1 // SPDX-License-Identifier: GPL-2.0
4 #include "alloc_background.h"
6 #include "btree_update.h"
7 #include "btree_update_interior.h"
13 #include "fs-common.h"
15 #include "journal_io.h"
16 #include "journal_reclaim.h"
17 #include "journal_seq_blacklist.h"
23 #include <linux/sort.h>
24 #include <linux/stat.h>
26 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
28 /* iterate over keys read from the journal: */
30 struct journal_iter bch2_journal_iter_init(struct journal_keys *keys,
33 return (struct journal_iter) {
40 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
43 if (iter->k == iter->keys->d + iter->keys->nr)
46 if (iter->k->btree_id == iter->btree_id)
47 return bkey_i_to_s_c(iter->k->k);
55 struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
57 if (iter->k == iter->keys->d + iter->keys->nr)
61 return bch2_journal_iter_peek(iter);
64 /* sort and dedup all keys in the journal: */
66 static void journal_entries_free(struct list_head *list)
69 while (!list_empty(list)) {
70 struct journal_replay *i =
71 list_first_entry(list, struct journal_replay, list);
73 kvpfree(i, offsetof(struct journal_replay, j) +
74 vstruct_bytes(&i->j));
78 static int journal_sort_key_cmp(const void *_l, const void *_r)
80 const struct journal_key *l = _l;
81 const struct journal_key *r = _r;
83 return cmp_int(l->btree_id, r->btree_id) ?:
84 bkey_cmp(l->pos, r->pos) ?:
85 cmp_int(l->journal_seq, r->journal_seq) ?:
86 cmp_int(l->journal_offset, r->journal_offset);
89 static int journal_sort_seq_cmp(const void *_l, const void *_r)
91 const struct journal_key *l = _l;
92 const struct journal_key *r = _r;
94 return cmp_int(l->journal_seq, r->journal_seq) ?:
95 cmp_int(l->btree_id, r->btree_id) ?:
96 bkey_cmp(l->pos, r->pos);
99 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
101 while (i + 1 < keys->d + keys->nr &&
102 journal_sort_key_cmp(i, i + 1) > 0) {
108 static void journal_keys_free(struct journal_keys *keys)
110 struct journal_key *i;
112 for_each_journal_key(*keys, i)
120 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
122 struct journal_replay *p;
123 struct jset_entry *entry;
124 struct bkey_i *k, *_n;
125 struct journal_keys keys = { NULL }, keys_deduped = { NULL };
126 struct journal_key *i;
129 list_for_each_entry(p, journal_entries, list)
130 for_each_jset_key(k, _n, entry, &p->j)
133 keys.journal_seq_base = keys_deduped.journal_seq_base =
134 le64_to_cpu(list_first_entry(journal_entries,
135 struct journal_replay,
138 keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
142 keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
146 list_for_each_entry(p, journal_entries, list)
147 for_each_jset_key(k, _n, entry, &p->j)
148 keys.d[keys.nr++] = (struct journal_key) {
149 .btree_id = entry->btree_id,
150 .pos = bkey_start_pos(&k->k),
152 .journal_seq = le64_to_cpu(p->j.seq) -
153 keys.journal_seq_base,
154 .journal_offset = k->_data - p->j._data,
157 sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
160 while (i < keys.d + keys.nr) {
161 if (i + 1 < keys.d + keys.nr &&
162 i[0].btree_id == i[1].btree_id &&
163 !bkey_cmp(i[0].pos, i[1].pos)) {
164 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
167 bch2_cut_front(i[1].k->k.p, i[0].k);
168 i[0].pos = i[1].k->k.p;
169 journal_keys_sift(&keys, i);
174 if (i + 1 < keys.d + keys.nr &&
175 i[0].btree_id == i[1].btree_id &&
176 bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
177 if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
178 cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
179 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
180 bch2_cut_back(bkey_start_pos(&i[1].k->k), i[0].k);
182 struct bkey_i *split =
183 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
188 bkey_copy(split, i[0].k);
189 bch2_cut_back(bkey_start_pos(&i[1].k->k), split);
190 keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
191 .btree_id = i[0].btree_id,
193 .pos = bkey_start_pos(&split->k),
195 .journal_seq = i[0].journal_seq,
196 .journal_offset = i[0].journal_offset,
199 bch2_cut_front(i[1].k->k.p, i[0].k);
200 i[0].pos = i[1].k->k.p;
201 journal_keys_sift(&keys, i);
205 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
210 bch2_cut_front(i[0].k->k.p, i[1].k);
211 i[1].pos = i[0].k->k.p;
212 journal_keys_sift(&keys, i + 1);
218 keys_deduped.d[keys_deduped.nr++] = *i++;
224 journal_keys_free(&keys_deduped);
226 return (struct journal_keys) { NULL };
229 /* journal replay: */
231 static void replay_now_at(struct journal *j, u64 seq)
233 BUG_ON(seq < j->replay_journal_seq);
234 BUG_ON(seq > j->replay_journal_seq_end);
236 while (j->replay_journal_seq < seq)
237 bch2_journal_pin_put(j, j->replay_journal_seq++);
240 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
243 struct btree_trans trans;
244 struct btree_iter *iter, *split_iter;
246 * We might cause compressed extents to be split, so we need to pass in
247 * a disk_reservation:
249 struct disk_reservation disk_res =
250 bch2_disk_reservation_init(c, 0);
251 struct bkey_i *split;
252 struct bpos atomic_end;
254 * Some extents aren't equivalent - w.r.t. what the triggers do
255 * - if they're split:
257 bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
258 k->k.type == KEY_TYPE_reflink_p;
262 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
264 bch2_trans_begin(&trans);
266 iter = bch2_trans_get_iter(&trans, btree_id,
267 bkey_start_pos(&k->k),
271 ret = bch2_btree_iter_traverse(iter);
275 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
277 split_iter = bch2_trans_copy_iter(&trans, iter);
278 ret = PTR_ERR_OR_ZERO(split_iter);
282 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
283 ret = PTR_ERR_OR_ZERO(split);
289 bkey_cmp(atomic_end, k->k.p) < 0) {
290 ret = bch2_disk_reservation_add(c, &disk_res,
292 bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
293 BCH_DISK_RESERVATION_NOFAIL);
300 bch2_cut_front(split_iter->pos, split);
301 bch2_cut_back(atomic_end, split);
303 bch2_trans_update(&trans, split_iter, split);
304 bch2_btree_iter_set_pos(iter, split->k.p);
305 } while (bkey_cmp(iter->pos, k->k.p) < 0);
308 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
309 0, -((s64) k->k.size),
310 BCH_BUCKET_MARK_OVERWRITE) ?:
311 bch2_trans_commit(&trans, &disk_res, NULL,
313 BTREE_INSERT_LAZY_RW|
314 BTREE_INSERT_NOMARK_OVERWRITES);
316 ret = bch2_trans_commit(&trans, &disk_res, NULL,
318 BTREE_INSERT_LAZY_RW|
319 BTREE_INSERT_JOURNAL_REPLAY|
320 BTREE_INSERT_NOMARK);
329 bch2_disk_reservation_put(c, &disk_res);
331 return bch2_trans_exit(&trans) ?: ret;
334 static int bch2_journal_replay(struct bch_fs *c,
335 struct journal_keys keys)
337 struct journal *j = &c->journal;
338 struct journal_key *i;
341 sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
343 for_each_journal_key(keys, i) {
344 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
346 if (i->btree_id == BTREE_ID_ALLOC)
347 ret = bch2_alloc_replay_key(c, i->k);
348 else if (btree_node_type_is_extents(i->btree_id))
349 ret = bch2_extent_replay_key(c, i->btree_id, i->k);
351 ret = bch2_btree_insert(c, i->btree_id, i->k,
354 BTREE_INSERT_LAZY_RW|
355 BTREE_INSERT_JOURNAL_REPLAY|
356 BTREE_INSERT_NOMARK);
359 bch_err(c, "journal replay: error %d while replaying key",
367 replay_now_at(j, j->replay_journal_seq_end);
368 j->replay_journal_seq = 0;
370 bch2_journal_set_replay_done(j);
371 bch2_journal_flush_all_pins(j);
372 return bch2_journal_error(j);
375 static bool journal_empty(struct list_head *journal)
377 return list_empty(journal) ||
378 journal_entry_empty(&list_last_entry(journal,
379 struct journal_replay, list)->j);
383 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
384 struct list_head *journal)
386 struct journal_replay *i =
387 list_last_entry(journal, struct journal_replay, list);
388 u64 start_seq = le64_to_cpu(i->j.last_seq);
389 u64 end_seq = le64_to_cpu(i->j.seq);
393 list_for_each_entry(i, journal, list) {
394 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
395 "journal entries %llu-%llu missing! (replaying %llu-%llu)",
396 seq, le64_to_cpu(i->j.seq) - 1,
399 seq = le64_to_cpu(i->j.seq);
401 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
402 "found blacklisted journal entry %llu", seq);
406 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
412 /* journal replay early: */
414 static int journal_replay_entry_early(struct bch_fs *c,
415 struct jset_entry *entry)
419 switch (entry->type) {
420 case BCH_JSET_ENTRY_btree_root: {
421 struct btree_root *r;
423 if (entry->btree_id >= BTREE_ID_NR) {
424 bch_err(c, "filesystem has unknown btree type %u",
429 r = &c->btree_roots[entry->btree_id];
432 r->level = entry->level;
433 bkey_copy(&r->key, &entry->start[0]);
441 case BCH_JSET_ENTRY_usage: {
442 struct jset_entry_usage *u =
443 container_of(entry, struct jset_entry_usage, entry);
445 switch (entry->btree_id) {
446 case FS_USAGE_RESERVED:
447 if (entry->level < BCH_REPLICAS_MAX)
448 c->usage_base->persistent_reserved[entry->level] =
451 case FS_USAGE_INODES:
452 c->usage_base->nr_inodes = le64_to_cpu(u->v);
454 case FS_USAGE_KEY_VERSION:
455 atomic64_set(&c->key_version,
462 case BCH_JSET_ENTRY_data_usage: {
463 struct jset_entry_data_usage *u =
464 container_of(entry, struct jset_entry_data_usage, entry);
465 ret = bch2_replicas_set_usage(c, &u->r,
469 case BCH_JSET_ENTRY_blacklist: {
470 struct jset_entry_blacklist *bl_entry =
471 container_of(entry, struct jset_entry_blacklist, entry);
473 ret = bch2_journal_seq_blacklist_add(c,
474 le64_to_cpu(bl_entry->seq),
475 le64_to_cpu(bl_entry->seq) + 1);
478 case BCH_JSET_ENTRY_blacklist_v2: {
479 struct jset_entry_blacklist_v2 *bl_entry =
480 container_of(entry, struct jset_entry_blacklist_v2, entry);
482 ret = bch2_journal_seq_blacklist_add(c,
483 le64_to_cpu(bl_entry->start),
484 le64_to_cpu(bl_entry->end) + 1);
492 static int journal_replay_early(struct bch_fs *c,
493 struct bch_sb_field_clean *clean,
494 struct list_head *journal)
496 struct jset_entry *entry;
500 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
501 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
503 for (entry = clean->start;
504 entry != vstruct_end(&clean->field);
505 entry = vstruct_next(entry)) {
506 ret = journal_replay_entry_early(c, entry);
511 struct journal_replay *i =
512 list_last_entry(journal, struct journal_replay, list);
514 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
515 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
517 list_for_each_entry(i, journal, list)
518 vstruct_for_each(&i->j, entry) {
519 ret = journal_replay_entry_early(c, entry);
525 bch2_fs_usage_initialize(c);
530 /* sb clean section: */
532 static struct bkey_i *btree_root_find(struct bch_fs *c,
533 struct bch_sb_field_clean *clean,
535 enum btree_id id, unsigned *level)
538 struct jset_entry *entry, *start, *end;
541 start = clean->start;
542 end = vstruct_end(&clean->field);
545 end = vstruct_last(j);
548 for (entry = start; entry < end; entry = vstruct_next(entry))
549 if (entry->type == BCH_JSET_ENTRY_btree_root &&
550 entry->btree_id == id)
556 return ERR_PTR(-EINVAL);
559 *level = entry->level;
563 static int verify_superblock_clean(struct bch_fs *c,
564 struct bch_sb_field_clean **cleanp,
568 struct bch_sb_field_clean *clean = *cleanp;
571 if (!c->sb.clean || !j)
574 if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
575 "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
576 le64_to_cpu(clean->journal_seq),
577 le64_to_cpu(j->seq))) {
583 mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
584 "superblock read clock doesn't match journal after clean shutdown");
585 mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
586 "superblock read clock doesn't match journal after clean shutdown");
588 for (i = 0; i < BTREE_ID_NR; i++) {
589 struct bkey_i *k1, *k2;
590 unsigned l1 = 0, l2 = 0;
592 k1 = btree_root_find(c, clean, NULL, i, &l1);
593 k2 = btree_root_find(c, NULL, j, i, &l2);
598 mustfix_fsck_err_on(!k1 || !k2 ||
601 k1->k.u64s != k2->k.u64s ||
602 memcmp(k1, k2, bkey_bytes(k1)) ||
604 "superblock btree root doesn't match journal after clean shutdown");
610 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
612 struct bch_sb_field_clean *clean, *sb_clean;
615 mutex_lock(&c->sb_lock);
616 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
618 if (fsck_err_on(!sb_clean, c,
619 "superblock marked clean but clean section not present")) {
620 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
622 mutex_unlock(&c->sb_lock);
626 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
629 mutex_unlock(&c->sb_lock);
630 return ERR_PTR(-ENOMEM);
633 if (le16_to_cpu(c->disk_sb.sb->version) <
634 bcachefs_metadata_version_bkey_renumber)
635 bch2_sb_clean_renumber(clean, READ);
637 mutex_unlock(&c->sb_lock);
641 mutex_unlock(&c->sb_lock);
645 static int read_btree_roots(struct bch_fs *c)
650 for (i = 0; i < BTREE_ID_NR; i++) {
651 struct btree_root *r = &c->btree_roots[i];
656 if (i == BTREE_ID_ALLOC &&
657 c->opts.reconstruct_alloc) {
658 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
664 __fsck_err(c, i == BTREE_ID_ALLOC
665 ? FSCK_CAN_IGNORE : 0,
666 "invalid btree root %s",
668 if (i == BTREE_ID_ALLOC)
669 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
672 ret = bch2_btree_root_read(c, i, &r->key, r->level);
674 __fsck_err(c, i == BTREE_ID_ALLOC
675 ? FSCK_CAN_IGNORE : 0,
676 "error reading btree root %s",
678 if (i == BTREE_ID_ALLOC)
679 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
683 for (i = 0; i < BTREE_ID_NR; i++)
684 if (!c->btree_roots[i].b)
685 bch2_btree_root_alloc(c, i);
690 int bch2_fs_recovery(struct bch_fs *c)
692 const char *err = "cannot allocate memory";
693 struct bch_sb_field_clean *clean = NULL;
695 LIST_HEAD(journal_entries);
696 struct journal_keys journal_keys = { NULL };
697 bool wrote = false, write_sb = false;
701 clean = read_superblock_clean(c);
702 ret = PTR_ERR_OR_ZERO(clean);
707 bch_info(c, "recovering from clean shutdown, journal seq %llu",
708 le64_to_cpu(clean->journal_seq));
710 if (!c->replicas.entries) {
711 bch_info(c, "building replicas info");
712 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
715 if (!c->sb.clean || c->opts.fsck) {
718 ret = bch2_journal_read(c, &journal_entries);
722 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
723 "filesystem marked clean but journal not empty")) {
724 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
725 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
729 if (!c->sb.clean && list_empty(&journal_entries)) {
730 bch_err(c, "no journal entries found");
731 ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
735 journal_keys = journal_keys_sort(&journal_entries);
736 if (!journal_keys.d) {
741 j = &list_last_entry(&journal_entries,
742 struct journal_replay, list)->j;
744 ret = verify_superblock_clean(c, &clean, j);
748 journal_seq = le64_to_cpu(j->seq) + 1;
750 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
753 ret = journal_replay_early(c, clean, &journal_entries);
758 ret = bch2_journal_seq_blacklist_add(c,
762 bch_err(c, "error creating new journal seq blacklist entry");
769 ret = bch2_blacklist_table_initialize(c);
771 if (!list_empty(&journal_entries)) {
772 ret = verify_journal_entries_not_blacklisted_or_missing(c,
778 ret = bch2_fs_journal_start(&c->journal, journal_seq,
783 ret = read_btree_roots(c);
787 bch_verbose(c, "starting alloc read");
788 err = "error reading allocation information";
789 ret = bch2_alloc_read(c, &journal_keys);
792 bch_verbose(c, "alloc read done");
794 bch_verbose(c, "starting stripes_read");
795 err = "error reading stripes";
796 ret = bch2_stripes_read(c, &journal_keys);
799 bch_verbose(c, "stripes_read done");
801 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
803 if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
804 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
806 * interior btree node updates aren't consistent with the
807 * journal; after an unclean shutdown we have to walk all
808 * pointers to metadata:
810 bch_info(c, "starting metadata mark and sweep");
811 err = "error in mark and sweep";
812 ret = bch2_gc(c, NULL, true, true);
815 bch_verbose(c, "mark and sweep done");
819 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
820 test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
821 bch_info(c, "starting mark and sweep");
822 err = "error in mark and sweep";
823 ret = bch2_gc(c, &journal_keys, true, false);
826 bch_verbose(c, "mark and sweep done");
829 clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
830 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
833 * Skip past versions that might have possibly been used (as nonces),
834 * but hadn't had their pointers written:
836 if (c->sb.encryption_type && !c->sb.clean)
837 atomic64_add(1 << 16, &c->key_version);
839 if (c->opts.norecovery)
842 bch_verbose(c, "starting journal replay");
843 err = "journal replay failed";
844 ret = bch2_journal_replay(c, journal_keys);
847 bch_verbose(c, "journal replay done");
849 if (!c->opts.nochanges) {
851 * note that even when filesystem was clean there might be work
852 * to do here, if we ran gc (because of fsck) which recalculated
855 bch_verbose(c, "writing allocation info");
856 err = "error writing out alloc info";
857 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
858 bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
860 bch_err(c, "error writing alloc info");
863 bch_verbose(c, "alloc write done");
865 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
869 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
870 bch_info(c, "checking inode link counts");
871 err = "error in recovery";
872 ret = bch2_fsck_inode_nlink(c);
875 bch_verbose(c, "check inodes done");
878 bch_verbose(c, "checking for deleted inodes");
879 err = "error in recovery";
880 ret = bch2_fsck_walk_inodes_only(c);
883 bch_verbose(c, "check inodes done");
888 bch_info(c, "starting fsck");
889 err = "error in fsck";
890 ret = bch2_fsck_full(c);
893 bch_verbose(c, "fsck done");
896 if (enabled_qtypes(c)) {
897 bch_verbose(c, "reading quotas");
898 ret = bch2_fs_quota_read(c);
901 bch_verbose(c, "quotas done");
904 mutex_lock(&c->sb_lock);
905 if (c->opts.version_upgrade) {
906 if (c->sb.version < bcachefs_metadata_version_new_versioning)
907 c->disk_sb.sb->version_min =
908 le16_to_cpu(bcachefs_metadata_version_min);
909 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
910 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
914 if (!test_bit(BCH_FS_ERROR, &c->flags)) {
915 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
920 !test_bit(BCH_FS_ERROR, &c->flags)) {
921 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
922 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
928 mutex_unlock(&c->sb_lock);
930 if (c->journal_seq_blacklist_table &&
931 c->journal_seq_blacklist_table->nr > 128)
932 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
937 set_bit(BCH_FS_FSCK_DONE, &c->flags);
938 bch2_flush_fsck_errs(c);
940 journal_keys_free(&journal_keys);
941 journal_entries_free(&journal_entries);
944 bch_err(c, "Error in recovery: %s (%i)", err, ret);
946 bch_verbose(c, "ret %i", ret);
950 int bch2_fs_initialize(struct bch_fs *c)
952 struct bch_inode_unpacked root_inode, lostfound_inode;
953 struct bkey_inode_buf packed_inode;
954 struct qstr lostfound = QSTR("lost+found");
955 const char *err = "cannot allocate memory";
961 bch_notice(c, "initializing new filesystem");
963 mutex_lock(&c->sb_lock);
964 for_each_online_member(ca, c, i)
965 bch2_mark_dev_superblock(c, ca, 0);
966 mutex_unlock(&c->sb_lock);
968 set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
969 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
971 for (i = 0; i < BTREE_ID_NR; i++)
972 bch2_btree_root_alloc(c, i);
974 err = "unable to allocate journal buckets";
975 for_each_online_member(ca, c, i) {
976 ret = bch2_dev_journal_alloc(ca);
978 percpu_ref_put(&ca->io_ref);
984 * journal_res_get() will crash if called before this has
985 * set up the journal.pin FIFO and journal.cur pointer:
987 bch2_fs_journal_start(&c->journal, 1, &journal);
988 bch2_journal_set_replay_done(&c->journal);
990 bch2_inode_init(c, &root_inode, 0, 0,
991 S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
992 root_inode.bi_inum = BCACHEFS_ROOT_INO;
993 bch2_inode_pack(&packed_inode, &root_inode);
995 err = "error creating root directory";
996 ret = bch2_btree_insert(c, BTREE_ID_INODES,
997 &packed_inode.inode.k_i,
998 NULL, NULL, BTREE_INSERT_LAZY_RW);
1002 bch2_inode_init_early(c, &lostfound_inode);
1004 err = "error creating lost+found";
1005 ret = bch2_trans_do(c, NULL, NULL, 0,
1006 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1007 &root_inode, &lostfound_inode,
1009 0, 0, S_IFDIR|0700, 0,
1014 if (enabled_qtypes(c)) {
1015 ret = bch2_fs_quota_read(c);
1020 err = "error writing first journal entry";
1021 ret = bch2_journal_meta(&c->journal);
1025 mutex_lock(&c->sb_lock);
1026 c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1027 le16_to_cpu(bcachefs_metadata_version_current);
1028 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1029 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
1031 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1032 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1034 bch2_write_super(c);
1035 mutex_unlock(&c->sb_lock);
1039 pr_err("Error initializing new filesystem: %s (%i)", err, ret);