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 "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) } }, .name = n }
36
37 void bch2_btree_lost_data(struct bch_fs *c, enum btree_id btree)
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 lost data", bch2_btree_id_str(btree));
46
47 mutex_lock(&c->sb_lock);
48 bch2_sb_field_get(c->disk_sb.sb, ext)->btrees_lost_data |= cpu_to_le64(b);
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_fs *c)
56 {
57 bch2_journal_log_msg(c, "dropping alloc info");
58 bch_info(c, "dropping and reconstructing all alloc info");
59
60 mutex_lock(&c->sb_lock);
61 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
62
63 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required);
64 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_info, ext->recovery_passes_required);
65 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_lrus, ext->recovery_passes_required);
66 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_extents_to_backpointers, ext->recovery_passes_required);
67 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_to_lru_refs, ext->recovery_passes_required);
68
69 __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_alloc_key, ext->errors_silent);
70 __set_bit_le64(BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen, ext->errors_silent);
71 __set_bit_le64(BCH_FSCK_ERR_stale_dirty_ptr, ext->errors_silent);
72
73 __set_bit_le64(BCH_FSCK_ERR_dev_usage_buckets_wrong, ext->errors_silent);
74 __set_bit_le64(BCH_FSCK_ERR_dev_usage_sectors_wrong, ext->errors_silent);
75 __set_bit_le64(BCH_FSCK_ERR_dev_usage_fragmented_wrong, ext->errors_silent);
76
77 __set_bit_le64(BCH_FSCK_ERR_fs_usage_btree_wrong, ext->errors_silent);
78 __set_bit_le64(BCH_FSCK_ERR_fs_usage_cached_wrong, ext->errors_silent);
79 __set_bit_le64(BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong, ext->errors_silent);
80 __set_bit_le64(BCH_FSCK_ERR_fs_usage_replicas_wrong, ext->errors_silent);
81
82 __set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
83 __set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
84 __set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
85 __set_bit_le64(BCH_FSCK_ERR_alloc_key_cached_sectors_wrong, ext->errors_silent);
86 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent);
87 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent);
88 __set_bit_le64(BCH_FSCK_ERR_need_discard_key_wrong, ext->errors_silent);
89 __set_bit_le64(BCH_FSCK_ERR_freespace_key_wrong, ext->errors_silent);
90 __set_bit_le64(BCH_FSCK_ERR_bucket_gens_key_wrong, ext->errors_silent);
91 __set_bit_le64(BCH_FSCK_ERR_freespace_hole_missing, ext->errors_silent);
92 __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_backpointer, ext->errors_silent);
93 __set_bit_le64(BCH_FSCK_ERR_lru_entry_bad, ext->errors_silent);
94 __set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent);
95 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
96
97 bch2_write_super(c);
98 mutex_unlock(&c->sb_lock);
99
100 c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
101
102
103 bch2_shoot_down_journal_keys(c, BTREE_ID_alloc,
104 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
105 bch2_shoot_down_journal_keys(c, BTREE_ID_backpointers,
106 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
107 bch2_shoot_down_journal_keys(c, BTREE_ID_need_discard,
108 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
109 bch2_shoot_down_journal_keys(c, BTREE_ID_freespace,
110 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
111 bch2_shoot_down_journal_keys(c, BTREE_ID_bucket_gens,
112 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
113 }
114
115 /*
116 * Btree node pointers have a field to stack a pointer to the in memory btree
117 * node; we need to zero out this field when reading in btree nodes, or when
118 * reading in keys from the journal:
119 */
120 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
121 {
122 darray_for_each(*keys, i)
123 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
124 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
125 }
126
127 /* journal replay: */
128
129 static void replay_now_at(struct journal *j, u64 seq)
130 {
131 BUG_ON(seq < j->replay_journal_seq);
132
133 seq = min(seq, j->replay_journal_seq_end);
134
135 while (j->replay_journal_seq < seq)
136 bch2_journal_pin_put(j, j->replay_journal_seq++);
137 }
138
139 static int bch2_journal_replay_accounting_key(struct btree_trans *trans,
140 struct journal_key *k)
141 {
142 struct btree_iter iter;
143 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
144 BTREE_MAX_DEPTH, k->level,
145 BTREE_ITER_intent);
146 int ret = bch2_btree_iter_traverse(&iter);
147 if (ret)
148 goto out;
149
150 struct bkey u;
151 struct bkey_s_c old = bch2_btree_path_peek_slot(btree_iter_path(trans, &iter), &u);
152
153 /* Has this delta already been applied to the btree? */
154 if (bversion_cmp(old.k->version, k->k->k.version) >= 0) {
155 ret = 0;
156 goto out;
157 }
158
159 struct bkey_i *new = k->k;
160 if (old.k->type == KEY_TYPE_accounting) {
161 new = bch2_bkey_make_mut_noupdate(trans, bkey_i_to_s_c(k->k));
162 ret = PTR_ERR_OR_ZERO(new);
163 if (ret)
164 goto out;
165
166 bch2_accounting_accumulate(bkey_i_to_accounting(new),
167 bkey_s_c_to_accounting(old));
168 }
169
170 trans->journal_res.seq = k->journal_seq;
171
172 ret = bch2_trans_update(trans, &iter, new, BTREE_TRIGGER_norun);
173 out:
174 bch2_trans_iter_exit(trans, &iter);
175 return ret;
176 }
177
178 static int bch2_journal_replay_key(struct btree_trans *trans,
179 struct journal_key *k)
180 {
181 struct btree_iter iter;
182 unsigned iter_flags =
183 BTREE_ITER_intent|
184 BTREE_ITER_not_extents;
185 unsigned update_flags = BTREE_TRIGGER_norun;
186 int ret;
187
188 if (k->overwritten)
189 return 0;
190
191 trans->journal_res.seq = k->journal_seq;
192
193 /*
194 * BTREE_UPDATE_key_cache_reclaim disables key cache lookup/update to
195 * keep the key cache coherent with the underlying btree. Nothing
196 * besides the allocator is doing updates yet so we don't need key cache
197 * coherency for non-alloc btrees, and key cache fills for snapshots
198 * btrees use BTREE_ITER_filter_snapshots, which isn't available until
199 * the snapshots recovery pass runs.
200 */
201 if (!k->level && k->btree_id == BTREE_ID_alloc)
202 iter_flags |= BTREE_ITER_cached;
203 else
204 update_flags |= BTREE_UPDATE_key_cache_reclaim;
205
206 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
207 BTREE_MAX_DEPTH, k->level,
208 iter_flags);
209 ret = bch2_btree_iter_traverse(&iter);
210 if (ret)
211 goto out;
212
213 struct btree_path *path = btree_iter_path(trans, &iter);
214 if (unlikely(!btree_path_node(path, k->level))) {
215 bch2_trans_iter_exit(trans, &iter);
216 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
217 BTREE_MAX_DEPTH, 0, iter_flags);
218 ret = bch2_btree_iter_traverse(&iter) ?:
219 bch2_btree_increase_depth(trans, iter.path, 0) ?:
220 -BCH_ERR_transaction_restart_nested;
221 goto out;
222 }
223
224 /* Must be checked with btree locked: */
225 if (k->overwritten)
226 goto out;
227
228 if (k->k->k.type == KEY_TYPE_accounting) {
229 ret = bch2_trans_update_buffered(trans, BTREE_ID_accounting, k->k);
230 goto out;
231 }
232
233 ret = bch2_trans_update(trans, &iter, k->k, update_flags);
234 out:
235 bch2_trans_iter_exit(trans, &iter);
236 return ret;
237 }
238
239 static int journal_sort_seq_cmp(const void *_l, const void *_r)
240 {
241 const struct journal_key *l = *((const struct journal_key **)_l);
242 const struct journal_key *r = *((const struct journal_key **)_r);
243
244 /*
245 * Map 0 to U64_MAX, so that keys with journal_seq === 0 come last
246 *
247 * journal_seq == 0 means that the key comes from early repair, and
248 * should be inserted last so as to avoid overflowing the journal
249 */
250 return cmp_int(l->journal_seq - 1, r->journal_seq - 1);
251 }
252
253 int bch2_journal_replay(struct bch_fs *c)
254 {
255 struct journal_keys *keys = &c->journal_keys;
256 DARRAY(struct journal_key *) keys_sorted = { 0 };
257 struct journal *j = &c->journal;
258 u64 start_seq = c->journal_replay_seq_start;
259 u64 end_seq = c->journal_replay_seq_start;
260 struct btree_trans *trans = NULL;
261 bool immediate_flush = false;
262 int ret = 0;
263
264 if (keys->nr) {
265 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
266 keys->nr, start_seq, end_seq);
267 if (ret)
268 goto err;
269 }
270
271 BUG_ON(!atomic_read(&keys->ref));
272
273 move_gap(keys, keys->nr);
274 trans = bch2_trans_get(c);
275
276 /*
277 * Replay accounting keys first: we can't allow the write buffer to
278 * flush accounting keys until we're done
279 */
280 darray_for_each(*keys, k) {
281 if (!(k->k->k.type == KEY_TYPE_accounting && !k->allocated))
282 continue;
283
284 cond_resched();
285
286 ret = commit_do(trans, NULL, NULL,
287 BCH_TRANS_COMMIT_no_enospc|
288 BCH_TRANS_COMMIT_journal_reclaim|
289 BCH_TRANS_COMMIT_skip_accounting_apply|
290 BCH_TRANS_COMMIT_no_journal_res,
291 bch2_journal_replay_accounting_key(trans, k));
292 if (bch2_fs_fatal_err_on(ret, c, "error replaying accounting; %s", bch2_err_str(ret)))
293 goto err;
294
295 k->overwritten = true;
296 }
297
298 set_bit(BCH_FS_accounting_replay_done, &c->flags);
299
300 /*
301 * First, attempt to replay keys in sorted order. This is more
302 * efficient - better locality of btree access - but some might fail if
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 wasn't read in from the journal,
310 * rather it was from early repair code
311 */
312 if (k->allocated)
313 immediate_flush = true;
314
315 /* Skip fastpath if we're low on space in the journal */
316 ret = c->journal.watermark ? -1 :
317 commit_do(trans, NULL, NULL,
318 BCH_TRANS_COMMIT_no_enospc|
319 BCH_TRANS_COMMIT_journal_reclaim|
320 BCH_TRANS_COMMIT_skip_accounting_apply|
321 (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0),
322 bch2_journal_replay_key(trans, k));
323 BUG_ON(!ret && !k->overwritten && k->k->k.type != KEY_TYPE_accounting);
324 if (ret) {
325 ret = darray_push(&keys_sorted, k);
326 if (ret)
327 goto err;
328 }
329 }
330
331 bch2_trans_unlock_long(trans);
332 /*
333 * Now, replay any remaining keys in the order in which they appear in
334 * the journal, unpinning those journal entries as we go:
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->journal_seq);
347 else
348 replay_now_at(j, j->replay_journal_seq_end);
349
350 ret = commit_do(trans, NULL, NULL,
351 BCH_TRANS_COMMIT_no_enospc|
352 BCH_TRANS_COMMIT_skip_accounting_apply|
353 (!k->allocated
354 ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim
355 : 0),
356 bch2_journal_replay_key(trans, k));
357 bch_err_msg(c, ret, "while replaying key at btree %s level %u:",
358 bch2_btree_id_str(k->btree_id), k->level);
359 if (ret)
360 goto err;
361
362 BUG_ON(k->btree_id != BTREE_ID_accounting && !k->overwritten);
363 }
364
365 /*
366 * We need to put our btree_trans before calling flush_all_pins(), since
367 * that will use a btree_trans internally
368 */
369 bch2_trans_put(trans);
370 trans = NULL;
371
372 if (!c->opts.retain_recovery_info &&
373 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay)
374 bch2_journal_keys_put_initial(c);
375
376 replay_now_at(j, j->replay_journal_seq_end);
377 j->replay_journal_seq = 0;
378
379 bch2_journal_set_replay_done(j);
380
381 /* if we did any repair, flush it immediately */
382 if (immediate_flush) {
383 bch2_journal_flush_all_pins(&c->journal);
384 ret = bch2_journal_meta(&c->journal);
385 }
386
387 if (keys->nr)
388 bch2_journal_log_msg(c, "journal replay finished");
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 bch_fs *c,
400 struct jset_entry *entry)
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_id >= BTREE_ID_NR_MAX,
409 c, invalid_btree_id,
410 "invalid btree id %u (max %u)",
411 entry->btree_id, BTREE_ID_NR_MAX))
412 return 0;
413
414 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
415 ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
416 if (ret)
417 return ret;
418 }
419
420 r = bch2_btree_id_root(c, entry->btree_id);
421
422 if (entry->u64s) {
423 r->level = entry->level;
424 bkey_copy(&r->key, (struct bkey_i *) entry->start);
425 r->error = 0;
426 } else {
427 r->error = -BCH_ERR_btree_node_read_error;
428 }
429 r->alive = true;
430 break;
431 }
432 case BCH_JSET_ENTRY_usage: {
433 struct jset_entry_usage *u =
434 container_of(entry, struct jset_entry_usage, entry);
435
436 switch (entry->btree_id) {
437 case BCH_FS_USAGE_key_version:
438 atomic64_set(&c->key_version, le64_to_cpu(u->v));
439 break;
440 }
441 break;
442 }
443 case BCH_JSET_ENTRY_blacklist: {
444 struct jset_entry_blacklist *bl_entry =
445 container_of(entry, struct jset_entry_blacklist, entry);
446
447 ret = bch2_journal_seq_blacklist_add(c,
448 le64_to_cpu(bl_entry->seq),
449 le64_to_cpu(bl_entry->seq) + 1);
450 break;
451 }
452 case BCH_JSET_ENTRY_blacklist_v2: {
453 struct jset_entry_blacklist_v2 *bl_entry =
454 container_of(entry, struct jset_entry_blacklist_v2, entry);
455
456 ret = bch2_journal_seq_blacklist_add(c,
457 le64_to_cpu(bl_entry->start),
458 le64_to_cpu(bl_entry->end) + 1);
459 break;
460 }
461 case BCH_JSET_ENTRY_clock: {
462 struct jset_entry_clock *clock =
463 container_of(entry, struct jset_entry_clock, entry);
464
465 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
466 }
467 }
468 fsck_err:
469 return ret;
470 }
471
472 static int journal_replay_early(struct bch_fs *c,
473 struct bch_sb_field_clean *clean)
474 {
475 if (clean) {
476 for (struct jset_entry *entry = clean->start;
477 entry != vstruct_end(&clean->field);
478 entry = vstruct_next(entry)) {
479 int ret = journal_replay_entry_early(c, entry);
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_entries, iter, _i) {
488 i = *_i;
489
490 if (journal_replay_ignore(i))
491 continue;
492
493 vstruct_for_each(&i->j, entry) {
494 int ret = journal_replay_entry_early(c, entry);
495 if (ret)
496 return ret;
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_alive(c); i++) {
511 struct btree_root *r = bch2_btree_id_root(c, i);
512
513 if (!r->alive)
514 continue;
515
516 if (btree_id_is_alloc(i) && c->opts.reconstruct_alloc)
517 continue;
518
519 if (mustfix_fsck_err_on((ret = r->error),
520 c, btree_root_bkey_invalid,
521 "invalid btree root %s",
522 bch2_btree_id_str(i)) ||
523 mustfix_fsck_err_on((ret = r->error = bch2_btree_root_read(c, i, &r->key, r->level)),
524 c, btree_root_read_error,
525 "error reading btree root %s l=%u: %s",
526 bch2_btree_id_str(i), r->level, bch2_err_str(ret))) {
527 if (btree_id_is_alloc(i)) {
528 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_allocations);
529 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_info);
530 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_lrus);
531 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers);
532 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_to_lru_refs);
533 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
534 r->error = 0;
535 } else if (!(c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes))) {
536 bch_info(c, "will run btree node scan");
537 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes);
538 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
539 }
540
541 ret = 0;
542 bch2_btree_lost_data(c, i);
543 }
544 }
545
546 for (unsigned i = 0; i < BTREE_ID_NR; i++) {
547 struct btree_root *r = bch2_btree_id_root(c, i);
548
549 if (!r->b && !r->error) {
550 r->alive = false;
551 r->level = 0;
552 bch2_btree_root_alloc_fake(c, i, 0);
553 }
554 }
555 fsck_err:
556 return ret;
557 }
558
559 static bool check_version_upgrade(struct bch_fs *c)
560 {
561 unsigned latest_version = bcachefs_metadata_version_current;
562 unsigned latest_compatible = min(latest_version,
563 bch2_latest_compatible_version(c->sb.version));
564 unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
565 unsigned new_version = 0;
566
567 if (old_version < bcachefs_metadata_required_upgrade_below) {
568 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
569 latest_compatible < bcachefs_metadata_required_upgrade_below)
570 new_version = latest_version;
571 else
572 new_version = latest_compatible;
573 } else {
574 switch (c->opts.version_upgrade) {
575 case BCH_VERSION_UPGRADE_compatible:
576 new_version = latest_compatible;
577 break;
578 case BCH_VERSION_UPGRADE_incompatible:
579 new_version = latest_version;
580 break;
581 case BCH_VERSION_UPGRADE_none:
582 new_version = min(old_version, latest_version);
583 break;
584 }
585 }
586
587 if (new_version > old_version) {
588 struct printbuf buf = PRINTBUF;
589
590 if (old_version < bcachefs_metadata_required_upgrade_below)
591 prt_str(&buf, "Version upgrade required:\n");
592
593 if (old_version != c->sb.version) {
594 prt_str(&buf, "Version upgrade from ");
595 bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
596 prt_str(&buf, " to ");
597 bch2_version_to_text(&buf, c->sb.version);
598 prt_str(&buf, " incomplete\n");
599 }
600
601 prt_printf(&buf, "Doing %s version upgrade from ",
602 BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
603 ? "incompatible" : "compatible");
604 bch2_version_to_text(&buf, old_version);
605 prt_str(&buf, " to ");
606 bch2_version_to_text(&buf, new_version);
607 prt_newline(&buf);
608
609 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
610 __le64 passes = ext->recovery_passes_required[0];
611 bch2_sb_set_upgrade(c, old_version, new_version);
612 passes = ext->recovery_passes_required[0] & ~passes;
613
614 if (passes) {
615 prt_str(&buf, " running recovery passes: ");
616 prt_bitflags(&buf, bch2_recovery_passes,
617 bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
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 = NULL;
634 struct jset *last_journal_entry = NULL;
635 u64 last_seq = 0, blacklist_seq, journal_seq;
636 int ret = 0;
637
638 if (c->sb.clean) {
639 clean = bch2_read_superblock_clean(c);
640 ret = PTR_ERR_OR_ZERO(clean);
641 if (ret)
642 goto err;
643
644 bch_info(c, "recovering from clean shutdown, journal seq %llu",
645 le64_to_cpu(clean->journal_seq));
646 } else {
647 bch_info(c, "recovering from unclean shutdown");
648 }
649
650 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
651 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
652 ret = -EINVAL;
653 goto err;
654 }
655
656 if (!c->sb.clean &&
657 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
658 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
659 ret = -EINVAL;
660 goto err;
661 }
662
663 if (c->opts.norecovery)
664 c->opts.recovery_pass_last = BCH_RECOVERY_PASS_journal_replay - 1;
665
666 mutex_lock(&c->sb_lock);
667 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
668 bool write_sb = false;
669
670 if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) {
671 ext->recovery_passes_required[0] |=
672 cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology)));
673 write_sb = true;
674 }
675
676 u64 sb_passes = bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
677 if (sb_passes) {
678 struct printbuf buf = PRINTBUF;
679 prt_str(&buf, "superblock requires following recovery passes to be run:\n ");
680 prt_bitflags(&buf, bch2_recovery_passes, sb_passes);
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 downgrade required:");
689
690 __le64 passes = ext->recovery_passes_required[0];
691 bch2_sb_set_downgrade(c,
692 BCH_VERSION_MINOR(bcachefs_metadata_version_current),
693 BCH_VERSION_MINOR(c->sb.version));
694 passes = ext->recovery_passes_required[0] & ~passes;
695 if (passes) {
696 prt_str(&buf, "\n running recovery passes: ");
697 prt_bitflags(&buf, bch2_recovery_passes,
698 bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
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->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
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_BCACHEFS_DEBUG))
716 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
717
718 if (c->opts.fsck)
719 set_bit(BCH_FS_fsck_running, &c->flags);
720
721 ret = bch2_blacklist_table_initialize(c);
722 if (ret) {
723 bch_err(c, "error initializing blacklist table");
724 goto err;
725 }
726
727 bch2_journal_pos_from_member_info_resume(c);
728
729 if (!c->sb.clean || c->opts.retain_recovery_info) {
730 struct genradix_iter iter;
731 struct journal_replay **i;
732
733 bch_verbose(c, "starting journal read");
734 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
735 if (ret)
736 goto err;
737
738 /*
739 * note: cmd_list_journal needs the blacklist table fully up to date so
740 * it can asterisk ignored journal entries:
741 */
742 if (c->opts.read_journal_only)
743 goto out;
744
745 genradix_for_each_reverse(&c->journal_entries, iter, i)
746 if (!journal_replay_ignore(*i)) {
747 last_journal_entry = &(*i)->j;
748 break;
749 }
750
751 if (mustfix_fsck_err_on(c->sb.clean &&
752 last_journal_entry &&
753 !journal_entry_empty(last_journal_entry), c,
754 clean_but_journal_not_empty,
755 "filesystem marked clean but journal not empty")) {
756 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
757 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
758 c->sb.clean = false;
759 }
760
761 if (!last_journal_entry) {
762 fsck_err_on(!c->sb.clean, c,
763 dirty_but_no_journal_entries,
764 "no journal entries found");
765 if (clean)
766 goto use_clean;
767
768 genradix_for_each_reverse(&c->journal_entries, iter, i)
769 if (*i) {
770 last_journal_entry = &(*i)->j;
771 (*i)->ignore_blacklisted = false;
772 (*i)->ignore_not_dirty= false;
773 /*
774 * This was probably a NO_FLUSH entry,
775 * so last_seq was garbage - but we know
776 * we're only using a single journal
777 * entry, set it here:
778 */
779 (*i)->j.last_seq = (*i)->j.seq;
780 break;
781 }
782 }
783
784 ret = bch2_journal_keys_sort(c);
785 if (ret)
786 goto err;
787
788 if (c->sb.clean && last_journal_entry) {
789 ret = bch2_verify_superblock_clean(c, &clean,
790 last_journal_entry);
791 if (ret)
792 goto err;
793 }
794 } else {
795 use_clean:
796 if (!clean) {
797 bch_err(c, "no superblock clean section found");
798 ret = -BCH_ERR_fsck_repair_impossible;
799 goto err;
800
801 }
802 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
803 }
804
805 c->journal_replay_seq_start = last_seq;
806 c->journal_replay_seq_end = blacklist_seq - 1;
807
808 if (c->opts.reconstruct_alloc)
809 bch2_reconstruct_alloc(c);
810
811 zero_out_btree_mem_ptr(&c->journal_keys);
812
813 ret = journal_replay_early(c, clean);
814 if (ret)
815 goto err;
816
817 /*
818 * After an unclean shutdown, skip then next few journal sequence
819 * numbers as they may have been referenced by btree writes that
820 * happened before their corresponding journal writes - those btree
821 * writes need to be ignored, by skipping and blacklisting the next few
822 * journal sequence numbers:
823 */
824 if (!c->sb.clean)
825 journal_seq += 8;
826
827 if (blacklist_seq != journal_seq) {
828 ret = bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
829 blacklist_seq, journal_seq) ?:
830 bch2_journal_seq_blacklist_add(c,
831 blacklist_seq, journal_seq);
832 if (ret) {
833 bch_err_msg(c, ret, "error creating new journal seq blacklist entry");
834 goto err;
835 }
836 }
837
838 ret = bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
839 journal_seq, last_seq, blacklist_seq - 1) ?:
840 bch2_fs_journal_start(&c->journal, journal_seq);
841 if (ret)
842 goto err;
843
844 /*
845 * Skip past versions that might have possibly been used (as nonces),
846 * but hadn't had their pointers written:
847 */
848 if (c->sb.encryption_type && !c->sb.clean)
849 atomic64_add(1 << 16, &c->key_version);
850
851 ret = read_btree_roots(c);
852 if (ret)
853 goto err;
854
855 set_bit(BCH_FS_btree_running, &c->flags);
856
857 ret = bch2_sb_set_upgrade_extra(c);
858
859 ret = bch2_run_recovery_passes(c);
860 if (ret)
861 goto err;
862
863 clear_bit(BCH_FS_fsck_running, &c->flags);
864
865 /* fsync if we fixed errors */
866 if (test_bit(BCH_FS_errors_fixed, &c->flags) &&
867 bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync)) {
868 bch2_journal_flush_all_pins(&c->journal);
869 bch2_journal_meta(&c->journal);
870 bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
871 }
872
873 /* If we fixed errors, verify that fs is actually clean now: */
874 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
875 test_bit(BCH_FS_errors_fixed, &c->flags) &&
876 !test_bit(BCH_FS_errors_not_fixed, &c->flags) &&
877 !test_bit(BCH_FS_error, &c->flags)) {
878 bch2_flush_fsck_errs(c);
879
880 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
881 clear_bit(BCH_FS_errors_fixed, &c->flags);
882
883 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
884
885 ret = bch2_run_recovery_passes(c);
886 if (ret)
887 goto err;
888
889 if (test_bit(BCH_FS_errors_fixed, &c->flags) ||
890 test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
891 bch_err(c, "Second fsck run was not clean");
892 set_bit(BCH_FS_errors_not_fixed, &c->flags);
893 }
894
895 set_bit(BCH_FS_errors_fixed, &c->flags);
896 }
897
898 if (enabled_qtypes(c)) {
899 bch_verbose(c, "reading quotas");
900 ret = bch2_fs_quota_read(c);
901 if (ret)
902 goto err;
903 bch_verbose(c, "quotas done");
904 }
905
906 mutex_lock(&c->sb_lock);
907 ext = bch2_sb_field_get(c->disk_sb.sb, ext);
908 write_sb = false;
909
910 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) {
911 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, le16_to_cpu(c->disk_sb.sb->version));
912 write_sb = true;
913 }
914
915 if (!test_bit(BCH_FS_error, &c->flags) &&
916 !(c->disk_sb.sb->compat[0] & cpu_to_le64(1ULL << BCH_COMPAT_alloc_info))) {
917 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
918 write_sb = true;
919 }
920
921 if (!test_bit(BCH_FS_error, &c->flags) &&
922 !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent))) {
923 memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
924 write_sb = true;
925 }
926
927 if (c->opts.fsck &&
928 !test_bit(BCH_FS_error, &c->flags) &&
929 c->recovery_pass_done == BCH_RECOVERY_PASS_NR - 1 &&
930 ext->btrees_lost_data) {
931 ext->btrees_lost_data = 0;
932 write_sb = true;
933 }
934
935 if (c->opts.fsck &&
936 !test_bit(BCH_FS_error, &c->flags) &&
937 !test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
938 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
939 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
940 write_sb = true;
941 }
942
943 if (bch2_blacklist_entries_gc(c))
944 write_sb = true;
945
946 if (write_sb)
947 bch2_write_super(c);
948 mutex_unlock(&c->sb_lock);
949
950 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
951 c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
952 struct bch_move_stats stats;
953
954 bch2_move_stats_init(&stats, "recovery");
955
956 struct printbuf buf = PRINTBUF;
957 bch2_version_to_text(&buf, c->sb.version_min);
958 bch_info(c, "scanning for old btree nodes: min_version %s", buf.buf);
959 printbuf_exit(&buf);
960
961 ret = bch2_fs_read_write_early(c) ?:
962 bch2_scan_old_btree_nodes(c, &stats);
963 if (ret)
964 goto err;
965 bch_info(c, "scanning for old btree nodes done");
966 }
967
968 ret = 0;
969 out:
970 bch2_flush_fsck_errs(c);
971
972 if (!c->opts.retain_recovery_info) {
973 bch2_journal_keys_put_initial(c);
974 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
975 }
976 if (!IS_ERR(clean))
977 kfree(clean);
978
979 if (!ret &&
980 test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags) &&
981 !c->opts.nochanges) {
982 bch2_fs_read_write_early(c);
983 bch2_delete_dead_snapshots_async(c);
984 }
985
986 bch_err_fn(c, ret);
987 return ret;
988 err:
989 fsck_err:
990 bch2_fs_emergency_read_only(c);
991 goto out;
992 }
993
994 int bch2_fs_initialize(struct bch_fs *c)
995 {
996 struct bch_inode_unpacked root_inode, lostfound_inode;
997 struct bkey_inode_buf packed_inode;
998 struct qstr lostfound = QSTR("lost+found");
999 int ret;
1000
1001 bch_notice(c, "initializing new filesystem");
1002 set_bit(BCH_FS_new_fs, &c->flags);
1003
1004 mutex_lock(&c->sb_lock);
1005 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1006 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1007
1008 bch2_check_version_downgrade(c);
1009
1010 if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
1011 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
1012 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1013 bch2_write_super(c);
1014 }
1015 mutex_unlock(&c->sb_lock);
1016
1017 c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
1018 set_bit(BCH_FS_btree_running, &c->flags);
1019 set_bit(BCH_FS_may_go_rw, &c->flags);
1020
1021 for (unsigned i = 0; i < BTREE_ID_NR; i++)
1022 bch2_btree_root_alloc_fake(c, i, 0);
1023
1024 ret = bch2_fs_journal_alloc(c);
1025 if (ret)
1026 goto err;
1027
1028 /*
1029 * journal_res_get() will crash if called before this has
1030 * set up the journal.pin FIFO and journal.cur pointer:
1031 */
1032 bch2_fs_journal_start(&c->journal, 1);
1033 set_bit(BCH_FS_accounting_replay_done, &c->flags);
1034 bch2_journal_set_replay_done(&c->journal);
1035
1036 ret = bch2_fs_read_write_early(c);
1037 if (ret)
1038 goto err;
1039
1040 for_each_member_device(c, ca) {
1041 ret = bch2_dev_usage_init(ca, false);
1042 if (ret) {
1043 bch2_dev_put(ca);
1044 goto err;
1045 }
1046 }
1047
1048 /*
1049 * Write out the superblock and journal buckets, now that we can do
1050 * btree updates
1051 */
1052 bch_verbose(c, "marking superblocks");
1053 ret = bch2_trans_mark_dev_sbs(c);
1054 bch_err_msg(c, ret, "marking superblocks");
1055 if (ret)
1056 goto err;
1057
1058 for_each_online_member(c, ca)
1059 ca->new_fs_bucket_idx = 0;
1060
1061 ret = bch2_fs_freespace_init(c);
1062 if (ret)
1063 goto err;
1064
1065 ret = bch2_initialize_subvolumes(c);
1066 if (ret)
1067 goto err;
1068
1069 bch_verbose(c, "reading snapshots table");
1070 ret = bch2_snapshots_read(c);
1071 if (ret)
1072 goto err;
1073 bch_verbose(c, "reading snapshots done");
1074
1075 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1076 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1077 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1078 bch2_inode_pack(&packed_inode, &root_inode);
1079 packed_inode.inode.k.p.snapshot = U32_MAX;
1080
1081 ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0, 0);
1082 bch_err_msg(c, ret, "creating root directory");
1083 if (ret)
1084 goto err;
1085
1086 bch2_inode_init_early(c, &lostfound_inode);
1087
1088 ret = bch2_trans_do(c, NULL, NULL, 0,
1089 bch2_create_trans(trans,
1090 BCACHEFS_ROOT_SUBVOL_INUM,
1091 &root_inode, &lostfound_inode,
1092 &lostfound,
1093 0, 0, S_IFDIR|0700, 0,
1094 NULL, NULL, (subvol_inum) { 0 }, 0));
1095 bch_err_msg(c, ret, "creating lost+found");
1096 if (ret)
1097 goto err;
1098
1099 c->recovery_pass_done = BCH_RECOVERY_PASS_NR - 1;
1100
1101 if (enabled_qtypes(c)) {
1102 ret = bch2_fs_quota_read(c);
1103 if (ret)
1104 goto err;
1105 }
1106
1107 ret = bch2_journal_flush(&c->journal);
1108 bch_err_msg(c, ret, "writing first journal entry");
1109 if (ret)
1110 goto err;
1111
1112 mutex_lock(&c->sb_lock);
1113 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1114 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1115
1116 bch2_write_super(c);
1117 mutex_unlock(&c->sb_lock);
1118
1119 return 0;
1120 err:
1121 bch_err_fn(c, ret);
1122 return ret;
1123 }
1124
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.