btrfs: fix refcount_t usage when deleting btrfs_delayed_nodes

refcounts have a generic implementation and an asm optimized one.  The
generic version has extra debugging to make sure that once a refcount
goes to zero, refcount_inc won't increase it.

The btrfs delayed inode code wasn't expecting this, and we're tripping
over the warnings when the generic refcounts are used.  We ended up with
this race:

Process A                                         Process B
                                                  btrfs_get_delayed_node()
						  spin_lock(root->inode_lock)
						  radix_tree_lookup()
__btrfs_release_delayed_node()
refcount_dec_and_test(&delayed_node->refs)
our refcount is now zero
						  refcount_add(2) <---
						  warning here, refcount
                                                  unchanged

spin_lock(root->inode_lock)
radix_tree_delete()

With the generic refcounts, we actually warn again when process B above
tries to release his refcount because refcount_add() turned into a
no-op.

We saw this in production on older kernels without the asm optimized
refcounts.

The fix used here is to use refcount_inc_not_zero() to detect when the
object is in the middle of being freed and return NULL.  This is almost
always the right answer anyway, since we usually end up pitching the
delayed_node if it didn't have fresh data in it.

This also changes __btrfs_release_delayed_node() to remove the extra
check for zero refcounts before radix tree deletion.
btrfs_get_delayed_node() was the only path that was allowing refcounts
to go from zero to one.

Fixes: 6de5f18e ("btrfs: fix refcount_t usage when deleting btrfs_delayed_node")
CC: <stable@vger.kernel.org> # 4.12+
Signed-off-by: Chris Mason <clm@fb.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/delayed-inode.c

@@ -87,6 +87,7 @@ static struct btrfs_delayed_node *btrfs_get_delayed_node(
 
 	spin_lock(&root->inode_lock);
 	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
+
 	if (node) {
 		if (btrfs_inode->delayed_node) {
 			refcount_inc(&node->refs);	/* can be accessed */
@@ -94,9 +95,30 @@ static struct btrfs_delayed_node *btrfs_get_delayed_node(
 			spin_unlock(&root->inode_lock);
 			return node;
 		}
+
+		/*
+		 * It's possible that we're racing into the middle of removing
+		 * this node from the radix tree.  In this case, the refcount
+		 * was zero and it should never go back to one.  Just return
+		 * NULL like it was never in the radix at all; our release
+		 * function is in the process of removing it.
+		 *
+		 * Some implementations of refcount_inc refuse to bump the
+		 * refcount once it has hit zero.  If we don't do this dance
+		 * here, refcount_inc() may decide to just WARN_ONCE() instead
+		 * of actually bumping the refcount.
+		 *
+		 * If this node is properly in the radix, we want to bump the
+		 * refcount twice, once for the inode and once for this get
+		 * operation.
+		 */
+		if (refcount_inc_not_zero(&node->refs)) {
+			refcount_inc(&node->refs);
 			btrfs_inode->delayed_node = node;
-		/* can be accessed and cached in the inode */
-		refcount_add(2, &node->refs);
+		} else {
+			node = NULL;
+		}
+
 		spin_unlock(&root->inode_lock);
 		return node;
 	}
@@ -254,16 +276,17 @@ static void __btrfs_release_delayed_node(
 	mutex_unlock(&delayed_node->mutex);
 
 	if (refcount_dec_and_test(&delayed_node->refs)) {
-		bool free = false;
 		struct btrfs_root *root = delayed_node->root;
+
 		spin_lock(&root->inode_lock);
-		if (refcount_read(&delayed_node->refs) == 0) {
+		/*
+		 * Once our refcount goes to zero, nobody is allowed to bump it
+		 * back up.  We can delete it now.
+		 */
+		ASSERT(refcount_read(&delayed_node->refs) == 0);
 		radix_tree_delete(&root->delayed_nodes_tree,
 				  delayed_node->inode_id);
-			free = true;
-		}
 		spin_unlock(&root->inode_lock);
-		if (free)
 		kmem_cache_free(delayed_node_cache, delayed_node);
 	}
 }