1 files changed, 140 insertions, 99 deletions

diff --git a/drivers/scsi/scsi_transport_iscsi.c b/drivers/scsi/scsi_transport_iscsi.c
index 27951ea05dd4..2c0dd64159b0 100644
--- a/drivers/scsi/scsi_transport_iscsi.c
+++ b/drivers/scsi/scsi_transport_iscsi.c
@@ -86,6 +86,9 @@ struct iscsi_internal {
 	struct transport_container session_cont;
 };
 
+static DEFINE_IDR(iscsi_ep_idr);
+static DEFINE_MUTEX(iscsi_ep_idr_mutex);
+
 static atomic_t iscsi_session_nr; /* sysfs session id for next new session */
 
 static struct workqueue_struct *iscsi_conn_cleanup_workq;
@@ -168,6 +171,11 @@ struct device_attribute dev_attr_##_prefix##_##_name =	\
 static void iscsi_endpoint_release(struct device *dev)
 {
 	struct iscsi_endpoint *ep = iscsi_dev_to_endpoint(dev);
+
+	mutex_lock(&iscsi_ep_idr_mutex);
+	idr_remove(&iscsi_ep_idr, ep->id);
+	mutex_unlock(&iscsi_ep_idr_mutex);
+
 	kfree(ep);
 }
 
@@ -180,7 +188,7 @@ static ssize_t
 show_ep_handle(struct device *dev, struct device_attribute *attr, char *buf)
 {
 	struct iscsi_endpoint *ep = iscsi_dev_to_endpoint(dev);
-	return sysfs_emit(buf, "%llu\n", (unsigned long long) ep->id);
+	return sysfs_emit(buf, "%d\n", ep->id);
 }
 static ISCSI_ATTR(ep, handle, S_IRUGO, show_ep_handle, NULL);
 
@@ -193,48 +201,32 @@ static struct attribute_group iscsi_endpoint_group = {
 	.attrs = iscsi_endpoint_attrs,
 };
 
-#define ISCSI_MAX_EPID -1
-
-static int iscsi_match_epid(struct device *dev, const void *data)
-{
-	struct iscsi_endpoint *ep = iscsi_dev_to_endpoint(dev);
-	const uint64_t *epid = data;
-
-	return *epid == ep->id;
-}
-
 struct iscsi_endpoint *
 iscsi_create_endpoint(int dd_size)
 {
-	struct device *dev;
 	struct iscsi_endpoint *ep;
-	uint64_t id;
-	int err;
-
-	for (id = 1; id < ISCSI_MAX_EPID; id++) {
-		dev = class_find_device(&iscsi_endpoint_class, NULL, &id,
-					iscsi_match_epid);
-		if (!dev)
-			break;
-		else
-			put_device(dev);
-	}
-	if (id == ISCSI_MAX_EPID) {
-		printk(KERN_ERR "Too many connections. Max supported %u\n",
-		       ISCSI_MAX_EPID - 1);
-		return NULL;
-	}
+	int err, id;
 
 	ep = kzalloc(sizeof(*ep) + dd_size, GFP_KERNEL);
 	if (!ep)
 		return NULL;
 
+	mutex_lock(&iscsi_ep_idr_mutex);
+	id = idr_alloc(&iscsi_ep_idr, ep, 0, -1, GFP_NOIO);
+	if (id < 0) {
+		mutex_unlock(&iscsi_ep_idr_mutex);
+		printk(KERN_ERR "Could not allocate endpoint ID. Error %d.\n",
+		       id);
+		goto free_ep;
+	}
+	mutex_unlock(&iscsi_ep_idr_mutex);
+
 	ep->id = id;
 	ep->dev.class = &iscsi_endpoint_class;
-	dev_set_name(&ep->dev, "ep-%llu", (unsigned long long) id);
+	dev_set_name(&ep->dev, "ep-%d", id);
 	err = device_register(&ep->dev);
         if (err)
-                goto free_ep;
+		goto free_id;
 
 	err = sysfs_create_group(&ep->dev.kobj, &iscsi_endpoint_group);
 	if (err)
@@ -248,6 +240,10 @@ unregister_dev:
 	device_unregister(&ep->dev);
 	return NULL;
 
+free_id:
+	mutex_lock(&iscsi_ep_idr_mutex);
+	idr_remove(&iscsi_ep_idr, id);
+	mutex_unlock(&iscsi_ep_idr_mutex);
 free_ep:
 	kfree(ep);
 	return NULL;
@@ -275,14 +271,17 @@ EXPORT_SYMBOL_GPL(iscsi_put_endpoint);
  */
 struct iscsi_endpoint *iscsi_lookup_endpoint(u64 handle)
 {
-	struct device *dev;
+	struct iscsi_endpoint *ep;
 
-	dev = class_find_device(&iscsi_endpoint_class, NULL, &handle,
-				iscsi_match_epid);
-	if (!dev)
-		return NULL;
+	mutex_lock(&iscsi_ep_idr_mutex);
+	ep = idr_find(&iscsi_ep_idr, handle);
+	if (!ep)
+		goto unlock;
 
-	return iscsi_dev_to_endpoint(dev);
+	get_device(&ep->dev);
+unlock:
+	mutex_unlock(&iscsi_ep_idr_mutex);
+	return ep;
 }
 EXPORT_SYMBOL_GPL(iscsi_lookup_endpoint);
 
@@ -2202,10 +2201,10 @@ static void iscsi_stop_conn(struct iscsi_cls_conn *conn, int flag)
 
 	switch (flag) {
 	case STOP_CONN_RECOVER:
-		conn->state = ISCSI_CONN_FAILED;
+		WRITE_ONCE(conn->state, ISCSI_CONN_FAILED);
 		break;
 	case STOP_CONN_TERM:
-		conn->state = ISCSI_CONN_DOWN;
+		WRITE_ONCE(conn->state, ISCSI_CONN_DOWN);
 		break;
 	default:
 		iscsi_cls_conn_printk(KERN_ERR, conn, "invalid stop flag %d\n",
@@ -2217,6 +2216,49 @@ static void iscsi_stop_conn(struct iscsi_cls_conn *conn, int flag)
 	ISCSI_DBG_TRANS_CONN(conn, "Stopping conn done.\n");
 }
 
+static void iscsi_ep_disconnect(struct iscsi_cls_conn *conn, bool is_active)
+{
+	struct iscsi_cls_session *session = iscsi_conn_to_session(conn);
+	struct iscsi_endpoint *ep;
+
+	ISCSI_DBG_TRANS_CONN(conn, "disconnect ep.\n");
+	WRITE_ONCE(conn->state, ISCSI_CONN_FAILED);
+
+	if (!conn->ep || !session->transport->ep_disconnect)
+		return;
+
+	ep = conn->ep;
+	conn->ep = NULL;
+
+	session->transport->unbind_conn(conn, is_active);
+	session->transport->ep_disconnect(ep);
+	ISCSI_DBG_TRANS_CONN(conn, "disconnect ep done.\n");
+}
+
+static void iscsi_if_disconnect_bound_ep(struct iscsi_cls_conn *conn,
+					 struct iscsi_endpoint *ep,
+					 bool is_active)
+{
+	/* Check if this was a conn error and the kernel took ownership */
+	spin_lock_irq(&conn->lock);
+	if (!test_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags)) {
+		spin_unlock_irq(&conn->lock);
+		iscsi_ep_disconnect(conn, is_active);
+	} else {
+		spin_unlock_irq(&conn->lock);
+		ISCSI_DBG_TRANS_CONN(conn, "flush kernel conn cleanup.\n");
+		mutex_unlock(&conn->ep_mutex);
+
+		flush_work(&conn->cleanup_work);
+		/*
+		 * Userspace is now done with the EP so we can release the ref
+		 * iscsi_cleanup_conn_work_fn took.
+		 */
+		iscsi_put_endpoint(ep);
+		mutex_lock(&conn->ep_mutex);
+	}
+}
+
 static int iscsi_if_stop_conn(struct iscsi_transport *transport,
 			      struct iscsi_uevent *ev)
 {
@@ -2238,11 +2280,24 @@ static int iscsi_if_stop_conn(struct iscsi_transport *transport,
 		iscsi_stop_conn(conn, flag);
 	} else {
 		/*
+		 * For offload, when iscsid is restarted it won't know about
+		 * existing endpoints so it can't do a ep_disconnect. We clean
+		 * it up here for userspace.
+		 */
+		mutex_lock(&conn->ep_mutex);
+		if (conn->ep)
+			iscsi_if_disconnect_bound_ep(conn, conn->ep, true);
+		mutex_unlock(&conn->ep_mutex);
+
+		/*
 		 * Figure out if it was the kernel or userspace initiating this.
 		 */
+		spin_lock_irq(&conn->lock);
 		if (!test_and_set_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags)) {
+			spin_unlock_irq(&conn->lock);
 			iscsi_stop_conn(conn, flag);
 		} else {
+			spin_unlock_irq(&conn->lock);
 			ISCSI_DBG_TRANS_CONN(conn,
 					     "flush kernel conn cleanup.\n");
 			flush_work(&conn->cleanup_work);
@@ -2251,31 +2306,14 @@ static int iscsi_if_stop_conn(struct iscsi_transport *transport,
 		 * Only clear for recovery to avoid extra cleanup runs during
 		 * termination.
 		 */
+		spin_lock_irq(&conn->lock);
 		clear_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags);
+		spin_unlock_irq(&conn->lock);
 	}
 	ISCSI_DBG_TRANS_CONN(conn, "iscsi if conn stop done.\n");
 	return 0;
 }
 
-static void iscsi_ep_disconnect(struct iscsi_cls_conn *conn, bool is_active)
-{
-	struct iscsi_cls_session *session = iscsi_conn_to_session(conn);
-	struct iscsi_endpoint *ep;
-
-	ISCSI_DBG_TRANS_CONN(conn, "disconnect ep.\n");
-	conn->state = ISCSI_CONN_FAILED;
-
-	if (!conn->ep || !session->transport->ep_disconnect)
-		return;
-
-	ep = conn->ep;
-	conn->ep = NULL;
-
-	session->transport->unbind_conn(conn, is_active);
-	session->transport->ep_disconnect(ep);
-	ISCSI_DBG_TRANS_CONN(conn, "disconnect ep done.\n");
-}
-
 static void iscsi_cleanup_conn_work_fn(struct work_struct *work)
 {
 	struct iscsi_cls_conn *conn = container_of(work, struct iscsi_cls_conn,
@@ -2284,18 +2322,11 @@ static void iscsi_cleanup_conn_work_fn(struct work_struct *work)
 
 	mutex_lock(&conn->ep_mutex);
 	/*
-	 * If we are not at least bound there is nothing for us to do. Userspace
-	 * will do a ep_disconnect call if offload is used, but will not be
-	 * doing a stop since there is nothing to clean up, so we have to clear
-	 * the cleanup bit here.
+	 * Get a ref to the ep, so we don't release its ID until after
+	 * userspace is done referencing it in iscsi_if_disconnect_bound_ep.
 	 */
-	if (conn->state != ISCSI_CONN_BOUND && conn->state != ISCSI_CONN_UP) {
-		ISCSI_DBG_TRANS_CONN(conn, "Got error while conn is already failed. Ignoring.\n");
-		clear_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags);
-		mutex_unlock(&conn->ep_mutex);
-		return;
-	}
-
+	if (conn->ep)
+		get_device(&conn->ep->dev);
 	iscsi_ep_disconnect(conn, false);
 
 	if (system_state != SYSTEM_RUNNING) {
@@ -2340,11 +2371,12 @@ iscsi_alloc_conn(struct iscsi_cls_session *session, int dd_size, uint32_t cid)
 		conn->dd_data = &conn[1];
 
 	mutex_init(&conn->ep_mutex);
+	spin_lock_init(&conn->lock);
 	INIT_LIST_HEAD(&conn->conn_list);
 	INIT_WORK(&conn->cleanup_work, iscsi_cleanup_conn_work_fn);
 	conn->transport = transport;
 	conn->cid = cid;
-	conn->state = ISCSI_CONN_DOWN;
+	WRITE_ONCE(conn->state, ISCSI_CONN_DOWN);
 
 	/* this is released in the dev's release function */
 	if (!get_device(&session->dev))
@@ -2542,9 +2574,32 @@ void iscsi_conn_error_event(struct iscsi_cls_conn *conn, enum iscsi_err error)
 	struct iscsi_uevent *ev;
 	struct iscsi_internal *priv;
 	int len = nlmsg_total_size(sizeof(*ev));
+	unsigned long flags;
+	int state;
 
-	if (!test_and_set_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags))
-		queue_work(iscsi_conn_cleanup_workq, &conn->cleanup_work);
+	spin_lock_irqsave(&conn->lock, flags);
+	/*
+	 * Userspace will only do a stop call if we are at least bound. And, we
+	 * only need to do the in kernel cleanup if in the UP state so cmds can
+	 * be released to upper layers. If in other states just wait for
+	 * userspace to avoid races that can leave the cleanup_work queued.
+	 */
+	state = READ_ONCE(conn->state);
+	switch (state) {
+	case ISCSI_CONN_BOUND:
+	case ISCSI_CONN_UP:
+		if (!test_and_set_bit(ISCSI_CLS_CONN_BIT_CLEANUP,
+				      &conn->flags)) {
+			queue_work(iscsi_conn_cleanup_workq,
+				   &conn->cleanup_work);
+		}
+		break;
+	default:
+		ISCSI_DBG_TRANS_CONN(conn, "Got conn error in state %d\n",
+				     state);
+		break;
+	}
+	spin_unlock_irqrestore(&conn->lock, flags);
 
 	priv = iscsi_if_transport_lookup(conn->transport);
 	if (!priv)
@@ -2894,7 +2949,7 @@ iscsi_set_param(struct iscsi_transport *transport, struct iscsi_uevent *ev)
 	char *data = (char*)ev + sizeof(*ev);
 	struct iscsi_cls_conn *conn;
 	struct iscsi_cls_session *session;
-	int err = 0, value = 0;
+	int err = 0, value = 0, state;
 
 	if (ev->u.set_param.len > PAGE_SIZE)
 		return -EINVAL;
@@ -2911,8 +2966,8 @@ iscsi_set_param(struct iscsi_transport *transport, struct iscsi_uevent *ev)
 			session->recovery_tmo = value;
 		break;
 	default:
-		if ((conn->state == ISCSI_CONN_BOUND) ||
-			(conn->state == ISCSI_CONN_UP)) {
+		state = READ_ONCE(conn->state);
+		if (state == ISCSI_CONN_BOUND || state == ISCSI_CONN_UP) {
 			err = transport->set_param(conn, ev->u.set_param.param,
 					data, ev->u.set_param.len);
 		} else {
@@ -2984,16 +3039,7 @@ static int iscsi_if_ep_disconnect(struct iscsi_transport *transport,
 	}
 
 	mutex_lock(&conn->ep_mutex);
-	/* Check if this was a conn error and the kernel took ownership */
-	if (test_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags)) {
-		ISCSI_DBG_TRANS_CONN(conn, "flush kernel conn cleanup.\n");
-		mutex_unlock(&conn->ep_mutex);
-
-		flush_work(&conn->cleanup_work);
-		goto put_ep;
-	}
-
-	iscsi_ep_disconnect(conn, false);
+	iscsi_if_disconnect_bound_ep(conn, ep, false);
 	mutex_unlock(&conn->ep_mutex);
 put_ep:
 	iscsi_put_endpoint(ep);
@@ -3696,24 +3742,17 @@ static int iscsi_if_transport_conn(struct iscsi_transport *transport,
 		return -EINVAL;
 
 	mutex_lock(&conn->ep_mutex);
+	spin_lock_irq(&conn->lock);
 	if (test_bit(ISCSI_CLS_CONN_BIT_CLEANUP, &conn->flags)) {
+		spin_unlock_irq(&conn->lock);
 		mutex_unlock(&conn->ep_mutex);
 		ev->r.retcode = -ENOTCONN;
 		return 0;
 	}
+	spin_unlock_irq(&conn->lock);
 
 	switch (nlh->nlmsg_type) {
 	case ISCSI_UEVENT_BIND_CONN:
-		if (conn->ep) {
-			/*
-			 * For offload boot support where iscsid is restarted
-			 * during the pivot root stage, the ep will be intact
-			 * here when the new iscsid instance starts up and
-			 * reconnects.
-			 */
-			iscsi_ep_disconnect(conn, true);
-		}
-
 		session = iscsi_session_lookup(ev->u.b_conn.sid);
 		if (!session) {
 			err = -EINVAL;
@@ -3724,7 +3763,7 @@ static int iscsi_if_transport_conn(struct iscsi_transport *transport,
 						ev->u.b_conn.transport_eph,
 						ev->u.b_conn.is_leading);
 		if (!ev->r.retcode)
-			conn->state = ISCSI_CONN_BOUND;
+			WRITE_ONCE(conn->state, ISCSI_CONN_BOUND);
 
 		if (ev->r.retcode || !transport->ep_connect)
 			break;
@@ -3743,7 +3782,8 @@ static int iscsi_if_transport_conn(struct iscsi_transport *transport,
 	case ISCSI_UEVENT_START_CONN:
 		ev->r.retcode = transport->start_conn(conn);
 		if (!ev->r.retcode)
-			conn->state = ISCSI_CONN_UP;
+			WRITE_ONCE(conn->state, ISCSI_CONN_UP);
+
 		break;
 	case ISCSI_UEVENT_SEND_PDU:
 		pdu_len = nlh->nlmsg_len - sizeof(*nlh) - sizeof(*ev);
@@ -4050,10 +4090,11 @@ static ssize_t show_conn_state(struct device *dev,
 {
 	struct iscsi_cls_conn *conn = iscsi_dev_to_conn(dev->parent);
 	const char *state = "unknown";
+	int conn_state = READ_ONCE(conn->state);
 
-	if (conn->state >= 0 &&
-	    conn->state < ARRAY_SIZE(connection_state_names))
-		state = connection_state_names[conn->state];
+	if (conn_state >= 0 &&
+	    conn_state < ARRAY_SIZE(connection_state_names))
+		state = connection_state_names[conn_state];
 
 	return sysfs_emit(buf, "%s\n", state);
 }