1 // SPDX-License-Identifier: GPL-2.0
5 #include "alloc_background.h"
7 #include "btree_update.h"
8 #include "btree_update_interior.h"
14 #include "fs-common.h"
16 #include "journal_io.h"
17 #include "journal_reclaim.h"
18 #include "journal_seq_blacklist.h"
23 #include "subvolume.h"
26 #include <linux/sort.h>
27 #include <linux/stat.h>
29 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
31 /* for -o reconstruct_alloc: */
32 static void drop_alloc_keys(struct journal_keys *keys)
36 for (src = 0, dst = 0; src < keys->nr; src++)
37 if (keys->d[src].btree_id != BTREE_ID_alloc)
38 keys->d[dst++] = keys->d[src];
44 * Btree node pointers have a field to stack a pointer to the in memory btree
45 * node; we need to zero out this field when reading in btree nodes, or when
46 * reading in keys from the journal:
48 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
50 struct journal_key *i;
52 for (i = keys->d; i < keys->d + keys->nr; i++)
53 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
54 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
57 /* iterate over keys read from the journal: */
59 static int __journal_key_cmp(enum btree_id l_btree_id,
62 const struct journal_key *r)
64 return (cmp_int(l_btree_id, r->btree_id) ?:
65 cmp_int(l_level, r->level) ?:
66 bpos_cmp(l_pos, r->k->k.p));
69 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
71 return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
74 size_t bch2_journal_key_search(struct journal_keys *journal_keys,
75 enum btree_id id, unsigned level,
78 size_t l = 0, r = journal_keys->nr, m;
81 m = l + ((r - l) >> 1);
82 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
88 BUG_ON(l < journal_keys->nr &&
89 __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
92 __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
97 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
99 struct bkey_i *n = iter->keys->d[idx].k;
100 struct btree_and_journal_iter *biter =
101 container_of(iter, struct btree_and_journal_iter, journal);
103 if (iter->idx > idx ||
106 bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
110 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
111 unsigned level, struct bkey_i *k)
113 struct journal_key n = {
119 * Ensure these keys are done last by journal replay, to unblock
122 .journal_seq = U32_MAX,
124 struct journal_keys *keys = &c->journal_keys;
125 struct journal_iter *iter;
126 size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
128 BUG_ON(test_bit(BCH_FS_RW, &c->flags));
130 if (idx < keys->nr &&
131 journal_key_cmp(&n, &keys->d[idx]) == 0) {
132 if (keys->d[idx].allocated)
133 kfree(keys->d[idx].k);
138 if (keys->nr == keys->size) {
139 struct journal_keys new_keys = {
141 .size = keys->size * 2,
142 .journal_seq_base = keys->journal_seq_base,
145 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
147 bch_err(c, "%s: error allocating new key array (size %zu)",
148 __func__, new_keys.size);
152 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
157 array_insert_item(keys->d, keys->nr, idx, n);
159 list_for_each_entry(iter, &c->journal_iters, list)
160 journal_iter_fix(c, iter, idx);
166 * Can only be used from the recovery thread while we're still RO - can't be
167 * used once we've got RW, as journal_keys is at that point used by multiple
170 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
171 unsigned level, struct bkey_i *k)
176 n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
181 ret = bch2_journal_key_insert_take(c, id, level, n);
187 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
188 unsigned level, struct bpos pos)
190 struct bkey_i whiteout;
192 bkey_init(&whiteout.k);
195 return bch2_journal_key_insert(c, id, level, &whiteout);
198 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
199 unsigned level, struct bpos pos)
201 struct journal_keys *keys = &c->journal_keys;
202 size_t idx = bch2_journal_key_search(keys, btree, level, pos);
204 if (idx < keys->nr &&
205 keys->d[idx].btree_id == btree &&
206 keys->d[idx].level == level &&
207 !bpos_cmp(keys->d[idx].k->k.p, pos))
208 keys->d[idx].overwritten = true;
211 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
213 struct journal_key *k = iter->keys->d + iter->idx;
215 while (k < iter->keys->d + iter->keys->nr &&
216 k->btree_id == iter->btree_id &&
217 k->level == iter->level) {
222 k = iter->keys->d + iter->idx;
228 static void bch2_journal_iter_advance(struct journal_iter *iter)
230 if (iter->idx < iter->keys->nr)
234 static void bch2_journal_iter_exit(struct journal_iter *iter)
236 list_del(&iter->list);
239 static void bch2_journal_iter_init(struct bch_fs *c,
240 struct journal_iter *iter,
241 enum btree_id id, unsigned level,
246 iter->keys = &c->journal_keys;
247 iter->idx = bch2_journal_key_search(&c->journal_keys, id, level, pos);
250 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
252 return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
253 iter->b, &iter->unpacked);
256 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
258 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
261 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
263 switch (iter->last) {
267 bch2_journal_iter_advance_btree(iter);
270 bch2_journal_iter_advance(&iter->journal);
277 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
282 struct bkey_s_c btree_k =
283 bch2_journal_iter_peek_btree(iter);
284 struct bkey_s_c journal_k =
285 bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
287 if (btree_k.k && journal_k.k) {
288 int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
291 bch2_journal_iter_advance_btree(iter);
293 iter->last = cmp < 0 ? btree : journal;
294 } else if (btree_k.k) {
296 } else if (journal_k.k) {
297 iter->last = journal;
300 return bkey_s_c_null;
303 ret = iter->last == journal ? journal_k : btree_k;
306 bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
307 iter->journal.idx = iter->journal.keys->nr;
309 return bkey_s_c_null;
312 if (!bkey_deleted(ret.k))
315 bch2_btree_and_journal_iter_advance(iter);
321 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
323 bch2_btree_and_journal_iter_advance(iter);
325 return bch2_btree_and_journal_iter_peek(iter);
328 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
330 bch2_journal_iter_exit(&iter->journal);
333 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
336 struct btree_node_iter node_iter,
339 memset(iter, 0, sizeof(*iter));
342 iter->node_iter = node_iter;
343 bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
344 INIT_LIST_HEAD(&iter->journal.list);
348 * this version is used by btree_gc before filesystem has gone RW and
349 * multithreaded, so uses the journal_iters list:
351 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
355 struct btree_node_iter node_iter;
357 bch2_btree_node_iter_init_from_start(&node_iter, b);
358 __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
359 list_add(&iter->journal.list, &c->journal_iters);
362 /* sort and dedup all keys in the journal: */
364 void bch2_journal_entries_free(struct list_head *list)
367 while (!list_empty(list)) {
368 struct journal_replay *i =
369 list_first_entry(list, struct journal_replay, list);
371 kvpfree(i, offsetof(struct journal_replay, j) +
372 vstruct_bytes(&i->j));
377 * When keys compare equal, oldest compares first:
379 static int journal_sort_key_cmp(const void *_l, const void *_r)
381 const struct journal_key *l = _l;
382 const struct journal_key *r = _r;
384 return journal_key_cmp(l, r) ?:
385 cmp_int(l->journal_seq, r->journal_seq) ?:
386 cmp_int(l->journal_offset, r->journal_offset);
389 void bch2_journal_keys_free(struct journal_keys *keys)
391 struct journal_key *i;
393 for (i = keys->d; i < keys->d + keys->nr; i++)
402 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
404 struct journal_replay *i;
405 struct jset_entry *entry;
406 struct bkey_i *k, *_n;
407 struct journal_keys keys = { NULL };
408 struct journal_key *src, *dst;
411 if (list_empty(journal_entries))
414 list_for_each_entry(i, journal_entries, list) {
418 if (!keys.journal_seq_base)
419 keys.journal_seq_base = le64_to_cpu(i->j.seq);
421 for_each_jset_key(k, _n, entry, &i->j)
425 keys.size = roundup_pow_of_two(nr_keys);
427 keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
431 list_for_each_entry(i, journal_entries, list) {
435 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
437 for_each_jset_key(k, _n, entry, &i->j)
438 keys.d[keys.nr++] = (struct journal_key) {
439 .btree_id = entry->btree_id,
440 .level = entry->level,
442 .journal_seq = le64_to_cpu(i->j.seq) -
443 keys.journal_seq_base,
444 .journal_offset = k->_data - i->j._data,
448 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
451 while (src < keys.d + keys.nr) {
452 while (src + 1 < keys.d + keys.nr &&
453 src[0].btree_id == src[1].btree_id &&
454 src[0].level == src[1].level &&
455 !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
461 keys.nr = dst - keys.d;
466 /* journal replay: */
468 static void replay_now_at(struct journal *j, u64 seq)
470 BUG_ON(seq < j->replay_journal_seq);
471 BUG_ON(seq > j->replay_journal_seq_end);
473 while (j->replay_journal_seq < seq)
474 bch2_journal_pin_put(j, j->replay_journal_seq++);
477 static int bch2_journal_replay_key(struct btree_trans *trans,
478 struct journal_key *k)
480 struct btree_iter iter;
481 unsigned iter_flags =
483 BTREE_ITER_NOT_EXTENTS;
486 if (!k->level && k->btree_id == BTREE_ID_alloc)
487 iter_flags |= BTREE_ITER_CACHED;
489 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
490 BTREE_MAX_DEPTH, k->level,
492 ret = bch2_btree_iter_traverse(&iter);
496 /* Must be checked with btree locked: */
500 ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
502 bch2_trans_iter_exit(trans, &iter);
506 static int journal_sort_seq_cmp(const void *_l, const void *_r)
508 const struct journal_key *l = *((const struct journal_key **)_l);
509 const struct journal_key *r = *((const struct journal_key **)_r);
511 return cmp_int(l->journal_seq, r->journal_seq);
514 static int bch2_journal_replay(struct bch_fs *c)
516 struct journal_keys *keys = &c->journal_keys;
517 struct journal_key **keys_sorted, *k;
518 struct journal *j = &c->journal;
522 keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
526 for (i = 0; i < keys->nr; i++)
527 keys_sorted[i] = &keys->d[i];
529 sort(keys_sorted, keys->nr,
530 sizeof(keys_sorted[0]),
531 journal_sort_seq_cmp, NULL);
534 replay_now_at(j, keys->journal_seq_base);
536 for (i = 0; i < keys->nr; i++) {
542 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
544 ret = bch2_trans_do(c, NULL, NULL,
545 BTREE_INSERT_LAZY_RW|
547 BTREE_INSERT_JOURNAL_RESERVED|
548 (!k->allocated ? BTREE_INSERT_JOURNAL_REPLAY : 0),
549 bch2_journal_replay_key(&trans, k));
551 bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
552 ret, bch2_btree_ids[k->btree_id], k->level);
557 replay_now_at(j, j->replay_journal_seq_end);
558 j->replay_journal_seq = 0;
560 bch2_journal_set_replay_done(j);
561 bch2_journal_flush_all_pins(j);
562 ret = bch2_journal_error(j);
568 /* journal replay early: */
570 static int journal_replay_entry_early(struct bch_fs *c,
571 struct jset_entry *entry)
575 switch (entry->type) {
576 case BCH_JSET_ENTRY_btree_root: {
577 struct btree_root *r;
579 if (entry->btree_id >= BTREE_ID_NR) {
580 bch_err(c, "filesystem has unknown btree type %u",
585 r = &c->btree_roots[entry->btree_id];
588 r->level = entry->level;
589 bkey_copy(&r->key, &entry->start[0]);
597 case BCH_JSET_ENTRY_usage: {
598 struct jset_entry_usage *u =
599 container_of(entry, struct jset_entry_usage, entry);
601 switch (entry->btree_id) {
602 case BCH_FS_USAGE_reserved:
603 if (entry->level < BCH_REPLICAS_MAX)
604 c->usage_base->persistent_reserved[entry->level] =
607 case BCH_FS_USAGE_inodes:
608 c->usage_base->nr_inodes = le64_to_cpu(u->v);
610 case BCH_FS_USAGE_key_version:
611 atomic64_set(&c->key_version,
618 case BCH_JSET_ENTRY_data_usage: {
619 struct jset_entry_data_usage *u =
620 container_of(entry, struct jset_entry_data_usage, entry);
622 ret = bch2_replicas_set_usage(c, &u->r,
626 case BCH_JSET_ENTRY_dev_usage: {
627 struct jset_entry_dev_usage *u =
628 container_of(entry, struct jset_entry_dev_usage, entry);
629 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
630 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
632 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
633 ca->usage_base->buckets_unavailable = le64_to_cpu(u->buckets_unavailable);
635 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
636 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
637 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
638 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
643 case BCH_JSET_ENTRY_blacklist: {
644 struct jset_entry_blacklist *bl_entry =
645 container_of(entry, struct jset_entry_blacklist, entry);
647 ret = bch2_journal_seq_blacklist_add(c,
648 le64_to_cpu(bl_entry->seq),
649 le64_to_cpu(bl_entry->seq) + 1);
652 case BCH_JSET_ENTRY_blacklist_v2: {
653 struct jset_entry_blacklist_v2 *bl_entry =
654 container_of(entry, struct jset_entry_blacklist_v2, entry);
656 ret = bch2_journal_seq_blacklist_add(c,
657 le64_to_cpu(bl_entry->start),
658 le64_to_cpu(bl_entry->end) + 1);
661 case BCH_JSET_ENTRY_clock: {
662 struct jset_entry_clock *clock =
663 container_of(entry, struct jset_entry_clock, entry);
665 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
672 static int journal_replay_early(struct bch_fs *c,
673 struct bch_sb_field_clean *clean,
674 struct list_head *journal)
676 struct journal_replay *i;
677 struct jset_entry *entry;
681 for (entry = clean->start;
682 entry != vstruct_end(&clean->field);
683 entry = vstruct_next(entry)) {
684 ret = journal_replay_entry_early(c, entry);
689 list_for_each_entry(i, journal, list) {
693 vstruct_for_each(&i->j, entry) {
694 ret = journal_replay_entry_early(c, entry);
701 bch2_fs_usage_initialize(c);
706 /* sb clean section: */
708 static struct bkey_i *btree_root_find(struct bch_fs *c,
709 struct bch_sb_field_clean *clean,
711 enum btree_id id, unsigned *level)
714 struct jset_entry *entry, *start, *end;
717 start = clean->start;
718 end = vstruct_end(&clean->field);
721 end = vstruct_last(j);
724 for (entry = start; entry < end; entry = vstruct_next(entry))
725 if (entry->type == BCH_JSET_ENTRY_btree_root &&
726 entry->btree_id == id)
732 return ERR_PTR(-EINVAL);
735 *level = entry->level;
739 static int verify_superblock_clean(struct bch_fs *c,
740 struct bch_sb_field_clean **cleanp,
744 struct bch_sb_field_clean *clean = *cleanp;
747 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
748 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
749 le64_to_cpu(clean->journal_seq),
750 le64_to_cpu(j->seq))) {
756 for (i = 0; i < BTREE_ID_NR; i++) {
757 char buf1[200], buf2[200];
758 struct bkey_i *k1, *k2;
759 unsigned l1 = 0, l2 = 0;
761 k1 = btree_root_find(c, clean, NULL, i, &l1);
762 k2 = btree_root_find(c, NULL, j, i, &l2);
767 mustfix_fsck_err_on(!k1 || !k2 ||
770 k1->k.u64s != k2->k.u64s ||
771 memcmp(k1, k2, bkey_bytes(k1)) ||
773 "superblock btree root %u doesn't match journal after clean shutdown\n"
775 "journal: l=%u %s\n", i,
776 l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
777 l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
783 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
785 struct bch_sb_field_clean *clean, *sb_clean;
788 mutex_lock(&c->sb_lock);
789 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
791 if (fsck_err_on(!sb_clean, c,
792 "superblock marked clean but clean section not present")) {
793 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
795 mutex_unlock(&c->sb_lock);
799 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
802 mutex_unlock(&c->sb_lock);
803 return ERR_PTR(-ENOMEM);
806 ret = bch2_sb_clean_validate(c, clean, READ);
808 mutex_unlock(&c->sb_lock);
812 mutex_unlock(&c->sb_lock);
816 mutex_unlock(&c->sb_lock);
820 static int read_btree_roots(struct bch_fs *c)
825 for (i = 0; i < BTREE_ID_NR; i++) {
826 struct btree_root *r = &c->btree_roots[i];
831 if (i == BTREE_ID_alloc &&
832 c->opts.reconstruct_alloc) {
833 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
838 __fsck_err(c, i == BTREE_ID_alloc
839 ? FSCK_CAN_IGNORE : 0,
840 "invalid btree root %s",
842 if (i == BTREE_ID_alloc)
843 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
846 ret = bch2_btree_root_read(c, i, &r->key, r->level);
848 __fsck_err(c, i == BTREE_ID_alloc
849 ? FSCK_CAN_IGNORE : 0,
850 "error reading btree root %s",
852 if (i == BTREE_ID_alloc)
853 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
857 for (i = 0; i < BTREE_ID_NR; i++)
858 if (!c->btree_roots[i].b)
859 bch2_btree_root_alloc(c, i);
864 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
866 struct bkey_i_snapshot root_snapshot;
867 struct bkey_i_subvolume root_volume;
870 bkey_snapshot_init(&root_snapshot.k_i);
871 root_snapshot.k.p.offset = U32_MAX;
872 root_snapshot.v.flags = 0;
873 root_snapshot.v.parent = 0;
874 root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL;
875 root_snapshot.v.pad = 0;
876 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
878 ret = bch2_btree_insert(c, BTREE_ID_snapshots,
885 bkey_subvolume_init(&root_volume.k_i);
886 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
887 root_volume.v.flags = 0;
888 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
889 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
891 ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
900 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
902 struct btree_iter iter;
904 struct bch_inode_unpacked inode;
907 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
908 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
909 k = bch2_btree_iter_peek_slot(&iter);
914 if (!bkey_is_inode(k.k)) {
915 bch_err(trans->c, "root inode not found");
920 ret = bch2_inode_unpack(k, &inode);
923 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
925 ret = bch2_inode_write(trans, &iter, &inode);
927 bch2_trans_iter_exit(trans, &iter);
931 int bch2_fs_recovery(struct bch_fs *c)
933 const char *err = "cannot allocate memory";
934 struct bch_sb_field_clean *clean = NULL;
935 struct jset *last_journal_entry = NULL;
936 u64 blacklist_seq, journal_seq;
937 bool write_sb = false;
941 clean = read_superblock_clean(c);
942 ret = PTR_ERR_OR_ZERO(clean);
947 bch_info(c, "recovering from clean shutdown, journal seq %llu",
948 le64_to_cpu(clean->journal_seq));
950 bch_info(c, "recovering from unclean shutdown");
952 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
953 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
959 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
960 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
965 if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
966 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
971 if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
972 bch_info(c, "alloc_v2 feature bit not set, fsck required");
974 c->opts.fix_errors = FSCK_OPT_YES;
977 if (!c->replicas.entries ||
978 c->opts.rebuild_replicas) {
979 bch_info(c, "building replicas info");
980 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
983 if (!c->opts.nochanges) {
984 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
985 bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
986 c->opts.version_upgrade = true;
988 c->opts.fix_errors = FSCK_OPT_YES;
989 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
990 bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
991 c->opts.version_upgrade = true;
993 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
994 bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
995 c->opts.version_upgrade = true;
999 ret = bch2_blacklist_table_initialize(c);
1001 bch_err(c, "error initializing blacklist table");
1005 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1006 struct journal_replay *i;
1008 bch_verbose(c, "starting journal read");
1009 ret = bch2_journal_read(c, &c->journal_entries,
1010 &blacklist_seq, &journal_seq);
1014 list_for_each_entry_reverse(i, &c->journal_entries, list)
1016 last_journal_entry = &i->j;
1020 if (mustfix_fsck_err_on(c->sb.clean &&
1021 last_journal_entry &&
1022 !journal_entry_empty(last_journal_entry), c,
1023 "filesystem marked clean but journal not empty")) {
1024 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1025 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1026 c->sb.clean = false;
1029 if (!last_journal_entry) {
1030 fsck_err_on(!c->sb.clean, c, "no journal entries found");
1034 c->journal_keys = journal_keys_sort(&c->journal_entries);
1035 if (!c->journal_keys.d) {
1040 if (c->sb.clean && last_journal_entry) {
1041 ret = verify_superblock_clean(c, &clean,
1042 last_journal_entry);
1049 bch_err(c, "no superblock clean section found");
1050 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1054 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1057 if (c->opts.reconstruct_alloc) {
1058 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1059 drop_alloc_keys(&c->journal_keys);
1062 zero_out_btree_mem_ptr(&c->journal_keys);
1064 ret = journal_replay_early(c, clean, &c->journal_entries);
1069 * After an unclean shutdown, skip then next few journal sequence
1070 * numbers as they may have been referenced by btree writes that
1071 * happened before their corresponding journal writes - those btree
1072 * writes need to be ignored, by skipping and blacklisting the next few
1073 * journal sequence numbers:
1078 if (blacklist_seq != journal_seq) {
1079 ret = bch2_journal_seq_blacklist_add(c,
1080 blacklist_seq, journal_seq);
1082 bch_err(c, "error creating new journal seq blacklist entry");
1087 ret = bch2_fs_journal_start(&c->journal, journal_seq,
1088 &c->journal_entries);
1092 ret = read_btree_roots(c);
1096 bch_verbose(c, "starting alloc read");
1097 err = "error reading allocation information";
1099 down_read(&c->gc_lock);
1100 ret = bch2_alloc_read(c, false, false);
1101 up_read(&c->gc_lock);
1105 bch_verbose(c, "alloc read done");
1107 bch_verbose(c, "starting stripes_read");
1108 err = "error reading stripes";
1109 ret = bch2_stripes_read(c);
1112 bch_verbose(c, "stripes_read done");
1114 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1117 * If we're not running fsck, this ensures bch2_fsck_err() calls are
1118 * instead interpreted as bch2_inconsistent_err() calls:
1121 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1124 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1125 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1126 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1127 bool metadata_only = c->opts.norecovery;
1129 bch_info(c, "checking allocations");
1130 err = "error in mark and sweep";
1131 ret = bch2_gc(c, true, metadata_only);
1134 bch_verbose(c, "done checking allocations");
1137 bch2_stripes_heap_start(c);
1139 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1140 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1143 * Skip past versions that might have possibly been used (as nonces),
1144 * but hadn't had their pointers written:
1146 if (c->sb.encryption_type && !c->sb.clean)
1147 atomic64_add(1 << 16, &c->key_version);
1149 if (c->opts.norecovery)
1152 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1153 err = "journal replay failed";
1154 ret = bch2_journal_replay(c);
1157 if (c->opts.verbose || !c->sb.clean)
1158 bch_info(c, "journal replay done");
1160 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1163 err = "error creating root snapshot node";
1164 ret = bch2_fs_initialize_subvolumes(c);
1169 bch_verbose(c, "reading snapshots table");
1170 err = "error reading snapshots table";
1171 ret = bch2_fs_snapshots_start(c);
1174 bch_verbose(c, "reading snapshots done");
1176 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1177 /* set bi_subvol on root inode */
1178 err = "error upgrade root inode for subvolumes";
1179 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1180 bch2_fs_upgrade_for_subvolumes(&trans));
1186 bch_info(c, "starting fsck");
1187 err = "error in fsck";
1188 ret = bch2_fsck_full(c);
1191 bch_verbose(c, "fsck done");
1192 } else if (!c->sb.clean) {
1193 bch_verbose(c, "checking for deleted inodes");
1194 err = "error in recovery";
1195 ret = bch2_fsck_walk_inodes_only(c);
1198 bch_verbose(c, "check inodes done");
1201 if (enabled_qtypes(c)) {
1202 bch_verbose(c, "reading quotas");
1203 ret = bch2_fs_quota_read(c);
1206 bch_verbose(c, "quotas done");
1209 mutex_lock(&c->sb_lock);
1210 if (c->opts.version_upgrade) {
1211 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1212 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1216 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1217 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1222 !test_bit(BCH_FS_ERROR, &c->flags) &&
1223 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1224 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1225 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1230 bch2_write_super(c);
1231 mutex_unlock(&c->sb_lock);
1233 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1234 !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1235 le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1236 struct bch_move_stats stats;
1238 bch_move_stats_init(&stats, "recovery");
1240 bch_info(c, "scanning for old btree nodes");
1241 ret = bch2_fs_read_write(c);
1245 ret = bch2_scan_old_btree_nodes(c, &stats);
1248 bch_info(c, "scanning for old btree nodes done");
1251 if (c->journal_seq_blacklist_table &&
1252 c->journal_seq_blacklist_table->nr > 128)
1253 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1257 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1258 bch2_flush_fsck_errs(c);
1260 if (!c->opts.keep_journal) {
1261 bch2_journal_keys_free(&c->journal_keys);
1262 bch2_journal_entries_free(&c->journal_entries);
1266 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1268 bch_verbose(c, "ret %i", ret);
1272 bch2_fs_emergency_read_only(c);
1276 int bch2_fs_initialize(struct bch_fs *c)
1278 struct bch_inode_unpacked root_inode, lostfound_inode;
1279 struct bkey_inode_buf packed_inode;
1280 struct qstr lostfound = QSTR("lost+found");
1281 const char *err = "cannot allocate memory";
1287 bch_notice(c, "initializing new filesystem");
1289 mutex_lock(&c->sb_lock);
1290 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1291 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1293 if (c->opts.version_upgrade) {
1294 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1295 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1296 bch2_write_super(c);
1298 mutex_unlock(&c->sb_lock);
1300 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1301 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1302 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1304 for (i = 0; i < BTREE_ID_NR; i++)
1305 bch2_btree_root_alloc(c, i);
1307 err = "unable to allocate journal buckets";
1308 for_each_online_member(ca, c, i) {
1309 ret = bch2_dev_journal_alloc(ca);
1311 percpu_ref_put(&ca->io_ref);
1317 * journal_res_get() will crash if called before this has
1318 * set up the journal.pin FIFO and journal.cur pointer:
1320 bch2_fs_journal_start(&c->journal, 1, &journal);
1321 bch2_journal_set_replay_done(&c->journal);
1323 err = "error going read-write";
1324 ret = bch2_fs_read_write_early(c);
1329 * Write out the superblock and journal buckets, now that we can do
1332 err = "error marking superblock and journal";
1333 for_each_member_device(ca, c, i) {
1334 ret = bch2_trans_mark_dev_sb(c, ca);
1336 percpu_ref_put(&ca->ref);
1340 ca->new_fs_bucket_idx = 0;
1343 err = "error creating root snapshot node";
1344 ret = bch2_fs_initialize_subvolumes(c);
1348 bch_verbose(c, "reading snapshots table");
1349 err = "error reading snapshots table";
1350 ret = bch2_fs_snapshots_start(c);
1353 bch_verbose(c, "reading snapshots done");
1355 bch2_inode_init(c, &root_inode, 0, 0,
1356 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1357 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1358 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1359 bch2_inode_pack(c, &packed_inode, &root_inode);
1360 packed_inode.inode.k.p.snapshot = U32_MAX;
1362 err = "error creating root directory";
1363 ret = bch2_btree_insert(c, BTREE_ID_inodes,
1364 &packed_inode.inode.k_i,
1369 bch2_inode_init_early(c, &lostfound_inode);
1371 err = "error creating lost+found";
1372 ret = bch2_trans_do(c, NULL, NULL, 0,
1373 bch2_create_trans(&trans,
1374 BCACHEFS_ROOT_SUBVOL_INUM,
1375 &root_inode, &lostfound_inode,
1377 0, 0, S_IFDIR|0700, 0,
1378 NULL, NULL, (subvol_inum) { 0 }, 0));
1380 bch_err(c, "error creating lost+found");
1384 if (enabled_qtypes(c)) {
1385 ret = bch2_fs_quota_read(c);
1390 err = "error writing first journal entry";
1391 ret = bch2_journal_flush(&c->journal);
1395 mutex_lock(&c->sb_lock);
1396 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1397 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1399 bch2_write_super(c);
1400 mutex_unlock(&c->sb_lock);
1404 pr_err("Error initializing new filesystem: %s (%i)", err, ret);