Date: Tue, 22 May 2012 22:07:05 +0100
From: Al Viro <viro@ZenIV.linux.org.uk>
To: Will Drewry <wad@chromium.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>, Indan Zupancic <indan@nul.nu>,
        Roland McGrath <mcgrathr@google.com>,
        Eric Paris <netdev@parisplace.org>, linux-kernel@vger.kernel.org,
        linux-security-module@vger.kernel.org,
        kernel-hardening@lists.openwall.com, mingo@redhat.com, oleg@redhat.com,
        peterz@infradead.org, rdunlap@xenotime.net, tglx@linutronix.de,
        luto@mit.edu, eparis@redhat.com, serge.hallyn@canonical.com,
        pmoore@redhat.com, akpm@linux-foundation.org, corbet@lwn.net,
        eric.dumazet@gmail.com, markus@chromium.org, coreyb@linux.vnet.ibm.com,
        keescook@chromium.org
Subject: Re: seccomp and ptrace. what is the correct order?
Message-ID: <20120522210704.GK11775@ZenIV.linux.org.uk>
References: <CACLa4puRJLUd6hAHO6zFYiqa5dFVum8eLmNMk_ef+=r8OGhjRg@mail.gmail.com>
 <CAB=4xho2b5fZV_jC15XW3mhOD9Z0QUMnV_UHy-P5EmQWfX59dA@mail.gmail.com>
 <5d6179f59222155b72d9aa9f171e883c.squirrel@webmail.greenhost.nl>
 <CABqD9hbzatHerMA1Nq57Qn8SKSZzZf=5t6aLumG9ZpAafgX=UQ@mail.gmail.com>
 <20120522173942.GJ11775@ZenIV.linux.org.uk>
 <CABqD9hbwnp-Kob0-Uc9L=qgJfS88po=RHNPkhQFcSf4RTrFCmQ@mail.gmail.com>
 <4FBBF84A.4070308@zytor.com>
 <CABqD9haSh=Tof9n2m8PHBaoqac1kcUZq-f5BcjDBv++5APGTCg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CABqD9haSh=Tof9n2m8PHBaoqac1kcUZq-f5BcjDBv++5APGTCg@mail.gmail.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1918
Lines: 36

On Tue, May 22, 2012 at 03:48:40PM -0500, Will Drewry wrote:
> On Tue, May 22, 2012 at 3:34 PM, H. Peter Anvin <hpa@zytor.com> wrote:
> > The proposed patch seems to duplicate the functionality in
> > <asm/syscall.h>. ?Those macros also try to do the right thing in the
> > presence of compat.
> 
> That was my first thought too, so I ran a few simple tests.  gcc isn't
> smart enough to not add ~344 bytes of code to get the number and
> arguments for the x86/kernel/ptrace.c case I included (in the
> naive-est of integrations).  But I don't know that it justifies the
> extra patchwork or enforcing shared code across arches.
> 
> Regardless, the syscall entry + trace code can use some attention
> across the architectures. I don't know that
> one-more-layer-of-abstraction is the right answer (rather than just
> fixing the code). The biggest benefit would be having one-true
> syscall_trace_entry slow path. That said, the fast paths will be
> forever divergent so the opportunity for bugs like the ones pointed
> out will still be there.

FWIW, I'd prefer to have all that done inside __audit_syscall_entry(),
with
        context->arch       = syscall_get_arch(current, regs);
        context->major      = syscall_get_nr(current, regs);
	syscall_get_arguments(current, regs, 0, 4, context->argv);
done instead of initializations from arguments we are doing there now.
I seriously doubt that it would lead to worse code than what we currently
have.  If nothing else, we won't be passing that pile of arguments around.

And yes, asm/syscall.h stuff is probably the right approach here.  For
biarch ones syscall_get_arguments() is saner than doing them one by one...
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/