2 * Copyright (c) 2004, 2005 Intel Corporation. All rights reserved.
3 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
4 * Copyright (c) 2004, 2005 Voltaire Corporation. All rights reserved.
5 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
6 * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
7 * Copyright (c) 2005 Network Appliance, Inc. All rights reserved.
9 * This software is available to you under a choice of one of two
10 * licenses. You may choose to be licensed under the terms of the GNU
11 * General Public License (GPL) Version 2, available from the file
12 * COPYING in the main directory of this source tree, or the
13 * OpenIB.org BSD license below:
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
23 * - Redistributions in binary form must reproduce the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer in the documentation and/or other materials
26 * provided with the distribution.
28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38 #include <linux/dma-mapping.h>
39 #include <linux/err.h>
40 #include <linux/idr.h>
41 #include <linux/interrupt.h>
42 #include <linux/rbtree.h>
43 #include <linux/sched.h>
44 #include <linux/spinlock.h>
45 #include <linux/workqueue.h>
46 #include <linux/completion.h>
47 #include <linux/slab.h>
48 #include <linux/module.h>
49 #include <linux/sysctl.h>
51 #include <rdma/iw_cm.h>
52 #include <rdma/ib_addr.h>
53 #include <rdma/iw_portmap.h>
54 #include <rdma/rdma_netlink.h>
58 MODULE_AUTHOR("Tom Tucker");
59 MODULE_DESCRIPTION("iWARP CM");
60 MODULE_LICENSE("Dual BSD/GPL");
62 static const char * const iwcm_rej_reason_strs[] = {
63 [ECONNRESET] = "reset by remote host",
64 [ECONNREFUSED] = "refused by remote application",
65 [ETIMEDOUT] = "setup timeout",
68 const char *__attribute_const__ iwcm_reject_msg(int reason)
72 /* iWARP uses negative errnos */
75 if (index < ARRAY_SIZE(iwcm_rej_reason_strs) &&
76 iwcm_rej_reason_strs[index])
77 return iwcm_rej_reason_strs[index];
79 return "unrecognized reason";
81 EXPORT_SYMBOL(iwcm_reject_msg);
83 static struct rdma_nl_cbs iwcm_nl_cb_table[RDMA_NL_IWPM_NUM_OPS] = {
84 [RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb},
85 [RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb},
86 [RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb},
87 [RDMA_NL_IWPM_REMOTE_INFO] = {.dump = iwpm_remote_info_cb},
88 [RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb},
89 [RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
90 [RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
93 static struct workqueue_struct *iwcm_wq;
95 struct work_struct work;
96 struct iwcm_id_private *cm_id;
97 struct list_head list;
98 struct iw_cm_event event;
99 struct list_head free_list;
102 static unsigned int default_backlog = 256;
104 static struct ctl_table_header *iwcm_ctl_table_hdr;
105 static struct ctl_table iwcm_ctl_table[] = {
107 .procname = "default_backlog",
108 .data = &default_backlog,
109 .maxlen = sizeof(default_backlog),
111 .proc_handler = proc_dointvec,
117 * The following services provide a mechanism for pre-allocating iwcm_work
118 * elements. The design pre-allocates them based on the cm_id type:
119 * LISTENING IDS: Get enough elements preallocated to handle the
121 * ACTIVE IDS: 4: CONNECT_REPLY, ESTABLISHED, DISCONNECT, CLOSE
122 * PASSIVE IDS: 3: ESTABLISHED, DISCONNECT, CLOSE
124 * Allocating them in connect and listen avoids having to deal
125 * with allocation failures on the event upcall from the provider (which
126 * is called in the interrupt context).
128 * One exception is when creating the cm_id for incoming connection requests.
129 * There are two cases:
130 * 1) in the event upcall, cm_event_handler(), for a listening cm_id. If
131 * the backlog is exceeded, then no more connection request events will
132 * be processed. cm_event_handler() returns -ENOMEM in this case. Its up
133 * to the provider to reject the connection request.
134 * 2) in the connection request workqueue handler, cm_conn_req_handler().
135 * If work elements cannot be allocated for the new connect request cm_id,
136 * then IWCM will call the provider reject method. This is ok since
137 * cm_conn_req_handler() runs in the workqueue thread context.
140 static struct iwcm_work *get_work(struct iwcm_id_private *cm_id_priv)
142 struct iwcm_work *work;
144 if (list_empty(&cm_id_priv->work_free_list))
146 work = list_entry(cm_id_priv->work_free_list.next, struct iwcm_work,
148 list_del_init(&work->free_list);
152 static void put_work(struct iwcm_work *work)
154 list_add(&work->free_list, &work->cm_id->work_free_list);
157 static void dealloc_work_entries(struct iwcm_id_private *cm_id_priv)
159 struct list_head *e, *tmp;
161 list_for_each_safe(e, tmp, &cm_id_priv->work_free_list) {
163 kfree(list_entry(e, struct iwcm_work, free_list));
167 static int alloc_work_entries(struct iwcm_id_private *cm_id_priv, int count)
169 struct iwcm_work *work;
171 BUG_ON(!list_empty(&cm_id_priv->work_free_list));
173 work = kmalloc(sizeof(struct iwcm_work), GFP_KERNEL);
175 dealloc_work_entries(cm_id_priv);
178 work->cm_id = cm_id_priv;
179 INIT_LIST_HEAD(&work->list);
186 * Save private data from incoming connection requests to
187 * iw_cm_event, so the low level driver doesn't have to. Adjust
188 * the event ptr to point to the local copy.
190 static int copy_private_data(struct iw_cm_event *event)
194 p = kmemdup(event->private_data, event->private_data_len, GFP_ATOMIC);
197 event->private_data = p;
201 static void free_cm_id(struct iwcm_id_private *cm_id_priv)
203 dealloc_work_entries(cm_id_priv);
208 * Release a reference on cm_id. If the last reference is being
209 * released, free the cm_id and return 1.
211 static int iwcm_deref_id(struct iwcm_id_private *cm_id_priv)
213 BUG_ON(atomic_read(&cm_id_priv->refcount)==0);
214 if (atomic_dec_and_test(&cm_id_priv->refcount)) {
215 BUG_ON(!list_empty(&cm_id_priv->work_list));
216 free_cm_id(cm_id_priv);
223 static void add_ref(struct iw_cm_id *cm_id)
225 struct iwcm_id_private *cm_id_priv;
226 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
227 atomic_inc(&cm_id_priv->refcount);
230 static void rem_ref(struct iw_cm_id *cm_id)
232 struct iwcm_id_private *cm_id_priv;
234 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
236 (void)iwcm_deref_id(cm_id_priv);
239 static int cm_event_handler(struct iw_cm_id *cm_id, struct iw_cm_event *event);
241 struct iw_cm_id *iw_create_cm_id(struct ib_device *device,
242 iw_cm_handler cm_handler,
245 struct iwcm_id_private *cm_id_priv;
247 cm_id_priv = kzalloc(sizeof(*cm_id_priv), GFP_KERNEL);
249 return ERR_PTR(-ENOMEM);
251 cm_id_priv->state = IW_CM_STATE_IDLE;
252 cm_id_priv->id.device = device;
253 cm_id_priv->id.cm_handler = cm_handler;
254 cm_id_priv->id.context = context;
255 cm_id_priv->id.event_handler = cm_event_handler;
256 cm_id_priv->id.add_ref = add_ref;
257 cm_id_priv->id.rem_ref = rem_ref;
258 spin_lock_init(&cm_id_priv->lock);
259 atomic_set(&cm_id_priv->refcount, 1);
260 init_waitqueue_head(&cm_id_priv->connect_wait);
261 init_completion(&cm_id_priv->destroy_comp);
262 INIT_LIST_HEAD(&cm_id_priv->work_list);
263 INIT_LIST_HEAD(&cm_id_priv->work_free_list);
265 return &cm_id_priv->id;
267 EXPORT_SYMBOL(iw_create_cm_id);
270 static int iwcm_modify_qp_err(struct ib_qp *qp)
272 struct ib_qp_attr qp_attr;
277 qp_attr.qp_state = IB_QPS_ERR;
278 return ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
282 * This is really the RDMAC CLOSING state. It is most similar to the
285 static int iwcm_modify_qp_sqd(struct ib_qp *qp)
287 struct ib_qp_attr qp_attr;
290 qp_attr.qp_state = IB_QPS_SQD;
291 return ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
297 * Block if a passive or active connection is currently being processed. Then
298 * process the event as follows:
299 * - If we are ESTABLISHED, move to CLOSING and modify the QP state
300 * based on the abrupt flag
301 * - If the connection is already in the CLOSING or IDLE state, the peer is
302 * disconnecting concurrently with us and we've already seen the
303 * DISCONNECT event -- ignore the request and return 0
304 * - Disconnect on a listening endpoint returns -EINVAL
306 int iw_cm_disconnect(struct iw_cm_id *cm_id, int abrupt)
308 struct iwcm_id_private *cm_id_priv;
311 struct ib_qp *qp = NULL;
313 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
314 /* Wait if we're currently in a connect or accept downcall */
315 wait_event(cm_id_priv->connect_wait,
316 !test_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags));
318 spin_lock_irqsave(&cm_id_priv->lock, flags);
319 switch (cm_id_priv->state) {
320 case IW_CM_STATE_ESTABLISHED:
321 cm_id_priv->state = IW_CM_STATE_CLOSING;
323 /* QP could be <nul> for user-mode client */
329 case IW_CM_STATE_LISTEN:
332 case IW_CM_STATE_CLOSING:
333 /* remote peer closed first */
334 case IW_CM_STATE_IDLE:
335 /* accept or connect returned !0 */
337 case IW_CM_STATE_CONN_RECV:
339 * App called disconnect before/without calling accept after
340 * connect_request event delivered.
343 case IW_CM_STATE_CONN_SENT:
344 /* Can only get here if wait above fails */
348 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
352 ret = iwcm_modify_qp_err(qp);
354 ret = iwcm_modify_qp_sqd(qp);
357 * If both sides are disconnecting the QP could
358 * already be in ERR or SQD states
365 EXPORT_SYMBOL(iw_cm_disconnect);
368 * CM_ID <-- DESTROYING
370 * Clean up all resources associated with the connection and release
371 * the initial reference taken by iw_create_cm_id.
373 static void destroy_cm_id(struct iw_cm_id *cm_id)
375 struct iwcm_id_private *cm_id_priv;
378 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
380 * Wait if we're currently in a connect or accept downcall. A
381 * listening endpoint should never block here.
383 wait_event(cm_id_priv->connect_wait,
384 !test_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags));
387 * Since we're deleting the cm_id, drop any events that
388 * might arrive before the last dereference.
390 set_bit(IWCM_F_DROP_EVENTS, &cm_id_priv->flags);
392 spin_lock_irqsave(&cm_id_priv->lock, flags);
393 switch (cm_id_priv->state) {
394 case IW_CM_STATE_LISTEN:
395 cm_id_priv->state = IW_CM_STATE_DESTROYING;
396 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
397 /* destroy the listening endpoint */
398 cm_id->device->iwcm->destroy_listen(cm_id);
399 spin_lock_irqsave(&cm_id_priv->lock, flags);
401 case IW_CM_STATE_ESTABLISHED:
402 cm_id_priv->state = IW_CM_STATE_DESTROYING;
403 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
404 /* Abrupt close of the connection */
405 (void)iwcm_modify_qp_err(cm_id_priv->qp);
406 spin_lock_irqsave(&cm_id_priv->lock, flags);
408 case IW_CM_STATE_IDLE:
409 case IW_CM_STATE_CLOSING:
410 cm_id_priv->state = IW_CM_STATE_DESTROYING;
412 case IW_CM_STATE_CONN_RECV:
414 * App called destroy before/without calling accept after
415 * receiving connection request event notification or
416 * returned non zero from the event callback function.
417 * In either case, must tell the provider to reject.
419 cm_id_priv->state = IW_CM_STATE_DESTROYING;
420 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
421 cm_id->device->iwcm->reject(cm_id, NULL, 0);
422 spin_lock_irqsave(&cm_id_priv->lock, flags);
424 case IW_CM_STATE_CONN_SENT:
425 case IW_CM_STATE_DESTROYING:
430 if (cm_id_priv->qp) {
431 cm_id_priv->id.device->iwcm->rem_ref(cm_id_priv->qp);
432 cm_id_priv->qp = NULL;
434 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
437 iwpm_remove_mapinfo(&cm_id->local_addr, &cm_id->m_local_addr);
438 iwpm_remove_mapping(&cm_id->local_addr, RDMA_NL_IWCM);
441 (void)iwcm_deref_id(cm_id_priv);
445 * This function is only called by the application thread and cannot
446 * be called by the event thread. The function will wait for all
447 * references to be released on the cm_id and then kfree the cm_id
450 void iw_destroy_cm_id(struct iw_cm_id *cm_id)
452 destroy_cm_id(cm_id);
454 EXPORT_SYMBOL(iw_destroy_cm_id);
457 * iw_cm_check_wildcard - If IP address is 0 then use original
458 * @pm_addr: sockaddr containing the ip to check for wildcard
459 * @cm_addr: sockaddr containing the actual IP address
460 * @cm_outaddr: sockaddr to set IP addr which leaving port
462 * Checks the pm_addr for wildcard and then sets cm_outaddr's
463 * IP to the actual (cm_addr).
465 static void iw_cm_check_wildcard(struct sockaddr_storage *pm_addr,
466 struct sockaddr_storage *cm_addr,
467 struct sockaddr_storage *cm_outaddr)
469 if (pm_addr->ss_family == AF_INET) {
470 struct sockaddr_in *pm4_addr = (struct sockaddr_in *)pm_addr;
472 if (pm4_addr->sin_addr.s_addr == htonl(INADDR_ANY)) {
473 struct sockaddr_in *cm4_addr =
474 (struct sockaddr_in *)cm_addr;
475 struct sockaddr_in *cm4_outaddr =
476 (struct sockaddr_in *)cm_outaddr;
478 cm4_outaddr->sin_addr = cm4_addr->sin_addr;
481 struct sockaddr_in6 *pm6_addr = (struct sockaddr_in6 *)pm_addr;
483 if (ipv6_addr_type(&pm6_addr->sin6_addr) == IPV6_ADDR_ANY) {
484 struct sockaddr_in6 *cm6_addr =
485 (struct sockaddr_in6 *)cm_addr;
486 struct sockaddr_in6 *cm6_outaddr =
487 (struct sockaddr_in6 *)cm_outaddr;
489 cm6_outaddr->sin6_addr = cm6_addr->sin6_addr;
495 * iw_cm_map - Use portmapper to map the ports
496 * @cm_id: connection manager pointer
497 * @active: Indicates the active side when true
498 * returns nonzero for error only if iwpm_create_mapinfo() fails
500 * Tries to add a mapping for a port using the Portmapper. If
501 * successful in mapping the IP/Port it will check the remote
502 * mapped IP address for a wildcard IP address and replace the
503 * zero IP address with the remote_addr.
505 static int iw_cm_map(struct iw_cm_id *cm_id, bool active)
507 struct iwpm_dev_data pm_reg_msg;
508 struct iwpm_sa_data pm_msg;
511 cm_id->m_local_addr = cm_id->local_addr;
512 cm_id->m_remote_addr = cm_id->remote_addr;
514 memcpy(pm_reg_msg.dev_name, cm_id->device->name,
515 sizeof(pm_reg_msg.dev_name));
516 memcpy(pm_reg_msg.if_name, cm_id->device->iwcm->ifname,
517 sizeof(pm_reg_msg.if_name));
519 if (iwpm_register_pid(&pm_reg_msg, RDMA_NL_IWCM) ||
523 cm_id->mapped = true;
524 pm_msg.loc_addr = cm_id->local_addr;
525 pm_msg.rem_addr = cm_id->remote_addr;
527 status = iwpm_add_and_query_mapping(&pm_msg,
530 status = iwpm_add_mapping(&pm_msg, RDMA_NL_IWCM);
533 cm_id->m_local_addr = pm_msg.mapped_loc_addr;
535 cm_id->m_remote_addr = pm_msg.mapped_rem_addr;
536 iw_cm_check_wildcard(&pm_msg.mapped_rem_addr,
538 &cm_id->m_remote_addr);
542 return iwpm_create_mapinfo(&cm_id->local_addr,
543 &cm_id->m_local_addr,
550 * Start listening for connect requests. Generates one CONNECT_REQUEST
551 * event for each inbound connect request.
553 int iw_cm_listen(struct iw_cm_id *cm_id, int backlog)
555 struct iwcm_id_private *cm_id_priv;
559 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
562 backlog = default_backlog;
564 ret = alloc_work_entries(cm_id_priv, backlog);
568 spin_lock_irqsave(&cm_id_priv->lock, flags);
569 switch (cm_id_priv->state) {
570 case IW_CM_STATE_IDLE:
571 cm_id_priv->state = IW_CM_STATE_LISTEN;
572 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
573 ret = iw_cm_map(cm_id, false);
575 ret = cm_id->device->iwcm->create_listen(cm_id, backlog);
577 cm_id_priv->state = IW_CM_STATE_IDLE;
578 spin_lock_irqsave(&cm_id_priv->lock, flags);
583 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
587 EXPORT_SYMBOL(iw_cm_listen);
592 * Rejects an inbound connection request. No events are generated.
594 int iw_cm_reject(struct iw_cm_id *cm_id,
595 const void *private_data,
598 struct iwcm_id_private *cm_id_priv;
602 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
603 set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
605 spin_lock_irqsave(&cm_id_priv->lock, flags);
606 if (cm_id_priv->state != IW_CM_STATE_CONN_RECV) {
607 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
608 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
609 wake_up_all(&cm_id_priv->connect_wait);
612 cm_id_priv->state = IW_CM_STATE_IDLE;
613 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
615 ret = cm_id->device->iwcm->reject(cm_id, private_data,
618 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
619 wake_up_all(&cm_id_priv->connect_wait);
623 EXPORT_SYMBOL(iw_cm_reject);
626 * CM_ID <-- ESTABLISHED
628 * Accepts an inbound connection request and generates an ESTABLISHED
629 * event. Callers of iw_cm_disconnect and iw_destroy_cm_id will block
630 * until the ESTABLISHED event is received from the provider.
632 int iw_cm_accept(struct iw_cm_id *cm_id,
633 struct iw_cm_conn_param *iw_param)
635 struct iwcm_id_private *cm_id_priv;
640 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
641 set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
643 spin_lock_irqsave(&cm_id_priv->lock, flags);
644 if (cm_id_priv->state != IW_CM_STATE_CONN_RECV) {
645 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
646 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
647 wake_up_all(&cm_id_priv->connect_wait);
650 /* Get the ib_qp given the QPN */
651 qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn);
653 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
654 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
655 wake_up_all(&cm_id_priv->connect_wait);
658 cm_id->device->iwcm->add_ref(qp);
660 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
662 ret = cm_id->device->iwcm->accept(cm_id, iw_param);
664 /* An error on accept precludes provider events */
665 BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_RECV);
666 cm_id_priv->state = IW_CM_STATE_IDLE;
667 spin_lock_irqsave(&cm_id_priv->lock, flags);
668 if (cm_id_priv->qp) {
669 cm_id->device->iwcm->rem_ref(qp);
670 cm_id_priv->qp = NULL;
672 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
673 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
674 wake_up_all(&cm_id_priv->connect_wait);
679 EXPORT_SYMBOL(iw_cm_accept);
682 * Active Side: CM_ID <-- CONN_SENT
684 * If successful, results in the generation of a CONNECT_REPLY
685 * event. iw_cm_disconnect and iw_cm_destroy will block until the
686 * CONNECT_REPLY event is received from the provider.
688 int iw_cm_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
690 struct iwcm_id_private *cm_id_priv;
695 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
697 ret = alloc_work_entries(cm_id_priv, 4);
701 set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
702 spin_lock_irqsave(&cm_id_priv->lock, flags);
704 if (cm_id_priv->state != IW_CM_STATE_IDLE) {
709 /* Get the ib_qp given the QPN */
710 qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn);
715 cm_id->device->iwcm->add_ref(qp);
717 cm_id_priv->state = IW_CM_STATE_CONN_SENT;
718 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
720 ret = iw_cm_map(cm_id, true);
722 ret = cm_id->device->iwcm->connect(cm_id, iw_param);
724 return 0; /* success */
726 spin_lock_irqsave(&cm_id_priv->lock, flags);
727 if (cm_id_priv->qp) {
728 cm_id->device->iwcm->rem_ref(qp);
729 cm_id_priv->qp = NULL;
731 cm_id_priv->state = IW_CM_STATE_IDLE;
733 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
734 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
735 wake_up_all(&cm_id_priv->connect_wait);
738 EXPORT_SYMBOL(iw_cm_connect);
741 * Passive Side: new CM_ID <-- CONN_RECV
743 * Handles an inbound connect request. The function creates a new
744 * iw_cm_id to represent the new connection and inherits the client
745 * callback function and other attributes from the listening parent.
747 * The work item contains a pointer to the listen_cm_id and the event. The
748 * listen_cm_id contains the client cm_handler, context and
749 * device. These are copied when the device is cloned. The event
750 * contains the new four tuple.
752 * An error on the child should not affect the parent, so this
753 * function does not return a value.
755 static void cm_conn_req_handler(struct iwcm_id_private *listen_id_priv,
756 struct iw_cm_event *iw_event)
759 struct iw_cm_id *cm_id;
760 struct iwcm_id_private *cm_id_priv;
764 * The provider should never generate a connection request
765 * event with a bad status.
767 BUG_ON(iw_event->status);
769 cm_id = iw_create_cm_id(listen_id_priv->id.device,
770 listen_id_priv->id.cm_handler,
771 listen_id_priv->id.context);
772 /* If the cm_id could not be created, ignore the request */
776 cm_id->provider_data = iw_event->provider_data;
777 cm_id->m_local_addr = iw_event->local_addr;
778 cm_id->m_remote_addr = iw_event->remote_addr;
779 cm_id->local_addr = listen_id_priv->id.local_addr;
781 ret = iwpm_get_remote_info(&listen_id_priv->id.m_local_addr,
782 &iw_event->remote_addr,
786 cm_id->remote_addr = iw_event->remote_addr;
788 iw_cm_check_wildcard(&listen_id_priv->id.m_local_addr,
789 &iw_event->local_addr,
791 iw_event->local_addr = cm_id->local_addr;
792 iw_event->remote_addr = cm_id->remote_addr;
795 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
796 cm_id_priv->state = IW_CM_STATE_CONN_RECV;
799 * We could be destroying the listening id. If so, ignore this
802 spin_lock_irqsave(&listen_id_priv->lock, flags);
803 if (listen_id_priv->state != IW_CM_STATE_LISTEN) {
804 spin_unlock_irqrestore(&listen_id_priv->lock, flags);
805 iw_cm_reject(cm_id, NULL, 0);
806 iw_destroy_cm_id(cm_id);
809 spin_unlock_irqrestore(&listen_id_priv->lock, flags);
811 ret = alloc_work_entries(cm_id_priv, 3);
813 iw_cm_reject(cm_id, NULL, 0);
814 iw_destroy_cm_id(cm_id);
818 /* Call the client CM handler */
819 ret = cm_id->cm_handler(cm_id, iw_event);
821 iw_cm_reject(cm_id, NULL, 0);
822 iw_destroy_cm_id(cm_id);
826 if (iw_event->private_data_len)
827 kfree(iw_event->private_data);
831 * Passive Side: CM_ID <-- ESTABLISHED
833 * The provider generated an ESTABLISHED event which means that
834 * the MPA negotion has completed successfully and we are now in MPA
837 * This event can only be received in the CONN_RECV state. If the
838 * remote peer closed, the ESTABLISHED event would be received followed
839 * by the CLOSE event. If the app closes, it will block until we wake
840 * it up after processing this event.
842 static int cm_conn_est_handler(struct iwcm_id_private *cm_id_priv,
843 struct iw_cm_event *iw_event)
848 spin_lock_irqsave(&cm_id_priv->lock, flags);
851 * We clear the CONNECT_WAIT bit here to allow the callback
852 * function to call iw_cm_disconnect. Calling iw_destroy_cm_id
853 * from a callback handler is not allowed.
855 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
856 BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_RECV);
857 cm_id_priv->state = IW_CM_STATE_ESTABLISHED;
858 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
859 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event);
860 wake_up_all(&cm_id_priv->connect_wait);
866 * Active Side: CM_ID <-- ESTABLISHED
868 * The app has called connect and is waiting for the established event to
869 * post it's requests to the server. This event will wake up anyone
870 * blocked in iw_cm_disconnect or iw_destroy_id.
872 static int cm_conn_rep_handler(struct iwcm_id_private *cm_id_priv,
873 struct iw_cm_event *iw_event)
878 spin_lock_irqsave(&cm_id_priv->lock, flags);
880 * Clear the connect wait bit so a callback function calling
881 * iw_cm_disconnect will not wait and deadlock this thread
883 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
884 BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_SENT);
885 if (iw_event->status == 0) {
886 cm_id_priv->id.m_local_addr = iw_event->local_addr;
887 cm_id_priv->id.m_remote_addr = iw_event->remote_addr;
888 iw_event->local_addr = cm_id_priv->id.local_addr;
889 iw_event->remote_addr = cm_id_priv->id.remote_addr;
890 cm_id_priv->state = IW_CM_STATE_ESTABLISHED;
892 /* REJECTED or RESET */
893 cm_id_priv->id.device->iwcm->rem_ref(cm_id_priv->qp);
894 cm_id_priv->qp = NULL;
895 cm_id_priv->state = IW_CM_STATE_IDLE;
897 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
898 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event);
900 if (iw_event->private_data_len)
901 kfree(iw_event->private_data);
903 /* Wake up waiters on connect complete */
904 wake_up_all(&cm_id_priv->connect_wait);
912 * If in the ESTABLISHED state, move to CLOSING.
914 static void cm_disconnect_handler(struct iwcm_id_private *cm_id_priv,
915 struct iw_cm_event *iw_event)
919 spin_lock_irqsave(&cm_id_priv->lock, flags);
920 if (cm_id_priv->state == IW_CM_STATE_ESTABLISHED)
921 cm_id_priv->state = IW_CM_STATE_CLOSING;
922 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
928 * If in the ESTBLISHED or CLOSING states, the QP will have have been
929 * moved by the provider to the ERR state. Disassociate the CM_ID from
930 * the QP, move to IDLE, and remove the 'connected' reference.
932 * If in some other state, the cm_id was destroyed asynchronously.
933 * This is the last reference that will result in waking up
934 * the app thread blocked in iw_destroy_cm_id.
936 static int cm_close_handler(struct iwcm_id_private *cm_id_priv,
937 struct iw_cm_event *iw_event)
941 spin_lock_irqsave(&cm_id_priv->lock, flags);
943 if (cm_id_priv->qp) {
944 cm_id_priv->id.device->iwcm->rem_ref(cm_id_priv->qp);
945 cm_id_priv->qp = NULL;
947 switch (cm_id_priv->state) {
948 case IW_CM_STATE_ESTABLISHED:
949 case IW_CM_STATE_CLOSING:
950 cm_id_priv->state = IW_CM_STATE_IDLE;
951 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
952 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event);
953 spin_lock_irqsave(&cm_id_priv->lock, flags);
955 case IW_CM_STATE_DESTROYING:
960 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
965 static int process_event(struct iwcm_id_private *cm_id_priv,
966 struct iw_cm_event *iw_event)
970 switch (iw_event->event) {
971 case IW_CM_EVENT_CONNECT_REQUEST:
972 cm_conn_req_handler(cm_id_priv, iw_event);
974 case IW_CM_EVENT_CONNECT_REPLY:
975 ret = cm_conn_rep_handler(cm_id_priv, iw_event);
977 case IW_CM_EVENT_ESTABLISHED:
978 ret = cm_conn_est_handler(cm_id_priv, iw_event);
980 case IW_CM_EVENT_DISCONNECT:
981 cm_disconnect_handler(cm_id_priv, iw_event);
983 case IW_CM_EVENT_CLOSE:
984 ret = cm_close_handler(cm_id_priv, iw_event);
994 * Process events on the work_list for the cm_id. If the callback
995 * function requests that the cm_id be deleted, a flag is set in the
996 * cm_id flags to indicate that when the last reference is
997 * removed, the cm_id is to be destroyed. This is necessary to
998 * distinguish between an object that will be destroyed by the app
999 * thread asleep on the destroy_comp list vs. an object destroyed
1000 * here synchronously when the last reference is removed.
1002 static void cm_work_handler(struct work_struct *_work)
1004 struct iwcm_work *work = container_of(_work, struct iwcm_work, work);
1005 struct iw_cm_event levent;
1006 struct iwcm_id_private *cm_id_priv = work->cm_id;
1007 unsigned long flags;
1011 spin_lock_irqsave(&cm_id_priv->lock, flags);
1012 empty = list_empty(&cm_id_priv->work_list);
1014 work = list_entry(cm_id_priv->work_list.next,
1015 struct iwcm_work, list);
1016 list_del_init(&work->list);
1017 empty = list_empty(&cm_id_priv->work_list);
1018 levent = work->event;
1020 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1022 if (!test_bit(IWCM_F_DROP_EVENTS, &cm_id_priv->flags)) {
1023 ret = process_event(cm_id_priv, &levent);
1025 destroy_cm_id(&cm_id_priv->id);
1027 pr_debug("dropping event %d\n", levent.event);
1028 if (iwcm_deref_id(cm_id_priv))
1032 spin_lock_irqsave(&cm_id_priv->lock, flags);
1034 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1038 * This function is called on interrupt context. Schedule events on
1039 * the iwcm_wq thread to allow callback functions to downcall into
1040 * the CM and/or block. Events are queued to a per-CM_ID
1041 * work_list. If this is the first event on the work_list, the work
1042 * element is also queued on the iwcm_wq thread.
1044 * Each event holds a reference on the cm_id. Until the last posted
1045 * event has been delivered and processed, the cm_id cannot be
1049 * 0 - the event was handled.
1050 * -ENOMEM - the event was not handled due to lack of resources.
1052 static int cm_event_handler(struct iw_cm_id *cm_id,
1053 struct iw_cm_event *iw_event)
1055 struct iwcm_work *work;
1056 struct iwcm_id_private *cm_id_priv;
1057 unsigned long flags;
1060 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
1062 spin_lock_irqsave(&cm_id_priv->lock, flags);
1063 work = get_work(cm_id_priv);
1069 INIT_WORK(&work->work, cm_work_handler);
1070 work->cm_id = cm_id_priv;
1071 work->event = *iw_event;
1073 if ((work->event.event == IW_CM_EVENT_CONNECT_REQUEST ||
1074 work->event.event == IW_CM_EVENT_CONNECT_REPLY) &&
1075 work->event.private_data_len) {
1076 ret = copy_private_data(&work->event);
1083 atomic_inc(&cm_id_priv->refcount);
1084 if (list_empty(&cm_id_priv->work_list)) {
1085 list_add_tail(&work->list, &cm_id_priv->work_list);
1086 queue_work(iwcm_wq, &work->work);
1088 list_add_tail(&work->list, &cm_id_priv->work_list);
1090 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1094 static int iwcm_init_qp_init_attr(struct iwcm_id_private *cm_id_priv,
1095 struct ib_qp_attr *qp_attr,
1098 unsigned long flags;
1101 spin_lock_irqsave(&cm_id_priv->lock, flags);
1102 switch (cm_id_priv->state) {
1103 case IW_CM_STATE_IDLE:
1104 case IW_CM_STATE_CONN_SENT:
1105 case IW_CM_STATE_CONN_RECV:
1106 case IW_CM_STATE_ESTABLISHED:
1107 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1108 qp_attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE|
1109 IB_ACCESS_REMOTE_READ;
1116 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1120 static int iwcm_init_qp_rts_attr(struct iwcm_id_private *cm_id_priv,
1121 struct ib_qp_attr *qp_attr,
1124 unsigned long flags;
1127 spin_lock_irqsave(&cm_id_priv->lock, flags);
1128 switch (cm_id_priv->state) {
1129 case IW_CM_STATE_IDLE:
1130 case IW_CM_STATE_CONN_SENT:
1131 case IW_CM_STATE_CONN_RECV:
1132 case IW_CM_STATE_ESTABLISHED:
1140 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1144 int iw_cm_init_qp_attr(struct iw_cm_id *cm_id,
1145 struct ib_qp_attr *qp_attr,
1148 struct iwcm_id_private *cm_id_priv;
1151 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
1152 switch (qp_attr->qp_state) {
1155 ret = iwcm_init_qp_init_attr(cm_id_priv,
1156 qp_attr, qp_attr_mask);
1159 ret = iwcm_init_qp_rts_attr(cm_id_priv,
1160 qp_attr, qp_attr_mask);
1168 EXPORT_SYMBOL(iw_cm_init_qp_attr);
1170 static int __init iw_cm_init(void)
1174 ret = iwpm_init(RDMA_NL_IWCM);
1178 iwcm_wq = alloc_ordered_workqueue("iw_cm_wq", 0);
1182 iwcm_ctl_table_hdr = register_net_sysctl(&init_net, "net/iw_cm",
1184 if (!iwcm_ctl_table_hdr) {
1185 pr_err("iw_cm: couldn't register sysctl paths\n");
1189 rdma_nl_register(RDMA_NL_IWCM, iwcm_nl_cb_table);
1193 destroy_workqueue(iwcm_wq);
1195 iwpm_exit(RDMA_NL_IWCM);
1199 static void __exit iw_cm_cleanup(void)
1201 rdma_nl_unregister(RDMA_NL_IWCM);
1202 unregister_net_sysctl_table(iwcm_ctl_table_hdr);
1203 destroy_workqueue(iwcm_wq);
1204 iwpm_exit(RDMA_NL_IWCM);
1207 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_IWCM, 2);
1209 module_init(iw_cm_init);
1210 module_exit(iw_cm_cleanup);