Comment 0 for bug 802081

On the host machine, a disk is configured with 4k sector sizes:
---
sudo fdisk /dev/sdd

Disk /dev/sdd: 4000.6 GB, 4000627818496 bytes
255 heads, 63 sectors/track, 60797 cylinders
Units = cylinders of 16065 * 4096 = 65802240 bytes
Sector size (logical/physical): 4096 bytes / 4096 bytes
I/O size (minimum/optimal): 4096 bytes / 4096 bytes
Disk identifier: 0xf8eabb69
---
... and when fdisking the lvm parition, also reports properly:
---
Disk /dev/VMsVG/mail: 21.0 GB, 20971520000 bytes
255 heads, 63 sectors/track, 318 cylinders
Units = cylinders of 16065 * 4096 = 65802240 bytes
Sector size (logical/physical): 4096 bytes / 4096 bytes
I/O size (minimum/optimal): 4096 bytes / 4096 bytes
Disk identifier: 0x6a3bdb21
---
However, within a VM using the LVM parition as a raw disk, its fdisk reports a 512 block size:
---
Disk /dev/vda: 21.0 GB, 20971520000 bytes
16 heads, 63 sectors/track, 5079 cylinders
Units = cylinders of 1008 * 4096 = 4128768 bytes
Sector size (logical/physical): 4096 bytes / 4096 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x915c306a

using cache=none, which is established practice for best performance on raw devices, the host system repeatedly fails to write to the disk, due to it trying to write 512 byte sectors while the backing is 4096.