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 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(struct journal_key *l, struct journal_key *r)
71 return (cmp_int(l->btree_id, r->btree_id) ?:
72 cmp_int(l->level, r->level) ?:
73 bpos_cmp(l->k->k.p, r->k->k.p));
76 static size_t journal_key_search(struct journal_keys *journal_keys,
77 enum btree_id id, unsigned level,
80 size_t l = 0, r = journal_keys->nr, m;
83 m = l + ((r - l) >> 1);
84 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
90 BUG_ON(l < journal_keys->nr &&
91 __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
94 __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
99 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
101 struct bkey_i *n = iter->keys->d[idx].k;
102 struct btree_and_journal_iter *biter =
103 container_of(iter, struct btree_and_journal_iter, journal);
105 if (iter->idx > idx ||
108 bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
112 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
113 unsigned level, struct bkey_i *k)
115 struct journal_key n = {
121 struct journal_keys *keys = &c->journal_keys;
122 struct journal_iter *iter;
123 unsigned idx = journal_key_search(keys, id, level, k->k.p);
125 if (idx < keys->nr &&
126 journal_key_cmp(&n, &keys->d[idx]) == 0) {
127 if (keys->d[idx].allocated)
128 kfree(keys->d[idx].k);
133 if (keys->nr == keys->size) {
134 struct journal_keys new_keys = {
136 .size = keys->size * 2,
137 .journal_seq_base = keys->journal_seq_base,
140 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
142 bch_err(c, "%s: error allocating new key array (size %zu)",
143 __func__, new_keys.size);
147 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
152 array_insert_item(keys->d, keys->nr, idx, n);
154 list_for_each_entry(iter, &c->journal_iters, list)
155 journal_iter_fix(c, iter, idx);
160 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
161 unsigned level, struct bpos pos)
163 struct bkey_i *whiteout =
164 kmalloc(sizeof(struct bkey), GFP_KERNEL);
168 bch_err(c, "%s: error allocating new key", __func__);
172 bkey_init(&whiteout->k);
175 ret = bch2_journal_key_insert(c, id, level, whiteout);
181 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
183 struct journal_key *k = iter->idx - iter->keys->nr
184 ? iter->keys->d + iter->idx : NULL;
187 k->btree_id == iter->btree_id &&
188 k->level == iter->level)
191 iter->idx = iter->keys->nr;
195 static void bch2_journal_iter_advance(struct journal_iter *iter)
197 if (iter->idx < iter->keys->nr)
201 static void bch2_journal_iter_exit(struct journal_iter *iter)
203 list_del(&iter->list);
206 static void bch2_journal_iter_init(struct bch_fs *c,
207 struct journal_iter *iter,
208 enum btree_id id, unsigned level,
213 iter->keys = &c->journal_keys;
214 iter->idx = journal_key_search(&c->journal_keys, id, level, pos);
215 list_add(&iter->list, &c->journal_iters);
218 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
220 return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
221 iter->b, &iter->unpacked);
224 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
226 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
229 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
231 switch (iter->last) {
235 bch2_journal_iter_advance_btree(iter);
238 bch2_journal_iter_advance(&iter->journal);
245 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
250 struct bkey_s_c btree_k =
251 bch2_journal_iter_peek_btree(iter);
252 struct bkey_s_c journal_k =
253 bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
255 if (btree_k.k && journal_k.k) {
256 int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
259 bch2_journal_iter_advance_btree(iter);
261 iter->last = cmp < 0 ? btree : journal;
262 } else if (btree_k.k) {
264 } else if (journal_k.k) {
265 iter->last = journal;
268 return bkey_s_c_null;
271 ret = iter->last == journal ? journal_k : btree_k;
274 bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
275 iter->journal.idx = iter->journal.keys->nr;
277 return bkey_s_c_null;
280 if (!bkey_deleted(ret.k))
283 bch2_btree_and_journal_iter_advance(iter);
289 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
291 bch2_btree_and_journal_iter_advance(iter);
293 return bch2_btree_and_journal_iter_peek(iter);
296 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
298 bch2_journal_iter_exit(&iter->journal);
301 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
305 memset(iter, 0, sizeof(*iter));
308 bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
309 bch2_journal_iter_init(c, &iter->journal,
310 b->c.btree_id, b->c.level, b->data->min_key);
313 /* Walk btree, overlaying keys from the journal: */
315 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
316 struct btree_and_journal_iter iter)
318 unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
324 bch2_bkey_buf_init(&tmp);
327 (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
328 bch2_bkey_buf_reassemble(&tmp, c, k);
330 bch2_btree_node_prefetch(c, NULL, NULL, tmp.k,
331 b->c.btree_id, b->c.level - 1);
333 bch2_btree_and_journal_iter_advance(&iter);
337 bch2_bkey_buf_exit(&tmp, c);
340 static int bch2_btree_and_journal_walk_recurse(struct btree_trans *trans, struct btree *b,
341 enum btree_id btree_id,
342 btree_walk_key_fn key_fn)
344 struct bch_fs *c = trans->c;
345 struct btree_and_journal_iter iter;
351 bch2_bkey_buf_init(&tmp);
352 bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
354 while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
356 bch2_bkey_buf_reassemble(&tmp, c, k);
358 child = bch2_btree_node_get_noiter(c, tmp.k,
359 b->c.btree_id, b->c.level - 1,
362 ret = PTR_ERR_OR_ZERO(child);
366 btree_and_journal_iter_prefetch(c, b, iter);
368 ret = bch2_btree_and_journal_walk_recurse(trans, child,
370 six_unlock_read(&child->c.lock);
372 ret = key_fn(trans, k);
378 bch2_btree_and_journal_iter_advance(&iter);
381 bch2_btree_and_journal_iter_exit(&iter);
382 bch2_bkey_buf_exit(&tmp, c);
386 int bch2_btree_and_journal_walk(struct btree_trans *trans, enum btree_id btree_id,
387 btree_walk_key_fn key_fn)
389 struct bch_fs *c = trans->c;
390 struct btree *b = c->btree_roots[btree_id].b;
393 if (btree_node_fake(b))
396 six_lock_read(&b->c.lock, NULL, NULL);
397 ret = bch2_btree_and_journal_walk_recurse(trans, b, btree_id, key_fn);
398 six_unlock_read(&b->c.lock);
403 /* sort and dedup all keys in the journal: */
405 void bch2_journal_entries_free(struct list_head *list)
408 while (!list_empty(list)) {
409 struct journal_replay *i =
410 list_first_entry(list, struct journal_replay, list);
412 kvpfree(i, offsetof(struct journal_replay, j) +
413 vstruct_bytes(&i->j));
418 * When keys compare equal, oldest compares first:
420 static int journal_sort_key_cmp(const void *_l, const void *_r)
422 const struct journal_key *l = _l;
423 const struct journal_key *r = _r;
425 return cmp_int(l->btree_id, r->btree_id) ?:
426 cmp_int(l->level, r->level) ?:
427 bpos_cmp(l->k->k.p, r->k->k.p) ?:
428 cmp_int(l->journal_seq, r->journal_seq) ?:
429 cmp_int(l->journal_offset, r->journal_offset);
432 void bch2_journal_keys_free(struct journal_keys *keys)
434 struct journal_key *i;
436 for (i = keys->d; i < keys->d + keys->nr; i++)
445 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
447 struct journal_replay *i;
448 struct jset_entry *entry;
449 struct bkey_i *k, *_n;
450 struct journal_keys keys = { NULL };
451 struct journal_key *src, *dst;
454 if (list_empty(journal_entries))
457 list_for_each_entry(i, journal_entries, list) {
461 if (!keys.journal_seq_base)
462 keys.journal_seq_base = le64_to_cpu(i->j.seq);
464 for_each_jset_key(k, _n, entry, &i->j)
468 keys.size = roundup_pow_of_two(nr_keys);
470 keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
474 list_for_each_entry(i, journal_entries, list) {
478 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
480 for_each_jset_key(k, _n, entry, &i->j)
481 keys.d[keys.nr++] = (struct journal_key) {
482 .btree_id = entry->btree_id,
483 .level = entry->level,
485 .journal_seq = le64_to_cpu(i->j.seq) -
486 keys.journal_seq_base,
487 .journal_offset = k->_data - i->j._data,
491 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
494 while (src < keys.d + keys.nr) {
495 while (src + 1 < keys.d + keys.nr &&
496 src[0].btree_id == src[1].btree_id &&
497 src[0].level == src[1].level &&
498 !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
504 keys.nr = dst - keys.d;
509 /* journal replay: */
511 static void replay_now_at(struct journal *j, u64 seq)
513 BUG_ON(seq < j->replay_journal_seq);
514 BUG_ON(seq > j->replay_journal_seq_end);
516 while (j->replay_journal_seq < seq)
517 bch2_journal_pin_put(j, j->replay_journal_seq++);
520 static int __bch2_journal_replay_key(struct btree_trans *trans,
521 struct journal_key *k)
523 struct btree_iter iter;
524 unsigned iter_flags =
526 BTREE_ITER_NOT_EXTENTS;
529 if (!k->level && k->btree_id == BTREE_ID_alloc)
530 iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
532 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
533 BTREE_MAX_DEPTH, k->level,
535 ret = bch2_btree_iter_traverse(&iter) ?:
536 bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
537 bch2_trans_iter_exit(trans, &iter);
541 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
543 unsigned commit_flags =
544 BTREE_INSERT_LAZY_RW|
546 BTREE_INSERT_JOURNAL_RESERVED;
549 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
551 return bch2_trans_do(c, NULL, NULL, commit_flags,
552 __bch2_journal_replay_key(&trans, k));
555 static int journal_sort_seq_cmp(const void *_l, const void *_r)
557 const struct journal_key *l = _l;
558 const struct journal_key *r = _r;
560 return cmp_int(r->level, l->level) ?:
561 cmp_int(l->journal_seq, r->journal_seq) ?:
562 cmp_int(l->btree_id, r->btree_id) ?:
563 bpos_cmp(l->k->k.p, r->k->k.p);
566 static int bch2_journal_replay(struct bch_fs *c,
567 struct journal_keys keys)
569 struct journal *j = &c->journal;
570 struct journal_key *i;
574 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
577 replay_now_at(j, keys.journal_seq_base);
579 seq = j->replay_journal_seq;
582 * First replay updates to the alloc btree - these will only update the
585 for_each_journal_key(keys, i) {
588 if (!i->level && i->btree_id == BTREE_ID_alloc) {
589 j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
590 ret = bch2_journal_replay_key(c, i);
597 * Next replay updates to interior btree nodes:
599 for_each_journal_key(keys, i) {
603 j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
604 ret = bch2_journal_replay_key(c, i);
611 * Now that the btree is in a consistent state, we can start journal
612 * reclaim (which will be flushing entries from the btree key cache back
615 set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
616 set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
617 journal_reclaim_kick(j);
619 j->replay_journal_seq = seq;
622 * Now replay leaf node updates:
624 for_each_journal_key(keys, i) {
627 if (i->level || i->btree_id == BTREE_ID_alloc)
630 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
632 ret = bch2_journal_replay_key(c, i);
637 replay_now_at(j, j->replay_journal_seq_end);
638 j->replay_journal_seq = 0;
640 bch2_journal_set_replay_done(j);
641 bch2_journal_flush_all_pins(j);
642 return bch2_journal_error(j);
644 bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
645 ret, bch2_btree_ids[i->btree_id], i->level);
649 /* journal replay early: */
651 static int journal_replay_entry_early(struct bch_fs *c,
652 struct jset_entry *entry)
656 switch (entry->type) {
657 case BCH_JSET_ENTRY_btree_root: {
658 struct btree_root *r;
660 if (entry->btree_id >= BTREE_ID_NR) {
661 bch_err(c, "filesystem has unknown btree type %u",
666 r = &c->btree_roots[entry->btree_id];
669 r->level = entry->level;
670 bkey_copy(&r->key, &entry->start[0]);
678 case BCH_JSET_ENTRY_usage: {
679 struct jset_entry_usage *u =
680 container_of(entry, struct jset_entry_usage, entry);
682 switch (entry->btree_id) {
683 case FS_USAGE_RESERVED:
684 if (entry->level < BCH_REPLICAS_MAX)
685 c->usage_base->persistent_reserved[entry->level] =
688 case FS_USAGE_INODES:
689 c->usage_base->nr_inodes = le64_to_cpu(u->v);
691 case FS_USAGE_KEY_VERSION:
692 atomic64_set(&c->key_version,
699 case BCH_JSET_ENTRY_data_usage: {
700 struct jset_entry_data_usage *u =
701 container_of(entry, struct jset_entry_data_usage, entry);
703 ret = bch2_replicas_set_usage(c, &u->r,
707 case BCH_JSET_ENTRY_dev_usage: {
708 struct jset_entry_dev_usage *u =
709 container_of(entry, struct jset_entry_dev_usage, entry);
710 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
711 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
712 unsigned nr_types = (bytes - sizeof(struct jset_entry_dev_usage)) /
713 sizeof(struct jset_entry_dev_usage_type);
716 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
717 ca->usage_base->buckets_unavailable = le64_to_cpu(u->buckets_unavailable);
719 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
720 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
721 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
722 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
727 case BCH_JSET_ENTRY_blacklist: {
728 struct jset_entry_blacklist *bl_entry =
729 container_of(entry, struct jset_entry_blacklist, entry);
731 ret = bch2_journal_seq_blacklist_add(c,
732 le64_to_cpu(bl_entry->seq),
733 le64_to_cpu(bl_entry->seq) + 1);
736 case BCH_JSET_ENTRY_blacklist_v2: {
737 struct jset_entry_blacklist_v2 *bl_entry =
738 container_of(entry, struct jset_entry_blacklist_v2, entry);
740 ret = bch2_journal_seq_blacklist_add(c,
741 le64_to_cpu(bl_entry->start),
742 le64_to_cpu(bl_entry->end) + 1);
745 case BCH_JSET_ENTRY_clock: {
746 struct jset_entry_clock *clock =
747 container_of(entry, struct jset_entry_clock, entry);
749 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
756 static int journal_replay_early(struct bch_fs *c,
757 struct bch_sb_field_clean *clean,
758 struct list_head *journal)
760 struct journal_replay *i;
761 struct jset_entry *entry;
765 for (entry = clean->start;
766 entry != vstruct_end(&clean->field);
767 entry = vstruct_next(entry)) {
768 ret = journal_replay_entry_early(c, entry);
773 list_for_each_entry(i, journal, list) {
777 vstruct_for_each(&i->j, entry) {
778 ret = journal_replay_entry_early(c, entry);
785 bch2_fs_usage_initialize(c);
790 /* sb clean section: */
792 static struct bkey_i *btree_root_find(struct bch_fs *c,
793 struct bch_sb_field_clean *clean,
795 enum btree_id id, unsigned *level)
798 struct jset_entry *entry, *start, *end;
801 start = clean->start;
802 end = vstruct_end(&clean->field);
805 end = vstruct_last(j);
808 for (entry = start; entry < end; entry = vstruct_next(entry))
809 if (entry->type == BCH_JSET_ENTRY_btree_root &&
810 entry->btree_id == id)
816 return ERR_PTR(-EINVAL);
819 *level = entry->level;
823 static int verify_superblock_clean(struct bch_fs *c,
824 struct bch_sb_field_clean **cleanp,
828 struct bch_sb_field_clean *clean = *cleanp;
831 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
832 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
833 le64_to_cpu(clean->journal_seq),
834 le64_to_cpu(j->seq))) {
840 for (i = 0; i < BTREE_ID_NR; i++) {
841 char buf1[200], buf2[200];
842 struct bkey_i *k1, *k2;
843 unsigned l1 = 0, l2 = 0;
845 k1 = btree_root_find(c, clean, NULL, i, &l1);
846 k2 = btree_root_find(c, NULL, j, i, &l2);
851 mustfix_fsck_err_on(!k1 || !k2 ||
854 k1->k.u64s != k2->k.u64s ||
855 memcmp(k1, k2, bkey_bytes(k1)) ||
857 "superblock btree root %u doesn't match journal after clean shutdown\n"
859 "journal: l=%u %s\n", i,
860 l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
861 l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
867 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
869 struct bch_sb_field_clean *clean, *sb_clean;
872 mutex_lock(&c->sb_lock);
873 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
875 if (fsck_err_on(!sb_clean, c,
876 "superblock marked clean but clean section not present")) {
877 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
879 mutex_unlock(&c->sb_lock);
883 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
886 mutex_unlock(&c->sb_lock);
887 return ERR_PTR(-ENOMEM);
890 ret = bch2_sb_clean_validate(c, clean, READ);
892 mutex_unlock(&c->sb_lock);
896 mutex_unlock(&c->sb_lock);
900 mutex_unlock(&c->sb_lock);
904 static int read_btree_roots(struct bch_fs *c)
909 for (i = 0; i < BTREE_ID_NR; i++) {
910 struct btree_root *r = &c->btree_roots[i];
915 if (i == BTREE_ID_alloc &&
916 c->opts.reconstruct_alloc) {
917 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
922 __fsck_err(c, i == BTREE_ID_alloc
923 ? FSCK_CAN_IGNORE : 0,
924 "invalid btree root %s",
926 if (i == BTREE_ID_alloc)
927 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
930 ret = bch2_btree_root_read(c, i, &r->key, r->level);
932 __fsck_err(c, i == BTREE_ID_alloc
933 ? FSCK_CAN_IGNORE : 0,
934 "error reading btree root %s",
936 if (i == BTREE_ID_alloc)
937 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
941 for (i = 0; i < BTREE_ID_NR; i++)
942 if (!c->btree_roots[i].b)
943 bch2_btree_root_alloc(c, i);
948 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
950 struct bkey_i_snapshot root_snapshot;
951 struct bkey_i_subvolume root_volume;
954 bkey_snapshot_init(&root_snapshot.k_i);
955 root_snapshot.k.p.offset = U32_MAX;
956 root_snapshot.v.flags = 0;
957 root_snapshot.v.parent = 0;
958 root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL;
959 root_snapshot.v.pad = 0;
960 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
962 ret = bch2_btree_insert(c, BTREE_ID_snapshots,
969 bkey_subvolume_init(&root_volume.k_i);
970 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
971 root_volume.v.flags = 0;
972 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
973 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
975 ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
984 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
986 struct bch_fs *c = trans->c;
987 struct btree_iter iter;
989 struct bch_inode_unpacked inode;
992 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
993 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
994 k = bch2_btree_iter_peek_slot(&iter);
999 if (!bkey_is_inode(k.k)) {
1000 bch_err(c, "root inode not found");
1005 ret = bch2_inode_unpack(k, &inode);
1008 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1010 ret = bch2_inode_write(trans, &iter, &inode);
1012 bch2_trans_iter_exit(trans, &iter);
1016 int bch2_fs_recovery(struct bch_fs *c)
1018 const char *err = "cannot allocate memory";
1019 struct bch_sb_field_clean *clean = NULL;
1020 struct jset *last_journal_entry = NULL;
1021 u64 blacklist_seq, journal_seq;
1022 bool write_sb = false;
1026 clean = read_superblock_clean(c);
1027 ret = PTR_ERR_OR_ZERO(clean);
1032 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1033 le64_to_cpu(clean->journal_seq));
1035 bch_info(c, "recovering from unclean shutdown");
1037 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1038 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1044 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1045 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1050 if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1051 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1056 if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1057 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1058 c->opts.fsck = true;
1059 c->opts.fix_errors = FSCK_OPT_YES;
1062 if (!c->replicas.entries ||
1063 c->opts.rebuild_replicas) {
1064 bch_info(c, "building replicas info");
1065 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1068 if (!c->opts.nochanges) {
1069 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1070 bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1071 c->opts.version_upgrade = true;
1072 c->opts.fsck = true;
1073 c->opts.fix_errors = FSCK_OPT_YES;
1074 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1075 bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1076 c->opts.version_upgrade = true;
1077 c->opts.fsck = true;
1078 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
1079 bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
1080 c->opts.version_upgrade = true;
1084 ret = bch2_blacklist_table_initialize(c);
1086 bch_err(c, "error initializing blacklist table");
1090 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1091 struct journal_replay *i;
1093 ret = bch2_journal_read(c, &c->journal_entries,
1094 &blacklist_seq, &journal_seq);
1098 list_for_each_entry_reverse(i, &c->journal_entries, list)
1100 last_journal_entry = &i->j;
1104 if (mustfix_fsck_err_on(c->sb.clean &&
1105 last_journal_entry &&
1106 !journal_entry_empty(last_journal_entry), c,
1107 "filesystem marked clean but journal not empty")) {
1108 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1109 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1110 c->sb.clean = false;
1113 if (!last_journal_entry) {
1114 fsck_err_on(!c->sb.clean, c, "no journal entries found");
1118 c->journal_keys = journal_keys_sort(&c->journal_entries);
1119 if (!c->journal_keys.d) {
1124 if (c->sb.clean && last_journal_entry) {
1125 ret = verify_superblock_clean(c, &clean,
1126 last_journal_entry);
1133 bch_err(c, "no superblock clean section found");
1134 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1138 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1141 if (c->opts.reconstruct_alloc) {
1142 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1143 drop_alloc_keys(&c->journal_keys);
1146 zero_out_btree_mem_ptr(&c->journal_keys);
1148 ret = journal_replay_early(c, clean, &c->journal_entries);
1153 * After an unclean shutdown, skip then next few journal sequence
1154 * numbers as they may have been referenced by btree writes that
1155 * happened before their corresponding journal writes - those btree
1156 * writes need to be ignored, by skipping and blacklisting the next few
1157 * journal sequence numbers:
1162 if (blacklist_seq != journal_seq) {
1163 ret = bch2_journal_seq_blacklist_add(c,
1164 blacklist_seq, journal_seq);
1166 bch_err(c, "error creating new journal seq blacklist entry");
1171 ret = bch2_fs_journal_start(&c->journal, journal_seq,
1172 &c->journal_entries);
1176 ret = read_btree_roots(c);
1180 bch_verbose(c, "starting alloc read");
1181 err = "error reading allocation information";
1182 ret = bch2_alloc_read(c);
1185 bch_verbose(c, "alloc read done");
1187 bch_verbose(c, "starting stripes_read");
1188 err = "error reading stripes";
1189 ret = bch2_stripes_read(c);
1192 bch_verbose(c, "stripes_read done");
1194 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1197 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1198 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1199 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1200 bool metadata_only = c->opts.norecovery;
1202 bch_info(c, "starting mark and sweep");
1203 err = "error in mark and sweep";
1204 ret = bch2_gc(c, true, metadata_only);
1207 bch_verbose(c, "mark and sweep done");
1210 bch2_stripes_heap_start(c);
1212 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1213 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1216 * Skip past versions that might have possibly been used (as nonces),
1217 * but hadn't had their pointers written:
1219 if (c->sb.encryption_type && !c->sb.clean)
1220 atomic64_add(1 << 16, &c->key_version);
1222 if (c->opts.norecovery)
1225 bch_verbose(c, "starting journal replay");
1226 err = "journal replay failed";
1227 ret = bch2_journal_replay(c, c->journal_keys);
1230 bch_verbose(c, "journal replay done");
1232 if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1233 !c->opts.nochanges) {
1235 * note that even when filesystem was clean there might be work
1236 * to do here, if we ran gc (because of fsck) which recalculated
1239 bch_verbose(c, "writing allocation info");
1240 err = "error writing out alloc info";
1241 ret = bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1243 bch_err(c, "error writing alloc info");
1246 bch_verbose(c, "alloc write done");
1249 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1252 err = "error creating root snapshot node";
1253 ret = bch2_fs_initialize_subvolumes(c);
1258 bch_verbose(c, "reading snapshots table");
1259 err = "error reading snapshots table";
1260 ret = bch2_fs_snapshots_start(c);
1263 bch_verbose(c, "reading snapshots done");
1265 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1266 /* set bi_subvol on root inode */
1267 err = "error upgrade root inode for subvolumes";
1268 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1269 bch2_fs_upgrade_for_subvolumes(&trans));
1275 bch_info(c, "starting fsck");
1276 err = "error in fsck";
1277 ret = bch2_fsck_full(c);
1280 bch_verbose(c, "fsck done");
1281 } else if (!c->sb.clean) {
1282 bch_verbose(c, "checking for deleted inodes");
1283 err = "error in recovery";
1284 ret = bch2_fsck_walk_inodes_only(c);
1287 bch_verbose(c, "check inodes done");
1290 if (enabled_qtypes(c)) {
1291 bch_verbose(c, "reading quotas");
1292 ret = bch2_fs_quota_read(c);
1295 bch_verbose(c, "quotas done");
1298 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1299 !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1300 struct bch_move_stats stats;
1302 bch_move_stats_init(&stats, "recovery");
1304 bch_info(c, "scanning for old btree nodes");
1305 ret = bch2_fs_read_write(c);
1309 ret = bch2_scan_old_btree_nodes(c, &stats);
1312 bch_info(c, "scanning for old btree nodes done");
1315 mutex_lock(&c->sb_lock);
1316 if (c->opts.version_upgrade) {
1317 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1318 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1322 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1323 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1328 !test_bit(BCH_FS_ERROR, &c->flags) &&
1329 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1330 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1331 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1336 bch2_write_super(c);
1337 mutex_unlock(&c->sb_lock);
1339 if (c->journal_seq_blacklist_table &&
1340 c->journal_seq_blacklist_table->nr > 128)
1341 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1345 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1346 bch2_flush_fsck_errs(c);
1348 if (!c->opts.keep_journal) {
1349 bch2_journal_keys_free(&c->journal_keys);
1350 bch2_journal_entries_free(&c->journal_entries);
1354 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1356 bch_verbose(c, "ret %i", ret);
1360 bch2_fs_emergency_read_only(c);
1364 int bch2_fs_initialize(struct bch_fs *c)
1366 struct bch_inode_unpacked root_inode, lostfound_inode;
1367 struct bkey_inode_buf packed_inode;
1368 struct qstr lostfound = QSTR("lost+found");
1369 const char *err = "cannot allocate memory";
1375 bch_notice(c, "initializing new filesystem");
1377 mutex_lock(&c->sb_lock);
1378 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1379 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1381 if (c->opts.version_upgrade) {
1382 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1383 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1384 bch2_write_super(c);
1387 for_each_online_member(ca, c, i)
1388 bch2_mark_dev_superblock(c, ca, 0);
1389 mutex_unlock(&c->sb_lock);
1391 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1392 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1394 for (i = 0; i < BTREE_ID_NR; i++)
1395 bch2_btree_root_alloc(c, i);
1397 set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1398 set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1400 err = "unable to allocate journal buckets";
1401 for_each_online_member(ca, c, i) {
1402 ret = bch2_dev_journal_alloc(ca);
1404 percpu_ref_put(&ca->io_ref);
1410 * journal_res_get() will crash if called before this has
1411 * set up the journal.pin FIFO and journal.cur pointer:
1413 bch2_fs_journal_start(&c->journal, 1, &journal);
1414 bch2_journal_set_replay_done(&c->journal);
1416 err = "error going read-write";
1417 ret = bch2_fs_read_write_early(c);
1422 * Write out the superblock and journal buckets, now that we can do
1425 err = "error marking superblock and journal";
1426 for_each_member_device(ca, c, i) {
1427 ret = bch2_trans_mark_dev_sb(c, ca);
1429 percpu_ref_put(&ca->ref);
1434 err = "error creating root snapshot node";
1435 ret = bch2_fs_initialize_subvolumes(c);
1439 bch_verbose(c, "reading snapshots table");
1440 err = "error reading snapshots table";
1441 ret = bch2_fs_snapshots_start(c);
1444 bch_verbose(c, "reading snapshots done");
1446 bch2_inode_init(c, &root_inode, 0, 0,
1447 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1448 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1449 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1450 bch2_inode_pack(c, &packed_inode, &root_inode);
1451 packed_inode.inode.k.p.snapshot = U32_MAX;
1453 err = "error creating root directory";
1454 ret = bch2_btree_insert(c, BTREE_ID_inodes,
1455 &packed_inode.inode.k_i,
1460 bch2_inode_init_early(c, &lostfound_inode);
1462 err = "error creating lost+found";
1463 ret = bch2_trans_do(c, NULL, NULL, 0,
1464 bch2_create_trans(&trans,
1465 BCACHEFS_ROOT_SUBVOL_INUM,
1466 &root_inode, &lostfound_inode,
1468 0, 0, S_IFDIR|0700, 0,
1469 NULL, NULL, (subvol_inum) { 0 }, 0));
1471 bch_err(c, "error creating lost+found");
1475 if (enabled_qtypes(c)) {
1476 ret = bch2_fs_quota_read(c);
1481 err = "error writing first journal entry";
1482 ret = bch2_journal_flush(&c->journal);
1486 mutex_lock(&c->sb_lock);
1487 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1488 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1490 bch2_write_super(c);
1491 mutex_unlock(&c->sb_lock);
1495 pr_err("Error initializing new filesystem: %s (%i)", err, ret);