1/*
2 * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
3 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
4 * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
5 * Copyright (c) 2005-2006 Intel Corporation.  All rights reserved.
6 *
7 * This software is available to you under a choice of one of two
8 * licenses.  You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
12 *
13 *     Redistribution and use in source and binary forms, with or
14 *     without modification, are permitted provided that the following
15 *     conditions are met:
16 *
17 *      - Redistributions of source code must retain the above
18 *        copyright notice, this list of conditions and the following
19 *        disclaimer.
20 *
21 *      - Redistributions in binary form must reproduce the above
22 *        copyright notice, this list of conditions and the following
23 *        disclaimer in the documentation and/or other materials
24 *        provided with the distribution.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 * SOFTWARE.
34 */
35
36#include <linux/completion.h>
37#include <linux/in.h>
38#include <linux/in6.h>
39#include <linux/mutex.h>
40#include <linux/random.h>
41#include <linux/idr.h>
42#include <linux/inetdevice.h>
43#include <linux/slab.h>
44#include <linux/module.h>
45#include <net/route.h>
46
47#include <net/tcp.h>
48#include <net/ipv6.h>
49
50#include <rdma/rdma_cm.h>
51#include <rdma/rdma_cm_ib.h>
52#include <rdma/rdma_netlink.h>
53#include <rdma/ib.h>
54#include <rdma/ib_cache.h>
55#include <rdma/ib_cm.h>
56#include <rdma/ib_sa.h>
57#include <rdma/iw_cm.h>
58
59MODULE_AUTHOR("Sean Hefty");
60MODULE_DESCRIPTION("Generic RDMA CM Agent");
61MODULE_LICENSE("Dual BSD/GPL");
62
63#define CMA_CM_RESPONSE_TIMEOUT 20
64#define CMA_MAX_CM_RETRIES 15
65#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
66#define CMA_IBOE_PACKET_LIFETIME 18
67
68static void cma_add_one(struct ib_device *device);
69static void cma_remove_one(struct ib_device *device);
70
71static struct ib_client cma_client = {
72	.name   = "cma",
73	.add    = cma_add_one,
74	.remove = cma_remove_one
75};
76
77static struct ib_sa_client sa_client;
78static struct rdma_addr_client addr_client;
79static LIST_HEAD(dev_list);
80static LIST_HEAD(listen_any_list);
81static DEFINE_MUTEX(lock);
82static struct workqueue_struct *cma_wq;
83static DEFINE_IDR(tcp_ps);
84static DEFINE_IDR(udp_ps);
85static DEFINE_IDR(ipoib_ps);
86static DEFINE_IDR(ib_ps);
87
88struct cma_device {
89	struct list_head	list;
90	struct ib_device	*device;
91	struct completion	comp;
92	atomic_t		refcount;
93	struct list_head	id_list;
94};
95
96struct rdma_bind_list {
97	struct idr		*ps;
98	struct hlist_head	owners;
99	unsigned short		port;
100};
101
102enum {
103	CMA_OPTION_AFONLY,
104};
105
106/*
107 * Device removal can occur at anytime, so we need extra handling to
108 * serialize notifying the user of device removal with other callbacks.
109 * We do this by disabling removal notification while a callback is in process,
110 * and reporting it after the callback completes.
111 */
112struct rdma_id_private {
113	struct rdma_cm_id	id;
114
115	struct rdma_bind_list	*bind_list;
116	struct hlist_node	node;
117	struct list_head	list; /* listen_any_list or cma_device.list */
118	struct list_head	listen_list; /* per device listens */
119	struct cma_device	*cma_dev;
120	struct list_head	mc_list;
121
122	int			internal_id;
123	enum rdma_cm_state	state;
124	spinlock_t		lock;
125	struct mutex		qp_mutex;
126
127	struct completion	comp;
128	atomic_t		refcount;
129	struct mutex		handler_mutex;
130
131	int			backlog;
132	int			timeout_ms;
133	struct ib_sa_query	*query;
134	int			query_id;
135	union {
136		struct ib_cm_id	*ib;
137		struct iw_cm_id	*iw;
138	} cm_id;
139
140	u32			seq_num;
141	u32			qkey;
142	u32			qp_num;
143	pid_t			owner;
144	u32			options;
145	u8			srq;
146	u8			tos;
147	u8			reuseaddr;
148	u8			afonly;
149};
150
151struct cma_multicast {
152	struct rdma_id_private *id_priv;
153	union {
154		struct ib_sa_multicast *ib;
155	} multicast;
156	struct list_head	list;
157	void			*context;
158	struct sockaddr_storage	addr;
159	struct kref		mcref;
160};
161
162struct cma_work {
163	struct work_struct	work;
164	struct rdma_id_private	*id;
165	enum rdma_cm_state	old_state;
166	enum rdma_cm_state	new_state;
167	struct rdma_cm_event	event;
168};
169
170struct cma_ndev_work {
171	struct work_struct	work;
172	struct rdma_id_private	*id;
173	struct rdma_cm_event	event;
174};
175
176struct iboe_mcast_work {
177	struct work_struct	 work;
178	struct rdma_id_private	*id;
179	struct cma_multicast	*mc;
180};
181
182union cma_ip_addr {
183	struct in6_addr ip6;
184	struct {
185		__be32 pad[3];
186		__be32 addr;
187	} ip4;
188};
189
190struct cma_hdr {
191	u8 cma_version;
192	u8 ip_version;	/* IP version: 7:4 */
193	__be16 port;
194	union cma_ip_addr src_addr;
195	union cma_ip_addr dst_addr;
196};
197
198#define CMA_VERSION 0x00
199
200static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
201{
202	unsigned long flags;
203	int ret;
204
205	spin_lock_irqsave(&id_priv->lock, flags);
206	ret = (id_priv->state == comp);
207	spin_unlock_irqrestore(&id_priv->lock, flags);
208	return ret;
209}
210
211static int cma_comp_exch(struct rdma_id_private *id_priv,
212			 enum rdma_cm_state comp, enum rdma_cm_state exch)
213{
214	unsigned long flags;
215	int ret;
216
217	spin_lock_irqsave(&id_priv->lock, flags);
218	if ((ret = (id_priv->state == comp)))
219		id_priv->state = exch;
220	spin_unlock_irqrestore(&id_priv->lock, flags);
221	return ret;
222}
223
224static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
225				   enum rdma_cm_state exch)
226{
227	unsigned long flags;
228	enum rdma_cm_state old;
229
230	spin_lock_irqsave(&id_priv->lock, flags);
231	old = id_priv->state;
232	id_priv->state = exch;
233	spin_unlock_irqrestore(&id_priv->lock, flags);
234	return old;
235}
236
237static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
238{
239	return hdr->ip_version >> 4;
240}
241
242static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
243{
244	hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
245}
246
247static void cma_attach_to_dev(struct rdma_id_private *id_priv,
248			      struct cma_device *cma_dev)
249{
250	atomic_inc(&cma_dev->refcount);
251	id_priv->cma_dev = cma_dev;
252	id_priv->id.device = cma_dev->device;
253	id_priv->id.route.addr.dev_addr.transport =
254		rdma_node_get_transport(cma_dev->device->node_type);
255	list_add_tail(&id_priv->list, &cma_dev->id_list);
256}
257
258static inline void cma_deref_dev(struct cma_device *cma_dev)
259{
260	if (atomic_dec_and_test(&cma_dev->refcount))
261		complete(&cma_dev->comp);
262}
263
264static inline void release_mc(struct kref *kref)
265{
266	struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
267
268	kfree(mc->multicast.ib);
269	kfree(mc);
270}
271
272static void cma_release_dev(struct rdma_id_private *id_priv)
273{
274	mutex_lock(&lock);
275	list_del(&id_priv->list);
276	cma_deref_dev(id_priv->cma_dev);
277	id_priv->cma_dev = NULL;
278	mutex_unlock(&lock);
279}
280
281static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
282{
283	return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
284}
285
286static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
287{
288	return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
289}
290
291static inline unsigned short cma_family(struct rdma_id_private *id_priv)
292{
293	return id_priv->id.route.addr.src_addr.ss_family;
294}
295
296static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
297{
298	struct ib_sa_mcmember_rec rec;
299	int ret = 0;
300
301	if (id_priv->qkey) {
302		if (qkey && id_priv->qkey != qkey)
303			return -EINVAL;
304		return 0;
305	}
306
307	if (qkey) {
308		id_priv->qkey = qkey;
309		return 0;
310	}
311
312	switch (id_priv->id.ps) {
313	case RDMA_PS_UDP:
314	case RDMA_PS_IB:
315		id_priv->qkey = RDMA_UDP_QKEY;
316		break;
317	case RDMA_PS_IPOIB:
318		ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
319		ret = ib_sa_get_mcmember_rec(id_priv->id.device,
320					     id_priv->id.port_num, &rec.mgid,
321					     &rec);
322		if (!ret)
323			id_priv->qkey = be32_to_cpu(rec.qkey);
324		break;
325	default:
326		break;
327	}
328	return ret;
329}
330
331static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
332{
333	dev_addr->dev_type = ARPHRD_INFINIBAND;
334	rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
335	ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
336}
337
338static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
339{
340	int ret;
341
342	if (addr->sa_family != AF_IB) {
343		ret = rdma_translate_ip(addr, dev_addr, NULL);
344	} else {
345		cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
346		ret = 0;
347	}
348
349	return ret;
350}
351
352static int cma_acquire_dev(struct rdma_id_private *id_priv,
353			   struct rdma_id_private *listen_id_priv)
354{
355	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
356	struct cma_device *cma_dev;
357	union ib_gid gid, iboe_gid;
358	int ret = -ENODEV;
359	u8 port, found_port;
360	enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
361		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
362
363	if (dev_ll != IB_LINK_LAYER_INFINIBAND &&
364	    id_priv->id.ps == RDMA_PS_IPOIB)
365		return -EINVAL;
366
367	mutex_lock(&lock);
368	rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
369		    &iboe_gid);
370
371	memcpy(&gid, dev_addr->src_dev_addr +
372	       rdma_addr_gid_offset(dev_addr), sizeof gid);
373	if (listen_id_priv &&
374	    rdma_port_get_link_layer(listen_id_priv->id.device,
375				     listen_id_priv->id.port_num) == dev_ll) {
376		cma_dev = listen_id_priv->cma_dev;
377		port = listen_id_priv->id.port_num;
378		if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
379		    rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
380			ret = ib_find_cached_gid(cma_dev->device, &iboe_gid,
381						 &found_port, NULL);
382		else
383			ret = ib_find_cached_gid(cma_dev->device, &gid,
384						 &found_port, NULL);
385
386		if (!ret && (port  == found_port)) {
387			id_priv->id.port_num = found_port;
388			goto out;
389		}
390	}
391	list_for_each_entry(cma_dev, &dev_list, list) {
392		for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
393			if (listen_id_priv &&
394			    listen_id_priv->cma_dev == cma_dev &&
395			    listen_id_priv->id.port_num == port)
396				continue;
397			if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
398				if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
399				    rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
400					ret = ib_find_cached_gid(cma_dev->device, &iboe_gid, &found_port, NULL);
401				else
402					ret = ib_find_cached_gid(cma_dev->device, &gid, &found_port, NULL);
403
404				if (!ret && (port == found_port)) {
405					id_priv->id.port_num = found_port;
406					goto out;
407				}
408			}
409		}
410	}
411
412out:
413	if (!ret)
414		cma_attach_to_dev(id_priv, cma_dev);
415
416	mutex_unlock(&lock);
417	return ret;
418}
419
420/*
421 * Select the source IB device and address to reach the destination IB address.
422 */
423static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
424{
425	struct cma_device *cma_dev, *cur_dev;
426	struct sockaddr_ib *addr;
427	union ib_gid gid, sgid, *dgid;
428	u16 pkey, index;
429	u8 p;
430	int i;
431
432	cma_dev = NULL;
433	addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
434	dgid = (union ib_gid *) &addr->sib_addr;
435	pkey = ntohs(addr->sib_pkey);
436
437	list_for_each_entry(cur_dev, &dev_list, list) {
438		if (rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
439			continue;
440
441		for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
442			if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
443				continue;
444
445			for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
446				if (!memcmp(&gid, dgid, sizeof(gid))) {
447					cma_dev = cur_dev;
448					sgid = gid;
449					id_priv->id.port_num = p;
450					goto found;
451				}
452
453				if (!cma_dev && (gid.global.subnet_prefix ==
454						 dgid->global.subnet_prefix)) {
455					cma_dev = cur_dev;
456					sgid = gid;
457					id_priv->id.port_num = p;
458				}
459			}
460		}
461	}
462
463	if (!cma_dev)
464		return -ENODEV;
465
466found:
467	cma_attach_to_dev(id_priv, cma_dev);
468	addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
469	memcpy(&addr->sib_addr, &sgid, sizeof sgid);
470	cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
471	return 0;
472}
473
474static void cma_deref_id(struct rdma_id_private *id_priv)
475{
476	if (atomic_dec_and_test(&id_priv->refcount))
477		complete(&id_priv->comp);
478}
479
480static int cma_disable_callback(struct rdma_id_private *id_priv,
481				enum rdma_cm_state state)
482{
483	mutex_lock(&id_priv->handler_mutex);
484	if (id_priv->state != state) {
485		mutex_unlock(&id_priv->handler_mutex);
486		return -EINVAL;
487	}
488	return 0;
489}
490
491struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
492				  void *context, enum rdma_port_space ps,
493				  enum ib_qp_type qp_type)
494{
495	struct rdma_id_private *id_priv;
496
497	id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
498	if (!id_priv)
499		return ERR_PTR(-ENOMEM);
500
501	id_priv->owner = task_pid_nr(current);
502	id_priv->state = RDMA_CM_IDLE;
503	id_priv->id.context = context;
504	id_priv->id.event_handler = event_handler;
505	id_priv->id.ps = ps;
506	id_priv->id.qp_type = qp_type;
507	spin_lock_init(&id_priv->lock);
508	mutex_init(&id_priv->qp_mutex);
509	init_completion(&id_priv->comp);
510	atomic_set(&id_priv->refcount, 1);
511	mutex_init(&id_priv->handler_mutex);
512	INIT_LIST_HEAD(&id_priv->listen_list);
513	INIT_LIST_HEAD(&id_priv->mc_list);
514	get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
515
516	return &id_priv->id;
517}
518EXPORT_SYMBOL(rdma_create_id);
519
520static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
521{
522	struct ib_qp_attr qp_attr;
523	int qp_attr_mask, ret;
524
525	qp_attr.qp_state = IB_QPS_INIT;
526	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
527	if (ret)
528		return ret;
529
530	ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
531	if (ret)
532		return ret;
533
534	qp_attr.qp_state = IB_QPS_RTR;
535	ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
536	if (ret)
537		return ret;
538
539	qp_attr.qp_state = IB_QPS_RTS;
540	qp_attr.sq_psn = 0;
541	ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
542
543	return ret;
544}
545
546static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
547{
548	struct ib_qp_attr qp_attr;
549	int qp_attr_mask, ret;
550
551	qp_attr.qp_state = IB_QPS_INIT;
552	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
553	if (ret)
554		return ret;
555
556	return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
557}
558
559int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
560		   struct ib_qp_init_attr *qp_init_attr)
561{
562	struct rdma_id_private *id_priv;
563	struct ib_qp *qp;
564	int ret;
565
566	id_priv = container_of(id, struct rdma_id_private, id);
567	if (id->device != pd->device)
568		return -EINVAL;
569
570	qp = ib_create_qp(pd, qp_init_attr);
571	if (IS_ERR(qp))
572		return PTR_ERR(qp);
573
574	if (id->qp_type == IB_QPT_UD)
575		ret = cma_init_ud_qp(id_priv, qp);
576	else
577		ret = cma_init_conn_qp(id_priv, qp);
578	if (ret)
579		goto err;
580
581	id->qp = qp;
582	id_priv->qp_num = qp->qp_num;
583	id_priv->srq = (qp->srq != NULL);
584	return 0;
585err:
586	ib_destroy_qp(qp);
587	return ret;
588}
589EXPORT_SYMBOL(rdma_create_qp);
590
591void rdma_destroy_qp(struct rdma_cm_id *id)
592{
593	struct rdma_id_private *id_priv;
594
595	id_priv = container_of(id, struct rdma_id_private, id);
596	mutex_lock(&id_priv->qp_mutex);
597	ib_destroy_qp(id_priv->id.qp);
598	id_priv->id.qp = NULL;
599	mutex_unlock(&id_priv->qp_mutex);
600}
601EXPORT_SYMBOL(rdma_destroy_qp);
602
603static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
604			     struct rdma_conn_param *conn_param)
605{
606	struct ib_qp_attr qp_attr;
607	int qp_attr_mask, ret;
608	union ib_gid sgid;
609
610	mutex_lock(&id_priv->qp_mutex);
611	if (!id_priv->id.qp) {
612		ret = 0;
613		goto out;
614	}
615
616	/* Need to update QP attributes from default values. */
617	qp_attr.qp_state = IB_QPS_INIT;
618	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
619	if (ret)
620		goto out;
621
622	ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
623	if (ret)
624		goto out;
625
626	qp_attr.qp_state = IB_QPS_RTR;
627	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
628	if (ret)
629		goto out;
630
631	ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
632			   qp_attr.ah_attr.grh.sgid_index, &sgid);
633	if (ret)
634		goto out;
635
636	if (rdma_node_get_transport(id_priv->cma_dev->device->node_type)
637	    == RDMA_TRANSPORT_IB &&
638	    rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)
639	    == IB_LINK_LAYER_ETHERNET) {
640		ret = rdma_addr_find_smac_by_sgid(&sgid, qp_attr.smac, NULL);
641
642		if (ret)
643			goto out;
644	}
645	if (conn_param)
646		qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
647	ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
648out:
649	mutex_unlock(&id_priv->qp_mutex);
650	return ret;
651}
652
653static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
654			     struct rdma_conn_param *conn_param)
655{
656	struct ib_qp_attr qp_attr;
657	int qp_attr_mask, ret;
658
659	mutex_lock(&id_priv->qp_mutex);
660	if (!id_priv->id.qp) {
661		ret = 0;
662		goto out;
663	}
664
665	qp_attr.qp_state = IB_QPS_RTS;
666	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
667	if (ret)
668		goto out;
669
670	if (conn_param)
671		qp_attr.max_rd_atomic = conn_param->initiator_depth;
672	ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
673out:
674	mutex_unlock(&id_priv->qp_mutex);
675	return ret;
676}
677
678static int cma_modify_qp_err(struct rdma_id_private *id_priv)
679{
680	struct ib_qp_attr qp_attr;
681	int ret;
682
683	mutex_lock(&id_priv->qp_mutex);
684	if (!id_priv->id.qp) {
685		ret = 0;
686		goto out;
687	}
688
689	qp_attr.qp_state = IB_QPS_ERR;
690	ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
691out:
692	mutex_unlock(&id_priv->qp_mutex);
693	return ret;
694}
695
696static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
697			       struct ib_qp_attr *qp_attr, int *qp_attr_mask)
698{
699	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
700	int ret;
701	u16 pkey;
702
703	if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
704	    IB_LINK_LAYER_INFINIBAND)
705		pkey = ib_addr_get_pkey(dev_addr);
706	else
707		pkey = 0xffff;
708
709	ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
710				  pkey, &qp_attr->pkey_index);
711	if (ret)
712		return ret;
713
714	qp_attr->port_num = id_priv->id.port_num;
715	*qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
716
717	if (id_priv->id.qp_type == IB_QPT_UD) {
718		ret = cma_set_qkey(id_priv, 0);
719		if (ret)
720			return ret;
721
722		qp_attr->qkey = id_priv->qkey;
723		*qp_attr_mask |= IB_QP_QKEY;
724	} else {
725		qp_attr->qp_access_flags = 0;
726		*qp_attr_mask |= IB_QP_ACCESS_FLAGS;
727	}
728	return 0;
729}
730
731int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
732		       int *qp_attr_mask)
733{
734	struct rdma_id_private *id_priv;
735	int ret = 0;
736
737	id_priv = container_of(id, struct rdma_id_private, id);
738	switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
739	case RDMA_TRANSPORT_IB:
740		if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
741			ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
742		else
743			ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
744						 qp_attr_mask);
745
746		if (qp_attr->qp_state == IB_QPS_RTR)
747			qp_attr->rq_psn = id_priv->seq_num;
748		break;
749	case RDMA_TRANSPORT_IWARP:
750		if (!id_priv->cm_id.iw) {
751			qp_attr->qp_access_flags = 0;
752			*qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
753		} else
754			ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
755						 qp_attr_mask);
756		break;
757	default:
758		ret = -ENOSYS;
759		break;
760	}
761
762	return ret;
763}
764EXPORT_SYMBOL(rdma_init_qp_attr);
765
766static inline int cma_zero_addr(struct sockaddr *addr)
767{
768	switch (addr->sa_family) {
769	case AF_INET:
770		return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
771	case AF_INET6:
772		return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
773	case AF_IB:
774		return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
775	default:
776		return 0;
777	}
778}
779
780static inline int cma_loopback_addr(struct sockaddr *addr)
781{
782	switch (addr->sa_family) {
783	case AF_INET:
784		return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
785	case AF_INET6:
786		return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
787	case AF_IB:
788		return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
789	default:
790		return 0;
791	}
792}
793
794static inline int cma_any_addr(struct sockaddr *addr)
795{
796	return cma_zero_addr(addr) || cma_loopback_addr(addr);
797}
798
799static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
800{
801	if (src->sa_family != dst->sa_family)
802		return -1;
803
804	switch (src->sa_family) {
805	case AF_INET:
806		return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
807		       ((struct sockaddr_in *) dst)->sin_addr.s_addr;
808	case AF_INET6:
809		return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
810				     &((struct sockaddr_in6 *) dst)->sin6_addr);
811	default:
812		return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
813				   &((struct sockaddr_ib *) dst)->sib_addr);
814	}
815}
816
817static __be16 cma_port(struct sockaddr *addr)
818{
819	struct sockaddr_ib *sib;
820
821	switch (addr->sa_family) {
822	case AF_INET:
823		return ((struct sockaddr_in *) addr)->sin_port;
824	case AF_INET6:
825		return ((struct sockaddr_in6 *) addr)->sin6_port;
826	case AF_IB:
827		sib = (struct sockaddr_ib *) addr;
828		return htons((u16) (be64_to_cpu(sib->sib_sid) &
829				    be64_to_cpu(sib->sib_sid_mask)));
830	default:
831		return 0;
832	}
833}
834
835static inline int cma_any_port(struct sockaddr *addr)
836{
837	return !cma_port(addr);
838}
839
840static void cma_save_ib_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
841			     struct ib_sa_path_rec *path)
842{
843	struct sockaddr_ib *listen_ib, *ib;
844
845	listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
846	ib = (struct sockaddr_ib *) &id->route.addr.src_addr;
847	ib->sib_family = listen_ib->sib_family;
848	if (path) {
849		ib->sib_pkey = path->pkey;
850		ib->sib_flowinfo = path->flow_label;
851		memcpy(&ib->sib_addr, &path->sgid, 16);
852	} else {
853		ib->sib_pkey = listen_ib->sib_pkey;
854		ib->sib_flowinfo = listen_ib->sib_flowinfo;
855		ib->sib_addr = listen_ib->sib_addr;
856	}
857	ib->sib_sid = listen_ib->sib_sid;
858	ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
859	ib->sib_scope_id = listen_ib->sib_scope_id;
860
861	if (path) {
862		ib = (struct sockaddr_ib *) &id->route.addr.dst_addr;
863		ib->sib_family = listen_ib->sib_family;
864		ib->sib_pkey = path->pkey;
865		ib->sib_flowinfo = path->flow_label;
866		memcpy(&ib->sib_addr, &path->dgid, 16);
867	}
868}
869
870static __be16 ss_get_port(const struct sockaddr_storage *ss)
871{
872	if (ss->ss_family == AF_INET)
873		return ((struct sockaddr_in *)ss)->sin_port;
874	else if (ss->ss_family == AF_INET6)
875		return ((struct sockaddr_in6 *)ss)->sin6_port;
876	BUG();
877}
878
879static void cma_save_ip4_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
880			      struct cma_hdr *hdr)
881{
882	struct sockaddr_in *ip4;
883
884	ip4 = (struct sockaddr_in *) &id->route.addr.src_addr;
885	ip4->sin_family = AF_INET;
886	ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
887	ip4->sin_port = ss_get_port(&listen_id->route.addr.src_addr);
888
889	ip4 = (struct sockaddr_in *) &id->route.addr.dst_addr;
890	ip4->sin_family = AF_INET;
891	ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
892	ip4->sin_port = hdr->port;
893}
894
895static void cma_save_ip6_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
896			      struct cma_hdr *hdr)
897{
898	struct sockaddr_in6 *ip6;
899
900	ip6 = (struct sockaddr_in6 *) &id->route.addr.src_addr;
901	ip6->sin6_family = AF_INET6;
902	ip6->sin6_addr = hdr->dst_addr.ip6;
903	ip6->sin6_port = ss_get_port(&listen_id->route.addr.src_addr);
904
905	ip6 = (struct sockaddr_in6 *) &id->route.addr.dst_addr;
906	ip6->sin6_family = AF_INET6;
907	ip6->sin6_addr = hdr->src_addr.ip6;
908	ip6->sin6_port = hdr->port;
909}
910
911static int cma_save_net_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
912			     struct ib_cm_event *ib_event)
913{
914	struct cma_hdr *hdr;
915
916	if (listen_id->route.addr.src_addr.ss_family == AF_IB) {
917		if (ib_event->event == IB_CM_REQ_RECEIVED)
918			cma_save_ib_info(id, listen_id, ib_event->param.req_rcvd.primary_path);
919		else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
920			cma_save_ib_info(id, listen_id, NULL);
921		return 0;
922	}
923
924	hdr = ib_event->private_data;
925	if (hdr->cma_version != CMA_VERSION)
926		return -EINVAL;
927
928	switch (cma_get_ip_ver(hdr)) {
929	case 4:
930		cma_save_ip4_info(id, listen_id, hdr);
931		break;
932	case 6:
933		cma_save_ip6_info(id, listen_id, hdr);
934		break;
935	default:
936		return -EINVAL;
937	}
938	return 0;
939}
940
941static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
942{
943	return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
944}
945
946static void cma_cancel_route(struct rdma_id_private *id_priv)
947{
948	switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
949	case IB_LINK_LAYER_INFINIBAND:
950		if (id_priv->query)
951			ib_sa_cancel_query(id_priv->query_id, id_priv->query);
952		break;
953	default:
954		break;
955	}
956}
957
958static void cma_cancel_listens(struct rdma_id_private *id_priv)
959{
960	struct rdma_id_private *dev_id_priv;
961
962	/*
963	 * Remove from listen_any_list to prevent added devices from spawning
964	 * additional listen requests.
965	 */
966	mutex_lock(&lock);
967	list_del(&id_priv->list);
968
969	while (!list_empty(&id_priv->listen_list)) {
970		dev_id_priv = list_entry(id_priv->listen_list.next,
971					 struct rdma_id_private, listen_list);
972		/* sync with device removal to avoid duplicate destruction */
973		list_del_init(&dev_id_priv->list);
974		list_del(&dev_id_priv->listen_list);
975		mutex_unlock(&lock);
976
977		rdma_destroy_id(&dev_id_priv->id);
978		mutex_lock(&lock);
979	}
980	mutex_unlock(&lock);
981}
982
983static void cma_cancel_operation(struct rdma_id_private *id_priv,
984				 enum rdma_cm_state state)
985{
986	switch (state) {
987	case RDMA_CM_ADDR_QUERY:
988		rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
989		break;
990	case RDMA_CM_ROUTE_QUERY:
991		cma_cancel_route(id_priv);
992		break;
993	case RDMA_CM_LISTEN:
994		if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
995			cma_cancel_listens(id_priv);
996		break;
997	default:
998		break;
999	}
1000}
1001
1002static void cma_release_port(struct rdma_id_private *id_priv)
1003{
1004	struct rdma_bind_list *bind_list = id_priv->bind_list;
1005
1006	if (!bind_list)
1007		return;
1008
1009	mutex_lock(&lock);
1010	hlist_del(&id_priv->node);
1011	if (hlist_empty(&bind_list->owners)) {
1012		idr_remove(bind_list->ps, bind_list->port);
1013		kfree(bind_list);
1014	}
1015	mutex_unlock(&lock);
1016}
1017
1018static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1019{
1020	struct cma_multicast *mc;
1021
1022	while (!list_empty(&id_priv->mc_list)) {
1023		mc = container_of(id_priv->mc_list.next,
1024				  struct cma_multicast, list);
1025		list_del(&mc->list);
1026		switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
1027		case IB_LINK_LAYER_INFINIBAND:
1028			ib_sa_free_multicast(mc->multicast.ib);
1029			kfree(mc);
1030			break;
1031		case IB_LINK_LAYER_ETHERNET:
1032			kref_put(&mc->mcref, release_mc);
1033			break;
1034		default:
1035			break;
1036		}
1037	}
1038}
1039
1040void rdma_destroy_id(struct rdma_cm_id *id)
1041{
1042	struct rdma_id_private *id_priv;
1043	enum rdma_cm_state state;
1044
1045	id_priv = container_of(id, struct rdma_id_private, id);
1046	state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1047	cma_cancel_operation(id_priv, state);
1048
1049	/*
1050	 * Wait for any active callback to finish.  New callbacks will find
1051	 * the id_priv state set to destroying and abort.
1052	 */
1053	mutex_lock(&id_priv->handler_mutex);
1054	mutex_unlock(&id_priv->handler_mutex);
1055
1056	if (id_priv->cma_dev) {
1057		switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
1058		case RDMA_TRANSPORT_IB:
1059			if (id_priv->cm_id.ib)
1060				ib_destroy_cm_id(id_priv->cm_id.ib);
1061			break;
1062		case RDMA_TRANSPORT_IWARP:
1063			if (id_priv->cm_id.iw)
1064				iw_destroy_cm_id(id_priv->cm_id.iw);
1065			break;
1066		default:
1067			break;
1068		}
1069		cma_leave_mc_groups(id_priv);
1070		cma_release_dev(id_priv);
1071	}
1072
1073	cma_release_port(id_priv);
1074	cma_deref_id(id_priv);
1075	wait_for_completion(&id_priv->comp);
1076
1077	if (id_priv->internal_id)
1078		cma_deref_id(id_priv->id.context);
1079
1080	kfree(id_priv->id.route.path_rec);
1081	kfree(id_priv);
1082}
1083EXPORT_SYMBOL(rdma_destroy_id);
1084
1085static int cma_rep_recv(struct rdma_id_private *id_priv)
1086{
1087	int ret;
1088
1089	ret = cma_modify_qp_rtr(id_priv, NULL);
1090	if (ret)
1091		goto reject;
1092
1093	ret = cma_modify_qp_rts(id_priv, NULL);
1094	if (ret)
1095		goto reject;
1096
1097	ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1098	if (ret)
1099		goto reject;
1100
1101	return 0;
1102reject:
1103	cma_modify_qp_err(id_priv);
1104	ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1105		       NULL, 0, NULL, 0);
1106	return ret;
1107}
1108
1109static void cma_set_rep_event_data(struct rdma_cm_event *event,
1110				   struct ib_cm_rep_event_param *rep_data,
1111				   void *private_data)
1112{
1113	event->param.conn.private_data = private_data;
1114	event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1115	event->param.conn.responder_resources = rep_data->responder_resources;
1116	event->param.conn.initiator_depth = rep_data->initiator_depth;
1117	event->param.conn.flow_control = rep_data->flow_control;
1118	event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1119	event->param.conn.srq = rep_data->srq;
1120	event->param.conn.qp_num = rep_data->remote_qpn;
1121}
1122
1123static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1124{
1125	struct rdma_id_private *id_priv = cm_id->context;
1126	struct rdma_cm_event event;
1127	int ret = 0;
1128
1129	if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1130		cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1131	    (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1132		cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1133		return 0;
1134
1135	memset(&event, 0, sizeof event);
1136	switch (ib_event->event) {
1137	case IB_CM_REQ_ERROR:
1138	case IB_CM_REP_ERROR:
1139		event.event = RDMA_CM_EVENT_UNREACHABLE;
1140		event.status = -ETIMEDOUT;
1141		break;
1142	case IB_CM_REP_RECEIVED:
1143		if (id_priv->id.qp) {
1144			event.status = cma_rep_recv(id_priv);
1145			event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1146						     RDMA_CM_EVENT_ESTABLISHED;
1147		} else {
1148			event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1149		}
1150		cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1151				       ib_event->private_data);
1152		break;
1153	case IB_CM_RTU_RECEIVED:
1154	case IB_CM_USER_ESTABLISHED:
1155		event.event = RDMA_CM_EVENT_ESTABLISHED;
1156		break;
1157	case IB_CM_DREQ_ERROR:
1158		event.status = -ETIMEDOUT; /* fall through */
1159	case IB_CM_DREQ_RECEIVED:
1160	case IB_CM_DREP_RECEIVED:
1161		if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1162				   RDMA_CM_DISCONNECT))
1163			goto out;
1164		event.event = RDMA_CM_EVENT_DISCONNECTED;
1165		break;
1166	case IB_CM_TIMEWAIT_EXIT:
1167		event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1168		break;
1169	case IB_CM_MRA_RECEIVED:
1170		/* ignore event */
1171		goto out;
1172	case IB_CM_REJ_RECEIVED:
1173		cma_modify_qp_err(id_priv);
1174		event.status = ib_event->param.rej_rcvd.reason;
1175		event.event = RDMA_CM_EVENT_REJECTED;
1176		event.param.conn.private_data = ib_event->private_data;
1177		event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1178		break;
1179	default:
1180		printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1181		       ib_event->event);
1182		goto out;
1183	}
1184
1185	ret = id_priv->id.event_handler(&id_priv->id, &event);
1186	if (ret) {
1187		/* Destroy the CM ID by returning a non-zero value. */
1188		id_priv->cm_id.ib = NULL;
1189		cma_exch(id_priv, RDMA_CM_DESTROYING);
1190		mutex_unlock(&id_priv->handler_mutex);
1191		rdma_destroy_id(&id_priv->id);
1192		return ret;
1193	}
1194out:
1195	mutex_unlock(&id_priv->handler_mutex);
1196	return ret;
1197}
1198
1199static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1200					       struct ib_cm_event *ib_event)
1201{
1202	struct rdma_id_private *id_priv;
1203	struct rdma_cm_id *id;
1204	struct rdma_route *rt;
1205	int ret;
1206
1207	id = rdma_create_id(listen_id->event_handler, listen_id->context,
1208			    listen_id->ps, ib_event->param.req_rcvd.qp_type);
1209	if (IS_ERR(id))
1210		return NULL;
1211
1212	id_priv = container_of(id, struct rdma_id_private, id);
1213	if (cma_save_net_info(id, listen_id, ib_event))
1214		goto err;
1215
1216	rt = &id->route;
1217	rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1218	rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1219			       GFP_KERNEL);
1220	if (!rt->path_rec)
1221		goto err;
1222
1223	rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1224	if (rt->num_paths == 2)
1225		rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1226
1227	if (cma_any_addr(cma_src_addr(id_priv))) {
1228		rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1229		rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1230		ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1231	} else {
1232		ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1233		if (ret)
1234			goto err;
1235	}
1236	rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1237
1238	id_priv->state = RDMA_CM_CONNECT;
1239	return id_priv;
1240
1241err:
1242	rdma_destroy_id(id);
1243	return NULL;
1244}
1245
1246static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1247					      struct ib_cm_event *ib_event)
1248{
1249	struct rdma_id_private *id_priv;
1250	struct rdma_cm_id *id;
1251	int ret;
1252
1253	id = rdma_create_id(listen_id->event_handler, listen_id->context,
1254			    listen_id->ps, IB_QPT_UD);
1255	if (IS_ERR(id))
1256		return NULL;
1257
1258	id_priv = container_of(id, struct rdma_id_private, id);
1259	if (cma_save_net_info(id, listen_id, ib_event))
1260		goto err;
1261
1262	if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1263		ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
1264		if (ret)
1265			goto err;
1266	}
1267
1268	id_priv->state = RDMA_CM_CONNECT;
1269	return id_priv;
1270err:
1271	rdma_destroy_id(id);
1272	return NULL;
1273}
1274
1275static void cma_set_req_event_data(struct rdma_cm_event *event,
1276				   struct ib_cm_req_event_param *req_data,
1277				   void *private_data, int offset)
1278{
1279	event->param.conn.private_data = private_data + offset;
1280	event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1281	event->param.conn.responder_resources = req_data->responder_resources;
1282	event->param.conn.initiator_depth = req_data->initiator_depth;
1283	event->param.conn.flow_control = req_data->flow_control;
1284	event->param.conn.retry_count = req_data->retry_count;
1285	event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1286	event->param.conn.srq = req_data->srq;
1287	event->param.conn.qp_num = req_data->remote_qpn;
1288}
1289
1290static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1291{
1292	return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1293		 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1294		((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1295		 (id->qp_type == IB_QPT_UD)) ||
1296		(!id->qp_type));
1297}
1298
1299static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1300{
1301	struct rdma_id_private *listen_id, *conn_id;
1302	struct rdma_cm_event event;
1303	int offset, ret;
1304
1305	listen_id = cm_id->context;
1306	if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1307		return -EINVAL;
1308
1309	if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1310		return -ECONNABORTED;
1311
1312	memset(&event, 0, sizeof event);
1313	offset = cma_user_data_offset(listen_id);
1314	event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1315	if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1316		conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1317		event.param.ud.private_data = ib_event->private_data + offset;
1318		event.param.ud.private_data_len =
1319				IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1320	} else {
1321		conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1322		cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1323				       ib_event->private_data, offset);
1324	}
1325	if (!conn_id) {
1326		ret = -ENOMEM;
1327		goto err1;
1328	}
1329
1330	mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1331	ret = cma_acquire_dev(conn_id, listen_id);
1332	if (ret)
1333		goto err2;
1334
1335	conn_id->cm_id.ib = cm_id;
1336	cm_id->context = conn_id;
1337	cm_id->cm_handler = cma_ib_handler;
1338
1339	/*
1340	 * Protect against the user destroying conn_id from another thread
1341	 * until we're done accessing it.
1342	 */
1343	atomic_inc(&conn_id->refcount);
1344	ret = conn_id->id.event_handler(&conn_id->id, &event);
1345	if (ret)
1346		goto err3;
1347	/*
1348	 * Acquire mutex to prevent user executing rdma_destroy_id()
1349	 * while we're accessing the cm_id.
1350	 */
1351	mutex_lock(&lock);
1352	if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1353	    (conn_id->id.qp_type != IB_QPT_UD))
1354		ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1355	mutex_unlock(&lock);
1356	mutex_unlock(&conn_id->handler_mutex);
1357	mutex_unlock(&listen_id->handler_mutex);
1358	cma_deref_id(conn_id);
1359	return 0;
1360
1361err3:
1362	cma_deref_id(conn_id);
1363	/* Destroy the CM ID by returning a non-zero value. */
1364	conn_id->cm_id.ib = NULL;
1365err2:
1366	cma_exch(conn_id, RDMA_CM_DESTROYING);
1367	mutex_unlock(&conn_id->handler_mutex);
1368err1:
1369	mutex_unlock(&listen_id->handler_mutex);
1370	if (conn_id)
1371		rdma_destroy_id(&conn_id->id);
1372	return ret;
1373}
1374
1375__be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1376{
1377	if (addr->sa_family == AF_IB)
1378		return ((struct sockaddr_ib *) addr)->sib_sid;
1379
1380	return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1381}
1382EXPORT_SYMBOL(rdma_get_service_id);
1383
1384static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1385				 struct ib_cm_compare_data *compare)
1386{
1387	struct cma_hdr *cma_data, *cma_mask;
1388	__be32 ip4_addr;
1389	struct in6_addr ip6_addr;
1390
1391	memset(compare, 0, sizeof *compare);
1392	cma_data = (void *) compare->data;
1393	cma_mask = (void *) compare->mask;
1394
1395	switch (addr->sa_family) {
1396	case AF_INET:
1397		ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1398		cma_set_ip_ver(cma_data, 4);
1399		cma_set_ip_ver(cma_mask, 0xF);
1400		if (!cma_any_addr(addr)) {
1401			cma_data->dst_addr.ip4.addr = ip4_addr;
1402			cma_mask->dst_addr.ip4.addr = htonl(~0);
1403		}
1404		break;
1405	case AF_INET6:
1406		ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1407		cma_set_ip_ver(cma_data, 6);
1408		cma_set_ip_ver(cma_mask, 0xF);
1409		if (!cma_any_addr(addr)) {
1410			cma_data->dst_addr.ip6 = ip6_addr;
1411			memset(&cma_mask->dst_addr.ip6, 0xFF,
1412			       sizeof cma_mask->dst_addr.ip6);
1413		}
1414		break;
1415	default:
1416		break;
1417	}
1418}
1419
1420static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1421{
1422	struct rdma_id_private *id_priv = iw_id->context;
1423	struct rdma_cm_event event;
1424	int ret = 0;
1425	struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1426	struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1427
1428	if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1429		return 0;
1430
1431	memset(&event, 0, sizeof event);
1432	switch (iw_event->event) {
1433	case IW_CM_EVENT_CLOSE:
1434		event.event = RDMA_CM_EVENT_DISCONNECTED;
1435		break;
1436	case IW_CM_EVENT_CONNECT_REPLY:
1437		memcpy(cma_src_addr(id_priv), laddr,
1438		       rdma_addr_size(laddr));
1439		memcpy(cma_dst_addr(id_priv), raddr,
1440		       rdma_addr_size(raddr));
1441		switch (iw_event->status) {
1442		case 0:
1443			event.event = RDMA_CM_EVENT_ESTABLISHED;
1444			event.param.conn.initiator_depth = iw_event->ird;
1445			event.param.conn.responder_resources = iw_event->ord;
1446			break;
1447		case -ECONNRESET:
1448		case -ECONNREFUSED:
1449			event.event = RDMA_CM_EVENT_REJECTED;
1450			break;
1451		case -ETIMEDOUT:
1452			event.event = RDMA_CM_EVENT_UNREACHABLE;
1453			break;
1454		default:
1455			event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1456			break;
1457		}
1458		break;
1459	case IW_CM_EVENT_ESTABLISHED:
1460		event.event = RDMA_CM_EVENT_ESTABLISHED;
1461		event.param.conn.initiator_depth = iw_event->ird;
1462		event.param.conn.responder_resources = iw_event->ord;
1463		break;
1464	default:
1465		BUG_ON(1);
1466	}
1467
1468	event.status = iw_event->status;
1469	event.param.conn.private_data = iw_event->private_data;
1470	event.param.conn.private_data_len = iw_event->private_data_len;
1471	ret = id_priv->id.event_handler(&id_priv->id, &event);
1472	if (ret) {
1473		/* Destroy the CM ID by returning a non-zero value. */
1474		id_priv->cm_id.iw = NULL;
1475		cma_exch(id_priv, RDMA_CM_DESTROYING);
1476		mutex_unlock(&id_priv->handler_mutex);
1477		rdma_destroy_id(&id_priv->id);
1478		return ret;
1479	}
1480
1481	mutex_unlock(&id_priv->handler_mutex);
1482	return ret;
1483}
1484
1485static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1486			       struct iw_cm_event *iw_event)
1487{
1488	struct rdma_cm_id *new_cm_id;
1489	struct rdma_id_private *listen_id, *conn_id;
1490	struct rdma_cm_event event;
1491	int ret;
1492	struct ib_device_attr attr;
1493	struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1494	struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1495
1496	listen_id = cm_id->context;
1497	if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1498		return -ECONNABORTED;
1499
1500	/* Create a new RDMA id for the new IW CM ID */
1501	new_cm_id = rdma_create_id(listen_id->id.event_handler,
1502				   listen_id->id.context,
1503				   RDMA_PS_TCP, IB_QPT_RC);
1504	if (IS_ERR(new_cm_id)) {
1505		ret = -ENOMEM;
1506		goto out;
1507	}
1508	conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1509	mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1510	conn_id->state = RDMA_CM_CONNECT;
1511
1512	ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1513	if (ret) {
1514		mutex_unlock(&conn_id->handler_mutex);
1515		rdma_destroy_id(new_cm_id);
1516		goto out;
1517	}
1518
1519	ret = cma_acquire_dev(conn_id, listen_id);
1520	if (ret) {
1521		mutex_unlock(&conn_id->handler_mutex);
1522		rdma_destroy_id(new_cm_id);
1523		goto out;
1524	}
1525
1526	conn_id->cm_id.iw = cm_id;
1527	cm_id->context = conn_id;
1528	cm_id->cm_handler = cma_iw_handler;
1529
1530	memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1531	memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1532
1533	ret = ib_query_device(conn_id->id.device, &attr);
1534	if (ret) {
1535		mutex_unlock(&conn_id->handler_mutex);
1536		rdma_destroy_id(new_cm_id);
1537		goto out;
1538	}
1539
1540	memset(&event, 0, sizeof event);
1541	event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1542	event.param.conn.private_data = iw_event->private_data;
1543	event.param.conn.private_data_len = iw_event->private_data_len;
1544	event.param.conn.initiator_depth = iw_event->ird;
1545	event.param.conn.responder_resources = iw_event->ord;
1546
1547	/*
1548	 * Protect against the user destroying conn_id from another thread
1549	 * until we're done accessing it.
1550	 */
1551	atomic_inc(&conn_id->refcount);
1552	ret = conn_id->id.event_handler(&conn_id->id, &event);
1553	if (ret) {
1554		/* User wants to destroy the CM ID */
1555		conn_id->cm_id.iw = NULL;
1556		cma_exch(conn_id, RDMA_CM_DESTROYING);
1557		mutex_unlock(&conn_id->handler_mutex);
1558		cma_deref_id(conn_id);
1559		rdma_destroy_id(&conn_id->id);
1560		goto out;
1561	}
1562
1563	mutex_unlock(&conn_id->handler_mutex);
1564	cma_deref_id(conn_id);
1565
1566out:
1567	mutex_unlock(&listen_id->handler_mutex);
1568	return ret;
1569}
1570
1571static int cma_ib_listen(struct rdma_id_private *id_priv)
1572{
1573	struct ib_cm_compare_data compare_data;
1574	struct sockaddr *addr;
1575	struct ib_cm_id	*id;
1576	__be64 svc_id;
1577	int ret;
1578
1579	id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1580	if (IS_ERR(id))
1581		return PTR_ERR(id);
1582
1583	id_priv->cm_id.ib = id;
1584
1585	addr = cma_src_addr(id_priv);
1586	svc_id = rdma_get_service_id(&id_priv->id, addr);
1587	if (cma_any_addr(addr) && !id_priv->afonly)
1588		ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1589	else {
1590		cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1591		ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1592	}
1593
1594	if (ret) {
1595		ib_destroy_cm_id(id_priv->cm_id.ib);
1596		id_priv->cm_id.ib = NULL;
1597	}
1598
1599	return ret;
1600}
1601
1602static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1603{
1604	int ret;
1605	struct iw_cm_id	*id;
1606
1607	id = iw_create_cm_id(id_priv->id.device,
1608			     iw_conn_req_handler,
1609			     id_priv);
1610	if (IS_ERR(id))
1611		return PTR_ERR(id);
1612
1613	id_priv->cm_id.iw = id;
1614
1615	memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
1616	       rdma_addr_size(cma_src_addr(id_priv)));
1617
1618	ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1619
1620	if (ret) {
1621		iw_destroy_cm_id(id_priv->cm_id.iw);
1622		id_priv->cm_id.iw = NULL;
1623	}
1624
1625	return ret;
1626}
1627
1628static int cma_listen_handler(struct rdma_cm_id *id,
1629			      struct rdma_cm_event *event)
1630{
1631	struct rdma_id_private *id_priv = id->context;
1632
1633	id->context = id_priv->id.context;
1634	id->event_handler = id_priv->id.event_handler;
1635	return id_priv->id.event_handler(id, event);
1636}
1637
1638static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1639			      struct cma_device *cma_dev)
1640{
1641	struct rdma_id_private *dev_id_priv;
1642	struct rdma_cm_id *id;
1643	int ret;
1644
1645	if (cma_family(id_priv) == AF_IB &&
1646	    rdma_node_get_transport(cma_dev->device->node_type) != RDMA_TRANSPORT_IB)
1647		return;
1648
1649	id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1650			    id_priv->id.qp_type);
1651	if (IS_ERR(id))
1652		return;
1653
1654	dev_id_priv = container_of(id, struct rdma_id_private, id);
1655
1656	dev_id_priv->state = RDMA_CM_ADDR_BOUND;
1657	memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1658	       rdma_addr_size(cma_src_addr(id_priv)));
1659
1660	cma_attach_to_dev(dev_id_priv, cma_dev);
1661	list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1662	atomic_inc(&id_priv->refcount);
1663	dev_id_priv->internal_id = 1;
1664	dev_id_priv->afonly = id_priv->afonly;
1665
1666	ret = rdma_listen(id, id_priv->backlog);
1667	if (ret)
1668		printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1669		       "listening on device %s\n", ret, cma_dev->device->name);
1670}
1671
1672static void cma_listen_on_all(struct rdma_id_private *id_priv)
1673{
1674	struct cma_device *cma_dev;
1675
1676	mutex_lock(&lock);
1677	list_add_tail(&id_priv->list, &listen_any_list);
1678	list_for_each_entry(cma_dev, &dev_list, list)
1679		cma_listen_on_dev(id_priv, cma_dev);
1680	mutex_unlock(&lock);
1681}
1682
1683void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1684{
1685	struct rdma_id_private *id_priv;
1686
1687	id_priv = container_of(id, struct rdma_id_private, id);
1688	id_priv->tos = (u8) tos;
1689}
1690EXPORT_SYMBOL(rdma_set_service_type);
1691
1692static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1693			      void *context)
1694{
1695	struct cma_work *work = context;
1696	struct rdma_route *route;
1697
1698	route = &work->id->id.route;
1699
1700	if (!status) {
1701		route->num_paths = 1;
1702		*route->path_rec = *path_rec;
1703	} else {
1704		work->old_state = RDMA_CM_ROUTE_QUERY;
1705		work->new_state = RDMA_CM_ADDR_RESOLVED;
1706		work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1707		work->event.status = status;
1708	}
1709
1710	queue_work(cma_wq, &work->work);
1711}
1712
1713static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1714			      struct cma_work *work)
1715{
1716	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1717	struct ib_sa_path_rec path_rec;
1718	ib_sa_comp_mask comp_mask;
1719	struct sockaddr_in6 *sin6;
1720	struct sockaddr_ib *sib;
1721
1722	memset(&path_rec, 0, sizeof path_rec);
1723	rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1724	rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1725	path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
1726	path_rec.numb_path = 1;
1727	path_rec.reversible = 1;
1728	path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
1729
1730	comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1731		    IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1732		    IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1733
1734	switch (cma_family(id_priv)) {
1735	case AF_INET:
1736		path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1737		comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1738		break;
1739	case AF_INET6:
1740		sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
1741		path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1742		comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1743		break;
1744	case AF_IB:
1745		sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
1746		path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
1747		comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1748		break;
1749	}
1750
1751	id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1752					       id_priv->id.port_num, &path_rec,
1753					       comp_mask, timeout_ms,
1754					       GFP_KERNEL, cma_query_handler,
1755					       work, &id_priv->query);
1756
1757	return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1758}
1759
1760static void cma_work_handler(struct work_struct *_work)
1761{
1762	struct cma_work *work = container_of(_work, struct cma_work, work);
1763	struct rdma_id_private *id_priv = work->id;
1764	int destroy = 0;
1765
1766	mutex_lock(&id_priv->handler_mutex);
1767	if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1768		goto out;
1769
1770	if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1771		cma_exch(id_priv, RDMA_CM_DESTROYING);
1772		destroy = 1;
1773	}
1774out:
1775	mutex_unlock(&id_priv->handler_mutex);
1776	cma_deref_id(id_priv);
1777	if (destroy)
1778		rdma_destroy_id(&id_priv->id);
1779	kfree(work);
1780}
1781
1782static void cma_ndev_work_handler(struct work_struct *_work)
1783{
1784	struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1785	struct rdma_id_private *id_priv = work->id;
1786	int destroy = 0;
1787
1788	mutex_lock(&id_priv->handler_mutex);
1789	if (id_priv->state == RDMA_CM_DESTROYING ||
1790	    id_priv->state == RDMA_CM_DEVICE_REMOVAL)
1791		goto out;
1792
1793	if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1794		cma_exch(id_priv, RDMA_CM_DESTROYING);
1795		destroy = 1;
1796	}
1797
1798out:
1799	mutex_unlock(&id_priv->handler_mutex);
1800	cma_deref_id(id_priv);
1801	if (destroy)
1802		rdma_destroy_id(&id_priv->id);
1803	kfree(work);
1804}
1805
1806static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1807{
1808	struct rdma_route *route = &id_priv->id.route;
1809	struct cma_work *work;
1810	int ret;
1811
1812	work = kzalloc(sizeof *work, GFP_KERNEL);
1813	if (!work)
1814		return -ENOMEM;
1815
1816	work->id = id_priv;
1817	INIT_WORK(&work->work, cma_work_handler);
1818	work->old_state = RDMA_CM_ROUTE_QUERY;
1819	work->new_state = RDMA_CM_ROUTE_RESOLVED;
1820	work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1821
1822	route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1823	if (!route->path_rec) {
1824		ret = -ENOMEM;
1825		goto err1;
1826	}
1827
1828	ret = cma_query_ib_route(id_priv, timeout_ms, work);
1829	if (ret)
1830		goto err2;
1831
1832	return 0;
1833err2:
1834	kfree(route->path_rec);
1835	route->path_rec = NULL;
1836err1:
1837	kfree(work);
1838	return ret;
1839}
1840
1841int rdma_set_ib_paths(struct rdma_cm_id *id,
1842		      struct ib_sa_path_rec *path_rec, int num_paths)
1843{
1844	struct rdma_id_private *id_priv;
1845	int ret;
1846
1847	id_priv = container_of(id, struct rdma_id_private, id);
1848	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1849			   RDMA_CM_ROUTE_RESOLVED))
1850		return -EINVAL;
1851
1852	id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1853				     GFP_KERNEL);
1854	if (!id->route.path_rec) {
1855		ret = -ENOMEM;
1856		goto err;
1857	}
1858
1859	id->route.num_paths = num_paths;
1860	return 0;
1861err:
1862	cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
1863	return ret;
1864}
1865EXPORT_SYMBOL(rdma_set_ib_paths);
1866
1867static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1868{
1869	struct cma_work *work;
1870
1871	work = kzalloc(sizeof *work, GFP_KERNEL);
1872	if (!work)
1873		return -ENOMEM;
1874
1875	work->id = id_priv;
1876	INIT_WORK(&work->work, cma_work_handler);
1877	work->old_state = RDMA_CM_ROUTE_QUERY;
1878	work->new_state = RDMA_CM_ROUTE_RESOLVED;
1879	work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1880	queue_work(cma_wq, &work->work);
1881	return 0;
1882}
1883
1884static int iboe_tos_to_sl(struct net_device *ndev, int tos)
1885{
1886	int prio;
1887	struct net_device *dev;
1888
1889	prio = rt_tos2priority(tos);
1890	dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
1891		vlan_dev_real_dev(ndev) : ndev;
1892
1893	if (dev->num_tc)
1894		return netdev_get_prio_tc_map(dev, prio);
1895
1896#if IS_ENABLED(CONFIG_VLAN_8021Q)
1897	if (ndev->priv_flags & IFF_802_1Q_VLAN)
1898		return (vlan_dev_get_egress_qos_mask(ndev, prio) &
1899			VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1900#endif
1901	return 0;
1902}
1903
1904static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1905{
1906	struct rdma_route *route = &id_priv->id.route;
1907	struct rdma_addr *addr = &route->addr;
1908	struct cma_work *work;
1909	int ret;
1910	struct net_device *ndev = NULL;
1911
1912
1913	work = kzalloc(sizeof *work, GFP_KERNEL);
1914	if (!work)
1915		return -ENOMEM;
1916
1917	work->id = id_priv;
1918	INIT_WORK(&work->work, cma_work_handler);
1919
1920	route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1921	if (!route->path_rec) {
1922		ret = -ENOMEM;
1923		goto err1;
1924	}
1925
1926	route->num_paths = 1;
1927
1928	if (addr->dev_addr.bound_dev_if)
1929		ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1930	if (!ndev) {
1931		ret = -ENODEV;
1932		goto err2;
1933	}
1934
1935	route->path_rec->vlan_id = rdma_vlan_dev_vlan_id(ndev);
1936	memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
1937	memcpy(route->path_rec->smac, ndev->dev_addr, ndev->addr_len);
1938
1939	rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
1940		    &route->path_rec->sgid);
1941	rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
1942		    &route->path_rec->dgid);
1943
1944	route->path_rec->hop_limit = 1;
1945	route->path_rec->reversible = 1;
1946	route->path_rec->pkey = cpu_to_be16(0xffff);
1947	route->path_rec->mtu_selector = IB_SA_EQ;
1948	route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
1949	route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1950	route->path_rec->rate_selector = IB_SA_EQ;
1951	route->path_rec->rate = iboe_get_rate(ndev);
1952	dev_put(ndev);
1953	route->path_rec->packet_life_time_selector = IB_SA_EQ;
1954	route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1955	if (!route->path_rec->mtu) {
1956		ret = -EINVAL;
1957		goto err2;
1958	}
1959
1960	work->old_state = RDMA_CM_ROUTE_QUERY;
1961	work->new_state = RDMA_CM_ROUTE_RESOLVED;
1962	work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1963	work->event.status = 0;
1964
1965	queue_work(cma_wq, &work->work);
1966
1967	return 0;
1968
1969err2:
1970	kfree(route->path_rec);
1971	route->path_rec = NULL;
1972err1:
1973	kfree(work);
1974	return ret;
1975}
1976
1977int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1978{
1979	struct rdma_id_private *id_priv;
1980	int ret;
1981
1982	id_priv = container_of(id, struct rdma_id_private, id);
1983	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
1984		return -EINVAL;
1985
1986	atomic_inc(&id_priv->refcount);
1987	switch (rdma_node_get_transport(id->device->node_type)) {
1988	case RDMA_TRANSPORT_IB:
1989		switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1990		case IB_LINK_LAYER_INFINIBAND:
1991			ret = cma_resolve_ib_route(id_priv, timeout_ms);
1992			break;
1993		case IB_LINK_LAYER_ETHERNET:
1994			ret = cma_resolve_iboe_route(id_priv);
1995			break;
1996		default:
1997			ret = -ENOSYS;
1998		}
1999		break;
2000	case RDMA_TRANSPORT_IWARP:
2001		ret = cma_resolve_iw_route(id_priv, timeout_ms);
2002		break;
2003	default:
2004		ret = -ENOSYS;
2005		break;
2006	}
2007	if (ret)
2008		goto err;
2009
2010	return 0;
2011err:
2012	cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2013	cma_deref_id(id_priv);
2014	return ret;
2015}
2016EXPORT_SYMBOL(rdma_resolve_route);
2017
2018static void cma_set_loopback(struct sockaddr *addr)
2019{
2020	switch (addr->sa_family) {
2021	case AF_INET:
2022		((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2023		break;
2024	case AF_INET6:
2025		ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2026			      0, 0, 0, htonl(1));
2027		break;
2028	default:
2029		ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2030			    0, 0, 0, htonl(1));
2031		break;
2032	}
2033}
2034
2035static int cma_bind_loopback(struct rdma_id_private *id_priv)
2036{
2037	struct cma_device *cma_dev, *cur_dev;
2038	struct ib_port_attr port_attr;
2039	union ib_gid gid;
2040	u16 pkey;
2041	int ret;
2042	u8 p;
2043
2044	cma_dev = NULL;
2045	mutex_lock(&lock);
2046	list_for_each_entry(cur_dev, &dev_list, list) {
2047		if (cma_family(id_priv) == AF_IB &&
2048		    rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
2049			continue;
2050
2051		if (!cma_dev)
2052			cma_dev = cur_dev;
2053
2054		for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2055			if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2056			    port_attr.state == IB_PORT_ACTIVE) {
2057				cma_dev = cur_dev;
2058				goto port_found;
2059			}
2060		}
2061	}
2062
2063	if (!cma_dev) {
2064		ret = -ENODEV;
2065		goto out;
2066	}
2067
2068	p = 1;
2069
2070port_found:
2071	ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2072	if (ret)
2073		goto out;
2074
2075	ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2076	if (ret)
2077		goto out;
2078
2079	id_priv->id.route.addr.dev_addr.dev_type =
2080		(rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
2081		ARPHRD_INFINIBAND : ARPHRD_ETHER;
2082
2083	rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2084	ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2085	id_priv->id.port_num = p;
2086	cma_attach_to_dev(id_priv, cma_dev);
2087	cma_set_loopback(cma_src_addr(id_priv));
2088out:
2089	mutex_unlock(&lock);
2090	return ret;
2091}
2092
2093static void addr_handler(int status, struct sockaddr *src_addr,
2094			 struct rdma_dev_addr *dev_addr, void *context)
2095{
2096	struct rdma_id_private *id_priv = context;
2097	struct rdma_cm_event event;
2098
2099	memset(&event, 0, sizeof event);
2100	mutex_lock(&id_priv->handler_mutex);
2101	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2102			   RDMA_CM_ADDR_RESOLVED))
2103		goto out;
2104
2105	memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2106	if (!status && !id_priv->cma_dev)
2107		status = cma_acquire_dev(id_priv, NULL);
2108
2109	if (status) {
2110		if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2111				   RDMA_CM_ADDR_BOUND))
2112			goto out;
2113		event.event = RDMA_CM_EVENT_ADDR_ERROR;
2114		event.status = status;
2115	} else
2116		event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2117
2118	if (id_priv->id.event_handler(&id_priv->id, &event)) {
2119		cma_exch(id_priv, RDMA_CM_DESTROYING);
2120		mutex_unlock(&id_priv->handler_mutex);
2121		cma_deref_id(id_priv);
2122		rdma_destroy_id(&id_priv->id);
2123		return;
2124	}
2125out:
2126	mutex_unlock(&id_priv->handler_mutex);
2127	cma_deref_id(id_priv);
2128}
2129
2130static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2131{
2132	struct cma_work *work;
2133	union ib_gid gid;
2134	int ret;
2135
2136	work = kzalloc(sizeof *work, GFP_KERNEL);
2137	if (!work)
2138		return -ENOMEM;
2139
2140	if (!id_priv->cma_dev) {
2141		ret = cma_bind_loopback(id_priv);
2142		if (ret)
2143			goto err;
2144	}
2145
2146	rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2147	rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2148
2149	work->id = id_priv;
2150	INIT_WORK(&work->work, cma_work_handler);
2151	work->old_state = RDMA_CM_ADDR_QUERY;
2152	work->new_state = RDMA_CM_ADDR_RESOLVED;
2153	work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2154	queue_work(cma_wq, &work->work);
2155	return 0;
2156err:
2157	kfree(work);
2158	return ret;
2159}
2160
2161static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2162{
2163	struct cma_work *work;
2164	int ret;
2165
2166	work = kzalloc(sizeof *work, GFP_KERNEL);
2167	if (!work)
2168		return -ENOMEM;
2169
2170	if (!id_priv->cma_dev) {
2171		ret = cma_resolve_ib_dev(id_priv);
2172		if (ret)
2173			goto err;
2174	}
2175
2176	rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2177		&(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2178
2179	work->id = id_priv;
2180	INIT_WORK(&work->work, cma_work_handler);
2181	work->old_state = RDMA_CM_ADDR_QUERY;
2182	work->new_state = RDMA_CM_ADDR_RESOLVED;
2183	work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2184	queue_work(cma_wq, &work->work);
2185	return 0;
2186err:
2187	kfree(work);
2188	return ret;
2189}
2190
2191static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2192			 struct sockaddr *dst_addr)
2193{
2194	if (!src_addr || !src_addr->sa_family) {
2195		src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2196		src_addr->sa_family = dst_addr->sa_family;
2197		if (dst_addr->sa_family == AF_INET6) {
2198			((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2199				((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2200		} else if (dst_addr->sa_family == AF_IB) {
2201			((struct sockaddr_ib *) src_addr)->sib_pkey =
2202				((struct sockaddr_ib *) dst_addr)->sib_pkey;
2203		}
2204	}
2205	return rdma_bind_addr(id, src_addr);
2206}
2207
2208int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2209		      struct sockaddr *dst_addr, int timeout_ms)
2210{
2211	struct rdma_id_private *id_priv;
2212	int ret;
2213
2214	id_priv = container_of(id, struct rdma_id_private, id);
2215	if (id_priv->state == RDMA_CM_IDLE) {
2216		ret = cma_bind_addr(id, src_addr, dst_addr);
2217		if (ret)
2218			return ret;
2219	}
2220
2221	if (cma_family(id_priv) != dst_addr->sa_family)
2222		return -EINVAL;
2223
2224	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2225		return -EINVAL;
2226
2227	atomic_inc(&id_priv->refcount);
2228	memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2229	if (cma_any_addr(dst_addr)) {
2230		ret = cma_resolve_loopback(id_priv);
2231	} else {
2232		if (dst_addr->sa_family == AF_IB) {
2233			ret = cma_resolve_ib_addr(id_priv);
2234		} else {
2235			ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2236					      dst_addr, &id->route.addr.dev_addr,
2237					      timeout_ms, addr_handler, id_priv);
2238		}
2239	}
2240	if (ret)
2241		goto err;
2242
2243	return 0;
2244err:
2245	cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2246	cma_deref_id(id_priv);
2247	return ret;
2248}
2249EXPORT_SYMBOL(rdma_resolve_addr);
2250
2251int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2252{
2253	struct rdma_id_private *id_priv;
2254	unsigned long flags;
2255	int ret;
2256
2257	id_priv = container_of(id, struct rdma_id_private, id);
2258	spin_lock_irqsave(&id_priv->lock, flags);
2259	if (reuse || id_priv->state == RDMA_CM_IDLE) {
2260		id_priv->reuseaddr = reuse;
2261		ret = 0;
2262	} else {
2263		ret = -EINVAL;
2264	}
2265	spin_unlock_irqrestore(&id_priv->lock, flags);
2266	return ret;
2267}
2268EXPORT_SYMBOL(rdma_set_reuseaddr);
2269
2270int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2271{
2272	struct rdma_id_private *id_priv;
2273	unsigned long flags;
2274	int ret;
2275
2276	id_priv = container_of(id, struct rdma_id_private, id);
2277	spin_lock_irqsave(&id_priv->lock, flags);
2278	if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2279		id_priv->options |= (1 << CMA_OPTION_AFONLY);
2280		id_priv->afonly = afonly;
2281		ret = 0;
2282	} else {
2283		ret = -EINVAL;
2284	}
2285	spin_unlock_irqrestore(&id_priv->lock, flags);
2286	return ret;
2287}
2288EXPORT_SYMBOL(rdma_set_afonly);
2289
2290static void cma_bind_port(struct rdma_bind_list *bind_list,
2291			  struct rdma_id_private *id_priv)
2292{
2293	struct sockaddr *addr;
2294	struct sockaddr_ib *sib;
2295	u64 sid, mask;
2296	__be16 port;
2297
2298	addr = cma_src_addr(id_priv);
2299	port = htons(bind_list->port);
2300
2301	switch (addr->sa_family) {
2302	case AF_INET:
2303		((struct sockaddr_in *) addr)->sin_port = port;
2304		break;
2305	case AF_INET6:
2306		((struct sockaddr_in6 *) addr)->sin6_port = port;
2307		break;
2308	case AF_IB:
2309		sib = (struct sockaddr_ib *) addr;
2310		sid = be64_to_cpu(sib->sib_sid);
2311		mask = be64_to_cpu(sib->sib_sid_mask);
2312		sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2313		sib->sib_sid_mask = cpu_to_be64(~0ULL);
2314		break;
2315	}
2316	id_priv->bind_list = bind_list;
2317	hlist_add_head(&id_priv->node, &bind_list->owners);
2318}
2319
2320static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2321			  unsigned short snum)
2322{
2323	struct rdma_bind_list *bind_list;
2324	int ret;
2325
2326	bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2327	if (!bind_list)
2328		return -ENOMEM;
2329
2330	ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2331	if (ret < 0)
2332		goto err;
2333
2334	bind_list->ps = ps;
2335	bind_list->port = (unsigned short)ret;
2336	cma_bind_port(bind_list, id_priv);
2337	return 0;
2338err:
2339	kfree(bind_list);
2340	return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2341}
2342
2343static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2344{
2345	static unsigned int last_used_port;
2346	int low, high, remaining;
2347	unsigned int rover;
2348
2349	inet_get_local_port_range(&init_net, &low, &high);
2350	remaining = (high - low) + 1;
2351	rover = prandom_u32() % remaining + low;
2352retry:
2353	if (last_used_port != rover &&
2354	    !idr_find(ps, (unsigned short) rover)) {
2355		int ret = cma_alloc_port(ps, id_priv, rover);
2356		/*
2357		 * Remember previously used port number in order to avoid
2358		 * re-using same port immediately after it is closed.
2359		 */
2360		if (!ret)
2361			last_used_port = rover;
2362		if (ret != -EADDRNOTAVAIL)
2363			return ret;
2364	}
2365	if (--remaining) {
2366		rover++;
2367		if ((rover < low) || (rover > high))
2368			rover = low;
2369		goto retry;
2370	}
2371	return -EADDRNOTAVAIL;
2372}
2373
2374/*
2375 * Check that the requested port is available.  This is called when trying to
2376 * bind to a specific port, or when trying to listen on a bound port.  In
2377 * the latter case, the provided id_priv may already be on the bind_list, but
2378 * we still need to check that it's okay to start listening.
2379 */
2380static int cma_check_port(struct rdma_bind_list *bind_list,
2381			  struct rdma_id_private *id_priv, uint8_t reuseaddr)
2382{
2383	struct rdma_id_private *cur_id;
2384	struct sockaddr *addr, *cur_addr;
2385
2386	addr = cma_src_addr(id_priv);
2387	hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2388		if (id_priv == cur_id)
2389			continue;
2390
2391		if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2392		    cur_id->reuseaddr)
2393			continue;
2394
2395		cur_addr = cma_src_addr(cur_id);
2396		if (id_priv->afonly && cur_id->afonly &&
2397		    (addr->sa_family != cur_addr->sa_family))
2398			continue;
2399
2400		if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2401			return -EADDRNOTAVAIL;
2402
2403		if (!cma_addr_cmp(addr, cur_addr))
2404			return -EADDRINUSE;
2405	}
2406	return 0;
2407}
2408
2409static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2410{
2411	struct rdma_bind_list *bind_list;
2412	unsigned short snum;
2413	int ret;
2414
2415	snum = ntohs(cma_port(cma_src_addr(id_priv)));
2416	if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2417		return -EACCES;
2418
2419	bind_list = idr_find(ps, snum);
2420	if (!bind_list) {
2421		ret = cma_alloc_port(ps, id_priv, snum);
2422	} else {
2423		ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2424		if (!ret)
2425			cma_bind_port(bind_list, id_priv);
2426	}
2427	return ret;
2428}
2429
2430static int cma_bind_listen(struct rdma_id_private *id_priv)
2431{
2432	struct rdma_bind_list *bind_list = id_priv->bind_list;
2433	int ret = 0;
2434
2435	mutex_lock(&lock);
2436	if (bind_list->owners.first->next)
2437		ret = cma_check_port(bind_list, id_priv, 0);
2438	mutex_unlock(&lock);
2439	return ret;
2440}
2441
2442static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
2443{
2444	switch (id_priv->id.ps) {
2445	case RDMA_PS_TCP:
2446		return &tcp_ps;
2447	case RDMA_PS_UDP:
2448		return &udp_ps;
2449	case RDMA_PS_IPOIB:
2450		return &ipoib_ps;
2451	case RDMA_PS_IB:
2452		return &ib_ps;
2453	default:
2454		return NULL;
2455	}
2456}
2457
2458static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2459{
2460	struct idr *ps = NULL;
2461	struct sockaddr_ib *sib;
2462	u64 sid_ps, mask, sid;
2463
2464	sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2465	mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2466	sid = be64_to_cpu(sib->sib_sid) & mask;
2467
2468	if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2469		sid_ps = RDMA_IB_IP_PS_IB;
2470		ps = &ib_ps;
2471	} else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2472		   (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2473		sid_ps = RDMA_IB_IP_PS_TCP;
2474		ps = &tcp_ps;
2475	} else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2476		   (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2477		sid_ps = RDMA_IB_IP_PS_UDP;
2478		ps = &udp_ps;
2479	}
2480
2481	if (ps) {
2482		sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2483		sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2484						be64_to_cpu(sib->sib_sid_mask));
2485	}
2486	return ps;
2487}
2488
2489static int cma_get_port(struct rdma_id_private *id_priv)
2490{
2491	struct idr *ps;
2492	int ret;
2493
2494	if (cma_family(id_priv) != AF_IB)
2495		ps = cma_select_inet_ps(id_priv);
2496	else
2497		ps = cma_select_ib_ps(id_priv);
2498	if (!ps)
2499		return -EPROTONOSUPPORT;
2500
2501	mutex_lock(&lock);
2502	if (cma_any_port(cma_src_addr(id_priv)))
2503		ret = cma_alloc_any_port(ps, id_priv);
2504	else
2505		ret = cma_use_port(ps, id_priv);
2506	mutex_unlock(&lock);
2507
2508	return ret;
2509}
2510
2511static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2512			       struct sockaddr *addr)
2513{
2514#if IS_ENABLED(CONFIG_IPV6)
2515	struct sockaddr_in6 *sin6;
2516
2517	if (addr->sa_family != AF_INET6)
2518		return 0;
2519
2520	sin6 = (struct sockaddr_in6 *) addr;
2521
2522	if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2523		return 0;
2524
2525	if (!sin6->sin6_scope_id)
2526			return -EINVAL;
2527
2528	dev_addr->bound_dev_if = sin6->sin6_scope_id;
2529#endif
2530	return 0;
2531}
2532
2533int rdma_listen(struct rdma_cm_id *id, int backlog)
2534{
2535	struct rdma_id_private *id_priv;
2536	int ret;
2537
2538	id_priv = container_of(id, struct rdma_id_private, id);
2539	if (id_priv->state == RDMA_CM_IDLE) {
2540		id->route.addr.src_addr.ss_family = AF_INET;
2541		ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2542		if (ret)
2543			return ret;
2544	}
2545
2546	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2547		return -EINVAL;
2548
2549	if (id_priv->reuseaddr) {
2550		ret = cma_bind_listen(id_priv);
2551		if (ret)
2552			goto err;
2553	}
2554
2555	id_priv->backlog = backlog;
2556	if (id->device) {
2557		switch (rdma_node_get_transport(id->device->node_type)) {
2558		case RDMA_TRANSPORT_IB:
2559			ret = cma_ib_listen(id_priv);
2560			if (ret)
2561				goto err;
2562			break;
2563		case RDMA_TRANSPORT_IWARP:
2564			ret = cma_iw_listen(id_priv, backlog);
2565			if (ret)
2566				goto err;
2567			break;
2568		default:
2569			ret = -ENOSYS;
2570			goto err;
2571		}
2572	} else
2573		cma_listen_on_all(id_priv);
2574
2575	return 0;
2576err:
2577	id_priv->backlog = 0;
2578	cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2579	return ret;
2580}
2581EXPORT_SYMBOL(rdma_listen);
2582
2583int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2584{
2585	struct rdma_id_private *id_priv;
2586	int ret;
2587
2588	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2589	    addr->sa_family != AF_IB)
2590		return -EAFNOSUPPORT;
2591
2592	id_priv = container_of(id, struct rdma_id_private, id);
2593	if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2594		return -EINVAL;
2595
2596	ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2597	if (ret)
2598		goto err1;
2599
2600	memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2601	if (!cma_any_addr(addr)) {
2602		ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2603		if (ret)
2604			goto err1;
2605
2606		ret = cma_acquire_dev(id_priv, NULL);
2607		if (ret)
2608			goto err1;
2609	}
2610
2611	if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2612		if (addr->sa_family == AF_INET)
2613			id_priv->afonly = 1;
2614#if IS_ENABLED(CONFIG_IPV6)
2615		else if (addr->sa_family == AF_INET6)
2616			id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2617#endif
2618	}
2619	ret = cma_get_port(id_priv);
2620	if (ret)
2621		goto err2;
2622
2623	return 0;
2624err2:
2625	if (id_priv->cma_dev)
2626		cma_release_dev(id_priv);
2627err1:
2628	cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2629	return ret;
2630}
2631EXPORT_SYMBOL(rdma_bind_addr);
2632
2633static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2634{
2635	struct cma_hdr *cma_hdr;
2636
2637	cma_hdr = hdr;
2638	cma_hdr->cma_version = CMA_VERSION;
2639	if (cma_family(id_priv) == AF_INET) {
2640		struct sockaddr_in *src4, *dst4;
2641
2642		src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2643		dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2644
2645		cma_set_ip_ver(cma_hdr, 4);
2646		cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2647		cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2648		cma_hdr->port = src4->sin_port;
2649	} else if (cma_family(id_priv) == AF_INET6) {
2650		struct sockaddr_in6 *src6, *dst6;
2651
2652		src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2653		dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2654
2655		cma_set_ip_ver(cma_hdr, 6);
2656		cma_hdr->src_addr.ip6 = src6->sin6_addr;
2657		cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2658		cma_hdr->port = src6->sin6_port;
2659	}
2660	return 0;
2661}
2662
2663static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2664				struct ib_cm_event *ib_event)
2665{
2666	struct rdma_id_private *id_priv = cm_id->context;
2667	struct rdma_cm_event event;
2668	struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2669	int ret = 0;
2670
2671	if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
2672		return 0;
2673
2674	memset(&event, 0, sizeof event);
2675	switch (ib_event->event) {
2676	case IB_CM_SIDR_REQ_ERROR:
2677		event.event = RDMA_CM_EVENT_UNREACHABLE;
2678		event.status = -ETIMEDOUT;
2679		break;
2680	case IB_CM_SIDR_REP_RECEIVED:
2681		event.param.ud.private_data = ib_event->private_data;
2682		event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2683		if (rep->status != IB_SIDR_SUCCESS) {
2684			event.event = RDMA_CM_EVENT_UNREACHABLE;
2685			event.status = ib_event->param.sidr_rep_rcvd.status;
2686			break;
2687		}
2688		ret = cma_set_qkey(id_priv, rep->qkey);
2689		if (ret) {
2690			event.event = RDMA_CM_EVENT_ADDR_ERROR;
2691			event.status = ret;
2692			break;
2693		}
2694		ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2695				     id_priv->id.route.path_rec,
2696				     &event.param.ud.ah_attr);
2697		event.param.ud.qp_num = rep->qpn;
2698		event.param.ud.qkey = rep->qkey;
2699		event.event = RDMA_CM_EVENT_ESTABLISHED;
2700		event.status = 0;
2701		break;
2702	default:
2703		printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2704		       ib_event->event);
2705		goto out;
2706	}
2707
2708	ret = id_priv->id.event_handler(&id_priv->id, &event);
2709	if (ret) {
2710		/* Destroy the CM ID by returning a non-zero value. */
2711		id_priv->cm_id.ib = NULL;
2712		cma_exch(id_priv, RDMA_CM_DESTROYING);
2713		mutex_unlock(&id_priv->handler_mutex);
2714		rdma_destroy_id(&id_priv->id);
2715		return ret;
2716	}
2717out:
2718	mutex_unlock(&id_priv->handler_mutex);
2719	return ret;
2720}
2721
2722static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2723			      struct rdma_conn_param *conn_param)
2724{
2725	struct ib_cm_sidr_req_param req;
2726	struct ib_cm_id	*id;
2727	void *private_data;
2728	int offset, ret;
2729
2730	memset(&req, 0, sizeof req);
2731	offset = cma_user_data_offset(id_priv);
2732	req.private_data_len = offset + conn_param->private_data_len;
2733	if (req.private_data_len < conn_param->private_data_len)
2734		return -EINVAL;
2735
2736	if (req.private_data_len) {
2737		private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2738		if (!private_data)
2739			return -ENOMEM;
2740	} else {
2741		private_data = NULL;
2742	}
2743
2744	if (conn_param->private_data && conn_param->private_data_len)
2745		memcpy(private_data + offset, conn_param->private_data,
2746		       conn_param->private_data_len);
2747
2748	if (private_data) {
2749		ret = cma_format_hdr(private_data, id_priv);
2750		if (ret)
2751			goto out;
2752		req.private_data = private_data;
2753	}
2754
2755	id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2756			     id_priv);
2757	if (IS_ERR(id)) {
2758		ret = PTR_ERR(id);
2759		goto out;
2760	}
2761	id_priv->cm_id.ib = id;
2762
2763	req.path = id_priv->id.route.path_rec;
2764	req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2765	req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2766	req.max_cm_retries = CMA_MAX_CM_RETRIES;
2767
2768	ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2769	if (ret) {
2770		ib_destroy_cm_id(id_priv->cm_id.ib);
2771		id_priv->cm_id.ib = NULL;
2772	}
2773out:
2774	kfree(private_data);
2775	return ret;
2776}
2777
2778static int cma_connect_ib(struct rdma_id_private *id_priv,
2779			  struct rdma_conn_param *conn_param)
2780{
2781	struct ib_cm_req_param req;
2782	struct rdma_route *route;
2783	void *private_data;
2784	struct ib_cm_id	*id;
2785	int offset, ret;
2786
2787	memset(&req, 0, sizeof req);
2788	offset = cma_user_data_offset(id_priv);
2789	req.private_data_len = offset + conn_param->private_data_len;
2790	if (req.private_data_len < conn_param->private_data_len)
2791		return -EINVAL;
2792
2793	if (req.private_data_len) {
2794		private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2795		if (!private_data)
2796			return -ENOMEM;
2797	} else {
2798		private_data = NULL;
2799	}
2800
2801	if (conn_param->private_data && conn_param->private_data_len)
2802		memcpy(private_data + offset, conn_param->private_data,
2803		       conn_param->private_data_len);
2804
2805	id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2806	if (IS_ERR(id)) {
2807		ret = PTR_ERR(id);
2808		goto out;
2809	}
2810	id_priv->cm_id.ib = id;
2811
2812	route = &id_priv->id.route;
2813	if (private_data) {
2814		ret = cma_format_hdr(private_data, id_priv);
2815		if (ret)
2816			goto out;
2817		req.private_data = private_data;
2818	}
2819
2820	req.primary_path = &route->path_rec[0];
2821	if (route->num_paths == 2)
2822		req.alternate_path = &route->path_rec[1];
2823
2824	req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2825	req.qp_num = id_priv->qp_num;
2826	req.qp_type = id_priv->id.qp_type;
2827	req.starting_psn = id_priv->seq_num;
2828	req.responder_resources = conn_param->responder_resources;
2829	req.initiator_depth = conn_param->initiator_depth;
2830	req.flow_control = conn_param->flow_control;
2831	req.retry_count = min_t(u8, 7, conn_param->retry_count);
2832	req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2833	req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2834	req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2835	req.max_cm_retries = CMA_MAX_CM_RETRIES;
2836	req.srq = id_priv->srq ? 1 : 0;
2837
2838	ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2839out:
2840	if (ret && !IS_ERR(id)) {
2841		ib_destroy_cm_id(id);
2842		id_priv->cm_id.ib = NULL;
2843	}
2844
2845	kfree(private_data);
2846	return ret;
2847}
2848
2849static int cma_connect_iw(struct rdma_id_private *id_priv,
2850			  struct rdma_conn_param *conn_param)
2851{
2852	struct iw_cm_id *cm_id;
2853	int ret;
2854	struct iw_cm_conn_param iw_param;
2855
2856	cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2857	if (IS_ERR(cm_id))
2858		return PTR_ERR(cm_id);
2859
2860	id_priv->cm_id.iw = cm_id;
2861
2862	memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
2863	       rdma_addr_size(cma_src_addr(id_priv)));
2864	memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
2865	       rdma_addr_size(cma_dst_addr(id_priv)));
2866
2867	ret = cma_modify_qp_rtr(id_priv, conn_param);
2868	if (ret)
2869		goto out;
2870
2871	if (conn_param) {
2872		iw_param.ord = conn_param->initiator_depth;
2873		iw_param.ird = conn_param->responder_resources;
2874		iw_param.private_data = conn_param->private_data;
2875		iw_param.private_data_len = conn_param->private_data_len;
2876		iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2877	} else {
2878		memset(&iw_param, 0, sizeof iw_param);
2879		iw_param.qpn = id_priv->qp_num;
2880	}
2881	ret = iw_cm_connect(cm_id, &iw_param);
2882out:
2883	if (ret) {
2884		iw_destroy_cm_id(cm_id);
2885		id_priv->cm_id.iw = NULL;
2886	}
2887	return ret;
2888}
2889
2890int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2891{
2892	struct rdma_id_private *id_priv;
2893	int ret;
2894
2895	id_priv = container_of(id, struct rdma_id_private, id);
2896	if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
2897		return -EINVAL;
2898
2899	if (!id->qp) {
2900		id_priv->qp_num = conn_param->qp_num;
2901		id_priv->srq = conn_param->srq;
2902	}
2903
2904	switch (rdma_node_get_transport(id->device->node_type)) {
2905	case RDMA_TRANSPORT_IB:
2906		if (id->qp_type == IB_QPT_UD)
2907			ret = cma_resolve_ib_udp(id_priv, conn_param);
2908		else
2909			ret = cma_connect_ib(id_priv, conn_param);
2910		break;
2911	case RDMA_TRANSPORT_IWARP:
2912		ret = cma_connect_iw(id_priv, conn_param);
2913		break;
2914	default:
2915		ret = -ENOSYS;
2916		break;
2917	}
2918	if (ret)
2919		goto err;
2920
2921	return 0;
2922err:
2923	cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
2924	return ret;
2925}
2926EXPORT_SYMBOL(rdma_connect);
2927
2928static int cma_accept_ib(struct rdma_id_private *id_priv,
2929			 struct rdma_conn_param *conn_param)
2930{
2931	struct ib_cm_rep_param rep;
2932	int ret;
2933
2934	ret = cma_modify_qp_rtr(id_priv, conn_param);
2935	if (ret)
2936		goto out;
2937
2938	ret = cma_modify_qp_rts(id_priv, conn_param);
2939	if (ret)
2940		goto out;
2941
2942	memset(&rep, 0, sizeof rep);
2943	rep.qp_num = id_priv->qp_num;
2944	rep.starting_psn = id_priv->seq_num;
2945	rep.private_data = conn_param->private_data;
2946	rep.private_data_len = conn_param->private_data_len;
2947	rep.responder_resources = conn_param->responder_resources;
2948	rep.initiator_depth = conn_param->initiator_depth;
2949	rep.failover_accepted = 0;
2950	rep.flow_control = conn_param->flow_control;
2951	rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2952	rep.srq = id_priv->srq ? 1 : 0;
2953
2954	ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2955out:
2956	return ret;
2957}
2958
2959static int cma_accept_iw(struct rdma_id_private *id_priv,
2960		  struct rdma_conn_param *conn_param)
2961{
2962	struct iw_cm_conn_param iw_param;
2963	int ret;
2964
2965	ret = cma_modify_qp_rtr(id_priv, conn_param);
2966	if (ret)
2967		return ret;
2968
2969	iw_param.ord = conn_param->initiator_depth;
2970	iw_param.ird = conn_param->responder_resources;
2971	iw_param.private_data = conn_param->private_data;
2972	iw_param.private_data_len = conn_param->private_data_len;
2973	if (id_priv->id.qp) {
2974		iw_param.qpn = id_priv->qp_num;
2975	} else
2976		iw_param.qpn = conn_param->qp_num;
2977
2978	return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2979}
2980
2981static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2982			     enum ib_cm_sidr_status status, u32 qkey,
2983			     const void *private_data, int private_data_len)
2984{
2985	struct ib_cm_sidr_rep_param rep;
2986	int ret;
2987
2988	memset(&rep, 0, sizeof rep);
2989	rep.status = status;
2990	if (status == IB_SIDR_SUCCESS) {
2991		ret = cma_set_qkey(id_priv, qkey);
2992		if (ret)
2993			return ret;
2994		rep.qp_num = id_priv->qp_num;
2995		rep.qkey = id_priv->qkey;
2996	}
2997	rep.private_data = private_data;
2998	rep.private_data_len = private_data_len;
2999
3000	return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3001}
3002
3003int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3004{
3005	struct rdma_id_private *id_priv;
3006	int ret;
3007
3008	id_priv = container_of(id, struct rdma_id_private, id);
3009
3010	id_priv->owner = task_pid_nr(current);
3011
3012	if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3013		return -EINVAL;
3014
3015	if (!id->qp && conn_param) {
3016		id_priv->qp_num = conn_param->qp_num;
3017		id_priv->srq = conn_param->srq;
3018	}
3019
3020	switch (rdma_node_get_transport(id->device->node_type)) {
3021	case RDMA_TRANSPORT_IB:
3022		if (id->qp_type == IB_QPT_UD) {
3023			if (conn_param)
3024				ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3025							conn_param->qkey,
3026							conn_param->private_data,
3027							conn_param->private_data_len);
3028			else
3029				ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3030							0, NULL, 0);
3031		} else {
3032			if (conn_param)
3033				ret = cma_accept_ib(id_priv, conn_param);
3034			else
3035				ret = cma_rep_recv(id_priv);
3036		}
3037		break;
3038	case RDMA_TRANSPORT_IWARP:
3039		ret = cma_accept_iw(id_priv, conn_param);
3040		break;
3041	default:
3042		ret = -ENOSYS;
3043		break;
3044	}
3045
3046	if (ret)
3047		goto reject;
3048
3049	return 0;
3050reject:
3051	cma_modify_qp_err(id_priv);
3052	rdma_reject(id, NULL, 0);
3053	return ret;
3054}
3055EXPORT_SYMBOL(rdma_accept);
3056
3057int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3058{
3059	struct rdma_id_private *id_priv;
3060	int ret;
3061
3062	id_priv = container_of(id, struct rdma_id_private, id);
3063	if (!id_priv->cm_id.ib)
3064		return -EINVAL;
3065
3066	switch (id->device->node_type) {
3067	case RDMA_NODE_IB_CA:
3068		ret = ib_cm_notify(id_priv->cm_id.ib, event);
3069		break;
3070	default:
3071		ret = 0;
3072		break;
3073	}
3074	return ret;
3075}
3076EXPORT_SYMBOL(rdma_notify);
3077
3078int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3079		u8 private_data_len)
3080{
3081	struct rdma_id_private *id_priv;
3082	int ret;
3083
3084	id_priv = container_of(id, struct rdma_id_private, id);
3085	if (!id_priv->cm_id.ib)
3086		return -EINVAL;
3087
3088	switch (rdma_node_get_transport(id->device->node_type)) {
3089	case RDMA_TRANSPORT_IB:
3090		if (id->qp_type == IB_QPT_UD)
3091			ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3092						private_data, private_data_len);
3093		else
3094			ret = ib_send_cm_rej(id_priv->cm_id.ib,
3095					     IB_CM_REJ_CONSUMER_DEFINED, NULL,
3096					     0, private_data, private_data_len);
3097		break;
3098	case RDMA_TRANSPORT_IWARP:
3099		ret = iw_cm_reject(id_priv->cm_id.iw,
3100				   private_data, private_data_len);
3101		break;
3102	default:
3103		ret = -ENOSYS;
3104		break;
3105	}
3106	return ret;
3107}
3108EXPORT_SYMBOL(rdma_reject);
3109
3110int rdma_disconnect(struct rdma_cm_id *id)
3111{
3112	struct rdma_id_private *id_priv;
3113	int ret;
3114
3115	id_priv = container_of(id, struct rdma_id_private, id);
3116	if (!id_priv->cm_id.ib)
3117		return -EINVAL;
3118
3119	switch (rdma_node_get_transport(id->device->node_type)) {
3120	case RDMA_TRANSPORT_IB:
3121		ret = cma_modify_qp_err(id_priv);
3122		if (ret)
3123			goto out;
3124		/* Initiate or respond to a disconnect. */
3125		if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3126			ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3127		break;
3128	case RDMA_TRANSPORT_IWARP:
3129		ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3130		break;
3131	default:
3132		ret = -EINVAL;
3133		break;
3134	}
3135out:
3136	return ret;
3137}
3138EXPORT_SYMBOL(rdma_disconnect);
3139
3140static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3141{
3142	struct rdma_id_private *id_priv;
3143	struct cma_multicast *mc = multicast->context;
3144	struct rdma_cm_event event;
3145	int ret;
3146
3147	id_priv = mc->id_priv;
3148	if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3149	    cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3150		return 0;
3151
3152	if (!status)
3153		status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3154	mutex_lock(&id_priv->qp_mutex);
3155	if (!status && id_priv->id.qp)
3156		status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3157					 be16_to_cpu(multicast->rec.mlid));
3158	mutex_unlock(&id_priv->qp_mutex);
3159
3160	memset(&event, 0, sizeof event);
3161	event.status = status;
3162	event.param.ud.private_data = mc->context;
3163	if (!status) {
3164		event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3165		ib_init_ah_from_mcmember(id_priv->id.device,
3166					 id_priv->id.port_num, &multicast->rec,
3167					 &event.param.ud.ah_attr);
3168		event.param.ud.qp_num = 0xFFFFFF;
3169		event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3170	} else
3171		event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3172
3173	ret = id_priv->id.event_handler(&id_priv->id, &event);
3174	if (ret) {
3175		cma_exch(id_priv, RDMA_CM_DESTROYING);
3176		mutex_unlock(&id_priv->handler_mutex);
3177		rdma_destroy_id(&id_priv->id);
3178		return 0;
3179	}
3180
3181	mutex_unlock(&id_priv->handler_mutex);
3182	return 0;
3183}
3184
3185static void cma_set_mgid(struct rdma_id_private *id_priv,
3186			 struct sockaddr *addr, union ib_gid *mgid)
3187{
3188	unsigned char mc_map[MAX_ADDR_LEN];
3189	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3190	struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3191	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3192
3193	if (cma_any_addr(addr)) {
3194		memset(mgid, 0, sizeof *mgid);
3195	} else if ((addr->sa_family == AF_INET6) &&
3196		   ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3197								 0xFF10A01B)) {
3198		/* IPv6 address is an SA assigned MGID. */
3199		memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3200	} else if (addr->sa_family == AF_IB) {
3201		memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3202	} else if ((addr->sa_family == AF_INET6)) {
3203		ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3204		if (id_priv->id.ps == RDMA_PS_UDP)
3205			mc_map[7] = 0x01;	/* Use RDMA CM signature */
3206		*mgid = *(union ib_gid *) (mc_map + 4);
3207	} else {
3208		ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3209		if (id_priv->id.ps == RDMA_PS_UDP)
3210			mc_map[7] = 0x01;	/* Use RDMA CM signature */
3211		*mgid = *(union ib_gid *) (mc_map + 4);
3212	}
3213}
3214
3215static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3216				 struct cma_multicast *mc)
3217{
3218	struct ib_sa_mcmember_rec rec;
3219	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3220	ib_sa_comp_mask comp_mask;
3221	int ret;
3222
3223	ib_addr_get_mgid(dev_addr, &rec.mgid);
3224	ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3225				     &rec.mgid, &rec);
3226	if (ret)
3227		return ret;
3228
3229	ret = cma_set_qkey(id_priv, 0);
3230	if (ret)
3231		return ret;
3232
3233	cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3234	rec.qkey = cpu_to_be32(id_priv->qkey);
3235	rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3236	rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3237	rec.join_state = 1;
3238
3239	comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3240		    IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3241		    IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3242		    IB_SA_MCMEMBER_REC_FLOW_LABEL |
3243		    IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3244
3245	if (id_priv->id.ps == RDMA_PS_IPOIB)
3246		comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3247			     IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3248			     IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3249			     IB_SA_MCMEMBER_REC_MTU |
3250			     IB_SA_MCMEMBER_REC_HOP_LIMIT;
3251
3252	mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3253						id_priv->id.port_num, &rec,
3254						comp_mask, GFP_KERNEL,
3255						cma_ib_mc_handler, mc);
3256	return PTR_ERR_OR_ZERO(mc->multicast.ib);
3257}
3258
3259static void iboe_mcast_work_handler(struct work_struct *work)
3260{
3261	struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3262	struct cma_multicast *mc = mw->mc;
3263	struct ib_sa_multicast *m = mc->multicast.ib;
3264
3265	mc->multicast.ib->context = mc;
3266	cma_ib_mc_handler(0, m);
3267	kref_put(&mc->mcref, release_mc);
3268	kfree(mw);
3269}
3270
3271static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3272{
3273	struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3274	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3275
3276	if (cma_any_addr(addr)) {
3277		memset(mgid, 0, sizeof *mgid);
3278	} else if (addr->sa_family == AF_INET6) {
3279		memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3280	} else {
3281		mgid->raw[0] = 0xff;
3282		mgid->raw[1] = 0x0e;
3283		mgid->raw[2] = 0;
3284		mgid->raw[3] = 0;
3285		mgid->raw[4] = 0;
3286		mgid->raw[5] = 0;
3287		mgid->raw[6] = 0;
3288		mgid->raw[7] = 0;
3289		mgid->raw[8] = 0;
3290		mgid->raw[9] = 0;
3291		mgid->raw[10] = 0xff;
3292		mgid->raw[11] = 0xff;
3293		*(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3294	}
3295}
3296
3297static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3298				   struct cma_multicast *mc)
3299{
3300	struct iboe_mcast_work *work;
3301	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3302	int err;
3303	struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3304	struct net_device *ndev = NULL;
3305
3306	if (cma_zero_addr((struct sockaddr *)&mc->addr))
3307		return -EINVAL;
3308
3309	work = kzalloc(sizeof *work, GFP_KERNEL);
3310	if (!work)
3311		return -ENOMEM;
3312
3313	mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3314	if (!mc->multicast.ib) {
3315		err = -ENOMEM;
3316		goto out1;
3317	}
3318
3319	cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3320
3321	mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3322	if (id_priv->id.ps == RDMA_PS_UDP)
3323		mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3324
3325	if (dev_addr->bound_dev_if)
3326		ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3327	if (!ndev) {
3328		err = -ENODEV;
3329		goto out2;
3330	}
3331	mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3332	mc->multicast.ib->rec.hop_limit = 1;
3333	mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3334	dev_put(ndev);
3335	if (!mc->multicast.ib->rec.mtu) {
3336		err = -EINVAL;
3337		goto out2;
3338	}
3339	rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3340		    &mc->multicast.ib->rec.port_gid);
3341	work->id = id_priv;
3342	work->mc = mc;
3343	INIT_WORK(&work->work, iboe_mcast_work_handler);
3344	kref_get(&mc->mcref);
3345	queue_work(cma_wq, &work->work);
3346
3347	return 0;
3348
3349out2:
3350	kfree(mc->multicast.ib);
3351out1:
3352	kfree(work);
3353	return err;
3354}
3355
3356int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3357			void *context)
3358{
3359	struct rdma_id_private *id_priv;
3360	struct cma_multicast *mc;
3361	int ret;
3362
3363	id_priv = container_of(id, struct rdma_id_private, id);
3364	if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3365	    !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3366		return -EINVAL;
3367
3368	mc = kmalloc(sizeof *mc, GFP_KERNEL);
3369	if (!mc)
3370		return -ENOMEM;
3371
3372	memcpy(&mc->addr, addr, rdma_addr_size(addr));
3373	mc->context = context;
3374	mc->id_priv = id_priv;
3375
3376	spin_lock(&id_priv->lock);
3377	list_add(&mc->list, &id_priv->mc_list);
3378	spin_unlock(&id_priv->lock);
3379
3380	switch (rdma_node_get_transport(id->device->node_type)) {
3381	case RDMA_TRANSPORT_IB:
3382		switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3383		case IB_LINK_LAYER_INFINIBAND:
3384			ret = cma_join_ib_multicast(id_priv, mc);
3385			break;
3386		case IB_LINK_LAYER_ETHERNET:
3387			kref_init(&mc->mcref);
3388			ret = cma_iboe_join_multicast(id_priv, mc);
3389			break;
3390		default:
3391			ret = -EINVAL;
3392		}
3393		break;
3394	default:
3395		ret = -ENOSYS;
3396		break;
3397	}
3398
3399	if (ret) {
3400		spin_lock_irq(&id_priv->lock);
3401		list_del(&mc->list);
3402		spin_unlock_irq(&id_priv->lock);
3403		kfree(mc);
3404	}
3405	return ret;
3406}
3407EXPORT_SYMBOL(rdma_join_multicast);
3408
3409void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3410{
3411	struct rdma_id_private *id_priv;
3412	struct cma_multicast *mc;
3413
3414	id_priv = container_of(id, struct rdma_id_private, id);
3415	spin_lock_irq(&id_priv->lock);
3416	list_for_each_entry(mc, &id_priv->mc_list, list) {
3417		if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3418			list_del(&mc->list);
3419			spin_unlock_irq(&id_priv->lock);
3420
3421			if (id->qp)
3422				ib_detach_mcast(id->qp,
3423						&mc->multicast.ib->rec.mgid,
3424						be16_to_cpu(mc->multicast.ib->rec.mlid));
3425			if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3426				switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3427				case IB_LINK_LAYER_INFINIBAND:
3428					ib_sa_free_multicast(mc->multicast.ib);
3429					kfree(mc);
3430					break;
3431				case IB_LINK_LAYER_ETHERNET:
3432					kref_put(&mc->mcref, release_mc);
3433					break;
3434				default:
3435					break;
3436				}
3437			}
3438			return;
3439		}
3440	}
3441	spin_unlock_irq(&id_priv->lock);
3442}
3443EXPORT_SYMBOL(rdma_leave_multicast);
3444
3445static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3446{
3447	struct rdma_dev_addr *dev_addr;
3448	struct cma_ndev_work *work;
3449
3450	dev_addr = &id_priv->id.route.addr.dev_addr;
3451
3452	if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3453	    memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3454		printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3455		       ndev->name, &id_priv->id);
3456		work = kzalloc(sizeof *work, GFP_KERNEL);
3457		if (!work)
3458			return -ENOMEM;
3459
3460		INIT_WORK(&work->work, cma_ndev_work_handler);
3461		work->id = id_priv;
3462		work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3463		atomic_inc(&id_priv->refcount);
3464		queue_work(cma_wq, &work->work);
3465	}
3466
3467	return 0;
3468}
3469
3470static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3471			       void *ptr)
3472{
3473	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3474	struct cma_device *cma_dev;
3475	struct rdma_id_private *id_priv;
3476	int ret = NOTIFY_DONE;
3477
3478	if (dev_net(ndev) != &init_net)
3479		return NOTIFY_DONE;
3480
3481	if (event != NETDEV_BONDING_FAILOVER)
3482		return NOTIFY_DONE;
3483
3484	if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3485		return NOTIFY_DONE;
3486
3487	mutex_lock(&lock);
3488	list_for_each_entry(cma_dev, &dev_list, list)
3489		list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3490			ret = cma_netdev_change(ndev, id_priv);
3491			if (ret)
3492				goto out;
3493		}
3494
3495out:
3496	mutex_unlock(&lock);
3497	return ret;
3498}
3499
3500static struct notifier_block cma_nb = {
3501	.notifier_call = cma_netdev_callback
3502};
3503
3504static void cma_add_one(struct ib_device *device)
3505{
3506	struct cma_device *cma_dev;
3507	struct rdma_id_private *id_priv;
3508
3509	cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3510	if (!cma_dev)
3511		return;
3512
3513	cma_dev->device = device;
3514
3515	init_completion(&cma_dev->comp);
3516	atomic_set(&cma_dev->refcount, 1);
3517	INIT_LIST_HEAD(&cma_dev->id_list);
3518	ib_set_client_data(device, &cma_client, cma_dev);
3519
3520	mutex_lock(&lock);
3521	list_add_tail(&cma_dev->list, &dev_list);
3522	list_for_each_entry(id_priv, &listen_any_list, list)
3523		cma_listen_on_dev(id_priv, cma_dev);
3524	mutex_unlock(&lock);
3525}
3526
3527static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3528{
3529	struct rdma_cm_event event;
3530	enum rdma_cm_state state;
3531	int ret = 0;
3532
3533	/* Record that we want to remove the device */
3534	state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3535	if (state == RDMA_CM_DESTROYING)
3536		return 0;
3537
3538	cma_cancel_operation(id_priv, state);
3539	mutex_lock(&id_priv->handler_mutex);
3540
3541	/* Check for destruction from another callback. */
3542	if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3543		goto out;
3544
3545	memset(&event, 0, sizeof event);
3546	event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3547	ret = id_priv->id.event_handler(&id_priv->id, &event);
3548out:
3549	mutex_unlock(&id_priv->handler_mutex);
3550	return ret;
3551}
3552
3553static void cma_process_remove(struct cma_device *cma_dev)
3554{
3555	struct rdma_id_private *id_priv;
3556	int ret;
3557
3558	mutex_lock(&lock);
3559	while (!list_empty(&cma_dev->id_list)) {
3560		id_priv = list_entry(cma_dev->id_list.next,
3561				     struct rdma_id_private, list);
3562
3563		list_del(&id_priv->listen_list);
3564		list_del_init(&id_priv->list);
3565		atomic_inc(&id_priv->refcount);
3566		mutex_unlock(&lock);
3567
3568		ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3569		cma_deref_id(id_priv);
3570		if (ret)
3571			rdma_destroy_id(&id_priv->id);
3572
3573		mutex_lock(&lock);
3574	}
3575	mutex_unlock(&lock);
3576
3577	cma_deref_dev(cma_dev);
3578	wait_for_completion(&cma_dev->comp);
3579}
3580
3581static void cma_remove_one(struct ib_device *device)
3582{
3583	struct cma_device *cma_dev;
3584
3585	cma_dev = ib_get_client_data(device, &cma_client);
3586	if (!cma_dev)
3587		return;
3588
3589	mutex_lock(&lock);
3590	list_del(&cma_dev->list);
3591	mutex_unlock(&lock);
3592
3593	cma_process_remove(cma_dev);
3594	kfree(cma_dev);
3595}
3596
3597static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3598{
3599	struct nlmsghdr *nlh;
3600	struct rdma_cm_id_stats *id_stats;
3601	struct rdma_id_private *id_priv;
3602	struct rdma_cm_id *id = NULL;
3603	struct cma_device *cma_dev;
3604	int i_dev = 0, i_id = 0;
3605
3606	/*
3607	 * We export all of the IDs as a sequence of messages.  Each
3608	 * ID gets its own netlink message.
3609	 */
3610	mutex_lock(&lock);
3611
3612	list_for_each_entry(cma_dev, &dev_list, list) {
3613		if (i_dev < cb->args[0]) {
3614			i_dev++;
3615			continue;
3616		}
3617
3618		i_id = 0;
3619		list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3620			if (i_id < cb->args[1]) {
3621				i_id++;
3622				continue;
3623			}
3624
3625			id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3626						sizeof *id_stats, RDMA_NL_RDMA_CM,
3627						RDMA_NL_RDMA_CM_ID_STATS,
3628						NLM_F_MULTI);
3629			if (!id_stats)
3630				goto out;
3631
3632			memset(id_stats, 0, sizeof *id_stats);
3633			id = &id_priv->id;
3634			id_stats->node_type = id->route.addr.dev_addr.dev_type;
3635			id_stats->port_num = id->port_num;
3636			id_stats->bound_dev_if =
3637				id->route.addr.dev_addr.bound_dev_if;
3638
3639			if (ibnl_put_attr(skb, nlh,
3640					  rdma_addr_size(cma_src_addr(id_priv)),
3641					  cma_src_addr(id_priv),
3642					  RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3643				goto out;
3644			if (ibnl_put_attr(skb, nlh,
3645					  rdma_addr_size(cma_src_addr(id_priv)),
3646					  cma_dst_addr(id_priv),
3647					  RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
3648				goto out;
3649
3650			id_stats->pid		= id_priv->owner;
3651			id_stats->port_space	= id->ps;
3652			id_stats->cm_state	= id_priv->state;
3653			id_stats->qp_num	= id_priv->qp_num;
3654			id_stats->qp_type	= id->qp_type;
3655
3656			i_id++;
3657		}
3658
3659		cb->args[1] = 0;
3660		i_dev++;
3661	}
3662
3663out:
3664	mutex_unlock(&lock);
3665	cb->args[0] = i_dev;
3666	cb->args[1] = i_id;
3667
3668	return skb->len;
3669}
3670
3671static const struct ibnl_client_cbs cma_cb_table[] = {
3672	[RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3673				       .module = THIS_MODULE },
3674};
3675
3676static int __init cma_init(void)
3677{
3678	int ret;
3679
3680	cma_wq = create_singlethread_workqueue("rdma_cm");
3681	if (!cma_wq)
3682		return -ENOMEM;
3683
3684	ib_sa_register_client(&sa_client);
3685	rdma_addr_register_client(&addr_client);
3686	register_netdevice_notifier(&cma_nb);
3687
3688	ret = ib_register_client(&cma_client);
3689	if (ret)
3690		goto err;
3691
3692	if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3693		printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3694
3695	return 0;
3696
3697err:
3698	unregister_netdevice_notifier(&cma_nb);
3699	rdma_addr_unregister_client(&addr_client);
3700	ib_sa_unregister_client(&sa_client);
3701	destroy_workqueue(cma_wq);
3702	return ret;
3703}
3704
3705static void __exit cma_cleanup(void)
3706{
3707	ibnl_remove_client(RDMA_NL_RDMA_CM);
3708	ib_unregister_client(&cma_client);
3709	unregister_netdevice_notifier(&cma_nb);
3710	rdma_addr_unregister_client(&addr_client);
3711	ib_sa_unregister_client(&sa_client);
3712	destroy_workqueue(cma_wq);
3713	idr_destroy(&tcp_ps);
3714	idr_destroy(&udp_ps);
3715	idr_destroy(&ipoib_ps);
3716	idr_destroy(&ib_ps);
3717}
3718
3719module_init(cma_init);
3720module_exit(cma_cleanup);
3721