bcache: don't flush journal unnecessarily

3 files changed, 43 insertions, 4 deletions

diff --git a/drivers/md/bcache/btree_update.c b/drivers/md/bcache/btree_update.c
index 8b5c5b43d0f8..e5771185155d 100644
--- a/drivers/md/bcache/btree_update.c
+++ b/drivers/md/bcache/btree_update.c
@@ -1033,7 +1033,7 @@ static void btree_interior_update_updated_btree(struct cache_set *c,
 
 	mutex_unlock(&c->btree_interior_update_lock);
 
-	bch_journal_flush_seq_async(&c->journal, as->journal_seq, &as->cl);
+	bch_journal_wait_on_seq(&c->journal, as->journal_seq, &as->cl);
 
 	continue_at(&as->cl, btree_interior_update_nodes_written,
 		    system_freezable_wq);
@@ -1068,12 +1068,20 @@ static void btree_interior_update_updated_root(struct cache_set *c,
 
 	mutex_unlock(&c->btree_interior_update_lock);
 
-	bch_journal_flush_seq_async(&c->journal, as->journal_seq, &as->cl);
+	bch_journal_wait_on_seq(&c->journal, as->journal_seq, &as->cl);
 
 	continue_at(&as->cl, btree_interior_update_nodes_written,
 		    system_freezable_wq);
 }
 
+static void interior_update_flush(struct journal *j, struct journal_entry_pin *pin)
+{
+	struct btree_interior_update *as =
+		container_of(pin, struct btree_interior_update, journal);
+
+	bch_journal_flush_seq_async(j, as->journal_seq, NULL);
+}
+
 /*
  * @b is being split/rewritten: it may have pointers to not-yet-written btree
  * nodes and thus outstanding btree_interior_updates - redirect @b's
@@ -1108,10 +1116,10 @@ void bch_btree_interior_update_will_free_node(struct cache_set *c,
 	 */
 	bch_journal_pin_add_if_older(&c->journal,
 				     &b->writes[0].journal,
-				     &as->journal, NULL);
+				     &as->journal, interior_update_flush);
 	bch_journal_pin_add_if_older(&c->journal,
 				     &b->writes[1].journal,
-				     &as->journal, NULL);
+				     &as->journal, interior_update_flush);
 
 	mutex_lock(&c->btree_interior_update_lock);
 
diff --git a/drivers/md/bcache/journal.c b/drivers/md/bcache/journal.c
index 3a30b102c093..3f26c36c9da5 100644
--- a/drivers/md/bcache/journal.c
+++ b/drivers/md/bcache/journal.c
@@ -2111,6 +2111,36 @@ int bch_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 	return ret < 0 ? ret : 0;
 }
 
+void bch_journal_wait_on_seq(struct journal *j, u64 seq, struct closure *parent)
+{
+	spin_lock(&j->lock);
+
+	BUG_ON(seq > atomic64_read(&j->seq));
+
+	if (bch_journal_error(j)) {
+		spin_unlock(&j->lock);
+		return;
+	}
+
+	if (seq == atomic64_read(&j->seq)) {
+		if (!closure_wait(&journal_cur_buf(j)->wait, parent))
+			BUG();
+	} else if (seq + 1 == atomic64_read(&j->seq) &&
+		   j->reservations.prev_buf_unwritten) {
+		if (!closure_wait(&journal_prev_buf(j)->wait, parent))
+			BUG();
+
+		smp_mb();
+
+		/* check if raced with write completion (or failure) */
+		if (!j->reservations.prev_buf_unwritten ||
+		    bch_journal_error(j))
+			closure_wake_up(&journal_prev_buf(j)->wait);
+	}
+
+	spin_unlock(&j->lock);
+}
+
 void bch_journal_flush_seq_async(struct journal *j, u64 seq, struct closure *parent)
 {
 	spin_lock(&j->lock);
diff --git a/drivers/md/bcache/journal.h b/drivers/md/bcache/journal.h
index 940e697d5d47..6fe7014ecd76 100644
--- a/drivers/md/bcache/journal.h
+++ b/drivers/md/bcache/journal.h
@@ -315,6 +315,7 @@ out:
 	return 0;
 }
 
+void bch_journal_wait_on_seq(struct journal *, u64, struct closure *);
 void bch_journal_flush_seq_async(struct journal *, u64, struct closure *);
 void bch_journal_flush_async(struct journal *, struct closure *);
 void bch_journal_meta_async(struct journal *, struct closure *);