Re: [PATCH 3.5 2/2] seccomp: Future-proof against silly tracers

[Andy Lutomirski]

On Wed, Jul 18, 2012 at 11:35 AM, Will Drewry <wad@xxxxxxxxxxxx> wrote:
> On Tue, Jul 17, 2012 at 9:13 PM, Will Drewry <wad@xxxxxxxxxxxx> wrote:
>> On Tue, Jul 17, 2012 at 6:19 PM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>>> Currently, if a tracer changes a syscall nr to __NR_future_enosys,
>>> behavior will differ between kernels that know about
>>> __NR_future_enosys (and return -ENOSYS) and older kernels (which
>>> return the value from pt_regs).  This is silly; we should just
>>> return -ENOSYS.
>>>
>>> This is unlikely to ever happen on x86 because the return value in
>>> pt_regs starts out as -ENOSYS, but a silly tracer can change that.
>>>
>>> Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
>>> Cc: Will Drewry <wad@xxxxxxxxxxxx>
>>> ---
>>>  arch/x86/include/asm/syscall.h |   11 +++++++++++
>>>  kernel/seccomp.c               |   15 +++++++++++++++
>>>  2 files changed, 26 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/arch/x86/include/asm/syscall.h b/arch/x86/include/asm/syscall.h
>>> index 1ace47b..8191e057 100644
>>> --- a/arch/x86/include/asm/syscall.h
>>> +++ b/arch/x86/include/asm/syscall.h
>>
>> Since this is used in an arch-wide location, the prototype and comment
>> should be in asm-generic/syscall.h too.
>>
>>> @@ -70,6 +70,17 @@ static inline void syscall_set_return_value(struct task_struct *task,
>>>         regs->ax = (long) error ?: val;
>>>  }
>>>
>>> +static inline bool syscall_is_nr_future(struct task_struct *task,
>>> +                                       struct pt_regs *regs)
>>> +{
>>> +#ifdef CONFIG_IA32_EMULATION
>>> +       if (task_thread_info(task)->status & TS_COMPAT)
>>> +               return syscall_get_nr(task, regs) > __NR_ia32_syscall_max;
>>> +#endif
>>> +
>>> +       return syscall_get_nr(task, regs) > __NR_syscall_max;
>>
>> I'm not sure how easy this will be to implement on some of the arches
>> where this data isn't bubbled up.  It'd be good if some non-x86 arch
>> maintainers chimed in (since x86 is easy and already works as expected
>> :).
>>
>
> Since x86 always returns -ENOSYS with an invalid syscall and only x86
> supports seccomp filter in 3.5, I'd propose pushing this off for 3.6+
> to get more feedback from the relevant parties.  Not doing it now
> doesn't expose any users of 3.5 to any sort of changing ABI.

It's (barely) visible.  See VSYS.trace_changenr_high here:

https://github.com/amluto/seccomp

>
> (Otherwise, it seems fine but may make adding new arches slightly more
> onerous, but I suspect ftrace needs this sort of info too as it
> spreads to other arches!)

Are there any arches for which the return value isn't its own
register?  If so, this is necessary.

Agreed, though: let's wait for 3.6.

-- 
Andy Lutomirski
AMA Capital Management, LLC
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/