Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 26 Nov 2018 14:36:10 +0100
From:   Ingo Molnar <mingo@kernel.org>
To:     Thomas Gleixner <tglx@linutronix.de>
Cc:     Andy Lutomirski <luto@amacapital.net>,
        Andi Kleen <ak@linux.intel.com>,
        LKML <linux-kernel@vger.kernel.org>, x86@kernel.org,
        Peter Zijlstra <peterz@infradead.org>,
        Andy Lutomirski <luto@kernel.org>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Jiri Kosina <jkosina@suse.cz>,
        Tom Lendacky <thomas.lendacky@amd.com>,
        Josh Poimboeuf <jpoimboe@redhat.com>,
        Andrea Arcangeli <aarcange@redhat.com>,
        David Woodhouse <dwmw@amazon.co.uk>,
        Tim Chen <tim.c.chen@linux.intel.com>,
        Dave Hansen <dave.hansen@intel.com>,
        Casey Schaufler <casey.schaufler@intel.com>,
        Asit Mallick <asit.k.mallick@intel.com>,
        Arjan van de Ven <arjan@linux.intel.com>,
        Jon Masters <jcm@redhat.com>,
        Waiman Long <longman9394@gmail.com>,
        Greg KH <gregkh@linuxfoundation.org>,
        Dave Stewart <david.c.stewart@intel.com>,
        Kees Cook <keescook@chromium.org>
Subject: Re: [patch V2 21/28] x86/speculation: Prepare for conditional IBPB
 in switch_mm()
Message-ID: <20181126133610.GB108014@gmail.com>
References: <20181125183328.318175777@linutronix.de>
 <20181125185005.466447057@linutronix.de>
 <20181125205330.GO13936@tassilo.jf.intel.com>
 <alpine.DEB.2.21.1811252316560.1684@nanos.tec.linutronix.de>
 <92833780-41BE-446E-A676-925BA1EC93D9@amacapital.net>
 <alpine.DEB.2.21.1811260751030.1684@nanos.tec.linutronix.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <alpine.DEB.2.21.1811260751030.1684@nanos.tec.linutronix.de>
User-Agent: Mutt/1.9.4 (2018-02-28)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


* Thomas Gleixner <tglx@linutronix.de> wrote:

> On Sun, 25 Nov 2018, Andy Lutomirski wrote:
> > > On Nov 25, 2018, at 2:20 PM, Thomas Gleixner <tglx@linutronix.de> wrote:
> > > On Sun, 25 Nov 2018, Andi Kleen wrote:
> > > 
> > >>> The current check whether two tasks belong to the same context is using the
> > >>> tasks context id. While correct, it's simpler to use the mm pointer because
> > >>> it allows to mangle the TIF_SPEC_IB bit into it. The context id based
> > >>> mechanism requires extra storage, which creates worse code.
> > >> 
> > >> [We tried similar in some really early versions, but it was replaced
> > >> with the context id later.]
> > >> 
> > >> One issue with using the pointer is that the pointer can be reused
> > >> when the original mm_struct is freed, and then gets reallocated
> > >> immediately to an attacker. Then the attacker may avoid the IBPB.
> > >> 
> > >> Given it's probably hard to generate any reasonable leak bandwidth with
> > >> such a complex scenario, but it still seemed better to close the hole.
> > > 
> > > Sorry, but that's really a purely academic exercise. 
> > 
> > I would guess that it’s actually very easy to force mm_struct* reuse.
> > Don’t the various allocators try to allocate hot memory?  There’s nothing
> > hotter than a just-freed allocation of the same size.
> 
> Sure, but this is about a indirect branch predictor attack against
> something which reuses the mm.
> 
> So you'd need to pull off:
> 
>    P1 poisons branch predictor
>    P1 exit
> 
>    P2 starts and resuses mm(P1) and uses the poisoned branch predictor
> 
> the only thing between P1 and P2 is either idle or some other kernel
> thread, but no other user task. If that happens then the code would not
> issue IBPB as it assumes to switch back to the same process.
> 
> Even if you can pull that off the speculation would hit the startup code of
> P2, which is truly a source of secret information. Creating a valuable
> attack based on mm reuse is really less proabable than a lottery jackpot.
> 
> So using mm is really good enough and results in better assembly code which
> is surely more valuable than addressing some hypothetical hole.

OTOH we could probably close even this with very little cost if we added 
an IBPB to non-threaded fork() and vfork()+exec() paths? Those are really 
slow paths compared to all the context switch paths we are trying to 
optimize here.

Alternatively we could IBPB on the post-exit() final task struct freeing, 
which too is a relative slow path compared to the context switch paths.

But no strong opinion.

Thanks,

	Ingo