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 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
77 size_t gap_size = keys->size - keys->nr;
84 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
86 return keys->d + idx_to_pos(keys, idx);
89 size_t bch2_journal_key_search(struct journal_keys *keys,
90 enum btree_id id, unsigned level,
93 size_t l = 0, r = keys->nr, m;
96 m = l + ((r - l) >> 1);
97 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
103 BUG_ON(l < keys->nr &&
104 __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
107 __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
109 return idx_to_pos(keys, l);
112 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
113 unsigned level, struct bpos pos,
116 struct journal_keys *keys = &c->journal_keys;
117 size_t idx = bch2_journal_key_search(keys, btree_id, level, pos);
119 while (idx < keys->size &&
120 keys->d[idx].btree_id == btree_id &&
121 keys->d[idx].level == level &&
122 bpos_cmp(keys->d[idx].k->k.p, end_pos) <= 0) {
123 if (!keys->d[idx].overwritten)
124 return keys->d[idx].k;
127 if (idx == keys->gap)
128 idx += keys->size - keys->nr;
134 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
135 unsigned level, struct bpos pos)
137 return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos);
140 static void journal_iters_fix(struct bch_fs *c)
142 struct journal_keys *keys = &c->journal_keys;
143 /* The key we just inserted is immediately before the gap: */
144 struct journal_key *n = &keys->d[keys->gap - 1];
145 size_t gap_end = keys->gap + (keys->size - keys->nr);
146 struct btree_and_journal_iter *iter;
149 * If an iterator points one after the key we just inserted,
150 * and the key we just inserted compares >= the iterator's position,
151 * decrement the iterator so it points at the key we just inserted:
153 list_for_each_entry(iter, &c->journal_iters, journal.list)
154 if (iter->journal.idx == gap_end &&
156 iter->b->c.btree_id == n->btree_id &&
157 iter->b->c.level == n->level &&
158 bpos_cmp(n->k->k.p, iter->unpacked.p) >= 0)
159 iter->journal.idx = keys->gap - 1;
162 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
164 struct journal_keys *keys = &c->journal_keys;
165 struct journal_iter *iter;
166 size_t gap_size = keys->size - keys->nr;
168 list_for_each_entry(iter, &c->journal_iters, list) {
169 if (iter->idx > old_gap)
170 iter->idx -= gap_size;
171 if (iter->idx >= new_gap)
172 iter->idx += gap_size;
176 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
177 unsigned level, struct bkey_i *k)
179 struct journal_key n = {
185 * Ensure these keys are done last by journal replay, to unblock
188 .journal_seq = U32_MAX,
190 struct journal_keys *keys = &c->journal_keys;
191 size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
193 BUG_ON(test_bit(BCH_FS_RW, &c->flags));
195 if (idx < keys->size &&
196 journal_key_cmp(&n, &keys->d[idx]) == 0) {
197 if (keys->d[idx].allocated)
198 kfree(keys->d[idx].k);
204 idx -= keys->size - keys->nr;
206 if (keys->nr == keys->size) {
207 struct journal_keys new_keys = {
209 .size = max_t(size_t, keys->size, 8) * 2,
210 .journal_seq_base = keys->journal_seq_base,
213 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
215 bch_err(c, "%s: error allocating new key array (size %zu)",
216 __func__, new_keys.size);
220 /* Since @keys was full, there was no gap: */
221 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
225 /* And now the gap is at the end: */
226 keys->gap = keys->nr;
229 journal_iters_move_gap(c, keys->gap, idx);
231 move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
235 keys->d[keys->gap++] = n;
237 journal_iters_fix(c);
243 * Can only be used from the recovery thread while we're still RO - can't be
244 * used once we've got RW, as journal_keys is at that point used by multiple
247 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
248 unsigned level, struct bkey_i *k)
253 n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
258 ret = bch2_journal_key_insert_take(c, id, level, n);
264 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
265 unsigned level, struct bpos pos)
267 struct bkey_i whiteout;
269 bkey_init(&whiteout.k);
272 return bch2_journal_key_insert(c, id, level, &whiteout);
275 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
276 unsigned level, struct bpos pos)
278 struct journal_keys *keys = &c->journal_keys;
279 size_t idx = bch2_journal_key_search(keys, btree, level, pos);
281 if (idx < keys->size &&
282 keys->d[idx].btree_id == btree &&
283 keys->d[idx].level == level &&
284 !bpos_cmp(keys->d[idx].k->k.p, pos))
285 keys->d[idx].overwritten = true;
288 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
290 struct journal_key *k = iter->keys->d + iter->idx;
292 while (k < iter->keys->d + iter->keys->nr &&
293 k->btree_id == iter->btree_id &&
294 k->level == iter->level) {
299 k = iter->keys->d + iter->idx;
305 static void bch2_journal_iter_advance(struct journal_iter *iter)
307 if (iter->idx < iter->keys->size) {
309 if (iter->idx == iter->keys->gap)
310 iter->idx += iter->keys->size - iter->keys->nr;
314 static void bch2_journal_iter_exit(struct journal_iter *iter)
316 list_del(&iter->list);
319 static void bch2_journal_iter_init(struct bch_fs *c,
320 struct journal_iter *iter,
321 enum btree_id id, unsigned level,
326 iter->keys = &c->journal_keys;
327 iter->idx = bch2_journal_key_search(&c->journal_keys, id, level, pos);
330 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
332 return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
333 iter->b, &iter->unpacked);
336 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
338 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
341 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
343 switch (iter->last) {
347 bch2_journal_iter_advance_btree(iter);
350 bch2_journal_iter_advance(&iter->journal);
357 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
362 struct bkey_s_c btree_k =
363 bch2_journal_iter_peek_btree(iter);
364 struct bkey_s_c journal_k =
365 bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
367 if (btree_k.k && journal_k.k) {
368 int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
371 bch2_journal_iter_advance_btree(iter);
373 iter->last = cmp < 0 ? btree : journal;
374 } else if (btree_k.k) {
376 } else if (journal_k.k) {
377 iter->last = journal;
380 return bkey_s_c_null;
383 ret = iter->last == journal ? journal_k : btree_k;
386 bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
387 iter->journal.idx = iter->journal.keys->nr;
389 return bkey_s_c_null;
392 if (!bkey_deleted(ret.k))
395 bch2_btree_and_journal_iter_advance(iter);
401 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
403 bch2_btree_and_journal_iter_advance(iter);
405 return bch2_btree_and_journal_iter_peek(iter);
408 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
410 bch2_journal_iter_exit(&iter->journal);
413 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
416 struct btree_node_iter node_iter,
419 memset(iter, 0, sizeof(*iter));
422 iter->node_iter = node_iter;
423 bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
424 INIT_LIST_HEAD(&iter->journal.list);
428 * this version is used by btree_gc before filesystem has gone RW and
429 * multithreaded, so uses the journal_iters list:
431 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
435 struct btree_node_iter node_iter;
437 bch2_btree_node_iter_init_from_start(&node_iter, b);
438 __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
439 list_add(&iter->journal.list, &c->journal_iters);
442 /* sort and dedup all keys in the journal: */
444 void bch2_journal_entries_free(struct bch_fs *c)
446 struct journal_replay **i;
447 struct genradix_iter iter;
449 genradix_for_each(&c->journal_entries, iter, i)
451 kvpfree(*i, offsetof(struct journal_replay, j) +
452 vstruct_bytes(&(*i)->j));
453 genradix_free(&c->journal_entries);
457 * When keys compare equal, oldest compares first:
459 static int journal_sort_key_cmp(const void *_l, const void *_r)
461 const struct journal_key *l = _l;
462 const struct journal_key *r = _r;
464 return journal_key_cmp(l, r) ?:
465 cmp_int(l->journal_seq, r->journal_seq) ?:
466 cmp_int(l->journal_offset, r->journal_offset);
469 void bch2_journal_keys_free(struct journal_keys *keys)
471 struct journal_key *i;
473 move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
474 keys->gap = keys->nr;
476 for (i = keys->d; i < keys->d + keys->nr; i++)
482 keys->nr = keys->gap = keys->size = 0;
485 static int journal_keys_sort(struct bch_fs *c)
487 struct genradix_iter iter;
488 struct journal_replay *i, **_i;
489 struct jset_entry *entry;
490 struct bkey_i *k, *_n;
491 struct journal_keys *keys = &c->journal_keys;
492 struct journal_key *src, *dst;
495 genradix_for_each(&c->journal_entries, iter, _i) {
501 if (!keys->journal_seq_base)
502 keys->journal_seq_base = le64_to_cpu(i->j.seq);
504 for_each_jset_key(k, _n, entry, &i->j)
511 keys->size = roundup_pow_of_two(nr_keys);
513 keys->d = kvmalloc(sizeof(keys->d[0]) * keys->size, GFP_KERNEL);
517 genradix_for_each(&c->journal_entries, iter, _i) {
523 BUG_ON(le64_to_cpu(i->j.seq) - keys->journal_seq_base > U32_MAX);
525 for_each_jset_key(k, _n, entry, &i->j)
526 keys->d[keys->nr++] = (struct journal_key) {
527 .btree_id = entry->btree_id,
528 .level = entry->level,
530 .journal_seq = le64_to_cpu(i->j.seq) -
531 keys->journal_seq_base,
532 .journal_offset = k->_data - i->j._data,
536 sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
539 while (src < keys->d + keys->nr) {
540 while (src + 1 < keys->d + keys->nr &&
541 src[0].btree_id == src[1].btree_id &&
542 src[0].level == src[1].level &&
543 !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
549 keys->nr = dst - keys->d;
550 keys->gap = keys->nr;
554 /* journal replay: */
556 static void replay_now_at(struct journal *j, u64 seq)
558 BUG_ON(seq < j->replay_journal_seq);
560 seq = min(seq, j->replay_journal_seq_end);
562 while (j->replay_journal_seq < seq)
563 bch2_journal_pin_put(j, j->replay_journal_seq++);
566 static int bch2_journal_replay_key(struct btree_trans *trans,
567 struct journal_key *k)
569 struct btree_iter iter;
570 unsigned iter_flags =
572 BTREE_ITER_NOT_EXTENTS;
575 if (!k->level && k->btree_id == BTREE_ID_alloc)
576 iter_flags |= BTREE_ITER_CACHED;
578 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
579 BTREE_MAX_DEPTH, k->level,
581 ret = bch2_btree_iter_traverse(&iter);
585 /* Must be checked with btree locked: */
589 ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
591 bch2_trans_iter_exit(trans, &iter);
595 static int journal_sort_seq_cmp(const void *_l, const void *_r)
597 const struct journal_key *l = *((const struct journal_key **)_l);
598 const struct journal_key *r = *((const struct journal_key **)_r);
600 return cmp_int(l->journal_seq, r->journal_seq);
603 static int bch2_journal_replay(struct bch_fs *c)
605 struct journal_keys *keys = &c->journal_keys;
606 struct journal_key **keys_sorted, *k;
607 struct journal *j = &c->journal;
611 move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
612 keys->gap = keys->nr;
614 keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
618 for (i = 0; i < keys->nr; i++)
619 keys_sorted[i] = &keys->d[i];
621 sort(keys_sorted, keys->nr,
622 sizeof(keys_sorted[0]),
623 journal_sort_seq_cmp, NULL);
626 replay_now_at(j, keys->journal_seq_base);
628 for (i = 0; i < keys->nr; i++) {
633 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
635 ret = bch2_trans_do(c, NULL, NULL,
636 BTREE_INSERT_LAZY_RW|
639 ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
641 bch2_journal_replay_key(&trans, k));
643 bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
644 ret, bch2_btree_ids[k->btree_id], k->level);
649 replay_now_at(j, j->replay_journal_seq_end);
650 j->replay_journal_seq = 0;
652 bch2_journal_set_replay_done(j);
653 bch2_journal_flush_all_pins(j);
654 ret = bch2_journal_error(j);
656 if (keys->nr && !ret)
657 bch2_journal_log_msg(&c->journal, "journal replay finished");
663 /* journal replay early: */
665 static int journal_replay_entry_early(struct bch_fs *c,
666 struct jset_entry *entry)
670 switch (entry->type) {
671 case BCH_JSET_ENTRY_btree_root: {
672 struct btree_root *r;
674 if (entry->btree_id >= BTREE_ID_NR) {
675 bch_err(c, "filesystem has unknown btree type %u",
680 r = &c->btree_roots[entry->btree_id];
683 r->level = entry->level;
684 bkey_copy(&r->key, &entry->start[0]);
692 case BCH_JSET_ENTRY_usage: {
693 struct jset_entry_usage *u =
694 container_of(entry, struct jset_entry_usage, entry);
696 switch (entry->btree_id) {
697 case BCH_FS_USAGE_reserved:
698 if (entry->level < BCH_REPLICAS_MAX)
699 c->usage_base->persistent_reserved[entry->level] =
702 case BCH_FS_USAGE_inodes:
703 c->usage_base->nr_inodes = le64_to_cpu(u->v);
705 case BCH_FS_USAGE_key_version:
706 atomic64_set(&c->key_version,
713 case BCH_JSET_ENTRY_data_usage: {
714 struct jset_entry_data_usage *u =
715 container_of(entry, struct jset_entry_data_usage, entry);
717 ret = bch2_replicas_set_usage(c, &u->r,
721 case BCH_JSET_ENTRY_dev_usage: {
722 struct jset_entry_dev_usage *u =
723 container_of(entry, struct jset_entry_dev_usage, entry);
724 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
725 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
727 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
729 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
730 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
731 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
732 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
737 case BCH_JSET_ENTRY_blacklist: {
738 struct jset_entry_blacklist *bl_entry =
739 container_of(entry, struct jset_entry_blacklist, entry);
741 ret = bch2_journal_seq_blacklist_add(c,
742 le64_to_cpu(bl_entry->seq),
743 le64_to_cpu(bl_entry->seq) + 1);
746 case BCH_JSET_ENTRY_blacklist_v2: {
747 struct jset_entry_blacklist_v2 *bl_entry =
748 container_of(entry, struct jset_entry_blacklist_v2, entry);
750 ret = bch2_journal_seq_blacklist_add(c,
751 le64_to_cpu(bl_entry->start),
752 le64_to_cpu(bl_entry->end) + 1);
755 case BCH_JSET_ENTRY_clock: {
756 struct jset_entry_clock *clock =
757 container_of(entry, struct jset_entry_clock, entry);
759 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
766 static int journal_replay_early(struct bch_fs *c,
767 struct bch_sb_field_clean *clean)
769 struct jset_entry *entry;
773 for (entry = clean->start;
774 entry != vstruct_end(&clean->field);
775 entry = vstruct_next(entry)) {
776 ret = journal_replay_entry_early(c, entry);
781 struct genradix_iter iter;
782 struct journal_replay *i, **_i;
784 genradix_for_each(&c->journal_entries, iter, _i) {
790 vstruct_for_each(&i->j, entry) {
791 ret = journal_replay_entry_early(c, entry);
798 bch2_fs_usage_initialize(c);
803 /* sb clean section: */
805 static struct bkey_i *btree_root_find(struct bch_fs *c,
806 struct bch_sb_field_clean *clean,
808 enum btree_id id, unsigned *level)
811 struct jset_entry *entry, *start, *end;
814 start = clean->start;
815 end = vstruct_end(&clean->field);
818 end = vstruct_last(j);
821 for (entry = start; entry < end; entry = vstruct_next(entry))
822 if (entry->type == BCH_JSET_ENTRY_btree_root &&
823 entry->btree_id == id)
829 return ERR_PTR(-EINVAL);
832 *level = entry->level;
836 static int verify_superblock_clean(struct bch_fs *c,
837 struct bch_sb_field_clean **cleanp,
841 struct bch_sb_field_clean *clean = *cleanp;
842 struct printbuf buf1 = PRINTBUF;
843 struct printbuf buf2 = PRINTBUF;
846 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
847 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
848 le64_to_cpu(clean->journal_seq),
849 le64_to_cpu(j->seq))) {
855 for (i = 0; i < BTREE_ID_NR; i++) {
856 struct bkey_i *k1, *k2;
857 unsigned l1 = 0, l2 = 0;
859 k1 = btree_root_find(c, clean, NULL, i, &l1);
860 k2 = btree_root_find(c, NULL, j, i, &l2);
865 printbuf_reset(&buf1);
866 printbuf_reset(&buf2);
869 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
871 pr_buf(&buf1, "(none)");
874 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
876 pr_buf(&buf2, "(none)");
878 mustfix_fsck_err_on(!k1 || !k2 ||
881 k1->k.u64s != k2->k.u64s ||
882 memcmp(k1, k2, bkey_bytes(k1)) ||
884 "superblock btree root %u doesn't match journal after clean shutdown\n"
886 "journal: l=%u %s\n", i,
891 printbuf_exit(&buf2);
892 printbuf_exit(&buf1);
896 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
898 struct bch_sb_field_clean *clean, *sb_clean;
901 mutex_lock(&c->sb_lock);
902 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
904 if (fsck_err_on(!sb_clean, c,
905 "superblock marked clean but clean section not present")) {
906 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
908 mutex_unlock(&c->sb_lock);
912 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
915 mutex_unlock(&c->sb_lock);
916 return ERR_PTR(-ENOMEM);
919 ret = bch2_sb_clean_validate_late(c, clean, READ);
921 mutex_unlock(&c->sb_lock);
925 mutex_unlock(&c->sb_lock);
929 mutex_unlock(&c->sb_lock);
933 static int read_btree_roots(struct bch_fs *c)
938 for (i = 0; i < BTREE_ID_NR; i++) {
939 struct btree_root *r = &c->btree_roots[i];
944 if (i == BTREE_ID_alloc &&
945 c->opts.reconstruct_alloc) {
946 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
951 __fsck_err(c, i == BTREE_ID_alloc
952 ? FSCK_CAN_IGNORE : 0,
953 "invalid btree root %s",
955 if (i == BTREE_ID_alloc)
956 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
959 ret = bch2_btree_root_read(c, i, &r->key, r->level);
961 __fsck_err(c, i == BTREE_ID_alloc
962 ? FSCK_CAN_IGNORE : 0,
963 "error reading btree root %s",
965 if (i == BTREE_ID_alloc)
966 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
970 for (i = 0; i < BTREE_ID_NR; i++)
971 if (!c->btree_roots[i].b)
972 bch2_btree_root_alloc(c, i);
977 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
979 struct bkey_i_snapshot root_snapshot;
980 struct bkey_i_subvolume root_volume;
983 bkey_snapshot_init(&root_snapshot.k_i);
984 root_snapshot.k.p.offset = U32_MAX;
985 root_snapshot.v.flags = 0;
986 root_snapshot.v.parent = 0;
987 root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL;
988 root_snapshot.v.pad = 0;
989 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
991 ret = bch2_btree_insert(c, BTREE_ID_snapshots,
998 bkey_subvolume_init(&root_volume.k_i);
999 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1000 root_volume.v.flags = 0;
1001 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
1002 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
1004 ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1013 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1015 struct btree_iter iter;
1017 struct bch_inode_unpacked inode;
1020 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1021 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1022 k = bch2_btree_iter_peek_slot(&iter);
1027 if (!bkey_is_inode(k.k)) {
1028 bch_err(trans->c, "root inode not found");
1033 ret = bch2_inode_unpack(k, &inode);
1036 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1038 ret = bch2_inode_write(trans, &iter, &inode);
1040 bch2_trans_iter_exit(trans, &iter);
1044 int bch2_fs_recovery(struct bch_fs *c)
1046 const char *err = "cannot allocate memory";
1047 struct bch_sb_field_clean *clean = NULL;
1048 struct jset *last_journal_entry = NULL;
1049 u64 blacklist_seq, journal_seq;
1050 bool write_sb = false;
1054 clean = read_superblock_clean(c);
1055 ret = PTR_ERR_OR_ZERO(clean);
1060 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1061 le64_to_cpu(clean->journal_seq));
1063 bch_info(c, "recovering from unclean shutdown");
1065 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1066 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1072 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1073 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1078 if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1079 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1084 if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1085 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1086 c->opts.fsck = true;
1087 c->opts.fix_errors = FSCK_OPT_YES;
1090 if (!c->replicas.entries ||
1091 c->opts.rebuild_replicas) {
1092 bch_info(c, "building replicas info");
1093 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1096 if (!c->opts.nochanges) {
1097 if (c->sb.version < bcachefs_metadata_version_new_data_types) {
1098 bch_info(c, "version prior to new_data_types, upgrade and fsck required");
1099 c->opts.version_upgrade = true;
1100 c->opts.fsck = true;
1101 c->opts.fix_errors = FSCK_OPT_YES;
1105 ret = bch2_blacklist_table_initialize(c);
1107 bch_err(c, "error initializing blacklist table");
1111 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1112 struct genradix_iter iter;
1113 struct journal_replay **i;
1115 bch_verbose(c, "starting journal read");
1116 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1120 genradix_for_each_reverse(&c->journal_entries, iter, i)
1121 if (*i && !(*i)->ignore) {
1122 last_journal_entry = &(*i)->j;
1126 if (mustfix_fsck_err_on(c->sb.clean &&
1127 last_journal_entry &&
1128 !journal_entry_empty(last_journal_entry), c,
1129 "filesystem marked clean but journal not empty")) {
1130 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1131 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1132 c->sb.clean = false;
1135 if (!last_journal_entry) {
1136 fsck_err_on(!c->sb.clean, c, "no journal entries found");
1140 ret = journal_keys_sort(c);
1144 if (c->sb.clean && last_journal_entry) {
1145 ret = verify_superblock_clean(c, &clean,
1146 last_journal_entry);
1153 bch_err(c, "no superblock clean section found");
1154 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1158 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1161 if (c->opts.read_journal_only)
1164 if (c->opts.reconstruct_alloc) {
1165 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1166 drop_alloc_keys(&c->journal_keys);
1169 zero_out_btree_mem_ptr(&c->journal_keys);
1171 ret = journal_replay_early(c, clean);
1176 * After an unclean shutdown, skip then next few journal sequence
1177 * numbers as they may have been referenced by btree writes that
1178 * happened before their corresponding journal writes - those btree
1179 * writes need to be ignored, by skipping and blacklisting the next few
1180 * journal sequence numbers:
1185 if (blacklist_seq != journal_seq) {
1186 ret = bch2_journal_seq_blacklist_add(c,
1187 blacklist_seq, journal_seq);
1189 bch_err(c, "error creating new journal seq blacklist entry");
1194 ret = bch2_fs_journal_start(&c->journal, journal_seq);
1198 ret = read_btree_roots(c);
1202 bch_verbose(c, "starting alloc read");
1203 err = "error reading allocation information";
1205 down_read(&c->gc_lock);
1206 ret = bch2_alloc_read(c);
1207 up_read(&c->gc_lock);
1211 bch_verbose(c, "alloc read done");
1213 bch_verbose(c, "starting stripes_read");
1214 err = "error reading stripes";
1215 ret = bch2_stripes_read(c);
1218 bch_verbose(c, "stripes_read done");
1221 * If we're not running fsck, this ensures bch2_fsck_err() calls are
1222 * instead interpreted as bch2_inconsistent_err() calls:
1225 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1228 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1229 !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1230 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1231 bool metadata_only = c->opts.norecovery;
1233 bch_info(c, "checking allocations");
1234 err = "error checking allocations";
1235 ret = bch2_gc(c, true, metadata_only);
1238 bch_verbose(c, "done checking allocations");
1242 bch_info(c, "checking need_discard and freespace btrees");
1243 err = "error checking need_discard and freespace btrees";
1244 ret = bch2_check_alloc_info(c);
1248 ret = bch2_check_lrus(c, true);
1251 bch_verbose(c, "done checking need_discard and freespace btrees");
1254 bch2_stripes_heap_start(c);
1256 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1257 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1258 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1261 * Skip past versions that might have possibly been used (as nonces),
1262 * but hadn't had their pointers written:
1264 if (c->sb.encryption_type && !c->sb.clean)
1265 atomic64_add(1 << 16, &c->key_version);
1267 if (c->opts.norecovery)
1270 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1271 err = "journal replay failed";
1272 ret = bch2_journal_replay(c);
1275 if (c->opts.verbose || !c->sb.clean)
1276 bch_info(c, "journal replay done");
1278 err = "error initializing freespace";
1279 ret = bch2_fs_freespace_init(c);
1284 bch_info(c, "checking alloc to lru refs");
1285 err = "error checking alloc to lru refs";
1286 ret = bch2_check_alloc_to_lru_refs(c);
1290 ret = bch2_check_lrus(c, true);
1293 bch_verbose(c, "done checking alloc to lru refs");
1296 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1299 err = "error creating root snapshot node";
1300 ret = bch2_fs_initialize_subvolumes(c);
1305 bch_verbose(c, "reading snapshots table");
1306 err = "error reading snapshots table";
1307 ret = bch2_fs_snapshots_start(c);
1310 bch_verbose(c, "reading snapshots done");
1312 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1313 /* set bi_subvol on root inode */
1314 err = "error upgrade root inode for subvolumes";
1315 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1316 bch2_fs_upgrade_for_subvolumes(&trans));
1322 bch_info(c, "starting fsck");
1323 err = "error in fsck";
1324 ret = bch2_fsck_full(c);
1327 bch_verbose(c, "fsck done");
1328 } else if (!c->sb.clean) {
1329 bch_verbose(c, "checking for deleted inodes");
1330 err = "error in recovery";
1331 ret = bch2_fsck_walk_inodes_only(c);
1334 bch_verbose(c, "check inodes done");
1337 if (enabled_qtypes(c)) {
1338 bch_verbose(c, "reading quotas");
1339 ret = bch2_fs_quota_read(c);
1342 bch_verbose(c, "quotas done");
1345 mutex_lock(&c->sb_lock);
1346 if (c->opts.version_upgrade) {
1347 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1348 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1352 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1353 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1358 !test_bit(BCH_FS_ERROR, &c->flags) &&
1359 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1360 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1361 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1366 bch2_write_super(c);
1367 mutex_unlock(&c->sb_lock);
1369 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1370 !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1371 le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1372 struct bch_move_stats stats;
1374 bch_move_stats_init(&stats, "recovery");
1376 bch_info(c, "scanning for old btree nodes");
1377 ret = bch2_fs_read_write(c);
1381 ret = bch2_scan_old_btree_nodes(c, &stats);
1384 bch_info(c, "scanning for old btree nodes done");
1387 if (c->journal_seq_blacklist_table &&
1388 c->journal_seq_blacklist_table->nr > 128)
1389 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1393 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1394 bch2_flush_fsck_errs(c);
1396 if (!c->opts.keep_journal) {
1397 bch2_journal_keys_free(&c->journal_keys);
1398 bch2_journal_entries_free(c);
1402 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1404 bch_verbose(c, "ret %i", ret);
1408 bch2_fs_emergency_read_only(c);
1412 int bch2_fs_initialize(struct bch_fs *c)
1414 struct bch_inode_unpacked root_inode, lostfound_inode;
1415 struct bkey_inode_buf packed_inode;
1416 struct qstr lostfound = QSTR("lost+found");
1417 const char *err = "cannot allocate memory";
1422 bch_notice(c, "initializing new filesystem");
1424 mutex_lock(&c->sb_lock);
1425 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1426 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1428 if (c->opts.version_upgrade) {
1429 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1430 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1431 bch2_write_super(c);
1433 mutex_unlock(&c->sb_lock);
1435 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1436 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1437 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1439 for (i = 0; i < BTREE_ID_NR; i++)
1440 bch2_btree_root_alloc(c, i);
1442 for_each_online_member(ca, c, i)
1443 bch2_dev_usage_init(ca);
1445 err = "unable to allocate journal buckets";
1446 for_each_online_member(ca, c, i) {
1447 ret = bch2_dev_journal_alloc(ca);
1449 percpu_ref_put(&ca->io_ref);
1455 * journal_res_get() will crash if called before this has
1456 * set up the journal.pin FIFO and journal.cur pointer:
1458 bch2_fs_journal_start(&c->journal, 1);
1459 bch2_journal_set_replay_done(&c->journal);
1461 err = "error going read-write";
1462 ret = bch2_fs_read_write_early(c);
1467 * Write out the superblock and journal buckets, now that we can do
1470 bch_verbose(c, "marking superblocks");
1471 err = "error marking superblock and journal";
1472 for_each_member_device(ca, c, i) {
1473 ret = bch2_trans_mark_dev_sb(c, ca);
1475 percpu_ref_put(&ca->ref);
1479 ca->new_fs_bucket_idx = 0;
1482 bch_verbose(c, "initializing freespace");
1483 err = "error initializing freespace";
1484 ret = bch2_fs_freespace_init(c);
1488 err = "error creating root snapshot node";
1489 ret = bch2_fs_initialize_subvolumes(c);
1493 bch_verbose(c, "reading snapshots table");
1494 err = "error reading snapshots table";
1495 ret = bch2_fs_snapshots_start(c);
1498 bch_verbose(c, "reading snapshots done");
1500 bch2_inode_init(c, &root_inode, 0, 0,
1501 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1502 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1503 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1504 bch2_inode_pack(c, &packed_inode, &root_inode);
1505 packed_inode.inode.k.p.snapshot = U32_MAX;
1507 err = "error creating root directory";
1508 ret = bch2_btree_insert(c, BTREE_ID_inodes,
1509 &packed_inode.inode.k_i,
1514 bch2_inode_init_early(c, &lostfound_inode);
1516 err = "error creating lost+found";
1517 ret = bch2_trans_do(c, NULL, NULL, 0,
1518 bch2_create_trans(&trans,
1519 BCACHEFS_ROOT_SUBVOL_INUM,
1520 &root_inode, &lostfound_inode,
1522 0, 0, S_IFDIR|0700, 0,
1523 NULL, NULL, (subvol_inum) { 0 }, 0));
1525 bch_err(c, "error creating lost+found");
1529 if (enabled_qtypes(c)) {
1530 ret = bch2_fs_quota_read(c);
1535 err = "error writing first journal entry";
1536 ret = bch2_journal_flush(&c->journal);
1540 mutex_lock(&c->sb_lock);
1541 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1542 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1544 bch2_write_super(c);
1545 mutex_unlock(&c->sb_lock);
1549 pr_err("Error initializing new filesystem: %s (%i)", err, ret);