Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Mon, 12 Mar 2018 09:42:23 -0700
From: Kees Cook <>
To: Ingo Molnar <>, P J P <>, 
	Florian Weimer <>
Cc: Ard Biesheuvel <>, 
	Linus Torvalds <>, Steven Rostedt <>, 
	Arnd Bergmann <>, Daniel Micay <>, 
	Dave Hansen <>, Alexander Popov <>, 
	Kernel Hardening <>, PaX Team <>, 
	Brad Spengler <>, Andy Lutomirski <>, 
	Tycho Andersen <>, Laura Abbott <>, Mark Rutland <>, 
	Borislav Petkov <>, Richard Sandiford <>, 
	Thomas Gleixner <>, "H . Peter Anvin" <>, 
	Peter Zijlstra <>, "Dmitry V . Levin" <>, 
	Emese Revfy <>, Jonathan Corbet <>, 
	Andrey Ryabinin <>, 
	"Kirill A . Shutemov" <>, Thomas Garnier <>, 
	Andrew Morton <>, Alexei Starovoitov <>, Josef Bacik <>, 
	Masami Hiramatsu <>, Nicholas Piggin <>, 
	Al Viro <>, "David S . Miller" <>, 
	Ding Tianhong <>, David Woodhouse <>, 
	Josh Poimboeuf <>, Dominik Brodowski <>, 
	Juergen Gross <>, Greg Kroah-Hartman <>, 
	Dan Williams <>, Mathias Krause <>, 
	Vikas Shivappa <>, Kyle Huey <>, 
	Dmitry Safonov <>, Will Deacon <>, X86 ML <>, 
	LKML <>
Subject: Fully initialized stack usage (was Re: [PATCH RFC v9 4/7] x86/entry:
 Erase kernel stack in syscall_trace_enter())

On Mon, Mar 12, 2018 at 2:21 AM, Ingo Molnar <> wrote:
> * Ard Biesheuvel <> wrote:
>> > Is it possible to implement this "safe automatic variable initialization" language
>> > feature via a GCC plugin robustly, while still keeping code generation sane? (i.e.
>> > no forced allocation of stack slots, etc.) It should be a superset of
>> I think that should be feasible, yes.
>> It would be worth trying whether the current code can be simplified, though: it
>> currently takes care not to add such an initialization if it can already spot
>> one, but I wonder whether just blindly adding one at the start and letting the
>> compiler optimize it away again is safer.
> Absolutely - followed by:
>  - a good look at the resulting code generation with a reasonably uptodate GCC
>  - a look at the resulting code with older GCCs, to make sure there's no
>    pathological code generation
> ... because IMHO in the end it is the practical effects that will make (or break)
> any such attempt.
> ( BTW., initializing all automatic variables might even speed up certain
>   optimization passes, FWIIW. )

[attempting to merge threads...]

On Tue, Feb 27, 2018 at 3:15 AM, P J P <> wrote:
> Please see:
>   ->
> This experimental patch by Florian Weimer(CC'd) adds an option
> '-finit-local-vars' to gcc(1) compiler. When a program(or kernel)
> is built using this option, its automatic(local) variables are
> initialised with zero(0). This could significantly reduce the kernel
> information leakage issues.
> A dnf(8) repository of the latest gcc-7.3.1 package built with the above
> patch and kernel-4.15.5 package built using '-finit-local-vars' option
> on Fedora-27 is available below
>   ->
> This same kernel is running on my F27 test machine as I write this.
> There is no slowness or notice-able performance impact as such.

Getting this implemented directly in the compiler would be preferred
over a plugin. (Prototyping may be easier in the plugin, though.)


- the kernel needs a way to "opt out" of this for places where later
functions will do the initialization. Something like
__attribute__((no_automatic_variable_init)) ?

- initialization _must include structure padding_. Without this, we're
only solving part of the exposure. Does -finit-local-vars do this?

- Can we retain the uninitialized variable usage warning? (with an
updated text; maybe "variable used without explicit initialization,
using zero-initialization"?)

It sounds like Fedora-27's default kernel build is already using this
option, is that correct?


Kees Cook
Pixel Security

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.