ceph: fix inode removal from snap realm when racing with migration

When an inode was dropped while being migrated between two MDSs,
i_cap_exporting_issued was non-zero such that issue caps were non-zero and
__ceph_is_any_caps(ci) was true.  This prevented the inode from being
removed from the snap realm, even as it was dropped from the cache.

Fix this by dropping any residual i_snap_realm ref in destroy_inode.

Signed-off-by: Sage Weil <sage@newdream.net>

fs/ceph/inode.c

@@ -378,6 +378,22 @@ void ceph_destroy_inode(struct inode *inode)
 
 	ceph_queue_caps_release(inode);
 
+	/*
+	 * we may still have a snap_realm reference if there are stray
+	 * caps in i_cap_exporting_issued or i_snap_caps.
+	 */
+	if (ci->i_snap_realm) {
+		struct ceph_mds_client *mdsc =
+			&ceph_client(ci->vfs_inode.i_sb)->mdsc;
+		struct ceph_snap_realm *realm = ci->i_snap_realm;
+
+		dout(" dropping residual ref to snap realm %p\n", realm);
+		spin_lock(&realm->inodes_with_caps_lock);
+		list_del_init(&ci->i_snap_realm_item);
+		spin_unlock(&realm->inodes_with_caps_lock);
+		ceph_put_snap_realm(mdsc, realm);
+	}
+
 	kfree(ci->i_symlink);
 	while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 		frag = rb_entry(n, struct ceph_inode_frag, node);