[dm-devel] Buffer I/O and journaling errors while using snapshot

[cramerj]

I'm not too familiar with device mapper or filesystems yet, so please bear with me.

So I'm attempting to create a writeable snapshot of a read-only device in order to create a ramdisk overlay for temporary changes that will be discarded later.   DM seems to do the job.  However, after tinkering with the ramdisk overlay for awhile, I encounter Buffer I/O errors with "__journal_remove_journal_head: freeing b_frozen_data" messages.

>From what I understand, the ramdisk size should simply cover the changes incurred, not necessarily the full size of the read-only device.  Even so, doesn't the ramdisk size grow as needed?  Either way, I don't know if it's a case of running out of ramdisk space or physical memory, or something else I'm missing.

Info:
2.6.8-rc1
ramdisk_size=32768
1GB physical ram
Size of MyImage.img is anywhere between 1.4GB and 3GB.

Here's the script and basic steps of what I'm doing to get the errors...

losetup /dev/loop0 MyImage.img
echo "0 `blockdev --getsize /dev/loop0` snapshot-origin /dev/loop0" | dmsetup create MyImage-ro
mkdir /mnt/MyImage-ro
mount -r /dev/mapper/MyImage-ro /mnt/MyImage-ro
sync
losetup /dev/loop1 /dev/ram0
echo "0 `blockdev --getsize /dev/loop0` snapshot /dev/loop0 /dev/loop1 n 128" | dmsetup create COWDrive-rw
mkdir /mnt/COWDrive-rw
mount /dev/mapper/COWDrive-rw /mnt/COWDrive-rw

- I copy some files over to /mnt/COWDrive-rw/home/me/,
- do a mount --bind /proc /mnt/COWDrive-rw/proc because I'm messing with RPM,
- chroot into /mn/COWDrive, then do some file ops.

...and then come the errors.  After exiting the chroot environment, tear down the mounts and remove the dm's, I reboot.  However not everything unmounts cleanly because I get fsck'd on boot-up.

Am I running out of memory?  Is what I'm doing completely silly?

Thanks,
-Jeb