On Wed, Jul 7, 2021 at 9:50 AM Alexey Gladkov <legion@xxxxxxxxxx> wrote:
>
> > + dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING, 1);
> > + fallthrough;
> > + case LONG_MAX:
>
> I think that the counter should be decreased in this case too.
> inc_rlimit_ucounts() increments the counter in all parent userns. If we
> don't decrease the counter then the parent userns will have a counter
> leak.
Ack. So basically that patch, but move the dec_rlimit_ucounts() into
the LONG_MAX case?
Would you mind making a real patch with a commit message, and trying
whatever test-case you had for that KASAN use-after-free report?
Linus
