Re: [PATCH v1 0/6] seccomp: Implement constant action bitmaps

From: Kees Cook <keescook@chromium.org>
To: YiFei Zhu <zhuyifei1999@gmail.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>,
	YiFei Zhu <yifeifz2@illinois.edu>,
	Andrea Arcangeli <aarcange@redhat.com>,
	Giuseppe Scrivano <gscrivan@redhat.com>,
	Will Drewry <wad@chromium.org>, bpf <bpf@vger.kernel.org>,
	Jann Horn <jannh@google.com>,
	Linux API <linux-api@vger.kernel.org>,
	Linux Containers <containers@lists.linux-foundation.org>,
	Tobin Feldman-Fitzthum <tobin@ibm.com>,
	Hubertus Franke <frankeh@us.ibm.com>,
	Andy Lutomirski <luto@amacapital.net>,
	Valentin Rothberg <vrothber@redhat.com>,
	Dimitrios Skarlatos <dskarlat@cs.cmu.edu>,
	Jack Chen <jianyan2@illinois.edu>,
	Josep Torrellas <torrella@illinois.edu>,
	Tianyin Xu <tyxu@illinois.edu>,
	kernel list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v1 0/6] seccomp: Implement constant action bitmaps
Date: Mon, 28 Sep 2020 13:04:50 -0700	[thread overview]
Message-ID: <202009281259.D7D18AE95@keescook> (raw)
In-Reply-To: <CABqSeASMObs7HtwfM=ua9Tbx1mfHZaxCMWD6AP6-6hR4-Xcn=Q@mail.gmail.com>

On Sat, Sep 26, 2020 at 01:11:50PM -0500, YiFei Zhu wrote:
> On Fri, Sep 25, 2020 at 2:07 AM YiFei Zhu <zhuyifei1999@gmail.com> wrote:
> > I'll try to profile the latter later on my qemu-kvm, with a recent
> > libsecomp with binary tree and docker's profile, probably both direct
> > filter attaches and filter attaches with fork(). I'm guessing if I
> > have fork() the cost of fork() will overshadow seccomp() though.
> 
> I'm surprised. That is not the case as far as I can tell.
> 
> I wrote a benchmark [1] that would fork() and in the child attach a
> seccomp filter, look at the CLOCK_MONOTONIC difference, then add it to
> a struct timespec shared with the parent. It checks the difference
> with the timespec before prctl and before fork. CLOCK_MONOTONIC
> instead of CLOCK_PROCESS_CPUTIME_ID because of fork.
> 
> I ran `./seccomp_emu_bench 100000` in my qemu-kvm and here are the results:
> without emulator:
> Benchmarking 100000 syscalls...
> 19799663603 (19.8s)
> seecomp attach without fork: 197996 ns
> 33911173847 (33.9s)
> seecomp attach with fork: 339111 ns
> 
> with emulator:
> Benchmarking 100000 syscalls...
> 54428289147 (54.4s)
> seecomp attach without fork: 544282 ns
> 69494235408 (69.5s)
> seecomp attach with fork: 694942 ns
> 
> fork seems to take around 150us, seccomp attach takes around 200us,
> and the filter emulation overhead is around 350us. I had no idea that
> fork was this fast. If I wrote my benchmark badly please criticise.

You're calling clock_gettime() inside your loop. That might change the
numbers. Why not just measure outside the loop, or better yet, use
"perf" to measure the time in prctl().

> Given that we are doubling the time to fork() + seccomp attach filter,
> I think yeah running the emulator on the first instance of a syscall,
> holding a lock, is a much better idea. If I naively divide 350us by
> the number of syscall + arch pairs emulated the overhead is less than
> 1 us and that should be okay since it only happens for the first
> invocation of the particular syscall.
> 
> [1] https://gist.github.com/zhuyifei1999/d7bee62bea14187e150fef59db8e30b1

Regardless, let's take things one step at a time. First, let's do
the simplest version of the feature, and then let's look at further
optimizations.

Can you send a v3 and we can continue from there?

-- 
Kees Cook