(usagi-users 01201) Re: Is this a bug?

[JINMEI Tatuya / 神明達哉 <jinmei@xxxxxxxxxxxxxxxxxxxxx>]

>>>>> On Thu, 17 Jan 2002 09:38:06 +0200 (EET), 
>>>>> Pekka Savola <pekkas@xxxxxxxxxx> said:

>> But whether such kind address selection may cause some routing problem for 
>> the reply packets?

> Yes, this is very bad IMO; this probably often leads to dropped packets 
> due to ingress filtering; the draft says:

I'd say that "very bad" is too strong.  As Yoshifuji-san said before,
there are so many parameters that affect the selection policy of the
source address; address scope, preferred vs deprecated, temporary vs
public, home address vs care-of address, etc...

It is not clear enough which one should be preferred in the following
situation:
  against a global destination address,
  A: a link-local address on the outgoing interface (assuming that the
     interface does not have a larger-scope address)
  B: a global address on another interface

according to the address selection draft, we'll choose B.  And I think
it is a reasonable decision, or at least not a "very bad" choice.

>    It is RECOMMENDED that the candidate source addresses be the set of 
>    unicast addresses assigned to the interface that will be used to 
>    send to the destination. (The "outgoing" interface.)

(If I remember correctly) It was MUST before, but was changed to
RECOMMENDED with the fact that we had several cases like the situation
above (i.e. A. vs B.).  This recommendation basically assumes that we
at least assign enough type of addresses (link-local, site-local,
global) on every interface.  We cannot always expect such
configuration in a practical operation.  Thus, this is rather an
operational issue.

					JINMEI, Tatuya
					Communication Platform Lab.
					Corporate R&D Center, Toshiba Corp.
					jinmei@xxxxxxxxxxxxxxxxxxxxx