Hi Samita and Tatsuya,
From: JINMEI Tatuya / 神明達哉 <[EMAIL PROTECTED]>
Date: Mon, 25 Jul 2005 16:34:51 +0900
> >>>>> On Fri, 22 Jul 2005 17:18:20 -0700 (PDT),
> >>>>> Samita Chakrabarti <[EMAIL PROTECTED]> said:
>
> > The idea is to use both IPV6_PREFER flags in conjunction with corresponding
> > AI_PREFER_* flags in getaddrinfo().
>
> > RFC3484 Destination address selection rule #8 ( prefer smaller scope):
>
> > Thus, IPV6_PREFER_DST_SMALLSCOPE is the default setting for a system.
>
> > One example might be that if sender wants to use global destination
> > address to talk to an onlink node for which it knows a smaller scope
> > address. In that situation, setting LARGESCOPE
> > socket option and passing AI flag to getaddrinfo() is recommended.
>
> For which socket are you intending to set the LARGESCOPE socket
> option? Could you show an example code fragment that uses this option
> (e.g., containing calls to getaddrinfo(), socket(), and connect() or
> sendto())?
I had the same feeling with Tatsuya.
For the particular case, that you quated, maybe we don't need
IPV6_PREFER_DST_ option. If were you, I would write the following code:
hints.ai_family = PF_UNSPEC;
hints.ai_socktype = SOCK_STREAM;
hints.ai_flags = AI_PREFER_SRC_LARGESCOPE|AI_PREFER_DST_LARGESCOPE;
getaddrinfo("www.foobar.com", "http", &hints, &res0);
for every res in res0 {
s = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
setsockopt(s, IPPROTO_IPV6, IPV6_PREFER_SRC_LARGESCOPE=on);
connect(s, res->ai_addr, res->ai_addrlen);
...
}
I think specifying IPV6_PREFER_DST_XXX doesn't have any meaning,
because the destination address has already been chosen in
getaddrinfo().
> > Now, for DST flags, it may be possible to get away with AI_PREFER_* flags,
> > but, implementations may vary in implementing getaddrinfo() for collecting
> > potential source addresses (RFC3484 section 8).
> > Hence for portability and consistency reason, both setsockopt() and AI
> > flags
> > are recommended for destination scope.
I agree that AI_PREFER_DST_XXX is necessary to give a user a chance to
choose preferred destination address based on the conditions you listed
in the draft.
> > From implementation point of view do you think IPV6_PREFER_DST_*SCOPE
> > is redundant across all systems ?
>
> I don't know about "all" systems, but I cannot just imagine meaningful
> usage for this option. And, if no one can provide concrete useful
> usage, I don't think it a good idea to provide such a meaningless
> option just for "consistency".
---
Keiichi SHIMA
IIJ Research Laboratory <[EMAIL PROTECTED]>
KAME Project <[EMAIL PROTECTED]>
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
