Re: [Qemu-devel] Re: [PATCH 2/2] Add flush=off parameter to -drive

[Anthony Liguori <anthony@codemonkey.ws>]

On 05/11/2010 11:32 AM, Jamie Lokier wrote:
> Anthony Liguori wrote:
>    
>> On 05/11/2010 08:12 AM, Paul Brook wrote:
>>      
>>>>> cache=always (or a more scary name like cache=lie to defend against
>>>>> idiots)
>>>>>
>>>>>     Reads and writes are cached. Guest flushes are ignored.  Useful for
>>>>>     dumb guests in non-critical environments.
>>>>>
>>>>>            
>>>> I really don't believe that we should support a cache=lie.  There are
>>>> many other obtain the same results.  For instance, mount your guest
>>>> filesystem with barrier=0.
>>>>
>>>>          
>>> Ideally yes. However in practice I suspect this is still a useful option.
>>> Is
>>> it even possible to disable barriers in all cases (e.g. NTFS under
>>> windows)?
>>>
>>> In a production environment it's probably not so useful - you're generally
>>> dealing with long lived, custom configured guests.
>>>
>>> In a development environment the rules can be a bit different. For example
>>> if
>>> you're testing an OS installer then you really don't want to be passing
>>> magic
>>> mount options. If the host machine dies then you don't care about the
>>> state of
>>> the guest because you're going to start from scratch anyway.
>>>
>>>        
>> Then create a mount point on your host and mount the host file system
>> under that mount with barrier=0.
>>      
> Two reasons that advice doesn't work:
>
> 1. It doesn't work in many environments.  You can't mount a filesystem
> with barrier=0 in one place and barrier=1 on a different point, and
> there's ofen only one host partition.
>    

qemu-img create -f raw foo.img 10G
mkfs.ext3 foo.img
mount -oloop,rw,barrier=1 -t ext3 foo.img mnt

Works perfectly fine.

> 2. barrier=0 does _not_ provide the cache=off behaviour.  It only
> disables barriers; it does not prevent writing to the disk hardware.
>    

The proposal has nothing to do with cache=off.

>> The problem with options added for developers is that those options are
>> very often accidentally used for production.
>>      
> We already have risky cache= options.  Also, do we call fdatasync
> (with barrier) on _every_ write for guests which disable the
> emulated disk cache?
>    

None of our cache= options should result in data corruption on power 
loss.  If they do, it's a bug.

Regards,

Anthony Liguori

> -- Jamie
>