1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "bkey_buf.h"
5 #include "btree_cache.h"
6 #include "btree_update.h"
7 #include "buckets.h"
8 #include "darray.h"
9 #include "dirent.h"
10 #include "error.h"
11 #include "fs.h"
12 #include "fs-common.h"
13 #include "fsck.h"
14 #include "inode.h"
15 #include "keylist.h"
16 #include "recovery_passes.h"
17 #include "snapshot.h"
18 #include "super.h"
19 #include "xattr.h"
20
21 #include <linux/bsearch.h>
22 #include <linux/dcache.h> /* struct qstr */
23
24 /*
25 * XXX: this is handling transaction restarts without returning
26 * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
27 */
28 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
29 u32 snapshot)
30 {
31 u64 sectors = 0;
32
33 int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
34 SPOS(inum, 0, snapshot),
35 POS(inum, U64_MAX),
36 0, k, ({
37 if (bkey_extent_is_allocation(k.k))
38 sectors += k.k->size;
39 0;
40 }));
41
42 return ret ?: sectors;
43 }
44
45 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
46 u32 snapshot)
47 {
48 u64 subdirs = 0;
49
50 int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
51 SPOS(inum, 0, snapshot),
52 POS(inum, U64_MAX),
53 0, k, ({
54 if (k.k->type == KEY_TYPE_dirent &&
55 bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
56 subdirs++;
57 0;
58 }));
59
60 return ret ?: subdirs;
61 }
62
63 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
64 u32 *snapshot, u64 *inum)
65 {
66 struct bch_subvolume s;
67 int ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
68
69 *snapshot = le32_to_cpu(s.snapshot);
70 *inum = le64_to_cpu(s.inode);
71 return ret;
72 }
73
74 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
75 struct bch_inode_unpacked *inode)
76 {
77 struct btree_iter iter;
78 struct bkey_s_c k;
79 int ret;
80
81 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inode_nr),
82 BTREE_ITER_all_snapshots, k, ret) {
83 if (k.k->p.offset != inode_nr)
84 break;
85 if (!bkey_is_inode(k.k))
86 continue;
87 ret = bch2_inode_unpack(k, inode);
88 goto found;
89 }
90 ret = -BCH_ERR_ENOENT_inode;
91 found:
92 bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
93 bch2_trans_iter_exit(trans, &iter);
94 return ret;
95 }
96
97 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
98 struct bch_inode_unpacked *inode,
99 u32 *snapshot)
100 {
101 struct btree_iter iter;
102 struct bkey_s_c k;
103 int ret;
104
105 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
106 SPOS(0, inode_nr, *snapshot), 0);
107 ret = bkey_err(k);
108 if (ret)
109 goto err;
110
111 ret = bkey_is_inode(k.k)
112 ? bch2_inode_unpack(k, inode)
113 : -BCH_ERR_ENOENT_inode;
114 if (!ret)
115 *snapshot = iter.pos.snapshot;
116 err:
117 bch2_trans_iter_exit(trans, &iter);
118 return ret;
119 }
120
121 static int lookup_dirent_in_snapshot(struct btree_trans *trans,
122 struct bch_hash_info hash_info,
123 subvol_inum dir, struct qstr *name,
124 u64 *target, unsigned *type, u32 snapshot)
125 {
126 struct btree_iter iter;
127 struct bkey_s_c k = bch2_hash_lookup_in_snapshot(trans, &iter, bch2_dirent_hash_desc,
128 &hash_info, dir, name, 0, snapshot);
129 int ret = bkey_err(k);
130 if (ret)
131 return ret;
132
133 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
134 *target = le64_to_cpu(d.v->d_inum);
135 *type = d.v->d_type;
136 bch2_trans_iter_exit(trans, &iter);
137 return 0;
138 }
139
140 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
141 {
142 struct bch_fs *c = trans->c;
143 struct btree_iter iter;
144 struct bch_inode_unpacked dir_inode;
145 struct bch_hash_info dir_hash_info;
146 int ret;
147
148 ret = lookup_first_inode(trans, pos.inode, &dir_inode);
149 if (ret)
150 goto err;
151
152 dir_hash_info = bch2_hash_info_init(c, &dir_inode);
153
154 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_intent);
155
156 ret = bch2_btree_iter_traverse(&iter) ?:
157 bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
158 &dir_hash_info, &iter,
159 BTREE_UPDATE_internal_snapshot_node);
160 bch2_trans_iter_exit(trans, &iter);
161 err:
162 bch_err_fn(c, ret);
163 return ret;
164 }
165
166 /* Get lost+found, create if it doesn't exist: */
167 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
168 struct bch_inode_unpacked *lostfound,
169 u64 reattaching_inum)
170 {
171 struct bch_fs *c = trans->c;
172 struct qstr lostfound_str = QSTR("lost+found");
173 u64 inum = 0;
174 unsigned d_type = 0;
175 int ret;
176
177 struct bch_snapshot_tree st;
178 ret = bch2_snapshot_tree_lookup(trans,
179 bch2_snapshot_tree(c, snapshot), &st);
180 if (ret)
181 return ret;
182
183 subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
184
185 struct bch_subvolume subvol;
186 ret = bch2_subvolume_get(trans, le32_to_cpu(st.master_subvol),
187 false, 0, &subvol);
188 bch_err_msg(c, ret, "looking up root subvol %u for snapshot %u",
189 le32_to_cpu(st.master_subvol), snapshot);
190 if (ret)
191 return ret;
192
193 if (!subvol.inode) {
194 struct btree_iter iter;
195 struct bkey_i_subvolume *subvol = bch2_bkey_get_mut_typed(trans, &iter,
196 BTREE_ID_subvolumes, POS(0, le32_to_cpu(st.master_subvol)),
197 0, subvolume);
198 ret = PTR_ERR_OR_ZERO(subvol);
199 if (ret)
200 return ret;
201
202 subvol->v.inode = cpu_to_le64(reattaching_inum);
203 bch2_trans_iter_exit(trans, &iter);
204 }
205
206 root_inum.inum = le64_to_cpu(subvol.inode);
207
208 struct bch_inode_unpacked root_inode;
209 struct bch_hash_info root_hash_info;
210 u32 root_inode_snapshot = snapshot;
211 ret = lookup_inode(trans, root_inum.inum, &root_inode, &root_inode_snapshot);
212 bch_err_msg(c, ret, "looking up root inode %llu for subvol %u",
213 root_inum.inum, le32_to_cpu(st.master_subvol));
214 if (ret)
215 return ret;
216
217 root_hash_info = bch2_hash_info_init(c, &root_inode);
218
219 ret = lookup_dirent_in_snapshot(trans, root_hash_info, root_inum,
220 &lostfound_str, &inum, &d_type, snapshot);
221 if (bch2_err_matches(ret, ENOENT))
222 goto create_lostfound;
223
224 bch_err_fn(c, ret);
225 if (ret)
226 return ret;
227
228 if (d_type != DT_DIR) {
229 bch_err(c, "error looking up lost+found: not a directory");
230 return -BCH_ERR_ENOENT_not_directory;
231 }
232
233 /*
234 * The bch2_check_dirents pass has already run, dangling dirents
235 * shouldn't exist here:
236 */
237 ret = lookup_inode(trans, inum, lostfound, &snapshot);
238 bch_err_msg(c, ret, "looking up lost+found %llu:%u in (root inode %llu, snapshot root %u)",
239 inum, snapshot, root_inum.inum, bch2_snapshot_root(c, snapshot));
240 return ret;
241
242 create_lostfound:
243 /*
244 * XXX: we could have a nicer log message here if we had a nice way to
245 * walk backpointers to print a path
246 */
247 bch_notice(c, "creating lost+found in snapshot %u", le32_to_cpu(st.root_snapshot));
248
249 u64 now = bch2_current_time(c);
250 struct btree_iter lostfound_iter = { NULL };
251 u64 cpu = raw_smp_processor_id();
252
253 bch2_inode_init_early(c, lostfound);
254 bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
255 lostfound->bi_dir = root_inode.bi_inum;
256
257 root_inode.bi_nlink++;
258
259 ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
260 if (ret)
261 goto err;
262
263 bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
264 ret = bch2_btree_iter_traverse(&lostfound_iter);
265 if (ret)
266 goto err;
267
268 ret = bch2_dirent_create_snapshot(trans,
269 0, root_inode.bi_inum, snapshot, &root_hash_info,
270 mode_to_type(lostfound->bi_mode),
271 &lostfound_str,
272 lostfound->bi_inum,
273 &lostfound->bi_dir_offset,
274 STR_HASH_must_create) ?:
275 bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
276 BTREE_UPDATE_internal_snapshot_node);
277 err:
278 bch_err_msg(c, ret, "creating lost+found");
279 bch2_trans_iter_exit(trans, &lostfound_iter);
280 return ret;
281 }
282
283 static int reattach_inode(struct btree_trans *trans,
284 struct bch_inode_unpacked *inode,
285 u32 inode_snapshot)
286 {
287 struct bch_fs *c = trans->c;
288 struct bch_hash_info dir_hash;
289 struct bch_inode_unpacked lostfound;
290 char name_buf[20];
291 struct qstr name;
292 u64 dir_offset = 0;
293 u32 dirent_snapshot = inode_snapshot;
294 int ret;
295
296 if (inode->bi_subvol) {
297 inode->bi_parent_subvol = BCACHEFS_ROOT_SUBVOL;
298
299 u64 root_inum;
300 ret = subvol_lookup(trans, inode->bi_parent_subvol,
301 &dirent_snapshot, &root_inum);
302 if (ret)
303 return ret;
304
305 snprintf(name_buf, sizeof(name_buf), "subvol-%u", inode->bi_subvol);
306 } else {
307 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
308 }
309
310 ret = lookup_lostfound(trans, dirent_snapshot, &lostfound, inode->bi_inum);
311 if (ret)
312 return ret;
313
314 if (S_ISDIR(inode->bi_mode)) {
315 lostfound.bi_nlink++;
316
317 ret = __bch2_fsck_write_inode(trans, &lostfound, U32_MAX);
318 if (ret)
319 return ret;
320 }
321
322 dir_hash = bch2_hash_info_init(c, &lostfound);
323
324 name = (struct qstr) QSTR(name_buf);
325
326 ret = bch2_dirent_create_snapshot(trans,
327 inode->bi_parent_subvol, lostfound.bi_inum,
328 dirent_snapshot,
329 &dir_hash,
330 inode_d_type(inode),
331 &name,
332 inode->bi_subvol ?: inode->bi_inum,
333 &dir_offset,
334 STR_HASH_must_create);
335 if (ret) {
336 bch_err_msg(c, ret, "error creating dirent");
337 return ret;
338 }
339
340 inode->bi_dir = lostfound.bi_inum;
341 inode->bi_dir_offset = dir_offset;
342
343 return __bch2_fsck_write_inode(trans, inode, inode_snapshot);
344 }
345
346 static int remove_backpointer(struct btree_trans *trans,
347 struct bch_inode_unpacked *inode)
348 {
349 struct btree_iter iter;
350 struct bkey_s_c_dirent d;
351 int ret;
352
353 d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
354 POS(inode->bi_dir, inode->bi_dir_offset), 0,
355 dirent);
356 ret = bkey_err(d) ?:
357 __remove_dirent(trans, d.k->p);
358 bch2_trans_iter_exit(trans, &iter);
359 return ret;
360 }
361
362 static int reattach_subvol(struct btree_trans *trans, struct bkey_s_c_subvolume s)
363 {
364 struct bch_fs *c = trans->c;
365
366 struct bch_inode_unpacked inode;
367 int ret = bch2_inode_find_by_inum_trans(trans,
368 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
369 &inode);
370 if (ret)
371 return ret;
372
373 ret = remove_backpointer(trans, &inode);
374 bch_err_msg(c, ret, "removing dirent");
375 if (ret)
376 return ret;
377
378 ret = reattach_inode(trans, &inode, le32_to_cpu(s.v->snapshot));
379 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
380 return ret;
381 }
382
383 static int reconstruct_subvol(struct btree_trans *trans, u32 snapshotid, u32 subvolid, u64 inum)
384 {
385 struct bch_fs *c = trans->c;
386
387 if (!bch2_snapshot_is_leaf(c, snapshotid)) {
388 bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
389 return -BCH_ERR_fsck_repair_unimplemented;
390 }
391
392 /*
393 * If inum isn't set, that means we're being called from check_dirents,
394 * not check_inodes - the root of this subvolume doesn't exist or we
395 * would have found it there:
396 */
397 if (!inum) {
398 struct btree_iter inode_iter = {};
399 struct bch_inode_unpacked new_inode;
400 u64 cpu = raw_smp_processor_id();
401
402 bch2_inode_init_early(c, &new_inode);
403 bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, S_IFDIR|0755, 0, NULL);
404
405 new_inode.bi_subvol = subvolid;
406
407 int ret = bch2_inode_create(trans, &inode_iter, &new_inode, snapshotid, cpu) ?:
408 bch2_btree_iter_traverse(&inode_iter) ?:
409 bch2_inode_write(trans, &inode_iter, &new_inode);
410 bch2_trans_iter_exit(trans, &inode_iter);
411 if (ret)
412 return ret;
413
414 inum = new_inode.bi_inum;
415 }
416
417 bch_info(c, "reconstructing subvol %u with root inode %llu", subvolid, inum);
418
419 struct bkey_i_subvolume *new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
420 int ret = PTR_ERR_OR_ZERO(new_subvol);
421 if (ret)
422 return ret;
423
424 bkey_subvolume_init(&new_subvol->k_i);
425 new_subvol->k.p.offset = subvolid;
426 new_subvol->v.snapshot = cpu_to_le32(snapshotid);
427 new_subvol->v.inode = cpu_to_le64(inum);
428 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &new_subvol->k_i, 0);
429 if (ret)
430 return ret;
431
432 struct btree_iter iter;
433 struct bkey_i_snapshot *s = bch2_bkey_get_mut_typed(trans, &iter,
434 BTREE_ID_snapshots, POS(0, snapshotid),
435 0, snapshot);
436 ret = PTR_ERR_OR_ZERO(s);
437 bch_err_msg(c, ret, "getting snapshot %u", snapshotid);
438 if (ret)
439 return ret;
440
441 u32 snapshot_tree = le32_to_cpu(s->v.tree);
442
443 s->v.subvol = cpu_to_le32(subvolid);
444 SET_BCH_SNAPSHOT_SUBVOL(&s->v, true);
445 bch2_trans_iter_exit(trans, &iter);
446
447 struct bkey_i_snapshot_tree *st = bch2_bkey_get_mut_typed(trans, &iter,
448 BTREE_ID_snapshot_trees, POS(0, snapshot_tree),
449 0, snapshot_tree);
450 ret = PTR_ERR_OR_ZERO(st);
451 bch_err_msg(c, ret, "getting snapshot tree %u", snapshot_tree);
452 if (ret)
453 return ret;
454
455 if (!st->v.master_subvol)
456 st->v.master_subvol = cpu_to_le32(subvolid);
457
458 bch2_trans_iter_exit(trans, &iter);
459 return 0;
460 }
461
462 static int reconstruct_inode(struct btree_trans *trans, enum btree_id btree, u32 snapshot, u64 inum)
463 {
464 struct bch_fs *c = trans->c;
465 unsigned i_mode = S_IFREG;
466 u64 i_size = 0;
467
468 switch (btree) {
469 case BTREE_ID_extents: {
470 struct btree_iter iter = {};
471
472 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, SPOS(inum, U64_MAX, snapshot), 0);
473 struct bkey_s_c k = bch2_btree_iter_peek_prev(&iter);
474 bch2_trans_iter_exit(trans, &iter);
475 int ret = bkey_err(k);
476 if (ret)
477 return ret;
478
479 i_size = k.k->p.offset << 9;
480 break;
481 }
482 case BTREE_ID_dirents:
483 i_mode = S_IFDIR;
484 break;
485 case BTREE_ID_xattrs:
486 break;
487 default:
488 BUG();
489 }
490
491 struct bch_inode_unpacked new_inode;
492 bch2_inode_init_early(c, &new_inode);
493 bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, i_mode|0600, 0, NULL);
494 new_inode.bi_size = i_size;
495 new_inode.bi_inum = inum;
496
497 return __bch2_fsck_write_inode(trans, &new_inode, snapshot);
498 }
499
500 struct snapshots_seen {
501 struct bpos pos;
502 snapshot_id_list ids;
503 };
504
505 static inline void snapshots_seen_exit(struct snapshots_seen *s)
506 {
507 darray_exit(&s->ids);
508 }
509
510 static inline void snapshots_seen_init(struct snapshots_seen *s)
511 {
512 memset(s, 0, sizeof(*s));
513 }
514
515 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
516 {
517 u32 *i;
518 __darray_for_each(s->ids, i) {
519 if (*i == id)
520 return 0;
521 if (*i > id)
522 break;
523 }
524
525 int ret = darray_insert_item(&s->ids, i - s->ids.data, id);
526 if (ret)
527 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
528 s->ids.size);
529 return ret;
530 }
531
532 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
533 enum btree_id btree_id, struct bpos pos)
534 {
535 if (!bkey_eq(s->pos, pos))
536 s->ids.nr = 0;
537 s->pos = pos;
538
539 return snapshot_list_add_nodup(c, &s->ids, pos.snapshot);
540 }
541
542 /**
543 * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
544 * and @ancestor hasn't been overwritten in @seen
545 *
546 * @c: filesystem handle
547 * @seen: list of snapshot ids already seen at current position
548 * @id: descendent snapshot id
549 * @ancestor: ancestor snapshot id
550 *
551 * Returns: whether key in @ancestor snapshot is visible in @id snapshot
552 */
553 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
554 u32 id, u32 ancestor)
555 {
556 ssize_t i;
557
558 EBUG_ON(id > ancestor);
559
560 /* @ancestor should be the snapshot most recently added to @seen */
561 EBUG_ON(ancestor != seen->pos.snapshot);
562 EBUG_ON(ancestor != darray_last(seen->ids));
563
564 if (id == ancestor)
565 return true;
566
567 if (!bch2_snapshot_is_ancestor(c, id, ancestor))
568 return false;
569
570 /*
571 * We know that @id is a descendant of @ancestor, we're checking if
572 * we've seen a key that overwrote @ancestor - i.e. also a descendent of
573 * @ascestor and with @id as a descendent.
574 *
575 * But we already know that we're scanning IDs between @id and @ancestor
576 * numerically, since snapshot ID lists are kept sorted, so if we find
577 * an id that's an ancestor of @id we're done:
578 */
579
580 for (i = seen->ids.nr - 2;
581 i >= 0 && seen->ids.data[i] >= id;
582 --i)
583 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i]))
584 return false;
585
586 return true;
587 }
588
589 /**
590 * ref_visible - given a key with snapshot id @src that points to a key with
591 * snapshot id @dst, test whether there is some snapshot in which @dst is
592 * visible.
593 *
594 * @c: filesystem handle
595 * @s: list of snapshot IDs already seen at @src
596 * @src: snapshot ID of src key
597 * @dst: snapshot ID of dst key
598 * Returns: true if there is some snapshot in which @dst is visible
599 *
600 * Assumes we're visiting @src keys in natural key order
601 */
602 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
603 u32 src, u32 dst)
604 {
605 return dst <= src
606 ? key_visible_in_snapshot(c, s, dst, src)
607 : bch2_snapshot_is_ancestor(c, src, dst);
608 }
609
610 static int ref_visible2(struct bch_fs *c,
611 u32 src, struct snapshots_seen *src_seen,
612 u32 dst, struct snapshots_seen *dst_seen)
613 {
614 if (dst > src) {
615 swap(dst, src);
616 swap(dst_seen, src_seen);
617 }
618 return key_visible_in_snapshot(c, src_seen, dst, src);
619 }
620
621 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \
622 for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && \
623 (_i)->snapshot <= (_snapshot); _i++) \
624 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
625
626 struct inode_walker_entry {
627 struct bch_inode_unpacked inode;
628 u32 snapshot;
629 bool seen_this_pos;
630 u64 count;
631 };
632
633 struct inode_walker {
634 bool first_this_inode;
635 bool recalculate_sums;
636 struct bpos last_pos;
637
638 DARRAY(struct inode_walker_entry) inodes;
639 };
640
641 static void inode_walker_exit(struct inode_walker *w)
642 {
643 darray_exit(&w->inodes);
644 }
645
646 static struct inode_walker inode_walker_init(void)
647 {
648 return (struct inode_walker) { 0, };
649 }
650
651 static int add_inode(struct bch_fs *c, struct inode_walker *w,
652 struct bkey_s_c inode)
653 {
654 struct bch_inode_unpacked u;
655
656 BUG_ON(bch2_inode_unpack(inode, &u));
657
658 return darray_push(&w->inodes, ((struct inode_walker_entry) {
659 .inode = u,
660 .snapshot = inode.k->p.snapshot,
661 }));
662 }
663
664 static int get_inodes_all_snapshots(struct btree_trans *trans,
665 struct inode_walker *w, u64 inum)
666 {
667 struct bch_fs *c = trans->c;
668 struct btree_iter iter;
669 struct bkey_s_c k;
670 int ret;
671
672 w->recalculate_sums = false;
673 w->inodes.nr = 0;
674
675 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
676 BTREE_ITER_all_snapshots, k, ret) {
677 if (k.k->p.offset != inum)
678 break;
679
680 if (bkey_is_inode(k.k))
681 add_inode(c, w, k);
682 }
683 bch2_trans_iter_exit(trans, &iter);
684
685 if (ret)
686 return ret;
687
688 w->first_this_inode = true;
689 return 0;
690 }
691
692 static struct inode_walker_entry *
693 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w, struct bkey_s_c k)
694 {
695 bool is_whiteout = k.k->type == KEY_TYPE_whiteout;
696
697 struct inode_walker_entry *i;
698 __darray_for_each(w->inodes, i)
699 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i->snapshot))
700 goto found;
701
702 return NULL;
703 found:
704 BUG_ON(k.k->p.snapshot > i->snapshot);
705
706 if (k.k->p.snapshot != i->snapshot && !is_whiteout) {
707 struct inode_walker_entry new = *i;
708
709 new.snapshot = k.k->p.snapshot;
710 new.count = 0;
711
712 struct printbuf buf = PRINTBUF;
713 bch2_bkey_val_to_text(&buf, c, k);
714
715 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u\n"
716 "unexpected because we should always update the inode when we update a key in that inode\n"
717 "%s",
718 w->last_pos.inode, k.k->p.snapshot, i->snapshot, buf.buf);
719 printbuf_exit(&buf);
720
721 while (i > w->inodes.data && i[-1].snapshot > k.k->p.snapshot)
722 --i;
723
724 size_t pos = i - w->inodes.data;
725 int ret = darray_insert_item(&w->inodes, pos, new);
726 if (ret)
727 return ERR_PTR(ret);
728
729 i = w->inodes.data + pos;
730 }
731
732 return i;
733 }
734
735 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
736 struct inode_walker *w,
737 struct bkey_s_c k)
738 {
739 if (w->last_pos.inode != k.k->p.inode) {
740 int ret = get_inodes_all_snapshots(trans, w, k.k->p.inode);
741 if (ret)
742 return ERR_PTR(ret);
743 } else if (bkey_cmp(w->last_pos, k.k->p)) {
744 darray_for_each(w->inodes, i)
745 i->seen_this_pos = false;
746 }
747
748 w->last_pos = k.k->p;
749
750 return lookup_inode_for_snapshot(trans->c, w, k);
751 }
752
753 static int get_visible_inodes(struct btree_trans *trans,
754 struct inode_walker *w,
755 struct snapshots_seen *s,
756 u64 inum)
757 {
758 struct bch_fs *c = trans->c;
759 struct btree_iter iter;
760 struct bkey_s_c k;
761 int ret;
762
763 w->inodes.nr = 0;
764
765 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
766 BTREE_ITER_all_snapshots, k, ret) {
767 if (k.k->p.offset != inum)
768 break;
769
770 if (!ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot))
771 continue;
772
773 if (bkey_is_inode(k.k))
774 add_inode(c, w, k);
775
776 if (k.k->p.snapshot >= s->pos.snapshot)
777 break;
778 }
779 bch2_trans_iter_exit(trans, &iter);
780
781 return ret;
782 }
783
784 static int hash_redo_key(struct btree_trans *trans,
785 const struct bch_hash_desc desc,
786 struct bch_hash_info *hash_info,
787 struct btree_iter *k_iter, struct bkey_s_c k)
788 {
789 struct bkey_i *delete;
790 struct bkey_i *tmp;
791
792 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
793 if (IS_ERR(delete))
794 return PTR_ERR(delete);
795
796 tmp = bch2_bkey_make_mut_noupdate(trans, k);
797 if (IS_ERR(tmp))
798 return PTR_ERR(tmp);
799
800 bkey_init(&delete->k);
801 delete->k.p = k_iter->pos;
802 return bch2_btree_iter_traverse(k_iter) ?:
803 bch2_trans_update(trans, k_iter, delete, 0) ?:
804 bch2_hash_set_in_snapshot(trans, desc, hash_info,
805 (subvol_inum) { 0, k.k->p.inode },
806 k.k->p.snapshot, tmp,
807 STR_HASH_must_create|
808 BTREE_UPDATE_internal_snapshot_node) ?:
809 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
810 }
811
812 static int hash_check_key(struct btree_trans *trans,
813 const struct bch_hash_desc desc,
814 struct bch_hash_info *hash_info,
815 struct btree_iter *k_iter, struct bkey_s_c hash_k)
816 {
817 struct bch_fs *c = trans->c;
818 struct btree_iter iter = { NULL };
819 struct printbuf buf = PRINTBUF;
820 struct bkey_s_c k;
821 u64 hash;
822 int ret = 0;
823
824 if (hash_k.k->type != desc.key_type)
825 return 0;
826
827 hash = desc.hash_bkey(hash_info, hash_k);
828
829 if (likely(hash == hash_k.k->p.offset))
830 return 0;
831
832 if (hash_k.k->p.offset < hash)
833 goto bad_hash;
834
835 for_each_btree_key_norestart(trans, iter, desc.btree_id,
836 SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
837 BTREE_ITER_slots, k, ret) {
838 if (bkey_eq(k.k->p, hash_k.k->p))
839 break;
840
841 if (fsck_err_on(k.k->type == desc.key_type &&
842 !desc.cmp_bkey(k, hash_k),
843 trans, hash_table_key_duplicate,
844 "duplicate hash table keys:\n%s",
845 (printbuf_reset(&buf),
846 bch2_bkey_val_to_text(&buf, c, hash_k),
847 buf.buf))) {
848 ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
849 break;
850 }
851
852 if (bkey_deleted(k.k)) {
853 bch2_trans_iter_exit(trans, &iter);
854 goto bad_hash;
855 }
856 }
857 out:
858 bch2_trans_iter_exit(trans, &iter);
859 printbuf_exit(&buf);
860 return ret;
861 bad_hash:
862 if (fsck_err(trans, hash_table_key_wrong_offset,
863 "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
864 bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
865 (printbuf_reset(&buf),
866 bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
867 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
868 bch_err_fn(c, ret);
869 if (ret)
870 return ret;
871 ret = -BCH_ERR_transaction_restart_nested;
872 }
873 fsck_err:
874 goto out;
875 }
876
877 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
878 struct btree_iter *iter,
879 struct bpos pos)
880 {
881 return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
882 }
883
884 static struct bkey_s_c_dirent inode_get_dirent(struct btree_trans *trans,
885 struct btree_iter *iter,
886 struct bch_inode_unpacked *inode,
887 u32 *snapshot)
888 {
889 if (inode->bi_subvol) {
890 u64 inum;
891 int ret = subvol_lookup(trans, inode->bi_parent_subvol, snapshot, &inum);
892 if (ret)
893 return ((struct bkey_s_c_dirent) { .k = ERR_PTR(ret) });
894 }
895
896 return dirent_get_by_pos(trans, iter, SPOS(inode->bi_dir, inode->bi_dir_offset, *snapshot));
897 }
898
899 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
900 struct bkey_s_c_dirent d)
901 {
902 return inode->bi_dir == d.k->p.inode &&
903 inode->bi_dir_offset == d.k->p.offset;
904 }
905
906 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
907 struct bch_inode_unpacked *inode)
908 {
909 return d.v->d_type == DT_SUBVOL
910 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
911 : le64_to_cpu(d.v->d_inum) == inode->bi_inum;
912 }
913
914 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
915 {
916 struct btree_iter iter;
917 struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
918 int ret = bkey_err(k) ?: k.k->type == KEY_TYPE_set;
919 bch2_trans_iter_exit(trans, &iter);
920 return ret;
921 }
922
923 static int check_inode_dirent_inode(struct btree_trans *trans, struct bkey_s_c inode_k,
924 struct bch_inode_unpacked *inode,
925 u32 inode_snapshot, bool *write_inode)
926 {
927 struct bch_fs *c = trans->c;
928 struct printbuf buf = PRINTBUF;
929
930 struct btree_iter dirent_iter = {};
931 struct bkey_s_c_dirent d = inode_get_dirent(trans, &dirent_iter, inode, &inode_snapshot);
932 int ret = bkey_err(d);
933 if (ret && !bch2_err_matches(ret, ENOENT))
934 return ret;
935
936 if (fsck_err_on(ret,
937 trans, inode_points_to_missing_dirent,
938 "inode points to missing dirent\n%s",
939 (bch2_bkey_val_to_text(&buf, c, inode_k), buf.buf)) ||
940 fsck_err_on(!ret && !dirent_points_to_inode(d, inode),
941 trans, inode_points_to_wrong_dirent,
942 "inode points to dirent that does not point back:\n%s",
943 (bch2_bkey_val_to_text(&buf, c, inode_k),
944 prt_newline(&buf),
945 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
946 /*
947 * We just clear the backpointer fields for now. If we find a
948 * dirent that points to this inode in check_dirents(), we'll
949 * update it then; then when we get to check_path() if the
950 * backpointer is still 0 we'll reattach it.
951 */
952 inode->bi_dir = 0;
953 inode->bi_dir_offset = 0;
954 inode->bi_flags &= ~BCH_INODE_backptr_untrusted;
955 *write_inode = true;
956 }
957
958 ret = 0;
959 fsck_err:
960 bch2_trans_iter_exit(trans, &dirent_iter);
961 printbuf_exit(&buf);
962 bch_err_fn(c, ret);
963 return ret;
964 }
965
966 static bool bch2_inode_open(struct bch_fs *c, struct bpos p)
967 {
968 subvol_inum inum = {
969 .subvol = snapshot_t(c, p.snapshot)->subvol,
970 .inum = p.offset,
971 };
972
973 /* snapshot tree corruption, can't safely delete */
974 if (!inum.subvol) {
975 bch_err_ratelimited(c, "%s(): snapshot %u has no subvol", __func__, p.snapshot);
976 return true;
977 }
978
979 return __bch2_inode_hash_find(c, inum) != NULL;
980 }
981
982 static int check_inode(struct btree_trans *trans,
983 struct btree_iter *iter,
984 struct bkey_s_c k,
985 struct bch_inode_unpacked *prev,
986 struct snapshots_seen *s,
987 bool full)
988 {
989 struct bch_fs *c = trans->c;
990 struct bch_inode_unpacked u;
991 bool do_update = false;
992 int ret;
993
994 ret = bch2_check_key_has_snapshot(trans, iter, k);
995 if (ret < 0)
996 goto err;
997 if (ret)
998 return 0;
999
1000 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1001 if (ret)
1002 goto err;
1003
1004 if (!bkey_is_inode(k.k))
1005 return 0;
1006
1007 BUG_ON(bch2_inode_unpack(k, &u));
1008
1009 if (!full &&
1010 !(u.bi_flags & (BCH_INODE_i_size_dirty|
1011 BCH_INODE_i_sectors_dirty|
1012 BCH_INODE_unlinked)))
1013 return 0;
1014
1015 if (prev->bi_inum != u.bi_inum)
1016 *prev = u;
1017
1018 if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed ||
1019 inode_d_type(prev) != inode_d_type(&u),
1020 trans, inode_snapshot_mismatch,
1021 "inodes in different snapshots don't match")) {
1022 bch_err(c, "repair not implemented yet");
1023 return -BCH_ERR_fsck_repair_unimplemented;
1024 }
1025
1026 if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
1027 bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
1028 struct bpos new_min_pos;
1029
1030 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
1031 if (ret)
1032 goto err;
1033
1034 u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
1035
1036 ret = __bch2_fsck_write_inode(trans, &u, iter->pos.snapshot);
1037
1038 bch_err_msg(c, ret, "in fsck updating inode");
1039 if (ret)
1040 return ret;
1041
1042 if (!bpos_eq(new_min_pos, POS_MIN))
1043 bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
1044 return 0;
1045 }
1046
1047 if (u.bi_flags & BCH_INODE_unlinked) {
1048 ret = check_inode_deleted_list(trans, k.k->p);
1049 if (ret < 0)
1050 return ret;
1051
1052 fsck_err_on(!ret,
1053 trans, unlinked_inode_not_on_deleted_list,
1054 "inode %llu:%u unlinked, but not on deleted list",
1055 u.bi_inum, k.k->p.snapshot);
1056 ret = 0;
1057 }
1058
1059 if (u.bi_flags & BCH_INODE_unlinked &&
1060 !bch2_inode_open(c, k.k->p) &&
1061 (!c->sb.clean ||
1062 fsck_err(trans, inode_unlinked_but_clean,
1063 "filesystem marked clean, but inode %llu unlinked",
1064 u.bi_inum))) {
1065 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
1066 bch_err_msg(c, ret, "in fsck deleting inode");
1067 return ret;
1068 }
1069
1070 if (u.bi_flags & BCH_INODE_i_size_dirty &&
1071 (!c->sb.clean ||
1072 fsck_err(trans, inode_i_size_dirty_but_clean,
1073 "filesystem marked clean, but inode %llu has i_size dirty",
1074 u.bi_inum))) {
1075 bch_verbose(c, "truncating inode %llu", u.bi_inum);
1076
1077 /*
1078 * XXX: need to truncate partial blocks too here - or ideally
1079 * just switch units to bytes and that issue goes away
1080 */
1081 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1082 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
1083 iter->pos.snapshot),
1084 POS(u.bi_inum, U64_MAX),
1085 0, NULL);
1086 bch_err_msg(c, ret, "in fsck truncating inode");
1087 if (ret)
1088 return ret;
1089
1090 /*
1091 * We truncated without our normal sector accounting hook, just
1092 * make sure we recalculate it:
1093 */
1094 u.bi_flags |= BCH_INODE_i_sectors_dirty;
1095
1096 u.bi_flags &= ~BCH_INODE_i_size_dirty;
1097 do_update = true;
1098 }
1099
1100 if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
1101 (!c->sb.clean ||
1102 fsck_err(trans, inode_i_sectors_dirty_but_clean,
1103 "filesystem marked clean, but inode %llu has i_sectors dirty",
1104 u.bi_inum))) {
1105 s64 sectors;
1106
1107 bch_verbose(c, "recounting sectors for inode %llu",
1108 u.bi_inum);
1109
1110 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
1111 if (sectors < 0) {
1112 bch_err_msg(c, sectors, "in fsck recounting inode sectors");
1113 return sectors;
1114 }
1115
1116 u.bi_sectors = sectors;
1117 u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
1118 do_update = true;
1119 }
1120
1121 if (u.bi_flags & BCH_INODE_backptr_untrusted) {
1122 u.bi_dir = 0;
1123 u.bi_dir_offset = 0;
1124 u.bi_flags &= ~BCH_INODE_backptr_untrusted;
1125 do_update = true;
1126 }
1127
1128 if (u.bi_dir || u.bi_dir_offset) {
1129 ret = check_inode_dirent_inode(trans, k, &u, k.k->p.snapshot, &do_update);
1130 if (ret)
1131 goto err;
1132 }
1133
1134 if (fsck_err_on(u.bi_parent_subvol &&
1135 (u.bi_subvol == 0 ||
1136 u.bi_subvol == BCACHEFS_ROOT_SUBVOL),
1137 trans, inode_bi_parent_nonzero,
1138 "inode %llu:%u has subvol %u but nonzero parent subvol %u",
1139 u.bi_inum, k.k->p.snapshot, u.bi_subvol, u.bi_parent_subvol)) {
1140 u.bi_parent_subvol = 0;
1141 do_update = true;
1142 }
1143
1144 if (u.bi_subvol) {
1145 struct bch_subvolume s;
1146
1147 ret = bch2_subvolume_get(trans, u.bi_subvol, false, 0, &s);
1148 if (ret && !bch2_err_matches(ret, ENOENT))
1149 goto err;
1150
1151 if (ret && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
1152 ret = reconstruct_subvol(trans, k.k->p.snapshot, u.bi_subvol, u.bi_inum);
1153 goto do_update;
1154 }
1155
1156 if (fsck_err_on(ret,
1157 trans, inode_bi_subvol_missing,
1158 "inode %llu:%u bi_subvol points to missing subvolume %u",
1159 u.bi_inum, k.k->p.snapshot, u.bi_subvol) ||
1160 fsck_err_on(le64_to_cpu(s.inode) != u.bi_inum ||
1161 !bch2_snapshot_is_ancestor(c, le32_to_cpu(s.snapshot),
1162 k.k->p.snapshot),
1163 trans, inode_bi_subvol_wrong,
1164 "inode %llu:%u points to subvol %u, but subvol points to %llu:%u",
1165 u.bi_inum, k.k->p.snapshot, u.bi_subvol,
1166 le64_to_cpu(s.inode),
1167 le32_to_cpu(s.snapshot))) {
1168 u.bi_subvol = 0;
1169 u.bi_parent_subvol = 0;
1170 do_update = true;
1171 }
1172 }
1173 do_update:
1174 if (do_update) {
1175 ret = __bch2_fsck_write_inode(trans, &u, iter->pos.snapshot);
1176 bch_err_msg(c, ret, "in fsck updating inode");
1177 if (ret)
1178 return ret;
1179 }
1180 err:
1181 fsck_err:
1182 bch_err_fn(c, ret);
1183 return ret;
1184 }
1185
1186 int bch2_check_inodes(struct bch_fs *c)
1187 {
1188 bool full = c->opts.fsck;
1189 struct bch_inode_unpacked prev = { 0 };
1190 struct snapshots_seen s;
1191
1192 snapshots_seen_init(&s);
1193
1194 int ret = bch2_trans_run(c,
1195 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1196 POS_MIN,
1197 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1198 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1199 check_inode(trans, &iter, k, &prev, &s, full)));
1200
1201 snapshots_seen_exit(&s);
1202 bch_err_fn(c, ret);
1203 return ret;
1204 }
1205
1206 static inline bool btree_matches_i_mode(enum btree_id btree, unsigned mode)
1207 {
1208 switch (btree) {
1209 case BTREE_ID_extents:
1210 return S_ISREG(mode) || S_ISLNK(mode);
1211 case BTREE_ID_dirents:
1212 return S_ISDIR(mode);
1213 case BTREE_ID_xattrs:
1214 return true;
1215 default:
1216 BUG();
1217 }
1218 }
1219
1220 static int check_key_has_inode(struct btree_trans *trans,
1221 struct btree_iter *iter,
1222 struct inode_walker *inode,
1223 struct inode_walker_entry *i,
1224 struct bkey_s_c k)
1225 {
1226 struct bch_fs *c = trans->c;
1227 struct printbuf buf = PRINTBUF;
1228 int ret = PTR_ERR_OR_ZERO(i);
1229 if (ret)
1230 return ret;
1231
1232 if (k.k->type == KEY_TYPE_whiteout)
1233 goto out;
1234
1235 if (!i && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_inodes))) {
1236 ret = reconstruct_inode(trans, iter->btree_id, k.k->p.snapshot, k.k->p.inode) ?:
1237 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
1238 if (ret)
1239 goto err;
1240
1241 inode->last_pos.inode--;
1242 ret = -BCH_ERR_transaction_restart_nested;
1243 goto err;
1244 }
1245
1246 if (fsck_err_on(!i,
1247 trans, key_in_missing_inode,
1248 "key in missing inode:\n %s",
1249 (printbuf_reset(&buf),
1250 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1251 goto delete;
1252
1253 if (fsck_err_on(i && !btree_matches_i_mode(iter->btree_id, i->inode.bi_mode),
1254 trans, key_in_wrong_inode_type,
1255 "key for wrong inode mode %o:\n %s",
1256 i->inode.bi_mode,
1257 (printbuf_reset(&buf),
1258 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1259 goto delete;
1260 out:
1261 err:
1262 fsck_err:
1263 printbuf_exit(&buf);
1264 bch_err_fn(c, ret);
1265 return ret;
1266 delete:
1267 ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_internal_snapshot_node);
1268 goto out;
1269 }
1270
1271 static int check_i_sectors_notnested(struct btree_trans *trans, struct inode_walker *w)
1272 {
1273 struct bch_fs *c = trans->c;
1274 int ret = 0;
1275 s64 count2;
1276
1277 darray_for_each(w->inodes, i) {
1278 if (i->inode.bi_sectors == i->count)
1279 continue;
1280
1281 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1282
1283 if (w->recalculate_sums)
1284 i->count = count2;
1285
1286 if (i->count != count2) {
1287 bch_err_ratelimited(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1288 w->last_pos.inode, i->snapshot, i->count, count2);
1289 return -BCH_ERR_internal_fsck_err;
1290 }
1291
1292 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1293 trans, inode_i_sectors_wrong,
1294 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1295 w->last_pos.inode, i->snapshot,
1296 i->inode.bi_sectors, i->count)) {
1297 i->inode.bi_sectors = i->count;
1298 ret = bch2_fsck_write_inode(trans, &i->inode, i->snapshot);
1299 if (ret)
1300 break;
1301 }
1302 }
1303 fsck_err:
1304 bch_err_fn(c, ret);
1305 return ret;
1306 }
1307
1308 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1309 {
1310 u32 restart_count = trans->restart_count;
1311 return check_i_sectors_notnested(trans, w) ?:
1312 trans_was_restarted(trans, restart_count);
1313 }
1314
1315 struct extent_end {
1316 u32 snapshot;
1317 u64 offset;
1318 struct snapshots_seen seen;
1319 };
1320
1321 struct extent_ends {
1322 struct bpos last_pos;
1323 DARRAY(struct extent_end) e;
1324 };
1325
1326 static void extent_ends_reset(struct extent_ends *extent_ends)
1327 {
1328 darray_for_each(extent_ends->e, i)
1329 snapshots_seen_exit(&i->seen);
1330 extent_ends->e.nr = 0;
1331 }
1332
1333 static void extent_ends_exit(struct extent_ends *extent_ends)
1334 {
1335 extent_ends_reset(extent_ends);
1336 darray_exit(&extent_ends->e);
1337 }
1338
1339 static void extent_ends_init(struct extent_ends *extent_ends)
1340 {
1341 memset(extent_ends, 0, sizeof(*extent_ends));
1342 }
1343
1344 static int extent_ends_at(struct bch_fs *c,
1345 struct extent_ends *extent_ends,
1346 struct snapshots_seen *seen,
1347 struct bkey_s_c k)
1348 {
1349 struct extent_end *i, n = (struct extent_end) {
1350 .offset = k.k->p.offset,
1351 .snapshot = k.k->p.snapshot,
1352 .seen = *seen,
1353 };
1354
1355 n.seen.ids.data = kmemdup(seen->ids.data,
1356 sizeof(seen->ids.data[0]) * seen->ids.size,
1357 GFP_KERNEL);
1358 if (!n.seen.ids.data)
1359 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1360
1361 __darray_for_each(extent_ends->e, i) {
1362 if (i->snapshot == k.k->p.snapshot) {
1363 snapshots_seen_exit(&i->seen);
1364 *i = n;
1365 return 0;
1366 }
1367
1368 if (i->snapshot >= k.k->p.snapshot)
1369 break;
1370 }
1371
1372 return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1373 }
1374
1375 static int overlapping_extents_found(struct btree_trans *trans,
1376 enum btree_id btree,
1377 struct bpos pos1, struct snapshots_seen *pos1_seen,
1378 struct bkey pos2,
1379 bool *fixed,
1380 struct extent_end *extent_end)
1381 {
1382 struct bch_fs *c = trans->c;
1383 struct printbuf buf = PRINTBUF;
1384 struct btree_iter iter1, iter2 = { NULL };
1385 struct bkey_s_c k1, k2;
1386 int ret;
1387
1388 BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1389
1390 bch2_trans_iter_init(trans, &iter1, btree, pos1,
1391 BTREE_ITER_all_snapshots|
1392 BTREE_ITER_not_extents);
1393 k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1394 ret = bkey_err(k1);
1395 if (ret)
1396 goto err;
1397
1398 prt_str(&buf, "\n ");
1399 bch2_bkey_val_to_text(&buf, c, k1);
1400
1401 if (!bpos_eq(pos1, k1.k->p)) {
1402 prt_str(&buf, "\n wanted\n ");
1403 bch2_bpos_to_text(&buf, pos1);
1404 prt_str(&buf, "\n ");
1405 bch2_bkey_to_text(&buf, &pos2);
1406
1407 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1408 __func__, buf.buf);
1409 ret = -BCH_ERR_internal_fsck_err;
1410 goto err;
1411 }
1412
1413 bch2_trans_copy_iter(&iter2, &iter1);
1414
1415 while (1) {
1416 bch2_btree_iter_advance(&iter2);
1417
1418 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1419 ret = bkey_err(k2);
1420 if (ret)
1421 goto err;
1422
1423 if (bpos_ge(k2.k->p, pos2.p))
1424 break;
1425 }
1426
1427 prt_str(&buf, "\n ");
1428 bch2_bkey_val_to_text(&buf, c, k2);
1429
1430 if (bpos_gt(k2.k->p, pos2.p) ||
1431 pos2.size != k2.k->size) {
1432 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1433 __func__, buf.buf);
1434 ret = -BCH_ERR_internal_fsck_err;
1435 goto err;
1436 }
1437
1438 prt_printf(&buf, "\n overwriting %s extent",
1439 pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1440
1441 if (fsck_err(trans, extent_overlapping,
1442 "overlapping extents%s", buf.buf)) {
1443 struct btree_iter *old_iter = &iter1;
1444 struct disk_reservation res = { 0 };
1445
1446 if (pos1.snapshot < pos2.p.snapshot) {
1447 old_iter = &iter2;
1448 swap(k1, k2);
1449 }
1450
1451 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1452
1453 ret = bch2_trans_update_extent_overwrite(trans, old_iter,
1454 BTREE_UPDATE_internal_snapshot_node,
1455 k1, k2) ?:
1456 bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1457 bch2_disk_reservation_put(c, &res);
1458
1459 if (ret)
1460 goto err;
1461
1462 *fixed = true;
1463
1464 if (pos1.snapshot == pos2.p.snapshot) {
1465 /*
1466 * We overwrote the first extent, and did the overwrite
1467 * in the same snapshot:
1468 */
1469 extent_end->offset = bkey_start_offset(&pos2);
1470 } else if (pos1.snapshot > pos2.p.snapshot) {
1471 /*
1472 * We overwrote the first extent in pos2's snapshot:
1473 */
1474 ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1475 } else {
1476 /*
1477 * We overwrote the second extent - restart
1478 * check_extent() from the top:
1479 */
1480 ret = -BCH_ERR_transaction_restart_nested;
1481 }
1482 }
1483 fsck_err:
1484 err:
1485 bch2_trans_iter_exit(trans, &iter2);
1486 bch2_trans_iter_exit(trans, &iter1);
1487 printbuf_exit(&buf);
1488 return ret;
1489 }
1490
1491 static int check_overlapping_extents(struct btree_trans *trans,
1492 struct snapshots_seen *seen,
1493 struct extent_ends *extent_ends,
1494 struct bkey_s_c k,
1495 struct btree_iter *iter,
1496 bool *fixed)
1497 {
1498 struct bch_fs *c = trans->c;
1499 int ret = 0;
1500
1501 /* transaction restart, running again */
1502 if (bpos_eq(extent_ends->last_pos, k.k->p))
1503 return 0;
1504
1505 if (extent_ends->last_pos.inode != k.k->p.inode)
1506 extent_ends_reset(extent_ends);
1507
1508 darray_for_each(extent_ends->e, i) {
1509 if (i->offset <= bkey_start_offset(k.k))
1510 continue;
1511
1512 if (!ref_visible2(c,
1513 k.k->p.snapshot, seen,
1514 i->snapshot, &i->seen))
1515 continue;
1516
1517 ret = overlapping_extents_found(trans, iter->btree_id,
1518 SPOS(iter->pos.inode,
1519 i->offset,
1520 i->snapshot),
1521 &i->seen,
1522 *k.k, fixed, i);
1523 if (ret)
1524 goto err;
1525 }
1526
1527 extent_ends->last_pos = k.k->p;
1528 err:
1529 return ret;
1530 }
1531
1532 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1533 struct bkey_s_c k)
1534 {
1535 struct bch_fs *c = trans->c;
1536 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1537 struct bch_extent_crc_unpacked crc;
1538 const union bch_extent_entry *i;
1539 unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1540
1541 bkey_for_each_crc(k.k, ptrs, crc, i)
1542 if (crc_is_encoded(crc) &&
1543 crc.uncompressed_size > encoded_extent_max_sectors) {
1544 struct printbuf buf = PRINTBUF;
1545
1546 bch2_bkey_val_to_text(&buf, c, k);
1547 bch_err(c, "overbig encoded extent, please report this:\n %s", buf.buf);
1548 printbuf_exit(&buf);
1549 }
1550
1551 return 0;
1552 }
1553
1554 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1555 struct bkey_s_c k,
1556 struct inode_walker *inode,
1557 struct snapshots_seen *s,
1558 struct extent_ends *extent_ends)
1559 {
1560 struct bch_fs *c = trans->c;
1561 struct inode_walker_entry *i;
1562 struct printbuf buf = PRINTBUF;
1563 int ret = 0;
1564
1565 ret = bch2_check_key_has_snapshot(trans, iter, k);
1566 if (ret) {
1567 ret = ret < 0 ? ret : 0;
1568 goto out;
1569 }
1570
1571 if (inode->last_pos.inode != k.k->p.inode) {
1572 ret = check_i_sectors(trans, inode);
1573 if (ret)
1574 goto err;
1575 }
1576
1577 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1578 if (ret)
1579 goto err;
1580
1581 i = walk_inode(trans, inode, k);
1582 ret = PTR_ERR_OR_ZERO(i);
1583 if (ret)
1584 goto err;
1585
1586 ret = check_key_has_inode(trans, iter, inode, i, k);
1587 if (ret)
1588 goto err;
1589
1590 if (k.k->type != KEY_TYPE_whiteout) {
1591 ret = check_overlapping_extents(trans, s, extent_ends, k, iter,
1592 &inode->recalculate_sums);
1593 if (ret)
1594 goto err;
1595 }
1596
1597 /*
1598 * Check inodes in reverse order, from oldest snapshots to newest,
1599 * starting from the inode that matches this extent's snapshot. If we
1600 * didn't have one, iterate over all inodes:
1601 */
1602 if (!i)
1603 i = &darray_last(inode->inodes);
1604
1605 for (;
1606 inode->inodes.data && i >= inode->inodes.data;
1607 --i) {
1608 if (i->snapshot > k.k->p.snapshot ||
1609 !key_visible_in_snapshot(c, s, i->snapshot, k.k->p.snapshot))
1610 continue;
1611
1612 if (k.k->type != KEY_TYPE_whiteout) {
1613 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1614 k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1615 !bkey_extent_is_reservation(k),
1616 trans, extent_past_end_of_inode,
1617 "extent type past end of inode %llu:%u, i_size %llu\n %s",
1618 i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1619 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1620 struct btree_iter iter2;
1621
1622 bch2_trans_copy_iter(&iter2, iter);
1623 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1624 ret = bch2_btree_iter_traverse(&iter2) ?:
1625 bch2_btree_delete_at(trans, &iter2,
1626 BTREE_UPDATE_internal_snapshot_node);
1627 bch2_trans_iter_exit(trans, &iter2);
1628 if (ret)
1629 goto err;
1630
1631 iter->k.type = KEY_TYPE_whiteout;
1632 }
1633
1634 if (bkey_extent_is_allocation(k.k))
1635 i->count += k.k->size;
1636 }
1637
1638 i->seen_this_pos = true;
1639 }
1640
1641 if (k.k->type != KEY_TYPE_whiteout) {
1642 ret = extent_ends_at(c, extent_ends, s, k);
1643 if (ret)
1644 goto err;
1645 }
1646 out:
1647 err:
1648 fsck_err:
1649 printbuf_exit(&buf);
1650 bch_err_fn(c, ret);
1651 return ret;
1652 }
1653
1654 /*
1655 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1656 * that i_size an i_sectors are consistent
1657 */
1658 int bch2_check_extents(struct bch_fs *c)
1659 {
1660 struct inode_walker w = inode_walker_init();
1661 struct snapshots_seen s;
1662 struct extent_ends extent_ends;
1663 struct disk_reservation res = { 0 };
1664
1665 snapshots_seen_init(&s);
1666 extent_ends_init(&extent_ends);
1667
1668 int ret = bch2_trans_run(c,
1669 for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1670 POS(BCACHEFS_ROOT_INO, 0),
1671 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1672 &res, NULL,
1673 BCH_TRANS_COMMIT_no_enospc, ({
1674 bch2_disk_reservation_put(c, &res);
1675 check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1676 check_extent_overbig(trans, &iter, k);
1677 })) ?:
1678 check_i_sectors_notnested(trans, &w));
1679
1680 bch2_disk_reservation_put(c, &res);
1681 extent_ends_exit(&extent_ends);
1682 inode_walker_exit(&w);
1683 snapshots_seen_exit(&s);
1684
1685 bch_err_fn(c, ret);
1686 return ret;
1687 }
1688
1689 int bch2_check_indirect_extents(struct bch_fs *c)
1690 {
1691 struct disk_reservation res = { 0 };
1692
1693 int ret = bch2_trans_run(c,
1694 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1695 POS_MIN,
1696 BTREE_ITER_prefetch, k,
1697 &res, NULL,
1698 BCH_TRANS_COMMIT_no_enospc, ({
1699 bch2_disk_reservation_put(c, &res);
1700 check_extent_overbig(trans, &iter, k);
1701 })));
1702
1703 bch2_disk_reservation_put(c, &res);
1704 bch_err_fn(c, ret);
1705 return ret;
1706 }
1707
1708 static int check_subdir_count_notnested(struct btree_trans *trans, struct inode_walker *w)
1709 {
1710 struct bch_fs *c = trans->c;
1711 int ret = 0;
1712 s64 count2;
1713
1714 darray_for_each(w->inodes, i) {
1715 if (i->inode.bi_nlink == i->count)
1716 continue;
1717
1718 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1719 if (count2 < 0)
1720 return count2;
1721
1722 if (i->count != count2) {
1723 bch_err_ratelimited(c, "fsck counted subdirectories wrong for inum %llu:%u: got %llu should be %llu",
1724 w->last_pos.inode, i->snapshot, i->count, count2);
1725 i->count = count2;
1726 if (i->inode.bi_nlink == i->count)
1727 continue;
1728 }
1729
1730 if (fsck_err_on(i->inode.bi_nlink != i->count,
1731 trans, inode_dir_wrong_nlink,
1732 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1733 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1734 i->inode.bi_nlink = i->count;
1735 ret = bch2_fsck_write_inode(trans, &i->inode, i->snapshot);
1736 if (ret)
1737 break;
1738 }
1739 }
1740 fsck_err:
1741 bch_err_fn(c, ret);
1742 return ret;
1743 }
1744
1745 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1746 {
1747 u32 restart_count = trans->restart_count;
1748 return check_subdir_count_notnested(trans, w) ?:
1749 trans_was_restarted(trans, restart_count);
1750 }
1751
1752 noinline_for_stack
1753 static int check_dirent_inode_dirent(struct btree_trans *trans,
1754 struct btree_iter *iter,
1755 struct bkey_s_c_dirent d,
1756 struct bch_inode_unpacked *target,
1757 u32 target_snapshot)
1758 {
1759 struct bch_fs *c = trans->c;
1760 struct printbuf buf = PRINTBUF;
1761 int ret = 0;
1762
1763 if (inode_points_to_dirent(target, d))
1764 return 0;
1765
1766 if (bch2_inode_should_have_bp(target) &&
1767 !fsck_err(trans, inode_wrong_backpointer,
1768 "dirent points to inode that does not point back:\n %s",
1769 (bch2_bkey_val_to_text(&buf, c, d.s_c),
1770 prt_printf(&buf, "\n "),
1771 bch2_inode_unpacked_to_text(&buf, target),
1772 buf.buf)))
1773 goto out_noiter;
1774
1775 if (!target->bi_dir &&
1776 !target->bi_dir_offset) {
1777 target->bi_dir = d.k->p.inode;
1778 target->bi_dir_offset = d.k->p.offset;
1779 return __bch2_fsck_write_inode(trans, target, target_snapshot);
1780 }
1781
1782 struct btree_iter bp_iter = { NULL };
1783 struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1784 SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1785 ret = bkey_err(bp_dirent);
1786 if (ret && !bch2_err_matches(ret, ENOENT))
1787 goto err;
1788
1789 bool backpointer_exists = !ret;
1790 ret = 0;
1791
1792 if (fsck_err_on(!backpointer_exists,
1793 trans, inode_wrong_backpointer,
1794 "inode %llu:%u has wrong backpointer:\n"
1795 "got %llu:%llu\n"
1796 "should be %llu:%llu",
1797 target->bi_inum, target_snapshot,
1798 target->bi_dir,
1799 target->bi_dir_offset,
1800 d.k->p.inode,
1801 d.k->p.offset)) {
1802 target->bi_dir = d.k->p.inode;
1803 target->bi_dir_offset = d.k->p.offset;
1804 ret = __bch2_fsck_write_inode(trans, target, target_snapshot);
1805 goto out;
1806 }
1807
1808 bch2_bkey_val_to_text(&buf, c, d.s_c);
1809 prt_newline(&buf);
1810 if (backpointer_exists)
1811 bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1812
1813 if (fsck_err_on(backpointer_exists &&
1814 (S_ISDIR(target->bi_mode) ||
1815 target->bi_subvol),
1816 trans, inode_dir_multiple_links,
1817 "%s %llu:%u with multiple links\n%s",
1818 S_ISDIR(target->bi_mode) ? "directory" : "subvolume",
1819 target->bi_inum, target_snapshot, buf.buf)) {
1820 ret = __remove_dirent(trans, d.k->p);
1821 goto out;
1822 }
1823
1824 /*
1825 * hardlinked file with nlink 0:
1826 * We're just adjusting nlink here so check_nlinks() will pick
1827 * it up, it ignores inodes with nlink 0
1828 */
1829 if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1830 trans, inode_multiple_links_but_nlink_0,
1831 "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1832 target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1833 target->bi_nlink++;
1834 target->bi_flags &= ~BCH_INODE_unlinked;
1835 ret = __bch2_fsck_write_inode(trans, target, target_snapshot);
1836 if (ret)
1837 goto err;
1838 }
1839 out:
1840 err:
1841 fsck_err:
1842 bch2_trans_iter_exit(trans, &bp_iter);
1843 out_noiter:
1844 printbuf_exit(&buf);
1845 bch_err_fn(c, ret);
1846 return ret;
1847 }
1848
1849 noinline_for_stack
1850 static int check_dirent_target(struct btree_trans *trans,
1851 struct btree_iter *iter,
1852 struct bkey_s_c_dirent d,
1853 struct bch_inode_unpacked *target,
1854 u32 target_snapshot)
1855 {
1856 struct bch_fs *c = trans->c;
1857 struct bkey_i_dirent *n;
1858 struct printbuf buf = PRINTBUF;
1859 int ret = 0;
1860
1861 ret = check_dirent_inode_dirent(trans, iter, d, target, target_snapshot);
1862 if (ret)
1863 goto err;
1864
1865 if (fsck_err_on(d.v->d_type != inode_d_type(target),
1866 trans, dirent_d_type_wrong,
1867 "incorrect d_type: got %s, should be %s:\n%s",
1868 bch2_d_type_str(d.v->d_type),
1869 bch2_d_type_str(inode_d_type(target)),
1870 (printbuf_reset(&buf),
1871 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1872 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1873 ret = PTR_ERR_OR_ZERO(n);
1874 if (ret)
1875 goto err;
1876
1877 bkey_reassemble(&n->k_i, d.s_c);
1878 n->v.d_type = inode_d_type(target);
1879 if (n->v.d_type == DT_SUBVOL) {
1880 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1881 n->v.d_child_subvol = cpu_to_le32(target->bi_subvol);
1882 } else {
1883 n->v.d_inum = cpu_to_le64(target->bi_inum);
1884 }
1885
1886 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1887 if (ret)
1888 goto err;
1889
1890 d = dirent_i_to_s_c(n);
1891 }
1892 err:
1893 fsck_err:
1894 printbuf_exit(&buf);
1895 bch_err_fn(c, ret);
1896 return ret;
1897 }
1898
1899 /* find a subvolume that's a descendent of @snapshot: */
1900 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
1901 {
1902 struct btree_iter iter;
1903 struct bkey_s_c k;
1904 int ret;
1905
1906 for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN, 0, k, ret) {
1907 if (k.k->type != KEY_TYPE_subvolume)
1908 continue;
1909
1910 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
1911 if (bch2_snapshot_is_ancestor(trans->c, le32_to_cpu(s.v->snapshot), snapshot)) {
1912 bch2_trans_iter_exit(trans, &iter);
1913 *subvolid = k.k->p.offset;
1914 goto found;
1915 }
1916 }
1917 if (!ret)
1918 ret = -ENOENT;
1919 found:
1920 bch2_trans_iter_exit(trans, &iter);
1921 return ret;
1922 }
1923
1924 noinline_for_stack
1925 static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *iter,
1926 struct bkey_s_c_dirent d)
1927 {
1928 struct bch_fs *c = trans->c;
1929 struct btree_iter subvol_iter = {};
1930 struct bch_inode_unpacked subvol_root;
1931 u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
1932 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1933 u32 parent_snapshot;
1934 u32 new_parent_subvol = 0;
1935 u64 parent_inum;
1936 struct printbuf buf = PRINTBUF;
1937 int ret = 0;
1938
1939 ret = subvol_lookup(trans, parent_subvol, &parent_snapshot, &parent_inum);
1940 if (ret && !bch2_err_matches(ret, ENOENT))
1941 return ret;
1942
1943 if (ret ||
1944 (!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot))) {
1945 int ret2 = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
1946 if (ret2 && !bch2_err_matches(ret, ENOENT))
1947 return ret2;
1948 }
1949
1950 if (ret &&
1951 !new_parent_subvol &&
1952 (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
1953 /*
1954 * Couldn't find a subvol for dirent's snapshot - but we lost
1955 * subvols, so we need to reconstruct:
1956 */
1957 ret = reconstruct_subvol(trans, d.k->p.snapshot, parent_subvol, 0);
1958 if (ret)
1959 return ret;
1960
1961 parent_snapshot = d.k->p.snapshot;
1962 }
1963
1964 if (fsck_err_on(ret,
1965 trans, dirent_to_missing_parent_subvol,
1966 "dirent parent_subvol points to missing subvolume\n%s",
1967 (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
1968 fsck_err_on(!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot),
1969 trans, dirent_not_visible_in_parent_subvol,
1970 "dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
1971 parent_snapshot,
1972 (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1973 if (!new_parent_subvol) {
1974 bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
1975 return -BCH_ERR_fsck_repair_unimplemented;
1976 }
1977
1978 struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
1979 ret = PTR_ERR_OR_ZERO(new_dirent);
1980 if (ret)
1981 goto err;
1982
1983 new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
1984 }
1985
1986 struct bkey_s_c_subvolume s =
1987 bch2_bkey_get_iter_typed(trans, &subvol_iter,
1988 BTREE_ID_subvolumes, POS(0, target_subvol),
1989 0, subvolume);
1990 ret = bkey_err(s.s_c);
1991 if (ret && !bch2_err_matches(ret, ENOENT))
1992 return ret;
1993
1994 if (ret) {
1995 if (fsck_err(trans, dirent_to_missing_subvol,
1996 "dirent points to missing subvolume\n%s",
1997 (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
1998 return __remove_dirent(trans, d.k->p);
1999 ret = 0;
2000 goto out;
2001 }
2002
2003 if (fsck_err_on(le32_to_cpu(s.v->fs_path_parent) != parent_subvol,
2004 trans, subvol_fs_path_parent_wrong,
2005 "subvol with wrong fs_path_parent, should be be %u\n%s",
2006 parent_subvol,
2007 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
2008 struct bkey_i_subvolume *n =
2009 bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
2010 ret = PTR_ERR_OR_ZERO(n);
2011 if (ret)
2012 goto err;
2013
2014 n->v.fs_path_parent = cpu_to_le32(parent_subvol);
2015 }
2016
2017 u64 target_inum = le64_to_cpu(s.v->inode);
2018 u32 target_snapshot = le32_to_cpu(s.v->snapshot);
2019
2020 ret = lookup_inode(trans, target_inum, &subvol_root, &target_snapshot);
2021 if (ret && !bch2_err_matches(ret, ENOENT))
2022 goto err;
2023
2024 if (ret) {
2025 bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
2026 ret = -BCH_ERR_fsck_repair_unimplemented;
2027 goto err;
2028 }
2029
2030 if (fsck_err_on(!ret && parent_subvol != subvol_root.bi_parent_subvol,
2031 trans, inode_bi_parent_wrong,
2032 "subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
2033 target_inum,
2034 subvol_root.bi_parent_subvol, parent_subvol)) {
2035 subvol_root.bi_parent_subvol = parent_subvol;
2036 ret = __bch2_fsck_write_inode(trans, &subvol_root, target_snapshot);
2037 if (ret)
2038 goto err;
2039 }
2040
2041 ret = check_dirent_target(trans, iter, d, &subvol_root,
2042 target_snapshot);
2043 if (ret)
2044 goto err;
2045 out:
2046 err:
2047 fsck_err:
2048 bch2_trans_iter_exit(trans, &subvol_iter);
2049 printbuf_exit(&buf);
2050 return ret;
2051 }
2052
2053 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
2054 struct bkey_s_c k,
2055 struct bch_hash_info *hash_info,
2056 struct inode_walker *dir,
2057 struct inode_walker *target,
2058 struct snapshots_seen *s)
2059 {
2060 struct bch_fs *c = trans->c;
2061 struct inode_walker_entry *i;
2062 struct printbuf buf = PRINTBUF;
2063 int ret = 0;
2064
2065 ret = bch2_check_key_has_snapshot(trans, iter, k);
2066 if (ret) {
2067 ret = ret < 0 ? ret : 0;
2068 goto out;
2069 }
2070
2071 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
2072 if (ret)
2073 goto err;
2074
2075 if (k.k->type == KEY_TYPE_whiteout)
2076 goto out;
2077
2078 if (dir->last_pos.inode != k.k->p.inode) {
2079 ret = check_subdir_count(trans, dir);
2080 if (ret)
2081 goto err;
2082 }
2083
2084 i = walk_inode(trans, dir, k);
2085 ret = PTR_ERR_OR_ZERO(i);
2086 if (ret < 0)
2087 goto err;
2088
2089 ret = check_key_has_inode(trans, iter, dir, i, k);
2090 if (ret)
2091 goto err;
2092
2093 if (!i)
2094 goto out;
2095
2096 if (dir->first_this_inode)
2097 *hash_info = bch2_hash_info_init(c, &i->inode);
2098 dir->first_this_inode = false;
2099
2100 ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
2101 if (ret < 0)
2102 goto err;
2103 if (ret) {
2104 /* dirent has been deleted */
2105 ret = 0;
2106 goto out;
2107 }
2108
2109 if (k.k->type != KEY_TYPE_dirent)
2110 goto out;
2111
2112 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2113
2114 if (d.v->d_type == DT_SUBVOL) {
2115 ret = check_dirent_to_subvol(trans, iter, d);
2116 if (ret)
2117 goto err;
2118 } else {
2119 ret = get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
2120 if (ret)
2121 goto err;
2122
2123 if (fsck_err_on(!target->inodes.nr,
2124 trans, dirent_to_missing_inode,
2125 "dirent points to missing inode:\n%s",
2126 (printbuf_reset(&buf),
2127 bch2_bkey_val_to_text(&buf, c, k),
2128 buf.buf))) {
2129 ret = __remove_dirent(trans, d.k->p);
2130 if (ret)
2131 goto err;
2132 }
2133
2134 darray_for_each(target->inodes, i) {
2135 ret = check_dirent_target(trans, iter, d,
2136 &i->inode, i->snapshot);
2137 if (ret)
2138 goto err;
2139 }
2140
2141 if (d.v->d_type == DT_DIR)
2142 for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
2143 i->count++;
2144 }
2145 out:
2146 err:
2147 fsck_err:
2148 printbuf_exit(&buf);
2149 bch_err_fn(c, ret);
2150 return ret;
2151 }
2152
2153 /*
2154 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
2155 * validate d_type
2156 */
2157 int bch2_check_dirents(struct bch_fs *c)
2158 {
2159 struct inode_walker dir = inode_walker_init();
2160 struct inode_walker target = inode_walker_init();
2161 struct snapshots_seen s;
2162 struct bch_hash_info hash_info;
2163
2164 snapshots_seen_init(&s);
2165
2166 int ret = bch2_trans_run(c,
2167 for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
2168 POS(BCACHEFS_ROOT_INO, 0),
2169 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots,
2170 k,
2171 NULL, NULL,
2172 BCH_TRANS_COMMIT_no_enospc,
2173 check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)) ?:
2174 check_subdir_count_notnested(trans, &dir));
2175
2176 snapshots_seen_exit(&s);
2177 inode_walker_exit(&dir);
2178 inode_walker_exit(&target);
2179 bch_err_fn(c, ret);
2180 return ret;
2181 }
2182
2183 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
2184 struct bkey_s_c k,
2185 struct bch_hash_info *hash_info,
2186 struct inode_walker *inode)
2187 {
2188 struct bch_fs *c = trans->c;
2189 struct inode_walker_entry *i;
2190 int ret;
2191
2192 ret = bch2_check_key_has_snapshot(trans, iter, k);
2193 if (ret < 0)
2194 return ret;
2195 if (ret)
2196 return 0;
2197
2198 i = walk_inode(trans, inode, k);
2199 ret = PTR_ERR_OR_ZERO(i);
2200 if (ret)
2201 return ret;
2202
2203 ret = check_key_has_inode(trans, iter, inode, i, k);
2204 if (ret)
2205 return ret;
2206
2207 if (!i)
2208 return 0;
2209
2210 if (inode->first_this_inode)
2211 *hash_info = bch2_hash_info_init(c, &i->inode);
2212 inode->first_this_inode = false;
2213
2214 ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
2215 bch_err_fn(c, ret);
2216 return ret;
2217 }
2218
2219 /*
2220 * Walk xattrs: verify that they all have a corresponding inode
2221 */
2222 int bch2_check_xattrs(struct bch_fs *c)
2223 {
2224 struct inode_walker inode = inode_walker_init();
2225 struct bch_hash_info hash_info;
2226 int ret = 0;
2227
2228 ret = bch2_trans_run(c,
2229 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
2230 POS(BCACHEFS_ROOT_INO, 0),
2231 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots,
2232 k,
2233 NULL, NULL,
2234 BCH_TRANS_COMMIT_no_enospc,
2235 check_xattr(trans, &iter, k, &hash_info, &inode)));
2236
2237 inode_walker_exit(&inode);
2238 bch_err_fn(c, ret);
2239 return ret;
2240 }
2241
2242 static int check_root_trans(struct btree_trans *trans)
2243 {
2244 struct bch_fs *c = trans->c;
2245 struct bch_inode_unpacked root_inode;
2246 u32 snapshot;
2247 u64 inum;
2248 int ret;
2249
2250 ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
2251 if (ret && !bch2_err_matches(ret, ENOENT))
2252 return ret;
2253
2254 if (mustfix_fsck_err_on(ret, trans, root_subvol_missing,
2255 "root subvol missing")) {
2256 struct bkey_i_subvolume *root_subvol =
2257 bch2_trans_kmalloc(trans, sizeof(*root_subvol));
2258 ret = PTR_ERR_OR_ZERO(root_subvol);
2259 if (ret)
2260 goto err;
2261
2262 snapshot = U32_MAX;
2263 inum = BCACHEFS_ROOT_INO;
2264
2265 bkey_subvolume_init(&root_subvol->k_i);
2266 root_subvol->k.p.offset = BCACHEFS_ROOT_SUBVOL;
2267 root_subvol->v.flags = 0;
2268 root_subvol->v.snapshot = cpu_to_le32(snapshot);
2269 root_subvol->v.inode = cpu_to_le64(inum);
2270 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol->k_i, 0);
2271 bch_err_msg(c, ret, "writing root subvol");
2272 if (ret)
2273 goto err;
2274 }
2275
2276 ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
2277 if (ret && !bch2_err_matches(ret, ENOENT))
2278 return ret;
2279
2280 if (mustfix_fsck_err_on(ret,
2281 trans, root_dir_missing,
2282 "root directory missing") ||
2283 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
2284 trans, root_inode_not_dir,
2285 "root inode not a directory")) {
2286 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
2287 0, NULL);
2288 root_inode.bi_inum = inum;
2289
2290 ret = __bch2_fsck_write_inode(trans, &root_inode, snapshot);
2291 bch_err_msg(c, ret, "writing root inode");
2292 }
2293 err:
2294 fsck_err:
2295 return ret;
2296 }
2297
2298 /* Get root directory, create if it doesn't exist: */
2299 int bch2_check_root(struct bch_fs *c)
2300 {
2301 int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2302 check_root_trans(trans));
2303 bch_err_fn(c, ret);
2304 return ret;
2305 }
2306
2307 typedef DARRAY(u32) darray_u32;
2308
2309 static bool darray_u32_has(darray_u32 *d, u32 v)
2310 {
2311 darray_for_each(*d, i)
2312 if (*i == v)
2313 return true;
2314 return false;
2315 }
2316
2317 /*
2318 * We've checked that inode backpointers point to valid dirents; here, it's
2319 * sufficient to check that the subvolume root has a dirent:
2320 */
2321 static int subvol_has_dirent(struct btree_trans *trans, struct bkey_s_c_subvolume s)
2322 {
2323 struct bch_inode_unpacked inode;
2324 int ret = bch2_inode_find_by_inum_trans(trans,
2325 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
2326 &inode);
2327 if (ret)
2328 return ret;
2329
2330 return inode.bi_dir != 0;
2331 }
2332
2333 static int check_subvol_path(struct btree_trans *trans, struct btree_iter *iter, struct bkey_s_c k)
2334 {
2335 struct bch_fs *c = trans->c;
2336 struct btree_iter parent_iter = {};
2337 darray_u32 subvol_path = {};
2338 struct printbuf buf = PRINTBUF;
2339 int ret = 0;
2340
2341 if (k.k->type != KEY_TYPE_subvolume)
2342 return 0;
2343
2344 while (k.k->p.offset != BCACHEFS_ROOT_SUBVOL) {
2345 ret = darray_push(&subvol_path, k.k->p.offset);
2346 if (ret)
2347 goto err;
2348
2349 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2350
2351 ret = subvol_has_dirent(trans, s);
2352 if (ret < 0)
2353 break;
2354
2355 if (fsck_err_on(!ret,
2356 trans, subvol_unreachable,
2357 "unreachable subvolume %s",
2358 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2359 buf.buf))) {
2360 ret = reattach_subvol(trans, s);
2361 break;
2362 }
2363
2364 u32 parent = le32_to_cpu(s.v->fs_path_parent);
2365
2366 if (darray_u32_has(&subvol_path, parent)) {
2367 if (fsck_err(c, subvol_loop, "subvolume loop"))
2368 ret = reattach_subvol(trans, s);
2369 break;
2370 }
2371
2372 bch2_trans_iter_exit(trans, &parent_iter);
2373 bch2_trans_iter_init(trans, &parent_iter,
2374 BTREE_ID_subvolumes, POS(0, parent), 0);
2375 k = bch2_btree_iter_peek_slot(&parent_iter);
2376 ret = bkey_err(k);
2377 if (ret)
2378 goto err;
2379
2380 if (fsck_err_on(k.k->type != KEY_TYPE_subvolume,
2381 trans, subvol_unreachable,
2382 "unreachable subvolume %s",
2383 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2384 buf.buf))) {
2385 ret = reattach_subvol(trans, s);
2386 break;
2387 }
2388 }
2389 fsck_err:
2390 err:
2391 printbuf_exit(&buf);
2392 darray_exit(&subvol_path);
2393 bch2_trans_iter_exit(trans, &parent_iter);
2394 return ret;
2395 }
2396
2397 int bch2_check_subvolume_structure(struct bch_fs *c)
2398 {
2399 int ret = bch2_trans_run(c,
2400 for_each_btree_key_commit(trans, iter,
2401 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
2402 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2403 check_subvol_path(trans, &iter, k)));
2404 bch_err_fn(c, ret);
2405 return ret;
2406 }
2407
2408 struct pathbuf_entry {
2409 u64 inum;
2410 u32 snapshot;
2411 };
2412
2413 typedef DARRAY(struct pathbuf_entry) pathbuf;
2414
2415 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
2416 {
2417 darray_for_each(*p, i)
2418 if (i->inum == inum &&
2419 i->snapshot == snapshot)
2420 return true;
2421 return false;
2422 }
2423
2424 /*
2425 * Check that a given inode is reachable from its subvolume root - we already
2426 * verified subvolume connectivity:
2427 *
2428 * XXX: we should also be verifying that inodes are in the right subvolumes
2429 */
2430 static int check_path(struct btree_trans *trans, pathbuf *p, struct bkey_s_c inode_k)
2431 {
2432 struct bch_fs *c = trans->c;
2433 struct btree_iter inode_iter = {};
2434 struct bch_inode_unpacked inode;
2435 struct printbuf buf = PRINTBUF;
2436 u32 snapshot = inode_k.k->p.snapshot;
2437 int ret = 0;
2438
2439 p->nr = 0;
2440
2441 BUG_ON(bch2_inode_unpack(inode_k, &inode));
2442
2443 while (!inode.bi_subvol) {
2444 struct btree_iter dirent_iter;
2445 struct bkey_s_c_dirent d;
2446 u32 parent_snapshot = snapshot;
2447
2448 d = inode_get_dirent(trans, &dirent_iter, &inode, &parent_snapshot);
2449 ret = bkey_err(d.s_c);
2450 if (ret && !bch2_err_matches(ret, ENOENT))
2451 break;
2452
2453 if (!ret && !dirent_points_to_inode(d, &inode)) {
2454 bch2_trans_iter_exit(trans, &dirent_iter);
2455 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2456 }
2457
2458 if (bch2_err_matches(ret, ENOENT)) {
2459 ret = 0;
2460 if (fsck_err(trans, inode_unreachable,
2461 "unreachable inode\n%s",
2462 (printbuf_reset(&buf),
2463 bch2_bkey_val_to_text(&buf, c, inode_k),
2464 buf.buf)))
2465 ret = reattach_inode(trans, &inode, snapshot);
2466 goto out;
2467 }
2468
2469 bch2_trans_iter_exit(trans, &dirent_iter);
2470
2471 if (!S_ISDIR(inode.bi_mode))
2472 break;
2473
2474 ret = darray_push(p, ((struct pathbuf_entry) {
2475 .inum = inode.bi_inum,
2476 .snapshot = snapshot,
2477 }));
2478 if (ret)
2479 return ret;
2480
2481 snapshot = parent_snapshot;
2482
2483 bch2_trans_iter_exit(trans, &inode_iter);
2484 inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
2485 SPOS(0, inode.bi_dir, snapshot), 0);
2486 ret = bkey_err(inode_k) ?:
2487 !bkey_is_inode(inode_k.k) ? -BCH_ERR_ENOENT_inode
2488 : bch2_inode_unpack(inode_k, &inode);
2489 if (ret) {
2490 /* Should have been caught in dirents pass */
2491 bch_err_msg(c, ret, "error looking up parent directory");
2492 break;
2493 }
2494
2495 snapshot = inode_k.k->p.snapshot;
2496
2497 if (path_is_dup(p, inode.bi_inum, snapshot)) {
2498 /* XXX print path */
2499 bch_err(c, "directory structure loop");
2500
2501 darray_for_each(*p, i)
2502 pr_err("%llu:%u", i->inum, i->snapshot);
2503 pr_err("%llu:%u", inode.bi_inum, snapshot);
2504
2505 if (fsck_err(trans, dir_loop, "directory structure loop")) {
2506 ret = remove_backpointer(trans, &inode);
2507 bch_err_msg(c, ret, "removing dirent");
2508 if (ret)
2509 break;
2510
2511 ret = reattach_inode(trans, &inode, snapshot);
2512 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
2513 }
2514 break;
2515 }
2516 }
2517 out:
2518 fsck_err:
2519 bch2_trans_iter_exit(trans, &inode_iter);
2520 printbuf_exit(&buf);
2521 bch_err_fn(c, ret);
2522 return ret;
2523 }
2524
2525 /*
2526 * Check for unreachable inodes, as well as loops in the directory structure:
2527 * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2528 * unreachable:
2529 */
2530 int bch2_check_directory_structure(struct bch_fs *c)
2531 {
2532 pathbuf path = { 0, };
2533 int ret;
2534
2535 ret = bch2_trans_run(c,
2536 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2537 BTREE_ITER_intent|
2538 BTREE_ITER_prefetch|
2539 BTREE_ITER_all_snapshots, k,
2540 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2541 if (!bkey_is_inode(k.k))
2542 continue;
2543
2544 if (bch2_inode_flags(k) & BCH_INODE_unlinked)
2545 continue;
2546
2547 check_path(trans, &path, k);
2548 })));
2549 darray_exit(&path);
2550
2551 bch_err_fn(c, ret);
2552 return ret;
2553 }
2554
2555 struct nlink_table {
2556 size_t nr;
2557 size_t size;
2558
2559 struct nlink {
2560 u64 inum;
2561 u32 snapshot;
2562 u32 count;
2563 } *d;
2564 };
2565
2566 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2567 u64 inum, u32 snapshot)
2568 {
2569 if (t->nr == t->size) {
2570 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2571 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2572
2573 if (!d) {
2574 bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2575 new_size);
2576 return -BCH_ERR_ENOMEM_fsck_add_nlink;
2577 }
2578
2579 if (t->d)
2580 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2581 kvfree(t->d);
2582
2583 t->d = d;
2584 t->size = new_size;
2585 }
2586
2587
2588 t->d[t->nr++] = (struct nlink) {
2589 .inum = inum,
2590 .snapshot = snapshot,
2591 };
2592
2593 return 0;
2594 }
2595
2596 static int nlink_cmp(const void *_l, const void *_r)
2597 {
2598 const struct nlink *l = _l;
2599 const struct nlink *r = _r;
2600
2601 return cmp_int(l->inum, r->inum);
2602 }
2603
2604 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2605 struct nlink_table *links,
2606 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2607 {
2608 struct nlink *link, key = {
2609 .inum = inum, .snapshot = U32_MAX,
2610 };
2611
2612 if (inum < range_start || inum >= range_end)
2613 return;
2614
2615 link = __inline_bsearch(&key, links->d, links->nr,
2616 sizeof(links->d[0]), nlink_cmp);
2617 if (!link)
2618 return;
2619
2620 while (link > links->d && link[0].inum == link[-1].inum)
2621 --link;
2622
2623 for (; link < links->d + links->nr && link->inum == inum; link++)
2624 if (ref_visible(c, s, snapshot, link->snapshot)) {
2625 link->count++;
2626 if (link->snapshot >= snapshot)
2627 break;
2628 }
2629 }
2630
2631 noinline_for_stack
2632 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2633 struct nlink_table *t,
2634 u64 start, u64 *end)
2635 {
2636 int ret = bch2_trans_run(c,
2637 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2638 POS(0, start),
2639 BTREE_ITER_intent|
2640 BTREE_ITER_prefetch|
2641 BTREE_ITER_all_snapshots, k, ({
2642 if (!bkey_is_inode(k.k))
2643 continue;
2644
2645 /* Should never fail, checked by bch2_inode_invalid: */
2646 struct bch_inode_unpacked u;
2647 BUG_ON(bch2_inode_unpack(k, &u));
2648
2649 /*
2650 * Backpointer and directory structure checks are sufficient for
2651 * directories, since they can't have hardlinks:
2652 */
2653 if (S_ISDIR(u.bi_mode))
2654 continue;
2655
2656 if (!u.bi_nlink)
2657 continue;
2658
2659 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2660 if (ret) {
2661 *end = k.k->p.offset;
2662 ret = 0;
2663 break;
2664 }
2665 0;
2666 })));
2667
2668 bch_err_fn(c, ret);
2669 return ret;
2670 }
2671
2672 noinline_for_stack
2673 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2674 u64 range_start, u64 range_end)
2675 {
2676 struct snapshots_seen s;
2677
2678 snapshots_seen_init(&s);
2679
2680 int ret = bch2_trans_run(c,
2681 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2682 BTREE_ITER_intent|
2683 BTREE_ITER_prefetch|
2684 BTREE_ITER_all_snapshots, k, ({
2685 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2686 if (ret)
2687 break;
2688
2689 if (k.k->type == KEY_TYPE_dirent) {
2690 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2691
2692 if (d.v->d_type != DT_DIR &&
2693 d.v->d_type != DT_SUBVOL)
2694 inc_link(c, &s, links, range_start, range_end,
2695 le64_to_cpu(d.v->d_inum), d.k->p.snapshot);
2696 }
2697 0;
2698 })));
2699
2700 snapshots_seen_exit(&s);
2701
2702 bch_err_fn(c, ret);
2703 return ret;
2704 }
2705
2706 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2707 struct bkey_s_c k,
2708 struct nlink_table *links,
2709 size_t *idx, u64 range_end)
2710 {
2711 struct bch_inode_unpacked u;
2712 struct nlink *link = &links->d[*idx];
2713 int ret = 0;
2714
2715 if (k.k->p.offset >= range_end)
2716 return 1;
2717
2718 if (!bkey_is_inode(k.k))
2719 return 0;
2720
2721 BUG_ON(bch2_inode_unpack(k, &u));
2722
2723 if (S_ISDIR(u.bi_mode))
2724 return 0;
2725
2726 if (!u.bi_nlink)
2727 return 0;
2728
2729 while ((cmp_int(link->inum, k.k->p.offset) ?:
2730 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2731 BUG_ON(*idx == links->nr);
2732 link = &links->d[++*idx];
2733 }
2734
2735 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2736 trans, inode_wrong_nlink,
2737 "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2738 u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2739 bch2_inode_nlink_get(&u), link->count)) {
2740 bch2_inode_nlink_set(&u, link->count);
2741 ret = __bch2_fsck_write_inode(trans, &u, k.k->p.snapshot);
2742 }
2743 fsck_err:
2744 return ret;
2745 }
2746
2747 noinline_for_stack
2748 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2749 struct nlink_table *links,
2750 u64 range_start, u64 range_end)
2751 {
2752 size_t idx = 0;
2753
2754 int ret = bch2_trans_run(c,
2755 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2756 POS(0, range_start),
2757 BTREE_ITER_intent|BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
2758 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2759 check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2760 if (ret < 0) {
2761 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2762 return ret;
2763 }
2764
2765 return 0;
2766 }
2767
2768 int bch2_check_nlinks(struct bch_fs *c)
2769 {
2770 struct nlink_table links = { 0 };
2771 u64 this_iter_range_start, next_iter_range_start = 0;
2772 int ret = 0;
2773
2774 do {
2775 this_iter_range_start = next_iter_range_start;
2776 next_iter_range_start = U64_MAX;
2777
2778 ret = check_nlinks_find_hardlinks(c, &links,
2779 this_iter_range_start,
2780 &next_iter_range_start);
2781
2782 ret = check_nlinks_walk_dirents(c, &links,
2783 this_iter_range_start,
2784 next_iter_range_start);
2785 if (ret)
2786 break;
2787
2788 ret = check_nlinks_update_hardlinks(c, &links,
2789 this_iter_range_start,
2790 next_iter_range_start);
2791 if (ret)
2792 break;
2793
2794 links.nr = 0;
2795 } while (next_iter_range_start != U64_MAX);
2796
2797 kvfree(links.d);
2798 bch_err_fn(c, ret);
2799 return ret;
2800 }
2801
2802 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2803 struct bkey_s_c k)
2804 {
2805 struct bkey_s_c_reflink_p p;
2806 struct bkey_i_reflink_p *u;
2807
2808 if (k.k->type != KEY_TYPE_reflink_p)
2809 return 0;
2810
2811 p = bkey_s_c_to_reflink_p(k);
2812
2813 if (!p.v->front_pad && !p.v->back_pad)
2814 return 0;
2815
2816 u = bch2_trans_kmalloc(trans, sizeof(*u));
2817 int ret = PTR_ERR_OR_ZERO(u);
2818 if (ret)
2819 return ret;
2820
2821 bkey_reassemble(&u->k_i, k);
2822 u->v.front_pad = 0;
2823 u->v.back_pad = 0;
2824
2825 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_norun);
2826 }
2827
2828 int bch2_fix_reflink_p(struct bch_fs *c)
2829 {
2830 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2831 return 0;
2832
2833 int ret = bch2_trans_run(c,
2834 for_each_btree_key_commit(trans, iter,
2835 BTREE_ID_extents, POS_MIN,
2836 BTREE_ITER_intent|BTREE_ITER_prefetch|
2837 BTREE_ITER_all_snapshots, k,
2838 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2839 fix_reflink_p_key(trans, &iter, k)));
2840 bch_err_fn(c, ret);
2841 return ret;
2842 }
2843
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.