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"
24 #include "subvolume.h"
27 #include <linux/sort.h>
28 #include <linux/stat.h>
30 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
32 /* for -o reconstruct_alloc: */
33 static void drop_alloc_keys(struct journal_keys *keys)
37 for (src = 0, dst = 0; src < keys->nr; src++)
38 if (keys->d[src].btree_id != BTREE_ID_alloc)
39 keys->d[dst++] = keys->d[src];
45 * Btree node pointers have a field to stack a pointer to the in memory btree
46 * node; we need to zero out this field when reading in btree nodes, or when
47 * reading in keys from the journal:
49 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
51 struct journal_key *i;
53 for (i = keys->d; i < keys->d + keys->nr; i++)
54 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
55 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
58 /* iterate over keys read from the journal: */
60 static int __journal_key_cmp(enum btree_id l_btree_id,
63 const struct journal_key *r)
65 return (cmp_int(l_btree_id, r->btree_id) ?:
66 cmp_int(l_level, r->level) ?:
67 bpos_cmp(l_pos, r->k->k.p));
70 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
72 return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
75 size_t bch2_journal_key_search(struct journal_keys *journal_keys,
76 enum btree_id id, unsigned level,
79 size_t l = 0, r = journal_keys->nr, m;
82 m = l + ((r - l) >> 1);
83 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
89 BUG_ON(l < journal_keys->nr &&
90 __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
93 __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
98 struct bkey_i *bch2_journal_keys_peek(struct bch_fs *c, enum btree_id btree_id,
99 unsigned level, struct bpos pos)
101 struct journal_keys *keys = &c->journal_keys;
102 struct journal_key *end = keys->d + keys->nr;
103 struct journal_key *k = keys->d +
104 bch2_journal_key_search(keys, btree_id, level, pos);
106 while (k < end && k->overwritten)
110 k->btree_id == btree_id &&
116 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
118 struct bkey_i *n = iter->keys->d[idx].k;
119 struct btree_and_journal_iter *biter =
120 container_of(iter, struct btree_and_journal_iter, journal);
122 if (iter->idx > idx ||
125 bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
129 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
130 unsigned level, struct bkey_i *k)
132 struct journal_key n = {
138 * Ensure these keys are done last by journal replay, to unblock
141 .journal_seq = U32_MAX,
143 struct journal_keys *keys = &c->journal_keys;
144 struct journal_iter *iter;
145 size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
147 BUG_ON(test_bit(BCH_FS_RW, &c->flags));
149 if (idx < keys->nr &&
150 journal_key_cmp(&n, &keys->d[idx]) == 0) {
151 if (keys->d[idx].allocated)
152 kfree(keys->d[idx].k);
157 if (keys->nr == keys->size) {
158 struct journal_keys new_keys = {
160 .size = keys->size * 2,
161 .journal_seq_base = keys->journal_seq_base,
164 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
166 bch_err(c, "%s: error allocating new key array (size %zu)",
167 __func__, new_keys.size);
171 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
176 array_insert_item(keys->d, keys->nr, idx, n);
178 list_for_each_entry(iter, &c->journal_iters, list)
179 journal_iter_fix(c, iter, idx);
185 * Can only be used from the recovery thread while we're still RO - can't be
186 * used once we've got RW, as journal_keys is at that point used by multiple
189 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
190 unsigned level, struct bkey_i *k)
195 n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
200 ret = bch2_journal_key_insert_take(c, id, level, n);
206 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
207 unsigned level, struct bpos pos)
209 struct bkey_i whiteout;
211 bkey_init(&whiteout.k);
214 return bch2_journal_key_insert(c, id, level, &whiteout);
217 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
218 unsigned level, struct bpos pos)
220 struct journal_keys *keys = &c->journal_keys;
221 size_t idx = bch2_journal_key_search(keys, btree, level, pos);
223 if (idx < keys->nr &&
224 keys->d[idx].btree_id == btree &&
225 keys->d[idx].level == level &&
226 !bpos_cmp(keys->d[idx].k->k.p, pos))
227 keys->d[idx].overwritten = true;
230 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
232 struct journal_key *k = iter->keys->d + iter->idx;
234 while (k < iter->keys->d + iter->keys->nr &&
235 k->btree_id == iter->btree_id &&
236 k->level == iter->level) {
241 k = iter->keys->d + iter->idx;
247 static void bch2_journal_iter_advance(struct journal_iter *iter)
249 if (iter->idx < iter->keys->nr)
253 static void bch2_journal_iter_exit(struct journal_iter *iter)
255 list_del(&iter->list);
258 static void bch2_journal_iter_init(struct bch_fs *c,
259 struct journal_iter *iter,
260 enum btree_id id, unsigned level,
265 iter->keys = &c->journal_keys;
266 iter->idx = bch2_journal_key_search(&c->journal_keys, id, level, pos);
269 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
271 return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
272 iter->b, &iter->unpacked);
275 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
277 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
280 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
282 switch (iter->last) {
286 bch2_journal_iter_advance_btree(iter);
289 bch2_journal_iter_advance(&iter->journal);
296 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
301 struct bkey_s_c btree_k =
302 bch2_journal_iter_peek_btree(iter);
303 struct bkey_s_c journal_k =
304 bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
306 if (btree_k.k && journal_k.k) {
307 int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
310 bch2_journal_iter_advance_btree(iter);
312 iter->last = cmp < 0 ? btree : journal;
313 } else if (btree_k.k) {
315 } else if (journal_k.k) {
316 iter->last = journal;
319 return bkey_s_c_null;
322 ret = iter->last == journal ? journal_k : btree_k;
325 bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
326 iter->journal.idx = iter->journal.keys->nr;
328 return bkey_s_c_null;
331 if (!bkey_deleted(ret.k))
334 bch2_btree_and_journal_iter_advance(iter);
340 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
342 bch2_btree_and_journal_iter_advance(iter);
344 return bch2_btree_and_journal_iter_peek(iter);
347 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
349 bch2_journal_iter_exit(&iter->journal);
352 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
355 struct btree_node_iter node_iter,
358 memset(iter, 0, sizeof(*iter));
361 iter->node_iter = node_iter;
362 bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
363 INIT_LIST_HEAD(&iter->journal.list);
367 * this version is used by btree_gc before filesystem has gone RW and
368 * multithreaded, so uses the journal_iters list:
370 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
374 struct btree_node_iter node_iter;
376 bch2_btree_node_iter_init_from_start(&node_iter, b);
377 __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
378 list_add(&iter->journal.list, &c->journal_iters);
381 /* sort and dedup all keys in the journal: */
383 void bch2_journal_entries_free(struct list_head *list)
386 while (!list_empty(list)) {
387 struct journal_replay *i =
388 list_first_entry(list, struct journal_replay, list);
390 kvpfree(i, offsetof(struct journal_replay, j) +
391 vstruct_bytes(&i->j));
396 * When keys compare equal, oldest compares first:
398 static int journal_sort_key_cmp(const void *_l, const void *_r)
400 const struct journal_key *l = _l;
401 const struct journal_key *r = _r;
403 return journal_key_cmp(l, r) ?:
404 cmp_int(l->journal_seq, r->journal_seq) ?:
405 cmp_int(l->journal_offset, r->journal_offset);
408 void bch2_journal_keys_free(struct journal_keys *keys)
410 struct journal_key *i;
412 for (i = keys->d; i < keys->d + keys->nr; i++)
421 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
423 struct journal_replay *i;
424 struct jset_entry *entry;
425 struct bkey_i *k, *_n;
426 struct journal_keys keys = { NULL };
427 struct journal_key *src, *dst;
430 if (list_empty(journal_entries))
433 list_for_each_entry(i, journal_entries, list) {
437 if (!keys.journal_seq_base)
438 keys.journal_seq_base = le64_to_cpu(i->j.seq);
440 for_each_jset_key(k, _n, entry, &i->j)
444 keys.size = roundup_pow_of_two(nr_keys);
446 keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
450 list_for_each_entry(i, journal_entries, list) {
454 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
456 for_each_jset_key(k, _n, entry, &i->j)
457 keys.d[keys.nr++] = (struct journal_key) {
458 .btree_id = entry->btree_id,
459 .level = entry->level,
461 .journal_seq = le64_to_cpu(i->j.seq) -
462 keys.journal_seq_base,
463 .journal_offset = k->_data - i->j._data,
467 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
470 while (src < keys.d + keys.nr) {
471 while (src + 1 < keys.d + keys.nr &&
472 src[0].btree_id == src[1].btree_id &&
473 src[0].level == src[1].level &&
474 !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
480 keys.nr = dst - keys.d;
485 /* journal replay: */
487 static void replay_now_at(struct journal *j, u64 seq)
489 BUG_ON(seq < j->replay_journal_seq);
490 BUG_ON(seq > j->replay_journal_seq_end);
492 while (j->replay_journal_seq < seq)
493 bch2_journal_pin_put(j, j->replay_journal_seq++);
496 static int bch2_journal_replay_key(struct btree_trans *trans,
497 struct journal_key *k)
499 struct btree_iter iter;
500 unsigned iter_flags =
502 BTREE_ITER_NOT_EXTENTS;
505 if (!k->level && k->btree_id == BTREE_ID_alloc)
506 iter_flags |= BTREE_ITER_CACHED;
508 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
509 BTREE_MAX_DEPTH, k->level,
511 ret = bch2_btree_iter_traverse(&iter);
515 /* Must be checked with btree locked: */
519 ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
521 bch2_trans_iter_exit(trans, &iter);
525 static int journal_sort_seq_cmp(const void *_l, const void *_r)
527 const struct journal_key *l = *((const struct journal_key **)_l);
528 const struct journal_key *r = *((const struct journal_key **)_r);
530 return cmp_int(l->journal_seq, r->journal_seq);
533 static int bch2_journal_replay(struct bch_fs *c)
535 struct journal_keys *keys = &c->journal_keys;
536 struct journal_key **keys_sorted, *k;
537 struct journal *j = &c->journal;
541 keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
545 for (i = 0; i < keys->nr; i++)
546 keys_sorted[i] = &keys->d[i];
548 sort(keys_sorted, keys->nr,
549 sizeof(keys_sorted[0]),
550 journal_sort_seq_cmp, NULL);
553 replay_now_at(j, keys->journal_seq_base);
555 for (i = 0; i < keys->nr; i++) {
561 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
563 ret = bch2_trans_do(c, NULL, NULL,
564 BTREE_INSERT_LAZY_RW|
567 ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
569 bch2_journal_replay_key(&trans, k));
571 bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
572 ret, bch2_btree_ids[k->btree_id], k->level);
577 replay_now_at(j, j->replay_journal_seq_end);
578 j->replay_journal_seq = 0;
580 bch2_journal_set_replay_done(j);
581 bch2_journal_flush_all_pins(j);
582 ret = bch2_journal_error(j);
584 if (keys->nr && !ret)
585 bch2_journal_log_msg(&c->journal, "journal replay finished");
591 /* journal replay early: */
593 static int journal_replay_entry_early(struct bch_fs *c,
594 struct jset_entry *entry)
598 switch (entry->type) {
599 case BCH_JSET_ENTRY_btree_root: {
600 struct btree_root *r;
602 if (entry->btree_id >= BTREE_ID_NR) {
603 bch_err(c, "filesystem has unknown btree type %u",
608 r = &c->btree_roots[entry->btree_id];
611 r->level = entry->level;
612 bkey_copy(&r->key, &entry->start[0]);
620 case BCH_JSET_ENTRY_usage: {
621 struct jset_entry_usage *u =
622 container_of(entry, struct jset_entry_usage, entry);
624 switch (entry->btree_id) {
625 case BCH_FS_USAGE_reserved:
626 if (entry->level < BCH_REPLICAS_MAX)
627 c->usage_base->persistent_reserved[entry->level] =
630 case BCH_FS_USAGE_inodes:
631 c->usage_base->nr_inodes = le64_to_cpu(u->v);
633 case BCH_FS_USAGE_key_version:
634 atomic64_set(&c->key_version,
641 case BCH_JSET_ENTRY_data_usage: {
642 struct jset_entry_data_usage *u =
643 container_of(entry, struct jset_entry_data_usage, entry);
645 ret = bch2_replicas_set_usage(c, &u->r,
649 case BCH_JSET_ENTRY_dev_usage: {
650 struct jset_entry_dev_usage *u =
651 container_of(entry, struct jset_entry_dev_usage, entry);
652 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
653 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
655 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
656 ca->usage_base->buckets_unavailable = le64_to_cpu(u->buckets_unavailable);
658 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
659 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
660 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
661 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
666 case BCH_JSET_ENTRY_blacklist: {
667 struct jset_entry_blacklist *bl_entry =
668 container_of(entry, struct jset_entry_blacklist, entry);
670 ret = bch2_journal_seq_blacklist_add(c,
671 le64_to_cpu(bl_entry->seq),
672 le64_to_cpu(bl_entry->seq) + 1);
675 case BCH_JSET_ENTRY_blacklist_v2: {
676 struct jset_entry_blacklist_v2 *bl_entry =
677 container_of(entry, struct jset_entry_blacklist_v2, entry);
679 ret = bch2_journal_seq_blacklist_add(c,
680 le64_to_cpu(bl_entry->start),
681 le64_to_cpu(bl_entry->end) + 1);
684 case BCH_JSET_ENTRY_clock: {
685 struct jset_entry_clock *clock =
686 container_of(entry, struct jset_entry_clock, entry);
688 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
695 static int journal_replay_early(struct bch_fs *c,
696 struct bch_sb_field_clean *clean,
697 struct list_head *journal)
699 struct journal_replay *i;
700 struct jset_entry *entry;
704 for (entry = clean->start;
705 entry != vstruct_end(&clean->field);
706 entry = vstruct_next(entry)) {
707 ret = journal_replay_entry_early(c, entry);
712 list_for_each_entry(i, journal, list) {
716 vstruct_for_each(&i->j, entry) {
717 ret = journal_replay_entry_early(c, entry);
724 bch2_fs_usage_initialize(c);
729 /* sb clean section: */
731 static struct bkey_i *btree_root_find(struct bch_fs *c,
732 struct bch_sb_field_clean *clean,
734 enum btree_id id, unsigned *level)
737 struct jset_entry *entry, *start, *end;
740 start = clean->start;
741 end = vstruct_end(&clean->field);
744 end = vstruct_last(j);
747 for (entry = start; entry < end; entry = vstruct_next(entry))
748 if (entry->type == BCH_JSET_ENTRY_btree_root &&
749 entry->btree_id == id)
755 return ERR_PTR(-EINVAL);
758 *level = entry->level;
762 static int verify_superblock_clean(struct bch_fs *c,
763 struct bch_sb_field_clean **cleanp,
767 struct bch_sb_field_clean *clean = *cleanp;
768 struct printbuf buf1 = PRINTBUF;
769 struct printbuf buf2 = PRINTBUF;
772 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
773 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
774 le64_to_cpu(clean->journal_seq),
775 le64_to_cpu(j->seq))) {
781 for (i = 0; i < BTREE_ID_NR; i++) {
782 struct bkey_i *k1, *k2;
783 unsigned l1 = 0, l2 = 0;
785 k1 = btree_root_find(c, clean, NULL, i, &l1);
786 k2 = btree_root_find(c, NULL, j, i, &l2);
791 printbuf_reset(&buf1);
792 printbuf_reset(&buf2);
795 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
797 pr_buf(&buf1, "(none)");
800 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
802 pr_buf(&buf2, "(none)");
804 mustfix_fsck_err_on(!k1 || !k2 ||
807 k1->k.u64s != k2->k.u64s ||
808 memcmp(k1, k2, bkey_bytes(k1)) ||
810 "superblock btree root %u doesn't match journal after clean shutdown\n"
812 "journal: l=%u %s\n", i,
817 printbuf_exit(&buf2);
818 printbuf_exit(&buf1);
822 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
824 struct bch_sb_field_clean *clean, *sb_clean;
827 mutex_lock(&c->sb_lock);
828 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
830 if (fsck_err_on(!sb_clean, c,
831 "superblock marked clean but clean section not present")) {
832 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
834 mutex_unlock(&c->sb_lock);
838 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
841 mutex_unlock(&c->sb_lock);
842 return ERR_PTR(-ENOMEM);
845 ret = bch2_sb_clean_validate_late(c, clean, READ);
847 mutex_unlock(&c->sb_lock);
851 mutex_unlock(&c->sb_lock);
855 mutex_unlock(&c->sb_lock);
859 static int read_btree_roots(struct bch_fs *c)
864 for (i = 0; i < BTREE_ID_NR; i++) {
865 struct btree_root *r = &c->btree_roots[i];
870 if (i == BTREE_ID_alloc &&
871 c->opts.reconstruct_alloc) {
872 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
877 __fsck_err(c, i == BTREE_ID_alloc
878 ? FSCK_CAN_IGNORE : 0,
879 "invalid btree root %s",
881 if (i == BTREE_ID_alloc)
882 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
885 ret = bch2_btree_root_read(c, i, &r->key, r->level);
887 __fsck_err(c, i == BTREE_ID_alloc
888 ? FSCK_CAN_IGNORE : 0,
889 "error reading btree root %s",
891 if (i == BTREE_ID_alloc)
892 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
896 for (i = 0; i < BTREE_ID_NR; i++)
897 if (!c->btree_roots[i].b)
898 bch2_btree_root_alloc(c, i);
903 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
905 struct bkey_i_snapshot root_snapshot;
906 struct bkey_i_subvolume root_volume;
909 bkey_snapshot_init(&root_snapshot.k_i);
910 root_snapshot.k.p.offset = U32_MAX;
911 root_snapshot.v.flags = 0;
912 root_snapshot.v.parent = 0;
913 root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL;
914 root_snapshot.v.pad = 0;
915 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
917 ret = bch2_btree_insert(c, BTREE_ID_snapshots,
924 bkey_subvolume_init(&root_volume.k_i);
925 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
926 root_volume.v.flags = 0;
927 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
928 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
930 ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
939 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
941 struct btree_iter iter;
943 struct bch_inode_unpacked inode;
946 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
947 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
948 k = bch2_btree_iter_peek_slot(&iter);
953 if (!bkey_is_inode(k.k)) {
954 bch_err(trans->c, "root inode not found");
959 ret = bch2_inode_unpack(k, &inode);
962 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
964 ret = bch2_inode_write(trans, &iter, &inode);
966 bch2_trans_iter_exit(trans, &iter);
970 int bch2_fs_recovery(struct bch_fs *c)
972 const char *err = "cannot allocate memory";
973 struct bch_sb_field_clean *clean = NULL;
974 struct jset *last_journal_entry = NULL;
975 u64 blacklist_seq, journal_seq;
976 bool write_sb = false;
980 clean = read_superblock_clean(c);
981 ret = PTR_ERR_OR_ZERO(clean);
986 bch_info(c, "recovering from clean shutdown, journal seq %llu",
987 le64_to_cpu(clean->journal_seq));
989 bch_info(c, "recovering from unclean shutdown");
991 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
992 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
998 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
999 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1004 if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1005 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1010 if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1011 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1012 c->opts.fsck = true;
1013 c->opts.fix_errors = FSCK_OPT_YES;
1016 if (!c->replicas.entries ||
1017 c->opts.rebuild_replicas) {
1018 bch_info(c, "building replicas info");
1019 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1022 if (!c->opts.nochanges) {
1023 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1024 bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1025 c->opts.version_upgrade = true;
1026 c->opts.fsck = true;
1027 c->opts.fix_errors = FSCK_OPT_YES;
1028 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1029 bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1030 c->opts.version_upgrade = true;
1031 c->opts.fsck = true;
1032 } else if (c->sb.version < bcachefs_metadata_version_alloc_v4) {
1033 bch_info(c, "filesystem version is prior to alloc_v4 - upgrading");
1034 c->opts.version_upgrade = true;
1038 ret = bch2_blacklist_table_initialize(c);
1040 bch_err(c, "error initializing blacklist table");
1044 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1045 struct journal_replay *i;
1047 bch_verbose(c, "starting journal read");
1048 ret = bch2_journal_read(c, &c->journal_entries,
1049 &blacklist_seq, &journal_seq);
1053 list_for_each_entry_reverse(i, &c->journal_entries, list)
1055 last_journal_entry = &i->j;
1059 if (mustfix_fsck_err_on(c->sb.clean &&
1060 last_journal_entry &&
1061 !journal_entry_empty(last_journal_entry), c,
1062 "filesystem marked clean but journal not empty")) {
1063 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1064 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1065 c->sb.clean = false;
1068 if (!last_journal_entry) {
1069 fsck_err_on(!c->sb.clean, c, "no journal entries found");
1073 c->journal_keys = journal_keys_sort(&c->journal_entries);
1074 if (!c->journal_keys.d) {
1079 if (c->sb.clean && last_journal_entry) {
1080 ret = verify_superblock_clean(c, &clean,
1081 last_journal_entry);
1088 bch_err(c, "no superblock clean section found");
1089 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1093 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1096 if (c->opts.read_journal_only)
1099 if (c->opts.reconstruct_alloc) {
1100 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1101 drop_alloc_keys(&c->journal_keys);
1104 zero_out_btree_mem_ptr(&c->journal_keys);
1106 ret = journal_replay_early(c, clean, &c->journal_entries);
1111 * After an unclean shutdown, skip then next few journal sequence
1112 * numbers as they may have been referenced by btree writes that
1113 * happened before their corresponding journal writes - those btree
1114 * writes need to be ignored, by skipping and blacklisting the next few
1115 * journal sequence numbers:
1120 if (blacklist_seq != journal_seq) {
1121 ret = bch2_journal_seq_blacklist_add(c,
1122 blacklist_seq, journal_seq);
1124 bch_err(c, "error creating new journal seq blacklist entry");
1129 ret = bch2_fs_journal_start(&c->journal, journal_seq,
1130 &c->journal_entries);
1134 ret = read_btree_roots(c);
1138 bch_verbose(c, "starting alloc read");
1139 err = "error reading allocation information";
1141 down_read(&c->gc_lock);
1142 ret = bch2_alloc_read(c);
1143 up_read(&c->gc_lock);
1147 bch_verbose(c, "alloc read done");
1149 bch_verbose(c, "starting stripes_read");
1150 err = "error reading stripes";
1151 ret = bch2_stripes_read(c);
1154 bch_verbose(c, "stripes_read done");
1157 * If we're not running fsck, this ensures bch2_fsck_err() calls are
1158 * instead interpreted as bch2_inconsistent_err() calls:
1161 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1164 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1165 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1166 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1167 bool metadata_only = c->opts.norecovery;
1169 bch_info(c, "checking allocations");
1170 err = "error checking allocations";
1171 ret = bch2_gc(c, true, metadata_only);
1174 bch_verbose(c, "done checking allocations");
1178 bch_info(c, "checking need_discard and freespace btrees");
1179 err = "error checking need_discard and freespace btrees";
1180 ret = bch2_check_alloc_info(c, true);
1184 ret = bch2_check_lrus(c, true);
1187 bch_verbose(c, "done checking need_discard and freespace btrees");
1190 bch2_stripes_heap_start(c);
1192 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1193 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1194 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1197 * Skip past versions that might have possibly been used (as nonces),
1198 * but hadn't had their pointers written:
1200 if (c->sb.encryption_type && !c->sb.clean)
1201 atomic64_add(1 << 16, &c->key_version);
1203 if (c->opts.norecovery)
1206 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1207 err = "journal replay failed";
1208 ret = bch2_journal_replay(c);
1211 if (c->opts.verbose || !c->sb.clean)
1212 bch_info(c, "journal replay done");
1214 err = "error initializing freespace";
1215 ret = bch2_fs_freespace_init(c);
1219 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1222 err = "error creating root snapshot node";
1223 ret = bch2_fs_initialize_subvolumes(c);
1228 bch_verbose(c, "reading snapshots table");
1229 err = "error reading snapshots table";
1230 ret = bch2_fs_snapshots_start(c);
1233 bch_verbose(c, "reading snapshots done");
1235 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1236 /* set bi_subvol on root inode */
1237 err = "error upgrade root inode for subvolumes";
1238 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1239 bch2_fs_upgrade_for_subvolumes(&trans));
1245 bch_info(c, "starting fsck");
1246 err = "error in fsck";
1247 ret = bch2_fsck_full(c);
1250 bch_verbose(c, "fsck done");
1251 } else if (!c->sb.clean) {
1252 bch_verbose(c, "checking for deleted inodes");
1253 err = "error in recovery";
1254 ret = bch2_fsck_walk_inodes_only(c);
1257 bch_verbose(c, "check inodes done");
1260 if (enabled_qtypes(c)) {
1261 bch_verbose(c, "reading quotas");
1262 ret = bch2_fs_quota_read(c);
1265 bch_verbose(c, "quotas done");
1268 mutex_lock(&c->sb_lock);
1269 if (c->opts.version_upgrade) {
1270 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1271 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1275 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1276 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1281 !test_bit(BCH_FS_ERROR, &c->flags) &&
1282 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1283 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1284 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1289 bch2_write_super(c);
1290 mutex_unlock(&c->sb_lock);
1292 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1293 !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1294 le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1295 struct bch_move_stats stats;
1297 bch_move_stats_init(&stats, "recovery");
1299 bch_info(c, "scanning for old btree nodes");
1300 ret = bch2_fs_read_write(c);
1304 ret = bch2_scan_old_btree_nodes(c, &stats);
1307 bch_info(c, "scanning for old btree nodes done");
1310 if (c->journal_seq_blacklist_table &&
1311 c->journal_seq_blacklist_table->nr > 128)
1312 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1316 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1317 bch2_flush_fsck_errs(c);
1319 if (!c->opts.keep_journal) {
1320 bch2_journal_keys_free(&c->journal_keys);
1321 bch2_journal_entries_free(&c->journal_entries);
1325 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1327 bch_verbose(c, "ret %i", ret);
1331 bch2_fs_emergency_read_only(c);
1335 int bch2_fs_initialize(struct bch_fs *c)
1337 struct bch_inode_unpacked root_inode, lostfound_inode;
1338 struct bkey_inode_buf packed_inode;
1339 struct qstr lostfound = QSTR("lost+found");
1340 const char *err = "cannot allocate memory";
1346 bch_notice(c, "initializing new filesystem");
1348 mutex_lock(&c->sb_lock);
1349 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1350 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1352 if (c->opts.version_upgrade) {
1353 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1354 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1355 bch2_write_super(c);
1357 mutex_unlock(&c->sb_lock);
1359 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1360 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1361 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1363 for (i = 0; i < BTREE_ID_NR; i++)
1364 bch2_btree_root_alloc(c, i);
1366 err = "unable to allocate journal buckets";
1367 for_each_online_member(ca, c, i) {
1368 ret = bch2_dev_journal_alloc(ca);
1370 percpu_ref_put(&ca->io_ref);
1376 * journal_res_get() will crash if called before this has
1377 * set up the journal.pin FIFO and journal.cur pointer:
1379 bch2_fs_journal_start(&c->journal, 1, &journal);
1380 bch2_journal_set_replay_done(&c->journal);
1382 err = "error going read-write";
1383 ret = bch2_fs_read_write_early(c);
1388 * Write out the superblock and journal buckets, now that we can do
1391 bch_verbose(c, "marking superblocks");
1392 err = "error marking superblock and journal";
1393 for_each_member_device(ca, c, i) {
1394 ret = bch2_trans_mark_dev_sb(c, ca);
1396 percpu_ref_put(&ca->ref);
1400 ca->new_fs_bucket_idx = 0;
1403 bch_verbose(c, "initializing freespace");
1404 err = "error initializing freespace";
1405 ret = bch2_fs_freespace_init(c);
1409 err = "error creating root snapshot node";
1410 ret = bch2_fs_initialize_subvolumes(c);
1414 bch_verbose(c, "reading snapshots table");
1415 err = "error reading snapshots table";
1416 ret = bch2_fs_snapshots_start(c);
1419 bch_verbose(c, "reading snapshots done");
1421 bch2_inode_init(c, &root_inode, 0, 0,
1422 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1423 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1424 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1425 bch2_inode_pack(c, &packed_inode, &root_inode);
1426 packed_inode.inode.k.p.snapshot = U32_MAX;
1428 err = "error creating root directory";
1429 ret = bch2_btree_insert(c, BTREE_ID_inodes,
1430 &packed_inode.inode.k_i,
1435 bch2_inode_init_early(c, &lostfound_inode);
1437 err = "error creating lost+found";
1438 ret = bch2_trans_do(c, NULL, NULL, 0,
1439 bch2_create_trans(&trans,
1440 BCACHEFS_ROOT_SUBVOL_INUM,
1441 &root_inode, &lostfound_inode,
1443 0, 0, S_IFDIR|0700, 0,
1444 NULL, NULL, (subvol_inum) { 0 }, 0));
1446 bch_err(c, "error creating lost+found");
1450 if (enabled_qtypes(c)) {
1451 ret = bch2_fs_quota_read(c);
1456 err = "error writing first journal entry";
1457 ret = bch2_journal_flush(&c->journal);
1461 mutex_lock(&c->sb_lock);
1462 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1463 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1465 bch2_write_super(c);
1466 mutex_unlock(&c->sb_lock);
1470 pr_err("Error initializing new filesystem: %s (%i)", err, ret);