Managing Inter-Process Communication (IPC)


Sven Vermeulen Sun 30 March 2014

As a Linux administrator, you'll eventually need to concern you about Inter-Process Communication (IPC). The IPC primitives that most POSIX operating systems provide are semaphores, shared memory and message queues. On Linux, the first utility that helps you with those primitives is ipcs. Let's start with semaphores first.

Semaphores in general are integer variables that have a positive value, and are accessible by multiple processes (users/tasks/whatever). The idea behind a semaphore is that it is used to streamline access to a shared resource. For instance, a device' control channel might be used by multiple applications, but only one application at a time is allowed to put something on the channel. Through semaphores, applications check the semaphore value. If it is zero, they wait. If it is higher, they attempt decrement the semaphore. If it fails (because another application in the mean time has decremented the semaphore) then the application waits, otherwise it continues as it has successfully decremented the semaphore. In effect, it acts as a sort-of lock towards a common resource.

An example you can come across is with ALSA. Some of the ALSA plugins (such as dmix) use IPC semaphores to allow multiple ALSA applications to connect to and use the sound subsystem. When an ALSA-enabled application is using the sound system, you'll see that a semaphore is active:

~$ ipcs -s
------ Semaphore Arrays --------
key        semid      owner      perms      nsems     
0x0056a4d5 32768      swift      660        1

More information about a particular semaphore can be obtained using ipcs -s -i SEMID where SEMID is the value in the semid column:

~$ ipcs -s -i 32768
Semaphore Array semid=32768
uid=1001         gid=18  cuid=1001       cgid=100
mode=0660, access_perms=0660
nsems = 1
otime = Sun Mar 30 12:33:46 2014  
ctime = Sun Mar 30 12:33:38 2014  
semnum     value      ncount     zcount     pid       
0          0          0          0          32061

As with all IPC resources, we have information about the owner of the semaphore (uid and gid), the creator of the semaphore (cuid and cgid) as well as its access mask, similar to the file access mask on Linux systems (mode and access_perms). Specific to the IPC semaphore, you can also notice the nsems = 1. Unlike the general semaphores, IPC semaphores are actually a wrapper around one or more "real" semaphores. The nsems variable shows how many "real" semaphores are handled by the IPC semaphore.

Another very popular IPC resource is shared memory. This is memory that is accessible by multiple applications, and provides a very versatile approach to sharing information and collaboration between processes. Usually, a semaphore is also used to govern writes and reads to the shared memory, so that one process that wants to update a part of the shared memory takes a semaphore (a sort-of lock), makes the updates, and then increments the semaphore again.

You can see the currently defined shared memory using ipcs -m:

~$ ipcs -m
------ Shared Memory Segments --------
key        shmid      owner      perms      bytes      nattch     status      
0x00000000 655370     swift      600        393216     2          dest

Again, more information can be obtained through -i SHMID. An interesting value to look at as well is the creator PID (just in case the process still runs, or through the audit logs) and the last PID used to operate on the shared memory (which also might no longer exist, but is still an important value to investigate).

~$ ipcs -m -p
------ Shared Memory Creator/Last-op PIDs --------
shmid      owner      cpid       lpid      
655370     swift      6147       6017

~$ ps -ef | grep -E '(6147|6017)'
root      6017  6016  0 09:49 tty1     00:01:30 /usr/bin/X -nolisten tcp :0 -auth /home/swift/.serverauth.6000
swift     6147     1  2 09:50 tty1     00:05:10 firefox

In this case, the shared memory is most likely used to handle the UI of firefox towards the X server.

A last IPC resource are message queues, through which processes can put messages on a queue and remove messages (by reading them) from the queue. I don't have an example at hand for the moment, but just like semaphores and shared memory, queues can be looked at through ipcs -q with more information being available through ipcs -q -i MSQID.

Now what if you need to operate these? For this, you can use ipcrm to remove an IPC resource whereas ipcmk can be used to create one (although the latter is not that often used for administrative purposes, whereas ipcrm can help you troubleshoot and fix issues without having to reboot a system). Of course, removing IPC resources from the system should only be done when there is a bug in the application(s) that use it (for instance, a process decreased a semaphore and then crashed - in that case, remove the semaphore and start one of the application(s) that also operates on the semaphore as they usually recreate it and continue happily).

Now before finishing this post, I do need to tell you about the difference between an IPC resource key and its identifier. The key is like a path or URL, and is a value used by the applications to find and obtain existing IPC resources (something like, "give me the list of semaphores that I can access with key 12345"). The identifier is a unique ID generated by the Linux kernel at the moment that the IPC resource is created. Unlike the key, which can be used for multiple IPC resources, the identifier is unique. This is why the identifier is used in the ipcs -i command rather than the key. Also, that means that if applications would properly document their IPC usage then we would easily know what an IPC resource is used for.