No Way, PHP Strikes Again! (CVE-2024-4577)

Orange Tsai tweeted a few hours ago about “One of [his] PHP vulnerabilities, which affects XAMPP by default”, and we were curious to say the least. XAMPP is a very popular way for administrators and developers to rapidly deploy Apache, PHP, and a bunch of other tools, and any bug that could give us RCE in its default installation sounds pretty tantalizing.

Fortunately, for defenders, the bug has only been exploited on Windows-based PHP installations (where PHP is specifically used in CGI mode), under some specific locales:

Chinese (both simplified and traditional), and
Japanese.

However, Orange cautions that other locales could be affected too, and urges users to upgrade to the latest version of PHP, which fixes these bugs (for detail, see their blogpost).

We are keen to point out that we are unsure how common this configuration, or deployment type, is in reality. It is also not our job to find out, outside of our client base. But, regardless, it's an interesting vulnerability due to the root cause. Enjoy with us.

Orange's blogpost, while informative, doesn’t tell us exactly what to do to get that sweet RCE. Unfortunately, the wide range of configuration options makes it difficult to conclusively prove an instance to be vulnerable (or not) at a passive glance and, obviously, because a Windows machine's 'locale' is not typically externally fingerprintable. Because of this, we set about reproducing the bug—if we can exploit it, that’s the best way of proving exploitability, right?

Reading Orange's blog, it is clear that the bug only affects CGI mode of PHP. In this mode, the webserver parses HTTP requests and passes them to a PHP script, which then performs some processing on them. For example, querystrings are parsed and passed to the PHP interpreter on the command line - a request such as as http://host/cgi.php?foo=bar might be executed as php.exe cgi.php foo=bar, for example.

This does, of course, introduce an avenue for command injection, which is why input is carefully handled and sanitized before calling php.exe (cough CVE-2012-1823). However, it seems there is a corner-case which the developers did not account for, which allows an attacker to break out of the command line and supply arguments that are interpreted by PHP itself. This corner-case relates to how unicode characters are converted into ASCII. This is best explained with an example.

Here are two invocations of php.exe, one malicious and one benign. Can you spot the difference?

No, neither can I. Let’s look at then in a hex editor and see if that give us any clue.

Hmm, interesting - here we can see that the first invocation uses a normal dash (0x2D), while the second, it seems, uses something else entirely (a ‘soft hyphen,’ apparently), with the code 0xAD (highlighted). While they both appear the same to you and me, they have vastly different meanings to the OS.

An important detail here is that Apache will escape the actual hyphen - 0x2D - but not the second ‘soft hyphen’, 0xAD. After all, it’s not a real hyphen, right? So there’s no need to escape it… right?

We don't care if it's the same joke as above, it's still funny.

Well. It turns out that, as part of unicode processing, PHP will apply what’s known as a ‘best fit’ mapping, and helpfully assume that, when the user entered a soft hyphen, they actually intended to type a real hyphen, and interpret it as such. Herein lies our vulnerability - if we supply a CGI handler with a soft hyphen (0xAD), the CGI handler won’t feel the need to escape it, and will pass it to PHP. PHP, however, will interpret it as if it were a real hyphen, which allows an attacker to sneak extra command line arguments, which begin with hyphens, into the PHP process.

This is remarkably similar to an older PHP bug (when in CGI mode), CVE-2012-1823, and so we can borrow some exploitation techniques developed for this older bug and adapt them to work with our new bug. A helpful writeup advises that, to translate our injection into RCE, we should aim to inject the following arguments:

-d allow_url_include=1 -d auto_prepend_file=php://input

This will accept input from our HTTP request body, and process it using PHP. Straightforward enough - let’s try a version of this equipped with our 0xAD ‘soft hyphen’ instead of the usual hyphen. Maybe it’s enough to slip through the escaping?

POST /test.php?%ADd+allow_url_include%3d1+%ADd+auto_prepend_file%3dphp://input HTTP/1.1
Host: {{host}}
User-Agent: curl/8.3.0
Accept: */*
Content-Length: 23
Content-Type: application/x-www-form-urlencoded
Connection: keep-alive

<?php
phpinfo();
?>

Oh joy - we’re rewarded with a phpinfo page, showing us we have indeed achieved RCE.

Conclusions

A nasty bug with a very simple exploit - perfect for a Friday afternoon.

Fortunately, though, patches are available, so we echo Orange Tsai’s advice to upgrade your PHP installation. As always, fantastic work and a salute to Orange Tsai.

Those running in an affected configuration under one of the affected locales - Chinese (simplified, or traditional) or Japanese - are urged to do this as fast as humanely possible, as the bug has a high chance of being exploited en-mass due to the low exploit complexity. Other users are still strongly encouraged to update:

For Windows running in other locales such as English, Korean, and Western European, due to the wide range of PHP usage scenarios, it is currently not possible to completely enumerate and eliminate all potential exploitation scenarios. Therefore, it is recommended that users conduct a comprehensive asset assessment, verify their usage scenarios, and update PHP to the latest version to ensure security.

We won’t duplicate the advisory here, instead, we advise those individuals seeking remediation advice to refer to the comprehensive advisory.

At watchTowr, we believe continuous security testing is the future, enabling the rapid identification of holistic high-impact vulnerabilities that affect your organisation.

It's our job to understand how emerging threats, vulnerabilities, and TTPs affect your organisation.

If you'd like to learn more about the watchTowr Platform, our Attack Surface Management and Continuous Automated Red Teaming solution, please get in touch.