in_ifinit() fix for SIOCSIFADDR

[Matthew Dillon]

:After patch, in_ifinit() will remove ia from hash table if in_ifinit()
:fails, so for 2) and 3) there is no need to remove ia again
:
:For 1), ia was installed into hash table only if
:ia->ia_addr.sin_family == AF_INET.  This forms the condition test
:before the XXX comment.  IMHO, using sin_family to indicate, whether
:ia is in hash table or not, is not a good idea, that's why I added XXX
:comment.  Once I parallelize ia hash table, I will add a flag to
:indicate whether ia is in hash table or not.

    Ok, then I'm fine with it.  After reading the first part of your
    answer I was going to suggest having a flag.  It's a good idea
    even if the flag state is known, along with adding a few KKASSERT's
    to catch coding mistakes.

:I mean following code segment in in_ifinit() (~839 in.c rev1.34):
:		if ((error = rtinit(&ia->ia_ifa, RTM_ADD, flags)) != 0) {
:			ia->ia_addr = oldaddr;
:			return (error);
:		}
:ia's address is restored to oldaddr, but ia itself is left in hash
:bucket indexed by newaddr's hash value.
:
:Best Regards,
:sephe
:
:-- 
:Live Free or Die

    Ok.  Yes, that should definitely be moved to the correct bucket.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>