Slashdot Mirror
How a Router's Missed Range Check Nearly Crashed the Internet
Barlaam writes "A bug by router vendor A (omitting a range check from a critical field in the configuration interface) tickled a bug from router vendor B (dropping BGP sessions when processing some ASPATH attributes with length very close to 256), causing a ripple effect that caused widespread global routing instability last week. The flaw lay dormant until one of vendor A's systems was deployed in an autonomous system whose ASN, modulo 256, was greater than 250. At that point, the Internet was one typo away from disaster. Other router vendors, who were not affected by the bug, happily propagated the trigger message to every vulnerable system on the planet in about 30 seconds. Few people appreciate how fragile and unsecured the Internet's trust-based critical infrastructure really is — this is just the latest example." Vendor A, in this case, is a Latvian router vendor called MikroTik.
Is this related to the story posted that stated:
"One Broken Router Takes Out Half the Internet?"
http://tech.slashdot.org/article.pl?sid=09/02/16/2233207
It just amazes me how differently presented this story is compared with the previous.
In fairness, there is much more information about this 'outage' now.
This news is alarming. Thanks for not making in alarmist this time.
Vendor B is Cisco btw. Dunno why they were being vague.
I'm sure nobody here would argue with me if I suggested that the internet would be a much safer place without routers.
If people had upgraded their routers this wouldn't have happened. Newsflash: software has bugs. Not upgrading your software will bite you in the ass eventually, especially if this software runs critical systems like your routers.
except in the kdawson style it was a single link to a message board posting about a router "taking out half the internet." Dupe? Correction? I dont care as long as kdawson is kept away from the site for a while.
I don't know about it nearly crashing the Internet. How many people actually noticed a difference that day, for that matter?
A lot of admins, especially after the alert went out over the NANOG list, set their routers to reject long ASPATHs (or I assume, from what I saw on those list, I am not a BGP admin myself.) Many routers simply rejected these ASPATHs as well; correct me if I'm wrong, but weren't old versions of IOS the only ones affected? It was a serious issue, but I'm not sure if it came anywhere near a disaster scenario.
"The Internet was back to normal in short order."
Well, not completely normal, not yet.
I always thought they did.
Most already do. The problem was not the ASPATH itself, it was the length of it. The routers affected did not handle updates for a prefix which required more than one AS_SEQUENCE segments in order to obtain the full AS path. The existence of the additional AS_SEQUENCE segment is what triggered the bug, causing the receiving router to treat the update as invalid, and the BGP session is dropped.
Few people appreciate how fragile and unsecured the Internet's trust-based critical infrastructure really is - this is just the latest example.
Yeah. Like how everyone is trusted not to google "google".
When I worked for *unnamed nw regional backbone here* we had peering agreements with everyone except uunet that we connected to, and it was pretty known that if we spat out an bad BGP route we could bring down the whole net by hitting enter ('cept uunet, although I'm pretty sure uunet woulda went down from everyone else routing around them to us)
How is this new? That was the 90's. and when we spent 100k+ on a Cisco 7513 with 64megs of ram so it could hold the BGP tables...
We even wrote our own manual ('cause none existed) on how to deal with BGP tables so junior admins working for us wouldn't fuq it up. (and on top of that, we wouldn't let them touch the routers either)
-meetme room in the westin in Seattle-
The critical bug is with the Cisco routers; a Mikrotik router merely nearly triggered the bug.
It would be possible to trigger this bug with any routing software that does not do range checking on the amount of times the ASN is pretended.
The summary is spreading FUD by making Mikrotik, the only named vendor in the summary, look like the vendor at fault.
The next time someone needs you to fix a computer problem and asks what went wrong, simply give them this article's summary as the reason why, replacing "router" and "Internet" with the the defective part in question. You're also guarenteed to look a bit sharper, too.
"A bug by power supply vendor A (omitting a range check from a critical field in the configuration interface) tickled a bug from power supply vendor B (dropping BGP sessions when processing some ASPATH attributes with length very close to 256), causing a ripple effect that caused widespread global routing instability last week. The flaw lay dormant until one of vendor A's systems was deployed in an autonomous system whose ASN, modulo 256, was greater than 250. At that point, the power supply was one typo away from disaster. Other power supply vendors, who were not affected by the bug, happily propagated the trigger message to every vulnerable system on the planet in about 30 seconds. Few people appreciate how fragile and unsecured the power supply's trust-based critical infrastructure really is â" this is just the latest example."
Mikrotik are known GPL violators, that use a modified Linux (they re-branded that as "RouterOS") and a terribly bad implementation of the BGP protocol..
In some custom community network, where MikroTik has been deployed internally, that stolen-Linux is being hacked to use the Quagga instead of MikroTik's BGP.
In short: that "RouterOS" has been higly unsuitable for the Internet. I can't believe somebody was so stupid to trust it.
Reminds me of a story that Keith Marzullo told our class in a graduate level reliability class. This was back in the days of using UUCP to send email, and the vendor that he worked for had just released a "failsafe" product they were very proud of -- essentially, it was a mail router that could detect if a path went down, and would try an alternate router instead. The company touted it as a bulletproof solution.
So they go to a conference, and set up some routers, unplug some of them, etc., and everything is going fine until they ask an audience member for his UUCP address. UUCP addresses are in the form of host1!host2!host3!username, with the routing for the username explicitly specified... the addresses could thus get quite long. In this case, the guy's email address was over the buffer limit the company's routers used.
Guess what happened?
The mail server tried sending an email to the next router in the chain. The router buffer overflowed and crashed. The reliable server than tried another router... and crashed it. It then went through the entire network, and crashed every single one of the nodes, turning a bug that would have been a single point of failure into a total network collapse.
=)
Yeah, one of my favorite stories from UCSD.
Maybe if they updated their IOS back in 2003 when Cisco came out with the fix they wouldn't have these problems. You wouldn't give an XP user a pass on not updating for 6 years and having a problem, don't give these upstreams any.
-zifr
Summary reads like the script for a bad disaster movie.
Carbon based humanoid in training.
...A Slashdot "Editor" notices these posts and mods them into oblivion.
But is that better or worse than having them modded down by sycophantic Slashdot readers?
My Slashdot login - a four-digit userid - is worthless now.
It's been stuck on Karma:-1, Terrible for a couple of years.
What did I do to deserve that terrible fate?
My sin was to post a message critical of dear Michael Sims and his editing methods and practices here on Slashdot.
At that point, the Internet was one typo away from disaster.
I wonder how long that took?
Pulp Audio Weekly - Geek News and Reviews
A bug by device vendor A (twiddling a framis panel instead of sparting the glinbo interface) patted a bug from device vendor B (elevating ALP packets when deferring some GALAS modifiers with size benath 176), yielding a domino effect that caused widespread universal switching instability last week. The flaw lay dormant until one of vendor A's systems was deployed in an autonomous system whose LKM, divisor 965, was less than 1250. At that point, the Internet was one typo away from disaster. Other router vendors, who were not affected by the bug, happily propagated the trigger message to every vulnerable system on the planet in about 30 seconds. Few people appreciate how fragile and unsecured the Internet's trust-based critical infrastructure really is -- this is just the latest example.
Reads just about the same to me. I can't make any sense of either description of the bug
Cisco update policy? Isn't that called Juniper or Huawei?
Cisco used to be the best option (they weren't that great in product terms, but everyone else was worse, and Cisco had good service and support).
They're getting squeezed from both the top and bottom.