Re: [Qemu-devel] [RFC PATCH] vpc: Ignore geometry for large images

[Peter Lieven <pl@kamp.de>]

Am 10.02.2015 um 14:54 schrieb Kevin Wolf:
> Am 10.02.2015 um 14:42 hat Jeff Cody geschrieben:
>> On Tue, Feb 10, 2015 at 02:34:14PM +0100, Kevin Wolf wrote:
>>> Am 10.02.2015 um 12:41 hat Peter Lieven geschrieben:
>>>> Am 09.02.2015 um 17:09 schrieb Kevin Wolf:
>>>>> The CHS calculation as done per the VHD spec imposes a maximum
>>>>> image size of ~127 GB. Real VHD images exist that are larger than
>>>>> that.
>>>>>
>>>>> Apparently there are two separate non-standard ways to achieve
>>>>> this: You could use more heads than the spec does - this is the
>>>>> option that qemu-img create chooses.
>>>>>
>>>>> However, other images exist where the geometry is set to the
>>>>> maximum (65536/16/255), but the actual image size is larger.
>>>>> Until now, such images are truncated at 127 GB when opening them
>>>>> with qemu.
>>>>>
>>>>> This patch changes the vpc driver to ignore geometry in this case
>>>>> and only trust the size field in the header.
>>>>>
>>>>> Signed-off-by: Kevin Wolf <kwolf@redhat.com> ---
>>>>>
>>>>> Peter, I'm replacing some of your code in the hope that the new
>>>>> approach is more generally valid. Of course, I haven't tested if
>>>>> your case with disk2vhd is still covered. Could you check this,
>>>>> please?
>>>> I checked this and found that disk2vhd always sets CHS to 65535ULL
>>>> * 16 * 255 independed of the real size.
>>>>
>>>> But, as the conversion to CHS may have an error its maybe the best
>>>> solution to ignore CHS completely and always derive total_sectors
>>>> from footer->size unconditionally.
>>>>
>>>> I had a look at what virtualbox does and they only rely on
>>>> footer->size. If they alter the size or create an image the write
>>>> the new size into the footer and recalculate CHS by the formula
>>>> found in the appendix of the original spec.
>>>>
>>>> Check vhdCreateImage, vhdOpen in
>>>> http://www.virtualbox.org/svn/vbox/trunk/src/VBox/Storage/VHD.cpp
>>>>
>>>> The original spec also says that CHS values purpose is the use in
>>>> an ATA controller only.
>>> The problem with just using footer->size back then when I
>>> implemented this was that from the perspective of a VirtualPC guest
>>> run in qemu, the size of its hard disk would change, which you don't
>>> want either. Going from VPC to qemu would be ugly, but mostly
>>> harmless as the disk only grows. But if you use an image in qemu
>>> where the disk looks larger and then go back to VPC which respects
>>> geometry, your data may be truncated.
>> I believe the vpc "creator" field is different if the image was
>> created by Virtual PC, versus created by Hyper-V ("vpc" and "win",
>> respectively, I think).  Perhaps we could use that to infer a guest
>> image came from VirtualPC, and thus not use footer->size in that
>> scenario?
> Right, I think we discussed that before. Do you remember the outcome of
> that discussion? I seem to remember that we had a conclusion, but
> apparently it was never actually implemented.
>
> Would your proposal be to special-case "vpc" to apply the geometry, and
> everything else (including "win", "d2v" and "qemu") would use the footer
> field?

That sounds reasonable. In any case we have to fix qemu-img create
to do not create out of spec geometry for images larger than 127G.
It should set the correct footer->size and then calculate the geometry.

Peter