They need to be quoted so that the shell does not parse them.

It might happen that the binary exists but the configuration file doesn't,
and then the insserv invocation will fail. Therefore don't consider insserv
installed until the configuration file exists.

Thanks to Petter Reinholdtsen for the patch.

Closes: #849284

This ensures systemd itself can order them the best way it sees fit.

Closes: #848192

See
https://www.freedesktop.org/software/systemd/man/systemd-escape.html
for details.

Closes: #861204

It is a hard to diagnose condition because trailing spaces are
invisible in most editors. Moreover, all the other parsers (systemd,
insserv) allow such trailing space.

Closes: #851730

In jessie, update-rc.d/invoke-rc.d/service are shipped by sysv-rc.
They got moved to init-system-helpers after that, so make sure we
have the new versions to ensure the scripts are available.

This dependency can probably be dropped after stretch is released.

See #851688

See #838966 and #838997 for background.

A debootstrap with --variant=minbase will have lsb-base installed,
so the person running into this issue must have:
 - not added a policy-rc.d in his init-less chroot
 - removed lsb-base manually (and what else?)

This patch auto-detects the situation with init-less chroot that's
missing a policy-rc.d and deals with it automatically (plus gives
a warning message).

Closes: #843761

Previously, when calling invoke-rc.d with a "custom" action, it
would print an error when a policy-rc.d file was installed, but
it would be silent if no such file was installed.

Closes: #728682

Closes: #839545

It is now provided by debhelper
Closes: #841914

This caused fallout at least in the autopkgtests of two packages, possibly many
more, thus could also break assumptions in more important environments, and was
just a defensive move, not a bug fix.

This reverts commit 9cfb6dfe.

There is a curious bug which sometimes causes "systemctl is-active
default.target" to say inactive until "show" or "status" gets called on the
unit. This needs to be investigated.

Until then, check for multi-user.target which by and large does the same job,
but seems to work reliably.

Reloading SysV init scripts has never been synchronous, so use --no-block to
reload them under systemd. This is another, and a better, line of defence
against if-up.d hook deadlocks during boot (see https://bugs.debian.org/777113)

"systemctl is-system-running" might still be false in case of running
jobs  for device/mount/hotplug/dynamic actions units. But in those cases we
already do want to respect unit dependencies, as the system is booted up
sufficiently to avoid dependency loops. Thus weaken the condition to
"default.target is active".

This does not change the behaviour for single-user: is-system-running has
always been false there, so dependencies continue to be ignored.

Fixes installation of packages like PostgreSQL under cloud-init or when
manually installing packages right after booting.

LP: #1576692

Because update-rc.d does not remove the /etc/rc?.d directories when
removing links if the directory is empty, every package that ships an
init script will leave behind the unowned /etc/rc?.d directories.

Let us mimic the behavior of sysv-rc (and even ourselves with the
/etc/systemd/system directory), and let dpkg know we own it.

Closes: #834524

Gbp-Dch: ignore

If there are any links, preserve them. In fallback mode we don't care
about the number encoded, but we do care about the S/K state. So instead
of trying to renumber back to 01, preserve the actually existing links.

This also fixes a bug that could cause multiple enable links to be
created for a single service.

Closes: #832945

This makes it easier to figure out if the fallback code in update-rc.d
was used or not.

insserv_toggle is not really insserv specific, so rename it to
sysv_toggle and also use it in case insserv is not installed for
enabling and disabling a SysV init script.

Closes: #832582

Restart services that are disabled but have been started manually.

A start request for a running service is a nop under systemd, so we
don't need to handle start and restart separately.

On start|restart we already enforce the local policy and RC will be set
to 101, if the service was disabled.
Also, checking for the output of "systemctl --quiet is-enabled" is
problematic, as it is currently broken for non-native services [1].
So drop it completely.

[1] https://github.com/systemd/systemd/issues/3813