ceph: disable cached readdir after dropping positive dentry

Ideally CEPH_CAP_FILE_SHARED should have been revoked before
postive dentry get dropped. But if something goes wrong, later
cached readdir may dereference the dropped dentry.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

fs/ceph/inode.c

@@ -1186,6 +1186,7 @@ retry_lookup:
 				    ceph_snap(d_inode(dn)) != tvino.snap)) {
 				dout(" dn %p points to wrong inode %p\n",
 				     dn, d_inode(dn));
+				ceph_dir_clear_ordered(dir);
 				d_delete(dn);
 				dput(dn);
 				goto retry_lookup;
@@ -1323,6 +1324,7 @@ retry_lookup:
 			dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
 			     dn, d_inode(dn), ceph_vinop(d_inode(dn)),
 			     ceph_vinop(in));
+			ceph_dir_clear_ordered(dir);
 			d_invalidate(dn);
 			have_lease = false;
 		}
@@ -1574,6 +1576,7 @@ retry_lookup:
 			    ceph_snap(d_inode(dn)) != tvino.snap)) {
 			dout(" dn %p points to wrong inode %p\n",
 			     dn, d_inode(dn));
+			__ceph_dir_clear_ordered(ci);
 			d_delete(dn);
 			dput(dn);
 			goto retry_lookup;
@@ -1598,7 +1601,9 @@ retry_lookup:
 				 &req->r_caps_reservation);
 		if (ret < 0) {
 			pr_err("fill_inode badness on %p\n", in);
-			if (d_really_is_negative(dn))
+			if (d_really_is_positive(dn))
+				__ceph_dir_clear_ordered(ci);
+			else
 				iput(in);
 			d_drop(dn);
 			err = ret;