From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kara Subject: Re: mmotm 2010-11-09 - lockdep splat in ext4/quota code Date: Sat, 13 Nov 2010 00:33:43 +0100 Message-ID: <20101112233343.GA12153@quack.suse.cz> References: <201011100003.oAA03O3c015222@imap1.linux-foundation.org> <19688.1289503541@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andrew Morton , Theodore Ts'o , Andreas Dilger , Jan Kara , linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org To: Valdis.Kletnieks@vt.edu Return-path: Received: from cantor2.suse.de ([195.135.220.15]:42161 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751313Ab0KLXdr (ORCPT ); Fri, 12 Nov 2010 18:33:47 -0500 Content-Disposition: inline In-Reply-To: <19688.1289503541@localhost> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Thu 11-11-10 14:25:41, Valdis.Kletnieks@vt.edu wrote: > On Tue, 09 Nov 2010 15:31:15 PST, akpm@linux-foundation.org said: > > The mm-of-the-moment snapshot 2010-11-09-15-31 has been uploaded to > > > > http://userweb.kernel.org/~akpm/mmotm/ > > Still seeing this just after quotas get turned on ext4. First seen in > 2.6.36-mmotm1022, but problem has apparently been there since > 2.6.36-rc5-mmotm0922 (oldest kernel I have lying around). Thanks for the report. In fact, it is a false positive caused by the fact that EXT4(inode)->i_data_sem for quota file inode would need a special locking class (or subclass) but does not have one. I'll have a look into implementing that sometime next week hopefully. Honza > The last few relevant commands from rc.sysinit: > > /sbin/quotaon -aug > dd if=/dev/urandom of=/var/lib/random-seed count=1 bs=512 <- this triggers the trace > > [ 48.967833] ======================================================= > [ 48.968162] [ INFO: possible circular locking dependency detected ] > [ 48.968411] 2.6.37-rc1-mmotm1109 #1 > [ 48.968553] ------------------------------------------------------- > [ 48.968592] dd/3253 is trying to acquire lock: > [ 48.968592] (&s->s_dquot.dqio_mutex){+.+...}, at: [] dquot_commit+0x26/0xda > [ 48.968592] > [ 48.968592] but task is already holding lock: > [ 48.968592] (&s->s_dquot.dqptr_sem){++++..}, at: [] __dquot_free_space+0x66/0x16b > [ 48.968592] > [ 48.968592] which lock already depends on the new lock. > [ 48.968592] > [ 48.968592] > [ 48.968592] the existing dependency chain (in reverse order) is: > [ 48.968592] > [ 48.968592] -> #2 (&s->s_dquot.dqptr_sem){++++..}: > [ 48.968592] [] lock_acquire+0x100/0x126 > [ 48.968592] [] down_read+0x42/0x51 > [ 48.968592] [] __dquot_free_space+0x66/0x16b > [ 48.968592] [] dquot_free_block+0x19/0x2a > [ 48.968592] [] ext4_free_blocks+0x75b/0x810 > [ 48.968592] [] ext4_ext_truncate+0x3af/0x853 > [ 48.968592] [] ext4_truncate+0x98/0x51a > [ 48.968592] [] vmtruncate+0x3f/0x4b > [ 48.968592] [] ext4_setattr+0x305/0x3a7 > [ 48.968592] [] notify_change+0x1a1/0x2a1 > [ 48.968592] [] do_truncate+0x67/0x84 > [ 48.968592] [] do_last+0x4ce/0x5b2 > [ 48.968592] [] do_filp_open+0x248/0x64a > [ 48.968592] [] do_sys_open+0x60/0xfb > [ 48.968592] [] sys_open+0x1b/0x1d > [ 48.968592] [] system_call_fastpath+0x16/0x1b > [ 48.968592] > [ 48.968592] -> #1 (&ei->i_data_sem){++++..}: > [ 48.968592] [] lock_acquire+0x100/0x126 > [ 48.968592] [] down_read+0x42/0x51 > [ 48.968592] [] ext4_map_blocks+0x3e/0x20e > [ 48.968592] [] ext4_getblk+0x71/0x19b > [ 48.968592] [] ext4_bread+0x11/0x61 > [ 48.968592] [] ext4_quota_read+0x86/0xf0 > [ 48.968592] [] v2_read_header+0x1e/0x4f > [ 48.968592] [] v2_read_file_info+0x20/0x2ae > [ 48.968592] [] vfs_load_quota_inode+0x292/0x448 > [ 48.968592] [] dquot_quota_on_path+0x4c/0x55 > [ 48.968592] [] ext4_quota_on+0x123/0x175 > [ 48.968592] [] do_quotactl+0x1d1/0x465 > [ 48.968592] [] sys_quotactl+0xfd/0x11a > [ 48.968592] [] system_call_fastpath+0x16/0x1b > [ 48.968592] > [ 48.968592] -> #0 (&s->s_dquot.dqio_mutex){+.+...}: > [ 48.968592] [] __lock_acquire+0xa4c/0xd4e > [ 48.968592] [] lock_acquire+0x100/0x126 > [ 48.968592] [] __mutex_lock_common+0x5d/0x56c > [ 48.968592] [] mutex_lock_nested+0x34/0x39 > [ 48.968592] [] dquot_commit+0x26/0xda > [ 48.968592] [] ext4_write_dquot+0x65/0x87 > [ 48.968592] [] ext4_mark_dquot_dirty+0x3e/0x49 > [ 48.968592] [] mark_all_dquot_dirty+0x22/0x48 > [ 48.968592] [] __dquot_free_space+0x139/0x16b > [ 48.968592] [] dquot_free_block+0x19/0x2a > [ 48.968592] [] ext4_free_blocks+0x75b/0x810 > [ 48.968592] [] ext4_ext_truncate+0x3af/0x853 > [ 48.968592] [] ext4_truncate+0x98/0x51a > [ 48.968592] [] vmtruncate+0x3f/0x4b > [ 48.968592] [] ext4_setattr+0x305/0x3a7 > [ 48.968592] [] notify_change+0x1a1/0x2a1 > [ 48.968592] [] do_truncate+0x67/0x84 > [ 48.968592] [] do_last+0x4ce/0x5b2 > [ 48.968592] [] do_filp_open+0x248/0x64a > [ 48.968592] [] do_sys_open+0x60/0xfb > [ 48.968592] [] sys_open+0x1b/0x1d > [ 48.968592] [] system_call_fastpath+0x16/0x1b > [ 48.968592] > [ 48.968592] other info that might help us debug this: > [ 48.968592] > [ 49.458584] 5 locks held by dd/3253: > [ 49.458584] #0: (&sb->s_type->i_mutex_key#13){+.+.+.}, at: [] do_truncate+0x5b/0x84 > [ 49.458584] #1: (&sb->s_type->i_alloc_sem_key#5){+.+...}, at: [] notify_change+0x189/0x2a1 > [ 49.458584] #2: (jbd2_handle){+.+...}, at: [] start_this_handle+0x57a/0x605 > [ 49.458584] #3: (&ei->i_data_sem){++++..}, at: [] ext4_ext_truncate+0x91/0x853 > [ 49.458584] #4: (&s->s_dquot.dqptr_sem){++++..}, at: [] __dquot_free_space+0x66/0x16b > [ 49.458584] > [ 49.458584] stack backtrace: > [ 49.458584] Pid: 3253, comm: dd Not tainted 2.6.37-rc1-mmotm1109 #1 > [ 49.458584] Call Trace: > [ 49.458584] [] print_circular_bug+0xa8/0xb7 > [ 49.458584] [] __lock_acquire+0xa4c/0xd4e > [ 49.458584] [] ? dquot_commit+0x26/0xda > [ 49.458584] [] lock_acquire+0x100/0x126 > [ 49.458584] [] ? dquot_commit+0x26/0xda > [ 49.458584] [] __mutex_lock_common+0x5d/0x56c > [ 49.458584] [] ? dquot_commit+0x26/0xda > [ 49.458584] [] ? dquot_mark_dquot_dirty+0x27/0x7a > [ 49.458584] [] ? dquot_commit+0x26/0xda > [ 49.458584] [] mutex_lock_nested+0x34/0x39 > [ 49.458584] [] dquot_commit+0x26/0xda > [ 49.458584] [] ext4_write_dquot+0x65/0x87 > [ 49.458584] [] ext4_mark_dquot_dirty+0x3e/0x49 > [ 49.458584] [] mark_all_dquot_dirty+0x22/0x48 > [ 49.458584] [] __dquot_free_space+0x139/0x16b > [ 49.458584] [] dquot_free_block+0x19/0x2a > [ 49.458584] [] ext4_free_blocks+0x75b/0x810 > [ 49.458584] [] ? jbd2_journal_extend+0x5d/0xb8 > [ 49.458584] [] ext4_ext_truncate+0x3af/0x853 > [ 49.458584] [] ? trace_preempt_on+0x15/0x28 > [ 49.458584] [] ? unmap_mapping_range+0x279/0x288 > [ 49.458584] [] ext4_truncate+0x98/0x51a > [ 49.458584] [] ? unmap_mapping_range+0x279/0x288 > [ 49.458584] [] ? trace_hardirqs_on_caller+0x117/0x13b > [ 49.458584] [] ? trace_hardirqs_on+0xd/0xf > [ 49.458584] [] vmtruncate+0x3f/0x4b > [ 49.458584] [] ext4_setattr+0x305/0x3a7 > [ 49.458584] [] notify_change+0x1a1/0x2a1 > [ 49.458584] [] do_truncate+0x67/0x84 > [ 49.458584] [] ? get_write_access+0x49/0x50 > [ 49.458584] [] ? _raw_spin_unlock+0x30/0x69 > [ 49.458584] [] ? sub_preempt_count+0x35/0x48 > [ 49.458584] [] ? security_path_truncate+0x48/0x4d > [ 49.458584] [] do_last+0x4ce/0x5b2 > [ 49.458584] [] do_filp_open+0x248/0x64a > [ 49.458584] [] ? trace_preempt_on+0x15/0x28 > [ 49.458584] [] ? alloc_fd+0x17c/0x18e > [ 49.458584] [] ? _raw_spin_unlock+0x30/0x69 > [ 49.458584] [] ? sub_preempt_count+0x35/0x48 > [ 49.458584] [] ? alloc_fd+0x17c/0x18e > [ 49.458584] [] do_sys_open+0x60/0xfb > [ 49.458584] [] ? trace_hardirqs_on_thunk+0x3a/0x3f > [ 49.458584] [] sys_open+0x1b/0x1d > [ 49.458584] [] system_call_fastpath+0x16/0x1b > > -- Jan Kara SUSE Labs, CR