diff --git a/net/rds/ib_send.c b/net/rds/ib_send.c
index 82459e5..209dbc6 100644
--- a/net/rds/ib_send.c
+++ b/net/rds/ib_send.c
@@ -757,6 +757,8 @@
 	send->s_wr.next = NULL;
 	send->s_wr.wr.atomic.remote_addr = op->op_remote_addr;
 	send->s_wr.wr.atomic.rkey = op->op_rkey;
+	send->s_op = op;
+	rds_message_addref(container_of(send->s_op, struct rds_message, atomic));
 
 	/* map 8 byte retval buffer to the device */
 	ret = ib_dma_map_sg(ic->i_cm_id->device, op->op_sg, 1, DMA_FROM_DEVICE);
@@ -857,13 +859,13 @@
 	for (i = 0; i < work_alloc && scat != &op->op_sg[op->op_count]; i++) {
 		send->s_wr.send_flags = 0;
 		send->s_queued = jiffies;
+		send->s_op = NULL;
 
 		rds_ib_set_wr_signal_state(ic, send, op->op_notify);
 
 		send->s_wr.opcode = op->op_write ? IB_WR_RDMA_WRITE : IB_WR_RDMA_READ;
 		send->s_wr.wr.rdma.remote_addr = remote_addr;
 		send->s_wr.wr.rdma.rkey = op->op_rkey;
-		send->s_op = op;
 
 		if (num_sge > rds_ibdev->max_sge) {
 			send->s_wr.num_sge = rds_ibdev->max_sge;
@@ -899,6 +901,12 @@
 			send = ic->i_sends;
 	}
 
+	/* give a reference to the last op */
+	if (scat == &op->op_sg[op->op_count]) {
+		prev->s_op = op;
+		rds_message_addref(container_of(op, struct rds_message, rdma));
+	}
+
 	if (i < work_alloc) {
 		rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - i);
 		work_alloc = i;
diff --git a/net/rds/send.c b/net/rds/send.c
index 5c6d4a0..437f1e7 100644
--- a/net/rds/send.c
+++ b/net/rds/send.c
@@ -233,13 +233,10 @@
 
 		/* The transport either sends the whole rdma or none of it */
 		if (rm->rdma.op_active && !conn->c_xmit_rdma_sent) {
-			rds_message_addref(rm);
 			rm->m_final_op = &rm->rdma;
 			ret = conn->c_trans->xmit_rdma(conn, &rm->rdma);
-			if (ret) {
-				rds_message_put(rm);
+			if (ret)
 				break;
-			}
 			conn->c_xmit_rdma_sent = 1;
 
 			/* The transport owns the mapped memory for now.
@@ -248,13 +245,10 @@
 		}
 
 		if (rm->atomic.op_active && !conn->c_xmit_atomic_sent) {
-			rds_message_addref(rm);
 			rm->m_final_op = &rm->atomic;
 			ret = conn->c_trans->xmit_atomic(conn, &rm->atomic);
-			if (ret) {
-				rds_message_put(rm);
+			if (ret)
 				break;
-			}
 			conn->c_xmit_atomic_sent = 1;
 
 			/* The transport owns the mapped memory for now.