Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!decwrl!shlump.nac.dec.com!shodha.dec.com!alan
From: alan@shodha.dec.com ( Alan's Home for Wayward Notes File.)
Newsgroups: comp.unix.ultrix
Subject: Re: uq0 being reset
Summary: Some comments on Forced Errors.
Keywords: ultrix, microvax, rd53, hard error
Message-ID: <709@shodha.dec.com>
Date: 13 Feb 90 16:28:42 GMT
References: <3888@ucrmath.UCR.EDU> <9648@cbmvax.commodore.com>
Distribution: na
Organization: Digital Equipment Corp. - Colorado Springs, CO.
Lines: 130

In article <9648@cbmvax.commodore.com>, grr@cbmvax.commodore.com (George Robbins) writes:
> In article <3888@ucrmath.UCR.EDU> russ@mays.ucr.edu () writes:
> > We have a VAXstation II running Ultrix 2.0 with
> > a TK50, and two RD53 drives (one recently added).
> > From time to time console messages appear saying
> > 	Force Error Modifier set LBN ......
> > 	ra1g: hard error sn .....
> 
> It is important to understand that these messages are basically *fatal* -
> meaning that you need to take action as soon you see them.  It is probably
> an indication that either your drive wasn't adequately formatted/tested
> initially or that it is picking up a new errors.

	One of the features of the Digitial Storage Architecture
	(DSA) is that it tries to provide applications a view
	of disks that make them appear to be error free.  It does
	this mapping bad sectors to good ones.  Any initially bad
	sectors are mapped when the disk is formatted.  For errors
	that occur after formatting there are parts of the architecture
	that describe what is to be done.

	For this commentary I'll call the process Bad Block Replacement
	(or BBR).  There two kinds of BBR, static and dynamic.  Pre-V2.0
	version of ULTRIX and BSD 4.2 (and probably 4.3) do static BBR.
	If a bad block appeared and had to be fixed you booted a stand-
	alone program (rabads I think) that would let you scan the disk	
	would do the BBR for you.  Dynamic BBR has been supported by
	every version of ULTRIX since V2.0 and some disk controllers.

	The UDA50, KDA50 and KDB50 disk controllers will report a bad
	block to the host and expect the host to perform BBR.  The
	RQDX3 and HSC family will do the BBR themselves.  Part of the
	BBR process is to attempt to read the block many times in order
	to get a good copy of the data.  If the attempt fails then the
	original copy of the data is written to a replacement block and
	a bit is set in the block header.  This is the "Forced Error"
	referred to in the error message.  The block is good, but the
	data is corrupted from what it should have been.  Rather than
	gloss over it, the drivers force an Input error when the block
	is accessed.  The bit gets cleared when it is written to.

	In V2.0 and later is a program called radisk(8) that has options
	to scan for bad blocks, clear forced errors and start the BBR
	algorithm for a specific block or set of blocks (more on this
	one later).  The command to clear a forced error is:

		radisk -c LBN length special

		LBN is the logical block number of where the forced
		error is.  The length is generally 1, but if you have
		set of sequential forced errors you can get them all
		at once.  The last argument is the special device file
		for the disk.

	NOTE:  Radisk should only be run with the system single user.
	This is a documented restriction of the program.

	The scan operation tells the controller to scan the disk and
	doesn't transfer any of the data back to host.  This makes it
	faster than doing something like a dd(1) to read every block.
	The command is:

		radisk -s LBN length special

	If you want to scan the entire disk you can use:

		radisk -s 0 -1 special

	and radisk(8) will figure out the length.  The command to
	force BBR is:

		radisk -r LBN special

	The algorithm doesn't automatically replace a block, but
	execises it to make sure that it is bad.  If the block isn't
	bad then it won't replace it.

> 
> Typically, after getting a hard error, you want to do a backup, address
> the error condition and the restore the filesystem.  You can use something
> like "tar cvf /dev/null /mount_point" to try to figure out which file(s)
> the bad spot(s) are in, if you care.

	Once you've cleared a Forced Error on a replaced block you
	need to determine if the block was important.  George's
	suggestion is ok, but if know the block numbers and can
	translate them into blocks numbers within the partition
	there are simpiler ways of finding the file.

	First identify where the block is:

		icheck -b block-number special

	Icheck(8) can take a list of block numbers and identify
	where the blocks are.  It will say whether the block is
	part of the inode list (and which inodes), a data block
	of a file, a free block, a superblock (or backup superblock),
	etc.  If the block belongs to a file you can track down the
	file name by the inode number with:

		ncheck -i inode-number special

	This can be slow, so if you can mount the file system another
	method is to use ls and grep:

		ls -Rli | grep inode-number

	Once you know the file you can replace it with a good copy of
	it from a backup or the distribution (or other system).  Some-
	times it will be a file that is easily recreated (object file
	for example).

	If a block of inodes is bad you'll have to determine if any
	of them are used.  Generally for this I use fsck so I can
	repair any damage that there is.  Sometimes the damage will
	bad enough that it's simplier to restore from a backup.
> 
> If the bad block(s) are in inodes or other unpleasant spots, your system may
> crash when accessing the mounted filesystem or the filesystem may becomre
> more corrupt.  

	For this reason its a good idea to avoid mounting the file
	system until you know where the problem is.

> George Robbins - now working for,     uucp:   {uunet|pyramid|rutgers}!cbmvax!grr
> but no way officially representing:   domain: grr@cbmvax.commodore.com
> Commodore, Engineering Department     phone:  215-431-9349 (only by moonlite)

-- 
Alan Rollow				alan@nabeth.enet.dec.com