[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-netdev
Subject:    Re: (usagi-users 00745) Re: [Patch 2of2] IPv6 routers don't
From:       "David Stevens" <dlstevens () us ! ibm ! com>
Date:       2001-08-30 22:44:55
[Download RAW message or body]


[sorry if this is duplicated; I'm having mailer problems ]
>
>>  - whether we forward a packet from it
>>  - whether we set is_router flag in NA to be sent on it
>>  - whether we join (some scope(s) of) all-routers multicast on it
>>
>> SHOULD be the same.
>
>Why? They should be different flags and the last is not flag at all.
>Kernel is not going to police you.
>
>What's about the first item, it is too complicated. Grabbing device
>lock in data path is not an option.

     I'm no expert on the linux implementation, but I'd think this
should just require a reference you already have. It doesn't really have
to be atomic, since reading a stale value for a short time during a
transition wouldn't be fatal.

>> ping6 ff02::2%eth0 ?
>
>Compare to ping4 224.0.0.2%eth0
>
>Routers lose expensive entry in multicast entry not for fun,
>but for work.

RFC 2373, section 2.8:

 A router is required to recognize all addresses that a host is
   required to recognize, plus the following addresses as identifying
   itself:
...
      o All-Routers Multicast Addresses

Even if it weren't a requirement, I think it's still a good idea. For
network debugging and for applications that simply can't (easily) join
and leave the group under the right circumstances, because appropriate
membership depends on the forwarding state of the interface, something
applications shouldn't have to check and recheck for correctness.

                                   +-DLS

[prev in list] [next in list] [prev in thread] [next in thread]