Re: [BUG] btrfs dev del causes 'possible circular locking dependency detected'
From: Stefan Behrens <hidden>
Date: 2012-08-09 16:44:21
On Thu, 9 Aug 2012 18:01:57 +0200, David Sterba wrote:
On Thu, Aug 09, 2012 at 05:36:54PM +0200, Stefan Behrens wrote:quoted
Aug 9 16:02:21 qvarne kernel: [ 543.479460] -> #2 (&fs_info->cleaner_mutex){+.+...}: Aug 9 16:02:21 qvarne kernel: [ 543.541341] [<ffffffff810da005>] lock_acquire+0x95/0x150 Aug 9 16:02:21 qvarne kernel: [ 543.612269] [<ffffffff819259a1>] mutex_lock_nested+0x71/0x360 Aug 9 16:02:21 qvarne kernel: [ 543.688396] [<ffffffffa00da06f>] btrfs_commit_super+0x1f/0x100 [btrfs] Aug 9 16:02:21 qvarne kernel: [ 543.773884] [<ffffffffa00db8a8>] close_ctree+0x318/0x360 [btrfs] Aug 9 16:02:21 qvarne kernel: [ 543.853132] [<ffffffffa00b3da4>] btrfs_put_super+0x14/0x20 [btrfs] Aug 9 16:02:21 qvarne kernel: [ 543.934459] [<ffffffff8118ce6d>] generic_shutdown_super+0x5d/0xf0 Aug 9 16:02:21 qvarne kernel: [ 544.014747] [<ffffffff8118cf91>] kill_anon_super+0x11/0x20 Aug 9 16:02:21 qvarne kernel: [ 544.087756] [<ffffffffa00b7405>] btrfs_kill_super+0x15/0x90 [btrfs] Aug 9 16:02:21 qvarne kernel: [ 544.170124] [<ffffffff8118d1bd>] deactivate_locked_super+0x3d/0x70 Aug 9 16:02:21 qvarne kernel: [ 544.251452] [<ffffffff8118dcf9>] deactivate_super+0x49/0x70 Aug 9 16:02:21 qvarne kernel: [ 544.325500] [<ffffffff811a8eae>] mntput_no_expire+0xfe/0x160 Aug 9 16:02:21 qvarne kernel: [ 544.400588] [<ffffffff811a9c69>] sys_umount+0x69/0x3a0 Aug 9 16:02:21 qvarne kernel: [ 544.469435] [<ffffffff81931162>] system_call_fastpath+0x16/0x1b Aug 9 16:02:21 qvarne kernel: [ 544.547644] Aug 9 16:02:21 qvarne kernel: [ 544.547644] -> #1 (&type->s_umount_key#31){+++++.}: Aug 9 16:02:21 qvarne kernel: [ 544.608588] [<ffffffff810da005>] lock_acquire+0x95/0x150 Aug 9 16:02:21 qvarne kernel: [ 544.679516] [<ffffffff81926ca7>] down_read+0x47/0x8e Aug 9 16:02:21 qvarne kernel: [ 544.746283] [<ffffffff811b3ae8>] writeback_inodes_sb_nr_if_idle+0x38/0x60this matches the "umount vs cleaner vs writeback_idle" deadlock pattern, Miao sent patches some time ago, I've tested them succesfully, but they're not merged upstream. http://thread.gmane.org/gmane.linux.file-systems/63900 vfs: re-implement writeback_inodes_sb(_nr)_if_idle() and rename them http://thread.gmane.org/gmane.comp.file-systems.btrfs/17040 [PATCH 2/2] Btrfs: flush all the dirty pages if try_to_writeback_inodes_sb_nr() fails
After applying these two patches from Miao Xie, logcheck does not complain anymore. Could we start a second attempt to add these VFS, ext4 and btrfs changes?