Re: 2.6 kernel failing lofs mount from lsb-runtime-test

On Thu, 5 Feb 2004, Christopher Yeoh wrote:

> So it looks like when a test is failing the harness is failing to
> unmount the loopback filesystem. Kind of get a train wreck effect
> after that as its unable to mount the loopback ro for other tests,
> leading to lots of unresolved failures.
> 
> > I've found that I need to uninstall/reinstall the test suite to get back 
> > to clean runs. Rebooting back to a 2.4 kernel isn't sufficient.
> 
> I suspect the loopback file is getting corrupted. Could you try saving
> a copy of loopback_disk.ext2 before you run the tests, then replacing
> it after the test run? You'll also need to do a "losetup -d
> /dev/loop0" before copying the backup in if you don't reboot first.
> 

Appears to be true. md5sum of my backup and the original file differ, 
after the failed run. (4 unresolved - mini streamio run)

losetup -d /dev/loop0 doesn't work, I get

ioctl: LOOP_CLR_FD: Device or resource busy

If I unmount the entries in /proc/mounts first, then it works.

A reboot into a 2.4 kernel with the modified loopback_disk.ext2
generates failures.  Same mount points in /proc/mounts.

umount, restore original loopback_disk.ext2, no failures with 2.4 run

md5sum does still differ though from my .bak file, after the good run. No 
lingering /dev/loop0 entries in /proc/mounts.

-- 
Stew Benedict
--
MandrakeSoft