Subject: Current fsck trashes old filesystems
To: Brian D. Carlstrom <>
From: Ivan Vazquez <>
List: current-users
Date: 07/21/1994 02:02:31
>>>> Michael L VanLoon writes:

>     > P.S. Current (July 12) fsck also freaked out on my SCSI /usr
>     > partition, which had been working fine on the May 1 system, after
>     > it had been working under the July 12 kernel fine for a few hours
>     > and I rebooted.  It complained about a corrupted directory, but
>     > only gave the name as "DIR=?" (it also complained about
>     > lost+found being corrupted).  Then, after complaining about these
>     > two dirs, each time it promptly proceeded to get a segmentation
>     > violation.  I rmdir'd the lost+found dir, which made it happy,
>     > but I couldn't even tell what the other dir was that it was
>     > talking about.  And every time it hit that dir it died with a
>     > segv.  I was only able to get that drive to pass fsck clean by
>     > fsck'ing it a couple times under my alternate IDE boot drive
>     > running the May 1 kernel and binaries (including fsck).  The
>     > older fsck called that "DIR=".  It seems the 4.4 version of fsck
>     > has a null pointer dereferenced somewhere if it gets strange dir
>     > entries, that didn't happen under the older fsck.  Sorry I can't
>     > provide any more detailed debugging info since the machine wasn't
>     > exactly in a stable state at the time.  Is this a known problem,
>     > or is my machine just screwey?

The pattern I have gone through is:

- Bring a machine up to 0.9a (May 2) using floppies,
- Download and extract all the current binaries,
- reboot 
- during initial fsck it goes into the " DIR=?" problem and starts
  corrupting the fs.

This current problem makes it require extraordinary measures to bring
a new machine up to current because the only existing floppy images
are 0.9.  Would someone in the know care to make some new ones
available?  They would allow people to install Current (and test it).

People running Current may be able to test this by booting off of the
distributed (0.9) floppies and newfs'ing a partition, then doing lots
of file writing and fscks on that partition under Current.  I hope
this tickles the bug.