Re: [PATCH 4.9 00/39] 4.9.75-stable review

[Andy Lutomirski]

On Wed, Jan 3, 2018 at 8:07 PM, Hugh Dickins <hughd@xxxxxxxxxx> wrote:
> On Wed, Jan 3, 2018 at 5:24 PM, Ben Hutchings <ben@xxxxxxxxxxxxxxx> wrote:
>> On Wed, 2018-01-03 at 21:11 +0100, Greg Kroah-Hartman wrote:
>>> This is the start of the stable review cycle for the 4.9.75 release.
>>> There are 39 patches in this series, all will be posted as a response
>>> to this one.  If anyone has any issues with these being applied, please
>>> let me know.
>> [...]
>>
>> This seems to be missing a backport of:
>>
>> commit 85900ea51577e31b186e523c8f4e068c79ecc7d3
>> Author: Andy Lutomirski <luto@xxxxxxxxxx>
>> Date:   Tue Dec 12 07:56:42 2017 -0800
>>
>>     x86/pti: Map the vsyscall page if needed
>>
>> without which, KPTI will break some older user-space.
>
> You're (kind of) right: I did warn Greg of that yesterday, and we were
> both vaguely hoping that someone else could fill in that blank more
> quickly than I'd get there.
>
> I doubt it's an actual backport of Andy's 85900ea51577 that's needed -
> I'd expect his to depend on at least some of the improved
> infrastructure in 4.15-rc, rather than the old-style kaiser-mapping
> employed in these backports.
>
> Andrea has a RHEL7 patch for this, and I have a 3.2 patch derived from
> Andrea's, but neither would be right for 4.9 or 4.4 as is: I think I'm
> going to have to drop today's priority, and attend to this lacuna
> tomorrow.
>

I think what you need to backport is:

"x86/vsyscall/64: Explicitly set _PAGE_USER in the pagetable
hierarchy" to get the permission bits right

and maybe the big simplification of
kaiser_whatever_its_called_set_pgd().  The latter is going to be a
mess if you're still using the old code.  Sorry.  But even with the
crappy old code, kaiser_add_mapping() on the vsyscall page plus the
latter _PAGE_USER bit plus the call to the new _PAGE_USER code in
"x86/pti: Map the vsyscall page if needed" should do it.

If you skip all this and only get vsyscall emulation working such that
vsyscall page *reads* don't work from user mode, you're going to
encounter weird problems with programs like "pin".  Apparently people
use these things.