Cisco patches bug that crashed 1% of Internet

IDG News Service - Cisco has fixed a bug in its IOS (Internetwork Operating System) router software that contributed to a brief Internet blackout last week, thought to have affected about 1% of the Internet.

The bug was discovered last Friday when the RIPE NCC (Reseaux IP Europeens Network Coordination Centre) and researchers at Duke University started distributing experimental BGP (Border Gateway Protocol) data via RIPE NCC's systems. A large number of routers on the Internet became unreachable within minutes and the experiment was quickly stopped.

The Border Gateway Protocol is used by routers to find the best ways to send traffic to each other on the Internet. Because it is very easy for bad BGP data to spread quickly, security experts have warned that it could someday be misused to seriously disrupt the Internet.

It turned out that routers that were running Cisco's IOS XR operating system took the experimental data -- which was much larger than typical BGP routing information -- corrupted it, and then passed that corrupted information on to other routers. Many of the routers that received this information simply closed connections with the Cisco routers that sent the buggy data, causing part of the Internet to become inaccessible.

In a security advisory released just hours after the incident, Cisco confirmed that Friday's incident disclosed the bug. "An advertisement of an unrecognized but valid BGP attribute resulted in resetting of several BGP neighbors on 27 August 2010. This advertisement was not malicious but inadvertently triggered this vulnerability," Cisco said in its advisory.

Cisco's IOS XR operating system is built for its carrier-grade CRS-1 routers, used by large telecommunications companies.

Reached via e-mail Friday, Duke University assistant professor Xiaowei Yang declined to explain the point of her experiment, but she said that all of the data that her team sent was "100% standard compliant."

The experiment made it difficult to reach some networks in more than 60 countries, according to Renesys General Manager Earl Zmijewski, who blogged about the issue on Friday. More than 3,500 "prefixes," or blocks of Internet Protocol address space, were affected, he said. There are just over 333,000 such prefixes on the Internet, according to the website Cidr-report.org.

Friday's disruption lasted less than half an hour.

In an interview Monday, Zmijewski said that while Cisco's buggy software caused the problems, the Duke team running the experiment should have been more careful. "The days of academics playing with a live network are kind of gone now," he said. "I think it would be foolhardy to try something like this in the future. ... I'm amazed that this happened in the first place."