Moved to tech-net from source-changes.

itojun@iijlab.net writes:

> 	this is definitely a post-1.5 item, just to be sure.

So, for the 1.5 I guess we can just pullup them.

(1) For the original problem that an existence of dangling struct
in_multi for ALLHOST_GROUP, just keeping track of the object should do
the right thing.

(2) What the multicast kludge table fixes is a different problem.
Without it, kernel may leave from a multicast group while it
shouldn't, after the first AF_INET address is deleted.

(3) Leaving from multicast group on an interface detach is anyway the
current behaviour for both v4/v6.

enami.