Sl6ckware

Using s6 + s6-rc + s6-linux-init as the init system on Slackware.

Note

• This repo is intended to be used on Slackware-current ("current" branch) and Slackware 15.0 ("15.0" branch).

• Before using this repo, read the s6, s6-rc, and s6-linux-init home page first to have a better understanding of what they are and what features or advantages that they provide compared to other init systems. Or for a quick introduction in video format, see this video.

Requirements

• skalibs
• execline
• s6
• s6-rc
• s6-linux-init

All of them are available on SBo.

Setup

Compiling a new s6-rc service database

1. Clone this repo.

2. Create a copy of the source directory from this repo and put it somewhere else in your system (e.g., /etc/s6/rc/source).

3. Edit these two files (all paths below are relative to the source directory that you just copied):

   • bundles/S/contents: this file contains a list of services that are needed to initialize the system, just like the rc.S script. In this file you can (for example) uncomment font service to load a custom screen font for the virtual console (see rc.font/font/up for instructions on how to load your desired custom screen font).

   • bundles/M/contents: this file contains a list of services that will be started when entering one of the multiuser runlevels (runlevel 2, 3, 4, and 5 by default), just like the rc.M script. In this file you can (for example) uncomment cpufreq and bluetooth service to set the default CPU governor and enable Bluetooth support, respectively, at boot time.

4. Feel free to edit all other files inside the source directory beside those two files.

5. Use s6-rc-compile (more details are available in /usr/doc/s6-rc-<version>/doc/s6-rc-compile.html) to compile a new s6-rc service database from your source directory that will be used by s6-rc at runtime to manage services:

   s6-rc-compile /etc/s6/rc/compiled /etc/s6/rc/source/*

   • /etc/s6/rc/compiled: tell s6-rc-compile to put the compiled database data in the directory /etc/s6/rc/compiled (it must not exist beforehand). Of course, you can change /etc/s6/rc/compiled to other places that you want.

   • /etc/s6/rc/source/*: tell s6-rc-compile to search for service definition source directories in all subdirectories of /etc/s6/rc/source (notice the trailing asterisk). Of course, you should change /etc/s6/rc/source to the path of your own copy of source directory.

Setting up s6-linux-init

1. Use s6-linux-init-maker (more details are available in /usr/doc/s6-linux-init-<version>/doc/s6-linux-init-maker.html) to create all the necessary files that the system needs to be able to properly boot and bring up a full s6 infrastructure. Below is the s6-linux-init-maker invocation that I use for my system (feel free to change them and add any other options that suit your needs):

   s6-linux-init-maker \
       -f ~/Documents/sl6ckware/skel \
       -c /etc/s6/init/current \
       -u adm \
       -G "agetty 38400 tty12 linux" \
       -1 \
       -p "/bin:/sbin:/usr/bin:/usr/sbin" \
       -t 2 \
       -D 4 \
       /etc/s6/init/current

   • -f ~/Documents/sl6ckware/skel: tell s6-linux-init to use the skeleton scripts provided by this repo (i.e., skel/rc.init, skel/rc.shutdown, skel/rc.shutdown.final, and skel/runlevel) instead of the default s6-linux-init skeleton scripts (don't forget to change ~/Documents/sl6ckware to the path where you clone this repo into).

   • -c /etc/s6/init/current: tell s6-linux-init to look inside /etc/s6/init/current at system startup for all the necessary files it needs in order to boot the system properly. We will reference this directory as basedir.

   • -u adm: the catch-all logger will run as the adm user. The catch-all logger is a service that is started early at boot time to log all system startup and shutdown time messages (except for kernel messages) to the logging directory /run/uncaught-logs (or tmpfsdir/uncaught-logs where tmpfsdir is the argument that was given to the --tmpfsdir= configure option at s6-linux-init build time). This is made possible by using s6-log which takes inputs from stdin and processes them based on the logging script given to it (like adding timestamps, deselect certain lines, etc.). See /usr/doc/s6-<version>/doc/s6-log.html for more details on s6-log.

   • -G "agetty 38400 tty12 linux": start an early getty service by executing the given command line (agetty 38400 tty12 linux). The command line agetty 38400 tty12 linux will start an agetty instance that connects to port /dev/tty12 with baud rate 38400 and TERM=linux.

   • -1: make it so that all the messages that are sent to the catch-all logger are also copied to /dev/console (without timestamps). This will make those messages also appear on the computer screen just like Slackware's default init system but better, since those messages can still be accessed later after the system has finished booting.

   • -p "/bin:/sbin:/usr/bin:/usr/sbin": set the initial value of the PATH environment variable to /bin:/sbin:/usr/bin:/usr/sbin. Almost all s6-rc service definitions in this repo do not use absolute path when invoking a program so at least these directories should be added to the initial PATH value.

   • -t 2: use the ISO 8601 format for the catch-all logger's timestamp format.

   • -D 4: set the default runlevel to runlevel 4.

   • /etc/s6/init/current: tell s6-linux-init-maker to put the generated files inside /etc/s6/init/current (this directory must not exist beforehand). You can change this argument to other places that you want (e.g., /tmp/init-current) but don't forget to move/copy the generated files back to the declared basedir (/etc/s6/init/current in my case). Use cp -a when copying the generated files since there are FIFOs, files with precise UID/GID permissions, and files with non-standard access rights, so be sure to copy it verbatim.

2. Edit bin/init inside your basedir and change s6-linux-init to /sbin/s6-linux-init since the PATH environment variable is undefined at early boot time. This is not necessary if you add /sbin to the default executable search path when compiling skalibs (see its SBo README file for more details on how to do that).

3. Lastly, make a backup copy of /sbin/halt, /sbin/init, /sbin/poweroff, /sbin/reboot, /sbin/shutdown, and /sbin/telinit. Then, copy all scripts inside your basedir's bin directory to /sbin (or alternatively, you can just create symbolic links inside /sbin that point to each script).

Tips

• The default init system and the s6-based init system can actually coexist by using different names when copying/linking the aforementioned scripts (e.g., instead of /sbin/init, use /sbin/init.s6). Then, when you want to use the s6-based init system, pass init=/sbin/init.s6 (following the example before) to the kernel as a command line argument. This is mainly useful for testing the s6-based init system before using it as the main init system.

• Use a symbolic link to manage the default service database that s6-rc uses at system startup. For example, each time you compile a new database, you run:

  s6-rc-compile "compiled-$(date +%y%m%d-%R:%S)" source/*

  Then, if you want to make it the default service database, make a symbolic link named compiled that points to it. With this approach, you don't have to modify your basedir's scripts/rc.init script every time you change your default service database.

• If you applied the previous tip, you probably want to modify your basedir's scripts/rc.init script and add the -d option to the s6-rc-init invocation. The -d option will tell s6-rc-init to dereference the argument that is given to the -c option (/etc/s6/rc/compiled in my example above) and use that instead. The symbolic link will then not be used by s6-rc and thus can be updated freely without affecting the currently running system. Without the -d option, you have to run s6-rc-update <another_compiled_database_absolute_path> first before updating the symbolic link.

Quick user guide

• Start a service:

  s6-rc -u change <service_name>

• Stop a service:

  s6-rc -d change <service_name>

• Restart a longrun service:

  s6-svc -r /run/service/<service_name>

• List all active services:

  s6-rc -a list

• List all nonactive services:

  s6-rc -da list

• Switch to another compiled service database:

  s6-rc-update <another_compiled_database_absolute_path>

Per-user supervision tree

To have a per-user supervision tree running under the system's supervision tree, we need to create a dedicated system service that runs an instance of s6-svscan as a normal user. Each user wanting to have a personal supervision tree will need to have this kind of service created for them. Luckily, there is the s6-usertree-maker program that we can use to easily create it.

Creating your user services

Before creating a per-user supervision tree service (using s6-usertree-maker) that will run as a system service, you need to create your own user services first and test them. If you don't need dependencies handling between your services, you don't need s6-rc. Just create a new scan directory containing the service directories that you have made and put it somewhere accessible by your user, like /home/<your_username>/.config/s6/service. Try running s6-svscan <your_scan_directory> directly on your terminal to test out the services you have made. If everything is working correctly, continue to the next section. Personally, I use s6-rc to manage all my user services and I don't create a scan directory myself. Rather, I let the per-user supervision tree service creates an empty scan directory /run/user/${UID}/s6/service at runtime, as you will see later in the next section. Some of the programs that I run under this per-user supervision tree are ssh-agent, dbus-daemon, pipewire, pipewire-media-session, and pipewire-pulse. See my s6-rc source definition directories for these programs here.

Creating the per-user supervision tree service

The s6-usertree-maker program provides many options that you can use to customize the generated per-user supervision tree service, see /usr/doc/s6-<version>/doc/s6-usertree-maker.html for more details. As an example, this is what I use on my system:

s6-usertree-maker \
    -d '/run/user/${UID}/s6/service' \
    -p '/usr/local/bin:/usr/bin:/bin' \
    -E env \
    -e XDG_CACHE_HOME \
    -e XDG_CONFIG_HOME \
    -e XDG_DATA_HOME \
    -e XDG_RUNTIME_DIR \
    -e XDG_STATE_HOME \
    -r usertree-${USER}-srv/usertree-${USER}-log/usertree-${USER} \
    -n 4 \
    -s 500000 \
    -t 2 \
    $USER /var/log/usertree-${USER} /etc/s6/rc/source/usertrees

Explanation:

• -d '/run/user/${UID}/s6/service': the per-user supervision tree will be run on the directory /run/user/${UID}/s6/service, which is subject to variable substitution (notice that I use single quote here to prevent ${UID} from being substituted by the shell).

• -p '/usr/local/bin:/usr/bin:/bin': the per-user supervision tree will be run with a PATH environment variable set to /usr/local/bin:/usr/bin:/bin, which is also subject to variable substitution (although, in this case, no variable is actually being used). You might want to add something like ${HOME}/.local/bin to that list if you have some programs in your home directory.

• -E env: the per-user supervision tree will be run with the environment variables defined in the directory env, which will be read via s6-envdir without options. Since it's not using an absolute path, this env directory will need to be located inside the generated service directory. If you have more than one user supervision trees, you might want to set it to something like /etc/user-env so all of them can share the same set of environment variables.

• -e XDG_CACHE_HOME, -e XDG_CONFIG_HOME, -e XDG_DATA_HOME, -e XDG_RUNTIME_DIR, -e XDG_STATE_HOME: perform variable substitution on XDG_CACHE_HOME, XDG_CONFIG_HOME, XDG_DATA_HOME, XDG_RUNTIME_DIR, and XDG_STATE_HOME. That is, if these environment variables contain ${USER}, ${HOME}, ${UID}, ${GID}, and ${GIDLIST}, they will be substituted with their respective values.

• -r usertree-${USER}-srv/usertree-${USER}-log/usertree-${USER}: create s6-rc source definition directories instead of a regular s6 service directory, where the per-user supervision tree service will be named usertree-<your_username>-srv, its logger will be named usertree-<your_username>-log, and a bundle containing both of them will be named usertree-<your_username>.

• -n 4, -s 500000, -t 2: use n4 s500000 T as the control directives for the s6-log instance that will log the per-user supervision tree service.

• $USER: the per-user supervision tree will be run as the user <your_username>.

• /var/log/usertree-${USER}: the per-user supervision tree logs will be saved in the directory /var/log/usertree-<your_username>.

• /etc/s6/rc/source/usertrees: put the generated s6-rc source definition directories in the directory /etc/s6/rc/source/usertrees.

After running that command, you can inspect the generated s6-rc source definition directories and modify them if necessary. As an example, on my system I have to add this small change to the generated usertree-<your_username>-srv/run script:

--- a/usertree-<your_username>-srv/run     2022-08-18 00:18:42.986512403 +0700
+++ b/usertree-<your_username>-srv/run     2022-08-18 01:35:12.639851358 +0700
@@ -28,4 +28,6 @@
 export "XDG_RUNTIME_DIR" ${"XDG_RUNTIME_DIR"}
 export "XDG_STATE_HOME" ${"XDG_STATE_HOME"}
 export PATH "/usr/local/bin:/usr/bin:/bin"
+# Creating an empty scan directory since all services will be handled by s6-rc
+if { mkdir -p "/run/user/${UID}/s6/service" }
 s6-svscan -d3 -- "/run/user/${UID}/s6/service"

Lastly, recompile your system's s6-rc service database to include these new services.

Starting the per-user supervision tree service

Probably the best way to start this service is by dynamically starting it upon the user's first log-in and stopping it upon the user's last log-out. To achieve this, I use the PAM module pam_exec that executes two scripts: pam-start-usertree when a user logs in and pam-stop-usertree when a user logs out. Here are the diffs that I have applied to /etc/pam.d/login and /etc/pam.d/sddm on my system:

• /etc/pam.d/login:

  --- a/etc/pam.d/login   2020-06-19 04:01:47.000000000 +0700
  +++ b/etc/pam.d/login   2022-08-17 16:58:49.716526301 +0700
  @@ -17,4 +17,18 @@
   session         include         postlogin
   session         required        pam_loginuid.so
   -session        optional        pam_ck_connector.so nox11
  +# Skip the pam-stop-usertree script for non-regular users
  +session         [success=ok default=1] pam_succeed_if.so quiet uid >= 1000 \
  +                                                         uid <= 60000
  +# Running the pam-stop-usertree script before $XDG_RUNTIME_DIR is deleted by
  +# pam_elogind
  +session         optional        pam_exec.so type=close_session \
  +                                            /usr/local/sbin/pam-stop-usertree
   -session        optional        pam_elogind.so
  +# Skip the pam-start-usertree script for non-regular users
  +session         [success=ok default=1] pam_succeed_if.so quiet uid >= 1000 \
  +                                                         uid <= 60000
  +# Running the pam-start-usertree script after $XDG_RUNTIME_DIR is set and
  +# created by pam_elogind
  +session         optional        pam_exec.so type=open_session \
  +                                            /usr/local/sbin/pam-start-usertree

• /etc/pam.d/sddm:

  --- a/etc/pam.d/sddm    2022-01-16 12:12:19.000000000 +0700
  +++ b/etc/pam.d/sddm    2022-08-17 16:58:49.716526301 +0700
  @@ -20,7 +20,21 @@
   session    substack     system-auth
   session    required     pam_loginuid.so
   -session   optional     pam_ck_connector.so      nox11
  +# Skip the pam-stop-usertree script for non-regular users
  +session    [success=ok default=1] pam_succeed_if.so quiet uid >= 1000 \
  +                                                    uid <= 60000
  +# Running the pam-stop-usertree script before $XDG_RUNTIME_DIR is deleted by
  +# pam_elogind
  +session    optional     pam_exec.so type=close_session \
  +                                    /usr/local/sbin/pam-stop-usertree
   -session   optional     pam_elogind.so
  +# Skip the pam-start-usertree script for non-regular users
  +session    [success=ok default=1] pam_succeed_if.so quiet uid >= 1000 \
  +                                                    uid <= 60000
  +# Running the pam-start-usertree script after $XDG_RUNTIME_DIR is set and
  +# created by pam_elogind
  +session    optional     pam_exec.so type=open_session \
  +                                    /usr/local/sbin/pam-start-usertree
   -session   optional     pam_gnome_keyring.so     auto_start
   -session   optional     pam_kwallet5.so          auto_start
   session    include      postlogin

As for the scripts themselves, here are what I use:

• pam-start-usertree:

  #!/bin/sh -e
  #
  # Start s6-based user services upon first login
  #
  
  # Setting $PATH first since it's probably undefined
  export PATH="/usr/bin:/bin:/usr/sbin:/sbin"
  
  # Defining some important constants that will be used later
  SYSTEM_SCANDIR="$(xargs -0a /proc/1/cmdline | grep -o ' /.*$' | cut -c 2-)"
  USERTREE_SVNAME="usertree-$PAM_USER"
  USERTREE_SVDIR="${SYSTEM_SCANDIR}/${USERTREE_SVNAME}-srv"
  S6RC_TIMEOUT="5000"
  USER_SCANDIR="${XDG_RUNTIME_DIR}/s6/service"
  USER_S6RC_LIVE="${XDG_RUNTIME_DIR}/s6/rc"
  USER_S6RC_BUNDLE="${XDG_SESSION_TYPE}-session" # unspecified/tty/x11/wayland/mir
  USER_S6RC_COMPILED="/home/${PAM_USER}/.config/s6/rc/compiled"
  
  # Aborting execution when not invoked by pam_exec
  if [ -z "$PAM_SERVICE" ]; then
      echo "We are not invoked by pam_exec, aborting..."
      exit 1
  fi
  
  # Exiting immediately if PID 1 is not s6-svscan
  grep -Fqwz s6-svscan /proc/1/cmdline || exit 0
  
  # Exiting immediately if the user doesn't have a supervision tree configured
  [ -d "$USERTREE_SVDIR" ] || exit 0
  
  # Checking if this is not the user's first session
  if [ $(loginctl -p Sessions --value show-user "$PAM_USER" | wc -w) -gt 1 ]; then
      # Exiting immediately if the user doesn't use s6-rc
      [ -d "$USER_S6RC_LIVE" ] || exit 0
      # Starting the appropriate user configured bundle (based on this session's
      # type) again in case the current and last sessions' type are different
      exec s6-setuidgid "$PAM_USER" \
          s6-rc -bl "$USER_S6RC_LIVE" -t "$S6RC_TIMEOUT" \
              change "$USER_S6RC_BUNDLE"
  fi
  
  # Starting the user's supervision tree
  s6-rc -bt "$S6RC_TIMEOUT" change "$USERTREE_SVNAME"
  
  # No need to continue further if the user doesn't use s6-rc
  [ -d "$USER_S6RC_COMPILED" ] || exit 0
  
  # Initializing s6-rc for the user's supervision tree
  s6-setuidgid "$PAM_USER" \
      s6-rc-init -dc "$USER_S6RC_COMPILED" -l "$USER_S6RC_LIVE" \
          -t "$S6RC_TIMEOUT" "$USER_SCANDIR"
  
  # Starting the appropriate user configured bundle (based on this session's type)
  exec s6-setuidgid "$PAM_USER" \
      s6-rc -l "$USER_S6RC_LIVE" -t "$S6RC_TIMEOUT" change "$USER_S6RC_BUNDLE"

• pam-stop-usertree:

  #!/bin/sh
  #
  # Stop s6-based user services upon last logout
  #
  
  # Setting $PATH first since it's probably undefined
  export PATH="/usr/bin:/bin:/usr/sbin:/sbin"
  
  # Defining some important constants that will be used later
  SYSTEM_SCANDIR="$(xargs -0a /proc/1/cmdline | grep -o ' /.*$' | cut -c 2-)"
  USERTREE_SVNAME="usertree-$PAM_USER"
  USERTREE_SVDIR="${SYSTEM_SCANDIR}/${USERTREE_SVNAME}-srv"
  S6RC_TIMEOUT="5000"
  USER_S6RC_LIVE="${XDG_RUNTIME_DIR}/s6/rc"
  
  # Aborting execution when not invoked by pam_exec
  if [ -z "$PAM_SERVICE" ]; then
      echo "We are not invoked by pam_exec, aborting..."
      exit 1
  fi
  
  # Exiting immediately if PID 1 is not s6-svscan
  grep -Fqwz s6-svscan /proc/1/cmdline || exit 0
  
  # Exiting immediately if the user doesn't have a supervision tree configured
  [ -d "$USERTREE_SVDIR" ] || exit 0
  
  # Exiting immediately if this is not the user's last session
  [ $(loginctl -p Sessions --value show-user "$PAM_USER" | wc -w) -gt 1 ] && \
      exit 0
  
  # Using s6-rc to bring down all the user's services (if applicable)
  [ -d "$USER_S6RC_LIVE" ] && s6-setuidgid "$PAM_USER" \
      s6-rc -Dal "$USER_S6RC_LIVE" -t "$S6RC_TIMEOUT" change
  
  # Stopping the user's supervision tree
  exec s6-rc -bdt "$S6RC_TIMEOUT" change "$USERTREE_SVNAME"

These scripts might be overkill for most people, since they also handle more than just starting and stopping the per-user supervision tree service. The pam-start-usertree script handles initializing s6-rc for the user (if it's used) and starting different user s6-rc bundles depending on the session type (logging in to a graphical or TTY session will start a different set of services). Then, upon the user's last log-out, the pam-stop-usertree script handles stopping all s6-rc managed user services that are still active.

Services not yet converted

Below rc scripts are not yet converted to s6-rc service definition:

• rc.sendmail
• rc.6 (pppd related stuff)

Notice of non-affiliation and disclaimer

Sl6ckware is not affiliated, associated, endorsed by, or in any way officially connected with Slackware. The official Slackware website can be found at http://www.slackware.com/.