Re: [PATCH] ublk_drv: fix NULL pointer dereference in ublk_ctrl_start_recovery()

[Ming Lei <ming.lei@redhat.com>]

On Wed, May 29, 2024 at 05:53:13PM +0800, linan666@huaweicloud.com wrote:
> From: Li Nan <linan122@huawei.com>
> 
> When two UBLK_CMD_START_USER_RECOVERY commands are submitted, the
> first one sets 'ubq->ubq_daemon' to NULL, and the second one triggers
> WARN in ublk_queue_reinit() and subsequently a NULL pointer dereference
> issue.
> 
> Continuing execution after WARN is incorrect, as 'ubq->ubq_daemon' is
> known to be NULL. Fix it by return directly if the WARN is triggered.
> 
> Note that WARN will still be triggered after the fix if anyone tries to
> start recovery twice.
> 
>   BUG: kernel NULL pointer dereference, address: 0000000000000028
>   RIP: 0010:ublk_ctrl_start_recovery.constprop.0+0x82/0x180
>   Call Trace:
>    <TASK>
>    ? __die+0x20/0x70
>    ? page_fault_oops+0x75/0x170
>    ? exc_page_fault+0x64/0x140
>    ? asm_exc_page_fault+0x22/0x30
>    ? ublk_ctrl_start_recovery.constprop.0+0x82/0x180
>    ublk_ctrl_uring_cmd+0x4f7/0x6c0
>    ? pick_next_task_idle+0x26/0x40
>    io_uring_cmd+0x9a/0x1b0
>    io_issue_sqe+0x193/0x3f0
>    io_wq_submit_work+0x9b/0x390
>    io_worker_handle_work+0x165/0x360
>    io_wq_worker+0xcb/0x2f0
>    ? finish_task_switch.isra.0+0x203/0x290
>    ? finish_task_switch.isra.0+0x203/0x290
>    ? __pfx_io_wq_worker+0x10/0x10
>    ret_from_fork+0x2d/0x50
>    ? __pfx_io_wq_worker+0x10/0x10
>    ret_from_fork_asm+0x1a/0x30
>    </TASK>
> 
> Fixes: c732a852b419 ("ublk_drv: add START_USER_RECOVERY and END_USER_RECOVERY support")
> Reported-and-tested-by: Changhui Zhong <czhong@redhat.com>
> Closes: https://lore.kernel.org/all/CAGVVp+UvLiS+bhNXV-h2icwX1dyybbYHeQUuH7RYqUvMQf6N3w@mail.gmail.com
> Signed-off-by: Li Nan <linan122@huawei.com>
> ---
>  drivers/block/ublk_drv.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> index 4e159948c912..99b621b2d40f 100644
> --- a/drivers/block/ublk_drv.c
> +++ b/drivers/block/ublk_drv.c
> @@ -2630,7 +2630,8 @@ static void ublk_queue_reinit(struct ublk_device *ub, struct ublk_queue *ubq)
>  {
>  	int i;
>  
> -	WARN_ON_ONCE(!(ubq->ubq_daemon && ubq_daemon_is_dying(ubq)));
> +	if (WARN_ON_ONCE(!(ubq->ubq_daemon && ubq_daemon_is_dying(ubq))))
> +		return;

Yeah, it is one bug. However, it could be addressed by adding the check in
ublk_ctrl_start_recovery() and return immediately in case of NULL ubq->ubq_daemon,
what do you think about this way?


Thanks,
Ming