[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Yann E. MORIN]

This toolchain uses glibc-2.18, which is tool old to provide O_TMPFILE,
which was added only in glibc-2.19.

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
---
 package/systemd/Config.in | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/package/systemd/Config.in b/package/systemd/Config.in
index 4a2aa33..48e6c45 100644
--- a/package/systemd/Config.in
+++ b/package/systemd/Config.in
@@ -145,6 +145,8 @@ config BR2_PACKAGE_SYSTEMD_IMPORTD
 	bool "enable import daemon"
 	depends on BR2_PACKAGE_LIBGPG_ERROR_ARCH_SUPPORTS # libgcrypt
 	depends on BR2_TOOLCHAIN_HEADERS_AT_LEAST_3_11 # O_TMPFILE
+	# This toolchain has glibc-2.18, too old to provide O_TMPFILE
+	depends on !BR2_TOOLCHAIN_EXTERNAL_CODESOURCERY_ARM
 	select BR2_PACKAGE_LIBCURL
 	select BR2_PACKAGE_LIBGCRYPT
 	select BR2_PACKAGE_BZIP2
@@ -160,8 +162,12 @@ config BR2_PACKAGE_SYSTEMD_IMPORTD
 
 comment "import daemon needs a toolchain w/ headers >= 3.11"
 	depends on BR2_PACKAGE_LIBGPG_ERROR_ARCH_SUPPORTS
+	depends on !BR2_TOOLCHAIN_EXTERNAL_CODESOURCERY_ARM
 	depends on !BR2_TOOLCHAIN_HEADERS_AT_LEAST_3_11
 
+comment "import daemon broken on Sourcery CodeBench ARM 2014.05"
+	depends on !BR2_TOOLCHAIN_EXTERNAL_CODESOURCERY_ARM
+
 config BR2_PACKAGE_SYSTEMD_KDBUS
 	bool "enable kdbus support"
 	help
-- 
2.7.4

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Thomas Petazzoni]

Hello,

On Sun, 10 Jul 2016 23:56:21 +0200, Yann E. MORIN wrote:
> This toolchain uses glibc-2.18, which is tool old to provide O_TMPFILE,
> which was added only in glibc-2.19.
> 
> Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
> ---
>  package/systemd/Config.in | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/package/systemd/Config.in b/package/systemd/Config.in
> index 4a2aa33..48e6c45 100644
> --- a/package/systemd/Config.in
> +++ b/package/systemd/Config.in
> @@ -145,6 +145,8 @@ config BR2_PACKAGE_SYSTEMD_IMPORTD
>  	bool "enable import daemon"
>  	depends on BR2_PACKAGE_LIBGPG_ERROR_ARCH_SUPPORTS # libgcrypt
>  	depends on BR2_TOOLCHAIN_HEADERS_AT_LEAST_3_11 # O_TMPFILE
> +	# This toolchain has glibc-2.18, too old to provide O_TMPFILE
> +	depends on !BR2_TOOLCHAIN_EXTERNAL_CODESOURCERY_ARM

So is the problem a kernel headers problem, or a glibc version problem ?

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Yann E. MORIN]

Thomas, All,

On 2016-07-11 11:02 +0200, Thomas Petazzoni spake thusly:
> On Sun, 10 Jul 2016 23:56:21 +0200, Yann E. MORIN wrote:
> > This toolchain uses glibc-2.18, which is tool old to provide O_TMPFILE,
> > which was added only in glibc-2.19.
> > 
> > Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
> > ---
> >  package/systemd/Config.in | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/package/systemd/Config.in b/package/systemd/Config.in
> > index 4a2aa33..48e6c45 100644
> > --- a/package/systemd/Config.in
> > +++ b/package/systemd/Config.in
> > @@ -145,6 +145,8 @@ config BR2_PACKAGE_SYSTEMD_IMPORTD
> >  	bool "enable import daemon"
> >  	depends on BR2_PACKAGE_LIBGPG_ERROR_ARCH_SUPPORTS # libgcrypt
> >  	depends on BR2_TOOLCHAIN_HEADERS_AT_LEAST_3_11 # O_TMPFILE
> > +	# This toolchain has glibc-2.18, too old to provide O_TMPFILE
> > +	depends on !BR2_TOOLCHAIN_EXTERNAL_CODESOURCERY_ARM
> 
> So is the problem a kernel headers problem, or a glibc version problem ?

Both. O_TMPFILE was introduced in Linux 3.11, so we need at least those
headers. Also, glibc-2.18 did not provide it in fcntl.h; it only
appeared in glibc-2.19.

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Thomas Petazzoni]

Hello,

On Mon, 11 Jul 2016 19:16:30 +0200, Yann E. MORIN wrote:

> > So is the problem a kernel headers problem, or a glibc version problem ?  
> 
> Both. O_TMPFILE was introduced in Linux 3.11, so we need at least those
> headers. Also, glibc-2.18 did not provide it in fcntl.h; it only
> appeared in glibc-2.19.

I'm still not sure to understand why both are needed.

In any case, this fix is not correct, since the issue appears with
other toolchains as well:

 http://autobuild.buildroot.net/results/e6f/e6f8189029487155a784b4e664178adfd92f8884/build-end.log

 This one is an ARM toolchain, that uses 3.12 kernel headers, so most
 likely it's a glibc version issue.

 http://autobuild.buildroot.net/results/339/33912eebb5f4ee3f73be078ba2035f48d76a6a46/build-end.log

 This one is a PowerPC toolchain.

I see two possibilities only:

 * Introduce some glibc version dependency config knobs.

 * Add a workaround in the systemd code to define O_TMPFILE to the
   proper value when not provided by the C library.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Yann E. MORIN]

Thomas, All,

On 2016-08-27 23:32 +0200, Thomas Petazzoni spake thusly:
> On Mon, 11 Jul 2016 19:16:30 +0200, Yann E. MORIN wrote:
> > > So is the problem a kernel headers problem, or a glibc version problem ?  
> > Both. O_TMPFILE was introduced in Linux 3.11, so we need at least those
> > headers. Also, glibc-2.18 did not provide it in fcntl.h; it only
> > appeared in glibc-2.19.
> 
> I'm still not sure to understand why both are needed.

Well, indeed, both are not needed. glibc-2.19 is enough, as it does
provide it if not defined by the kernel headers.

And if the kernel headers provide it but not glibc, then it is not
available. (Weird).

> In any case, this fix is not correct, since the issue appears with
> other toolchains as well:
> 
>  http://autobuild.buildroot.net/results/e6f/e6f8189029487155a784b4e664178adfd92f8884/build-end.log
> 
>  This one is an ARM toolchain, that uses 3.12 kernel headers, so most
>  likely it's a glibc version issue.

glibc-2.18, so too old.

>  http://autobuild.buildroot.net/results/339/33912eebb5f4ee3f73be078ba2035f48d76a6a46/build-end.log
> 
>  This one is a PowerPC toolchain.

glibc-2.18 too.

> I see two possibilities only:
> 
>  * Introduce some glibc version dependency config knobs.

That's unfortunately the only sane solution, I'm araid.

But that just for systemd is big a gun... :-/

>  * Add a workaround in the systemd code to define O_TMPFILE to the
>    proper value when not provided by the C library.

At the same time, I'm not really happy with patching systemd...

O_TMPFILE is not the same value on all archs: generic, alpha, parisc and
sparc.

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Thomas Petazzoni]

Hello,

On Sat, 27 Aug 2016 23:48:49 +0200, Yann E. MORIN wrote:

> >  * Introduce some glibc version dependency config knobs.  
> 
> That's unfortunately the only sane solution, I'm araid.
> 
> But that just for systemd is big a gun... :-/

Yes, it sucks :-/

> >  * Add a workaround in the systemd code to define O_TMPFILE to the
> >    proper value when not provided by the C library.  
> 
> At the same time, I'm not really happy with patching systemd...
> 
> O_TMPFILE is not the same value on all archs: generic, alpha, parisc and
> sparc.

I must say I still don't understand the overlap of the kernel headers
and glibc here. 

In the kernel, we have:

include/uapi/asm-generic/fcntl.h:#define __O_TMPFILE    020000000

In glibc, we have:

sysdeps/unix/sysv/linux/bits/fcntl-linux.h:# define __O_TMPFILE   (020000000 | __O_DIRECTORY)

Which is not even the same value.

Same for SPARC.

I'd still say that patching systemd is the easiest solution, and we
will be able to remove this patch once we get rid of those too old
toolchains (which we will have to do at some point, when they become
too annoying / irrelevant).

Thanks,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[Buildroot] [PATCH] package/systemd: importd broken on Sourcery CodeBench ARM 2014.05

[Yann E. MORIN]

Thomas, All,

On 2016-08-28 09:29 +0200, Thomas Petazzoni spake thusly:
> > >  * Add a workaround in the systemd code to define O_TMPFILE to the
> > >    proper value when not provided by the C library.  
> > 
> > At the same time, I'm not really happy with patching systemd...
> > 
> > O_TMPFILE is not the same value on all archs: generic, alpha, parisc and
> > sparc.
> 
> I must say I still don't understand the overlap of the kernel headers
> and glibc here. 
> 
> In the kernel, we have:
> 
> include/uapi/asm-generic/fcntl.h:#define __O_TMPFILE    020000000
> 
> In glibc, we have:
> 
> sysdeps/unix/sysv/linux/bits/fcntl-linux.h:# define __O_TMPFILE   (020000000 | __O_DIRECTORY)
> 
> Which is not even the same value.

Ah, I got the answer to that one! :-)

O_TMPFILE is defined as such to be sure to break at runtime on old
kernels that do not have O_TMPFILE, becasue open() silently ignores
unknown flags. So, a userland app may request O_TMPFILE but silently not
get it:

    https://lwn.net/Articles/558940/

> I'd still say that patching systemd is the easiest solution, and we
> will be able to remove this patch once we get rid of those too old
> toolchains (which we will have to do at some point, when they become
> too annoying / irrelevant).

Fact is, upstream has a missing.h header that defines O_TMPFILE if not
already defined.

But they do it only for __i386__ and __x86_64__:

    https://github.com/systemd/systemd/blob/master/src/basic/missing.h#L540

So the fix will be two-fold:

  - use missing.h from importd, 

  - enhance missing.h to also define O_TMPFILE for other archs.

I'll be doing that shortly (first already done).

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'