lkml.org 
[lkml]   [2018]   [Apr]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectQ: Do si_time and si_utime need to be 64bit for y2038?

Arnd,

I am looking at the siginfo si_utime and si_stime fields of type clock_t
on 32bit architectures except for x32 these are 32bit fields. For y2038
do we want to extend these fields to 64bit like x32 does? Or is it not
a problem for these fields to be 32bit?

I care right now because I am trying to figure out how
copy_siginfo_to_user32 and copy_siginfo_to_user need to evolve.

If we are going to extend existing architectures with 64bit variations
of si_utime and si_stime copy_siginfo_to_user and copy_siginfo_to_user32
needs an additional parameter describing which variant they should be
copying.

It looks like posix does not define si_stime and and si_utime so we only
have to be backwards compatible with ourselves for whatever that is
worth.

I am wondering if perhaps the general solution might be to just add
two extra fields si_stime64 and si_utime64 and always fill those in.

Arnd do you have any ideas?


Andy Lutomirski <luto@amacapital.net> writes:

>> On Apr 10, 2018, at 6:26 PM, Eric W. Biederman <ebiederm@xmission.com> wrote:
>>
>>
>> Andy,
>>
>> I am looking at copy_siginfo_to_user32 and find it very unfortunate
>> that x86 with _sigchld_x32 needs to be the odd man out. I am looking
>> at ways to simplify the special case.
>>
>> The core of the special case comes from:
>> exit_to_usermode_loop
>> do_signal
>> handle_signal
>> setup_rt_frame
>>
>>
>> In setup_rt_frame the code looks at ksig to see which kind of signal
>> frame should be written for the signal.
>>
>> This leads to the one case in the kernel where copy_siginfo_to_user32
>> does not use is_ia32_syscall() or is_x32_syscall() to see which kind of
>> signal frame it needs to create.
>>
>> Andy, since you have been all over the entry point code in recent years
>> do you know if we allow tasks that can do both ia32 and x86_64 system
>> calls? That seems to be what we the testing of ksig to see which kind
>> of signal frame to setup is all about.
>
> We do :(
>
>> If we don't allow mixed abi's on x86_64 then can I see if I have a ia32
>> task in setup_rt_frame by just calling is_ia32_syscall()?
>>
>> If we do allow mixed abi's do you know if it would be safe to
>> temporarily play with orig_ax or current_thread_info()->status?
>
> Maybe, but it’s a real minefield. I think the right fix is to use
> sa_flags's SA_X32_ABI bit instead for the sigchld bit. In general,
> the is_..._syscall() helpers can't be expected to return anything
> valid in any context other than a syscall, and handle_signal() is not
> a syscall.
>
>>
>> My goal is to write two wrappers: copy_siginfo_to_user32_ia32, and
>> copy_siginfo_to_user32_x32 around the ordinary copy_siginfo_to_user32.
>> With only a runtime test to see which ABI we need to implement.
>>
>> Aka change:
>>> case SIL_CHLD:
>>> to->si_pid = from->si_pid;
>>> to->si_uid = from->si_uid;
>>> to->si_status = from->si_status;
>>> #ifdef CONFIG_X86_X32_ABI
>>> if (x32_ABI) {
>>> to->_sifields._sigchld_x32._utime = from->si_utime;
>>> to->_sifields._sigchld_x32._stime = from->si_stime;
>>> } else
>>> #endif
>>> {
>>> to->si_utime = from->si_utime;
>>> to->si_stime = from->si_stime;
>>> }
>>> break;
>> to something like:
>>> case SIL_CHLD:
>>> to->si_pid = from->si_pid;
>>> to->si_uid = from->si_uid;
>>> to->si_status = from->si_status;
>>> #ifdef CONFIG_X86_X32_ABI
>>> if (!is_ia32_syscall()) {
>>> to->_sifields._sigchld_x32._utime = from->si_utime;
>>> to->_sifields._sigchld_x32._stime = from->si_stime;
>>> } else
>>> #endif
>>> {
>>> to->si_utime = from->si_utime;
>>> to->si_stime = from->si_stime;
>>> }
>>> break;
>>
>
> Makes sense, but can you get to sa_flags in there instead?

Almost. copy_siginfo_to_user32 is called in 3 places: setup_rt_frame32
(or whatever the arch names the function for setting up the 32bit signal
frame), ptrace, and compat_binfmt_elf.

So except for ptrace and compat_binfmt_elf sa_flags are available so
that is a possibility. And for those we can fake something up if
needed.

Stepping back it really looks like the question is really do
we want/need 64bit time in siginfo for 32bit architectures to
make the code y2038 safe?

If so passing an extra parameter to copy_siginfo_to_user32 and
copy_siginfo_to_user is a no-brainer. If not we are at x86 and
in particular x32 is weird. So I am asking Arnd above if he
has any idea which way things should evolve.

> FWIW, I have a branch here:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/luto/linux.git/log/?h=execve
>
> that contains a *massive* cleanup of some other x86 signal stuff. I
> need to dust it off and test it better.

It looks interesting, and except for the last patch "Drop the separate
compat signal delivery code" looks orthogonal to what I am doing.

What I have seen other architectures do in that last case are instead of
#ifdefs to #define functions to their compat counterparts on 64bit.
Something like:
#define copy_siginfo_to_user copy_siginfo_to_user32

Eric

\
 
 \ /
  Last update: 2018-04-11 18:13    [W:0.062 / U:1.168 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site