Re: [Qemu-devel] [PATCH] Disk image shared and exclusive locks.

["Richard W.M. Jones" <rjones@redhat.com>]

On Mon, Dec 07, 2009 at 08:22:24AM -0600, Anthony Liguori wrote:
> Richard W.M. Jones wrote:
>> On Mon, Dec 07, 2009 at 07:39:11AM -0600, Anthony Liguori wrote:
>>   
>>> Richard W.M. Jones wrote:
>>>     
>>>> Also if we only acquire the lock during the commit operation then
>>>> we'll end up with disk corruption.
>>>>       
>>> Why do we end up with disk corruption?
>>>     
>>
>> Forget about locking for a minute, I don't think this is safe
>> currently.  If you have two VMs set up like:
>>
>>   qemu-img create -b backing.img foo.img
>>   qemu-img create -b backing.img bar.img
>>
>>   qemu -drive file=foo.img     # VM1
>>   qemu -drive file=bar.img     # VM2
>>
>> If VM1 does a commit to the backing image, then VM2 may be caching (in
>> its kernel memory) bits of the old backing image, and will
>> subsequently fetch bits of the new backing image, so it'll see a
>> mixture of old and new data.  How is VM2 supposed to cope with this?
>> It sounds like massive disk corruption to me ...
>>   
>
> Yes, this will cause corruption.  Implementing locking in the fashion  
> I've previously described will prevent 'commit' from being run (since  
> you can't upgrade the lock since someone else is holding a read-lock).

So to be clear, the use case is that all the other VMs must be shut
down, then the VM which wants to commit will upgrade its lock and
commit, and then all the other VMs will restart?  I agree this should
avoid corruption, although it sounds like something which is fairly
unlikely to be done in practice.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
New in Fedora 11: Fedora Windows cross-compiler. Compile Windows
programs, test, and build Windows installers. Over 70 libraries supprt'd
http://fedoraproject.org/wiki/MinGW http://www.annexia.org/fedora_mingw