Re: [PATCH mm] vmalloc: back off when the current task is OOM-killed

From: Vasily Averin
Date: Mon Sep 27 2021 - 05:36:27 EST

On 9/24/21 10:55 AM, Michal Hocko wrote:
> On Thu 23-09-21 09:49:57, Vasily Averin wrote:
> [...]
>> I'm agree that vmalloc callers should expect and handle single vnalloc failures.
>> I think it is acceptable to enable fatal_signal_pending check to quickly
>> detect such kind of iussues.
>> However fatal_signal_pending check can cause serial vmalloc failures
>> and I doubt it is acceptable.
>> Rollback after failed vmalloc can call new vmalloc calls that will be failed too,
>> even properly handled such serial failures can cause troubles.
> Could you be more specific? Also how would this be any different from
> similar failures for an oom victim? Except that the later is less likely
> so (as already mentioend) any potential bugs would be just lurking there
> for a longer time.
>> Hypothetically, cancelled vmalloc called inside some filesystem's transaction
>> forces its rollback, that in own turn it can call own vmalloc.
> Do you have any specific example?

No, it was pure hypothetical assumption.
I was thinking about it over the weekend, and decided that:
a) such kind of issue (i.e. vmalloc call in rollback after failed vmalloc)
is very unlikely
b) if it still exists -- it must have own failback with kmalloc(NOFAIL)
or just accept/ignore such failure and should not lead to critical failures.
If this still happen -- ihis is a bug, we should detect and fix it ASAP.

>> Should we perhaps interrupt the first vmalloc only?
> This doesn't make much sense to me TBH. It doesn't address the very
> problem you are describing in the changelog.

Last question:
how do you think, should we perhaps, instead, detect such vmallocs
(called in rollback after failed vmalloc) and generate a warnings,
to prevent such kind of problems in future?

Thank you,
Vasily Averin