Issues with md RAID-1, LVM, and snapshots

I'm having trouble with snapshot volumes when combining software RAID-1 with LVM2 striping.

I have a server with four 18GB drives. I would like to combine them into two RAID-1 mirrors and then stripe across them with LVM, with the goal of data redundancy and the ability to make hot backups via snapshots.

System:
Suse Linux Enterprise Server 9 for x86_64
kernel 2.6.5-7.147-smp
lvm2-2.00.15-0.8

When I set up the drives in one big volume group (no RAID), snapshots work correctly:

pvcreate /dev/sdc1
pvcreate /dev/sdd1
pvcreate /dev/sdf1
pvcreate /dev/sde1
vgcreate -M2 vg1 /dev/sdc1 /dev/sdd1 /dev/sde1 /dev/sdf1 lvcreate -i4 -I64 -L32G -n lv1 vg1
mkfs.ext2 /dev/vg1/lv1
mount /dev/vg1/lv1 /mnt/bigdisk
{put some data in /mnt/bigdisk}
lvcreate -L1G -s -n snap1 /dev/vg1/lv1
mount /dev/vg1/snap1 /mnt/backup
find /mnt/backup | cpio -ovHnewc > /tmp/backup.cpio

This works fine. However, if I set up two RAID-1 volumes and stripe across them with LVM, I get errors when reading from a snapshot:

mdadm --create -l1 -n2 /dev/sdc1 /dev/sdd1 /dev/md2 mdadm --create -l1 -n2 /dev/sde1 /dev/sdf1 /dev/md3 pvcreate /dev/md2
pvcreate /dev/md3
vgcreate -M2 vg1 /dev/md2 /dev/md3
lvcreate -i2 -I64 -L32G -n lv1 vg1 mkfs.ext2 /dev/vg1/lv1
mount /dev/vg1/lv1 /mnt/bigdisk
{put some data in /mnt/bigdisk}
lvcreate -L1G -s -n snap1 /dev/vg1/lv1 mount /dev/vg1/snap1 /mnt/backup

I get this error:
cpio: Read error at byte 2129920 in file /mnt/backup/file1, padding with zeros

dmesg says:
attempt to access beyond end of device
dm-3: rw=0, want=17060337672, limit=67108864
attempt to access beyond end of device
dm-3: rw=0, want=134226096, limit=67108864
attempt to access beyond end of device
dm-3: rw=0, want=3499657224, limit=67108864
attempt to access beyond end of device
dm-3: rw=0, want=3094910976, limit=67108864
attempt to access beyond end of device
dm-3: rw=0, want=8894171864, limit=67108864
attempt to access beyond end of device
dm-3: rw=0, want=17060337672, limit=67108864

When I extract the archive, file1 is corrupted.

Is there any special magic that needs to be done with LVM+MD, or have I hit a bug?

--
Kevin Murphy

I've got exactly the same problem here. Raid1(md)+LVM2 (2.6.11, Xen Domain0), can't snapshot correctly. In dmesg, I've got :

attempt to access beyond end of device
dm-30: rw=0, want=1905355624, limit=1228800 attempt to access beyond end of device
dm-30: rw=0, want=1905357216, limit=1228800 attempt to access beyond end of device
dm-30: rw=0, want=8390664, limit=1228800

As soon as I remove the Raid1, the problem disappears.

Is it a known problem ? Is there a workaround ?