Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

On Fri, Feb 24, 2012 at 08:15:09PM +0100, Sven Joachim wrote:
> Short summary for readers new to the bug: boot hangs with i386 systemd
> and an x86_64 kernel.

Care to explain why it hangs and not dies?

Bastian

--
Conquest is easy. Control is not.
-- Kirk, "Mirror, Mirror", stardate unknown



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120224194425.GB11361@wavehammer.waldi.eu.org">ht tp://lists.debian.org/20120224194425.GB11361@wavehammer.waldi.eu.org

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

found 633423 linux-2.6/3.2.6-1
tags 633423 + upstream
# broken interface introduced in v2.6.17-rc1~444
# amd64 kernel image on i386 introduced in 2.6.18.dfsg.1-10
found 633423 linux-2.6/2.6.18.dfsg.1-10
forwarded 633423 http://thread.gmane.org/gmane.linux.kernel/1256405/focus=1256966
quit

Sven Joachim wrote:

> reassign 633423 linux-2.6
> retitle 633423 autofs4 interface is broken between x86 and x86_64
> affects 633423 systemd
[...]
> http://thread.gmane.org/gmane.linux.kernel/1255125
> http://thread.gmane.org/gmane.linux.kernel/1256405
>
> I will keep an eye on this bug and send an update when a fix appears in
> Linus' tree that I can test.

Thanks.



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120224194600.GA14534@burratino">http://lists.debian.org/20120224194600.GA14534@burratino

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

On Fri, Feb 24, 2012 at 08:15:09PM +0100, Sven Joachim wrote:
[...]
> Thankfully this problem has now been communicated to Linus himself, and
> he agreed that it should be fixed in the kernel.

Of course it should. A shame that some kernel maintainers think
they don't have to care about 32-bit compat.

(But why hasn't this been spotted on sparc or powerpc already,
where 32-bit userland on 64-bit kernel is the normal case?)

> See the following long > threads on the LKML:
>
> http://thread.gmane.org/gmane.linux.kernel/1255125
> http://thread.gmane.org/gmane.linux.kernel/1256405
>
> I will keep an eye on this bug and send an update when a fix appears in
> Linus' tree that I can test.

Thanks.

Ben.

--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus


--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120224195005.GI12704@decadent.org.uk">http://lists.debian.org/20120224195005.GI12704@decadent.org.uk

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

Ben Hutchings wrote:

> (But why hasn't this been spotted on sparc or powerpc already,
> where 32-bit userland on 64-bit kernel is the normal case?)

__u64 is 64-bit aligned on those arches in 32-bit mode.

http://thread.gmane.org/gmane.linux.kernel/1255890



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120224195439.GB14534@burratino">http://lists.debian.org/20120224195439.GB14534@burratino

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

On Fri, Feb 24, 2012 at 07:50:05PM +0000, Ben Hutchings wrote:
> On Fri, Feb 24, 2012 at 08:15:09PM +0100, Sven Joachim wrote:
> [...]
> > Thankfully this problem has now been communicated to Linus himself, and
> > he agreed that it should be fixed in the kernel.
> Of course it should. A shame that some kernel maintainers think
> they don't have to care about 32-bit compat.

They tried, but failed.

> (But why hasn't this been spotted on sparc or powerpc already,
> where 32-bit userland on 64-bit kernel is the normal case?)

Sane allignment rules. Only x86-32 and m68k aligns 8-byte values on
4-byte.

Bastian

--
Captain's Log, star date 21:34.5...



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120224200233.GA13036@wavehammer.waldi.eu.org">ht tp://lists.debian.org/20120224200233.GA13036@wavehammer.waldi.eu.org

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

On Fri, Feb 24, 2012 at 01:54:39PM -0600, Jonathan Nieder wrote:
> Ben Hutchings wrote:
>
> > (But why hasn't this been spotted on sparc or powerpc already,
> > where 32-bit userland on 64-bit kernel is the normal case?)
>
> __u64 is 64-bit aligned on those arches in 32-bit mode.
>
> http://thread.gmane.org/gmane.linux.kernel/1255890

OK, so it was all supposed to be non-word-size-dependent and someone
messed it up with padding.

The same thing happened with some newer ethtool command structures.
:-/ Would be nice if we had some systematic testing to detect this
before such definitions escape into a release.

Ben.

--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120224200602.GK12704@decadent.org.uk">http://lists.debian.org/20120224200602.GK12704@decadent.org.uk

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

On 2012-02-24 20:44 +0100, Bastian Blank wrote:

> On Fri, Feb 24, 2012 at 08:15:09PM +0100, Sven Joachim wrote:
>> Short summary for readers new to the bug: boot hangs with i386 systemd
>> and an x86_64 kernel.
>
> Care to explain why it hangs and not dies?

Here is Thomas Meyer's explanation from [1]:

,----
| the size of autofs_v5_packet_union is 300 bytes on x86 and 304 bytes on x86_64 kernels.
| when running systemd (x86) on an x86_64 kernel this leads to a hang in automount_fd_event->loop_read
| as a second fd_event is trigged for the remaining 4 bytes. but the loop_read tries to read
| 300 bytes again.
`----

Sven


1. http://lists.freedesktop.org/archives/systemd-devel/2011-September/003396.html



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 87haygx9qc.fsf@turtle.gmx.de">http://lists.debian.org/87haygx9qc.fsf@turtle.gmx.de

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

tags 633423 + fixed-upstream patch
thanks

On 2012-02-24 20:15 +0100, Sven Joachim wrote:

> Short summary for readers new to the bug: boot hangs with i386 systemd
> and an x86_64 kernel.
>
> On 2011-10-15 21:51 +0200, Sven Joachim wrote:
>
>> It seems that somebody who is both smarter and more pertinacious than
>> myself has tried to use a 32-bit systemd under 64-bit kernel and
>> experienced the hangs, see the thread starting at [1].
>>
>> Thomas Meyer proposed a patch in [2], but it was rejected, arguing that
>> the incompatibility should be fixed in the kernel. Of course the autofs
>> maintainer disagrees and wants workarounds in userspace, so we're stuck
>> in a deadlock. :-/
>>
>> 1. http://lists.freedesktop.org/archives/systemd-devel/2011-September/003338.html
>> 2. http://lists.freedesktop.org/archives/systemd-devel/2011-September/003396.html
>
> Thankfully this problem has now been communicated to Linus himself, and
> he agreed that it should be fixed in the kernel. See the following long
> threads on the LKML:
>
> http://thread.gmane.org/gmane.linux.kernel/1255125
> http://thread.gmane.org/gmane.linux.kernel/1256405
>
> I will keep an eye on this bug and send an update when a fix appears in
> Linus' tree that I can test.

3.3-rc5 works for me, as does 3.2.7 with the following changes
cherry-picked:

a32744d4abae (autofs: work around unhappy compat problem on x86-64)
3c761ea05a89 (Fix autofs compile without CONFIG_COMPAT)
048cd4e51d24 (compat: fix compile breakage on s390)

The latter two are necessary to fix FTBFS problems on other
architectures introduced by a32744d4abae.

Cheers,
Sven



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 87wr763dtb.fsf@turtle.gmx.de">http://lists.debian.org/87wr763dtb.fsf@turtle.gmx.de

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

tags 633423 = patch upstream
quit

Sven Joachim wrote:

> 3.3-rc5 works for me, as does 3.2.7 with the following changes
> cherry-picked:
>
> a32744d4abae (autofs: work around unhappy compat problem on x86-64)
> 3c761ea05a89 (Fix autofs compile without CONFIG_COMPAT)
> 048cd4e51d24 (compat: fix compile breakage on s390)

Thanks! I've passed this information to Greg (though I'm a bit
nervous since it hasn't shown up in mail archives yet). Hopefully we
can see this fix in a 3.2.y-stable kernel soon.



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120229193253.GD5828@burratino">http://lists.debian.org/20120229193253.GD5828@burratino

Bug#633423: Problem with autofs 64-bit kernel/32-bit userspace compatibility
 

Hi,

I'm having the same problem (autofs hangs when accessing an auto mount point)
with recent kernels, so I wonder if this is really fixed upstream or if there
is another cause.

Affected kernel: vanilla 3.2.13 and 3.3.1 (both amd64)
Working kernel: 3.1.10 (amd64)

Userspace: autofs5 5.0.4-3.2+b1 (i386)



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20120403140620.48f3d8ee@hr1">http://lists.debian.org/20120403140620.48f3d8ee@hr1