[mpls-linux-general] [PATCH] ldp-portable
Status: Beta
Brought to you by:
jleu
From: Yon U. <hav...@gm...> - 2002-04-25 03:29:30
|
Hi, this is a raw (unlinted, unindented, it will eat your boot sector) diff of my ldp-portable tree vs cvs (i hope). lib: * be quiet, remove prints. * exit decode loop on error, it will endlessly loop otherwise. (i tested ldp-portable against the indian (bangalore) ldp stuff, it would endlessly loop (they are using some private definition of ldp, apparently). * assorted random patches I've forgotten what they were about (mostly to avoid asserts and NULLs for some particular states I've encountered and I couldnt explain) zebra-port: * the start of 2 snmp implementations, one of my own, the other an ugly hacked version of smidumps version (now that they are cleaning the mpls mib mess and smidump can swallow all of them, btw the OIDS havent been assigned yet, I use transmission.188.646 for the start of the ldp mib). They dont quite work, but I think there were some ppl interested in doing snmp development, this might be helpful. * make depend * outdated README * GPL boiler plate & (c)s , jim is missing from some (C). * changed the memory allocation to use differentiated names, so leaks are easier to pinpoint. (mostly bgp names)(not that I'm debugging that yet) * zebra will send funny routes when it doesnt manage to instatiate the route in the kernel FIB, avoid dying. * it compiles with an april zebra cvs version * interface startup doesnt have a 3 second hold time anymore * "show ldp bindings" can filter by binding type (usually only map_(recv|sent) are of any interest (and present at all longer than 1 second, AFAIK)) * configure per interface hello/keepalive hold/interval timers commands * configure per interface distribution mode (aka "make me core dump") * commands to configure resources, paths and tunnels. Nothing useful. I implemented the following: When the egress/32 for a tunnel (show mpls ldp egress) gets mapped and installed an interface is brought up and bound to that label. That interface follows the label should it ever change. If you want routes to follow that interface just add routes in zebra (ip route bgp-nexthop/32 tunnel_44t0). Handle with care. Do not loop your route recursion. The zebra port is still for the old kernel interface, I'm waiting for Jorge Boncompte <jo...@dt...> to release his port to the new world order kernel interface, I hope to integrate both making it a -D option (before september, I hope). BTW, why do have mpls tunnel interface have an mtu of underlying interface minus 4 bytes? We are sending oversized labeled packets over the hw interface, why backoff now? IMHO label stacking is not a reason, as it applies to the non-interface case, too. Jim? regards yon |