I've been experimenting with what happens when a normal user fills the
disk on an LFS (fixes not yet integrated into CVS, but which I hope 
to get into 1.5 if I can clean up some details...at
  ftp://ftp.hhhh.org/pub/perseant/lfs/diskfull.patch 
if anyone wants to see for themselves, otherwise "don't try this at
home").

My question is, given that LFS has dynamically changing amounts of
low-level filestore metadata (in particular, inode blocks and
partial-segment summaries), what should the disk usage numbers reflect?

In the patch I mention above I estimate the amount of space this metadata
will take if the whole disk were full based on the part that *is* full,
and subtract that estimate from the total partition size to fill in
f_blocks.  f_bfree and f_avail are similarly estimated.

This seems to work pretty well in that the "df" avail and used numbers
look right, but has the disquieting side-effect that the size of the disk
seems to change over time.  While this might be "correct" I don't like it
much when I put on my system administrator hat.

Thoughts?
						Konrad Schroder
						perseant@hhhh.org