Now that we've talked a bit about how processes and system calls work at a high level, it's time to apply these concepts by doing some exercises related to process creation and making system calls. We'll be utilizing the fork(), exec(), wait(), and pipe() system calls in order to create processes and even have them pass messages to each other.
To introduce and familiarize yourself with some basic system calls pertaining to process creation, to spawn some new processes, and to practice writing more C!
The fork() system call is used by a parent process to create a new child process. Its
actual implementation isn't as intuitive as it could be, though. When a parent process
executes fork(), the new child process that is created is an almost exact copy of the
calling process from the operating system's perspective. We say almost an exact copy
to delineate the fact that while the new child process has the same instruction set
(i.e. code) as its parent process, the child process starts executing at the point right
after fork() is called in the parent process.
Let's look at a program that calls fork() to try to give an example of what this means:
// p1.c
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
int main(int argc, char *argv[])
{
printf("hello world (pid: %d)\n", (int) getpid());
int rc = fork();
// ------------------------------------------------ child process starts executing here
if (rc < 0) { // fork failed; exit
fprintf(stderr, "fork failed\n");
exit(1);
} else if (rc == 0) { // child process satisfies this branch
printf("hello, child here (pid: %d) \n", (int) getpid());
} else {
printf("hello, parent here (pid: %d) of child %d\n", (int) getpid(), rc);
}
return 0;
}Running this program, one of the possible outputs could be something like this:
prompt> ./p1
hello world (pid: 29146)
hello, parent here (pid: 29146) of child 29147
hello, child here (pid: 29147)
prompt>
So as we can see, the child process doesn't return the exact same output as its parent process, which is good, because otherwise spawning child processes wouldn't be nearly as useful. Notice that the child process has its own process identifier (PID); it also has its own address space, program counter, and execution context that are not the same as its parent's.
More specifically, if we look at the int rc = fork(); line, which both parent and chlid execute, they
receive different results here. The parent process receives the PID of the child process it just spawned,
while the child process receives 0 if the fork() call was successful.
So fork() allows us to spawn new child processes, but the thing with having multiple processes is that
the order in which they execute is not deterministic, i.e., the parent may execute before the child or
the child may execute before the parent; we can't say for certain. The wait() system call (along with
the more complete waitpid() system call) allows us to get around this non-determinism if that is something
that needs to be accounted for. Let's add a call to waitpid() in our previous example program to ensure that
the child always runs before its parent:
// p2.c
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/wait.h> // `waitpid` needs to be included separately
int main(int argc, char *argv[])
{
printf("hello world (pid: %d)\n", (int) getpid());
int rc = fork();
// ------------------------------------------------ child process starts executing here
if (rc < 0) { // fork failed; exit
fprintf(stderr, "fork failed\n");
exit(1);
} else if (rc == 0) { // child process satisfies this branch
printf("hello, child here (pid: %d) \n", (int) getpid());
} else {
int wc = waitpid(rc, NULL, 0); // `waitpid` call added here
printf("hello, parent here (pid: %d) of child %d\n", (int) getpid(), rc);
}
return 0;
}Here, the waitpid() function suspends the parent process until the child process pointed at by rc terminates. Thus, we ensure that the parent process only runs after the child process has finished its execution.
The exec() system call is used in order to run a program that is different from the calling program (since fork only executes copies of the program that called it).
Let's say we wanted to spin up a child process to execute a word count program. Here's how what a program that does that might look like:
// p3.c
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/wait.h>
#include <string.h> // this header file is needed in order to work with strings
int main(int argc, char *argv[])
{
printf("hello world (pid: %d)\n", (int) getpid());
int rc = fork();
// ------------------------------------------------ child process starts executing here
if (rc < 0) { // fork failed; exit
fprintf(stderr, "fork failed\n");
exit(1);
} else if (rc == 0) { // child process satisfies this branch
printf("hello, child here (pid: %d) \n", (int) getpid());
char *myargs[3]; // allocate an array of chars to hold 3 bytes
// `strdup` duplicates the given input string
myargs[0] = strdup("wc"); // pass the name of the program we want to run as the first array element
myargs[1] = strdup("p3.c"); // argument: the file to count
myargs[2] = NULL; // marks the end of the array
execvp(myargs[0], myargs); // runs the word count program, passing in the `myargs` array to the word count program as arguments
printf("this should not be seen");
} else {
int wc = waitpid(rc, NULL, 0); // `waitpid` call added here
printf("hello, parent here (pid: %d) of child %d\n", (int) getpid(), rc);
}
return 0;
}Here, we're doing the same thing as before, forking a new child process from a parent process. Then, inside that
child process, we're calling execvp() with the arguments it needs to run the word count program. Note that exec
does not spin up another child process from the child process in which we called it. It transforms the process
that called it into the new program that was passed to exec, in this case, wc, the word count process. That's
why we still had to have the parent process fork a new child process.
Conceptually, a pipe is a uni-directional channel between two processes that would otherwise be isolated from each other. When a pipe is established between two processes, one process has access to the write end of the pipe, while the other has read access to the other end of the pipe. Thus, if you want two-way communication between two processes, two pipes will have to be created between both processes, one in each direction.
Some things to keep in mind when working with pipes is that when a process writes to a pipe, the other process receives the data in FIFO order (which makes sense when you think about the pipe analogy in real life). Additionally, if the process with read access tries to read from the pipe before anything has been written to it, the reading process is suspended until there is some data to read.
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define MSGSIZE 16 // define a constant message size for each message
char* msg1 = "hello world #1";
char* msg2 = "hello world #2";
char* msg3 = "hello world #3";
int main()
{
char inbuf[MSGSIZE]; // a buffer that will hold the incoming data that is being written
int p[2]; // a two-element array to hold the read and write file descriptors that are used by the pipe
// establish our pipe, passing it the p array so that it gets populated by the read and write file descriptors
if (pipe(p) < 0) {
fprintf(stderr, "pipe failed\n");
exit(1);
}
// write 16 bytes of data to the write file descriptor
write(p[1], msg1, MSGSIZE);
write(p[1], msg2, MSGSIZE);
write(p[1], msg3, MSGSIZE);
for (int i = 0; i < 3; i++) {
// read 16 bytes of data from the read file descriptor
read(p[0], inbuf, MSGSIZE);
printf("% s\n", inbuf);
}
return 0;
}This program isn't actually sending data between two different processes. The process that is doing the writing is also doing the reading. This process is essentially just talking to itself. But at least it gives you an idea of how to create pipes and send data between both ends.
This was your first introduction to making system calls to an operating system kernel. Obviously, there are many more system calls that we'll talk about in a future lesson, but for now we'll just practice with these. Once you've cloned this repo, go into each directory, open up the exercise file, read the prompt, and write some C code! Compile your code with gcc [NAME OF FILE].c -o [NAME OF EXECUTABLE], then run the executable with ./[NAME OF EXECUTABLE].
If you would like to read more on this topic, these chapters from Operating Systems: Three Easy Pieces heavily informed this particular topic and material: http://pages.cs.wisc.edu/~remzi/OSTEP/cpu-api.pdf http://pages.cs.wisc.edu/~remzi/OSTEP/cpu-mechanisms.pdf
Additionally, it's always a good idea to read the man pages for any system calls you're trying to use:
- The man page for
fork: https://linux.die.net/man/2/fork - The man page for
waitandwaitpid: https://linux.die.net/man/3/waitpid - The man page for the
execfamily: https://linux.die.net/man/3/exec - The man page for the
pipesystem call: https://linux.die.net/man/2/pipe
Open up the /stretch directory. In there, you'll find an involved exercise pertaining to file locking and concurrency. Read the README included in that directory for instructions on what to do.