1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2021-2024 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans.h"
14 #include "xfs_inode.h"
15 #include "xfs_ialloc.h"
16 #include "xfs_quota.h"
17 #include "xfs_trans_space.h"
18 #include "xfs_dir2.h"
19 #include "xfs_icache.h"
20 #include "xfs_bmap.h"
21 #include "xfs_bmap_btree.h"
22 #include "xfs_parent.h"
23 #include "xfs_attr_sf.h"
24 #include "scrub/scrub.h"
25 #include "scrub/common.h"
26 #include "scrub/repair.h"
27 #include "scrub/trace.h"
28 #include "scrub/orphanage.h"
29 #include "scrub/readdir.h"
30
31 #include <linux/namei.h>
32
33 /*
34 * The Orphanage
35 * =============
36 *
37 * If the directory tree is damaged, children of that directory become
38 * inaccessible via that file path. If a child has no other parents, the file
39 * is said to be orphaned. xfs_repair fixes this situation by creating a
40 * orphanage directory (specifically, /lost+found) and creating a directory
41 * entry pointing to the orphaned file.
42 *
43 * Online repair follows this tactic by creating a root-owned /lost+found
44 * directory if one does not exist. If an orphan is found, it will move that
45 * files into orphanage.
46 */
47
48 /* Make the orphanage owned by root. */
49 STATIC int
50 xrep_chown_orphanage(
51 struct xfs_scrub *sc,
52 struct xfs_inode *dp)
53 {
54 struct xfs_trans *tp;
55 struct xfs_mount *mp = sc->mp;
56 struct xfs_dquot *udqp = NULL, *gdqp = NULL, *pdqp = NULL;
57 struct xfs_dquot *oldu = NULL, *oldg = NULL, *oldp = NULL;
58 struct inode *inode = VFS_I(dp);
59 int error;
60
61 error = xfs_qm_vop_dqalloc(dp, GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
62 XFS_QMOPT_QUOTALL, &udqp, &gdqp, &pdqp);
63 if (error)
64 return error;
65
66 error = xfs_trans_alloc_ichange(dp, udqp, gdqp, pdqp, true, &tp);
67 if (error)
68 goto out_dqrele;
69
70 /*
71 * Always clear setuid/setgid/sticky on the orphanage since we don't
72 * normally want that functionality on this directory and xfs_repair
73 * doesn't create it this way either. Leave the other access bits
74 * unchanged.
75 */
76 inode->i_mode &= ~(S_ISUID | S_ISGID | S_ISVTX);
77
78 /*
79 * Change the ownerships and register quota modifications
80 * in the transaction.
81 */
82 if (!uid_eq(inode->i_uid, GLOBAL_ROOT_UID)) {
83 if (XFS_IS_UQUOTA_ON(mp))
84 oldu = xfs_qm_vop_chown(tp, dp, &dp->i_udquot, udqp);
85 inode->i_uid = GLOBAL_ROOT_UID;
86 }
87 if (!gid_eq(inode->i_gid, GLOBAL_ROOT_GID)) {
88 if (XFS_IS_GQUOTA_ON(mp))
89 oldg = xfs_qm_vop_chown(tp, dp, &dp->i_gdquot, gdqp);
90 inode->i_gid = GLOBAL_ROOT_GID;
91 }
92 if (dp->i_projid != 0) {
93 if (XFS_IS_PQUOTA_ON(mp))
94 oldp = xfs_qm_vop_chown(tp, dp, &dp->i_pdquot, pdqp);
95 dp->i_projid = 0;
96 }
97
98 dp->i_diflags &= ~(XFS_DIFLAG_REALTIME | XFS_DIFLAG_RTINHERIT);
99 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
100
101 XFS_STATS_INC(mp, xs_ig_attrchg);
102
103 if (xfs_has_wsync(mp))
104 xfs_trans_set_sync(tp);
105 error = xfs_trans_commit(tp);
106
107 xfs_qm_dqrele(oldu);
108 xfs_qm_dqrele(oldg);
109 xfs_qm_dqrele(oldp);
110
111 out_dqrele:
112 xfs_qm_dqrele(udqp);
113 xfs_qm_dqrele(gdqp);
114 xfs_qm_dqrele(pdqp);
115 return error;
116 }
117
118 #define ORPHANAGE "lost+found"
119
120 /* Create the orphanage directory, and set sc->orphanage to it. */
121 int
122 xrep_orphanage_create(
123 struct xfs_scrub *sc)
124 {
125 struct xfs_mount *mp = sc->mp;
126 struct dentry *root_dentry, *orphanage_dentry;
127 struct inode *root_inode = VFS_I(sc->mp->m_rootip);
128 struct inode *orphanage_inode;
129 int error;
130
131 if (xfs_is_shutdown(mp))
132 return -EIO;
133 if (xfs_is_readonly(mp)) {
134 sc->orphanage = NULL;
135 return 0;
136 }
137
138 ASSERT(sc->tp == NULL);
139 ASSERT(sc->orphanage == NULL);
140
141 /* Find the dentry for the root directory... */
142 root_dentry = d_find_alias(root_inode);
143 if (!root_dentry) {
144 error = -EFSCORRUPTED;
145 goto out;
146 }
147
148 /* ...which is a directory, right? */
149 if (!d_is_dir(root_dentry)) {
150 error = -EFSCORRUPTED;
151 goto out_dput_root;
152 }
153
154 /* Try to find the orphanage directory. */
155 inode_lock_nested(root_inode, I_MUTEX_PARENT);
156 orphanage_dentry = lookup_one_len(ORPHANAGE, root_dentry,
157 strlen(ORPHANAGE));
158 if (IS_ERR(orphanage_dentry)) {
159 error = PTR_ERR(orphanage_dentry);
160 goto out_unlock_root;
161 }
162
163 /*
164 * Nothing found? Call mkdir to create the orphanage. Create the
165 * directory without other-user access because we're live and someone
166 * could have been relying partly on minimal access to a parent
167 * directory to control access to a file we put in here.
168 */
169 if (d_really_is_negative(orphanage_dentry)) {
170 error = vfs_mkdir(&nop_mnt_idmap, root_inode, orphanage_dentry,
171 0750);
172 if (error)
173 goto out_dput_orphanage;
174 }
175
176 /* Not a directory? Bail out. */
177 if (!d_is_dir(orphanage_dentry)) {
178 error = -ENOTDIR;
179 goto out_dput_orphanage;
180 }
181
182 /*
183 * Grab a reference to the orphanage. This /should/ succeed since
184 * we hold the root directory locked and therefore nobody can delete
185 * the orphanage.
186 */
187 orphanage_inode = igrab(d_inode(orphanage_dentry));
188 if (!orphanage_inode) {
189 error = -ENOENT;
190 goto out_dput_orphanage;
191 }
192
193 /* Make sure the orphanage is owned by root. */
194 error = xrep_chown_orphanage(sc, XFS_I(orphanage_inode));
195 if (error)
196 goto out_dput_orphanage;
197
198 /* Stash the reference for later and bail out. */
199 sc->orphanage = XFS_I(orphanage_inode);
200 sc->orphanage_ilock_flags = 0;
201
202 out_dput_orphanage:
203 dput(orphanage_dentry);
204 out_unlock_root:
205 inode_unlock(VFS_I(sc->mp->m_rootip));
206 out_dput_root:
207 dput(root_dentry);
208 out:
209 return error;
210 }
211
212 void
213 xrep_orphanage_ilock(
214 struct xfs_scrub *sc,
215 unsigned int ilock_flags)
216 {
217 sc->orphanage_ilock_flags |= ilock_flags;
218 xfs_ilock(sc->orphanage, ilock_flags);
219 }
220
221 bool
222 xrep_orphanage_ilock_nowait(
223 struct xfs_scrub *sc,
224 unsigned int ilock_flags)
225 {
226 if (xfs_ilock_nowait(sc->orphanage, ilock_flags)) {
227 sc->orphanage_ilock_flags |= ilock_flags;
228 return true;
229 }
230
231 return false;
232 }
233
234 void
235 xrep_orphanage_iunlock(
236 struct xfs_scrub *sc,
237 unsigned int ilock_flags)
238 {
239 xfs_iunlock(sc->orphanage, ilock_flags);
240 sc->orphanage_ilock_flags &= ~ilock_flags;
241 }
242
243 /* Grab the IOLOCK of the orphanage and sc->ip. */
244 int
245 xrep_orphanage_iolock_two(
246 struct xfs_scrub *sc)
247 {
248 int error = 0;
249
250 while (true) {
251 if (xchk_should_terminate(sc, &error))
252 return error;
253
254 /*
255 * Normal XFS takes the IOLOCK before grabbing a transaction.
256 * Scrub holds a transaction, which means that we can't block
257 * on either IOLOCK.
258 */
259 if (xrep_orphanage_ilock_nowait(sc, XFS_IOLOCK_EXCL)) {
260 if (xchk_ilock_nowait(sc, XFS_IOLOCK_EXCL))
261 break;
262 xrep_orphanage_iunlock(sc, XFS_IOLOCK_EXCL);
263 }
264 delay(1);
265 }
266
267 return 0;
268 }
269
270 /* Release the orphanage. */
271 void
272 xrep_orphanage_rele(
273 struct xfs_scrub *sc)
274 {
275 if (!sc->orphanage)
276 return;
277
278 if (sc->orphanage_ilock_flags)
279 xfs_iunlock(sc->orphanage, sc->orphanage_ilock_flags);
280
281 xchk_irele(sc, sc->orphanage);
282 sc->orphanage = NULL;
283 }
284
285 /* Adoption moves a file into /lost+found */
286
287 /* Can the orphanage adopt @sc->ip? */
288 bool
289 xrep_orphanage_can_adopt(
290 struct xfs_scrub *sc)
291 {
292 ASSERT(sc->ip != NULL);
293
294 if (!sc->orphanage)
295 return false;
296 if (sc->ip == sc->orphanage)
297 return false;
298 if (xfs_internal_inum(sc->mp, sc->ip->i_ino))
299 return false;
300 return true;
301 }
302
303 /*
304 * Create a new transaction to send a child to the orphanage.
305 *
306 * Allocate a new transaction with sufficient disk space to handle the
307 * adoption, take ILOCK_EXCL of the orphanage and sc->ip, joins them to the
308 * transaction, and reserve quota to reparent the latter. Caller must hold the
309 * IOLOCK of the orphanage and sc->ip.
310 */
311 int
312 xrep_adoption_trans_alloc(
313 struct xfs_scrub *sc,
314 struct xrep_adoption *adopt)
315 {
316 struct xfs_mount *mp = sc->mp;
317 unsigned int child_blkres = 0;
318 int error;
319
320 ASSERT(sc->tp == NULL);
321 ASSERT(sc->ip != NULL);
322 ASSERT(sc->orphanage != NULL);
323 ASSERT(sc->ilock_flags & XFS_IOLOCK_EXCL);
324 ASSERT(sc->orphanage_ilock_flags & XFS_IOLOCK_EXCL);
325 ASSERT(!(sc->ilock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)));
326 ASSERT(!(sc->orphanage_ilock_flags &
327 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)));
328
329 /* Compute the worst case space reservation that we need. */
330 adopt->sc = sc;
331 adopt->orphanage_blkres = xfs_link_space_res(mp, MAXNAMELEN);
332 if (S_ISDIR(VFS_I(sc->ip)->i_mode))
333 child_blkres = xfs_rename_space_res(mp, 0, false,
334 xfs_name_dotdot.len, false);
335 if (xfs_has_parent(mp))
336 child_blkres += XFS_ADDAFORK_SPACE_RES(mp);
337 adopt->child_blkres = child_blkres;
338
339 /*
340 * Allocate a transaction to link the child into the parent, along with
341 * enough disk space to handle expansion of both the orphanage and the
342 * dotdot entry of a child directory.
343 */
344 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link,
345 adopt->orphanage_blkres + adopt->child_blkres, 0, 0,
346 &sc->tp);
347 if (error)
348 return error;
349
350 xfs_lock_two_inodes(sc->orphanage, XFS_ILOCK_EXCL,
351 sc->ip, XFS_ILOCK_EXCL);
352 sc->ilock_flags |= XFS_ILOCK_EXCL;
353 sc->orphanage_ilock_flags |= XFS_ILOCK_EXCL;
354
355 xfs_trans_ijoin(sc->tp, sc->orphanage, 0);
356 xfs_trans_ijoin(sc->tp, sc->ip, 0);
357
358 /*
359 * Reserve enough quota in the orphan directory to add the new name.
360 * Normally the orphanage should have user/group/project ids of zero
361 * and hence is not subject to quota enforcement, but we're allowed to
362 * exceed quota to reattach disconnected parts of the directory tree.
363 */
364 error = xfs_trans_reserve_quota_nblks(sc->tp, sc->orphanage,
365 adopt->orphanage_blkres, 0, true);
366 if (error)
367 goto out_cancel;
368
369 /*
370 * Reserve enough quota in the child directory to change dotdot.
371 * Here we're also allowed to exceed file quota to repair inconsistent
372 * metadata.
373 */
374 if (adopt->child_blkres) {
375 error = xfs_trans_reserve_quota_nblks(sc->tp, sc->ip,
376 adopt->child_blkres, 0, true);
377 if (error)
378 goto out_cancel;
379 }
380
381 return 0;
382 out_cancel:
383 xchk_trans_cancel(sc);
384 xrep_orphanage_iunlock(sc, XFS_ILOCK_EXCL);
385 xchk_iunlock(sc, XFS_ILOCK_EXCL);
386 return error;
387 }
388
389 /*
390 * Compute the xfs_name for the directory entry that we're adding to the
391 * orphanage. Caller must hold ILOCKs of sc->ip and the orphanage and must not
392 * reuse namebuf until the adoption completes or is dissolved.
393 */
394 int
395 xrep_adoption_compute_name(
396 struct xrep_adoption *adopt,
397 struct xfs_name *xname)
398 {
399 struct xfs_scrub *sc = adopt->sc;
400 char *namebuf = (void *)xname->name;
401 xfs_ino_t ino;
402 unsigned int incr = 0;
403 int error = 0;
404
405 adopt->xname = xname;
406 xname->len = snprintf(namebuf, MAXNAMELEN, "%llu", sc->ip->i_ino);
407 xname->type = xfs_mode_to_ftype(VFS_I(sc->ip)->i_mode);
408
409 /* Make sure the filename is unique in the lost+found. */
410 error = xchk_dir_lookup(sc, sc->orphanage, xname, &ino);
411 while (error == 0 && incr < 10000) {
412 xname->len = snprintf(namebuf, MAXNAMELEN, "%llu.%u",
413 sc->ip->i_ino, ++incr);
414 error = xchk_dir_lookup(sc, sc->orphanage, xname, &ino);
415 }
416 if (error == 0) {
417 /* We already have 10,000 entries in the orphanage? */
418 return -EFSCORRUPTED;
419 }
420
421 if (error != -ENOENT)
422 return error;
423 return 0;
424 }
425
426 /*
427 * Make sure the dcache does not have a positive dentry for the name we've
428 * chosen. The caller should have checked with the ondisk directory, so any
429 * discrepancy is a sign that something is seriously wrong.
430 */
431 static int
432 xrep_adoption_check_dcache(
433 struct xrep_adoption *adopt)
434 {
435 struct qstr qname = QSTR_INIT(adopt->xname->name,
436 adopt->xname->len);
437 struct xfs_scrub *sc = adopt->sc;
438 struct dentry *d_orphanage, *d_child;
439 int error = 0;
440
441 d_orphanage = d_find_alias(VFS_I(sc->orphanage));
442 if (!d_orphanage)
443 return 0;
444
445 d_child = d_hash_and_lookup(d_orphanage, &qname);
446 if (d_child) {
447 trace_xrep_adoption_check_child(sc->mp, d_child);
448
449 if (d_is_positive(d_child)) {
450 ASSERT(d_is_negative(d_child));
451 error = -EFSCORRUPTED;
452 }
453
454 dput(d_child);
455 }
456
457 dput(d_orphanage);
458 return error;
459 }
460
461 /*
462 * Invalidate all dentries for the name that was added to the orphanage
463 * directory, and all dentries pointing to the child inode that was moved.
464 *
465 * There should not be any positive entries for the name, since we've
466 * maintained our lock on the orphanage directory.
467 */
468 static void
469 xrep_adoption_zap_dcache(
470 struct xrep_adoption *adopt)
471 {
472 struct qstr qname = QSTR_INIT(adopt->xname->name,
473 adopt->xname->len);
474 struct xfs_scrub *sc = adopt->sc;
475 struct dentry *d_orphanage, *d_child;
476
477 /* Invalidate all dentries for the adoption name */
478 d_orphanage = d_find_alias(VFS_I(sc->orphanage));
479 if (!d_orphanage)
480 return;
481
482 d_child = d_hash_and_lookup(d_orphanage, &qname);
483 while (d_child != NULL) {
484 trace_xrep_adoption_invalidate_child(sc->mp, d_child);
485
486 ASSERT(d_is_negative(d_child));
487 d_invalidate(d_child);
488 dput(d_child);
489 d_child = d_lookup(d_orphanage, &qname);
490 }
491
492 dput(d_orphanage);
493
494 /* Invalidate all the dentries pointing down to this file. */
495 while ((d_child = d_find_alias(VFS_I(sc->ip))) != NULL) {
496 trace_xrep_adoption_invalidate_child(sc->mp, d_child);
497
498 d_invalidate(d_child);
499 dput(d_child);
500 }
501 }
502
503 /*
504 * If we have to add an attr fork ahead of a parent pointer update, how much
505 * space should we ask for?
506 */
507 static inline int
508 xrep_adoption_attr_sizeof(
509 const struct xrep_adoption *adopt)
510 {
511 return sizeof(struct xfs_attr_sf_hdr) +
512 xfs_attr_sf_entsize_byname(sizeof(struct xfs_parent_rec),
513 adopt->xname->len);
514 }
515
516 /*
517 * Move the current file to the orphanage under the computed name.
518 *
519 * Returns with a dirty transaction so that the caller can handle any other
520 * work, such as fixing up unlinked lists or resetting link counts.
521 */
522 int
523 xrep_adoption_move(
524 struct xrep_adoption *adopt)
525 {
526 struct xfs_scrub *sc = adopt->sc;
527 bool isdir = S_ISDIR(VFS_I(sc->ip)->i_mode);
528 int error;
529
530 trace_xrep_adoption_reparent(sc->orphanage, adopt->xname,
531 sc->ip->i_ino);
532
533 error = xrep_adoption_check_dcache(adopt);
534 if (error)
535 return error;
536
537 /*
538 * If this filesystem has parent pointers, ensure that the file being
539 * moved to the orphanage has an attribute fork. This is required
540 * because the parent pointer code does not itself add attr forks.
541 */
542 if (!xfs_inode_has_attr_fork(sc->ip) && xfs_has_parent(sc->mp)) {
543 int sf_size = xrep_adoption_attr_sizeof(adopt);
544
545 error = xfs_bmap_add_attrfork(sc->tp, sc->ip, sf_size, true);
546 if (error)
547 return error;
548 }
549
550 /* Create the new name in the orphanage. */
551 error = xfs_dir_createname(sc->tp, sc->orphanage, adopt->xname,
552 sc->ip->i_ino, adopt->orphanage_blkres);
553 if (error)
554 return error;
555
556 /*
557 * Bump the link count of the orphanage if we just added a
558 * subdirectory, and update its timestamps.
559 */
560 xfs_trans_ichgtime(sc->tp, sc->orphanage,
561 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
562 if (isdir)
563 xfs_bumplink(sc->tp, sc->orphanage);
564 xfs_trans_log_inode(sc->tp, sc->orphanage, XFS_ILOG_CORE);
565
566 /* Bump the link count of the child. */
567 if (adopt->bump_child_nlink) {
568 xfs_bumplink(sc->tp, sc->ip);
569 xfs_trans_log_inode(sc->tp, sc->ip, XFS_ILOG_CORE);
570 }
571
572 /* Replace the dotdot entry if the child is a subdirectory. */
573 if (isdir) {
574 error = xfs_dir_replace(sc->tp, sc->ip, &xfs_name_dotdot,
575 sc->orphanage->i_ino, adopt->child_blkres);
576 if (error)
577 return error;
578 }
579
580 /* Add a parent pointer from the file back to the lost+found. */
581 if (xfs_has_parent(sc->mp)) {
582 error = xfs_parent_addname(sc->tp, &adopt->ppargs,
583 sc->orphanage, adopt->xname, sc->ip);
584 if (error)
585 return error;
586 }
587
588 /*
589 * Notify dirent hooks that we moved the file to /lost+found, and
590 * finish all the deferred work so that we know the adoption is fully
591 * recorded in the log.
592 */
593 xfs_dir_update_hook(sc->orphanage, sc->ip, 1, adopt->xname);
594
595 /* Remove negative dentries from the lost+found's dcache */
596 xrep_adoption_zap_dcache(adopt);
597 return 0;
598 }
599
600 /*
601 * Roll to a clean scrub transaction so that we can release the orphanage,
602 * even if xrep_adoption_move was not called.
603 *
604 * Commits all the work and deferred ops attached to an adoption request and
605 * rolls to a clean scrub transaction. On success, returns 0 with the scrub
606 * context holding a clean transaction with no inodes joined. On failure,
607 * returns negative errno with no scrub transaction. All inode locks are
608 * still held after this function returns.
609 */
610 int
611 xrep_adoption_trans_roll(
612 struct xrep_adoption *adopt)
613 {
614 struct xfs_scrub *sc = adopt->sc;
615 int error;
616
617 trace_xrep_adoption_trans_roll(sc->orphanage, sc->ip,
618 !!(sc->tp->t_flags & XFS_TRANS_DIRTY));
619
620 /* Finish all the deferred ops to commit all repairs. */
621 error = xrep_defer_finish(sc);
622 if (error)
623 return error;
624
625 /* Roll the transaction once more to detach the inodes. */
626 return xfs_trans_roll(&sc->tp);
627 }
628
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.