A remote attacker could trick an authenticated victim (with “autodiscovery job” creation privileges) to visit a malicious URL and obtain a remote root shell via a reflected Cross-Site Scripting (XSS), an authenticated Remote Code Execution (RCE) and a Local Privilege Escalation (LPE).
A few months ago I read about some Nagios XI vulnerabilities which got me interested in studying it a bit by myself. For those of you who don’t know what Nagios XI is I suggest you have a look at their website.
Fortunately, around that same time the team I am part of in Shielder chose to start spending one week each month to research or 0day discovery projects. These vulnerabilities are part of the ones I have found during that week, you can read about all of them at the security disclosures page. My target was to find an unauthenticated remote code execution with zero interaction needed, which I couldn’t find in that time span, maybe I’ll have a second look sometime in the future 🙂
Nagios offers quite a few options in order to try Nagios XI, with a 60 days trial which allows you to understand the architecture and try all the functionalities. During my test I used the OVA provided, however I suppose that’s a standard installation and the other options are the same.
By reading the code used on the web interface we can see a lot of files are not obfuscated and seemingly even commented. The first vulnerability I’ve found is a reflected XSS through an iframe tag creation, which is in
nagiosxi/basedir/html/includes/pageparts.inc.php at line 552, function
parse_url() can be tricked into parsing a malicious URL via the
xiwindow parameter, we can inject any URL in the resulting iframe
Now that we have the privileges of an authenticated user we can start looking at the authenticated pages. As the documentation suggests, autodiscovery jobs allow a user to setup a scheduled scan of a specific subnet, along with many other options. That functionality resides in
nagiosxi/basedir/html/includes/components/autodiscovery/autodiscovery.inc.php, at line 191 there’s an interesting function called
I know, this code looks super-vulnerable ™, but if you analyze it most of the time the inputs are sanitized at different steps (many times using escapeshellcmd instead of escapeshellarg).
As you can see the
system_dns parameter ends up in the command line string which is going to be executed. The other variables which end in the string are sanitized, not under user-control or called differently than the user-supplied ones so this “trick” doesn’t work.
Now we have command execution with
It is possible to escalate our privileges to root by exploiting the script
/usr/local/nagiosxi/scripts/repair_databases.sh which is runnable as root by our user without password, as
sudo -l states.
Reading that script we find on line 12:
We do not have write privileges on that file, but let see what it does:
For those of you not familiar with PHP, all require_once does is interpret the source code of another file during the interpretation of the current file. It is useful in modular and object-oriented projects.
Checking the permissions on such file confirms we do have read/write privileges:
$ ls -lah '/usr/local/nagiosxi/html/config/config.inc.php' -rw-rw-r--. 1 nagios nagios 8.4K Feb 18 18:38 /usr/local/nagiosxi/html/config/config.inc.php
We can poison it in order to inject arbitrary commands during the
repair_databases.sh script execution and obtain root privileges.
These vulnerabilities can be chained together in order to craft a malicious URL which when visited by a victim (authenticated in Nagios XI and with the ‘autodiscover job’ privileges) is going to trigger our vulnerabilities and provide us with a remote root shell.
PoC creation is left as exercise for the reader 🙂
As I said earlier, by looking at the source code the security bugs seems to be patched on single vulnerabilities basis instead of implementing a safe way or guidelines to do common actions (such as executing CLI commands). However, that’s just a feeling I got by reading about the historical security bugs and the source code itself, the reality might be different.
Besides that, the communication with the developers was really smooth and they released a patch quickly.
27/02: MITRE assigned CVEs