]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/infiniband/core/cma.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
IB/SA: Modify SA to implicitly cache Class Port info
[mirror_ubuntu-artful-kernel.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_QUERY_CLASSPORT_INFO_TIMEOUT 3000
72 #define CMA_MAX_CM_RETRIES 15
73 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
74 #define CMA_IBOE_PACKET_LIFETIME 18
75
76 static const char * const cma_events[] = {
77 [RDMA_CM_EVENT_ADDR_RESOLVED] = "address resolved",
78 [RDMA_CM_EVENT_ADDR_ERROR] = "address error",
79 [RDMA_CM_EVENT_ROUTE_RESOLVED] = "route resolved ",
80 [RDMA_CM_EVENT_ROUTE_ERROR] = "route error",
81 [RDMA_CM_EVENT_CONNECT_REQUEST] = "connect request",
82 [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
83 [RDMA_CM_EVENT_CONNECT_ERROR] = "connect error",
84 [RDMA_CM_EVENT_UNREACHABLE] = "unreachable",
85 [RDMA_CM_EVENT_REJECTED] = "rejected",
86 [RDMA_CM_EVENT_ESTABLISHED] = "established",
87 [RDMA_CM_EVENT_DISCONNECTED] = "disconnected",
88 [RDMA_CM_EVENT_DEVICE_REMOVAL] = "device removal",
89 [RDMA_CM_EVENT_MULTICAST_JOIN] = "multicast join",
90 [RDMA_CM_EVENT_MULTICAST_ERROR] = "multicast error",
91 [RDMA_CM_EVENT_ADDR_CHANGE] = "address change",
92 [RDMA_CM_EVENT_TIMEWAIT_EXIT] = "timewait exit",
93 };
94
95 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
96 {
97 size_t index = event;
98
99 return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
100 cma_events[index] : "unrecognized event";
101 }
102 EXPORT_SYMBOL(rdma_event_msg);
103
104 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
105 int reason)
106 {
107 if (rdma_ib_or_roce(id->device, id->port_num))
108 return ibcm_reject_msg(reason);
109
110 if (rdma_protocol_iwarp(id->device, id->port_num))
111 return iwcm_reject_msg(reason);
112
113 WARN_ON_ONCE(1);
114 return "unrecognized transport";
115 }
116 EXPORT_SYMBOL(rdma_reject_msg);
117
118 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
119 {
120 if (rdma_ib_or_roce(id->device, id->port_num))
121 return reason == IB_CM_REJ_CONSUMER_DEFINED;
122
123 if (rdma_protocol_iwarp(id->device, id->port_num))
124 return reason == -ECONNREFUSED;
125
126 WARN_ON_ONCE(1);
127 return false;
128 }
129 EXPORT_SYMBOL(rdma_is_consumer_reject);
130
131 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
132 struct rdma_cm_event *ev, u8 *data_len)
133 {
134 const void *p;
135
136 if (rdma_is_consumer_reject(id, ev->status)) {
137 *data_len = ev->param.conn.private_data_len;
138 p = ev->param.conn.private_data;
139 } else {
140 *data_len = 0;
141 p = NULL;
142 }
143 return p;
144 }
145 EXPORT_SYMBOL(rdma_consumer_reject_data);
146
147 static void cma_add_one(struct ib_device *device);
148 static void cma_remove_one(struct ib_device *device, void *client_data);
149
150 static struct ib_client cma_client = {
151 .name = "cma",
152 .add = cma_add_one,
153 .remove = cma_remove_one
154 };
155
156 static struct ib_sa_client sa_client;
157 static struct rdma_addr_client addr_client;
158 static LIST_HEAD(dev_list);
159 static LIST_HEAD(listen_any_list);
160 static DEFINE_MUTEX(lock);
161 static struct workqueue_struct *cma_wq;
162 static unsigned int cma_pernet_id;
163
164 struct cma_pernet {
165 struct idr tcp_ps;
166 struct idr udp_ps;
167 struct idr ipoib_ps;
168 struct idr ib_ps;
169 };
170
171 static struct cma_pernet *cma_pernet(struct net *net)
172 {
173 return net_generic(net, cma_pernet_id);
174 }
175
176 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
177 {
178 struct cma_pernet *pernet = cma_pernet(net);
179
180 switch (ps) {
181 case RDMA_PS_TCP:
182 return &pernet->tcp_ps;
183 case RDMA_PS_UDP:
184 return &pernet->udp_ps;
185 case RDMA_PS_IPOIB:
186 return &pernet->ipoib_ps;
187 case RDMA_PS_IB:
188 return &pernet->ib_ps;
189 default:
190 return NULL;
191 }
192 }
193
194 struct cma_device {
195 struct list_head list;
196 struct ib_device *device;
197 struct completion comp;
198 atomic_t refcount;
199 struct list_head id_list;
200 enum ib_gid_type *default_gid_type;
201 u8 *default_roce_tos;
202 };
203
204 struct rdma_bind_list {
205 enum rdma_port_space ps;
206 struct hlist_head owners;
207 unsigned short port;
208 };
209
210 struct class_port_info_context {
211 struct ib_class_port_info *class_port_info;
212 struct ib_device *device;
213 struct completion done;
214 struct ib_sa_query *sa_query;
215 u8 port_num;
216 };
217
218 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
219 struct rdma_bind_list *bind_list, int snum)
220 {
221 struct idr *idr = cma_pernet_idr(net, ps);
222
223 return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
224 }
225
226 static struct rdma_bind_list *cma_ps_find(struct net *net,
227 enum rdma_port_space ps, int snum)
228 {
229 struct idr *idr = cma_pernet_idr(net, ps);
230
231 return idr_find(idr, snum);
232 }
233
234 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
235 {
236 struct idr *idr = cma_pernet_idr(net, ps);
237
238 idr_remove(idr, snum);
239 }
240
241 enum {
242 CMA_OPTION_AFONLY,
243 };
244
245 void cma_ref_dev(struct cma_device *cma_dev)
246 {
247 atomic_inc(&cma_dev->refcount);
248 }
249
250 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter filter,
251 void *cookie)
252 {
253 struct cma_device *cma_dev;
254 struct cma_device *found_cma_dev = NULL;
255
256 mutex_lock(&lock);
257
258 list_for_each_entry(cma_dev, &dev_list, list)
259 if (filter(cma_dev->device, cookie)) {
260 found_cma_dev = cma_dev;
261 break;
262 }
263
264 if (found_cma_dev)
265 cma_ref_dev(found_cma_dev);
266 mutex_unlock(&lock);
267 return found_cma_dev;
268 }
269
270 int cma_get_default_gid_type(struct cma_device *cma_dev,
271 unsigned int port)
272 {
273 if (!rdma_is_port_valid(cma_dev->device, port))
274 return -EINVAL;
275
276 return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
277 }
278
279 int cma_set_default_gid_type(struct cma_device *cma_dev,
280 unsigned int port,
281 enum ib_gid_type default_gid_type)
282 {
283 unsigned long supported_gids;
284
285 if (!rdma_is_port_valid(cma_dev->device, port))
286 return -EINVAL;
287
288 supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
289
290 if (!(supported_gids & 1 << default_gid_type))
291 return -EINVAL;
292
293 cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
294 default_gid_type;
295
296 return 0;
297 }
298
299 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
300 {
301 if (!rdma_is_port_valid(cma_dev->device, port))
302 return -EINVAL;
303
304 return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
305 }
306
307 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
308 u8 default_roce_tos)
309 {
310 if (!rdma_is_port_valid(cma_dev->device, port))
311 return -EINVAL;
312
313 cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
314 default_roce_tos;
315
316 return 0;
317 }
318 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
319 {
320 return cma_dev->device;
321 }
322
323 /*
324 * Device removal can occur at anytime, so we need extra handling to
325 * serialize notifying the user of device removal with other callbacks.
326 * We do this by disabling removal notification while a callback is in process,
327 * and reporting it after the callback completes.
328 */
329 struct rdma_id_private {
330 struct rdma_cm_id id;
331
332 struct rdma_bind_list *bind_list;
333 struct hlist_node node;
334 struct list_head list; /* listen_any_list or cma_device.list */
335 struct list_head listen_list; /* per device listens */
336 struct cma_device *cma_dev;
337 struct list_head mc_list;
338
339 int internal_id;
340 enum rdma_cm_state state;
341 spinlock_t lock;
342 struct mutex qp_mutex;
343
344 struct completion comp;
345 atomic_t refcount;
346 struct mutex handler_mutex;
347
348 int backlog;
349 int timeout_ms;
350 struct ib_sa_query *query;
351 int query_id;
352 union {
353 struct ib_cm_id *ib;
354 struct iw_cm_id *iw;
355 } cm_id;
356
357 u32 seq_num;
358 u32 qkey;
359 u32 qp_num;
360 pid_t owner;
361 u32 options;
362 u8 srq;
363 u8 tos;
364 bool tos_set;
365 u8 reuseaddr;
366 u8 afonly;
367 enum ib_gid_type gid_type;
368 };
369
370 struct cma_multicast {
371 struct rdma_id_private *id_priv;
372 union {
373 struct ib_sa_multicast *ib;
374 } multicast;
375 struct list_head list;
376 void *context;
377 struct sockaddr_storage addr;
378 struct kref mcref;
379 bool igmp_joined;
380 u8 join_state;
381 };
382
383 struct cma_work {
384 struct work_struct work;
385 struct rdma_id_private *id;
386 enum rdma_cm_state old_state;
387 enum rdma_cm_state new_state;
388 struct rdma_cm_event event;
389 };
390
391 struct cma_ndev_work {
392 struct work_struct work;
393 struct rdma_id_private *id;
394 struct rdma_cm_event event;
395 };
396
397 struct iboe_mcast_work {
398 struct work_struct work;
399 struct rdma_id_private *id;
400 struct cma_multicast *mc;
401 };
402
403 union cma_ip_addr {
404 struct in6_addr ip6;
405 struct {
406 __be32 pad[3];
407 __be32 addr;
408 } ip4;
409 };
410
411 struct cma_hdr {
412 u8 cma_version;
413 u8 ip_version; /* IP version: 7:4 */
414 __be16 port;
415 union cma_ip_addr src_addr;
416 union cma_ip_addr dst_addr;
417 };
418
419 #define CMA_VERSION 0x00
420
421 struct cma_req_info {
422 struct ib_device *device;
423 int port;
424 union ib_gid local_gid;
425 __be64 service_id;
426 u16 pkey;
427 bool has_gid:1;
428 };
429
430 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
431 {
432 unsigned long flags;
433 int ret;
434
435 spin_lock_irqsave(&id_priv->lock, flags);
436 ret = (id_priv->state == comp);
437 spin_unlock_irqrestore(&id_priv->lock, flags);
438 return ret;
439 }
440
441 static int cma_comp_exch(struct rdma_id_private *id_priv,
442 enum rdma_cm_state comp, enum rdma_cm_state exch)
443 {
444 unsigned long flags;
445 int ret;
446
447 spin_lock_irqsave(&id_priv->lock, flags);
448 if ((ret = (id_priv->state == comp)))
449 id_priv->state = exch;
450 spin_unlock_irqrestore(&id_priv->lock, flags);
451 return ret;
452 }
453
454 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
455 enum rdma_cm_state exch)
456 {
457 unsigned long flags;
458 enum rdma_cm_state old;
459
460 spin_lock_irqsave(&id_priv->lock, flags);
461 old = id_priv->state;
462 id_priv->state = exch;
463 spin_unlock_irqrestore(&id_priv->lock, flags);
464 return old;
465 }
466
467 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
468 {
469 return hdr->ip_version >> 4;
470 }
471
472 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
473 {
474 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
475 }
476
477 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
478 {
479 struct in_device *in_dev = NULL;
480
481 if (ndev) {
482 rtnl_lock();
483 in_dev = __in_dev_get_rtnl(ndev);
484 if (in_dev) {
485 if (join)
486 ip_mc_inc_group(in_dev,
487 *(__be32 *)(mgid->raw + 12));
488 else
489 ip_mc_dec_group(in_dev,
490 *(__be32 *)(mgid->raw + 12));
491 }
492 rtnl_unlock();
493 }
494 return (in_dev) ? 0 : -ENODEV;
495 }
496
497 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
498 struct cma_device *cma_dev)
499 {
500 cma_ref_dev(cma_dev);
501 id_priv->cma_dev = cma_dev;
502 id_priv->gid_type = 0;
503 id_priv->id.device = cma_dev->device;
504 id_priv->id.route.addr.dev_addr.transport =
505 rdma_node_get_transport(cma_dev->device->node_type);
506 list_add_tail(&id_priv->list, &cma_dev->id_list);
507 }
508
509 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
510 struct cma_device *cma_dev)
511 {
512 _cma_attach_to_dev(id_priv, cma_dev);
513 id_priv->gid_type =
514 cma_dev->default_gid_type[id_priv->id.port_num -
515 rdma_start_port(cma_dev->device)];
516 }
517
518 void cma_deref_dev(struct cma_device *cma_dev)
519 {
520 if (atomic_dec_and_test(&cma_dev->refcount))
521 complete(&cma_dev->comp);
522 }
523
524 static inline void release_mc(struct kref *kref)
525 {
526 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
527
528 kfree(mc->multicast.ib);
529 kfree(mc);
530 }
531
532 static void cma_release_dev(struct rdma_id_private *id_priv)
533 {
534 mutex_lock(&lock);
535 list_del(&id_priv->list);
536 cma_deref_dev(id_priv->cma_dev);
537 id_priv->cma_dev = NULL;
538 mutex_unlock(&lock);
539 }
540
541 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
542 {
543 return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
544 }
545
546 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
547 {
548 return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
549 }
550
551 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
552 {
553 return id_priv->id.route.addr.src_addr.ss_family;
554 }
555
556 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
557 {
558 struct ib_sa_mcmember_rec rec;
559 int ret = 0;
560
561 if (id_priv->qkey) {
562 if (qkey && id_priv->qkey != qkey)
563 return -EINVAL;
564 return 0;
565 }
566
567 if (qkey) {
568 id_priv->qkey = qkey;
569 return 0;
570 }
571
572 switch (id_priv->id.ps) {
573 case RDMA_PS_UDP:
574 case RDMA_PS_IB:
575 id_priv->qkey = RDMA_UDP_QKEY;
576 break;
577 case RDMA_PS_IPOIB:
578 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
579 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
580 id_priv->id.port_num, &rec.mgid,
581 &rec);
582 if (!ret)
583 id_priv->qkey = be32_to_cpu(rec.qkey);
584 break;
585 default:
586 break;
587 }
588 return ret;
589 }
590
591 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
592 {
593 dev_addr->dev_type = ARPHRD_INFINIBAND;
594 rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
595 ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
596 }
597
598 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
599 {
600 int ret;
601
602 if (addr->sa_family != AF_IB) {
603 ret = rdma_translate_ip(addr, dev_addr, NULL);
604 } else {
605 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
606 ret = 0;
607 }
608
609 return ret;
610 }
611
612 static inline int cma_validate_port(struct ib_device *device, u8 port,
613 enum ib_gid_type gid_type,
614 union ib_gid *gid, int dev_type,
615 int bound_if_index)
616 {
617 int ret = -ENODEV;
618 struct net_device *ndev = NULL;
619
620 if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
621 return ret;
622
623 if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
624 return ret;
625
626 if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
627 ndev = dev_get_by_index(&init_net, bound_if_index);
628 if (ndev && ndev->flags & IFF_LOOPBACK) {
629 pr_info("detected loopback device\n");
630 dev_put(ndev);
631
632 if (!device->get_netdev)
633 return -EOPNOTSUPP;
634
635 ndev = device->get_netdev(device, port);
636 if (!ndev)
637 return -ENODEV;
638 }
639 } else {
640 gid_type = IB_GID_TYPE_IB;
641 }
642
643 ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
644 ndev, NULL);
645
646 if (ndev)
647 dev_put(ndev);
648
649 return ret;
650 }
651
652 static int cma_acquire_dev(struct rdma_id_private *id_priv,
653 struct rdma_id_private *listen_id_priv)
654 {
655 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
656 struct cma_device *cma_dev;
657 union ib_gid gid, iboe_gid, *gidp;
658 int ret = -ENODEV;
659 u8 port;
660
661 if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
662 id_priv->id.ps == RDMA_PS_IPOIB)
663 return -EINVAL;
664
665 mutex_lock(&lock);
666 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
667 &iboe_gid);
668
669 memcpy(&gid, dev_addr->src_dev_addr +
670 rdma_addr_gid_offset(dev_addr), sizeof gid);
671
672 if (listen_id_priv) {
673 cma_dev = listen_id_priv->cma_dev;
674 port = listen_id_priv->id.port_num;
675 gidp = rdma_protocol_roce(cma_dev->device, port) ?
676 &iboe_gid : &gid;
677
678 ret = cma_validate_port(cma_dev->device, port,
679 rdma_protocol_ib(cma_dev->device, port) ?
680 IB_GID_TYPE_IB :
681 listen_id_priv->gid_type, gidp,
682 dev_addr->dev_type,
683 dev_addr->bound_dev_if);
684 if (!ret) {
685 id_priv->id.port_num = port;
686 goto out;
687 }
688 }
689
690 list_for_each_entry(cma_dev, &dev_list, list) {
691 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
692 if (listen_id_priv &&
693 listen_id_priv->cma_dev == cma_dev &&
694 listen_id_priv->id.port_num == port)
695 continue;
696
697 gidp = rdma_protocol_roce(cma_dev->device, port) ?
698 &iboe_gid : &gid;
699
700 ret = cma_validate_port(cma_dev->device, port,
701 rdma_protocol_ib(cma_dev->device, port) ?
702 IB_GID_TYPE_IB :
703 cma_dev->default_gid_type[port - 1],
704 gidp, dev_addr->dev_type,
705 dev_addr->bound_dev_if);
706 if (!ret) {
707 id_priv->id.port_num = port;
708 goto out;
709 }
710 }
711 }
712
713 out:
714 if (!ret)
715 cma_attach_to_dev(id_priv, cma_dev);
716
717 mutex_unlock(&lock);
718 return ret;
719 }
720
721 /*
722 * Select the source IB device and address to reach the destination IB address.
723 */
724 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
725 {
726 struct cma_device *cma_dev, *cur_dev;
727 struct sockaddr_ib *addr;
728 union ib_gid gid, sgid, *dgid;
729 u16 pkey, index;
730 u8 p;
731 enum ib_port_state port_state;
732 int i;
733
734 cma_dev = NULL;
735 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
736 dgid = (union ib_gid *) &addr->sib_addr;
737 pkey = ntohs(addr->sib_pkey);
738
739 list_for_each_entry(cur_dev, &dev_list, list) {
740 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
741 if (!rdma_cap_af_ib(cur_dev->device, p))
742 continue;
743
744 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
745 continue;
746
747 if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
748 continue;
749 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
750 &gid, NULL);
751 i++) {
752 if (!memcmp(&gid, dgid, sizeof(gid))) {
753 cma_dev = cur_dev;
754 sgid = gid;
755 id_priv->id.port_num = p;
756 goto found;
757 }
758
759 if (!cma_dev && (gid.global.subnet_prefix ==
760 dgid->global.subnet_prefix) &&
761 port_state == IB_PORT_ACTIVE) {
762 cma_dev = cur_dev;
763 sgid = gid;
764 id_priv->id.port_num = p;
765 }
766 }
767 }
768 }
769
770 if (!cma_dev)
771 return -ENODEV;
772
773 found:
774 cma_attach_to_dev(id_priv, cma_dev);
775 addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
776 memcpy(&addr->sib_addr, &sgid, sizeof sgid);
777 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
778 return 0;
779 }
780
781 static void cma_deref_id(struct rdma_id_private *id_priv)
782 {
783 if (atomic_dec_and_test(&id_priv->refcount))
784 complete(&id_priv->comp);
785 }
786
787 struct rdma_cm_id *rdma_create_id(struct net *net,
788 rdma_cm_event_handler event_handler,
789 void *context, enum rdma_port_space ps,
790 enum ib_qp_type qp_type)
791 {
792 struct rdma_id_private *id_priv;
793
794 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
795 if (!id_priv)
796 return ERR_PTR(-ENOMEM);
797
798 id_priv->owner = task_pid_nr(current);
799 id_priv->state = RDMA_CM_IDLE;
800 id_priv->id.context = context;
801 id_priv->id.event_handler = event_handler;
802 id_priv->id.ps = ps;
803 id_priv->id.qp_type = qp_type;
804 id_priv->tos_set = false;
805 spin_lock_init(&id_priv->lock);
806 mutex_init(&id_priv->qp_mutex);
807 init_completion(&id_priv->comp);
808 atomic_set(&id_priv->refcount, 1);
809 mutex_init(&id_priv->handler_mutex);
810 INIT_LIST_HEAD(&id_priv->listen_list);
811 INIT_LIST_HEAD(&id_priv->mc_list);
812 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
813 id_priv->id.route.addr.dev_addr.net = get_net(net);
814
815 return &id_priv->id;
816 }
817 EXPORT_SYMBOL(rdma_create_id);
818
819 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
820 {
821 struct ib_qp_attr qp_attr;
822 int qp_attr_mask, ret;
823
824 qp_attr.qp_state = IB_QPS_INIT;
825 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
826 if (ret)
827 return ret;
828
829 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
830 if (ret)
831 return ret;
832
833 qp_attr.qp_state = IB_QPS_RTR;
834 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
835 if (ret)
836 return ret;
837
838 qp_attr.qp_state = IB_QPS_RTS;
839 qp_attr.sq_psn = 0;
840 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
841
842 return ret;
843 }
844
845 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
846 {
847 struct ib_qp_attr qp_attr;
848 int qp_attr_mask, ret;
849
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 return ret;
854
855 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
856 }
857
858 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
859 struct ib_qp_init_attr *qp_init_attr)
860 {
861 struct rdma_id_private *id_priv;
862 struct ib_qp *qp;
863 int ret;
864
865 id_priv = container_of(id, struct rdma_id_private, id);
866 if (id->device != pd->device)
867 return -EINVAL;
868
869 qp_init_attr->port_num = id->port_num;
870 qp = ib_create_qp(pd, qp_init_attr);
871 if (IS_ERR(qp))
872 return PTR_ERR(qp);
873
874 if (id->qp_type == IB_QPT_UD)
875 ret = cma_init_ud_qp(id_priv, qp);
876 else
877 ret = cma_init_conn_qp(id_priv, qp);
878 if (ret)
879 goto err;
880
881 id->qp = qp;
882 id_priv->qp_num = qp->qp_num;
883 id_priv->srq = (qp->srq != NULL);
884 return 0;
885 err:
886 ib_destroy_qp(qp);
887 return ret;
888 }
889 EXPORT_SYMBOL(rdma_create_qp);
890
891 void rdma_destroy_qp(struct rdma_cm_id *id)
892 {
893 struct rdma_id_private *id_priv;
894
895 id_priv = container_of(id, struct rdma_id_private, id);
896 mutex_lock(&id_priv->qp_mutex);
897 ib_destroy_qp(id_priv->id.qp);
898 id_priv->id.qp = NULL;
899 mutex_unlock(&id_priv->qp_mutex);
900 }
901 EXPORT_SYMBOL(rdma_destroy_qp);
902
903 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
904 struct rdma_conn_param *conn_param)
905 {
906 struct ib_qp_attr qp_attr;
907 int qp_attr_mask, ret;
908 union ib_gid sgid;
909
910 mutex_lock(&id_priv->qp_mutex);
911 if (!id_priv->id.qp) {
912 ret = 0;
913 goto out;
914 }
915
916 /* Need to update QP attributes from default values. */
917 qp_attr.qp_state = IB_QPS_INIT;
918 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
919 if (ret)
920 goto out;
921
922 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
923 if (ret)
924 goto out;
925
926 qp_attr.qp_state = IB_QPS_RTR;
927 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
928 if (ret)
929 goto out;
930
931 ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
932 qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
933 if (ret)
934 goto out;
935
936 BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
937
938 if (conn_param)
939 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
940 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
941 out:
942 mutex_unlock(&id_priv->qp_mutex);
943 return ret;
944 }
945
946 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
947 struct rdma_conn_param *conn_param)
948 {
949 struct ib_qp_attr qp_attr;
950 int qp_attr_mask, ret;
951
952 mutex_lock(&id_priv->qp_mutex);
953 if (!id_priv->id.qp) {
954 ret = 0;
955 goto out;
956 }
957
958 qp_attr.qp_state = IB_QPS_RTS;
959 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
960 if (ret)
961 goto out;
962
963 if (conn_param)
964 qp_attr.max_rd_atomic = conn_param->initiator_depth;
965 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
966 out:
967 mutex_unlock(&id_priv->qp_mutex);
968 return ret;
969 }
970
971 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
972 {
973 struct ib_qp_attr qp_attr;
974 int ret;
975
976 mutex_lock(&id_priv->qp_mutex);
977 if (!id_priv->id.qp) {
978 ret = 0;
979 goto out;
980 }
981
982 qp_attr.qp_state = IB_QPS_ERR;
983 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
984 out:
985 mutex_unlock(&id_priv->qp_mutex);
986 return ret;
987 }
988
989 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
990 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
991 {
992 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
993 int ret;
994 u16 pkey;
995
996 if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
997 pkey = 0xffff;
998 else
999 pkey = ib_addr_get_pkey(dev_addr);
1000
1001 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
1002 pkey, &qp_attr->pkey_index);
1003 if (ret)
1004 return ret;
1005
1006 qp_attr->port_num = id_priv->id.port_num;
1007 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1008
1009 if (id_priv->id.qp_type == IB_QPT_UD) {
1010 ret = cma_set_qkey(id_priv, 0);
1011 if (ret)
1012 return ret;
1013
1014 qp_attr->qkey = id_priv->qkey;
1015 *qp_attr_mask |= IB_QP_QKEY;
1016 } else {
1017 qp_attr->qp_access_flags = 0;
1018 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1019 }
1020 return 0;
1021 }
1022
1023 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1024 int *qp_attr_mask)
1025 {
1026 struct rdma_id_private *id_priv;
1027 int ret = 0;
1028
1029 id_priv = container_of(id, struct rdma_id_private, id);
1030 if (rdma_cap_ib_cm(id->device, id->port_num)) {
1031 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1032 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1033 else
1034 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1035 qp_attr_mask);
1036
1037 if (qp_attr->qp_state == IB_QPS_RTR)
1038 qp_attr->rq_psn = id_priv->seq_num;
1039 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1040 if (!id_priv->cm_id.iw) {
1041 qp_attr->qp_access_flags = 0;
1042 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1043 } else
1044 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1045 qp_attr_mask);
1046 } else
1047 ret = -ENOSYS;
1048
1049 return ret;
1050 }
1051 EXPORT_SYMBOL(rdma_init_qp_attr);
1052
1053 static inline int cma_zero_addr(struct sockaddr *addr)
1054 {
1055 switch (addr->sa_family) {
1056 case AF_INET:
1057 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1058 case AF_INET6:
1059 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1060 case AF_IB:
1061 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1062 default:
1063 return 0;
1064 }
1065 }
1066
1067 static inline int cma_loopback_addr(struct sockaddr *addr)
1068 {
1069 switch (addr->sa_family) {
1070 case AF_INET:
1071 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1072 case AF_INET6:
1073 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1074 case AF_IB:
1075 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1076 default:
1077 return 0;
1078 }
1079 }
1080
1081 static inline int cma_any_addr(struct sockaddr *addr)
1082 {
1083 return cma_zero_addr(addr) || cma_loopback_addr(addr);
1084 }
1085
1086 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1087 {
1088 if (src->sa_family != dst->sa_family)
1089 return -1;
1090
1091 switch (src->sa_family) {
1092 case AF_INET:
1093 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1094 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1095 case AF_INET6:
1096 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1097 &((struct sockaddr_in6 *) dst)->sin6_addr);
1098 default:
1099 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1100 &((struct sockaddr_ib *) dst)->sib_addr);
1101 }
1102 }
1103
1104 static __be16 cma_port(struct sockaddr *addr)
1105 {
1106 struct sockaddr_ib *sib;
1107
1108 switch (addr->sa_family) {
1109 case AF_INET:
1110 return ((struct sockaddr_in *) addr)->sin_port;
1111 case AF_INET6:
1112 return ((struct sockaddr_in6 *) addr)->sin6_port;
1113 case AF_IB:
1114 sib = (struct sockaddr_ib *) addr;
1115 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1116 be64_to_cpu(sib->sib_sid_mask)));
1117 default:
1118 return 0;
1119 }
1120 }
1121
1122 static inline int cma_any_port(struct sockaddr *addr)
1123 {
1124 return !cma_port(addr);
1125 }
1126
1127 static void cma_save_ib_info(struct sockaddr *src_addr,
1128 struct sockaddr *dst_addr,
1129 struct rdma_cm_id *listen_id,
1130 struct ib_sa_path_rec *path)
1131 {
1132 struct sockaddr_ib *listen_ib, *ib;
1133
1134 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1135 if (src_addr) {
1136 ib = (struct sockaddr_ib *)src_addr;
1137 ib->sib_family = AF_IB;
1138 if (path) {
1139 ib->sib_pkey = path->pkey;
1140 ib->sib_flowinfo = path->flow_label;
1141 memcpy(&ib->sib_addr, &path->sgid, 16);
1142 ib->sib_sid = path->service_id;
1143 ib->sib_scope_id = 0;
1144 } else {
1145 ib->sib_pkey = listen_ib->sib_pkey;
1146 ib->sib_flowinfo = listen_ib->sib_flowinfo;
1147 ib->sib_addr = listen_ib->sib_addr;
1148 ib->sib_sid = listen_ib->sib_sid;
1149 ib->sib_scope_id = listen_ib->sib_scope_id;
1150 }
1151 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1152 }
1153 if (dst_addr) {
1154 ib = (struct sockaddr_ib *)dst_addr;
1155 ib->sib_family = AF_IB;
1156 if (path) {
1157 ib->sib_pkey = path->pkey;
1158 ib->sib_flowinfo = path->flow_label;
1159 memcpy(&ib->sib_addr, &path->dgid, 16);
1160 }
1161 }
1162 }
1163
1164 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1165 struct sockaddr_in *dst_addr,
1166 struct cma_hdr *hdr,
1167 __be16 local_port)
1168 {
1169 if (src_addr) {
1170 *src_addr = (struct sockaddr_in) {
1171 .sin_family = AF_INET,
1172 .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1173 .sin_port = local_port,
1174 };
1175 }
1176
1177 if (dst_addr) {
1178 *dst_addr = (struct sockaddr_in) {
1179 .sin_family = AF_INET,
1180 .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1181 .sin_port = hdr->port,
1182 };
1183 }
1184 }
1185
1186 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1187 struct sockaddr_in6 *dst_addr,
1188 struct cma_hdr *hdr,
1189 __be16 local_port)
1190 {
1191 if (src_addr) {
1192 *src_addr = (struct sockaddr_in6) {
1193 .sin6_family = AF_INET6,
1194 .sin6_addr = hdr->dst_addr.ip6,
1195 .sin6_port = local_port,
1196 };
1197 }
1198
1199 if (dst_addr) {
1200 *dst_addr = (struct sockaddr_in6) {
1201 .sin6_family = AF_INET6,
1202 .sin6_addr = hdr->src_addr.ip6,
1203 .sin6_port = hdr->port,
1204 };
1205 }
1206 }
1207
1208 static u16 cma_port_from_service_id(__be64 service_id)
1209 {
1210 return (u16)be64_to_cpu(service_id);
1211 }
1212
1213 static int cma_save_ip_info(struct sockaddr *src_addr,
1214 struct sockaddr *dst_addr,
1215 struct ib_cm_event *ib_event,
1216 __be64 service_id)
1217 {
1218 struct cma_hdr *hdr;
1219 __be16 port;
1220
1221 hdr = ib_event->private_data;
1222 if (hdr->cma_version != CMA_VERSION)
1223 return -EINVAL;
1224
1225 port = htons(cma_port_from_service_id(service_id));
1226
1227 switch (cma_get_ip_ver(hdr)) {
1228 case 4:
1229 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1230 (struct sockaddr_in *)dst_addr, hdr, port);
1231 break;
1232 case 6:
1233 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1234 (struct sockaddr_in6 *)dst_addr, hdr, port);
1235 break;
1236 default:
1237 return -EAFNOSUPPORT;
1238 }
1239
1240 return 0;
1241 }
1242
1243 static int cma_save_net_info(struct sockaddr *src_addr,
1244 struct sockaddr *dst_addr,
1245 struct rdma_cm_id *listen_id,
1246 struct ib_cm_event *ib_event,
1247 sa_family_t sa_family, __be64 service_id)
1248 {
1249 if (sa_family == AF_IB) {
1250 if (ib_event->event == IB_CM_REQ_RECEIVED)
1251 cma_save_ib_info(src_addr, dst_addr, listen_id,
1252 ib_event->param.req_rcvd.primary_path);
1253 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1254 cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1255 return 0;
1256 }
1257
1258 return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1259 }
1260
1261 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1262 struct cma_req_info *req)
1263 {
1264 const struct ib_cm_req_event_param *req_param =
1265 &ib_event->param.req_rcvd;
1266 const struct ib_cm_sidr_req_event_param *sidr_param =
1267 &ib_event->param.sidr_req_rcvd;
1268
1269 switch (ib_event->event) {
1270 case IB_CM_REQ_RECEIVED:
1271 req->device = req_param->listen_id->device;
1272 req->port = req_param->port;
1273 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1274 sizeof(req->local_gid));
1275 req->has_gid = true;
1276 req->service_id = req_param->primary_path->service_id;
1277 req->pkey = be16_to_cpu(req_param->primary_path->pkey);
1278 if (req->pkey != req_param->bth_pkey)
1279 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1280 "RDMA CMA: in the future this may cause the request to be dropped\n",
1281 req_param->bth_pkey, req->pkey);
1282 break;
1283 case IB_CM_SIDR_REQ_RECEIVED:
1284 req->device = sidr_param->listen_id->device;
1285 req->port = sidr_param->port;
1286 req->has_gid = false;
1287 req->service_id = sidr_param->service_id;
1288 req->pkey = sidr_param->pkey;
1289 if (req->pkey != sidr_param->bth_pkey)
1290 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1291 "RDMA CMA: in the future this may cause the request to be dropped\n",
1292 sidr_param->bth_pkey, req->pkey);
1293 break;
1294 default:
1295 return -EINVAL;
1296 }
1297
1298 return 0;
1299 }
1300
1301 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1302 const struct sockaddr_in *dst_addr,
1303 const struct sockaddr_in *src_addr)
1304 {
1305 __be32 daddr = dst_addr->sin_addr.s_addr,
1306 saddr = src_addr->sin_addr.s_addr;
1307 struct fib_result res;
1308 struct flowi4 fl4;
1309 int err;
1310 bool ret;
1311
1312 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1313 ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1314 ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1315 ipv4_is_loopback(saddr))
1316 return false;
1317
1318 memset(&fl4, 0, sizeof(fl4));
1319 fl4.flowi4_iif = net_dev->ifindex;
1320 fl4.daddr = daddr;
1321 fl4.saddr = saddr;
1322
1323 rcu_read_lock();
1324 err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1325 ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1326 rcu_read_unlock();
1327
1328 return ret;
1329 }
1330
1331 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1332 const struct sockaddr_in6 *dst_addr,
1333 const struct sockaddr_in6 *src_addr)
1334 {
1335 #if IS_ENABLED(CONFIG_IPV6)
1336 const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1337 IPV6_ADDR_LINKLOCAL;
1338 struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1339 &src_addr->sin6_addr, net_dev->ifindex,
1340 strict);
1341 bool ret;
1342
1343 if (!rt)
1344 return false;
1345
1346 ret = rt->rt6i_idev->dev == net_dev;
1347 ip6_rt_put(rt);
1348
1349 return ret;
1350 #else
1351 return false;
1352 #endif
1353 }
1354
1355 static bool validate_net_dev(struct net_device *net_dev,
1356 const struct sockaddr *daddr,
1357 const struct sockaddr *saddr)
1358 {
1359 const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1360 const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1361 const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1362 const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1363
1364 switch (daddr->sa_family) {
1365 case AF_INET:
1366 return saddr->sa_family == AF_INET &&
1367 validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1368
1369 case AF_INET6:
1370 return saddr->sa_family == AF_INET6 &&
1371 validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1372
1373 default:
1374 return false;
1375 }
1376 }
1377
1378 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1379 const struct cma_req_info *req)
1380 {
1381 struct sockaddr_storage listen_addr_storage, src_addr_storage;
1382 struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1383 *src_addr = (struct sockaddr *)&src_addr_storage;
1384 struct net_device *net_dev;
1385 const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1386 int err;
1387
1388 err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1389 req->service_id);
1390 if (err)
1391 return ERR_PTR(err);
1392
1393 net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1394 gid, listen_addr);
1395 if (!net_dev)
1396 return ERR_PTR(-ENODEV);
1397
1398 if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1399 dev_put(net_dev);
1400 return ERR_PTR(-EHOSTUNREACH);
1401 }
1402
1403 return net_dev;
1404 }
1405
1406 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1407 {
1408 return (be64_to_cpu(service_id) >> 16) & 0xffff;
1409 }
1410
1411 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1412 const struct cma_hdr *hdr)
1413 {
1414 struct sockaddr *addr = cma_src_addr(id_priv);
1415 __be32 ip4_addr;
1416 struct in6_addr ip6_addr;
1417
1418 if (cma_any_addr(addr) && !id_priv->afonly)
1419 return true;
1420
1421 switch (addr->sa_family) {
1422 case AF_INET:
1423 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1424 if (cma_get_ip_ver(hdr) != 4)
1425 return false;
1426 if (!cma_any_addr(addr) &&
1427 hdr->dst_addr.ip4.addr != ip4_addr)
1428 return false;
1429 break;
1430 case AF_INET6:
1431 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1432 if (cma_get_ip_ver(hdr) != 6)
1433 return false;
1434 if (!cma_any_addr(addr) &&
1435 memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1436 return false;
1437 break;
1438 case AF_IB:
1439 return true;
1440 default:
1441 return false;
1442 }
1443
1444 return true;
1445 }
1446
1447 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1448 {
1449 enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1450 enum rdma_transport_type transport =
1451 rdma_node_get_transport(device->node_type);
1452
1453 return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1454 }
1455
1456 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1457 {
1458 struct ib_device *device = id->device;
1459 const int port_num = id->port_num ?: rdma_start_port(device);
1460
1461 return cma_protocol_roce_dev_port(device, port_num);
1462 }
1463
1464 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1465 const struct net_device *net_dev,
1466 u8 port_num)
1467 {
1468 const struct rdma_addr *addr = &id->route.addr;
1469
1470 if (!net_dev)
1471 /* This request is an AF_IB request or a RoCE request */
1472 return (!id->port_num || id->port_num == port_num) &&
1473 (addr->src_addr.ss_family == AF_IB ||
1474 cma_protocol_roce_dev_port(id->device, port_num));
1475
1476 return !addr->dev_addr.bound_dev_if ||
1477 (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1478 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1479 }
1480
1481 static struct rdma_id_private *cma_find_listener(
1482 const struct rdma_bind_list *bind_list,
1483 const struct ib_cm_id *cm_id,
1484 const struct ib_cm_event *ib_event,
1485 const struct cma_req_info *req,
1486 const struct net_device *net_dev)
1487 {
1488 struct rdma_id_private *id_priv, *id_priv_dev;
1489
1490 if (!bind_list)
1491 return ERR_PTR(-EINVAL);
1492
1493 hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1494 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1495 if (id_priv->id.device == cm_id->device &&
1496 cma_match_net_dev(&id_priv->id, net_dev, req->port))
1497 return id_priv;
1498 list_for_each_entry(id_priv_dev,
1499 &id_priv->listen_list,
1500 listen_list) {
1501 if (id_priv_dev->id.device == cm_id->device &&
1502 cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1503 return id_priv_dev;
1504 }
1505 }
1506 }
1507
1508 return ERR_PTR(-EINVAL);
1509 }
1510
1511 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1512 struct ib_cm_event *ib_event,
1513 struct net_device **net_dev)
1514 {
1515 struct cma_req_info req;
1516 struct rdma_bind_list *bind_list;
1517 struct rdma_id_private *id_priv;
1518 int err;
1519
1520 err = cma_save_req_info(ib_event, &req);
1521 if (err)
1522 return ERR_PTR(err);
1523
1524 *net_dev = cma_get_net_dev(ib_event, &req);
1525 if (IS_ERR(*net_dev)) {
1526 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1527 /* Assuming the protocol is AF_IB */
1528 *net_dev = NULL;
1529 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1530 /* TODO find the net dev matching the request parameters
1531 * through the RoCE GID table */
1532 *net_dev = NULL;
1533 } else {
1534 return ERR_CAST(*net_dev);
1535 }
1536 }
1537
1538 bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1539 rdma_ps_from_service_id(req.service_id),
1540 cma_port_from_service_id(req.service_id));
1541 id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1542 if (IS_ERR(id_priv) && *net_dev) {
1543 dev_put(*net_dev);
1544 *net_dev = NULL;
1545 }
1546
1547 return id_priv;
1548 }
1549
1550 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1551 {
1552 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1553 }
1554
1555 static void cma_cancel_route(struct rdma_id_private *id_priv)
1556 {
1557 if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1558 if (id_priv->query)
1559 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1560 }
1561 }
1562
1563 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1564 {
1565 struct rdma_id_private *dev_id_priv;
1566
1567 /*
1568 * Remove from listen_any_list to prevent added devices from spawning
1569 * additional listen requests.
1570 */
1571 mutex_lock(&lock);
1572 list_del(&id_priv->list);
1573
1574 while (!list_empty(&id_priv->listen_list)) {
1575 dev_id_priv = list_entry(id_priv->listen_list.next,
1576 struct rdma_id_private, listen_list);
1577 /* sync with device removal to avoid duplicate destruction */
1578 list_del_init(&dev_id_priv->list);
1579 list_del(&dev_id_priv->listen_list);
1580 mutex_unlock(&lock);
1581
1582 rdma_destroy_id(&dev_id_priv->id);
1583 mutex_lock(&lock);
1584 }
1585 mutex_unlock(&lock);
1586 }
1587
1588 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1589 enum rdma_cm_state state)
1590 {
1591 switch (state) {
1592 case RDMA_CM_ADDR_QUERY:
1593 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1594 break;
1595 case RDMA_CM_ROUTE_QUERY:
1596 cma_cancel_route(id_priv);
1597 break;
1598 case RDMA_CM_LISTEN:
1599 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1600 cma_cancel_listens(id_priv);
1601 break;
1602 default:
1603 break;
1604 }
1605 }
1606
1607 static void cma_release_port(struct rdma_id_private *id_priv)
1608 {
1609 struct rdma_bind_list *bind_list = id_priv->bind_list;
1610 struct net *net = id_priv->id.route.addr.dev_addr.net;
1611
1612 if (!bind_list)
1613 return;
1614
1615 mutex_lock(&lock);
1616 hlist_del(&id_priv->node);
1617 if (hlist_empty(&bind_list->owners)) {
1618 cma_ps_remove(net, bind_list->ps, bind_list->port);
1619 kfree(bind_list);
1620 }
1621 mutex_unlock(&lock);
1622 }
1623
1624 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1625 {
1626 struct cma_multicast *mc;
1627
1628 while (!list_empty(&id_priv->mc_list)) {
1629 mc = container_of(id_priv->mc_list.next,
1630 struct cma_multicast, list);
1631 list_del(&mc->list);
1632 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1633 id_priv->id.port_num)) {
1634 ib_sa_free_multicast(mc->multicast.ib);
1635 kfree(mc);
1636 } else {
1637 if (mc->igmp_joined) {
1638 struct rdma_dev_addr *dev_addr =
1639 &id_priv->id.route.addr.dev_addr;
1640 struct net_device *ndev = NULL;
1641
1642 if (dev_addr->bound_dev_if)
1643 ndev = dev_get_by_index(&init_net,
1644 dev_addr->bound_dev_if);
1645 if (ndev) {
1646 cma_igmp_send(ndev,
1647 &mc->multicast.ib->rec.mgid,
1648 false);
1649 dev_put(ndev);
1650 }
1651 }
1652 kref_put(&mc->mcref, release_mc);
1653 }
1654 }
1655 }
1656
1657 void rdma_destroy_id(struct rdma_cm_id *id)
1658 {
1659 struct rdma_id_private *id_priv;
1660 enum rdma_cm_state state;
1661
1662 id_priv = container_of(id, struct rdma_id_private, id);
1663 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1664 cma_cancel_operation(id_priv, state);
1665
1666 /*
1667 * Wait for any active callback to finish. New callbacks will find
1668 * the id_priv state set to destroying and abort.
1669 */
1670 mutex_lock(&id_priv->handler_mutex);
1671 mutex_unlock(&id_priv->handler_mutex);
1672
1673 if (id_priv->cma_dev) {
1674 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1675 if (id_priv->cm_id.ib)
1676 ib_destroy_cm_id(id_priv->cm_id.ib);
1677 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1678 if (id_priv->cm_id.iw)
1679 iw_destroy_cm_id(id_priv->cm_id.iw);
1680 }
1681 cma_leave_mc_groups(id_priv);
1682 cma_release_dev(id_priv);
1683 }
1684
1685 cma_release_port(id_priv);
1686 cma_deref_id(id_priv);
1687 wait_for_completion(&id_priv->comp);
1688
1689 if (id_priv->internal_id)
1690 cma_deref_id(id_priv->id.context);
1691
1692 kfree(id_priv->id.route.path_rec);
1693 put_net(id_priv->id.route.addr.dev_addr.net);
1694 kfree(id_priv);
1695 }
1696 EXPORT_SYMBOL(rdma_destroy_id);
1697
1698 static int cma_rep_recv(struct rdma_id_private *id_priv)
1699 {
1700 int ret;
1701
1702 ret = cma_modify_qp_rtr(id_priv, NULL);
1703 if (ret)
1704 goto reject;
1705
1706 ret = cma_modify_qp_rts(id_priv, NULL);
1707 if (ret)
1708 goto reject;
1709
1710 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1711 if (ret)
1712 goto reject;
1713
1714 return 0;
1715 reject:
1716 pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1717 cma_modify_qp_err(id_priv);
1718 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1719 NULL, 0, NULL, 0);
1720 return ret;
1721 }
1722
1723 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1724 struct ib_cm_rep_event_param *rep_data,
1725 void *private_data)
1726 {
1727 event->param.conn.private_data = private_data;
1728 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1729 event->param.conn.responder_resources = rep_data->responder_resources;
1730 event->param.conn.initiator_depth = rep_data->initiator_depth;
1731 event->param.conn.flow_control = rep_data->flow_control;
1732 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1733 event->param.conn.srq = rep_data->srq;
1734 event->param.conn.qp_num = rep_data->remote_qpn;
1735 }
1736
1737 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1738 {
1739 struct rdma_id_private *id_priv = cm_id->context;
1740 struct rdma_cm_event event;
1741 int ret = 0;
1742
1743 mutex_lock(&id_priv->handler_mutex);
1744 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1745 id_priv->state != RDMA_CM_CONNECT) ||
1746 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1747 id_priv->state != RDMA_CM_DISCONNECT))
1748 goto out;
1749
1750 memset(&event, 0, sizeof event);
1751 switch (ib_event->event) {
1752 case IB_CM_REQ_ERROR:
1753 case IB_CM_REP_ERROR:
1754 event.event = RDMA_CM_EVENT_UNREACHABLE;
1755 event.status = -ETIMEDOUT;
1756 break;
1757 case IB_CM_REP_RECEIVED:
1758 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1759 (id_priv->id.qp_type != IB_QPT_UD))
1760 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1761 if (id_priv->id.qp) {
1762 event.status = cma_rep_recv(id_priv);
1763 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1764 RDMA_CM_EVENT_ESTABLISHED;
1765 } else {
1766 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1767 }
1768 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1769 ib_event->private_data);
1770 break;
1771 case IB_CM_RTU_RECEIVED:
1772 case IB_CM_USER_ESTABLISHED:
1773 event.event = RDMA_CM_EVENT_ESTABLISHED;
1774 break;
1775 case IB_CM_DREQ_ERROR:
1776 event.status = -ETIMEDOUT; /* fall through */
1777 case IB_CM_DREQ_RECEIVED:
1778 case IB_CM_DREP_RECEIVED:
1779 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1780 RDMA_CM_DISCONNECT))
1781 goto out;
1782 event.event = RDMA_CM_EVENT_DISCONNECTED;
1783 break;
1784 case IB_CM_TIMEWAIT_EXIT:
1785 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1786 break;
1787 case IB_CM_MRA_RECEIVED:
1788 /* ignore event */
1789 goto out;
1790 case IB_CM_REJ_RECEIVED:
1791 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1792 ib_event->param.rej_rcvd.reason));
1793 cma_modify_qp_err(id_priv);
1794 event.status = ib_event->param.rej_rcvd.reason;
1795 event.event = RDMA_CM_EVENT_REJECTED;
1796 event.param.conn.private_data = ib_event->private_data;
1797 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1798 break;
1799 default:
1800 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1801 ib_event->event);
1802 goto out;
1803 }
1804
1805 ret = id_priv->id.event_handler(&id_priv->id, &event);
1806 if (ret) {
1807 /* Destroy the CM ID by returning a non-zero value. */
1808 id_priv->cm_id.ib = NULL;
1809 cma_exch(id_priv, RDMA_CM_DESTROYING);
1810 mutex_unlock(&id_priv->handler_mutex);
1811 rdma_destroy_id(&id_priv->id);
1812 return ret;
1813 }
1814 out:
1815 mutex_unlock(&id_priv->handler_mutex);
1816 return ret;
1817 }
1818
1819 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1820 struct ib_cm_event *ib_event,
1821 struct net_device *net_dev)
1822 {
1823 struct rdma_id_private *id_priv;
1824 struct rdma_cm_id *id;
1825 struct rdma_route *rt;
1826 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1827 const __be64 service_id =
1828 ib_event->param.req_rcvd.primary_path->service_id;
1829 int ret;
1830
1831 id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1832 listen_id->event_handler, listen_id->context,
1833 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1834 if (IS_ERR(id))
1835 return NULL;
1836
1837 id_priv = container_of(id, struct rdma_id_private, id);
1838 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1839 (struct sockaddr *)&id->route.addr.dst_addr,
1840 listen_id, ib_event, ss_family, service_id))
1841 goto err;
1842
1843 rt = &id->route;
1844 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1845 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1846 GFP_KERNEL);
1847 if (!rt->path_rec)
1848 goto err;
1849
1850 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1851 if (rt->num_paths == 2)
1852 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1853
1854 if (net_dev) {
1855 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1856 if (ret)
1857 goto err;
1858 } else {
1859 if (!cma_protocol_roce(listen_id) &&
1860 cma_any_addr(cma_src_addr(id_priv))) {
1861 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1862 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1863 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1864 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1865 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1866 if (ret)
1867 goto err;
1868 }
1869 }
1870 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1871
1872 id_priv->state = RDMA_CM_CONNECT;
1873 return id_priv;
1874
1875 err:
1876 rdma_destroy_id(id);
1877 return NULL;
1878 }
1879
1880 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1881 struct ib_cm_event *ib_event,
1882 struct net_device *net_dev)
1883 {
1884 struct rdma_id_private *id_priv;
1885 struct rdma_cm_id *id;
1886 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1887 struct net *net = listen_id->route.addr.dev_addr.net;
1888 int ret;
1889
1890 id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1891 listen_id->ps, IB_QPT_UD);
1892 if (IS_ERR(id))
1893 return NULL;
1894
1895 id_priv = container_of(id, struct rdma_id_private, id);
1896 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1897 (struct sockaddr *)&id->route.addr.dst_addr,
1898 listen_id, ib_event, ss_family,
1899 ib_event->param.sidr_req_rcvd.service_id))
1900 goto err;
1901
1902 if (net_dev) {
1903 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1904 if (ret)
1905 goto err;
1906 } else {
1907 if (!cma_any_addr(cma_src_addr(id_priv))) {
1908 ret = cma_translate_addr(cma_src_addr(id_priv),
1909 &id->route.addr.dev_addr);
1910 if (ret)
1911 goto err;
1912 }
1913 }
1914
1915 id_priv->state = RDMA_CM_CONNECT;
1916 return id_priv;
1917 err:
1918 rdma_destroy_id(id);
1919 return NULL;
1920 }
1921
1922 static void cma_set_req_event_data(struct rdma_cm_event *event,
1923 struct ib_cm_req_event_param *req_data,
1924 void *private_data, int offset)
1925 {
1926 event->param.conn.private_data = private_data + offset;
1927 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1928 event->param.conn.responder_resources = req_data->responder_resources;
1929 event->param.conn.initiator_depth = req_data->initiator_depth;
1930 event->param.conn.flow_control = req_data->flow_control;
1931 event->param.conn.retry_count = req_data->retry_count;
1932 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1933 event->param.conn.srq = req_data->srq;
1934 event->param.conn.qp_num = req_data->remote_qpn;
1935 }
1936
1937 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1938 {
1939 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1940 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1941 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1942 (id->qp_type == IB_QPT_UD)) ||
1943 (!id->qp_type));
1944 }
1945
1946 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1947 {
1948 struct rdma_id_private *listen_id, *conn_id = NULL;
1949 struct rdma_cm_event event;
1950 struct net_device *net_dev;
1951 int offset, ret;
1952
1953 listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1954 if (IS_ERR(listen_id))
1955 return PTR_ERR(listen_id);
1956
1957 if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1958 ret = -EINVAL;
1959 goto net_dev_put;
1960 }
1961
1962 mutex_lock(&listen_id->handler_mutex);
1963 if (listen_id->state != RDMA_CM_LISTEN) {
1964 ret = -ECONNABORTED;
1965 goto err1;
1966 }
1967
1968 memset(&event, 0, sizeof event);
1969 offset = cma_user_data_offset(listen_id);
1970 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1971 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1972 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1973 event.param.ud.private_data = ib_event->private_data + offset;
1974 event.param.ud.private_data_len =
1975 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1976 } else {
1977 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1978 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1979 ib_event->private_data, offset);
1980 }
1981 if (!conn_id) {
1982 ret = -ENOMEM;
1983 goto err1;
1984 }
1985
1986 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1987 ret = cma_acquire_dev(conn_id, listen_id);
1988 if (ret)
1989 goto err2;
1990
1991 conn_id->cm_id.ib = cm_id;
1992 cm_id->context = conn_id;
1993 cm_id->cm_handler = cma_ib_handler;
1994
1995 /*
1996 * Protect against the user destroying conn_id from another thread
1997 * until we're done accessing it.
1998 */
1999 atomic_inc(&conn_id->refcount);
2000 ret = conn_id->id.event_handler(&conn_id->id, &event);
2001 if (ret)
2002 goto err3;
2003 /*
2004 * Acquire mutex to prevent user executing rdma_destroy_id()
2005 * while we're accessing the cm_id.
2006 */
2007 mutex_lock(&lock);
2008 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2009 (conn_id->id.qp_type != IB_QPT_UD))
2010 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2011 mutex_unlock(&lock);
2012 mutex_unlock(&conn_id->handler_mutex);
2013 mutex_unlock(&listen_id->handler_mutex);
2014 cma_deref_id(conn_id);
2015 if (net_dev)
2016 dev_put(net_dev);
2017 return 0;
2018
2019 err3:
2020 cma_deref_id(conn_id);
2021 /* Destroy the CM ID by returning a non-zero value. */
2022 conn_id->cm_id.ib = NULL;
2023 err2:
2024 cma_exch(conn_id, RDMA_CM_DESTROYING);
2025 mutex_unlock(&conn_id->handler_mutex);
2026 err1:
2027 mutex_unlock(&listen_id->handler_mutex);
2028 if (conn_id)
2029 rdma_destroy_id(&conn_id->id);
2030
2031 net_dev_put:
2032 if (net_dev)
2033 dev_put(net_dev);
2034
2035 return ret;
2036 }
2037
2038 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2039 {
2040 if (addr->sa_family == AF_IB)
2041 return ((struct sockaddr_ib *) addr)->sib_sid;
2042
2043 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2044 }
2045 EXPORT_SYMBOL(rdma_get_service_id);
2046
2047 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2048 {
2049 struct rdma_id_private *id_priv = iw_id->context;
2050 struct rdma_cm_event event;
2051 int ret = 0;
2052 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2053 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2054
2055 mutex_lock(&id_priv->handler_mutex);
2056 if (id_priv->state != RDMA_CM_CONNECT)
2057 goto out;
2058
2059 memset(&event, 0, sizeof event);
2060 switch (iw_event->event) {
2061 case IW_CM_EVENT_CLOSE:
2062 event.event = RDMA_CM_EVENT_DISCONNECTED;
2063 break;
2064 case IW_CM_EVENT_CONNECT_REPLY:
2065 memcpy(cma_src_addr(id_priv), laddr,
2066 rdma_addr_size(laddr));
2067 memcpy(cma_dst_addr(id_priv), raddr,
2068 rdma_addr_size(raddr));
2069 switch (iw_event->status) {
2070 case 0:
2071 event.event = RDMA_CM_EVENT_ESTABLISHED;
2072 event.param.conn.initiator_depth = iw_event->ird;
2073 event.param.conn.responder_resources = iw_event->ord;
2074 break;
2075 case -ECONNRESET:
2076 case -ECONNREFUSED:
2077 event.event = RDMA_CM_EVENT_REJECTED;
2078 break;
2079 case -ETIMEDOUT:
2080 event.event = RDMA_CM_EVENT_UNREACHABLE;
2081 break;
2082 default:
2083 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2084 break;
2085 }
2086 break;
2087 case IW_CM_EVENT_ESTABLISHED:
2088 event.event = RDMA_CM_EVENT_ESTABLISHED;
2089 event.param.conn.initiator_depth = iw_event->ird;
2090 event.param.conn.responder_resources = iw_event->ord;
2091 break;
2092 default:
2093 BUG_ON(1);
2094 }
2095
2096 event.status = iw_event->status;
2097 event.param.conn.private_data = iw_event->private_data;
2098 event.param.conn.private_data_len = iw_event->private_data_len;
2099 ret = id_priv->id.event_handler(&id_priv->id, &event);
2100 if (ret) {
2101 /* Destroy the CM ID by returning a non-zero value. */
2102 id_priv->cm_id.iw = NULL;
2103 cma_exch(id_priv, RDMA_CM_DESTROYING);
2104 mutex_unlock(&id_priv->handler_mutex);
2105 rdma_destroy_id(&id_priv->id);
2106 return ret;
2107 }
2108
2109 out:
2110 mutex_unlock(&id_priv->handler_mutex);
2111 return ret;
2112 }
2113
2114 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2115 struct iw_cm_event *iw_event)
2116 {
2117 struct rdma_cm_id *new_cm_id;
2118 struct rdma_id_private *listen_id, *conn_id;
2119 struct rdma_cm_event event;
2120 int ret = -ECONNABORTED;
2121 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2122 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2123
2124 listen_id = cm_id->context;
2125
2126 mutex_lock(&listen_id->handler_mutex);
2127 if (listen_id->state != RDMA_CM_LISTEN)
2128 goto out;
2129
2130 /* Create a new RDMA id for the new IW CM ID */
2131 new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2132 listen_id->id.event_handler,
2133 listen_id->id.context,
2134 RDMA_PS_TCP, IB_QPT_RC);
2135 if (IS_ERR(new_cm_id)) {
2136 ret = -ENOMEM;
2137 goto out;
2138 }
2139 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2140 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2141 conn_id->state = RDMA_CM_CONNECT;
2142
2143 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2144 if (ret) {
2145 mutex_unlock(&conn_id->handler_mutex);
2146 rdma_destroy_id(new_cm_id);
2147 goto out;
2148 }
2149
2150 ret = cma_acquire_dev(conn_id, listen_id);
2151 if (ret) {
2152 mutex_unlock(&conn_id->handler_mutex);
2153 rdma_destroy_id(new_cm_id);
2154 goto out;
2155 }
2156
2157 conn_id->cm_id.iw = cm_id;
2158 cm_id->context = conn_id;
2159 cm_id->cm_handler = cma_iw_handler;
2160
2161 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2162 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2163
2164 memset(&event, 0, sizeof event);
2165 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2166 event.param.conn.private_data = iw_event->private_data;
2167 event.param.conn.private_data_len = iw_event->private_data_len;
2168 event.param.conn.initiator_depth = iw_event->ird;
2169 event.param.conn.responder_resources = iw_event->ord;
2170
2171 /*
2172 * Protect against the user destroying conn_id from another thread
2173 * until we're done accessing it.
2174 */
2175 atomic_inc(&conn_id->refcount);
2176 ret = conn_id->id.event_handler(&conn_id->id, &event);
2177 if (ret) {
2178 /* User wants to destroy the CM ID */
2179 conn_id->cm_id.iw = NULL;
2180 cma_exch(conn_id, RDMA_CM_DESTROYING);
2181 mutex_unlock(&conn_id->handler_mutex);
2182 cma_deref_id(conn_id);
2183 rdma_destroy_id(&conn_id->id);
2184 goto out;
2185 }
2186
2187 mutex_unlock(&conn_id->handler_mutex);
2188 cma_deref_id(conn_id);
2189
2190 out:
2191 mutex_unlock(&listen_id->handler_mutex);
2192 return ret;
2193 }
2194
2195 static int cma_ib_listen(struct rdma_id_private *id_priv)
2196 {
2197 struct sockaddr *addr;
2198 struct ib_cm_id *id;
2199 __be64 svc_id;
2200
2201 addr = cma_src_addr(id_priv);
2202 svc_id = rdma_get_service_id(&id_priv->id, addr);
2203 id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2204 if (IS_ERR(id))
2205 return PTR_ERR(id);
2206 id_priv->cm_id.ib = id;
2207
2208 return 0;
2209 }
2210
2211 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2212 {
2213 int ret;
2214 struct iw_cm_id *id;
2215
2216 id = iw_create_cm_id(id_priv->id.device,
2217 iw_conn_req_handler,
2218 id_priv);
2219 if (IS_ERR(id))
2220 return PTR_ERR(id);
2221
2222 id->tos = id_priv->tos;
2223 id_priv->cm_id.iw = id;
2224
2225 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2226 rdma_addr_size(cma_src_addr(id_priv)));
2227
2228 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2229
2230 if (ret) {
2231 iw_destroy_cm_id(id_priv->cm_id.iw);
2232 id_priv->cm_id.iw = NULL;
2233 }
2234
2235 return ret;
2236 }
2237
2238 static int cma_listen_handler(struct rdma_cm_id *id,
2239 struct rdma_cm_event *event)
2240 {
2241 struct rdma_id_private *id_priv = id->context;
2242
2243 id->context = id_priv->id.context;
2244 id->event_handler = id_priv->id.event_handler;
2245 return id_priv->id.event_handler(id, event);
2246 }
2247
2248 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2249 struct cma_device *cma_dev)
2250 {
2251 struct rdma_id_private *dev_id_priv;
2252 struct rdma_cm_id *id;
2253 struct net *net = id_priv->id.route.addr.dev_addr.net;
2254 int ret;
2255
2256 if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2257 return;
2258
2259 id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2260 id_priv->id.qp_type);
2261 if (IS_ERR(id))
2262 return;
2263
2264 dev_id_priv = container_of(id, struct rdma_id_private, id);
2265
2266 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2267 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2268 rdma_addr_size(cma_src_addr(id_priv)));
2269
2270 _cma_attach_to_dev(dev_id_priv, cma_dev);
2271 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2272 atomic_inc(&id_priv->refcount);
2273 dev_id_priv->internal_id = 1;
2274 dev_id_priv->afonly = id_priv->afonly;
2275
2276 ret = rdma_listen(id, id_priv->backlog);
2277 if (ret)
2278 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2279 ret, cma_dev->device->name);
2280 }
2281
2282 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2283 {
2284 struct cma_device *cma_dev;
2285
2286 mutex_lock(&lock);
2287 list_add_tail(&id_priv->list, &listen_any_list);
2288 list_for_each_entry(cma_dev, &dev_list, list)
2289 cma_listen_on_dev(id_priv, cma_dev);
2290 mutex_unlock(&lock);
2291 }
2292
2293 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2294 {
2295 struct rdma_id_private *id_priv;
2296
2297 id_priv = container_of(id, struct rdma_id_private, id);
2298 id_priv->tos = (u8) tos;
2299 id_priv->tos_set = true;
2300 }
2301 EXPORT_SYMBOL(rdma_set_service_type);
2302
2303 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2304 void *context)
2305 {
2306 struct cma_work *work = context;
2307 struct rdma_route *route;
2308
2309 route = &work->id->id.route;
2310
2311 if (!status) {
2312 route->num_paths = 1;
2313 *route->path_rec = *path_rec;
2314 } else {
2315 work->old_state = RDMA_CM_ROUTE_QUERY;
2316 work->new_state = RDMA_CM_ADDR_RESOLVED;
2317 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2318 work->event.status = status;
2319 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2320 status);
2321 }
2322
2323 queue_work(cma_wq, &work->work);
2324 }
2325
2326 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2327 struct cma_work *work)
2328 {
2329 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2330 struct ib_sa_path_rec path_rec;
2331 ib_sa_comp_mask comp_mask;
2332 struct sockaddr_in6 *sin6;
2333 struct sockaddr_ib *sib;
2334
2335 memset(&path_rec, 0, sizeof path_rec);
2336 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2337 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2338 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2339 path_rec.numb_path = 1;
2340 path_rec.reversible = 1;
2341 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2342
2343 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2344 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2345 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2346
2347 switch (cma_family(id_priv)) {
2348 case AF_INET:
2349 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2350 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2351 break;
2352 case AF_INET6:
2353 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2354 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2355 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2356 break;
2357 case AF_IB:
2358 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2359 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2360 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2361 break;
2362 }
2363
2364 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2365 id_priv->id.port_num, &path_rec,
2366 comp_mask, timeout_ms,
2367 GFP_KERNEL, cma_query_handler,
2368 work, &id_priv->query);
2369
2370 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2371 }
2372
2373 static void cma_work_handler(struct work_struct *_work)
2374 {
2375 struct cma_work *work = container_of(_work, struct cma_work, work);
2376 struct rdma_id_private *id_priv = work->id;
2377 int destroy = 0;
2378
2379 mutex_lock(&id_priv->handler_mutex);
2380 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2381 goto out;
2382
2383 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2384 cma_exch(id_priv, RDMA_CM_DESTROYING);
2385 destroy = 1;
2386 }
2387 out:
2388 mutex_unlock(&id_priv->handler_mutex);
2389 cma_deref_id(id_priv);
2390 if (destroy)
2391 rdma_destroy_id(&id_priv->id);
2392 kfree(work);
2393 }
2394
2395 static void cma_ndev_work_handler(struct work_struct *_work)
2396 {
2397 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2398 struct rdma_id_private *id_priv = work->id;
2399 int destroy = 0;
2400
2401 mutex_lock(&id_priv->handler_mutex);
2402 if (id_priv->state == RDMA_CM_DESTROYING ||
2403 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2404 goto out;
2405
2406 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2407 cma_exch(id_priv, RDMA_CM_DESTROYING);
2408 destroy = 1;
2409 }
2410
2411 out:
2412 mutex_unlock(&id_priv->handler_mutex);
2413 cma_deref_id(id_priv);
2414 if (destroy)
2415 rdma_destroy_id(&id_priv->id);
2416 kfree(work);
2417 }
2418
2419 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2420 {
2421 struct rdma_route *route = &id_priv->id.route;
2422 struct cma_work *work;
2423 int ret;
2424
2425 work = kzalloc(sizeof *work, GFP_KERNEL);
2426 if (!work)
2427 return -ENOMEM;
2428
2429 work->id = id_priv;
2430 INIT_WORK(&work->work, cma_work_handler);
2431 work->old_state = RDMA_CM_ROUTE_QUERY;
2432 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2433 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2434
2435 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2436 if (!route->path_rec) {
2437 ret = -ENOMEM;
2438 goto err1;
2439 }
2440
2441 ret = cma_query_ib_route(id_priv, timeout_ms, work);
2442 if (ret)
2443 goto err2;
2444
2445 return 0;
2446 err2:
2447 kfree(route->path_rec);
2448 route->path_rec = NULL;
2449 err1:
2450 kfree(work);
2451 return ret;
2452 }
2453
2454 int rdma_set_ib_paths(struct rdma_cm_id *id,
2455 struct ib_sa_path_rec *path_rec, int num_paths)
2456 {
2457 struct rdma_id_private *id_priv;
2458 int ret;
2459
2460 id_priv = container_of(id, struct rdma_id_private, id);
2461 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2462 RDMA_CM_ROUTE_RESOLVED))
2463 return -EINVAL;
2464
2465 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2466 GFP_KERNEL);
2467 if (!id->route.path_rec) {
2468 ret = -ENOMEM;
2469 goto err;
2470 }
2471
2472 id->route.num_paths = num_paths;
2473 return 0;
2474 err:
2475 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2476 return ret;
2477 }
2478 EXPORT_SYMBOL(rdma_set_ib_paths);
2479
2480 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2481 {
2482 struct cma_work *work;
2483
2484 work = kzalloc(sizeof *work, GFP_KERNEL);
2485 if (!work)
2486 return -ENOMEM;
2487
2488 work->id = id_priv;
2489 INIT_WORK(&work->work, cma_work_handler);
2490 work->old_state = RDMA_CM_ROUTE_QUERY;
2491 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2492 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2493 queue_work(cma_wq, &work->work);
2494 return 0;
2495 }
2496
2497 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2498 {
2499 int prio;
2500 struct net_device *dev;
2501
2502 prio = rt_tos2priority(tos);
2503 dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2504 if (dev->num_tc)
2505 return netdev_get_prio_tc_map(dev, prio);
2506
2507 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2508 if (is_vlan_dev(ndev))
2509 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2510 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2511 #endif
2512 return 0;
2513 }
2514
2515 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2516 unsigned long supported_gids,
2517 enum ib_gid_type default_gid)
2518 {
2519 if ((network_type == RDMA_NETWORK_IPV4 ||
2520 network_type == RDMA_NETWORK_IPV6) &&
2521 test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2522 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2523
2524 return default_gid;
2525 }
2526
2527 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2528 {
2529 struct rdma_route *route = &id_priv->id.route;
2530 struct rdma_addr *addr = &route->addr;
2531 struct cma_work *work;
2532 int ret;
2533 struct net_device *ndev = NULL;
2534 u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2535 rdma_start_port(id_priv->cma_dev->device)];
2536 u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2537
2538
2539 work = kzalloc(sizeof *work, GFP_KERNEL);
2540 if (!work)
2541 return -ENOMEM;
2542
2543 work->id = id_priv;
2544 INIT_WORK(&work->work, cma_work_handler);
2545
2546 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2547 if (!route->path_rec) {
2548 ret = -ENOMEM;
2549 goto err1;
2550 }
2551
2552 route->num_paths = 1;
2553
2554 if (addr->dev_addr.bound_dev_if) {
2555 unsigned long supported_gids;
2556
2557 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2558 if (!ndev) {
2559 ret = -ENODEV;
2560 goto err2;
2561 }
2562
2563 if (ndev->flags & IFF_LOOPBACK) {
2564 dev_put(ndev);
2565 if (!id_priv->id.device->get_netdev) {
2566 ret = -EOPNOTSUPP;
2567 goto err2;
2568 }
2569
2570 ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2571 id_priv->id.port_num);
2572 if (!ndev) {
2573 ret = -ENODEV;
2574 goto err2;
2575 }
2576 }
2577
2578 route->path_rec->net = &init_net;
2579 route->path_rec->ifindex = ndev->ifindex;
2580 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2581 id_priv->id.port_num);
2582 route->path_rec->gid_type =
2583 cma_route_gid_type(addr->dev_addr.network,
2584 supported_gids,
2585 id_priv->gid_type);
2586 }
2587 if (!ndev) {
2588 ret = -ENODEV;
2589 goto err2;
2590 }
2591
2592 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2593
2594 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2595 &route->path_rec->sgid);
2596 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2597 &route->path_rec->dgid);
2598
2599 /* Use the hint from IP Stack to select GID Type */
2600 if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2601 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2602 if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2603 /* TODO: get the hoplimit from the inet/inet6 device */
2604 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2605 else
2606 route->path_rec->hop_limit = 1;
2607 route->path_rec->reversible = 1;
2608 route->path_rec->pkey = cpu_to_be16(0xffff);
2609 route->path_rec->mtu_selector = IB_SA_EQ;
2610 route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2611 route->path_rec->traffic_class = tos;
2612 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2613 route->path_rec->rate_selector = IB_SA_EQ;
2614 route->path_rec->rate = iboe_get_rate(ndev);
2615 dev_put(ndev);
2616 route->path_rec->packet_life_time_selector = IB_SA_EQ;
2617 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2618 if (!route->path_rec->mtu) {
2619 ret = -EINVAL;
2620 goto err2;
2621 }
2622
2623 work->old_state = RDMA_CM_ROUTE_QUERY;
2624 work->new_state = RDMA_CM_ROUTE_RESOLVED;
2625 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2626 work->event.status = 0;
2627
2628 queue_work(cma_wq, &work->work);
2629
2630 return 0;
2631
2632 err2:
2633 kfree(route->path_rec);
2634 route->path_rec = NULL;
2635 err1:
2636 kfree(work);
2637 return ret;
2638 }
2639
2640 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2641 {
2642 struct rdma_id_private *id_priv;
2643 int ret;
2644
2645 id_priv = container_of(id, struct rdma_id_private, id);
2646 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2647 return -EINVAL;
2648
2649 atomic_inc(&id_priv->refcount);
2650 if (rdma_cap_ib_sa(id->device, id->port_num))
2651 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2652 else if (rdma_protocol_roce(id->device, id->port_num))
2653 ret = cma_resolve_iboe_route(id_priv);
2654 else if (rdma_protocol_iwarp(id->device, id->port_num))
2655 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2656 else
2657 ret = -ENOSYS;
2658
2659 if (ret)
2660 goto err;
2661
2662 return 0;
2663 err:
2664 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2665 cma_deref_id(id_priv);
2666 return ret;
2667 }
2668 EXPORT_SYMBOL(rdma_resolve_route);
2669
2670 static void cma_set_loopback(struct sockaddr *addr)
2671 {
2672 switch (addr->sa_family) {
2673 case AF_INET:
2674 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2675 break;
2676 case AF_INET6:
2677 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2678 0, 0, 0, htonl(1));
2679 break;
2680 default:
2681 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2682 0, 0, 0, htonl(1));
2683 break;
2684 }
2685 }
2686
2687 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2688 {
2689 struct cma_device *cma_dev, *cur_dev;
2690 union ib_gid gid;
2691 enum ib_port_state port_state;
2692 u16 pkey;
2693 int ret;
2694 u8 p;
2695
2696 cma_dev = NULL;
2697 mutex_lock(&lock);
2698 list_for_each_entry(cur_dev, &dev_list, list) {
2699 if (cma_family(id_priv) == AF_IB &&
2700 !rdma_cap_ib_cm(cur_dev->device, 1))
2701 continue;
2702
2703 if (!cma_dev)
2704 cma_dev = cur_dev;
2705
2706 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2707 if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2708 port_state == IB_PORT_ACTIVE) {
2709 cma_dev = cur_dev;
2710 goto port_found;
2711 }
2712 }
2713 }
2714
2715 if (!cma_dev) {
2716 ret = -ENODEV;
2717 goto out;
2718 }
2719
2720 p = 1;
2721
2722 port_found:
2723 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2724 if (ret)
2725 goto out;
2726
2727 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2728 if (ret)
2729 goto out;
2730
2731 id_priv->id.route.addr.dev_addr.dev_type =
2732 (rdma_protocol_ib(cma_dev->device, p)) ?
2733 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2734
2735 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2736 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2737 id_priv->id.port_num = p;
2738 cma_attach_to_dev(id_priv, cma_dev);
2739 cma_set_loopback(cma_src_addr(id_priv));
2740 out:
2741 mutex_unlock(&lock);
2742 return ret;
2743 }
2744
2745 static void addr_handler(int status, struct sockaddr *src_addr,
2746 struct rdma_dev_addr *dev_addr, void *context)
2747 {
2748 struct rdma_id_private *id_priv = context;
2749 struct rdma_cm_event event;
2750
2751 memset(&event, 0, sizeof event);
2752 mutex_lock(&id_priv->handler_mutex);
2753 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2754 RDMA_CM_ADDR_RESOLVED))
2755 goto out;
2756
2757 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2758 if (!status && !id_priv->cma_dev) {
2759 status = cma_acquire_dev(id_priv, NULL);
2760 if (status)
2761 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2762 status);
2763 } else {
2764 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2765 }
2766
2767 if (status) {
2768 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2769 RDMA_CM_ADDR_BOUND))
2770 goto out;
2771 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2772 event.status = status;
2773 } else
2774 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2775
2776 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2777 cma_exch(id_priv, RDMA_CM_DESTROYING);
2778 mutex_unlock(&id_priv->handler_mutex);
2779 cma_deref_id(id_priv);
2780 rdma_destroy_id(&id_priv->id);
2781 return;
2782 }
2783 out:
2784 mutex_unlock(&id_priv->handler_mutex);
2785 cma_deref_id(id_priv);
2786 }
2787
2788 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2789 {
2790 struct cma_work *work;
2791 union ib_gid gid;
2792 int ret;
2793
2794 work = kzalloc(sizeof *work, GFP_KERNEL);
2795 if (!work)
2796 return -ENOMEM;
2797
2798 if (!id_priv->cma_dev) {
2799 ret = cma_bind_loopback(id_priv);
2800 if (ret)
2801 goto err;
2802 }
2803
2804 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2805 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2806
2807 work->id = id_priv;
2808 INIT_WORK(&work->work, cma_work_handler);
2809 work->old_state = RDMA_CM_ADDR_QUERY;
2810 work->new_state = RDMA_CM_ADDR_RESOLVED;
2811 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2812 queue_work(cma_wq, &work->work);
2813 return 0;
2814 err:
2815 kfree(work);
2816 return ret;
2817 }
2818
2819 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2820 {
2821 struct cma_work *work;
2822 int ret;
2823
2824 work = kzalloc(sizeof *work, GFP_KERNEL);
2825 if (!work)
2826 return -ENOMEM;
2827
2828 if (!id_priv->cma_dev) {
2829 ret = cma_resolve_ib_dev(id_priv);
2830 if (ret)
2831 goto err;
2832 }
2833
2834 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2835 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2836
2837 work->id = id_priv;
2838 INIT_WORK(&work->work, cma_work_handler);
2839 work->old_state = RDMA_CM_ADDR_QUERY;
2840 work->new_state = RDMA_CM_ADDR_RESOLVED;
2841 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2842 queue_work(cma_wq, &work->work);
2843 return 0;
2844 err:
2845 kfree(work);
2846 return ret;
2847 }
2848
2849 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2850 struct sockaddr *dst_addr)
2851 {
2852 if (!src_addr || !src_addr->sa_family) {
2853 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2854 src_addr->sa_family = dst_addr->sa_family;
2855 if (IS_ENABLED(CONFIG_IPV6) &&
2856 dst_addr->sa_family == AF_INET6) {
2857 struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2858 struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2859 src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2860 if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2861 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2862 } else if (dst_addr->sa_family == AF_IB) {
2863 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2864 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2865 }
2866 }
2867 return rdma_bind_addr(id, src_addr);
2868 }
2869
2870 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2871 struct sockaddr *dst_addr, int timeout_ms)
2872 {
2873 struct rdma_id_private *id_priv;
2874 int ret;
2875
2876 id_priv = container_of(id, struct rdma_id_private, id);
2877 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2878 if (id_priv->state == RDMA_CM_IDLE) {
2879 ret = cma_bind_addr(id, src_addr, dst_addr);
2880 if (ret) {
2881 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2882 return ret;
2883 }
2884 }
2885
2886 if (cma_family(id_priv) != dst_addr->sa_family) {
2887 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2888 return -EINVAL;
2889 }
2890
2891 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2892 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2893 return -EINVAL;
2894 }
2895
2896 atomic_inc(&id_priv->refcount);
2897 if (cma_any_addr(dst_addr)) {
2898 ret = cma_resolve_loopback(id_priv);
2899 } else {
2900 if (dst_addr->sa_family == AF_IB) {
2901 ret = cma_resolve_ib_addr(id_priv);
2902 } else {
2903 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2904 dst_addr, &id->route.addr.dev_addr,
2905 timeout_ms, addr_handler, id_priv);
2906 }
2907 }
2908 if (ret)
2909 goto err;
2910
2911 return 0;
2912 err:
2913 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2914 cma_deref_id(id_priv);
2915 return ret;
2916 }
2917 EXPORT_SYMBOL(rdma_resolve_addr);
2918
2919 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2920 {
2921 struct rdma_id_private *id_priv;
2922 unsigned long flags;
2923 int ret;
2924
2925 id_priv = container_of(id, struct rdma_id_private, id);
2926 spin_lock_irqsave(&id_priv->lock, flags);
2927 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2928 id_priv->reuseaddr = reuse;
2929 ret = 0;
2930 } else {
2931 ret = -EINVAL;
2932 }
2933 spin_unlock_irqrestore(&id_priv->lock, flags);
2934 return ret;
2935 }
2936 EXPORT_SYMBOL(rdma_set_reuseaddr);
2937
2938 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2939 {
2940 struct rdma_id_private *id_priv;
2941 unsigned long flags;
2942 int ret;
2943
2944 id_priv = container_of(id, struct rdma_id_private, id);
2945 spin_lock_irqsave(&id_priv->lock, flags);
2946 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2947 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2948 id_priv->afonly = afonly;
2949 ret = 0;
2950 } else {
2951 ret = -EINVAL;
2952 }
2953 spin_unlock_irqrestore(&id_priv->lock, flags);
2954 return ret;
2955 }
2956 EXPORT_SYMBOL(rdma_set_afonly);
2957
2958 static void cma_bind_port(struct rdma_bind_list *bind_list,
2959 struct rdma_id_private *id_priv)
2960 {
2961 struct sockaddr *addr;
2962 struct sockaddr_ib *sib;
2963 u64 sid, mask;
2964 __be16 port;
2965
2966 addr = cma_src_addr(id_priv);
2967 port = htons(bind_list->port);
2968
2969 switch (addr->sa_family) {
2970 case AF_INET:
2971 ((struct sockaddr_in *) addr)->sin_port = port;
2972 break;
2973 case AF_INET6:
2974 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2975 break;
2976 case AF_IB:
2977 sib = (struct sockaddr_ib *) addr;
2978 sid = be64_to_cpu(sib->sib_sid);
2979 mask = be64_to_cpu(sib->sib_sid_mask);
2980 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2981 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2982 break;
2983 }
2984 id_priv->bind_list = bind_list;
2985 hlist_add_head(&id_priv->node, &bind_list->owners);
2986 }
2987
2988 static int cma_alloc_port(enum rdma_port_space ps,
2989 struct rdma_id_private *id_priv, unsigned short snum)
2990 {
2991 struct rdma_bind_list *bind_list;
2992 int ret;
2993
2994 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2995 if (!bind_list)
2996 return -ENOMEM;
2997
2998 ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2999 snum);
3000 if (ret < 0)
3001 goto err;
3002
3003 bind_list->ps = ps;
3004 bind_list->port = (unsigned short)ret;
3005 cma_bind_port(bind_list, id_priv);
3006 return 0;
3007 err:
3008 kfree(bind_list);
3009 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3010 }
3011
3012 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3013 struct rdma_id_private *id_priv)
3014 {
3015 struct rdma_id_private *cur_id;
3016 struct sockaddr *daddr = cma_dst_addr(id_priv);
3017 struct sockaddr *saddr = cma_src_addr(id_priv);
3018 __be16 dport = cma_port(daddr);
3019
3020 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3021 struct sockaddr *cur_daddr = cma_dst_addr(cur_id);
3022 struct sockaddr *cur_saddr = cma_src_addr(cur_id);
3023 __be16 cur_dport = cma_port(cur_daddr);
3024
3025 if (id_priv == cur_id)
3026 continue;
3027
3028 /* different dest port -> unique */
3029 if (!cma_any_port(cur_daddr) &&
3030 (dport != cur_dport))
3031 continue;
3032
3033 /* different src address -> unique */
3034 if (!cma_any_addr(saddr) &&
3035 !cma_any_addr(cur_saddr) &&
3036 cma_addr_cmp(saddr, cur_saddr))
3037 continue;
3038
3039 /* different dst address -> unique */
3040 if (!cma_any_addr(cur_daddr) &&
3041 cma_addr_cmp(daddr, cur_daddr))
3042 continue;
3043
3044 return -EADDRNOTAVAIL;
3045 }
3046 return 0;
3047 }
3048
3049 static int cma_alloc_any_port(enum rdma_port_space ps,
3050 struct rdma_id_private *id_priv)
3051 {
3052 static unsigned int last_used_port;
3053 int low, high, remaining;
3054 unsigned int rover;
3055 struct net *net = id_priv->id.route.addr.dev_addr.net;
3056
3057 inet_get_local_port_range(net, &low, &high);
3058 remaining = (high - low) + 1;
3059 rover = prandom_u32() % remaining + low;
3060 retry:
3061 if (last_used_port != rover) {
3062 struct rdma_bind_list *bind_list;
3063 int ret;
3064
3065 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3066
3067 if (!bind_list) {
3068 ret = cma_alloc_port(ps, id_priv, rover);
3069 } else {
3070 ret = cma_port_is_unique(bind_list, id_priv);
3071 if (!ret)
3072 cma_bind_port(bind_list, id_priv);
3073 }
3074 /*
3075 * Remember previously used port number in order to avoid
3076 * re-using same port immediately after it is closed.
3077 */
3078 if (!ret)
3079 last_used_port = rover;
3080 if (ret != -EADDRNOTAVAIL)
3081 return ret;
3082 }
3083 if (--remaining) {
3084 rover++;
3085 if ((rover < low) || (rover > high))
3086 rover = low;
3087 goto retry;
3088 }
3089 return -EADDRNOTAVAIL;
3090 }
3091
3092 /*
3093 * Check that the requested port is available. This is called when trying to
3094 * bind to a specific port, or when trying to listen on a bound port. In
3095 * the latter case, the provided id_priv may already be on the bind_list, but
3096 * we still need to check that it's okay to start listening.
3097 */
3098 static int cma_check_port(struct rdma_bind_list *bind_list,
3099 struct rdma_id_private *id_priv, uint8_t reuseaddr)
3100 {
3101 struct rdma_id_private *cur_id;
3102 struct sockaddr *addr, *cur_addr;
3103
3104 addr = cma_src_addr(id_priv);
3105 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3106 if (id_priv == cur_id)
3107 continue;
3108
3109 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3110 cur_id->reuseaddr)
3111 continue;
3112
3113 cur_addr = cma_src_addr(cur_id);
3114 if (id_priv->afonly && cur_id->afonly &&
3115 (addr->sa_family != cur_addr->sa_family))
3116 continue;
3117
3118 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3119 return -EADDRNOTAVAIL;
3120
3121 if (!cma_addr_cmp(addr, cur_addr))
3122 return -EADDRINUSE;
3123 }
3124 return 0;
3125 }
3126
3127 static int cma_use_port(enum rdma_port_space ps,
3128 struct rdma_id_private *id_priv)
3129 {
3130 struct rdma_bind_list *bind_list;
3131 unsigned short snum;
3132 int ret;
3133
3134 snum = ntohs(cma_port(cma_src_addr(id_priv)));
3135 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3136 return -EACCES;
3137
3138 bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3139 if (!bind_list) {
3140 ret = cma_alloc_port(ps, id_priv, snum);
3141 } else {
3142 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3143 if (!ret)
3144 cma_bind_port(bind_list, id_priv);
3145 }
3146 return ret;
3147 }
3148
3149 static int cma_bind_listen(struct rdma_id_private *id_priv)
3150 {
3151 struct rdma_bind_list *bind_list = id_priv->bind_list;
3152 int ret = 0;
3153
3154 mutex_lock(&lock);
3155 if (bind_list->owners.first->next)
3156 ret = cma_check_port(bind_list, id_priv, 0);
3157 mutex_unlock(&lock);
3158 return ret;
3159 }
3160
3161 static enum rdma_port_space cma_select_inet_ps(
3162 struct rdma_id_private *id_priv)
3163 {
3164 switch (id_priv->id.ps) {
3165 case RDMA_PS_TCP:
3166 case RDMA_PS_UDP:
3167 case RDMA_PS_IPOIB:
3168 case RDMA_PS_IB:
3169 return id_priv->id.ps;
3170 default:
3171
3172 return 0;
3173 }
3174 }
3175
3176 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3177 {
3178 enum rdma_port_space ps = 0;
3179 struct sockaddr_ib *sib;
3180 u64 sid_ps, mask, sid;
3181
3182 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3183 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3184 sid = be64_to_cpu(sib->sib_sid) & mask;
3185
3186 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3187 sid_ps = RDMA_IB_IP_PS_IB;
3188 ps = RDMA_PS_IB;
3189 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3190 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3191 sid_ps = RDMA_IB_IP_PS_TCP;
3192 ps = RDMA_PS_TCP;
3193 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3194 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3195 sid_ps = RDMA_IB_IP_PS_UDP;
3196 ps = RDMA_PS_UDP;
3197 }
3198
3199 if (ps) {
3200 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3201 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3202 be64_to_cpu(sib->sib_sid_mask));
3203 }
3204 return ps;
3205 }
3206
3207 static int cma_get_port(struct rdma_id_private *id_priv)
3208 {
3209 enum rdma_port_space ps;
3210 int ret;
3211
3212 if (cma_family(id_priv) != AF_IB)
3213 ps = cma_select_inet_ps(id_priv);
3214 else
3215 ps = cma_select_ib_ps(id_priv);
3216 if (!ps)
3217 return -EPROTONOSUPPORT;
3218
3219 mutex_lock(&lock);
3220 if (cma_any_port(cma_src_addr(id_priv)))
3221 ret = cma_alloc_any_port(ps, id_priv);
3222 else
3223 ret = cma_use_port(ps, id_priv);
3224 mutex_unlock(&lock);
3225
3226 return ret;
3227 }
3228
3229 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3230 struct sockaddr *addr)
3231 {
3232 #if IS_ENABLED(CONFIG_IPV6)
3233 struct sockaddr_in6 *sin6;
3234
3235 if (addr->sa_family != AF_INET6)
3236 return 0;
3237
3238 sin6 = (struct sockaddr_in6 *) addr;
3239
3240 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3241 return 0;
3242
3243 if (!sin6->sin6_scope_id)
3244 return -EINVAL;
3245
3246 dev_addr->bound_dev_if = sin6->sin6_scope_id;
3247 #endif
3248 return 0;
3249 }
3250
3251 int rdma_listen(struct rdma_cm_id *id, int backlog)
3252 {
3253 struct rdma_id_private *id_priv;
3254 int ret;
3255
3256 id_priv = container_of(id, struct rdma_id_private, id);
3257 if (id_priv->state == RDMA_CM_IDLE) {
3258 id->route.addr.src_addr.ss_family = AF_INET;
3259 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3260 if (ret)
3261 return ret;
3262 }
3263
3264 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3265 return -EINVAL;
3266
3267 if (id_priv->reuseaddr) {
3268 ret = cma_bind_listen(id_priv);
3269 if (ret)
3270 goto err;
3271 }
3272
3273 id_priv->backlog = backlog;
3274 if (id->device) {
3275 if (rdma_cap_ib_cm(id->device, 1)) {
3276 ret = cma_ib_listen(id_priv);
3277 if (ret)
3278 goto err;
3279 } else if (rdma_cap_iw_cm(id->device, 1)) {
3280 ret = cma_iw_listen(id_priv, backlog);
3281 if (ret)
3282 goto err;
3283 } else {
3284 ret = -ENOSYS;
3285 goto err;
3286 }
3287 } else
3288 cma_listen_on_all(id_priv);
3289
3290 return 0;
3291 err:
3292 id_priv->backlog = 0;
3293 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3294 return ret;
3295 }
3296 EXPORT_SYMBOL(rdma_listen);
3297
3298 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3299 {
3300 struct rdma_id_private *id_priv;
3301 int ret;
3302 struct sockaddr *daddr;
3303
3304 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3305 addr->sa_family != AF_IB)
3306 return -EAFNOSUPPORT;
3307
3308 id_priv = container_of(id, struct rdma_id_private, id);
3309 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3310 return -EINVAL;
3311
3312 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3313 if (ret)
3314 goto err1;
3315
3316 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3317 if (!cma_any_addr(addr)) {
3318 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3319 if (ret)
3320 goto err1;
3321
3322 ret = cma_acquire_dev(id_priv, NULL);
3323 if (ret)
3324 goto err1;
3325 }
3326
3327 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3328 if (addr->sa_family == AF_INET)
3329 id_priv->afonly = 1;
3330 #if IS_ENABLED(CONFIG_IPV6)
3331 else if (addr->sa_family == AF_INET6) {
3332 struct net *net = id_priv->id.route.addr.dev_addr.net;
3333
3334 id_priv->afonly = net->ipv6.sysctl.bindv6only;
3335 }
3336 #endif
3337 }
3338 ret = cma_get_port(id_priv);
3339 if (ret)
3340 goto err2;
3341
3342 daddr = cma_dst_addr(id_priv);
3343 daddr->sa_family = addr->sa_family;
3344
3345 return 0;
3346 err2:
3347 if (id_priv->cma_dev)
3348 cma_release_dev(id_priv);
3349 err1:
3350 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3351 return ret;
3352 }
3353 EXPORT_SYMBOL(rdma_bind_addr);
3354
3355 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3356 {
3357 struct cma_hdr *cma_hdr;
3358
3359 cma_hdr = hdr;
3360 cma_hdr->cma_version = CMA_VERSION;
3361 if (cma_family(id_priv) == AF_INET) {
3362 struct sockaddr_in *src4, *dst4;
3363
3364 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3365 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3366
3367 cma_set_ip_ver(cma_hdr, 4);
3368 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3369 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3370 cma_hdr->port = src4->sin_port;
3371 } else if (cma_family(id_priv) == AF_INET6) {
3372 struct sockaddr_in6 *src6, *dst6;
3373
3374 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3375 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3376
3377 cma_set_ip_ver(cma_hdr, 6);
3378 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3379 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3380 cma_hdr->port = src6->sin6_port;
3381 }
3382 return 0;
3383 }
3384
3385 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3386 struct ib_cm_event *ib_event)
3387 {
3388 struct rdma_id_private *id_priv = cm_id->context;
3389 struct rdma_cm_event event;
3390 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3391 int ret = 0;
3392
3393 mutex_lock(&id_priv->handler_mutex);
3394 if (id_priv->state != RDMA_CM_CONNECT)
3395 goto out;
3396
3397 memset(&event, 0, sizeof event);
3398 switch (ib_event->event) {
3399 case IB_CM_SIDR_REQ_ERROR:
3400 event.event = RDMA_CM_EVENT_UNREACHABLE;
3401 event.status = -ETIMEDOUT;
3402 break;
3403 case IB_CM_SIDR_REP_RECEIVED:
3404 event.param.ud.private_data = ib_event->private_data;
3405 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3406 if (rep->status != IB_SIDR_SUCCESS) {
3407 event.event = RDMA_CM_EVENT_UNREACHABLE;
3408 event.status = ib_event->param.sidr_rep_rcvd.status;
3409 pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3410 event.status);
3411 break;
3412 }
3413 ret = cma_set_qkey(id_priv, rep->qkey);
3414 if (ret) {
3415 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3416 event.event = RDMA_CM_EVENT_ADDR_ERROR;
3417 event.status = ret;
3418 break;
3419 }
3420 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3421 id_priv->id.route.path_rec,
3422 &event.param.ud.ah_attr);
3423 event.param.ud.qp_num = rep->qpn;
3424 event.param.ud.qkey = rep->qkey;
3425 event.event = RDMA_CM_EVENT_ESTABLISHED;
3426 event.status = 0;
3427 break;
3428 default:
3429 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3430 ib_event->event);
3431 goto out;
3432 }
3433
3434 ret = id_priv->id.event_handler(&id_priv->id, &event);
3435 if (ret) {
3436 /* Destroy the CM ID by returning a non-zero value. */
3437 id_priv->cm_id.ib = NULL;
3438 cma_exch(id_priv, RDMA_CM_DESTROYING);
3439 mutex_unlock(&id_priv->handler_mutex);
3440 rdma_destroy_id(&id_priv->id);
3441 return ret;
3442 }
3443 out:
3444 mutex_unlock(&id_priv->handler_mutex);
3445 return ret;
3446 }
3447
3448 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3449 struct rdma_conn_param *conn_param)
3450 {
3451 struct ib_cm_sidr_req_param req;
3452 struct ib_cm_id *id;
3453 void *private_data;
3454 int offset, ret;
3455
3456 memset(&req, 0, sizeof req);
3457 offset = cma_user_data_offset(id_priv);
3458 req.private_data_len = offset + conn_param->private_data_len;
3459 if (req.private_data_len < conn_param->private_data_len)
3460 return -EINVAL;
3461
3462 if (req.private_data_len) {
3463 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3464 if (!private_data)
3465 return -ENOMEM;
3466 } else {
3467 private_data = NULL;
3468 }
3469
3470 if (conn_param->private_data && conn_param->private_data_len)
3471 memcpy(private_data + offset, conn_param->private_data,
3472 conn_param->private_data_len);
3473
3474 if (private_data) {
3475 ret = cma_format_hdr(private_data, id_priv);
3476 if (ret)
3477 goto out;
3478 req.private_data = private_data;
3479 }
3480
3481 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3482 id_priv);
3483 if (IS_ERR(id)) {
3484 ret = PTR_ERR(id);
3485 goto out;
3486 }
3487 id_priv->cm_id.ib = id;
3488
3489 req.path = id_priv->id.route.path_rec;
3490 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3491 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3492 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3493
3494 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3495 if (ret) {
3496 ib_destroy_cm_id(id_priv->cm_id.ib);
3497 id_priv->cm_id.ib = NULL;
3498 }
3499 out:
3500 kfree(private_data);
3501 return ret;
3502 }
3503
3504 static int cma_connect_ib(struct rdma_id_private *id_priv,
3505 struct rdma_conn_param *conn_param)
3506 {
3507 struct ib_cm_req_param req;
3508 struct rdma_route *route;
3509 void *private_data;
3510 struct ib_cm_id *id;
3511 int offset, ret;
3512
3513 memset(&req, 0, sizeof req);
3514 offset = cma_user_data_offset(id_priv);
3515 req.private_data_len = offset + conn_param->private_data_len;
3516 if (req.private_data_len < conn_param->private_data_len)
3517 return -EINVAL;
3518
3519 if (req.private_data_len) {
3520 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3521 if (!private_data)
3522 return -ENOMEM;
3523 } else {
3524 private_data = NULL;
3525 }
3526
3527 if (conn_param->private_data && conn_param->private_data_len)
3528 memcpy(private_data + offset, conn_param->private_data,
3529 conn_param->private_data_len);
3530
3531 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3532 if (IS_ERR(id)) {
3533 ret = PTR_ERR(id);
3534 goto out;
3535 }
3536 id_priv->cm_id.ib = id;
3537
3538 route = &id_priv->id.route;
3539 if (private_data) {
3540 ret = cma_format_hdr(private_data, id_priv);
3541 if (ret)
3542 goto out;
3543 req.private_data = private_data;
3544 }
3545
3546 req.primary_path = &route->path_rec[0];
3547 if (route->num_paths == 2)
3548 req.alternate_path = &route->path_rec[1];
3549
3550 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3551 req.qp_num = id_priv->qp_num;
3552 req.qp_type = id_priv->id.qp_type;
3553 req.starting_psn = id_priv->seq_num;
3554 req.responder_resources = conn_param->responder_resources;
3555 req.initiator_depth = conn_param->initiator_depth;
3556 req.flow_control = conn_param->flow_control;
3557 req.retry_count = min_t(u8, 7, conn_param->retry_count);
3558 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3559 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3560 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3561 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3562 req.srq = id_priv->srq ? 1 : 0;
3563
3564 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3565 out:
3566 if (ret && !IS_ERR(id)) {
3567 ib_destroy_cm_id(id);
3568 id_priv->cm_id.ib = NULL;
3569 }
3570
3571 kfree(private_data);
3572 return ret;
3573 }
3574
3575 static int cma_connect_iw(struct rdma_id_private *id_priv,
3576 struct rdma_conn_param *conn_param)
3577 {
3578 struct iw_cm_id *cm_id;
3579 int ret;
3580 struct iw_cm_conn_param iw_param;
3581
3582 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3583 if (IS_ERR(cm_id))
3584 return PTR_ERR(cm_id);
3585
3586 cm_id->tos = id_priv->tos;
3587 id_priv->cm_id.iw = cm_id;
3588
3589 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3590 rdma_addr_size(cma_src_addr(id_priv)));
3591 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3592 rdma_addr_size(cma_dst_addr(id_priv)));
3593
3594 ret = cma_modify_qp_rtr(id_priv, conn_param);
3595 if (ret)
3596 goto out;
3597
3598 if (conn_param) {
3599 iw_param.ord = conn_param->initiator_depth;
3600 iw_param.ird = conn_param->responder_resources;
3601 iw_param.private_data = conn_param->private_data;
3602 iw_param.private_data_len = conn_param->private_data_len;
3603 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3604 } else {
3605 memset(&iw_param, 0, sizeof iw_param);
3606 iw_param.qpn = id_priv->qp_num;
3607 }
3608 ret = iw_cm_connect(cm_id, &iw_param);
3609 out:
3610 if (ret) {
3611 iw_destroy_cm_id(cm_id);
3612 id_priv->cm_id.iw = NULL;
3613 }
3614 return ret;
3615 }
3616
3617 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
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 (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3624 return -EINVAL;
3625
3626 if (!id->qp) {
3627 id_priv->qp_num = conn_param->qp_num;
3628 id_priv->srq = conn_param->srq;
3629 }
3630
3631 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3632 if (id->qp_type == IB_QPT_UD)
3633 ret = cma_resolve_ib_udp(id_priv, conn_param);
3634 else
3635 ret = cma_connect_ib(id_priv, conn_param);
3636 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3637 ret = cma_connect_iw(id_priv, conn_param);
3638 else
3639 ret = -ENOSYS;
3640 if (ret)
3641 goto err;
3642
3643 return 0;
3644 err:
3645 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3646 return ret;
3647 }
3648 EXPORT_SYMBOL(rdma_connect);
3649
3650 static int cma_accept_ib(struct rdma_id_private *id_priv,
3651 struct rdma_conn_param *conn_param)
3652 {
3653 struct ib_cm_rep_param rep;
3654 int ret;
3655
3656 ret = cma_modify_qp_rtr(id_priv, conn_param);
3657 if (ret)
3658 goto out;
3659
3660 ret = cma_modify_qp_rts(id_priv, conn_param);
3661 if (ret)
3662 goto out;
3663
3664 memset(&rep, 0, sizeof rep);
3665 rep.qp_num = id_priv->qp_num;
3666 rep.starting_psn = id_priv->seq_num;
3667 rep.private_data = conn_param->private_data;
3668 rep.private_data_len = conn_param->private_data_len;
3669 rep.responder_resources = conn_param->responder_resources;
3670 rep.initiator_depth = conn_param->initiator_depth;
3671 rep.failover_accepted = 0;
3672 rep.flow_control = conn_param->flow_control;
3673 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3674 rep.srq = id_priv->srq ? 1 : 0;
3675
3676 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3677 out:
3678 return ret;
3679 }
3680
3681 static int cma_accept_iw(struct rdma_id_private *id_priv,
3682 struct rdma_conn_param *conn_param)
3683 {
3684 struct iw_cm_conn_param iw_param;
3685 int ret;
3686
3687 if (!conn_param)
3688 return -EINVAL;
3689
3690 ret = cma_modify_qp_rtr(id_priv, conn_param);
3691 if (ret)
3692 return ret;
3693
3694 iw_param.ord = conn_param->initiator_depth;
3695 iw_param.ird = conn_param->responder_resources;
3696 iw_param.private_data = conn_param->private_data;
3697 iw_param.private_data_len = conn_param->private_data_len;
3698 if (id_priv->id.qp) {
3699 iw_param.qpn = id_priv->qp_num;
3700 } else
3701 iw_param.qpn = conn_param->qp_num;
3702
3703 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3704 }
3705
3706 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3707 enum ib_cm_sidr_status status, u32 qkey,
3708 const void *private_data, int private_data_len)
3709 {
3710 struct ib_cm_sidr_rep_param rep;
3711 int ret;
3712
3713 memset(&rep, 0, sizeof rep);
3714 rep.status = status;
3715 if (status == IB_SIDR_SUCCESS) {
3716 ret = cma_set_qkey(id_priv, qkey);
3717 if (ret)
3718 return ret;
3719 rep.qp_num = id_priv->qp_num;
3720 rep.qkey = id_priv->qkey;
3721 }
3722 rep.private_data = private_data;
3723 rep.private_data_len = private_data_len;
3724
3725 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3726 }
3727
3728 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3729 {
3730 struct rdma_id_private *id_priv;
3731 int ret;
3732
3733 id_priv = container_of(id, struct rdma_id_private, id);
3734
3735 id_priv->owner = task_pid_nr(current);
3736
3737 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3738 return -EINVAL;
3739
3740 if (!id->qp && conn_param) {
3741 id_priv->qp_num = conn_param->qp_num;
3742 id_priv->srq = conn_param->srq;
3743 }
3744
3745 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3746 if (id->qp_type == IB_QPT_UD) {
3747 if (conn_param)
3748 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3749 conn_param->qkey,
3750 conn_param->private_data,
3751 conn_param->private_data_len);
3752 else
3753 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3754 0, NULL, 0);
3755 } else {
3756 if (conn_param)
3757 ret = cma_accept_ib(id_priv, conn_param);
3758 else
3759 ret = cma_rep_recv(id_priv);
3760 }
3761 } else if (rdma_cap_iw_cm(id->device, id->port_num))
3762 ret = cma_accept_iw(id_priv, conn_param);
3763 else
3764 ret = -ENOSYS;
3765
3766 if (ret)
3767 goto reject;
3768
3769 return 0;
3770 reject:
3771 cma_modify_qp_err(id_priv);
3772 rdma_reject(id, NULL, 0);
3773 return ret;
3774 }
3775 EXPORT_SYMBOL(rdma_accept);
3776
3777 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3778 {
3779 struct rdma_id_private *id_priv;
3780 int ret;
3781
3782 id_priv = container_of(id, struct rdma_id_private, id);
3783 if (!id_priv->cm_id.ib)
3784 return -EINVAL;
3785
3786 switch (id->device->node_type) {
3787 case RDMA_NODE_IB_CA:
3788 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3789 break;
3790 default:
3791 ret = 0;
3792 break;
3793 }
3794 return ret;
3795 }
3796 EXPORT_SYMBOL(rdma_notify);
3797
3798 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3799 u8 private_data_len)
3800 {
3801 struct rdma_id_private *id_priv;
3802 int ret;
3803
3804 id_priv = container_of(id, struct rdma_id_private, id);
3805 if (!id_priv->cm_id.ib)
3806 return -EINVAL;
3807
3808 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3809 if (id->qp_type == IB_QPT_UD)
3810 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3811 private_data, private_data_len);
3812 else
3813 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3814 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3815 0, private_data, private_data_len);
3816 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3817 ret = iw_cm_reject(id_priv->cm_id.iw,
3818 private_data, private_data_len);
3819 } else
3820 ret = -ENOSYS;
3821
3822 return ret;
3823 }
3824 EXPORT_SYMBOL(rdma_reject);
3825
3826 int rdma_disconnect(struct rdma_cm_id *id)
3827 {
3828 struct rdma_id_private *id_priv;
3829 int ret;
3830
3831 id_priv = container_of(id, struct rdma_id_private, id);
3832 if (!id_priv->cm_id.ib)
3833 return -EINVAL;
3834
3835 if (rdma_cap_ib_cm(id->device, id->port_num)) {
3836 ret = cma_modify_qp_err(id_priv);
3837 if (ret)
3838 goto out;
3839 /* Initiate or respond to a disconnect. */
3840 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3841 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3842 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3843 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3844 } else
3845 ret = -EINVAL;
3846
3847 out:
3848 return ret;
3849 }
3850 EXPORT_SYMBOL(rdma_disconnect);
3851
3852 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3853 {
3854 struct rdma_id_private *id_priv;
3855 struct cma_multicast *mc = multicast->context;
3856 struct rdma_cm_event event;
3857 int ret = 0;
3858
3859 id_priv = mc->id_priv;
3860 mutex_lock(&id_priv->handler_mutex);
3861 if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3862 id_priv->state != RDMA_CM_ADDR_RESOLVED)
3863 goto out;
3864
3865 if (!status)
3866 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3867 else
3868 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3869 status);
3870 mutex_lock(&id_priv->qp_mutex);
3871 if (!status && id_priv->id.qp) {
3872 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3873 be16_to_cpu(multicast->rec.mlid));
3874 if (status)
3875 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3876 status);
3877 }
3878 mutex_unlock(&id_priv->qp_mutex);
3879
3880 memset(&event, 0, sizeof event);
3881 event.status = status;
3882 event.param.ud.private_data = mc->context;
3883 if (!status) {
3884 struct rdma_dev_addr *dev_addr =
3885 &id_priv->id.route.addr.dev_addr;
3886 struct net_device *ndev =
3887 dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3888 enum ib_gid_type gid_type =
3889 id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3890 rdma_start_port(id_priv->cma_dev->device)];
3891
3892 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3893 ib_init_ah_from_mcmember(id_priv->id.device,
3894 id_priv->id.port_num, &multicast->rec,
3895 ndev, gid_type,
3896 &event.param.ud.ah_attr);
3897 event.param.ud.qp_num = 0xFFFFFF;
3898 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3899 if (ndev)
3900 dev_put(ndev);
3901 } else
3902 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3903
3904 ret = id_priv->id.event_handler(&id_priv->id, &event);
3905 if (ret) {
3906 cma_exch(id_priv, RDMA_CM_DESTROYING);
3907 mutex_unlock(&id_priv->handler_mutex);
3908 rdma_destroy_id(&id_priv->id);
3909 return 0;
3910 }
3911
3912 out:
3913 mutex_unlock(&id_priv->handler_mutex);
3914 return 0;
3915 }
3916
3917 static void cma_set_mgid(struct rdma_id_private *id_priv,
3918 struct sockaddr *addr, union ib_gid *mgid)
3919 {
3920 unsigned char mc_map[MAX_ADDR_LEN];
3921 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3922 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3923 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3924
3925 if (cma_any_addr(addr)) {
3926 memset(mgid, 0, sizeof *mgid);
3927 } else if ((addr->sa_family == AF_INET6) &&
3928 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3929 0xFF10A01B)) {
3930 /* IPv6 address is an SA assigned MGID. */
3931 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3932 } else if (addr->sa_family == AF_IB) {
3933 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3934 } else if ((addr->sa_family == AF_INET6)) {
3935 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3936 if (id_priv->id.ps == RDMA_PS_UDP)
3937 mc_map[7] = 0x01; /* Use RDMA CM signature */
3938 *mgid = *(union ib_gid *) (mc_map + 4);
3939 } else {
3940 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3941 if (id_priv->id.ps == RDMA_PS_UDP)
3942 mc_map[7] = 0x01; /* Use RDMA CM signature */
3943 *mgid = *(union ib_gid *) (mc_map + 4);
3944 }
3945 }
3946
3947 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3948 struct cma_multicast *mc)
3949 {
3950 struct ib_sa_mcmember_rec rec;
3951 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3952 ib_sa_comp_mask comp_mask;
3953 int ret;
3954
3955 ib_addr_get_mgid(dev_addr, &rec.mgid);
3956 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3957 &rec.mgid, &rec);
3958 if (ret)
3959 return ret;
3960
3961 ret = cma_set_qkey(id_priv, 0);
3962 if (ret)
3963 return ret;
3964
3965 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3966 rec.qkey = cpu_to_be32(id_priv->qkey);
3967 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3968 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3969 rec.join_state = mc->join_state;
3970
3971 if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
3972 (!ib_sa_sendonly_fullmem_support(&sa_client,
3973 id_priv->id.device,
3974 id_priv->id.port_num))) {
3975 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3976 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3977 id_priv->id.device->name, id_priv->id.port_num);
3978 return -EOPNOTSUPP;
3979 }
3980
3981 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3982 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3983 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3984 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3985 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3986
3987 if (id_priv->id.ps == RDMA_PS_IPOIB)
3988 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3989 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3990 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3991 IB_SA_MCMEMBER_REC_MTU |
3992 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3993
3994 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3995 id_priv->id.port_num, &rec,
3996 comp_mask, GFP_KERNEL,
3997 cma_ib_mc_handler, mc);
3998 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3999 }
4000
4001 static void iboe_mcast_work_handler(struct work_struct *work)
4002 {
4003 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4004 struct cma_multicast *mc = mw->mc;
4005 struct ib_sa_multicast *m = mc->multicast.ib;
4006
4007 mc->multicast.ib->context = mc;
4008 cma_ib_mc_handler(0, m);
4009 kref_put(&mc->mcref, release_mc);
4010 kfree(mw);
4011 }
4012
4013 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
4014 {
4015 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4016 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4017
4018 if (cma_any_addr(addr)) {
4019 memset(mgid, 0, sizeof *mgid);
4020 } else if (addr->sa_family == AF_INET6) {
4021 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4022 } else {
4023 mgid->raw[0] = 0xff;
4024 mgid->raw[1] = 0x0e;
4025 mgid->raw[2] = 0;
4026 mgid->raw[3] = 0;
4027 mgid->raw[4] = 0;
4028 mgid->raw[5] = 0;
4029 mgid->raw[6] = 0;
4030 mgid->raw[7] = 0;
4031 mgid->raw[8] = 0;
4032 mgid->raw[9] = 0;
4033 mgid->raw[10] = 0xff;
4034 mgid->raw[11] = 0xff;
4035 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4036 }
4037 }
4038
4039 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4040 struct cma_multicast *mc)
4041 {
4042 struct iboe_mcast_work *work;
4043 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4044 int err = 0;
4045 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4046 struct net_device *ndev = NULL;
4047 enum ib_gid_type gid_type;
4048 bool send_only;
4049
4050 send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4051
4052 if (cma_zero_addr((struct sockaddr *)&mc->addr))
4053 return -EINVAL;
4054
4055 work = kzalloc(sizeof *work, GFP_KERNEL);
4056 if (!work)
4057 return -ENOMEM;
4058
4059 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4060 if (!mc->multicast.ib) {
4061 err = -ENOMEM;
4062 goto out1;
4063 }
4064
4065 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4066
4067 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4068 if (id_priv->id.ps == RDMA_PS_UDP)
4069 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4070
4071 if (dev_addr->bound_dev_if)
4072 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4073 if (!ndev) {
4074 err = -ENODEV;
4075 goto out2;
4076 }
4077 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4078 mc->multicast.ib->rec.hop_limit = 1;
4079 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4080
4081 gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4082 rdma_start_port(id_priv->cma_dev->device)];
4083 if (addr->sa_family == AF_INET) {
4084 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4085 mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4086 if (!send_only) {
4087 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4088 true);
4089 if (!err)
4090 mc->igmp_joined = true;
4091 }
4092 }
4093 } else {
4094 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4095 err = -ENOTSUPP;
4096 }
4097 dev_put(ndev);
4098 if (err || !mc->multicast.ib->rec.mtu) {
4099 if (!err)
4100 err = -EINVAL;
4101 goto out2;
4102 }
4103 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4104 &mc->multicast.ib->rec.port_gid);
4105 work->id = id_priv;
4106 work->mc = mc;
4107 INIT_WORK(&work->work, iboe_mcast_work_handler);
4108 kref_get(&mc->mcref);
4109 queue_work(cma_wq, &work->work);
4110
4111 return 0;
4112
4113 out2:
4114 kfree(mc->multicast.ib);
4115 out1:
4116 kfree(work);
4117 return err;
4118 }
4119
4120 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4121 u8 join_state, void *context)
4122 {
4123 struct rdma_id_private *id_priv;
4124 struct cma_multicast *mc;
4125 int ret;
4126
4127 id_priv = container_of(id, struct rdma_id_private, id);
4128 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4129 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4130 return -EINVAL;
4131
4132 mc = kmalloc(sizeof *mc, GFP_KERNEL);
4133 if (!mc)
4134 return -ENOMEM;
4135
4136 memcpy(&mc->addr, addr, rdma_addr_size(addr));
4137 mc->context = context;
4138 mc->id_priv = id_priv;
4139 mc->igmp_joined = false;
4140 mc->join_state = join_state;
4141 spin_lock(&id_priv->lock);
4142 list_add(&mc->list, &id_priv->mc_list);
4143 spin_unlock(&id_priv->lock);
4144
4145 if (rdma_protocol_roce(id->device, id->port_num)) {
4146 kref_init(&mc->mcref);
4147 ret = cma_iboe_join_multicast(id_priv, mc);
4148 } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4149 ret = cma_join_ib_multicast(id_priv, mc);
4150 else
4151 ret = -ENOSYS;
4152
4153 if (ret) {
4154 spin_lock_irq(&id_priv->lock);
4155 list_del(&mc->list);
4156 spin_unlock_irq(&id_priv->lock);
4157 kfree(mc);
4158 }
4159 return ret;
4160 }
4161 EXPORT_SYMBOL(rdma_join_multicast);
4162
4163 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4164 {
4165 struct rdma_id_private *id_priv;
4166 struct cma_multicast *mc;
4167
4168 id_priv = container_of(id, struct rdma_id_private, id);
4169 spin_lock_irq(&id_priv->lock);
4170 list_for_each_entry(mc, &id_priv->mc_list, list) {
4171 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4172 list_del(&mc->list);
4173 spin_unlock_irq(&id_priv->lock);
4174
4175 if (id->qp)
4176 ib_detach_mcast(id->qp,
4177 &mc->multicast.ib->rec.mgid,
4178 be16_to_cpu(mc->multicast.ib->rec.mlid));
4179
4180 BUG_ON(id_priv->cma_dev->device != id->device);
4181
4182 if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4183 ib_sa_free_multicast(mc->multicast.ib);
4184 kfree(mc);
4185 } else if (rdma_protocol_roce(id->device, id->port_num)) {
4186 if (mc->igmp_joined) {
4187 struct rdma_dev_addr *dev_addr =
4188 &id->route.addr.dev_addr;
4189 struct net_device *ndev = NULL;
4190
4191 if (dev_addr->bound_dev_if)
4192 ndev = dev_get_by_index(&init_net,
4193 dev_addr->bound_dev_if);
4194 if (ndev) {
4195 cma_igmp_send(ndev,
4196 &mc->multicast.ib->rec.mgid,
4197 false);
4198 dev_put(ndev);
4199 }
4200 mc->igmp_joined = false;
4201 }
4202 kref_put(&mc->mcref, release_mc);
4203 }
4204 return;
4205 }
4206 }
4207 spin_unlock_irq(&id_priv->lock);
4208 }
4209 EXPORT_SYMBOL(rdma_leave_multicast);
4210
4211 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4212 {
4213 struct rdma_dev_addr *dev_addr;
4214 struct cma_ndev_work *work;
4215
4216 dev_addr = &id_priv->id.route.addr.dev_addr;
4217
4218 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4219 (net_eq(dev_net(ndev), dev_addr->net)) &&
4220 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4221 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4222 ndev->name, &id_priv->id);
4223 work = kzalloc(sizeof *work, GFP_KERNEL);
4224 if (!work)
4225 return -ENOMEM;
4226
4227 INIT_WORK(&work->work, cma_ndev_work_handler);
4228 work->id = id_priv;
4229 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4230 atomic_inc(&id_priv->refcount);
4231 queue_work(cma_wq, &work->work);
4232 }
4233
4234 return 0;
4235 }
4236
4237 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4238 void *ptr)
4239 {
4240 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4241 struct cma_device *cma_dev;
4242 struct rdma_id_private *id_priv;
4243 int ret = NOTIFY_DONE;
4244
4245 if (event != NETDEV_BONDING_FAILOVER)
4246 return NOTIFY_DONE;
4247
4248 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4249 return NOTIFY_DONE;
4250
4251 mutex_lock(&lock);
4252 list_for_each_entry(cma_dev, &dev_list, list)
4253 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4254 ret = cma_netdev_change(ndev, id_priv);
4255 if (ret)
4256 goto out;
4257 }
4258
4259 out:
4260 mutex_unlock(&lock);
4261 return ret;
4262 }
4263
4264 static struct notifier_block cma_nb = {
4265 .notifier_call = cma_netdev_callback
4266 };
4267
4268 static void cma_add_one(struct ib_device *device)
4269 {
4270 struct cma_device *cma_dev;
4271 struct rdma_id_private *id_priv;
4272 unsigned int i;
4273 unsigned long supported_gids = 0;
4274
4275 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4276 if (!cma_dev)
4277 return;
4278
4279 cma_dev->device = device;
4280 cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4281 sizeof(*cma_dev->default_gid_type),
4282 GFP_KERNEL);
4283 if (!cma_dev->default_gid_type)
4284 goto free_cma_dev;
4285
4286 cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4287 sizeof(*cma_dev->default_roce_tos),
4288 GFP_KERNEL);
4289 if (!cma_dev->default_roce_tos)
4290 goto free_gid_type;
4291
4292 for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4293 supported_gids = roce_gid_type_mask_support(device, i);
4294 WARN_ON(!supported_gids);
4295 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4296 find_first_bit(&supported_gids, BITS_PER_LONG);
4297 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4298 }
4299
4300 init_completion(&cma_dev->comp);
4301 atomic_set(&cma_dev->refcount, 1);
4302 INIT_LIST_HEAD(&cma_dev->id_list);
4303 ib_set_client_data(device, &cma_client, cma_dev);
4304
4305 mutex_lock(&lock);
4306 list_add_tail(&cma_dev->list, &dev_list);
4307 list_for_each_entry(id_priv, &listen_any_list, list)
4308 cma_listen_on_dev(id_priv, cma_dev);
4309 mutex_unlock(&lock);
4310
4311 return;
4312
4313 free_gid_type:
4314 kfree(cma_dev->default_gid_type);
4315
4316 free_cma_dev:
4317 kfree(cma_dev);
4318
4319 return;
4320 }
4321
4322 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4323 {
4324 struct rdma_cm_event event;
4325 enum rdma_cm_state state;
4326 int ret = 0;
4327
4328 /* Record that we want to remove the device */
4329 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4330 if (state == RDMA_CM_DESTROYING)
4331 return 0;
4332
4333 cma_cancel_operation(id_priv, state);
4334 mutex_lock(&id_priv->handler_mutex);
4335
4336 /* Check for destruction from another callback. */
4337 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4338 goto out;
4339
4340 memset(&event, 0, sizeof event);
4341 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4342 ret = id_priv->id.event_handler(&id_priv->id, &event);
4343 out:
4344 mutex_unlock(&id_priv->handler_mutex);
4345 return ret;
4346 }
4347
4348 static void cma_process_remove(struct cma_device *cma_dev)
4349 {
4350 struct rdma_id_private *id_priv;
4351 int ret;
4352
4353 mutex_lock(&lock);
4354 while (!list_empty(&cma_dev->id_list)) {
4355 id_priv = list_entry(cma_dev->id_list.next,
4356 struct rdma_id_private, list);
4357
4358 list_del(&id_priv->listen_list);
4359 list_del_init(&id_priv->list);
4360 atomic_inc(&id_priv->refcount);
4361 mutex_unlock(&lock);
4362
4363 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4364 cma_deref_id(id_priv);
4365 if (ret)
4366 rdma_destroy_id(&id_priv->id);
4367
4368 mutex_lock(&lock);
4369 }
4370 mutex_unlock(&lock);
4371
4372 cma_deref_dev(cma_dev);
4373 wait_for_completion(&cma_dev->comp);
4374 }
4375
4376 static void cma_remove_one(struct ib_device *device, void *client_data)
4377 {
4378 struct cma_device *cma_dev = client_data;
4379
4380 if (!cma_dev)
4381 return;
4382
4383 mutex_lock(&lock);
4384 list_del(&cma_dev->list);
4385 mutex_unlock(&lock);
4386
4387 cma_process_remove(cma_dev);
4388 kfree(cma_dev->default_roce_tos);
4389 kfree(cma_dev->default_gid_type);
4390 kfree(cma_dev);
4391 }
4392
4393 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4394 {
4395 struct nlmsghdr *nlh;
4396 struct rdma_cm_id_stats *id_stats;
4397 struct rdma_id_private *id_priv;
4398 struct rdma_cm_id *id = NULL;
4399 struct cma_device *cma_dev;
4400 int i_dev = 0, i_id = 0;
4401
4402 /*
4403 * We export all of the IDs as a sequence of messages. Each
4404 * ID gets its own netlink message.
4405 */
4406 mutex_lock(&lock);
4407
4408 list_for_each_entry(cma_dev, &dev_list, list) {
4409 if (i_dev < cb->args[0]) {
4410 i_dev++;
4411 continue;
4412 }
4413
4414 i_id = 0;
4415 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4416 if (i_id < cb->args[1]) {
4417 i_id++;
4418 continue;
4419 }
4420
4421 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4422 sizeof *id_stats, RDMA_NL_RDMA_CM,
4423 RDMA_NL_RDMA_CM_ID_STATS,
4424 NLM_F_MULTI);
4425 if (!id_stats)
4426 goto out;
4427
4428 memset(id_stats, 0, sizeof *id_stats);
4429 id = &id_priv->id;
4430 id_stats->node_type = id->route.addr.dev_addr.dev_type;
4431 id_stats->port_num = id->port_num;
4432 id_stats->bound_dev_if =
4433 id->route.addr.dev_addr.bound_dev_if;
4434
4435 if (ibnl_put_attr(skb, nlh,
4436 rdma_addr_size(cma_src_addr(id_priv)),
4437 cma_src_addr(id_priv),
4438 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4439 goto out;
4440 if (ibnl_put_attr(skb, nlh,
4441 rdma_addr_size(cma_src_addr(id_priv)),
4442 cma_dst_addr(id_priv),
4443 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4444 goto out;
4445
4446 id_stats->pid = id_priv->owner;
4447 id_stats->port_space = id->ps;
4448 id_stats->cm_state = id_priv->state;
4449 id_stats->qp_num = id_priv->qp_num;
4450 id_stats->qp_type = id->qp_type;
4451
4452 i_id++;
4453 }
4454
4455 cb->args[1] = 0;
4456 i_dev++;
4457 }
4458
4459 out:
4460 mutex_unlock(&lock);
4461 cb->args[0] = i_dev;
4462 cb->args[1] = i_id;
4463
4464 return skb->len;
4465 }
4466
4467 static const struct ibnl_client_cbs cma_cb_table[] = {
4468 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4469 .module = THIS_MODULE },
4470 };
4471
4472 static int cma_init_net(struct net *net)
4473 {
4474 struct cma_pernet *pernet = cma_pernet(net);
4475
4476 idr_init(&pernet->tcp_ps);
4477 idr_init(&pernet->udp_ps);
4478 idr_init(&pernet->ipoib_ps);
4479 idr_init(&pernet->ib_ps);
4480
4481 return 0;
4482 }
4483
4484 static void cma_exit_net(struct net *net)
4485 {
4486 struct cma_pernet *pernet = cma_pernet(net);
4487
4488 idr_destroy(&pernet->tcp_ps);
4489 idr_destroy(&pernet->udp_ps);
4490 idr_destroy(&pernet->ipoib_ps);
4491 idr_destroy(&pernet->ib_ps);
4492 }
4493
4494 static struct pernet_operations cma_pernet_operations = {
4495 .init = cma_init_net,
4496 .exit = cma_exit_net,
4497 .id = &cma_pernet_id,
4498 .size = sizeof(struct cma_pernet),
4499 };
4500
4501 static int __init cma_init(void)
4502 {
4503 int ret;
4504
4505 cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4506 if (!cma_wq)
4507 return -ENOMEM;
4508
4509 ret = register_pernet_subsys(&cma_pernet_operations);
4510 if (ret)
4511 goto err_wq;
4512
4513 ib_sa_register_client(&sa_client);
4514 rdma_addr_register_client(&addr_client);
4515 register_netdevice_notifier(&cma_nb);
4516
4517 ret = ib_register_client(&cma_client);
4518 if (ret)
4519 goto err;
4520
4521 if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4522 cma_cb_table))
4523 pr_warn("RDMA CMA: failed to add netlink callback\n");
4524 cma_configfs_init();
4525
4526 return 0;
4527
4528 err:
4529 unregister_netdevice_notifier(&cma_nb);
4530 rdma_addr_unregister_client(&addr_client);
4531 ib_sa_unregister_client(&sa_client);
4532 err_wq:
4533 destroy_workqueue(cma_wq);
4534 return ret;
4535 }
4536
4537 static void __exit cma_cleanup(void)
4538 {
4539 cma_configfs_exit();
4540 ibnl_remove_client(RDMA_NL_RDMA_CM);
4541 ib_unregister_client(&cma_client);
4542 unregister_netdevice_notifier(&cma_nb);
4543 rdma_addr_unregister_client(&addr_client);
4544 ib_sa_unregister_client(&sa_client);
4545 unregister_pernet_subsys(&cma_pernet_operations);
4546 destroy_workqueue(cma_wq);
4547 }
4548
4549 module_init(cma_init);
4550 module_exit(cma_cleanup);
Ubuntu Artful kernel mirror