]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/infiniband/core/cma.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: replace macros net_random and net_srandom with direct calls to prandom
[mirror_ubuntu-jammy-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/idr.h>
42 #include <linux/inetdevice.h>
43 #include <linux/slab.h>
44 #include <linux/module.h>
45 #include <net/route.h>
46
47 #include <net/tcp.h>
48 #include <net/ipv6.h>
49
50 #include <rdma/rdma_cm.h>
51 #include <rdma/rdma_cm_ib.h>
52 #include <rdma/rdma_netlink.h>
53 #include <rdma/ib.h>
54 #include <rdma/ib_cache.h>
55 #include <rdma/ib_cm.h>
56 #include <rdma/ib_sa.h>
57 #include <rdma/iw_cm.h>
58
59 MODULE_AUTHOR("Sean Hefty");
60 MODULE_DESCRIPTION("Generic RDMA CM Agent");
61 MODULE_LICENSE("Dual BSD/GPL");
62
63 #define CMA_CM_RESPONSE_TIMEOUT 20
64 #define CMA_MAX_CM_RETRIES 15
65 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
66 #define CMA_IBOE_PACKET_LIFETIME 18
67
68 static void cma_add_one(struct ib_device *device);
69 static void cma_remove_one(struct ib_device *device);
70
71 static struct ib_client cma_client = {
72 .name = "cma",
73 .add = cma_add_one,
74 .remove = cma_remove_one
75 };
76
77 static struct ib_sa_client sa_client;
78 static struct rdma_addr_client addr_client;
79 static LIST_HEAD(dev_list);
80 static LIST_HEAD(listen_any_list);
81 static DEFINE_MUTEX(lock);
82 static struct workqueue_struct *cma_wq;
83 static DEFINE_IDR(tcp_ps);
84 static DEFINE_IDR(udp_ps);
85 static DEFINE_IDR(ipoib_ps);
86 static DEFINE_IDR(ib_ps);
87
88 struct cma_device {
89 struct list_head list;
90 struct ib_device *device;
91 struct completion comp;
92 atomic_t refcount;
93 struct list_head id_list;
94 };
95
96 struct rdma_bind_list {
97 struct idr *ps;
98 struct hlist_head owners;
99 unsigned short port;
100 };
101
102 enum {
103 CMA_OPTION_AFONLY,
104 };
105
106 /*
107 * Device removal can occur at anytime, so we need extra handling to
108 * serialize notifying the user of device removal with other callbacks.
109 * We do this by disabling removal notification while a callback is in process,
110 * and reporting it after the callback completes.
111 */
112 struct rdma_id_private {
113 struct rdma_cm_id id;
114
115 struct rdma_bind_list *bind_list;
116 struct hlist_node node;
117 struct list_head list; /* listen_any_list or cma_device.list */
118 struct list_head listen_list; /* per device listens */
119 struct cma_device *cma_dev;
120 struct list_head mc_list;
121
122 int internal_id;
123 enum rdma_cm_state state;
124 spinlock_t lock;
125 struct mutex qp_mutex;
126
127 struct completion comp;
128 atomic_t refcount;
129 struct mutex handler_mutex;
130
131 int backlog;
132 int timeout_ms;
133 struct ib_sa_query *query;
134 int query_id;
135 union {
136 struct ib_cm_id *ib;
137 struct iw_cm_id *iw;
138 } cm_id;
139
140 u32 seq_num;
141 u32 qkey;
142 u32 qp_num;
143 pid_t owner;
144 u32 options;
145 u8 srq;
146 u8 tos;
147 u8 reuseaddr;
148 u8 afonly;
149 };
150
151 struct cma_multicast {
152 struct rdma_id_private *id_priv;
153 union {
154 struct ib_sa_multicast *ib;
155 } multicast;
156 struct list_head list;
157 void *context;
158 struct sockaddr_storage addr;
159 struct kref mcref;
160 };
161
162 struct cma_work {
163 struct work_struct work;
164 struct rdma_id_private *id;
165 enum rdma_cm_state old_state;
166 enum rdma_cm_state new_state;
167 struct rdma_cm_event event;
168 };
169
170 struct cma_ndev_work {
171 struct work_struct work;
172 struct rdma_id_private *id;
173 struct rdma_cm_event event;
174 };
175
176 struct iboe_mcast_work {
177 struct work_struct work;
178 struct rdma_id_private *id;
179 struct cma_multicast *mc;
180 };
181
182 union cma_ip_addr {
183 struct in6_addr ip6;
184 struct {
185 __be32 pad[3];
186 __be32 addr;
187 } ip4;
188 };
189
190 struct cma_hdr {
191 u8 cma_version;
192 u8 ip_version; /* IP version: 7:4 */
193 __be16 port;
194 union cma_ip_addr src_addr;
195 union cma_ip_addr dst_addr;
196 };
197
198 #define CMA_VERSION 0x00
199
200 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
201 {
202 unsigned long flags;
203 int ret;
204
205 spin_lock_irqsave(&id_priv->lock, flags);
206 ret = (id_priv->state == comp);
207 spin_unlock_irqrestore(&id_priv->lock, flags);
208 return ret;
209 }
210
211 static int cma_comp_exch(struct rdma_id_private *id_priv,
212 enum rdma_cm_state comp, enum rdma_cm_state exch)
213 {
214 unsigned long flags;
215 int ret;
216
217 spin_lock_irqsave(&id_priv->lock, flags);
218 if ((ret = (id_priv->state == comp)))
219 id_priv->state = exch;
220 spin_unlock_irqrestore(&id_priv->lock, flags);
221 return ret;
222 }
223
224 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
225 enum rdma_cm_state exch)
226 {
227 unsigned long flags;
228 enum rdma_cm_state old;
229
230 spin_lock_irqsave(&id_priv->lock, flags);
231 old = id_priv->state;
232 id_priv->state = exch;
233 spin_unlock_irqrestore(&id_priv->lock, flags);
234 return old;
235 }
236
237 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
238 {
239 return hdr->ip_version >> 4;
240 }
241
242 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
243 {
244 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
245 }
246
247 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
248 struct cma_device *cma_dev)
249 {
250 atomic_inc(&cma_dev->refcount);
251 id_priv->cma_dev = cma_dev;
252 id_priv->id.device = cma_dev->device;
253 id_priv->id.route.addr.dev_addr.transport =
254 rdma_node_get_transport(cma_dev->device->node_type);
255 list_add_tail(&id_priv->list, &cma_dev->id_list);
256 }
257
258 static inline void cma_deref_dev(struct cma_device *cma_dev)
259 {
260 if (atomic_dec_and_test(&cma_dev->refcount))
261 complete(&cma_dev->comp);
262 }
263
264 static inline void release_mc(struct kref *kref)
265 {
266 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
267
268 kfree(mc->multicast.ib);
269 kfree(mc);
270 }
271
272 static void cma_release_dev(struct rdma_id_private *id_priv)
273 {
274 mutex_lock(&lock);
275 list_del(&id_priv->list);
276 cma_deref_dev(id_priv->cma_dev);
277 id_priv->cma_dev = NULL;
278 mutex_unlock(&lock);
279 }
280
281 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
282 {
283 return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
284 }
285
286 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
287 {
288 return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
289 }
290
291 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
292 {
293 return id_priv->id.route.addr.src_addr.ss_family;
294 }
295
296 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
297 {
298 struct ib_sa_mcmember_rec rec;
299 int ret = 0;
300
301 if (id_priv->qkey) {
302 if (qkey && id_priv->qkey != qkey)
303 return -EINVAL;
304 return 0;
305 }
306
307 if (qkey) {
308 id_priv->qkey = qkey;
309 return 0;
310 }
311
312 switch (id_priv->id.ps) {
313 case RDMA_PS_UDP:
314 case RDMA_PS_IB:
315 id_priv->qkey = RDMA_UDP_QKEY;
316 break;
317 case RDMA_PS_IPOIB:
318 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
319 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
320 id_priv->id.port_num, &rec.mgid,
321 &rec);
322 if (!ret)
323 id_priv->qkey = be32_to_cpu(rec.qkey);
324 break;
325 default:
326 break;
327 }
328 return ret;
329 }
330
331 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
332 {
333 dev_addr->dev_type = ARPHRD_INFINIBAND;
334 rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
335 ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
336 }
337
338 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
339 {
340 int ret;
341
342 if (addr->sa_family != AF_IB) {
343 ret = rdma_translate_ip(addr, dev_addr);
344 } else {
345 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
346 ret = 0;
347 }
348
349 return ret;
350 }
351
352 static int cma_acquire_dev(struct rdma_id_private *id_priv,
353 struct rdma_id_private *listen_id_priv)
354 {
355 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
356 struct cma_device *cma_dev;
357 union ib_gid gid, iboe_gid;
358 int ret = -ENODEV;
359 u8 port, found_port;
360 enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
361 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
362
363 if (dev_ll != IB_LINK_LAYER_INFINIBAND &&
364 id_priv->id.ps == RDMA_PS_IPOIB)
365 return -EINVAL;
366
367 mutex_lock(&lock);
368 iboe_addr_get_sgid(dev_addr, &iboe_gid);
369 memcpy(&gid, dev_addr->src_dev_addr +
370 rdma_addr_gid_offset(dev_addr), sizeof gid);
371 if (listen_id_priv &&
372 rdma_port_get_link_layer(listen_id_priv->id.device,
373 listen_id_priv->id.port_num) == dev_ll) {
374 cma_dev = listen_id_priv->cma_dev;
375 port = listen_id_priv->id.port_num;
376 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
377 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
378 ret = ib_find_cached_gid(cma_dev->device, &iboe_gid,
379 &found_port, NULL);
380 else
381 ret = ib_find_cached_gid(cma_dev->device, &gid,
382 &found_port, NULL);
383
384 if (!ret && (port == found_port)) {
385 id_priv->id.port_num = found_port;
386 goto out;
387 }
388 }
389 list_for_each_entry(cma_dev, &dev_list, list) {
390 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
391 if (listen_id_priv &&
392 listen_id_priv->cma_dev == cma_dev &&
393 listen_id_priv->id.port_num == port)
394 continue;
395 if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
396 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
397 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
398 ret = ib_find_cached_gid(cma_dev->device, &iboe_gid, &found_port, NULL);
399 else
400 ret = ib_find_cached_gid(cma_dev->device, &gid, &found_port, NULL);
401
402 if (!ret && (port == found_port)) {
403 id_priv->id.port_num = found_port;
404 goto out;
405 }
406 }
407 }
408 }
409
410 out:
411 if (!ret)
412 cma_attach_to_dev(id_priv, cma_dev);
413
414 mutex_unlock(&lock);
415 return ret;
416 }
417
418 /*
419 * Select the source IB device and address to reach the destination IB address.
420 */
421 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
422 {
423 struct cma_device *cma_dev, *cur_dev;
424 struct sockaddr_ib *addr;
425 union ib_gid gid, sgid, *dgid;
426 u16 pkey, index;
427 u8 p;
428 int i;
429
430 cma_dev = NULL;
431 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
432 dgid = (union ib_gid *) &addr->sib_addr;
433 pkey = ntohs(addr->sib_pkey);
434
435 list_for_each_entry(cur_dev, &dev_list, list) {
436 if (rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
437 continue;
438
439 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
440 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
441 continue;
442
443 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
444 if (!memcmp(&gid, dgid, sizeof(gid))) {
445 cma_dev = cur_dev;
446 sgid = gid;
447 id_priv->id.port_num = p;
448 goto found;
449 }
450
451 if (!cma_dev && (gid.global.subnet_prefix ==
452 dgid->global.subnet_prefix)) {
453 cma_dev = cur_dev;
454 sgid = gid;
455 id_priv->id.port_num = p;
456 }
457 }
458 }
459 }
460
461 if (!cma_dev)
462 return -ENODEV;
463
464 found:
465 cma_attach_to_dev(id_priv, cma_dev);
466 addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
467 memcpy(&addr->sib_addr, &sgid, sizeof sgid);
468 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
469 return 0;
470 }
471
472 static void cma_deref_id(struct rdma_id_private *id_priv)
473 {
474 if (atomic_dec_and_test(&id_priv->refcount))
475 complete(&id_priv->comp);
476 }
477
478 static int cma_disable_callback(struct rdma_id_private *id_priv,
479 enum rdma_cm_state state)
480 {
481 mutex_lock(&id_priv->handler_mutex);
482 if (id_priv->state != state) {
483 mutex_unlock(&id_priv->handler_mutex);
484 return -EINVAL;
485 }
486 return 0;
487 }
488
489 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
490 void *context, enum rdma_port_space ps,
491 enum ib_qp_type qp_type)
492 {
493 struct rdma_id_private *id_priv;
494
495 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
496 if (!id_priv)
497 return ERR_PTR(-ENOMEM);
498
499 id_priv->owner = task_pid_nr(current);
500 id_priv->state = RDMA_CM_IDLE;
501 id_priv->id.context = context;
502 id_priv->id.event_handler = event_handler;
503 id_priv->id.ps = ps;
504 id_priv->id.qp_type = qp_type;
505 spin_lock_init(&id_priv->lock);
506 mutex_init(&id_priv->qp_mutex);
507 init_completion(&id_priv->comp);
508 atomic_set(&id_priv->refcount, 1);
509 mutex_init(&id_priv->handler_mutex);
510 INIT_LIST_HEAD(&id_priv->listen_list);
511 INIT_LIST_HEAD(&id_priv->mc_list);
512 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
513
514 return &id_priv->id;
515 }
516 EXPORT_SYMBOL(rdma_create_id);
517
518 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
519 {
520 struct ib_qp_attr qp_attr;
521 int qp_attr_mask, ret;
522
523 qp_attr.qp_state = IB_QPS_INIT;
524 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
525 if (ret)
526 return ret;
527
528 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
529 if (ret)
530 return ret;
531
532 qp_attr.qp_state = IB_QPS_RTR;
533 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
534 if (ret)
535 return ret;
536
537 qp_attr.qp_state = IB_QPS_RTS;
538 qp_attr.sq_psn = 0;
539 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
540
541 return ret;
542 }
543
544 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
545 {
546 struct ib_qp_attr qp_attr;
547 int qp_attr_mask, ret;
548
549 qp_attr.qp_state = IB_QPS_INIT;
550 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
551 if (ret)
552 return ret;
553
554 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
555 }
556
557 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
558 struct ib_qp_init_attr *qp_init_attr)
559 {
560 struct rdma_id_private *id_priv;
561 struct ib_qp *qp;
562 int ret;
563
564 id_priv = container_of(id, struct rdma_id_private, id);
565 if (id->device != pd->device)
566 return -EINVAL;
567
568 qp = ib_create_qp(pd, qp_init_attr);
569 if (IS_ERR(qp))
570 return PTR_ERR(qp);
571
572 if (id->qp_type == IB_QPT_UD)
573 ret = cma_init_ud_qp(id_priv, qp);
574 else
575 ret = cma_init_conn_qp(id_priv, qp);
576 if (ret)
577 goto err;
578
579 id->qp = qp;
580 id_priv->qp_num = qp->qp_num;
581 id_priv->srq = (qp->srq != NULL);
582 return 0;
583 err:
584 ib_destroy_qp(qp);
585 return ret;
586 }
587 EXPORT_SYMBOL(rdma_create_qp);
588
589 void rdma_destroy_qp(struct rdma_cm_id *id)
590 {
591 struct rdma_id_private *id_priv;
592
593 id_priv = container_of(id, struct rdma_id_private, id);
594 mutex_lock(&id_priv->qp_mutex);
595 ib_destroy_qp(id_priv->id.qp);
596 id_priv->id.qp = NULL;
597 mutex_unlock(&id_priv->qp_mutex);
598 }
599 EXPORT_SYMBOL(rdma_destroy_qp);
600
601 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
602 struct rdma_conn_param *conn_param)
603 {
604 struct ib_qp_attr qp_attr;
605 int qp_attr_mask, ret;
606
607 mutex_lock(&id_priv->qp_mutex);
608 if (!id_priv->id.qp) {
609 ret = 0;
610 goto out;
611 }
612
613 /* Need to update QP attributes from default values. */
614 qp_attr.qp_state = IB_QPS_INIT;
615 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
616 if (ret)
617 goto out;
618
619 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
620 if (ret)
621 goto out;
622
623 qp_attr.qp_state = IB_QPS_RTR;
624 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
625 if (ret)
626 goto out;
627
628 if (conn_param)
629 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
630 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
631 out:
632 mutex_unlock(&id_priv->qp_mutex);
633 return ret;
634 }
635
636 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
637 struct rdma_conn_param *conn_param)
638 {
639 struct ib_qp_attr qp_attr;
640 int qp_attr_mask, ret;
641
642 mutex_lock(&id_priv->qp_mutex);
643 if (!id_priv->id.qp) {
644 ret = 0;
645 goto out;
646 }
647
648 qp_attr.qp_state = IB_QPS_RTS;
649 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
650 if (ret)
651 goto out;
652
653 if (conn_param)
654 qp_attr.max_rd_atomic = conn_param->initiator_depth;
655 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
656 out:
657 mutex_unlock(&id_priv->qp_mutex);
658 return ret;
659 }
660
661 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
662 {
663 struct ib_qp_attr qp_attr;
664 int ret;
665
666 mutex_lock(&id_priv->qp_mutex);
667 if (!id_priv->id.qp) {
668 ret = 0;
669 goto out;
670 }
671
672 qp_attr.qp_state = IB_QPS_ERR;
673 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
674 out:
675 mutex_unlock(&id_priv->qp_mutex);
676 return ret;
677 }
678
679 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
680 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
681 {
682 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
683 int ret;
684 u16 pkey;
685
686 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
687 IB_LINK_LAYER_INFINIBAND)
688 pkey = ib_addr_get_pkey(dev_addr);
689 else
690 pkey = 0xffff;
691
692 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
693 pkey, &qp_attr->pkey_index);
694 if (ret)
695 return ret;
696
697 qp_attr->port_num = id_priv->id.port_num;
698 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
699
700 if (id_priv->id.qp_type == IB_QPT_UD) {
701 ret = cma_set_qkey(id_priv, 0);
702 if (ret)
703 return ret;
704
705 qp_attr->qkey = id_priv->qkey;
706 *qp_attr_mask |= IB_QP_QKEY;
707 } else {
708 qp_attr->qp_access_flags = 0;
709 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
710 }
711 return 0;
712 }
713
714 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
715 int *qp_attr_mask)
716 {
717 struct rdma_id_private *id_priv;
718 int ret = 0;
719
720 id_priv = container_of(id, struct rdma_id_private, id);
721 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
722 case RDMA_TRANSPORT_IB:
723 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
724 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
725 else
726 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
727 qp_attr_mask);
728 if (qp_attr->qp_state == IB_QPS_RTR)
729 qp_attr->rq_psn = id_priv->seq_num;
730 break;
731 case RDMA_TRANSPORT_IWARP:
732 if (!id_priv->cm_id.iw) {
733 qp_attr->qp_access_flags = 0;
734 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
735 } else
736 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
737 qp_attr_mask);
738 break;
739 default:
740 ret = -ENOSYS;
741 break;
742 }
743
744 return ret;
745 }
746 EXPORT_SYMBOL(rdma_init_qp_attr);
747
748 static inline int cma_zero_addr(struct sockaddr *addr)
749 {
750 switch (addr->sa_family) {
751 case AF_INET:
752 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
753 case AF_INET6:
754 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
755 case AF_IB:
756 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
757 default:
758 return 0;
759 }
760 }
761
762 static inline int cma_loopback_addr(struct sockaddr *addr)
763 {
764 switch (addr->sa_family) {
765 case AF_INET:
766 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
767 case AF_INET6:
768 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
769 case AF_IB:
770 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
771 default:
772 return 0;
773 }
774 }
775
776 static inline int cma_any_addr(struct sockaddr *addr)
777 {
778 return cma_zero_addr(addr) || cma_loopback_addr(addr);
779 }
780
781 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
782 {
783 if (src->sa_family != dst->sa_family)
784 return -1;
785
786 switch (src->sa_family) {
787 case AF_INET:
788 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
789 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
790 case AF_INET6:
791 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
792 &((struct sockaddr_in6 *) dst)->sin6_addr);
793 default:
794 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
795 &((struct sockaddr_ib *) dst)->sib_addr);
796 }
797 }
798
799 static __be16 cma_port(struct sockaddr *addr)
800 {
801 struct sockaddr_ib *sib;
802
803 switch (addr->sa_family) {
804 case AF_INET:
805 return ((struct sockaddr_in *) addr)->sin_port;
806 case AF_INET6:
807 return ((struct sockaddr_in6 *) addr)->sin6_port;
808 case AF_IB:
809 sib = (struct sockaddr_ib *) addr;
810 return htons((u16) (be64_to_cpu(sib->sib_sid) &
811 be64_to_cpu(sib->sib_sid_mask)));
812 default:
813 return 0;
814 }
815 }
816
817 static inline int cma_any_port(struct sockaddr *addr)
818 {
819 return !cma_port(addr);
820 }
821
822 static void cma_save_ib_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
823 struct ib_sa_path_rec *path)
824 {
825 struct sockaddr_ib *listen_ib, *ib;
826
827 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
828 ib = (struct sockaddr_ib *) &id->route.addr.src_addr;
829 ib->sib_family = listen_ib->sib_family;
830 ib->sib_pkey = path->pkey;
831 ib->sib_flowinfo = path->flow_label;
832 memcpy(&ib->sib_addr, &path->sgid, 16);
833 ib->sib_sid = listen_ib->sib_sid;
834 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
835 ib->sib_scope_id = listen_ib->sib_scope_id;
836
837 ib = (struct sockaddr_ib *) &id->route.addr.dst_addr;
838 ib->sib_family = listen_ib->sib_family;
839 ib->sib_pkey = path->pkey;
840 ib->sib_flowinfo = path->flow_label;
841 memcpy(&ib->sib_addr, &path->dgid, 16);
842 }
843
844 static void cma_save_ip4_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
845 struct cma_hdr *hdr)
846 {
847 struct sockaddr_in *listen4, *ip4;
848
849 listen4 = (struct sockaddr_in *) &listen_id->route.addr.src_addr;
850 ip4 = (struct sockaddr_in *) &id->route.addr.src_addr;
851 ip4->sin_family = listen4->sin_family;
852 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
853 ip4->sin_port = listen4->sin_port;
854
855 ip4 = (struct sockaddr_in *) &id->route.addr.dst_addr;
856 ip4->sin_family = listen4->sin_family;
857 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
858 ip4->sin_port = hdr->port;
859 }
860
861 static void cma_save_ip6_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
862 struct cma_hdr *hdr)
863 {
864 struct sockaddr_in6 *listen6, *ip6;
865
866 listen6 = (struct sockaddr_in6 *) &listen_id->route.addr.src_addr;
867 ip6 = (struct sockaddr_in6 *) &id->route.addr.src_addr;
868 ip6->sin6_family = listen6->sin6_family;
869 ip6->sin6_addr = hdr->dst_addr.ip6;
870 ip6->sin6_port = listen6->sin6_port;
871
872 ip6 = (struct sockaddr_in6 *) &id->route.addr.dst_addr;
873 ip6->sin6_family = listen6->sin6_family;
874 ip6->sin6_addr = hdr->src_addr.ip6;
875 ip6->sin6_port = hdr->port;
876 }
877
878 static int cma_save_net_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
879 struct ib_cm_event *ib_event)
880 {
881 struct cma_hdr *hdr;
882
883 if ((listen_id->route.addr.src_addr.ss_family == AF_IB) &&
884 (ib_event->event == IB_CM_REQ_RECEIVED)) {
885 cma_save_ib_info(id, listen_id, ib_event->param.req_rcvd.primary_path);
886 return 0;
887 }
888
889 hdr = ib_event->private_data;
890 if (hdr->cma_version != CMA_VERSION)
891 return -EINVAL;
892
893 switch (cma_get_ip_ver(hdr)) {
894 case 4:
895 cma_save_ip4_info(id, listen_id, hdr);
896 break;
897 case 6:
898 cma_save_ip6_info(id, listen_id, hdr);
899 break;
900 default:
901 return -EINVAL;
902 }
903 return 0;
904 }
905
906 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
907 {
908 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
909 }
910
911 static void cma_cancel_route(struct rdma_id_private *id_priv)
912 {
913 switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
914 case IB_LINK_LAYER_INFINIBAND:
915 if (id_priv->query)
916 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
917 break;
918 default:
919 break;
920 }
921 }
922
923 static void cma_cancel_listens(struct rdma_id_private *id_priv)
924 {
925 struct rdma_id_private *dev_id_priv;
926
927 /*
928 * Remove from listen_any_list to prevent added devices from spawning
929 * additional listen requests.
930 */
931 mutex_lock(&lock);
932 list_del(&id_priv->list);
933
934 while (!list_empty(&id_priv->listen_list)) {
935 dev_id_priv = list_entry(id_priv->listen_list.next,
936 struct rdma_id_private, listen_list);
937 /* sync with device removal to avoid duplicate destruction */
938 list_del_init(&dev_id_priv->list);
939 list_del(&dev_id_priv->listen_list);
940 mutex_unlock(&lock);
941
942 rdma_destroy_id(&dev_id_priv->id);
943 mutex_lock(&lock);
944 }
945 mutex_unlock(&lock);
946 }
947
948 static void cma_cancel_operation(struct rdma_id_private *id_priv,
949 enum rdma_cm_state state)
950 {
951 switch (state) {
952 case RDMA_CM_ADDR_QUERY:
953 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
954 break;
955 case RDMA_CM_ROUTE_QUERY:
956 cma_cancel_route(id_priv);
957 break;
958 case RDMA_CM_LISTEN:
959 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
960 cma_cancel_listens(id_priv);
961 break;
962 default:
963 break;
964 }
965 }
966
967 static void cma_release_port(struct rdma_id_private *id_priv)
968 {
969 struct rdma_bind_list *bind_list = id_priv->bind_list;
970
971 if (!bind_list)
972 return;
973
974 mutex_lock(&lock);
975 hlist_del(&id_priv->node);
976 if (hlist_empty(&bind_list->owners)) {
977 idr_remove(bind_list->ps, bind_list->port);
978 kfree(bind_list);
979 }
980 mutex_unlock(&lock);
981 }
982
983 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
984 {
985 struct cma_multicast *mc;
986
987 while (!list_empty(&id_priv->mc_list)) {
988 mc = container_of(id_priv->mc_list.next,
989 struct cma_multicast, list);
990 list_del(&mc->list);
991 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
992 case IB_LINK_LAYER_INFINIBAND:
993 ib_sa_free_multicast(mc->multicast.ib);
994 kfree(mc);
995 break;
996 case IB_LINK_LAYER_ETHERNET:
997 kref_put(&mc->mcref, release_mc);
998 break;
999 default:
1000 break;
1001 }
1002 }
1003 }
1004
1005 void rdma_destroy_id(struct rdma_cm_id *id)
1006 {
1007 struct rdma_id_private *id_priv;
1008 enum rdma_cm_state state;
1009
1010 id_priv = container_of(id, struct rdma_id_private, id);
1011 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1012 cma_cancel_operation(id_priv, state);
1013
1014 /*
1015 * Wait for any active callback to finish. New callbacks will find
1016 * the id_priv state set to destroying and abort.
1017 */
1018 mutex_lock(&id_priv->handler_mutex);
1019 mutex_unlock(&id_priv->handler_mutex);
1020
1021 if (id_priv->cma_dev) {
1022 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
1023 case RDMA_TRANSPORT_IB:
1024 if (id_priv->cm_id.ib)
1025 ib_destroy_cm_id(id_priv->cm_id.ib);
1026 break;
1027 case RDMA_TRANSPORT_IWARP:
1028 if (id_priv->cm_id.iw)
1029 iw_destroy_cm_id(id_priv->cm_id.iw);
1030 break;
1031 default:
1032 break;
1033 }
1034 cma_leave_mc_groups(id_priv);
1035 cma_release_dev(id_priv);
1036 }
1037
1038 cma_release_port(id_priv);
1039 cma_deref_id(id_priv);
1040 wait_for_completion(&id_priv->comp);
1041
1042 if (id_priv->internal_id)
1043 cma_deref_id(id_priv->id.context);
1044
1045 kfree(id_priv->id.route.path_rec);
1046 kfree(id_priv);
1047 }
1048 EXPORT_SYMBOL(rdma_destroy_id);
1049
1050 static int cma_rep_recv(struct rdma_id_private *id_priv)
1051 {
1052 int ret;
1053
1054 ret = cma_modify_qp_rtr(id_priv, NULL);
1055 if (ret)
1056 goto reject;
1057
1058 ret = cma_modify_qp_rts(id_priv, NULL);
1059 if (ret)
1060 goto reject;
1061
1062 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1063 if (ret)
1064 goto reject;
1065
1066 return 0;
1067 reject:
1068 cma_modify_qp_err(id_priv);
1069 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1070 NULL, 0, NULL, 0);
1071 return ret;
1072 }
1073
1074 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1075 struct ib_cm_rep_event_param *rep_data,
1076 void *private_data)
1077 {
1078 event->param.conn.private_data = private_data;
1079 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1080 event->param.conn.responder_resources = rep_data->responder_resources;
1081 event->param.conn.initiator_depth = rep_data->initiator_depth;
1082 event->param.conn.flow_control = rep_data->flow_control;
1083 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1084 event->param.conn.srq = rep_data->srq;
1085 event->param.conn.qp_num = rep_data->remote_qpn;
1086 }
1087
1088 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1089 {
1090 struct rdma_id_private *id_priv = cm_id->context;
1091 struct rdma_cm_event event;
1092 int ret = 0;
1093
1094 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1095 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1096 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1097 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1098 return 0;
1099
1100 memset(&event, 0, sizeof event);
1101 switch (ib_event->event) {
1102 case IB_CM_REQ_ERROR:
1103 case IB_CM_REP_ERROR:
1104 event.event = RDMA_CM_EVENT_UNREACHABLE;
1105 event.status = -ETIMEDOUT;
1106 break;
1107 case IB_CM_REP_RECEIVED:
1108 if (id_priv->id.qp) {
1109 event.status = cma_rep_recv(id_priv);
1110 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1111 RDMA_CM_EVENT_ESTABLISHED;
1112 } else {
1113 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1114 }
1115 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1116 ib_event->private_data);
1117 break;
1118 case IB_CM_RTU_RECEIVED:
1119 case IB_CM_USER_ESTABLISHED:
1120 event.event = RDMA_CM_EVENT_ESTABLISHED;
1121 break;
1122 case IB_CM_DREQ_ERROR:
1123 event.status = -ETIMEDOUT; /* fall through */
1124 case IB_CM_DREQ_RECEIVED:
1125 case IB_CM_DREP_RECEIVED:
1126 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1127 RDMA_CM_DISCONNECT))
1128 goto out;
1129 event.event = RDMA_CM_EVENT_DISCONNECTED;
1130 break;
1131 case IB_CM_TIMEWAIT_EXIT:
1132 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1133 break;
1134 case IB_CM_MRA_RECEIVED:
1135 /* ignore event */
1136 goto out;
1137 case IB_CM_REJ_RECEIVED:
1138 cma_modify_qp_err(id_priv);
1139 event.status = ib_event->param.rej_rcvd.reason;
1140 event.event = RDMA_CM_EVENT_REJECTED;
1141 event.param.conn.private_data = ib_event->private_data;
1142 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1143 break;
1144 default:
1145 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1146 ib_event->event);
1147 goto out;
1148 }
1149
1150 ret = id_priv->id.event_handler(&id_priv->id, &event);
1151 if (ret) {
1152 /* Destroy the CM ID by returning a non-zero value. */
1153 id_priv->cm_id.ib = NULL;
1154 cma_exch(id_priv, RDMA_CM_DESTROYING);
1155 mutex_unlock(&id_priv->handler_mutex);
1156 rdma_destroy_id(&id_priv->id);
1157 return ret;
1158 }
1159 out:
1160 mutex_unlock(&id_priv->handler_mutex);
1161 return ret;
1162 }
1163
1164 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1165 struct ib_cm_event *ib_event)
1166 {
1167 struct rdma_id_private *id_priv;
1168 struct rdma_cm_id *id;
1169 struct rdma_route *rt;
1170 int ret;
1171
1172 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1173 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1174 if (IS_ERR(id))
1175 return NULL;
1176
1177 id_priv = container_of(id, struct rdma_id_private, id);
1178 if (cma_save_net_info(id, listen_id, ib_event))
1179 goto err;
1180
1181 rt = &id->route;
1182 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1183 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1184 GFP_KERNEL);
1185 if (!rt->path_rec)
1186 goto err;
1187
1188 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1189 if (rt->num_paths == 2)
1190 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1191
1192 if (cma_any_addr(cma_src_addr(id_priv))) {
1193 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1194 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1195 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1196 } else {
1197 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1198 if (ret)
1199 goto err;
1200 }
1201 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1202
1203 id_priv->state = RDMA_CM_CONNECT;
1204 return id_priv;
1205
1206 err:
1207 rdma_destroy_id(id);
1208 return NULL;
1209 }
1210
1211 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1212 struct ib_cm_event *ib_event)
1213 {
1214 struct rdma_id_private *id_priv;
1215 struct rdma_cm_id *id;
1216 int ret;
1217
1218 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1219 listen_id->ps, IB_QPT_UD);
1220 if (IS_ERR(id))
1221 return NULL;
1222
1223 id_priv = container_of(id, struct rdma_id_private, id);
1224 if (cma_save_net_info(id, listen_id, ib_event))
1225 goto err;
1226
1227 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1228 ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
1229 if (ret)
1230 goto err;
1231 }
1232
1233 id_priv->state = RDMA_CM_CONNECT;
1234 return id_priv;
1235 err:
1236 rdma_destroy_id(id);
1237 return NULL;
1238 }
1239
1240 static void cma_set_req_event_data(struct rdma_cm_event *event,
1241 struct ib_cm_req_event_param *req_data,
1242 void *private_data, int offset)
1243 {
1244 event->param.conn.private_data = private_data + offset;
1245 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1246 event->param.conn.responder_resources = req_data->responder_resources;
1247 event->param.conn.initiator_depth = req_data->initiator_depth;
1248 event->param.conn.flow_control = req_data->flow_control;
1249 event->param.conn.retry_count = req_data->retry_count;
1250 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1251 event->param.conn.srq = req_data->srq;
1252 event->param.conn.qp_num = req_data->remote_qpn;
1253 }
1254
1255 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1256 {
1257 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1258 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1259 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1260 (id->qp_type == IB_QPT_UD)) ||
1261 (!id->qp_type));
1262 }
1263
1264 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1265 {
1266 struct rdma_id_private *listen_id, *conn_id;
1267 struct rdma_cm_event event;
1268 int offset, ret;
1269
1270 listen_id = cm_id->context;
1271 if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1272 return -EINVAL;
1273
1274 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1275 return -ECONNABORTED;
1276
1277 memset(&event, 0, sizeof event);
1278 offset = cma_user_data_offset(listen_id);
1279 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1280 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1281 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1282 event.param.ud.private_data = ib_event->private_data + offset;
1283 event.param.ud.private_data_len =
1284 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1285 } else {
1286 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1287 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1288 ib_event->private_data, offset);
1289 }
1290 if (!conn_id) {
1291 ret = -ENOMEM;
1292 goto err1;
1293 }
1294
1295 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1296 ret = cma_acquire_dev(conn_id, listen_id);
1297 if (ret)
1298 goto err2;
1299
1300 conn_id->cm_id.ib = cm_id;
1301 cm_id->context = conn_id;
1302 cm_id->cm_handler = cma_ib_handler;
1303
1304 /*
1305 * Protect against the user destroying conn_id from another thread
1306 * until we're done accessing it.
1307 */
1308 atomic_inc(&conn_id->refcount);
1309 ret = conn_id->id.event_handler(&conn_id->id, &event);
1310 if (ret)
1311 goto err3;
1312
1313 /*
1314 * Acquire mutex to prevent user executing rdma_destroy_id()
1315 * while we're accessing the cm_id.
1316 */
1317 mutex_lock(&lock);
1318 if (cma_comp(conn_id, RDMA_CM_CONNECT) && (conn_id->id.qp_type != IB_QPT_UD))
1319 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1320 mutex_unlock(&lock);
1321 mutex_unlock(&conn_id->handler_mutex);
1322 mutex_unlock(&listen_id->handler_mutex);
1323 cma_deref_id(conn_id);
1324 return 0;
1325
1326 err3:
1327 cma_deref_id(conn_id);
1328 /* Destroy the CM ID by returning a non-zero value. */
1329 conn_id->cm_id.ib = NULL;
1330 err2:
1331 cma_exch(conn_id, RDMA_CM_DESTROYING);
1332 mutex_unlock(&conn_id->handler_mutex);
1333 err1:
1334 mutex_unlock(&listen_id->handler_mutex);
1335 if (conn_id)
1336 rdma_destroy_id(&conn_id->id);
1337 return ret;
1338 }
1339
1340 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1341 {
1342 if (addr->sa_family == AF_IB)
1343 return ((struct sockaddr_ib *) addr)->sib_sid;
1344
1345 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1346 }
1347 EXPORT_SYMBOL(rdma_get_service_id);
1348
1349 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1350 struct ib_cm_compare_data *compare)
1351 {
1352 struct cma_hdr *cma_data, *cma_mask;
1353 __be32 ip4_addr;
1354 struct in6_addr ip6_addr;
1355
1356 memset(compare, 0, sizeof *compare);
1357 cma_data = (void *) compare->data;
1358 cma_mask = (void *) compare->mask;
1359
1360 switch (addr->sa_family) {
1361 case AF_INET:
1362 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1363 cma_set_ip_ver(cma_data, 4);
1364 cma_set_ip_ver(cma_mask, 0xF);
1365 if (!cma_any_addr(addr)) {
1366 cma_data->dst_addr.ip4.addr = ip4_addr;
1367 cma_mask->dst_addr.ip4.addr = htonl(~0);
1368 }
1369 break;
1370 case AF_INET6:
1371 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1372 cma_set_ip_ver(cma_data, 6);
1373 cma_set_ip_ver(cma_mask, 0xF);
1374 if (!cma_any_addr(addr)) {
1375 cma_data->dst_addr.ip6 = ip6_addr;
1376 memset(&cma_mask->dst_addr.ip6, 0xFF,
1377 sizeof cma_mask->dst_addr.ip6);
1378 }
1379 break;
1380 default:
1381 break;
1382 }
1383 }
1384
1385 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1386 {
1387 struct rdma_id_private *id_priv = iw_id->context;
1388 struct rdma_cm_event event;
1389 int ret = 0;
1390 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1391 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1392
1393 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1394 return 0;
1395
1396 memset(&event, 0, sizeof event);
1397 switch (iw_event->event) {
1398 case IW_CM_EVENT_CLOSE:
1399 event.event = RDMA_CM_EVENT_DISCONNECTED;
1400 break;
1401 case IW_CM_EVENT_CONNECT_REPLY:
1402 memcpy(cma_src_addr(id_priv), laddr,
1403 rdma_addr_size(laddr));
1404 memcpy(cma_dst_addr(id_priv), raddr,
1405 rdma_addr_size(raddr));
1406 switch (iw_event->status) {
1407 case 0:
1408 event.event = RDMA_CM_EVENT_ESTABLISHED;
1409 event.param.conn.initiator_depth = iw_event->ird;
1410 event.param.conn.responder_resources = iw_event->ord;
1411 break;
1412 case -ECONNRESET:
1413 case -ECONNREFUSED:
1414 event.event = RDMA_CM_EVENT_REJECTED;
1415 break;
1416 case -ETIMEDOUT:
1417 event.event = RDMA_CM_EVENT_UNREACHABLE;
1418 break;
1419 default:
1420 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1421 break;
1422 }
1423 break;
1424 case IW_CM_EVENT_ESTABLISHED:
1425 event.event = RDMA_CM_EVENT_ESTABLISHED;
1426 event.param.conn.initiator_depth = iw_event->ird;
1427 event.param.conn.responder_resources = iw_event->ord;
1428 break;
1429 default:
1430 BUG_ON(1);
1431 }
1432
1433 event.status = iw_event->status;
1434 event.param.conn.private_data = iw_event->private_data;
1435 event.param.conn.private_data_len = iw_event->private_data_len;
1436 ret = id_priv->id.event_handler(&id_priv->id, &event);
1437 if (ret) {
1438 /* Destroy the CM ID by returning a non-zero value. */
1439 id_priv->cm_id.iw = NULL;
1440 cma_exch(id_priv, RDMA_CM_DESTROYING);
1441 mutex_unlock(&id_priv->handler_mutex);
1442 rdma_destroy_id(&id_priv->id);
1443 return ret;
1444 }
1445
1446 mutex_unlock(&id_priv->handler_mutex);
1447 return ret;
1448 }
1449
1450 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1451 struct iw_cm_event *iw_event)
1452 {
1453 struct rdma_cm_id *new_cm_id;
1454 struct rdma_id_private *listen_id, *conn_id;
1455 struct rdma_cm_event event;
1456 int ret;
1457 struct ib_device_attr attr;
1458 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1459 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1460
1461 listen_id = cm_id->context;
1462 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1463 return -ECONNABORTED;
1464
1465 /* Create a new RDMA id for the new IW CM ID */
1466 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1467 listen_id->id.context,
1468 RDMA_PS_TCP, IB_QPT_RC);
1469 if (IS_ERR(new_cm_id)) {
1470 ret = -ENOMEM;
1471 goto out;
1472 }
1473 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1474 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1475 conn_id->state = RDMA_CM_CONNECT;
1476
1477 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
1478 if (ret) {
1479 mutex_unlock(&conn_id->handler_mutex);
1480 rdma_destroy_id(new_cm_id);
1481 goto out;
1482 }
1483
1484 ret = cma_acquire_dev(conn_id, listen_id);
1485 if (ret) {
1486 mutex_unlock(&conn_id->handler_mutex);
1487 rdma_destroy_id(new_cm_id);
1488 goto out;
1489 }
1490
1491 conn_id->cm_id.iw = cm_id;
1492 cm_id->context = conn_id;
1493 cm_id->cm_handler = cma_iw_handler;
1494
1495 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1496 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1497
1498 ret = ib_query_device(conn_id->id.device, &attr);
1499 if (ret) {
1500 mutex_unlock(&conn_id->handler_mutex);
1501 rdma_destroy_id(new_cm_id);
1502 goto out;
1503 }
1504
1505 memset(&event, 0, sizeof event);
1506 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1507 event.param.conn.private_data = iw_event->private_data;
1508 event.param.conn.private_data_len = iw_event->private_data_len;
1509 event.param.conn.initiator_depth = iw_event->ird;
1510 event.param.conn.responder_resources = iw_event->ord;
1511
1512 /*
1513 * Protect against the user destroying conn_id from another thread
1514 * until we're done accessing it.
1515 */
1516 atomic_inc(&conn_id->refcount);
1517 ret = conn_id->id.event_handler(&conn_id->id, &event);
1518 if (ret) {
1519 /* User wants to destroy the CM ID */
1520 conn_id->cm_id.iw = NULL;
1521 cma_exch(conn_id, RDMA_CM_DESTROYING);
1522 mutex_unlock(&conn_id->handler_mutex);
1523 cma_deref_id(conn_id);
1524 rdma_destroy_id(&conn_id->id);
1525 goto out;
1526 }
1527
1528 mutex_unlock(&conn_id->handler_mutex);
1529 cma_deref_id(conn_id);
1530
1531 out:
1532 mutex_unlock(&listen_id->handler_mutex);
1533 return ret;
1534 }
1535
1536 static int cma_ib_listen(struct rdma_id_private *id_priv)
1537 {
1538 struct ib_cm_compare_data compare_data;
1539 struct sockaddr *addr;
1540 struct ib_cm_id *id;
1541 __be64 svc_id;
1542 int ret;
1543
1544 id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1545 if (IS_ERR(id))
1546 return PTR_ERR(id);
1547
1548 id_priv->cm_id.ib = id;
1549
1550 addr = cma_src_addr(id_priv);
1551 svc_id = rdma_get_service_id(&id_priv->id, addr);
1552 if (cma_any_addr(addr) && !id_priv->afonly)
1553 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1554 else {
1555 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1556 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1557 }
1558
1559 if (ret) {
1560 ib_destroy_cm_id(id_priv->cm_id.ib);
1561 id_priv->cm_id.ib = NULL;
1562 }
1563
1564 return ret;
1565 }
1566
1567 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1568 {
1569 int ret;
1570 struct iw_cm_id *id;
1571
1572 id = iw_create_cm_id(id_priv->id.device,
1573 iw_conn_req_handler,
1574 id_priv);
1575 if (IS_ERR(id))
1576 return PTR_ERR(id);
1577
1578 id_priv->cm_id.iw = id;
1579
1580 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
1581 rdma_addr_size(cma_src_addr(id_priv)));
1582
1583 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1584
1585 if (ret) {
1586 iw_destroy_cm_id(id_priv->cm_id.iw);
1587 id_priv->cm_id.iw = NULL;
1588 }
1589
1590 return ret;
1591 }
1592
1593 static int cma_listen_handler(struct rdma_cm_id *id,
1594 struct rdma_cm_event *event)
1595 {
1596 struct rdma_id_private *id_priv = id->context;
1597
1598 id->context = id_priv->id.context;
1599 id->event_handler = id_priv->id.event_handler;
1600 return id_priv->id.event_handler(id, event);
1601 }
1602
1603 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1604 struct cma_device *cma_dev)
1605 {
1606 struct rdma_id_private *dev_id_priv;
1607 struct rdma_cm_id *id;
1608 int ret;
1609
1610 if (cma_family(id_priv) == AF_IB &&
1611 rdma_node_get_transport(cma_dev->device->node_type) != RDMA_TRANSPORT_IB)
1612 return;
1613
1614 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1615 id_priv->id.qp_type);
1616 if (IS_ERR(id))
1617 return;
1618
1619 dev_id_priv = container_of(id, struct rdma_id_private, id);
1620
1621 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
1622 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1623 rdma_addr_size(cma_src_addr(id_priv)));
1624
1625 cma_attach_to_dev(dev_id_priv, cma_dev);
1626 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1627 atomic_inc(&id_priv->refcount);
1628 dev_id_priv->internal_id = 1;
1629 dev_id_priv->afonly = id_priv->afonly;
1630
1631 ret = rdma_listen(id, id_priv->backlog);
1632 if (ret)
1633 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1634 "listening on device %s\n", ret, cma_dev->device->name);
1635 }
1636
1637 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1638 {
1639 struct cma_device *cma_dev;
1640
1641 mutex_lock(&lock);
1642 list_add_tail(&id_priv->list, &listen_any_list);
1643 list_for_each_entry(cma_dev, &dev_list, list)
1644 cma_listen_on_dev(id_priv, cma_dev);
1645 mutex_unlock(&lock);
1646 }
1647
1648 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1649 {
1650 struct rdma_id_private *id_priv;
1651
1652 id_priv = container_of(id, struct rdma_id_private, id);
1653 id_priv->tos = (u8) tos;
1654 }
1655 EXPORT_SYMBOL(rdma_set_service_type);
1656
1657 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1658 void *context)
1659 {
1660 struct cma_work *work = context;
1661 struct rdma_route *route;
1662
1663 route = &work->id->id.route;
1664
1665 if (!status) {
1666 route->num_paths = 1;
1667 *route->path_rec = *path_rec;
1668 } else {
1669 work->old_state = RDMA_CM_ROUTE_QUERY;
1670 work->new_state = RDMA_CM_ADDR_RESOLVED;
1671 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1672 work->event.status = status;
1673 }
1674
1675 queue_work(cma_wq, &work->work);
1676 }
1677
1678 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1679 struct cma_work *work)
1680 {
1681 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1682 struct ib_sa_path_rec path_rec;
1683 ib_sa_comp_mask comp_mask;
1684 struct sockaddr_in6 *sin6;
1685 struct sockaddr_ib *sib;
1686
1687 memset(&path_rec, 0, sizeof path_rec);
1688 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1689 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1690 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
1691 path_rec.numb_path = 1;
1692 path_rec.reversible = 1;
1693 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
1694
1695 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1696 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1697 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1698
1699 switch (cma_family(id_priv)) {
1700 case AF_INET:
1701 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1702 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1703 break;
1704 case AF_INET6:
1705 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
1706 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1707 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1708 break;
1709 case AF_IB:
1710 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
1711 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
1712 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1713 break;
1714 }
1715
1716 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1717 id_priv->id.port_num, &path_rec,
1718 comp_mask, timeout_ms,
1719 GFP_KERNEL, cma_query_handler,
1720 work, &id_priv->query);
1721
1722 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1723 }
1724
1725 static void cma_work_handler(struct work_struct *_work)
1726 {
1727 struct cma_work *work = container_of(_work, struct cma_work, work);
1728 struct rdma_id_private *id_priv = work->id;
1729 int destroy = 0;
1730
1731 mutex_lock(&id_priv->handler_mutex);
1732 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1733 goto out;
1734
1735 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1736 cma_exch(id_priv, RDMA_CM_DESTROYING);
1737 destroy = 1;
1738 }
1739 out:
1740 mutex_unlock(&id_priv->handler_mutex);
1741 cma_deref_id(id_priv);
1742 if (destroy)
1743 rdma_destroy_id(&id_priv->id);
1744 kfree(work);
1745 }
1746
1747 static void cma_ndev_work_handler(struct work_struct *_work)
1748 {
1749 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1750 struct rdma_id_private *id_priv = work->id;
1751 int destroy = 0;
1752
1753 mutex_lock(&id_priv->handler_mutex);
1754 if (id_priv->state == RDMA_CM_DESTROYING ||
1755 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
1756 goto out;
1757
1758 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1759 cma_exch(id_priv, RDMA_CM_DESTROYING);
1760 destroy = 1;
1761 }
1762
1763 out:
1764 mutex_unlock(&id_priv->handler_mutex);
1765 cma_deref_id(id_priv);
1766 if (destroy)
1767 rdma_destroy_id(&id_priv->id);
1768 kfree(work);
1769 }
1770
1771 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1772 {
1773 struct rdma_route *route = &id_priv->id.route;
1774 struct cma_work *work;
1775 int ret;
1776
1777 work = kzalloc(sizeof *work, GFP_KERNEL);
1778 if (!work)
1779 return -ENOMEM;
1780
1781 work->id = id_priv;
1782 INIT_WORK(&work->work, cma_work_handler);
1783 work->old_state = RDMA_CM_ROUTE_QUERY;
1784 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1785 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1786
1787 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1788 if (!route->path_rec) {
1789 ret = -ENOMEM;
1790 goto err1;
1791 }
1792
1793 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1794 if (ret)
1795 goto err2;
1796
1797 return 0;
1798 err2:
1799 kfree(route->path_rec);
1800 route->path_rec = NULL;
1801 err1:
1802 kfree(work);
1803 return ret;
1804 }
1805
1806 int rdma_set_ib_paths(struct rdma_cm_id *id,
1807 struct ib_sa_path_rec *path_rec, int num_paths)
1808 {
1809 struct rdma_id_private *id_priv;
1810 int ret;
1811
1812 id_priv = container_of(id, struct rdma_id_private, id);
1813 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1814 RDMA_CM_ROUTE_RESOLVED))
1815 return -EINVAL;
1816
1817 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1818 GFP_KERNEL);
1819 if (!id->route.path_rec) {
1820 ret = -ENOMEM;
1821 goto err;
1822 }
1823
1824 id->route.num_paths = num_paths;
1825 return 0;
1826 err:
1827 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
1828 return ret;
1829 }
1830 EXPORT_SYMBOL(rdma_set_ib_paths);
1831
1832 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1833 {
1834 struct cma_work *work;
1835
1836 work = kzalloc(sizeof *work, GFP_KERNEL);
1837 if (!work)
1838 return -ENOMEM;
1839
1840 work->id = id_priv;
1841 INIT_WORK(&work->work, cma_work_handler);
1842 work->old_state = RDMA_CM_ROUTE_QUERY;
1843 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1844 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1845 queue_work(cma_wq, &work->work);
1846 return 0;
1847 }
1848
1849 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
1850 {
1851 int prio;
1852 struct net_device *dev;
1853
1854 prio = rt_tos2priority(tos);
1855 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
1856 vlan_dev_real_dev(ndev) : ndev;
1857
1858 if (dev->num_tc)
1859 return netdev_get_prio_tc_map(dev, prio);
1860
1861 #if IS_ENABLED(CONFIG_VLAN_8021Q)
1862 if (ndev->priv_flags & IFF_802_1Q_VLAN)
1863 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
1864 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1865 #endif
1866 return 0;
1867 }
1868
1869 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1870 {
1871 struct rdma_route *route = &id_priv->id.route;
1872 struct rdma_addr *addr = &route->addr;
1873 struct cma_work *work;
1874 int ret;
1875 struct net_device *ndev = NULL;
1876 u16 vid;
1877
1878 work = kzalloc(sizeof *work, GFP_KERNEL);
1879 if (!work)
1880 return -ENOMEM;
1881
1882 work->id = id_priv;
1883 INIT_WORK(&work->work, cma_work_handler);
1884
1885 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1886 if (!route->path_rec) {
1887 ret = -ENOMEM;
1888 goto err1;
1889 }
1890
1891 route->num_paths = 1;
1892
1893 if (addr->dev_addr.bound_dev_if)
1894 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1895 if (!ndev) {
1896 ret = -ENODEV;
1897 goto err2;
1898 }
1899
1900 vid = rdma_vlan_dev_vlan_id(ndev);
1901
1902 iboe_mac_vlan_to_ll(&route->path_rec->sgid, addr->dev_addr.src_dev_addr, vid);
1903 iboe_mac_vlan_to_ll(&route->path_rec->dgid, addr->dev_addr.dst_dev_addr, vid);
1904
1905 route->path_rec->hop_limit = 1;
1906 route->path_rec->reversible = 1;
1907 route->path_rec->pkey = cpu_to_be16(0xffff);
1908 route->path_rec->mtu_selector = IB_SA_EQ;
1909 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
1910 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1911 route->path_rec->rate_selector = IB_SA_EQ;
1912 route->path_rec->rate = iboe_get_rate(ndev);
1913 dev_put(ndev);
1914 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1915 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1916 if (!route->path_rec->mtu) {
1917 ret = -EINVAL;
1918 goto err2;
1919 }
1920
1921 work->old_state = RDMA_CM_ROUTE_QUERY;
1922 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1923 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1924 work->event.status = 0;
1925
1926 queue_work(cma_wq, &work->work);
1927
1928 return 0;
1929
1930 err2:
1931 kfree(route->path_rec);
1932 route->path_rec = NULL;
1933 err1:
1934 kfree(work);
1935 return ret;
1936 }
1937
1938 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1939 {
1940 struct rdma_id_private *id_priv;
1941 int ret;
1942
1943 id_priv = container_of(id, struct rdma_id_private, id);
1944 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
1945 return -EINVAL;
1946
1947 atomic_inc(&id_priv->refcount);
1948 switch (rdma_node_get_transport(id->device->node_type)) {
1949 case RDMA_TRANSPORT_IB:
1950 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1951 case IB_LINK_LAYER_INFINIBAND:
1952 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1953 break;
1954 case IB_LINK_LAYER_ETHERNET:
1955 ret = cma_resolve_iboe_route(id_priv);
1956 break;
1957 default:
1958 ret = -ENOSYS;
1959 }
1960 break;
1961 case RDMA_TRANSPORT_IWARP:
1962 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1963 break;
1964 default:
1965 ret = -ENOSYS;
1966 break;
1967 }
1968 if (ret)
1969 goto err;
1970
1971 return 0;
1972 err:
1973 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
1974 cma_deref_id(id_priv);
1975 return ret;
1976 }
1977 EXPORT_SYMBOL(rdma_resolve_route);
1978
1979 static void cma_set_loopback(struct sockaddr *addr)
1980 {
1981 switch (addr->sa_family) {
1982 case AF_INET:
1983 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1984 break;
1985 case AF_INET6:
1986 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
1987 0, 0, 0, htonl(1));
1988 break;
1989 default:
1990 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
1991 0, 0, 0, htonl(1));
1992 break;
1993 }
1994 }
1995
1996 static int cma_bind_loopback(struct rdma_id_private *id_priv)
1997 {
1998 struct cma_device *cma_dev, *cur_dev;
1999 struct ib_port_attr port_attr;
2000 union ib_gid gid;
2001 u16 pkey;
2002 int ret;
2003 u8 p;
2004
2005 cma_dev = NULL;
2006 mutex_lock(&lock);
2007 list_for_each_entry(cur_dev, &dev_list, list) {
2008 if (cma_family(id_priv) == AF_IB &&
2009 rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
2010 continue;
2011
2012 if (!cma_dev)
2013 cma_dev = cur_dev;
2014
2015 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2016 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2017 port_attr.state == IB_PORT_ACTIVE) {
2018 cma_dev = cur_dev;
2019 goto port_found;
2020 }
2021 }
2022 }
2023
2024 if (!cma_dev) {
2025 ret = -ENODEV;
2026 goto out;
2027 }
2028
2029 p = 1;
2030
2031 port_found:
2032 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2033 if (ret)
2034 goto out;
2035
2036 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2037 if (ret)
2038 goto out;
2039
2040 id_priv->id.route.addr.dev_addr.dev_type =
2041 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
2042 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2043
2044 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2045 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2046 id_priv->id.port_num = p;
2047 cma_attach_to_dev(id_priv, cma_dev);
2048 cma_set_loopback(cma_src_addr(id_priv));
2049 out:
2050 mutex_unlock(&lock);
2051 return ret;
2052 }
2053
2054 static void addr_handler(int status, struct sockaddr *src_addr,
2055 struct rdma_dev_addr *dev_addr, void *context)
2056 {
2057 struct rdma_id_private *id_priv = context;
2058 struct rdma_cm_event event;
2059
2060 memset(&event, 0, sizeof event);
2061 mutex_lock(&id_priv->handler_mutex);
2062 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2063 RDMA_CM_ADDR_RESOLVED))
2064 goto out;
2065
2066 if (!status && !id_priv->cma_dev)
2067 status = cma_acquire_dev(id_priv, NULL);
2068
2069 if (status) {
2070 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2071 RDMA_CM_ADDR_BOUND))
2072 goto out;
2073 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2074 event.status = status;
2075 } else {
2076 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2077 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2078 }
2079
2080 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2081 cma_exch(id_priv, RDMA_CM_DESTROYING);
2082 mutex_unlock(&id_priv->handler_mutex);
2083 cma_deref_id(id_priv);
2084 rdma_destroy_id(&id_priv->id);
2085 return;
2086 }
2087 out:
2088 mutex_unlock(&id_priv->handler_mutex);
2089 cma_deref_id(id_priv);
2090 }
2091
2092 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2093 {
2094 struct cma_work *work;
2095 union ib_gid gid;
2096 int ret;
2097
2098 work = kzalloc(sizeof *work, GFP_KERNEL);
2099 if (!work)
2100 return -ENOMEM;
2101
2102 if (!id_priv->cma_dev) {
2103 ret = cma_bind_loopback(id_priv);
2104 if (ret)
2105 goto err;
2106 }
2107
2108 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2109 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2110
2111 work->id = id_priv;
2112 INIT_WORK(&work->work, cma_work_handler);
2113 work->old_state = RDMA_CM_ADDR_QUERY;
2114 work->new_state = RDMA_CM_ADDR_RESOLVED;
2115 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2116 queue_work(cma_wq, &work->work);
2117 return 0;
2118 err:
2119 kfree(work);
2120 return ret;
2121 }
2122
2123 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2124 {
2125 struct cma_work *work;
2126 int ret;
2127
2128 work = kzalloc(sizeof *work, GFP_KERNEL);
2129 if (!work)
2130 return -ENOMEM;
2131
2132 if (!id_priv->cma_dev) {
2133 ret = cma_resolve_ib_dev(id_priv);
2134 if (ret)
2135 goto err;
2136 }
2137
2138 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2139 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2140
2141 work->id = id_priv;
2142 INIT_WORK(&work->work, cma_work_handler);
2143 work->old_state = RDMA_CM_ADDR_QUERY;
2144 work->new_state = RDMA_CM_ADDR_RESOLVED;
2145 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2146 queue_work(cma_wq, &work->work);
2147 return 0;
2148 err:
2149 kfree(work);
2150 return ret;
2151 }
2152
2153 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2154 struct sockaddr *dst_addr)
2155 {
2156 if (!src_addr || !src_addr->sa_family) {
2157 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2158 src_addr->sa_family = dst_addr->sa_family;
2159 if (dst_addr->sa_family == AF_INET6) {
2160 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2161 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2162 } else if (dst_addr->sa_family == AF_IB) {
2163 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2164 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2165 }
2166 }
2167 return rdma_bind_addr(id, src_addr);
2168 }
2169
2170 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2171 struct sockaddr *dst_addr, int timeout_ms)
2172 {
2173 struct rdma_id_private *id_priv;
2174 int ret;
2175
2176 id_priv = container_of(id, struct rdma_id_private, id);
2177 if (id_priv->state == RDMA_CM_IDLE) {
2178 ret = cma_bind_addr(id, src_addr, dst_addr);
2179 if (ret)
2180 return ret;
2181 }
2182
2183 if (cma_family(id_priv) != dst_addr->sa_family)
2184 return -EINVAL;
2185
2186 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2187 return -EINVAL;
2188
2189 atomic_inc(&id_priv->refcount);
2190 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2191 if (cma_any_addr(dst_addr)) {
2192 ret = cma_resolve_loopback(id_priv);
2193 } else {
2194 if (dst_addr->sa_family == AF_IB) {
2195 ret = cma_resolve_ib_addr(id_priv);
2196 } else {
2197 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2198 dst_addr, &id->route.addr.dev_addr,
2199 timeout_ms, addr_handler, id_priv);
2200 }
2201 }
2202 if (ret)
2203 goto err;
2204
2205 return 0;
2206 err:
2207 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2208 cma_deref_id(id_priv);
2209 return ret;
2210 }
2211 EXPORT_SYMBOL(rdma_resolve_addr);
2212
2213 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2214 {
2215 struct rdma_id_private *id_priv;
2216 unsigned long flags;
2217 int ret;
2218
2219 id_priv = container_of(id, struct rdma_id_private, id);
2220 spin_lock_irqsave(&id_priv->lock, flags);
2221 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2222 id_priv->reuseaddr = reuse;
2223 ret = 0;
2224 } else {
2225 ret = -EINVAL;
2226 }
2227 spin_unlock_irqrestore(&id_priv->lock, flags);
2228 return ret;
2229 }
2230 EXPORT_SYMBOL(rdma_set_reuseaddr);
2231
2232 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2233 {
2234 struct rdma_id_private *id_priv;
2235 unsigned long flags;
2236 int ret;
2237
2238 id_priv = container_of(id, struct rdma_id_private, id);
2239 spin_lock_irqsave(&id_priv->lock, flags);
2240 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2241 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2242 id_priv->afonly = afonly;
2243 ret = 0;
2244 } else {
2245 ret = -EINVAL;
2246 }
2247 spin_unlock_irqrestore(&id_priv->lock, flags);
2248 return ret;
2249 }
2250 EXPORT_SYMBOL(rdma_set_afonly);
2251
2252 static void cma_bind_port(struct rdma_bind_list *bind_list,
2253 struct rdma_id_private *id_priv)
2254 {
2255 struct sockaddr *addr;
2256 struct sockaddr_ib *sib;
2257 u64 sid, mask;
2258 __be16 port;
2259
2260 addr = cma_src_addr(id_priv);
2261 port = htons(bind_list->port);
2262
2263 switch (addr->sa_family) {
2264 case AF_INET:
2265 ((struct sockaddr_in *) addr)->sin_port = port;
2266 break;
2267 case AF_INET6:
2268 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2269 break;
2270 case AF_IB:
2271 sib = (struct sockaddr_ib *) addr;
2272 sid = be64_to_cpu(sib->sib_sid);
2273 mask = be64_to_cpu(sib->sib_sid_mask);
2274 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2275 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2276 break;
2277 }
2278 id_priv->bind_list = bind_list;
2279 hlist_add_head(&id_priv->node, &bind_list->owners);
2280 }
2281
2282 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2283 unsigned short snum)
2284 {
2285 struct rdma_bind_list *bind_list;
2286 int ret;
2287
2288 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2289 if (!bind_list)
2290 return -ENOMEM;
2291
2292 ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2293 if (ret < 0)
2294 goto err;
2295
2296 bind_list->ps = ps;
2297 bind_list->port = (unsigned short)ret;
2298 cma_bind_port(bind_list, id_priv);
2299 return 0;
2300 err:
2301 kfree(bind_list);
2302 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2303 }
2304
2305 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2306 {
2307 static unsigned int last_used_port;
2308 int low, high, remaining;
2309 unsigned int rover;
2310
2311 inet_get_local_port_range(&init_net, &low, &high);
2312 remaining = (high - low) + 1;
2313 rover = prandom_u32() % remaining + low;
2314 retry:
2315 if (last_used_port != rover &&
2316 !idr_find(ps, (unsigned short) rover)) {
2317 int ret = cma_alloc_port(ps, id_priv, rover);
2318 /*
2319 * Remember previously used port number in order to avoid
2320 * re-using same port immediately after it is closed.
2321 */
2322 if (!ret)
2323 last_used_port = rover;
2324 if (ret != -EADDRNOTAVAIL)
2325 return ret;
2326 }
2327 if (--remaining) {
2328 rover++;
2329 if ((rover < low) || (rover > high))
2330 rover = low;
2331 goto retry;
2332 }
2333 return -EADDRNOTAVAIL;
2334 }
2335
2336 /*
2337 * Check that the requested port is available. This is called when trying to
2338 * bind to a specific port, or when trying to listen on a bound port. In
2339 * the latter case, the provided id_priv may already be on the bind_list, but
2340 * we still need to check that it's okay to start listening.
2341 */
2342 static int cma_check_port(struct rdma_bind_list *bind_list,
2343 struct rdma_id_private *id_priv, uint8_t reuseaddr)
2344 {
2345 struct rdma_id_private *cur_id;
2346 struct sockaddr *addr, *cur_addr;
2347
2348 addr = cma_src_addr(id_priv);
2349 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2350 if (id_priv == cur_id)
2351 continue;
2352
2353 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2354 cur_id->reuseaddr)
2355 continue;
2356
2357 cur_addr = cma_src_addr(cur_id);
2358 if (id_priv->afonly && cur_id->afonly &&
2359 (addr->sa_family != cur_addr->sa_family))
2360 continue;
2361
2362 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2363 return -EADDRNOTAVAIL;
2364
2365 if (!cma_addr_cmp(addr, cur_addr))
2366 return -EADDRINUSE;
2367 }
2368 return 0;
2369 }
2370
2371 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2372 {
2373 struct rdma_bind_list *bind_list;
2374 unsigned short snum;
2375 int ret;
2376
2377 snum = ntohs(cma_port(cma_src_addr(id_priv)));
2378 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2379 return -EACCES;
2380
2381 bind_list = idr_find(ps, snum);
2382 if (!bind_list) {
2383 ret = cma_alloc_port(ps, id_priv, snum);
2384 } else {
2385 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2386 if (!ret)
2387 cma_bind_port(bind_list, id_priv);
2388 }
2389 return ret;
2390 }
2391
2392 static int cma_bind_listen(struct rdma_id_private *id_priv)
2393 {
2394 struct rdma_bind_list *bind_list = id_priv->bind_list;
2395 int ret = 0;
2396
2397 mutex_lock(&lock);
2398 if (bind_list->owners.first->next)
2399 ret = cma_check_port(bind_list, id_priv, 0);
2400 mutex_unlock(&lock);
2401 return ret;
2402 }
2403
2404 static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
2405 {
2406 switch (id_priv->id.ps) {
2407 case RDMA_PS_TCP:
2408 return &tcp_ps;
2409 case RDMA_PS_UDP:
2410 return &udp_ps;
2411 case RDMA_PS_IPOIB:
2412 return &ipoib_ps;
2413 case RDMA_PS_IB:
2414 return &ib_ps;
2415 default:
2416 return NULL;
2417 }
2418 }
2419
2420 static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2421 {
2422 struct idr *ps = NULL;
2423 struct sockaddr_ib *sib;
2424 u64 sid_ps, mask, sid;
2425
2426 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2427 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2428 sid = be64_to_cpu(sib->sib_sid) & mask;
2429
2430 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2431 sid_ps = RDMA_IB_IP_PS_IB;
2432 ps = &ib_ps;
2433 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2434 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2435 sid_ps = RDMA_IB_IP_PS_TCP;
2436 ps = &tcp_ps;
2437 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2438 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2439 sid_ps = RDMA_IB_IP_PS_UDP;
2440 ps = &udp_ps;
2441 }
2442
2443 if (ps) {
2444 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2445 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2446 be64_to_cpu(sib->sib_sid_mask));
2447 }
2448 return ps;
2449 }
2450
2451 static int cma_get_port(struct rdma_id_private *id_priv)
2452 {
2453 struct idr *ps;
2454 int ret;
2455
2456 if (cma_family(id_priv) != AF_IB)
2457 ps = cma_select_inet_ps(id_priv);
2458 else
2459 ps = cma_select_ib_ps(id_priv);
2460 if (!ps)
2461 return -EPROTONOSUPPORT;
2462
2463 mutex_lock(&lock);
2464 if (cma_any_port(cma_src_addr(id_priv)))
2465 ret = cma_alloc_any_port(ps, id_priv);
2466 else
2467 ret = cma_use_port(ps, id_priv);
2468 mutex_unlock(&lock);
2469
2470 return ret;
2471 }
2472
2473 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2474 struct sockaddr *addr)
2475 {
2476 #if IS_ENABLED(CONFIG_IPV6)
2477 struct sockaddr_in6 *sin6;
2478
2479 if (addr->sa_family != AF_INET6)
2480 return 0;
2481
2482 sin6 = (struct sockaddr_in6 *) addr;
2483 if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) &&
2484 !sin6->sin6_scope_id)
2485 return -EINVAL;
2486
2487 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2488 #endif
2489 return 0;
2490 }
2491
2492 int rdma_listen(struct rdma_cm_id *id, int backlog)
2493 {
2494 struct rdma_id_private *id_priv;
2495 int ret;
2496
2497 id_priv = container_of(id, struct rdma_id_private, id);
2498 if (id_priv->state == RDMA_CM_IDLE) {
2499 id->route.addr.src_addr.ss_family = AF_INET;
2500 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2501 if (ret)
2502 return ret;
2503 }
2504
2505 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2506 return -EINVAL;
2507
2508 if (id_priv->reuseaddr) {
2509 ret = cma_bind_listen(id_priv);
2510 if (ret)
2511 goto err;
2512 }
2513
2514 id_priv->backlog = backlog;
2515 if (id->device) {
2516 switch (rdma_node_get_transport(id->device->node_type)) {
2517 case RDMA_TRANSPORT_IB:
2518 ret = cma_ib_listen(id_priv);
2519 if (ret)
2520 goto err;
2521 break;
2522 case RDMA_TRANSPORT_IWARP:
2523 ret = cma_iw_listen(id_priv, backlog);
2524 if (ret)
2525 goto err;
2526 break;
2527 default:
2528 ret = -ENOSYS;
2529 goto err;
2530 }
2531 } else
2532 cma_listen_on_all(id_priv);
2533
2534 return 0;
2535 err:
2536 id_priv->backlog = 0;
2537 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2538 return ret;
2539 }
2540 EXPORT_SYMBOL(rdma_listen);
2541
2542 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2543 {
2544 struct rdma_id_private *id_priv;
2545 int ret;
2546
2547 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2548 addr->sa_family != AF_IB)
2549 return -EAFNOSUPPORT;
2550
2551 id_priv = container_of(id, struct rdma_id_private, id);
2552 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2553 return -EINVAL;
2554
2555 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2556 if (ret)
2557 goto err1;
2558
2559 if (!cma_any_addr(addr)) {
2560 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2561 if (ret)
2562 goto err1;
2563
2564 ret = cma_acquire_dev(id_priv, NULL);
2565 if (ret)
2566 goto err1;
2567 }
2568
2569 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2570 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2571 if (addr->sa_family == AF_INET)
2572 id_priv->afonly = 1;
2573 #if IS_ENABLED(CONFIG_IPV6)
2574 else if (addr->sa_family == AF_INET6)
2575 id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2576 #endif
2577 }
2578 ret = cma_get_port(id_priv);
2579 if (ret)
2580 goto err2;
2581
2582 return 0;
2583 err2:
2584 if (id_priv->cma_dev)
2585 cma_release_dev(id_priv);
2586 err1:
2587 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2588 return ret;
2589 }
2590 EXPORT_SYMBOL(rdma_bind_addr);
2591
2592 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2593 {
2594 struct cma_hdr *cma_hdr;
2595
2596 cma_hdr = hdr;
2597 cma_hdr->cma_version = CMA_VERSION;
2598 if (cma_family(id_priv) == AF_INET) {
2599 struct sockaddr_in *src4, *dst4;
2600
2601 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2602 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2603
2604 cma_set_ip_ver(cma_hdr, 4);
2605 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2606 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2607 cma_hdr->port = src4->sin_port;
2608 } else if (cma_family(id_priv) == AF_INET6) {
2609 struct sockaddr_in6 *src6, *dst6;
2610
2611 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2612 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2613
2614 cma_set_ip_ver(cma_hdr, 6);
2615 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2616 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2617 cma_hdr->port = src6->sin6_port;
2618 }
2619 return 0;
2620 }
2621
2622 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2623 struct ib_cm_event *ib_event)
2624 {
2625 struct rdma_id_private *id_priv = cm_id->context;
2626 struct rdma_cm_event event;
2627 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2628 int ret = 0;
2629
2630 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
2631 return 0;
2632
2633 memset(&event, 0, sizeof event);
2634 switch (ib_event->event) {
2635 case IB_CM_SIDR_REQ_ERROR:
2636 event.event = RDMA_CM_EVENT_UNREACHABLE;
2637 event.status = -ETIMEDOUT;
2638 break;
2639 case IB_CM_SIDR_REP_RECEIVED:
2640 event.param.ud.private_data = ib_event->private_data;
2641 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2642 if (rep->status != IB_SIDR_SUCCESS) {
2643 event.event = RDMA_CM_EVENT_UNREACHABLE;
2644 event.status = ib_event->param.sidr_rep_rcvd.status;
2645 break;
2646 }
2647 ret = cma_set_qkey(id_priv, rep->qkey);
2648 if (ret) {
2649 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2650 event.status = ret;
2651 break;
2652 }
2653 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2654 id_priv->id.route.path_rec,
2655 &event.param.ud.ah_attr);
2656 event.param.ud.qp_num = rep->qpn;
2657 event.param.ud.qkey = rep->qkey;
2658 event.event = RDMA_CM_EVENT_ESTABLISHED;
2659 event.status = 0;
2660 break;
2661 default:
2662 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2663 ib_event->event);
2664 goto out;
2665 }
2666
2667 ret = id_priv->id.event_handler(&id_priv->id, &event);
2668 if (ret) {
2669 /* Destroy the CM ID by returning a non-zero value. */
2670 id_priv->cm_id.ib = NULL;
2671 cma_exch(id_priv, RDMA_CM_DESTROYING);
2672 mutex_unlock(&id_priv->handler_mutex);
2673 rdma_destroy_id(&id_priv->id);
2674 return ret;
2675 }
2676 out:
2677 mutex_unlock(&id_priv->handler_mutex);
2678 return ret;
2679 }
2680
2681 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2682 struct rdma_conn_param *conn_param)
2683 {
2684 struct ib_cm_sidr_req_param req;
2685 struct ib_cm_id *id;
2686 void *private_data;
2687 int offset, ret;
2688
2689 memset(&req, 0, sizeof req);
2690 offset = cma_user_data_offset(id_priv);
2691 req.private_data_len = offset + conn_param->private_data_len;
2692 if (req.private_data_len < conn_param->private_data_len)
2693 return -EINVAL;
2694
2695 if (req.private_data_len) {
2696 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2697 if (!private_data)
2698 return -ENOMEM;
2699 } else {
2700 private_data = NULL;
2701 }
2702
2703 if (conn_param->private_data && conn_param->private_data_len)
2704 memcpy(private_data + offset, conn_param->private_data,
2705 conn_param->private_data_len);
2706
2707 if (private_data) {
2708 ret = cma_format_hdr(private_data, id_priv);
2709 if (ret)
2710 goto out;
2711 req.private_data = private_data;
2712 }
2713
2714 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2715 id_priv);
2716 if (IS_ERR(id)) {
2717 ret = PTR_ERR(id);
2718 goto out;
2719 }
2720 id_priv->cm_id.ib = id;
2721
2722 req.path = id_priv->id.route.path_rec;
2723 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2724 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2725 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2726
2727 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2728 if (ret) {
2729 ib_destroy_cm_id(id_priv->cm_id.ib);
2730 id_priv->cm_id.ib = NULL;
2731 }
2732 out:
2733 kfree(private_data);
2734 return ret;
2735 }
2736
2737 static int cma_connect_ib(struct rdma_id_private *id_priv,
2738 struct rdma_conn_param *conn_param)
2739 {
2740 struct ib_cm_req_param req;
2741 struct rdma_route *route;
2742 void *private_data;
2743 struct ib_cm_id *id;
2744 int offset, ret;
2745
2746 memset(&req, 0, sizeof req);
2747 offset = cma_user_data_offset(id_priv);
2748 req.private_data_len = offset + conn_param->private_data_len;
2749 if (req.private_data_len < conn_param->private_data_len)
2750 return -EINVAL;
2751
2752 if (req.private_data_len) {
2753 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2754 if (!private_data)
2755 return -ENOMEM;
2756 } else {
2757 private_data = NULL;
2758 }
2759
2760 if (conn_param->private_data && conn_param->private_data_len)
2761 memcpy(private_data + offset, conn_param->private_data,
2762 conn_param->private_data_len);
2763
2764 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2765 if (IS_ERR(id)) {
2766 ret = PTR_ERR(id);
2767 goto out;
2768 }
2769 id_priv->cm_id.ib = id;
2770
2771 route = &id_priv->id.route;
2772 if (private_data) {
2773 ret = cma_format_hdr(private_data, id_priv);
2774 if (ret)
2775 goto out;
2776 req.private_data = private_data;
2777 }
2778
2779 req.primary_path = &route->path_rec[0];
2780 if (route->num_paths == 2)
2781 req.alternate_path = &route->path_rec[1];
2782
2783 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2784 req.qp_num = id_priv->qp_num;
2785 req.qp_type = id_priv->id.qp_type;
2786 req.starting_psn = id_priv->seq_num;
2787 req.responder_resources = conn_param->responder_resources;
2788 req.initiator_depth = conn_param->initiator_depth;
2789 req.flow_control = conn_param->flow_control;
2790 req.retry_count = min_t(u8, 7, conn_param->retry_count);
2791 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2792 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2793 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2794 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2795 req.srq = id_priv->srq ? 1 : 0;
2796
2797 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2798 out:
2799 if (ret && !IS_ERR(id)) {
2800 ib_destroy_cm_id(id);
2801 id_priv->cm_id.ib = NULL;
2802 }
2803
2804 kfree(private_data);
2805 return ret;
2806 }
2807
2808 static int cma_connect_iw(struct rdma_id_private *id_priv,
2809 struct rdma_conn_param *conn_param)
2810 {
2811 struct iw_cm_id *cm_id;
2812 int ret;
2813 struct iw_cm_conn_param iw_param;
2814
2815 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2816 if (IS_ERR(cm_id))
2817 return PTR_ERR(cm_id);
2818
2819 id_priv->cm_id.iw = cm_id;
2820
2821 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
2822 rdma_addr_size(cma_src_addr(id_priv)));
2823 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
2824 rdma_addr_size(cma_dst_addr(id_priv)));
2825
2826 ret = cma_modify_qp_rtr(id_priv, conn_param);
2827 if (ret)
2828 goto out;
2829
2830 if (conn_param) {
2831 iw_param.ord = conn_param->initiator_depth;
2832 iw_param.ird = conn_param->responder_resources;
2833 iw_param.private_data = conn_param->private_data;
2834 iw_param.private_data_len = conn_param->private_data_len;
2835 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2836 } else {
2837 memset(&iw_param, 0, sizeof iw_param);
2838 iw_param.qpn = id_priv->qp_num;
2839 }
2840 ret = iw_cm_connect(cm_id, &iw_param);
2841 out:
2842 if (ret) {
2843 iw_destroy_cm_id(cm_id);
2844 id_priv->cm_id.iw = NULL;
2845 }
2846 return ret;
2847 }
2848
2849 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2850 {
2851 struct rdma_id_private *id_priv;
2852 int ret;
2853
2854 id_priv = container_of(id, struct rdma_id_private, id);
2855 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
2856 return -EINVAL;
2857
2858 if (!id->qp) {
2859 id_priv->qp_num = conn_param->qp_num;
2860 id_priv->srq = conn_param->srq;
2861 }
2862
2863 switch (rdma_node_get_transport(id->device->node_type)) {
2864 case RDMA_TRANSPORT_IB:
2865 if (id->qp_type == IB_QPT_UD)
2866 ret = cma_resolve_ib_udp(id_priv, conn_param);
2867 else
2868 ret = cma_connect_ib(id_priv, conn_param);
2869 break;
2870 case RDMA_TRANSPORT_IWARP:
2871 ret = cma_connect_iw(id_priv, conn_param);
2872 break;
2873 default:
2874 ret = -ENOSYS;
2875 break;
2876 }
2877 if (ret)
2878 goto err;
2879
2880 return 0;
2881 err:
2882 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
2883 return ret;
2884 }
2885 EXPORT_SYMBOL(rdma_connect);
2886
2887 static int cma_accept_ib(struct rdma_id_private *id_priv,
2888 struct rdma_conn_param *conn_param)
2889 {
2890 struct ib_cm_rep_param rep;
2891 int ret;
2892
2893 ret = cma_modify_qp_rtr(id_priv, conn_param);
2894 if (ret)
2895 goto out;
2896
2897 ret = cma_modify_qp_rts(id_priv, conn_param);
2898 if (ret)
2899 goto out;
2900
2901 memset(&rep, 0, sizeof rep);
2902 rep.qp_num = id_priv->qp_num;
2903 rep.starting_psn = id_priv->seq_num;
2904 rep.private_data = conn_param->private_data;
2905 rep.private_data_len = conn_param->private_data_len;
2906 rep.responder_resources = conn_param->responder_resources;
2907 rep.initiator_depth = conn_param->initiator_depth;
2908 rep.failover_accepted = 0;
2909 rep.flow_control = conn_param->flow_control;
2910 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2911 rep.srq = id_priv->srq ? 1 : 0;
2912
2913 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2914 out:
2915 return ret;
2916 }
2917
2918 static int cma_accept_iw(struct rdma_id_private *id_priv,
2919 struct rdma_conn_param *conn_param)
2920 {
2921 struct iw_cm_conn_param iw_param;
2922 int ret;
2923
2924 ret = cma_modify_qp_rtr(id_priv, conn_param);
2925 if (ret)
2926 return ret;
2927
2928 iw_param.ord = conn_param->initiator_depth;
2929 iw_param.ird = conn_param->responder_resources;
2930 iw_param.private_data = conn_param->private_data;
2931 iw_param.private_data_len = conn_param->private_data_len;
2932 if (id_priv->id.qp) {
2933 iw_param.qpn = id_priv->qp_num;
2934 } else
2935 iw_param.qpn = conn_param->qp_num;
2936
2937 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2938 }
2939
2940 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2941 enum ib_cm_sidr_status status, u32 qkey,
2942 const void *private_data, int private_data_len)
2943 {
2944 struct ib_cm_sidr_rep_param rep;
2945 int ret;
2946
2947 memset(&rep, 0, sizeof rep);
2948 rep.status = status;
2949 if (status == IB_SIDR_SUCCESS) {
2950 ret = cma_set_qkey(id_priv, qkey);
2951 if (ret)
2952 return ret;
2953 rep.qp_num = id_priv->qp_num;
2954 rep.qkey = id_priv->qkey;
2955 }
2956 rep.private_data = private_data;
2957 rep.private_data_len = private_data_len;
2958
2959 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2960 }
2961
2962 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2963 {
2964 struct rdma_id_private *id_priv;
2965 int ret;
2966
2967 id_priv = container_of(id, struct rdma_id_private, id);
2968
2969 id_priv->owner = task_pid_nr(current);
2970
2971 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
2972 return -EINVAL;
2973
2974 if (!id->qp && conn_param) {
2975 id_priv->qp_num = conn_param->qp_num;
2976 id_priv->srq = conn_param->srq;
2977 }
2978
2979 switch (rdma_node_get_transport(id->device->node_type)) {
2980 case RDMA_TRANSPORT_IB:
2981 if (id->qp_type == IB_QPT_UD) {
2982 if (conn_param)
2983 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2984 conn_param->qkey,
2985 conn_param->private_data,
2986 conn_param->private_data_len);
2987 else
2988 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2989 0, NULL, 0);
2990 } else {
2991 if (conn_param)
2992 ret = cma_accept_ib(id_priv, conn_param);
2993 else
2994 ret = cma_rep_recv(id_priv);
2995 }
2996 break;
2997 case RDMA_TRANSPORT_IWARP:
2998 ret = cma_accept_iw(id_priv, conn_param);
2999 break;
3000 default:
3001 ret = -ENOSYS;
3002 break;
3003 }
3004
3005 if (ret)
3006 goto reject;
3007
3008 return 0;
3009 reject:
3010 cma_modify_qp_err(id_priv);
3011 rdma_reject(id, NULL, 0);
3012 return ret;
3013 }
3014 EXPORT_SYMBOL(rdma_accept);
3015
3016 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3017 {
3018 struct rdma_id_private *id_priv;
3019 int ret;
3020
3021 id_priv = container_of(id, struct rdma_id_private, id);
3022 if (!id_priv->cm_id.ib)
3023 return -EINVAL;
3024
3025 switch (id->device->node_type) {
3026 case RDMA_NODE_IB_CA:
3027 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3028 break;
3029 default:
3030 ret = 0;
3031 break;
3032 }
3033 return ret;
3034 }
3035 EXPORT_SYMBOL(rdma_notify);
3036
3037 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3038 u8 private_data_len)
3039 {
3040 struct rdma_id_private *id_priv;
3041 int ret;
3042
3043 id_priv = container_of(id, struct rdma_id_private, id);
3044 if (!id_priv->cm_id.ib)
3045 return -EINVAL;
3046
3047 switch (rdma_node_get_transport(id->device->node_type)) {
3048 case RDMA_TRANSPORT_IB:
3049 if (id->qp_type == IB_QPT_UD)
3050 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3051 private_data, private_data_len);
3052 else
3053 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3054 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3055 0, private_data, private_data_len);
3056 break;
3057 case RDMA_TRANSPORT_IWARP:
3058 ret = iw_cm_reject(id_priv->cm_id.iw,
3059 private_data, private_data_len);
3060 break;
3061 default:
3062 ret = -ENOSYS;
3063 break;
3064 }
3065 return ret;
3066 }
3067 EXPORT_SYMBOL(rdma_reject);
3068
3069 int rdma_disconnect(struct rdma_cm_id *id)
3070 {
3071 struct rdma_id_private *id_priv;
3072 int ret;
3073
3074 id_priv = container_of(id, struct rdma_id_private, id);
3075 if (!id_priv->cm_id.ib)
3076 return -EINVAL;
3077
3078 switch (rdma_node_get_transport(id->device->node_type)) {
3079 case RDMA_TRANSPORT_IB:
3080 ret = cma_modify_qp_err(id_priv);
3081 if (ret)
3082 goto out;
3083 /* Initiate or respond to a disconnect. */
3084 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3085 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3086 break;
3087 case RDMA_TRANSPORT_IWARP:
3088 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3089 break;
3090 default:
3091 ret = -EINVAL;
3092 break;
3093 }
3094 out:
3095 return ret;
3096 }
3097 EXPORT_SYMBOL(rdma_disconnect);
3098
3099 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3100 {
3101 struct rdma_id_private *id_priv;
3102 struct cma_multicast *mc = multicast->context;
3103 struct rdma_cm_event event;
3104 int ret;
3105
3106 id_priv = mc->id_priv;
3107 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3108 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3109 return 0;
3110
3111 if (!status)
3112 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3113 mutex_lock(&id_priv->qp_mutex);
3114 if (!status && id_priv->id.qp)
3115 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3116 be16_to_cpu(multicast->rec.mlid));
3117 mutex_unlock(&id_priv->qp_mutex);
3118
3119 memset(&event, 0, sizeof event);
3120 event.status = status;
3121 event.param.ud.private_data = mc->context;
3122 if (!status) {
3123 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3124 ib_init_ah_from_mcmember(id_priv->id.device,
3125 id_priv->id.port_num, &multicast->rec,
3126 &event.param.ud.ah_attr);
3127 event.param.ud.qp_num = 0xFFFFFF;
3128 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3129 } else
3130 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3131
3132 ret = id_priv->id.event_handler(&id_priv->id, &event);
3133 if (ret) {
3134 cma_exch(id_priv, RDMA_CM_DESTROYING);
3135 mutex_unlock(&id_priv->handler_mutex);
3136 rdma_destroy_id(&id_priv->id);
3137 return 0;
3138 }
3139
3140 mutex_unlock(&id_priv->handler_mutex);
3141 return 0;
3142 }
3143
3144 static void cma_set_mgid(struct rdma_id_private *id_priv,
3145 struct sockaddr *addr, union ib_gid *mgid)
3146 {
3147 unsigned char mc_map[MAX_ADDR_LEN];
3148 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3149 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3150 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3151
3152 if (cma_any_addr(addr)) {
3153 memset(mgid, 0, sizeof *mgid);
3154 } else if ((addr->sa_family == AF_INET6) &&
3155 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3156 0xFF10A01B)) {
3157 /* IPv6 address is an SA assigned MGID. */
3158 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3159 } else if (addr->sa_family == AF_IB) {
3160 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3161 } else if ((addr->sa_family == AF_INET6)) {
3162 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3163 if (id_priv->id.ps == RDMA_PS_UDP)
3164 mc_map[7] = 0x01; /* Use RDMA CM signature */
3165 *mgid = *(union ib_gid *) (mc_map + 4);
3166 } else {
3167 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3168 if (id_priv->id.ps == RDMA_PS_UDP)
3169 mc_map[7] = 0x01; /* Use RDMA CM signature */
3170 *mgid = *(union ib_gid *) (mc_map + 4);
3171 }
3172 }
3173
3174 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3175 struct cma_multicast *mc)
3176 {
3177 struct ib_sa_mcmember_rec rec;
3178 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3179 ib_sa_comp_mask comp_mask;
3180 int ret;
3181
3182 ib_addr_get_mgid(dev_addr, &rec.mgid);
3183 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3184 &rec.mgid, &rec);
3185 if (ret)
3186 return ret;
3187
3188 ret = cma_set_qkey(id_priv, 0);
3189 if (ret)
3190 return ret;
3191
3192 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3193 rec.qkey = cpu_to_be32(id_priv->qkey);
3194 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3195 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3196 rec.join_state = 1;
3197
3198 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3199 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3200 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3201 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3202 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3203
3204 if (id_priv->id.ps == RDMA_PS_IPOIB)
3205 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3206 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3207 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3208 IB_SA_MCMEMBER_REC_MTU |
3209 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3210
3211 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3212 id_priv->id.port_num, &rec,
3213 comp_mask, GFP_KERNEL,
3214 cma_ib_mc_handler, mc);
3215 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3216 }
3217
3218 static void iboe_mcast_work_handler(struct work_struct *work)
3219 {
3220 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3221 struct cma_multicast *mc = mw->mc;
3222 struct ib_sa_multicast *m = mc->multicast.ib;
3223
3224 mc->multicast.ib->context = mc;
3225 cma_ib_mc_handler(0, m);
3226 kref_put(&mc->mcref, release_mc);
3227 kfree(mw);
3228 }
3229
3230 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3231 {
3232 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3233 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3234
3235 if (cma_any_addr(addr)) {
3236 memset(mgid, 0, sizeof *mgid);
3237 } else if (addr->sa_family == AF_INET6) {
3238 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3239 } else {
3240 mgid->raw[0] = 0xff;
3241 mgid->raw[1] = 0x0e;
3242 mgid->raw[2] = 0;
3243 mgid->raw[3] = 0;
3244 mgid->raw[4] = 0;
3245 mgid->raw[5] = 0;
3246 mgid->raw[6] = 0;
3247 mgid->raw[7] = 0;
3248 mgid->raw[8] = 0;
3249 mgid->raw[9] = 0;
3250 mgid->raw[10] = 0xff;
3251 mgid->raw[11] = 0xff;
3252 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3253 }
3254 }
3255
3256 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3257 struct cma_multicast *mc)
3258 {
3259 struct iboe_mcast_work *work;
3260 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3261 int err;
3262 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3263 struct net_device *ndev = NULL;
3264
3265 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3266 return -EINVAL;
3267
3268 work = kzalloc(sizeof *work, GFP_KERNEL);
3269 if (!work)
3270 return -ENOMEM;
3271
3272 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3273 if (!mc->multicast.ib) {
3274 err = -ENOMEM;
3275 goto out1;
3276 }
3277
3278 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3279
3280 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3281 if (id_priv->id.ps == RDMA_PS_UDP)
3282 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3283
3284 if (dev_addr->bound_dev_if)
3285 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3286 if (!ndev) {
3287 err = -ENODEV;
3288 goto out2;
3289 }
3290 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3291 mc->multicast.ib->rec.hop_limit = 1;
3292 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3293 dev_put(ndev);
3294 if (!mc->multicast.ib->rec.mtu) {
3295 err = -EINVAL;
3296 goto out2;
3297 }
3298 iboe_addr_get_sgid(dev_addr, &mc->multicast.ib->rec.port_gid);
3299 work->id = id_priv;
3300 work->mc = mc;
3301 INIT_WORK(&work->work, iboe_mcast_work_handler);
3302 kref_get(&mc->mcref);
3303 queue_work(cma_wq, &work->work);
3304
3305 return 0;
3306
3307 out2:
3308 kfree(mc->multicast.ib);
3309 out1:
3310 kfree(work);
3311 return err;
3312 }
3313
3314 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3315 void *context)
3316 {
3317 struct rdma_id_private *id_priv;
3318 struct cma_multicast *mc;
3319 int ret;
3320
3321 id_priv = container_of(id, struct rdma_id_private, id);
3322 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3323 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3324 return -EINVAL;
3325
3326 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3327 if (!mc)
3328 return -ENOMEM;
3329
3330 memcpy(&mc->addr, addr, rdma_addr_size(addr));
3331 mc->context = context;
3332 mc->id_priv = id_priv;
3333
3334 spin_lock(&id_priv->lock);
3335 list_add(&mc->list, &id_priv->mc_list);
3336 spin_unlock(&id_priv->lock);
3337
3338 switch (rdma_node_get_transport(id->device->node_type)) {
3339 case RDMA_TRANSPORT_IB:
3340 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3341 case IB_LINK_LAYER_INFINIBAND:
3342 ret = cma_join_ib_multicast(id_priv, mc);
3343 break;
3344 case IB_LINK_LAYER_ETHERNET:
3345 kref_init(&mc->mcref);
3346 ret = cma_iboe_join_multicast(id_priv, mc);
3347 break;
3348 default:
3349 ret = -EINVAL;
3350 }
3351 break;
3352 default:
3353 ret = -ENOSYS;
3354 break;
3355 }
3356
3357 if (ret) {
3358 spin_lock_irq(&id_priv->lock);
3359 list_del(&mc->list);
3360 spin_unlock_irq(&id_priv->lock);
3361 kfree(mc);
3362 }
3363 return ret;
3364 }
3365 EXPORT_SYMBOL(rdma_join_multicast);
3366
3367 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3368 {
3369 struct rdma_id_private *id_priv;
3370 struct cma_multicast *mc;
3371
3372 id_priv = container_of(id, struct rdma_id_private, id);
3373 spin_lock_irq(&id_priv->lock);
3374 list_for_each_entry(mc, &id_priv->mc_list, list) {
3375 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3376 list_del(&mc->list);
3377 spin_unlock_irq(&id_priv->lock);
3378
3379 if (id->qp)
3380 ib_detach_mcast(id->qp,
3381 &mc->multicast.ib->rec.mgid,
3382 be16_to_cpu(mc->multicast.ib->rec.mlid));
3383 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3384 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3385 case IB_LINK_LAYER_INFINIBAND:
3386 ib_sa_free_multicast(mc->multicast.ib);
3387 kfree(mc);
3388 break;
3389 case IB_LINK_LAYER_ETHERNET:
3390 kref_put(&mc->mcref, release_mc);
3391 break;
3392 default:
3393 break;
3394 }
3395 }
3396 return;
3397 }
3398 }
3399 spin_unlock_irq(&id_priv->lock);
3400 }
3401 EXPORT_SYMBOL(rdma_leave_multicast);
3402
3403 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3404 {
3405 struct rdma_dev_addr *dev_addr;
3406 struct cma_ndev_work *work;
3407
3408 dev_addr = &id_priv->id.route.addr.dev_addr;
3409
3410 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3411 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3412 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3413 ndev->name, &id_priv->id);
3414 work = kzalloc(sizeof *work, GFP_KERNEL);
3415 if (!work)
3416 return -ENOMEM;
3417
3418 INIT_WORK(&work->work, cma_ndev_work_handler);
3419 work->id = id_priv;
3420 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3421 atomic_inc(&id_priv->refcount);
3422 queue_work(cma_wq, &work->work);
3423 }
3424
3425 return 0;
3426 }
3427
3428 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3429 void *ptr)
3430 {
3431 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3432 struct cma_device *cma_dev;
3433 struct rdma_id_private *id_priv;
3434 int ret = NOTIFY_DONE;
3435
3436 if (dev_net(ndev) != &init_net)
3437 return NOTIFY_DONE;
3438
3439 if (event != NETDEV_BONDING_FAILOVER)
3440 return NOTIFY_DONE;
3441
3442 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3443 return NOTIFY_DONE;
3444
3445 mutex_lock(&lock);
3446 list_for_each_entry(cma_dev, &dev_list, list)
3447 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3448 ret = cma_netdev_change(ndev, id_priv);
3449 if (ret)
3450 goto out;
3451 }
3452
3453 out:
3454 mutex_unlock(&lock);
3455 return ret;
3456 }
3457
3458 static struct notifier_block cma_nb = {
3459 .notifier_call = cma_netdev_callback
3460 };
3461
3462 static void cma_add_one(struct ib_device *device)
3463 {
3464 struct cma_device *cma_dev;
3465 struct rdma_id_private *id_priv;
3466
3467 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3468 if (!cma_dev)
3469 return;
3470
3471 cma_dev->device = device;
3472
3473 init_completion(&cma_dev->comp);
3474 atomic_set(&cma_dev->refcount, 1);
3475 INIT_LIST_HEAD(&cma_dev->id_list);
3476 ib_set_client_data(device, &cma_client, cma_dev);
3477
3478 mutex_lock(&lock);
3479 list_add_tail(&cma_dev->list, &dev_list);
3480 list_for_each_entry(id_priv, &listen_any_list, list)
3481 cma_listen_on_dev(id_priv, cma_dev);
3482 mutex_unlock(&lock);
3483 }
3484
3485 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3486 {
3487 struct rdma_cm_event event;
3488 enum rdma_cm_state state;
3489 int ret = 0;
3490
3491 /* Record that we want to remove the device */
3492 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3493 if (state == RDMA_CM_DESTROYING)
3494 return 0;
3495
3496 cma_cancel_operation(id_priv, state);
3497 mutex_lock(&id_priv->handler_mutex);
3498
3499 /* Check for destruction from another callback. */
3500 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3501 goto out;
3502
3503 memset(&event, 0, sizeof event);
3504 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3505 ret = id_priv->id.event_handler(&id_priv->id, &event);
3506 out:
3507 mutex_unlock(&id_priv->handler_mutex);
3508 return ret;
3509 }
3510
3511 static void cma_process_remove(struct cma_device *cma_dev)
3512 {
3513 struct rdma_id_private *id_priv;
3514 int ret;
3515
3516 mutex_lock(&lock);
3517 while (!list_empty(&cma_dev->id_list)) {
3518 id_priv = list_entry(cma_dev->id_list.next,
3519 struct rdma_id_private, list);
3520
3521 list_del(&id_priv->listen_list);
3522 list_del_init(&id_priv->list);
3523 atomic_inc(&id_priv->refcount);
3524 mutex_unlock(&lock);
3525
3526 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3527 cma_deref_id(id_priv);
3528 if (ret)
3529 rdma_destroy_id(&id_priv->id);
3530
3531 mutex_lock(&lock);
3532 }
3533 mutex_unlock(&lock);
3534
3535 cma_deref_dev(cma_dev);
3536 wait_for_completion(&cma_dev->comp);
3537 }
3538
3539 static void cma_remove_one(struct ib_device *device)
3540 {
3541 struct cma_device *cma_dev;
3542
3543 cma_dev = ib_get_client_data(device, &cma_client);
3544 if (!cma_dev)
3545 return;
3546
3547 mutex_lock(&lock);
3548 list_del(&cma_dev->list);
3549 mutex_unlock(&lock);
3550
3551 cma_process_remove(cma_dev);
3552 kfree(cma_dev);
3553 }
3554
3555 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3556 {
3557 struct nlmsghdr *nlh;
3558 struct rdma_cm_id_stats *id_stats;
3559 struct rdma_id_private *id_priv;
3560 struct rdma_cm_id *id = NULL;
3561 struct cma_device *cma_dev;
3562 int i_dev = 0, i_id = 0;
3563
3564 /*
3565 * We export all of the IDs as a sequence of messages. Each
3566 * ID gets its own netlink message.
3567 */
3568 mutex_lock(&lock);
3569
3570 list_for_each_entry(cma_dev, &dev_list, list) {
3571 if (i_dev < cb->args[0]) {
3572 i_dev++;
3573 continue;
3574 }
3575
3576 i_id = 0;
3577 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3578 if (i_id < cb->args[1]) {
3579 i_id++;
3580 continue;
3581 }
3582
3583 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3584 sizeof *id_stats, RDMA_NL_RDMA_CM,
3585 RDMA_NL_RDMA_CM_ID_STATS);
3586 if (!id_stats)
3587 goto out;
3588
3589 memset(id_stats, 0, sizeof *id_stats);
3590 id = &id_priv->id;
3591 id_stats->node_type = id->route.addr.dev_addr.dev_type;
3592 id_stats->port_num = id->port_num;
3593 id_stats->bound_dev_if =
3594 id->route.addr.dev_addr.bound_dev_if;
3595
3596 if (ibnl_put_attr(skb, nlh,
3597 rdma_addr_size(cma_src_addr(id_priv)),
3598 cma_src_addr(id_priv),
3599 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3600 goto out;
3601 if (ibnl_put_attr(skb, nlh,
3602 rdma_addr_size(cma_src_addr(id_priv)),
3603 cma_dst_addr(id_priv),
3604 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
3605 goto out;
3606
3607 id_stats->pid = id_priv->owner;
3608 id_stats->port_space = id->ps;
3609 id_stats->cm_state = id_priv->state;
3610 id_stats->qp_num = id_priv->qp_num;
3611 id_stats->qp_type = id->qp_type;
3612
3613 i_id++;
3614 }
3615
3616 cb->args[1] = 0;
3617 i_dev++;
3618 }
3619
3620 out:
3621 mutex_unlock(&lock);
3622 cb->args[0] = i_dev;
3623 cb->args[1] = i_id;
3624
3625 return skb->len;
3626 }
3627
3628 static const struct ibnl_client_cbs cma_cb_table[] = {
3629 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3630 .module = THIS_MODULE },
3631 };
3632
3633 static int __init cma_init(void)
3634 {
3635 int ret;
3636
3637 cma_wq = create_singlethread_workqueue("rdma_cm");
3638 if (!cma_wq)
3639 return -ENOMEM;
3640
3641 ib_sa_register_client(&sa_client);
3642 rdma_addr_register_client(&addr_client);
3643 register_netdevice_notifier(&cma_nb);
3644
3645 ret = ib_register_client(&cma_client);
3646 if (ret)
3647 goto err;
3648
3649 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3650 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3651
3652 return 0;
3653
3654 err:
3655 unregister_netdevice_notifier(&cma_nb);
3656 rdma_addr_unregister_client(&addr_client);
3657 ib_sa_unregister_client(&sa_client);
3658 destroy_workqueue(cma_wq);
3659 return ret;
3660 }
3661
3662 static void __exit cma_cleanup(void)
3663 {
3664 ibnl_remove_client(RDMA_NL_RDMA_CM);
3665 ib_unregister_client(&cma_client);
3666 unregister_netdevice_notifier(&cma_nb);
3667 rdma_addr_unregister_client(&addr_client);
3668 ib_sa_unregister_client(&sa_client);
3669 destroy_workqueue(cma_wq);
3670 idr_destroy(&tcp_ps);
3671 idr_destroy(&udp_ps);
3672 idr_destroy(&ipoib_ps);
3673 idr_destroy(&ib_ps);
3674 }
3675
3676 module_init(cma_init);
3677 module_exit(cma_cleanup);
Ubuntu Jammy Linux kernel mirror