BMCs are those little mini-computers in your bare-metal servers. You know, the things with the LEDs in the big cabinets that generate heat and noise. But I love them :)
Without getting much too technical here, IPMI is the name of the interface specification and RMCP is the underlying protocol (for LAN transmission it runs on UDP, TCP’s connectionless sister protocol).
Since these little “BMC computers” are tightly interwoven with the rest of the server’s components and have their own network interfaces, they are of course a good target for attacks of all kinds.
In the past, among others, two types of attacks from the network were very popular: cipher 0 and dumping the hashes.
Cipher 0 (as in “zero”) made it possible to make changes to the system without a correct password and ultimately log in with elevated privileges. And this for the usual manufacturers HP (iLO), Dell (iDRAC) and Supermicro, who have since fixed this with patches.
Another attack is to read the password hashes of IPMI users. When I try this on test servers using the usual pentest frameworks or just ipmitool, I get partial hits. Does the management software find this suspicious? Well, at least HP doesn’t, there are only IPMI/RMCP login/logoff events when scanning with severity “informational”. So if SNMB traps are configured, none is generated here.
Dan Farmer has written some useful articles: