]> git.ipfire.org Git - thirdparty/kernel/linux.git/blob
projects / thirdparty / kernel / linux.git / blob
? search:


4080401
[thirdparty/kernel/linux.git] /
1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "bkey_buf.h"
5 #include "btree_update.h"
6 #include "buckets.h"
7 #include "darray.h"
8 #include "dirent.h"
9 #include "error.h"
10 #include "fs-common.h"
11 #include "fsck.h"
12 #include "inode.h"
13 #include "keylist.h"
14 #include "subvolume.h"
15 #include "super.h"
16 #include "xattr.h"
17
18 #include <linux/bsearch.h>
19 #include <linux/dcache.h> /* struct qstr */
20
21 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
22
23 /*
24 * XXX: this is handling transaction restarts without returning
25 * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
26 */
27 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
28 u32 snapshot)
29 {
30 struct btree_iter iter;
31 struct bkey_s_c k;
32 u64 sectors = 0;
33 int ret;
34
35 for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
36 SPOS(inum, 0, snapshot),
37 POS(inum, U64_MAX),
38 0, k, ret)
39 if (bkey_extent_is_allocation(k.k))
40 sectors += k.k->size;
41
42 bch2_trans_iter_exit(trans, &iter);
43
44 return ret ?: sectors;
45 }
46
47 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
48 u32 snapshot)
49 {
50 struct btree_iter iter;
51 struct bkey_s_c k;
52 struct bkey_s_c_dirent d;
53 u64 subdirs = 0;
54 int ret;
55
56 for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
57 SPOS(inum, 0, snapshot),
58 POS(inum, U64_MAX),
59 0, k, ret) {
60 if (k.k->type != KEY_TYPE_dirent)
61 continue;
62
63 d = bkey_s_c_to_dirent(k);
64 if (d.v->d_type == DT_DIR)
65 subdirs++;
66 }
67 bch2_trans_iter_exit(trans, &iter);
68
69 return ret ?: subdirs;
70 }
71
72 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
73 u32 *subvol)
74 {
75 struct bch_snapshot s;
76 int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots,
77 POS(0, snapshot), 0,
78 snapshot, &s);
79 if (!ret)
80 *subvol = le32_to_cpu(s.subvol);
81 else if (bch2_err_matches(ret, ENOENT))
82 bch_err(trans->c, "snapshot %u not fonud", snapshot);
83 return ret;
84
85 }
86
87 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
88 u32 *snapshot, u64 *inum)
89 {
90 struct bch_subvolume s;
91 int ret;
92
93 ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
94
95 *snapshot = le32_to_cpu(s.snapshot);
96 *inum = le64_to_cpu(s.inode);
97 return ret;
98 }
99
100 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
101 u32 *snapshot, u64 *inum)
102 {
103 return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
104 }
105
106 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
107 struct bch_inode_unpacked *inode)
108 {
109 struct btree_iter iter;
110 struct bkey_s_c k;
111 int ret;
112
113 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
114 POS(0, inode_nr),
115 BTREE_ITER_ALL_SNAPSHOTS);
116 k = bch2_btree_iter_peek(&iter);
117 ret = bkey_err(k);
118 if (ret)
119 goto err;
120
121 if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
122 ret = -BCH_ERR_ENOENT_inode;
123 goto err;
124 }
125
126 ret = bch2_inode_unpack(k, inode);
127 err:
128 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
129 bch_err(trans->c, "error fetching inode %llu: %s",
130 inode_nr, bch2_err_str(ret));
131 bch2_trans_iter_exit(trans, &iter);
132 return ret;
133 }
134
135 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
136 struct bch_inode_unpacked *inode,
137 u32 *snapshot)
138 {
139 struct btree_iter iter;
140 struct bkey_s_c k;
141 int ret;
142
143 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
144 SPOS(0, inode_nr, *snapshot), 0);
145 ret = bkey_err(k);
146 if (ret)
147 goto err;
148
149 ret = bkey_is_inode(k.k)
150 ? bch2_inode_unpack(k, inode)
151 : -BCH_ERR_ENOENT_inode;
152 if (!ret)
153 *snapshot = iter.pos.snapshot;
154 err:
155 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
156 bch_err(trans->c, "error fetching inode %llu:%u: %s",
157 inode_nr, *snapshot, bch2_err_str(ret));
158 bch2_trans_iter_exit(trans, &iter);
159 return ret;
160 }
161
162 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
163 struct bch_inode_unpacked *inode,
164 u32 *snapshot)
165 {
166 return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
167 }
168
169 static int __lookup_dirent(struct btree_trans *trans,
170 struct bch_hash_info hash_info,
171 subvol_inum dir, struct qstr *name,
172 u64 *target, unsigned *type)
173 {
174 struct btree_iter iter;
175 struct bkey_s_c_dirent d;
176 int ret;
177
178 ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
179 &hash_info, dir, name, 0);
180 if (ret)
181 return ret;
182
183 d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
184 *target = le64_to_cpu(d.v->d_inum);
185 *type = d.v->d_type;
186 bch2_trans_iter_exit(trans, &iter);
187 return 0;
188 }
189
190 static int __write_inode(struct btree_trans *trans,
191 struct bch_inode_unpacked *inode,
192 u32 snapshot)
193 {
194 struct btree_iter iter;
195 int ret;
196
197 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
198 SPOS(0, inode->bi_inum, snapshot),
199 BTREE_ITER_INTENT);
200
201 ret = bch2_btree_iter_traverse(&iter) ?:
202 bch2_inode_write(trans, &iter, inode);
203 bch2_trans_iter_exit(trans, &iter);
204 return ret;
205 }
206
207 static int write_inode(struct btree_trans *trans,
208 struct bch_inode_unpacked *inode,
209 u32 snapshot)
210 {
211 int ret = commit_do(trans, NULL, NULL,
212 BTREE_INSERT_NOFAIL|
213 BTREE_INSERT_LAZY_RW,
214 __write_inode(trans, inode, snapshot));
215 if (ret)
216 bch_err(trans->c, "error in fsck: error updating inode: %s",
217 bch2_err_str(ret));
218 return ret;
219 }
220
221 static int fsck_inode_rm(struct btree_trans *trans, u64 inum, u32 snapshot)
222 {
223 struct bch_fs *c = trans->c;
224 struct btree_iter iter = { NULL };
225 struct bkey_i_inode_generation delete;
226 struct bch_inode_unpacked inode_u;
227 struct bkey_s_c k;
228 int ret;
229
230 do {
231 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
232 SPOS(inum, 0, snapshot),
233 SPOS(inum, U64_MAX, snapshot),
234 0, NULL) ?:
235 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
236 SPOS(inum, 0, snapshot),
237 SPOS(inum, U64_MAX, snapshot),
238 0, NULL) ?:
239 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
240 SPOS(inum, 0, snapshot),
241 SPOS(inum, U64_MAX, snapshot),
242 0, NULL);
243 } while (ret == -BCH_ERR_transaction_restart_nested);
244 if (ret)
245 goto err;
246 retry:
247 bch2_trans_begin(trans);
248
249 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
250 SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
251 ret = bkey_err(k);
252 if (ret)
253 goto err;
254
255 if (!bkey_is_inode(k.k)) {
256 bch2_fs_inconsistent(c,
257 "inode %llu:%u not found when deleting",
258 inum, snapshot);
259 ret = -EIO;
260 goto err;
261 }
262
263 bch2_inode_unpack(k, &inode_u);
264
265 /* Subvolume root? */
266 if (inode_u.bi_subvol)
267 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
268
269 bkey_inode_generation_init(&delete.k_i);
270 delete.k.p = iter.pos;
271 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
272
273 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
274 bch2_trans_commit(trans, NULL, NULL,
275 BTREE_INSERT_NOFAIL);
276 err:
277 bch2_trans_iter_exit(trans, &iter);
278 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
279 goto retry;
280
281 return ret ?: -BCH_ERR_transaction_restart_nested;
282 }
283
284 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
285 {
286 struct bch_fs *c = trans->c;
287 struct btree_iter iter;
288 struct bch_inode_unpacked dir_inode;
289 struct bch_hash_info dir_hash_info;
290 int ret;
291
292 ret = lookup_first_inode(trans, pos.inode, &dir_inode);
293 if (ret)
294 goto err;
295
296 dir_hash_info = bch2_hash_info_init(c, &dir_inode);
297
298 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
299
300 ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
301 &dir_hash_info, &iter,
302 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
303 bch2_trans_iter_exit(trans, &iter);
304 err:
305 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
306 bch_err_fn(c, ret);
307 return ret;
308 }
309
310 /* Get lost+found, create if it doesn't exist: */
311 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
312 struct bch_inode_unpacked *lostfound)
313 {
314 struct bch_fs *c = trans->c;
315 struct bch_inode_unpacked root;
316 struct bch_hash_info root_hash_info;
317 struct qstr lostfound_str = QSTR("lost+found");
318 subvol_inum root_inum = { .subvol = subvol };
319 u64 inum = 0;
320 unsigned d_type = 0;
321 u32 snapshot;
322 int ret;
323
324 ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
325 if (ret)
326 return ret;
327
328 ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
329 if (ret)
330 return ret;
331
332 root_hash_info = bch2_hash_info_init(c, &root);
333
334 ret = __lookup_dirent(trans, root_hash_info, root_inum,
335 &lostfound_str, &inum, &d_type);
336 if (bch2_err_matches(ret, ENOENT)) {
337 bch_notice(c, "creating lost+found");
338 goto create_lostfound;
339 }
340
341 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
342 bch_err(c, "error looking up lost+found: %s", bch2_err_str(ret));
343 if (ret)
344 return ret;
345
346 if (d_type != DT_DIR) {
347 bch_err(c, "error looking up lost+found: not a directory");
348 return ret;
349 }
350
351 /*
352 * The check_dirents pass has already run, dangling dirents
353 * shouldn't exist here:
354 */
355 return __lookup_inode(trans, inum, lostfound, &snapshot);
356
357 create_lostfound:
358 bch2_inode_init_early(c, lostfound);
359
360 ret = bch2_create_trans(trans, root_inum, &root,
361 lostfound, &lostfound_str,
362 0, 0, S_IFDIR|0700, 0, NULL, NULL,
363 (subvol_inum) { }, 0);
364 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
365 bch_err(c, "error creating lost+found: %s", bch2_err_str(ret));
366 return ret;
367 }
368
369 static int __reattach_inode(struct btree_trans *trans,
370 struct bch_inode_unpacked *inode,
371 u32 inode_snapshot)
372 {
373 struct bch_hash_info dir_hash;
374 struct bch_inode_unpacked lostfound;
375 char name_buf[20];
376 struct qstr name;
377 u64 dir_offset = 0;
378 u32 subvol;
379 int ret;
380
381 ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
382 if (ret)
383 return ret;
384
385 ret = lookup_lostfound(trans, subvol, &lostfound);
386 if (ret)
387 return ret;
388
389 if (S_ISDIR(inode->bi_mode)) {
390 lostfound.bi_nlink++;
391
392 ret = __write_inode(trans, &lostfound, U32_MAX);
393 if (ret)
394 return ret;
395 }
396
397 dir_hash = bch2_hash_info_init(trans->c, &lostfound);
398
399 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
400 name = (struct qstr) QSTR(name_buf);
401
402 ret = bch2_dirent_create(trans,
403 (subvol_inum) {
404 .subvol = subvol,
405 .inum = lostfound.bi_inum,
406 },
407 &dir_hash,
408 inode_d_type(inode),
409 &name, inode->bi_inum, &dir_offset,
410 BCH_HASH_SET_MUST_CREATE);
411 if (ret)
412 return ret;
413
414 inode->bi_dir = lostfound.bi_inum;
415 inode->bi_dir_offset = dir_offset;
416
417 return __write_inode(trans, inode, inode_snapshot);
418 }
419
420 static int reattach_inode(struct btree_trans *trans,
421 struct bch_inode_unpacked *inode,
422 u32 inode_snapshot)
423 {
424 int ret = commit_do(trans, NULL, NULL,
425 BTREE_INSERT_LAZY_RW|
426 BTREE_INSERT_NOFAIL,
427 __reattach_inode(trans, inode, inode_snapshot));
428 if (ret) {
429 bch_err(trans->c, "error reattaching inode %llu: %s",
430 inode->bi_inum, bch2_err_str(ret));
431 return ret;
432 }
433
434 return ret;
435 }
436
437 static int remove_backpointer(struct btree_trans *trans,
438 struct bch_inode_unpacked *inode)
439 {
440 struct btree_iter iter;
441 struct bkey_s_c_dirent d;
442 int ret;
443
444 d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
445 POS(inode->bi_dir, inode->bi_dir_offset), 0,
446 dirent);
447 ret = bkey_err(d) ?:
448 __remove_dirent(trans, d.k->p);
449 bch2_trans_iter_exit(trans, &iter);
450 return ret;
451 }
452
453 struct snapshots_seen_entry {
454 u32 id;
455 u32 equiv;
456 };
457
458 struct snapshots_seen {
459 struct bpos pos;
460 DARRAY(struct snapshots_seen_entry) ids;
461 };
462
463 static inline void snapshots_seen_exit(struct snapshots_seen *s)
464 {
465 darray_exit(&s->ids);
466 }
467
468 static inline void snapshots_seen_init(struct snapshots_seen *s)
469 {
470 memset(s, 0, sizeof(*s));
471 }
472
473 static int snapshots_seen_add(struct bch_fs *c, struct snapshots_seen *s, u32 id)
474 {
475 struct snapshots_seen_entry *i, n = { id, id };
476 int ret;
477
478 darray_for_each(s->ids, i) {
479 if (n.equiv < i->equiv)
480 break;
481
482 if (i->equiv == n.equiv) {
483 bch_err(c, "%s(): adding duplicate snapshot", __func__);
484 return -EINVAL;
485 }
486 }
487
488 ret = darray_insert_item(&s->ids, i - s->ids.data, n);
489 if (ret)
490 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
491 s->ids.size);
492 return ret;
493 }
494
495 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
496 enum btree_id btree_id, struct bpos pos)
497 {
498 struct snapshots_seen_entry *i, n = {
499 .id = pos.snapshot,
500 .equiv = bch2_snapshot_equiv(c, pos.snapshot),
501 };
502 int ret = 0;
503
504 if (!bkey_eq(s->pos, pos))
505 s->ids.nr = 0;
506
507 pos.snapshot = n.equiv;
508 s->pos = pos;
509
510 darray_for_each(s->ids, i)
511 if (i->equiv == n.equiv) {
512 if (fsck_err_on(i->id != n.id, c,
513 "snapshot deletion did not run correctly:\n"
514 " duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
515 bch2_btree_ids[btree_id],
516 pos.inode, pos.offset,
517 i->id, n.id, n.equiv))
518 return -BCH_ERR_need_snapshot_cleanup;
519
520 return 0;
521 }
522
523 ret = darray_push(&s->ids, n);
524 if (ret)
525 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
526 s->ids.size);
527 fsck_err:
528 return ret;
529 }
530
531 /**
532 * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
533 * and @ancestor hasn't been overwritten in @seen
534 *
535 * That is, returns whether key in @ancestor snapshot is visible in @id snapshot
536 */
537 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
538 u32 id, u32 ancestor)
539 {
540 ssize_t i;
541 u32 top = seen->ids.nr ? seen->ids.data[seen->ids.nr - 1].equiv : 0;
542
543 BUG_ON(id > ancestor);
544 BUG_ON(!bch2_snapshot_is_equiv(c, id));
545 BUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
546
547 /* @ancestor should be the snapshot most recently added to @seen */
548 BUG_ON(ancestor != seen->pos.snapshot);
549 BUG_ON(ancestor != top);
550
551 if (id == ancestor)
552 return true;
553
554 if (!bch2_snapshot_is_ancestor(c, id, ancestor))
555 return false;
556
557 for (i = seen->ids.nr - 2;
558 i >= 0 && seen->ids.data[i].equiv >= id;
559 --i)
560 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv) &&
561 bch2_snapshot_is_ancestor(c, seen->ids.data[i].equiv, ancestor))
562 return false;
563
564 return true;
565 }
566
567 /**
568 * ref_visible - given a key with snapshot id @src that points to a key with
569 * snapshot id @dst, test whether there is some snapshot in which @dst is
570 * visible.
571 *
572 * This assumes we're visiting @src keys in natural key order.
573 *
574 * @s - list of snapshot IDs already seen at @src
575 * @src - snapshot ID of src key
576 * @dst - snapshot ID of dst key
577 */
578 static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
579 u32 src, u32 dst)
580 {
581 return dst <= src
582 ? key_visible_in_snapshot(c, s, dst, src)
583 : bch2_snapshot_is_ancestor(c, src, dst);
584 }
585
586 static int ref_visible2(struct bch_fs *c,
587 u32 src, struct snapshots_seen *src_seen,
588 u32 dst, struct snapshots_seen *dst_seen)
589 {
590 src = bch2_snapshot_equiv(c, src);
591 dst = bch2_snapshot_equiv(c, dst);
592
593 if (dst > src) {
594 swap(dst, src);
595 swap(dst_seen, src_seen);
596 }
597 return key_visible_in_snapshot(c, src_seen, dst, src);
598 }
599
600 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \
601 for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && \
602 (_i)->snapshot <= (_snapshot); _i++) \
603 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
604
605 struct inode_walker_entry {
606 struct bch_inode_unpacked inode;
607 u32 snapshot;
608 u64 count;
609 };
610
611 struct inode_walker {
612 bool first_this_inode;
613 u64 cur_inum;
614
615 DARRAY(struct inode_walker_entry) inodes;
616 };
617
618 static void inode_walker_exit(struct inode_walker *w)
619 {
620 darray_exit(&w->inodes);
621 }
622
623 static struct inode_walker inode_walker_init(void)
624 {
625 return (struct inode_walker) { 0, };
626 }
627
628 static int add_inode(struct bch_fs *c, struct inode_walker *w,
629 struct bkey_s_c inode)
630 {
631 struct bch_inode_unpacked u;
632
633 BUG_ON(bch2_inode_unpack(inode, &u));
634
635 return darray_push(&w->inodes, ((struct inode_walker_entry) {
636 .inode = u,
637 .snapshot = bch2_snapshot_equiv(c, inode.k->p.snapshot),
638 }));
639 }
640
641 static int __walk_inode(struct btree_trans *trans,
642 struct inode_walker *w, struct bpos pos)
643 {
644 struct bch_fs *c = trans->c;
645 struct btree_iter iter;
646 struct bkey_s_c k;
647 u32 restart_count = trans->restart_count;
648 unsigned i;
649 int ret;
650
651 pos.snapshot = bch2_snapshot_equiv(c, pos.snapshot);
652
653 if (pos.inode == w->cur_inum)
654 goto lookup_snapshot;
655
656 w->inodes.nr = 0;
657
658 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
659 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
660 if (k.k->p.offset != pos.inode)
661 break;
662
663 if (bkey_is_inode(k.k))
664 add_inode(c, w, k);
665 }
666 bch2_trans_iter_exit(trans, &iter);
667
668 if (ret)
669 return ret;
670
671 w->cur_inum = pos.inode;
672 w->first_this_inode = true;
673
674 if (trans_was_restarted(trans, restart_count))
675 return -BCH_ERR_transaction_restart_nested;
676
677 lookup_snapshot:
678 for (i = 0; i < w->inodes.nr; i++)
679 if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->inodes.data[i].snapshot))
680 goto found;
681 return INT_MAX;
682 found:
683 BUG_ON(pos.snapshot > w->inodes.data[i].snapshot);
684
685 if (pos.snapshot != w->inodes.data[i].snapshot) {
686 struct inode_walker_entry e = w->inodes.data[i];
687
688 e.snapshot = pos.snapshot;
689 e.count = 0;
690
691 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
692 pos.inode, pos.snapshot, w->inodes.data[i].snapshot);
693
694 while (i && w->inodes.data[i - 1].snapshot > pos.snapshot)
695 --i;
696
697 ret = darray_insert_item(&w->inodes, i, e);
698 if (ret)
699 return ret;
700 }
701
702 return i;
703 }
704
705 static int __get_visible_inodes(struct btree_trans *trans,
706 struct inode_walker *w,
707 struct snapshots_seen *s,
708 u64 inum)
709 {
710 struct bch_fs *c = trans->c;
711 struct btree_iter iter;
712 struct bkey_s_c k;
713 int ret;
714
715 w->inodes.nr = 0;
716
717 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
718 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
719 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
720
721 if (k.k->p.offset != inum)
722 break;
723
724 if (!ref_visible(c, s, s->pos.snapshot, equiv))
725 continue;
726
727 if (bkey_is_inode(k.k))
728 add_inode(c, w, k);
729
730 if (equiv >= s->pos.snapshot)
731 break;
732 }
733 bch2_trans_iter_exit(trans, &iter);
734
735 return ret;
736 }
737
738 static int check_key_has_snapshot(struct btree_trans *trans,
739 struct btree_iter *iter,
740 struct bkey_s_c k)
741 {
742 struct bch_fs *c = trans->c;
743 struct printbuf buf = PRINTBUF;
744 int ret = 0;
745
746 if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
747 "key in missing snapshot: %s",
748 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
749 ret = bch2_btree_delete_at(trans, iter,
750 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
751 fsck_err:
752 printbuf_exit(&buf);
753 return ret;
754 }
755
756 static int hash_redo_key(struct btree_trans *trans,
757 const struct bch_hash_desc desc,
758 struct bch_hash_info *hash_info,
759 struct btree_iter *k_iter, struct bkey_s_c k)
760 {
761 struct bkey_i *delete;
762 struct bkey_i *tmp;
763
764 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
765 if (IS_ERR(delete))
766 return PTR_ERR(delete);
767
768 tmp = bch2_bkey_make_mut_noupdate(trans, k);
769 if (IS_ERR(tmp))
770 return PTR_ERR(tmp);
771
772 bkey_init(&delete->k);
773 delete->k.p = k_iter->pos;
774 return bch2_btree_iter_traverse(k_iter) ?:
775 bch2_trans_update(trans, k_iter, delete, 0) ?:
776 bch2_hash_set_snapshot(trans, desc, hash_info,
777 (subvol_inum) { 0, k.k->p.inode },
778 k.k->p.snapshot, tmp,
779 BCH_HASH_SET_MUST_CREATE,
780 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
781 bch2_trans_commit(trans, NULL, NULL,
782 BTREE_INSERT_NOFAIL|
783 BTREE_INSERT_LAZY_RW);
784 }
785
786 static int hash_check_key(struct btree_trans *trans,
787 const struct bch_hash_desc desc,
788 struct bch_hash_info *hash_info,
789 struct btree_iter *k_iter, struct bkey_s_c hash_k)
790 {
791 struct bch_fs *c = trans->c;
792 struct btree_iter iter = { NULL };
793 struct printbuf buf = PRINTBUF;
794 struct bkey_s_c k;
795 u64 hash;
796 int ret = 0;
797
798 if (hash_k.k->type != desc.key_type)
799 return 0;
800
801 hash = desc.hash_bkey(hash_info, hash_k);
802
803 if (likely(hash == hash_k.k->p.offset))
804 return 0;
805
806 if (hash_k.k->p.offset < hash)
807 goto bad_hash;
808
809 for_each_btree_key_norestart(trans, iter, desc.btree_id,
810 SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
811 BTREE_ITER_SLOTS, k, ret) {
812 if (bkey_eq(k.k->p, hash_k.k->p))
813 break;
814
815 if (fsck_err_on(k.k->type == desc.key_type &&
816 !desc.cmp_bkey(k, hash_k), c,
817 "duplicate hash table keys:\n%s",
818 (printbuf_reset(&buf),
819 bch2_bkey_val_to_text(&buf, c, hash_k),
820 buf.buf))) {
821 ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
822 break;
823 }
824
825 if (bkey_deleted(k.k)) {
826 bch2_trans_iter_exit(trans, &iter);
827 goto bad_hash;
828 }
829 }
830 out:
831 bch2_trans_iter_exit(trans, &iter);
832 printbuf_exit(&buf);
833 return ret;
834 bad_hash:
835 if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
836 bch2_btree_ids[desc.btree_id], hash_k.k->p.inode, hash_k.k->p.offset, hash,
837 (printbuf_reset(&buf),
838 bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
839 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
840 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
841 bch_err(c, "hash_redo_key err %s", bch2_err_str(ret));
842 if (ret)
843 return ret;
844 ret = -BCH_ERR_transaction_restart_nested;
845 }
846 fsck_err:
847 goto out;
848 }
849
850 static int check_inode(struct btree_trans *trans,
851 struct btree_iter *iter,
852 struct bkey_s_c k,
853 struct bch_inode_unpacked *prev,
854 struct snapshots_seen *s,
855 bool full)
856 {
857 struct bch_fs *c = trans->c;
858 struct bch_inode_unpacked u;
859 bool do_update = false;
860 int ret;
861
862 ret = check_key_has_snapshot(trans, iter, k);
863 if (ret < 0)
864 goto err;
865 if (ret)
866 return 0;
867
868 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
869 if (ret)
870 goto err;
871
872 /*
873 * if snapshot id isn't a leaf node, skip it - deletion in
874 * particular is not atomic, so on the internal snapshot nodes
875 * we can see inodes marked for deletion after a clean shutdown
876 */
877 if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
878 return 0;
879
880 if (!bkey_is_inode(k.k))
881 return 0;
882
883 BUG_ON(bch2_inode_unpack(k, &u));
884
885 if (!full &&
886 !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
887 BCH_INODE_I_SECTORS_DIRTY|
888 BCH_INODE_UNLINKED)))
889 return 0;
890
891 if (prev->bi_inum != u.bi_inum)
892 *prev = u;
893
894 if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed ||
895 inode_d_type(prev) != inode_d_type(&u), c,
896 "inodes in different snapshots don't match")) {
897 bch_err(c, "repair not implemented yet");
898 return -EINVAL;
899 }
900
901 if (u.bi_flags & BCH_INODE_UNLINKED &&
902 (!c->sb.clean ||
903 fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
904 u.bi_inum))) {
905 bch2_trans_unlock(trans);
906 bch2_fs_lazy_rw(c);
907
908 ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
909 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
910 bch_err(c, "error in fsck: error while deleting inode: %s",
911 bch2_err_str(ret));
912 return ret;
913 }
914
915 if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
916 (!c->sb.clean ||
917 fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
918 u.bi_inum))) {
919 bch_verbose(c, "truncating inode %llu", u.bi_inum);
920
921 bch2_trans_unlock(trans);
922 bch2_fs_lazy_rw(c);
923
924 /*
925 * XXX: need to truncate partial blocks too here - or ideally
926 * just switch units to bytes and that issue goes away
927 */
928 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
929 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
930 iter->pos.snapshot),
931 POS(u.bi_inum, U64_MAX),
932 0, NULL);
933 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
934 bch_err(c, "error in fsck: error truncating inode: %s",
935 bch2_err_str(ret));
936 if (ret)
937 return ret;
938
939 /*
940 * We truncated without our normal sector accounting hook, just
941 * make sure we recalculate it:
942 */
943 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
944
945 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
946 do_update = true;
947 }
948
949 if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
950 (!c->sb.clean ||
951 fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
952 u.bi_inum))) {
953 s64 sectors;
954
955 bch_verbose(c, "recounting sectors for inode %llu",
956 u.bi_inum);
957
958 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
959 if (sectors < 0) {
960 bch_err(c, "error in fsck: error recounting inode sectors: %s",
961 bch2_err_str(sectors));
962 return sectors;
963 }
964
965 u.bi_sectors = sectors;
966 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
967 do_update = true;
968 }
969
970 if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
971 u.bi_dir = 0;
972 u.bi_dir_offset = 0;
973 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
974 do_update = true;
975 }
976
977 if (do_update) {
978 ret = __write_inode(trans, &u, iter->pos.snapshot);
979 if (ret)
980 bch_err(c, "error in fsck: error updating inode: %s",
981 bch2_err_str(ret));
982 }
983 err:
984 fsck_err:
985 if (ret)
986 bch_err_fn(c, ret);
987 return ret;
988 }
989
990 noinline_for_stack
991 static int check_inodes(struct bch_fs *c, bool full)
992 {
993 struct btree_trans trans;
994 struct btree_iter iter;
995 struct bch_inode_unpacked prev = { 0 };
996 struct snapshots_seen s;
997 struct bkey_s_c k;
998 int ret;
999
1000 snapshots_seen_init(&s);
1001 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1002
1003 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
1004 POS_MIN,
1005 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1006 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1007 check_inode(&trans, &iter, k, &prev, &s, full));
1008
1009 bch2_trans_exit(&trans);
1010 snapshots_seen_exit(&s);
1011 if (ret)
1012 bch_err_fn(c, ret);
1013 return ret;
1014 }
1015
1016 /*
1017 * Checking for overlapping extents needs to be reimplemented
1018 */
1019 #if 0
1020 static int fix_overlapping_extent(struct btree_trans *trans,
1021 struct bkey_s_c k, struct bpos cut_at)
1022 {
1023 struct btree_iter iter;
1024 struct bkey_i *u;
1025 int ret;
1026
1027 u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
1028 ret = PTR_ERR_OR_ZERO(u);
1029 if (ret)
1030 return ret;
1031
1032 bkey_reassemble(u, k);
1033 bch2_cut_front(cut_at, u);
1034
1035
1036 /*
1037 * We don't want to go through the extent_handle_overwrites path:
1038 *
1039 * XXX: this is going to screw up disk accounting, extent triggers
1040 * assume things about extent overwrites - we should be running the
1041 * triggers manually here
1042 */
1043 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p,
1044 BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
1045
1046 BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS);
1047 ret = bch2_btree_iter_traverse(&iter) ?:
1048 bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?:
1049 bch2_trans_commit(trans, NULL, NULL,
1050 BTREE_INSERT_NOFAIL|
1051 BTREE_INSERT_LAZY_RW);
1052 bch2_trans_iter_exit(trans, &iter);
1053 return ret;
1054 }
1055 #endif
1056
1057 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1058 struct btree_iter *iter,
1059 struct bpos pos)
1060 {
1061 return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1062 }
1063
1064 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1065 struct bkey_s_c_dirent d)
1066 {
1067 return inode->bi_dir == d.k->p.inode &&
1068 inode->bi_dir_offset == d.k->p.offset;
1069 }
1070
1071 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1072 struct bch_inode_unpacked *inode)
1073 {
1074 return d.v->d_type == DT_SUBVOL
1075 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
1076 : le64_to_cpu(d.v->d_inum) == inode->bi_inum;
1077 }
1078
1079 static int inode_backpointer_exists(struct btree_trans *trans,
1080 struct bch_inode_unpacked *inode,
1081 u32 snapshot)
1082 {
1083 struct btree_iter iter;
1084 struct bkey_s_c_dirent d;
1085 int ret;
1086
1087 d = dirent_get_by_pos(trans, &iter,
1088 SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1089 ret = bkey_err(d);
1090 if (ret)
1091 return bch2_err_matches(ret, ENOENT) ? 0 : ret;
1092
1093 ret = dirent_points_to_inode(d, inode);
1094 bch2_trans_iter_exit(trans, &iter);
1095 return ret;
1096 }
1097
1098 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1099 {
1100 struct bch_fs *c = trans->c;
1101 struct inode_walker_entry *i;
1102 u32 restart_count = trans->restart_count;
1103 int ret = 0;
1104 s64 count2;
1105
1106 darray_for_each(w->inodes, i) {
1107 if (i->inode.bi_sectors == i->count)
1108 continue;
1109
1110 count2 = bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot);
1111
1112 if (i->count != count2) {
1113 bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu",
1114 i->count, count2);
1115 i->count = count2;
1116 if (i->inode.bi_sectors == i->count)
1117 continue;
1118 }
1119
1120 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1121 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1122 w->cur_inum, i->snapshot,
1123 i->inode.bi_sectors, i->count)) {
1124 i->inode.bi_sectors = i->count;
1125 ret = write_inode(trans, &i->inode, i->snapshot);
1126 if (ret)
1127 break;
1128 }
1129 }
1130 fsck_err:
1131 if (ret)
1132 bch_err_fn(c, ret);
1133 if (!ret && trans_was_restarted(trans, restart_count))
1134 ret = -BCH_ERR_transaction_restart_nested;
1135 return ret;
1136 }
1137
1138 struct extent_end {
1139 u32 snapshot;
1140 u64 offset;
1141 struct snapshots_seen seen;
1142 };
1143
1144 typedef DARRAY(struct extent_end) extent_ends;
1145
1146 static int get_print_extent(struct btree_trans *trans, struct bpos pos, struct printbuf *out)
1147 {
1148 struct btree_iter iter;
1149 struct bkey_s_c k;
1150 int ret;
1151
1152 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_extents, pos,
1153 BTREE_ITER_SLOTS|
1154 BTREE_ITER_ALL_SNAPSHOTS|
1155 BTREE_ITER_NOT_EXTENTS);
1156 ret = bkey_err(k);
1157 if (ret)
1158 return ret;
1159
1160 bch2_bkey_val_to_text(out, trans->c, k);
1161 bch2_trans_iter_exit(trans, &iter);
1162 return 0;
1163 }
1164
1165 static int check_overlapping_extents(struct btree_trans *trans,
1166 struct snapshots_seen *seen,
1167 extent_ends *extent_ends,
1168 struct bkey_s_c k,
1169 struct btree_iter *iter)
1170 {
1171 struct bch_fs *c = trans->c;
1172 struct extent_end *i;
1173 struct printbuf buf = PRINTBUF;
1174 int ret = 0;
1175
1176 darray_for_each(*extent_ends, i) {
1177 /* duplicate, due to transaction restart: */
1178 if (i->offset == k.k->p.offset &&
1179 i->snapshot == k.k->p.snapshot)
1180 continue;
1181
1182 if (!ref_visible2(c,
1183 k.k->p.snapshot, seen,
1184 i->snapshot, &i->seen))
1185 continue;
1186
1187 if (i->offset <= bkey_start_offset(k.k))
1188 continue;
1189
1190 printbuf_reset(&buf);
1191 prt_str(&buf, "overlapping extents:\n ");
1192 bch2_bkey_val_to_text(&buf, c, k);
1193 prt_str(&buf, "\n ");
1194
1195 ret = get_print_extent(trans, SPOS(k.k->p.inode, i->offset, i->snapshot), &buf);
1196 if (ret)
1197 break;
1198
1199 if (fsck_err(c, buf.buf)) {
1200 struct bkey_i *update = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
1201 if ((ret = PTR_ERR_OR_ZERO(update)))
1202 goto err;
1203 bkey_reassemble(update, k);
1204 ret = bch2_trans_update_extent(trans, iter, update, 0);
1205 if (ret)
1206 goto err;
1207 }
1208 }
1209 err:
1210 fsck_err:
1211 printbuf_exit(&buf);
1212 return ret;
1213 }
1214
1215 static int extent_ends_at(extent_ends *extent_ends,
1216 struct snapshots_seen *seen,
1217 struct bkey_s_c k)
1218 {
1219 struct extent_end *i, n = (struct extent_end) {
1220 .snapshot = k.k->p.snapshot,
1221 .offset = k.k->p.offset,
1222 .seen = *seen,
1223 };
1224
1225 n.seen.ids.data = kmemdup(seen->ids.data,
1226 sizeof(seen->ids.data[0]) * seen->ids.size,
1227 GFP_KERNEL);
1228 if (!n.seen.ids.data)
1229 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1230
1231 darray_for_each(*extent_ends, i) {
1232 if (i->snapshot == k.k->p.snapshot) {
1233 snapshots_seen_exit(&i->seen);
1234 *i = n;
1235 return 0;
1236 }
1237
1238 if (i->snapshot >= k.k->p.snapshot)
1239 break;
1240 }
1241
1242 return darray_insert_item(extent_ends, i - extent_ends->data, n);
1243 }
1244
1245 static void extent_ends_reset(extent_ends *extent_ends)
1246 {
1247 struct extent_end *i;
1248
1249 darray_for_each(*extent_ends, i)
1250 snapshots_seen_exit(&i->seen);
1251
1252 extent_ends->nr = 0;
1253 }
1254
1255 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1256 struct bkey_s_c k,
1257 struct inode_walker *inode,
1258 struct snapshots_seen *s,
1259 extent_ends *extent_ends)
1260 {
1261 struct bch_fs *c = trans->c;
1262 struct inode_walker_entry *i;
1263 struct printbuf buf = PRINTBUF;
1264 struct bpos equiv;
1265 int ret = 0;
1266
1267 ret = check_key_has_snapshot(trans, iter, k);
1268 if (ret) {
1269 ret = ret < 0 ? ret : 0;
1270 goto out;
1271 }
1272
1273 equiv = k.k->p;
1274 equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1275
1276 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1277 if (ret)
1278 goto err;
1279
1280 if (k.k->type == KEY_TYPE_whiteout)
1281 goto out;
1282
1283 if (inode->cur_inum != k.k->p.inode) {
1284 ret = check_i_sectors(trans, inode);
1285 if (ret)
1286 goto err;
1287
1288 extent_ends_reset(extent_ends);
1289 }
1290
1291 BUG_ON(!iter->path->should_be_locked);
1292
1293 ret = check_overlapping_extents(trans, s, extent_ends, k, iter);
1294 if (ret)
1295 goto err;
1296
1297 ret = extent_ends_at(extent_ends, s, k);
1298 if (ret)
1299 goto err;
1300
1301 ret = __walk_inode(trans, inode, equiv);
1302 if (ret < 0)
1303 goto err;
1304
1305 if (fsck_err_on(ret == INT_MAX, c,
1306 "extent in missing inode:\n %s",
1307 (printbuf_reset(&buf),
1308 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1309 ret = bch2_btree_delete_at(trans, iter,
1310 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1311 goto out;
1312 }
1313
1314 if (ret == INT_MAX) {
1315 ret = 0;
1316 goto out;
1317 }
1318
1319 i = inode->inodes.data + ret;
1320 ret = 0;
1321
1322 if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
1323 !S_ISLNK(i->inode.bi_mode), c,
1324 "extent in non regular inode mode %o:\n %s",
1325 i->inode.bi_mode,
1326 (printbuf_reset(&buf),
1327 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1328 ret = bch2_btree_delete_at(trans, iter,
1329 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1330 goto out;
1331 }
1332
1333 /*
1334 * Check inodes in reverse order, from oldest snapshots to newest, so
1335 * that we emit the fewest number of whiteouts necessary:
1336 */
1337 for (i = inode->inodes.data + inode->inodes.nr - 1;
1338 i >= inode->inodes.data;
1339 --i) {
1340 if (i->snapshot > equiv.snapshot ||
1341 !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1342 continue;
1343
1344 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1345 k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1346 !bkey_extent_is_reservation(k), c,
1347 "extent type past end of inode %llu:%u, i_size %llu\n %s",
1348 i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1349 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1350 struct btree_iter iter2;
1351
1352 bch2_trans_copy_iter(&iter2, iter);
1353 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1354 ret = bch2_btree_iter_traverse(&iter2) ?:
1355 bch2_btree_delete_at(trans, &iter2,
1356 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1357 bch2_trans_iter_exit(trans, &iter2);
1358 if (ret)
1359 goto err;
1360
1361 if (i->snapshot != equiv.snapshot) {
1362 ret = snapshots_seen_add(c, s, i->snapshot);
1363 if (ret)
1364 goto err;
1365 }
1366 }
1367 }
1368
1369 if (bkey_extent_is_allocation(k.k))
1370 for_each_visible_inode(c, s, inode, equiv.snapshot, i)
1371 i->count += k.k->size;
1372 #if 0
1373 bch2_bkey_buf_reassemble(&prev, c, k);
1374 #endif
1375
1376 out:
1377 err:
1378 fsck_err:
1379 printbuf_exit(&buf);
1380
1381 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1382 bch_err_fn(c, ret);
1383 return ret;
1384 }
1385
1386 /*
1387 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1388 * that i_size an i_sectors are consistent
1389 */
1390 noinline_for_stack
1391 static int check_extents(struct bch_fs *c)
1392 {
1393 struct inode_walker w = inode_walker_init();
1394 struct snapshots_seen s;
1395 struct btree_trans trans;
1396 struct btree_iter iter;
1397 struct bkey_s_c k;
1398 extent_ends extent_ends = { 0 };
1399 struct disk_reservation res = { 0 };
1400 int ret = 0;
1401
1402 snapshots_seen_init(&s);
1403 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1404
1405 bch_verbose(c, "checking extents");
1406
1407 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_extents,
1408 POS(BCACHEFS_ROOT_INO, 0),
1409 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1410 &res, NULL,
1411 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
1412 bch2_disk_reservation_put(c, &res);
1413 check_extent(&trans, &iter, k, &w, &s, &extent_ends);
1414 }));
1415
1416 bch2_disk_reservation_put(c, &res);
1417 extent_ends_reset(&extent_ends);
1418 darray_exit(&extent_ends);
1419 inode_walker_exit(&w);
1420 bch2_trans_exit(&trans);
1421 snapshots_seen_exit(&s);
1422
1423 if (ret)
1424 bch_err_fn(c, ret);
1425 return ret;
1426 }
1427
1428 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1429 {
1430 struct bch_fs *c = trans->c;
1431 struct inode_walker_entry *i;
1432 u32 restart_count = trans->restart_count;
1433 int ret = 0;
1434 s64 count2;
1435
1436 darray_for_each(w->inodes, i) {
1437 if (i->inode.bi_nlink == i->count)
1438 continue;
1439
1440 count2 = bch2_count_subdirs(trans, w->cur_inum, i->snapshot);
1441 if (count2 < 0)
1442 return count2;
1443
1444 if (i->count != count2) {
1445 bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1446 i->count, count2);
1447 i->count = count2;
1448 if (i->inode.bi_nlink == i->count)
1449 continue;
1450 }
1451
1452 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1453 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1454 w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) {
1455 i->inode.bi_nlink = i->count;
1456 ret = write_inode(trans, &i->inode, i->snapshot);
1457 if (ret)
1458 break;
1459 }
1460 }
1461 fsck_err:
1462 if (ret)
1463 bch_err_fn(c, ret);
1464 if (!ret && trans_was_restarted(trans, restart_count))
1465 ret = -BCH_ERR_transaction_restart_nested;
1466 return ret;
1467 }
1468
1469 static int check_dirent_target(struct btree_trans *trans,
1470 struct btree_iter *iter,
1471 struct bkey_s_c_dirent d,
1472 struct bch_inode_unpacked *target,
1473 u32 target_snapshot)
1474 {
1475 struct bch_fs *c = trans->c;
1476 struct bkey_i_dirent *n;
1477 bool backpointer_exists = true;
1478 struct printbuf buf = PRINTBUF;
1479 int ret = 0;
1480
1481 if (!target->bi_dir &&
1482 !target->bi_dir_offset) {
1483 target->bi_dir = d.k->p.inode;
1484 target->bi_dir_offset = d.k->p.offset;
1485
1486 ret = __write_inode(trans, target, target_snapshot);
1487 if (ret)
1488 goto err;
1489 }
1490
1491 if (!inode_points_to_dirent(target, d)) {
1492 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1493 if (ret < 0)
1494 goto err;
1495
1496 backpointer_exists = ret;
1497 ret = 0;
1498
1499 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1500 backpointer_exists, c,
1501 "directory %llu with multiple links",
1502 target->bi_inum)) {
1503 ret = __remove_dirent(trans, d.k->p);
1504 goto out;
1505 }
1506
1507 if (fsck_err_on(backpointer_exists &&
1508 !target->bi_nlink, c,
1509 "inode %llu type %s has multiple links but i_nlink 0",
1510 target->bi_inum, bch2_d_types[d.v->d_type])) {
1511 target->bi_nlink++;
1512 target->bi_flags &= ~BCH_INODE_UNLINKED;
1513
1514 ret = __write_inode(trans, target, target_snapshot);
1515 if (ret)
1516 goto err;
1517 }
1518
1519 if (fsck_err_on(!backpointer_exists, c,
1520 "inode %llu:%u has wrong backpointer:\n"
1521 "got %llu:%llu\n"
1522 "should be %llu:%llu",
1523 target->bi_inum, target_snapshot,
1524 target->bi_dir,
1525 target->bi_dir_offset,
1526 d.k->p.inode,
1527 d.k->p.offset)) {
1528 target->bi_dir = d.k->p.inode;
1529 target->bi_dir_offset = d.k->p.offset;
1530
1531 ret = __write_inode(trans, target, target_snapshot);
1532 if (ret)
1533 goto err;
1534 }
1535 }
1536
1537 if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1538 "incorrect d_type: got %s, should be %s:\n%s",
1539 bch2_d_type_str(d.v->d_type),
1540 bch2_d_type_str(inode_d_type(target)),
1541 (printbuf_reset(&buf),
1542 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1543 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1544 ret = PTR_ERR_OR_ZERO(n);
1545 if (ret)
1546 goto err;
1547
1548 bkey_reassemble(&n->k_i, d.s_c);
1549 n->v.d_type = inode_d_type(target);
1550
1551 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1552 if (ret)
1553 goto err;
1554
1555 d = dirent_i_to_s_c(n);
1556 }
1557
1558 if (d.v->d_type == DT_SUBVOL &&
1559 target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1560 (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1561 fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1562 le32_to_cpu(d.v->d_parent_subvol),
1563 target->bi_parent_subvol))) {
1564 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1565 ret = PTR_ERR_OR_ZERO(n);
1566 if (ret)
1567 goto err;
1568
1569 bkey_reassemble(&n->k_i, d.s_c);
1570 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1571
1572 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1573 if (ret)
1574 goto err;
1575
1576 d = dirent_i_to_s_c(n);
1577 }
1578 out:
1579 err:
1580 fsck_err:
1581 printbuf_exit(&buf);
1582
1583 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1584 bch_err_fn(c, ret);
1585 return ret;
1586 }
1587
1588 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1589 struct bkey_s_c k,
1590 struct bch_hash_info *hash_info,
1591 struct inode_walker *dir,
1592 struct inode_walker *target,
1593 struct snapshots_seen *s)
1594 {
1595 struct bch_fs *c = trans->c;
1596 struct bkey_s_c_dirent d;
1597 struct inode_walker_entry *i;
1598 struct printbuf buf = PRINTBUF;
1599 struct bpos equiv;
1600 int ret = 0;
1601
1602 ret = check_key_has_snapshot(trans, iter, k);
1603 if (ret) {
1604 ret = ret < 0 ? ret : 0;
1605 goto out;
1606 }
1607
1608 equiv = k.k->p;
1609 equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1610
1611 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1612 if (ret)
1613 goto err;
1614
1615 if (k.k->type == KEY_TYPE_whiteout)
1616 goto out;
1617
1618 if (dir->cur_inum != k.k->p.inode) {
1619 ret = check_subdir_count(trans, dir);
1620 if (ret)
1621 goto err;
1622 }
1623
1624 BUG_ON(!iter->path->should_be_locked);
1625
1626 ret = __walk_inode(trans, dir, equiv);
1627 if (ret < 0)
1628 goto err;
1629
1630 if (dir->first_this_inode)
1631 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1632 dir->first_this_inode = false;
1633
1634 if (fsck_err_on(ret == INT_MAX, c,
1635 "dirent in nonexisting directory:\n%s",
1636 (printbuf_reset(&buf),
1637 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1638 ret = bch2_btree_delete_at(trans, iter,
1639 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1640 goto out;
1641 }
1642
1643 if (ret == INT_MAX) {
1644 ret = 0;
1645 goto out;
1646 }
1647
1648 i = dir->inodes.data + ret;
1649 ret = 0;
1650
1651 if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1652 "dirent in non directory inode type %s:\n%s",
1653 bch2_d_type_str(inode_d_type(&i->inode)),
1654 (printbuf_reset(&buf),
1655 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1656 ret = bch2_btree_delete_at(trans, iter, 0);
1657 goto out;
1658 }
1659
1660 ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1661 if (ret < 0)
1662 goto err;
1663 if (ret) {
1664 /* dirent has been deleted */
1665 ret = 0;
1666 goto out;
1667 }
1668
1669 if (k.k->type != KEY_TYPE_dirent)
1670 goto out;
1671
1672 d = bkey_s_c_to_dirent(k);
1673
1674 if (d.v->d_type == DT_SUBVOL) {
1675 struct bch_inode_unpacked subvol_root;
1676 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1677 u32 target_snapshot;
1678 u64 target_inum;
1679
1680 ret = __subvol_lookup(trans, target_subvol,
1681 &target_snapshot, &target_inum);
1682 if (ret && !bch2_err_matches(ret, ENOENT))
1683 goto err;
1684
1685 if (fsck_err_on(ret, c,
1686 "dirent points to missing subvolume %llu",
1687 le64_to_cpu(d.v->d_child_subvol))) {
1688 ret = __remove_dirent(trans, d.k->p);
1689 goto err;
1690 }
1691
1692 ret = __lookup_inode(trans, target_inum,
1693 &subvol_root, &target_snapshot);
1694 if (ret && !bch2_err_matches(ret, ENOENT))
1695 goto err;
1696
1697 if (fsck_err_on(ret, c,
1698 "subvolume %u points to missing subvolume root %llu",
1699 target_subvol,
1700 target_inum)) {
1701 bch_err(c, "repair not implemented yet");
1702 ret = -EINVAL;
1703 goto err;
1704 }
1705
1706 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1707 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1708 target_inum,
1709 subvol_root.bi_subvol, target_subvol)) {
1710 subvol_root.bi_subvol = target_subvol;
1711 ret = __write_inode(trans, &subvol_root, target_snapshot);
1712 if (ret)
1713 goto err;
1714 }
1715
1716 ret = check_dirent_target(trans, iter, d, &subvol_root,
1717 target_snapshot);
1718 if (ret)
1719 goto err;
1720 } else {
1721 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1722 if (ret)
1723 goto err;
1724
1725 if (fsck_err_on(!target->inodes.nr, c,
1726 "dirent points to missing inode: (equiv %u)\n%s",
1727 equiv.snapshot,
1728 (printbuf_reset(&buf),
1729 bch2_bkey_val_to_text(&buf, c, k),
1730 buf.buf))) {
1731 ret = __remove_dirent(trans, d.k->p);
1732 if (ret)
1733 goto err;
1734 }
1735
1736 darray_for_each(target->inodes, i) {
1737 ret = check_dirent_target(trans, iter, d,
1738 &i->inode, i->snapshot);
1739 if (ret)
1740 goto err;
1741 }
1742 }
1743
1744 if (d.v->d_type == DT_DIR)
1745 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1746 i->count++;
1747
1748 out:
1749 err:
1750 fsck_err:
1751 printbuf_exit(&buf);
1752
1753 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1754 bch_err_fn(c, ret);
1755 return ret;
1756 }
1757
1758 /*
1759 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1760 * validate d_type
1761 */
1762 noinline_for_stack
1763 static int check_dirents(struct bch_fs *c)
1764 {
1765 struct inode_walker dir = inode_walker_init();
1766 struct inode_walker target = inode_walker_init();
1767 struct snapshots_seen s;
1768 struct bch_hash_info hash_info;
1769 struct btree_trans trans;
1770 struct btree_iter iter;
1771 struct bkey_s_c k;
1772 int ret = 0;
1773
1774 bch_verbose(c, "checking dirents");
1775
1776 snapshots_seen_init(&s);
1777 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1778
1779 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_dirents,
1780 POS(BCACHEFS_ROOT_INO, 0),
1781 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1782 k,
1783 NULL, NULL,
1784 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1785 check_dirent(&trans, &iter, k, &hash_info, &dir, &target, &s));
1786
1787 bch2_trans_exit(&trans);
1788 snapshots_seen_exit(&s);
1789 inode_walker_exit(&dir);
1790 inode_walker_exit(&target);
1791
1792 if (ret)
1793 bch_err_fn(c, ret);
1794 return ret;
1795 }
1796
1797 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1798 struct bkey_s_c k,
1799 struct bch_hash_info *hash_info,
1800 struct inode_walker *inode)
1801 {
1802 struct bch_fs *c = trans->c;
1803 int ret;
1804
1805 ret = check_key_has_snapshot(trans, iter, k);
1806 if (ret)
1807 return ret;
1808
1809 ret = __walk_inode(trans, inode, k.k->p);
1810 if (ret < 0)
1811 return ret;
1812
1813 if (inode->first_this_inode)
1814 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1815 inode->first_this_inode = false;
1816
1817 if (fsck_err_on(ret == INT_MAX, c,
1818 "xattr for missing inode %llu",
1819 k.k->p.inode))
1820 return bch2_btree_delete_at(trans, iter, 0);
1821
1822 if (ret == INT_MAX)
1823 return 0;
1824
1825 ret = 0;
1826
1827 ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1828 fsck_err:
1829 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1830 bch_err_fn(c, ret);
1831 return ret;
1832 }
1833
1834 /*
1835 * Walk xattrs: verify that they all have a corresponding inode
1836 */
1837 noinline_for_stack
1838 static int check_xattrs(struct bch_fs *c)
1839 {
1840 struct inode_walker inode = inode_walker_init();
1841 struct bch_hash_info hash_info;
1842 struct btree_trans trans;
1843 struct btree_iter iter;
1844 struct bkey_s_c k;
1845 int ret = 0;
1846
1847 bch_verbose(c, "checking xattrs");
1848
1849 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1850
1851 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_xattrs,
1852 POS(BCACHEFS_ROOT_INO, 0),
1853 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1854 k,
1855 NULL, NULL,
1856 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1857 check_xattr(&trans, &iter, k, &hash_info, &inode));
1858
1859 bch2_trans_exit(&trans);
1860
1861 if (ret)
1862 bch_err_fn(c, ret);
1863 return ret;
1864 }
1865
1866 static int check_root_trans(struct btree_trans *trans)
1867 {
1868 struct bch_fs *c = trans->c;
1869 struct bch_inode_unpacked root_inode;
1870 u32 snapshot;
1871 u64 inum;
1872 int ret;
1873
1874 ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1875 if (ret && !bch2_err_matches(ret, ENOENT))
1876 return ret;
1877
1878 if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1879 struct bkey_i_subvolume root_subvol;
1880
1881 snapshot = U32_MAX;
1882 inum = BCACHEFS_ROOT_INO;
1883
1884 bkey_subvolume_init(&root_subvol.k_i);
1885 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1886 root_subvol.v.flags = 0;
1887 root_subvol.v.snapshot = cpu_to_le32(snapshot);
1888 root_subvol.v.inode = cpu_to_le64(inum);
1889 ret = commit_do(trans, NULL, NULL,
1890 BTREE_INSERT_NOFAIL|
1891 BTREE_INSERT_LAZY_RW,
1892 __bch2_btree_insert(trans, BTREE_ID_subvolumes,
1893 &root_subvol.k_i, 0));
1894 if (ret) {
1895 bch_err(c, "error writing root subvol: %s", bch2_err_str(ret));
1896 goto err;
1897 }
1898
1899 }
1900
1901 ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1902 if (ret && !bch2_err_matches(ret, ENOENT))
1903 return ret;
1904
1905 if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1906 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1907 "root inode not a directory")) {
1908 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1909 0, NULL);
1910 root_inode.bi_inum = inum;
1911
1912 ret = __write_inode(trans, &root_inode, snapshot);
1913 if (ret)
1914 bch_err(c, "error writing root inode: %s", bch2_err_str(ret));
1915 }
1916 err:
1917 fsck_err:
1918 return ret;
1919 }
1920
1921 /* Get root directory, create if it doesn't exist: */
1922 noinline_for_stack
1923 static int check_root(struct bch_fs *c)
1924 {
1925 int ret;
1926
1927 bch_verbose(c, "checking root directory");
1928
1929 ret = bch2_trans_do(c, NULL, NULL,
1930 BTREE_INSERT_NOFAIL|
1931 BTREE_INSERT_LAZY_RW,
1932 check_root_trans(&trans));
1933
1934 if (ret)
1935 bch_err_fn(c, ret);
1936 return ret;
1937 }
1938
1939 struct pathbuf_entry {
1940 u64 inum;
1941 u32 snapshot;
1942 };
1943
1944 typedef DARRAY(struct pathbuf_entry) pathbuf;
1945
1946 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1947 {
1948 struct pathbuf_entry *i;
1949
1950 darray_for_each(*p, i)
1951 if (i->inum == inum &&
1952 i->snapshot == snapshot)
1953 return true;
1954
1955 return false;
1956 }
1957
1958 static int path_down(struct bch_fs *c, pathbuf *p,
1959 u64 inum, u32 snapshot)
1960 {
1961 int ret = darray_push(p, ((struct pathbuf_entry) {
1962 .inum = inum,
1963 .snapshot = snapshot,
1964 }));
1965
1966 if (ret)
1967 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1968 p->size);
1969 return ret;
1970 }
1971
1972 /*
1973 * Check that a given inode is reachable from the root:
1974 *
1975 * XXX: we should also be verifying that inodes are in the right subvolumes
1976 */
1977 static int check_path(struct btree_trans *trans,
1978 pathbuf *p,
1979 struct bch_inode_unpacked *inode,
1980 u32 snapshot)
1981 {
1982 struct bch_fs *c = trans->c;
1983 int ret = 0;
1984
1985 snapshot = bch2_snapshot_equiv(c, snapshot);
1986 p->nr = 0;
1987
1988 while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1989 inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1990 struct btree_iter dirent_iter;
1991 struct bkey_s_c_dirent d;
1992 u32 parent_snapshot = snapshot;
1993
1994 if (inode->bi_subvol) {
1995 u64 inum;
1996
1997 ret = subvol_lookup(trans, inode->bi_parent_subvol,
1998 &parent_snapshot, &inum);
1999 if (ret)
2000 break;
2001 }
2002
2003 ret = lockrestart_do(trans,
2004 PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
2005 SPOS(inode->bi_dir, inode->bi_dir_offset,
2006 parent_snapshot))).k));
2007 if (ret && !bch2_err_matches(ret, ENOENT))
2008 break;
2009
2010 if (!ret && !dirent_points_to_inode(d, inode)) {
2011 bch2_trans_iter_exit(trans, &dirent_iter);
2012 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2013 }
2014
2015 if (bch2_err_matches(ret, ENOENT)) {
2016 if (fsck_err(c, "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2017 inode->bi_inum, snapshot,
2018 bch2_d_type_str(inode_d_type(inode)),
2019 inode->bi_nlink,
2020 inode->bi_dir,
2021 inode->bi_dir_offset))
2022 ret = reattach_inode(trans, inode, snapshot);
2023 break;
2024 }
2025
2026 bch2_trans_iter_exit(trans, &dirent_iter);
2027
2028 if (!S_ISDIR(inode->bi_mode))
2029 break;
2030
2031 ret = path_down(c, p, inode->bi_inum, snapshot);
2032 if (ret) {
2033 bch_err(c, "memory allocation failure");
2034 return ret;
2035 }
2036
2037 snapshot = parent_snapshot;
2038
2039 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2040 if (ret) {
2041 /* Should have been caught in dirents pass */
2042 bch_err(c, "error looking up parent directory: %i", ret);
2043 break;
2044 }
2045
2046 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2047 struct pathbuf_entry *i;
2048
2049 /* XXX print path */
2050 bch_err(c, "directory structure loop");
2051
2052 darray_for_each(*p, i)
2053 pr_err("%llu:%u", i->inum, i->snapshot);
2054 pr_err("%llu:%u", inode->bi_inum, snapshot);
2055
2056 if (!fsck_err(c, "directory structure loop"))
2057 return 0;
2058
2059 ret = commit_do(trans, NULL, NULL,
2060 BTREE_INSERT_NOFAIL|
2061 BTREE_INSERT_LAZY_RW,
2062 remove_backpointer(trans, inode));
2063 if (ret) {
2064 bch_err(c, "error removing dirent: %i", ret);
2065 break;
2066 }
2067
2068 ret = reattach_inode(trans, inode, snapshot);
2069 }
2070 }
2071 fsck_err:
2072 if (ret)
2073 bch_err_fn(c, ret);
2074 return ret;
2075 }
2076
2077 /*
2078 * Check for unreachable inodes, as well as loops in the directory structure:
2079 * After check_dirents(), if an inode backpointer doesn't exist that means it's
2080 * unreachable:
2081 */
2082 noinline_for_stack
2083 static int check_directory_structure(struct bch_fs *c)
2084 {
2085 struct btree_trans trans;
2086 struct btree_iter iter;
2087 struct bkey_s_c k;
2088 struct bch_inode_unpacked u;
2089 pathbuf path = { 0, };
2090 int ret;
2091
2092 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2093
2094 for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
2095 BTREE_ITER_INTENT|
2096 BTREE_ITER_PREFETCH|
2097 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2098 if (!bkey_is_inode(k.k))
2099 continue;
2100
2101 ret = bch2_inode_unpack(k, &u);
2102 if (ret) {
2103 /* Should have been caught earlier in fsck: */
2104 bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2105 break;
2106 }
2107
2108 if (u.bi_flags & BCH_INODE_UNLINKED)
2109 continue;
2110
2111 ret = check_path(&trans, &path, &u, iter.pos.snapshot);
2112 if (ret)
2113 break;
2114 }
2115 bch2_trans_iter_exit(&trans, &iter);
2116 bch2_trans_exit(&trans);
2117 darray_exit(&path);
2118
2119 if (ret)
2120 bch_err_fn(c, ret);
2121 return ret;
2122 }
2123
2124 struct nlink_table {
2125 size_t nr;
2126 size_t size;
2127
2128 struct nlink {
2129 u64 inum;
2130 u32 snapshot;
2131 u32 count;
2132 } *d;
2133 };
2134
2135 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2136 u64 inum, u32 snapshot)
2137 {
2138 if (t->nr == t->size) {
2139 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2140 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2141
2142 if (!d) {
2143 bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2144 new_size);
2145 return -BCH_ERR_ENOMEM_fsck_add_nlink;
2146 }
2147
2148 if (t->d)
2149 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2150 kvfree(t->d);
2151
2152 t->d = d;
2153 t->size = new_size;
2154 }
2155
2156
2157 t->d[t->nr++] = (struct nlink) {
2158 .inum = inum,
2159 .snapshot = snapshot,
2160 };
2161
2162 return 0;
2163 }
2164
2165 static int nlink_cmp(const void *_l, const void *_r)
2166 {
2167 const struct nlink *l = _l;
2168 const struct nlink *r = _r;
2169
2170 return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2171 }
2172
2173 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2174 struct nlink_table *links,
2175 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2176 {
2177 struct nlink *link, key = {
2178 .inum = inum, .snapshot = U32_MAX,
2179 };
2180
2181 if (inum < range_start || inum >= range_end)
2182 return;
2183
2184 link = __inline_bsearch(&key, links->d, links->nr,
2185 sizeof(links->d[0]), nlink_cmp);
2186 if (!link)
2187 return;
2188
2189 while (link > links->d && link[0].inum == link[-1].inum)
2190 --link;
2191
2192 for (; link < links->d + links->nr && link->inum == inum; link++)
2193 if (ref_visible(c, s, snapshot, link->snapshot)) {
2194 link->count++;
2195 if (link->snapshot >= snapshot)
2196 break;
2197 }
2198 }
2199
2200 noinline_for_stack
2201 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2202 struct nlink_table *t,
2203 u64 start, u64 *end)
2204 {
2205 struct btree_trans trans;
2206 struct btree_iter iter;
2207 struct bkey_s_c k;
2208 struct bch_inode_unpacked u;
2209 int ret = 0;
2210
2211 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2212
2213 for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2214 POS(0, start),
2215 BTREE_ITER_INTENT|
2216 BTREE_ITER_PREFETCH|
2217 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2218 if (!bkey_is_inode(k.k))
2219 continue;
2220
2221 /* Should never fail, checked by bch2_inode_invalid: */
2222 BUG_ON(bch2_inode_unpack(k, &u));
2223
2224 /*
2225 * Backpointer and directory structure checks are sufficient for
2226 * directories, since they can't have hardlinks:
2227 */
2228 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2229 continue;
2230
2231 if (!u.bi_nlink)
2232 continue;
2233
2234 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2235 if (ret) {
2236 *end = k.k->p.offset;
2237 ret = 0;
2238 break;
2239 }
2240
2241 }
2242 bch2_trans_iter_exit(&trans, &iter);
2243 bch2_trans_exit(&trans);
2244
2245 if (ret)
2246 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2247
2248 return ret;
2249 }
2250
2251 noinline_for_stack
2252 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2253 u64 range_start, u64 range_end)
2254 {
2255 struct btree_trans trans;
2256 struct snapshots_seen s;
2257 struct btree_iter iter;
2258 struct bkey_s_c k;
2259 struct bkey_s_c_dirent d;
2260 int ret;
2261
2262 snapshots_seen_init(&s);
2263
2264 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2265
2266 for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
2267 BTREE_ITER_INTENT|
2268 BTREE_ITER_PREFETCH|
2269 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2270 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2271 if (ret)
2272 break;
2273
2274 switch (k.k->type) {
2275 case KEY_TYPE_dirent:
2276 d = bkey_s_c_to_dirent(k);
2277
2278 if (d.v->d_type != DT_DIR &&
2279 d.v->d_type != DT_SUBVOL)
2280 inc_link(c, &s, links, range_start, range_end,
2281 le64_to_cpu(d.v->d_inum),
2282 bch2_snapshot_equiv(c, d.k->p.snapshot));
2283 break;
2284 }
2285 }
2286 bch2_trans_iter_exit(&trans, &iter);
2287
2288 if (ret)
2289 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2290
2291 bch2_trans_exit(&trans);
2292 snapshots_seen_exit(&s);
2293 return ret;
2294 }
2295
2296 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2297 struct bkey_s_c k,
2298 struct nlink_table *links,
2299 size_t *idx, u64 range_end)
2300 {
2301 struct bch_fs *c = trans->c;
2302 struct bch_inode_unpacked u;
2303 struct nlink *link = &links->d[*idx];
2304 int ret = 0;
2305
2306 if (k.k->p.offset >= range_end)
2307 return 1;
2308
2309 if (!bkey_is_inode(k.k))
2310 return 0;
2311
2312 BUG_ON(bch2_inode_unpack(k, &u));
2313
2314 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2315 return 0;
2316
2317 if (!u.bi_nlink)
2318 return 0;
2319
2320 while ((cmp_int(link->inum, k.k->p.offset) ?:
2321 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2322 BUG_ON(*idx == links->nr);
2323 link = &links->d[++*idx];
2324 }
2325
2326 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2327 "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2328 u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2329 bch2_inode_nlink_get(&u), link->count)) {
2330 bch2_inode_nlink_set(&u, link->count);
2331 ret = __write_inode(trans, &u, k.k->p.snapshot);
2332 }
2333 fsck_err:
2334 return ret;
2335 }
2336
2337 noinline_for_stack
2338 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2339 struct nlink_table *links,
2340 u64 range_start, u64 range_end)
2341 {
2342 struct btree_trans trans;
2343 struct btree_iter iter;
2344 struct bkey_s_c k;
2345 size_t idx = 0;
2346 int ret = 0;
2347
2348 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2349
2350 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
2351 POS(0, range_start),
2352 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2353 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
2354 check_nlinks_update_inode(&trans, &iter, k, links, &idx, range_end));
2355
2356 bch2_trans_exit(&trans);
2357
2358 if (ret < 0) {
2359 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2360 return ret;
2361 }
2362
2363 return 0;
2364 }
2365
2366 noinline_for_stack
2367 static int check_nlinks(struct bch_fs *c)
2368 {
2369 struct nlink_table links = { 0 };
2370 u64 this_iter_range_start, next_iter_range_start = 0;
2371 int ret = 0;
2372
2373 bch_verbose(c, "checking inode nlinks");
2374
2375 do {
2376 this_iter_range_start = next_iter_range_start;
2377 next_iter_range_start = U64_MAX;
2378
2379 ret = check_nlinks_find_hardlinks(c, &links,
2380 this_iter_range_start,
2381 &next_iter_range_start);
2382
2383 ret = check_nlinks_walk_dirents(c, &links,
2384 this_iter_range_start,
2385 next_iter_range_start);
2386 if (ret)
2387 break;
2388
2389 ret = check_nlinks_update_hardlinks(c, &links,
2390 this_iter_range_start,
2391 next_iter_range_start);
2392 if (ret)
2393 break;
2394
2395 links.nr = 0;
2396 } while (next_iter_range_start != U64_MAX);
2397
2398 kvfree(links.d);
2399
2400 if (ret)
2401 bch_err_fn(c, ret);
2402 return ret;
2403 }
2404
2405 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2406 struct bkey_s_c k)
2407 {
2408 struct bkey_s_c_reflink_p p;
2409 struct bkey_i_reflink_p *u;
2410 int ret;
2411
2412 if (k.k->type != KEY_TYPE_reflink_p)
2413 return 0;
2414
2415 p = bkey_s_c_to_reflink_p(k);
2416
2417 if (!p.v->front_pad && !p.v->back_pad)
2418 return 0;
2419
2420 u = bch2_trans_kmalloc(trans, sizeof(*u));
2421 ret = PTR_ERR_OR_ZERO(u);
2422 if (ret)
2423 return ret;
2424
2425 bkey_reassemble(&u->k_i, k);
2426 u->v.front_pad = 0;
2427 u->v.back_pad = 0;
2428
2429 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2430 }
2431
2432 noinline_for_stack
2433 static int fix_reflink_p(struct bch_fs *c)
2434 {
2435 struct btree_iter iter;
2436 struct bkey_s_c k;
2437 int ret;
2438
2439 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2440 return 0;
2441
2442 bch_verbose(c, "fixing reflink_p keys");
2443
2444 ret = bch2_trans_run(c,
2445 for_each_btree_key_commit(&trans, iter,
2446 BTREE_ID_extents, POS_MIN,
2447 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2448 BTREE_ITER_ALL_SNAPSHOTS, k,
2449 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
2450 fix_reflink_p_key(&trans, &iter, k)));
2451
2452 if (ret)
2453 bch_err_fn(c, ret);
2454 return ret;
2455 }
2456
2457 /*
2458 * Checks for inconsistencies that shouldn't happen, unless we have a bug.
2459 * Doesn't fix them yet, mainly because they haven't yet been observed:
2460 */
2461 int bch2_fsck_full(struct bch_fs *c)
2462 {
2463 int ret;
2464 again:
2465 ret = bch2_fs_check_snapshot_trees(c);
2466 bch2_fs_check_snapshots(c) ?:
2467 bch2_fs_check_subvols(c) ?:
2468 bch2_delete_dead_snapshots(c) ?:
2469 check_inodes(c, true) ?:
2470 check_extents(c) ?:
2471 check_dirents(c) ?:
2472 check_xattrs(c) ?:
2473 check_root(c) ?:
2474 check_directory_structure(c) ?:
2475 check_nlinks(c) ?:
2476 fix_reflink_p(c);
2477
2478 if (bch2_err_matches(ret, BCH_ERR_need_snapshot_cleanup)) {
2479 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
2480 goto again;
2481 }
2482
2483 return ret;
2484 }
2485
2486 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
2487 {
2488 return bch2_fs_check_snapshots(c) ?:
2489 bch2_fs_check_subvols(c) ?:
2490 bch2_delete_dead_snapshots(c) ?:
2491 check_inodes(c, false);
2492 }
A mirror of Linus' kernel repository