[PATCH 1/1] cr: fix use of __d_path()

[Serge Hallyn]

__d_path():
	1. should be used under dcache_lock
	2. can change root->{mnt,dentry} without changing refcounts
The second point was the cause of my BUGs.  The ctx->root was passed
in, and do_checkpoint() had taken a path_get on the vfsroot.  So now
at cleanup it was doing path_put() using another mnt+dentry.

(Why they are different, I'm not sure - but my guess would be that
stdin or stdout is inherited from the parent task in parent mntns,
hence file->mnt is different from root->mnt as it's a different
namespace.)

Signed-off-by: Serge Hallyn <serue@xxxxxxxxxx>
---
 checkpoint/checkpoint.c |   13 ++++++++++++-
 1 files changed, 12 insertions(+), 1 deletions(-)

diff --git a/checkpoint/checkpoint.c b/checkpoint/checkpoint.c
index 173b637..7f0c1e7 100644
--- a/checkpoint/checkpoint.c
+++ b/checkpoint/checkpoint.c
@@ -70,9 +70,20 @@ static char *
 cr_fill_fname(struct path *path, struct path *root, char *buf, int *n)
 {
 	char *fname;
+	struct path root2;
+
+	root2.mnt = root->mnt;
+	root2.dentry = root->dentry;
 
 	BUG_ON(!buf);
-	fname = __d_path(path, root, buf, *n);
+	spin_lock(&dcache_lock);
+	fname = __d_path(path, &root2, buf, *n);
+	spin_unlock(&dcache_lock);
+	if (root2.mnt != root->mnt)
+		printk(KERN_NOTICE "%s: mnt changed\n", __func__);
+	if (root2.dentry != root->dentry)
+		printk(KERN_NOTICE "%s: dentry changed\n", __func__);
+	fname = buf+10;
 	if (!IS_ERR(fname))
 		*n = (buf + (*n) - fname);
 	return fname;
-- 
1.5.6.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/