[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20180716101337.GA30279@gmail.com>
Date: Mon, 16 Jul 2018 12:13:38 +0200
From: Ingo Molnar <mingo@...nel.org>
To: Alexander Popov <alex.popov@...ux.com>
Cc: Kees Cook <keescook@...omium.org>,
Linus Torvalds <torvalds@...ux-foundation.org>,
Kernel Hardening <kernel-hardening@...ts.openwall.com>,
Pax Team <pageexec@...email.hu>,
Brad Spengler <spender@...ecurity.net>,
Andrew Lutomirski <luto@...nel.org>,
Tycho Andersen <tycho@...ho.ws>, Laura Abbott <labbott@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Ard Biesheuvel <ard.biesheuvel@...aro.org>,
Borislav Petkov <bp@...en8.de>,
Richard Sandiford <richard.sandiford@....com>,
Thomas Gleixner <tglx@...utronix.de>, Peter Anvin <hpa@...or.com>,
Peter Zijlstra <a.p.zijlstra@...llo.nl>,
"Dmitry V. Levin" <ldv@...linux.org>,
Emese Revfy <re.emese@...il.com>, Jonathan Corbet <corbet@....net>,
Andrey Ryabinin <aryabinin@...tuozzo.com>,
"Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
Thomas Garnier <thgarnie@...gle.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Alexei Starovoitov <ast@...nel.org>, Josef Bacik <jbacik@...com>,
Masami Hiramatsu <mhiramat@...nel.org>,
Nick Piggin <npiggin@...il.com>, Al Viro <viro@...iv.linux.org.uk>,
David Miller <davem@...emloft.net>,
dingtianhong <dingtianhong@...wei.com>,
David Woodhouse <dwmw@...zon.co.uk>,
Josh Poimboeuf <jpoimboe@...hat.com>,
Steven Rostedt <rostedt@...dmis.org>,
Dominik Brodowski <linux@...inikbrodowski.net>,
Jürgen Groß <jgross@...e.com>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Dan Williams <dan.j.williams@...el.com>,
Dave Hansen <dave.hansen@...ux.intel.com>,
Mathias Krause <minipli@...glemail.com>,
Vikas Shivappa <vikas.shivappa@...ux.intel.com>,
Kyle Huey <me@...ehuey.com>,
Dmitry Safonov <dsafonov@...tuozzo.com>,
Will Deacon <will.deacon@....com>, Arnd Bergmann <arnd@...db.de>,
Florian Weimer <fweimer@...hat.com>,
Boris Lukashev <blukashev@...pervictus.com>,
Andrey Konovalov <andreyknvl@...gle.com>,
the arch/x86 maintainers <x86@...nel.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v14 0/6] Introduce the STACKLEAK feature and a test for it
* Alexander Popov <alex.popov@...ux.com> wrote:
> On 16.07.2018 01:44, Ingo Molnar wrote:
> >
> > * Kees Cook <keescook@...omium.org> wrote:
> >
> >> On Thu, Jul 12, 2018 at 2:22 PM, Alexander Popov <alex.popov@...ux.com> wrote:
> >>> On 12.07.2018 23:50, Ingo Molnar wrote:
> >>>> Let's make sure informed users have an easy runtime way out
> >>>> from the worst of the overhead that doesn't involve "recompile your distro
> >>>> kernel". Also, make it easier to measure and fingerpoint the overhead...
> >>>
> >>> Would you like the following solution?
> >>>
> >>> I'll create the CONFIG_STACKLEAK_RUNTIME_DISABLE config option, which would be
> >>> similar to CONFIG_SECURITY_SELINUX_DISABLE. That option will provide a sysctl
> >>> switch disabling stackleak_erase(), which provides the most of performance penalty.
> >>
> >> I don't think CONFIG/sysctl is the way to go. I'd recommend making it
> >> a boot arg and using a static key, similar to what's happening to
> >> hardened_usercopy:
> >
> > Why no sysctl? It's a PITA to reboot systems just to turn a stupid knob off.
> >
> > Also, it's _much_ easier to measure performance impact when there's a sysctl.
>
> Yes, you are right.
>
> But I looked carefully and now see the troubles which sysctl would bring us.
> Each 'task_struct' has 'lowest_stack', which must be initialized before enabling
> STACKLEAK. So runtime enabling via sysctl is not plain and may bring race
> conditions.
Firstly, a sysctl could still allow it to be *disabled*, once - which is the most
important usecase of the sysctl anyway.
Secondly, in the first iteration this could be kept included unconditionally:
current->lowest_stack = current_top_of_stack() - THREAD_SIZE/64;
... which would keep it initialized and wouldn't be racy, right?
I.e. only the most expensive part of the function, the scanning, would be turned
off via the sysctl. I submit that this will avoid all measurable aspects of the 1%
kbuild performance overhead.
A more involved approach can be done in the future if warranted, and the feature
could be disabled/enabled more thoroughly - but the runtime sysctl would be
acceptable for me for now, as a first iteration.
> On the other hand, a boot param + static key that can only disable STACKLEAK
> during __init are much more robust. I'm preparing the patch and performance
> measurements.
A boot parameter does *not* address the concern I outlined. Why force admins
reboot a box just to disable something that causes overhead?
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.
