Debian Bug report logs - #657990
systemd: /dev/initctl has moved to /run/initctl

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Laurent Bigonville <bigon@debian.org>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: systemd: /dev/initctl has moved to /run/initctl

Date: Mon, 30 Jan 2012 17:05:38 +0100

Package: systemd
Version: 37-1
Severity: grave
Justification: renders package unusable

Hi,

/dev/initctl has moved to /run/initctl.

This means that shutdown and other commands are now failing with errors
like "init: timeout opening/writing control channel /run/initctl"
preventing the user to reboot or shutdown his machine.

Cheers

Laurent Bigonville


-- System Information:
Debian Release: wheezy/sid
  APT prefers unstable
  APT policy: (500, 'unstable'), (1, 'experimental')
Architecture: amd64 (x86_64)

Kernel: Linux 3.2.0-1-amd64 (SMP w/4 CPU cores)
Locale: LANG=fr_BE.utf8, LC_CTYPE=fr_BE.utf8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash

Versions of packages systemd depends on:
ii  initscripts         2.88dsf-21
ii  libacl1             2.2.51-5
ii  libaudit0           1.7.18-1
ii  libc6               2.13-24
ii  libcap2             1:2.22-1
ii  libcryptsetup1      2:1.3.0-3.1
ii  libdbus-1-3         1.4.16-1
ii  libpam0g            1.1.3-7
ii  libselinux1         2.1.0-4.1
ii  libsystemd-daemon0  37-1
ii  libsystemd-login0   37-1
ii  libudev0            175-3
ii  libwrap0            7.6.q-22
ii  udev                175-3
ii  util-linux          2.20.1-1.2

Versions of packages systemd recommends:
ii  libpam-systemd  37-1

Versions of packages systemd suggests:
ii  python       2.7.2-10
ii  systemd-gui  37-1

-- no debconf information

Message #10 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Adrien CLERC <bugs-debian@antipoul.fr>

To: 657990@bugs.debian.org

Subject: Temporary workaround

Date: Tue, 31 Jan 2012 07:29:49 +0100

Hi,

This is a very annoying bug, since everything rely on "shutdown" and 
"reboot" command.

However calling directly "systemctl poweroff" or "systemctl reboot" works.

But it could be nice for systemd to listen to the right FIFO ;)

Have a good day,
Adrien

Message #15 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Alexander Hofbauer <alex@derhofbauer.at>

To: 657990@bugs.debian.org

Subject: Re: Temporary workaround

Date: Tue, 31 Jan 2012 15:04:58 +0100

[Message part 1 (text/plain, inline)]

Real workaround for those annoyed by the bug until it gets fixed:

Replace the FIFO-name (ListenFIFO) in systemd's socket file
(/lib/systemd/system/systemd-initctl.socket) with /run/initctl.

[signature.asc (application/pgp-signature, inline)]

Message #20 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org

Cc: Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Tue, 31 Jan 2012 19:46:03 +0100

[Message part 1 (text/plain, inline)]

On 30.01.2012 17:05, Laurent Bigonville wrote:
> Package: systemd
> Version: 37-1
> Severity: grave
> Justification: renders package unusable

I'd argue it's sysvinit breaking systemd, but ok :-)

> 
> Hi,
> 
> /dev/initctl has moved to /run/initctl.
> 
> This means that shutdown and other commands are now failing with errors
> like "init: timeout opening/writing control channel /run/initctl"
> preventing the user to reboot or shutdown his machine.

I guess you don't have systemd-sysv installed, i.e. halt and shutdown
are provided by sysvinit?

As already mentioned, you can either use systemctl directly or create a
/dev/log → /run/log symlink manually.

Roger, would it be feasible if halt/shutdown could gracefully fallback
to /dev/log?

Cheers,
Michael


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

[signature.asc (application/pgp-signature, attachment)]

Message #25 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Roger Leigh <rleigh@codelibre.net>

To: Michael Biebl <biebl@debian.org>

Cc: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Tue, 31 Jan 2012 20:56:51 +0000

On Tue, Jan 31, 2012 at 07:46:03PM +0100, Michael Biebl wrote:
> I guess you don't have systemd-sysv installed, i.e. halt and shutdown
> are provided by sysvinit?
> 
> As already mentioned, you can either use systemctl directly or create a
> /dev/log → /run/log symlink manually.
> 
> Roger, would it be feasible if halt/shutdown could gracefully fallback
> to /dev/log?

I don't think /dev/log is at issue here.  I would hazard a guess
that the sysvinit telinit/init binary is being invoked with
"telinit (0|6)" and is trying to open /run/initctl, which does
not exist and/or has no listener due to /sbin/init not being
started when systemd is running.  Assuming that the systemd socket
listener is listening on /dev/initctl, it's simply not talking to
the service.

I would suggest that the systemd package switch to using the new
sysvinit path, since if the service is there for backward
compatibility with the sysvinit binaries installed on the system
(I might be wrong here), it should be compatible with those.

I'll look and see if we could add a fallback to sysvinit, but I
would suspect not--it's simply hardcoded as a
"#define INIT_FIFO ...." without any support for fallbacks.  While
we could add support, it gains us nothing from the sysvinit side.


Regards,
Roger

-- 
  .''`.  Roger Leigh
 : :' :  Debian GNU/Linux             http://people.debian.org/~rleigh/
 `. `'   Printing on GNU/Linux?       http://gutenprint.sourceforge.net/
   `-    GPG Public Key: 0x25BFB848   Please GPG sign your mail.

Message #30 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

Cc: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Tue, 31 Jan 2012 22:05:43 +0100

[Message part 1 (text/plain, inline)]

On 31.01.2012 21:56, Roger Leigh wrote:
> On Tue, Jan 31, 2012 at 07:46:03PM +0100, Michael Biebl wrote:
>> I guess you don't have systemd-sysv installed, i.e. halt and shutdown
>> are provided by sysvinit?
>>
>> As already mentioned, you can either use systemctl directly or create a
>> /dev/log → /run/log symlink manually.
>>
>> Roger, would it be feasible if halt/shutdown could gracefully fallback
>> to /dev/log?

Sorry for the brainfuck here. I meant /dev/initctl and /run/initctl,
obviously.

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

[signature.asc (application/pgp-signature, attachment)]

Message #35 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Adrien CLERC <bugs-debian@antipoul.fr>

To: 657990@bugs.debian.org

Subject: Single line patch

Date: Wed, 01 Feb 2012 07:33:25 +0100

Hi,

As Alexander mentionned, the real workaround is to replace the listening 
socket in /lib/systemd/system/systemd-initctl.socket with /run/initctl.
It's not only a workaround, it is the thing to do, because this is the 
exact role of this file: provide SysV init compatibility, in a flexible way.

Sorry if I don't provide a patch for this, but it's really a simple 
change to do.

Have a nice day!

Message #40 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Laurent Bigonville <bigon@debian.org>

To: 657990@bugs.debian.org

Subject: Re: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 1 Feb 2012 17:20:44 +0100

[Message part 1 (text/plain, inline)]

tag 657990 + patch
thanks

Hi,

Please see the attached patch.

Cheers

Laurent Bigonville

[657990.patch (text/x-patch, attachment)]

Message #47 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 01 Feb 2012 17:42:28 +0100

[Message part 1 (text/plain, inline)]

On 01.02.2012 17:20, Laurent Bigonville wrote:
> tag 657990 + patch
> thanks
> 
> Hi,
> 
> Please see the attached patch.
> 

> -Breaks: lvm2 (<< 2.02.84-1)
> +Breaks: lvm2 (<< 2.02.84-1), sysvinit (<< 2.88dsf-19)

If we provide a /dev/initctl → /run/initctl symlink, we don't need that
Breaks afaics and I think we should add such a a symlink anyway for
backwards compatibility.
I'd guess there is more code out there hardcoding the location of the
initctl socket (I could do an archive wide grep for "/dev/initctl" fwiw).

Michael


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

[signature.asc (application/pgp-signature, attachment)]

Message #52 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 01 Feb 2012 17:47:11 +0100

[Message part 1 (text/plain, inline)]

On 01.02.2012 17:42, Michael Biebl wrote:
> On 01.02.2012 17:20, Laurent Bigonville wrote:
>> tag 657990 + patch
>> thanks
>>
>> Hi,
>>
>> Please see the attached patch.
>>
> 
>> -Breaks: lvm2 (<< 2.02.84-1)
>> +Breaks: lvm2 (<< 2.02.84-1), sysvinit (<< 2.88dsf-19)
> 
> If we provide a /dev/initctl → /run/initctl symlink, we don't need that
> Breaks afaics and I think we should add such a a symlink anyway for
> backwards compatibility.

That's said, I'm not sure if socket activation works on symlinks.
systemd-initctl.service is started on-demand via socket activation.


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

[signature.asc (application/pgp-signature, attachment)]

Message #57 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Roger Leigh <rleigh@codelibre.net>

To: Michael Biebl <biebl@debian.org>

Cc: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 1 Feb 2012 16:57:52 +0000

On Wed, Feb 01, 2012 at 05:42:28PM +0100, Michael Biebl wrote:
> If we provide a /dev/initctl → /run/initctl symlink, we don't need that
> Breaks afaics and I think we should add such a a symlink anyway for
> backwards compatibility.
> I'd guess there is more code out there hardcoding the location of the
> initctl socket (I could do an archive wide grep for "/dev/initctl" fwiw).

They shouldn't, and I wouldn't worry too much about this possibility.
The protocol is private and undocumented, so there should be no
third-party users.  The location is hard-coded directly into the
binaries and is likely only intended to be used by sysvinit binaries.
systemd is an exception to this rule, and I would suspect the only
one.


Regards,
Roger

-- 
  .''`.  Roger Leigh
 : :' :  Debian GNU/Linux             http://people.debian.org/~rleigh/
 `. `'   Printing on GNU/Linux?       http://gutenprint.sourceforge.net/
   `-    GPG Public Key: 0x25BFB848   Please GPG sign your mail.

Message #62 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: 657990@bugs.debian.org

Cc: Laurent Bigonville <bigon@debian.org>, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 01 Feb 2012 19:40:00 +0100

[Message part 1 (text/plain, inline)]

Two more issues:

After installing the updated systemd-initctl.{service,socket}, they need
to be restarted in postinst. systemctl try-restart for the service and
restart for the socket is most likely what we want here.

On 01.02.2012 17:57, Roger Leigh wrote:
> On Wed, Feb 01, 2012 at 05:42:28PM +0100, Michael Biebl wrote:
>> If we provide a /dev/initctl → /run/initctl symlink, we don't need that
>> Breaks afaics and I think we should add such a a symlink anyway for
>> backwards compatibility.
>> I'd guess there is more code out there hardcoding the location of the
>> initctl socket (I could do an archive wide grep for "/dev/initctl" fwiw).
> 
> They shouldn't, and I wouldn't worry too much about this possibility.
> The protocol is private and undocumented, so there should be no
> third-party users.  The location is hard-coded directly into the
> binaries and is likely only intended to be used by sysvinit binaries.
> systemd is an exception to this rule, and I would suspect the only
> one.

If we don't provide backwards compatibility via a /dev/initctl symlink
in systemd, sytemd needs a versioned breaks against sysvinit.
On the other hand sysvinit needs to declare a versioned breaks against
systemd.
This means we have a kind of circular dep and the order in which the
packages are updated is not deterministic.

As the order of in which the postinst of sysvinit and systemd are run is
not deterministic either, they need to work in both cases.
But I *think* it should work.
If systemd.postinst is run first (and the try-restart has been added),
/run/initctl should exist and sysvinit.postinst does nothing.
If sysvinit.postint is run first, it will shuffle the initctl socket
around but systemd-initctl.socket will still listen on /dev/initctl.

That means the "init u" call in sysvinit.postinst will not work and we
try it for 7 times before we give up.
Later then systemd.postinst is run, which will restart the socket, so we
have a working /run/initctl, and in case the systemd-initctl service was
running, we restart it, too.



Michael


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

[signature.asc (application/pgp-signature, attachment)]

Message #67 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Michael Biebl <biebl@debian.org>

To: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 01 Feb 2012 21:44:58 +0100

[Message part 1 (text/plain, inline)]

On 01.02.2012 17:42, Michael Biebl wrote:
> I'd guess there is more code out there hardcoding the location of the
> initctl socket (I could do an archive wide grep for "/dev/initctl" fwiw).

FYI: http://people.debian.org/~biebl/results.dev-initctl



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

[signature.asc (application/pgp-signature, attachment)]

Message #72 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Roger Leigh <rleigh@codelibre.net>

To: Michael Biebl <biebl@debian.org>

Cc: Laurent Bigonville <bigon@debian.org>, 657990@bugs.debian.org, Roger Leigh <rleigh@debian.org>

Subject: Re: Bug#657990: systemd: /dev/initctl has moved to /run/initctl

Date: Wed, 1 Feb 2012 21:36:22 +0000

On Wed, Feb 01, 2012 at 09:44:58PM +0100, Michael Biebl wrote:
> On 01.02.2012 17:42, Michael Biebl wrote:
> > I'd guess there is more code out there hardcoding the location of the
> > initctl socket (I could do an archive wide grep for "/dev/initctl" fwiw).
> 
> FYI: http://people.debian.org/~biebl/results.dev-initctl

Many thanks.  Looking through the list, many are false positives:

bup: only in examples
[mindi-]busybox: just avoids lookups if present
cdebootstrap: cleans if present
debootstrap: changelog only
gradm2: in policy; not harmful
lvm2: part of list in unit tests
loop-aes-utils: implements lightweight initctl utility; does not
  appear to be built or installed; originated from util-linux-ng
nut: Only in FAQ/TODO docs
ppbuttonsd: Only in docs/comment.
pbuilder: Only in logs of chroot update, due to timeout in a chroot.
rpm: Only in changelogs
util-linux: In source; does not appear to be built or used.

Needs fixing:

systemd: Using old paths.

sysvinit: Unpatched.

upstart: Using control FIFO directly.  Not clear why it can't just
  use telinit; probably due to removing sysvinit with Conflicts.

libsepol-2: Checks for socket before running telinit.  Should
  probably use both old and new paths.  Or not bother checking;
  the worst case is it times out or fails.

manpages: These translations will be picked up by po4a.

So some definitely need fixing; others could have bugs filed
e.g. gradm2.


Regards,
Roger

-- 
  .''`.  Roger Leigh
 : :' :  Debian GNU/Linux             http://people.debian.org/~rleigh/
 `. `'   Printing on GNU/Linux?       http://gutenprint.sourceforge.net/
   `-    GPG Public Key: 0x25BFB848   Please GPG sign your mail.

Message #81 received at 657990@bugs.debian.org (full text, mbox, reply):

From: Lutz Reinhardt <lutz.reinhardt@tu-bs.de>

To: 657990@bugs.debian.org

Subject: It is back in systemd 43

Date: Sat, 24 Mar 2012 14:00:45 +0100

Hi,

I upgraded to systemd v43 and it is back. I had to do this again, to be
able to poweroff and reboot:
> Replace the FIFO-name (ListenFIFO) in systemd's socket file
> (/lib/systemd/system/systemd-initctl.socket) with /run/initctl.

mfg Lutz

Send a report that this bug log contains spam.