bcachefs: Really don't hold btree locks while btree IOs are in flight

This is something we've attempted to stick to for quite some time, as it
helps guarantee filesystem latency - but there's a few remaining paths
that this patch fixes.

This is also necessary for an upcoming patch to update btree pointers
after every btree write - since the btree write completion path will now
be doing btree operations.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/btree_cache.c

@@ -187,6 +187,17 @@ static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush)
 	int ret = 0;
 
 	lockdep_assert_held(&bc->lock);
+wait_on_io:
+	if (b->flags & ((1U << BTREE_NODE_dirty)|
+			(1U << BTREE_NODE_read_in_flight)|
+			(1U << BTREE_NODE_write_in_flight))) {
+		if (!flush)
+			return -ENOMEM;
+
+		/* XXX: waiting on IO with btree cache lock held */
+		bch2_btree_node_wait_on_read(b);
+		bch2_btree_node_wait_on_write(b);
+	}
 
 	if (!six_trylock_intent(&b->c.lock))
 		return -ENOMEM;
@@ -194,25 +205,26 @@ static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush)
 	if (!six_trylock_write(&b->c.lock))
 		goto out_unlock_intent;
 
+	/* recheck under lock */
+	if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
+			(1U << BTREE_NODE_write_in_flight))) {
+		if (!flush)
+			goto out_unlock;
+		six_unlock_write(&b->c.lock);
+		six_unlock_intent(&b->c.lock);
+		goto wait_on_io;
+	}
+
 	if (btree_node_noevict(b))
 		goto out_unlock;
 
 	if (!btree_node_may_write(b))
 		goto out_unlock;
 
-	if (btree_node_dirty(b) &&
-	    test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags))
-		goto out_unlock;
-
-	if (btree_node_dirty(b) ||
-	    btree_node_write_in_flight(b) ||
-	    btree_node_read_in_flight(b)) {
-		if (!flush)
+	if (btree_node_dirty(b)) {
+		if (!flush ||
+		    test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags))
 			goto out_unlock;
-
-		wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
-			       TASK_UNINTERRUPTIBLE);
-
 		/*
 		 * Using the underscore version because we don't want to compact
 		 * bsets after the write, since this node is about to be evicted
@@ -224,8 +236,9 @@ static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush)
 		else
 			__bch2_btree_node_write(c, b);
 
-		/* wait for any in flight btree write */
-		btree_node_wait_on_io(b);
+		six_unlock_write(&b->c.lock);
+		six_unlock_intent(&b->c.lock);
+		goto wait_on_io;
 	}
 out:
 	if (b->hash_val && !ret)
@@ -581,6 +594,7 @@ struct btree *bch2_btree_node_mem_alloc(struct bch_fs *c)
 	}
 
 	BUG_ON(btree_node_hashed(b));
+	BUG_ON(btree_node_dirty(b));
 	BUG_ON(btree_node_write_in_flight(b));
 out:
 	b->flags		= 0;
@@ -634,6 +648,7 @@ static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
 {
 	struct btree_cache *bc = &c->btree_cache;
 	struct btree *b;
+	u32 seq;
 
 	BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
 	/*
@@ -663,31 +678,31 @@ static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
 		return NULL;
 	}
 
+	set_btree_node_read_in_flight(b);
+
+	six_unlock_write(&b->c.lock);
+	seq = b->c.lock.state.seq;
+	six_unlock_intent(&b->c.lock);
+
 	/* Unlock before doing IO: */
 	if (iter && sync)
 		bch2_trans_unlock(iter->trans);
 
 	bch2_btree_node_read(c, b, sync);
 
-	six_unlock_write(&b->c.lock);
-
-	if (!sync) {
-		six_unlock_intent(&b->c.lock);
+	if (!sync)
 		return NULL;
-	}
 
 	/*
 	 * XXX: this will probably always fail because btree_iter_relock()
 	 * currently fails for iterators that aren't pointed at a valid btree
 	 * node
 	 */
-	if (iter && !bch2_trans_relock(iter->trans)) {
-		six_unlock_intent(&b->c.lock);
+	if (iter && !bch2_trans_relock(iter->trans))
 		return ERR_PTR(-EINTR);
-	}
 
-	if (lock_type == SIX_LOCK_read)
-		six_lock_downgrade(&b->c.lock);
+	if (!six_relock_type(&b->c.lock, lock_type, seq))
+		return ERR_PTR(-EINTR);
 
 	return b;
 }
@@ -831,11 +846,12 @@ struct btree *bch2_btree_node_get(struct bch_fs *c, struct btree_iter *iter,
 	}
 
 	if (unlikely(btree_node_read_in_flight(b))) {
+		u32 seq = b->c.lock.state.seq;
+
 		six_unlock_type(&b->c.lock, lock_type);
 		bch2_trans_unlock(iter->trans);
 
-		wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
-			       TASK_UNINTERRUPTIBLE);
+		bch2_btree_node_wait_on_read(b);
 
 		/*
 		 * XXX: check if this always fails - btree_iter_relock()
@@ -844,7 +860,9 @@ struct btree *bch2_btree_node_get(struct bch_fs *c, struct btree_iter *iter,
 		 */
 		if (iter && !bch2_trans_relock(iter->trans))
 			return ERR_PTR(-EINTR);
-		goto retry;
+
+		if (!six_relock_type(&b->c.lock, lock_type, seq))
+			goto retry;
 	}
 
 	prefetch(b->aux_data);
@@ -923,8 +941,7 @@ struct btree *bch2_btree_node_get_noiter(struct bch_fs *c,
 	}
 
 	/* XXX: waiting on IO with btree locks held: */
-	wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
-		       TASK_UNINTERRUPTIBLE);
+	__bch2_btree_node_wait_on_read(b);
 
 	prefetch(b->aux_data);
 
@@ -979,16 +996,24 @@ void bch2_btree_node_evict(struct bch_fs *c, const struct bkey_i *k)
 	b = btree_cache_find(bc, k);
 	if (!b)
 		return;
+wait_on_io:
+	/* not allowed to wait on io with btree locks held: */
+
+	/* XXX we're called from btree_gc which will be holding other btree
+	 * nodes locked
+	 * */
+	__bch2_btree_node_wait_on_read(b);
+	__bch2_btree_node_wait_on_write(b);
 
 	six_lock_intent(&b->c.lock, NULL, NULL);
 	six_lock_write(&b->c.lock, NULL, NULL);
 
-	wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
-		       TASK_UNINTERRUPTIBLE);
-	__bch2_btree_node_write(c, b);
-
-	/* wait for any in flight btree write */
-	btree_node_wait_on_io(b);
+	if (btree_node_dirty(b)) {
+		__bch2_btree_node_write(c, b);
+		six_unlock_write(&b->c.lock);
+		six_unlock_intent(&b->c.lock);
+		goto wait_on_io;
+	}
 
 	BUG_ON(btree_node_dirty(b));
 

fs/bcachefs/btree_io.c

@@ -22,6 +22,44 @@
 
 #include <linux/sched/mm.h>
 
+void bch2_btree_node_io_unlock(struct btree *b)
+{
+	EBUG_ON(!btree_node_write_in_flight(b));
+
+	clear_btree_node_write_in_flight(b);
+	wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
+}
+
+void bch2_btree_node_io_lock(struct btree *b)
+{
+	wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
+			    TASK_UNINTERRUPTIBLE);
+}
+
+void __bch2_btree_node_wait_on_read(struct btree *b)
+{
+	wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
+		       TASK_UNINTERRUPTIBLE);
+}
+
+void __bch2_btree_node_wait_on_write(struct btree *b)
+{
+	wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
+		       TASK_UNINTERRUPTIBLE);
+}
+
+void bch2_btree_node_wait_on_read(struct btree *b)
+{
+	wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
+		       TASK_UNINTERRUPTIBLE);
+}
+
+void bch2_btree_node_wait_on_write(struct btree *b)
+{
+	wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
+		       TASK_UNINTERRUPTIBLE);
+}
+
 static void verify_no_dups(struct btree *b,
 			   struct bkey_packed *start,
 			   struct bkey_packed *end)
@@ -432,7 +470,8 @@ void bch2_btree_init_next(struct btree_trans *trans,
 	EBUG_ON(iter && iter->l[b->c.level].b != b);
 	BUG_ON(bset_written(b, bset(b, &b->set[1])));
 
-	if (b->nsets == MAX_BSETS) {
+	if (b->nsets == MAX_BSETS &&
+	    !btree_node_write_in_flight(b)) {
 		unsigned log_u64s[] = {
 			ilog2(bset_u64s(&b->set[0])),
 			ilog2(bset_u64s(&b->set[1])),
@@ -1402,8 +1441,6 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *b,
 	btree_pos_to_text(&PBUF(buf), c, b);
 	trace_btree_read(c, b);
 
-	set_btree_node_read_in_flight(b);
-
 	if (bch2_verify_all_btree_replicas &&
 	    !btree_node_read_all_replicas(c, b, sync))
 		return;
@@ -1480,6 +1517,8 @@ int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
 	bkey_copy(&b->key, k);
 	BUG_ON(bch2_btree_node_hash_insert(&c->btree_cache, b, level, id));
 
+	set_btree_node_read_in_flight(b);
+
 	bch2_btree_node_read(c, b, true);
 
 	if (btree_node_read_error(b)) {
@@ -1525,7 +1564,7 @@ static void btree_node_write_done(struct bch_fs *c, struct btree *b)
 	struct btree_write *w = btree_prev_write(b);
 
 	bch2_btree_complete_write(c, b, w);
-	btree_node_io_unlock(b);
+	bch2_btree_node_io_unlock(b);
 }
 
 static void bch2_btree_node_write_error(struct bch_fs *c,
@@ -1707,6 +1746,8 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b)
 	bool validate_before_checksum = false;
 	void *data;
 
+	BUG_ON(btree_node_write_in_flight(b));
+
 	if (test_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags))
 		return;
 
@@ -1734,7 +1775,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b)
 			 * XXX waiting on btree writes with btree locks held -
 			 * this can deadlock, and we hit the write error path
 			 */
-			btree_node_wait_on_io(b);
+			bch2_btree_node_wait_on_write(b);
 			continue;
 		}
 

fs/bcachefs/btree_io.h

@@ -52,24 +52,12 @@ struct btree_write_bio {
 	struct bch_write_bio	wbio;
 };
 
-static inline void btree_node_io_unlock(struct btree *b)
-{
-	EBUG_ON(!btree_node_write_in_flight(b));
-	clear_btree_node_write_in_flight(b);
-	wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
-}
-
-static inline void btree_node_io_lock(struct btree *b)
-{
-	wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
-			    TASK_UNINTERRUPTIBLE);
-}
-
-static inline void btree_node_wait_on_io(struct btree *b)
-{
-	wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
-		       TASK_UNINTERRUPTIBLE);
-}
+void bch2_btree_node_io_unlock(struct btree *);
+void bch2_btree_node_io_lock(struct btree *);
+void __bch2_btree_node_wait_on_read(struct btree *);
+void __bch2_btree_node_wait_on_write(struct btree *);
+void bch2_btree_node_wait_on_read(struct btree *);
+void bch2_btree_node_wait_on_write(struct btree *);
 
 static inline bool btree_node_may_write(struct btree *b)
 {
@@ -169,7 +157,7 @@ static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 		}
 
 		six_unlock_type(&b->c.lock, lock_held);
-		btree_node_wait_on_io(b);
+		bch2_btree_node_wait_on_write(b);
 		btree_node_lock_type(c, b, lock_held);
 	}
 }

fs/bcachefs/btree_update_interior.c

@@ -567,7 +567,7 @@ static void btree_update_nodes_written(struct btree_update *as)
 		six_unlock_read(&old->c.lock);
 
 		if (seq == as->old_nodes_seq[i])
-			btree_node_wait_on_io(old);
+			bch2_btree_node_wait_on_write(old);
 	}
 
 	/*

fs/bcachefs/debug.c

@@ -133,7 +133,7 @@ void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 	if (c->opts.nochanges)
 		return;
 
-	btree_node_io_lock(b);
+	bch2_btree_node_io_lock(b);
 	mutex_lock(&c->verify_lock);
 
 	if (!c->verify_ondisk) {
@@ -176,7 +176,7 @@ void __bch2_btree_verify(struct bch_fs *c, struct btree *b)
 	}
 out:
 	mutex_unlock(&c->verify_lock);
-	btree_node_io_unlock(b);
+	bch2_btree_node_io_unlock(b);
 }
 
 #ifdef CONFIG_DEBUG_FS