Today’s topic

•Inter-process communication with pipes

• More about files and I/O.– cin read (0, …)

– cout write (1, …)

– What happens when we use both I/O mechanisms in the same program?• Check out the output of mix.cpp?

– Why?– The liberty in the implementation of the

C/C++ runtime library

• How to fix the order? fflush(0).

• More on the timestamp of a file– How to get the info? Stat– See newstat.c for comparing the timestamp of

two files.

• Inter-Process Communication (IPC):– processes may be collaborated on tasks– e.g,

ps -ef | grep a.out | more– e.g,

– OS needs to provide mechanisms for IPC

client server

• IPC related system calls:– the most general IPC mechanism in UNIX is

socket (send and receive paradigm with read/write interface, works also for processes on different machines).

– What we will discuss:• pipes: allow transfer of data between processes in a

first-in-first-out manner.

• signal: send a flag to another process

• Pipes:– two types of pipes, named pipes and unnamed

pipes– name pipes:

• like a file (create a named pipe (mknod), open, read/write)

• can be shared by any number of processes

• will not be discussed in detail.

– Unnamed pipes:• an unnamed pipe does not associated with any file

• can only be shared by related processes (descendants of a process that creates the unnamed pipe).

• Created using system call pipe().

• Pipes and files– Both can be used to share information among

processes– Both share the file API

– Performance difference: • A pipe is usually realized in memory, a file is not.• Pipe operations are memory operations, file

operations are I/O operations.

– Semantic difference:• A pipe is a fifo queue:

– The content in a fifo queue can only be read once (the information is gone after the read operation).

• A storage in a file is persistent– The content can be used many times.

• The pipe system call– open unnamed pipes– syntax

int pipe(int fds[2])

– semanticcreate a pipe and returns two file descriptors fds[0] and

fds[1]

a read from fds[0] accesses the data written to fds[1] on a fifo basis.

the pipe has a limited size (64K in most systems) -- cannot write to the pipe infinitely.

#include <unistd.h>

#include <stdio.h>

main()

{

char *s, buf[1024];

int fds[2];

s = “hello world\n”;

pipe(fds);

write(fds[1], s, strlen(s));

read(fds[0], buf, strlen(s));

printf(“fds[0] = %d, fds[1] = %d\n”, fds[0], fds[1]);

write(1, buf, strlen(s));

}

/* example1.c */

#include <unistd.h>#include <stdio.h>main(){ char *s, buf[1024]; int fds[2]; s = “hello world\n”; pipe(fds); if (fork() == 0) { printf(“child process: \n”); write(fds[1], s, 12); exit(0); } read(fds[0], buf, 6); write(1, buf, 6);}/* example2.c : using pipe with fork*/IPC can be used to enforce the order of the execution of processes.

main()

{

char *s, buf[1024]; 11111 33333 11111 33333

int fds[2]; 22222 44444 33333 11111

s = “hello world\n”; 33333 11111 22222 22222

pipe(fds); 44444 22222 44444 44444

if (fork() == 0) {

printf(“11111 \n”); /* how to make 111111 before 444444 */

read(fds[0], s, 6);

printf(“22222\n”);

} else {

printf(“33333\n”);

write(fds[1], buf, 6);

printf(“44444\n”)

}

} /* example3.c */

• Anyone writes any programs that take advantage of multi-core in a CPU?– A multiple process solution for computing

PI?• See pi1.c and pi2.c?

• Implementing Pipe in shell.E.g. /usr/bin/ps -ef | /usr/bin/more

• How shell realizes this command?– Create a process to run ps -ef– Create a process to run more– Create a pipe from ps -ef to more

• the standard output of the process to run ps -ef is redirected to a pipe streaming to the process to run more

• the standard input of the process to run more is redirected to be the pipe from the process running ps -ef

• Implement “/bin/ps -ef | /bin/more” – first trymain() { int fds[2]; char *argv[3]; pipe(fds); // create pipe if (fork() == 0) { close(0); dup(fds[0]); // redirect standard input to fds[0] argv[0] = “/bin/more”; argv[1] = 0; if (execv(argv[0], argv) == -1) exit(0); } if (fork() == 0) { close(1); dup(fds[1]); // redirect standard output to fds[1]; argv[0] = “/bin/ps”; argv[1] = “-ef”; argv[2] = 0; if (execv(argv[0], argv) == -1) exit(0); } wait(); wait();} /* example4a.c */

• Example4a.c not the same as “/bin/ps –ef | /bin/more”, what is missing?– When can ‘more’ be done?

• When the end of the file is reached.

– What is “the end of file” of a pipe?• No data in pipe?

• No data in pipe and no potential writer to the pipe!!!!

– In example4a.c, the more program will not finish since there are potential writers.

• How to fix this program?

• Implement “/bin/ps -ef | /bin/more”main() { int fds[2]; char *argv[3]; pipe(fds); // create pipe if (fork() == 0) { close(0); dup(fds[1]); // redirect standard input to fds[1] close(fds[0]); close(fds[1]); // close unused files argv[0] = “/bin/more”; argv[1] = 0; execv(argv[0], argv); exit(0); } if (fork() == 0) { close(1); dup(fds[0]); // redirect standard output to fds[0]; close(fds[0]); close(fds[1]); // close unused files argv[0] = “/bin/ps”; argv[1] = “-ef”; argv[2] = 0; execv(argv[0], argv); exit(0); } close(fds[0]); close(fds[1]); wait(); wait();} /* example4.c */