Learning Objectives

At the end of this lecture, you should be able to:

• Define inter-process-communication (IPC) and some of its mechanisms.
• Implement signal sending, receiving, and handling in a Linux environment.

Topics

In this lecture, we will cover the following topics:

• Inter Process Communication.
• Signals.

Notes

• Definition
• The life of a signal
• Signal Delivery

Definition

• Consider an example where you are working at your desk, and you get a text message from your crush or your significant other, what would you do at that point in time?
• What did your partner do?
  • Communicate with you by sending a signal
• Your current work will be interrupted, you go to handle the corresponding signal, and then you return to where you were in the studying
• What if the text message was from your mom or dad? What would be your reaction then? probably ignore right?!
• Now in terms of the processes that are running, they can communicate with each other with the use of signals
• A signal is a short message sent to a particular task, process, or group of processes
  • Signals allow for IPC: Inter Process Communication
• In most circumstances, only the signal number is delivered the signalled task, and no further information
  • So you cannot know who sent you the signal, only that a signal has been delivered to you
  • There are other ways to tell the a task more information, using the siginfo method to deliver more information, but we won’t worry about that one for now.
  • A signal serves two roles:
    1. Notification: A task is notified that a signal has been sent to it
    2. Handling action: A task is caused to break current execution to handle the signal, but it can customize the way it can handle signals, including blocking or ignoring them.
       • Cannot ignore to substitute a handler for SIGSTOP and SIGKILL
       • Imagine if all processes could not be stopped or killed, what would happen in that case?

The life of a signal

• A signal goes through two phases:
  1. The signal generation phase: This is sent from the sending task with a specified destination, the OS will handle the delivery
  2. The signal delivery phase: The OS will try to deliver the signal to the destination task or process or group of processes
• The OS acts as the mailman between the sending and the receiving tasks.
• Examples:

/* The hardware will send the SIGFPE to the task */
float x, y = 0.0;

x = x / y;

/* We will receive a SIGSEGV to the task from the kernel */
int *ptr = 0;

*ptr = 3;

• Most common signals that we will use:
  • SIGSEGV: segmentation fault
  • SIGINT: Interrupt from keyboard (Ctl - C)
  • SIGSTOP: Stop process, cannot be caught or ignored
  • SIGCONT: Continue signal, sent after a stop
  • SIGKILL: Kill process, cannot be caught or ignored
  • SIGTERM: Terminate signal, can be caught or ignored
• From your source code, you can use the routine kill(PID, signal_number)to send a signal to a specific task having a certain PID

Signal Delivery

• When attempting to deliver a signal, the OS will check what the process has chosen to do with that signal
• A process can choose one of several default actions:
  • Terminate: The process or task is terminated
  • Dump: Terminate by generating a core dump (i.e., a stack trace of the current process) that can be read using gdb
  • Ignore: Simply ignore the signal
  • Stop: Put the process in a stopped state and do not schedule it any further
  • Continue: Continue from a stopped state
• In addition, a task may choose to install a customer handler to handle receiving a specific signal (with some exceptions as noted above)
  • We can achieve that by registering a function with the operating system
  • It is like we tell the OS that when we receive a specific signal, break the current execution, execute that custom function, and then go back to current state if not terminated
• Two ways to achieve this:

/* This way is now deprecated, it's okay to use in this class, but better
 * learn the correct way to do so as well */
void signal_handler(int signum) {
	/* Add code here */
}

signal(SIGINT, signal_handler);

struct sigaction new_action, old_action;

new_action.sa_handler = signal_handler;
sigmeptyset(&new_action.sa_mask);
new_action.sa_flags = 0;

sigaction(SIGINT, &new_action, NULL);

• Finally, a task may chose to temporarily block a signal. Blocking a signal is NOT the same as ignoring a signal. All what happens is that the signal is not instantly delivered, it remains as pending until the task decides to unblock that signal type.
  • The signal is delivered but not received
• Signals are NOT queued
  • Only one signal of a given type is dealt with a time
  • If a signal is blocked, it is set to pending until the task unblocks it and consumes the signal