-
Notifications
You must be signed in to change notification settings - Fork 49
Phone Conference, December 15th 2011
A request has been made to IBM for having GPFS libraries and header files so that it would be possible to compile FSAL_GPFS without installing the whole of GPFS. This is critical in order to have comprehensive compilation tests, including all possible FSALs.
How should the DSs and the MDS communication ? Suggestion was made to have stateids propagated from MDS to DSs but it sounds like it could be a way to produce race condition in-between NFSv4 request and inter servers communication. Using the filehandle passed via NFSv4.1 layout operation is probably safer for exchanging information. Part of the FH could be encrypted so that only DSs can read it (a single initial handshake is required to exchange encryption keys). This approach is race-free. It is also kind of similar to kerberos5 sign-off mechanism.
Structures used by Adam are different from those in the kernel, preventing from reusing already existing code. On the other hand, the semantics are pretty different in NFS-Ganesha and in the kernel, why should we keep exactly the same logic in both cases ? An opportunity seems to exists : talking both approaches and keep the best of both. This discussion is not over, more will probably come in later phone conferences. On his side, LinuxBox will integrate elements provided my Boaz to the pNFS code.
Boaz announced that a FSAL_EXOFS was about to come.
The release 1.3.0 of NFS-Ganesha will be delivered before Christmas. It contains Matt's improvement on readdirs (by using AVL tree) and the integration of CIDR aware code to ease writing access lists in export entries.
The need for a security dedicated list is clear. A new (private) mailing list will be created. It will have restricted access and will be fully moderated.