projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
IB/addr: Simplify resolving IPv4 addresses
[deliverable/linux.git] / drivers / infiniband / core / cma.c
1 /*
2 * Copyright (c) 2005 Voltaire Inc. All rights reserved.
3 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
4 * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
5 * Copyright (c) 2005-2006 Intel Corporation. All rights reserved.
6 *
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 * SOFTWARE.
34 */
35
36 #include <linux/completion.h>
37 #include <linux/in.h>
38 #include <linux/in6.h>
39 #include <linux/mutex.h>
40 #include <linux/random.h>
41 #include <linux/idr.h>
42 #include <linux/inetdevice.h>
43
44 #include <net/tcp.h>
45 #include <net/ipv6.h>
46
47 #include <rdma/rdma_cm.h>
48 #include <rdma/rdma_cm_ib.h>
49 #include <rdma/ib_cache.h>
50 #include <rdma/ib_cm.h>
51 #include <rdma/ib_sa.h>
52 #include <rdma/iw_cm.h>
53
54 MODULE_AUTHOR("Sean Hefty");
55 MODULE_DESCRIPTION("Generic RDMA CM Agent");
56 MODULE_LICENSE("Dual BSD/GPL");
57
58 #define CMA_CM_RESPONSE_TIMEOUT 20
59 #define CMA_MAX_CM_RETRIES 15
60 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
61
62 static void cma_add_one(struct ib_device *device);
63 static void cma_remove_one(struct ib_device *device);
64
65 static struct ib_client cma_client = {
66 .name = "cma",
67 .add = cma_add_one,
68 .remove = cma_remove_one
69 };
70
71 static struct ib_sa_client sa_client;
72 static struct rdma_addr_client addr_client;
73 static LIST_HEAD(dev_list);
74 static LIST_HEAD(listen_any_list);
75 static DEFINE_MUTEX(lock);
76 static struct workqueue_struct *cma_wq;
77 static DEFINE_IDR(sdp_ps);
78 static DEFINE_IDR(tcp_ps);
79 static DEFINE_IDR(udp_ps);
80 static DEFINE_IDR(ipoib_ps);
81 static int next_port;
82
83 struct cma_device {
84 struct list_head list;
85 struct ib_device *device;
86 struct completion comp;
87 atomic_t refcount;
88 struct list_head id_list;
89 };
90
91 enum cma_state {
92 CMA_IDLE,
93 CMA_ADDR_QUERY,
94 CMA_ADDR_RESOLVED,
95 CMA_ROUTE_QUERY,
96 CMA_ROUTE_RESOLVED,
97 CMA_CONNECT,
98 CMA_DISCONNECT,
99 CMA_ADDR_BOUND,
100 CMA_LISTEN,
101 CMA_DEVICE_REMOVAL,
102 CMA_DESTROYING
103 };
104
105 struct rdma_bind_list {
106 struct idr *ps;
107 struct hlist_head owners;
108 unsigned short port;
109 };
110
111 /*
112 * Device removal can occur at anytime, so we need extra handling to
113 * serialize notifying the user of device removal with other callbacks.
114 * We do this by disabling removal notification while a callback is in process,
115 * and reporting it after the callback completes.
116 */
117 struct rdma_id_private {
118 struct rdma_cm_id id;
119
120 struct rdma_bind_list *bind_list;
121 struct hlist_node node;
122 struct list_head list; /* listen_any_list or cma_device.list */
123 struct list_head listen_list; /* per device listens */
124 struct cma_device *cma_dev;
125 struct list_head mc_list;
126
127 int internal_id;
128 enum cma_state state;
129 spinlock_t lock;
130 struct mutex qp_mutex;
131
132 struct completion comp;
133 atomic_t refcount;
134 struct mutex handler_mutex;
135
136 int backlog;
137 int timeout_ms;
138 struct ib_sa_query *query;
139 int query_id;
140 union {
141 struct ib_cm_id *ib;
142 struct iw_cm_id *iw;
143 } cm_id;
144
145 u32 seq_num;
146 u32 qkey;
147 u32 qp_num;
148 u8 srq;
149 u8 tos;
150 };
151
152 struct cma_multicast {
153 struct rdma_id_private *id_priv;
154 union {
155 struct ib_sa_multicast *ib;
156 } multicast;
157 struct list_head list;
158 void *context;
159 struct sockaddr_storage addr;
160 };
161
162 struct cma_work {
163 struct work_struct work;
164 struct rdma_id_private *id;
165 enum cma_state old_state;
166 enum cma_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 union cma_ip_addr {
177 struct in6_addr ip6;
178 struct {
179 __be32 pad[3];
180 __be32 addr;
181 } ip4;
182 };
183
184 struct cma_hdr {
185 u8 cma_version;
186 u8 ip_version; /* IP version: 7:4 */
187 __be16 port;
188 union cma_ip_addr src_addr;
189 union cma_ip_addr dst_addr;
190 };
191
192 struct sdp_hh {
193 u8 bsdh[16];
194 u8 sdp_version; /* Major version: 7:4 */
195 u8 ip_version; /* IP version: 7:4 */
196 u8 sdp_specific1[10];
197 __be16 port;
198 __be16 sdp_specific2;
199 union cma_ip_addr src_addr;
200 union cma_ip_addr dst_addr;
201 };
202
203 struct sdp_hah {
204 u8 bsdh[16];
205 u8 sdp_version;
206 };
207
208 #define CMA_VERSION 0x00
209 #define SDP_MAJ_VERSION 0x2
210
211 static int cma_comp(struct rdma_id_private *id_priv, enum cma_state comp)
212 {
213 unsigned long flags;
214 int ret;
215
216 spin_lock_irqsave(&id_priv->lock, flags);
217 ret = (id_priv->state == comp);
218 spin_unlock_irqrestore(&id_priv->lock, flags);
219 return ret;
220 }
221
222 static int cma_comp_exch(struct rdma_id_private *id_priv,
223 enum cma_state comp, enum cma_state exch)
224 {
225 unsigned long flags;
226 int ret;
227
228 spin_lock_irqsave(&id_priv->lock, flags);
229 if ((ret = (id_priv->state == comp)))
230 id_priv->state = exch;
231 spin_unlock_irqrestore(&id_priv->lock, flags);
232 return ret;
233 }
234
235 static enum cma_state cma_exch(struct rdma_id_private *id_priv,
236 enum cma_state exch)
237 {
238 unsigned long flags;
239 enum cma_state old;
240
241 spin_lock_irqsave(&id_priv->lock, flags);
242 old = id_priv->state;
243 id_priv->state = exch;
244 spin_unlock_irqrestore(&id_priv->lock, flags);
245 return old;
246 }
247
248 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
249 {
250 return hdr->ip_version >> 4;
251 }
252
253 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
254 {
255 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
256 }
257
258 static inline u8 sdp_get_majv(u8 sdp_version)
259 {
260 return sdp_version >> 4;
261 }
262
263 static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
264 {
265 return hh->ip_version >> 4;
266 }
267
268 static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
269 {
270 hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
271 }
272
273 static inline int cma_is_ud_ps(enum rdma_port_space ps)
274 {
275 return (ps == RDMA_PS_UDP || ps == RDMA_PS_IPOIB);
276 }
277
278 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
279 struct cma_device *cma_dev)
280 {
281 atomic_inc(&cma_dev->refcount);
282 id_priv->cma_dev = cma_dev;
283 id_priv->id.device = cma_dev->device;
284 list_add_tail(&id_priv->list, &cma_dev->id_list);
285 }
286
287 static inline void cma_deref_dev(struct cma_device *cma_dev)
288 {
289 if (atomic_dec_and_test(&cma_dev->refcount))
290 complete(&cma_dev->comp);
291 }
292
293 static void cma_detach_from_dev(struct rdma_id_private *id_priv)
294 {
295 list_del(&id_priv->list);
296 cma_deref_dev(id_priv->cma_dev);
297 id_priv->cma_dev = NULL;
298 }
299
300 static int cma_set_qkey(struct rdma_id_private *id_priv)
301 {
302 struct ib_sa_mcmember_rec rec;
303 int ret = 0;
304
305 if (id_priv->qkey)
306 return 0;
307
308 switch (id_priv->id.ps) {
309 case RDMA_PS_UDP:
310 id_priv->qkey = RDMA_UDP_QKEY;
311 break;
312 case RDMA_PS_IPOIB:
313 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
314 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
315 id_priv->id.port_num, &rec.mgid,
316 &rec);
317 if (!ret)
318 id_priv->qkey = be32_to_cpu(rec.qkey);
319 break;
320 default:
321 break;
322 }
323 return ret;
324 }
325
326 static int cma_acquire_dev(struct rdma_id_private *id_priv)
327 {
328 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
329 struct cma_device *cma_dev;
330 union ib_gid gid;
331 int ret = -ENODEV;
332
333 rdma_addr_get_sgid(dev_addr, &gid);
334 list_for_each_entry(cma_dev, &dev_list, list) {
335 ret = ib_find_cached_gid(cma_dev->device, &gid,
336 &id_priv->id.port_num, NULL);
337 if (!ret) {
338 cma_attach_to_dev(id_priv, cma_dev);
339 break;
340 }
341 }
342 return ret;
343 }
344
345 static void cma_deref_id(struct rdma_id_private *id_priv)
346 {
347 if (atomic_dec_and_test(&id_priv->refcount))
348 complete(&id_priv->comp);
349 }
350
351 static int cma_disable_callback(struct rdma_id_private *id_priv,
352 enum cma_state state)
353 {
354 mutex_lock(&id_priv->handler_mutex);
355 if (id_priv->state != state) {
356 mutex_unlock(&id_priv->handler_mutex);
357 return -EINVAL;
358 }
359 return 0;
360 }
361
362 static int cma_has_cm_dev(struct rdma_id_private *id_priv)
363 {
364 return (id_priv->id.device && id_priv->cm_id.ib);
365 }
366
367 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
368 void *context, enum rdma_port_space ps)
369 {
370 struct rdma_id_private *id_priv;
371
372 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
373 if (!id_priv)
374 return ERR_PTR(-ENOMEM);
375
376 id_priv->state = CMA_IDLE;
377 id_priv->id.context = context;
378 id_priv->id.event_handler = event_handler;
379 id_priv->id.ps = ps;
380 spin_lock_init(&id_priv->lock);
381 mutex_init(&id_priv->qp_mutex);
382 init_completion(&id_priv->comp);
383 atomic_set(&id_priv->refcount, 1);
384 mutex_init(&id_priv->handler_mutex);
385 INIT_LIST_HEAD(&id_priv->listen_list);
386 INIT_LIST_HEAD(&id_priv->mc_list);
387 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
388
389 return &id_priv->id;
390 }
391 EXPORT_SYMBOL(rdma_create_id);
392
393 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
394 {
395 struct ib_qp_attr qp_attr;
396 int qp_attr_mask, ret;
397
398 qp_attr.qp_state = IB_QPS_INIT;
399 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
400 if (ret)
401 return ret;
402
403 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
404 if (ret)
405 return ret;
406
407 qp_attr.qp_state = IB_QPS_RTR;
408 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
409 if (ret)
410 return ret;
411
412 qp_attr.qp_state = IB_QPS_RTS;
413 qp_attr.sq_psn = 0;
414 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
415
416 return ret;
417 }
418
419 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
420 {
421 struct ib_qp_attr qp_attr;
422 int qp_attr_mask, ret;
423
424 qp_attr.qp_state = IB_QPS_INIT;
425 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
426 if (ret)
427 return ret;
428
429 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
430 }
431
432 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
433 struct ib_qp_init_attr *qp_init_attr)
434 {
435 struct rdma_id_private *id_priv;
436 struct ib_qp *qp;
437 int ret;
438
439 id_priv = container_of(id, struct rdma_id_private, id);
440 if (id->device != pd->device)
441 return -EINVAL;
442
443 qp = ib_create_qp(pd, qp_init_attr);
444 if (IS_ERR(qp))
445 return PTR_ERR(qp);
446
447 if (cma_is_ud_ps(id_priv->id.ps))
448 ret = cma_init_ud_qp(id_priv, qp);
449 else
450 ret = cma_init_conn_qp(id_priv, qp);
451 if (ret)
452 goto err;
453
454 id->qp = qp;
455 id_priv->qp_num = qp->qp_num;
456 id_priv->srq = (qp->srq != NULL);
457 return 0;
458 err:
459 ib_destroy_qp(qp);
460 return ret;
461 }
462 EXPORT_SYMBOL(rdma_create_qp);
463
464 void rdma_destroy_qp(struct rdma_cm_id *id)
465 {
466 struct rdma_id_private *id_priv;
467
468 id_priv = container_of(id, struct rdma_id_private, id);
469 mutex_lock(&id_priv->qp_mutex);
470 ib_destroy_qp(id_priv->id.qp);
471 id_priv->id.qp = NULL;
472 mutex_unlock(&id_priv->qp_mutex);
473 }
474 EXPORT_SYMBOL(rdma_destroy_qp);
475
476 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
477 struct rdma_conn_param *conn_param)
478 {
479 struct ib_qp_attr qp_attr;
480 int qp_attr_mask, ret;
481
482 mutex_lock(&id_priv->qp_mutex);
483 if (!id_priv->id.qp) {
484 ret = 0;
485 goto out;
486 }
487
488 /* Need to update QP attributes from default values. */
489 qp_attr.qp_state = IB_QPS_INIT;
490 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
491 if (ret)
492 goto out;
493
494 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
495 if (ret)
496 goto out;
497
498 qp_attr.qp_state = IB_QPS_RTR;
499 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
500 if (ret)
501 goto out;
502
503 if (conn_param)
504 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
505 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
506 out:
507 mutex_unlock(&id_priv->qp_mutex);
508 return ret;
509 }
510
511 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
512 struct rdma_conn_param *conn_param)
513 {
514 struct ib_qp_attr qp_attr;
515 int qp_attr_mask, ret;
516
517 mutex_lock(&id_priv->qp_mutex);
518 if (!id_priv->id.qp) {
519 ret = 0;
520 goto out;
521 }
522
523 qp_attr.qp_state = IB_QPS_RTS;
524 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
525 if (ret)
526 goto out;
527
528 if (conn_param)
529 qp_attr.max_rd_atomic = conn_param->initiator_depth;
530 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
531 out:
532 mutex_unlock(&id_priv->qp_mutex);
533 return ret;
534 }
535
536 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
537 {
538 struct ib_qp_attr qp_attr;
539 int ret;
540
541 mutex_lock(&id_priv->qp_mutex);
542 if (!id_priv->id.qp) {
543 ret = 0;
544 goto out;
545 }
546
547 qp_attr.qp_state = IB_QPS_ERR;
548 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
549 out:
550 mutex_unlock(&id_priv->qp_mutex);
551 return ret;
552 }
553
554 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
555 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
556 {
557 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
558 int ret;
559
560 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
561 ib_addr_get_pkey(dev_addr),
562 &qp_attr->pkey_index);
563 if (ret)
564 return ret;
565
566 qp_attr->port_num = id_priv->id.port_num;
567 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
568
569 if (cma_is_ud_ps(id_priv->id.ps)) {
570 ret = cma_set_qkey(id_priv);
571 if (ret)
572 return ret;
573
574 qp_attr->qkey = id_priv->qkey;
575 *qp_attr_mask |= IB_QP_QKEY;
576 } else {
577 qp_attr->qp_access_flags = 0;
578 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
579 }
580 return 0;
581 }
582
583 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
584 int *qp_attr_mask)
585 {
586 struct rdma_id_private *id_priv;
587 int ret = 0;
588
589 id_priv = container_of(id, struct rdma_id_private, id);
590 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
591 case RDMA_TRANSPORT_IB:
592 if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps))
593 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
594 else
595 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
596 qp_attr_mask);
597 if (qp_attr->qp_state == IB_QPS_RTR)
598 qp_attr->rq_psn = id_priv->seq_num;
599 break;
600 case RDMA_TRANSPORT_IWARP:
601 if (!id_priv->cm_id.iw) {
602 qp_attr->qp_access_flags = 0;
603 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
604 } else
605 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
606 qp_attr_mask);
607 break;
608 default:
609 ret = -ENOSYS;
610 break;
611 }
612
613 return ret;
614 }
615 EXPORT_SYMBOL(rdma_init_qp_attr);
616
617 static inline int cma_zero_addr(struct sockaddr *addr)
618 {
619 struct in6_addr *ip6;
620
621 if (addr->sa_family == AF_INET)
622 return ipv4_is_zeronet(
623 ((struct sockaddr_in *)addr)->sin_addr.s_addr);
624 else {
625 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
626 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
627 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
628 }
629 }
630
631 static inline int cma_loopback_addr(struct sockaddr *addr)
632 {
633 if (addr->sa_family == AF_INET)
634 return ipv4_is_loopback(
635 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
636 else
637 return ipv6_addr_loopback(
638 &((struct sockaddr_in6 *) addr)->sin6_addr);
639 }
640
641 static inline int cma_any_addr(struct sockaddr *addr)
642 {
643 return cma_zero_addr(addr) || cma_loopback_addr(addr);
644 }
645
646 static inline __be16 cma_port(struct sockaddr *addr)
647 {
648 if (addr->sa_family == AF_INET)
649 return ((struct sockaddr_in *) addr)->sin_port;
650 else
651 return ((struct sockaddr_in6 *) addr)->sin6_port;
652 }
653
654 static inline int cma_any_port(struct sockaddr *addr)
655 {
656 return !cma_port(addr);
657 }
658
659 static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
660 u8 *ip_ver, __be16 *port,
661 union cma_ip_addr **src, union cma_ip_addr **dst)
662 {
663 switch (ps) {
664 case RDMA_PS_SDP:
665 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
666 SDP_MAJ_VERSION)
667 return -EINVAL;
668
669 *ip_ver = sdp_get_ip_ver(hdr);
670 *port = ((struct sdp_hh *) hdr)->port;
671 *src = &((struct sdp_hh *) hdr)->src_addr;
672 *dst = &((struct sdp_hh *) hdr)->dst_addr;
673 break;
674 default:
675 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
676 return -EINVAL;
677
678 *ip_ver = cma_get_ip_ver(hdr);
679 *port = ((struct cma_hdr *) hdr)->port;
680 *src = &((struct cma_hdr *) hdr)->src_addr;
681 *dst = &((struct cma_hdr *) hdr)->dst_addr;
682 break;
683 }
684
685 if (*ip_ver != 4 && *ip_ver != 6)
686 return -EINVAL;
687 return 0;
688 }
689
690 static void cma_save_net_info(struct rdma_addr *addr,
691 struct rdma_addr *listen_addr,
692 u8 ip_ver, __be16 port,
693 union cma_ip_addr *src, union cma_ip_addr *dst)
694 {
695 struct sockaddr_in *listen4, *ip4;
696 struct sockaddr_in6 *listen6, *ip6;
697
698 switch (ip_ver) {
699 case 4:
700 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
701 ip4 = (struct sockaddr_in *) &addr->src_addr;
702 ip4->sin_family = listen4->sin_family;
703 ip4->sin_addr.s_addr = dst->ip4.addr;
704 ip4->sin_port = listen4->sin_port;
705
706 ip4 = (struct sockaddr_in *) &addr->dst_addr;
707 ip4->sin_family = listen4->sin_family;
708 ip4->sin_addr.s_addr = src->ip4.addr;
709 ip4->sin_port = port;
710 break;
711 case 6:
712 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
713 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
714 ip6->sin6_family = listen6->sin6_family;
715 ip6->sin6_addr = dst->ip6;
716 ip6->sin6_port = listen6->sin6_port;
717
718 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
719 ip6->sin6_family = listen6->sin6_family;
720 ip6->sin6_addr = src->ip6;
721 ip6->sin6_port = port;
722 break;
723 default:
724 break;
725 }
726 }
727
728 static inline int cma_user_data_offset(enum rdma_port_space ps)
729 {
730 switch (ps) {
731 case RDMA_PS_SDP:
732 return 0;
733 default:
734 return sizeof(struct cma_hdr);
735 }
736 }
737
738 static void cma_cancel_route(struct rdma_id_private *id_priv)
739 {
740 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
741 case RDMA_TRANSPORT_IB:
742 if (id_priv->query)
743 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
744 break;
745 default:
746 break;
747 }
748 }
749
750 static void cma_cancel_listens(struct rdma_id_private *id_priv)
751 {
752 struct rdma_id_private *dev_id_priv;
753
754 /*
755 * Remove from listen_any_list to prevent added devices from spawning
756 * additional listen requests.
757 */
758 mutex_lock(&lock);
759 list_del(&id_priv->list);
760
761 while (!list_empty(&id_priv->listen_list)) {
762 dev_id_priv = list_entry(id_priv->listen_list.next,
763 struct rdma_id_private, listen_list);
764 /* sync with device removal to avoid duplicate destruction */
765 list_del_init(&dev_id_priv->list);
766 list_del(&dev_id_priv->listen_list);
767 mutex_unlock(&lock);
768
769 rdma_destroy_id(&dev_id_priv->id);
770 mutex_lock(&lock);
771 }
772 mutex_unlock(&lock);
773 }
774
775 static void cma_cancel_operation(struct rdma_id_private *id_priv,
776 enum cma_state state)
777 {
778 switch (state) {
779 case CMA_ADDR_QUERY:
780 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
781 break;
782 case CMA_ROUTE_QUERY:
783 cma_cancel_route(id_priv);
784 break;
785 case CMA_LISTEN:
786 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr)
787 && !id_priv->cma_dev)
788 cma_cancel_listens(id_priv);
789 break;
790 default:
791 break;
792 }
793 }
794
795 static void cma_release_port(struct rdma_id_private *id_priv)
796 {
797 struct rdma_bind_list *bind_list = id_priv->bind_list;
798
799 if (!bind_list)
800 return;
801
802 mutex_lock(&lock);
803 hlist_del(&id_priv->node);
804 if (hlist_empty(&bind_list->owners)) {
805 idr_remove(bind_list->ps, bind_list->port);
806 kfree(bind_list);
807 }
808 mutex_unlock(&lock);
809 }
810
811 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
812 {
813 struct cma_multicast *mc;
814
815 while (!list_empty(&id_priv->mc_list)) {
816 mc = container_of(id_priv->mc_list.next,
817 struct cma_multicast, list);
818 list_del(&mc->list);
819 ib_sa_free_multicast(mc->multicast.ib);
820 kfree(mc);
821 }
822 }
823
824 void rdma_destroy_id(struct rdma_cm_id *id)
825 {
826 struct rdma_id_private *id_priv;
827 enum cma_state state;
828
829 id_priv = container_of(id, struct rdma_id_private, id);
830 state = cma_exch(id_priv, CMA_DESTROYING);
831 cma_cancel_operation(id_priv, state);
832
833 mutex_lock(&lock);
834 if (id_priv->cma_dev) {
835 mutex_unlock(&lock);
836 switch (rdma_node_get_transport(id->device->node_type)) {
837 case RDMA_TRANSPORT_IB:
838 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib))
839 ib_destroy_cm_id(id_priv->cm_id.ib);
840 break;
841 case RDMA_TRANSPORT_IWARP:
842 if (id_priv->cm_id.iw && !IS_ERR(id_priv->cm_id.iw))
843 iw_destroy_cm_id(id_priv->cm_id.iw);
844 break;
845 default:
846 break;
847 }
848 cma_leave_mc_groups(id_priv);
849 mutex_lock(&lock);
850 cma_detach_from_dev(id_priv);
851 }
852 mutex_unlock(&lock);
853
854 cma_release_port(id_priv);
855 cma_deref_id(id_priv);
856 wait_for_completion(&id_priv->comp);
857
858 if (id_priv->internal_id)
859 cma_deref_id(id_priv->id.context);
860
861 kfree(id_priv->id.route.path_rec);
862 kfree(id_priv);
863 }
864 EXPORT_SYMBOL(rdma_destroy_id);
865
866 static int cma_rep_recv(struct rdma_id_private *id_priv)
867 {
868 int ret;
869
870 ret = cma_modify_qp_rtr(id_priv, NULL);
871 if (ret)
872 goto reject;
873
874 ret = cma_modify_qp_rts(id_priv, NULL);
875 if (ret)
876 goto reject;
877
878 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
879 if (ret)
880 goto reject;
881
882 return 0;
883 reject:
884 cma_modify_qp_err(id_priv);
885 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
886 NULL, 0, NULL, 0);
887 return ret;
888 }
889
890 static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
891 {
892 if (id_priv->id.ps == RDMA_PS_SDP &&
893 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
894 SDP_MAJ_VERSION)
895 return -EINVAL;
896
897 return 0;
898 }
899
900 static void cma_set_rep_event_data(struct rdma_cm_event *event,
901 struct ib_cm_rep_event_param *rep_data,
902 void *private_data)
903 {
904 event->param.conn.private_data = private_data;
905 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
906 event->param.conn.responder_resources = rep_data->responder_resources;
907 event->param.conn.initiator_depth = rep_data->initiator_depth;
908 event->param.conn.flow_control = rep_data->flow_control;
909 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
910 event->param.conn.srq = rep_data->srq;
911 event->param.conn.qp_num = rep_data->remote_qpn;
912 }
913
914 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
915 {
916 struct rdma_id_private *id_priv = cm_id->context;
917 struct rdma_cm_event event;
918 int ret = 0;
919
920 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
921 cma_disable_callback(id_priv, CMA_CONNECT)) ||
922 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
923 cma_disable_callback(id_priv, CMA_DISCONNECT)))
924 return 0;
925
926 memset(&event, 0, sizeof event);
927 switch (ib_event->event) {
928 case IB_CM_REQ_ERROR:
929 case IB_CM_REP_ERROR:
930 event.event = RDMA_CM_EVENT_UNREACHABLE;
931 event.status = -ETIMEDOUT;
932 break;
933 case IB_CM_REP_RECEIVED:
934 event.status = cma_verify_rep(id_priv, ib_event->private_data);
935 if (event.status)
936 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
937 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
938 event.status = cma_rep_recv(id_priv);
939 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
940 RDMA_CM_EVENT_ESTABLISHED;
941 } else
942 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
943 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
944 ib_event->private_data);
945 break;
946 case IB_CM_RTU_RECEIVED:
947 case IB_CM_USER_ESTABLISHED:
948 event.event = RDMA_CM_EVENT_ESTABLISHED;
949 break;
950 case IB_CM_DREQ_ERROR:
951 event.status = -ETIMEDOUT; /* fall through */
952 case IB_CM_DREQ_RECEIVED:
953 case IB_CM_DREP_RECEIVED:
954 if (!cma_comp_exch(id_priv, CMA_CONNECT, CMA_DISCONNECT))
955 goto out;
956 event.event = RDMA_CM_EVENT_DISCONNECTED;
957 break;
958 case IB_CM_TIMEWAIT_EXIT:
959 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
960 break;
961 case IB_CM_MRA_RECEIVED:
962 /* ignore event */
963 goto out;
964 case IB_CM_REJ_RECEIVED:
965 cma_modify_qp_err(id_priv);
966 event.status = ib_event->param.rej_rcvd.reason;
967 event.event = RDMA_CM_EVENT_REJECTED;
968 event.param.conn.private_data = ib_event->private_data;
969 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
970 break;
971 default:
972 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
973 ib_event->event);
974 goto out;
975 }
976
977 ret = id_priv->id.event_handler(&id_priv->id, &event);
978 if (ret) {
979 /* Destroy the CM ID by returning a non-zero value. */
980 id_priv->cm_id.ib = NULL;
981 cma_exch(id_priv, CMA_DESTROYING);
982 mutex_unlock(&id_priv->handler_mutex);
983 rdma_destroy_id(&id_priv->id);
984 return ret;
985 }
986 out:
987 mutex_unlock(&id_priv->handler_mutex);
988 return ret;
989 }
990
991 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
992 struct ib_cm_event *ib_event)
993 {
994 struct rdma_id_private *id_priv;
995 struct rdma_cm_id *id;
996 struct rdma_route *rt;
997 union cma_ip_addr *src, *dst;
998 __be16 port;
999 u8 ip_ver;
1000 int ret;
1001
1002 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1003 &ip_ver, &port, &src, &dst))
1004 goto err;
1005
1006 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1007 listen_id->ps);
1008 if (IS_ERR(id))
1009 goto err;
1010
1011 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1012 ip_ver, port, src, dst);
1013
1014 rt = &id->route;
1015 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1016 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1017 GFP_KERNEL);
1018 if (!rt->path_rec)
1019 goto destroy_id;
1020
1021 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1022 if (rt->num_paths == 2)
1023 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1024
1025 if (cma_any_addr((struct sockaddr *) &rt->addr.src_addr)) {
1026 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1027 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1028 ib_addr_set_pkey(&rt->addr.dev_addr, rt->path_rec[0].pkey);
1029 } else {
1030 ret = rdma_translate_ip((struct sockaddr *) &rt->addr.src_addr,
1031 &rt->addr.dev_addr);
1032 if (ret)
1033 goto destroy_id;
1034 }
1035 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1036
1037 id_priv = container_of(id, struct rdma_id_private, id);
1038 id_priv->state = CMA_CONNECT;
1039 return id_priv;
1040
1041 destroy_id:
1042 rdma_destroy_id(id);
1043 err:
1044 return NULL;
1045 }
1046
1047 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1048 struct ib_cm_event *ib_event)
1049 {
1050 struct rdma_id_private *id_priv;
1051 struct rdma_cm_id *id;
1052 union cma_ip_addr *src, *dst;
1053 __be16 port;
1054 u8 ip_ver;
1055 int ret;
1056
1057 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1058 listen_id->ps);
1059 if (IS_ERR(id))
1060 return NULL;
1061
1062
1063 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1064 &ip_ver, &port, &src, &dst))
1065 goto err;
1066
1067 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1068 ip_ver, port, src, dst);
1069
1070 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1071 ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr,
1072 &id->route.addr.dev_addr);
1073 if (ret)
1074 goto err;
1075 }
1076
1077 id_priv = container_of(id, struct rdma_id_private, id);
1078 id_priv->state = CMA_CONNECT;
1079 return id_priv;
1080 err:
1081 rdma_destroy_id(id);
1082 return NULL;
1083 }
1084
1085 static void cma_set_req_event_data(struct rdma_cm_event *event,
1086 struct ib_cm_req_event_param *req_data,
1087 void *private_data, int offset)
1088 {
1089 event->param.conn.private_data = private_data + offset;
1090 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1091 event->param.conn.responder_resources = req_data->responder_resources;
1092 event->param.conn.initiator_depth = req_data->initiator_depth;
1093 event->param.conn.flow_control = req_data->flow_control;
1094 event->param.conn.retry_count = req_data->retry_count;
1095 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1096 event->param.conn.srq = req_data->srq;
1097 event->param.conn.qp_num = req_data->remote_qpn;
1098 }
1099
1100 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1101 {
1102 struct rdma_id_private *listen_id, *conn_id;
1103 struct rdma_cm_event event;
1104 int offset, ret;
1105
1106 listen_id = cm_id->context;
1107 if (cma_disable_callback(listen_id, CMA_LISTEN))
1108 return -ECONNABORTED;
1109
1110 memset(&event, 0, sizeof event);
1111 offset = cma_user_data_offset(listen_id->id.ps);
1112 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1113 if (cma_is_ud_ps(listen_id->id.ps)) {
1114 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1115 event.param.ud.private_data = ib_event->private_data + offset;
1116 event.param.ud.private_data_len =
1117 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1118 } else {
1119 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1120 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1121 ib_event->private_data, offset);
1122 }
1123 if (!conn_id) {
1124 ret = -ENOMEM;
1125 goto out;
1126 }
1127
1128 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1129 mutex_lock(&lock);
1130 ret = cma_acquire_dev(conn_id);
1131 mutex_unlock(&lock);
1132 if (ret)
1133 goto release_conn_id;
1134
1135 conn_id->cm_id.ib = cm_id;
1136 cm_id->context = conn_id;
1137 cm_id->cm_handler = cma_ib_handler;
1138
1139 ret = conn_id->id.event_handler(&conn_id->id, &event);
1140 if (!ret) {
1141 /*
1142 * Acquire mutex to prevent user executing rdma_destroy_id()
1143 * while we're accessing the cm_id.
1144 */
1145 mutex_lock(&lock);
1146 if (cma_comp(conn_id, CMA_CONNECT) &&
1147 !cma_is_ud_ps(conn_id->id.ps))
1148 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1149 mutex_unlock(&lock);
1150 mutex_unlock(&conn_id->handler_mutex);
1151 goto out;
1152 }
1153
1154 /* Destroy the CM ID by returning a non-zero value. */
1155 conn_id->cm_id.ib = NULL;
1156
1157 release_conn_id:
1158 cma_exch(conn_id, CMA_DESTROYING);
1159 mutex_unlock(&conn_id->handler_mutex);
1160 rdma_destroy_id(&conn_id->id);
1161
1162 out:
1163 mutex_unlock(&listen_id->handler_mutex);
1164 return ret;
1165 }
1166
1167 static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1168 {
1169 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
1170 }
1171
1172 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1173 struct ib_cm_compare_data *compare)
1174 {
1175 struct cma_hdr *cma_data, *cma_mask;
1176 struct sdp_hh *sdp_data, *sdp_mask;
1177 __be32 ip4_addr;
1178 struct in6_addr ip6_addr;
1179
1180 memset(compare, 0, sizeof *compare);
1181 cma_data = (void *) compare->data;
1182 cma_mask = (void *) compare->mask;
1183 sdp_data = (void *) compare->data;
1184 sdp_mask = (void *) compare->mask;
1185
1186 switch (addr->sa_family) {
1187 case AF_INET:
1188 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1189 if (ps == RDMA_PS_SDP) {
1190 sdp_set_ip_ver(sdp_data, 4);
1191 sdp_set_ip_ver(sdp_mask, 0xF);
1192 sdp_data->dst_addr.ip4.addr = ip4_addr;
1193 sdp_mask->dst_addr.ip4.addr = htonl(~0);
1194 } else {
1195 cma_set_ip_ver(cma_data, 4);
1196 cma_set_ip_ver(cma_mask, 0xF);
1197 cma_data->dst_addr.ip4.addr = ip4_addr;
1198 cma_mask->dst_addr.ip4.addr = htonl(~0);
1199 }
1200 break;
1201 case AF_INET6:
1202 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1203 if (ps == RDMA_PS_SDP) {
1204 sdp_set_ip_ver(sdp_data, 6);
1205 sdp_set_ip_ver(sdp_mask, 0xF);
1206 sdp_data->dst_addr.ip6 = ip6_addr;
1207 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1208 sizeof sdp_mask->dst_addr.ip6);
1209 } else {
1210 cma_set_ip_ver(cma_data, 6);
1211 cma_set_ip_ver(cma_mask, 0xF);
1212 cma_data->dst_addr.ip6 = ip6_addr;
1213 memset(&cma_mask->dst_addr.ip6, 0xFF,
1214 sizeof cma_mask->dst_addr.ip6);
1215 }
1216 break;
1217 default:
1218 break;
1219 }
1220 }
1221
1222 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1223 {
1224 struct rdma_id_private *id_priv = iw_id->context;
1225 struct rdma_cm_event event;
1226 struct sockaddr_in *sin;
1227 int ret = 0;
1228
1229 if (cma_disable_callback(id_priv, CMA_CONNECT))
1230 return 0;
1231
1232 memset(&event, 0, sizeof event);
1233 switch (iw_event->event) {
1234 case IW_CM_EVENT_CLOSE:
1235 event.event = RDMA_CM_EVENT_DISCONNECTED;
1236 break;
1237 case IW_CM_EVENT_CONNECT_REPLY:
1238 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1239 *sin = iw_event->local_addr;
1240 sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
1241 *sin = iw_event->remote_addr;
1242 switch (iw_event->status) {
1243 case 0:
1244 event.event = RDMA_CM_EVENT_ESTABLISHED;
1245 break;
1246 case -ECONNRESET:
1247 case -ECONNREFUSED:
1248 event.event = RDMA_CM_EVENT_REJECTED;
1249 break;
1250 case -ETIMEDOUT:
1251 event.event = RDMA_CM_EVENT_UNREACHABLE;
1252 break;
1253 default:
1254 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1255 break;
1256 }
1257 break;
1258 case IW_CM_EVENT_ESTABLISHED:
1259 event.event = RDMA_CM_EVENT_ESTABLISHED;
1260 break;
1261 default:
1262 BUG_ON(1);
1263 }
1264
1265 event.status = iw_event->status;
1266 event.param.conn.private_data = iw_event->private_data;
1267 event.param.conn.private_data_len = iw_event->private_data_len;
1268 ret = id_priv->id.event_handler(&id_priv->id, &event);
1269 if (ret) {
1270 /* Destroy the CM ID by returning a non-zero value. */
1271 id_priv->cm_id.iw = NULL;
1272 cma_exch(id_priv, CMA_DESTROYING);
1273 mutex_unlock(&id_priv->handler_mutex);
1274 rdma_destroy_id(&id_priv->id);
1275 return ret;
1276 }
1277
1278 mutex_unlock(&id_priv->handler_mutex);
1279 return ret;
1280 }
1281
1282 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1283 struct iw_cm_event *iw_event)
1284 {
1285 struct rdma_cm_id *new_cm_id;
1286 struct rdma_id_private *listen_id, *conn_id;
1287 struct sockaddr_in *sin;
1288 struct net_device *dev = NULL;
1289 struct rdma_cm_event event;
1290 int ret;
1291 struct ib_device_attr attr;
1292
1293 listen_id = cm_id->context;
1294 if (cma_disable_callback(listen_id, CMA_LISTEN))
1295 return -ECONNABORTED;
1296
1297 /* Create a new RDMA id for the new IW CM ID */
1298 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1299 listen_id->id.context,
1300 RDMA_PS_TCP);
1301 if (IS_ERR(new_cm_id)) {
1302 ret = -ENOMEM;
1303 goto out;
1304 }
1305 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1306 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1307 conn_id->state = CMA_CONNECT;
1308
1309 dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
1310 if (!dev) {
1311 ret = -EADDRNOTAVAIL;
1312 mutex_unlock(&conn_id->handler_mutex);
1313 rdma_destroy_id(new_cm_id);
1314 goto out;
1315 }
1316 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1317 if (ret) {
1318 mutex_unlock(&conn_id->handler_mutex);
1319 rdma_destroy_id(new_cm_id);
1320 goto out;
1321 }
1322
1323 mutex_lock(&lock);
1324 ret = cma_acquire_dev(conn_id);
1325 mutex_unlock(&lock);
1326 if (ret) {
1327 mutex_unlock(&conn_id->handler_mutex);
1328 rdma_destroy_id(new_cm_id);
1329 goto out;
1330 }
1331
1332 conn_id->cm_id.iw = cm_id;
1333 cm_id->context = conn_id;
1334 cm_id->cm_handler = cma_iw_handler;
1335
1336 sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
1337 *sin = iw_event->local_addr;
1338 sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
1339 *sin = iw_event->remote_addr;
1340
1341 ret = ib_query_device(conn_id->id.device, &attr);
1342 if (ret) {
1343 mutex_unlock(&conn_id->handler_mutex);
1344 rdma_destroy_id(new_cm_id);
1345 goto out;
1346 }
1347
1348 memset(&event, 0, sizeof event);
1349 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1350 event.param.conn.private_data = iw_event->private_data;
1351 event.param.conn.private_data_len = iw_event->private_data_len;
1352 event.param.conn.initiator_depth = attr.max_qp_init_rd_atom;
1353 event.param.conn.responder_resources = attr.max_qp_rd_atom;
1354 ret = conn_id->id.event_handler(&conn_id->id, &event);
1355 if (ret) {
1356 /* User wants to destroy the CM ID */
1357 conn_id->cm_id.iw = NULL;
1358 cma_exch(conn_id, CMA_DESTROYING);
1359 mutex_unlock(&conn_id->handler_mutex);
1360 rdma_destroy_id(&conn_id->id);
1361 goto out;
1362 }
1363
1364 mutex_unlock(&conn_id->handler_mutex);
1365
1366 out:
1367 if (dev)
1368 dev_put(dev);
1369 mutex_unlock(&listen_id->handler_mutex);
1370 return ret;
1371 }
1372
1373 static int cma_ib_listen(struct rdma_id_private *id_priv)
1374 {
1375 struct ib_cm_compare_data compare_data;
1376 struct sockaddr *addr;
1377 __be64 svc_id;
1378 int ret;
1379
1380 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_req_handler,
1381 id_priv);
1382 if (IS_ERR(id_priv->cm_id.ib))
1383 return PTR_ERR(id_priv->cm_id.ib);
1384
1385 addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
1386 svc_id = cma_get_service_id(id_priv->id.ps, addr);
1387 if (cma_any_addr(addr))
1388 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1389 else {
1390 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1391 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1392 }
1393
1394 if (ret) {
1395 ib_destroy_cm_id(id_priv->cm_id.ib);
1396 id_priv->cm_id.ib = NULL;
1397 }
1398
1399 return ret;
1400 }
1401
1402 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1403 {
1404 int ret;
1405 struct sockaddr_in *sin;
1406
1407 id_priv->cm_id.iw = iw_create_cm_id(id_priv->id.device,
1408 iw_conn_req_handler,
1409 id_priv);
1410 if (IS_ERR(id_priv->cm_id.iw))
1411 return PTR_ERR(id_priv->cm_id.iw);
1412
1413 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1414 id_priv->cm_id.iw->local_addr = *sin;
1415
1416 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1417
1418 if (ret) {
1419 iw_destroy_cm_id(id_priv->cm_id.iw);
1420 id_priv->cm_id.iw = NULL;
1421 }
1422
1423 return ret;
1424 }
1425
1426 static int cma_listen_handler(struct rdma_cm_id *id,
1427 struct rdma_cm_event *event)
1428 {
1429 struct rdma_id_private *id_priv = id->context;
1430
1431 id->context = id_priv->id.context;
1432 id->event_handler = id_priv->id.event_handler;
1433 return id_priv->id.event_handler(id, event);
1434 }
1435
1436 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1437 struct cma_device *cma_dev)
1438 {
1439 struct rdma_id_private *dev_id_priv;
1440 struct rdma_cm_id *id;
1441 int ret;
1442
1443 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps);
1444 if (IS_ERR(id))
1445 return;
1446
1447 dev_id_priv = container_of(id, struct rdma_id_private, id);
1448
1449 dev_id_priv->state = CMA_ADDR_BOUND;
1450 memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
1451 ip_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
1452
1453 cma_attach_to_dev(dev_id_priv, cma_dev);
1454 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1455 atomic_inc(&id_priv->refcount);
1456 dev_id_priv->internal_id = 1;
1457
1458 ret = rdma_listen(id, id_priv->backlog);
1459 if (ret)
1460 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1461 "listening on device %s\n", ret, cma_dev->device->name);
1462 }
1463
1464 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1465 {
1466 struct cma_device *cma_dev;
1467
1468 mutex_lock(&lock);
1469 list_add_tail(&id_priv->list, &listen_any_list);
1470 list_for_each_entry(cma_dev, &dev_list, list)
1471 cma_listen_on_dev(id_priv, cma_dev);
1472 mutex_unlock(&lock);
1473 }
1474
1475 static int cma_bind_any(struct rdma_cm_id *id, sa_family_t af)
1476 {
1477 struct sockaddr_storage addr_in;
1478
1479 memset(&addr_in, 0, sizeof addr_in);
1480 addr_in.ss_family = af;
1481 return rdma_bind_addr(id, (struct sockaddr *) &addr_in);
1482 }
1483
1484 int rdma_listen(struct rdma_cm_id *id, int backlog)
1485 {
1486 struct rdma_id_private *id_priv;
1487 int ret;
1488
1489 id_priv = container_of(id, struct rdma_id_private, id);
1490 if (id_priv->state == CMA_IDLE) {
1491 ret = cma_bind_any(id, AF_INET);
1492 if (ret)
1493 return ret;
1494 }
1495
1496 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
1497 return -EINVAL;
1498
1499 id_priv->backlog = backlog;
1500 if (id->device) {
1501 switch (rdma_node_get_transport(id->device->node_type)) {
1502 case RDMA_TRANSPORT_IB:
1503 ret = cma_ib_listen(id_priv);
1504 if (ret)
1505 goto err;
1506 break;
1507 case RDMA_TRANSPORT_IWARP:
1508 ret = cma_iw_listen(id_priv, backlog);
1509 if (ret)
1510 goto err;
1511 break;
1512 default:
1513 ret = -ENOSYS;
1514 goto err;
1515 }
1516 } else
1517 cma_listen_on_all(id_priv);
1518
1519 return 0;
1520 err:
1521 id_priv->backlog = 0;
1522 cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
1523 return ret;
1524 }
1525 EXPORT_SYMBOL(rdma_listen);
1526
1527 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1528 {
1529 struct rdma_id_private *id_priv;
1530
1531 id_priv = container_of(id, struct rdma_id_private, id);
1532 id_priv->tos = (u8) tos;
1533 }
1534 EXPORT_SYMBOL(rdma_set_service_type);
1535
1536 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1537 void *context)
1538 {
1539 struct cma_work *work = context;
1540 struct rdma_route *route;
1541
1542 route = &work->id->id.route;
1543
1544 if (!status) {
1545 route->num_paths = 1;
1546 *route->path_rec = *path_rec;
1547 } else {
1548 work->old_state = CMA_ROUTE_QUERY;
1549 work->new_state = CMA_ADDR_RESOLVED;
1550 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1551 work->event.status = status;
1552 }
1553
1554 queue_work(cma_wq, &work->work);
1555 }
1556
1557 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1558 struct cma_work *work)
1559 {
1560 struct rdma_addr *addr = &id_priv->id.route.addr;
1561 struct ib_sa_path_rec path_rec;
1562 ib_sa_comp_mask comp_mask;
1563 struct sockaddr_in6 *sin6;
1564
1565 memset(&path_rec, 0, sizeof path_rec);
1566 rdma_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
1567 rdma_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
1568 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
1569 path_rec.numb_path = 1;
1570 path_rec.reversible = 1;
1571 path_rec.service_id = cma_get_service_id(id_priv->id.ps,
1572 (struct sockaddr *) &addr->dst_addr);
1573
1574 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1575 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1576 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1577
1578 if (addr->src_addr.ss_family == AF_INET) {
1579 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1580 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1581 } else {
1582 sin6 = (struct sockaddr_in6 *) &addr->src_addr;
1583 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1584 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1585 }
1586
1587 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1588 id_priv->id.port_num, &path_rec,
1589 comp_mask, timeout_ms,
1590 GFP_KERNEL, cma_query_handler,
1591 work, &id_priv->query);
1592
1593 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1594 }
1595
1596 static void cma_work_handler(struct work_struct *_work)
1597 {
1598 struct cma_work *work = container_of(_work, struct cma_work, work);
1599 struct rdma_id_private *id_priv = work->id;
1600 int destroy = 0;
1601
1602 mutex_lock(&id_priv->handler_mutex);
1603 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1604 goto out;
1605
1606 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1607 cma_exch(id_priv, CMA_DESTROYING);
1608 destroy = 1;
1609 }
1610 out:
1611 mutex_unlock(&id_priv->handler_mutex);
1612 cma_deref_id(id_priv);
1613 if (destroy)
1614 rdma_destroy_id(&id_priv->id);
1615 kfree(work);
1616 }
1617
1618 static void cma_ndev_work_handler(struct work_struct *_work)
1619 {
1620 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1621 struct rdma_id_private *id_priv = work->id;
1622 int destroy = 0;
1623
1624 mutex_lock(&id_priv->handler_mutex);
1625 if (id_priv->state == CMA_DESTROYING ||
1626 id_priv->state == CMA_DEVICE_REMOVAL)
1627 goto out;
1628
1629 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1630 cma_exch(id_priv, CMA_DESTROYING);
1631 destroy = 1;
1632 }
1633
1634 out:
1635 mutex_unlock(&id_priv->handler_mutex);
1636 cma_deref_id(id_priv);
1637 if (destroy)
1638 rdma_destroy_id(&id_priv->id);
1639 kfree(work);
1640 }
1641
1642 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1643 {
1644 struct rdma_route *route = &id_priv->id.route;
1645 struct cma_work *work;
1646 int ret;
1647
1648 work = kzalloc(sizeof *work, GFP_KERNEL);
1649 if (!work)
1650 return -ENOMEM;
1651
1652 work->id = id_priv;
1653 INIT_WORK(&work->work, cma_work_handler);
1654 work->old_state = CMA_ROUTE_QUERY;
1655 work->new_state = CMA_ROUTE_RESOLVED;
1656 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1657
1658 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1659 if (!route->path_rec) {
1660 ret = -ENOMEM;
1661 goto err1;
1662 }
1663
1664 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1665 if (ret)
1666 goto err2;
1667
1668 return 0;
1669 err2:
1670 kfree(route->path_rec);
1671 route->path_rec = NULL;
1672 err1:
1673 kfree(work);
1674 return ret;
1675 }
1676
1677 int rdma_set_ib_paths(struct rdma_cm_id *id,
1678 struct ib_sa_path_rec *path_rec, int num_paths)
1679 {
1680 struct rdma_id_private *id_priv;
1681 int ret;
1682
1683 id_priv = container_of(id, struct rdma_id_private, id);
1684 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_RESOLVED))
1685 return -EINVAL;
1686
1687 id->route.path_rec = kmalloc(sizeof *path_rec * num_paths, GFP_KERNEL);
1688 if (!id->route.path_rec) {
1689 ret = -ENOMEM;
1690 goto err;
1691 }
1692
1693 memcpy(id->route.path_rec, path_rec, sizeof *path_rec * num_paths);
1694 return 0;
1695 err:
1696 cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_ADDR_RESOLVED);
1697 return ret;
1698 }
1699 EXPORT_SYMBOL(rdma_set_ib_paths);
1700
1701 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1702 {
1703 struct cma_work *work;
1704
1705 work = kzalloc(sizeof *work, GFP_KERNEL);
1706 if (!work)
1707 return -ENOMEM;
1708
1709 work->id = id_priv;
1710 INIT_WORK(&work->work, cma_work_handler);
1711 work->old_state = CMA_ROUTE_QUERY;
1712 work->new_state = CMA_ROUTE_RESOLVED;
1713 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1714 queue_work(cma_wq, &work->work);
1715 return 0;
1716 }
1717
1718 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1719 {
1720 struct rdma_id_private *id_priv;
1721 int ret;
1722
1723 id_priv = container_of(id, struct rdma_id_private, id);
1724 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_QUERY))
1725 return -EINVAL;
1726
1727 atomic_inc(&id_priv->refcount);
1728 switch (rdma_node_get_transport(id->device->node_type)) {
1729 case RDMA_TRANSPORT_IB:
1730 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1731 break;
1732 case RDMA_TRANSPORT_IWARP:
1733 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1734 break;
1735 default:
1736 ret = -ENOSYS;
1737 break;
1738 }
1739 if (ret)
1740 goto err;
1741
1742 return 0;
1743 err:
1744 cma_comp_exch(id_priv, CMA_ROUTE_QUERY, CMA_ADDR_RESOLVED);
1745 cma_deref_id(id_priv);
1746 return ret;
1747 }
1748 EXPORT_SYMBOL(rdma_resolve_route);
1749
1750 static int cma_bind_loopback(struct rdma_id_private *id_priv)
1751 {
1752 struct cma_device *cma_dev;
1753 struct ib_port_attr port_attr;
1754 union ib_gid gid;
1755 u16 pkey;
1756 int ret;
1757 u8 p;
1758
1759 mutex_lock(&lock);
1760 if (list_empty(&dev_list)) {
1761 ret = -ENODEV;
1762 goto out;
1763 }
1764 list_for_each_entry(cma_dev, &dev_list, list)
1765 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
1766 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
1767 port_attr.state == IB_PORT_ACTIVE)
1768 goto port_found;
1769
1770 p = 1;
1771 cma_dev = list_entry(dev_list.next, struct cma_device, list);
1772
1773 port_found:
1774 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
1775 if (ret)
1776 goto out;
1777
1778 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
1779 if (ret)
1780 goto out;
1781
1782 id_priv->id.route.addr.dev_addr.dev_type =
1783 (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB) ?
1784 ARPHRD_INFINIBAND : ARPHRD_ETHER;
1785
1786 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1787 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1788 id_priv->id.port_num = p;
1789 cma_attach_to_dev(id_priv, cma_dev);
1790 out:
1791 mutex_unlock(&lock);
1792 return ret;
1793 }
1794
1795 static void addr_handler(int status, struct sockaddr *src_addr,
1796 struct rdma_dev_addr *dev_addr, void *context)
1797 {
1798 struct rdma_id_private *id_priv = context;
1799 struct rdma_cm_event event;
1800
1801 memset(&event, 0, sizeof event);
1802 mutex_lock(&id_priv->handler_mutex);
1803
1804 /*
1805 * Grab mutex to block rdma_destroy_id() from removing the device while
1806 * we're trying to acquire it.
1807 */
1808 mutex_lock(&lock);
1809 if (!cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_RESOLVED)) {
1810 mutex_unlock(&lock);
1811 goto out;
1812 }
1813
1814 if (!status && !id_priv->cma_dev)
1815 status = cma_acquire_dev(id_priv);
1816 mutex_unlock(&lock);
1817
1818 if (status) {
1819 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ADDR_BOUND))
1820 goto out;
1821 event.event = RDMA_CM_EVENT_ADDR_ERROR;
1822 event.status = status;
1823 } else {
1824 memcpy(&id_priv->id.route.addr.src_addr, src_addr,
1825 ip_addr_size(src_addr));
1826 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1827 }
1828
1829 if (id_priv->id.event_handler(&id_priv->id, &event)) {
1830 cma_exch(id_priv, CMA_DESTROYING);
1831 mutex_unlock(&id_priv->handler_mutex);
1832 cma_deref_id(id_priv);
1833 rdma_destroy_id(&id_priv->id);
1834 return;
1835 }
1836 out:
1837 mutex_unlock(&id_priv->handler_mutex);
1838 cma_deref_id(id_priv);
1839 }
1840
1841 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1842 {
1843 struct cma_work *work;
1844 struct sockaddr *src, *dst;
1845 union ib_gid gid;
1846 int ret;
1847
1848 work = kzalloc(sizeof *work, GFP_KERNEL);
1849 if (!work)
1850 return -ENOMEM;
1851
1852 if (!id_priv->cma_dev) {
1853 ret = cma_bind_loopback(id_priv);
1854 if (ret)
1855 goto err;
1856 }
1857
1858 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1859 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
1860
1861 src = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
1862 if (cma_zero_addr(src)) {
1863 dst = (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
1864 if ((src->sa_family = dst->sa_family) == AF_INET) {
1865 ((struct sockaddr_in *) src)->sin_addr.s_addr =
1866 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1867 } else {
1868 ipv6_addr_copy(&((struct sockaddr_in6 *) src)->sin6_addr,
1869 &((struct sockaddr_in6 *) dst)->sin6_addr);
1870 }
1871 }
1872
1873 work->id = id_priv;
1874 INIT_WORK(&work->work, cma_work_handler);
1875 work->old_state = CMA_ADDR_QUERY;
1876 work->new_state = CMA_ADDR_RESOLVED;
1877 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1878 queue_work(cma_wq, &work->work);
1879 return 0;
1880 err:
1881 kfree(work);
1882 return ret;
1883 }
1884
1885 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1886 struct sockaddr *dst_addr)
1887 {
1888 if (src_addr && src_addr->sa_family)
1889 return rdma_bind_addr(id, src_addr);
1890 else
1891 return cma_bind_any(id, dst_addr->sa_family);
1892 }
1893
1894 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1895 struct sockaddr *dst_addr, int timeout_ms)
1896 {
1897 struct rdma_id_private *id_priv;
1898 int ret;
1899
1900 id_priv = container_of(id, struct rdma_id_private, id);
1901 if (id_priv->state == CMA_IDLE) {
1902 ret = cma_bind_addr(id, src_addr, dst_addr);
1903 if (ret)
1904 return ret;
1905 }
1906
1907 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_ADDR_QUERY))
1908 return -EINVAL;
1909
1910 atomic_inc(&id_priv->refcount);
1911 memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
1912 if (cma_any_addr(dst_addr))
1913 ret = cma_resolve_loopback(id_priv);
1914 else
1915 ret = rdma_resolve_ip(&addr_client, (struct sockaddr *) &id->route.addr.src_addr,
1916 dst_addr, &id->route.addr.dev_addr,
1917 timeout_ms, addr_handler, id_priv);
1918 if (ret)
1919 goto err;
1920
1921 return 0;
1922 err:
1923 cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_BOUND);
1924 cma_deref_id(id_priv);
1925 return ret;
1926 }
1927 EXPORT_SYMBOL(rdma_resolve_addr);
1928
1929 static void cma_bind_port(struct rdma_bind_list *bind_list,
1930 struct rdma_id_private *id_priv)
1931 {
1932 struct sockaddr_in *sin;
1933
1934 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1935 sin->sin_port = htons(bind_list->port);
1936 id_priv->bind_list = bind_list;
1937 hlist_add_head(&id_priv->node, &bind_list->owners);
1938 }
1939
1940 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
1941 unsigned short snum)
1942 {
1943 struct rdma_bind_list *bind_list;
1944 int port, ret;
1945
1946 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1947 if (!bind_list)
1948 return -ENOMEM;
1949
1950 do {
1951 ret = idr_get_new_above(ps, bind_list, snum, &port);
1952 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1953
1954 if (ret)
1955 goto err1;
1956
1957 if (port != snum) {
1958 ret = -EADDRNOTAVAIL;
1959 goto err2;
1960 }
1961
1962 bind_list->ps = ps;
1963 bind_list->port = (unsigned short) port;
1964 cma_bind_port(bind_list, id_priv);
1965 return 0;
1966 err2:
1967 idr_remove(ps, port);
1968 err1:
1969 kfree(bind_list);
1970 return ret;
1971 }
1972
1973 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
1974 {
1975 struct rdma_bind_list *bind_list;
1976 int port, ret, low, high;
1977
1978 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1979 if (!bind_list)
1980 return -ENOMEM;
1981
1982 retry:
1983 /* FIXME: add proper port randomization per like inet_csk_get_port */
1984 do {
1985 ret = idr_get_new_above(ps, bind_list, next_port, &port);
1986 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1987
1988 if (ret)
1989 goto err1;
1990
1991 inet_get_local_port_range(&low, &high);
1992 if (port > high) {
1993 if (next_port != low) {
1994 idr_remove(ps, port);
1995 next_port = low;
1996 goto retry;
1997 }
1998 ret = -EADDRNOTAVAIL;
1999 goto err2;
2000 }
2001
2002 if (port == high)
2003 next_port = low;
2004 else
2005 next_port = port + 1;
2006
2007 bind_list->ps = ps;
2008 bind_list->port = (unsigned short) port;
2009 cma_bind_port(bind_list, id_priv);
2010 return 0;
2011 err2:
2012 idr_remove(ps, port);
2013 err1:
2014 kfree(bind_list);
2015 return ret;
2016 }
2017
2018 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2019 {
2020 struct rdma_id_private *cur_id;
2021 struct sockaddr_in *sin, *cur_sin;
2022 struct rdma_bind_list *bind_list;
2023 struct hlist_node *node;
2024 unsigned short snum;
2025
2026 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
2027 snum = ntohs(sin->sin_port);
2028 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2029 return -EACCES;
2030
2031 bind_list = idr_find(ps, snum);
2032 if (!bind_list)
2033 return cma_alloc_port(ps, id_priv, snum);
2034
2035 /*
2036 * We don't support binding to any address if anyone is bound to
2037 * a specific address on the same port.
2038 */
2039 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr))
2040 return -EADDRNOTAVAIL;
2041
2042 hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
2043 if (cma_any_addr((struct sockaddr *) &cur_id->id.route.addr.src_addr))
2044 return -EADDRNOTAVAIL;
2045
2046 cur_sin = (struct sockaddr_in *) &cur_id->id.route.addr.src_addr;
2047 if (sin->sin_addr.s_addr == cur_sin->sin_addr.s_addr)
2048 return -EADDRINUSE;
2049 }
2050
2051 cma_bind_port(bind_list, id_priv);
2052 return 0;
2053 }
2054
2055 static int cma_get_port(struct rdma_id_private *id_priv)
2056 {
2057 struct idr *ps;
2058 int ret;
2059
2060 switch (id_priv->id.ps) {
2061 case RDMA_PS_SDP:
2062 ps = &sdp_ps;
2063 break;
2064 case RDMA_PS_TCP:
2065 ps = &tcp_ps;
2066 break;
2067 case RDMA_PS_UDP:
2068 ps = &udp_ps;
2069 break;
2070 case RDMA_PS_IPOIB:
2071 ps = &ipoib_ps;
2072 break;
2073 default:
2074 return -EPROTONOSUPPORT;
2075 }
2076
2077 mutex_lock(&lock);
2078 if (cma_any_port((struct sockaddr *) &id_priv->id.route.addr.src_addr))
2079 ret = cma_alloc_any_port(ps, id_priv);
2080 else
2081 ret = cma_use_port(ps, id_priv);
2082 mutex_unlock(&lock);
2083
2084 return ret;
2085 }
2086
2087 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2088 {
2089 struct rdma_id_private *id_priv;
2090 int ret;
2091
2092 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
2093 return -EAFNOSUPPORT;
2094
2095 id_priv = container_of(id, struct rdma_id_private, id);
2096 if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND))
2097 return -EINVAL;
2098
2099 if (cma_loopback_addr(addr)) {
2100 ret = cma_bind_loopback(id_priv);
2101 } else if (!cma_zero_addr(addr)) {
2102 ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
2103 if (ret)
2104 goto err1;
2105
2106 mutex_lock(&lock);
2107 ret = cma_acquire_dev(id_priv);
2108 mutex_unlock(&lock);
2109 if (ret)
2110 goto err1;
2111 }
2112
2113 memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
2114 ret = cma_get_port(id_priv);
2115 if (ret)
2116 goto err2;
2117
2118 return 0;
2119 err2:
2120 if (id_priv->cma_dev) {
2121 mutex_lock(&lock);
2122 cma_detach_from_dev(id_priv);
2123 mutex_unlock(&lock);
2124 }
2125 err1:
2126 cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_IDLE);
2127 return ret;
2128 }
2129 EXPORT_SYMBOL(rdma_bind_addr);
2130
2131 static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
2132 struct rdma_route *route)
2133 {
2134 struct cma_hdr *cma_hdr;
2135 struct sdp_hh *sdp_hdr;
2136
2137 if (route->addr.src_addr.ss_family == AF_INET) {
2138 struct sockaddr_in *src4, *dst4;
2139
2140 src4 = (struct sockaddr_in *) &route->addr.src_addr;
2141 dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
2142
2143 switch (ps) {
2144 case RDMA_PS_SDP:
2145 sdp_hdr = hdr;
2146 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2147 return -EINVAL;
2148 sdp_set_ip_ver(sdp_hdr, 4);
2149 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2150 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2151 sdp_hdr->port = src4->sin_port;
2152 break;
2153 default:
2154 cma_hdr = hdr;
2155 cma_hdr->cma_version = CMA_VERSION;
2156 cma_set_ip_ver(cma_hdr, 4);
2157 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2158 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2159 cma_hdr->port = src4->sin_port;
2160 break;
2161 }
2162 } else {
2163 struct sockaddr_in6 *src6, *dst6;
2164
2165 src6 = (struct sockaddr_in6 *) &route->addr.src_addr;
2166 dst6 = (struct sockaddr_in6 *) &route->addr.dst_addr;
2167
2168 switch (ps) {
2169 case RDMA_PS_SDP:
2170 sdp_hdr = hdr;
2171 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2172 return -EINVAL;
2173 sdp_set_ip_ver(sdp_hdr, 6);
2174 sdp_hdr->src_addr.ip6 = src6->sin6_addr;
2175 sdp_hdr->dst_addr.ip6 = dst6->sin6_addr;
2176 sdp_hdr->port = src6->sin6_port;
2177 break;
2178 default:
2179 cma_hdr = hdr;
2180 cma_hdr->cma_version = CMA_VERSION;
2181 cma_set_ip_ver(cma_hdr, 6);
2182 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2183 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2184 cma_hdr->port = src6->sin6_port;
2185 break;
2186 }
2187 }
2188 return 0;
2189 }
2190
2191 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2192 struct ib_cm_event *ib_event)
2193 {
2194 struct rdma_id_private *id_priv = cm_id->context;
2195 struct rdma_cm_event event;
2196 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2197 int ret = 0;
2198
2199 if (cma_disable_callback(id_priv, CMA_CONNECT))
2200 return 0;
2201
2202 memset(&event, 0, sizeof event);
2203 switch (ib_event->event) {
2204 case IB_CM_SIDR_REQ_ERROR:
2205 event.event = RDMA_CM_EVENT_UNREACHABLE;
2206 event.status = -ETIMEDOUT;
2207 break;
2208 case IB_CM_SIDR_REP_RECEIVED:
2209 event.param.ud.private_data = ib_event->private_data;
2210 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2211 if (rep->status != IB_SIDR_SUCCESS) {
2212 event.event = RDMA_CM_EVENT_UNREACHABLE;
2213 event.status = ib_event->param.sidr_rep_rcvd.status;
2214 break;
2215 }
2216 ret = cma_set_qkey(id_priv);
2217 if (ret) {
2218 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2219 event.status = -EINVAL;
2220 break;
2221 }
2222 if (id_priv->qkey != rep->qkey) {
2223 event.event = RDMA_CM_EVENT_UNREACHABLE;
2224 event.status = -EINVAL;
2225 break;
2226 }
2227 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2228 id_priv->id.route.path_rec,
2229 &event.param.ud.ah_attr);
2230 event.param.ud.qp_num = rep->qpn;
2231 event.param.ud.qkey = rep->qkey;
2232 event.event = RDMA_CM_EVENT_ESTABLISHED;
2233 event.status = 0;
2234 break;
2235 default:
2236 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2237 ib_event->event);
2238 goto out;
2239 }
2240
2241 ret = id_priv->id.event_handler(&id_priv->id, &event);
2242 if (ret) {
2243 /* Destroy the CM ID by returning a non-zero value. */
2244 id_priv->cm_id.ib = NULL;
2245 cma_exch(id_priv, CMA_DESTROYING);
2246 mutex_unlock(&id_priv->handler_mutex);
2247 rdma_destroy_id(&id_priv->id);
2248 return ret;
2249 }
2250 out:
2251 mutex_unlock(&id_priv->handler_mutex);
2252 return ret;
2253 }
2254
2255 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2256 struct rdma_conn_param *conn_param)
2257 {
2258 struct ib_cm_sidr_req_param req;
2259 struct rdma_route *route;
2260 int ret;
2261
2262 req.private_data_len = sizeof(struct cma_hdr) +
2263 conn_param->private_data_len;
2264 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2265 if (!req.private_data)
2266 return -ENOMEM;
2267
2268 if (conn_param->private_data && conn_param->private_data_len)
2269 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2270 conn_param->private_data, conn_param->private_data_len);
2271
2272 route = &id_priv->id.route;
2273 ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
2274 if (ret)
2275 goto out;
2276
2277 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device,
2278 cma_sidr_rep_handler, id_priv);
2279 if (IS_ERR(id_priv->cm_id.ib)) {
2280 ret = PTR_ERR(id_priv->cm_id.ib);
2281 goto out;
2282 }
2283
2284 req.path = route->path_rec;
2285 req.service_id = cma_get_service_id(id_priv->id.ps,
2286 (struct sockaddr *) &route->addr.dst_addr);
2287 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2288 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2289
2290 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2291 if (ret) {
2292 ib_destroy_cm_id(id_priv->cm_id.ib);
2293 id_priv->cm_id.ib = NULL;
2294 }
2295 out:
2296 kfree(req.private_data);
2297 return ret;
2298 }
2299
2300 static int cma_connect_ib(struct rdma_id_private *id_priv,
2301 struct rdma_conn_param *conn_param)
2302 {
2303 struct ib_cm_req_param req;
2304 struct rdma_route *route;
2305 void *private_data;
2306 int offset, ret;
2307
2308 memset(&req, 0, sizeof req);
2309 offset = cma_user_data_offset(id_priv->id.ps);
2310 req.private_data_len = offset + conn_param->private_data_len;
2311 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2312 if (!private_data)
2313 return -ENOMEM;
2314
2315 if (conn_param->private_data && conn_param->private_data_len)
2316 memcpy(private_data + offset, conn_param->private_data,
2317 conn_param->private_data_len);
2318
2319 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_ib_handler,
2320 id_priv);
2321 if (IS_ERR(id_priv->cm_id.ib)) {
2322 ret = PTR_ERR(id_priv->cm_id.ib);
2323 goto out;
2324 }
2325
2326 route = &id_priv->id.route;
2327 ret = cma_format_hdr(private_data, id_priv->id.ps, route);
2328 if (ret)
2329 goto out;
2330 req.private_data = private_data;
2331
2332 req.primary_path = &route->path_rec[0];
2333 if (route->num_paths == 2)
2334 req.alternate_path = &route->path_rec[1];
2335
2336 req.service_id = cma_get_service_id(id_priv->id.ps,
2337 (struct sockaddr *) &route->addr.dst_addr);
2338 req.qp_num = id_priv->qp_num;
2339 req.qp_type = IB_QPT_RC;
2340 req.starting_psn = id_priv->seq_num;
2341 req.responder_resources = conn_param->responder_resources;
2342 req.initiator_depth = conn_param->initiator_depth;
2343 req.flow_control = conn_param->flow_control;
2344 req.retry_count = conn_param->retry_count;
2345 req.rnr_retry_count = conn_param->rnr_retry_count;
2346 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2347 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2348 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2349 req.srq = id_priv->srq ? 1 : 0;
2350
2351 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2352 out:
2353 if (ret && !IS_ERR(id_priv->cm_id.ib)) {
2354 ib_destroy_cm_id(id_priv->cm_id.ib);
2355 id_priv->cm_id.ib = NULL;
2356 }
2357
2358 kfree(private_data);
2359 return ret;
2360 }
2361
2362 static int cma_connect_iw(struct rdma_id_private *id_priv,
2363 struct rdma_conn_param *conn_param)
2364 {
2365 struct iw_cm_id *cm_id;
2366 struct sockaddr_in* sin;
2367 int ret;
2368 struct iw_cm_conn_param iw_param;
2369
2370 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2371 if (IS_ERR(cm_id)) {
2372 ret = PTR_ERR(cm_id);
2373 goto out;
2374 }
2375
2376 id_priv->cm_id.iw = cm_id;
2377
2378 sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
2379 cm_id->local_addr = *sin;
2380
2381 sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
2382 cm_id->remote_addr = *sin;
2383
2384 ret = cma_modify_qp_rtr(id_priv, conn_param);
2385 if (ret)
2386 goto out;
2387
2388 iw_param.ord = conn_param->initiator_depth;
2389 iw_param.ird = conn_param->responder_resources;
2390 iw_param.private_data = conn_param->private_data;
2391 iw_param.private_data_len = conn_param->private_data_len;
2392 if (id_priv->id.qp)
2393 iw_param.qpn = id_priv->qp_num;
2394 else
2395 iw_param.qpn = conn_param->qp_num;
2396 ret = iw_cm_connect(cm_id, &iw_param);
2397 out:
2398 if (ret && !IS_ERR(cm_id)) {
2399 iw_destroy_cm_id(cm_id);
2400 id_priv->cm_id.iw = NULL;
2401 }
2402 return ret;
2403 }
2404
2405 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2406 {
2407 struct rdma_id_private *id_priv;
2408 int ret;
2409
2410 id_priv = container_of(id, struct rdma_id_private, id);
2411 if (!cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_CONNECT))
2412 return -EINVAL;
2413
2414 if (!id->qp) {
2415 id_priv->qp_num = conn_param->qp_num;
2416 id_priv->srq = conn_param->srq;
2417 }
2418
2419 switch (rdma_node_get_transport(id->device->node_type)) {
2420 case RDMA_TRANSPORT_IB:
2421 if (cma_is_ud_ps(id->ps))
2422 ret = cma_resolve_ib_udp(id_priv, conn_param);
2423 else
2424 ret = cma_connect_ib(id_priv, conn_param);
2425 break;
2426 case RDMA_TRANSPORT_IWARP:
2427 ret = cma_connect_iw(id_priv, conn_param);
2428 break;
2429 default:
2430 ret = -ENOSYS;
2431 break;
2432 }
2433 if (ret)
2434 goto err;
2435
2436 return 0;
2437 err:
2438 cma_comp_exch(id_priv, CMA_CONNECT, CMA_ROUTE_RESOLVED);
2439 return ret;
2440 }
2441 EXPORT_SYMBOL(rdma_connect);
2442
2443 static int cma_accept_ib(struct rdma_id_private *id_priv,
2444 struct rdma_conn_param *conn_param)
2445 {
2446 struct ib_cm_rep_param rep;
2447 int ret;
2448
2449 ret = cma_modify_qp_rtr(id_priv, conn_param);
2450 if (ret)
2451 goto out;
2452
2453 ret = cma_modify_qp_rts(id_priv, conn_param);
2454 if (ret)
2455 goto out;
2456
2457 memset(&rep, 0, sizeof rep);
2458 rep.qp_num = id_priv->qp_num;
2459 rep.starting_psn = id_priv->seq_num;
2460 rep.private_data = conn_param->private_data;
2461 rep.private_data_len = conn_param->private_data_len;
2462 rep.responder_resources = conn_param->responder_resources;
2463 rep.initiator_depth = conn_param->initiator_depth;
2464 rep.failover_accepted = 0;
2465 rep.flow_control = conn_param->flow_control;
2466 rep.rnr_retry_count = conn_param->rnr_retry_count;
2467 rep.srq = id_priv->srq ? 1 : 0;
2468
2469 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2470 out:
2471 return ret;
2472 }
2473
2474 static int cma_accept_iw(struct rdma_id_private *id_priv,
2475 struct rdma_conn_param *conn_param)
2476 {
2477 struct iw_cm_conn_param iw_param;
2478 int ret;
2479
2480 ret = cma_modify_qp_rtr(id_priv, conn_param);
2481 if (ret)
2482 return ret;
2483
2484 iw_param.ord = conn_param->initiator_depth;
2485 iw_param.ird = conn_param->responder_resources;
2486 iw_param.private_data = conn_param->private_data;
2487 iw_param.private_data_len = conn_param->private_data_len;
2488 if (id_priv->id.qp) {
2489 iw_param.qpn = id_priv->qp_num;
2490 } else
2491 iw_param.qpn = conn_param->qp_num;
2492
2493 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2494 }
2495
2496 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2497 enum ib_cm_sidr_status status,
2498 const void *private_data, int private_data_len)
2499 {
2500 struct ib_cm_sidr_rep_param rep;
2501 int ret;
2502
2503 memset(&rep, 0, sizeof rep);
2504 rep.status = status;
2505 if (status == IB_SIDR_SUCCESS) {
2506 ret = cma_set_qkey(id_priv);
2507 if (ret)
2508 return ret;
2509 rep.qp_num = id_priv->qp_num;
2510 rep.qkey = id_priv->qkey;
2511 }
2512 rep.private_data = private_data;
2513 rep.private_data_len = private_data_len;
2514
2515 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2516 }
2517
2518 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2519 {
2520 struct rdma_id_private *id_priv;
2521 int ret;
2522
2523 id_priv = container_of(id, struct rdma_id_private, id);
2524 if (!cma_comp(id_priv, CMA_CONNECT))
2525 return -EINVAL;
2526
2527 if (!id->qp && conn_param) {
2528 id_priv->qp_num = conn_param->qp_num;
2529 id_priv->srq = conn_param->srq;
2530 }
2531
2532 switch (rdma_node_get_transport(id->device->node_type)) {
2533 case RDMA_TRANSPORT_IB:
2534 if (cma_is_ud_ps(id->ps))
2535 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2536 conn_param->private_data,
2537 conn_param->private_data_len);
2538 else if (conn_param)
2539 ret = cma_accept_ib(id_priv, conn_param);
2540 else
2541 ret = cma_rep_recv(id_priv);
2542 break;
2543 case RDMA_TRANSPORT_IWARP:
2544 ret = cma_accept_iw(id_priv, conn_param);
2545 break;
2546 default:
2547 ret = -ENOSYS;
2548 break;
2549 }
2550
2551 if (ret)
2552 goto reject;
2553
2554 return 0;
2555 reject:
2556 cma_modify_qp_err(id_priv);
2557 rdma_reject(id, NULL, 0);
2558 return ret;
2559 }
2560 EXPORT_SYMBOL(rdma_accept);
2561
2562 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2563 {
2564 struct rdma_id_private *id_priv;
2565 int ret;
2566
2567 id_priv = container_of(id, struct rdma_id_private, id);
2568 if (!cma_has_cm_dev(id_priv))
2569 return -EINVAL;
2570
2571 switch (id->device->node_type) {
2572 case RDMA_NODE_IB_CA:
2573 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2574 break;
2575 default:
2576 ret = 0;
2577 break;
2578 }
2579 return ret;
2580 }
2581 EXPORT_SYMBOL(rdma_notify);
2582
2583 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2584 u8 private_data_len)
2585 {
2586 struct rdma_id_private *id_priv;
2587 int ret;
2588
2589 id_priv = container_of(id, struct rdma_id_private, id);
2590 if (!cma_has_cm_dev(id_priv))
2591 return -EINVAL;
2592
2593 switch (rdma_node_get_transport(id->device->node_type)) {
2594 case RDMA_TRANSPORT_IB:
2595 if (cma_is_ud_ps(id->ps))
2596 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
2597 private_data, private_data_len);
2598 else
2599 ret = ib_send_cm_rej(id_priv->cm_id.ib,
2600 IB_CM_REJ_CONSUMER_DEFINED, NULL,
2601 0, private_data, private_data_len);
2602 break;
2603 case RDMA_TRANSPORT_IWARP:
2604 ret = iw_cm_reject(id_priv->cm_id.iw,
2605 private_data, private_data_len);
2606 break;
2607 default:
2608 ret = -ENOSYS;
2609 break;
2610 }
2611 return ret;
2612 }
2613 EXPORT_SYMBOL(rdma_reject);
2614
2615 int rdma_disconnect(struct rdma_cm_id *id)
2616 {
2617 struct rdma_id_private *id_priv;
2618 int ret;
2619
2620 id_priv = container_of(id, struct rdma_id_private, id);
2621 if (!cma_has_cm_dev(id_priv))
2622 return -EINVAL;
2623
2624 switch (rdma_node_get_transport(id->device->node_type)) {
2625 case RDMA_TRANSPORT_IB:
2626 ret = cma_modify_qp_err(id_priv);
2627 if (ret)
2628 goto out;
2629 /* Initiate or respond to a disconnect. */
2630 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
2631 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
2632 break;
2633 case RDMA_TRANSPORT_IWARP:
2634 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
2635 break;
2636 default:
2637 ret = -EINVAL;
2638 break;
2639 }
2640 out:
2641 return ret;
2642 }
2643 EXPORT_SYMBOL(rdma_disconnect);
2644
2645 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
2646 {
2647 struct rdma_id_private *id_priv;
2648 struct cma_multicast *mc = multicast->context;
2649 struct rdma_cm_event event;
2650 int ret;
2651
2652 id_priv = mc->id_priv;
2653 if (cma_disable_callback(id_priv, CMA_ADDR_BOUND) &&
2654 cma_disable_callback(id_priv, CMA_ADDR_RESOLVED))
2655 return 0;
2656
2657 mutex_lock(&id_priv->qp_mutex);
2658 if (!status && id_priv->id.qp)
2659 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
2660 multicast->rec.mlid);
2661 mutex_unlock(&id_priv->qp_mutex);
2662
2663 memset(&event, 0, sizeof event);
2664 event.status = status;
2665 event.param.ud.private_data = mc->context;
2666 if (!status) {
2667 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
2668 ib_init_ah_from_mcmember(id_priv->id.device,
2669 id_priv->id.port_num, &multicast->rec,
2670 &event.param.ud.ah_attr);
2671 event.param.ud.qp_num = 0xFFFFFF;
2672 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
2673 } else
2674 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
2675
2676 ret = id_priv->id.event_handler(&id_priv->id, &event);
2677 if (ret) {
2678 cma_exch(id_priv, CMA_DESTROYING);
2679 mutex_unlock(&id_priv->handler_mutex);
2680 rdma_destroy_id(&id_priv->id);
2681 return 0;
2682 }
2683
2684 mutex_unlock(&id_priv->handler_mutex);
2685 return 0;
2686 }
2687
2688 static void cma_set_mgid(struct rdma_id_private *id_priv,
2689 struct sockaddr *addr, union ib_gid *mgid)
2690 {
2691 unsigned char mc_map[MAX_ADDR_LEN];
2692 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2693 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
2694 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
2695
2696 if (cma_any_addr(addr)) {
2697 memset(mgid, 0, sizeof *mgid);
2698 } else if ((addr->sa_family == AF_INET6) &&
2699 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
2700 0xFF10A01B)) {
2701 /* IPv6 address is an SA assigned MGID. */
2702 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
2703 } else if ((addr->sa_family == AF_INET6)) {
2704 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
2705 if (id_priv->id.ps == RDMA_PS_UDP)
2706 mc_map[7] = 0x01; /* Use RDMA CM signature */
2707 *mgid = *(union ib_gid *) (mc_map + 4);
2708 } else {
2709 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
2710 if (id_priv->id.ps == RDMA_PS_UDP)
2711 mc_map[7] = 0x01; /* Use RDMA CM signature */
2712 *mgid = *(union ib_gid *) (mc_map + 4);
2713 }
2714 }
2715
2716 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
2717 struct cma_multicast *mc)
2718 {
2719 struct ib_sa_mcmember_rec rec;
2720 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2721 ib_sa_comp_mask comp_mask;
2722 int ret;
2723
2724 ib_addr_get_mgid(dev_addr, &rec.mgid);
2725 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
2726 &rec.mgid, &rec);
2727 if (ret)
2728 return ret;
2729
2730 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
2731 if (id_priv->id.ps == RDMA_PS_UDP)
2732 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
2733 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
2734 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2735 rec.join_state = 1;
2736
2737 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
2738 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
2739 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
2740 IB_SA_MCMEMBER_REC_FLOW_LABEL |
2741 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
2742
2743 if (id_priv->id.ps == RDMA_PS_IPOIB)
2744 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
2745 IB_SA_MCMEMBER_REC_RATE_SELECTOR;
2746
2747 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
2748 id_priv->id.port_num, &rec,
2749 comp_mask, GFP_KERNEL,
2750 cma_ib_mc_handler, mc);
2751 if (IS_ERR(mc->multicast.ib))
2752 return PTR_ERR(mc->multicast.ib);
2753
2754 return 0;
2755 }
2756
2757 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
2758 void *context)
2759 {
2760 struct rdma_id_private *id_priv;
2761 struct cma_multicast *mc;
2762 int ret;
2763
2764 id_priv = container_of(id, struct rdma_id_private, id);
2765 if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
2766 !cma_comp(id_priv, CMA_ADDR_RESOLVED))
2767 return -EINVAL;
2768
2769 mc = kmalloc(sizeof *mc, GFP_KERNEL);
2770 if (!mc)
2771 return -ENOMEM;
2772
2773 memcpy(&mc->addr, addr, ip_addr_size(addr));
2774 mc->context = context;
2775 mc->id_priv = id_priv;
2776
2777 spin_lock(&id_priv->lock);
2778 list_add(&mc->list, &id_priv->mc_list);
2779 spin_unlock(&id_priv->lock);
2780
2781 switch (rdma_node_get_transport(id->device->node_type)) {
2782 case RDMA_TRANSPORT_IB:
2783 ret = cma_join_ib_multicast(id_priv, mc);
2784 break;
2785 default:
2786 ret = -ENOSYS;
2787 break;
2788 }
2789
2790 if (ret) {
2791 spin_lock_irq(&id_priv->lock);
2792 list_del(&mc->list);
2793 spin_unlock_irq(&id_priv->lock);
2794 kfree(mc);
2795 }
2796 return ret;
2797 }
2798 EXPORT_SYMBOL(rdma_join_multicast);
2799
2800 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
2801 {
2802 struct rdma_id_private *id_priv;
2803 struct cma_multicast *mc;
2804
2805 id_priv = container_of(id, struct rdma_id_private, id);
2806 spin_lock_irq(&id_priv->lock);
2807 list_for_each_entry(mc, &id_priv->mc_list, list) {
2808 if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
2809 list_del(&mc->list);
2810 spin_unlock_irq(&id_priv->lock);
2811
2812 if (id->qp)
2813 ib_detach_mcast(id->qp,
2814 &mc->multicast.ib->rec.mgid,
2815 mc->multicast.ib->rec.mlid);
2816 ib_sa_free_multicast(mc->multicast.ib);
2817 kfree(mc);
2818 return;
2819 }
2820 }
2821 spin_unlock_irq(&id_priv->lock);
2822 }
2823 EXPORT_SYMBOL(rdma_leave_multicast);
2824
2825 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
2826 {
2827 struct rdma_dev_addr *dev_addr;
2828 struct cma_ndev_work *work;
2829
2830 dev_addr = &id_priv->id.route.addr.dev_addr;
2831
2832 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
2833 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
2834 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
2835 ndev->name, &id_priv->id);
2836 work = kzalloc(sizeof *work, GFP_KERNEL);
2837 if (!work)
2838 return -ENOMEM;
2839
2840 INIT_WORK(&work->work, cma_ndev_work_handler);
2841 work->id = id_priv;
2842 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
2843 atomic_inc(&id_priv->refcount);
2844 queue_work(cma_wq, &work->work);
2845 }
2846
2847 return 0;
2848 }
2849
2850 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
2851 void *ctx)
2852 {
2853 struct net_device *ndev = (struct net_device *)ctx;
2854 struct cma_device *cma_dev;
2855 struct rdma_id_private *id_priv;
2856 int ret = NOTIFY_DONE;
2857
2858 if (dev_net(ndev) != &init_net)
2859 return NOTIFY_DONE;
2860
2861 if (event != NETDEV_BONDING_FAILOVER)
2862 return NOTIFY_DONE;
2863
2864 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
2865 return NOTIFY_DONE;
2866
2867 mutex_lock(&lock);
2868 list_for_each_entry(cma_dev, &dev_list, list)
2869 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
2870 ret = cma_netdev_change(ndev, id_priv);
2871 if (ret)
2872 goto out;
2873 }
2874
2875 out:
2876 mutex_unlock(&lock);
2877 return ret;
2878 }
2879
2880 static struct notifier_block cma_nb = {
2881 .notifier_call = cma_netdev_callback
2882 };
2883
2884 static void cma_add_one(struct ib_device *device)
2885 {
2886 struct cma_device *cma_dev;
2887 struct rdma_id_private *id_priv;
2888
2889 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
2890 if (!cma_dev)
2891 return;
2892
2893 cma_dev->device = device;
2894
2895 init_completion(&cma_dev->comp);
2896 atomic_set(&cma_dev->refcount, 1);
2897 INIT_LIST_HEAD(&cma_dev->id_list);
2898 ib_set_client_data(device, &cma_client, cma_dev);
2899
2900 mutex_lock(&lock);
2901 list_add_tail(&cma_dev->list, &dev_list);
2902 list_for_each_entry(id_priv, &listen_any_list, list)
2903 cma_listen_on_dev(id_priv, cma_dev);
2904 mutex_unlock(&lock);
2905 }
2906
2907 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
2908 {
2909 struct rdma_cm_event event;
2910 enum cma_state state;
2911 int ret = 0;
2912
2913 /* Record that we want to remove the device */
2914 state = cma_exch(id_priv, CMA_DEVICE_REMOVAL);
2915 if (state == CMA_DESTROYING)
2916 return 0;
2917
2918 cma_cancel_operation(id_priv, state);
2919 mutex_lock(&id_priv->handler_mutex);
2920
2921 /* Check for destruction from another callback. */
2922 if (!cma_comp(id_priv, CMA_DEVICE_REMOVAL))
2923 goto out;
2924
2925 memset(&event, 0, sizeof event);
2926 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
2927 ret = id_priv->id.event_handler(&id_priv->id, &event);
2928 out:
2929 mutex_unlock(&id_priv->handler_mutex);
2930 return ret;
2931 }
2932
2933 static void cma_process_remove(struct cma_device *cma_dev)
2934 {
2935 struct rdma_id_private *id_priv;
2936 int ret;
2937
2938 mutex_lock(&lock);
2939 while (!list_empty(&cma_dev->id_list)) {
2940 id_priv = list_entry(cma_dev->id_list.next,
2941 struct rdma_id_private, list);
2942
2943 list_del(&id_priv->listen_list);
2944 list_del_init(&id_priv->list);
2945 atomic_inc(&id_priv->refcount);
2946 mutex_unlock(&lock);
2947
2948 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
2949 cma_deref_id(id_priv);
2950 if (ret)
2951 rdma_destroy_id(&id_priv->id);
2952
2953 mutex_lock(&lock);
2954 }
2955 mutex_unlock(&lock);
2956
2957 cma_deref_dev(cma_dev);
2958 wait_for_completion(&cma_dev->comp);
2959 }
2960
2961 static void cma_remove_one(struct ib_device *device)
2962 {
2963 struct cma_device *cma_dev;
2964
2965 cma_dev = ib_get_client_data(device, &cma_client);
2966 if (!cma_dev)
2967 return;
2968
2969 mutex_lock(&lock);
2970 list_del(&cma_dev->list);
2971 mutex_unlock(&lock);
2972
2973 cma_process_remove(cma_dev);
2974 kfree(cma_dev);
2975 }
2976
2977 static int __init cma_init(void)
2978 {
2979 int ret, low, high, remaining;
2980
2981 get_random_bytes(&next_port, sizeof next_port);
2982 inet_get_local_port_range(&low, &high);
2983 remaining = (high - low) + 1;
2984 next_port = ((unsigned int) next_port % remaining) + low;
2985
2986 cma_wq = create_singlethread_workqueue("rdma_cm");
2987 if (!cma_wq)
2988 return -ENOMEM;
2989
2990 ib_sa_register_client(&sa_client);
2991 rdma_addr_register_client(&addr_client);
2992 register_netdevice_notifier(&cma_nb);
2993
2994 ret = ib_register_client(&cma_client);
2995 if (ret)
2996 goto err;
2997 return 0;
2998
2999 err:
3000 unregister_netdevice_notifier(&cma_nb);
3001 rdma_addr_unregister_client(&addr_client);
3002 ib_sa_unregister_client(&sa_client);
3003 destroy_workqueue(cma_wq);
3004 return ret;
3005 }
3006
3007 static void __exit cma_cleanup(void)
3008 {
3009 ib_unregister_client(&cma_client);
3010 unregister_netdevice_notifier(&cma_nb);
3011 rdma_addr_unregister_client(&addr_client);
3012 ib_sa_unregister_client(&sa_client);
3013 destroy_workqueue(cma_wq);
3014 idr_destroy(&sdp_ps);
3015 idr_destroy(&tcp_ps);
3016 idr_destroy(&udp_ps);
3017 idr_destroy(&ipoib_ps);
3018 }
3019
3020 module_init(cma_init);
3021 module_exit(cma_cleanup);
Linux kernel - Deliverables
This page took 0.14777 seconds and 6 git commands to generate.