+++ This bug was initially created as a clone of Bug #853674 +++

Created attachment 609013 [details]
strace -f ./testumount.sh >testumount.txt 2>&1

Description of problem:
Mount a floppy image, and write some files, then immediately umount. The umount will fail.


Version-Release number of selected component (if applicable):
uname -a:
Linux zhshzhouf17 3.5.2-3.fc17.x86_64 #1 SMP Tue Aug 21 19:06:52 UTC 2012 x86_64 x86_64 x86_64 GNU/Linux

rpm -q glibc:
glibc-2.15-56.fc17.x86_64


How reproducible:
On my machine, always, on kernel 3.4 and 3.5 release. Other release are not tested. On some of my colleagues's machines, it can be reproduced. It can not be reporduced on virtual machines.


Steps to Reproduce:
1. I write a script to reproduce the problem.

#!/usr/bin/bash
mkdir tmpMount
/usr/sbin/mkfs.msdos -C floppy.img 1440
/usr/bin/mount -o loop floppy.img tmpMount
echo '123'>tmpMount/tst1.txt
echo '123'>tmpMount/tst2.txt
echo '123'>tmpMount/tst3.txt
echo '123'>tmpMount/tst4.txt
echo '123'>tmpMount/tst5.txt
echo '123'>tmpMount/tst6.txt
echo '123'>tmpMount/tst7.txt
echo '123'>tmpMount/tst8.txt
echo '123'>tmpMount/tst9.txt
#sleep 2
/usr/bin/umount -f tmpMount
rm -rf tmpMount floppy.img
  
Actual results:
When I run the script, umount fails. The script prints some error message as follow.
mkfs.msdos 3.0.12 (29 Oct 2011)
umount: /home/edward/src/tmpMount: target is busy.
        (In some cases useful info about processes that use
         the device is found by lsof(8) or fuser(1))
rm: cannot remove `tmpMount': Device or resource busy

Expected results:
The device can be umounted and no error are reported.

Additional info:
1. sleep several seconds before umount
If I uncomment the "sleep" command in the script, umount works again. This means data is waiting to be sync to disk, so the device is busy, and umount can not detach the devices. On my system, if I sync before umount, umount will work.

2. lsof the mount point prints nothing
If I lsof the mount point between before umount, nothing will be printed, but umount works again

So I guess the driver/kernel holds the devices because the data is in the a kernel buffer but not written to disk.

By the way, lsof gives the following waring when I lsof the mount point. lsof: WARNING: can't stat() fuse.gvfs-fuse-daemon file system /run/user/edward/gvfs Output information may be incomplete

I have fuse and gvfs-fuse packages installed, but fuse.service can not be started. When I try to start the service, it fails and says "No such file or directory".

I don't know if the fuse problem is related to this problem.

3. lazy umount does not work, but fsync + lazy umount works
When I use lazy umount in the script, it works. However when I let the script mount the image again immediately, the image contains nothing. The image should contain the files created by the script.

When I fsync the FDs of those files before the lazy umount, umount successed. Then I mount the image again immediately, I can see the files there.

4. strace
The detialed strace report is in the attachment. I use the following command to generate that report.
strace -f ./testumount.sh >testumount.txt 2>&1

--- Additional comment from danken on 2012-09-02 13:50:04 CST ---

Could you check if this is also the case with other filesystems, or only with mkfs.msdos ?

--- Additional comment from zhshzhou.ibm.com on 2012-09-02 19:02:12 CST ---

Thanks for the quick reply. I tested ext2, ext3, ext4 and vfat images, all can reproduce this bug.

--- Additional comment from danken on 2012-09-03 05:42:39 CST ---

A kernel developer failed to reproduce the issue on his system today. Could you give more details on the hosts where this shows up? Which storage is backing your filesystem? Is there a host where you can grant access to a developer?

--- Additional comment from zhshzhou.ibm.com on 2012-09-03 10:39:53 CST ---

I'm sorry that I can not grant access because of the security policy.

My machine is ThinkPad T420, manufactured by Lennovo. The colleagues' machines are ThinkPad T410.

I use hard drive and Fedora 17 default LVM layout created by the OS installation.

lvm versions are
lvm2-libs-2.02.95-6.fc17.x86_64
lvm2-2.02.95-6.fc17.x86_64

df -l gives the following information
Filesystem                      1K-blocks     Used Available Use% Mounted on
rootfs                           93857720 33665472  55424504  38% /
devtmpfs                          4012256        0   4012256   0% /dev
tmpfs                             4029556      304   4029252   1% /dev/shm
tmpfs                             4029556     2320   4027236   1% /run
/dev/mapper/vg_zhshzhou-lvroot   93857720 33665472  55424504  38% /
tmpfs                             4029556        0   4029556   0% /sys/fs/cgroup
tmpfs                             4029556        0   4029556   0% /media
/dev/mapper/35000cca68ac8e2fdp2    495844    86730    383514  19% /boot
/dev/mapper/vg_zhshzhou-lvhome  103212320 33563604  64405836  35% /home



--- Additional comment from wudxw.ibm.com on 2012-09-07 15:29:28 CST ---

I can reproduce this problem, and the problem disappear after I kill the process of gvfsd-trash.  It looks it's the same problem reported in: http://code.google.com/p/chromium-os/issues/detail?id=23443. 
So I think we should ask gvfs guys take a look at first.

gvfs version:
gvfs-1.12.3-1.fc17