Bug 760138

Description of problem:

Setup: 4 Xen-VMs share the same disk image that should be mounted read-only by Xen.

When I start the VMs in sequential order (one after another) using python-libvirt, everything works.  If I start the VMs in parallel (using multiple python threads) I hit a race condition and get the following error from Xen:

  Device 51713 (vbd) could not be connected.
  File /home/xge/imagepool/2/a64/disk.img is loopback-mounted through 
  /dev/loop0 
  /dev/loop3,
  which is mounted read-write in a guest domain,
  and so cannot be mounted read-only now

It seems libvirt mounts the disk image read-write for a short period of time and then switches to read-only.  If I add a small delay between the thread creation, it also works. Nevertheless, this is not a good solution :)

Version-Release number of selected component (if applicable):

Using the package from Debian-Squeeze 0.8.3-5+squeeze2

Xen in Version 4.0.1 and the Kernel is 2.6.32-5-xen-amd64


How reproducible:

Use libvirt and start multiple VMs using the same read-only disk image in parallel.

Steps to Reproduce:
1. Create a VM disk image
2. Write some python-libvirt code to start multiple VMs in parallel
3. Set the <readonly> flag in the <disk> section of the XML
  
Actual results:

Xen reports an error. Some VMs are started, some not.

Expected results:

All VMs are running and share the same disk image that is accessed r/o-only.

Additional info: