Re: [PATCH v1 2/3] NFSD: Implement the COPY call

["J. Bruce Fields" <bfields@fieldses.org>]

On Fri, Dec 04, 2015 at 10:49:24AM -0500, Anna Schumaker wrote:
> On 12/04/2015 10:45 AM, J. Bruce Fields wrote:
> > On Thu, Dec 03, 2015 at 03:55:35PM -0500, Anna Schumaker wrote:
> >> @@ -498,6 +499,22 @@ __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
> >>  }
> >>  #endif
> >>  
> >> +ssize_t nfsd_copy_range(struct file *src, u64 src_pos,
> >> +		       struct file *dst, u64 dst_pos,
> >> +		       u64 count)
> >> +{
> >> +	ssize_t bytes;
> >> +	u64 limit = 0x10000000;
> > 
> > Why that value?  Could I get a comment here?
> 
> Whoops!  I had a comment there at one point, but I must have deleted it :(.  That value is to cap copies to 256MB.

Could you include some justification for the choice of that particular
value?

> >> +	if (count > limit)
> >> +		count = limit;
> >> +
> >> +	bytes = vfs_copy_file_range(src, src_pos, dst, dst_pos, count, 0);
> > 
> > Sorry, I lost track of the copy discussions: does this only work on
> > filesystems with special support, or does it fall back on doing the copy
> > by hand?  Which filesystems (of the exportable filesystems) support
> > this?
> 
> The system call falls back on doing the copy by hand if there is no filesystem acceleration.

Is this practical?  It means a huge range in possible latency of the
single COPY call depending on filesystem.

I guess I can live with it and we can see if people run into problems in
practice.  But let's make sure this is documented.

--b.