Error reading VMDK4 with footer instead of header

[bbgordonn]

VMDK4 files can have a footer in the last block, which is the same datastructure as the header but must be used instead if present. In this case, the gd_offset in the usual header at the beginning of the file is the special flag -1 (VMDK 1.1 spec, page 17, "GD_AT_END
"). qemu-img doesn't know about this flag so it goes on to try to read extents with a bogus l1_table from the wrong location in the file.

I have regression-tested this with various OVAs exported from VSphere/ESXi 3 and 4. Current master and all previous QEMU versions were unable to import any compressed VMDKs with a footer. It now works on all the ones I have.

bb45ded93115ad4303471c9a492579dc36716547 changed the order of gd_offset and rgd_offset in the VMDK4Header struct. Page 8 of the VMDK 1.1 spec from VMWare shows the structure as rgd_ then gd_, while QEMU now has gd_ *before* rgd_offset. I was only able to get VMDK conversion to work by switching the order back to that specified by VMWare and previously used by QEMU. I don't know what VMDK this commit is referring to, so I can't test to see if I've broken it. :(

I will submit this patch to the mailing list if I get a chance, but I'm also uploading it here so I don't lose it.

[Stefan Hajnoczi (stefanha)]

On Tue, Dec 20, 2011 at 8:53 PM, bbgordonn <email address hidden> wrote:
> Public bug reported:
>
> VMDK4 files can have a footer in the last block, which is the same datastructure as the header but must be used instead if present. In this case, the gd_offset in the usual header at the beginning of the file is the special flag -1 (VMDK 1.1 spec, page 17, "GD_AT_END
> "). qemu-img doesn't know about this flag so it goes on to try to read extents with a bogus l1_table from the wrong location in the file.
>
> I have regression-tested this with various OVAs exported from
> VSphere/ESXi 3 and 4. Current master and all previous QEMU versions were
> unable to import any compressed VMDKs with a footer. It now works on all
> the ones I have.
>
> bb45ded93115ad4303471c9a492579dc36716547 changed the order of gd_offset
> and rgd_offset in the VMDK4Header struct. Page 8 of the VMDK 1.1 spec
> from VMWare shows the structure as rgd_ then gd_, while QEMU now has gd_
> *before* rgd_offset. I was only able to get VMDK conversion to work by
> switching the order back to that specified by VMWare and previously used
> by QEMU. I don't know what VMDK this commit is referring to, so I can't
> test to see if I've broken it. :(
>
> I will submit this patch to the mailing list if I get a chance, but I'm
> also uploading it here so I don't lose it.
>
> ** Affects: qemu
>     Importance: Undecided
>         Status: New
>
> --
> You received this bug notification because you are a member of qemu-
> devel-ml, which is subscribed to QEMU.
> https://bugs.launchpad.net/bugs/907063

Thanks for reporting this. I have CCed Fam who worked on VMDK this summer.

Please submit patches to the mailing list according to the guidelines here:

http://wiki.qemu.org/Contribute/SubmitAPatch

Stefan

[Thomas Huth (th-huth)]

Looks like something similar has been commited here:
http://git.qemu.org/?p=qemu.git;a=commitdiff;h=65bd155c7356d448ffee7
So is this problem fixed nowadays?

[Launchpad Janitor (janitor)]

[Expired for QEMU because there has been no activity for 60 days.]