1 // SPDX-License-Identifier: GPL-2.0
4 #include "backpointers.h"
6 #include "alloc_background.h"
8 #include "btree_journal_iter.h"
9 #include "btree_update.h"
10 #include "btree_update_interior.h"
17 #include "fs-common.h"
19 #include "journal_io.h"
20 #include "journal_reclaim.h"
21 #include "journal_seq_blacklist.h"
23 #include "logged_ops.h"
26 #include "rebalance.h"
31 #include "subvolume.h"
34 #include <linux/sort.h>
35 #include <linux/stat.h>
37 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
39 static bool btree_id_is_alloc(enum btree_id id)
43 case BTREE_ID_backpointers:
44 case BTREE_ID_need_discard:
45 case BTREE_ID_freespace:
46 case BTREE_ID_bucket_gens:
53 /* for -o reconstruct_alloc: */
54 static void drop_alloc_keys(struct journal_keys *keys)
58 for (src = 0, dst = 0; src < keys->nr; src++)
59 if (!btree_id_is_alloc(keys->d[src].btree_id))
60 keys->d[dst++] = keys->d[src];
66 * Btree node pointers have a field to stack a pointer to the in memory btree
67 * node; we need to zero out this field when reading in btree nodes, or when
68 * reading in keys from the journal:
70 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
72 struct journal_key *i;
74 for (i = keys->d; i < keys->d + keys->nr; i++)
75 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
76 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
81 static void replay_now_at(struct journal *j, u64 seq)
83 BUG_ON(seq < j->replay_journal_seq);
85 seq = min(seq, j->replay_journal_seq_end);
87 while (j->replay_journal_seq < seq)
88 bch2_journal_pin_put(j, j->replay_journal_seq++);
91 static int bch2_journal_replay_key(struct btree_trans *trans,
92 struct journal_key *k)
94 struct btree_iter iter;
97 BTREE_ITER_NOT_EXTENTS;
98 unsigned update_flags = BTREE_TRIGGER_NORUN;
102 * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
103 * keep the key cache coherent with the underlying btree. Nothing
104 * besides the allocator is doing updates yet so we don't need key cache
105 * coherency for non-alloc btrees, and key cache fills for snapshots
106 * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
107 * the snapshots recovery pass runs.
109 if (!k->level && k->btree_id == BTREE_ID_alloc)
110 iter_flags |= BTREE_ITER_CACHED;
112 update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
114 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
115 BTREE_MAX_DEPTH, k->level,
117 ret = bch2_btree_iter_traverse(&iter);
121 /* Must be checked with btree locked: */
125 ret = bch2_trans_update(trans, &iter, k->k, update_flags);
127 bch2_trans_iter_exit(trans, &iter);
131 static int journal_sort_seq_cmp(const void *_l, const void *_r)
133 const struct journal_key *l = *((const struct journal_key **)_l);
134 const struct journal_key *r = *((const struct journal_key **)_r);
136 return cmp_int(l->journal_seq, r->journal_seq);
139 static int bch2_journal_replay(struct bch_fs *c)
141 struct journal_keys *keys = &c->journal_keys;
142 struct journal_key **keys_sorted, *k;
143 struct journal *j = &c->journal;
144 u64 start_seq = c->journal_replay_seq_start;
145 u64 end_seq = c->journal_replay_seq_start;
149 move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
150 keys->gap = keys->nr;
152 keys_sorted = kvmalloc_array(keys->nr, sizeof(*keys_sorted), GFP_KERNEL);
154 return -BCH_ERR_ENOMEM_journal_replay;
156 for (i = 0; i < keys->nr; i++)
157 keys_sorted[i] = &keys->d[i];
159 sort(keys_sorted, keys->nr,
160 sizeof(keys_sorted[0]),
161 journal_sort_seq_cmp, NULL);
164 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
165 keys->nr, start_seq, end_seq);
170 for (i = 0; i < keys->nr; i++) {
175 replay_now_at(j, k->journal_seq);
177 ret = bch2_trans_do(c, NULL, NULL,
178 BTREE_INSERT_LAZY_RW|
181 ? BTREE_INSERT_JOURNAL_REPLAY|BCH_WATERMARK_reclaim
183 bch2_journal_replay_key(trans, k));
185 bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
186 bch2_btree_id_str(k->btree_id), k->level, bch2_err_str(ret));
191 replay_now_at(j, j->replay_journal_seq_end);
192 j->replay_journal_seq = 0;
194 bch2_journal_set_replay_done(j);
195 bch2_journal_flush_all_pins(j);
196 ret = bch2_journal_error(j);
198 if (keys->nr && !ret)
199 bch2_journal_log_msg(c, "journal replay finished");
208 /* journal replay early: */
210 static int journal_replay_entry_early(struct bch_fs *c,
211 struct jset_entry *entry)
215 switch (entry->type) {
216 case BCH_JSET_ENTRY_btree_root: {
217 struct btree_root *r;
219 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
220 ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
225 r = bch2_btree_id_root(c, entry->btree_id);
228 r->level = entry->level;
229 bkey_copy(&r->key, (struct bkey_i *) entry->start);
237 case BCH_JSET_ENTRY_usage: {
238 struct jset_entry_usage *u =
239 container_of(entry, struct jset_entry_usage, entry);
241 switch (entry->btree_id) {
242 case BCH_FS_USAGE_reserved:
243 if (entry->level < BCH_REPLICAS_MAX)
244 c->usage_base->persistent_reserved[entry->level] =
247 case BCH_FS_USAGE_inodes:
248 c->usage_base->nr_inodes = le64_to_cpu(u->v);
250 case BCH_FS_USAGE_key_version:
251 atomic64_set(&c->key_version,
258 case BCH_JSET_ENTRY_data_usage: {
259 struct jset_entry_data_usage *u =
260 container_of(entry, struct jset_entry_data_usage, entry);
262 ret = bch2_replicas_set_usage(c, &u->r,
266 case BCH_JSET_ENTRY_dev_usage: {
267 struct jset_entry_dev_usage *u =
268 container_of(entry, struct jset_entry_dev_usage, entry);
269 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
270 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
272 ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec);
274 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
275 ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets);
276 ca->usage_base->d[i].sectors = le64_to_cpu(u->d[i].sectors);
277 ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
282 case BCH_JSET_ENTRY_blacklist: {
283 struct jset_entry_blacklist *bl_entry =
284 container_of(entry, struct jset_entry_blacklist, entry);
286 ret = bch2_journal_seq_blacklist_add(c,
287 le64_to_cpu(bl_entry->seq),
288 le64_to_cpu(bl_entry->seq) + 1);
291 case BCH_JSET_ENTRY_blacklist_v2: {
292 struct jset_entry_blacklist_v2 *bl_entry =
293 container_of(entry, struct jset_entry_blacklist_v2, entry);
295 ret = bch2_journal_seq_blacklist_add(c,
296 le64_to_cpu(bl_entry->start),
297 le64_to_cpu(bl_entry->end) + 1);
300 case BCH_JSET_ENTRY_clock: {
301 struct jset_entry_clock *clock =
302 container_of(entry, struct jset_entry_clock, entry);
304 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
311 static int journal_replay_early(struct bch_fs *c,
312 struct bch_sb_field_clean *clean)
314 struct jset_entry *entry;
318 for (entry = clean->start;
319 entry != vstruct_end(&clean->field);
320 entry = vstruct_next(entry)) {
321 ret = journal_replay_entry_early(c, entry);
326 struct genradix_iter iter;
327 struct journal_replay *i, **_i;
329 genradix_for_each(&c->journal_entries, iter, _i) {
335 vstruct_for_each(&i->j, entry) {
336 ret = journal_replay_entry_early(c, entry);
343 bch2_fs_usage_initialize(c);
348 /* sb clean section: */
350 static int read_btree_roots(struct bch_fs *c)
355 for (i = 0; i < btree_id_nr_alive(c); i++) {
356 struct btree_root *r = bch2_btree_id_root(c, i);
361 if (btree_id_is_alloc(i) &&
362 c->opts.reconstruct_alloc) {
363 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
370 ? FSCK_CAN_IGNORE : 0,
371 btree_root_bkey_invalid,
372 "invalid btree root %s",
373 bch2_btree_id_str(i));
374 if (i == BTREE_ID_alloc)
375 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
378 ret = bch2_btree_root_read(c, i, &r->key, r->level);
381 btree_root_read_error,
382 "error reading btree root %s",
383 bch2_btree_id_str(i));
384 if (btree_id_is_alloc(i))
385 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
390 for (i = 0; i < BTREE_ID_NR; i++) {
391 struct btree_root *r = bch2_btree_id_root(c, i);
396 bch2_btree_root_alloc(c, i);
403 static int bch2_initialize_subvolumes(struct bch_fs *c)
405 struct bkey_i_snapshot_tree root_tree;
406 struct bkey_i_snapshot root_snapshot;
407 struct bkey_i_subvolume root_volume;
410 bkey_snapshot_tree_init(&root_tree.k_i);
411 root_tree.k.p.offset = 1;
412 root_tree.v.master_subvol = cpu_to_le32(1);
413 root_tree.v.root_snapshot = cpu_to_le32(U32_MAX);
415 bkey_snapshot_init(&root_snapshot.k_i);
416 root_snapshot.k.p.offset = U32_MAX;
417 root_snapshot.v.flags = 0;
418 root_snapshot.v.parent = 0;
419 root_snapshot.v.subvol = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
420 root_snapshot.v.tree = cpu_to_le32(1);
421 SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
423 bkey_subvolume_init(&root_volume.k_i);
424 root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
425 root_volume.v.flags = 0;
426 root_volume.v.snapshot = cpu_to_le32(U32_MAX);
427 root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO);
429 ret = bch2_btree_insert(c, BTREE_ID_snapshot_trees, &root_tree.k_i, NULL, 0) ?:
430 bch2_btree_insert(c, BTREE_ID_snapshots, &root_snapshot.k_i, NULL, 0) ?:
431 bch2_btree_insert(c, BTREE_ID_subvolumes, &root_volume.k_i, NULL, 0);
437 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
439 struct btree_iter iter;
441 struct bch_inode_unpacked inode;
444 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
445 SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
450 if (!bkey_is_inode(k.k)) {
451 bch_err(trans->c, "root inode not found");
452 ret = -BCH_ERR_ENOENT_inode;
456 ret = bch2_inode_unpack(k, &inode);
459 inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
461 ret = bch2_inode_write(trans, &iter, &inode);
463 bch2_trans_iter_exit(trans, &iter);
467 /* set bi_subvol on root inode */
469 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
471 int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
472 __bch2_fs_upgrade_for_subvolumes(trans));
478 const char * const bch2_recovery_passes[] = {
479 #define x(_fn, _when) #_fn,
480 BCH_RECOVERY_PASSES()
485 static int bch2_check_allocations(struct bch_fs *c)
487 return bch2_gc(c, true, c->opts.norecovery);
490 static int bch2_set_may_go_rw(struct bch_fs *c)
492 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
496 struct recovery_pass_fn {
497 int (*fn)(struct bch_fs *);
501 static struct recovery_pass_fn recovery_pass_fns[] = {
502 #define x(_fn, _when) { .fn = bch2_##_fn, .when = _when },
503 BCH_RECOVERY_PASSES()
507 static void check_version_upgrade(struct bch_fs *c)
509 unsigned latest_compatible = bch2_latest_compatible_version(c->sb.version);
510 unsigned latest_version = bcachefs_metadata_version_current;
511 unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
512 unsigned new_version = 0;
515 if (old_version < bcachefs_metadata_required_upgrade_below) {
516 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
517 latest_compatible < bcachefs_metadata_required_upgrade_below)
518 new_version = latest_version;
520 new_version = latest_compatible;
522 switch (c->opts.version_upgrade) {
523 case BCH_VERSION_UPGRADE_compatible:
524 new_version = latest_compatible;
526 case BCH_VERSION_UPGRADE_incompatible:
527 new_version = latest_version;
529 case BCH_VERSION_UPGRADE_none:
530 new_version = old_version;
535 if (new_version > old_version) {
536 struct printbuf buf = PRINTBUF;
538 if (old_version < bcachefs_metadata_required_upgrade_below)
539 prt_str(&buf, "Version upgrade required:\n");
541 if (old_version != c->sb.version) {
542 prt_str(&buf, "Version upgrade from ");
543 bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
544 prt_str(&buf, " to ");
545 bch2_version_to_text(&buf, c->sb.version);
546 prt_str(&buf, " incomplete\n");
549 prt_printf(&buf, "Doing %s version upgrade from ",
550 BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
551 ? "incompatible" : "compatible");
552 bch2_version_to_text(&buf, old_version);
553 prt_str(&buf, " to ");
554 bch2_version_to_text(&buf, new_version);
557 recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version);
558 if (recovery_passes) {
559 if ((recovery_passes & RECOVERY_PASS_ALL_FSCK) == RECOVERY_PASS_ALL_FSCK)
560 prt_str(&buf, "fsck required");
562 prt_str(&buf, "running recovery passes: ");
563 prt_bitflags(&buf, bch2_recovery_passes, recovery_passes);
566 c->recovery_passes_explicit |= recovery_passes;
567 c->opts.fix_errors = FSCK_FIX_yes;
570 bch_info(c, "%s", buf.buf);
572 mutex_lock(&c->sb_lock);
573 bch2_sb_upgrade(c, new_version);
574 mutex_unlock(&c->sb_lock);
580 u64 bch2_fsck_recovery_passes(void)
584 for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
585 if (recovery_pass_fns[i].when & PASS_FSCK)
590 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
592 struct recovery_pass_fn *p = recovery_pass_fns + c->curr_recovery_pass;
594 if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
596 if (c->recovery_passes_explicit & BIT_ULL(pass))
598 if ((p->when & PASS_FSCK) && c->opts.fsck)
600 if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
602 if (p->when & PASS_ALWAYS)
607 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
611 c->curr_recovery_pass = pass;
613 if (should_run_recovery_pass(c, pass)) {
614 struct recovery_pass_fn *p = recovery_pass_fns + pass;
616 if (!(p->when & PASS_SILENT))
617 printk(KERN_INFO bch2_log_msg(c, "%s..."),
618 bch2_recovery_passes[pass]);
622 if (!(p->when & PASS_SILENT))
623 printk(KERN_CONT " done\n");
625 c->recovery_passes_complete |= BIT_ULL(pass);
631 static int bch2_run_recovery_passes(struct bch_fs *c)
635 while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
636 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
637 if (bch2_err_matches(ret, BCH_ERR_restart_recovery))
641 c->curr_recovery_pass++;
647 int bch2_fs_recovery(struct bch_fs *c)
649 struct bch_sb_field_clean *clean = NULL;
650 struct jset *last_journal_entry = NULL;
651 u64 last_seq = 0, blacklist_seq, journal_seq;
652 bool write_sb = false;
656 clean = bch2_read_superblock_clean(c);
657 ret = PTR_ERR_OR_ZERO(clean);
661 bch_info(c, "recovering from clean shutdown, journal seq %llu",
662 le64_to_cpu(clean->journal_seq));
664 bch_info(c, "recovering from unclean shutdown");
667 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
668 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
674 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
675 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
680 if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery))
681 check_version_upgrade(c);
683 if (c->opts.fsck && c->opts.norecovery) {
684 bch_err(c, "cannot select both norecovery and fsck");
689 ret = bch2_blacklist_table_initialize(c);
691 bch_err(c, "error initializing blacklist table");
695 if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
696 struct genradix_iter iter;
697 struct journal_replay **i;
699 bch_verbose(c, "starting journal read");
700 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
705 * note: cmd_list_journal needs the blacklist table fully up to date so
706 * it can asterisk ignored journal entries:
708 if (c->opts.read_journal_only)
711 genradix_for_each_reverse(&c->journal_entries, iter, i)
712 if (*i && !(*i)->ignore) {
713 last_journal_entry = &(*i)->j;
717 if (mustfix_fsck_err_on(c->sb.clean &&
718 last_journal_entry &&
719 !journal_entry_empty(last_journal_entry), c,
720 clean_but_journal_not_empty,
721 "filesystem marked clean but journal not empty")) {
722 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
723 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
727 if (!last_journal_entry) {
728 fsck_err_on(!c->sb.clean, c,
729 dirty_but_no_journal_entries,
730 "no journal entries found");
734 genradix_for_each_reverse(&c->journal_entries, iter, i)
736 last_journal_entry = &(*i)->j;
737 (*i)->ignore = false;
739 * This was probably a NO_FLUSH entry,
740 * so last_seq was garbage - but we know
741 * we're only using a single journal
742 * entry, set it here:
744 (*i)->j.last_seq = (*i)->j.seq;
749 ret = bch2_journal_keys_sort(c);
753 if (c->sb.clean && last_journal_entry) {
754 ret = bch2_verify_superblock_clean(c, &clean,
762 bch_err(c, "no superblock clean section found");
763 ret = -BCH_ERR_fsck_repair_impossible;
767 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
770 c->journal_replay_seq_start = last_seq;
771 c->journal_replay_seq_end = blacklist_seq - 1;
773 if (c->opts.reconstruct_alloc) {
774 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
775 drop_alloc_keys(&c->journal_keys);
778 zero_out_btree_mem_ptr(&c->journal_keys);
780 ret = journal_replay_early(c, clean);
785 * After an unclean shutdown, skip then next few journal sequence
786 * numbers as they may have been referenced by btree writes that
787 * happened before their corresponding journal writes - those btree
788 * writes need to be ignored, by skipping and blacklisting the next few
789 * journal sequence numbers:
794 if (blacklist_seq != journal_seq) {
795 ret = bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
796 blacklist_seq, journal_seq) ?:
797 bch2_journal_seq_blacklist_add(c,
798 blacklist_seq, journal_seq);
800 bch_err(c, "error creating new journal seq blacklist entry");
805 ret = bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
806 journal_seq, last_seq, blacklist_seq - 1) ?:
807 bch2_fs_journal_start(&c->journal, journal_seq);
811 if (c->opts.reconstruct_alloc)
812 bch2_journal_log_msg(c, "dropping alloc info");
815 * Skip past versions that might have possibly been used (as nonces),
816 * but hadn't had their pointers written:
818 if (c->sb.encryption_type && !c->sb.clean)
819 atomic64_add(1 << 16, &c->key_version);
821 ret = read_btree_roots(c);
826 (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) ||
827 BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)))
828 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
830 ret = bch2_run_recovery_passes(c);
834 /* If we fixed errors, verify that fs is actually clean now: */
835 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
836 test_bit(BCH_FS_ERRORS_FIXED, &c->flags) &&
837 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags) &&
838 !test_bit(BCH_FS_ERROR, &c->flags)) {
839 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
840 clear_bit(BCH_FS_ERRORS_FIXED, &c->flags);
842 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
844 ret = bch2_run_recovery_passes(c);
848 if (test_bit(BCH_FS_ERRORS_FIXED, &c->flags) ||
849 test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
850 bch_err(c, "Second fsck run was not clean");
851 set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
854 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
857 if (enabled_qtypes(c)) {
858 bch_verbose(c, "reading quotas");
859 ret = bch2_fs_quota_read(c);
862 bch_verbose(c, "quotas done");
865 mutex_lock(&c->sb_lock);
866 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
867 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
871 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
872 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
877 !test_bit(BCH_FS_ERROR, &c->flags) &&
878 !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
879 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
880 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
886 mutex_unlock(&c->sb_lock);
888 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
889 c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
890 struct bch_move_stats stats;
892 bch2_move_stats_init(&stats, "recovery");
894 bch_info(c, "scanning for old btree nodes");
895 ret = bch2_fs_read_write(c) ?:
896 bch2_scan_old_btree_nodes(c, &stats);
899 bch_info(c, "scanning for old btree nodes done");
902 if (c->journal_seq_blacklist_table &&
903 c->journal_seq_blacklist_table->nr > 128)
904 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
908 set_bit(BCH_FS_FSCK_DONE, &c->flags);
909 bch2_flush_fsck_errs(c);
911 if (!c->opts.keep_journal &&
912 test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
913 bch2_journal_keys_free(&c->journal_keys);
914 bch2_journal_entries_free(c);
918 if (!ret && test_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags)) {
919 bch2_fs_read_write_early(c);
920 bch2_delete_dead_snapshots_async(c);
928 bch2_fs_emergency_read_only(c);
932 int bch2_fs_initialize(struct bch_fs *c)
934 struct bch_inode_unpacked root_inode, lostfound_inode;
935 struct bkey_inode_buf packed_inode;
936 struct qstr lostfound = QSTR("lost+found");
941 bch_notice(c, "initializing new filesystem");
943 mutex_lock(&c->sb_lock);
944 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
945 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
947 bch2_sb_maybe_downgrade(c);
949 if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
950 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
951 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
954 mutex_unlock(&c->sb_lock);
956 c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
957 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
958 set_bit(BCH_FS_FSCK_DONE, &c->flags);
960 for (i = 0; i < BTREE_ID_NR; i++)
961 bch2_btree_root_alloc(c, i);
963 for_each_member_device(ca, c, i)
964 bch2_dev_usage_init(ca);
966 ret = bch2_fs_journal_alloc(c);
971 * journal_res_get() will crash if called before this has
972 * set up the journal.pin FIFO and journal.cur pointer:
974 bch2_fs_journal_start(&c->journal, 1);
975 bch2_journal_set_replay_done(&c->journal);
977 ret = bch2_fs_read_write_early(c);
982 * Write out the superblock and journal buckets, now that we can do
985 bch_verbose(c, "marking superblocks");
986 ret = bch2_trans_mark_dev_sbs(c);
987 bch_err_msg(c, ret, "marking superblocks");
991 for_each_online_member(ca, c, i)
992 ca->new_fs_bucket_idx = 0;
994 ret = bch2_fs_freespace_init(c);
998 ret = bch2_initialize_subvolumes(c);
1002 bch_verbose(c, "reading snapshots table");
1003 ret = bch2_snapshots_read(c);
1006 bch_verbose(c, "reading snapshots done");
1008 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1009 root_inode.bi_inum = BCACHEFS_ROOT_INO;
1010 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1011 bch2_inode_pack(&packed_inode, &root_inode);
1012 packed_inode.inode.k.p.snapshot = U32_MAX;
1014 ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
1016 bch_err_msg(c, ret, "creating root directory");
1020 bch2_inode_init_early(c, &lostfound_inode);
1022 ret = bch2_trans_do(c, NULL, NULL, 0,
1023 bch2_create_trans(trans,
1024 BCACHEFS_ROOT_SUBVOL_INUM,
1025 &root_inode, &lostfound_inode,
1027 0, 0, S_IFDIR|0700, 0,
1028 NULL, NULL, (subvol_inum) { 0 }, 0));
1030 bch_err_msg(c, ret, "creating lost+found");
1034 if (enabled_qtypes(c)) {
1035 ret = bch2_fs_quota_read(c);
1040 ret = bch2_journal_flush(&c->journal);
1042 bch_err_msg(c, ret, "writing first journal entry");
1046 mutex_lock(&c->sb_lock);
1047 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1048 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1050 bch2_write_super(c);
1051 mutex_unlock(&c->sb_lock);
1055 bch_err_fn(ca, ret);