I've been looking on ipv6 lately, and even though I got a global /64 for free from he.net, I'm not that amused about ipv6 yet.
- ipv6 link-local scope : if you have multiple interfaces with ipv6 link-local addresses, the operating system does not know which interface to use, so you have to append the interface to the hostname/ip when connecting hosts in link-local scope. If you do not use getaddrinfo, this information has to be passed to the bind/connect using
struct sockaddr_in6.sin6_scope_id = if_nametoindex(devicename);
This sounds weird, and it actually is:
nc6 -6 -vv fe80::21f:d0ff:fe23:9b77%eth1 80
may work for some people, but encoding the interface in url renders the whole url-idea useless
- getaddrinfo() is meant to resolve a domain for a service to its A&AAAA records, report the required family etc.
- problem: some cheap home routers drop AAAA requests, so the getaddrinfo() call -as well as the application/user- has to wait for a timeout
- check for AF_INET6 support, if ipv6 is supported, resolve AAAA
- problem: loading a module does not mean you really use ipv6, your nics always get the ipv6 link-local addresses assigned
- if hint AI_ADDRCONFIG is provided, check if we have a ipv6 address with site or global scope, if there is none, report only A/ipv4
- problem: all programm using getaddrinfo() do no work for local-link scope any longer, including simple things like netcat6
Ubuntu, has chosen to disable getaddrinfo() for ipv6 if only link-local scope addresses are availible, breaking ipv6 support for all major applications, but improving the user experience for people with b0rked routers. netcat6 on Ubuntu hardy:
nc6 -6 -vv -l -p 4711
nc6: forward host lookup failed for local endpoint [unspecified] (4711): Name or service not known
Debian had the same patch applied, but removed it due to regression. Summing up the problems with ipv6 link-local scope, it is a mess. I'll have to provide site/global ipv6 for my network to circumvent these problems, rendering link-local completely useless.