#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: */
static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
iter->b = b;
bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
bch2_journal_iter_init(&iter->journal, journal_keys,
- b->btree_id, b->level, b->data->min_key);
+ b->c.btree_id, b->c.level, b->data->min_key);
}
/* Walk btree, overlaying keys from the journal: */
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->level, k);
+ ret = key_fn(c, btree_id, b->c.level, k);
if (ret)
break;
- if (b->level) {
+ if (b->c.level) {
struct btree *child;
BKEY_PADDED(k) tmp;
bch2_btree_and_journal_iter_advance(&iter);
- if (b->level > 0) {
+ if (b->c.level > 0) {
child = bch2_btree_node_get_noiter(c, &tmp.k,
- b->btree_id, b->level - 1);
+ 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->lock);
+ six_unlock_read(&child->c.lock);
if (ret)
break;
if (btree_node_fake(b))
return 0;
- six_lock_read(&b->lock);
+ 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->level + 1, bkey_i_to_s_c(&b->key));
- six_unlock_read(&b->lock);
+ key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
+ six_unlock_read(&b->c.lock);
return ret;
}
cmp_int(l->journal_offset, r->journal_offset);
}
-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);
-}
-
void bch2_journal_keys_free(struct journal_keys *keys)
{
kvfree(keys->d);
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 =
- 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;
- list_for_each_entry(p, journal_entries, list)
+ 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)
keys.d[keys.nr++] = (struct journal_key) {
.btree_id = entry->btree_id,
keys.journal_seq_base,
.journal_offset = k->_data - p->j._data,
};
+ }
sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
* regular keys
*/
__bch2_btree_iter_set_pos(split_iter, split->k.p, false);
- bch2_trans_update(&trans, split_iter, split, !remark
- ? BTREE_TRIGGER_NORUN
- : BTREE_TRIGGER_NOOVERWRITES);
+ bch2_trans_update(&trans, split_iter, split,
+ BTREE_TRIGGER_NORUN);
+ bch2_trans_iter_put(&trans, split_iter);
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) {
BTREE_INSERT_LAZY_RW|
BTREE_INSERT_JOURNAL_REPLAY);
err:
+ bch2_trans_iter_put(&trans, iter);
+
if (ret == -EINTR)
goto retry;
__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) {
- if (!i->level)
- replay_now_at(j, keys.journal_seq_base + i->journal_seq);
+ cond_resched();
- if (i->level)
- ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
- 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 (i->k->k.size)
- ret = bch2_extent_replay_key(c, i->btree_id, i->k);
- else
- ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
+ 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);
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)
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,
}
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];
continue;
}
-
if (r->error) {
__fsck_err(c, i == BTREE_ID_ALLOC
? FSCK_CAN_IGNORE : 0,
const char *err = "cannot allocate memory";
struct bch_sb_field_clean *clean = NULL;
u64 journal_seq;
- bool wrote = false, write_sb = false;
+ bool write_sb = false, need_write_alloc = false;
int ret;
if (c->sb.clean)
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);
}
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;
}
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);
bch_info(c, "starting metadata mark and sweep");
err = "error in mark and sweep";
ret = bch2_gc(c, &c->journal_keys, true, true);
- if (ret)
+ if (ret < 0)
goto err;
+ if (ret)
+ need_write_alloc = true;
bch_verbose(c, "mark and sweep done");
}
bch_info(c, "starting mark and sweep");
err = "error in mark and sweep";
ret = bch2_gc(c, &c->journal_keys, true, false);
- if (ret)
+ if (ret < 0)
goto err;
+ if (ret)
+ need_write_alloc = true;
bch_verbose(c, "mark and sweep done");
}
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
*/
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;
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);
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_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,
&packed_inode.inode.k_i,
- NULL, NULL, BTREE_INSERT_LAZY_RW);
+ NULL, NULL, 0);
if (ret)
goto err;