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 (*idx < keys->nr &&
133 (k = idx_to_key(keys, *idx),
134 k->btree_id == btree_id &&
136 bpos_le(k->k->k.p, end_pos))) {
137 if (bpos_ge(k->k->k.p, pos) && !k->overwritten)
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);
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);
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 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 int journal_keys_sort(struct bch_fs *c)
481 struct genradix_iter iter;
482 struct journal_replay *i, **_i;
483 struct jset_entry *entry;
484 struct bkey_i *k, *_n;
485 struct journal_keys *keys = &c->journal_keys;
486 struct journal_key *src, *dst;
489 genradix_for_each(&c->journal_entries, iter, _i) {
495 for_each_jset_key(k, _n, entry, &i->j)
502 keys->size = roundup_pow_of_two(nr_keys);
504 keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
508 genradix_for_each(&c->journal_entries, iter, _i) {
514 for_each_jset_key(k, _n, entry, &i->j)
515 keys->d[keys->nr++] = (struct journal_key) {
516 .btree_id = entry->btree_id,
517 .level = entry->level,
519 .journal_seq = le64_to_cpu(i->j.seq),
520 .journal_offset = k->_data - i->j._data,
524 sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
527 while (src < keys->d + keys->nr) {
528 while (src + 1 < keys->d + keys->nr &&
529 src[0].btree_id == src[1].btree_id &&
530 src[0].level == src[1].level &&
531 bpos_eq(src[0].k->k.p, src[1].k->k.p))
537 keys->nr = dst - keys->d;
538 keys->gap = keys->nr;
542 /* journal replay: */
544 static void replay_now_at(struct journal *j, u64 seq)
546 BUG_ON(seq < j->replay_journal_seq);
548 seq = min(seq, j->replay_journal_seq_end);
550 while (j->replay_journal_seq < seq)
551 bch2_journal_pin_put(j, j->replay_journal_seq++);
554 static int bch2_journal_replay_key(struct btree_trans *trans,
555 struct journal_key *k)
557 struct btree_iter iter;
558 unsigned iter_flags =
560 BTREE_ITER_NOT_EXTENTS;
563 if (!k->level && k->btree_id == BTREE_ID_alloc)
564 iter_flags |= BTREE_ITER_CACHED;
566 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
567 BTREE_MAX_DEPTH, k->level,
569 ret = bch2_btree_iter_traverse(&iter);
573 /* Must be checked with btree locked: */
577 ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
579 bch2_trans_iter_exit(trans, &iter);
583 static int journal_sort_seq_cmp(const void *_l, const void *_r)
585 const struct journal_key *l = *((const struct journal_key **)_l);
586 const struct journal_key *r = *((const struct journal_key **)_r);
588 return cmp_int(l->journal_seq, r->journal_seq);
591 static int bch2_journal_replay(struct bch_fs *c)
593 struct journal_keys *keys = &c->journal_keys;
594 struct journal_key **keys_sorted, *k;
595 struct journal *j = &c->journal;
599 move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
600 keys->gap = keys->nr;
602 keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
606 for (i = 0; i < keys->nr; i++)
607 keys_sorted[i] = &keys->d[i];
609 sort(keys_sorted, keys->nr,
610 sizeof(keys_sorted[0]),
611 journal_sort_seq_cmp, NULL);
613 for (i = 0; i < keys->nr; i++) {
618 replay_now_at(j, k->journal_seq);
620 ret = bch2_trans_do(c, NULL, NULL,
621 BTREE_INSERT_LAZY_RW|
624 ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
626 bch2_journal_replay_key(&trans, k));
628 bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
629 ret, bch2_btree_ids[k->btree_id], k->level);
634 replay_now_at(j, j->replay_journal_seq_end);
635 j->replay_journal_seq = 0;
637 bch2_journal_set_replay_done(j);
638 bch2_journal_flush_all_pins(j);
639 ret = bch2_journal_error(j);
641 if (keys->nr && !ret)
642 bch2_journal_log_msg(&c->journal, "journal replay finished");
648 /* journal replay early: */
650 static int journal_replay_entry_early(struct bch_fs *c,
651 struct jset_entry *entry)
655 switch (entry->type) {
656 case BCH_JSET_ENTRY_btree_root: {
657 struct btree_root *r;
659 if (entry->btree_id >= BTREE_ID_NR) {
660 bch_err(c, "filesystem has unknown btree type %u",
665 r = &c->btree_roots[entry->btree_id];
668 r->level = entry->level;
669 bkey_copy(&r->key, &entry->start[0]);
677 case BCH_JSET_ENTRY_usage: {
678 struct jset_entry_usage *u =
679 container_of(entry, struct jset_entry_usage, entry);
681 switch (entry->btree_id) {
682 case BCH_FS_USAGE_reserved:
683 if (entry->level < BCH_REPLICAS_MAX)
684 c->usage_base->persistent_reserved[entry->level] =
687 case BCH_FS_USAGE_inodes:
688 c->usage_base->nr_inodes = le64_to_cpu(u->v);
690 case BCH_FS_USAGE_key_version:
691 atomic64_set(&c->key_version,
698 case BCH_JSET_ENTRY_data_usage: {
699 struct jset_entry_data_usage *u =
700 container_of(entry, struct jset_entry_data_usage, entry);
702 ret = bch2_replicas_set_usage(c, &u->r,
706 case BCH_JSET_ENTRY_dev_usage: {
707 struct jset_entry_dev_usage *u =
708 container_of(entry, struct jset_entry_dev_usage, entry);
709 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
710 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
712 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
714 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
715 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
716 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
717 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
722 case BCH_JSET_ENTRY_blacklist: {
723 struct jset_entry_blacklist *bl_entry =
724 container_of(entry, struct jset_entry_blacklist, entry);
726 ret = bch2_journal_seq_blacklist_add(c,
727 le64_to_cpu(bl_entry->seq),
728 le64_to_cpu(bl_entry->seq) + 1);
731 case BCH_JSET_ENTRY_blacklist_v2: {
732 struct jset_entry_blacklist_v2 *bl_entry =
733 container_of(entry, struct jset_entry_blacklist_v2, entry);
735 ret = bch2_journal_seq_blacklist_add(c,
736 le64_to_cpu(bl_entry->start),
737 le64_to_cpu(bl_entry->end) + 1);
740 case BCH_JSET_ENTRY_clock: {
741 struct jset_entry_clock *clock =
742 container_of(entry, struct jset_entry_clock, entry);
744 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
751 static int journal_replay_early(struct bch_fs *c,
752 struct bch_sb_field_clean *clean)
754 struct jset_entry *entry;
758 for (entry = clean->start;
759 entry != vstruct_end(&clean->field);
760 entry = vstruct_next(entry)) {
761 ret = journal_replay_entry_early(c, entry);
766 struct genradix_iter iter;
767 struct journal_replay *i, **_i;
769 genradix_for_each(&c->journal_entries, iter, _i) {
775 vstruct_for_each(&i->j, entry) {
776 ret = journal_replay_entry_early(c, entry);
783 bch2_fs_usage_initialize(c);
788 /* sb clean section: */
790 static struct bkey_i *btree_root_find(struct bch_fs *c,
791 struct bch_sb_field_clean *clean,
793 enum btree_id id, unsigned *level)
796 struct jset_entry *entry, *start, *end;
799 start = clean->start;
800 end = vstruct_end(&clean->field);
803 end = vstruct_last(j);
806 for (entry = start; entry < end; entry = vstruct_next(entry))
807 if (entry->type == BCH_JSET_ENTRY_btree_root &&
808 entry->btree_id == id)
814 return ERR_PTR(-EINVAL);
817 *level = entry->level;
821 static int verify_superblock_clean(struct bch_fs *c,
822 struct bch_sb_field_clean **cleanp,
826 struct bch_sb_field_clean *clean = *cleanp;
827 struct printbuf buf1 = PRINTBUF;
828 struct printbuf buf2 = PRINTBUF;
831 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
832 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
833 le64_to_cpu(clean->journal_seq),
834 le64_to_cpu(j->seq))) {
840 for (i = 0; i < BTREE_ID_NR; i++) {
841 struct bkey_i *k1, *k2;
842 unsigned l1 = 0, l2 = 0;
844 k1 = btree_root_find(c, clean, NULL, i, &l1);
845 k2 = btree_root_find(c, NULL, j, i, &l2);
850 printbuf_reset(&buf1);
851 printbuf_reset(&buf2);
854 bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
856 prt_printf(&buf1, "(none)");
859 bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
861 prt_printf(&buf2, "(none)");
863 mustfix_fsck_err_on(!k1 || !k2 ||
866 k1->k.u64s != k2->k.u64s ||
867 memcmp(k1, k2, bkey_bytes(k1)) ||
869 "superblock btree root %u doesn't match journal after clean shutdown\n"
871 "journal: l=%u %s\n", i,
876 printbuf_exit(&buf2);
877 printbuf_exit(&buf1);
881 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
883 struct bch_sb_field_clean *clean, *sb_clean;
886 mutex_lock(&c->sb_lock);
887 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
889 if (fsck_err_on(!sb_clean, c,
890 "superblock marked clean but clean section not present")) {
891 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
893 mutex_unlock(&c->sb_lock);
897 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
900 mutex_unlock(&c->sb_lock);
901 return ERR_PTR(-ENOMEM);
904 ret = bch2_sb_clean_validate_late(c, clean, READ);
906 mutex_unlock(&c->sb_lock);
910 mutex_unlock(&c->sb_lock);
914 mutex_unlock(&c->sb_lock);
918 static bool btree_id_is_alloc(enum btree_id id)
922 case BTREE_ID_backpointers:
923 case BTREE_ID_need_discard:
924 case BTREE_ID_freespace:
931 static int read_btree_roots(struct bch_fs *c)
936 for (i = 0; i < BTREE_ID_NR; i++) {
937 struct btree_root *r = &c->btree_roots[i];
942 if (btree_id_is_alloc(i) &&
943 c->opts.reconstruct_alloc) {
944 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
949 __fsck_err(c, btree_id_is_alloc(i)
950 ? FSCK_CAN_IGNORE : 0,
951 "invalid btree root %s",
953 if (i == BTREE_ID_alloc)
954 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
957 ret = bch2_btree_root_read(c, i, &r->key, r->level);
961 ? FSCK_CAN_IGNORE : 0,
962 "error reading btree root %s",
964 if (i == BTREE_ID_alloc)
965 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
969 for (i = 0; i < BTREE_ID_NR; i++)
970 if (!c->btree_roots[i].b)
971 bch2_btree_root_alloc(c, i);
976 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
978 struct bkey_i_snapshot root_snapshot;
979 struct bkey_i_subvolume root_volume;
982 bkey_snapshot_init(&root_snapshot.k_i);
983 root_snapshot.k.p.offset = U32_MAX;
984 root_snapshot.v.flags = 0;
985 root_snapshot.v.parent = 0;
986 root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL;
987 root_snapshot.v.pad = 0;
988 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
990 ret = bch2_btree_insert(c, BTREE_ID_snapshots,
996 bkey_subvolume_init(&root_volume.k_i);
997 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
998 root_volume.v.flags = 0;
999 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
1000 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
1002 ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1011 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1013 struct btree_iter iter;
1015 struct bch_inode_unpacked inode;
1018 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1019 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1020 k = bch2_btree_iter_peek_slot(&iter);
1025 if (!bkey_is_inode(k.k)) {
1026 bch_err(trans->c, "root inode not found");
1031 ret = bch2_inode_unpack(k, &inode);
1034 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1036 ret = bch2_inode_write(trans, &iter, &inode);
1038 bch2_trans_iter_exit(trans, &iter);
1042 int bch2_fs_recovery(struct bch_fs *c)
1044 const char *err = "cannot allocate memory";
1045 struct bch_sb_field_clean *clean = NULL;
1046 struct jset *last_journal_entry = NULL;
1047 u64 blacklist_seq, journal_seq;
1048 bool write_sb = false;
1052 clean = read_superblock_clean(c);
1053 ret = PTR_ERR_OR_ZERO(clean);
1058 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1059 le64_to_cpu(clean->journal_seq));
1061 bch_info(c, "recovering from unclean shutdown");
1063 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1064 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1070 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1071 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1076 if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1077 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1082 if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1083 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1084 c->opts.fsck = true;
1085 c->opts.fix_errors = FSCK_OPT_YES;
1088 if (!c->opts.nochanges) {
1089 if (c->sb.version < bcachefs_metadata_version_backpointers) {
1090 bch_info(c, "version prior to backpointers, upgrade and fsck required");
1091 c->opts.version_upgrade = true;
1092 c->opts.fsck = true;
1093 c->opts.fix_errors = FSCK_OPT_YES;
1094 } else if (c->sb.version < bcachefs_metadata_version_inode_v3) {
1095 bch_info(c, "version prior to inode_v3, upgrade required");
1096 c->opts.version_upgrade = true;
1100 if (c->opts.fsck && c->opts.norecovery) {
1101 bch_err(c, "cannot select both norecovery and fsck");
1106 ret = bch2_blacklist_table_initialize(c);
1108 bch_err(c, "error initializing blacklist table");
1112 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1113 struct genradix_iter iter;
1114 struct journal_replay **i;
1116 bch_verbose(c, "starting journal read");
1117 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1121 genradix_for_each_reverse(&c->journal_entries, iter, i)
1122 if (*i && !(*i)->ignore) {
1123 last_journal_entry = &(*i)->j;
1127 if (mustfix_fsck_err_on(c->sb.clean &&
1128 last_journal_entry &&
1129 !journal_entry_empty(last_journal_entry), c,
1130 "filesystem marked clean but journal not empty")) {
1131 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1132 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1133 c->sb.clean = false;
1136 if (!last_journal_entry) {
1137 fsck_err_on(!c->sb.clean, c, "no journal entries found");
1141 ret = journal_keys_sort(c);
1145 if (c->sb.clean && last_journal_entry) {
1146 ret = verify_superblock_clean(c, &clean,
1147 last_journal_entry);
1154 bch_err(c, "no superblock clean section found");
1155 ret = -BCH_ERR_fsck_repair_impossible;
1159 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1162 if (c->opts.reconstruct_alloc) {
1163 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1164 drop_alloc_keys(&c->journal_keys);
1167 zero_out_btree_mem_ptr(&c->journal_keys);
1169 ret = journal_replay_early(c, clean);
1174 * After an unclean shutdown, skip then next few journal sequence
1175 * numbers as they may have been referenced by btree writes that
1176 * happened before their corresponding journal writes - those btree
1177 * writes need to be ignored, by skipping and blacklisting the next few
1178 * journal sequence numbers:
1183 if (blacklist_seq != journal_seq) {
1184 ret = bch2_journal_seq_blacklist_add(c,
1185 blacklist_seq, journal_seq);
1187 bch_err(c, "error creating new journal seq blacklist entry");
1193 * note: cmd_list_journal needs the blacklist table fully up to date so
1194 * it can asterisk ignored journal entries:
1196 if (c->opts.read_journal_only)
1199 ret = bch2_fs_journal_start(&c->journal, journal_seq);
1204 * Skip past versions that might have possibly been used (as nonces),
1205 * but hadn't had their pointers written:
1207 if (c->sb.encryption_type && !c->sb.clean)
1208 atomic64_add(1 << 16, &c->key_version);
1210 ret = read_btree_roots(c);
1214 bch_verbose(c, "starting alloc read");
1215 err = "error reading allocation information";
1217 down_read(&c->gc_lock);
1218 ret = bch2_alloc_read(c);
1219 up_read(&c->gc_lock);
1223 bch_verbose(c, "alloc read done");
1225 bch_verbose(c, "starting stripes_read");
1226 err = "error reading stripes";
1227 ret = bch2_stripes_read(c);
1230 bch_verbose(c, "stripes_read done");
1232 bch2_stripes_heap_start(c);
1235 bool metadata_only = c->opts.norecovery;
1237 bch_info(c, "checking allocations");
1238 err = "error checking allocations";
1239 ret = bch2_gc(c, true, metadata_only);
1242 bch_verbose(c, "done checking allocations");
1244 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1246 bch_info(c, "checking need_discard and freespace btrees");
1247 err = "error checking need_discard and freespace btrees";
1248 ret = bch2_check_alloc_info(c);
1251 bch_verbose(c, "done checking need_discard and freespace btrees");
1253 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1254 err = "error creating root snapshot node";
1255 ret = bch2_fs_initialize_subvolumes(c);
1260 bch_verbose(c, "reading snapshots table");
1261 err = "error reading snapshots table";
1262 ret = bch2_fs_snapshots_start(c);
1265 bch_verbose(c, "reading snapshots done");
1267 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1269 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1270 err = "journal replay failed";
1271 ret = bch2_journal_replay(c);
1274 if (c->opts.verbose || !c->sb.clean)
1275 bch_info(c, "journal replay done");
1277 bch_info(c, "checking lrus");
1278 err = "error checking lrus";
1279 ret = bch2_check_lrus(c);
1282 bch_verbose(c, "done checking lrus");
1283 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1285 bch_info(c, "checking backpointers to alloc keys");
1286 err = "error checking backpointers to alloc keys";
1287 ret = bch2_check_btree_backpointers(c);
1290 bch_verbose(c, "done checking backpointers to alloc keys");
1292 bch_info(c, "checking backpointers to extents");
1293 err = "error checking backpointers to extents";
1294 ret = bch2_check_backpointers_to_extents(c);
1297 bch_verbose(c, "done checking backpointers to extents");
1299 bch_info(c, "checking extents to backpointers");
1300 err = "error checking extents to backpointers";
1301 ret = bch2_check_extents_to_backpointers(c);
1304 bch_verbose(c, "done checking extents to backpointers");
1305 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1307 bch_info(c, "checking alloc to lru refs");
1308 err = "error checking alloc to lru refs";
1309 ret = bch2_check_alloc_to_lru_refs(c);
1312 bch_verbose(c, "done checking alloc to lru refs");
1313 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1315 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1316 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1317 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1318 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1319 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1321 if (c->opts.norecovery)
1324 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1325 err = "error creating root snapshot node";
1326 ret = bch2_fs_initialize_subvolumes(c);
1331 bch_verbose(c, "reading snapshots table");
1332 err = "error reading snapshots table";
1333 ret = bch2_fs_snapshots_start(c);
1336 bch_verbose(c, "reading snapshots done");
1338 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1340 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1341 err = "journal replay failed";
1342 ret = bch2_journal_replay(c);
1345 if (c->opts.verbose || !c->sb.clean)
1346 bch_info(c, "journal replay done");
1349 err = "error initializing freespace";
1350 ret = bch2_fs_freespace_init(c);
1354 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1355 /* set bi_subvol on root inode */
1356 err = "error upgrade root inode for subvolumes";
1357 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1358 bch2_fs_upgrade_for_subvolumes(&trans));
1364 bch_info(c, "starting fsck");
1365 err = "error in fsck";
1366 ret = bch2_fsck_full(c);
1369 bch_verbose(c, "fsck done");
1370 } else if (!c->sb.clean) {
1371 bch_verbose(c, "checking for deleted inodes");
1372 err = "error in recovery";
1373 ret = bch2_fsck_walk_inodes_only(c);
1376 bch_verbose(c, "check inodes done");
1379 if (enabled_qtypes(c)) {
1380 bch_verbose(c, "reading quotas");
1381 ret = bch2_fs_quota_read(c);
1384 bch_verbose(c, "quotas done");
1387 mutex_lock(&c->sb_lock);
1388 if (c->opts.version_upgrade) {
1389 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1390 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1394 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1395 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1400 !test_bit(BCH_FS_ERROR, &c->flags) &&
1401 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1402 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1403 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1408 bch2_write_super(c);
1409 mutex_unlock(&c->sb_lock);
1411 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1412 !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1413 le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1414 struct bch_move_stats stats;
1416 bch2_move_stats_init(&stats, "recovery");
1418 bch_info(c, "scanning for old btree nodes");
1419 ret = bch2_fs_read_write(c);
1423 ret = bch2_scan_old_btree_nodes(c, &stats);
1426 bch_info(c, "scanning for old btree nodes done");
1429 if (c->journal_seq_blacklist_table &&
1430 c->journal_seq_blacklist_table->nr > 128)
1431 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1435 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1436 bch2_flush_fsck_errs(c);
1438 if (!c->opts.keep_journal &&
1439 test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1440 bch2_journal_keys_free(&c->journal_keys);
1441 bch2_journal_entries_free(c);
1445 if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1446 bch2_fs_read_write_early(c);
1447 bch2_delete_dead_snapshots_async(c);
1451 bch_err(c, "Error in recovery: %s (%s)", err, bch2_err_str(ret));
1453 bch_verbose(c, "ret %s", bch2_err_str(ret));
1457 bch2_fs_emergency_read_only(c);
1461 int bch2_fs_initialize(struct bch_fs *c)
1463 struct bch_inode_unpacked root_inode, lostfound_inode;
1464 struct bkey_inode_buf packed_inode;
1465 struct qstr lostfound = QSTR("lost+found");
1466 const char *err = "cannot allocate memory";
1471 bch_notice(c, "initializing new filesystem");
1473 mutex_lock(&c->sb_lock);
1474 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1475 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1477 if (c->sb.version < bcachefs_metadata_version_inode_v3)
1478 c->opts.version_upgrade = true;
1480 if (c->opts.version_upgrade) {
1481 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1482 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1483 bch2_write_super(c);
1485 mutex_unlock(&c->sb_lock);
1487 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1488 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1489 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1490 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1491 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1492 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1494 for (i = 0; i < BTREE_ID_NR; i++)
1495 bch2_btree_root_alloc(c, i);
1497 for_each_online_member(ca, c, i)
1498 bch2_dev_usage_init(ca);
1500 err = "unable to allocate journal buckets";
1501 for_each_online_member(ca, c, i) {
1502 ret = bch2_dev_journal_alloc(ca);
1504 percpu_ref_put(&ca->io_ref);
1510 * journal_res_get() will crash if called before this has
1511 * set up the journal.pin FIFO and journal.cur pointer:
1513 bch2_fs_journal_start(&c->journal, 1);
1514 bch2_journal_set_replay_done(&c->journal);
1516 err = "error going read-write";
1517 ret = bch2_fs_read_write_early(c);
1522 * Write out the superblock and journal buckets, now that we can do
1525 bch_verbose(c, "marking superblocks");
1526 err = "error marking superblock and journal";
1527 for_each_member_device(ca, c, i) {
1528 ret = bch2_trans_mark_dev_sb(c, ca);
1530 percpu_ref_put(&ca->ref);
1534 ca->new_fs_bucket_idx = 0;
1537 bch_verbose(c, "initializing freespace");
1538 err = "error initializing freespace";
1539 ret = bch2_fs_freespace_init(c);
1543 err = "error creating root snapshot node";
1544 ret = bch2_fs_initialize_subvolumes(c);
1548 bch_verbose(c, "reading snapshots table");
1549 err = "error reading snapshots table";
1550 ret = bch2_fs_snapshots_start(c);
1553 bch_verbose(c, "reading snapshots done");
1555 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1556 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1557 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1558 bch2_inode_pack(&packed_inode, &root_inode);
1559 packed_inode.inode.k.p.snapshot = U32_MAX;
1561 err = "error creating root directory";
1562 ret = bch2_btree_insert(c, BTREE_ID_inodes,
1563 &packed_inode.inode.k_i,
1568 bch2_inode_init_early(c, &lostfound_inode);
1570 err = "error creating lost+found";
1571 ret = bch2_trans_do(c, NULL, NULL, 0,
1572 bch2_create_trans(&trans,
1573 BCACHEFS_ROOT_SUBVOL_INUM,
1574 &root_inode, &lostfound_inode,
1576 0, 0, S_IFDIR|0700, 0,
1577 NULL, NULL, (subvol_inum) { 0 }, 0));
1579 bch_err(c, "error creating lost+found");
1583 if (enabled_qtypes(c)) {
1584 ret = bch2_fs_quota_read(c);
1589 err = "error writing first journal entry";
1590 ret = bch2_journal_flush(&c->journal);
1594 mutex_lock(&c->sb_lock);
1595 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1596 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1598 bch2_write_super(c);
1599 mutex_unlock(&c->sb_lock);
1603 pr_err("Error initializing new filesystem: %s (%i)", err, ret);