On 2018-07-02 23:07, Max Reitz wrote:
> The VMDK driver assumes that its backing file is always a VMDK file,
> too, because it verifies that when creating the overlay. However, that
> verification means nothing at runtime, and consequently you can assign
> non-VMDK backing files to a VMDK node. This then breaks because the
> driver accesses the backing node as a VMDK node to read its supposed CID
> to compare it with the overlay's parentCID entry -- which usually fails,
> either in a benign way (we read from a garbage offset, and then we read
> garbage or get a read error straight away), or we get a segfault
> (because the backing node does not have a respective file child).
>
> Anyway, we just shouldn't do it and instead check whether the backing
> file is a VMDK node before treating it like one.
>
>
> (This fixes
> http://lists.nongnu.org/archive/html/qemu-block/2018-06/msg01268.html)
>
>
> Max Reitz (2):
> vmdk: Fix possible segfault with non-VMDK backing
> iotests: Add VMDK backing file correlation test
>
> block/vmdk.c | 6 ++
> tests/qemu-iotests/225 | 132 +++++++++++++++++++++++++++++++++++++
> tests/qemu-iotests/225.out | 24 +++++++
> tests/qemu-iotests/group | 1 +
> 4 files changed, 163 insertions(+)
> create mode 100755 tests/qemu-iotests/225
> create mode 100644 tests/qemu-iotests/225.out
Applied to my block branch.
Max