While systemd has been around for a few years, it is gaining more and more attention as more Linux distributions switch to it. Lots of developers are extremely opposed to systemd as it goes against the Unix philosophy. Most Linux users, however, have no idea what it is or why it is a good or bad thing. Here is a basic overview of init systems and systemd.
What is it?
When you boot a Linux machine, once the kernel has finished getting itself ready a new process is launched which is termed the init process. This process is a daemon and its responsibilities are to load and manage the other daemons on the system. By this, I mean that once your kernel has booted, the init process gets your computer into the much more usable, familiar state that you are used to. Essentially the init process establishes and operates the entire user space.
Some of the more typical tasks which fall to the init process are starting any user services (That game server you like to run on start up? Probably is launched through the init process.), loading a desktop manager such as GNOME or KDE, and even checking and mounting file systems.
The init process is generally started with what is known as a run level. This level defines what should and should not be loaded (and unloaded). Each operating system uses different run levels for different things but they tend to have some things in common. Generally there are 7 of these run levels, numbered 0-6 and some are fairly common. Some of the more common ones are referred to by name since they don’t always have the same run level. Examples of this are “single user mode”, “multi user mode without network services”, and “system reboot”. While there is no guarantee, of consistency between systems, run level 0 is generally a halt which shuts down the system and level 6 is generally reboot.
When the init system is loaded with a run level, it executes all the processes for that run level and that run level only. For instance if it starts with run level 3 (which is a fairly typical selection for normal use), it will load all the regular user processes including your desktop manager and networking. The scripts and instructions for all other run levels are ignored. When you decide to shutdown or reboot your machine you must change the run level. This change in run level is achieved through a separate binary. Upon the change, the init system will execute the scripts and instructions for that run level. These run levels are generally only seen in System-V style init methods (this will be explained later).
The idea of this init process was originally found in Unix, with the most well known model coming from System-V. The init method here really set the foundation of having one daemon manage and load all of the others. This method involved executing a series of scripts in turn when initially launched or when the run level changed. This method is used today in many Linux distributions which use System-V style init systems. So lets look at the various current options for Linux.
Often just named init, also known as sysvinit, this was one of the most popular
init systems found on Linux. It’s main configuration file is found in
/etc/inittab which is also where the run level is specified for launch. This
init method follows the principle outlined above, where a scripts are executed
in series until the system is ready. These scripts tend be found in directories
/etc/rc.******. Once all processes are loaded it waits for one of three
events: The parent process dies, a power failure signal or a request via
/sbin/telinit to change the run level (for example, on shutdown).
Upstart is a replacement for sysvinit. It is fairly well known due to its use in Ubuntu. Created by former Canonical engineer Scott James Remnant, it was designed to try and remove as much of the serial architecture of sysvinit as possible. In sysvinit, if something is loaded from HDD then the entire system had to wait for this to finish. IO takes a huge amount of time relatively speaking, and so the process was rather wasteful. Upstart is asynchronous, which allows it to load multiple things at the same time, reducing the time spent waiting on IO. Further more, it is also able to respond to events which allows it to gracefully handle actions such as inserting a USB drive. One of the greatest strengths of upstart is that it is also backwards compatible with sysvinit scripts.
Just like upstart, systemd is a replacement for sysvinit, but includes a lot more software with it such as systemd, journald and logind. It was also designed to be more efficient than sysvinit, but works slightly differently. The idea is that you can clearly express the dependencies on different daemons and that way systemd can solve the problem and load the various daemons concurrently as efficiently as possible. Instead of the standard scripts used for sysvinit and upstart, systemd configuration files are declarative plain text files.
So basically, what we have seen is that each of these pieces of software successfully does its job of initialising and monitoring daemons. The difference is that upstart and systemd parallelise this task and so they load everything much faster than sysvinit.
Why should I care?
So this is where things get more complicated. Basically, the Unix philosophy is that each piece of software should do one thing and do it well. The idea means that we create a bunch of extremely powerful, modular components which can be stacked together to solve problems. This is most apparent with simple tools used on the command line where you pipe output from one program into another. This applies throughout Linux.
Systemd goes against this, according to some people, as it tries to do too much. Systemd refers not just to an init process, but an entire software bundle along with the init process. These other applications are used for various things from new login managers to logging managers. However, while it may be bundled with lots of other functionality, this functionality is all separated into different binaries which interact with each other, thus keeping the modularity.
So does it really matter? Well that depends. While systemd has achieved its goal of being a better init system, it has continued to grow and take over functions which should be left to other software where that is the software’s only responsibility (remember that Unix philosophy). This presents a some problems.
The first problem is that it means that systemd is a single point of failure. The entire user space is dependent on it, and without it, the machine is useless to most people. If there is a bug in the code, its required position due to many dependencies, will make it a prime target for attackers. The Linux kernel is already a huge target so why give attackers another?
The second problem is the fact that systemd is very tightly coupled with the Linux kernel and therefore it is very easy to introduce incompatibilities. Software which runs in user space should have minimal dependencies on the kernel. With these dependencies, it makes it very difficult to ensure that systems can remain stable as a software update is much more likely to require a kernel update.
The third problem stems from the tight coupling with the Linux kernel. It means that systemd itself can’t be ported to other platforms. You might think that this isn’t a huge issue, and normally you would be right. However, lots of software is starting to list systemd as a dependency. If it has systemd as a dependency then it means that software also can only run on Linux. As an example of how serious this is, both KDE and GNOME have a dependency on systemd. For the time being, this is only for “non-basic functionality”, but that means things like power management, which one might argue, is pretty important. It seems to only be a matter of time before the dependencies grow.
The list goes on. Personally, the biggest warning sign to me is how they handle logging. No longer are text files used for logging, instead a binary format is used. Furthermore, this format does not have ACID compliant transactions, which means that your logs can easily become corrupted. When this happens, the software simple ignores it, rotates the file out, and then continues on as though nothing happened. The best part of this one is that the developers have stated that it is unlikely that this will be fixed in the near future.1 If such a critical component can’t get logging right, what chance do they have for everything else?
I set out when writing this post to try and give a balanced view on systemd and I have to admit that I have completely failed so far. Almost everything I’ve said so far has been negative. So let’s look at some of the good parts of systemd:
- It has a much more efficient init process than sysvinit.
- It tracks processes using cgroups rather than PIDs which helps with process management.
- In theory, it is easier to launch a daemon as you no longer have to write a bash script.
- The support for D-Bus and sockets makes inter-daemon communication easier.
Basically, you have to make your own decision, but hopefully you can see that this is an extremely important issue and you most definitely should care.