如何使进程知道同一程序的其他进程

Question

I must write a program that must be aware of another instance of itself running on that machine, and communicate with it, then die. I want to know if there is a canonical way of doing that in Linux.

My first thought was to write a file containing the PID of the process somewere, and look for that file every time the program executes, but where is the "right" place and name for that file? Is there a better, or more "correct" way?

Then I must communicate, saying the user tried to run it, but since there is another instance it will hand over the job and exit. I thought of just sending a signal, like SIGUSR1, but that would not allow me to send more information, like the X11 display from where the user executed the second process. How to send this info?

The program is linked against Gtk, so a solution that uses the glib is OK.

Answer 1

Putting the pid in a file is a common way of achieving this. For daemons ("system programs"), the common place to put such a file is /var/run/PROGRAM.pid . For user programs, put the pid file hidden in the user's homedir (if the program also has configuration files, then put both config files and the pid file in a subdir of the home dir).

Sending information to the "master" instance is most commonly achieved using Unix domain sockets, also known as local sockets. With a socket, you won't need a pid file (if no-one listens on the socket, the process knows it's master).

Answer 2

Unix domain sockets. Have the first instance create one in a temporary directory, then have other instances communicate with it via that.

Answer 3

Writing a PID file is a common approach. Check the pidfile(3) library.

Answer 4

Does linux have the equivalent of a named mutex or semaphore? So you can check to see if it's 'locked' and then warn the user they already have one out there and close it out?

does this make sense from this link? http://www.linuxquestions.org/questions/programming-9/named-mutex-in-linux-296816/

Answer 5

There are many ways to do this. The way you proposed (using a file containing the PID) is a valid one and is used by many applications.

Some times the application's configuration file contains the path for the PID file, other times a hardcoded path is used. Usually application put the PID file in /tmp , in /var (if they run with uid 0) or in their local directory ( ~/.application/ ).

Wouldn't have a general suggestion on where to put your PID file, just choose the place you prefer.

Answer 6

You can certainly use a Unix domain socket; I think most applications (which don't use a higher-level system like DCOP or DBUS) use these.

If you're happy for it to be Linux-specific, you can use an "abstract namespace" unix socket; these are rather nice because they don't need to exist in the filesystem.

If your program is user-oriented, it should probably be multiuser aware; one user should not be able to trigger behaviour in another user's copy of the app, and security needs to be in place to ensure that users cannot DoS each other easily either (Example: if user A's copy of the program hangs, does it stop user B's from starting?).