On Jun 13, 2008, at 3:15 PM, Danhua Shao wrote:
> Hi, Sam
>
> I am working on adding new DTrace probes for operations NFSv4.1 and
> am confused about setup their provider: Add all new probes into a
> new provider for NFSv4.1 or add them into the current NFSv4.0
> provider?
>
> I just compare the source code for operations in nfs4_srv.c and in
> nfs41_srv.c.
>
> Some operations share code between the two versions. For example,
> in nfs41_srv.c
> static void
> mds_op_readdir_free(nfs_resop4 *resop, compound_node_t *cn)
> {
> /* Common function used for NFSv4.0 and NFSv4.1 */
> rfs4_op_readdir_free(resop);
> }
>
> This function reuse the operation definition in NFSv4.0
>
> But other operations have similar name and arguments, but have
> different implementation code.
>
> in nfsv41_srv.c:
> static void
> mds_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
> compound_node_t *cn)
>
> in nfsv4_srv.c
> static void
> rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req
> *req,
> struct compound_state *cs)
>
> So, if the new probes for NFSv4.1 are put under the NFSv4.0 provider
> ('nfsv4'), it is good for the shared code, but be confused for
> operations with different implementation.
>
> If the new probes for NFSv4.1 are under a new provider, for example,
> nfsv41, it is good to distinguish the different implementation for
> operations in the two NFS versions, but there may be some redundant
> probes for shared code operations.
>
> My favorite is the second solution, setup a new provider. But I am
> not sure if it is fit the requirement for the users of these probes.
Since protocol state is separate between 40 and 41 my preference would
be the second
solution, *but* then i'm a developer.. and that is my main focus right
now.. as an end-user
wanting to know basic v4 information (which clients, which files, IO
performance), i don't care about this silly minor version thing..
The 4.1 code is in flux; but i would recommend that the existing NFSv4
probes be placed in the corresponding mds_op* functions.
Robert.
