Reproducible Filesystem Corruption on FC4 (Long)

[Tom Sightler]

Hi List,

I was excited about the developments in FC4 and, over the weekend,
decided that I would convert my laptop from CentOS 4 to FC4 and see how
things went.  I backed up my current system to an external USB2 drive
and decided reformatted the system to prepare it for a fresh install.

After wiping the system I booted from the DVD and began the install.  I
pretty much accepted the defaults for everything (I did change the
default names of the LVM2 volumes and customized my package
selection).  

On the first pass the install exited with an error formatting the /boot
partition.  I didn't think to much about it at the time, and just
rebooted the start the install again.  On this second attempt the
install seemingly went through without a problem.  I rebooted, walked
through the initial welcome pages and logged in to my new FC4 desktop.

Things seemed OK, so I decided to restore the data from my external
drive.  I connected the drive and it appeared on the desktop, and I
began copying the data, basically my /usr/local and /home directories.
The copies seemed to take a longer than expected, but I didn't find any
errors on the system.

After the copies were complete I logged back in with my username and
things looked great at first.  Then I noticed that I was getting errors
trying to run certain applications.  Upon investigating I found that
there were ext3 errors in the log and the system had remounted the
volume read-only.

I rebooted the system and ran fsck.  It found tons of errors, mostly in
the directories that I had copied from my backup device.  After several
cycles of fsck it was continuing to find errors and I pretty much gave
up.

I decided to reinstall and try again.  This time, immediately after the
install I ran fsck and found no errors.  I copied my directories from my
backup again, and the corruption also returned.  I repeated again, this
time I booted with ide=nodma before restoring my backup, this caused the
restore to take so long that I wasn't sure it would ever finish.  I did
not get corruption, but the system was far to slow to use with this
option.

I eventually reinstalled CentOS 4 and restored my backup without issues.
I looked around but don't see anyone else with a similar problem.  I'm
suspicious that the corruption is somehow happening when reading from
the USB2 external drive since the install seems to work without issues,
but if that was the case why would running with ide=nodma keep the
corruption from happening?

Has anyone else seen anything like this?  Are the any suggestions on
what I might be able to do to either get FC4 installed and usable or
gather enough information to report a bug?  I thought of installing the
CentOS 4 kernel before restoring my backup just to prove it was a kernel
related issue.  Any thoughts would be appreciated.

The hardware in question is a Dell D800, 1.7Ghz P4, 1GB RAM with a
Hitichi 60GB internal hard drive.  The external drive is a pretty
generic USB2/Firewire 250GB drive, I don't know the manufacturer off the
top of my head.  This combination had run FC2 without major issues for a
while, and has run CentOS 4 since around March without issues as well,
so I'm pretty sure the hardware is in decent shape.

Thanks,
Tom