This is essentially what a shell does if build a redirection chain, i.e. something like
ls | grep foo | sort | uniq
There are some excellent introducionary texts on Unix programming, in which a simple shell is implemented through the book. And one of the tasks of a shell is redirection. One of these books is "Linux Application Programming" by Michael K. Johnson and Erik W. Troan.
The book's homepage: http://ladweb.net/
To build a chain of redirections for N processes you need N-1 pipes. For each redirection you create a pipe using the pipe(int fds[2]) system call. After fork()ing, but before execving use dup2(int from, int to) to "connect" a pipe's end to the standard input (0) or standard output of each process. Here's a overly simplified code, without error checking:
int pipe_A[2];
int pipe_B[2];
pipe(pipe_A);
pipe(pipe_B);
pid_t pid_A, pid_B, pid_C;
if( !(pid_A = fork()) ) {
dup2(pipe_A[1], 1); /* redirect standard output to pipe_A write end */
execv(...);
}
if( !(pid_B = fork()) ) {
dup2(pipe_A[0], 0); /* redirect standard input to pipe_A read end */
dup2(pipe_B[1], 1); /* redirect standard output to pipe_B write end */
execv(...);
}
if( !(pid_C = fork()) ) {
dup2(pipe_B[0], 0); /* redirect standard input to pipe_B read end */
execv(...);
}
Take note that the pipe's array indices have been choosen in a way that they reflect the standard input/output file descriptors if they're used for stdio redirection. This choice was not arbitrary.
Of course you can connect pipes to any file descriptors (e.g. there are some applications, which expect their parent to open, say fd 3 and 4, connected to pipes) and most shells directly support this, too (for example 1>&3 will redirect stdout into fd 3). However the array indices for pipe(int fds[2]) are 0 and 1 of course. I'm just telling this, because I had some cargo-cult-programming students, which mindlessly took the target fds also for the pipe syscall array.
To wait for all the children to finish use waitpid(-1, NULL, 0) – I think that's the -1 my pre-answerer meant, which means: Wait for all child processes to finish. The other option was calling wait() in a loop which will return the pid of the just terminated child. If called again and there's still a child running, it will block again. If there's no child left, it will return -1; I prefer the waitpid solution.
