RE: IS-IS protocol implementation problem

[rdobbins () netmore net]

I had a bizarre event occur on Thursday night/Friday morning, and this is
likely the culprit.

I peer with AS701.  At approximately 11:15PM PDT or thereabouts (2:15AM
EDT), the 7507 which provides my connectivity to uu.net went belly-up in a
very strange way.  The BGP session with 701 showed active, with full tables;
existing TCP connections stayed up.  However, it appeared that all new
connections inbound from 701 were being dropped on the floor, and my
outbound traffic with them dropped from 40mb/sec down to about 5kb/sec.  The
same router was also handling a secondary connection to pbi.net; because the
BGP stayed active and in a supposedly functional state, traffic didn't get
routed in that direction as it should've been.

I had to reload the router to get it to function properly.  Very odd.
Nothing in the logs, etc.   The router just essentially went on strike, and
I've no idea why.

I don't run IS-IS, needless to say, especially with a foreign AS.  This
particular 7507 was running an 11.3.x CC-train IOS, and hadn't had any of
the ISO/CLNS family of protocols enabled, ever.

This is a bit earlier than the timeframe Sean cited, but I don't think it
was a coincidence, either.

-----------------------------------------------------------
Roland Dobbins <rdobbins () netmore net> // 818.535.5024 voice 

-----Original Message-----
From: Neil J. McRae [mailto:neil () COLT NET]
Sent: Monday, October 30, 2000 1:28 AM
To: sean () donelan com
Cc: smd () clock org; nanog () merit edu
Subject: Re: IS-IS protocol implementation problem



> At approximately 7:37am EDT on Friday, about 258 Cisco 12000's on UUNET's
> primary backbone reloaded. This appeared to be isolated to routers
> in ASN 701. It disrupted reachability to about 15% of the world-wide
Internet
> based on data from Matrix measurements.  A contributing cause was a bad
> IS-IS packet which confused certain IOS versions in the 12.0 IOS software
> train. I haven't heard what the root cause was or what originated the
> bad IS-IS packet. The Cisco bug id is CSCdr05779. Any provider running the
> affected IOS version may be vulnerable depending on what the root cause
> turns out to be.
>
> Although the bad IS-IS packet didn't propagate to other providers, several
> other providers did report BGP resets and route flaps about the same time.

If a large AS such as AS701 starts flapping I wouldn't be surprised
if other ASes start seeing BGP resets and route-flaps. Could be
that crud routing information was exchange when that chaos started
[jeez 258 routers I'd hate to have been the on duty NOC guy on that
morning :-)]

Interestingly though we still see alot routes with bad AS-PATH information
people should be setting more stringent configurations on the routes
the learn and subsequentally pass on to avoid this.

Regards,
Neil.