Re: [GIT PULL] ucounts: Count rlimits in each user namespace

[Eric W. Biederman]

Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> writes:

> On Mon, Jun 28, 2021 at 3:35 PM Eric W. Biederman <ebiederm@xxxxxxxxxxxx> wrote:
>>
>> This is the work mainly by Alexey Gladkov to limit rlimits to the
>> rlimits of the user that created a user namespace, and to allow users to
>> have stricter limits on the resources created within a user namespace.
>
> I guess all the performance issues got sorted, since I haven't seen
> any reports from the test robots.

Yes.  The structure was made to not change anything unnecessarily
(such as the ordering in sigqueue_alloc) and the performances
differences went away.

With the code in linux-next the entire cycle I think that is a reliable
result.  There are probably some things we could do to further optimize
things but we did not need them to avoid regressions.

> I do end up with two questions, mainly because of looking at the
> result of the conflict resolution.
>
> In particular, in __sigqueue_alloc(), two oddities..
>
> Why the "sigpending < LONG_MAX" test in that
>
>         if (override_rlimit || (sigpending < LONG_MAX && sigpending <=
> task_rlimit(t, RLIMIT_SIGPENDING))) {
>
> thing?

> And why test for "ucounts" being non-NULL in
>
>                 if (ucounts && dec_rlimit_ucounts(ucounts,
> UCOUNT_RLIMIT_SIGPENDING, 1))
>                         put_ucounts(ucounts);
>
> when afaik both of those should be happy with a NULL 'ucounts' pointer
> (if it was NULL, we certainly already used it for the reverse
> operations for get_ucounts() and inc_rlimit_ucounts()..)
>
> Hmm?

Yes.  I suspect that those tests are left over from a previous version
of the change.  Alex do you remember why those tests are there?

> And somebody should verify that I didn't screw anything up in my merge
> resolution. It all looked very straightforward, but mistakes happen..

Just reading through the resolution looks correct.

Eric