volume access I/O error with libvirt-xen and iscsi

[Anthony PERARD]

On a single-node devstack installation, on Ubuntu LTS 14.4, with the Xen hypervisor.

To reproduce:
nova volume-create --image-id cirros-0.3.2-x86_64-uec --display-name vol-cirros 1
nova boot --key-name `hostname` --block-device source=volume,id=$volume_id,dest=volume,bootindex=0,shutdown=preserve --image '' --flavor 42 cirros

The instance cirros does not finish to boot, and `nova console-log cirros` shows "I/O error" while accessing the block device.

console-log:
info: copying initramfs to /dev/xvda
[ 79.327661] end_request: I/O error, dev xvda, sector 2
[ 79.327678] Buffer I/O error on device xvda, logical block 1
[ 79.327686] lost page write due to I/O error on xvda
[ 79.327728] EXT3-fs (xvda): I/O error while writing superblock
[ 160.463559] end_request: I/O error, dev xvda, sector 25154
[...]

Another way to reproduce would be to use `nova volume-attach` instead of the --block-device option of nova boot.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/177311

[OpenStack Infra (hudson-openstack)]

Change abandoned by Anthony PERARD (<email address hidden>) on branch: master
Review: https://review.openstack.org/177311
Reason: This is only an issue if the volume and the instance are on the same host. The bug that this patch tries to workaround is been fixed in Linux 4.0.