Merge branch 'net-smc-more-socket-closing-improvements'

Ursula Braun says:

====================
net/smc: more socket closing improvements

these patches improve the smc behavior for abnormal socket closing.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

net/smc/af_smc.c

@@ -377,6 +377,15 @@ static void smc_link_save_peer_info(struct smc_link *link,
 	link->peer_mtu = clc->qp_mtu;
 }
 
+static void smc_lgr_forget(struct smc_link_group *lgr)
+{
+	spin_lock_bh(&smc_lgr_list.lock);
+	/* do not use this link group for new connections */
+	if (!list_empty(&lgr->list))
+		list_del_init(&lgr->list);
+	spin_unlock_bh(&smc_lgr_list.lock);
+}
+
 /* setup for RDMA connection of client */
 static int smc_connect_rdma(struct smc_sock *smc)
 {
@@ -513,6 +522,8 @@ out_connected:
 	return rc ? rc : local_contact;
 
 decline_rdma_unlock:
+	if (local_contact == SMC_FIRST_CONTACT)
+		smc_lgr_forget(smc->conn.lgr);
 	mutex_unlock(&smc_create_lgr_pending);
 	smc_conn_free(&smc->conn);
 decline_rdma:
@@ -526,6 +537,8 @@ decline_rdma:
 	goto out_connected;
 
 out_err_unlock:
+	if (local_contact == SMC_FIRST_CONTACT)
+		smc_lgr_forget(smc->conn.lgr);
 	mutex_unlock(&smc_create_lgr_pending);
 	smc_conn_free(&smc->conn);
 out_err:
@@ -906,6 +919,8 @@ enqueue:
 	return;
 
 decline_rdma_unlock:
+	if (local_contact == SMC_FIRST_CONTACT)
+		smc_lgr_forget(new_smc->conn.lgr);
 	mutex_unlock(&smc_create_lgr_pending);
 decline_rdma:
 	/* RDMA setup failed, switch back to TCP */
@@ -918,6 +933,8 @@ decline_rdma:
 	goto out_connected;
 
 out_err_unlock:
+	if (local_contact == SMC_FIRST_CONTACT)
+		smc_lgr_forget(new_smc->conn.lgr);
 	mutex_unlock(&smc_create_lgr_pending);
 out_err:
 	newsmcsk->sk_state = SMC_CLOSED;

net/smc/smc_cdc.c

@@ -65,9 +65,14 @@ int smc_cdc_get_free_slot(struct smc_connection *conn,
 			  struct smc_cdc_tx_pend **pend)
 {
 	struct smc_link *link = &conn->lgr->lnk[SMC_SINGLE_LINK];
+	int rc;
 
-	return smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
-				       (struct smc_wr_tx_pend_priv **)pend);
+	rc = smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
+				     (struct smc_wr_tx_pend_priv **)pend);
+	if (!conn->alert_token_local)
+		/* abnormal termination */
+		rc = -EPIPE;
+	return rc;
 }
 
 static inline void smc_cdc_add_pending_send(struct smc_connection *conn,

net/smc/smc_close.c

@@ -89,7 +89,7 @@ static int smc_close_abort(struct smc_connection *conn)
 }
 
 /* terminate smc socket abnormally - active abort
- * RDMA communication no longer possible
+ * link group is terminated, i.e. RDMA communication no longer possible
  */
 static void smc_close_active_abort(struct smc_sock *smc)
 {
@@ -107,21 +107,25 @@ static void smc_close_active_abort(struct smc_sock *smc)
 	case SMC_INIT:
 	case SMC_ACTIVE:
 		sk->sk_state = SMC_PEERABORTWAIT;
+		release_sock(sk);
+		cancel_delayed_work_sync(&smc->conn.tx_work);
+		lock_sock(sk);
 		break;
 	case SMC_APPCLOSEWAIT1:
 	case SMC_APPCLOSEWAIT2:
-		txflags->peer_conn_abort = 1;
 		sock_release(smc->clcsock);
 		if (!smc_cdc_rxed_any_close(&smc->conn))
 			sk->sk_state = SMC_PEERABORTWAIT;
 		else
 			sk->sk_state = SMC_CLOSED;
+		release_sock(sk);
+		cancel_delayed_work_sync(&smc->conn.tx_work);
+		lock_sock(sk);
 		break;
 	case SMC_PEERCLOSEWAIT1:
 	case SMC_PEERCLOSEWAIT2:
 		if (!txflags->peer_conn_closed) {
 			sk->sk_state = SMC_PEERABORTWAIT;
-			txflags->peer_conn_abort = 1;
 			sock_release(smc->clcsock);
 		} else {
 			sk->sk_state = SMC_CLOSED;
@@ -129,10 +133,8 @@ static void smc_close_active_abort(struct smc_sock *smc)
 		break;
 	case SMC_PROCESSABORT:
 	case SMC_APPFINCLOSEWAIT:
-		if (!txflags->peer_conn_closed) {
-			txflags->peer_conn_abort = 1;
+		if (!txflags->peer_conn_closed)
 			sock_release(smc->clcsock);
-		}
 		sk->sk_state = SMC_CLOSED;
 		break;
 	case SMC_PEERFINCLOSEWAIT:
@@ -249,9 +251,6 @@ again:
 		/* peer sending PeerConnectionClosed will cause transition */
 		break;
 	case SMC_PROCESSABORT:
-		release_sock(sk);
-		cancel_delayed_work_sync(&conn->tx_work);
-		lock_sock(sk);
 		smc_close_abort(conn);
 		sk->sk_state = SMC_CLOSED;
 		break;
@@ -277,7 +276,6 @@ static void smc_close_passive_abort_received(struct smc_sock *smc)
 	case SMC_APPFINCLOSEWAIT:
 	case SMC_APPCLOSEWAIT1:
 	case SMC_APPCLOSEWAIT2:
-		smc_close_abort(&smc->conn);
 		sk->sk_state = SMC_PROCESSABORT;
 		break;
 	case SMC_PEERCLOSEWAIT1:
@@ -285,7 +283,6 @@ static void smc_close_passive_abort_received(struct smc_sock *smc)
 		if (txflags->peer_done_writing &&
 		    !smc_close_sent_any_close(&smc->conn)) {
 			/* just shutdown, but not yet closed locally */
-			smc_close_abort(&smc->conn);
 			sk->sk_state = SMC_PROCESSABORT;
 		} else {
 			sk->sk_state = SMC_CLOSED;
@@ -302,8 +299,9 @@ static void smc_close_passive_abort_received(struct smc_sock *smc)
 	}
 }
 
-/* Some kind of closing has been received: peer_conn_closed, peer_conn_abort,
- * or peer_done_writing.
+/* Either some kind of closing has been received: peer_conn_closed,
+ * peer_conn_abort, or peer_done_writing
+ * or the link group of the connection terminates abnormally.
  */
 static void smc_close_passive_work(struct work_struct *work)
 {
@@ -326,7 +324,11 @@ static void smc_close_passive_work(struct work_struct *work)
 
 	rxflags = &conn->local_rx_ctrl.conn_state_flags;
 	if (rxflags->peer_conn_abort) {
+		/* peer has not received all data */
 		smc_close_passive_abort_received(smc);
+		release_sock(&smc->sk);
+		cancel_delayed_work_sync(&conn->tx_work);
+		lock_sock(&smc->sk);
 		goto wakeup;
 	}
 
@@ -348,7 +350,6 @@ static void smc_close_passive_work(struct work_struct *work)
 		/* fall through */
 		/* to check for closing */
 	case SMC_PEERCLOSEWAIT2:
-	case SMC_PEERFINCLOSEWAIT:
 		if (!smc_cdc_rxed_any_close(conn))
 			break;
 		if (sock_flag(sk, SOCK_DEAD) &&
@@ -360,6 +361,10 @@ static void smc_close_passive_work(struct work_struct *work)
 			sk->sk_state = SMC_APPFINCLOSEWAIT;
 		}
 		break;
+	case SMC_PEERFINCLOSEWAIT:
+		if (smc_cdc_rxed_any_close(conn))
+			sk->sk_state = SMC_CLOSED;
+		break;
 	case SMC_APPCLOSEWAIT1:
 	case SMC_APPCLOSEWAIT2:
 	case SMC_APPFINCLOSEWAIT:

net/smc/smc_core.c

@@ -128,6 +128,8 @@ static void smc_lgr_free_work(struct work_struct *work)
 	bool conns;
 
 	spin_lock_bh(&smc_lgr_list.lock);
+	if (list_empty(&lgr->list))
+		goto free;
 	read_lock_bh(&lgr->conns_lock);
 	conns = RB_EMPTY_ROOT(&lgr->conns_all);
 	read_unlock_bh(&lgr->conns_lock);
@@ -136,6 +138,7 @@ static void smc_lgr_free_work(struct work_struct *work)
 		return;
 	}
 	list_del_init(&lgr->list); /* remove from smc_lgr_list */
+free:
 	spin_unlock_bh(&smc_lgr_list.lock);
 	smc_lgr_free(lgr);
 }
@@ -231,9 +234,7 @@ static void smc_buf_unuse(struct smc_connection *conn)
 /* remove a finished connection from its link group */
 void smc_conn_free(struct smc_connection *conn)
 {
-	struct smc_link_group *lgr = conn->lgr;
-
-	if (!lgr)
+	if (!conn->lgr)
 		return;
 	smc_cdc_tx_dismiss_slots(conn);
 	smc_lgr_unregister_conn(conn);
@@ -328,12 +329,16 @@ void smc_lgr_terminate(struct smc_link_group *lgr)
 		conn = rb_entry(node, struct smc_connection, alert_node);
 		smc = container_of(conn, struct smc_sock, conn);
 		sock_hold(&smc->sk);
+		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
 		__smc_lgr_unregister_conn(conn);
+		write_unlock_bh(&lgr->conns_lock);
 		schedule_work(&conn->close_work);
+		write_lock_bh(&lgr->conns_lock);
 		sock_put(&smc->sk);
 		node = rb_first(&lgr->conns_all);
 	}
 	write_unlock_bh(&lgr->conns_lock);
+	wake_up(&lgr->lnk[SMC_SINGLE_LINK].wr_reg_wait);
 }
 
 /* Determine vlan of internal TCP socket.

net/smc/smc_diag.c

@@ -86,7 +86,8 @@ static int __smc_diag_dump(struct sock *sk, struct sk_buff *skb,
 	if (smc_diag_msg_attrs_fill(sk, skb, r, user_ns))
 		goto errout;
 
-	if ((req->diag_ext & (1 << (SMC_DIAG_CONNINFO - 1))) && smc->conn.lgr) {
+	if ((req->diag_ext & (1 << (SMC_DIAG_CONNINFO - 1))) &&
+	    smc->conn.alert_token_local) {
 		struct smc_connection *conn = &smc->conn;
 		struct smc_diag_conninfo cinfo = {
 			.token = conn->alert_token_local,
@@ -124,7 +125,8 @@ static int __smc_diag_dump(struct sock *sk, struct sk_buff *skb,
 			goto errout;
 	}
 
-	if ((req->diag_ext & (1 << (SMC_DIAG_LGRINFO - 1))) && smc->conn.lgr) {
+	if ((req->diag_ext & (1 << (SMC_DIAG_LGRINFO - 1))) && smc->conn.lgr &&
+	    !list_empty(&smc->conn.lgr->list)) {
 		struct smc_diag_lgrinfo linfo = {
 			.role = smc->conn.lgr->role,
 			.lnk[0].ibport = smc->conn.lgr->lnk[0].ibport,

net/smc/smc_tx.c

@@ -248,8 +248,10 @@ static int smc_tx_rdma_write(struct smc_connection *conn, int peer_rmbe_offset,
 		peer_rmbe_offset;
 	rdma_wr.rkey = lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].rkey;
 	rc = ib_post_send(link->roce_qp, &rdma_wr.wr, &failed_wr);
-	if (rc)
+	if (rc) {
 		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
+		smc_lgr_terminate(lgr);
+	}
 	return rc;
 }
 
@@ -406,8 +408,9 @@ int smc_tx_sndbuf_nonempty(struct smc_connection *conn)
 				goto out_unlock;
 			}
 			rc = 0;
-			schedule_delayed_work(&conn->tx_work,
-					      SMC_TX_WORK_DELAY);
+			if (conn->alert_token_local) /* connection healthy */
+				schedule_delayed_work(&conn->tx_work,
+						      SMC_TX_WORK_DELAY);
 		}
 		goto out_unlock;
 	}
@@ -438,10 +441,17 @@ static void smc_tx_work(struct work_struct *work)
 	int rc;
 
 	lock_sock(&smc->sk);
+	if (smc->sk.sk_err ||
+	    !conn->alert_token_local ||
+	    conn->local_rx_ctrl.conn_state_flags.peer_conn_abort)
+		goto out;
+
 	rc = smc_tx_sndbuf_nonempty(conn);
 	if (!rc && conn->local_rx_ctrl.prod_flags.write_blocked &&
 	    !atomic_read(&conn->bytes_to_rcv))
 		conn->local_rx_ctrl.prod_flags.write_blocked = 0;
+
+out:
 	release_sock(&smc->sk);
 }
 
@@ -462,7 +472,8 @@ void smc_tx_consumer_update(struct smc_connection *conn)
 	    ((to_confirm > conn->rmbe_update_limit) &&
 	     ((to_confirm > (conn->rmbe_size / 2)) ||
 	      conn->local_rx_ctrl.prod_flags.write_blocked))) {
-		if (smc_cdc_get_slot_and_msg_send(conn) < 0) {
+		if ((smc_cdc_get_slot_and_msg_send(conn) < 0) &&
+		    conn->alert_token_local) { /* connection healthy */
 			schedule_delayed_work(&conn->tx_work,
 					      SMC_TX_WORK_DELAY);
 			return;

net/smc/smc_wr.c

@@ -174,9 +174,9 @@ int smc_wr_tx_get_free_slot(struct smc_link *link,
 			    struct smc_wr_tx_pend_priv **wr_pend_priv)
 {
 	struct smc_wr_tx_pend *wr_pend;
+	u32 idx = link->wr_tx_cnt;
 	struct ib_send_wr *wr_ib;
 	u64 wr_id;
-	u32 idx;
 	int rc;
 
 	*wr_buf = NULL;
@@ -186,16 +186,17 @@ int smc_wr_tx_get_free_slot(struct smc_link *link,
 		if (rc)
 			return rc;
 	} else {
+		struct smc_link_group *lgr;
+
+		lgr = container_of(link, struct smc_link_group,
+				   lnk[SMC_SINGLE_LINK]);
 		rc = wait_event_timeout(
 			link->wr_tx_wait,
+			list_empty(&lgr->list) || /* lgr terminated */
 			(smc_wr_tx_get_free_slot_index(link, &idx) != -EBUSY),
 			SMC_WR_TX_WAIT_FREE_SLOT_TIME);
 		if (!rc) {
 			/* timeout - terminate connections */
-			struct smc_link_group *lgr;
-
-			lgr = container_of(link, struct smc_link_group,
-					   lnk[SMC_SINGLE_LINK]);
 			smc_lgr_terminate(lgr);
 			return -EPIPE;
 		}
@@ -248,8 +249,14 @@ int smc_wr_tx_send(struct smc_link *link, struct smc_wr_tx_pend_priv *priv)
 	pend = container_of(priv, struct smc_wr_tx_pend, priv);
 	rc = ib_post_send(link->roce_qp, &link->wr_tx_ibs[pend->idx],
 			  &failed_wr);
-	if (rc)
+	if (rc) {
+		struct smc_link_group *lgr =
+			container_of(link, struct smc_link_group,
+				     lnk[SMC_SINGLE_LINK]);
+
 		smc_wr_tx_put_slot(link, priv);
+		smc_lgr_terminate(lgr);
+	}
 	return rc;
 }