1 // SPDX-License-Identifier: GPL-2.0
4 #include "alloc_background.h"
6 #include "btree_update.h"
7 #include "btree_update_interior.h"
13 #include "fs-common.h"
15 #include "journal_io.h"
16 #include "journal_reclaim.h"
17 #include "journal_seq_blacklist.h"
23 #include <linux/sort.h>
24 #include <linux/stat.h>
26 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
28 /* iterate over keys read from the journal: */
30 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
33 if (iter->k->btree_id == iter->btree_id)
34 return bkey_i_to_s_c(iter->k->k);
37 if (iter->k == iter->keys->d + iter->keys->nr)
44 struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
50 if (iter->k == iter->keys->d + iter->keys->nr)
53 return bch2_journal_iter_peek(iter);
56 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
62 bch2_btree_iter_next(iter->btree);
65 bch2_journal_iter_next(&iter->journal);
72 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
77 struct bkey_s_c btree_k = bch2_btree_iter_peek(iter->btree);
78 struct bkey_s_c journal_k = bch2_journal_iter_peek(&iter->journal);
80 if (btree_k.k && journal_k.k) {
81 int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
84 bch2_btree_iter_next(iter->btree);
86 iter->last = cmp < 0 ? btree : journal;
87 } else if (btree_k.k) {
89 } else if (journal_k.k) {
96 ret = iter->last == journal ? journal_k : btree_k;
97 if (!bkey_deleted(ret.k))
100 bch2_btree_and_journal_iter_advance(iter);
106 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
108 bch2_btree_and_journal_iter_advance(iter);
110 return bch2_btree_and_journal_iter_peek(iter);
113 struct journal_key *journal_key_search(struct journal_keys *journal_keys,
114 enum btree_id id, struct bpos pos)
116 size_t l = 0, r = journal_keys->nr, m;
119 m = l + ((r - l) >> 1);
120 if ((cmp_int(id, journal_keys->d[m].btree_id) ?:
121 bkey_cmp(pos, journal_keys->d[m].k->k.p)) > 0)
127 BUG_ON(l < journal_keys->nr &&
128 (cmp_int(id, journal_keys->d[l].btree_id) ?:
129 bkey_cmp(pos, journal_keys->d[l].k->k.p)) > 0);
132 (cmp_int(id, journal_keys->d[l - 1].btree_id) ?:
133 bkey_cmp(pos, journal_keys->d[l - 1].k->k.p)) <= 0);
135 return l < journal_keys->nr ? journal_keys->d + l : NULL;
138 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
139 struct btree_trans *trans,
140 struct journal_keys *journal_keys,
141 enum btree_id id, struct bpos pos)
143 iter->journal.keys = journal_keys;
144 iter->journal.k = journal_key_search(journal_keys, id, pos);
145 iter->journal.btree_id = id;
147 iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
150 /* sort and dedup all keys in the journal: */
152 static void journal_entries_free(struct list_head *list)
155 while (!list_empty(list)) {
156 struct journal_replay *i =
157 list_first_entry(list, struct journal_replay, list);
159 kvpfree(i, offsetof(struct journal_replay, j) +
160 vstruct_bytes(&i->j));
164 static int journal_sort_key_cmp(const void *_l, const void *_r)
166 const struct journal_key *l = _l;
167 const struct journal_key *r = _r;
169 return cmp_int(l->btree_id, r->btree_id) ?:
170 bkey_cmp(l->pos, r->pos) ?:
171 cmp_int(l->journal_seq, r->journal_seq) ?:
172 cmp_int(l->journal_offset, r->journal_offset);
175 static int journal_sort_seq_cmp(const void *_l, const void *_r)
177 const struct journal_key *l = _l;
178 const struct journal_key *r = _r;
180 return cmp_int(l->journal_seq, r->journal_seq) ?:
181 cmp_int(l->btree_id, r->btree_id) ?:
182 bkey_cmp(l->pos, r->pos);
185 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
187 while (i + 1 < keys->d + keys->nr &&
188 journal_sort_key_cmp(i, i + 1) > 0) {
194 static void journal_keys_free(struct journal_keys *keys)
196 struct journal_key *i;
198 for_each_journal_key(*keys, i)
206 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
208 struct journal_replay *p;
209 struct jset_entry *entry;
210 struct bkey_i *k, *_n;
211 struct journal_keys keys = { NULL }, keys_deduped = { NULL };
212 struct journal_key *i;
215 list_for_each_entry(p, journal_entries, list)
216 for_each_jset_key(k, _n, entry, &p->j)
219 keys.journal_seq_base = keys_deduped.journal_seq_base =
220 le64_to_cpu(list_first_entry(journal_entries,
221 struct journal_replay,
224 keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
228 keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
232 list_for_each_entry(p, journal_entries, list)
233 for_each_jset_key(k, _n, entry, &p->j)
234 keys.d[keys.nr++] = (struct journal_key) {
235 .btree_id = entry->btree_id,
236 .pos = bkey_start_pos(&k->k),
238 .journal_seq = le64_to_cpu(p->j.seq) -
239 keys.journal_seq_base,
240 .journal_offset = k->_data - p->j._data,
243 sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
246 while (i < keys.d + keys.nr) {
247 if (i + 1 < keys.d + keys.nr &&
248 i[0].btree_id == i[1].btree_id &&
249 !bkey_cmp(i[0].pos, i[1].pos)) {
250 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
253 bch2_cut_front(i[1].k->k.p, i[0].k);
254 i[0].pos = i[1].k->k.p;
255 journal_keys_sift(&keys, i);
260 if (i + 1 < keys.d + keys.nr &&
261 i[0].btree_id == i[1].btree_id &&
262 bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
263 if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
264 cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
265 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
266 bch2_cut_back(bkey_start_pos(&i[1].k->k), i[0].k);
268 struct bkey_i *split =
269 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
274 bkey_copy(split, i[0].k);
275 bch2_cut_back(bkey_start_pos(&i[1].k->k), split);
276 keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
277 .btree_id = i[0].btree_id,
279 .pos = bkey_start_pos(&split->k),
281 .journal_seq = i[0].journal_seq,
282 .journal_offset = i[0].journal_offset,
285 bch2_cut_front(i[1].k->k.p, i[0].k);
286 i[0].pos = i[1].k->k.p;
287 journal_keys_sift(&keys, i);
291 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
296 bch2_cut_front(i[0].k->k.p, i[1].k);
297 i[1].pos = i[0].k->k.p;
298 journal_keys_sift(&keys, i + 1);
304 keys_deduped.d[keys_deduped.nr++] = *i++;
310 journal_keys_free(&keys_deduped);
312 return (struct journal_keys) { NULL };
315 /* journal replay: */
317 static void replay_now_at(struct journal *j, u64 seq)
319 BUG_ON(seq < j->replay_journal_seq);
320 BUG_ON(seq > j->replay_journal_seq_end);
322 while (j->replay_journal_seq < seq)
323 bch2_journal_pin_put(j, j->replay_journal_seq++);
326 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
329 struct btree_trans trans;
330 struct btree_iter *iter, *split_iter;
332 * We might cause compressed extents to be split, so we need to pass in
333 * a disk_reservation:
335 struct disk_reservation disk_res =
336 bch2_disk_reservation_init(c, 0);
337 struct bkey_i *split;
338 struct bpos atomic_end;
340 * Some extents aren't equivalent - w.r.t. what the triggers do
341 * - if they're split:
343 bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
344 k->k.type == KEY_TYPE_reflink_p;
348 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
350 bch2_trans_begin(&trans);
352 iter = bch2_trans_get_iter(&trans, btree_id,
353 bkey_start_pos(&k->k),
357 ret = bch2_btree_iter_traverse(iter);
361 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
363 split_iter = bch2_trans_copy_iter(&trans, iter);
364 ret = PTR_ERR_OR_ZERO(split_iter);
368 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
369 ret = PTR_ERR_OR_ZERO(split);
375 bkey_cmp(atomic_end, k->k.p) < 0) {
376 ret = bch2_disk_reservation_add(c, &disk_res,
378 bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
379 BCH_DISK_RESERVATION_NOFAIL);
386 bch2_cut_front(split_iter->pos, split);
387 bch2_cut_back(atomic_end, split);
389 bch2_trans_update(&trans, split_iter, split, !remark
390 ? BTREE_TRIGGER_NORUN
391 : BTREE_TRIGGER_NOOVERWRITES);
392 bch2_btree_iter_set_pos(iter, split->k.p);
393 } while (bkey_cmp(iter->pos, k->k.p) < 0);
396 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
397 0, -((s64) k->k.size),
398 BTREE_TRIGGER_OVERWRITE);
403 ret = bch2_trans_commit(&trans, &disk_res, NULL,
405 BTREE_INSERT_LAZY_RW|
406 BTREE_INSERT_JOURNAL_REPLAY);
411 bch2_disk_reservation_put(c, &disk_res);
413 return bch2_trans_exit(&trans) ?: ret;
416 static int __bch2_journal_replay_key(struct btree_trans *trans,
417 enum btree_id id, struct bkey_i *k)
419 struct btree_iter *iter;
421 iter = bch2_trans_get_iter(trans, id, bkey_start_pos(&k->k),
424 return PTR_ERR(iter);
426 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
430 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
433 return bch2_trans_do(c, NULL, NULL,
435 BTREE_INSERT_LAZY_RW|
436 BTREE_INSERT_JOURNAL_REPLAY,
437 __bch2_journal_replay_key(&trans, id, k));
440 static int bch2_journal_replay(struct bch_fs *c,
441 struct journal_keys keys)
443 struct journal *j = &c->journal;
444 struct journal_key *i;
447 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
449 for_each_journal_key(keys, i) {
450 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
452 if (i->btree_id == BTREE_ID_ALLOC)
453 ret = bch2_alloc_replay_key(c, i->k);
454 else if (btree_node_type_is_extents(i->btree_id))
455 ret = bch2_extent_replay_key(c, i->btree_id, i->k);
457 ret = bch2_journal_replay_key(c, i->btree_id, i->k);
460 bch_err(c, "journal replay: error %d while replaying key",
468 replay_now_at(j, j->replay_journal_seq_end);
469 j->replay_journal_seq = 0;
471 bch2_journal_set_replay_done(j);
472 bch2_journal_flush_all_pins(j);
473 return bch2_journal_error(j);
476 static bool journal_empty(struct list_head *journal)
478 return list_empty(journal) ||
479 journal_entry_empty(&list_last_entry(journal,
480 struct journal_replay, list)->j);
484 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
485 struct list_head *journal)
487 struct journal_replay *i =
488 list_last_entry(journal, struct journal_replay, list);
489 u64 start_seq = le64_to_cpu(i->j.last_seq);
490 u64 end_seq = le64_to_cpu(i->j.seq);
494 list_for_each_entry(i, journal, list) {
495 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
496 "journal entries %llu-%llu missing! (replaying %llu-%llu)",
497 seq, le64_to_cpu(i->j.seq) - 1,
500 seq = le64_to_cpu(i->j.seq);
502 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
503 "found blacklisted journal entry %llu", seq);
507 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
513 /* journal replay early: */
515 static int journal_replay_entry_early(struct bch_fs *c,
516 struct jset_entry *entry)
520 switch (entry->type) {
521 case BCH_JSET_ENTRY_btree_root: {
522 struct btree_root *r;
524 if (entry->btree_id >= BTREE_ID_NR) {
525 bch_err(c, "filesystem has unknown btree type %u",
530 r = &c->btree_roots[entry->btree_id];
533 r->level = entry->level;
534 bkey_copy(&r->key, &entry->start[0]);
542 case BCH_JSET_ENTRY_usage: {
543 struct jset_entry_usage *u =
544 container_of(entry, struct jset_entry_usage, entry);
546 switch (entry->btree_id) {
547 case FS_USAGE_RESERVED:
548 if (entry->level < BCH_REPLICAS_MAX)
549 c->usage_base->persistent_reserved[entry->level] =
552 case FS_USAGE_INODES:
553 c->usage_base->nr_inodes = le64_to_cpu(u->v);
555 case FS_USAGE_KEY_VERSION:
556 atomic64_set(&c->key_version,
563 case BCH_JSET_ENTRY_data_usage: {
564 struct jset_entry_data_usage *u =
565 container_of(entry, struct jset_entry_data_usage, entry);
566 ret = bch2_replicas_set_usage(c, &u->r,
570 case BCH_JSET_ENTRY_blacklist: {
571 struct jset_entry_blacklist *bl_entry =
572 container_of(entry, struct jset_entry_blacklist, entry);
574 ret = bch2_journal_seq_blacklist_add(c,
575 le64_to_cpu(bl_entry->seq),
576 le64_to_cpu(bl_entry->seq) + 1);
579 case BCH_JSET_ENTRY_blacklist_v2: {
580 struct jset_entry_blacklist_v2 *bl_entry =
581 container_of(entry, struct jset_entry_blacklist_v2, entry);
583 ret = bch2_journal_seq_blacklist_add(c,
584 le64_to_cpu(bl_entry->start),
585 le64_to_cpu(bl_entry->end) + 1);
593 static int journal_replay_early(struct bch_fs *c,
594 struct bch_sb_field_clean *clean,
595 struct list_head *journal)
597 struct jset_entry *entry;
601 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
602 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
604 for (entry = clean->start;
605 entry != vstruct_end(&clean->field);
606 entry = vstruct_next(entry)) {
607 ret = journal_replay_entry_early(c, entry);
612 struct journal_replay *i =
613 list_last_entry(journal, struct journal_replay, list);
615 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
616 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
618 list_for_each_entry(i, journal, list)
619 vstruct_for_each(&i->j, entry) {
620 ret = journal_replay_entry_early(c, entry);
626 bch2_fs_usage_initialize(c);
631 /* sb clean section: */
633 static struct bkey_i *btree_root_find(struct bch_fs *c,
634 struct bch_sb_field_clean *clean,
636 enum btree_id id, unsigned *level)
639 struct jset_entry *entry, *start, *end;
642 start = clean->start;
643 end = vstruct_end(&clean->field);
646 end = vstruct_last(j);
649 for (entry = start; entry < end; entry = vstruct_next(entry))
650 if (entry->type == BCH_JSET_ENTRY_btree_root &&
651 entry->btree_id == id)
657 return ERR_PTR(-EINVAL);
660 *level = entry->level;
664 static int verify_superblock_clean(struct bch_fs *c,
665 struct bch_sb_field_clean **cleanp,
669 struct bch_sb_field_clean *clean = *cleanp;
672 if (!c->sb.clean || !j)
675 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
676 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
677 le64_to_cpu(clean->journal_seq),
678 le64_to_cpu(j->seq))) {
684 mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
685 "superblock read clock doesn't match journal after clean shutdown");
686 mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
687 "superblock read clock doesn't match journal after clean shutdown");
689 for (i = 0; i < BTREE_ID_NR; i++) {
690 struct bkey_i *k1, *k2;
691 unsigned l1 = 0, l2 = 0;
693 k1 = btree_root_find(c, clean, NULL, i, &l1);
694 k2 = btree_root_find(c, NULL, j, i, &l2);
699 mustfix_fsck_err_on(!k1 || !k2 ||
702 k1->k.u64s != k2->k.u64s ||
703 memcmp(k1, k2, bkey_bytes(k1)) ||
705 "superblock btree root doesn't match journal after clean shutdown");
711 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
713 struct bch_sb_field_clean *clean, *sb_clean;
716 mutex_lock(&c->sb_lock);
717 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
719 if (fsck_err_on(!sb_clean, c,
720 "superblock marked clean but clean section not present")) {
721 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
723 mutex_unlock(&c->sb_lock);
727 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
730 mutex_unlock(&c->sb_lock);
731 return ERR_PTR(-ENOMEM);
734 if (le16_to_cpu(c->disk_sb.sb->version) <
735 bcachefs_metadata_version_bkey_renumber)
736 bch2_sb_clean_renumber(clean, READ);
738 mutex_unlock(&c->sb_lock);
742 mutex_unlock(&c->sb_lock);
746 static int read_btree_roots(struct bch_fs *c)
751 for (i = 0; i < BTREE_ID_NR; i++) {
752 struct btree_root *r = &c->btree_roots[i];
757 if (i == BTREE_ID_ALLOC &&
758 c->opts.reconstruct_alloc) {
759 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
765 __fsck_err(c, i == BTREE_ID_ALLOC
766 ? FSCK_CAN_IGNORE : 0,
767 "invalid btree root %s",
769 if (i == BTREE_ID_ALLOC)
770 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
773 ret = bch2_btree_root_read(c, i, &r->key, r->level);
775 __fsck_err(c, i == BTREE_ID_ALLOC
776 ? FSCK_CAN_IGNORE : 0,
777 "error reading btree root %s",
779 if (i == BTREE_ID_ALLOC)
780 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
784 for (i = 0; i < BTREE_ID_NR; i++)
785 if (!c->btree_roots[i].b)
786 bch2_btree_root_alloc(c, i);
791 int bch2_fs_recovery(struct bch_fs *c)
793 const char *err = "cannot allocate memory";
794 struct bch_sb_field_clean *clean = NULL;
796 LIST_HEAD(journal_entries);
797 struct journal_keys journal_keys = { NULL };
798 bool wrote = false, write_sb = false;
802 clean = read_superblock_clean(c);
803 ret = PTR_ERR_OR_ZERO(clean);
808 bch_info(c, "recovering from clean shutdown, journal seq %llu",
809 le64_to_cpu(clean->journal_seq));
811 if (!c->replicas.entries) {
812 bch_info(c, "building replicas info");
813 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
816 if (!c->sb.clean || c->opts.fsck) {
819 ret = bch2_journal_read(c, &journal_entries);
823 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
824 "filesystem marked clean but journal not empty")) {
825 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
826 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
830 if (!c->sb.clean && list_empty(&journal_entries)) {
831 bch_err(c, "no journal entries found");
832 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
836 journal_keys = journal_keys_sort(&journal_entries);
837 if (!journal_keys.d) {
842 j = &list_last_entry(&journal_entries,
843 struct journal_replay, list)->j;
845 ret = verify_superblock_clean(c, &clean, j);
849 journal_seq = le64_to_cpu(j->seq) + 1;
851 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
854 ret = journal_replay_early(c, clean, &journal_entries);
859 ret = bch2_journal_seq_blacklist_add(c,
863 bch_err(c, "error creating new journal seq blacklist entry");
870 ret = bch2_blacklist_table_initialize(c);
872 if (!list_empty(&journal_entries)) {
873 ret = verify_journal_entries_not_blacklisted_or_missing(c,
879 ret = bch2_fs_journal_start(&c->journal, journal_seq,
884 ret = read_btree_roots(c);
888 bch_verbose(c, "starting alloc read");
889 err = "error reading allocation information";
890 ret = bch2_alloc_read(c, &journal_keys);
893 bch_verbose(c, "alloc read done");
895 bch_verbose(c, "starting stripes_read");
896 err = "error reading stripes";
897 ret = bch2_stripes_read(c, &journal_keys);
900 bch_verbose(c, "stripes_read done");
902 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
904 if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
905 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
907 * interior btree node updates aren't consistent with the
908 * journal; after an unclean shutdown we have to walk all
909 * pointers to metadata:
911 bch_info(c, "starting metadata mark and sweep");
912 err = "error in mark and sweep";
913 ret = bch2_gc(c, NULL, true, true);
916 bch_verbose(c, "mark and sweep done");
920 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
921 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
922 bch_info(c, "starting mark and sweep");
923 err = "error in mark and sweep";
924 ret = bch2_gc(c, &journal_keys, true, false);
927 bch_verbose(c, "mark and sweep done");
930 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
931 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
934 * Skip past versions that might have possibly been used (as nonces),
935 * but hadn't had their pointers written:
937 if (c->sb.encryption_type && !c->sb.clean)
938 atomic64_add(1 << 16, &c->key_version);
940 if (c->opts.norecovery)
943 bch_verbose(c, "starting journal replay");
944 err = "journal replay failed";
945 ret = bch2_journal_replay(c, journal_keys);
948 bch_verbose(c, "journal replay done");
950 if (!c->opts.nochanges) {
952 * note that even when filesystem was clean there might be work
953 * to do here, if we ran gc (because of fsck) which recalculated
956 bch_verbose(c, "writing allocation info");
957 err = "error writing out alloc info";
958 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
959 bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
961 bch_err(c, "error writing alloc info");
964 bch_verbose(c, "alloc write done");
966 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
970 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
971 bch_info(c, "checking inode link counts");
972 err = "error in recovery";
973 ret = bch2_fsck_inode_nlink(c);
976 bch_verbose(c, "check inodes done");
979 bch_verbose(c, "checking for deleted inodes");
980 err = "error in recovery";
981 ret = bch2_fsck_walk_inodes_only(c);
984 bch_verbose(c, "check inodes done");
989 bch_info(c, "starting fsck");
990 err = "error in fsck";
991 ret = bch2_fsck_full(c);
994 bch_verbose(c, "fsck done");
997 if (enabled_qtypes(c)) {
998 bch_verbose(c, "reading quotas");
999 ret = bch2_fs_quota_read(c);
1002 bch_verbose(c, "quotas done");
1005 mutex_lock(&c->sb_lock);
1006 if (c->opts.version_upgrade) {
1007 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1008 c->disk_sb.sb->version_min =
1009 le16_to_cpu(bcachefs_metadata_version_min);
1010 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1011 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
1012 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
1013 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_ptr_v2;
1017 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1018 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1023 !test_bit(BCH_FS_ERROR, &c->flags)) {
1024 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1025 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1030 bch2_write_super(c);
1031 mutex_unlock(&c->sb_lock);
1033 if (c->journal_seq_blacklist_table &&
1034 c->journal_seq_blacklist_table->nr > 128)
1035 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1040 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1041 bch2_flush_fsck_errs(c);
1043 journal_keys_free(&journal_keys);
1044 journal_entries_free(&journal_entries);
1047 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1049 bch_verbose(c, "ret %i", ret);
1053 int bch2_fs_initialize(struct bch_fs *c)
1055 struct bch_inode_unpacked root_inode, lostfound_inode;
1056 struct bkey_inode_buf packed_inode;
1057 struct qstr lostfound = QSTR("lost+found");
1058 const char *err = "cannot allocate memory";
1064 bch_notice(c, "initializing new filesystem");
1066 mutex_lock(&c->sb_lock);
1067 for_each_online_member(ca, c, i)
1068 bch2_mark_dev_superblock(c, ca, 0);
1069 mutex_unlock(&c->sb_lock);
1071 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1072 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1074 for (i = 0; i < BTREE_ID_NR; i++)
1075 bch2_btree_root_alloc(c, i);
1077 err = "unable to allocate journal buckets";
1078 for_each_online_member(ca, c, i) {
1079 ret = bch2_dev_journal_alloc(ca);
1081 percpu_ref_put(&ca->io_ref);
1087 * journal_res_get() will crash if called before this has
1088 * set up the journal.pin FIFO and journal.cur pointer:
1090 bch2_fs_journal_start(&c->journal, 1, &journal);
1091 bch2_journal_set_replay_done(&c->journal);
1093 bch2_inode_init(c, &root_inode, 0, 0,
1094 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1095 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1096 bch2_inode_pack(&packed_inode, &root_inode);
1098 err = "error creating root directory";
1099 ret = bch2_btree_insert(c, BTREE_ID_INODES,
1100 &packed_inode.inode.k_i,
1101 NULL, NULL, BTREE_INSERT_LAZY_RW);
1105 bch2_inode_init_early(c, &lostfound_inode);
1107 err = "error creating lost+found";
1108 ret = bch2_trans_do(c, NULL, NULL, 0,
1109 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1110 &root_inode, &lostfound_inode,
1112 0, 0, S_IFDIR|0700, 0,
1117 if (enabled_qtypes(c)) {
1118 ret = bch2_fs_quota_read(c);
1123 err = "error writing first journal entry";
1124 ret = bch2_journal_meta(&c->journal);
1128 mutex_lock(&c->sb_lock);
1129 c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1130 le16_to_cpu(bcachefs_metadata_version_current);
1131 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1132 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
1133 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
1135 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1136 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1138 bch2_write_super(c);
1139 mutex_unlock(&c->sb_lock);
1143 pr_err("Error initializing new filesystem: %s (%i)", err, ret);