Re: RFC: Marking file system va_filerev style

From: Rick Macklem <rick.macklem_at_gmail.com>
Date: Thu, 09 Jan 2025 01:45:47 UTC
On Wed, Jan 8, 2025 at 5:37 PM Konstantin Belousov <kostikbel@gmail.com> wrote:
>
> On Wed, Jan 08, 2025 at 04:39:36PM -0800, Rick Macklem wrote:
> > Hi,
> >
> > Back in Sept. Richard reported a problem with a
> > FreeBSD server serving NFSv4 to Linux clients,
> > where the file attributes would get messed up
> > sometimes. (I can't recall if he posted me or a mailing
> > list.)
> >
> > Anyhow, He tried a patch I gave him that told the
> > Linux client that the Change attribute (va_filerev)
> > increments for each change.
> > --> This appears to have fixed the problem.
> >
> > However, this patch should really report what the
> > exported fs does w.r.t. va_filerev and not just guess
> > that it increments it.
> >
> > So, the question is, how do I have file systems report
> > how they generate va_filerev?
> > My current thought is a new MNTK_xxx flag set by
> > the filesystem in its mount point to indicate it increments
> > it upon each change (which is what UFS and ZFS currently
> > does, I think?)
> >
> > Does this sound reasonable?
>
> Is this a per-mount property, or a generic filesystem-type property (for
> me, the nature of the indicated feature tends to mean per-type).  Then
> it might be that a VFCF_ flag is the more natural solution than MNTK_ one.
>
Yes, it is a file system property. Thanks for the suggestion.
(I'll come up with a patch and put it on phabricator.)

Thanks, rick