ceph: stop on-going cached readdir if mds revokes FILE_SHARED cap

If directory's FILE_SHARED cap get revoked, dentry in the directory
can get spliced into other directory (Eg, other client move the
dentry into directory B, then we do readdir on directory B). So we
should stop on-going cached readdir. this can be achieved by marking
dir not complete, because __dcache_readdir() checks dir completeness
before emitting each dentry.
Signed-off-by: N"Yan, Zheng" <zyan@redhat.com>
Signed-off-by: NIlya Dryomov <idryomov@gmail.com>

fs/ceph/caps.c

@@ -490,13 +490,14 @@ static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
 	}
 
 	/*
-	 * if we are newly issued FILE_SHARED, mark dir not complete; we
-	 * don't know what happened to this directory while we didn't
-	 * have the cap.
+	 * If FILE_SHARED is newly issued, mark dir not complete. We don't
+	 * know what happened to this directory while we didn't have the cap.
+	 * If FILE_SHARED is being revoked, also mark dir not complete. It
+	 * stops on-going cached readdir.
 	 */
-	if ((issued & CEPH_CAP_FILE_SHARED) &&
-	    (had & CEPH_CAP_FILE_SHARED) == 0) {
-		ci->i_shared_gen++;
+	if ((issued & CEPH_CAP_FILE_SHARED) != (had & CEPH_CAP_FILE_SHARED)) {
+		if (issued & CEPH_CAP_FILE_SHARED)
+			ci->i_shared_gen++;
 		if (S_ISDIR(ci->vfs_inode.i_mode)) {
 			dout(" marking %p NOT complete\n", &ci->vfs_inode);
 			__ceph_dir_clear_complete(ci);