Merge branch 'net-smc-improve-termination-handling-part-3'

Karsten Graul says:

====================
net/smc: improve termination handling (part 3)

Part 3 of the SMC termination patches improves the link group
termination processing and introduces the ability to immediately
terminate a link group.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/s390/net/ism.h

@@ -32,8 +32,6 @@
 #define ISM_UNREG_SBA	0x11
 #define ISM_UNREG_IEQ	0x12
 
-#define ISM_ERROR	0xFFFF
-
 struct ism_req_hdr {
 	u32 cmd;
 	u16 : 16;

include/net/smc.h

@@ -37,6 +37,8 @@ struct smcd_dmb {
 #define ISM_EVENT_GID	1
 #define ISM_EVENT_SWR	2
 
+#define ISM_ERROR	0xFFFF
+
 struct smcd_event {
 	u32 type;
 	u32 code;
@@ -77,6 +79,8 @@ struct smcd_dev {
 	bool pnetid_by_user;
 	struct list_head lgr_list;
 	spinlock_t lgr_lock;
+	atomic_t lgr_cnt;
+	wait_queue_head_t lgrs_deleted;
 	u8 going_away : 1;
 };
 

net/smc/smc_cdc.c

@@ -131,6 +131,9 @@ int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
 {
 	int rc;
 
+	if (!conn->lgr || (conn->lgr->is_smcd && conn->lgr->peer_shutdown))
+		return -EPIPE;
+
 	if (conn->lgr->is_smcd) {
 		spin_lock_bh(&conn->send_lock);
 		rc = smcd_cdc_msg_send(conn);

net/smc/smc_clc.c

@@ -349,7 +349,7 @@ int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
 		smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
 		if (((struct smc_clc_msg_decline *)buf)->hdr.flag) {
 			smc->conn.lgr->sync_err = 1;
-			smc_lgr_terminate(smc->conn.lgr);
+			smc_lgr_terminate(smc->conn.lgr, true);
 		}
 	}
 

net/smc/smc_close.c

@@ -110,6 +110,17 @@ int smc_close_abort(struct smc_connection *conn)
 	return smc_cdc_get_slot_and_msg_send(conn);
 }
 
+static void smc_close_cancel_work(struct smc_sock *smc)
+{
+	struct sock *sk = &smc->sk;
+
+	release_sock(sk);
+	cancel_work_sync(&smc->conn.close_work);
+	cancel_delayed_work_sync(&smc->conn.tx_work);
+	lock_sock(sk);
+	sk->sk_state = SMC_CLOSED;
+}
+
 /* terminate smc socket abnormally - active abort
  * link group is terminated, i.e. RDMA communication no longer possible
  */
@@ -126,23 +137,21 @@ void smc_close_active_abort(struct smc_sock *smc)
 	switch (sk->sk_state) {
 	case SMC_ACTIVE:
 		sk->sk_state = SMC_PEERABORTWAIT;
-		release_sock(sk);
-		cancel_delayed_work_sync(&smc->conn.tx_work);
-		lock_sock(sk);
+		smc_close_cancel_work(smc);
 		sk->sk_state = SMC_CLOSED;
 		sock_put(sk); /* passive closing */
 		break;
 	case SMC_APPCLOSEWAIT1:
 	case SMC_APPCLOSEWAIT2:
-		release_sock(sk);
-		cancel_delayed_work_sync(&smc->conn.tx_work);
-		lock_sock(sk);
+		smc_close_cancel_work(smc);
 		sk->sk_state = SMC_CLOSED;
 		sock_put(sk); /* postponed passive closing */
 		break;
 	case SMC_PEERCLOSEWAIT1:
 	case SMC_PEERCLOSEWAIT2:
 	case SMC_PEERFINCLOSEWAIT:
+		sk->sk_state = SMC_PEERABORTWAIT;
+		smc_close_cancel_work(smc);
 		sk->sk_state = SMC_CLOSED;
 		smc_conn_free(&smc->conn);
 		release_clcsock = true;
@@ -150,7 +159,11 @@ void smc_close_active_abort(struct smc_sock *smc)
 		break;
 	case SMC_PROCESSABORT:
 	case SMC_APPFINCLOSEWAIT:
+		sk->sk_state = SMC_PEERABORTWAIT;
+		smc_close_cancel_work(smc);
 		sk->sk_state = SMC_CLOSED;
+		smc_conn_free(&smc->conn);
+		release_clcsock = true;
 		break;
 	case SMC_INIT:
 	case SMC_PEERABORTWAIT:

net/smc/smc_core.h

@@ -228,6 +228,8 @@ struct smc_link_group {
 						/* Peer GID (remote) */
 			struct smcd_dev		*smcd;
 						/* ISM device for VLAN reg. */
+			u8			peer_shutdown : 1;
+						/* peer triggered shutdownn */
 		};
 	};
 };
@@ -285,7 +287,7 @@ static inline struct smc_connection *smc_lgr_find_conn(
 
 static inline void smc_lgr_terminate_sched(struct smc_link_group *lgr)
 {
-	if (!lgr->terminating)
+	if (!lgr->terminating && !lgr->freeing)
 		schedule_work(&lgr->terminate_work);
 }
 
@@ -294,10 +296,12 @@ struct smc_clc_msg_accept_confirm;
 struct smc_clc_msg_local;
 
 void smc_lgr_forget(struct smc_link_group *lgr);
-void smc_lgr_terminate(struct smc_link_group *lgr);
+void smc_lgr_terminate(struct smc_link_group *lgr, bool soft);
 void smc_port_terminate(struct smc_ib_device *smcibdev, u8 ibport);
 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid,
 			unsigned short vlan);
+void smc_smcd_terminate_all(struct smcd_dev *dev);
+void smc_smcr_terminate_all(struct smc_ib_device *smcibdev);
 int smc_buf_create(struct smc_sock *smc, bool is_smcd);
 int smc_uncompress_bufsize(u8 compressed);
 int smc_rmb_rtoken_handling(struct smc_connection *conn,

net/smc/smc_ib.c

@@ -520,9 +520,9 @@ static void smc_ib_cleanup_per_ibdev(struct smc_ib_device *smcibdev)
 	if (!smcibdev->initialized)
 		return;
 	smcibdev->initialized = 0;
-	smc_wr_remove_dev(smcibdev);
 	ib_destroy_cq(smcibdev->roce_cq_recv);
 	ib_destroy_cq(smcibdev->roce_cq_send);
+	smc_wr_remove_dev(smcibdev);
 }
 
 static struct ib_client smc_ib_client;
@@ -565,7 +565,7 @@ static void smc_ib_add_dev(struct ib_device *ibdev)
 	schedule_work(&smcibdev->port_event_work);
 }
 
-/* callback function for ib_register_client() */
+/* callback function for ib_unregister_client() */
 static void smc_ib_remove_dev(struct ib_device *ibdev, void *client_data)
 {
 	struct smc_ib_device *smcibdev;
@@ -575,6 +575,7 @@ static void smc_ib_remove_dev(struct ib_device *ibdev, void *client_data)
 	spin_lock(&smc_ib_devices.lock);
 	list_del_init(&smcibdev->list); /* remove from smc_ib_devices */
 	spin_unlock(&smc_ib_devices.lock);
+	smc_smcr_terminate_all(smcibdev);
 	smc_ib_cleanup_per_ibdev(smcibdev);
 	ib_unregister_event_handler(&smcibdev->event_handler);
 	kfree(smcibdev);

net/smc/smc_ism.c

@@ -146,6 +146,10 @@ int smc_ism_put_vlan(struct smcd_dev *smcd, unsigned short vlanid)
 int smc_ism_unregister_dmb(struct smcd_dev *smcd, struct smc_buf_desc *dmb_desc)
 {
 	struct smcd_dmb dmb;
+	int rc = 0;
+
+	if (!dmb_desc->dma_addr)
+		return rc;
 
 	memset(&dmb, 0, sizeof(dmb));
 	dmb.dmb_tok = dmb_desc->token;
@@ -153,7 +157,13 @@ int smc_ism_unregister_dmb(struct smcd_dev *smcd, struct smc_buf_desc *dmb_desc)
 	dmb.cpu_addr = dmb_desc->cpu_addr;
 	dmb.dma_addr = dmb_desc->dma_addr;
 	dmb.dmb_len = dmb_desc->len;
-	return smcd->ops->unregister_dmb(smcd, &dmb);
+	rc = smcd->ops->unregister_dmb(smcd, &dmb);
+	if (!rc || rc == ISM_ERROR) {
+		dmb_desc->cpu_addr = NULL;
+		dmb_desc->dma_addr = 0;
+	}
+
+	return rc;
 }
 
 int smc_ism_register_dmb(struct smc_link_group *lgr, int dmb_len,
@@ -226,6 +236,9 @@ int smc_ism_signal_shutdown(struct smc_link_group *lgr)
 	int rc;
 	union smcd_sw_event_info ev_info;
 
+	if (lgr->peer_shutdown)
+		return 0;
+
 	memcpy(ev_info.uid, lgr->id, SMC_LGR_ID_SIZE);
 	ev_info.vlan_id = lgr->vlan_id;
 	ev_info.code = ISM_EVENT_REQUEST;
@@ -289,6 +302,7 @@ struct smcd_dev *smcd_alloc_dev(struct device *parent, const char *name,
 	spin_lock_init(&smcd->lgr_lock);
 	INIT_LIST_HEAD(&smcd->vlan);
 	INIT_LIST_HEAD(&smcd->lgr_list);
+	init_waitqueue_head(&smcd->lgrs_deleted);
 	smcd->event_wq = alloc_ordered_workqueue("ism_evt_wq-%s)",
 						 WQ_MEM_RECLAIM, name);
 	if (!smcd->event_wq) {
@@ -313,12 +327,12 @@ EXPORT_SYMBOL_GPL(smcd_register_dev);
 void smcd_unregister_dev(struct smcd_dev *smcd)
 {
 	spin_lock(&smcd_dev_list.lock);
-	list_del(&smcd->list);
+	list_del_init(&smcd->list);
 	spin_unlock(&smcd_dev_list.lock);
 	smcd->going_away = 1;
+	smc_smcd_terminate_all(smcd);
 	flush_workqueue(smcd->event_wq);
 	destroy_workqueue(smcd->event_wq);
-	smc_smcd_terminate(smcd, 0, VLAN_VID_MASK);
 
 	device_del(&smcd->dev);
 }
@@ -372,7 +386,7 @@ void smcd_handle_irq(struct smcd_dev *smcd, unsigned int dmbno)
 
 	spin_lock_irqsave(&smcd->lock, flags);
 	conn = smcd->conn[dmbno];
-	if (conn)
+	if (conn && !conn->killed)
 		tasklet_schedule(&conn->rx_tsklet);
 	spin_unlock_irqrestore(&smcd->lock, flags);
 }

net/smc/smc_llc.c

@@ -614,7 +614,7 @@ static void smc_llc_testlink_work(struct work_struct *work)
 	rc = wait_for_completion_interruptible_timeout(&link->llc_testlink_resp,
 						       SMC_LLC_WAIT_TIME);
 	if (rc <= 0) {
-		smc_lgr_terminate(smc_get_lgr(link));
+		smc_lgr_terminate(smc_get_lgr(link), true);
 		return;
 	}
 	next_interval = link->llc_testlink_time;
@@ -656,6 +656,7 @@ void smc_llc_link_active(struct smc_link *link, int testlink_time)
 void smc_llc_link_deleting(struct smc_link *link)
 {
 	link->state = SMC_LNK_DELETING;
+	smc_wr_wakeup_tx_wait(link);
 }
 
 /* called in tasklet context */
@@ -663,6 +664,8 @@ void smc_llc_link_inactive(struct smc_link *link)
 {
 	link->state = SMC_LNK_INACTIVE;
 	cancel_delayed_work(&link->llc_testlink_wrk);
+	smc_wr_wakeup_reg_wait(link);
+	smc_wr_wakeup_tx_wait(link);
 }
 
 /* called in worker context */
@@ -695,9 +698,11 @@ int smc_llc_do_confirm_rkey(struct smc_link *link,
 int smc_llc_do_delete_rkey(struct smc_link *link,
 			   struct smc_buf_desc *rmb_desc)
 {
-	int rc;
+	int rc = 0;
 
 	mutex_lock(&link->llc_delete_rkey_mutex);
+	if (link->state != SMC_LNK_ACTIVE)
+		goto out;
 	reinit_completion(&link->llc_delete_rkey);
 	rc = smc_llc_send_delete_rkey(link, rmb_desc);
 	if (rc)

net/smc/smc_tx.c

@@ -284,7 +284,7 @@ static int smc_tx_rdma_write(struct smc_connection *conn, int 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, NULL);
 	if (rc)
-		smc_lgr_terminate(lgr);
+		smc_lgr_terminate(lgr, true);
 	return rc;
 }
 

net/smc/smc_wr.c

@@ -50,6 +50,26 @@ struct smc_wr_tx_pend {	/* control data for a pending send request */
 
 /*------------------------------- completion --------------------------------*/
 
+/* returns true if at least one tx work request is pending on the given link */
+static inline bool smc_wr_is_tx_pend(struct smc_link *link)
+{
+	if (find_first_bit(link->wr_tx_mask, link->wr_tx_cnt) !=
+							link->wr_tx_cnt) {
+		return true;
+	}
+	return false;
+}
+
+/* wait till all pending tx work requests on the given link are completed */
+static inline int smc_wr_tx_wait_no_pending_sends(struct smc_link *link)
+{
+	if (wait_event_timeout(link->wr_tx_wait, !smc_wr_is_tx_pend(link),
+			       SMC_WR_TX_WAIT_PENDING_TIME))
+		return 0;
+	else /* timeout */
+		return -EPIPE;
+}
+
 static inline int smc_wr_tx_find_pending_index(struct smc_link *link, u64 wr_id)
 {
 	u32 i;
@@ -75,7 +95,7 @@ static inline void smc_wr_tx_process_cqe(struct ib_wc *wc)
 			link->wr_reg_state = FAILED;
 		else
 			link->wr_reg_state = CONFIRMED;
-		wake_up(&link->wr_reg_wait);
+		smc_wr_wakeup_reg_wait(link);
 		return;
 	}
 
@@ -171,6 +191,7 @@ int smc_wr_tx_get_free_slot(struct smc_link *link,
 			    struct smc_rdma_wr **wr_rdma_buf,
 			    struct smc_wr_tx_pend_priv **wr_pend_priv)
 {
+	struct smc_link_group *lgr = smc_get_lgr(link);
 	struct smc_wr_tx_pend *wr_pend;
 	u32 idx = link->wr_tx_cnt;
 	struct ib_send_wr *wr_ib;
@@ -179,19 +200,20 @@ int smc_wr_tx_get_free_slot(struct smc_link *link,
 
 	*wr_buf = NULL;
 	*wr_pend_priv = NULL;
-	if (in_softirq()) {
+	if (in_softirq() || lgr->terminating) {
 		rc = smc_wr_tx_get_free_slot_index(link, &idx);
 		if (rc)
 			return rc;
 	} else {
-		rc = wait_event_timeout(
+		rc = wait_event_interruptible_timeout(
 			link->wr_tx_wait,
 			link->state == SMC_LNK_INACTIVE ||
+			lgr->terminating ||
 			(smc_wr_tx_get_free_slot_index(link, &idx) != -EBUSY),
 			SMC_WR_TX_WAIT_FREE_SLOT_TIME);
 		if (!rc) {
 			/* timeout - terminate connections */
-			smc_lgr_terminate_sched(smc_get_lgr(link));
+			smc_lgr_terminate_sched(lgr);
 			return -EPIPE;
 		}
 		if (idx == link->wr_tx_cnt)
@@ -227,6 +249,7 @@ int smc_wr_tx_put_slot(struct smc_link *link,
 		memset(&link->wr_tx_bufs[idx], 0,
 		       sizeof(link->wr_tx_bufs[idx]));
 		test_and_clear_bit(idx, link->wr_tx_mask);
+		wake_up(&link->wr_tx_wait);
 		return 1;
 	}
 
@@ -510,8 +533,10 @@ void smc_wr_free_link(struct smc_link *lnk)
 {
 	struct ib_device *ibdev;
 
-	memset(lnk->wr_tx_mask, 0,
-	       BITS_TO_LONGS(SMC_WR_BUF_CNT) * sizeof(*lnk->wr_tx_mask));
+	if (smc_wr_tx_wait_no_pending_sends(lnk))
+		memset(lnk->wr_tx_mask, 0,
+		       BITS_TO_LONGS(SMC_WR_BUF_CNT) *
+						sizeof(*lnk->wr_tx_mask));
 
 	if (!lnk->smcibdev)
 		return;

net/smc/smc_wr.h

@@ -60,6 +60,16 @@ static inline void smc_wr_tx_set_wr_id(atomic_long_t *wr_tx_id, long val)
 	atomic_long_set(wr_tx_id, val);
 }
 
+static inline void smc_wr_wakeup_tx_wait(struct smc_link *lnk)
+{
+	wake_up_all(&lnk->wr_tx_wait);
+}
+
+static inline void smc_wr_wakeup_reg_wait(struct smc_link *lnk)
+{
+	wake_up(&lnk->wr_reg_wait);
+}
+
 /* post a new receive work request to fill a completed old work request entry */
 static inline int smc_wr_rx_post(struct smc_link *link)
 {