BUG/DANGER in Bridging-kernel code

Holger P. Krekel (hpk@prim.han.de)
4 Nov 1996 12:47:23 GMT


Well, it costed me about 70 german marks and more than 10 hours:

Enable bridging (yes, i know it's experimental) and strange things will happen,
i.e. packets will get routed at link-level between (in this case)
isdn and ethernet-devices. I never told him so and I haven't installed
the briding-software so far, but enabled kernel-support for it.

This is how I tracked it down:
* boot kernel 2.0.24 (2.0.21 is as good) with tcp and if you like
without forwarding/masquerading etc, but bridging enabled, in SINGLE mode
* no processes (except init, sh, syslogd) should work, no devices etc.
* create an eth0 - device and give it some adress (doesn't really matter)
and create a net-routing entry for that device., say 123.123.123.0
* load the isdn-modules and create an ordinary isdn-device isdn0
with some ip-adress and a phone-number, NO route-entry
* ping 123.123.123.255

-----> look in your syslog and you see your isdn-subsystem beeing busy
calling to the world with no ip-pakets to transport.

So, what is it?

1. A bug in bridging code?
1. Me beeing too optimistic about defaults in the bridging code?
2. a more general problem I haven't realized so far?
3. me beeing not afraid of experimental features?

Any comments welcome. Maybe this message saves other peoples' money.

Aeeh, yes, by the way, if you test it yourself be sure to disconnect
the isdn-cable :-).

Holger (Hildesheim, Germany)