1 // SPDX-License-Identifier: GPL-2.0
4 #include "backpointers.h"
6 #include "alloc_background.h"
8 #include "btree_update.h"
9 #include "btree_update_interior.h"
16 #include "fs-common.h"
18 #include "journal_io.h"
19 #include "journal_reclaim.h"
20 #include "journal_seq_blacklist.h"
26 #include "subvolume.h"
29 #include <linux/sort.h>
30 #include <linux/stat.h>
32 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
34 /* for -o reconstruct_alloc: */
35 static void drop_alloc_keys(struct journal_keys *keys)
39 for (src = 0, dst = 0; src < keys->nr; src++)
40 if (keys->d[src].btree_id != BTREE_ID_alloc)
41 keys->d[dst++] = keys->d[src];
47 * Btree node pointers have a field to stack a pointer to the in memory btree
48 * node; we need to zero out this field when reading in btree nodes, or when
49 * reading in keys from the journal:
51 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
53 struct journal_key *i;
55 for (i = keys->d; i < keys->d + keys->nr; i++)
56 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
57 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
60 /* iterate over keys read from the journal: */
62 static int __journal_key_cmp(enum btree_id l_btree_id,
65 const struct journal_key *r)
67 return (cmp_int(l_btree_id, r->btree_id) ?:
68 cmp_int(l_level, r->level) ?:
69 bpos_cmp(l_pos, r->k->k.p));
72 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
74 return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
77 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
79 size_t gap_size = keys->size - keys->nr;
86 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
88 return keys->d + idx_to_pos(keys, idx);
91 static size_t __bch2_journal_key_search(struct journal_keys *keys,
92 enum btree_id id, unsigned level,
95 size_t l = 0, r = keys->nr, m;
98 m = l + ((r - l) >> 1);
99 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
105 BUG_ON(l < keys->nr &&
106 __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
109 __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
114 static size_t bch2_journal_key_search(struct journal_keys *keys,
115 enum btree_id id, unsigned level,
118 return idx_to_pos(keys, __bch2_journal_key_search(keys, id, level, pos));
121 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
122 unsigned level, struct bpos pos,
123 struct bpos end_pos, size_t *idx)
125 struct journal_keys *keys = &c->journal_keys;
127 struct journal_key *k;
130 *idx = __bch2_journal_key_search(keys, btree_id, level, pos);
132 while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
133 if (__journal_key_cmp(btree_id, level, end_pos, k) < 0)
136 if (__journal_key_cmp(btree_id, level, pos, k) <= 0 &&
151 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
152 unsigned level, struct bpos pos)
156 return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
159 static void journal_iters_fix(struct bch_fs *c)
161 struct journal_keys *keys = &c->journal_keys;
162 /* The key we just inserted is immediately before the gap: */
163 size_t gap_end = keys->gap + (keys->size - keys->nr);
164 struct btree_and_journal_iter *iter;
167 * If an iterator points one after the key we just inserted, decrement
168 * the iterator so it points at the key we just inserted - if the
169 * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
172 list_for_each_entry(iter, &c->journal_iters, journal.list)
173 if (iter->journal.idx == gap_end)
174 iter->journal.idx = keys->gap - 1;
177 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
179 struct journal_keys *keys = &c->journal_keys;
180 struct journal_iter *iter;
181 size_t gap_size = keys->size - keys->nr;
183 list_for_each_entry(iter, &c->journal_iters, list) {
184 if (iter->idx > old_gap)
185 iter->idx -= gap_size;
186 if (iter->idx >= new_gap)
187 iter->idx += gap_size;
191 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
192 unsigned level, struct bkey_i *k)
194 struct journal_key n = {
200 * Ensure these keys are done last by journal replay, to unblock
203 .journal_seq = U32_MAX,
205 struct journal_keys *keys = &c->journal_keys;
206 size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
208 BUG_ON(test_bit(BCH_FS_RW, &c->flags));
210 if (idx < keys->size &&
211 journal_key_cmp(&n, &keys->d[idx]) == 0) {
212 if (keys->d[idx].allocated)
213 kfree(keys->d[idx].k);
219 idx -= keys->size - keys->nr;
221 if (keys->nr == keys->size) {
222 struct journal_keys new_keys = {
224 .size = max_t(size_t, keys->size, 8) * 2,
227 new_keys.d = kvmalloc_array(new_keys.size, sizeof(new_keys.d[0]), GFP_KERNEL);
229 bch_err(c, "%s: error allocating new key array (size %zu)",
230 __func__, new_keys.size);
231 return -BCH_ERR_ENOMEM_journal_key_insert;
234 /* Since @keys was full, there was no gap: */
235 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
239 /* And now the gap is at the end: */
240 keys->gap = keys->nr;
243 journal_iters_move_gap(c, keys->gap, idx);
245 move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
249 keys->d[keys->gap++] = n;
251 journal_iters_fix(c);
257 * Can only be used from the recovery thread while we're still RO - can't be
258 * used once we've got RW, as journal_keys is at that point used by multiple
261 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
262 unsigned level, struct bkey_i *k)
267 n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
269 return -BCH_ERR_ENOMEM_journal_key_insert;
272 ret = bch2_journal_key_insert_take(c, id, level, n);
278 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
279 unsigned level, struct bpos pos)
281 struct bkey_i whiteout;
283 bkey_init(&whiteout.k);
286 return bch2_journal_key_insert(c, id, level, &whiteout);
289 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
290 unsigned level, struct bpos pos)
292 struct journal_keys *keys = &c->journal_keys;
293 size_t idx = bch2_journal_key_search(keys, btree, level, pos);
295 if (idx < keys->size &&
296 keys->d[idx].btree_id == btree &&
297 keys->d[idx].level == level &&
298 bpos_eq(keys->d[idx].k->k.p, pos))
299 keys->d[idx].overwritten = true;
302 static void bch2_journal_iter_advance(struct journal_iter *iter)
304 if (iter->idx < iter->keys->size) {
306 if (iter->idx == iter->keys->gap)
307 iter->idx += iter->keys->size - iter->keys->nr;
311 static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
313 struct journal_key *k = iter->keys->d + iter->idx;
315 while (k < iter->keys->d + iter->keys->size &&
316 k->btree_id == iter->btree_id &&
317 k->level == iter->level) {
319 return bkey_i_to_s_c(k->k);
321 bch2_journal_iter_advance(iter);
322 k = iter->keys->d + iter->idx;
325 return bkey_s_c_null;
328 static void bch2_journal_iter_exit(struct journal_iter *iter)
330 list_del(&iter->list);
333 static void bch2_journal_iter_init(struct bch_fs *c,
334 struct journal_iter *iter,
335 enum btree_id id, unsigned level,
340 iter->keys = &c->journal_keys;
341 iter->idx = bch2_journal_key_search(&c->journal_keys, id, level, pos);
344 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
346 return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
347 iter->b, &iter->unpacked);
350 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
352 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
355 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
357 if (bpos_eq(iter->pos, SPOS_MAX))
360 iter->pos = bpos_successor(iter->pos);
363 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
365 struct bkey_s_c btree_k, journal_k, ret;
368 return bkey_s_c_null;
370 while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
371 bpos_lt(btree_k.k->p, iter->pos))
372 bch2_journal_iter_advance_btree(iter);
374 while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
375 bpos_lt(journal_k.k->p, iter->pos))
376 bch2_journal_iter_advance(&iter->journal);
379 (!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
383 if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key))
387 iter->pos = ret.k->p;
388 if (bkey_deleted(ret.k)) {
389 bch2_btree_and_journal_iter_advance(iter);
393 iter->pos = SPOS_MAX;
400 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
402 bch2_journal_iter_exit(&iter->journal);
405 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
408 struct btree_node_iter node_iter,
411 memset(iter, 0, sizeof(*iter));
414 iter->node_iter = node_iter;
415 bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
416 INIT_LIST_HEAD(&iter->journal.list);
417 iter->pos = b->data->min_key;
418 iter->at_end = false;
422 * this version is used by btree_gc before filesystem has gone RW and
423 * multithreaded, so uses the journal_iters list:
425 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
429 struct btree_node_iter node_iter;
431 bch2_btree_node_iter_init_from_start(&node_iter, b);
432 __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
433 list_add(&iter->journal.list, &c->journal_iters);
436 /* sort and dedup all keys in the journal: */
438 void bch2_journal_entries_free(struct bch_fs *c)
440 struct journal_replay **i;
441 struct genradix_iter iter;
443 genradix_for_each(&c->journal_entries, iter, i)
445 kvpfree(*i, offsetof(struct journal_replay, j) +
446 vstruct_bytes(&(*i)->j));
447 genradix_free(&c->journal_entries);
451 * When keys compare equal, oldest compares first:
453 static int journal_sort_key_cmp(const void *_l, const void *_r)
455 const struct journal_key *l = _l;
456 const struct journal_key *r = _r;
458 return journal_key_cmp(l, r) ?:
459 cmp_int(l->journal_seq, r->journal_seq) ?:
460 cmp_int(l->journal_offset, r->journal_offset);
463 void bch2_journal_keys_free(struct journal_keys *keys)
465 struct journal_key *i;
467 move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
468 keys->gap = keys->nr;
470 for (i = keys->d; i < keys->d + keys->nr; i++)
476 keys->nr = keys->gap = keys->size = 0;
479 static void __journal_keys_sort(struct journal_keys *keys)
481 struct journal_key *src, *dst;
483 sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
486 while (src < keys->d + keys->nr) {
487 while (src + 1 < keys->d + keys->nr &&
488 src[0].btree_id == src[1].btree_id &&
489 src[0].level == src[1].level &&
490 bpos_eq(src[0].k->k.p, src[1].k->k.p))
496 keys->nr = dst - keys->d;
499 static int journal_keys_sort(struct bch_fs *c)
501 struct genradix_iter iter;
502 struct journal_replay *i, **_i;
503 struct jset_entry *entry;
505 struct journal_keys *keys = &c->journal_keys;
506 size_t nr_keys = 0, nr_read = 0;
508 genradix_for_each(&c->journal_entries, iter, _i) {
514 for_each_jset_key(k, entry, &i->j)
521 keys->size = roundup_pow_of_two(nr_keys);
523 keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
525 bch_err(c, "Failed to allocate buffer for sorted journal keys (%zu keys); trying slowpath",
530 keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
531 } while (!keys->d && keys->size > nr_keys / 8);
534 bch_err(c, "Failed to allocate %zu size buffer for sorted journal keys; exiting",
536 return -BCH_ERR_ENOMEM_journal_keys_sort;
540 genradix_for_each(&c->journal_entries, iter, _i) {
548 for_each_jset_key(k, entry, &i->j) {
549 if (keys->nr == keys->size) {
550 __journal_keys_sort(keys);
552 if (keys->nr > keys->size * 7 / 8) {
553 bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu/%zu",
554 keys->nr, keys->size, nr_read, nr_keys);
555 return -BCH_ERR_ENOMEM_journal_keys_sort;
559 keys->d[keys->nr++] = (struct journal_key) {
560 .btree_id = entry->btree_id,
561 .level = entry->level,
563 .journal_seq = le64_to_cpu(i->j.seq),
564 .journal_offset = k->_data - i->j._data,
571 __journal_keys_sort(keys);
572 keys->gap = keys->nr;
574 bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_keys, keys->nr);
578 /* journal replay: */
580 static void replay_now_at(struct journal *j, u64 seq)
582 BUG_ON(seq < j->replay_journal_seq);
584 seq = min(seq, j->replay_journal_seq_end);
586 while (j->replay_journal_seq < seq)
587 bch2_journal_pin_put(j, j->replay_journal_seq++);
590 static int bch2_journal_replay_key(struct btree_trans *trans,
591 struct journal_key *k)
593 struct btree_iter iter;
594 unsigned iter_flags =
596 BTREE_ITER_NOT_EXTENTS;
599 if (!k->level && k->btree_id == BTREE_ID_alloc)
600 iter_flags |= BTREE_ITER_CACHED;
602 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
603 BTREE_MAX_DEPTH, k->level,
605 ret = bch2_btree_iter_traverse(&iter);
609 /* Must be checked with btree locked: */
613 ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
615 bch2_trans_iter_exit(trans, &iter);
619 static int journal_sort_seq_cmp(const void *_l, const void *_r)
621 const struct journal_key *l = *((const struct journal_key **)_l);
622 const struct journal_key *r = *((const struct journal_key **)_r);
624 return cmp_int(l->journal_seq, r->journal_seq);
627 static int bch2_journal_replay(struct bch_fs *c, u64 start_seq, u64 end_seq)
629 struct journal_keys *keys = &c->journal_keys;
630 struct journal_key **keys_sorted, *k;
631 struct journal *j = &c->journal;
635 move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
636 keys->gap = keys->nr;
638 keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
640 return -BCH_ERR_ENOMEM_journal_replay;
642 for (i = 0; i < keys->nr; i++)
643 keys_sorted[i] = &keys->d[i];
645 sort(keys_sorted, keys->nr,
646 sizeof(keys_sorted[0]),
647 journal_sort_seq_cmp, NULL);
650 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
651 keys->nr, start_seq, end_seq);
656 for (i = 0; i < keys->nr; i++) {
661 replay_now_at(j, k->journal_seq);
663 ret = bch2_trans_do(c, NULL, NULL,
664 BTREE_INSERT_LAZY_RW|
667 ? BTREE_INSERT_JOURNAL_REPLAY|BCH_WATERMARK_reclaim
669 bch2_journal_replay_key(&trans, k));
671 bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
672 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
677 replay_now_at(j, j->replay_journal_seq_end);
678 j->replay_journal_seq = 0;
680 bch2_journal_set_replay_done(j);
681 bch2_journal_flush_all_pins(j);
682 ret = bch2_journal_error(j);
684 if (keys->nr && !ret)
685 bch2_journal_log_msg(c, "journal replay finished");
694 /* journal replay early: */
696 static int journal_replay_entry_early(struct bch_fs *c,
697 struct jset_entry *entry)
701 switch (entry->type) {
702 case BCH_JSET_ENTRY_btree_root: {
703 struct btree_root *r;
705 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
706 ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
711 r = bch2_btree_id_root(c, entry->btree_id);
714 r->level = entry->level;
715 bkey_copy(&r->key, &entry->start[0]);
723 case BCH_JSET_ENTRY_usage: {
724 struct jset_entry_usage *u =
725 container_of(entry, struct jset_entry_usage, entry);
727 switch (entry->btree_id) {
728 case BCH_FS_USAGE_reserved:
729 if (entry->level < BCH_REPLICAS_MAX)
730 c->usage_base->persistent_reserved[entry->level] =
733 case BCH_FS_USAGE_inodes:
734 c->usage_base->nr_inodes = le64_to_cpu(u->v);
736 case BCH_FS_USAGE_key_version:
737 atomic64_set(&c->key_version,
744 case BCH_JSET_ENTRY_data_usage: {
745 struct jset_entry_data_usage *u =
746 container_of(entry, struct jset_entry_data_usage, entry);
748 ret = bch2_replicas_set_usage(c, &u->r,
752 case BCH_JSET_ENTRY_dev_usage: {
753 struct jset_entry_dev_usage *u =
754 container_of(entry, struct jset_entry_dev_usage, entry);
755 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
756 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
758 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
760 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
761 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
762 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
763 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
768 case BCH_JSET_ENTRY_blacklist: {
769 struct jset_entry_blacklist *bl_entry =
770 container_of(entry, struct jset_entry_blacklist, entry);
772 ret = bch2_journal_seq_blacklist_add(c,
773 le64_to_cpu(bl_entry->seq),
774 le64_to_cpu(bl_entry->seq) + 1);
777 case BCH_JSET_ENTRY_blacklist_v2: {
778 struct jset_entry_blacklist_v2 *bl_entry =
779 container_of(entry, struct jset_entry_blacklist_v2, entry);
781 ret = bch2_journal_seq_blacklist_add(c,
782 le64_to_cpu(bl_entry->start),
783 le64_to_cpu(bl_entry->end) + 1);
786 case BCH_JSET_ENTRY_clock: {
787 struct jset_entry_clock *clock =
788 container_of(entry, struct jset_entry_clock, entry);
790 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
797 static int journal_replay_early(struct bch_fs *c,
798 struct bch_sb_field_clean *clean)
800 struct jset_entry *entry;
804 for (entry = clean->start;
805 entry != vstruct_end(&clean->field);
806 entry = vstruct_next(entry)) {
807 ret = journal_replay_entry_early(c, entry);
812 struct genradix_iter iter;
813 struct journal_replay *i, **_i;
815 genradix_for_each(&c->journal_entries, iter, _i) {
821 vstruct_for_each(&i->j, entry) {
822 ret = journal_replay_entry_early(c, entry);
829 bch2_fs_usage_initialize(c);
834 /* sb clean section: */
836 static struct bkey_i *btree_root_find(struct bch_fs *c,
837 struct bch_sb_field_clean *clean,
839 enum btree_id id, unsigned *level)
842 struct jset_entry *entry, *start, *end;
845 start = clean->start;
846 end = vstruct_end(&clean->field);
849 end = vstruct_last(j);
852 for (entry = start; entry < end; entry = vstruct_next(entry))
853 if (entry->type == BCH_JSET_ENTRY_btree_root &&
854 entry->btree_id == id)
860 return ERR_PTR(-EINVAL);
863 *level = entry->level;
867 static int verify_superblock_clean(struct bch_fs *c,
868 struct bch_sb_field_clean **cleanp,
872 struct bch_sb_field_clean *clean = *cleanp;
873 struct printbuf buf1 = PRINTBUF;
874 struct printbuf buf2 = PRINTBUF;
877 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
878 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
879 le64_to_cpu(clean->journal_seq),
880 le64_to_cpu(j->seq))) {
886 for (i = 0; i < BTREE_ID_NR; i++) {
887 struct bkey_i *k1, *k2;
888 unsigned l1 = 0, l2 = 0;
890 k1 = btree_root_find(c, clean, NULL, i, &l1);
891 k2 = btree_root_find(c, NULL, j, i, &l2);
896 printbuf_reset(&buf1);
897 printbuf_reset(&buf2);
900 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
902 prt_printf(&buf1, "(none)");
905 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
907 prt_printf(&buf2, "(none)");
909 mustfix_fsck_err_on(!k1 || !k2 ||
912 k1->k.u64s != k2->k.u64s ||
913 memcmp(k1, k2, bkey_bytes(&k1->k)) ||
915 "superblock btree root %u doesn't match journal after clean shutdown\n"
917 "journal: l=%u %s\n", i,
922 printbuf_exit(&buf2);
923 printbuf_exit(&buf1);
927 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
929 struct bch_sb_field_clean *clean, *sb_clean;
932 mutex_lock(&c->sb_lock);
933 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
935 if (fsck_err_on(!sb_clean, c,
936 "superblock marked clean but clean section not present")) {
937 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
939 mutex_unlock(&c->sb_lock);
943 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
946 mutex_unlock(&c->sb_lock);
947 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
950 ret = bch2_sb_clean_validate_late(c, clean, READ);
952 mutex_unlock(&c->sb_lock);
956 mutex_unlock(&c->sb_lock);
960 mutex_unlock(&c->sb_lock);
964 static bool btree_id_is_alloc(enum btree_id id)
968 case BTREE_ID_backpointers:
969 case BTREE_ID_need_discard:
970 case BTREE_ID_freespace:
971 case BTREE_ID_bucket_gens:
978 static int read_btree_roots(struct bch_fs *c)
983 for (i = 0; i < btree_id_nr_alive(c); i++) {
984 struct btree_root *r = bch2_btree_id_root(c, i);
989 if (btree_id_is_alloc(i) &&
990 c->opts.reconstruct_alloc) {
991 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
996 __fsck_err(c, btree_id_is_alloc(i)
997 ? FSCK_CAN_IGNORE : 0,
998 "invalid btree root %s",
1000 if (i == BTREE_ID_alloc)
1001 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1004 ret = bch2_btree_root_read(c, i, &r->key, r->level);
1007 btree_id_is_alloc(i)
1008 ? FSCK_CAN_IGNORE : 0,
1009 "error reading btree root %s",
1011 if (btree_id_is_alloc(i))
1012 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1016 for (i = 0; i < BTREE_ID_NR; i++) {
1017 struct btree_root *r = bch2_btree_id_root(c, i);
1022 bch2_btree_root_alloc(c, i);
1029 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
1031 struct bkey_i_snapshot_tree root_tree;
1032 struct bkey_i_snapshot root_snapshot;
1033 struct bkey_i_subvolume root_volume;
1036 bkey_snapshot_tree_init(&root_tree.k_i);
1037 root_tree.k.p.offset = 1;
1038 root_tree.v.master_subvol = cpu_to_le32(1);
1039 root_tree.v.root_snapshot = cpu_to_le32(U32_MAX);
1041 bkey_snapshot_init(&root_snapshot.k_i);
1042 root_snapshot.k.p.offset = U32_MAX;
1043 root_snapshot.v.flags = 0;
1044 root_snapshot.v.parent = 0;
1045 root_snapshot.v.subvol = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
1046 root_snapshot.v.tree = cpu_to_le32(1);
1047 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
1049 bkey_subvolume_init(&root_volume.k_i);
1050 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1051 root_volume.v.flags = 0;
1052 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
1053 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
1055 ret = bch2_btree_insert(c, BTREE_ID_snapshot_trees,
1058 bch2_btree_insert(c, BTREE_ID_snapshots,
1061 bch2_btree_insert(c, BTREE_ID_subvolumes,
1069 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1071 struct btree_iter iter;
1073 struct bch_inode_unpacked inode;
1076 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1077 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1082 if (!bkey_is_inode(k.k)) {
1083 bch_err(trans->c, "root inode not found");
1084 ret = -BCH_ERR_ENOENT_inode;
1088 ret = bch2_inode_unpack(k, &inode);
1091 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1093 ret = bch2_inode_write(trans, &iter, &inode);
1095 bch2_trans_iter_exit(trans, &iter);
1099 /* set bi_subvol on root inode */
1101 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
1103 int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1104 __bch2_fs_upgrade_for_subvolumes(&trans));
1110 int bch2_fs_recovery(struct bch_fs *c)
1112 struct bch_sb_field_clean *clean = NULL;
1113 struct jset *last_journal_entry = NULL;
1114 u64 last_seq, blacklist_seq, journal_seq;
1115 bool write_sb = false;
1119 clean = read_superblock_clean(c);
1120 ret = PTR_ERR_OR_ZERO(clean);
1125 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1126 le64_to_cpu(clean->journal_seq));
1128 bch_info(c, "recovering from unclean shutdown");
1130 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1131 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1137 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1138 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1143 if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1144 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1149 if (!c->opts.nochanges &&
1150 c->sb.version < bcachefs_metadata_required_upgrade_below) {
1151 struct printbuf buf = PRINTBUF;
1153 prt_str(&buf, "version ");
1154 bch2_version_to_text(&buf, c->sb.version);
1155 prt_str(&buf, " prior to ");
1156 bch2_version_to_text(&buf, bcachefs_metadata_required_upgrade_below);
1157 prt_str(&buf, ", upgrade and fsck required");
1159 bch_info(c, "%s", buf.buf);
1160 printbuf_exit(&buf);
1162 c->opts.version_upgrade = true;
1163 c->opts.fsck = true;
1164 c->opts.fix_errors = FSCK_OPT_YES;
1167 if (c->opts.fsck && c->opts.norecovery) {
1168 bch_err(c, "cannot select both norecovery and fsck");
1173 ret = bch2_blacklist_table_initialize(c);
1175 bch_err(c, "error initializing blacklist table");
1179 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1180 struct genradix_iter iter;
1181 struct journal_replay **i;
1183 bch_verbose(c, "starting journal read");
1184 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
1189 * note: cmd_list_journal needs the blacklist table fully up to date so
1190 * it can asterisk ignored journal entries:
1192 if (c->opts.read_journal_only)
1195 genradix_for_each_reverse(&c->journal_entries, iter, i)
1196 if (*i && !(*i)->ignore) {
1197 last_journal_entry = &(*i)->j;
1201 if (mustfix_fsck_err_on(c->sb.clean &&
1202 last_journal_entry &&
1203 !journal_entry_empty(last_journal_entry), c,
1204 "filesystem marked clean but journal not empty")) {
1205 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1206 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1207 c->sb.clean = false;
1210 if (!last_journal_entry) {
1211 fsck_err_on(!c->sb.clean, c, "no journal entries found");
1215 genradix_for_each_reverse(&c->journal_entries, iter, i)
1217 last_journal_entry = &(*i)->j;
1218 (*i)->ignore = false;
1223 ret = journal_keys_sort(c);
1227 if (c->sb.clean && last_journal_entry) {
1228 ret = verify_superblock_clean(c, &clean,
1229 last_journal_entry);
1236 bch_err(c, "no superblock clean section found");
1237 ret = -BCH_ERR_fsck_repair_impossible;
1241 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1244 if (c->opts.reconstruct_alloc) {
1245 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1246 drop_alloc_keys(&c->journal_keys);
1249 zero_out_btree_mem_ptr(&c->journal_keys);
1251 ret = journal_replay_early(c, clean);
1256 * After an unclean shutdown, skip then next few journal sequence
1257 * numbers as they may have been referenced by btree writes that
1258 * happened before their corresponding journal writes - those btree
1259 * writes need to be ignored, by skipping and blacklisting the next few
1260 * journal sequence numbers:
1265 if (blacklist_seq != journal_seq) {
1266 ret = bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
1267 blacklist_seq, journal_seq) ?:
1268 bch2_journal_seq_blacklist_add(c,
1269 blacklist_seq, journal_seq);
1271 bch_err(c, "error creating new journal seq blacklist entry");
1276 ret = bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
1277 journal_seq, last_seq, blacklist_seq - 1) ?:
1278 bch2_fs_journal_start(&c->journal, journal_seq);
1282 if (c->opts.reconstruct_alloc)
1283 bch2_journal_log_msg(c, "dropping alloc info");
1286 * Skip past versions that might have possibly been used (as nonces),
1287 * but hadn't had their pointers written:
1289 if (c->sb.encryption_type && !c->sb.clean)
1290 atomic64_add(1 << 16, &c->key_version);
1292 ret = read_btree_roots(c);
1296 bch_verbose(c, "starting alloc read");
1297 down_read(&c->gc_lock);
1298 ret = c->sb.version < bcachefs_metadata_version_bucket_gens
1299 ? bch2_alloc_read(c)
1300 : bch2_bucket_gens_read(c);
1301 up_read(&c->gc_lock);
1304 bch_verbose(c, "alloc read done");
1306 bch_verbose(c, "starting stripes_read");
1307 ret = bch2_stripes_read(c);
1310 bch_verbose(c, "stripes_read done");
1312 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1313 ret = bch2_fs_initialize_subvolumes(c);
1318 bch_verbose(c, "reading snapshots table");
1319 ret = bch2_fs_snapshots_start(c);
1322 bch_verbose(c, "reading snapshots done");
1325 bool metadata_only = c->opts.norecovery;
1327 bch_info(c, "checking allocations");
1328 ret = bch2_gc(c, true, metadata_only);
1331 bch_verbose(c, "done checking allocations");
1333 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1335 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1337 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1338 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1341 if (c->opts.verbose || !c->sb.clean)
1342 bch_info(c, "journal replay done");
1344 bch_info(c, "checking need_discard and freespace btrees");
1345 ret = bch2_check_alloc_info(c);
1348 bch_verbose(c, "done checking need_discard and freespace btrees");
1350 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1352 bch_info(c, "checking lrus");
1353 ret = bch2_check_lrus(c);
1356 bch_verbose(c, "done checking lrus");
1357 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1359 bch_info(c, "checking backpointers to alloc keys");
1360 ret = bch2_check_btree_backpointers(c);
1363 bch_verbose(c, "done checking backpointers to alloc keys");
1365 bch_info(c, "checking backpointers to extents");
1366 ret = bch2_check_backpointers_to_extents(c);
1369 bch_verbose(c, "done checking backpointers to extents");
1371 bch_info(c, "checking extents to backpointers");
1372 ret = bch2_check_extents_to_backpointers(c);
1375 bch_verbose(c, "done checking extents to backpointers");
1376 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1378 bch_info(c, "checking alloc to lru refs");
1379 ret = bch2_check_alloc_to_lru_refs(c);
1382 bch_verbose(c, "done checking alloc to lru refs");
1383 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1385 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1386 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1387 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1388 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1389 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1390 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1392 if (c->opts.norecovery)
1395 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1397 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1398 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1401 if (c->opts.verbose || !c->sb.clean)
1402 bch_info(c, "journal replay done");
1405 ret = bch2_fs_freespace_init(c);
1409 if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1410 c->opts.version_upgrade) {
1411 bch_info(c, "initializing bucket_gens");
1412 ret = bch2_bucket_gens_init(c);
1415 bch_verbose(c, "bucket_gens init done");
1418 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1419 ret = bch2_fs_upgrade_for_subvolumes(c);
1425 ret = bch2_fsck_full(c);
1428 bch_verbose(c, "fsck done");
1429 } else if (!c->sb.clean) {
1430 bch_verbose(c, "checking for deleted inodes");
1431 ret = bch2_fsck_walk_inodes_only(c);
1434 bch_verbose(c, "check inodes done");
1437 if (enabled_qtypes(c)) {
1438 bch_verbose(c, "reading quotas");
1439 ret = bch2_fs_quota_read(c);
1442 bch_verbose(c, "quotas done");
1445 mutex_lock(&c->sb_lock);
1446 if (c->opts.version_upgrade) {
1447 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1448 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1452 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1453 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1458 !test_bit(BCH_FS_ERROR, &c->flags) &&
1459 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1460 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1461 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1466 bch2_write_super(c);
1467 mutex_unlock(&c->sb_lock);
1469 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1470 !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1471 c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
1472 struct bch_move_stats stats;
1474 bch2_move_stats_init(&stats, "recovery");
1476 bch_info(c, "scanning for old btree nodes");
1477 ret = bch2_fs_read_write(c) ?:
1478 bch2_scan_old_btree_nodes(c, &stats);
1481 bch_info(c, "scanning for old btree nodes done");
1484 if (c->journal_seq_blacklist_table &&
1485 c->journal_seq_blacklist_table->nr > 128)
1486 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1490 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1491 bch2_flush_fsck_errs(c);
1493 if (!c->opts.keep_journal &&
1494 test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1495 bch2_journal_keys_free(&c->journal_keys);
1496 bch2_journal_entries_free(c);
1500 if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1501 bch2_fs_read_write_early(c);
1502 bch2_delete_dead_snapshots_async(c);
1508 bch_verbose(c, "ret %s", bch2_err_str(ret));
1512 bch2_fs_emergency_read_only(c);
1516 int bch2_fs_initialize(struct bch_fs *c)
1518 struct bch_inode_unpacked root_inode, lostfound_inode;
1519 struct bkey_inode_buf packed_inode;
1520 struct qstr lostfound = QSTR("lost+found");
1525 bch_notice(c, "initializing new filesystem");
1527 mutex_lock(&c->sb_lock);
1528 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1529 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1531 if (c->sb.version < bcachefs_metadata_version_inode_v3)
1532 c->opts.version_upgrade = true;
1534 if (c->opts.version_upgrade) {
1535 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1536 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1537 bch2_write_super(c);
1539 mutex_unlock(&c->sb_lock);
1541 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1542 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1543 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1544 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1545 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1546 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1548 for (i = 0; i < BTREE_ID_NR; i++)
1549 bch2_btree_root_alloc(c, i);
1551 for_each_online_member(ca, c, i)
1552 bch2_dev_usage_init(ca);
1554 for_each_online_member(ca, c, i) {
1555 ret = bch2_dev_journal_alloc(ca);
1557 percpu_ref_put(&ca->io_ref);
1563 * journal_res_get() will crash if called before this has
1564 * set up the journal.pin FIFO and journal.cur pointer:
1566 bch2_fs_journal_start(&c->journal, 1);
1567 bch2_journal_set_replay_done(&c->journal);
1569 ret = bch2_fs_read_write_early(c);
1574 * Write out the superblock and journal buckets, now that we can do
1577 bch_verbose(c, "marking superblocks");
1578 for_each_member_device(ca, c, i) {
1579 ret = bch2_trans_mark_dev_sb(c, ca);
1581 percpu_ref_put(&ca->ref);
1585 ca->new_fs_bucket_idx = 0;
1588 ret = bch2_fs_freespace_init(c);
1592 ret = bch2_fs_initialize_subvolumes(c);
1596 bch_verbose(c, "reading snapshots table");
1597 ret = bch2_fs_snapshots_start(c);
1600 bch_verbose(c, "reading snapshots done");
1602 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1603 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1604 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1605 bch2_inode_pack(&packed_inode, &root_inode);
1606 packed_inode.inode.k.p.snapshot = U32_MAX;
1608 ret = bch2_btree_insert(c, BTREE_ID_inodes,
1609 &packed_inode.inode.k_i,
1612 bch_err_msg(c, ret, "creating root directory");
1616 bch2_inode_init_early(c, &lostfound_inode);
1618 ret = bch2_trans_do(c, NULL, NULL, 0,
1619 bch2_create_trans(&trans,
1620 BCACHEFS_ROOT_SUBVOL_INUM,
1621 &root_inode, &lostfound_inode,
1623 0, 0, S_IFDIR|0700, 0,
1624 NULL, NULL, (subvol_inum) { 0 }, 0));
1626 bch_err_msg(c, ret, "creating lost+found");
1630 if (enabled_qtypes(c)) {
1631 ret = bch2_fs_quota_read(c);
1636 ret = bch2_journal_flush(&c->journal);
1638 bch_err_msg(c, ret, "writing first journal entry");
1642 mutex_lock(&c->sb_lock);
1643 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1644 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1646 bch2_write_super(c);
1647 mutex_unlock(&c->sb_lock);
1651 bch_err_fn(ca, ret);