CS 550 Operating Systems Spring Inter Process Communication

Transcription

1 CS 550 Operating Systems Spring 2019 Inter Process Communication 1

2 Question? How processes communicate with each other? 2

3 Some simple forms of IPC Parent-child Command-line arguments, wait( ), waitpid( ) exit( ) Reading/modifying common files Servers commonly use pid file to determine other active servers. Signals Event notification from one process to another 3

4 Some more forms of IPC Shared Memory Common piece of read/write memory. Needs synchronization for access Semaphores Pipes Locking and event signaling mechanism between processes Uni-directional (if used cleanly) ps -aux more Sockets Bi-directional Not just across the network, but also between processes. 4

5 Pipes 5

6 Pipe Abstraction Write to one end, read from another pipe( ) fd[1] Pipe fd[0] write( ) read( ) 6

7 Pipe provides a "byte-stream" abstraction You can read and write at arbitrary byte boundaries. E.g. Byte lengths sequence written 10, 10, 10, 10 byte lengths sequence read 5, 15, 15, 5 As opposed to message abstraction, which provides explicit message boundaries. E.g. network packets 7

8 Parent-child communication using pipe Parent pipe( ) fd 0 1 Here s an example. 8

9 int main() { int pfds[2]; char buf[30]; pid_t pid; /* create a pipe */ if (pipe(pfds) == -1) { perror("pipe"); exit(1); /* fork a child */ if ( (pid = fork()) < 0) { perror("fork"); exit(1); if ( pid == 0 ) { printf("child: writing to the pipe\n"); /* close the read end */ close(pfds[0]); /* write message to parent through the write end */ if(write(pfds[1], "Hello", 6) <= 0) { perror("child"); exit(1); printf("child: exiting\n"); exit(0); else { printf("parent: reading from pipe\n"); /* close the write end */ close(pfds[1]); /* read message from child through the read end */ if( read(pfds[0], buf, 6) <= 0 ) { perror("parent"); exit(1); printf("parent: read \"%s\"\n", buf); /* wait for child to complete */ wait(null); 9

10 Parent-child communication using pipe Parent fork( ) Child pipe( ) fd 0 1 fd 0 1 Here s an example. 10

11 int main() { int pfds[2]; char buf[30]; pid_t pid; /* create a pipe */ if (pipe(pfds) == -1) { perror("pipe"); exit(1); /* fork a child */ if ( (pid = fork()) < 0) { perror("fork"); exit(1); if ( pid == 0 ) { printf("child: writing to the pipe\n"); /* close the read end */ close(pfds[0]); /* write message to parent through the write end */ if(write(pfds[1], "Hello", 6) <= 0) { perror("child"); exit(1); printf("child: exiting\n"); exit(0); else { printf("parent: reading from pipe\n"); /* close the write end */ close(pfds[1]); /* read message from child through the read end */ if( read(pfds[0], buf, 6) <= 0 ) { perror("parent"); exit(1); printf("parent: read \"%s\"\n", buf); /* wait for child to complete */ wait(null); 11

12 Parent-child communication using pipe Parent fork( ) Child pipe( ) fd 0 1 X fd X 0 1 Here s an example. 12

13 int main() { int pfds[2]; char buf[30]; pid_t pid; /* create a pipe */ if (pipe(pfds) == -1) { perror("pipe"); exit(1); /* fork a child */ if ( (pid = fork()) < 0) { perror("fork"); exit(1); if ( pid == 0 ) { printf("child: writing to the pipe\n"); /* close the read end */ close(pfds[0]); /* write message to parent through the write end */ if(write(pfds[1], "Hello", 6) <= 0) { perror("child"); exit(1); printf("child: exiting\n"); exit(0); else { printf("parent: reading from pipe\n"); /* close the write end */ close(pfds[1]); /* read message from child through the read end */ if( read(pfds[0], buf, 6) <= 0 ) { perror("parent"); exit(1); printf("parent: read \"%s\"\n", buf); /* wait for child to complete */ wait(null); 13

14 Discussion Now what if the parent wants to send data to the child? What should we do? 14

15 Filters in shell command-line ps -elf less Shell fork() fork() less stdin 0 1 pipe() stdout ps -elf Here s an example. 15

16 int main() { int fds[2]; char buf[30]; pid_t pid1, pid2, pid; int status, i; /* create a pipe */ if (pipe(fds) == -1) { perror("pipe"); exit(1); /* fork first child */ if ( (pid1 = fork()) < 0) { perror("fork"); exit(1); if ( pid1 == 0 ) { close(1); /* close normal stdout (fd = 1) */ dup2(fds[1], 1); /* make stdout same as fds[1] */ close(fds[0]); /* we don't need the read end -- fds[0] */ if( execlp("ps", "ps", "-elf", (char *) 0) < 0) { perror("child"); exit(0); /* control never reaches here */ /* fork second child */ if ( (pid2 = fork()) < 0) { perror("fork"); exit(1); if ( pid2 == 0 ) { close(0); /* close normal stdin (fd = 0)*/ dup2(fds[0],0); /* make stdin same as fds[0] */ close(fds[1]); /* we don't need the write end -- fds[1]*/ if( execlp("less", "less", (char *) 0) < 0) { perror("child"); exit(0); /* control never reaches here */ /* parent doesn't need fds - MUST close - WHY? */ close(fds[0]); close(fds[1]); /* parent waits for children to complete */ for( i=0; i<2; i++) { pid = wait(&status); printf("parent: Child %d completed with status %d\n", pid, status); 16

17 Understanding fds: File-Descriptor Table Each process has a file-descriptor table One entry for each open file File = regular files, stdin, stdout, pipes, I/O devices etc. 0 1 stdin stdout 2 fds[0] fds[1] read end of the pipe write end of the pipe 17

18 Handling long chain of filters Recursive approach ps -elf grep "/usr" wc l 1 bonus point for this practice (show me during office hours before NEXT FRIDAY Feb. 2) 18

19 Being careful with read()/write() read(fds[0], buf, 6); Doesn t mean read will return with 6 bytes of data! It could be less. Why? Some reasons read() could reach end of input stream (EOF). Other endpoint may abruptly close the connection read() could return on a signal. So you MUST incorporate error handling with every I/O call (actually with any system call) 19

20 Error handling { You must First check the return value of every read( )/write( ) system call. Then either Wait to read/write more data OR Handle any error conditions More convinient to write a wrapper function /* Write "n" bytes to a descriptor. */ ssize_t writen(int fd, const void *vptr, size_t n) size_t size_t const char nleft; nwritten; *ptr; ptr = vptr; nleft = n; while (nleft > 0) { if ((nwritten = write(fd, ptr, nleft))<=0){ if (errno == EINTR) nwritten = 0; /* call write() again*/ else return(-1); /* error */ nleft -= nwritten; ptr += nwritten; return(n); 20

21 Shared Memory, Semaphores Man pages : shmget, shmat, shmdt, shmctl, semget, semop, semctl 21

22 Shared Memory Common chunk of read/write memory among processes MAX Create Shared Memory (unique key) PTR Proc. 1 Attach 0 Attach PTR Proc. 2 PTR PTR PTR Proc. 3 Proc. 4 Proc. 5 22

23 Creating Shared Memory int shmget(key_t key, size_t size, int shmflg); Example: key_t key; int shmid; key = ftok( <somefile>", A ); shmid = shmget(key, 1024, 0644 IPC_CREAT); Here s an example. reate.c 23

24 Attach and Detach Shared Memory void *shmat(int shmid, void *shmaddr, int shmflg); int shmdt(void *shmaddr); Example: key_t key; int shmid; char *data; key = ftok("<somefile>", A'); shmid = shmget(key, 1024, 0644); data = shmat(shmid, (void *)0, 0); // read or write something to data here. shmdt(data); Here s an example. ch.c 24

25 Deleting Shared Memory int shmctl(int shmid, int cmd, struct shmid_ds *buf); shmctl(shmid, IPC_RMID, NULL); Example: delete.c 25

26 Command-line IPC control ipcs Lists all IPC objects owned by the user ipcrm Removes specific IPC object 26

27 Signals 27

28 Signals Overview Signal is a notification to a process that an event has occurred. Could come from another process or from the OS Type of event determined by type of signal Try listing all signal types using % kill l Some interesting signals SIGCHLD, SIGKILL, SIGSTOP 28

29 Handling Signals Signals can be caught i.e. an action can be associated with them SIGKILL and SIGSTOP cannot be caught. Actions to signals can be customized using sigaction( ) which associates a signal handler with the signal. Default action for most signals is to terminate the process Except SIGCHLD and SIGURG are ignored by default. Unwanted signals can be ignored Except SIGKILL or SIGSTOP Here s an example. als_ex.c 29

30 More on SIGCHLD Sent to parent when a child process terminates or stops. If act.sa_handler is SIG_IGN SIGCHLD will be ignored (default behavior) If act.sa_flags is SA_NOCLDSTOP SIGCHLD won't be generated when children stop act.sa_flags is SA_NOCLDWAIT children of the calling process will not be transformed into zombies when they terminate. These need to be set in sigaction() before parent calls fork() 30

31 Reading child s exit status without blocking on wait() Parent could install a signal handler for SIGCHLD Call wait( )/waitpid( )inside the signal handler void handle_sigchld(int signo) { pid_t pid; int stat; pid = wait(&stat); //returns without blocking printf( child %d terminated\n, pid); Here s an example. hld.c 31

32 More information Check man sigaction( ) Understand what happens when signal is delivered in the middle of a system call? Different OSes have different behavior. Google for keywords Unix Signals Tons of useful links 32

33 References Unix man pages Advanced Programming in Unix Environment by Richard Stevens 33