]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/infiniband/core/cma.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
IPoIB: improve IPv4/IPv6 to IB mcast mapping functions
[mirror_ubuntu-zesty-kernel.git] / drivers / infiniband / core / cma.c
1 /*
2 * Copyright (c) 2005 Voltaire Inc. All rights reserved.
3 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
4 * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
5 * Copyright (c) 2005-2006 Intel Corporation. All rights reserved.
6 *
7 * This Software is licensed under one of the following licenses:
8 *
9 * 1) under the terms of the "Common Public License 1.0" a copy of which is
10 * available from the Open Source Initiative, see
11 * http://www.opensource.org/licenses/cpl.php.
12 *
13 * 2) under the terms of the "The BSD License" a copy of which is
14 * available from the Open Source Initiative, see
15 * http://www.opensource.org/licenses/bsd-license.php.
16 *
17 * 3) under the terms of the "GNU General Public License (GPL) Version 2" a
18 * copy of which is available from the Open Source Initiative, see
19 * http://www.opensource.org/licenses/gpl-license.php.
20 *
21 * Licensee has the right to choose one of the above licenses.
22 *
23 * Redistributions of source code must retain the above copyright
24 * notice and one of the license notices.
25 *
26 * Redistributions in binary form must reproduce both the above copyright
27 * notice, one of the license notices in the documentation
28 * and/or other materials provided with the distribution.
29 *
30 */
31
32 #include <linux/completion.h>
33 #include <linux/in.h>
34 #include <linux/in6.h>
35 #include <linux/mutex.h>
36 #include <linux/random.h>
37 #include <linux/idr.h>
38 #include <linux/inetdevice.h>
39
40 #include <net/tcp.h>
41
42 #include <rdma/rdma_cm.h>
43 #include <rdma/rdma_cm_ib.h>
44 #include <rdma/ib_cache.h>
45 #include <rdma/ib_cm.h>
46 #include <rdma/ib_sa.h>
47 #include <rdma/iw_cm.h>
48
49 MODULE_AUTHOR("Sean Hefty");
50 MODULE_DESCRIPTION("Generic RDMA CM Agent");
51 MODULE_LICENSE("Dual BSD/GPL");
52
53 #define CMA_CM_RESPONSE_TIMEOUT 20
54 #define CMA_MAX_CM_RETRIES 15
55 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
56
57 static void cma_add_one(struct ib_device *device);
58 static void cma_remove_one(struct ib_device *device);
59
60 static struct ib_client cma_client = {
61 .name = "cma",
62 .add = cma_add_one,
63 .remove = cma_remove_one
64 };
65
66 static struct ib_sa_client sa_client;
67 static struct rdma_addr_client addr_client;
68 static LIST_HEAD(dev_list);
69 static LIST_HEAD(listen_any_list);
70 static DEFINE_MUTEX(lock);
71 static struct workqueue_struct *cma_wq;
72 static DEFINE_IDR(sdp_ps);
73 static DEFINE_IDR(tcp_ps);
74 static DEFINE_IDR(udp_ps);
75 static DEFINE_IDR(ipoib_ps);
76 static int next_port;
77
78 struct cma_device {
79 struct list_head list;
80 struct ib_device *device;
81 struct completion comp;
82 atomic_t refcount;
83 struct list_head id_list;
84 };
85
86 enum cma_state {
87 CMA_IDLE,
88 CMA_ADDR_QUERY,
89 CMA_ADDR_RESOLVED,
90 CMA_ROUTE_QUERY,
91 CMA_ROUTE_RESOLVED,
92 CMA_CONNECT,
93 CMA_DISCONNECT,
94 CMA_ADDR_BOUND,
95 CMA_LISTEN,
96 CMA_DEVICE_REMOVAL,
97 CMA_DESTROYING
98 };
99
100 struct rdma_bind_list {
101 struct idr *ps;
102 struct hlist_head owners;
103 unsigned short port;
104 };
105
106 /*
107 * Device removal can occur at anytime, so we need extra handling to
108 * serialize notifying the user of device removal with other callbacks.
109 * We do this by disabling removal notification while a callback is in process,
110 * and reporting it after the callback completes.
111 */
112 struct rdma_id_private {
113 struct rdma_cm_id id;
114
115 struct rdma_bind_list *bind_list;
116 struct hlist_node node;
117 struct list_head list; /* listen_any_list or cma_device.list */
118 struct list_head listen_list; /* per device listens */
119 struct cma_device *cma_dev;
120 struct list_head mc_list;
121
122 int internal_id;
123 enum cma_state state;
124 spinlock_t lock;
125 struct mutex qp_mutex;
126
127 struct completion comp;
128 atomic_t refcount;
129 wait_queue_head_t wait_remove;
130 atomic_t dev_remove;
131
132 int backlog;
133 int timeout_ms;
134 struct ib_sa_query *query;
135 int query_id;
136 union {
137 struct ib_cm_id *ib;
138 struct iw_cm_id *iw;
139 } cm_id;
140
141 u32 seq_num;
142 u32 qkey;
143 u32 qp_num;
144 u8 srq;
145 u8 tos;
146 };
147
148 struct cma_multicast {
149 struct rdma_id_private *id_priv;
150 union {
151 struct ib_sa_multicast *ib;
152 } multicast;
153 struct list_head list;
154 void *context;
155 struct sockaddr addr;
156 u8 pad[sizeof(struct sockaddr_in6) -
157 sizeof(struct sockaddr)];
158 };
159
160 struct cma_work {
161 struct work_struct work;
162 struct rdma_id_private *id;
163 enum cma_state old_state;
164 enum cma_state new_state;
165 struct rdma_cm_event event;
166 };
167
168 union cma_ip_addr {
169 struct in6_addr ip6;
170 struct {
171 __u32 pad[3];
172 __u32 addr;
173 } ip4;
174 };
175
176 struct cma_hdr {
177 u8 cma_version;
178 u8 ip_version; /* IP version: 7:4 */
179 __u16 port;
180 union cma_ip_addr src_addr;
181 union cma_ip_addr dst_addr;
182 };
183
184 struct sdp_hh {
185 u8 bsdh[16];
186 u8 sdp_version; /* Major version: 7:4 */
187 u8 ip_version; /* IP version: 7:4 */
188 u8 sdp_specific1[10];
189 __u16 port;
190 __u16 sdp_specific2;
191 union cma_ip_addr src_addr;
192 union cma_ip_addr dst_addr;
193 };
194
195 struct sdp_hah {
196 u8 bsdh[16];
197 u8 sdp_version;
198 };
199
200 #define CMA_VERSION 0x00
201 #define SDP_MAJ_VERSION 0x2
202
203 static int cma_comp(struct rdma_id_private *id_priv, enum cma_state comp)
204 {
205 unsigned long flags;
206 int ret;
207
208 spin_lock_irqsave(&id_priv->lock, flags);
209 ret = (id_priv->state == comp);
210 spin_unlock_irqrestore(&id_priv->lock, flags);
211 return ret;
212 }
213
214 static int cma_comp_exch(struct rdma_id_private *id_priv,
215 enum cma_state comp, enum cma_state exch)
216 {
217 unsigned long flags;
218 int ret;
219
220 spin_lock_irqsave(&id_priv->lock, flags);
221 if ((ret = (id_priv->state == comp)))
222 id_priv->state = exch;
223 spin_unlock_irqrestore(&id_priv->lock, flags);
224 return ret;
225 }
226
227 static enum cma_state cma_exch(struct rdma_id_private *id_priv,
228 enum cma_state exch)
229 {
230 unsigned long flags;
231 enum cma_state old;
232
233 spin_lock_irqsave(&id_priv->lock, flags);
234 old = id_priv->state;
235 id_priv->state = exch;
236 spin_unlock_irqrestore(&id_priv->lock, flags);
237 return old;
238 }
239
240 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
241 {
242 return hdr->ip_version >> 4;
243 }
244
245 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
246 {
247 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
248 }
249
250 static inline u8 sdp_get_majv(u8 sdp_version)
251 {
252 return sdp_version >> 4;
253 }
254
255 static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
256 {
257 return hh->ip_version >> 4;
258 }
259
260 static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
261 {
262 hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
263 }
264
265 static inline int cma_is_ud_ps(enum rdma_port_space ps)
266 {
267 return (ps == RDMA_PS_UDP || ps == RDMA_PS_IPOIB);
268 }
269
270 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
271 struct cma_device *cma_dev)
272 {
273 atomic_inc(&cma_dev->refcount);
274 id_priv->cma_dev = cma_dev;
275 id_priv->id.device = cma_dev->device;
276 list_add_tail(&id_priv->list, &cma_dev->id_list);
277 }
278
279 static inline void cma_deref_dev(struct cma_device *cma_dev)
280 {
281 if (atomic_dec_and_test(&cma_dev->refcount))
282 complete(&cma_dev->comp);
283 }
284
285 static void cma_detach_from_dev(struct rdma_id_private *id_priv)
286 {
287 list_del(&id_priv->list);
288 cma_deref_dev(id_priv->cma_dev);
289 id_priv->cma_dev = NULL;
290 }
291
292 static int cma_set_qkey(struct ib_device *device, u8 port_num,
293 enum rdma_port_space ps,
294 struct rdma_dev_addr *dev_addr, u32 *qkey)
295 {
296 struct ib_sa_mcmember_rec rec;
297 int ret = 0;
298
299 switch (ps) {
300 case RDMA_PS_UDP:
301 *qkey = RDMA_UDP_QKEY;
302 break;
303 case RDMA_PS_IPOIB:
304 ib_addr_get_mgid(dev_addr, &rec.mgid);
305 ret = ib_sa_get_mcmember_rec(device, port_num, &rec.mgid, &rec);
306 *qkey = be32_to_cpu(rec.qkey);
307 break;
308 default:
309 break;
310 }
311 return ret;
312 }
313
314 static int cma_acquire_dev(struct rdma_id_private *id_priv)
315 {
316 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
317 struct cma_device *cma_dev;
318 union ib_gid gid;
319 int ret = -ENODEV;
320
321 switch (rdma_node_get_transport(dev_addr->dev_type)) {
322 case RDMA_TRANSPORT_IB:
323 ib_addr_get_sgid(dev_addr, &gid);
324 break;
325 case RDMA_TRANSPORT_IWARP:
326 iw_addr_get_sgid(dev_addr, &gid);
327 break;
328 default:
329 return -ENODEV;
330 }
331
332 list_for_each_entry(cma_dev, &dev_list, list) {
333 ret = ib_find_cached_gid(cma_dev->device, &gid,
334 &id_priv->id.port_num, NULL);
335 if (!ret) {
336 ret = cma_set_qkey(cma_dev->device,
337 id_priv->id.port_num,
338 id_priv->id.ps, dev_addr,
339 &id_priv->qkey);
340 if (!ret)
341 cma_attach_to_dev(id_priv, cma_dev);
342 break;
343 }
344 }
345 return ret;
346 }
347
348 static void cma_deref_id(struct rdma_id_private *id_priv)
349 {
350 if (atomic_dec_and_test(&id_priv->refcount))
351 complete(&id_priv->comp);
352 }
353
354 static int cma_disable_remove(struct rdma_id_private *id_priv,
355 enum cma_state state)
356 {
357 unsigned long flags;
358 int ret;
359
360 spin_lock_irqsave(&id_priv->lock, flags);
361 if (id_priv->state == state) {
362 atomic_inc(&id_priv->dev_remove);
363 ret = 0;
364 } else
365 ret = -EINVAL;
366 spin_unlock_irqrestore(&id_priv->lock, flags);
367 return ret;
368 }
369
370 static void cma_enable_remove(struct rdma_id_private *id_priv)
371 {
372 if (atomic_dec_and_test(&id_priv->dev_remove))
373 wake_up(&id_priv->wait_remove);
374 }
375
376 static int cma_has_cm_dev(struct rdma_id_private *id_priv)
377 {
378 return (id_priv->id.device && id_priv->cm_id.ib);
379 }
380
381 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
382 void *context, enum rdma_port_space ps)
383 {
384 struct rdma_id_private *id_priv;
385
386 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
387 if (!id_priv)
388 return ERR_PTR(-ENOMEM);
389
390 id_priv->state = CMA_IDLE;
391 id_priv->id.context = context;
392 id_priv->id.event_handler = event_handler;
393 id_priv->id.ps = ps;
394 spin_lock_init(&id_priv->lock);
395 mutex_init(&id_priv->qp_mutex);
396 init_completion(&id_priv->comp);
397 atomic_set(&id_priv->refcount, 1);
398 init_waitqueue_head(&id_priv->wait_remove);
399 atomic_set(&id_priv->dev_remove, 0);
400 INIT_LIST_HEAD(&id_priv->listen_list);
401 INIT_LIST_HEAD(&id_priv->mc_list);
402 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
403
404 return &id_priv->id;
405 }
406 EXPORT_SYMBOL(rdma_create_id);
407
408 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
409 {
410 struct ib_qp_attr qp_attr;
411 int qp_attr_mask, ret;
412
413 qp_attr.qp_state = IB_QPS_INIT;
414 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
415 if (ret)
416 return ret;
417
418 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
419 if (ret)
420 return ret;
421
422 qp_attr.qp_state = IB_QPS_RTR;
423 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
424 if (ret)
425 return ret;
426
427 qp_attr.qp_state = IB_QPS_RTS;
428 qp_attr.sq_psn = 0;
429 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
430
431 return ret;
432 }
433
434 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
435 {
436 struct ib_qp_attr qp_attr;
437 int qp_attr_mask, ret;
438
439 qp_attr.qp_state = IB_QPS_INIT;
440 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
441 if (ret)
442 return ret;
443
444 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
445 }
446
447 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
448 struct ib_qp_init_attr *qp_init_attr)
449 {
450 struct rdma_id_private *id_priv;
451 struct ib_qp *qp;
452 int ret;
453
454 id_priv = container_of(id, struct rdma_id_private, id);
455 if (id->device != pd->device)
456 return -EINVAL;
457
458 qp = ib_create_qp(pd, qp_init_attr);
459 if (IS_ERR(qp))
460 return PTR_ERR(qp);
461
462 if (cma_is_ud_ps(id_priv->id.ps))
463 ret = cma_init_ud_qp(id_priv, qp);
464 else
465 ret = cma_init_conn_qp(id_priv, qp);
466 if (ret)
467 goto err;
468
469 id->qp = qp;
470 id_priv->qp_num = qp->qp_num;
471 id_priv->srq = (qp->srq != NULL);
472 return 0;
473 err:
474 ib_destroy_qp(qp);
475 return ret;
476 }
477 EXPORT_SYMBOL(rdma_create_qp);
478
479 void rdma_destroy_qp(struct rdma_cm_id *id)
480 {
481 struct rdma_id_private *id_priv;
482
483 id_priv = container_of(id, struct rdma_id_private, id);
484 mutex_lock(&id_priv->qp_mutex);
485 ib_destroy_qp(id_priv->id.qp);
486 id_priv->id.qp = NULL;
487 mutex_unlock(&id_priv->qp_mutex);
488 }
489 EXPORT_SYMBOL(rdma_destroy_qp);
490
491 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv)
492 {
493 struct ib_qp_attr qp_attr;
494 int qp_attr_mask, ret;
495
496 mutex_lock(&id_priv->qp_mutex);
497 if (!id_priv->id.qp) {
498 ret = 0;
499 goto out;
500 }
501
502 /* Need to update QP attributes from default values. */
503 qp_attr.qp_state = IB_QPS_INIT;
504 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
505 if (ret)
506 goto out;
507
508 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
509 if (ret)
510 goto out;
511
512 qp_attr.qp_state = IB_QPS_RTR;
513 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
514 if (ret)
515 goto out;
516
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 {
525 struct ib_qp_attr qp_attr;
526 int qp_attr_mask, ret;
527
528 mutex_lock(&id_priv->qp_mutex);
529 if (!id_priv->id.qp) {
530 ret = 0;
531 goto out;
532 }
533
534 qp_attr.qp_state = IB_QPS_RTS;
535 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
536 if (ret)
537 goto out;
538
539 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
540 out:
541 mutex_unlock(&id_priv->qp_mutex);
542 return ret;
543 }
544
545 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
546 {
547 struct ib_qp_attr qp_attr;
548 int ret;
549
550 mutex_lock(&id_priv->qp_mutex);
551 if (!id_priv->id.qp) {
552 ret = 0;
553 goto out;
554 }
555
556 qp_attr.qp_state = IB_QPS_ERR;
557 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
558 out:
559 mutex_unlock(&id_priv->qp_mutex);
560 return ret;
561 }
562
563 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
564 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
565 {
566 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
567 int ret;
568
569 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
570 ib_addr_get_pkey(dev_addr),
571 &qp_attr->pkey_index);
572 if (ret)
573 return ret;
574
575 qp_attr->port_num = id_priv->id.port_num;
576 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
577
578 if (cma_is_ud_ps(id_priv->id.ps)) {
579 qp_attr->qkey = id_priv->qkey;
580 *qp_attr_mask |= IB_QP_QKEY;
581 } else {
582 qp_attr->qp_access_flags = 0;
583 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
584 }
585 return 0;
586 }
587
588 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
589 int *qp_attr_mask)
590 {
591 struct rdma_id_private *id_priv;
592 int ret = 0;
593
594 id_priv = container_of(id, struct rdma_id_private, id);
595 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
596 case RDMA_TRANSPORT_IB:
597 if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps))
598 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
599 else
600 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
601 qp_attr_mask);
602 if (qp_attr->qp_state == IB_QPS_RTR)
603 qp_attr->rq_psn = id_priv->seq_num;
604 break;
605 case RDMA_TRANSPORT_IWARP:
606 if (!id_priv->cm_id.iw) {
607 qp_attr->qp_access_flags = 0;
608 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
609 } else
610 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
611 qp_attr_mask);
612 break;
613 default:
614 ret = -ENOSYS;
615 break;
616 }
617
618 return ret;
619 }
620 EXPORT_SYMBOL(rdma_init_qp_attr);
621
622 static inline int cma_zero_addr(struct sockaddr *addr)
623 {
624 struct in6_addr *ip6;
625
626 if (addr->sa_family == AF_INET)
627 return ZERONET(((struct sockaddr_in *) addr)->sin_addr.s_addr);
628 else {
629 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
630 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
631 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
632 }
633 }
634
635 static inline int cma_loopback_addr(struct sockaddr *addr)
636 {
637 return LOOPBACK(((struct sockaddr_in *) addr)->sin_addr.s_addr);
638 }
639
640 static inline int cma_any_addr(struct sockaddr *addr)
641 {
642 return cma_zero_addr(addr) || cma_loopback_addr(addr);
643 }
644
645 static inline __be16 cma_port(struct sockaddr *addr)
646 {
647 if (addr->sa_family == AF_INET)
648 return ((struct sockaddr_in *) addr)->sin_port;
649 else
650 return ((struct sockaddr_in6 *) addr)->sin6_port;
651 }
652
653 static inline int cma_any_port(struct sockaddr *addr)
654 {
655 return !cma_port(addr);
656 }
657
658 static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
659 u8 *ip_ver, __u16 *port,
660 union cma_ip_addr **src, union cma_ip_addr **dst)
661 {
662 switch (ps) {
663 case RDMA_PS_SDP:
664 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
665 SDP_MAJ_VERSION)
666 return -EINVAL;
667
668 *ip_ver = sdp_get_ip_ver(hdr);
669 *port = ((struct sdp_hh *) hdr)->port;
670 *src = &((struct sdp_hh *) hdr)->src_addr;
671 *dst = &((struct sdp_hh *) hdr)->dst_addr;
672 break;
673 default:
674 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
675 return -EINVAL;
676
677 *ip_ver = cma_get_ip_ver(hdr);
678 *port = ((struct cma_hdr *) hdr)->port;
679 *src = &((struct cma_hdr *) hdr)->src_addr;
680 *dst = &((struct cma_hdr *) hdr)->dst_addr;
681 break;
682 }
683
684 if (*ip_ver != 4 && *ip_ver != 6)
685 return -EINVAL;
686 return 0;
687 }
688
689 static void cma_save_net_info(struct rdma_addr *addr,
690 struct rdma_addr *listen_addr,
691 u8 ip_ver, __u16 port,
692 union cma_ip_addr *src, union cma_ip_addr *dst)
693 {
694 struct sockaddr_in *listen4, *ip4;
695 struct sockaddr_in6 *listen6, *ip6;
696
697 switch (ip_ver) {
698 case 4:
699 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
700 ip4 = (struct sockaddr_in *) &addr->src_addr;
701 ip4->sin_family = listen4->sin_family;
702 ip4->sin_addr.s_addr = dst->ip4.addr;
703 ip4->sin_port = listen4->sin_port;
704
705 ip4 = (struct sockaddr_in *) &addr->dst_addr;
706 ip4->sin_family = listen4->sin_family;
707 ip4->sin_addr.s_addr = src->ip4.addr;
708 ip4->sin_port = port;
709 break;
710 case 6:
711 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
712 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
713 ip6->sin6_family = listen6->sin6_family;
714 ip6->sin6_addr = dst->ip6;
715 ip6->sin6_port = listen6->sin6_port;
716
717 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
718 ip6->sin6_family = listen6->sin6_family;
719 ip6->sin6_addr = src->ip6;
720 ip6->sin6_port = port;
721 break;
722 default:
723 break;
724 }
725 }
726
727 static inline int cma_user_data_offset(enum rdma_port_space ps)
728 {
729 switch (ps) {
730 case RDMA_PS_SDP:
731 return 0;
732 default:
733 return sizeof(struct cma_hdr);
734 }
735 }
736
737 static void cma_cancel_route(struct rdma_id_private *id_priv)
738 {
739 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
740 case RDMA_TRANSPORT_IB:
741 if (id_priv->query)
742 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
743 break;
744 default:
745 break;
746 }
747 }
748
749 static void cma_cancel_listens(struct rdma_id_private *id_priv)
750 {
751 struct rdma_id_private *dev_id_priv;
752
753 /*
754 * Remove from listen_any_list to prevent added devices from spawning
755 * additional listen requests.
756 */
757 mutex_lock(&lock);
758 list_del(&id_priv->list);
759
760 while (!list_empty(&id_priv->listen_list)) {
761 dev_id_priv = list_entry(id_priv->listen_list.next,
762 struct rdma_id_private, listen_list);
763 /* sync with device removal to avoid duplicate destruction */
764 list_del_init(&dev_id_priv->list);
765 list_del(&dev_id_priv->listen_list);
766 mutex_unlock(&lock);
767
768 rdma_destroy_id(&dev_id_priv->id);
769 mutex_lock(&lock);
770 }
771 mutex_unlock(&lock);
772 }
773
774 static void cma_cancel_operation(struct rdma_id_private *id_priv,
775 enum cma_state state)
776 {
777 switch (state) {
778 case CMA_ADDR_QUERY:
779 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
780 break;
781 case CMA_ROUTE_QUERY:
782 cma_cancel_route(id_priv);
783 break;
784 case CMA_LISTEN:
785 if (cma_any_addr(&id_priv->id.route.addr.src_addr) &&
786 !id_priv->cma_dev)
787 cma_cancel_listens(id_priv);
788 break;
789 default:
790 break;
791 }
792 }
793
794 static void cma_release_port(struct rdma_id_private *id_priv)
795 {
796 struct rdma_bind_list *bind_list = id_priv->bind_list;
797
798 if (!bind_list)
799 return;
800
801 mutex_lock(&lock);
802 hlist_del(&id_priv->node);
803 if (hlist_empty(&bind_list->owners)) {
804 idr_remove(bind_list->ps, bind_list->port);
805 kfree(bind_list);
806 }
807 mutex_unlock(&lock);
808 }
809
810 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
811 {
812 struct cma_multicast *mc;
813
814 while (!list_empty(&id_priv->mc_list)) {
815 mc = container_of(id_priv->mc_list.next,
816 struct cma_multicast, list);
817 list_del(&mc->list);
818 ib_sa_free_multicast(mc->multicast.ib);
819 kfree(mc);
820 }
821 }
822
823 void rdma_destroy_id(struct rdma_cm_id *id)
824 {
825 struct rdma_id_private *id_priv;
826 enum cma_state state;
827
828 id_priv = container_of(id, struct rdma_id_private, id);
829 state = cma_exch(id_priv, CMA_DESTROYING);
830 cma_cancel_operation(id_priv, state);
831
832 mutex_lock(&lock);
833 if (id_priv->cma_dev) {
834 mutex_unlock(&lock);
835 switch (rdma_node_get_transport(id->device->node_type)) {
836 case RDMA_TRANSPORT_IB:
837 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib))
838 ib_destroy_cm_id(id_priv->cm_id.ib);
839 break;
840 case RDMA_TRANSPORT_IWARP:
841 if (id_priv->cm_id.iw && !IS_ERR(id_priv->cm_id.iw))
842 iw_destroy_cm_id(id_priv->cm_id.iw);
843 break;
844 default:
845 break;
846 }
847 cma_leave_mc_groups(id_priv);
848 mutex_lock(&lock);
849 cma_detach_from_dev(id_priv);
850 }
851 mutex_unlock(&lock);
852
853 cma_release_port(id_priv);
854 cma_deref_id(id_priv);
855 wait_for_completion(&id_priv->comp);
856
857 if (id_priv->internal_id)
858 cma_deref_id(id_priv->id.context);
859
860 kfree(id_priv->id.route.path_rec);
861 kfree(id_priv);
862 }
863 EXPORT_SYMBOL(rdma_destroy_id);
864
865 static int cma_rep_recv(struct rdma_id_private *id_priv)
866 {
867 int ret;
868
869 ret = cma_modify_qp_rtr(id_priv);
870 if (ret)
871 goto reject;
872
873 ret = cma_modify_qp_rts(id_priv);
874 if (ret)
875 goto reject;
876
877 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
878 if (ret)
879 goto reject;
880
881 return 0;
882 reject:
883 cma_modify_qp_err(id_priv);
884 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
885 NULL, 0, NULL, 0);
886 return ret;
887 }
888
889 static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
890 {
891 if (id_priv->id.ps == RDMA_PS_SDP &&
892 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
893 SDP_MAJ_VERSION)
894 return -EINVAL;
895
896 return 0;
897 }
898
899 static void cma_set_rep_event_data(struct rdma_cm_event *event,
900 struct ib_cm_rep_event_param *rep_data,
901 void *private_data)
902 {
903 event->param.conn.private_data = private_data;
904 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
905 event->param.conn.responder_resources = rep_data->responder_resources;
906 event->param.conn.initiator_depth = rep_data->initiator_depth;
907 event->param.conn.flow_control = rep_data->flow_control;
908 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
909 event->param.conn.srq = rep_data->srq;
910 event->param.conn.qp_num = rep_data->remote_qpn;
911 }
912
913 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
914 {
915 struct rdma_id_private *id_priv = cm_id->context;
916 struct rdma_cm_event event;
917 int ret = 0;
918
919 if (cma_disable_remove(id_priv, CMA_CONNECT))
920 return 0;
921
922 memset(&event, 0, sizeof event);
923 switch (ib_event->event) {
924 case IB_CM_REQ_ERROR:
925 case IB_CM_REP_ERROR:
926 event.event = RDMA_CM_EVENT_UNREACHABLE;
927 event.status = -ETIMEDOUT;
928 break;
929 case IB_CM_REP_RECEIVED:
930 event.status = cma_verify_rep(id_priv, ib_event->private_data);
931 if (event.status)
932 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
933 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
934 event.status = cma_rep_recv(id_priv);
935 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
936 RDMA_CM_EVENT_ESTABLISHED;
937 } else
938 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
939 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
940 ib_event->private_data);
941 break;
942 case IB_CM_RTU_RECEIVED:
943 case IB_CM_USER_ESTABLISHED:
944 event.event = RDMA_CM_EVENT_ESTABLISHED;
945 break;
946 case IB_CM_DREQ_ERROR:
947 event.status = -ETIMEDOUT; /* fall through */
948 case IB_CM_DREQ_RECEIVED:
949 case IB_CM_DREP_RECEIVED:
950 if (!cma_comp_exch(id_priv, CMA_CONNECT, CMA_DISCONNECT))
951 goto out;
952 event.event = RDMA_CM_EVENT_DISCONNECTED;
953 break;
954 case IB_CM_TIMEWAIT_EXIT:
955 case IB_CM_MRA_RECEIVED:
956 /* ignore event */
957 goto out;
958 case IB_CM_REJ_RECEIVED:
959 cma_modify_qp_err(id_priv);
960 event.status = ib_event->param.rej_rcvd.reason;
961 event.event = RDMA_CM_EVENT_REJECTED;
962 event.param.conn.private_data = ib_event->private_data;
963 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
964 break;
965 default:
966 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d",
967 ib_event->event);
968 goto out;
969 }
970
971 ret = id_priv->id.event_handler(&id_priv->id, &event);
972 if (ret) {
973 /* Destroy the CM ID by returning a non-zero value. */
974 id_priv->cm_id.ib = NULL;
975 cma_exch(id_priv, CMA_DESTROYING);
976 cma_enable_remove(id_priv);
977 rdma_destroy_id(&id_priv->id);
978 return ret;
979 }
980 out:
981 cma_enable_remove(id_priv);
982 return ret;
983 }
984
985 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
986 struct ib_cm_event *ib_event)
987 {
988 struct rdma_id_private *id_priv;
989 struct rdma_cm_id *id;
990 struct rdma_route *rt;
991 union cma_ip_addr *src, *dst;
992 __u16 port;
993 u8 ip_ver;
994
995 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
996 &ip_ver, &port, &src, &dst))
997 goto err;
998
999 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1000 listen_id->ps);
1001 if (IS_ERR(id))
1002 goto err;
1003
1004 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1005 ip_ver, port, src, dst);
1006
1007 rt = &id->route;
1008 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1009 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1010 GFP_KERNEL);
1011 if (!rt->path_rec)
1012 goto destroy_id;
1013
1014 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1015 if (rt->num_paths == 2)
1016 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1017
1018 ib_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1019 ib_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1020 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1021 rt->addr.dev_addr.dev_type = RDMA_NODE_IB_CA;
1022
1023 id_priv = container_of(id, struct rdma_id_private, id);
1024 id_priv->state = CMA_CONNECT;
1025 return id_priv;
1026
1027 destroy_id:
1028 rdma_destroy_id(id);
1029 err:
1030 return NULL;
1031 }
1032
1033 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1034 struct ib_cm_event *ib_event)
1035 {
1036 struct rdma_id_private *id_priv;
1037 struct rdma_cm_id *id;
1038 union cma_ip_addr *src, *dst;
1039 __u16 port;
1040 u8 ip_ver;
1041 int ret;
1042
1043 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1044 listen_id->ps);
1045 if (IS_ERR(id))
1046 return NULL;
1047
1048
1049 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1050 &ip_ver, &port, &src, &dst))
1051 goto err;
1052
1053 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1054 ip_ver, port, src, dst);
1055
1056 ret = rdma_translate_ip(&id->route.addr.src_addr,
1057 &id->route.addr.dev_addr);
1058 if (ret)
1059 goto err;
1060
1061 id_priv = container_of(id, struct rdma_id_private, id);
1062 id_priv->state = CMA_CONNECT;
1063 return id_priv;
1064 err:
1065 rdma_destroy_id(id);
1066 return NULL;
1067 }
1068
1069 static void cma_set_req_event_data(struct rdma_cm_event *event,
1070 struct ib_cm_req_event_param *req_data,
1071 void *private_data, int offset)
1072 {
1073 event->param.conn.private_data = private_data + offset;
1074 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1075 event->param.conn.responder_resources = req_data->responder_resources;
1076 event->param.conn.initiator_depth = req_data->initiator_depth;
1077 event->param.conn.flow_control = req_data->flow_control;
1078 event->param.conn.retry_count = req_data->retry_count;
1079 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1080 event->param.conn.srq = req_data->srq;
1081 event->param.conn.qp_num = req_data->remote_qpn;
1082 }
1083
1084 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1085 {
1086 struct rdma_id_private *listen_id, *conn_id;
1087 struct rdma_cm_event event;
1088 int offset, ret;
1089
1090 listen_id = cm_id->context;
1091 if (cma_disable_remove(listen_id, CMA_LISTEN))
1092 return -ECONNABORTED;
1093
1094 memset(&event, 0, sizeof event);
1095 offset = cma_user_data_offset(listen_id->id.ps);
1096 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1097 if (cma_is_ud_ps(listen_id->id.ps)) {
1098 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1099 event.param.ud.private_data = ib_event->private_data + offset;
1100 event.param.ud.private_data_len =
1101 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1102 } else {
1103 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1104 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1105 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1106 ib_event->private_data, offset);
1107 }
1108 if (!conn_id) {
1109 ret = -ENOMEM;
1110 goto out;
1111 }
1112
1113 atomic_inc(&conn_id->dev_remove);
1114 mutex_lock(&lock);
1115 ret = cma_acquire_dev(conn_id);
1116 mutex_unlock(&lock);
1117 if (ret)
1118 goto release_conn_id;
1119
1120 conn_id->cm_id.ib = cm_id;
1121 cm_id->context = conn_id;
1122 cm_id->cm_handler = cma_ib_handler;
1123
1124 ret = conn_id->id.event_handler(&conn_id->id, &event);
1125 if (!ret) {
1126 cma_enable_remove(conn_id);
1127 goto out;
1128 }
1129
1130 /* Destroy the CM ID by returning a non-zero value. */
1131 conn_id->cm_id.ib = NULL;
1132
1133 release_conn_id:
1134 cma_exch(conn_id, CMA_DESTROYING);
1135 cma_enable_remove(conn_id);
1136 rdma_destroy_id(&conn_id->id);
1137
1138 out:
1139 cma_enable_remove(listen_id);
1140 return ret;
1141 }
1142
1143 static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1144 {
1145 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
1146 }
1147
1148 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1149 struct ib_cm_compare_data *compare)
1150 {
1151 struct cma_hdr *cma_data, *cma_mask;
1152 struct sdp_hh *sdp_data, *sdp_mask;
1153 __u32 ip4_addr;
1154 struct in6_addr ip6_addr;
1155
1156 memset(compare, 0, sizeof *compare);
1157 cma_data = (void *) compare->data;
1158 cma_mask = (void *) compare->mask;
1159 sdp_data = (void *) compare->data;
1160 sdp_mask = (void *) compare->mask;
1161
1162 switch (addr->sa_family) {
1163 case AF_INET:
1164 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1165 if (ps == RDMA_PS_SDP) {
1166 sdp_set_ip_ver(sdp_data, 4);
1167 sdp_set_ip_ver(sdp_mask, 0xF);
1168 sdp_data->dst_addr.ip4.addr = ip4_addr;
1169 sdp_mask->dst_addr.ip4.addr = ~0;
1170 } else {
1171 cma_set_ip_ver(cma_data, 4);
1172 cma_set_ip_ver(cma_mask, 0xF);
1173 cma_data->dst_addr.ip4.addr = ip4_addr;
1174 cma_mask->dst_addr.ip4.addr = ~0;
1175 }
1176 break;
1177 case AF_INET6:
1178 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1179 if (ps == RDMA_PS_SDP) {
1180 sdp_set_ip_ver(sdp_data, 6);
1181 sdp_set_ip_ver(sdp_mask, 0xF);
1182 sdp_data->dst_addr.ip6 = ip6_addr;
1183 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1184 sizeof sdp_mask->dst_addr.ip6);
1185 } else {
1186 cma_set_ip_ver(cma_data, 6);
1187 cma_set_ip_ver(cma_mask, 0xF);
1188 cma_data->dst_addr.ip6 = ip6_addr;
1189 memset(&cma_mask->dst_addr.ip6, 0xFF,
1190 sizeof cma_mask->dst_addr.ip6);
1191 }
1192 break;
1193 default:
1194 break;
1195 }
1196 }
1197
1198 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1199 {
1200 struct rdma_id_private *id_priv = iw_id->context;
1201 struct rdma_cm_event event;
1202 struct sockaddr_in *sin;
1203 int ret = 0;
1204
1205 if (cma_disable_remove(id_priv, CMA_CONNECT))
1206 return 0;
1207
1208 memset(&event, 0, sizeof event);
1209 switch (iw_event->event) {
1210 case IW_CM_EVENT_CLOSE:
1211 event.event = RDMA_CM_EVENT_DISCONNECTED;
1212 break;
1213 case IW_CM_EVENT_CONNECT_REPLY:
1214 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1215 *sin = iw_event->local_addr;
1216 sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
1217 *sin = iw_event->remote_addr;
1218 switch (iw_event->status) {
1219 case 0:
1220 event.event = RDMA_CM_EVENT_ESTABLISHED;
1221 break;
1222 case -ECONNRESET:
1223 case -ECONNREFUSED:
1224 event.event = RDMA_CM_EVENT_REJECTED;
1225 break;
1226 case -ETIMEDOUT:
1227 event.event = RDMA_CM_EVENT_UNREACHABLE;
1228 break;
1229 default:
1230 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1231 break;
1232 }
1233 break;
1234 case IW_CM_EVENT_ESTABLISHED:
1235 event.event = RDMA_CM_EVENT_ESTABLISHED;
1236 break;
1237 default:
1238 BUG_ON(1);
1239 }
1240
1241 event.status = iw_event->status;
1242 event.param.conn.private_data = iw_event->private_data;
1243 event.param.conn.private_data_len = iw_event->private_data_len;
1244 ret = id_priv->id.event_handler(&id_priv->id, &event);
1245 if (ret) {
1246 /* Destroy the CM ID by returning a non-zero value. */
1247 id_priv->cm_id.iw = NULL;
1248 cma_exch(id_priv, CMA_DESTROYING);
1249 cma_enable_remove(id_priv);
1250 rdma_destroy_id(&id_priv->id);
1251 return ret;
1252 }
1253
1254 cma_enable_remove(id_priv);
1255 return ret;
1256 }
1257
1258 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1259 struct iw_cm_event *iw_event)
1260 {
1261 struct rdma_cm_id *new_cm_id;
1262 struct rdma_id_private *listen_id, *conn_id;
1263 struct sockaddr_in *sin;
1264 struct net_device *dev = NULL;
1265 struct rdma_cm_event event;
1266 int ret;
1267 struct ib_device_attr attr;
1268
1269 listen_id = cm_id->context;
1270 if (cma_disable_remove(listen_id, CMA_LISTEN))
1271 return -ECONNABORTED;
1272
1273 /* Create a new RDMA id for the new IW CM ID */
1274 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1275 listen_id->id.context,
1276 RDMA_PS_TCP);
1277 if (!new_cm_id) {
1278 ret = -ENOMEM;
1279 goto out;
1280 }
1281 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1282 atomic_inc(&conn_id->dev_remove);
1283 conn_id->state = CMA_CONNECT;
1284
1285 dev = ip_dev_find(iw_event->local_addr.sin_addr.s_addr);
1286 if (!dev) {
1287 ret = -EADDRNOTAVAIL;
1288 cma_enable_remove(conn_id);
1289 rdma_destroy_id(new_cm_id);
1290 goto out;
1291 }
1292 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1293 if (ret) {
1294 cma_enable_remove(conn_id);
1295 rdma_destroy_id(new_cm_id);
1296 goto out;
1297 }
1298
1299 mutex_lock(&lock);
1300 ret = cma_acquire_dev(conn_id);
1301 mutex_unlock(&lock);
1302 if (ret) {
1303 cma_enable_remove(conn_id);
1304 rdma_destroy_id(new_cm_id);
1305 goto out;
1306 }
1307
1308 conn_id->cm_id.iw = cm_id;
1309 cm_id->context = conn_id;
1310 cm_id->cm_handler = cma_iw_handler;
1311
1312 sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
1313 *sin = iw_event->local_addr;
1314 sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
1315 *sin = iw_event->remote_addr;
1316
1317 ret = ib_query_device(conn_id->id.device, &attr);
1318 if (ret) {
1319 cma_enable_remove(conn_id);
1320 rdma_destroy_id(new_cm_id);
1321 goto out;
1322 }
1323
1324 memset(&event, 0, sizeof event);
1325 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1326 event.param.conn.private_data = iw_event->private_data;
1327 event.param.conn.private_data_len = iw_event->private_data_len;
1328 event.param.conn.initiator_depth = attr.max_qp_init_rd_atom;
1329 event.param.conn.responder_resources = attr.max_qp_rd_atom;
1330 ret = conn_id->id.event_handler(&conn_id->id, &event);
1331 if (ret) {
1332 /* User wants to destroy the CM ID */
1333 conn_id->cm_id.iw = NULL;
1334 cma_exch(conn_id, CMA_DESTROYING);
1335 cma_enable_remove(conn_id);
1336 rdma_destroy_id(&conn_id->id);
1337 }
1338
1339 out:
1340 if (dev)
1341 dev_put(dev);
1342 cma_enable_remove(listen_id);
1343 return ret;
1344 }
1345
1346 static int cma_ib_listen(struct rdma_id_private *id_priv)
1347 {
1348 struct ib_cm_compare_data compare_data;
1349 struct sockaddr *addr;
1350 __be64 svc_id;
1351 int ret;
1352
1353 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_req_handler,
1354 id_priv);
1355 if (IS_ERR(id_priv->cm_id.ib))
1356 return PTR_ERR(id_priv->cm_id.ib);
1357
1358 addr = &id_priv->id.route.addr.src_addr;
1359 svc_id = cma_get_service_id(id_priv->id.ps, addr);
1360 if (cma_any_addr(addr))
1361 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1362 else {
1363 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1364 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1365 }
1366
1367 if (ret) {
1368 ib_destroy_cm_id(id_priv->cm_id.ib);
1369 id_priv->cm_id.ib = NULL;
1370 }
1371
1372 return ret;
1373 }
1374
1375 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1376 {
1377 int ret;
1378 struct sockaddr_in *sin;
1379
1380 id_priv->cm_id.iw = iw_create_cm_id(id_priv->id.device,
1381 iw_conn_req_handler,
1382 id_priv);
1383 if (IS_ERR(id_priv->cm_id.iw))
1384 return PTR_ERR(id_priv->cm_id.iw);
1385
1386 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1387 id_priv->cm_id.iw->local_addr = *sin;
1388
1389 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1390
1391 if (ret) {
1392 iw_destroy_cm_id(id_priv->cm_id.iw);
1393 id_priv->cm_id.iw = NULL;
1394 }
1395
1396 return ret;
1397 }
1398
1399 static int cma_listen_handler(struct rdma_cm_id *id,
1400 struct rdma_cm_event *event)
1401 {
1402 struct rdma_id_private *id_priv = id->context;
1403
1404 id->context = id_priv->id.context;
1405 id->event_handler = id_priv->id.event_handler;
1406 return id_priv->id.event_handler(id, event);
1407 }
1408
1409 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1410 struct cma_device *cma_dev)
1411 {
1412 struct rdma_id_private *dev_id_priv;
1413 struct rdma_cm_id *id;
1414 int ret;
1415
1416 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps);
1417 if (IS_ERR(id))
1418 return;
1419
1420 dev_id_priv = container_of(id, struct rdma_id_private, id);
1421
1422 dev_id_priv->state = CMA_ADDR_BOUND;
1423 memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
1424 ip_addr_size(&id_priv->id.route.addr.src_addr));
1425
1426 cma_attach_to_dev(dev_id_priv, cma_dev);
1427 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1428 atomic_inc(&id_priv->refcount);
1429 dev_id_priv->internal_id = 1;
1430
1431 ret = rdma_listen(id, id_priv->backlog);
1432 if (ret)
1433 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1434 "listening on device %s", ret, cma_dev->device->name);
1435 }
1436
1437 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1438 {
1439 struct cma_device *cma_dev;
1440
1441 mutex_lock(&lock);
1442 list_add_tail(&id_priv->list, &listen_any_list);
1443 list_for_each_entry(cma_dev, &dev_list, list)
1444 cma_listen_on_dev(id_priv, cma_dev);
1445 mutex_unlock(&lock);
1446 }
1447
1448 static int cma_bind_any(struct rdma_cm_id *id, sa_family_t af)
1449 {
1450 struct sockaddr_in addr_in;
1451
1452 memset(&addr_in, 0, sizeof addr_in);
1453 addr_in.sin_family = af;
1454 return rdma_bind_addr(id, (struct sockaddr *) &addr_in);
1455 }
1456
1457 int rdma_listen(struct rdma_cm_id *id, int backlog)
1458 {
1459 struct rdma_id_private *id_priv;
1460 int ret;
1461
1462 id_priv = container_of(id, struct rdma_id_private, id);
1463 if (id_priv->state == CMA_IDLE) {
1464 ret = cma_bind_any(id, AF_INET);
1465 if (ret)
1466 return ret;
1467 }
1468
1469 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
1470 return -EINVAL;
1471
1472 id_priv->backlog = backlog;
1473 if (id->device) {
1474 switch (rdma_node_get_transport(id->device->node_type)) {
1475 case RDMA_TRANSPORT_IB:
1476 ret = cma_ib_listen(id_priv);
1477 if (ret)
1478 goto err;
1479 break;
1480 case RDMA_TRANSPORT_IWARP:
1481 ret = cma_iw_listen(id_priv, backlog);
1482 if (ret)
1483 goto err;
1484 break;
1485 default:
1486 ret = -ENOSYS;
1487 goto err;
1488 }
1489 } else
1490 cma_listen_on_all(id_priv);
1491
1492 return 0;
1493 err:
1494 id_priv->backlog = 0;
1495 cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
1496 return ret;
1497 }
1498 EXPORT_SYMBOL(rdma_listen);
1499
1500 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1501 {
1502 struct rdma_id_private *id_priv;
1503
1504 id_priv = container_of(id, struct rdma_id_private, id);
1505 id_priv->tos = (u8) tos;
1506 }
1507 EXPORT_SYMBOL(rdma_set_service_type);
1508
1509 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1510 void *context)
1511 {
1512 struct cma_work *work = context;
1513 struct rdma_route *route;
1514
1515 route = &work->id->id.route;
1516
1517 if (!status) {
1518 route->num_paths = 1;
1519 *route->path_rec = *path_rec;
1520 } else {
1521 work->old_state = CMA_ROUTE_QUERY;
1522 work->new_state = CMA_ADDR_RESOLVED;
1523 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1524 work->event.status = status;
1525 }
1526
1527 queue_work(cma_wq, &work->work);
1528 }
1529
1530 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1531 struct cma_work *work)
1532 {
1533 struct rdma_addr *addr = &id_priv->id.route.addr;
1534 struct ib_sa_path_rec path_rec;
1535 ib_sa_comp_mask comp_mask;
1536 struct sockaddr_in6 *sin6;
1537
1538 memset(&path_rec, 0, sizeof path_rec);
1539 ib_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
1540 ib_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
1541 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
1542 path_rec.numb_path = 1;
1543 path_rec.reversible = 1;
1544 path_rec.service_id = cma_get_service_id(id_priv->id.ps, &addr->dst_addr);
1545
1546 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1547 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1548 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1549
1550 if (addr->src_addr.sa_family == AF_INET) {
1551 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1552 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1553 } else {
1554 sin6 = (struct sockaddr_in6 *) &addr->src_addr;
1555 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1556 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1557 }
1558
1559 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1560 id_priv->id.port_num, &path_rec,
1561 comp_mask, timeout_ms,
1562 GFP_KERNEL, cma_query_handler,
1563 work, &id_priv->query);
1564
1565 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1566 }
1567
1568 static void cma_work_handler(struct work_struct *_work)
1569 {
1570 struct cma_work *work = container_of(_work, struct cma_work, work);
1571 struct rdma_id_private *id_priv = work->id;
1572 int destroy = 0;
1573
1574 atomic_inc(&id_priv->dev_remove);
1575 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1576 goto out;
1577
1578 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1579 cma_exch(id_priv, CMA_DESTROYING);
1580 destroy = 1;
1581 }
1582 out:
1583 cma_enable_remove(id_priv);
1584 cma_deref_id(id_priv);
1585 if (destroy)
1586 rdma_destroy_id(&id_priv->id);
1587 kfree(work);
1588 }
1589
1590 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1591 {
1592 struct rdma_route *route = &id_priv->id.route;
1593 struct cma_work *work;
1594 int ret;
1595
1596 work = kzalloc(sizeof *work, GFP_KERNEL);
1597 if (!work)
1598 return -ENOMEM;
1599
1600 work->id = id_priv;
1601 INIT_WORK(&work->work, cma_work_handler);
1602 work->old_state = CMA_ROUTE_QUERY;
1603 work->new_state = CMA_ROUTE_RESOLVED;
1604 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1605
1606 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1607 if (!route->path_rec) {
1608 ret = -ENOMEM;
1609 goto err1;
1610 }
1611
1612 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1613 if (ret)
1614 goto err2;
1615
1616 return 0;
1617 err2:
1618 kfree(route->path_rec);
1619 route->path_rec = NULL;
1620 err1:
1621 kfree(work);
1622 return ret;
1623 }
1624
1625 int rdma_set_ib_paths(struct rdma_cm_id *id,
1626 struct ib_sa_path_rec *path_rec, int num_paths)
1627 {
1628 struct rdma_id_private *id_priv;
1629 int ret;
1630
1631 id_priv = container_of(id, struct rdma_id_private, id);
1632 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_RESOLVED))
1633 return -EINVAL;
1634
1635 id->route.path_rec = kmalloc(sizeof *path_rec * num_paths, GFP_KERNEL);
1636 if (!id->route.path_rec) {
1637 ret = -ENOMEM;
1638 goto err;
1639 }
1640
1641 memcpy(id->route.path_rec, path_rec, sizeof *path_rec * num_paths);
1642 return 0;
1643 err:
1644 cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_ADDR_RESOLVED);
1645 return ret;
1646 }
1647 EXPORT_SYMBOL(rdma_set_ib_paths);
1648
1649 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1650 {
1651 struct cma_work *work;
1652
1653 work = kzalloc(sizeof *work, GFP_KERNEL);
1654 if (!work)
1655 return -ENOMEM;
1656
1657 work->id = id_priv;
1658 INIT_WORK(&work->work, cma_work_handler);
1659 work->old_state = CMA_ROUTE_QUERY;
1660 work->new_state = CMA_ROUTE_RESOLVED;
1661 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1662 queue_work(cma_wq, &work->work);
1663 return 0;
1664 }
1665
1666 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1667 {
1668 struct rdma_id_private *id_priv;
1669 int ret;
1670
1671 id_priv = container_of(id, struct rdma_id_private, id);
1672 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_QUERY))
1673 return -EINVAL;
1674
1675 atomic_inc(&id_priv->refcount);
1676 switch (rdma_node_get_transport(id->device->node_type)) {
1677 case RDMA_TRANSPORT_IB:
1678 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1679 break;
1680 case RDMA_TRANSPORT_IWARP:
1681 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1682 break;
1683 default:
1684 ret = -ENOSYS;
1685 break;
1686 }
1687 if (ret)
1688 goto err;
1689
1690 return 0;
1691 err:
1692 cma_comp_exch(id_priv, CMA_ROUTE_QUERY, CMA_ADDR_RESOLVED);
1693 cma_deref_id(id_priv);
1694 return ret;
1695 }
1696 EXPORT_SYMBOL(rdma_resolve_route);
1697
1698 static int cma_bind_loopback(struct rdma_id_private *id_priv)
1699 {
1700 struct cma_device *cma_dev;
1701 struct ib_port_attr port_attr;
1702 union ib_gid gid;
1703 u16 pkey;
1704 int ret;
1705 u8 p;
1706
1707 mutex_lock(&lock);
1708 if (list_empty(&dev_list)) {
1709 ret = -ENODEV;
1710 goto out;
1711 }
1712 list_for_each_entry(cma_dev, &dev_list, list)
1713 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
1714 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
1715 port_attr.state == IB_PORT_ACTIVE)
1716 goto port_found;
1717
1718 p = 1;
1719 cma_dev = list_entry(dev_list.next, struct cma_device, list);
1720
1721 port_found:
1722 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
1723 if (ret)
1724 goto out;
1725
1726 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
1727 if (ret)
1728 goto out;
1729
1730 ib_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1731 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1732 id_priv->id.port_num = p;
1733 cma_attach_to_dev(id_priv, cma_dev);
1734 out:
1735 mutex_unlock(&lock);
1736 return ret;
1737 }
1738
1739 static void addr_handler(int status, struct sockaddr *src_addr,
1740 struct rdma_dev_addr *dev_addr, void *context)
1741 {
1742 struct rdma_id_private *id_priv = context;
1743 struct rdma_cm_event event;
1744
1745 memset(&event, 0, sizeof event);
1746 atomic_inc(&id_priv->dev_remove);
1747
1748 /*
1749 * Grab mutex to block rdma_destroy_id() from removing the device while
1750 * we're trying to acquire it.
1751 */
1752 mutex_lock(&lock);
1753 if (!cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_RESOLVED)) {
1754 mutex_unlock(&lock);
1755 goto out;
1756 }
1757
1758 if (!status && !id_priv->cma_dev)
1759 status = cma_acquire_dev(id_priv);
1760 mutex_unlock(&lock);
1761
1762 if (status) {
1763 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ADDR_BOUND))
1764 goto out;
1765 event.event = RDMA_CM_EVENT_ADDR_ERROR;
1766 event.status = status;
1767 } else {
1768 memcpy(&id_priv->id.route.addr.src_addr, src_addr,
1769 ip_addr_size(src_addr));
1770 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1771 }
1772
1773 if (id_priv->id.event_handler(&id_priv->id, &event)) {
1774 cma_exch(id_priv, CMA_DESTROYING);
1775 cma_enable_remove(id_priv);
1776 cma_deref_id(id_priv);
1777 rdma_destroy_id(&id_priv->id);
1778 return;
1779 }
1780 out:
1781 cma_enable_remove(id_priv);
1782 cma_deref_id(id_priv);
1783 }
1784
1785 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1786 {
1787 struct cma_work *work;
1788 struct sockaddr_in *src_in, *dst_in;
1789 union ib_gid gid;
1790 int ret;
1791
1792 work = kzalloc(sizeof *work, GFP_KERNEL);
1793 if (!work)
1794 return -ENOMEM;
1795
1796 if (!id_priv->cma_dev) {
1797 ret = cma_bind_loopback(id_priv);
1798 if (ret)
1799 goto err;
1800 }
1801
1802 ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1803 ib_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
1804
1805 if (cma_zero_addr(&id_priv->id.route.addr.src_addr)) {
1806 src_in = (struct sockaddr_in *)&id_priv->id.route.addr.src_addr;
1807 dst_in = (struct sockaddr_in *)&id_priv->id.route.addr.dst_addr;
1808 src_in->sin_family = dst_in->sin_family;
1809 src_in->sin_addr.s_addr = dst_in->sin_addr.s_addr;
1810 }
1811
1812 work->id = id_priv;
1813 INIT_WORK(&work->work, cma_work_handler);
1814 work->old_state = CMA_ADDR_QUERY;
1815 work->new_state = CMA_ADDR_RESOLVED;
1816 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1817 queue_work(cma_wq, &work->work);
1818 return 0;
1819 err:
1820 kfree(work);
1821 return ret;
1822 }
1823
1824 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1825 struct sockaddr *dst_addr)
1826 {
1827 if (src_addr && src_addr->sa_family)
1828 return rdma_bind_addr(id, src_addr);
1829 else
1830 return cma_bind_any(id, dst_addr->sa_family);
1831 }
1832
1833 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1834 struct sockaddr *dst_addr, int timeout_ms)
1835 {
1836 struct rdma_id_private *id_priv;
1837 int ret;
1838
1839 id_priv = container_of(id, struct rdma_id_private, id);
1840 if (id_priv->state == CMA_IDLE) {
1841 ret = cma_bind_addr(id, src_addr, dst_addr);
1842 if (ret)
1843 return ret;
1844 }
1845
1846 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_ADDR_QUERY))
1847 return -EINVAL;
1848
1849 atomic_inc(&id_priv->refcount);
1850 memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
1851 if (cma_any_addr(dst_addr))
1852 ret = cma_resolve_loopback(id_priv);
1853 else
1854 ret = rdma_resolve_ip(&addr_client, &id->route.addr.src_addr,
1855 dst_addr, &id->route.addr.dev_addr,
1856 timeout_ms, addr_handler, id_priv);
1857 if (ret)
1858 goto err;
1859
1860 return 0;
1861 err:
1862 cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_BOUND);
1863 cma_deref_id(id_priv);
1864 return ret;
1865 }
1866 EXPORT_SYMBOL(rdma_resolve_addr);
1867
1868 static void cma_bind_port(struct rdma_bind_list *bind_list,
1869 struct rdma_id_private *id_priv)
1870 {
1871 struct sockaddr_in *sin;
1872
1873 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1874 sin->sin_port = htons(bind_list->port);
1875 id_priv->bind_list = bind_list;
1876 hlist_add_head(&id_priv->node, &bind_list->owners);
1877 }
1878
1879 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
1880 unsigned short snum)
1881 {
1882 struct rdma_bind_list *bind_list;
1883 int port, ret;
1884
1885 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1886 if (!bind_list)
1887 return -ENOMEM;
1888
1889 do {
1890 ret = idr_get_new_above(ps, bind_list, snum, &port);
1891 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1892
1893 if (ret)
1894 goto err1;
1895
1896 if (port != snum) {
1897 ret = -EADDRNOTAVAIL;
1898 goto err2;
1899 }
1900
1901 bind_list->ps = ps;
1902 bind_list->port = (unsigned short) port;
1903 cma_bind_port(bind_list, id_priv);
1904 return 0;
1905 err2:
1906 idr_remove(ps, port);
1907 err1:
1908 kfree(bind_list);
1909 return ret;
1910 }
1911
1912 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
1913 {
1914 struct rdma_bind_list *bind_list;
1915 int port, ret, low, high;
1916
1917 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1918 if (!bind_list)
1919 return -ENOMEM;
1920
1921 retry:
1922 /* FIXME: add proper port randomization per like inet_csk_get_port */
1923 do {
1924 ret = idr_get_new_above(ps, bind_list, next_port, &port);
1925 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1926
1927 if (ret)
1928 goto err1;
1929
1930 inet_get_local_port_range(&low, &high);
1931 if (port > high) {
1932 if (next_port != low) {
1933 idr_remove(ps, port);
1934 next_port = low;
1935 goto retry;
1936 }
1937 ret = -EADDRNOTAVAIL;
1938 goto err2;
1939 }
1940
1941 if (port == high)
1942 next_port = low;
1943 else
1944 next_port = port + 1;
1945
1946 bind_list->ps = ps;
1947 bind_list->port = (unsigned short) port;
1948 cma_bind_port(bind_list, id_priv);
1949 return 0;
1950 err2:
1951 idr_remove(ps, port);
1952 err1:
1953 kfree(bind_list);
1954 return ret;
1955 }
1956
1957 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
1958 {
1959 struct rdma_id_private *cur_id;
1960 struct sockaddr_in *sin, *cur_sin;
1961 struct rdma_bind_list *bind_list;
1962 struct hlist_node *node;
1963 unsigned short snum;
1964
1965 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1966 snum = ntohs(sin->sin_port);
1967 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
1968 return -EACCES;
1969
1970 bind_list = idr_find(ps, snum);
1971 if (!bind_list)
1972 return cma_alloc_port(ps, id_priv, snum);
1973
1974 /*
1975 * We don't support binding to any address if anyone is bound to
1976 * a specific address on the same port.
1977 */
1978 if (cma_any_addr(&id_priv->id.route.addr.src_addr))
1979 return -EADDRNOTAVAIL;
1980
1981 hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
1982 if (cma_any_addr(&cur_id->id.route.addr.src_addr))
1983 return -EADDRNOTAVAIL;
1984
1985 cur_sin = (struct sockaddr_in *) &cur_id->id.route.addr.src_addr;
1986 if (sin->sin_addr.s_addr == cur_sin->sin_addr.s_addr)
1987 return -EADDRINUSE;
1988 }
1989
1990 cma_bind_port(bind_list, id_priv);
1991 return 0;
1992 }
1993
1994 static int cma_get_port(struct rdma_id_private *id_priv)
1995 {
1996 struct idr *ps;
1997 int ret;
1998
1999 switch (id_priv->id.ps) {
2000 case RDMA_PS_SDP:
2001 ps = &sdp_ps;
2002 break;
2003 case RDMA_PS_TCP:
2004 ps = &tcp_ps;
2005 break;
2006 case RDMA_PS_UDP:
2007 ps = &udp_ps;
2008 break;
2009 case RDMA_PS_IPOIB:
2010 ps = &ipoib_ps;
2011 break;
2012 default:
2013 return -EPROTONOSUPPORT;
2014 }
2015
2016 mutex_lock(&lock);
2017 if (cma_any_port(&id_priv->id.route.addr.src_addr))
2018 ret = cma_alloc_any_port(ps, id_priv);
2019 else
2020 ret = cma_use_port(ps, id_priv);
2021 mutex_unlock(&lock);
2022
2023 return ret;
2024 }
2025
2026 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2027 {
2028 struct rdma_id_private *id_priv;
2029 int ret;
2030
2031 if (addr->sa_family != AF_INET)
2032 return -EAFNOSUPPORT;
2033
2034 id_priv = container_of(id, struct rdma_id_private, id);
2035 if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND))
2036 return -EINVAL;
2037
2038 if (!cma_any_addr(addr)) {
2039 ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
2040 if (ret)
2041 goto err1;
2042
2043 mutex_lock(&lock);
2044 ret = cma_acquire_dev(id_priv);
2045 mutex_unlock(&lock);
2046 if (ret)
2047 goto err1;
2048 }
2049
2050 memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
2051 ret = cma_get_port(id_priv);
2052 if (ret)
2053 goto err2;
2054
2055 return 0;
2056 err2:
2057 if (!cma_any_addr(addr)) {
2058 mutex_lock(&lock);
2059 cma_detach_from_dev(id_priv);
2060 mutex_unlock(&lock);
2061 }
2062 err1:
2063 cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_IDLE);
2064 return ret;
2065 }
2066 EXPORT_SYMBOL(rdma_bind_addr);
2067
2068 static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
2069 struct rdma_route *route)
2070 {
2071 struct sockaddr_in *src4, *dst4;
2072 struct cma_hdr *cma_hdr;
2073 struct sdp_hh *sdp_hdr;
2074
2075 src4 = (struct sockaddr_in *) &route->addr.src_addr;
2076 dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
2077
2078 switch (ps) {
2079 case RDMA_PS_SDP:
2080 sdp_hdr = hdr;
2081 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2082 return -EINVAL;
2083 sdp_set_ip_ver(sdp_hdr, 4);
2084 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2085 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2086 sdp_hdr->port = src4->sin_port;
2087 break;
2088 default:
2089 cma_hdr = hdr;
2090 cma_hdr->cma_version = CMA_VERSION;
2091 cma_set_ip_ver(cma_hdr, 4);
2092 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2093 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2094 cma_hdr->port = src4->sin_port;
2095 break;
2096 }
2097 return 0;
2098 }
2099
2100 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2101 struct ib_cm_event *ib_event)
2102 {
2103 struct rdma_id_private *id_priv = cm_id->context;
2104 struct rdma_cm_event event;
2105 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2106 int ret = 0;
2107
2108 if (cma_disable_remove(id_priv, CMA_CONNECT))
2109 return 0;
2110
2111 memset(&event, 0, sizeof event);
2112 switch (ib_event->event) {
2113 case IB_CM_SIDR_REQ_ERROR:
2114 event.event = RDMA_CM_EVENT_UNREACHABLE;
2115 event.status = -ETIMEDOUT;
2116 break;
2117 case IB_CM_SIDR_REP_RECEIVED:
2118 event.param.ud.private_data = ib_event->private_data;
2119 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2120 if (rep->status != IB_SIDR_SUCCESS) {
2121 event.event = RDMA_CM_EVENT_UNREACHABLE;
2122 event.status = ib_event->param.sidr_rep_rcvd.status;
2123 break;
2124 }
2125 if (id_priv->qkey != rep->qkey) {
2126 event.event = RDMA_CM_EVENT_UNREACHABLE;
2127 event.status = -EINVAL;
2128 break;
2129 }
2130 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2131 id_priv->id.route.path_rec,
2132 &event.param.ud.ah_attr);
2133 event.param.ud.qp_num = rep->qpn;
2134 event.param.ud.qkey = rep->qkey;
2135 event.event = RDMA_CM_EVENT_ESTABLISHED;
2136 event.status = 0;
2137 break;
2138 default:
2139 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d",
2140 ib_event->event);
2141 goto out;
2142 }
2143
2144 ret = id_priv->id.event_handler(&id_priv->id, &event);
2145 if (ret) {
2146 /* Destroy the CM ID by returning a non-zero value. */
2147 id_priv->cm_id.ib = NULL;
2148 cma_exch(id_priv, CMA_DESTROYING);
2149 cma_enable_remove(id_priv);
2150 rdma_destroy_id(&id_priv->id);
2151 return ret;
2152 }
2153 out:
2154 cma_enable_remove(id_priv);
2155 return ret;
2156 }
2157
2158 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2159 struct rdma_conn_param *conn_param)
2160 {
2161 struct ib_cm_sidr_req_param req;
2162 struct rdma_route *route;
2163 int ret;
2164
2165 req.private_data_len = sizeof(struct cma_hdr) +
2166 conn_param->private_data_len;
2167 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2168 if (!req.private_data)
2169 return -ENOMEM;
2170
2171 if (conn_param->private_data && conn_param->private_data_len)
2172 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2173 conn_param->private_data, conn_param->private_data_len);
2174
2175 route = &id_priv->id.route;
2176 ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
2177 if (ret)
2178 goto out;
2179
2180 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device,
2181 cma_sidr_rep_handler, id_priv);
2182 if (IS_ERR(id_priv->cm_id.ib)) {
2183 ret = PTR_ERR(id_priv->cm_id.ib);
2184 goto out;
2185 }
2186
2187 req.path = route->path_rec;
2188 req.service_id = cma_get_service_id(id_priv->id.ps,
2189 &route->addr.dst_addr);
2190 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2191 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2192
2193 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2194 if (ret) {
2195 ib_destroy_cm_id(id_priv->cm_id.ib);
2196 id_priv->cm_id.ib = NULL;
2197 }
2198 out:
2199 kfree(req.private_data);
2200 return ret;
2201 }
2202
2203 static int cma_connect_ib(struct rdma_id_private *id_priv,
2204 struct rdma_conn_param *conn_param)
2205 {
2206 struct ib_cm_req_param req;
2207 struct rdma_route *route;
2208 void *private_data;
2209 int offset, ret;
2210
2211 memset(&req, 0, sizeof req);
2212 offset = cma_user_data_offset(id_priv->id.ps);
2213 req.private_data_len = offset + conn_param->private_data_len;
2214 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2215 if (!private_data)
2216 return -ENOMEM;
2217
2218 if (conn_param->private_data && conn_param->private_data_len)
2219 memcpy(private_data + offset, conn_param->private_data,
2220 conn_param->private_data_len);
2221
2222 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_ib_handler,
2223 id_priv);
2224 if (IS_ERR(id_priv->cm_id.ib)) {
2225 ret = PTR_ERR(id_priv->cm_id.ib);
2226 goto out;
2227 }
2228
2229 route = &id_priv->id.route;
2230 ret = cma_format_hdr(private_data, id_priv->id.ps, route);
2231 if (ret)
2232 goto out;
2233 req.private_data = private_data;
2234
2235 req.primary_path = &route->path_rec[0];
2236 if (route->num_paths == 2)
2237 req.alternate_path = &route->path_rec[1];
2238
2239 req.service_id = cma_get_service_id(id_priv->id.ps,
2240 &route->addr.dst_addr);
2241 req.qp_num = id_priv->qp_num;
2242 req.qp_type = IB_QPT_RC;
2243 req.starting_psn = id_priv->seq_num;
2244 req.responder_resources = conn_param->responder_resources;
2245 req.initiator_depth = conn_param->initiator_depth;
2246 req.flow_control = conn_param->flow_control;
2247 req.retry_count = conn_param->retry_count;
2248 req.rnr_retry_count = conn_param->rnr_retry_count;
2249 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2250 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2251 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2252 req.srq = id_priv->srq ? 1 : 0;
2253
2254 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2255 out:
2256 if (ret && !IS_ERR(id_priv->cm_id.ib)) {
2257 ib_destroy_cm_id(id_priv->cm_id.ib);
2258 id_priv->cm_id.ib = NULL;
2259 }
2260
2261 kfree(private_data);
2262 return ret;
2263 }
2264
2265 static int cma_connect_iw(struct rdma_id_private *id_priv,
2266 struct rdma_conn_param *conn_param)
2267 {
2268 struct iw_cm_id *cm_id;
2269 struct sockaddr_in* sin;
2270 int ret;
2271 struct iw_cm_conn_param iw_param;
2272
2273 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2274 if (IS_ERR(cm_id)) {
2275 ret = PTR_ERR(cm_id);
2276 goto out;
2277 }
2278
2279 id_priv->cm_id.iw = cm_id;
2280
2281 sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
2282 cm_id->local_addr = *sin;
2283
2284 sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
2285 cm_id->remote_addr = *sin;
2286
2287 ret = cma_modify_qp_rtr(id_priv);
2288 if (ret)
2289 goto out;
2290
2291 iw_param.ord = conn_param->initiator_depth;
2292 iw_param.ird = conn_param->responder_resources;
2293 iw_param.private_data = conn_param->private_data;
2294 iw_param.private_data_len = conn_param->private_data_len;
2295 if (id_priv->id.qp)
2296 iw_param.qpn = id_priv->qp_num;
2297 else
2298 iw_param.qpn = conn_param->qp_num;
2299 ret = iw_cm_connect(cm_id, &iw_param);
2300 out:
2301 if (ret && !IS_ERR(cm_id)) {
2302 iw_destroy_cm_id(cm_id);
2303 id_priv->cm_id.iw = NULL;
2304 }
2305 return ret;
2306 }
2307
2308 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2309 {
2310 struct rdma_id_private *id_priv;
2311 int ret;
2312
2313 id_priv = container_of(id, struct rdma_id_private, id);
2314 if (!cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_CONNECT))
2315 return -EINVAL;
2316
2317 if (!id->qp) {
2318 id_priv->qp_num = conn_param->qp_num;
2319 id_priv->srq = conn_param->srq;
2320 }
2321
2322 switch (rdma_node_get_transport(id->device->node_type)) {
2323 case RDMA_TRANSPORT_IB:
2324 if (cma_is_ud_ps(id->ps))
2325 ret = cma_resolve_ib_udp(id_priv, conn_param);
2326 else
2327 ret = cma_connect_ib(id_priv, conn_param);
2328 break;
2329 case RDMA_TRANSPORT_IWARP:
2330 ret = cma_connect_iw(id_priv, conn_param);
2331 break;
2332 default:
2333 ret = -ENOSYS;
2334 break;
2335 }
2336 if (ret)
2337 goto err;
2338
2339 return 0;
2340 err:
2341 cma_comp_exch(id_priv, CMA_CONNECT, CMA_ROUTE_RESOLVED);
2342 return ret;
2343 }
2344 EXPORT_SYMBOL(rdma_connect);
2345
2346 static int cma_accept_ib(struct rdma_id_private *id_priv,
2347 struct rdma_conn_param *conn_param)
2348 {
2349 struct ib_cm_rep_param rep;
2350 struct ib_qp_attr qp_attr;
2351 int qp_attr_mask, ret;
2352
2353 if (id_priv->id.qp) {
2354 ret = cma_modify_qp_rtr(id_priv);
2355 if (ret)
2356 goto out;
2357
2358 qp_attr.qp_state = IB_QPS_RTS;
2359 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, &qp_attr,
2360 &qp_attr_mask);
2361 if (ret)
2362 goto out;
2363
2364 qp_attr.max_rd_atomic = conn_param->initiator_depth;
2365 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
2366 if (ret)
2367 goto out;
2368 }
2369
2370 memset(&rep, 0, sizeof rep);
2371 rep.qp_num = id_priv->qp_num;
2372 rep.starting_psn = id_priv->seq_num;
2373 rep.private_data = conn_param->private_data;
2374 rep.private_data_len = conn_param->private_data_len;
2375 rep.responder_resources = conn_param->responder_resources;
2376 rep.initiator_depth = conn_param->initiator_depth;
2377 rep.failover_accepted = 0;
2378 rep.flow_control = conn_param->flow_control;
2379 rep.rnr_retry_count = conn_param->rnr_retry_count;
2380 rep.srq = id_priv->srq ? 1 : 0;
2381
2382 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2383 out:
2384 return ret;
2385 }
2386
2387 static int cma_accept_iw(struct rdma_id_private *id_priv,
2388 struct rdma_conn_param *conn_param)
2389 {
2390 struct iw_cm_conn_param iw_param;
2391 int ret;
2392
2393 ret = cma_modify_qp_rtr(id_priv);
2394 if (ret)
2395 return ret;
2396
2397 iw_param.ord = conn_param->initiator_depth;
2398 iw_param.ird = conn_param->responder_resources;
2399 iw_param.private_data = conn_param->private_data;
2400 iw_param.private_data_len = conn_param->private_data_len;
2401 if (id_priv->id.qp) {
2402 iw_param.qpn = id_priv->qp_num;
2403 } else
2404 iw_param.qpn = conn_param->qp_num;
2405
2406 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2407 }
2408
2409 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2410 enum ib_cm_sidr_status status,
2411 const void *private_data, int private_data_len)
2412 {
2413 struct ib_cm_sidr_rep_param rep;
2414
2415 memset(&rep, 0, sizeof rep);
2416 rep.status = status;
2417 if (status == IB_SIDR_SUCCESS) {
2418 rep.qp_num = id_priv->qp_num;
2419 rep.qkey = id_priv->qkey;
2420 }
2421 rep.private_data = private_data;
2422 rep.private_data_len = private_data_len;
2423
2424 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2425 }
2426
2427 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2428 {
2429 struct rdma_id_private *id_priv;
2430 int ret;
2431
2432 id_priv = container_of(id, struct rdma_id_private, id);
2433 if (!cma_comp(id_priv, CMA_CONNECT))
2434 return -EINVAL;
2435
2436 if (!id->qp && conn_param) {
2437 id_priv->qp_num = conn_param->qp_num;
2438 id_priv->srq = conn_param->srq;
2439 }
2440
2441 switch (rdma_node_get_transport(id->device->node_type)) {
2442 case RDMA_TRANSPORT_IB:
2443 if (cma_is_ud_ps(id->ps))
2444 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2445 conn_param->private_data,
2446 conn_param->private_data_len);
2447 else if (conn_param)
2448 ret = cma_accept_ib(id_priv, conn_param);
2449 else
2450 ret = cma_rep_recv(id_priv);
2451 break;
2452 case RDMA_TRANSPORT_IWARP:
2453 ret = cma_accept_iw(id_priv, conn_param);
2454 break;
2455 default:
2456 ret = -ENOSYS;
2457 break;
2458 }
2459
2460 if (ret)
2461 goto reject;
2462
2463 return 0;
2464 reject:
2465 cma_modify_qp_err(id_priv);
2466 rdma_reject(id, NULL, 0);
2467 return ret;
2468 }
2469 EXPORT_SYMBOL(rdma_accept);
2470
2471 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2472 {
2473 struct rdma_id_private *id_priv;
2474 int ret;
2475
2476 id_priv = container_of(id, struct rdma_id_private, id);
2477 if (!cma_has_cm_dev(id_priv))
2478 return -EINVAL;
2479
2480 switch (id->device->node_type) {
2481 case RDMA_NODE_IB_CA:
2482 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2483 break;
2484 default:
2485 ret = 0;
2486 break;
2487 }
2488 return ret;
2489 }
2490 EXPORT_SYMBOL(rdma_notify);
2491
2492 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2493 u8 private_data_len)
2494 {
2495 struct rdma_id_private *id_priv;
2496 int ret;
2497
2498 id_priv = container_of(id, struct rdma_id_private, id);
2499 if (!cma_has_cm_dev(id_priv))
2500 return -EINVAL;
2501
2502 switch (rdma_node_get_transport(id->device->node_type)) {
2503 case RDMA_TRANSPORT_IB:
2504 if (cma_is_ud_ps(id->ps))
2505 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
2506 private_data, private_data_len);
2507 else
2508 ret = ib_send_cm_rej(id_priv->cm_id.ib,
2509 IB_CM_REJ_CONSUMER_DEFINED, NULL,
2510 0, private_data, private_data_len);
2511 break;
2512 case RDMA_TRANSPORT_IWARP:
2513 ret = iw_cm_reject(id_priv->cm_id.iw,
2514 private_data, private_data_len);
2515 break;
2516 default:
2517 ret = -ENOSYS;
2518 break;
2519 }
2520 return ret;
2521 }
2522 EXPORT_SYMBOL(rdma_reject);
2523
2524 int rdma_disconnect(struct rdma_cm_id *id)
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 ret = cma_modify_qp_err(id_priv);
2536 if (ret)
2537 goto out;
2538 /* Initiate or respond to a disconnect. */
2539 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
2540 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
2541 break;
2542 case RDMA_TRANSPORT_IWARP:
2543 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
2544 break;
2545 default:
2546 ret = -EINVAL;
2547 break;
2548 }
2549 out:
2550 return ret;
2551 }
2552 EXPORT_SYMBOL(rdma_disconnect);
2553
2554 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
2555 {
2556 struct rdma_id_private *id_priv;
2557 struct cma_multicast *mc = multicast->context;
2558 struct rdma_cm_event event;
2559 int ret;
2560
2561 id_priv = mc->id_priv;
2562 if (cma_disable_remove(id_priv, CMA_ADDR_BOUND) &&
2563 cma_disable_remove(id_priv, CMA_ADDR_RESOLVED))
2564 return 0;
2565
2566 mutex_lock(&id_priv->qp_mutex);
2567 if (!status && id_priv->id.qp)
2568 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
2569 multicast->rec.mlid);
2570 mutex_unlock(&id_priv->qp_mutex);
2571
2572 memset(&event, 0, sizeof event);
2573 event.status = status;
2574 event.param.ud.private_data = mc->context;
2575 if (!status) {
2576 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
2577 ib_init_ah_from_mcmember(id_priv->id.device,
2578 id_priv->id.port_num, &multicast->rec,
2579 &event.param.ud.ah_attr);
2580 event.param.ud.qp_num = 0xFFFFFF;
2581 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
2582 } else
2583 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
2584
2585 ret = id_priv->id.event_handler(&id_priv->id, &event);
2586 if (ret) {
2587 cma_exch(id_priv, CMA_DESTROYING);
2588 cma_enable_remove(id_priv);
2589 rdma_destroy_id(&id_priv->id);
2590 return 0;
2591 }
2592
2593 cma_enable_remove(id_priv);
2594 return 0;
2595 }
2596
2597 static void cma_set_mgid(struct rdma_id_private *id_priv,
2598 struct sockaddr *addr, union ib_gid *mgid)
2599 {
2600 unsigned char mc_map[MAX_ADDR_LEN];
2601 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2602 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
2603 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
2604
2605 if (cma_any_addr(addr)) {
2606 memset(mgid, 0, sizeof *mgid);
2607 } else if ((addr->sa_family == AF_INET6) &&
2608 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFF10A01B) ==
2609 0xFF10A01B)) {
2610 /* IPv6 address is an SA assigned MGID. */
2611 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
2612 } else {
2613 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
2614 if (id_priv->id.ps == RDMA_PS_UDP)
2615 mc_map[7] = 0x01; /* Use RDMA CM signature */
2616 *mgid = *(union ib_gid *) (mc_map + 4);
2617 }
2618 }
2619
2620 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
2621 struct cma_multicast *mc)
2622 {
2623 struct ib_sa_mcmember_rec rec;
2624 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2625 ib_sa_comp_mask comp_mask;
2626 int ret;
2627
2628 ib_addr_get_mgid(dev_addr, &rec.mgid);
2629 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
2630 &rec.mgid, &rec);
2631 if (ret)
2632 return ret;
2633
2634 cma_set_mgid(id_priv, &mc->addr, &rec.mgid);
2635 if (id_priv->id.ps == RDMA_PS_UDP)
2636 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
2637 ib_addr_get_sgid(dev_addr, &rec.port_gid);
2638 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2639 rec.join_state = 1;
2640
2641 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
2642 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
2643 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
2644 IB_SA_MCMEMBER_REC_FLOW_LABEL |
2645 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
2646
2647 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
2648 id_priv->id.port_num, &rec,
2649 comp_mask, GFP_KERNEL,
2650 cma_ib_mc_handler, mc);
2651 if (IS_ERR(mc->multicast.ib))
2652 return PTR_ERR(mc->multicast.ib);
2653
2654 return 0;
2655 }
2656
2657 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
2658 void *context)
2659 {
2660 struct rdma_id_private *id_priv;
2661 struct cma_multicast *mc;
2662 int ret;
2663
2664 id_priv = container_of(id, struct rdma_id_private, id);
2665 if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
2666 !cma_comp(id_priv, CMA_ADDR_RESOLVED))
2667 return -EINVAL;
2668
2669 mc = kmalloc(sizeof *mc, GFP_KERNEL);
2670 if (!mc)
2671 return -ENOMEM;
2672
2673 memcpy(&mc->addr, addr, ip_addr_size(addr));
2674 mc->context = context;
2675 mc->id_priv = id_priv;
2676
2677 spin_lock(&id_priv->lock);
2678 list_add(&mc->list, &id_priv->mc_list);
2679 spin_unlock(&id_priv->lock);
2680
2681 switch (rdma_node_get_transport(id->device->node_type)) {
2682 case RDMA_TRANSPORT_IB:
2683 ret = cma_join_ib_multicast(id_priv, mc);
2684 break;
2685 default:
2686 ret = -ENOSYS;
2687 break;
2688 }
2689
2690 if (ret) {
2691 spin_lock_irq(&id_priv->lock);
2692 list_del(&mc->list);
2693 spin_unlock_irq(&id_priv->lock);
2694 kfree(mc);
2695 }
2696 return ret;
2697 }
2698 EXPORT_SYMBOL(rdma_join_multicast);
2699
2700 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
2701 {
2702 struct rdma_id_private *id_priv;
2703 struct cma_multicast *mc;
2704
2705 id_priv = container_of(id, struct rdma_id_private, id);
2706 spin_lock_irq(&id_priv->lock);
2707 list_for_each_entry(mc, &id_priv->mc_list, list) {
2708 if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
2709 list_del(&mc->list);
2710 spin_unlock_irq(&id_priv->lock);
2711
2712 if (id->qp)
2713 ib_detach_mcast(id->qp,
2714 &mc->multicast.ib->rec.mgid,
2715 mc->multicast.ib->rec.mlid);
2716 ib_sa_free_multicast(mc->multicast.ib);
2717 kfree(mc);
2718 return;
2719 }
2720 }
2721 spin_unlock_irq(&id_priv->lock);
2722 }
2723 EXPORT_SYMBOL(rdma_leave_multicast);
2724
2725 static void cma_add_one(struct ib_device *device)
2726 {
2727 struct cma_device *cma_dev;
2728 struct rdma_id_private *id_priv;
2729
2730 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
2731 if (!cma_dev)
2732 return;
2733
2734 cma_dev->device = device;
2735
2736 init_completion(&cma_dev->comp);
2737 atomic_set(&cma_dev->refcount, 1);
2738 INIT_LIST_HEAD(&cma_dev->id_list);
2739 ib_set_client_data(device, &cma_client, cma_dev);
2740
2741 mutex_lock(&lock);
2742 list_add_tail(&cma_dev->list, &dev_list);
2743 list_for_each_entry(id_priv, &listen_any_list, list)
2744 cma_listen_on_dev(id_priv, cma_dev);
2745 mutex_unlock(&lock);
2746 }
2747
2748 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
2749 {
2750 struct rdma_cm_event event;
2751 enum cma_state state;
2752
2753 /* Record that we want to remove the device */
2754 state = cma_exch(id_priv, CMA_DEVICE_REMOVAL);
2755 if (state == CMA_DESTROYING)
2756 return 0;
2757
2758 cma_cancel_operation(id_priv, state);
2759 wait_event(id_priv->wait_remove, !atomic_read(&id_priv->dev_remove));
2760
2761 /* Check for destruction from another callback. */
2762 if (!cma_comp(id_priv, CMA_DEVICE_REMOVAL))
2763 return 0;
2764
2765 memset(&event, 0, sizeof event);
2766 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
2767 return id_priv->id.event_handler(&id_priv->id, &event);
2768 }
2769
2770 static void cma_process_remove(struct cma_device *cma_dev)
2771 {
2772 struct rdma_id_private *id_priv;
2773 int ret;
2774
2775 mutex_lock(&lock);
2776 while (!list_empty(&cma_dev->id_list)) {
2777 id_priv = list_entry(cma_dev->id_list.next,
2778 struct rdma_id_private, list);
2779
2780 list_del(&id_priv->listen_list);
2781 list_del_init(&id_priv->list);
2782 atomic_inc(&id_priv->refcount);
2783 mutex_unlock(&lock);
2784
2785 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
2786 cma_deref_id(id_priv);
2787 if (ret)
2788 rdma_destroy_id(&id_priv->id);
2789
2790 mutex_lock(&lock);
2791 }
2792 mutex_unlock(&lock);
2793
2794 cma_deref_dev(cma_dev);
2795 wait_for_completion(&cma_dev->comp);
2796 }
2797
2798 static void cma_remove_one(struct ib_device *device)
2799 {
2800 struct cma_device *cma_dev;
2801
2802 cma_dev = ib_get_client_data(device, &cma_client);
2803 if (!cma_dev)
2804 return;
2805
2806 mutex_lock(&lock);
2807 list_del(&cma_dev->list);
2808 mutex_unlock(&lock);
2809
2810 cma_process_remove(cma_dev);
2811 kfree(cma_dev);
2812 }
2813
2814 static int cma_init(void)
2815 {
2816 int ret, low, high, remaining;
2817
2818 get_random_bytes(&next_port, sizeof next_port);
2819 inet_get_local_port_range(&low, &high);
2820 remaining = (high - low) + 1;
2821 next_port = ((unsigned int) next_port % remaining) + low;
2822
2823 cma_wq = create_singlethread_workqueue("rdma_cm");
2824 if (!cma_wq)
2825 return -ENOMEM;
2826
2827 ib_sa_register_client(&sa_client);
2828 rdma_addr_register_client(&addr_client);
2829
2830 ret = ib_register_client(&cma_client);
2831 if (ret)
2832 goto err;
2833 return 0;
2834
2835 err:
2836 rdma_addr_unregister_client(&addr_client);
2837 ib_sa_unregister_client(&sa_client);
2838 destroy_workqueue(cma_wq);
2839 return ret;
2840 }
2841
2842 static void cma_cleanup(void)
2843 {
2844 ib_unregister_client(&cma_client);
2845 rdma_addr_unregister_client(&addr_client);
2846 ib_sa_unregister_client(&sa_client);
2847 destroy_workqueue(cma_wq);
2848 idr_destroy(&sdp_ps);
2849 idr_destroy(&tcp_ps);
2850 idr_destroy(&udp_ps);
2851 idr_destroy(&ipoib_ps);
2852 }
2853
2854 module_init(cma_init);
2855 module_exit(cma_cleanup);
ubuntu-zesty-kernel mirror