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));
165 * When keys compare equal, oldest compares first:
167 static int journal_sort_key_cmp(const void *_l, const void *_r)
169 const struct journal_key *l = _l;
170 const struct journal_key *r = _r;
172 return cmp_int(l->btree_id, r->btree_id) ?:
173 bkey_cmp(l->k->k.p, r->k->k.p) ?:
174 cmp_int(l->journal_seq, r->journal_seq) ?:
175 cmp_int(l->journal_offset, r->journal_offset);
178 static int journal_sort_seq_cmp(const void *_l, const void *_r)
180 const struct journal_key *l = _l;
181 const struct journal_key *r = _r;
183 return cmp_int(l->journal_seq, r->journal_seq) ?:
184 cmp_int(l->btree_id, r->btree_id) ?:
185 bkey_cmp(l->k->k.p, r->k->k.p);
188 static void journal_keys_free(struct journal_keys *keys)
195 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
197 struct journal_replay *p;
198 struct jset_entry *entry;
199 struct bkey_i *k, *_n;
200 struct journal_keys keys = { NULL };
201 struct journal_key *src, *dst;
204 list_for_each_entry(p, journal_entries, list)
205 for_each_jset_key(k, _n, entry, &p->j)
208 keys.journal_seq_base =
209 le64_to_cpu(list_first_entry(journal_entries,
210 struct journal_replay,
213 keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
217 list_for_each_entry(p, journal_entries, list)
218 for_each_jset_key(k, _n, entry, &p->j)
219 keys.d[keys.nr++] = (struct journal_key) {
220 .btree_id = entry->btree_id,
222 .journal_seq = le64_to_cpu(p->j.seq) -
223 keys.journal_seq_base,
224 .journal_offset = k->_data - p->j._data,
227 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
230 while (src < keys.d + keys.nr) {
231 while (src + 1 < keys.d + keys.nr &&
232 src[0].btree_id == src[1].btree_id &&
233 !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
239 keys.nr = dst - keys.d;
244 /* journal replay: */
246 static void replay_now_at(struct journal *j, u64 seq)
248 BUG_ON(seq < j->replay_journal_seq);
249 BUG_ON(seq > j->replay_journal_seq_end);
251 while (j->replay_journal_seq < seq)
252 bch2_journal_pin_put(j, j->replay_journal_seq++);
255 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
258 struct btree_trans trans;
259 struct btree_iter *iter, *split_iter;
261 * We might cause compressed extents to be split, so we need to pass in
262 * a disk_reservation:
264 struct disk_reservation disk_res =
265 bch2_disk_reservation_init(c, 0);
266 struct bkey_i *split;
267 struct bpos atomic_end;
269 * Some extents aren't equivalent - w.r.t. what the triggers do
270 * - if they're split:
272 bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
273 k->k.type == KEY_TYPE_reflink_p;
277 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
279 bch2_trans_begin(&trans);
281 iter = bch2_trans_get_iter(&trans, btree_id,
282 bkey_start_pos(&k->k),
286 ret = bch2_btree_iter_traverse(iter);
290 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
292 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
293 ret = PTR_ERR_OR_ZERO(split);
299 bkey_cmp(atomic_end, k->k.p) < 0) {
300 ret = bch2_disk_reservation_add(c, &disk_res,
302 bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
303 BCH_DISK_RESERVATION_NOFAIL);
310 bch2_cut_front(iter->pos, split);
311 bch2_cut_back(atomic_end, split);
313 split_iter = bch2_trans_copy_iter(&trans, iter);
314 ret = PTR_ERR_OR_ZERO(split_iter);
319 * It's important that we don't go through the
320 * extent_handle_overwrites() and extent_update_to_keys() path
321 * here: journal replay is supposed to treat extents like
324 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
325 bch2_trans_update(&trans, split_iter, split, !remark
326 ? BTREE_TRIGGER_NORUN
327 : BTREE_TRIGGER_NOOVERWRITES);
329 bch2_btree_iter_set_pos(iter, split->k.p);
330 } while (bkey_cmp(iter->pos, k->k.p) < 0);
333 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
334 0, -((s64) k->k.size),
335 BTREE_TRIGGER_OVERWRITE);
340 ret = bch2_trans_commit(&trans, &disk_res, NULL,
342 BTREE_INSERT_LAZY_RW|
343 BTREE_INSERT_JOURNAL_REPLAY);
348 bch2_disk_reservation_put(c, &disk_res);
350 return bch2_trans_exit(&trans) ?: ret;
353 static int __bch2_journal_replay_key(struct btree_trans *trans,
354 enum btree_id id, struct bkey_i *k)
356 struct btree_iter *iter;
359 iter = bch2_trans_get_iter(trans, id, k->k.p, BTREE_ITER_INTENT);
361 return PTR_ERR(iter);
364 * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
365 * extent_handle_overwrites() and extent_update_to_keys() - but we don't
366 * want that here, journal replay is supposed to treat extents like
369 __bch2_btree_iter_set_pos(iter, k->k.p, false);
371 ret = bch2_btree_iter_traverse(iter) ?:
372 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
373 bch2_trans_iter_put(trans, iter);
377 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
380 return bch2_trans_do(c, NULL, NULL,
382 BTREE_INSERT_LAZY_RW|
383 BTREE_INSERT_JOURNAL_REPLAY,
384 __bch2_journal_replay_key(&trans, id, k));
387 static int bch2_journal_replay(struct bch_fs *c,
388 struct journal_keys keys)
390 struct journal *j = &c->journal;
391 struct journal_key *i;
394 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
396 for_each_journal_key(keys, i) {
397 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
399 if (i->btree_id == BTREE_ID_ALLOC)
400 ret = bch2_alloc_replay_key(c, i->k);
401 else if (i->k->k.size)
402 ret = bch2_extent_replay_key(c, i->btree_id, i->k);
404 ret = bch2_journal_replay_key(c, i->btree_id, i->k);
407 bch_err(c, "journal replay: error %d while replaying key",
415 replay_now_at(j, j->replay_journal_seq_end);
416 j->replay_journal_seq = 0;
418 bch2_journal_set_replay_done(j);
419 bch2_journal_flush_all_pins(j);
420 return bch2_journal_error(j);
423 static bool journal_empty(struct list_head *journal)
425 return list_empty(journal) ||
426 journal_entry_empty(&list_last_entry(journal,
427 struct journal_replay, list)->j);
431 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
432 struct list_head *journal)
434 struct journal_replay *i =
435 list_last_entry(journal, struct journal_replay, list);
436 u64 start_seq = le64_to_cpu(i->j.last_seq);
437 u64 end_seq = le64_to_cpu(i->j.seq);
441 list_for_each_entry(i, journal, list) {
442 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
443 "journal entries %llu-%llu missing! (replaying %llu-%llu)",
444 seq, le64_to_cpu(i->j.seq) - 1,
447 seq = le64_to_cpu(i->j.seq);
449 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
450 "found blacklisted journal entry %llu", seq);
454 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
460 /* journal replay early: */
462 static int journal_replay_entry_early(struct bch_fs *c,
463 struct jset_entry *entry)
467 switch (entry->type) {
468 case BCH_JSET_ENTRY_btree_root: {
469 struct btree_root *r;
471 if (entry->btree_id >= BTREE_ID_NR) {
472 bch_err(c, "filesystem has unknown btree type %u",
477 r = &c->btree_roots[entry->btree_id];
480 r->level = entry->level;
481 bkey_copy(&r->key, &entry->start[0]);
489 case BCH_JSET_ENTRY_usage: {
490 struct jset_entry_usage *u =
491 container_of(entry, struct jset_entry_usage, entry);
493 switch (entry->btree_id) {
494 case FS_USAGE_RESERVED:
495 if (entry->level < BCH_REPLICAS_MAX)
496 c->usage_base->persistent_reserved[entry->level] =
499 case FS_USAGE_INODES:
500 c->usage_base->nr_inodes = le64_to_cpu(u->v);
502 case FS_USAGE_KEY_VERSION:
503 atomic64_set(&c->key_version,
510 case BCH_JSET_ENTRY_data_usage: {
511 struct jset_entry_data_usage *u =
512 container_of(entry, struct jset_entry_data_usage, entry);
513 ret = bch2_replicas_set_usage(c, &u->r,
517 case BCH_JSET_ENTRY_blacklist: {
518 struct jset_entry_blacklist *bl_entry =
519 container_of(entry, struct jset_entry_blacklist, entry);
521 ret = bch2_journal_seq_blacklist_add(c,
522 le64_to_cpu(bl_entry->seq),
523 le64_to_cpu(bl_entry->seq) + 1);
526 case BCH_JSET_ENTRY_blacklist_v2: {
527 struct jset_entry_blacklist_v2 *bl_entry =
528 container_of(entry, struct jset_entry_blacklist_v2, entry);
530 ret = bch2_journal_seq_blacklist_add(c,
531 le64_to_cpu(bl_entry->start),
532 le64_to_cpu(bl_entry->end) + 1);
540 static int journal_replay_early(struct bch_fs *c,
541 struct bch_sb_field_clean *clean,
542 struct list_head *journal)
544 struct jset_entry *entry;
548 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
549 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
551 for (entry = clean->start;
552 entry != vstruct_end(&clean->field);
553 entry = vstruct_next(entry)) {
554 ret = journal_replay_entry_early(c, entry);
559 struct journal_replay *i =
560 list_last_entry(journal, struct journal_replay, list);
562 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
563 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
565 list_for_each_entry(i, journal, list)
566 vstruct_for_each(&i->j, entry) {
567 ret = journal_replay_entry_early(c, entry);
573 bch2_fs_usage_initialize(c);
578 /* sb clean section: */
580 static struct bkey_i *btree_root_find(struct bch_fs *c,
581 struct bch_sb_field_clean *clean,
583 enum btree_id id, unsigned *level)
586 struct jset_entry *entry, *start, *end;
589 start = clean->start;
590 end = vstruct_end(&clean->field);
593 end = vstruct_last(j);
596 for (entry = start; entry < end; entry = vstruct_next(entry))
597 if (entry->type == BCH_JSET_ENTRY_btree_root &&
598 entry->btree_id == id)
604 return ERR_PTR(-EINVAL);
607 *level = entry->level;
611 static int verify_superblock_clean(struct bch_fs *c,
612 struct bch_sb_field_clean **cleanp,
616 struct bch_sb_field_clean *clean = *cleanp;
619 if (!c->sb.clean || !j)
622 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
623 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
624 le64_to_cpu(clean->journal_seq),
625 le64_to_cpu(j->seq))) {
631 mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
632 "superblock read clock doesn't match journal after clean shutdown");
633 mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
634 "superblock read clock doesn't match journal after clean shutdown");
636 for (i = 0; i < BTREE_ID_NR; i++) {
637 struct bkey_i *k1, *k2;
638 unsigned l1 = 0, l2 = 0;
640 k1 = btree_root_find(c, clean, NULL, i, &l1);
641 k2 = btree_root_find(c, NULL, j, i, &l2);
646 mustfix_fsck_err_on(!k1 || !k2 ||
649 k1->k.u64s != k2->k.u64s ||
650 memcmp(k1, k2, bkey_bytes(k1)) ||
652 "superblock btree root doesn't match journal after clean shutdown");
658 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
660 struct bch_sb_field_clean *clean, *sb_clean;
663 mutex_lock(&c->sb_lock);
664 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
666 if (fsck_err_on(!sb_clean, c,
667 "superblock marked clean but clean section not present")) {
668 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
670 mutex_unlock(&c->sb_lock);
674 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
677 mutex_unlock(&c->sb_lock);
678 return ERR_PTR(-ENOMEM);
681 if (le16_to_cpu(c->disk_sb.sb->version) <
682 bcachefs_metadata_version_bkey_renumber)
683 bch2_sb_clean_renumber(clean, READ);
685 mutex_unlock(&c->sb_lock);
689 mutex_unlock(&c->sb_lock);
693 static int read_btree_roots(struct bch_fs *c)
698 for (i = 0; i < BTREE_ID_NR; i++) {
699 struct btree_root *r = &c->btree_roots[i];
704 if (i == BTREE_ID_ALLOC &&
705 c->opts.reconstruct_alloc) {
706 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
712 __fsck_err(c, i == BTREE_ID_ALLOC
713 ? FSCK_CAN_IGNORE : 0,
714 "invalid btree root %s",
716 if (i == BTREE_ID_ALLOC)
717 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
720 ret = bch2_btree_root_read(c, i, &r->key, r->level);
722 __fsck_err(c, i == BTREE_ID_ALLOC
723 ? FSCK_CAN_IGNORE : 0,
724 "error reading btree root %s",
726 if (i == BTREE_ID_ALLOC)
727 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
731 for (i = 0; i < BTREE_ID_NR; i++)
732 if (!c->btree_roots[i].b)
733 bch2_btree_root_alloc(c, i);
738 int bch2_fs_recovery(struct bch_fs *c)
740 const char *err = "cannot allocate memory";
741 struct bch_sb_field_clean *clean = NULL;
743 LIST_HEAD(journal_entries);
744 struct journal_keys journal_keys = { NULL };
745 bool wrote = false, write_sb = false;
749 clean = read_superblock_clean(c);
750 ret = PTR_ERR_OR_ZERO(clean);
755 bch_info(c, "recovering from clean shutdown, journal seq %llu",
756 le64_to_cpu(clean->journal_seq));
758 if (!c->replicas.entries) {
759 bch_info(c, "building replicas info");
760 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
763 if (!c->sb.clean || c->opts.fsck) {
766 ret = bch2_journal_read(c, &journal_entries);
770 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
771 "filesystem marked clean but journal not empty")) {
772 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
773 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
777 if (!c->sb.clean && list_empty(&journal_entries)) {
778 bch_err(c, "no journal entries found");
779 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
783 journal_keys = journal_keys_sort(&journal_entries);
784 if (!journal_keys.d) {
789 j = &list_last_entry(&journal_entries,
790 struct journal_replay, list)->j;
792 ret = verify_superblock_clean(c, &clean, j);
796 journal_seq = le64_to_cpu(j->seq) + 1;
798 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
802 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
803 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
808 ret = journal_replay_early(c, clean, &journal_entries);
813 ret = bch2_journal_seq_blacklist_add(c,
817 bch_err(c, "error creating new journal seq blacklist entry");
824 ret = bch2_blacklist_table_initialize(c);
826 if (!list_empty(&journal_entries)) {
827 ret = verify_journal_entries_not_blacklisted_or_missing(c,
833 ret = bch2_fs_journal_start(&c->journal, journal_seq,
838 ret = read_btree_roots(c);
842 bch_verbose(c, "starting alloc read");
843 err = "error reading allocation information";
844 ret = bch2_alloc_read(c, &journal_keys);
847 bch_verbose(c, "alloc read done");
849 bch_verbose(c, "starting stripes_read");
850 err = "error reading stripes";
851 ret = bch2_stripes_read(c, &journal_keys);
854 bch_verbose(c, "stripes_read done");
856 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
858 if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
859 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
861 * interior btree node updates aren't consistent with the
862 * journal; after an unclean shutdown we have to walk all
863 * pointers to metadata:
865 bch_info(c, "starting metadata mark and sweep");
866 err = "error in mark and sweep";
867 ret = bch2_gc(c, NULL, true, true);
870 bch_verbose(c, "mark and sweep done");
874 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
875 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
876 bch_info(c, "starting mark and sweep");
877 err = "error in mark and sweep";
878 ret = bch2_gc(c, &journal_keys, true, false);
881 bch_verbose(c, "mark and sweep done");
884 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
885 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
888 * Skip past versions that might have possibly been used (as nonces),
889 * but hadn't had their pointers written:
891 if (c->sb.encryption_type && !c->sb.clean)
892 atomic64_add(1 << 16, &c->key_version);
894 if (c->opts.norecovery)
897 bch_verbose(c, "starting journal replay");
898 err = "journal replay failed";
899 ret = bch2_journal_replay(c, journal_keys);
902 bch_verbose(c, "journal replay done");
904 if (!c->opts.nochanges) {
906 * note that even when filesystem was clean there might be work
907 * to do here, if we ran gc (because of fsck) which recalculated
910 bch_verbose(c, "writing allocation info");
911 err = "error writing out alloc info";
912 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
913 bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
915 bch_err(c, "error writing alloc info");
918 bch_verbose(c, "alloc write done");
920 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
924 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
925 bch_info(c, "checking inode link counts");
926 err = "error in recovery";
927 ret = bch2_fsck_inode_nlink(c);
930 bch_verbose(c, "check inodes done");
933 bch_verbose(c, "checking for deleted inodes");
934 err = "error in recovery";
935 ret = bch2_fsck_walk_inodes_only(c);
938 bch_verbose(c, "check inodes done");
943 bch_info(c, "starting fsck");
944 err = "error in fsck";
945 ret = bch2_fsck_full(c);
948 bch_verbose(c, "fsck done");
951 if (enabled_qtypes(c)) {
952 bch_verbose(c, "reading quotas");
953 ret = bch2_fs_quota_read(c);
956 bch_verbose(c, "quotas done");
959 mutex_lock(&c->sb_lock);
960 if (c->opts.version_upgrade) {
961 if (c->sb.version < bcachefs_metadata_version_new_versioning)
962 c->disk_sb.sb->version_min =
963 le16_to_cpu(bcachefs_metadata_version_min);
964 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
965 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
969 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
970 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
975 !test_bit(BCH_FS_ERROR, &c->flags)) {
976 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
977 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
983 mutex_unlock(&c->sb_lock);
985 if (c->journal_seq_blacklist_table &&
986 c->journal_seq_blacklist_table->nr > 128)
987 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
992 set_bit(BCH_FS_FSCK_DONE, &c->flags);
993 bch2_flush_fsck_errs(c);
995 journal_keys_free(&journal_keys);
996 journal_entries_free(&journal_entries);
999 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1001 bch_verbose(c, "ret %i", ret);
1005 int bch2_fs_initialize(struct bch_fs *c)
1007 struct bch_inode_unpacked root_inode, lostfound_inode;
1008 struct bkey_inode_buf packed_inode;
1009 struct qstr lostfound = QSTR("lost+found");
1010 const char *err = "cannot allocate memory";
1016 bch_notice(c, "initializing new filesystem");
1018 mutex_lock(&c->sb_lock);
1019 for_each_online_member(ca, c, i)
1020 bch2_mark_dev_superblock(c, ca, 0);
1021 mutex_unlock(&c->sb_lock);
1023 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1024 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1026 for (i = 0; i < BTREE_ID_NR; i++)
1027 bch2_btree_root_alloc(c, i);
1029 err = "unable to allocate journal buckets";
1030 for_each_online_member(ca, c, i) {
1031 ret = bch2_dev_journal_alloc(ca);
1033 percpu_ref_put(&ca->io_ref);
1039 * journal_res_get() will crash if called before this has
1040 * set up the journal.pin FIFO and journal.cur pointer:
1042 bch2_fs_journal_start(&c->journal, 1, &journal);
1043 bch2_journal_set_replay_done(&c->journal);
1045 bch2_inode_init(c, &root_inode, 0, 0,
1046 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1047 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1048 bch2_inode_pack(&packed_inode, &root_inode);
1050 err = "error creating root directory";
1051 ret = bch2_btree_insert(c, BTREE_ID_INODES,
1052 &packed_inode.inode.k_i,
1053 NULL, NULL, BTREE_INSERT_LAZY_RW);
1057 bch2_inode_init_early(c, &lostfound_inode);
1059 err = "error creating lost+found";
1060 ret = bch2_trans_do(c, NULL, NULL, 0,
1061 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1062 &root_inode, &lostfound_inode,
1064 0, 0, S_IFDIR|0700, 0,
1069 if (enabled_qtypes(c)) {
1070 ret = bch2_fs_quota_read(c);
1075 err = "error writing first journal entry";
1076 ret = bch2_journal_meta(&c->journal);
1080 mutex_lock(&c->sb_lock);
1081 c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1082 le16_to_cpu(bcachefs_metadata_version_current);
1083 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1084 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1086 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1087 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1089 bch2_write_super(c);
1090 mutex_unlock(&c->sb_lock);
1094 pr_err("Error initializing new filesystem: %s (%i)", err, ret);