After having introduced different types of bots, we now want to take a closer look at what these bots normally do and how they work. This section will in detail explain how bots spread and how they are controlled by their masters.
After successful exploitation, a bot uses Trivial File Transfer Protocol (TFTP), File Transfer Protocol (FTP), HyperText Transfer Protocol (HTTP), or CSend (an IRC extension to send files to other users, comparable to DCC) to transfer itself to the compromised host. The binary is started, and tries to connect to the hard-coded master IRC server. Often a dynamic DNS name is provided (for example one from www.dyndns.org) rather than a hard coded IP address, so the bot can be easily relocated. Some bots even remove themselves if the given master server is localhost or in a private subnet, since this indicates an unusual situations. Using a special crafted nickname like USA|743634 or [UrX]-98439854 the bot tries to join the master's channel, sometimes using a password to keep strangers out of the channel. A typical communication that can be observed after a successful infection looks like:
Afterwards, the server accepts the bot as a client and sends him RPL_ISUPPORT, RPL_MOTDSTART, RPL_MOTD, RPL_ENDOFMOTD or ERR_NOMOTD. Replies starting with RPL_ contain information for the client, for example RPL_ISUPPORT tells the client which features the server understands and RPL_MOTD indicates the Message Of The Day (MOTD). In contrast to this, ERR_NOMOTD is an error message if no MOTD is available. In the following listing, these replies are highlihted with colors:
On RPL_ENDOFMOTD or ERR_NOMOTD, the bot will try to join his master's channel with the provided password:
The bot receives the topic of the channel and interprets it as a command:
Most botnets use a topic command like
The first topic tells the bot to spread further with the help of the LSASS vulnerability. 200 concurrent threads should scan with a delay of 5 seconds for an unlimited time (parameter 0). The scans should be random (parameter -r) and silent (parameter -s), thus avoiding too much traffic due to status reports. In contrast to this, the second example of a possible topic instructs the bot to download a binary from the web and execute it (parameter 1). And if the topic does not contain any instructions for the bot, then it does nothing but idling in the channel, awaiting commands. That is fundamental for most current bots: They do not spread if they are not told to spread in their master's channel.
Upon successful exploitation the bot will message the owner about it, if it has been advised to do so.
Then the IRC server (also called IRC daemon, abbreviated IRCd) will provide the channels userlist. But most botnet owners have modified the IRCd to just send the channel operators to save traffic and disguise the number of bots in the channel.
The controller of a botnet has to authenticate himself to take control over the bots. This authentication is done with the help of a command prefix and the "auth" command. The command prefix is used to login the master on the bots and afterwards he has to authenticate himself. For example,
are commands used on different bots to approve the controller. Again, the "-s" switch in the last example tells the bots to be silent when authenticating their master. Else they reply something like
which can be a lot of traffic if you have 10,000 bots on your network. Once an attacker is authenticated, they can do whatever they want with the bots: Searching for sensitive information on all compromised machines and DCC-sending these files to another machine, DDoS-ing individuals or organizations, or enabling a keylogger and looking for PayPal or eBay account information. These are just a few possible commands, other options have been presented in the previous section. The IRC server that is used to connect all bots is in most cases a compromised box. This is probably because an attacker would not receive
operator-rights on a normal chat network and thus has to set-up their own IRC server which offers more flexibility. Furthermore, we made some other interesting observations: Only beginners start a botnet on a normal IRCd. It is just too obvious you are doing something nasty if you got 1.200 clients named as rbot-<6-digits> reporting scanning results in a channel.
Two different IRC servers software implementation are commonly used to run a botnet: Unreal IRCd and ConferenceRoom:
Common modifications we have noticed are stripping "JOIN", "PART" and "QUIT" messages on channels to avoid unnecessary traffic. In addition, the messages "LUSERS" (information about number of connected clients) and "RPL_ISUPPORT" are removed to hide identity and botnet size. We recently got a win32 binary only copy of a heavily modified Unreal IRCd that was stripped down and optimized. The filenames suggest that this modified IRCd is able to serve 80.000 bots:
As we don't run a 80,000 user botnet and lack 80,000 developers in our group we are not able to verify that information. But probably such huge botnets are used by cyber criminals for "professional" attacks. These kind of networks can cause severe damage since they offer a lot of bandwidth and many targets for identity theft.
Since the people who run botnets often share the same motives (DDoS attacks or other crimes) every bot family has its own set of commands to implement the same goals. Agobot is really nice here: Just grep the source for
RegisterCommand and get the whole command-list with a complete description of all features. Due to the lack of clean design, the whole SDBot family is harder to analyze. Often the command set is changed in various forks of the same bot and thus an automated analysis of the implemented commands is nearly impossible.
If you are interested in learning more about the different bot commands, we have a more detailed overview of command analysis in botnet commands. In addition, if you are interested in learning more about source code of bots, you can find more detail in the separate page on botnet source code.