Desired behaviour of "ifconfig -alias"

[Bruce M. Simpson]

JoaoBR wrote:
>
> The only correct thing you say here is that all IPs are equal - and - nobody 
> EVER said something different.
>
> Aliasing does not say anything about priority of the Ip it is simply related 
> to the time the interface was set with the IP so the first IP is the one 
> which was set first and the first alias is the one which was set after.
>
>   
The thing is, source selection policy and IP address scope blow the 
assumptions in this discussion out of the water, as does the concept of 
unnumbered IP interfaces.

Scoping and selection policy is necessary to support link-local IPv4 
(Zeroconf) and IPv6 as we move swiftly into a world where these things 
are a fact of life, and make deployment of useful IP networks for non 
computer users a reality. Interface preference is desirable in a stack 
supporting multipathing.

Unnumbered IP, for example, is not dealt with at all well by the BSD 
stack. There are situations in which it is perfectly valid; a newly 
booting machine; a router at the end of a PPP link. We deal with 
something like 30% of the problem space for unnumbered IP. We need to be 
able to tell the IP stack 'be attached to this interface without a 
configured IP'.

ifconfig syntax currently doesn't capture this requirement; the keywords 
'add/delete' are closer to this intent. In the meantime, attachment of 
an address family to an ifnet in the kernel depends upon having at least 
one address configured, therefore 'alias/-alias' is the best fit to the 
current reality of the code.

I look forward to seeing patches as a result of this discussion if 
anybody actually wishes to change anything.

Regards,
BMS