dhcpcd-discuss

On 12/11/2017 07:49, Neal P. Murphy wrote:
> After some time figuring out how to deal with dnsmasq, I didn't see substantive differences betweeen RC3 and HEAD. Both handle addresses and delegations with aplomb.

Yay

>    - Should dhcpcd remove the internet IF's IP6 address on '-k'? It does remove the internal
>      IF's addresses. Or, more generally, should it remove all addresses it assigned when it exits?

-k forces a lease drop and exit.
This has the same effect as using -x and not having the persistent 
keyword in dhcpcd.conf.

All addresses, except for IPv6 LL addresses should be removed.
This has been improved in HEAD from RC3.

> General iaid questions:
>    - What are the limitations on iaids? Short-ish numbers or strings? Long-ish strings?

32 bit unsigned int.
In dhcpcd, you can express it as a number or hex string.

>    - Is every interface-address combination supposed to have a unique iaid? That is, if my
>      eth0 had 5 addresses assigned, should it have 5 iaids? And if my eth3 has a /128 and
>      a /64 assigned, should it have two iaids? Is there clear documentation for this anywhere
>      (your docs, RFCs, ec.)?

Yes, it's supposed to be unique for every address requested per protocol.
So IPv4 and DHCPv6 can share the default (based on mac address).
For all others (Prefix Delegation, a second address or temporary 
address) you should roll your own.

But note this is different from Prefix Delegation assignment to 
downstream interfaces. Here we use a SLA to append to the leased address 
to make it unique per downstream interface internally to dhcpcd.

> Now that I'm correctly telling dnsmasq to add/remove IP6 DNS servers, bringing the links up and down is fairly reliable. Comcrash *still* like to give me a /64 instead of a /60 now and again. I guess they don't like something I save, recover and dynamically include in the dhcpcd.conf; removing those files and starting from scratch seems to get it working again. Query? Would it be reasonable to have dhcpcd reject the /64 offer when it explicitly requested a /60? (Instead of taking it, assigning it to the first interface--eth3 in my case--and then failing to assign anything to the other IFs?) Of course, I have to verify that that is really what happens.

Well, that could still be a bug in dhcpcd.
Can you email a full tcpdump of the DHCPv6 transations please?
Binary, so I can see em in wireshark.

Thanks

Roy