FYI. Comments, especially on the RDMA portions, are most welcome. Tom.
> ---------- Forwarded Message ---------- >Date: Mon, 10 Sep 2007 13:41:55 -0400 >To: [EMAIL PROTECTED] >From: "Talpey, Thomas" <[EMAIL PROTECTED]> >Subject: [NFS] [PATCH 00/19] NFS/RDMA client support >List-Id: "Discussion of NFS under Linux development, interoperability, > and testing." <nfs.lists.sourceforge.net> >List-Archive: <http://sourceforge.net/mailarchive/forum.php?forum_name=nfs> >List-Post: <mailto:[EMAIL PROTECTED]> >List-Help: <mailto:[EMAIL PROTECTED]> >List-Subscribe: <https://lists.sourceforge.net/lists/listinfo/nfs>, > <mailto:[EMAIL PROTECTED]> >Sender: [EMAIL PROTECTED] > >The following 19 messages contain a patch series to implement the >fully integrated NFS/RDMA client into kernel 2.6.23-rc5 + NFS_ALL. >They also integrate cleanly into the current <git://linux-nfs.org/nfs-2.6>. > >I would like to see them considered for inclusion into 2.6.24, most >especially the first 14 which are purely infrastructure related. > >The patches are sequenced into the following functional groupings: > >01-02 implement RPCBIND netid's in each transport, instead of being >decided by the rpcbind client before sending. This change also corrects >the problem that IPv6 netid's were not supported. > >03-05 "invert" the parsing of NFS mount options in kernel to use the >existing kernel-private nfs_parsed_mount_options structure for NFS[234], >instead of continuing to use the legacy nfs_mount_data in both kernel >and user space. Coupled with the new string-based NFS mount API, this >allows extension of NFS mount options with no required changes to user >space. It's needed for adding RDMA, but may also be useful for other >NFS-related projects (such as fscache?). > >06 adds a flag to the xdrbuf to permit the rdma transport to marshal >data appropriately. > >07-09 implement dynamic RPC transport registration, and logically >reorganize the socket support as built-in TCP/UDP transports. These >patches originated with Chuck Lever's transport switch work. > >10-11 rearrange a few sockets-specific RPC transport definitions into >their own logical components. This is in preparation for adding the first >new RPC transport. > >12-14 change the way RPC transports are selected from a raw IP >protocol to a new RPC-layer identifier. Also #14 allows NFS to >accurately print the "proto=" argument via /proc/mounts. > >15-19 implement the new RPC RDMA transport: > > 15 declares the RPC/RDMA protocol, RPC RDMA transport definitions, >and configuration option. > > 16 adds support for the "-o rdma" option to string-based NFS mounts > > 17 adds the core RPC RDMA transport switch implementation. It has >stubs for the RDMA API below it, and the RPC/RDMA protocol marshaling >used when sending and receiving RPC messages. > > 18 implements the RPC message handling and connection management. > > 19 implements the kernel RDMA verbs interface. > >Since this version is fully integrated with the new string-based NFS >kernel mount API, it requires the latest nfs-utils mount.nfs command, >which must be invoked with the new "-i" flag. Additionally, until the >server implements the necessary rpcbindv3 support, the RDMA port >number must be provided in the mount command line. Currently, most >NFS/RDMA servers are listening on 2050, this is likely to change. > > mount -i [-t nfs4] -o rdma,port=2050 server:/filesystem /mountpoint > >The core net/sunrpc/xprtrdma files are little changed from the July >RFC Patches, except to fit into the above infrastructure, fix two >issues in corner case RDMA chunk marshaling, and to correct kernel >coding style nits. > >Comments on any and all issues are welcome. > >Tom. > >------------------------------------------------------------------------- >This SF.net email is sponsored by: Microsoft >Defy all challenges. Microsoft(R) Visual Studio 2005. >http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ >_______________________________________________ >NFS maillist - [EMAIL PROTECTED] >https://lists.sourceforge.net/lists/listinfo/nfs > ---------- End of Forwarded Message ---------- _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
