bcachefs: Simplify journal replay

With BTREE_ITER_WITH_JOURNAL, there's no longer any restrictions on the
order we have to replay keys from the journal in, and we can also start
up journal reclaim right away - and delete a bunch of code.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

fs/bcachefs/alloc_background.c

@@ -902,8 +902,7 @@ static void discard_one_bucket(struct bch_fs *c, struct bch_dev *ca, u64 b)
 static bool allocator_thread_running(struct bch_dev *ca)
 {
 	unsigned state = ca->mi.state == BCH_MEMBER_STATE_rw &&
-		test_bit(BCH_FS_ALLOCATOR_RUNNING, &ca->fs->flags) &&
-		test_bit(BCH_FS_ALLOC_REPLAY_DONE, &ca->fs->flags)
+		test_bit(BCH_FS_ALLOCATOR_RUNNING, &ca->fs->flags)
 		? ALLOCATOR_running
 		: ALLOCATOR_stopped;
 	alloc_thread_set_state(ca, state);

fs/bcachefs/bcachefs.h

@@ -510,8 +510,6 @@ enum {
 	BCH_FS_INITIAL_GC_DONE,
 	BCH_FS_INITIAL_GC_UNFIXED,
 	BCH_FS_TOPOLOGY_REPAIR_DONE,
-	BCH_FS_ALLOC_REPLAY_DONE,
-	BCH_FS_BTREE_INTERIOR_REPLAY_DONE,
 	BCH_FS_FSCK_DONE,
 	BCH_FS_STARTED,
 	BCH_FS_RW,

fs/bcachefs/btree_key_cache.h

@@ -16,8 +16,7 @@ static inline bool bch2_btree_key_cache_must_wait(struct bch_fs *c)
 	size_t nr_keys = atomic_long_read(&c->btree_key_cache.nr_keys);
 	size_t max_dirty = 4096 + (nr_keys * 3) / 4;
 
-	return nr_dirty > max_dirty &&
-		test_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
+	return nr_dirty > max_dirty;
 }
 
 int bch2_btree_key_cache_journal_flush(struct journal *,

fs/bcachefs/btree_update_interior.c

@@ -45,7 +45,7 @@ static void btree_node_interior_verify(struct bch_fs *c, struct btree *b)
 
 	BUG_ON(!b->c.level);
 
-	if (!test_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags))
+	if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
 		return;
 
 	bch2_btree_node_iter_init_from_start(&iter, b);
@@ -1851,9 +1851,6 @@ void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b)
 {
 	struct async_btree_rewrite *a;
 
-	if (!test_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags))
-		return;
-
 	if (!percpu_ref_tryget(&c->writes))
 		return;
 

fs/bcachefs/btree_update_leaf.c

@@ -206,9 +206,6 @@ static bool btree_insert_key_leaf(struct btree_trans *trans,
 	int old_live_u64s = b->nr.live_u64s;
 	int live_u64s_added, u64s_added;
 
-	EBUG_ON(!insert->level &&
-		!test_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags));
-
 	if (unlikely(!bch2_btree_bset_insert_key(trans, insert->path, b,
 					&insert_l(insert)->iter, insert->k)))
 		return false;

fs/bcachefs/journal_reclaim.c

@@ -489,9 +489,6 @@ static size_t journal_flush_pins(struct journal *j, u64 seq_to_flush,
 	u64 seq;
 	int err;
 
-	if (!test_bit(JOURNAL_RECLAIM_STARTED, &j->flags))
-		return 0;
-
 	lockdep_assert_held(&j->reclaim_lock);
 
 	while (1) {
@@ -689,8 +686,6 @@ static int bch2_journal_reclaim_thread(void *arg)
 
 	set_freezable();
 
-	kthread_wait_freezable(test_bit(JOURNAL_RECLAIM_STARTED, &j->flags));
-
 	j->last_flushed = jiffies;
 
 	while (!ret && !kthread_should_stop()) {

fs/bcachefs/journal_types.h

@@ -148,7 +148,6 @@ enum journal_space_from {
 enum {
 	JOURNAL_REPLAY_DONE,
 	JOURNAL_STARTED,
-	JOURNAL_RECLAIM_STARTED,
 	JOURNAL_NEED_WRITE,
 	JOURNAL_MAY_GET_UNRESERVED,
 	JOURNAL_MAY_SKIP_FLUSH,

fs/bcachefs/recovery.c

@@ -474,8 +474,8 @@ static void replay_now_at(struct journal *j, u64 seq)
 		bch2_journal_pin_put(j, j->replay_journal_seq++);
 }
 
-static int __bch2_journal_replay_key(struct btree_trans *trans,
-				     struct journal_key *k)
+static int bch2_journal_replay_key(struct btree_trans *trans,
+				   struct journal_key *k)
 {
 	struct btree_iter iter;
 	unsigned iter_flags =
@@ -484,7 +484,7 @@ static int __bch2_journal_replay_key(struct btree_trans *trans,
 	int ret;
 
 	if (!k->level && k->btree_id == BTREE_ID_alloc)
-		iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
+		iter_flags |= BTREE_ITER_CACHED;
 
 	bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 				  BTREE_MAX_DEPTH, k->level,
@@ -503,29 +503,12 @@ static int __bch2_journal_replay_key(struct btree_trans *trans,
 	return ret;
 }
 
-static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
-{
-	unsigned commit_flags =
-		BTREE_INSERT_LAZY_RW|
-		BTREE_INSERT_NOFAIL|
-		BTREE_INSERT_JOURNAL_RESERVED;
-
-	if (!k->allocated)
-		commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
-
-	return bch2_trans_do(c, NULL, NULL, commit_flags,
-			     __bch2_journal_replay_key(&trans, k));
-}
-
 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 {
 	const struct journal_key *l = *((const struct journal_key **)_l);
 	const struct journal_key *r = *((const struct journal_key **)_r);
 
-	return  cmp_int(r->level,	l->level) ?:
-		cmp_int(l->journal_seq, r->journal_seq) ?:
-		cmp_int(l->btree_id,	r->btree_id) ?:
-		bpos_cmp(l->k->k.p,	r->k->k.p);
+	return cmp_int(l->journal_seq, r->journal_seq);
 }
 
 static int bch2_journal_replay(struct bch_fs *c)
@@ -533,10 +516,7 @@ 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;
-	struct bch_dev *ca;
-	unsigned idx;
 	size_t i;
-	u64 seq;
 	int ret;
 
 	keys_sorted = kmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
@@ -555,73 +535,25 @@ static int bch2_journal_replay(struct bch_fs *c)
 		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 (i = 0; i < keys->nr; i++) {
-		k = keys_sorted[i];
-
-		cond_resched();
-
-		if (!k->level && k->btree_id == BTREE_ID_alloc) {
-			j->replay_journal_seq = keys->journal_seq_base + k->journal_seq;
-			ret = bch2_journal_replay_key(c, k);
-			if (ret)
-				goto err;
-		}
-	}
-
-	/* Now we can start the allocator threads: */
-	set_bit(BCH_FS_ALLOC_REPLAY_DONE, &c->flags);
-	for_each_member_device(ca, c, idx)
-		bch2_wake_allocator(ca);
-
-	/*
-	 * Next replay updates to interior btree nodes:
-	 */
-	for (i = 0; i < keys->nr; i++) {
-		k = keys_sorted[i];
-
-		cond_resched();
-
-		if (k->level) {
-			j->replay_journal_seq = keys->journal_seq_base + k->journal_seq;
-			ret = bch2_journal_replay_key(c, 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 (i = 0; i < keys->nr; i++) {
 		k = keys_sorted[i];
 
 		cond_resched();
 
-		if (k->level || k->btree_id == BTREE_ID_alloc)
-			continue;
-
-		replay_now_at(j, keys->journal_seq_base + k->journal_seq);
+		if (!k->allocated)
+			replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 
-		ret = bch2_journal_replay_key(c, k);
-		if (ret)
+		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),
+			     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);
 			goto err;
+		}
 	}
 
 	replay_now_at(j, j->replay_journal_seq_end);
@@ -629,14 +561,9 @@ static int bch2_journal_replay(struct bch_fs *c)
 
 	bch2_journal_set_replay_done(j);
 	bch2_journal_flush_all_pins(j);
-	kfree(keys_sorted);
-
-	return bch2_journal_error(j);
+	ret = bch2_journal_error(j);
 err:
-	bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
-		ret, bch2_btree_ids[k->btree_id], k->level);
 	kfree(keys_sorted);
-
 	return ret;
 }
 
@@ -1215,7 +1142,8 @@ int bch2_fs_recovery(struct bch_fs *c)
 	ret = bch2_journal_replay(c);
 	if (ret)
 		goto err;
-	bch_verbose(c, "journal replay done");
+	if (c->opts.verbose || !c->sb.clean)
+		bch_info(c, "journal replay done");
 
 	if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
 	    !c->opts.nochanges) {
@@ -1385,10 +1313,6 @@ int bch2_fs_initialize(struct bch_fs *c)
 	for (i = 0; i < BTREE_ID_NR; i++)
 		bch2_btree_root_alloc(c, i);
 
-	set_bit(BCH_FS_ALLOC_REPLAY_DONE, &c->flags);
-	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);