This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 23/27] [AARCH64] delouse input arguments in system functions
- From: Andrew Pinski <pinskia at gmail dot com>
- To: Arnd Bergmann <arnd at arndb dot de>
- Cc: Joseph Myers <joseph at codesourcery dot com>, Yury Norov <ynorov at caviumnetworks dot com>, GNU C Library <libc-alpha at sourceware dot org>, LKML <linux-kernel at vger dot kernel dot org>, Catalin Marinas <catalin dot marinas at arm dot com>, Marcus Shawcroft <marcus dot shawcroft at arm dot com>, philb at gnu dot org, David Miller <davem at davemloft dot net>, Szabolcs Nagy <szabolcs dot nagy at arm dot com>, Maxim Kuvyrkov <maxim dot kuvyrkov at linaro dot org>
- Date: Tue, 21 Jun 2016 09:37:47 -0700
- Subject: Re: [PATCH 23/27] [AARCH64] delouse input arguments in system functions
- Authentication-results: sourceware.org; auth=none
- References: <1466485631-3532-1-git-send-email-ynorov at caviumnetworks dot com> <1466485631-3532-25-git-send-email-ynorov at caviumnetworks dot com> <alpine dot DEB dot 2 dot 20 dot 1606211035510 dot 4526 at digraph dot polyomino dot org dot uk> <4477439 dot hqSlFvdEBH at wuerfel>
On Tue, Jun 21, 2016 at 8:42 AM, Arnd Bergmann <arnd@arndb.de> wrote:
> On Tuesday, June 21, 2016 10:36:53 AM CEST Joseph Myers wrote:
>> On Tue, 21 Jun 2016, Yury Norov wrote:
>>
>> > Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
>>
>> You're missing a patch description. What does "delouse" even mean? What
>> is the ABI peculiarity that means there are ABI-conforming arguments to
>> these functions that need such a manipulation?
>>
>
> This is the term the kernel uses for making sure that no system call
> from user space passes data in the upper bits of the argument registers,
> which could end up being used in an exploit when the calling conventions
> between functions assume that the upper bits contain zeroes.
>
> I don't think there is any point in doing this in glibc though: we
> can safely assume that any application calling into glibc follows
> the documented calling conventions (it would otherwise be a bug),
> but the kernel still has to zero those registers because malicious
> user space code would simply execute the system call instruction
> directly instead of calling into glibc...
The documented abi leaves the top 32bits undefined which is why the
zeroing is needed in these cases. Maybe a different name is needed
here.
It is the same reason why it is needed for vdso assembly code. If
these functions were written in C, the compiler would be emitting the
zero extend for you.
Thanks,
Andrew
>
> Arnd