dhcpcd-discuss

Graham,
you're not alone ;) and you have been bitten by the confusion around that IPv6 isn't IPv4 with long addresses. That's why there is RFC 5942 ... but why should I really dive into those details without reason, I only found out about it rather late.
Please forgive me for talking about architecture and politics, but this has helped me understanding what really is going on...
It's actually quite funny that Microsoft got this part of IPv6 right from the start ... as I guess that's because they had several people heavily involved in the architectural work at the IETF and ensured timely implementation. So the joke actually is on the Linux community who got it wrong.
So please ignore any /xxx prefix information in an IPv6 address table -- from the architecture it doesn't belong there and might be even in conflict. IPv6 addresses are nothing without (subnet) IPv6 routes. Of course, that causes quite some mishaps and unexpected failures when migrating from IPv4, and people are scratching their heads what the ... I could even warn people in advance, but they had to fall into the trap to take the RFC definitions seriously. And this is where Microsoft helped a lot, because in this area it closely follows the IPv6 architecture.
Regards,Harald

-------- Ursprüngliche Nachricht --------Von: Graham Breed <graham.breed@xxxxxxxxxxxxxx> Datum: 29.09.17  12:39  (GMT+01:00) An: "harald.albrecht" <harald.albrecht@xxxxxxx>, Roy Marples <roy@xxxxxxxxxxxx>, dhcpcd-discuss@xxxxxxxxxxxx Betreff: Re: [dhcpcd-discuss] DHCPv6 addresses being assigned with /128
  prefixes 
On 28/09/17 19:03, harald.albrecht wrote:

> Well, IPv6 addresses don't have prefix lengths outside the Linux kernel universe. Linux uses the prefix length attached to its addresses to autogenerate route table entries ... probably in an attempt to do IPv4 in IPv6. After all, IPv6 is just IPv4 with longer addresses ... what could probably go wrong ;) So with dhcpcd you should never see an address/128. Or am I mixing up the route table entries with the address table entries here, that is, you refer to the routing table entries?

You know, I'm wondering if I've been getting this wrong.  I assumed that
the IP addresses as shown by the "ip address" Linux command should show
/64 if I know the subnet is /64.  But it doesn't seem to matter because
the routing table knows that the address is in the /64 subnet and does
the right thing.

A simple test confirms this: I make two DHCPV6-configured clients and I
can get an SSH connection from one to the other with no problem.  The
only apparent issue is the output of "ip address".

> From my experience with dhcpcd it correctly creates the required route table entries. And the IPv6 address table just contains /128 addresses, but not any prefix information. But I'm also dazzled why your setup doesn't work. Disclaimer: I use dhcpcd for PD and SLAAC only, never had a need to deploy IA-NAs.

My experience of dhcpcd so far has been that it wasn't working with
DHCPv6 because we didn't have the router advertisements correct.  My
experience with DHCPv6 was, therefore, with the older versions of
dhclient that used a /64 in violation of the RFC.  So my expectations
were that a /64 should come back (as it does with SLAAC).  But maybe my
expectations are wrong and this is the way it's supposed to be.

It does make some sense, in that the routing table might be changed
after the DHCPv6 address is assigned.  Choosing a prefix based on the
routing table at the point the address was assigned doesn't guarantee
that the prefix will be valid for the whole life of the address.  I
think I set the lifetime of the router advertisement to be longer than
the lifetime of the DHCPv6 lease, but maybe I missed something, or maybe
dhcpcd doesn't care.


                 Graham


> Best regards,Harald
> 
> -------- Ursprüngliche Nachricht --------Von: Graham Breed <graham.breed@xxxxxxxxxxxxxx> Datum: 28.09.17  15:17  (GMT+01:00) An: Roy Marples <roy@xxxxxxxxxxxx>, dhcpcd-discuss@xxxxxxxxxxxx Betreff: Re: [dhcpcd-discuss] DHCPv6 addresses being assigned with /128
>   prefixes 
> On 28/09/17 13:43, Roy Marples wrote:
> 
>> ISC dhclient **only** does DHCPv6. It doesn't handle any RA recieved
>> inforamtion such as prefixes, not will it generate any routes.
>> AFAIK, no patches for it do either - instead a 3rd party will do this.
>> NetworkManager, systemd, Linux kernel (if dhcpcd hasn't turned that off).
>>
>> So what you say about dhclient not receiving SLAAC addressess makes
>> perfect sense.
> 
> The interface is configured to receive the addresses, however Ubuntu
> does that.  This is a separate test VM without dhcpcd.  And SLAAC works
> with a "route" section in radvd.conf along with the "prefix" (or maybe
> it just takes a while to apply).  But DHCPv6 is still giving /128
> addresses in both cases.
> 
> So I'm still mystified in that both implementations think there's
> something wrong, but I don't know what it is.
> 
>> dhcpcd is the only all-in-one solution :)
> 
> Indeed, and we find it useful to use it like that to control all dynamic
> address and route assignment.  And it makes this all the more
> perplexing, because if dhcpcd can use the prefix length to assign a
> stateless address, why not use the same prefix length for the DHCPv6
> address?  It wouldn't attempt to get the address at all if it hadn't
> received the correct advertisement.
> 
> 
>> dhcpcd is saying that a 3rd party deleted the address.
> 
> In that case, I know what the 3rd party is, so never mind this detail.
> 
> 
>                    Graham
> 

-- 
------------------------------

*Graham Breed*
Systems Developer, ApplianSys

ApplianSys House
Harry Weston Road
Coventry CV3 2UB

www.appliansys.com

*Our sales team sells by referral:*
Less time looking for customers, more time looking after them

ApplianSys Limited registered in England & Wales #: 07608382 at ApplianSys
House, Harry Weston Road, Coventry CV3 2UB