[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20151126104229.GA8530@gmail.com>
Date: Thu, 26 Nov 2015 11:42:29 +0100
From: Ingo Molnar <mingo@...nel.org>
To: PaX Team <pageexec@...email.hu>
Cc: kernel-hardening@...ts.openwall.com,
Mathias Krause <minipli@...glemail.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Kees Cook <keescook@...omium.org>,
Andy Lutomirski <luto@...capital.net>,
Ingo Molnar <mingo@...hat.com>,
Thomas Gleixner <tglx@...utronix.de>,
"H. Peter Anvin" <hpa@...or.com>, x86-ml <x86@...nel.org>,
Arnd Bergmann <arnd@...db.de>,
Michael Ellerman <mpe@...erman.id.au>, linux-arch@...r.kernel.org,
Emese Revfy <re.emese@...il.com>
Subject: Re: [PATCH 0/2] introduce post-init read-only
memory
* PaX Team <pageexec@...email.hu> wrote:
> On 26 Nov 2015 at 9:54, Ingo Molnar wrote:
>
> > * PaX Team <pageexec@...email.hu> wrote:
> >
> > > actually the kernel could silently recover from this given how the page fault
> > > handler could easily determine that the fault address fell into the
> > > data..read_only section and just silently undo the read-only property, log the
> > > event to dmesg and retry the faulting access.
> >
> > So a safer method would be to decode the faulting instruction, to skip it by
> > fixing up the return RIP and to log the event. It would be mostly equivalent
> > to trying to write to ROM (which get ignored as well), so it's a recoverable
> > (and debuggable) event.
>
> if by skipping you mean ignoring the write attempt then it's not a good idea as
> it has a good chance to cause unexpected behaviour down the line.
>
> e.g., imagine that the write was to a function pointer (even an entire ops
> structure) or a boolean that controls some important feature for after-init
> code. ignoring/dropping such writes could cause all kinds of logic bugs (if not
> worse).
Well, the typical case is that it's a logic bug to _do_ the write: the structure
was marked readonly for a reason but some init code re-runs during suspend or so.
But yes, logic bugs might trigger - but that is true in the opposite case as well,
if we do the write despite it being marked readonly:
> my somewhat related war story is that i once tried to constify machine_ops (both
> the struct and the variable of the same name) directly and just forced the
> writes in kvm/xen/etc via type casts. now i knew it was all undefined behaviour
> but i didn't expect gcc to take advantage of it but it did (const propagated the
> *initial* fptr values into the indirect calls by turning them into direct calls)
> and which in turn prevented proper reboots for guests (an event which obviously
> happens much later after init/boot to the great puzzlement of end users and
> myself).
>
> misusing __read_only and ignoring write attempts would effectively produce the
> same misbehaviour as above so i strongly advise against it.
No, the difference to the GCC related aliasing bug is that with my technique the
kernel would immediately produce a very visible kernel warning, which is a very
clear sign that is wrong - and with a very clear backtrace in the warning that
points right to the problematic code - which signature shows us (and users) what
is wrong.
So your example is not comparable at all.
Plus the truly paranoid might panic/halt the system on such warnings, so for
highly secure systems there's a way to not even allow the possibility of logic
bugs. (at the cost of stopping the system when a bug triggers.)
Thanks,
Ingo
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
