Re: Obscure crashes due to gcc 4.9 -O2 => -fisolate-erroneous-paths-dereference

[Florian Weimer <fweimer@redhat.com>]

On 02/20/2015 12:43 PM, Jonathan Wakely wrote:
> On 20 February 2015 at 11:06, Florian Weimer wrote:
>> On 02/19/2015 09:56 PM, Sandra Loosemore wrote:
>>> Hmmmm,  Passing the additional option in user code would be one thing,
>>> but what about library code?  E.g., using memcpy (either explicitly or
>>> implicitly for a structure copy)?
>>
>> The memcpy problem isn't restricted to embedded architectures.
>>
>>   size_t size;
>>   const unsigned char *source;
>>   std::vector<char> vec;
>>   …
>>   vec.resize(size);
>>   memcpy(vec.data(), source, size);
>>
>> std::vector<T>::data() can return a null pointer if the vector is empty,
>> which means that this code is invalid for empty inputs.
>>
>> I think the C standard is wrong here.  We should extend it, as a QoI
>> matter, and support null pointers for variable-length inputs and outputs
>> if the size is 0.  But I suspect this is still a minority view.
> 
> I'm inclined to agree.
> 
> Most developers aren't aware of the preconditions on memcpy, but GCC
> optimizes aggressively based on those preconditions, so we have a
> large and potentially dangerous gap between what developers expect and
> what actually happens.

Maybe we can add, as a compromise, an always-inline wrapper like this?

void *memcpy(void *dst, const *void src, size_t size)
{
  if (__builtin_constant_p(size > 0) && size > 0) {
    // Or whatever else is needed as non-null assertions.
    *(char *)dst;
    *(const char *)src;
  }
  return memcpy_real(dst, src, size); // Without non-null assertion.
}

Then we'll still get the non-NULL optimization for the common positive
size case.

-- 
Florian Weimer / Red Hat Product Security