NfsRdmaClient/Home
From Linux NFS
Chucklever (Talk | contribs) (→Submitting patches) |
Chucklever (Talk | contribs) (Add link for ffsb) |
||
(5 intermediate revisions not shown) | |||
Line 22: | Line 22: | ||
* cthon04: git://git.linux-nfs.org/projects/steved/cthon04.git | * cthon04: git://git.linux-nfs.org/projects/steved/cthon04.git | ||
* xfstests: git://oss.sgi.com/xfs/cmds/xfstests | * xfstests: git://oss.sgi.com/xfs/cmds/xfstests | ||
+ | * nfs-o-meter: git://git.linux-nfs.org/projects/dros/nfsometer.git | ||
* [http://codemonkey.org.uk/projects/fsx/ fsx] | * [http://codemonkey.org.uk/projects/fsx/ fsx] | ||
* [http://www.iozone.org/ IOzone] | * [http://www.iozone.org/ IOzone] | ||
* [http://freecode.com/projects/fio fio] | * [http://freecode.com/projects/fio fio] | ||
* Multi-process Linux kernel builds | * Multi-process Linux kernel builds | ||
+ | * [http://downloads.sourceforge.net/project/ffsb/ffsb/ffsb-6.0-rc2/ffsb-6.0-rc2.tar.bz2 ffsb] | ||
Typically, for testing, the NFS/RDMA server exports a tmpfs or ram disk-based local FS to drive the transport fast enough. | Typically, for testing, the NFS/RDMA server exports a tmpfs or ram disk-based local FS to drive the transport fast enough. | ||
Line 41: | Line 43: | ||
General advice for preparing upstream kernel or nfs-utils patches: see [https://www.kernel.org/doc/Documentation/SubmittingPatches SubmittingPatches]. | General advice for preparing upstream kernel or nfs-utils patches: see [https://www.kernel.org/doc/Documentation/SubmittingPatches SubmittingPatches]. | ||
- | It is strongly recommended that patches be compile-tested, tested with sparse, and run by scripts/checkpatch.pl before submission. Patches should be tested with the tools above before submission, though that's not needed if you are posting only for review | + | It is strongly recommended that patches be compile-tested, tested with sparse, and run by scripts/checkpatch.pl before submission. Patches should be tested with the tools above before submission, though that's not needed if you are posting only for review. |
- | Submit patches to linux-nfs@vger.kernel.org for review. | + | ==== Review ==== |
+ | |||
+ | Review is done via e-mail. In the kernel source tree, see Documentation/email-clients.txt for important advice on how to use e-mail to submit patches. | ||
+ | |||
+ | Label your patches "[RFC ... ]" or "untested" "DO NOT APPLY" or "for review only" so maintainers know the patch is for review and may not be broadly tested yet. | ||
+ | |||
+ | Submit patches to linux-nfs@vger.kernel.org and linux-rdma@vger.kernel.org for review. | ||
+ | |||
+ | If you have reviewed a posted patch, reply with a "Reviewed-by:". If you've tested it, reply with a "Tested-by:" | ||
+ | |||
+ | ==== Merge request ==== | ||
+ | |||
+ | When the community feels a client-side patch is ready to be merged, submit the patches again (without "for review" disclaimers) to linux-nfs@vger.kernel.org and linux-rdma@vger.kernel.org, and cc: Anna.Schumaker@netapp.com. Anna will collect ready patches for Trond to merge upstream. | ||
+ | |||
+ | We are still developing a mechanism for requesting server-side fixes be merged. | ||
=== Engineering Notes === | === Engineering Notes === | ||
* [[NfsRdmaClient/MemRegModes|Simplifying memory registration modes]] | * [[NfsRdmaClient/MemRegModes|Simplifying memory registration modes]] |
Latest revision as of 20:54, 14 October 2014
Contents |
Support for Linux NFS/RDMA Client upstream
Goals
Provide maintenance for NFS/RDMA and RPC/RDMA client-side code in the Linux kernel. Support continuous testing of NFS/RDMA feature in upstream Linux kernels. Provide enhancements (see below). Identify resources for supporting server-side code.
Bug tracking
Upstream bug tracking is here.
Feature planning
Our Pivotal Tracking project lists individual work items and defines several long-term arcs. At a high level, these include:
- Broad support for a variety of RDMA-capable hardware
- Performance and scalability enhancements
- Support for NFSv4.1
- Observability features that enable distributor support teams and customers to diagnose and address problems
Developer tools
For anyone working directly on Linux NFS/RDMA or RPC/RDMA, please consider the following tools for validating your work.
- cthon04: git://git.linux-nfs.org/projects/steved/cthon04.git
- xfstests: git://oss.sgi.com/xfs/cmds/xfstests
- nfs-o-meter: git://git.linux-nfs.org/projects/dros/nfsometer.git
- fsx
- IOzone
- fio
- Multi-process Linux kernel builds
- ffsb
Typically, for testing, the NFS/RDMA server exports a tmpfs or ram disk-based local FS to drive the transport fast enough.
Asking for help
You can find Linux NFS developers at linux-nfs@vger.kernel.org, or in the linux-nfs chatroom at oftc.net.
Submitting patches
Clone the upstream Linux kernel with:
$ git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git
General advice for preparing upstream kernel or nfs-utils patches: see SubmittingPatches.
It is strongly recommended that patches be compile-tested, tested with sparse, and run by scripts/checkpatch.pl before submission. Patches should be tested with the tools above before submission, though that's not needed if you are posting only for review.
Review
Review is done via e-mail. In the kernel source tree, see Documentation/email-clients.txt for important advice on how to use e-mail to submit patches.
Label your patches "[RFC ... ]" or "untested" "DO NOT APPLY" or "for review only" so maintainers know the patch is for review and may not be broadly tested yet.
Submit patches to linux-nfs@vger.kernel.org and linux-rdma@vger.kernel.org for review.
If you have reviewed a posted patch, reply with a "Reviewed-by:". If you've tested it, reply with a "Tested-by:"
Merge request
When the community feels a client-side patch is ready to be merged, submit the patches again (without "for review" disclaimers) to linux-nfs@vger.kernel.org and linux-rdma@vger.kernel.org, and cc: Anna.Schumaker@netapp.com. Anna will collect ready patches for Trond to merge upstream.
We are still developing a mechanism for requesting server-side fixes be merged.