2017-12-13 03:18:21

by Jia-Ju Bai

[permalink] [raw]
Subject: [BUG] scsi/qla2xxx: a possible sleep-in-atomic bug in qlt_get_tag

The driver may sleep under a spinlock.
The function call paths are:
qlt_handle_abts_recv_work (acquire the spinlock)
qlt_response_pkt_all_vps
qlt_response_pkt
qlt_handle_cmd_for_atio
qlt_get_tag
percpu_ida_alloc --> may sleep

qla82xx_msix_rsp_q (acquire the spinlock)
qla24xx_process_response_queue
qlt_handle_abts_recv
qlt_response_pkt_all_vps
qlt_response_pkt
qlt_handle_cmd_for_atio
qlt_get_tag
percpu_ida_alloc --> may sleep-in-atomic

qla24xx_intr_handler (acquire the spinlock)
qla24xx_process_response_queue
qlt_handle_abts_recv
qlt_response_pkt
qlt_handle_cmd_for_atio
qlt_get_tag
percpu_ida_alloc --> may sleep

I do not find a good way to fix it, so I only report.
This possible bug is found by my static analysis tool (DSAC) and checked
by my code review.


Thanks,
Jia-Ju Bai


2017-12-13 04:42:11

by James Bottomley

[permalink] [raw]
Subject: Re: [BUG] scsi/qla2xxx: a possible sleep-in-atomic bug in qlt_get_tag

On Wed, 2017-12-13 at 11:18 +0800, Jia-Ju Bai wrote:
> The driver may sleep under a spinlock.
> The function call paths are:
> qlt_handle_abts_recv_work (acquire the spinlock)
>    qlt_response_pkt_all_vps
>      qlt_response_pkt
>        qlt_handle_cmd_for_atio
>          qlt_get_tag
>            percpu_ida_alloc --> may sleep
>
> qla82xx_msix_rsp_q (acquire the spinlock)
>    qla24xx_process_response_queue
>      qlt_handle_abts_recv
>        qlt_response_pkt_all_vps
>          qlt_response_pkt
>            qlt_handle_cmd_for_atio
>              qlt_get_tag
>                percpu_ida_alloc --> may sleep-in-atomic
>
> qla24xx_intr_handler (acquire the spinlock)
>    qla24xx_process_response_queue
>      qlt_handle_abts_recv
>        qlt_response_pkt
>          qlt_handle_cmd_for_atio
>            qlt_get_tag
>              percpu_ida_alloc --> may sleep
>
> I do not find a good way to fix it, so I only report.
> This possible bug is found by my static analysis tool (DSAC) and
> checked by my code review.

The report is incorrect: percpu_ida_alloc with state==TASK_RUNNING is
atomic (and interrupt) safe which appears to be the case here.

James

2017-12-13 07:38:36

by Jia-Ju Bai

[permalink] [raw]
Subject: Re: [BUG] scsi/qla2xxx: a possible sleep-in-atomic bug in qlt_get_tag


On 2017/12/13 12:42, James Bottomley wrote:
> On Wed, 2017-12-13 at 11:18 +0800, Jia-Ju Bai wrote:
>> The driver may sleep under a spinlock.
>> The function call paths are:
>> qlt_handle_abts_recv_work (acquire the spinlock)
>> qlt_response_pkt_all_vps
>> qlt_response_pkt
>> qlt_handle_cmd_for_atio
>> qlt_get_tag
>> percpu_ida_alloc --> may sleep
>>
>> qla82xx_msix_rsp_q (acquire the spinlock)
>> qla24xx_process_response_queue
>> qlt_handle_abts_recv
>> qlt_response_pkt_all_vps
>> qlt_response_pkt
>> qlt_handle_cmd_for_atio
>> qlt_get_tag
>> percpu_ida_alloc --> may sleep-in-atomic
>>
>> qla24xx_intr_handler (acquire the spinlock)
>> qla24xx_process_response_queue
>> qlt_handle_abts_recv
>> qlt_response_pkt
>> qlt_handle_cmd_for_atio
>> qlt_get_tag
>> percpu_ida_alloc --> may sleep
>>
>> I do not find a good way to fix it, so I only report.
>> This possible bug is found by my static analysis tool (DSAC) and
>> checked by my code review.
> The report is incorrect: percpu_ida_alloc with state==TASK_RUNNING is
> atomic (and interrupt) safe which appears to be the case here.
>
> James
>

Thanks for your reply :)
I have checked the definition of percpu_ida_alloc, and I think you are
right.
Sorry for my incorrect bug report.


Thanks,
Jia-Ju Bai