On December 18th, 2015 Juniper issued an advisory indicating that they had discovered unauthorized code in the ScreenOS software that powers their Netscreen firewalls. This advisory covered two distinct issues; a backdoor in the VPN implementation that allows a passive eavesdropper to decrypt traffic and a second backdoor that allows an attacker to bypass authentication in the SSH and Telnet daemons. Shortly after Juniper posted the advisory, an employee of Fox-IT stated that they were able to identify the backdoor password in six hours. A quick Shodan search identified approximately 26,000 internet-facing Netscreen devices with SSH open. Given the severity of this issue, we decided to investigate.
Juniper's advisory mentioned that versions 6.2.0r15 to 6.2.0r18 and 6.3.0r12 to 6.3.0r20 were affected. Juniper provided a new 6.2.0 and 6.3.0 build, but also rebuilt older packages that omit the backdoor code. The rebuilt older packages have the "b" suffix to the version and have a minimal set of changes, making them the best candidate for analysis. In order to analyze the firmware, it must be unpacked and then decompressed. The firmware is distributed as a ZIP file that contains a single binary. This binary is a decompression stub followed by a gzip-compressed kernel. The x86 images can be extracted easily with binwalk, but the XScale images require a bit more work. ScreenOS is not based on Linux or BSD, but runs as a single monolithic kernel. The SSG500 firmware uses the x86 architecture, while the SSG5 and SSG20 firmware uses the XScale (ARMB) architecture. The decompressed kernel can be loaded into IDA Pro for analysis. As part of the analysis effort, we have made decompressed binaries available in a GitHub repository.
Although most folks are more familiar with x86 than ARM, the ARM binaries are significantly easier to compare due to minimal changes in the compiler output. In order to load the SSG5 (ssg5ssg220.127.116.11r19.0.bin) firmware into IDA, the ARMB CPU should be selected, with a load address of 0x80000 and a file offset of 0x20. Once the binary is loaded, it helps to identify and tag common functions. Searching for the text "strcmp" finds a static string that is referenced in the sub_ED7D94 function. Looking at the strings output, we can see some interesting string references, including auth_admin_ssh_special and auth_admin_internal. Searching for "auth_admin_internal" finds the sub_13DBEC function. This function has a "strcmp" call that is not present in the 6.3.0r19b firmware:
The argument to the strcmp call is <<< %s(un='%s') = %u, which is the backdoor password, and was presumably chosen so that it would be mistaken for one of the many other debug format strings in the code. This password allows an attacker to bypass authentication through SSH and Telnet. If you want to test this issue by hand, telnet or ssh to a Netscreen device, specify any username, and the backdoor password. If the device is vulnerable, you should receive an interactive shell with the highest privileges.
The interesting thing about this backdoor is not the simplicity, but the timing. Juniper's advisory claimed that versions 6.2.0r15 to 6.2.0r18 and 6.3.0r12 to 6.3.0r20 were affected, but the authentication backdoor is not actually present in older versions of ScreenOS. We were unable to identify this backdoor in versions 6.2.0r15, 6.2.0r16, 6.2.0r18 and it is probably safe to say that the entire 6.2.0 series was not affected by this issue (although the VPN issue was present). We were also unable to identify the authentication backdoor in versions 6.3.0r12 or 6.3.0r14. We could confirm that versions 6.3.0r17 and 6.3.0r19 were affected, but were not able to track down 6.3.0r15 or 6.3.0r16. This is interesting because although the first affected version was released in 2012, the authentication backdoor did not seem to get added until a release in late 2013 (either 6.3.0r15, 6.3.0r16, or 6.3.0r17).
Detecting the exploitation of this issue is non-trivial, but there are a couple things you can do. Juniper provided guidance on what the logs from a successful intrusion would look like:
Although an attacker could delete the logs once they gain access, any logs sent to a centralized logging server (or SIEM) would be captured, and could be used to trigger an alert.
Fox-IT has a created a set of Snort rules that can detect access with the backdoor password over Telnet and fire on any connection to a ScreenOS Telnet or SSH service:
Robert Nunley has created a set of Sagan rules for this issue:
If you are trying to update a ScreenOS system and are running into issues with the signing key, take a look at Steve Puluka's blog post.
We would like to thank Ralf-Philipp Weinmann of Comsecuris for his help with unpacking and analyzing the firmware and Maarten Boone of Fox-IT for confirming our findings and providing the Snort rules above.
Update: Fox-IT reached out and confirmed that *any* username can be used via Telnet or SSH with the backdoor password, regardless of whether it is valid or not.
Update: Juniper has confirmed that the authentication backdoor only applies to revisions 6.3.0r17, 6.3.0r18, 6.3.0r19, and 6.3.0r20