Re: [PATCH 0/3] Enable the setting of a kernel module parameter from nfs.conf

[Chuck Lever III <chuck.lever@oracle.com>]

> On Apr 17, 2021, at 12:18 PM, Steve Dickson <SteveD@RedHat.com> wrote:
> 
> 
> 
> On 4/15/21 12:37 PM, Chuck Lever III wrote:
>> 
>> 
>>> On Apr 15, 2021, at 11:33 AM, Steve Dickson <SteveD@RedHat.com> wrote:
>>> 
>>> Hey Chuck! 
>>> 
>>> On 4/14/21 7:26 PM, Chuck Lever III wrote:
>>>> Hi Steve-
>>>> 
>>>>> On Apr 14, 2021, at 2:10 PM, Steve Dickson <SteveD@redhat.com> wrote:
>>>>> 
>>>>> This is a tweak of the patch set Alice Mitchell posted last July [1].
>>>> 
>>>> That approach was dropped last July because it is not container-aware.
>>>> It should be simple for someone to write a udev script that uses the
>>>> existing sysfs API that can update nfs4_client_id in a namespace. I
>>>> would prefer the sysfs/udev approach for setting nfs4_client_id,
>>>> since it is container-aware and makes this setting completely
>>>> automatic (zero touch).
>>> As I said in in my cover letter, I see this more as introduction of
>>> a mechanism more than a way to set the unique id.
>> 
>> Yep, I got that.
>> 
>> I'm not addressing the question of whether adding a
>> mechanism to set a module parameter in nfs.conf is good
>> or not. I'm saying nfs4_client_id is not an appropriate
>> parameter to add to nfs.conf. Can you pick another
>> module parameter as an example for your mechanism?
> The request was to set that parameter...

The cover letter and the quoted e-mail above state that
you are just demonstrating a mechanism to set module
parameters via nfs.conf. I guess that statement was not
entirely accurate, then. Thanks for clarifying.

If there's a bug report somewhere that requests a
feature, it's normal practice to include a URL pointing
to that report in the patch description.


>>> The mechanism being
>>> a way to set kernel module params from nfs.conf. The setting of
>>> the id is just a side effect... 
>>> 
>>> Why spread out the NFS configuration?  Why not
>>> just keep it in one place... aka nfs.conf?
>> 
>> We need to understand whether a module parameter is not
>> going to work in nfs.conf because that setting needs to
>> be namespace-aware. In this case, this setting does indeed
>> need to be namespace-aware. nfs.conf is not aware of
>> network namespaces.
> You probably can say that for every /etc/*.conf file...  
> not being namespace aware... how can they be...
> 
> In the kernel document you say "Specifying a uniquifier string is 
> not support for NFS clients running in containers."
> 
> Why isn't that enough?

Because that statement is noting a limitation which is a
bug that has to be addressed to support NFSv4 properly in
containers.


>>> As far as not being container-aware... that might true
>>> but it does not mean its not useful to set the id from
>>> nfs.conf...
>> 
>> Yes, it does mean that in that case. It's completely
>> broken to use the same nfs4_client_id in every network
>> namespace.
> How does this break? If all the clients have unique ids
> what breaks?
> 
> Or are you saying setting the unique id like this:
> 
> options nfs nfs4_unique_id="64fd26280451566d648ab0e0b7384421"
> 
> in modprobe.d/nfs.conf is not namespace safe?

Setting the client_id via module parameter is not
namespace-aware. That's the bug that the sysfs/udev
contraption is designed to fix.


>>> Actual I have customers asking for this type
>>> of functionality
>> 
>> Ask yourself why they might want it. It's probably because
>> we don't set it correctly currently. If we have a way to
>> automatically get it right every time, there's really no
>> need for this setting to be exposed.
> If we shouldn't expose it... Lets get rid of it...
> You added the param in the fall 2012... If it hasn't
> been used correctly or can't be used correctly for
> all theses years... why does it exist?

Because back then we didn't care about container awareness
enough to make it an initial part of the feature. We were
trying to address the problem of how to ensure that the
nfs4_client_id is unique. But clearly it was only half a
solution.

The module parameter still exists because the general rule
about these things is that module parameters are part of the
kernel API, and can't just be removed. If there's a process
for removing it, then I would agree to that now that we have
a sysfs API to manage the nfs4_client_id setting.


>> I do agree that it's long past time we should be setting
>> nfs4_client_id properly. I would rather see a udev script
>> developed (you, me, or Alice could do it in an afternoon)
>> first. If that doesn't meet the actual customer need, then
>> we can revisit.
> I'll address this in Trond's reply... 
> 
> thanks!
> 
> steved.

--
Chuck Lever