This is version 3 (I think) of my patches to improve NFS writeaback.

Changes:
 - the code for adding legacy values to /proc/vmstat was broken.
   I haven't taken the approach that Michal and Jan discussed but
   a simpler (I hope) approach that just seq_puts() the lines at an
   appropriate place.

 - I've modified the handling of PF_LOCAL_THROTTLE - sufficiently that
   I dropped Jan's reviewed-by.
   Rather than invoking the same behaviour as BDI_CAP_STRICTLIMIT,
   I now just use the part I needed.
   So if the global threshold is not exceeded, PF_LOCAL_THROTTLE tasks
   are not throttled.  This is the case for normal processes, but not
   when BDI_CAP_STRICTLIMIT is in effect.
   If the global threshold *is* exceeded, only then to we check the
   local per-bdi threshold.  If that is not exceeded then
   PF_LOCAL_THROTTLE again avoid any throttling.  Only if both the
   thresholds are exceeded are these tasks throttled.
   I think this is more refletive if what we actually need.


Thanks,
NeilBrown