Re: [PATCH] ocfs2: cancel dqi_sync_work before freeing oinfo

[Heming Zhao]

On 9/4/24 10:09, Joseph Qi wrote:
> ocfs2_global_read_info() will schedule dqi_sync_work. So any error
> occurs after it, we have to cancel this delayed work first before
> freeing oinfo, otherwise it will trigger the following warning with
> CONFIG_DEBUG_OBJECTS_* enabled:
>
> ODEBUG: free active (active state 0) object: 00000000d8b0ce28 object type: timer_list hint: qsync_work_fn+0x0/0x16c
>
> Link: https://syzkaller.appspot.com/bug?extid=f7af59df5d6b25f0febd
> Reported-by: syzbot+f7af59df5d6b25f0febd@xxxxxxxxxxxxxxxxxxxxxxxxx
> Tested-by: syzbot+f7af59df5d6b25f0febd@xxxxxxxxxxxxxxxxxxxxxxxxx
> Fixes: 171bf93ce11f ("ocfs2: Periodic quota syncing")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Joseph Qi <joseph.qi@xxxxxxxxxxxxxxxxx>
> ---
>   fs/ocfs2/quota_local.c | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/fs/ocfs2/quota_local.c b/fs/ocfs2/quota_local.c
> index 8ce462c64c51..ebe0dbc8db4a 100644
> --- a/fs/ocfs2/quota_local.c
> +++ b/fs/ocfs2/quota_local.c
> @@ -782,6 +782,7 @@ static int ocfs2_local_read_info(struct super_block *sb, int type)
>   		if (locked)
>   			ocfs2_inode_unlock(lqinode, 1);
>   		ocfs2_release_local_quota_bitmaps(&oinfo->dqi_chunk);
> +		cancel_delayed_work_sync(&oinfo->dqi_sync_work);
>   		kfree(oinfo);
>   	}
>   	brelse(bh);

In my view, there is one problem:
If ocfs2_global_read_info() returns error before ->dqi_sync_work is
initialized, above code will trigger wild pointer error.

-Heming