mock: chroot proc busy when unmounting

[Alan Franzoni]

Hello, I'm experiencing a rather strange error with mock. This happens
almost every time, hence I'd like to find out what's happening.

Build host machine: Fedora 12, 64 bit
Build target: Centos 5 64 bit, using some additional repos in the
config, otherwise the config is the stock epel-5-x86_64.cfg
Mock version: 1.0.2
Running mock as root.

After building (either with success or without) I get this problem:

DEBUG: Executing command: umount -n
/var/lib/mock/rhel5-unstable-x86_64/root/proc
DEBUG: umount: /var/lib/mock/rhel5-unstable-x86_64/root/proc: device is
busy.
DEBUG:         (In some cases useful info about processes that use
DEBUG:          the device is found by lsof(8) or fuser(1))

I tried modifying mock's backend.py in order to run lsof before and
after running umount and grepping for proc, but without success - there
seems to be no open file there. I tried running orphanskill before
umounting, I tried sleeping before umount, but nothing changed.

While that error has no great impact, that one and some subsequent
(involving some failing chattr calls) really get my build logs messed up
and hard to crawl when something goes wrong.

Has anybody had similar issues? Can I provide you with any other useful
debugging data?

[Alan Franzoni]

Hello,
I tried catting /proc/mounts as well and I discovered this nested mount
under proc in the chroot:

none on /var/lib/mock/rhel5-unstable-x86_64/root/proc/sys/fs/binfmt_misc
type binfmt_misc (rw)

Adding it to umountCmds fixed my issue.

This doesn't seem to get mounted explicitly by mock, hence I think it's
something happening inside the chroot that triggers its mount? I'm not
really an expert in rhel/fedora internals, any clue?

Alan Franzoni

[Clark Williams]

On Wed, 24 Feb 2010 11:15:05 +0100
Alan Franzoni  wrote:
> Hello,
> I tried catting /proc/mounts as well and I discovered this nested mount
> under proc in the chroot:
>
> none on /var/lib/mock/rhel5-unstable-x86_64/root/proc/sys/fs/binfmt_misc
> type binfmt_misc (rw)
>
> Adding it to umountCmds fixed my issue.
>
> This doesn't seem to get mounted explicitly by mock, hence I think it's
> something happening inside the chroot that triggers its mount? I'm not
> really an expert in rhel/fedora internals, any clue?
>
> Alan Franzoni

Alan,

The only things we mount from mock are: /proc, /sys, /dev/pts
and /dev/shm.

What package(s) are you building?

Clark

[Alan Franzoni]

> Alan,
>
> The only things we mount from mock are: /proc, /sys, /dev/pts
> and /dev/shm.

I have seen that in mountCmds and I'd tell you're right. Hence I guess
they get mounted by some other means, but i don't know why and how.
> What package(s) are you building?

It happens to me all the time with all kind of packages. Most of them
are just jars with some config files that are actually built by a
separate machine and are just rpm-packaged via mock, others are python
projects which are treated in a similar fashion; we never build anything
too strictly system or kernel related. The chroot is a centos5-64 bit one.

I'll try reproducing the issue on a freshly-installed system with stock
configuration, in order to find out what triggers it.