[EMAIL PROTECTED] wrote:
>> As far as I can tell, the key.src value which is preset, is determined
>> by the address the socket is bound to. Meaning: it seems like this
>> patch doesn't mess up anything. If someone can explain me otherwise,
>> speak up.
>Oops! Does it solve the problem discussed in gated list thread
>"Wrong sender address despite best efforts"?
>(and probably "Sequence # Mismatch")
In one word: yes.
>BTW could you summarize this thread for me in several sentences?
Here goes:
1. First I tried:
- One IP-number for the ethernet.
- One (other) alias IP-number for the ethernet.
- Run gated in the backbone OSPF area.
- Gated turns out to be originating multicast packets from the
alias sometimes (not often), messing up any formed adjacencies.
This caused the "Sequence # Mismatch" error.
2. I solved this by making sure that I specify the IP-number instead
of the interface name (eth0) in the ospf configuration.
3. Then I tried to get gated to take part in several areas at the
same time using the same interface.
I.e. I have: one ethernet, several routers on the (switched) ethernet.
I'd like the Linux box running gated to form a separate area with
each of the routers present on the ethernet (ideally, an nssa).
I discovered that gated seems to be incapable of taking part in
more than one area using one interface (I tried tricks with
point-to-point definitions, didn't help).
4. I created several virtual interfaces on the same ethernet.
I.e. 10.0.1.0/29 10.0.1.8/29 10.0.1.16/29 etc.
This way gated gladly accepts my configuration and opens up a separate
area on each virtual network (all on the same ethernet).
The trouble now turned out to be that all multicast messages originate
from the *master* interface (eth0), instead of from the alias interface
the OSPF-area and multicast belongs to.
5. First I thought this was a gated problem. After some investigation
it turns out that gated is doing everything correctly, but Linux is
messing things up.
I.e. gated is binding to the right socket and IP-address, the kernel
however, is blissfully ignoring the IP-number the socket has been bound
to as soon as it needs to send out a multicast packet.
My patch solves that.
Which is not to say that it is still kind of silly that gated cannot
join more than one area on one interface (i.e. I get to waste a bit
of IP-address-space).
One more problem to solve (working on it as we speak): teaching gated
to work with NSSAs.
--
Sincerely, [EMAIL PROTECTED]
Stephen R. van den Berg (AKA BuGless).
"Be spontaneous!"
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
