2016 twenty four merry days of Perl Feed

Using containers with Linux

Linux::Clone Linux::Unshare Linux::Setns - 2016-12-19

Linux containers are a hot topic these days. This is why I decided to share with you, how you can interact with containers directly from Perl.

Linux Namespaces

Linux containers are an OS-level virtualization technique to help processes co- exist on the same machine regardless of what other processes are already running on that machine (for example, having two processes listening on port 80 on the same physical hardware). Rather than starting a whole new virtual machine, including a new operating system and system daemons, Linux containers are a way to simply and sufficiently isolate a process from other processes running on the same operating system beyond the normal process management Linux provides.

Currently in Linux there are six namespaces which we can control to allow processes to:

  • UTS - have different hostname and domain name

  • PID - have its own view of the PIDs of the machine. This allows the process to create processes with PIDs that are already existing on the machine.

  • NET - see different, limited version of the network infrastructure of the kernel. As if the process has its own network.

  • IPC - have a separate SHM, SEM and MQ identifiers

  • USER - have separate user and group identifiers

  • MOUNT - see different view of the mounted filesystems

For the purpose of this article I will assume that a "container" is a processes or group of processes that has one or more namespaces different from the initial namespaces.

There are 3 modules on the CPAN for working with "containers":

Linux::Clone

So let's start with Linux::Clone. This module is basically wrapper to the glibc clone(2) wrapper function. I will cover only the parts of it related to Linux namespaces. These are CLONE_NEWNET, CLONE_NEWPID, CLONE_NEWIPC, CLONE_NEWUTS, CLONE_NEWNS(mount namespace), CLONE_NEWUSER.

Namespaces are used to create isolated environment where your process have only limited visibility to the system.

For example if you want to make sure that only processes you have started can use the SHM (shared memory) you created, you can use Linux::Clone to create a new process with its own IPC namespace and after that create your SHM. This way you are both protecting your SHM from others on the machine and protecting everyone else from your process.

This is how the code will look in your Perl application:


1: 
2: 
3: 
4: 
5: 
6: 
7: 
8: 

 

use Linux::Clone;
use POSIX;

sub child {
    print "In the child\n";
    system("ipcs");
}
Linux::Clone::clone sub { child; }, 0, Linux::Clone::NEWIPC || POSIX::SIGCHILD;

 

Normally running the ipcs command on a normal laptop/desktop machine will print out a few entries listing all the ipc facilities that your process normally has access to. However, when ipcs is called in the above script it prints nothing - the new namespace the process is in has nothing in it.

Linux::Unshare

If you don't want to create a new process, but you want to change some of the namespaces for your current process you can use Linux::Unshare. This module implements the glibc unshare(2) wrapper function, which does exactly this.


1: 
2: 
3: 
4: 

 

use Linux::Unshare qw(unshare CLONE_NEWIPC);

unshare(CLONE_NEWIPC);
system("ipcs");

 

Now the above example will create a new IPC namespace without creating new process and it will replace your current IPC namespace with the newly created one. However, if you want to return to your previous IPC namespace, you would need to make sure you have a file descriptor from that IPC namespace and use Linux::Setns (which I'll cover next) to switch your current IPC namespace to the previous one.

Linux::Setns

Usually you would not create containers directly from your application, instead you probably would use something like LXC, LXD, Rocket or Docker to create and mange your containers. So most likely what you would want to do is attach/enter into these containers and do some work there.

For this you can use the Linux::Setns module, a wrapper for the setns(2) glibc function.

In order to identify a namespace you would use the files in /proc/PID/ns/{ipc,mnt,net,pid,user,uts}. With the setns function, you can join one or more of these namespaces or all of them. This way your process can actually enter in each namespace and do the rest of its work within the confines of that namespace.


1: 
2: 
3: 
4: 

 

use Linux::Setns qw(setns CLONE_NEWIPC)

setns("/proc/213/ns/ipc", CLONE_NEWIPC);
system("ipcs");

 

When you are entering a single namespace, setns(2) requires that you give it a file descriptor from that precise namespace. So if we look at the example above, we are entering the IPC namespace (by using CLONE_NEWIPC) and we are supplying the path to the ipc file descriptor (/proc/$PID/ns/ipc). For example, if we want to enter the network namespace, we would use CLONE_NEWNET and /proc/$PID/ns/net.

Additional

For any kind of container manipulation (creation or entering) you will need root (CAP_SYS_ADMIN) privileges.

In addition to namespaces, the Linux Kernel also offer resource limits/isolation in the form of Control Groups. Currently Perl lacks module for managing them, but I'm going to present my proposal for such a module at FOSDEM 2017.

Gravatar Image This article contributed by: Marian HackMan Marinov <mm@yuhu.biz>