X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Frecovery.c;h=0b3521c9cc19ef3ba285805a05de94bcc8bfd717;hb=1e574cb1aa07ab3a796c7d6c5501b96f3056ef4d;hp=d1184bf62cae4e120cc61adf56df7351e2ff1921;hpb=f597e81aab00d0f5174f104f967cbf20da7c51c7;p=bcachefs-tools-debian diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c index d1184bf..0b3521c 100644 --- a/libbcachefs/recovery.c +++ b/libbcachefs/recovery.c @@ -25,45 +25,259 @@ #define QSTR(n) { { { .len = strlen(n) } }, .name = n } +/* for -o reconstruct_alloc: */ +static void drop_alloc_keys(struct journal_keys *keys) +{ + size_t src, dst; + + for (src = 0, dst = 0; src < keys->nr; src++) + if (keys->d[src].btree_id != BTREE_ID_ALLOC) + keys->d[dst++] = keys->d[src]; + + keys->nr = dst; +} + /* iterate over keys read from the journal: */ -struct journal_iter bch2_journal_iter_init(struct journal_keys *keys, - enum btree_id id) +static struct journal_key *journal_key_search(struct journal_keys *journal_keys, + enum btree_id id, unsigned level, + struct bpos pos) +{ + size_t l = 0, r = journal_keys->nr, m; + + while (l < r) { + m = l + ((r - l) >> 1); + if ((cmp_int(id, journal_keys->d[m].btree_id) ?: + cmp_int(level, journal_keys->d[m].level) ?: + bkey_cmp(pos, journal_keys->d[m].k->k.p)) > 0) + l = m + 1; + else + r = m; + } + + BUG_ON(l < journal_keys->nr && + (cmp_int(id, journal_keys->d[l].btree_id) ?: + cmp_int(level, journal_keys->d[l].level) ?: + bkey_cmp(pos, journal_keys->d[l].k->k.p)) > 0); + + BUG_ON(l && + (cmp_int(id, journal_keys->d[l - 1].btree_id) ?: + cmp_int(level, journal_keys->d[l - 1].level) ?: + bkey_cmp(pos, journal_keys->d[l - 1].k->k.p)) <= 0); + + return l < journal_keys->nr ? journal_keys->d + l : NULL; +} + +static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter) +{ + if (iter->k && + iter->k < iter->keys->d + iter->keys->nr && + iter->k->btree_id == iter->btree_id && + iter->k->level == iter->level) + return iter->k->k; + + iter->k = NULL; + return NULL; +} + +static void bch2_journal_iter_advance(struct journal_iter *iter) +{ + if (iter->k) + iter->k++; +} + +static void bch2_journal_iter_init(struct journal_iter *iter, + struct journal_keys *journal_keys, + enum btree_id id, unsigned level, + struct bpos pos) { - return (struct journal_iter) { - .keys = keys, - .k = keys->d, - .btree_id = id, - }; + iter->btree_id = id; + iter->level = level; + iter->keys = journal_keys; + iter->k = journal_key_search(journal_keys, id, level, pos); } -struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter) +static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter) { + return iter->btree + ? bch2_btree_iter_peek(iter->btree) + : bch2_btree_node_iter_peek_unpack(&iter->node_iter, + iter->b, &iter->unpacked); +} + +static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter) +{ + if (iter->btree) + bch2_btree_iter_next(iter->btree); + else + bch2_btree_node_iter_advance(&iter->node_iter, iter->b); +} + +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; +} + +struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter) +{ + struct bkey_s_c ret; + while (1) { - if (iter->k == iter->keys->d + iter->keys->nr) + 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 = bkey_cmp(btree_k.k->p, journal_k.k->p); + + if (!cmp) + bch2_journal_iter_advance_btree(iter); + + 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; + } - if (iter->k->btree_id == iter->btree_id) - return bkey_i_to_s_c(iter->k->k); + ret = iter->last == journal ? journal_k : btree_k; - iter->k++; + if (iter->b && + bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) { + iter->journal.k = NULL; + iter->last = none; + return bkey_s_c_null; + } + + if (!bkey_deleted(ret.k)) + break; + + bch2_btree_and_journal_iter_advance(iter); + } + + 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_init(struct btree_and_journal_iter *iter, + struct btree_trans *trans, + struct journal_keys *journal_keys, + enum btree_id id, struct bpos pos) +{ + memset(iter, 0, sizeof(*iter)); + + iter->btree = bch2_trans_get_iter(trans, id, pos, 0); + bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos); +} + +void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter, + struct journal_keys *journal_keys, + struct btree *b) +{ + memset(iter, 0, sizeof(*iter)); + + iter->b = b; + bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b); + bch2_journal_iter_init(&iter->journal, journal_keys, + b->c.btree_id, b->c.level, b->data->min_key); +} + +/* Walk btree, overlaying keys from the journal: */ + +static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b, + struct journal_keys *journal_keys, + enum btree_id btree_id, + btree_walk_node_fn node_fn, + btree_walk_key_fn key_fn) +{ + struct btree_and_journal_iter iter; + struct bkey_s_c k; + int ret = 0; + + bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b); + + while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { + ret = key_fn(c, btree_id, b->c.level, k); + if (ret) + break; + + if (b->c.level) { + struct btree *child; + BKEY_PADDED(k) tmp; + + bkey_reassemble(&tmp.k, k); + k = bkey_i_to_s_c(&tmp.k); + + bch2_btree_and_journal_iter_advance(&iter); + + if (b->c.level > 0) { + child = bch2_btree_node_get_noiter(c, &tmp.k, + b->c.btree_id, b->c.level - 1); + ret = PTR_ERR_OR_ZERO(child); + if (ret) + break; + + ret = (node_fn ? node_fn(c, b) : 0) ?: + bch2_btree_and_journal_walk_recurse(c, child, + journal_keys, btree_id, node_fn, key_fn); + six_unlock_read(&child->c.lock); + + if (ret) + break; + } + } else { + bch2_btree_and_journal_iter_advance(&iter); + } } - return bkey_s_c_null; + return ret; } -struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter) +int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys, + enum btree_id btree_id, + btree_walk_node_fn node_fn, + btree_walk_key_fn key_fn) { - if (iter->k == iter->keys->d + iter->keys->nr) - return bkey_s_c_null; + struct btree *b = c->btree_roots[btree_id].b; + int ret = 0; + + if (btree_node_fake(b)) + return 0; + + six_lock_read(&b->c.lock, NULL, NULL); + ret = (node_fn ? node_fn(c, b) : 0) ?: + bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id, + node_fn, key_fn) ?: + key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key)); + six_unlock_read(&b->c.lock); - iter->k++; - return bch2_journal_iter_peek(iter); + return ret; } /* sort and dedup all keys in the journal: */ -static void journal_entries_free(struct list_head *list) +void bch2_journal_entries_free(struct list_head *list) { while (!list_empty(list)) { @@ -75,43 +289,23 @@ static void journal_entries_free(struct list_head *list) } } +/* + * When keys compare equal, oldest compares first: + */ static int journal_sort_key_cmp(const void *_l, const void *_r) { const struct journal_key *l = _l; const struct journal_key *r = _r; - return cmp_int(l->btree_id, r->btree_id) ?: - bkey_cmp(l->pos, r->pos) ?: + return cmp_int(l->btree_id, r->btree_id) ?: + cmp_int(l->level, r->level) ?: + bkey_cmp(l->k->k.p, r->k->k.p) ?: cmp_int(l->journal_seq, r->journal_seq) ?: cmp_int(l->journal_offset, r->journal_offset); } -static int journal_sort_seq_cmp(const void *_l, const void *_r) +void bch2_journal_keys_free(struct journal_keys *keys) { - const struct journal_key *l = _l; - const struct journal_key *r = _r; - - return cmp_int(l->journal_seq, r->journal_seq) ?: - cmp_int(l->btree_id, r->btree_id) ?: - bkey_cmp(l->pos, r->pos); -} - -static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i) -{ - while (i + 1 < keys->d + keys->nr && - journal_sort_key_cmp(i, i + 1) > 0) { - swap(i[0], i[1]); - i++; - } -} - -static void journal_keys_free(struct journal_keys *keys) -{ - struct journal_key *i; - - for_each_journal_key(*keys, i) - if (i->allocated) - kfree(i->k); kvfree(keys->d); keys->d = NULL; keys->nr = 0; @@ -122,108 +316,61 @@ static struct journal_keys journal_keys_sort(struct list_head *journal_entries) struct journal_replay *p; struct jset_entry *entry; struct bkey_i *k, *_n; - struct journal_keys keys = { NULL }, keys_deduped = { NULL }; - struct journal_key *i; + struct journal_keys keys = { NULL }; + struct journal_key *src, *dst; size_t nr_keys = 0; - list_for_each_entry(p, journal_entries, list) + if (list_empty(journal_entries)) + return keys; + + keys.journal_seq_base = + le64_to_cpu(list_last_entry(journal_entries, + struct journal_replay, list)->j.last_seq); + + list_for_each_entry(p, journal_entries, list) { + if (le64_to_cpu(p->j.seq) < keys.journal_seq_base) + continue; + for_each_jset_key(k, _n, entry, &p->j) nr_keys++; + } - keys.journal_seq_base = keys_deduped.journal_seq_base = - le64_to_cpu(list_first_entry(journal_entries, - struct journal_replay, - list)->j.seq); keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL); if (!keys.d) goto err; - keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL); - if (!keys_deduped.d) - goto err; + list_for_each_entry(p, journal_entries, list) { + if (le64_to_cpu(p->j.seq) < keys.journal_seq_base) + continue; - list_for_each_entry(p, journal_entries, list) for_each_jset_key(k, _n, entry, &p->j) keys.d[keys.nr++] = (struct journal_key) { .btree_id = entry->btree_id, - .pos = bkey_start_pos(&k->k), + .level = entry->level, .k = k, .journal_seq = le64_to_cpu(p->j.seq) - keys.journal_seq_base, .journal_offset = k->_data - p->j._data, }; + } - sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL); - - i = keys.d; - while (i < keys.d + keys.nr) { - if (i + 1 < keys.d + keys.nr && - i[0].btree_id == i[1].btree_id && - !bkey_cmp(i[0].pos, i[1].pos)) { - if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) { - i++; - } else { - bch2_cut_front(i[1].k->k.p, i[0].k); - i[0].pos = i[1].k->k.p; - journal_keys_sift(&keys, i); - } - continue; - } + sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL); - if (i + 1 < keys.d + keys.nr && - i[0].btree_id == i[1].btree_id && - bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) { - if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?: - cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) { - if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) { - bch2_cut_back(bkey_start_pos(&i[1].k->k), &i[0].k->k); - } else { - struct bkey_i *split = - kmalloc(bkey_bytes(i[0].k), GFP_KERNEL); - - if (!split) - goto err; - - bkey_copy(split, i[0].k); - bch2_cut_back(bkey_start_pos(&i[1].k->k), &split->k); - keys_deduped.d[keys_deduped.nr++] = (struct journal_key) { - .btree_id = i[0].btree_id, - .allocated = true, - .pos = bkey_start_pos(&split->k), - .k = split, - .journal_seq = i[0].journal_seq, - .journal_offset = i[0].journal_offset, - }; - - bch2_cut_front(i[1].k->k.p, i[0].k); - i[0].pos = i[1].k->k.p; - journal_keys_sift(&keys, i); - continue; - } - } else { - if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) { - i[1] = i[0]; - i++; - continue; - } else { - bch2_cut_front(i[0].k->k.p, i[1].k); - i[1].pos = i[0].k->k.p; - journal_keys_sift(&keys, i + 1); - 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 && + !bkey_cmp(src[0].k->k.p, src[1].k->k.p)) + src++; - keys_deduped.d[keys_deduped.nr++] = *i++; + *dst++ = *src++; } - kvfree(keys.d); - return keys_deduped; + keys.nr = dst - keys.d; err: - journal_keys_free(&keys_deduped); - kvfree(keys.d); - return (struct journal_keys) { NULL }; + return keys; } /* journal replay: */ @@ -254,7 +401,7 @@ static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id, * Some extents aren't equivalent - w.r.t. what the triggers do * - if they're split: */ - bool remark_if_split = bch2_extent_is_compressed(bkey_i_to_s_c(k)) || + bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) || k->k.type == KEY_TYPE_reflink_p; bool remark = false; int ret; @@ -274,11 +421,6 @@ retry: atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p); - split_iter = bch2_trans_copy_iter(&trans, iter); - ret = PTR_ERR_OR_ZERO(split_iter); - if (ret) - goto err; - split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k)); ret = PTR_ERR_OR_ZERO(split); if (ret) @@ -289,7 +431,7 @@ retry: bkey_cmp(atomic_end, k->k.p) < 0) { ret = bch2_disk_reservation_add(c, &disk_res, k->k.size * - bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(k)), + bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)), BCH_DISK_RESERVATION_NOFAIL); BUG_ON(ret); @@ -297,35 +439,51 @@ retry: } bkey_copy(split, k); - bch2_cut_front(split_iter->pos, split); - bch2_cut_back(atomic_end, &split->k); + bch2_cut_front(iter->pos, split); + bch2_cut_back(atomic_end, split); + + split_iter = bch2_trans_copy_iter(&trans, iter); + ret = PTR_ERR_OR_ZERO(split_iter); + if (ret) + goto err; + + /* + * It's important that we don't go through the + * extent_handle_overwrites() and extent_update_to_keys() path + * here: journal replay is supposed to treat extents like + * regular keys + */ + __bch2_btree_iter_set_pos(split_iter, split->k.p, false); + bch2_trans_update(&trans, split_iter, split, + BTREE_TRIGGER_NORUN); + bch2_trans_iter_put(&trans, split_iter); - bch2_trans_update(&trans, split_iter, split); bch2_btree_iter_set_pos(iter, split->k.p); + + if (remark) { + ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(split), + 0, split->k.size, + BTREE_TRIGGER_INSERT); + if (ret) + goto err; + } } while (bkey_cmp(iter->pos, k->k.p) < 0); if (remark) { ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k), 0, -((s64) k->k.size), - BCH_BUCKET_MARK_OVERWRITE) ?: - bch2_trans_commit(&trans, &disk_res, NULL, - BTREE_INSERT_ATOMIC| - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW| - BTREE_INSERT_NOMARK_OVERWRITES| - BTREE_INSERT_NO_CLEAR_REPLICAS); - } else { - ret = bch2_trans_commit(&trans, &disk_res, NULL, - BTREE_INSERT_ATOMIC| - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW| - BTREE_INSERT_JOURNAL_REPLAY| - BTREE_INSERT_NOMARK); + BTREE_TRIGGER_OVERWRITE); + if (ret) + goto err; } - if (ret) - goto err; + ret = bch2_trans_commit(&trans, &disk_res, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW| + BTREE_INSERT_JOURNAL_REPLAY); err: + bch2_trans_iter_put(&trans, iter); + if (ret == -EINTR) goto retry; @@ -334,37 +492,150 @@ err: return bch2_trans_exit(&trans) ?: ret; } +static int __bch2_journal_replay_key(struct btree_trans *trans, + enum btree_id id, unsigned level, + struct bkey_i *k) +{ + struct btree_iter *iter; + int ret; + + iter = bch2_trans_get_node_iter(trans, id, k->k.p, + BTREE_MAX_DEPTH, level, + BTREE_ITER_INTENT); + if (IS_ERR(iter)) + return PTR_ERR(iter); + + /* + * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run + * extent_handle_overwrites() and extent_update_to_keys() - but we don't + * want that here, journal replay is supposed to treat extents like + * regular keys: + */ + __bch2_btree_iter_set_pos(iter, k->k.p, false); + + ret = bch2_btree_iter_traverse(iter) ?: + bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN); + bch2_trans_iter_put(trans, iter); + return ret; +} + +static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id, + unsigned level, struct bkey_i *k) +{ + return bch2_trans_do(c, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW| + BTREE_INSERT_JOURNAL_REPLAY, + __bch2_journal_replay_key(&trans, id, level, k)); +} + +static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k) +{ + struct btree_iter *iter; + int ret; + + iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p, + BTREE_ITER_CACHED| + BTREE_ITER_CACHED_NOFILL| + BTREE_ITER_INTENT); + ret = PTR_ERR_OR_ZERO(iter) ?: + bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN); + bch2_trans_iter_put(trans, iter); + return ret; +} + +static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k) +{ + return bch2_trans_do(c, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_USE_RESERVE| + BTREE_INSERT_LAZY_RW| + BTREE_INSERT_JOURNAL_REPLAY, + __bch2_alloc_replay_key(&trans, k)); +} + +static int journal_sort_seq_cmp(const void *_l, const void *_r) +{ + const struct journal_key *l = _l; + const struct journal_key *r = _r; + + return cmp_int(r->level, l->level) ?: + cmp_int(l->journal_seq, r->journal_seq) ?: + cmp_int(l->btree_id, r->btree_id) ?: + bkey_cmp(l->k->k.p, r->k->k.p); +} + static int bch2_journal_replay(struct bch_fs *c, struct journal_keys keys) { struct journal *j = &c->journal; struct journal_key *i; + u64 seq; int ret; sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL); + if (keys.nr) + replay_now_at(j, keys.journal_seq_base); + + seq = j->replay_journal_seq; + + /* + * First replay updates to the alloc btree - these will only update the + * btree key cache: + */ for_each_journal_key(keys, i) { - replay_now_at(j, keys.journal_seq_base + i->journal_seq); + cond_resched(); - if (i->btree_id == BTREE_ID_ALLOC) + if (!i->level && i->btree_id == BTREE_ID_ALLOC) { + j->replay_journal_seq = keys.journal_seq_base + i->journal_seq; ret = bch2_alloc_replay_key(c, i->k); - else if (btree_node_type_is_extents(i->btree_id)) - ret = bch2_extent_replay_key(c, i->btree_id, i->k); - else - ret = bch2_btree_insert(c, i->btree_id, i->k, - NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW| - BTREE_INSERT_JOURNAL_REPLAY| - BTREE_INSERT_NOMARK); + if (ret) + goto err; + } + } - if (ret) { - bch_err(c, "journal replay: error %d while replaying key", - ret); - return ret; + /* + * Next replay updates to interior btree nodes: + */ + for_each_journal_key(keys, i) { + cond_resched(); + + if (i->level) { + j->replay_journal_seq = keys.journal_seq_base + i->journal_seq; + ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k); + if (ret) + goto err; } + } + + /* + * Now that the btree is in a consistent state, we can start journal + * reclaim (which will be flushing entries from the btree key cache back + * to the btree: + */ + set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags); + set_bit(JOURNAL_RECLAIM_STARTED, &j->flags); + journal_reclaim_kick(j); + + j->replay_journal_seq = seq; + /* + * Now replay leaf node updates: + */ + for_each_journal_key(keys, i) { cond_resched(); + + if (i->level || i->btree_id == BTREE_ID_ALLOC) + continue; + + replay_now_at(j, keys.journal_seq_base + i->journal_seq); + + ret = i->k->k.size + ? bch2_extent_replay_key(c, i->btree_id, i->k) + : bch2_journal_replay_key(c, i->btree_id, i->level, i->k); + if (ret) + goto err; } replay_now_at(j, j->replay_journal_seq_end); @@ -373,6 +644,9 @@ static int bch2_journal_replay(struct bch_fs *c, bch2_journal_set_replay_done(j); bch2_journal_flush_all_pins(j); return bch2_journal_error(j); +err: + bch_err(c, "journal replay: error %d while replaying key", ret); + return ret; } static bool journal_empty(struct list_head *journal) @@ -394,6 +668,9 @@ verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c, int ret = 0; list_for_each_entry(i, journal, list) { + if (le64_to_cpu(i->j.seq) < start_seq) + continue; + fsck_err_on(seq != le64_to_cpu(i->j.seq), c, "journal entries %llu-%llu missing! (replaying %llu-%llu)", seq, le64_to_cpu(i->j.seq) - 1, @@ -584,11 +861,14 @@ static int verify_superblock_clean(struct bch_fs *c, } mustfix_fsck_err_on(j->read_clock != clean->read_clock, c, - "superblock read clock doesn't match journal after clean shutdown"); + "superblock read clock %u doesn't match journal %u after clean shutdown", + clean->read_clock, j->read_clock); mustfix_fsck_err_on(j->write_clock != clean->write_clock, c, - "superblock read clock doesn't match journal after clean shutdown"); + "superblock write clock %u doesn't match journal %u after clean shutdown", + clean->write_clock, j->write_clock); for (i = 0; i < BTREE_ID_NR; i++) { + char buf1[200], buf2[200]; struct bkey_i *k1, *k2; unsigned l1 = 0, l2 = 0; @@ -604,7 +884,11 @@ static int verify_superblock_clean(struct bch_fs *c, k1->k.u64s != k2->k.u64s || memcmp(k1, k2, bkey_bytes(k1)) || l1 != l2, c, - "superblock btree root doesn't match journal after clean shutdown"); + "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)); } fsck_err: return ret; @@ -662,7 +946,6 @@ static int read_btree_roots(struct bch_fs *c) continue; } - if (r->error) { __fsck_err(c, i == BTREE_ID_ALLOC ? FSCK_CAN_IGNORE : 0, @@ -695,9 +978,7 @@ int bch2_fs_recovery(struct bch_fs *c) const char *err = "cannot allocate memory"; struct bch_sb_field_clean *clean = NULL; u64 journal_seq; - LIST_HEAD(journal_entries); - struct journal_keys journal_keys = { NULL }; - bool wrote = false, write_sb = false; + bool write_sb = false, need_write_alloc = false; int ret; if (c->sb.clean) @@ -710,38 +991,39 @@ int bch2_fs_recovery(struct bch_fs *c) bch_info(c, "recovering from clean shutdown, journal seq %llu", le64_to_cpu(clean->journal_seq)); - if (!c->replicas.entries) { + if (!c->replicas.entries || + c->opts.rebuild_replicas) { bch_info(c, "building replicas info"); set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags); } - if (!c->sb.clean || c->opts.fsck) { + if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) { struct jset *j; - ret = bch2_journal_read(c, &journal_entries); + ret = bch2_journal_read(c, &c->journal_entries); if (ret) goto err; - if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c, + if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c, "filesystem marked clean but journal not empty")) { c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO); SET_BCH_SB_CLEAN(c->disk_sb.sb, false); c->sb.clean = false; } - if (!c->sb.clean && list_empty(&journal_entries)) { + if (!c->sb.clean && list_empty(&c->journal_entries)) { bch_err(c, "no journal entries found"); ret = BCH_FSCK_REPAIR_IMPOSSIBLE; goto err; } - journal_keys = journal_keys_sort(&journal_entries); - if (!journal_keys.d) { + c->journal_keys = journal_keys_sort(&c->journal_entries); + if (!c->journal_keys.d) { ret = -ENOMEM; goto err; } - j = &list_last_entry(&journal_entries, + j = &list_last_entry(&c->journal_entries, struct journal_replay, list)->j; ret = verify_superblock_clean(c, &clean, j); @@ -753,7 +1035,19 @@ int bch2_fs_recovery(struct bch_fs *c) journal_seq = le64_to_cpu(clean->journal_seq) + 1; } - ret = journal_replay_early(c, clean, &journal_entries); + if (!c->sb.clean && + !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) { + bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix"); + ret = -EINVAL; + goto err; + } + + if (c->opts.reconstruct_alloc) { + c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO); + drop_alloc_keys(&c->journal_keys); + } + + ret = journal_replay_early(c, clean, &c->journal_entries); if (ret) goto err; @@ -767,19 +1061,24 @@ int bch2_fs_recovery(struct bch_fs *c) } journal_seq += 4; + + /* + * The superblock needs to be written before we do any btree + * node writes: it will be in the read_write() path + */ } ret = bch2_blacklist_table_initialize(c); - if (!list_empty(&journal_entries)) { + if (!list_empty(&c->journal_entries)) { ret = verify_journal_entries_not_blacklisted_or_missing(c, - &journal_entries); + &c->journal_entries); if (ret) goto err; } ret = bch2_fs_journal_start(&c->journal, journal_seq, - &journal_entries); + &c->journal_entries); if (ret) goto err; @@ -789,14 +1088,14 @@ int bch2_fs_recovery(struct bch_fs *c) bch_verbose(c, "starting alloc read"); err = "error reading allocation information"; - ret = bch2_alloc_read(c, &journal_keys); + ret = bch2_alloc_read(c, &c->journal_keys); 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, &journal_keys); + ret = bch2_stripes_read(c, &c->journal_keys); if (ret) goto err; bch_verbose(c, "stripes_read done"); @@ -812,9 +1111,11 @@ int bch2_fs_recovery(struct bch_fs *c) */ bch_info(c, "starting metadata mark and sweep"); err = "error in mark and sweep"; - ret = bch2_gc(c, NULL, true, true); - if (ret) + ret = bch2_gc(c, &c->journal_keys, true, true); + if (ret < 0) goto err; + if (ret) + need_write_alloc = true; bch_verbose(c, "mark and sweep done"); } @@ -823,9 +1124,11 @@ int bch2_fs_recovery(struct bch_fs *c) test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) { bch_info(c, "starting mark and sweep"); err = "error in mark and sweep"; - ret = bch2_gc(c, &journal_keys, true, false); - if (ret) + ret = bch2_gc(c, &c->journal_keys, true, false); + if (ret < 0) goto err; + if (ret) + need_write_alloc = true; bch_verbose(c, "mark and sweep done"); } @@ -844,12 +1147,12 @@ int bch2_fs_recovery(struct bch_fs *c) bch_verbose(c, "starting journal replay"); err = "journal replay failed"; - ret = bch2_journal_replay(c, journal_keys); + ret = bch2_journal_replay(c, c->journal_keys); if (ret) goto err; bch_verbose(c, "journal replay done"); - if (!c->opts.nochanges) { + if (need_write_alloc && !c->opts.nochanges) { /* * note that even when filesystem was clean there might be work * to do here, if we ran gc (because of fsck) which recalculated @@ -857,8 +1160,8 @@ int bch2_fs_recovery(struct bch_fs *c) */ bch_verbose(c, "writing allocation info"); err = "error writing out alloc info"; - ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?: - bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote); + ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?: + bch2_alloc_write(c, BTREE_INSERT_LAZY_RW); if (ret) { bch_err(c, "error writing alloc info"); goto err; @@ -869,7 +1172,7 @@ int bch2_fs_recovery(struct bch_fs *c) } if (!c->sb.clean) { - if (!(c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) { + if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) { bch_info(c, "checking inode link counts"); err = "error in recovery"; ret = bch2_fsck_inode_nlink(c); @@ -910,6 +1213,7 @@ int bch2_fs_recovery(struct bch_fs *c) c->disk_sb.sb->version_min = le16_to_cpu(bcachefs_metadata_version_min); c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current); + c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL; write_sb = true; } @@ -920,7 +1224,7 @@ int bch2_fs_recovery(struct bch_fs *c) if (c->opts.fsck && !test_bit(BCH_FS_ERROR, &c->flags)) { - c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK; + c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink; SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0); write_sb = true; } @@ -939,8 +1243,10 @@ fsck_err: set_bit(BCH_FS_FSCK_DONE, &c->flags); bch2_flush_fsck_errs(c); - journal_keys_free(&journal_keys); - journal_entries_free(&journal_entries); + if (!c->opts.keep_journal) { + bch2_journal_keys_free(&c->journal_keys); + bch2_journal_entries_free(&c->journal_entries); + } kfree(clean); if (ret) bch_err(c, "Error in recovery: %s (%i)", err, ret); @@ -967,12 +1273,24 @@ int bch2_fs_initialize(struct bch_fs *c) bch2_mark_dev_superblock(c, ca, 0); mutex_unlock(&c->sb_lock); + mutex_lock(&c->sb_lock); + c->disk_sb.sb->version = c->disk_sb.sb->version_min = + le16_to_cpu(bcachefs_metadata_version_current); + c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink; + c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL; + + 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); for (i = 0; i < BTREE_ID_NR; i++) bch2_btree_root_alloc(c, i); + set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags); + set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags); + err = "unable to allocate journal buckets"; for_each_online_member(ca, c, i) { ret = bch2_dev_journal_alloc(ca); @@ -989,15 +1307,24 @@ int bch2_fs_initialize(struct bch_fs *c) bch2_fs_journal_start(&c->journal, 1, &journal); bch2_journal_set_replay_done(&c->journal); - err = "error going read write"; - ret = __bch2_fs_read_write(c, true); + err = "error going read-write"; + ret = bch2_fs_read_write_early(c); + if (ret) + goto err; + + /* + * Write out the superblock and journal buckets, now that we can do + * btree updates + */ + err = "error writing alloc info"; + ret = bch2_alloc_write(c, 0); if (ret) goto err; bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL); root_inode.bi_inum = BCACHEFS_ROOT_INO; - bch2_inode_pack(&packed_inode, &root_inode); + bch2_inode_pack(c, &packed_inode, &root_inode); err = "error creating root directory"; ret = bch2_btree_insert(c, BTREE_ID_INODES, @@ -1009,7 +1336,7 @@ int bch2_fs_initialize(struct bch_fs *c) bch2_inode_init_early(c, &lostfound_inode); err = "error creating lost+found"; - ret = bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC, + ret = bch2_trans_do(c, NULL, NULL, 0, bch2_create_trans(&trans, BCACHEFS_ROOT_INO, &root_inode, &lostfound_inode, &lostfound, @@ -1030,10 +1357,6 @@ int bch2_fs_initialize(struct bch_fs *c) goto err; mutex_lock(&c->sb_lock); - c->disk_sb.sb->version = c->disk_sb.sb->version_min = - le16_to_cpu(bcachefs_metadata_version_current); - c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK; - SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true); SET_BCH_SB_CLEAN(c->disk_sb.sb, false);