Skip to contentSkip to conversation

NFS server development update

DISCUSSION J. Bruce Fields 1 participant 2 messages Sep 27, 2007

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

These days Neil Brown and I are maintaining the NFS server together, and I'm currently (as of the last few weeks) tracking the to-be-submitted patches.

I have limited time and am not expert on all of the relevant code, so I depend on your help! I'm still experimenting a bit with the process (suggestions welcomed), but most of the following should be obvious to anyone who's been working on the kernel for a while.

I keep all the pending NFS and sunrpc server patches that I believe are done in the "nfs-server-stable" branch here[1]:

git://linux-nfs.org/~bfields/linux.git nfs-server-stable

I also include nfs-server-stable in the "for-mm" branch, which is normally included in Andrew's -mm kernels automatically, and which may sometimes also contain other works-in-progress.

I'll try to submit urgent[2] patches to upstream immediately. All the other patches in nfs-server-stable will be submitted in one batch just after each new kernel release (that's about once every two months).

If you want patches included, email them to me, one patch per message, in the format described in[3]

http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt

and also cc: [email protected], Neil Brown <[email protected]>, and anyone you consider particularly qualified to review your work. Also, please state clearly that you're asking me to apply your patches; otherwise I may just assume you're asking people for early feedback on work in progress (which is also encouraged, of course). If you get no response and your work doesn't show up in my git tree, I may have forgotten; please remind me. If I ignore something urgent, please try Neil, Andrew, lkml, and/or Linus as necessary.

If you're fixing a mistake in a patch I already have, I'd usually prefer an incremental patch as opposed to a complete replacement.

If you have large changes you'd like included in the next kernel, try to submit them well in advance of the release of the current kernel. (For example, with 2.6.23 about to be released, it may already be too late for major changes to be included in 2.6.24.)

I'll follow up with a summary of the patches I currently have.

--b.

Notes:

[1] My suggestion for how to access my git tree:
1. Install git version 1.5.0 or later
2. If you haven't already, get Linus's tree with:
git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git
cd linux-2.6.git
3. Tell git about my tree with:
git remote add bfields git://linux-nfs.org/~bfields/linux.git
git fetch bfields
4. View current server patches with:
gitk origin..bfields/nfs-server-stable
4. Update as necessary using:
git fetch bfields

[2] A patch will be more likely to be treated as "urgent" if it is obviously correct and if it fixes a severe problem (for example, one that corrupts data or compromises security). As a release approaches, and the time remaining to test new patches decreases, the standard for such patches will become harder to meet.

[3] If you're new to the process, the contents of

http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt

and Documentation/SubmittingPatches are useful, but possibly also overwhelming. The important points are:

- Each patch should make one set of logically related changes and come with a comment that explains what the patch does and why.
- No patch should introduce a regression: after applying the first n patches of a series (for any n), the kernel should still compile and run. It's not OK to introduce a bug in one patch and then fix it in some later patch; instead, you need to figure out how to reorder your changes.

Our goal is to improve the kernel through a long series of small, incremental, easy to understand, and individually testable changes.

#1 lore raw
Quoted 1 line · J. Bruce Fields

I'll follow up with a summary of the patches I currently have.

... but this time around that's nothing major--mostly just small bugfixes and cleanup, including 64 bit inode support from Peter Staubach and some preparation for supporting callbacks (hence delegations) over krb5.

One outstanding question is what to do about the server RDMA and transport-switch code, not yet included below. It would nice to finally see that start going in, but it's a lot of code to add this late.

--b.

Quoted 1 line

From git://linux-nfs.org/~bfields/linux.git nfs-server-stable:

Andrew Morton (1):
nfsd warning fix

Christoph Hellwig (1):
nfsd: fix horrible indentation in nfsd_setattr

Dr. David Alan Gilbert (1):
knfsd: Add source address to sunrpc svc errors

J. Bruce Fields (14):
nfsd: tone down inaccurate dprintk
nfsd: remove unused cache_for_each macro
knfsd: delete code made redundant by map_new_errors
knfsd: cleanup of nfsd4 cmp_* functions
knfsd: demote some printk()s to dprintk()s
knfsd: nfs4 name->id mapping not correctly parsing negative downcall
knfsd: spawn kernel thread to probe callback channel
knfsd: move nfsv4 slab creation/destruction to module init/exit
knfsd: fix callback rpc cred
knfsd: remove code duplication in nfsd4_setclientid()
svcgss: move init code into separate function
knfsd: let nfsd manage timing out its own leases
knfsd: don't shutdown callbacks until nfsv4 client is freed
knfsd: nfsv4 delegation recall should take reference on client

Peter Staubach (1):
knfsd: 64 bit ino support for NFS server