, SecurityFocus 2005-09-07
Fernando Gont is nothing if not tenacious.
Earlier this year, the Argentinian researcher highlighted several attacks that could disrupt network connections using the Internet control message protocol, or ICMP, and proposed four changes to the structure and handling of network-data packets that would essentially eliminate the risk.
However, rather than open up a discussion on the flaws and their fixes, Gont's disclosure marked the start of a months-long debate over whether the vulnerabilities--the general details of which have been known for some time--are serious enough to require fixing. While many researchers have lauded his research, others in the security community have criticized the work on public mailing lists. The few companies that Gont has contacted have not generally cooperated, and very few makers of operating systems and network software have implemented his fixes.
Yet, the researcher is at it again. This week, Gont updated his proposal to the Internet Engineering Task Force (IETF), the body that handles technical standards for the Internet, to add more information sought by some researchers.
"Some people say 'this is old stuff,'" said Gont, currently on staff at the Universidad Tecnologica Nacional (National University of Technology) in Argentina. "But they miss a very important point: While these attacks have been known to many people for many years, there have never been proposals on how to deal with them."
The flaws are essentially issues caused by the lack of a requirement in the Internet standards to check ICMP packets for specific, problematic data. By sending enough malformed ICMP packets at a vulnerable server or network router, an attack could shutdown a connection, cause a degradation in network bandwidth, or cause the host to burn processor cycles, Gont said.
The issues affect a large number of networking products and operating systems. While a vulnerability note released by the United States' Computer Emergency Readiness Team (US-CERT) in April indicates that the lion's share of vendors have not declared if their products are vulnerable, major vendors--such as Cisco Systems, Juniper Networks, Microsoft and Sun Microsystems--have confirmed the security issue affects their systems, according to the vulnerability note. An analysis of the vulnerabilities by the National Infrastructure Security Co-ordination Centre, United Kingdom's version of US-CERT, contains more information about vulnerable systems.
Cisco addressed the issues in an update to a wide variety of products in April, following Gont's first presentation of the issues at CanSecWest in Vancouver. Cisco confirmed all three attacks but did not comment on their severity.
"We take all research like this very seriously," said Cisco spokesman John Noh.
To Gont, the problems are serious. To others, the potential attacks are too complicated and gain the attacker too little to be much of a threat.
"To some people, any possible avenue of attack is a serious issue," said Mark Allman, co-chair of the TCP Maintenance and Minor Extensions working group at the IETF. "Sometimes we need the paranoid folk's beating of the drum, but often times, these sorts of folks work in the minutia."
Gont readily admits that he did not discover the broad vulnerabilities, but refined their definitions and quantified the threat to transmission control protocol, or TCP, connections. Data sent via TCP makes up the lion's share of Internet activity.
Gont identified three attacks that could affect connections made between a host and a client over the Internet. The first could allow an attacker to reset an arbitrary TCP connection. The second could allow an attacker to degrade the throughput of a TCP connection until the target could only send a single packet at a time. The third could allow the attacker to reduce the throughput of the connection and increase the workload of the processor. All are considered "blind" attacks, since they don't require the attacker to sniff the targeted network.
While the attacks could affect Web applications, the serious threat is to the invisible infrastructure that helps the Internet run efficiently, Gont said. For example, using the attacks against routers that use the border gateway protocol, or BGP, to determine the best path to send large quantities of data could result in serious disruptions.
"Being able to use the attacks to successfully attack BGP means you can take entire networks off the Internet," Gont said.
Some have taken the threat seriously. The developers for the OpenBSD operating system fixed the issues using all four of Gont's suggested changes to the protocols, said Theo de Raadt, project leader for OpenBSD.
"This thing has been so frustrating to many of us, because Gont was so careful to write a very good paper about the issues and his fixes, and it is quite clear that no one read his paper," de Raadt said.
Other researchers have agreed with Gont about the severity of the problem, but have couched their assessment in more general terms.
"Any security compromise is serious," said John Day, chief technology officer for Netnostics and a member of the Network Working Group (NWG) that developed the original Internet application protocols. "This attack alone may not be a big deal but it may turn out that in combination with others it has greater impact."
Gont intends to continue to answer his critics and lobby to have his fixes added to the relevant Internet specification.
The IETF working group to which he has submitted his draft has not accepted the document, even though adding some sort of "sanity checks" similar to those proposed by Gont is an idea under consideration, said the IETF's Allman.
"This whole thing has been way overblown," Allman said. "This has been so publicized that if there was a large-scale danger, someone would have exploited it and caused large-scale problems by now."
In one way at least, Allman's point seems to have been made. In July, Gont posted three attack tools to his Web site to demonstrate the three vulnerabilities. So far, the publicly available tools have not resulted in any obvious major attacks on Internet infrastructure.
In the end, only such a wake-up call may be able to settle the debate. Unless a major event shakes up the security community, most researchers continue to believe that the threat is not a critical one.
That's a serious mistake, said Gont.
"For some reason, it seems the security community does not understand what matters is which threats are current, rather than which threats are new," he said. "If someone can take your entire network off the Internet, does it make a difference whether he did it with a zero-day exploit or with a 20-year-old bug?"