Re: [tip:x86/urgent] x86/entry/32: Add an ASM_CLAC to entry_SYSENTER_32

[Andy Lutomirski]

On Thu, Feb 25, 2016 at 12:07 AM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> [resend -- thank you Gmail for sucking]
>
> On Wed, Feb 24, 2016 at 10:00 PM, H. Peter Anvin <hpa@xxxxxxxxx> wrote:
>> On 02/24/16 21:53, tip-bot for Andy Lutomirski wrote:
>>> Commit-ID:  04d1d281dcfe683a53cddfab8371fc8bb302b069
>>> Gitweb:     http://git.kernel.org/tip/04d1d281dcfe683a53cddfab8371fc8bb302b069
>>> Author:     Andy Lutomirski <luto@xxxxxxxxxx>
>>> AuthorDate: Tue, 23 Feb 2016 13:19:29 -0800
>>> Committer:  Ingo Molnar <mingo@xxxxxxxxxx>
>>> CommitDate: Wed, 24 Feb 2016 08:43:04 +0100
>>>
>>> x86/entry/32: Add an ASM_CLAC to entry_SYSENTER_32
>>>
>>> Both before and after 5f310f739b4c ("x86/entry/32: Re-implement
>>> SYSENTER using the new C path"), we relied on a uaccess very early
>>> in the SYSENTER path to clear AC.  After that change, though, we can
>>> potentially make it all the way into C code with AC set, which
>>> enlarges the attack surface for SMAP bypass by doing SYSENTER with
>>> AC set.
>>>
>>> Strengthen the SMAP protection by addding the missing ASM_CLAC right
>>> at the beginning.
>>>
>>
>> Hmmm... this potentially adds a *lot* of unnecessary cycles to this
>> path.  Could we reinstate the early uaccess?
>>
>
> I think that's more trouble than it's worth, and it'll undo a bunch of
> the context tracking cleanups that deferring it made possible,
> especially since this only matters in a configuration (32-bit SMAP)
> that no one uses. [1]
>
> *However*, I just realized that I have no idea why the 32-bit sysenter
> path is safe against NT being set.  I fixed it on compat, and now I'm
> confused as to the status on 32-bit.  If we need to fix up NT, I think
> we can fold AC into that.
>

It is, indeed, broken.  My test case doesn't notice because
opportunistic sysexit papers over the issue.  Grump.

> Also, I'll try to benchmark this soon.
>
> [1] http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=de9e478b9d49f3a0214310d921450cf5bb4a21e6
> (it didn't even boot through most of 4.5-rc)



-- 
Andy Lutomirski
AMA Capital Management, LLC