projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'fs.acl.rework.v6.2' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs...
[linux-block.git] / fs / overlayfs / dir.c
... / ...
CommitLineData
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 *
4 * Copyright (C) 2011 Novell Inc.
5 */
6
7#include <linux/fs.h>
8#include <linux/namei.h>
9#include <linux/xattr.h>
10#include <linux/security.h>
11#include <linux/cred.h>
12#include <linux/module.h>
13#include <linux/posix_acl.h>
14#include <linux/posix_acl_xattr.h>
15#include <linux/atomic.h>
16#include <linux/ratelimit.h>
17#include "overlayfs.h"
18
19static unsigned short ovl_redirect_max = 256;
20module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
21MODULE_PARM_DESC(redirect_max,
22 "Maximum length of absolute redirect xattr value");
23
24static int ovl_set_redirect(struct dentry *dentry, bool samedir);
25
26int ovl_cleanup(struct ovl_fs *ofs, struct inode *wdir, struct dentry *wdentry)
27{
28 int err;
29
30 dget(wdentry);
31 if (d_is_dir(wdentry))
32 err = ovl_do_rmdir(ofs, wdir, wdentry);
33 else
34 err = ovl_do_unlink(ofs, wdir, wdentry);
35 dput(wdentry);
36
37 if (err) {
38 pr_err("cleanup of '%pd2' failed (%i)\n",
39 wdentry, err);
40 }
41
42 return err;
43}
44
45struct dentry *ovl_lookup_temp(struct ovl_fs *ofs, struct dentry *workdir)
46{
47 struct dentry *temp;
48 char name[20];
49 static atomic_t temp_id = ATOMIC_INIT(0);
50
51 /* counter is allowed to wrap, since temp dentries are ephemeral */
52 snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
53
54 temp = ovl_lookup_upper(ofs, name, workdir, strlen(name));
55 if (!IS_ERR(temp) && temp->d_inode) {
56 pr_err("workdir/%s already exists\n", name);
57 dput(temp);
58 temp = ERR_PTR(-EIO);
59 }
60
61 return temp;
62}
63
64/* caller holds i_mutex on workdir */
65static struct dentry *ovl_whiteout(struct ovl_fs *ofs)
66{
67 int err;
68 struct dentry *whiteout;
69 struct dentry *workdir = ofs->workdir;
70 struct inode *wdir = workdir->d_inode;
71
72 if (!ofs->whiteout) {
73 whiteout = ovl_lookup_temp(ofs, workdir);
74 if (IS_ERR(whiteout))
75 goto out;
76
77 err = ovl_do_whiteout(ofs, wdir, whiteout);
78 if (err) {
79 dput(whiteout);
80 whiteout = ERR_PTR(err);
81 goto out;
82 }
83 ofs->whiteout = whiteout;
84 }
85
86 if (ofs->share_whiteout) {
87 whiteout = ovl_lookup_temp(ofs, workdir);
88 if (IS_ERR(whiteout))
89 goto out;
90
91 err = ovl_do_link(ofs, ofs->whiteout, wdir, whiteout);
92 if (!err)
93 goto out;
94
95 if (err != -EMLINK) {
96 pr_warn("Failed to link whiteout - disabling whiteout inode sharing(nlink=%u, err=%i)\n",
97 ofs->whiteout->d_inode->i_nlink, err);
98 ofs->share_whiteout = false;
99 }
100 dput(whiteout);
101 }
102 whiteout = ofs->whiteout;
103 ofs->whiteout = NULL;
104out:
105 return whiteout;
106}
107
108/* Caller must hold i_mutex on both workdir and dir */
109int ovl_cleanup_and_whiteout(struct ovl_fs *ofs, struct inode *dir,
110 struct dentry *dentry)
111{
112 struct inode *wdir = ofs->workdir->d_inode;
113 struct dentry *whiteout;
114 int err;
115 int flags = 0;
116
117 whiteout = ovl_whiteout(ofs);
118 err = PTR_ERR(whiteout);
119 if (IS_ERR(whiteout))
120 return err;
121
122 if (d_is_dir(dentry))
123 flags = RENAME_EXCHANGE;
124
125 err = ovl_do_rename(ofs, wdir, whiteout, dir, dentry, flags);
126 if (err)
127 goto kill_whiteout;
128 if (flags)
129 ovl_cleanup(ofs, wdir, dentry);
130
131out:
132 dput(whiteout);
133 return err;
134
135kill_whiteout:
136 ovl_cleanup(ofs, wdir, whiteout);
137 goto out;
138}
139
140int ovl_mkdir_real(struct ovl_fs *ofs, struct inode *dir,
141 struct dentry **newdentry, umode_t mode)
142{
143 int err;
144 struct dentry *d, *dentry = *newdentry;
145
146 err = ovl_do_mkdir(ofs, dir, dentry, mode);
147 if (err)
148 return err;
149
150 if (likely(!d_unhashed(dentry)))
151 return 0;
152
153 /*
154 * vfs_mkdir() may succeed and leave the dentry passed
155 * to it unhashed and negative. If that happens, try to
156 * lookup a new hashed and positive dentry.
157 */
158 d = ovl_lookup_upper(ofs, dentry->d_name.name, dentry->d_parent,
159 dentry->d_name.len);
160 if (IS_ERR(d)) {
161 pr_warn("failed lookup after mkdir (%pd2, err=%i).\n",
162 dentry, err);
163 return PTR_ERR(d);
164 }
165 dput(dentry);
166 *newdentry = d;
167
168 return 0;
169}
170
171struct dentry *ovl_create_real(struct ovl_fs *ofs, struct inode *dir,
172 struct dentry *newdentry, struct ovl_cattr *attr)
173{
174 int err;
175
176 if (IS_ERR(newdentry))
177 return newdentry;
178
179 err = -ESTALE;
180 if (newdentry->d_inode)
181 goto out;
182
183 if (attr->hardlink) {
184 err = ovl_do_link(ofs, attr->hardlink, dir, newdentry);
185 } else {
186 switch (attr->mode & S_IFMT) {
187 case S_IFREG:
188 err = ovl_do_create(ofs, dir, newdentry, attr->mode);
189 break;
190
191 case S_IFDIR:
192 /* mkdir is special... */
193 err = ovl_mkdir_real(ofs, dir, &newdentry, attr->mode);
194 break;
195
196 case S_IFCHR:
197 case S_IFBLK:
198 case S_IFIFO:
199 case S_IFSOCK:
200 err = ovl_do_mknod(ofs, dir, newdentry, attr->mode,
201 attr->rdev);
202 break;
203
204 case S_IFLNK:
205 err = ovl_do_symlink(ofs, dir, newdentry, attr->link);
206 break;
207
208 default:
209 err = -EPERM;
210 }
211 }
212 if (!err && WARN_ON(!newdentry->d_inode)) {
213 /*
214 * Not quite sure if non-instantiated dentry is legal or not.
215 * VFS doesn't seem to care so check and warn here.
216 */
217 err = -EIO;
218 }
219out:
220 if (err) {
221 dput(newdentry);
222 return ERR_PTR(err);
223 }
224 return newdentry;
225}
226
227struct dentry *ovl_create_temp(struct ovl_fs *ofs, struct dentry *workdir,
228 struct ovl_cattr *attr)
229{
230 return ovl_create_real(ofs, d_inode(workdir),
231 ovl_lookup_temp(ofs, workdir), attr);
232}
233
234static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
235 int xerr)
236{
237 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
238 int err;
239
240 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
241 if (!err)
242 ovl_dentry_set_opaque(dentry);
243
244 return err;
245}
246
247static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
248{
249 /*
250 * Fail with -EIO when trying to create opaque dir and upper doesn't
251 * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
252 * return a specific error for noxattr case.
253 */
254 return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
255}
256
257/*
258 * Common operations required to be done after creation of file on upper.
259 * If @hardlink is false, then @inode is a pre-allocated inode, we may or
260 * may not use to instantiate the new dentry.
261 */
262static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
263 struct dentry *newdentry, bool hardlink)
264{
265 struct ovl_inode_params oip = {
266 .upperdentry = newdentry,
267 .newinode = inode,
268 };
269
270 ovl_dir_modified(dentry->d_parent, false);
271 ovl_dentry_set_upper_alias(dentry);
272 ovl_dentry_update_reval(dentry, newdentry,
273 DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
274
275 if (!hardlink) {
276 /*
277 * ovl_obtain_alias() can be called after ovl_create_real()
278 * and before we get here, so we may get an inode from cache
279 * with the same real upperdentry that is not the inode we
280 * pre-allocated. In this case we will use the cached inode
281 * to instantiate the new dentry.
282 *
283 * XXX: if we ever use ovl_obtain_alias() to decode directory
284 * file handles, need to use ovl_get_inode_locked() and
285 * d_instantiate_new() here to prevent from creating two
286 * hashed directory inode aliases.
287 */
288 inode = ovl_get_inode(dentry->d_sb, &oip);
289 if (IS_ERR(inode))
290 return PTR_ERR(inode);
291 if (inode == oip.newinode)
292 ovl_set_flag(OVL_UPPERDATA, inode);
293 } else {
294 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
295 dput(newdentry);
296 inc_nlink(inode);
297 }
298
299 d_instantiate(dentry, inode);
300 if (inode != oip.newinode) {
301 pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
302 dentry);
303 }
304
305 /* Force lookup of new upper hardlink to find its lower */
306 if (hardlink)
307 d_drop(dentry);
308
309 return 0;
310}
311
312static bool ovl_type_merge(struct dentry *dentry)
313{
314 return OVL_TYPE_MERGE(ovl_path_type(dentry));
315}
316
317static bool ovl_type_origin(struct dentry *dentry)
318{
319 return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
320}
321
322static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
323 struct ovl_cattr *attr)
324{
325 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
326 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
327 struct inode *udir = upperdir->d_inode;
328 struct dentry *newdentry;
329 int err;
330
331 if (!attr->hardlink && !IS_POSIXACL(udir))
332 attr->mode &= ~current_umask();
333
334 inode_lock_nested(udir, I_MUTEX_PARENT);
335 newdentry = ovl_create_real(ofs, udir,
336 ovl_lookup_upper(ofs, dentry->d_name.name,
337 upperdir, dentry->d_name.len),
338 attr);
339 err = PTR_ERR(newdentry);
340 if (IS_ERR(newdentry))
341 goto out_unlock;
342
343 if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
344 !ovl_allow_offline_changes(ofs)) {
345 /* Setting opaque here is just an optimization, allow to fail */
346 ovl_set_opaque(dentry, newdentry);
347 }
348
349 err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
350 if (err)
351 goto out_cleanup;
352out_unlock:
353 inode_unlock(udir);
354 return err;
355
356out_cleanup:
357 ovl_cleanup(ofs, udir, newdentry);
358 dput(newdentry);
359 goto out_unlock;
360}
361
362static struct dentry *ovl_clear_empty(struct dentry *dentry,
363 struct list_head *list)
364{
365 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
366 struct dentry *workdir = ovl_workdir(dentry);
367 struct inode *wdir = workdir->d_inode;
368 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
369 struct inode *udir = upperdir->d_inode;
370 struct path upperpath;
371 struct dentry *upper;
372 struct dentry *opaquedir;
373 struct kstat stat;
374 int err;
375
376 if (WARN_ON(!workdir))
377 return ERR_PTR(-EROFS);
378
379 err = ovl_lock_rename_workdir(workdir, upperdir);
380 if (err)
381 goto out;
382
383 ovl_path_upper(dentry, &upperpath);
384 err = vfs_getattr(&upperpath, &stat,
385 STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
386 if (err)
387 goto out_unlock;
388
389 err = -ESTALE;
390 if (!S_ISDIR(stat.mode))
391 goto out_unlock;
392 upper = upperpath.dentry;
393 if (upper->d_parent->d_inode != udir)
394 goto out_unlock;
395
396 opaquedir = ovl_create_temp(ofs, workdir, OVL_CATTR(stat.mode));
397 err = PTR_ERR(opaquedir);
398 if (IS_ERR(opaquedir))
399 goto out_unlock;
400
401 err = ovl_copy_xattr(dentry->d_sb, &upperpath, opaquedir);
402 if (err)
403 goto out_cleanup;
404
405 err = ovl_set_opaque(dentry, opaquedir);
406 if (err)
407 goto out_cleanup;
408
409 inode_lock(opaquedir->d_inode);
410 err = ovl_set_attr(ofs, opaquedir, &stat);
411 inode_unlock(opaquedir->d_inode);
412 if (err)
413 goto out_cleanup;
414
415 err = ovl_do_rename(ofs, wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
416 if (err)
417 goto out_cleanup;
418
419 ovl_cleanup_whiteouts(ofs, upper, list);
420 ovl_cleanup(ofs, wdir, upper);
421 unlock_rename(workdir, upperdir);
422
423 /* dentry's upper doesn't match now, get rid of it */
424 d_drop(dentry);
425
426 return opaquedir;
427
428out_cleanup:
429 ovl_cleanup(ofs, wdir, opaquedir);
430 dput(opaquedir);
431out_unlock:
432 unlock_rename(workdir, upperdir);
433out:
434 return ERR_PTR(err);
435}
436
437static int ovl_set_upper_acl(struct ovl_fs *ofs, struct dentry *upperdentry,
438 const char *acl_name, struct posix_acl *acl)
439{
440 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
441 return 0;
442
443 return ovl_do_set_acl(ofs, upperdentry, acl_name, acl);
444}
445
446static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
447 struct ovl_cattr *cattr)
448{
449 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
450 struct dentry *workdir = ovl_workdir(dentry);
451 struct inode *wdir = workdir->d_inode;
452 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
453 struct inode *udir = upperdir->d_inode;
454 struct dentry *upper;
455 struct dentry *newdentry;
456 int err;
457 struct posix_acl *acl, *default_acl;
458 bool hardlink = !!cattr->hardlink;
459
460 if (WARN_ON(!workdir))
461 return -EROFS;
462
463 if (!hardlink) {
464 err = posix_acl_create(dentry->d_parent->d_inode,
465 &cattr->mode, &default_acl, &acl);
466 if (err)
467 return err;
468 }
469
470 err = ovl_lock_rename_workdir(workdir, upperdir);
471 if (err)
472 goto out;
473
474 upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
475 dentry->d_name.len);
476 err = PTR_ERR(upper);
477 if (IS_ERR(upper))
478 goto out_unlock;
479
480 err = -ESTALE;
481 if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
482 goto out_dput;
483
484 newdentry = ovl_create_temp(ofs, workdir, cattr);
485 err = PTR_ERR(newdentry);
486 if (IS_ERR(newdentry))
487 goto out_dput;
488
489 /*
490 * mode could have been mutilated due to umask (e.g. sgid directory)
491 */
492 if (!hardlink &&
493 !S_ISLNK(cattr->mode) &&
494 newdentry->d_inode->i_mode != cattr->mode) {
495 struct iattr attr = {
496 .ia_valid = ATTR_MODE,
497 .ia_mode = cattr->mode,
498 };
499 inode_lock(newdentry->d_inode);
500 err = ovl_do_notify_change(ofs, newdentry, &attr);
501 inode_unlock(newdentry->d_inode);
502 if (err)
503 goto out_cleanup;
504 }
505 if (!hardlink) {
506 err = ovl_set_upper_acl(ofs, newdentry,
507 XATTR_NAME_POSIX_ACL_ACCESS, acl);
508 if (err)
509 goto out_cleanup;
510
511 err = ovl_set_upper_acl(ofs, newdentry,
512 XATTR_NAME_POSIX_ACL_DEFAULT, default_acl);
513 if (err)
514 goto out_cleanup;
515 }
516
517 if (!hardlink && S_ISDIR(cattr->mode)) {
518 err = ovl_set_opaque(dentry, newdentry);
519 if (err)
520 goto out_cleanup;
521
522 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper,
523 RENAME_EXCHANGE);
524 if (err)
525 goto out_cleanup;
526
527 ovl_cleanup(ofs, wdir, upper);
528 } else {
529 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper, 0);
530 if (err)
531 goto out_cleanup;
532 }
533 err = ovl_instantiate(dentry, inode, newdentry, hardlink);
534 if (err) {
535 ovl_cleanup(ofs, udir, newdentry);
536 dput(newdentry);
537 }
538out_dput:
539 dput(upper);
540out_unlock:
541 unlock_rename(workdir, upperdir);
542out:
543 if (!hardlink) {
544 posix_acl_release(acl);
545 posix_acl_release(default_acl);
546 }
547 return err;
548
549out_cleanup:
550 ovl_cleanup(ofs, wdir, newdentry);
551 dput(newdentry);
552 goto out_dput;
553}
554
555static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
556 struct ovl_cattr *attr, bool origin)
557{
558 int err;
559 const struct cred *old_cred;
560 struct cred *override_cred;
561 struct dentry *parent = dentry->d_parent;
562
563 err = ovl_copy_up(parent);
564 if (err)
565 return err;
566
567 old_cred = ovl_override_creds(dentry->d_sb);
568
569 /*
570 * When linking a file with copy up origin into a new parent, mark the
571 * new parent dir "impure".
572 */
573 if (origin) {
574 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
575 if (err)
576 goto out_revert_creds;
577 }
578
579 err = -ENOMEM;
580 override_cred = prepare_creds();
581 if (override_cred) {
582 override_cred->fsuid = inode->i_uid;
583 override_cred->fsgid = inode->i_gid;
584 if (!attr->hardlink) {
585 err = security_dentry_create_files_as(dentry,
586 attr->mode, &dentry->d_name, old_cred,
587 override_cred);
588 if (err) {
589 put_cred(override_cred);
590 goto out_revert_creds;
591 }
592 }
593 put_cred(override_creds(override_cred));
594 put_cred(override_cred);
595
596 if (!ovl_dentry_is_whiteout(dentry))
597 err = ovl_create_upper(dentry, inode, attr);
598 else
599 err = ovl_create_over_whiteout(dentry, inode, attr);
600 }
601out_revert_creds:
602 revert_creds(old_cred);
603 return err;
604}
605
606static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
607 const char *link)
608{
609 int err;
610 struct inode *inode;
611 struct ovl_cattr attr = {
612 .rdev = rdev,
613 .link = link,
614 };
615
616 err = ovl_want_write(dentry);
617 if (err)
618 goto out;
619
620 /* Preallocate inode to be used by ovl_get_inode() */
621 err = -ENOMEM;
622 inode = ovl_new_inode(dentry->d_sb, mode, rdev);
623 if (!inode)
624 goto out_drop_write;
625
626 spin_lock(&inode->i_lock);
627 inode->i_state |= I_CREATING;
628 spin_unlock(&inode->i_lock);
629
630 inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
631 attr.mode = inode->i_mode;
632
633 err = ovl_create_or_link(dentry, inode, &attr, false);
634 /* Did we end up using the preallocated inode? */
635 if (inode != d_inode(dentry))
636 iput(inode);
637
638out_drop_write:
639 ovl_drop_write(dentry);
640out:
641 return err;
642}
643
644static int ovl_create(struct user_namespace *mnt_userns, struct inode *dir,
645 struct dentry *dentry, umode_t mode, bool excl)
646{
647 return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
648}
649
650static int ovl_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
651 struct dentry *dentry, umode_t mode)
652{
653 return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
654}
655
656static int ovl_mknod(struct user_namespace *mnt_userns, struct inode *dir,
657 struct dentry *dentry, umode_t mode, dev_t rdev)
658{
659 /* Don't allow creation of "whiteout" on overlay */
660 if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
661 return -EPERM;
662
663 return ovl_create_object(dentry, mode, rdev, NULL);
664}
665
666static int ovl_symlink(struct user_namespace *mnt_userns, struct inode *dir,
667 struct dentry *dentry, const char *link)
668{
669 return ovl_create_object(dentry, S_IFLNK, 0, link);
670}
671
672static int ovl_set_link_redirect(struct dentry *dentry)
673{
674 const struct cred *old_cred;
675 int err;
676
677 old_cred = ovl_override_creds(dentry->d_sb);
678 err = ovl_set_redirect(dentry, false);
679 revert_creds(old_cred);
680
681 return err;
682}
683
684static int ovl_link(struct dentry *old, struct inode *newdir,
685 struct dentry *new)
686{
687 int err;
688 struct inode *inode;
689
690 err = ovl_want_write(old);
691 if (err)
692 goto out;
693
694 err = ovl_copy_up(old);
695 if (err)
696 goto out_drop_write;
697
698 err = ovl_copy_up(new->d_parent);
699 if (err)
700 goto out_drop_write;
701
702 if (ovl_is_metacopy_dentry(old)) {
703 err = ovl_set_link_redirect(old);
704 if (err)
705 goto out_drop_write;
706 }
707
708 err = ovl_nlink_start(old);
709 if (err)
710 goto out_drop_write;
711
712 inode = d_inode(old);
713 ihold(inode);
714
715 err = ovl_create_or_link(new, inode,
716 &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
717 ovl_type_origin(old));
718 if (err)
719 iput(inode);
720
721 ovl_nlink_end(old);
722out_drop_write:
723 ovl_drop_write(old);
724out:
725 return err;
726}
727
728static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
729{
730 return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
731}
732
733static int ovl_remove_and_whiteout(struct dentry *dentry,
734 struct list_head *list)
735{
736 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
737 struct dentry *workdir = ovl_workdir(dentry);
738 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
739 struct dentry *upper;
740 struct dentry *opaquedir = NULL;
741 int err;
742
743 if (WARN_ON(!workdir))
744 return -EROFS;
745
746 if (!list_empty(list)) {
747 opaquedir = ovl_clear_empty(dentry, list);
748 err = PTR_ERR(opaquedir);
749 if (IS_ERR(opaquedir))
750 goto out;
751 }
752
753 err = ovl_lock_rename_workdir(workdir, upperdir);
754 if (err)
755 goto out_dput;
756
757 upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
758 dentry->d_name.len);
759 err = PTR_ERR(upper);
760 if (IS_ERR(upper))
761 goto out_unlock;
762
763 err = -ESTALE;
764 if ((opaquedir && upper != opaquedir) ||
765 (!opaquedir && ovl_dentry_upper(dentry) &&
766 !ovl_matches_upper(dentry, upper))) {
767 goto out_dput_upper;
768 }
769
770 err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
771 if (err)
772 goto out_d_drop;
773
774 ovl_dir_modified(dentry->d_parent, true);
775out_d_drop:
776 d_drop(dentry);
777out_dput_upper:
778 dput(upper);
779out_unlock:
780 unlock_rename(workdir, upperdir);
781out_dput:
782 dput(opaquedir);
783out:
784 return err;
785}
786
787static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
788 struct list_head *list)
789{
790 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
791 struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
792 struct inode *dir = upperdir->d_inode;
793 struct dentry *upper;
794 struct dentry *opaquedir = NULL;
795 int err;
796
797 if (!list_empty(list)) {
798 opaquedir = ovl_clear_empty(dentry, list);
799 err = PTR_ERR(opaquedir);
800 if (IS_ERR(opaquedir))
801 goto out;
802 }
803
804 inode_lock_nested(dir, I_MUTEX_PARENT);
805 upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
806 dentry->d_name.len);
807 err = PTR_ERR(upper);
808 if (IS_ERR(upper))
809 goto out_unlock;
810
811 err = -ESTALE;
812 if ((opaquedir && upper != opaquedir) ||
813 (!opaquedir && !ovl_matches_upper(dentry, upper)))
814 goto out_dput_upper;
815
816 if (is_dir)
817 err = ovl_do_rmdir(ofs, dir, upper);
818 else
819 err = ovl_do_unlink(ofs, dir, upper);
820 ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
821
822 /*
823 * Keeping this dentry hashed would mean having to release
824 * upperpath/lowerpath, which could only be done if we are the
825 * sole user of this dentry. Too tricky... Just unhash for
826 * now.
827 */
828 if (!err)
829 d_drop(dentry);
830out_dput_upper:
831 dput(upper);
832out_unlock:
833 inode_unlock(dir);
834 dput(opaquedir);
835out:
836 return err;
837}
838
839static bool ovl_pure_upper(struct dentry *dentry)
840{
841 return !ovl_dentry_lower(dentry) &&
842 !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
843}
844
845static void ovl_drop_nlink(struct dentry *dentry)
846{
847 struct inode *inode = d_inode(dentry);
848 struct dentry *alias;
849
850 /* Try to find another, hashed alias */
851 spin_lock(&inode->i_lock);
852 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
853 if (alias != dentry && !d_unhashed(alias))
854 break;
855 }
856 spin_unlock(&inode->i_lock);
857
858 /*
859 * Changes to underlying layers may cause i_nlink to lose sync with
860 * reality. In this case prevent the link count from going to zero
861 * prematurely.
862 */
863 if (inode->i_nlink > !!alias)
864 drop_nlink(inode);
865}
866
867static int ovl_do_remove(struct dentry *dentry, bool is_dir)
868{
869 int err;
870 const struct cred *old_cred;
871 bool lower_positive = ovl_lower_positive(dentry);
872 LIST_HEAD(list);
873
874 /* No need to clean pure upper removed by vfs_rmdir() */
875 if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
876 err = ovl_check_empty_dir(dentry, &list);
877 if (err)
878 goto out;
879 }
880
881 err = ovl_want_write(dentry);
882 if (err)
883 goto out;
884
885 err = ovl_copy_up(dentry->d_parent);
886 if (err)
887 goto out_drop_write;
888
889 err = ovl_nlink_start(dentry);
890 if (err)
891 goto out_drop_write;
892
893 old_cred = ovl_override_creds(dentry->d_sb);
894 if (!lower_positive)
895 err = ovl_remove_upper(dentry, is_dir, &list);
896 else
897 err = ovl_remove_and_whiteout(dentry, &list);
898 revert_creds(old_cred);
899 if (!err) {
900 if (is_dir)
901 clear_nlink(dentry->d_inode);
902 else
903 ovl_drop_nlink(dentry);
904 }
905 ovl_nlink_end(dentry);
906
907 /*
908 * Copy ctime
909 *
910 * Note: we fail to update ctime if there was no copy-up, only a
911 * whiteout
912 */
913 if (ovl_dentry_upper(dentry))
914 ovl_copyattr(d_inode(dentry));
915
916out_drop_write:
917 ovl_drop_write(dentry);
918out:
919 ovl_cache_free(&list);
920 return err;
921}
922
923static int ovl_unlink(struct inode *dir, struct dentry *dentry)
924{
925 return ovl_do_remove(dentry, false);
926}
927
928static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
929{
930 return ovl_do_remove(dentry, true);
931}
932
933static bool ovl_type_merge_or_lower(struct dentry *dentry)
934{
935 enum ovl_path_type type = ovl_path_type(dentry);
936
937 return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
938}
939
940static bool ovl_can_move(struct dentry *dentry)
941{
942 return ovl_redirect_dir(dentry->d_sb) ||
943 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
944}
945
946static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
947{
948 char *buf, *ret;
949 struct dentry *d, *tmp;
950 int buflen = ovl_redirect_max + 1;
951
952 if (!abs_redirect) {
953 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
954 GFP_KERNEL);
955 goto out;
956 }
957
958 buf = ret = kmalloc(buflen, GFP_KERNEL);
959 if (!buf)
960 goto out;
961
962 buflen--;
963 buf[buflen] = '\0';
964 for (d = dget(dentry); !IS_ROOT(d);) {
965 const char *name;
966 int thislen;
967
968 spin_lock(&d->d_lock);
969 name = ovl_dentry_get_redirect(d);
970 if (name) {
971 thislen = strlen(name);
972 } else {
973 name = d->d_name.name;
974 thislen = d->d_name.len;
975 }
976
977 /* If path is too long, fall back to userspace move */
978 if (thislen + (name[0] != '/') > buflen) {
979 ret = ERR_PTR(-EXDEV);
980 spin_unlock(&d->d_lock);
981 goto out_put;
982 }
983
984 buflen -= thislen;
985 memcpy(&buf[buflen], name, thislen);
986 spin_unlock(&d->d_lock);
987 tmp = dget_parent(d);
988
989 dput(d);
990 d = tmp;
991
992 /* Absolute redirect: finished */
993 if (buf[buflen] == '/')
994 break;
995 buflen--;
996 buf[buflen] = '/';
997 }
998 ret = kstrdup(&buf[buflen], GFP_KERNEL);
999out_put:
1000 dput(d);
1001 kfree(buf);
1002out:
1003 return ret ? ret : ERR_PTR(-ENOMEM);
1004}
1005
1006static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1007{
1008 struct dentry *lowerdentry;
1009
1010 if (!samedir)
1011 return true;
1012
1013 if (d_is_dir(dentry))
1014 return false;
1015
1016 /*
1017 * For non-dir hardlinked files, we need absolute redirects
1018 * in general as two upper hardlinks could be in different
1019 * dirs. We could put a relative redirect now and convert
1020 * it to absolute redirect later. But when nlink > 1 and
1021 * indexing is on, that means relative redirect needs to be
1022 * converted to absolute during copy up of another lower
1023 * hardllink as well.
1024 *
1025 * So without optimizing too much, just check if lower is
1026 * a hard link or not. If lower is hard link, put absolute
1027 * redirect.
1028 */
1029 lowerdentry = ovl_dentry_lower(dentry);
1030 return (d_inode(lowerdentry)->i_nlink > 1);
1031}
1032
1033static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1034{
1035 int err;
1036 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1037 const char *redirect = ovl_dentry_get_redirect(dentry);
1038 bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1039
1040 if (redirect && (!absolute_redirect || redirect[0] == '/'))
1041 return 0;
1042
1043 redirect = ovl_get_redirect(dentry, absolute_redirect);
1044 if (IS_ERR(redirect))
1045 return PTR_ERR(redirect);
1046
1047 err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1048 OVL_XATTR_REDIRECT,
1049 redirect, strlen(redirect), -EXDEV);
1050 if (!err) {
1051 spin_lock(&dentry->d_lock);
1052 ovl_dentry_set_redirect(dentry, redirect);
1053 spin_unlock(&dentry->d_lock);
1054 } else {
1055 kfree(redirect);
1056 pr_warn_ratelimited("failed to set redirect (%i)\n",
1057 err);
1058 /* Fall back to userspace copy-up */
1059 err = -EXDEV;
1060 }
1061 return err;
1062}
1063
1064static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
1065 struct dentry *old, struct inode *newdir,
1066 struct dentry *new, unsigned int flags)
1067{
1068 int err;
1069 struct dentry *old_upperdir;
1070 struct dentry *new_upperdir;
1071 struct dentry *olddentry;
1072 struct dentry *newdentry;
1073 struct dentry *trap;
1074 bool old_opaque;
1075 bool new_opaque;
1076 bool cleanup_whiteout = false;
1077 bool update_nlink = false;
1078 bool overwrite = !(flags & RENAME_EXCHANGE);
1079 bool is_dir = d_is_dir(old);
1080 bool new_is_dir = d_is_dir(new);
1081 bool samedir = olddir == newdir;
1082 struct dentry *opaquedir = NULL;
1083 const struct cred *old_cred = NULL;
1084 struct ovl_fs *ofs = OVL_FS(old->d_sb);
1085 LIST_HEAD(list);
1086
1087 err = -EINVAL;
1088 if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1089 goto out;
1090
1091 flags &= ~RENAME_NOREPLACE;
1092
1093 /* Don't copy up directory trees */
1094 err = -EXDEV;
1095 if (!ovl_can_move(old))
1096 goto out;
1097 if (!overwrite && !ovl_can_move(new))
1098 goto out;
1099
1100 if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1101 err = ovl_check_empty_dir(new, &list);
1102 if (err)
1103 goto out;
1104 }
1105
1106 if (overwrite) {
1107 if (ovl_lower_positive(old)) {
1108 if (!ovl_dentry_is_whiteout(new)) {
1109 /* Whiteout source */
1110 flags |= RENAME_WHITEOUT;
1111 } else {
1112 /* Switch whiteouts */
1113 flags |= RENAME_EXCHANGE;
1114 }
1115 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1116 flags |= RENAME_EXCHANGE;
1117 cleanup_whiteout = true;
1118 }
1119 }
1120
1121 err = ovl_want_write(old);
1122 if (err)
1123 goto out;
1124
1125 err = ovl_copy_up(old);
1126 if (err)
1127 goto out_drop_write;
1128
1129 err = ovl_copy_up(new->d_parent);
1130 if (err)
1131 goto out_drop_write;
1132 if (!overwrite) {
1133 err = ovl_copy_up(new);
1134 if (err)
1135 goto out_drop_write;
1136 } else if (d_inode(new)) {
1137 err = ovl_nlink_start(new);
1138 if (err)
1139 goto out_drop_write;
1140
1141 update_nlink = true;
1142 }
1143
1144 old_cred = ovl_override_creds(old->d_sb);
1145
1146 if (!list_empty(&list)) {
1147 opaquedir = ovl_clear_empty(new, &list);
1148 err = PTR_ERR(opaquedir);
1149 if (IS_ERR(opaquedir)) {
1150 opaquedir = NULL;
1151 goto out_revert_creds;
1152 }
1153 }
1154
1155 old_upperdir = ovl_dentry_upper(old->d_parent);
1156 new_upperdir = ovl_dentry_upper(new->d_parent);
1157
1158 if (!samedir) {
1159 /*
1160 * When moving a merge dir or non-dir with copy up origin into
1161 * a new parent, we are marking the new parent dir "impure".
1162 * When ovl_iterate() iterates an "impure" upper dir, it will
1163 * lookup the origin inodes of the entries to fill d_ino.
1164 */
1165 if (ovl_type_origin(old)) {
1166 err = ovl_set_impure(new->d_parent, new_upperdir);
1167 if (err)
1168 goto out_revert_creds;
1169 }
1170 if (!overwrite && ovl_type_origin(new)) {
1171 err = ovl_set_impure(old->d_parent, old_upperdir);
1172 if (err)
1173 goto out_revert_creds;
1174 }
1175 }
1176
1177 trap = lock_rename(new_upperdir, old_upperdir);
1178
1179 olddentry = ovl_lookup_upper(ofs, old->d_name.name, old_upperdir,
1180 old->d_name.len);
1181 err = PTR_ERR(olddentry);
1182 if (IS_ERR(olddentry))
1183 goto out_unlock;
1184
1185 err = -ESTALE;
1186 if (!ovl_matches_upper(old, olddentry))
1187 goto out_dput_old;
1188
1189 newdentry = ovl_lookup_upper(ofs, new->d_name.name, new_upperdir,
1190 new->d_name.len);
1191 err = PTR_ERR(newdentry);
1192 if (IS_ERR(newdentry))
1193 goto out_dput_old;
1194
1195 old_opaque = ovl_dentry_is_opaque(old);
1196 new_opaque = ovl_dentry_is_opaque(new);
1197
1198 err = -ESTALE;
1199 if (d_inode(new) && ovl_dentry_upper(new)) {
1200 if (opaquedir) {
1201 if (newdentry != opaquedir)
1202 goto out_dput;
1203 } else {
1204 if (!ovl_matches_upper(new, newdentry))
1205 goto out_dput;
1206 }
1207 } else {
1208 if (!d_is_negative(newdentry)) {
1209 if (!new_opaque || !ovl_is_whiteout(newdentry))
1210 goto out_dput;
1211 } else {
1212 if (flags & RENAME_EXCHANGE)
1213 goto out_dput;
1214 }
1215 }
1216
1217 if (olddentry == trap)
1218 goto out_dput;
1219 if (newdentry == trap)
1220 goto out_dput;
1221
1222 if (olddentry->d_inode == newdentry->d_inode)
1223 goto out_dput;
1224
1225 err = 0;
1226 if (ovl_type_merge_or_lower(old))
1227 err = ovl_set_redirect(old, samedir);
1228 else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1229 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1230 if (err)
1231 goto out_dput;
1232
1233 if (!overwrite && ovl_type_merge_or_lower(new))
1234 err = ovl_set_redirect(new, samedir);
1235 else if (!overwrite && new_is_dir && !new_opaque &&
1236 ovl_type_merge(old->d_parent))
1237 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1238 if (err)
1239 goto out_dput;
1240
1241 err = ovl_do_rename(ofs, old_upperdir->d_inode, olddentry,
1242 new_upperdir->d_inode, newdentry, flags);
1243 if (err)
1244 goto out_dput;
1245
1246 if (cleanup_whiteout)
1247 ovl_cleanup(ofs, old_upperdir->d_inode, newdentry);
1248
1249 if (overwrite && d_inode(new)) {
1250 if (new_is_dir)
1251 clear_nlink(d_inode(new));
1252 else
1253 ovl_drop_nlink(new);
1254 }
1255
1256 ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1257 (!overwrite && ovl_type_origin(new)));
1258 ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1259 (d_inode(new) && ovl_type_origin(new)));
1260
1261 /* copy ctime: */
1262 ovl_copyattr(d_inode(old));
1263 if (d_inode(new) && ovl_dentry_upper(new))
1264 ovl_copyattr(d_inode(new));
1265
1266out_dput:
1267 dput(newdentry);
1268out_dput_old:
1269 dput(olddentry);
1270out_unlock:
1271 unlock_rename(new_upperdir, old_upperdir);
1272out_revert_creds:
1273 revert_creds(old_cred);
1274 if (update_nlink)
1275 ovl_nlink_end(new);
1276out_drop_write:
1277 ovl_drop_write(old);
1278out:
1279 dput(opaquedir);
1280 ovl_cache_free(&list);
1281 return err;
1282}
1283
1284const struct inode_operations ovl_dir_inode_operations = {
1285 .lookup = ovl_lookup,
1286 .mkdir = ovl_mkdir,
1287 .symlink = ovl_symlink,
1288 .unlink = ovl_unlink,
1289 .rmdir = ovl_rmdir,
1290 .rename = ovl_rename,
1291 .link = ovl_link,
1292 .setattr = ovl_setattr,
1293 .create = ovl_create,
1294 .mknod = ovl_mknod,
1295 .permission = ovl_permission,
1296 .getattr = ovl_getattr,
1297 .listxattr = ovl_listxattr,
1298 .get_inode_acl = ovl_get_inode_acl,
1299 .get_acl = ovl_get_acl,
1300 .set_acl = ovl_set_acl,
1301 .update_time = ovl_update_time,
1302 .fileattr_get = ovl_fileattr_get,
1303 .fileattr_set = ovl_fileattr_set,
1304};
Linux 6.x block layer and io_uring tree(s)