projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
IB/mad: pass ib_mad_send_buf explicitly to the recv_handler
[deliverable/linux.git] / drivers / infiniband / core / cma.c
1 /*
2 * Copyright (c) 2005 Voltaire Inc. All rights reserved.
3 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
4 * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
5 * Copyright (c) 2005-2006 Intel Corporation. All rights reserved.
6 *
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 * SOFTWARE.
34 */
35
36 #include <linux/completion.h>
37 #include <linux/in.h>
38 #include <linux/in6.h>
39 #include <linux/mutex.h>
40 #include <linux/random.h>
41 #include <linux/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_MAX_CM_RETRIES 15
72 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
73 #define CMA_IBOE_PACKET_LIFETIME 18
74
75 static const char * const cma_events[] = {
76 [RDMA_CM_EVENT_ADDR_RESOLVED] = "address resolved",
77 [RDMA_CM_EVENT_ADDR_ERROR] = "address error",
78 [RDMA_CM_EVENT_ROUTE_RESOLVED] = "route resolved ",
79 [RDMA_CM_EVENT_ROUTE_ERROR] = "route error",
80 [RDMA_CM_EVENT_CONNECT_REQUEST] = "connect request",
81 [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
82 [RDMA_CM_EVENT_CONNECT_ERROR] = "connect error",
83 [RDMA_CM_EVENT_UNREACHABLE] = "unreachable",
84 [RDMA_CM_EVENT_REJECTED] = "rejected",
85 [RDMA_CM_EVENT_ESTABLISHED] = "established",
86 [RDMA_CM_EVENT_DISCONNECTED] = "disconnected",
87 [RDMA_CM_EVENT_DEVICE_REMOVAL] = "device removal",
88 [RDMA_CM_EVENT_MULTICAST_JOIN] = "multicast join",
89 [RDMA_CM_EVENT_MULTICAST_ERROR] = "multicast error",
90 [RDMA_CM_EVENT_ADDR_CHANGE] = "address change",
91 [RDMA_CM_EVENT_TIMEWAIT_EXIT] = "timewait exit",
92 };
93
94 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
95 {
96 size_t index = event;
97
98 return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
99 cma_events[index] : "unrecognized event";
100 }
101 EXPORT_SYMBOL(rdma_event_msg);
102
103 static void cma_add_one(struct ib_device *device);
104 static void cma_remove_one(struct ib_device *device, void *client_data);
105
106 static struct ib_client cma_client = {
107 .name = "cma",
108 .add = cma_add_one,
109 .remove = cma_remove_one
110 };
111
112 static struct ib_sa_client sa_client;
113 static struct rdma_addr_client addr_client;
114 static LIST_HEAD(dev_list);
115 static LIST_HEAD(listen_any_list);
116 static DEFINE_MUTEX(lock);
117 static struct workqueue_struct *cma_wq;
118 static int cma_pernet_id;
119
120 struct cma_pernet {
121 struct idr tcp_ps;
122 struct idr udp_ps;
123 struct idr ipoib_ps;
124 struct idr ib_ps;
125 };
126
127 static struct cma_pernet *cma_pernet(struct net *net)
128 {
129 return net_generic(net, cma_pernet_id);
130 }
131
132 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
133 {
134 struct cma_pernet *pernet = cma_pernet(net);
135
136 switch (ps) {
137 case RDMA_PS_TCP:
138 return &pernet->tcp_ps;
139 case RDMA_PS_UDP:
140 return &pernet->udp_ps;
141 case RDMA_PS_IPOIB:
142 return &pernet->ipoib_ps;
143 case RDMA_PS_IB:
144 return &pernet->ib_ps;
145 default:
146 return NULL;
147 }
148 }
149
150 struct cma_device {
151 struct list_head list;
152 struct ib_device *device;
153 struct completion comp;
154 atomic_t refcount;
155 struct list_head id_list;
156 enum ib_gid_type *default_gid_type;
157 };
158
159 struct rdma_bind_list {
160 enum rdma_port_space ps;
161 struct hlist_head owners;
162 unsigned short port;
163 };
164
165 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
166 struct rdma_bind_list *bind_list, int snum)
167 {
168 struct idr *idr = cma_pernet_idr(net, ps);
169
170 return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
171 }
172
173 static struct rdma_bind_list *cma_ps_find(struct net *net,
174 enum rdma_port_space ps, int snum)
175 {
176 struct idr *idr = cma_pernet_idr(net, ps);
177
178 return idr_find(idr, snum);
179 }
180
181 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
182 {
183 struct idr *idr = cma_pernet_idr(net, ps);
184
185 idr_remove(idr, snum);
186 }
187
188 enum {
189 CMA_OPTION_AFONLY,
190 };
191
192 void cma_ref_dev(struct cma_device *cma_dev)
193 {
194 atomic_inc(&cma_dev->refcount);
195 }
196
197 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter filter,
198 void *cookie)
199 {
200 struct cma_device *cma_dev;
201 struct cma_device *found_cma_dev = NULL;
202
203 mutex_lock(&lock);
204
205 list_for_each_entry(cma_dev, &dev_list, list)
206 if (filter(cma_dev->device, cookie)) {
207 found_cma_dev = cma_dev;
208 break;
209 }
210
211 if (found_cma_dev)
212 cma_ref_dev(found_cma_dev);
213 mutex_unlock(&lock);
214 return found_cma_dev;
215 }
216
217 int cma_get_default_gid_type(struct cma_device *cma_dev,
218 unsigned int port)
219 {
220 if (port < rdma_start_port(cma_dev->device) ||
221 port > rdma_end_port(cma_dev->device))
222 return -EINVAL;
223
224 return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
225 }
226
227 int cma_set_default_gid_type(struct cma_device *cma_dev,
228 unsigned int port,
229 enum ib_gid_type default_gid_type)
230 {
231 unsigned long supported_gids;
232
233 if (port < rdma_start_port(cma_dev->device) ||
234 port > rdma_end_port(cma_dev->device))
235 return -EINVAL;
236
237 supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
238
239 if (!(supported_gids & 1 << default_gid_type))
240 return -EINVAL;
241
242 cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
243 default_gid_type;
244
245 return 0;
246 }
247
248 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
249 {
250 return cma_dev->device;
251 }
252
253 /*
254 * Device removal can occur at anytime, so we need extra handling to
255 * serialize notifying the user of device removal with other callbacks.
256 * We do this by disabling removal notification while a callback is in process,
257 * and reporting it after the callback completes.
258 */
259 struct rdma_id_private {
260 struct rdma_cm_id id;
261
262 struct rdma_bind_list *bind_list;
263 struct hlist_node node;
264 struct list_head list; /* listen_any_list or cma_device.list */
265 struct list_head listen_list; /* per device listens */
266 struct cma_device *cma_dev;
267 struct list_head mc_list;
268
269 int internal_id;
270 enum rdma_cm_state state;
271 spinlock_t lock;
272 struct mutex qp_mutex;
273
274 struct completion comp;
275 atomic_t refcount;
276 struct mutex handler_mutex;
277
278 int backlog;
279 int timeout_ms;
280 struct ib_sa_query *query;
281 int query_id;
282 union {
283 struct ib_cm_id *ib;
284 struct iw_cm_id *iw;
285 } cm_id;
286
287 u32 seq_num;
288 u32 qkey;
289 u32 qp_num;
290 pid_t owner;
291 u32 options;
292 u8 srq;
293 u8 tos;
294 u8 reuseaddr;
295 u8 afonly;
296 enum ib_gid_type gid_type;
297 };
298
299 struct cma_multicast {
300 struct rdma_id_private *id_priv;
301 union {
302 struct ib_sa_multicast *ib;
303 } multicast;
304 struct list_head list;
305 void *context;
306 struct sockaddr_storage addr;
307 struct kref mcref;
308 bool igmp_joined;
309 };
310
311 struct cma_work {
312 struct work_struct work;
313 struct rdma_id_private *id;
314 enum rdma_cm_state old_state;
315 enum rdma_cm_state new_state;
316 struct rdma_cm_event event;
317 };
318
319 struct cma_ndev_work {
320 struct work_struct work;
321 struct rdma_id_private *id;
322 struct rdma_cm_event event;
323 };
324
325 struct iboe_mcast_work {
326 struct work_struct work;
327 struct rdma_id_private *id;
328 struct cma_multicast *mc;
329 };
330
331 union cma_ip_addr {
332 struct in6_addr ip6;
333 struct {
334 __be32 pad[3];
335 __be32 addr;
336 } ip4;
337 };
338
339 struct cma_hdr {
340 u8 cma_version;
341 u8 ip_version; /* IP version: 7:4 */
342 __be16 port;
343 union cma_ip_addr src_addr;
344 union cma_ip_addr dst_addr;
345 };
346
347 #define CMA_VERSION 0x00
348
349 struct cma_req_info {
350 struct ib_device *device;
351 int port;
352 union ib_gid local_gid;
353 __be64 service_id;
354 u16 pkey;
355 bool has_gid:1;
356 };
357
358 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
359 {
360 unsigned long flags;
361 int ret;
362
363 spin_lock_irqsave(&id_priv->lock, flags);
364 ret = (id_priv->state == comp);
365 spin_unlock_irqrestore(&id_priv->lock, flags);
366 return ret;
367 }
368
369 static int cma_comp_exch(struct rdma_id_private *id_priv,
370 enum rdma_cm_state comp, enum rdma_cm_state exch)
371 {
372 unsigned long flags;
373 int ret;
374
375 spin_lock_irqsave(&id_priv->lock, flags);
376 if ((ret = (id_priv->state == comp)))
377 id_priv->state = exch;
378 spin_unlock_irqrestore(&id_priv->lock, flags);
379 return ret;
380 }
381
382 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
383 enum rdma_cm_state exch)
384 {
385 unsigned long flags;
386 enum rdma_cm_state old;
387
388 spin_lock_irqsave(&id_priv->lock, flags);
389 old = id_priv->state;
390 id_priv->state = exch;
391 spin_unlock_irqrestore(&id_priv->lock, flags);
392 return old;
393 }
394
395 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
396 {
397 return hdr->ip_version >> 4;
398 }
399
400 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
401 {
402 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
403 }
404
405 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
406 {
407 struct in_device *in_dev = NULL;
408
409 if (ndev) {
410 rtnl_lock();
411 in_dev = __in_dev_get_rtnl(ndev);
412 if (in_dev) {
413 if (join)
414 ip_mc_inc_group(in_dev,
415 *(__be32 *)(mgid->raw + 12));
416 else
417 ip_mc_dec_group(in_dev,
418 *(__be32 *)(mgid->raw + 12));
419 }
420 rtnl_unlock();
421 }
422 return (in_dev) ? 0 : -ENODEV;
423 }
424
425 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
426 struct cma_device *cma_dev)
427 {
428 cma_ref_dev(cma_dev);
429 id_priv->cma_dev = cma_dev;
430 id_priv->gid_type = 0;
431 id_priv->id.device = cma_dev->device;
432 id_priv->id.route.addr.dev_addr.transport =
433 rdma_node_get_transport(cma_dev->device->node_type);
434 list_add_tail(&id_priv->list, &cma_dev->id_list);
435 }
436
437 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
438 struct cma_device *cma_dev)
439 {
440 _cma_attach_to_dev(id_priv, cma_dev);
441 id_priv->gid_type =
442 cma_dev->default_gid_type[id_priv->id.port_num -
443 rdma_start_port(cma_dev->device)];
444 }
445
446 void cma_deref_dev(struct cma_device *cma_dev)
447 {
448 if (atomic_dec_and_test(&cma_dev->refcount))
449 complete(&cma_dev->comp);
450 }
451
452 static inline void release_mc(struct kref *kref)
453 {
454 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
455
456 kfree(mc->multicast.ib);
457 kfree(mc);
458 }
459
460 static void cma_release_dev(struct rdma_id_private *id_priv)
461 {
462 mutex_lock(&lock);
463 list_del(&id_priv->list);
464 cma_deref_dev(id_priv->cma_dev);
465 id_priv->cma_dev = NULL;
466 mutex_unlock(&lock);
467 }
468
469 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
470 {
471 return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
472 }
473
474 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
475 {
476 return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
477 }
478
479 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
480 {
481 return id_priv->id.route.addr.src_addr.ss_family;
482 }
483
484 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
485 {
486 struct ib_sa_mcmember_rec rec;
487 int ret = 0;
488
489 if (id_priv->qkey) {
490 if (qkey && id_priv->qkey != qkey)
491 return -EINVAL;
492 return 0;
493 }
494
495 if (qkey) {
496 id_priv->qkey = qkey;
497 return 0;
498 }
499
500 switch (id_priv->id.ps) {
501 case RDMA_PS_UDP:
502 case RDMA_PS_IB:
503 id_priv->qkey = RDMA_UDP_QKEY;
504 break;
505 case RDMA_PS_IPOIB:
506 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
507 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
508 id_priv->id.port_num, &rec.mgid,
509 &rec);
510 if (!ret)
511 id_priv->qkey = be32_to_cpu(rec.qkey);
512 break;
513 default:
514 break;
515 }
516 return ret;
517 }
518
519 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
520 {
521 dev_addr->dev_type = ARPHRD_INFINIBAND;
522 rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
523 ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
524 }
525
526 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
527 {
528 int ret;
529
530 if (addr->sa_family != AF_IB) {
531 ret = rdma_translate_ip(addr, dev_addr, NULL);
532 } else {
533 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
534 ret = 0;
535 }
536
537 return ret;
538 }
539
540 static inline int cma_validate_port(struct ib_device *device, u8 port,
541 enum ib_gid_type gid_type,
542 union ib_gid *gid, int dev_type,
543 int bound_if_index)
544 {
545 int ret = -ENODEV;
546 struct net_device *ndev = NULL;
547
548 if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
549 return ret;
550
551 if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
552 return ret;
553
554 if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
555 ndev = dev_get_by_index(&init_net, bound_if_index);
556 if (ndev && ndev->flags & IFF_LOOPBACK) {
557 pr_info("detected loopback device\n");
558 dev_put(ndev);
559
560 if (!device->get_netdev)
561 return -EOPNOTSUPP;
562
563 ndev = device->get_netdev(device, port);
564 if (!ndev)
565 return -ENODEV;
566 }
567 } else {
568 gid_type = IB_GID_TYPE_IB;
569 }
570
571 ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
572 ndev, NULL);
573
574 if (ndev)
575 dev_put(ndev);
576
577 return ret;
578 }
579
580 static int cma_acquire_dev(struct rdma_id_private *id_priv,
581 struct rdma_id_private *listen_id_priv)
582 {
583 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
584 struct cma_device *cma_dev;
585 union ib_gid gid, iboe_gid, *gidp;
586 int ret = -ENODEV;
587 u8 port;
588
589 if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
590 id_priv->id.ps == RDMA_PS_IPOIB)
591 return -EINVAL;
592
593 mutex_lock(&lock);
594 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
595 &iboe_gid);
596
597 memcpy(&gid, dev_addr->src_dev_addr +
598 rdma_addr_gid_offset(dev_addr), sizeof gid);
599
600 if (listen_id_priv) {
601 cma_dev = listen_id_priv->cma_dev;
602 port = listen_id_priv->id.port_num;
603 gidp = rdma_protocol_roce(cma_dev->device, port) ?
604 &iboe_gid : &gid;
605
606 ret = cma_validate_port(cma_dev->device, port,
607 rdma_protocol_ib(cma_dev->device, port) ?
608 IB_GID_TYPE_IB :
609 listen_id_priv->gid_type, gidp,
610 dev_addr->dev_type,
611 dev_addr->bound_dev_if);
612 if (!ret) {
613 id_priv->id.port_num = port;
614 goto out;
615 }
616 }
617
618 list_for_each_entry(cma_dev, &dev_list, list) {
619 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
620 if (listen_id_priv &&
621 listen_id_priv->cma_dev == cma_dev &&
622 listen_id_priv->id.port_num == port)
623 continue;
624
625 gidp = rdma_protocol_roce(cma_dev->device, port) ?
626 &iboe_gid : &gid;
627
628 ret = cma_validate_port(cma_dev->device, port,
629 rdma_protocol_ib(cma_dev->device, port) ?
630 IB_GID_TYPE_IB :
631 cma_dev->default_gid_type[port - 1],
632 gidp, dev_addr->dev_type,
633 dev_addr->bound_dev_if);
634 if (!ret) {
635 id_priv->id.port_num = port;
636 goto out;
637 }
638 }
639 }
640
641 out:
642 if (!ret)
643 cma_attach_to_dev(id_priv, cma_dev);
644
645 mutex_unlock(&lock);
646 return ret;
647 }
648
649 /*
650 * Select the source IB device and address to reach the destination IB address.
651 */
652 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
653 {
654 struct cma_device *cma_dev, *cur_dev;
655 struct sockaddr_ib *addr;
656 union ib_gid gid, sgid, *dgid;
657 u16 pkey, index;
658 u8 p;
659 int i;
660
661 cma_dev = NULL;
662 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
663 dgid = (union ib_gid *) &addr->sib_addr;
664 pkey = ntohs(addr->sib_pkey);
665
666 list_for_each_entry(cur_dev, &dev_list, list) {
667 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
668 if (!rdma_cap_af_ib(cur_dev->device, p))
669 continue;
670
671 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
672 continue;
673
674 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
675 &gid, NULL);
676 i++) {
677 if (!memcmp(&gid, dgid, sizeof(gid))) {
678 cma_dev = cur_dev;
679 sgid = gid;
680 id_priv->id.port_num = p;
681 goto found;
682 }
683
684 if (!cma_dev && (gid.global.subnet_prefix ==
685 dgid->global.subnet_prefix)) {
686 cma_dev = cur_dev;
687 sgid = gid;
688 id_priv->id.port_num = p;
689 }
690 }
691 }
692 }
693
694 if (!cma_dev)
695 return -ENODEV;
696
697 found:
698 cma_attach_to_dev(id_priv, cma_dev);
699 addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
700 memcpy(&addr->sib_addr, &sgid, sizeof sgid);
701 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
702 return 0;
703 }
704
705 static void cma_deref_id(struct rdma_id_private *id_priv)
706 {
707 if (atomic_dec_and_test(&id_priv->refcount))
708 complete(&id_priv->comp);
709 }
710
711 static int cma_disable_callback(struct rdma_id_private *id_priv,
712 enum rdma_cm_state state)
713 {
714 mutex_lock(&id_priv->handler_mutex);
715 if (id_priv->state != state) {
716 mutex_unlock(&id_priv->handler_mutex);
717 return -EINVAL;
718 }
719 return 0;
720 }
721
722 struct rdma_cm_id *rdma_create_id(struct net *net,
723 rdma_cm_event_handler event_handler,
724 void *context, enum rdma_port_space ps,
725 enum ib_qp_type qp_type)
726 {
727 struct rdma_id_private *id_priv;
728
729 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
730 if (!id_priv)
731 return ERR_PTR(-ENOMEM);
732
733 id_priv->owner = task_pid_nr(current);
734 id_priv->state = RDMA_CM_IDLE;
735 id_priv->id.context = context;
736 id_priv->id.event_handler = event_handler;
737 id_priv->id.ps = ps;
738 id_priv->id.qp_type = qp_type;
739 spin_lock_init(&id_priv->lock);
740 mutex_init(&id_priv->qp_mutex);
741 init_completion(&id_priv->comp);
742 atomic_set(&id_priv->refcount, 1);
743 mutex_init(&id_priv->handler_mutex);
744 INIT_LIST_HEAD(&id_priv->listen_list);
745 INIT_LIST_HEAD(&id_priv->mc_list);
746 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
747 id_priv->id.route.addr.dev_addr.net = get_net(net);
748
749 return &id_priv->id;
750 }
751 EXPORT_SYMBOL(rdma_create_id);
752
753 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
754 {
755 struct ib_qp_attr qp_attr;
756 int qp_attr_mask, ret;
757
758 qp_attr.qp_state = IB_QPS_INIT;
759 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
760 if (ret)
761 return ret;
762
763 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
764 if (ret)
765 return ret;
766
767 qp_attr.qp_state = IB_QPS_RTR;
768 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
769 if (ret)
770 return ret;
771
772 qp_attr.qp_state = IB_QPS_RTS;
773 qp_attr.sq_psn = 0;
774 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
775
776 return ret;
777 }
778
779 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
780 {
781 struct ib_qp_attr qp_attr;
782 int qp_attr_mask, ret;
783
784 qp_attr.qp_state = IB_QPS_INIT;
785 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
786 if (ret)
787 return ret;
788
789 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
790 }
791
792 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
793 struct ib_qp_init_attr *qp_init_attr)
794 {
795 struct rdma_id_private *id_priv;
796 struct ib_qp *qp;
797 int ret;
798
799 id_priv = container_of(id, struct rdma_id_private, id);
800 if (id->device != pd->device)
801 return -EINVAL;
802
803 qp = ib_create_qp(pd, qp_init_attr);
804 if (IS_ERR(qp))
805 return PTR_ERR(qp);
806
807 if (id->qp_type == IB_QPT_UD)
808 ret = cma_init_ud_qp(id_priv, qp);
809 else
810 ret = cma_init_conn_qp(id_priv, qp);
811 if (ret)
812 goto err;
813
814 id->qp = qp;
815 id_priv->qp_num = qp->qp_num;
816 id_priv->srq = (qp->srq != NULL);
817 return 0;
818 err:
819 ib_destroy_qp(qp);
820 return ret;
821 }
822 EXPORT_SYMBOL(rdma_create_qp);
823
824 void rdma_destroy_qp(struct rdma_cm_id *id)
825 {
826 struct rdma_id_private *id_priv;
827
828 id_priv = container_of(id, struct rdma_id_private, id);
829 mutex_lock(&id_priv->qp_mutex);
830 ib_destroy_qp(id_priv->id.qp);
831 id_priv->id.qp = NULL;
832 mutex_unlock(&id_priv->qp_mutex);
833 }
834 EXPORT_SYMBOL(rdma_destroy_qp);
835
836 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
837 struct rdma_conn_param *conn_param)
838 {
839 struct ib_qp_attr qp_attr;
840 int qp_attr_mask, ret;
841 union ib_gid sgid;
842
843 mutex_lock(&id_priv->qp_mutex);
844 if (!id_priv->id.qp) {
845 ret = 0;
846 goto out;
847 }
848
849 /* Need to update QP attributes from default values. */
850 qp_attr.qp_state = IB_QPS_INIT;
851 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
852 if (ret)
853 goto out;
854
855 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
856 if (ret)
857 goto out;
858
859 qp_attr.qp_state = IB_QPS_RTR;
860 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
861 if (ret)
862 goto out;
863
864 ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
865 qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
866 if (ret)
867 goto out;
868
869 BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
870
871 if (conn_param)
872 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
873 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
874 out:
875 mutex_unlock(&id_priv->qp_mutex);
876 return ret;
877 }
878
879 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
880 struct rdma_conn_param *conn_param)
881 {
882 struct ib_qp_attr qp_attr;
883 int qp_attr_mask, ret;
884
885 mutex_lock(&id_priv->qp_mutex);
886 if (!id_priv->id.qp) {
887 ret = 0;
888 goto out;
889 }
890
891 qp_attr.qp_state = IB_QPS_RTS;
892 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
893 if (ret)
894 goto out;
895
896 if (conn_param)
897 qp_attr.max_rd_atomic = conn_param->initiator_depth;
898 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
899 out:
900 mutex_unlock(&id_priv->qp_mutex);
901 return ret;
902 }
903
904 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
905 {
906 struct ib_qp_attr qp_attr;
907 int ret;
908
909 mutex_lock(&id_priv->qp_mutex);
910 if (!id_priv->id.qp) {
911 ret = 0;
912 goto out;
913 }
914
915 qp_attr.qp_state = IB_QPS_ERR;
916 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
917 out:
918 mutex_unlock(&id_priv->qp_mutex);
919 return ret;
920 }
921
922 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
923 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
924 {
925 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
926 int ret;
927 u16 pkey;
928
929 if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
930 pkey = 0xffff;
931 else
932 pkey = ib_addr_get_pkey(dev_addr);
933
934 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
935 pkey, &qp_attr->pkey_index);
936 if (ret)
937 return ret;
938
939 qp_attr->port_num = id_priv->id.port_num;
940 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
941
942 if (id_priv->id.qp_type == IB_QPT_UD) {
943 ret = cma_set_qkey(id_priv, 0);
944 if (ret)
945 return ret;
946
947 qp_attr->qkey = id_priv->qkey;
948 *qp_attr_mask |= IB_QP_QKEY;
949 } else {
950 qp_attr->qp_access_flags = 0;
951 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
952 }
953 return 0;
954 }
955
956 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
957 int *qp_attr_mask)
958 {
959 struct rdma_id_private *id_priv;
960 int ret = 0;
961
962 id_priv = container_of(id, struct rdma_id_private, id);
963 if (rdma_cap_ib_cm(id->device, id->port_num)) {
964 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
965 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
966 else
967 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
968 qp_attr_mask);
969
970 if (qp_attr->qp_state == IB_QPS_RTR)
971 qp_attr->rq_psn = id_priv->seq_num;
972 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
973 if (!id_priv->cm_id.iw) {
974 qp_attr->qp_access_flags = 0;
975 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
976 } else
977 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
978 qp_attr_mask);
979 } else
980 ret = -ENOSYS;
981
982 return ret;
983 }
984 EXPORT_SYMBOL(rdma_init_qp_attr);
985
986 static inline int cma_zero_addr(struct sockaddr *addr)
987 {
988 switch (addr->sa_family) {
989 case AF_INET:
990 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
991 case AF_INET6:
992 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
993 case AF_IB:
994 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
995 default:
996 return 0;
997 }
998 }
999
1000 static inline int cma_loopback_addr(struct sockaddr *addr)
1001 {
1002 switch (addr->sa_family) {
1003 case AF_INET:
1004 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1005 case AF_INET6:
1006 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1007 case AF_IB:
1008 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1009 default:
1010 return 0;
1011 }
1012 }
1013
1014 static inline int cma_any_addr(struct sockaddr *addr)
1015 {
1016 return cma_zero_addr(addr) || cma_loopback_addr(addr);
1017 }
1018
1019 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1020 {
1021 if (src->sa_family != dst->sa_family)
1022 return -1;
1023
1024 switch (src->sa_family) {
1025 case AF_INET:
1026 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1027 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1028 case AF_INET6:
1029 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1030 &((struct sockaddr_in6 *) dst)->sin6_addr);
1031 default:
1032 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1033 &((struct sockaddr_ib *) dst)->sib_addr);
1034 }
1035 }
1036
1037 static __be16 cma_port(struct sockaddr *addr)
1038 {
1039 struct sockaddr_ib *sib;
1040
1041 switch (addr->sa_family) {
1042 case AF_INET:
1043 return ((struct sockaddr_in *) addr)->sin_port;
1044 case AF_INET6:
1045 return ((struct sockaddr_in6 *) addr)->sin6_port;
1046 case AF_IB:
1047 sib = (struct sockaddr_ib *) addr;
1048 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1049 be64_to_cpu(sib->sib_sid_mask)));
1050 default:
1051 return 0;
1052 }
1053 }
1054
1055 static inline int cma_any_port(struct sockaddr *addr)
1056 {
1057 return !cma_port(addr);
1058 }
1059
1060 static void cma_save_ib_info(struct sockaddr *src_addr,
1061 struct sockaddr *dst_addr,
1062 struct rdma_cm_id *listen_id,
1063 struct ib_sa_path_rec *path)
1064 {
1065 struct sockaddr_ib *listen_ib, *ib;
1066
1067 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1068 if (src_addr) {
1069 ib = (struct sockaddr_ib *)src_addr;
1070 ib->sib_family = AF_IB;
1071 if (path) {
1072 ib->sib_pkey = path->pkey;
1073 ib->sib_flowinfo = path->flow_label;
1074 memcpy(&ib->sib_addr, &path->sgid, 16);
1075 ib->sib_sid = path->service_id;
1076 ib->sib_scope_id = 0;
1077 } else {
1078 ib->sib_pkey = listen_ib->sib_pkey;
1079 ib->sib_flowinfo = listen_ib->sib_flowinfo;
1080 ib->sib_addr = listen_ib->sib_addr;
1081 ib->sib_sid = listen_ib->sib_sid;
1082 ib->sib_scope_id = listen_ib->sib_scope_id;
1083 }
1084 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1085 }
1086 if (dst_addr) {
1087 ib = (struct sockaddr_ib *)dst_addr;
1088 ib->sib_family = AF_IB;
1089 if (path) {
1090 ib->sib_pkey = path->pkey;
1091 ib->sib_flowinfo = path->flow_label;
1092 memcpy(&ib->sib_addr, &path->dgid, 16);
1093 }
1094 }
1095 }
1096
1097 static void cma_save_ip4_info(struct sockaddr *src_addr,
1098 struct sockaddr *dst_addr,
1099 struct cma_hdr *hdr,
1100 __be16 local_port)
1101 {
1102 struct sockaddr_in *ip4;
1103
1104 if (src_addr) {
1105 ip4 = (struct sockaddr_in *)src_addr;
1106 ip4->sin_family = AF_INET;
1107 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
1108 ip4->sin_port = local_port;
1109 }
1110
1111 if (dst_addr) {
1112 ip4 = (struct sockaddr_in *)dst_addr;
1113 ip4->sin_family = AF_INET;
1114 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
1115 ip4->sin_port = hdr->port;
1116 }
1117 }
1118
1119 static void cma_save_ip6_info(struct sockaddr *src_addr,
1120 struct sockaddr *dst_addr,
1121 struct cma_hdr *hdr,
1122 __be16 local_port)
1123 {
1124 struct sockaddr_in6 *ip6;
1125
1126 if (src_addr) {
1127 ip6 = (struct sockaddr_in6 *)src_addr;
1128 ip6->sin6_family = AF_INET6;
1129 ip6->sin6_addr = hdr->dst_addr.ip6;
1130 ip6->sin6_port = local_port;
1131 }
1132
1133 if (dst_addr) {
1134 ip6 = (struct sockaddr_in6 *)dst_addr;
1135 ip6->sin6_family = AF_INET6;
1136 ip6->sin6_addr = hdr->src_addr.ip6;
1137 ip6->sin6_port = hdr->port;
1138 }
1139 }
1140
1141 static u16 cma_port_from_service_id(__be64 service_id)
1142 {
1143 return (u16)be64_to_cpu(service_id);
1144 }
1145
1146 static int cma_save_ip_info(struct sockaddr *src_addr,
1147 struct sockaddr *dst_addr,
1148 struct ib_cm_event *ib_event,
1149 __be64 service_id)
1150 {
1151 struct cma_hdr *hdr;
1152 __be16 port;
1153
1154 hdr = ib_event->private_data;
1155 if (hdr->cma_version != CMA_VERSION)
1156 return -EINVAL;
1157
1158 port = htons(cma_port_from_service_id(service_id));
1159
1160 switch (cma_get_ip_ver(hdr)) {
1161 case 4:
1162 cma_save_ip4_info(src_addr, dst_addr, hdr, port);
1163 break;
1164 case 6:
1165 cma_save_ip6_info(src_addr, dst_addr, hdr, port);
1166 break;
1167 default:
1168 return -EAFNOSUPPORT;
1169 }
1170
1171 return 0;
1172 }
1173
1174 static int cma_save_net_info(struct sockaddr *src_addr,
1175 struct sockaddr *dst_addr,
1176 struct rdma_cm_id *listen_id,
1177 struct ib_cm_event *ib_event,
1178 sa_family_t sa_family, __be64 service_id)
1179 {
1180 if (sa_family == AF_IB) {
1181 if (ib_event->event == IB_CM_REQ_RECEIVED)
1182 cma_save_ib_info(src_addr, dst_addr, listen_id,
1183 ib_event->param.req_rcvd.primary_path);
1184 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1185 cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1186 return 0;
1187 }
1188
1189 return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1190 }
1191
1192 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1193 struct cma_req_info *req)
1194 {
1195 const struct ib_cm_req_event_param *req_param =
1196 &ib_event->param.req_rcvd;
1197 const struct ib_cm_sidr_req_event_param *sidr_param =
1198 &ib_event->param.sidr_req_rcvd;
1199
1200 switch (ib_event->event) {
1201 case IB_CM_REQ_RECEIVED:
1202 req->device = req_param->listen_id->device;
1203 req->port = req_param->port;
1204 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1205 sizeof(req->local_gid));
1206 req->has_gid = true;
1207 req->service_id = req_param->primary_path->service_id;
1208 req->pkey = be16_to_cpu(req_param->primary_path->pkey);
1209 break;
1210 case IB_CM_SIDR_REQ_RECEIVED:
1211 req->device = sidr_param->listen_id->device;
1212 req->port = sidr_param->port;
1213 req->has_gid = false;
1214 req->service_id = sidr_param->service_id;
1215 req->pkey = sidr_param->pkey;
1216 break;
1217 default:
1218 return -EINVAL;
1219 }
1220
1221 return 0;
1222 }
1223
1224 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1225 const struct sockaddr_in *dst_addr,
1226 const struct sockaddr_in *src_addr)
1227 {
1228 __be32 daddr = dst_addr->sin_addr.s_addr,
1229 saddr = src_addr->sin_addr.s_addr;
1230 struct fib_result res;
1231 struct flowi4 fl4;
1232 int err;
1233 bool ret;
1234
1235 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1236 ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1237 ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1238 ipv4_is_loopback(saddr))
1239 return false;
1240
1241 memset(&fl4, 0, sizeof(fl4));
1242 fl4.flowi4_iif = net_dev->ifindex;
1243 fl4.daddr = daddr;
1244 fl4.saddr = saddr;
1245
1246 rcu_read_lock();
1247 err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1248 ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1249 rcu_read_unlock();
1250
1251 return ret;
1252 }
1253
1254 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1255 const struct sockaddr_in6 *dst_addr,
1256 const struct sockaddr_in6 *src_addr)
1257 {
1258 #if IS_ENABLED(CONFIG_IPV6)
1259 const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1260 IPV6_ADDR_LINKLOCAL;
1261 struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1262 &src_addr->sin6_addr, net_dev->ifindex,
1263 strict);
1264 bool ret;
1265
1266 if (!rt)
1267 return false;
1268
1269 ret = rt->rt6i_idev->dev == net_dev;
1270 ip6_rt_put(rt);
1271
1272 return ret;
1273 #else
1274 return false;
1275 #endif
1276 }
1277
1278 static bool validate_net_dev(struct net_device *net_dev,
1279 const struct sockaddr *daddr,
1280 const struct sockaddr *saddr)
1281 {
1282 const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1283 const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1284 const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1285 const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1286
1287 switch (daddr->sa_family) {
1288 case AF_INET:
1289 return saddr->sa_family == AF_INET &&
1290 validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1291
1292 case AF_INET6:
1293 return saddr->sa_family == AF_INET6 &&
1294 validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1295
1296 default:
1297 return false;
1298 }
1299 }
1300
1301 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1302 const struct cma_req_info *req)
1303 {
1304 struct sockaddr_storage listen_addr_storage, src_addr_storage;
1305 struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1306 *src_addr = (struct sockaddr *)&src_addr_storage;
1307 struct net_device *net_dev;
1308 const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1309 int err;
1310
1311 err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1312 req->service_id);
1313 if (err)
1314 return ERR_PTR(err);
1315
1316 net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1317 gid, listen_addr);
1318 if (!net_dev)
1319 return ERR_PTR(-ENODEV);
1320
1321 if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1322 dev_put(net_dev);
1323 return ERR_PTR(-EHOSTUNREACH);
1324 }
1325
1326 return net_dev;
1327 }
1328
1329 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1330 {
1331 return (be64_to_cpu(service_id) >> 16) & 0xffff;
1332 }
1333
1334 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1335 const struct cma_hdr *hdr)
1336 {
1337 struct sockaddr *addr = cma_src_addr(id_priv);
1338 __be32 ip4_addr;
1339 struct in6_addr ip6_addr;
1340
1341 if (cma_any_addr(addr) && !id_priv->afonly)
1342 return true;
1343
1344 switch (addr->sa_family) {
1345 case AF_INET:
1346 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1347 if (cma_get_ip_ver(hdr) != 4)
1348 return false;
1349 if (!cma_any_addr(addr) &&
1350 hdr->dst_addr.ip4.addr != ip4_addr)
1351 return false;
1352 break;
1353 case AF_INET6:
1354 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1355 if (cma_get_ip_ver(hdr) != 6)
1356 return false;
1357 if (!cma_any_addr(addr) &&
1358 memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1359 return false;
1360 break;
1361 case AF_IB:
1362 return true;
1363 default:
1364 return false;
1365 }
1366
1367 return true;
1368 }
1369
1370 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1371 {
1372 enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1373 enum rdma_transport_type transport =
1374 rdma_node_get_transport(device->node_type);
1375
1376 return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1377 }
1378
1379 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1380 {
1381 struct ib_device *device = id->device;
1382 const int port_num = id->port_num ?: rdma_start_port(device);
1383
1384 return cma_protocol_roce_dev_port(device, port_num);
1385 }
1386
1387 static bool cma_match_net_dev(const struct rdma_id_private *id_priv,
1388 const struct net_device *net_dev)
1389 {
1390 const struct rdma_addr *addr = &id_priv->id.route.addr;
1391
1392 if (!net_dev)
1393 /* This request is an AF_IB request or a RoCE request */
1394 return addr->src_addr.ss_family == AF_IB ||
1395 cma_protocol_roce(&id_priv->id);
1396
1397 return !addr->dev_addr.bound_dev_if ||
1398 (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1399 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1400 }
1401
1402 static struct rdma_id_private *cma_find_listener(
1403 const struct rdma_bind_list *bind_list,
1404 const struct ib_cm_id *cm_id,
1405 const struct ib_cm_event *ib_event,
1406 const struct cma_req_info *req,
1407 const struct net_device *net_dev)
1408 {
1409 struct rdma_id_private *id_priv, *id_priv_dev;
1410
1411 if (!bind_list)
1412 return ERR_PTR(-EINVAL);
1413
1414 hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1415 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1416 if (id_priv->id.device == cm_id->device &&
1417 cma_match_net_dev(id_priv, net_dev))
1418 return id_priv;
1419 list_for_each_entry(id_priv_dev,
1420 &id_priv->listen_list,
1421 listen_list) {
1422 if (id_priv_dev->id.device == cm_id->device &&
1423 cma_match_net_dev(id_priv_dev, net_dev))
1424 return id_priv_dev;
1425 }
1426 }
1427 }
1428
1429 return ERR_PTR(-EINVAL);
1430 }
1431
1432 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1433 struct ib_cm_event *ib_event,
1434 struct net_device **net_dev)
1435 {
1436 struct cma_req_info req;
1437 struct rdma_bind_list *bind_list;
1438 struct rdma_id_private *id_priv;
1439 int err;
1440
1441 err = cma_save_req_info(ib_event, &req);
1442 if (err)
1443 return ERR_PTR(err);
1444
1445 *net_dev = cma_get_net_dev(ib_event, &req);
1446 if (IS_ERR(*net_dev)) {
1447 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1448 /* Assuming the protocol is AF_IB */
1449 *net_dev = NULL;
1450 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1451 /* TODO find the net dev matching the request parameters
1452 * through the RoCE GID table */
1453 *net_dev = NULL;
1454 } else {
1455 return ERR_CAST(*net_dev);
1456 }
1457 }
1458
1459 bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1460 rdma_ps_from_service_id(req.service_id),
1461 cma_port_from_service_id(req.service_id));
1462 id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1463 if (IS_ERR(id_priv) && *net_dev) {
1464 dev_put(*net_dev);
1465 *net_dev = NULL;
1466 }
1467
1468 return id_priv;
1469 }
1470
1471 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1472 {
1473 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1474 }
1475
1476 static void cma_cancel_route(struct rdma_id_private *id_priv)
1477 {
1478 if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1479 if (id_priv->query)
1480 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1481 }
1482 }
1483
1484 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1485 {
1486 struct rdma_id_private *dev_id_priv;
1487
1488 /*
1489 * Remove from listen_any_list to prevent added devices from spawning
1490 * additional listen requests.
1491 */
1492 mutex_lock(&lock);
1493 list_del(&id_priv->list);
1494
1495 while (!list_empty(&id_priv->listen_list)) {
1496 dev_id_priv = list_entry(id_priv->listen_list.next,
1497 struct rdma_id_private, listen_list);
1498 /* sync with device removal to avoid duplicate destruction */
1499 list_del_init(&dev_id_priv->list);
1500 list_del(&dev_id_priv->listen_list);
1501 mutex_unlock(&lock);
1502
1503 rdma_destroy_id(&dev_id_priv->id);
1504 mutex_lock(&lock);
1505 }
1506 mutex_unlock(&lock);
1507 }
1508
1509 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1510 enum rdma_cm_state state)
1511 {
1512 switch (state) {
1513 case RDMA_CM_ADDR_QUERY:
1514 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1515 break;
1516 case RDMA_CM_ROUTE_QUERY:
1517 cma_cancel_route(id_priv);
1518 break;
1519 case RDMA_CM_LISTEN:
1520 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1521 cma_cancel_listens(id_priv);
1522 break;
1523 default:
1524 break;
1525 }
1526 }
1527
1528 static void cma_release_port(struct rdma_id_private *id_priv)
1529 {
1530 struct rdma_bind_list *bind_list = id_priv->bind_list;
1531 struct net *net = id_priv->id.route.addr.dev_addr.net;
1532
1533 if (!bind_list)
1534 return;
1535
1536 mutex_lock(&lock);
1537 hlist_del(&id_priv->node);
1538 if (hlist_empty(&bind_list->owners)) {
1539 cma_ps_remove(net, bind_list->ps, bind_list->port);
1540 kfree(bind_list);
1541 }
1542 mutex_unlock(&lock);
1543 }
1544
1545 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1546 {
1547 struct cma_multicast *mc;
1548
1549 while (!list_empty(&id_priv->mc_list)) {
1550 mc = container_of(id_priv->mc_list.next,
1551 struct cma_multicast, list);
1552 list_del(&mc->list);
1553 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1554 id_priv->id.port_num)) {
1555 ib_sa_free_multicast(mc->multicast.ib);
1556 kfree(mc);
1557 } else {
1558 if (mc->igmp_joined) {
1559 struct rdma_dev_addr *dev_addr =
1560 &id_priv->id.route.addr.dev_addr;
1561 struct net_device *ndev = NULL;
1562
1563 if (dev_addr->bound_dev_if)
1564 ndev = dev_get_by_index(&init_net,
1565 dev_addr->bound_dev_if);
1566 if (ndev) {
1567 cma_igmp_send(ndev,
1568 &mc->multicast.ib->rec.mgid,
1569 false);
1570 dev_put(ndev);
1571 }
1572 }
1573 kref_put(&mc->mcref, release_mc);
1574 }
1575 }
1576 }
1577
1578 void rdma_destroy_id(struct rdma_cm_id *id)
1579 {
1580 struct rdma_id_private *id_priv;
1581 enum rdma_cm_state state;
1582
1583 id_priv = container_of(id, struct rdma_id_private, id);
1584 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1585 cma_cancel_operation(id_priv, state);
1586
1587 /*
1588 * Wait for any active callback to finish. New callbacks will find
1589 * the id_priv state set to destroying and abort.
1590 */
1591 mutex_lock(&id_priv->handler_mutex);
1592 mutex_unlock(&id_priv->handler_mutex);
1593
1594 if (id_priv->cma_dev) {
1595 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1596 if (id_priv->cm_id.ib)
1597 ib_destroy_cm_id(id_priv->cm_id.ib);
1598 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1599 if (id_priv->cm_id.iw)
1600 iw_destroy_cm_id(id_priv->cm_id.iw);
1601 }
1602 cma_leave_mc_groups(id_priv);
1603 cma_release_dev(id_priv);
1604 }
1605
1606 cma_release_port(id_priv);
1607 cma_deref_id(id_priv);
1608 wait_for_completion(&id_priv->comp);
1609
1610 if (id_priv->internal_id)
1611 cma_deref_id(id_priv->id.context);
1612
1613 kfree(id_priv->id.route.path_rec);
1614 put_net(id_priv->id.route.addr.dev_addr.net);
1615 kfree(id_priv);
1616 }
1617 EXPORT_SYMBOL(rdma_destroy_id);
1618
1619 static int cma_rep_recv(struct rdma_id_private *id_priv)
1620 {
1621 int ret;
1622
1623 ret = cma_modify_qp_rtr(id_priv, NULL);
1624 if (ret)
1625 goto reject;
1626
1627 ret = cma_modify_qp_rts(id_priv, NULL);
1628 if (ret)
1629 goto reject;
1630
1631 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1632 if (ret)
1633 goto reject;
1634
1635 return 0;
1636 reject:
1637 cma_modify_qp_err(id_priv);
1638 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1639 NULL, 0, NULL, 0);
1640 return ret;
1641 }
1642
1643 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1644 struct ib_cm_rep_event_param *rep_data,
1645 void *private_data)
1646 {
1647 event->param.conn.private_data = private_data;
1648 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1649 event->param.conn.responder_resources = rep_data->responder_resources;
1650 event->param.conn.initiator_depth = rep_data->initiator_depth;
1651 event->param.conn.flow_control = rep_data->flow_control;
1652 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1653 event->param.conn.srq = rep_data->srq;
1654 event->param.conn.qp_num = rep_data->remote_qpn;
1655 }
1656
1657 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1658 {
1659 struct rdma_id_private *id_priv = cm_id->context;
1660 struct rdma_cm_event event;
1661 int ret = 0;
1662
1663 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1664 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1665 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1666 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1667 return 0;
1668
1669 memset(&event, 0, sizeof event);
1670 switch (ib_event->event) {
1671 case IB_CM_REQ_ERROR:
1672 case IB_CM_REP_ERROR:
1673 event.event = RDMA_CM_EVENT_UNREACHABLE;
1674 event.status = -ETIMEDOUT;
1675 break;
1676 case IB_CM_REP_RECEIVED:
1677 if (id_priv->id.qp) {
1678 event.status = cma_rep_recv(id_priv);
1679 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1680 RDMA_CM_EVENT_ESTABLISHED;
1681 } else {
1682 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1683 }
1684 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1685 ib_event->private_data);
1686 break;
1687 case IB_CM_RTU_RECEIVED:
1688 case IB_CM_USER_ESTABLISHED:
1689 event.event = RDMA_CM_EVENT_ESTABLISHED;
1690 break;
1691 case IB_CM_DREQ_ERROR:
1692 event.status = -ETIMEDOUT; /* fall through */
1693 case IB_CM_DREQ_RECEIVED:
1694 case IB_CM_DREP_RECEIVED:
1695 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1696 RDMA_CM_DISCONNECT))
1697 goto out;
1698 event.event = RDMA_CM_EVENT_DISCONNECTED;
1699 break;
1700 case IB_CM_TIMEWAIT_EXIT:
1701 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1702 break;
1703 case IB_CM_MRA_RECEIVED:
1704 /* ignore event */
1705 goto out;
1706 case IB_CM_REJ_RECEIVED:
1707 cma_modify_qp_err(id_priv);
1708 event.status = ib_event->param.rej_rcvd.reason;
1709 event.event = RDMA_CM_EVENT_REJECTED;
1710 event.param.conn.private_data = ib_event->private_data;
1711 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1712 break;
1713 default:
1714 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1715 ib_event->event);
1716 goto out;
1717 }
1718
1719 ret = id_priv->id.event_handler(&id_priv->id, &event);
1720 if (ret) {
1721 /* Destroy the CM ID by returning a non-zero value. */
1722 id_priv->cm_id.ib = NULL;
1723 cma_exch(id_priv, RDMA_CM_DESTROYING);
1724 mutex_unlock(&id_priv->handler_mutex);
1725 rdma_destroy_id(&id_priv->id);
1726 return ret;
1727 }
1728 out:
1729 mutex_unlock(&id_priv->handler_mutex);
1730 return ret;
1731 }
1732
1733 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1734 struct ib_cm_event *ib_event,
1735 struct net_device *net_dev)
1736 {
1737 struct rdma_id_private *id_priv;
1738 struct rdma_cm_id *id;
1739 struct rdma_route *rt;
1740 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1741 const __be64 service_id =
1742 ib_event->param.req_rcvd.primary_path->service_id;
1743 int ret;
1744
1745 id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1746 listen_id->event_handler, listen_id->context,
1747 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1748 if (IS_ERR(id))
1749 return NULL;
1750
1751 id_priv = container_of(id, struct rdma_id_private, id);
1752 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1753 (struct sockaddr *)&id->route.addr.dst_addr,
1754 listen_id, ib_event, ss_family, service_id))
1755 goto err;
1756
1757 rt = &id->route;
1758 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1759 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1760 GFP_KERNEL);
1761 if (!rt->path_rec)
1762 goto err;
1763
1764 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1765 if (rt->num_paths == 2)
1766 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1767
1768 if (net_dev) {
1769 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1770 if (ret)
1771 goto err;
1772 } else {
1773 if (!cma_protocol_roce(listen_id) &&
1774 cma_any_addr(cma_src_addr(id_priv))) {
1775 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1776 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1777 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1778 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1779 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1780 if (ret)
1781 goto err;
1782 }
1783 }
1784 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1785
1786 id_priv->state = RDMA_CM_CONNECT;
1787 return id_priv;
1788
1789 err:
1790 rdma_destroy_id(id);
1791 return NULL;
1792 }
1793
1794 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1795 struct ib_cm_event *ib_event,
1796 struct net_device *net_dev)
1797 {
1798 struct rdma_id_private *id_priv;
1799 struct rdma_cm_id *id;
1800 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1801 struct net *net = listen_id->route.addr.dev_addr.net;
1802 int ret;
1803
1804 id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1805 listen_id->ps, IB_QPT_UD);
1806 if (IS_ERR(id))
1807 return NULL;
1808
1809 id_priv = container_of(id, struct rdma_id_private, id);
1810 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1811 (struct sockaddr *)&id->route.addr.dst_addr,
1812 listen_id, ib_event, ss_family,
1813 ib_event->param.sidr_req_rcvd.service_id))
1814 goto err;
1815
1816 if (net_dev) {
1817 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1818 if (ret)
1819 goto err;
1820 } else {
1821 if (!cma_any_addr(cma_src_addr(id_priv))) {
1822 ret = cma_translate_addr(cma_src_addr(id_priv),
1823 &id->route.addr.dev_addr);
1824 if (ret)
1825 goto err;
1826 }
1827 }
1828
1829 id_priv->state = RDMA_CM_CONNECT;
1830 return id_priv;
1831 err:
1832 rdma_destroy_id(id);
1833 return NULL;
1834 }
1835
1836 static void cma_set_req_event_data(struct rdma_cm_event *event,
1837 struct ib_cm_req_event_param *req_data,
1838 void *private_data, int offset)
1839 {
1840 event->param.conn.private_data = private_data + offset;
1841 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1842 event->param.conn.responder_resources = req_data->responder_resources;
1843 event->param.conn.initiator_depth = req_data->initiator_depth;
1844 event->param.conn.flow_control = req_data->flow_control;
1845 event->param.conn.retry_count = req_data->retry_count;
1846 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1847 event->param.conn.srq = req_data->srq;
1848 event->param.conn.qp_num = req_data->remote_qpn;
1849 }
1850
1851 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1852 {
1853 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1854 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1855 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1856 (id->qp_type == IB_QPT_UD)) ||
1857 (!id->qp_type));
1858 }
1859
1860 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1861 {
1862 struct rdma_id_private *listen_id, *conn_id;
1863 struct rdma_cm_event event;
1864 struct net_device *net_dev;
1865 int offset, ret;
1866
1867 listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1868 if (IS_ERR(listen_id))
1869 return PTR_ERR(listen_id);
1870
1871 if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1872 ret = -EINVAL;
1873 goto net_dev_put;
1874 }
1875
1876 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) {
1877 ret = -ECONNABORTED;
1878 goto net_dev_put;
1879 }
1880
1881 memset(&event, 0, sizeof event);
1882 offset = cma_user_data_offset(listen_id);
1883 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1884 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1885 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1886 event.param.ud.private_data = ib_event->private_data + offset;
1887 event.param.ud.private_data_len =
1888 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1889 } else {
1890 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1891 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1892 ib_event->private_data, offset);
1893 }
1894 if (!conn_id) {
1895 ret = -ENOMEM;
1896 goto err1;
1897 }
1898
1899 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1900 ret = cma_acquire_dev(conn_id, listen_id);
1901 if (ret)
1902 goto err2;
1903
1904 conn_id->cm_id.ib = cm_id;
1905 cm_id->context = conn_id;
1906 cm_id->cm_handler = cma_ib_handler;
1907
1908 /*
1909 * Protect against the user destroying conn_id from another thread
1910 * until we're done accessing it.
1911 */
1912 atomic_inc(&conn_id->refcount);
1913 ret = conn_id->id.event_handler(&conn_id->id, &event);
1914 if (ret)
1915 goto err3;
1916 /*
1917 * Acquire mutex to prevent user executing rdma_destroy_id()
1918 * while we're accessing the cm_id.
1919 */
1920 mutex_lock(&lock);
1921 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1922 (conn_id->id.qp_type != IB_QPT_UD))
1923 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1924 mutex_unlock(&lock);
1925 mutex_unlock(&conn_id->handler_mutex);
1926 mutex_unlock(&listen_id->handler_mutex);
1927 cma_deref_id(conn_id);
1928 if (net_dev)
1929 dev_put(net_dev);
1930 return 0;
1931
1932 err3:
1933 cma_deref_id(conn_id);
1934 /* Destroy the CM ID by returning a non-zero value. */
1935 conn_id->cm_id.ib = NULL;
1936 err2:
1937 cma_exch(conn_id, RDMA_CM_DESTROYING);
1938 mutex_unlock(&conn_id->handler_mutex);
1939 err1:
1940 mutex_unlock(&listen_id->handler_mutex);
1941 if (conn_id)
1942 rdma_destroy_id(&conn_id->id);
1943
1944 net_dev_put:
1945 if (net_dev)
1946 dev_put(net_dev);
1947
1948 return ret;
1949 }
1950
1951 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1952 {
1953 if (addr->sa_family == AF_IB)
1954 return ((struct sockaddr_ib *) addr)->sib_sid;
1955
1956 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1957 }
1958 EXPORT_SYMBOL(rdma_get_service_id);
1959
1960 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1961 {
1962 struct rdma_id_private *id_priv = iw_id->context;
1963 struct rdma_cm_event event;
1964 int ret = 0;
1965 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1966 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1967
1968 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1969 return 0;
1970
1971 memset(&event, 0, sizeof event);
1972 switch (iw_event->event) {
1973 case IW_CM_EVENT_CLOSE:
1974 event.event = RDMA_CM_EVENT_DISCONNECTED;
1975 break;
1976 case IW_CM_EVENT_CONNECT_REPLY:
1977 memcpy(cma_src_addr(id_priv), laddr,
1978 rdma_addr_size(laddr));
1979 memcpy(cma_dst_addr(id_priv), raddr,
1980 rdma_addr_size(raddr));
1981 switch (iw_event->status) {
1982 case 0:
1983 event.event = RDMA_CM_EVENT_ESTABLISHED;
1984 event.param.conn.initiator_depth = iw_event->ird;
1985 event.param.conn.responder_resources = iw_event->ord;
1986 break;
1987 case -ECONNRESET:
1988 case -ECONNREFUSED:
1989 event.event = RDMA_CM_EVENT_REJECTED;
1990 break;
1991 case -ETIMEDOUT:
1992 event.event = RDMA_CM_EVENT_UNREACHABLE;
1993 break;
1994 default:
1995 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1996 break;
1997 }
1998 break;
1999 case IW_CM_EVENT_ESTABLISHED:
2000 event.event = RDMA_CM_EVENT_ESTABLISHED;
2001 event.param.conn.initiator_depth = iw_event->ird;
2002 event.param.conn.responder_resources = iw_event->ord;
2003 break;
2004 default:
2005 BUG_ON(1);
2006 }
2007
2008 event.status = iw_event->status;
2009 event.param.conn.private_data = iw_event->private_data;
2010 event.param.conn.private_data_len = iw_event->private_data_len;
2011 ret = id_priv->id.event_handler(&id_priv->id, &event);
2012 if (ret) {
2013 /* Destroy the CM ID by returning a non-zero value. */
2014 id_priv->cm_id.iw = NULL;
2015 cma_exch(id_priv, RDMA_CM_DESTROYING);
2016 mutex_unlock(&id_priv->handler_mutex);
2017 rdma_destroy_id(&id_priv->id);
2018 return ret;
2019 }
2020
2021 mutex_unlock(&id_priv->handler_mutex);
2022 return ret;
2023 }
2024
2025 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2026 struct iw_cm_event *iw_event)
2027 {
2028 struct rdma_cm_id *new_cm_id;
2029 struct rdma_id_private *listen_id, *conn_id;
2030 struct rdma_cm_event event;
2031 int ret;
2032 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2033 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2034
2035 listen_id = cm_id->context;
2036 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
2037 return -ECONNABORTED;
2038
2039 /* Create a new RDMA id for the new IW CM ID */
2040 new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2041 listen_id->id.event_handler,
2042 listen_id->id.context,
2043 RDMA_PS_TCP, IB_QPT_RC);
2044 if (IS_ERR(new_cm_id)) {
2045 ret = -ENOMEM;
2046 goto out;
2047 }
2048 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2049 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2050 conn_id->state = RDMA_CM_CONNECT;
2051
2052 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2053 if (ret) {
2054 mutex_unlock(&conn_id->handler_mutex);
2055 rdma_destroy_id(new_cm_id);
2056 goto out;
2057 }
2058
2059 ret = cma_acquire_dev(conn_id, listen_id);
2060 if (ret) {
2061 mutex_unlock(&conn_id->handler_mutex);
2062 rdma_destroy_id(new_cm_id);
2063 goto out;
2064 }
2065
2066 conn_id->cm_id.iw = cm_id;
2067 cm_id->context = conn_id;
2068 cm_id->cm_handler = cma_iw_handler;
2069
2070 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2071 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2072
2073 memset(&event, 0, sizeof event);
2074 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2075 event.param.conn.private_data = iw_event->private_data;
2076 event.param.conn.private_data_len = iw_event->private_data_len;
2077 event.param.conn.initiator_depth = iw_event->ird;
2078 event.param.conn.responder_resources = iw_event->ord;
2079
2080 /*
2081 * Protect against the user destroying conn_id from another thread
2082 * until we're done accessing it.
2083 */
2084 atomic_inc(&conn_id->refcount);
2085 ret = conn_id->id.event_handler(&conn_id->id, &event);
2086 if (ret) {
2087 /* User wants to destroy the CM ID */
2088 conn_id->cm_id.iw = NULL;
2089 cma_exch(conn_id, RDMA_CM_DESTROYING);
2090 mutex_unlock(&conn_id->handler_mutex);
2091 cma_deref_id(conn_id);
2092 rdma_destroy_id(&conn_id->id);
2093 goto out;
2094 }
2095
2096 mutex_unlock(&conn_id->handler_mutex);
2097 cma_deref_id(conn_id);
2098
2099 out:
2100 mutex_unlock(&listen_id->handler_mutex);
2101 return ret;
2102 }
2103
2104 static int cma_ib_listen(struct rdma_id_private *id_priv)
2105 {
2106 struct sockaddr *addr;
2107 struct ib_cm_id *id;
2108 __be64 svc_id;
2109
2110 addr = cma_src_addr(id_priv);
2111 svc_id = rdma_get_service_id(&id_priv->id, addr);
2112 id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2113 if (IS_ERR(id))
2114 return PTR_ERR(id);
2115 id_priv->cm_id.ib = id;
2116
2117 return 0;
2118 }
2119
2120 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2121 {
2122 int ret;
2123 struct iw_cm_id *id;
2124
2125 id = iw_create_cm_id(id_priv->id.device,
2126 iw_conn_req_handler,
2127 id_priv);
2128 if (IS_ERR(id))
2129 return PTR_ERR(id);
2130
2131 id->tos = id_priv->tos;
2132 id_priv->cm_id.iw = id;
2133
2134 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2135 rdma_addr_size(cma_src_addr(id_priv)));
2136
2137 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2138
2139 if (ret) {
2140 iw_destroy_cm_id(id_priv->cm_id.iw);
2141 id_priv->cm_id.iw = NULL;
2142 }
2143
2144 return ret;
2145 }
2146
2147 static int cma_listen_handler(struct rdma_cm_id *id,
2148 struct rdma_cm_event *event)
2149 {
2150 struct rdma_id_private *id_priv = id->context;
2151
2152 id->context = id_priv->id.context;
2153 id->event_handler = id_priv->id.event_handler;
2154 return id_priv->id.event_handler(id, event);
2155 }
2156
2157 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2158 struct cma_device *cma_dev)
2159 {
2160 struct rdma_id_private *dev_id_priv;
2161 struct rdma_cm_id *id;
2162 struct net *net = id_priv->id.route.addr.dev_addr.net;
2163 int ret;
2164
2165 if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2166 return;
2167
2168 id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2169 id_priv->id.qp_type);
2170 if (IS_ERR(id))
2171 return;
2172
2173 dev_id_priv = container_of(id, struct rdma_id_private, id);
2174
2175 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2176 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2177 rdma_addr_size(cma_src_addr(id_priv)));
2178
2179 _cma_attach_to_dev(dev_id_priv, cma_dev);
2180 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2181 atomic_inc(&id_priv->refcount);
2182 dev_id_priv->internal_id = 1;
2183 dev_id_priv->afonly = id_priv->afonly;
2184
2185 ret = rdma_listen(id, id_priv->backlog);
2186 if (ret)
2187 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
2188 "listening on device %s\n", ret, cma_dev->device->name);
2189 }
2190
2191 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2192 {
2193 struct cma_device *cma_dev;
2194
2195 mutex_lock(&lock);
2196 list_add_tail(&id_priv->list, &listen_any_list);
2197 list_for_each_entry(cma_dev, &dev_list, list)
2198 cma_listen_on_dev(id_priv, cma_dev);
2199 mutex_unlock(&lock);
2200 }
2201
2202 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2203 {
2204 struct rdma_id_private *id_priv;
2205
2206 id_priv = container_of(id, struct rdma_id_private, id);
2207 id_priv->tos = (u8) tos;
2208 }
2209 EXPORT_SYMBOL(rdma_set_service_type);
2210
2211 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2212 void *context)
2213 {
2214 struct cma_work *work = context;
2215 struct rdma_route *route;
2216
2217 route = &work->id->id.route;
2218
2219 if (!status) {
2220 route->num_paths = 1;
2221 *route->path_rec = *path_rec;
2222 } else {
2223 work->old_state = RDMA_CM_ROUTE_QUERY;
2224 work->new_state = RDMA_CM_ADDR_RESOLVED;
2225 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2226 work->event.status = status;
2227 }
2228
2229 queue_work(cma_wq, &work->work);
2230 }
2231
2232 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2233 struct cma_work *work)
2234 {
2235 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2236 struct ib_sa_path_rec path_rec;
2237 ib_sa_comp_mask comp_mask;
2238 struct sockaddr_in6 *sin6;
2239 struct sockaddr_ib *sib;
2240
2241 memset(&path_rec, 0, sizeof path_rec);
2242 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2243 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2244 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2245 path_rec.numb_path = 1;
2246 path_rec.reversible = 1;
2247 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2248
2249 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2250 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2251 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2252
2253 switch (cma_family(id_priv)) {
2254 case AF_INET:
2255 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2256 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2257 break;
2258 case AF_INET6:
2259 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2260 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2261 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2262 break;
2263 case AF_IB:
2264 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2265 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2266 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2267 break;
2268 }
2269
2270 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2271 id_priv->id.port_num, &path_rec,
2272 comp_mask, timeout_ms,
2273 GFP_KERNEL, cma_query_handler,
2274 work, &id_priv->query);
2275
2276 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2277 }
2278
2279 static void cma_work_handler(struct work_struct *_work)
2280 {
2281 struct cma_work *work = container_of(_work, struct cma_work, work);
2282 struct rdma_id_private *id_priv = work->id;
2283 int destroy = 0;
2284
2285 mutex_lock(&id_priv->handler_mutex);
2286 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2287 goto out;
2288
2289 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2290 cma_exch(id_priv, RDMA_CM_DESTROYING);
2291 destroy = 1;
2292 }
2293 out:
2294 mutex_unlock(&id_priv->handler_mutex);
2295 cma_deref_id(id_priv);
2296 if (destroy)
2297 rdma_destroy_id(&id_priv->id);
2298 kfree(work);
2299 }
2300
2301 static void cma_ndev_work_handler(struct work_struct *_work)
2302 {
2303 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2304 struct rdma_id_private *id_priv = work->id;
2305 int destroy = 0;
2306
2307 mutex_lock(&id_priv->handler_mutex);
2308 if (id_priv->state == RDMA_CM_DESTROYING ||
2309 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2310 goto out;
2311
2312 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2313 cma_exch(id_priv, RDMA_CM_DESTROYING);
2314 destroy = 1;
2315 }
2316
2317 out:
2318 mutex_unlock(&id_priv->handler_mutex);
2319 cma_deref_id(id_priv);
2320 if (destroy)
2321 rdma_destroy_id(&id_priv->id);
2322 kfree(work);
2323 }
2324
2325 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2326 {
2327 struct rdma_route *route = &id_priv->id.route;
2328 struct cma_work *work;
2329 int ret;
2330
2331 work = kzalloc(sizeof *work, GFP_KERNEL);
2332 if (!work)
2333 return -ENOMEM;
2334
2335 work->id = id_priv;
2336 INIT_WORK(&work->work, cma_work_handler);
2337 work->old_state = RDMA_CM_ROUTE_QUERY;
2338 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2339 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2340
2341 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2342 if (!route->path_rec) {
2343 ret = -ENOMEM;
2344 goto err1;
2345 }
2346
2347 ret = cma_query_ib_route(id_priv, timeout_ms, work);
2348 if (ret)
2349 goto err2;
2350
2351 return 0;
2352 err2:
2353 kfree(route->path_rec);
2354 route->path_rec = NULL;
2355 err1:
2356 kfree(work);
2357 return ret;
2358 }
2359
2360 int rdma_set_ib_paths(struct rdma_cm_id *id,
2361 struct ib_sa_path_rec *path_rec, int num_paths)
2362 {
2363 struct rdma_id_private *id_priv;
2364 int ret;
2365
2366 id_priv = container_of(id, struct rdma_id_private, id);
2367 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2368 RDMA_CM_ROUTE_RESOLVED))
2369 return -EINVAL;
2370
2371 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2372 GFP_KERNEL);
2373 if (!id->route.path_rec) {
2374 ret = -ENOMEM;
2375 goto err;
2376 }
2377
2378 id->route.num_paths = num_paths;
2379 return 0;
2380 err:
2381 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2382 return ret;
2383 }
2384 EXPORT_SYMBOL(rdma_set_ib_paths);
2385
2386 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2387 {
2388 struct cma_work *work;
2389
2390 work = kzalloc(sizeof *work, GFP_KERNEL);
2391 if (!work)
2392 return -ENOMEM;
2393
2394 work->id = id_priv;
2395 INIT_WORK(&work->work, cma_work_handler);
2396 work->old_state = RDMA_CM_ROUTE_QUERY;
2397 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2398 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2399 queue_work(cma_wq, &work->work);
2400 return 0;
2401 }
2402
2403 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2404 {
2405 int prio;
2406 struct net_device *dev;
2407
2408 prio = rt_tos2priority(tos);
2409 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2410 vlan_dev_real_dev(ndev) : ndev;
2411
2412 if (dev->num_tc)
2413 return netdev_get_prio_tc_map(dev, prio);
2414
2415 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2416 if (ndev->priv_flags & IFF_802_1Q_VLAN)
2417 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2418 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2419 #endif
2420 return 0;
2421 }
2422
2423 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2424 {
2425 struct rdma_route *route = &id_priv->id.route;
2426 struct rdma_addr *addr = &route->addr;
2427 enum ib_gid_type network_gid_type;
2428 struct cma_work *work;
2429 int ret;
2430 struct net_device *ndev = NULL;
2431
2432
2433 work = kzalloc(sizeof *work, GFP_KERNEL);
2434 if (!work)
2435 return -ENOMEM;
2436
2437 work->id = id_priv;
2438 INIT_WORK(&work->work, cma_work_handler);
2439
2440 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2441 if (!route->path_rec) {
2442 ret = -ENOMEM;
2443 goto err1;
2444 }
2445
2446 route->num_paths = 1;
2447
2448 if (addr->dev_addr.bound_dev_if) {
2449 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2450 if (!ndev)
2451 return -ENODEV;
2452
2453 if (ndev->flags & IFF_LOOPBACK) {
2454 dev_put(ndev);
2455 if (!id_priv->id.device->get_netdev)
2456 return -EOPNOTSUPP;
2457
2458 ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2459 id_priv->id.port_num);
2460 if (!ndev)
2461 return -ENODEV;
2462 }
2463
2464 route->path_rec->net = &init_net;
2465 route->path_rec->ifindex = ndev->ifindex;
2466 route->path_rec->gid_type = id_priv->gid_type;
2467 }
2468 if (!ndev) {
2469 ret = -ENODEV;
2470 goto err2;
2471 }
2472
2473 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2474
2475 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2476 &route->path_rec->sgid);
2477 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2478 &route->path_rec->dgid);
2479
2480 /* Use the hint from IP Stack to select GID Type */
2481 network_gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2482 if (addr->dev_addr.network != RDMA_NETWORK_IB) {
2483 route->path_rec->gid_type = network_gid_type;
2484 /* TODO: get the hoplimit from the inet/inet6 device */
2485 route->path_rec->hop_limit = IPV6_DEFAULT_HOPLIMIT;
2486 } else {
2487 route->path_rec->hop_limit = 1;
2488 }
2489 route->path_rec->reversible = 1;
2490 route->path_rec->pkey = cpu_to_be16(0xffff);
2491 route->path_rec->mtu_selector = IB_SA_EQ;
2492 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2493 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2494 route->path_rec->rate_selector = IB_SA_EQ;
2495 route->path_rec->rate = iboe_get_rate(ndev);
2496 dev_put(ndev);
2497 route->path_rec->packet_life_time_selector = IB_SA_EQ;
2498 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2499 if (!route->path_rec->mtu) {
2500 ret = -EINVAL;
2501 goto err2;
2502 }
2503
2504 work->old_state = RDMA_CM_ROUTE_QUERY;
2505 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2506 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2507 work->event.status = 0;
2508
2509 queue_work(cma_wq, &work->work);
2510
2511 return 0;
2512
2513 err2:
2514 kfree(route->path_rec);
2515 route->path_rec = NULL;
2516 err1:
2517 kfree(work);
2518 return ret;
2519 }
2520
2521 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2522 {
2523 struct rdma_id_private *id_priv;
2524 int ret;
2525
2526 id_priv = container_of(id, struct rdma_id_private, id);
2527 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2528 return -EINVAL;
2529
2530 atomic_inc(&id_priv->refcount);
2531 if (rdma_cap_ib_sa(id->device, id->port_num))
2532 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2533 else if (rdma_protocol_roce(id->device, id->port_num))
2534 ret = cma_resolve_iboe_route(id_priv);
2535 else if (rdma_protocol_iwarp(id->device, id->port_num))
2536 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2537 else
2538 ret = -ENOSYS;
2539
2540 if (ret)
2541 goto err;
2542
2543 return 0;
2544 err:
2545 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2546 cma_deref_id(id_priv);
2547 return ret;
2548 }
2549 EXPORT_SYMBOL(rdma_resolve_route);
2550
2551 static void cma_set_loopback(struct sockaddr *addr)
2552 {
2553 switch (addr->sa_family) {
2554 case AF_INET:
2555 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2556 break;
2557 case AF_INET6:
2558 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2559 0, 0, 0, htonl(1));
2560 break;
2561 default:
2562 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2563 0, 0, 0, htonl(1));
2564 break;
2565 }
2566 }
2567
2568 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2569 {
2570 struct cma_device *cma_dev, *cur_dev;
2571 struct ib_port_attr port_attr;
2572 union ib_gid gid;
2573 u16 pkey;
2574 int ret;
2575 u8 p;
2576
2577 cma_dev = NULL;
2578 mutex_lock(&lock);
2579 list_for_each_entry(cur_dev, &dev_list, list) {
2580 if (cma_family(id_priv) == AF_IB &&
2581 !rdma_cap_ib_cm(cur_dev->device, 1))
2582 continue;
2583
2584 if (!cma_dev)
2585 cma_dev = cur_dev;
2586
2587 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2588 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2589 port_attr.state == IB_PORT_ACTIVE) {
2590 cma_dev = cur_dev;
2591 goto port_found;
2592 }
2593 }
2594 }
2595
2596 if (!cma_dev) {
2597 ret = -ENODEV;
2598 goto out;
2599 }
2600
2601 p = 1;
2602
2603 port_found:
2604 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2605 if (ret)
2606 goto out;
2607
2608 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2609 if (ret)
2610 goto out;
2611
2612 id_priv->id.route.addr.dev_addr.dev_type =
2613 (rdma_protocol_ib(cma_dev->device, p)) ?
2614 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2615
2616 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2617 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2618 id_priv->id.port_num = p;
2619 cma_attach_to_dev(id_priv, cma_dev);
2620 cma_set_loopback(cma_src_addr(id_priv));
2621 out:
2622 mutex_unlock(&lock);
2623 return ret;
2624 }
2625
2626 static void addr_handler(int status, struct sockaddr *src_addr,
2627 struct rdma_dev_addr *dev_addr, void *context)
2628 {
2629 struct rdma_id_private *id_priv = context;
2630 struct rdma_cm_event event;
2631
2632 memset(&event, 0, sizeof event);
2633 mutex_lock(&id_priv->handler_mutex);
2634 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2635 RDMA_CM_ADDR_RESOLVED))
2636 goto out;
2637
2638 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2639 if (!status && !id_priv->cma_dev)
2640 status = cma_acquire_dev(id_priv, NULL);
2641
2642 if (status) {
2643 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2644 RDMA_CM_ADDR_BOUND))
2645 goto out;
2646 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2647 event.status = status;
2648 } else
2649 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2650
2651 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2652 cma_exch(id_priv, RDMA_CM_DESTROYING);
2653 mutex_unlock(&id_priv->handler_mutex);
2654 cma_deref_id(id_priv);
2655 rdma_destroy_id(&id_priv->id);
2656 return;
2657 }
2658 out:
2659 mutex_unlock(&id_priv->handler_mutex);
2660 cma_deref_id(id_priv);
2661 }
2662
2663 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2664 {
2665 struct cma_work *work;
2666 union ib_gid gid;
2667 int ret;
2668
2669 work = kzalloc(sizeof *work, GFP_KERNEL);
2670 if (!work)
2671 return -ENOMEM;
2672
2673 if (!id_priv->cma_dev) {
2674 ret = cma_bind_loopback(id_priv);
2675 if (ret)
2676 goto err;
2677 }
2678
2679 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2680 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2681
2682 work->id = id_priv;
2683 INIT_WORK(&work->work, cma_work_handler);
2684 work->old_state = RDMA_CM_ADDR_QUERY;
2685 work->new_state = RDMA_CM_ADDR_RESOLVED;
2686 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2687 queue_work(cma_wq, &work->work);
2688 return 0;
2689 err:
2690 kfree(work);
2691 return ret;
2692 }
2693
2694 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2695 {
2696 struct cma_work *work;
2697 int ret;
2698
2699 work = kzalloc(sizeof *work, GFP_KERNEL);
2700 if (!work)
2701 return -ENOMEM;
2702
2703 if (!id_priv->cma_dev) {
2704 ret = cma_resolve_ib_dev(id_priv);
2705 if (ret)
2706 goto err;
2707 }
2708
2709 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2710 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2711
2712 work->id = id_priv;
2713 INIT_WORK(&work->work, cma_work_handler);
2714 work->old_state = RDMA_CM_ADDR_QUERY;
2715 work->new_state = RDMA_CM_ADDR_RESOLVED;
2716 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2717 queue_work(cma_wq, &work->work);
2718 return 0;
2719 err:
2720 kfree(work);
2721 return ret;
2722 }
2723
2724 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2725 struct sockaddr *dst_addr)
2726 {
2727 if (!src_addr || !src_addr->sa_family) {
2728 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2729 src_addr->sa_family = dst_addr->sa_family;
2730 if (dst_addr->sa_family == AF_INET6) {
2731 struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2732 struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2733 src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2734 if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2735 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2736 } else if (dst_addr->sa_family == AF_IB) {
2737 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2738 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2739 }
2740 }
2741 return rdma_bind_addr(id, src_addr);
2742 }
2743
2744 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2745 struct sockaddr *dst_addr, int timeout_ms)
2746 {
2747 struct rdma_id_private *id_priv;
2748 int ret;
2749
2750 id_priv = container_of(id, struct rdma_id_private, id);
2751 if (id_priv->state == RDMA_CM_IDLE) {
2752 ret = cma_bind_addr(id, src_addr, dst_addr);
2753 if (ret)
2754 return ret;
2755 }
2756
2757 if (cma_family(id_priv) != dst_addr->sa_family)
2758 return -EINVAL;
2759
2760 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2761 return -EINVAL;
2762
2763 atomic_inc(&id_priv->refcount);
2764 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2765 if (cma_any_addr(dst_addr)) {
2766 ret = cma_resolve_loopback(id_priv);
2767 } else {
2768 if (dst_addr->sa_family == AF_IB) {
2769 ret = cma_resolve_ib_addr(id_priv);
2770 } else {
2771 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2772 dst_addr, &id->route.addr.dev_addr,
2773 timeout_ms, addr_handler, id_priv);
2774 }
2775 }
2776 if (ret)
2777 goto err;
2778
2779 return 0;
2780 err:
2781 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2782 cma_deref_id(id_priv);
2783 return ret;
2784 }
2785 EXPORT_SYMBOL(rdma_resolve_addr);
2786
2787 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2788 {
2789 struct rdma_id_private *id_priv;
2790 unsigned long flags;
2791 int ret;
2792
2793 id_priv = container_of(id, struct rdma_id_private, id);
2794 spin_lock_irqsave(&id_priv->lock, flags);
2795 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2796 id_priv->reuseaddr = reuse;
2797 ret = 0;
2798 } else {
2799 ret = -EINVAL;
2800 }
2801 spin_unlock_irqrestore(&id_priv->lock, flags);
2802 return ret;
2803 }
2804 EXPORT_SYMBOL(rdma_set_reuseaddr);
2805
2806 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2807 {
2808 struct rdma_id_private *id_priv;
2809 unsigned long flags;
2810 int ret;
2811
2812 id_priv = container_of(id, struct rdma_id_private, id);
2813 spin_lock_irqsave(&id_priv->lock, flags);
2814 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2815 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2816 id_priv->afonly = afonly;
2817 ret = 0;
2818 } else {
2819 ret = -EINVAL;
2820 }
2821 spin_unlock_irqrestore(&id_priv->lock, flags);
2822 return ret;
2823 }
2824 EXPORT_SYMBOL(rdma_set_afonly);
2825
2826 static void cma_bind_port(struct rdma_bind_list *bind_list,
2827 struct rdma_id_private *id_priv)
2828 {
2829 struct sockaddr *addr;
2830 struct sockaddr_ib *sib;
2831 u64 sid, mask;
2832 __be16 port;
2833
2834 addr = cma_src_addr(id_priv);
2835 port = htons(bind_list->port);
2836
2837 switch (addr->sa_family) {
2838 case AF_INET:
2839 ((struct sockaddr_in *) addr)->sin_port = port;
2840 break;
2841 case AF_INET6:
2842 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2843 break;
2844 case AF_IB:
2845 sib = (struct sockaddr_ib *) addr;
2846 sid = be64_to_cpu(sib->sib_sid);
2847 mask = be64_to_cpu(sib->sib_sid_mask);
2848 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2849 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2850 break;
2851 }
2852 id_priv->bind_list = bind_list;
2853 hlist_add_head(&id_priv->node, &bind_list->owners);
2854 }
2855
2856 static int cma_alloc_port(enum rdma_port_space ps,
2857 struct rdma_id_private *id_priv, unsigned short snum)
2858 {
2859 struct rdma_bind_list *bind_list;
2860 int ret;
2861
2862 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2863 if (!bind_list)
2864 return -ENOMEM;
2865
2866 ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2867 snum);
2868 if (ret < 0)
2869 goto err;
2870
2871 bind_list->ps = ps;
2872 bind_list->port = (unsigned short)ret;
2873 cma_bind_port(bind_list, id_priv);
2874 return 0;
2875 err:
2876 kfree(bind_list);
2877 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2878 }
2879
2880 static int cma_alloc_any_port(enum rdma_port_space ps,
2881 struct rdma_id_private *id_priv)
2882 {
2883 static unsigned int last_used_port;
2884 int low, high, remaining;
2885 unsigned int rover;
2886 struct net *net = id_priv->id.route.addr.dev_addr.net;
2887
2888 inet_get_local_port_range(net, &low, &high);
2889 remaining = (high - low) + 1;
2890 rover = prandom_u32() % remaining + low;
2891 retry:
2892 if (last_used_port != rover &&
2893 !cma_ps_find(net, ps, (unsigned short)rover)) {
2894 int ret = cma_alloc_port(ps, id_priv, rover);
2895 /*
2896 * Remember previously used port number in order to avoid
2897 * re-using same port immediately after it is closed.
2898 */
2899 if (!ret)
2900 last_used_port = rover;
2901 if (ret != -EADDRNOTAVAIL)
2902 return ret;
2903 }
2904 if (--remaining) {
2905 rover++;
2906 if ((rover < low) || (rover > high))
2907 rover = low;
2908 goto retry;
2909 }
2910 return -EADDRNOTAVAIL;
2911 }
2912
2913 /*
2914 * Check that the requested port is available. This is called when trying to
2915 * bind to a specific port, or when trying to listen on a bound port. In
2916 * the latter case, the provided id_priv may already be on the bind_list, but
2917 * we still need to check that it's okay to start listening.
2918 */
2919 static int cma_check_port(struct rdma_bind_list *bind_list,
2920 struct rdma_id_private *id_priv, uint8_t reuseaddr)
2921 {
2922 struct rdma_id_private *cur_id;
2923 struct sockaddr *addr, *cur_addr;
2924
2925 addr = cma_src_addr(id_priv);
2926 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2927 if (id_priv == cur_id)
2928 continue;
2929
2930 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2931 cur_id->reuseaddr)
2932 continue;
2933
2934 cur_addr = cma_src_addr(cur_id);
2935 if (id_priv->afonly && cur_id->afonly &&
2936 (addr->sa_family != cur_addr->sa_family))
2937 continue;
2938
2939 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2940 return -EADDRNOTAVAIL;
2941
2942 if (!cma_addr_cmp(addr, cur_addr))
2943 return -EADDRINUSE;
2944 }
2945 return 0;
2946 }
2947
2948 static int cma_use_port(enum rdma_port_space ps,
2949 struct rdma_id_private *id_priv)
2950 {
2951 struct rdma_bind_list *bind_list;
2952 unsigned short snum;
2953 int ret;
2954
2955 snum = ntohs(cma_port(cma_src_addr(id_priv)));
2956 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2957 return -EACCES;
2958
2959 bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
2960 if (!bind_list) {
2961 ret = cma_alloc_port(ps, id_priv, snum);
2962 } else {
2963 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2964 if (!ret)
2965 cma_bind_port(bind_list, id_priv);
2966 }
2967 return ret;
2968 }
2969
2970 static int cma_bind_listen(struct rdma_id_private *id_priv)
2971 {
2972 struct rdma_bind_list *bind_list = id_priv->bind_list;
2973 int ret = 0;
2974
2975 mutex_lock(&lock);
2976 if (bind_list->owners.first->next)
2977 ret = cma_check_port(bind_list, id_priv, 0);
2978 mutex_unlock(&lock);
2979 return ret;
2980 }
2981
2982 static enum rdma_port_space cma_select_inet_ps(
2983 struct rdma_id_private *id_priv)
2984 {
2985 switch (id_priv->id.ps) {
2986 case RDMA_PS_TCP:
2987 case RDMA_PS_UDP:
2988 case RDMA_PS_IPOIB:
2989 case RDMA_PS_IB:
2990 return id_priv->id.ps;
2991 default:
2992
2993 return 0;
2994 }
2995 }
2996
2997 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
2998 {
2999 enum rdma_port_space ps = 0;
3000 struct sockaddr_ib *sib;
3001 u64 sid_ps, mask, sid;
3002
3003 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3004 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3005 sid = be64_to_cpu(sib->sib_sid) & mask;
3006
3007 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3008 sid_ps = RDMA_IB_IP_PS_IB;
3009 ps = RDMA_PS_IB;
3010 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3011 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3012 sid_ps = RDMA_IB_IP_PS_TCP;
3013 ps = RDMA_PS_TCP;
3014 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3015 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3016 sid_ps = RDMA_IB_IP_PS_UDP;
3017 ps = RDMA_PS_UDP;
3018 }
3019
3020 if (ps) {
3021 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3022 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3023 be64_to_cpu(sib->sib_sid_mask));
3024 }
3025 return ps;
3026 }
3027
3028 static int cma_get_port(struct rdma_id_private *id_priv)
3029 {
3030 enum rdma_port_space ps;
3031 int ret;
3032
3033 if (cma_family(id_priv) != AF_IB)
3034 ps = cma_select_inet_ps(id_priv);
3035 else
3036 ps = cma_select_ib_ps(id_priv);
3037 if (!ps)
3038 return -EPROTONOSUPPORT;
3039
3040 mutex_lock(&lock);
3041 if (cma_any_port(cma_src_addr(id_priv)))
3042 ret = cma_alloc_any_port(ps, id_priv);
3043 else
3044 ret = cma_use_port(ps, id_priv);
3045 mutex_unlock(&lock);
3046
3047 return ret;
3048 }
3049
3050 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3051 struct sockaddr *addr)
3052 {
3053 #if IS_ENABLED(CONFIG_IPV6)
3054 struct sockaddr_in6 *sin6;
3055
3056 if (addr->sa_family != AF_INET6)
3057 return 0;
3058
3059 sin6 = (struct sockaddr_in6 *) addr;
3060
3061 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3062 return 0;
3063
3064 if (!sin6->sin6_scope_id)
3065 return -EINVAL;
3066
3067 dev_addr->bound_dev_if = sin6->sin6_scope_id;
3068 #endif
3069 return 0;
3070 }
3071
3072 int rdma_listen(struct rdma_cm_id *id, int backlog)
3073 {
3074 struct rdma_id_private *id_priv;
3075 int ret;
3076
3077 id_priv = container_of(id, struct rdma_id_private, id);
3078 if (id_priv->state == RDMA_CM_IDLE) {
3079 id->route.addr.src_addr.ss_family = AF_INET;
3080 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3081 if (ret)
3082 return ret;
3083 }
3084
3085 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3086 return -EINVAL;
3087
3088 if (id_priv->reuseaddr) {
3089 ret = cma_bind_listen(id_priv);
3090 if (ret)
3091 goto err;
3092 }
3093
3094 id_priv->backlog = backlog;
3095 if (id->device) {
3096 if (rdma_cap_ib_cm(id->device, 1)) {
3097 ret = cma_ib_listen(id_priv);
3098 if (ret)
3099 goto err;
3100 } else if (rdma_cap_iw_cm(id->device, 1)) {
3101 ret = cma_iw_listen(id_priv, backlog);
3102 if (ret)
3103 goto err;
3104 } else {
3105 ret = -ENOSYS;
3106 goto err;
3107 }
3108 } else
3109 cma_listen_on_all(id_priv);
3110
3111 return 0;
3112 err:
3113 id_priv->backlog = 0;
3114 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3115 return ret;
3116 }
3117 EXPORT_SYMBOL(rdma_listen);
3118
3119 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3120 {
3121 struct rdma_id_private *id_priv;
3122 int ret;
3123
3124 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3125 addr->sa_family != AF_IB)
3126 return -EAFNOSUPPORT;
3127
3128 id_priv = container_of(id, struct rdma_id_private, id);
3129 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3130 return -EINVAL;
3131
3132 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3133 if (ret)
3134 goto err1;
3135
3136 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3137 if (!cma_any_addr(addr)) {
3138 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3139 if (ret)
3140 goto err1;
3141
3142 ret = cma_acquire_dev(id_priv, NULL);
3143 if (ret)
3144 goto err1;
3145 }
3146
3147 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3148 if (addr->sa_family == AF_INET)
3149 id_priv->afonly = 1;
3150 #if IS_ENABLED(CONFIG_IPV6)
3151 else if (addr->sa_family == AF_INET6) {
3152 struct net *net = id_priv->id.route.addr.dev_addr.net;
3153
3154 id_priv->afonly = net->ipv6.sysctl.bindv6only;
3155 }
3156 #endif
3157 }
3158 ret = cma_get_port(id_priv);
3159 if (ret)
3160 goto err2;
3161
3162 return 0;
3163 err2:
3164 if (id_priv->cma_dev)
3165 cma_release_dev(id_priv);
3166 err1:
3167 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3168 return ret;
3169 }
3170 EXPORT_SYMBOL(rdma_bind_addr);
3171
3172 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3173 {
3174 struct cma_hdr *cma_hdr;
3175
3176 cma_hdr = hdr;
3177 cma_hdr->cma_version = CMA_VERSION;
3178 if (cma_family(id_priv) == AF_INET) {
3179 struct sockaddr_in *src4, *dst4;
3180
3181 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3182 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3183
3184 cma_set_ip_ver(cma_hdr, 4);
3185 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3186 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3187 cma_hdr->port = src4->sin_port;
3188 } else if (cma_family(id_priv) == AF_INET6) {
3189 struct sockaddr_in6 *src6, *dst6;
3190
3191 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3192 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3193
3194 cma_set_ip_ver(cma_hdr, 6);
3195 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3196 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3197 cma_hdr->port = src6->sin6_port;
3198 }
3199 return 0;
3200 }
3201
3202 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3203 struct ib_cm_event *ib_event)
3204 {
3205 struct rdma_id_private *id_priv = cm_id->context;
3206 struct rdma_cm_event event;
3207 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3208 int ret = 0;
3209
3210 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
3211 return 0;
3212
3213 memset(&event, 0, sizeof event);
3214 switch (ib_event->event) {
3215 case IB_CM_SIDR_REQ_ERROR:
3216 event.event = RDMA_CM_EVENT_UNREACHABLE;
3217 event.status = -ETIMEDOUT;
3218 break;
3219 case IB_CM_SIDR_REP_RECEIVED:
3220 event.param.ud.private_data = ib_event->private_data;
3221 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3222 if (rep->status != IB_SIDR_SUCCESS) {
3223 event.event = RDMA_CM_EVENT_UNREACHABLE;
3224 event.status = ib_event->param.sidr_rep_rcvd.status;
3225 break;
3226 }
3227 ret = cma_set_qkey(id_priv, rep->qkey);
3228 if (ret) {
3229 event.event = RDMA_CM_EVENT_ADDR_ERROR;
3230 event.status = ret;
3231 break;
3232 }
3233 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3234 id_priv->id.route.path_rec,
3235 &event.param.ud.ah_attr);
3236 event.param.ud.qp_num = rep->qpn;
3237 event.param.ud.qkey = rep->qkey;
3238 event.event = RDMA_CM_EVENT_ESTABLISHED;
3239 event.status = 0;
3240 break;
3241 default:
3242 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
3243 ib_event->event);
3244 goto out;
3245 }
3246
3247 ret = id_priv->id.event_handler(&id_priv->id, &event);
3248 if (ret) {
3249 /* Destroy the CM ID by returning a non-zero value. */
3250 id_priv->cm_id.ib = NULL;
3251 cma_exch(id_priv, RDMA_CM_DESTROYING);
3252 mutex_unlock(&id_priv->handler_mutex);
3253 rdma_destroy_id(&id_priv->id);
3254 return ret;
3255 }
3256 out:
3257 mutex_unlock(&id_priv->handler_mutex);
3258 return ret;
3259 }
3260
3261 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3262 struct rdma_conn_param *conn_param)
3263 {
3264 struct ib_cm_sidr_req_param req;
3265 struct ib_cm_id *id;
3266 void *private_data;
3267 int offset, ret;
3268
3269 memset(&req, 0, sizeof req);
3270 offset = cma_user_data_offset(id_priv);
3271 req.private_data_len = offset + conn_param->private_data_len;
3272 if (req.private_data_len < conn_param->private_data_len)
3273 return -EINVAL;
3274
3275 if (req.private_data_len) {
3276 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3277 if (!private_data)
3278 return -ENOMEM;
3279 } else {
3280 private_data = NULL;
3281 }
3282
3283 if (conn_param->private_data && conn_param->private_data_len)
3284 memcpy(private_data + offset, conn_param->private_data,
3285 conn_param->private_data_len);
3286
3287 if (private_data) {
3288 ret = cma_format_hdr(private_data, id_priv);
3289 if (ret)
3290 goto out;
3291 req.private_data = private_data;
3292 }
3293
3294 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3295 id_priv);
3296 if (IS_ERR(id)) {
3297 ret = PTR_ERR(id);
3298 goto out;
3299 }
3300 id_priv->cm_id.ib = id;
3301
3302 req.path = id_priv->id.route.path_rec;
3303 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3304 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3305 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3306
3307 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3308 if (ret) {
3309 ib_destroy_cm_id(id_priv->cm_id.ib);
3310 id_priv->cm_id.ib = NULL;
3311 }
3312 out:
3313 kfree(private_data);
3314 return ret;
3315 }
3316
3317 static int cma_connect_ib(struct rdma_id_private *id_priv,
3318 struct rdma_conn_param *conn_param)
3319 {
3320 struct ib_cm_req_param req;
3321 struct rdma_route *route;
3322 void *private_data;
3323 struct ib_cm_id *id;
3324 int offset, ret;
3325
3326 memset(&req, 0, sizeof req);
3327 offset = cma_user_data_offset(id_priv);
3328 req.private_data_len = offset + conn_param->private_data_len;
3329 if (req.private_data_len < conn_param->private_data_len)
3330 return -EINVAL;
3331
3332 if (req.private_data_len) {
3333 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3334 if (!private_data)
3335 return -ENOMEM;
3336 } else {
3337 private_data = NULL;
3338 }
3339
3340 if (conn_param->private_data && conn_param->private_data_len)
3341 memcpy(private_data + offset, conn_param->private_data,
3342 conn_param->private_data_len);
3343
3344 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3345 if (IS_ERR(id)) {
3346 ret = PTR_ERR(id);
3347 goto out;
3348 }
3349 id_priv->cm_id.ib = id;
3350
3351 route = &id_priv->id.route;
3352 if (private_data) {
3353 ret = cma_format_hdr(private_data, id_priv);
3354 if (ret)
3355 goto out;
3356 req.private_data = private_data;
3357 }
3358
3359 req.primary_path = &route->path_rec[0];
3360 if (route->num_paths == 2)
3361 req.alternate_path = &route->path_rec[1];
3362
3363 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3364 req.qp_num = id_priv->qp_num;
3365 req.qp_type = id_priv->id.qp_type;
3366 req.starting_psn = id_priv->seq_num;
3367 req.responder_resources = conn_param->responder_resources;
3368 req.initiator_depth = conn_param->initiator_depth;
3369 req.flow_control = conn_param->flow_control;
3370 req.retry_count = min_t(u8, 7, conn_param->retry_count);
3371 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3372 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3373 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3374 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3375 req.srq = id_priv->srq ? 1 : 0;
3376
3377 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3378 out:
3379 if (ret && !IS_ERR(id)) {
3380 ib_destroy_cm_id(id);
3381 id_priv->cm_id.ib = NULL;
3382 }
3383
3384 kfree(private_data);
3385 return ret;
3386 }
3387
3388 static int cma_connect_iw(struct rdma_id_private *id_priv,
3389 struct rdma_conn_param *conn_param)
3390 {
3391 struct iw_cm_id *cm_id;
3392 int ret;
3393 struct iw_cm_conn_param iw_param;
3394
3395 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3396 if (IS_ERR(cm_id))
3397 return PTR_ERR(cm_id);
3398
3399 cm_id->tos = id_priv->tos;
3400 id_priv->cm_id.iw = cm_id;
3401
3402 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3403 rdma_addr_size(cma_src_addr(id_priv)));
3404 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3405 rdma_addr_size(cma_dst_addr(id_priv)));
3406
3407 ret = cma_modify_qp_rtr(id_priv, conn_param);
3408 if (ret)
3409 goto out;
3410
3411 if (conn_param) {
3412 iw_param.ord = conn_param->initiator_depth;
3413 iw_param.ird = conn_param->responder_resources;
3414 iw_param.private_data = conn_param->private_data;
3415 iw_param.private_data_len = conn_param->private_data_len;
3416 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3417 } else {
3418 memset(&iw_param, 0, sizeof iw_param);
3419 iw_param.qpn = id_priv->qp_num;
3420 }
3421 ret = iw_cm_connect(cm_id, &iw_param);
3422 out:
3423 if (ret) {
3424 iw_destroy_cm_id(cm_id);
3425 id_priv->cm_id.iw = NULL;
3426 }
3427 return ret;
3428 }
3429
3430 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3431 {
3432 struct rdma_id_private *id_priv;
3433 int ret;
3434
3435 id_priv = container_of(id, struct rdma_id_private, id);
3436 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3437 return -EINVAL;
3438
3439 if (!id->qp) {
3440 id_priv->qp_num = conn_param->qp_num;
3441 id_priv->srq = conn_param->srq;
3442 }
3443
3444 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3445 if (id->qp_type == IB_QPT_UD)
3446 ret = cma_resolve_ib_udp(id_priv, conn_param);
3447 else
3448 ret = cma_connect_ib(id_priv, conn_param);
3449 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3450 ret = cma_connect_iw(id_priv, conn_param);
3451 else
3452 ret = -ENOSYS;
3453 if (ret)
3454 goto err;
3455
3456 return 0;
3457 err:
3458 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3459 return ret;
3460 }
3461 EXPORT_SYMBOL(rdma_connect);
3462
3463 static int cma_accept_ib(struct rdma_id_private *id_priv,
3464 struct rdma_conn_param *conn_param)
3465 {
3466 struct ib_cm_rep_param rep;
3467 int ret;
3468
3469 ret = cma_modify_qp_rtr(id_priv, conn_param);
3470 if (ret)
3471 goto out;
3472
3473 ret = cma_modify_qp_rts(id_priv, conn_param);
3474 if (ret)
3475 goto out;
3476
3477 memset(&rep, 0, sizeof rep);
3478 rep.qp_num = id_priv->qp_num;
3479 rep.starting_psn = id_priv->seq_num;
3480 rep.private_data = conn_param->private_data;
3481 rep.private_data_len = conn_param->private_data_len;
3482 rep.responder_resources = conn_param->responder_resources;
3483 rep.initiator_depth = conn_param->initiator_depth;
3484 rep.failover_accepted = 0;
3485 rep.flow_control = conn_param->flow_control;
3486 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3487 rep.srq = id_priv->srq ? 1 : 0;
3488
3489 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3490 out:
3491 return ret;
3492 }
3493
3494 static int cma_accept_iw(struct rdma_id_private *id_priv,
3495 struct rdma_conn_param *conn_param)
3496 {
3497 struct iw_cm_conn_param iw_param;
3498 int ret;
3499
3500 ret = cma_modify_qp_rtr(id_priv, conn_param);
3501 if (ret)
3502 return ret;
3503
3504 iw_param.ord = conn_param->initiator_depth;
3505 iw_param.ird = conn_param->responder_resources;
3506 iw_param.private_data = conn_param->private_data;
3507 iw_param.private_data_len = conn_param->private_data_len;
3508 if (id_priv->id.qp) {
3509 iw_param.qpn = id_priv->qp_num;
3510 } else
3511 iw_param.qpn = conn_param->qp_num;
3512
3513 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3514 }
3515
3516 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3517 enum ib_cm_sidr_status status, u32 qkey,
3518 const void *private_data, int private_data_len)
3519 {
3520 struct ib_cm_sidr_rep_param rep;
3521 int ret;
3522
3523 memset(&rep, 0, sizeof rep);
3524 rep.status = status;
3525 if (status == IB_SIDR_SUCCESS) {
3526 ret = cma_set_qkey(id_priv, qkey);
3527 if (ret)
3528 return ret;
3529 rep.qp_num = id_priv->qp_num;
3530 rep.qkey = id_priv->qkey;
3531 }
3532 rep.private_data = private_data;
3533 rep.private_data_len = private_data_len;
3534
3535 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3536 }
3537
3538 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3539 {
3540 struct rdma_id_private *id_priv;
3541 int ret;
3542
3543 id_priv = container_of(id, struct rdma_id_private, id);
3544
3545 id_priv->owner = task_pid_nr(current);
3546
3547 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3548 return -EINVAL;
3549
3550 if (!id->qp && conn_param) {
3551 id_priv->qp_num = conn_param->qp_num;
3552 id_priv->srq = conn_param->srq;
3553 }
3554
3555 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3556 if (id->qp_type == IB_QPT_UD) {
3557 if (conn_param)
3558 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3559 conn_param->qkey,
3560 conn_param->private_data,
3561 conn_param->private_data_len);
3562 else
3563 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3564 0, NULL, 0);
3565 } else {
3566 if (conn_param)
3567 ret = cma_accept_ib(id_priv, conn_param);
3568 else
3569 ret = cma_rep_recv(id_priv);
3570 }
3571 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3572 ret = cma_accept_iw(id_priv, conn_param);
3573 else
3574 ret = -ENOSYS;
3575
3576 if (ret)
3577 goto reject;
3578
3579 return 0;
3580 reject:
3581 cma_modify_qp_err(id_priv);
3582 rdma_reject(id, NULL, 0);
3583 return ret;
3584 }
3585 EXPORT_SYMBOL(rdma_accept);
3586
3587 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3588 {
3589 struct rdma_id_private *id_priv;
3590 int ret;
3591
3592 id_priv = container_of(id, struct rdma_id_private, id);
3593 if (!id_priv->cm_id.ib)
3594 return -EINVAL;
3595
3596 switch (id->device->node_type) {
3597 case RDMA_NODE_IB_CA:
3598 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3599 break;
3600 default:
3601 ret = 0;
3602 break;
3603 }
3604 return ret;
3605 }
3606 EXPORT_SYMBOL(rdma_notify);
3607
3608 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3609 u8 private_data_len)
3610 {
3611 struct rdma_id_private *id_priv;
3612 int ret;
3613
3614 id_priv = container_of(id, struct rdma_id_private, id);
3615 if (!id_priv->cm_id.ib)
3616 return -EINVAL;
3617
3618 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3619 if (id->qp_type == IB_QPT_UD)
3620 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3621 private_data, private_data_len);
3622 else
3623 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3624 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3625 0, private_data, private_data_len);
3626 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3627 ret = iw_cm_reject(id_priv->cm_id.iw,
3628 private_data, private_data_len);
3629 } else
3630 ret = -ENOSYS;
3631
3632 return ret;
3633 }
3634 EXPORT_SYMBOL(rdma_reject);
3635
3636 int rdma_disconnect(struct rdma_cm_id *id)
3637 {
3638 struct rdma_id_private *id_priv;
3639 int ret;
3640
3641 id_priv = container_of(id, struct rdma_id_private, id);
3642 if (!id_priv->cm_id.ib)
3643 return -EINVAL;
3644
3645 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3646 ret = cma_modify_qp_err(id_priv);
3647 if (ret)
3648 goto out;
3649 /* Initiate or respond to a disconnect. */
3650 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3651 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3652 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3653 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3654 } else
3655 ret = -EINVAL;
3656
3657 out:
3658 return ret;
3659 }
3660 EXPORT_SYMBOL(rdma_disconnect);
3661
3662 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3663 {
3664 struct rdma_id_private *id_priv;
3665 struct cma_multicast *mc = multicast->context;
3666 struct rdma_cm_event event;
3667 int ret;
3668
3669 id_priv = mc->id_priv;
3670 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3671 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3672 return 0;
3673
3674 if (!status)
3675 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3676 mutex_lock(&id_priv->qp_mutex);
3677 if (!status && id_priv->id.qp)
3678 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3679 be16_to_cpu(multicast->rec.mlid));
3680 mutex_unlock(&id_priv->qp_mutex);
3681
3682 memset(&event, 0, sizeof event);
3683 event.status = status;
3684 event.param.ud.private_data = mc->context;
3685 if (!status) {
3686 struct rdma_dev_addr *dev_addr =
3687 &id_priv->id.route.addr.dev_addr;
3688 struct net_device *ndev =
3689 dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3690 enum ib_gid_type gid_type =
3691 id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3692 rdma_start_port(id_priv->cma_dev->device)];
3693
3694 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3695 ib_init_ah_from_mcmember(id_priv->id.device,
3696 id_priv->id.port_num, &multicast->rec,
3697 ndev, gid_type,
3698 &event.param.ud.ah_attr);
3699 event.param.ud.qp_num = 0xFFFFFF;
3700 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3701 if (ndev)
3702 dev_put(ndev);
3703 } else
3704 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3705
3706 ret = id_priv->id.event_handler(&id_priv->id, &event);
3707 if (ret) {
3708 cma_exch(id_priv, RDMA_CM_DESTROYING);
3709 mutex_unlock(&id_priv->handler_mutex);
3710 rdma_destroy_id(&id_priv->id);
3711 return 0;
3712 }
3713
3714 mutex_unlock(&id_priv->handler_mutex);
3715 return 0;
3716 }
3717
3718 static void cma_set_mgid(struct rdma_id_private *id_priv,
3719 struct sockaddr *addr, union ib_gid *mgid)
3720 {
3721 unsigned char mc_map[MAX_ADDR_LEN];
3722 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3723 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3724 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3725
3726 if (cma_any_addr(addr)) {
3727 memset(mgid, 0, sizeof *mgid);
3728 } else if ((addr->sa_family == AF_INET6) &&
3729 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3730 0xFF10A01B)) {
3731 /* IPv6 address is an SA assigned MGID. */
3732 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3733 } else if (addr->sa_family == AF_IB) {
3734 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3735 } else if ((addr->sa_family == AF_INET6)) {
3736 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3737 if (id_priv->id.ps == RDMA_PS_UDP)
3738 mc_map[7] = 0x01; /* Use RDMA CM signature */
3739 *mgid = *(union ib_gid *) (mc_map + 4);
3740 } else {
3741 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3742 if (id_priv->id.ps == RDMA_PS_UDP)
3743 mc_map[7] = 0x01; /* Use RDMA CM signature */
3744 *mgid = *(union ib_gid *) (mc_map + 4);
3745 }
3746 }
3747
3748 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3749 struct cma_multicast *mc)
3750 {
3751 struct ib_sa_mcmember_rec rec;
3752 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3753 ib_sa_comp_mask comp_mask;
3754 int ret;
3755
3756 ib_addr_get_mgid(dev_addr, &rec.mgid);
3757 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3758 &rec.mgid, &rec);
3759 if (ret)
3760 return ret;
3761
3762 ret = cma_set_qkey(id_priv, 0);
3763 if (ret)
3764 return ret;
3765
3766 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3767 rec.qkey = cpu_to_be32(id_priv->qkey);
3768 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3769 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3770 rec.join_state = 1;
3771
3772 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3773 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3774 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3775 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3776 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3777
3778 if (id_priv->id.ps == RDMA_PS_IPOIB)
3779 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3780 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3781 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3782 IB_SA_MCMEMBER_REC_MTU |
3783 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3784
3785 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3786 id_priv->id.port_num, &rec,
3787 comp_mask, GFP_KERNEL,
3788 cma_ib_mc_handler, mc);
3789 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3790 }
3791
3792 static void iboe_mcast_work_handler(struct work_struct *work)
3793 {
3794 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3795 struct cma_multicast *mc = mw->mc;
3796 struct ib_sa_multicast *m = mc->multicast.ib;
3797
3798 mc->multicast.ib->context = mc;
3799 cma_ib_mc_handler(0, m);
3800 kref_put(&mc->mcref, release_mc);
3801 kfree(mw);
3802 }
3803
3804 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3805 {
3806 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3807 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3808
3809 if (cma_any_addr(addr)) {
3810 memset(mgid, 0, sizeof *mgid);
3811 } else if (addr->sa_family == AF_INET6) {
3812 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3813 } else {
3814 mgid->raw[0] = 0xff;
3815 mgid->raw[1] = 0x0e;
3816 mgid->raw[2] = 0;
3817 mgid->raw[3] = 0;
3818 mgid->raw[4] = 0;
3819 mgid->raw[5] = 0;
3820 mgid->raw[6] = 0;
3821 mgid->raw[7] = 0;
3822 mgid->raw[8] = 0;
3823 mgid->raw[9] = 0;
3824 mgid->raw[10] = 0xff;
3825 mgid->raw[11] = 0xff;
3826 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3827 }
3828 }
3829
3830 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3831 struct cma_multicast *mc)
3832 {
3833 struct iboe_mcast_work *work;
3834 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3835 int err = 0;
3836 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3837 struct net_device *ndev = NULL;
3838 enum ib_gid_type gid_type;
3839
3840 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3841 return -EINVAL;
3842
3843 work = kzalloc(sizeof *work, GFP_KERNEL);
3844 if (!work)
3845 return -ENOMEM;
3846
3847 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3848 if (!mc->multicast.ib) {
3849 err = -ENOMEM;
3850 goto out1;
3851 }
3852
3853 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3854
3855 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3856 if (id_priv->id.ps == RDMA_PS_UDP)
3857 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3858
3859 if (dev_addr->bound_dev_if)
3860 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3861 if (!ndev) {
3862 err = -ENODEV;
3863 goto out2;
3864 }
3865 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3866 mc->multicast.ib->rec.hop_limit = 1;
3867 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3868
3869 gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3870 rdma_start_port(id_priv->cma_dev->device)];
3871 if (addr->sa_family == AF_INET) {
3872 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
3873 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
3874 true);
3875 if (!err) {
3876 mc->igmp_joined = true;
3877 mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
3878 }
3879 } else {
3880 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
3881 err = -ENOTSUPP;
3882 }
3883 dev_put(ndev);
3884 if (err || !mc->multicast.ib->rec.mtu) {
3885 if (!err)
3886 err = -EINVAL;
3887 goto out2;
3888 }
3889 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3890 &mc->multicast.ib->rec.port_gid);
3891 work->id = id_priv;
3892 work->mc = mc;
3893 INIT_WORK(&work->work, iboe_mcast_work_handler);
3894 kref_get(&mc->mcref);
3895 queue_work(cma_wq, &work->work);
3896
3897 return 0;
3898
3899 out2:
3900 kfree(mc->multicast.ib);
3901 out1:
3902 kfree(work);
3903 return err;
3904 }
3905
3906 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3907 void *context)
3908 {
3909 struct rdma_id_private *id_priv;
3910 struct cma_multicast *mc;
3911 int ret;
3912
3913 id_priv = container_of(id, struct rdma_id_private, id);
3914 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3915 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3916 return -EINVAL;
3917
3918 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3919 if (!mc)
3920 return -ENOMEM;
3921
3922 memcpy(&mc->addr, addr, rdma_addr_size(addr));
3923 mc->context = context;
3924 mc->id_priv = id_priv;
3925 mc->igmp_joined = false;
3926 spin_lock(&id_priv->lock);
3927 list_add(&mc->list, &id_priv->mc_list);
3928 spin_unlock(&id_priv->lock);
3929
3930 if (rdma_protocol_roce(id->device, id->port_num)) {
3931 kref_init(&mc->mcref);
3932 ret = cma_iboe_join_multicast(id_priv, mc);
3933 } else if (rdma_cap_ib_mcast(id->device, id->port_num))
3934 ret = cma_join_ib_multicast(id_priv, mc);
3935 else
3936 ret = -ENOSYS;
3937
3938 if (ret) {
3939 spin_lock_irq(&id_priv->lock);
3940 list_del(&mc->list);
3941 spin_unlock_irq(&id_priv->lock);
3942 kfree(mc);
3943 }
3944 return ret;
3945 }
3946 EXPORT_SYMBOL(rdma_join_multicast);
3947
3948 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3949 {
3950 struct rdma_id_private *id_priv;
3951 struct cma_multicast *mc;
3952
3953 id_priv = container_of(id, struct rdma_id_private, id);
3954 spin_lock_irq(&id_priv->lock);
3955 list_for_each_entry(mc, &id_priv->mc_list, list) {
3956 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3957 list_del(&mc->list);
3958 spin_unlock_irq(&id_priv->lock);
3959
3960 if (id->qp)
3961 ib_detach_mcast(id->qp,
3962 &mc->multicast.ib->rec.mgid,
3963 be16_to_cpu(mc->multicast.ib->rec.mlid));
3964
3965 BUG_ON(id_priv->cma_dev->device != id->device);
3966
3967 if (rdma_cap_ib_mcast(id->device, id->port_num)) {
3968 ib_sa_free_multicast(mc->multicast.ib);
3969 kfree(mc);
3970 } else if (rdma_protocol_roce(id->device, id->port_num)) {
3971 if (mc->igmp_joined) {
3972 struct rdma_dev_addr *dev_addr =
3973 &id->route.addr.dev_addr;
3974 struct net_device *ndev = NULL;
3975
3976 if (dev_addr->bound_dev_if)
3977 ndev = dev_get_by_index(&init_net,
3978 dev_addr->bound_dev_if);
3979 if (ndev) {
3980 cma_igmp_send(ndev,
3981 &mc->multicast.ib->rec.mgid,
3982 false);
3983 dev_put(ndev);
3984 }
3985 mc->igmp_joined = false;
3986 }
3987 kref_put(&mc->mcref, release_mc);
3988 }
3989 return;
3990 }
3991 }
3992 spin_unlock_irq(&id_priv->lock);
3993 }
3994 EXPORT_SYMBOL(rdma_leave_multicast);
3995
3996 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3997 {
3998 struct rdma_dev_addr *dev_addr;
3999 struct cma_ndev_work *work;
4000
4001 dev_addr = &id_priv->id.route.addr.dev_addr;
4002
4003 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4004 (net_eq(dev_net(ndev), dev_addr->net)) &&
4005 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4006 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
4007 ndev->name, &id_priv->id);
4008 work = kzalloc(sizeof *work, GFP_KERNEL);
4009 if (!work)
4010 return -ENOMEM;
4011
4012 INIT_WORK(&work->work, cma_ndev_work_handler);
4013 work->id = id_priv;
4014 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4015 atomic_inc(&id_priv->refcount);
4016 queue_work(cma_wq, &work->work);
4017 }
4018
4019 return 0;
4020 }
4021
4022 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4023 void *ptr)
4024 {
4025 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4026 struct cma_device *cma_dev;
4027 struct rdma_id_private *id_priv;
4028 int ret = NOTIFY_DONE;
4029
4030 if (event != NETDEV_BONDING_FAILOVER)
4031 return NOTIFY_DONE;
4032
4033 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4034 return NOTIFY_DONE;
4035
4036 mutex_lock(&lock);
4037 list_for_each_entry(cma_dev, &dev_list, list)
4038 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4039 ret = cma_netdev_change(ndev, id_priv);
4040 if (ret)
4041 goto out;
4042 }
4043
4044 out:
4045 mutex_unlock(&lock);
4046 return ret;
4047 }
4048
4049 static struct notifier_block cma_nb = {
4050 .notifier_call = cma_netdev_callback
4051 };
4052
4053 static void cma_add_one(struct ib_device *device)
4054 {
4055 struct cma_device *cma_dev;
4056 struct rdma_id_private *id_priv;
4057 unsigned int i;
4058 unsigned long supported_gids = 0;
4059
4060 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4061 if (!cma_dev)
4062 return;
4063
4064 cma_dev->device = device;
4065 cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4066 sizeof(*cma_dev->default_gid_type),
4067 GFP_KERNEL);
4068 if (!cma_dev->default_gid_type) {
4069 kfree(cma_dev);
4070 return;
4071 }
4072 for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4073 supported_gids = roce_gid_type_mask_support(device, i);
4074 WARN_ON(!supported_gids);
4075 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4076 find_first_bit(&supported_gids, BITS_PER_LONG);
4077 }
4078
4079 init_completion(&cma_dev->comp);
4080 atomic_set(&cma_dev->refcount, 1);
4081 INIT_LIST_HEAD(&cma_dev->id_list);
4082 ib_set_client_data(device, &cma_client, cma_dev);
4083
4084 mutex_lock(&lock);
4085 list_add_tail(&cma_dev->list, &dev_list);
4086 list_for_each_entry(id_priv, &listen_any_list, list)
4087 cma_listen_on_dev(id_priv, cma_dev);
4088 mutex_unlock(&lock);
4089 }
4090
4091 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4092 {
4093 struct rdma_cm_event event;
4094 enum rdma_cm_state state;
4095 int ret = 0;
4096
4097 /* Record that we want to remove the device */
4098 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4099 if (state == RDMA_CM_DESTROYING)
4100 return 0;
4101
4102 cma_cancel_operation(id_priv, state);
4103 mutex_lock(&id_priv->handler_mutex);
4104
4105 /* Check for destruction from another callback. */
4106 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4107 goto out;
4108
4109 memset(&event, 0, sizeof event);
4110 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4111 ret = id_priv->id.event_handler(&id_priv->id, &event);
4112 out:
4113 mutex_unlock(&id_priv->handler_mutex);
4114 return ret;
4115 }
4116
4117 static void cma_process_remove(struct cma_device *cma_dev)
4118 {
4119 struct rdma_id_private *id_priv;
4120 int ret;
4121
4122 mutex_lock(&lock);
4123 while (!list_empty(&cma_dev->id_list)) {
4124 id_priv = list_entry(cma_dev->id_list.next,
4125 struct rdma_id_private, list);
4126
4127 list_del(&id_priv->listen_list);
4128 list_del_init(&id_priv->list);
4129 atomic_inc(&id_priv->refcount);
4130 mutex_unlock(&lock);
4131
4132 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4133 cma_deref_id(id_priv);
4134 if (ret)
4135 rdma_destroy_id(&id_priv->id);
4136
4137 mutex_lock(&lock);
4138 }
4139 mutex_unlock(&lock);
4140
4141 cma_deref_dev(cma_dev);
4142 wait_for_completion(&cma_dev->comp);
4143 }
4144
4145 static void cma_remove_one(struct ib_device *device, void *client_data)
4146 {
4147 struct cma_device *cma_dev = client_data;
4148
4149 if (!cma_dev)
4150 return;
4151
4152 mutex_lock(&lock);
4153 list_del(&cma_dev->list);
4154 mutex_unlock(&lock);
4155
4156 cma_process_remove(cma_dev);
4157 kfree(cma_dev->default_gid_type);
4158 kfree(cma_dev);
4159 }
4160
4161 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4162 {
4163 struct nlmsghdr *nlh;
4164 struct rdma_cm_id_stats *id_stats;
4165 struct rdma_id_private *id_priv;
4166 struct rdma_cm_id *id = NULL;
4167 struct cma_device *cma_dev;
4168 int i_dev = 0, i_id = 0;
4169
4170 /*
4171 * We export all of the IDs as a sequence of messages. Each
4172 * ID gets its own netlink message.
4173 */
4174 mutex_lock(&lock);
4175
4176 list_for_each_entry(cma_dev, &dev_list, list) {
4177 if (i_dev < cb->args[0]) {
4178 i_dev++;
4179 continue;
4180 }
4181
4182 i_id = 0;
4183 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4184 if (i_id < cb->args[1]) {
4185 i_id++;
4186 continue;
4187 }
4188
4189 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4190 sizeof *id_stats, RDMA_NL_RDMA_CM,
4191 RDMA_NL_RDMA_CM_ID_STATS,
4192 NLM_F_MULTI);
4193 if (!id_stats)
4194 goto out;
4195
4196 memset(id_stats, 0, sizeof *id_stats);
4197 id = &id_priv->id;
4198 id_stats->node_type = id->route.addr.dev_addr.dev_type;
4199 id_stats->port_num = id->port_num;
4200 id_stats->bound_dev_if =
4201 id->route.addr.dev_addr.bound_dev_if;
4202
4203 if (ibnl_put_attr(skb, nlh,
4204 rdma_addr_size(cma_src_addr(id_priv)),
4205 cma_src_addr(id_priv),
4206 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4207 goto out;
4208 if (ibnl_put_attr(skb, nlh,
4209 rdma_addr_size(cma_src_addr(id_priv)),
4210 cma_dst_addr(id_priv),
4211 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4212 goto out;
4213
4214 id_stats->pid = id_priv->owner;
4215 id_stats->port_space = id->ps;
4216 id_stats->cm_state = id_priv->state;
4217 id_stats->qp_num = id_priv->qp_num;
4218 id_stats->qp_type = id->qp_type;
4219
4220 i_id++;
4221 }
4222
4223 cb->args[1] = 0;
4224 i_dev++;
4225 }
4226
4227 out:
4228 mutex_unlock(&lock);
4229 cb->args[0] = i_dev;
4230 cb->args[1] = i_id;
4231
4232 return skb->len;
4233 }
4234
4235 static const struct ibnl_client_cbs cma_cb_table[] = {
4236 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4237 .module = THIS_MODULE },
4238 };
4239
4240 static int cma_init_net(struct net *net)
4241 {
4242 struct cma_pernet *pernet = cma_pernet(net);
4243
4244 idr_init(&pernet->tcp_ps);
4245 idr_init(&pernet->udp_ps);
4246 idr_init(&pernet->ipoib_ps);
4247 idr_init(&pernet->ib_ps);
4248
4249 return 0;
4250 }
4251
4252 static void cma_exit_net(struct net *net)
4253 {
4254 struct cma_pernet *pernet = cma_pernet(net);
4255
4256 idr_destroy(&pernet->tcp_ps);
4257 idr_destroy(&pernet->udp_ps);
4258 idr_destroy(&pernet->ipoib_ps);
4259 idr_destroy(&pernet->ib_ps);
4260 }
4261
4262 static struct pernet_operations cma_pernet_operations = {
4263 .init = cma_init_net,
4264 .exit = cma_exit_net,
4265 .id = &cma_pernet_id,
4266 .size = sizeof(struct cma_pernet),
4267 };
4268
4269 static int __init cma_init(void)
4270 {
4271 int ret;
4272
4273 cma_wq = create_singlethread_workqueue("rdma_cm");
4274 if (!cma_wq)
4275 return -ENOMEM;
4276
4277 ret = register_pernet_subsys(&cma_pernet_operations);
4278 if (ret)
4279 goto err_wq;
4280
4281 ib_sa_register_client(&sa_client);
4282 rdma_addr_register_client(&addr_client);
4283 register_netdevice_notifier(&cma_nb);
4284
4285 ret = ib_register_client(&cma_client);
4286 if (ret)
4287 goto err;
4288
4289 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
4290 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
4291 cma_configfs_init();
4292
4293 return 0;
4294
4295 err:
4296 unregister_netdevice_notifier(&cma_nb);
4297 rdma_addr_unregister_client(&addr_client);
4298 ib_sa_unregister_client(&sa_client);
4299 err_wq:
4300 destroy_workqueue(cma_wq);
4301 return ret;
4302 }
4303
4304 static void __exit cma_cleanup(void)
4305 {
4306 cma_configfs_exit();
4307 ibnl_remove_client(RDMA_NL_RDMA_CM);
4308 ib_unregister_client(&cma_client);
4309 unregister_netdevice_notifier(&cma_nb);
4310 rdma_addr_unregister_client(&addr_client);
4311 ib_sa_unregister_client(&sa_client);
4312 unregister_pernet_subsys(&cma_pernet_operations);
4313 destroy_workqueue(cma_wq);
4314 }
4315
4316 module_init(cma_init);
4317 module_exit(cma_cleanup);
Linux kernel - Deliverables
This page took 0.124419 seconds and 5 git commands to generate.