projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
IB/Verbs: Reform cma_acquire_dev()
[deliverable/linux.git] / drivers / infiniband / core / cma.c
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
59 MODULE_AUTHOR("Sean Hefty");
60 MODULE_DESCRIPTION("Generic RDMA CM Agent");
61 MODULE_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
68 static void cma_add_one(struct ib_device *device);
69 static void cma_remove_one(struct ib_device *device);
70
71 static struct ib_client cma_client = {
72 .name = "cma",
73 .add = cma_add_one,
74 .remove = cma_remove_one
75 };
76
77 static struct ib_sa_client sa_client;
78 static struct rdma_addr_client addr_client;
79 static LIST_HEAD(dev_list);
80 static LIST_HEAD(listen_any_list);
81 static DEFINE_MUTEX(lock);
82 static struct workqueue_struct *cma_wq;
83 static DEFINE_IDR(tcp_ps);
84 static DEFINE_IDR(udp_ps);
85 static DEFINE_IDR(ipoib_ps);
86 static DEFINE_IDR(ib_ps);
87
88 struct 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
96 struct rdma_bind_list {
97 struct idr *ps;
98 struct hlist_head owners;
99 unsigned short port;
100 };
101
102 enum {
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 */
112 struct 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
151 struct 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
162 struct 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
170 struct cma_ndev_work {
171 struct work_struct work;
172 struct rdma_id_private *id;
173 struct rdma_cm_event event;
174 };
175
176 struct iboe_mcast_work {
177 struct work_struct work;
178 struct rdma_id_private *id;
179 struct cma_multicast *mc;
180 };
181
182 union cma_ip_addr {
183 struct in6_addr ip6;
184 struct {
185 __be32 pad[3];
186 __be32 addr;
187 } ip4;
188 };
189
190 struct 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
200 static 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
211 static 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
224 static 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
237 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
238 {
239 return hdr->ip_version >> 4;
240 }
241
242 static 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
247 static 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
258 static 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
264 static 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
272 static 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
281 static 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
286 static 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
291 static 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
296 static 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
331 static 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
338 static 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
352 static inline int cma_validate_port(struct ib_device *device, u8 port,
353 union ib_gid *gid, int dev_type)
354 {
355 u8 found_port;
356 int ret = -ENODEV;
357
358 if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
359 return ret;
360
361 if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
362 return ret;
363
364 ret = ib_find_cached_gid(device, gid, &found_port, NULL);
365 if (port != found_port)
366 return -ENODEV;
367
368 return ret;
369 }
370
371 static int cma_acquire_dev(struct rdma_id_private *id_priv,
372 struct rdma_id_private *listen_id_priv)
373 {
374 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
375 struct cma_device *cma_dev;
376 union ib_gid gid, iboe_gid, *gidp;
377 int ret = -ENODEV;
378 u8 port;
379
380 if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
381 id_priv->id.ps == RDMA_PS_IPOIB)
382 return -EINVAL;
383
384 mutex_lock(&lock);
385 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
386 &iboe_gid);
387
388 memcpy(&gid, dev_addr->src_dev_addr +
389 rdma_addr_gid_offset(dev_addr), sizeof gid);
390
391 if (listen_id_priv) {
392 cma_dev = listen_id_priv->cma_dev;
393 port = listen_id_priv->id.port_num;
394 gidp = rdma_protocol_iboe(cma_dev->device, port) ?
395 &iboe_gid : &gid;
396
397 ret = cma_validate_port(cma_dev->device, port, gidp,
398 dev_addr->dev_type);
399 if (!ret) {
400 id_priv->id.port_num = port;
401 goto out;
402 }
403 }
404
405 list_for_each_entry(cma_dev, &dev_list, list) {
406 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
407 if (listen_id_priv &&
408 listen_id_priv->cma_dev == cma_dev &&
409 listen_id_priv->id.port_num == port)
410 continue;
411
412 gidp = rdma_protocol_iboe(cma_dev->device, port) ?
413 &iboe_gid : &gid;
414
415 ret = cma_validate_port(cma_dev->device, port, gidp,
416 dev_addr->dev_type);
417 if (!ret) {
418 id_priv->id.port_num = port;
419 goto out;
420 }
421 }
422 }
423
424 out:
425 if (!ret)
426 cma_attach_to_dev(id_priv, cma_dev);
427
428 mutex_unlock(&lock);
429 return ret;
430 }
431
432 /*
433 * Select the source IB device and address to reach the destination IB address.
434 */
435 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
436 {
437 struct cma_device *cma_dev, *cur_dev;
438 struct sockaddr_ib *addr;
439 union ib_gid gid, sgid, *dgid;
440 u16 pkey, index;
441 u8 p;
442 int i;
443
444 cma_dev = NULL;
445 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
446 dgid = (union ib_gid *) &addr->sib_addr;
447 pkey = ntohs(addr->sib_pkey);
448
449 list_for_each_entry(cur_dev, &dev_list, list) {
450 if (rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
451 continue;
452
453 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
454 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
455 continue;
456
457 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
458 if (!memcmp(&gid, dgid, sizeof(gid))) {
459 cma_dev = cur_dev;
460 sgid = gid;
461 id_priv->id.port_num = p;
462 goto found;
463 }
464
465 if (!cma_dev && (gid.global.subnet_prefix ==
466 dgid->global.subnet_prefix)) {
467 cma_dev = cur_dev;
468 sgid = gid;
469 id_priv->id.port_num = p;
470 }
471 }
472 }
473 }
474
475 if (!cma_dev)
476 return -ENODEV;
477
478 found:
479 cma_attach_to_dev(id_priv, cma_dev);
480 addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
481 memcpy(&addr->sib_addr, &sgid, sizeof sgid);
482 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
483 return 0;
484 }
485
486 static void cma_deref_id(struct rdma_id_private *id_priv)
487 {
488 if (atomic_dec_and_test(&id_priv->refcount))
489 complete(&id_priv->comp);
490 }
491
492 static int cma_disable_callback(struct rdma_id_private *id_priv,
493 enum rdma_cm_state state)
494 {
495 mutex_lock(&id_priv->handler_mutex);
496 if (id_priv->state != state) {
497 mutex_unlock(&id_priv->handler_mutex);
498 return -EINVAL;
499 }
500 return 0;
501 }
502
503 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
504 void *context, enum rdma_port_space ps,
505 enum ib_qp_type qp_type)
506 {
507 struct rdma_id_private *id_priv;
508
509 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
510 if (!id_priv)
511 return ERR_PTR(-ENOMEM);
512
513 id_priv->owner = task_pid_nr(current);
514 id_priv->state = RDMA_CM_IDLE;
515 id_priv->id.context = context;
516 id_priv->id.event_handler = event_handler;
517 id_priv->id.ps = ps;
518 id_priv->id.qp_type = qp_type;
519 spin_lock_init(&id_priv->lock);
520 mutex_init(&id_priv->qp_mutex);
521 init_completion(&id_priv->comp);
522 atomic_set(&id_priv->refcount, 1);
523 mutex_init(&id_priv->handler_mutex);
524 INIT_LIST_HEAD(&id_priv->listen_list);
525 INIT_LIST_HEAD(&id_priv->mc_list);
526 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
527
528 return &id_priv->id;
529 }
530 EXPORT_SYMBOL(rdma_create_id);
531
532 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
533 {
534 struct ib_qp_attr qp_attr;
535 int qp_attr_mask, ret;
536
537 qp_attr.qp_state = IB_QPS_INIT;
538 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
539 if (ret)
540 return ret;
541
542 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
543 if (ret)
544 return ret;
545
546 qp_attr.qp_state = IB_QPS_RTR;
547 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
548 if (ret)
549 return ret;
550
551 qp_attr.qp_state = IB_QPS_RTS;
552 qp_attr.sq_psn = 0;
553 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
554
555 return ret;
556 }
557
558 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
559 {
560 struct ib_qp_attr qp_attr;
561 int qp_attr_mask, ret;
562
563 qp_attr.qp_state = IB_QPS_INIT;
564 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
565 if (ret)
566 return ret;
567
568 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
569 }
570
571 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
572 struct ib_qp_init_attr *qp_init_attr)
573 {
574 struct rdma_id_private *id_priv;
575 struct ib_qp *qp;
576 int ret;
577
578 id_priv = container_of(id, struct rdma_id_private, id);
579 if (id->device != pd->device)
580 return -EINVAL;
581
582 qp = ib_create_qp(pd, qp_init_attr);
583 if (IS_ERR(qp))
584 return PTR_ERR(qp);
585
586 if (id->qp_type == IB_QPT_UD)
587 ret = cma_init_ud_qp(id_priv, qp);
588 else
589 ret = cma_init_conn_qp(id_priv, qp);
590 if (ret)
591 goto err;
592
593 id->qp = qp;
594 id_priv->qp_num = qp->qp_num;
595 id_priv->srq = (qp->srq != NULL);
596 return 0;
597 err:
598 ib_destroy_qp(qp);
599 return ret;
600 }
601 EXPORT_SYMBOL(rdma_create_qp);
602
603 void rdma_destroy_qp(struct rdma_cm_id *id)
604 {
605 struct rdma_id_private *id_priv;
606
607 id_priv = container_of(id, struct rdma_id_private, id);
608 mutex_lock(&id_priv->qp_mutex);
609 ib_destroy_qp(id_priv->id.qp);
610 id_priv->id.qp = NULL;
611 mutex_unlock(&id_priv->qp_mutex);
612 }
613 EXPORT_SYMBOL(rdma_destroy_qp);
614
615 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
616 struct rdma_conn_param *conn_param)
617 {
618 struct ib_qp_attr qp_attr;
619 int qp_attr_mask, ret;
620 union ib_gid sgid;
621
622 mutex_lock(&id_priv->qp_mutex);
623 if (!id_priv->id.qp) {
624 ret = 0;
625 goto out;
626 }
627
628 /* Need to update QP attributes from default values. */
629 qp_attr.qp_state = IB_QPS_INIT;
630 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
631 if (ret)
632 goto out;
633
634 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
635 if (ret)
636 goto out;
637
638 qp_attr.qp_state = IB_QPS_RTR;
639 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
640 if (ret)
641 goto out;
642
643 ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
644 qp_attr.ah_attr.grh.sgid_index, &sgid);
645 if (ret)
646 goto out;
647
648 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type)
649 == RDMA_TRANSPORT_IB &&
650 rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)
651 == IB_LINK_LAYER_ETHERNET) {
652 ret = rdma_addr_find_smac_by_sgid(&sgid, qp_attr.smac, NULL);
653
654 if (ret)
655 goto out;
656 }
657 if (conn_param)
658 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
659 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
660 out:
661 mutex_unlock(&id_priv->qp_mutex);
662 return ret;
663 }
664
665 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
666 struct rdma_conn_param *conn_param)
667 {
668 struct ib_qp_attr qp_attr;
669 int qp_attr_mask, ret;
670
671 mutex_lock(&id_priv->qp_mutex);
672 if (!id_priv->id.qp) {
673 ret = 0;
674 goto out;
675 }
676
677 qp_attr.qp_state = IB_QPS_RTS;
678 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
679 if (ret)
680 goto out;
681
682 if (conn_param)
683 qp_attr.max_rd_atomic = conn_param->initiator_depth;
684 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
685 out:
686 mutex_unlock(&id_priv->qp_mutex);
687 return ret;
688 }
689
690 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
691 {
692 struct ib_qp_attr qp_attr;
693 int ret;
694
695 mutex_lock(&id_priv->qp_mutex);
696 if (!id_priv->id.qp) {
697 ret = 0;
698 goto out;
699 }
700
701 qp_attr.qp_state = IB_QPS_ERR;
702 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
703 out:
704 mutex_unlock(&id_priv->qp_mutex);
705 return ret;
706 }
707
708 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
709 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
710 {
711 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
712 int ret;
713 u16 pkey;
714
715 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
716 IB_LINK_LAYER_INFINIBAND)
717 pkey = ib_addr_get_pkey(dev_addr);
718 else
719 pkey = 0xffff;
720
721 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
722 pkey, &qp_attr->pkey_index);
723 if (ret)
724 return ret;
725
726 qp_attr->port_num = id_priv->id.port_num;
727 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
728
729 if (id_priv->id.qp_type == IB_QPT_UD) {
730 ret = cma_set_qkey(id_priv, 0);
731 if (ret)
732 return ret;
733
734 qp_attr->qkey = id_priv->qkey;
735 *qp_attr_mask |= IB_QP_QKEY;
736 } else {
737 qp_attr->qp_access_flags = 0;
738 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
739 }
740 return 0;
741 }
742
743 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
744 int *qp_attr_mask)
745 {
746 struct rdma_id_private *id_priv;
747 int ret = 0;
748
749 id_priv = container_of(id, struct rdma_id_private, id);
750 if (rdma_ib_or_iboe(id->device, id->port_num)) {
751 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
752 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
753 else
754 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
755 qp_attr_mask);
756
757 if (qp_attr->qp_state == IB_QPS_RTR)
758 qp_attr->rq_psn = id_priv->seq_num;
759 } else if (rdma_protocol_iwarp(id->device, id->port_num)) {
760 if (!id_priv->cm_id.iw) {
761 qp_attr->qp_access_flags = 0;
762 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
763 } else
764 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
765 qp_attr_mask);
766 } else
767 ret = -ENOSYS;
768
769 return ret;
770 }
771 EXPORT_SYMBOL(rdma_init_qp_attr);
772
773 static inline int cma_zero_addr(struct sockaddr *addr)
774 {
775 switch (addr->sa_family) {
776 case AF_INET:
777 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
778 case AF_INET6:
779 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
780 case AF_IB:
781 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
782 default:
783 return 0;
784 }
785 }
786
787 static inline int cma_loopback_addr(struct sockaddr *addr)
788 {
789 switch (addr->sa_family) {
790 case AF_INET:
791 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
792 case AF_INET6:
793 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
794 case AF_IB:
795 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
796 default:
797 return 0;
798 }
799 }
800
801 static inline int cma_any_addr(struct sockaddr *addr)
802 {
803 return cma_zero_addr(addr) || cma_loopback_addr(addr);
804 }
805
806 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
807 {
808 if (src->sa_family != dst->sa_family)
809 return -1;
810
811 switch (src->sa_family) {
812 case AF_INET:
813 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
814 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
815 case AF_INET6:
816 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
817 &((struct sockaddr_in6 *) dst)->sin6_addr);
818 default:
819 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
820 &((struct sockaddr_ib *) dst)->sib_addr);
821 }
822 }
823
824 static __be16 cma_port(struct sockaddr *addr)
825 {
826 struct sockaddr_ib *sib;
827
828 switch (addr->sa_family) {
829 case AF_INET:
830 return ((struct sockaddr_in *) addr)->sin_port;
831 case AF_INET6:
832 return ((struct sockaddr_in6 *) addr)->sin6_port;
833 case AF_IB:
834 sib = (struct sockaddr_ib *) addr;
835 return htons((u16) (be64_to_cpu(sib->sib_sid) &
836 be64_to_cpu(sib->sib_sid_mask)));
837 default:
838 return 0;
839 }
840 }
841
842 static inline int cma_any_port(struct sockaddr *addr)
843 {
844 return !cma_port(addr);
845 }
846
847 static void cma_save_ib_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
848 struct ib_sa_path_rec *path)
849 {
850 struct sockaddr_ib *listen_ib, *ib;
851
852 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
853 ib = (struct sockaddr_ib *) &id->route.addr.src_addr;
854 ib->sib_family = listen_ib->sib_family;
855 ib->sib_pkey = path->pkey;
856 ib->sib_flowinfo = path->flow_label;
857 memcpy(&ib->sib_addr, &path->sgid, 16);
858 ib->sib_sid = listen_ib->sib_sid;
859 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
860 ib->sib_scope_id = listen_ib->sib_scope_id;
861
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 static __be16 ss_get_port(const struct sockaddr_storage *ss)
870 {
871 if (ss->ss_family == AF_INET)
872 return ((struct sockaddr_in *)ss)->sin_port;
873 else if (ss->ss_family == AF_INET6)
874 return ((struct sockaddr_in6 *)ss)->sin6_port;
875 BUG();
876 }
877
878 static void cma_save_ip4_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
879 struct cma_hdr *hdr)
880 {
881 struct sockaddr_in *ip4;
882
883 ip4 = (struct sockaddr_in *) &id->route.addr.src_addr;
884 ip4->sin_family = AF_INET;
885 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
886 ip4->sin_port = ss_get_port(&listen_id->route.addr.src_addr);
887
888 ip4 = (struct sockaddr_in *) &id->route.addr.dst_addr;
889 ip4->sin_family = AF_INET;
890 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
891 ip4->sin_port = hdr->port;
892 }
893
894 static void cma_save_ip6_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
895 struct cma_hdr *hdr)
896 {
897 struct sockaddr_in6 *ip6;
898
899 ip6 = (struct sockaddr_in6 *) &id->route.addr.src_addr;
900 ip6->sin6_family = AF_INET6;
901 ip6->sin6_addr = hdr->dst_addr.ip6;
902 ip6->sin6_port = ss_get_port(&listen_id->route.addr.src_addr);
903
904 ip6 = (struct sockaddr_in6 *) &id->route.addr.dst_addr;
905 ip6->sin6_family = AF_INET6;
906 ip6->sin6_addr = hdr->src_addr.ip6;
907 ip6->sin6_port = hdr->port;
908 }
909
910 static int cma_save_net_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
911 struct ib_cm_event *ib_event)
912 {
913 struct cma_hdr *hdr;
914
915 if ((listen_id->route.addr.src_addr.ss_family == AF_IB) &&
916 (ib_event->event == IB_CM_REQ_RECEIVED)) {
917 cma_save_ib_info(id, listen_id, ib_event->param.req_rcvd.primary_path);
918 return 0;
919 }
920
921 hdr = ib_event->private_data;
922 if (hdr->cma_version != CMA_VERSION)
923 return -EINVAL;
924
925 switch (cma_get_ip_ver(hdr)) {
926 case 4:
927 cma_save_ip4_info(id, listen_id, hdr);
928 break;
929 case 6:
930 cma_save_ip6_info(id, listen_id, hdr);
931 break;
932 default:
933 return -EINVAL;
934 }
935 return 0;
936 }
937
938 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
939 {
940 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
941 }
942
943 static void cma_cancel_route(struct rdma_id_private *id_priv)
944 {
945 if (rdma_protocol_ib(id_priv->id.device, id_priv->id.port_num)) {
946 if (id_priv->query)
947 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
948 }
949 }
950
951 static void cma_cancel_listens(struct rdma_id_private *id_priv)
952 {
953 struct rdma_id_private *dev_id_priv;
954
955 /*
956 * Remove from listen_any_list to prevent added devices from spawning
957 * additional listen requests.
958 */
959 mutex_lock(&lock);
960 list_del(&id_priv->list);
961
962 while (!list_empty(&id_priv->listen_list)) {
963 dev_id_priv = list_entry(id_priv->listen_list.next,
964 struct rdma_id_private, listen_list);
965 /* sync with device removal to avoid duplicate destruction */
966 list_del_init(&dev_id_priv->list);
967 list_del(&dev_id_priv->listen_list);
968 mutex_unlock(&lock);
969
970 rdma_destroy_id(&dev_id_priv->id);
971 mutex_lock(&lock);
972 }
973 mutex_unlock(&lock);
974 }
975
976 static void cma_cancel_operation(struct rdma_id_private *id_priv,
977 enum rdma_cm_state state)
978 {
979 switch (state) {
980 case RDMA_CM_ADDR_QUERY:
981 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
982 break;
983 case RDMA_CM_ROUTE_QUERY:
984 cma_cancel_route(id_priv);
985 break;
986 case RDMA_CM_LISTEN:
987 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
988 cma_cancel_listens(id_priv);
989 break;
990 default:
991 break;
992 }
993 }
994
995 static void cma_release_port(struct rdma_id_private *id_priv)
996 {
997 struct rdma_bind_list *bind_list = id_priv->bind_list;
998
999 if (!bind_list)
1000 return;
1001
1002 mutex_lock(&lock);
1003 hlist_del(&id_priv->node);
1004 if (hlist_empty(&bind_list->owners)) {
1005 idr_remove(bind_list->ps, bind_list->port);
1006 kfree(bind_list);
1007 }
1008 mutex_unlock(&lock);
1009 }
1010
1011 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1012 {
1013 struct cma_multicast *mc;
1014
1015 while (!list_empty(&id_priv->mc_list)) {
1016 mc = container_of(id_priv->mc_list.next,
1017 struct cma_multicast, list);
1018 list_del(&mc->list);
1019 if (rdma_protocol_ib(id_priv->cma_dev->device,
1020 id_priv->id.port_num)) {
1021 ib_sa_free_multicast(mc->multicast.ib);
1022 kfree(mc);
1023 } else
1024 kref_put(&mc->mcref, release_mc);
1025 }
1026 }
1027
1028 void rdma_destroy_id(struct rdma_cm_id *id)
1029 {
1030 struct rdma_id_private *id_priv;
1031 enum rdma_cm_state state;
1032
1033 id_priv = container_of(id, struct rdma_id_private, id);
1034 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1035 cma_cancel_operation(id_priv, state);
1036
1037 /*
1038 * Wait for any active callback to finish. New callbacks will find
1039 * the id_priv state set to destroying and abort.
1040 */
1041 mutex_lock(&id_priv->handler_mutex);
1042 mutex_unlock(&id_priv->handler_mutex);
1043
1044 if (id_priv->cma_dev) {
1045 if (rdma_ib_or_iboe(id_priv->id.device, 1)) {
1046 if (id_priv->cm_id.ib)
1047 ib_destroy_cm_id(id_priv->cm_id.ib);
1048 } else if (rdma_protocol_iwarp(id_priv->id.device, 1)) {
1049 if (id_priv->cm_id.iw)
1050 iw_destroy_cm_id(id_priv->cm_id.iw);
1051 }
1052 cma_leave_mc_groups(id_priv);
1053 cma_release_dev(id_priv);
1054 }
1055
1056 cma_release_port(id_priv);
1057 cma_deref_id(id_priv);
1058 wait_for_completion(&id_priv->comp);
1059
1060 if (id_priv->internal_id)
1061 cma_deref_id(id_priv->id.context);
1062
1063 kfree(id_priv->id.route.path_rec);
1064 kfree(id_priv);
1065 }
1066 EXPORT_SYMBOL(rdma_destroy_id);
1067
1068 static int cma_rep_recv(struct rdma_id_private *id_priv)
1069 {
1070 int ret;
1071
1072 ret = cma_modify_qp_rtr(id_priv, NULL);
1073 if (ret)
1074 goto reject;
1075
1076 ret = cma_modify_qp_rts(id_priv, NULL);
1077 if (ret)
1078 goto reject;
1079
1080 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1081 if (ret)
1082 goto reject;
1083
1084 return 0;
1085 reject:
1086 cma_modify_qp_err(id_priv);
1087 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1088 NULL, 0, NULL, 0);
1089 return ret;
1090 }
1091
1092 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1093 struct ib_cm_rep_event_param *rep_data,
1094 void *private_data)
1095 {
1096 event->param.conn.private_data = private_data;
1097 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1098 event->param.conn.responder_resources = rep_data->responder_resources;
1099 event->param.conn.initiator_depth = rep_data->initiator_depth;
1100 event->param.conn.flow_control = rep_data->flow_control;
1101 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1102 event->param.conn.srq = rep_data->srq;
1103 event->param.conn.qp_num = rep_data->remote_qpn;
1104 }
1105
1106 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1107 {
1108 struct rdma_id_private *id_priv = cm_id->context;
1109 struct rdma_cm_event event;
1110 int ret = 0;
1111
1112 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1113 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1114 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1115 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1116 return 0;
1117
1118 memset(&event, 0, sizeof event);
1119 switch (ib_event->event) {
1120 case IB_CM_REQ_ERROR:
1121 case IB_CM_REP_ERROR:
1122 event.event = RDMA_CM_EVENT_UNREACHABLE;
1123 event.status = -ETIMEDOUT;
1124 break;
1125 case IB_CM_REP_RECEIVED:
1126 if (id_priv->id.qp) {
1127 event.status = cma_rep_recv(id_priv);
1128 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1129 RDMA_CM_EVENT_ESTABLISHED;
1130 } else {
1131 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1132 }
1133 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1134 ib_event->private_data);
1135 break;
1136 case IB_CM_RTU_RECEIVED:
1137 case IB_CM_USER_ESTABLISHED:
1138 event.event = RDMA_CM_EVENT_ESTABLISHED;
1139 break;
1140 case IB_CM_DREQ_ERROR:
1141 event.status = -ETIMEDOUT; /* fall through */
1142 case IB_CM_DREQ_RECEIVED:
1143 case IB_CM_DREP_RECEIVED:
1144 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1145 RDMA_CM_DISCONNECT))
1146 goto out;
1147 event.event = RDMA_CM_EVENT_DISCONNECTED;
1148 break;
1149 case IB_CM_TIMEWAIT_EXIT:
1150 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1151 break;
1152 case IB_CM_MRA_RECEIVED:
1153 /* ignore event */
1154 goto out;
1155 case IB_CM_REJ_RECEIVED:
1156 cma_modify_qp_err(id_priv);
1157 event.status = ib_event->param.rej_rcvd.reason;
1158 event.event = RDMA_CM_EVENT_REJECTED;
1159 event.param.conn.private_data = ib_event->private_data;
1160 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1161 break;
1162 default:
1163 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1164 ib_event->event);
1165 goto out;
1166 }
1167
1168 ret = id_priv->id.event_handler(&id_priv->id, &event);
1169 if (ret) {
1170 /* Destroy the CM ID by returning a non-zero value. */
1171 id_priv->cm_id.ib = NULL;
1172 cma_exch(id_priv, RDMA_CM_DESTROYING);
1173 mutex_unlock(&id_priv->handler_mutex);
1174 rdma_destroy_id(&id_priv->id);
1175 return ret;
1176 }
1177 out:
1178 mutex_unlock(&id_priv->handler_mutex);
1179 return ret;
1180 }
1181
1182 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1183 struct ib_cm_event *ib_event)
1184 {
1185 struct rdma_id_private *id_priv;
1186 struct rdma_cm_id *id;
1187 struct rdma_route *rt;
1188 int ret;
1189
1190 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1191 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1192 if (IS_ERR(id))
1193 return NULL;
1194
1195 id_priv = container_of(id, struct rdma_id_private, id);
1196 if (cma_save_net_info(id, listen_id, ib_event))
1197 goto err;
1198
1199 rt = &id->route;
1200 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1201 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1202 GFP_KERNEL);
1203 if (!rt->path_rec)
1204 goto err;
1205
1206 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1207 if (rt->num_paths == 2)
1208 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1209
1210 if (cma_any_addr(cma_src_addr(id_priv))) {
1211 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1212 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1213 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1214 } else {
1215 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1216 if (ret)
1217 goto err;
1218 }
1219 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1220
1221 id_priv->state = RDMA_CM_CONNECT;
1222 return id_priv;
1223
1224 err:
1225 rdma_destroy_id(id);
1226 return NULL;
1227 }
1228
1229 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1230 struct ib_cm_event *ib_event)
1231 {
1232 struct rdma_id_private *id_priv;
1233 struct rdma_cm_id *id;
1234 int ret;
1235
1236 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1237 listen_id->ps, IB_QPT_UD);
1238 if (IS_ERR(id))
1239 return NULL;
1240
1241 id_priv = container_of(id, struct rdma_id_private, id);
1242 if (cma_save_net_info(id, listen_id, ib_event))
1243 goto err;
1244
1245 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1246 ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
1247 if (ret)
1248 goto err;
1249 }
1250
1251 id_priv->state = RDMA_CM_CONNECT;
1252 return id_priv;
1253 err:
1254 rdma_destroy_id(id);
1255 return NULL;
1256 }
1257
1258 static void cma_set_req_event_data(struct rdma_cm_event *event,
1259 struct ib_cm_req_event_param *req_data,
1260 void *private_data, int offset)
1261 {
1262 event->param.conn.private_data = private_data + offset;
1263 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1264 event->param.conn.responder_resources = req_data->responder_resources;
1265 event->param.conn.initiator_depth = req_data->initiator_depth;
1266 event->param.conn.flow_control = req_data->flow_control;
1267 event->param.conn.retry_count = req_data->retry_count;
1268 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1269 event->param.conn.srq = req_data->srq;
1270 event->param.conn.qp_num = req_data->remote_qpn;
1271 }
1272
1273 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1274 {
1275 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1276 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1277 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1278 (id->qp_type == IB_QPT_UD)) ||
1279 (!id->qp_type));
1280 }
1281
1282 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1283 {
1284 struct rdma_id_private *listen_id, *conn_id;
1285 struct rdma_cm_event event;
1286 int offset, ret;
1287
1288 listen_id = cm_id->context;
1289 if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1290 return -EINVAL;
1291
1292 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1293 return -ECONNABORTED;
1294
1295 memset(&event, 0, sizeof event);
1296 offset = cma_user_data_offset(listen_id);
1297 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1298 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1299 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1300 event.param.ud.private_data = ib_event->private_data + offset;
1301 event.param.ud.private_data_len =
1302 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1303 } else {
1304 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1305 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1306 ib_event->private_data, offset);
1307 }
1308 if (!conn_id) {
1309 ret = -ENOMEM;
1310 goto err1;
1311 }
1312
1313 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1314 ret = cma_acquire_dev(conn_id, listen_id);
1315 if (ret)
1316 goto err2;
1317
1318 conn_id->cm_id.ib = cm_id;
1319 cm_id->context = conn_id;
1320 cm_id->cm_handler = cma_ib_handler;
1321
1322 /*
1323 * Protect against the user destroying conn_id from another thread
1324 * until we're done accessing it.
1325 */
1326 atomic_inc(&conn_id->refcount);
1327 ret = conn_id->id.event_handler(&conn_id->id, &event);
1328 if (ret)
1329 goto err3;
1330 /*
1331 * Acquire mutex to prevent user executing rdma_destroy_id()
1332 * while we're accessing the cm_id.
1333 */
1334 mutex_lock(&lock);
1335 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1336 (conn_id->id.qp_type != IB_QPT_UD))
1337 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1338 mutex_unlock(&lock);
1339 mutex_unlock(&conn_id->handler_mutex);
1340 mutex_unlock(&listen_id->handler_mutex);
1341 cma_deref_id(conn_id);
1342 return 0;
1343
1344 err3:
1345 cma_deref_id(conn_id);
1346 /* Destroy the CM ID by returning a non-zero value. */
1347 conn_id->cm_id.ib = NULL;
1348 err2:
1349 cma_exch(conn_id, RDMA_CM_DESTROYING);
1350 mutex_unlock(&conn_id->handler_mutex);
1351 err1:
1352 mutex_unlock(&listen_id->handler_mutex);
1353 if (conn_id)
1354 rdma_destroy_id(&conn_id->id);
1355 return ret;
1356 }
1357
1358 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1359 {
1360 if (addr->sa_family == AF_IB)
1361 return ((struct sockaddr_ib *) addr)->sib_sid;
1362
1363 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1364 }
1365 EXPORT_SYMBOL(rdma_get_service_id);
1366
1367 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1368 struct ib_cm_compare_data *compare)
1369 {
1370 struct cma_hdr *cma_data, *cma_mask;
1371 __be32 ip4_addr;
1372 struct in6_addr ip6_addr;
1373
1374 memset(compare, 0, sizeof *compare);
1375 cma_data = (void *) compare->data;
1376 cma_mask = (void *) compare->mask;
1377
1378 switch (addr->sa_family) {
1379 case AF_INET:
1380 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1381 cma_set_ip_ver(cma_data, 4);
1382 cma_set_ip_ver(cma_mask, 0xF);
1383 if (!cma_any_addr(addr)) {
1384 cma_data->dst_addr.ip4.addr = ip4_addr;
1385 cma_mask->dst_addr.ip4.addr = htonl(~0);
1386 }
1387 break;
1388 case AF_INET6:
1389 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1390 cma_set_ip_ver(cma_data, 6);
1391 cma_set_ip_ver(cma_mask, 0xF);
1392 if (!cma_any_addr(addr)) {
1393 cma_data->dst_addr.ip6 = ip6_addr;
1394 memset(&cma_mask->dst_addr.ip6, 0xFF,
1395 sizeof cma_mask->dst_addr.ip6);
1396 }
1397 break;
1398 default:
1399 break;
1400 }
1401 }
1402
1403 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1404 {
1405 struct rdma_id_private *id_priv = iw_id->context;
1406 struct rdma_cm_event event;
1407 int ret = 0;
1408 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1409 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1410
1411 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1412 return 0;
1413
1414 memset(&event, 0, sizeof event);
1415 switch (iw_event->event) {
1416 case IW_CM_EVENT_CLOSE:
1417 event.event = RDMA_CM_EVENT_DISCONNECTED;
1418 break;
1419 case IW_CM_EVENT_CONNECT_REPLY:
1420 memcpy(cma_src_addr(id_priv), laddr,
1421 rdma_addr_size(laddr));
1422 memcpy(cma_dst_addr(id_priv), raddr,
1423 rdma_addr_size(raddr));
1424 switch (iw_event->status) {
1425 case 0:
1426 event.event = RDMA_CM_EVENT_ESTABLISHED;
1427 event.param.conn.initiator_depth = iw_event->ird;
1428 event.param.conn.responder_resources = iw_event->ord;
1429 break;
1430 case -ECONNRESET:
1431 case -ECONNREFUSED:
1432 event.event = RDMA_CM_EVENT_REJECTED;
1433 break;
1434 case -ETIMEDOUT:
1435 event.event = RDMA_CM_EVENT_UNREACHABLE;
1436 break;
1437 default:
1438 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1439 break;
1440 }
1441 break;
1442 case IW_CM_EVENT_ESTABLISHED:
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 default:
1448 BUG_ON(1);
1449 }
1450
1451 event.status = iw_event->status;
1452 event.param.conn.private_data = iw_event->private_data;
1453 event.param.conn.private_data_len = iw_event->private_data_len;
1454 ret = id_priv->id.event_handler(&id_priv->id, &event);
1455 if (ret) {
1456 /* Destroy the CM ID by returning a non-zero value. */
1457 id_priv->cm_id.iw = NULL;
1458 cma_exch(id_priv, RDMA_CM_DESTROYING);
1459 mutex_unlock(&id_priv->handler_mutex);
1460 rdma_destroy_id(&id_priv->id);
1461 return ret;
1462 }
1463
1464 mutex_unlock(&id_priv->handler_mutex);
1465 return ret;
1466 }
1467
1468 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1469 struct iw_cm_event *iw_event)
1470 {
1471 struct rdma_cm_id *new_cm_id;
1472 struct rdma_id_private *listen_id, *conn_id;
1473 struct rdma_cm_event event;
1474 int ret;
1475 struct ib_device_attr attr;
1476 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1477 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1478
1479 listen_id = cm_id->context;
1480 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1481 return -ECONNABORTED;
1482
1483 /* Create a new RDMA id for the new IW CM ID */
1484 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1485 listen_id->id.context,
1486 RDMA_PS_TCP, IB_QPT_RC);
1487 if (IS_ERR(new_cm_id)) {
1488 ret = -ENOMEM;
1489 goto out;
1490 }
1491 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1492 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1493 conn_id->state = RDMA_CM_CONNECT;
1494
1495 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1496 if (ret) {
1497 mutex_unlock(&conn_id->handler_mutex);
1498 rdma_destroy_id(new_cm_id);
1499 goto out;
1500 }
1501
1502 ret = cma_acquire_dev(conn_id, listen_id);
1503 if (ret) {
1504 mutex_unlock(&conn_id->handler_mutex);
1505 rdma_destroy_id(new_cm_id);
1506 goto out;
1507 }
1508
1509 conn_id->cm_id.iw = cm_id;
1510 cm_id->context = conn_id;
1511 cm_id->cm_handler = cma_iw_handler;
1512
1513 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1514 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1515
1516 ret = ib_query_device(conn_id->id.device, &attr);
1517 if (ret) {
1518 mutex_unlock(&conn_id->handler_mutex);
1519 rdma_destroy_id(new_cm_id);
1520 goto out;
1521 }
1522
1523 memset(&event, 0, sizeof event);
1524 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1525 event.param.conn.private_data = iw_event->private_data;
1526 event.param.conn.private_data_len = iw_event->private_data_len;
1527 event.param.conn.initiator_depth = iw_event->ird;
1528 event.param.conn.responder_resources = iw_event->ord;
1529
1530 /*
1531 * Protect against the user destroying conn_id from another thread
1532 * until we're done accessing it.
1533 */
1534 atomic_inc(&conn_id->refcount);
1535 ret = conn_id->id.event_handler(&conn_id->id, &event);
1536 if (ret) {
1537 /* User wants to destroy the CM ID */
1538 conn_id->cm_id.iw = NULL;
1539 cma_exch(conn_id, RDMA_CM_DESTROYING);
1540 mutex_unlock(&conn_id->handler_mutex);
1541 cma_deref_id(conn_id);
1542 rdma_destroy_id(&conn_id->id);
1543 goto out;
1544 }
1545
1546 mutex_unlock(&conn_id->handler_mutex);
1547 cma_deref_id(conn_id);
1548
1549 out:
1550 mutex_unlock(&listen_id->handler_mutex);
1551 return ret;
1552 }
1553
1554 static int cma_ib_listen(struct rdma_id_private *id_priv)
1555 {
1556 struct ib_cm_compare_data compare_data;
1557 struct sockaddr *addr;
1558 struct ib_cm_id *id;
1559 __be64 svc_id;
1560 int ret;
1561
1562 id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1563 if (IS_ERR(id))
1564 return PTR_ERR(id);
1565
1566 id_priv->cm_id.ib = id;
1567
1568 addr = cma_src_addr(id_priv);
1569 svc_id = rdma_get_service_id(&id_priv->id, addr);
1570 if (cma_any_addr(addr) && !id_priv->afonly)
1571 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1572 else {
1573 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1574 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1575 }
1576
1577 if (ret) {
1578 ib_destroy_cm_id(id_priv->cm_id.ib);
1579 id_priv->cm_id.ib = NULL;
1580 }
1581
1582 return ret;
1583 }
1584
1585 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1586 {
1587 int ret;
1588 struct iw_cm_id *id;
1589
1590 id = iw_create_cm_id(id_priv->id.device,
1591 iw_conn_req_handler,
1592 id_priv);
1593 if (IS_ERR(id))
1594 return PTR_ERR(id);
1595
1596 id_priv->cm_id.iw = id;
1597
1598 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
1599 rdma_addr_size(cma_src_addr(id_priv)));
1600
1601 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1602
1603 if (ret) {
1604 iw_destroy_cm_id(id_priv->cm_id.iw);
1605 id_priv->cm_id.iw = NULL;
1606 }
1607
1608 return ret;
1609 }
1610
1611 static int cma_listen_handler(struct rdma_cm_id *id,
1612 struct rdma_cm_event *event)
1613 {
1614 struct rdma_id_private *id_priv = id->context;
1615
1616 id->context = id_priv->id.context;
1617 id->event_handler = id_priv->id.event_handler;
1618 return id_priv->id.event_handler(id, event);
1619 }
1620
1621 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1622 struct cma_device *cma_dev)
1623 {
1624 struct rdma_id_private *dev_id_priv;
1625 struct rdma_cm_id *id;
1626 int ret;
1627
1628 if (cma_family(id_priv) == AF_IB &&
1629 !rdma_ib_or_iboe(cma_dev->device, 1))
1630 return;
1631
1632 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1633 id_priv->id.qp_type);
1634 if (IS_ERR(id))
1635 return;
1636
1637 dev_id_priv = container_of(id, struct rdma_id_private, id);
1638
1639 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
1640 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1641 rdma_addr_size(cma_src_addr(id_priv)));
1642
1643 cma_attach_to_dev(dev_id_priv, cma_dev);
1644 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1645 atomic_inc(&id_priv->refcount);
1646 dev_id_priv->internal_id = 1;
1647 dev_id_priv->afonly = id_priv->afonly;
1648
1649 ret = rdma_listen(id, id_priv->backlog);
1650 if (ret)
1651 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1652 "listening on device %s\n", ret, cma_dev->device->name);
1653 }
1654
1655 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1656 {
1657 struct cma_device *cma_dev;
1658
1659 mutex_lock(&lock);
1660 list_add_tail(&id_priv->list, &listen_any_list);
1661 list_for_each_entry(cma_dev, &dev_list, list)
1662 cma_listen_on_dev(id_priv, cma_dev);
1663 mutex_unlock(&lock);
1664 }
1665
1666 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1667 {
1668 struct rdma_id_private *id_priv;
1669
1670 id_priv = container_of(id, struct rdma_id_private, id);
1671 id_priv->tos = (u8) tos;
1672 }
1673 EXPORT_SYMBOL(rdma_set_service_type);
1674
1675 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1676 void *context)
1677 {
1678 struct cma_work *work = context;
1679 struct rdma_route *route;
1680
1681 route = &work->id->id.route;
1682
1683 if (!status) {
1684 route->num_paths = 1;
1685 *route->path_rec = *path_rec;
1686 } else {
1687 work->old_state = RDMA_CM_ROUTE_QUERY;
1688 work->new_state = RDMA_CM_ADDR_RESOLVED;
1689 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1690 work->event.status = status;
1691 }
1692
1693 queue_work(cma_wq, &work->work);
1694 }
1695
1696 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1697 struct cma_work *work)
1698 {
1699 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1700 struct ib_sa_path_rec path_rec;
1701 ib_sa_comp_mask comp_mask;
1702 struct sockaddr_in6 *sin6;
1703 struct sockaddr_ib *sib;
1704
1705 memset(&path_rec, 0, sizeof path_rec);
1706 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1707 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1708 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
1709 path_rec.numb_path = 1;
1710 path_rec.reversible = 1;
1711 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
1712
1713 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1714 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1715 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1716
1717 switch (cma_family(id_priv)) {
1718 case AF_INET:
1719 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1720 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1721 break;
1722 case AF_INET6:
1723 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
1724 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1725 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1726 break;
1727 case AF_IB:
1728 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
1729 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
1730 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1731 break;
1732 }
1733
1734 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1735 id_priv->id.port_num, &path_rec,
1736 comp_mask, timeout_ms,
1737 GFP_KERNEL, cma_query_handler,
1738 work, &id_priv->query);
1739
1740 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1741 }
1742
1743 static void cma_work_handler(struct work_struct *_work)
1744 {
1745 struct cma_work *work = container_of(_work, struct cma_work, work);
1746 struct rdma_id_private *id_priv = work->id;
1747 int destroy = 0;
1748
1749 mutex_lock(&id_priv->handler_mutex);
1750 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1751 goto out;
1752
1753 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1754 cma_exch(id_priv, RDMA_CM_DESTROYING);
1755 destroy = 1;
1756 }
1757 out:
1758 mutex_unlock(&id_priv->handler_mutex);
1759 cma_deref_id(id_priv);
1760 if (destroy)
1761 rdma_destroy_id(&id_priv->id);
1762 kfree(work);
1763 }
1764
1765 static void cma_ndev_work_handler(struct work_struct *_work)
1766 {
1767 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1768 struct rdma_id_private *id_priv = work->id;
1769 int destroy = 0;
1770
1771 mutex_lock(&id_priv->handler_mutex);
1772 if (id_priv->state == RDMA_CM_DESTROYING ||
1773 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
1774 goto out;
1775
1776 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1777 cma_exch(id_priv, RDMA_CM_DESTROYING);
1778 destroy = 1;
1779 }
1780
1781 out:
1782 mutex_unlock(&id_priv->handler_mutex);
1783 cma_deref_id(id_priv);
1784 if (destroy)
1785 rdma_destroy_id(&id_priv->id);
1786 kfree(work);
1787 }
1788
1789 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1790 {
1791 struct rdma_route *route = &id_priv->id.route;
1792 struct cma_work *work;
1793 int ret;
1794
1795 work = kzalloc(sizeof *work, GFP_KERNEL);
1796 if (!work)
1797 return -ENOMEM;
1798
1799 work->id = id_priv;
1800 INIT_WORK(&work->work, cma_work_handler);
1801 work->old_state = RDMA_CM_ROUTE_QUERY;
1802 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1803 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1804
1805 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1806 if (!route->path_rec) {
1807 ret = -ENOMEM;
1808 goto err1;
1809 }
1810
1811 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1812 if (ret)
1813 goto err2;
1814
1815 return 0;
1816 err2:
1817 kfree(route->path_rec);
1818 route->path_rec = NULL;
1819 err1:
1820 kfree(work);
1821 return ret;
1822 }
1823
1824 int rdma_set_ib_paths(struct rdma_cm_id *id,
1825 struct ib_sa_path_rec *path_rec, int num_paths)
1826 {
1827 struct rdma_id_private *id_priv;
1828 int ret;
1829
1830 id_priv = container_of(id, struct rdma_id_private, id);
1831 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1832 RDMA_CM_ROUTE_RESOLVED))
1833 return -EINVAL;
1834
1835 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1836 GFP_KERNEL);
1837 if (!id->route.path_rec) {
1838 ret = -ENOMEM;
1839 goto err;
1840 }
1841
1842 id->route.num_paths = num_paths;
1843 return 0;
1844 err:
1845 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
1846 return ret;
1847 }
1848 EXPORT_SYMBOL(rdma_set_ib_paths);
1849
1850 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1851 {
1852 struct cma_work *work;
1853
1854 work = kzalloc(sizeof *work, GFP_KERNEL);
1855 if (!work)
1856 return -ENOMEM;
1857
1858 work->id = id_priv;
1859 INIT_WORK(&work->work, cma_work_handler);
1860 work->old_state = RDMA_CM_ROUTE_QUERY;
1861 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1862 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1863 queue_work(cma_wq, &work->work);
1864 return 0;
1865 }
1866
1867 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
1868 {
1869 int prio;
1870 struct net_device *dev;
1871
1872 prio = rt_tos2priority(tos);
1873 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
1874 vlan_dev_real_dev(ndev) : ndev;
1875
1876 if (dev->num_tc)
1877 return netdev_get_prio_tc_map(dev, prio);
1878
1879 #if IS_ENABLED(CONFIG_VLAN_8021Q)
1880 if (ndev->priv_flags & IFF_802_1Q_VLAN)
1881 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
1882 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1883 #endif
1884 return 0;
1885 }
1886
1887 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1888 {
1889 struct rdma_route *route = &id_priv->id.route;
1890 struct rdma_addr *addr = &route->addr;
1891 struct cma_work *work;
1892 int ret;
1893 struct net_device *ndev = NULL;
1894
1895
1896 work = kzalloc(sizeof *work, GFP_KERNEL);
1897 if (!work)
1898 return -ENOMEM;
1899
1900 work->id = id_priv;
1901 INIT_WORK(&work->work, cma_work_handler);
1902
1903 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1904 if (!route->path_rec) {
1905 ret = -ENOMEM;
1906 goto err1;
1907 }
1908
1909 route->num_paths = 1;
1910
1911 if (addr->dev_addr.bound_dev_if)
1912 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1913 if (!ndev) {
1914 ret = -ENODEV;
1915 goto err2;
1916 }
1917
1918 route->path_rec->vlan_id = rdma_vlan_dev_vlan_id(ndev);
1919 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
1920 memcpy(route->path_rec->smac, ndev->dev_addr, ndev->addr_len);
1921
1922 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
1923 &route->path_rec->sgid);
1924 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
1925 &route->path_rec->dgid);
1926
1927 route->path_rec->hop_limit = 1;
1928 route->path_rec->reversible = 1;
1929 route->path_rec->pkey = cpu_to_be16(0xffff);
1930 route->path_rec->mtu_selector = IB_SA_EQ;
1931 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
1932 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1933 route->path_rec->rate_selector = IB_SA_EQ;
1934 route->path_rec->rate = iboe_get_rate(ndev);
1935 dev_put(ndev);
1936 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1937 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1938 if (!route->path_rec->mtu) {
1939 ret = -EINVAL;
1940 goto err2;
1941 }
1942
1943 work->old_state = RDMA_CM_ROUTE_QUERY;
1944 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1945 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1946 work->event.status = 0;
1947
1948 queue_work(cma_wq, &work->work);
1949
1950 return 0;
1951
1952 err2:
1953 kfree(route->path_rec);
1954 route->path_rec = NULL;
1955 err1:
1956 kfree(work);
1957 return ret;
1958 }
1959
1960 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1961 {
1962 struct rdma_id_private *id_priv;
1963 int ret;
1964
1965 id_priv = container_of(id, struct rdma_id_private, id);
1966 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
1967 return -EINVAL;
1968
1969 atomic_inc(&id_priv->refcount);
1970 if (rdma_protocol_ib(id->device, id->port_num))
1971 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1972 else if (rdma_protocol_iboe(id->device, id->port_num))
1973 ret = cma_resolve_iboe_route(id_priv);
1974 else if (rdma_protocol_iwarp(id->device, id->port_num))
1975 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1976 else
1977 ret = -ENOSYS;
1978
1979 if (ret)
1980 goto err;
1981
1982 return 0;
1983 err:
1984 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
1985 cma_deref_id(id_priv);
1986 return ret;
1987 }
1988 EXPORT_SYMBOL(rdma_resolve_route);
1989
1990 static void cma_set_loopback(struct sockaddr *addr)
1991 {
1992 switch (addr->sa_family) {
1993 case AF_INET:
1994 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1995 break;
1996 case AF_INET6:
1997 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
1998 0, 0, 0, htonl(1));
1999 break;
2000 default:
2001 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2002 0, 0, 0, htonl(1));
2003 break;
2004 }
2005 }
2006
2007 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2008 {
2009 struct cma_device *cma_dev, *cur_dev;
2010 struct ib_port_attr port_attr;
2011 union ib_gid gid;
2012 u16 pkey;
2013 int ret;
2014 u8 p;
2015
2016 cma_dev = NULL;
2017 mutex_lock(&lock);
2018 list_for_each_entry(cur_dev, &dev_list, list) {
2019 if (cma_family(id_priv) == AF_IB &&
2020 !rdma_ib_or_iboe(cur_dev->device, 1))
2021 continue;
2022
2023 if (!cma_dev)
2024 cma_dev = cur_dev;
2025
2026 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2027 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2028 port_attr.state == IB_PORT_ACTIVE) {
2029 cma_dev = cur_dev;
2030 goto port_found;
2031 }
2032 }
2033 }
2034
2035 if (!cma_dev) {
2036 ret = -ENODEV;
2037 goto out;
2038 }
2039
2040 p = 1;
2041
2042 port_found:
2043 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2044 if (ret)
2045 goto out;
2046
2047 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2048 if (ret)
2049 goto out;
2050
2051 id_priv->id.route.addr.dev_addr.dev_type =
2052 (rdma_protocol_ib(cma_dev->device, p)) ?
2053 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2054
2055 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2056 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2057 id_priv->id.port_num = p;
2058 cma_attach_to_dev(id_priv, cma_dev);
2059 cma_set_loopback(cma_src_addr(id_priv));
2060 out:
2061 mutex_unlock(&lock);
2062 return ret;
2063 }
2064
2065 static void addr_handler(int status, struct sockaddr *src_addr,
2066 struct rdma_dev_addr *dev_addr, void *context)
2067 {
2068 struct rdma_id_private *id_priv = context;
2069 struct rdma_cm_event event;
2070
2071 memset(&event, 0, sizeof event);
2072 mutex_lock(&id_priv->handler_mutex);
2073 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2074 RDMA_CM_ADDR_RESOLVED))
2075 goto out;
2076
2077 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2078 if (!status && !id_priv->cma_dev)
2079 status = cma_acquire_dev(id_priv, NULL);
2080
2081 if (status) {
2082 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2083 RDMA_CM_ADDR_BOUND))
2084 goto out;
2085 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2086 event.status = status;
2087 } else
2088 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2089
2090 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2091 cma_exch(id_priv, RDMA_CM_DESTROYING);
2092 mutex_unlock(&id_priv->handler_mutex);
2093 cma_deref_id(id_priv);
2094 rdma_destroy_id(&id_priv->id);
2095 return;
2096 }
2097 out:
2098 mutex_unlock(&id_priv->handler_mutex);
2099 cma_deref_id(id_priv);
2100 }
2101
2102 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2103 {
2104 struct cma_work *work;
2105 union ib_gid gid;
2106 int ret;
2107
2108 work = kzalloc(sizeof *work, GFP_KERNEL);
2109 if (!work)
2110 return -ENOMEM;
2111
2112 if (!id_priv->cma_dev) {
2113 ret = cma_bind_loopback(id_priv);
2114 if (ret)
2115 goto err;
2116 }
2117
2118 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2119 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2120
2121 work->id = id_priv;
2122 INIT_WORK(&work->work, cma_work_handler);
2123 work->old_state = RDMA_CM_ADDR_QUERY;
2124 work->new_state = RDMA_CM_ADDR_RESOLVED;
2125 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2126 queue_work(cma_wq, &work->work);
2127 return 0;
2128 err:
2129 kfree(work);
2130 return ret;
2131 }
2132
2133 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2134 {
2135 struct cma_work *work;
2136 int ret;
2137
2138 work = kzalloc(sizeof *work, GFP_KERNEL);
2139 if (!work)
2140 return -ENOMEM;
2141
2142 if (!id_priv->cma_dev) {
2143 ret = cma_resolve_ib_dev(id_priv);
2144 if (ret)
2145 goto err;
2146 }
2147
2148 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2149 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2150
2151 work->id = id_priv;
2152 INIT_WORK(&work->work, cma_work_handler);
2153 work->old_state = RDMA_CM_ADDR_QUERY;
2154 work->new_state = RDMA_CM_ADDR_RESOLVED;
2155 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2156 queue_work(cma_wq, &work->work);
2157 return 0;
2158 err:
2159 kfree(work);
2160 return ret;
2161 }
2162
2163 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2164 struct sockaddr *dst_addr)
2165 {
2166 if (!src_addr || !src_addr->sa_family) {
2167 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2168 src_addr->sa_family = dst_addr->sa_family;
2169 if (dst_addr->sa_family == AF_INET6) {
2170 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2171 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2172 } else if (dst_addr->sa_family == AF_IB) {
2173 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2174 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2175 }
2176 }
2177 return rdma_bind_addr(id, src_addr);
2178 }
2179
2180 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2181 struct sockaddr *dst_addr, int timeout_ms)
2182 {
2183 struct rdma_id_private *id_priv;
2184 int ret;
2185
2186 id_priv = container_of(id, struct rdma_id_private, id);
2187 if (id_priv->state == RDMA_CM_IDLE) {
2188 ret = cma_bind_addr(id, src_addr, dst_addr);
2189 if (ret)
2190 return ret;
2191 }
2192
2193 if (cma_family(id_priv) != dst_addr->sa_family)
2194 return -EINVAL;
2195
2196 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2197 return -EINVAL;
2198
2199 atomic_inc(&id_priv->refcount);
2200 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2201 if (cma_any_addr(dst_addr)) {
2202 ret = cma_resolve_loopback(id_priv);
2203 } else {
2204 if (dst_addr->sa_family == AF_IB) {
2205 ret = cma_resolve_ib_addr(id_priv);
2206 } else {
2207 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2208 dst_addr, &id->route.addr.dev_addr,
2209 timeout_ms, addr_handler, id_priv);
2210 }
2211 }
2212 if (ret)
2213 goto err;
2214
2215 return 0;
2216 err:
2217 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2218 cma_deref_id(id_priv);
2219 return ret;
2220 }
2221 EXPORT_SYMBOL(rdma_resolve_addr);
2222
2223 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2224 {
2225 struct rdma_id_private *id_priv;
2226 unsigned long flags;
2227 int ret;
2228
2229 id_priv = container_of(id, struct rdma_id_private, id);
2230 spin_lock_irqsave(&id_priv->lock, flags);
2231 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2232 id_priv->reuseaddr = reuse;
2233 ret = 0;
2234 } else {
2235 ret = -EINVAL;
2236 }
2237 spin_unlock_irqrestore(&id_priv->lock, flags);
2238 return ret;
2239 }
2240 EXPORT_SYMBOL(rdma_set_reuseaddr);
2241
2242 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2243 {
2244 struct rdma_id_private *id_priv;
2245 unsigned long flags;
2246 int ret;
2247
2248 id_priv = container_of(id, struct rdma_id_private, id);
2249 spin_lock_irqsave(&id_priv->lock, flags);
2250 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2251 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2252 id_priv->afonly = afonly;
2253 ret = 0;
2254 } else {
2255 ret = -EINVAL;
2256 }
2257 spin_unlock_irqrestore(&id_priv->lock, flags);
2258 return ret;
2259 }
2260 EXPORT_SYMBOL(rdma_set_afonly);
2261
2262 static void cma_bind_port(struct rdma_bind_list *bind_list,
2263 struct rdma_id_private *id_priv)
2264 {
2265 struct sockaddr *addr;
2266 struct sockaddr_ib *sib;
2267 u64 sid, mask;
2268 __be16 port;
2269
2270 addr = cma_src_addr(id_priv);
2271 port = htons(bind_list->port);
2272
2273 switch (addr->sa_family) {
2274 case AF_INET:
2275 ((struct sockaddr_in *) addr)->sin_port = port;
2276 break;
2277 case AF_INET6:
2278 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2279 break;
2280 case AF_IB:
2281 sib = (struct sockaddr_ib *) addr;
2282 sid = be64_to_cpu(sib->sib_sid);
2283 mask = be64_to_cpu(sib->sib_sid_mask);
2284 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2285 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2286 break;
2287 }
2288 id_priv->bind_list = bind_list;
2289 hlist_add_head(&id_priv->node, &bind_list->owners);
2290 }
2291
2292 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2293 unsigned short snum)
2294 {
2295 struct rdma_bind_list *bind_list;
2296 int ret;
2297
2298 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2299 if (!bind_list)
2300 return -ENOMEM;
2301
2302 ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2303 if (ret < 0)
2304 goto err;
2305
2306 bind_list->ps = ps;
2307 bind_list->port = (unsigned short)ret;
2308 cma_bind_port(bind_list, id_priv);
2309 return 0;
2310 err:
2311 kfree(bind_list);
2312 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2313 }
2314
2315 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2316 {
2317 static unsigned int last_used_port;
2318 int low, high, remaining;
2319 unsigned int rover;
2320
2321 inet_get_local_port_range(&init_net, &low, &high);
2322 remaining = (high - low) + 1;
2323 rover = prandom_u32() % remaining + low;
2324 retry:
2325 if (last_used_port != rover &&
2326 !idr_find(ps, (unsigned short) rover)) {
2327 int ret = cma_alloc_port(ps, id_priv, rover);
2328 /*
2329 * Remember previously used port number in order to avoid
2330 * re-using same port immediately after it is closed.
2331 */
2332 if (!ret)
2333 last_used_port = rover;
2334 if (ret != -EADDRNOTAVAIL)
2335 return ret;
2336 }
2337 if (--remaining) {
2338 rover++;
2339 if ((rover < low) || (rover > high))
2340 rover = low;
2341 goto retry;
2342 }
2343 return -EADDRNOTAVAIL;
2344 }
2345
2346 /*
2347 * Check that the requested port is available. This is called when trying to
2348 * bind to a specific port, or when trying to listen on a bound port. In
2349 * the latter case, the provided id_priv may already be on the bind_list, but
2350 * we still need to check that it's okay to start listening.
2351 */
2352 static int cma_check_port(struct rdma_bind_list *bind_list,
2353 struct rdma_id_private *id_priv, uint8_t reuseaddr)
2354 {
2355 struct rdma_id_private *cur_id;
2356 struct sockaddr *addr, *cur_addr;
2357
2358 addr = cma_src_addr(id_priv);
2359 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2360 if (id_priv == cur_id)
2361 continue;
2362
2363 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2364 cur_id->reuseaddr)
2365 continue;
2366
2367 cur_addr = cma_src_addr(cur_id);
2368 if (id_priv->afonly && cur_id->afonly &&
2369 (addr->sa_family != cur_addr->sa_family))
2370 continue;
2371
2372 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2373 return -EADDRNOTAVAIL;
2374
2375 if (!cma_addr_cmp(addr, cur_addr))
2376 return -EADDRINUSE;
2377 }
2378 return 0;
2379 }
2380
2381 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2382 {
2383 struct rdma_bind_list *bind_list;
2384 unsigned short snum;
2385 int ret;
2386
2387 snum = ntohs(cma_port(cma_src_addr(id_priv)));
2388 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2389 return -EACCES;
2390
2391 bind_list = idr_find(ps, snum);
2392 if (!bind_list) {
2393 ret = cma_alloc_port(ps, id_priv, snum);
2394 } else {
2395 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2396 if (!ret)
2397 cma_bind_port(bind_list, id_priv);
2398 }
2399 return ret;
2400 }
2401
2402 static int cma_bind_listen(struct rdma_id_private *id_priv)
2403 {
2404 struct rdma_bind_list *bind_list = id_priv->bind_list;
2405 int ret = 0;
2406
2407 mutex_lock(&lock);
2408 if (bind_list->owners.first->next)
2409 ret = cma_check_port(bind_list, id_priv, 0);
2410 mutex_unlock(&lock);
2411 return ret;
2412 }
2413
2414 static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
2415 {
2416 switch (id_priv->id.ps) {
2417 case RDMA_PS_TCP:
2418 return &tcp_ps;
2419 case RDMA_PS_UDP:
2420 return &udp_ps;
2421 case RDMA_PS_IPOIB:
2422 return &ipoib_ps;
2423 case RDMA_PS_IB:
2424 return &ib_ps;
2425 default:
2426 return NULL;
2427 }
2428 }
2429
2430 static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2431 {
2432 struct idr *ps = NULL;
2433 struct sockaddr_ib *sib;
2434 u64 sid_ps, mask, sid;
2435
2436 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2437 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2438 sid = be64_to_cpu(sib->sib_sid) & mask;
2439
2440 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2441 sid_ps = RDMA_IB_IP_PS_IB;
2442 ps = &ib_ps;
2443 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2444 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2445 sid_ps = RDMA_IB_IP_PS_TCP;
2446 ps = &tcp_ps;
2447 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2448 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2449 sid_ps = RDMA_IB_IP_PS_UDP;
2450 ps = &udp_ps;
2451 }
2452
2453 if (ps) {
2454 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2455 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2456 be64_to_cpu(sib->sib_sid_mask));
2457 }
2458 return ps;
2459 }
2460
2461 static int cma_get_port(struct rdma_id_private *id_priv)
2462 {
2463 struct idr *ps;
2464 int ret;
2465
2466 if (cma_family(id_priv) != AF_IB)
2467 ps = cma_select_inet_ps(id_priv);
2468 else
2469 ps = cma_select_ib_ps(id_priv);
2470 if (!ps)
2471 return -EPROTONOSUPPORT;
2472
2473 mutex_lock(&lock);
2474 if (cma_any_port(cma_src_addr(id_priv)))
2475 ret = cma_alloc_any_port(ps, id_priv);
2476 else
2477 ret = cma_use_port(ps, id_priv);
2478 mutex_unlock(&lock);
2479
2480 return ret;
2481 }
2482
2483 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2484 struct sockaddr *addr)
2485 {
2486 #if IS_ENABLED(CONFIG_IPV6)
2487 struct sockaddr_in6 *sin6;
2488
2489 if (addr->sa_family != AF_INET6)
2490 return 0;
2491
2492 sin6 = (struct sockaddr_in6 *) addr;
2493
2494 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2495 return 0;
2496
2497 if (!sin6->sin6_scope_id)
2498 return -EINVAL;
2499
2500 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2501 #endif
2502 return 0;
2503 }
2504
2505 int rdma_listen(struct rdma_cm_id *id, int backlog)
2506 {
2507 struct rdma_id_private *id_priv;
2508 int ret;
2509
2510 id_priv = container_of(id, struct rdma_id_private, id);
2511 if (id_priv->state == RDMA_CM_IDLE) {
2512 id->route.addr.src_addr.ss_family = AF_INET;
2513 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2514 if (ret)
2515 return ret;
2516 }
2517
2518 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2519 return -EINVAL;
2520
2521 if (id_priv->reuseaddr) {
2522 ret = cma_bind_listen(id_priv);
2523 if (ret)
2524 goto err;
2525 }
2526
2527 id_priv->backlog = backlog;
2528 if (id->device) {
2529 if (rdma_ib_or_iboe(id->device, 1)) {
2530 ret = cma_ib_listen(id_priv);
2531 if (ret)
2532 goto err;
2533 } else if (rdma_protocol_iwarp(id->device, 1)) {
2534 ret = cma_iw_listen(id_priv, backlog);
2535 if (ret)
2536 goto err;
2537 } else {
2538 ret = -ENOSYS;
2539 goto err;
2540 }
2541 } else
2542 cma_listen_on_all(id_priv);
2543
2544 return 0;
2545 err:
2546 id_priv->backlog = 0;
2547 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2548 return ret;
2549 }
2550 EXPORT_SYMBOL(rdma_listen);
2551
2552 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2553 {
2554 struct rdma_id_private *id_priv;
2555 int ret;
2556
2557 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2558 addr->sa_family != AF_IB)
2559 return -EAFNOSUPPORT;
2560
2561 id_priv = container_of(id, struct rdma_id_private, id);
2562 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2563 return -EINVAL;
2564
2565 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2566 if (ret)
2567 goto err1;
2568
2569 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2570 if (!cma_any_addr(addr)) {
2571 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2572 if (ret)
2573 goto err1;
2574
2575 ret = cma_acquire_dev(id_priv, NULL);
2576 if (ret)
2577 goto err1;
2578 }
2579
2580 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2581 if (addr->sa_family == AF_INET)
2582 id_priv->afonly = 1;
2583 #if IS_ENABLED(CONFIG_IPV6)
2584 else if (addr->sa_family == AF_INET6)
2585 id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2586 #endif
2587 }
2588 ret = cma_get_port(id_priv);
2589 if (ret)
2590 goto err2;
2591
2592 return 0;
2593 err2:
2594 if (id_priv->cma_dev)
2595 cma_release_dev(id_priv);
2596 err1:
2597 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2598 return ret;
2599 }
2600 EXPORT_SYMBOL(rdma_bind_addr);
2601
2602 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2603 {
2604 struct cma_hdr *cma_hdr;
2605
2606 cma_hdr = hdr;
2607 cma_hdr->cma_version = CMA_VERSION;
2608 if (cma_family(id_priv) == AF_INET) {
2609 struct sockaddr_in *src4, *dst4;
2610
2611 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2612 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2613
2614 cma_set_ip_ver(cma_hdr, 4);
2615 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2616 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2617 cma_hdr->port = src4->sin_port;
2618 } else if (cma_family(id_priv) == AF_INET6) {
2619 struct sockaddr_in6 *src6, *dst6;
2620
2621 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2622 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2623
2624 cma_set_ip_ver(cma_hdr, 6);
2625 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2626 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2627 cma_hdr->port = src6->sin6_port;
2628 }
2629 return 0;
2630 }
2631
2632 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2633 struct ib_cm_event *ib_event)
2634 {
2635 struct rdma_id_private *id_priv = cm_id->context;
2636 struct rdma_cm_event event;
2637 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2638 int ret = 0;
2639
2640 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
2641 return 0;
2642
2643 memset(&event, 0, sizeof event);
2644 switch (ib_event->event) {
2645 case IB_CM_SIDR_REQ_ERROR:
2646 event.event = RDMA_CM_EVENT_UNREACHABLE;
2647 event.status = -ETIMEDOUT;
2648 break;
2649 case IB_CM_SIDR_REP_RECEIVED:
2650 event.param.ud.private_data = ib_event->private_data;
2651 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2652 if (rep->status != IB_SIDR_SUCCESS) {
2653 event.event = RDMA_CM_EVENT_UNREACHABLE;
2654 event.status = ib_event->param.sidr_rep_rcvd.status;
2655 break;
2656 }
2657 ret = cma_set_qkey(id_priv, rep->qkey);
2658 if (ret) {
2659 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2660 event.status = ret;
2661 break;
2662 }
2663 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2664 id_priv->id.route.path_rec,
2665 &event.param.ud.ah_attr);
2666 event.param.ud.qp_num = rep->qpn;
2667 event.param.ud.qkey = rep->qkey;
2668 event.event = RDMA_CM_EVENT_ESTABLISHED;
2669 event.status = 0;
2670 break;
2671 default:
2672 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2673 ib_event->event);
2674 goto out;
2675 }
2676
2677 ret = id_priv->id.event_handler(&id_priv->id, &event);
2678 if (ret) {
2679 /* Destroy the CM ID by returning a non-zero value. */
2680 id_priv->cm_id.ib = NULL;
2681 cma_exch(id_priv, RDMA_CM_DESTROYING);
2682 mutex_unlock(&id_priv->handler_mutex);
2683 rdma_destroy_id(&id_priv->id);
2684 return ret;
2685 }
2686 out:
2687 mutex_unlock(&id_priv->handler_mutex);
2688 return ret;
2689 }
2690
2691 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2692 struct rdma_conn_param *conn_param)
2693 {
2694 struct ib_cm_sidr_req_param req;
2695 struct ib_cm_id *id;
2696 void *private_data;
2697 int offset, ret;
2698
2699 memset(&req, 0, sizeof req);
2700 offset = cma_user_data_offset(id_priv);
2701 req.private_data_len = offset + conn_param->private_data_len;
2702 if (req.private_data_len < conn_param->private_data_len)
2703 return -EINVAL;
2704
2705 if (req.private_data_len) {
2706 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2707 if (!private_data)
2708 return -ENOMEM;
2709 } else {
2710 private_data = NULL;
2711 }
2712
2713 if (conn_param->private_data && conn_param->private_data_len)
2714 memcpy(private_data + offset, conn_param->private_data,
2715 conn_param->private_data_len);
2716
2717 if (private_data) {
2718 ret = cma_format_hdr(private_data, id_priv);
2719 if (ret)
2720 goto out;
2721 req.private_data = private_data;
2722 }
2723
2724 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2725 id_priv);
2726 if (IS_ERR(id)) {
2727 ret = PTR_ERR(id);
2728 goto out;
2729 }
2730 id_priv->cm_id.ib = id;
2731
2732 req.path = id_priv->id.route.path_rec;
2733 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2734 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2735 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2736
2737 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2738 if (ret) {
2739 ib_destroy_cm_id(id_priv->cm_id.ib);
2740 id_priv->cm_id.ib = NULL;
2741 }
2742 out:
2743 kfree(private_data);
2744 return ret;
2745 }
2746
2747 static int cma_connect_ib(struct rdma_id_private *id_priv,
2748 struct rdma_conn_param *conn_param)
2749 {
2750 struct ib_cm_req_param req;
2751 struct rdma_route *route;
2752 void *private_data;
2753 struct ib_cm_id *id;
2754 int offset, ret;
2755
2756 memset(&req, 0, sizeof req);
2757 offset = cma_user_data_offset(id_priv);
2758 req.private_data_len = offset + conn_param->private_data_len;
2759 if (req.private_data_len < conn_param->private_data_len)
2760 return -EINVAL;
2761
2762 if (req.private_data_len) {
2763 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2764 if (!private_data)
2765 return -ENOMEM;
2766 } else {
2767 private_data = NULL;
2768 }
2769
2770 if (conn_param->private_data && conn_param->private_data_len)
2771 memcpy(private_data + offset, conn_param->private_data,
2772 conn_param->private_data_len);
2773
2774 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2775 if (IS_ERR(id)) {
2776 ret = PTR_ERR(id);
2777 goto out;
2778 }
2779 id_priv->cm_id.ib = id;
2780
2781 route = &id_priv->id.route;
2782 if (private_data) {
2783 ret = cma_format_hdr(private_data, id_priv);
2784 if (ret)
2785 goto out;
2786 req.private_data = private_data;
2787 }
2788
2789 req.primary_path = &route->path_rec[0];
2790 if (route->num_paths == 2)
2791 req.alternate_path = &route->path_rec[1];
2792
2793 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2794 req.qp_num = id_priv->qp_num;
2795 req.qp_type = id_priv->id.qp_type;
2796 req.starting_psn = id_priv->seq_num;
2797 req.responder_resources = conn_param->responder_resources;
2798 req.initiator_depth = conn_param->initiator_depth;
2799 req.flow_control = conn_param->flow_control;
2800 req.retry_count = min_t(u8, 7, conn_param->retry_count);
2801 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2802 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2803 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2804 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2805 req.srq = id_priv->srq ? 1 : 0;
2806
2807 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2808 out:
2809 if (ret && !IS_ERR(id)) {
2810 ib_destroy_cm_id(id);
2811 id_priv->cm_id.ib = NULL;
2812 }
2813
2814 kfree(private_data);
2815 return ret;
2816 }
2817
2818 static int cma_connect_iw(struct rdma_id_private *id_priv,
2819 struct rdma_conn_param *conn_param)
2820 {
2821 struct iw_cm_id *cm_id;
2822 int ret;
2823 struct iw_cm_conn_param iw_param;
2824
2825 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2826 if (IS_ERR(cm_id))
2827 return PTR_ERR(cm_id);
2828
2829 id_priv->cm_id.iw = cm_id;
2830
2831 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
2832 rdma_addr_size(cma_src_addr(id_priv)));
2833 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
2834 rdma_addr_size(cma_dst_addr(id_priv)));
2835
2836 ret = cma_modify_qp_rtr(id_priv, conn_param);
2837 if (ret)
2838 goto out;
2839
2840 if (conn_param) {
2841 iw_param.ord = conn_param->initiator_depth;
2842 iw_param.ird = conn_param->responder_resources;
2843 iw_param.private_data = conn_param->private_data;
2844 iw_param.private_data_len = conn_param->private_data_len;
2845 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2846 } else {
2847 memset(&iw_param, 0, sizeof iw_param);
2848 iw_param.qpn = id_priv->qp_num;
2849 }
2850 ret = iw_cm_connect(cm_id, &iw_param);
2851 out:
2852 if (ret) {
2853 iw_destroy_cm_id(cm_id);
2854 id_priv->cm_id.iw = NULL;
2855 }
2856 return ret;
2857 }
2858
2859 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2860 {
2861 struct rdma_id_private *id_priv;
2862 int ret;
2863
2864 id_priv = container_of(id, struct rdma_id_private, id);
2865 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
2866 return -EINVAL;
2867
2868 if (!id->qp) {
2869 id_priv->qp_num = conn_param->qp_num;
2870 id_priv->srq = conn_param->srq;
2871 }
2872
2873 if (rdma_ib_or_iboe(id->device, id->port_num)) {
2874 if (id->qp_type == IB_QPT_UD)
2875 ret = cma_resolve_ib_udp(id_priv, conn_param);
2876 else
2877 ret = cma_connect_ib(id_priv, conn_param);
2878 } else if (rdma_protocol_iwarp(id->device, id->port_num))
2879 ret = cma_connect_iw(id_priv, conn_param);
2880 else
2881 ret = -ENOSYS;
2882 if (ret)
2883 goto err;
2884
2885 return 0;
2886 err:
2887 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
2888 return ret;
2889 }
2890 EXPORT_SYMBOL(rdma_connect);
2891
2892 static int cma_accept_ib(struct rdma_id_private *id_priv,
2893 struct rdma_conn_param *conn_param)
2894 {
2895 struct ib_cm_rep_param rep;
2896 int ret;
2897
2898 ret = cma_modify_qp_rtr(id_priv, conn_param);
2899 if (ret)
2900 goto out;
2901
2902 ret = cma_modify_qp_rts(id_priv, conn_param);
2903 if (ret)
2904 goto out;
2905
2906 memset(&rep, 0, sizeof rep);
2907 rep.qp_num = id_priv->qp_num;
2908 rep.starting_psn = id_priv->seq_num;
2909 rep.private_data = conn_param->private_data;
2910 rep.private_data_len = conn_param->private_data_len;
2911 rep.responder_resources = conn_param->responder_resources;
2912 rep.initiator_depth = conn_param->initiator_depth;
2913 rep.failover_accepted = 0;
2914 rep.flow_control = conn_param->flow_control;
2915 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2916 rep.srq = id_priv->srq ? 1 : 0;
2917
2918 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2919 out:
2920 return ret;
2921 }
2922
2923 static int cma_accept_iw(struct rdma_id_private *id_priv,
2924 struct rdma_conn_param *conn_param)
2925 {
2926 struct iw_cm_conn_param iw_param;
2927 int ret;
2928
2929 ret = cma_modify_qp_rtr(id_priv, conn_param);
2930 if (ret)
2931 return ret;
2932
2933 iw_param.ord = conn_param->initiator_depth;
2934 iw_param.ird = conn_param->responder_resources;
2935 iw_param.private_data = conn_param->private_data;
2936 iw_param.private_data_len = conn_param->private_data_len;
2937 if (id_priv->id.qp) {
2938 iw_param.qpn = id_priv->qp_num;
2939 } else
2940 iw_param.qpn = conn_param->qp_num;
2941
2942 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2943 }
2944
2945 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2946 enum ib_cm_sidr_status status, u32 qkey,
2947 const void *private_data, int private_data_len)
2948 {
2949 struct ib_cm_sidr_rep_param rep;
2950 int ret;
2951
2952 memset(&rep, 0, sizeof rep);
2953 rep.status = status;
2954 if (status == IB_SIDR_SUCCESS) {
2955 ret = cma_set_qkey(id_priv, qkey);
2956 if (ret)
2957 return ret;
2958 rep.qp_num = id_priv->qp_num;
2959 rep.qkey = id_priv->qkey;
2960 }
2961 rep.private_data = private_data;
2962 rep.private_data_len = private_data_len;
2963
2964 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2965 }
2966
2967 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2968 {
2969 struct rdma_id_private *id_priv;
2970 int ret;
2971
2972 id_priv = container_of(id, struct rdma_id_private, id);
2973
2974 id_priv->owner = task_pid_nr(current);
2975
2976 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
2977 return -EINVAL;
2978
2979 if (!id->qp && conn_param) {
2980 id_priv->qp_num = conn_param->qp_num;
2981 id_priv->srq = conn_param->srq;
2982 }
2983
2984 if (rdma_ib_or_iboe(id->device, id->port_num)) {
2985 if (id->qp_type == IB_QPT_UD) {
2986 if (conn_param)
2987 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2988 conn_param->qkey,
2989 conn_param->private_data,
2990 conn_param->private_data_len);
2991 else
2992 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2993 0, NULL, 0);
2994 } else {
2995 if (conn_param)
2996 ret = cma_accept_ib(id_priv, conn_param);
2997 else
2998 ret = cma_rep_recv(id_priv);
2999 }
3000 } else if (rdma_protocol_iwarp(id->device, id->port_num))
3001 ret = cma_accept_iw(id_priv, conn_param);
3002 else
3003 ret = -ENOSYS;
3004
3005 if (ret)
3006 goto reject;
3007
3008 return 0;
3009 reject:
3010 cma_modify_qp_err(id_priv);
3011 rdma_reject(id, NULL, 0);
3012 return ret;
3013 }
3014 EXPORT_SYMBOL(rdma_accept);
3015
3016 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3017 {
3018 struct rdma_id_private *id_priv;
3019 int ret;
3020
3021 id_priv = container_of(id, struct rdma_id_private, id);
3022 if (!id_priv->cm_id.ib)
3023 return -EINVAL;
3024
3025 switch (id->device->node_type) {
3026 case RDMA_NODE_IB_CA:
3027 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3028 break;
3029 default:
3030 ret = 0;
3031 break;
3032 }
3033 return ret;
3034 }
3035 EXPORT_SYMBOL(rdma_notify);
3036
3037 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3038 u8 private_data_len)
3039 {
3040 struct rdma_id_private *id_priv;
3041 int ret;
3042
3043 id_priv = container_of(id, struct rdma_id_private, id);
3044 if (!id_priv->cm_id.ib)
3045 return -EINVAL;
3046
3047 if (rdma_ib_or_iboe(id->device, id->port_num)) {
3048 if (id->qp_type == IB_QPT_UD)
3049 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3050 private_data, private_data_len);
3051 else
3052 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3053 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3054 0, private_data, private_data_len);
3055 } else if (rdma_protocol_iwarp(id->device, id->port_num)) {
3056 ret = iw_cm_reject(id_priv->cm_id.iw,
3057 private_data, private_data_len);
3058 } else
3059 ret = -ENOSYS;
3060
3061 return ret;
3062 }
3063 EXPORT_SYMBOL(rdma_reject);
3064
3065 int rdma_disconnect(struct rdma_cm_id *id)
3066 {
3067 struct rdma_id_private *id_priv;
3068 int ret;
3069
3070 id_priv = container_of(id, struct rdma_id_private, id);
3071 if (!id_priv->cm_id.ib)
3072 return -EINVAL;
3073
3074 if (rdma_ib_or_iboe(id->device, id->port_num)) {
3075 ret = cma_modify_qp_err(id_priv);
3076 if (ret)
3077 goto out;
3078 /* Initiate or respond to a disconnect. */
3079 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3080 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3081 } else if (rdma_protocol_iwarp(id->device, id->port_num)) {
3082 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3083 } else
3084 ret = -EINVAL;
3085
3086 out:
3087 return ret;
3088 }
3089 EXPORT_SYMBOL(rdma_disconnect);
3090
3091 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3092 {
3093 struct rdma_id_private *id_priv;
3094 struct cma_multicast *mc = multicast->context;
3095 struct rdma_cm_event event;
3096 int ret;
3097
3098 id_priv = mc->id_priv;
3099 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3100 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3101 return 0;
3102
3103 if (!status)
3104 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3105 mutex_lock(&id_priv->qp_mutex);
3106 if (!status && id_priv->id.qp)
3107 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3108 be16_to_cpu(multicast->rec.mlid));
3109 mutex_unlock(&id_priv->qp_mutex);
3110
3111 memset(&event, 0, sizeof event);
3112 event.status = status;
3113 event.param.ud.private_data = mc->context;
3114 if (!status) {
3115 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3116 ib_init_ah_from_mcmember(id_priv->id.device,
3117 id_priv->id.port_num, &multicast->rec,
3118 &event.param.ud.ah_attr);
3119 event.param.ud.qp_num = 0xFFFFFF;
3120 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3121 } else
3122 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3123
3124 ret = id_priv->id.event_handler(&id_priv->id, &event);
3125 if (ret) {
3126 cma_exch(id_priv, RDMA_CM_DESTROYING);
3127 mutex_unlock(&id_priv->handler_mutex);
3128 rdma_destroy_id(&id_priv->id);
3129 return 0;
3130 }
3131
3132 mutex_unlock(&id_priv->handler_mutex);
3133 return 0;
3134 }
3135
3136 static void cma_set_mgid(struct rdma_id_private *id_priv,
3137 struct sockaddr *addr, union ib_gid *mgid)
3138 {
3139 unsigned char mc_map[MAX_ADDR_LEN];
3140 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3141 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3142 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3143
3144 if (cma_any_addr(addr)) {
3145 memset(mgid, 0, sizeof *mgid);
3146 } else if ((addr->sa_family == AF_INET6) &&
3147 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3148 0xFF10A01B)) {
3149 /* IPv6 address is an SA assigned MGID. */
3150 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3151 } else if (addr->sa_family == AF_IB) {
3152 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3153 } else if ((addr->sa_family == AF_INET6)) {
3154 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3155 if (id_priv->id.ps == RDMA_PS_UDP)
3156 mc_map[7] = 0x01; /* Use RDMA CM signature */
3157 *mgid = *(union ib_gid *) (mc_map + 4);
3158 } else {
3159 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3160 if (id_priv->id.ps == RDMA_PS_UDP)
3161 mc_map[7] = 0x01; /* Use RDMA CM signature */
3162 *mgid = *(union ib_gid *) (mc_map + 4);
3163 }
3164 }
3165
3166 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3167 struct cma_multicast *mc)
3168 {
3169 struct ib_sa_mcmember_rec rec;
3170 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3171 ib_sa_comp_mask comp_mask;
3172 int ret;
3173
3174 ib_addr_get_mgid(dev_addr, &rec.mgid);
3175 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3176 &rec.mgid, &rec);
3177 if (ret)
3178 return ret;
3179
3180 ret = cma_set_qkey(id_priv, 0);
3181 if (ret)
3182 return ret;
3183
3184 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3185 rec.qkey = cpu_to_be32(id_priv->qkey);
3186 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3187 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3188 rec.join_state = 1;
3189
3190 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3191 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3192 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3193 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3194 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3195
3196 if (id_priv->id.ps == RDMA_PS_IPOIB)
3197 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3198 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3199 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3200 IB_SA_MCMEMBER_REC_MTU |
3201 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3202
3203 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3204 id_priv->id.port_num, &rec,
3205 comp_mask, GFP_KERNEL,
3206 cma_ib_mc_handler, mc);
3207 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3208 }
3209
3210 static void iboe_mcast_work_handler(struct work_struct *work)
3211 {
3212 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3213 struct cma_multicast *mc = mw->mc;
3214 struct ib_sa_multicast *m = mc->multicast.ib;
3215
3216 mc->multicast.ib->context = mc;
3217 cma_ib_mc_handler(0, m);
3218 kref_put(&mc->mcref, release_mc);
3219 kfree(mw);
3220 }
3221
3222 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3223 {
3224 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3225 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3226
3227 if (cma_any_addr(addr)) {
3228 memset(mgid, 0, sizeof *mgid);
3229 } else if (addr->sa_family == AF_INET6) {
3230 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3231 } else {
3232 mgid->raw[0] = 0xff;
3233 mgid->raw[1] = 0x0e;
3234 mgid->raw[2] = 0;
3235 mgid->raw[3] = 0;
3236 mgid->raw[4] = 0;
3237 mgid->raw[5] = 0;
3238 mgid->raw[6] = 0;
3239 mgid->raw[7] = 0;
3240 mgid->raw[8] = 0;
3241 mgid->raw[9] = 0;
3242 mgid->raw[10] = 0xff;
3243 mgid->raw[11] = 0xff;
3244 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3245 }
3246 }
3247
3248 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3249 struct cma_multicast *mc)
3250 {
3251 struct iboe_mcast_work *work;
3252 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3253 int err;
3254 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3255 struct net_device *ndev = NULL;
3256
3257 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3258 return -EINVAL;
3259
3260 work = kzalloc(sizeof *work, GFP_KERNEL);
3261 if (!work)
3262 return -ENOMEM;
3263
3264 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3265 if (!mc->multicast.ib) {
3266 err = -ENOMEM;
3267 goto out1;
3268 }
3269
3270 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3271
3272 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3273 if (id_priv->id.ps == RDMA_PS_UDP)
3274 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3275
3276 if (dev_addr->bound_dev_if)
3277 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3278 if (!ndev) {
3279 err = -ENODEV;
3280 goto out2;
3281 }
3282 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3283 mc->multicast.ib->rec.hop_limit = 1;
3284 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3285 dev_put(ndev);
3286 if (!mc->multicast.ib->rec.mtu) {
3287 err = -EINVAL;
3288 goto out2;
3289 }
3290 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3291 &mc->multicast.ib->rec.port_gid);
3292 work->id = id_priv;
3293 work->mc = mc;
3294 INIT_WORK(&work->work, iboe_mcast_work_handler);
3295 kref_get(&mc->mcref);
3296 queue_work(cma_wq, &work->work);
3297
3298 return 0;
3299
3300 out2:
3301 kfree(mc->multicast.ib);
3302 out1:
3303 kfree(work);
3304 return err;
3305 }
3306
3307 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3308 void *context)
3309 {
3310 struct rdma_id_private *id_priv;
3311 struct cma_multicast *mc;
3312 int ret;
3313
3314 id_priv = container_of(id, struct rdma_id_private, id);
3315 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3316 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3317 return -EINVAL;
3318
3319 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3320 if (!mc)
3321 return -ENOMEM;
3322
3323 memcpy(&mc->addr, addr, rdma_addr_size(addr));
3324 mc->context = context;
3325 mc->id_priv = id_priv;
3326
3327 spin_lock(&id_priv->lock);
3328 list_add(&mc->list, &id_priv->mc_list);
3329 spin_unlock(&id_priv->lock);
3330
3331 if (rdma_protocol_iboe(id->device, id->port_num)) {
3332 kref_init(&mc->mcref);
3333 ret = cma_iboe_join_multicast(id_priv, mc);
3334 } else if (rdma_protocol_ib(id->device, id->port_num))
3335 ret = cma_join_ib_multicast(id_priv, mc);
3336 else
3337 ret = -ENOSYS;
3338
3339 if (ret) {
3340 spin_lock_irq(&id_priv->lock);
3341 list_del(&mc->list);
3342 spin_unlock_irq(&id_priv->lock);
3343 kfree(mc);
3344 }
3345 return ret;
3346 }
3347 EXPORT_SYMBOL(rdma_join_multicast);
3348
3349 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3350 {
3351 struct rdma_id_private *id_priv;
3352 struct cma_multicast *mc;
3353
3354 id_priv = container_of(id, struct rdma_id_private, id);
3355 spin_lock_irq(&id_priv->lock);
3356 list_for_each_entry(mc, &id_priv->mc_list, list) {
3357 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3358 list_del(&mc->list);
3359 spin_unlock_irq(&id_priv->lock);
3360
3361 if (id->qp)
3362 ib_detach_mcast(id->qp,
3363 &mc->multicast.ib->rec.mgid,
3364 be16_to_cpu(mc->multicast.ib->rec.mlid));
3365
3366 BUG_ON(id_priv->cma_dev->device != id->device);
3367
3368 if (rdma_protocol_ib(id->device, id->port_num)) {
3369 ib_sa_free_multicast(mc->multicast.ib);
3370 kfree(mc);
3371 } else if (rdma_protocol_iboe(id->device, id->port_num))
3372 kref_put(&mc->mcref, release_mc);
3373
3374 return;
3375 }
3376 }
3377 spin_unlock_irq(&id_priv->lock);
3378 }
3379 EXPORT_SYMBOL(rdma_leave_multicast);
3380
3381 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3382 {
3383 struct rdma_dev_addr *dev_addr;
3384 struct cma_ndev_work *work;
3385
3386 dev_addr = &id_priv->id.route.addr.dev_addr;
3387
3388 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3389 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3390 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3391 ndev->name, &id_priv->id);
3392 work = kzalloc(sizeof *work, GFP_KERNEL);
3393 if (!work)
3394 return -ENOMEM;
3395
3396 INIT_WORK(&work->work, cma_ndev_work_handler);
3397 work->id = id_priv;
3398 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3399 atomic_inc(&id_priv->refcount);
3400 queue_work(cma_wq, &work->work);
3401 }
3402
3403 return 0;
3404 }
3405
3406 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3407 void *ptr)
3408 {
3409 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3410 struct cma_device *cma_dev;
3411 struct rdma_id_private *id_priv;
3412 int ret = NOTIFY_DONE;
3413
3414 if (dev_net(ndev) != &init_net)
3415 return NOTIFY_DONE;
3416
3417 if (event != NETDEV_BONDING_FAILOVER)
3418 return NOTIFY_DONE;
3419
3420 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3421 return NOTIFY_DONE;
3422
3423 mutex_lock(&lock);
3424 list_for_each_entry(cma_dev, &dev_list, list)
3425 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3426 ret = cma_netdev_change(ndev, id_priv);
3427 if (ret)
3428 goto out;
3429 }
3430
3431 out:
3432 mutex_unlock(&lock);
3433 return ret;
3434 }
3435
3436 static struct notifier_block cma_nb = {
3437 .notifier_call = cma_netdev_callback
3438 };
3439
3440 static void cma_add_one(struct ib_device *device)
3441 {
3442 struct cma_device *cma_dev;
3443 struct rdma_id_private *id_priv;
3444
3445 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3446 if (!cma_dev)
3447 return;
3448
3449 cma_dev->device = device;
3450
3451 init_completion(&cma_dev->comp);
3452 atomic_set(&cma_dev->refcount, 1);
3453 INIT_LIST_HEAD(&cma_dev->id_list);
3454 ib_set_client_data(device, &cma_client, cma_dev);
3455
3456 mutex_lock(&lock);
3457 list_add_tail(&cma_dev->list, &dev_list);
3458 list_for_each_entry(id_priv, &listen_any_list, list)
3459 cma_listen_on_dev(id_priv, cma_dev);
3460 mutex_unlock(&lock);
3461 }
3462
3463 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3464 {
3465 struct rdma_cm_event event;
3466 enum rdma_cm_state state;
3467 int ret = 0;
3468
3469 /* Record that we want to remove the device */
3470 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3471 if (state == RDMA_CM_DESTROYING)
3472 return 0;
3473
3474 cma_cancel_operation(id_priv, state);
3475 mutex_lock(&id_priv->handler_mutex);
3476
3477 /* Check for destruction from another callback. */
3478 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3479 goto out;
3480
3481 memset(&event, 0, sizeof event);
3482 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3483 ret = id_priv->id.event_handler(&id_priv->id, &event);
3484 out:
3485 mutex_unlock(&id_priv->handler_mutex);
3486 return ret;
3487 }
3488
3489 static void cma_process_remove(struct cma_device *cma_dev)
3490 {
3491 struct rdma_id_private *id_priv;
3492 int ret;
3493
3494 mutex_lock(&lock);
3495 while (!list_empty(&cma_dev->id_list)) {
3496 id_priv = list_entry(cma_dev->id_list.next,
3497 struct rdma_id_private, list);
3498
3499 list_del(&id_priv->listen_list);
3500 list_del_init(&id_priv->list);
3501 atomic_inc(&id_priv->refcount);
3502 mutex_unlock(&lock);
3503
3504 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3505 cma_deref_id(id_priv);
3506 if (ret)
3507 rdma_destroy_id(&id_priv->id);
3508
3509 mutex_lock(&lock);
3510 }
3511 mutex_unlock(&lock);
3512
3513 cma_deref_dev(cma_dev);
3514 wait_for_completion(&cma_dev->comp);
3515 }
3516
3517 static void cma_remove_one(struct ib_device *device)
3518 {
3519 struct cma_device *cma_dev;
3520
3521 cma_dev = ib_get_client_data(device, &cma_client);
3522 if (!cma_dev)
3523 return;
3524
3525 mutex_lock(&lock);
3526 list_del(&cma_dev->list);
3527 mutex_unlock(&lock);
3528
3529 cma_process_remove(cma_dev);
3530 kfree(cma_dev);
3531 }
3532
3533 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3534 {
3535 struct nlmsghdr *nlh;
3536 struct rdma_cm_id_stats *id_stats;
3537 struct rdma_id_private *id_priv;
3538 struct rdma_cm_id *id = NULL;
3539 struct cma_device *cma_dev;
3540 int i_dev = 0, i_id = 0;
3541
3542 /*
3543 * We export all of the IDs as a sequence of messages. Each
3544 * ID gets its own netlink message.
3545 */
3546 mutex_lock(&lock);
3547
3548 list_for_each_entry(cma_dev, &dev_list, list) {
3549 if (i_dev < cb->args[0]) {
3550 i_dev++;
3551 continue;
3552 }
3553
3554 i_id = 0;
3555 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3556 if (i_id < cb->args[1]) {
3557 i_id++;
3558 continue;
3559 }
3560
3561 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3562 sizeof *id_stats, RDMA_NL_RDMA_CM,
3563 RDMA_NL_RDMA_CM_ID_STATS,
3564 NLM_F_MULTI);
3565 if (!id_stats)
3566 goto out;
3567
3568 memset(id_stats, 0, sizeof *id_stats);
3569 id = &id_priv->id;
3570 id_stats->node_type = id->route.addr.dev_addr.dev_type;
3571 id_stats->port_num = id->port_num;
3572 id_stats->bound_dev_if =
3573 id->route.addr.dev_addr.bound_dev_if;
3574
3575 if (ibnl_put_attr(skb, nlh,
3576 rdma_addr_size(cma_src_addr(id_priv)),
3577 cma_src_addr(id_priv),
3578 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3579 goto out;
3580 if (ibnl_put_attr(skb, nlh,
3581 rdma_addr_size(cma_src_addr(id_priv)),
3582 cma_dst_addr(id_priv),
3583 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
3584 goto out;
3585
3586 id_stats->pid = id_priv->owner;
3587 id_stats->port_space = id->ps;
3588 id_stats->cm_state = id_priv->state;
3589 id_stats->qp_num = id_priv->qp_num;
3590 id_stats->qp_type = id->qp_type;
3591
3592 i_id++;
3593 }
3594
3595 cb->args[1] = 0;
3596 i_dev++;
3597 }
3598
3599 out:
3600 mutex_unlock(&lock);
3601 cb->args[0] = i_dev;
3602 cb->args[1] = i_id;
3603
3604 return skb->len;
3605 }
3606
3607 static const struct ibnl_client_cbs cma_cb_table[] = {
3608 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3609 .module = THIS_MODULE },
3610 };
3611
3612 static int __init cma_init(void)
3613 {
3614 int ret;
3615
3616 cma_wq = create_singlethread_workqueue("rdma_cm");
3617 if (!cma_wq)
3618 return -ENOMEM;
3619
3620 ib_sa_register_client(&sa_client);
3621 rdma_addr_register_client(&addr_client);
3622 register_netdevice_notifier(&cma_nb);
3623
3624 ret = ib_register_client(&cma_client);
3625 if (ret)
3626 goto err;
3627
3628 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3629 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3630
3631 return 0;
3632
3633 err:
3634 unregister_netdevice_notifier(&cma_nb);
3635 rdma_addr_unregister_client(&addr_client);
3636 ib_sa_unregister_client(&sa_client);
3637 destroy_workqueue(cma_wq);
3638 return ret;
3639 }
3640
3641 static void __exit cma_cleanup(void)
3642 {
3643 ibnl_remove_client(RDMA_NL_RDMA_CM);
3644 ib_unregister_client(&cma_client);
3645 unregister_netdevice_notifier(&cma_nb);
3646 rdma_addr_unregister_client(&addr_client);
3647 ib_sa_unregister_client(&sa_client);
3648 destroy_workqueue(cma_wq);
3649 idr_destroy(&tcp_ps);
3650 idr_destroy(&udp_ps);
3651 idr_destroy(&ipoib_ps);
3652 idr_destroy(&ib_ps);
3653 }
3654
3655 module_init(cma_init);
3656 module_exit(cma_cleanup);
Linux kernel - Deliverables
This page took 0.185426 seconds and 5 git commands to generate.