?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fix a kmap leak in virtio_console
[linux/fpc-iii.git] / drivers / infiniband / core / cma.c
blob199958d9ddc8088d7576904ab96c838fb193990c
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, NULL);
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 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
369 &iboe_gid);
370
371 memcpy(&gid, dev_addr->src_dev_addr +
372 rdma_addr_gid_offset(dev_addr), sizeof gid);
373 if (listen_id_priv &&
374 rdma_port_get_link_layer(listen_id_priv->id.device,
375 listen_id_priv->id.port_num) == dev_ll) {
376 cma_dev = listen_id_priv->cma_dev;
377 port = listen_id_priv->id.port_num;
378 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
379 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
380 ret = ib_find_cached_gid(cma_dev->device, &iboe_gid,
381 &found_port, NULL);
382 else
383 ret = ib_find_cached_gid(cma_dev->device, &gid,
384 &found_port, NULL);
385
386 if (!ret && (port == found_port)) {
387 id_priv->id.port_num = found_port;
388 goto out;
389 }
390 }
391 list_for_each_entry(cma_dev, &dev_list, list) {
392 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
393 if (listen_id_priv &&
394 listen_id_priv->cma_dev == cma_dev &&
395 listen_id_priv->id.port_num == port)
396 continue;
397 if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
398 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
399 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
400 ret = ib_find_cached_gid(cma_dev->device, &iboe_gid, &found_port, NULL);
401 else
402 ret = ib_find_cached_gid(cma_dev->device, &gid, &found_port, NULL);
403
404 if (!ret && (port == found_port)) {
405 id_priv->id.port_num = found_port;
406 goto out;
407 }
408 }
409 }
410 }
411
412 out:
413 if (!ret)
414 cma_attach_to_dev(id_priv, cma_dev);
415
416 mutex_unlock(&lock);
417 return ret;
418 }
419
420 /*
421 * Select the source IB device and address to reach the destination IB address.
422 */
423 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
424 {
425 struct cma_device *cma_dev, *cur_dev;
426 struct sockaddr_ib *addr;
427 union ib_gid gid, sgid, *dgid;
428 u16 pkey, index;
429 u8 p;
430 int i;
431
432 cma_dev = NULL;
433 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
434 dgid = (union ib_gid *) &addr->sib_addr;
435 pkey = ntohs(addr->sib_pkey);
436
437 list_for_each_entry(cur_dev, &dev_list, list) {
438 if (rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
439 continue;
440
441 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
442 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
443 continue;
444
445 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
446 if (!memcmp(&gid, dgid, sizeof(gid))) {
447 cma_dev = cur_dev;
448 sgid = gid;
449 id_priv->id.port_num = p;
450 goto found;
451 }
452
453 if (!cma_dev && (gid.global.subnet_prefix ==
454 dgid->global.subnet_prefix)) {
455 cma_dev = cur_dev;
456 sgid = gid;
457 id_priv->id.port_num = p;
458 }
459 }
460 }
461 }
462
463 if (!cma_dev)
464 return -ENODEV;
465
466 found:
467 cma_attach_to_dev(id_priv, cma_dev);
468 addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
469 memcpy(&addr->sib_addr, &sgid, sizeof sgid);
470 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
471 return 0;
472 }
473
474 static void cma_deref_id(struct rdma_id_private *id_priv)
475 {
476 if (atomic_dec_and_test(&id_priv->refcount))
477 complete(&id_priv->comp);
478 }
479
480 static int cma_disable_callback(struct rdma_id_private *id_priv,
481 enum rdma_cm_state state)
482 {
483 mutex_lock(&id_priv->handler_mutex);
484 if (id_priv->state != state) {
485 mutex_unlock(&id_priv->handler_mutex);
486 return -EINVAL;
487 }
488 return 0;
489 }
490
491 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
492 void *context, enum rdma_port_space ps,
493 enum ib_qp_type qp_type)
494 {
495 struct rdma_id_private *id_priv;
496
497 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
498 if (!id_priv)
499 return ERR_PTR(-ENOMEM);
500
501 id_priv->owner = task_pid_nr(current);
502 id_priv->state = RDMA_CM_IDLE;
503 id_priv->id.context = context;
504 id_priv->id.event_handler = event_handler;
505 id_priv->id.ps = ps;
506 id_priv->id.qp_type = qp_type;
507 spin_lock_init(&id_priv->lock);
508 mutex_init(&id_priv->qp_mutex);
509 init_completion(&id_priv->comp);
510 atomic_set(&id_priv->refcount, 1);
511 mutex_init(&id_priv->handler_mutex);
512 INIT_LIST_HEAD(&id_priv->listen_list);
513 INIT_LIST_HEAD(&id_priv->mc_list);
514 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
515
516 return &id_priv->id;
517 }
518 EXPORT_SYMBOL(rdma_create_id);
519
520 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
521 {
522 struct ib_qp_attr qp_attr;
523 int qp_attr_mask, ret;
524
525 qp_attr.qp_state = IB_QPS_INIT;
526 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
527 if (ret)
528 return ret;
529
530 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
531 if (ret)
532 return ret;
533
534 qp_attr.qp_state = IB_QPS_RTR;
535 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
536 if (ret)
537 return ret;
538
539 qp_attr.qp_state = IB_QPS_RTS;
540 qp_attr.sq_psn = 0;
541 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
542
543 return ret;
544 }
545
546 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
547 {
548 struct ib_qp_attr qp_attr;
549 int qp_attr_mask, ret;
550
551 qp_attr.qp_state = IB_QPS_INIT;
552 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
553 if (ret)
554 return ret;
555
556 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
557 }
558
559 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
560 struct ib_qp_init_attr *qp_init_attr)
561 {
562 struct rdma_id_private *id_priv;
563 struct ib_qp *qp;
564 int ret;
565
566 id_priv = container_of(id, struct rdma_id_private, id);
567 if (id->device != pd->device)
568 return -EINVAL;
569
570 qp = ib_create_qp(pd, qp_init_attr);
571 if (IS_ERR(qp))
572 return PTR_ERR(qp);
573
574 if (id->qp_type == IB_QPT_UD)
575 ret = cma_init_ud_qp(id_priv, qp);
576 else
577 ret = cma_init_conn_qp(id_priv, qp);
578 if (ret)
579 goto err;
580
581 id->qp = qp;
582 id_priv->qp_num = qp->qp_num;
583 id_priv->srq = (qp->srq != NULL);
584 return 0;
585 err:
586 ib_destroy_qp(qp);
587 return ret;
588 }
589 EXPORT_SYMBOL(rdma_create_qp);
590
591 void rdma_destroy_qp(struct rdma_cm_id *id)
592 {
593 struct rdma_id_private *id_priv;
594
595 id_priv = container_of(id, struct rdma_id_private, id);
596 mutex_lock(&id_priv->qp_mutex);
597 ib_destroy_qp(id_priv->id.qp);
598 id_priv->id.qp = NULL;
599 mutex_unlock(&id_priv->qp_mutex);
600 }
601 EXPORT_SYMBOL(rdma_destroy_qp);
602
603 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
604 struct rdma_conn_param *conn_param)
605 {
606 struct ib_qp_attr qp_attr;
607 int qp_attr_mask, ret;
608 union ib_gid sgid;
609
610 mutex_lock(&id_priv->qp_mutex);
611 if (!id_priv->id.qp) {
612 ret = 0;
613 goto out;
614 }
615
616 /* Need to update QP attributes from default values. */
617 qp_attr.qp_state = IB_QPS_INIT;
618 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
619 if (ret)
620 goto out;
621
622 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
623 if (ret)
624 goto out;
625
626 qp_attr.qp_state = IB_QPS_RTR;
627 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
628 if (ret)
629 goto out;
630
631 ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
632 qp_attr.ah_attr.grh.sgid_index, &sgid);
633 if (ret)
634 goto out;
635
636 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type)
637 == RDMA_TRANSPORT_IB &&
638 rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)
639 == IB_LINK_LAYER_ETHERNET) {
640 ret = rdma_addr_find_smac_by_sgid(&sgid, qp_attr.smac, NULL);
641
642 if (ret)
643 goto out;
644 }
645 if (conn_param)
646 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
647 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
648 out:
649 mutex_unlock(&id_priv->qp_mutex);
650 return ret;
651 }
652
653 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
654 struct rdma_conn_param *conn_param)
655 {
656 struct ib_qp_attr qp_attr;
657 int qp_attr_mask, ret;
658
659 mutex_lock(&id_priv->qp_mutex);
660 if (!id_priv->id.qp) {
661 ret = 0;
662 goto out;
663 }
664
665 qp_attr.qp_state = IB_QPS_RTS;
666 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
667 if (ret)
668 goto out;
669
670 if (conn_param)
671 qp_attr.max_rd_atomic = conn_param->initiator_depth;
672 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
673 out:
674 mutex_unlock(&id_priv->qp_mutex);
675 return ret;
676 }
677
678 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
679 {
680 struct ib_qp_attr qp_attr;
681 int ret;
682
683 mutex_lock(&id_priv->qp_mutex);
684 if (!id_priv->id.qp) {
685 ret = 0;
686 goto out;
687 }
688
689 qp_attr.qp_state = IB_QPS_ERR;
690 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
691 out:
692 mutex_unlock(&id_priv->qp_mutex);
693 return ret;
694 }
695
696 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
697 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
698 {
699 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
700 int ret;
701 u16 pkey;
702
703 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
704 IB_LINK_LAYER_INFINIBAND)
705 pkey = ib_addr_get_pkey(dev_addr);
706 else
707 pkey = 0xffff;
708
709 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
710 pkey, &qp_attr->pkey_index);
711 if (ret)
712 return ret;
713
714 qp_attr->port_num = id_priv->id.port_num;
715 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
716
717 if (id_priv->id.qp_type == IB_QPT_UD) {
718 ret = cma_set_qkey(id_priv, 0);
719 if (ret)
720 return ret;
721
722 qp_attr->qkey = id_priv->qkey;
723 *qp_attr_mask |= IB_QP_QKEY;
724 } else {
725 qp_attr->qp_access_flags = 0;
726 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
727 }
728 return 0;
729 }
730
731 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
732 int *qp_attr_mask)
733 {
734 struct rdma_id_private *id_priv;
735 int ret = 0;
736
737 id_priv = container_of(id, struct rdma_id_private, id);
738 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
739 case RDMA_TRANSPORT_IB:
740 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
741 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
742 else
743 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
744 qp_attr_mask);
745
746 if (qp_attr->qp_state == IB_QPS_RTR)
747 qp_attr->rq_psn = id_priv->seq_num;
748 break;
749 case RDMA_TRANSPORT_IWARP:
750 if (!id_priv->cm_id.iw) {
751 qp_attr->qp_access_flags = 0;
752 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
753 } else
754 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
755 qp_attr_mask);
756 break;
757 default:
758 ret = -ENOSYS;
759 break;
760 }
761
762 return ret;
763 }
764 EXPORT_SYMBOL(rdma_init_qp_attr);
765
766 static inline int cma_zero_addr(struct sockaddr *addr)
767 {
768 switch (addr->sa_family) {
769 case AF_INET:
770 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
771 case AF_INET6:
772 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
773 case AF_IB:
774 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
775 default:
776 return 0;
777 }
778 }
779
780 static inline int cma_loopback_addr(struct sockaddr *addr)
781 {
782 switch (addr->sa_family) {
783 case AF_INET:
784 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
785 case AF_INET6:
786 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
787 case AF_IB:
788 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
789 default:
790 return 0;
791 }
792 }
793
794 static inline int cma_any_addr(struct sockaddr *addr)
795 {
796 return cma_zero_addr(addr) || cma_loopback_addr(addr);
797 }
798
799 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
800 {
801 if (src->sa_family != dst->sa_family)
802 return -1;
803
804 switch (src->sa_family) {
805 case AF_INET:
806 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
807 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
808 case AF_INET6:
809 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
810 &((struct sockaddr_in6 *) dst)->sin6_addr);
811 default:
812 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
813 &((struct sockaddr_ib *) dst)->sib_addr);
814 }
815 }
816
817 static __be16 cma_port(struct sockaddr *addr)
818 {
819 struct sockaddr_ib *sib;
820
821 switch (addr->sa_family) {
822 case AF_INET:
823 return ((struct sockaddr_in *) addr)->sin_port;
824 case AF_INET6:
825 return ((struct sockaddr_in6 *) addr)->sin6_port;
826 case AF_IB:
827 sib = (struct sockaddr_ib *) addr;
828 return htons((u16) (be64_to_cpu(sib->sib_sid) &
829 be64_to_cpu(sib->sib_sid_mask)));
830 default:
831 return 0;
832 }
833 }
834
835 static inline int cma_any_port(struct sockaddr *addr)
836 {
837 return !cma_port(addr);
838 }
839
840 static void cma_save_ib_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
841 struct ib_sa_path_rec *path)
842 {
843 struct sockaddr_ib *listen_ib, *ib;
844
845 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
846 ib = (struct sockaddr_ib *) &id->route.addr.src_addr;
847 ib->sib_family = listen_ib->sib_family;
848 ib->sib_pkey = path->pkey;
849 ib->sib_flowinfo = path->flow_label;
850 memcpy(&ib->sib_addr, &path->sgid, 16);
851 ib->sib_sid = listen_ib->sib_sid;
852 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
853 ib->sib_scope_id = listen_ib->sib_scope_id;
854
855 ib = (struct sockaddr_ib *) &id->route.addr.dst_addr;
856 ib->sib_family = listen_ib->sib_family;
857 ib->sib_pkey = path->pkey;
858 ib->sib_flowinfo = path->flow_label;
859 memcpy(&ib->sib_addr, &path->dgid, 16);
860 }
861
862 static void cma_save_ip4_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
863 struct cma_hdr *hdr)
864 {
865 struct sockaddr_in *listen4, *ip4;
866
867 listen4 = (struct sockaddr_in *) &listen_id->route.addr.src_addr;
868 ip4 = (struct sockaddr_in *) &id->route.addr.src_addr;
869 ip4->sin_family = listen4->sin_family;
870 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
871 ip4->sin_port = listen4->sin_port;
872
873 ip4 = (struct sockaddr_in *) &id->route.addr.dst_addr;
874 ip4->sin_family = listen4->sin_family;
875 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
876 ip4->sin_port = hdr->port;
877 }
878
879 static void cma_save_ip6_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
880 struct cma_hdr *hdr)
881 {
882 struct sockaddr_in6 *listen6, *ip6;
883
884 listen6 = (struct sockaddr_in6 *) &listen_id->route.addr.src_addr;
885 ip6 = (struct sockaddr_in6 *) &id->route.addr.src_addr;
886 ip6->sin6_family = listen6->sin6_family;
887 ip6->sin6_addr = hdr->dst_addr.ip6;
888 ip6->sin6_port = listen6->sin6_port;
889
890 ip6 = (struct sockaddr_in6 *) &id->route.addr.dst_addr;
891 ip6->sin6_family = listen6->sin6_family;
892 ip6->sin6_addr = hdr->src_addr.ip6;
893 ip6->sin6_port = hdr->port;
894 }
895
896 static int cma_save_net_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
897 struct ib_cm_event *ib_event)
898 {
899 struct cma_hdr *hdr;
900
901 if ((listen_id->route.addr.src_addr.ss_family == AF_IB) &&
902 (ib_event->event == IB_CM_REQ_RECEIVED)) {
903 cma_save_ib_info(id, listen_id, ib_event->param.req_rcvd.primary_path);
904 return 0;
905 }
906
907 hdr = ib_event->private_data;
908 if (hdr->cma_version != CMA_VERSION)
909 return -EINVAL;
910
911 switch (cma_get_ip_ver(hdr)) {
912 case 4:
913 cma_save_ip4_info(id, listen_id, hdr);
914 break;
915 case 6:
916 cma_save_ip6_info(id, listen_id, hdr);
917 break;
918 default:
919 return -EINVAL;
920 }
921 return 0;
922 }
923
924 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
925 {
926 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
927 }
928
929 static void cma_cancel_route(struct rdma_id_private *id_priv)
930 {
931 switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
932 case IB_LINK_LAYER_INFINIBAND:
933 if (id_priv->query)
934 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
935 break;
936 default:
937 break;
938 }
939 }
940
941 static void cma_cancel_listens(struct rdma_id_private *id_priv)
942 {
943 struct rdma_id_private *dev_id_priv;
944
945 /*
946 * Remove from listen_any_list to prevent added devices from spawning
947 * additional listen requests.
948 */
949 mutex_lock(&lock);
950 list_del(&id_priv->list);
951
952 while (!list_empty(&id_priv->listen_list)) {
953 dev_id_priv = list_entry(id_priv->listen_list.next,
954 struct rdma_id_private, listen_list);
955 /* sync with device removal to avoid duplicate destruction */
956 list_del_init(&dev_id_priv->list);
957 list_del(&dev_id_priv->listen_list);
958 mutex_unlock(&lock);
959
960 rdma_destroy_id(&dev_id_priv->id);
961 mutex_lock(&lock);
962 }
963 mutex_unlock(&lock);
964 }
965
966 static void cma_cancel_operation(struct rdma_id_private *id_priv,
967 enum rdma_cm_state state)
968 {
969 switch (state) {
970 case RDMA_CM_ADDR_QUERY:
971 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
972 break;
973 case RDMA_CM_ROUTE_QUERY:
974 cma_cancel_route(id_priv);
975 break;
976 case RDMA_CM_LISTEN:
977 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
978 cma_cancel_listens(id_priv);
979 break;
980 default:
981 break;
982 }
983 }
984
985 static void cma_release_port(struct rdma_id_private *id_priv)
986 {
987 struct rdma_bind_list *bind_list = id_priv->bind_list;
988
989 if (!bind_list)
990 return;
991
992 mutex_lock(&lock);
993 hlist_del(&id_priv->node);
994 if (hlist_empty(&bind_list->owners)) {
995 idr_remove(bind_list->ps, bind_list->port);
996 kfree(bind_list);
997 }
998 mutex_unlock(&lock);
999 }
1000
1001 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1002 {
1003 struct cma_multicast *mc;
1004
1005 while (!list_empty(&id_priv->mc_list)) {
1006 mc = container_of(id_priv->mc_list.next,
1007 struct cma_multicast, list);
1008 list_del(&mc->list);
1009 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
1010 case IB_LINK_LAYER_INFINIBAND:
1011 ib_sa_free_multicast(mc->multicast.ib);
1012 kfree(mc);
1013 break;
1014 case IB_LINK_LAYER_ETHERNET:
1015 kref_put(&mc->mcref, release_mc);
1016 break;
1017 default:
1018 break;
1019 }
1020 }
1021 }
1022
1023 void rdma_destroy_id(struct rdma_cm_id *id)
1024 {
1025 struct rdma_id_private *id_priv;
1026 enum rdma_cm_state state;
1027
1028 id_priv = container_of(id, struct rdma_id_private, id);
1029 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1030 cma_cancel_operation(id_priv, state);
1031
1032 /*
1033 * Wait for any active callback to finish. New callbacks will find
1034 * the id_priv state set to destroying and abort.
1035 */
1036 mutex_lock(&id_priv->handler_mutex);
1037 mutex_unlock(&id_priv->handler_mutex);
1038
1039 if (id_priv->cma_dev) {
1040 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
1041 case RDMA_TRANSPORT_IB:
1042 if (id_priv->cm_id.ib)
1043 ib_destroy_cm_id(id_priv->cm_id.ib);
1044 break;
1045 case RDMA_TRANSPORT_IWARP:
1046 if (id_priv->cm_id.iw)
1047 iw_destroy_cm_id(id_priv->cm_id.iw);
1048 break;
1049 default:
1050 break;
1051 }
1052 cma_leave_mc_groups(id_priv);
1053 cma_release_dev(id_priv);
1054 }
1055
1056 cma_release_port(id_priv);
1057 cma_deref_id(id_priv);
1058 wait_for_completion(&id_priv->comp);
1059
1060 if (id_priv->internal_id)
1061 cma_deref_id(id_priv->id.context);
1062
1063 kfree(id_priv->id.route.path_rec);
1064 kfree(id_priv);
1065 }
1066 EXPORT_SYMBOL(rdma_destroy_id);
1067
1068 static int cma_rep_recv(struct rdma_id_private *id_priv)
1069 {
1070 int ret;
1071
1072 ret = cma_modify_qp_rtr(id_priv, NULL);
1073 if (ret)
1074 goto reject;
1075
1076 ret = cma_modify_qp_rts(id_priv, NULL);
1077 if (ret)
1078 goto reject;
1079
1080 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1081 if (ret)
1082 goto reject;
1083
1084 return 0;
1085 reject:
1086 cma_modify_qp_err(id_priv);
1087 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1088 NULL, 0, NULL, 0);
1089 return ret;
1090 }
1091
1092 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1093 struct ib_cm_rep_event_param *rep_data,
1094 void *private_data)
1095 {
1096 event->param.conn.private_data = private_data;
1097 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1098 event->param.conn.responder_resources = rep_data->responder_resources;
1099 event->param.conn.initiator_depth = rep_data->initiator_depth;
1100 event->param.conn.flow_control = rep_data->flow_control;
1101 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1102 event->param.conn.srq = rep_data->srq;
1103 event->param.conn.qp_num = rep_data->remote_qpn;
1104 }
1105
1106 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1107 {
1108 struct rdma_id_private *id_priv = cm_id->context;
1109 struct rdma_cm_event event;
1110 int ret = 0;
1111
1112 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1113 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1114 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1115 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1116 return 0;
1117
1118 memset(&event, 0, sizeof event);
1119 switch (ib_event->event) {
1120 case IB_CM_REQ_ERROR:
1121 case IB_CM_REP_ERROR:
1122 event.event = RDMA_CM_EVENT_UNREACHABLE;
1123 event.status = -ETIMEDOUT;
1124 break;
1125 case IB_CM_REP_RECEIVED:
1126 if (id_priv->id.qp) {
1127 event.status = cma_rep_recv(id_priv);
1128 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1129 RDMA_CM_EVENT_ESTABLISHED;
1130 } else {
1131 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1132 }
1133 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1134 ib_event->private_data);
1135 break;
1136 case IB_CM_RTU_RECEIVED:
1137 case IB_CM_USER_ESTABLISHED:
1138 event.event = RDMA_CM_EVENT_ESTABLISHED;
1139 break;
1140 case IB_CM_DREQ_ERROR:
1141 event.status = -ETIMEDOUT; /* fall through */
1142 case IB_CM_DREQ_RECEIVED:
1143 case IB_CM_DREP_RECEIVED:
1144 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1145 RDMA_CM_DISCONNECT))
1146 goto out;
1147 event.event = RDMA_CM_EVENT_DISCONNECTED;
1148 break;
1149 case IB_CM_TIMEWAIT_EXIT:
1150 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1151 break;
1152 case IB_CM_MRA_RECEIVED:
1153 /* ignore event */
1154 goto out;
1155 case IB_CM_REJ_RECEIVED:
1156 cma_modify_qp_err(id_priv);
1157 event.status = ib_event->param.rej_rcvd.reason;
1158 event.event = RDMA_CM_EVENT_REJECTED;
1159 event.param.conn.private_data = ib_event->private_data;
1160 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1161 break;
1162 default:
1163 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1164 ib_event->event);
1165 goto out;
1166 }
1167
1168 ret = id_priv->id.event_handler(&id_priv->id, &event);
1169 if (ret) {
1170 /* Destroy the CM ID by returning a non-zero value. */
1171 id_priv->cm_id.ib = NULL;
1172 cma_exch(id_priv, RDMA_CM_DESTROYING);
1173 mutex_unlock(&id_priv->handler_mutex);
1174 rdma_destroy_id(&id_priv->id);
1175 return ret;
1176 }
1177 out:
1178 mutex_unlock(&id_priv->handler_mutex);
1179 return ret;
1180 }
1181
1182 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1183 struct ib_cm_event *ib_event)
1184 {
1185 struct rdma_id_private *id_priv;
1186 struct rdma_cm_id *id;
1187 struct rdma_route *rt;
1188 int ret;
1189
1190 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1191 listen_id->ps, ib_event->param.req_rcvd.qp_type);
1192 if (IS_ERR(id))
1193 return NULL;
1194
1195 id_priv = container_of(id, struct rdma_id_private, id);
1196 if (cma_save_net_info(id, listen_id, ib_event))
1197 goto err;
1198
1199 rt = &id->route;
1200 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1201 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1202 GFP_KERNEL);
1203 if (!rt->path_rec)
1204 goto err;
1205
1206 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1207 if (rt->num_paths == 2)
1208 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1209
1210 if (cma_any_addr(cma_src_addr(id_priv))) {
1211 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1212 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1213 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1214 } else {
1215 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1216 if (ret)
1217 goto err;
1218 }
1219 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1220
1221 id_priv->state = RDMA_CM_CONNECT;
1222 return id_priv;
1223
1224 err:
1225 rdma_destroy_id(id);
1226 return NULL;
1227 }
1228
1229 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1230 struct ib_cm_event *ib_event)
1231 {
1232 struct rdma_id_private *id_priv;
1233 struct rdma_cm_id *id;
1234 int ret;
1235
1236 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1237 listen_id->ps, IB_QPT_UD);
1238 if (IS_ERR(id))
1239 return NULL;
1240
1241 id_priv = container_of(id, struct rdma_id_private, id);
1242 if (cma_save_net_info(id, listen_id, ib_event))
1243 goto err;
1244
1245 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1246 ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
1247 if (ret)
1248 goto err;
1249 }
1250
1251 id_priv->state = RDMA_CM_CONNECT;
1252 return id_priv;
1253 err:
1254 rdma_destroy_id(id);
1255 return NULL;
1256 }
1257
1258 static void cma_set_req_event_data(struct rdma_cm_event *event,
1259 struct ib_cm_req_event_param *req_data,
1260 void *private_data, int offset)
1261 {
1262 event->param.conn.private_data = private_data + offset;
1263 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1264 event->param.conn.responder_resources = req_data->responder_resources;
1265 event->param.conn.initiator_depth = req_data->initiator_depth;
1266 event->param.conn.flow_control = req_data->flow_control;
1267 event->param.conn.retry_count = req_data->retry_count;
1268 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1269 event->param.conn.srq = req_data->srq;
1270 event->param.conn.qp_num = req_data->remote_qpn;
1271 }
1272
1273 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1274 {
1275 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1276 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1277 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1278 (id->qp_type == IB_QPT_UD)) ||
1279 (!id->qp_type));
1280 }
1281
1282 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1283 {
1284 struct rdma_id_private *listen_id, *conn_id;
1285 struct rdma_cm_event event;
1286 int offset, ret;
1287 u8 smac[ETH_ALEN];
1288 u8 alt_smac[ETH_ALEN];
1289 u8 *psmac = smac;
1290 u8 *palt_smac = alt_smac;
1291 int is_iboe = ((rdma_node_get_transport(cm_id->device->node_type) ==
1292 RDMA_TRANSPORT_IB) &&
1293 (rdma_port_get_link_layer(cm_id->device,
1294 ib_event->param.req_rcvd.port) ==
1295 IB_LINK_LAYER_ETHERNET));
1296
1297 listen_id = cm_id->context;
1298 if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1299 return -EINVAL;
1300
1301 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1302 return -ECONNABORTED;
1303
1304 memset(&event, 0, sizeof event);
1305 offset = cma_user_data_offset(listen_id);
1306 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1307 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1308 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1309 event.param.ud.private_data = ib_event->private_data + offset;
1310 event.param.ud.private_data_len =
1311 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1312 } else {
1313 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1314 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1315 ib_event->private_data, offset);
1316 }
1317 if (!conn_id) {
1318 ret = -ENOMEM;
1319 goto err1;
1320 }
1321
1322 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1323 ret = cma_acquire_dev(conn_id, listen_id);
1324 if (ret)
1325 goto err2;
1326
1327 conn_id->cm_id.ib = cm_id;
1328 cm_id->context = conn_id;
1329 cm_id->cm_handler = cma_ib_handler;
1330
1331 /*
1332 * Protect against the user destroying conn_id from another thread
1333 * until we're done accessing it.
1334 */
1335 atomic_inc(&conn_id->refcount);
1336 ret = conn_id->id.event_handler(&conn_id->id, &event);
1337 if (ret)
1338 goto err3;
1339
1340 if (is_iboe) {
1341 if (ib_event->param.req_rcvd.primary_path != NULL)
1342 rdma_addr_find_smac_by_sgid(
1343 &ib_event->param.req_rcvd.primary_path->sgid,
1344 psmac, NULL);
1345 else
1346 psmac = NULL;
1347 if (ib_event->param.req_rcvd.alternate_path != NULL)
1348 rdma_addr_find_smac_by_sgid(
1349 &ib_event->param.req_rcvd.alternate_path->sgid,
1350 palt_smac, NULL);
1351 else
1352 palt_smac = NULL;
1353 }
1354 /*
1355 * Acquire mutex to prevent user executing rdma_destroy_id()
1356 * while we're accessing the cm_id.
1357 */
1358 mutex_lock(&lock);
1359 if (is_iboe)
1360 ib_update_cm_av(cm_id, psmac, palt_smac);
1361 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1362 (conn_id->id.qp_type != IB_QPT_UD))
1363 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1364 mutex_unlock(&lock);
1365 mutex_unlock(&conn_id->handler_mutex);
1366 mutex_unlock(&listen_id->handler_mutex);
1367 cma_deref_id(conn_id);
1368 return 0;
1369
1370 err3:
1371 cma_deref_id(conn_id);
1372 /* Destroy the CM ID by returning a non-zero value. */
1373 conn_id->cm_id.ib = NULL;
1374 err2:
1375 cma_exch(conn_id, RDMA_CM_DESTROYING);
1376 mutex_unlock(&conn_id->handler_mutex);
1377 err1:
1378 mutex_unlock(&listen_id->handler_mutex);
1379 if (conn_id)
1380 rdma_destroy_id(&conn_id->id);
1381 return ret;
1382 }
1383
1384 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1385 {
1386 if (addr->sa_family == AF_IB)
1387 return ((struct sockaddr_ib *) addr)->sib_sid;
1388
1389 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1390 }
1391 EXPORT_SYMBOL(rdma_get_service_id);
1392
1393 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1394 struct ib_cm_compare_data *compare)
1395 {
1396 struct cma_hdr *cma_data, *cma_mask;
1397 __be32 ip4_addr;
1398 struct in6_addr ip6_addr;
1399
1400 memset(compare, 0, sizeof *compare);
1401 cma_data = (void *) compare->data;
1402 cma_mask = (void *) compare->mask;
1403
1404 switch (addr->sa_family) {
1405 case AF_INET:
1406 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1407 cma_set_ip_ver(cma_data, 4);
1408 cma_set_ip_ver(cma_mask, 0xF);
1409 if (!cma_any_addr(addr)) {
1410 cma_data->dst_addr.ip4.addr = ip4_addr;
1411 cma_mask->dst_addr.ip4.addr = htonl(~0);
1412 }
1413 break;
1414 case AF_INET6:
1415 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1416 cma_set_ip_ver(cma_data, 6);
1417 cma_set_ip_ver(cma_mask, 0xF);
1418 if (!cma_any_addr(addr)) {
1419 cma_data->dst_addr.ip6 = ip6_addr;
1420 memset(&cma_mask->dst_addr.ip6, 0xFF,
1421 sizeof cma_mask->dst_addr.ip6);
1422 }
1423 break;
1424 default:
1425 break;
1426 }
1427 }
1428
1429 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1430 {
1431 struct rdma_id_private *id_priv = iw_id->context;
1432 struct rdma_cm_event event;
1433 int ret = 0;
1434 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1435 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1436
1437 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1438 return 0;
1439
1440 memset(&event, 0, sizeof event);
1441 switch (iw_event->event) {
1442 case IW_CM_EVENT_CLOSE:
1443 event.event = RDMA_CM_EVENT_DISCONNECTED;
1444 break;
1445 case IW_CM_EVENT_CONNECT_REPLY:
1446 memcpy(cma_src_addr(id_priv), laddr,
1447 rdma_addr_size(laddr));
1448 memcpy(cma_dst_addr(id_priv), raddr,
1449 rdma_addr_size(raddr));
1450 switch (iw_event->status) {
1451 case 0:
1452 event.event = RDMA_CM_EVENT_ESTABLISHED;
1453 event.param.conn.initiator_depth = iw_event->ird;
1454 event.param.conn.responder_resources = iw_event->ord;
1455 break;
1456 case -ECONNRESET:
1457 case -ECONNREFUSED:
1458 event.event = RDMA_CM_EVENT_REJECTED;
1459 break;
1460 case -ETIMEDOUT:
1461 event.event = RDMA_CM_EVENT_UNREACHABLE;
1462 break;
1463 default:
1464 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1465 break;
1466 }
1467 break;
1468 case IW_CM_EVENT_ESTABLISHED:
1469 event.event = RDMA_CM_EVENT_ESTABLISHED;
1470 event.param.conn.initiator_depth = iw_event->ird;
1471 event.param.conn.responder_resources = iw_event->ord;
1472 break;
1473 default:
1474 BUG_ON(1);
1475 }
1476
1477 event.status = iw_event->status;
1478 event.param.conn.private_data = iw_event->private_data;
1479 event.param.conn.private_data_len = iw_event->private_data_len;
1480 ret = id_priv->id.event_handler(&id_priv->id, &event);
1481 if (ret) {
1482 /* Destroy the CM ID by returning a non-zero value. */
1483 id_priv->cm_id.iw = NULL;
1484 cma_exch(id_priv, RDMA_CM_DESTROYING);
1485 mutex_unlock(&id_priv->handler_mutex);
1486 rdma_destroy_id(&id_priv->id);
1487 return ret;
1488 }
1489
1490 mutex_unlock(&id_priv->handler_mutex);
1491 return ret;
1492 }
1493
1494 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1495 struct iw_cm_event *iw_event)
1496 {
1497 struct rdma_cm_id *new_cm_id;
1498 struct rdma_id_private *listen_id, *conn_id;
1499 struct rdma_cm_event event;
1500 int ret;
1501 struct ib_device_attr attr;
1502 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1503 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1504
1505 listen_id = cm_id->context;
1506 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1507 return -ECONNABORTED;
1508
1509 /* Create a new RDMA id for the new IW CM ID */
1510 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1511 listen_id->id.context,
1512 RDMA_PS_TCP, IB_QPT_RC);
1513 if (IS_ERR(new_cm_id)) {
1514 ret = -ENOMEM;
1515 goto out;
1516 }
1517 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1518 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1519 conn_id->state = RDMA_CM_CONNECT;
1520
1521 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1522 if (ret) {
1523 mutex_unlock(&conn_id->handler_mutex);
1524 rdma_destroy_id(new_cm_id);
1525 goto out;
1526 }
1527
1528 ret = cma_acquire_dev(conn_id, listen_id);
1529 if (ret) {
1530 mutex_unlock(&conn_id->handler_mutex);
1531 rdma_destroy_id(new_cm_id);
1532 goto out;
1533 }
1534
1535 conn_id->cm_id.iw = cm_id;
1536 cm_id->context = conn_id;
1537 cm_id->cm_handler = cma_iw_handler;
1538
1539 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1540 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1541
1542 ret = ib_query_device(conn_id->id.device, &attr);
1543 if (ret) {
1544 mutex_unlock(&conn_id->handler_mutex);
1545 rdma_destroy_id(new_cm_id);
1546 goto out;
1547 }
1548
1549 memset(&event, 0, sizeof event);
1550 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1551 event.param.conn.private_data = iw_event->private_data;
1552 event.param.conn.private_data_len = iw_event->private_data_len;
1553 event.param.conn.initiator_depth = iw_event->ird;
1554 event.param.conn.responder_resources = iw_event->ord;
1555
1556 /*
1557 * Protect against the user destroying conn_id from another thread
1558 * until we're done accessing it.
1559 */
1560 atomic_inc(&conn_id->refcount);
1561 ret = conn_id->id.event_handler(&conn_id->id, &event);
1562 if (ret) {
1563 /* User wants to destroy the CM ID */
1564 conn_id->cm_id.iw = NULL;
1565 cma_exch(conn_id, RDMA_CM_DESTROYING);
1566 mutex_unlock(&conn_id->handler_mutex);
1567 cma_deref_id(conn_id);
1568 rdma_destroy_id(&conn_id->id);
1569 goto out;
1570 }
1571
1572 mutex_unlock(&conn_id->handler_mutex);
1573 cma_deref_id(conn_id);
1574
1575 out:
1576 mutex_unlock(&listen_id->handler_mutex);
1577 return ret;
1578 }
1579
1580 static int cma_ib_listen(struct rdma_id_private *id_priv)
1581 {
1582 struct ib_cm_compare_data compare_data;
1583 struct sockaddr *addr;
1584 struct ib_cm_id *id;
1585 __be64 svc_id;
1586 int ret;
1587
1588 id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1589 if (IS_ERR(id))
1590 return PTR_ERR(id);
1591
1592 id_priv->cm_id.ib = id;
1593
1594 addr = cma_src_addr(id_priv);
1595 svc_id = rdma_get_service_id(&id_priv->id, addr);
1596 if (cma_any_addr(addr) && !id_priv->afonly)
1597 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1598 else {
1599 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1600 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1601 }
1602
1603 if (ret) {
1604 ib_destroy_cm_id(id_priv->cm_id.ib);
1605 id_priv->cm_id.ib = NULL;
1606 }
1607
1608 return ret;
1609 }
1610
1611 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1612 {
1613 int ret;
1614 struct iw_cm_id *id;
1615
1616 id = iw_create_cm_id(id_priv->id.device,
1617 iw_conn_req_handler,
1618 id_priv);
1619 if (IS_ERR(id))
1620 return PTR_ERR(id);
1621
1622 id_priv->cm_id.iw = id;
1623
1624 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
1625 rdma_addr_size(cma_src_addr(id_priv)));
1626
1627 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1628
1629 if (ret) {
1630 iw_destroy_cm_id(id_priv->cm_id.iw);
1631 id_priv->cm_id.iw = NULL;
1632 }
1633
1634 return ret;
1635 }
1636
1637 static int cma_listen_handler(struct rdma_cm_id *id,
1638 struct rdma_cm_event *event)
1639 {
1640 struct rdma_id_private *id_priv = id->context;
1641
1642 id->context = id_priv->id.context;
1643 id->event_handler = id_priv->id.event_handler;
1644 return id_priv->id.event_handler(id, event);
1645 }
1646
1647 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1648 struct cma_device *cma_dev)
1649 {
1650 struct rdma_id_private *dev_id_priv;
1651 struct rdma_cm_id *id;
1652 int ret;
1653
1654 if (cma_family(id_priv) == AF_IB &&
1655 rdma_node_get_transport(cma_dev->device->node_type) != RDMA_TRANSPORT_IB)
1656 return;
1657
1658 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1659 id_priv->id.qp_type);
1660 if (IS_ERR(id))
1661 return;
1662
1663 dev_id_priv = container_of(id, struct rdma_id_private, id);
1664
1665 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
1666 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1667 rdma_addr_size(cma_src_addr(id_priv)));
1668
1669 cma_attach_to_dev(dev_id_priv, cma_dev);
1670 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1671 atomic_inc(&id_priv->refcount);
1672 dev_id_priv->internal_id = 1;
1673 dev_id_priv->afonly = id_priv->afonly;
1674
1675 ret = rdma_listen(id, id_priv->backlog);
1676 if (ret)
1677 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1678 "listening on device %s\n", ret, cma_dev->device->name);
1679 }
1680
1681 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1682 {
1683 struct cma_device *cma_dev;
1684
1685 mutex_lock(&lock);
1686 list_add_tail(&id_priv->list, &listen_any_list);
1687 list_for_each_entry(cma_dev, &dev_list, list)
1688 cma_listen_on_dev(id_priv, cma_dev);
1689 mutex_unlock(&lock);
1690 }
1691
1692 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1693 {
1694 struct rdma_id_private *id_priv;
1695
1696 id_priv = container_of(id, struct rdma_id_private, id);
1697 id_priv->tos = (u8) tos;
1698 }
1699 EXPORT_SYMBOL(rdma_set_service_type);
1700
1701 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1702 void *context)
1703 {
1704 struct cma_work *work = context;
1705 struct rdma_route *route;
1706
1707 route = &work->id->id.route;
1708
1709 if (!status) {
1710 route->num_paths = 1;
1711 *route->path_rec = *path_rec;
1712 } else {
1713 work->old_state = RDMA_CM_ROUTE_QUERY;
1714 work->new_state = RDMA_CM_ADDR_RESOLVED;
1715 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1716 work->event.status = status;
1717 }
1718
1719 queue_work(cma_wq, &work->work);
1720 }
1721
1722 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1723 struct cma_work *work)
1724 {
1725 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1726 struct ib_sa_path_rec path_rec;
1727 ib_sa_comp_mask comp_mask;
1728 struct sockaddr_in6 *sin6;
1729 struct sockaddr_ib *sib;
1730
1731 memset(&path_rec, 0, sizeof path_rec);
1732 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1733 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1734 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
1735 path_rec.numb_path = 1;
1736 path_rec.reversible = 1;
1737 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
1738
1739 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1740 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1741 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1742
1743 switch (cma_family(id_priv)) {
1744 case AF_INET:
1745 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1746 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1747 break;
1748 case AF_INET6:
1749 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
1750 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1751 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1752 break;
1753 case AF_IB:
1754 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
1755 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
1756 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1757 break;
1758 }
1759
1760 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1761 id_priv->id.port_num, &path_rec,
1762 comp_mask, timeout_ms,
1763 GFP_KERNEL, cma_query_handler,
1764 work, &id_priv->query);
1765
1766 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1767 }
1768
1769 static void cma_work_handler(struct work_struct *_work)
1770 {
1771 struct cma_work *work = container_of(_work, struct cma_work, work);
1772 struct rdma_id_private *id_priv = work->id;
1773 int destroy = 0;
1774
1775 mutex_lock(&id_priv->handler_mutex);
1776 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1777 goto out;
1778
1779 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1780 cma_exch(id_priv, RDMA_CM_DESTROYING);
1781 destroy = 1;
1782 }
1783 out:
1784 mutex_unlock(&id_priv->handler_mutex);
1785 cma_deref_id(id_priv);
1786 if (destroy)
1787 rdma_destroy_id(&id_priv->id);
1788 kfree(work);
1789 }
1790
1791 static void cma_ndev_work_handler(struct work_struct *_work)
1792 {
1793 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1794 struct rdma_id_private *id_priv = work->id;
1795 int destroy = 0;
1796
1797 mutex_lock(&id_priv->handler_mutex);
1798 if (id_priv->state == RDMA_CM_DESTROYING ||
1799 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
1800 goto out;
1801
1802 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1803 cma_exch(id_priv, RDMA_CM_DESTROYING);
1804 destroy = 1;
1805 }
1806
1807 out:
1808 mutex_unlock(&id_priv->handler_mutex);
1809 cma_deref_id(id_priv);
1810 if (destroy)
1811 rdma_destroy_id(&id_priv->id);
1812 kfree(work);
1813 }
1814
1815 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1816 {
1817 struct rdma_route *route = &id_priv->id.route;
1818 struct cma_work *work;
1819 int ret;
1820
1821 work = kzalloc(sizeof *work, GFP_KERNEL);
1822 if (!work)
1823 return -ENOMEM;
1824
1825 work->id = id_priv;
1826 INIT_WORK(&work->work, cma_work_handler);
1827 work->old_state = RDMA_CM_ROUTE_QUERY;
1828 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1829 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1830
1831 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1832 if (!route->path_rec) {
1833 ret = -ENOMEM;
1834 goto err1;
1835 }
1836
1837 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1838 if (ret)
1839 goto err2;
1840
1841 return 0;
1842 err2:
1843 kfree(route->path_rec);
1844 route->path_rec = NULL;
1845 err1:
1846 kfree(work);
1847 return ret;
1848 }
1849
1850 int rdma_set_ib_paths(struct rdma_cm_id *id,
1851 struct ib_sa_path_rec *path_rec, int num_paths)
1852 {
1853 struct rdma_id_private *id_priv;
1854 int ret;
1855
1856 id_priv = container_of(id, struct rdma_id_private, id);
1857 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1858 RDMA_CM_ROUTE_RESOLVED))
1859 return -EINVAL;
1860
1861 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1862 GFP_KERNEL);
1863 if (!id->route.path_rec) {
1864 ret = -ENOMEM;
1865 goto err;
1866 }
1867
1868 id->route.num_paths = num_paths;
1869 return 0;
1870 err:
1871 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
1872 return ret;
1873 }
1874 EXPORT_SYMBOL(rdma_set_ib_paths);
1875
1876 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1877 {
1878 struct cma_work *work;
1879
1880 work = kzalloc(sizeof *work, GFP_KERNEL);
1881 if (!work)
1882 return -ENOMEM;
1883
1884 work->id = id_priv;
1885 INIT_WORK(&work->work, cma_work_handler);
1886 work->old_state = RDMA_CM_ROUTE_QUERY;
1887 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1888 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1889 queue_work(cma_wq, &work->work);
1890 return 0;
1891 }
1892
1893 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
1894 {
1895 int prio;
1896 struct net_device *dev;
1897
1898 prio = rt_tos2priority(tos);
1899 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
1900 vlan_dev_real_dev(ndev) : ndev;
1901
1902 if (dev->num_tc)
1903 return netdev_get_prio_tc_map(dev, prio);
1904
1905 #if IS_ENABLED(CONFIG_VLAN_8021Q)
1906 if (ndev->priv_flags & IFF_802_1Q_VLAN)
1907 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
1908 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1909 #endif
1910 return 0;
1911 }
1912
1913 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1914 {
1915 struct rdma_route *route = &id_priv->id.route;
1916 struct rdma_addr *addr = &route->addr;
1917 struct cma_work *work;
1918 int ret;
1919 struct net_device *ndev = NULL;
1920
1921
1922 work = kzalloc(sizeof *work, GFP_KERNEL);
1923 if (!work)
1924 return -ENOMEM;
1925
1926 work->id = id_priv;
1927 INIT_WORK(&work->work, cma_work_handler);
1928
1929 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1930 if (!route->path_rec) {
1931 ret = -ENOMEM;
1932 goto err1;
1933 }
1934
1935 route->num_paths = 1;
1936
1937 if (addr->dev_addr.bound_dev_if)
1938 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1939 if (!ndev) {
1940 ret = -ENODEV;
1941 goto err2;
1942 }
1943
1944 route->path_rec->vlan_id = rdma_vlan_dev_vlan_id(ndev);
1945 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
1946 memcpy(route->path_rec->smac, ndev->dev_addr, ndev->addr_len);
1947
1948 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
1949 &route->path_rec->sgid);
1950 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
1951 &route->path_rec->dgid);
1952
1953 route->path_rec->hop_limit = 1;
1954 route->path_rec->reversible = 1;
1955 route->path_rec->pkey = cpu_to_be16(0xffff);
1956 route->path_rec->mtu_selector = IB_SA_EQ;
1957 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
1958 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1959 route->path_rec->rate_selector = IB_SA_EQ;
1960 route->path_rec->rate = iboe_get_rate(ndev);
1961 dev_put(ndev);
1962 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1963 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1964 if (!route->path_rec->mtu) {
1965 ret = -EINVAL;
1966 goto err2;
1967 }
1968
1969 work->old_state = RDMA_CM_ROUTE_QUERY;
1970 work->new_state = RDMA_CM_ROUTE_RESOLVED;
1971 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1972 work->event.status = 0;
1973
1974 queue_work(cma_wq, &work->work);
1975
1976 return 0;
1977
1978 err2:
1979 kfree(route->path_rec);
1980 route->path_rec = NULL;
1981 err1:
1982 kfree(work);
1983 return ret;
1984 }
1985
1986 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1987 {
1988 struct rdma_id_private *id_priv;
1989 int ret;
1990
1991 id_priv = container_of(id, struct rdma_id_private, id);
1992 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
1993 return -EINVAL;
1994
1995 atomic_inc(&id_priv->refcount);
1996 switch (rdma_node_get_transport(id->device->node_type)) {
1997 case RDMA_TRANSPORT_IB:
1998 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1999 case IB_LINK_LAYER_INFINIBAND:
2000 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2001 break;
2002 case IB_LINK_LAYER_ETHERNET:
2003 ret = cma_resolve_iboe_route(id_priv);
2004 break;
2005 default:
2006 ret = -ENOSYS;
2007 }
2008 break;
2009 case RDMA_TRANSPORT_IWARP:
2010 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2011 break;
2012 default:
2013 ret = -ENOSYS;
2014 break;
2015 }
2016 if (ret)
2017 goto err;
2018
2019 return 0;
2020 err:
2021 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2022 cma_deref_id(id_priv);
2023 return ret;
2024 }
2025 EXPORT_SYMBOL(rdma_resolve_route);
2026
2027 static void cma_set_loopback(struct sockaddr *addr)
2028 {
2029 switch (addr->sa_family) {
2030 case AF_INET:
2031 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2032 break;
2033 case AF_INET6:
2034 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2035 0, 0, 0, htonl(1));
2036 break;
2037 default:
2038 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2039 0, 0, 0, htonl(1));
2040 break;
2041 }
2042 }
2043
2044 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2045 {
2046 struct cma_device *cma_dev, *cur_dev;
2047 struct ib_port_attr port_attr;
2048 union ib_gid gid;
2049 u16 pkey;
2050 int ret;
2051 u8 p;
2052
2053 cma_dev = NULL;
2054 mutex_lock(&lock);
2055 list_for_each_entry(cur_dev, &dev_list, list) {
2056 if (cma_family(id_priv) == AF_IB &&
2057 rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
2058 continue;
2059
2060 if (!cma_dev)
2061 cma_dev = cur_dev;
2062
2063 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2064 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2065 port_attr.state == IB_PORT_ACTIVE) {
2066 cma_dev = cur_dev;
2067 goto port_found;
2068 }
2069 }
2070 }
2071
2072 if (!cma_dev) {
2073 ret = -ENODEV;
2074 goto out;
2075 }
2076
2077 p = 1;
2078
2079 port_found:
2080 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2081 if (ret)
2082 goto out;
2083
2084 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2085 if (ret)
2086 goto out;
2087
2088 id_priv->id.route.addr.dev_addr.dev_type =
2089 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
2090 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2091
2092 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2093 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2094 id_priv->id.port_num = p;
2095 cma_attach_to_dev(id_priv, cma_dev);
2096 cma_set_loopback(cma_src_addr(id_priv));
2097 out:
2098 mutex_unlock(&lock);
2099 return ret;
2100 }
2101
2102 static void addr_handler(int status, struct sockaddr *src_addr,
2103 struct rdma_dev_addr *dev_addr, void *context)
2104 {
2105 struct rdma_id_private *id_priv = context;
2106 struct rdma_cm_event event;
2107
2108 memset(&event, 0, sizeof event);
2109 mutex_lock(&id_priv->handler_mutex);
2110 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2111 RDMA_CM_ADDR_RESOLVED))
2112 goto out;
2113
2114 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2115 if (!status && !id_priv->cma_dev)
2116 status = cma_acquire_dev(id_priv, NULL);
2117
2118 if (status) {
2119 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2120 RDMA_CM_ADDR_BOUND))
2121 goto out;
2122 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2123 event.status = status;
2124 } else
2125 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2126
2127 if (id_priv->id.event_handler(&id_priv->id, &event)) {
2128 cma_exch(id_priv, RDMA_CM_DESTROYING);
2129 mutex_unlock(&id_priv->handler_mutex);
2130 cma_deref_id(id_priv);
2131 rdma_destroy_id(&id_priv->id);
2132 return;
2133 }
2134 out:
2135 mutex_unlock(&id_priv->handler_mutex);
2136 cma_deref_id(id_priv);
2137 }
2138
2139 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2140 {
2141 struct cma_work *work;
2142 union ib_gid gid;
2143 int ret;
2144
2145 work = kzalloc(sizeof *work, GFP_KERNEL);
2146 if (!work)
2147 return -ENOMEM;
2148
2149 if (!id_priv->cma_dev) {
2150 ret = cma_bind_loopback(id_priv);
2151 if (ret)
2152 goto err;
2153 }
2154
2155 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2156 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2157
2158 work->id = id_priv;
2159 INIT_WORK(&work->work, cma_work_handler);
2160 work->old_state = RDMA_CM_ADDR_QUERY;
2161 work->new_state = RDMA_CM_ADDR_RESOLVED;
2162 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2163 queue_work(cma_wq, &work->work);
2164 return 0;
2165 err:
2166 kfree(work);
2167 return ret;
2168 }
2169
2170 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2171 {
2172 struct cma_work *work;
2173 int ret;
2174
2175 work = kzalloc(sizeof *work, GFP_KERNEL);
2176 if (!work)
2177 return -ENOMEM;
2178
2179 if (!id_priv->cma_dev) {
2180 ret = cma_resolve_ib_dev(id_priv);
2181 if (ret)
2182 goto err;
2183 }
2184
2185 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2186 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2187
2188 work->id = id_priv;
2189 INIT_WORK(&work->work, cma_work_handler);
2190 work->old_state = RDMA_CM_ADDR_QUERY;
2191 work->new_state = RDMA_CM_ADDR_RESOLVED;
2192 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2193 queue_work(cma_wq, &work->work);
2194 return 0;
2195 err:
2196 kfree(work);
2197 return ret;
2198 }
2199
2200 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2201 struct sockaddr *dst_addr)
2202 {
2203 if (!src_addr || !src_addr->sa_family) {
2204 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2205 src_addr->sa_family = dst_addr->sa_family;
2206 if (dst_addr->sa_family == AF_INET6) {
2207 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2208 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2209 } else if (dst_addr->sa_family == AF_IB) {
2210 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2211 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2212 }
2213 }
2214 return rdma_bind_addr(id, src_addr);
2215 }
2216
2217 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2218 struct sockaddr *dst_addr, int timeout_ms)
2219 {
2220 struct rdma_id_private *id_priv;
2221 int ret;
2222
2223 id_priv = container_of(id, struct rdma_id_private, id);
2224 if (id_priv->state == RDMA_CM_IDLE) {
2225 ret = cma_bind_addr(id, src_addr, dst_addr);
2226 if (ret)
2227 return ret;
2228 }
2229
2230 if (cma_family(id_priv) != dst_addr->sa_family)
2231 return -EINVAL;
2232
2233 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2234 return -EINVAL;
2235
2236 atomic_inc(&id_priv->refcount);
2237 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2238 if (cma_any_addr(dst_addr)) {
2239 ret = cma_resolve_loopback(id_priv);
2240 } else {
2241 if (dst_addr->sa_family == AF_IB) {
2242 ret = cma_resolve_ib_addr(id_priv);
2243 } else {
2244 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2245 dst_addr, &id->route.addr.dev_addr,
2246 timeout_ms, addr_handler, id_priv);
2247 }
2248 }
2249 if (ret)
2250 goto err;
2251
2252 return 0;
2253 err:
2254 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2255 cma_deref_id(id_priv);
2256 return ret;
2257 }
2258 EXPORT_SYMBOL(rdma_resolve_addr);
2259
2260 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2261 {
2262 struct rdma_id_private *id_priv;
2263 unsigned long flags;
2264 int ret;
2265
2266 id_priv = container_of(id, struct rdma_id_private, id);
2267 spin_lock_irqsave(&id_priv->lock, flags);
2268 if (reuse || id_priv->state == RDMA_CM_IDLE) {
2269 id_priv->reuseaddr = reuse;
2270 ret = 0;
2271 } else {
2272 ret = -EINVAL;
2273 }
2274 spin_unlock_irqrestore(&id_priv->lock, flags);
2275 return ret;
2276 }
2277 EXPORT_SYMBOL(rdma_set_reuseaddr);
2278
2279 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2280 {
2281 struct rdma_id_private *id_priv;
2282 unsigned long flags;
2283 int ret;
2284
2285 id_priv = container_of(id, struct rdma_id_private, id);
2286 spin_lock_irqsave(&id_priv->lock, flags);
2287 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2288 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2289 id_priv->afonly = afonly;
2290 ret = 0;
2291 } else {
2292 ret = -EINVAL;
2293 }
2294 spin_unlock_irqrestore(&id_priv->lock, flags);
2295 return ret;
2296 }
2297 EXPORT_SYMBOL(rdma_set_afonly);
2298
2299 static void cma_bind_port(struct rdma_bind_list *bind_list,
2300 struct rdma_id_private *id_priv)
2301 {
2302 struct sockaddr *addr;
2303 struct sockaddr_ib *sib;
2304 u64 sid, mask;
2305 __be16 port;
2306
2307 addr = cma_src_addr(id_priv);
2308 port = htons(bind_list->port);
2309
2310 switch (addr->sa_family) {
2311 case AF_INET:
2312 ((struct sockaddr_in *) addr)->sin_port = port;
2313 break;
2314 case AF_INET6:
2315 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2316 break;
2317 case AF_IB:
2318 sib = (struct sockaddr_ib *) addr;
2319 sid = be64_to_cpu(sib->sib_sid);
2320 mask = be64_to_cpu(sib->sib_sid_mask);
2321 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2322 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2323 break;
2324 }
2325 id_priv->bind_list = bind_list;
2326 hlist_add_head(&id_priv->node, &bind_list->owners);
2327 }
2328
2329 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2330 unsigned short snum)
2331 {
2332 struct rdma_bind_list *bind_list;
2333 int ret;
2334
2335 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2336 if (!bind_list)
2337 return -ENOMEM;
2338
2339 ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2340 if (ret < 0)
2341 goto err;
2342
2343 bind_list->ps = ps;
2344 bind_list->port = (unsigned short)ret;
2345 cma_bind_port(bind_list, id_priv);
2346 return 0;
2347 err:
2348 kfree(bind_list);
2349 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2350 }
2351
2352 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2353 {
2354 static unsigned int last_used_port;
2355 int low, high, remaining;
2356 unsigned int rover;
2357
2358 inet_get_local_port_range(&init_net, &low, &high);
2359 remaining = (high - low) + 1;
2360 rover = prandom_u32() % remaining + low;
2361 retry:
2362 if (last_used_port != rover &&
2363 !idr_find(ps, (unsigned short) rover)) {
2364 int ret = cma_alloc_port(ps, id_priv, rover);
2365 /*
2366 * Remember previously used port number in order to avoid
2367 * re-using same port immediately after it is closed.
2368 */
2369 if (!ret)
2370 last_used_port = rover;
2371 if (ret != -EADDRNOTAVAIL)
2372 return ret;
2373 }
2374 if (--remaining) {
2375 rover++;
2376 if ((rover < low) || (rover > high))
2377 rover = low;
2378 goto retry;
2379 }
2380 return -EADDRNOTAVAIL;
2381 }
2382
2383 /*
2384 * Check that the requested port is available. This is called when trying to
2385 * bind to a specific port, or when trying to listen on a bound port. In
2386 * the latter case, the provided id_priv may already be on the bind_list, but
2387 * we still need to check that it's okay to start listening.
2388 */
2389 static int cma_check_port(struct rdma_bind_list *bind_list,
2390 struct rdma_id_private *id_priv, uint8_t reuseaddr)
2391 {
2392 struct rdma_id_private *cur_id;
2393 struct sockaddr *addr, *cur_addr;
2394
2395 addr = cma_src_addr(id_priv);
2396 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2397 if (id_priv == cur_id)
2398 continue;
2399
2400 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2401 cur_id->reuseaddr)
2402 continue;
2403
2404 cur_addr = cma_src_addr(cur_id);
2405 if (id_priv->afonly && cur_id->afonly &&
2406 (addr->sa_family != cur_addr->sa_family))
2407 continue;
2408
2409 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2410 return -EADDRNOTAVAIL;
2411
2412 if (!cma_addr_cmp(addr, cur_addr))
2413 return -EADDRINUSE;
2414 }
2415 return 0;
2416 }
2417
2418 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2419 {
2420 struct rdma_bind_list *bind_list;
2421 unsigned short snum;
2422 int ret;
2423
2424 snum = ntohs(cma_port(cma_src_addr(id_priv)));
2425 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2426 return -EACCES;
2427
2428 bind_list = idr_find(ps, snum);
2429 if (!bind_list) {
2430 ret = cma_alloc_port(ps, id_priv, snum);
2431 } else {
2432 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2433 if (!ret)
2434 cma_bind_port(bind_list, id_priv);
2435 }
2436 return ret;
2437 }
2438
2439 static int cma_bind_listen(struct rdma_id_private *id_priv)
2440 {
2441 struct rdma_bind_list *bind_list = id_priv->bind_list;
2442 int ret = 0;
2443
2444 mutex_lock(&lock);
2445 if (bind_list->owners.first->next)
2446 ret = cma_check_port(bind_list, id_priv, 0);
2447 mutex_unlock(&lock);
2448 return ret;
2449 }
2450
2451 static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
2452 {
2453 switch (id_priv->id.ps) {
2454 case RDMA_PS_TCP:
2455 return &tcp_ps;
2456 case RDMA_PS_UDP:
2457 return &udp_ps;
2458 case RDMA_PS_IPOIB:
2459 return &ipoib_ps;
2460 case RDMA_PS_IB:
2461 return &ib_ps;
2462 default:
2463 return NULL;
2464 }
2465 }
2466
2467 static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2468 {
2469 struct idr *ps = NULL;
2470 struct sockaddr_ib *sib;
2471 u64 sid_ps, mask, sid;
2472
2473 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2474 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2475 sid = be64_to_cpu(sib->sib_sid) & mask;
2476
2477 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2478 sid_ps = RDMA_IB_IP_PS_IB;
2479 ps = &ib_ps;
2480 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2481 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2482 sid_ps = RDMA_IB_IP_PS_TCP;
2483 ps = &tcp_ps;
2484 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2485 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2486 sid_ps = RDMA_IB_IP_PS_UDP;
2487 ps = &udp_ps;
2488 }
2489
2490 if (ps) {
2491 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2492 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2493 be64_to_cpu(sib->sib_sid_mask));
2494 }
2495 return ps;
2496 }
2497
2498 static int cma_get_port(struct rdma_id_private *id_priv)
2499 {
2500 struct idr *ps;
2501 int ret;
2502
2503 if (cma_family(id_priv) != AF_IB)
2504 ps = cma_select_inet_ps(id_priv);
2505 else
2506 ps = cma_select_ib_ps(id_priv);
2507 if (!ps)
2508 return -EPROTONOSUPPORT;
2509
2510 mutex_lock(&lock);
2511 if (cma_any_port(cma_src_addr(id_priv)))
2512 ret = cma_alloc_any_port(ps, id_priv);
2513 else
2514 ret = cma_use_port(ps, id_priv);
2515 mutex_unlock(&lock);
2516
2517 return ret;
2518 }
2519
2520 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2521 struct sockaddr *addr)
2522 {
2523 #if IS_ENABLED(CONFIG_IPV6)
2524 struct sockaddr_in6 *sin6;
2525
2526 if (addr->sa_family != AF_INET6)
2527 return 0;
2528
2529 sin6 = (struct sockaddr_in6 *) addr;
2530
2531 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2532 return 0;
2533
2534 if (!sin6->sin6_scope_id)
2535 return -EINVAL;
2536
2537 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2538 #endif
2539 return 0;
2540 }
2541
2542 int rdma_listen(struct rdma_cm_id *id, int backlog)
2543 {
2544 struct rdma_id_private *id_priv;
2545 int ret;
2546
2547 id_priv = container_of(id, struct rdma_id_private, id);
2548 if (id_priv->state == RDMA_CM_IDLE) {
2549 id->route.addr.src_addr.ss_family = AF_INET;
2550 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2551 if (ret)
2552 return ret;
2553 }
2554
2555 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2556 return -EINVAL;
2557
2558 if (id_priv->reuseaddr) {
2559 ret = cma_bind_listen(id_priv);
2560 if (ret)
2561 goto err;
2562 }
2563
2564 id_priv->backlog = backlog;
2565 if (id->device) {
2566 switch (rdma_node_get_transport(id->device->node_type)) {
2567 case RDMA_TRANSPORT_IB:
2568 ret = cma_ib_listen(id_priv);
2569 if (ret)
2570 goto err;
2571 break;
2572 case RDMA_TRANSPORT_IWARP:
2573 ret = cma_iw_listen(id_priv, backlog);
2574 if (ret)
2575 goto err;
2576 break;
2577 default:
2578 ret = -ENOSYS;
2579 goto err;
2580 }
2581 } else
2582 cma_listen_on_all(id_priv);
2583
2584 return 0;
2585 err:
2586 id_priv->backlog = 0;
2587 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2588 return ret;
2589 }
2590 EXPORT_SYMBOL(rdma_listen);
2591
2592 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2593 {
2594 struct rdma_id_private *id_priv;
2595 int ret;
2596
2597 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2598 addr->sa_family != AF_IB)
2599 return -EAFNOSUPPORT;
2600
2601 id_priv = container_of(id, struct rdma_id_private, id);
2602 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2603 return -EINVAL;
2604
2605 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2606 if (ret)
2607 goto err1;
2608
2609 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2610 if (!cma_any_addr(addr)) {
2611 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2612 if (ret)
2613 goto err1;
2614
2615 ret = cma_acquire_dev(id_priv, NULL);
2616 if (ret)
2617 goto err1;
2618 }
2619
2620 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2621 if (addr->sa_family == AF_INET)
2622 id_priv->afonly = 1;
2623 #if IS_ENABLED(CONFIG_IPV6)
2624 else if (addr->sa_family == AF_INET6)
2625 id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2626 #endif
2627 }
2628 ret = cma_get_port(id_priv);
2629 if (ret)
2630 goto err2;
2631
2632 return 0;
2633 err2:
2634 if (id_priv->cma_dev)
2635 cma_release_dev(id_priv);
2636 err1:
2637 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2638 return ret;
2639 }
2640 EXPORT_SYMBOL(rdma_bind_addr);
2641
2642 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2643 {
2644 struct cma_hdr *cma_hdr;
2645
2646 cma_hdr = hdr;
2647 cma_hdr->cma_version = CMA_VERSION;
2648 if (cma_family(id_priv) == AF_INET) {
2649 struct sockaddr_in *src4, *dst4;
2650
2651 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2652 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2653
2654 cma_set_ip_ver(cma_hdr, 4);
2655 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2656 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2657 cma_hdr->port = src4->sin_port;
2658 } else if (cma_family(id_priv) == AF_INET6) {
2659 struct sockaddr_in6 *src6, *dst6;
2660
2661 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2662 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2663
2664 cma_set_ip_ver(cma_hdr, 6);
2665 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2666 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2667 cma_hdr->port = src6->sin6_port;
2668 }
2669 return 0;
2670 }
2671
2672 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2673 struct ib_cm_event *ib_event)
2674 {
2675 struct rdma_id_private *id_priv = cm_id->context;
2676 struct rdma_cm_event event;
2677 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2678 int ret = 0;
2679
2680 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
2681 return 0;
2682
2683 memset(&event, 0, sizeof event);
2684 switch (ib_event->event) {
2685 case IB_CM_SIDR_REQ_ERROR:
2686 event.event = RDMA_CM_EVENT_UNREACHABLE;
2687 event.status = -ETIMEDOUT;
2688 break;
2689 case IB_CM_SIDR_REP_RECEIVED:
2690 event.param.ud.private_data = ib_event->private_data;
2691 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2692 if (rep->status != IB_SIDR_SUCCESS) {
2693 event.event = RDMA_CM_EVENT_UNREACHABLE;
2694 event.status = ib_event->param.sidr_rep_rcvd.status;
2695 break;
2696 }
2697 ret = cma_set_qkey(id_priv, rep->qkey);
2698 if (ret) {
2699 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2700 event.status = ret;
2701 break;
2702 }
2703 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2704 id_priv->id.route.path_rec,
2705 &event.param.ud.ah_attr);
2706 event.param.ud.qp_num = rep->qpn;
2707 event.param.ud.qkey = rep->qkey;
2708 event.event = RDMA_CM_EVENT_ESTABLISHED;
2709 event.status = 0;
2710 break;
2711 default:
2712 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2713 ib_event->event);
2714 goto out;
2715 }
2716
2717 ret = id_priv->id.event_handler(&id_priv->id, &event);
2718 if (ret) {
2719 /* Destroy the CM ID by returning a non-zero value. */
2720 id_priv->cm_id.ib = NULL;
2721 cma_exch(id_priv, RDMA_CM_DESTROYING);
2722 mutex_unlock(&id_priv->handler_mutex);
2723 rdma_destroy_id(&id_priv->id);
2724 return ret;
2725 }
2726 out:
2727 mutex_unlock(&id_priv->handler_mutex);
2728 return ret;
2729 }
2730
2731 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2732 struct rdma_conn_param *conn_param)
2733 {
2734 struct ib_cm_sidr_req_param req;
2735 struct ib_cm_id *id;
2736 void *private_data;
2737 int offset, ret;
2738
2739 memset(&req, 0, sizeof req);
2740 offset = cma_user_data_offset(id_priv);
2741 req.private_data_len = offset + conn_param->private_data_len;
2742 if (req.private_data_len < conn_param->private_data_len)
2743 return -EINVAL;
2744
2745 if (req.private_data_len) {
2746 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2747 if (!private_data)
2748 return -ENOMEM;
2749 } else {
2750 private_data = NULL;
2751 }
2752
2753 if (conn_param->private_data && conn_param->private_data_len)
2754 memcpy(private_data + offset, conn_param->private_data,
2755 conn_param->private_data_len);
2756
2757 if (private_data) {
2758 ret = cma_format_hdr(private_data, id_priv);
2759 if (ret)
2760 goto out;
2761 req.private_data = private_data;
2762 }
2763
2764 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2765 id_priv);
2766 if (IS_ERR(id)) {
2767 ret = PTR_ERR(id);
2768 goto out;
2769 }
2770 id_priv->cm_id.ib = id;
2771
2772 req.path = id_priv->id.route.path_rec;
2773 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2774 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2775 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2776
2777 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2778 if (ret) {
2779 ib_destroy_cm_id(id_priv->cm_id.ib);
2780 id_priv->cm_id.ib = NULL;
2781 }
2782 out:
2783 kfree(private_data);
2784 return ret;
2785 }
2786
2787 static int cma_connect_ib(struct rdma_id_private *id_priv,
2788 struct rdma_conn_param *conn_param)
2789 {
2790 struct ib_cm_req_param req;
2791 struct rdma_route *route;
2792 void *private_data;
2793 struct ib_cm_id *id;
2794 int offset, ret;
2795
2796 memset(&req, 0, sizeof req);
2797 offset = cma_user_data_offset(id_priv);
2798 req.private_data_len = offset + conn_param->private_data_len;
2799 if (req.private_data_len < conn_param->private_data_len)
2800 return -EINVAL;
2801
2802 if (req.private_data_len) {
2803 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2804 if (!private_data)
2805 return -ENOMEM;
2806 } else {
2807 private_data = NULL;
2808 }
2809
2810 if (conn_param->private_data && conn_param->private_data_len)
2811 memcpy(private_data + offset, conn_param->private_data,
2812 conn_param->private_data_len);
2813
2814 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2815 if (IS_ERR(id)) {
2816 ret = PTR_ERR(id);
2817 goto out;
2818 }
2819 id_priv->cm_id.ib = id;
2820
2821 route = &id_priv->id.route;
2822 if (private_data) {
2823 ret = cma_format_hdr(private_data, id_priv);
2824 if (ret)
2825 goto out;
2826 req.private_data = private_data;
2827 }
2828
2829 req.primary_path = &route->path_rec[0];
2830 if (route->num_paths == 2)
2831 req.alternate_path = &route->path_rec[1];
2832
2833 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2834 req.qp_num = id_priv->qp_num;
2835 req.qp_type = id_priv->id.qp_type;
2836 req.starting_psn = id_priv->seq_num;
2837 req.responder_resources = conn_param->responder_resources;
2838 req.initiator_depth = conn_param->initiator_depth;
2839 req.flow_control = conn_param->flow_control;
2840 req.retry_count = min_t(u8, 7, conn_param->retry_count);
2841 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2842 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2843 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2844 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2845 req.srq = id_priv->srq ? 1 : 0;
2846
2847 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2848 out:
2849 if (ret && !IS_ERR(id)) {
2850 ib_destroy_cm_id(id);
2851 id_priv->cm_id.ib = NULL;
2852 }
2853
2854 kfree(private_data);
2855 return ret;
2856 }
2857
2858 static int cma_connect_iw(struct rdma_id_private *id_priv,
2859 struct rdma_conn_param *conn_param)
2860 {
2861 struct iw_cm_id *cm_id;
2862 int ret;
2863 struct iw_cm_conn_param iw_param;
2864
2865 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2866 if (IS_ERR(cm_id))
2867 return PTR_ERR(cm_id);
2868
2869 id_priv->cm_id.iw = cm_id;
2870
2871 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
2872 rdma_addr_size(cma_src_addr(id_priv)));
2873 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
2874 rdma_addr_size(cma_dst_addr(id_priv)));
2875
2876 ret = cma_modify_qp_rtr(id_priv, conn_param);
2877 if (ret)
2878 goto out;
2879
2880 if (conn_param) {
2881 iw_param.ord = conn_param->initiator_depth;
2882 iw_param.ird = conn_param->responder_resources;
2883 iw_param.private_data = conn_param->private_data;
2884 iw_param.private_data_len = conn_param->private_data_len;
2885 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2886 } else {
2887 memset(&iw_param, 0, sizeof iw_param);
2888 iw_param.qpn = id_priv->qp_num;
2889 }
2890 ret = iw_cm_connect(cm_id, &iw_param);
2891 out:
2892 if (ret) {
2893 iw_destroy_cm_id(cm_id);
2894 id_priv->cm_id.iw = NULL;
2895 }
2896 return ret;
2897 }
2898
2899 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2900 {
2901 struct rdma_id_private *id_priv;
2902 int ret;
2903
2904 id_priv = container_of(id, struct rdma_id_private, id);
2905 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
2906 return -EINVAL;
2907
2908 if (!id->qp) {
2909 id_priv->qp_num = conn_param->qp_num;
2910 id_priv->srq = conn_param->srq;
2911 }
2912
2913 switch (rdma_node_get_transport(id->device->node_type)) {
2914 case RDMA_TRANSPORT_IB:
2915 if (id->qp_type == IB_QPT_UD)
2916 ret = cma_resolve_ib_udp(id_priv, conn_param);
2917 else
2918 ret = cma_connect_ib(id_priv, conn_param);
2919 break;
2920 case RDMA_TRANSPORT_IWARP:
2921 ret = cma_connect_iw(id_priv, conn_param);
2922 break;
2923 default:
2924 ret = -ENOSYS;
2925 break;
2926 }
2927 if (ret)
2928 goto err;
2929
2930 return 0;
2931 err:
2932 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
2933 return ret;
2934 }
2935 EXPORT_SYMBOL(rdma_connect);
2936
2937 static int cma_accept_ib(struct rdma_id_private *id_priv,
2938 struct rdma_conn_param *conn_param)
2939 {
2940 struct ib_cm_rep_param rep;
2941 int ret;
2942
2943 ret = cma_modify_qp_rtr(id_priv, conn_param);
2944 if (ret)
2945 goto out;
2946
2947 ret = cma_modify_qp_rts(id_priv, conn_param);
2948 if (ret)
2949 goto out;
2950
2951 memset(&rep, 0, sizeof rep);
2952 rep.qp_num = id_priv->qp_num;
2953 rep.starting_psn = id_priv->seq_num;
2954 rep.private_data = conn_param->private_data;
2955 rep.private_data_len = conn_param->private_data_len;
2956 rep.responder_resources = conn_param->responder_resources;
2957 rep.initiator_depth = conn_param->initiator_depth;
2958 rep.failover_accepted = 0;
2959 rep.flow_control = conn_param->flow_control;
2960 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2961 rep.srq = id_priv->srq ? 1 : 0;
2962
2963 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2964 out:
2965 return ret;
2966 }
2967
2968 static int cma_accept_iw(struct rdma_id_private *id_priv,
2969 struct rdma_conn_param *conn_param)
2970 {
2971 struct iw_cm_conn_param iw_param;
2972 int ret;
2973
2974 ret = cma_modify_qp_rtr(id_priv, conn_param);
2975 if (ret)
2976 return ret;
2977
2978 iw_param.ord = conn_param->initiator_depth;
2979 iw_param.ird = conn_param->responder_resources;
2980 iw_param.private_data = conn_param->private_data;
2981 iw_param.private_data_len = conn_param->private_data_len;
2982 if (id_priv->id.qp) {
2983 iw_param.qpn = id_priv->qp_num;
2984 } else
2985 iw_param.qpn = conn_param->qp_num;
2986
2987 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2988 }
2989
2990 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2991 enum ib_cm_sidr_status status, u32 qkey,
2992 const void *private_data, int private_data_len)
2993 {
2994 struct ib_cm_sidr_rep_param rep;
2995 int ret;
2996
2997 memset(&rep, 0, sizeof rep);
2998 rep.status = status;
2999 if (status == IB_SIDR_SUCCESS) {
3000 ret = cma_set_qkey(id_priv, qkey);
3001 if (ret)
3002 return ret;
3003 rep.qp_num = id_priv->qp_num;
3004 rep.qkey = id_priv->qkey;
3005 }
3006 rep.private_data = private_data;
3007 rep.private_data_len = private_data_len;
3008
3009 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3010 }
3011
3012 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3013 {
3014 struct rdma_id_private *id_priv;
3015 int ret;
3016
3017 id_priv = container_of(id, struct rdma_id_private, id);
3018
3019 id_priv->owner = task_pid_nr(current);
3020
3021 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3022 return -EINVAL;
3023
3024 if (!id->qp && conn_param) {
3025 id_priv->qp_num = conn_param->qp_num;
3026 id_priv->srq = conn_param->srq;
3027 }
3028
3029 switch (rdma_node_get_transport(id->device->node_type)) {
3030 case RDMA_TRANSPORT_IB:
3031 if (id->qp_type == IB_QPT_UD) {
3032 if (conn_param)
3033 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3034 conn_param->qkey,
3035 conn_param->private_data,
3036 conn_param->private_data_len);
3037 else
3038 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3039 0, NULL, 0);
3040 } else {
3041 if (conn_param)
3042 ret = cma_accept_ib(id_priv, conn_param);
3043 else
3044 ret = cma_rep_recv(id_priv);
3045 }
3046 break;
3047 case RDMA_TRANSPORT_IWARP:
3048 ret = cma_accept_iw(id_priv, conn_param);
3049 break;
3050 default:
3051 ret = -ENOSYS;
3052 break;
3053 }
3054
3055 if (ret)
3056 goto reject;
3057
3058 return 0;
3059 reject:
3060 cma_modify_qp_err(id_priv);
3061 rdma_reject(id, NULL, 0);
3062 return ret;
3063 }
3064 EXPORT_SYMBOL(rdma_accept);
3065
3066 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3067 {
3068 struct rdma_id_private *id_priv;
3069 int ret;
3070
3071 id_priv = container_of(id, struct rdma_id_private, id);
3072 if (!id_priv->cm_id.ib)
3073 return -EINVAL;
3074
3075 switch (id->device->node_type) {
3076 case RDMA_NODE_IB_CA:
3077 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3078 break;
3079 default:
3080 ret = 0;
3081 break;
3082 }
3083 return ret;
3084 }
3085 EXPORT_SYMBOL(rdma_notify);
3086
3087 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3088 u8 private_data_len)
3089 {
3090 struct rdma_id_private *id_priv;
3091 int ret;
3092
3093 id_priv = container_of(id, struct rdma_id_private, id);
3094 if (!id_priv->cm_id.ib)
3095 return -EINVAL;
3096
3097 switch (rdma_node_get_transport(id->device->node_type)) {
3098 case RDMA_TRANSPORT_IB:
3099 if (id->qp_type == IB_QPT_UD)
3100 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3101 private_data, private_data_len);
3102 else
3103 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3104 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3105 0, private_data, private_data_len);
3106 break;
3107 case RDMA_TRANSPORT_IWARP:
3108 ret = iw_cm_reject(id_priv->cm_id.iw,
3109 private_data, private_data_len);
3110 break;
3111 default:
3112 ret = -ENOSYS;
3113 break;
3114 }
3115 return ret;
3116 }
3117 EXPORT_SYMBOL(rdma_reject);
3118
3119 int rdma_disconnect(struct rdma_cm_id *id)
3120 {
3121 struct rdma_id_private *id_priv;
3122 int ret;
3123
3124 id_priv = container_of(id, struct rdma_id_private, id);
3125 if (!id_priv->cm_id.ib)
3126 return -EINVAL;
3127
3128 switch (rdma_node_get_transport(id->device->node_type)) {
3129 case RDMA_TRANSPORT_IB:
3130 ret = cma_modify_qp_err(id_priv);
3131 if (ret)
3132 goto out;
3133 /* Initiate or respond to a disconnect. */
3134 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3135 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3136 break;
3137 case RDMA_TRANSPORT_IWARP:
3138 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3139 break;
3140 default:
3141 ret = -EINVAL;
3142 break;
3143 }
3144 out:
3145 return ret;
3146 }
3147 EXPORT_SYMBOL(rdma_disconnect);
3148
3149 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3150 {
3151 struct rdma_id_private *id_priv;
3152 struct cma_multicast *mc = multicast->context;
3153 struct rdma_cm_event event;
3154 int ret;
3155
3156 id_priv = mc->id_priv;
3157 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3158 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3159 return 0;
3160
3161 if (!status)
3162 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3163 mutex_lock(&id_priv->qp_mutex);
3164 if (!status && id_priv->id.qp)
3165 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3166 be16_to_cpu(multicast->rec.mlid));
3167 mutex_unlock(&id_priv->qp_mutex);
3168
3169 memset(&event, 0, sizeof event);
3170 event.status = status;
3171 event.param.ud.private_data = mc->context;
3172 if (!status) {
3173 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3174 ib_init_ah_from_mcmember(id_priv->id.device,
3175 id_priv->id.port_num, &multicast->rec,
3176 &event.param.ud.ah_attr);
3177 event.param.ud.qp_num = 0xFFFFFF;
3178 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3179 } else
3180 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3181
3182 ret = id_priv->id.event_handler(&id_priv->id, &event);
3183 if (ret) {
3184 cma_exch(id_priv, RDMA_CM_DESTROYING);
3185 mutex_unlock(&id_priv->handler_mutex);
3186 rdma_destroy_id(&id_priv->id);
3187 return 0;
3188 }
3189
3190 mutex_unlock(&id_priv->handler_mutex);
3191 return 0;
3192 }
3193
3194 static void cma_set_mgid(struct rdma_id_private *id_priv,
3195 struct sockaddr *addr, union ib_gid *mgid)
3196 {
3197 unsigned char mc_map[MAX_ADDR_LEN];
3198 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3199 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3200 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3201
3202 if (cma_any_addr(addr)) {
3203 memset(mgid, 0, sizeof *mgid);
3204 } else if ((addr->sa_family == AF_INET6) &&
3205 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3206 0xFF10A01B)) {
3207 /* IPv6 address is an SA assigned MGID. */
3208 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3209 } else if (addr->sa_family == AF_IB) {
3210 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3211 } else if ((addr->sa_family == AF_INET6)) {
3212 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3213 if (id_priv->id.ps == RDMA_PS_UDP)
3214 mc_map[7] = 0x01; /* Use RDMA CM signature */
3215 *mgid = *(union ib_gid *) (mc_map + 4);
3216 } else {
3217 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3218 if (id_priv->id.ps == RDMA_PS_UDP)
3219 mc_map[7] = 0x01; /* Use RDMA CM signature */
3220 *mgid = *(union ib_gid *) (mc_map + 4);
3221 }
3222 }
3223
3224 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3225 struct cma_multicast *mc)
3226 {
3227 struct ib_sa_mcmember_rec rec;
3228 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3229 ib_sa_comp_mask comp_mask;
3230 int ret;
3231
3232 ib_addr_get_mgid(dev_addr, &rec.mgid);
3233 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3234 &rec.mgid, &rec);
3235 if (ret)
3236 return ret;
3237
3238 ret = cma_set_qkey(id_priv, 0);
3239 if (ret)
3240 return ret;
3241
3242 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3243 rec.qkey = cpu_to_be32(id_priv->qkey);
3244 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3245 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3246 rec.join_state = 1;
3247
3248 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3249 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3250 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3251 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3252 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3253
3254 if (id_priv->id.ps == RDMA_PS_IPOIB)
3255 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3256 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3257 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3258 IB_SA_MCMEMBER_REC_MTU |
3259 IB_SA_MCMEMBER_REC_HOP_LIMIT;
3260
3261 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3262 id_priv->id.port_num, &rec,
3263 comp_mask, GFP_KERNEL,
3264 cma_ib_mc_handler, mc);
3265 return PTR_ERR_OR_ZERO(mc->multicast.ib);
3266 }
3267
3268 static void iboe_mcast_work_handler(struct work_struct *work)
3269 {
3270 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3271 struct cma_multicast *mc = mw->mc;
3272 struct ib_sa_multicast *m = mc->multicast.ib;
3273
3274 mc->multicast.ib->context = mc;
3275 cma_ib_mc_handler(0, m);
3276 kref_put(&mc->mcref, release_mc);
3277 kfree(mw);
3278 }
3279
3280 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3281 {
3282 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3283 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3284
3285 if (cma_any_addr(addr)) {
3286 memset(mgid, 0, sizeof *mgid);
3287 } else if (addr->sa_family == AF_INET6) {
3288 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3289 } else {
3290 mgid->raw[0] = 0xff;
3291 mgid->raw[1] = 0x0e;
3292 mgid->raw[2] = 0;
3293 mgid->raw[3] = 0;
3294 mgid->raw[4] = 0;
3295 mgid->raw[5] = 0;
3296 mgid->raw[6] = 0;
3297 mgid->raw[7] = 0;
3298 mgid->raw[8] = 0;
3299 mgid->raw[9] = 0;
3300 mgid->raw[10] = 0xff;
3301 mgid->raw[11] = 0xff;
3302 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3303 }
3304 }
3305
3306 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3307 struct cma_multicast *mc)
3308 {
3309 struct iboe_mcast_work *work;
3310 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3311 int err;
3312 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3313 struct net_device *ndev = NULL;
3314
3315 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3316 return -EINVAL;
3317
3318 work = kzalloc(sizeof *work, GFP_KERNEL);
3319 if (!work)
3320 return -ENOMEM;
3321
3322 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3323 if (!mc->multicast.ib) {
3324 err = -ENOMEM;
3325 goto out1;
3326 }
3327
3328 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3329
3330 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3331 if (id_priv->id.ps == RDMA_PS_UDP)
3332 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3333
3334 if (dev_addr->bound_dev_if)
3335 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3336 if (!ndev) {
3337 err = -ENODEV;
3338 goto out2;
3339 }
3340 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3341 mc->multicast.ib->rec.hop_limit = 1;
3342 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3343 dev_put(ndev);
3344 if (!mc->multicast.ib->rec.mtu) {
3345 err = -EINVAL;
3346 goto out2;
3347 }
3348 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3349 &mc->multicast.ib->rec.port_gid);
3350 work->id = id_priv;
3351 work->mc = mc;
3352 INIT_WORK(&work->work, iboe_mcast_work_handler);
3353 kref_get(&mc->mcref);
3354 queue_work(cma_wq, &work->work);
3355
3356 return 0;
3357
3358 out2:
3359 kfree(mc->multicast.ib);
3360 out1:
3361 kfree(work);
3362 return err;
3363 }
3364
3365 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3366 void *context)
3367 {
3368 struct rdma_id_private *id_priv;
3369 struct cma_multicast *mc;
3370 int ret;
3371
3372 id_priv = container_of(id, struct rdma_id_private, id);
3373 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3374 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3375 return -EINVAL;
3376
3377 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3378 if (!mc)
3379 return -ENOMEM;
3380
3381 memcpy(&mc->addr, addr, rdma_addr_size(addr));
3382 mc->context = context;
3383 mc->id_priv = id_priv;
3384
3385 spin_lock(&id_priv->lock);
3386 list_add(&mc->list, &id_priv->mc_list);
3387 spin_unlock(&id_priv->lock);
3388
3389 switch (rdma_node_get_transport(id->device->node_type)) {
3390 case RDMA_TRANSPORT_IB:
3391 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3392 case IB_LINK_LAYER_INFINIBAND:
3393 ret = cma_join_ib_multicast(id_priv, mc);
3394 break;
3395 case IB_LINK_LAYER_ETHERNET:
3396 kref_init(&mc->mcref);
3397 ret = cma_iboe_join_multicast(id_priv, mc);
3398 break;
3399 default:
3400 ret = -EINVAL;
3401 }
3402 break;
3403 default:
3404 ret = -ENOSYS;
3405 break;
3406 }
3407
3408 if (ret) {
3409 spin_lock_irq(&id_priv->lock);
3410 list_del(&mc->list);
3411 spin_unlock_irq(&id_priv->lock);
3412 kfree(mc);
3413 }
3414 return ret;
3415 }
3416 EXPORT_SYMBOL(rdma_join_multicast);
3417
3418 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3419 {
3420 struct rdma_id_private *id_priv;
3421 struct cma_multicast *mc;
3422
3423 id_priv = container_of(id, struct rdma_id_private, id);
3424 spin_lock_irq(&id_priv->lock);
3425 list_for_each_entry(mc, &id_priv->mc_list, list) {
3426 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3427 list_del(&mc->list);
3428 spin_unlock_irq(&id_priv->lock);
3429
3430 if (id->qp)
3431 ib_detach_mcast(id->qp,
3432 &mc->multicast.ib->rec.mgid,
3433 be16_to_cpu(mc->multicast.ib->rec.mlid));
3434 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3435 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3436 case IB_LINK_LAYER_INFINIBAND:
3437 ib_sa_free_multicast(mc->multicast.ib);
3438 kfree(mc);
3439 break;
3440 case IB_LINK_LAYER_ETHERNET:
3441 kref_put(&mc->mcref, release_mc);
3442 break;
3443 default:
3444 break;
3445 }
3446 }
3447 return;
3448 }
3449 }
3450 spin_unlock_irq(&id_priv->lock);
3451 }
3452 EXPORT_SYMBOL(rdma_leave_multicast);
3453
3454 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3455 {
3456 struct rdma_dev_addr *dev_addr;
3457 struct cma_ndev_work *work;
3458
3459 dev_addr = &id_priv->id.route.addr.dev_addr;
3460
3461 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3462 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3463 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3464 ndev->name, &id_priv->id);
3465 work = kzalloc(sizeof *work, GFP_KERNEL);
3466 if (!work)
3467 return -ENOMEM;
3468
3469 INIT_WORK(&work->work, cma_ndev_work_handler);
3470 work->id = id_priv;
3471 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3472 atomic_inc(&id_priv->refcount);
3473 queue_work(cma_wq, &work->work);
3474 }
3475
3476 return 0;
3477 }
3478
3479 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3480 void *ptr)
3481 {
3482 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3483 struct cma_device *cma_dev;
3484 struct rdma_id_private *id_priv;
3485 int ret = NOTIFY_DONE;
3486
3487 if (dev_net(ndev) != &init_net)
3488 return NOTIFY_DONE;
3489
3490 if (event != NETDEV_BONDING_FAILOVER)
3491 return NOTIFY_DONE;
3492
3493 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3494 return NOTIFY_DONE;
3495
3496 mutex_lock(&lock);
3497 list_for_each_entry(cma_dev, &dev_list, list)
3498 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3499 ret = cma_netdev_change(ndev, id_priv);
3500 if (ret)
3501 goto out;
3502 }
3503
3504 out:
3505 mutex_unlock(&lock);
3506 return ret;
3507 }
3508
3509 static struct notifier_block cma_nb = {
3510 .notifier_call = cma_netdev_callback
3511 };
3512
3513 static void cma_add_one(struct ib_device *device)
3514 {
3515 struct cma_device *cma_dev;
3516 struct rdma_id_private *id_priv;
3517
3518 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3519 if (!cma_dev)
3520 return;
3521
3522 cma_dev->device = device;
3523
3524 init_completion(&cma_dev->comp);
3525 atomic_set(&cma_dev->refcount, 1);
3526 INIT_LIST_HEAD(&cma_dev->id_list);
3527 ib_set_client_data(device, &cma_client, cma_dev);
3528
3529 mutex_lock(&lock);
3530 list_add_tail(&cma_dev->list, &dev_list);
3531 list_for_each_entry(id_priv, &listen_any_list, list)
3532 cma_listen_on_dev(id_priv, cma_dev);
3533 mutex_unlock(&lock);
3534 }
3535
3536 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3537 {
3538 struct rdma_cm_event event;
3539 enum rdma_cm_state state;
3540 int ret = 0;
3541
3542 /* Record that we want to remove the device */
3543 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3544 if (state == RDMA_CM_DESTROYING)
3545 return 0;
3546
3547 cma_cancel_operation(id_priv, state);
3548 mutex_lock(&id_priv->handler_mutex);
3549
3550 /* Check for destruction from another callback. */
3551 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3552 goto out;
3553
3554 memset(&event, 0, sizeof event);
3555 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3556 ret = id_priv->id.event_handler(&id_priv->id, &event);
3557 out:
3558 mutex_unlock(&id_priv->handler_mutex);
3559 return ret;
3560 }
3561
3562 static void cma_process_remove(struct cma_device *cma_dev)
3563 {
3564 struct rdma_id_private *id_priv;
3565 int ret;
3566
3567 mutex_lock(&lock);
3568 while (!list_empty(&cma_dev->id_list)) {
3569 id_priv = list_entry(cma_dev->id_list.next,
3570 struct rdma_id_private, list);
3571
3572 list_del(&id_priv->listen_list);
3573 list_del_init(&id_priv->list);
3574 atomic_inc(&id_priv->refcount);
3575 mutex_unlock(&lock);
3576
3577 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3578 cma_deref_id(id_priv);
3579 if (ret)
3580 rdma_destroy_id(&id_priv->id);
3581
3582 mutex_lock(&lock);
3583 }
3584 mutex_unlock(&lock);
3585
3586 cma_deref_dev(cma_dev);
3587 wait_for_completion(&cma_dev->comp);
3588 }
3589
3590 static void cma_remove_one(struct ib_device *device)
3591 {
3592 struct cma_device *cma_dev;
3593
3594 cma_dev = ib_get_client_data(device, &cma_client);
3595 if (!cma_dev)
3596 return;
3597
3598 mutex_lock(&lock);
3599 list_del(&cma_dev->list);
3600 mutex_unlock(&lock);
3601
3602 cma_process_remove(cma_dev);
3603 kfree(cma_dev);
3604 }
3605
3606 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3607 {
3608 struct nlmsghdr *nlh;
3609 struct rdma_cm_id_stats *id_stats;
3610 struct rdma_id_private *id_priv;
3611 struct rdma_cm_id *id = NULL;
3612 struct cma_device *cma_dev;
3613 int i_dev = 0, i_id = 0;
3614
3615 /*
3616 * We export all of the IDs as a sequence of messages. Each
3617 * ID gets its own netlink message.
3618 */
3619 mutex_lock(&lock);
3620
3621 list_for_each_entry(cma_dev, &dev_list, list) {
3622 if (i_dev < cb->args[0]) {
3623 i_dev++;
3624 continue;
3625 }
3626
3627 i_id = 0;
3628 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3629 if (i_id < cb->args[1]) {
3630 i_id++;
3631 continue;
3632 }
3633
3634 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3635 sizeof *id_stats, RDMA_NL_RDMA_CM,
3636 RDMA_NL_RDMA_CM_ID_STATS);
3637 if (!id_stats)
3638 goto out;
3639
3640 memset(id_stats, 0, sizeof *id_stats);
3641 id = &id_priv->id;
3642 id_stats->node_type = id->route.addr.dev_addr.dev_type;
3643 id_stats->port_num = id->port_num;
3644 id_stats->bound_dev_if =
3645 id->route.addr.dev_addr.bound_dev_if;
3646
3647 if (ibnl_put_attr(skb, nlh,
3648 rdma_addr_size(cma_src_addr(id_priv)),
3649 cma_src_addr(id_priv),
3650 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3651 goto out;
3652 if (ibnl_put_attr(skb, nlh,
3653 rdma_addr_size(cma_src_addr(id_priv)),
3654 cma_dst_addr(id_priv),
3655 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
3656 goto out;
3657
3658 id_stats->pid = id_priv->owner;
3659 id_stats->port_space = id->ps;
3660 id_stats->cm_state = id_priv->state;
3661 id_stats->qp_num = id_priv->qp_num;
3662 id_stats->qp_type = id->qp_type;
3663
3664 i_id++;
3665 }
3666
3667 cb->args[1] = 0;
3668 i_dev++;
3669 }
3670
3671 out:
3672 mutex_unlock(&lock);
3673 cb->args[0] = i_dev;
3674 cb->args[1] = i_id;
3675
3676 return skb->len;
3677 }
3678
3679 static const struct ibnl_client_cbs cma_cb_table[] = {
3680 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3681 .module = THIS_MODULE },
3682 };
3683
3684 static int __init cma_init(void)
3685 {
3686 int ret;
3687
3688 cma_wq = create_singlethread_workqueue("rdma_cm");
3689 if (!cma_wq)
3690 return -ENOMEM;
3691
3692 ib_sa_register_client(&sa_client);
3693 rdma_addr_register_client(&addr_client);
3694 register_netdevice_notifier(&cma_nb);
3695
3696 ret = ib_register_client(&cma_client);
3697 if (ret)
3698 goto err;
3699
3700 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3701 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3702
3703 return 0;
3704
3705 err:
3706 unregister_netdevice_notifier(&cma_nb);
3707 rdma_addr_unregister_client(&addr_client);
3708 ib_sa_unregister_client(&sa_client);
3709 destroy_workqueue(cma_wq);
3710 return ret;
3711 }
3712
3713 static void __exit cma_cleanup(void)
3714 {
3715 ibnl_remove_client(RDMA_NL_RDMA_CM);
3716 ib_unregister_client(&cma_client);
3717 unregister_netdevice_notifier(&cma_nb);
3718 rdma_addr_unregister_client(&addr_client);
3719 ib_sa_unregister_client(&sa_client);
3720 destroy_workqueue(cma_wq);
3721 idr_destroy(&tcp_ps);
3722 idr_destroy(&udp_ps);
3723 idr_destroy(&ipoib_ps);
3724 idr_destroy(&ib_ps);
3725 }
3726
3727 module_init(cma_init);
3728 module_exit(cma_cleanup);
Linux with added FPC-III support