Possible race in mlx5_ib.ko

Possible race in mlx5_ib.ko

[Anton Volkov]

Hello.

While searching for races in the Linux kernel I've come across
"drivers/infiniband/hw/mlx5/mlx5_ib.ko" module. Here are questions that 
I came up with while analyzing results. Lines are given using the info 
from Linux v4.12.

Consider the following case:

Thread 1:              Thread 2:
size_write
->remove_keys          limit_write
     ent->cur--;          if (ent->cur < ent->limit)
     (mr.c: line 234)     (mr.c: line 335)
                             err = add_keys(... ent->limit - ent->cur);

If size_write and limit_write are able to work concurrently with the 
same ent then there is a possibility of a race between the accesses to 
ent->cur. In worst case in limit_write new keys wouldn't be added. Is it 
feasible from your point of view? If so, is it a benign race or a 
serious one?

Thank you for your time.

-- Anton Volkov
Linux Verification Center, ISPRAS
web: http://linuxtesting.org
e-mail: avolkov@ispras.ru

Re: Possible race in mlx5_ib.ko

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1441 bytes --]

On Fri, Aug 18, 2017 at 06:17:15PM +0300, Anton Volkov wrote:
> Hello.
>
> While searching for races in the Linux kernel I've come across
> "drivers/infiniband/hw/mlx5/mlx5_ib.ko" module. Here are questions that I
> came up with while analyzing results. Lines are given using the info from
> Linux v4.12.
>
> Consider the following case:
>
> Thread 1:              Thread 2:
> size_write
> ->remove_keys          limit_write
>     ent->cur--;          if (ent->cur < ent->limit)
>     (mr.c: line 234)     (mr.c: line 335)
>                             err = add_keys(... ent->limit - ent->cur);
>
> If size_write and limit_write are able to work concurrently with the same
> ent then there is a possibility of a race between the accesses to ent->cur.
> In worst case in limit_write new keys wouldn't be added. Is it feasible from
> your point of view? If so, is it a benign race or a serious one?

Technically yes, it can race, but it gives a little value to protect
the ent->check with lock, because the limit_eize instruction is triggered
by debugfs write and is used for the debug.

Thanks

>
> Thank you for your time.
>
> -- Anton Volkov
> Linux Verification Center, ISPRAS
> web: http://linuxtesting.org
> e-mail: avolkov@ispras.ru
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]