This is just a little update about my JFS corruption problem.  After making a few tests about ecryptfs, I discovered it may not be as stable as the author claims it to.  I'm starting to doubt if it would not corrupt the underlying filesystem somehow as it hooks calls in the VFS.  I decided to dump it and use the more widely adopted and tested dm-crypt / cryptsetup.  I prefered encryption on a per-file basis as the cpu is a duron 750mhz, but after testing dm-crypt seems, surprisingly, to be fast enough.

So right now, I created a dm-crypt device on top of my mdadm raid 1 device, and formatted the encrypted device to JFS.  Up to now, no new corruption arised and the system is undergoing heavy stress tests.  It's copying via rsync/rsh about 500 gb of files, then it deletes everything and starts over.

I'll keep you informed if new corruption arise.  If the filesystem stays rock solid after two weeks of testing, I'll consider the problem closed and file a bug request at ecryptfs.

Thank you all,


Dave Kleikamp wrote:
On Thu, 2007-10-18 at 08:11 -0400, Charles Perreault wrote:
So before repairing it, I ran

jfs_fsck -e /dev/md0 -f current.log
jfs_fsck -p -e /dev/md0 -f previous.log

Hope that can help a bit.

I'm not sure if it will lead to the root cause, but it's probably all
there is to look at after the fact.  I'll get back to you after I've
looked at it some more.