[150714] in North American Network Operators' Group
Re: dns and software, was Re: Reliable Cloud host ?
daemon@ATHENA.MIT.EDU (David Conrad)
Thu Mar 1 14:28:49 2012
From: David Conrad <drc@virtualized.org>
In-Reply-To: <4F4FB0EC.9060807@unfix.org>
Date: Thu, 1 Mar 2012 11:27:51 -0800
To: Jeroen Massar <jeroen@unfix.org>
Cc: Nanog list <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
Jeroen,
On Mar 1, 2012, at 9:25 AM, Jeroen Massar wrote:
>> I always thought the right way to deal with IPv6 would have been to
>> use a 32-bit number from the class E space as a 'network handle'
>> where the actual address (be it IPv4 or IPv6) was handled by the
>> kernel.
>=20
> This is the case when you pass in a sockaddr. Note, not a sockaddr_in =
or
> a sockaddr_in6, but just a sockaddr.
Sorry? On which system? As far as I'm aware, there are no libraries =
that make use of class E addresses to act as a layer of indirection =
similar to file handles. Would love to know such exists.
> There is a nice 14 year old article about this:
> http://www.kame.net/newsletter/19980604/
Quoting from that article: "This way the network address and address =
family is will not live together, and leads to bunch of if/switch =
statement and mistakes in programming. " which is exactly the point. It =
has been 14 years and people are _STILL_ discussing this.
> And actually those addresses
> are only 'copied' once: during accept() or connect(),
Assuming the application doesn't need to copy the address, ever.
> Ever heard of sockaddr_storage?
Oddly, yes. It still astonishes me that sizeof(struct sockaddr) < =
sizeof(struct sockaddr_storage).
> It was made to solve that little issue. See also, that article above.
Thus requiring people to go in and muck with code thereby increasing the =
cost of migration with obvious effect.
> What you are talking about is an API wrapper. Depending on platform
> these have existed for years already. Quite a few do not expose
> addresses at all to the calling code.
And yet, look at the code Mark Andrews just referenced as his recommend =
way of dealing with initiating connections. How many applications =
actually do anything like that? More to the point, how many =
books/article/etc. exist that reference these APIs you're talking about =
vs. how many reference the traditional way one goes about dealing with =
networks?
Rhetorical questions, no need to answer. Got tired of tilting at this =
windmill some time ago and I know nothing will change. I'm just amazed =
that people defend the abominable kludge that are the existing common =
sockets/resolver APIs.
Regards,
-drc
