2021-07-15 13:17:15

by guy keren

Subject: nfs4.1 and nconnect - is this supported?

hi,

i wonder if the linux client's nfs nconnect feature was designed to
support NFS4.1 (or higher) versions? according to our experimentation,
the linux client seems to just alternate messages between the multiple
RPC/TCP connections, and does not seem to adhere to the NFS 4.1 protocol
requirement, that when using multiple connections, the client needs to
use BIND_CONN_TO_SESSION when trying to user a 2nd connection with the
same NFS4.1 session.

was this done on purpose? or is this configuration not supported by
linux client's 'nconnect'? or am i missing something?

thanks,
--guy

2021-07-15 14:44:42

by Rick Macklem

Subject: Re: nfs4.1 and nconnect - is this supported?

guy keren <[email protected]> wrote:
>hi,
>
>i wonder if the linux client's nfs nconnect feature was designed to
>support NFS4.1 (or higher) versions? according to our experimentation,
>the linux client seems to just alternate messages between the multiple
>RPC/TCP connections, and does not seem to adhere to the NFS 4.1 protocol
>requirement, that when using multiple connections, the client needs to
>use BIND_CONN_TO_SESSION when trying to user a 2nd connection with the
>same NFS4.1 session.
>
>was this done on purpose? or is this configuration not supported by
>linux client's 'nconnect'? or am i missing something?
Yep, you're missing something.

Snippet from RFC 5661 pg. 43:
If the client specifies no state
protection (Section 18.35) when the session is created, then when
SEQUENCE is transmitted on a different connection, the connection is
automatically associated with the fore channel of the session
specified in the SEQUENCE operation.

As such, BIND_CONN_TO_SESSION is only required to associate the
backchannel to the connection.

rick

thanks,
--guy

2021-07-15 20:28:14

by guy keren

Subject: Re: nfs4.1 and nconnect - is this supported?

On 7/15/21 5:43 PM, Rick Macklem wrote:
> guy keren <[email protected]> wrote:
>> hi,
>>
>> i wonder if the linux client's nfs nconnect feature was designed to
>> support NFS4.1 (or higher) versions? according to our experimentation,
>> the linux client seems to just alternate messages between the multiple
>> RPC/TCP connections, and does not seem to adhere to the NFS 4.1 protocol
>> requirement, that when using multiple connections, the client needs to
>> use BIND_CONN_TO_SESSION when trying to user a 2nd connection with the
>> same NFS4.1 session.
>>
>> was this done on purpose? or is this configuration not supported by
>> linux client's 'nconnect'? or am i missing something?
> Yep, you're missing something.
>
> Snippet from RFC 5661 pg. 43:
> If the client specifies no state
> protection (Section 18.35) when the session is created, then when
> SEQUENCE is transmitted on a different connection, the connection is
> automatically associated with the fore channel of the session
> specified in the SEQUENCE operation.
>
> As such, BIND_CONN_TO_SESSION is only required to associate the
> backchannel to the connection.
>
> rick

thanks for pointing this out, rick - i forgot about this feature.


does it mean that 'nconnect' cannot work with MACHCRED or SSV client
authentication then?

will it work with kerberos (which does authenticate the client machine
normally, as far as i know)?


thanks,

--guy