Comment 2 for bug 741643

This is a known issue in upstream; it appears to hurt usb devices more than sata.

The btrfs module doesn't release the inode properly when the device disappears (you should see the device is still in /dev). When the next transaction commit occurs (in this case, due to a presumably automated umount call), writing out the superblock fails, which triggers the sanity check (the reasoning being that it's much better to noisily and visibly die than to silently corrupt data).