Memcached security is a hot topic since the sensepost guys released go-derper at blackhat.
The presentation was pretty good and informative, but it seems like the hype around it has left a bunch of people confused. Although much of this was covered in the presentation, it needs to be restated as much as possible.
This is really part of the sysadmin placement test and has nothing to do with memcached in particular, but I’m going to go ahead and mention it anyway.
You always start by firewalling everything and then allowing only stuff you need to pass through to the places you need it to pass through.
I won’t teach you how to use your firewall, but start with the setting that disables all connectivity to your box.
If you’re running a web server, allow connections to port
you also want non-ssl connections, allow port
80. If that’s the
only service you’re providing, then your firewalling is now complete!
I’d like to note that Amazon EC2 does this by default, yet enough firewalls are misconfigured that they felt the need to send out a form mail to many of their users to let them know that they “have at least one security group that allows the whole internet to have access to the port most commonly used by memcached (11211)”.
If your application only runs on one server (with the app and memcached on the same box), you can bind it to localhost by adding
to the memcached flags. Now even though you’ve firewalled access to memcached, you have to be on the machine to even contact the cache when someone breaks your firewall settings.
The latest versions of memcached support SASL authentication.
Although you’ve already firewalled your memcached services off, you can require clients to perform strong authentication before using the service.
You can read more about setting this up in the SASL howto page of the wiki.
memcached does not want to run as root. It tries hard to prevent this. Yet many people have a “workaround” that allows memcached to start as root (which I will not repeat) just for the sake of making their infrastructure less secure.
If someone somehow bypasses the firewall you have set up preventing access to memcached and somehow manages to find a security hole in memcached allowing code execution, do you really want to just hand over root access?
There are no such known issues, but we don’t audit the code to ensure it’s safe to run as root. That’s OK, though, because no responsible sysadmin would ever run a service as root without very strong justification, and probably a lot of work in creating a jailed environment.
Look, I’m not doubting that you know how to set up your firewall, but just bear with me.
Grab nmap or similar. Run a full port scan across your box – one from a trusted system, one from a semi-trusted system, and one from a completely untrusted system.
If there’s any response for any service you cannot justify running, you now know about it and can fix it.
That’s not just memcached – that’s gearman, beanstalkd, snmpd, a mail server, a DNS server, LDAP server, etc…
For any service you do have running and publicly available, make sure you completely understand the security implications of running this service.
Do not be embarrassed to ask if you don’t understand everything. It’s a lot better than being an example in a presentation at the next black hat because you’re running a service you didn’t intend to and you leaked important information.