[Devel] [PATCH vz7 2/2] block: store exec_ub on struct request and use it

[Andrey Ryabinin]

On 08/02/2018 05:11 PM, Konstantin Khorenko wrote:
> Good calltrace: we set proper exec_ub in ploop_req_state_process()
> and later process requests - accounting goes to correct ub.
> 
> loop35172  6275 [002]   755.716683:  probe:ub_writeback_io_3:
> 			(ffffffff813547aa) ub=0xffff880394a04000
>                   5547ab deadline_add_request
>                   52e76a __elv_add_request
>                   536350 blk_flush_plug_list
>                   5365aa blk_queue_bio
>                   5345bb generic_make_request
>                   534843 submit_bio
>                     43e4 dio_submit ([pio_direct])
>                     616f ploop_entry_request ([ploop])
>                     7209 ploop_req_state_process ([ploop])
>                     7735 ploop_thread ([ploop])
> 
> Bad calltrace: due to task->plug io batching we may miss calling
> ploop_req_state_process() and process requests with ploop thread's ub == ub0:
> 
> ploop35172  6275 [002]   755.716763:  probe:ub_writeback_io_2:
> 			 (ffffffff81354785) ub=0xffffffff81f7b880 (ub0!!!)
>                   554786 deadline_add_request
>                   52e76a __elv_add_request
>                   536350 blk_flush_plug_list
>                   536714 blk_finish_plug
>                     1a93 ploop_wait ([ploop])
>                     77a1 ploop_thread ([ploop])
> 
> => need to store exec_ub in struct request as well.
> 
> We are safe to check for NULL request:req_ub to detect set it or not
> because request is zeroed on creation:
> 
>    get_request
>     __get_request
>      blk_rq_init
>       memset(rq, 0, sizeof(*rq));
> 
> https://jira.sw.ru/browse/PSBM-86910
> 
> Signed-off-by: Konstantin Khorenko <khorenko at virtuozzo.com>

Reviewed-by: Andrey Ryabinin <aryabinin at virtuozzo.com>