Re: kern/45305: umount says device busy without any process having current directory in the mount or file open

[christos%zoulas.com@localhost (Christos Zoulas)]

The following reply was made to PR kern/45305; it has been noted by GNATS.

From: christos%zoulas.com@localhost (Christos Zoulas)
To: Matthias Kretschmer <kretschm%cs.uni-bonn.de@localhost>, 
gnats-bugs%NetBSD.org@localhost
Cc: kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, 
        netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/45305: umount says device busy without any process having 
current directory in the mount or file open
Date: Tue, 27 Sep 2011 10:15:51 -0400

 On Sep 27,  4:07pm, kretschm%cs.uni-bonn.de@localhost (Matthias Kretschmer) 
wrote:
 -- Subject: Re: kern/45305: umount says device busy without any process havin
 
 | Hi
 | 
 | On Tue, Sep 27, 2011 at 01:50:04PM +0000, Christos Zoulas wrote:
 | >  Do you have a recipe for recreating the problem?
 | 
 | no recipe which is quite worth anything.  I'll stumble upon the problem,
 | when I update large packages.  I perform compiling of packages inside a
 | chroot to not have to touch the current installation.  As this iBook
 | with G3 600MHz (where I have this problem) is quite slow and has not
 | much disk-space, I perform compiling of packages via a NetBSD/macppc
 | installation (same userland) hosted on another box which is mounted to
 | /netboot (basically it is the same installation which is netboot-able).
 | To speed up compilations I mount_null a /netboot/tmp and
 | /netboot/wrkobjdir and set PKGSRC_COMPILER to ccache distcc gcc, where
 | the only DISTCC_HOST is the nfs-host.  So only preprocessing is done
 | locally.
 | 
 | The problem seems to always occur when I compile www/firefox (including
 | devel/xulrunner) from pkgsrc HEAD (so last time version 6.0.2).  The
 | last time, when even the nullfs didn't want to unmount without -f, I was
 | additionally building chat/ircII.  At the moment I do not have a clue if
 | creating, opening, writing, reading, or chdir'ing causes this problem
 | (or a probably parallel combination of all of them).  Even though
 | writing a test program wouldn't be much work, I have not found the time
 | to write it...
 | 
 | Are there any counters of open/etc. files for a mount-point?  Am I able
 | to output these somehow?  As there is no record of anything accessing
 | the corresponding mount-points I'm wondering why the systems thinks it
 | is busy.
 
 I would start adding printfs in the unmount path in vfs_mount.c following
 them down to the one that returns EBUSY...
 
 christos