|
From: Camille T. <ca...@os...> - 2011-11-22 14:15:27
|
Hi Stephen, On 21 nov. 2011, at 20:09, Stephen Sinclair wrote: > After a month of working on an unrelated project I finally have time > to get back to this :) > Sorry for the delay. Thank you! I appreciate you spend time on this. > I've just been testing this IPV6 patch on Linux (Ubuntu 10.10) and > iMac (10.6.8), and can report the following. With PF_UNSPEC and > V6ONLY enabled on the server, if the client specifies the hostname as: > > Ubuntu10.10 msg received lo_send returned > "localhost" yes number of bytes > "::1" yes number of bytes > "127.0.0.1" no number of bytes > "::FFFF:127.0.0.1" no number of bytes > > OSX 10.6.8 > "localhost" no number of bytes > "::1" yes number of bytes > "127.0.0.1" no number of bytes > "::FFFF:127.0.0.1" no number of bytes > > With IPV6ONLY disabled, > > Ubuntu 10.10: > "localhost" yes number of bytes > "::1" yes number of bytes > "127.0.0.1" yes number of bytes > "::FFFF:127.0.0.1" yes number of bytes > > OSX 10.6.8: > "localhost" yes number of bytes > "::1" yes number of bytes > "127.0.0.1" yes number of bytes > "::FFFF:127.0.0.1" yes number of bytes > > So, I never observed lo_send() returning error 22, contrary to your > results, which is strange. It could simply be a difference between > the order of the NICs in the two machines, the wifi vs. wired > configuration, etc. (I was testing on a wired ethernet connection.) Right, I was testing using the wifi adapter. > The only difference with Ubuntu seems to be that the result of looking > up "localhost" results in IPv4 on Mac OS X but in IPv6 on Linux. I prefer the results of Ubuntu. But if you have a look at /etc/hosts on Mac OS X it does indeed have the IPv6 addresses for localhost. Oh well ... > However, it's clear that V6ONLY is working on both operating systems, > and using PF_UNSPEC is correct when "::FFFF" is prepended. Moreover I > never observed a case where the loop over addrinfo actually ran more > than once. But I see no problem in iterating and saving the results, > if that potentially solves a problem. It's easy to simply save the > result in lo_address if the loop discovers a successful addrinfo later > in the list, so it should only execute the loop once. Sounds good to me! > I think I can safely commit this patch (done) but some independent > testing would be appreciated. Thank you Steve. I'll be back in a short while with some questions about liblo and TCP. I think having IPv6 and TCP support in liblo is a real plus. Best, Cam |