The exploit uses a vulnerability within MariaDB, PerconaDB (and/or XtraDB Cluster) and MySQL to, first, gain access to the 'mysql' system user using any mysql user that has CREATE / INSERT / UPDATE permissions. The first part revolves around a race condition when sql generates temporary files as part of the `REPAIR table` command. Then using the mysql system user the second vulnerability grants the attacker root access to the server using a clever hack that takes advantage of mysql_safe's approach to writing to file based error logs. Below I've provided a list of vulnerable server versions. Just about any server using the more recent (unpatched) stable releases of MySQL or MariaDB through CentOS is vulnerable (Percona isn't part of the standard CentOS repositories), with a few of caveats.
The first caveat is that an unpatched vulnerable server can prevent at least the 2nd exploit by disabling symlinks through /etc/my.cnf using skip-symbolic-links or symbolic-links=0
The next caveat is that the 2nd exploit also depends on using file-based mysql logging. Using syslog will avoid trouble.
The third caveat is that for the 1st exploit to work an attacker needs a mysql user and password.
There is some good news here. The latest stable versions of MariaDB at least disable symbolic links in my.cnf by default (its been a while since I installed MySQL through the repo but I'm fairly sure its disabled here as well). And how would an attacker get a MySQL user anyway?
Consider that because *any* MySQL user to be used, an un-patched shared server used by a hosting company would depend on the security competency of every one of that c customers to securely handle database authentication. Not only are there a variety of exploits available for obtaining a standard database user, but its depressingly common for web designers to place their connection strings with un-encrypted database username and password into world-readable files. There are a variety of feeds and sites that scan the internet for and compile such files.
And even without the use of the 2nd exploit, an attacker can still do an enormous amount of damage without server root with only the mysql system user. The attacker will have full access to the MySQL system files. An attacker could easily delete an entire database instance, for example.
Of course the best part is that this is a vulnerability in MySQL itself. Upgrading MySQL is the scariest, riskiest upgrade there is among standard repo software. A lot of admins compile it from source or install it from a direct RPM (in which cases symlinks are enabled by default). And applications are closely linked with the database version. Even successful upgrades can easily break applications that run on that database as calls used by the application become deprecated. Upgrading applications has substantial costs, whether you develop the application itself or license it. A patch was already in circulation before these exploits were posted, but for all of the reasons listed above, vulnerable databases will be active for years.
Here are the impacted DB versions:
MariaDB < 5.5.52 < 10.1.18 < 10.0.28 MySQL <= 5.5.51 <= 5.6.32 <= 5.7.14 Percona Server < 5.5.51-38.2 < 5.6.32-78-1 < 5.7.14-8 Percona XtraDB Cluster < 5.6.32-25.17 < 5.7.14-26.17 < 5.5.41-37.0
Here the first two links below contain a comprehensive breakdown of both exploits with example scripts that you can run to test.
https://legalhackers.com/advisories/MySQL-Maria-Percona-RootPrivEsc-CVE-2016-6664-5617-Exploit.html This link includes a video illustrating how a compromise takes place using the example scripts: