[RFC] syscalls: Restore address limit after a syscall

[Andy Lutomirski]

On Thu, Feb 9, 2017 at 3:41 PM, Thomas Garnier <thgarnie at google.com> wrote:
> On Thu, Feb 9, 2017 at 3:05 PM, Andy Lutomirski <luto at amacapital.net> wrote:
>> On Thu, Feb 9, 2017 at 11:31 AM, Kees Cook <keescook at chromium.org> wrote:
>>> On Thu, Feb 9, 2017 at 10:33 AM, Thomas Garnier <thgarnie at google.com> wrote:
>>>> This patch prevents a syscall to modify the address limit of the
>>>> caller. The address limit is kept by the syscall wrapper and restored
>>>> just after the syscall ends.
>>>>
>>>> For example, it would mitigation this bug:
>>>>
>>>> - https://bugs.chromium.org/p/project-zero/issues/detail?id=990
>>>>
>>>> Signed-off-by: Thomas Garnier <thgarnie at google.com>
>>>> ---
>>>> Based on next-20170209
>>>> ---
>>>>  include/linux/syscalls.h | 5 ++++-
>>>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
>>>> index 91a740f6b884..a1b6a62a9849 100644
>>>> --- a/include/linux/syscalls.h
>>>> +++ b/include/linux/syscalls.h
>>>> @@ -198,7 +198,10 @@ extern struct trace_event_functions exit_syscall_print_funcs;
>>>>         asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__));      \
>>>>         asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__))       \
>>>>         {                                                               \
>>>> -               long ret = SYSC##name(__MAP(x,__SC_CAST,__VA_ARGS__));  \
>>>> +               long ret;                                               \
>>>> +               mm_segment_t fs = get_fs();                             \
>>>> +               ret = SYSC##name(__MAP(x,__SC_CAST,__VA_ARGS__));       \
>>>> +               set_fs(fs);                                             \
>>>>                 __MAP(x,__SC_TEST,__VA_ARGS__);                         \
>>>>                 __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__));       \
>>>>                 return ret;                                             \
>>>> --
>>>> 2.11.0.483.g087da7b7c-goog
>>>>
>>>
>>> I have a memory of Andy looking at this before, and there was some
>>> problem with how a bunch of compat code would set fs and then re-call
>>> the syscall... but I can't quite find the conversation. Andy, do you
>>> remember the details?
>>>
>>> This seems like an entirely reasonable thing to enforce for syscalls,
>>> though I'm sure there's a gotcha somewhere. :)
>>
>> This sounds vaguely familiar, but that's about all.
>>
>> Anyway, it seems reasonable that the SyS_foobar wrappers are genuinely
>> only used for syscalls and not for other things, so the code should
>> *work*.  That being said, I think there's room for several
>> improvements.
>>
>> 1. Why save the old "fs" value?  For that matter, why restore it?
>> IOW, I'd rather see BUG_ON(get_fs() != USER_DS) at the end.
>>
>
> I guess that make sense in the wrapper.
>
>> 2. I'd rather see the mechanism be more general.  If we had, effectively:
>>
>> asmlinkage long SyS_foo(...) {
>>   sys_foo();
>>   verify_pre_usermode_state();
>> }
>>
>> and let verify_pre_usermode_state() potentially do more things, we'd
>> get a more flexible mechanism.  On arches like x86_32, we could save a
>> decent amount of code size by moving verify_pre_usermode_state() into
>> prepare_exit_to_usermode(), but that would have to be a per-arch
>> opt-in.  x86_64 probably would *not* select this due to the fast path
>> (or it would do it in asm.  hmm.).
>>
>
> I will look into that. I like this design better.
>
>> 3. If this thing gets factored out, then arch code can call it for
>> non-syscall entries, too.
>>
>
> Yes, it makes sense.
>
>> 4. Can we make this configurable?
>>
>>
>> For x86, a nice implementation might be:
>>
>> select ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
>>
>> ... in prepare_exit_to_usermode():
>>
>> verify_pre_usermode_state();  // right at the beginning
>>
>> ... in the asm syscall fast path:
>>
>> #ifdef CONFIG_VERIFY_PRE_USERMODE_STATE
>> call verify_pre_usermode_staet
>> #endif
>>
>> (or just inline the interesting bit)
>>
>
> So by default it is in the wrapper. If selected, an architecture can
> disable the wrapper put it in the best places. Understood correctly?

Sounds good to me.

Presumably the result should go through -mm.  Want to cc: akpm and
linux-arch@ on the next version?

I've also cc'd arm and s390 folks -- those are the other arches that
try to be on top of hardening.