git.infradead.org Git - nvme.git/blob

1 // SPDX-License-Identifier: GPL-2.0-only

2 /*

3 *

5 */

6

7 #include <uapi/linux/magic.h>

8 #include <linux/fs.h>

9 #include <linux/namei.h>

10 #include <linux/xattr.h>

11 #include <linux/mount.h>

12 #include <linux/parser.h>

13 #include <linux/module.h>

14 #include <linux/statfs.h>

15 #include <linux/seq_file.h>

16 #include <linux/posix_acl_xattr.h>

17 #include <linux/exportfs.h>

18 #include <linux/file.h>

19 #include <linux/fs_context.h>

20 #include <linux/fs_parser.h>

21 #include "overlayfs.h"

22 #include "params.h"

23

24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");

25 MODULE_DESCRIPTION("Overlay filesystem");

26 MODULE_LICENSE("GPL");

27

28

29 struct ovl_dir_cache;

30

31 static struct dentry *ovl_d_real(struct dentry *dentry,

32 const struct inode *inode)

33 {

34 struct dentry *real = NULL, *lower;

35 int err;

36

37 /*

38 * vfs is only expected to call d_real() with NULL from d_real_inode()

39 * and with overlay inode from file_dentry() on an overlay file.

40 *

41 * TODO: remove @inode argument from d_real() API, remove code in this

42 * function that deals with non-NULL @inode and remove d_real() call

43 * from file_dentry().

44 */

45 if (inode && d_inode(dentry) == inode)

46 return dentry;

47 else if (inode)

48 goto bug;

49

50 if (!d_is_reg(dentry)) {

51 /* d_real_inode() is only relevant for regular files */

52 return dentry;

53 }

54

55 real = ovl_dentry_upper(dentry);

56 if (real && (inode == d_inode(real)))

57 return real;

58

59 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))

60 return real;

61

62 /*

63 * Best effort lazy lookup of lowerdata for !inode case to return

64 * the real lowerdata dentry. The only current caller of d_real() with

65 * NULL inode is d_real_inode() from trace_uprobe and this caller is

66 * likely going to be followed reading from the file, before placing

67 * uprobes on offset within the file, so lowerdata should be available

68 * when setting the uprobe.

69 */

70 err = ovl_verify_lowerdata(dentry);

71 if (err)

72 goto bug;

73 lower = ovl_dentry_lowerdata(dentry);

74 if (!lower)

75 goto bug;

76 real = lower;

77

78 /* Handle recursion */

79 real = d_real(real, inode);

80

81 if (!inode || inode == d_inode(real))

82 return real;

83 bug:

84 WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",

85 __func__, dentry, inode ? inode->i_sb->s_id : "NULL",

86 inode ? inode->i_ino : 0, real,

87 real && d_inode(real) ? d_inode(real)->i_ino : 0);

88 return dentry;

89 }

90

91 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)

92 {

93 int ret = 1;

94

95 if (!d)

96 return 1;

97

98 if (weak) {

99 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)

100 ret = d->d_op->d_weak_revalidate(d, flags);

101 } else if (d->d_flags & DCACHE_OP_REVALIDATE) {

102 ret = d->d_op->d_revalidate(d, flags);

103 if (!ret) {

104 if (!(flags & LOOKUP_RCU))

105 d_invalidate(d);

106 ret = -ESTALE;

107 }

108 }

109 return ret;

110 }

111

112 static int ovl_dentry_revalidate_common(struct dentry *dentry,

113 unsigned int flags, bool weak)

114 {

115 struct ovl_entry *oe;

116 struct ovl_path *lowerstack;

117 struct inode *inode = d_inode_rcu(dentry);

118 struct dentry *upper;

119 unsigned int i;

120 int ret = 1;

121

122 /* Careful in RCU mode */

123 if (!inode)

124 return -ECHILD;

125

126 oe = OVL_I_E(inode);

127 lowerstack = ovl_lowerstack(oe);

128 upper = ovl_i_dentry_upper(inode);

129 if (upper)

130 ret = ovl_revalidate_real(upper, flags, weak);

131

132 for (i = 0; ret > 0 && i < ovl_numlower(oe); i++)

133 ret = ovl_revalidate_real(lowerstack[i].dentry, flags, weak);

134

135 return ret;

136 }

137

138 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)

139 {

140 return ovl_dentry_revalidate_common(dentry, flags, false);

141 }

142

143 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)

144 {

145 return ovl_dentry_revalidate_common(dentry, flags, true);

146 }

147

148 static const struct dentry_operations ovl_dentry_operations = {

149 .d_real = ovl_d_real,

150 .d_revalidate = ovl_dentry_revalidate,

151 .d_weak_revalidate = ovl_dentry_weak_revalidate,

152 };

153

154 static struct kmem_cache *ovl_inode_cachep;

155

156 static struct inode *ovl_alloc_inode(struct super_block *sb)

157 {

158 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);

159

160 if (!oi)

161 return NULL;

162

163 oi->cache = NULL;

164 oi->redirect = NULL;

165 oi->version = 0;

166 oi->flags = 0;

167 oi->__upperdentry = NULL;

168 oi->lowerdata_redirect = NULL;

169 oi->oe = NULL;

170 mutex_init(&oi->lock);

171

172 return &oi->vfs_inode;

173 }

174

175 static void ovl_free_inode(struct inode *inode)

176 {

177 struct ovl_inode *oi = OVL_I(inode);

178

179 kfree(oi->redirect);

180 kfree(oi->oe);

181 mutex_destroy(&oi->lock);

182 kmem_cache_free(ovl_inode_cachep, oi);

183 }

184

185 static void ovl_destroy_inode(struct inode *inode)

186 {

187 struct ovl_inode *oi = OVL_I(inode);

188

189 dput(oi->__upperdentry);

190 ovl_stack_put(ovl_lowerstack(oi->oe), ovl_numlower(oi->oe));

191 if (S_ISDIR(inode->i_mode))

192 ovl_dir_cache_free(inode);

193 else

194 kfree(oi->lowerdata_redirect);

195 }

196

197 static void ovl_put_super(struct super_block *sb)

198 {

199 struct ovl_fs *ofs = OVL_FS(sb);

200

201 if (ofs)

202 ovl_free_fs(ofs);

203 }

204

205 /* Sync real dirty inodes in upper filesystem (if it exists) */

206 static int ovl_sync_fs(struct super_block *sb, int wait)

207 {

208 struct ovl_fs *ofs = OVL_FS(sb);

209 struct super_block *upper_sb;

210 int ret;

211

212 ret = ovl_sync_status(ofs);

213 /*

214 * We have to always set the err, because the return value isn't

215 * checked in syncfs, and instead indirectly return an error via

216 * the sb's writeback errseq, which VFS inspects after this call.

217 */

218 if (ret < 0) {

219 errseq_set(&sb->s_wb_err, -EIO);

220 return -EIO;

221 }

222

223 if (!ret)

224 return ret;

225

226 /*

227 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).

228 * All the super blocks will be iterated, including upper_sb.

229 *

230 * If this is a syncfs(2) call, then we do need to call

231 * sync_filesystem() on upper_sb, but enough if we do it when being

232 * called with wait == 1.

233 */

234 if (!wait)

235 return 0;

236

237 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;

238

239 down_read(&upper_sb->s_umount);

240 ret = sync_filesystem(upper_sb);

241 up_read(&upper_sb->s_umount);

242

243 return ret;

244 }

245

246 /**

247 * ovl_statfs

248 * @dentry: The dentry to query

249 * @buf: The struct kstatfs to fill in with stats

250 *

251 * Get the filesystem statistics. As writes always target the upper layer

252 * filesystem pass the statfs to the upper filesystem (if it exists)

253 */

254 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)

255 {

256 struct super_block *sb = dentry->d_sb;

257 struct ovl_fs *ofs = OVL_FS(sb);

258 struct dentry *root_dentry = sb->s_root;

259 struct path path;

260 int err;

261

262 ovl_path_real(root_dentry, &path);

263

264 err = vfs_statfs(&path, buf);

265 if (!err) {

266 buf->f_namelen = ofs->namelen;

267 buf->f_type = OVERLAYFS_SUPER_MAGIC;

268 if (ovl_has_fsid(ofs))

269 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);

270 }

271

272 return err;

273 }

274

275 static const struct super_operations ovl_super_operations = {

276 .alloc_inode = ovl_alloc_inode,

277 .free_inode = ovl_free_inode,

278 .destroy_inode = ovl_destroy_inode,

279 .drop_inode = generic_delete_inode,

280 .put_super = ovl_put_super,

281 .sync_fs = ovl_sync_fs,

282 .statfs = ovl_statfs,

283 .show_options = ovl_show_options,

284 };

285

286 #define OVL_WORKDIR_NAME "work"

287 #define OVL_INDEXDIR_NAME "index"

288

289 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,

290 const char *name, bool persist)

291 {

292 struct inode *dir = ofs->workbasedir->d_inode;

293 struct vfsmount *mnt = ovl_upper_mnt(ofs);

294 struct dentry *work;

295 int err;

296 bool retried = false;

297

298 inode_lock_nested(dir, I_MUTEX_PARENT);

299 retry:

300 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));

301

302 if (!IS_ERR(work)) {

303 struct iattr attr = {

304 .ia_valid = ATTR_MODE,

305 .ia_mode = S_IFDIR | 0,

306 };

307

308 if (work->d_inode) {

309 err = -EEXIST;

310 if (retried)

311 goto out_dput;

312

313 if (persist)

314 goto out_unlock;

315

316 retried = true;

317 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);

318 dput(work);

319 if (err == -EINVAL) {

320 work = ERR_PTR(err);

321 goto out_unlock;

322 }

323 goto retry;

324 }

325

326 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);

327 if (err)

328 goto out_dput;

329

330 /* Weird filesystem returning with hashed negative (kernfs)? */

331 err = -EINVAL;

332 if (d_really_is_negative(work))

333 goto out_dput;

334

335 /*

336 * Try to remove POSIX ACL xattrs from workdir. We are good if:

337 *

338 * a) success (there was a POSIX ACL xattr and was removed)

339 * b) -ENODATA (there was no POSIX ACL xattr)

340 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)

341 *

342 * There are various other error values that could effectively

343 * mean that the xattr doesn't exist (e.g. -ERANGE is returned

344 * if the xattr name is too long), but the set of filesystems

345 * allowed as upper are limited to "normal" ones, where checking

346 * for the above two errors is sufficient.

347 */

348 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);

349 if (err && err != -ENODATA && err != -EOPNOTSUPP)

350 goto out_dput;

351

352 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);

353 if (err && err != -ENODATA && err != -EOPNOTSUPP)

354 goto out_dput;

355

356 /* Clear any inherited mode bits */

357 inode_lock(work->d_inode);

358 err = ovl_do_notify_change(ofs, work, &attr);

359 inode_unlock(work->d_inode);

360 if (err)

361 goto out_dput;

362 } else {

363 err = PTR_ERR(work);

364 goto out_err;

365 }

366 out_unlock:

367 inode_unlock(dir);

368 return work;

369

370 out_dput:

371 dput(work);

372 out_err:

373 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",

374 ofs->config.workdir, name, -err);

375 work = NULL;

376 goto out_unlock;

377 }

378

379 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,

380 const char *name)

381 {

382 struct kstatfs statfs;

383 int err = vfs_statfs(path, &statfs);

384

385 if (err)

386 pr_err("statfs failed on '%s'\n", name);

387 else

388 ofs->namelen = max(ofs->namelen, statfs.f_namelen);

389

390 return err;

391 }

392

393 static int ovl_lower_dir(const char *name, struct path *path,

394 struct ovl_fs *ofs, int *stack_depth)

395 {

396 int fh_type;

397 int err;

398

399 err = ovl_check_namelen(path, ofs, name);

400 if (err)

401 return err;

402

403 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);

404

405 /*

406 * The inodes index feature and NFS export need to encode and decode

407 * file handles, so they require that all layers support them.

408 */

409 fh_type = ovl_can_decode_fh(path->dentry->d_sb);

410 if ((ofs->config.nfs_export ||

411 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {

412 ofs->config.index = false;

413 ofs->config.nfs_export = false;

414 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",

415 name);

416 }

417 ofs->nofh |= !fh_type;

418 /*

419 * Decoding origin file handle is required for persistent st_ino.

420 * Without persistent st_ino, xino=auto falls back to xino=off.

421 */

422 if (ofs->config.xino == OVL_XINO_AUTO &&

423 ofs->config.upperdir && !fh_type) {

424 ofs->config.xino = OVL_XINO_OFF;

425 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",

426 name);

427 }

428

429 /* Check if lower fs has 32bit inode numbers */

430 if (fh_type != FILEID_INO32_GEN)

431 ofs->xino_mode = -1;

432

433 return 0;

434 }

435

436 /* Workdir should not be subdir of upperdir and vice versa */

437 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)

438 {

439 bool ok = false;

440

441 if (workdir != upperdir) {

442 struct dentry *trap = lock_rename(workdir, upperdir);

443 if (!IS_ERR(trap))

444 unlock_rename(workdir, upperdir);

445 ok = (trap == NULL);

446 }

447 return ok;

448 }

449

450 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,

451 struct inode **ptrap, const char *name)

452 {

453 struct inode *trap;

454 int err;

455

456 trap = ovl_get_trap_inode(sb, dir);

457 err = PTR_ERR_OR_ZERO(trap);

458 if (err) {

459 if (err == -ELOOP)

460 pr_err("conflicting %s path\n", name);

461 return err;

462 }

463

464 *ptrap = trap;

465 return 0;

466 }

467

468 /*

469 * Determine how we treat concurrent use of upperdir/workdir based on the

470 * index feature. This is papering over mount leaks of container runtimes,

471 * for example, an old overlay mount is leaked and now its upperdir is

472 * attempted to be used as a lower layer in a new overlay mount.

473 */

474 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)

475 {

476 if (ofs->config.index) {

477 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",

478 name);

479 return -EBUSY;

480 } else {

481 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",

482 name);

483 return 0;

484 }

485 }

486

487 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,

488 struct ovl_layer *upper_layer,

489 const struct path *upperpath)

490 {

491 struct vfsmount *upper_mnt;

492 int err;

493

494 /* Upperdir path should not be r/o */

495 if (__mnt_is_readonly(upperpath->mnt)) {

496 pr_err("upper fs is r/o, try multi-lower layers mount\n");

497 err = -EINVAL;

498 goto out;

499 }

500

501 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);

502 if (err)

503 goto out;

504

505 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,

506 "upperdir");

507 if (err)

508 goto out;

509

510 upper_mnt = clone_private_mount(upperpath);

511 err = PTR_ERR(upper_mnt);

512 if (IS_ERR(upper_mnt)) {

513 pr_err("failed to clone upperpath\n");

514 goto out;

515 }

516

517 /* Don't inherit atime flags */

518 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);

519 upper_layer->mnt = upper_mnt;

520 upper_layer->idx = 0;

521 upper_layer->fsid = 0;

522

523 /*

524 * Inherit SB_NOSEC flag from upperdir.

525 *

526 * This optimization changes behavior when a security related attribute

527 * (suid/sgid/security.*) is changed on an underlying layer. This is

528 * okay because we don't yet have guarantees in that case, but it will

529 * need careful treatment once we want to honour changes to underlying

530 * filesystems.

531 */

532 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)

533 sb->s_flags |= SB_NOSEC;

534

535 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {

536 ofs->upperdir_locked = true;

537 } else {

538 err = ovl_report_in_use(ofs, "upperdir");

539 if (err)

540 goto out;

541 }

542

543 err = 0;

544 out:

545 return err;

546 }

547

548 /*

549 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and

550 * negative values if error is encountered.

551 */

552 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)

553 {

554 struct dentry *workdir = ofs->workdir;

555 struct inode *dir = d_inode(workdir);

556 struct dentry *temp;

557 struct dentry *dest;

558 struct dentry *whiteout;

559 struct name_snapshot name;

560 int err;

561

562 inode_lock_nested(dir, I_MUTEX_PARENT);

563

564 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));

565 err = PTR_ERR(temp);

566 if (IS_ERR(temp))

567 goto out_unlock;

568

569 dest = ovl_lookup_temp(ofs, workdir);

570 err = PTR_ERR(dest);

571 if (IS_ERR(dest)) {

572 dput(temp);

573 goto out_unlock;

574 }

575

576 /* Name is inline and stable - using snapshot as a copy helper */

577 take_dentry_name_snapshot(&name, temp);

578 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);

579 if (err) {

580 if (err == -EINVAL)

581 err = 0;

582 goto cleanup_temp;

583 }

584

585 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);

586 err = PTR_ERR(whiteout);

587 if (IS_ERR(whiteout))

588 goto cleanup_temp;

589

590 err = ovl_upper_is_whiteout(ofs, whiteout);

591

592 /* Best effort cleanup of whiteout and temp file */

593 if (err)

594 ovl_cleanup(ofs, dir, whiteout);

595 dput(whiteout);

596

597 cleanup_temp:

598 ovl_cleanup(ofs, dir, temp);

599 release_dentry_name_snapshot(&name);

600 dput(temp);

601 dput(dest);

602

603 out_unlock:

604 inode_unlock(dir);

605

606 return err;

607 }

608

609 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,

610 struct dentry *parent,

611 const char *name, umode_t mode)

612 {

613 size_t len = strlen(name);

614 struct dentry *child;

615

616 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);

617 child = ovl_lookup_upper(ofs, name, parent, len);

618 if (!IS_ERR(child) && !child->d_inode)

619 child = ovl_create_real(ofs, parent->d_inode, child,

620 OVL_CATTR(mode));

621 inode_unlock(parent->d_inode);

622 dput(parent);

623

624 return child;

625 }

626

627 /*

628 * Creates $workdir/work/incompat/volatile/dirty file if it is not already

629 * present.

630 */

631 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)

632 {

633 unsigned int ctr;

634 struct dentry *d = dget(ofs->workbasedir);

635 static const char *const volatile_path[] = {

636 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"

637 };

638 const char *const *name = volatile_path;

639

640 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {

641 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);

642 if (IS_ERR(d))

643 return PTR_ERR(d);

644 }

645 dput(d);

646 return 0;

647 }

648

649 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,

650 const struct path *workpath)

651 {

652 struct vfsmount *mnt = ovl_upper_mnt(ofs);

653 struct dentry *workdir;

654 struct file *tmpfile;

655 bool rename_whiteout;

656 bool d_type;

657 int fh_type;

658 int err;

659

660 err = mnt_want_write(mnt);

661 if (err)

662 return err;

663

664 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);

665 err = PTR_ERR(workdir);

666 if (IS_ERR_OR_NULL(workdir))

667 goto out;

668

669 ofs->workdir = workdir;

670

671 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");

672 if (err)

673 goto out;

674

675 /*

676 * Upper should support d_type, else whiteouts are visible. Given

677 * workdir and upper are on same fs, we can do iterate_dir() on

678 * workdir. This check requires successful creation of workdir in

679 * previous step.

680 */

681 err = ovl_check_d_type_supported(workpath);

682 if (err < 0)

683 goto out;

684

685 d_type = err;

686 if (!d_type)

687 pr_warn("upper fs needs to support d_type.\n");

688

689 /* Check if upper/work fs supports O_TMPFILE */

690 tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);

691 ofs->tmpfile = !IS_ERR(tmpfile);

692 if (ofs->tmpfile)

693 fput(tmpfile);

694 else

695 pr_warn("upper fs does not support tmpfile.\n");

696

697

698 /* Check if upper/work fs supports RENAME_WHITEOUT */

699 err = ovl_check_rename_whiteout(ofs);

700 if (err < 0)

701 goto out;

702

703 rename_whiteout = err;

704 if (!rename_whiteout)

705 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");

706

707 /*

708 * Check if upper/work fs supports (trusted|user).overlay.* xattr

709 */

710 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);

711 if (err) {

712 pr_warn("failed to set xattr on upper\n");

713 ofs->noxattr = true;

714 if (ovl_redirect_follow(ofs)) {

715 ofs->config.redirect_mode = OVL_REDIRECT_NOFOLLOW;

716 pr_warn("...falling back to redirect_dir=nofollow.\n");

717 }

718 if (ofs->config.metacopy) {

719 ofs->config.metacopy = false;

720 pr_warn("...falling back to metacopy=off.\n");

721 }

722 if (ofs->config.index) {

723 ofs->config.index = false;

724 pr_warn("...falling back to index=off.\n");

725 }

726 if (ovl_has_fsid(ofs)) {

727 ofs->config.uuid = OVL_UUID_NULL;

728 pr_warn("...falling back to uuid=null.\n");

729 }

730 /*

731 * xattr support is required for persistent st_ino.

732 * Without persistent st_ino, xino=auto falls back to xino=off.

733 */

734 if (ofs->config.xino == OVL_XINO_AUTO) {

735 ofs->config.xino = OVL_XINO_OFF;

736 pr_warn("...falling back to xino=off.\n");

737 }

738 if (err == -EPERM && !ofs->config.userxattr)

739 pr_info("try mounting with 'userxattr' option\n");

740 err = 0;

741 } else {

742 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);

743 }

744

745 /*

746 * We allowed sub-optimal upper fs configuration and don't want to break

747 * users over kernel upgrade, but we never allowed remote upper fs, so

748 * we can enforce strict requirements for remote upper fs.

749 */

750 if (ovl_dentry_remote(ofs->workdir) &&

751 (!d_type || !rename_whiteout || ofs->noxattr)) {

752 pr_err("upper fs missing required features.\n");

753 err = -EINVAL;

754 goto out;

755 }

756

757 /*

758 * For volatile mount, create a incompat/volatile/dirty file to keep

759 * track of it.

760 */

761 if (ofs->config.ovl_volatile) {

762 err = ovl_create_volatile_dirty(ofs);

763 if (err < 0) {

764 pr_err("Failed to create volatile/dirty file.\n");

765 goto out;

766 }

767 }

768

769 /* Check if upper/work fs supports file handles */

770 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);

771 if (ofs->config.index && !fh_type) {

772 ofs->config.index = false;

773 pr_warn("upper fs does not support file handles, falling back to index=off.\n");

774 }

775 ofs->nofh |= !fh_type;

776

777 /* Check if upper fs has 32bit inode numbers */

778 if (fh_type != FILEID_INO32_GEN)

779 ofs->xino_mode = -1;

780

781 /* NFS export of r/w mount depends on index */

782 if (ofs->config.nfs_export && !ofs->config.index) {

783 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");

784 ofs->config.nfs_export = false;

785 }

786 out:

787 mnt_drop_write(mnt);

788 return err;

789 }

790

791 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,

792 const struct path *upperpath,

793 const struct path *workpath)

794 {

795 int err;

796

797 err = -EINVAL;

798 if (upperpath->mnt != workpath->mnt) {

799 pr_err("workdir and upperdir must reside under the same mount\n");

800 return err;

801 }

802 if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {

803 pr_err("workdir and upperdir must be separate subtrees\n");

804 return err;

805 }

806

807 ofs->workbasedir = dget(workpath->dentry);

808

809 if (ovl_inuse_trylock(ofs->workbasedir)) {

810 ofs->workdir_locked = true;

811 } else {

812 err = ovl_report_in_use(ofs, "workdir");

813 if (err)

814 return err;

815 }

816

817 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,

818 "workdir");

819 if (err)

820 return err;

821

822 return ovl_make_workdir(sb, ofs, workpath);

823 }

824

825 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,

826 struct ovl_entry *oe, const struct path *upperpath)

827 {

828 struct vfsmount *mnt = ovl_upper_mnt(ofs);

829 struct dentry *indexdir;

830 struct dentry *origin = ovl_lowerstack(oe)->dentry;

831 const struct ovl_fh *fh;

832 int err;

833

834 fh = ovl_get_origin_fh(ofs, origin);

835 if (IS_ERR(fh))

836 return PTR_ERR(fh);

837

838 err = mnt_want_write(mnt);

839 if (err)

840 goto out_free_fh;

841

842 /* Verify lower root is upper root origin */

843 err = ovl_verify_origin_fh(ofs, upperpath->dentry, fh, true);

844 if (err) {

845 pr_err("failed to verify upper root origin\n");

846 goto out;

847 }

848

849 /* index dir will act also as workdir */

850 iput(ofs->workdir_trap);

851 ofs->workdir_trap = NULL;

852 dput(ofs->workdir);

853 ofs->workdir = NULL;

854 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);

855 if (IS_ERR(indexdir)) {

856 err = PTR_ERR(indexdir);

857 } else if (indexdir) {

858 ofs->workdir = indexdir;

859 err = ovl_setup_trap(sb, indexdir, &ofs->workdir_trap,

860 "indexdir");

861 if (err)

862 goto out;

863

864 /*

865 * Verify upper root is exclusively associated with index dir.

866 * Older kernels stored upper fh in ".overlay.origin"

867 * xattr. If that xattr exists, verify that it is a match to

868 * upper dir file handle. In any case, verify or set xattr

869 * ".overlay.upper" to indicate that index may have

870 * directory entries.

871 */

872 if (ovl_check_origin_xattr(ofs, indexdir)) {

873 err = ovl_verify_origin_xattr(ofs, indexdir,

874 OVL_XATTR_ORIGIN,

875 upperpath->dentry, true,

876 false);

877 if (err)

878 pr_err("failed to verify index dir 'origin' xattr\n");

879 }

880 err = ovl_verify_upper(ofs, indexdir, upperpath->dentry, true);

881 if (err)

882 pr_err("failed to verify index dir 'upper' xattr\n");

883

884 /* Cleanup bad/stale/orphan index entries */

885 if (!err)

886 err = ovl_indexdir_cleanup(ofs);

887 }

888 if (err || !indexdir)

889 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");

890

891 out:

892 mnt_drop_write(mnt);

893 out_free_fh:

894 kfree(fh);

895 return err;

896 }

897

898 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)

899 {

900 unsigned int i;

901

902 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))

903 return true;

904

905 /*

906 * We allow using single lower with null uuid for index and nfs_export

907 * for example to support those features with single lower squashfs.

908 * To avoid regressions in setups of overlay with re-formatted lower

909 * squashfs, do not allow decoding origin with lower null uuid unless

910 * user opted-in to one of the new features that require following the

911 * lower inode of non-dir upper.

912 */

913 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))

914 return false;

915

916 for (i = 0; i < ofs->numfs; i++) {

917 /*

918 * We use uuid to associate an overlay lower file handle with a

919 * lower layer, so we can accept lower fs with null uuid as long

920 * as all lower layers with null uuid are on the same fs.

921 * if we detect multiple lower fs with the same uuid, we

922 * disable lower file handle decoding on all of them.

923 */

924 if (ofs->fs[i].is_lower &&

925 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {

926 ofs->fs[i].bad_uuid = true;

927 return false;

928 }

929 }

930 return true;

931 }

932

933 /* Get a unique fsid for the layer */

934 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)

935 {

936 struct super_block *sb = path->mnt->mnt_sb;

937 unsigned int i;

938 dev_t dev;

939 int err;

940 bool bad_uuid = false;

941 bool warn = false;

942

943 for (i = 0; i < ofs->numfs; i++) {

944 if (ofs->fs[i].sb == sb)

945 return i;

946 }

947

948 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {

949 bad_uuid = true;

950 if (ofs->config.xino == OVL_XINO_AUTO) {

951 ofs->config.xino = OVL_XINO_OFF;

952 warn = true;

953 }

954 if (ofs->config.index || ofs->config.nfs_export) {

955 ofs->config.index = false;

956 ofs->config.nfs_export = false;

957 warn = true;

958 }

959 if (warn) {

960 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",

961 uuid_is_null(&sb->s_uuid) ? "null" :

962 "conflicting",

963 path->dentry, ovl_xino_mode(&ofs->config));

964 }

965 }

966

967 err = get_anon_bdev(&dev);

968 if (err) {

969 pr_err("failed to get anonymous bdev for lowerpath\n");

970 return err;

971 }

972

973 ofs->fs[ofs->numfs].sb = sb;

974 ofs->fs[ofs->numfs].pseudo_dev = dev;

975 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;

976

977 return ofs->numfs++;

978 }

979

980 /*

981 * The fsid after the last lower fsid is used for the data layers.

982 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.

983 */

984 static int ovl_get_data_fsid(struct ovl_fs *ofs)

985 {

986 return ofs->numfs;

987 }

988

989

990 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,

991 struct ovl_fs_context *ctx, struct ovl_layer *layers)

992 {

993 int err;

994 unsigned int i;

995 size_t nr_merged_lower;

996

997 ofs->fs = kcalloc(ctx->nr + 2, sizeof(struct ovl_sb), GFP_KERNEL);

998 if (ofs->fs == NULL)

999 return -ENOMEM;

1000

1001 /*

1002 * idx/fsid 0 are reserved for upper fs even with lower only overlay

1003 * and the last fsid is reserved for "null fs" of the data layers.

1004 */

1005 ofs->numfs++;

1006

1007 /*

1008 * All lower layers that share the same fs as upper layer, use the same

1009 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower

1010 * only overlay to simplify ovl_fs_free().

1011 * is_lower will be set if upper fs is shared with a lower layer.

1012 */

1013 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);

1014 if (err) {

1015 pr_err("failed to get anonymous bdev for upper fs\n");

1016 return err;

1017 }

1018

1019 if (ovl_upper_mnt(ofs)) {

1020 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;

1021 ofs->fs[0].is_lower = false;

1022 }

1023

1024 nr_merged_lower = ctx->nr - ctx->nr_data;

1025 for (i = 0; i < ctx->nr; i++) {

1026 struct ovl_fs_context_layer *l = &ctx->lower[i];

1027 struct vfsmount *mnt;

1028 struct inode *trap;

1029 int fsid;

1030

1031 if (i < nr_merged_lower)

1032 fsid = ovl_get_fsid(ofs, &l->path);

1033 else

1034 fsid = ovl_get_data_fsid(ofs);

1035 if (fsid < 0)

1036 return fsid;

1037

1038 /*

1039 * Check if lower root conflicts with this overlay layers before

1040 * checking if it is in-use as upperdir/workdir of "another"

1041 * mount, because we do not bother to check in ovl_is_inuse() if

1042 * the upperdir/workdir is in fact in-use by our

1043 * upperdir/workdir.

1044 */

1045 err = ovl_setup_trap(sb, l->path.dentry, &trap, "lowerdir");

1046 if (err)

1047 return err;

1048

1049 if (ovl_is_inuse(l->path.dentry)) {

1050 err = ovl_report_in_use(ofs, "lowerdir");

1051 if (err) {

1052 iput(trap);

1053 return err;

1054 }

1055 }

1056

1057 mnt = clone_private_mount(&l->path);

1058 err = PTR_ERR(mnt);

1059 if (IS_ERR(mnt)) {

1060 pr_err("failed to clone lowerpath\n");

1061 iput(trap);

1062 return err;

1063 }

1064

1065 /*

1066 * Make lower layers R/O. That way fchmod/fchown on lower file

1067 * will fail instead of modifying lower fs.

1068 */

1069 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;

1070

1071 layers[ofs->numlayer].trap = trap;

1072 layers[ofs->numlayer].mnt = mnt;

1073 layers[ofs->numlayer].idx = ofs->numlayer;

1074 layers[ofs->numlayer].fsid = fsid;

1075 layers[ofs->numlayer].fs = &ofs->fs[fsid];

1076 /* Store for printing lowerdir=... in ovl_show_options() */

1077 ofs->config.lowerdirs[ofs->numlayer] = l->name;

1078 l->name = NULL;

1079 ofs->numlayer++;

1080 ofs->fs[fsid].is_lower = true;

1081 }

1082

1083 /*

1084 * When all layers on same fs, overlay can use real inode numbers.

1085 * With mount option "xino=<on|auto>", mounter declares that there are

1086 * enough free high bits in underlying fs to hold the unique fsid.

1087 * If overlayfs does encounter underlying inodes using the high xino

1088 * bits reserved for fsid, it emits a warning and uses the original

1089 * inode number or a non persistent inode number allocated from a

1090 * dedicated range.

1091 */

1092 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {

1093 if (ofs->config.xino == OVL_XINO_ON)

1094 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");

1095 ofs->xino_mode = 0;

1096 } else if (ofs->config.xino == OVL_XINO_OFF) {

1097 ofs->xino_mode = -1;

1098 } else if (ofs->xino_mode < 0) {

1099 /*

1100 * This is a roundup of number of bits needed for encoding

1101 * fsid, where fsid 0 is reserved for upper fs (even with

1102 * lower only overlay) +1 extra bit is reserved for the non

1103 * persistent inode number range that is used for resolving

1104 * xino lower bits overflow.

1105 */

1106 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);

1107 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;

1108 }

1109

1110 if (ofs->xino_mode > 0) {

1111 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",

1112 ofs->xino_mode);

1113 }

1114

1115 return 0;

1116 }

1117

1118 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,

1119 struct ovl_fs_context *ctx,

1120 struct ovl_fs *ofs,

1121 struct ovl_layer *layers)

1122 {

1123 int err;

1124 unsigned int i;

1125 size_t nr_merged_lower;

1126 struct ovl_entry *oe;

1127 struct ovl_path *lowerstack;

1128

1129 struct ovl_fs_context_layer *l;

1130

1131 if (!ofs->config.upperdir && ctx->nr == 1) {

1132 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");

1133 return ERR_PTR(-EINVAL);

1134 }

1135

1136 err = -EINVAL;

1137 for (i = 0; i < ctx->nr; i++) {

1138 l = &ctx->lower[i];

1139

1140 err = ovl_lower_dir(l->name, &l->path, ofs, &sb->s_stack_depth);

1141 if (err)

1142 return ERR_PTR(err);

1143 }

1144

1145 err = -EINVAL;

1146 sb->s_stack_depth++;

1147 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {

1148 pr_err("maximum fs stacking depth exceeded\n");

1149 return ERR_PTR(err);

1150 }

1151

1152 err = ovl_get_layers(sb, ofs, ctx, layers);

1153 if (err)

1154 return ERR_PTR(err);

1155

1156 err = -ENOMEM;

1157 /* Data-only layers are not merged in root directory */

1158 nr_merged_lower = ctx->nr - ctx->nr_data;

1159 oe = ovl_alloc_entry(nr_merged_lower);

1160 if (!oe)

1161 return ERR_PTR(err);

1162

1163 lowerstack = ovl_lowerstack(oe);

1164 for (i = 0; i < nr_merged_lower; i++) {

1165 l = &ctx->lower[i];

1166 lowerstack[i].dentry = dget(l->path.dentry);

1167 lowerstack[i].layer = &ofs->layers[i + 1];

1168 }

1169 ofs->numdatalayer = ctx->nr_data;

1170

1171 return oe;

1172 }

1173

1174 /*

1175 * Check if this layer root is a descendant of:

1176 * - another layer of this overlayfs instance

1177 * - upper/work dir of any overlayfs instance

1178 */

1179 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,

1180 struct dentry *dentry, const char *name,

1181 bool is_lower)

1182 {

1183 struct dentry *next = dentry, *parent;

1184 int err = 0;

1185

1186 if (!dentry)

1187 return 0;

1188

1189 parent = dget_parent(next);

1190

1191 /* Walk back ancestors to root (inclusive) looking for traps */

1192 while (!err && parent != next) {

1193 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {

1194 err = -ELOOP;

1195 pr_err("overlapping %s path\n", name);

1196 } else if (ovl_is_inuse(parent)) {

1197 err = ovl_report_in_use(ofs, name);

1198 }

1199 next = parent;

1200 parent = dget_parent(next);

1201 dput(next);

1202 }

1203

1204 dput(parent);

1205

1206 return err;

1207 }

1208

1209 /*

1210 * Check if any of the layers or work dirs overlap.

1211 */

1212 static int ovl_check_overlapping_layers(struct super_block *sb,

1213 struct ovl_fs *ofs)

1214 {

1215 int i, err;

1216

1217 if (ovl_upper_mnt(ofs)) {

1218 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,

1219 "upperdir", false);

1220 if (err)

1221 return err;

1222

1223 /*

1224 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of

1225 * this instance and covers overlapping work and index dirs,

1226 * unless work or index dir have been moved since created inside

1227 * workbasedir. In that case, we already have their traps in

1228 * inode cache and we will catch that case on lookup.

1229 */

1230 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",

1231 false);

1232 if (err)

1233 return err;

1234 }

1235

1236 for (i = 1; i < ofs->numlayer; i++) {

1237 err = ovl_check_layer(sb, ofs,

1238 ofs->layers[i].mnt->mnt_root,

1239 "lowerdir", true);

1240 if (err)

1241 return err;

1242 }

1243

1244 return 0;

1245 }

1246

1247 static struct dentry *ovl_get_root(struct super_block *sb,

1248 struct dentry *upperdentry,

1249 struct ovl_entry *oe)

1250 {

1251 struct dentry *root;

1252 struct ovl_fs *ofs = OVL_FS(sb);

1253 struct ovl_path *lowerpath = ovl_lowerstack(oe);

1254 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;

1255 int fsid = lowerpath->layer->fsid;

1256 struct ovl_inode_params oip = {

1257 .upperdentry = upperdentry,

1258 .oe = oe,

1259 };

1260

1261 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));

1262 if (!root)

1263 return NULL;

1264

1265 if (upperdentry) {

1266 /* Root inode uses upper st_ino/i_ino */

1267 ino = d_inode(upperdentry)->i_ino;

1268 fsid = 0;

1269 ovl_dentry_set_upper_alias(root);

1270 if (ovl_is_impuredir(sb, upperdentry))

1271 ovl_set_flag(OVL_IMPURE, d_inode(root));

1272 }

1273

1274 /* Look for xwhiteouts marker except in the lowermost layer */

1275 for (int i = 0; i < ovl_numlower(oe) - 1; i++, lowerpath++) {

1276 struct path path = {

1277 .mnt = lowerpath->layer->mnt,

1278 .dentry = lowerpath->dentry,

1279 };

1280

1281 /* overlay.opaque=x means xwhiteouts directory */

1282 if (ovl_get_opaquedir_val(ofs, &path) == 'x') {

1283 ovl_layer_set_xwhiteouts(ofs, lowerpath->layer);

1284 ovl_dentry_set_xwhiteouts(root);

1285 }

1286 }

1287

1288 /* Root is always merge -> can have whiteouts */

1289 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));

1290 ovl_dentry_set_flag(OVL_E_CONNECTED, root);

1291 ovl_set_upperdata(d_inode(root));

1292 ovl_inode_init(d_inode(root), &oip, ino, fsid);

1293 ovl_dentry_init_flags(root, upperdentry, oe, DCACHE_OP_WEAK_REVALIDATE);

1294 /* root keeps a reference of upperdentry */

1295 dget(upperdentry);

1296

1297 return root;

1298 }

1299

1300 int ovl_fill_super(struct super_block *sb, struct fs_context *fc)

1301 {

1302 struct ovl_fs *ofs = sb->s_fs_info;

1303 struct ovl_fs_context *ctx = fc->fs_private;

1304 struct dentry *root_dentry;

1305 struct ovl_entry *oe;

1306 struct ovl_layer *layers;

1307 struct cred *cred;

1308 int err;

1309

1310 err = -EIO;

1311 if (WARN_ON(fc->user_ns != current_user_ns()))

1312 goto out_err;

1313

1314 sb->s_d_op = &ovl_dentry_operations;

1315

1316 err = -ENOMEM;

1317 ofs->creator_cred = cred = prepare_creds();

1318 if (!cred)

1319 goto out_err;

1320

1321 err = ovl_fs_params_verify(ctx, &ofs->config);

1322 if (err)

1323 goto out_err;

1324

1325 err = -EINVAL;

1326 if (ctx->nr == 0) {

1327 if (!(fc->sb_flags & SB_SILENT))

1328 pr_err("missing 'lowerdir'\n");

1329 goto out_err;

1330 }

1331

1332 err = -ENOMEM;

1333 layers = kcalloc(ctx->nr + 1, sizeof(struct ovl_layer), GFP_KERNEL);

1334 if (!layers)

1335 goto out_err;

1336

1337 ofs->config.lowerdirs = kcalloc(ctx->nr + 1, sizeof(char *), GFP_KERNEL);

1338 if (!ofs->config.lowerdirs) {

1339 kfree(layers);

1340 goto out_err;

1341 }

1342 ofs->layers = layers;

1343 /*

1344 * Layer 0 is reserved for upper even if there's no upper.

1345 * config.lowerdirs[0] is used for storing the user provided colon

1346 * separated lowerdir string.

1347 */

1348 ofs->config.lowerdirs[0] = ctx->lowerdir_all;

1349 ctx->lowerdir_all = NULL;

1350 ofs->numlayer = 1;

1351

1352 sb->s_stack_depth = 0;

1353 sb->s_maxbytes = MAX_LFS_FILESIZE;

1354 atomic_long_set(&ofs->last_ino, 1);

1355 /* Assume underlying fs uses 32bit inodes unless proven otherwise */

1356 if (ofs->config.xino != OVL_XINO_OFF) {

1357 ofs->xino_mode = BITS_PER_LONG - 32;

1358 if (!ofs->xino_mode) {

1359 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");

1360 ofs->config.xino = OVL_XINO_OFF;

1361 }

1362 }

1363

1364 /* alloc/destroy_inode needed for setting up traps in inode cache */

1365 sb->s_op = &ovl_super_operations;

1366

1367 if (ofs->config.upperdir) {

1368 struct super_block *upper_sb;

1369

1370 err = -EINVAL;

1371 if (!ofs->config.workdir) {

1372 pr_err("missing 'workdir'\n");

1373 goto out_err;

1374 }

1375

1376 err = ovl_get_upper(sb, ofs, &layers[0], &ctx->upper);

1377 if (err)

1378 goto out_err;

1379

1380 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;

1381 if (!ovl_should_sync(ofs)) {

1382 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);

1383 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {

1384 err = -EIO;

1385 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");

1386 goto out_err;

1387 }

1388 }

1389

1390 err = ovl_get_workdir(sb, ofs, &ctx->upper, &ctx->work);

1391 if (err)

1392 goto out_err;

1393

1394 if (!ofs->workdir)

1395 sb->s_flags |= SB_RDONLY;

1396

1397 sb->s_stack_depth = upper_sb->s_stack_depth;

1398 sb->s_time_gran = upper_sb->s_time_gran;

1399 }

1400 oe = ovl_get_lowerstack(sb, ctx, ofs, layers);

1401 err = PTR_ERR(oe);

1402 if (IS_ERR(oe))

1403 goto out_err;

1404

1405 /* If the upper fs is nonexistent, we mark overlayfs r/o too */

1406 if (!ovl_upper_mnt(ofs))

1407 sb->s_flags |= SB_RDONLY;

1408

1409 if (!ovl_origin_uuid(ofs) && ofs->numfs > 1) {

1410 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=null.\n");

1411 ofs->config.uuid = OVL_UUID_NULL;

1412 } else if (ovl_has_fsid(ofs) && ovl_upper_mnt(ofs)) {

1413 /* Use per instance persistent uuid/fsid */

1414 ovl_init_uuid_xattr(sb, ofs, &ctx->upper);

1415 }

1416

1417 if (!ovl_force_readonly(ofs) && ofs->config.index) {

1418 err = ovl_get_indexdir(sb, ofs, oe, &ctx->upper);

1419 if (err)

1420 goto out_free_oe;

1421

1422 /* Force r/o mount with no index dir */

1423 if (!ofs->workdir)

1424 sb->s_flags |= SB_RDONLY;

1425 }

1426

1427 err = ovl_check_overlapping_layers(sb, ofs);

1428 if (err)

1429 goto out_free_oe;

1430

1431 /* Show index=off in /proc/mounts for forced r/o mount */

1432 if (!ofs->workdir) {

1433 ofs->config.index = false;

1434 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {

1435 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");

1436 ofs->config.nfs_export = false;

1437 }

1438 }

1439

1440 if (ofs->config.metacopy && ofs->config.nfs_export) {

1441 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");

1442 ofs->config.nfs_export = false;

1443 }

1444

1445 /*

1446 * Support encoding decodable file handles with nfs_export=on

1447 * and encoding non-decodable file handles with nfs_export=off

1448 * if all layers support file handles.

1449 */

1450 if (ofs->config.nfs_export)

1451 sb->s_export_op = &ovl_export_operations;

1452 else if (!ofs->nofh)

1453 sb->s_export_op = &ovl_export_fid_operations;

1454

1455 /* Never override disk quota limits or use reserved space */

1456 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);

1457

1458 sb->s_magic = OVERLAYFS_SUPER_MAGIC;

1459 sb->s_xattr = ovl_xattr_handlers(ofs);

1460 sb->s_fs_info = ofs;

1461 #ifdef CONFIG_FS_POSIX_ACL

1462 sb->s_flags |= SB_POSIXACL;

1463 #endif

1464 sb->s_iflags |= SB_I_SKIP_SYNC;

1465 /*

1466 * Ensure that umask handling is done by the filesystems used

1467 * for the the upper layer instead of overlayfs as that would

1468 * lead to unexpected results.

1469 */

1470 sb->s_iflags |= SB_I_NOUMASK;

1471 sb->s_iflags |= SB_I_EVM_UNSUPPORTED;

1472

1473 err = -ENOMEM;

1474 root_dentry = ovl_get_root(sb, ctx->upper.dentry, oe);

1475 if (!root_dentry)

1476 goto out_free_oe;

1477

1478 sb->s_root = root_dentry;

1479

1480 return 0;

1481

1482 out_free_oe:

1483 ovl_free_entry(oe);

1484 out_err:

1485 ovl_free_fs(ofs);

1486 sb->s_fs_info = NULL;

1487 return err;

1488 }

1489

1490 struct file_system_type ovl_fs_type = {

1491 .owner = THIS_MODULE,

1492 .name = "overlay",

1493 .init_fs_context = ovl_init_fs_context,

1494 .parameters = ovl_parameter_spec,

1495 .fs_flags = FS_USERNS_MOUNT,

1496 .kill_sb = kill_anon_super,

1497 };

1498 MODULE_ALIAS_FS("overlay");

1499

1500 static void ovl_inode_init_once(void *foo)

1501 {

1502 struct ovl_inode *oi = foo;

1503

1504 inode_init_once(&oi->vfs_inode);

1505 }

1506

1507 static int __init ovl_init(void)

1508 {

1509 int err;

1510

1511 ovl_inode_cachep = kmem_cache_create("ovl_inode",

1512 sizeof(struct ovl_inode), 0,

1513 (SLAB_RECLAIM_ACCOUNT|

1514 SLAB_MEM_SPREAD|SLAB_ACCOUNT),

1515 ovl_inode_init_once);

1516 if (ovl_inode_cachep == NULL)

1517 return -ENOMEM;

1518

1519 err = register_filesystem(&ovl_fs_type);

1520 if (!err)

1521 return 0;

1522

1523 kmem_cache_destroy(ovl_inode_cachep);

1524

1525 return err;

1526 }

1527

1528 static void __exit ovl_exit(void)

1529 {

1530 unregister_filesystem(&ovl_fs_type);

1531

1532 /*

1533 * Make sure all delayed rcu free inodes are flushed before we

1534 * destroy cache.

1535 */

1536 rcu_barrier();

1537 kmem_cache_destroy(ovl_inode_cachep);

1538 }

1539

1540 module_init(ovl_init);

1541 module_exit(ovl_exit);