Re: still nfs problems [Was: Linux 2.6.37-rc8]

[Trond Myklebust <Trond.Myklebust@netapp.com>]

On Sat, 2011-01-08 at 11:49 -0500, Trond Myklebust wrote: 
> On Fri, 2011-01-07 at 13:11 -0600, James Bottomley wrote: 
> > On the other hand, the xdr routines, since they take the pages anyway,
> > could use a scatterlist approach to writing through the kernel mapping
> > instead of using vmap ... we have all the machinery for this in
> > lib/scatterlist.c ... it's not designed for this case, since it's
> > designed to allow arbitrary linear reads and writes on a block
> > scatterlist, but the principle is the same ... it looks like it would be
> > rather a big patch, though ... 
> 
> The following alternative seems to work for me, but has only been
> lightly tested so far. It's a bit large for a stable patch, but not too
> ungainly.
> 
> It modifies xdr_stream, adding the ability to iterate through page data.
> To avoid kmap()/kunmap(), it does require that pages be allocated in
> lowmem, but since the only use case here is when using page arrays as
> temporary buffers, that seems like an acceptable compromise.

...and here is an update which makes the whole process transparent to
the decoder. It basically teaches xdr_inline_decode() how to switch
buffers when it hits the end of the current iovec and/or page.

Cheers
  Trond
-----------------------------------------------------------------------------------