Re: [Qemu-devel] [PATCH 2/8] qcow2: add dirty-bitmaps feature

[John Snow <jsnow@redhat.com>]

On 06/15/2015 10:42 AM, Stefan Hajnoczi wrote:
> On Fri, Jun 12, 2015 at 03:02:33PM -0400, John Snow wrote:
>>
>>
>> On 06/10/2015 10:30 AM, Stefan Hajnoczi wrote:
>>> On Mon, Jun 08, 2015 at 06:21:20PM +0300, Vladimir Sementsov-Ogievskiy wrote:
>>>
>>> I noticed a corner case, it's probably not a problem in practice:
>>>
>>> Since the dirty bitmap is stored with the help of a BlockDriverState
>>> (and its bs->file), it's possible that writing the bitmap will cause
>>> bits in the bitmap to be dirtied!
>>>
>>
>> But since it's metadata and not stored within a disk sector, can this
>> actually happen? Do you have an example of a scenario where this might
>> come up?
> 
> The persistent dirty bitmap for bs->file is storeed in the qcow2 BDS.
> This results in recursion.
> 
> This is a misconfiguration but I just want to understand what happens
> when someone does this by mistake.
> 
> Stefan
> 

I still don't follow you, actually.

The dirty bitmap only tracks changed virtual disk sectors, not actual
file sectors, right? Writing a bitmap that describes foo.qcow2 to
foo.qcow2 won't dirty bitmaps, it's an out-of-channel write as far as
the bitmap is concerned.

Right? Am I fatally misunderstanding the situation?