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;
MIME-Version: 1.0
References: <20180807172920.8766-1-sean.j.christopherson@intel.com> <b7fe6fad-7ce4-c7c1-31d7-cb609cf45b2f@intel.com>
In-Reply-To: <b7fe6fad-7ce4-c7c1-31d7-cb609cf45b2f@intel.com>
From:   Jann Horn <jannh@google.com>
Date:   Fri, 31 Aug 2018 04:38:44 +0200
Message-ID: <CAG48ez1iPr0q-PBYnc9Q4GA0uMEyph9THiV2CrN6aRMzqpEfng@mail.gmail.com>
Subject: Re: [PATCH] x86/pkeys: Explicitly treat PK #PF on kernel address as a
 bad area
To:     Dave Hansen <dave.hansen@intel.com>,
        Andy Lutomirski <luto@kernel.org>
Cc:     sean.j.christopherson@intel.com,
        kernel list <linux-kernel@vger.kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>,
        "H . Peter Anvin" <hpa@zytor.com>,
        "the arch/x86 maintainers" <x86@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, 7 Aug 2018 Dave Hansen <dave.hansen@intel.com> wrote:
>
> On 08/07/2018 10:29 AM, Sean Christopherson wrote:
> >       if (unlikely(fault_in_kernel_space(address))) {
> > +             /*
> > +              * We should never encounter a protection keys fault on a
> > +              * kernel address as kernel address are always mapped with
> > +              * _PAGE_USER=0, i.e. PKRU isn't enforced.
> > +              */
> > +             if (WARN_ON_ONCE(error_code & X86_PF_PK))
> > +                     goto bad_kernel_address;
>
> I just realized one more thing: the vsyscall page can bite us here.
> It's at a fault_in_kernel_space() address and we *can* trigger a pkey
> fault on it if we jump to an instruction that reads from a
> pkey-protected area.
>
> We can make a gadget out of unaligned vsyscall instructions that does
> that.  See:
>
> 0xffffffffff600002:  shlb   $0x0,0x0(%rax)
>
> Then, we turn off access to all pkeys, including pkey-0, then jump to
> the unaligned vsyscall instruction, which reads %rax, which is a kernel
> address:

Andy got rid of the (native) vsyscall page in
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=076ca272a14cea558b1092ec85cea08510283f2a
('x86/vsyscall/64: Drop "native" vsyscalls') a few months ago, right?
At this point, the vsyscall page should never be executable.