Re: [Qemu-devel] [PATCH] libseccomp: add cacheflush to whitelist

[Andrew Jones <drjones@redhat.com>]

On Thu, Sep 24, 2015 at 11:31:19AM +0200, Eduardo Otubo wrote:
> On Wed, Jul 01, 2015 at 09=12=33AM -0400, Andrew Jones wrote:
> > cacheflush is an arm-specific syscall that qemu built for arm
> > uses. Add it to the whitelist.
> > 
> > Signed-off-by: Andrew Jones <drjones@redhat.com>
> > 
> > ---
> > 
> > I'm not sure about the priority selection. Maybe cacheflush gets
> > used frequently enough that it deserves a higher one?
> 
> The frequency is measured using strace and comparing the frequency they
> appear among other syscalls. Can you run this analysis and double check
> if the lowest priority is still accurate?

Hi Eduardo,

Short answer: The lowest priority is definitely correct.

Long answer:

I ran strace while installing a new guest, of 3.6 million syscalls,
only 5 were cacheflush. Of course the syscalls used (and their frequency)
is host-type, qemu machine-type, config (qemu command line), and guest
workload specific. So, ideally, qemu machine-types would register their
own whitelists, possibly modified by host-type. For example, I ran the
mach-virt machine-type on both a midway and a mustang. In both cases it
was a basic guest config and an install-type workload. For the mustang,
over 55% of the syscalls were ioctl, but, for the midway, ioctls were
16% and 43% were clock_gettime. I generated a most-used-first list for
each. Neither list really matched up well with seccomp_whitelist (except
for futex).

Besides allowing machine types to help set priorities, it may also be
nice if both compile-time and run-time configs could further reduce the
whitelist. For example, mlockall is only necessary if '-realtime mlock=on'
is passed on the command line.

Thanks,
drew

> 
> Thanks for the patch.
> 
> > 
> > This patch isn't really necessary yet due to ae6e8ef11e6c: "Revert
> > seccomp tests that allow it to be used on non-x86 architectures",
> > which we can't revert until libseccomp has released a fix for
> > arm-specific syscall symbol naming, but when linking to a patched
> > libseccomp and reverting ae6e8ef11e6c, then this patch allows
> > guests to boot with '-sandbox on'.
> > 
> > Signed-off-by: Andrew Jones <drjones@redhat.com>
> > ---
> >  qemu-seccomp.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/qemu-seccomp.c b/qemu-seccomp.c
> > index f9de0d3390feb..33644a4e3c3d3 100644
> > --- a/qemu-seccomp.c
> > +++ b/qemu-seccomp.c
> > @@ -237,7 +237,8 @@ static const struct QemuSeccompSyscall seccomp_whitelist[] = {
> >      { SCMP_SYS(fadvise64), 240 },
> >      { SCMP_SYS(inotify_init1), 240 },
> >      { SCMP_SYS(inotify_add_watch), 240 },
> > -    { SCMP_SYS(mbind), 240 }
> > +    { SCMP_SYS(mbind), 240 },
> > +    { SCMP_SYS(cacheflush), 240 },
> >  };
> >  
> >  int seccomp_start(void)
> > -- 
> > 2.1.0
> > 
> 
> -- 
> Eduardo Otubo
> ProfitBricks GmbH