Re: [Qemu-devel] [PATCH 1/4] block: Correct bs->growable

[Max Reitz <mreitz@redhat.com>]

On 21.08.2014 10:19, Kevin Wolf wrote:
> Am 20.08.2014 um 21:13 hat Max Reitz geschrieben:
>> On 20.08.2014 13:40, Kevin Wolf wrote:
>>> Am 12.07.2014 um 00:23 hat Max Reitz geschrieben:
>>>> Currently, the field "growable" in a BDS is set iff the BDS is opened in
>>>> protocol mode (with O_BDRV_PROTOCOL). However, not every protocol block
>>>> driver allows growing: NBD, for instance, does not. On the other hand,
>>>> a non-protocol block driver may allow growing: The raw driver does.
>>>>
>>>> Fix this by correcting the "growable" field in the driver-specific open
>>>> function for the BDS, if necessary.
>>>>
>>>> Signed-off-by: Max Reitz <mreitz@redhat.com>
>>> I'm not sure I agree with bs->growable = true for raw. It's certainly
>>> true that the backend can technically provide the functionality that
>>> writes beyond EOF grow the file. That's not the point of bs->growable,
>>> though.
>>>
>>> The point of it was to _forbid_ it to grow even when it's technically
>>> possible (non-file protocols weren't really a thing back then, apart
>> >from vvfat, so the assumption was that it's always technically
>>> possible). growable was introduced with bdrv_check_request(), which is
>>> supposed to reject guest requests after the end of the virtual disk (and
>>> this fixed a CVE, see commit 71d0770c). You're now disabling this check
>>> for raw.
>>>
>>> I think we need to make sure that bs->growable is only set if it is
>>> opened for an image that has drv->requires_growing_file set and
>>> therefore not directly used by a guest.
>>>
>>> Well, except that with node-name a guest will be able to use any image
>>> in the chain... Might this mean that it's really a BlockBackend
>>> property?
>> I guess I can make things easy for me by just introducing some
>> "really_growable" or "writes_beyond_eof" field or something for the
>> sake of this series. ;-)
> Nah, don't evade the real solution... Using BDRV_O_PROTOCOL like we
> currently do isn't quite right either. If you clear growable when
> requires_growing_file isn't set for the parent, you should be fine. I
> think. Or hope.

But then using qcow2 over raw over file will throw a warning. *g*

...Okay, you're right, I can go with growable and just don't set it for raw.

Max