Btrfs: make sure the async caching thread advances the key

The async caching thread can end up looping forever if a given
search puts it at the last key in a leaf.  It will end up calling
btrfs_next_leaf and then checking if it needs to politely drop
the read semaphore.

Most of the time this looping isn't noticed because it is able to
make progress the next time around.  But, during log replay,
we wait on the async caching thread to finish, and the async thread
is waiting on the commit, and no progress is really made.

The fix used here is to copy the key out of the next leaf,
that way our search lands there properly.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/extent-tree.c

@@ -265,10 +265,6 @@ static int caching_kthread(void *data)
 
 	atomic_inc(&block_group->space_info->caching_threads);
 	last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
-again:
-	/* need to make sure the commit_root doesn't disappear */
-	down_read(&fs_info->extent_commit_sem);
-
 	/*
 	 * We don't want to deadlock with somebody trying to allocate a new
 	 * extent for the extent root while also trying to search the extent
@@ -282,6 +278,10 @@ again:
 	key.objectid = last;
 	key.offset = 0;
 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+again:
+	/* need to make sure the commit_root doesn't disappear */
+	down_read(&fs_info->extent_commit_sem);
+
 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
 	if (ret < 0)
 		goto err;
@@ -304,6 +304,19 @@ again:
 
 			if (need_resched() ||
 			    btrfs_transaction_in_commit(fs_info)) {
+				leaf = path->nodes[0];
+
+				/* this shouldn't happen, but if the
+				 * leaf is empty just move on.
+				 */
+				if (btrfs_header_nritems(leaf) == 0)
+					break;
+				/*
+				 * we need to copy the key out so that
+				 * we are sure the next search advances
+				 * us forward in the btree.
+				 */
+				btrfs_item_key_to_cpu(leaf, &key, 0);
 				btrfs_release_path(fs_info->extent_root, path);
 				up_read(&fs_info->extent_commit_sem);
 				schedule_timeout(1);