs.Inode. +func (s *StaticDirectory) DecRef(context.Context) { + s.StaticDirectoryRefs.DecRef(s.Destroy) +} + // AlwaysValid partially implements kernfs.inodeDynamicLookup. type AlwaysValid struct{} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 51dbc050c..ca3685800 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -57,7 +57,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -161,9 +160,9 @@ const ( // // Must be initialized by Init prior to first use. type Dentry struct { - vfsd vfs.Dentry + DentryRefs - refs.AtomicRefCount + vfsd vfs.Dentry // flags caches useful information about the dentry from the inode. See the // dflags* consts above. Must be accessed by atomic ops. @@ -194,6 +193,7 @@ func (d *Dentry) Init(inode Inode) { if ftype == linux.ModeSymlink { d.flags |= dflagsIsSymlink } + d.EnableLeakCheck() } // VFSDentry returns the generic vfs dentry for this kernfs dentry. @@ -213,16 +213,14 @@ func (d *Dentry) isSymlink() bool { // DecRef implements vfs.DentryImpl.DecRef. func (d *Dentry) DecRef(ctx context.Context) { - d.AtomicRefCount.DecRefWithDestructor(ctx, d.destroy) -} - -// Precondition: Dentry must be removed from VFS' dentry cache. -func (d *Dentry) destroy(ctx context.Context) { - d.inode.DecRef(ctx) // IncRef from Init. - d.inode = nil - if d.parent != nil { - d.parent.DecRef(ctx) // IncRef from Dentry.InsertChild. - } + // Before the destructor is called, Dentry must be removed from VFS' dentry cache. + d.DentryRefs.DecRef(func() { + d.inode.DecRef(ctx) // IncRef from Init. + d.inode = nil + if d.parent != nil { + d.parent.DecRef(ctx) // IncRef from Dentry.InsertChild. + } + }) } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index e5c28c0e4..e376d1736 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -96,6 +96,7 @@ func (*attrs) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.S } type readonlyDir struct { + readonlyDirRefs attrs kernfs.InodeNotSymlink kernfs.InodeNoDynamicLookup @@ -111,6 +112,7 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod dir := &readonlyDir{} dir.attrs.Init(creds, 0 /* devMajor */, 0 /* devMinor */, fs.NextIno(), linux.ModeDirectory|mode) dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + dir.EnableLeakCheck() dir.dentry.Init(dir) dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) @@ -128,7 +130,12 @@ func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs return fd.VFSFileDescription(), nil } +func (d *readonlyDir) DecRef(context.Context) { + d.readonlyDirRefs.DecRef(d.Destroy) +} + type dir struct { + dirRefs attrs kernfs.InodeNotSymlink kernfs.InodeNoDynamicLookup @@ -145,6 +152,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte dir.fs = fs dir.attrs.Init(creds, 0 /* devMajor */, 0 /* devMinor */, fs.NextIno(), linux.ModeDirectory|mode) dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{Writable: true}) + dir.EnableLeakCheck() dir.dentry.Init(dir) dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) @@ -162,6 +170,10 @@ func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, return fd.VFSFileDescription(), nil } +func (d *dir) DecRef(context.Context) { + d.dirRefs.DecRef(d.Destroy) +} + func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { creds := auth.CredentialsFromContext(ctx) dir := d.fs.newDir(creds, opts.Mode, nil) diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index a3cee4047..e720bfb0b 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -30,7 +30,7 @@ import ( // _OVL_XATTR_OPAQUE is an extended attribute key whose value is set to "y" for // opaque directories. // Linux: fs/overlayfs/overlayfs.h:OVL_XATTR_OPAQUE -const _OVL_XATTR_OPAQUE = "trusted.overlay.opaque" +const _OVL_XATTR_OPAQUE = linux.XATTR_TRUSTED_PREFIX + "overlay.opaque" func isWhiteout(stat *linux.Statx) bool { return stat.Mode&linux.S_IFMT == linux.S_IFCHR && stat.RdevMajor == 0 && stat.RdevMinor == 0 diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index d3060a481..268b32537 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -121,7 +121,6 @@ func (fd *nonDirectoryFD) OnClose(ctx context.Context) error { fd.cachedFlags = statusFlags } wrappedFD := fd.cachedFD - defer wrappedFD.IncRef() fd.mu.Unlock() return wrappedFD.OnClose(ctx) } diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 14ecfd300..a45b44440 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -1,18 +1,79 @@ load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "fd_dir_inode_refs", + out = "fd_dir_inode_refs.go", + package = "proc", + prefix = "fdDirInode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "fdDirInode", + }, +) + +go_template_instance( + name = "fd_info_dir_inode_refs", + out = "fd_info_dir_inode_refs.go", + package = "proc", + prefix = "fdInfoDirInode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "fdInfoDirInode", + }, +) + +go_template_instance( + name = "subtasks_inode_refs", + out = "subtasks_inode_refs.go", + package = "proc", + prefix = "subtasksInode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "subtasksInode", + }, +) + +go_template_instance( + name = "task_inode_refs", + out = "task_inode_refs.go", + package = "proc", + prefix = "taskInode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "taskInode", + }, +) + +go_template_instance( + name = "tasks_inode_refs", + out = "tasks_inode_refs.go", + package = "proc", + prefix = "tasksInode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "tasksInode", + }, +) + go_library( name = "proc", srcs = [ + "fd_dir_inode_refs.go", + "fd_info_dir_inode_refs.go", "filesystem.go", "subtasks.go", + "subtasks_inode_refs.go", "task.go", "task_fds.go", "task_files.go", + "task_inode_refs.go", "task_net.go", "tasks.go", "tasks_files.go", + "tasks_inode_refs.go", "tasks_sys.go", ], visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index f25747da3..01c0efb3a 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -31,6 +31,7 @@ import ( // // +stateify savable type subtasksInode struct { + subtasksInodeRefs kernfs.InodeNotSymlink kernfs.InodeDirectoryNoNewChildren kernfs.InodeAttrs @@ -57,6 +58,7 @@ func (fs *filesystem) newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, // Note: credentials are overridden by taskOwnedInode. subInode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + subInode.EnableLeakCheck() inode := &taskOwnedInode{Inode: subInode, owner: task} dentry := &kernfs.Dentry{} @@ -182,3 +184,8 @@ func (i *subtasksInode) Stat(ctx context.Context, vsfs *vfs.Filesystem, opts vfs func (*subtasksInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } + +// DecRef implements kernfs.Inode. +func (i *subtasksInode) DecRef(context.Context) { + i.subtasksInodeRefs.DecRef(i.Destroy) +} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 109b31b4c..66b557abd 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -32,6 +32,7 @@ import ( // // +stateify savable type taskInode struct { + taskInodeRefs kernfs.InodeNotSymlink kernfs.InodeDirectoryNoNewChildren kernfs.InodeNoDynamicLookup @@ -84,6 +85,7 @@ func (fs *filesystem) newTaskInode(task *kernel.Task, pidns *kernel.PIDNamespace taskInode := &taskInode{task: task} // Note: credentials are overridden by taskOwnedInode. taskInode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) + taskInode.EnableLeakCheck() inode := &taskOwnedInode{Inode: taskInode, owner: task} dentry := &kernfs.Dentry{} @@ -119,6 +121,11 @@ func (*taskInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, v return syserror.EPERM } +// DecRef implements kernfs.Inode. +func (i *taskInode) DecRef(context.Context) { + i.taskInodeRefs.DecRef(i.Destroy) +} + // taskOwnedInode implements kernfs.Inode and overrides inode owner with task // effective user and group. type taskOwnedInode struct { @@ -147,6 +154,7 @@ func (fs *filesystem) newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux. dir.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, perm, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) + dir.EnableLeakCheck() inode := &taskOwnedInode{Inode: dir, owner: task} d := &kernfs.Dentry{} diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index e8fcb9aa1..0527b2de8 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -101,6 +100,7 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, off // // +stateify savable type fdDirInode struct { + fdDirInodeRefs kernfs.InodeNotSymlink kernfs.InodeDirectoryNoNewChildren kernfs.InodeAttrs @@ -120,6 +120,7 @@ func (fs *filesystem) newFDDirInode(task *kernel.Task) *kernfs.Dentry { }, } inode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) + inode.EnableLeakCheck() dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -175,6 +176,11 @@ func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentia return err } +// DecRef implements kernfs.Inode. +func (i *fdDirInode) DecRef(context.Context) { + i.fdDirInodeRefs.DecRef(i.Destroy) +} + // fdSymlink is an symlink for the /proc/[pid]/fd/[fd] file. // // +stateify savable @@ -227,6 +233,7 @@ func (s *fdSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDen // // +stateify savable type fdInfoDirInode struct { + fdInfoDirInodeRefs kernfs.InodeNotSymlink kernfs.InodeDirectoryNoNewChildren kernfs.InodeAttrs @@ -245,6 +252,7 @@ func (fs *filesystem) newFDInfoDirInode(task *kernel.Task) *kernfs.Dentry { }, } inode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) + inode.EnableLeakCheck() dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -282,12 +290,16 @@ func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd * return fd.VFSFileDescription(), nil } +// DecRef implements kernfs.Inode. +func (i *fdInfoDirInode) DecRef(context.Context) { + i.fdInfoDirInodeRefs.DecRef(i.Destroy) +} + // fdInfoData implements vfs.DynamicBytesSource for /proc/[pid]/fdinfo/[fd]. // // +stateify savable type fdInfoData struct { kernfs.DynamicBytesFile - refs.AtomicRefCount task *kernel.Task fd int32 diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index a4c884bf9..4e69782c7 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -262,7 +262,7 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { // For now, we always redact this pointer. fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %8d", (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - s.Refs()-1, // RefCount, don't count our own ref. + s.ReadRefs()-1, // RefCount, don't count our own ref. 0, // Protocol, always 0 for UDS. sockFlags, // Flags. sops.Endpoint().Type(), // Type. @@ -430,7 +430,7 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, // Field: refcount. Don't count the ref we obtain while deferencing // the weakref to this socket. - fmt.Fprintf(buf, "%d ", s.Refs()-1) + fmt.Fprintf(buf, "%d ", s.ReadRefs()-1) // Field: Socket struct address. Redacted due to the same reason as // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. @@ -589,7 +589,7 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { // 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 ", s.Refs()-1) + fmt.Fprintf(buf, "%d ", s.ReadRefs()-1) // Field: Socket struct address. Redacted due to the same reason as // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 1391992b7..863c4467e 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -37,6 +37,7 @@ const ( // // +stateify savable type tasksInode struct { + tasksInodeRefs kernfs.InodeNotSymlink kernfs.InodeDirectoryNoNewChildren kernfs.InodeAttrs @@ -84,6 +85,7 @@ func (fs *filesystem) newTasksInode(k *kernel.Kernel, pidns *kernel.PIDNamespace cgroupControllers: cgroupControllers, } inode.InodeAttrs.Init(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) + inode.EnableLeakCheck() dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -226,6 +228,11 @@ func (i *tasksInode) Stat(ctx context.Context, vsfs *vfs.Filesystem, opts vfs.St return stat, nil } +// DecRef implements kernfs.Inode. +func (i *tasksInode) DecRef(context.Context) { + i.tasksInodeRefs.DecRef(i.Destroy) +} + // staticFileSetStat implements a special static file that allows inode // attributes to be set. This is to support /proc files that are readonly, but // allow attributes to be set. diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index f9b232da6..906cd52cb 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -1,10 +1,23 @@ load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "dir_refs", + out = "dir_refs.go", + package = "sys", + prefix = "dir", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "dir", + }, +) + go_library( name = "sys", srcs = [ + "dir_refs.go", "kcov.go", "sys.go", ], @@ -13,6 +26,8 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/coverage", + "//pkg/log", + "//pkg/refs", "//pkg/sentry/arch", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 1f042d9f7..ea30a4ec2 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -118,6 +118,7 @@ func (fs *filesystem) Release(ctx context.Context) { // dir implements kernfs.Inode. type dir struct { + dirRefs kernfs.InodeAttrs kernfs.InodeNoDynamicLookup kernfs.InodeNotSymlink @@ -133,6 +134,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte d := &dir{} d.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) d.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + d.EnableLeakCheck() d.dentry.Init(d) d.IncLinks(d.OrderedChildren.Populate(&d.dentry, contents)) @@ -140,7 +142,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte return &d.dentry } -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +// SetStat implements kernfs.Inode.SetStat not allowing inode attributes to be changed. func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } @@ -156,6 +158,11 @@ func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, return fd.VFSFileDescription(), nil } +// DecRef implements kernfs.Inode.DecRef. +func (d *dir) DecRef(context.Context) { + d.dirRefs.DecRef(d.Destroy) +} + // cpuFile implements kernfs.Inode. type cpuFile struct { kernfs.DynamicBytesFile diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index a7fdf19ca..c4cec4130 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -631,49 +631,65 @@ func (i *inode) listxattr(size uint64) ([]string, error) { } func (i *inode) getxattr(creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) { - if err := i.checkPermissions(creds, vfs.MayRead); err != nil { + if err := i.checkXattrPermissions(creds, opts.Name, vfs.MayRead); err != nil { return "", err } - if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { - return "", syserror.EOPNOTSUPP - } - if !i.userXattrSupported() { - return "", syserror.ENODATA - } return i.xattrs.Getxattr(opts) } func (i *inode) setxattr(creds *auth.Credentials, opts *vfs.SetxattrOptions) error { - if err := i.checkPermissions(creds, vfs.MayWrite); err != nil { + if err := i.checkXattrPermissions(creds, opts.Name, vfs.MayWrite); err != nil { return err } - if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { - return syserror.EOPNOTSUPP - } - if !i.userXattrSupported() { - return syserror.EPERM - } return i.xattrs.Setxattr(opts) } func (i *inode) removexattr(creds *auth.Credentials, name string) error { - if err := i.checkPermissions(creds, vfs.MayWrite); err != nil { + if err := i.checkXattrPermissions(creds, name, vfs.MayWrite); err != nil { return err } - if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { - return syserror.EOPNOTSUPP - } - if !i.userXattrSupported() { - return syserror.EPERM - } return i.xattrs.Removexattr(name) } -// Extended attributes in the user.* namespace are only supported for regular -// files and directories. -func (i *inode) userXattrSupported() bool { - filetype := linux.S_IFMT & atomic.LoadUint32(&i.mode) - return filetype == linux.S_IFREG || filetype == linux.S_IFDIR +func (i *inode) checkXattrPermissions(creds *auth.Credentials, name string, ats vfs.AccessTypes) error { + switch { + case ats&vfs.MayRead == vfs.MayRead: + if err := i.checkPermissions(creds, vfs.MayRead); err != nil { + return err + } + case ats&vfs.MayWrite == vfs.MayWrite: + if err := i.checkPermissions(creds, vfs.MayWrite); err != nil { + return err + } + default: + panic(fmt.Sprintf("checkXattrPermissions called with impossible AccessTypes: %v", ats)) + } + + switch { + case strings.HasPrefix(name, linux.XATTR_TRUSTED_PREFIX): + // The trusted.* namespace can only be accessed by privileged + // users. + if creds.HasCapability(linux.CAP_SYS_ADMIN) { + return nil + } + if ats&vfs.MayWrite == vfs.MayWrite { + return syserror.EPERM + } + return syserror.ENODATA + case strings.HasPrefix(name, linux.XATTR_USER_PREFIX): + // Extended attributes in the user.* namespace are only + // supported for regular files and directories. + filetype := linux.S_IFMT & atomic.LoadUint32(&i.mode) + if filetype == linux.S_IFREG || filetype == linux.S_IFDIR { + return nil + } + if ats&vfs.MayWrite == vfs.MayWrite { + return syserror.EPERM + } + return syserror.ENODATA + + } + return syserror.EOPNOTSUPP } // fileDescription is embedded by tmpfs implementations of |