1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
|
// 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 cmd
import (
"context"
"encoding/json"
"fmt"
"os"
"path/filepath"
"strings"
"github.com/google/subcommands"
specs "github.com/opencontainers/runtime-spec/specs-go"
"golang.org/x/sys/unix"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/p9"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/unet"
"gvisor.dev/gvisor/runsc/config"
"gvisor.dev/gvisor/runsc/flag"
"gvisor.dev/gvisor/runsc/fsgofer"
"gvisor.dev/gvisor/runsc/fsgofer/filter"
"gvisor.dev/gvisor/runsc/specutils"
)
var caps = []string{
"CAP_CHOWN",
"CAP_DAC_OVERRIDE",
"CAP_DAC_READ_SEARCH",
"CAP_FOWNER",
"CAP_FSETID",
"CAP_SYS_CHROOT",
}
// goferCaps is the minimal set of capabilities needed by the Gofer to operate
// on files.
var goferCaps = &specs.LinuxCapabilities{
Bounding: caps,
Effective: caps,
Permitted: caps,
}
// Gofer implements subcommands.Command for the "gofer" command, which starts a
// filesystem gofer. This command should not be called directly.
type Gofer struct {
bundleDir string
ioFDs intFlags
applyCaps bool
setUpRoot bool
specFD int
mountsFD int
}
// Name implements subcommands.Command.
func (*Gofer) Name() string {
return "gofer"
}
// Synopsis implements subcommands.Command.
func (*Gofer) Synopsis() string {
return "launch a gofer process that serves files over 9P protocol (internal use only)"
}
// Usage implements subcommands.Command.
func (*Gofer) Usage() string {
return `gofer [flags]`
}
// SetFlags implements subcommands.Command.
func (g *Gofer) SetFlags(f *flag.FlagSet) {
f.StringVar(&g.bundleDir, "bundle", "", "path to the root of the bundle directory, defaults to the current directory")
f.Var(&g.ioFDs, "io-fds", "list of FDs to connect 9P servers. They must follow this order: root first, then mounts as defined in the spec")
f.BoolVar(&g.applyCaps, "apply-caps", true, "if true, apply capabilities to restrict what the Gofer process can do")
f.BoolVar(&g.setUpRoot, "setup-root", true, "if true, set up an empty root for the process")
f.IntVar(&g.specFD, "spec-fd", -1, "required fd with the container spec")
f.IntVar(&g.mountsFD, "mounts-fd", -1, "mountsFD is the file descriptor to write list of mounts after they have been resolved (direct paths, no symlinks).")
}
// Execute implements subcommands.Command.
func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus {
if g.bundleDir == "" || len(g.ioFDs) < 1 || g.specFD < 0 {
f.Usage()
return subcommands.ExitUsageError
}
conf := args[0].(*config.Config)
specFile := os.NewFile(uintptr(g.specFD), "spec file")
defer specFile.Close()
spec, err := specutils.ReadSpecFromFile(g.bundleDir, specFile, conf)
if err != nil {
Fatalf("reading spec: %v", err)
}
if g.setUpRoot {
if err := setupRootFS(spec, conf); err != nil {
Fatalf("Error setting up root FS: %v", err)
}
}
if g.applyCaps {
// Disable caps when calling myself again.
// Note: minimal argument handling for the default case to keep it simple.
args := os.Args
args = append(args, "--apply-caps=false", "--setup-root=false")
Fatalf("setCapsAndCallSelf(%v, %v): %v", args, goferCaps, setCapsAndCallSelf(args, goferCaps))
panic("unreachable")
}
// Find what path is going to be served by this gofer.
root := spec.Root.Path
if !conf.TestOnlyAllowRunAsCurrentUserWithoutChroot {
root = "/root"
}
// Resolve mount points paths, then replace mounts from our spec and send the
// mount list over to the sandbox, so they are both in sync.
//
// Note that all mount points have been mounted in the proper location in
// setupRootFS().
cleanMounts, err := resolveMounts(conf, spec.Mounts, root)
if err != nil {
Fatalf("Failure to resolve mounts: %v", err)
}
spec.Mounts = cleanMounts
go func() {
if err := g.writeMounts(cleanMounts); err != nil {
panic(fmt.Sprintf("Failed to write mounts: %v", err))
}
}()
specutils.LogSpec(spec)
// fsgofer should run with a umask of 0, because we want to preserve file
// modes exactly as sent by the sandbox, which will have applied its own umask.
unix.Umask(0)
if err := fsgofer.OpenProcSelfFD(); err != nil {
Fatalf("failed to open /proc/self/fd: %v", err)
}
if err := unix.Chroot(root); err != nil {
Fatalf("failed to chroot to %q: %v", root, err)
}
if err := unix.Chdir("/"); err != nil {
Fatalf("changing working dir: %v", err)
}
log.Infof("Process chroot'd to %q", root)
// Start with root mount, then add any other additional mount as needed.
ats := make([]p9.Attacher, 0, len(spec.Mounts)+1)
ap, err := fsgofer.NewAttachPoint("/", fsgofer.Config{
ROMount: spec.Root.Readonly || conf.Overlay,
EnableVerityXattr: conf.Verity,
})
if err != nil {
Fatalf("creating attach point: %v", err)
}
ats = append(ats, ap)
log.Infof("Serving %q mapped to %q on FD %d (ro: %t)", "/", root, g.ioFDs[0], spec.Root.Readonly)
mountIdx := 1 // first one is the root
for _, m := range spec.Mounts {
if specutils.Is9PMount(m, conf.VFS2) {
cfg := fsgofer.Config{
ROMount: isReadonlyMount(m.Options) || conf.Overlay,
HostUDS: conf.FSGoferHostUDS,
EnableVerityXattr: conf.Verity,
}
ap, err := fsgofer.NewAttachPoint(m.Destination, cfg)
if err != nil {
Fatalf("creating attach point: %v", err)
}
ats = append(ats, ap)
if mountIdx >= len(g.ioFDs) {
Fatalf("no FD found for mount. Did you forget --io-fd? mount: %d, %v", len(g.ioFDs), m)
}
log.Infof("Serving %q mapped on FD %d (ro: %t)", m.Destination, g.ioFDs[mountIdx], cfg.ROMount)
mountIdx++
}
}
if mountIdx != len(g.ioFDs) {
Fatalf("too many FDs passed for mounts. mounts: %d, FDs: %d", mountIdx, len(g.ioFDs))
}
if conf.FSGoferHostUDS {
filter.InstallUDSFilters()
}
if conf.Verity {
filter.InstallXattrFilters()
}
if err := filter.Install(); err != nil {
Fatalf("installing seccomp filters: %v", err)
}
runServers(ats, g.ioFDs)
return subcommands.ExitSuccess
}
func runServers(ats []p9.Attacher, ioFDs []int) {
// Run the loops and wait for all to exit.
var wg sync.WaitGroup
for i, ioFD := range ioFDs {
wg.Add(1)
go func(ioFD int, at p9.Attacher) {
socket, err := unet.NewSocket(ioFD)
if err != nil {
Fatalf("creating server on FD %d: %v", ioFD, err)
}
s := p9.NewServer(at)
if err := s.Handle(socket); err != nil {
Fatalf("P9 server returned error. Gofer is shutting down. FD: %d, err: %v", ioFD, err)
}
wg.Done()
}(ioFD, ats[i])
}
wg.Wait()
log.Infof("All 9P servers exited.")
}
func (g *Gofer) writeMounts(mounts []specs.Mount) error {
bytes, err := json.Marshal(mounts)
if err != nil {
return err
}
f := os.NewFile(uintptr(g.mountsFD), "mounts file")
defer f.Close()
for written := 0; written < len(bytes); {
w, err := f.Write(bytes[written:])
if err != nil {
return err
}
written += w
}
return nil
}
func isReadonlyMount(opts []string) bool {
for _, o := range opts {
if o == "ro" {
return true
}
}
return false
}
func setupRootFS(spec *specs.Spec, conf *config.Config) error {
// Convert all shared mounts into slaves to be sure that nothing will be
// propagated outside of our namespace.
procPath := "/proc"
if err := specutils.SafeMount("", "/", "", unix.MS_SLAVE|unix.MS_REC, "", procPath); err != nil {
Fatalf("error converting mounts: %v", err)
}
root := spec.Root.Path
if !conf.TestOnlyAllowRunAsCurrentUserWithoutChroot {
// runsc can't be re-executed without /proc, so we create a tmpfs mount,
// mount ./proc and ./root there, then move this mount to the root and after
// setCapsAndCallSelf, runsc will chroot into /root.
//
// We need a directory to construct a new root and we know that
// runsc can't start without /proc, so we can use it for this.
flags := uintptr(unix.MS_NOSUID | unix.MS_NODEV | unix.MS_NOEXEC)
if err := specutils.SafeMount("runsc-root", "/proc", "tmpfs", flags, "", procPath); err != nil {
Fatalf("error mounting tmpfs: %v", err)
}
// Prepare tree structure for pivot_root(2).
if err := os.Mkdir("/proc/proc", 0755); err != nil {
Fatalf("error creating /proc/proc: %v", err)
}
if err := os.Mkdir("/proc/root", 0755); err != nil {
Fatalf("error creating /proc/root: %v", err)
}
// This cannot use SafeMount because there's no available procfs. But we
// know that /proc is an empty tmpfs mount, so this is safe.
if err := unix.Mount("runsc-proc", "/proc/proc", "proc", flags|unix.MS_RDONLY, ""); err != nil {
Fatalf("error mounting proc: %v", err)
}
root = "/proc/root"
procPath = "/proc/proc"
}
// Mount root path followed by submounts.
if err := specutils.SafeMount(spec.Root.Path, root, "bind", unix.MS_BIND|unix.MS_REC, "", procPath); err != nil {
return fmt.Errorf("mounting root on root (%q) err: %v", root, err)
}
flags := uint32(unix.MS_SLAVE | unix.MS_REC)
if spec.Linux != nil && spec.Linux.RootfsPropagation != "" {
flags = specutils.PropOptionsToFlags([]string{spec.Linux.RootfsPropagation})
}
if err := specutils.SafeMount("", root, "", uintptr(flags), "", procPath); err != nil {
return fmt.Errorf("mounting root (%q) with flags: %#x, err: %v", root, flags, err)
}
// Replace the current spec, with the clean spec with symlinks resolved.
if err := setupMounts(conf, spec.Mounts, root, procPath); err != nil {
Fatalf("error setting up FS: %v", err)
}
// Create working directory if needed.
if spec.Process.Cwd != "" {
dst, err := resolveSymlinks(root, spec.Process.Cwd)
if err != nil {
return fmt.Errorf("resolving symlinks to %q: %v", spec.Process.Cwd, err)
}
log.Infof("Create working directory %q if needed", spec.Process.Cwd)
if err := os.MkdirAll(dst, 0755); err != nil {
return fmt.Errorf("creating working directory %q: %v", spec.Process.Cwd, err)
}
}
// Check if root needs to be remounted as readonly.
if spec.Root.Readonly || conf.Overlay {
// If root is a mount point but not read-only, we can change mount options
// to make it read-only for extra safety.
log.Infof("Remounting root as readonly: %q", root)
flags := uintptr(unix.MS_BIND | unix.MS_REMOUNT | unix.MS_RDONLY | unix.MS_REC)
if err := specutils.SafeMount(root, root, "bind", flags, "", procPath); err != nil {
return fmt.Errorf("remounting root as read-only with source: %q, target: %q, flags: %#x, err: %v", root, root, flags, err)
}
}
if !conf.TestOnlyAllowRunAsCurrentUserWithoutChroot {
if err := pivotRoot("/proc"); err != nil {
Fatalf("failed to change the root file system: %v", err)
}
if err := os.Chdir("/"); err != nil {
Fatalf("failed to change working directory")
}
}
return nil
}
// setupMounts bind mounts all mounts specified in the spec in their correct
// location inside root. It will resolve relative paths and symlinks. It also
// creates directories as needed.
func setupMounts(conf *config.Config, mounts []specs.Mount, root, procPath string) error {
for _, m := range mounts {
if !specutils.Is9PMount(m, conf.VFS2) {
continue
}
dst, err := resolveSymlinks(root, m.Destination)
if err != nil {
return fmt.Errorf("resolving symlinks to %q: %v", m.Destination, err)
}
flags := specutils.OptionsToFlags(m.Options) | unix.MS_BIND
if conf.Overlay {
// Force mount read-only if writes are not going to be sent to it.
flags |= unix.MS_RDONLY
}
log.Infof("Mounting src: %q, dst: %q, flags: %#x", m.Source, dst, flags)
if err := specutils.Mount(m.Source, dst, m.Type, flags, procPath); err != nil {
return fmt.Errorf("mounting %+v: %v", m, err)
}
// Set propagation options that cannot be set together with other options.
flags = specutils.PropOptionsToFlags(m.Options)
if flags != 0 {
if err := specutils.SafeMount("", dst, "", uintptr(flags), "", procPath); err != nil {
return fmt.Errorf("mount dst: %q, flags: %#x, err: %v", dst, flags, err)
}
}
}
return nil
}
// resolveMounts resolved relative paths and symlinks to mount points.
//
// Note: mount points must already be in place for resolution to work.
// Otherwise, it may follow symlinks to locations that would be overwritten
// with another mount point and return the wrong location. In short, make sure
// setupMounts() has been called before.
func resolveMounts(conf *config.Config, mounts []specs.Mount, root string) ([]specs.Mount, error) {
cleanMounts := make([]specs.Mount, 0, len(mounts))
for _, m := range mounts {
if !specutils.Is9PMount(m, conf.VFS2) {
cleanMounts = append(cleanMounts, m)
continue
}
dst, err := resolveSymlinks(root, m.Destination)
if err != nil {
return nil, fmt.Errorf("resolving symlinks to %q: %v", m.Destination, err)
}
relDst, err := filepath.Rel(root, dst)
if err != nil {
panic(fmt.Sprintf("%q could not be made relative to %q: %v", dst, root, err))
}
opts, err := adjustMountOptions(conf, filepath.Join(root, relDst), m.Options)
if err != nil {
return nil, err
}
cpy := m
cpy.Destination = filepath.Join("/", relDst)
cpy.Options = opts
cleanMounts = append(cleanMounts, cpy)
}
return cleanMounts, nil
}
// ResolveSymlinks walks 'rel' having 'root' as the root directory. If there are
// symlinks, they are evaluated relative to 'root' to ensure the end result is
// the same as if the process was running inside the container.
func resolveSymlinks(root, rel string) (string, error) {
return resolveSymlinksImpl(root, root, rel, 255)
}
func resolveSymlinksImpl(root, base, rel string, followCount uint) (string, error) {
if followCount == 0 {
return "", fmt.Errorf("too many symlinks to follow, path: %q", filepath.Join(base, rel))
}
rel = filepath.Clean(rel)
for _, name := range strings.Split(rel, string(filepath.Separator)) {
if name == "" {
continue
}
// Note that Join() resolves things like ".." and returns a clean path.
path := filepath.Join(base, name)
if !strings.HasPrefix(path, root) {
// One cannot '..' their way out of root.
base = root
continue
}
fi, err := os.Lstat(path)
if err != nil {
if !os.IsNotExist(err) {
return "", err
}
// Not found means there is no symlink to check. Just keep walking dirs.
base = path
continue
}
if fi.Mode()&os.ModeSymlink != 0 {
link, err := os.Readlink(path)
if err != nil {
return "", err
}
if filepath.IsAbs(link) {
base = root
}
base, err = resolveSymlinksImpl(root, base, link, followCount-1)
if err != nil {
return "", err
}
continue
}
base = path
}
return base, nil
}
// adjustMountOptions adds 'overlayfs_stale_read' if mounting over overlayfs.
func adjustMountOptions(conf *config.Config, path string, opts []string) ([]string, error) {
rv := make([]string, len(opts))
copy(rv, opts)
statfs := unix.Statfs_t{}
if err := unix.Statfs(path, &statfs); err != nil {
return nil, err
}
if statfs.Type == unix.OVERLAYFS_SUPER_MAGIC {
rv = append(rv, "overlayfs_stale_read")
}
return rv, nil
}
|