Re: [PATCHv2] x86/mm: set x32 syscall bit in SET_PERSONALITY()

[Dmitry Safonov <dsafonov@virtuozzo.com>]

On 03/21/2017 10:31 PM, Andy Lutomirski wrote:
> On Tue, Mar 21, 2017 at 11:09 AM, Dmitry Safonov <dsafonov@virtuozzo.com> wrote:
>> On 03/21/2017 08:45 PM, Andy Lutomirski wrote:
>>>
>>> On Tue, Mar 21, 2017 at 10:17 AM, Cyrill Gorcunov <gorcunov@gmail.com>
>>> wrote:
>>>>
>>>> On Tue, Mar 21, 2017 at 07:37:12PM +0300, Dmitry Safonov wrote:
>>>> ...
>>>>>
>>>>> diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
>>>>> index d6b784a5520d..d3d4d9abcaf8 100644
>>>>> --- a/arch/x86/kernel/process_64.c
>>>>> +++ b/arch/x86/kernel/process_64.c
>>>>> @@ -519,8 +519,14 @@ void set_personality_ia32(bool x32)
>>>>>               if (current->mm)
>>>>>                       current->mm->context.ia32_compat = TIF_X32;
>>>>>               current->personality &= ~READ_IMPLIES_EXEC;
>>>>> -             /* in_compat_syscall() uses the presence of the x32
>>>>> -                syscall bit flag to determine compat status */
>>>>> +             /*
>>>>> +              * in_compat_syscall() uses the presence of the x32
>>>>> +              * syscall bit flag to determine compat status.
>>>>> +              * On the bitness of syscall relies x86 mmap() code,
>>>>> +              * so set x32 syscall bit right here to make
>>>>> +              * in_compat_syscall() work during exec().
>>>>> +              */
>>>>> +             task_pt_regs(current)->orig_ax |= __X32_SYSCALL_BIT;
>>>>>               current->thread.status &= ~TS_COMPAT;
>>>>
>>>>
>>>> Hi! I must admit I didn't follow close the overall series (so can't
>>>> comment much here :) but I have a slightly unrelated question -- is
>>>> there a way to figure out if task is running in x32 mode say with
>>>> some ptrace or procfs sign?
>>>
>>>
>>> You should be able to figure out of a *syscall* is x32 by simply
>>> looking at bit 30 in the syscall number.  (This is unlike i386, which
>>> is currently not reflected in ptrace.)
>>
>>
>> The process could be stopped with PTRACE_SEIZE and I think, it'll not
>> have x32 syscall bit at that moment.
>>
>> I guess the question comes from that we're releasing CRIU 3.0 with
>> 32-bit C/R and some other cool stuff, but we don't support x32 yet.
>> As we don't want release a thing that we aren't properly testing.
>> So for a while we should error on dumping x32 applications.
>
> I'm curious: shouldn't x32 CRIU just work?  What goes wrong?

I also think, it should be quite easy to add, as we have arch_prctl() 
for vdso and etc.
But there are things, which will not work if we just dump application
as 64-bit.

For example, what comes to mind: sys_get_robust_list(), it has different 
pointers for 64-bit or for x32/ia32 applications: robust_list
and compat_robust_list. So during C/R we should sometimes call
compatible syscalls for x32 applications to dump/restore, as for futex
list e.g., native will return NULL or empty list.

>
> --Andy
>


-- 
              Dmitry