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

Karsten Graul says:

====================
net/smc: improve termination handling

First set of patches to improve termination handling.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>

include/net/smc.h

@@ -75,6 +75,9 @@ struct smcd_dev {
 	struct workqueue_struct *event_wq;
 	u8 pnetid[SMC_MAX_PNETID_LEN];
 	bool pnetid_by_user;
+	struct list_head lgr_list;
+	spinlock_t lgr_lock;
+	u8 going_away : 1;
 };
 
 struct smcd_dev *smcd_alloc_dev(struct device *parent, const char *name,

net/smc/smc_close.c

@@ -65,8 +65,8 @@ static void smc_close_stream_wait(struct smc_sock *smc, long timeout)
 
 		rc = sk_wait_event(sk, &timeout,
 				   !smc_tx_prepared_sends(&smc->conn) ||
-				   (sk->sk_err == ECONNABORTED) ||
-				   (sk->sk_err == ECONNRESET),
+				   sk->sk_err == ECONNABORTED ||
+				   sk->sk_err == ECONNRESET,
 				   &wait);
 		if (rc)
 			break;
@@ -113,9 +113,6 @@ static void smc_close_active_abort(struct smc_sock *smc)
 {
 	struct sock *sk = &smc->sk;
 
-	struct smc_cdc_conn_state_flags *txflags =
-		&smc->conn.local_tx_ctrl.conn_state_flags;
-
 	if (sk->sk_state != SMC_INIT && smc->clcsock && smc->clcsock->sk) {
 		sk->sk_err = ECONNABORTED;
 		if (smc->clcsock && smc->clcsock->sk) {
@@ -129,35 +126,26 @@ static void smc_close_active_abort(struct smc_sock *smc)
 		release_sock(sk);
 		cancel_delayed_work_sync(&smc->conn.tx_work);
 		lock_sock(sk);
+		sk->sk_state = SMC_CLOSED;
 		sock_put(sk); /* passive closing */
 		break;
 	case SMC_APPCLOSEWAIT1:
 	case SMC_APPCLOSEWAIT2:
-		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);
+		sk->sk_state = SMC_CLOSED;
 		break;
 	case SMC_PEERCLOSEWAIT1:
 	case SMC_PEERCLOSEWAIT2:
-		if (!txflags->peer_conn_closed) {
-			/* just SHUTDOWN_SEND done */
-			sk->sk_state = SMC_PEERABORTWAIT;
-		} else {
+	case SMC_PEERFINCLOSEWAIT:
 		sk->sk_state = SMC_CLOSED;
-		}
 		sock_put(sk); /* passive closing */
 		break;
 	case SMC_PROCESSABORT:
 	case SMC_APPFINCLOSEWAIT:
 		sk->sk_state = SMC_CLOSED;
 		break;
-	case SMC_PEERFINCLOSEWAIT:
-		sock_put(sk); /* passive closing */
-		break;
 	case SMC_INIT:
 	case SMC_PEERABORTWAIT:
 	case SMC_CLOSED:
@@ -215,8 +203,6 @@ int smc_close_active(struct smc_sock *smc)
 		if (sk->sk_state == SMC_ACTIVE) {
 			/* send close request */
 			rc = smc_close_final(conn);
-			if (rc)
-				break;
 			sk->sk_state = SMC_PEERCLOSEWAIT1;
 		} else {
 			/* peer event has changed the state */
@@ -229,8 +215,6 @@ int smc_close_active(struct smc_sock *smc)
 		    !smc_close_sent_any_close(conn)) {
 			/* just shutdown wr done, send close request */
 			rc = smc_close_final(conn);
-			if (rc)
-				break;
 		}
 		sk->sk_state = SMC_CLOSED;
 		break;
@@ -246,8 +230,6 @@ int smc_close_active(struct smc_sock *smc)
 			goto again;
 		/* confirm close from peer */
 		rc = smc_close_final(conn);
-		if (rc)
-			break;
 		if (smc_cdc_rxed_any_close(conn)) {
 			/* peer has closed the socket already */
 			sk->sk_state = SMC_CLOSED;
@@ -263,8 +245,6 @@ int smc_close_active(struct smc_sock *smc)
 		    !smc_close_sent_any_close(conn)) {
 			/* just shutdown wr done, send close request */
 			rc = smc_close_final(conn);
-			if (rc)
-				break;
 		}
 		/* peer sending PeerConnectionClosed will cause transition */
 		break;
@@ -272,10 +252,12 @@ int smc_close_active(struct smc_sock *smc)
 		/* peer sending PeerConnectionClosed will cause transition */
 		break;
 	case SMC_PROCESSABORT:
-		smc_close_abort(conn);
+		rc = smc_close_abort(conn);
 		sk->sk_state = SMC_CLOSED;
 		break;
 	case SMC_PEERABORTWAIT:
+		sk->sk_state = SMC_CLOSED;
+		break;
 	case SMC_CLOSED:
 		/* nothing to do, add tracing in future patch */
 		break;
@@ -451,8 +433,6 @@ int smc_close_shutdown_write(struct smc_sock *smc)
 			goto again;
 		/* send close wr request */
 		rc = smc_close_wr(conn);
-		if (rc)
-			break;
 		sk->sk_state = SMC_PEERCLOSEWAIT1;
 		break;
 	case SMC_APPCLOSEWAIT1:
@@ -466,8 +446,6 @@ int smc_close_shutdown_write(struct smc_sock *smc)
 			goto again;
 		/* confirm close from peer */
 		rc = smc_close_wr(conn);
-		if (rc)
-			break;
 		sk->sk_state = SMC_APPCLOSEWAIT2;
 		break;
 	case SMC_APPCLOSEWAIT2:

net/smc/smc_core.c

@@ -42,6 +42,19 @@ static struct smc_lgr_list smc_lgr_list = {	/* established link groups */
 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
 			 struct smc_buf_desc *buf_desc);
 
+/* return head of link group list and its lock for a given link group */
+static inline struct list_head *smc_lgr_list_head(struct smc_link_group *lgr,
+						  spinlock_t **lgr_lock)
+{
+	if (lgr->is_smcd) {
+		*lgr_lock = &lgr->smcd->lgr_lock;
+		return &lgr->smcd->lgr_list;
+	}
+
+	*lgr_lock = &smc_lgr_list.lock;
+	return &smc_lgr_list.list;
+}
+
 static void smc_lgr_schedule_free_work(struct smc_link_group *lgr)
 {
 	/* client link group creation always follows the server link group
@@ -157,19 +170,21 @@ static void smc_lgr_free_work(struct work_struct *work)
 	struct smc_link_group *lgr = container_of(to_delayed_work(work),
 						  struct smc_link_group,
 						  free_work);
+	spinlock_t *lgr_lock;
 	bool conns;
 
-	spin_lock_bh(&smc_lgr_list.lock);
+	smc_lgr_list_head(lgr, &lgr_lock);
+	spin_lock_bh(lgr_lock);
 	read_lock_bh(&lgr->conns_lock);
 	conns = RB_EMPTY_ROOT(&lgr->conns_all);
 	read_unlock_bh(&lgr->conns_lock);
 	if (!conns) { /* number of lgr connections is no longer zero */
-		spin_unlock_bh(&smc_lgr_list.lock);
+		spin_unlock_bh(lgr_lock);
 		return;
 	}
 	if (!list_empty(&lgr->list))
 		list_del_init(&lgr->list); /* remove from smc_lgr_list */
-	spin_unlock_bh(&smc_lgr_list.lock);
+	spin_unlock_bh(lgr_lock);
 
 	if (!lgr->is_smcd && !lgr->terminating)	{
 		struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK];
@@ -198,7 +213,9 @@ static void smc_lgr_free_work(struct work_struct *work)
 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
 {
 	struct smc_link_group *lgr;
+	struct list_head *lgr_list;
 	struct smc_link *lnk;
+	spinlock_t *lgr_lock;
 	u8 rndvec[3];
 	int rc = 0;
 	int i;
@@ -231,10 +248,14 @@ static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
 	lgr->conns_all = RB_ROOT;
 	if (ini->is_smcd) {
 		/* SMC-D specific settings */
+		get_device(&ini->ism_dev->dev);
 		lgr->peer_gid = ini->ism_gid;
 		lgr->smcd = ini->ism_dev;
+		lgr_list = &ini->ism_dev->lgr_list;
+		lgr_lock = &lgr->smcd->lgr_lock;
 	} else {
 		/* SMC-R specific settings */
+		get_device(&ini->ib_dev->ibdev->dev);
 		lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
 		memcpy(lgr->peer_systemid, ini->ib_lcl->id_for_peer,
 		       SMC_SYSTEMID_LEN);
@@ -245,6 +266,8 @@ static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
 		lnk->link_id = SMC_SINGLE_LINK;
 		lnk->smcibdev = ini->ib_dev;
 		lnk->ibport = ini->ib_port;
+		lgr_list = &smc_lgr_list.list;
+		lgr_lock = &smc_lgr_list.lock;
 		lnk->path_mtu =
 			ini->ib_dev->pattr[ini->ib_port - 1].active_mtu;
 		if (!ini->ib_dev->initialized)
@@ -274,9 +297,9 @@ static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
 			goto destroy_qp;
 	}
 	smc->conn.lgr = lgr;
-	spin_lock_bh(&smc_lgr_list.lock);
-	list_add(&lgr->list, &smc_lgr_list.list);
-	spin_unlock_bh(&smc_lgr_list.lock);
+	spin_lock_bh(lgr_lock);
+	list_add(&lgr->list, lgr_list);
+	spin_unlock_bh(lgr_lock);
 	return 0;
 
 destroy_qp:
@@ -430,20 +453,27 @@ static void smc_lgr_free_bufs(struct smc_link_group *lgr)
 static void smc_lgr_free(struct smc_link_group *lgr)
 {
 	smc_lgr_free_bufs(lgr);
-	if (lgr->is_smcd)
+	if (lgr->is_smcd) {
 		smc_ism_put_vlan(lgr->smcd, lgr->vlan_id);
-	else
+		put_device(&lgr->smcd->dev);
+	} else {
 		smc_link_clear(&lgr->lnk[SMC_SINGLE_LINK]);
+		put_device(&lgr->lnk[SMC_SINGLE_LINK].smcibdev->ibdev->dev);
+	}
 	kfree(lgr);
 }
 
 void smc_lgr_forget(struct smc_link_group *lgr)
 {
-	spin_lock_bh(&smc_lgr_list.lock);
+	struct list_head *lgr_list;
+	spinlock_t *lgr_lock;
+
+	lgr_list = smc_lgr_list_head(lgr, &lgr_lock);
+	spin_lock_bh(lgr_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);
+	if (!list_empty(lgr_list))
+		list_del_init(lgr_list);
+	spin_unlock_bh(lgr_lock);
 }
 
 /* terminate linkgroup abnormally */
@@ -484,9 +514,12 @@ static void __smc_lgr_terminate(struct smc_link_group *lgr)
 
 void smc_lgr_terminate(struct smc_link_group *lgr)
 {
-	spin_lock_bh(&smc_lgr_list.lock);
+	spinlock_t *lgr_lock;
+
+	smc_lgr_list_head(lgr, &lgr_lock);
+	spin_lock_bh(lgr_lock);
 	__smc_lgr_terminate(lgr);
-	spin_unlock_bh(&smc_lgr_list.lock);
+	spin_unlock_bh(lgr_lock);
 }
 
 /* Called when IB port is terminated */
@@ -511,16 +544,15 @@ void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid, unsigned short vlan)
 	LIST_HEAD(lgr_free_list);
 
 	/* run common cleanup function and build free list */
-	spin_lock_bh(&smc_lgr_list.lock);
-	list_for_each_entry_safe(lgr, l, &smc_lgr_list.list, list) {
-		if (lgr->is_smcd && lgr->smcd == dev &&
-		    (!peer_gid || lgr->peer_gid == peer_gid) &&
+	spin_lock_bh(&dev->lgr_lock);
+	list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) {
+		if ((!peer_gid || lgr->peer_gid == peer_gid) &&
 		    (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) {
 			__smc_lgr_terminate(lgr);
 			list_move(&lgr->list, &lgr_free_list);
 		}
 	}
-	spin_unlock_bh(&smc_lgr_list.lock);
+	spin_unlock_bh(&dev->lgr_lock);
 
 	/* cancel the regular free workers and actually free lgrs */
 	list_for_each_entry_safe(lgr, l, &lgr_free_list, list) {
@@ -604,10 +636,14 @@ static bool smcd_lgr_match(struct smc_link_group *lgr,
 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
 {
 	struct smc_connection *conn = &smc->conn;
+	struct list_head *lgr_list;
 	struct smc_link_group *lgr;
 	enum smc_lgr_role role;
+	spinlock_t *lgr_lock;
 	int rc = 0;
 
+	lgr_list = ini->is_smcd ? &ini->ism_dev->lgr_list : &smc_lgr_list.list;
+	lgr_lock = ini->is_smcd ? &ini->ism_dev->lgr_lock : &smc_lgr_list.lock;
 	ini->cln_first_contact = SMC_FIRST_CONTACT;
 	role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
 	if (role == SMC_CLNT && ini->srv_first_contact)
@@ -615,8 +651,8 @@ int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
 		goto create;
 
 	/* determine if an existing link group can be reused */
-	spin_lock_bh(&smc_lgr_list.lock);
-	list_for_each_entry(lgr, &smc_lgr_list.list, list) {
+	spin_lock_bh(lgr_lock);
+	list_for_each_entry(lgr, lgr_list, list) {
 		write_lock_bh(&lgr->conns_lock);
 		if ((ini->is_smcd ?
 		     smcd_lgr_match(lgr, ini->ism_dev, ini->ism_gid) :
@@ -636,7 +672,7 @@ int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
 		}
 		write_unlock_bh(&lgr->conns_lock);
 	}
-	spin_unlock_bh(&smc_lgr_list.lock);
+	spin_unlock_bh(lgr_lock);
 
 	if (role == SMC_CLNT && !ini->srv_first_contact &&
 	    ini->cln_first_contact == SMC_FIRST_CONTACT) {
@@ -1024,16 +1060,45 @@ int smc_rmb_rtoken_handling(struct smc_connection *conn,
 	return 0;
 }
 
+static void smc_core_going_away(void)
+{
+	struct smc_ib_device *smcibdev;
+	struct smcd_dev *smcd;
+
+	spin_lock(&smc_ib_devices.lock);
+	list_for_each_entry(smcibdev, &smc_ib_devices.list, list) {
+		int i;
+
+		for (i = 0; i < SMC_MAX_PORTS; i++)
+			set_bit(i, smcibdev->ports_going_away);
+	}
+	spin_unlock(&smc_ib_devices.lock);
+
+	spin_lock(&smcd_dev_list.lock);
+	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
+		smcd->going_away = 1;
+	}
+	spin_unlock(&smcd_dev_list.lock);
+}
+
 /* Called (from smc_exit) when module is removed */
 void smc_core_exit(void)
 {
 	struct smc_link_group *lgr, *lg;
 	LIST_HEAD(lgr_freeing_list);
+	struct smcd_dev *smcd;
+
+	smc_core_going_away();
 
 	spin_lock_bh(&smc_lgr_list.lock);
-	if (!list_empty(&smc_lgr_list.list))
 	list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
 	spin_unlock_bh(&smc_lgr_list.lock);
+
+	spin_lock(&smcd_dev_list.lock);
+	list_for_each_entry(smcd, &smcd_dev_list.list, list)
+		list_splice_init(&smcd->lgr_list, &lgr_freeing_list);
+	spin_unlock(&smcd_dev_list.lock);
+
 	list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
 		list_del_init(&lgr->list);
 		if (!lgr->is_smcd) {

net/smc/smc_ib.c

@@ -242,8 +242,12 @@ static void smc_ib_port_event_work(struct work_struct *work)
 	for_each_set_bit(port_idx, &smcibdev->port_event_mask, SMC_MAX_PORTS) {
 		smc_ib_remember_port_attr(smcibdev, port_idx + 1);
 		clear_bit(port_idx, &smcibdev->port_event_mask);
-		if (!smc_ib_port_active(smcibdev, port_idx + 1))
+		if (!smc_ib_port_active(smcibdev, port_idx + 1)) {
+			set_bit(port_idx, smcibdev->ports_going_away);
 			smc_port_terminate(smcibdev, port_idx + 1);
+		} else {
+			clear_bit(port_idx, smcibdev->ports_going_away);
+		}
 	}
 }
 
@@ -259,8 +263,10 @@ static void smc_ib_global_event_handler(struct ib_event_handler *handler,
 	switch (ibevent->event) {
 	case IB_EVENT_DEVICE_FATAL:
 		/* terminate all ports on device */
-		for (port_idx = 0; port_idx < SMC_MAX_PORTS; port_idx++)
+		for (port_idx = 0; port_idx < SMC_MAX_PORTS; port_idx++) {
 			set_bit(port_idx, &smcibdev->port_event_mask);
+			set_bit(port_idx, smcibdev->ports_going_away);
+		}
 		schedule_work(&smcibdev->port_event_work);
 		break;
 	case IB_EVENT_PORT_ERR:
@@ -269,6 +275,10 @@ static void smc_ib_global_event_handler(struct ib_event_handler *handler,
 		port_idx = ibevent->element.port_num - 1;
 		if (port_idx < SMC_MAX_PORTS) {
 			set_bit(port_idx, &smcibdev->port_event_mask);
+			if (ibevent->event == IB_EVENT_PORT_ERR)
+				set_bit(port_idx, smcibdev->ports_going_away);
+			else if (ibevent->event == IB_EVENT_PORT_ACTIVE)
+				clear_bit(port_idx, smcibdev->ports_going_away);
 			schedule_work(&smcibdev->port_event_work);
 		}
 		break;
@@ -307,6 +317,7 @@ static void smc_ib_qp_event_handler(struct ib_event *ibevent, void *priv)
 		port_idx = ibevent->element.qp->port - 1;
 		if (port_idx < SMC_MAX_PORTS) {
 			set_bit(port_idx, &smcibdev->port_event_mask);
+			set_bit(port_idx, smcibdev->ports_going_away);
 			schedule_work(&smcibdev->port_event_work);
 		}
 		break;

net/smc/smc_ib.h

@@ -47,6 +47,7 @@ struct smc_ib_device {				/* ib-device infos for smc */
 	u8			initialized : 1; /* ib dev CQ, evthdl done */
 	struct work_struct	port_event_work;
 	unsigned long		port_event_mask;
+	DECLARE_BITMAP(ports_going_away, SMC_MAX_PORTS);
 };
 
 struct smc_buf_desc;

net/smc/smc_ism.c

@@ -286,7 +286,9 @@ struct smcd_dev *smcd_alloc_dev(struct device *parent, const char *name,
 	smc_pnetid_by_dev_port(parent, 0, smcd->pnetid);
 
 	spin_lock_init(&smcd->lock);
+	spin_lock_init(&smcd->lgr_lock);
 	INIT_LIST_HEAD(&smcd->vlan);
+	INIT_LIST_HEAD(&smcd->lgr_list);
 	smcd->event_wq = alloc_ordered_workqueue("ism_evt_wq-%s)",
 						 WQ_MEM_RECLAIM, name);
 	if (!smcd->event_wq) {
@@ -313,6 +315,7 @@ void smcd_unregister_dev(struct smcd_dev *smcd)
 	spin_lock(&smcd_dev_list.lock);
 	list_del(&smcd->list);
 	spin_unlock(&smcd_dev_list.lock);
+	smcd->going_away = 1;
 	flush_workqueue(smcd->event_wq);
 	destroy_workqueue(smcd->event_wq);
 	smc_smcd_terminate(smcd, 0, VLAN_VID_MASK);
@@ -342,6 +345,8 @@ void smcd_handle_event(struct smcd_dev *smcd, struct smcd_event *event)
 {
 	struct smc_ism_event_work *wrk;
 
+	if (smcd->going_away)
+		return;
 	/* copy event to event work queue, and let it be handled there */
 	wrk = kmalloc(sizeof(*wrk), GFP_ATOMIC);
 	if (!wrk)

net/smc/smc_pnet.c

@@ -781,6 +781,7 @@ static void smc_pnet_find_rdma_dev(struct net_device *netdev,
 			dev_put(ndev);
 			if (netdev == ndev &&
 			    smc_ib_port_active(ibdev, i) &&
+			    !test_bit(i - 1, ibdev->ports_going_away) &&
 			    !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
 						  ini->ib_gid, NULL)) {
 				ini->ib_dev = ibdev;
@@ -820,6 +821,7 @@ static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev,
 				continue;
 			if (smc_pnet_match(ibdev->pnetid[i - 1], ndev_pnetid) &&
 			    smc_ib_port_active(ibdev, i) &&
+			    !test_bit(i - 1, ibdev->ports_going_away) &&
 			    !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
 						  ini->ib_gid, NULL)) {
 				ini->ib_dev = ibdev;
@@ -846,7 +848,8 @@ static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev,
 
 	spin_lock(&smcd_dev_list.lock);
 	list_for_each_entry(ismdev, &smcd_dev_list.list, list) {
-		if (smc_pnet_match(ismdev->pnetid, ndev_pnetid)) {
+		if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) &&
+		    !ismdev->going_away) {
 			ini->ism_dev = ismdev;
 			break;
 		}