[Buildroot] Has anybody got systemd to work?

[Buildroot] Has anybody got systemd to work?

[Dmitry Golubovsky]

Hi,

I am trying to boot a root FS built with systemd acting as init.

Looks like it fails in src/mount-setup.c, function
mount-group-controllers, ca. line 193.

I get an error message

Failed to parse /proc/cgroups.

as a result of

if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2)

failing to parse /proc/cgroups.

I changed %ms to %s and added explicit malloc right before (only in
that file, mount-setup.c). I was able to get a bit further, but failed
on another instance of fscanf with %ms (anf there seem to be more
files with %ms

Is %ms format (pre-malloc a value buffer) or %as format supported in
uClibc in principle, or can it be enabled via specific configutration,
or expected in future versions?

I use uClibc 0.9.33

Has anybody seen this?

Thanks.

-- 
Dmitry Golubovsky

Anywhere on the Web

[Buildroot] Has anybody got systemd to work?

[Attila Kinali]

On Sun, 17 Jun 2012 22:45:20 -0400
Dmitry Golubovsky <golubovsky@gmail.com> wrote:

> 
> I am trying to boot a root FS built with systemd acting as init.

Out of curiosity: Is there any reason why you'd want to use
systemd in an embedded system? What does it provide that a normal init
cannot?

			Attila Kinali

-- 
It is upon moral qualities that a society is ultimately founded. All 
the prosperity and technological sophistication in the world is of no 
use without that foundation.
                 -- Miss Matheson, The Diamond Age, Neil Stephenson

[Buildroot] Has anybody got systemd to work?

[Dmitry Golubovsky]

Attila,

On Mon, Jun 18, 2012 at 3:15 AM, Attila Kinali <attila@kinali.ch> wrote:

>> I am trying to boot a root FS built with systemd acting as init.
>
> Out of curiosity: Is there any reason why you'd want to use
> systemd in an embedded system? What does it provide that a normal init
> cannot?

I am not targeting an embedded system; rather portable to desktop
range (laptops, small desktops, and those new pocket-size boxes that
can be hooked to a TV/monitor and worked with - I need to build a
compact initial root FS).

Systemd provides asynchronous dependency-based booting (something I
liked in Android init), and some level of containerization (LXC would
be too much for my purpose, but I'd rather use systemd than some
clumsy shell scripts which I tried to achieve the same with).

I use Buildroot because it is a mature toolkit being many years in
active use, with convenient menu configuration.

OTOH I see other developers suggesting more systemd patches (like one
by Maxime Ripard [1] that showed up in the list recently)

Maxime: do you use an alternative toolchain with systemd?

Per my postings some time ago where I suggested adding PAM
capabilities to systemd: I was able to compile systemd with PAM but
was unable to test it because of the parsing problem I ran into (see
the OP).

Thanks.

[1]  http://lists.busybox.net/pipermail/buildroot/2012-June/054903.html

-- 
Dmitry Golubovsky

Anywhere on the Web

[Buildroot] Has anybody got systemd to work?

[Maxime Ripard]

Hi Dmitry,

I was just replying to your previous mail.

Le 18/06/2012 14:46, Dmitry Golubovsky a ?crit :
> Maxime: do you use an alternative toolchain with systemd?

After a bit of testing, it seems indeed that the bug is uclibc-only, and
didn't encountered it as I do most of my testing using glibc.

Anyway, if it is only the %ms thing that is breaking systemd on uclibc,
it shouldn't be such a big problem. Quickly grepping the systemd source
code returns 17 occurences of such case, so a patch is definitely doable.

I won't have much time, do you mind doing so and sending us the patch so
that it could get included ?

Maxime

-- 
Maxime Ripard, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

[Buildroot] Has anybody got systemd to work?

[Maxime Ripard]

Le 18/06/2012 14:53, Maxime Ripard a ?crit :
> Hi Dmitry,
> 
> I was just replying to your previous mail.
> 
> Le 18/06/2012 14:46, Dmitry Golubovsky a ?crit :
>> Maxime: do you use an alternative toolchain with systemd?
> 
> After a bit of testing, it seems indeed that the bug is uclibc-only, and
> didn't encountered it as I do most of my testing using glibc.
> 
> Anyway, if it is only the %ms thing that is breaking systemd on uclibc,
> it shouldn't be such a big problem. Quickly grepping the systemd source
> code returns 17 occurences of such case, so a patch is definitely doable.
> 
> I won't have much time, do you mind doing so and sending us the patch so
> that it could get included ?

Also, it seems that the %m modifier support has been merged 3 weeks ago
in uclibc:

http://git.uclibc.org/uClibc/commit/libc/stdio/_scanf.c?id=8cfb43de636faa401634340d1a18404844f9ba5a

Looks like we will definitely need a patch :)

-- 
Maxime Ripard, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

[Buildroot] Has anybody got systemd to work?

[Dmitry Golubovsky]

Maxime,

On Mon, Jun 18, 2012 at 8:53 AM, Maxime Ripard
<maxime.ripard@free-electrons.com> wrote:

> Anyway, if it is only the %ms thing that is breaking systemd on uclibc,

This has been found so far ;)

> it shouldn't be such a big problem. Quickly grepping the systemd source
> code returns 17 occurences of such case, so a patch is definitely doable.

Yes, I got the same.

> I won't have much time, do you mind doing so and sending us the patch so
> that it could get included ?

Yes, I am going to work on this, and will provide the patch once I have it.

I cannot promise exactly when ;)

Thanks.

-- 
Dmitry Golubovsky

Anywhere on the Web

[Buildroot] Has anybody got systemd to work?

[Attila Kinali]

On Mon, 18 Jun 2012 14:53:54 +0200
Maxime Ripard <maxime.ripard@free-electrons.com> wrote:

> Le 18/06/2012 14:46, Dmitry Golubovsky a ?crit :
> > Maxime: do you use an alternative toolchain with systemd?
> 
> After a bit of testing, it seems indeed that the bug is uclibc-only, and
> didn't encountered it as I do most of my testing using glibc.

Actually no: The bug is in systemd.
The m modifier is a gnu extension of the scanf format and is not
in the C99 standard. Because systemd silently assume that you are building
in a gnu environment (ie with glibc gcc etc) it fails at this point.

			Attila Kinali
-- 
Why does it take years to find the answers to
the questions one should have asked long ago?

[Buildroot] Has anybody got systemd to work?

[Dmitry Golubovsky]

Hi,

On Mon, Jun 18, 2012 at 9:21 AM, Attila Kinali <attila@kinali.ch> wrote:

> Actually no: The bug is in systemd.
> The m modifier is a gnu extension of the scanf format and is not
> in the C99 standard. Because systemd silently assume that you are building
> in a gnu environment (ie with glibc gcc etc) it fails at this point.

From the portability standpoint, it is..

They likely did not provide testing and fallback for %ms support in
their autoconf scripts.

I'll try to look into their source tree as well to see if this has
been addressed there, and maybe will try to propose a fix for
upstream, however local patching might be a faster solution
(unfortunately).

Thanks.

-- 
Dmitry Golubovsky

Anywhere on the Web

[Buildroot] Has anybody got systemd to work?

[Dmitry Golubovsky]

Hi,

On Mon, Jun 18, 2012 at 9:32 AM, Dmitry Golubovsky <golubovsky@gmail.com> wrote:

>
> I'll try to look into their source tree as well to see if this has
> been addressed there, and maybe will try to propose a fix for
> upstream, however local patching might be a faster solution
> (unfortunately).

From the quick look at the repo [1] it seems like after v44 they
completely reorganized the sources (and integrated udev): the
offending file mount-setup.c is now in src/core (but %ms is still
there).

Why was version 44 selected for Buildroot rather than any later
version - just to avoid conflict with already included udev?

Thanks.

[1] http://cgit.freedesktop.org/systemd/systemd/tree/src?id=v185 (v185
is the last recorded tag)

-- 
Dmitry Golubovsky

Anywhere on the Web

[Buildroot] Has anybody got systemd to work?

[Thomas Petazzoni]

Le Mon, 18 Jun 2012 14:59:36 -0400,
Dmitry Golubovsky <golubovsky@gmail.com> a ?crit :

> Why was version 44 selected for Buildroot rather than any later
> version - just to avoid conflict with already included udev?

I guess just because Maxime started the work of packaging systemd into
Buildroot a while ago, and at the time, version 44 was the latest one.
It would of course be good to bump to the newest version.

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com