?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Revert "ALSA: hda: Flush interrupts on disabling"
[linux/fpc-iii.git] / drivers / infiniband / core / cma.c
blobdcfbf326f45c9110c57da13fea7142bfa9116712
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/igmp.h>
42 #include <linux/idr.h>
43 #include <linux/inetdevice.h>
44 #include <linux/slab.h>
45 #include <linux/module.h>
46 #include <net/route.h>
47
48 #include <net/net_namespace.h>
49 #include <net/netns/generic.h>
50 #include <net/tcp.h>
51 #include <net/ipv6.h>
52 #include <net/ip_fib.h>
53 #include <net/ip6_route.h>
54
55 #include <rdma/rdma_cm.h>
56 #include <rdma/rdma_cm_ib.h>
57 #include <rdma/rdma_netlink.h>
58 #include <rdma/ib.h>
59 #include <rdma/ib_cache.h>
60 #include <rdma/ib_cm.h>
61 #include <rdma/ib_sa.h>
62 #include <rdma/iw_cm.h>
63
64 #include "core_priv.h"
65
66 MODULE_AUTHOR("Sean Hefty");
67 MODULE_DESCRIPTION("Generic RDMA CM Agent");
68 MODULE_LICENSE("Dual BSD/GPL");
69
70 #define CMA_CM_RESPONSE_TIMEOUT 20
71 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
72 #define CMA_MAX_CM_RETRIES 15
73 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
74 #define CMA_IBOE_PACKET_LIFETIME 18
75
76 static const char * const cma_events[] = {
77 [RDMA_CM_EVENT_ADDR_RESOLVED] = "address resolved",
78 [RDMA_CM_EVENT_ADDR_ERROR] = "address error",
79 [RDMA_CM_EVENT_ROUTE_RESOLVED] = "route resolved ",
80 [RDMA_CM_EVENT_ROUTE_ERROR] = "route error",
81 [RDMA_CM_EVENT_CONNECT_REQUEST] = "connect request",
82 [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
83 [RDMA_CM_EVENT_CONNECT_ERROR] = "connect error",
84 [RDMA_CM_EVENT_UNREACHABLE] = "unreachable",
85 [RDMA_CM_EVENT_REJECTED] = "rejected",
86 [RDMA_CM_EVENT_ESTABLISHED] = "established",
87 [RDMA_CM_EVENT_DISCONNECTED] = "disconnected",
88 [RDMA_CM_EVENT_DEVICE_REMOVAL] = "device removal",
89 [RDMA_CM_EVENT_MULTICAST_JOIN] = "multicast join",
90 [RDMA_CM_EVENT_MULTICAST_ERROR] = "multicast error",
91 [RDMA_CM_EVENT_ADDR_CHANGE] = "address change",
92 [RDMA_CM_EVENT_TIMEWAIT_EXIT] = "timewait exit",
93 };
94
95 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
96 {
97 size_t index = event;
98
99 return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
100 cma_events[index] : "unrecognized event";
101 }
102 EXPORT_SYMBOL(rdma_event_msg);
103
104 static void cma_add_one(struct ib_device *device);
105 static void cma_remove_one(struct ib_device *device, void *client_data);
106
107 static struct ib_client cma_client = {
108 .name = "cma",
109 .add = cma_add_one,
110 .remove = cma_remove_one
111 };
112
113 static struct ib_sa_client sa_client;
114 static struct rdma_addr_client addr_client;
115 static LIST_HEAD(dev_list);
116 static LIST_HEAD(listen_any_list);
117 static DEFINE_MUTEX(lock);
118 static struct workqueue_struct *cma_wq;
119 static int cma_pernet_id;
120
121 struct cma_pernet {
122 struct idr tcp_ps;
123 struct idr udp_ps;
124 struct idr ipoib_ps;
125 struct idr ib_ps;
126 };
127
128 static struct cma_pernet *cma_pernet(struct net *net)
129 {
130 return net_generic(net, cma_pernet_id);
131 }
132
133 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
134 {
135 struct cma_pernet *pernet = cma_pernet(net);
136
137 switch (ps) {
138 case RDMA_PS_TCP:
139 return &pernet->tcp_ps;
140 case RDMA_PS_UDP:
141 return &pernet->udp_ps;
142 case RDMA_PS_IPOIB:
143 return &pernet->ipoib_ps;
144 case RDMA_PS_IB:
145 return &pernet->ib_ps;
146 default:
147 return NULL;
148 }
149 }
150
151 struct cma_device {
152 struct list_head list;
153 struct ib_device *device;
154 struct completion comp;
155 atomic_t refcount;
156 struct list_head id_list;
157 enum ib_gid_type *default_gid_type;
158 };
159
160 struct rdma_bind_list {
161 enum rdma_port_space ps;
162 struct hlist_head owners;
163 unsigned short port;
164 };
165
166 struct class_port_info_context {
167 struct ib_class_port_info *class_port_info;
168 struct ib_device *device;
169 struct completion done;
170 struct ib_sa_query *sa_query;
171 u8 port_num;
172 };
173
174 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
175 struct rdma_bind_list *bind_list, int snum)
176 {
177 struct idr *idr = cma_pernet_idr(net, ps);
178
179 return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
180 }
181
182 static struct rdma_bind_list *cma_ps_find(struct net *net,
183 enum rdma_port_space ps, int snum)
184 {
185 struct idr *idr = cma_pernet_idr(net, ps);
186
187 return idr_find(idr, snum);
188 }
189
190 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
191 {
192 struct idr *idr = cma_pernet_idr(net, ps);
193
194 idr_remove(idr, snum);
195 }
196
197 enum {
198 CMA_OPTION_AFONLY,
199 };
200
201 void cma_ref_dev(struct cma_device *cma_dev)
202 {
203 atomic_inc(&cma_dev->refcount);
204 }
205
206 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter filter,
207 void *cookie)
208 {
209 struct cma_device *cma_dev;
210 struct cma_device *found_cma_dev = NULL;
211
212 mutex_lock(&lock);
213
214 list_for_each_entry(cma_dev, &dev_list, list)
215 if (filter(cma_dev->device, cookie)) {
216 found_cma_dev = cma_dev;
217 break;
218 }
219
220 if (found_cma_dev)
221 cma_ref_dev(found_cma_dev);
222 mutex_unlock(&lock);
223 return found_cma_dev;
224 }
225
226 int cma_get_default_gid_type(struct cma_device *cma_dev,
227 unsigned int port)
228 {
229 if (port < rdma_start_port(cma_dev->device) ||
230 port > rdma_end_port(cma_dev->device))
231 return -EINVAL;
232
233 return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
234 }
235
236 int cma_set_default_gid_type(struct cma_device *cma_dev,
237 unsigned int port,
238 enum ib_gid_type default_gid_type)
239 {
240 unsigned long supported_gids;
241
242 if (port < rdma_start_port(cma_dev->device) ||
243 port > rdma_end_port(cma_dev->device))
244 return -EINVAL;
245
246 supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
247
248 if (!(supported_gids & 1 << default_gid_type))
249 return -EINVAL;
250
251 cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
252 default_gid_type;
253
254 return 0;
255 }
256
257 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
258 {
259 return cma_dev->device;
260 }
261
262 /*
263 * Device removal can occur at anytime, so we need extra handling to
264 * serialize notifying the user of device removal with other callbacks.
265 * We do this by disabling removal notification while a callback is in process,
266 * and reporting it after the callback completes.
267 */
268 struct rdma_id_private {
269 struct rdma_cm_id id;
270
271 struct rdma_bind_list *bind_list;
272 struct hlist_node node;
273 struct list_head list; /* listen_any_list or cma_device.list */
274 struct list_head listen_list; /* per device listens */
275 struct cma_device *cma_dev;
276 struct list_head mc_list;
277
278 int internal_id;
279 enum rdma_cm_state state;
280 spinlock_t lock;
281 struct mutex qp_mutex;
282
283 struct completion comp;
284 atomic_t refcount;
285 struct mutex handler_mutex;
286
287 int backlog;
288 int timeout_ms;
289 struct ib_sa_query *query;
290 int query_id;
291 union {
292 struct ib_cm_id *ib;
293 struct iw_cm_id *iw;
294 } cm_id;
295
296 u32 seq_num;
297 u32 qkey;
298 u32 qp_num;
299 pid_t owner;
300 u32 options;
301 u8 srq;
302 u8 tos;
303 u8 reuseaddr;
304 u8 afonly;
305 enum ib_gid_type gid_type;
306 };
307
308 struct cma_multicast {
309 struct rdma_id_private *id_priv;
310 union {
311 struct ib_sa_multicast *ib;
312 } multicast;
313 struct list_head list;
314 void *context;
315 struct sockaddr_storage addr;
316 struct kref mcref;
317 bool igmp_joined;
318 u8 join_state;
319 };
320
321 struct cma_work {
322 struct work_struct work;
323 struct rdma_id_private *id;
324 enum rdma_cm_state old_state;
325 enum rdma_cm_state new_state;
326 struct rdma_cm_event event;
327 };
328
329 struct cma_ndev_work {
330 struct work_struct work;
331 struct rdma_id_private *id;
332 struct rdma_cm_event event;
333 };
334
335 struct iboe_mcast_work {
336 struct work_struct work;
337 struct rdma_id_private *id;
338 struct cma_multicast *mc;
339 };
340
341 union cma_ip_addr {
342 struct in6_addr ip6;
343 struct {
344 __be32 pad[3];
345 __be32 addr;
346 } ip4;
347 };
348
349 struct cma_hdr {
350 u8 cma_version;
351 u8 ip_version; /* IP version: 7:4 */
352 __be16 port;
353 union cma_ip_addr src_addr;
354 union cma_ip_addr dst_addr;
355 };
356
357 #define CMA_VERSION 0x00
358
359 struct cma_req_info {
360 struct ib_device *device;
361 int port;
362 union ib_gid local_gid;
363 __be64 service_id;
364 u16 pkey;
365 bool has_gid:1;
366 };
367
368 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
369 {
370 unsigned long flags;
371 int ret;
372
373 spin_lock_irqsave(&id_priv->lock, flags);
374 ret = (id_priv->state == comp);
375 spin_unlock_irqrestore(&id_priv->lock, flags);
376 return ret;
377 }
378
379 static int cma_comp_exch(struct rdma_id_private *id_priv,
380 enum rdma_cm_state comp, enum rdma_cm_state exch)
381 {
382 unsigned long flags;
383 int ret;
384
385 spin_lock_irqsave(&id_priv->lock, flags);
386 if ((ret = (id_priv->state == comp)))
387 id_priv->state = exch;
388 spin_unlock_irqrestore(&id_priv->lock, flags);
389 return ret;
390 }
391
392 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
393 enum rdma_cm_state exch)
394 {
395 unsigned long flags;
396 enum rdma_cm_state old;
397
398 spin_lock_irqsave(&id_priv->lock, flags);
399 old = id_priv->state;
400 id_priv->state = exch;
401 spin_unlock_irqrestore(&id_priv->lock, flags);
402 return old;
403 }
404
405 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
406 {
407 return hdr->ip_version >> 4;
408 }
409
410 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
411 {
412 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
413 }
414
415 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
416 {
417 struct in_device *in_dev = NULL;
418
419 if (ndev) {
420 rtnl_lock();
421 in_dev = __in_dev_get_rtnl(ndev);
422 if (in_dev) {
423 if (join)
424 ip_mc_inc_group(in_dev,
425 *(__be32 *)(mgid->raw + 12));
426 else
427 ip_mc_dec_group(in_dev,
428 *(__be32 *)(mgid->raw + 12));
429 }
430 rtnl_unlock();
431 }
432 return (in_dev) ? 0 : -ENODEV;
433 }
434
435 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
436 struct cma_device *cma_dev)
437 {
438 cma_ref_dev(cma_dev);
439 id_priv->cma_dev = cma_dev;
440 id_priv->gid_type = 0;
441 id_priv->id.device = cma_dev->device;
442 id_priv->id.route.addr.dev_addr.transport =
443 rdma_node_get_transport(cma_dev->device->node_type);
444 list_add_tail(&id_priv->list, &cma_dev->id_list);
445 }
446
447 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
448 struct cma_device *cma_dev)
449 {
450 _cma_attach_to_dev(id_priv, cma_dev);
451 id_priv->gid_type =
452 cma_dev->default_gid_type[id_priv->id.port_num -
453 rdma_start_port(cma_dev->device)];
454 }
455
456 void cma_deref_dev(struct cma_device *cma_dev)
457 {
458 if (atomic_dec_and_test(&cma_dev->refcount))
459 complete(&cma_dev->comp);
460 }
461
462 static inline void release_mc(struct kref *kref)
463 {
464 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
465
466 kfree(mc->multicast.ib);
467 kfree(mc);
468 }
469
470 static void cma_release_dev(struct rdma_id_private *id_priv)
471 {
472 mutex_lock(&lock);
473 list_del(&id_priv->list);
474 cma_deref_dev(id_priv->cma_dev);
475 id_priv->cma_dev = NULL;
476 mutex_unlock(&lock);
477 }
478
479 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
480 {
481 return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
482 }
483
484 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
485 {
486 return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
487 }
488
489 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
490 {
491 return id_priv->id.route.addr.src_addr.ss_family;
492 }
493
494 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
495 {
496 struct ib_sa_mcmember_rec rec;
497 int ret = 0;
498
499 if (id_priv->qkey) {
500 if (qkey && id_priv->qkey != qkey)
501 return -EINVAL;
502 return 0;
503 }
504
505 if (qkey) {
506 id_priv->qkey = qkey;
507 return 0;
508 }
509
510 switch (id_priv->id.ps) {
511 case RDMA_PS_UDP:
512 case RDMA_PS_IB:
513 id_priv->qkey = RDMA_UDP_QKEY;
514 break;
515 case RDMA_PS_IPOIB:
516 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
517 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
518 id_priv->id.port_num, &rec.mgid,
519 &rec);
520 if (!ret)
521 id_priv->qkey = be32_to_cpu(rec.qkey);
522 break;
523 default:
524 break;
525 }
526 return ret;
527 }
528
529 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
530 {
531 dev_addr->dev_type = ARPHRD_INFINIBAND;
532 rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
533 ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
534 }
535
536 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
537 {
538 int ret;
539
540 if (addr->sa_family != AF_IB) {
541 ret = rdma_translate_ip(addr, dev_addr, NULL);
542 } else {
543 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
544 ret = 0;
545 }
546
547 return ret;
548 }
549
550 static inline int cma_validate_port(struct ib_device *device, u8 port,
551 enum ib_gid_type gid_type,
552 union ib_gid *gid, int dev_type,
553 int bound_if_index)
554 {
555 int ret = -ENODEV;
556 struct net_device *ndev = NULL;
557
558 if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
559 return ret;
560
561 if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
562 return ret;
563
564 if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
565 ndev = dev_get_by_index(&init_net, bound_if_index);
566 if (ndev && ndev->flags & IFF_LOOPBACK) {
567 pr_info("detected loopback device\n");
568 dev_put(ndev);
569
570 if (!device->get_netdev)
571 return -EOPNOTSUPP;
572
573 ndev = device->get_netdev(device, port);
574 if (!ndev)
575 return -ENODEV;
576 }
577 } else {
578 gid_type = IB_GID_TYPE_IB;
579 }
580
581 ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
582 ndev, NULL);
583
584 if (ndev)
585 dev_put(ndev);
586
587 return ret;
588 }
589
590 static int cma_acquire_dev(struct rdma_id_private *id_priv,
591 struct rdma_id_private *listen_id_priv)
592 {
593 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
594 struct cma_device *cma_dev;
595 union ib_gid gid, iboe_gid, *gidp;
596 int ret = -ENODEV;
597 u8 port;
598
599 if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
600 id_priv->id.ps == RDMA_PS_IPOIB)
601 return -EINVAL;
602
603 mutex_lock(&lock);
604 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
605 &iboe_gid);
606
607 memcpy(&gid, dev_addr->src_dev_addr +
608 rdma_addr_gid_offset(dev_addr), sizeof gid);
609
610 if (listen_id_priv) {
611 cma_dev = listen_id_priv->cma_dev;
612 port = listen_id_priv->id.port_num;
613 gidp = rdma_protocol_roce(cma_dev->device, port) ?
614 &iboe_gid : &gid;
615
616 ret = cma_validate_port(cma_dev->device, port,
617 rdma_protocol_ib(cma_dev->device, port) ?
618 IB_GID_TYPE_IB :
619 listen_id_priv->gid_type, gidp,
620 dev_addr->dev_type,
621 dev_addr->bound_dev_if);
622 if (!ret) {
623 id_priv->id.port_num = port;
624 goto out;
625 }
626 }
627
628 list_for_each_entry(cma_dev, &dev_list, list) {
629 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
630 if (listen_id_priv &&
631 listen_id_priv->cma_dev == cma_dev &&
632 listen_id_priv->id.port_num == port)
633 continue;
634
635 gidp = rdma_protocol_roce(cma_dev->device, port) ?
636 &iboe_gid : &gid;
637
638 ret = cma_validate_port(cma_dev->device, port,
639 rdma_protocol_ib(cma_dev->device, port) ?
640 IB_GID_TYPE_IB :
641 cma_dev->default_gid_type[port - 1],
642 gidp, dev_addr->dev_type,
643 dev_addr->bound_dev_if);
644 if (!ret) {
645 id_priv->id.port_num = port;
646 goto out;
647 }
648 }
649 }
650
651 out:
652 if (!ret)
653 cma_attach_to_dev(id_priv, cma_dev);
654
655 mutex_unlock(&lock);
656 return ret;
657 }
658
659 /*
660 * Select the source IB device and address to reach the destination IB address.
661 */
662 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
663 {
664 struct cma_device *cma_dev, *cur_dev;
665 struct sockaddr_ib *addr;
666 union ib_gid gid, sgid, *dgid;
667 u16 pkey, index;
668 u8 p;
669 int i;
670
671 cma_dev = NULL;
672 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
673 dgid = (union ib_gid *) &addr->sib_addr;
674 pkey = ntohs(addr->sib_pkey);
675
676 mutex_lock(&lock);
677 list_for_each_entry(cur_dev, &dev_list, list) {
678 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
679 if (!rdma_cap_af_ib(cur_dev->device, p))
680 continue;
681
682 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
683 continue;
684
685 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
686 &gid, NULL);
687 i++) {
688 if (!memcmp(&gid, dgid, sizeof(gid))) {
689 cma_dev = cur_dev;
690 sgid = gid;
691 id_priv->id.port_num = p;
692 goto found;
693 }
694
695 if (!cma_dev && (gid.global.subnet_prefix ==
696 dgid->global.subnet_prefix)) {
697 cma_dev = cur_dev;
698 sgid = gid;
699 id_priv->id.port_num = p;
700 goto found;
701 }
702 }
703 }
704 }
705 mutex_unlock(&lock);
706 return -ENODEV;
707
708 found:
709 cma_attach_to_dev(id_priv, cma_dev);
710 mutex_unlock(&lock);
711 addr = (struct sockaddr_ib *)cma_src_addr(id_priv);
712 memcpy(&addr->sib_addr, &sgid, sizeof(sgid));
713 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
714 return 0;
715 }
716
717 static void cma_deref_id(struct rdma_id_private *id_priv)
718 {
719 if (atomic_dec_and_test(&id_priv->refcount))
720 complete(&id_priv->comp);
721 }
722
723 struct rdma_cm_id *rdma_create_id(struct net *net,
724 rdma_cm_event_handler event_handler,
725 void *context, enum rdma_port_space ps,
726 enum ib_qp_type qp_type)
727 {
728 struct rdma_id_private *id_priv;
729
730 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
731 if (!id_priv)
732 return ERR_PTR(-ENOMEM);
733
734 id_priv->owner = task_pid_nr(current);
735 id_priv->state = RDMA_CM_IDLE;
736 id_priv->id.context = context;
737 id_priv->id.event_handler = event_handler;
738 id_priv->id.ps = ps;
739 id_priv->id.qp_type = qp_type;
740 spin_lock_init(&id_priv->lock);
741 mutex_init(&id_priv->qp_mutex);
742 init_completion(&id_priv->comp);
743 atomic_set(&id_priv->refcount, 1);
744 mutex_init(&id_priv->handler_mutex);
745 INIT_LIST_HEAD(&id_priv->listen_list);
746 INIT_LIST_HEAD(&id_priv->mc_list);
747 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
748 id_priv->id.route.addr.dev_addr.net = get_net(net);
749 id_priv->seq_num &= 0x00ffffff;
750
751 return &id_priv->id;
752 }
753 EXPORT_SYMBOL(rdma_create_id);
754
755 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
756 {
757 struct ib_qp_attr qp_attr;
758 int qp_attr_mask, ret;
759
760 qp_attr.qp_state = IB_QPS_INIT;
761 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
762 if (ret)
763 return ret;
764
765 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
766 if (ret)
767 return ret;
768
769 qp_attr.qp_state = IB_QPS_RTR;
770 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
771 if (ret)
772 return ret;
773
774 qp_attr.qp_state = IB_QPS_RTS;
775 qp_attr.sq_psn = 0;
776 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
777
778 return ret;
779 }
780
781 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
782 {
783 struct ib_qp_attr qp_attr;
784 int qp_attr_mask, ret;
785
786 qp_attr.qp_state = IB_QPS_INIT;
787 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
788 if (ret)
789 return ret;
790
791 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
792 }
793
794 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
795 struct ib_qp_init_attr *qp_init_attr)
796 {
797 struct rdma_id_private *id_priv;
798 struct ib_qp *qp;
799 int ret;
800
801 id_priv = container_of(id, struct rdma_id_private, id);
802 if (id->device != pd->device)
803 return -EINVAL;
804
805 qp_init_attr->port_num = id->port_num;
806 qp = ib_create_qp(pd, qp_init_attr);
807 if (IS_ERR(qp))
808 return PTR_ERR(qp);
809
810 if (id->qp_type == IB_QPT_UD)
811 ret = cma_init_ud_qp(id_priv, qp);
812 else
813 ret = cma_init_conn_qp(id_priv, qp);
814 if (ret)
815 goto err;
816
817 id->qp = qp;
818 id_priv->qp_num = qp->qp_num;
819 id_priv->srq = (qp->srq != NULL);
820 return 0;
821 err:
822 ib_destroy_qp(qp);
823 return ret;
824 }
825 EXPORT_SYMBOL(rdma_create_qp);
826
827 void rdma_destroy_qp(struct rdma_cm_id *id)
828 {
829 struct rdma_id_private *id_priv;
830
831 id_priv = container_of(id, struct rdma_id_private, id);
832 mutex_lock(&id_priv->qp_mutex);
833 ib_destroy_qp(id_priv->id.qp);
834 id_priv->id.qp = NULL;
835 mutex_unlock(&id_priv->qp_mutex);
836 }
837 EXPORT_SYMBOL(rdma_destroy_qp);
838
839 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
840 struct rdma_conn_param *conn_param)
841 {
842 struct ib_qp_attr qp_attr;
843 int qp_attr_mask, ret;
844 union ib_gid sgid;
845
846 mutex_lock(&id_priv->qp_mutex);
847 if (!id_priv->id.qp) {
848 ret = 0;
849 goto out;
850 }
851
852 /* Need to update QP attributes from default values. */
853 qp_attr.qp_state = IB_QPS_INIT;
854 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
855 if (ret)
856 goto out;
857
858 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
859 if (ret)
860 goto out;
861
862 qp_attr.qp_state = IB_QPS_RTR;
863 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
864 if (ret)
865 goto out;
866
867 ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
868 qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
869 if (ret)
870 goto out;
871
872 BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
873
874 if (conn_param)
875 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
876 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
877 out:
878 mutex_unlock(&id_priv->qp_mutex);
879 return ret;
880 }
881
882 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
883 struct rdma_conn_param *conn_param)
884 {
885 struct ib_qp_attr qp_attr;
886 int qp_attr_mask, ret;
887
888 mutex_lock(&id_priv->qp_mutex);
889 if (!id_priv->id.qp) {
890 ret = 0;
891 goto out;
892 }
893
894 qp_attr.qp_state = IB_QPS_RTS;
895 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
896 if (ret)
897 goto out;
898
899 if (conn_param)
900 qp_attr.max_rd_atomic = conn_param->initiator_depth;
901 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
902 out:
903 mutex_unlock(&id_priv->qp_mutex);
904 return ret;
905 }
906
907 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
908 {
909 struct ib_qp_attr qp_attr;
910 int ret;
911
912 mutex_lock(&id_priv->qp_mutex);
913 if (!id_priv->id.qp) {
914 ret = 0;
915 goto out;
916 }
917
918 qp_attr.qp_state = IB_QPS_ERR;
919 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
920 out:
921 mutex_unlock(&id_priv->qp_mutex);
922 return ret;
923 }
924
925 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
926 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
927 {
928 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
929 int ret;
930 u16 pkey;
931
932 if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
933 pkey = 0xffff;
934 else
935 pkey = ib_addr_get_pkey(dev_addr);
936
937 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
938 pkey, &qp_attr->pkey_index);
939 if (ret)
940 return ret;
941
942 qp_attr->port_num = id_priv->id.port_num;
943 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
944
945 if (id_priv->id.qp_type == IB_QPT_UD) {
946 ret = cma_set_qkey(id_priv, 0);
947 if (ret)
948 return ret;
949
950 qp_attr->qkey = id_priv->qkey;
951 *qp_attr_mask |= IB_QP_QKEY;
952 } else {
953 qp_attr->qp_access_flags = 0;
954 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
955 }
956 return 0;
957 }
958
959 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
960 int *qp_attr_mask)
961 {
962 struct rdma_id_private *id_priv;
963 int ret = 0;
964
965 id_priv = container_of(id, struct rdma_id_private, id);
966 if (rdma_cap_ib_cm(id->device, id->port_num)) {
967 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
968 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
969 else
970 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
971 qp_attr_mask);
972
973 if (qp_attr->qp_state == IB_QPS_RTR)
974 qp_attr->rq_psn = id_priv->seq_num;
975 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
976 if (!id_priv->cm_id.iw) {
977 qp_attr->qp_access_flags = 0;
978 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
979 } else
980 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
981 qp_attr_mask);
982 qp_attr->port_num = id_priv->id.port_num;
983 *qp_attr_mask |= IB_QP_PORT;
984 } else
985 ret = -ENOSYS;
986
987 return ret;
988 }
989 EXPORT_SYMBOL(rdma_init_qp_attr);
990
991 static inline int cma_zero_addr(struct sockaddr *addr)
992 {
993 switch (addr->sa_family) {
994 case AF_INET:
995 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
996 case AF_INET6:
997 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
998 case AF_IB:
999 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1000 default:
1001 return 0;
1002 }
1003 }
1004
1005 static inline int cma_loopback_addr(struct sockaddr *addr)
1006 {
1007 switch (addr->sa_family) {
1008 case AF_INET:
1009 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1010 case AF_INET6:
1011 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1012 case AF_IB:
1013 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1014 default:
1015 return 0;
1016 }
1017 }
1018
1019 static inline int cma_any_addr(struct sockaddr *addr)
1020 {
1021 return cma_zero_addr(addr) || cma_loopback_addr(addr);
1022 }
1023
1024 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1025 {
1026 if (src->sa_family != dst->sa_family)
1027 return -1;
1028
1029 switch (src->sa_family) {
1030 case AF_INET:
1031 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1032 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1033 case AF_INET6:
1034 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1035 &((struct sockaddr_in6 *) dst)->sin6_addr);
1036 default:
1037 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1038 &((struct sockaddr_ib *) dst)->sib_addr);
1039 }
1040 }
1041
1042 static __be16 cma_port(struct sockaddr *addr)
1043 {
1044 struct sockaddr_ib *sib;
1045
1046 switch (addr->sa_family) {
1047 case AF_INET:
1048 return ((struct sockaddr_in *) addr)->sin_port;
1049 case AF_INET6:
1050 return ((struct sockaddr_in6 *) addr)->sin6_port;
1051 case AF_IB:
1052 sib = (struct sockaddr_ib *) addr;
1053 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1054 be64_to_cpu(sib->sib_sid_mask)));
1055 default:
1056 return 0;
1057 }
1058 }
1059
1060 static inline int cma_any_port(struct sockaddr *addr)
1061 {
1062 return !cma_port(addr);
1063 }
1064
1065 static void cma_save_ib_info(struct sockaddr *src_addr,
1066 struct sockaddr *dst_addr,
1067 struct rdma_cm_id *listen_id,
1068 struct ib_sa_path_rec *path)
1069 {
1070 struct sockaddr_ib *listen_ib, *ib;
1071
1072 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1073 if (src_addr) {
1074 ib = (struct sockaddr_ib *)src_addr;
1075 ib->sib_family = AF_IB;
1076 if (path) {
1077 ib->sib_pkey = path->pkey;
1078 ib->sib_flowinfo = path->flow_label;
1079 memcpy(&ib->sib_addr, &path->sgid, 16);
1080 ib->sib_sid = path->service_id;
1081 ib->sib_scope_id = 0;
1082 } else {
1083 ib->sib_pkey = listen_ib->sib_pkey;
1084 ib->sib_flowinfo = listen_ib->sib_flowinfo;
1085 ib->sib_addr = listen_ib->sib_addr;
1086 ib->sib_sid = listen_ib->sib_sid;
1087 ib->sib_scope_id = listen_ib->sib_scope_id;
1088 }
1089 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1090 }
1091 if (dst_addr) {
1092 ib = (struct sockaddr_ib *)dst_addr;
1093 ib->sib_family = AF_IB;
1094 if (path) {
1095 ib->sib_pkey = path->pkey;
1096 ib->sib_flowinfo = path->flow_label;
1097 memcpy(&ib->sib_addr, &path->dgid, 16);
1098 }
1099 }
1100 }
1101
1102 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1103 struct sockaddr_in *dst_addr,
1104 struct cma_hdr *hdr,
1105 __be16 local_port)
1106 {
1107 if (src_addr) {
1108 *src_addr = (struct sockaddr_in) {
1109 .sin_family = AF_INET,
1110 .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1111 .sin_port = local_port,
1112 };
1113 }
1114
1115 if (dst_addr) {
1116 *dst_addr = (struct sockaddr_in) {
1117 .sin_family = AF_INET,
1118 .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1119 .sin_port = hdr->port,
1120 };
1121 }
1122 }
1123
1124 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1125 struct sockaddr_in6 *dst_addr,
1126 struct cma_hdr *hdr,
1127 __be16 local_port)
1128 {
1129 if (src_addr) {
1130 *src_addr = (struct sockaddr_in6) {
1131 .sin6_family = AF_INET6,
1132 .sin6_addr = hdr->dst_addr.ip6,
1133 .sin6_port = local_port,
1134 };
1135 }
1136
1137 if (dst_addr) {
1138 *dst_addr = (struct sockaddr_in6) {
1139 .sin6_family = AF_INET6,
1140 .sin6_addr = hdr->src_addr.ip6,
1141 .sin6_port = hdr->port,
1142 };
1143 }
1144 }
1145
1146 static u16 cma_port_from_service_id(__be64 service_id)
1147 {
1148 return (u16)be64_to_cpu(service_id);
1149 }
1150
1151 static int cma_save_ip_info(struct sockaddr *src_addr,
1152 struct sockaddr *dst_addr,
1153 struct ib_cm_event *ib_event,
1154 __be64 service_id)
1155 {
1156 struct cma_hdr *hdr;
1157 __be16 port;
1158
1159 hdr = ib_event->private_data;
1160 if (hdr->cma_version != CMA_VERSION)
1161 return -EINVAL;
1162
1163 port = htons(cma_port_from_service_id(service_id));
1164
1165 switch (cma_get_ip_ver(hdr)) {
1166 case 4:
1167 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1168 (struct sockaddr_in *)dst_addr, hdr, port);
1169 break;
1170 case 6:
1171 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1172 (struct sockaddr_in6 *)dst_addr, hdr, port);
1173 break;
1174 default:
1175 return -EAFNOSUPPORT;
1176 }
1177
1178 return 0;
1179 }
1180
1181 static int cma_save_net_info(struct sockaddr *src_addr,
1182 struct sockaddr *dst_addr,
1183 struct rdma_cm_id *listen_id,
1184 struct ib_cm_event *ib_event,
1185 sa_family_t sa_family, __be64 service_id)
1186 {
1187 if (sa_family == AF_IB) {
1188 if (ib_event->event == IB_CM_REQ_RECEIVED)
1189 cma_save_ib_info(src_addr, dst_addr, listen_id,
1190 ib_event->param.req_rcvd.primary_path);
1191 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1192 cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1193 return 0;
1194 }
1195
1196 return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1197 }
1198
1199 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1200 struct cma_req_info *req)
1201 {
1202 const struct ib_cm_req_event_param *req_param =
1203 &ib_event->param.req_rcvd;
1204 const struct ib_cm_sidr_req_event_param *sidr_param =
1205 &ib_event->param.sidr_req_rcvd;
1206
1207 switch (ib_event->event) {
1208 case IB_CM_REQ_RECEIVED:
1209 req->device = req_param->listen_id->device;
1210 req->port = req_param->port;
1211 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1212 sizeof(req->local_gid));
1213 req->has_gid = true;
1214 req->service_id = req_param->primary_path->service_id;
1215 req->pkey = be16_to_cpu(req_param->primary_path->pkey);
1216 if (req->pkey != req_param->bth_pkey)
1217 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1218 "RDMA CMA: in the future this may cause the request to be dropped\n",
1219 req_param->bth_pkey, req->pkey);
1220 break;
1221 case IB_CM_SIDR_REQ_RECEIVED:
1222 req->device = sidr_param->listen_id->device;
1223 req->port = sidr_param->port;
1224 req->has_gid = false;
1225 req->service_id = sidr_param->service_id;
1226 req->pkey = sidr_param->pkey;
1227 if (req->pkey != sidr_param->bth_pkey)
1228 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1229 "RDMA CMA: in the future this may cause the request to be dropped\n",
1230 sidr_param->bth_pkey, req->pkey);
1231 break;
1232 default:
1233 return -EINVAL;
1234 }
1235
1236 return 0;
1237 }
1238
1239 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1240 const struct sockaddr_in *dst_addr,
1241 const struct sockaddr_in *src_addr)
1242 {
1243 __be32 daddr = dst_addr->sin_addr.s_addr,
1244 saddr = src_addr->sin_addr.s_addr;
1245 struct fib_result res;
1246 struct flowi4 fl4;
1247 int err;
1248 bool ret;
1249
1250 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1251 ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1252 ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1253 ipv4_is_loopback(saddr))
1254 return false;
1255
1256 memset(&fl4, 0, sizeof(fl4));
1257 fl4.flowi4_iif = net_dev->ifindex;
1258 fl4.daddr = daddr;
1259 fl4.saddr = saddr;
1260
1261 rcu_read_lock();
1262 err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1263 ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1264 rcu_read_unlock();
1265
1266 return ret;
1267 }
1268
1269 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1270 const struct sockaddr_in6 *dst_addr,
1271 const struct sockaddr_in6 *src_addr)
1272 {
1273 #if IS_ENABLED(CONFIG_IPV6)
1274 const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1275 IPV6_ADDR_LINKLOCAL;
1276 struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1277 &src_addr->sin6_addr, net_dev->ifindex,
1278 strict);
1279 bool ret;
1280
1281 if (!rt)
1282 return false;
1283
1284 ret = rt->rt6i_idev->dev == net_dev;
1285 ip6_rt_put(rt);
1286
1287 return ret;
1288 #else
1289 return false;
1290 #endif
1291 }
1292
1293 static bool validate_net_dev(struct net_device *net_dev,
1294 const struct sockaddr *daddr,
1295 const struct sockaddr *saddr)
1296 {
1297 const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1298 const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1299 const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1300 const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1301
1302 switch (daddr->sa_family) {
1303 case AF_INET:
1304 return saddr->sa_family == AF_INET &&
1305 validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1306
1307 case AF_INET6:
1308 return saddr->sa_family == AF_INET6 &&
1309 validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1310
1311 default:
1312 return false;
1313 }
1314 }
1315
1316 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1317 const struct cma_req_info *req)
1318 {
1319 struct sockaddr_storage listen_addr_storage, src_addr_storage;
1320 struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1321 *src_addr = (struct sockaddr *)&src_addr_storage;
1322 struct net_device *net_dev;
1323 const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1324 int err;
1325
1326 err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1327 req->service_id);
1328 if (err)
1329 return ERR_PTR(err);
1330
1331 net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1332 gid, listen_addr);
1333 if (!net_dev)
1334 return ERR_PTR(-ENODEV);
1335
1336 if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1337 dev_put(net_dev);
1338 return ERR_PTR(-EHOSTUNREACH);
1339 }
1340
1341 return net_dev;
1342 }
1343
1344 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1345 {
1346 return (be64_to_cpu(service_id) >> 16) & 0xffff;
1347 }
1348
1349 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1350 const struct cma_hdr *hdr)
1351 {
1352 struct sockaddr *addr = cma_src_addr(id_priv);
1353 __be32 ip4_addr;
1354 struct in6_addr ip6_addr;
1355
1356 if (cma_any_addr(addr) && !id_priv->afonly)
1357 return true;
1358
1359 switch (addr->sa_family) {
1360 case AF_INET:
1361 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1362 if (cma_get_ip_ver(hdr) != 4)
1363 return false;
1364 if (!cma_any_addr(addr) &&
1365 hdr->dst_addr.ip4.addr != ip4_addr)
1366 return false;
1367 break;
1368 case AF_INET6:
1369 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1370 if (cma_get_ip_ver(hdr) != 6)
1371 return false;
1372 if (!cma_any_addr(addr) &&
1373 memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1374 return false;
1375 break;
1376 case AF_IB:
1377 return true;
1378 default:
1379 return false;
1380 }
1381
1382 return true;
1383 }
1384
1385 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1386 {
1387 enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1388 enum rdma_transport_type transport =
1389 rdma_node_get_transport(device->node_type);
1390
1391 return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1392 }
1393
1394 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1395 {
1396 struct ib_device *device = id->device;
1397 const int port_num = id->port_num ?: rdma_start_port(device);
1398
1399 return cma_protocol_roce_dev_port(device, port_num);
1400 }
1401
1402 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1403 const struct net_device *net_dev,
1404 u8 port_num)
1405 {
1406 const struct rdma_addr *addr = &id->route.addr;
1407
1408 if (!net_dev)
1409 /* This request is an AF_IB request or a RoCE request */
1410 return (!id->port_num || id->port_num == port_num) &&
1411 (addr->src_addr.ss_family == AF_IB ||
1412 cma_protocol_roce_dev_port(id->device, port_num));
1413
1414 /*
1415 * Net namespaces must match, and if the listner is listening
1416 * on a specific netdevice than netdevice must match as well.
1417 */
1418 if (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1419 (!!addr->dev_addr.bound_dev_if ==
1420 (addr->dev_addr.bound_dev_if == net_dev->ifindex)))
1421 return true;
1422 else
1423 return false;
1424 }
1425
1426 static struct rdma_id_private *cma_find_listener(
1427 const struct rdma_bind_list *bind_list,
1428 const struct ib_cm_id *cm_id,
1429 const struct ib_cm_event *ib_event,
1430 const struct cma_req_info *req,
1431 const struct net_device *net_dev)
1432 {
1433 struct rdma_id_private *id_priv, *id_priv_dev;
1434
1435 if (!bind_list)
1436 return ERR_PTR(-EINVAL);
1437
1438 hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1439 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1440 if (id_priv->id.device == cm_id->device &&
1441 cma_match_net_dev(&id_priv->id, net_dev, req->port))
1442 return id_priv;
1443 list_for_each_entry(id_priv_dev,
1444 &id_priv->listen_list,
1445 listen_list) {
1446 if (id_priv_dev->id.device == cm_id->device &&
1447 cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1448 return id_priv_dev;
1449 }
1450 }
1451 }
1452
1453 return ERR_PTR(-EINVAL);
1454 }
1455
1456 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1457 struct ib_cm_event *ib_event,
1458 struct net_device **net_dev)
1459 {
1460 struct cma_req_info req;
1461 struct rdma_bind_list *bind_list;
1462 struct rdma_id_private *id_priv;
1463 int err;
1464
1465 err = cma_save_req_info(ib_event, &req);
1466 if (err)
1467 return ERR_PTR(err);
1468
1469 *net_dev = cma_get_net_dev(ib_event, &req);
1470 if (IS_ERR(*net_dev)) {
1471 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1472 /* Assuming the protocol is AF_IB */
1473 *net_dev = NULL;
1474 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1475 /* TODO find the net dev matching the request parameters
1476 * through the RoCE GID table */
1477 *net_dev = NULL;
1478 } else {
1479 return ERR_CAST(*net_dev);
1480 }
1481 }
1482
1483 bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1484 rdma_ps_from_service_id(req.service_id),
1485 cma_port_from_service_id(req.service_id));
1486 id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1487 if (IS_ERR(id_priv) && *net_dev) {
1488 dev_put(*net_dev);
1489 *net_dev = NULL;
1490 }
1491
1492 return id_priv;
1493 }
1494
1495 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1496 {
1497 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1498 }
1499
1500 static void cma_cancel_route(struct rdma_id_private *id_priv)
1501 {
1502 if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1503 if (id_priv->query)
1504 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1505 }
1506 }
1507
1508 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1509 {
1510 struct rdma_id_private *dev_id_priv;
1511
1512 /*
1513 * Remove from listen_any_list to prevent added devices from spawning
1514 * additional listen requests.
1515 */
1516 mutex_lock(&lock);
1517 list_del(&id_priv->list);
1518
1519 while (!list_empty(&id_priv->listen_list)) {
1520 dev_id_priv = list_entry(id_priv->listen_list.next,
1521 struct rdma_id_private, listen_list);
1522 /* sync with device removal to avoid duplicate destruction */
1523 list_del_init(&dev_id_priv->list);
1524 list_del(&dev_id_priv->listen_list);
1525 mutex_unlock(&lock);
1526
1527 rdma_destroy_id(&dev_id_priv->id);
1528 mutex_lock(&lock);
1529 }
1530 mutex_unlock(&lock);
1531 }
1532
1533 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1534 enum rdma_cm_state state)
1535 {
1536 switch (state) {
1537 case RDMA_CM_ADDR_QUERY:
1538 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1539 break;
1540 case RDMA_CM_ROUTE_QUERY:
1541 cma_cancel_route(id_priv);
1542 break;
1543 case RDMA_CM_LISTEN:
1544 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1545 cma_cancel_listens(id_priv);
1546 break;
1547 default:
1548 break;
1549 }
1550 }
1551
1552 static void cma_release_port(struct rdma_id_private *id_priv)
1553 {
1554 struct rdma_bind_list *bind_list = id_priv->bind_list;
1555 struct net *net = id_priv->id.route.addr.dev_addr.net;
1556
1557 if (!bind_list)
1558 return;
1559
1560 mutex_lock(&lock);
1561 hlist_del(&id_priv->node);
1562 if (hlist_empty(&bind_list->owners)) {
1563 cma_ps_remove(net, bind_list->ps, bind_list->port);
1564 kfree(bind_list);
1565 }
1566 mutex_unlock(&lock);
1567 }
1568
1569 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1570 {
1571 struct cma_multicast *mc;
1572
1573 while (!list_empty(&id_priv->mc_list)) {
1574 mc = container_of(id_priv->mc_list.next,
1575 struct cma_multicast, list);
1576 list_del(&mc->list);
1577 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1578 id_priv->id.port_num)) {
1579 ib_sa_free_multicast(mc->multicast.ib);
1580 kfree(mc);
1581 } else {
1582 if (mc->igmp_joined) {
1583 struct rdma_dev_addr *dev_addr =
1584 &id_priv->id.route.addr.dev_addr;
1585 struct net_device *ndev = NULL;
1586
1587 if (dev_addr->bound_dev_if)
1588 ndev = dev_get_by_index(&init_net,
1589 dev_addr->bound_dev_if);
1590 if (ndev) {
1591 cma_igmp_send(ndev,
1592 &mc->multicast.ib->rec.mgid,
1593 false);
1594 dev_put(ndev);
1595 }
1596 }
1597 kref_put(&mc->mcref, release_mc);
1598 }
1599 }
1600 }
1601
1602 void rdma_destroy_id(struct rdma_cm_id *id)
1603 {
1604 struct rdma_id_private *id_priv;
1605 enum rdma_cm_state state;
1606
1607 id_priv = container_of(id, struct rdma_id_private, id);
1608 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1609 cma_cancel_operation(id_priv, state);
1610
1611 /*
1612 * Wait for any active callback to finish. New callbacks will find
1613 * the id_priv state set to destroying and abort.
1614 */
1615 mutex_lock(&id_priv->handler_mutex);
1616 mutex_unlock(&id_priv->handler_mutex);
1617
1618 if (id_priv->cma_dev) {
1619 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1620 if (id_priv->cm_id.ib)
1621 ib_destroy_cm_id(id_priv->cm_id.ib);
1622 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1623 if (id_priv->cm_id.iw)
1624 iw_destroy_cm_id(id_priv->cm_id.iw);
1625 }
1626 cma_leave_mc_groups(id_priv);
1627 cma_release_dev(id_priv);
1628 }
1629
1630 cma_release_port(id_priv);
1631 cma_deref_id(id_priv);
1632 wait_for_completion(&id_priv->comp);
1633
1634 if (id_priv->internal_id)
1635 cma_deref_id(id_priv->id.context);
1636
1637 kfree(id_priv->id.route.path_rec);
1638 put_net(id_priv->id.route.addr.dev_addr.net);
1639 kfree(id_priv);
1640 }
1641 EXPORT_SYMBOL(rdma_destroy_id);
1642
1643 static int cma_rep_recv(struct rdma_id_private *id_priv)
1644 {
1645 int ret;
1646
1647 ret = cma_modify_qp_rtr(id_priv, NULL);
1648 if (ret)
1649 goto reject;
1650
1651 ret = cma_modify_qp_rts(id_priv, NULL);
1652 if (ret)
1653 goto reject;
1654
1655 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1656 if (ret)
1657 goto reject;
1658
1659 return 0;
1660 reject:
1661 cma_modify_qp_err(id_priv);
1662 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1663 NULL, 0, NULL, 0);
1664 return ret;
1665 }
1666
1667 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1668 struct ib_cm_rep_event_param *rep_data,
1669 void *private_data)
1670 {
1671 event->param.conn.private_data = private_data;
1672 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1673 event->param.conn.responder_resources = rep_data->responder_resources;
1674 event->param.conn.initiator_depth = rep_data->initiator_depth;
1675 event->param.conn.flow_control = rep_data->flow_control;
1676 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1677 event->param.conn.srq = rep_data->srq;
1678 event->param.conn.qp_num = rep_data->remote_qpn;
1679 }
1680
1681 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1682 {
1683 struct rdma_id_private *id_priv = cm_id->context;
1684 struct rdma_cm_event event;
1685 int ret = 0;
1686
1687 mutex_lock(&id_priv->handler_mutex);
1688 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1689 id_priv->state != RDMA_CM_CONNECT) ||
1690 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1691 id_priv->state != RDMA_CM_DISCONNECT))
1692 goto out;
1693
1694 memset(&event, 0, sizeof event);
1695 switch (ib_event->event) {
1696 case IB_CM_REQ_ERROR:
1697 case IB_CM_REP_ERROR:
1698 event.event = RDMA_CM_EVENT_UNREACHABLE;
1699 event.status = -ETIMEDOUT;
1700 break;
1701 case IB_CM_REP_RECEIVED:
1702 if (id_priv->id.qp) {
1703 event.status = cma_rep_recv(id_priv);
1704 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1705 RDMA_CM_EVENT_ESTABLISHED;
1706 } else {
1707 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1708 }
1709 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1710 ib_event->private_data);
1711 break;
1712 case IB_CM_RTU_RECEIVED:
1713 case IB_CM_USER_ESTABLISHED:
1714 event.event = RDMA_CM_EVENT_ESTABLISHED;
1715 break;
1716 case IB_CM_DREQ_ERROR:
1717 event.status = -ETIMEDOUT; /* fall through */
1718 case IB_CM_DREQ_RECEIVED:
1719 case IB_CM_DREP_RECEIVED:
1720 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1721 RDMA_CM_DISCONNECT))
1722 goto out;
1723 event.event = RDMA_CM_EVENT_DISCONNECTED;
1724 break;
1725 case IB_CM_TIMEWAIT_EXIT:
1726 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1727 break;
1728 case IB_CM_MRA_RECEIVED:
1729 /* ignore event */
1730 goto out;
1731 case IB_CM_REJ_RECEIVED:
1732 cma_modify_qp_err(id_priv);
1733 event.status = ib_event->param.rej_rcvd.reason;
1734 event.event = RDMA_CM_EVENT_REJECTED;
1735 event.param.conn.private_data = ib_event->private_data;
1736 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1737 break;
1738 default:
1739 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1740 ib_event->event);
1741 goto out;
1742 }
1743
1744 ret = id_priv->id.event_handler(&id_priv->id, &event);
1745 if (ret) {
1746 /* Destroy the CM ID by returning a non-zero value. */
1747 id_priv->cm_id.ib = NULL;
1748 cma_exch(id_priv, RDMA_CM_DESTROYING);
1749 mutex_unlock(&id_priv->handler_mutex);
1750 rdma_destroy_id(&id_priv->id);
1751 return ret;
1752 }
1753 out:
1754 mutex_unlock(&id_priv->handler_mutex);
1755 return ret;
1756 }
1757
1758 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1759 struct ib_cm_event *ib_event,
1760 struct net_device *net_dev)
1761 {
1762 struct rdma_id_private *id_priv;
1763 struct rdma_cm_id *id;
1764 struct rdma_route *rt;
1765 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1766 const __be64 service_id =
1767 ib_event->param.req_rcvd.primary_path->service_id;
1768 int ret;
1769
1770 id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1771 listen_id->event_handler, listen_id->context,
1772 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1773 if (IS_ERR(id))
1774 return NULL;
1775
1776 id_priv = container_of(id, struct rdma_id_private, id);
1777 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1778 (struct sockaddr *)&id->route.addr.dst_addr,
1779 listen_id, ib_event, ss_family, service_id))
1780 goto err;
1781
1782 rt = &id->route;
1783 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1784 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1785 GFP_KERNEL);
1786 if (!rt->path_rec)
1787 goto err;
1788
1789 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1790 if (rt->num_paths == 2)
1791 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1792
1793 if (net_dev) {
1794 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1795 if (ret)
1796 goto err;
1797 } else {
1798 if (!cma_protocol_roce(listen_id) &&
1799 cma_any_addr(cma_src_addr(id_priv))) {
1800 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1801 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1802 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1803 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1804 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1805 if (ret)
1806 goto err;
1807 }
1808 }
1809 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1810
1811 id_priv->state = RDMA_CM_CONNECT;
1812 return id_priv;
1813
1814 err:
1815 rdma_destroy_id(id);
1816 return NULL;
1817 }
1818
1819 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1820 struct ib_cm_event *ib_event,
1821 struct net_device *net_dev)
1822 {
1823 struct rdma_id_private *id_priv;
1824 struct rdma_cm_id *id;
1825 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1826 struct net *net = listen_id->route.addr.dev_addr.net;
1827 int ret;
1828
1829 id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1830 listen_id->ps, IB_QPT_UD);
1831 if (IS_ERR(id))
1832 return NULL;
1833
1834 id_priv = container_of(id, struct rdma_id_private, id);
1835 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1836 (struct sockaddr *)&id->route.addr.dst_addr,
1837 listen_id, ib_event, ss_family,
1838 ib_event->param.sidr_req_rcvd.service_id))
1839 goto err;
1840
1841 if (net_dev) {
1842 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1843 if (ret)
1844 goto err;
1845 } else {
1846 if (!cma_any_addr(cma_src_addr(id_priv))) {
1847 ret = cma_translate_addr(cma_src_addr(id_priv),
1848 &id->route.addr.dev_addr);
1849 if (ret)
1850 goto err;
1851 }
1852 }
1853
1854 id_priv->state = RDMA_CM_CONNECT;
1855 return id_priv;
1856 err:
1857 rdma_destroy_id(id);
1858 return NULL;
1859 }
1860
1861 static void cma_set_req_event_data(struct rdma_cm_event *event,
1862 struct ib_cm_req_event_param *req_data,
1863 void *private_data, int offset)
1864 {
1865 event->param.conn.private_data = private_data + offset;
1866 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1867 event->param.conn.responder_resources = req_data->responder_resources;
1868 event->param.conn.initiator_depth = req_data->initiator_depth;
1869 event->param.conn.flow_control = req_data->flow_control;
1870 event->param.conn.retry_count = req_data->retry_count;
1871 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1872 event->param.conn.srq = req_data->srq;
1873 event->param.conn.qp_num = req_data->remote_qpn;
1874 }
1875
1876 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1877 {
1878 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1879 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1880 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1881 (id->qp_type == IB_QPT_UD)) ||
1882 (!id->qp_type));
1883 }
1884
1885 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1886 {
1887 struct rdma_id_private *listen_id, *conn_id = NULL;
1888 struct rdma_cm_event event;
1889 struct net_device *net_dev;
1890 u8 offset;
1891 int ret;
1892
1893 listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1894 if (IS_ERR(listen_id))
1895 return PTR_ERR(listen_id);
1896
1897 if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1898 ret = -EINVAL;
1899 goto net_dev_put;
1900 }
1901
1902 mutex_lock(&listen_id->handler_mutex);
1903 if (listen_id->state != RDMA_CM_LISTEN) {
1904 ret = -ECONNABORTED;
1905 goto err1;
1906 }
1907
1908 memset(&event, 0, sizeof event);
1909 offset = cma_user_data_offset(listen_id);
1910 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1911 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1912 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1913 event.param.ud.private_data = ib_event->private_data + offset;
1914 event.param.ud.private_data_len =
1915 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1916 } else {
1917 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1918 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1919 ib_event->private_data, offset);
1920 }
1921 if (!conn_id) {
1922 ret = -ENOMEM;
1923 goto err1;
1924 }
1925
1926 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1927 ret = cma_acquire_dev(conn_id, listen_id);
1928 if (ret)
1929 goto err2;
1930
1931 conn_id->cm_id.ib = cm_id;
1932 cm_id->context = conn_id;
1933 cm_id->cm_handler = cma_ib_handler;
1934
1935 /*
1936 * Protect against the user destroying conn_id from another thread
1937 * until we're done accessing it.
1938 */
1939 atomic_inc(&conn_id->refcount);
1940 ret = conn_id->id.event_handler(&conn_id->id, &event);
1941 if (ret)
1942 goto err3;
1943 /*
1944 * Acquire mutex to prevent user executing rdma_destroy_id()
1945 * while we're accessing the cm_id.
1946 */
1947 mutex_lock(&lock);
1948 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1949 (conn_id->id.qp_type != IB_QPT_UD))
1950 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1951 mutex_unlock(&lock);
1952 mutex_unlock(&conn_id->handler_mutex);
1953 mutex_unlock(&listen_id->handler_mutex);
1954 cma_deref_id(conn_id);
1955 if (net_dev)
1956 dev_put(net_dev);
1957 return 0;
1958
1959 err3:
1960 cma_deref_id(conn_id);
1961 /* Destroy the CM ID by returning a non-zero value. */
1962 conn_id->cm_id.ib = NULL;
1963 err2:
1964 cma_exch(conn_id, RDMA_CM_DESTROYING);
1965 mutex_unlock(&conn_id->handler_mutex);
1966 err1:
1967 mutex_unlock(&listen_id->handler_mutex);
1968 if (conn_id)
1969 rdma_destroy_id(&conn_id->id);
1970
1971 net_dev_put:
1972 if (net_dev)
1973 dev_put(net_dev);
1974
1975 return ret;
1976 }
1977
1978 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1979 {
1980 if (addr->sa_family == AF_IB)
1981 return ((struct sockaddr_ib *) addr)->sib_sid;
1982
1983 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1984 }
1985 EXPORT_SYMBOL(rdma_get_service_id);
1986
1987 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1988 {
1989 struct rdma_id_private *id_priv = iw_id->context;
1990 struct rdma_cm_event event;
1991 int ret = 0;
1992 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1993 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1994
1995 mutex_lock(&id_priv->handler_mutex);
1996 if (id_priv->state != RDMA_CM_CONNECT)
1997 goto out;
1998
1999 memset(&event, 0, sizeof event);
2000 switch (iw_event->event) {
2001 case IW_CM_EVENT_CLOSE:
2002 event.event = RDMA_CM_EVENT_DISCONNECTED;
2003 break;
2004 case IW_CM_EVENT_CONNECT_REPLY:
2005 memcpy(cma_src_addr(id_priv), laddr,
2006 rdma_addr_size(laddr));
2007 memcpy(cma_dst_addr(id_priv), raddr,
2008 rdma_addr_size(raddr));
2009 switch (iw_event->status) {
2010 case 0:
2011 event.event = RDMA_CM_EVENT_ESTABLISHED;
2012 event.param.conn.initiator_depth = iw_event->ird;
2013 event.param.conn.responder_resources = iw_event->ord;
2014 break;
2015 case -ECONNRESET:
2016 case -ECONNREFUSED:
2017 event.event = RDMA_CM_EVENT_REJECTED;
2018 break;
2019 case -ETIMEDOUT:
2020 event.event = RDMA_CM_EVENT_UNREACHABLE;
2021 break;
2022 default:
2023 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2024 break;
2025 }
2026 break;
2027 case IW_CM_EVENT_ESTABLISHED:
2028 event.event = RDMA_CM_EVENT_ESTABLISHED;
2029 event.param.conn.initiator_depth = iw_event->ird;
2030 event.param.conn.responder_resources = iw_event->ord;
2031 break;
2032 default:
2033 BUG_ON(1);
2034 }
2035
2036 event.status = iw_event->status;
2037 event.param.conn.private_data = iw_event->private_data;
2038 event.param.conn.private_data_len = iw_event->private_data_len;
2039 ret = id_priv->id.event_handler(&id_priv->id, &event);
2040 if (ret) {
2041 /* Destroy the CM ID by returning a non-zero value. */
2042 id_priv->cm_id.iw = NULL;
2043 cma_exch(id_priv, RDMA_CM_DESTROYING);
2044 mutex_unlock(&id_priv->handler_mutex);
2045 rdma_destroy_id(&id_priv->id);
2046 return ret;
2047 }
2048
2049 out:
2050 mutex_unlock(&id_priv->handler_mutex);
2051 return ret;
2052 }
2053
2054 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2055 struct iw_cm_event *iw_event)
2056 {
2057 struct rdma_cm_id *new_cm_id;
2058 struct rdma_id_private *listen_id, *conn_id;
2059 struct rdma_cm_event event;
2060 int ret = -ECONNABORTED;
2061 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2062 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2063
2064 listen_id = cm_id->context;
2065
2066 mutex_lock(&listen_id->handler_mutex);
2067 if (listen_id->state != RDMA_CM_LISTEN)
2068 goto out;
2069
2070 /* Create a new RDMA id for the new IW CM ID */
2071 new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2072 listen_id->id.event_handler,
2073 listen_id->id.context,
2074 RDMA_PS_TCP, IB_QPT_RC);
2075 if (IS_ERR(new_cm_id)) {
2076 ret = -ENOMEM;
2077 goto out;
2078 }
2079 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2080 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2081 conn_id->state = RDMA_CM_CONNECT;
2082
2083 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2084 if (ret) {
2085 mutex_unlock(&conn_id->handler_mutex);
2086 rdma_destroy_id(new_cm_id);
2087 goto out;
2088 }
2089
2090 ret = cma_acquire_dev(conn_id, listen_id);
2091 if (ret) {
2092 mutex_unlock(&conn_id->handler_mutex);
2093 rdma_destroy_id(new_cm_id);
2094 goto out;
2095 }
2096
2097 conn_id->cm_id.iw = cm_id;
2098 cm_id->context = conn_id;
2099 cm_id->cm_handler = cma_iw_handler;
2100
2101 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2102 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2103
2104 memset(&event, 0, sizeof event);
2105 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2106 event.param.conn.private_data = iw_event->private_data;
2107 event.param.conn.private_data_len = iw_event->private_data_len;
2108 event.param.conn.initiator_depth = iw_event->ird;
2109 event.param.conn.responder_resources = iw_event->ord;
2110
2111 /*
2112 * Protect against the user destroying conn_id from another thread
2113 * until we're done accessing it.
2114 */
2115 atomic_inc(&conn_id->refcount);
2116 ret = conn_id->id.event_handler(&conn_id->id, &event);
2117 if (ret) {
2118 /* User wants to destroy the CM ID */
2119 conn_id->cm_id.iw = NULL;
2120 cma_exch(conn_id, RDMA_CM_DESTROYING);
2121 mutex_unlock(&conn_id->handler_mutex);
2122 mutex_unlock(&listen_id->handler_mutex);
2123 cma_deref_id(conn_id);
2124 rdma_destroy_id(&conn_id->id);
2125 return ret;
2126 }
2127
2128 mutex_unlock(&conn_id->handler_mutex);
2129 cma_deref_id(conn_id);
2130
2131 out:
2132 mutex_unlock(&listen_id->handler_mutex);
2133 return ret;
2134 }
2135
2136 static int cma_ib_listen(struct rdma_id_private *id_priv)
2137 {
2138 struct sockaddr *addr;
2139 struct ib_cm_id *id;
2140 __be64 svc_id;
2141
2142 addr = cma_src_addr(id_priv);
2143 svc_id = rdma_get_service_id(&id_priv->id, addr);
2144 id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2145 if (IS_ERR(id))
2146 return PTR_ERR(id);
2147 id_priv->cm_id.ib = id;
2148
2149 return 0;
2150 }
2151
2152 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2153 {
2154 int ret;
2155 struct iw_cm_id *id;
2156
2157 id = iw_create_cm_id(id_priv->id.device,
2158 iw_conn_req_handler,
2159 id_priv);
2160 if (IS_ERR(id))
2161 return PTR_ERR(id);
2162
2163 id->tos = id_priv->tos;
2164 id_priv->cm_id.iw = id;
2165
2166 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2167 rdma_addr_size(cma_src_addr(id_priv)));
2168
2169 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2170
2171 if (ret) {
2172 iw_destroy_cm_id(id_priv->cm_id.iw);
2173 id_priv->cm_id.iw = NULL;
2174 }
2175
2176 return ret;
2177 }
2178
2179 static int cma_listen_handler(struct rdma_cm_id *id,
2180 struct rdma_cm_event *event)
2181 {
2182 struct rdma_id_private *id_priv = id->context;
2183
2184 id->context = id_priv->id.context;
2185 id->event_handler = id_priv->id.event_handler;
2186 return id_priv->id.event_handler(id, event);
2187 }
2188
2189 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2190 struct cma_device *cma_dev)
2191 {
2192 struct rdma_id_private *dev_id_priv;
2193 struct rdma_cm_id *id;
2194 struct net *net = id_priv->id.route.addr.dev_addr.net;
2195 int ret;
2196
2197 if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2198 return;
2199
2200 id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2201 id_priv->id.qp_type);
2202 if (IS_ERR(id))
2203 return;
2204
2205 dev_id_priv = container_of(id, struct rdma_id_private, id);
2206
2207 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2208 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2209 rdma_addr_size(cma_src_addr(id_priv)));
2210
2211 _cma_attach_to_dev(dev_id_priv, cma_dev);
2212 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2213 atomic_inc(&id_priv->refcount);
2214 dev_id_priv->internal_id = 1;
2215 dev_id_priv->afonly = id_priv->afonly;
2216
2217 ret = rdma_listen(id, id_priv->backlog);
2218 if (ret)
2219 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2220 ret, cma_dev->device->name);
2221 }
2222
2223 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2224 {
2225 struct cma_device *cma_dev;
2226
2227 mutex_lock(&lock);
2228 list_add_tail(&id_priv->list, &listen_any_list);
2229 list_for_each_entry(cma_dev, &dev_list, list)
2230 cma_listen_on_dev(id_priv, cma_dev);
2231 mutex_unlock(&lock);
2232 }
2233
2234 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2235 {
2236 struct rdma_id_private *id_priv;
2237
2238 id_priv = container_of(id, struct rdma_id_private, id);
2239 id_priv->tos = (u8) tos;
2240 }
2241 EXPORT_SYMBOL(rdma_set_service_type);
2242
2243 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2244 void *context)
2245 {
2246 struct cma_work *work = context;
2247 struct rdma_route *route;
2248
2249 route = &work->id->id.route;
2250
2251 if (!status) {
2252 route->num_paths = 1;
2253 *route->path_rec = *path_rec;
2254 } else {
2255 work->old_state = RDMA_CM_ROUTE_QUERY;
2256 work->new_state = RDMA_CM_ADDR_RESOLVED;
2257 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2258 work->event.status = status;
2259 }
2260
2261 queue_work(cma_wq, &work->work);
2262 }
2263
2264 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2265 struct cma_work *work)
2266 {
2267 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2268 struct ib_sa_path_rec path_rec;
2269 ib_sa_comp_mask comp_mask;
2270 struct sockaddr_in6 *sin6;
2271 struct sockaddr_ib *sib;
2272
2273 memset(&path_rec, 0, sizeof path_rec);
2274 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2275 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2276 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2277 path_rec.numb_path = 1;
2278 path_rec.reversible = 1;
2279 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2280
2281 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2282 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2283 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2284
2285 switch (cma_family(id_priv)) {
2286 case AF_INET:
2287 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2288 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2289 break;
2290 case AF_INET6:
2291 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2292 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2293 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2294 break;
2295 case AF_IB:
2296 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2297 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2298 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2299 break;
2300 }
2301
2302 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2303 id_priv->id.port_num, &path_rec,
2304 comp_mask, timeout_ms,
2305 GFP_KERNEL, cma_query_handler,
2306 work, &id_priv->query);
2307
2308 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2309 }
2310
2311 static void cma_work_handler(struct work_struct *_work)
2312 {
2313 struct cma_work *work = container_of(_work, struct cma_work, work);
2314 struct rdma_id_private *id_priv = work->id;
2315 int destroy = 0;
2316
2317 mutex_lock(&id_priv->handler_mutex);
2318 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2319 goto out;
2320
2321 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2322 cma_exch(id_priv, RDMA_CM_DESTROYING);
2323 destroy = 1;
2324 }
2325 out:
2326 mutex_unlock(&id_priv->handler_mutex);
2327 cma_deref_id(id_priv);
2328 if (destroy)
2329 rdma_destroy_id(&id_priv->id);
2330 kfree(work);
2331 }
2332
2333 static void cma_ndev_work_handler(struct work_struct *_work)
2334 {
2335 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2336 struct rdma_id_private *id_priv = work->id;
2337 int destroy = 0;
2338
2339 mutex_lock(&id_priv->handler_mutex);
2340 if (id_priv->state == RDMA_CM_DESTROYING ||
2341 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2342 goto out;
2343
2344 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2345 cma_exch(id_priv, RDMA_CM_DESTROYING);
2346 destroy = 1;
2347 }
2348
2349 out:
2350 mutex_unlock(&id_priv->handler_mutex);
2351 cma_deref_id(id_priv);
2352 if (destroy)
2353 rdma_destroy_id(&id_priv->id);
2354 kfree(work);
2355 }
2356
2357 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2358 {
2359 struct rdma_route *route = &id_priv->id.route;
2360 struct cma_work *work;
2361 int ret;
2362
2363 work = kzalloc(sizeof *work, GFP_KERNEL);
2364 if (!work)
2365 return -ENOMEM;
2366
2367 work->id = id_priv;
2368 INIT_WORK(&work->work, cma_work_handler);
2369 work->old_state = RDMA_CM_ROUTE_QUERY;
2370 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2371 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2372
2373 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2374 if (!route->path_rec) {
2375 ret = -ENOMEM;
2376 goto err1;
2377 }
2378
2379 ret = cma_query_ib_route(id_priv, timeout_ms, work);
2380 if (ret)
2381 goto err2;
2382
2383 return 0;
2384 err2:
2385 kfree(route->path_rec);
2386 route->path_rec = NULL;
2387 err1:
2388 kfree(work);
2389 return ret;
2390 }
2391
2392 int rdma_set_ib_paths(struct rdma_cm_id *id,
2393 struct ib_sa_path_rec *path_rec, int num_paths)
2394 {
2395 struct rdma_id_private *id_priv;
2396 int ret;
2397
2398 id_priv = container_of(id, struct rdma_id_private, id);
2399 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2400 RDMA_CM_ROUTE_RESOLVED))
2401 return -EINVAL;
2402
2403 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2404 GFP_KERNEL);
2405 if (!id->route.path_rec) {
2406 ret = -ENOMEM;
2407 goto err;
2408 }
2409
2410 id->route.num_paths = num_paths;
2411 return 0;
2412 err:
2413 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2414 return ret;
2415 }
2416 EXPORT_SYMBOL(rdma_set_ib_paths);
2417
2418 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2419 {
2420 struct cma_work *work;
2421
2422 work = kzalloc(sizeof *work, GFP_KERNEL);
2423 if (!work)
2424 return -ENOMEM;
2425
2426 work->id = id_priv;
2427 INIT_WORK(&work->work, cma_work_handler);
2428 work->old_state = RDMA_CM_ROUTE_QUERY;
2429 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2430 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2431 queue_work(cma_wq, &work->work);
2432 return 0;
2433 }
2434
2435 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2436 {
2437 int prio;
2438 struct net_device *dev;
2439
2440 prio = rt_tos2priority(tos);
2441 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2442 vlan_dev_real_dev(ndev) : ndev;
2443
2444 if (dev->num_tc)
2445 return netdev_get_prio_tc_map(dev, prio);
2446
2447 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2448 if (ndev->priv_flags & IFF_802_1Q_VLAN)
2449 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2450 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2451 #endif
2452 return 0;
2453 }
2454
2455 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2456 unsigned long supported_gids,
2457 enum ib_gid_type default_gid)
2458 {
2459 if ((network_type == RDMA_NETWORK_IPV4 ||
2460 network_type == RDMA_NETWORK_IPV6) &&
2461 test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2462 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2463
2464 return default_gid;
2465 }
2466
2467 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2468 {
2469 struct rdma_route *route = &id_priv->id.route;
2470 struct rdma_addr *addr = &route->addr;
2471 struct cma_work *work;
2472 int ret;
2473 struct net_device *ndev = NULL;
2474
2475
2476 work = kzalloc(sizeof *work, GFP_KERNEL);
2477 if (!work)
2478 return -ENOMEM;
2479
2480 work->id = id_priv;
2481 INIT_WORK(&work->work, cma_work_handler);
2482
2483 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2484 if (!route->path_rec) {
2485 ret = -ENOMEM;
2486 goto err1;
2487 }
2488
2489 route->num_paths = 1;
2490
2491 if (addr->dev_addr.bound_dev_if) {
2492 unsigned long supported_gids;
2493
2494 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2495 if (!ndev) {
2496 ret = -ENODEV;
2497 goto err2;
2498 }
2499
2500 if (ndev->flags & IFF_LOOPBACK) {
2501 dev_put(ndev);
2502 if (!id_priv->id.device->get_netdev) {
2503 ret = -EOPNOTSUPP;
2504 goto err2;
2505 }
2506
2507 ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2508 id_priv->id.port_num);
2509 if (!ndev) {
2510 ret = -ENODEV;
2511 goto err2;
2512 }
2513 }
2514
2515 route->path_rec->net = &init_net;
2516 route->path_rec->ifindex = ndev->ifindex;
2517 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2518 id_priv->id.port_num);
2519 route->path_rec->gid_type =
2520 cma_route_gid_type(addr->dev_addr.network,
2521 supported_gids,
2522 id_priv->gid_type);
2523 }
2524 if (!ndev) {
2525 ret = -ENODEV;
2526 goto err2;
2527 }
2528
2529 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2530
2531 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2532 &route->path_rec->sgid);
2533 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2534 &route->path_rec->dgid);
2535
2536 /* Use the hint from IP Stack to select GID Type */
2537 if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2538 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2539 if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2540 /* TODO: get the hoplimit from the inet/inet6 device */
2541 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2542 else
2543 route->path_rec->hop_limit = 1;
2544 route->path_rec->reversible = 1;
2545 route->path_rec->pkey = cpu_to_be16(0xffff);
2546 route->path_rec->mtu_selector = IB_SA_EQ;
2547 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2548 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2549 route->path_rec->rate_selector = IB_SA_EQ;
2550 route->path_rec->rate = iboe_get_rate(ndev);
2551 dev_put(ndev);
2552 route->path_rec->packet_life_time_selector = IB_SA_EQ;
2553 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2554 if (!route->path_rec->mtu) {
2555 ret = -EINVAL;
2556 goto err2;
2557 }
2558
2559 work->old_state = RDMA_CM_ROUTE_QUERY;
2560 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2561 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2562 work->event.status = 0;
2563
2564 queue_work(cma_wq, &work->work);
2565
2566 return 0;
2567
2568 err2:
2569 kfree(route->path_rec);
2570 route->path_rec = NULL;
2571 err1:
2572 kfree(work);
2573 return ret;
2574 }
2575
2576 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2577 {
2578 struct rdma_id_private *id_priv;
2579 int ret;
2580
2581 id_priv = container_of(id, struct rdma_id_private, id);
2582 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2583 return -EINVAL;
2584
2585 atomic_inc(&id_priv->refcount);
2586 if (rdma_cap_ib_sa(id->device, id->port_num))
2587 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2588 else if (rdma_protocol_roce(id->device, id->port_num))
2589 ret = cma_resolve_iboe_route(id_priv);
2590 else if (rdma_protocol_iwarp(id->device, id->port_num))
2591 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2592 else
2593 ret = -ENOSYS;
2594
2595 if (ret)
2596 goto err;
2597
2598 return 0;
2599 err:
2600 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2601 cma_deref_id(id_priv);
2602 return ret;
2603 }
2604 EXPORT_SYMBOL(rdma_resolve_route);
2605
2606 static void cma_set_loopback(struct sockaddr *addr)
2607 {
2608 switch (addr->sa_family) {
2609 case AF_INET:
2610 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2611 break;
2612 case AF_INET6:
2613 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2614 0, 0, 0, htonl(1));
2615 break;
2616 default:
2617 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2618 0, 0, 0, htonl(1));
2619 break;
2620 }
2621 }
2622
2623 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2624 {
2625 struct cma_device *cma_dev, *cur_dev;
2626 struct ib_port_attr port_attr;
2627 union ib_gid gid;
2628 u16 pkey;
2629 int ret;
2630 u8 p;
2631
2632 cma_dev = NULL;
2633 mutex_lock(&lock);
2634 list_for_each_entry(cur_dev, &dev_list, list) {
2635 if (cma_family(id_priv) == AF_IB &&
2636 !rdma_cap_ib_cm(cur_dev->device, 1))
2637 continue;
2638
2639 if (!cma_dev)
2640 cma_dev = cur_dev;
2641
2642 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2643 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2644 port_attr.state == IB_PORT_ACTIVE) {
2645 cma_dev = cur_dev;
2646 goto port_found;
2647 }
2648 }
2649 }
2650
2651 if (!cma_dev) {
2652 ret = -ENODEV;
2653 goto out;
2654 }
2655
2656 p = 1;
2657
2658 port_found:
2659 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2660 if (ret)
2661 goto out;
2662
2663 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2664 if (ret)
2665 goto out;
2666
2667 id_priv->id.route.addr.dev_addr.dev_type =
2668 (rdma_protocol_ib(cma_dev->device, p)) ?
2669 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2670
2671 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2672 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2673 id_priv->id.port_num = p;
2674 cma_attach_to_dev(id_priv, cma_dev);
2675 cma_set_loopback(cma_src_addr(id_priv));
2676 out:
2677 mutex_unlock(&lock);
2678 return ret;
2679 }
2680
2681 static void addr_handler(int status, struct sockaddr *src_addr,
2682 struct rdma_dev_addr *dev_addr, void *context)
2683 {
2684 struct rdma_id_private *id_priv = context;
2685 struct rdma_cm_event event;
2686
2687 memset(&event, 0, sizeof event);
2688 mutex_lock(&id_priv->handler_mutex);
2689 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2690 RDMA_CM_ADDR_RESOLVED))
2691 goto out;
2692
2693 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2694 if (!status && !id_priv->cma_dev)
2695 status = cma_acquire_dev(id_priv, NULL);
2696
2697 if (status) {
2698 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2699 RDMA_CM_ADDR_BOUND))
2700 goto out;
2701 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2702 event.status = status;
2703 } else
2704 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2705
2706 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2707 cma_exch(id_priv, RDMA_CM_DESTROYING);
2708 mutex_unlock(&id_priv->handler_mutex);
2709 cma_deref_id(id_priv);
2710 rdma_destroy_id(&id_priv->id);
2711 return;
2712 }
2713 out:
2714 mutex_unlock(&id_priv->handler_mutex);
2715 cma_deref_id(id_priv);
2716 }
2717
2718 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2719 {
2720 struct cma_work *work;
2721 union ib_gid gid;
2722 int ret;
2723
2724 work = kzalloc(sizeof *work, GFP_KERNEL);
2725 if (!work)
2726 return -ENOMEM;
2727
2728 if (!id_priv->cma_dev) {
2729 ret = cma_bind_loopback(id_priv);
2730 if (ret)
2731 goto err;
2732 }
2733
2734 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2735 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2736
2737 work->id = id_priv;
2738 INIT_WORK(&work->work, cma_work_handler);
2739 work->old_state = RDMA_CM_ADDR_QUERY;
2740 work->new_state = RDMA_CM_ADDR_RESOLVED;
2741 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2742 queue_work(cma_wq, &work->work);
2743 return 0;
2744 err:
2745 kfree(work);
2746 return ret;
2747 }
2748
2749 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2750 {
2751 struct cma_work *work;
2752 int ret;
2753
2754 work = kzalloc(sizeof *work, GFP_KERNEL);
2755 if (!work)
2756 return -ENOMEM;
2757
2758 if (!id_priv->cma_dev) {
2759 ret = cma_resolve_ib_dev(id_priv);
2760 if (ret)
2761 goto err;
2762 }
2763
2764 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2765 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2766
2767 work->id = id_priv;
2768 INIT_WORK(&work->work, cma_work_handler);
2769 work->old_state = RDMA_CM_ADDR_QUERY;
2770 work->new_state = RDMA_CM_ADDR_RESOLVED;
2771 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2772 queue_work(cma_wq, &work->work);
2773 return 0;
2774 err:
2775 kfree(work);
2776 return ret;
2777 }
2778
2779 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2780 struct sockaddr *dst_addr)
2781 {
2782 if (!src_addr || !src_addr->sa_family) {
2783 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2784 src_addr->sa_family = dst_addr->sa_family;
2785 if (IS_ENABLED(CONFIG_IPV6) &&
2786 dst_addr->sa_family == AF_INET6) {
2787 struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2788 struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2789 src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2790 if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2791 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2792 } else if (dst_addr->sa_family == AF_IB) {
2793 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2794 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2795 }
2796 }
2797 return rdma_bind_addr(id, src_addr);
2798 }
2799
2800 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2801 struct sockaddr *dst_addr, int timeout_ms)
2802 {
2803 struct rdma_id_private *id_priv;
2804 int ret;
2805
2806 id_priv = container_of(id, struct rdma_id_private, id);
2807 if (id_priv->state == RDMA_CM_IDLE) {
2808 ret = cma_bind_addr(id, src_addr, dst_addr);
2809 if (ret)
2810 return ret;
2811 }
2812
2813 if (cma_family(id_priv) != dst_addr->sa_family)
2814 return -EINVAL;
2815
2816 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2817 return -EINVAL;
2818
2819 atomic_inc(&id_priv->refcount);
2820 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2821 if (cma_any_addr(dst_addr)) {
2822 ret = cma_resolve_loopback(id_priv);
2823 } else {
2824 if (dst_addr->sa_family == AF_IB) {
2825 ret = cma_resolve_ib_addr(id_priv);
2826 } else {
2827 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2828 dst_addr, &id->route.addr.dev_addr,
2829 timeout_ms, addr_handler, id_priv);
2830 }
2831 }
2832 if (ret)
2833 goto err;
2834
2835 return 0;
2836 err:
2837 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2838 cma_deref_id(id_priv);
2839 return ret;
2840 }
2841 EXPORT_SYMBOL(rdma_resolve_addr);
2842
2843 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2844 {
2845 struct rdma_id_private *id_priv;
2846 unsigned long flags;
2847 int ret;
2848
2849 id_priv = container_of(id, struct rdma_id_private, id);
2850 spin_lock_irqsave(&id_priv->lock, flags);
2851 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2852 id_priv->reuseaddr = reuse;
2853 ret = 0;
2854 } else {
2855 ret = -EINVAL;
2856 }
2857 spin_unlock_irqrestore(&id_priv->lock, flags);
2858 return ret;
2859 }
2860 EXPORT_SYMBOL(rdma_set_reuseaddr);
2861
2862 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2863 {
2864 struct rdma_id_private *id_priv;
2865 unsigned long flags;
2866 int ret;
2867
2868 id_priv = container_of(id, struct rdma_id_private, id);
2869 spin_lock_irqsave(&id_priv->lock, flags);
2870 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2871 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2872 id_priv->afonly = afonly;
2873 ret = 0;
2874 } else {
2875 ret = -EINVAL;
2876 }
2877 spin_unlock_irqrestore(&id_priv->lock, flags);
2878 return ret;
2879 }
2880 EXPORT_SYMBOL(rdma_set_afonly);
2881
2882 static void cma_bind_port(struct rdma_bind_list *bind_list,
2883 struct rdma_id_private *id_priv)
2884 {
2885 struct sockaddr *addr;
2886 struct sockaddr_ib *sib;
2887 u64 sid, mask;
2888 __be16 port;
2889
2890 addr = cma_src_addr(id_priv);
2891 port = htons(bind_list->port);
2892
2893 switch (addr->sa_family) {
2894 case AF_INET:
2895 ((struct sockaddr_in *) addr)->sin_port = port;
2896 break;
2897 case AF_INET6:
2898 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2899 break;
2900 case AF_IB:
2901 sib = (struct sockaddr_ib *) addr;
2902 sid = be64_to_cpu(sib->sib_sid);
2903 mask = be64_to_cpu(sib->sib_sid_mask);
2904 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2905 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2906 break;
2907 }
2908 id_priv->bind_list = bind_list;
2909 hlist_add_head(&id_priv->node, &bind_list->owners);
2910 }
2911
2912 static int cma_alloc_port(enum rdma_port_space ps,
2913 struct rdma_id_private *id_priv, unsigned short snum)
2914 {
2915 struct rdma_bind_list *bind_list;
2916 int ret;
2917
2918 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2919 if (!bind_list)
2920 return -ENOMEM;
2921
2922 ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2923 snum);
2924 if (ret < 0)
2925 goto err;
2926
2927 bind_list->ps = ps;
2928 bind_list->port = (unsigned short)ret;
2929 cma_bind_port(bind_list, id_priv);
2930 return 0;
2931 err:
2932 kfree(bind_list);
2933 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2934 }
2935
2936 static int cma_alloc_any_port(enum rdma_port_space ps,
2937 struct rdma_id_private *id_priv)
2938 {
2939 static unsigned int last_used_port;
2940 int low, high, remaining;
2941 unsigned int rover;
2942 struct net *net = id_priv->id.route.addr.dev_addr.net;
2943
2944 inet_get_local_port_range(net, &low, &high);
2945 remaining = (high - low) + 1;
2946 rover = prandom_u32() % remaining + low;
2947 retry:
2948 if (last_used_port != rover &&
2949 !cma_ps_find(net, ps, (unsigned short)rover)) {
2950 int ret = cma_alloc_port(ps, id_priv, rover);
2951 /*
2952 * Remember previously used port number in order to avoid
2953 * re-using same port immediately after it is closed.
2954 */
2955 if (!ret)
2956 last_used_port = rover;
2957 if (ret != -EADDRNOTAVAIL)
2958 return ret;
2959 }
2960 if (--remaining) {
2961 rover++;
2962 if ((rover < low) || (rover > high))
2963 rover = low;
2964 goto retry;
2965 }
2966 return -EADDRNOTAVAIL;
2967 }
2968
2969 /*
2970 * Check that the requested port is available. This is called when trying to
2971 * bind to a specific port, or when trying to listen on a bound port. In
2972 * the latter case, the provided id_priv may already be on the bind_list, but
2973 * we still need to check that it's okay to start listening.
2974 */
2975 static int cma_check_port(struct rdma_bind_list *bind_list,
2976 struct rdma_id_private *id_priv, uint8_t reuseaddr)
2977 {
2978 struct rdma_id_private *cur_id;
2979 struct sockaddr *addr, *cur_addr;
2980
2981 addr = cma_src_addr(id_priv);
2982 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2983 if (id_priv == cur_id)
2984 continue;
2985
2986 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2987 cur_id->reuseaddr)
2988 continue;
2989
2990 cur_addr = cma_src_addr(cur_id);
2991 if (id_priv->afonly && cur_id->afonly &&
2992 (addr->sa_family != cur_addr->sa_family))
2993 continue;
2994
2995 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2996 return -EADDRNOTAVAIL;
2997
2998 if (!cma_addr_cmp(addr, cur_addr))
2999 return -EADDRINUSE;
3000 }
3001 return 0;
3002 }
3003
3004 static int cma_use_port(enum rdma_port_space ps,
3005 struct rdma_id_private *id_priv)
3006 {
3007 struct rdma_bind_list *bind_list;
3008 unsigned short snum;
3009 int ret;
3010
3011 snum = ntohs(cma_port(cma_src_addr(id_priv)));
3012 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3013 return -EACCES;
3014
3015 bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3016 if (!bind_list) {
3017 ret = cma_alloc_port(ps, id_priv, snum);
3018 } else {
3019 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3020 if (!ret)
3021 cma_bind_port(bind_list, id_priv);
3022 }
3023 return ret;
3024 }
3025
3026 static int cma_bind_listen(struct rdma_id_private *id_priv)
3027 {
3028 struct rdma_bind_list *bind_list = id_priv->bind_list;
3029 int ret = 0;
3030
3031 mutex_lock(&lock);
3032 if (bind_list->owners.first->next)
3033 ret = cma_check_port(bind_list, id_priv, 0);
3034 mutex_unlock(&lock);
3035 return ret;
3036 }
3037
3038 static enum rdma_port_space cma_select_inet_ps(
3039 struct rdma_id_private *id_priv)
3040 {
3041 switch (id_priv->id.ps) {
3042 case RDMA_PS_TCP:
3043 case RDMA_PS_UDP:
3044 case RDMA_PS_IPOIB:
3045 case RDMA_PS_IB:
3046 return id_priv->id.ps;
3047 default:
3048
3049 return 0;
3050 }
3051 }
3052
3053 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3054 {
3055 enum rdma_port_space ps = 0;
3056 struct sockaddr_ib *sib;
3057 u64 sid_ps, mask, sid;
3058
3059 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3060 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3061 sid = be64_to_cpu(sib->sib_sid) & mask;
3062
3063 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3064 sid_ps = RDMA_IB_IP_PS_IB;
3065 ps = RDMA_PS_IB;
3066 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3067 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3068 sid_ps = RDMA_IB_IP_PS_TCP;
3069 ps = RDMA_PS_TCP;
3070 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3071 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3072 sid_ps = RDMA_IB_IP_PS_UDP;
3073 ps = RDMA_PS_UDP;
3074 }
3075
3076 if (ps) {
3077 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3078 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3079 be64_to_cpu(sib->sib_sid_mask));
3080 }
3081 return ps;
3082 }
3083
3084 static int cma_get_port(struct rdma_id_private *id_priv)
3085 {
3086 enum rdma_port_space ps;
3087 int ret;
3088
3089 if (cma_family(id_priv) != AF_IB)
3090 ps = cma_select_inet_ps(id_priv);
3091 else
3092 ps = cma_select_ib_ps(id_priv);
3093 if (!ps)
3094 return -EPROTONOSUPPORT;
3095
3096 mutex_lock(&lock);
3097 if (cma_any_port(cma_src_addr(id_priv)))
3098 ret = cma_alloc_any_port(ps, id_priv);
3099 else
3100 ret = cma_use_port(ps, id_priv);
3101 mutex_unlock(&lock);
3102
3103 return ret;
3104 }
3105
3106 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3107 struct sockaddr *addr)
3108 {
3109 #if IS_ENABLED(CONFIG_IPV6)
3110 struct sockaddr_in6 *sin6;
3111
3112 if (addr->sa_family != AF_INET6)
3113 return 0;
3114
3115 sin6 = (struct sockaddr_in6 *) addr;
3116
3117 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3118 return 0;
3119
3120 if (!sin6->sin6_scope_id)
3121 return -EINVAL;
3122
3123 dev_addr->bound_dev_if = sin6->sin6_scope_id;
3124 #endif
3125 return 0;
3126 }
3127
3128 int rdma_listen(struct rdma_cm_id *id, int backlog)
3129 {
3130 struct rdma_id_private *id_priv;
3131 int ret;
3132
3133 id_priv = container_of(id, struct rdma_id_private, id);
3134 if (id_priv->state == RDMA_CM_IDLE) {
3135 id->route.addr.src_addr.ss_family = AF_INET;
3136 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3137 if (ret)
3138 return ret;
3139 }
3140
3141 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3142 return -EINVAL;
3143
3144 if (id_priv->reuseaddr) {
3145 ret = cma_bind_listen(id_priv);
3146 if (ret)
3147 goto err;
3148 }
3149
3150 id_priv->backlog = backlog;
3151 if (id->device) {
3152 if (rdma_cap_ib_cm(id->device, 1)) {
3153 ret = cma_ib_listen(id_priv);
3154 if (ret)
3155 goto err;
3156 } else if (rdma_cap_iw_cm(id->device, 1)) {
3157 ret = cma_iw_listen(id_priv, backlog);
3158 if (ret)
3159 goto err;
3160 } else {
3161 ret = -ENOSYS;
3162 goto err;
3163 }
3164 } else
3165 cma_listen_on_all(id_priv);
3166
3167 return 0;
3168 err:
3169 id_priv->backlog = 0;
3170 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3171 return ret;
3172 }
3173 EXPORT_SYMBOL(rdma_listen);
3174
3175 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3176 {
3177 struct rdma_id_private *id_priv;
3178 int ret;
3179
3180 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3181 addr->sa_family != AF_IB)
3182 return -EAFNOSUPPORT;
3183
3184 id_priv = container_of(id, struct rdma_id_private, id);
3185 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3186 return -EINVAL;
3187
3188 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3189 if (ret)
3190 goto err1;
3191
3192 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3193 if (!cma_any_addr(addr)) {
3194 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3195 if (ret)
3196 goto err1;
3197
3198 ret = cma_acquire_dev(id_priv, NULL);
3199 if (ret)
3200 goto err1;
3201 }
3202
3203 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3204 if (addr->sa_family == AF_INET)
3205 id_priv->afonly = 1;
3206 #if IS_ENABLED(CONFIG_IPV6)
3207 else if (addr->sa_family == AF_INET6) {
3208 struct net *net = id_priv->id.route.addr.dev_addr.net;
3209
3210 id_priv->afonly = net->ipv6.sysctl.bindv6only;
3211 }
3212 #endif
3213 }
3214 ret = cma_get_port(id_priv);
3215 if (ret)
3216 goto err2;
3217
3218 return 0;
3219 err2:
3220 if (id_priv->cma_dev)
3221 cma_release_dev(id_priv);
3222 err1:
3223 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3224 return ret;
3225 }
3226 EXPORT_SYMBOL(rdma_bind_addr);
3227
3228 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3229 {
3230 struct cma_hdr *cma_hdr;
3231
3232 cma_hdr = hdr;
3233 cma_hdr->cma_version = CMA_VERSION;
3234 if (cma_family(id_priv) == AF_INET) {
3235 struct sockaddr_in *src4, *dst4;
3236
3237 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3238 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3239
3240 cma_set_ip_ver(cma_hdr, 4);
3241 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3242 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3243 cma_hdr->port = src4->sin_port;
3244 } else if (cma_family(id_priv) == AF_INET6) {
3245 struct sockaddr_in6 *src6, *dst6;
3246
3247 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3248 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3249
3250 cma_set_ip_ver(cma_hdr, 6);
3251 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3252 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3253 cma_hdr->port = src6->sin6_port;
3254 }
3255 return 0;
3256 }
3257
3258 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3259 struct ib_cm_event *ib_event)
3260 {
3261 struct rdma_id_private *id_priv = cm_id->context;
3262 struct rdma_cm_event event;
3263 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3264 int ret = 0;
3265
3266 mutex_lock(&id_priv->handler_mutex);
3267 if (id_priv->state != RDMA_CM_CONNECT)
3268 goto out;
3269
3270 memset(&event, 0, sizeof event);
3271 switch (ib_event->event) {
3272 case IB_CM_SIDR_REQ_ERROR:
3273 event.event = RDMA_CM_EVENT_UNREACHABLE;
3274 event.status = -ETIMEDOUT;
3275 break;
3276 case IB_CM_SIDR_REP_RECEIVED:
3277 event.param.ud.private_data = ib_event->private_data;
3278 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3279 if (rep->status != IB_SIDR_SUCCESS) {
3280 event.event = RDMA_CM_EVENT_UNREACHABLE;
3281 event.status = ib_event->param.sidr_rep_rcvd.status;
3282 break;
3283 }
3284 ret = cma_set_qkey(id_priv, rep->qkey);
3285 if (ret) {
3286 event.event = RDMA_CM_EVENT_ADDR_ERROR;
3287 event.status = ret;
3288 break;
3289 }
3290 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3291 id_priv->id.route.path_rec,
3292 &event.param.ud.ah_attr);
3293 event.param.ud.qp_num = rep->qpn;
3294 event.param.ud.qkey = rep->qkey;
3295 event.event = RDMA_CM_EVENT_ESTABLISHED;
3296 event.status = 0;
3297 break;
3298 default:
3299 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3300 ib_event->event);
3301 goto out;
3302 }
3303
3304 ret = id_priv->id.event_handler(&id_priv->id, &event);
3305 if (ret) {
3306 /* Destroy the CM ID by returning a non-zero value. */
3307 id_priv->cm_id.ib = NULL;
3308 cma_exch(id_priv, RDMA_CM_DESTROYING);
3309 mutex_unlock(&id_priv->handler_mutex);
3310 rdma_destroy_id(&id_priv->id);
3311 return ret;
3312 }
3313 out:
3314 mutex_unlock(&id_priv->handler_mutex);
3315 return ret;
3316 }
3317
3318 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3319 struct rdma_conn_param *conn_param)
3320 {
3321 struct ib_cm_sidr_req_param req;
3322 struct ib_cm_id *id;
3323 void *private_data;
3324 u8 offset;
3325 int ret;
3326
3327 memset(&req, 0, sizeof req);
3328 offset = cma_user_data_offset(id_priv);
3329 req.private_data_len = offset + conn_param->private_data_len;
3330 if (req.private_data_len < conn_param->private_data_len)
3331 return -EINVAL;
3332
3333 if (req.private_data_len) {
3334 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3335 if (!private_data)
3336 return -ENOMEM;
3337 } else {
3338 private_data = NULL;
3339 }
3340
3341 if (conn_param->private_data && conn_param->private_data_len)
3342 memcpy(private_data + offset, conn_param->private_data,
3343 conn_param->private_data_len);
3344
3345 if (private_data) {
3346 ret = cma_format_hdr(private_data, id_priv);
3347 if (ret)
3348 goto out;
3349 req.private_data = private_data;
3350 }
3351
3352 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3353 id_priv);
3354 if (IS_ERR(id)) {
3355 ret = PTR_ERR(id);
3356 goto out;
3357 }
3358 id_priv->cm_id.ib = id;
3359
3360 req.path = id_priv->id.route.path_rec;
3361 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3362 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3363 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3364
3365 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3366 if (ret) {
3367 ib_destroy_cm_id(id_priv->cm_id.ib);
3368 id_priv->cm_id.ib = NULL;
3369 }
3370 out:
3371 kfree(private_data);
3372 return ret;
3373 }
3374
3375 static int cma_connect_ib(struct rdma_id_private *id_priv,
3376 struct rdma_conn_param *conn_param)
3377 {
3378 struct ib_cm_req_param req;
3379 struct rdma_route *route;
3380 void *private_data;
3381 struct ib_cm_id *id;
3382 u8 offset;
3383 int ret;
3384
3385 memset(&req, 0, sizeof req);
3386 offset = cma_user_data_offset(id_priv);
3387 req.private_data_len = offset + conn_param->private_data_len;
3388 if (req.private_data_len < conn_param->private_data_len)
3389 return -EINVAL;
3390
3391 if (req.private_data_len) {
3392 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3393 if (!private_data)
3394 return -ENOMEM;
3395 } else {
3396 private_data = NULL;
3397 }
3398
3399 if (conn_param->private_data && conn_param->private_data_len)
3400 memcpy(private_data + offset, conn_param->private_data,
3401 conn_param->private_data_len);
3402
3403 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3404 if (IS_ERR(id)) {
3405 ret = PTR_ERR(id);
3406 goto out;
3407 }
3408 id_priv->cm_id.ib = id;
3409
3410 route = &id_priv->id.route;
3411 if (private_data) {
3412 ret = cma_format_hdr(private_data, id_priv);
3413 if (ret)
3414 goto out;
3415 req.private_data = private_data;
3416 }
3417
3418 req.primary_path = &route->path_rec[0];
3419 if (route->num_paths == 2)
3420 req.alternate_path = &route->path_rec[1];
3421
3422 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3423 req.qp_num = id_priv->qp_num;
3424 req.qp_type = id_priv->id.qp_type;
3425 req.starting_psn = id_priv->seq_num;
3426 req.responder_resources = conn_param->responder_resources;
3427 req.initiator_depth = conn_param->initiator_depth;
3428 req.flow_control = conn_param->flow_control;
3429 req.retry_count = min_t(u8, 7, conn_param->retry_count);
3430 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3431 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3432 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3433 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3434 req.srq = id_priv->srq ? 1 : 0;
3435
3436 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3437 out:
3438 if (ret && !IS_ERR(id)) {
3439 ib_destroy_cm_id(id);
3440 id_priv->cm_id.ib = NULL;
3441 }
3442
3443 kfree(private_data);
3444 return ret;
3445 }
3446
3447 static int cma_connect_iw(struct rdma_id_private *id_priv,
3448 struct rdma_conn_param *conn_param)
3449 {
3450 struct iw_cm_id *cm_id;
3451 int ret;
3452 struct iw_cm_conn_param iw_param;
3453
3454 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3455 if (IS_ERR(cm_id))
3456 return PTR_ERR(cm_id);
3457
3458 cm_id->tos = id_priv->tos;
3459 id_priv->cm_id.iw = cm_id;
3460
3461 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3462 rdma_addr_size(cma_src_addr(id_priv)));
3463 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3464 rdma_addr_size(cma_dst_addr(id_priv)));
3465
3466 ret = cma_modify_qp_rtr(id_priv, conn_param);
3467 if (ret)
3468 goto out;
3469
3470 if (conn_param) {
3471 iw_param.ord = conn_param->initiator_depth;
3472 iw_param.ird = conn_param->responder_resources;
3473 iw_param.private_data = conn_param->private_data;
3474 iw_param.private_data_len = conn_param->private_data_len;
3475 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3476 } else {
3477 memset(&iw_param, 0, sizeof iw_param);
3478 iw_param.qpn = id_priv->qp_num;
3479 }
3480 ret = iw_cm_connect(cm_id, &iw_param);
3481 out:
3482 if (ret) {
3483 iw_destroy_cm_id(cm_id);
3484 id_priv->cm_id.iw = NULL;
3485 }
3486 return ret;
3487 }
3488
3489 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3490 {
3491 struct rdma_id_private *id_priv;
3492 int ret;
3493
3494 id_priv = container_of(id, struct rdma_id_private, id);
3495 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3496 return -EINVAL;
3497
3498 if (!id->qp) {
3499 id_priv->qp_num = conn_param->qp_num;
3500 id_priv->srq = conn_param->srq;
3501 }
3502
3503 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3504 if (id->qp_type == IB_QPT_UD)
3505 ret = cma_resolve_ib_udp(id_priv, conn_param);
3506 else
3507 ret = cma_connect_ib(id_priv, conn_param);
3508 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3509 ret = cma_connect_iw(id_priv, conn_param);
3510 else
3511 ret = -ENOSYS;
3512 if (ret)
3513 goto err;
3514
3515 return 0;
3516 err:
3517 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3518 return ret;
3519 }
3520 EXPORT_SYMBOL(rdma_connect);
3521
3522 static int cma_accept_ib(struct rdma_id_private *id_priv,
3523 struct rdma_conn_param *conn_param)
3524 {
3525 struct ib_cm_rep_param rep;
3526 int ret;
3527
3528 ret = cma_modify_qp_rtr(id_priv, conn_param);
3529 if (ret)
3530 goto out;
3531
3532 ret = cma_modify_qp_rts(id_priv, conn_param);
3533 if (ret)
3534 goto out;
3535
3536 memset(&rep, 0, sizeof rep);
3537 rep.qp_num = id_priv->qp_num;
3538 rep.starting_psn = id_priv->seq_num;
3539 rep.private_data = conn_param->private_data;
3540 rep.private_data_len = conn_param->private_data_len;
3541 rep.responder_resources = conn_param->responder_resources;
3542 rep.initiator_depth = conn_param->initiator_depth;
3543 rep.failover_accepted = 0;
3544 rep.flow_control = conn_param->flow_control;
3545 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3546 rep.srq = id_priv->srq ? 1 : 0;
3547
3548 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3549 out:
3550 return ret;
3551 }
3552
3553 static int cma_accept_iw(struct rdma_id_private *id_priv,
3554 struct rdma_conn_param *conn_param)
3555 {
3556 struct iw_cm_conn_param iw_param;
3557 int ret;
3558
3559 if (!conn_param)
3560 return -EINVAL;
3561
3562 ret = cma_modify_qp_rtr(id_priv, conn_param);
3563 if (ret)
3564 return ret;
3565
3566 iw_param.ord = conn_param->initiator_depth;
3567 iw_param.ird = conn_param->responder_resources;
3568 iw_param.private_data = conn_param->private_data;
3569 iw_param.private_data_len = conn_param->private_data_len;
3570 if (id_priv->id.qp) {
3571 iw_param.qpn = id_priv->qp_num;
3572 } else
3573 iw_param.qpn = conn_param->qp_num;
3574
3575 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3576 }
3577
3578 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3579 enum ib_cm_sidr_status status, u32 qkey,
3580 const void *private_data, int private_data_len)
3581 {
3582 struct ib_cm_sidr_rep_param rep;
3583 int ret;
3584
3585 memset(&rep, 0, sizeof rep);
3586 rep.status = status;
3587 if (status == IB_SIDR_SUCCESS) {
3588 ret = cma_set_qkey(id_priv, qkey);
3589 if (ret)
3590 return ret;
3591 rep.qp_num = id_priv->qp_num;
3592 rep.qkey = id_priv->qkey;
3593 }
3594 rep.private_data = private_data;
3595 rep.private_data_len = private_data_len;
3596
3597 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3598 }
3599
3600 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3601 {
3602 struct rdma_id_private *id_priv;
3603 int ret;
3604
3605 id_priv = container_of(id, struct rdma_id_private, id);
3606
3607 id_priv->owner = task_pid_nr(current);
3608
3609 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3610 return -EINVAL;
3611
3612 if (!id->qp && conn_param) {
3613 id_priv->qp_num = conn_param->qp_num;
3614 id_priv->srq = conn_param->srq;
3615 }
3616
3617 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3618 if (id->qp_type == IB_QPT_UD) {
3619 if (conn_param)
3620 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3621 conn_param->qkey,
3622 conn_param->private_data,
3623 conn_param->private_data_len);
3624 else
3625 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3626 0, NULL, 0);
3627 } else {
3628 if (conn_param)
3629 ret = cma_accept_ib(id_priv, conn_param);
3630 else
3631 ret = cma_rep_recv(id_priv);
3632 }
3633 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3634 ret = cma_accept_iw(id_priv, conn_param);
3635 else
3636 ret = -ENOSYS;
3637
3638 if (ret)
3639 goto reject;
3640
3641 return 0;
3642 reject:
3643 cma_modify_qp_err(id_priv);
3644 rdma_reject(id, NULL, 0);
3645 return ret;
3646 }
3647 EXPORT_SYMBOL(rdma_accept);
3648
3649 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3650 {
3651 struct rdma_id_private *id_priv;
3652 int ret;
3653
3654 id_priv = container_of(id, struct rdma_id_private, id);
3655 if (!id_priv->cm_id.ib)
3656 return -EINVAL;
3657
3658 switch (id->device->node_type) {
3659 case RDMA_NODE_IB_CA:
3660 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3661 break;
3662 default:
3663 ret = 0;
3664 break;
3665 }
3666 return ret;
3667 }
3668 EXPORT_SYMBOL(rdma_notify);
3669
3670 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3671 u8 private_data_len)
3672 {
3673 struct rdma_id_private *id_priv;
3674 int ret;
3675
3676 id_priv = container_of(id, struct rdma_id_private, id);
3677 if (!id_priv->cm_id.ib)
3678 return -EINVAL;
3679
3680 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3681 if (id->qp_type == IB_QPT_UD)
3682 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3683 private_data, private_data_len);
3684 else
3685 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3686 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3687 0, private_data, private_data_len);
3688 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3689 ret = iw_cm_reject(id_priv->cm_id.iw,
3690 private_data, private_data_len);
3691 } else
3692 ret = -ENOSYS;
3693
3694 return ret;
3695 }
3696 EXPORT_SYMBOL(rdma_reject);
3697
3698 int rdma_disconnect(struct rdma_cm_id *id)
3699 {
3700 struct rdma_id_private *id_priv;
3701 int ret;
3702
3703 id_priv = container_of(id, struct rdma_id_private, id);
3704 if (!id_priv->cm_id.ib)
3705 return -EINVAL;
3706
3707 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3708 ret = cma_modify_qp_err(id_priv);
3709 if (ret)
3710 goto out;
3711 /* Initiate or respond to a disconnect. */
3712 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3713 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3714 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3715 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3716 } else
3717 ret = -EINVAL;
3718
3719 out:
3720 return ret;
3721 }
3722 EXPORT_SYMBOL(rdma_disconnect);
3723
3724 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3725 {
3726 struct rdma_id_private *id_priv;
3727 struct cma_multicast *mc = multicast->context;
3728 struct rdma_cm_event event;
3729 int ret = 0;
3730
3731 id_priv = mc->id_priv;
3732 mutex_lock(&id_priv->handler_mutex);
3733 if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3734 id_priv->state != RDMA_CM_ADDR_RESOLVED)
3735 goto out;
3736
3737 if (!status)
3738 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3739 mutex_lock(&id_priv->qp_mutex);
3740 if (!status && id_priv->id.qp)
3741 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3742 be16_to_cpu(multicast->rec.mlid));
3743 mutex_unlock(&id_priv->qp_mutex);
3744
3745 memset(&event, 0, sizeof event);
3746 event.status = status;
3747 event.param.ud.private_data = mc->context;
3748 if (!status) {
3749 struct rdma_dev_addr *dev_addr =
3750 &id_priv->id.route.addr.dev_addr;
3751 struct net_device *ndev =
3752 dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3753 enum ib_gid_type gid_type =
3754 id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3755 rdma_start_port(id_priv->cma_dev->device)];
3756
3757 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3758 ib_init_ah_from_mcmember(id_priv->id.device,
3759 id_priv->id.port_num, &multicast->rec,
3760 ndev, gid_type,
3761 &event.param.ud.ah_attr);
3762 event.param.ud.qp_num = 0xFFFFFF;
3763 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3764 if (ndev)
3765 dev_put(ndev);
3766 } else
3767 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3768
3769 ret = id_priv->id.event_handler(&id_priv->id, &event);
3770 if (ret) {
3771 cma_exch(id_priv, RDMA_CM_DESTROYING);
3772 mutex_unlock(&id_priv->handler_mutex);
3773 rdma_destroy_id(&id_priv->id);
3774 return 0;
3775 }
3776
3777 out:
3778 mutex_unlock(&id_priv->handler_mutex);
3779 return 0;
3780 }
3781
3782 static void cma_set_mgid(struct rdma_id_private *id_priv,
3783 struct sockaddr *addr, union ib_gid *mgid)
3784 {
3785 unsigned char mc_map[MAX_ADDR_LEN];
3786 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3787 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3788 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3789
3790 if (cma_any_addr(addr)) {
3791 memset(mgid, 0, sizeof *mgid);
3792 } else if ((addr->sa_family == AF_INET6) &&
3793 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3794 0xFF10A01B)) {
3795 /* IPv6 address is an SA assigned MGID. */
3796 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3797 } else if (addr->sa_family == AF_IB) {
3798 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3799 } else if ((addr->sa_family == AF_INET6)) {
3800 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3801 if (id_priv->id.ps == RDMA_PS_UDP)
3802 mc_map[7] = 0x01; /* Use RDMA CM signature */
3803 *mgid = *(union ib_gid *) (mc_map + 4);
3804 } else {
3805 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3806 if (id_priv->id.ps == RDMA_PS_UDP)
3807 mc_map[7] = 0x01; /* Use RDMA CM signature */
3808 *mgid = *(union ib_gid *) (mc_map + 4);
3809 }
3810 }
3811
3812 static void cma_query_sa_classport_info_cb(int status,
3813 struct ib_class_port_info *rec,
3814 void *context)
3815 {
3816 struct class_port_info_context *cb_ctx = context;
3817
3818 WARN_ON(!context);
3819
3820 if (status || !rec) {
3821 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3822 cb_ctx->device->name, cb_ctx->port_num, status);
3823 goto out;
3824 }
3825
3826 memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3827
3828 out:
3829 complete(&cb_ctx->done);
3830 }
3831
3832 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3833 struct ib_class_port_info *class_port_info)
3834 {
3835 struct class_port_info_context *cb_ctx;
3836 int ret;
3837
3838 cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3839 if (!cb_ctx)
3840 return -ENOMEM;
3841
3842 cb_ctx->device = device;
3843 cb_ctx->class_port_info = class_port_info;
3844 cb_ctx->port_num = port_num;
3845 init_completion(&cb_ctx->done);
3846
3847 ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3848 CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3849 GFP_KERNEL, cma_query_sa_classport_info_cb,
3850 cb_ctx, &cb_ctx->sa_query);
3851 if (ret < 0) {
3852 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3853 device->name, port_num, ret);
3854 goto out;
3855 }
3856
3857 wait_for_completion(&cb_ctx->done);
3858
3859 out:
3860 kfree(cb_ctx);
3861 return ret;
3862 }
3863
3864 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3865 struct cma_multicast *mc)
3866 {
3867 struct ib_sa_mcmember_rec rec;
3868 struct ib_class_port_info class_port_info;
3869 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3870 ib_sa_comp_mask comp_mask;
3871 int ret;
3872
3873 ib_addr_get_mgid(dev_addr, &rec.mgid);
3874 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3875 &rec.mgid, &rec);
3876 if (ret)
3877 return ret;
3878
3879 ret = cma_set_qkey(id_priv, 0);
3880 if (ret)
3881 return ret;
3882
3883 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3884 rec.qkey = cpu_to_be32(id_priv->qkey);
3885 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3886 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3887 rec.join_state = mc->join_state;
3888
3889 if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3890 ret = cma_query_sa_classport_info(id_priv->id.device,
3891 id_priv->id.port_num,
3892 &class_port_info);
3893
3894 if (ret)
3895 return ret;
3896
3897 if (!(ib_get_cpi_capmask2(&class_port_info) &
3898 IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3899 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3900 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3901 id_priv->id.device->name, id_priv->id.port_num);
3902 return -EOPNOTSUPP;
3903 }
3904 }
3905
3906 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3907 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3908 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3909 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3910 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3911
3912 if (id_priv->id.ps == RDMA_PS_IPOIB)
3913 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3914 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3915 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3916 IB_SA_MCMEMBER_REC_MTU |
3917 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3918
3919 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3920 id_priv->id.port_num, &rec,
3921 comp_mask, GFP_KERNEL,
3922 cma_ib_mc_handler, mc);
3923 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3924 }
3925
3926 static void iboe_mcast_work_handler(struct work_struct *work)
3927 {
3928 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3929 struct cma_multicast *mc = mw->mc;
3930 struct ib_sa_multicast *m = mc->multicast.ib;
3931
3932 mc->multicast.ib->context = mc;
3933 cma_ib_mc_handler(0, m);
3934 kref_put(&mc->mcref, release_mc);
3935 kfree(mw);
3936 }
3937
3938 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3939 {
3940 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3941 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3942
3943 if (cma_any_addr(addr)) {
3944 memset(mgid, 0, sizeof *mgid);
3945 } else if (addr->sa_family == AF_INET6) {
3946 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3947 } else {
3948 mgid->raw[0] = 0xff;
3949 mgid->raw[1] = 0x0e;
3950 mgid->raw[2] = 0;
3951 mgid->raw[3] = 0;
3952 mgid->raw[4] = 0;
3953 mgid->raw[5] = 0;
3954 mgid->raw[6] = 0;
3955 mgid->raw[7] = 0;
3956 mgid->raw[8] = 0;
3957 mgid->raw[9] = 0;
3958 mgid->raw[10] = 0xff;
3959 mgid->raw[11] = 0xff;
3960 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3961 }
3962 }
3963
3964 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3965 struct cma_multicast *mc)
3966 {
3967 struct iboe_mcast_work *work;
3968 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3969 int err = 0;
3970 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3971 struct net_device *ndev = NULL;
3972 enum ib_gid_type gid_type;
3973 bool send_only;
3974
3975 send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
3976
3977 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3978 return -EINVAL;
3979
3980 work = kzalloc(sizeof *work, GFP_KERNEL);
3981 if (!work)
3982 return -ENOMEM;
3983
3984 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3985 if (!mc->multicast.ib) {
3986 err = -ENOMEM;
3987 goto out1;
3988 }
3989
3990 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3991
3992 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3993 if (id_priv->id.ps == RDMA_PS_UDP)
3994 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3995
3996 if (dev_addr->bound_dev_if)
3997 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3998 if (!ndev) {
3999 err = -ENODEV;
4000 goto out2;
4001 }
4002 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4003 mc->multicast.ib->rec.hop_limit = 1;
4004 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4005
4006 gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4007 rdma_start_port(id_priv->cma_dev->device)];
4008 if (addr->sa_family == AF_INET) {
4009 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4010 mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4011 if (!send_only) {
4012 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4013 true);
4014 if (!err)
4015 mc->igmp_joined = true;
4016 }
4017 }
4018 } else {
4019 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4020 err = -ENOTSUPP;
4021 }
4022 dev_put(ndev);
4023 if (err || !mc->multicast.ib->rec.mtu) {
4024 if (!err)
4025 err = -EINVAL;
4026 goto out2;
4027 }
4028 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4029 &mc->multicast.ib->rec.port_gid);
4030 work->id = id_priv;
4031 work->mc = mc;
4032 INIT_WORK(&work->work, iboe_mcast_work_handler);
4033 kref_get(&mc->mcref);
4034 queue_work(cma_wq, &work->work);
4035
4036 return 0;
4037
4038 out2:
4039 kfree(mc->multicast.ib);
4040 out1:
4041 kfree(work);
4042 return err;
4043 }
4044
4045 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4046 u8 join_state, void *context)
4047 {
4048 struct rdma_id_private *id_priv;
4049 struct cma_multicast *mc;
4050 int ret;
4051
4052 if (!id->device)
4053 return -EINVAL;
4054
4055 id_priv = container_of(id, struct rdma_id_private, id);
4056 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4057 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4058 return -EINVAL;
4059
4060 mc = kmalloc(sizeof *mc, GFP_KERNEL);
4061 if (!mc)
4062 return -ENOMEM;
4063
4064 memcpy(&mc->addr, addr, rdma_addr_size(addr));
4065 mc->context = context;
4066 mc->id_priv = id_priv;
4067 mc->igmp_joined = false;
4068 mc->join_state = join_state;
4069 spin_lock(&id_priv->lock);
4070 list_add(&mc->list, &id_priv->mc_list);
4071 spin_unlock(&id_priv->lock);
4072
4073 if (rdma_protocol_roce(id->device, id->port_num)) {
4074 kref_init(&mc->mcref);
4075 ret = cma_iboe_join_multicast(id_priv, mc);
4076 } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4077 ret = cma_join_ib_multicast(id_priv, mc);
4078 else
4079 ret = -ENOSYS;
4080
4081 if (ret) {
4082 spin_lock_irq(&id_priv->lock);
4083 list_del(&mc->list);
4084 spin_unlock_irq(&id_priv->lock);
4085 kfree(mc);
4086 }
4087 return ret;
4088 }
4089 EXPORT_SYMBOL(rdma_join_multicast);
4090
4091 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4092 {
4093 struct rdma_id_private *id_priv;
4094 struct cma_multicast *mc;
4095
4096 id_priv = container_of(id, struct rdma_id_private, id);
4097 spin_lock_irq(&id_priv->lock);
4098 list_for_each_entry(mc, &id_priv->mc_list, list) {
4099 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4100 list_del(&mc->list);
4101 spin_unlock_irq(&id_priv->lock);
4102
4103 if (id->qp)
4104 ib_detach_mcast(id->qp,
4105 &mc->multicast.ib->rec.mgid,
4106 be16_to_cpu(mc->multicast.ib->rec.mlid));
4107
4108 BUG_ON(id_priv->cma_dev->device != id->device);
4109
4110 if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4111 ib_sa_free_multicast(mc->multicast.ib);
4112 kfree(mc);
4113 } else if (rdma_protocol_roce(id->device, id->port_num)) {
4114 if (mc->igmp_joined) {
4115 struct rdma_dev_addr *dev_addr =
4116 &id->route.addr.dev_addr;
4117 struct net_device *ndev = NULL;
4118
4119 if (dev_addr->bound_dev_if)
4120 ndev = dev_get_by_index(&init_net,
4121 dev_addr->bound_dev_if);
4122 if (ndev) {
4123 cma_igmp_send(ndev,
4124 &mc->multicast.ib->rec.mgid,
4125 false);
4126 dev_put(ndev);
4127 }
4128 mc->igmp_joined = false;
4129 }
4130 kref_put(&mc->mcref, release_mc);
4131 }
4132 return;
4133 }
4134 }
4135 spin_unlock_irq(&id_priv->lock);
4136 }
4137 EXPORT_SYMBOL(rdma_leave_multicast);
4138
4139 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4140 {
4141 struct rdma_dev_addr *dev_addr;
4142 struct cma_ndev_work *work;
4143
4144 dev_addr = &id_priv->id.route.addr.dev_addr;
4145
4146 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4147 (net_eq(dev_net(ndev), dev_addr->net)) &&
4148 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4149 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4150 ndev->name, &id_priv->id);
4151 work = kzalloc(sizeof *work, GFP_KERNEL);
4152 if (!work)
4153 return -ENOMEM;
4154
4155 INIT_WORK(&work->work, cma_ndev_work_handler);
4156 work->id = id_priv;
4157 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4158 atomic_inc(&id_priv->refcount);
4159 queue_work(cma_wq, &work->work);
4160 }
4161
4162 return 0;
4163 }
4164
4165 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4166 void *ptr)
4167 {
4168 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4169 struct cma_device *cma_dev;
4170 struct rdma_id_private *id_priv;
4171 int ret = NOTIFY_DONE;
4172
4173 if (event != NETDEV_BONDING_FAILOVER)
4174 return NOTIFY_DONE;
4175
4176 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4177 return NOTIFY_DONE;
4178
4179 mutex_lock(&lock);
4180 list_for_each_entry(cma_dev, &dev_list, list)
4181 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4182 ret = cma_netdev_change(ndev, id_priv);
4183 if (ret)
4184 goto out;
4185 }
4186
4187 out:
4188 mutex_unlock(&lock);
4189 return ret;
4190 }
4191
4192 static struct notifier_block cma_nb = {
4193 .notifier_call = cma_netdev_callback
4194 };
4195
4196 static void cma_add_one(struct ib_device *device)
4197 {
4198 struct cma_device *cma_dev;
4199 struct rdma_id_private *id_priv;
4200 unsigned int i;
4201 unsigned long supported_gids = 0;
4202
4203 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4204 if (!cma_dev)
4205 return;
4206
4207 cma_dev->device = device;
4208 cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4209 sizeof(*cma_dev->default_gid_type),
4210 GFP_KERNEL);
4211 if (!cma_dev->default_gid_type) {
4212 kfree(cma_dev);
4213 return;
4214 }
4215 for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4216 supported_gids = roce_gid_type_mask_support(device, i);
4217 WARN_ON(!supported_gids);
4218 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4219 find_first_bit(&supported_gids, BITS_PER_LONG);
4220 }
4221
4222 init_completion(&cma_dev->comp);
4223 atomic_set(&cma_dev->refcount, 1);
4224 INIT_LIST_HEAD(&cma_dev->id_list);
4225 ib_set_client_data(device, &cma_client, cma_dev);
4226
4227 mutex_lock(&lock);
4228 list_add_tail(&cma_dev->list, &dev_list);
4229 list_for_each_entry(id_priv, &listen_any_list, list)
4230 cma_listen_on_dev(id_priv, cma_dev);
4231 mutex_unlock(&lock);
4232 }
4233
4234 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4235 {
4236 struct rdma_cm_event event;
4237 enum rdma_cm_state state;
4238 int ret = 0;
4239
4240 /* Record that we want to remove the device */
4241 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4242 if (state == RDMA_CM_DESTROYING)
4243 return 0;
4244
4245 cma_cancel_operation(id_priv, state);
4246 mutex_lock(&id_priv->handler_mutex);
4247
4248 /* Check for destruction from another callback. */
4249 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4250 goto out;
4251
4252 memset(&event, 0, sizeof event);
4253 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4254 ret = id_priv->id.event_handler(&id_priv->id, &event);
4255 out:
4256 mutex_unlock(&id_priv->handler_mutex);
4257 return ret;
4258 }
4259
4260 static void cma_process_remove(struct cma_device *cma_dev)
4261 {
4262 struct rdma_id_private *id_priv;
4263 int ret;
4264
4265 mutex_lock(&lock);
4266 while (!list_empty(&cma_dev->id_list)) {
4267 id_priv = list_entry(cma_dev->id_list.next,
4268 struct rdma_id_private, list);
4269
4270 list_del(&id_priv->listen_list);
4271 list_del_init(&id_priv->list);
4272 atomic_inc(&id_priv->refcount);
4273 mutex_unlock(&lock);
4274
4275 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4276 cma_deref_id(id_priv);
4277 if (ret)
4278 rdma_destroy_id(&id_priv->id);
4279
4280 mutex_lock(&lock);
4281 }
4282 mutex_unlock(&lock);
4283
4284 cma_deref_dev(cma_dev);
4285 wait_for_completion(&cma_dev->comp);
4286 }
4287
4288 static void cma_remove_one(struct ib_device *device, void *client_data)
4289 {
4290 struct cma_device *cma_dev = client_data;
4291
4292 if (!cma_dev)
4293 return;
4294
4295 mutex_lock(&lock);
4296 list_del(&cma_dev->list);
4297 mutex_unlock(&lock);
4298
4299 cma_process_remove(cma_dev);
4300 kfree(cma_dev->default_gid_type);
4301 kfree(cma_dev);
4302 }
4303
4304 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4305 {
4306 struct nlmsghdr *nlh;
4307 struct rdma_cm_id_stats *id_stats;
4308 struct rdma_id_private *id_priv;
4309 struct rdma_cm_id *id = NULL;
4310 struct cma_device *cma_dev;
4311 int i_dev = 0, i_id = 0;
4312
4313 /*
4314 * We export all of the IDs as a sequence of messages. Each
4315 * ID gets its own netlink message.
4316 */
4317 mutex_lock(&lock);
4318
4319 list_for_each_entry(cma_dev, &dev_list, list) {
4320 if (i_dev < cb->args[0]) {
4321 i_dev++;
4322 continue;
4323 }
4324
4325 i_id = 0;
4326 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4327 if (i_id < cb->args[1]) {
4328 i_id++;
4329 continue;
4330 }
4331
4332 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4333 sizeof *id_stats, RDMA_NL_RDMA_CM,
4334 RDMA_NL_RDMA_CM_ID_STATS,
4335 NLM_F_MULTI);
4336 if (!id_stats)
4337 goto out;
4338
4339 memset(id_stats, 0, sizeof *id_stats);
4340 id = &id_priv->id;
4341 id_stats->node_type = id->route.addr.dev_addr.dev_type;
4342 id_stats->port_num = id->port_num;
4343 id_stats->bound_dev_if =
4344 id->route.addr.dev_addr.bound_dev_if;
4345
4346 if (ibnl_put_attr(skb, nlh,
4347 rdma_addr_size(cma_src_addr(id_priv)),
4348 cma_src_addr(id_priv),
4349 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4350 goto out;
4351 if (ibnl_put_attr(skb, nlh,
4352 rdma_addr_size(cma_dst_addr(id_priv)),
4353 cma_dst_addr(id_priv),
4354 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4355 goto out;
4356
4357 id_stats->pid = id_priv->owner;
4358 id_stats->port_space = id->ps;
4359 id_stats->cm_state = id_priv->state;
4360 id_stats->qp_num = id_priv->qp_num;
4361 id_stats->qp_type = id->qp_type;
4362
4363 i_id++;
4364 }
4365
4366 cb->args[1] = 0;
4367 i_dev++;
4368 }
4369
4370 out:
4371 mutex_unlock(&lock);
4372 cb->args[0] = i_dev;
4373 cb->args[1] = i_id;
4374
4375 return skb->len;
4376 }
4377
4378 static const struct ibnl_client_cbs cma_cb_table[] = {
4379 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4380 .module = THIS_MODULE },
4381 };
4382
4383 static int cma_init_net(struct net *net)
4384 {
4385 struct cma_pernet *pernet = cma_pernet(net);
4386
4387 idr_init(&pernet->tcp_ps);
4388 idr_init(&pernet->udp_ps);
4389 idr_init(&pernet->ipoib_ps);
4390 idr_init(&pernet->ib_ps);
4391
4392 return 0;
4393 }
4394
4395 static void cma_exit_net(struct net *net)
4396 {
4397 struct cma_pernet *pernet = cma_pernet(net);
4398
4399 idr_destroy(&pernet->tcp_ps);
4400 idr_destroy(&pernet->udp_ps);
4401 idr_destroy(&pernet->ipoib_ps);
4402 idr_destroy(&pernet->ib_ps);
4403 }
4404
4405 static struct pernet_operations cma_pernet_operations = {
4406 .init = cma_init_net,
4407 .exit = cma_exit_net,
4408 .id = &cma_pernet_id,
4409 .size = sizeof(struct cma_pernet),
4410 };
4411
4412 static int __init cma_init(void)
4413 {
4414 int ret;
4415
4416 cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4417 if (!cma_wq)
4418 return -ENOMEM;
4419
4420 ret = register_pernet_subsys(&cma_pernet_operations);
4421 if (ret)
4422 goto err_wq;
4423
4424 ib_sa_register_client(&sa_client);
4425 rdma_addr_register_client(&addr_client);
4426 register_netdevice_notifier(&cma_nb);
4427
4428 ret = ib_register_client(&cma_client);
4429 if (ret)
4430 goto err;
4431
4432 if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4433 cma_cb_table))
4434 pr_warn("RDMA CMA: failed to add netlink callback\n");
4435 cma_configfs_init();
4436
4437 return 0;
4438
4439 err:
4440 unregister_netdevice_notifier(&cma_nb);
4441 rdma_addr_unregister_client(&addr_client);
4442 ib_sa_unregister_client(&sa_client);
4443 err_wq:
4444 destroy_workqueue(cma_wq);
4445 return ret;
4446 }
4447
4448 static void __exit cma_cleanup(void)
4449 {
4450 cma_configfs_exit();
4451 ibnl_remove_client(RDMA_NL_RDMA_CM);
4452 ib_unregister_client(&cma_client);
4453 unregister_netdevice_notifier(&cma_nb);
4454 rdma_addr_unregister_client(&addr_client);
4455 ib_sa_unregister_client(&sa_client);
4456 unregister_pernet_subsys(&cma_pernet_operations);
4457 destroy_workqueue(cma_wq);
4458 }
4459
4460 module_init(cma_init);
4461 module_exit(cma_cleanup);
Linux with added FPC-III support