On Dec 4, 10:44am, liman@autonomica.se (Lars-Johan Liman) wrote:
-- Subject: Re: Truly bizarre problem with GRE tunnel.

| christos@astron.com:
| > I think that we had this backwards and it was causing interoperability
| > problems, and it was changed. Can you recompile your ZC system with this
| > patch:
| 
| Unfortunatelly not easily. My src tree has been uppgraded numerous
| times since then. I try to keep the sys part of all -currents that I
| actually have in operation, but sometimes I forget to tar it up
| before I do an upgrade. This is such a case. :-(
| 
| I could check out an old tree using "date" I suppose. It'll take some
| time, though.
| 
| BTW, when was the patch added to the tree?

if_gre.c    11 Dec 2003 00:22:29

You could also apply the patch to the current tree, thus breaking gre
again, but making it compatible with the old kernel.

| > Alternatively use a gif tunnel. You'll be much happier.
| 
| I'm not fond of tunnels, and I have no prejudice against one type or
| another, so that's OK with me.
| 
| What are the advantages with gif versus gre?

I don't recall exactly. I remember that gif is the "sanctioned" way, and
that gre is a strange hack flipping the last bit of the point to point
address, in order to notify the routing table that things have changed.

| ... AND, my big question is:
| 
| Why did the outgoing packet look different from the incoming one in my
| tcpdumps? Are you telling me that the code in the patch is applied
| _before_ tcpdump picks up the packet on the incoming interface on my
| server?!
| 
| If there is such code - then what use is tcpdump?

That I don't know. Try changing the current kernel and running tcpdump
again.

christos