dhcpcd-discuss

On 22/04/17 20:55, Stuart Henderson wrote:
> On 2017/04/22 15:36, Roy Marples wrote:
>> On 22/04/17 11:49, Stuart Henderson wrote:
>>> As far as OpenBSD goes, calling ioctl SIOCGVNETID with struct ifreq
>>> and checking for ifr_vnetid > 0 might be a good way. It will be set to
>>> the vlan tag number on a vlan interface, and also act as a stable id
>>> for vxlan(4) interfaces (http://man.openbsd.org/vxlan) - these do
>>> have a MAC address but it's typically randomly generated.
>>>
>>> Example code in getvnetid() in
>>> https://cvsweb.openbsd.org/cgi-bin/cvsweb/src/sbin/ifconfig/ifconfig.c?rev=HEAD&content-type=text/x-cvsweb-markup
>>>
>>> The only problem is where you have the same vnetid number on multiple
>>> interfaces (e.g. vlan with a certain tag number on multiple interfaces),
>>> but it would cope with the common cases automatically, and is no worse
>>> than the status quo if there are duplicates.
>>
>> Initial patch for this is here:
>> https://dev.marples.name/D107
>>
>> It seems to work well enough, but it relies on vlan interfaces being
>> existing and configured with a vlanid before dhcpcd starts. This is
>> sub-optimal and requires a bit more work yet, but the OS specific parts
>> should be done (sans Illumos, but I can work on that later).
>> Saying that, dhcpcd currently doesn't bring up vlan interfaces on NetBSD
>> at least that are created while dhcpcd is running as the hardware
>> address is not yet assigned, so it's no worse for sure.
>>
>> Hopefully you can test it on OpenBSD to ensure it works there at least.
>> If so, you could accept the revision via the web interface or just
>> comment here.
>>
>> Roy
>>
> 
> The code works on OpenBSD, but trying it with my setup shows a
> problem. The pppoe interface doesn't have a hwaddr so unless I hardcode
> IAID for that dhcpcd falls back to using the ifindex, which then conflicts
> with a vlan's vnetid.
> 
> Would forcing the first byte of IAID to 0xff or something be sane for
> this case?

Yes it would because the maximal VLANID is 4095 which easily fits in the
last two bytes.

I updated the revision and added you as a reviewer.

Roy