Date: Mon, 02 Jul 2001 10:15:21 -0700 From: "Kevin Oberman" <oberman@es.net> To: JINMEI Tatuya / =?ISO-2022-JP?B?GyRCP0BMQEMjOkgbKEI=?= <jinmei@isl.rdc.toshiba.co.jp> Cc: freebsd-net@FreeBSD.ORG Subject: Re: Odd IPv6 behavior when not connected to IPv6 net Message-ID: <200107021715.f62HFLc25924@ptavv.es.net> In-Reply-To: Your message of "Mon, 02 Jul 2001 15:27:26 %2B0900." <y7vr8vzkfqp.wl@condor2.jinmei.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> Date: Mon, 02 Jul 2001 15:27:26 +0900 > From: JINMEI Tatuya / =?ISO-2022-JP?B?GyRCP0BMQEMjOkgbKEI=?= <jinmei@isl.rdc.toshiba.co.jp> > Sender: owner-freebsd-net@FreeBSD.ORG > > >>>>> On Fri, 29 Jun 2001 09:28:54 -0700, > >>>>> "Kevin Oberman" <oberman@es.net> said: > > >> That is, if we do not have any default router (hear from RA), we > >> should regard all IPv6 prefixes as on-link. To implement this trick, > >> we use the "default interface", and install the default route as an > >> interface direct route to the interface. > > > I think the basic idea in the RFC may be reasonable. It only breaks > > when the link selected is loopback. Had the stack tried to connect to > > a physical link, this would have worked as intended, but loopback will > > always be the wrong answer if it is the link used. > > Just to make it sure, even if you specify a non-loopback interface as > the default, you should still (usually) see a long delay before the > connection attempt by IPv6 fails, because this type of error is not a > hard error for TCP (like "no route to host"). The delay would be > about 1 minute. I don't think most users will tolerate the delay, > especially when the IPv4 connection can be established. I absolutely agree. I have users up in arms when there is an 8 second delay due to a DNS failure. 1 minute is a complete non-starter. > >> Thus, for the moment, I agree that we should turn the default > >> interface off by default. For a longer term solution, we might have > >> to consider a better source address selection algorithm, > >> e.g. described in draft-ietf-ipngwg-default-addr-select-04.txt. Then > >> IPv4 would be preferred in this case. > > > This is certainly reasonable, too. But the step of not allowing the lo > > interface to qualify as the link for the default route seems like > > something that should be done as well. > > > Of course, there may be some reason to want default to point at lo, > > but I can't think of a good one. > > I don't have any objection to changing the default interface to a > non-loopback one, *if the default is ever defined*. I'm arguing that > it would be safe *not to specify the default interface by default*. Ah. I guess I missed your obvious point. I'm still thinking about the possible effects of not specifying a default route by default. At the moment I don't see any other "solution" to the problem. I think it's a "wrong" answer, but less "wrong" than any other answer I've seen. The real issue, from my perspective, is that nothing that discourages the use of IPv6 should be done. Default to loopback is unacceptable on both POLA and simple functional basis. (And I was VERY astonished!) A one minute delay is also unacceptable. So, we're stuck with whatever is left. I am a bit unclear why the timeout for a "local" connection should be as long as a minute, but even if it was significantly reduced, it would remain unacceptable, so I guess I won't worry about it. R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: oberman@es.net Phone: +1 510 486-8634 To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200107021715.f62HFLc25924>