public wiki contents of
git clone git://
Log | Files | Refs (20388B)

      1 systemd is the best example of Suck.
      2 ====================================
      3 There is a menace which is spreading like a disease throughout the Linux
      4 world, it is called [systemd](
      6 Short Historical Summary
      7 ------------------------
      8 systemd is a replacement for the standard init command, which normally runs
      9 as process id 1 on initialisation of a UNIX bootup. There has been a movement,
     10 especially around the [Red Hat]( developers to
     11 copy [Microsoft Windows]( and all of its features.
     12 Now this interpretation of how a userspace should look like is implemented and
     13 was introduced with big criticism and change in the Open Source world into
     14 many distributions. The [debacle in Debian](
     15 is the best example in how to not introduce such a changing technology into a
     16 distribution.
     18 What PID 1 Should Do
     19 ------------------------
     20 When your system boots up the kernel is executing a given binary in its known
     21 namespace. To
     22 see what are the only tasks the application running as pid 1 has to do, see
     23 [sinit](// Just wait for child process to reap
     24 and run some other init scripts.
     26 The systemd Chronicles
     27 ----------------------
     28 Now follows a collection of links related to all the features systemd tries to
     29 enforce upon you as a Linux user, because »they know better«.
     31 Please add all the links you can find! Generations after us have to know how
     32 to not do software and system development! See the
     33 [wiki instructions](// in how to contribute to this
     34 page.
     36 The Never Ending Part
     37 ---------------------
     38 * Your link here.
     40 * [The SystemD Platform: an Unbiased Critique](
     42 Part 4
     43 ------
     44 * [systemd-logind deletes your message queues](
     45 * [systemd breaks the mouse(again)](
     47 Part 3
     48 ------
     49 [NEWS revision 76153ad45f09b6ae45464f2e03d3afefbb4b2afe](
     51 * [systemd does {,U}EFI bootload](
     52   * Should systemd's PID be changed from 1 to a negative, or imaginary,
     53     number? It now exists before the kernel itself, during a bootup. 
     54     See also [systemd-boot](
     55 * [systemd replaces sudo and su](
     56   * Please note the command name, machinectl and 
     57     its [features at the manpage]( 
     58     In exchange for a program which contains sudo, su and kill 
     59     (and does some functions which historically ssh/telnet did), bare metal
     60     users have a tons of bloat and a lot of things to disable, if even 
     61     possible, useful only to people which deal with virtual machines.
     62 * [systemd-journald can do log-rotate](
     63   * Being journal files binaries written with easily corruptable
     64     transactions, does this feature make the log unreadable at times?
     65 * [Transient units](
     66   * Temporary services, because we love to reinvent procps, forking, 
     67     nohup and lsof.
     68 * [systemd does socat/netcat](
     69 * [systemd-logind does sighup and nohup](
     70   * Logout is equivalent to shutting off the machine, so you will NOT 
     71     have any running program after logout, unless you inform your init system.
     72 * [systemd-nspawn can patch at will any kind of file in a container](
     73   * Paired with transient units and user escalation performable remotely,
     74     this can mean that if you house VPS instances somewhere, your 
     75     hosting provider has means and tools to spy, modify, delete any 
     76     kind of content you store there. Encrypt everything, read your TOS.
     77 * [systemd does UNIX nice](
     79 from 2016-08-06
     81 Part 2
     82 ------
     83 * [systemd locks down /etc and makes it read-only](
     84 * [systemd now does your DNS](
     85 * [systemd hates when you adapt your system (graphics on other than vt1)](
     87 Part 1
     88 ------
     89 (If you are adding more parts, link to the right revision and not just the
     90 plain file.)
     92 [NEWS revision 3f80af378331748e1373d4c189a8e20cddb735c9](
     94 * [logind should wait](
     95   * systemd was introduced to decrease the boot up time. Now that they
     96     do not understand all implications and dependencies, let us add some
     97     artifical time we found out might work for the developers
     98     laptops. More on this small world hypothesis of the systemd
     99     developers below.
    100 * [screen brightness](
    101   * Screen brightness is something that should crash your boot up when
    102     it is not working.
    103 * [hostnamed](
    104   * There really should be a process running which exposes the content
    105     of a file. Complexity is without cost.
    106 * [seqnum removed](
    107   * The sequential ordering of requests was one reason why udevd was
    108     introduced. Now remove it, because the developer laptops do not have
    109     a problem anymore.
    110 * [floppy group removed](
    111   * Because we know what is right to know about groups. This is just one
    112     example of the mass of group name dependencies systemd is adding.
    113     See sinit for how to not need such dependencies.
    114 * [sysv removed](
    115   * We have won. Now remove all remains of our defeated enemy as
    116     fast as we can. As said in the beginning of the systemd crusade
    117     against the UNIX infidels: »You can patch it out.« It is no more
    118     there.
    119 * [abnormal processes](
    120   * Now systemd is getting deep into philosophy. What is »abnormal«?
    121     Well, let us just define it. There is no technical merit to accept
    122     this.
    123 * [systemd-resolved](
    124   * Every configuration file needs its own process and service.
    125   * Symlinks are a good way to solve all world problems.
    126 * [new is better](
    127   * The systemd development process is flawed by always assuming »new is
    128     best«.
    129   * Network configuration should be in my init process.
    130 * [remote pid 1](
    131   * »Everything will end up having a remote API.« I wonder when systemd
    132     will understand MIME and e-mail.
    133 * [init does man](
    134   * My init process is too big, it needs its own file hierarchy and an
    135     abstraction layer to find paths.
    136 * [factory reset](
    137   * Welcome to the Windows OEM world: Factory reset for Linux! Of course
    138     it is in your init process.
    139 * [system runs runs](
    140   * Exactly. The predisposition of being able to call such a complex
    141     command does not imply the running system. Let's check it again.
    142 * [clean up directories](
    143   * There is another monster in systemd, it handles tmp files.
    144     There are just some cases before it was introduced to have to clean up
    145     a directory in the file tree. Now there are hundreds. And easily
    146     another case can be added! Of course your init process does that.
    147 * [firstboot](
    148   * »Interactive queries« pulls in many dependencies. Let us have it in
    149     every installation out there on by default. Of course in pid 1.
    150 * [journald](
    151   * This is a bit longer. Because the systemd developers were not able
    152     to contribute to any syslog daemon they had to write their own with
    153     some binary format so the principle of being able to read your log
    154     files after a critical crash was violated. To be nice invaders the old
    155     systems were allowed to order the system log through the specified
    156     mechanism. Now that they implemented our specifics, turn off the
    157     neutral syslog delivery. You will see this pattern of »now that we
    158     conquered your culture, obey« more often in systemd.
    159 * [systemd-terminal](
    160   * Why does the kernel have tty handling? So in serious situations you
    161     will be able to debug it over the last standing PIN on your
    162     motherboard. Let us remove this, run it in pid 1.
    163 * [networkd is your oppressor](
    164   * Premature optimisation of IP configurations always leads to misery.
    165 * [We do not understand broadcast](
    166   * With the growth of systemd in complexity and the new depending
    167     software the implications of the added hacks are increasing.
    168 * [Timezone hack](
    169   * systemd is too complex for such a simple transaction with the
    170     kernel. Do not inform the kernel and add another assumption which is
    171     only documented in the changelog.
    172 * [the web is a session]( 
    173   * For the future generations: In 2015 nearly everyone was trying to
    174     connect all you do to the web. In the UNIX model it is just a client
    175     which should be simple and integrate into the system. systemd is
    176     doing it the other way around and does integrate itself into the
    177     web.
    178   * The discussion about why my pid 1 is handling sessions is discussed
    179     in other points.
    180 * [pid 1 does DNS](
    181   * This change adds another open door to your pid 1, adds caches, new
    182     APIs and of course it will never fail and break systemd on your
    183     initial boot.
    184   * They are planning (2015-07) to add multicast DNS to your pid 1.
    185     Multicast DNS implies an enforced timeout to know if someone answered.
    186     systemd was about speed, you know. Look at avahi in how to ruin this
    187     concept for Linux. It has the same developer flaws as systemd: Too
    188     many forced dependencies, you need to implement its API and when it is
    189     taking over you are doomed.
    190 * [policykit](
    191   * There is a bus inside your pid 1 and if it crashes you are flawed.
    192     Exactly, there are a thousand more cases of errors that could occur
    193     and make your system unbootable. Instead of using a separation of
    194     functions, add everything to a big bus.
    195   * Of course when you are using a misdesign like dbus you need to add
    196     interactions over the bus to add features you forgot in the initial
    197     design. Now let us have our pid 1 have to query for the permission
    198     to boot.
    199   * »Interactive authentication« will produce you a graphical input form
    200     on your serial line.
    201 * [Calendar](
    202   * As you see, your pid 1 should handle your calendars and cron jobs
    203     too.
    204 * [utmp should go](
    205   * We have taken over your culture, now die! Another flaw in the
    206     systemd small world theory: When something is getting optional it
    207     will be removed.
    208 * [password agent](
    209   * »Interactive authentication«
    210 * [udev timeout](
    211   * Instead of patching the kernel to add a simple solution, add a hack.
    212     Only the systemd developers tell you when it is allowed to wait or
    213     sleep in userspace. The rest obey our orders!
    214 * [systemd-pm](
    215   * Power management is required on boot up.
    216 * [user systemd units](
    217   * What can go wrong when you are adding more paths that are read,
    218     parsed and executed?
    219 * [hack the reload](
    220   * First systemd was adding »better features« like socket activation to
    221     make developers use their mechanism for daemons. They hit the
    222     proprietary wall of disgust with this changelog entry. systemd is
    223     too big and you will lose your face if you change the misdesign. Now
    224     add another hack because we can do it. Big empires fall too and
    225     sadly have too many casualties when they are falling. :(
    226 * [X11 in systemd](
    227   * Of course graphics were missing in pid 1.
    228 * [complexity is purity](
    229   * You will of course need PPPoE when you do parallel bootup. Every
    230     1000 lines of code add one critical bug you never fix.
    231 * [gateway hostname](
    232   * We rule the world so we are above IETF and IANA. Now add our own
    233     hostnames that of course won't add another assumption.
    234 * [no editor in systemd](
    235   * This one is a setback. Why is there no default editor in systemd in
    236     case of factory reset?
    237 * [8x ctrl + alt + del](
    238   * In systemd you press eight times Ctrl+Alt+Del to trigger reboot.
    239 * [privacy policy](
    240   * For the next generations: In 2015 privacy was a big issue because of
    241     the mentioned hard-wiring between the web and software.
    242     As you can see, every commit which adds some preparation for a 
    243     feature adds another intepretation of what will be a major
    244     assumption in a next release. If you
    245     handle privacy you will have some features depending on that user
    246     decision and of course the factory reset default value.
    247   * Why didn't they use XML for /etc/os-release?
    248 * [fds cache](
    249   * We have talked about misdesign, too-big-to-fail and world
    250     domination. This is the next example of a hack that is prone to
    251     fail.
    252 * [umount -rf](
    253   * This is umount for dummies. Just do one thing – right.
    254 * [libudev will be orphaned](
    255   * With the advent of udevd there was a compatibility to its complexity
    256     called libudev. X11 uses it to query the changing devices. And of
    257     course make it a non-independent API in systemd. Why? You can guess
    258     why: Defeating the infidels.
    259 * [fsck indirections](
    260   * When there was syslog fsck did output errors to your display as
    261     easy as possible. Now add a hack to have this possible again.
    262 * [systemd-importd](
    263   * This is pure evil. Your pid 1 is now able to import complete system
    264     images over the network and show them to you as your running system.
    265     There is nothing that can go wrong.
    266 * [CGI for systemd](
    267   * The web thing has been discussed before.
    268 * [kdbus](
    269   * As of 2015-07-31 kdbus is not in the mainline Linux kernel. systemd
    270     made kdbus non-optional in its release. The kernel maintainers are
    271     still debating the kdbus ABI or possible alternatives, but if
    272     systemd depends on the current state of kdbus the kernel maintainers
    273     are faced with the hard decision to either break Fedora userspace or
    274     accept the current kdbus proposal into the kernel with its security
    275     and maintainability issues. This is the best example how
    276     systemd is forcing you into decisions. Of course if you are  a mindless
    277     bureaucrat it helps you to keep your job.
    278 * [readahead removed](
    279   * The first thing swallowed in on Fedora was readahead. Now that (of
    280     course!) everyone is using an SSD (at least the developers of
    281     systemd do that) it can be removed. Why was it there? Is it possible
    282     to make it a separate tool again?  There's no time for that, we are implementing
    283     new features.
    285 from 2015-07-31
    287 Practical systemd
    288 -----------------
    290 Here is what happens on a stock Arch Linux system, powered by systemd,
    291 when a non-root user tries to restart the system:
    293     $ reboot
    294     Failed to set wall message, ignoring: The name org.freedesktop.PolicyKit1 was not provided by any .service files
    295     Failed to reboot system via logind: The name org.freedesktop.PolicyKit1 was not provided by any .service files
    296     Failed to talk to init daemon.
    298 In contrast, here is the equivalent error message on a system powered by
    299 runit:
    301     $ reboot
    302     init: fatal: unable to create /etc/runit/stopit: access denied
    304 And on the oldest and best, Slackware:
    306     $ reboot
    307     reboot: must be superuser.
    309 systemd is driving "just google the problem" attitude, because how the hell are
    310 you expected to troubleshoot this kind of error otherwise?
    312 Further Reading
    313 ---------------
    314 * [Without systemd](
    315 * [Arguments Against systemd](