Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!nuchat!steve
From: steve@nuchat.UUCP (Steve Nuchia)
Newsgroups: comp.unix.wizards
Subject: Re: Bad disk blocks
Message-ID: <404@nuchat.UUCP>
Date: Sat, 17-Oct-87 12:25:00 EDT
Article-I.D.: nuchat.404
Posted: Sat Oct 17 12:25:00 1987
Date-Received: Sun, 18-Oct-87 12:58:04 EDT
References: <9762@brl-adm.ARPA>
Organization: Public Access - Houston, Tx
Lines: 47
Summary: bad block handling is a wide-open area.

In article <9762@brl-adm.ARPA>, ebersman@sysu-0 (mt700) writes:
> I had always been told that UNIX had no means of flagging
> bad blocks on a disk so that they would not be allocated.

This is true - there is no unix-wide method for doing this.
Obviously, however, any given unix is going to have _some_ way
to deal with the problem.

Early systems allocated bad blocks to a "bad blocks" file,
which is similar to what many mainframe systems do.

A few systems may have been taught to leave bad blocks out
of the free list, but I haven't heard of them.

The remaining schemes involve sparing - mapping bad
{cylinders,tracks,sectors} to a reserved area at one
end of the disk.  This is done either in the controller
circuitry or in the driver logic.

DEC, in their newer drives, has a standard that involves
formatting a spare sector onto each track - these are 30+
sector/track drives, so the overhead is not onerous.  I don't
know how well any particular unix deals with this standard,
unless it is implemented in the controllers (likely).

> However, I was reading the V.2 man pages on our system,
> and bumped into addbad(1LM) and bdblk(1M).
> We are running on PDP 11/74's with RP05 disk drives.
> Has anyone out there used these functions? Will they

My client's 11/73 is running 2.9, and at least in that
one case the bad blocks are handled in the (overly) smart
controller board, so the driver sees a clean (and
distorted) drive.  I don't know what sysV does, but
at a guess (and that's all most of us can do - anything
this vendor-dependant in unix is almost never documented)
I'd say it is marking blocks bad in whatever place the
controller or driver is looking for that info, and it is
a pretty safe bet it will work, modulo the effect that
moving the sparing quantum will have on the filesystem.
Unless you can confirm that you are working with a _sector_
sparing scheme you should plan on rebuilding the involved
filesystems from backup.
-- 
Steve Nuchia	    | [...] but the machine would probably be allowed no mercy.
uunet!nuchat!steve  | In other words then, if a machine is expected to be
(713) 334 6720	    | infallible, it cannot be intelligent.  - Alan Turing, 1947