[Cluster-devel] Re: [Linux-cluster] GFS2 interesting death with error

[Cluster-devel] Re: [Linux-cluster] GFS2 interesting death with error

[Steven Whitehouse]

Hi,

On Thu, 2009-11-05 at 14:36 -0500, Allen Belletti wrote:
> Saw an interesting and different GFS2 death this morning that I wanted 
> to pass along in case anyone has insights.  We have not seen any of the 
> "hanging in dlm_posix_lock" since fsck'ing early Sunday morning.  In any 
> case I'm pretty confident that's being triggered by the creation & 
> deletion of ".lock" files within Dovecot.  This was something completely 
> different and it left some potentially useful debug info in the logs.
> 
I've made an educated guess as to what this might be. The attached patch
should  fix it, if my hunch is correct. If you have the back trace I
mentioned in my previous email, we can confirm that this really is the
cause,

Steve.


From 89fc5489d25fc0a34a367b119448a037ed162c00 Mon Sep 17 00:00:00 2001
From: Steven Whitehouse <swhiteho@redhat.com>
Date: Fri, 6 Nov 2009 11:10:51 +0000
Subject: [PATCH 27/27] GFS2: Locking order fix in gfs2_check_blk_state

In some cases we already have the rindex lock when
we enter this function.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
---
 fs/gfs2/rgrp.c |   14 ++++++++++----
 1 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/fs/gfs2/rgrp.c b/fs/gfs2/rgrp.c
index 8f1cfb0..0608f49 100644
--- a/fs/gfs2/rgrp.c
+++ b/fs/gfs2/rgrp.c
@@ -1710,11 +1710,16 @@ int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
 {
 	struct gfs2_rgrpd *rgd;
 	struct gfs2_holder ri_gh, rgd_gh;
+	struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex);
+	int ri_locked = 0;
 	int error;
 
-	error = gfs2_rindex_hold(sdp, &ri_gh);
-	if (error)
-		goto fail;
+	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
+		error = gfs2_rindex_hold(sdp, &ri_gh);
+		if (error)
+			goto fail;
+		ri_locked = 1;
+	}
 
 	error = -EINVAL;
 	rgd = gfs2_blk2rgrpd(sdp, no_addr);
@@ -1730,7 +1735,8 @@ int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type)
 
 	gfs2_glock_dq_uninit(&rgd_gh);
 fail_rindex:
-	gfs2_glock_dq_uninit(&ri_gh);
+	if (ri_locked)
+		gfs2_glock_dq_uninit(&ri_gh);
 fail:
 	return error;
 }
-- 
1.6.2.5