Re: [PATCH v6 0/2] rbind mounts for /dev and /sys

[Henning Schild <henning.schild@siemens.com>]

Am Tue, 27 Nov 2018 14:48:37 +0000
schrieb "Hombourger, Cedric" <Cedric_Hombourger@mentor.com>:

> Hi Maxim,
> 
> We may need to revert Henning's suggestion to rbind /sys and
> mount /sys as a regular mount (the chroot may not need all of the
> sub-mounts from /sys) What do people think?

I already expected the umount not being able to clean up, but Cedric
tested that and made the umount recursive.
I guess it would be a good idea to understand what is going on, and why
those mounts do not get cleaned up. Maybe we still have a problem in
the umounting logic and mounting less will not fix but just hide it.

But we can also revert to mounting as little as needed and accept new
mounts on demand, just like we did for sys. However the demand for
something under sys would not be a big surprise and we might be talking
about that topic again soon ... or never. I do not know.

Henning

> Cedric
> 
> -----Original Message-----
> From: Maxim Yu. Osipov [mailto:mosipov@ilbers.de] 
> Sent: Tuesday, November 27, 2018 5:45 PM
> To: Hombourger, Cedric <Cedric_Hombourger@mentor.com>;
> isar-users@googlegroups.com Subject: Re: [PATCH v6 0/2] rbind mounts
> for /dev and /sys
> 
> Hi Cedric,
> 
> After applying these patches I've recognized that after running
> ci_build.sh script, the number of mounts significantly increases.
> 
> On my PC it looks like that
> 
> $ mount | wc -l
> 34
> isar$ ./scripts/ci_build.sh -q -f
> ...
> $ mount | wc -l
> 100
> 
> If I revert these patches the number of mount points stays the same
> after running ci_build.sh script.
> 
> This behavior significantly affects CI - builds start to fail. This
> is a blocking point for CI and if we don't find a quick solution I
> will have to revert these patches in 'next'.
> 
> Thanks,
> Maxim.
> 
> On 11/20/18 3:40 PM, Cedric Hombourger wrote:
> > Some of the packages I have been rebuilding within Isar
> > require /sys mounted. For instance, golang-google-grpc runs some
> > package tests after it builds itself and reads properties
> > from /sys/class/dmi.
> > 
> > The initial version of this patch series only added a mount
> > of /sys. Henning then suggested that we follow Gentoo's guidelines
> > for chroots. As such, this patch series is now mounting /dev
> > and /sys as rbind mounts but leaves /proc as a regular procfs mount
> > (like Gentoo does).
> > 
> > ChangeLog:
> > 
> > v6: remove unnecessary mkdir for /sys
> > v5: cover letter added as suggested by Jan (private e-mail)
> > v4: limit rbind mounts to /dev and /sys, leave /proc as a regular 
> > mount
> > v3: rbind mounts for /dev, /proc and /sys, add explicit umount calls
> > v2: create /sys directory
> > v1: initial submission, simply mount /sys
> > 
> > Cedric Hombourger (2):
> >    buildchroot+bootstrap: rbind mount for /dev
> >    buildchroot: add sysfs mount since expected by some packages
> > 
> >   meta/classes/buildchroot.bbclass                    | 5 ++++-
> >   meta/classes/image.bbclass                          | 3 ++-
> >   meta/classes/isar-bootstrap-helper.bbclass          | 6 +++++-
> >   meta/classes/isar-image.bbclass                     | 3 ++-
> >   meta/recipes-core/isar-bootstrap/isar-bootstrap.inc | 8 ++++++--
> >   5 files changed, 19 insertions(+), 6 deletions(-)
> >   
> 
> 
> --
> Maxim Osipov
> ilbers GmbH
> Maria-Merian-Str. 8
> 85521 Ottobrunn
> Germany
> +49 (151) 6517 6917
> mosipov@ilbers.de
> http://ilbers.de/
> Commercial register Munich, HRB 214197
> General Manager: Baurzhan Ismagulov
>