dhcpcd-discuss

On Sat, Mar 10, 2018 at 3:52 AM, Roy Marples <roy@xxxxxxxxxxxx> wrote:
> On 09/03/2018 13:13, Roy Marples wrote:
>>
>> On 08/03/2018 22:31, Brenden Blanco wrote:
>>>>
>>>> I'm running dhcpcd on a single interface in dual stack and IPv6-PD.
>>>> Archlinux recently updated to 7.0.1 which broke IPv6 for me.
>>>>
>>>> After doing git bisect I found that commit ef53a17 (DHCPv6: listen on
>>>> all
>>>> IPv6 addresses for active interface) responsible for this. Reverting it
>>>> on
>>>> master solves the issue for me.
>>>>
>>>> I'm running archlinux with systemd-networkd for everything else but the
>>>> WAN
>>>> interface.
>>>> I can provide additional information if needed.
>>>
>>>
>>> I came across this thread while googling, and I believe that I am
>>> hitting it as well. Specifically, it seems that my ISP (wave-G
>>> broadband) is ignoring any solicit requests that do not come from port
>>> 546 (packets are blackholed).
>>>
>>> Besides reverting ef53a17, could we consider introducing a config
>>> option to allow the source port to be overridden, as a buggy ISP
>>> workaround?
>>
>>
>> The problem here is simple, but hard to solve.
>> dhcpcd is not running in master mode, which means you could have >1 dhcpcd
>> instance handling dhcp6 on different interfaces. However, only one instance
>> can actually bind the unspecified address to port 546. As we don't know
>> which instance can win, we just don't bind it in non master mode, so the
>> source port is random.
Yup, understand that part, it's kinda ugly.
>>
>> We could bind to the LL address for the interface, but that in turn means
>> we *may* not be able to unicast messages to the DHCPv6 server if it's asking
>> for that.
>> So a short term fix would be to ignore the unicast option in non master
>> mode and bind to the LL address.
>
>
> Fixed here hopefully:
> https://roy.marples.name/git/dhcpcd.git/commit/?id=3715c0a1b3fbb92dd08d07d5f6e8f3afa833ff0d
>
> Can you both please test it and report back?
I will give it a try this weekend, hopefully.
>
> Roy