Subject: Recent LFS changes (warning)
To: None <>
From: Konrad Schroder <>
List: current-users
Date: 01/17/2000 18:37:24
Over the past week or so I've made several changes to the LFS code that
should make it more usable.  Most of these are addressed to run-time
problems: LFS should just work better now that they're fixed.  Two of
these problems were more serious, though, in that they could cause on-disk

If you are using an LFS you may want to tar (not dump_lfs) the contents of
your filesystem, and recreate it using newfs_lfs.

(If you're not using LFS, try it out!  But be aware that it is not exactly
rock-solid.  If you find any problems, please let me know ... preferably
using send-pr so the reports don't get lost when I get busy.)

Problem #1:

Kernels built from sources between 14 Dec 1999 and 16 Jan 2000 may suffer
from the problem that some removed files are left on the disk (removed
from the filesystem but not from the filestore).  This has the effect of
making your disk fuller than it should be, and given the problems LFS has
with full disks, this is not a good scenario.  (IIRC "dump" dumps the
whole filestore, which is why I recommend using "tar" to clean this
up...I'm not sure if "restore" will filter out files that have no name.)

Problem #2:

Under certain conditions ifile updates could be "unwritten" by the
cleaner.  The chief symptom of this problem is that at some later time,
after the cleaned segments are filled with new data, lfs_ifind can no
longer find the inode where it thinks it ought to be, giving a panic with
"lfs_ifind: dinode %u not found".  This is an old problem.  In this case
dump should work as well as tar (or better, you can dump without the fs
						Konrad Schroder