summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fs
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/fs')
-rw-r--r--pkg/sentry/fs/dirent_cache.go2
-rw-r--r--pkg/sentry/fs/host/BUILD4
-rw-r--r--pkg/sentry/fs/host/control.go2
-rw-r--r--pkg/sentry/fs/host/descriptor.go37
-rw-r--r--pkg/sentry/fs/host/descriptor_state.go2
-rw-r--r--pkg/sentry/fs/host/descriptor_test.go4
-rw-r--r--pkg/sentry/fs/host/file.go4
-rw-r--r--pkg/sentry/fs/host/fs.go339
-rw-r--r--pkg/sentry/fs/host/fs_test.go380
-rw-r--r--pkg/sentry/fs/host/host.go59
-rw-r--r--pkg/sentry/fs/host/inode.go141
-rw-r--r--pkg/sentry/fs/host/inode_state.go32
-rw-r--r--pkg/sentry/fs/host/inode_test.go66
-rw-r--r--pkg/sentry/fs/host/util.go96
-rw-r--r--pkg/sentry/fs/host/util_unsafe.go41
-rw-r--r--pkg/sentry/fs/inotify.go5
-rw-r--r--pkg/sentry/fs/proc/net.go53
-rw-r--r--pkg/sentry/fs/proc/proc.go2
-rw-r--r--pkg/sentry/fs/proc/task.go130
19 files changed, 248 insertions, 1151 deletions
diff --git a/pkg/sentry/fs/dirent_cache.go b/pkg/sentry/fs/dirent_cache.go
index 25514ace4..33de32c69 100644
--- a/pkg/sentry/fs/dirent_cache.go
+++ b/pkg/sentry/fs/dirent_cache.go
@@ -101,8 +101,6 @@ func (c *DirentCache) remove(d *Dirent) {
panic(fmt.Sprintf("trying to remove %v, which is not in the dirent cache", d))
}
c.list.Remove(d)
- d.SetPrev(nil)
- d.SetNext(nil)
d.DecRef()
c.currentSize--
if c.limit != nil {
diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD
index 21003ea45..011625c80 100644
--- a/pkg/sentry/fs/host/BUILD
+++ b/pkg/sentry/fs/host/BUILD
@@ -10,7 +10,7 @@ go_library(
"descriptor_state.go",
"device.go",
"file.go",
- "fs.go",
+ "host.go",
"inode.go",
"inode_state.go",
"ioctl_unsafe.go",
@@ -62,14 +62,12 @@ go_test(
size = "small",
srcs = [
"descriptor_test.go",
- "fs_test.go",
"inode_test.go",
"socket_test.go",
"wait_test.go",
],
library = ":host",
deps = [
- "//pkg/context",
"//pkg/fd",
"//pkg/fdnotifier",
"//pkg/sentry/contexttest",
diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go
index 1658979fc..cd84e1337 100644
--- a/pkg/sentry/fs/host/control.go
+++ b/pkg/sentry/fs/host/control.go
@@ -32,6 +32,8 @@ func newSCMRights(fds []int) control.SCMRights {
}
// Files implements control.SCMRights.Files.
+//
+// TODO(gvisor.dev/issue/2017): Port to VFS2.
func (c *scmRights) Files(ctx context.Context, max int) (control.RightsFiles, bool) {
n := max
var trunc bool
diff --git a/pkg/sentry/fs/host/descriptor.go b/pkg/sentry/fs/host/descriptor.go
index 2a4d1b291..cfdce6a74 100644
--- a/pkg/sentry/fs/host/descriptor.go
+++ b/pkg/sentry/fs/host/descriptor.go
@@ -16,7 +16,6 @@ package host
import (
"fmt"
- "path"
"syscall"
"gvisor.dev/gvisor/pkg/fdnotifier"
@@ -28,12 +27,9 @@ import (
//
// +stateify savable
type descriptor struct {
- // donated is true if the host fd was donated by another process.
- donated bool
-
// If origFD >= 0, it is the host fd that this file was originally created
// from, which must be available at time of restore. The FD can be closed
- // after descriptor is created. Only set if donated is true.
+ // after descriptor is created.
origFD int
// wouldBlock is true if value (below) points to a file that can
@@ -41,15 +37,13 @@ type descriptor struct {
wouldBlock bool
// value is the wrapped host fd. It is never saved or restored
- // directly. How it is restored depends on whether it was
- // donated and the fs.MountSource it was originally
- // opened/created from.
+ // directly.
value int `state:"nosave"`
}
// newDescriptor returns a wrapped host file descriptor. On success,
// the descriptor is registered for event notifications with queue.
-func newDescriptor(fd int, donated bool, saveable bool, wouldBlock bool, queue *waiter.Queue) (*descriptor, error) {
+func newDescriptor(fd int, saveable bool, wouldBlock bool, queue *waiter.Queue) (*descriptor, error) {
ownedFD := fd
origFD := -1
if saveable {
@@ -69,7 +63,6 @@ func newDescriptor(fd int, donated bool, saveable bool, wouldBlock bool, queue *
}
}
return &descriptor{
- donated: donated,
origFD: origFD,
wouldBlock: wouldBlock,
value: ownedFD,
@@ -77,25 +70,11 @@ func newDescriptor(fd int, donated bool, saveable bool, wouldBlock bool, queue *
}
// initAfterLoad initializes the value of the descriptor after Load.
-func (d *descriptor) initAfterLoad(mo *superOperations, id uint64, queue *waiter.Queue) error {
- if d.donated {
- var err error
- d.value, err = syscall.Dup(d.origFD)
- if err != nil {
- return fmt.Errorf("failed to dup restored fd %d: %v", d.origFD, err)
- }
- } else {
- name, ok := mo.inodeMappings[id]
- if !ok {
- return fmt.Errorf("failed to find path for inode number %d", id)
- }
- fullpath := path.Join(mo.root, name)
-
- var err error
- d.value, err = open(nil, fullpath)
- if err != nil {
- return fmt.Errorf("failed to open %q: %v", fullpath, err)
- }
+func (d *descriptor) initAfterLoad(id uint64, queue *waiter.Queue) error {
+ var err error
+ d.value, err = syscall.Dup(d.origFD)
+ if err != nil {
+ return fmt.Errorf("failed to dup restored fd %d: %v", d.origFD, err)
}
if d.wouldBlock {
if err := syscall.SetNonblock(d.value, true); err != nil {
diff --git a/pkg/sentry/fs/host/descriptor_state.go b/pkg/sentry/fs/host/descriptor_state.go
index 8167390a9..e880582ab 100644
--- a/pkg/sentry/fs/host/descriptor_state.go
+++ b/pkg/sentry/fs/host/descriptor_state.go
@@ -16,7 +16,7 @@ package host
// beforeSave is invoked by stateify.
func (d *descriptor) beforeSave() {
- if d.donated && d.origFD < 0 {
+ if d.origFD < 0 {
panic("donated file descriptor cannot be saved")
}
}
diff --git a/pkg/sentry/fs/host/descriptor_test.go b/pkg/sentry/fs/host/descriptor_test.go
index 4205981f5..d8e4605b6 100644
--- a/pkg/sentry/fs/host/descriptor_test.go
+++ b/pkg/sentry/fs/host/descriptor_test.go
@@ -47,10 +47,10 @@ func TestDescriptorRelease(t *testing.T) {
// FD ownership is transferred to the descritor.
queue := &waiter.Queue{}
- d, err := newDescriptor(fd, false /* donated*/, tc.saveable, tc.wouldBlock, queue)
+ d, err := newDescriptor(fd, tc.saveable, tc.wouldBlock, queue)
if err != nil {
syscall.Close(fd)
- t.Fatalf("newDescriptor(%d, %t, false, %t, queue) failed, err: %v", fd, tc.saveable, tc.wouldBlock, err)
+ t.Fatalf("newDescriptor(%d, %t, %t, queue) failed, err: %v", fd, tc.saveable, tc.wouldBlock, err)
}
if tc.saveable {
if d.origFD < 0 {
diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go
index e08f56d04..034862694 100644
--- a/pkg/sentry/fs/host/file.go
+++ b/pkg/sentry/fs/host/file.go
@@ -101,8 +101,8 @@ func newFileFromDonatedFD(ctx context.Context, donated int, mounter fs.FileOwner
})
return s, nil
default:
- msrc := newMountSource(ctx, "/", mounter, &Filesystem{}, fs.MountSourceFlags{}, false /* dontTranslateOwnership */)
- inode, err := newInode(ctx, msrc, donated, saveable, true /* donated */)
+ msrc := fs.NewNonCachingMountSource(ctx, &filesystem{}, fs.MountSourceFlags{})
+ inode, err := newInode(ctx, msrc, donated, saveable)
if err != nil {
return nil, err
}
diff --git a/pkg/sentry/fs/host/fs.go b/pkg/sentry/fs/host/fs.go
deleted file mode 100644
index d3e8e3a36..000000000
--- a/pkg/sentry/fs/host/fs.go
+++ /dev/null
@@ -1,339 +0,0 @@
-// Copyright 2018 The gVisor Authors.
-//
-// 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 host implements an fs.Filesystem for files backed by host
-// file descriptors.
-package host
-
-import (
- "fmt"
- "path"
- "path/filepath"
- "strconv"
- "strings"
-
- "gvisor.dev/gvisor/pkg/context"
- "gvisor.dev/gvisor/pkg/log"
- "gvisor.dev/gvisor/pkg/sentry/fs"
-)
-
-// FilesystemName is the name under which Filesystem is registered.
-const FilesystemName = "whitelistfs"
-
-const (
- // whitelistKey is the mount option containing a comma-separated list
- // of host paths to whitelist.
- whitelistKey = "whitelist"
-
- // rootPathKey is the mount option containing the root path of the
- // mount.
- rootPathKey = "root"
-
- // dontTranslateOwnershipKey is the key to superOperations.dontTranslateOwnership.
- dontTranslateOwnershipKey = "dont_translate_ownership"
-)
-
-// maxTraversals determines link traversals in building the whitelist.
-const maxTraversals = 10
-
-// Filesystem is a pseudo file system that is only available during the setup
-// to lock down the configurations. This filesystem should only be mounted at root.
-//
-// Think twice before exposing this to applications.
-//
-// +stateify savable
-type Filesystem struct {
- // whitelist is a set of host paths to whitelist.
- paths []string
-}
-
-var _ fs.Filesystem = (*Filesystem)(nil)
-
-// Name is the identifier of this file system.
-func (*Filesystem) Name() string {
- return FilesystemName
-}
-
-// AllowUserMount prohibits users from using mount(2) with this file system.
-func (*Filesystem) AllowUserMount() bool {
- return false
-}
-
-// AllowUserList allows this filesystem to be listed in /proc/filesystems.
-func (*Filesystem) AllowUserList() bool {
- return true
-}
-
-// Flags returns that there is nothing special about this file system.
-func (*Filesystem) Flags() fs.FilesystemFlags {
- return 0
-}
-
-// Mount returns an fs.Inode exposing the host file system. It is intended to be locked
-// down in PreExec below.
-func (f *Filesystem) Mount(ctx context.Context, _ string, flags fs.MountSourceFlags, data string, _ interface{}) (*fs.Inode, error) {
- // Parse generic comma-separated key=value options.
- options := fs.GenericMountSourceOptions(data)
-
- // Grab the whitelist if one was specified.
- // TODO(edahlgren/mpratt/hzy): require another option "testonly" in order to allow
- // no whitelist.
- if wl, ok := options[whitelistKey]; ok {
- f.paths = strings.Split(wl, "|")
- delete(options, whitelistKey)
- }
-
- // If the rootPath was set, use it. Othewise default to the root of the
- // host fs.
- rootPath := "/"
- if rp, ok := options[rootPathKey]; ok {
- rootPath = rp
- delete(options, rootPathKey)
-
- // We must relativize the whitelisted paths to the new root.
- for i, p := range f.paths {
- rel, err := filepath.Rel(rootPath, p)
- if err != nil {
- return nil, fmt.Errorf("whitelist path %q must be a child of root path %q", p, rootPath)
- }
- f.paths[i] = path.Join("/", rel)
- }
- }
- fd, err := open(nil, rootPath)
- if err != nil {
- return nil, fmt.Errorf("failed to find root: %v", err)
- }
-
- var dontTranslateOwnership bool
- if v, ok := options[dontTranslateOwnershipKey]; ok {
- b, err := strconv.ParseBool(v)
- if err != nil {
- return nil, fmt.Errorf("invalid value for %q: %v", dontTranslateOwnershipKey, err)
- }
- dontTranslateOwnership = b
- delete(options, dontTranslateOwnershipKey)
- }
-
- // Fail if the caller passed us more options than we know about.
- if len(options) > 0 {
- return nil, fmt.Errorf("unsupported mount options: %v", options)
- }
-
- // The mounting EUID/EGID will be cached by this file system. This will
- // be used to assign ownership to files that we own.
- owner := fs.FileOwnerFromContext(ctx)
-
- // Construct the host file system mount and inode.
- msrc := newMountSource(ctx, rootPath, owner, f, flags, dontTranslateOwnership)
- return newInode(ctx, msrc, fd, false /* saveable */, false /* donated */)
-}
-
-// InstallWhitelist locks down the MountNamespace to only the currently installed
-// Dirents and the given paths.
-func (f *Filesystem) InstallWhitelist(ctx context.Context, m *fs.MountNamespace) error {
- return installWhitelist(ctx, m, f.paths)
-}
-
-func installWhitelist(ctx context.Context, m *fs.MountNamespace, paths []string) error {
- if len(paths) == 0 || (len(paths) == 1 && paths[0] == "") {
- // Warning will be logged during filter installation if the empty
- // whitelist matters (allows for host file access).
- return nil
- }
-
- // Done tracks entries already added.
- done := make(map[string]bool)
- root := m.Root()
- defer root.DecRef()
-
- for i := 0; i < len(paths); i++ {
- // Make sure the path is absolute. This is a sanity check.
- if !path.IsAbs(paths[i]) {
- return fmt.Errorf("path %q is not absolute", paths[i])
- }
-
- // We need to add all the intermediate paths, in case one of
- // them is a symlink that needs to be resolved.
- for j := 1; j <= len(paths[i]); j++ {
- if j < len(paths[i]) && paths[i][j] != '/' {
- continue
- }
- current := paths[i][:j]
-
- // Lookup the given component in the tree.
- remainingTraversals := uint(maxTraversals)
- d, err := m.FindLink(ctx, root, nil, current, &remainingTraversals)
- if err != nil {
- log.Warningf("populate failed for %q: %v", current, err)
- continue
- }
-
- // It's critical that this DecRef happens after the
- // freeze below. This ensures that the dentry is in
- // place to be frozen. Otherwise, we freeze without
- // these entries.
- defer d.DecRef()
-
- // Expand the last component if necessary.
- if current == paths[i] {
- // Is it a directory or symlink?
- sattr := d.Inode.StableAttr
- if fs.IsDir(sattr) {
- for name := range childDentAttrs(ctx, d) {
- paths = append(paths, path.Join(current, name))
- }
- }
- if fs.IsSymlink(sattr) {
- // Only expand symlinks once. The
- // folder structure may contain
- // recursive symlinks and we don't want
- // to end up infinitely expanding this
- // symlink. This is safe because this
- // is the last component. If a later
- // path wants to symlink something
- // beneath this symlink that will still
- // be handled by the FindLink above.
- if done[current] {
- continue
- }
-
- s, err := d.Inode.Readlink(ctx)
- if err != nil {
- log.Warningf("readlink failed for %q: %v", current, err)
- continue
- }
- if path.IsAbs(s) {
- paths = append(paths, s)
- } else {
- target := path.Join(path.Dir(current), s)
- paths = append(paths, target)
- }
- }
- }
-
- // Only report this one once even though we may look
- // it up more than once. If we whitelist /a/b,/a then
- // /a will be "done" when it is looked up for /a/b,
- // however we still need to expand all of its contents
- // when whitelisting /a.
- if !done[current] {
- log.Debugf("whitelisted: %s", current)
- }
- done[current] = true
- }
- }
-
- // Freeze the mount tree in place. This prevents any new paths from
- // being opened and any old ones from being removed. If we do provide
- // tmpfs mounts, we'll want to freeze/thaw those separately.
- m.Freeze()
- return nil
-}
-
-func childDentAttrs(ctx context.Context, d *fs.Dirent) map[string]fs.DentAttr {
- dirname, _ := d.FullName(nil /* root */)
- dir, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true})
- if err != nil {
- log.Warningf("failed to open directory %q: %v", dirname, err)
- return nil
- }
- dir.DecRef()
- var stubSerializer fs.CollectEntriesSerializer
- if err := dir.Readdir(ctx, &stubSerializer); err != nil {
- log.Warningf("failed to iterate on host directory %q: %v", dirname, err)
- return nil
- }
- delete(stubSerializer.Entries, ".")
- delete(stubSerializer.Entries, "..")
- return stubSerializer.Entries
-}
-
-// newMountSource constructs a new host fs.MountSource
-// relative to a root path. The root should match the mount point.
-func newMountSource(ctx context.Context, root string, mounter fs.FileOwner, filesystem fs.Filesystem, flags fs.MountSourceFlags, dontTranslateOwnership bool) *fs.MountSource {
- return fs.NewMountSource(ctx, &superOperations{
- root: root,
- inodeMappings: make(map[uint64]string),
- mounter: mounter,
- dontTranslateOwnership: dontTranslateOwnership,
- }, filesystem, flags)
-}
-
-// superOperations implements fs.MountSourceOperations.
-//
-// +stateify savable
-type superOperations struct {
- fs.SimpleMountSourceOperations
-
- // root is the path of the mount point. All inode mappings
- // are relative to this root.
- root string
-
- // inodeMappings contains mappings of fs.Inodes associated
- // with this MountSource to paths under root.
- inodeMappings map[uint64]string
-
- // mounter is the cached EUID/EGID that mounted this file system.
- mounter fs.FileOwner
-
- // dontTranslateOwnership indicates whether to not translate file
- // ownership.
- //
- // By default, files/directories owned by the sandbox uses UID/GID
- // of the mounter. For files/directories that are not owned by the
- // sandbox, file UID/GID is translated to a UID/GID which cannot
- // be mapped in the sandboxed application's user namespace. The
- // UID/GID will look like the nobody UID/GID (65534) but is not
- // strictly owned by the user "nobody".
- //
- // If whitelistfs is a lower filesystem in an overlay, set
- // dont_translate_ownership=true in mount options.
- dontTranslateOwnership bool
-}
-
-var _ fs.MountSourceOperations = (*superOperations)(nil)
-
-// ResetInodeMappings implements fs.MountSourceOperations.ResetInodeMappings.
-func (m *superOperations) ResetInodeMappings() {
- m.inodeMappings = make(map[uint64]string)
-}
-
-// SaveInodeMapping implements fs.MountSourceOperations.SaveInodeMapping.
-func (m *superOperations) SaveInodeMapping(inode *fs.Inode, path string) {
- // This is very unintuitive. We *CANNOT* trust the inode's StableAttrs,
- // because overlay copyUp may have changed them out from under us.
- // So much for "immutable".
- sattr := inode.InodeOperations.(*inodeOperations).fileState.sattr
- m.inodeMappings[sattr.InodeID] = path
-}
-
-// Keep implements fs.MountSourceOperations.Keep.
-//
-// TODO(b/72455313,b/77596690): It is possible to change the permissions on a
-// host file while it is in the dirent cache (say from RO to RW), but it is not
-// possible to re-open the file with more relaxed permissions, since the host
-// FD is already open and stored in the inode.
-//
-// Using the dirent LRU cache increases the odds that this bug is encountered.
-// Since host file access is relatively fast anyways, we disable the LRU cache
-// for host fs files. Once we can properly deal with permissions changes and
-// re-opening host files, we should revisit whether or not to make use of the
-// LRU cache.
-func (*superOperations) Keep(*fs.Dirent) bool {
- return false
-}
-
-func init() {
- fs.RegisterFilesystem(&Filesystem{})
-}
diff --git a/pkg/sentry/fs/host/fs_test.go b/pkg/sentry/fs/host/fs_test.go
deleted file mode 100644
index 3111d2df9..000000000
--- a/pkg/sentry/fs/host/fs_test.go
+++ /dev/null
@@ -1,380 +0,0 @@
-// Copyright 2018 The gVisor Authors.
-//
-// 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 host
-
-import (
- "fmt"
- "io/ioutil"
- "os"
- "path"
- "reflect"
- "sort"
- "testing"
-
- "gvisor.dev/gvisor/pkg/context"
- "gvisor.dev/gvisor/pkg/sentry/contexttest"
- "gvisor.dev/gvisor/pkg/sentry/fs"
-)
-
-// newTestMountNamespace creates a MountNamespace with a ramfs root.
-// It returns the host folder created, which should be removed when done.
-func newTestMountNamespace(t *testing.T) (*fs.MountNamespace, string, error) {
- p, err := ioutil.TempDir("", "root")
- if err != nil {
- return nil, "", err
- }
-
- fd, err := open(nil, p)
- if err != nil {
- os.RemoveAll(p)
- return nil, "", err
- }
- ctx := contexttest.Context(t)
- root, err := newInode(ctx, newMountSource(ctx, p, fs.RootOwner, &Filesystem{}, fs.MountSourceFlags{}, false), fd, false, false)
- if err != nil {
- os.RemoveAll(p)
- return nil, "", err
- }
- mm, err := fs.NewMountNamespace(ctx, root)
- if err != nil {
- os.RemoveAll(p)
- return nil, "", err
- }
- return mm, p, nil
-}
-
-// createTestDirs populates the root with some test files and directories.
-// /a/a1.txt
-// /a/a2.txt
-// /b/b1.txt
-// /b/c/c1.txt
-// /symlinks/normal.txt
-// /symlinks/to_normal.txt -> /symlinks/normal.txt
-// /symlinks/recursive -> /symlinks
-func createTestDirs(ctx context.Context, t *testing.T, m *fs.MountNamespace) error {
- r := m.Root()
- defer r.DecRef()
-
- if err := r.CreateDirectory(ctx, r, "a", fs.FilePermsFromMode(0777)); err != nil {
- return err
- }
-
- a, err := r.Walk(ctx, r, "a")
- if err != nil {
- return err
- }
- defer a.DecRef()
-
- a1, err := a.Create(ctx, r, "a1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666))
- if err != nil {
- return err
- }
- a1.DecRef()
-
- a2, err := a.Create(ctx, r, "a2.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666))
- if err != nil {
- return err
- }
- a2.DecRef()
-
- if err := r.CreateDirectory(ctx, r, "b", fs.FilePermsFromMode(0777)); err != nil {
- return err
- }
-
- b, err := r.Walk(ctx, r, "b")
- if err != nil {
- return err
- }
- defer b.DecRef()
-
- b1, err := b.Create(ctx, r, "b1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666))
- if err != nil {
- return err
- }
- b1.DecRef()
-
- if err := b.CreateDirectory(ctx, r, "c", fs.FilePermsFromMode(0777)); err != nil {
- return err
- }
-
- c, err := b.Walk(ctx, r, "c")
- if err != nil {
- return err
- }
- defer c.DecRef()
-
- c1, err := c.Create(ctx, r, "c1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666))
- if err != nil {
- return err
- }
- c1.DecRef()
-
- if err := r.CreateDirectory(ctx, r, "symlinks", fs.FilePermsFromMode(0777)); err != nil {
- return err
- }
-
- symlinks, err := r.Walk(ctx, r, "symlinks")
- if err != nil {
- return err
- }
- defer symlinks.DecRef()
-
- normal, err := symlinks.Create(ctx, r, "normal.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666))
- if err != nil {
- return err
- }
- normal.DecRef()
-
- if err := symlinks.CreateLink(ctx, r, "/symlinks/normal.txt", "to_normal.txt"); err != nil {
- return err
- }
-
- return symlinks.CreateLink(ctx, r, "/symlinks", "recursive")
-}
-
-// allPaths returns a slice of all paths of entries visible in the rootfs.
-func allPaths(ctx context.Context, t *testing.T, m *fs.MountNamespace, base string) ([]string, error) {
- var paths []string
- root := m.Root()
- defer root.DecRef()
-
- maxTraversals := uint(1)
- d, err := m.FindLink(ctx, root, nil, base, &maxTraversals)
- if err != nil {
- t.Logf("FindLink failed for %q", base)
- return paths, err
- }
- defer d.DecRef()
-
- if fs.IsDir(d.Inode.StableAttr) {
- dir, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true})
- if err != nil {
- return nil, fmt.Errorf("failed to open directory %q: %v", base, err)
- }
- iter, ok := dir.FileOperations.(fs.DirIterator)
- if !ok {
- return nil, fmt.Errorf("cannot directly iterate on host directory %q", base)
- }
- dirCtx := &fs.DirCtx{
- Serializer: noopDentrySerializer{},
- }
- if _, err := fs.DirentReaddir(ctx, d, iter, root, dirCtx, 0); err != nil {
- return nil, err
- }
- for name := range dirCtx.DentAttrs() {
- if name == "." || name == ".." {
- continue
- }
-
- fullName := path.Join(base, name)
- paths = append(paths, fullName)
-
- // Recurse.
- subpaths, err := allPaths(ctx, t, m, fullName)
- if err != nil {
- return paths, err
- }
- paths = append(paths, subpaths...)
- }
- }
-
- return paths, nil
-}
-
-type noopDentrySerializer struct{}
-
-func (noopDentrySerializer) CopyOut(string, fs.DentAttr) error {
- return nil
-}
-func (noopDentrySerializer) Written() int {
- return 4096
-}
-
-// pathsEqual returns true if the two string slices contain the same entries.
-func pathsEqual(got, want []string) bool {
- sort.Strings(got)
- sort.Strings(want)
-
- if len(got) != len(want) {
- return false
- }
-
- for i := range got {
- if got[i] != want[i] {
- return false
- }
- }
-
- return true
-}
-
-func TestWhitelist(t *testing.T) {
- for _, test := range []struct {
- // description of the test.
- desc string
- // paths are the paths to whitelist
- paths []string
- // want are all of the directory entries that should be
- // visible (nothing beyond this set should be visible).
- want []string
- }{
- {
- desc: "root",
- paths: []string{"/"},
- want: []string{"/a", "/a/a1.txt", "/a/a2.txt", "/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt", "/symlinks", "/symlinks/normal.txt", "/symlinks/to_normal.txt", "/symlinks/recursive"},
- },
- {
- desc: "top-level directories",
- paths: []string{"/a", "/b"},
- want: []string{"/a", "/a/a1.txt", "/a/a2.txt", "/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"},
- },
- {
- desc: "nested directories (1/2)",
- paths: []string{"/b", "/b/c"},
- want: []string{"/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"},
- },
- {
- desc: "nested directories (2/2)",
- paths: []string{"/b/c", "/b"},
- want: []string{"/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"},
- },
- {
- desc: "single file",
- paths: []string{"/b/c/c1.txt"},
- want: []string{"/b", "/b/c", "/b/c/c1.txt"},
- },
- {
- desc: "single file and directory",
- paths: []string{"/a/a1.txt", "/b/c"},
- want: []string{"/a", "/a/a1.txt", "/b", "/b/c", "/b/c/c1.txt"},
- },
- {
- desc: "symlink",
- paths: []string{"/symlinks/to_normal.txt"},
- want: []string{"/symlinks", "/symlinks/normal.txt", "/symlinks/to_normal.txt"},
- },
- {
- desc: "recursive symlink",
- paths: []string{"/symlinks/recursive/normal.txt"},
- want: []string{"/symlinks", "/symlinks/normal.txt", "/symlinks/recursive"},
- },
- } {
- t.Run(test.desc, func(t *testing.T) {
- m, p, err := newTestMountNamespace(t)
- if err != nil {
- t.Errorf("Failed to create MountNamespace: %v", err)
- }
- defer os.RemoveAll(p)
-
- ctx := withRoot(contexttest.RootContext(t), m.Root())
- if err := createTestDirs(ctx, t, m); err != nil {
- t.Errorf("Failed to create test dirs: %v", err)
- }
-
- if err := installWhitelist(ctx, m, test.paths); err != nil {
- t.Errorf("installWhitelist(%v) err got %v want nil", test.paths, err)
- }
-
- got, err := allPaths(ctx, t, m, "/")
- if err != nil {
- t.Fatalf("Failed to lookup paths (whitelisted: %v): %v", test.paths, err)
- }
-
- if !pathsEqual(got, test.want) {
- t.Errorf("For paths %v got %v want %v", test.paths, got, test.want)
- }
- })
- }
-}
-
-func TestRootPath(t *testing.T) {
- // Create a temp dir, which will be the root of our mounted fs.
- rootPath, err := ioutil.TempDir(os.TempDir(), "root")
- if err != nil {
- t.Fatalf("TempDir failed: %v", err)
- }
- defer os.RemoveAll(rootPath)
-
- // Create two files inside the new root, one which will be whitelisted
- // and one not.
- whitelisted, err := ioutil.TempFile(rootPath, "white")
- if err != nil {
- t.Fatalf("TempFile failed: %v", err)
- }
- if _, err := ioutil.TempFile(rootPath, "black"); err != nil {
- t.Fatalf("TempFile failed: %v", err)
- }
-
- // Create a mount with a root path and single whitelisted file.
- hostFS := &Filesystem{}
- ctx := contexttest.Context(t)
- data := fmt.Sprintf("%s=%s,%s=%s", rootPathKey, rootPath, whitelistKey, whitelisted.Name())
- inode, err := hostFS.Mount(ctx, "", fs.MountSourceFlags{}, data, nil)
- if err != nil {
- t.Fatalf("Mount failed: %v", err)
- }
- mm, err := fs.NewMountNamespace(ctx, inode)
- if err != nil {
- t.Fatalf("NewMountNamespace failed: %v", err)
- }
- if err := hostFS.InstallWhitelist(ctx, mm); err != nil {
- t.Fatalf("InstallWhitelist failed: %v", err)
- }
-
- // Get the contents of the root directory.
- rootDir := mm.Root()
- rctx := withRoot(ctx, rootDir)
- f, err := rootDir.Inode.GetFile(rctx, rootDir, fs.FileFlags{})
- if err != nil {
- t.Fatalf("GetFile failed: %v", err)
- }
- c := &fs.CollectEntriesSerializer{}
- if err := f.Readdir(rctx, c); err != nil {
- t.Fatalf("Readdir failed: %v", err)
- }
-
- // We should have only our whitelisted file, plus the dots.
- want := []string{path.Base(whitelisted.Name()), ".", ".."}
- got := c.Order
- sort.Strings(want)
- sort.Strings(got)
- if !reflect.DeepEqual(got, want) {
- t.Errorf("Readdir got %v, wanted %v", got, want)
- }
-}
-
-type rootContext struct {
- context.Context
- root *fs.Dirent
-}
-
-// withRoot returns a copy of ctx with the given root.
-func withRoot(ctx context.Context, root *fs.Dirent) context.Context {
- return &rootContext{
- Context: ctx,
- root: root,
- }
-}
-
-// Value implements Context.Value.
-func (rc rootContext) Value(key interface{}) interface{} {
- switch key {
- case fs.CtxRoot:
- rc.root.IncRef()
- return rc.root
- default:
- return rc.Context.Value(key)
- }
-}
diff --git a/pkg/sentry/fs/host/host.go b/pkg/sentry/fs/host/host.go
new file mode 100644
index 000000000..081ba1dd8
--- /dev/null
+++ b/pkg/sentry/fs/host/host.go
@@ -0,0 +1,59 @@
+// Copyright 2020 The gVisor Authors.
+//
+// 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 host supports file descriptors imported directly.
+package host
+
+import (
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/sentry/fs"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+// filesystem is a host filesystem.
+//
+// +stateify savable
+type filesystem struct{}
+
+func init() {
+ fs.RegisterFilesystem(&filesystem{})
+}
+
+// FilesystemName is the name under which the filesystem is registered.
+const FilesystemName = "host"
+
+// Name is the name of the filesystem.
+func (*filesystem) Name() string {
+ return FilesystemName
+}
+
+// Mount returns an error. Mounting hostfs is not allowed.
+func (*filesystem) Mount(ctx context.Context, device string, flags fs.MountSourceFlags, data string, dataObj interface{}) (*fs.Inode, error) {
+ return nil, syserror.EPERM
+}
+
+// AllowUserMount prohibits users from using mount(2) with this file system.
+func (*filesystem) AllowUserMount() bool {
+ return false
+}
+
+// AllowUserList prohibits this filesystem to be listed in /proc/filesystems.
+func (*filesystem) AllowUserList() bool {
+ return false
+}
+
+// Flags returns that there is nothing special about this file system.
+func (*filesystem) Flags() fs.FilesystemFlags {
+ return 0
+}
diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go
index 6fa39caab..1da3c0a17 100644
--- a/pkg/sentry/fs/host/inode.go
+++ b/pkg/sentry/fs/host/inode.go
@@ -17,12 +17,10 @@ package host
import (
"syscall"
- "gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fd"
"gvisor.dev/gvisor/pkg/safemem"
"gvisor.dev/gvisor/pkg/secio"
- "gvisor.dev/gvisor/pkg/sentry/device"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
"gvisor.dev/gvisor/pkg/sentry/memmap"
@@ -69,9 +67,6 @@ type inodeOperations struct {
//
// +stateify savable
type inodeFileState struct {
- // Common file system state.
- mops *superOperations `state:"wait"`
-
// descriptor is the backing host FD.
descriptor *descriptor `state:"wait"`
@@ -160,7 +155,7 @@ func (i *inodeFileState) unstableAttr(ctx context.Context) (fs.UnstableAttr, err
if err := syscall.Fstat(i.FD(), &s); err != nil {
return fs.UnstableAttr{}, err
}
- return unstableAttr(i.mops, &s), nil
+ return unstableAttr(&s), nil
}
// Allocate implements fsutil.CachedFileObject.Allocate.
@@ -172,7 +167,7 @@ func (i *inodeFileState) Allocate(_ context.Context, offset, length int64) error
var _ fs.InodeOperations = (*inodeOperations)(nil)
// newInode returns a new fs.Inode backed by the host FD.
-func newInode(ctx context.Context, msrc *fs.MountSource, fd int, saveable bool, donated bool) (*fs.Inode, error) {
+func newInode(ctx context.Context, msrc *fs.MountSource, fd int, saveable bool) (*fs.Inode, error) {
// Retrieve metadata.
var s syscall.Stat_t
err := syscall.Fstat(fd, &s)
@@ -181,24 +176,17 @@ func newInode(ctx context.Context, msrc *fs.MountSource, fd int, saveable bool,
}
fileState := &inodeFileState{
- mops: msrc.MountSourceOperations.(*superOperations),
sattr: stableAttr(&s),
}
// Initialize the wrapped host file descriptor.
- fileState.descriptor, err = newDescriptor(
- fd,
- donated,
- saveable,
- wouldBlock(&s),
- &fileState.queue,
- )
+ fileState.descriptor, err = newDescriptor(fd, saveable, wouldBlock(&s), &fileState.queue)
if err != nil {
return nil, err
}
// Build the fs.InodeOperations.
- uattr := unstableAttr(msrc.MountSourceOperations.(*superOperations), &s)
+ uattr := unstableAttr(&s)
iops := &inodeOperations{
fileState: fileState,
cachingInodeOps: fsutil.NewCachingInodeOperations(ctx, fileState, uattr, fsutil.CachingInodeOperationsOptions{
@@ -232,54 +220,23 @@ func (i *inodeOperations) Release(context.Context) {
// Lookup implements fs.InodeOperations.Lookup.
func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dirent, error) {
- // Get a new FD relative to i at name.
- fd, err := open(i, name)
- if err != nil {
- if err == syserror.ENOENT {
- return nil, syserror.ENOENT
- }
- return nil, err
- }
-
- inode, err := newInode(ctx, dir.MountSource, fd, false /* saveable */, false /* donated */)
- if err != nil {
- return nil, err
- }
-
- // Return the fs.Dirent.
- return fs.NewDirent(ctx, inode, name), nil
+ return nil, syserror.ENOENT
}
// Create implements fs.InodeOperations.Create.
func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.FileFlags, perm fs.FilePermissions) (*fs.File, error) {
- // Create a file relative to i at name.
- //
- // N.B. We always open this file O_RDWR regardless of flags because a
- // future GetFile might want more access. Open allows this regardless
- // of perm.
- fd, err := openAt(i, name, syscall.O_RDWR|syscall.O_CREAT|syscall.O_EXCL, perm.LinuxMode())
- if err != nil {
- return nil, err
- }
-
- inode, err := newInode(ctx, dir.MountSource, fd, false /* saveable */, false /* donated */)
- if err != nil {
- return nil, err
- }
+ return nil, syserror.EPERM
- d := fs.NewDirent(ctx, inode, name)
- defer d.DecRef()
- return inode.GetFile(ctx, d, flags)
}
// CreateDirectory implements fs.InodeOperations.CreateDirectory.
func (i *inodeOperations) CreateDirectory(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions) error {
- return syscall.Mkdirat(i.fileState.FD(), name, uint32(perm.LinuxMode()))
+ return syserror.EPERM
}
// CreateLink implements fs.InodeOperations.CreateLink.
func (i *inodeOperations) CreateLink(ctx context.Context, dir *fs.Inode, oldname string, newname string) error {
- return createLink(i.fileState.FD(), oldname, newname)
+ return syserror.EPERM
}
// CreateHardLink implements fs.InodeOperations.CreateHardLink.
@@ -294,25 +251,17 @@ func (*inodeOperations) CreateFifo(context.Context, *fs.Inode, string, fs.FilePe
// Remove implements fs.InodeOperations.Remove.
func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string) error {
- return unlinkAt(i.fileState.FD(), name, false /* dir */)
+ return syserror.EPERM
}
// RemoveDirectory implements fs.InodeOperations.RemoveDirectory.
func (i *inodeOperations) RemoveDirectory(ctx context.Context, dir *fs.Inode, name string) error {
- return unlinkAt(i.fileState.FD(), name, true /* dir */)
+ return syserror.EPERM
}
// Rename implements fs.InodeOperations.Rename.
func (i *inodeOperations) Rename(ctx context.Context, inode *fs.Inode, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string, replacement bool) error {
- op, ok := oldParent.InodeOperations.(*inodeOperations)
- if !ok {
- return syscall.EXDEV
- }
- np, ok := newParent.InodeOperations.(*inodeOperations)
- if !ok {
- return syscall.EXDEV
- }
- return syscall.Renameat(op.fileState.FD(), oldName, np.fileState.FD(), newName)
+ return syserror.EPERM
}
// Bind implements fs.InodeOperations.Bind.
@@ -461,69 +410,7 @@ func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {}
// readdirAll returns all of the directory entries in i.
func (i *inodeOperations) readdirAll(d *dirInfo) (map[string]fs.DentAttr, error) {
- i.readdirMu.Lock()
- defer i.readdirMu.Unlock()
-
- fd := i.fileState.FD()
-
- // syscall.ReadDirent will use getdents, which will seek the file past
- // the last directory entry. To read the directory entries a second
- // time, we need to seek back to the beginning.
- if _, err := syscall.Seek(fd, 0, 0); err != nil {
- if err == syscall.ESPIPE {
- // All directories should be seekable. If this file
- // isn't seekable, it is not a directory and we should
- // return that more sane error.
- err = syscall.ENOTDIR
- }
- return nil, err
- }
-
- names := make([]string, 0, 100)
- for {
- // Refill the buffer if necessary
- if d.bufp >= d.nbuf {
- d.bufp = 0
- // ReadDirent will just do a sys_getdents64 to the kernel.
- n, err := syscall.ReadDirent(fd, d.buf)
- if err != nil {
- return nil, err
- }
- if n == 0 {
- break // EOF
- }
- d.nbuf = n
- }
-
- var nb int
- // Parse the dirent buffer we just get and return the directory names along
- // with the number of bytes consumed in the buffer.
- nb, _, names = syscall.ParseDirent(d.buf[d.bufp:d.nbuf], -1, names)
- d.bufp += nb
- }
-
- entries := make(map[string]fs.DentAttr)
- for _, filename := range names {
- // Lookup the type and host device and inode.
- stat, lerr := fstatat(fd, filename, linux.AT_SYMLINK_NOFOLLOW)
- if lerr == syscall.ENOENT {
- // File disappeared between readdir and lstat.
- // Just treat it as if it didn't exist.
- continue
- }
-
- // There was a serious problem, we should probably report it.
- if lerr != nil {
- return nil, lerr
- }
-
- entries[filename] = fs.DentAttr{
- Type: nodeType(&stat),
- InodeID: hostFileDevice.Map(device.MultiDeviceKey{
- Device: stat.Dev,
- Inode: stat.Ino,
- }),
- }
- }
- return entries, nil
+ // We only support non-directory file descriptors that have been
+ // imported, so just claim that this isn't a directory, even if it is.
+ return nil, syscall.ENOTDIR
}
diff --git a/pkg/sentry/fs/host/inode_state.go b/pkg/sentry/fs/host/inode_state.go
index 299e0e0b0..1adbd4562 100644
--- a/pkg/sentry/fs/host/inode_state.go
+++ b/pkg/sentry/fs/host/inode_state.go
@@ -18,29 +18,14 @@ import (
"fmt"
"syscall"
- "gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/device"
"gvisor.dev/gvisor/pkg/sentry/fs"
)
-// beforeSave is invoked by stateify.
-func (i *inodeFileState) beforeSave() {
- if !i.queue.IsEmpty() {
- panic("event queue must be empty")
- }
- if !i.descriptor.donated && i.sattr.Type == fs.RegularFile {
- uattr, err := i.unstableAttr(context.Background())
- if err != nil {
- panic(fs.ErrSaveRejection{fmt.Errorf("failed to get unstable atttribute of %s: %v", i.mops.inodeMappings[i.sattr.InodeID], err)})
- }
- i.savedUAttr = &uattr
- }
-}
-
// afterLoad is invoked by stateify.
func (i *inodeFileState) afterLoad() {
// Initialize the descriptor value.
- if err := i.descriptor.initAfterLoad(i.mops, i.sattr.InodeID, &i.queue); err != nil {
+ if err := i.descriptor.initAfterLoad(i.sattr.InodeID, &i.queue); err != nil {
panic(fmt.Sprintf("failed to load value of descriptor: %v", err))
}
@@ -61,19 +46,4 @@ func (i *inodeFileState) afterLoad() {
// change across save and restore, error out.
panic(fs.ErrCorruption{fmt.Errorf("host %s conflict in host device mappings: %s", key, hostFileDevice)})
}
-
- if !i.descriptor.donated && i.sattr.Type == fs.RegularFile {
- env, ok := fs.CurrentRestoreEnvironment()
- if !ok {
- panic("missing restore environment")
- }
- uattr := unstableAttr(i.mops, &s)
- if env.ValidateFileSize && uattr.Size != i.savedUAttr.Size {
- panic(fs.ErrCorruption{fmt.Errorf("file size has changed for %s: previously %d, now %d", i.mops.inodeMappings[i.sattr.InodeID], i.savedUAttr.Size, uattr.Size)})
- }
- if env.ValidateFileTimestamp && uattr.ModificationTime != i.savedUAttr.ModificationTime {
- panic(fs.ErrCorruption{fmt.Errorf("file modification time has changed for %s: previously %v, now %v", i.mops.inodeMappings[i.sattr.InodeID], i.savedUAttr.ModificationTime, uattr.ModificationTime)})
- }
- i.savedUAttr = nil
- }
}
diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go
index 7221bc825..4c374681c 100644
--- a/pkg/sentry/fs/host/inode_test.go
+++ b/pkg/sentry/fs/host/inode_test.go
@@ -15,9 +15,6 @@
package host
import (
- "io/ioutil"
- "os"
- "path"
"syscall"
"testing"
@@ -25,69 +22,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/fs"
)
-// TestMultipleReaddir verifies that multiple Readdir calls return the same
-// thing if they use different dir contexts.
-func TestMultipleReaddir(t *testing.T) {
- p, err := ioutil.TempDir("", "readdir")
- if err != nil {
- t.Fatalf("Failed to create test dir: %v", err)
- }
- defer os.RemoveAll(p)
-
- f, err := os.Create(path.Join(p, "a.txt"))
- if err != nil {
- t.Fatalf("Failed to create a.txt: %v", err)
- }
- f.Close()
-
- f, err = os.Create(path.Join(p, "b.txt"))
- if err != nil {
- t.Fatalf("Failed to create b.txt: %v", err)
- }
- f.Close()
-
- fd, err := open(nil, p)
- if err != nil {
- t.Fatalf("Failed to open %q: %v", p, err)
- }
- ctx := contexttest.Context(t)
- n, err := newInode(ctx, newMountSource(ctx, p, fs.RootOwner, &Filesystem{}, fs.MountSourceFlags{}, false), fd, false, false)
- if err != nil {
- t.Fatalf("Failed to create inode: %v", err)
- }
-
- dirent := fs.NewDirent(ctx, n, "readdir")
- openFile, err := n.GetFile(ctx, dirent, fs.FileFlags{Read: true})
- if err != nil {
- t.Fatalf("Failed to get file: %v", err)
- }
- defer openFile.DecRef()
-
- c1 := &fs.DirCtx{DirCursor: new(string)}
- if _, err := openFile.FileOperations.(*fileOperations).IterateDir(ctx, dirent, c1, 0); err != nil {
- t.Fatalf("First Readdir failed: %v", err)
- }
-
- c2 := &fs.DirCtx{DirCursor: new(string)}
- if _, err := openFile.FileOperations.(*fileOperations).IterateDir(ctx, dirent, c2, 0); err != nil {
- t.Errorf("Second Readdir failed: %v", err)
- }
-
- if _, ok := c1.DentAttrs()["a.txt"]; !ok {
- t.Errorf("want a.txt in first Readdir, got %v", c1.DentAttrs())
- }
- if _, ok := c1.DentAttrs()["b.txt"]; !ok {
- t.Errorf("want b.txt in first Readdir, got %v", c1.DentAttrs())
- }
-
- if _, ok := c2.DentAttrs()["a.txt"]; !ok {
- t.Errorf("want a.txt in second Readdir, got %v", c2.DentAttrs())
- }
- if _, ok := c2.DentAttrs()["b.txt"]; !ok {
- t.Errorf("want b.txt in second Readdir, got %v", c2.DentAttrs())
- }
-}
-
// TestCloseFD verifies fds will be closed.
func TestCloseFD(t *testing.T) {
var p [2]int
diff --git a/pkg/sentry/fs/host/util.go b/pkg/sentry/fs/host/util.go
index e37e687c6..388108fdf 100644
--- a/pkg/sentry/fs/host/util.go
+++ b/pkg/sentry/fs/host/util.go
@@ -16,7 +16,6 @@ package host
import (
"os"
- "path"
"syscall"
"gvisor.dev/gvisor/pkg/abi/linux"
@@ -24,49 +23,10 @@ import (
"gvisor.dev/gvisor/pkg/sentry/device"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
- ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/time"
"gvisor.dev/gvisor/pkg/syserror"
)
-func open(parent *inodeOperations, name string) (int, error) {
- if parent == nil && !path.IsAbs(name) {
- return -1, syserror.EINVAL
- }
- name = path.Clean(name)
-
- // Don't follow through symlinks.
- flags := syscall.O_NOFOLLOW
-
- if fd, err := openAt(parent, name, flags|syscall.O_RDWR, 0); err == nil {
- return fd, nil
- }
- // Retry as read-only.
- if fd, err := openAt(parent, name, flags|syscall.O_RDONLY, 0); err == nil {
- return fd, nil
- }
-
- // Retry as write-only.
- if fd, err := openAt(parent, name, flags|syscall.O_WRONLY, 0); err == nil {
- return fd, nil
- }
-
- // Retry as a symlink, by including O_PATH as an option.
- fd, err := openAt(parent, name, linux.O_PATH|flags, 0)
- if err == nil {
- return fd, nil
- }
-
- // Everything failed.
- return -1, err
-}
-
-func openAt(parent *inodeOperations, name string, flags int, perm linux.FileMode) (int, error) {
- if parent == nil {
- return syscall.Open(name, flags, uint32(perm))
- }
- return syscall.Openat(parent.fileState.FD(), name, flags, uint32(perm))
-}
-
func nodeType(s *syscall.Stat_t) fs.InodeType {
switch x := (s.Mode & syscall.S_IFMT); x {
case syscall.S_IFLNK:
@@ -107,54 +67,22 @@ func stableAttr(s *syscall.Stat_t) fs.StableAttr {
}
}
-func owner(mo *superOperations, s *syscall.Stat_t) fs.FileOwner {
- // User requested no translation, just return actual owner.
- if mo.dontTranslateOwnership {
- return fs.FileOwner{auth.KUID(s.Uid), auth.KGID(s.Gid)}
+func owner(s *syscall.Stat_t) fs.FileOwner {
+ return fs.FileOwner{
+ UID: auth.KUID(s.Uid),
+ GID: auth.KGID(s.Gid),
}
-
- // Show only IDs relevant to the sandboxed task. I.e. if we not own the
- // file, no sandboxed task can own the file. In that case, we
- // use OverflowID for UID, implying that the IDs are not mapped in the
- // "root" user namespace.
- //
- // E.g.
- // sandbox's host EUID/EGID is 1/1.
- // some_dir's host UID/GID is 2/1.
- // Task that mounted this fs has virtualized EUID/EGID 5/5.
- //
- // If you executed `ls -n` in the sandboxed task, it would show:
- // drwxwrxwrx [...] 65534 5 [...] some_dir
-
- // Files are owned by OverflowID by default.
- owner := fs.FileOwner{auth.KUID(auth.OverflowUID), auth.KGID(auth.OverflowGID)}
-
- // If we own file on host, let mounting task's initial EUID own
- // the file.
- if s.Uid == hostUID {
- owner.UID = mo.mounter.UID
- }
-
- // If our group matches file's group, make file's group match
- // the mounting task's initial EGID.
- for _, gid := range hostGIDs {
- if s.Gid == gid {
- owner.GID = mo.mounter.GID
- break
- }
- }
- return owner
}
-func unstableAttr(mo *superOperations, s *syscall.Stat_t) fs.UnstableAttr {
+func unstableAttr(s *syscall.Stat_t) fs.UnstableAttr {
return fs.UnstableAttr{
Size: s.Size,
Usage: s.Blocks * 512,
Perms: fs.FilePermsFromMode(linux.FileMode(s.Mode)),
- Owner: owner(mo, s),
- AccessTime: ktime.FromUnix(s.Atim.Sec, s.Atim.Nsec),
- ModificationTime: ktime.FromUnix(s.Mtim.Sec, s.Mtim.Nsec),
- StatusChangeTime: ktime.FromUnix(s.Ctim.Sec, s.Ctim.Nsec),
+ Owner: owner(s),
+ AccessTime: time.FromUnix(s.Atim.Sec, s.Atim.Nsec),
+ ModificationTime: time.FromUnix(s.Mtim.Sec, s.Mtim.Nsec),
+ StatusChangeTime: time.FromUnix(s.Ctim.Sec, s.Ctim.Nsec),
Links: uint64(s.Nlink),
}
}
@@ -165,6 +93,8 @@ type dirInfo struct {
bufp int // location of next record in buf.
}
+// LINT.IfChange
+
// isBlockError unwraps os errors and checks if they are caused by EAGAIN or
// EWOULDBLOCK. This is so they can be transformed into syserror.ErrWouldBlock.
func isBlockError(err error) bool {
@@ -177,6 +107,8 @@ func isBlockError(err error) bool {
return false
}
+// LINT.ThenChange(../../fsimpl/host/util.go)
+
func hostEffectiveKIDs() (uint32, []uint32, error) {
gids, err := os.Getgroups()
if err != nil {
diff --git a/pkg/sentry/fs/host/util_unsafe.go b/pkg/sentry/fs/host/util_unsafe.go
index 3ab36b088..23bd35d64 100644
--- a/pkg/sentry/fs/host/util_unsafe.go
+++ b/pkg/sentry/fs/host/util_unsafe.go
@@ -26,26 +26,6 @@ import (
// NulByte is a single NUL byte. It is passed to readlinkat as an empty string.
var NulByte byte = '\x00'
-func createLink(fd int, name string, linkName string) error {
- namePtr, err := syscall.BytePtrFromString(name)
- if err != nil {
- return err
- }
- linkNamePtr, err := syscall.BytePtrFromString(linkName)
- if err != nil {
- return err
- }
- _, _, errno := syscall.Syscall(
- syscall.SYS_SYMLINKAT,
- uintptr(unsafe.Pointer(namePtr)),
- uintptr(fd),
- uintptr(unsafe.Pointer(linkNamePtr)))
- if errno != 0 {
- return errno
- }
- return nil
-}
-
func readLink(fd int) (string, error) {
// Buffer sizing copied from os.Readlink.
for l := 128; ; l *= 2 {
@@ -66,27 +46,6 @@ func readLink(fd int) (string, error) {
}
}
-func unlinkAt(fd int, name string, dir bool) error {
- namePtr, err := syscall.BytePtrFromString(name)
- if err != nil {
- return err
- }
- var flags uintptr
- if dir {
- flags = linux.AT_REMOVEDIR
- }
- _, _, errno := syscall.Syscall(
- syscall.SYS_UNLINKAT,
- uintptr(fd),
- uintptr(unsafe.Pointer(namePtr)),
- flags,
- )
- if errno != 0 {
- return errno
- }
- return nil
-}
-
func timespecFromTimestamp(t ktime.Time, omit, setSysTime bool) syscall.Timespec {
if omit {
return syscall.Timespec{0, linux.UTIME_OMIT}
diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go
index 928c90aa0..e3a715c1f 100644
--- a/pkg/sentry/fs/inotify.go
+++ b/pkg/sentry/fs/inotify.go
@@ -143,7 +143,10 @@ func (i *Inotify) Read(ctx context.Context, _ *File, dst usermem.IOSequence, _ i
}
var writeLen int64
- for event := i.events.Front(); event != nil; event = event.Next() {
+ for it := i.events.Front(); it != nil; {
+ event := it
+ it = it.Next()
+
// Does the buffer have enough remaining space to hold the event we're
// about to write out?
if dst.NumBytes() < int64(event.sizeOf()) {
diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go
index 95d5817ff..bd18177d4 100644
--- a/pkg/sentry/fs/proc/net.go
+++ b/pkg/sentry/fs/proc/net.go
@@ -40,47 +40,48 @@ import (
// LINT.IfChange
-// newNet creates a new proc net entry.
-func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode {
+// newNetDir creates a new proc net entry.
+func newNetDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ k := t.Kernel()
+
var contents map[string]*fs.Inode
- // TODO(gvisor.dev/issue/1833): Support for using the network stack in the
- // network namespace of the calling process. We should make this per-process,
- // a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net.
- if s := p.k.RootNetworkNamespace().Stack(); s != nil {
+ if s := t.NetworkNamespace().Stack(); s != nil {
+ // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task
+ // network namespace.
contents = map[string]*fs.Inode{
- "dev": seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc),
- "snmp": seqfile.NewSeqFileInode(ctx, &netSnmp{s: s}, msrc),
+ "dev": seqfile.NewSeqFileInode(t, &netDev{s: s}, msrc),
+ "snmp": seqfile.NewSeqFileInode(t, &netSnmp{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\n")),
+ "arp": newStaticProcInode(t, msrc, []byte("IP address HW type Flags HW address Mask Device\n")),
- "netlink": newStaticProcInode(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n")),
- "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\n")),
- "packet": newStaticProcInode(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode\n")),
- "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\n")),
+ "netlink": newStaticProcInode(t, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n")),
+ "netstat": newStaticProcInode(t, 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\n")),
+ "packet": newStaticProcInode(t, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode\n")),
+ "protocols": newStaticProcInode(t, 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\n")),
// 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\n")),
- "route": seqfile.NewSeqFileInode(ctx, &netRoute{s: s}, msrc),
- "tcp": seqfile.NewSeqFileInode(ctx, &netTCP{k: k}, msrc),
- "udp": seqfile.NewSeqFileInode(ctx, &netUDP{k: k}, msrc),
- "unix": seqfile.NewSeqFileInode(ctx, &netUnix{k: k}, msrc),
+ "psched": newStaticProcInode(t, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))),
+ "ptype": newStaticProcInode(t, msrc, []byte("Type Device Function\n")),
+ "route": seqfile.NewSeqFileInode(t, &netRoute{s: s}, msrc),
+ "tcp": seqfile.NewSeqFileInode(t, &netTCP{k: k}, msrc),
+ "udp": seqfile.NewSeqFileInode(t, &netUDP{k: k}, msrc),
+ "unix": seqfile.NewSeqFileInode(t, &netUnix{k: k}, msrc),
}
if s.SupportsIPv6() {
- contents["if_inet6"] = seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc)
- contents["ipv6_route"] = newStaticProcInode(ctx, msrc, []byte(""))
- contents["tcp6"] = seqfile.NewSeqFileInode(ctx, &netTCP6{k: k}, msrc)
- contents["udp6"] = newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n"))
+ contents["if_inet6"] = seqfile.NewSeqFileInode(t, &ifinet6{s: s}, msrc)
+ contents["ipv6_route"] = newStaticProcInode(t, msrc, []byte(""))
+ contents["tcp6"] = seqfile.NewSeqFileInode(t, &netTCP6{k: k}, msrc)
+ contents["udp6"] = newStaticProcInode(t, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n"))
}
}
- d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555))
- return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil)
+ d := ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0555))
+ return newProcInode(t, d, msrc, fs.SpecialDirectory, t)
}
// ifinet6 implements seqfile.SeqSource for /proc/net/if_inet6.
@@ -837,4 +838,4 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se
return data, 0
}
-// LINT.ThenChange(../../fsimpl/proc/tasks_net.go)
+// LINT.ThenChange(../../fsimpl/proc/task_net.go)
diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go
index c8abb5052..c659224a7 100644
--- a/pkg/sentry/fs/proc/proc.go
+++ b/pkg/sentry/fs/proc/proc.go
@@ -70,6 +70,7 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string
"loadavg": seqfile.NewSeqFileInode(ctx, &loadavgData{}, msrc),
"meminfo": seqfile.NewSeqFileInode(ctx, &meminfoData{k}, msrc),
"mounts": newProcInode(ctx, ramfs.NewSymlink(ctx, fs.RootOwner, "self/mounts"), msrc, fs.Symlink, nil),
+ "net": newProcInode(ctx, ramfs.NewSymlink(ctx, fs.RootOwner, "self/net"), msrc, fs.Symlink, nil),
"self": newSelf(ctx, pidns, msrc),
"stat": seqfile.NewSeqFileInode(ctx, &statData{k}, msrc),
"thread-self": newThreadSelf(ctx, pidns, msrc),
@@ -86,7 +87,6 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string
}
// Add more contents that need proc to be initialized.
- p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc))
p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc))
return newProcInode(ctx, p, msrc, fs.SpecialDirectory, nil), nil
diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go
index 8ab8d8a02..d6c5dd2c1 100644
--- a/pkg/sentry/fs/proc/task.go
+++ b/pkg/sentry/fs/proc/task.go
@@ -72,24 +72,27 @@ var _ fs.InodeOperations = (*taskDir)(nil)
// newTaskDir creates a new proc task entry.
func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode {
contents := map[string]*fs.Inode{
- "auxv": newAuxvec(t, msrc),
- "cmdline": newExecArgInode(t, msrc, cmdlineExecArg),
- "comm": newComm(t, msrc),
- "environ": newExecArgInode(t, msrc, environExecArg),
- "exe": newExe(t, msrc),
- "fd": newFdDir(t, msrc),
- "fdinfo": newFdInfoDir(t, msrc),
- "gid_map": newGIDMap(t, msrc),
- "io": newIO(t, msrc, isThreadGroup),
- "maps": newMaps(t, msrc),
- "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
- "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
- "ns": newNamespaceDir(t, msrc),
- "smaps": newSmaps(t, msrc),
- "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns),
- "statm": newStatm(t, msrc),
- "status": newStatus(t, msrc, p.pidns),
- "uid_map": newUIDMap(t, msrc),
+ "auxv": newAuxvec(t, msrc),
+ "cmdline": newExecArgInode(t, msrc, cmdlineExecArg),
+ "comm": newComm(t, msrc),
+ "environ": newExecArgInode(t, msrc, environExecArg),
+ "exe": newExe(t, msrc),
+ "fd": newFdDir(t, msrc),
+ "fdinfo": newFdInfoDir(t, msrc),
+ "gid_map": newGIDMap(t, msrc),
+ "io": newIO(t, msrc, isThreadGroup),
+ "maps": newMaps(t, msrc),
+ "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
+ "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
+ "net": newNetDir(t, msrc),
+ "ns": newNamespaceDir(t, msrc),
+ "oom_score": newOOMScore(t, msrc),
+ "oom_score_adj": newOOMScoreAdj(t, msrc),
+ "smaps": newSmaps(t, msrc),
+ "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns),
+ "statm": newStatm(t, msrc),
+ "status": newStatus(t, msrc, p.pidns),
+ "uid_map": newUIDMap(t, msrc),
}
if isThreadGroup {
contents["task"] = p.newSubtasks(t, msrc)
@@ -796,4 +799,95 @@ func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc
return int64(n), err
}
+// newOOMScore returns a oom_score file. It is a stub that always returns 0.
+// TODO(gvisor.dev/issue/1967)
+func newOOMScore(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ return newStaticProcInode(t, msrc, []byte("0\n"))
+}
+
+// oomScoreAdj is a file containing the oom_score adjustment for a task.
+//
+// +stateify savable
+type oomScoreAdj struct {
+ fsutil.SimpleFileInode
+
+ t *kernel.Task
+}
+
+// +stateify savable
+type oomScoreAdjFile struct {
+ fsutil.FileGenericSeek `state:"nosave"`
+ fsutil.FileNoIoctl `state:"nosave"`
+ fsutil.FileNoMMap `state:"nosave"`
+ fsutil.FileNoSplice `state:"nosave"`
+ fsutil.FileNoopFlush `state:"nosave"`
+ fsutil.FileNoopFsync `state:"nosave"`
+ fsutil.FileNoopRelease `state:"nosave"`
+ fsutil.FileNotDirReaddir `state:"nosave"`
+ fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+ waiter.AlwaysReady `state:"nosave"`
+
+ t *kernel.Task
+}
+
+// newOOMScoreAdj returns a oom_score_adj file.
+func newOOMScoreAdj(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ i := &oomScoreAdj{
+ SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC),
+ t: t,
+ }
+ return newProcInode(t, i, msrc, fs.SpecialFile, t)
+}
+
+// Truncate implements fs.InodeOperations.Truncate. Truncate is called when
+// O_TRUNC is specified for any kind of existing Dirent but is not called via
+// (f)truncate for proc files.
+func (*oomScoreAdj) Truncate(context.Context, *fs.Inode, int64) error {
+ return nil
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+func (o *oomScoreAdj) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+ return fs.NewFile(ctx, dirent, flags, &oomScoreAdjFile{t: o.t}), nil
+}
+
+// Read implements fs.FileOperations.Read.
+func (f *oomScoreAdjFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) {
+ if f.t.ExitState() == kernel.TaskExitDead {
+ return 0, syserror.ESRCH
+ }
+ var buf bytes.Buffer
+ fmt.Fprintf(&buf, "%d\n", f.t.OOMScoreAdj())
+ if offset >= int64(buf.Len()) {
+ return 0, io.EOF
+ }
+ n, err := dst.CopyOut(ctx, buf.Bytes()[offset:])
+ return int64(n), err
+}
+
+// Write implements fs.FileOperations.Write.
+func (f *oomScoreAdjFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) {
+ if src.NumBytes() == 0 {
+ return 0, nil
+ }
+
+ // Limit input size so as not to impact performance if input size is large.
+ src = src.TakeFirst(usermem.PageSize - 1)
+
+ var v int32
+ n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts)
+ if err != nil {
+ return 0, err
+ }
+
+ if f.t.ExitState() == kernel.TaskExitDead {
+ return 0, syserror.ESRCH
+ }
+ if err := f.t.SetOOMScoreAdj(v); err != nil {
+ return 0, err
+ }
+
+ return n, nil
+}
+
// LINT.ThenChange(../../fsimpl/proc/task.go|../../fsimpl/proc/task_files.go)
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-28 08:23:21 +0000