]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - fs/bcachefs/recovery.c
projects / thirdparty / kernel / stable.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
bcachefs: Repair pass for scanning for btree nodes
[thirdparty/kernel/stable.git] / fs / bcachefs / recovery.c
1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "alloc_background.h"
5 #include "bkey_buf.h"
6 #include "btree_journal_iter.h"
7 #include "btree_node_scan.h"
8 #include "btree_update.h"
9 #include "btree_update_interior.h"
10 #include "btree_io.h"
11 #include "buckets.h"
12 #include "dirent.h"
13 #include "errcode.h"
14 #include "error.h"
15 #include "fs-common.h"
16 #include "journal_io.h"
17 #include "journal_reclaim.h"
18 #include "journal_seq_blacklist.h"
19 #include "logged_ops.h"
20 #include "move.h"
21 #include "quota.h"
22 #include "rebalance.h"
23 #include "recovery.h"
24 #include "recovery_passes.h"
25 #include "replicas.h"
26 #include "sb-clean.h"
27 #include "sb-downgrade.h"
28 #include "snapshot.h"
29 #include "super-io.h"
30
31 #include <linux/sort.h>
32 #include <linux/stat.h>
33
34 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
35
36 static bool btree_id_is_alloc(enum btree_id id)
37 {
38 switch (id) {
39 case BTREE_ID_alloc:
40 case BTREE_ID_backpointers:
41 case BTREE_ID_need_discard:
42 case BTREE_ID_freespace:
43 case BTREE_ID_bucket_gens:
44 return true;
45 default:
46 return false;
47 }
48 }
49
50 /* for -o reconstruct_alloc: */
51 static void bch2_reconstruct_alloc(struct bch_fs *c)
52 {
53 bch2_journal_log_msg(c, "dropping alloc info");
54 bch_info(c, "dropping and reconstructing all alloc info");
55
56 mutex_lock(&c->sb_lock);
57 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
58
59 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required);
60 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_info, ext->recovery_passes_required);
61 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_lrus, ext->recovery_passes_required);
62 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_extents_to_backpointers, ext->recovery_passes_required);
63 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_to_lru_refs, ext->recovery_passes_required);
64
65 __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_alloc_key, ext->errors_silent);
66 __set_bit_le64(BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen, ext->errors_silent);
67 __set_bit_le64(BCH_FSCK_ERR_stale_dirty_ptr, ext->errors_silent);
68 __set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
69 __set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
70 __set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
71 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent);
72 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent);
73 __set_bit_le64(BCH_FSCK_ERR_need_discard_key_wrong, ext->errors_silent);
74 __set_bit_le64(BCH_FSCK_ERR_freespace_key_wrong, ext->errors_silent);
75 __set_bit_le64(BCH_FSCK_ERR_bucket_gens_key_wrong, ext->errors_silent);
76 __set_bit_le64(BCH_FSCK_ERR_freespace_hole_missing, ext->errors_silent);
77 __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_backpointer, ext->errors_silent);
78 __set_bit_le64(BCH_FSCK_ERR_lru_entry_bad, ext->errors_silent);
79 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
80
81 bch2_write_super(c);
82 mutex_unlock(&c->sb_lock);
83
84 c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
85
86
87 bch2_shoot_down_journal_keys(c, BTREE_ID_alloc,
88 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
89 bch2_shoot_down_journal_keys(c, BTREE_ID_backpointers,
90 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
91 bch2_shoot_down_journal_keys(c, BTREE_ID_need_discard,
92 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
93 bch2_shoot_down_journal_keys(c, BTREE_ID_freespace,
94 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
95 bch2_shoot_down_journal_keys(c, BTREE_ID_bucket_gens,
96 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
97 }
98
99 /*
100 * Btree node pointers have a field to stack a pointer to the in memory btree
101 * node; we need to zero out this field when reading in btree nodes, or when
102 * reading in keys from the journal:
103 */
104 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
105 {
106 darray_for_each(*keys, i)
107 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
108 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
109 }
110
111 /* journal replay: */
112
113 static void replay_now_at(struct journal *j, u64 seq)
114 {
115 BUG_ON(seq < j->replay_journal_seq);
116
117 seq = min(seq, j->replay_journal_seq_end);
118
119 while (j->replay_journal_seq < seq)
120 bch2_journal_pin_put(j, j->replay_journal_seq++);
121 }
122
123 static int bch2_journal_replay_key(struct btree_trans *trans,
124 struct journal_key *k)
125 {
126 struct btree_iter iter;
127 unsigned iter_flags =
128 BTREE_ITER_INTENT|
129 BTREE_ITER_NOT_EXTENTS;
130 unsigned update_flags = BTREE_TRIGGER_NORUN;
131 int ret;
132
133 if (k->overwritten)
134 return 0;
135
136 trans->journal_res.seq = k->journal_seq;
137
138 /*
139 * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
140 * keep the key cache coherent with the underlying btree. Nothing
141 * besides the allocator is doing updates yet so we don't need key cache
142 * coherency for non-alloc btrees, and key cache fills for snapshots
143 * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
144 * the snapshots recovery pass runs.
145 */
146 if (!k->level && k->btree_id == BTREE_ID_alloc)
147 iter_flags |= BTREE_ITER_CACHED;
148 else
149 update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
150
151 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
152 BTREE_MAX_DEPTH, k->level,
153 iter_flags);
154 ret = bch2_btree_iter_traverse(&iter);
155 if (ret)
156 goto out;
157
158 struct btree_path *path = btree_iter_path(trans, &iter);
159 if (unlikely(!btree_path_node(path, k->level))) {
160 bch2_trans_iter_exit(trans, &iter);
161 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
162 BTREE_MAX_DEPTH, 0, iter_flags);
163 ret = bch2_btree_iter_traverse(&iter) ?:
164 bch2_btree_increase_depth(trans, iter.path, 0) ?:
165 -BCH_ERR_transaction_restart_nested;
166 goto out;
167 }
168
169 /* Must be checked with btree locked: */
170 if (k->overwritten)
171 goto out;
172
173 ret = bch2_trans_update(trans, &iter, k->k, update_flags);
174 out:
175 bch2_trans_iter_exit(trans, &iter);
176 return ret;
177 }
178
179 static int journal_sort_seq_cmp(const void *_l, const void *_r)
180 {
181 const struct journal_key *l = *((const struct journal_key **)_l);
182 const struct journal_key *r = *((const struct journal_key **)_r);
183
184 return cmp_int(l->journal_seq, r->journal_seq);
185 }
186
187 int bch2_journal_replay(struct bch_fs *c)
188 {
189 struct journal_keys *keys = &c->journal_keys;
190 DARRAY(struct journal_key *) keys_sorted = { 0 };
191 struct journal *j = &c->journal;
192 u64 start_seq = c->journal_replay_seq_start;
193 u64 end_seq = c->journal_replay_seq_start;
194 struct btree_trans *trans = bch2_trans_get(c);
195 bool immediate_flush = false;
196 int ret = 0;
197
198 if (keys->nr) {
199 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
200 keys->nr, start_seq, end_seq);
201 if (ret)
202 goto err;
203 }
204
205 BUG_ON(!atomic_read(&keys->ref));
206
207 move_gap(keys, keys->nr);
208
209 /*
210 * First, attempt to replay keys in sorted order. This is more
211 * efficient - better locality of btree access - but some might fail if
212 * that would cause a journal deadlock.
213 */
214 darray_for_each(*keys, k) {
215 cond_resched();
216
217 /*
218 * k->allocated means the key wasn't read in from the journal,
219 * rather it was from early repair code
220 */
221 if (k->allocated)
222 immediate_flush = true;
223
224 /* Skip fastpath if we're low on space in the journal */
225 ret = c->journal.watermark ? -1 :
226 commit_do(trans, NULL, NULL,
227 BCH_TRANS_COMMIT_no_enospc|
228 BCH_TRANS_COMMIT_journal_reclaim|
229 (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0),
230 bch2_journal_replay_key(trans, k));
231 BUG_ON(!ret && !k->overwritten);
232 if (ret) {
233 ret = darray_push(&keys_sorted, k);
234 if (ret)
235 goto err;
236 }
237 }
238
239 /*
240 * Now, replay any remaining keys in the order in which they appear in
241 * the journal, unpinning those journal entries as we go:
242 */
243 sort(keys_sorted.data, keys_sorted.nr,
244 sizeof(keys_sorted.data[0]),
245 journal_sort_seq_cmp, NULL);
246
247 darray_for_each(keys_sorted, kp) {
248 cond_resched();
249
250 struct journal_key *k = *kp;
251
252 replay_now_at(j, k->journal_seq);
253
254 ret = commit_do(trans, NULL, NULL,
255 BCH_TRANS_COMMIT_no_enospc|
256 (!k->allocated
257 ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim
258 : 0),
259 bch2_journal_replay_key(trans, k));
260 bch_err_msg(c, ret, "while replaying key at btree %s level %u:",
261 bch2_btree_id_str(k->btree_id), k->level);
262 if (ret)
263 goto err;
264
265 BUG_ON(!k->overwritten);
266 }
267
268 /*
269 * We need to put our btree_trans before calling flush_all_pins(), since
270 * that will use a btree_trans internally
271 */
272 bch2_trans_put(trans);
273 trans = NULL;
274
275 if (!c->opts.retain_recovery_info &&
276 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay)
277 bch2_journal_keys_put_initial(c);
278
279 replay_now_at(j, j->replay_journal_seq_end);
280 j->replay_journal_seq = 0;
281
282 bch2_journal_set_replay_done(j);
283
284 /* if we did any repair, flush it immediately */
285 if (immediate_flush) {
286 bch2_journal_flush_all_pins(&c->journal);
287 ret = bch2_journal_meta(&c->journal);
288 }
289
290 if (keys->nr)
291 bch2_journal_log_msg(c, "journal replay finished");
292 err:
293 if (trans)
294 bch2_trans_put(trans);
295 darray_exit(&keys_sorted);
296 bch_err_fn(c, ret);
297 return ret;
298 }
299
300 /* journal replay early: */
301
302 static int journal_replay_entry_early(struct bch_fs *c,
303 struct jset_entry *entry)
304 {
305 int ret = 0;
306
307 switch (entry->type) {
308 case BCH_JSET_ENTRY_btree_root: {
309 struct btree_root *r;
310
311 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
312 ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
313 if (ret)
314 return ret;
315 }
316
317 r = bch2_btree_id_root(c, entry->btree_id);
318
319 if (entry->u64s) {
320 r->level = entry->level;
321 bkey_copy(&r->key, (struct bkey_i *) entry->start);
322 r->error = 0;
323 } else {
324 r->error = -BCH_ERR_btree_node_read_error;
325 }
326 r->alive = true;
327 break;
328 }
329 case BCH_JSET_ENTRY_usage: {
330 struct jset_entry_usage *u =
331 container_of(entry, struct jset_entry_usage, entry);
332
333 switch (entry->btree_id) {
334 case BCH_FS_USAGE_reserved:
335 if (entry->level < BCH_REPLICAS_MAX)
336 c->usage_base->persistent_reserved[entry->level] =
337 le64_to_cpu(u->v);
338 break;
339 case BCH_FS_USAGE_inodes:
340 c->usage_base->b.nr_inodes = le64_to_cpu(u->v);
341 break;
342 case BCH_FS_USAGE_key_version:
343 atomic64_set(&c->key_version,
344 le64_to_cpu(u->v));
345 break;
346 }
347
348 break;
349 }
350 case BCH_JSET_ENTRY_data_usage: {
351 struct jset_entry_data_usage *u =
352 container_of(entry, struct jset_entry_data_usage, entry);
353
354 ret = bch2_replicas_set_usage(c, &u->r,
355 le64_to_cpu(u->v));
356 break;
357 }
358 case BCH_JSET_ENTRY_dev_usage: {
359 struct jset_entry_dev_usage *u =
360 container_of(entry, struct jset_entry_dev_usage, entry);
361 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
362 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
363
364 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
365 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
366 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
367 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
368 }
369
370 break;
371 }
372 case BCH_JSET_ENTRY_blacklist: {
373 struct jset_entry_blacklist *bl_entry =
374 container_of(entry, struct jset_entry_blacklist, entry);
375
376 ret = bch2_journal_seq_blacklist_add(c,
377 le64_to_cpu(bl_entry->seq),
378 le64_to_cpu(bl_entry->seq) + 1);
379 break;
380 }
381 case BCH_JSET_ENTRY_blacklist_v2: {
382 struct jset_entry_blacklist_v2 *bl_entry =
383 container_of(entry, struct jset_entry_blacklist_v2, entry);
384
385 ret = bch2_journal_seq_blacklist_add(c,
386 le64_to_cpu(bl_entry->start),
387 le64_to_cpu(bl_entry->end) + 1);
388 break;
389 }
390 case BCH_JSET_ENTRY_clock: {
391 struct jset_entry_clock *clock =
392 container_of(entry, struct jset_entry_clock, entry);
393
394 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
395 }
396 }
397
398 return ret;
399 }
400
401 static int journal_replay_early(struct bch_fs *c,
402 struct bch_sb_field_clean *clean)
403 {
404 if (clean) {
405 for (struct jset_entry *entry = clean->start;
406 entry != vstruct_end(&clean->field);
407 entry = vstruct_next(entry)) {
408 int ret = journal_replay_entry_early(c, entry);
409 if (ret)
410 return ret;
411 }
412 } else {
413 struct genradix_iter iter;
414 struct journal_replay *i, **_i;
415
416 genradix_for_each(&c->journal_entries, iter, _i) {
417 i = *_i;
418
419 if (journal_replay_ignore(i))
420 continue;
421
422 vstruct_for_each(&i->j, entry) {
423 int ret = journal_replay_entry_early(c, entry);
424 if (ret)
425 return ret;
426 }
427 }
428 }
429
430 bch2_fs_usage_initialize(c);
431
432 return 0;
433 }
434
435 /* sb clean section: */
436
437 static int read_btree_roots(struct bch_fs *c)
438 {
439 int ret = 0;
440
441 for (unsigned i = 0; i < btree_id_nr_alive(c); i++) {
442 struct btree_root *r = bch2_btree_id_root(c, i);
443
444 if (!r->alive)
445 continue;
446
447 if (btree_id_is_alloc(i) && c->opts.reconstruct_alloc)
448 continue;
449
450 if (mustfix_fsck_err_on((ret = r->error),
451 c, btree_root_bkey_invalid,
452 "invalid btree root %s",
453 bch2_btree_id_str(i)) ||
454 mustfix_fsck_err_on((ret = r->error = bch2_btree_root_read(c, i, &r->key, r->level)),
455 c, btree_root_read_error,
456 "error reading btree root %s l=%u: %s",
457 bch2_btree_id_str(i), r->level, bch2_err_str(ret))) {
458 if (btree_id_is_alloc(i)) {
459 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_allocations);
460 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_info);
461 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_lrus);
462 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers);
463 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_to_lru_refs);
464 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
465 r->error = 0;
466 } else if (!(c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes))) {
467 bch_info(c, "will run btree node scan");
468 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes);
469 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
470 }
471
472 ret = 0;
473 }
474 }
475
476 for (unsigned i = 0; i < BTREE_ID_NR; i++) {
477 struct btree_root *r = bch2_btree_id_root(c, i);
478
479 if (!r->b && !r->error) {
480 r->alive = false;
481 r->level = 0;
482 bch2_btree_root_alloc_fake(c, i, 0);
483 }
484 }
485 fsck_err:
486 return ret;
487 }
488
489 static bool check_version_upgrade(struct bch_fs *c)
490 {
491 unsigned latest_version = bcachefs_metadata_version_current;
492 unsigned latest_compatible = min(latest_version,
493 bch2_latest_compatible_version(c->sb.version));
494 unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
495 unsigned new_version = 0;
496
497 if (old_version < bcachefs_metadata_required_upgrade_below) {
498 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
499 latest_compatible < bcachefs_metadata_required_upgrade_below)
500 new_version = latest_version;
501 else
502 new_version = latest_compatible;
503 } else {
504 switch (c->opts.version_upgrade) {
505 case BCH_VERSION_UPGRADE_compatible:
506 new_version = latest_compatible;
507 break;
508 case BCH_VERSION_UPGRADE_incompatible:
509 new_version = latest_version;
510 break;
511 case BCH_VERSION_UPGRADE_none:
512 new_version = min(old_version, latest_version);
513 break;
514 }
515 }
516
517 if (new_version > old_version) {
518 struct printbuf buf = PRINTBUF;
519
520 if (old_version < bcachefs_metadata_required_upgrade_below)
521 prt_str(&buf, "Version upgrade required:\n");
522
523 if (old_version != c->sb.version) {
524 prt_str(&buf, "Version upgrade from ");
525 bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
526 prt_str(&buf, " to ");
527 bch2_version_to_text(&buf, c->sb.version);
528 prt_str(&buf, " incomplete\n");
529 }
530
531 prt_printf(&buf, "Doing %s version upgrade from ",
532 BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
533 ? "incompatible" : "compatible");
534 bch2_version_to_text(&buf, old_version);
535 prt_str(&buf, " to ");
536 bch2_version_to_text(&buf, new_version);
537 prt_newline(&buf);
538
539 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
540 __le64 passes = ext->recovery_passes_required[0];
541 bch2_sb_set_upgrade(c, old_version, new_version);
542 passes = ext->recovery_passes_required[0] & ~passes;
543
544 if (passes) {
545 prt_str(&buf, " running recovery passes: ");
546 prt_bitflags(&buf, bch2_recovery_passes,
547 bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
548 }
549
550 bch_info(c, "%s", buf.buf);
551
552 bch2_sb_upgrade(c, new_version);
553
554 printbuf_exit(&buf);
555 return true;
556 }
557
558 return false;
559 }
560
561 int bch2_fs_recovery(struct bch_fs *c)
562 {
563 struct bch_sb_field_clean *clean = NULL;
564 struct jset *last_journal_entry = NULL;
565 u64 last_seq = 0, blacklist_seq, journal_seq;
566 int ret = 0;
567
568 if (c->sb.clean) {
569 clean = bch2_read_superblock_clean(c);
570 ret = PTR_ERR_OR_ZERO(clean);
571 if (ret)
572 goto err;
573
574 bch_info(c, "recovering from clean shutdown, journal seq %llu",
575 le64_to_cpu(clean->journal_seq));
576 } else {
577 bch_info(c, "recovering from unclean shutdown");
578 }
579
580 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
581 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
582 ret = -EINVAL;
583 goto err;
584 }
585
586 if (!c->sb.clean &&
587 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
588 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
589 ret = -EINVAL;
590 goto err;
591 }
592
593 if (c->opts.norecovery)
594 c->opts.recovery_pass_last = BCH_RECOVERY_PASS_journal_replay - 1;
595
596 if (!c->opts.nochanges) {
597 mutex_lock(&c->sb_lock);
598 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
599 bool write_sb = false;
600
601 if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) {
602 ext->recovery_passes_required[0] |=
603 cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology)));
604 write_sb = true;
605 }
606
607 u64 sb_passes = bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
608 if (sb_passes) {
609 struct printbuf buf = PRINTBUF;
610 prt_str(&buf, "superblock requires following recovery passes to be run:\n ");
611 prt_bitflags(&buf, bch2_recovery_passes, sb_passes);
612 bch_info(c, "%s", buf.buf);
613 printbuf_exit(&buf);
614 }
615
616 if (bch2_check_version_downgrade(c)) {
617 struct printbuf buf = PRINTBUF;
618
619 prt_str(&buf, "Version downgrade required:");
620
621 __le64 passes = ext->recovery_passes_required[0];
622 bch2_sb_set_downgrade(c,
623 BCH_VERSION_MINOR(bcachefs_metadata_version_current),
624 BCH_VERSION_MINOR(c->sb.version));
625 passes = ext->recovery_passes_required[0] & ~passes;
626 if (passes) {
627 prt_str(&buf, "\n running recovery passes: ");
628 prt_bitflags(&buf, bch2_recovery_passes,
629 bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
630 }
631
632 bch_info(c, "%s", buf.buf);
633 printbuf_exit(&buf);
634 write_sb = true;
635 }
636
637 if (check_version_upgrade(c))
638 write_sb = true;
639
640 if (write_sb)
641 bch2_write_super(c);
642
643 c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
644 mutex_unlock(&c->sb_lock);
645 }
646
647 if (c->opts.fsck && IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
648 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
649
650 if (c->opts.fsck)
651 set_bit(BCH_FS_fsck_running, &c->flags);
652
653 ret = bch2_blacklist_table_initialize(c);
654 if (ret) {
655 bch_err(c, "error initializing blacklist table");
656 goto err;
657 }
658
659 if (!c->sb.clean || c->opts.fsck || c->opts.retain_recovery_info) {
660 struct genradix_iter iter;
661 struct journal_replay **i;
662
663 bch_verbose(c, "starting journal read");
664 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
665 if (ret)
666 goto err;
667
668 /*
669 * note: cmd_list_journal needs the blacklist table fully up to date so
670 * it can asterisk ignored journal entries:
671 */
672 if (c->opts.read_journal_only)
673 goto out;
674
675 genradix_for_each_reverse(&c->journal_entries, iter, i)
676 if (!journal_replay_ignore(*i)) {
677 last_journal_entry = &(*i)->j;
678 break;
679 }
680
681 if (mustfix_fsck_err_on(c->sb.clean &&
682 last_journal_entry &&
683 !journal_entry_empty(last_journal_entry), c,
684 clean_but_journal_not_empty,
685 "filesystem marked clean but journal not empty")) {
686 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
687 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
688 c->sb.clean = false;
689 }
690
691 if (!last_journal_entry) {
692 fsck_err_on(!c->sb.clean, c,
693 dirty_but_no_journal_entries,
694 "no journal entries found");
695 if (clean)
696 goto use_clean;
697
698 genradix_for_each_reverse(&c->journal_entries, iter, i)
699 if (*i) {
700 last_journal_entry = &(*i)->j;
701 (*i)->ignore_blacklisted = false;
702 (*i)->ignore_not_dirty= false;
703 /*
704 * This was probably a NO_FLUSH entry,
705 * so last_seq was garbage - but we know
706 * we're only using a single journal
707 * entry, set it here:
708 */
709 (*i)->j.last_seq = (*i)->j.seq;
710 break;
711 }
712 }
713
714 ret = bch2_journal_keys_sort(c);
715 if (ret)
716 goto err;
717
718 if (c->sb.clean && last_journal_entry) {
719 ret = bch2_verify_superblock_clean(c, &clean,
720 last_journal_entry);
721 if (ret)
722 goto err;
723 }
724 } else {
725 use_clean:
726 if (!clean) {
727 bch_err(c, "no superblock clean section found");
728 ret = -BCH_ERR_fsck_repair_impossible;
729 goto err;
730
731 }
732 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
733 }
734
735 c->journal_replay_seq_start = last_seq;
736 c->journal_replay_seq_end = blacklist_seq - 1;
737
738 if (c->opts.reconstruct_alloc)
739 bch2_reconstruct_alloc(c);
740
741 zero_out_btree_mem_ptr(&c->journal_keys);
742
743 ret = journal_replay_early(c, clean);
744 if (ret)
745 goto err;
746
747 /*
748 * After an unclean shutdown, skip then next few journal sequence
749 * numbers as they may have been referenced by btree writes that
750 * happened before their corresponding journal writes - those btree
751 * writes need to be ignored, by skipping and blacklisting the next few
752 * journal sequence numbers:
753 */
754 if (!c->sb.clean)
755 journal_seq += 8;
756
757 if (blacklist_seq != journal_seq) {
758 ret = bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
759 blacklist_seq, journal_seq) ?:
760 bch2_journal_seq_blacklist_add(c,
761 blacklist_seq, journal_seq);
762 if (ret) {
763 bch_err_msg(c, ret, "error creating new journal seq blacklist entry");
764 goto err;
765 }
766 }
767
768 ret = bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
769 journal_seq, last_seq, blacklist_seq - 1) ?:
770 bch2_fs_journal_start(&c->journal, journal_seq);
771 if (ret)
772 goto err;
773
774 /*
775 * Skip past versions that might have possibly been used (as nonces),
776 * but hadn't had their pointers written:
777 */
778 if (c->sb.encryption_type && !c->sb.clean)
779 atomic64_add(1 << 16, &c->key_version);
780
781 ret = read_btree_roots(c);
782 if (ret)
783 goto err;
784
785 ret = bch2_run_recovery_passes(c);
786 if (ret)
787 goto err;
788
789 clear_bit(BCH_FS_fsck_running, &c->flags);
790
791 /* fsync if we fixed errors */
792 if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
793 bch2_journal_flush_all_pins(&c->journal);
794 bch2_journal_meta(&c->journal);
795 }
796
797 /* If we fixed errors, verify that fs is actually clean now: */
798 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
799 test_bit(BCH_FS_errors_fixed, &c->flags) &&
800 !test_bit(BCH_FS_errors_not_fixed, &c->flags) &&
801 !test_bit(BCH_FS_error, &c->flags)) {
802 bch2_flush_fsck_errs(c);
803
804 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
805 clear_bit(BCH_FS_errors_fixed, &c->flags);
806
807 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
808
809 ret = bch2_run_recovery_passes(c);
810 if (ret)
811 goto err;
812
813 if (test_bit(BCH_FS_errors_fixed, &c->flags) ||
814 test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
815 bch_err(c, "Second fsck run was not clean");
816 set_bit(BCH_FS_errors_not_fixed, &c->flags);
817 }
818
819 set_bit(BCH_FS_errors_fixed, &c->flags);
820 }
821
822 if (enabled_qtypes(c)) {
823 bch_verbose(c, "reading quotas");
824 ret = bch2_fs_quota_read(c);
825 if (ret)
826 goto err;
827 bch_verbose(c, "quotas done");
828 }
829
830 mutex_lock(&c->sb_lock);
831 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
832 bool write_sb = false;
833
834 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) {
835 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, le16_to_cpu(c->disk_sb.sb->version));
836 write_sb = true;
837 }
838
839 if (!test_bit(BCH_FS_error, &c->flags) &&
840 !(c->disk_sb.sb->compat[0] & cpu_to_le64(1ULL << BCH_COMPAT_alloc_info))) {
841 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
842 write_sb = true;
843 }
844
845 if (!test_bit(BCH_FS_error, &c->flags) &&
846 !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent))) {
847 memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
848 write_sb = true;
849 }
850
851 if (c->opts.fsck &&
852 !test_bit(BCH_FS_error, &c->flags) &&
853 !test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
854 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
855 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
856 write_sb = true;
857 }
858
859 if (write_sb)
860 bch2_write_super(c);
861 mutex_unlock(&c->sb_lock);
862
863 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
864 c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
865 struct bch_move_stats stats;
866
867 bch2_move_stats_init(&stats, "recovery");
868
869 struct printbuf buf = PRINTBUF;
870 bch2_version_to_text(&buf, c->sb.version_min);
871 bch_info(c, "scanning for old btree nodes: min_version %s", buf.buf);
872 printbuf_exit(&buf);
873
874 ret = bch2_fs_read_write_early(c) ?:
875 bch2_scan_old_btree_nodes(c, &stats);
876 if (ret)
877 goto err;
878 bch_info(c, "scanning for old btree nodes done");
879 }
880
881 if (c->journal_seq_blacklist_table &&
882 c->journal_seq_blacklist_table->nr > 128)
883 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
884
885 ret = 0;
886 out:
887 bch2_flush_fsck_errs(c);
888
889 if (!c->opts.retain_recovery_info) {
890 bch2_journal_keys_put_initial(c);
891 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
892 }
893 kfree(clean);
894
895 if (!ret &&
896 test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags) &&
897 !c->opts.nochanges) {
898 bch2_fs_read_write_early(c);
899 bch2_delete_dead_snapshots_async(c);
900 }
901
902 bch_err_fn(c, ret);
903 return ret;
904 err:
905 fsck_err:
906 bch2_fs_emergency_read_only(c);
907 goto out;
908 }
909
910 int bch2_fs_initialize(struct bch_fs *c)
911 {
912 struct bch_inode_unpacked root_inode, lostfound_inode;
913 struct bkey_inode_buf packed_inode;
914 struct qstr lostfound = QSTR("lost+found");
915 int ret;
916
917 bch_notice(c, "initializing new filesystem");
918
919 mutex_lock(&c->sb_lock);
920 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
921 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
922
923 bch2_check_version_downgrade(c);
924
925 if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
926 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
927 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
928 bch2_write_super(c);
929 }
930 mutex_unlock(&c->sb_lock);
931
932 c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
933 set_bit(BCH_FS_may_go_rw, &c->flags);
934
935 for (unsigned i = 0; i < BTREE_ID_NR; i++)
936 bch2_btree_root_alloc_fake(c, i, 0);
937
938 for_each_member_device(c, ca)
939 bch2_dev_usage_init(ca);
940
941 ret = bch2_fs_journal_alloc(c);
942 if (ret)
943 goto err;
944
945 /*
946 * journal_res_get() will crash if called before this has
947 * set up the journal.pin FIFO and journal.cur pointer:
948 */
949 bch2_fs_journal_start(&c->journal, 1);
950 bch2_journal_set_replay_done(&c->journal);
951
952 ret = bch2_fs_read_write_early(c);
953 if (ret)
954 goto err;
955
956 /*
957 * Write out the superblock and journal buckets, now that we can do
958 * btree updates
959 */
960 bch_verbose(c, "marking superblocks");
961 ret = bch2_trans_mark_dev_sbs(c);
962 bch_err_msg(c, ret, "marking superblocks");
963 if (ret)
964 goto err;
965
966 for_each_online_member(c, ca)
967 ca->new_fs_bucket_idx = 0;
968
969 ret = bch2_fs_freespace_init(c);
970 if (ret)
971 goto err;
972
973 ret = bch2_initialize_subvolumes(c);
974 if (ret)
975 goto err;
976
977 bch_verbose(c, "reading snapshots table");
978 ret = bch2_snapshots_read(c);
979 if (ret)
980 goto err;
981 bch_verbose(c, "reading snapshots done");
982
983 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
984 root_inode.bi_inum = BCACHEFS_ROOT_INO;
985 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
986 bch2_inode_pack(&packed_inode, &root_inode);
987 packed_inode.inode.k.p.snapshot = U32_MAX;
988
989 ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
990 bch_err_msg(c, ret, "creating root directory");
991 if (ret)
992 goto err;
993
994 bch2_inode_init_early(c, &lostfound_inode);
995
996 ret = bch2_trans_do(c, NULL, NULL, 0,
997 bch2_create_trans(trans,
998 BCACHEFS_ROOT_SUBVOL_INUM,
999 &root_inode, &lostfound_inode,
1000 &lostfound,
1001 0, 0, S_IFDIR|0700, 0,
1002 NULL, NULL, (subvol_inum) { 0 }, 0));
1003 bch_err_msg(c, ret, "creating lost+found");
1004 if (ret)
1005 goto err;
1006
1007 c->recovery_pass_done = BCH_RECOVERY_PASS_NR - 1;
1008
1009 if (enabled_qtypes(c)) {
1010 ret = bch2_fs_quota_read(c);
1011 if (ret)
1012 goto err;
1013 }
1014
1015 ret = bch2_journal_flush(&c->journal);
1016 bch_err_msg(c, ret, "writing first journal entry");
1017 if (ret)
1018 goto err;
1019
1020 mutex_lock(&c->sb_lock);
1021 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1022 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1023
1024 bch2_write_super(c);
1025 mutex_unlock(&c->sb_lock);
1026
1027 return 0;
1028 err:
1029 bch_err_fn(c, ret);
1030 return ret;
1031 }
Mirror https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git