sysvinit Configuration — The Complete File Map

Overview

sysvinit has no monolithic configuration daemon to talk to — no central store, no schema, no reload API. Configuration is a constellation of small files, each read by exactly one consumer at one well-defined moment: /etc/inittab is read by PID 1, a handful of /etc/default/* panels are read by specific boot and shutdown scripts, /etc/fstab is read by the fsck-and-mount pass, and internal artifacts like /etc/init.d/.depend.* are read by the rc transition engine itself. /etc/inittab is only the entry point of this map, not the whole of it — treating it as “the sysvinit config file” is the single most common mental-model error.

This page is the full map, verified against the actual Debian 12 (bookworm) packages rather than folklore. The shipped initscripts and sysv-rc binaries were unpacked and grepped script by script, and every variable documented below is quoted from the real file contents or from the man pages those packages ship (rcS(5), halt(5), tmpfs-config(5)). Where tradition says one thing and bookworm does another — for example, the widely repeated claim that “the rcS.d scripts source /etc/default/rcS” — this page follows the packages, and says what actually happens.

The tour proceeds in dependency order: the who-reads-what table first, then /etc/inittab in brief (the full grammar lives in inittab.md), then each /etc/default file with its verified variable set, then the rc script that ties the runlevel machinery together, the remaining rcS-pass inputs, the per-service defaults convention, runtime-reconfiguration semantics, and a cookbook of exact tuning edits. The grammar details of individual files are deliberately not repeated here — each deep dive links the page that owns them.

Who Reads What — the File Map

Every file in the table below was verified against the bookworm packages; the consumer column names the script that literally sources or reads it:

Two structural observations. First, no boot script sources /etc/default/rcS directly except rc itself — the values reach the rcS.d scripts indirectly, through /lib/init/vars.sh (proven below). Second, the /etc/default/locale row proves the defaults-file convention is not limited to daemons: even a mount script reads one. That convention is the subject of its own section near the end of this page.

The same map as a picture — every arrow is a literal . file or read inside a bookworm script:

flowchart TB
    INIT["/sbin/init reads /etc/inittab"] --> RC["/lib/init/rc via the /etc/init.d/rc symlink"]
    RC -->|"sources it, exports VERBOSE"| DRC["/etc/default/rcS"]
    DRC --> VARS["/lib/init/vars.sh"]
    VARS -->|"defaults, quiet and INIT_VERBOSE handling"| SCRIPTS["rcS.d and rcN.d scripts"]
    RC -->|"makefile mode via startpar"| DEP["/etc/init.d/.depend.*"]
    RC --> SCRIPTS
    HALT["/etc/init.d/halt and /etc/init.d/reboot"] --> HALTD["/etc/default/halt"]
    MDS["/etc/init.d/mountdevsubfs.sh"] --> DEVD["/etc/default/devpts"]
    TMP["/lib/init/tmpfs.sh"] --> TMPD["/etc/default/tmpfs"]

/etc/inittab in One Screen

The full id:runlevels:action:process grammar, all fifteen actions, and the annotated bookworm file live in inittab.md. For the configuration story, three lines matter daily:

id:2:initdefault:              # which runlevel boot ends in
si::sysinit:/etc/init.d/rcS    # the one-shot rcS pass (fsck, mounts, udev)
ca:12345:ctrlaltdel:/sbin/shutdown -t1 -a -r now

initdefault selects the destination runlevel (Debian boots to 2); the sysinit line runs /etc/init.d/rcS — itself a two-line wrapper that execs rc S, so the entire boot pass is the same engine as a runlevel change; ctrlaltdel decides what the three-finger salute does. Everything else in the stock file is the seven lN:...:wait:/etc/init.d/rc N lines, getty respawns, and UPS hooks.

/etc/inittab is also the only sysvinit file with a live-reload path: PID 1 re-reads it on telinit q. No other file on this page is ever re-read by a running process — they are consumed once, at script invocation or at boot.

/etc/default/rcS — the Boot Defaults Panel

/etc/default/rcS belongs to the initscripts package and documents itself as “Default settings for the scripts in /etc/rcS.d/”. rcS(5) specifies the format: POSIX VAR=VAL, one assignment per line, # comments. The shipped file opens with a header that matters on any dual-init machine:

##################################################################
# NOTE: This file is ignored when systemd is used as init system #
##################################################################
#
# /etc/default/rcS
#
# Default settings for the scripts in /etc/rcS.d/
#
# For information about these variables see the rcS(5) manual page.
#
# This file belongs to the "initscripts" package.

# delete files in /tmp during boot older than x days.
# '0' means always, -1 or 'infinite' disables the feature
#TMPTIME=0

# spawn sulogin during boot, continue normal boot if not used in 30 seconds
#SULOGIN=no

# do not allow users to log in until the boot has completed
#DELAYLOGIN=no

# be more verbose during the boot process
#VERBOSE=no

# automatically repair filesystems with inconsistencies during boot
#FSCKFIX=no

# be verbose even if kernel command line contains "quiet"
#INIT_VERBOSE=yes

# Mount NFS filesystems asynchronously
#ASYNCMOUNTNFS=yes

(abridged: comments above are verbatim; nothing else was removed). Note that every assignment ships commented out. The shipped file is documentation; the effective defaults live in code.

The variables

Who actually reads it

Grepping the bookworm initscripts package settles a point most documentation gets wrong: no rcS.d boot script sources /etc/default/rcS directly. The verified chain is:

• /lib/init/rc (the runlevel engine) sources it unconditionally when present and exports VERBOSE to every script it runs.
• Every boot script — checkroot.sh, checkfs.sh, hostname.sh, mountnfs.sh, rmnologin, bootmisc.sh, mountall.sh — sources /lib/init/vars.sh instead, which carries the code defaults, sources /etc/default/rcS, and then applies the kernel-command-line and INIT_VERBOSE logic.
• mountnfs.sh says so in its own comment: “Using ‘no !=’ instead of ‘yes =’ to make sure async nfs mounting is the default even without a value in /etc/default/rcS.”

The bookworm /lib/init/vars.sh (from the sysvinit source, shipped to /lib/init/vars.sh) is short enough to quote in full — it is the real defaults panel:

# Because /etc/default/rcS isn't a conffile, it's never updated
# automatically.  So that an empty or outdated file missing newer
# options works correctly, set the default values here.
TMPTIME=0
SULOGIN=no
DELAYLOGIN=no
UTC=yes
VERBOSE=no
FSCKFIX=no

# Source conffile
if [ -f /etc/default/rcS ]; then
    . /etc/default/rcS
fi

# Unset old unused options
unset EDITMOTD
unset RAMRUN
unset RAMLOCK
# Don't unset RAMSHM and RAMTMP for now.

# Parse kernel command line
if [ -r /proc/cmdline ]; then
    for ARG in $(cat /proc/cmdline); do
        case $ARG in
            noswap)     NOSWAP=yes ;;
            quiet)      # only during boot/shutdown, when rc's vars are present
                        if [ "$RUNLEVEL" ] && [ "$PREVLEVEL" ] ; then
                            VERBOSE="no"
                        fi ;;
        esac
    done
fi

# But allow both rcS and the kernel options 'quiet' to be overrided
# when INIT_VERBOSE=yes is used as well.
if [ "$INIT_VERBOSE" ] ; then
    VERBOSE="$INIT_VERBOSE"
fi

Retired variables

rcS(5) closes with a graveyard list: EDITMOTD, RAMRUN, CONCURRENCY, and UTC are no longer used (UTC moved to the UTC/LOCAL setting in /etc/adjtime; concurrent boot is inhibited with the concurrency=none kernel argument, not a variable). RAMLOCK, RAMSHM, and RAMTMP moved to /etc/default/tmpfs — old rcS values for RAMSHM/RAMTMP are still honored for backward compatibility but lose to /etc/default/tmpfs. Setting them in rcS today is a no-op waiting to confuse the next administrator.

/etc/default/halt, /etc/default/devpts, /etc/default/tmpfs

Three more defaults panels, each with a single verified consumer. Unlike rcS, these ship their assignments active (uncommented).

/etc/default/halt — shutdown behavior

The whole file, verbatim:

# Default behaviour of shutdown -h / halt. Set to "halt" or "poweroff".
HALT=poweroff

# Bring networking down right before halting/rebooting system. You
# should only need this to handle issues with Wake-on-LAN or network
# filesystems.  Set to "yes" or "no".
NETDOWN=yes

The consumer is verified and double: /etc/init.d/halt (line 15) and /etc/init.d/reboot (line 15) both source this file — NETDOWN applies on the reboot path too, which the filename alone would never suggest. halt(5) documents both variables; the surrounding shutdown choreography is covered in shutdown-halt.md.

/etc/default/devpts — pty ownership and mesg

# GID of the `tty' group
TTYGRP=5

# Set to 620 to have `mesg y' be the default
TTYMODE=600

mountdevsubfs.sh is the only consumer; it sets its own inline defaults (TTYGRP=5, TTYMODE=620), then sources this file, then mounts devpts with -onoexec,nosuid,gid=$TTYGRP,mode=$TTYMODE. The interplay is the whole point: mode 620 (group tty writable) is what makes write access to your terminal the default — the mesg y state that talk/write and wall need — while the shipped 600 leaves terminals private unless mesg y is run by hand (see mesg(1)). So the file’s comment is the manual’s hint in disguise: to restore old-school mesg y-by-default behavior, set TTYMODE=620 here rather than editing the script.

/etc/default/tmpfs — early-boot tmpfs mounts

The panel for the tmpfs filesystems mounted before fstab mounts: /run (always tmpfs), plus optionally /run/lock, /run/shm, and /tmp. Consumer: /lib/init/tmpfs.sh, which sets the code defaults, sources this file, and computes the mount options. Documented by tmpfs-config(5).

The size grammar, implemented by tmpfs_size_vm in tmpfs.sh: a bare integer is KiB; N% is N percent of physical RAM; N%VM is N percent of virtual memory (RAM + swap), computed at boot because the kernel cannot parse that suffix — which is also why tmpfs-config(5) warns that %VM may be used in these variables but not in /etc/fstab. An fstab entry for any of these mount points overrides this file entirely (tmpfs /tmp tmpfs nodev,nosuid,size=2g,mode=1777 0 0 beats RAMTMP), and a /tmp symlink trick (pointing /tmp into a tmpfs like /run/tmp) is documented in the package’s own README.

The rc Script Itself

Bookworm layout: /etc/init.d/rc is a symlink

The package that owns the runlevel machinery is sysv-rc, and in bookworm it ships the engine at /lib/init/rc with compatibility symlinks in the traditional place — verified from the package’s own file listing:

/etc/init.d/rc  -> /lib/init/rc
/etc/init.d/rcS -> /lib/init/rcS

/lib/init/rcS is a two-line wrapper — exec /etc/init.d/rc S — so inittab’s si::sysinit line and the seven lN:...:wait lines all reach the same engine. The move out of /etc/init.d separates implementation from the service-script directory that admins and tools enumerate; /etc/init.d/rc remains only so old muscle memory and old tooling still resolve.

What rc sources, and the environment contract

The engine consumes more than it is given credit for. At startup it reads the runlevel transition from its environment — $RUNLEVEL (current) and $PREVLEVEL (previous, N on first boot), both exported by /sbin/init per init(8) — with the single command-line argument overriding $RUNLEVEL. It then sources /etc/default/rcS when present, explicitly exports VERBOSE, and loads /lib/lsb/init-functions. If the exit flow is ever broken it fires an emergency trap printing error: '/etc/init.d/rc' exited outside the expected code flow., and it ignores INT/QUIT/TSTP so a stray Ctrl-C cannot kill a transition. There is no re-exec and no failure-driven panic path: runlevels 0 and 6 flip the action to stop, S means a start pass, and the K-scripts run first only when $PREVLEVEL is not N.

CONCURRENCY and the silent fallback

The concurrency decision is made fresh on every invocation, and its failure mode is silence:

CONCURRENCY=makefile
test -s /etc/init.d/.depend.boot  || CONCURRENCY="none"
test -s /etc/init.d/.depend.start || CONCURRENCY="none"
test -s /etc/init.d/.depend.stop  || CONCURRENCY="none"
if test -e /etc/init.d/.legacy-bootordering; then
        CONCURRENCY="none"
fi
if grep -wqs concurrency=none /proc/cmdline; then
        CONCURRENCY="none"
fi

Makefile-style concurrent boot requires insserv’s .depend.{boot,start,stop} graph and startpar(1) (invoked as startpar -p 4 -t 20 -T 3 -M boot|stop|start -P $previous -R $runlevel, with failed/skipped service names reported back). If any graph file is missing or empty, if the .legacy-bootordering marker exists, or if the kernel command line says concurrency=none, rc quietly drops to the serial loop — no error, just a slower boot. So a boot that mysteriously stopped being parallel is diagnosed by checking for those three files, not by hunting for an error message. The full mechanics of the parallel path are in parallel-booting.md.

Why you must not hand-edit it

/lib/init/rc is dpkg-owned by sysv-rc and is replaced verbatim on every package upgrade — anything you patch into it is silently lost, and because it is not a conffile you get no merge prompt, no dpkg-dist leftover, nothing. It is also on the critical path of every boot and every runlevel change, where an experimental edit can leave a machine that boots to nothing. Patch points exist for every purpose the script serves: boot defaults go in /etc/default/rcS, service ordering in LSB headers or insserv overrides (regenerate with insserv -d — see tooling.md), and end-of-boot customization in /etc/rc.local. Edit the panel, not the engine.

Files Consumed by the rcS Pass

The one-shot boot pass reads the rest of the constellation, script by script:

• /etc/fstab — read by four consumers with different jobs. checkroot.sh verifies the root entry against the actual root device and activates swap (swapon -a -e); checkfs.sh runs fsck -A, which walks fstab honoring fs_passno — pass 1 filesystems (usually just root) are checked serially, pass 2 filesystems can be checked in parallel by fsck itself; mountall.sh does mount -a -t nonfs,nfs4,smbfs,cifs,... -O no_netdev for local filesystems; mountnfs.sh waits on the async NFS mount path (or mounts synchronously when ASYNCMOUNTNFS=no). The pass field is therefore an ordering contract between fstab and fsck(8), not a sysvinit feature — but the rcS scripts are what execute it.
• /etc/hostname — hostname.sh reads it verbatim (HOSTNAME="$(cat /etc/hostname)", returning silently if the file is absent) and applies it with hostname(1). One line, no comments, no FQDN tricks — the domain side belongs to /etc/hosts and resolver configuration.
• /etc/network/interfaces — consumed by ifup(8) (the ifupdown package) via /etc/init.d/networking, which runs during the rcS pass to bring up interfaces marked auto. Bookworm sysvinit installs keep this path; the file is ifupdown-owned, not initscripts-owned, which is why it is absent from the initscripts package listing.
• /run/nologin (historically /etc/nologin) — the DELAYLOGIN flag file. Verified round trip: bootmisc.sh writes System bootup in progress - please wait into it when DELAYLOGIN is yes; rmnologin (run at the end of the multiuser levels) removes it; while it exists, login(1) refuses non-root logins. rcS(5) adds the operational footnote: if you set DELAYLOGIN=no, make sure no stale /run/nologin is lying around, or you have built a permanent login blocker.

Per-Service Configuration: /etc/default/<service>

The same panel pattern scales to every service. The Debian convention: an init script sources /etc/default/<name> near the top (usually conditionally — [ -r /etc/default/myapp ] && . /etc/default/myapp — so a missing file cannot break the boot), sets its own code defaults first, and lets the file’s assignments win. Package upgrades overwrite /etc/init.d/<name> but never your /etc/default/<name>, which is the entire point: maintainer defaults in package-owned code, operator intent in a file dpkg does not touch.

The bookworm tree is full of instances beyond the boot panel: /etc/default/locale is read even by mountall.sh (to export LANG for filesystem helpers), and /etc/default/ssh carries SSHD_OPTS, the standard hook for extra sshd command-line options without touching the packaged script. A complete worked example — a myapp service with a DAEMON_ARGS override, the conditional-sourcing habit, and the full init script around it — is built end-to-end in custom-init-scripts.md; the pattern to internalize here is one line of that page:

[ -r /etc/default/myapp ] && . /etc/default/myapp

Variables are not exported by the sourcing line, so a defaults file that wants to influence child processes must export explicitly — a detail that explains most “my defaults file only half-works” reports.

Root Login Policy: /etc/securetty

/etc/securetty lists device names, one per line, on which root is permitted to log in — console, tty1–tty6, ttyS0, and so on; a device not listed refuses root. The consumer is login(1), which consults it for root sessions; the interface itself is documented in securetty(5). Its scope is deliberately narrow: it gates where root may log in from, not whether the account works — disable root logins entirely with passwd -l or SSH configuration, not with an empty securetty (an empty file merely confines root to zero terminals, which can lock you out of the console you needed).

The interaction with single-user mode is the subtle part. The sysvinit path into single-user runs sulogin(8) (inittab’s ~~:S:wait:/sbin/sulogin, and checkroot.sh’s SULOGIN=yes hook), and the util-linux sulogin shipped in bookworm documents no /etc/securetty check at all — single-user access is gated by the root password, and with -e/--force sulogin will even start a shell without a password when the root account is locked, on the explicit assumption that the console is physically protected. So the two root-entry paths have different policies: login(1) honors securetty, sulogin(8) does not, and console physical security is the control that covers the difference. Deep coverage of that mode lives in single-user-sulogin.md and getty-terminals.md.

Runtime Reconfiguration vs Reboot

Nothing in sysvinit is globally hot-reloaded; each file has its own moment of consumption:

• /etc/inittab — the exception: PID 1 re-reads it on telinit q, spawning or killing respawn entries immediately (the full procedure is in inittab.md).
• /etc/default/<service> — takes effect on the next invocation of the consuming script: service myapp restart picks up new daemon options, but the rcS boot panels (rcS, halt, devpts, tmpfs) are read only during boot or shutdown, so they need a reboot (or at minimum the next shutdown) to matter.
• /etc/fstab — read on every script invocation that consumes it, so mount -a and the boot pass agree; but fsck pass ordering is a boot-time property.
• .depend.* — regenerated with insserv -d after header or override changes (tooling.md has the exact session); rc re-tests their presence at every invocation, so a regeneration changes the next transition, not the running one. Link surgery in rcN.d is covered in rc-symlinks.md.

The design consequence: sysvinit reconfiguration is idempotent and per-consumer. There is no daemon whose state can drift from the files — but also no single command that says “apply everything.”

Tuning Recipes

Exact edits, each verified against the consumers above:

1. Verbose boot. Uncomment VERBOSE=yes in /etc/default/rcS — it overrides both the no default and a quiet kernel argument (the quiet handling in vars.sh only lowers VERBOSE before INIT_VERBOSE gets its turn; VERBOSE=yes set after that logic wins because the file is sourced between default and override). To silence a normal boot instead, keep the kernel quiet argument and leave rcS alone.
2. Always fsck -y at boot. FSCKFIX=yes in /etc/default/rcS — checkroot.sh and checkfs.sh then pass -y instead of -a to every fsck. Understand the trade: genuine hardware-caused inconsistency gets “repaired” automatically; a manual-intervention prompt (which would have stopped for a reason) never happens.
3. /tmp as tmpfs with a cap. RAMTMP=yes plus TMP_SIZE=1G (or rely on TMPFS_SIZE=20%VM) in /etc/default/tmpfs; note /tmp contents vanish at reboot, and an fstab entry for /tmp would override this file entirely. The TMP_OVERFLOW_LIMIT emergency path still applies.
4. poweroff vs halt on shutdown -h. HALT=poweroff (default) cuts power; HALT=halt stops the OS and leaves the machine on, per halt(5). Consumed at the end of runlevel 0, not by shutdown(8) itself.
5. Keep Wake-on-LAN through shutdown. NETDOWN=no in /etc/default/halt — the interfaces stay up because halt/reboot are called without -i. Works for both poweroff and reboot paths since both scripts source the file.
6. Boot-time maintenance login. SULOGIN=yes in /etc/default/rcS — checkroot.sh spawns sulogin -t 30 $CONSOLE before anything else; thirty idle seconds and boot continues unattended.
7. Stop wiping /tmp on reboot. TMPTIME=-1 (or infinite) in /etc/default/rcS — bootclean.sh skips the whole sweep. TMPTIME=0 is the opposite extreme: no age filter at all, so even future-dated files go. Only the lock files and lost+found-class exclusions survive any setting.
8. Synchronous NFS mounting. ASYNCMOUNTNFS=no in /etc/default/rcS — mountnfs.sh stops waiting on the async mount done when interfaces come up and instead runs the mount path itself. Intended for NFS-root setups, per rcS(5).

Interview Questions

Q: Who actually sources /etc/default/rcS — and why is “the rcS.d scripts” the wrong answer?

/lib/init/rc sources it directly and exports VERBOSE; the rcS.d boot scripts do not touch it. They source /lib/init/vars.sh, which carries the code defaults (TMPTIME=0, SULOGIN=no, DELAYLOGIN=no, VERBOSE=no, FSCKFIX=no), sources /etc/default/rcS on their behalf, unsets dead options, parses the kernel command line for quiet/noswap, and applies the INIT_VERBOSE override. The proof is a grep: the bookworm initscripts package contains no . /etc/default/rcS outside rc — and mountnfs.sh’s own comment confirms values arrive via this chain.

Q: What does FSCKFIX=yes change, and what is the risk you accept?

checkroot.sh and checkfs.sh switch the fsck invocation from -a (autorepair of trivial issues, bail out and give a sulogin prompt on serious inconsistency) to -y (answer yes to every repair). You accept that a filesystem inconsistency caused by failing hardware gets auto-“repaired” aggressively — possibly destroying recoverable data — in exchange for a boot that never blocks on an fsck prompt. Appropriate for unattended appliances; a calculated risk everywhere else.

Q: HALT and NETDOWN in /etc/default/halt — what does each control, and which script consumes them?

HALT selects the final action of runlevel 0: poweroff (default) cuts power, halt leaves the box powered on the login screen of its firmware/console. NETDOWN decides whether the last halt/reboot step passes -i, tearing interfaces down; no keeps them up so Wake-on-LAN still works. Both /etc/init.d/halt and /etc/init.d/reboot source the file — the reboot consumer is the non-obvious part.

Q: Under what conditions does rc abandon makefile-style concurrent boot?

Four, all checked on every invocation: any of .depend.boot, .depend.start, .depend.stop missing or zero-length; the presence of /etc/init.d/.legacy-bootordering; concurrency=none on the kernel command line; and (implicitly) startpar being unavailable. The fallback is silent — the serial walk just runs — which is why “my boot stopped being parallel” is diagnosed by checking for those files, not by reading logs for an error.

Q: What is /etc/securetty for, and does it protect single-user mode?

It lists the devices on which login(1) will permit a root login; the interface is securetty(5). It does not protect single-user mode: the sulogin shipped in bookworm’s util-linux performs no securetty check — its man page documents none — and its --force path can even skip the password on a locked root account, assuming a physically protected console. The control that covers sulogin is physical/console security (and BIOS/bootloader hardening), not this file.

Q: Why is /etc/init.d/rc a symlink to /lib/init/rc in bookworm, and what should you patch instead?

Because the engine is implementation, not a service script: sysv-rc owns /lib/init/rc and leaves the traditional name as a symlink so inittab lines and old tooling still resolve. The real file is dpkg-owned and replaced verbatim on every upgrade — not a conffile, so hand edits vanish silently with no merge prompt. Everything rc reads is a patch point: /etc/default/rcS for boot behavior, insserv headers/overrides plus insserv -d for ordering, /etc/rc.local for end-of-boot work.

References

• rcS(5) — initscripts man page, Debian bookworm — the normative variable list for /etc/default/rcS, including the retired-variable graveyard.
• halt(5) — initscripts man page, Debian bookworm — documents HALT and NETDOWN.
• tmpfs-config(5) — initscripts man page, Debian bookworm — the /etc/default/tmpfs variables, the %VM size grammar, and fstab precedence.
• init(8) — sysvinit-core man page, Debian bookworm — the RUNLEVEL/PREVLEVEL environment contract and telinit control.
• inittab(5) — sysvinit-core man page, Debian bookworm — the grammar behind the one-screen section above.
• telinit(8) — sysvinit-core man page, Debian bookworm — the q re-read and runlevel change interface.
• halt(8) — sysvinit-core man page, Debian bookworm — the -i/-p/-h flags the halt script assembles.
• shutdown(8) — sysvinit-core man page, Debian bookworm — how shutdown -h reaches the halt script.
• sulogin(8) — util-linux man page, Debian bookworm — the single-user login, its -t/-e options, and the absence of a securetty check.
• login(1) — login man page, Debian bookworm — the consumer of /etc/securetty.
• mesg(1) — util-linux man page, Debian bookworm — the terminal-write permission tied to TTYMODE.

Cross-References

• inittab — The Master Configuration — the full grammar this page compresses to three knobs.
• The sysvinit Boot Sequence — when in the timeline each file above is consumed.
• rc0.d–rc6.d, rcS.d — Symlink Sequencing Mechanics — the link sets whose regeneration the runtime-reconfiguration section describes.
• Parallel Booting and startpar — the makefile-mode path rc selects when .depend.* are intact.
• Writing a Production init.d Script — the complete worked /etc/default/<service> example.
• Single-User Mode and sulogin — the sulogin side of the securetty discussion.
• Getty and Terminal Management — how respawned terminals relate to securetty device names.
• Shutdown and Halt Mechanics — the runlevel 0/6 choreography that ends at /etc/default/halt.
• SysVinit Service Management Tooling — insserv, update-rc.d, and the .depend.* regeneration session.
• systemd — Configuration — the sibling config map on the systemd side of the divide.
• SysVinit admin overview — the summary-level admin page for this init system.
• Init Systems Hub — section map and reading order.
• Init System Comparison — how each init system answers the who-reads-what question.