We have an QM8790 Switch utilized in an HPC cluster. It seems that with the Mellanox infiniband the they are using CIFS dummy drivers due to CIFS not beein supported to run alongside the infiniband.
However we have now tried to get NFS working with extended ACLs and a windows environment in parallel, but have not managed to get this working and i want to check if anyone has experience or knows if it possible to replace the CIFS dummy drivers with actual working CIFS drivers.
And potentially what implications this could bring.
Adding inn the root cause post from redhat
Root Cause
Using the mellanox dummy cifs module instead of Red Hat Enterprise Linux built-in module. As per Nvidia/Mellanox community usage of CIFS along with Mallanox OFED is not supported, but consult this situation with Mellanox/Nvidia support and if it’s possible to remove the module and use the Red Hat Enterprise Linux built-in.
If you look at NVIDIA KB, as CIFS is not supported with OFED, there would be no way to make it successful.
Please refer to NVIDIA KB Article number : 000005052 (CIFS Support with MLNX_OFED)
==========================================================================
In the ‘General Support’ section of the MLNX_OFED release notes, under the ‘Unsupported Functionalities/Features/NICs’ section, the following statement is made:
CIFS (Common Internet File System) module installation
This statement is accurate, and no other statement should be made to the customer.
Engineering is not road mapping any support for CIFS, and no effort will be made to accommodate this functionality.
There have been cases where old/outdated information has been provided to the customer, leading to confusion.
Information in past community posts regarding this topic do not override this statement in the official MLNX_OFED release notes, and cannot be used for case resolution.
There is no way to support CIFS with MLNX_OFED. If the customer requires the use of CIFS, they need to use inbox drivers for their distribution, and can not use MLNX_OFED.
We are not able to assist customers with rebuilding kernel modules. We are unable to assume liability for any issues which arise as a result of such action.