bcache: rework some -ENOSPC stuff

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

7 files changed, 44 insertions, 25 deletions

diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index b6f8046ac8e1..1c69d9457d69 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -1614,18 +1614,24 @@ struct btree *btree_node_alloc_replacement(struct btree *b)
 static int __btree_check_reserve(struct cache_set *c,
 				 enum alloc_reserve reserve,
 				 unsigned required,
-				 struct closure *cl)
+				 struct closure *cl,
+				 bool check_enospc)
 {
+	struct cache_group *devs = &c->cache_tiers[0];
 	struct cache *ca;
 	unsigned i;
 	int ret;
 
-	rcu_read_lock();
+	/*
+	 * XXX: racy... the whole btree node reserve thing needs to be
+	 * completely reworked
+	 */
+	if (!devs->nr_devices)
+		return -ENOSPC;
 
-	for_each_cache_rcu(ca, c, i) {
-		if (CACHE_STATE(&ca->mi) != CACHE_ACTIVE)
-			continue;
+	rcu_read_lock();
 
+	group_for_each_cache_rcu(ca, devs, i) {
 		spin_lock(&ca->freelist_lock);
 
 		if (fifo_used(&ca->free[reserve]) < required) {
@@ -1633,11 +1639,12 @@ static int __btree_check_reserve(struct cache_set *c,
 					fifo_used(&ca->free[reserve]),
 					required, cl);
 
-			if (cl) {
+			if (!cl ||
+			    (check_enospc && cache_set_full(c))) {
+				ret = -ENOSPC;
+			} else {
 				closure_wait(&c->freelist_wait, cl);
 				ret = -EAGAIN;
-			} else {
-				ret = -ENOSPC;
 			}
 
 			spin_unlock(&ca->freelist_lock);
@@ -1655,13 +1662,13 @@ static int __btree_check_reserve(struct cache_set *c,
 
 int btree_check_reserve(struct btree *b, struct btree_iter *iter,
 			enum alloc_reserve reserve,
-			unsigned extra_nodes)
+			unsigned extra_nodes, bool check_enospc)
 {
 	unsigned depth = btree_node_root(b)->level - b->level;
 
 	return __btree_check_reserve(b->c, reserve,
 			btree_reserve_required_nodes(depth) + extra_nodes,
-			iter ? &iter->cl : NULL);
+			iter ? &iter->cl : NULL, check_enospc);
 }
 
 static struct btree *__btree_root_alloc(struct cache_set *c, unsigned level,
@@ -1684,12 +1691,16 @@ int bch_btree_root_alloc(struct cache_set *c, enum btree_id id,
 {
 	struct closure cl;
 	struct btree *b;
+	int ret;
 
 	closure_init_stack(&cl);
 
-	while (__btree_check_reserve(c, id, 1, &cl))
+	while ((ret = __btree_check_reserve(c, id, 1, &cl, true)) == -EAGAIN)
 		closure_sync(&cl);
 
+	if (ret == -ENOSPC)
+		return ret;
+
 	b = __btree_root_alloc(c, 0, id);
 
 	bch_btree_node_write(b, writes, NULL);
@@ -1746,7 +1757,8 @@ int bch_btree_node_rewrite(struct btree *b, struct btree_iter *iter, bool wait)
 	if (!bch_btree_iter_upgrade(iter))
 		return -EINTR;
 
-	ret = btree_check_reserve(b, wait ? iter : NULL, iter->btree_id, 1);
+	ret = btree_check_reserve(b, wait ? iter : NULL,
+				  iter->btree_id, 1, true);
 	if (ret) {
 		trace_bcache_btree_gc_rewrite_node_fail(b);
 		return ret;
@@ -2073,7 +2085,8 @@ static int btree_split(struct btree *b,
 	BUG_ON(!btree_node_intent_locked(iter, btree_node_root(b)->level));
 
 	/* After this check we cannot return -EAGAIN anymore */
-	ret = btree_check_reserve(b, iter, iter->btree_id, 0);
+	ret = btree_check_reserve(b, iter, iter->btree_id, 0,
+				  !(flags & BTREE_INSERT_NOFAIL));
 	if (ret) {
 		/* If splitting an interior node, we've already split a leaf,
 		 * so we should have checked for sufficient reserve. We can't
@@ -2467,7 +2480,7 @@ int bch_btree_insert_check_key(struct btree_iter *iter,
  */
 int bch_btree_insert(struct cache_set *c, enum btree_id id,
 		     struct keylist *keys, struct bch_replace_info *replace,
-		     struct closure *persistent, u64 *journal_seq)
+		     struct closure *persistent, u64 *journal_seq, int flags)
 {
 	struct btree_iter iter;
 	int ret, ret2;
@@ -2480,7 +2493,7 @@ int bch_btree_insert(struct cache_set *c, enum btree_id id,
 		goto out;
 
 	ret = bch_btree_insert_at(&iter, keys, replace,
-				  persistent, journal_seq, 0);
+				  persistent, journal_seq, flags);
 out:	ret2 = bch_btree_iter_unlock(&iter);
 
 	return ret ?: ret2;
diff --git a/drivers/md/bcache/btree.h b/drivers/md/bcache/btree.h
index e777ec4bdb8e..557c2f77fa29 100644
--- a/drivers/md/bcache/btree.h
+++ b/drivers/md/bcache/btree.h
@@ -434,7 +434,7 @@ struct btree *__btree_node_alloc_replacement(struct btree *,
 					     struct bkey_format);
 struct btree *btree_node_alloc_replacement(struct btree *);
 int btree_check_reserve(struct btree *, struct btree_iter *,
-			enum alloc_reserve, unsigned);
+			enum alloc_reserve, unsigned, bool);
 
 int bch_btree_root_alloc(struct cache_set *, enum btree_id, struct closure *);
 int bch_btree_root_read(struct cache_set *, enum btree_id,
@@ -455,7 +455,10 @@ int bch_btree_insert_node(struct btree *, struct btree_iter *,
  * Don't drop/retake locks: instead return -EINTR if need to upgrade to intent
  * locks, -EAGAIN if need to wait on btree reserve
  */
-#define BTREE_INSERT_ATOMIC		1
+#define BTREE_INSERT_ATOMIC		(1 << 0)
+
+/* Don't check for -ENOSPC: */
+#define BTREE_INSERT_NOFAIL		(1 << 1)
 
 /*
  * Fail a btree insert if dirty stale pointers are being added
@@ -463,14 +466,15 @@ int bch_btree_insert_node(struct btree *, struct btree_iter *,
  * Needs to be set for compare exchange and device removal, and not
  * set for journal replay. See big comment in bch_insert_fixup_extent()
  */
-#define FAIL_IF_STALE			2
+#define FAIL_IF_STALE			(1 << 2)
 
 int bch_btree_insert_at(struct btree_iter *, struct keylist *,
 			struct bch_replace_info *, struct closure *,
 			u64 *, unsigned);
 int bch_btree_insert_check_key(struct btree_iter *, struct bkey_i *);
 int bch_btree_insert(struct cache_set *, enum btree_id, struct keylist *,
-		     struct bch_replace_info *, struct closure *, u64 *);
+		     struct bch_replace_info *, struct closure *,
+		     u64 *, int flags);
 int bch_btree_update(struct cache_set *, enum btree_id, struct bkey_i *,
 		     struct closure *, u64 *);
 
diff --git a/drivers/md/bcache/gc.c b/drivers/md/bcache/gc.c
index 4668cd7445a4..db433b50ec63 100644
--- a/drivers/md/bcache/gc.c
+++ b/drivers/md/bcache/gc.c
@@ -420,7 +420,8 @@ static void bch_coalesce_nodes(struct btree *old_nodes[GC_MERGE_NODES],
 			 block_bytes(c)) > blocks)
 		return;
 
-	if (btree_check_reserve(parent, NULL, iter->btree_id, nr_old_nodes) ||
+	if (btree_check_reserve(parent, NULL, iter->btree_id,
+				nr_old_nodes, false) ||
 	    bch_keylist_realloc(&keylist,
 			(BKEY_U64s + BKEY_EXTENT_MAX_U64s) * nr_old_nodes)) {
 		trace_bcache_btree_gc_coalesce_fail(c);
diff --git a/drivers/md/bcache/inode.c b/drivers/md/bcache/inode.c
index 09b21f509013..f54c89428574 100644
--- a/drivers/md/bcache/inode.c
+++ b/drivers/md/bcache/inode.c
@@ -175,7 +175,7 @@ int bch_inode_rm(struct cache_set *c, u64 inode_nr)
 
 	return bch_btree_insert(c, BTREE_ID_INODES,
 				&keylist_single(&delete),
-				NULL, NULL, NULL);
+				NULL, NULL, NULL, 0);
 }
 
 int bch_blockdev_inode_find_by_uuid(struct cache_set *c, uuid_le *uuid,
diff --git a/drivers/md/bcache/io.c b/drivers/md/bcache/io.c
index 9b638562a912..2cb909fa60ce 100644
--- a/drivers/md/bcache/io.c
+++ b/drivers/md/bcache/io.c
@@ -322,7 +322,8 @@ static void bch_write_index(struct closure *cl)
 	int ret = bch_btree_insert(op->c, BTREE_ID_EXTENTS, &op->insert_keys,
 				   op->replace ? &op->replace_info : NULL,
 				   op->flush ? &op->cl : NULL,
-				   op->journal_seq);
+				   op->journal_seq,
+				   op->check_enospc ? 0 : BTREE_INSERT_NOFAIL);
 	if (ret) {
 		__bcache_io_error(op->c, "btree IO error");
 		op->error = ret;
diff --git a/drivers/md/bcache/journal.c b/drivers/md/bcache/journal.c
index 9df11b1594e9..c616eb4caa59 100644
--- a/drivers/md/bcache/journal.c
+++ b/drivers/md/bcache/journal.c
@@ -843,7 +843,7 @@ static int bch_journal_replay_key(struct cache_set *c, enum btree_id id,
 	if (do_subtract)
 		bkey_copy(&temp.key, k);
 
-	ret = bch_btree_insert(c, id, &keylist_single(k), NULL, NULL, NULL);
+	ret = bch_btree_insert(c, id, &keylist_single(k), NULL, NULL, NULL, 0);
 	if (ret)
 		return ret;
 
diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c
index 4f42adc75e9a..8e3854108735 100644
--- a/drivers/md/bcache/writeback.c
+++ b/drivers/md/bcache/writeback.c
@@ -119,8 +119,8 @@ static void write_dirty_finish(struct closure *cl)
 		SET_EXTENT_CACHED(&bkey_i_to_extent(keys.top)->v, true);
 		bch_keylist_enqueue(&keys);
 
-		ret = bch_btree_insert(dc->disk.c, BTREE_ID_EXTENTS,
-				       &keys, &io->replace, NULL, NULL);
+		ret = bch_btree_insert(dc->disk.c, BTREE_ID_EXTENTS, &keys,
+				       &io->replace, NULL, NULL, 0);
 		if (io->replace.successes == 0)
 			trace_bcache_writeback_collision(&io->replace.key.k);