[systemd-devel] How do I disable old init.d scripts?

Post by Paul Menzel
Dear systemd folks,
how can I disable init.d scripts which systemd loads for compatibility
reasons?
$ ls /etc/init.d/motd (or any other init.d script)
/etc/init.d/motd
$ systemd-analyze blame | grep motd
543ms motd.service
$ sudo systemctl disable motd.service
Failed to issue method call: No such file or directory

Try:

systemctl mask motd.service

regards,

Koen

Paul Menzel

2012-06-26 22:55:33 UTC

Post by Koen Kooi

Post by Paul Menzel
how can I disable init.d scripts which systemd loads for compatibility
reasons?
$ ls /etc/init.d/motd (or any other init.d script)
/etc/init.d/motd
$ systemd-analyze blame | grep motd
543ms motd.service
$ sudo systemctl disable motd.service
Failed to issue method call: No such file or directory

systemctl mask motd.service

That worked and userspace now takes less than three seconds on the
Debian Sid/unstable system.

Thanks a million,

Paul

Lennart Poettering

2012-06-27 09:14:50 UTC

Post by Koen Kooi

systemctl mask motd.service

That worked and userspace now takes less than three seconds on the
Debian Sid/unstable system.

You should be able to pull this down to < 1s with some love... ;-)

Just for he sake of curiosity, could you post the output of
"systemd-analyze plot" somewhere?

Lennart

--
Lennart Poettering - Red Hat, Inc.

Paul Menzel

2012-06-27 10:18:02 UTC

Post by Koen Kooi

systemctl mask motd.service

That worked and userspace now takes less than three seconds on the
Debian Sid/unstable system.

You should be able to pull this down to < 1s with some love... ;-)
Just for the sake of curiosity, could you post the output of
"systemd-analyze plot" somewhere?

I was hoping you will write that. Please find the files attached.

chrony and VDR still use the init.d scripts. For chrony I have not yet
come around to try the service files [1]. For VDR I started a thread on
***@linuxtv.org [2] (although this looks to be complicated thinking of
all the use cases).

Looking at the `systemd-analyze plot` output I would think PulseAudio
not depending on NetworkManager and GDM3 â maybe start in parallel to
NetworkManager and then wait for network connection â could start
earlier too.

And lastly the mainboard is an ASRock E350M1 (AMD Fusion E-350) and it
is a crucial m4 256 GB SSD [3]. I added the GDM3 service file and
disabled the init.d script motd. Everything else should be a standard
Debian Sid/unstable system.

Thanks,

Paul

[1] http://lists.freedesktop.org/archives/systemd-devel/2012-June/005596.html
[2] http://linuxtv.org/pipermail/vdr/2012-June/026487.html
[3] http://lists.freedesktop.org/archives/systemd-devel/2012-May/005181.html

Lennart Poettering

2012-06-27 10:35:46 UTC

Post by Lennart Poettering
You should be able to pull this down to < 1s with some love... ;-)
Just for the sake of curiosity, could you post the output of
"systemd-analyze plot" somewhere?

I was hoping you will write that. Please find the files attached.
chrony and VDR still use the init.d scripts. For chrony I have not yet
come around to try the service files [1]. For VDR I started a thread on
all the use cases).
Looking at the `systemd-analyze plot` output I would think PulseAudio
not depending on NetworkManager and GDM3 – maybe start in parallel to
NetworkManager and then wait for network connection – could start
earlier too.
And lastly the mainboard is an ASRock E350M1 (AMD Fusion E-350) and it
is a crucial m4 256 GB SSD [3]. I added the GDM3 service file and
disabled the init.d script motd. Everything else should be a standard
Debian Sid/unstable system.
acpid.service static

Should be pretty much redundant with systemd logind from 185 and newer.

Post by Paul Menzel
checkroot.service static

Should be redundant, done by systemd-fsck-root.service

Post by Paul Menzel
kmod.service static

Should be redundant, done by systemd-modules-load.service, no?

Post by Paul Menzel
module-init-tools.service static

Obsoleted by kmod, no?

Post by Paul Menzel
urandom.service static

Hmm, this is already done by systemd-random-load.service, no?

Post by Paul Menzel
712ms systemd-logind.service
615ms console-kit-log-system-start.service

No need for CK anymore, gdm and friends should support logind just
fine.

Post by Paul Menzel
580ms fglrx-atieventsd.service
540ms chrony.service
515ms rc.local.service

The rc-local generator should be smart enough to pull this in only if it
exists. It's a really slow service and most likely just a NOP anyway.

Post by Paul Menzel
509ms vdr.service
487ms sysfsutils.service

What is this? This stuff sounds like something that can just go away...

Post by Paul Menzel
484ms ssh.service
444ms cron.service
444ms udev.service
407ms systemd-user-sessions.service
360ms media.mount
359ms systemd-remount-api-vfs.service
328ms network-manager.service
328ms dev-mqueue.mount
319ms udev-trigger.service
319ms systemd-modules-load.service
300ms sys-kernel-debug.mount
288ms dev-hugepages.mount
287ms systemd-sysctl.service
256ms sys-kernel-security.mount
239ms networking.service
205ms gdm3.service
195ms hdparm.service

This really should go. It's relly unnecessary these days and if it is
really desired then should be done via a udev rule, not as a
service. (see other discussion on the ML)

Post by Paul Menzel
191ms console-setup.service

Appears to be something that systemd-vconsole-setup should already handle.

Post by Paul Menzel
191ms screen-cleanup.service

Something for tmpfiles?

Post by Paul Menzel
167ms systemd-tmpfiles-setup.service
147ms pulseaudio.service

Hmm, as a system service? Meh..

Post by Paul Menzel
87ms polkitd.service
79ms debian-fixup.service
71ms keyboard-setup.service

Also something systemd-vconsole-setup could do?

Post by Paul Menzel
63ms remount-rootfs.service
44ms console-kit-daemon.service

Again, CK is obsolete.

Post by Paul Menzel
40ms accounts-daemon.service
29ms boot-efi.mount

Hmm, just out of curiosity, what does Debian do here? they automatically
mount the EFI partition to /boot/efi? Is that listed in fstab? Can you
point me to some details on this?

Post by Paul Menzel
28ms upower.service
25ms udisks.service
18ms rtkit-daemon.service

Otherwise looks good!

Lennart

--
Lennart Poettering - Red Hat, Inc.

Michael Biebl

2012-06-27 12:50:35 UTC

Hi,

checkroot.service static

Should be redundant, done by systemd-fsck-root.service

/lib/systemd/system/checkroot.service -> remount-rootfs.service

This is just a symlink as synchronization point for LSB sysv services.

kmod.service static

Should be redundant, done by systemd-modules-load.service, no?

/lib/systemd/system/kmod.service -> systemd-modules-load.service

same here

module-init-tools.service static

Obsoleted by kmod, no?

/lib/systemd/system/module-init-tools.service -> systemd-modules-load.service

same here

urandom.service static

Hmm, this is already done by systemd-random-load.service, no?

/lib/systemd/system/urandom.service -> systemd-random-seed-load.service

same here

712ms systemd-logind.service
615ms console-kit-log-system-start.service

No need for CK anymore, gdm and friends should support logind just
fine.

gdm is currently not built with systemd support.

There are quite a few packages which don't support a runtime fallback
for CK, so we decided to stay with CK for wheezy.

Cheers,
Michael

--
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

Michael Biebl

2012-06-27 13:03:19 UTC

Post by Michael Biebl
Hi,

checkroot.service static

Should be redundant, done by systemd-fsck-root.service

/lib/systemd/system/checkroot.service -> remount-rootfs.service

FWIW, this is with systemd v44, where we don't have a
systemd-fsck-root.service, but fsck-root.service.

IIRC, the sysv checkroot init script also does a remount /, so
symlinking it to remount-rootfs.service seemed like the better fit.

For more info, see also [1] where we mask resp. symlink the sysv
services with native implementations.

Lennart,
if you find anything missing/incorrect in that list, let us know.

Michael

[1] http://git.err.no/cgi-bin/gitweb.cgi?p=systemd;a=blob;f=debian/systemd.links;h=55a7306daadd518d3644e9502e13fe2fafd15a40;hb=refs/heads/debian

--
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

Lennart Poettering

2012-06-27 16:39:12 UTC

Post by Michael Biebl

Post by Michael Biebl
Hi,

checkroot.service static

Should be redundant, done by systemd-fsck-root.service

/lib/systemd/system/checkroot.service -> remount-rootfs.service

FWIW, this is with systemd v44, where we don't have a
systemd-fsck-root.service, but fsck-root.service.
IIRC, the sysv checkroot init script also does a remount /, so
symlinking it to remount-rootfs.service seemed like the better fit.
For more info, see also [1] where we mask resp. symlink the sysv
services with native implementations.
Lennart,
if you find anything missing/incorrect in that list, let us know.
Michael
[1] http://git.err.no/cgi-bin/gitweb.cgi?p=systemd;a=blob;f=debian/systemd.links;h=55a7306daadd518d3644e9502e13fe2fafd15a40;hb=refs/heads/debian

Ah, this looks actually pretty good (I mean, for what it does). It's
kinda cool that we can do almost all Debian hookups just with a series
of symlinks. Neat!

Lennart

--
Lennart Poettering - Red Hat, Inc

Michael Biebl

2012-06-27 13:28:00 UTC

191ms console-setup.service
71ms keyboard-setup.service

Also something systemd-vconsole-setup could do?

The Debian systemd package doesn't do a migration of the
console/keyboard configuration yet so if you decide to blacklist those
sysv services you need to set up and manage /etc/vconsole.conf
yourself and you also need to enable systemd-vconsole-setup.service.
The Debian package explicitly removes the symlink from
sysinit.target.wants.

Michael

--
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

Lennart Poettering

2012-06-27 16:41:29 UTC

Post by Michael Biebl

191ms console-setup.service
71ms keyboard-setup.service

Also something systemd-vconsole-setup could do?

Hmm, the other distributions have an #ifdef TARGET_FOOBAR section in
vconsole-setup for that. Debian currently doesn't. I'd be willing to add
a patch that parses the old DEbian configuration files, to implement the
fallback.

Lennart

--
Lennart Poettering - Red Hat, Inc.

Tollef Fog Heen

2012-06-27 19:59:17 UTC

]] Lennart Poettering

Post by Lennart Poettering
Hmm, the other distributions have an #ifdef TARGET_FOOBAR section in
vconsole-setup for that. Debian currently doesn't. I'd be willing to add
a patch that parses the old DEbian configuration files, to implement the
fallback.

We use the X11 keyboard definitions. Are you sure you want to
completely reimplement console-setup? :-)

--
Tollef Fog Heen
UNIX is user friendly, it's just picky about who its friends are

Lennart Poettering

2012-06-27 20:32:51 UTC

Post by Tollef Fog Heen
]] Lennart Poettering

We use the X11 keyboard definitions. Are you sure you want to
completely reimplement console-setup? :-)

Ah, hmm. I heard of that. Can you explain how precisely that works? Do
you actually use the same keybaord definitions, i.e. have converted them
to console definitions? Or do you just share the same naming namespace?

Presumably the console kbd subsystem is much less powerful than X, so i
wonder how this maps.

Any docs available for this?

Honestly this actually sounds a lot better than the non-sense we are
currently doing. This probably something we should just follow debian
in. (and iirc i saw a feature flying by about this for fedora...)

Lennart

--
Lennart Poettering - Red Hat, Inc.

Tollef Fog Heen

2012-06-28 05:54:02 UTC

]] Lennart Poettering

Post by Tollef Fog Heen
]] Lennart Poettering

We use the X11 keyboard definitions. Are you sure you want to
completely reimplement console-setup? :-)

console-setup reads the xkb definitions and downsamples them to
something the console can understand. IIRC, it doesn't work for the
most complex keymaps and the cases where you have multiple keymaps
defined which you can switch between, but it works well in the simple
case of one language. (IIRC, this is due to kernel limitations in how
many keys can be defined and such. Also, I don't think the kernel
understands Compose.)

Post by Lennart Poettering
Any docs available for this?

There might be some in
http://ftp.debian.org/debian/pool/main/c/console-setup/console-setup_1.78.tar.gz

--
Tollef Fog Heen
UNIX is user friendly, it's just picky about who its friends are

Paul Menzel

2012-06-28 07:05:42 UTC

Post by Tollef Fog Heen
]] Lennart Poettering

Post by Tollef Fog Heen
]] Lennart Poettering

We use the X11 keyboard definitions. Are you sure you want to
completely reimplement console-setup? :-)

Post by Lennart Poettering
Any docs available for this?

There might be some in
http://ftp.debian.org/debian/pool/main/c/console-setup/console-setup_1.78.tar.gz

You can also clone the Git repository.

$ debcheckout -d console-setup
type git
url git://git.debian.org/d-i/console-setup.git
top-bases
topgit no

Thanks,

Paul

Lennart Poettering

2012-06-28 11:21:44 UTC

On Thu, 28.06.12 07:54, Tollef Fog Heen (***@err.no) wrote:

Heya,

Post by Tollef Fog Heen

Post by Tollef Fog Heen
]] Lennart Poettering

We use the X11 keyboard definitions. Are you sure you want to
completely reimplement console-setup? :-)

Ah, pretty cool. I have now added an item to the TODO list, so that we
support this scheme natively in localed and systemd-vconsole-setup. It
appears the much better solution over what we are currently using to
setup the kbd on Fedora.

Lennart

--
Lennart Poettering - Red Hat, Inc.

Colin Guthrie

2012-06-29 01:00:22 UTC

This post might be inappropriate. Click to display it.

Paul Menzel

2012-06-30 11:47:15 UTC

[âŠ]

Post by Paul Menzel
540ms chrony.service
515ms rc.local.service

The rc-local generator should be smart enough to pull this in only if it
exists. It's a really slow service and most likely just a NOP anyway.

Right. The default file `/etc/rc.local` just contains `exit 0`. So I
masked that service file.

$ sudo systemctl mask rc.local.service

Post by Paul Menzel
509ms vdr.service
487ms sysfsutils.service

What is this? This stuff sounds like something that can just go away...

I was surprised too. It looks like this was installed by
`grml-debootstrap` and I removed it using the following command.

$ sudo aptitude purge --purge-unused sysfsutils

[âŠ]

Post by Paul Menzel
191ms screen-cleanup.service

Something for tmpfiles?

I have to look into that.

Post by Paul Menzel
167ms systemd-tmpfiles-setup.service
147ms pulseaudio.service

Hmm, as a system service? Meh..

Well, no. Debian still ships an init.d script which reads
`/etc/default/pulseaudio` and there system mode is disabled by default.
So I masked this one too.

$ sudo systemctl mask pulseaudio.service

[âŠ]

Thanks,

Paul

Lennart Poettering

2012-07-03 00:26:55 UTC

Post by Paul Menzel
509ms vdr.service
487ms sysfsutils.service

What is this? This stuff sounds like something that can just go away...

I was surprised too. It looks like this was installed by
`grml-debootstrap` and I removed it using the following command.
$ sudo aptitude purge --purge-unused sysfsutils

I pretty sure this is a package to remove from the distribution
entirely.

Post by Lennart Poettering
Hmm, as a system service? Meh..

Well, no. Debian still ships an init.d script which reads
`/etc/default/pulseaudio` and there system mode is disabled by default.
So I masked this one too.
$ sudo systemctl mask pulseaudio.service

Hmm, dounds a bit strange that on Debian you enable/disable that script
via a file in /etc/default, even though sysv already has a mechanism for
this, which is the done by update-rcd, right? Sounds like two levels of
turning things off where one would be sufficient and less confusing...

Lennart

--
Lennart Poettering - Red Hat, Inc.

Michael Biebl

2012-07-03 01:09:29 UTC

Post by Paul Menzel
Well, no. Debian still ships an init.d script which reads
`/etc/default/pulseaudio` and there system mode is disabled by default.
So I masked this one too.
$ sudo systemctl mask pulseaudio.service

Those /etc/default/<foo> files with ENABLE/DISABLE switches are a plague.

http://lists.alioth.debian.org/pipermail/initscripts-ng-devel/2007-June/000461.html

--
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

Colin Guthrie

2012-07-03 09:37:25 UTC

This post might be inappropriate. Click to display it.

Paul Menzel

2012-06-30 12:43:44 UTC

Am Mittwoch, den 27.06.2012, 12:35 +0200 schrieb Lennart Poettering:

[âŠ]

Post by Paul Menzel
29ms boot-efi.mount

Hmm, just out of curiosity, what does Debian do here? they automatically
mount the EFI partition to /boot/efi? Is that listed in fstab? Can you
point me to some details on this?

It is listed in `/etc/fstab`.

/dev/sda3 /boot/efi vfat defaults 0 2

Since I did not find any official information regarding Debian and UEFI
I followed the blog post from Tanguy Ortolo [1] which states to put that
entry into `/etc/fstab`. Although I think it still does not work since
`efibootmgr` does not work.

Thanks,

Paul

[1] http://tanguy.ortolo.eu/blog/article51/debian-efi

Lennart Poettering

2012-06-25 13:38:15 UTC

On Fedora we defer to chkconfig automatically for sysv services. This
hookup is currently missing for Debian, but I'd be willing to merge a
patch for that.

Lennart

--
Lennart Poettering - Red Hat, Inc.

Paul Menzel

2012-06-25 19:33:21 UTC

[âŠ]

On Fedora we defer to chkconfig automatically for sysv services.

Looking at the code

$ git grep -i chkconfig

this seems to happen for the following distributions.

#if defined (HAVE_SYSV_COMPAT) && (defined(TARGET_FEDORA) || defined(TARGET_MANDRIVA) || defined(TARGET_SUSE) || defined(TARGET_MEEGO) || defined(TARGET_ALTLINUX) || defined(TARGET_MAGEIA))

Post by Lennart Poettering
This hookup is currently missing for Debian, but I'd be willing to merge a
patch for that.

If I am not mistaken, Debian uses `update-rc.d` for that purpose.

Does someone know if it is feasible to just change the one variable (and
options) containing the path to the utilities and reuse the error
handling or will that not work?

Thanks,

Paul

Lennart Poettering

2012-06-27 09:13:54 UTC