Re: [Qemu-devel][PATCH] Qemu image over raw devices

[Shahar Frank <sfrank@redhat.com>]

Kevin Wolf wrote:
> Shahar Frank schrieb:
>> Kevin, Daniel,
>>
>> What do you think is the right thing to do here?
>>
>> Option one:
>>
>> 1. I can suggest a new extension mechanism for QCOW2 (it can be even
>> backwards compatible) and store the backing file format there
>> 2. I can add a flag to qemu-img create to force backing file format
>> (that is stored in the above extension).
>>
>> Or:
>>
>> Option two:
>>
>> I can register qcow2 as a protocol and allow users to explicitly force
>> qcow2 image logic.
>> It is true that the ":" notation is still used, but as this notation is
>> still required even for the drive option (vvfat, for example) it won't
>> be the first to introduce it to the system.
>>
>> I have a feeling you both prefer option one ;-)
> 
> I think I'm not really decided on that one. For me, the most important
> thing is that you don't force users to rely on guessing. So while I
> slightly tend towards the first option, I also wouldn't be opposed to
> the second one.
> 
> Kevin
> 
> 

Kevin, thanks for the input. I think I will try to suggest a patch to 
implement option 2.

My reasons are:
1. The raw image + probing is not caused by the current patch and not 
worsen by the current patch. It just allow you to use probing if you want.
2. I think it much more dangerous to mess with the qcow2 header than 
adding this probe option.
3. The security problem can be solved by various ways. For example, use 
probing just for non raw entities. If you know it is raw, force it to be 
  raw (using one of the current ways). This will solve the security 
problem. I have some other ideas to solve this issue, but none of them 
are related to the current patch - it is completely different issue. I 
hope to post such (additional) patch soon.

Shahar