Re: [nfs-utils PATCH] nfs4id: a tool to create and persist nfs4 client uniquifiers

["Benjamin Coddington" <bcodding@redhat.com>]

On 8 Feb 2022, at 14:52, Steve Dickson wrote:

> On 2/8/22 11:22 AM, Benjamin Coddington wrote:
>> On 8 Feb 2022, at 11:04, Steve Dickson wrote:
>>
>>> Hello,
>>>
>>> On 2/4/22 7:56 AM, Benjamin Coddington wrote:
>>>> The nfs4id program will either create a new UUID from a random 
>>>> source or
>>>> derive it from /etc/machine-id, else it returns a UUID that has 
>>>> already
>>>> been written to /etc/nfs4-id.  This small, lightweight tool is 
>>>> suitable for
>>>> execution by systemd-udev in rules to populate the nfs4 client 
>>>> uniquifier.
>>>>
>>>> Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
>>>> ---
>>>>   .gitignore               |   1 +
>>>>   configure.ac             |   4 +
>>>>   tools/Makefile.am        |   1 +
>>>>   tools/nfs4id/Makefile.am |   8 ++
>>>>   tools/nfs4id/nfs4id.c    | 184 
>>>> +++++++++++++++++++++++++++++++++++++++
>>>>   tools/nfs4id/nfs4id.man  |  29 ++++++
>>>>   6 files changed, 227 insertions(+)
>>>>   create mode 100644 tools/nfs4id/Makefile.am
>>>>   create mode 100644 tools/nfs4id/nfs4id.c
>>>>   create mode 100644 tools/nfs4id/nfs4id.man
>>> Just a nit... naming convention... In the past
>>> we never put the protocol version in the name.
>>> Do a ls tools and utils directory and you
>>> see what I mean....
>>>
>>> Would it be a problem to change the name from
>>> nfs4id to nfsid?
>>
>> Not at all..
> Good...

I didn't orginally do that because I thought it might be confusing for 
some
folks who want `nfsid` to display their kerberos identity.  There's a BZ
open for that!

Do you think that's a problem?  I feel like it's a problem.

>> and I think there's a lot of room for naming discussions about
>> the file to store the id too:
>>
>> Trond sent /etc/nfs4_uuid
>> Neil suggests /etc/netns/NAME/nfs.conf.d/identity.conf
>> Ben sent /etc/nfs4-id (to match /etc/machine-id)
> Question... is it kosher to be writing /etc which is
> generally on the root filesystem?

Sure, why not?

> By far Neil suggestion is the most intriguing... but
> on the containers I'm looking at there no /etc/netns
> directory.

Not yet -- you can create it.

> I had to install the iproute package to do the
> "ip netns identify" which returns NULL...
> also adds yet another dependency on nfs-utils.

We don't need the dependency..this little binary is just a helper for a 
udev
rule.  Trond already wrote his version of this.  :)  This one's just 
trying
to be a little lighter (whoa, we don't need all of bash!).

> So if "ip netns identify" does return NULL what directory
> path should be used?

I think thats out of scope..  if udevd is running in a container, and 
has a
rule that uses this tool, then the container either is likely to have
already customized its /etc.  Or perhaps we can make the udev rule 
namespace
aware (I need to look into what's available to udev's rule environment 
when
it runs in a namespace).

> I'm all for making things container friendly but I'm
> also a fan of keeping things simple... So
> how about /etc/nfs.conf.d/identity.conf or
> /etc/nfs.conf.d/nfsid.conf?

Really, because its not a configuration.  Its value persistence, and we 
/really/
don't want people configuring it.

>>
>> Maybe the tool wants an option to specify the file?
> This is probably not a bad idea... It is a common practice

OK, I'll do that.

Ben