Section A (March 14) Outline Exceptions Process Signals Non-local jumps Reminders Lab4: Due Next Thursday TA: Kun Gao Shamelessly Modified from Minglong Shao’s Recitation, Fall 2004

Exceptional control flow (ECF) Abrupt changes in the control flow React to changes in system state that are not captured by internal program variables and are not necessarily related to the execution of the program Happens at all levels of a computer system Exceptions Concurrent processes Signals Non-local jumps

Exceptions Interrupt (asynchronous exceptions) I/O interrupt, hardware reset, software reset, etc. Traps System calls, breakpoint traps, etc. Faults Page fault, protection fault, etc. Aborts Parity error, machine check, etc.

Process concept An instance of running program Multiple processes run “concurrently” by time slicing Context switching Control flow passes from one process to another Preemptive scheduler of OS Process vs Threads (a favorite interview question) Threads are logical flows that run in the context of a single process

Process IDs & process groups A process has its own, unique process ID pid_t getpid(); A process belongs to exactly one process group pid_t getpgrp(); A new process belongs to which process group? Its parent’s process group A process can make a process group for itself and its children setpgid(0, 0);

Create a new process int fork(void) Create a new process that is identical to the parent process Return 0 to child process Return child’s pid to the parent process Call once, return twice Test your understanding… Problem 1

#include int cnt = 0; int main(void) { if (fork() == 0){ cnt ++; fork(); cnt++; } cnt ++; printf("%d", cnt); return 0; } Possible output:

Reaping child process Child process becomes zombie when terminates Still consume system resources Parent performs reaping on terminated child pid_t wait(int *status) pid_t waitpid(pid_t pid, int *status, int options) Straightforward for reaping a single child Tricky for Shell implementation! Multiple child processes Both foreground and background

Signals Section 8.5 in text Read at least twice … really! A signal tells our program that some event has occurred

Important signals (Fig 8.23) SIGINT Interrupt signal from terminal (ctrl-c) SIGTSTP Stop signal from terminal (ctrl-z) SIGCHLD A child process has stopped or terminated

How to Send Signals Process int kill(pid_t pid, int sig) Groups int kill(pid_t gid, int sig), where gid is negative Process can also send a signal to itself int alarm(unsigned int secs) to send a SIGALRM signal Can we use signals to count events? No Why? Signals not queued!!

Signals: sending OS Kernel blocked pending 1 Process 1 Process 2 other events OS procedure kill(pid, SIGINT) divide by zero: SIGFPE ctrl-c: SIGINT child process exit: SIGCHLD

Signals: receiving OS Kernel blocked pending 1 Process 2 OS procedure 0 Check when schedule the process to run

Receiving a signal Default action The process terminates [and dumps core] The process stops until restarted by a SIGCONT signal (ctrl-z) The process ignore the signal Can modify (additional action) “Handle the signal” -- install signal handler void sigint_handler(int sig); signal(SIGINT, sigint_handler); An example: problem 3

Problem 3 void handler(int sig){ static int beeps = 0; printf("YO\n"); if (++beeps < 2) alarm(1); /* next SIGALRM will be delivered in 1s */ else{ printf("MA\n"); kill(getpid(), SIGKILL); } int main(){ signal(SIGALRM, handler); alarm(1); /* next SIGALRM will be delivered in 1s */ while (1) ; printf(" is Great!\n"); return 0; } 1. Output: YO MA 2. The program will terminate

Signals not queued int counter = 0; void handler(int sig) { counter++; sleep(1); return; } int main() { int i; signal(SIGUSER2, handler); if (fork() == 0){ for (i = 0; i < 5; i++){ kill(getppid(), SIGUSR2); printf(“sent SIGUSR2 to parent\n”); } exit(0); } wait(NULL); printf(“counter = %d\n”, counter); exit(0); } Output: sent SIGUSR2 to parent counter = 1

Race hazard A data structure is shared by two pieces of code that can run concurrently Different behaviors of program depending upon how the schedule interleaves the execution of code.

An example of race hazard sigchld_handler() { pid = waitpid(…); deletejob(pid); } eval() { pid = fork(); if(pid == 0) { /* child */ execve(…); } /* parent */ /* signal handler may run BEFORE addjob()*/ addjob(…); }

An okay schedule Shell Signal HandlerChild fork() addjob() execve() exit() sigchld_handler() deletejobs() time

A problematic schedule Shell Signal HandlerChild fork() execve() exit() sigchld_handler() deletejobs() time addjob() Job added to job list after the signal handler tried to delete it!

Solution: blocking signals sigchld_handler() { pid = waitpid(…); deletejob(pid); } eval() { sigprocmask(SIG_BLOCK, …) pid = fork(); if(pid == 0) { /* child */ sigprocmask(SIG_UNBLOCK, …) /* child inherits parents block set */ execve(…); } /* parent */ /* signal handler might run BEFORE addjob() */ addjob(…); sigprocmask(SIG_UNBLOCK, …) } More details (page 633)

Non-local jump int setjmp(jmp_buf env) Must called before longjmp Stores current register context, stack pointer, and PC in the jump buffer env First called, return 0 if no error void longjmp(jmp_buf env, int i) Restores register context from jump buffer env Jumps back to where previous setjmp is called, behaves like setjmp just completes. But this time it returns i, not 0 Can only jump to an active context A function that has been called but not yet completed

Non-local jump (cont) int sigsetjmp(jmp_buf env) Also saves blocked signals void siglongjmp(jmp_buf env, int i) Restores blocked signals besides others Let’s see an example: problem 4

Problem 4 jmp_buf stuff; int foo(){ char c = getc(stdin); if (c == 'x') longjmp(stuff, 42); else return 3; return -1; } int main() { int n; n = setjmp(stuff); while ((n+=foo())<0) sleep(1); printf("%d\n", n); } Answer:

Summary Process fork(), waitpid(), execve() Reaping child processes Signals signal(), install handler, signals not queued Non-local jumps Check man page to understand the system calls better man waitpid (fork, signal, …) Read text book!