Re: [PATCH] ext4: remove abnormal set for I_DATA_SEM subclass

From: Theodore Y. Ts'o
Date: Sun Jul 22 2018 - 18:08:52 EST


On Mon, Jul 09, 2018 at 04:58:28PM +0900, Junil Lee wrote:
> The -EBUSY return value of dquot_enable() function means that just
> want to update flags. If some users make a duplicate request to update
> flags, lockdep could catch the false positive casued by needing to
> allocate a quota block from inside ext4_map_blocks(), while holding
> i_data_sem for a data inode. This results in this complaint:
>
> CPU0 CPU1
> ---- ----
> lock(&s->s_dquot.dqio_mutex);
> lock(&ei->i_data_sem);
> lock(&s->s_dquot.dqio_mutex);
> lock(&ei->i_data_sem);

How does this happen in practice? The function ext4_quota_enable() is
only called by ext4_enable_quotas(), and I don't see the code path
where this would happen. And if it does it would be resulting an
EXT4-fs warning message getting printing indicating that a failure to
enable quotas with an error of EBUSY. So how does this happen that
"users would make a duplicate request to update flags"?

- Ted