Date: Fri, 19 Jan 2001 14:37:20 +0100 From: Rogier Mulhuijzen <rmulhuijzen@Transfer-Solutions.com> To: freebsd-current@freebsd.org Cc: John Baldwin <jhb@FreeBSD.org> Subject: NTFS issue (was: Panic w/crash dump (looks like atomic.h problem)) Message-ID: <5.0.2.1.0.20010119135940.00a565d0@tfrsolcom> In-Reply-To: <XFMail.010118130907.jhb@FreeBSD.org> References: <5.0.2.1.0.20010118201134.00a542c0@tfrsolcom>
next in thread | previous in thread | raw e-mail | index | archive | help
a quick look at kldstat output pinpointed the ??'s in my stacktrace(see
previous posts) to ntfs.ko. Compiled NTFS into my kernel and reproduced the
crash.
Taking a look at the trace now:
#11 0xc01c0c97 in vref (vp=0x0) at machine/atomic.h:332
#12 0xc0239592 in ntfs_vgetex (mp=0xc0f7dc00, ino=4458, attrtype=128,
attrname=0x0, lkflags=2, flags=4, p=0xc8d82640, vpp=0xc7f85db8)
at ../../ntfs/ntfs_vfsops.c:987
#13 0xc0235a0b in ntfs_ntvattrget (ntmp=0xc0f7da00, ip=0xc1043000, type=128,
name=0xc02c3fe8 "", vcn=0, vapp=0xc7f85e3c) at ../../ntfs/ntfs_subr.c:231
#14 0xc0238134 in ntfs_readattr (ntmp=0xc0f7da00, ip=0xc1043000, attrnum=128,
attrname=0x0, roff=1644032, rsize=512, rdata=0x0, uio=0xc7f85ee4)
at ../../ntfs/ntfs_subr.c:1699
#15 0xc02396c5 in ntfs_read (ap=0xc7f85e9c) at ../../ntfs/ntfs_vnops.c:184
#16 0xc01c91e4 in vn_read (fp=0xc103e540, uio=0xc7f85ee4, cred=0xc1024600,
flags=0, p=0xc8d82640) at vnode_if.h:279
I took a look at the code in src/sys/ntfs/ntfs_vfsops.c
I compared the code surrounding VREF there with code in other FS drivers.
in ntfs/ntfs_vfsops.c
if (lkflags & LK_TYPE_MASK) {
error = VN_LOCK(vp, lkflags, p);
if (error) {
vput(vp);
return (error);
}
}
VREF(ip->i_devvp);
in isofs/cd9660/cd9660_vfsops.c
} else
bp = 0;
ip->i_mnt = imp;
ip->i_devvp = imp->im_devvp;
VREF(ip->i_devvp);
in ufs/ffs/ffs_vfsops.c
/*
* Finish inode initialization now that aliasing has been resolved.
*/
ip->i_devvp = ump->um_devvp;
VREF(ip->i_devvp);
in ufs/ifs/ifs_vfsops.c
/*
* Finish inode initialization now that aliasing has been resolved.
*/
ip->i_devvp = ump->um_devvp;
VREF(ip->i_devvp);
Seeing this I added a check to see if ip->i_devvp was equal to
ntmp->ntm_devvp. And a check to see if it was 0.
During normal operation on my NTFS partition I got no messages, so I took a
bold step and added 'ip->i_devvp = ntmp->ntm_devvp;' just before the VREF call.
After installing new kernel and reboot I once again started playback of an
mp3 on my NTFS partition and started make -j 128 world on my UFS partition.
And 'lo behold, every time I got a message that ip->i_devvp was equal not
to ntmp->ntm_devvp, ip->i_devvp was equal to 0.
I can now stress the hell out of my box and the NTFS filesystem, and it
keeps running.
Why exactly ip->i_devvp is 0 I don't know, I have to dig into the code some
deeper, but every other FS that uses VREF copies it from their mountpoint
structure before calling vref().
Am I on the right track here?
DocWilco
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?5.0.2.1.0.20010119135940.00a565d0>