From jroberson@chesapeake.net Sat Mar 29 10:19:11 2003
Date: Fri, 28 Mar 2003 19:47:14 -0500 (EST)
From: Jeff Roberson <jroberson@chesapeake.net>
To: Kirk McKusick <mckusick@beastie.mckusick.com>
Cc: Robert Watson <rwatson@FreeBSD.org>, jeff@FreeBSD.org
Subject: Re: panic: softdep_disk_io_initiation: read 

Please try the attached patch.  This will fix it.  This happens if we try
to do a second read ahead attempt and there is only one contiguous block
on disk.  Ooops. :-/

Cheers,
Jeff

Index: vfs_cluster.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/vfs_cluster.c,v
retrieving revision 1.135
diff -u -r1.135 vfs_cluster.c
--- vfs_cluster.c       18 Mar 2003 08:45:23 -0000      1.135
+++ vfs_cluster.c       29 Mar 2003 00:20:51 -0000
@@ -266,10 +266,14 @@
                        lblkno += (rbp->b_bufsize / size);
                } else {
                        rbp = getblk(vp, lblkno, size, 0, 0, 0);
+                       lblkno += 1;
+                       if (rbp->b_flags & B_DELWRI) {
+                               bqrelese(rbp);
+                               continue;
+                       }
                        rbp->b_flags |= B_ASYNC | B_RAM;
                        rbp->b_iocmd = BIO_READ;
                        rbp->b_blkno = blkno;
-                       lblkno += 1;
                }
                if (rbp->b_flags & B_CACHE) {
                        rbp->b_flags &= ~B_ASYNC;


On Fri, 28 Mar 2003, Kirk McKusick wrote:

> We have traced this back to a point in the read-clustering code
> where it is changing a dirty BIO_WRITE buffer to be a BIO_READ
> buffer. I am guessing that it is being a bit too aggressive about
> trying to aggregate buffers together for reading and is accidentally
> picking up one that has not yet been written. Anyway, Jeff is now
> hard on the trail to track it down and I am hopeful that he will
> soon have a fix :-)
>
> 	~Kirk
>
> =-=-=-=-=-=
>
> Date: Fri, 28 Mar 2003 15:19:23 -0500 (EST)
> From: Robert Watson <rwatson@FreeBSD.org>
> To: kirk@mckusick.com, jeff@FreeBSD.org
> Subject: Re: panic: softdep_disk_io_initiation: read
> In-Reply-To: <Pine.NEB.3.96L.1030328144323.10348A-100000@fledge.watson.org>
> X-ASK-Info: Whitelist match
>
> Just had an identical "second panic" of the same sort.  It appears to be
> happening in the following scenario:
>
> sproing.gw.tislabs.com:
>    sproing:/data/p4/rwatson/trustedbsd/misc/kugidfw> !make
>    make
>    c++  -Wall -I/usr/local/include -I/usr/X11R6/include -g -c kruleedit.cc
>    g++ -o kugidfw main.o kugidfw.o kruleedit.o -L/usr/local/lib -lkdeui
>    -lkdecore -L/usr/X11R6/lib -lqt-mt -pthread -lugidfw
>    sproing:/data/p4/rwatson/trustedbsd/misc/kugidfw> ./k
> <hit tab and it hangs>
>
> cboss.gw.tislabs.com:
> <crash>
>
> This is the exactly the same scenario -- the moment I did a tab completion
> using tcsh on sproing.gw.tislabs.com.  I'm going to try it one more time
> without changing any settings, and then Jeff has asked me to try setting
> vfs.read_max=0, so I'll try it again with that.
>
> Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
> robert@fledge.watson.org      Network Associates Laboratories
>