1 // SPDX-License-Identifier: GPL-2.0 1
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 "disk_accounting.h"
14 #include "errcode.h"
15 #include "error.h"
16 #include "fs-common.h"
17 #include "journal_io.h"
18 #include "journal_reclaim.h"
19 #include "journal_seq_blacklist.h"
20 #include "logged_ops.h"
21 #include "move.h"
22 #include "quota.h"
23 #include "rebalance.h"
24 #include "recovery.h"
25 #include "recovery_passes.h"
26 #include "replicas.h"
27 #include "sb-clean.h"
28 #include "sb-downgrade.h"
29 #include "snapshot.h"
30 #include "super-io.h"
31
32 #include <linux/sort.h>
33 #include <linux/stat.h>
34
35 #define QSTR(n) { { { .len = strlen(n) } }, .n
36
37 void bch2_btree_lost_data(struct bch_fs *c, en
38 {
39 if (btree >= BTREE_ID_NR_MAX)
40 return;
41
42 u64 b = BIT_ULL(btree);
43
44 if (!(c->sb.btrees_lost_data & b)) {
45 bch_err(c, "flagging btree %s
46
47 mutex_lock(&c->sb_lock);
48 bch2_sb_field_get(c->disk_sb.s
49 bch2_write_super(c);
50 mutex_unlock(&c->sb_lock);
51 }
52 }
53
54 /* for -o reconstruct_alloc: */
55 static void bch2_reconstruct_alloc(struct bch_
56 {
57 bch2_journal_log_msg(c, "dropping allo
58 bch_info(c, "dropping and reconstructi
59
60 mutex_lock(&c->sb_lock);
61 struct bch_sb_field_ext *ext = bch2_sb
62
63 __set_bit_le64(BCH_RECOVERY_PASS_STABL
64 __set_bit_le64(BCH_RECOVERY_PASS_STABL
65 __set_bit_le64(BCH_RECOVERY_PASS_STABL
66 __set_bit_le64(BCH_RECOVERY_PASS_STABL
67 __set_bit_le64(BCH_RECOVERY_PASS_STABL
68
69 __set_bit_le64(BCH_FSCK_ERR_ptr_to_mis
70 __set_bit_le64(BCH_FSCK_ERR_ptr_gen_ne
71 __set_bit_le64(BCH_FSCK_ERR_stale_dirt
72
73 __set_bit_le64(BCH_FSCK_ERR_dev_usage_
74 __set_bit_le64(BCH_FSCK_ERR_dev_usage_
75 __set_bit_le64(BCH_FSCK_ERR_dev_usage_
76
77 __set_bit_le64(BCH_FSCK_ERR_fs_usage_b
78 __set_bit_le64(BCH_FSCK_ERR_fs_usage_c
79 __set_bit_le64(BCH_FSCK_ERR_fs_usage_p
80 __set_bit_le64(BCH_FSCK_ERR_fs_usage_r
81
82 __set_bit_le64(BCH_FSCK_ERR_alloc_key_
83 __set_bit_le64(BCH_FSCK_ERR_alloc_key_
84 __set_bit_le64(BCH_FSCK_ERR_alloc_key_
85 __set_bit_le64(BCH_FSCK_ERR_alloc_key_
86 __set_bit_le64(BCH_FSCK_ERR_alloc_key_
87 __set_bit_le64(BCH_FSCK_ERR_alloc_key_
88 __set_bit_le64(BCH_FSCK_ERR_need_disca
89 __set_bit_le64(BCH_FSCK_ERR_freespace_
90 __set_bit_le64(BCH_FSCK_ERR_bucket_gen
91 __set_bit_le64(BCH_FSCK_ERR_freespace_
92 __set_bit_le64(BCH_FSCK_ERR_ptr_to_mis
93 __set_bit_le64(BCH_FSCK_ERR_lru_entry_
94 __set_bit_le64(BCH_FSCK_ERR_accounting
95 c->sb.compat &= ~(1ULL << BCH_COMPAT_a
96
97 c->opts.recovery_passes |= bch2_recove
98
99 bch2_write_super(c);
100 mutex_unlock(&c->sb_lock);
101
102 bch2_shoot_down_journal_keys(c, BTREE_
103 0, BTREE_
104 bch2_shoot_down_journal_keys(c, BTREE_
105 0, BTREE_
106 bch2_shoot_down_journal_keys(c, BTREE_
107 0, BTREE_
108 bch2_shoot_down_journal_keys(c, BTREE_
109 0, BTREE_
110 bch2_shoot_down_journal_keys(c, BTREE_
111 0, BTREE_
112 }
113
114 /*
115 * Btree node pointers have a field to stack a
116 * node; we need to zero out this field when r
117 * reading in keys from the journal:
118 */
119 static void zero_out_btree_mem_ptr(struct jour
120 {
121 darray_for_each(*keys, i)
122 if (i->k->k.type == KEY_TYPE_b
123 bkey_i_to_btree_ptr_v2
124 }
125
126 /* journal replay: */
127
128 static void replay_now_at(struct journal *j, u
129 {
130 BUG_ON(seq < j->replay_journal_seq);
131
132 seq = min(seq, j->replay_journal_seq_e
133
134 while (j->replay_journal_seq < seq)
135 bch2_journal_pin_put(j, j->rep
136 }
137
138 static int bch2_journal_replay_accounting_key(
139
140 {
141 struct btree_iter iter;
142 bch2_trans_node_iter_init(trans, &iter
143 BTREE_MAX_DE
144 BTREE_ITER_i
145 int ret = bch2_btree_iter_traverse(&it
146 if (ret)
147 goto out;
148
149 struct bkey u;
150 struct bkey_s_c old = bch2_btree_path_
151
152 /* Has this delta already been applied
153 if (bversion_cmp(old.k->bversion, k->k
154 ret = 0;
155 goto out;
156 }
157
158 struct bkey_i *new = k->k;
159 if (old.k->type == KEY_TYPE_accounting
160 new = bch2_bkey_make_mut_noupd
161 ret = PTR_ERR_OR_ZERO(new);
162 if (ret)
163 goto out;
164
165 bch2_accounting_accumulate(bke
166 bke
167 }
168
169 trans->journal_res.seq = k->journal_se
170
171 ret = bch2_trans_update(trans, &iter,
172 out:
173 bch2_trans_iter_exit(trans, &iter);
174 return ret;
175 }
176
177 static int bch2_journal_replay_key(struct btre
178 struct jour
179 {
180 struct btree_iter iter;
181 unsigned iter_flags =
182 BTREE_ITER_intent|
183 BTREE_ITER_not_extents;
184 unsigned update_flags = BTREE_TRIGGER_
185 int ret;
186
187 if (k->overwritten)
188 return 0;
189
190 trans->journal_res.seq = k->journal_se
191
192 /*
193 * BTREE_UPDATE_key_cache_reclaim disa
194 * keep the key cache coherent with th
195 * besides the allocator is doing upda
196 * coherency for non-alloc btrees, and
197 * btrees use BTREE_ITER_filter_snapsh
198 * the snapshots recovery pass runs.
199 */
200 if (!k->level && k->btree_id == BTREE_
201 iter_flags |= BTREE_ITER_cache
202 else
203 update_flags |= BTREE_UPDATE_k
204
205 bch2_trans_node_iter_init(trans, &iter
206 BTREE_MAX_DE
207 iter_flags);
208 ret = bch2_btree_iter_traverse(&iter);
209 if (ret)
210 goto out;
211
212 struct btree_path *path = btree_iter_p
213 if (unlikely(!btree_path_node(path, k-
214 bch2_trans_iter_exit(trans, &i
215 bch2_trans_node_iter_init(tran
216 BTRE
217 ret = bch2_btree_iter_traver
218 bch2_btree_increase_de
219 -BCH_ERR_transaction_r
220 goto out;
221 }
222
223 /* Must be checked with btree locked:
224 if (k->overwritten)
225 goto out;
226
227 if (k->k->k.type == KEY_TYPE_accountin
228 ret = bch2_trans_update_buffer
229 goto out;
230 }
231
232 ret = bch2_trans_update(trans, &iter,
233 out:
234 bch2_trans_iter_exit(trans, &iter);
235 return ret;
236 }
237
238 static int journal_sort_seq_cmp(const void *_l
239 {
240 const struct journal_key *l = *((const
241 const struct journal_key *r = *((const
242
243 /*
244 * Map 0 to U64_MAX, so that keys with
245 *
246 * journal_seq == 0 means that the key
247 * should be inserted last so as to av
248 */
249 return cmp_int(l->journal_seq - 1, r->
250 }
251
252 int bch2_journal_replay(struct bch_fs *c)
253 {
254 struct journal_keys *keys = &c->journa
255 DARRAY(struct journal_key *) keys_sort
256 struct journal *j = &c->journal;
257 u64 start_seq = c->journal_replay_se
258 u64 end_seq = c->journal_replay_se
259 struct btree_trans *trans = NULL;
260 bool immediate_flush = false;
261 int ret = 0;
262
263 if (keys->nr) {
264 ret = bch2_journal_log_msg(c,
265 key
266 if (ret)
267 goto err;
268 }
269
270 BUG_ON(!atomic_read(&keys->ref));
271
272 move_gap(keys, keys->nr);
273 trans = bch2_trans_get(c);
274
275 /*
276 * Replay accounting keys first: we ca
277 * flush accounting keys until we're d
278 */
279 darray_for_each(*keys, k) {
280 if (!(k->k->k.type == KEY_TYPE
281 continue;
282
283 cond_resched();
284
285 ret = commit_do(trans, NULL, N
286 BCH_TRANS_COMM
287 BCH_TRANS_COMM
288 BCH_TRANS_COMM
289 BCH_TRANS_COMM
290 BCH_WATERMARK_
291 bch2_journal_repl
292 if (bch2_fs_fatal_err_on(ret,
293 goto err;
294
295 k->overwritten = true;
296 }
297
298 set_bit(BCH_FS_accounting_replay_done,
299
300 /*
301 * First, attempt to replay keys in so
302 * efficient - better locality of btre
303 * that would cause a journal deadlock
304 */
305 darray_for_each(*keys, k) {
306 cond_resched();
307
308 /*
309 * k->allocated means the key
310 * rather it was from early re
311 */
312 if (k->allocated)
313 immediate_flush = true
314
315 /* Skip fastpath if we're low
316 ret = c->journal.watermark ? -
317 commit_do(trans, NULL,
318 BCH_TRANS_CO
319 BCH_TRANS_CO
320 BCH_TRANS_CO
321 (!k->allocat
322 bch2_journal_repl
323 BUG_ON(!ret && !k->overwritten
324 if (ret) {
325 ret = darray_push(&key
326 if (ret)
327 goto err;
328 }
329 }
330
331 bch2_trans_unlock_long(trans);
332 /*
333 * Now, replay any remaining keys in t
334 * the journal, unpinning those journa
335 */
336 sort(keys_sorted.data, keys_sorted.nr,
337 sizeof(keys_sorted.data[0]),
338 journal_sort_seq_cmp, NULL);
339
340 darray_for_each(keys_sorted, kp) {
341 cond_resched();
342
343 struct journal_key *k = *kp;
344
345 if (k->journal_seq)
346 replay_now_at(j, k->jo
347 else
348 replay_now_at(j, j->re
349
350 ret = commit_do(trans, NULL, N
351 BCH_TRANS_COMM
352 BCH_TRANS_COMM
353 (!k->allocated
354 ? BCH_TRANS_C
355 : 0),
356 bch2_journal_repl
357 bch_err_msg(c, ret, "while rep
358 bch2_btree_id_str(
359 if (ret)
360 goto err;
361
362 BUG_ON(k->btree_id != BTREE_ID
363 }
364
365 /*
366 * We need to put our btree_trans befo
367 * that will use a btree_trans interna
368 */
369 bch2_trans_put(trans);
370 trans = NULL;
371
372 if (!c->opts.retain_recovery_info &&
373 c->recovery_pass_done >= BCH_RECOV
374 bch2_journal_keys_put_initial(
375
376 replay_now_at(j, j->replay_journal_seq
377 j->replay_journal_seq = 0;
378
379 bch2_journal_set_replay_done(j);
380
381 /* if we did any repair, flush it imme
382 if (immediate_flush) {
383 bch2_journal_flush_all_pins(&c
384 ret = bch2_journal_meta(&c->jo
385 }
386
387 if (keys->nr)
388 bch2_journal_log_msg(c, "journ
389 err:
390 if (trans)
391 bch2_trans_put(trans);
392 darray_exit(&keys_sorted);
393 bch_err_fn(c, ret);
394 return ret;
395 }
396
397 /* journal replay early: */
398
399 static int journal_replay_entry_early(struct b
400 struct j
401 {
402 int ret = 0;
403
404 switch (entry->type) {
405 case BCH_JSET_ENTRY_btree_root: {
406 struct btree_root *r;
407
408 if (fsck_err_on(entry->btree_i
409 c, invalid_btr
410 "invalid btree
411 entry->btree_i
412 return 0;
413
414 while (entry->btree_id >= c->b
415 ret = darray_push(&c->
416 if (ret)
417 return ret;
418 }
419
420 r = bch2_btree_id_root(c, entr
421
422 if (entry->u64s) {
423 r->level = entry->leve
424 bkey_copy(&r->key, (st
425 r->error = 0;
426 } else {
427 r->error = -BCH_ERR_bt
428 }
429 r->alive = true;
430 break;
431 }
432 case BCH_JSET_ENTRY_usage: {
433 struct jset_entry_usage *u =
434 container_of(entry, st
435
436 switch (entry->btree_id) {
437 case BCH_FS_USAGE_key_version:
438 atomic64_set(&c->key_v
439 break;
440 }
441 break;
442 }
443 case BCH_JSET_ENTRY_blacklist: {
444 struct jset_entry_blacklist *b
445 container_of(entry, st
446
447 ret = bch2_journal_seq_blackli
448 le64_to_cpu(bl
449 le64_to_cpu(bl
450 break;
451 }
452 case BCH_JSET_ENTRY_blacklist_v2: {
453 struct jset_entry_blacklist_v2
454 container_of(entry, st
455
456 ret = bch2_journal_seq_blackli
457 le64_to_cpu(bl
458 le64_to_cpu(bl
459 break;
460 }
461 case BCH_JSET_ENTRY_clock: {
462 struct jset_entry_clock *clock
463 container_of(entry, st
464
465 atomic64_set(&c->io_clock[cloc
466 }
467 }
468 fsck_err:
469 return ret;
470 }
471
472 static int journal_replay_early(struct bch_fs
473 struct bch_sb_
474 {
475 if (clean) {
476 for (struct jset_entry *entry
477 entry != vstruct_end(&cle
478 entry = vstruct_next(entr
479 int ret = journal_repl
480 if (ret)
481 return ret;
482 }
483 } else {
484 struct genradix_iter iter;
485 struct journal_replay *i, **_i
486
487 genradix_for_each(&c->journal_
488 i = *_i;
489
490 if (journal_replay_ign
491 continue;
492
493 vstruct_for_each(&i->j
494 int ret = jour
495 if (ret)
496 return
497 }
498 }
499 }
500
501 return 0;
502 }
503
504 /* sb clean section: */
505
506 static int read_btree_roots(struct bch_fs *c)
507 {
508 int ret = 0;
509
510 for (unsigned i = 0; i < btree_id_nr_a
511 struct btree_root *r = bch2_bt
512
513 if (!r->alive)
514 continue;
515
516 if (btree_id_is_alloc(i) && c-
517 continue;
518
519 if (mustfix_fsck_err_on((ret =
520 c, btr
521 "inval
522 bch2_b
523 mustfix_fsck_err_on((ret =
524 c, btr
525 "error
526 bch2_b
527 if (btree_id_is_alloc(
528 c->opts.recove
529 c->opts.recove
530 c->opts.recove
531 c->opts.recove
532 c->opts.recove
533 c->sb.compat &
534 r->error = 0;
535 } else if (!(c->opts.r
536 bch_info(c, "w
537 c->opts.recove
538 c->opts.recove
539 }
540
541 ret = 0;
542 bch2_btree_lost_data(c
543 }
544 }
545
546 for (unsigned i = 0; i < BTREE_ID_NR;
547 struct btree_root *r = bch2_bt
548
549 if (!r->b && !r->error) {
550 r->alive = false;
551 r->level = 0;
552 bch2_btree_root_alloc_
553 }
554 }
555 fsck_err:
556 return ret;
557 }
558
559 static bool check_version_upgrade(struct bch_f
560 {
561 unsigned latest_version = bcachefs_met
562 unsigned latest_compatible = min(lates
563 bch2_
564 unsigned old_version = c->sb.version_u
565 unsigned new_version = 0;
566
567 if (old_version < bcachefs_metadata_re
568 if (c->opts.version_upgrade ==
569 latest_compatible < bcache
570 new_version = latest_v
571 else
572 new_version = latest_c
573 } else {
574 switch (c->opts.version_upgrad
575 case BCH_VERSION_UPGRADE_compa
576 new_version = latest_c
577 break;
578 case BCH_VERSION_UPGRADE_incom
579 new_version = latest_v
580 break;
581 case BCH_VERSION_UPGRADE_none:
582 new_version = min(old_
583 break;
584 }
585 }
586
587 if (new_version > old_version) {
588 struct printbuf buf = PRINTBUF
589
590 if (old_version < bcachefs_met
591 prt_str(&buf, "Version
592
593 if (old_version != c->sb.versi
594 prt_str(&buf, "Version
595 bch2_version_to_text(&
596 prt_str(&buf, " to ");
597 bch2_version_to_text(&
598 prt_str(&buf, " incomp
599 }
600
601 prt_printf(&buf, "Doing %s ver
602 BCH_VERSION_MAJOR(o
603 ? "incompatible" :
604 bch2_version_to_text(&buf, old
605 prt_str(&buf, " to ");
606 bch2_version_to_text(&buf, new
607 prt_newline(&buf);
608
609 struct bch_sb_field_ext *ext =
610 __le64 passes = ext->recovery_
611 bch2_sb_set_upgrade(c, old_ver
612 passes = ext->recovery_passes_
613
614 if (passes) {
615 prt_str(&buf, " runni
616 prt_bitflags(&buf, bch
617 bch2_reco
618 }
619
620 bch_info(c, "%s", buf.buf);
621
622 bch2_sb_upgrade(c, new_version
623
624 printbuf_exit(&buf);
625 return true;
626 }
627
628 return false;
629 }
630
631 int bch2_fs_recovery(struct bch_fs *c)
632 {
633 struct bch_sb_field_clean *clean = NUL
634 struct jset *last_journal_entry = NULL
635 u64 last_seq = 0, blacklist_seq, journ
636 int ret = 0;
637
638 if (c->sb.clean) {
639 clean = bch2_read_superblock_c
640 ret = PTR_ERR_OR_ZERO(clean);
641 if (ret)
642 goto err;
643
644 bch_info(c, "recovering from c
645 le64_to_cpu(clean->jo
646 } else {
647 bch_info(c, "recovering from u
648 }
649
650 if (!(c->sb.features & (1ULL << BCH_FE
651 bch_err(c, "feature new_extent
652 ret = -EINVAL;
653 goto err;
654 }
655
656 if (!c->sb.clean &&
657 !(c->sb.features & (1ULL << BCH_FE
658 bch_err(c, "filesystem needs r
659 ret = -EINVAL;
660 goto err;
661 }
662
663 if (c->opts.norecovery)
664 c->opts.recovery_pass_last = B
665
666 mutex_lock(&c->sb_lock);
667 struct bch_sb_field_ext *ext = bch2_sb
668 bool write_sb = false;
669
670 if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk
671 ext->recovery_passes_required[
672 cpu_to_le64(bch2_recov
673 write_sb = true;
674 }
675
676 u64 sb_passes = bch2_recovery_passes_f
677 if (sb_passes) {
678 struct printbuf buf = PRINTBUF
679 prt_str(&buf, "superblock requ
680 prt_bitflags(&buf, bch2_recove
681 bch_info(c, "%s", buf.buf);
682 printbuf_exit(&buf);
683 }
684
685 if (bch2_check_version_downgrade(c)) {
686 struct printbuf buf = PRINTBUF
687
688 prt_str(&buf, "Version downgra
689
690 __le64 passes = ext->recovery_
691 bch2_sb_set_downgrade(c,
692 BCH_VERS
693 BCH_VERS
694 passes = ext->recovery_passes_
695 if (passes) {
696 prt_str(&buf, "\n run
697 prt_bitflags(&buf, bch
698 bch2_reco
699 }
700
701 bch_info(c, "%s", buf.buf);
702 printbuf_exit(&buf);
703 write_sb = true;
704 }
705
706 if (check_version_upgrade(c))
707 write_sb = true;
708
709 c->opts.recovery_passes |= bch2_recove
710
711 if (write_sb)
712 bch2_write_super(c);
713 mutex_unlock(&c->sb_lock);
714
715 if (c->opts.fsck && IS_ENABLED(CONFIG_
716 c->opts.recovery_passes |= BIT
717
718 if (c->opts.fsck)
719 set_bit(BCH_FS_fsck_running, &
720 if (c->sb.clean)
721 set_bit(BCH_FS_clean_recovery,
722
723 ret = bch2_blacklist_table_initialize(
724 if (ret) {
725 bch_err(c, "error initializing
726 goto err;
727 }
728
729 bch2_journal_pos_from_member_info_resu
730
731 if (!c->sb.clean || c->opts.retain_rec
732 struct genradix_iter iter;
733 struct journal_replay **i;
734
735 bch_verbose(c, "starting journ
736 ret = bch2_journal_read(c, &la
737 if (ret)
738 goto err;
739
740 /*
741 * note: cmd_list_journal need
742 * it can asterisk ignored jou
743 */
744 if (c->opts.read_journal_only)
745 goto out;
746
747 genradix_for_each_reverse(&c->
748 if (!journal_replay_ig
749 last_journal_e
750 break;
751 }
752
753 if (mustfix_fsck_err_on(c->sb.
754 last_j
755 !journ
756 clean_but_jour
757 "filesystem ma
758 c->sb.compat &= ~(1ULL
759 SET_BCH_SB_CLEAN(c->di
760 c->sb.clean = false;
761 }
762
763 if (!last_journal_entry) {
764 fsck_err_on(!c->sb.cle
765 dirty_but_
766 "no journa
767 if (clean)
768 goto use_clean
769
770 genradix_for_each_reve
771 if (*i) {
772 last_j
773 (*i)->
774 (*i)->
775 /*
776 * Thi
777 * so
778 * we'
779 * ent
780 */
781 (*i)->
782 break;
783 }
784 }
785
786 ret = bch2_journal_keys_sort(c
787 if (ret)
788 goto err;
789
790 if (c->sb.clean && last_journa
791 ret = bch2_verify_supe
792
793 if (ret)
794 goto err;
795 }
796 } else {
797 use_clean:
798 if (!clean) {
799 bch_err(c, "no superbl
800 ret = -BCH_ERR_fsck_re
801 goto err;
802
803 }
804 blacklist_seq = journal_seq =
805 }
806
807 c->journal_replay_seq_start = last
808 c->journal_replay_seq_end = blac
809
810 if (c->opts.reconstruct_alloc)
811 bch2_reconstruct_alloc(c);
812
813 zero_out_btree_mem_ptr(&c->journal_key
814
815 ret = journal_replay_early(c, clean);
816 if (ret)
817 goto err;
818
819 /*
820 * After an unclean shutdown, skip the
821 * numbers as they may have been refer
822 * happened before their corresponding
823 * writes need to be ignored, by skipp
824 * journal sequence numbers:
825 */
826 if (!c->sb.clean)
827 journal_seq += 8;
828
829 if (blacklist_seq != journal_seq) {
830 ret = bch2_journal_log_msg(c
831 b
832 bch2_journal_seq_black
833 blackl
834 if (ret) {
835 bch_err_msg(c, ret, "e
836 goto err;
837 }
838 }
839
840 ret = bch2_journal_log_msg(c, "start
841 journal_s
842 bch2_fs_journal_start(&c->jour
843 if (ret)
844 goto err;
845
846 /*
847 * Skip past versions that might have
848 * but hadn't had their pointers writt
849 */
850 if (c->sb.encryption_type && !c->sb.cl
851 atomic64_add(1 << 16, &c->key_
852
853 ret = read_btree_roots(c);
854 if (ret)
855 goto err;
856
857 set_bit(BCH_FS_btree_running, &c->flag
858
859 ret = bch2_sb_set_upgrade_extra(c);
860
861 ret = bch2_run_recovery_passes(c);
862 if (ret)
863 goto err;
864
865 /*
866 * Normally set by the appropriate rec
867 * indicates we're in early recovery a
868 * being applied to the journal replay
869 * multithreaded use:
870 */
871 set_bit(BCH_FS_may_go_rw, &c->flags);
872 clear_bit(BCH_FS_fsck_running, &c->fla
873
874 /* in case we don't run journal replay
875 set_bit(BCH_FS_accounting_replay_done,
876
877 /* fsync if we fixed errors */
878 if (test_bit(BCH_FS_errors_fixed, &c->
879 bch2_write_ref_tryget(c, BCH_WRITE
880 bch2_journal_flush_all_pins(&c
881 bch2_journal_meta(&c->journal)
882 bch2_write_ref_put(c, BCH_WRIT
883 }
884
885 /* If we fixed errors, verify that fs
886 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)
887 test_bit(BCH_FS_errors_fixed, &c->
888 !test_bit(BCH_FS_errors_not_fixed,
889 !test_bit(BCH_FS_error, &c->flags)
890 bch2_flush_fsck_errs(c);
891
892 bch_info(c, "Fixed errors, run
893 clear_bit(BCH_FS_errors_fixed,
894
895 c->curr_recovery_pass = BCH_RE
896
897 ret = bch2_run_recovery_passes
898 if (ret)
899 goto err;
900
901 if (test_bit(BCH_FS_errors_fix
902 test_bit(BCH_FS_errors_not
903 bch_err(c, "Second fsc
904 set_bit(BCH_FS_errors_
905 }
906
907 set_bit(BCH_FS_errors_fixed, &
908 }
909
910 if (enabled_qtypes(c)) {
911 bch_verbose(c, "reading quotas
912 ret = bch2_fs_quota_read(c);
913 if (ret)
914 goto err;
915 bch_verbose(c, "quotas done");
916 }
917
918 mutex_lock(&c->sb_lock);
919 ext = bch2_sb_field_get(c->disk_sb.sb,
920 write_sb = false;
921
922 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c-
923 SET_BCH_SB_VERSION_UPGRADE_COM
924 write_sb = true;
925 }
926
927 if (!test_bit(BCH_FS_error, &c->flags)
928 !(c->disk_sb.sb->compat[0] & cpu_t
929 c->disk_sb.sb->compat[0] |= cp
930 write_sb = true;
931 }
932
933 if (!test_bit(BCH_FS_error, &c->flags)
934 !bch2_is_zero(ext->errors_silent,
935 memset(ext->errors_silent, 0,
936 write_sb = true;
937 }
938
939 if (c->opts.fsck &&
940 !test_bit(BCH_FS_error, &c->flags)
941 c->recovery_pass_done == BCH_RECOV
942 ext->btrees_lost_data) {
943 ext->btrees_lost_data = 0;
944 write_sb = true;
945 }
946
947 if (c->opts.fsck &&
948 !test_bit(BCH_FS_error, &c->flags)
949 !test_bit(BCH_FS_errors_not_fixed,
950 SET_BCH_SB_HAS_ERRORS(c->disk_
951 SET_BCH_SB_HAS_TOPOLOGY_ERRORS
952 write_sb = true;
953 }
954
955 if (bch2_blacklist_entries_gc(c))
956 write_sb = true;
957
958 if (write_sb)
959 bch2_write_super(c);
960 mutex_unlock(&c->sb_lock);
961
962 if (!(c->sb.compat & (1ULL << BCH_COMP
963 c->sb.version_min < bcachefs_metad
964 struct bch_move_stats stats;
965
966 bch2_move_stats_init(&stats, "
967
968 struct printbuf buf = PRINTBUF
969 bch2_version_to_text(&buf, c->
970 bch_info(c, "scanning for old
971 printbuf_exit(&buf);
972
973 ret = bch2_fs_read_write_ear
974 bch2_scan_old_btree_no
975 if (ret)
976 goto err;
977 bch_info(c, "scanning for old
978 }
979
980 ret = 0;
981 out:
982 bch2_flush_fsck_errs(c);
983
984 if (!c->opts.retain_recovery_info) {
985 bch2_journal_keys_put_initial(
986 bch2_find_btree_nodes_exit(&c-
987 }
988 if (!IS_ERR(clean))
989 kfree(clean);
990
991 if (!ret &&
992 test_bit(BCH_FS_need_delete_dead_s
993 !c->opts.nochanges) {
994 bch2_fs_read_write_early(c);
995 bch2_delete_dead_snapshots_asy
996 }
997
998 bch_err_fn(c, ret);
999 return ret;
1000 err:
1001 fsck_err:
1002 bch2_fs_emergency_read_only(c);
1003 goto out;
1004 }
1005
1006 int bch2_fs_initialize(struct bch_fs *c)
1007 {
1008 struct bch_inode_unpacked root_inode,
1009 struct bkey_inode_buf packed_inode;
1010 struct qstr lostfound = QSTR("lost+fo
1011 struct bch_member *m;
1012 int ret;
1013
1014 bch_notice(c, "initializing new files
1015 set_bit(BCH_FS_new_fs, &c->flags);
1016
1017 mutex_lock(&c->sb_lock);
1018 c->disk_sb.sb->compat[0] |= cpu_to_le
1019 c->disk_sb.sb->compat[0] |= cpu_to_le
1020
1021 bch2_check_version_downgrade(c);
1022
1023 if (c->opts.version_upgrade != BCH_VE
1024 bch2_sb_upgrade(c, bcachefs_m
1025 SET_BCH_SB_VERSION_UPGRADE_CO
1026 bch2_write_super(c);
1027 }
1028
1029 for_each_member_device(c, ca) {
1030 m = bch2_members_v2_get_mut(c
1031 SET_BCH_MEMBER_FREESPACE_INIT
1032 ca->mi = bch2_mi_to_cpu(m);
1033 }
1034
1035 bch2_write_super(c);
1036 mutex_unlock(&c->sb_lock);
1037
1038 c->curr_recovery_pass = BCH_RECOVERY_
1039 set_bit(BCH_FS_btree_running, &c->fla
1040 set_bit(BCH_FS_may_go_rw, &c->flags);
1041
1042 for (unsigned i = 0; i < BTREE_ID_NR;
1043 bch2_btree_root_alloc_fake(c,
1044
1045 ret = bch2_fs_journal_alloc(c);
1046 if (ret)
1047 goto err;
1048
1049 /*
1050 * journal_res_get() will crash if ca
1051 * set up the journal.pin FIFO and jo
1052 */
1053 bch2_fs_journal_start(&c->journal, 1)
1054 set_bit(BCH_FS_accounting_replay_done
1055 bch2_journal_set_replay_done(&c->jour
1056
1057 ret = bch2_fs_read_write_early(c);
1058 if (ret)
1059 goto err;
1060
1061 for_each_member_device(c, ca) {
1062 ret = bch2_dev_usage_init(ca,
1063 if (ret) {
1064 bch2_dev_put(ca);
1065 goto err;
1066 }
1067 }
1068
1069 /*
1070 * Write out the superblock and journ
1071 * btree updates
1072 */
1073 bch_verbose(c, "marking superblocks")
1074 ret = bch2_trans_mark_dev_sbs(c);
1075 bch_err_msg(c, ret, "marking superblo
1076 if (ret)
1077 goto err;
1078
1079 for_each_online_member(c, ca)
1080 ca->new_fs_bucket_idx = 0;
1081
1082 ret = bch2_fs_freespace_init(c);
1083 if (ret)
1084 goto err;
1085
1086 ret = bch2_initialize_subvolumes(c);
1087 if (ret)
1088 goto err;
1089
1090 bch_verbose(c, "reading snapshots tab
1091 ret = bch2_snapshots_read(c);
1092 if (ret)
1093 goto err;
1094 bch_verbose(c, "reading snapshots don
1095
1096 bch2_inode_init(c, &root_inode, 0, 0,
1097 root_inode.bi_inum = BCACHEFS_RO
1098 root_inode.bi_subvol = BCACHEFS_RO
1099 bch2_inode_pack(&packed_inode, &root_
1100 packed_inode.inode.k.p.snapshot = U32
1101
1102 ret = bch2_btree_insert(c, BTREE_ID_i
1103 bch_err_msg(c, ret, "creating root di
1104 if (ret)
1105 goto err;
1106
1107 bch2_inode_init_early(c, &lostfound_i
1108
1109 ret = bch2_trans_commit_do(c, NULL, N
1110 bch2_create_trans(trans,
1111 BCACHEFS_RO
1112 &root_inode
1113 &lostfound,
1114 0, 0, S_IFD
1115 NULL, NULL,
1116 bch_err_msg(c, ret, "creating lost+fo
1117 if (ret)
1118 goto err;
1119
1120 c->recovery_pass_done = BCH_RECOVERY_
1121
1122 if (enabled_qtypes(c)) {
1123 ret = bch2_fs_quota_read(c);
1124 if (ret)
1125 goto err;
1126 }
1127
1128 ret = bch2_journal_flush(&c->journal)
1129 bch_err_msg(c, ret, "writing first jo
1130 if (ret)
1131 goto err;
1132
1133 mutex_lock(&c->sb_lock);
1134 SET_BCH_SB_INITIALIZED(c->disk_sb.sb,
1135 SET_BCH_SB_CLEAN(c->disk_sb.sb, false
1136
1137 bch2_write_super(c);
1138 mutex_unlock(&c->sb_lock);
1139
1140 return 0;
1141 err:
1142 bch_err_fn(c, ret);
1143 return ret;
1144 }
1145
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.