ceph: document unlocked d_parent accesses

For the most part we don't care about racing with rename when directing
MDS requests; either the old or new parent is fine.  Document that, and
do some minor cleanup.

Reviewed-by: Yehuda Sadeh <yehuda@hq.newdream.net>
Signed-off-by: Sage Weil <sage@newdream.net>

fs/ceph/dir.c

@@ -506,7 +506,7 @@ int ceph_handle_snapdir(struct ceph_mds_request *req,
 			struct dentry *dentry, int err)
 {
 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
-	struct inode *parent = dentry->d_parent->d_inode;
+	struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
 
 	/* .snap dir? */
 	if (err == -ENOENT &&

fs/ceph/mds_client.c

@@ -621,6 +621,12 @@ static void __unregister_request(struct ceph_mds_client *mdsc,
  */
 struct dentry *get_nonsnap_parent(struct dentry *dentry)
 {
+	/*
+	 * we don't need to worry about protecting the d_parent access
+	 * here because we never renaming inside the snapped namespace
+	 * except to resplice to another snapdir, and either the old or new
+	 * result is a valid result.
+	 */
 	while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
 		dentry = dentry->d_parent;
 	return dentry;
@@ -656,16 +662,17 @@ static int __choose_mds(struct ceph_mds_client *mdsc,
 	if (req->r_inode) {
 		inode = req->r_inode;
 	} else if (req->r_dentry) {
-		struct inode *dir = req->r_dentry->d_parent->d_inode;
+		/* ignore race with rename; old or new d_parent is okay */
+		struct dentry *parent = req->r_dentry->d_parent;
+		struct inode *dir = parent->d_inode;
 
 		if (dir->i_sb != mdsc->fsc->sb) {
 			/* not this fs! */
 			inode = req->r_dentry->d_inode;
 		} else if (ceph_snap(dir) != CEPH_NOSNAP) {
 			/* direct snapped/virtual snapdir requests
 			 * based on parent dir inode */
-			struct dentry *dn =
-				get_nonsnap_parent(req->r_dentry->d_parent);
+			struct dentry *dn = get_nonsnap_parent(parent);
 			inode = dn->d_inode;
 			dout("__choose_mds using nonsnap parent %p\n", inode);
 		} else if (req->r_dentry->d_inode) {