IPv6 breakage in recent changes - Re: cvs commit: src/sys/netinet tcp_syncache.c

Andre Oppermann andre at freebsd.org
Mon Mar 19 09:15:34 UTC 2007


Kip Macy wrote:
>> With Andre's most recent changes what is the result?
>>
>> Also, is this all over localhost or?
> 
> Andre's commit fixed the problem. Looking at the fix the problem
> couldn't have been isolated to localhost.

My original patch had two problems in syncache that broke IPv6
independently.  The first was a C error where I missed a 'else'
in the #ifdef IPv6 mess and the second got the offset for the
IPv6 checksum calculation wrong because I missed to adjust it
to the new realities.  Tor's quickly diagnosed the problem and
sent a patch which, due to some scheduling mismatch, I committed
only with delay.  I'm sorry for the extended breakage period this
caused.

I thought I had tested IPv6 as well but as it turns out I have
changed /etc/hosts to use IPv4 first instead of IPv6 as is the
default.  We should think of doing this in -current for 7.0 as
well.  IPv6 is quite a bit slower than IPv4 still and everything
that uses localhost is affected by it.

-- 
Andre



More information about the cvs-src mailing list