Re: [Qemu-devel] [PATCHv2 00/18] qemu-iotests: adjust tests to work with the NFS protocol

[Peter Lieven]

Am 24.02.2014 13:54, schrieb Stefan Hajnoczi:
> On Sun, Jan 05, 2014 at 06:21:50PM +0100, Peter Lieven wrote:
>> In order to proceed with the integration of the NFS protocol driver into
>> qemu I was asked by Stefan to add integration for NFS into the qemu-iotests.
>>
>> Unfortunately, this became more complex than I had expected because most
>> of the tests for non RAW formats only work with the file protocol because
>> they use shell commands like cp, rm or mv which obviously don't work
>> on nfs:// URLs.
>>
>> This series first changes all those tests that are not working out of the
>> box to support only the file protocol.
>>
>> After the NFS protocol is introduced in Patch 2 I fix most of them
>> to work with any protocol.
>>
>> After this series the qemu-iotests for NFS run through gracefully with
>> the RAW, QCOW2 and VMDK formats.
>>
>> There are 3 topics open:
>>  - test 051 fails regardless which protocol is used (I already send a msg
>>    to the list)
>>  - test 052 should work, but it seems there is a bug in the bdrv_open
>>    logic if the BDRV_O_SNAPSHOT flag is set and the protocol is anything
>>    else than file. Maybe someone with more understanding of the whole
>>    open logic could look at this. I do not believe that its sth which
>>    has to do with the NFS driver since the test fails while opening the
>>    backing file and other backing file tests run without problems.
>>  - other protocols like sheepdog or ssh that are allowed to use other formats
>>    than raw should be tested. they actually can't never have run qemu-iotests
>>    with qcow2 protocol for instance.
>>
>> If you want to do your tests please make sure to have
>>    [PATCHv5] block: add native support for NFS
>>    [PATCH v2] vmdk: Allow vmdk_create to work with protocol
> Going through my email backlog.  What is the status of this series?
The NFS driver and bare qemu-iotests support is merged. All the patches
that try to fix the limit for protocol != file where left out because there
was no consensus how the filename construction etc. should be handled.

Peter

>
> Stefan