Re: Point-to-Point interfaces regressions

Li, Qing wrote:
>> 1) I am going to reuse Ethernet address as local for PtP link:
>> %ifconfig ng0 10.0.0.1 10.0.0.2
>> ifconfig: ioctl (SIOCAIFADDR): File exists
>> %ifconfig ng0
>> ng0: flags=88d1<UP,POINTOPOINT,RUNNING,NOARP,SIMPLEX,MULTICAST> metric
>> 0
>> mtu 1500
>>         inet 10.0.0.1 --> 10.0.0.2 netmask 0xff000000
>> So as you can see, address was assigned, but request returned error
>> status. Probably it is related to the recent local address route
>> changes.
> 
> I have written several times about this particular topic here.
> 
> Yes, the behavior changed. Previously if you try to ping 10.0.0.1 the
> packets leak onto the wire even though that's a local address. And you
> cannot connect to the local end, again, packets go on the wire towards
> possibly the default gateway.

Me and many other people running net/mpd handling thousands of PtP
interfaces sharing local addresses with each other and with some
Ethernet interface. This change makes such setup inoperable, as mpd will
constantly receive errors while trying to set addresses and drop
connections.

There was legacy hack in mpd, doing the same, but in user level. I have
removed it few years ago, because it never worked as expected with more
then one interface. I would like kernel implementation to be more correct.

>> 2) I am going to configure arp-proxy:
>> %arp -s 10.0.0.2 00:1b:24:c5:5b:09 pub
>> arp: writing to routing socket: Invalid argument
>> And again it doesn't work, writing to syslog:
>> kernel: lla_rt_output: RTM_ADD publish (proxy only) is invalid
>>
> 
> I will take a look and let you know what the issue might be.

Thank you.

-- 
Alexander Motin