X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Frecovery.c;h=dcd4f9f410ae5f72ad9fcd396116a5681a99c0e3;hb=44fc32e7ef89d794854409d98d3d1c3fe8fb3580;hp=5da6b3b403ab5e8ae01055907ca8d1d7f3c040c4;hpb=c50379128bb027d43c76ac27d7cade7260db7f66;p=bcachefs-tools-debian diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c index 5da6b3b..dcd4f9f 100644 --- a/libbcachefs/recovery.c +++ b/libbcachefs/recovery.c @@ -1,6 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 #include "bcachefs.h" +#include "backpointers.h" #include "bkey_buf.h" #include "alloc_background.h" #include "btree_gc.h" @@ -10,12 +11,14 @@ #include "buckets.h" #include "dirent.h" #include "ec.h" +#include "errcode.h" #include "error.h" #include "fs-common.h" #include "fsck.h" #include "journal_io.h" #include "journal_reclaim.h" #include "journal_seq_blacklist.h" +#include "lru.h" #include "move.h" #include "quota.h" #include "recovery.h" @@ -71,40 +74,118 @@ static int journal_key_cmp(const struct journal_key *l, const struct journal_key return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r); } -size_t bch2_journal_key_search(struct journal_keys *journal_keys, - enum btree_id id, unsigned level, - struct bpos pos) +static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx) { - size_t l = 0, r = journal_keys->nr, m; + size_t gap_size = keys->size - keys->nr; + + if (idx >= keys->gap) + idx += gap_size; + return idx; +} + +static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx) +{ + return keys->d + idx_to_pos(keys, idx); +} + +static size_t __bch2_journal_key_search(struct journal_keys *keys, + enum btree_id id, unsigned level, + struct bpos pos) +{ + size_t l = 0, r = keys->nr, m; while (l < r) { m = l + ((r - l) >> 1); - if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0) + if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0) l = m + 1; else r = m; } - BUG_ON(l < journal_keys->nr && - __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0); + BUG_ON(l < keys->nr && + __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0); BUG_ON(l && - __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0); + __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0); return l; } -static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx) +static size_t bch2_journal_key_search(struct journal_keys *keys, + enum btree_id id, unsigned level, + struct bpos pos) { - struct bkey_i *n = iter->keys->d[idx].k; - struct btree_and_journal_iter *biter = - container_of(iter, struct btree_and_journal_iter, journal); - - if (iter->idx > idx || - (iter->idx == idx && - biter->last && - bpos_cmp(n->k.p, biter->unpacked.p) <= 0)) - iter->idx++; + return idx_to_pos(keys, __bch2_journal_key_search(keys, id, level, pos)); +} + +struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id, + unsigned level, struct bpos pos, + struct bpos end_pos, size_t *idx) +{ + struct journal_keys *keys = &c->journal_keys; + unsigned iters = 0; + struct journal_key *k; +search: + if (!*idx) + *idx = __bch2_journal_key_search(keys, btree_id, level, pos); + + while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) { + if (__journal_key_cmp(btree_id, level, end_pos, k) < 0) + return NULL; + + if (__journal_key_cmp(btree_id, level, pos, k) <= 0 && + !k->overwritten) + return k->k; + + (*idx)++; + iters++; + if (iters == 10) { + *idx = 0; + goto search; + } + } + + return NULL; +} + +struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id, + unsigned level, struct bpos pos) +{ + size_t idx = 0; + + return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx); +} + +static void journal_iters_fix(struct bch_fs *c) +{ + struct journal_keys *keys = &c->journal_keys; + /* The key we just inserted is immediately before the gap: */ + size_t gap_end = keys->gap + (keys->size - keys->nr); + struct btree_and_journal_iter *iter; + + /* + * If an iterator points one after the key we just inserted, decrement + * the iterator so it points at the key we just inserted - if the + * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will + * handle that: + */ + list_for_each_entry(iter, &c->journal_iters, journal.list) + if (iter->journal.idx == gap_end) + iter->journal.idx = keys->gap - 1; +} + +static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap) +{ + struct journal_keys *keys = &c->journal_keys; + struct journal_iter *iter; + size_t gap_size = keys->size - keys->nr; + + list_for_each_entry(iter, &c->journal_iters, list) { + if (iter->idx > old_gap) + iter->idx -= gap_size; + if (iter->idx >= new_gap) + iter->idx += gap_size; + } } int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id, @@ -122,12 +203,11 @@ int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id, .journal_seq = U32_MAX, }; struct journal_keys *keys = &c->journal_keys; - struct journal_iter *iter; size_t idx = bch2_journal_key_search(keys, id, level, k->k.p); BUG_ON(test_bit(BCH_FS_RW, &c->flags)); - if (idx < keys->nr && + if (idx < keys->size && journal_key_cmp(&n, &keys->d[idx]) == 0) { if (keys->d[idx].allocated) kfree(keys->d[idx].k); @@ -135,29 +215,40 @@ int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id, return 0; } + if (idx > keys->gap) + idx -= keys->size - keys->nr; + if (keys->nr == keys->size) { struct journal_keys new_keys = { .nr = keys->nr, - .size = keys->size * 2, - .journal_seq_base = keys->journal_seq_base, + .size = max_t(size_t, keys->size, 8) * 2, }; - new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL); + new_keys.d = kvmalloc_array(new_keys.size, sizeof(new_keys.d[0]), GFP_KERNEL); if (!new_keys.d) { bch_err(c, "%s: error allocating new key array (size %zu)", __func__, new_keys.size); - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_key_insert; } + /* Since @keys was full, there was no gap: */ memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr); kvfree(keys->d); *keys = new_keys; + + /* And now the gap is at the end: */ + keys->gap = keys->nr; } - array_insert_item(keys->d, keys->nr, idx, n); + journal_iters_move_gap(c, keys->gap, idx); - list_for_each_entry(iter, &c->journal_iters, list) - journal_iter_fix(c, iter, idx); + move_gap(keys->d, keys->nr, keys->size, keys->gap, idx); + keys->gap = idx; + + keys->nr++; + keys->d[keys->gap++] = n; + + journal_iters_fix(c); return 0; } @@ -175,7 +266,7 @@ int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id, n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL); if (!n) - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_key_insert; bkey_copy(n, k); ret = bch2_journal_key_insert_take(c, id, level, n); @@ -201,34 +292,37 @@ void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree, struct journal_keys *keys = &c->journal_keys; size_t idx = bch2_journal_key_search(keys, btree, level, pos); - if (idx < keys->nr && + if (idx < keys->size && keys->d[idx].btree_id == btree && keys->d[idx].level == level && - !bpos_cmp(keys->d[idx].k->k.p, pos)) + bpos_eq(keys->d[idx].k->k.p, pos)) keys->d[idx].overwritten = true; } -static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter) +static void bch2_journal_iter_advance(struct journal_iter *iter) +{ + if (iter->idx < iter->keys->size) { + iter->idx++; + if (iter->idx == iter->keys->gap) + iter->idx += iter->keys->size - iter->keys->nr; + } +} + +static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter) { struct journal_key *k = iter->keys->d + iter->idx; - while (k < iter->keys->d + iter->keys->nr && + while (k < iter->keys->d + iter->keys->size && k->btree_id == iter->btree_id && k->level == iter->level) { if (!k->overwritten) - return k->k; + return bkey_i_to_s_c(k->k); - iter->idx++; + bch2_journal_iter_advance(iter); k = iter->keys->d + iter->idx; } - return NULL; -} - -static void bch2_journal_iter_advance(struct journal_iter *iter) -{ - if (iter->idx < iter->keys->nr) - iter->idx++; + return bkey_s_c_null; } static void bch2_journal_iter_exit(struct journal_iter *iter) @@ -260,71 +354,49 @@ static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter) void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter) { - switch (iter->last) { - case none: - break; - case btree: - bch2_journal_iter_advance_btree(iter); - break; - case journal: - bch2_journal_iter_advance(&iter->journal); - break; - } - - iter->last = none; + if (bpos_eq(iter->pos, SPOS_MAX)) + iter->at_end = true; + else + iter->pos = bpos_successor(iter->pos); } struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter) { - struct bkey_s_c ret; + struct bkey_s_c btree_k, journal_k, ret; +again: + if (iter->at_end) + return bkey_s_c_null; - while (1) { - struct bkey_s_c btree_k = - bch2_journal_iter_peek_btree(iter); - struct bkey_s_c journal_k = - bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal)); - - if (btree_k.k && journal_k.k) { - int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p); + while ((btree_k = bch2_journal_iter_peek_btree(iter)).k && + bpos_lt(btree_k.k->p, iter->pos)) + bch2_journal_iter_advance_btree(iter); - if (!cmp) - bch2_journal_iter_advance_btree(iter); + while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k && + bpos_lt(journal_k.k->p, iter->pos)) + bch2_journal_iter_advance(&iter->journal); - iter->last = cmp < 0 ? btree : journal; - } else if (btree_k.k) { - iter->last = btree; - } else if (journal_k.k) { - iter->last = journal; - } else { - iter->last = none; - return bkey_s_c_null; - } + ret = journal_k.k && + (!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p)) + ? journal_k + : btree_k; - ret = iter->last == journal ? journal_k : btree_k; + if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key)) + ret = bkey_s_c_null; - if (iter->b && - bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) { - iter->journal.idx = iter->journal.keys->nr; - iter->last = none; - return bkey_s_c_null; + if (ret.k) { + iter->pos = ret.k->p; + if (bkey_deleted(ret.k)) { + bch2_btree_and_journal_iter_advance(iter); + goto again; } - - if (!bkey_deleted(ret.k)) - break; - - bch2_btree_and_journal_iter_advance(iter); + } else { + iter->pos = SPOS_MAX; + iter->at_end = true; } return ret; } -struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter) -{ - bch2_btree_and_journal_iter_advance(iter); - - return bch2_btree_and_journal_iter_peek(iter); -} - void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter) { bch2_journal_iter_exit(&iter->journal); @@ -342,6 +414,8 @@ void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter iter->node_iter = node_iter; bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos); INIT_LIST_HEAD(&iter->journal.list); + iter->pos = b->data->min_key; + iter->at_end = false; } /* @@ -361,16 +435,16 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *i /* sort and dedup all keys in the journal: */ -void bch2_journal_entries_free(struct list_head *list) +void bch2_journal_entries_free(struct bch_fs *c) { - - while (!list_empty(list)) { - struct journal_replay *i = - list_first_entry(list, struct journal_replay, list); - list_del(&i->list); - kvpfree(i, offsetof(struct journal_replay, j) + - vstruct_bytes(&i->j)); - } + struct journal_replay **i; + struct genradix_iter iter; + + genradix_for_each(&c->journal_entries, iter, i) + if (*i) + kvpfree(*i, offsetof(struct journal_replay, j) + + vstruct_bytes(&(*i)->j)); + genradix_free(&c->journal_entries); } /* @@ -390,77 +464,115 @@ void bch2_journal_keys_free(struct journal_keys *keys) { struct journal_key *i; + move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr); + keys->gap = keys->nr; + for (i = keys->d; i < keys->d + keys->nr; i++) if (i->allocated) kfree(i->k); kvfree(keys->d); keys->d = NULL; - keys->nr = 0; + keys->nr = keys->gap = keys->size = 0; } -static struct journal_keys journal_keys_sort(struct list_head *journal_entries) +static void __journal_keys_sort(struct journal_keys *keys) { - struct journal_replay *i; - struct jset_entry *entry; - struct bkey_i *k, *_n; - struct journal_keys keys = { NULL }; struct journal_key *src, *dst; - size_t nr_keys = 0; - if (list_empty(journal_entries)) - return keys; + sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL); - list_for_each_entry(i, journal_entries, list) { - if (i->ignore) - continue; + src = dst = keys->d; + while (src < keys->d + keys->nr) { + while (src + 1 < keys->d + keys->nr && + src[0].btree_id == src[1].btree_id && + src[0].level == src[1].level && + bpos_eq(src[0].k->k.p, src[1].k->k.p)) + src++; + + *dst++ = *src++; + } - if (!keys.journal_seq_base) - keys.journal_seq_base = le64_to_cpu(i->j.seq); + keys->nr = dst - keys->d; +} + +static int journal_keys_sort(struct bch_fs *c) +{ + struct genradix_iter iter; + struct journal_replay *i, **_i; + struct jset_entry *entry; + struct bkey_i *k; + struct journal_keys *keys = &c->journal_keys; + size_t nr_keys = 0, nr_read = 0; + + genradix_for_each(&c->journal_entries, iter, _i) { + i = *_i; - for_each_jset_key(k, _n, entry, &i->j) + if (!i || i->ignore) + continue; + + for_each_jset_key(k, entry, &i->j) nr_keys++; } - keys.size = roundup_pow_of_two(nr_keys); + if (!nr_keys) + return 0; - keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL); - if (!keys.d) - goto err; + keys->size = roundup_pow_of_two(nr_keys); + + keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL); + if (!keys->d) { + bch_err(c, "Failed to allocate buffer for sorted journal keys (%zu keys); trying slowpath", + nr_keys); + + do { + keys->size >>= 1; + keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL); + } while (!keys->d && keys->size > nr_keys / 8); + + if (!keys->d) { + bch_err(c, "Failed to allocate %zu size buffer for sorted journal keys; exiting", + keys->size); + return -BCH_ERR_ENOMEM_journal_keys_sort; + } + } + + genradix_for_each(&c->journal_entries, iter, _i) { + i = *_i; - list_for_each_entry(i, journal_entries, list) { - if (i->ignore) + if (!i || i->ignore) continue; - BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX); + cond_resched(); - for_each_jset_key(k, _n, entry, &i->j) - keys.d[keys.nr++] = (struct journal_key) { + for_each_jset_key(k, entry, &i->j) { + if (keys->nr == keys->size) { + __journal_keys_sort(keys); + + if (keys->nr > keys->size * 7 / 8) { + bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu/%zu", + keys->nr, keys->size, nr_read, nr_keys); + return -BCH_ERR_ENOMEM_journal_keys_sort; + } + } + + keys->d[keys->nr++] = (struct journal_key) { .btree_id = entry->btree_id, .level = entry->level, .k = k, - .journal_seq = le64_to_cpu(i->j.seq) - - keys.journal_seq_base, + .journal_seq = le64_to_cpu(i->j.seq), .journal_offset = k->_data - i->j._data, }; - } - - sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL); - src = dst = keys.d; - while (src < keys.d + keys.nr) { - while (src + 1 < keys.d + keys.nr && - src[0].btree_id == src[1].btree_id && - src[0].level == src[1].level && - !bpos_cmp(src[0].k->k.p, src[1].k->k.p)) - src++; - - *dst++ = *src++; + nr_read++; + } } - keys.nr = dst - keys.d; -err: - return keys; + __journal_keys_sort(keys); + keys->gap = keys->nr; + + bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_keys, keys->nr); + return 0; } /* journal replay: */ @@ -468,7 +580,8 @@ err: static void replay_now_at(struct journal *j, u64 seq) { BUG_ON(seq < j->replay_journal_seq); - BUG_ON(seq > j->replay_journal_seq_end); + + seq = min(seq, j->replay_journal_seq_end); while (j->replay_journal_seq < seq) bch2_journal_pin_put(j, j->replay_journal_seq++); @@ -481,10 +594,21 @@ static int bch2_journal_replay_key(struct btree_trans *trans, unsigned iter_flags = BTREE_ITER_INTENT| BTREE_ITER_NOT_EXTENTS; + unsigned update_flags = BTREE_TRIGGER_NORUN; int ret; + /* + * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to + * keep the key cache coherent with the underlying btree. Nothing + * besides the allocator is doing updates yet so we don't need key cache + * coherency for non-alloc btrees, and key cache fills for snapshots + * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until + * the snapshots recovery pass runs. + */ if (!k->level && k->btree_id == BTREE_ID_alloc) iter_flags |= BTREE_ITER_CACHED; + else + update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM; bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p, BTREE_MAX_DEPTH, k->level, @@ -497,7 +621,7 @@ static int bch2_journal_replay_key(struct btree_trans *trans, if (k->overwritten) goto out; - ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN); + ret = bch2_trans_update(trans, &iter, k->k, update_flags); out: bch2_trans_iter_exit(trans, &iter); return ret; @@ -516,12 +640,17 @@ static int bch2_journal_replay(struct bch_fs *c) struct journal_keys *keys = &c->journal_keys; struct journal_key **keys_sorted, *k; struct journal *j = &c->journal; + u64 start_seq = c->journal_replay_seq_start; + u64 end_seq = c->journal_replay_seq_start; size_t i; int ret; - keys_sorted = kmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL); + move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr); + keys->gap = keys->nr; + + keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL); if (!keys_sorted) - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_replay; for (i = 0; i < keys->nr; i++) keys_sorted[i] = &keys->d[i]; @@ -531,8 +660,10 @@ static int bch2_journal_replay(struct bch_fs *c) journal_sort_seq_cmp, NULL); if (keys->nr) { - bch_verbose(c, "starting journal replay, %zu keys", keys->nr); - replay_now_at(j, keys->journal_seq_base); + ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)", + keys->nr, start_seq, end_seq); + if (ret) + goto err; } for (i = 0; i < keys->nr; i++) { @@ -540,18 +671,18 @@ static int bch2_journal_replay(struct bch_fs *c) cond_resched(); - if (!k->allocated) - replay_now_at(j, keys->journal_seq_base + k->journal_seq); + replay_now_at(j, k->journal_seq); ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW| BTREE_INSERT_NOFAIL| - BTREE_INSERT_JOURNAL_RESERVED| - (!k->allocated ? BTREE_INSERT_JOURNAL_REPLAY : 0), + (!k->allocated + ? BTREE_INSERT_JOURNAL_REPLAY|BCH_WATERMARK_reclaim + : 0), bch2_journal_replay_key(&trans, k)); if (ret) { - bch_err(c, "journal replay: error %d while replaying key at btree %s level %u", - ret, bch2_btree_ids[k->btree_id], k->level); + bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s", + bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret)); goto err; } } @@ -562,8 +693,14 @@ static int bch2_journal_replay(struct bch_fs *c) bch2_journal_set_replay_done(j); bch2_journal_flush_all_pins(j); ret = bch2_journal_error(j); + + if (keys->nr && !ret) + bch2_journal_log_msg(c, "journal replay finished"); err: - kfree(keys_sorted); + kvfree(keys_sorted); + + if (ret) + bch_err_fn(c, ret); return ret; } @@ -578,13 +715,13 @@ static int journal_replay_entry_early(struct bch_fs *c, case BCH_JSET_ENTRY_btree_root: { struct btree_root *r; - if (entry->btree_id >= BTREE_ID_NR) { - bch_err(c, "filesystem has unknown btree type %u", - entry->btree_id); - return -EINVAL; + while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) { + ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL }); + if (ret) + return ret; } - r = &c->btree_roots[entry->btree_id]; + r = bch2_btree_id_root(c, entry->btree_id); if (entry->u64s) { r->level = entry->level; @@ -632,7 +769,6 @@ static int journal_replay_entry_early(struct bch_fs *c, unsigned i, nr_types = jset_entry_dev_usage_nr_types(u); ca->usage_base->buckets_ec = le64_to_cpu(u->buckets_ec); - ca->usage_base->buckets_unavailable = le64_to_cpu(u->buckets_unavailable); for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) { ca->usage_base->d[i].buckets = le64_to_cpu(u->d[i].buckets); @@ -672,10 +808,8 @@ static int journal_replay_entry_early(struct bch_fs *c, } static int journal_replay_early(struct bch_fs *c, - struct bch_sb_field_clean *clean, - struct list_head *journal) + struct bch_sb_field_clean *clean) { - struct journal_replay *i; struct jset_entry *entry; int ret; @@ -688,8 +822,13 @@ static int journal_replay_early(struct bch_fs *c, return ret; } } else { - list_for_each_entry(i, journal, list) { - if (i->ignore) + struct genradix_iter iter; + struct journal_replay *i, **_i; + + genradix_for_each(&c->journal_entries, iter, _i) { + i = *_i; + + if (!i || i->ignore) continue; vstruct_for_each(&i->j, entry) { @@ -744,6 +883,8 @@ static int verify_superblock_clean(struct bch_fs *c, { unsigned i; struct bch_sb_field_clean *clean = *cleanp; + struct printbuf buf1 = PRINTBUF; + struct printbuf buf2 = PRINTBUF; int ret = 0; if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c, @@ -756,7 +897,6 @@ static int verify_superblock_clean(struct bch_fs *c, } for (i = 0; i < BTREE_ID_NR; i++) { - char buf1[200], buf2[200]; struct bkey_i *k1, *k2; unsigned l1 = 0, l2 = 0; @@ -766,19 +906,34 @@ static int verify_superblock_clean(struct bch_fs *c, if (!k1 && !k2) continue; + printbuf_reset(&buf1); + printbuf_reset(&buf2); + + if (k1) + bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1)); + else + prt_printf(&buf1, "(none)"); + + if (k2) + bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2)); + else + prt_printf(&buf2, "(none)"); + mustfix_fsck_err_on(!k1 || !k2 || IS_ERR(k1) || IS_ERR(k2) || k1->k.u64s != k2->k.u64s || - memcmp(k1, k2, bkey_bytes(k1)) || + memcmp(k1, k2, bkey_bytes(&k1->k)) || l1 != l2, c, "superblock btree root %u doesn't match journal after clean shutdown\n" "sb: l=%u %s\n" "journal: l=%u %s\n", i, - l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1), - l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2)); + l1, buf1.buf, + l2, buf2.buf); } fsck_err: + printbuf_exit(&buf2); + printbuf_exit(&buf1); return ret; } @@ -802,10 +957,10 @@ static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c) GFP_KERNEL); if (!clean) { mutex_unlock(&c->sb_lock); - return ERR_PTR(-ENOMEM); + return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean); } - ret = bch2_sb_clean_validate(c, clean, READ); + ret = bch2_sb_clean_validate_late(c, clean, READ); if (ret) { mutex_unlock(&c->sb_lock); return ERR_PTR(ret); @@ -819,25 +974,39 @@ fsck_err: return ERR_PTR(ret); } +static bool btree_id_is_alloc(enum btree_id id) +{ + switch (id) { + case BTREE_ID_alloc: + case BTREE_ID_backpointers: + case BTREE_ID_need_discard: + case BTREE_ID_freespace: + case BTREE_ID_bucket_gens: + return true; + default: + return false; + } +} + static int read_btree_roots(struct bch_fs *c) { unsigned i; int ret = 0; - for (i = 0; i < BTREE_ID_NR; i++) { - struct btree_root *r = &c->btree_roots[i]; + for (i = 0; i < btree_id_nr_alive(c); i++) { + struct btree_root *r = bch2_btree_id_root(c, i); if (!r->alive) continue; - if (i == BTREE_ID_alloc && + if (btree_id_is_alloc(i) && c->opts.reconstruct_alloc) { c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info); continue; } if (r->error) { - __fsck_err(c, i == BTREE_ID_alloc + __fsck_err(c, btree_id_is_alloc(i) ? FSCK_CAN_IGNORE : 0, "invalid btree root %s", bch2_btree_ids[i]); @@ -847,76 +1016,85 @@ static int read_btree_roots(struct bch_fs *c) ret = bch2_btree_root_read(c, i, &r->key, r->level); if (ret) { - __fsck_err(c, i == BTREE_ID_alloc + __fsck_err(c, + btree_id_is_alloc(i) ? FSCK_CAN_IGNORE : 0, "error reading btree root %s", bch2_btree_ids[i]); - if (i == BTREE_ID_alloc) + if (btree_id_is_alloc(i)) c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info); } } - for (i = 0; i < BTREE_ID_NR; i++) - if (!c->btree_roots[i].b) + for (i = 0; i < BTREE_ID_NR; i++) { + struct btree_root *r = bch2_btree_id_root(c, i); + + if (!r->b) { + r->alive = false; + r->level = 0; bch2_btree_root_alloc(c, i); + } + } fsck_err: return ret; } -static int bch2_fs_initialize_subvolumes(struct bch_fs *c) +static int bch2_initialize_subvolumes(struct bch_fs *c) { - struct bkey_i_snapshot root_snapshot; - struct bkey_i_subvolume root_volume; + struct bkey_i_snapshot_tree root_tree; + struct bkey_i_snapshot root_snapshot; + struct bkey_i_subvolume root_volume; int ret; + bkey_snapshot_tree_init(&root_tree.k_i); + root_tree.k.p.offset = 1; + root_tree.v.master_subvol = cpu_to_le32(1); + root_tree.v.root_snapshot = cpu_to_le32(U32_MAX); + bkey_snapshot_init(&root_snapshot.k_i); root_snapshot.k.p.offset = U32_MAX; root_snapshot.v.flags = 0; root_snapshot.v.parent = 0; - root_snapshot.v.subvol = BCACHEFS_ROOT_SUBVOL; - root_snapshot.v.pad = 0; + root_snapshot.v.subvol = cpu_to_le32(BCACHEFS_ROOT_SUBVOL); + root_snapshot.v.tree = cpu_to_le32(1); SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true); - ret = bch2_btree_insert(c, BTREE_ID_snapshots, - &root_snapshot.k_i, - NULL, NULL, 0); - if (ret) - return ret; - - bkey_subvolume_init(&root_volume.k_i); root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL; root_volume.v.flags = 0; root_volume.v.snapshot = cpu_to_le32(U32_MAX); root_volume.v.inode = cpu_to_le64(BCACHEFS_ROOT_INO); - ret = bch2_btree_insert(c, BTREE_ID_subvolumes, - &root_volume.k_i, - NULL, NULL, 0); + ret = bch2_btree_insert(c, BTREE_ID_snapshot_trees, + &root_tree.k_i, + NULL, NULL, 0) ?: + bch2_btree_insert(c, BTREE_ID_snapshots, + &root_snapshot.k_i, + NULL, NULL, 0) ?: + bch2_btree_insert(c, BTREE_ID_subvolumes, + &root_volume.k_i, + NULL, NULL, 0); if (ret) - return ret; - - return 0; + bch_err_fn(c, ret); + return ret; } -static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans) +static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans) { - struct bch_fs *c = trans->c; struct btree_iter iter; struct bkey_s_c k; struct bch_inode_unpacked inode; int ret; - bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, - SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0); - k = bch2_btree_iter_peek_slot(&iter); + k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, + SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0); ret = bkey_err(k); if (ret) - goto err; + return ret; if (!bkey_is_inode(k.k)) { - bch_err(c, "root inode not found"); - ret = -ENOENT; + bch_err(trans->c, "root inode not found"); + ret = -BCH_ERR_ENOENT_inode; goto err; } @@ -931,12 +1109,177 @@ err: return ret; } +/* set bi_subvol on root inode */ +noinline_for_stack +static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c) +{ + int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW, + __bch2_fs_upgrade_for_subvolumes(&trans)); + if (ret) + bch_err_fn(c, ret); + return ret; +} + +static void check_version_upgrade(struct bch_fs *c) +{ + unsigned latest_compatible = bch2_version_compatible(c->sb.version); + unsigned latest_version = bcachefs_metadata_version_current; + unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version; + unsigned new_version = 0; + u64 recovery_passes; + + if (old_version < bcachefs_metadata_required_upgrade_below) { + if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible || + latest_compatible < bcachefs_metadata_required_upgrade_below) + new_version = latest_version; + else + new_version = latest_compatible; + } else { + switch (c->opts.version_upgrade) { + case BCH_VERSION_UPGRADE_compatible: + new_version = latest_compatible; + break; + case BCH_VERSION_UPGRADE_incompatible: + new_version = latest_version; + break; + case BCH_VERSION_UPGRADE_none: + new_version = old_version; + break; + } + } + + if (new_version > old_version) { + struct printbuf buf = PRINTBUF; + + if (old_version < bcachefs_metadata_required_upgrade_below) + prt_str(&buf, "Version upgrade required:\n"); + + if (old_version != c->sb.version) { + prt_str(&buf, "Version upgrade from "); + bch2_version_to_text(&buf, c->sb.version_upgrade_complete); + prt_str(&buf, " to "); + bch2_version_to_text(&buf, c->sb.version); + prt_str(&buf, " incomplete\n"); + } + + prt_printf(&buf, "Doing %s version upgrade from ", + BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version) + ? "incompatible" : "compatible"); + bch2_version_to_text(&buf, old_version); + prt_str(&buf, " to "); + bch2_version_to_text(&buf, new_version); + prt_newline(&buf); + + recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version); + if (recovery_passes) { + prt_str(&buf, "fsck required"); + + c->recovery_passes_explicit |= recovery_passes; + c->opts.fix_errors = FSCK_FIX_yes; + } + + bch_info(c, "%s", buf.buf); + + mutex_lock(&c->sb_lock); + bch2_sb_upgrade(c, new_version); + mutex_unlock(&c->sb_lock); + + printbuf_exit(&buf); + } +} + +static int bch2_check_allocations(struct bch_fs *c) +{ + return bch2_gc(c, true, c->opts.norecovery); +} + +static int bch2_set_may_go_rw(struct bch_fs *c) +{ + set_bit(BCH_FS_MAY_GO_RW, &c->flags); + return 0; +} + +struct recovery_pass_fn { + int (*fn)(struct bch_fs *); + const char *name; + unsigned when; +}; + +static struct recovery_pass_fn recovery_passes[] = { +#define x(_fn, _when) { .fn = bch2_##_fn, .name = #_fn, .when = _when }, + BCH_RECOVERY_PASSES() +#undef x +}; + +u64 bch2_fsck_recovery_passes(void) +{ + u64 ret = 0; + + for (unsigned i = 0; i < ARRAY_SIZE(recovery_passes); i++) + if (recovery_passes[i].when & PASS_FSCK) + ret |= BIT_ULL(i); + return ret; +} + +static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass) +{ + struct recovery_pass_fn *p = recovery_passes + c->curr_recovery_pass; + + if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read) + return false; + if (c->recovery_passes_explicit & BIT_ULL(pass)) + return true; + if ((p->when & PASS_FSCK) && c->opts.fsck) + return true; + if ((p->when & PASS_UNCLEAN) && !c->sb.clean) + return true; + if (p->when & PASS_ALWAYS) + return true; + return false; +} + +static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass) +{ + int ret; + + c->curr_recovery_pass = pass; + + if (should_run_recovery_pass(c, pass)) { + struct recovery_pass_fn *p = recovery_passes + pass; + + if (!(p->when & PASS_SILENT)) + printk(KERN_INFO bch2_log_msg(c, "%s..."), p->name); + ret = p->fn(c); + if (ret) + return ret; + if (!(p->when & PASS_SILENT)) + printk(KERN_CONT " done\n"); + } + + return 0; +} + +static int bch2_run_recovery_passes(struct bch_fs *c) +{ + int ret = 0; + + while (c->curr_recovery_pass < ARRAY_SIZE(recovery_passes)) { + ret = bch2_run_recovery_pass(c, c->curr_recovery_pass); + if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) + continue; + if (ret) + break; + c->curr_recovery_pass++; + } + + return ret; +} + int bch2_fs_recovery(struct bch_fs *c) { - const char *err = "cannot allocate memory"; struct bch_sb_field_clean *clean = NULL; struct jset *last_journal_entry = NULL; - u64 blacklist_seq, journal_seq; + u64 last_seq, blacklist_seq, journal_seq; bool write_sb = false; int ret = 0; @@ -971,32 +1314,13 @@ int bch2_fs_recovery(struct bch_fs *c) goto err; } - if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) { - bch_info(c, "alloc_v2 feature bit not set, fsck required"); - c->opts.fsck = true; - c->opts.fix_errors = FSCK_OPT_YES; - } + if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery)) + check_version_upgrade(c); - if (!c->replicas.entries || - c->opts.rebuild_replicas) { - bch_info(c, "building replicas info"); - set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags); - } - - if (!c->opts.nochanges) { - if (c->sb.version < bcachefs_metadata_version_inode_backpointers) { - bch_info(c, "version prior to inode backpointers, upgrade and fsck required"); - c->opts.version_upgrade = true; - c->opts.fsck = true; - c->opts.fix_errors = FSCK_OPT_YES; - } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) { - bch_info(c, "filesystem version is prior to subvol_dirent - upgrading"); - c->opts.version_upgrade = true; - c->opts.fsck = true; - } else if (c->sb.version < bcachefs_metadata_version_inode_v2) { - bch_info(c, "filesystem version is prior to inode_v2 - upgrading"); - c->opts.version_upgrade = true; - } + if (c->opts.fsck && c->opts.norecovery) { + bch_err(c, "cannot select both norecovery and fsck"); + ret = -EINVAL; + goto err; } ret = bch2_blacklist_table_initialize(c); @@ -1006,16 +1330,24 @@ int bch2_fs_recovery(struct bch_fs *c) } if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) { - struct journal_replay *i; + struct genradix_iter iter; + struct journal_replay **i; - ret = bch2_journal_read(c, &c->journal_entries, - &blacklist_seq, &journal_seq); + bch_verbose(c, "starting journal read"); + ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq); if (ret) goto err; - list_for_each_entry_reverse(i, &c->journal_entries, list) - if (!i->ignore) { - last_journal_entry = &i->j; + /* + * note: cmd_list_journal needs the blacklist table fully up to date so + * it can asterisk ignored journal entries: + */ + if (c->opts.read_journal_only) + goto out; + + genradix_for_each_reverse(&c->journal_entries, iter, i) + if (*i && !(*i)->ignore) { + last_journal_entry = &(*i)->j; break; } @@ -1030,14 +1362,20 @@ int bch2_fs_recovery(struct bch_fs *c) if (!last_journal_entry) { fsck_err_on(!c->sb.clean, c, "no journal entries found"); - goto use_clean; + if (clean) + goto use_clean; + + genradix_for_each_reverse(&c->journal_entries, iter, i) + if (*i) { + last_journal_entry = &(*i)->j; + (*i)->ignore = false; + break; + } } - c->journal_keys = journal_keys_sort(&c->journal_entries); - if (!c->journal_keys.d) { - ret = -ENOMEM; + ret = journal_keys_sort(c); + if (ret) goto err; - } if (c->sb.clean && last_journal_entry) { ret = verify_superblock_clean(c, &clean, @@ -1049,13 +1387,16 @@ int bch2_fs_recovery(struct bch_fs *c) use_clean: if (!clean) { bch_err(c, "no superblock clean section found"); - ret = BCH_FSCK_REPAIR_IMPOSSIBLE; + ret = -BCH_ERR_fsck_repair_impossible; goto err; } blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1; } + c->journal_replay_seq_start = last_seq; + c->journal_replay_seq_end = blacklist_seq - 1;; + if (c->opts.reconstruct_alloc) { c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info); drop_alloc_keys(&c->journal_keys); @@ -1063,12 +1404,24 @@ use_clean: zero_out_btree_mem_ptr(&c->journal_keys); - ret = journal_replay_early(c, clean, &c->journal_entries); + ret = journal_replay_early(c, clean); if (ret) goto err; + /* + * After an unclean shutdown, skip then next few journal sequence + * numbers as they may have been referenced by btree writes that + * happened before their corresponding journal writes - those btree + * writes need to be ignored, by skipping and blacklisting the next few + * journal sequence numbers: + */ + if (!c->sb.clean) + journal_seq += 8; + if (blacklist_seq != journal_seq) { - ret = bch2_journal_seq_blacklist_add(c, + ret = bch2_journal_log_msg(c, "blacklisting entries %llu-%llu", + blacklist_seq, journal_seq) ?: + bch2_journal_seq_blacklist_add(c, blacklist_seq, journal_seq); if (ret) { bch_err(c, "error creating new journal seq blacklist entry"); @@ -1076,60 +1429,14 @@ use_clean: } } - ret = bch2_fs_journal_start(&c->journal, journal_seq, - &c->journal_entries); - if (ret) - goto err; - - ret = read_btree_roots(c); - if (ret) - goto err; - - bch_verbose(c, "starting alloc read"); - err = "error reading allocation information"; - - down_read(&c->gc_lock); - ret = bch2_alloc_read(c, false, false); - up_read(&c->gc_lock); - - if (ret) - goto err; - bch_verbose(c, "alloc read done"); - - bch_verbose(c, "starting stripes_read"); - err = "error reading stripes"; - ret = bch2_stripes_read(c); + ret = bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu", + journal_seq, last_seq, blacklist_seq - 1) ?: + bch2_fs_journal_start(&c->journal, journal_seq); if (ret) goto err; - bch_verbose(c, "stripes_read done"); - set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags); - - /* - * If we're not running fsck, this ensures bch2_fsck_err() calls are - * instead interpreted as bch2_inconsistent_err() calls: - */ - if (!c->opts.fsck) - set_bit(BCH_FS_FSCK_DONE, &c->flags); - - if (c->opts.fsck || - !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) || - !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) || - test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) { - bool metadata_only = c->opts.norecovery; - - bch_info(c, "starting mark and sweep"); - err = "error in mark and sweep"; - ret = bch2_gc(c, true, metadata_only); - if (ret) - goto err; - bch_verbose(c, "mark and sweep done"); - } - - bch2_stripes_heap_start(c); - - clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags); - set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags); + if (c->opts.reconstruct_alloc) + bch2_journal_log_msg(c, "dropping alloc info"); /* * Skip past versions that might have possibly been used (as nonces), @@ -1138,57 +1445,18 @@ use_clean: if (c->sb.encryption_type && !c->sb.clean) atomic64_add(1 << 16, &c->key_version); - if (c->opts.norecovery) - goto out; - - bch_verbose(c, "starting journal replay"); - err = "journal replay failed"; - ret = bch2_journal_replay(c); + ret = read_btree_roots(c); if (ret) goto err; - if (c->opts.verbose || !c->sb.clean) - bch_info(c, "journal replay done"); - - if (c->sb.version < bcachefs_metadata_version_snapshot_2) { - bch2_fs_lazy_rw(c); - err = "error creating root snapshot node"; - ret = bch2_fs_initialize_subvolumes(c); - if (ret) - goto err; - } + if (c->opts.fsck && + (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) || + BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb))) + c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology); - bch_verbose(c, "reading snapshots table"); - err = "error reading snapshots table"; - ret = bch2_fs_snapshots_start(c); + ret = bch2_run_recovery_passes(c); if (ret) goto err; - bch_verbose(c, "reading snapshots done"); - - if (c->sb.version < bcachefs_metadata_version_snapshot_2) { - /* set bi_subvol on root inode */ - err = "error upgrade root inode for subvolumes"; - ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW, - bch2_fs_upgrade_for_subvolumes(&trans)); - if (ret) - goto err; - } - - if (c->opts.fsck) { - bch_info(c, "starting fsck"); - err = "error in fsck"; - ret = bch2_fsck_full(c); - if (ret) - goto err; - bch_verbose(c, "fsck done"); - } else if (!c->sb.clean) { - bch_verbose(c, "checking for deleted inodes"); - err = "error in recovery"; - ret = bch2_fsck_walk_inodes_only(c); - if (ret) - goto err; - bch_verbose(c, "check inodes done"); - } if (enabled_qtypes(c)) { bch_verbose(c, "reading quotas"); @@ -1199,17 +1467,8 @@ use_clean: } mutex_lock(&c->sb_lock); - /* - * With journal replay done, we can clear the journal seq blacklist - * table: - */ - BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)); - if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written) - bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0); - - if (c->opts.version_upgrade) { - c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current); - c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL); + if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) { + SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version); write_sb = true; } @@ -1232,36 +1491,42 @@ use_clean: if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) || !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) || - le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) { + c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) { struct bch_move_stats stats; - bch_move_stats_init(&stats, "recovery"); + bch2_move_stats_init(&stats, "recovery"); bch_info(c, "scanning for old btree nodes"); - ret = bch2_fs_read_write(c); - if (ret) - goto err; - - ret = bch2_scan_old_btree_nodes(c, &stats); + ret = bch2_fs_read_write(c) ?: + bch2_scan_old_btree_nodes(c, &stats); if (ret) goto err; bch_info(c, "scanning for old btree nodes done"); } + if (c->journal_seq_blacklist_table && + c->journal_seq_blacklist_table->nr > 128) + queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work); + ret = 0; out: set_bit(BCH_FS_FSCK_DONE, &c->flags); bch2_flush_fsck_errs(c); - if (!c->opts.keep_journal) { + if (!c->opts.keep_journal && + test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) { bch2_journal_keys_free(&c->journal_keys); - bch2_journal_entries_free(&c->journal_entries); + bch2_journal_entries_free(c); } kfree(clean); + + if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) { + bch2_fs_read_write_early(c); + bch2_delete_dead_snapshots_async(c); + } + if (ret) - bch_err(c, "Error in recovery: %s (%i)", err, ret); - else - bch_verbose(c, "ret %i", ret); + bch_err_fn(c, ret); return ret; err: fsck_err: @@ -1274,9 +1539,7 @@ int bch2_fs_initialize(struct bch_fs *c) struct bch_inode_unpacked root_inode, lostfound_inode; struct bkey_inode_buf packed_inode; struct qstr lostfound = QSTR("lost+found"); - const char *err = "cannot allocate memory"; struct bch_dev *ca; - LIST_HEAD(journal); unsigned i; int ret; @@ -1286,21 +1549,25 @@ int bch2_fs_initialize(struct bch_fs *c) c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done); c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done); - if (c->opts.version_upgrade) { - c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current); - c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL); + bch2_sb_maybe_downgrade(c); + + if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) { + bch2_sb_upgrade(c, bcachefs_metadata_version_current); + SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current); bch2_write_super(c); } mutex_unlock(&c->sb_lock); - set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags); - set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags); + c->curr_recovery_pass = ARRAY_SIZE(recovery_passes); + set_bit(BCH_FS_MAY_GO_RW, &c->flags); set_bit(BCH_FS_FSCK_DONE, &c->flags); for (i = 0; i < BTREE_ID_NR; i++) bch2_btree_root_alloc(c, i); - err = "unable to allocate journal buckets"; + for_each_online_member(ca, c, i) + bch2_dev_usage_init(ca); + for_each_online_member(ca, c, i) { ret = bch2_dev_journal_alloc(ca); if (ret) { @@ -1313,10 +1580,9 @@ int bch2_fs_initialize(struct bch_fs *c) * journal_res_get() will crash if called before this has * set up the journal.pin FIFO and journal.cur pointer: */ - bch2_fs_journal_start(&c->journal, 1, &journal); + bch2_fs_journal_start(&c->journal, 1); bch2_journal_set_replay_done(&c->journal); - err = "error going read-write"; ret = bch2_fs_read_write_early(c); if (ret) goto err; @@ -1325,7 +1591,7 @@ int bch2_fs_initialize(struct bch_fs *c) * Write out the superblock and journal buckets, now that we can do * btree updates */ - err = "error marking superblock and journal"; + bch_verbose(c, "marking superblocks"); for_each_member_device(ca, c, i) { ret = bch2_trans_mark_dev_sb(c, ca); if (ret) { @@ -1336,35 +1602,36 @@ int bch2_fs_initialize(struct bch_fs *c) ca->new_fs_bucket_idx = 0; } - err = "error creating root snapshot node"; - ret = bch2_fs_initialize_subvolumes(c); + ret = bch2_fs_freespace_init(c); + if (ret) + goto err; + + ret = bch2_initialize_subvolumes(c); if (ret) goto err; bch_verbose(c, "reading snapshots table"); - err = "error reading snapshots table"; - ret = bch2_fs_snapshots_start(c); + ret = bch2_snapshots_read(c); if (ret) goto err; bch_verbose(c, "reading snapshots done"); - bch2_inode_init(c, &root_inode, 0, 0, - S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL); + bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL); root_inode.bi_inum = BCACHEFS_ROOT_INO; root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL; - bch2_inode_pack(c, &packed_inode, &root_inode); + bch2_inode_pack(&packed_inode, &root_inode); packed_inode.inode.k.p.snapshot = U32_MAX; - err = "error creating root directory"; ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, NULL, 0); - if (ret) + if (ret) { + bch_err_msg(c, ret, "creating root directory"); goto err; + } bch2_inode_init_early(c, &lostfound_inode); - err = "error creating lost+found"; ret = bch2_trans_do(c, NULL, NULL, 0, bch2_create_trans(&trans, BCACHEFS_ROOT_SUBVOL_INUM, @@ -1373,7 +1640,7 @@ int bch2_fs_initialize(struct bch_fs *c) 0, 0, S_IFDIR|0700, 0, NULL, NULL, (subvol_inum) { 0 }, 0)); if (ret) { - bch_err(c, "error creating lost+found"); + bch_err_msg(c, ret, "creating lost+found"); goto err; } @@ -1383,10 +1650,11 @@ int bch2_fs_initialize(struct bch_fs *c) goto err; } - err = "error writing first journal entry"; ret = bch2_journal_flush(&c->journal); - if (ret) + if (ret) { + bch_err_msg(c, ret, "writing first journal entry"); goto err; + } mutex_lock(&c->sb_lock); SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true); @@ -1397,6 +1665,6 @@ int bch2_fs_initialize(struct bch_fs *c) return 0; err: - pr_err("Error initializing new filesystem: %s (%i)", err, ret); + bch_err_fn(ca, ret); return ret; }