Re: [PATCH net-next] ipv6, token: allow for clearing the current device token

[Daniel Borkmann <daniel@iogearbox.net>]

On 04/08/2016 05:36 PM, Hannes Frederic Sowa wrote:
> On 08.04.2016 17:25, Bjørn Mork wrote:
>> Hannes Frederic Sowa <hannes@stressinduktion.org> writes:
>>> On Fri, Apr 8, 2016, at 16:18, Bjørn Mork wrote:
>>>> Daniel Borkmann <daniel@iogearbox.net> writes:
>>>>
>>>>>       if (!token)
>>>>>           return -EINVAL;
>>>>> -    if (ipv6_addr_any(token))
>>>>> -        return -EINVAL;
>>>>>       if (dev->flags & (IFF_LOOPBACK | IFF_NOARP))
>>>>>           return -EINVAL;
>>>>
>>>> Not directly related to the patch in question.  It just made me aware of
>>>> this restriction...
>>>>
>>>> I realize that I'm a few years late here, but what's with the IFF_NOARP?
>>>> Is that just because we can't do DAD for the token based addresses?  How
>>>> is that different from manually configuring the whole address?
>>>
>>> IFF_NOARP is kind of the equivalent to no neighbor discovery. If you set
>>> a token and never get in a router advertisement you never create a
>>> tokenized ip address, thus the feature is useless.
>>
>> You can get router advertisements with IFF_NOARP. You cannot lookup L2
>> addresses, but the L3 prefix info is still as useful as with any other
>> interface.
>
> Of course router advertisements can be send and received with IFF_NOARP and probably we act on them as usual, as you showed. Looking in the source we don't really specify what those flags mean/do for IPv6. So I think you can assume that it is in there because of history.
>
> I would absolutely not mind if you remove the limitation for IFF_ARP.

Agreed me neither, the code should be able to handle it as far as I see.

Thanks,
Daniel