Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 8 Jun 1998 09:36:38 -0400 (EDT)
From:      Andrew Gallatin <gallatin@cs.duke.edu>
To:        current@FreeBSD.ORG
Subject:   NFS / softupdates interaction problems?
Message-ID:  <199806081336.JAA06735@grasshopper.cs.duke.edu>

next in thread | raw e-mail | index | archive | help


I'm running softupdates (ffs_softdep.c 1.7) in a kernel cvsupped on
Saturday afternoon.  I have a ccd stripe set (3 older f/w barracudas)
mounted with softupdates and NFS exported to a few FreeBSD 2.2.5 and
NetBSD/alpha(1.3E) clients as /usr/src and /usr/obj via a very fast
network.  Each client its repeatedly building the world (this is
intended as a torture test for my Myrinet drivers).

Under heavy write load,  the FreeBSD-current NFS server complains
repeatedly of problems like the following:

ffs_fsync: dirty: f749a880: type VREG, usecount 1, writecount 0, refcount 1023, flags (VOBJBUF)
        tag VT_UFS, ino 7937, on dev 21, 2 lock type inode: EXCL (count 1) by pid 293

(pid 293 is an nfsd).

These messages stop if I turn off softupdates, or if I set
vfs.nfs.async=1 (or, obviously, if I compile w/o DIAGNOSTIC).  This
only happens with remote writes, local writes do not trigger this
case. 


Occasionally I'll get a crash like the following.  I've seen this
stacktrace on -current, so I guess its a known problem.

kgdb) where
#0  boot (howto=256) at ../../kern/kern_shutdown.c:281
#1  0xf0116de7 in panic (
    fmt=0xf01a25b4 "handle_workitem_freeblocks: block count")
    at ../../kern/kern_shutdown.c:421
#2  0xf01a2827 in handle_workitem_freeblocks (freeblks=0xf57b7500)
    at ../../ufs/ffs/ffs_softdep.c:1968
#3  0xf01a08dc in softdep_process_worklist (matchmnt=0x0)
    at ../../ufs/ffs/ffs_softdep.c:542
#4  0xf01364be in sched_sync () at ../../kern/vfs_subr.c:958
#5  0xf0109f4e in kproc_start (udata=0xf01fae44) at ../../kern/init_main.c:245
#6  0xf01c5eb9 in fork_trampoline ()
#7  0x66da89f0 in ?? ()
Cannot access memory at address 0x1c5e10bb.

I just wanted to report these; neither of these problems is a show
stopper for my purposes...

------------------------------------------------------------------------------
Andrew Gallatin, Sr Systems Programmer	http://www.cs.duke.edu/~gallatin
Duke University				Email: gallatin@cs.duke.edu
Department of Computer Science		Phone: (919) 660-6590



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199806081336.JAA06735>