Re: [PATCH 00/10] nfsd:refactor nfsd4_create_file()

On Wed, Jul 1, 2026, at 9:34 PM, NeilBrown wrote:
> nfs4_create_file() duplicates knowledge about opening a file which
> exists in the VFS, mostly in lookup_open().  It does use dentry_create()
> which shares some code, but there is more code that could be shared.
>
> The nfsd code doesn't get some details quite right, particularly patch
> 05 shows this.
>
> I hope to introduce a new VFS interface which encapulates more of what
> nfsd needs and shares more code with lookup_open().  I particularly want
> this as it will simplify some changes to locking rules that I am working
> on.
>
> This series re-arranges the nfsd code to get it ready for switching to
> the new interface.  Once that interface lands we can then switch over
> fairly easily.  Hopefully the series helps clarity even before that
> happens.

A couple of high-level remarks (and full disclosure, I haven't looked
closely at these yet):

- Sashiko found something remarkable in 9 of the 10 patches in this
  series: https://sashiko.dev/#/patchset/20260702014000.3397240-1-neilb%40ownmail.net

  Which I commend to you for study ;-)

- We'll need to decide how to merge these given the other related
  series you have in flight that targets VFS code that NFSD consumes.
  Are these two series truly independent?


-- 
Chuck Lever