Re: [kernel-hardening] Re: [PATCH 1/2] security, perf: allow further restriction of perf_event_open

[Daniel Micay <danielmicay@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2015 bytes --]

On Thu, 2016-08-04 at 17:10 +0100, Mark Rutland wrote:
> On Thu, Aug 04, 2016 at 11:44:28AM -0400, Daniel Micay wrote:
> > 
> > Qualcomm's drivers might be lower quality than core kernel code, but
> > they're way above the baseline set by mainline kernel drivers...
> 
> I don't think that's true for the arm/arm64 perf code.

The baseline architecture support is essentially core kernel code. I
agree it's much better than the SoC vendor code. You're spending a lot
of time auditing, fuzzing and improving the code in general, which is
not true for most drivers. They don't get that attention.

> I think we've done a reasonable job of testing and fixing those, along
> with core infrastructure issues. The perf fuzzer runs for a very long
> time on a mainline kernel without issues, while on my Nexus 5x I get a
> hard lockup after ~85 seconds (and prior to the last android update
> the
> lockup was instantaneous).
>
> From my personal experience (and as above), and talking specifically
> about PMU drivers, I think that the opposite is true. This is not to
> say
> there aren't issues; I would not be surprised if there are. But it's
> disingenuous to say that mainline code is worse than that which exists
> in a vendor kernel when the latter is demonstrably much easier to
> break
> than the former.

I wasn't talking specifically about perf.

> If there are issues you are aware of, please report them. If those
> issues only exist in non-upstream code, then the applicable concerns
> are
> somewhat different (though certainly still exist).

I'm not going to do volunteer work for a corporation. I've learned that
lesson after spending years doing it.

> But please, let's frame the argument to match reality.

The argument is framed in reality. Stating that it now often takes a few
hours to find a vulnerability with the unaltered, widely known public
perf fuzzer is not impressive. It's really an argument for claiming that
it's a significant security issue.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 851 bytes --]