Bug #70644 [Com]: trivial hash complexity DoS attack

From: Date: Tue, 06 Oct 2015 07:26:52 +0000
Subject: Bug #70644 [Com]: trivial hash complexity DoS attack
References: 1  Groups: php.bugs 
Request: Send a blank email to php-bugs+get-196419@lists.php.net to get a copy of this message
Edit report at https://bugs.php.net/bug.php?id=70644&edit=1 ID: 70644 Comment by: phpmpan at mpan dot pl Reported by: ondrej@php.net Summary: trivial hash complexity DoS attack Status: Open Type: Bug Package: *Encryption and hash functions PHP Version: 5.6.14 Block user comment: N Private report: N New Comment: Note: there already exists a PHP extension for SipHash [1]. Unfortunely of unknown licence, but it is based on public domain version [2], which also may be used as a quick fix. Both of them linked from Aumasson's website. [1] <https://github.com/jedisct1/siphash-php> [2] <https://github.com/floodyberry/siphash> Previous Comments: ------------------------------------------------------------------------ [2015-10-05 15:34:49] ondrej@php.net Description: ------------ As reported here: https://bugs.debian.org/800564 by brian m. carlson: Applies to all PHP versions. PHP uses the DJB "times 33" hash to hash strings in its hash tables, without the use of any secret key. Hash values are therefore the same between multiple invocations. As a result, it's trivial to precompute a set of values that all hash to the same bucket and cause positively abysmal performance. If a script accepts untrusted hash keys, such as from JSON input, it is subject to a DoS attack. PHP implemented the max_input_vars option, but this is not effective in the general case, especially in the era of JSON-laden POST requests. Perl, Python, and Ruby have all addressed their CVEs properly, but PHP has not and as a result is still vulnerable. Cloning my example repository[0] and running "php scripts/exploited.php < example/1048576.json" demonstrates the problem very quickly. The similar Perl and Python scripts are not vulnerable to this attack. A JSON file containing only 65536 entries takes PHP 5.6 22 seconds to process. A new CVE should probably be allocated and the bug should be fixed correctly this time, probably by seeding a key from /dev/urandom and using SipHash-2-4 or the like. Python had CVE-2012-1150 and CVE-2013-7040. Ruby had CVE-2011-4815. I can't find a CVE for Perl's 2003 fix, if one exists. The fix, which went into 5.8, was incomplete and was addressed by CVE-2013-1667. [0] https://github.com/bk2204/php-hash-dos Test script: --------------- https://github.com/bk2204/php-hash-dos ------------------------------------------------------------------------ -- Edit this bug report at https://bugs.php.net/bug.php?id=70644&edit=1

« previous php.bugs (#196419) next »