Re: Detecting IP_TOS (and IPPROTO_IP vs. SOL_IP)

[Eric Blake]

[moving to gnulib, replies can drop autoconf]

On 03/28/2010 03:54 PM, Philip Prindeville wrote:
>>> Calling the sequence:
>>>
>>> setsockopt(fd, XXXX, IP_TOS,&foo, sizeof(foo));
>>>
>>> can be tricky.  Linux uses SOL_IP for XXXX, whereas Solaris and BSD use
>>> IPPROTO_IP for this instead.
>>>
>>> Not sure what MacOS or HP-UX do.
>>>
>>> Do the newer versions of autoconf have canned code for detecting what to
>>> use?
>>>      
>> No.
>>
>> My version of Linux defines both SOL_IP and IPPROTO_IP to the same
>> value.  Since changing it would seem to break the API, I *think*
>> if looking at Linux alone you should be able to just go with IPPROTO_IP.
>> Untested, you get to keep the pieces if it breaks.
>>    
> 
> Except, of course, that most people widely agree that using the protocol
> number as a synonym for the layer number was a poor design choice, and
> one that was eventually corrected (via the introduction of SOL_IP).
> 
> Thus I would argue the opposite: that always using SOL_IP is the correct
> approach, i.e.
> 
> #ifndef SOL_IP
> #define SOL_IP IPPROTO_IP
> #endif
> ...
> 
>     r = setsockopt(s, SOL_IP,&qos, sizeof(qos));
> 
> 
> In summary, the API was already broken, and propagating that usage model
> is probably not a good thing.
> 
> 
> 
>>> And if not, can someone please add it?
>>>
>>> I just grepped the 2.65 tarball but didn't find anything in there.
>>>      
>> Generally, for such issues it might be a good idea to just use gnulib if
>> you can; but I just looked and it doesn't address this particular issue
>> yet.

Anyone want to tackle making gnulib give better SOL_IP support, for
those platforms that don't yet implement it?

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature