On Fri, Oct 09, 2020 at 04:12:38PM -0400, Eric Whitney wrote:
> * Ritesh Harjani <riteshh@xxxxxxxxxxxxx>:
> > Hello All,
> >
> > While running generic/219 fstests on a 1k blocksize on x86 box, I see
> > below dmesg warning msg and generic/219 fails. I haven't yet analyzed
> > it, but I remember I have seen such warnings before as well in my testing.
> > I was wondering if others have also seen it in their testing or not and
> > if this is any known issue?
> >
> >
> > Here it goes, 219.dmesg file.
> >
> >
> > [14459.933253] run fstests generic/219 at 2020-10-08 20:03:27
> > [14462.947295] EXT4-fs (vdc): mounted filesystem with ordered data mode.
> > Opts: acl,user_xattr,block_validity,usrquota,grpquota
> > [14462.992869] EXT4-fs (vdc): re-mounted. Opts: (null)
> > [14463.017058] EXT4-fs (vdc): re-mounted. Opts: (null)
> > [14463.727095] EXT4-fs (vdc): mounted filesystem with ordered data mode.
> > Opts: usrquota
> > [14463.837130] EXT4-fs (vdc): re-mounted. Opts: (null)
> > [14463.874293] EXT4-fs (vdc): re-mounted. Opts: (null)
> >
> > [14464.149407] ======================================================
> > [14464.149407] WARNING: possible circular locking dependency detected
> > [14464.149407] 5.9.0-rc8-next-20201008 #16 Not tainted
> > [14464.149407] ------------------------------------------------------
> > [14464.149407] chown/23358 is trying to acquire lock:
> > [14464.149407] ffff8b9b8ff60be0 (&ei->i_data_sem){++++}-{3:3}, at:
> > ext4_map_blocks+0xd1/0x640
> > [14464.149407]
> > but task is already holding lock:
> > [14464.149407] ffff8b9b8963c208 (&s->s_dquot.dqio_sem){++++}-{3:3}, at:
> > v2_write_dquot+0x2d/0xb0
> > [14464.149407]
> > which lock already depends on the new lock.
> >
> > [14464.149407]
> > the existing dependency chain (in reverse order) is:
> > [14464.149407]
> > -> #2 (&s->s_dquot.dqio_sem){++++}-{3:3}:
> > [14464.149407] down_read+0x41/0x200
> > [14464.149407] v2_read_dquot+0x23/0x60
> > [14464.149407] dquot_acquire+0x4c/0x100
> > [14464.149407] ext4_acquire_dquot+0x72/0xc0
> > [14464.149407] dqget+0x24a/0x4a0
> > [14464.149407] __dquot_initialize+0x1f0/0x330
> > [14464.149407] ext4_create+0x38/0x190
> > [14464.149407] lookup_open+0x4cd/0x630
> > [14464.149407] path_openat+0x2c4/0xa10
> > [14464.149407] do_filp_open+0x91/0x100
> > [14464.149407] do_sys_openat2+0x20d/0x2d0
> > [14464.149407] do_sys_open+0x44/0x80
> > [14464.149407] do_syscall_64+0x33/0x40
> > [14464.149407] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > [14464.149407]
> > -> #1 (&dquot->dq_lock){+.+.}-{3:3}:
> > [14464.149407] __mutex_lock+0xaa/0x9c0
> > [14464.149407] dquot_commit+0x23/0xf0
> > [14464.149407] ext4_write_dquot+0x78/0xb0
> > [14464.149407] __dquot_alloc_space+0x151/0x310
> > [14464.149407] ext4_mb_new_blocks+0x1f6/0x12e0
> > [14464.149407] ext4_ext_map_blocks+0x9c3/0x1470
> > [14464.149407] ext4_map_blocks+0xf4/0x640
> > [14464.149407] _ext4_get_block+0x90/0x110
> > [14464.149407] ext4_block_write_begin+0x15f/0x5a0
> > [14464.149407] ext4_write_begin+0x267/0x5b0
> > [14464.149407] generic_perform_write+0xc2/0x1e0
> > [14464.149407] ext4_buffered_write_iter+0x8b/0x130
> > [14464.149407] ext4_file_write_iter+0x6c/0x6d0
> > [14464.149407] new_sync_write+0x122/0x1b0
> > [14464.149407] vfs_write+0x1ca/0x230
> > [14464.149407] ksys_pwrite64+0x68/0xa0
> > [14464.149407] do_syscall_64+0x33/0x40
> > [14464.149407] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > [14464.149407]
> > -> #0 (&ei->i_data_sem){++++}-{3:3}:
> > [14464.149407] __lock_acquire+0x147b/0x2830
> > [14464.149407] lock_acquire+0xc6/0x3a0
> > [14464.149407] down_write+0x40/0x110
> > [14464.149407] ext4_map_blocks+0xd1/0x640
> > [14464.149407] ext4_getblk+0x54/0x1c0
> > [14464.149407] ext4_bread+0x1f/0xd0
> > [14464.149407] ext4_quota_write+0xbf/0x280
> > [14464.149407] write_blk+0x35/0x70
> > [14464.149407] get_free_dqblk+0x42/0xa0
> > [14464.149407] do_insert_tree+0x1d6/0x480
> > [14464.149407] do_insert_tree+0x464/0x480
> > [14464.149407] do_insert_tree+0x218/0x480
> > [14464.149407] do_insert_tree+0x218/0x480
> > [14464.149407] qtree_write_dquot+0x79/0x1b0
> > [14464.149407] v2_write_dquot+0x52/0xb0
> > [14464.149407] dquot_acquire+0x8e/0x100
> > [14464.149407] ext4_acquire_dquot+0x72/0xc0
> > [14464.149407] dqget+0x24a/0x4a0
> > [14464.149407] dquot_transfer+0xfe/0x140
> > [14464.149407] ext4_setattr+0x134/0x9c0
> > [14464.149407] notify_change+0x34b/0x490
> > [14464.149407] chown_common+0x96/0x150
> > [14464.149407] do_fchownat+0x8d/0xe0
> > [14464.149407] __x64_sys_fchownat+0x21/0x30
> > [14464.149407] do_syscall_64+0x33/0x40
> > [14464.149407] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > [14464.149407]
> > other info that might help us debug this:
> >
> > [14464.149407] Chain exists of:
> > &ei->i_data_sem --> &dquot->dq_lock -->
> > &s->s_dquot.dqio_sem
> >
> > [14464.149407] Possible unsafe locking scenario:
> >
> > [14464.149407] CPU0 CPU1
> > [14464.149407] ---- ----
> > [14464.149407] lock(&s->s_dquot.dqio_sem);
> > [14464.149407] lock(&dquot->dq_lock);
> > [14464.149407] lock(&s->s_dquot.dqio_sem);
> > [14464.149407] lock(&ei->i_data_sem);
It's complaining that chown on a regular file takes i_data_sem, then
takes dq_lock to update the quotas, and then takes the i_data_sem of the
quota file. This /should/ be harmless as long as quota files themselves
are not accounted for in the quota file, but you could also shut up
lockdep by doing something like:
static struct lock_class_key ext4_dquot_lock_class;
and then when initializing the incore quota inodes:
lockdep_set_class(&inode->i_data_sem, &ext4_dquot_lock_class);
That way lockdep always knows that the quota inodes are "special". Even
better, I think this also means that lockdep actually /will/ call this
out specifically if someone screws up and tries to start a regular write
after taking the quota lock.
The downside is that now you have special inodes that are not like the
others.
--D
> > [14464.149407]
> > *** DEADLOCK ***
> >
> > [14464.149407] 6 locks held by chown/23358:
> > [14464.149407] #0: ffff8b9b8963c480 (sb_writers#3){++++}-{0:0}, at:
> > mnt_want_write+0x20/0x50
> > [14464.149407] #1: ffff8b9b8ff64bc8
> > (&sb->s_type->i_mutex_key#9){++++}-{3:3}, at: chown_common+0x85/0x150
> > [14464.149407] #2: ffff8b9b670c48d8 (jbd2_handle){++++}-{0:0}, at:
> > start_this_handle+0x1ad/0x690
> > [14464.149407] #3: ffff8b9b8ff648d0 (&ei->xattr_sem){++++}-{3:3}, at:
> > ext4_setattr+0x129/0x9c0
> > [14464.149407] #4: ffff8b9b1313f5f0 (&dquot->dq_lock){+.+.}-{3:3}, at:
> > dquot_acquire+0x23/0x100
> > [14464.149407] #5: ffff8b9b8963c208 (&s->s_dquot.dqio_sem){++++}-{3:3}, at:
> > v2_write_dquot+0x2d/0xb0
> > [14464.149407]
> > stack backtrace:
> > [14464.149407] CPU: 5 PID: 23358 Comm: chown Not tainted
> > 5.9.0-rc8-next-20201008 #16
> > [14464.149407] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
> > 1.13.0-1ubuntu1 04/01/2014
> > [14464.149407] Call Trace:
> > [14464.149407] dump_stack+0x77/0x97
> > [14464.149407] check_noncircular+0x132/0x150
> > [14464.149407] ? sched_clock_local+0x12/0x80
> > [14464.149407] __lock_acquire+0x147b/0x2830
> > [14464.149407] lock_acquire+0xc6/0x3a0
> > [14464.149407] ? ext4_map_blocks+0xd1/0x640
> > [14464.149407] down_write+0x40/0x110
> > [14464.149407] ? ext4_map_blocks+0xd1/0x640
> > [14464.149407] ext4_map_blocks+0xd1/0x640
> > [14464.149407] ? sched_clock_local+0x12/0x80
> > [14464.149407] ext4_getblk+0x54/0x1c0
> > [14464.149407] ext4_bread+0x1f/0xd0
> > [14464.149407] ext4_quota_write+0xbf/0x280
> > [14464.149407] write_blk+0x35/0x70
> > [14464.149407] get_free_dqblk+0x42/0xa0
> > [14464.149407] do_insert_tree+0x1d6/0x480
> > [14464.149407] ? ext4_quota_read+0x9d/0x110
> > [14464.149407] do_insert_tree+0x464/0x480
> > [14464.149407] ? ext4_quota_read+0x9d/0x110
> > [14464.149407] do_insert_tree+0x218/0x480
> > [14464.149407] ? ext4_quota_read+0x9d/0x110
> > [14464.149407] do_insert_tree+0x218/0x480
> > [14464.149407] ? __kmalloc+0x319/0x350
> > [14464.149407] qtree_write_dquot+0x79/0x1b0
> > [14464.149407] v2_write_dquot+0x52/0xb0
> > [14464.149407] dquot_acquire+0x8e/0x100
> > [14464.149407] ext4_acquire_dquot+0x72/0xc0
> > [14464.149407] dqget+0x24a/0x4a0
> > [14464.149407] dquot_transfer+0xfe/0x140
> > [14464.149407] ext4_setattr+0x134/0x9c0
> > [14464.149407] notify_change+0x34b/0x490
> > [14464.149407] ? chown_common+0x96/0x150
> > [14464.149407] chown_common+0x96/0x150
> > [14464.149407] ? preempt_count_add+0x49/0xa0
> > [14464.149407] do_fchownat+0x8d/0xe0
> > [14464.149407] __x64_sys_fchownat+0x21/0x30
> > [14464.149407] do_syscall_64+0x33/0x40
> > [14464.149407] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > [14464.149407] RIP: 0033:0x7efd3d2666ca
> > [14464.149407] Code: 48 8b 0d c9 f7 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66
> > 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 49 89 ca b8 04 01 00 00 0f 05 <48>
> > 3d 01 f0 ff ff 73 01 c3 48 8b 0d 96 f7 0c 00 f7 d8 64 89 01 48
> > [14464.149407] RSP: 002b:00007ffdbe6ae418 EFLAGS: 00000206 ORIG_RAX:
> > 0000000000000104
> > [14464.149407] RAX: ffffffffffffffda RBX: 00007ffdbe6ae650 RCX:
> > 00007efd3d2666ca
> > [14464.149407] RDX: 0000000000007ab7 RSI: 00005598d654ef60 RDI:
> > 00000000ffffff9c
> > [14464.149407] RBP: 00000000ffffff9c R08: 0000000000000000 R09:
> > 00000000ffffffff
> > [14464.149407] R10: 00000000ffffffff R11: 0000000000000206 R12:
> > 00005598d654e1f0
> > [14464.149407] R13: 00005598d654e268 R14: 0000000000007ab7 R15:
> > 00000000ffffffff
> > [14465.070206] EXT4-fs (vdc): mounted filesystem with ordered data mode.
> > Opts: grpquota
> > [14465.165678] EXT4-fs (vdc): re-mounted. Opts: (null)
> > [14465.201132] EXT4-fs (vdc): re-mounted. Opts: (null)
>
> Hi Ritesh:
>
> I've seen that exact failure on 1k at least as far back as 5.0. It fails
> on every 1k run for me using the test appliance. I've been treating this
> as a "known problem, likely harmless, no fix likely anytime soon" for so long
> I unfortunately don't remember what the original analysis was. IIRC,
> Jan Kara took a look at this one back when I first encountered it.
>
> Eric
