From owner-freebsd-fs  Wed Aug  4 20:16:45 1999
Delivered-To: freebsd-fs@freebsd.org
Received: from chrome.dreamscape.com (chrome.dreamscape.com [206.64.128.4])
	by hub.freebsd.org (Postfix) with ESMTP id 4EEB115064
	for <freebsd-fs@FreeBSD.ORG>; Wed,  4 Aug 1999 20:16:38 -0700 (PDT)
	(envelope-from krentel@dreamscape.com)
Received: from europa.dreamscape.com (mail.dreamscape.com [206.64.128.147])
	by chrome.dreamscape.com (8.9.1/8.9.1) with ESMTP id XAA00290;
	Wed, 4 Aug 1999 23:15:47 -0400 (EDT)
X-Dreamscape-Track-Chrome-A: mail.dreamscape.com [206.64.128.147]
X-Dreamscape-Track-Chrome-B: Wed, 4 Aug 1999 23:15:47 -0400 (EDT)
Received: from dreamscape.com (sA9-p62.dreamscape.com [207.198.13.254])
          by europa.dreamscape.com (8.8.5/8.8.4) with ESMTP
	  id WAA03034; Wed, 4 Aug 1999 22:19:40 -0400 (EDT)
X-Dreamscape-Track-A: sA9-p62.dreamscape.com [207.198.13.254]
X-Dreamscape-Track-B: Wed, 4 Aug 1999 22:19:40 -0400 (EDT)
Received: (from krentel@localhost)
	by dreamscape.com (8.9.3/8.9.3) id WAA03204;
	Wed, 4 Aug 1999 22:18:44 -0400 (EDT)
	(envelope-from krentel)
Date: Wed, 4 Aug 1999 22:18:44 -0400 (EDT)
From: "Mark W. Krentel" <krentel@dreamscape.com>
Message-Id: <199908050218.WAA03204@dreamscape.com>
To: bde@zeta.org.au, freebsd-fs@FreeBSD.ORG
Subject: Re: link counts in ext2fs are all 0
Sender: owner-freebsd-fs@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.org

> Link counts in ext2fs were broken by the soft updates changes (although
> soft updates are not implemented for ext2fs).  Partial fix:

I see.  This happens even though I'm not using soft updates?

I tried your patch in 3.2-release and indeed it does what you said it
would.  Old files have a correct link count while new ones appear to
have 0 links.  And even the new ones appear correct if you wait long
enough (even without umount/mount).  I guess that happens when the
vnode gets flushed out of memory?  A bazillion sync's doesn't do it,
but unpacking a large tar file does.

Should I consider the patch dangerous?  Suppose I create a new file,
the data gets sync'd but the power goes off before the vnode gets
flushed or the partition is unmounted.  Would the link count be wrong
for that file?

> This works for old links but not for new ones, since ext2_vnops.c doesn't
> know anything about i_effnlink.  E.g., link(2) appears to corrupt the
> link count (by not changing it), but the problem is magically fixed
> by umount/mount when the above changes do the translation.  Bugs like
> this show why ext2fs shouldn't use any ufs vnops.  (I just noticed some
> related ones for chflags(2).

I'm confused.  Is your comment, "ext2fs shouldn't use any ufs vnops",
directed at developers (suggesting that ext2fs should be reorganized)
or at users?

Thanks!

--Mark Krentel


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