[Cluster-devel] [GFS2 PATCH] GFS2: Check if iopen is held when deleting inode

[Bob Peterson]

Hi,

This patch fixes an error condition in which an inode is partially
created in gfs2_create_inode() but then some error is discovered,
which causes it to fail and call iput() before the iopen glock is
created or held. In that case, gfs2_delete_inode would try to
unlock an iopen glock that doesn't yet exist. Therefore, we test
its holder (which must exist) for the HIF_HOLDER bit before trying
to dq it.

Signed-off-by: Bob Peterson <rpeterso at redhat.com>

diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
index 8f960a5..d354e72 100644
--- a/fs/gfs2/super.c
+++ b/fs/gfs2/super.c
@@ -1551,12 +1551,15 @@ static void gfs2_evict_inode(struct inode *inode)
 			goto out_truncate;
 	}
 
-	ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
-	gfs2_glock_dq_wait(&ip->i_iopen_gh);
-	gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
-	error = gfs2_glock_nq(&ip->i_iopen_gh);
-	if (error)
-		goto out_truncate;
+	if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
+		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
+		gfs2_glock_dq_wait(&ip->i_iopen_gh);
+		gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
+				   &ip->i_iopen_gh);
+		error = gfs2_glock_nq(&ip->i_iopen_gh);
+		if (error)
+			goto out_truncate;
+	}
 
 	/* Case 1 starts here */