how to use kill syscall

shifter · 05-27-2007, 01:52 PM

Quote:

What parameters do you pass to the command ?

I'd like create a daemon to monitor filesystem, your name is 'myprogram'.

./myprogram -a /directorytomonitor

./myprogram -s (stop daemon)

Quote:

What is the intended purpose of the daemonize function ?

Making daemon, that is to say daemonize() permit to run program in background, without daemonize() this program remain in foreground.

Quote:

What process do you want to kill ?

I want kill myprogram using kill syscall, or other system calls to monitor to kill processes...

Quote:

Why don't you check and print the kill return value to exhibit the problem you are complaining of ?

It is done, there are many tests:

case 's': pid = getpid();
printf("pid = getpid() : %d \n",pid);
--pid;
printf("--pid : %d \n",pid);
kill(pid,15);
printf("PID %d \n",pid);
break;

jtshaw · 05-27-2007, 02:28 PM

That is never going to work the way you have set it up. You can't guarantee that getPid() - 1 is going to equal the pid of the program the last time you ran it. You never know what you might be killing if you just do a getPid() and kill the result - 1.

What most programs which need to do this do is write a file to /var/run like /var/run/myprogram.pid which just contains the pid of the daemonized program. When the program exits normally it should remove that file. If you want to kill the program then you'd read the pid out of the file, kill that pid, and if the kill was successful you'd remove the file.

This make sense?

jlliagre · 05-27-2007, 03:03 PM

As jtshaw wrote, you need to store the daemon pid somewhere. Successive execution of the same program won't share variables magically.

Two more comments:

- Your daemonize function should take care of file descriptors 0, 1 and 2 (stdin, out and err) otherwise it may mess the launching shell.

Quote:

It is done, there are many tests:
...

I see no test for the kill syscall, you should do something like:

Code:

#include <signal.h>
..
if(kill(pid,SIGTERM)==-1)
{
  perror("kill");
}
...

shifter · 05-28-2007, 09:47 AM

The test is following:

#./myprogram -a /root

#pstree -p
.....
myprogram(2756)
.....

#./myprogram -s

pid = getpid() : 2774
--pid : 2773
kill: No such process
PID 2773

How can I to obtain process of my daemon? I don't want to use system syscall....

jtshaw · 05-28-2007, 10:04 AM

As we have already told you... you cannot assume that subtracting 1 from your pid is going to get the daemon.

A little background.....

The process ID is simply just some unique number for each individual process on a system. Typically what linux does is just take the next unused PID when a process starts. Once it reaches the maximum value for a PID it starts over again at the beginning and finds the next unused.

If you use some of the extra security options when you build the kernel (grsec/pax patches provides these) then it simply picks a pseudo-random pid from the unused list each time a new process starts.

Either way, while it isn't impossible that starting two processes back to back will cause you to get sequential pids you certainly cannot count on that behavior given the true multiuser multitasking nature of the Linux OS.

When you start up your daemon you are going to have to store the PID of the daemon process somewhere (often in a file such as /var/run/<programname>.pid or /var/run/<programname>/pid). Then you will have to look for that file and read the pid from it to then kill the appropriate thing.

You could also search in running process... but that gets a bit messy, especially if your process ever becomes multithreaded.

I also what to emphasis something that Jilliagre said... kill, like many syscalls, returns a value. You should ALWAYS check the return values of these system calls and handle them accordingly, even if all you do is log it and die. This may seem annoying at times, but it can save you from some nasty debugging down the road.