In a multithreaded process, multiple threads are concurrently executing the same program. All of the threads share the same global and heap variables, but each thread has a private stack for local variables. The threads in a process also share a number of other attributes, including process ID, open file descriptors, signal dispositions, current working directory, and resource limits.
The key difference between threads and processes is the easier sharing of information that threads provide, and this is the main reason that some application designs map better onto a multithread design than onto a multiprocess design. Threads can also provide better performance for some operations (e.g., thread creation is faster than process creation), but this factor is usually secondary in influencing the choice of threads versus processes.
Threads are created using pthread_create(). Each thread can then independently terminate using pthread_exit(). (If any thread calls exit(), then all threads immediately terminate.) Unless a thread has been marked as detached (e.g., via a call to pthread_detach()), it must be joined by another thread using pthread_join(), which returns the termination status of the joined thread.
[Butenhof, 1996] provides an exposition of Pthreads that is both readable and thorough. [Robbins & Robbins, 2003] also provides good coverage of Pthreads. [Tanenbaum, 2007] provides a more theoretical introduction to thread concepts, covering topics such as mutexes, critical regions, conditional variables, and deadlock detection and avoidance. [Vahalia, 1996] provides background on the implementation of threads.