From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Yan, Zheng" Subject: [PATCH] properly wait log writers during log sync Date: Wed, 14 Oct 2009 16:00:34 +0800 Message-ID: <4AD58522.2070707@oracle.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 To: linux-btrfs@vger.kernel.org, chris Mason Return-path: List-ID: A recently fsync optimization make btrfs_sync_log skip calling wait_for_writer in the single log writer case. This is incorrect since the writer count can also be increased by btrfs_pin_log. Signed-off-by: Yan Zheng --- diff -urp 1/fs/btrfs/tree-log.c 2/fs/btrfs/tree-log.c --- 1/fs/btrfs/tree-log.c 2009-10-14 09:31:46.631606303 +0800 +++ 2/fs/btrfs/tree-log.c 2009-10-14 15:56:15.095844476 +0800 @@ -1995,12 +1995,13 @@ int btrfs_sync_log(struct btrfs_trans_ha if (atomic_read(&root->log_commit[(index1 + 1) % 2])) wait_log_commit(trans, root, root->log_transid - 1); - while (root->log_multiple_pids) { + while (1) { unsigned long batch = root->log_batch; - mutex_unlock(&root->log_mutex); - schedule_timeout_uninterruptible(1); - mutex_lock(&root->log_mutex); - + if (root->log_multiple_pids) { + mutex_unlock(&root->log_mutex); + schedule_timeout_uninterruptible(1); + mutex_lock(&root->log_mutex); + } wait_for_writer(trans, root); if (batch == root->log_batch) break;