On Sat, 28 Aug 2021, Christoph Hellwig wrote:

On Thu, Aug 26, 2021 at 02:28:15PM +1000, NeilBrown wrote:

quoted

This patch also moves the nfsfh.h from the include/uapi directory into
fs/nfsd.  I can find no evidence of it being used anywhere outside the
kernel.  Certainly nfs-utils and wireshark do not use it.

That sounds fine, but I'd split this into a separate patch.

Thanks for your review.  Your suggestions seem appropriate.
I'll send out revised patches in a couple of days.

NeilBrown

quoted

fh_base and fh_pad are occasionally used to refer to the whole
filehandle.  These are replaced with "fh_raw" which is hopefully more
meaningful.

I think that kind of cleanup should also be a separate patch.  That
being said as far as I can tell fh_raw is only ever used in context
where we can just pass a void pointer.  So just giving the struct
for the "new" file handle after fh_size a name and passing that
would be much cleaner than a union with a char array.

quoted

I found
 https://www.spinics.net/lists/linux-nfs/msg43280.html
 "Re: [PATCH] nfsd: clean up fh_auth usage"
from 2014 where moving nfsfh.h out of uapi was considered but not
actioned. Christoph said he would "do some research if the
uapi <linux/nfsd/*.h> headers are used anywhere at all".  I can find no
report on the result of that research.  My own research turned up
nothing.

I can't remember doing much of research, and certainly not of finding
anything.

quoted

-	memcpy((char*)&fh.fh_handle.fh_base, f->data, f->size);
+	memcpy((char*)&fh.fh_handle.fh_raw, f->data, f->size);

Indedpendnt on what we're going to pass here, I don't think we
should need cast like this one (there are a few more).