TCP session hijacking is when a hacker takes over a TCP session between two machines. Since most authentication only occurs at the start of a TCP session, this allows the hacker to gain access to a machine.
A popular method is using source-routed IP packets. This allows a hacker at point A on the network to participate in a conversation between B and C by encouraging the IP packets to pass through its machine.
If source-routing is turned off, the hacker can use “blind” hijacking, whereby it guesses the responses of the two machines. Thus, the hacker can send a command, but can never see the response. However, a common command would be to set a password allowing access from somewhere else on the net.
A hacker can also be “inline” between B and C using a sniffing program to watch the conversation. This is known as a “man-in-the-middle attack”.
A common component of such an attack is to execute a denial-of-service (DoS) attack against one end-point to stop it from responding. This attack can be either against the machine to force it to crash, or against the network connection to force heavy packet loss.