projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
RDMA/cma: Add RDMA_CM_EVENT_ADDR_CHANGE event
[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
46 #include <rdma/rdma_cm.h>
47 #include <rdma/rdma_cm_ib.h>
48 #include <rdma/ib_cache.h>
49 #include <rdma/ib_cm.h>
50 #include <rdma/ib_sa.h>
51 #include <rdma/iw_cm.h>
52
53 MODULE_AUTHOR("Sean Hefty");
54 MODULE_DESCRIPTION("Generic RDMA CM Agent");
55 MODULE_LICENSE("Dual BSD/GPL");
56
57 #define CMA_CM_RESPONSE_TIMEOUT 20
58 #define CMA_MAX_CM_RETRIES 15
59 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
60
61 static void cma_add_one(struct ib_device *device);
62 static void cma_remove_one(struct ib_device *device);
63
64 static struct ib_client cma_client = {
65 .name = "cma",
66 .add = cma_add_one,
67 .remove = cma_remove_one
68 };
69
70 static struct ib_sa_client sa_client;
71 static struct rdma_addr_client addr_client;
72 static LIST_HEAD(dev_list);
73 static LIST_HEAD(listen_any_list);
74 static DEFINE_MUTEX(lock);
75 static struct workqueue_struct *cma_wq;
76 static DEFINE_IDR(sdp_ps);
77 static DEFINE_IDR(tcp_ps);
78 static DEFINE_IDR(udp_ps);
79 static DEFINE_IDR(ipoib_ps);
80 static int next_port;
81
82 struct cma_device {
83 struct list_head list;
84 struct ib_device *device;
85 struct completion comp;
86 atomic_t refcount;
87 struct list_head id_list;
88 };
89
90 enum cma_state {
91 CMA_IDLE,
92 CMA_ADDR_QUERY,
93 CMA_ADDR_RESOLVED,
94 CMA_ROUTE_QUERY,
95 CMA_ROUTE_RESOLVED,
96 CMA_CONNECT,
97 CMA_DISCONNECT,
98 CMA_ADDR_BOUND,
99 CMA_LISTEN,
100 CMA_DEVICE_REMOVAL,
101 CMA_DESTROYING
102 };
103
104 struct rdma_bind_list {
105 struct idr *ps;
106 struct hlist_head owners;
107 unsigned short port;
108 };
109
110 /*
111 * Device removal can occur at anytime, so we need extra handling to
112 * serialize notifying the user of device removal with other callbacks.
113 * We do this by disabling removal notification while a callback is in process,
114 * and reporting it after the callback completes.
115 */
116 struct rdma_id_private {
117 struct rdma_cm_id id;
118
119 struct rdma_bind_list *bind_list;
120 struct hlist_node node;
121 struct list_head list; /* listen_any_list or cma_device.list */
122 struct list_head listen_list; /* per device listens */
123 struct cma_device *cma_dev;
124 struct list_head mc_list;
125
126 int internal_id;
127 enum cma_state state;
128 spinlock_t lock;
129 struct mutex qp_mutex;
130
131 struct completion comp;
132 atomic_t refcount;
133 struct mutex handler_mutex;
134
135 int backlog;
136 int timeout_ms;
137 struct ib_sa_query *query;
138 int query_id;
139 union {
140 struct ib_cm_id *ib;
141 struct iw_cm_id *iw;
142 } cm_id;
143
144 u32 seq_num;
145 u32 qkey;
146 u32 qp_num;
147 u8 srq;
148 u8 tos;
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 addr;
159 u8 pad[sizeof(struct sockaddr_in6) -
160 sizeof(struct sockaddr)];
161 };
162
163 struct cma_work {
164 struct work_struct work;
165 struct rdma_id_private *id;
166 enum cma_state old_state;
167 enum cma_state new_state;
168 struct rdma_cm_event event;
169 };
170
171 struct cma_ndev_work {
172 struct work_struct work;
173 struct rdma_id_private *id;
174 struct rdma_cm_event event;
175 };
176
177 union cma_ip_addr {
178 struct in6_addr ip6;
179 struct {
180 __be32 pad[3];
181 __be32 addr;
182 } ip4;
183 };
184
185 struct cma_hdr {
186 u8 cma_version;
187 u8 ip_version; /* IP version: 7:4 */
188 __be16 port;
189 union cma_ip_addr src_addr;
190 union cma_ip_addr dst_addr;
191 };
192
193 struct sdp_hh {
194 u8 bsdh[16];
195 u8 sdp_version; /* Major version: 7:4 */
196 u8 ip_version; /* IP version: 7:4 */
197 u8 sdp_specific1[10];
198 __be16 port;
199 __be16 sdp_specific2;
200 union cma_ip_addr src_addr;
201 union cma_ip_addr dst_addr;
202 };
203
204 struct sdp_hah {
205 u8 bsdh[16];
206 u8 sdp_version;
207 };
208
209 #define CMA_VERSION 0x00
210 #define SDP_MAJ_VERSION 0x2
211
212 static int cma_comp(struct rdma_id_private *id_priv, enum cma_state comp)
213 {
214 unsigned long flags;
215 int ret;
216
217 spin_lock_irqsave(&id_priv->lock, flags);
218 ret = (id_priv->state == comp);
219 spin_unlock_irqrestore(&id_priv->lock, flags);
220 return ret;
221 }
222
223 static int cma_comp_exch(struct rdma_id_private *id_priv,
224 enum cma_state comp, enum cma_state exch)
225 {
226 unsigned long flags;
227 int ret;
228
229 spin_lock_irqsave(&id_priv->lock, flags);
230 if ((ret = (id_priv->state == comp)))
231 id_priv->state = exch;
232 spin_unlock_irqrestore(&id_priv->lock, flags);
233 return ret;
234 }
235
236 static enum cma_state cma_exch(struct rdma_id_private *id_priv,
237 enum cma_state exch)
238 {
239 unsigned long flags;
240 enum cma_state old;
241
242 spin_lock_irqsave(&id_priv->lock, flags);
243 old = id_priv->state;
244 id_priv->state = exch;
245 spin_unlock_irqrestore(&id_priv->lock, flags);
246 return old;
247 }
248
249 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
250 {
251 return hdr->ip_version >> 4;
252 }
253
254 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
255 {
256 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
257 }
258
259 static inline u8 sdp_get_majv(u8 sdp_version)
260 {
261 return sdp_version >> 4;
262 }
263
264 static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
265 {
266 return hh->ip_version >> 4;
267 }
268
269 static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
270 {
271 hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
272 }
273
274 static inline int cma_is_ud_ps(enum rdma_port_space ps)
275 {
276 return (ps == RDMA_PS_UDP || ps == RDMA_PS_IPOIB);
277 }
278
279 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
280 struct cma_device *cma_dev)
281 {
282 atomic_inc(&cma_dev->refcount);
283 id_priv->cma_dev = cma_dev;
284 id_priv->id.device = cma_dev->device;
285 list_add_tail(&id_priv->list, &cma_dev->id_list);
286 }
287
288 static inline void cma_deref_dev(struct cma_device *cma_dev)
289 {
290 if (atomic_dec_and_test(&cma_dev->refcount))
291 complete(&cma_dev->comp);
292 }
293
294 static void cma_detach_from_dev(struct rdma_id_private *id_priv)
295 {
296 list_del(&id_priv->list);
297 cma_deref_dev(id_priv->cma_dev);
298 id_priv->cma_dev = NULL;
299 }
300
301 static int cma_set_qkey(struct ib_device *device, u8 port_num,
302 enum rdma_port_space ps,
303 struct rdma_dev_addr *dev_addr, u32 *qkey)
304 {
305 struct ib_sa_mcmember_rec rec;
306 int ret = 0;
307
308 switch (ps) {
309 case RDMA_PS_UDP:
310 *qkey = RDMA_UDP_QKEY;
311 break;
312 case RDMA_PS_IPOIB:
313 ib_addr_get_mgid(dev_addr, &rec.mgid);
314 ret = ib_sa_get_mcmember_rec(device, port_num, &rec.mgid, &rec);
315 *qkey = be32_to_cpu(rec.qkey);
316 break;
317 default:
318 break;
319 }
320 return ret;
321 }
322
323 static int cma_acquire_dev(struct rdma_id_private *id_priv)
324 {
325 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
326 struct cma_device *cma_dev;
327 union ib_gid gid;
328 int ret = -ENODEV;
329
330 switch (rdma_node_get_transport(dev_addr->dev_type)) {
331 case RDMA_TRANSPORT_IB:
332 ib_addr_get_sgid(dev_addr, &gid);
333 break;
334 case RDMA_TRANSPORT_IWARP:
335 iw_addr_get_sgid(dev_addr, &gid);
336 break;
337 default:
338 return -ENODEV;
339 }
340
341 list_for_each_entry(cma_dev, &dev_list, list) {
342 ret = ib_find_cached_gid(cma_dev->device, &gid,
343 &id_priv->id.port_num, NULL);
344 if (!ret) {
345 ret = cma_set_qkey(cma_dev->device,
346 id_priv->id.port_num,
347 id_priv->id.ps, dev_addr,
348 &id_priv->qkey);
349 if (!ret)
350 cma_attach_to_dev(id_priv, cma_dev);
351 break;
352 }
353 }
354 return ret;
355 }
356
357 static void cma_deref_id(struct rdma_id_private *id_priv)
358 {
359 if (atomic_dec_and_test(&id_priv->refcount))
360 complete(&id_priv->comp);
361 }
362
363 static int cma_disable_callback(struct rdma_id_private *id_priv,
364 enum cma_state state)
365 {
366 mutex_lock(&id_priv->handler_mutex);
367 if (id_priv->state != state) {
368 mutex_unlock(&id_priv->handler_mutex);
369 return -EINVAL;
370 }
371 return 0;
372 }
373
374 static int cma_has_cm_dev(struct rdma_id_private *id_priv)
375 {
376 return (id_priv->id.device && id_priv->cm_id.ib);
377 }
378
379 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
380 void *context, enum rdma_port_space ps)
381 {
382 struct rdma_id_private *id_priv;
383
384 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
385 if (!id_priv)
386 return ERR_PTR(-ENOMEM);
387
388 id_priv->state = CMA_IDLE;
389 id_priv->id.context = context;
390 id_priv->id.event_handler = event_handler;
391 id_priv->id.ps = ps;
392 spin_lock_init(&id_priv->lock);
393 mutex_init(&id_priv->qp_mutex);
394 init_completion(&id_priv->comp);
395 atomic_set(&id_priv->refcount, 1);
396 mutex_init(&id_priv->handler_mutex);
397 INIT_LIST_HEAD(&id_priv->listen_list);
398 INIT_LIST_HEAD(&id_priv->mc_list);
399 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
400
401 return &id_priv->id;
402 }
403 EXPORT_SYMBOL(rdma_create_id);
404
405 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
406 {
407 struct ib_qp_attr qp_attr;
408 int qp_attr_mask, ret;
409
410 qp_attr.qp_state = IB_QPS_INIT;
411 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
412 if (ret)
413 return ret;
414
415 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
416 if (ret)
417 return ret;
418
419 qp_attr.qp_state = IB_QPS_RTR;
420 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
421 if (ret)
422 return ret;
423
424 qp_attr.qp_state = IB_QPS_RTS;
425 qp_attr.sq_psn = 0;
426 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
427
428 return ret;
429 }
430
431 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
432 {
433 struct ib_qp_attr qp_attr;
434 int qp_attr_mask, ret;
435
436 qp_attr.qp_state = IB_QPS_INIT;
437 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
438 if (ret)
439 return ret;
440
441 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
442 }
443
444 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
445 struct ib_qp_init_attr *qp_init_attr)
446 {
447 struct rdma_id_private *id_priv;
448 struct ib_qp *qp;
449 int ret;
450
451 id_priv = container_of(id, struct rdma_id_private, id);
452 if (id->device != pd->device)
453 return -EINVAL;
454
455 qp = ib_create_qp(pd, qp_init_attr);
456 if (IS_ERR(qp))
457 return PTR_ERR(qp);
458
459 if (cma_is_ud_ps(id_priv->id.ps))
460 ret = cma_init_ud_qp(id_priv, qp);
461 else
462 ret = cma_init_conn_qp(id_priv, qp);
463 if (ret)
464 goto err;
465
466 id->qp = qp;
467 id_priv->qp_num = qp->qp_num;
468 id_priv->srq = (qp->srq != NULL);
469 return 0;
470 err:
471 ib_destroy_qp(qp);
472 return ret;
473 }
474 EXPORT_SYMBOL(rdma_create_qp);
475
476 void rdma_destroy_qp(struct rdma_cm_id *id)
477 {
478 struct rdma_id_private *id_priv;
479
480 id_priv = container_of(id, struct rdma_id_private, id);
481 mutex_lock(&id_priv->qp_mutex);
482 ib_destroy_qp(id_priv->id.qp);
483 id_priv->id.qp = NULL;
484 mutex_unlock(&id_priv->qp_mutex);
485 }
486 EXPORT_SYMBOL(rdma_destroy_qp);
487
488 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
489 struct rdma_conn_param *conn_param)
490 {
491 struct ib_qp_attr qp_attr;
492 int qp_attr_mask, ret;
493
494 mutex_lock(&id_priv->qp_mutex);
495 if (!id_priv->id.qp) {
496 ret = 0;
497 goto out;
498 }
499
500 /* Need to update QP attributes from default values. */
501 qp_attr.qp_state = IB_QPS_INIT;
502 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
503 if (ret)
504 goto out;
505
506 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
507 if (ret)
508 goto out;
509
510 qp_attr.qp_state = IB_QPS_RTR;
511 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
512 if (ret)
513 goto out;
514
515 if (conn_param)
516 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
517 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
518 out:
519 mutex_unlock(&id_priv->qp_mutex);
520 return ret;
521 }
522
523 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
524 struct rdma_conn_param *conn_param)
525 {
526 struct ib_qp_attr qp_attr;
527 int qp_attr_mask, ret;
528
529 mutex_lock(&id_priv->qp_mutex);
530 if (!id_priv->id.qp) {
531 ret = 0;
532 goto out;
533 }
534
535 qp_attr.qp_state = IB_QPS_RTS;
536 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
537 if (ret)
538 goto out;
539
540 if (conn_param)
541 qp_attr.max_rd_atomic = conn_param->initiator_depth;
542 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
543 out:
544 mutex_unlock(&id_priv->qp_mutex);
545 return ret;
546 }
547
548 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
549 {
550 struct ib_qp_attr qp_attr;
551 int ret;
552
553 mutex_lock(&id_priv->qp_mutex);
554 if (!id_priv->id.qp) {
555 ret = 0;
556 goto out;
557 }
558
559 qp_attr.qp_state = IB_QPS_ERR;
560 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
561 out:
562 mutex_unlock(&id_priv->qp_mutex);
563 return ret;
564 }
565
566 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
567 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
568 {
569 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
570 int ret;
571
572 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
573 ib_addr_get_pkey(dev_addr),
574 &qp_attr->pkey_index);
575 if (ret)
576 return ret;
577
578 qp_attr->port_num = id_priv->id.port_num;
579 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
580
581 if (cma_is_ud_ps(id_priv->id.ps)) {
582 qp_attr->qkey = id_priv->qkey;
583 *qp_attr_mask |= IB_QP_QKEY;
584 } else {
585 qp_attr->qp_access_flags = 0;
586 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
587 }
588 return 0;
589 }
590
591 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
592 int *qp_attr_mask)
593 {
594 struct rdma_id_private *id_priv;
595 int ret = 0;
596
597 id_priv = container_of(id, struct rdma_id_private, id);
598 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
599 case RDMA_TRANSPORT_IB:
600 if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps))
601 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
602 else
603 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
604 qp_attr_mask);
605 if (qp_attr->qp_state == IB_QPS_RTR)
606 qp_attr->rq_psn = id_priv->seq_num;
607 break;
608 case RDMA_TRANSPORT_IWARP:
609 if (!id_priv->cm_id.iw) {
610 qp_attr->qp_access_flags = 0;
611 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
612 } else
613 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
614 qp_attr_mask);
615 break;
616 default:
617 ret = -ENOSYS;
618 break;
619 }
620
621 return ret;
622 }
623 EXPORT_SYMBOL(rdma_init_qp_attr);
624
625 static inline int cma_zero_addr(struct sockaddr *addr)
626 {
627 struct in6_addr *ip6;
628
629 if (addr->sa_family == AF_INET)
630 return ipv4_is_zeronet(
631 ((struct sockaddr_in *)addr)->sin_addr.s_addr);
632 else {
633 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
634 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
635 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
636 }
637 }
638
639 static inline int cma_loopback_addr(struct sockaddr *addr)
640 {
641 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
642 }
643
644 static inline int cma_any_addr(struct sockaddr *addr)
645 {
646 return cma_zero_addr(addr) || cma_loopback_addr(addr);
647 }
648
649 static inline __be16 cma_port(struct sockaddr *addr)
650 {
651 if (addr->sa_family == AF_INET)
652 return ((struct sockaddr_in *) addr)->sin_port;
653 else
654 return ((struct sockaddr_in6 *) addr)->sin6_port;
655 }
656
657 static inline int cma_any_port(struct sockaddr *addr)
658 {
659 return !cma_port(addr);
660 }
661
662 static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
663 u8 *ip_ver, __be16 *port,
664 union cma_ip_addr **src, union cma_ip_addr **dst)
665 {
666 switch (ps) {
667 case RDMA_PS_SDP:
668 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
669 SDP_MAJ_VERSION)
670 return -EINVAL;
671
672 *ip_ver = sdp_get_ip_ver(hdr);
673 *port = ((struct sdp_hh *) hdr)->port;
674 *src = &((struct sdp_hh *) hdr)->src_addr;
675 *dst = &((struct sdp_hh *) hdr)->dst_addr;
676 break;
677 default:
678 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
679 return -EINVAL;
680
681 *ip_ver = cma_get_ip_ver(hdr);
682 *port = ((struct cma_hdr *) hdr)->port;
683 *src = &((struct cma_hdr *) hdr)->src_addr;
684 *dst = &((struct cma_hdr *) hdr)->dst_addr;
685 break;
686 }
687
688 if (*ip_ver != 4 && *ip_ver != 6)
689 return -EINVAL;
690 return 0;
691 }
692
693 static void cma_save_net_info(struct rdma_addr *addr,
694 struct rdma_addr *listen_addr,
695 u8 ip_ver, __be16 port,
696 union cma_ip_addr *src, union cma_ip_addr *dst)
697 {
698 struct sockaddr_in *listen4, *ip4;
699 struct sockaddr_in6 *listen6, *ip6;
700
701 switch (ip_ver) {
702 case 4:
703 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
704 ip4 = (struct sockaddr_in *) &addr->src_addr;
705 ip4->sin_family = listen4->sin_family;
706 ip4->sin_addr.s_addr = dst->ip4.addr;
707 ip4->sin_port = listen4->sin_port;
708
709 ip4 = (struct sockaddr_in *) &addr->dst_addr;
710 ip4->sin_family = listen4->sin_family;
711 ip4->sin_addr.s_addr = src->ip4.addr;
712 ip4->sin_port = port;
713 break;
714 case 6:
715 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
716 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
717 ip6->sin6_family = listen6->sin6_family;
718 ip6->sin6_addr = dst->ip6;
719 ip6->sin6_port = listen6->sin6_port;
720
721 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
722 ip6->sin6_family = listen6->sin6_family;
723 ip6->sin6_addr = src->ip6;
724 ip6->sin6_port = port;
725 break;
726 default:
727 break;
728 }
729 }
730
731 static inline int cma_user_data_offset(enum rdma_port_space ps)
732 {
733 switch (ps) {
734 case RDMA_PS_SDP:
735 return 0;
736 default:
737 return sizeof(struct cma_hdr);
738 }
739 }
740
741 static void cma_cancel_route(struct rdma_id_private *id_priv)
742 {
743 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
744 case RDMA_TRANSPORT_IB:
745 if (id_priv->query)
746 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
747 break;
748 default:
749 break;
750 }
751 }
752
753 static void cma_cancel_listens(struct rdma_id_private *id_priv)
754 {
755 struct rdma_id_private *dev_id_priv;
756
757 /*
758 * Remove from listen_any_list to prevent added devices from spawning
759 * additional listen requests.
760 */
761 mutex_lock(&lock);
762 list_del(&id_priv->list);
763
764 while (!list_empty(&id_priv->listen_list)) {
765 dev_id_priv = list_entry(id_priv->listen_list.next,
766 struct rdma_id_private, listen_list);
767 /* sync with device removal to avoid duplicate destruction */
768 list_del_init(&dev_id_priv->list);
769 list_del(&dev_id_priv->listen_list);
770 mutex_unlock(&lock);
771
772 rdma_destroy_id(&dev_id_priv->id);
773 mutex_lock(&lock);
774 }
775 mutex_unlock(&lock);
776 }
777
778 static void cma_cancel_operation(struct rdma_id_private *id_priv,
779 enum cma_state state)
780 {
781 switch (state) {
782 case CMA_ADDR_QUERY:
783 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
784 break;
785 case CMA_ROUTE_QUERY:
786 cma_cancel_route(id_priv);
787 break;
788 case CMA_LISTEN:
789 if (cma_any_addr(&id_priv->id.route.addr.src_addr) &&
790 !id_priv->cma_dev)
791 cma_cancel_listens(id_priv);
792 break;
793 default:
794 break;
795 }
796 }
797
798 static void cma_release_port(struct rdma_id_private *id_priv)
799 {
800 struct rdma_bind_list *bind_list = id_priv->bind_list;
801
802 if (!bind_list)
803 return;
804
805 mutex_lock(&lock);
806 hlist_del(&id_priv->node);
807 if (hlist_empty(&bind_list->owners)) {
808 idr_remove(bind_list->ps, bind_list->port);
809 kfree(bind_list);
810 }
811 mutex_unlock(&lock);
812 }
813
814 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
815 {
816 struct cma_multicast *mc;
817
818 while (!list_empty(&id_priv->mc_list)) {
819 mc = container_of(id_priv->mc_list.next,
820 struct cma_multicast, list);
821 list_del(&mc->list);
822 ib_sa_free_multicast(mc->multicast.ib);
823 kfree(mc);
824 }
825 }
826
827 void rdma_destroy_id(struct rdma_cm_id *id)
828 {
829 struct rdma_id_private *id_priv;
830 enum cma_state state;
831
832 id_priv = container_of(id, struct rdma_id_private, id);
833 state = cma_exch(id_priv, CMA_DESTROYING);
834 cma_cancel_operation(id_priv, state);
835
836 mutex_lock(&lock);
837 if (id_priv->cma_dev) {
838 mutex_unlock(&lock);
839 switch (rdma_node_get_transport(id->device->node_type)) {
840 case RDMA_TRANSPORT_IB:
841 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib))
842 ib_destroy_cm_id(id_priv->cm_id.ib);
843 break;
844 case RDMA_TRANSPORT_IWARP:
845 if (id_priv->cm_id.iw && !IS_ERR(id_priv->cm_id.iw))
846 iw_destroy_cm_id(id_priv->cm_id.iw);
847 break;
848 default:
849 break;
850 }
851 cma_leave_mc_groups(id_priv);
852 mutex_lock(&lock);
853 cma_detach_from_dev(id_priv);
854 }
855 mutex_unlock(&lock);
856
857 cma_release_port(id_priv);
858 cma_deref_id(id_priv);
859 wait_for_completion(&id_priv->comp);
860
861 if (id_priv->internal_id)
862 cma_deref_id(id_priv->id.context);
863
864 kfree(id_priv->id.route.path_rec);
865 kfree(id_priv);
866 }
867 EXPORT_SYMBOL(rdma_destroy_id);
868
869 static int cma_rep_recv(struct rdma_id_private *id_priv)
870 {
871 int ret;
872
873 ret = cma_modify_qp_rtr(id_priv, NULL);
874 if (ret)
875 goto reject;
876
877 ret = cma_modify_qp_rts(id_priv, NULL);
878 if (ret)
879 goto reject;
880
881 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
882 if (ret)
883 goto reject;
884
885 return 0;
886 reject:
887 cma_modify_qp_err(id_priv);
888 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
889 NULL, 0, NULL, 0);
890 return ret;
891 }
892
893 static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
894 {
895 if (id_priv->id.ps == RDMA_PS_SDP &&
896 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
897 SDP_MAJ_VERSION)
898 return -EINVAL;
899
900 return 0;
901 }
902
903 static void cma_set_rep_event_data(struct rdma_cm_event *event,
904 struct ib_cm_rep_event_param *rep_data,
905 void *private_data)
906 {
907 event->param.conn.private_data = private_data;
908 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
909 event->param.conn.responder_resources = rep_data->responder_resources;
910 event->param.conn.initiator_depth = rep_data->initiator_depth;
911 event->param.conn.flow_control = rep_data->flow_control;
912 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
913 event->param.conn.srq = rep_data->srq;
914 event->param.conn.qp_num = rep_data->remote_qpn;
915 }
916
917 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
918 {
919 struct rdma_id_private *id_priv = cm_id->context;
920 struct rdma_cm_event event;
921 int ret = 0;
922
923 if (cma_disable_callback(id_priv, CMA_CONNECT))
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 case IB_CM_MRA_RECEIVED:
960 /* ignore event */
961 goto out;
962 case IB_CM_REJ_RECEIVED:
963 cma_modify_qp_err(id_priv);
964 event.status = ib_event->param.rej_rcvd.reason;
965 event.event = RDMA_CM_EVENT_REJECTED;
966 event.param.conn.private_data = ib_event->private_data;
967 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
968 break;
969 default:
970 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
971 ib_event->event);
972 goto out;
973 }
974
975 ret = id_priv->id.event_handler(&id_priv->id, &event);
976 if (ret) {
977 /* Destroy the CM ID by returning a non-zero value. */
978 id_priv->cm_id.ib = NULL;
979 cma_exch(id_priv, CMA_DESTROYING);
980 mutex_unlock(&id_priv->handler_mutex);
981 rdma_destroy_id(&id_priv->id);
982 return ret;
983 }
984 out:
985 mutex_unlock(&id_priv->handler_mutex);
986 return ret;
987 }
988
989 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
990 struct ib_cm_event *ib_event)
991 {
992 struct rdma_id_private *id_priv;
993 struct rdma_cm_id *id;
994 struct rdma_route *rt;
995 union cma_ip_addr *src, *dst;
996 __be16 port;
997 u8 ip_ver;
998 int ret;
999
1000 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1001 &ip_ver, &port, &src, &dst))
1002 goto err;
1003
1004 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1005 listen_id->ps);
1006 if (IS_ERR(id))
1007 goto err;
1008
1009 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1010 ip_ver, port, src, dst);
1011
1012 rt = &id->route;
1013 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1014 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1015 GFP_KERNEL);
1016 if (!rt->path_rec)
1017 goto destroy_id;
1018
1019 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1020 if (rt->num_paths == 2)
1021 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1022
1023 ib_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1024 ret = rdma_translate_ip(&id->route.addr.src_addr,
1025 &id->route.addr.dev_addr);
1026 if (ret)
1027 goto destroy_id;
1028
1029 id_priv = container_of(id, struct rdma_id_private, id);
1030 id_priv->state = CMA_CONNECT;
1031 return id_priv;
1032
1033 destroy_id:
1034 rdma_destroy_id(id);
1035 err:
1036 return NULL;
1037 }
1038
1039 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1040 struct ib_cm_event *ib_event)
1041 {
1042 struct rdma_id_private *id_priv;
1043 struct rdma_cm_id *id;
1044 union cma_ip_addr *src, *dst;
1045 __be16 port;
1046 u8 ip_ver;
1047 int ret;
1048
1049 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1050 listen_id->ps);
1051 if (IS_ERR(id))
1052 return NULL;
1053
1054
1055 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1056 &ip_ver, &port, &src, &dst))
1057 goto err;
1058
1059 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1060 ip_ver, port, src, dst);
1061
1062 ret = rdma_translate_ip(&id->route.addr.src_addr,
1063 &id->route.addr.dev_addr);
1064 if (ret)
1065 goto err;
1066
1067 id_priv = container_of(id, struct rdma_id_private, id);
1068 id_priv->state = CMA_CONNECT;
1069 return id_priv;
1070 err:
1071 rdma_destroy_id(id);
1072 return NULL;
1073 }
1074
1075 static void cma_set_req_event_data(struct rdma_cm_event *event,
1076 struct ib_cm_req_event_param *req_data,
1077 void *private_data, int offset)
1078 {
1079 event->param.conn.private_data = private_data + offset;
1080 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1081 event->param.conn.responder_resources = req_data->responder_resources;
1082 event->param.conn.initiator_depth = req_data->initiator_depth;
1083 event->param.conn.flow_control = req_data->flow_control;
1084 event->param.conn.retry_count = req_data->retry_count;
1085 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1086 event->param.conn.srq = req_data->srq;
1087 event->param.conn.qp_num = req_data->remote_qpn;
1088 }
1089
1090 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1091 {
1092 struct rdma_id_private *listen_id, *conn_id;
1093 struct rdma_cm_event event;
1094 int offset, ret;
1095
1096 listen_id = cm_id->context;
1097 if (cma_disable_callback(listen_id, CMA_LISTEN))
1098 return -ECONNABORTED;
1099
1100 memset(&event, 0, sizeof event);
1101 offset = cma_user_data_offset(listen_id->id.ps);
1102 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1103 if (cma_is_ud_ps(listen_id->id.ps)) {
1104 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1105 event.param.ud.private_data = ib_event->private_data + offset;
1106 event.param.ud.private_data_len =
1107 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1108 } else {
1109 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1110 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1111 ib_event->private_data, offset);
1112 }
1113 if (!conn_id) {
1114 ret = -ENOMEM;
1115 goto out;
1116 }
1117
1118 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1119 mutex_lock(&lock);
1120 ret = cma_acquire_dev(conn_id);
1121 mutex_unlock(&lock);
1122 if (ret)
1123 goto release_conn_id;
1124
1125 conn_id->cm_id.ib = cm_id;
1126 cm_id->context = conn_id;
1127 cm_id->cm_handler = cma_ib_handler;
1128
1129 ret = conn_id->id.event_handler(&conn_id->id, &event);
1130 if (!ret) {
1131 /*
1132 * Acquire mutex to prevent user executing rdma_destroy_id()
1133 * while we're accessing the cm_id.
1134 */
1135 mutex_lock(&lock);
1136 if (cma_comp(conn_id, CMA_CONNECT) &&
1137 !cma_is_ud_ps(conn_id->id.ps))
1138 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1139 mutex_unlock(&lock);
1140 mutex_unlock(&conn_id->handler_mutex);
1141 goto out;
1142 }
1143
1144 /* Destroy the CM ID by returning a non-zero value. */
1145 conn_id->cm_id.ib = NULL;
1146
1147 release_conn_id:
1148 cma_exch(conn_id, CMA_DESTROYING);
1149 mutex_unlock(&conn_id->handler_mutex);
1150 rdma_destroy_id(&conn_id->id);
1151
1152 out:
1153 mutex_unlock(&listen_id->handler_mutex);
1154 return ret;
1155 }
1156
1157 static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1158 {
1159 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
1160 }
1161
1162 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1163 struct ib_cm_compare_data *compare)
1164 {
1165 struct cma_hdr *cma_data, *cma_mask;
1166 struct sdp_hh *sdp_data, *sdp_mask;
1167 __be32 ip4_addr;
1168 struct in6_addr ip6_addr;
1169
1170 memset(compare, 0, sizeof *compare);
1171 cma_data = (void *) compare->data;
1172 cma_mask = (void *) compare->mask;
1173 sdp_data = (void *) compare->data;
1174 sdp_mask = (void *) compare->mask;
1175
1176 switch (addr->sa_family) {
1177 case AF_INET:
1178 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1179 if (ps == RDMA_PS_SDP) {
1180 sdp_set_ip_ver(sdp_data, 4);
1181 sdp_set_ip_ver(sdp_mask, 0xF);
1182 sdp_data->dst_addr.ip4.addr = ip4_addr;
1183 sdp_mask->dst_addr.ip4.addr = htonl(~0);
1184 } else {
1185 cma_set_ip_ver(cma_data, 4);
1186 cma_set_ip_ver(cma_mask, 0xF);
1187 cma_data->dst_addr.ip4.addr = ip4_addr;
1188 cma_mask->dst_addr.ip4.addr = htonl(~0);
1189 }
1190 break;
1191 case AF_INET6:
1192 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1193 if (ps == RDMA_PS_SDP) {
1194 sdp_set_ip_ver(sdp_data, 6);
1195 sdp_set_ip_ver(sdp_mask, 0xF);
1196 sdp_data->dst_addr.ip6 = ip6_addr;
1197 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1198 sizeof sdp_mask->dst_addr.ip6);
1199 } else {
1200 cma_set_ip_ver(cma_data, 6);
1201 cma_set_ip_ver(cma_mask, 0xF);
1202 cma_data->dst_addr.ip6 = ip6_addr;
1203 memset(&cma_mask->dst_addr.ip6, 0xFF,
1204 sizeof cma_mask->dst_addr.ip6);
1205 }
1206 break;
1207 default:
1208 break;
1209 }
1210 }
1211
1212 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1213 {
1214 struct rdma_id_private *id_priv = iw_id->context;
1215 struct rdma_cm_event event;
1216 struct sockaddr_in *sin;
1217 int ret = 0;
1218
1219 if (cma_disable_callback(id_priv, CMA_CONNECT))
1220 return 0;
1221
1222 memset(&event, 0, sizeof event);
1223 switch (iw_event->event) {
1224 case IW_CM_EVENT_CLOSE:
1225 event.event = RDMA_CM_EVENT_DISCONNECTED;
1226 break;
1227 case IW_CM_EVENT_CONNECT_REPLY:
1228 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1229 *sin = iw_event->local_addr;
1230 sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
1231 *sin = iw_event->remote_addr;
1232 switch (iw_event->status) {
1233 case 0:
1234 event.event = RDMA_CM_EVENT_ESTABLISHED;
1235 break;
1236 case -ECONNRESET:
1237 case -ECONNREFUSED:
1238 event.event = RDMA_CM_EVENT_REJECTED;
1239 break;
1240 case -ETIMEDOUT:
1241 event.event = RDMA_CM_EVENT_UNREACHABLE;
1242 break;
1243 default:
1244 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1245 break;
1246 }
1247 break;
1248 case IW_CM_EVENT_ESTABLISHED:
1249 event.event = RDMA_CM_EVENT_ESTABLISHED;
1250 break;
1251 default:
1252 BUG_ON(1);
1253 }
1254
1255 event.status = iw_event->status;
1256 event.param.conn.private_data = iw_event->private_data;
1257 event.param.conn.private_data_len = iw_event->private_data_len;
1258 ret = id_priv->id.event_handler(&id_priv->id, &event);
1259 if (ret) {
1260 /* Destroy the CM ID by returning a non-zero value. */
1261 id_priv->cm_id.iw = NULL;
1262 cma_exch(id_priv, CMA_DESTROYING);
1263 mutex_unlock(&id_priv->handler_mutex);
1264 rdma_destroy_id(&id_priv->id);
1265 return ret;
1266 }
1267
1268 mutex_unlock(&id_priv->handler_mutex);
1269 return ret;
1270 }
1271
1272 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1273 struct iw_cm_event *iw_event)
1274 {
1275 struct rdma_cm_id *new_cm_id;
1276 struct rdma_id_private *listen_id, *conn_id;
1277 struct sockaddr_in *sin;
1278 struct net_device *dev = NULL;
1279 struct rdma_cm_event event;
1280 int ret;
1281 struct ib_device_attr attr;
1282
1283 listen_id = cm_id->context;
1284 if (cma_disable_callback(listen_id, CMA_LISTEN))
1285 return -ECONNABORTED;
1286
1287 /* Create a new RDMA id for the new IW CM ID */
1288 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1289 listen_id->id.context,
1290 RDMA_PS_TCP);
1291 if (IS_ERR(new_cm_id)) {
1292 ret = -ENOMEM;
1293 goto out;
1294 }
1295 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1296 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1297 conn_id->state = CMA_CONNECT;
1298
1299 dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
1300 if (!dev) {
1301 ret = -EADDRNOTAVAIL;
1302 mutex_unlock(&conn_id->handler_mutex);
1303 rdma_destroy_id(new_cm_id);
1304 goto out;
1305 }
1306 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1307 if (ret) {
1308 mutex_unlock(&conn_id->handler_mutex);
1309 rdma_destroy_id(new_cm_id);
1310 goto out;
1311 }
1312
1313 mutex_lock(&lock);
1314 ret = cma_acquire_dev(conn_id);
1315 mutex_unlock(&lock);
1316 if (ret) {
1317 mutex_unlock(&conn_id->handler_mutex);
1318 rdma_destroy_id(new_cm_id);
1319 goto out;
1320 }
1321
1322 conn_id->cm_id.iw = cm_id;
1323 cm_id->context = conn_id;
1324 cm_id->cm_handler = cma_iw_handler;
1325
1326 sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
1327 *sin = iw_event->local_addr;
1328 sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
1329 *sin = iw_event->remote_addr;
1330
1331 ret = ib_query_device(conn_id->id.device, &attr);
1332 if (ret) {
1333 mutex_unlock(&conn_id->handler_mutex);
1334 rdma_destroy_id(new_cm_id);
1335 goto out;
1336 }
1337
1338 memset(&event, 0, sizeof event);
1339 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1340 event.param.conn.private_data = iw_event->private_data;
1341 event.param.conn.private_data_len = iw_event->private_data_len;
1342 event.param.conn.initiator_depth = attr.max_qp_init_rd_atom;
1343 event.param.conn.responder_resources = attr.max_qp_rd_atom;
1344 ret = conn_id->id.event_handler(&conn_id->id, &event);
1345 if (ret) {
1346 /* User wants to destroy the CM ID */
1347 conn_id->cm_id.iw = NULL;
1348 cma_exch(conn_id, CMA_DESTROYING);
1349 mutex_unlock(&conn_id->handler_mutex);
1350 rdma_destroy_id(&conn_id->id);
1351 goto out;
1352 }
1353
1354 mutex_unlock(&conn_id->handler_mutex);
1355
1356 out:
1357 if (dev)
1358 dev_put(dev);
1359 mutex_unlock(&listen_id->handler_mutex);
1360 return ret;
1361 }
1362
1363 static int cma_ib_listen(struct rdma_id_private *id_priv)
1364 {
1365 struct ib_cm_compare_data compare_data;
1366 struct sockaddr *addr;
1367 __be64 svc_id;
1368 int ret;
1369
1370 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_req_handler,
1371 id_priv);
1372 if (IS_ERR(id_priv->cm_id.ib))
1373 return PTR_ERR(id_priv->cm_id.ib);
1374
1375 addr = &id_priv->id.route.addr.src_addr;
1376 svc_id = cma_get_service_id(id_priv->id.ps, addr);
1377 if (cma_any_addr(addr))
1378 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1379 else {
1380 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1381 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1382 }
1383
1384 if (ret) {
1385 ib_destroy_cm_id(id_priv->cm_id.ib);
1386 id_priv->cm_id.ib = NULL;
1387 }
1388
1389 return ret;
1390 }
1391
1392 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1393 {
1394 int ret;
1395 struct sockaddr_in *sin;
1396
1397 id_priv->cm_id.iw = iw_create_cm_id(id_priv->id.device,
1398 iw_conn_req_handler,
1399 id_priv);
1400 if (IS_ERR(id_priv->cm_id.iw))
1401 return PTR_ERR(id_priv->cm_id.iw);
1402
1403 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1404 id_priv->cm_id.iw->local_addr = *sin;
1405
1406 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1407
1408 if (ret) {
1409 iw_destroy_cm_id(id_priv->cm_id.iw);
1410 id_priv->cm_id.iw = NULL;
1411 }
1412
1413 return ret;
1414 }
1415
1416 static int cma_listen_handler(struct rdma_cm_id *id,
1417 struct rdma_cm_event *event)
1418 {
1419 struct rdma_id_private *id_priv = id->context;
1420
1421 id->context = id_priv->id.context;
1422 id->event_handler = id_priv->id.event_handler;
1423 return id_priv->id.event_handler(id, event);
1424 }
1425
1426 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1427 struct cma_device *cma_dev)
1428 {
1429 struct rdma_id_private *dev_id_priv;
1430 struct rdma_cm_id *id;
1431 int ret;
1432
1433 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps);
1434 if (IS_ERR(id))
1435 return;
1436
1437 dev_id_priv = container_of(id, struct rdma_id_private, id);
1438
1439 dev_id_priv->state = CMA_ADDR_BOUND;
1440 memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
1441 ip_addr_size(&id_priv->id.route.addr.src_addr));
1442
1443 cma_attach_to_dev(dev_id_priv, cma_dev);
1444 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1445 atomic_inc(&id_priv->refcount);
1446 dev_id_priv->internal_id = 1;
1447
1448 ret = rdma_listen(id, id_priv->backlog);
1449 if (ret)
1450 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1451 "listening on device %s\n", ret, cma_dev->device->name);
1452 }
1453
1454 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1455 {
1456 struct cma_device *cma_dev;
1457
1458 mutex_lock(&lock);
1459 list_add_tail(&id_priv->list, &listen_any_list);
1460 list_for_each_entry(cma_dev, &dev_list, list)
1461 cma_listen_on_dev(id_priv, cma_dev);
1462 mutex_unlock(&lock);
1463 }
1464
1465 static int cma_bind_any(struct rdma_cm_id *id, sa_family_t af)
1466 {
1467 struct sockaddr_in addr_in;
1468
1469 memset(&addr_in, 0, sizeof addr_in);
1470 addr_in.sin_family = af;
1471 return rdma_bind_addr(id, (struct sockaddr *) &addr_in);
1472 }
1473
1474 int rdma_listen(struct rdma_cm_id *id, int backlog)
1475 {
1476 struct rdma_id_private *id_priv;
1477 int ret;
1478
1479 id_priv = container_of(id, struct rdma_id_private, id);
1480 if (id_priv->state == CMA_IDLE) {
1481 ret = cma_bind_any(id, AF_INET);
1482 if (ret)
1483 return ret;
1484 }
1485
1486 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
1487 return -EINVAL;
1488
1489 id_priv->backlog = backlog;
1490 if (id->device) {
1491 switch (rdma_node_get_transport(id->device->node_type)) {
1492 case RDMA_TRANSPORT_IB:
1493 ret = cma_ib_listen(id_priv);
1494 if (ret)
1495 goto err;
1496 break;
1497 case RDMA_TRANSPORT_IWARP:
1498 ret = cma_iw_listen(id_priv, backlog);
1499 if (ret)
1500 goto err;
1501 break;
1502 default:
1503 ret = -ENOSYS;
1504 goto err;
1505 }
1506 } else
1507 cma_listen_on_all(id_priv);
1508
1509 return 0;
1510 err:
1511 id_priv->backlog = 0;
1512 cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
1513 return ret;
1514 }
1515 EXPORT_SYMBOL(rdma_listen);
1516
1517 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1518 {
1519 struct rdma_id_private *id_priv;
1520
1521 id_priv = container_of(id, struct rdma_id_private, id);
1522 id_priv->tos = (u8) tos;
1523 }
1524 EXPORT_SYMBOL(rdma_set_service_type);
1525
1526 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1527 void *context)
1528 {
1529 struct cma_work *work = context;
1530 struct rdma_route *route;
1531
1532 route = &work->id->id.route;
1533
1534 if (!status) {
1535 route->num_paths = 1;
1536 *route->path_rec = *path_rec;
1537 } else {
1538 work->old_state = CMA_ROUTE_QUERY;
1539 work->new_state = CMA_ADDR_RESOLVED;
1540 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1541 work->event.status = status;
1542 }
1543
1544 queue_work(cma_wq, &work->work);
1545 }
1546
1547 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1548 struct cma_work *work)
1549 {
1550 struct rdma_addr *addr = &id_priv->id.route.addr;
1551 struct ib_sa_path_rec path_rec;
1552 ib_sa_comp_mask comp_mask;
1553 struct sockaddr_in6 *sin6;
1554
1555 memset(&path_rec, 0, sizeof path_rec);
1556 ib_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
1557 ib_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
1558 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
1559 path_rec.numb_path = 1;
1560 path_rec.reversible = 1;
1561 path_rec.service_id = cma_get_service_id(id_priv->id.ps, &addr->dst_addr);
1562
1563 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1564 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1565 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1566
1567 if (addr->src_addr.sa_family == AF_INET) {
1568 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1569 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1570 } else {
1571 sin6 = (struct sockaddr_in6 *) &addr->src_addr;
1572 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1573 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1574 }
1575
1576 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1577 id_priv->id.port_num, &path_rec,
1578 comp_mask, timeout_ms,
1579 GFP_KERNEL, cma_query_handler,
1580 work, &id_priv->query);
1581
1582 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1583 }
1584
1585 static void cma_work_handler(struct work_struct *_work)
1586 {
1587 struct cma_work *work = container_of(_work, struct cma_work, work);
1588 struct rdma_id_private *id_priv = work->id;
1589 int destroy = 0;
1590
1591 mutex_lock(&id_priv->handler_mutex);
1592 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1593 goto out;
1594
1595 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1596 cma_exch(id_priv, CMA_DESTROYING);
1597 destroy = 1;
1598 }
1599 out:
1600 mutex_unlock(&id_priv->handler_mutex);
1601 cma_deref_id(id_priv);
1602 if (destroy)
1603 rdma_destroy_id(&id_priv->id);
1604 kfree(work);
1605 }
1606
1607 static void cma_ndev_work_handler(struct work_struct *_work)
1608 {
1609 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1610 struct rdma_id_private *id_priv = work->id;
1611 int destroy = 0;
1612
1613 mutex_lock(&id_priv->handler_mutex);
1614 if (id_priv->state == CMA_DESTROYING ||
1615 id_priv->state == CMA_DEVICE_REMOVAL)
1616 goto out;
1617
1618 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1619 cma_exch(id_priv, CMA_DESTROYING);
1620 destroy = 1;
1621 }
1622
1623 out:
1624 mutex_unlock(&id_priv->handler_mutex);
1625 cma_deref_id(id_priv);
1626 if (destroy)
1627 rdma_destroy_id(&id_priv->id);
1628 kfree(work);
1629 }
1630
1631 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1632 {
1633 struct rdma_route *route = &id_priv->id.route;
1634 struct cma_work *work;
1635 int ret;
1636
1637 work = kzalloc(sizeof *work, GFP_KERNEL);
1638 if (!work)
1639 return -ENOMEM;
1640
1641 work->id = id_priv;
1642 INIT_WORK(&work->work, cma_work_handler);
1643 work->old_state = CMA_ROUTE_QUERY;
1644 work->new_state = CMA_ROUTE_RESOLVED;
1645 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1646
1647 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1648 if (!route->path_rec) {
1649 ret = -ENOMEM;
1650 goto err1;
1651 }
1652
1653 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1654 if (ret)
1655 goto err2;
1656
1657 return 0;
1658 err2:
1659 kfree(route->path_rec);
1660 route->path_rec = NULL;
1661 err1:
1662 kfree(work);
1663 return ret;
1664 }
1665
1666 int rdma_set_ib_paths(struct rdma_cm_id *id,
1667 struct ib_sa_path_rec *path_rec, int num_paths)
1668 {
1669 struct rdma_id_private *id_priv;
1670 int ret;
1671
1672 id_priv = container_of(id, struct rdma_id_private, id);
1673 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_RESOLVED))
1674 return -EINVAL;
1675
1676 id->route.path_rec = kmalloc(sizeof *path_rec * num_paths, GFP_KERNEL);
1677 if (!id->route.path_rec) {
1678 ret = -ENOMEM;
1679 goto err;
1680 }
1681
1682 memcpy(id->route.path_rec, path_rec, sizeof *path_rec * num_paths);
1683 return 0;
1684 err:
1685 cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_ADDR_RESOLVED);
1686 return ret;
1687 }
1688 EXPORT_SYMBOL(rdma_set_ib_paths);
1689
1690 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1691 {
1692 struct cma_work *work;
1693
1694 work = kzalloc(sizeof *work, GFP_KERNEL);
1695 if (!work)
1696 return -ENOMEM;
1697
1698 work->id = id_priv;
1699 INIT_WORK(&work->work, cma_work_handler);
1700 work->old_state = CMA_ROUTE_QUERY;
1701 work->new_state = CMA_ROUTE_RESOLVED;
1702 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1703 queue_work(cma_wq, &work->work);
1704 return 0;
1705 }
1706
1707 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1708 {
1709 struct rdma_id_private *id_priv;
1710 int ret;
1711
1712 id_priv = container_of(id, struct rdma_id_private, id);
1713 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_QUERY))
1714 return -EINVAL;
1715
1716 atomic_inc(&id_priv->refcount);
1717 switch (rdma_node_get_transport(id->device->node_type)) {
1718 case RDMA_TRANSPORT_IB:
1719 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1720 break;
1721 case RDMA_TRANSPORT_IWARP:
1722 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1723 break;
1724 default:
1725 ret = -ENOSYS;
1726 break;
1727 }
1728 if (ret)
1729 goto err;
1730
1731 return 0;
1732 err:
1733 cma_comp_exch(id_priv, CMA_ROUTE_QUERY, CMA_ADDR_RESOLVED);
1734 cma_deref_id(id_priv);
1735 return ret;
1736 }
1737 EXPORT_SYMBOL(rdma_resolve_route);
1738
1739 static int cma_bind_loopback(struct rdma_id_private *id_priv)
1740 {
1741 struct cma_device *cma_dev;
1742 struct ib_port_attr port_attr;
1743 union ib_gid gid;
1744 u16 pkey;
1745 int ret;
1746 u8 p;
1747
1748 mutex_lock(&lock);
1749 if (list_empty(&dev_list)) {
1750 ret = -ENODEV;
1751 goto out;
1752 }
1753 list_for_each_entry(cma_dev, &dev_list, list)
1754 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
1755 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
1756 port_attr.state == IB_PORT_ACTIVE)
1757 goto port_found;
1758
1759 p = 1;
1760 cma_dev = list_entry(dev_list.next, struct cma_device, list);
1761
1762 port_found:
1763 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
1764 if (ret)
1765 goto out;
1766
1767 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
1768 if (ret)
1769 goto out;
1770
1771 ib_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1772 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1773 id_priv->id.port_num = p;
1774 cma_attach_to_dev(id_priv, cma_dev);
1775 out:
1776 mutex_unlock(&lock);
1777 return ret;
1778 }
1779
1780 static void addr_handler(int status, struct sockaddr *src_addr,
1781 struct rdma_dev_addr *dev_addr, void *context)
1782 {
1783 struct rdma_id_private *id_priv = context;
1784 struct rdma_cm_event event;
1785
1786 memset(&event, 0, sizeof event);
1787 mutex_lock(&id_priv->handler_mutex);
1788
1789 /*
1790 * Grab mutex to block rdma_destroy_id() from removing the device while
1791 * we're trying to acquire it.
1792 */
1793 mutex_lock(&lock);
1794 if (!cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_RESOLVED)) {
1795 mutex_unlock(&lock);
1796 goto out;
1797 }
1798
1799 if (!status && !id_priv->cma_dev)
1800 status = cma_acquire_dev(id_priv);
1801 mutex_unlock(&lock);
1802
1803 if (status) {
1804 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ADDR_BOUND))
1805 goto out;
1806 event.event = RDMA_CM_EVENT_ADDR_ERROR;
1807 event.status = status;
1808 } else {
1809 memcpy(&id_priv->id.route.addr.src_addr, src_addr,
1810 ip_addr_size(src_addr));
1811 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1812 }
1813
1814 if (id_priv->id.event_handler(&id_priv->id, &event)) {
1815 cma_exch(id_priv, CMA_DESTROYING);
1816 mutex_unlock(&id_priv->handler_mutex);
1817 cma_deref_id(id_priv);
1818 rdma_destroy_id(&id_priv->id);
1819 return;
1820 }
1821 out:
1822 mutex_unlock(&id_priv->handler_mutex);
1823 cma_deref_id(id_priv);
1824 }
1825
1826 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1827 {
1828 struct cma_work *work;
1829 struct sockaddr_in *src_in, *dst_in;
1830 union ib_gid gid;
1831 int ret;
1832
1833 work = kzalloc(sizeof *work, GFP_KERNEL);
1834 if (!work)
1835 return -ENOMEM;
1836
1837 if (!id_priv->cma_dev) {
1838 ret = cma_bind_loopback(id_priv);
1839 if (ret)
1840 goto err;
1841 }
1842
1843 ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1844 ib_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
1845
1846 if (cma_zero_addr(&id_priv->id.route.addr.src_addr)) {
1847 src_in = (struct sockaddr_in *)&id_priv->id.route.addr.src_addr;
1848 dst_in = (struct sockaddr_in *)&id_priv->id.route.addr.dst_addr;
1849 src_in->sin_family = dst_in->sin_family;
1850 src_in->sin_addr.s_addr = dst_in->sin_addr.s_addr;
1851 }
1852
1853 work->id = id_priv;
1854 INIT_WORK(&work->work, cma_work_handler);
1855 work->old_state = CMA_ADDR_QUERY;
1856 work->new_state = CMA_ADDR_RESOLVED;
1857 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1858 queue_work(cma_wq, &work->work);
1859 return 0;
1860 err:
1861 kfree(work);
1862 return ret;
1863 }
1864
1865 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1866 struct sockaddr *dst_addr)
1867 {
1868 if (src_addr && src_addr->sa_family)
1869 return rdma_bind_addr(id, src_addr);
1870 else
1871 return cma_bind_any(id, dst_addr->sa_family);
1872 }
1873
1874 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1875 struct sockaddr *dst_addr, int timeout_ms)
1876 {
1877 struct rdma_id_private *id_priv;
1878 int ret;
1879
1880 id_priv = container_of(id, struct rdma_id_private, id);
1881 if (id_priv->state == CMA_IDLE) {
1882 ret = cma_bind_addr(id, src_addr, dst_addr);
1883 if (ret)
1884 return ret;
1885 }
1886
1887 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_ADDR_QUERY))
1888 return -EINVAL;
1889
1890 atomic_inc(&id_priv->refcount);
1891 memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
1892 if (cma_any_addr(dst_addr))
1893 ret = cma_resolve_loopback(id_priv);
1894 else
1895 ret = rdma_resolve_ip(&addr_client, &id->route.addr.src_addr,
1896 dst_addr, &id->route.addr.dev_addr,
1897 timeout_ms, addr_handler, id_priv);
1898 if (ret)
1899 goto err;
1900
1901 return 0;
1902 err:
1903 cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_BOUND);
1904 cma_deref_id(id_priv);
1905 return ret;
1906 }
1907 EXPORT_SYMBOL(rdma_resolve_addr);
1908
1909 static void cma_bind_port(struct rdma_bind_list *bind_list,
1910 struct rdma_id_private *id_priv)
1911 {
1912 struct sockaddr_in *sin;
1913
1914 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1915 sin->sin_port = htons(bind_list->port);
1916 id_priv->bind_list = bind_list;
1917 hlist_add_head(&id_priv->node, &bind_list->owners);
1918 }
1919
1920 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
1921 unsigned short snum)
1922 {
1923 struct rdma_bind_list *bind_list;
1924 int port, ret;
1925
1926 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1927 if (!bind_list)
1928 return -ENOMEM;
1929
1930 do {
1931 ret = idr_get_new_above(ps, bind_list, snum, &port);
1932 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1933
1934 if (ret)
1935 goto err1;
1936
1937 if (port != snum) {
1938 ret = -EADDRNOTAVAIL;
1939 goto err2;
1940 }
1941
1942 bind_list->ps = ps;
1943 bind_list->port = (unsigned short) port;
1944 cma_bind_port(bind_list, id_priv);
1945 return 0;
1946 err2:
1947 idr_remove(ps, port);
1948 err1:
1949 kfree(bind_list);
1950 return ret;
1951 }
1952
1953 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
1954 {
1955 struct rdma_bind_list *bind_list;
1956 int port, ret, low, high;
1957
1958 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1959 if (!bind_list)
1960 return -ENOMEM;
1961
1962 retry:
1963 /* FIXME: add proper port randomization per like inet_csk_get_port */
1964 do {
1965 ret = idr_get_new_above(ps, bind_list, next_port, &port);
1966 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1967
1968 if (ret)
1969 goto err1;
1970
1971 inet_get_local_port_range(&low, &high);
1972 if (port > high) {
1973 if (next_port != low) {
1974 idr_remove(ps, port);
1975 next_port = low;
1976 goto retry;
1977 }
1978 ret = -EADDRNOTAVAIL;
1979 goto err2;
1980 }
1981
1982 if (port == high)
1983 next_port = low;
1984 else
1985 next_port = port + 1;
1986
1987 bind_list->ps = ps;
1988 bind_list->port = (unsigned short) port;
1989 cma_bind_port(bind_list, id_priv);
1990 return 0;
1991 err2:
1992 idr_remove(ps, port);
1993 err1:
1994 kfree(bind_list);
1995 return ret;
1996 }
1997
1998 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
1999 {
2000 struct rdma_id_private *cur_id;
2001 struct sockaddr_in *sin, *cur_sin;
2002 struct rdma_bind_list *bind_list;
2003 struct hlist_node *node;
2004 unsigned short snum;
2005
2006 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
2007 snum = ntohs(sin->sin_port);
2008 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2009 return -EACCES;
2010
2011 bind_list = idr_find(ps, snum);
2012 if (!bind_list)
2013 return cma_alloc_port(ps, id_priv, snum);
2014
2015 /*
2016 * We don't support binding to any address if anyone is bound to
2017 * a specific address on the same port.
2018 */
2019 if (cma_any_addr(&id_priv->id.route.addr.src_addr))
2020 return -EADDRNOTAVAIL;
2021
2022 hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
2023 if (cma_any_addr(&cur_id->id.route.addr.src_addr))
2024 return -EADDRNOTAVAIL;
2025
2026 cur_sin = (struct sockaddr_in *) &cur_id->id.route.addr.src_addr;
2027 if (sin->sin_addr.s_addr == cur_sin->sin_addr.s_addr)
2028 return -EADDRINUSE;
2029 }
2030
2031 cma_bind_port(bind_list, id_priv);
2032 return 0;
2033 }
2034
2035 static int cma_get_port(struct rdma_id_private *id_priv)
2036 {
2037 struct idr *ps;
2038 int ret;
2039
2040 switch (id_priv->id.ps) {
2041 case RDMA_PS_SDP:
2042 ps = &sdp_ps;
2043 break;
2044 case RDMA_PS_TCP:
2045 ps = &tcp_ps;
2046 break;
2047 case RDMA_PS_UDP:
2048 ps = &udp_ps;
2049 break;
2050 case RDMA_PS_IPOIB:
2051 ps = &ipoib_ps;
2052 break;
2053 default:
2054 return -EPROTONOSUPPORT;
2055 }
2056
2057 mutex_lock(&lock);
2058 if (cma_any_port(&id_priv->id.route.addr.src_addr))
2059 ret = cma_alloc_any_port(ps, id_priv);
2060 else
2061 ret = cma_use_port(ps, id_priv);
2062 mutex_unlock(&lock);
2063
2064 return ret;
2065 }
2066
2067 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2068 {
2069 struct rdma_id_private *id_priv;
2070 int ret;
2071
2072 if (addr->sa_family != AF_INET)
2073 return -EAFNOSUPPORT;
2074
2075 id_priv = container_of(id, struct rdma_id_private, id);
2076 if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND))
2077 return -EINVAL;
2078
2079 if (!cma_any_addr(addr)) {
2080 ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
2081 if (ret)
2082 goto err1;
2083
2084 mutex_lock(&lock);
2085 ret = cma_acquire_dev(id_priv);
2086 mutex_unlock(&lock);
2087 if (ret)
2088 goto err1;
2089 }
2090
2091 memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
2092 ret = cma_get_port(id_priv);
2093 if (ret)
2094 goto err2;
2095
2096 return 0;
2097 err2:
2098 if (!cma_any_addr(addr)) {
2099 mutex_lock(&lock);
2100 cma_detach_from_dev(id_priv);
2101 mutex_unlock(&lock);
2102 }
2103 err1:
2104 cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_IDLE);
2105 return ret;
2106 }
2107 EXPORT_SYMBOL(rdma_bind_addr);
2108
2109 static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
2110 struct rdma_route *route)
2111 {
2112 struct sockaddr_in *src4, *dst4;
2113 struct cma_hdr *cma_hdr;
2114 struct sdp_hh *sdp_hdr;
2115
2116 src4 = (struct sockaddr_in *) &route->addr.src_addr;
2117 dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
2118
2119 switch (ps) {
2120 case RDMA_PS_SDP:
2121 sdp_hdr = hdr;
2122 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2123 return -EINVAL;
2124 sdp_set_ip_ver(sdp_hdr, 4);
2125 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2126 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2127 sdp_hdr->port = src4->sin_port;
2128 break;
2129 default:
2130 cma_hdr = hdr;
2131 cma_hdr->cma_version = CMA_VERSION;
2132 cma_set_ip_ver(cma_hdr, 4);
2133 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2134 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2135 cma_hdr->port = src4->sin_port;
2136 break;
2137 }
2138 return 0;
2139 }
2140
2141 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2142 struct ib_cm_event *ib_event)
2143 {
2144 struct rdma_id_private *id_priv = cm_id->context;
2145 struct rdma_cm_event event;
2146 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2147 int ret = 0;
2148
2149 if (cma_disable_callback(id_priv, CMA_CONNECT))
2150 return 0;
2151
2152 memset(&event, 0, sizeof event);
2153 switch (ib_event->event) {
2154 case IB_CM_SIDR_REQ_ERROR:
2155 event.event = RDMA_CM_EVENT_UNREACHABLE;
2156 event.status = -ETIMEDOUT;
2157 break;
2158 case IB_CM_SIDR_REP_RECEIVED:
2159 event.param.ud.private_data = ib_event->private_data;
2160 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2161 if (rep->status != IB_SIDR_SUCCESS) {
2162 event.event = RDMA_CM_EVENT_UNREACHABLE;
2163 event.status = ib_event->param.sidr_rep_rcvd.status;
2164 break;
2165 }
2166 if (id_priv->qkey != rep->qkey) {
2167 event.event = RDMA_CM_EVENT_UNREACHABLE;
2168 event.status = -EINVAL;
2169 break;
2170 }
2171 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2172 id_priv->id.route.path_rec,
2173 &event.param.ud.ah_attr);
2174 event.param.ud.qp_num = rep->qpn;
2175 event.param.ud.qkey = rep->qkey;
2176 event.event = RDMA_CM_EVENT_ESTABLISHED;
2177 event.status = 0;
2178 break;
2179 default:
2180 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2181 ib_event->event);
2182 goto out;
2183 }
2184
2185 ret = id_priv->id.event_handler(&id_priv->id, &event);
2186 if (ret) {
2187 /* Destroy the CM ID by returning a non-zero value. */
2188 id_priv->cm_id.ib = NULL;
2189 cma_exch(id_priv, CMA_DESTROYING);
2190 mutex_unlock(&id_priv->handler_mutex);
2191 rdma_destroy_id(&id_priv->id);
2192 return ret;
2193 }
2194 out:
2195 mutex_unlock(&id_priv->handler_mutex);
2196 return ret;
2197 }
2198
2199 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2200 struct rdma_conn_param *conn_param)
2201 {
2202 struct ib_cm_sidr_req_param req;
2203 struct rdma_route *route;
2204 int ret;
2205
2206 req.private_data_len = sizeof(struct cma_hdr) +
2207 conn_param->private_data_len;
2208 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2209 if (!req.private_data)
2210 return -ENOMEM;
2211
2212 if (conn_param->private_data && conn_param->private_data_len)
2213 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2214 conn_param->private_data, conn_param->private_data_len);
2215
2216 route = &id_priv->id.route;
2217 ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
2218 if (ret)
2219 goto out;
2220
2221 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device,
2222 cma_sidr_rep_handler, id_priv);
2223 if (IS_ERR(id_priv->cm_id.ib)) {
2224 ret = PTR_ERR(id_priv->cm_id.ib);
2225 goto out;
2226 }
2227
2228 req.path = route->path_rec;
2229 req.service_id = cma_get_service_id(id_priv->id.ps,
2230 &route->addr.dst_addr);
2231 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2232 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2233
2234 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2235 if (ret) {
2236 ib_destroy_cm_id(id_priv->cm_id.ib);
2237 id_priv->cm_id.ib = NULL;
2238 }
2239 out:
2240 kfree(req.private_data);
2241 return ret;
2242 }
2243
2244 static int cma_connect_ib(struct rdma_id_private *id_priv,
2245 struct rdma_conn_param *conn_param)
2246 {
2247 struct ib_cm_req_param req;
2248 struct rdma_route *route;
2249 void *private_data;
2250 int offset, ret;
2251
2252 memset(&req, 0, sizeof req);
2253 offset = cma_user_data_offset(id_priv->id.ps);
2254 req.private_data_len = offset + conn_param->private_data_len;
2255 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2256 if (!private_data)
2257 return -ENOMEM;
2258
2259 if (conn_param->private_data && conn_param->private_data_len)
2260 memcpy(private_data + offset, conn_param->private_data,
2261 conn_param->private_data_len);
2262
2263 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_ib_handler,
2264 id_priv);
2265 if (IS_ERR(id_priv->cm_id.ib)) {
2266 ret = PTR_ERR(id_priv->cm_id.ib);
2267 goto out;
2268 }
2269
2270 route = &id_priv->id.route;
2271 ret = cma_format_hdr(private_data, id_priv->id.ps, route);
2272 if (ret)
2273 goto out;
2274 req.private_data = private_data;
2275
2276 req.primary_path = &route->path_rec[0];
2277 if (route->num_paths == 2)
2278 req.alternate_path = &route->path_rec[1];
2279
2280 req.service_id = cma_get_service_id(id_priv->id.ps,
2281 &route->addr.dst_addr);
2282 req.qp_num = id_priv->qp_num;
2283 req.qp_type = IB_QPT_RC;
2284 req.starting_psn = id_priv->seq_num;
2285 req.responder_resources = conn_param->responder_resources;
2286 req.initiator_depth = conn_param->initiator_depth;
2287 req.flow_control = conn_param->flow_control;
2288 req.retry_count = conn_param->retry_count;
2289 req.rnr_retry_count = conn_param->rnr_retry_count;
2290 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2291 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2292 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2293 req.srq = id_priv->srq ? 1 : 0;
2294
2295 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2296 out:
2297 if (ret && !IS_ERR(id_priv->cm_id.ib)) {
2298 ib_destroy_cm_id(id_priv->cm_id.ib);
2299 id_priv->cm_id.ib = NULL;
2300 }
2301
2302 kfree(private_data);
2303 return ret;
2304 }
2305
2306 static int cma_connect_iw(struct rdma_id_private *id_priv,
2307 struct rdma_conn_param *conn_param)
2308 {
2309 struct iw_cm_id *cm_id;
2310 struct sockaddr_in* sin;
2311 int ret;
2312 struct iw_cm_conn_param iw_param;
2313
2314 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2315 if (IS_ERR(cm_id)) {
2316 ret = PTR_ERR(cm_id);
2317 goto out;
2318 }
2319
2320 id_priv->cm_id.iw = cm_id;
2321
2322 sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
2323 cm_id->local_addr = *sin;
2324
2325 sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
2326 cm_id->remote_addr = *sin;
2327
2328 ret = cma_modify_qp_rtr(id_priv, conn_param);
2329 if (ret)
2330 goto out;
2331
2332 iw_param.ord = conn_param->initiator_depth;
2333 iw_param.ird = conn_param->responder_resources;
2334 iw_param.private_data = conn_param->private_data;
2335 iw_param.private_data_len = conn_param->private_data_len;
2336 if (id_priv->id.qp)
2337 iw_param.qpn = id_priv->qp_num;
2338 else
2339 iw_param.qpn = conn_param->qp_num;
2340 ret = iw_cm_connect(cm_id, &iw_param);
2341 out:
2342 if (ret && !IS_ERR(cm_id)) {
2343 iw_destroy_cm_id(cm_id);
2344 id_priv->cm_id.iw = NULL;
2345 }
2346 return ret;
2347 }
2348
2349 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2350 {
2351 struct rdma_id_private *id_priv;
2352 int ret;
2353
2354 id_priv = container_of(id, struct rdma_id_private, id);
2355 if (!cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_CONNECT))
2356 return -EINVAL;
2357
2358 if (!id->qp) {
2359 id_priv->qp_num = conn_param->qp_num;
2360 id_priv->srq = conn_param->srq;
2361 }
2362
2363 switch (rdma_node_get_transport(id->device->node_type)) {
2364 case RDMA_TRANSPORT_IB:
2365 if (cma_is_ud_ps(id->ps))
2366 ret = cma_resolve_ib_udp(id_priv, conn_param);
2367 else
2368 ret = cma_connect_ib(id_priv, conn_param);
2369 break;
2370 case RDMA_TRANSPORT_IWARP:
2371 ret = cma_connect_iw(id_priv, conn_param);
2372 break;
2373 default:
2374 ret = -ENOSYS;
2375 break;
2376 }
2377 if (ret)
2378 goto err;
2379
2380 return 0;
2381 err:
2382 cma_comp_exch(id_priv, CMA_CONNECT, CMA_ROUTE_RESOLVED);
2383 return ret;
2384 }
2385 EXPORT_SYMBOL(rdma_connect);
2386
2387 static int cma_accept_ib(struct rdma_id_private *id_priv,
2388 struct rdma_conn_param *conn_param)
2389 {
2390 struct ib_cm_rep_param rep;
2391 int ret;
2392
2393 ret = cma_modify_qp_rtr(id_priv, conn_param);
2394 if (ret)
2395 goto out;
2396
2397 ret = cma_modify_qp_rts(id_priv, conn_param);
2398 if (ret)
2399 goto out;
2400
2401 memset(&rep, 0, sizeof rep);
2402 rep.qp_num = id_priv->qp_num;
2403 rep.starting_psn = id_priv->seq_num;
2404 rep.private_data = conn_param->private_data;
2405 rep.private_data_len = conn_param->private_data_len;
2406 rep.responder_resources = conn_param->responder_resources;
2407 rep.initiator_depth = conn_param->initiator_depth;
2408 rep.failover_accepted = 0;
2409 rep.flow_control = conn_param->flow_control;
2410 rep.rnr_retry_count = conn_param->rnr_retry_count;
2411 rep.srq = id_priv->srq ? 1 : 0;
2412
2413 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2414 out:
2415 return ret;
2416 }
2417
2418 static int cma_accept_iw(struct rdma_id_private *id_priv,
2419 struct rdma_conn_param *conn_param)
2420 {
2421 struct iw_cm_conn_param iw_param;
2422 int ret;
2423
2424 ret = cma_modify_qp_rtr(id_priv, conn_param);
2425 if (ret)
2426 return ret;
2427
2428 iw_param.ord = conn_param->initiator_depth;
2429 iw_param.ird = conn_param->responder_resources;
2430 iw_param.private_data = conn_param->private_data;
2431 iw_param.private_data_len = conn_param->private_data_len;
2432 if (id_priv->id.qp) {
2433 iw_param.qpn = id_priv->qp_num;
2434 } else
2435 iw_param.qpn = conn_param->qp_num;
2436
2437 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2438 }
2439
2440 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2441 enum ib_cm_sidr_status status,
2442 const void *private_data, int private_data_len)
2443 {
2444 struct ib_cm_sidr_rep_param rep;
2445
2446 memset(&rep, 0, sizeof rep);
2447 rep.status = status;
2448 if (status == IB_SIDR_SUCCESS) {
2449 rep.qp_num = id_priv->qp_num;
2450 rep.qkey = id_priv->qkey;
2451 }
2452 rep.private_data = private_data;
2453 rep.private_data_len = private_data_len;
2454
2455 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2456 }
2457
2458 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2459 {
2460 struct rdma_id_private *id_priv;
2461 int ret;
2462
2463 id_priv = container_of(id, struct rdma_id_private, id);
2464 if (!cma_comp(id_priv, CMA_CONNECT))
2465 return -EINVAL;
2466
2467 if (!id->qp && conn_param) {
2468 id_priv->qp_num = conn_param->qp_num;
2469 id_priv->srq = conn_param->srq;
2470 }
2471
2472 switch (rdma_node_get_transport(id->device->node_type)) {
2473 case RDMA_TRANSPORT_IB:
2474 if (cma_is_ud_ps(id->ps))
2475 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2476 conn_param->private_data,
2477 conn_param->private_data_len);
2478 else if (conn_param)
2479 ret = cma_accept_ib(id_priv, conn_param);
2480 else
2481 ret = cma_rep_recv(id_priv);
2482 break;
2483 case RDMA_TRANSPORT_IWARP:
2484 ret = cma_accept_iw(id_priv, conn_param);
2485 break;
2486 default:
2487 ret = -ENOSYS;
2488 break;
2489 }
2490
2491 if (ret)
2492 goto reject;
2493
2494 return 0;
2495 reject:
2496 cma_modify_qp_err(id_priv);
2497 rdma_reject(id, NULL, 0);
2498 return ret;
2499 }
2500 EXPORT_SYMBOL(rdma_accept);
2501
2502 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2503 {
2504 struct rdma_id_private *id_priv;
2505 int ret;
2506
2507 id_priv = container_of(id, struct rdma_id_private, id);
2508 if (!cma_has_cm_dev(id_priv))
2509 return -EINVAL;
2510
2511 switch (id->device->node_type) {
2512 case RDMA_NODE_IB_CA:
2513 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2514 break;
2515 default:
2516 ret = 0;
2517 break;
2518 }
2519 return ret;
2520 }
2521 EXPORT_SYMBOL(rdma_notify);
2522
2523 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2524 u8 private_data_len)
2525 {
2526 struct rdma_id_private *id_priv;
2527 int ret;
2528
2529 id_priv = container_of(id, struct rdma_id_private, id);
2530 if (!cma_has_cm_dev(id_priv))
2531 return -EINVAL;
2532
2533 switch (rdma_node_get_transport(id->device->node_type)) {
2534 case RDMA_TRANSPORT_IB:
2535 if (cma_is_ud_ps(id->ps))
2536 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
2537 private_data, private_data_len);
2538 else
2539 ret = ib_send_cm_rej(id_priv->cm_id.ib,
2540 IB_CM_REJ_CONSUMER_DEFINED, NULL,
2541 0, private_data, private_data_len);
2542 break;
2543 case RDMA_TRANSPORT_IWARP:
2544 ret = iw_cm_reject(id_priv->cm_id.iw,
2545 private_data, private_data_len);
2546 break;
2547 default:
2548 ret = -ENOSYS;
2549 break;
2550 }
2551 return ret;
2552 }
2553 EXPORT_SYMBOL(rdma_reject);
2554
2555 int rdma_disconnect(struct rdma_cm_id *id)
2556 {
2557 struct rdma_id_private *id_priv;
2558 int ret;
2559
2560 id_priv = container_of(id, struct rdma_id_private, id);
2561 if (!cma_has_cm_dev(id_priv))
2562 return -EINVAL;
2563
2564 switch (rdma_node_get_transport(id->device->node_type)) {
2565 case RDMA_TRANSPORT_IB:
2566 ret = cma_modify_qp_err(id_priv);
2567 if (ret)
2568 goto out;
2569 /* Initiate or respond to a disconnect. */
2570 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
2571 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
2572 break;
2573 case RDMA_TRANSPORT_IWARP:
2574 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
2575 break;
2576 default:
2577 ret = -EINVAL;
2578 break;
2579 }
2580 out:
2581 return ret;
2582 }
2583 EXPORT_SYMBOL(rdma_disconnect);
2584
2585 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
2586 {
2587 struct rdma_id_private *id_priv;
2588 struct cma_multicast *mc = multicast->context;
2589 struct rdma_cm_event event;
2590 int ret;
2591
2592 id_priv = mc->id_priv;
2593 if (cma_disable_callback(id_priv, CMA_ADDR_BOUND) &&
2594 cma_disable_callback(id_priv, CMA_ADDR_RESOLVED))
2595 return 0;
2596
2597 mutex_lock(&id_priv->qp_mutex);
2598 if (!status && id_priv->id.qp)
2599 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
2600 multicast->rec.mlid);
2601 mutex_unlock(&id_priv->qp_mutex);
2602
2603 memset(&event, 0, sizeof event);
2604 event.status = status;
2605 event.param.ud.private_data = mc->context;
2606 if (!status) {
2607 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
2608 ib_init_ah_from_mcmember(id_priv->id.device,
2609 id_priv->id.port_num, &multicast->rec,
2610 &event.param.ud.ah_attr);
2611 event.param.ud.qp_num = 0xFFFFFF;
2612 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
2613 } else
2614 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
2615
2616 ret = id_priv->id.event_handler(&id_priv->id, &event);
2617 if (ret) {
2618 cma_exch(id_priv, CMA_DESTROYING);
2619 mutex_unlock(&id_priv->handler_mutex);
2620 rdma_destroy_id(&id_priv->id);
2621 return 0;
2622 }
2623
2624 mutex_unlock(&id_priv->handler_mutex);
2625 return 0;
2626 }
2627
2628 static void cma_set_mgid(struct rdma_id_private *id_priv,
2629 struct sockaddr *addr, union ib_gid *mgid)
2630 {
2631 unsigned char mc_map[MAX_ADDR_LEN];
2632 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2633 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
2634 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
2635
2636 if (cma_any_addr(addr)) {
2637 memset(mgid, 0, sizeof *mgid);
2638 } else if ((addr->sa_family == AF_INET6) &&
2639 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFF10A01B) ==
2640 0xFF10A01B)) {
2641 /* IPv6 address is an SA assigned MGID. */
2642 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
2643 } else {
2644 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
2645 if (id_priv->id.ps == RDMA_PS_UDP)
2646 mc_map[7] = 0x01; /* Use RDMA CM signature */
2647 *mgid = *(union ib_gid *) (mc_map + 4);
2648 }
2649 }
2650
2651 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
2652 struct cma_multicast *mc)
2653 {
2654 struct ib_sa_mcmember_rec rec;
2655 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2656 ib_sa_comp_mask comp_mask;
2657 int ret;
2658
2659 ib_addr_get_mgid(dev_addr, &rec.mgid);
2660 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
2661 &rec.mgid, &rec);
2662 if (ret)
2663 return ret;
2664
2665 cma_set_mgid(id_priv, &mc->addr, &rec.mgid);
2666 if (id_priv->id.ps == RDMA_PS_UDP)
2667 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
2668 ib_addr_get_sgid(dev_addr, &rec.port_gid);
2669 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2670 rec.join_state = 1;
2671
2672 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
2673 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
2674 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
2675 IB_SA_MCMEMBER_REC_FLOW_LABEL |
2676 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
2677
2678 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
2679 id_priv->id.port_num, &rec,
2680 comp_mask, GFP_KERNEL,
2681 cma_ib_mc_handler, mc);
2682 if (IS_ERR(mc->multicast.ib))
2683 return PTR_ERR(mc->multicast.ib);
2684
2685 return 0;
2686 }
2687
2688 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
2689 void *context)
2690 {
2691 struct rdma_id_private *id_priv;
2692 struct cma_multicast *mc;
2693 int ret;
2694
2695 id_priv = container_of(id, struct rdma_id_private, id);
2696 if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
2697 !cma_comp(id_priv, CMA_ADDR_RESOLVED))
2698 return -EINVAL;
2699
2700 mc = kmalloc(sizeof *mc, GFP_KERNEL);
2701 if (!mc)
2702 return -ENOMEM;
2703
2704 memcpy(&mc->addr, addr, ip_addr_size(addr));
2705 mc->context = context;
2706 mc->id_priv = id_priv;
2707
2708 spin_lock(&id_priv->lock);
2709 list_add(&mc->list, &id_priv->mc_list);
2710 spin_unlock(&id_priv->lock);
2711
2712 switch (rdma_node_get_transport(id->device->node_type)) {
2713 case RDMA_TRANSPORT_IB:
2714 ret = cma_join_ib_multicast(id_priv, mc);
2715 break;
2716 default:
2717 ret = -ENOSYS;
2718 break;
2719 }
2720
2721 if (ret) {
2722 spin_lock_irq(&id_priv->lock);
2723 list_del(&mc->list);
2724 spin_unlock_irq(&id_priv->lock);
2725 kfree(mc);
2726 }
2727 return ret;
2728 }
2729 EXPORT_SYMBOL(rdma_join_multicast);
2730
2731 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
2732 {
2733 struct rdma_id_private *id_priv;
2734 struct cma_multicast *mc;
2735
2736 id_priv = container_of(id, struct rdma_id_private, id);
2737 spin_lock_irq(&id_priv->lock);
2738 list_for_each_entry(mc, &id_priv->mc_list, list) {
2739 if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
2740 list_del(&mc->list);
2741 spin_unlock_irq(&id_priv->lock);
2742
2743 if (id->qp)
2744 ib_detach_mcast(id->qp,
2745 &mc->multicast.ib->rec.mgid,
2746 mc->multicast.ib->rec.mlid);
2747 ib_sa_free_multicast(mc->multicast.ib);
2748 kfree(mc);
2749 return;
2750 }
2751 }
2752 spin_unlock_irq(&id_priv->lock);
2753 }
2754 EXPORT_SYMBOL(rdma_leave_multicast);
2755
2756 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
2757 {
2758 struct rdma_dev_addr *dev_addr;
2759 struct cma_ndev_work *work;
2760
2761 dev_addr = &id_priv->id.route.addr.dev_addr;
2762
2763 if ((dev_addr->src_dev == ndev) &&
2764 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
2765 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
2766 ndev->name, &id_priv->id);
2767 work = kzalloc(sizeof *work, GFP_KERNEL);
2768 if (!work)
2769 return -ENOMEM;
2770
2771 INIT_WORK(&work->work, cma_ndev_work_handler);
2772 work->id = id_priv;
2773 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
2774 atomic_inc(&id_priv->refcount);
2775 queue_work(cma_wq, &work->work);
2776 }
2777
2778 return 0;
2779 }
2780
2781 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
2782 void *ctx)
2783 {
2784 struct net_device *ndev = (struct net_device *)ctx;
2785 struct cma_device *cma_dev;
2786 struct rdma_id_private *id_priv;
2787 int ret = NOTIFY_DONE;
2788
2789 if (dev_net(ndev) != &init_net)
2790 return NOTIFY_DONE;
2791
2792 if (event != NETDEV_BONDING_FAILOVER)
2793 return NOTIFY_DONE;
2794
2795 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
2796 return NOTIFY_DONE;
2797
2798 mutex_lock(&lock);
2799 list_for_each_entry(cma_dev, &dev_list, list)
2800 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
2801 ret = cma_netdev_change(ndev, id_priv);
2802 if (ret)
2803 goto out;
2804 }
2805
2806 out:
2807 mutex_unlock(&lock);
2808 return ret;
2809 }
2810
2811 static struct notifier_block cma_nb = {
2812 .notifier_call = cma_netdev_callback
2813 };
2814
2815 static void cma_add_one(struct ib_device *device)
2816 {
2817 struct cma_device *cma_dev;
2818 struct rdma_id_private *id_priv;
2819
2820 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
2821 if (!cma_dev)
2822 return;
2823
2824 cma_dev->device = device;
2825
2826 init_completion(&cma_dev->comp);
2827 atomic_set(&cma_dev->refcount, 1);
2828 INIT_LIST_HEAD(&cma_dev->id_list);
2829 ib_set_client_data(device, &cma_client, cma_dev);
2830
2831 mutex_lock(&lock);
2832 list_add_tail(&cma_dev->list, &dev_list);
2833 list_for_each_entry(id_priv, &listen_any_list, list)
2834 cma_listen_on_dev(id_priv, cma_dev);
2835 mutex_unlock(&lock);
2836 }
2837
2838 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
2839 {
2840 struct rdma_cm_event event;
2841 enum cma_state state;
2842 int ret = 0;
2843
2844 /* Record that we want to remove the device */
2845 state = cma_exch(id_priv, CMA_DEVICE_REMOVAL);
2846 if (state == CMA_DESTROYING)
2847 return 0;
2848
2849 cma_cancel_operation(id_priv, state);
2850 mutex_lock(&id_priv->handler_mutex);
2851
2852 /* Check for destruction from another callback. */
2853 if (!cma_comp(id_priv, CMA_DEVICE_REMOVAL))
2854 goto out;
2855
2856 memset(&event, 0, sizeof event);
2857 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
2858 ret = id_priv->id.event_handler(&id_priv->id, &event);
2859 out:
2860 mutex_unlock(&id_priv->handler_mutex);
2861 return ret;
2862 }
2863
2864 static void cma_process_remove(struct cma_device *cma_dev)
2865 {
2866 struct rdma_id_private *id_priv;
2867 int ret;
2868
2869 mutex_lock(&lock);
2870 while (!list_empty(&cma_dev->id_list)) {
2871 id_priv = list_entry(cma_dev->id_list.next,
2872 struct rdma_id_private, list);
2873
2874 list_del(&id_priv->listen_list);
2875 list_del_init(&id_priv->list);
2876 atomic_inc(&id_priv->refcount);
2877 mutex_unlock(&lock);
2878
2879 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
2880 cma_deref_id(id_priv);
2881 if (ret)
2882 rdma_destroy_id(&id_priv->id);
2883
2884 mutex_lock(&lock);
2885 }
2886 mutex_unlock(&lock);
2887
2888 cma_deref_dev(cma_dev);
2889 wait_for_completion(&cma_dev->comp);
2890 }
2891
2892 static void cma_remove_one(struct ib_device *device)
2893 {
2894 struct cma_device *cma_dev;
2895
2896 cma_dev = ib_get_client_data(device, &cma_client);
2897 if (!cma_dev)
2898 return;
2899
2900 mutex_lock(&lock);
2901 list_del(&cma_dev->list);
2902 mutex_unlock(&lock);
2903
2904 cma_process_remove(cma_dev);
2905 kfree(cma_dev);
2906 }
2907
2908 static int cma_init(void)
2909 {
2910 int ret, low, high, remaining;
2911
2912 get_random_bytes(&next_port, sizeof next_port);
2913 inet_get_local_port_range(&low, &high);
2914 remaining = (high - low) + 1;
2915 next_port = ((unsigned int) next_port % remaining) + low;
2916
2917 cma_wq = create_singlethread_workqueue("rdma_cm");
2918 if (!cma_wq)
2919 return -ENOMEM;
2920
2921 ib_sa_register_client(&sa_client);
2922 rdma_addr_register_client(&addr_client);
2923 register_netdevice_notifier(&cma_nb);
2924
2925 ret = ib_register_client(&cma_client);
2926 if (ret)
2927 goto err;
2928 return 0;
2929
2930 err:
2931 unregister_netdevice_notifier(&cma_nb);
2932 rdma_addr_unregister_client(&addr_client);
2933 ib_sa_unregister_client(&sa_client);
2934 destroy_workqueue(cma_wq);
2935 return ret;
2936 }
2937
2938 static void cma_cleanup(void)
2939 {
2940 ib_unregister_client(&cma_client);
2941 unregister_netdevice_notifier(&cma_nb);
2942 rdma_addr_unregister_client(&addr_client);
2943 ib_sa_unregister_client(&sa_client);
2944 destroy_workqueue(cma_wq);
2945 idr_destroy(&sdp_ps);
2946 idr_destroy(&tcp_ps);
2947 idr_destroy(&udp_ps);
2948 idr_destroy(&ipoib_ps);
2949 }
2950
2951 module_init(cma_init);
2952 module_exit(cma_cleanup);
Linux kernel - Deliverables
This page took 0.109245 seconds and 6 git commands to generate.