On Thu, 2024-02-15 at 11:02 -0500, Chuck Lever wrote: > On Wed, Feb 14, 2024 at 05:35:01PM -0500, trondmy@xxxxxxxxxx wrote: > > From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx> > > > > The main point of the guarded SETATTR is to prevent races with > > other > > WRITE and SETATTR calls. That requires that the check of the guard > > time > > against the inode ctime be done after taking the inode lock. > > > > Furthermore, we need to take into account the 32-bit nature of > > timestamps in NFSv3, and the possibility that files may change at a > > faster rate than once a second. > > > > Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx> > > LGTM! Applied to nfsd-next. Sigh... Unfortunately this patch wasn't sufficient. I just found a rather nasty regression in the same area of code. v2 is on its way, with the fix for the regression as a first standalone patch. -- Trond Myklebust Linux NFS client maintainer, Hammerspace trond.myklebust@xxxxxxxxxxxxxxx
