(usagi-users 00555) Re: Bug in 2.4.x multicast code?

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

>>>>> On 25 May 2001 15:17:10 +0200, 
>>>>> Bjorn Lindgren <bjorn.e.lindgren@xxxxxxxxx> said:

>> I'm afraid that this reply (of mine) is rather confusing, but
>> "interface" is not necessarily a correct wording.  Since "interface"
>> is a smaller scope than "link", we just have to specify the "link" for
>> sending a packet to a link-local destination.  Specifing "interface"
>> is theoretically an overspecification.  However, some systems
>> (e.g. KAME/BSD) assume a one-to-one mapping between interfaces and
>> links, and thus specifying an interface causes the same result of
>> specifing a link.  I don't know USAGI has the same assumption, though.

> Seams resonable for multicast addresses, but specifying "link" for
> normal interface link-local as fe80::/10 ?

> Should'nt the /128 entry pointing to loopback in the routing table
> match?

> Kernel IPv6 routing table
> Destination                    Next Hop    Flags Metric Ref    Use Iface
> fe80::a00:2bff:fe98:2f62/128   ::          U     0      0        0 lo

Sorry, but I don't understand your point.  In my understanding, the
issue here is how a node communicates with another node using a
link-local address.  For example, in the following topology,

    node 1
       |fe80::1
       |
       |
       |(link 1)
    node A
       |(link 2)
       |
       |
       |fe80::1
    node 1'
    
both node 1 and node 1' can have a same link-local address, say,
fe80::1.  When the node A wants to communicate with one of other
nodes, it should specify the outgoing link as well as the address
"fe80::1".  How does a route to the loopback interface solve this
issue?

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