Re: [libvirt] [PATCH] Add basic support for VDI images
On 02/04/2013 07:03 PM, Daniel P. Berrange wrote:
On Mon, Feb 04, 2013 at 06:53:34PM +0100, Martin Kletzander wrote:
> On 02/04/2013 04:53 PM, Daniel P. Berrange wrote:
>> On Mon, Feb 04, 2013 at 04:46:53PM +0100, Martin Kletzander wrote:
> [...]
>>> +
>>> + /* not fully supported, but qemu knows it, so we should be able to
>>> + * handle this at least basically */
>>> + [VIR_STORAGE_FILE_VDI] = { NULL, ".vdi", LV_LITTLE_ENDIAN,
>>> + -2, 0, 0, 0, 0, -1, NULL},
>>
>> We can do better than that - look at block/vdi.c in QEMU GIT tree
>> to find out the offsets / values for version number, magic signature
>> at least, even if you ignore backing files for now.
>>
>
> Thanks for pointing that out, I've got a v2 and will send it
> after some testing. However, it looks like VDI doesn't have a
> magic, because there is a 40 Byte string at the start that cacn
> contain anything the creating binary puts there. For example
> qemu-img uses "<<< QEMU VM Virtual Disk Image >>>\n".
According
> to [1], I'm adding "<<<" as the magic, because we don't
fail if
> it doesn't match and we can still match it by the extension then.
No, you need to look at the 'signature' field of the header
instead, which has a fixed magic value
#define VDI_SIGNATURE 0xbeda107f
But the struct mapped to the header starts like this:
typedef struct {
char text[0x40];
uint32_t signature;
...
Of course that would just mean adding one more parameter to the
FileTypeInfo struct, but I thought it's not magic then. However, I
checked that and you're right, I was wrong, because according to
magic(5) it can be anywhere. The magic file '/usr/share/misc/magic.mgc'
also matches the signature, so I'm using that.
Martin