Researchers at Finnish security firm Outpost 24 claim to have discovered a flaw in the Internet Protocol that can disrupt any computer or server.

After keeping the flaw quiet for years, the researchers hope that going public will help accelerate the creation of a solution.

The flaw allows attackers to cripple computers and servers by sending a few specially formed TCP/IP packets. The result can be compared to a denial of service attack, in which networks are flooded with traffic. But in the case of the newly revealed flaw, only a minimum of traffic is required. "We're talking 10 packets per second to take down one service," Jack Lewis, a senior researcher with Outpost24.

Researchers at Fox-IT, a Dutch security firm, confirm the issue. "Based on the available information, this vulnerability may be a serious problem for system availability," observed Erwin Paternotte, a researcher with Fox-IT. "If the technical details are publicly disclosed, performing a denial-of-service attack will become relatively trivial."

The problem surfaced during a test scan of 67 million Internet hosts. The researchers were alerted when a test caused some hosts to become unresponsive. Further investigation led to an issue in the TCP/IP stack. After a connection is successfully made, important system resources are at the attacker's disposal.

Each operating system is affected by the flaw, although different systems respond in different ways. "Each operating system does behave differently, of course. You might notice with OS X that a couple attacks that don't seem to bother too much completely devastate Windows XP and the other way around," said Lewis.

The researchers have crafted proof-of-concept code that demonstrates the issues. They claim that they hadn't seen a single implementation of TCP/IP that wasn't vulnerable. Systems remain unresponsive after an attack. "After the attack is over, the system never seems to recover until it is rebooted," said Robert Lee, Outpost24's chief security officer.

Firewalls or intrusion prevention systems are unable to mitigate the flaw, because they too support TCP/IP and are therefore a potential attack target.

The researchers so far have conceived five different attack scenarios, but they argue that as many as 30 would be feasible. "You basically have to sit there and stare through code and figure out what stages you can get to," Lewis said.

The researchers are publicising their finding after keeping quiet for three years, although they don't plan to fully disclose all the flaw's details. "We hope we can raise awareness and get more people that are smarter than us involved in looking for a solution," Lee said. Just because we can't think of a solution doesn't mean there isn't one, just that we haven't thought of it yet."

But there is another reason. Lee and Lewis see the migration toward IPv6 as a risk that could aggravate the situation. IPv6 "only makes the issue bigger, because the address space is bigger," Lee said.

The Finnish CERT is coordinating research into the security issue and education to software vendors affected by the flaw.