projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
[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 if (req->pkey != req_param->bth_pkey)
1210 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1211 "RDMA CMA: in the future this may cause the request to be dropped\n",
1212 req_param->bth_pkey, req->pkey);
1213 break;
1214 case IB_CM_SIDR_REQ_RECEIVED:
1215 req->device = sidr_param->listen_id->device;
1216 req->port = sidr_param->port;
1217 req->has_gid = false;
1218 req->service_id = sidr_param->service_id;
1219 req->pkey = sidr_param->pkey;
1220 if (req->pkey != sidr_param->bth_pkey)
1221 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1222 "RDMA CMA: in the future this may cause the request to be dropped\n",
1223 sidr_param->bth_pkey, req->pkey);
1224 break;
1225 default:
1226 return -EINVAL;
1227 }
1228
1229 return 0;
1230 }
1231
1232 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1233 const struct sockaddr_in *dst_addr,
1234 const struct sockaddr_in *src_addr)
1235 {
1236 __be32 daddr = dst_addr->sin_addr.s_addr,
1237 saddr = src_addr->sin_addr.s_addr;
1238 struct fib_result res;
1239 struct flowi4 fl4;
1240 int err;
1241 bool ret;
1242
1243 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1244 ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1245 ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1246 ipv4_is_loopback(saddr))
1247 return false;
1248
1249 memset(&fl4, 0, sizeof(fl4));
1250 fl4.flowi4_iif = net_dev->ifindex;
1251 fl4.daddr = daddr;
1252 fl4.saddr = saddr;
1253
1254 rcu_read_lock();
1255 err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1256 ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1257 rcu_read_unlock();
1258
1259 return ret;
1260 }
1261
1262 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1263 const struct sockaddr_in6 *dst_addr,
1264 const struct sockaddr_in6 *src_addr)
1265 {
1266 #if IS_ENABLED(CONFIG_IPV6)
1267 const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1268 IPV6_ADDR_LINKLOCAL;
1269 struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1270 &src_addr->sin6_addr, net_dev->ifindex,
1271 strict);
1272 bool ret;
1273
1274 if (!rt)
1275 return false;
1276
1277 ret = rt->rt6i_idev->dev == net_dev;
1278 ip6_rt_put(rt);
1279
1280 return ret;
1281 #else
1282 return false;
1283 #endif
1284 }
1285
1286 static bool validate_net_dev(struct net_device *net_dev,
1287 const struct sockaddr *daddr,
1288 const struct sockaddr *saddr)
1289 {
1290 const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1291 const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1292 const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1293 const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1294
1295 switch (daddr->sa_family) {
1296 case AF_INET:
1297 return saddr->sa_family == AF_INET &&
1298 validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1299
1300 case AF_INET6:
1301 return saddr->sa_family == AF_INET6 &&
1302 validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1303
1304 default:
1305 return false;
1306 }
1307 }
1308
1309 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1310 const struct cma_req_info *req)
1311 {
1312 struct sockaddr_storage listen_addr_storage, src_addr_storage;
1313 struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1314 *src_addr = (struct sockaddr *)&src_addr_storage;
1315 struct net_device *net_dev;
1316 const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1317 int err;
1318
1319 err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1320 req->service_id);
1321 if (err)
1322 return ERR_PTR(err);
1323
1324 net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1325 gid, listen_addr);
1326 if (!net_dev)
1327 return ERR_PTR(-ENODEV);
1328
1329 if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1330 dev_put(net_dev);
1331 return ERR_PTR(-EHOSTUNREACH);
1332 }
1333
1334 return net_dev;
1335 }
1336
1337 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1338 {
1339 return (be64_to_cpu(service_id) >> 16) & 0xffff;
1340 }
1341
1342 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1343 const struct cma_hdr *hdr)
1344 {
1345 struct sockaddr *addr = cma_src_addr(id_priv);
1346 __be32 ip4_addr;
1347 struct in6_addr ip6_addr;
1348
1349 if (cma_any_addr(addr) && !id_priv->afonly)
1350 return true;
1351
1352 switch (addr->sa_family) {
1353 case AF_INET:
1354 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1355 if (cma_get_ip_ver(hdr) != 4)
1356 return false;
1357 if (!cma_any_addr(addr) &&
1358 hdr->dst_addr.ip4.addr != ip4_addr)
1359 return false;
1360 break;
1361 case AF_INET6:
1362 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1363 if (cma_get_ip_ver(hdr) != 6)
1364 return false;
1365 if (!cma_any_addr(addr) &&
1366 memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1367 return false;
1368 break;
1369 case AF_IB:
1370 return true;
1371 default:
1372 return false;
1373 }
1374
1375 return true;
1376 }
1377
1378 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1379 {
1380 enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1381 enum rdma_transport_type transport =
1382 rdma_node_get_transport(device->node_type);
1383
1384 return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1385 }
1386
1387 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1388 {
1389 struct ib_device *device = id->device;
1390 const int port_num = id->port_num ?: rdma_start_port(device);
1391
1392 return cma_protocol_roce_dev_port(device, port_num);
1393 }
1394
1395 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1396 const struct net_device *net_dev,
1397 u8 port_num)
1398 {
1399 const struct rdma_addr *addr = &id->route.addr;
1400
1401 if (!net_dev)
1402 /* This request is an AF_IB request or a RoCE request */
1403 return (!id->port_num || id->port_num == port_num) &&
1404 (addr->src_addr.ss_family == AF_IB ||
1405 cma_protocol_roce_dev_port(id->device, port_num));
1406
1407 return !addr->dev_addr.bound_dev_if ||
1408 (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1409 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1410 }
1411
1412 static struct rdma_id_private *cma_find_listener(
1413 const struct rdma_bind_list *bind_list,
1414 const struct ib_cm_id *cm_id,
1415 const struct ib_cm_event *ib_event,
1416 const struct cma_req_info *req,
1417 const struct net_device *net_dev)
1418 {
1419 struct rdma_id_private *id_priv, *id_priv_dev;
1420
1421 if (!bind_list)
1422 return ERR_PTR(-EINVAL);
1423
1424 hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1425 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1426 if (id_priv->id.device == cm_id->device &&
1427 cma_match_net_dev(&id_priv->id, net_dev, req->port))
1428 return id_priv;
1429 list_for_each_entry(id_priv_dev,
1430 &id_priv->listen_list,
1431 listen_list) {
1432 if (id_priv_dev->id.device == cm_id->device &&
1433 cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1434 return id_priv_dev;
1435 }
1436 }
1437 }
1438
1439 return ERR_PTR(-EINVAL);
1440 }
1441
1442 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1443 struct ib_cm_event *ib_event,
1444 struct net_device **net_dev)
1445 {
1446 struct cma_req_info req;
1447 struct rdma_bind_list *bind_list;
1448 struct rdma_id_private *id_priv;
1449 int err;
1450
1451 err = cma_save_req_info(ib_event, &req);
1452 if (err)
1453 return ERR_PTR(err);
1454
1455 *net_dev = cma_get_net_dev(ib_event, &req);
1456 if (IS_ERR(*net_dev)) {
1457 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1458 /* Assuming the protocol is AF_IB */
1459 *net_dev = NULL;
1460 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1461 /* TODO find the net dev matching the request parameters
1462 * through the RoCE GID table */
1463 *net_dev = NULL;
1464 } else {
1465 return ERR_CAST(*net_dev);
1466 }
1467 }
1468
1469 bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1470 rdma_ps_from_service_id(req.service_id),
1471 cma_port_from_service_id(req.service_id));
1472 id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1473 if (IS_ERR(id_priv) && *net_dev) {
1474 dev_put(*net_dev);
1475 *net_dev = NULL;
1476 }
1477
1478 return id_priv;
1479 }
1480
1481 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1482 {
1483 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1484 }
1485
1486 static void cma_cancel_route(struct rdma_id_private *id_priv)
1487 {
1488 if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1489 if (id_priv->query)
1490 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1491 }
1492 }
1493
1494 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1495 {
1496 struct rdma_id_private *dev_id_priv;
1497
1498 /*
1499 * Remove from listen_any_list to prevent added devices from spawning
1500 * additional listen requests.
1501 */
1502 mutex_lock(&lock);
1503 list_del(&id_priv->list);
1504
1505 while (!list_empty(&id_priv->listen_list)) {
1506 dev_id_priv = list_entry(id_priv->listen_list.next,
1507 struct rdma_id_private, listen_list);
1508 /* sync with device removal to avoid duplicate destruction */
1509 list_del_init(&dev_id_priv->list);
1510 list_del(&dev_id_priv->listen_list);
1511 mutex_unlock(&lock);
1512
1513 rdma_destroy_id(&dev_id_priv->id);
1514 mutex_lock(&lock);
1515 }
1516 mutex_unlock(&lock);
1517 }
1518
1519 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1520 enum rdma_cm_state state)
1521 {
1522 switch (state) {
1523 case RDMA_CM_ADDR_QUERY:
1524 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1525 break;
1526 case RDMA_CM_ROUTE_QUERY:
1527 cma_cancel_route(id_priv);
1528 break;
1529 case RDMA_CM_LISTEN:
1530 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1531 cma_cancel_listens(id_priv);
1532 break;
1533 default:
1534 break;
1535 }
1536 }
1537
1538 static void cma_release_port(struct rdma_id_private *id_priv)
1539 {
1540 struct rdma_bind_list *bind_list = id_priv->bind_list;
1541 struct net *net = id_priv->id.route.addr.dev_addr.net;
1542
1543 if (!bind_list)
1544 return;
1545
1546 mutex_lock(&lock);
1547 hlist_del(&id_priv->node);
1548 if (hlist_empty(&bind_list->owners)) {
1549 cma_ps_remove(net, bind_list->ps, bind_list->port);
1550 kfree(bind_list);
1551 }
1552 mutex_unlock(&lock);
1553 }
1554
1555 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1556 {
1557 struct cma_multicast *mc;
1558
1559 while (!list_empty(&id_priv->mc_list)) {
1560 mc = container_of(id_priv->mc_list.next,
1561 struct cma_multicast, list);
1562 list_del(&mc->list);
1563 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1564 id_priv->id.port_num)) {
1565 ib_sa_free_multicast(mc->multicast.ib);
1566 kfree(mc);
1567 } else {
1568 if (mc->igmp_joined) {
1569 struct rdma_dev_addr *dev_addr =
1570 &id_priv->id.route.addr.dev_addr;
1571 struct net_device *ndev = NULL;
1572
1573 if (dev_addr->bound_dev_if)
1574 ndev = dev_get_by_index(&init_net,
1575 dev_addr->bound_dev_if);
1576 if (ndev) {
1577 cma_igmp_send(ndev,
1578 &mc->multicast.ib->rec.mgid,
1579 false);
1580 dev_put(ndev);
1581 }
1582 }
1583 kref_put(&mc->mcref, release_mc);
1584 }
1585 }
1586 }
1587
1588 void rdma_destroy_id(struct rdma_cm_id *id)
1589 {
1590 struct rdma_id_private *id_priv;
1591 enum rdma_cm_state state;
1592
1593 id_priv = container_of(id, struct rdma_id_private, id);
1594 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1595 cma_cancel_operation(id_priv, state);
1596
1597 /*
1598 * Wait for any active callback to finish. New callbacks will find
1599 * the id_priv state set to destroying and abort.
1600 */
1601 mutex_lock(&id_priv->handler_mutex);
1602 mutex_unlock(&id_priv->handler_mutex);
1603
1604 if (id_priv->cma_dev) {
1605 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1606 if (id_priv->cm_id.ib)
1607 ib_destroy_cm_id(id_priv->cm_id.ib);
1608 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1609 if (id_priv->cm_id.iw)
1610 iw_destroy_cm_id(id_priv->cm_id.iw);
1611 }
1612 cma_leave_mc_groups(id_priv);
1613 cma_release_dev(id_priv);
1614 }
1615
1616 cma_release_port(id_priv);
1617 cma_deref_id(id_priv);
1618 wait_for_completion(&id_priv->comp);
1619
1620 if (id_priv->internal_id)
1621 cma_deref_id(id_priv->id.context);
1622
1623 kfree(id_priv->id.route.path_rec);
1624 put_net(id_priv->id.route.addr.dev_addr.net);
1625 kfree(id_priv);
1626 }
1627 EXPORT_SYMBOL(rdma_destroy_id);
1628
1629 static int cma_rep_recv(struct rdma_id_private *id_priv)
1630 {
1631 int ret;
1632
1633 ret = cma_modify_qp_rtr(id_priv, NULL);
1634 if (ret)
1635 goto reject;
1636
1637 ret = cma_modify_qp_rts(id_priv, NULL);
1638 if (ret)
1639 goto reject;
1640
1641 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1642 if (ret)
1643 goto reject;
1644
1645 return 0;
1646 reject:
1647 cma_modify_qp_err(id_priv);
1648 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1649 NULL, 0, NULL, 0);
1650 return ret;
1651 }
1652
1653 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1654 struct ib_cm_rep_event_param *rep_data,
1655 void *private_data)
1656 {
1657 event->param.conn.private_data = private_data;
1658 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1659 event->param.conn.responder_resources = rep_data->responder_resources;
1660 event->param.conn.initiator_depth = rep_data->initiator_depth;
1661 event->param.conn.flow_control = rep_data->flow_control;
1662 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1663 event->param.conn.srq = rep_data->srq;
1664 event->param.conn.qp_num = rep_data->remote_qpn;
1665 }
1666
1667 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1668 {
1669 struct rdma_id_private *id_priv = cm_id->context;
1670 struct rdma_cm_event event;
1671 int ret = 0;
1672
1673 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1674 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1675 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1676 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1677 return 0;
1678
1679 memset(&event, 0, sizeof event);
1680 switch (ib_event->event) {
1681 case IB_CM_REQ_ERROR:
1682 case IB_CM_REP_ERROR:
1683 event.event = RDMA_CM_EVENT_UNREACHABLE;
1684 event.status = -ETIMEDOUT;
1685 break;
1686 case IB_CM_REP_RECEIVED:
1687 if (id_priv->id.qp) {
1688 event.status = cma_rep_recv(id_priv);
1689 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1690 RDMA_CM_EVENT_ESTABLISHED;
1691 } else {
1692 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1693 }
1694 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1695 ib_event->private_data);
1696 break;
1697 case IB_CM_RTU_RECEIVED:
1698 case IB_CM_USER_ESTABLISHED:
1699 event.event = RDMA_CM_EVENT_ESTABLISHED;
1700 break;
1701 case IB_CM_DREQ_ERROR:
1702 event.status = -ETIMEDOUT; /* fall through */
1703 case IB_CM_DREQ_RECEIVED:
1704 case IB_CM_DREP_RECEIVED:
1705 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1706 RDMA_CM_DISCONNECT))
1707 goto out;
1708 event.event = RDMA_CM_EVENT_DISCONNECTED;
1709 break;
1710 case IB_CM_TIMEWAIT_EXIT:
1711 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1712 break;
1713 case IB_CM_MRA_RECEIVED:
1714 /* ignore event */
1715 goto out;
1716 case IB_CM_REJ_RECEIVED:
1717 cma_modify_qp_err(id_priv);
1718 event.status = ib_event->param.rej_rcvd.reason;
1719 event.event = RDMA_CM_EVENT_REJECTED;
1720 event.param.conn.private_data = ib_event->private_data;
1721 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1722 break;
1723 default:
1724 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1725 ib_event->event);
1726 goto out;
1727 }
1728
1729 ret = id_priv->id.event_handler(&id_priv->id, &event);
1730 if (ret) {
1731 /* Destroy the CM ID by returning a non-zero value. */
1732 id_priv->cm_id.ib = NULL;
1733 cma_exch(id_priv, RDMA_CM_DESTROYING);
1734 mutex_unlock(&id_priv->handler_mutex);
1735 rdma_destroy_id(&id_priv->id);
1736 return ret;
1737 }
1738 out:
1739 mutex_unlock(&id_priv->handler_mutex);
1740 return ret;
1741 }
1742
1743 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1744 struct ib_cm_event *ib_event,
1745 struct net_device *net_dev)
1746 {
1747 struct rdma_id_private *id_priv;
1748 struct rdma_cm_id *id;
1749 struct rdma_route *rt;
1750 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1751 const __be64 service_id =
1752 ib_event->param.req_rcvd.primary_path->service_id;
1753 int ret;
1754
1755 id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1756 listen_id->event_handler, listen_id->context,
1757 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1758 if (IS_ERR(id))
1759 return NULL;
1760
1761 id_priv = container_of(id, struct rdma_id_private, id);
1762 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1763 (struct sockaddr *)&id->route.addr.dst_addr,
1764 listen_id, ib_event, ss_family, service_id))
1765 goto err;
1766
1767 rt = &id->route;
1768 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1769 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1770 GFP_KERNEL);
1771 if (!rt->path_rec)
1772 goto err;
1773
1774 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1775 if (rt->num_paths == 2)
1776 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1777
1778 if (net_dev) {
1779 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1780 if (ret)
1781 goto err;
1782 } else {
1783 if (!cma_protocol_roce(listen_id) &&
1784 cma_any_addr(cma_src_addr(id_priv))) {
1785 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1786 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1787 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1788 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1789 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1790 if (ret)
1791 goto err;
1792 }
1793 }
1794 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1795
1796 id_priv->state = RDMA_CM_CONNECT;
1797 return id_priv;
1798
1799 err:
1800 rdma_destroy_id(id);
1801 return NULL;
1802 }
1803
1804 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1805 struct ib_cm_event *ib_event,
1806 struct net_device *net_dev)
1807 {
1808 struct rdma_id_private *id_priv;
1809 struct rdma_cm_id *id;
1810 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1811 struct net *net = listen_id->route.addr.dev_addr.net;
1812 int ret;
1813
1814 id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1815 listen_id->ps, IB_QPT_UD);
1816 if (IS_ERR(id))
1817 return NULL;
1818
1819 id_priv = container_of(id, struct rdma_id_private, id);
1820 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1821 (struct sockaddr *)&id->route.addr.dst_addr,
1822 listen_id, ib_event, ss_family,
1823 ib_event->param.sidr_req_rcvd.service_id))
1824 goto err;
1825
1826 if (net_dev) {
1827 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1828 if (ret)
1829 goto err;
1830 } else {
1831 if (!cma_any_addr(cma_src_addr(id_priv))) {
1832 ret = cma_translate_addr(cma_src_addr(id_priv),
1833 &id->route.addr.dev_addr);
1834 if (ret)
1835 goto err;
1836 }
1837 }
1838
1839 id_priv->state = RDMA_CM_CONNECT;
1840 return id_priv;
1841 err:
1842 rdma_destroy_id(id);
1843 return NULL;
1844 }
1845
1846 static void cma_set_req_event_data(struct rdma_cm_event *event,
1847 struct ib_cm_req_event_param *req_data,
1848 void *private_data, int offset)
1849 {
1850 event->param.conn.private_data = private_data + offset;
1851 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1852 event->param.conn.responder_resources = req_data->responder_resources;
1853 event->param.conn.initiator_depth = req_data->initiator_depth;
1854 event->param.conn.flow_control = req_data->flow_control;
1855 event->param.conn.retry_count = req_data->retry_count;
1856 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1857 event->param.conn.srq = req_data->srq;
1858 event->param.conn.qp_num = req_data->remote_qpn;
1859 }
1860
1861 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1862 {
1863 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1864 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1865 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1866 (id->qp_type == IB_QPT_UD)) ||
1867 (!id->qp_type));
1868 }
1869
1870 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1871 {
1872 struct rdma_id_private *listen_id, *conn_id;
1873 struct rdma_cm_event event;
1874 struct net_device *net_dev;
1875 int offset, ret;
1876
1877 listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1878 if (IS_ERR(listen_id))
1879 return PTR_ERR(listen_id);
1880
1881 if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1882 ret = -EINVAL;
1883 goto net_dev_put;
1884 }
1885
1886 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) {
1887 ret = -ECONNABORTED;
1888 goto net_dev_put;
1889 }
1890
1891 memset(&event, 0, sizeof event);
1892 offset = cma_user_data_offset(listen_id);
1893 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1894 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1895 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1896 event.param.ud.private_data = ib_event->private_data + offset;
1897 event.param.ud.private_data_len =
1898 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1899 } else {
1900 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1901 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1902 ib_event->private_data, offset);
1903 }
1904 if (!conn_id) {
1905 ret = -ENOMEM;
1906 goto err1;
1907 }
1908
1909 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1910 ret = cma_acquire_dev(conn_id, listen_id);
1911 if (ret)
1912 goto err2;
1913
1914 conn_id->cm_id.ib = cm_id;
1915 cm_id->context = conn_id;
1916 cm_id->cm_handler = cma_ib_handler;
1917
1918 /*
1919 * Protect against the user destroying conn_id from another thread
1920 * until we're done accessing it.
1921 */
1922 atomic_inc(&conn_id->refcount);
1923 ret = conn_id->id.event_handler(&conn_id->id, &event);
1924 if (ret)
1925 goto err3;
1926 /*
1927 * Acquire mutex to prevent user executing rdma_destroy_id()
1928 * while we're accessing the cm_id.
1929 */
1930 mutex_lock(&lock);
1931 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1932 (conn_id->id.qp_type != IB_QPT_UD))
1933 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1934 mutex_unlock(&lock);
1935 mutex_unlock(&conn_id->handler_mutex);
1936 mutex_unlock(&listen_id->handler_mutex);
1937 cma_deref_id(conn_id);
1938 if (net_dev)
1939 dev_put(net_dev);
1940 return 0;
1941
1942 err3:
1943 cma_deref_id(conn_id);
1944 /* Destroy the CM ID by returning a non-zero value. */
1945 conn_id->cm_id.ib = NULL;
1946 err2:
1947 cma_exch(conn_id, RDMA_CM_DESTROYING);
1948 mutex_unlock(&conn_id->handler_mutex);
1949 err1:
1950 mutex_unlock(&listen_id->handler_mutex);
1951 if (conn_id)
1952 rdma_destroy_id(&conn_id->id);
1953
1954 net_dev_put:
1955 if (net_dev)
1956 dev_put(net_dev);
1957
1958 return ret;
1959 }
1960
1961 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1962 {
1963 if (addr->sa_family == AF_IB)
1964 return ((struct sockaddr_ib *) addr)->sib_sid;
1965
1966 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1967 }
1968 EXPORT_SYMBOL(rdma_get_service_id);
1969
1970 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1971 {
1972 struct rdma_id_private *id_priv = iw_id->context;
1973 struct rdma_cm_event event;
1974 int ret = 0;
1975 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1976 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1977
1978 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1979 return 0;
1980
1981 memset(&event, 0, sizeof event);
1982 switch (iw_event->event) {
1983 case IW_CM_EVENT_CLOSE:
1984 event.event = RDMA_CM_EVENT_DISCONNECTED;
1985 break;
1986 case IW_CM_EVENT_CONNECT_REPLY:
1987 memcpy(cma_src_addr(id_priv), laddr,
1988 rdma_addr_size(laddr));
1989 memcpy(cma_dst_addr(id_priv), raddr,
1990 rdma_addr_size(raddr));
1991 switch (iw_event->status) {
1992 case 0:
1993 event.event = RDMA_CM_EVENT_ESTABLISHED;
1994 event.param.conn.initiator_depth = iw_event->ird;
1995 event.param.conn.responder_resources = iw_event->ord;
1996 break;
1997 case -ECONNRESET:
1998 case -ECONNREFUSED:
1999 event.event = RDMA_CM_EVENT_REJECTED;
2000 break;
2001 case -ETIMEDOUT:
2002 event.event = RDMA_CM_EVENT_UNREACHABLE;
2003 break;
2004 default:
2005 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2006 break;
2007 }
2008 break;
2009 case IW_CM_EVENT_ESTABLISHED:
2010 event.event = RDMA_CM_EVENT_ESTABLISHED;
2011 event.param.conn.initiator_depth = iw_event->ird;
2012 event.param.conn.responder_resources = iw_event->ord;
2013 break;
2014 default:
2015 BUG_ON(1);
2016 }
2017
2018 event.status = iw_event->status;
2019 event.param.conn.private_data = iw_event->private_data;
2020 event.param.conn.private_data_len = iw_event->private_data_len;
2021 ret = id_priv->id.event_handler(&id_priv->id, &event);
2022 if (ret) {
2023 /* Destroy the CM ID by returning a non-zero value. */
2024 id_priv->cm_id.iw = NULL;
2025 cma_exch(id_priv, RDMA_CM_DESTROYING);
2026 mutex_unlock(&id_priv->handler_mutex);
2027 rdma_destroy_id(&id_priv->id);
2028 return ret;
2029 }
2030
2031 mutex_unlock(&id_priv->handler_mutex);
2032 return ret;
2033 }
2034
2035 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2036 struct iw_cm_event *iw_event)
2037 {
2038 struct rdma_cm_id *new_cm_id;
2039 struct rdma_id_private *listen_id, *conn_id;
2040 struct rdma_cm_event event;
2041 int ret;
2042 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2043 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2044
2045 listen_id = cm_id->context;
2046 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
2047 return -ECONNABORTED;
2048
2049 /* Create a new RDMA id for the new IW CM ID */
2050 new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2051 listen_id->id.event_handler,
2052 listen_id->id.context,
2053 RDMA_PS_TCP, IB_QPT_RC);
2054 if (IS_ERR(new_cm_id)) {
2055 ret = -ENOMEM;
2056 goto out;
2057 }
2058 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2059 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2060 conn_id->state = RDMA_CM_CONNECT;
2061
2062 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2063 if (ret) {
2064 mutex_unlock(&conn_id->handler_mutex);
2065 rdma_destroy_id(new_cm_id);
2066 goto out;
2067 }
2068
2069 ret = cma_acquire_dev(conn_id, listen_id);
2070 if (ret) {
2071 mutex_unlock(&conn_id->handler_mutex);
2072 rdma_destroy_id(new_cm_id);
2073 goto out;
2074 }
2075
2076 conn_id->cm_id.iw = cm_id;
2077 cm_id->context = conn_id;
2078 cm_id->cm_handler = cma_iw_handler;
2079
2080 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2081 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2082
2083 memset(&event, 0, sizeof event);
2084 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2085 event.param.conn.private_data = iw_event->private_data;
2086 event.param.conn.private_data_len = iw_event->private_data_len;
2087 event.param.conn.initiator_depth = iw_event->ird;
2088 event.param.conn.responder_resources = iw_event->ord;
2089
2090 /*
2091 * Protect against the user destroying conn_id from another thread
2092 * until we're done accessing it.
2093 */
2094 atomic_inc(&conn_id->refcount);
2095 ret = conn_id->id.event_handler(&conn_id->id, &event);
2096 if (ret) {
2097 /* User wants to destroy the CM ID */
2098 conn_id->cm_id.iw = NULL;
2099 cma_exch(conn_id, RDMA_CM_DESTROYING);
2100 mutex_unlock(&conn_id->handler_mutex);
2101 cma_deref_id(conn_id);
2102 rdma_destroy_id(&conn_id->id);
2103 goto out;
2104 }
2105
2106 mutex_unlock(&conn_id->handler_mutex);
2107 cma_deref_id(conn_id);
2108
2109 out:
2110 mutex_unlock(&listen_id->handler_mutex);
2111 return ret;
2112 }
2113
2114 static int cma_ib_listen(struct rdma_id_private *id_priv)
2115 {
2116 struct sockaddr *addr;
2117 struct ib_cm_id *id;
2118 __be64 svc_id;
2119
2120 addr = cma_src_addr(id_priv);
2121 svc_id = rdma_get_service_id(&id_priv->id, addr);
2122 id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2123 if (IS_ERR(id))
2124 return PTR_ERR(id);
2125 id_priv->cm_id.ib = id;
2126
2127 return 0;
2128 }
2129
2130 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2131 {
2132 int ret;
2133 struct iw_cm_id *id;
2134
2135 id = iw_create_cm_id(id_priv->id.device,
2136 iw_conn_req_handler,
2137 id_priv);
2138 if (IS_ERR(id))
2139 return PTR_ERR(id);
2140
2141 id->tos = id_priv->tos;
2142 id_priv->cm_id.iw = id;
2143
2144 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2145 rdma_addr_size(cma_src_addr(id_priv)));
2146
2147 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2148
2149 if (ret) {
2150 iw_destroy_cm_id(id_priv->cm_id.iw);
2151 id_priv->cm_id.iw = NULL;
2152 }
2153
2154 return ret;
2155 }
2156
2157 static int cma_listen_handler(struct rdma_cm_id *id,
2158 struct rdma_cm_event *event)
2159 {
2160 struct rdma_id_private *id_priv = id->context;
2161
2162 id->context = id_priv->id.context;
2163 id->event_handler = id_priv->id.event_handler;
2164 return id_priv->id.event_handler(id, event);
2165 }
2166
2167 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2168 struct cma_device *cma_dev)
2169 {
2170 struct rdma_id_private *dev_id_priv;
2171 struct rdma_cm_id *id;
2172 struct net *net = id_priv->id.route.addr.dev_addr.net;
2173 int ret;
2174
2175 if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2176 return;
2177
2178 id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2179 id_priv->id.qp_type);
2180 if (IS_ERR(id))
2181 return;
2182
2183 dev_id_priv = container_of(id, struct rdma_id_private, id);
2184
2185 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2186 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2187 rdma_addr_size(cma_src_addr(id_priv)));
2188
2189 _cma_attach_to_dev(dev_id_priv, cma_dev);
2190 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2191 atomic_inc(&id_priv->refcount);
2192 dev_id_priv->internal_id = 1;
2193 dev_id_priv->afonly = id_priv->afonly;
2194
2195 ret = rdma_listen(id, id_priv->backlog);
2196 if (ret)
2197 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2198 ret, cma_dev->device->name);
2199 }
2200
2201 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2202 {
2203 struct cma_device *cma_dev;
2204
2205 mutex_lock(&lock);
2206 list_add_tail(&id_priv->list, &listen_any_list);
2207 list_for_each_entry(cma_dev, &dev_list, list)
2208 cma_listen_on_dev(id_priv, cma_dev);
2209 mutex_unlock(&lock);
2210 }
2211
2212 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2213 {
2214 struct rdma_id_private *id_priv;
2215
2216 id_priv = container_of(id, struct rdma_id_private, id);
2217 id_priv->tos = (u8) tos;
2218 }
2219 EXPORT_SYMBOL(rdma_set_service_type);
2220
2221 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2222 void *context)
2223 {
2224 struct cma_work *work = context;
2225 struct rdma_route *route;
2226
2227 route = &work->id->id.route;
2228
2229 if (!status) {
2230 route->num_paths = 1;
2231 *route->path_rec = *path_rec;
2232 } else {
2233 work->old_state = RDMA_CM_ROUTE_QUERY;
2234 work->new_state = RDMA_CM_ADDR_RESOLVED;
2235 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2236 work->event.status = status;
2237 }
2238
2239 queue_work(cma_wq, &work->work);
2240 }
2241
2242 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2243 struct cma_work *work)
2244 {
2245 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2246 struct ib_sa_path_rec path_rec;
2247 ib_sa_comp_mask comp_mask;
2248 struct sockaddr_in6 *sin6;
2249 struct sockaddr_ib *sib;
2250
2251 memset(&path_rec, 0, sizeof path_rec);
2252 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2253 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2254 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2255 path_rec.numb_path = 1;
2256 path_rec.reversible = 1;
2257 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2258
2259 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2260 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2261 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2262
2263 switch (cma_family(id_priv)) {
2264 case AF_INET:
2265 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2266 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2267 break;
2268 case AF_INET6:
2269 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2270 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2271 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2272 break;
2273 case AF_IB:
2274 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2275 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2276 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2277 break;
2278 }
2279
2280 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2281 id_priv->id.port_num, &path_rec,
2282 comp_mask, timeout_ms,
2283 GFP_KERNEL, cma_query_handler,
2284 work, &id_priv->query);
2285
2286 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2287 }
2288
2289 static void cma_work_handler(struct work_struct *_work)
2290 {
2291 struct cma_work *work = container_of(_work, struct cma_work, work);
2292 struct rdma_id_private *id_priv = work->id;
2293 int destroy = 0;
2294
2295 mutex_lock(&id_priv->handler_mutex);
2296 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2297 goto out;
2298
2299 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2300 cma_exch(id_priv, RDMA_CM_DESTROYING);
2301 destroy = 1;
2302 }
2303 out:
2304 mutex_unlock(&id_priv->handler_mutex);
2305 cma_deref_id(id_priv);
2306 if (destroy)
2307 rdma_destroy_id(&id_priv->id);
2308 kfree(work);
2309 }
2310
2311 static void cma_ndev_work_handler(struct work_struct *_work)
2312 {
2313 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2314 struct rdma_id_private *id_priv = work->id;
2315 int destroy = 0;
2316
2317 mutex_lock(&id_priv->handler_mutex);
2318 if (id_priv->state == RDMA_CM_DESTROYING ||
2319 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2320 goto out;
2321
2322 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2323 cma_exch(id_priv, RDMA_CM_DESTROYING);
2324 destroy = 1;
2325 }
2326
2327 out:
2328 mutex_unlock(&id_priv->handler_mutex);
2329 cma_deref_id(id_priv);
2330 if (destroy)
2331 rdma_destroy_id(&id_priv->id);
2332 kfree(work);
2333 }
2334
2335 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2336 {
2337 struct rdma_route *route = &id_priv->id.route;
2338 struct cma_work *work;
2339 int ret;
2340
2341 work = kzalloc(sizeof *work, GFP_KERNEL);
2342 if (!work)
2343 return -ENOMEM;
2344
2345 work->id = id_priv;
2346 INIT_WORK(&work->work, cma_work_handler);
2347 work->old_state = RDMA_CM_ROUTE_QUERY;
2348 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2349 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2350
2351 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2352 if (!route->path_rec) {
2353 ret = -ENOMEM;
2354 goto err1;
2355 }
2356
2357 ret = cma_query_ib_route(id_priv, timeout_ms, work);
2358 if (ret)
2359 goto err2;
2360
2361 return 0;
2362 err2:
2363 kfree(route->path_rec);
2364 route->path_rec = NULL;
2365 err1:
2366 kfree(work);
2367 return ret;
2368 }
2369
2370 int rdma_set_ib_paths(struct rdma_cm_id *id,
2371 struct ib_sa_path_rec *path_rec, int num_paths)
2372 {
2373 struct rdma_id_private *id_priv;
2374 int ret;
2375
2376 id_priv = container_of(id, struct rdma_id_private, id);
2377 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2378 RDMA_CM_ROUTE_RESOLVED))
2379 return -EINVAL;
2380
2381 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2382 GFP_KERNEL);
2383 if (!id->route.path_rec) {
2384 ret = -ENOMEM;
2385 goto err;
2386 }
2387
2388 id->route.num_paths = num_paths;
2389 return 0;
2390 err:
2391 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2392 return ret;
2393 }
2394 EXPORT_SYMBOL(rdma_set_ib_paths);
2395
2396 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2397 {
2398 struct cma_work *work;
2399
2400 work = kzalloc(sizeof *work, GFP_KERNEL);
2401 if (!work)
2402 return -ENOMEM;
2403
2404 work->id = id_priv;
2405 INIT_WORK(&work->work, cma_work_handler);
2406 work->old_state = RDMA_CM_ROUTE_QUERY;
2407 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2408 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2409 queue_work(cma_wq, &work->work);
2410 return 0;
2411 }
2412
2413 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2414 {
2415 int prio;
2416 struct net_device *dev;
2417
2418 prio = rt_tos2priority(tos);
2419 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2420 vlan_dev_real_dev(ndev) : ndev;
2421
2422 if (dev->num_tc)
2423 return netdev_get_prio_tc_map(dev, prio);
2424
2425 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2426 if (ndev->priv_flags & IFF_802_1Q_VLAN)
2427 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2428 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2429 #endif
2430 return 0;
2431 }
2432
2433 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2434 {
2435 struct rdma_route *route = &id_priv->id.route;
2436 struct rdma_addr *addr = &route->addr;
2437 struct cma_work *work;
2438 int ret;
2439 struct net_device *ndev = NULL;
2440
2441
2442 work = kzalloc(sizeof *work, GFP_KERNEL);
2443 if (!work)
2444 return -ENOMEM;
2445
2446 work->id = id_priv;
2447 INIT_WORK(&work->work, cma_work_handler);
2448
2449 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2450 if (!route->path_rec) {
2451 ret = -ENOMEM;
2452 goto err1;
2453 }
2454
2455 route->num_paths = 1;
2456
2457 if (addr->dev_addr.bound_dev_if) {
2458 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2459 if (!ndev)
2460 return -ENODEV;
2461
2462 if (ndev->flags & IFF_LOOPBACK) {
2463 dev_put(ndev);
2464 if (!id_priv->id.device->get_netdev)
2465 return -EOPNOTSUPP;
2466
2467 ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2468 id_priv->id.port_num);
2469 if (!ndev)
2470 return -ENODEV;
2471 }
2472
2473 route->path_rec->net = &init_net;
2474 route->path_rec->ifindex = ndev->ifindex;
2475 route->path_rec->gid_type = id_priv->gid_type;
2476 }
2477 if (!ndev) {
2478 ret = -ENODEV;
2479 goto err2;
2480 }
2481
2482 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2483
2484 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2485 &route->path_rec->sgid);
2486 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2487 &route->path_rec->dgid);
2488
2489 /* Use the hint from IP Stack to select GID Type */
2490 if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2491 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2492 if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2493 /* TODO: get the hoplimit from the inet/inet6 device */
2494 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2495 else
2496 route->path_rec->hop_limit = 1;
2497 route->path_rec->reversible = 1;
2498 route->path_rec->pkey = cpu_to_be16(0xffff);
2499 route->path_rec->mtu_selector = IB_SA_EQ;
2500 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2501 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2502 route->path_rec->rate_selector = IB_SA_EQ;
2503 route->path_rec->rate = iboe_get_rate(ndev);
2504 dev_put(ndev);
2505 route->path_rec->packet_life_time_selector = IB_SA_EQ;
2506 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2507 if (!route->path_rec->mtu) {
2508 ret = -EINVAL;
2509 goto err2;
2510 }
2511
2512 work->old_state = RDMA_CM_ROUTE_QUERY;
2513 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2514 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2515 work->event.status = 0;
2516
2517 queue_work(cma_wq, &work->work);
2518
2519 return 0;
2520
2521 err2:
2522 kfree(route->path_rec);
2523 route->path_rec = NULL;
2524 err1:
2525 kfree(work);
2526 return ret;
2527 }
2528
2529 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2530 {
2531 struct rdma_id_private *id_priv;
2532 int ret;
2533
2534 id_priv = container_of(id, struct rdma_id_private, id);
2535 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2536 return -EINVAL;
2537
2538 atomic_inc(&id_priv->refcount);
2539 if (rdma_cap_ib_sa(id->device, id->port_num))
2540 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2541 else if (rdma_protocol_roce(id->device, id->port_num))
2542 ret = cma_resolve_iboe_route(id_priv);
2543 else if (rdma_protocol_iwarp(id->device, id->port_num))
2544 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2545 else
2546 ret = -ENOSYS;
2547
2548 if (ret)
2549 goto err;
2550
2551 return 0;
2552 err:
2553 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2554 cma_deref_id(id_priv);
2555 return ret;
2556 }
2557 EXPORT_SYMBOL(rdma_resolve_route);
2558
2559 static void cma_set_loopback(struct sockaddr *addr)
2560 {
2561 switch (addr->sa_family) {
2562 case AF_INET:
2563 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2564 break;
2565 case AF_INET6:
2566 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2567 0, 0, 0, htonl(1));
2568 break;
2569 default:
2570 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2571 0, 0, 0, htonl(1));
2572 break;
2573 }
2574 }
2575
2576 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2577 {
2578 struct cma_device *cma_dev, *cur_dev;
2579 struct ib_port_attr port_attr;
2580 union ib_gid gid;
2581 u16 pkey;
2582 int ret;
2583 u8 p;
2584
2585 cma_dev = NULL;
2586 mutex_lock(&lock);
2587 list_for_each_entry(cur_dev, &dev_list, list) {
2588 if (cma_family(id_priv) == AF_IB &&
2589 !rdma_cap_ib_cm(cur_dev->device, 1))
2590 continue;
2591
2592 if (!cma_dev)
2593 cma_dev = cur_dev;
2594
2595 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2596 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2597 port_attr.state == IB_PORT_ACTIVE) {
2598 cma_dev = cur_dev;
2599 goto port_found;
2600 }
2601 }
2602 }
2603
2604 if (!cma_dev) {
2605 ret = -ENODEV;
2606 goto out;
2607 }
2608
2609 p = 1;
2610
2611 port_found:
2612 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2613 if (ret)
2614 goto out;
2615
2616 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2617 if (ret)
2618 goto out;
2619
2620 id_priv->id.route.addr.dev_addr.dev_type =
2621 (rdma_protocol_ib(cma_dev->device, p)) ?
2622 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2623
2624 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2625 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2626 id_priv->id.port_num = p;
2627 cma_attach_to_dev(id_priv, cma_dev);
2628 cma_set_loopback(cma_src_addr(id_priv));
2629 out:
2630 mutex_unlock(&lock);
2631 return ret;
2632 }
2633
2634 static void addr_handler(int status, struct sockaddr *src_addr,
2635 struct rdma_dev_addr *dev_addr, void *context)
2636 {
2637 struct rdma_id_private *id_priv = context;
2638 struct rdma_cm_event event;
2639
2640 memset(&event, 0, sizeof event);
2641 mutex_lock(&id_priv->handler_mutex);
2642 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2643 RDMA_CM_ADDR_RESOLVED))
2644 goto out;
2645
2646 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2647 if (!status && !id_priv->cma_dev)
2648 status = cma_acquire_dev(id_priv, NULL);
2649
2650 if (status) {
2651 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2652 RDMA_CM_ADDR_BOUND))
2653 goto out;
2654 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2655 event.status = status;
2656 } else
2657 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2658
2659 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2660 cma_exch(id_priv, RDMA_CM_DESTROYING);
2661 mutex_unlock(&id_priv->handler_mutex);
2662 cma_deref_id(id_priv);
2663 rdma_destroy_id(&id_priv->id);
2664 return;
2665 }
2666 out:
2667 mutex_unlock(&id_priv->handler_mutex);
2668 cma_deref_id(id_priv);
2669 }
2670
2671 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2672 {
2673 struct cma_work *work;
2674 union ib_gid gid;
2675 int ret;
2676
2677 work = kzalloc(sizeof *work, GFP_KERNEL);
2678 if (!work)
2679 return -ENOMEM;
2680
2681 if (!id_priv->cma_dev) {
2682 ret = cma_bind_loopback(id_priv);
2683 if (ret)
2684 goto err;
2685 }
2686
2687 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2688 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2689
2690 work->id = id_priv;
2691 INIT_WORK(&work->work, cma_work_handler);
2692 work->old_state = RDMA_CM_ADDR_QUERY;
2693 work->new_state = RDMA_CM_ADDR_RESOLVED;
2694 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2695 queue_work(cma_wq, &work->work);
2696 return 0;
2697 err:
2698 kfree(work);
2699 return ret;
2700 }
2701
2702 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2703 {
2704 struct cma_work *work;
2705 int ret;
2706
2707 work = kzalloc(sizeof *work, GFP_KERNEL);
2708 if (!work)
2709 return -ENOMEM;
2710
2711 if (!id_priv->cma_dev) {
2712 ret = cma_resolve_ib_dev(id_priv);
2713 if (ret)
2714 goto err;
2715 }
2716
2717 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2718 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2719
2720 work->id = id_priv;
2721 INIT_WORK(&work->work, cma_work_handler);
2722 work->old_state = RDMA_CM_ADDR_QUERY;
2723 work->new_state = RDMA_CM_ADDR_RESOLVED;
2724 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2725 queue_work(cma_wq, &work->work);
2726 return 0;
2727 err:
2728 kfree(work);
2729 return ret;
2730 }
2731
2732 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2733 struct sockaddr *dst_addr)
2734 {
2735 if (!src_addr || !src_addr->sa_family) {
2736 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2737 src_addr->sa_family = dst_addr->sa_family;
2738 if (dst_addr->sa_family == AF_INET6) {
2739 struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2740 struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2741 src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2742 if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2743 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2744 } else if (dst_addr->sa_family == AF_IB) {
2745 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2746 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2747 }
2748 }
2749 return rdma_bind_addr(id, src_addr);
2750 }
2751
2752 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2753 struct sockaddr *dst_addr, int timeout_ms)
2754 {
2755 struct rdma_id_private *id_priv;
2756 int ret;
2757
2758 id_priv = container_of(id, struct rdma_id_private, id);
2759 if (id_priv->state == RDMA_CM_IDLE) {
2760 ret = cma_bind_addr(id, src_addr, dst_addr);
2761 if (ret)
2762 return ret;
2763 }
2764
2765 if (cma_family(id_priv) != dst_addr->sa_family)
2766 return -EINVAL;
2767
2768 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2769 return -EINVAL;
2770
2771 atomic_inc(&id_priv->refcount);
2772 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2773 if (cma_any_addr(dst_addr)) {
2774 ret = cma_resolve_loopback(id_priv);
2775 } else {
2776 if (dst_addr->sa_family == AF_IB) {
2777 ret = cma_resolve_ib_addr(id_priv);
2778 } else {
2779 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2780 dst_addr, &id->route.addr.dev_addr,
2781 timeout_ms, addr_handler, id_priv);
2782 }
2783 }
2784 if (ret)
2785 goto err;
2786
2787 return 0;
2788 err:
2789 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2790 cma_deref_id(id_priv);
2791 return ret;
2792 }
2793 EXPORT_SYMBOL(rdma_resolve_addr);
2794
2795 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2796 {
2797 struct rdma_id_private *id_priv;
2798 unsigned long flags;
2799 int ret;
2800
2801 id_priv = container_of(id, struct rdma_id_private, id);
2802 spin_lock_irqsave(&id_priv->lock, flags);
2803 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2804 id_priv->reuseaddr = reuse;
2805 ret = 0;
2806 } else {
2807 ret = -EINVAL;
2808 }
2809 spin_unlock_irqrestore(&id_priv->lock, flags);
2810 return ret;
2811 }
2812 EXPORT_SYMBOL(rdma_set_reuseaddr);
2813
2814 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2815 {
2816 struct rdma_id_private *id_priv;
2817 unsigned long flags;
2818 int ret;
2819
2820 id_priv = container_of(id, struct rdma_id_private, id);
2821 spin_lock_irqsave(&id_priv->lock, flags);
2822 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2823 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2824 id_priv->afonly = afonly;
2825 ret = 0;
2826 } else {
2827 ret = -EINVAL;
2828 }
2829 spin_unlock_irqrestore(&id_priv->lock, flags);
2830 return ret;
2831 }
2832 EXPORT_SYMBOL(rdma_set_afonly);
2833
2834 static void cma_bind_port(struct rdma_bind_list *bind_list,
2835 struct rdma_id_private *id_priv)
2836 {
2837 struct sockaddr *addr;
2838 struct sockaddr_ib *sib;
2839 u64 sid, mask;
2840 __be16 port;
2841
2842 addr = cma_src_addr(id_priv);
2843 port = htons(bind_list->port);
2844
2845 switch (addr->sa_family) {
2846 case AF_INET:
2847 ((struct sockaddr_in *) addr)->sin_port = port;
2848 break;
2849 case AF_INET6:
2850 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2851 break;
2852 case AF_IB:
2853 sib = (struct sockaddr_ib *) addr;
2854 sid = be64_to_cpu(sib->sib_sid);
2855 mask = be64_to_cpu(sib->sib_sid_mask);
2856 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2857 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2858 break;
2859 }
2860 id_priv->bind_list = bind_list;
2861 hlist_add_head(&id_priv->node, &bind_list->owners);
2862 }
2863
2864 static int cma_alloc_port(enum rdma_port_space ps,
2865 struct rdma_id_private *id_priv, unsigned short snum)
2866 {
2867 struct rdma_bind_list *bind_list;
2868 int ret;
2869
2870 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2871 if (!bind_list)
2872 return -ENOMEM;
2873
2874 ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2875 snum);
2876 if (ret < 0)
2877 goto err;
2878
2879 bind_list->ps = ps;
2880 bind_list->port = (unsigned short)ret;
2881 cma_bind_port(bind_list, id_priv);
2882 return 0;
2883 err:
2884 kfree(bind_list);
2885 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2886 }
2887
2888 static int cma_alloc_any_port(enum rdma_port_space ps,
2889 struct rdma_id_private *id_priv)
2890 {
2891 static unsigned int last_used_port;
2892 int low, high, remaining;
2893 unsigned int rover;
2894 struct net *net = id_priv->id.route.addr.dev_addr.net;
2895
2896 inet_get_local_port_range(net, &low, &high);
2897 remaining = (high - low) + 1;
2898 rover = prandom_u32() % remaining + low;
2899 retry:
2900 if (last_used_port != rover &&
2901 !cma_ps_find(net, ps, (unsigned short)rover)) {
2902 int ret = cma_alloc_port(ps, id_priv, rover);
2903 /*
2904 * Remember previously used port number in order to avoid
2905 * re-using same port immediately after it is closed.
2906 */
2907 if (!ret)
2908 last_used_port = rover;
2909 if (ret != -EADDRNOTAVAIL)
2910 return ret;
2911 }
2912 if (--remaining) {
2913 rover++;
2914 if ((rover < low) || (rover > high))
2915 rover = low;
2916 goto retry;
2917 }
2918 return -EADDRNOTAVAIL;
2919 }
2920
2921 /*
2922 * Check that the requested port is available. This is called when trying to
2923 * bind to a specific port, or when trying to listen on a bound port. In
2924 * the latter case, the provided id_priv may already be on the bind_list, but
2925 * we still need to check that it's okay to start listening.
2926 */
2927 static int cma_check_port(struct rdma_bind_list *bind_list,
2928 struct rdma_id_private *id_priv, uint8_t reuseaddr)
2929 {
2930 struct rdma_id_private *cur_id;
2931 struct sockaddr *addr, *cur_addr;
2932
2933 addr = cma_src_addr(id_priv);
2934 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2935 if (id_priv == cur_id)
2936 continue;
2937
2938 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2939 cur_id->reuseaddr)
2940 continue;
2941
2942 cur_addr = cma_src_addr(cur_id);
2943 if (id_priv->afonly && cur_id->afonly &&
2944 (addr->sa_family != cur_addr->sa_family))
2945 continue;
2946
2947 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2948 return -EADDRNOTAVAIL;
2949
2950 if (!cma_addr_cmp(addr, cur_addr))
2951 return -EADDRINUSE;
2952 }
2953 return 0;
2954 }
2955
2956 static int cma_use_port(enum rdma_port_space ps,
2957 struct rdma_id_private *id_priv)
2958 {
2959 struct rdma_bind_list *bind_list;
2960 unsigned short snum;
2961 int ret;
2962
2963 snum = ntohs(cma_port(cma_src_addr(id_priv)));
2964 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2965 return -EACCES;
2966
2967 bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
2968 if (!bind_list) {
2969 ret = cma_alloc_port(ps, id_priv, snum);
2970 } else {
2971 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2972 if (!ret)
2973 cma_bind_port(bind_list, id_priv);
2974 }
2975 return ret;
2976 }
2977
2978 static int cma_bind_listen(struct rdma_id_private *id_priv)
2979 {
2980 struct rdma_bind_list *bind_list = id_priv->bind_list;
2981 int ret = 0;
2982
2983 mutex_lock(&lock);
2984 if (bind_list->owners.first->next)
2985 ret = cma_check_port(bind_list, id_priv, 0);
2986 mutex_unlock(&lock);
2987 return ret;
2988 }
2989
2990 static enum rdma_port_space cma_select_inet_ps(
2991 struct rdma_id_private *id_priv)
2992 {
2993 switch (id_priv->id.ps) {
2994 case RDMA_PS_TCP:
2995 case RDMA_PS_UDP:
2996 case RDMA_PS_IPOIB:
2997 case RDMA_PS_IB:
2998 return id_priv->id.ps;
2999 default:
3000
3001 return 0;
3002 }
3003 }
3004
3005 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3006 {
3007 enum rdma_port_space ps = 0;
3008 struct sockaddr_ib *sib;
3009 u64 sid_ps, mask, sid;
3010
3011 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3012 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3013 sid = be64_to_cpu(sib->sib_sid) & mask;
3014
3015 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3016 sid_ps = RDMA_IB_IP_PS_IB;
3017 ps = RDMA_PS_IB;
3018 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3019 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3020 sid_ps = RDMA_IB_IP_PS_TCP;
3021 ps = RDMA_PS_TCP;
3022 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3023 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3024 sid_ps = RDMA_IB_IP_PS_UDP;
3025 ps = RDMA_PS_UDP;
3026 }
3027
3028 if (ps) {
3029 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3030 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3031 be64_to_cpu(sib->sib_sid_mask));
3032 }
3033 return ps;
3034 }
3035
3036 static int cma_get_port(struct rdma_id_private *id_priv)
3037 {
3038 enum rdma_port_space ps;
3039 int ret;
3040
3041 if (cma_family(id_priv) != AF_IB)
3042 ps = cma_select_inet_ps(id_priv);
3043 else
3044 ps = cma_select_ib_ps(id_priv);
3045 if (!ps)
3046 return -EPROTONOSUPPORT;
3047
3048 mutex_lock(&lock);
3049 if (cma_any_port(cma_src_addr(id_priv)))
3050 ret = cma_alloc_any_port(ps, id_priv);
3051 else
3052 ret = cma_use_port(ps, id_priv);
3053 mutex_unlock(&lock);
3054
3055 return ret;
3056 }
3057
3058 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3059 struct sockaddr *addr)
3060 {
3061 #if IS_ENABLED(CONFIG_IPV6)
3062 struct sockaddr_in6 *sin6;
3063
3064 if (addr->sa_family != AF_INET6)
3065 return 0;
3066
3067 sin6 = (struct sockaddr_in6 *) addr;
3068
3069 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3070 return 0;
3071
3072 if (!sin6->sin6_scope_id)
3073 return -EINVAL;
3074
3075 dev_addr->bound_dev_if = sin6->sin6_scope_id;
3076 #endif
3077 return 0;
3078 }
3079
3080 int rdma_listen(struct rdma_cm_id *id, int backlog)
3081 {
3082 struct rdma_id_private *id_priv;
3083 int ret;
3084
3085 id_priv = container_of(id, struct rdma_id_private, id);
3086 if (id_priv->state == RDMA_CM_IDLE) {
3087 id->route.addr.src_addr.ss_family = AF_INET;
3088 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3089 if (ret)
3090 return ret;
3091 }
3092
3093 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3094 return -EINVAL;
3095
3096 if (id_priv->reuseaddr) {
3097 ret = cma_bind_listen(id_priv);
3098 if (ret)
3099 goto err;
3100 }
3101
3102 id_priv->backlog = backlog;
3103 if (id->device) {
3104 if (rdma_cap_ib_cm(id->device, 1)) {
3105 ret = cma_ib_listen(id_priv);
3106 if (ret)
3107 goto err;
3108 } else if (rdma_cap_iw_cm(id->device, 1)) {
3109 ret = cma_iw_listen(id_priv, backlog);
3110 if (ret)
3111 goto err;
3112 } else {
3113 ret = -ENOSYS;
3114 goto err;
3115 }
3116 } else
3117 cma_listen_on_all(id_priv);
3118
3119 return 0;
3120 err:
3121 id_priv->backlog = 0;
3122 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3123 return ret;
3124 }
3125 EXPORT_SYMBOL(rdma_listen);
3126
3127 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3128 {
3129 struct rdma_id_private *id_priv;
3130 int ret;
3131
3132 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3133 addr->sa_family != AF_IB)
3134 return -EAFNOSUPPORT;
3135
3136 id_priv = container_of(id, struct rdma_id_private, id);
3137 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3138 return -EINVAL;
3139
3140 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3141 if (ret)
3142 goto err1;
3143
3144 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3145 if (!cma_any_addr(addr)) {
3146 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3147 if (ret)
3148 goto err1;
3149
3150 ret = cma_acquire_dev(id_priv, NULL);
3151 if (ret)
3152 goto err1;
3153 }
3154
3155 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3156 if (addr->sa_family == AF_INET)
3157 id_priv->afonly = 1;
3158 #if IS_ENABLED(CONFIG_IPV6)
3159 else if (addr->sa_family == AF_INET6) {
3160 struct net *net = id_priv->id.route.addr.dev_addr.net;
3161
3162 id_priv->afonly = net->ipv6.sysctl.bindv6only;
3163 }
3164 #endif
3165 }
3166 ret = cma_get_port(id_priv);
3167 if (ret)
3168 goto err2;
3169
3170 return 0;
3171 err2:
3172 if (id_priv->cma_dev)
3173 cma_release_dev(id_priv);
3174 err1:
3175 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3176 return ret;
3177 }
3178 EXPORT_SYMBOL(rdma_bind_addr);
3179
3180 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3181 {
3182 struct cma_hdr *cma_hdr;
3183
3184 cma_hdr = hdr;
3185 cma_hdr->cma_version = CMA_VERSION;
3186 if (cma_family(id_priv) == AF_INET) {
3187 struct sockaddr_in *src4, *dst4;
3188
3189 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3190 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3191
3192 cma_set_ip_ver(cma_hdr, 4);
3193 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3194 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3195 cma_hdr->port = src4->sin_port;
3196 } else if (cma_family(id_priv) == AF_INET6) {
3197 struct sockaddr_in6 *src6, *dst6;
3198
3199 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3200 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3201
3202 cma_set_ip_ver(cma_hdr, 6);
3203 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3204 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3205 cma_hdr->port = src6->sin6_port;
3206 }
3207 return 0;
3208 }
3209
3210 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3211 struct ib_cm_event *ib_event)
3212 {
3213 struct rdma_id_private *id_priv = cm_id->context;
3214 struct rdma_cm_event event;
3215 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3216 int ret = 0;
3217
3218 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
3219 return 0;
3220
3221 memset(&event, 0, sizeof event);
3222 switch (ib_event->event) {
3223 case IB_CM_SIDR_REQ_ERROR:
3224 event.event = RDMA_CM_EVENT_UNREACHABLE;
3225 event.status = -ETIMEDOUT;
3226 break;
3227 case IB_CM_SIDR_REP_RECEIVED:
3228 event.param.ud.private_data = ib_event->private_data;
3229 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3230 if (rep->status != IB_SIDR_SUCCESS) {
3231 event.event = RDMA_CM_EVENT_UNREACHABLE;
3232 event.status = ib_event->param.sidr_rep_rcvd.status;
3233 break;
3234 }
3235 ret = cma_set_qkey(id_priv, rep->qkey);
3236 if (ret) {
3237 event.event = RDMA_CM_EVENT_ADDR_ERROR;
3238 event.status = ret;
3239 break;
3240 }
3241 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3242 id_priv->id.route.path_rec,
3243 &event.param.ud.ah_attr);
3244 event.param.ud.qp_num = rep->qpn;
3245 event.param.ud.qkey = rep->qkey;
3246 event.event = RDMA_CM_EVENT_ESTABLISHED;
3247 event.status = 0;
3248 break;
3249 default:
3250 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3251 ib_event->event);
3252 goto out;
3253 }
3254
3255 ret = id_priv->id.event_handler(&id_priv->id, &event);
3256 if (ret) {
3257 /* Destroy the CM ID by returning a non-zero value. */
3258 id_priv->cm_id.ib = NULL;
3259 cma_exch(id_priv, RDMA_CM_DESTROYING);
3260 mutex_unlock(&id_priv->handler_mutex);
3261 rdma_destroy_id(&id_priv->id);
3262 return ret;
3263 }
3264 out:
3265 mutex_unlock(&id_priv->handler_mutex);
3266 return ret;
3267 }
3268
3269 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3270 struct rdma_conn_param *conn_param)
3271 {
3272 struct ib_cm_sidr_req_param req;
3273 struct ib_cm_id *id;
3274 void *private_data;
3275 int offset, ret;
3276
3277 memset(&req, 0, sizeof req);
3278 offset = cma_user_data_offset(id_priv);
3279 req.private_data_len = offset + conn_param->private_data_len;
3280 if (req.private_data_len < conn_param->private_data_len)
3281 return -EINVAL;
3282
3283 if (req.private_data_len) {
3284 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3285 if (!private_data)
3286 return -ENOMEM;
3287 } else {
3288 private_data = NULL;
3289 }
3290
3291 if (conn_param->private_data && conn_param->private_data_len)
3292 memcpy(private_data + offset, conn_param->private_data,
3293 conn_param->private_data_len);
3294
3295 if (private_data) {
3296 ret = cma_format_hdr(private_data, id_priv);
3297 if (ret)
3298 goto out;
3299 req.private_data = private_data;
3300 }
3301
3302 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3303 id_priv);
3304 if (IS_ERR(id)) {
3305 ret = PTR_ERR(id);
3306 goto out;
3307 }
3308 id_priv->cm_id.ib = id;
3309
3310 req.path = id_priv->id.route.path_rec;
3311 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3312 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3313 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3314
3315 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3316 if (ret) {
3317 ib_destroy_cm_id(id_priv->cm_id.ib);
3318 id_priv->cm_id.ib = NULL;
3319 }
3320 out:
3321 kfree(private_data);
3322 return ret;
3323 }
3324
3325 static int cma_connect_ib(struct rdma_id_private *id_priv,
3326 struct rdma_conn_param *conn_param)
3327 {
3328 struct ib_cm_req_param req;
3329 struct rdma_route *route;
3330 void *private_data;
3331 struct ib_cm_id *id;
3332 int offset, ret;
3333
3334 memset(&req, 0, sizeof req);
3335 offset = cma_user_data_offset(id_priv);
3336 req.private_data_len = offset + conn_param->private_data_len;
3337 if (req.private_data_len < conn_param->private_data_len)
3338 return -EINVAL;
3339
3340 if (req.private_data_len) {
3341 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3342 if (!private_data)
3343 return -ENOMEM;
3344 } else {
3345 private_data = NULL;
3346 }
3347
3348 if (conn_param->private_data && conn_param->private_data_len)
3349 memcpy(private_data + offset, conn_param->private_data,
3350 conn_param->private_data_len);
3351
3352 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3353 if (IS_ERR(id)) {
3354 ret = PTR_ERR(id);
3355 goto out;
3356 }
3357 id_priv->cm_id.ib = id;
3358
3359 route = &id_priv->id.route;
3360 if (private_data) {
3361 ret = cma_format_hdr(private_data, id_priv);
3362 if (ret)
3363 goto out;
3364 req.private_data = private_data;
3365 }
3366
3367 req.primary_path = &route->path_rec[0];
3368 if (route->num_paths == 2)
3369 req.alternate_path = &route->path_rec[1];
3370
3371 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3372 req.qp_num = id_priv->qp_num;
3373 req.qp_type = id_priv->id.qp_type;
3374 req.starting_psn = id_priv->seq_num;
3375 req.responder_resources = conn_param->responder_resources;
3376 req.initiator_depth = conn_param->initiator_depth;
3377 req.flow_control = conn_param->flow_control;
3378 req.retry_count = min_t(u8, 7, conn_param->retry_count);
3379 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3380 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3381 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3382 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3383 req.srq = id_priv->srq ? 1 : 0;
3384
3385 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3386 out:
3387 if (ret && !IS_ERR(id)) {
3388 ib_destroy_cm_id(id);
3389 id_priv->cm_id.ib = NULL;
3390 }
3391
3392 kfree(private_data);
3393 return ret;
3394 }
3395
3396 static int cma_connect_iw(struct rdma_id_private *id_priv,
3397 struct rdma_conn_param *conn_param)
3398 {
3399 struct iw_cm_id *cm_id;
3400 int ret;
3401 struct iw_cm_conn_param iw_param;
3402
3403 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3404 if (IS_ERR(cm_id))
3405 return PTR_ERR(cm_id);
3406
3407 cm_id->tos = id_priv->tos;
3408 id_priv->cm_id.iw = cm_id;
3409
3410 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3411 rdma_addr_size(cma_src_addr(id_priv)));
3412 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3413 rdma_addr_size(cma_dst_addr(id_priv)));
3414
3415 ret = cma_modify_qp_rtr(id_priv, conn_param);
3416 if (ret)
3417 goto out;
3418
3419 if (conn_param) {
3420 iw_param.ord = conn_param->initiator_depth;
3421 iw_param.ird = conn_param->responder_resources;
3422 iw_param.private_data = conn_param->private_data;
3423 iw_param.private_data_len = conn_param->private_data_len;
3424 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3425 } else {
3426 memset(&iw_param, 0, sizeof iw_param);
3427 iw_param.qpn = id_priv->qp_num;
3428 }
3429 ret = iw_cm_connect(cm_id, &iw_param);
3430 out:
3431 if (ret) {
3432 iw_destroy_cm_id(cm_id);
3433 id_priv->cm_id.iw = NULL;
3434 }
3435 return ret;
3436 }
3437
3438 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3439 {
3440 struct rdma_id_private *id_priv;
3441 int ret;
3442
3443 id_priv = container_of(id, struct rdma_id_private, id);
3444 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3445 return -EINVAL;
3446
3447 if (!id->qp) {
3448 id_priv->qp_num = conn_param->qp_num;
3449 id_priv->srq = conn_param->srq;
3450 }
3451
3452 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3453 if (id->qp_type == IB_QPT_UD)
3454 ret = cma_resolve_ib_udp(id_priv, conn_param);
3455 else
3456 ret = cma_connect_ib(id_priv, conn_param);
3457 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3458 ret = cma_connect_iw(id_priv, conn_param);
3459 else
3460 ret = -ENOSYS;
3461 if (ret)
3462 goto err;
3463
3464 return 0;
3465 err:
3466 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3467 return ret;
3468 }
3469 EXPORT_SYMBOL(rdma_connect);
3470
3471 static int cma_accept_ib(struct rdma_id_private *id_priv,
3472 struct rdma_conn_param *conn_param)
3473 {
3474 struct ib_cm_rep_param rep;
3475 int ret;
3476
3477 ret = cma_modify_qp_rtr(id_priv, conn_param);
3478 if (ret)
3479 goto out;
3480
3481 ret = cma_modify_qp_rts(id_priv, conn_param);
3482 if (ret)
3483 goto out;
3484
3485 memset(&rep, 0, sizeof rep);
3486 rep.qp_num = id_priv->qp_num;
3487 rep.starting_psn = id_priv->seq_num;
3488 rep.private_data = conn_param->private_data;
3489 rep.private_data_len = conn_param->private_data_len;
3490 rep.responder_resources = conn_param->responder_resources;
3491 rep.initiator_depth = conn_param->initiator_depth;
3492 rep.failover_accepted = 0;
3493 rep.flow_control = conn_param->flow_control;
3494 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3495 rep.srq = id_priv->srq ? 1 : 0;
3496
3497 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3498 out:
3499 return ret;
3500 }
3501
3502 static int cma_accept_iw(struct rdma_id_private *id_priv,
3503 struct rdma_conn_param *conn_param)
3504 {
3505 struct iw_cm_conn_param iw_param;
3506 int ret;
3507
3508 ret = cma_modify_qp_rtr(id_priv, conn_param);
3509 if (ret)
3510 return ret;
3511
3512 iw_param.ord = conn_param->initiator_depth;
3513 iw_param.ird = conn_param->responder_resources;
3514 iw_param.private_data = conn_param->private_data;
3515 iw_param.private_data_len = conn_param->private_data_len;
3516 if (id_priv->id.qp) {
3517 iw_param.qpn = id_priv->qp_num;
3518 } else
3519 iw_param.qpn = conn_param->qp_num;
3520
3521 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3522 }
3523
3524 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3525 enum ib_cm_sidr_status status, u32 qkey,
3526 const void *private_data, int private_data_len)
3527 {
3528 struct ib_cm_sidr_rep_param rep;
3529 int ret;
3530
3531 memset(&rep, 0, sizeof rep);
3532 rep.status = status;
3533 if (status == IB_SIDR_SUCCESS) {
3534 ret = cma_set_qkey(id_priv, qkey);
3535 if (ret)
3536 return ret;
3537 rep.qp_num = id_priv->qp_num;
3538 rep.qkey = id_priv->qkey;
3539 }
3540 rep.private_data = private_data;
3541 rep.private_data_len = private_data_len;
3542
3543 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3544 }
3545
3546 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3547 {
3548 struct rdma_id_private *id_priv;
3549 int ret;
3550
3551 id_priv = container_of(id, struct rdma_id_private, id);
3552
3553 id_priv->owner = task_pid_nr(current);
3554
3555 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3556 return -EINVAL;
3557
3558 if (!id->qp && conn_param) {
3559 id_priv->qp_num = conn_param->qp_num;
3560 id_priv->srq = conn_param->srq;
3561 }
3562
3563 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3564 if (id->qp_type == IB_QPT_UD) {
3565 if (conn_param)
3566 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3567 conn_param->qkey,
3568 conn_param->private_data,
3569 conn_param->private_data_len);
3570 else
3571 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3572 0, NULL, 0);
3573 } else {
3574 if (conn_param)
3575 ret = cma_accept_ib(id_priv, conn_param);
3576 else
3577 ret = cma_rep_recv(id_priv);
3578 }
3579 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3580 ret = cma_accept_iw(id_priv, conn_param);
3581 else
3582 ret = -ENOSYS;
3583
3584 if (ret)
3585 goto reject;
3586
3587 return 0;
3588 reject:
3589 cma_modify_qp_err(id_priv);
3590 rdma_reject(id, NULL, 0);
3591 return ret;
3592 }
3593 EXPORT_SYMBOL(rdma_accept);
3594
3595 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3596 {
3597 struct rdma_id_private *id_priv;
3598 int ret;
3599
3600 id_priv = container_of(id, struct rdma_id_private, id);
3601 if (!id_priv->cm_id.ib)
3602 return -EINVAL;
3603
3604 switch (id->device->node_type) {
3605 case RDMA_NODE_IB_CA:
3606 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3607 break;
3608 default:
3609 ret = 0;
3610 break;
3611 }
3612 return ret;
3613 }
3614 EXPORT_SYMBOL(rdma_notify);
3615
3616 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3617 u8 private_data_len)
3618 {
3619 struct rdma_id_private *id_priv;
3620 int ret;
3621
3622 id_priv = container_of(id, struct rdma_id_private, id);
3623 if (!id_priv->cm_id.ib)
3624 return -EINVAL;
3625
3626 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3627 if (id->qp_type == IB_QPT_UD)
3628 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3629 private_data, private_data_len);
3630 else
3631 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3632 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3633 0, private_data, private_data_len);
3634 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3635 ret = iw_cm_reject(id_priv->cm_id.iw,
3636 private_data, private_data_len);
3637 } else
3638 ret = -ENOSYS;
3639
3640 return ret;
3641 }
3642 EXPORT_SYMBOL(rdma_reject);
3643
3644 int rdma_disconnect(struct rdma_cm_id *id)
3645 {
3646 struct rdma_id_private *id_priv;
3647 int ret;
3648
3649 id_priv = container_of(id, struct rdma_id_private, id);
3650 if (!id_priv->cm_id.ib)
3651 return -EINVAL;
3652
3653 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3654 ret = cma_modify_qp_err(id_priv);
3655 if (ret)
3656 goto out;
3657 /* Initiate or respond to a disconnect. */
3658 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3659 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3660 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3661 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3662 } else
3663 ret = -EINVAL;
3664
3665 out:
3666 return ret;
3667 }
3668 EXPORT_SYMBOL(rdma_disconnect);
3669
3670 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3671 {
3672 struct rdma_id_private *id_priv;
3673 struct cma_multicast *mc = multicast->context;
3674 struct rdma_cm_event event;
3675 int ret;
3676
3677 id_priv = mc->id_priv;
3678 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3679 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3680 return 0;
3681
3682 if (!status)
3683 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3684 mutex_lock(&id_priv->qp_mutex);
3685 if (!status && id_priv->id.qp)
3686 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3687 be16_to_cpu(multicast->rec.mlid));
3688 mutex_unlock(&id_priv->qp_mutex);
3689
3690 memset(&event, 0, sizeof event);
3691 event.status = status;
3692 event.param.ud.private_data = mc->context;
3693 if (!status) {
3694 struct rdma_dev_addr *dev_addr =
3695 &id_priv->id.route.addr.dev_addr;
3696 struct net_device *ndev =
3697 dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3698 enum ib_gid_type gid_type =
3699 id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3700 rdma_start_port(id_priv->cma_dev->device)];
3701
3702 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3703 ib_init_ah_from_mcmember(id_priv->id.device,
3704 id_priv->id.port_num, &multicast->rec,
3705 ndev, gid_type,
3706 &event.param.ud.ah_attr);
3707 event.param.ud.qp_num = 0xFFFFFF;
3708 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3709 if (ndev)
3710 dev_put(ndev);
3711 } else
3712 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3713
3714 ret = id_priv->id.event_handler(&id_priv->id, &event);
3715 if (ret) {
3716 cma_exch(id_priv, RDMA_CM_DESTROYING);
3717 mutex_unlock(&id_priv->handler_mutex);
3718 rdma_destroy_id(&id_priv->id);
3719 return 0;
3720 }
3721
3722 mutex_unlock(&id_priv->handler_mutex);
3723 return 0;
3724 }
3725
3726 static void cma_set_mgid(struct rdma_id_private *id_priv,
3727 struct sockaddr *addr, union ib_gid *mgid)
3728 {
3729 unsigned char mc_map[MAX_ADDR_LEN];
3730 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3731 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3732 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3733
3734 if (cma_any_addr(addr)) {
3735 memset(mgid, 0, sizeof *mgid);
3736 } else if ((addr->sa_family == AF_INET6) &&
3737 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3738 0xFF10A01B)) {
3739 /* IPv6 address is an SA assigned MGID. */
3740 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3741 } else if (addr->sa_family == AF_IB) {
3742 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3743 } else if ((addr->sa_family == AF_INET6)) {
3744 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3745 if (id_priv->id.ps == RDMA_PS_UDP)
3746 mc_map[7] = 0x01; /* Use RDMA CM signature */
3747 *mgid = *(union ib_gid *) (mc_map + 4);
3748 } else {
3749 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3750 if (id_priv->id.ps == RDMA_PS_UDP)
3751 mc_map[7] = 0x01; /* Use RDMA CM signature */
3752 *mgid = *(union ib_gid *) (mc_map + 4);
3753 }
3754 }
3755
3756 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3757 struct cma_multicast *mc)
3758 {
3759 struct ib_sa_mcmember_rec rec;
3760 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3761 ib_sa_comp_mask comp_mask;
3762 int ret;
3763
3764 ib_addr_get_mgid(dev_addr, &rec.mgid);
3765 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3766 &rec.mgid, &rec);
3767 if (ret)
3768 return ret;
3769
3770 ret = cma_set_qkey(id_priv, 0);
3771 if (ret)
3772 return ret;
3773
3774 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3775 rec.qkey = cpu_to_be32(id_priv->qkey);
3776 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3777 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3778 rec.join_state = 1;
3779
3780 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3781 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3782 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3783 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3784 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3785
3786 if (id_priv->id.ps == RDMA_PS_IPOIB)
3787 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3788 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3789 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3790 IB_SA_MCMEMBER_REC_MTU |
3791 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3792
3793 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3794 id_priv->id.port_num, &rec,
3795 comp_mask, GFP_KERNEL,
3796 cma_ib_mc_handler, mc);
3797 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3798 }
3799
3800 static void iboe_mcast_work_handler(struct work_struct *work)
3801 {
3802 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3803 struct cma_multicast *mc = mw->mc;
3804 struct ib_sa_multicast *m = mc->multicast.ib;
3805
3806 mc->multicast.ib->context = mc;
3807 cma_ib_mc_handler(0, m);
3808 kref_put(&mc->mcref, release_mc);
3809 kfree(mw);
3810 }
3811
3812 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3813 {
3814 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3815 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3816
3817 if (cma_any_addr(addr)) {
3818 memset(mgid, 0, sizeof *mgid);
3819 } else if (addr->sa_family == AF_INET6) {
3820 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3821 } else {
3822 mgid->raw[0] = 0xff;
3823 mgid->raw[1] = 0x0e;
3824 mgid->raw[2] = 0;
3825 mgid->raw[3] = 0;
3826 mgid->raw[4] = 0;
3827 mgid->raw[5] = 0;
3828 mgid->raw[6] = 0;
3829 mgid->raw[7] = 0;
3830 mgid->raw[8] = 0;
3831 mgid->raw[9] = 0;
3832 mgid->raw[10] = 0xff;
3833 mgid->raw[11] = 0xff;
3834 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3835 }
3836 }
3837
3838 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3839 struct cma_multicast *mc)
3840 {
3841 struct iboe_mcast_work *work;
3842 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3843 int err = 0;
3844 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3845 struct net_device *ndev = NULL;
3846 enum ib_gid_type gid_type;
3847
3848 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3849 return -EINVAL;
3850
3851 work = kzalloc(sizeof *work, GFP_KERNEL);
3852 if (!work)
3853 return -ENOMEM;
3854
3855 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3856 if (!mc->multicast.ib) {
3857 err = -ENOMEM;
3858 goto out1;
3859 }
3860
3861 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3862
3863 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3864 if (id_priv->id.ps == RDMA_PS_UDP)
3865 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3866
3867 if (dev_addr->bound_dev_if)
3868 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3869 if (!ndev) {
3870 err = -ENODEV;
3871 goto out2;
3872 }
3873 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3874 mc->multicast.ib->rec.hop_limit = 1;
3875 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3876
3877 gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3878 rdma_start_port(id_priv->cma_dev->device)];
3879 if (addr->sa_family == AF_INET) {
3880 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
3881 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
3882 true);
3883 if (!err) {
3884 mc->igmp_joined = true;
3885 mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
3886 }
3887 } else {
3888 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
3889 err = -ENOTSUPP;
3890 }
3891 dev_put(ndev);
3892 if (err || !mc->multicast.ib->rec.mtu) {
3893 if (!err)
3894 err = -EINVAL;
3895 goto out2;
3896 }
3897 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3898 &mc->multicast.ib->rec.port_gid);
3899 work->id = id_priv;
3900 work->mc = mc;
3901 INIT_WORK(&work->work, iboe_mcast_work_handler);
3902 kref_get(&mc->mcref);
3903 queue_work(cma_wq, &work->work);
3904
3905 return 0;
3906
3907 out2:
3908 kfree(mc->multicast.ib);
3909 out1:
3910 kfree(work);
3911 return err;
3912 }
3913
3914 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3915 void *context)
3916 {
3917 struct rdma_id_private *id_priv;
3918 struct cma_multicast *mc;
3919 int ret;
3920
3921 id_priv = container_of(id, struct rdma_id_private, id);
3922 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3923 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3924 return -EINVAL;
3925
3926 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3927 if (!mc)
3928 return -ENOMEM;
3929
3930 memcpy(&mc->addr, addr, rdma_addr_size(addr));
3931 mc->context = context;
3932 mc->id_priv = id_priv;
3933 mc->igmp_joined = false;
3934 spin_lock(&id_priv->lock);
3935 list_add(&mc->list, &id_priv->mc_list);
3936 spin_unlock(&id_priv->lock);
3937
3938 if (rdma_protocol_roce(id->device, id->port_num)) {
3939 kref_init(&mc->mcref);
3940 ret = cma_iboe_join_multicast(id_priv, mc);
3941 } else if (rdma_cap_ib_mcast(id->device, id->port_num))
3942 ret = cma_join_ib_multicast(id_priv, mc);
3943 else
3944 ret = -ENOSYS;
3945
3946 if (ret) {
3947 spin_lock_irq(&id_priv->lock);
3948 list_del(&mc->list);
3949 spin_unlock_irq(&id_priv->lock);
3950 kfree(mc);
3951 }
3952 return ret;
3953 }
3954 EXPORT_SYMBOL(rdma_join_multicast);
3955
3956 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3957 {
3958 struct rdma_id_private *id_priv;
3959 struct cma_multicast *mc;
3960
3961 id_priv = container_of(id, struct rdma_id_private, id);
3962 spin_lock_irq(&id_priv->lock);
3963 list_for_each_entry(mc, &id_priv->mc_list, list) {
3964 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3965 list_del(&mc->list);
3966 spin_unlock_irq(&id_priv->lock);
3967
3968 if (id->qp)
3969 ib_detach_mcast(id->qp,
3970 &mc->multicast.ib->rec.mgid,
3971 be16_to_cpu(mc->multicast.ib->rec.mlid));
3972
3973 BUG_ON(id_priv->cma_dev->device != id->device);
3974
3975 if (rdma_cap_ib_mcast(id->device, id->port_num)) {
3976 ib_sa_free_multicast(mc->multicast.ib);
3977 kfree(mc);
3978 } else if (rdma_protocol_roce(id->device, id->port_num)) {
3979 if (mc->igmp_joined) {
3980 struct rdma_dev_addr *dev_addr =
3981 &id->route.addr.dev_addr;
3982 struct net_device *ndev = NULL;
3983
3984 if (dev_addr->bound_dev_if)
3985 ndev = dev_get_by_index(&init_net,
3986 dev_addr->bound_dev_if);
3987 if (ndev) {
3988 cma_igmp_send(ndev,
3989 &mc->multicast.ib->rec.mgid,
3990 false);
3991 dev_put(ndev);
3992 }
3993 mc->igmp_joined = false;
3994 }
3995 kref_put(&mc->mcref, release_mc);
3996 }
3997 return;
3998 }
3999 }
4000 spin_unlock_irq(&id_priv->lock);
4001 }
4002 EXPORT_SYMBOL(rdma_leave_multicast);
4003
4004 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4005 {
4006 struct rdma_dev_addr *dev_addr;
4007 struct cma_ndev_work *work;
4008
4009 dev_addr = &id_priv->id.route.addr.dev_addr;
4010
4011 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4012 (net_eq(dev_net(ndev), dev_addr->net)) &&
4013 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4014 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4015 ndev->name, &id_priv->id);
4016 work = kzalloc(sizeof *work, GFP_KERNEL);
4017 if (!work)
4018 return -ENOMEM;
4019
4020 INIT_WORK(&work->work, cma_ndev_work_handler);
4021 work->id = id_priv;
4022 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4023 atomic_inc(&id_priv->refcount);
4024 queue_work(cma_wq, &work->work);
4025 }
4026
4027 return 0;
4028 }
4029
4030 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4031 void *ptr)
4032 {
4033 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4034 struct cma_device *cma_dev;
4035 struct rdma_id_private *id_priv;
4036 int ret = NOTIFY_DONE;
4037
4038 if (event != NETDEV_BONDING_FAILOVER)
4039 return NOTIFY_DONE;
4040
4041 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4042 return NOTIFY_DONE;
4043
4044 mutex_lock(&lock);
4045 list_for_each_entry(cma_dev, &dev_list, list)
4046 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4047 ret = cma_netdev_change(ndev, id_priv);
4048 if (ret)
4049 goto out;
4050 }
4051
4052 out:
4053 mutex_unlock(&lock);
4054 return ret;
4055 }
4056
4057 static struct notifier_block cma_nb = {
4058 .notifier_call = cma_netdev_callback
4059 };
4060
4061 static void cma_add_one(struct ib_device *device)
4062 {
4063 struct cma_device *cma_dev;
4064 struct rdma_id_private *id_priv;
4065 unsigned int i;
4066 unsigned long supported_gids = 0;
4067
4068 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4069 if (!cma_dev)
4070 return;
4071
4072 cma_dev->device = device;
4073 cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4074 sizeof(*cma_dev->default_gid_type),
4075 GFP_KERNEL);
4076 if (!cma_dev->default_gid_type) {
4077 kfree(cma_dev);
4078 return;
4079 }
4080 for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4081 supported_gids = roce_gid_type_mask_support(device, i);
4082 WARN_ON(!supported_gids);
4083 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4084 find_first_bit(&supported_gids, BITS_PER_LONG);
4085 }
4086
4087 init_completion(&cma_dev->comp);
4088 atomic_set(&cma_dev->refcount, 1);
4089 INIT_LIST_HEAD(&cma_dev->id_list);
4090 ib_set_client_data(device, &cma_client, cma_dev);
4091
4092 mutex_lock(&lock);
4093 list_add_tail(&cma_dev->list, &dev_list);
4094 list_for_each_entry(id_priv, &listen_any_list, list)
4095 cma_listen_on_dev(id_priv, cma_dev);
4096 mutex_unlock(&lock);
4097 }
4098
4099 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4100 {
4101 struct rdma_cm_event event;
4102 enum rdma_cm_state state;
4103 int ret = 0;
4104
4105 /* Record that we want to remove the device */
4106 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4107 if (state == RDMA_CM_DESTROYING)
4108 return 0;
4109
4110 cma_cancel_operation(id_priv, state);
4111 mutex_lock(&id_priv->handler_mutex);
4112
4113 /* Check for destruction from another callback. */
4114 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4115 goto out;
4116
4117 memset(&event, 0, sizeof event);
4118 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4119 ret = id_priv->id.event_handler(&id_priv->id, &event);
4120 out:
4121 mutex_unlock(&id_priv->handler_mutex);
4122 return ret;
4123 }
4124
4125 static void cma_process_remove(struct cma_device *cma_dev)
4126 {
4127 struct rdma_id_private *id_priv;
4128 int ret;
4129
4130 mutex_lock(&lock);
4131 while (!list_empty(&cma_dev->id_list)) {
4132 id_priv = list_entry(cma_dev->id_list.next,
4133 struct rdma_id_private, list);
4134
4135 list_del(&id_priv->listen_list);
4136 list_del_init(&id_priv->list);
4137 atomic_inc(&id_priv->refcount);
4138 mutex_unlock(&lock);
4139
4140 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4141 cma_deref_id(id_priv);
4142 if (ret)
4143 rdma_destroy_id(&id_priv->id);
4144
4145 mutex_lock(&lock);
4146 }
4147 mutex_unlock(&lock);
4148
4149 cma_deref_dev(cma_dev);
4150 wait_for_completion(&cma_dev->comp);
4151 }
4152
4153 static void cma_remove_one(struct ib_device *device, void *client_data)
4154 {
4155 struct cma_device *cma_dev = client_data;
4156
4157 if (!cma_dev)
4158 return;
4159
4160 mutex_lock(&lock);
4161 list_del(&cma_dev->list);
4162 mutex_unlock(&lock);
4163
4164 cma_process_remove(cma_dev);
4165 kfree(cma_dev->default_gid_type);
4166 kfree(cma_dev);
4167 }
4168
4169 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4170 {
4171 struct nlmsghdr *nlh;
4172 struct rdma_cm_id_stats *id_stats;
4173 struct rdma_id_private *id_priv;
4174 struct rdma_cm_id *id = NULL;
4175 struct cma_device *cma_dev;
4176 int i_dev = 0, i_id = 0;
4177
4178 /*
4179 * We export all of the IDs as a sequence of messages. Each
4180 * ID gets its own netlink message.
4181 */
4182 mutex_lock(&lock);
4183
4184 list_for_each_entry(cma_dev, &dev_list, list) {
4185 if (i_dev < cb->args[0]) {
4186 i_dev++;
4187 continue;
4188 }
4189
4190 i_id = 0;
4191 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4192 if (i_id < cb->args[1]) {
4193 i_id++;
4194 continue;
4195 }
4196
4197 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4198 sizeof *id_stats, RDMA_NL_RDMA_CM,
4199 RDMA_NL_RDMA_CM_ID_STATS,
4200 NLM_F_MULTI);
4201 if (!id_stats)
4202 goto out;
4203
4204 memset(id_stats, 0, sizeof *id_stats);
4205 id = &id_priv->id;
4206 id_stats->node_type = id->route.addr.dev_addr.dev_type;
4207 id_stats->port_num = id->port_num;
4208 id_stats->bound_dev_if =
4209 id->route.addr.dev_addr.bound_dev_if;
4210
4211 if (ibnl_put_attr(skb, nlh,
4212 rdma_addr_size(cma_src_addr(id_priv)),
4213 cma_src_addr(id_priv),
4214 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4215 goto out;
4216 if (ibnl_put_attr(skb, nlh,
4217 rdma_addr_size(cma_src_addr(id_priv)),
4218 cma_dst_addr(id_priv),
4219 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4220 goto out;
4221
4222 id_stats->pid = id_priv->owner;
4223 id_stats->port_space = id->ps;
4224 id_stats->cm_state = id_priv->state;
4225 id_stats->qp_num = id_priv->qp_num;
4226 id_stats->qp_type = id->qp_type;
4227
4228 i_id++;
4229 }
4230
4231 cb->args[1] = 0;
4232 i_dev++;
4233 }
4234
4235 out:
4236 mutex_unlock(&lock);
4237 cb->args[0] = i_dev;
4238 cb->args[1] = i_id;
4239
4240 return skb->len;
4241 }
4242
4243 static const struct ibnl_client_cbs cma_cb_table[] = {
4244 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4245 .module = THIS_MODULE },
4246 };
4247
4248 static int cma_init_net(struct net *net)
4249 {
4250 struct cma_pernet *pernet = cma_pernet(net);
4251
4252 idr_init(&pernet->tcp_ps);
4253 idr_init(&pernet->udp_ps);
4254 idr_init(&pernet->ipoib_ps);
4255 idr_init(&pernet->ib_ps);
4256
4257 return 0;
4258 }
4259
4260 static void cma_exit_net(struct net *net)
4261 {
4262 struct cma_pernet *pernet = cma_pernet(net);
4263
4264 idr_destroy(&pernet->tcp_ps);
4265 idr_destroy(&pernet->udp_ps);
4266 idr_destroy(&pernet->ipoib_ps);
4267 idr_destroy(&pernet->ib_ps);
4268 }
4269
4270 static struct pernet_operations cma_pernet_operations = {
4271 .init = cma_init_net,
4272 .exit = cma_exit_net,
4273 .id = &cma_pernet_id,
4274 .size = sizeof(struct cma_pernet),
4275 };
4276
4277 static int __init cma_init(void)
4278 {
4279 int ret;
4280
4281 cma_wq = create_singlethread_workqueue("rdma_cm");
4282 if (!cma_wq)
4283 return -ENOMEM;
4284
4285 ret = register_pernet_subsys(&cma_pernet_operations);
4286 if (ret)
4287 goto err_wq;
4288
4289 ib_sa_register_client(&sa_client);
4290 rdma_addr_register_client(&addr_client);
4291 register_netdevice_notifier(&cma_nb);
4292
4293 ret = ib_register_client(&cma_client);
4294 if (ret)
4295 goto err;
4296
4297 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
4298 pr_warn("RDMA CMA: failed to add netlink callback\n");
4299 cma_configfs_init();
4300
4301 return 0;
4302
4303 err:
4304 unregister_netdevice_notifier(&cma_nb);
4305 rdma_addr_unregister_client(&addr_client);
4306 ib_sa_unregister_client(&sa_client);
4307 err_wq:
4308 destroy_workqueue(cma_wq);
4309 return ret;
4310 }
4311
4312 static void __exit cma_cleanup(void)
4313 {
4314 cma_configfs_exit();
4315 ibnl_remove_client(RDMA_NL_RDMA_CM);
4316 ib_unregister_client(&cma_client);
4317 unregister_netdevice_notifier(&cma_nb);
4318 rdma_addr_unregister_client(&addr_client);
4319 ib_sa_unregister_client(&sa_client);
4320 unregister_pernet_subsys(&cma_pernet_operations);
4321 destroy_workqueue(cma_wq);
4322 }
4323
4324 module_init(cma_init);
4325 module_exit(cma_cleanup);
Linux kernel - Deliverables
This page took 0.195447 seconds and 5 git commands to generate.