[openib-general] [PATCH 3/3] iWARP CM

Tom Tucker tom at opengridcomputing.com
Wed Mar 15 13:18:20 PST 2006


This patch provides the implementation of the iWARP CM: 
        core/Makefile
        core/iwcm.c

Signed-off-by: Tom Tucker <tom at opengridcomputing.com>

Index: infiniband/core/Makefile
===================================================================
--- infiniband/core/Makefile	(revision 5842)
+++ infiniband/core/Makefile	(working copy)
@@ -1,8 +1,9 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include -Idrivers/infiniband/ulp/ipoib
 
-obj-$(CONFIG_INFINIBAND) +=		ib_core.o ib_mad.o ib_ping.o ib_cm.o \
+obj-$(CONFIG_INFINIBAND) +=		ib_core.o ib_mad.o ib_ping.o \
+					ib_cm.o iw_cm.o \
 					ib_sa.o ib_at.o ib_addr.o rdma_cm.o \
-					ib_local_sa.o findex.o
+					ib_local_sa.o findex.o 
 obj-$(CONFIG_INFINIBAND_USER_MAD) += 	ib_umad.o
 obj-$(CONFIG_INFINIBAND_USER_ACCESS) += ib_uverbs.o ib_ucm.o ib_uat.o rdma_ucm.o
 
@@ -17,6 +18,8 @@
 
 ib_cm-y :=			cm.o
 
+iw_cm-y :=			iwcm.o
+
 rdma_cm-y :=			cma.o
 
 rdma_ucm-y :=			ucma.o
Index: infiniband/core/iwcm.c
===================================================================
--- infiniband/core/iwcm.c	(revision 0)
+++ infiniband/core/iwcm.c	(revision 0)
@@ -0,0 +1,713 @@
+/*
+ * Copyright (c) 2004, 2005 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004, 2005 Voltaire Corporation.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
+ * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
+ * Copyright (c) 2005 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/idr.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/rbtree.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_addr.h>
+
+MODULE_AUTHOR("Tom Tucker");
+MODULE_DESCRIPTION("iWARP CM");
+MODULE_LICENSE("Dual BSD/GPL");
+
+struct iwcm_id_private {
+	struct iw_cm_id	id;
+
+	spinlock_t lock;
+	atomic_t refcount;
+};
+
+struct iwcm_work {
+	struct work_struct work;
+	struct iwcm_id_private *cm_id;
+	struct iw_cm_event event;
+};
+
+/* Release a reference on cm_id and delete the object if the
+* reference count goes to zero. 
+ */
+static inline void iwcm_deref_id(struct iwcm_id_private *cm_id_priv)
+{
+	if (atomic_dec_and_test(&cm_id_priv->refcount))
+		kfree(cm_id_priv);
+}
+
+static void cm_event_handler(struct iw_cm_id *cm_id, struct iw_cm_event *event);
+
+struct iw_cm_id *iw_create_cm_id(struct ib_device *device,
+				 iw_cm_handler cm_handler,
+				 void *context)
+{
+	struct iwcm_id_private *iwcm_id_priv;
+
+	iwcm_id_priv = kzalloc(sizeof *iwcm_id_priv, GFP_KERNEL);
+	if (!iwcm_id_priv)
+		return ERR_PTR(-ENOMEM);
+
+	iwcm_id_priv->id.state = IW_CM_STATE_IDLE;
+	iwcm_id_priv->id.device = device;
+	iwcm_id_priv->id.cm_handler = cm_handler;
+	iwcm_id_priv->id.context = context;
+	iwcm_id_priv->id.event_handler = cm_event_handler;
+
+	spin_lock_init(&iwcm_id_priv->lock);
+	atomic_set(&iwcm_id_priv->refcount, 1);
+
+	return &iwcm_id_priv->id;
+
+}
+EXPORT_SYMBOL(iw_create_cm_id);
+
+
+static int iwcm_modify_qp_err(struct ib_qp *qp)
+{
+	struct ib_qp_attr qp_attr;
+
+	if (!qp)
+		return 0;
+
+	qp_attr.qp_state = IB_QPS_ERR;
+	return ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
+}
+
+/* This is really the RDMAC CLOSING state. It is most similar to the
+ * IB SQD QP state. 
+ */
+static int iwcm_modify_qp_sqd(struct ib_qp *qp)
+{
+	struct ib_qp_attr qp_attr;
+
+	if (!qp)
+		return 0;
+
+	qp_attr.qp_state = IB_QPS_SQD;
+	return ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
+}
+
+/* CM_ID <-- CLOSING
+ *
+ * - If we are established move to CLOSING and modify the QP state
+ *   based on the abrupt flag 
+ * - If the connection is already in the CLOSING state, the peer is
+ *   disconnecting concurrently with us and we've already seen the
+ *   DISCONNECT event -- ignore the request and return 0
+ * - If the connection was never established, return -EINVAL
+ */
+int iw_cm_disconnect(struct iw_cm_id *cm_id, int abrupt)
+{
+	struct iwcm_id_private *iwcm_id_priv;
+	unsigned long flags;
+	struct ib_qp *qp;
+	int ret = 0;
+
+	iwcm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+	spin_lock_irqsave(&iwcm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IW_CM_STATE_ESTABLISHED:
+		cm_id->state = IW_CM_STATE_CLOSING;
+		qp = cm_id->qp;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		if (abrupt)
+			ret = iwcm_modify_qp_err(qp);
+		else
+			ret = iwcm_modify_qp_sqd(qp);
+		break;
+	case IW_CM_STATE_CONN_SENT:
+	case IW_CM_STATE_LISTEN:
+	case IW_CM_STATE_CONN_RECV:
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = -EINVAL;
+		break;
+	case IW_CM_STATE_CLOSING:
+	default:		
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL(iw_cm_disconnect);
+
+/* Clean up all resources associated with the connection. Set the
+ * cm_id state to DESTROYING so queued events will be ignored and
+ * subsequent events will not be queued. 
+ */
+void iw_destroy_cm_id(struct iw_cm_id *cm_id)
+{
+	struct iwcm_id_private *iwcm_id_priv;
+	unsigned long flags;
+	struct ib_qp *qp;
+	int ret = 0;
+
+	iwcm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+
+	spin_lock_irqsave(&iwcm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IW_CM_STATE_LISTEN:
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = cm_id->device->iwcm->destroy_listen(cm_id);
+		break;
+	case IW_CM_STATE_CONN_RECV:
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		/* Need to call reject on behalf of client */
+		(void)iw_cm_reject(cm_id, NULL, 0);
+		break;
+	case IW_CM_STATE_CONN_SENT:
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		cm_id->qp = NULL;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		break;
+	case IW_CM_STATE_ESTABLISHED:
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		qp = cm_id->qp;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = iwcm_modify_qp_err(qp);
+		break;
+	case IW_CM_STATE_IDLE:
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		break;
+	case IW_CM_STATE_CLOSING:
+		/* Lost race with cm_disconnect_handler. Need to 
+		 * release 'connect' reference, since setting state to
+		 * DESTROYING will case CLOSE event to be ignored.
+		 */
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		iwcm_deref_id(iwcm_id_priv);
+		break;
+	default:
+		/* Lost race with cm_close_handler. */ 
+		cm_id->state = IW_CM_STATE_DESTROYING;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+	}
+	iwcm_deref_id(iwcm_id_priv);
+}
+EXPORT_SYMBOL(iw_destroy_cm_id);
+
+/* Start listening for connect requests. Generates one CONNECT_REQUEST
+ * event for each inbound connect request. 
+ */
+int iw_cm_listen(struct iw_cm_id *cm_id, int backlog)
+{
+	struct iwcm_id_private *iwcm_id_priv;
+	unsigned long flags;
+	int ret = 0;
+
+	iwcm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+	spin_lock_irqsave(&iwcm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IW_CM_STATE_IDLE:
+		cm_id->state = IW_CM_STATE_LISTEN;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = cm_id->device->iwcm->create_listen(cm_id, backlog);
+		if (ret)
+			cm_id->state = IW_CM_STATE_IDLE;
+		break;
+	default:
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = -EINVAL;
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL(iw_cm_listen);
+
+/* Rejects a connection. No events are generated */
+int iw_cm_reject(struct iw_cm_id *cm_id,
+		 const void *private_data,
+		 u8 private_data_len)
+{
+	struct iwcm_id_private *iwcm_id_priv;
+	unsigned long flags;
+	int ret;
+
+	iwcm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+
+	spin_lock_irqsave(&iwcm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IW_CM_STATE_CONN_RECV:
+		cm_id->state = IW_CM_STATE_IDLE;
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = cm_id->device->iwcm->reject(cm_id, private_data, 
+						  private_data_len);
+		break;
+	default:
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = -EINVAL;
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL(iw_cm_reject);
+
+/* Accepts a connection and generates an ESTABLISHED event. */
+int iw_cm_accept(struct iw_cm_id *cm_id,
+		 const void *private_data,
+		 u8 private_data_len)
+{
+	struct iwcm_id_private *iwcm_id_priv;
+	unsigned long flags;
+	int ret;
+
+	iwcm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+
+	spin_lock_irqsave(&iwcm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IW_CM_STATE_CONN_RECV:
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = cm_id->device->iwcm->accept(cm_id, private_data, 
+						  private_data_len);
+		if (ret == 0) {
+			struct iw_cm_event event;
+			event.event = IW_CM_EVENT_ESTABLISHED;
+			event.provider_id = cm_id->provider_id;
+			event.status = 0;
+			event.local_addr = cm_id->local_addr;
+			event.remote_addr = cm_id->remote_addr;
+			event.private_data = NULL;
+			event.private_data_len = 0;
+			cm_event_handler(cm_id, &event);
+		} else
+			cm_id->state = IW_CM_STATE_IDLE;
+			
+		break;
+	default:
+		spin_unlock_irqrestore(&iwcm_id_priv->lock, flags);
+		ret = -EINVAL;
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL(iw_cm_accept);
+
+/*
+ * Active Side: CM_ID <-- CONN_SENT
+ *
+ * If successful, results in the generation of a CONNECT_REPLY event. 
+ */
+int iw_cm_connect(struct iw_cm_id *cm_id, 
+		  const void *pdata,
+		  u8 pdata_len)
+{
+	struct iwcm_id_private *cm_id_priv;
+	int ret = 0;
+	unsigned long flags;
+
+	cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IW_CM_STATE_IDLE) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		return -EINVAL;
+	}
+	cm_id->state = IW_CM_STATE_CONN_SENT;
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	ret = cm_id->device->iwcm->connect(cm_id, pdata, pdata_len);
+	if (ret)
+		cm_id->state = IW_CM_STATE_IDLE;
+
+	return ret;
+}
+EXPORT_SYMBOL(iw_cm_connect);
+
+
+/*
+ * Passive Side: new CM_ID <-- CONN_RECV
+ *
+ * Handles an inbound connect request. The function creates a new
+ * iw_cm_id to represent the new connection and inherits the client
+ * callback function and other attributes from the listening parent. 
+ * 
+ * The work item contains a pointer to the listen_cm_id and the event. The
+ * listen_cm_id contains the client cm_handler, context and
+ * device. These are copied when the device is cloned. The event
+ * contains the new four tuple.
+ *
+ * An error on the child should not affect the parent, so this
+ * function does not return a value.
+ */
+static void cm_conn_req_handler(struct iwcm_work *work)
+{
+	struct iw_cm_id *cm_id;
+	struct iwcm_id_private *cm_id_priv;
+	struct iwcm_id_private *listen_id_priv = work->cm_id;
+	iw_cm_handler cm_handler;
+	void *context;
+	struct ib_device *device;
+	unsigned long flags;
+	int ret;
+
+	/* If the status was not successful, ignore request */
+	BUG_ON(work->event.status);
+
+	/* We could be destroying the listening id. If so, ignore this
+	 * upcall. */
+	spin_lock_irqsave(&listen_id_priv->lock, flags);
+	if (listen_id_priv->id.state != IW_CM_STATE_LISTEN) {
+		spin_unlock_irqrestore(&listen_id_priv->lock, flags);
+		return;
+	}
+	device = listen_id_priv->id.device;
+	cm_handler = listen_id_priv->id.cm_handler;
+	context = listen_id_priv->id.context;
+	spin_unlock_irqrestore(&listen_id_priv->lock, flags);
+
+	cm_id = iw_create_cm_id(device,	cm_handler, context);
+
+	/* If the cm_id could not be created, ignore the request */
+	if (IS_ERR(cm_id)) 
+		return;
+
+	cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+	cm_id_priv->id.local_addr = work->event.local_addr;
+	cm_id_priv->id.remote_addr = work->event.remote_addr;
+	cm_id_priv->id.provider_id = work->event.provider_id;
+	cm_id_priv->id.state = IW_CM_STATE_CONN_RECV;
+
+	/* Call the client CM handler */
+	atomic_inc(&cm_id_priv->refcount);
+	ret = cm_id->cm_handler(cm_id, &work->event);
+	iwcm_deref_id(cm_id_priv);
+	if (ret) {
+		cm_id->state = IW_CM_STATE_IDLE;
+		iw_destroy_cm_id(cm_id);
+	}
+
+	return;
+}
+
+/*
+ * Passive Side: CM_ID <-- ESTABLISHED
+ * 
+ * If we're in the wrong state, ignore the request. This could
+ * happen if we got a close from the peer.
+ */
+static int cm_conn_est_handler(struct iwcm_work *work)
+{
+	struct iwcm_id_private *cm_id_priv;
+	unsigned long flags;
+	
+	cm_id_priv = work->cm_id;
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IW_CM_STATE_CONN_RECV) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		return 0;
+	}
+
+	cm_id_priv = work->cm_id;
+	atomic_inc(&cm_id_priv->refcount); /* close decrements */
+	cm_id_priv->id.state = IW_CM_STATE_ESTABLISHED;
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	/* Call the client CM handler */
+	return cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->event);
+}
+
+/*
+ * Active Side: CM_ID <-- ESTABLISHED
+ *
+ * Handles the reply to our connect request:
+ * - If the cm_id is in the wrong state when the event is
+ *   delivered, the event is ignored. 
+ * - If the remote peer accepts the connection, we update the 4-tuple
+ *   in the cm_id with the remote peer info, move the cm_id to the
+ *   ESTABLISHED state and deliver the event to the client.
+ * - If the remote peer rejects the connection, or there is some
+ *   connection error, move the cm_id to the IDLE state, and deliver
+ *   the event to the client.
+ */
+static int cm_conn_rep_handler(struct iwcm_work *work)
+{
+	struct iwcm_id_private *cm_id_priv;
+	unsigned long flags;
+	
+	cm_id_priv = work->cm_id;
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IW_CM_STATE_CONN_SENT) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		return 0;
+	}
+
+	if (work->event.status == IW_CM_EVENT_STATUS_ACCEPTED) {
+		cm_id_priv = work->cm_id;
+		cm_id_priv->id.local_addr = work->event.local_addr;
+		cm_id_priv->id.remote_addr = work->event.remote_addr;
+		atomic_inc(&cm_id_priv->refcount); /* close decrements */
+		cm_id_priv->id.state = IW_CM_STATE_ESTABLISHED;
+	} else 
+		cm_id_priv->id.state = IW_CM_STATE_IDLE;
+
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	/* Call the client CM handler */
+	return cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->event);
+}
+
+/*
+ * CM_ID <-- CLOSING 
+ *
+ * If in the ESTABLISHED state, move to closing and disassociate the
+ * CM_ID from the QP. The provider will have already moved the QP to
+ * CLOSING.
+ */
+static void cm_disconnect_handler(struct iwcm_work *work) 
+{
+	struct iwcm_id_private *cm_id_priv;
+	unsigned long flags;
+	
+	cm_id_priv = work->cm_id;
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state == IW_CM_STATE_ESTABLISHED) {
+		cm_id_priv->id.state = IW_CM_STATE_CLOSING;
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	} else
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+}
+
+/*
+ * CM_ID <-- IDLE
+ *
+ * If in the ESTBLISHED state, the QP will have have been moved by the
+ * provider to the ERR state. Disassociate the CM_ID from the QP, 
+ * move to IDLE, and remove the 'connected' reference.
+ *
+ * If in the CLOSING state, the QP has already been disassociated,
+ * move to IDLE and remove 'connected' reference. 
+ *
+ */
+static int cm_close_handler(struct iwcm_work *work)
+{
+	struct iwcm_id_private *cm_id_priv;
+	unsigned long flags;
+
+	cm_id_priv = work->cm_id;
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	cm_id_priv->id.qp = NULL;
+	switch (cm_id_priv->id.state) {
+	case IW_CM_STATE_ESTABLISHED:
+	case IW_CM_STATE_CLOSING:
+		cm_id_priv->id.state = IW_CM_STATE_IDLE;
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		iwcm_deref_id(cm_id_priv);
+		break;
+	default:
+		/* lost race with iw_cm_disconnect */
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		return 0;
+	}
+
+	/* Call the client CM handler */
+	return cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->event);
+}
+
+static void cm_work_handler(void *arg)
+{
+	struct iwcm_work *work = (struct iwcm_work*)arg;
+	struct iwcm_id_private *cm_id_priv = work->cm_id;
+	int ret = 0;
+
+	/* Ignore scheduled events on a destroyed connection */
+	if (cm_id_priv->id.state != IW_CM_STATE_DESTROYING) {
+		switch (work->event.event) {
+		case IW_CM_EVENT_CONNECT_REQUEST:
+			cm_conn_req_handler(work);
+			break;
+		case IW_CM_EVENT_CONNECT_REPLY:
+			ret = cm_conn_rep_handler(work);
+			break;
+		case IW_CM_EVENT_ESTABLISHED:
+			ret = cm_conn_est_handler(work);
+			break;
+		case IW_CM_EVENT_DISCONNECT:
+			cm_disconnect_handler(work);
+			break;
+		case IW_CM_EVENT_CLOSE:
+			ret = cm_close_handler(work);
+			break;
+		default:
+			BUG_ON(1);
+		}
+	}
+	kfree(work);
+
+	/* The reference was done in cm_event_handler. */
+	iwcm_deref_id(cm_id_priv);
+
+	if (ret)
+		iw_destroy_cm_id(&cm_id_priv->id);
+}
+
+/* IW CM provider event callback handler. This function is called on
+ * interrupt context. Schedule events on the rdma_wq thread to allow
+ * callback functions to downcall into the CM and/or block. 
+ *
+ * If the cm_id is being destroyed, ignore all events except CLOSE. In
+ * this case, dump the the ESTABLISHED reference and allow the cm_id
+ * to be freed.
+ */
+static void cm_event_handler(struct iw_cm_id *cm_id,
+			     struct iw_cm_event *iw_event) 
+{
+	struct iwcm_work *work;
+	struct iwcm_id_private *cm_id_priv;
+	unsigned long flags;
+
+	work = kmalloc(sizeof *work, GFP_ATOMIC);
+	if (!work)
+		return;
+
+	cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+	
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IW_CM_STATE_DESTROYING:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		if (iw_event->event == IW_CM_EVENT_CLOSE)
+			iwcm_deref_id(cm_id_priv);
+		kfree(work);
+		break;
+	default:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		/* 
+		 * Reference cm_id_priv until the work is completed in 
+		 * cm_work_handler.
+		 */
+		atomic_inc(&cm_id_priv->refcount);
+		INIT_WORK(&work->work, cm_work_handler, work);
+		work->cm_id = cm_id_priv;
+		work->event = *iw_event;
+		queue_work(rdma_wq, &work->work);
+	}
+}
+
+static int iwcm_init_qp_init_attr(struct iwcm_id_private *cm_id_priv,
+				  struct ib_qp_attr *qp_attr,
+				  int *qp_attr_mask)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IW_CM_STATE_IDLE:
+	case IW_CM_STATE_CONN_SENT:
+	case IW_CM_STATE_CONN_RECV:
+	case IW_CM_STATE_ESTABLISHED:
+		*qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
+		qp_attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE |
+					   IB_ACCESS_REMOTE_WRITE|
+					   IB_ACCESS_REMOTE_READ;
+		ret = 0;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+
+static int iwcm_init_qp_rts_attr(struct iwcm_id_private *cm_id_priv,
+				  struct ib_qp_attr *qp_attr,
+				  int *qp_attr_mask)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IW_CM_STATE_IDLE:
+	case IW_CM_STATE_CONN_SENT:
+	case IW_CM_STATE_CONN_RECV:
+	case IW_CM_STATE_ESTABLISHED:
+		*qp_attr_mask = IB_QP_STATE;
+		ret = 0;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+
+int iw_cm_init_qp_attr(struct iw_cm_id *cm_id,
+		       struct ib_qp_attr *qp_attr,
+		       int *qp_attr_mask)
+{
+	struct iwcm_id_private *cm_id_priv;
+	int ret;
+
+	cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
+	switch (qp_attr->qp_state) {
+	case IB_QPS_INIT:
+	case IB_QPS_RTR:
+		ret = iwcm_init_qp_init_attr(cm_id_priv, 
+					     qp_attr, qp_attr_mask);
+		break;
+	case IB_QPS_RTS:
+		ret = iwcm_init_qp_rts_attr(cm_id_priv, 
+					    qp_attr, qp_attr_mask);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	return ret;
+}
+EXPORT_SYMBOL(iw_cm_init_qp_attr);
+
+static int __init iw_cm_init(void)
+{
+	return 0;
+}
+
+static void __exit iw_cm_cleanup(void)
+{
+}
+
+module_init(iw_cm_init);
+module_exit(iw_cm_cleanup);




More information about the general mailing list