Btrfs: fix joining same transaction handle more than twice

Wang Shilong · masoncl · commit f017f15f7c18 · 2014-06-09T17:20:46.000-07:00
We hit something like the following function call flows:

|-&gt;run_delalloc_range()
 |-&gt;btrfs_join_transaction()
   |-&gt;cow_file_range()
     |-&gt;btrfs_join_transaction()
       |-&gt;find_free_extent()
         |-&gt;btrfs_join_transaction()

Trace infomation can be seen as:

[ 7411.127040] ------------[ cut here ]------------
[ 7411.127060] WARNING: CPU: 0 PID: 11557 at fs/btrfs/transaction.c:383 start_transaction+0x561/0x580 [btrfs]()
[ 7411.127079] CPU: 0 PID: 11557 Comm: kworker/u8:9 Tainted: G           O 3.13.0+ #4
[ 7411.127080] Hardware name: LENOVO QiTianM4350/ , BIOS F1KT52AUS 05/24/2013
[ 7411.127085] Workqueue: writeback bdi_writeback_workfn (flush-btrfs-5)
[ 7411.127092] Call Trace:
[ 7411.127097]  [&lt;ffffffff815b87b0&gt;] dump_stack+0x45/0x56
[ 7411.127101]  [&lt;ffffffff81051ffd&gt;] warn_slowpath_common+0x7d/0xa0
[ 7411.127102]  [&lt;ffffffff810520da&gt;] warn_slowpath_null+0x1a/0x20
[ 7411.127109]  [&lt;ffffffffa0444fb1&gt;] start_transaction+0x561/0x580 [btrfs]
[ 7411.127115]  [&lt;ffffffffa0445027&gt;] btrfs_join_transaction+0x17/0x20 [btrfs]
[ 7411.127120]  [&lt;ffffffffa0431c91&gt;] find_free_extent+0xa21/0xb50 [btrfs]
[ 7411.127126]  [&lt;ffffffffa0431f68&gt;] btrfs_reserve_extent+0xa8/0x1a0 [btrfs]
[ 7411.127131]  [&lt;ffffffffa04322ce&gt;] btrfs_alloc_free_block+0xee/0x440 [btrfs]
[ 7411.127137]  [&lt;ffffffffa043bd6e&gt;] ? btree_set_page_dirty+0xe/0x10 [btrfs]
[ 7411.127142]  [&lt;ffffffffa041da51&gt;] __btrfs_cow_block+0x121/0x530 [btrfs]
[ 7411.127146]  [&lt;ffffffffa041dfff&gt;] btrfs_cow_block+0x11f/0x1c0 [btrfs]
[ 7411.127151]  [&lt;ffffffffa0421b74&gt;] btrfs_search_slot+0x1d4/0x9c0 [btrfs]
[ 7411.127157]  [&lt;ffffffffa0438567&gt;] btrfs_lookup_file_extent+0x37/0x40 [btrfs]
[ 7411.127163]  [&lt;ffffffffa0456bfc&gt;] __btrfs_drop_extents+0x16c/0xd90 [btrfs]
[ 7411.127169]  [&lt;ffffffffa0444ae3&gt;] ? start_transaction+0x93/0x580 [btrfs]
[ 7411.127171]  [&lt;ffffffff811663e2&gt;] ? kmem_cache_alloc+0x132/0x140
[ 7411.127176]  [&lt;ffffffffa041cd9a&gt;] ? btrfs_alloc_path+0x1a/0x20 [btrfs]
[ 7411.127182]  [&lt;ffffffffa044aa61&gt;] cow_file_range_inline+0x181/0x2e0 [btrfs]
[ 7411.127187]  [&lt;ffffffffa044aead&gt;] cow_file_range+0x2ed/0x440 [btrfs]
[ 7411.127194]  [&lt;ffffffffa0464d7f&gt;] ? free_extent_buffer+0x4f/0xb0 [btrfs]
[ 7411.127200]  [&lt;ffffffffa044b38f&gt;] run_delalloc_nocow+0x38f/0xa60 [btrfs]
[ 7411.127207]  [&lt;ffffffffa0461600&gt;] ? test_range_bit+0x30/0x180 [btrfs]
[ 7411.127212]  [&lt;ffffffffa044bd48&gt;] run_delalloc_range+0x2e8/0x350 [btrfs]
[ 7411.127219]  [&lt;ffffffffa04618f9&gt;] ? find_lock_delalloc_range+0x1a9/0x1e0 [btrfs]
[ 7411.127222]  [&lt;ffffffff812a1e71&gt;] ? blk_queue_bio+0x2c1/0x330
[ 7411.127228]  [&lt;ffffffffa0462ad4&gt;] __extent_writepage+0x2f4/0x760 [btrfs]

Here we fix it by avoiding joining transaction again if we have held
a transaction handle when allocating chunk in find_free_extent().

Signed-off-by: Wang Shilong &lt;wangsl.fnst@cn.fujitsu.com&gt;
Signed-off-by: Chris Mason &lt;clm@fb.com&gt;
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
@@ -6616,8 +6616,14 @@ static noinline int find_free_extent(struct btrfs_root *orig_root,
 		loop++;
 		if (loop == LOOP_ALLOC_CHUNK) {
 			struct btrfs_trans_handle *trans;
+			int exist = 0;
+
+			trans = current->journal_info;
+			if (trans)
+				exist = 1;
+			else
+				trans = btrfs_join_transaction(root);
 
-			trans = btrfs_join_transaction(root);
 			if (IS_ERR(trans)) {
 				ret = PTR_ERR(trans);
 				goto out;
@@ -6634,7 +6640,8 @@ static noinline int find_free_extent(struct btrfs_root *orig_root,
 							root, ret);
 			else
 				ret = 0;
-			btrfs_end_transaction(trans, root);
+			if (!exist)
+				btrfs_end_transaction(trans, root);
 			if (ret)
 				goto out;
 		}
