Some of the most commonly-used firewalls are subject to a hacker exploit that lets an attacker trick a firewall and get into an internal network as a trusted IP connection.
More on security: 20 hot IT security issues
NSS Labs recently tested half a dozen network firewalls to evaluate security weaknesses, and all but one of them was found not to be vulnerable to a type of attack called the "TCP Split Handshake Attack" that lets a hacker remotely fool the firewall into thinking an IP connection is a trusted one behind the firewall.
"If the firewall thinks you're inside, the security policy it applies to you is an internal one, and you can run a scan to see where machines are," says Rick Moy, president of NSS Labs. An attacker can then pretty much run wild in the network because the firewall mistakenly considers the IP address as a trusted one coming from behind the firewall.
This week NSS Labs published its "Network Firewall 2011 Comparative Test Results" research paper about the findings. NSS Labs is a well-known product testing organization that evaluates a wide range of security gear, sometimes as vendor-sponsored comparative tests, sometimes as completely independent tests under its own determination. The Network Firewall 2011 Comparative Test published this week is in the latter category, where costs were assumed wholly by NSS Labs itself.
NSS Labs independently tested the Check Point Power-1 11065, the Cisco ASA 5585-40, the Fortinet Fortigate 3950, the Juniper SRX 5800, the Palo Alto Networks PA-4020, and the SonicWall NSA E8500.
Moy pointed out that vendors were generally reluctant to participate in the battery of tests that NSS Labs did and that in fact about half the firewall equipment in the tests was contributed directly by end-user customers, such as financial services firms, which supported the tests because they wanted to find out about possible vulnerabilities in their firewalls.
The NSS Labs report says, "Five of the six products allowed external attackers to bypass the firewall and become an internal 'trusted machine.'" The only firewall tested by NSS labs that didn't was the Check Point one.
Moy says the exploit used in the test is known as the "TCP Split Handshake," which begins during the point that the firewall and any connection is being initiated during the TCP "handshake" process to set up a connection. Moy says attack code in the wild has been known for about a year. It's '"an easy way for an attacker to become part of the network," he says. What's particularly insidious about it is that since it occurs at the handshake stage, they are unlikely to be logs and alerts associated with the attack, Moy says.
The vendors whose equipment did not pass the "TCP Split Handshake" security test are in varying stages of remediation, according to the report.
Cisco is said to be currently working with NSS Labs on this issue and "recommendations will be provided as soon as they are available."
"Fortinet does not currently provide their customers protection against the TCP Split handshake attack," the report says, but NSS Labs says Fortinet has advised the lab that one will be included in an upcoming release in May.
"By default, Juniper does not enable protection against the TCP Split Handshake attack," the report states, but NSS labs recommends that Juniper customers examine their firewall configuration and follow the guidelines described in the report. NSS Labs warns the "protection may have a negative impact on performance and/or break applications that are not using TCP properly."
Palo Alto has indicated they are targeting an official fix in an upcoming release, according to NSS Labs, adding there may be "a negative impact on performance and/or break applications that are not using TCP properly."
By default, SonicWall does not enable protection against the TCP Split handshake, and NSS Labs advises those customers "to examine their firewall configuration at the earliest opportunity."
Other findings in the NSS Labs security evaluation include insight into what performance throughput rates were in specific conditions for all the half dozen different firewalls tests in comparison to the line speed rates advertised publicly by vendors.
"Performance claims in vendor data sheets are generally grossly overstated," NSS Labs points out.
In addition, three of the six products tested crashed when subject to certain types of stability tests, a troubling situation because an attacker could exploit this over time, especially as the instability may be due to a software flaw, the report states. The Check Point Power-1 and the Cisco ASA firewall 5585-40 and the Palo Alto PA-4020 passed the test, called a protocol fuzzing and mutation test, but the Fortinet 3950B and the SonicWall NSA E8500 did not.
The NSS Labs report also includes analysis related to purchase price and total cost of ownership for all the firewalls tested.
Read more about wide area network in Network World's Wide Area Network section.