1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "bcachefs_ioctl.h"
5 #include "bkey_buf.h"
6 #include "btree_cache.h"
7 #include "btree_update.h"
8 #include "buckets.h"
9 #include "darray.h"
10 #include "dirent.h"
11 #include "error.h"
12 #include "fs.h"
13 #include "fs-common.h"
14 #include "fsck.h"
15 #include "inode.h"
16 #include "keylist.h"
17 #include "recovery_passes.h"
18 #include "snapshot.h"
19 #include "super.h"
20 #include "thread_with_file.h"
21 #include "xattr.h"
22
23 #include <linux/bsearch.h>
24 #include <linux/dcache.h> /* struct qstr */
25
inode_points_to_dirent(struct bch_inode_unpacked * inode,struct bkey_s_c_dirent d)26 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
27 struct bkey_s_c_dirent d)
28 {
29 return inode->bi_dir == d.k->p.inode &&
30 inode->bi_dir_offset == d.k->p.offset;
31 }
32
dirent_points_to_inode_nowarn(struct bkey_s_c_dirent d,struct bch_inode_unpacked * inode)33 static int dirent_points_to_inode_nowarn(struct bkey_s_c_dirent d,
34 struct bch_inode_unpacked *inode)
35 {
36 if (d.v->d_type == DT_SUBVOL
37 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
38 : le64_to_cpu(d.v->d_inum) == inode->bi_inum)
39 return 0;
40 return -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
41 }
42
dirent_inode_mismatch_msg(struct printbuf * out,struct bch_fs * c,struct bkey_s_c_dirent dirent,struct bch_inode_unpacked * inode)43 static void dirent_inode_mismatch_msg(struct printbuf *out,
44 struct bch_fs *c,
45 struct bkey_s_c_dirent dirent,
46 struct bch_inode_unpacked *inode)
47 {
48 prt_str(out, "inode points to dirent that does not point back:");
49 prt_newline(out);
50 bch2_bkey_val_to_text(out, c, dirent.s_c);
51 prt_newline(out);
52 bch2_inode_unpacked_to_text(out, inode);
53 }
54
dirent_points_to_inode(struct bch_fs * c,struct bkey_s_c_dirent dirent,struct bch_inode_unpacked * inode)55 static int dirent_points_to_inode(struct bch_fs *c,
56 struct bkey_s_c_dirent dirent,
57 struct bch_inode_unpacked *inode)
58 {
59 int ret = dirent_points_to_inode_nowarn(dirent, inode);
60 if (ret) {
61 struct printbuf buf = PRINTBUF;
62 dirent_inode_mismatch_msg(&buf, c, dirent, inode);
63 bch_warn(c, "%s", buf.buf);
64 printbuf_exit(&buf);
65 }
66 return ret;
67 }
68
69 /*
70 * XXX: this is handling transaction restarts without returning
71 * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
72 */
bch2_count_inode_sectors(struct btree_trans * trans,u64 inum,u32 snapshot)73 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
74 u32 snapshot)
75 {
76 u64 sectors = 0;
77
78 int ret = for_each_btree_key_max(trans, iter, BTREE_ID_extents,
79 SPOS(inum, 0, snapshot),
80 POS(inum, U64_MAX),
81 0, k, ({
82 if (bkey_extent_is_allocation(k.k))
83 sectors += k.k->size;
84 0;
85 }));
86
87 return ret ?: sectors;
88 }
89
bch2_count_subdirs(struct btree_trans * trans,u64 inum,u32 snapshot)90 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
91 u32 snapshot)
92 {
93 u64 subdirs = 0;
94
95 int ret = for_each_btree_key_max(trans, iter, BTREE_ID_dirents,
96 SPOS(inum, 0, snapshot),
97 POS(inum, U64_MAX),
98 0, k, ({
99 if (k.k->type == KEY_TYPE_dirent &&
100 bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
101 subdirs++;
102 0;
103 }));
104
105 return ret ?: subdirs;
106 }
107
subvol_lookup(struct btree_trans * trans,u32 subvol,u32 * snapshot,u64 * inum)108 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
109 u32 *snapshot, u64 *inum)
110 {
111 struct bch_subvolume s;
112 int ret = bch2_subvolume_get(trans, subvol, false, &s);
113
114 *snapshot = le32_to_cpu(s.snapshot);
115 *inum = le64_to_cpu(s.inode);
116 return ret;
117 }
118
lookup_first_inode(struct btree_trans * trans,u64 inode_nr,struct bch_inode_unpacked * inode)119 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
120 struct bch_inode_unpacked *inode)
121 {
122 struct btree_iter iter;
123 struct bkey_s_c k;
124 int ret;
125
126 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inode_nr),
127 BTREE_ITER_all_snapshots, k, ret) {
128 if (k.k->p.offset != inode_nr)
129 break;
130 if (!bkey_is_inode(k.k))
131 continue;
132 ret = bch2_inode_unpack(k, inode);
133 goto found;
134 }
135 ret = -BCH_ERR_ENOENT_inode;
136 found:
137 bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
138 bch2_trans_iter_exit(trans, &iter);
139 return ret;
140 }
141
lookup_inode(struct btree_trans * trans,u64 inode_nr,u32 snapshot,struct bch_inode_unpacked * inode)142 static int lookup_inode(struct btree_trans *trans, u64 inode_nr, u32 snapshot,
143 struct bch_inode_unpacked *inode)
144 {
145 struct btree_iter iter;
146 struct bkey_s_c k;
147 int ret;
148
149 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
150 SPOS(0, inode_nr, snapshot), 0);
151 ret = bkey_err(k);
152 if (ret)
153 goto err;
154
155 ret = bkey_is_inode(k.k)
156 ? bch2_inode_unpack(k, inode)
157 : -BCH_ERR_ENOENT_inode;
158 err:
159 bch2_trans_iter_exit(trans, &iter);
160 return ret;
161 }
162
lookup_dirent_in_snapshot(struct btree_trans * trans,struct bch_hash_info hash_info,subvol_inum dir,struct qstr * name,u64 * target,unsigned * type,u32 snapshot)163 static int lookup_dirent_in_snapshot(struct btree_trans *trans,
164 struct bch_hash_info hash_info,
165 subvol_inum dir, struct qstr *name,
166 u64 *target, unsigned *type, u32 snapshot)
167 {
168 struct btree_iter iter;
169 struct bkey_s_c k = bch2_hash_lookup_in_snapshot(trans, &iter, bch2_dirent_hash_desc,
170 &hash_info, dir, name, 0, snapshot);
171 int ret = bkey_err(k);
172 if (ret)
173 return ret;
174
175 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
176 *target = le64_to_cpu(d.v->d_inum);
177 *type = d.v->d_type;
178 bch2_trans_iter_exit(trans, &iter);
179 return 0;
180 }
181
__remove_dirent(struct btree_trans * trans,struct bpos pos)182 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
183 {
184 struct bch_fs *c = trans->c;
185 struct btree_iter iter;
186 struct bch_inode_unpacked dir_inode;
187 struct bch_hash_info dir_hash_info;
188 int ret;
189
190 ret = lookup_first_inode(trans, pos.inode, &dir_inode);
191 if (ret)
192 goto err;
193
194 dir_hash_info = bch2_hash_info_init(c, &dir_inode);
195
196 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_intent);
197
198 ret = bch2_btree_iter_traverse(&iter) ?:
199 bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
200 &dir_hash_info, &iter,
201 BTREE_UPDATE_internal_snapshot_node);
202 bch2_trans_iter_exit(trans, &iter);
203 err:
204 bch_err_fn(c, ret);
205 return ret;
206 }
207
208 /*
209 * Find any subvolume associated with a tree of snapshots
210 * We can't rely on master_subvol - it might have been deleted.
211 */
find_snapshot_tree_subvol(struct btree_trans * trans,u32 tree_id,u32 * subvol)212 static int find_snapshot_tree_subvol(struct btree_trans *trans,
213 u32 tree_id, u32 *subvol)
214 {
215 struct btree_iter iter;
216 struct bkey_s_c k;
217 int ret;
218
219 for_each_btree_key_norestart(trans, iter, BTREE_ID_snapshots, POS_MIN, 0, k, ret) {
220 if (k.k->type != KEY_TYPE_snapshot)
221 continue;
222
223 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
224 if (le32_to_cpu(s.v->tree) != tree_id)
225 continue;
226
227 if (s.v->subvol) {
228 *subvol = le32_to_cpu(s.v->subvol);
229 goto found;
230 }
231 }
232 ret = -BCH_ERR_ENOENT_no_snapshot_tree_subvol;
233 found:
234 bch2_trans_iter_exit(trans, &iter);
235 return ret;
236 }
237
238 /* Get lost+found, create if it doesn't exist: */
lookup_lostfound(struct btree_trans * trans,u32 snapshot,struct bch_inode_unpacked * lostfound,u64 reattaching_inum)239 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
240 struct bch_inode_unpacked *lostfound,
241 u64 reattaching_inum)
242 {
243 struct bch_fs *c = trans->c;
244 struct qstr lostfound_str = QSTR("lost+found");
245 struct btree_iter lostfound_iter = { NULL };
246 u64 inum = 0;
247 unsigned d_type = 0;
248 int ret;
249
250 struct bch_snapshot_tree st;
251 ret = bch2_snapshot_tree_lookup(trans,
252 bch2_snapshot_tree(c, snapshot), &st);
253 if (ret)
254 return ret;
255
256 u32 subvolid;
257 ret = find_snapshot_tree_subvol(trans,
258 bch2_snapshot_tree(c, snapshot), &subvolid);
259 bch_err_msg(c, ret, "finding subvol associated with snapshot tree %u",
260 bch2_snapshot_tree(c, snapshot));
261 if (ret)
262 return ret;
263
264 struct bch_subvolume subvol;
265 ret = bch2_subvolume_get(trans, subvolid, false, &subvol);
266 bch_err_msg(c, ret, "looking up subvol %u for snapshot %u", subvolid, snapshot);
267 if (ret)
268 return ret;
269
270 if (!subvol.inode) {
271 struct btree_iter iter;
272 struct bkey_i_subvolume *subvol = bch2_bkey_get_mut_typed(trans, &iter,
273 BTREE_ID_subvolumes, POS(0, subvolid),
274 0, subvolume);
275 ret = PTR_ERR_OR_ZERO(subvol);
276 if (ret)
277 return ret;
278
279 subvol->v.inode = cpu_to_le64(reattaching_inum);
280 bch2_trans_iter_exit(trans, &iter);
281 }
282
283 subvol_inum root_inum = {
284 .subvol = subvolid,
285 .inum = le64_to_cpu(subvol.inode)
286 };
287
288 struct bch_inode_unpacked root_inode;
289 struct bch_hash_info root_hash_info;
290 ret = lookup_inode(trans, root_inum.inum, snapshot, &root_inode);
291 bch_err_msg(c, ret, "looking up root inode %llu for subvol %u",
292 root_inum.inum, subvolid);
293 if (ret)
294 return ret;
295
296 root_hash_info = bch2_hash_info_init(c, &root_inode);
297
298 ret = lookup_dirent_in_snapshot(trans, root_hash_info, root_inum,
299 &lostfound_str, &inum, &d_type, snapshot);
300 if (bch2_err_matches(ret, ENOENT))
301 goto create_lostfound;
302
303 bch_err_fn(c, ret);
304 if (ret)
305 return ret;
306
307 if (d_type != DT_DIR) {
308 bch_err(c, "error looking up lost+found: not a directory");
309 return -BCH_ERR_ENOENT_not_directory;
310 }
311
312 /*
313 * The bch2_check_dirents pass has already run, dangling dirents
314 * shouldn't exist here:
315 */
316 ret = lookup_inode(trans, inum, snapshot, lostfound);
317 bch_err_msg(c, ret, "looking up lost+found %llu:%u in (root inode %llu, snapshot root %u)",
318 inum, snapshot, root_inum.inum, bch2_snapshot_root(c, snapshot));
319 return ret;
320
321 create_lostfound:
322 /*
323 * we always create lost+found in the root snapshot; we don't want
324 * different branches of the snapshot tree to have different lost+found
325 */
326 snapshot = le32_to_cpu(st.root_snapshot);
327 /*
328 * XXX: we could have a nicer log message here if we had a nice way to
329 * walk backpointers to print a path
330 */
331 struct printbuf path = PRINTBUF;
332 ret = bch2_inum_to_path(trans, root_inum, &path);
333 if (ret)
334 goto err;
335
336 bch_notice(c, "creating %s/lost+found in subvol %llu snapshot %u",
337 path.buf, root_inum.subvol, snapshot);
338 printbuf_exit(&path);
339
340 u64 now = bch2_current_time(c);
341 u64 cpu = raw_smp_processor_id();
342
343 bch2_inode_init_early(c, lostfound);
344 bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
345 lostfound->bi_dir = root_inode.bi_inum;
346 lostfound->bi_snapshot = le32_to_cpu(st.root_snapshot);
347
348 root_inode.bi_nlink++;
349
350 ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
351 if (ret)
352 goto err;
353
354 bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
355 ret = bch2_btree_iter_traverse(&lostfound_iter);
356 if (ret)
357 goto err;
358
359 ret = bch2_dirent_create_snapshot(trans,
360 0, root_inode.bi_inum, snapshot, &root_hash_info,
361 mode_to_type(lostfound->bi_mode),
362 &lostfound_str,
363 lostfound->bi_inum,
364 &lostfound->bi_dir_offset,
365 STR_HASH_must_create) ?:
366 bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
367 BTREE_UPDATE_internal_snapshot_node);
368 err:
369 bch_err_msg(c, ret, "creating lost+found");
370 bch2_trans_iter_exit(trans, &lostfound_iter);
371 return ret;
372 }
373
inode_should_reattach(struct bch_inode_unpacked * inode)374 static inline bool inode_should_reattach(struct bch_inode_unpacked *inode)
375 {
376 if (inode->bi_inum == BCACHEFS_ROOT_INO &&
377 inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)
378 return false;
379
380 return !inode->bi_dir && !(inode->bi_flags & BCH_INODE_unlinked);
381 }
382
maybe_delete_dirent(struct btree_trans * trans,struct bpos d_pos,u32 snapshot)383 static int maybe_delete_dirent(struct btree_trans *trans, struct bpos d_pos, u32 snapshot)
384 {
385 struct btree_iter iter;
386 struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_dirents,
387 SPOS(d_pos.inode, d_pos.offset, snapshot),
388 BTREE_ITER_intent|
389 BTREE_ITER_with_updates);
390 int ret = bkey_err(k);
391 if (ret)
392 return ret;
393
394 if (bpos_eq(k.k->p, d_pos)) {
395 /*
396 * delet_at() doesn't work because the update path doesn't
397 * internally use BTREE_ITER_with_updates yet
398 */
399 struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
400 ret = PTR_ERR_OR_ZERO(k);
401 if (ret)
402 goto err;
403
404 bkey_init(&k->k);
405 k->k.type = KEY_TYPE_whiteout;
406 k->k.p = iter.pos;
407 ret = bch2_trans_update(trans, &iter, k, BTREE_UPDATE_internal_snapshot_node);
408 }
409 err:
410 bch2_trans_iter_exit(trans, &iter);
411 return ret;
412 }
413
reattach_inode(struct btree_trans * trans,struct bch_inode_unpacked * inode)414 static int reattach_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
415 {
416 struct bch_fs *c = trans->c;
417 struct bch_inode_unpacked lostfound;
418 char name_buf[20];
419 int ret;
420
421 u32 dirent_snapshot = inode->bi_snapshot;
422 if (inode->bi_subvol) {
423 inode->bi_parent_subvol = BCACHEFS_ROOT_SUBVOL;
424
425 u64 root_inum;
426 ret = subvol_lookup(trans, inode->bi_parent_subvol,
427 &dirent_snapshot, &root_inum);
428 if (ret)
429 return ret;
430
431 snprintf(name_buf, sizeof(name_buf), "subvol-%u", inode->bi_subvol);
432 } else {
433 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
434 }
435
436 ret = lookup_lostfound(trans, dirent_snapshot, &lostfound, inode->bi_inum);
437 if (ret)
438 return ret;
439
440 lostfound.bi_nlink += S_ISDIR(inode->bi_mode);
441
442 /* ensure lost+found inode is also present in inode snapshot */
443 if (!inode->bi_subvol) {
444 BUG_ON(!bch2_snapshot_is_ancestor(c, inode->bi_snapshot, lostfound.bi_snapshot));
445 lostfound.bi_snapshot = inode->bi_snapshot;
446 }
447
448 ret = __bch2_fsck_write_inode(trans, &lostfound);
449 if (ret)
450 return ret;
451
452 struct bch_hash_info dir_hash = bch2_hash_info_init(c, &lostfound);
453 struct qstr name = QSTR(name_buf);
454
455 inode->bi_dir = lostfound.bi_inum;
456
457 ret = bch2_dirent_create_snapshot(trans,
458 inode->bi_parent_subvol, lostfound.bi_inum,
459 dirent_snapshot,
460 &dir_hash,
461 inode_d_type(inode),
462 &name,
463 inode->bi_subvol ?: inode->bi_inum,
464 &inode->bi_dir_offset,
465 STR_HASH_must_create);
466 if (ret) {
467 bch_err_msg(c, ret, "error creating dirent");
468 return ret;
469 }
470
471 ret = __bch2_fsck_write_inode(trans, inode);
472 if (ret)
473 return ret;
474
475 /*
476 * Fix up inodes in child snapshots: if they should also be reattached
477 * update the backpointer field, if they should not be we need to emit
478 * whiteouts for the dirent we just created.
479 */
480 if (!inode->bi_subvol && bch2_snapshot_is_leaf(c, inode->bi_snapshot) <= 0) {
481 snapshot_id_list whiteouts_done;
482 struct btree_iter iter;
483 struct bkey_s_c k;
484
485 darray_init(&whiteouts_done);
486
487 for_each_btree_key_reverse_norestart(trans, iter,
488 BTREE_ID_inodes, SPOS(0, inode->bi_inum, inode->bi_snapshot - 1),
489 BTREE_ITER_all_snapshots|BTREE_ITER_intent, k, ret) {
490 if (k.k->p.offset != inode->bi_inum)
491 break;
492
493 if (!bkey_is_inode(k.k) ||
494 !bch2_snapshot_is_ancestor(c, k.k->p.snapshot, inode->bi_snapshot) ||
495 snapshot_list_has_ancestor(c, &whiteouts_done, k.k->p.snapshot))
496 continue;
497
498 struct bch_inode_unpacked child_inode;
499 ret = bch2_inode_unpack(k, &child_inode);
500 if (ret)
501 break;
502
503 if (!inode_should_reattach(&child_inode)) {
504 ret = maybe_delete_dirent(trans,
505 SPOS(lostfound.bi_inum, inode->bi_dir_offset,
506 dirent_snapshot),
507 k.k->p.snapshot);
508 if (ret)
509 break;
510
511 ret = snapshot_list_add(c, &whiteouts_done, k.k->p.snapshot);
512 if (ret)
513 break;
514 } else {
515 iter.snapshot = k.k->p.snapshot;
516 child_inode.bi_dir = inode->bi_dir;
517 child_inode.bi_dir_offset = inode->bi_dir_offset;
518
519 ret = bch2_inode_write_flags(trans, &iter, &child_inode,
520 BTREE_UPDATE_internal_snapshot_node);
521 if (ret)
522 break;
523 }
524 }
525 darray_exit(&whiteouts_done);
526 bch2_trans_iter_exit(trans, &iter);
527 }
528
529 return ret;
530 }
531
dirent_get_by_pos(struct btree_trans * trans,struct btree_iter * iter,struct bpos pos)532 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
533 struct btree_iter *iter,
534 struct bpos pos)
535 {
536 return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
537 }
538
remove_backpointer(struct btree_trans * trans,struct bch_inode_unpacked * inode)539 static int remove_backpointer(struct btree_trans *trans,
540 struct bch_inode_unpacked *inode)
541 {
542 if (!inode->bi_dir)
543 return 0;
544
545 struct bch_fs *c = trans->c;
546 struct btree_iter iter;
547 struct bkey_s_c_dirent d = dirent_get_by_pos(trans, &iter,
548 SPOS(inode->bi_dir, inode->bi_dir_offset, inode->bi_snapshot));
549 int ret = bkey_err(d) ?:
550 dirent_points_to_inode(c, d, inode) ?:
551 __remove_dirent(trans, d.k->p);
552 bch2_trans_iter_exit(trans, &iter);
553 return ret;
554 }
555
reattach_subvol(struct btree_trans * trans,struct bkey_s_c_subvolume s)556 static int reattach_subvol(struct btree_trans *trans, struct bkey_s_c_subvolume s)
557 {
558 struct bch_fs *c = trans->c;
559
560 struct bch_inode_unpacked inode;
561 int ret = bch2_inode_find_by_inum_trans(trans,
562 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
563 &inode);
564 if (ret)
565 return ret;
566
567 ret = remove_backpointer(trans, &inode);
568 if (!bch2_err_matches(ret, ENOENT))
569 bch_err_msg(c, ret, "removing dirent");
570 if (ret)
571 return ret;
572
573 ret = reattach_inode(trans, &inode);
574 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
575 return ret;
576 }
577
reconstruct_subvol(struct btree_trans * trans,u32 snapshotid,u32 subvolid,u64 inum)578 static int reconstruct_subvol(struct btree_trans *trans, u32 snapshotid, u32 subvolid, u64 inum)
579 {
580 struct bch_fs *c = trans->c;
581
582 if (!bch2_snapshot_is_leaf(c, snapshotid)) {
583 bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
584 return -BCH_ERR_fsck_repair_unimplemented;
585 }
586
587 /*
588 * If inum isn't set, that means we're being called from check_dirents,
589 * not check_inodes - the root of this subvolume doesn't exist or we
590 * would have found it there:
591 */
592 if (!inum) {
593 struct btree_iter inode_iter = {};
594 struct bch_inode_unpacked new_inode;
595 u64 cpu = raw_smp_processor_id();
596
597 bch2_inode_init_early(c, &new_inode);
598 bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, S_IFDIR|0755, 0, NULL);
599
600 new_inode.bi_subvol = subvolid;
601
602 int ret = bch2_inode_create(trans, &inode_iter, &new_inode, snapshotid, cpu) ?:
603 bch2_btree_iter_traverse(&inode_iter) ?:
604 bch2_inode_write(trans, &inode_iter, &new_inode);
605 bch2_trans_iter_exit(trans, &inode_iter);
606 if (ret)
607 return ret;
608
609 inum = new_inode.bi_inum;
610 }
611
612 bch_info(c, "reconstructing subvol %u with root inode %llu", subvolid, inum);
613
614 struct bkey_i_subvolume *new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
615 int ret = PTR_ERR_OR_ZERO(new_subvol);
616 if (ret)
617 return ret;
618
619 bkey_subvolume_init(&new_subvol->k_i);
620 new_subvol->k.p.offset = subvolid;
621 new_subvol->v.snapshot = cpu_to_le32(snapshotid);
622 new_subvol->v.inode = cpu_to_le64(inum);
623 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &new_subvol->k_i, 0);
624 if (ret)
625 return ret;
626
627 struct btree_iter iter;
628 struct bkey_i_snapshot *s = bch2_bkey_get_mut_typed(trans, &iter,
629 BTREE_ID_snapshots, POS(0, snapshotid),
630 0, snapshot);
631 ret = PTR_ERR_OR_ZERO(s);
632 bch_err_msg(c, ret, "getting snapshot %u", snapshotid);
633 if (ret)
634 return ret;
635
636 u32 snapshot_tree = le32_to_cpu(s->v.tree);
637
638 s->v.subvol = cpu_to_le32(subvolid);
639 SET_BCH_SNAPSHOT_SUBVOL(&s->v, true);
640 bch2_trans_iter_exit(trans, &iter);
641
642 struct bkey_i_snapshot_tree *st = bch2_bkey_get_mut_typed(trans, &iter,
643 BTREE_ID_snapshot_trees, POS(0, snapshot_tree),
644 0, snapshot_tree);
645 ret = PTR_ERR_OR_ZERO(st);
646 bch_err_msg(c, ret, "getting snapshot tree %u", snapshot_tree);
647 if (ret)
648 return ret;
649
650 if (!st->v.master_subvol)
651 st->v.master_subvol = cpu_to_le32(subvolid);
652
653 bch2_trans_iter_exit(trans, &iter);
654 return 0;
655 }
656
reconstruct_inode(struct btree_trans * trans,enum btree_id btree,u32 snapshot,u64 inum)657 static int reconstruct_inode(struct btree_trans *trans, enum btree_id btree, u32 snapshot, u64 inum)
658 {
659 struct bch_fs *c = trans->c;
660 unsigned i_mode = S_IFREG;
661 u64 i_size = 0;
662
663 switch (btree) {
664 case BTREE_ID_extents: {
665 struct btree_iter iter = {};
666
667 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, SPOS(inum, U64_MAX, snapshot), 0);
668 struct bkey_s_c k = bch2_btree_iter_peek_prev_min(&iter, POS(inum, 0));
669 bch2_trans_iter_exit(trans, &iter);
670 int ret = bkey_err(k);
671 if (ret)
672 return ret;
673
674 i_size = k.k->p.offset << 9;
675 break;
676 }
677 case BTREE_ID_dirents:
678 i_mode = S_IFDIR;
679 break;
680 case BTREE_ID_xattrs:
681 break;
682 default:
683 BUG();
684 }
685
686 struct bch_inode_unpacked new_inode;
687 bch2_inode_init_early(c, &new_inode);
688 bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, i_mode|0600, 0, NULL);
689 new_inode.bi_size = i_size;
690 new_inode.bi_inum = inum;
691 new_inode.bi_snapshot = snapshot;
692
693 return __bch2_fsck_write_inode(trans, &new_inode);
694 }
695
696 struct snapshots_seen {
697 struct bpos pos;
698 snapshot_id_list ids;
699 };
700
snapshots_seen_exit(struct snapshots_seen * s)701 static inline void snapshots_seen_exit(struct snapshots_seen *s)
702 {
703 darray_exit(&s->ids);
704 }
705
snapshots_seen_init(struct snapshots_seen * s)706 static inline void snapshots_seen_init(struct snapshots_seen *s)
707 {
708 memset(s, 0, sizeof(*s));
709 }
710
snapshots_seen_add_inorder(struct bch_fs * c,struct snapshots_seen * s,u32 id)711 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
712 {
713 u32 *i;
714 __darray_for_each(s->ids, i) {
715 if (*i == id)
716 return 0;
717 if (*i > id)
718 break;
719 }
720
721 int ret = darray_insert_item(&s->ids, i - s->ids.data, id);
722 if (ret)
723 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
724 s->ids.size);
725 return ret;
726 }
727
snapshots_seen_update(struct bch_fs * c,struct snapshots_seen * s,enum btree_id btree_id,struct bpos pos)728 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
729 enum btree_id btree_id, struct bpos pos)
730 {
731 if (!bkey_eq(s->pos, pos))
732 s->ids.nr = 0;
733 s->pos = pos;
734
735 return snapshot_list_add_nodup(c, &s->ids, pos.snapshot);
736 }
737
738 /**
739 * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
740 * and @ancestor hasn't been overwritten in @seen
741 *
742 * @c: filesystem handle
743 * @seen: list of snapshot ids already seen at current position
744 * @id: descendent snapshot id
745 * @ancestor: ancestor snapshot id
746 *
747 * Returns: whether key in @ancestor snapshot is visible in @id snapshot
748 */
key_visible_in_snapshot(struct bch_fs * c,struct snapshots_seen * seen,u32 id,u32 ancestor)749 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
750 u32 id, u32 ancestor)
751 {
752 ssize_t i;
753
754 EBUG_ON(id > ancestor);
755
756 /* @ancestor should be the snapshot most recently added to @seen */
757 EBUG_ON(ancestor != seen->pos.snapshot);
758 EBUG_ON(ancestor != darray_last(seen->ids));
759
760 if (id == ancestor)
761 return true;
762
763 if (!bch2_snapshot_is_ancestor(c, id, ancestor))
764 return false;
765
766 /*
767 * We know that @id is a descendant of @ancestor, we're checking if
768 * we've seen a key that overwrote @ancestor - i.e. also a descendent of
769 * @ascestor and with @id as a descendent.
770 *
771 * But we already know that we're scanning IDs between @id and @ancestor
772 * numerically, since snapshot ID lists are kept sorted, so if we find
773 * an id that's an ancestor of @id we're done:
774 */
775
776 for (i = seen->ids.nr - 2;
777 i >= 0 && seen->ids.data[i] >= id;
778 --i)
779 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i]))
780 return false;
781
782 return true;
783 }
784
785 /**
786 * ref_visible - given a key with snapshot id @src that points to a key with
787 * snapshot id @dst, test whether there is some snapshot in which @dst is
788 * visible.
789 *
790 * @c: filesystem handle
791 * @s: list of snapshot IDs already seen at @src
792 * @src: snapshot ID of src key
793 * @dst: snapshot ID of dst key
794 * Returns: true if there is some snapshot in which @dst is visible
795 *
796 * Assumes we're visiting @src keys in natural key order
797 */
ref_visible(struct bch_fs * c,struct snapshots_seen * s,u32 src,u32 dst)798 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
799 u32 src, u32 dst)
800 {
801 return dst <= src
802 ? key_visible_in_snapshot(c, s, dst, src)
803 : bch2_snapshot_is_ancestor(c, src, dst);
804 }
805
ref_visible2(struct bch_fs * c,u32 src,struct snapshots_seen * src_seen,u32 dst,struct snapshots_seen * dst_seen)806 static int ref_visible2(struct bch_fs *c,
807 u32 src, struct snapshots_seen *src_seen,
808 u32 dst, struct snapshots_seen *dst_seen)
809 {
810 if (dst > src) {
811 swap(dst, src);
812 swap(dst_seen, src_seen);
813 }
814 return key_visible_in_snapshot(c, src_seen, dst, src);
815 }
816
817 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \
818 for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && \
819 (_i)->snapshot <= (_snapshot); _i++) \
820 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
821
822 struct inode_walker_entry {
823 struct bch_inode_unpacked inode;
824 u32 snapshot;
825 u64 count;
826 u64 i_size;
827 };
828
829 struct inode_walker {
830 bool first_this_inode;
831 bool have_inodes;
832 bool recalculate_sums;
833 struct bpos last_pos;
834
835 DARRAY(struct inode_walker_entry) inodes;
836 snapshot_id_list deletes;
837 };
838
inode_walker_exit(struct inode_walker * w)839 static void inode_walker_exit(struct inode_walker *w)
840 {
841 darray_exit(&w->inodes);
842 darray_exit(&w->deletes);
843 }
844
inode_walker_init(void)845 static struct inode_walker inode_walker_init(void)
846 {
847 return (struct inode_walker) { 0, };
848 }
849
add_inode(struct bch_fs * c,struct inode_walker * w,struct bkey_s_c inode)850 static int add_inode(struct bch_fs *c, struct inode_walker *w,
851 struct bkey_s_c inode)
852 {
853 struct bch_inode_unpacked u;
854
855 return bch2_inode_unpack(inode, &u) ?:
856 darray_push(&w->inodes, ((struct inode_walker_entry) {
857 .inode = u,
858 .snapshot = inode.k->p.snapshot,
859 }));
860 }
861
get_inodes_all_snapshots(struct btree_trans * trans,struct inode_walker * w,u64 inum)862 static int get_inodes_all_snapshots(struct btree_trans *trans,
863 struct inode_walker *w, u64 inum)
864 {
865 struct bch_fs *c = trans->c;
866 struct btree_iter iter;
867 struct bkey_s_c k;
868 int ret;
869
870 /*
871 * We no longer have inodes for w->last_pos; clear this to avoid
872 * screwing up check_i_sectors/check_subdir_count if we take a
873 * transaction restart here:
874 */
875 w->have_inodes = false;
876 w->recalculate_sums = false;
877 w->inodes.nr = 0;
878
879 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
880 BTREE_ITER_all_snapshots, k, ret) {
881 if (k.k->p.offset != inum)
882 break;
883
884 if (bkey_is_inode(k.k))
885 add_inode(c, w, k);
886 }
887 bch2_trans_iter_exit(trans, &iter);
888
889 if (ret)
890 return ret;
891
892 w->first_this_inode = true;
893 w->have_inodes = true;
894 return 0;
895 }
896
897 static struct inode_walker_entry *
lookup_inode_for_snapshot(struct bch_fs * c,struct inode_walker * w,struct bkey_s_c k)898 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w, struct bkey_s_c k)
899 {
900 bool is_whiteout = k.k->type == KEY_TYPE_whiteout;
901
902 struct inode_walker_entry *i;
903 __darray_for_each(w->inodes, i)
904 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i->snapshot))
905 goto found;
906
907 return NULL;
908 found:
909 BUG_ON(k.k->p.snapshot > i->snapshot);
910
911 if (k.k->p.snapshot != i->snapshot && !is_whiteout) {
912 struct inode_walker_entry new = *i;
913
914 new.snapshot = k.k->p.snapshot;
915 new.count = 0;
916 new.i_size = 0;
917
918 struct printbuf buf = PRINTBUF;
919 bch2_bkey_val_to_text(&buf, c, k);
920
921 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u\n"
922 "unexpected because we should always update the inode when we update a key in that inode\n"
923 "%s",
924 w->last_pos.inode, k.k->p.snapshot, i->snapshot, buf.buf);
925 printbuf_exit(&buf);
926
927 while (i > w->inodes.data && i[-1].snapshot > k.k->p.snapshot)
928 --i;
929
930 size_t pos = i - w->inodes.data;
931 int ret = darray_insert_item(&w->inodes, pos, new);
932 if (ret)
933 return ERR_PTR(ret);
934
935 i = w->inodes.data + pos;
936 }
937
938 return i;
939 }
940
walk_inode(struct btree_trans * trans,struct inode_walker * w,struct bkey_s_c k)941 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
942 struct inode_walker *w,
943 struct bkey_s_c k)
944 {
945 if (w->last_pos.inode != k.k->p.inode) {
946 int ret = get_inodes_all_snapshots(trans, w, k.k->p.inode);
947 if (ret)
948 return ERR_PTR(ret);
949 }
950
951 w->last_pos = k.k->p;
952
953 return lookup_inode_for_snapshot(trans->c, w, k);
954 }
955
get_visible_inodes(struct btree_trans * trans,struct inode_walker * w,struct snapshots_seen * s,u64 inum)956 static int get_visible_inodes(struct btree_trans *trans,
957 struct inode_walker *w,
958 struct snapshots_seen *s,
959 u64 inum)
960 {
961 struct bch_fs *c = trans->c;
962 struct btree_iter iter;
963 struct bkey_s_c k;
964 int ret;
965
966 w->inodes.nr = 0;
967 w->deletes.nr = 0;
968
969 for_each_btree_key_reverse_norestart(trans, iter, BTREE_ID_inodes, SPOS(0, inum, s->pos.snapshot),
970 BTREE_ITER_all_snapshots, k, ret) {
971 if (k.k->p.offset != inum)
972 break;
973
974 if (!ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot))
975 continue;
976
977 if (snapshot_list_has_ancestor(c, &w->deletes, k.k->p.snapshot))
978 continue;
979
980 ret = bkey_is_inode(k.k)
981 ? add_inode(c, w, k)
982 : snapshot_list_add(c, &w->deletes, k.k->p.snapshot);
983 if (ret)
984 break;
985 }
986 bch2_trans_iter_exit(trans, &iter);
987
988 return ret;
989 }
990
991 /*
992 * Prefer to delete the first one, since that will be the one at the wrong
993 * offset:
994 * return value: 0 -> delete k1, 1 -> delete k2
995 */
bch2_fsck_update_backpointers(struct btree_trans * trans,struct snapshots_seen * s,const struct bch_hash_desc desc,struct bch_hash_info * hash_info,struct bkey_i * new)996 int bch2_fsck_update_backpointers(struct btree_trans *trans,
997 struct snapshots_seen *s,
998 const struct bch_hash_desc desc,
999 struct bch_hash_info *hash_info,
1000 struct bkey_i *new)
1001 {
1002 if (new->k.type != KEY_TYPE_dirent)
1003 return 0;
1004
1005 struct bkey_i_dirent *d = bkey_i_to_dirent(new);
1006 struct inode_walker target = inode_walker_init();
1007 int ret = 0;
1008
1009 if (d->v.d_type == DT_SUBVOL) {
1010 BUG();
1011 } else {
1012 ret = get_visible_inodes(trans, &target, s, le64_to_cpu(d->v.d_inum));
1013 if (ret)
1014 goto err;
1015
1016 darray_for_each(target.inodes, i) {
1017 i->inode.bi_dir_offset = d->k.p.offset;
1018 ret = __bch2_fsck_write_inode(trans, &i->inode);
1019 if (ret)
1020 goto err;
1021 }
1022 }
1023 err:
1024 inode_walker_exit(&target);
1025 return ret;
1026 }
1027
inode_get_dirent(struct btree_trans * trans,struct btree_iter * iter,struct bch_inode_unpacked * inode,u32 * snapshot)1028 static struct bkey_s_c_dirent inode_get_dirent(struct btree_trans *trans,
1029 struct btree_iter *iter,
1030 struct bch_inode_unpacked *inode,
1031 u32 *snapshot)
1032 {
1033 if (inode->bi_subvol) {
1034 u64 inum;
1035 int ret = subvol_lookup(trans, inode->bi_parent_subvol, snapshot, &inum);
1036 if (ret)
1037 return ((struct bkey_s_c_dirent) { .k = ERR_PTR(ret) });
1038 }
1039
1040 return dirent_get_by_pos(trans, iter, SPOS(inode->bi_dir, inode->bi_dir_offset, *snapshot));
1041 }
1042
check_inode_deleted_list(struct btree_trans * trans,struct bpos p)1043 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
1044 {
1045 struct btree_iter iter;
1046 struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
1047 int ret = bkey_err(k) ?: k.k->type == KEY_TYPE_set;
1048 bch2_trans_iter_exit(trans, &iter);
1049 return ret;
1050 }
1051
check_inode_dirent_inode(struct btree_trans * trans,struct bch_inode_unpacked * inode,bool * write_inode)1052 static int check_inode_dirent_inode(struct btree_trans *trans,
1053 struct bch_inode_unpacked *inode,
1054 bool *write_inode)
1055 {
1056 struct bch_fs *c = trans->c;
1057 struct printbuf buf = PRINTBUF;
1058
1059 u32 inode_snapshot = inode->bi_snapshot;
1060 struct btree_iter dirent_iter = {};
1061 struct bkey_s_c_dirent d = inode_get_dirent(trans, &dirent_iter, inode, &inode_snapshot);
1062 int ret = bkey_err(d);
1063 if (ret && !bch2_err_matches(ret, ENOENT))
1064 return ret;
1065
1066 if (fsck_err_on(ret,
1067 trans, inode_points_to_missing_dirent,
1068 "inode points to missing dirent\n%s",
1069 (bch2_inode_unpacked_to_text(&buf, inode), buf.buf)) ||
1070 fsck_err_on(!ret && dirent_points_to_inode_nowarn(d, inode),
1071 trans, inode_points_to_wrong_dirent,
1072 "%s",
1073 (printbuf_reset(&buf),
1074 dirent_inode_mismatch_msg(&buf, c, d, inode),
1075 buf.buf))) {
1076 /*
1077 * We just clear the backpointer fields for now. If we find a
1078 * dirent that points to this inode in check_dirents(), we'll
1079 * update it then; then when we get to check_path() if the
1080 * backpointer is still 0 we'll reattach it.
1081 */
1082 inode->bi_dir = 0;
1083 inode->bi_dir_offset = 0;
1084 *write_inode = true;
1085 }
1086
1087 ret = 0;
1088 fsck_err:
1089 bch2_trans_iter_exit(trans, &dirent_iter);
1090 printbuf_exit(&buf);
1091 bch_err_fn(c, ret);
1092 return ret;
1093 }
1094
get_snapshot_root_inode(struct btree_trans * trans,struct bch_inode_unpacked * root,u64 inum)1095 static int get_snapshot_root_inode(struct btree_trans *trans,
1096 struct bch_inode_unpacked *root,
1097 u64 inum)
1098 {
1099 struct btree_iter iter;
1100 struct bkey_s_c k;
1101 int ret = 0;
1102
1103 for_each_btree_key_reverse_norestart(trans, iter, BTREE_ID_inodes,
1104 SPOS(0, inum, U32_MAX),
1105 BTREE_ITER_all_snapshots, k, ret) {
1106 if (k.k->p.offset != inum)
1107 break;
1108 if (bkey_is_inode(k.k))
1109 goto found_root;
1110 }
1111 if (ret)
1112 goto err;
1113 BUG();
1114 found_root:
1115 ret = bch2_inode_unpack(k, root);
1116 err:
1117 bch2_trans_iter_exit(trans, &iter);
1118 return ret;
1119 }
1120
check_inode(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,struct bch_inode_unpacked * snapshot_root,struct snapshots_seen * s)1121 static int check_inode(struct btree_trans *trans,
1122 struct btree_iter *iter,
1123 struct bkey_s_c k,
1124 struct bch_inode_unpacked *snapshot_root,
1125 struct snapshots_seen *s)
1126 {
1127 struct bch_fs *c = trans->c;
1128 struct printbuf buf = PRINTBUF;
1129 struct bch_inode_unpacked u;
1130 bool do_update = false;
1131 int ret;
1132
1133 ret = bch2_check_key_has_snapshot(trans, iter, k);
1134 if (ret < 0)
1135 goto err;
1136 if (ret)
1137 return 0;
1138
1139 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1140 if (ret)
1141 goto err;
1142
1143 if (!bkey_is_inode(k.k))
1144 return 0;
1145
1146 ret = bch2_inode_unpack(k, &u);
1147 if (ret)
1148 goto err;
1149
1150 if (snapshot_root->bi_inum != u.bi_inum) {
1151 ret = get_snapshot_root_inode(trans, snapshot_root, u.bi_inum);
1152 if (ret)
1153 goto err;
1154 }
1155
1156 if (fsck_err_on(u.bi_hash_seed != snapshot_root->bi_hash_seed ||
1157 INODE_STR_HASH(&u) != INODE_STR_HASH(snapshot_root),
1158 trans, inode_snapshot_mismatch,
1159 "inode hash info in different snapshots don't match")) {
1160 u.bi_hash_seed = snapshot_root->bi_hash_seed;
1161 SET_INODE_STR_HASH(&u, INODE_STR_HASH(snapshot_root));
1162 do_update = true;
1163 }
1164
1165 if (u.bi_dir || u.bi_dir_offset) {
1166 ret = check_inode_dirent_inode(trans, &u, &do_update);
1167 if (ret)
1168 goto err;
1169 }
1170
1171 if (fsck_err_on(u.bi_dir && (u.bi_flags & BCH_INODE_unlinked),
1172 trans, inode_unlinked_but_has_dirent,
1173 "inode unlinked but has dirent\n%s",
1174 (printbuf_reset(&buf),
1175 bch2_inode_unpacked_to_text(&buf, &u),
1176 buf.buf))) {
1177 u.bi_flags &= ~BCH_INODE_unlinked;
1178 do_update = true;
1179 }
1180
1181 if (S_ISDIR(u.bi_mode) && (u.bi_flags & BCH_INODE_unlinked)) {
1182 /* Check for this early so that check_unreachable_inode() will reattach it */
1183
1184 ret = bch2_empty_dir_snapshot(trans, k.k->p.offset, 0, k.k->p.snapshot);
1185 if (ret && ret != -BCH_ERR_ENOTEMPTY_dir_not_empty)
1186 goto err;
1187
1188 fsck_err_on(ret, trans, inode_dir_unlinked_but_not_empty,
1189 "dir unlinked but not empty\n%s",
1190 (printbuf_reset(&buf),
1191 bch2_inode_unpacked_to_text(&buf, &u),
1192 buf.buf));
1193 u.bi_flags &= ~BCH_INODE_unlinked;
1194 do_update = true;
1195 ret = 0;
1196 }
1197
1198 ret = bch2_inode_has_child_snapshots(trans, k.k->p);
1199 if (ret < 0)
1200 goto err;
1201
1202 if (fsck_err_on(ret != !!(u.bi_flags & BCH_INODE_has_child_snapshot),
1203 trans, inode_has_child_snapshots_wrong,
1204 "inode has_child_snapshots flag wrong (should be %u)\n%s",
1205 ret,
1206 (printbuf_reset(&buf),
1207 bch2_inode_unpacked_to_text(&buf, &u),
1208 buf.buf))) {
1209 if (ret)
1210 u.bi_flags |= BCH_INODE_has_child_snapshot;
1211 else
1212 u.bi_flags &= ~BCH_INODE_has_child_snapshot;
1213 do_update = true;
1214 }
1215 ret = 0;
1216
1217 if ((u.bi_flags & BCH_INODE_unlinked) &&
1218 !(u.bi_flags & BCH_INODE_has_child_snapshot)) {
1219 if (!test_bit(BCH_FS_started, &c->flags)) {
1220 /*
1221 * If we're not in online fsck, don't delete unlinked
1222 * inodes, just make sure they're on the deleted list.
1223 *
1224 * They might be referred to by a logged operation -
1225 * i.e. we might have crashed in the middle of a
1226 * truncate on an unlinked but open file - so we want to
1227 * let the delete_dead_inodes kill it after resuming
1228 * logged ops.
1229 */
1230 ret = check_inode_deleted_list(trans, k.k->p);
1231 if (ret < 0)
1232 goto err_noprint;
1233
1234 fsck_err_on(!ret,
1235 trans, unlinked_inode_not_on_deleted_list,
1236 "inode %llu:%u unlinked, but not on deleted list",
1237 u.bi_inum, k.k->p.snapshot);
1238
1239 ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, k.k->p, 1);
1240 if (ret)
1241 goto err;
1242 } else {
1243 ret = bch2_inode_or_descendents_is_open(trans, k.k->p);
1244 if (ret < 0)
1245 goto err;
1246
1247 if (fsck_err_on(!ret,
1248 trans, inode_unlinked_and_not_open,
1249 "inode %llu:%u unlinked and not open",
1250 u.bi_inum, u.bi_snapshot)) {
1251 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
1252 bch_err_msg(c, ret, "in fsck deleting inode");
1253 goto err_noprint;
1254 }
1255 ret = 0;
1256 }
1257 }
1258
1259 if (fsck_err_on(u.bi_parent_subvol &&
1260 (u.bi_subvol == 0 ||
1261 u.bi_subvol == BCACHEFS_ROOT_SUBVOL),
1262 trans, inode_bi_parent_nonzero,
1263 "inode %llu:%u has subvol %u but nonzero parent subvol %u",
1264 u.bi_inum, k.k->p.snapshot, u.bi_subvol, u.bi_parent_subvol)) {
1265 u.bi_parent_subvol = 0;
1266 do_update = true;
1267 }
1268
1269 if (u.bi_subvol) {
1270 struct bch_subvolume s;
1271
1272 ret = bch2_subvolume_get(trans, u.bi_subvol, false, &s);
1273 if (ret && !bch2_err_matches(ret, ENOENT))
1274 goto err;
1275
1276 if (ret && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
1277 ret = reconstruct_subvol(trans, k.k->p.snapshot, u.bi_subvol, u.bi_inum);
1278 goto do_update;
1279 }
1280
1281 if (fsck_err_on(ret,
1282 trans, inode_bi_subvol_missing,
1283 "inode %llu:%u bi_subvol points to missing subvolume %u",
1284 u.bi_inum, k.k->p.snapshot, u.bi_subvol) ||
1285 fsck_err_on(le64_to_cpu(s.inode) != u.bi_inum ||
1286 !bch2_snapshot_is_ancestor(c, le32_to_cpu(s.snapshot),
1287 k.k->p.snapshot),
1288 trans, inode_bi_subvol_wrong,
1289 "inode %llu:%u points to subvol %u, but subvol points to %llu:%u",
1290 u.bi_inum, k.k->p.snapshot, u.bi_subvol,
1291 le64_to_cpu(s.inode),
1292 le32_to_cpu(s.snapshot))) {
1293 u.bi_subvol = 0;
1294 u.bi_parent_subvol = 0;
1295 do_update = true;
1296 }
1297 }
1298
1299 if (fsck_err_on(u.bi_journal_seq > journal_cur_seq(&c->journal),
1300 trans, inode_journal_seq_in_future,
1301 "inode journal seq in future (currently at %llu)\n%s",
1302 journal_cur_seq(&c->journal),
1303 (printbuf_reset(&buf),
1304 bch2_inode_unpacked_to_text(&buf, &u),
1305 buf.buf))) {
1306 u.bi_journal_seq = journal_cur_seq(&c->journal);
1307 do_update = true;
1308 }
1309 do_update:
1310 if (do_update) {
1311 ret = __bch2_fsck_write_inode(trans, &u);
1312 bch_err_msg(c, ret, "in fsck updating inode");
1313 if (ret)
1314 goto err_noprint;
1315 }
1316 err:
1317 fsck_err:
1318 bch_err_fn(c, ret);
1319 err_noprint:
1320 printbuf_exit(&buf);
1321 return ret;
1322 }
1323
bch2_check_inodes(struct bch_fs * c)1324 int bch2_check_inodes(struct bch_fs *c)
1325 {
1326 struct bch_inode_unpacked snapshot_root = {};
1327 struct snapshots_seen s;
1328
1329 snapshots_seen_init(&s);
1330
1331 int ret = bch2_trans_run(c,
1332 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1333 POS_MIN,
1334 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1335 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1336 check_inode(trans, &iter, k, &snapshot_root, &s)));
1337
1338 snapshots_seen_exit(&s);
1339 bch_err_fn(c, ret);
1340 return ret;
1341 }
1342
find_oldest_inode_needs_reattach(struct btree_trans * trans,struct bch_inode_unpacked * inode)1343 static int find_oldest_inode_needs_reattach(struct btree_trans *trans,
1344 struct bch_inode_unpacked *inode)
1345 {
1346 struct bch_fs *c = trans->c;
1347 struct btree_iter iter;
1348 struct bkey_s_c k;
1349 int ret = 0;
1350
1351 /*
1352 * We look for inodes to reattach in natural key order, leaves first,
1353 * but we should do the reattach at the oldest version that needs to be
1354 * reattached:
1355 */
1356 for_each_btree_key_norestart(trans, iter,
1357 BTREE_ID_inodes,
1358 SPOS(0, inode->bi_inum, inode->bi_snapshot + 1),
1359 BTREE_ITER_all_snapshots, k, ret) {
1360 if (k.k->p.offset != inode->bi_inum)
1361 break;
1362
1363 if (!bch2_snapshot_is_ancestor(c, inode->bi_snapshot, k.k->p.snapshot))
1364 continue;
1365
1366 if (!bkey_is_inode(k.k))
1367 break;
1368
1369 struct bch_inode_unpacked parent_inode;
1370 ret = bch2_inode_unpack(k, &parent_inode);
1371 if (ret)
1372 break;
1373
1374 if (!inode_should_reattach(&parent_inode))
1375 break;
1376
1377 *inode = parent_inode;
1378 }
1379 bch2_trans_iter_exit(trans, &iter);
1380
1381 return ret;
1382 }
1383
check_unreachable_inode(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k)1384 static int check_unreachable_inode(struct btree_trans *trans,
1385 struct btree_iter *iter,
1386 struct bkey_s_c k)
1387 {
1388 struct printbuf buf = PRINTBUF;
1389 int ret = 0;
1390
1391 if (!bkey_is_inode(k.k))
1392 return 0;
1393
1394 struct bch_inode_unpacked inode;
1395 ret = bch2_inode_unpack(k, &inode);
1396 if (ret)
1397 return ret;
1398
1399 if (!inode_should_reattach(&inode))
1400 return 0;
1401
1402 ret = find_oldest_inode_needs_reattach(trans, &inode);
1403 if (ret)
1404 return ret;
1405
1406 if (fsck_err(trans, inode_unreachable,
1407 "unreachable inode:\n%s",
1408 (bch2_inode_unpacked_to_text(&buf, &inode),
1409 buf.buf)))
1410 ret = reattach_inode(trans, &inode);
1411 fsck_err:
1412 printbuf_exit(&buf);
1413 return ret;
1414 }
1415
1416 /*
1417 * Reattach unreachable (but not unlinked) inodes
1418 *
1419 * Run after check_inodes() and check_dirents(), so we node that inode
1420 * backpointer fields point to valid dirents, and every inode that has a dirent
1421 * that points to it has its backpointer field set - so we're just looking for
1422 * non-unlinked inodes without backpointers:
1423 *
1424 * XXX: this is racy w.r.t. hardlink removal in online fsck
1425 */
bch2_check_unreachable_inodes(struct bch_fs * c)1426 int bch2_check_unreachable_inodes(struct bch_fs *c)
1427 {
1428 int ret = bch2_trans_run(c,
1429 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1430 POS_MIN,
1431 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1432 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1433 check_unreachable_inode(trans, &iter, k)));
1434 bch_err_fn(c, ret);
1435 return ret;
1436 }
1437
btree_matches_i_mode(enum btree_id btree,unsigned mode)1438 static inline bool btree_matches_i_mode(enum btree_id btree, unsigned mode)
1439 {
1440 switch (btree) {
1441 case BTREE_ID_extents:
1442 return S_ISREG(mode) || S_ISLNK(mode);
1443 case BTREE_ID_dirents:
1444 return S_ISDIR(mode);
1445 case BTREE_ID_xattrs:
1446 return true;
1447 default:
1448 BUG();
1449 }
1450 }
1451
check_key_has_inode(struct btree_trans * trans,struct btree_iter * iter,struct inode_walker * inode,struct inode_walker_entry * i,struct bkey_s_c k)1452 static int check_key_has_inode(struct btree_trans *trans,
1453 struct btree_iter *iter,
1454 struct inode_walker *inode,
1455 struct inode_walker_entry *i,
1456 struct bkey_s_c k)
1457 {
1458 struct bch_fs *c = trans->c;
1459 struct printbuf buf = PRINTBUF;
1460 int ret = PTR_ERR_OR_ZERO(i);
1461 if (ret)
1462 return ret;
1463
1464 if (k.k->type == KEY_TYPE_whiteout)
1465 goto out;
1466
1467 if (!i && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_inodes))) {
1468 ret = reconstruct_inode(trans, iter->btree_id, k.k->p.snapshot, k.k->p.inode) ?:
1469 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
1470 if (ret)
1471 goto err;
1472
1473 inode->last_pos.inode--;
1474 ret = -BCH_ERR_transaction_restart_nested;
1475 goto err;
1476 }
1477
1478 if (fsck_err_on(!i,
1479 trans, key_in_missing_inode,
1480 "key in missing inode:\n %s",
1481 (printbuf_reset(&buf),
1482 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1483 goto delete;
1484
1485 if (fsck_err_on(i && !btree_matches_i_mode(iter->btree_id, i->inode.bi_mode),
1486 trans, key_in_wrong_inode_type,
1487 "key for wrong inode mode %o:\n %s",
1488 i->inode.bi_mode,
1489 (printbuf_reset(&buf),
1490 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1491 goto delete;
1492 out:
1493 err:
1494 fsck_err:
1495 printbuf_exit(&buf);
1496 bch_err_fn(c, ret);
1497 return ret;
1498 delete:
1499 ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_internal_snapshot_node);
1500 goto out;
1501 }
1502
check_i_sectors_notnested(struct btree_trans * trans,struct inode_walker * w)1503 static int check_i_sectors_notnested(struct btree_trans *trans, struct inode_walker *w)
1504 {
1505 struct bch_fs *c = trans->c;
1506 int ret = 0;
1507 s64 count2;
1508
1509 darray_for_each(w->inodes, i) {
1510 if (i->inode.bi_sectors == i->count)
1511 continue;
1512
1513 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1514
1515 if (w->recalculate_sums)
1516 i->count = count2;
1517
1518 if (i->count != count2) {
1519 bch_err_ratelimited(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1520 w->last_pos.inode, i->snapshot, i->count, count2);
1521 i->count = count2;
1522 }
1523
1524 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1525 trans, inode_i_sectors_wrong,
1526 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1527 w->last_pos.inode, i->snapshot,
1528 i->inode.bi_sectors, i->count)) {
1529 i->inode.bi_sectors = i->count;
1530 ret = bch2_fsck_write_inode(trans, &i->inode);
1531 if (ret)
1532 break;
1533 }
1534 }
1535 fsck_err:
1536 bch_err_fn(c, ret);
1537 return ret;
1538 }
1539
check_i_sectors(struct btree_trans * trans,struct inode_walker * w)1540 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1541 {
1542 u32 restart_count = trans->restart_count;
1543 return check_i_sectors_notnested(trans, w) ?:
1544 trans_was_restarted(trans, restart_count);
1545 }
1546
1547 struct extent_end {
1548 u32 snapshot;
1549 u64 offset;
1550 struct snapshots_seen seen;
1551 };
1552
1553 struct extent_ends {
1554 struct bpos last_pos;
1555 DARRAY(struct extent_end) e;
1556 };
1557
extent_ends_reset(struct extent_ends * extent_ends)1558 static void extent_ends_reset(struct extent_ends *extent_ends)
1559 {
1560 darray_for_each(extent_ends->e, i)
1561 snapshots_seen_exit(&i->seen);
1562 extent_ends->e.nr = 0;
1563 }
1564
extent_ends_exit(struct extent_ends * extent_ends)1565 static void extent_ends_exit(struct extent_ends *extent_ends)
1566 {
1567 extent_ends_reset(extent_ends);
1568 darray_exit(&extent_ends->e);
1569 }
1570
extent_ends_init(struct extent_ends * extent_ends)1571 static void extent_ends_init(struct extent_ends *extent_ends)
1572 {
1573 memset(extent_ends, 0, sizeof(*extent_ends));
1574 }
1575
extent_ends_at(struct bch_fs * c,struct extent_ends * extent_ends,struct snapshots_seen * seen,struct bkey_s_c k)1576 static int extent_ends_at(struct bch_fs *c,
1577 struct extent_ends *extent_ends,
1578 struct snapshots_seen *seen,
1579 struct bkey_s_c k)
1580 {
1581 struct extent_end *i, n = (struct extent_end) {
1582 .offset = k.k->p.offset,
1583 .snapshot = k.k->p.snapshot,
1584 .seen = *seen,
1585 };
1586
1587 n.seen.ids.data = kmemdup(seen->ids.data,
1588 sizeof(seen->ids.data[0]) * seen->ids.size,
1589 GFP_KERNEL);
1590 if (!n.seen.ids.data)
1591 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1592
1593 __darray_for_each(extent_ends->e, i) {
1594 if (i->snapshot == k.k->p.snapshot) {
1595 snapshots_seen_exit(&i->seen);
1596 *i = n;
1597 return 0;
1598 }
1599
1600 if (i->snapshot >= k.k->p.snapshot)
1601 break;
1602 }
1603
1604 return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1605 }
1606
overlapping_extents_found(struct btree_trans * trans,enum btree_id btree,struct bpos pos1,struct snapshots_seen * pos1_seen,struct bkey pos2,bool * fixed,struct extent_end * extent_end)1607 static int overlapping_extents_found(struct btree_trans *trans,
1608 enum btree_id btree,
1609 struct bpos pos1, struct snapshots_seen *pos1_seen,
1610 struct bkey pos2,
1611 bool *fixed,
1612 struct extent_end *extent_end)
1613 {
1614 struct bch_fs *c = trans->c;
1615 struct printbuf buf = PRINTBUF;
1616 struct btree_iter iter1, iter2 = { NULL };
1617 struct bkey_s_c k1, k2;
1618 int ret;
1619
1620 BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1621
1622 bch2_trans_iter_init(trans, &iter1, btree, pos1,
1623 BTREE_ITER_all_snapshots|
1624 BTREE_ITER_not_extents);
1625 k1 = bch2_btree_iter_peek_max(&iter1, POS(pos1.inode, U64_MAX));
1626 ret = bkey_err(k1);
1627 if (ret)
1628 goto err;
1629
1630 prt_str(&buf, "\n ");
1631 bch2_bkey_val_to_text(&buf, c, k1);
1632
1633 if (!bpos_eq(pos1, k1.k->p)) {
1634 prt_str(&buf, "\n wanted\n ");
1635 bch2_bpos_to_text(&buf, pos1);
1636 prt_str(&buf, "\n ");
1637 bch2_bkey_to_text(&buf, &pos2);
1638
1639 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1640 __func__, buf.buf);
1641 ret = -BCH_ERR_internal_fsck_err;
1642 goto err;
1643 }
1644
1645 bch2_trans_copy_iter(&iter2, &iter1);
1646
1647 while (1) {
1648 bch2_btree_iter_advance(&iter2);
1649
1650 k2 = bch2_btree_iter_peek_max(&iter2, POS(pos1.inode, U64_MAX));
1651 ret = bkey_err(k2);
1652 if (ret)
1653 goto err;
1654
1655 if (bpos_ge(k2.k->p, pos2.p))
1656 break;
1657 }
1658
1659 prt_str(&buf, "\n ");
1660 bch2_bkey_val_to_text(&buf, c, k2);
1661
1662 if (bpos_gt(k2.k->p, pos2.p) ||
1663 pos2.size != k2.k->size) {
1664 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1665 __func__, buf.buf);
1666 ret = -BCH_ERR_internal_fsck_err;
1667 goto err;
1668 }
1669
1670 prt_printf(&buf, "\n overwriting %s extent",
1671 pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1672
1673 if (fsck_err(trans, extent_overlapping,
1674 "overlapping extents%s", buf.buf)) {
1675 struct btree_iter *old_iter = &iter1;
1676 struct disk_reservation res = { 0 };
1677
1678 if (pos1.snapshot < pos2.p.snapshot) {
1679 old_iter = &iter2;
1680 swap(k1, k2);
1681 }
1682
1683 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1684
1685 ret = bch2_trans_update_extent_overwrite(trans, old_iter,
1686 BTREE_UPDATE_internal_snapshot_node,
1687 k1, k2) ?:
1688 bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1689 bch2_disk_reservation_put(c, &res);
1690
1691 if (ret)
1692 goto err;
1693
1694 *fixed = true;
1695
1696 if (pos1.snapshot == pos2.p.snapshot) {
1697 /*
1698 * We overwrote the first extent, and did the overwrite
1699 * in the same snapshot:
1700 */
1701 extent_end->offset = bkey_start_offset(&pos2);
1702 } else if (pos1.snapshot > pos2.p.snapshot) {
1703 /*
1704 * We overwrote the first extent in pos2's snapshot:
1705 */
1706 ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1707 } else {
1708 /*
1709 * We overwrote the second extent - restart
1710 * check_extent() from the top:
1711 */
1712 ret = -BCH_ERR_transaction_restart_nested;
1713 }
1714 }
1715 fsck_err:
1716 err:
1717 bch2_trans_iter_exit(trans, &iter2);
1718 bch2_trans_iter_exit(trans, &iter1);
1719 printbuf_exit(&buf);
1720 return ret;
1721 }
1722
check_overlapping_extents(struct btree_trans * trans,struct snapshots_seen * seen,struct extent_ends * extent_ends,struct bkey_s_c k,struct btree_iter * iter,bool * fixed)1723 static int check_overlapping_extents(struct btree_trans *trans,
1724 struct snapshots_seen *seen,
1725 struct extent_ends *extent_ends,
1726 struct bkey_s_c k,
1727 struct btree_iter *iter,
1728 bool *fixed)
1729 {
1730 struct bch_fs *c = trans->c;
1731 int ret = 0;
1732
1733 /* transaction restart, running again */
1734 if (bpos_eq(extent_ends->last_pos, k.k->p))
1735 return 0;
1736
1737 if (extent_ends->last_pos.inode != k.k->p.inode)
1738 extent_ends_reset(extent_ends);
1739
1740 darray_for_each(extent_ends->e, i) {
1741 if (i->offset <= bkey_start_offset(k.k))
1742 continue;
1743
1744 if (!ref_visible2(c,
1745 k.k->p.snapshot, seen,
1746 i->snapshot, &i->seen))
1747 continue;
1748
1749 ret = overlapping_extents_found(trans, iter->btree_id,
1750 SPOS(iter->pos.inode,
1751 i->offset,
1752 i->snapshot),
1753 &i->seen,
1754 *k.k, fixed, i);
1755 if (ret)
1756 goto err;
1757 }
1758
1759 extent_ends->last_pos = k.k->p;
1760 err:
1761 return ret;
1762 }
1763
check_extent_overbig(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k)1764 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1765 struct bkey_s_c k)
1766 {
1767 struct bch_fs *c = trans->c;
1768 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1769 struct bch_extent_crc_unpacked crc;
1770 const union bch_extent_entry *i;
1771 unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1772
1773 bkey_for_each_crc(k.k, ptrs, crc, i)
1774 if (crc_is_encoded(crc) &&
1775 crc.uncompressed_size > encoded_extent_max_sectors) {
1776 struct printbuf buf = PRINTBUF;
1777
1778 bch2_bkey_val_to_text(&buf, c, k);
1779 bch_err(c, "overbig encoded extent, please report this:\n %s", buf.buf);
1780 printbuf_exit(&buf);
1781 }
1782
1783 return 0;
1784 }
1785
check_extent(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,struct inode_walker * inode,struct snapshots_seen * s,struct extent_ends * extent_ends,struct disk_reservation * res)1786 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1787 struct bkey_s_c k,
1788 struct inode_walker *inode,
1789 struct snapshots_seen *s,
1790 struct extent_ends *extent_ends,
1791 struct disk_reservation *res)
1792 {
1793 struct bch_fs *c = trans->c;
1794 struct printbuf buf = PRINTBUF;
1795 int ret = 0;
1796
1797 ret = bch2_check_key_has_snapshot(trans, iter, k);
1798 if (ret) {
1799 ret = ret < 0 ? ret : 0;
1800 goto out;
1801 }
1802
1803 if (inode->last_pos.inode != k.k->p.inode && inode->have_inodes) {
1804 ret = check_i_sectors(trans, inode);
1805 if (ret)
1806 goto err;
1807 }
1808
1809 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1810 if (ret)
1811 goto err;
1812
1813 struct inode_walker_entry *extent_i = walk_inode(trans, inode, k);
1814 ret = PTR_ERR_OR_ZERO(extent_i);
1815 if (ret)
1816 goto err;
1817
1818 ret = check_key_has_inode(trans, iter, inode, extent_i, k);
1819 if (ret)
1820 goto err;
1821
1822 if (k.k->type != KEY_TYPE_whiteout) {
1823 ret = check_overlapping_extents(trans, s, extent_ends, k, iter,
1824 &inode->recalculate_sums);
1825 if (ret)
1826 goto err;
1827
1828 /*
1829 * Check inodes in reverse order, from oldest snapshots to
1830 * newest, starting from the inode that matches this extent's
1831 * snapshot. If we didn't have one, iterate over all inodes:
1832 */
1833 for (struct inode_walker_entry *i = extent_i ?: &darray_last(inode->inodes);
1834 inode->inodes.data && i >= inode->inodes.data;
1835 --i) {
1836 if (i->snapshot > k.k->p.snapshot ||
1837 !key_visible_in_snapshot(c, s, i->snapshot, k.k->p.snapshot))
1838 continue;
1839
1840 if (fsck_err_on(k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1841 !bkey_extent_is_reservation(k),
1842 trans, extent_past_end_of_inode,
1843 "extent type past end of inode %llu:%u, i_size %llu\n %s",
1844 i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1845 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1846 struct btree_iter iter2;
1847
1848 bch2_trans_copy_iter(&iter2, iter);
1849 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1850 ret = bch2_btree_iter_traverse(&iter2) ?:
1851 bch2_btree_delete_at(trans, &iter2,
1852 BTREE_UPDATE_internal_snapshot_node);
1853 bch2_trans_iter_exit(trans, &iter2);
1854 if (ret)
1855 goto err;
1856
1857 iter->k.type = KEY_TYPE_whiteout;
1858 break;
1859 }
1860 }
1861 }
1862
1863 ret = bch2_trans_commit(trans, res, NULL, BCH_TRANS_COMMIT_no_enospc);
1864 if (ret)
1865 goto err;
1866
1867 if (bkey_extent_is_allocation(k.k)) {
1868 for (struct inode_walker_entry *i = extent_i ?: &darray_last(inode->inodes);
1869 inode->inodes.data && i >= inode->inodes.data;
1870 --i) {
1871 if (i->snapshot > k.k->p.snapshot ||
1872 !key_visible_in_snapshot(c, s, i->snapshot, k.k->p.snapshot))
1873 continue;
1874
1875 i->count += k.k->size;
1876 }
1877 }
1878
1879 if (k.k->type != KEY_TYPE_whiteout) {
1880 ret = extent_ends_at(c, extent_ends, s, k);
1881 if (ret)
1882 goto err;
1883 }
1884 out:
1885 err:
1886 fsck_err:
1887 printbuf_exit(&buf);
1888 bch_err_fn(c, ret);
1889 return ret;
1890 }
1891
1892 /*
1893 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1894 * that i_size an i_sectors are consistent
1895 */
bch2_check_extents(struct bch_fs * c)1896 int bch2_check_extents(struct bch_fs *c)
1897 {
1898 struct inode_walker w = inode_walker_init();
1899 struct snapshots_seen s;
1900 struct extent_ends extent_ends;
1901 struct disk_reservation res = { 0 };
1902
1903 snapshots_seen_init(&s);
1904 extent_ends_init(&extent_ends);
1905
1906 int ret = bch2_trans_run(c,
1907 for_each_btree_key(trans, iter, BTREE_ID_extents,
1908 POS(BCACHEFS_ROOT_INO, 0),
1909 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, ({
1910 bch2_disk_reservation_put(c, &res);
1911 check_extent(trans, &iter, k, &w, &s, &extent_ends, &res) ?:
1912 check_extent_overbig(trans, &iter, k);
1913 })) ?:
1914 check_i_sectors_notnested(trans, &w));
1915
1916 bch2_disk_reservation_put(c, &res);
1917 extent_ends_exit(&extent_ends);
1918 inode_walker_exit(&w);
1919 snapshots_seen_exit(&s);
1920
1921 bch_err_fn(c, ret);
1922 return ret;
1923 }
1924
bch2_check_indirect_extents(struct bch_fs * c)1925 int bch2_check_indirect_extents(struct bch_fs *c)
1926 {
1927 struct disk_reservation res = { 0 };
1928
1929 int ret = bch2_trans_run(c,
1930 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1931 POS_MIN,
1932 BTREE_ITER_prefetch, k,
1933 &res, NULL,
1934 BCH_TRANS_COMMIT_no_enospc, ({
1935 bch2_disk_reservation_put(c, &res);
1936 check_extent_overbig(trans, &iter, k);
1937 })));
1938
1939 bch2_disk_reservation_put(c, &res);
1940 bch_err_fn(c, ret);
1941 return ret;
1942 }
1943
check_subdir_count_notnested(struct btree_trans * trans,struct inode_walker * w)1944 static int check_subdir_count_notnested(struct btree_trans *trans, struct inode_walker *w)
1945 {
1946 struct bch_fs *c = trans->c;
1947 int ret = 0;
1948 s64 count2;
1949
1950 darray_for_each(w->inodes, i) {
1951 if (i->inode.bi_nlink == i->count)
1952 continue;
1953
1954 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1955 if (count2 < 0)
1956 return count2;
1957
1958 if (i->count != count2) {
1959 bch_err_ratelimited(c, "fsck counted subdirectories wrong for inum %llu:%u: got %llu should be %llu",
1960 w->last_pos.inode, i->snapshot, i->count, count2);
1961 i->count = count2;
1962 if (i->inode.bi_nlink == i->count)
1963 continue;
1964 }
1965
1966 if (fsck_err_on(i->inode.bi_nlink != i->count,
1967 trans, inode_dir_wrong_nlink,
1968 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1969 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1970 i->inode.bi_nlink = i->count;
1971 ret = bch2_fsck_write_inode(trans, &i->inode);
1972 if (ret)
1973 break;
1974 }
1975 }
1976 fsck_err:
1977 bch_err_fn(c, ret);
1978 return ret;
1979 }
1980
check_subdir_dirents_count(struct btree_trans * trans,struct inode_walker * w)1981 static int check_subdir_dirents_count(struct btree_trans *trans, struct inode_walker *w)
1982 {
1983 u32 restart_count = trans->restart_count;
1984 return check_subdir_count_notnested(trans, w) ?:
1985 trans_was_restarted(trans, restart_count);
1986 }
1987
1988 noinline_for_stack
check_dirent_inode_dirent(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c_dirent d,struct bch_inode_unpacked * target)1989 static int check_dirent_inode_dirent(struct btree_trans *trans,
1990 struct btree_iter *iter,
1991 struct bkey_s_c_dirent d,
1992 struct bch_inode_unpacked *target)
1993 {
1994 struct bch_fs *c = trans->c;
1995 struct printbuf buf = PRINTBUF;
1996 struct btree_iter bp_iter = { NULL };
1997 int ret = 0;
1998
1999 if (inode_points_to_dirent(target, d))
2000 return 0;
2001
2002 if (!target->bi_dir &&
2003 !target->bi_dir_offset) {
2004 fsck_err_on(S_ISDIR(target->bi_mode),
2005 trans, inode_dir_missing_backpointer,
2006 "directory with missing backpointer\n%s",
2007 (printbuf_reset(&buf),
2008 bch2_bkey_val_to_text(&buf, c, d.s_c),
2009 prt_printf(&buf, "\n"),
2010 bch2_inode_unpacked_to_text(&buf, target),
2011 buf.buf));
2012
2013 fsck_err_on(target->bi_flags & BCH_INODE_unlinked,
2014 trans, inode_unlinked_but_has_dirent,
2015 "inode unlinked but has dirent\n%s",
2016 (printbuf_reset(&buf),
2017 bch2_bkey_val_to_text(&buf, c, d.s_c),
2018 prt_printf(&buf, "\n"),
2019 bch2_inode_unpacked_to_text(&buf, target),
2020 buf.buf));
2021
2022 target->bi_flags &= ~BCH_INODE_unlinked;
2023 target->bi_dir = d.k->p.inode;
2024 target->bi_dir_offset = d.k->p.offset;
2025 return __bch2_fsck_write_inode(trans, target);
2026 }
2027
2028 if (bch2_inode_should_have_single_bp(target) &&
2029 !fsck_err(trans, inode_wrong_backpointer,
2030 "dirent points to inode that does not point back:\n %s",
2031 (bch2_bkey_val_to_text(&buf, c, d.s_c),
2032 prt_printf(&buf, "\n "),
2033 bch2_inode_unpacked_to_text(&buf, target),
2034 buf.buf)))
2035 goto err;
2036
2037 struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
2038 SPOS(target->bi_dir, target->bi_dir_offset, target->bi_snapshot));
2039 ret = bkey_err(bp_dirent);
2040 if (ret && !bch2_err_matches(ret, ENOENT))
2041 goto err;
2042
2043 bool backpointer_exists = !ret;
2044 ret = 0;
2045
2046 if (fsck_err_on(!backpointer_exists,
2047 trans, inode_wrong_backpointer,
2048 "inode %llu:%u has wrong backpointer:\n"
2049 "got %llu:%llu\n"
2050 "should be %llu:%llu",
2051 target->bi_inum, target->bi_snapshot,
2052 target->bi_dir,
2053 target->bi_dir_offset,
2054 d.k->p.inode,
2055 d.k->p.offset)) {
2056 target->bi_dir = d.k->p.inode;
2057 target->bi_dir_offset = d.k->p.offset;
2058 ret = __bch2_fsck_write_inode(trans, target);
2059 goto out;
2060 }
2061
2062 bch2_bkey_val_to_text(&buf, c, d.s_c);
2063 prt_newline(&buf);
2064 if (backpointer_exists)
2065 bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
2066
2067 if (fsck_err_on(backpointer_exists &&
2068 (S_ISDIR(target->bi_mode) ||
2069 target->bi_subvol),
2070 trans, inode_dir_multiple_links,
2071 "%s %llu:%u with multiple links\n%s",
2072 S_ISDIR(target->bi_mode) ? "directory" : "subvolume",
2073 target->bi_inum, target->bi_snapshot, buf.buf)) {
2074 ret = __remove_dirent(trans, d.k->p);
2075 goto out;
2076 }
2077
2078 /*
2079 * hardlinked file with nlink 0:
2080 * We're just adjusting nlink here so check_nlinks() will pick
2081 * it up, it ignores inodes with nlink 0
2082 */
2083 if (fsck_err_on(backpointer_exists && !target->bi_nlink,
2084 trans, inode_multiple_links_but_nlink_0,
2085 "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
2086 target->bi_inum, target->bi_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
2087 target->bi_nlink++;
2088 target->bi_flags &= ~BCH_INODE_unlinked;
2089 ret = __bch2_fsck_write_inode(trans, target);
2090 if (ret)
2091 goto err;
2092 }
2093 out:
2094 err:
2095 fsck_err:
2096 bch2_trans_iter_exit(trans, &bp_iter);
2097 printbuf_exit(&buf);
2098 bch_err_fn(c, ret);
2099 return ret;
2100 }
2101
2102 noinline_for_stack
check_dirent_target(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c_dirent d,struct bch_inode_unpacked * target)2103 static int check_dirent_target(struct btree_trans *trans,
2104 struct btree_iter *iter,
2105 struct bkey_s_c_dirent d,
2106 struct bch_inode_unpacked *target)
2107 {
2108 struct bch_fs *c = trans->c;
2109 struct bkey_i_dirent *n;
2110 struct printbuf buf = PRINTBUF;
2111 int ret = 0;
2112
2113 ret = check_dirent_inode_dirent(trans, iter, d, target);
2114 if (ret)
2115 goto err;
2116
2117 if (fsck_err_on(d.v->d_type != inode_d_type(target),
2118 trans, dirent_d_type_wrong,
2119 "incorrect d_type: got %s, should be %s:\n%s",
2120 bch2_d_type_str(d.v->d_type),
2121 bch2_d_type_str(inode_d_type(target)),
2122 (printbuf_reset(&buf),
2123 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
2124 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
2125 ret = PTR_ERR_OR_ZERO(n);
2126 if (ret)
2127 goto err;
2128
2129 bkey_reassemble(&n->k_i, d.s_c);
2130 n->v.d_type = inode_d_type(target);
2131 if (n->v.d_type == DT_SUBVOL) {
2132 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
2133 n->v.d_child_subvol = cpu_to_le32(target->bi_subvol);
2134 } else {
2135 n->v.d_inum = cpu_to_le64(target->bi_inum);
2136 }
2137
2138 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
2139 if (ret)
2140 goto err;
2141
2142 d = dirent_i_to_s_c(n);
2143 }
2144 err:
2145 fsck_err:
2146 printbuf_exit(&buf);
2147 bch_err_fn(c, ret);
2148 return ret;
2149 }
2150
2151 /* find a subvolume that's a descendent of @snapshot: */
find_snapshot_subvol(struct btree_trans * trans,u32 snapshot,u32 * subvolid)2152 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
2153 {
2154 struct btree_iter iter;
2155 struct bkey_s_c k;
2156 int ret;
2157
2158 for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN, 0, k, ret) {
2159 if (k.k->type != KEY_TYPE_subvolume)
2160 continue;
2161
2162 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2163 if (bch2_snapshot_is_ancestor(trans->c, le32_to_cpu(s.v->snapshot), snapshot)) {
2164 bch2_trans_iter_exit(trans, &iter);
2165 *subvolid = k.k->p.offset;
2166 goto found;
2167 }
2168 }
2169 if (!ret)
2170 ret = -ENOENT;
2171 found:
2172 bch2_trans_iter_exit(trans, &iter);
2173 return ret;
2174 }
2175
2176 noinline_for_stack
check_dirent_to_subvol(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c_dirent d)2177 static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *iter,
2178 struct bkey_s_c_dirent d)
2179 {
2180 struct bch_fs *c = trans->c;
2181 struct btree_iter subvol_iter = {};
2182 struct bch_inode_unpacked subvol_root;
2183 u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
2184 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
2185 u32 parent_snapshot;
2186 u32 new_parent_subvol = 0;
2187 u64 parent_inum;
2188 struct printbuf buf = PRINTBUF;
2189 int ret = 0;
2190
2191 ret = subvol_lookup(trans, parent_subvol, &parent_snapshot, &parent_inum);
2192 if (ret && !bch2_err_matches(ret, ENOENT))
2193 return ret;
2194
2195 if (ret ||
2196 (!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot))) {
2197 int ret2 = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
2198 if (ret2 && !bch2_err_matches(ret, ENOENT))
2199 return ret2;
2200 }
2201
2202 if (ret &&
2203 !new_parent_subvol &&
2204 (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
2205 /*
2206 * Couldn't find a subvol for dirent's snapshot - but we lost
2207 * subvols, so we need to reconstruct:
2208 */
2209 ret = reconstruct_subvol(trans, d.k->p.snapshot, parent_subvol, 0);
2210 if (ret)
2211 return ret;
2212
2213 parent_snapshot = d.k->p.snapshot;
2214 }
2215
2216 if (fsck_err_on(ret,
2217 trans, dirent_to_missing_parent_subvol,
2218 "dirent parent_subvol points to missing subvolume\n%s",
2219 (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
2220 fsck_err_on(!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot),
2221 trans, dirent_not_visible_in_parent_subvol,
2222 "dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
2223 parent_snapshot,
2224 (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
2225 if (!new_parent_subvol) {
2226 bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
2227 return -BCH_ERR_fsck_repair_unimplemented;
2228 }
2229
2230 struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
2231 ret = PTR_ERR_OR_ZERO(new_dirent);
2232 if (ret)
2233 goto err;
2234
2235 new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
2236 }
2237
2238 struct bkey_s_c_subvolume s =
2239 bch2_bkey_get_iter_typed(trans, &subvol_iter,
2240 BTREE_ID_subvolumes, POS(0, target_subvol),
2241 0, subvolume);
2242 ret = bkey_err(s.s_c);
2243 if (ret && !bch2_err_matches(ret, ENOENT))
2244 return ret;
2245
2246 if (ret) {
2247 if (fsck_err(trans, dirent_to_missing_subvol,
2248 "dirent points to missing subvolume\n%s",
2249 (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
2250 return __remove_dirent(trans, d.k->p);
2251 ret = 0;
2252 goto out;
2253 }
2254
2255 if (fsck_err_on(le32_to_cpu(s.v->fs_path_parent) != parent_subvol,
2256 trans, subvol_fs_path_parent_wrong,
2257 "subvol with wrong fs_path_parent, should be be %u\n%s",
2258 parent_subvol,
2259 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
2260 struct bkey_i_subvolume *n =
2261 bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
2262 ret = PTR_ERR_OR_ZERO(n);
2263 if (ret)
2264 goto err;
2265
2266 n->v.fs_path_parent = cpu_to_le32(parent_subvol);
2267 }
2268
2269 u64 target_inum = le64_to_cpu(s.v->inode);
2270 u32 target_snapshot = le32_to_cpu(s.v->snapshot);
2271
2272 ret = lookup_inode(trans, target_inum, target_snapshot, &subvol_root);
2273 if (ret && !bch2_err_matches(ret, ENOENT))
2274 goto err;
2275
2276 if (ret) {
2277 bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
2278 ret = -BCH_ERR_fsck_repair_unimplemented;
2279 goto err;
2280 }
2281
2282 if (fsck_err_on(!ret && parent_subvol != subvol_root.bi_parent_subvol,
2283 trans, inode_bi_parent_wrong,
2284 "subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
2285 target_inum,
2286 subvol_root.bi_parent_subvol, parent_subvol)) {
2287 subvol_root.bi_parent_subvol = parent_subvol;
2288 subvol_root.bi_snapshot = le32_to_cpu(s.v->snapshot);
2289 ret = __bch2_fsck_write_inode(trans, &subvol_root);
2290 if (ret)
2291 goto err;
2292 }
2293
2294 ret = check_dirent_target(trans, iter, d, &subvol_root);
2295 if (ret)
2296 goto err;
2297 out:
2298 err:
2299 fsck_err:
2300 bch2_trans_iter_exit(trans, &subvol_iter);
2301 printbuf_exit(&buf);
2302 return ret;
2303 }
2304
check_dirent(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,struct bch_hash_info * hash_info,struct inode_walker * dir,struct inode_walker * target,struct snapshots_seen * s)2305 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
2306 struct bkey_s_c k,
2307 struct bch_hash_info *hash_info,
2308 struct inode_walker *dir,
2309 struct inode_walker *target,
2310 struct snapshots_seen *s)
2311 {
2312 struct bch_fs *c = trans->c;
2313 struct inode_walker_entry *i;
2314 struct printbuf buf = PRINTBUF;
2315 int ret = 0;
2316
2317 ret = bch2_check_key_has_snapshot(trans, iter, k);
2318 if (ret) {
2319 ret = ret < 0 ? ret : 0;
2320 goto out;
2321 }
2322
2323 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
2324 if (ret)
2325 goto err;
2326
2327 if (k.k->type == KEY_TYPE_whiteout)
2328 goto out;
2329
2330 if (dir->last_pos.inode != k.k->p.inode && dir->have_inodes) {
2331 ret = check_subdir_dirents_count(trans, dir);
2332 if (ret)
2333 goto err;
2334 }
2335
2336 i = walk_inode(trans, dir, k);
2337 ret = PTR_ERR_OR_ZERO(i);
2338 if (ret < 0)
2339 goto err;
2340
2341 ret = check_key_has_inode(trans, iter, dir, i, k);
2342 if (ret)
2343 goto err;
2344
2345 if (!i)
2346 goto out;
2347
2348 if (dir->first_this_inode)
2349 *hash_info = bch2_hash_info_init(c, &i->inode);
2350 dir->first_this_inode = false;
2351
2352 ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info, iter, k);
2353 if (ret < 0)
2354 goto err;
2355 if (ret) {
2356 /* dirent has been deleted */
2357 ret = 0;
2358 goto out;
2359 }
2360
2361 if (k.k->type != KEY_TYPE_dirent)
2362 goto out;
2363
2364 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2365
2366 if (d.v->d_type == DT_SUBVOL) {
2367 ret = check_dirent_to_subvol(trans, iter, d);
2368 if (ret)
2369 goto err;
2370 } else {
2371 ret = get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
2372 if (ret)
2373 goto err;
2374
2375 if (fsck_err_on(!target->inodes.nr,
2376 trans, dirent_to_missing_inode,
2377 "dirent points to missing inode:\n%s",
2378 (printbuf_reset(&buf),
2379 bch2_bkey_val_to_text(&buf, c, k),
2380 buf.buf))) {
2381 ret = __remove_dirent(trans, d.k->p);
2382 if (ret)
2383 goto err;
2384 }
2385
2386 darray_for_each(target->inodes, i) {
2387 ret = check_dirent_target(trans, iter, d, &i->inode);
2388 if (ret)
2389 goto err;
2390 }
2391
2392 darray_for_each(target->deletes, i)
2393 if (fsck_err_on(!snapshot_list_has_id(&s->ids, *i),
2394 trans, dirent_to_overwritten_inode,
2395 "dirent points to inode overwritten in snapshot %u:\n%s",
2396 *i,
2397 (printbuf_reset(&buf),
2398 bch2_bkey_val_to_text(&buf, c, k),
2399 buf.buf))) {
2400 struct btree_iter delete_iter;
2401 bch2_trans_iter_init(trans, &delete_iter,
2402 BTREE_ID_dirents,
2403 SPOS(k.k->p.inode, k.k->p.offset, *i),
2404 BTREE_ITER_intent);
2405 ret = bch2_btree_iter_traverse(&delete_iter) ?:
2406 bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
2407 hash_info,
2408 &delete_iter,
2409 BTREE_UPDATE_internal_snapshot_node);
2410 bch2_trans_iter_exit(trans, &delete_iter);
2411 if (ret)
2412 goto err;
2413
2414 }
2415 }
2416
2417 ret = bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
2418 if (ret)
2419 goto err;
2420
2421 for_each_visible_inode(c, s, dir, d.k->p.snapshot, i) {
2422 if (d.v->d_type == DT_DIR)
2423 i->count++;
2424 i->i_size += bkey_bytes(d.k);
2425 }
2426 out:
2427 err:
2428 fsck_err:
2429 printbuf_exit(&buf);
2430 bch_err_fn(c, ret);
2431 return ret;
2432 }
2433
2434 /*
2435 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
2436 * validate d_type
2437 */
bch2_check_dirents(struct bch_fs * c)2438 int bch2_check_dirents(struct bch_fs *c)
2439 {
2440 struct inode_walker dir = inode_walker_init();
2441 struct inode_walker target = inode_walker_init();
2442 struct snapshots_seen s;
2443 struct bch_hash_info hash_info;
2444
2445 snapshots_seen_init(&s);
2446
2447 int ret = bch2_trans_run(c,
2448 for_each_btree_key(trans, iter, BTREE_ID_dirents,
2449 POS(BCACHEFS_ROOT_INO, 0),
2450 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
2451 check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)) ?:
2452 check_subdir_count_notnested(trans, &dir));
2453
2454 snapshots_seen_exit(&s);
2455 inode_walker_exit(&dir);
2456 inode_walker_exit(&target);
2457 bch_err_fn(c, ret);
2458 return ret;
2459 }
2460
check_xattr(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,struct bch_hash_info * hash_info,struct inode_walker * inode)2461 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
2462 struct bkey_s_c k,
2463 struct bch_hash_info *hash_info,
2464 struct inode_walker *inode)
2465 {
2466 struct bch_fs *c = trans->c;
2467 struct inode_walker_entry *i;
2468 int ret;
2469
2470 ret = bch2_check_key_has_snapshot(trans, iter, k);
2471 if (ret < 0)
2472 return ret;
2473 if (ret)
2474 return 0;
2475
2476 i = walk_inode(trans, inode, k);
2477 ret = PTR_ERR_OR_ZERO(i);
2478 if (ret)
2479 return ret;
2480
2481 ret = check_key_has_inode(trans, iter, inode, i, k);
2482 if (ret)
2483 return ret;
2484
2485 if (!i)
2486 return 0;
2487
2488 if (inode->first_this_inode)
2489 *hash_info = bch2_hash_info_init(c, &i->inode);
2490 inode->first_this_inode = false;
2491
2492 ret = bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info, iter, k);
2493 bch_err_fn(c, ret);
2494 return ret;
2495 }
2496
2497 /*
2498 * Walk xattrs: verify that they all have a corresponding inode
2499 */
bch2_check_xattrs(struct bch_fs * c)2500 int bch2_check_xattrs(struct bch_fs *c)
2501 {
2502 struct inode_walker inode = inode_walker_init();
2503 struct bch_hash_info hash_info;
2504 int ret = 0;
2505
2506 ret = bch2_trans_run(c,
2507 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
2508 POS(BCACHEFS_ROOT_INO, 0),
2509 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots,
2510 k,
2511 NULL, NULL,
2512 BCH_TRANS_COMMIT_no_enospc,
2513 check_xattr(trans, &iter, k, &hash_info, &inode)));
2514
2515 inode_walker_exit(&inode);
2516 bch_err_fn(c, ret);
2517 return ret;
2518 }
2519
check_root_trans(struct btree_trans * trans)2520 static int check_root_trans(struct btree_trans *trans)
2521 {
2522 struct bch_fs *c = trans->c;
2523 struct bch_inode_unpacked root_inode;
2524 u32 snapshot;
2525 u64 inum;
2526 int ret;
2527
2528 ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
2529 if (ret && !bch2_err_matches(ret, ENOENT))
2530 return ret;
2531
2532 if (mustfix_fsck_err_on(ret, trans, root_subvol_missing,
2533 "root subvol missing")) {
2534 struct bkey_i_subvolume *root_subvol =
2535 bch2_trans_kmalloc(trans, sizeof(*root_subvol));
2536 ret = PTR_ERR_OR_ZERO(root_subvol);
2537 if (ret)
2538 goto err;
2539
2540 snapshot = U32_MAX;
2541 inum = BCACHEFS_ROOT_INO;
2542
2543 bkey_subvolume_init(&root_subvol->k_i);
2544 root_subvol->k.p.offset = BCACHEFS_ROOT_SUBVOL;
2545 root_subvol->v.flags = 0;
2546 root_subvol->v.snapshot = cpu_to_le32(snapshot);
2547 root_subvol->v.inode = cpu_to_le64(inum);
2548 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol->k_i, 0);
2549 bch_err_msg(c, ret, "writing root subvol");
2550 if (ret)
2551 goto err;
2552 }
2553
2554 ret = lookup_inode(trans, BCACHEFS_ROOT_INO, snapshot, &root_inode);
2555 if (ret && !bch2_err_matches(ret, ENOENT))
2556 return ret;
2557
2558 if (mustfix_fsck_err_on(ret,
2559 trans, root_dir_missing,
2560 "root directory missing") ||
2561 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
2562 trans, root_inode_not_dir,
2563 "root inode not a directory")) {
2564 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
2565 0, NULL);
2566 root_inode.bi_inum = inum;
2567 root_inode.bi_snapshot = snapshot;
2568
2569 ret = __bch2_fsck_write_inode(trans, &root_inode);
2570 bch_err_msg(c, ret, "writing root inode");
2571 }
2572 err:
2573 fsck_err:
2574 return ret;
2575 }
2576
2577 /* Get root directory, create if it doesn't exist: */
bch2_check_root(struct bch_fs * c)2578 int bch2_check_root(struct bch_fs *c)
2579 {
2580 int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2581 check_root_trans(trans));
2582 bch_err_fn(c, ret);
2583 return ret;
2584 }
2585
2586 typedef DARRAY(u32) darray_u32;
2587
darray_u32_has(darray_u32 * d,u32 v)2588 static bool darray_u32_has(darray_u32 *d, u32 v)
2589 {
2590 darray_for_each(*d, i)
2591 if (*i == v)
2592 return true;
2593 return false;
2594 }
2595
check_subvol_path(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k)2596 static int check_subvol_path(struct btree_trans *trans, struct btree_iter *iter, struct bkey_s_c k)
2597 {
2598 struct bch_fs *c = trans->c;
2599 struct btree_iter parent_iter = {};
2600 darray_u32 subvol_path = {};
2601 struct printbuf buf = PRINTBUF;
2602 int ret = 0;
2603
2604 if (k.k->type != KEY_TYPE_subvolume)
2605 return 0;
2606
2607 while (k.k->p.offset != BCACHEFS_ROOT_SUBVOL) {
2608 ret = darray_push(&subvol_path, k.k->p.offset);
2609 if (ret)
2610 goto err;
2611
2612 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2613
2614 struct bch_inode_unpacked subvol_root;
2615 ret = bch2_inode_find_by_inum_trans(trans,
2616 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
2617 &subvol_root);
2618 if (ret)
2619 break;
2620
2621 u32 parent = le32_to_cpu(s.v->fs_path_parent);
2622
2623 if (darray_u32_has(&subvol_path, parent)) {
2624 if (fsck_err(c, subvol_loop, "subvolume loop"))
2625 ret = reattach_subvol(trans, s);
2626 break;
2627 }
2628
2629 bch2_trans_iter_exit(trans, &parent_iter);
2630 bch2_trans_iter_init(trans, &parent_iter,
2631 BTREE_ID_subvolumes, POS(0, parent), 0);
2632 k = bch2_btree_iter_peek_slot(&parent_iter);
2633 ret = bkey_err(k);
2634 if (ret)
2635 goto err;
2636
2637 if (fsck_err_on(k.k->type != KEY_TYPE_subvolume,
2638 trans, subvol_unreachable,
2639 "unreachable subvolume %s",
2640 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2641 buf.buf))) {
2642 ret = reattach_subvol(trans, s);
2643 break;
2644 }
2645 }
2646 fsck_err:
2647 err:
2648 printbuf_exit(&buf);
2649 darray_exit(&subvol_path);
2650 bch2_trans_iter_exit(trans, &parent_iter);
2651 return ret;
2652 }
2653
bch2_check_subvolume_structure(struct bch_fs * c)2654 int bch2_check_subvolume_structure(struct bch_fs *c)
2655 {
2656 int ret = bch2_trans_run(c,
2657 for_each_btree_key_commit(trans, iter,
2658 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
2659 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2660 check_subvol_path(trans, &iter, k)));
2661 bch_err_fn(c, ret);
2662 return ret;
2663 }
2664
2665 struct pathbuf_entry {
2666 u64 inum;
2667 u32 snapshot;
2668 };
2669
2670 typedef DARRAY(struct pathbuf_entry) pathbuf;
2671
bch2_bi_depth_renumber_one(struct btree_trans * trans,struct pathbuf_entry * p,u32 new_depth)2672 static int bch2_bi_depth_renumber_one(struct btree_trans *trans, struct pathbuf_entry *p,
2673 u32 new_depth)
2674 {
2675 struct btree_iter iter;
2676 struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
2677 SPOS(0, p->inum, p->snapshot), 0);
2678
2679 struct bch_inode_unpacked inode;
2680 int ret = bkey_err(k) ?:
2681 !bkey_is_inode(k.k) ? -BCH_ERR_ENOENT_inode
2682 : bch2_inode_unpack(k, &inode);
2683 if (ret)
2684 goto err;
2685
2686 if (inode.bi_depth != new_depth) {
2687 inode.bi_depth = new_depth;
2688 ret = __bch2_fsck_write_inode(trans, &inode) ?:
2689 bch2_trans_commit(trans, NULL, NULL, 0);
2690 }
2691 err:
2692 bch2_trans_iter_exit(trans, &iter);
2693 return ret;
2694 }
2695
bch2_bi_depth_renumber(struct btree_trans * trans,pathbuf * path,u32 new_bi_depth)2696 static int bch2_bi_depth_renumber(struct btree_trans *trans, pathbuf *path, u32 new_bi_depth)
2697 {
2698 u32 restart_count = trans->restart_count;
2699 int ret = 0;
2700
2701 darray_for_each_reverse(*path, i) {
2702 ret = nested_lockrestart_do(trans,
2703 bch2_bi_depth_renumber_one(trans, i, new_bi_depth));
2704 bch_err_fn(trans->c, ret);
2705 if (ret)
2706 break;
2707
2708 new_bi_depth++;
2709 }
2710
2711 return ret ?: trans_was_restarted(trans, restart_count);
2712 }
2713
path_is_dup(pathbuf * p,u64 inum,u32 snapshot)2714 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
2715 {
2716 darray_for_each(*p, i)
2717 if (i->inum == inum &&
2718 i->snapshot == snapshot)
2719 return true;
2720 return false;
2721 }
2722
check_path_loop(struct btree_trans * trans,struct bkey_s_c inode_k)2723 static int check_path_loop(struct btree_trans *trans, struct bkey_s_c inode_k)
2724 {
2725 struct bch_fs *c = trans->c;
2726 struct btree_iter inode_iter = {};
2727 pathbuf path = {};
2728 struct printbuf buf = PRINTBUF;
2729 u32 snapshot = inode_k.k->p.snapshot;
2730 bool redo_bi_depth = false;
2731 u32 min_bi_depth = U32_MAX;
2732 int ret = 0;
2733
2734 struct bch_inode_unpacked inode;
2735 ret = bch2_inode_unpack(inode_k, &inode);
2736 if (ret)
2737 return ret;
2738
2739 while (!inode.bi_subvol) {
2740 struct btree_iter dirent_iter;
2741 struct bkey_s_c_dirent d;
2742 u32 parent_snapshot = snapshot;
2743
2744 d = inode_get_dirent(trans, &dirent_iter, &inode, &parent_snapshot);
2745 ret = bkey_err(d.s_c);
2746 if (ret && !bch2_err_matches(ret, ENOENT))
2747 goto out;
2748
2749 if (!ret && (ret = dirent_points_to_inode(c, d, &inode)))
2750 bch2_trans_iter_exit(trans, &dirent_iter);
2751
2752 if (bch2_err_matches(ret, ENOENT)) {
2753 printbuf_reset(&buf);
2754 bch2_bkey_val_to_text(&buf, c, inode_k);
2755 bch_err(c, "unreachable inode in check_directory_structure: %s\n%s",
2756 bch2_err_str(ret), buf.buf);
2757 goto out;
2758 }
2759
2760 bch2_trans_iter_exit(trans, &dirent_iter);
2761
2762 ret = darray_push(&path, ((struct pathbuf_entry) {
2763 .inum = inode.bi_inum,
2764 .snapshot = snapshot,
2765 }));
2766 if (ret)
2767 return ret;
2768
2769 snapshot = parent_snapshot;
2770
2771 bch2_trans_iter_exit(trans, &inode_iter);
2772 inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
2773 SPOS(0, inode.bi_dir, snapshot), 0);
2774
2775 struct bch_inode_unpacked parent_inode;
2776 ret = bkey_err(inode_k) ?:
2777 !bkey_is_inode(inode_k.k) ? -BCH_ERR_ENOENT_inode
2778 : bch2_inode_unpack(inode_k, &parent_inode);
2779 if (ret) {
2780 /* Should have been caught in dirents pass */
2781 bch_err_msg(c, ret, "error looking up parent directory");
2782 goto out;
2783 }
2784
2785 min_bi_depth = parent_inode.bi_depth;
2786
2787 if (parent_inode.bi_depth < inode.bi_depth &&
2788 min_bi_depth < U16_MAX)
2789 break;
2790
2791 inode = parent_inode;
2792 snapshot = inode_k.k->p.snapshot;
2793 redo_bi_depth = true;
2794
2795 if (path_is_dup(&path, inode.bi_inum, snapshot)) {
2796 /* XXX print path */
2797 bch_err(c, "directory structure loop");
2798
2799 darray_for_each(path, i)
2800 pr_err("%llu:%u", i->inum, i->snapshot);
2801 pr_err("%llu:%u", inode.bi_inum, snapshot);
2802
2803 if (fsck_err(trans, dir_loop, "directory structure loop")) {
2804 ret = remove_backpointer(trans, &inode);
2805 bch_err_msg(c, ret, "removing dirent");
2806 if (ret)
2807 break;
2808
2809 ret = reattach_inode(trans, &inode);
2810 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
2811 }
2812
2813 goto out;
2814 }
2815 }
2816
2817 if (inode.bi_subvol)
2818 min_bi_depth = 0;
2819
2820 if (redo_bi_depth)
2821 ret = bch2_bi_depth_renumber(trans, &path, min_bi_depth);
2822 out:
2823 fsck_err:
2824 bch2_trans_iter_exit(trans, &inode_iter);
2825 darray_exit(&path);
2826 printbuf_exit(&buf);
2827 bch_err_fn(c, ret);
2828 return ret;
2829 }
2830
2831 /*
2832 * Check for loops in the directory structure: all other connectivity issues
2833 * have been fixed by prior passes
2834 */
bch2_check_directory_structure(struct bch_fs * c)2835 int bch2_check_directory_structure(struct bch_fs *c)
2836 {
2837 int ret = bch2_trans_run(c,
2838 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2839 BTREE_ITER_intent|
2840 BTREE_ITER_prefetch|
2841 BTREE_ITER_all_snapshots, k,
2842 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2843 if (!S_ISDIR(bkey_inode_mode(k)))
2844 continue;
2845
2846 if (bch2_inode_flags(k) & BCH_INODE_unlinked)
2847 continue;
2848
2849 check_path_loop(trans, k);
2850 })));
2851
2852 bch_err_fn(c, ret);
2853 return ret;
2854 }
2855
2856 struct nlink_table {
2857 size_t nr;
2858 size_t size;
2859
2860 struct nlink {
2861 u64 inum;
2862 u32 snapshot;
2863 u32 count;
2864 } *d;
2865 };
2866
add_nlink(struct bch_fs * c,struct nlink_table * t,u64 inum,u32 snapshot)2867 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2868 u64 inum, u32 snapshot)
2869 {
2870 if (t->nr == t->size) {
2871 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2872 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2873
2874 if (!d) {
2875 bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2876 new_size);
2877 return -BCH_ERR_ENOMEM_fsck_add_nlink;
2878 }
2879
2880 if (t->d)
2881 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2882 kvfree(t->d);
2883
2884 t->d = d;
2885 t->size = new_size;
2886 }
2887
2888
2889 t->d[t->nr++] = (struct nlink) {
2890 .inum = inum,
2891 .snapshot = snapshot,
2892 };
2893
2894 return 0;
2895 }
2896
nlink_cmp(const void * _l,const void * _r)2897 static int nlink_cmp(const void *_l, const void *_r)
2898 {
2899 const struct nlink *l = _l;
2900 const struct nlink *r = _r;
2901
2902 return cmp_int(l->inum, r->inum);
2903 }
2904
inc_link(struct bch_fs * c,struct snapshots_seen * s,struct nlink_table * links,u64 range_start,u64 range_end,u64 inum,u32 snapshot)2905 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2906 struct nlink_table *links,
2907 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2908 {
2909 struct nlink *link, key = {
2910 .inum = inum, .snapshot = U32_MAX,
2911 };
2912
2913 if (inum < range_start || inum >= range_end)
2914 return;
2915
2916 link = __inline_bsearch(&key, links->d, links->nr,
2917 sizeof(links->d[0]), nlink_cmp);
2918 if (!link)
2919 return;
2920
2921 while (link > links->d && link[0].inum == link[-1].inum)
2922 --link;
2923
2924 for (; link < links->d + links->nr && link->inum == inum; link++)
2925 if (ref_visible(c, s, snapshot, link->snapshot)) {
2926 link->count++;
2927 if (link->snapshot >= snapshot)
2928 break;
2929 }
2930 }
2931
2932 noinline_for_stack
check_nlinks_find_hardlinks(struct bch_fs * c,struct nlink_table * t,u64 start,u64 * end)2933 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2934 struct nlink_table *t,
2935 u64 start, u64 *end)
2936 {
2937 int ret = bch2_trans_run(c,
2938 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2939 POS(0, start),
2940 BTREE_ITER_intent|
2941 BTREE_ITER_prefetch|
2942 BTREE_ITER_all_snapshots, k, ({
2943 if (!bkey_is_inode(k.k))
2944 continue;
2945
2946 /* Should never fail, checked by bch2_inode_invalid: */
2947 struct bch_inode_unpacked u;
2948 _ret3 = bch2_inode_unpack(k, &u);
2949 if (_ret3)
2950 break;
2951
2952 /*
2953 * Backpointer and directory structure checks are sufficient for
2954 * directories, since they can't have hardlinks:
2955 */
2956 if (S_ISDIR(u.bi_mode))
2957 continue;
2958
2959 /*
2960 * Previous passes ensured that bi_nlink is nonzero if
2961 * it had multiple hardlinks:
2962 */
2963 if (!u.bi_nlink)
2964 continue;
2965
2966 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2967 if (ret) {
2968 *end = k.k->p.offset;
2969 ret = 0;
2970 break;
2971 }
2972 0;
2973 })));
2974
2975 bch_err_fn(c, ret);
2976 return ret;
2977 }
2978
2979 noinline_for_stack
check_nlinks_walk_dirents(struct bch_fs * c,struct nlink_table * links,u64 range_start,u64 range_end)2980 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2981 u64 range_start, u64 range_end)
2982 {
2983 struct snapshots_seen s;
2984
2985 snapshots_seen_init(&s);
2986
2987 int ret = bch2_trans_run(c,
2988 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2989 BTREE_ITER_intent|
2990 BTREE_ITER_prefetch|
2991 BTREE_ITER_all_snapshots, k, ({
2992 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2993 if (ret)
2994 break;
2995
2996 if (k.k->type == KEY_TYPE_dirent) {
2997 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2998
2999 if (d.v->d_type != DT_DIR &&
3000 d.v->d_type != DT_SUBVOL)
3001 inc_link(c, &s, links, range_start, range_end,
3002 le64_to_cpu(d.v->d_inum), d.k->p.snapshot);
3003 }
3004 0;
3005 })));
3006
3007 snapshots_seen_exit(&s);
3008
3009 bch_err_fn(c, ret);
3010 return ret;
3011 }
3012
check_nlinks_update_inode(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,struct nlink_table * links,size_t * idx,u64 range_end)3013 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
3014 struct bkey_s_c k,
3015 struct nlink_table *links,
3016 size_t *idx, u64 range_end)
3017 {
3018 struct bch_inode_unpacked u;
3019 struct nlink *link = &links->d[*idx];
3020 int ret = 0;
3021
3022 if (k.k->p.offset >= range_end)
3023 return 1;
3024
3025 if (!bkey_is_inode(k.k))
3026 return 0;
3027
3028 ret = bch2_inode_unpack(k, &u);
3029 if (ret)
3030 return ret;
3031
3032 if (S_ISDIR(u.bi_mode))
3033 return 0;
3034
3035 if (!u.bi_nlink)
3036 return 0;
3037
3038 while ((cmp_int(link->inum, k.k->p.offset) ?:
3039 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
3040 BUG_ON(*idx == links->nr);
3041 link = &links->d[++*idx];
3042 }
3043
3044 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
3045 trans, inode_wrong_nlink,
3046 "inode %llu type %s has wrong i_nlink (%u, should be %u)",
3047 u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
3048 bch2_inode_nlink_get(&u), link->count)) {
3049 bch2_inode_nlink_set(&u, link->count);
3050 ret = __bch2_fsck_write_inode(trans, &u);
3051 }
3052 fsck_err:
3053 return ret;
3054 }
3055
3056 noinline_for_stack
check_nlinks_update_hardlinks(struct bch_fs * c,struct nlink_table * links,u64 range_start,u64 range_end)3057 static int check_nlinks_update_hardlinks(struct bch_fs *c,
3058 struct nlink_table *links,
3059 u64 range_start, u64 range_end)
3060 {
3061 size_t idx = 0;
3062
3063 int ret = bch2_trans_run(c,
3064 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
3065 POS(0, range_start),
3066 BTREE_ITER_intent|BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
3067 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
3068 check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
3069 if (ret < 0) {
3070 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
3071 return ret;
3072 }
3073
3074 return 0;
3075 }
3076
bch2_check_nlinks(struct bch_fs * c)3077 int bch2_check_nlinks(struct bch_fs *c)
3078 {
3079 struct nlink_table links = { 0 };
3080 u64 this_iter_range_start, next_iter_range_start = 0;
3081 int ret = 0;
3082
3083 do {
3084 this_iter_range_start = next_iter_range_start;
3085 next_iter_range_start = U64_MAX;
3086
3087 ret = check_nlinks_find_hardlinks(c, &links,
3088 this_iter_range_start,
3089 &next_iter_range_start);
3090
3091 ret = check_nlinks_walk_dirents(c, &links,
3092 this_iter_range_start,
3093 next_iter_range_start);
3094 if (ret)
3095 break;
3096
3097 ret = check_nlinks_update_hardlinks(c, &links,
3098 this_iter_range_start,
3099 next_iter_range_start);
3100 if (ret)
3101 break;
3102
3103 links.nr = 0;
3104 } while (next_iter_range_start != U64_MAX);
3105
3106 kvfree(links.d);
3107 bch_err_fn(c, ret);
3108 return ret;
3109 }
3110
fix_reflink_p_key(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k)3111 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
3112 struct bkey_s_c k)
3113 {
3114 struct bkey_s_c_reflink_p p;
3115 struct bkey_i_reflink_p *u;
3116
3117 if (k.k->type != KEY_TYPE_reflink_p)
3118 return 0;
3119
3120 p = bkey_s_c_to_reflink_p(k);
3121
3122 if (!p.v->front_pad && !p.v->back_pad)
3123 return 0;
3124
3125 u = bch2_trans_kmalloc(trans, sizeof(*u));
3126 int ret = PTR_ERR_OR_ZERO(u);
3127 if (ret)
3128 return ret;
3129
3130 bkey_reassemble(&u->k_i, k);
3131 u->v.front_pad = 0;
3132 u->v.back_pad = 0;
3133
3134 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_norun);
3135 }
3136
bch2_fix_reflink_p(struct bch_fs * c)3137 int bch2_fix_reflink_p(struct bch_fs *c)
3138 {
3139 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
3140 return 0;
3141
3142 int ret = bch2_trans_run(c,
3143 for_each_btree_key_commit(trans, iter,
3144 BTREE_ID_extents, POS_MIN,
3145 BTREE_ITER_intent|BTREE_ITER_prefetch|
3146 BTREE_ITER_all_snapshots, k,
3147 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
3148 fix_reflink_p_key(trans, &iter, k)));
3149 bch_err_fn(c, ret);
3150 return ret;
3151 }
3152
3153 #ifndef NO_BCACHEFS_CHARDEV
3154
3155 struct fsck_thread {
3156 struct thread_with_stdio thr;
3157 struct bch_fs *c;
3158 struct bch_opts opts;
3159 };
3160
bch2_fsck_thread_exit(struct thread_with_stdio * _thr)3161 static void bch2_fsck_thread_exit(struct thread_with_stdio *_thr)
3162 {
3163 struct fsck_thread *thr = container_of(_thr, struct fsck_thread, thr);
3164 kfree(thr);
3165 }
3166
bch2_fsck_offline_thread_fn(struct thread_with_stdio * stdio)3167 static int bch2_fsck_offline_thread_fn(struct thread_with_stdio *stdio)
3168 {
3169 struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
3170 struct bch_fs *c = thr->c;
3171
3172 int ret = PTR_ERR_OR_ZERO(c);
3173 if (ret)
3174 return ret;
3175
3176 ret = bch2_fs_start(thr->c);
3177 if (ret)
3178 goto err;
3179
3180 if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
3181 bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: errors fixed\n", c->name);
3182 ret |= 1;
3183 }
3184 if (test_bit(BCH_FS_error, &c->flags)) {
3185 bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: still has errors\n", c->name);
3186 ret |= 4;
3187 }
3188 err:
3189 bch2_fs_stop(c);
3190 return ret;
3191 }
3192
3193 static const struct thread_with_stdio_ops bch2_offline_fsck_ops = {
3194 .exit = bch2_fsck_thread_exit,
3195 .fn = bch2_fsck_offline_thread_fn,
3196 };
3197
bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user * user_arg)3198 long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_arg)
3199 {
3200 struct bch_ioctl_fsck_offline arg;
3201 struct fsck_thread *thr = NULL;
3202 darray_str(devs) = {};
3203 long ret = 0;
3204
3205 if (copy_from_user(&arg, user_arg, sizeof(arg)))
3206 return -EFAULT;
3207
3208 if (arg.flags)
3209 return -EINVAL;
3210
3211 if (!capable(CAP_SYS_ADMIN))
3212 return -EPERM;
3213
3214 for (size_t i = 0; i < arg.nr_devs; i++) {
3215 u64 dev_u64;
3216 ret = copy_from_user_errcode(&dev_u64, &user_arg->devs[i], sizeof(u64));
3217 if (ret)
3218 goto err;
3219
3220 char *dev_str = strndup_user((char __user *)(unsigned long) dev_u64, PATH_MAX);
3221 ret = PTR_ERR_OR_ZERO(dev_str);
3222 if (ret)
3223 goto err;
3224
3225 ret = darray_push(&devs, dev_str);
3226 if (ret) {
3227 kfree(dev_str);
3228 goto err;
3229 }
3230 }
3231
3232 thr = kzalloc(sizeof(*thr), GFP_KERNEL);
3233 if (!thr) {
3234 ret = -ENOMEM;
3235 goto err;
3236 }
3237
3238 thr->opts = bch2_opts_empty();
3239
3240 if (arg.opts) {
3241 char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
3242 ret = PTR_ERR_OR_ZERO(optstr) ?:
3243 bch2_parse_mount_opts(NULL, &thr->opts, NULL, optstr);
3244 if (!IS_ERR(optstr))
3245 kfree(optstr);
3246
3247 if (ret)
3248 goto err;
3249 }
3250
3251 opt_set(thr->opts, stdio, (u64)(unsigned long)&thr->thr.stdio);
3252 opt_set(thr->opts, read_only, 1);
3253 opt_set(thr->opts, ratelimit_errors, 0);
3254
3255 /* We need request_key() to be called before we punt to kthread: */
3256 opt_set(thr->opts, nostart, true);
3257
3258 bch2_thread_with_stdio_init(&thr->thr, &bch2_offline_fsck_ops);
3259
3260 thr->c = bch2_fs_open(devs.data, arg.nr_devs, thr->opts);
3261
3262 if (!IS_ERR(thr->c) &&
3263 thr->c->opts.errors == BCH_ON_ERROR_panic)
3264 thr->c->opts.errors = BCH_ON_ERROR_ro;
3265
3266 ret = __bch2_run_thread_with_stdio(&thr->thr);
3267 out:
3268 darray_for_each(devs, i)
3269 kfree(*i);
3270 darray_exit(&devs);
3271 return ret;
3272 err:
3273 if (thr)
3274 bch2_fsck_thread_exit(&thr->thr);
3275 pr_err("ret %s", bch2_err_str(ret));
3276 goto out;
3277 }
3278
bch2_fsck_online_thread_fn(struct thread_with_stdio * stdio)3279 static int bch2_fsck_online_thread_fn(struct thread_with_stdio *stdio)
3280 {
3281 struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
3282 struct bch_fs *c = thr->c;
3283
3284 c->stdio_filter = current;
3285 c->stdio = &thr->thr.stdio;
3286
3287 /*
3288 * XXX: can we figure out a way to do this without mucking with c->opts?
3289 */
3290 unsigned old_fix_errors = c->opts.fix_errors;
3291 if (opt_defined(thr->opts, fix_errors))
3292 c->opts.fix_errors = thr->opts.fix_errors;
3293 else
3294 c->opts.fix_errors = FSCK_FIX_ask;
3295
3296 c->opts.fsck = true;
3297 set_bit(BCH_FS_fsck_running, &c->flags);
3298
3299 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
3300 int ret = bch2_run_online_recovery_passes(c);
3301
3302 clear_bit(BCH_FS_fsck_running, &c->flags);
3303 bch_err_fn(c, ret);
3304
3305 c->stdio = NULL;
3306 c->stdio_filter = NULL;
3307 c->opts.fix_errors = old_fix_errors;
3308
3309 up(&c->online_fsck_mutex);
3310 bch2_ro_ref_put(c);
3311 return ret;
3312 }
3313
3314 static const struct thread_with_stdio_ops bch2_online_fsck_ops = {
3315 .exit = bch2_fsck_thread_exit,
3316 .fn = bch2_fsck_online_thread_fn,
3317 };
3318
bch2_ioctl_fsck_online(struct bch_fs * c,struct bch_ioctl_fsck_online arg)3319 long bch2_ioctl_fsck_online(struct bch_fs *c, struct bch_ioctl_fsck_online arg)
3320 {
3321 struct fsck_thread *thr = NULL;
3322 long ret = 0;
3323
3324 if (arg.flags)
3325 return -EINVAL;
3326
3327 if (!capable(CAP_SYS_ADMIN))
3328 return -EPERM;
3329
3330 if (!bch2_ro_ref_tryget(c))
3331 return -EROFS;
3332
3333 if (down_trylock(&c->online_fsck_mutex)) {
3334 bch2_ro_ref_put(c);
3335 return -EAGAIN;
3336 }
3337
3338 thr = kzalloc(sizeof(*thr), GFP_KERNEL);
3339 if (!thr) {
3340 ret = -ENOMEM;
3341 goto err;
3342 }
3343
3344 thr->c = c;
3345 thr->opts = bch2_opts_empty();
3346
3347 if (arg.opts) {
3348 char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
3349
3350 ret = PTR_ERR_OR_ZERO(optstr) ?:
3351 bch2_parse_mount_opts(c, &thr->opts, NULL, optstr);
3352 if (!IS_ERR(optstr))
3353 kfree(optstr);
3354
3355 if (ret)
3356 goto err;
3357 }
3358
3359 ret = bch2_run_thread_with_stdio(&thr->thr, &bch2_online_fsck_ops);
3360 err:
3361 if (ret < 0) {
3362 bch_err_fn(c, ret);
3363 if (thr)
3364 bch2_fsck_thread_exit(&thr->thr);
3365 up(&c->online_fsck_mutex);
3366 bch2_ro_ref_put(c);
3367 }
3368 return ret;
3369 }
3370
3371 #endif /* NO_BCACHEFS_CHARDEV */
3372