Btrfs: remove residual code in delayed inode async helper
authorMiao Xie <miaox@cn.fujitsu.com>
Thu, 26 Dec 2013 05:07:01 +0000 (13:07 +0800)
committerChris Mason <clm@fb.com>
Tue, 28 Jan 2014 21:20:06 +0000 (13:20 -0800)
Before applying the patch
  commit de3cb945db4d8eb3b046dc7a5ea89a893372750c
  title: Btrfs: improve the delayed inode throttling

We need requeue the async work after the current work was done, it
introduced a deadlock problem. So we wrote the code that this patch
removes to avoid the above problem. But after applying the above
patch, the deadlock problem didn't exist. So we should remove that
fix code.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/delayed-inode.c

index 5841949386123efa052e1c090628690abff4c0ea..608efeb24aea68b97ce27a1431a005b2a3565a0e 100644 (file)
@@ -1300,33 +1300,6 @@ again:
        trans->block_rsv = &root->fs_info->delayed_block_rsv;
 
        __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
-       /*
-        * Maybe new delayed items have been inserted, so we need requeue
-        * the work. Besides that, we must dequeue the empty delayed nodes
-        * to avoid the race between delayed items balance and the worker.
-        * The race like this:
-        *      Task1                           Worker thread
-        *                                      count == 0, needn't requeue
-        *                                        also needn't insert the
-        *                                        delayed node into prepare
-        *                                        list again.
-        *      add lots of delayed items
-        *      queue the delayed node
-        *        already in the list,
-        *        and not in the prepare
-        *        list, it means the delayed
-        *        node is being dealt with
-        *        by the worker.
-        *      do delayed items balance
-        *        the delayed node is being
-        *        dealt with by the worker
-        *        now, just wait.
-        *                                      the worker goto idle.
-        * Task1 will sleep until the transaction is commited.
-        */
-       mutex_lock(&delayed_node->mutex);
-       btrfs_dequeue_delayed_node(root->fs_info->delayed_root, delayed_node);
-       mutex_unlock(&delayed_node->mutex);
 
        trans->block_rsv = block_rsv;
        btrfs_end_transaction_dmeta(trans, root);