[ovirt-users] oVirt 4.0.1 on FC23: problem mounting NFSv4 with fsid=0

[Nir Soffer]

On Wed, Jul 27, 2016 at 10:41 PM, Nir Soffer <nsoffer at redhat.com> wrote:
> On Wed, Jul 27, 2016 at 6:07 PM, Nir Soffer <nsoffer at redhat.com> wrote:
>> On Wed, Jul 27, 2016 at 4:57 PM, Idan Shaby <ishaby at redhat.com> wrote:
>>> Hi,
>>>
>>> Can you please attach the vdsm log so we can get all the details?
>>>
>>>
>>> Thanks,
>>> Idan
>>>
>>> On Wed, Jul 27, 2016 at 4:24 PM, wodel youchi <wodel.youchi at gmail.com>
>>> wrote:
>>>>
>>>> Hi,
>>>>
>>>> I wanted to test the new version of oVirt the 4.0.1 on FC23, using HE
>>>> method and using NFSv4 for storage.
>>>>
>>>> When launching the deploy command, and choosing nfsv4 and filling the
>>>> mount point, the setup exits with error : Connection to storage server
>>>> failed.
>>>>
>>>> In vdsm log I have this error:
>>>> Could not connect to storageServer
>>>> Traceback (most recent call last):
>>>> ...
>>>> OSError: [Errno 2] Mount of `noveria.wodel.wd:/engine_vm` at
>>>> `/rhev/data-center/mnt/noveria.wodel.wd:_engine__vm` does not exist
>>>>
>>>> the df -h command shows that the nfs mount was done, but the device path
>>>> contains a double slash.
>>>> noveria.wodel.wd://vmengine  656G  111G  513G  18%
>>>> /rhev/data-center/mnt/noveria.wodel.wd:_vmengine
>>
>> Can you share the output of:
>>
>>     cat /proc/mounts
>>
>> This is what vdsm is using the check if a mount exists.
>>
>>>>
>>>>
>>>> I am not sure of the problem, but I think it's the double slash on the
>>>> nfs4 path when using fsid=0: noveria.wodel.wd://vmengine
>>
>> Vdsm is normalizing the mount path - when you try to mount:
>>
>>      noveria.wodel.wd://vmengine
>>
>> Vdsm normalize the path to:
>>
>>      noveria.wodel.wd:/vmengine
>>
>> And it pass this value to the mount command.
>>
>> Vdsm expect that the kernel will not modify the value and return unormalized
>> paths.
>>
>> This looks like an issue reported in the util-linux list:
>> http://www.spinics.net/lists/util-linux-ng/msg13078.html
>>
>> According to the report, this is known issue in the kernel and unlikely to
>> be fixed soon.
>>
>> So it seems that we do need to normalize also what the kernel returns
>> in /proc/mounts.
>
> I posted these patches that should fix this issue.
> https://gerrit.ovirt.org/#/q/status:open+project:vdsm+branch:ovirt-4.0+topic:backport/4.0/normalize-mounts

Testing the nfs4 patch show that we need more work.

In the meantime, can you check if using mount path with double slash
(e.g. server://path) in engine works?

Also, can you file a bug about this?

Cheers,
Nir