systemd For Learner Drivers

A graphic to help out: hpr1672.svg

This is a subject that attracts controversy, but I am not today going to be controversial, I hope. Many Linux systems are moving away from SysV Init and adopting systemd instead; both Linuxes that I use, Fedora and Mint have adopted systemd, and I understand that Debian has now forked to allow both sides of the argument to have their way. I am not going to get into the debate here. My personal stance is that I see both sides of the argument and I will continue to perch on top of the fence until systemd either proves itself or fails to do so.

In this HPR I am going to try to fill a gap that I have seen in the systemd discussion; that is - how to operate it. I am not an expert on systemd, I have just tried to work it, and in doing so I have fished around in my file system and in the documentation. If you want to know what I found, then keep on listening. By way of opening I will remind myself, and you also, what systemd is replacing.

SysV initd works with runlevels, the most common being

• 5 for graphical multiuser networked
• 3 for cli multiuser networked
• 1 for single user
• 6 for reboot
• 0 for halt

In moving to a runlevel, unwanted services are shut down and wanted services are started up. For most users on most systems the most appropriate default runlevel is 5 giving multiuser, GUI & networking. Services can be started and stopped on demand by inetd.

systemd works differently. It has target units. For most users on most systems the most appropriate default target is the graphical.target, which does a similar thing to runlevel 5 . Units are configured by unit configuration files. These files may start other units and stop other units. They can impose sequence and dependancies. There is a lot of cascading going on, with unit launching unit launching unit. Units also can be started and stopped on demand by systemd.

Units

The term Unit refers to a resource that systemd is taking under its control. There are 12 different types of Unit.

systemd.service

that starts/stops daemons

systemd.socket

activates network connections

systemd.device

activates kernel devices

systemd.mount

controls mount points

systemd.automount

provides on-demand mounting of file systems

systemd.swap

does for swap what systemd.mount does for filesystems

systemd.scope

starts/stops external processes

systemd.target

groups of services akin to init level 3, init level 5

systemd.snapshot

saves/restores the momentary state of other units

systemd.timer

triggers units based on date/time

systemd.path

trigger units based on changes in file system objects

organises units in a hierarchical tree of cgroups, for resource management purposes

Units files called by systemd live in /etc/systemd/system. But these are symbolic links to the real ones stored in /usr/lib/systemd/system

There is a parallel /etc/systemd/user structure which does not seem to do anything on my computers, so I work for now like its not there.

There is also a /run/systemd/system structure which appears to contain runtime configuration files with names like session-xxxx.scope. These are the unit type for external processes.

Table 1. Directory structure for systemd

Path	Description
/etc/systemd/system	Local configuration
/etc/systemd/user	User configuration
/run/systemd/system	Runtime units
/usr/lib/systemd/system	Units of installed packages

Directives

The next thing we need is Directives.

The unit configuration files contain directives to start/stop a unit, and directives that cascade to other unit configuration files that start/stop dependant units. Directives may impose conditions on whether or when to call a unit. There are a whole bunch of different directives listed in man systemd.unit. These are a few.

• Requires= list of units to start. If any required units fail then abort this one
• Conflicts= list of units to stop
• After= the order in which units will start
• Before= the order in which units will start
• Wants= list of units to start. If any fail just continue anyway

As well wanted units listed by the WANTS directive, there may also be a 'wants' directory below the unit directory. So the unit conf file /etc/systemd/system/default.target will cause two further unit conf files to be read in from the /etc/systemd/system/default.target.wants/ directory.

Each required unit and wanted unit from the directives, as well as those in the wants directory are added to a job queue. If directives cascade to other unit files containing more directives then all of these dependences are also added to the job queue. A directive may start or stop another unit, or that change the detail of a job already in the queue. All directives ultimately cascade down to starting or stopping one of the base units in /usr/lib/systemd/system.

To get a feel for how this all pans out in practice I will walk us through the cascade of unit files from bootup.

From Bootup

First, the default.taget is activated, which on my system is just a link to graphical.target

graphical.target

[Unit]
Description=Graphical Interface
Documentation=man:systemd.special(7)
Requires=multi-user.target
After=multi-user.target
Conflicts=rescue.target
Wants=display-manager.service
AllowIsolate=yes

Cascades to

• start multi-user.target
• start display-manager.service
• stop rescue.target

Also we have a wants directory /etc/systemd/system/graphical.target.wants/ that

• starts accounts-daemon.service (for logging)
• starts rtkit-daemon.service (for realtime scheduling)

multi-user.target

graphical target cascaded to multi-user.target.

[Unit]
Description=Multi-User System
Documentation=man:systemd.special(7)
Requires=basic.target
Conflicts=rescue.service rescue.target
After=basic.target rescue.service rescue.target
AllowIsolate=yes

Cascades to

• start basic.target
• stop rescue.service
• stop rescue.target (again)

Also we have a wants directory /etc/systemd/system/multi-user.target.wants/ that

- abrt-ccpp.service
- abrtd.service
- abrt-oops.service
- abrt-vmcore.service
- abrt-xorg.service
- atd.service
- auditd.service
- avahi-daemon.service
- chronyd.service
- crond.service
- cups.path
- irqbalance.service
- libvirtd.service
- mcelog.service
- mdmonitor.service
- NetworkManager.service
- nfs.target
- remote-fs.target
- rngd.service
- rpcbind.service
- rsyslog.service
- smartd.service
- vmtoolsd.service

display-manager.service

graphical.target also cascaded to display-manager.service which is not present on F20 so I guess we don't need it.

basic.target

So multiuser.target cascaded to basic.target, which itself cascades to

- sysinit.target
- sockets.target
- timers.target
- paths.target
- slices.target
- firewalld.service

sysinit.target

basic.target cascaded to sysinit.target which itself cascades to

- local-fs.target 
- swap.target
- dmraid-activation.service
- iscsi.service
- lvm2-monitor.service
- multipathd.service ( which looks like all the file system daemons)

sockets.target

basic.target also cascaded to sockets.target which itself cascades to

- avahi-daemon.socket
- cups.socket
- dm-event.socket
- iscsid.socket
- iscsiuio.socket
- lvm2-lvmetad.socket
- rpcbind.socket

End point

Now we start reaching the end-points of this trail at

- systemd.sockets
- systemd.timer
- systemd.path
- systemd.slice
- systemd-fstab-generator

By the time all of that has finished, if I type the command

# systemctl list-units --type service

I see that 58 services are listed as running

Running and Configuring Services

If we are going to work with systemd we will have to give it instructions. In systemd parlance

• active = running, currently in use
• loaded = enabled, available for use

These terms crop up in the output from commands

Many instructions are given to systemd by the systemctl command.

Now to compare line up some common SysV init tasks with their systemd equivalent

Table 2. SysV init commands and their systemd equivalents

command	SysV Init	systemd
Check status	# service bluetooth status	# systemctl status bluetooth
Start	# service bluetooth start	# systemctl start bluetooth
Stop	# service bluetooth stop	# systemctl stop bluetooth
Enable	# chkconfig --level 35 ntpd on	# systemctl enable ntpd
Disable	# chkconfig --level 35 ntpd off	# systemctl disable ntpd

Journalctl Logging

Much has been said about the desirability or otherwise of binary logs, but systemd gives us these so we had better know what to do with them.

Journal instructions are given to systemd by the journalctl command

To view all log entries in one go. This is verbose, mine came out at ~9000 lines

# journalctl

To view from a specific date

# journalctl --since="2014-05-07"

To view kernel logs

# journalctl -k

To follow a log in realtime ... and then to close

# journalctl -f
...
# ctl-c

To view log entries associated with a given PID

# journalctl _PID=1

To view log entries associated with a given service

# journatlctl -u bluetooth

Interrogating the system

More systemd information

Get/Set system information. Works like uname, but is more verbose

# hostnamectl

Get/Set timezone & timedate info

# timedatectl

Table 3. SysV init information and their systemd equivalents

SysV Init Info	SysV Init command	systemd info	systemd command
What services are available for init.d to manage	# ls /etc/init.d	What service units are available for systemd to run	# systemctl list-units --type service --all
What services are configured to be run by init.d for each run level	# chkconfig --list	What service units are currently active	# systemctl list-units --type service

References

• man pages for systemd, systemd.unit, systemctl, journalctl.
  Plus a bunch of other man pages that are SEE ALSO in these man pages.
• systemd homepage
• A useful systemd blog posting, by Adam Dean of Asterisk Telephones
• A useful youtube vid by George Magklaras, UNiversity of Oslo, Norway