[gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Robin H. Johnson]

[-- Attachment #1: Type: text/plain, Size: 1648 bytes --]

I'm raising this as an extension of bug 253076, but also because I see
the potential for danger.

To date, for an init script that has baselayout2-specific behavior, we
have had some variant of [ -e /lib/librc.so ] in the init script.

On a multilib profile with no symlinks and a 64-bit userspace, the .so
file would be installed in /lib64/librc.so, and the check would
mistakenly have the wrong result. 

There's one fix that has started to turn up already, but I'm not sure if
it's going to be safe always: [ -f /etc/init.d/sysfs ]
This happens to work as openrc installs that init script.

As a long term solution, can we just consume an inode and have some file
like /etc/baselayout2? The file must reside on the / partition even when
the major trees /usr, /var, /tmp, /opt, /boot, /home, /dev, /root are
separate mountpoints.

Affected and broken profiles:
- amd64's 2005.0/no-symlink (it was very widely deployed. It's not
  supported or in the tree anymore, but there ARE still people using it)
- default/linux/sparc/2008.0 (unconfirmed)
- Any profile that sets SYMLINK_LIB="no" AND the userspace is 64-bit.
- Wouldn't surprise me if some of the N32 style mips stuff was broken
  too.

Affected, but not broken profiles:
These are multilib, but with the default as 32-bit causes the lib to be
in right place.
- arch/powerpc/ppc64/32ul
- hardened/linux/powerpc/ppc64/2008.0/32bit-userland
- Any other profile that inherits features/32bit-userland

-- 
Robin Hugh Johnson
Gentoo Linux Developer & Infra Guy
E-Mail     : robbat2@gentoo.org
GnuPG FP   : 11AC BA4F 4778 E3F6 E4ED  F38E B27B 944E 3488 4E85

[-- Attachment #2: Type: application/pgp-signature, Size: 329 bytes --]

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Friedrich Oslage]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Robin H. Johnson schrieb:
> I'm raising this as an extension of bug 253076, but also because I see
> the potential for danger.
> 
> To date, for an init script that has baselayout2-specific behavior, we
> have had some variant of [ -e /lib/librc.so ] in the init script.
> 
> On a multilib profile with no symlinks and a 64-bit userspace, the .so
> file would be installed in /lib64/librc.so, and the check would
> mistakenly have the wrong result. 
> 
> There's one fix that has started to turn up already, but I'm not sure if
> it's going to be safe always: [ -f /etc/init.d/sysfs ]
> This happens to work as openrc installs that init script.
> 
> As a long term solution, can we just consume an inode and have some file
> like /etc/baselayout2? The file must reside on the / partition even when
> the major trees /usr, /var, /tmp, /opt, /boot, /home, /dev, /root are
> separate mountpoints.

Using a file in /etc would break if someone would downgrade from
baselayout2/openrc to baselayout1.

Why not teach /sbin/runscript it's own version? With something like this
 we could also do stuff depending on a specific version of openrc:

- --- src/rc/runscript.c
+++ src/rc/runscript.c
@@ -1244,6 +1244,7 @@ runscript(int argc, char **argv)

        setenv("EINFO_LOG", service, 1);
        setenv("RC_SVCNAME", applet, 1);
+       setenv("OPENRC_VERSION", "0.4.2", 1);

        /* Set an env var so that we always know our pid regardless of
any
           subshells the init script may create so that our mark_service_*


and in the ebuild:

[ -n "${OPENRC_VERSION}" ]

> 
> Affected and broken profiles:
> - amd64's 2005.0/no-symlink (it was very widely deployed. It's not
>   supported or in the tree anymore, but there ARE still people using it)
> - default/linux/sparc/2008.0 (unconfirmed)

That one should be fine. It has a pure 32bit userland, no lib64 or
lib32, just plain lib as libdir.

> - Any profile that sets SYMLINK_LIB="no" AND the userspace is 64-bit.
> - Wouldn't surprise me if some of the N32 style mips stuff was broken
>   too.
> 
> Affected, but not broken profiles:
> These are multilib, but with the default as 32-bit causes the lib to be
> in right place.
> - arch/powerpc/ppc64/32ul
> - hardened/linux/powerpc/ppc64/2008.0/32bit-userland
> - Any other profile that inherits features/32bit-userland
> 

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAklybyAACgkQknxn9PmJ76VL5QCfXLgKLtrObX2FzbQ3XMHhAqZi
nT8An1zTWYJSmdRs30eLfDIy4LpqslRp
=80mR
-----END PGP SIGNATURE-----

[gentoo-dev] Re: Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Duncan]

Friedrich Oslage <bluebird@gentoo.org> posted 49726F21.8010004@gentoo.org,
excerpted below, on  Sun, 18 Jan 2009 00:52:01 +0100:

> Why not teach /sbin/runscript it's own version? With something like this
>  we could also do stuff depending on a specific version of openrc:

I happened to read the OP right before I headed in to work today, and 
thus had some time to think it over.  This was the exact same idea I came 
up with and was going to post it if someone hadn't, when I got back.  You 
beat me to it! =:^)

In fact, if Roy hasn't instituted something like this already upstream, 
I'd be surprised if he doesn't take the idea and run with it as well, as 
it's certainly convenient functionality to have and expose, and I can 
think of all sorts of reasons it might come in handy and no reason not to 
do it.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Peter Volkov]

[-- Attachment #1: Type: text/plain, Size: 1277 bytes --]

В Вск, 18/01/2009 в 00:52 +0100, Friedrich Oslage пишет:
> Robin H. Johnson schrieb:
> > I'm raising this as an extension of bug 253076, but also because I see
> > the potential for danger.

> > As a long term solution, can we just consume an inode and have some file
> > like /etc/baselayout2?

What package will install this file? Also is it possible to save openrc
${PVR} inside this file? Version is required to introduce workarounds in
reliable way for openrc versions which does not work with openvz startup
scripts. Also why not call this files /etc/*openrc*?

> > To date, for an init script that has baselayout2-specific behavior, we
> > have had some variant of [ -e /lib/librc.so ] in the init script.
> >
> Why not teach /sbin/runscript it's own version? 

In some cases this will break with openvz. openvz container (CT)
start/stop scripts need to check openrc existence inside container and
depending on that generate either openrc or baselayout-1 configuration
files. This start/stop scripts work in host node (HN) environment and
since it's possible to have different versions of everything (e.g. libc)
in HN and CT /sbin/runscript will not work in some cases and thus break
networking and other things for CT.

-- 
Peter.

[-- Attachment #2: Эта часть сообщения подписана цифровой подписью --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Santiago M. Mola]

On Sun, Jan 18, 2009 at 12:52 AM, Friedrich Oslage <bluebird@gentoo.org> wrote:
>
> Why not teach /sbin/runscript it's own version? With something like this
>  we could also do stuff depending on a specific version of openrc:
>

That would be a good solution since it'll allow us to check higher
versions for new features. Otherwise, it's very likely that we need to
solve the same problem again in the near future.

Regards,
-- 
Santiago M. Mola
Jabber ID: cooldwind@gmail.com

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Peter Volkov]

[-- Attachment #1: Type: text/plain, Size: 455 bytes --]

В Пнд, 19/01/2009 в 17:18 +0100, Santiago M. Mola пишет:
> On Sun, Jan 18, 2009 at 12:52 AM, Friedrich Oslage <bluebird@gentoo.org> wrote:
> > Why not teach /sbin/runscript it's own version? With something like this
> >  we could also do stuff depending on a specific version of openrc:
> 
> That would be a good solution since [[snip...]]

This solution does not work in some cases. See my previous mail to this
thread.

-- 
Peter.

[-- Attachment #2: Эта часть сообщения подписана цифровой подписью --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Donnie Berkholz]

[-- Attachment #1: Type: text/plain, Size: 532 bytes --]

On 15:07 Sat 17 Jan     , Robin H. Johnson wrote:
> As a long term solution, can we just consume an inode and have some file
> like /etc/baselayout2? The file must reside on the / partition even when
> the major trees /usr, /var, /tmp, /opt, /boot, /home, /dev, /root are
> separate mountpoints.

How about using the version in /etc/gentoo-release?

comet $ cat /etc/gentoo-release 
Gentoo Base System release 2.0.0

-- 
Thanks,
Donnie

Donnie Berkholz
Developer, Gentoo Linux
Blog: http://dberkholz.wordpress.com

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Robin H. Johnson]

[-- Attachment #1: Type: text/plain, Size: 732 bytes --]

On Sun, Jan 18, 2009 at 07:11:52PM -0800, Donnie Berkholz wrote:
> On 15:07 Sat 17 Jan     , Robin H. Johnson wrote:
> > As a long term solution, can we just consume an inode and have some file
> > like /etc/baselayout2? The file must reside on the / partition even when
> > the major trees /usr, /var, /tmp, /opt, /boot, /home, /dev, /root are
> > separate mountpoints.
> How about using the version in /etc/gentoo-release?
I should clarify that I want to check simply the existence of a file,
rather than parse out a version number or cause a fork to execute.

-- 
Robin Hugh Johnson
Gentoo Linux Developer & Infra Guy
E-Mail     : robbat2@gentoo.org
GnuPG FP   : 11AC BA4F 4778 E3F6 E4ED  F38E B27B 944E 3488 4E85

[-- Attachment #2: Type: application/pgp-signature, Size: 329 bytes --]

Re: [gentoo-dev] Detecting Baselayout2/openrc - no-symlink profiles leading to breakage

[Matthias Schwarzott]

On Sonntag, 18. Januar 2009, Robin H. Johnson wrote:
> I'm raising this as an extension of bug 253076, but also because I see
> the potential for danger.
>
> To date, for an init script that has baselayout2-specific behavior, we
> have had some variant of [ -e /lib/librc.so ] in the init script.
>
> On a multilib profile with no symlinks and a 64-bit userspace, the .so
> file would be installed in /lib64/librc.so, and the check would
> mistakenly have the wrong result.
>
> There's one fix that has started to turn up already, but I'm not sure if
> it's going to be safe always: [ -f /etc/init.d/sysfs ]
> This happens to work as openrc installs that init script.
>

I changed udev to only check for /etc/init.d/sysfs. See Bug #252493.
The only place where librc checking is kept is only run on older 
openrc-versions which are no longer available via ebuild.

I hope this will make udev work on any system regardless of how /lib and stuff 
is linked. (For further analysis perhaps someone on a multilib profile can 
check where udev still has /lib/xxx hardcoded instead of /lib64/xxx or 
similar).

Should I nevertheless add such a blocker to udev or will that make update 
unnecessary complicated?
RDEPEND="!<sys-apps/openrc-0.4.0"

Regards
Matthias