]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/infiniband/core/cma.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[mirror_ubuntu-bionic-kernel.git] / drivers / infiniband / core / cma.c
1 /*
2 * Copyright (c) 2005 Voltaire Inc. All rights reserved.
3 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
4 * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
5 * Copyright (c) 2005-2006 Intel Corporation. All rights reserved.
6 *
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 * SOFTWARE.
34 */
35
36 #include <linux/completion.h>
37 #include <linux/in.h>
38 #include <linux/in6.h>
39 #include <linux/mutex.h>
40 #include <linux/random.h>
41 #include <linux/idr.h>
42 #include <linux/inetdevice.h>
43 #include <linux/slab.h>
44
45 #include <net/tcp.h>
46 #include <net/ipv6.h>
47
48 #include <rdma/rdma_cm.h>
49 #include <rdma/rdma_cm_ib.h>
50 #include <rdma/ib_cache.h>
51 #include <rdma/ib_cm.h>
52 #include <rdma/ib_sa.h>
53 #include <rdma/iw_cm.h>
54
55 MODULE_AUTHOR("Sean Hefty");
56 MODULE_DESCRIPTION("Generic RDMA CM Agent");
57 MODULE_LICENSE("Dual BSD/GPL");
58
59 #define CMA_CM_RESPONSE_TIMEOUT 20
60 #define CMA_MAX_CM_RETRIES 15
61 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
62
63 static void cma_add_one(struct ib_device *device);
64 static void cma_remove_one(struct ib_device *device);
65
66 static struct ib_client cma_client = {
67 .name = "cma",
68 .add = cma_add_one,
69 .remove = cma_remove_one
70 };
71
72 static struct ib_sa_client sa_client;
73 static struct rdma_addr_client addr_client;
74 static LIST_HEAD(dev_list);
75 static LIST_HEAD(listen_any_list);
76 static DEFINE_MUTEX(lock);
77 static struct workqueue_struct *cma_wq;
78 static DEFINE_IDR(sdp_ps);
79 static DEFINE_IDR(tcp_ps);
80 static DEFINE_IDR(udp_ps);
81 static DEFINE_IDR(ipoib_ps);
82 static int next_port;
83
84 struct cma_device {
85 struct list_head list;
86 struct ib_device *device;
87 struct completion comp;
88 atomic_t refcount;
89 struct list_head id_list;
90 };
91
92 enum cma_state {
93 CMA_IDLE,
94 CMA_ADDR_QUERY,
95 CMA_ADDR_RESOLVED,
96 CMA_ROUTE_QUERY,
97 CMA_ROUTE_RESOLVED,
98 CMA_CONNECT,
99 CMA_DISCONNECT,
100 CMA_ADDR_BOUND,
101 CMA_LISTEN,
102 CMA_DEVICE_REMOVAL,
103 CMA_DESTROYING
104 };
105
106 struct rdma_bind_list {
107 struct idr *ps;
108 struct hlist_head owners;
109 unsigned short port;
110 };
111
112 /*
113 * Device removal can occur at anytime, so we need extra handling to
114 * serialize notifying the user of device removal with other callbacks.
115 * We do this by disabling removal notification while a callback is in process,
116 * and reporting it after the callback completes.
117 */
118 struct rdma_id_private {
119 struct rdma_cm_id id;
120
121 struct rdma_bind_list *bind_list;
122 struct hlist_node node;
123 struct list_head list; /* listen_any_list or cma_device.list */
124 struct list_head listen_list; /* per device listens */
125 struct cma_device *cma_dev;
126 struct list_head mc_list;
127
128 int internal_id;
129 enum cma_state state;
130 spinlock_t lock;
131 struct mutex qp_mutex;
132
133 struct completion comp;
134 atomic_t refcount;
135 struct mutex handler_mutex;
136
137 int backlog;
138 int timeout_ms;
139 struct ib_sa_query *query;
140 int query_id;
141 union {
142 struct ib_cm_id *ib;
143 struct iw_cm_id *iw;
144 } cm_id;
145
146 u32 seq_num;
147 u32 qkey;
148 u32 qp_num;
149 u8 srq;
150 u8 tos;
151 };
152
153 struct cma_multicast {
154 struct rdma_id_private *id_priv;
155 union {
156 struct ib_sa_multicast *ib;
157 } multicast;
158 struct list_head list;
159 void *context;
160 struct sockaddr_storage addr;
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 rdma_id_private *id_priv)
302 {
303 struct ib_sa_mcmember_rec rec;
304 int ret = 0;
305
306 if (id_priv->qkey)
307 return 0;
308
309 switch (id_priv->id.ps) {
310 case RDMA_PS_UDP:
311 id_priv->qkey = RDMA_UDP_QKEY;
312 break;
313 case RDMA_PS_IPOIB:
314 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
315 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
316 id_priv->id.port_num, &rec.mgid,
317 &rec);
318 if (!ret)
319 id_priv->qkey = be32_to_cpu(rec.qkey);
320 break;
321 default:
322 break;
323 }
324 return ret;
325 }
326
327 static int cma_acquire_dev(struct rdma_id_private *id_priv)
328 {
329 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
330 struct cma_device *cma_dev;
331 union ib_gid gid;
332 int ret = -ENODEV;
333
334 rdma_addr_get_sgid(dev_addr, &gid);
335 list_for_each_entry(cma_dev, &dev_list, list) {
336 ret = ib_find_cached_gid(cma_dev->device, &gid,
337 &id_priv->id.port_num, NULL);
338 if (!ret) {
339 cma_attach_to_dev(id_priv, cma_dev);
340 break;
341 }
342 }
343 return ret;
344 }
345
346 static void cma_deref_id(struct rdma_id_private *id_priv)
347 {
348 if (atomic_dec_and_test(&id_priv->refcount))
349 complete(&id_priv->comp);
350 }
351
352 static int cma_disable_callback(struct rdma_id_private *id_priv,
353 enum cma_state state)
354 {
355 mutex_lock(&id_priv->handler_mutex);
356 if (id_priv->state != state) {
357 mutex_unlock(&id_priv->handler_mutex);
358 return -EINVAL;
359 }
360 return 0;
361 }
362
363 static int cma_has_cm_dev(struct rdma_id_private *id_priv)
364 {
365 return (id_priv->id.device && id_priv->cm_id.ib);
366 }
367
368 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
369 void *context, enum rdma_port_space ps)
370 {
371 struct rdma_id_private *id_priv;
372
373 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
374 if (!id_priv)
375 return ERR_PTR(-ENOMEM);
376
377 id_priv->state = CMA_IDLE;
378 id_priv->id.context = context;
379 id_priv->id.event_handler = event_handler;
380 id_priv->id.ps = ps;
381 spin_lock_init(&id_priv->lock);
382 mutex_init(&id_priv->qp_mutex);
383 init_completion(&id_priv->comp);
384 atomic_set(&id_priv->refcount, 1);
385 mutex_init(&id_priv->handler_mutex);
386 INIT_LIST_HEAD(&id_priv->listen_list);
387 INIT_LIST_HEAD(&id_priv->mc_list);
388 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
389
390 return &id_priv->id;
391 }
392 EXPORT_SYMBOL(rdma_create_id);
393
394 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
395 {
396 struct ib_qp_attr qp_attr;
397 int qp_attr_mask, ret;
398
399 qp_attr.qp_state = IB_QPS_INIT;
400 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
401 if (ret)
402 return ret;
403
404 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
405 if (ret)
406 return ret;
407
408 qp_attr.qp_state = IB_QPS_RTR;
409 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
410 if (ret)
411 return ret;
412
413 qp_attr.qp_state = IB_QPS_RTS;
414 qp_attr.sq_psn = 0;
415 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
416
417 return ret;
418 }
419
420 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
421 {
422 struct ib_qp_attr qp_attr;
423 int qp_attr_mask, ret;
424
425 qp_attr.qp_state = IB_QPS_INIT;
426 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
427 if (ret)
428 return ret;
429
430 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
431 }
432
433 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
434 struct ib_qp_init_attr *qp_init_attr)
435 {
436 struct rdma_id_private *id_priv;
437 struct ib_qp *qp;
438 int ret;
439
440 id_priv = container_of(id, struct rdma_id_private, id);
441 if (id->device != pd->device)
442 return -EINVAL;
443
444 qp = ib_create_qp(pd, qp_init_attr);
445 if (IS_ERR(qp))
446 return PTR_ERR(qp);
447
448 if (cma_is_ud_ps(id_priv->id.ps))
449 ret = cma_init_ud_qp(id_priv, qp);
450 else
451 ret = cma_init_conn_qp(id_priv, qp);
452 if (ret)
453 goto err;
454
455 id->qp = qp;
456 id_priv->qp_num = qp->qp_num;
457 id_priv->srq = (qp->srq != NULL);
458 return 0;
459 err:
460 ib_destroy_qp(qp);
461 return ret;
462 }
463 EXPORT_SYMBOL(rdma_create_qp);
464
465 void rdma_destroy_qp(struct rdma_cm_id *id)
466 {
467 struct rdma_id_private *id_priv;
468
469 id_priv = container_of(id, struct rdma_id_private, id);
470 mutex_lock(&id_priv->qp_mutex);
471 ib_destroy_qp(id_priv->id.qp);
472 id_priv->id.qp = NULL;
473 mutex_unlock(&id_priv->qp_mutex);
474 }
475 EXPORT_SYMBOL(rdma_destroy_qp);
476
477 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
478 struct rdma_conn_param *conn_param)
479 {
480 struct ib_qp_attr qp_attr;
481 int qp_attr_mask, ret;
482
483 mutex_lock(&id_priv->qp_mutex);
484 if (!id_priv->id.qp) {
485 ret = 0;
486 goto out;
487 }
488
489 /* Need to update QP attributes from default values. */
490 qp_attr.qp_state = IB_QPS_INIT;
491 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
492 if (ret)
493 goto out;
494
495 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
496 if (ret)
497 goto out;
498
499 qp_attr.qp_state = IB_QPS_RTR;
500 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
501 if (ret)
502 goto out;
503
504 if (conn_param)
505 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
506 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
507 out:
508 mutex_unlock(&id_priv->qp_mutex);
509 return ret;
510 }
511
512 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
513 struct rdma_conn_param *conn_param)
514 {
515 struct ib_qp_attr qp_attr;
516 int qp_attr_mask, ret;
517
518 mutex_lock(&id_priv->qp_mutex);
519 if (!id_priv->id.qp) {
520 ret = 0;
521 goto out;
522 }
523
524 qp_attr.qp_state = IB_QPS_RTS;
525 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
526 if (ret)
527 goto out;
528
529 if (conn_param)
530 qp_attr.max_rd_atomic = conn_param->initiator_depth;
531 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
532 out:
533 mutex_unlock(&id_priv->qp_mutex);
534 return ret;
535 }
536
537 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
538 {
539 struct ib_qp_attr qp_attr;
540 int ret;
541
542 mutex_lock(&id_priv->qp_mutex);
543 if (!id_priv->id.qp) {
544 ret = 0;
545 goto out;
546 }
547
548 qp_attr.qp_state = IB_QPS_ERR;
549 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
550 out:
551 mutex_unlock(&id_priv->qp_mutex);
552 return ret;
553 }
554
555 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
556 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
557 {
558 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
559 int ret;
560
561 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
562 ib_addr_get_pkey(dev_addr),
563 &qp_attr->pkey_index);
564 if (ret)
565 return ret;
566
567 qp_attr->port_num = id_priv->id.port_num;
568 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
569
570 if (cma_is_ud_ps(id_priv->id.ps)) {
571 ret = cma_set_qkey(id_priv);
572 if (ret)
573 return ret;
574
575 qp_attr->qkey = id_priv->qkey;
576 *qp_attr_mask |= IB_QP_QKEY;
577 } else {
578 qp_attr->qp_access_flags = 0;
579 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
580 }
581 return 0;
582 }
583
584 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
585 int *qp_attr_mask)
586 {
587 struct rdma_id_private *id_priv;
588 int ret = 0;
589
590 id_priv = container_of(id, struct rdma_id_private, id);
591 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
592 case RDMA_TRANSPORT_IB:
593 if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps))
594 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
595 else
596 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
597 qp_attr_mask);
598 if (qp_attr->qp_state == IB_QPS_RTR)
599 qp_attr->rq_psn = id_priv->seq_num;
600 break;
601 case RDMA_TRANSPORT_IWARP:
602 if (!id_priv->cm_id.iw) {
603 qp_attr->qp_access_flags = 0;
604 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
605 } else
606 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
607 qp_attr_mask);
608 break;
609 default:
610 ret = -ENOSYS;
611 break;
612 }
613
614 return ret;
615 }
616 EXPORT_SYMBOL(rdma_init_qp_attr);
617
618 static inline int cma_zero_addr(struct sockaddr *addr)
619 {
620 struct in6_addr *ip6;
621
622 if (addr->sa_family == AF_INET)
623 return ipv4_is_zeronet(
624 ((struct sockaddr_in *)addr)->sin_addr.s_addr);
625 else {
626 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
627 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
628 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
629 }
630 }
631
632 static inline int cma_loopback_addr(struct sockaddr *addr)
633 {
634 if (addr->sa_family == AF_INET)
635 return ipv4_is_loopback(
636 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
637 else
638 return ipv6_addr_loopback(
639 &((struct sockaddr_in6 *) addr)->sin6_addr);
640 }
641
642 static inline int cma_any_addr(struct sockaddr *addr)
643 {
644 return cma_zero_addr(addr) || cma_loopback_addr(addr);
645 }
646
647 static inline __be16 cma_port(struct sockaddr *addr)
648 {
649 if (addr->sa_family == AF_INET)
650 return ((struct sockaddr_in *) addr)->sin_port;
651 else
652 return ((struct sockaddr_in6 *) addr)->sin6_port;
653 }
654
655 static inline int cma_any_port(struct sockaddr *addr)
656 {
657 return !cma_port(addr);
658 }
659
660 static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
661 u8 *ip_ver, __be16 *port,
662 union cma_ip_addr **src, union cma_ip_addr **dst)
663 {
664 switch (ps) {
665 case RDMA_PS_SDP:
666 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
667 SDP_MAJ_VERSION)
668 return -EINVAL;
669
670 *ip_ver = sdp_get_ip_ver(hdr);
671 *port = ((struct sdp_hh *) hdr)->port;
672 *src = &((struct sdp_hh *) hdr)->src_addr;
673 *dst = &((struct sdp_hh *) hdr)->dst_addr;
674 break;
675 default:
676 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
677 return -EINVAL;
678
679 *ip_ver = cma_get_ip_ver(hdr);
680 *port = ((struct cma_hdr *) hdr)->port;
681 *src = &((struct cma_hdr *) hdr)->src_addr;
682 *dst = &((struct cma_hdr *) hdr)->dst_addr;
683 break;
684 }
685
686 if (*ip_ver != 4 && *ip_ver != 6)
687 return -EINVAL;
688 return 0;
689 }
690
691 static void cma_save_net_info(struct rdma_addr *addr,
692 struct rdma_addr *listen_addr,
693 u8 ip_ver, __be16 port,
694 union cma_ip_addr *src, union cma_ip_addr *dst)
695 {
696 struct sockaddr_in *listen4, *ip4;
697 struct sockaddr_in6 *listen6, *ip6;
698
699 switch (ip_ver) {
700 case 4:
701 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
702 ip4 = (struct sockaddr_in *) &addr->src_addr;
703 ip4->sin_family = listen4->sin_family;
704 ip4->sin_addr.s_addr = dst->ip4.addr;
705 ip4->sin_port = listen4->sin_port;
706
707 ip4 = (struct sockaddr_in *) &addr->dst_addr;
708 ip4->sin_family = listen4->sin_family;
709 ip4->sin_addr.s_addr = src->ip4.addr;
710 ip4->sin_port = port;
711 break;
712 case 6:
713 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
714 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
715 ip6->sin6_family = listen6->sin6_family;
716 ip6->sin6_addr = dst->ip6;
717 ip6->sin6_port = listen6->sin6_port;
718
719 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
720 ip6->sin6_family = listen6->sin6_family;
721 ip6->sin6_addr = src->ip6;
722 ip6->sin6_port = port;
723 break;
724 default:
725 break;
726 }
727 }
728
729 static inline int cma_user_data_offset(enum rdma_port_space ps)
730 {
731 switch (ps) {
732 case RDMA_PS_SDP:
733 return 0;
734 default:
735 return sizeof(struct cma_hdr);
736 }
737 }
738
739 static void cma_cancel_route(struct rdma_id_private *id_priv)
740 {
741 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
742 case RDMA_TRANSPORT_IB:
743 if (id_priv->query)
744 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
745 break;
746 default:
747 break;
748 }
749 }
750
751 static void cma_cancel_listens(struct rdma_id_private *id_priv)
752 {
753 struct rdma_id_private *dev_id_priv;
754
755 /*
756 * Remove from listen_any_list to prevent added devices from spawning
757 * additional listen requests.
758 */
759 mutex_lock(&lock);
760 list_del(&id_priv->list);
761
762 while (!list_empty(&id_priv->listen_list)) {
763 dev_id_priv = list_entry(id_priv->listen_list.next,
764 struct rdma_id_private, listen_list);
765 /* sync with device removal to avoid duplicate destruction */
766 list_del_init(&dev_id_priv->list);
767 list_del(&dev_id_priv->listen_list);
768 mutex_unlock(&lock);
769
770 rdma_destroy_id(&dev_id_priv->id);
771 mutex_lock(&lock);
772 }
773 mutex_unlock(&lock);
774 }
775
776 static void cma_cancel_operation(struct rdma_id_private *id_priv,
777 enum cma_state state)
778 {
779 switch (state) {
780 case CMA_ADDR_QUERY:
781 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
782 break;
783 case CMA_ROUTE_QUERY:
784 cma_cancel_route(id_priv);
785 break;
786 case CMA_LISTEN:
787 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr)
788 && !id_priv->cma_dev)
789 cma_cancel_listens(id_priv);
790 break;
791 default:
792 break;
793 }
794 }
795
796 static void cma_release_port(struct rdma_id_private *id_priv)
797 {
798 struct rdma_bind_list *bind_list = id_priv->bind_list;
799
800 if (!bind_list)
801 return;
802
803 mutex_lock(&lock);
804 hlist_del(&id_priv->node);
805 if (hlist_empty(&bind_list->owners)) {
806 idr_remove(bind_list->ps, bind_list->port);
807 kfree(bind_list);
808 }
809 mutex_unlock(&lock);
810 }
811
812 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
813 {
814 struct cma_multicast *mc;
815
816 while (!list_empty(&id_priv->mc_list)) {
817 mc = container_of(id_priv->mc_list.next,
818 struct cma_multicast, list);
819 list_del(&mc->list);
820 ib_sa_free_multicast(mc->multicast.ib);
821 kfree(mc);
822 }
823 }
824
825 void rdma_destroy_id(struct rdma_cm_id *id)
826 {
827 struct rdma_id_private *id_priv;
828 enum cma_state state;
829
830 id_priv = container_of(id, struct rdma_id_private, id);
831 state = cma_exch(id_priv, CMA_DESTROYING);
832 cma_cancel_operation(id_priv, state);
833
834 mutex_lock(&lock);
835 if (id_priv->cma_dev) {
836 mutex_unlock(&lock);
837 switch (rdma_node_get_transport(id->device->node_type)) {
838 case RDMA_TRANSPORT_IB:
839 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib))
840 ib_destroy_cm_id(id_priv->cm_id.ib);
841 break;
842 case RDMA_TRANSPORT_IWARP:
843 if (id_priv->cm_id.iw && !IS_ERR(id_priv->cm_id.iw))
844 iw_destroy_cm_id(id_priv->cm_id.iw);
845 break;
846 default:
847 break;
848 }
849 cma_leave_mc_groups(id_priv);
850 mutex_lock(&lock);
851 cma_detach_from_dev(id_priv);
852 }
853 mutex_unlock(&lock);
854
855 cma_release_port(id_priv);
856 cma_deref_id(id_priv);
857 wait_for_completion(&id_priv->comp);
858
859 if (id_priv->internal_id)
860 cma_deref_id(id_priv->id.context);
861
862 kfree(id_priv->id.route.path_rec);
863 kfree(id_priv);
864 }
865 EXPORT_SYMBOL(rdma_destroy_id);
866
867 static int cma_rep_recv(struct rdma_id_private *id_priv)
868 {
869 int ret;
870
871 ret = cma_modify_qp_rtr(id_priv, NULL);
872 if (ret)
873 goto reject;
874
875 ret = cma_modify_qp_rts(id_priv, NULL);
876 if (ret)
877 goto reject;
878
879 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
880 if (ret)
881 goto reject;
882
883 return 0;
884 reject:
885 cma_modify_qp_err(id_priv);
886 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
887 NULL, 0, NULL, 0);
888 return ret;
889 }
890
891 static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
892 {
893 if (id_priv->id.ps == RDMA_PS_SDP &&
894 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
895 SDP_MAJ_VERSION)
896 return -EINVAL;
897
898 return 0;
899 }
900
901 static void cma_set_rep_event_data(struct rdma_cm_event *event,
902 struct ib_cm_rep_event_param *rep_data,
903 void *private_data)
904 {
905 event->param.conn.private_data = private_data;
906 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
907 event->param.conn.responder_resources = rep_data->responder_resources;
908 event->param.conn.initiator_depth = rep_data->initiator_depth;
909 event->param.conn.flow_control = rep_data->flow_control;
910 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
911 event->param.conn.srq = rep_data->srq;
912 event->param.conn.qp_num = rep_data->remote_qpn;
913 }
914
915 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
916 {
917 struct rdma_id_private *id_priv = cm_id->context;
918 struct rdma_cm_event event;
919 int ret = 0;
920
921 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
922 cma_disable_callback(id_priv, CMA_CONNECT)) ||
923 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
924 cma_disable_callback(id_priv, CMA_DISCONNECT)))
925 return 0;
926
927 memset(&event, 0, sizeof event);
928 switch (ib_event->event) {
929 case IB_CM_REQ_ERROR:
930 case IB_CM_REP_ERROR:
931 event.event = RDMA_CM_EVENT_UNREACHABLE;
932 event.status = -ETIMEDOUT;
933 break;
934 case IB_CM_REP_RECEIVED:
935 event.status = cma_verify_rep(id_priv, ib_event->private_data);
936 if (event.status)
937 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
938 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
939 event.status = cma_rep_recv(id_priv);
940 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
941 RDMA_CM_EVENT_ESTABLISHED;
942 } else
943 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
944 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
945 ib_event->private_data);
946 break;
947 case IB_CM_RTU_RECEIVED:
948 case IB_CM_USER_ESTABLISHED:
949 event.event = RDMA_CM_EVENT_ESTABLISHED;
950 break;
951 case IB_CM_DREQ_ERROR:
952 event.status = -ETIMEDOUT; /* fall through */
953 case IB_CM_DREQ_RECEIVED:
954 case IB_CM_DREP_RECEIVED:
955 if (!cma_comp_exch(id_priv, CMA_CONNECT, CMA_DISCONNECT))
956 goto out;
957 event.event = RDMA_CM_EVENT_DISCONNECTED;
958 break;
959 case IB_CM_TIMEWAIT_EXIT:
960 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
961 break;
962 case IB_CM_MRA_RECEIVED:
963 /* ignore event */
964 goto out;
965 case IB_CM_REJ_RECEIVED:
966 cma_modify_qp_err(id_priv);
967 event.status = ib_event->param.rej_rcvd.reason;
968 event.event = RDMA_CM_EVENT_REJECTED;
969 event.param.conn.private_data = ib_event->private_data;
970 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
971 break;
972 default:
973 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
974 ib_event->event);
975 goto out;
976 }
977
978 ret = id_priv->id.event_handler(&id_priv->id, &event);
979 if (ret) {
980 /* Destroy the CM ID by returning a non-zero value. */
981 id_priv->cm_id.ib = NULL;
982 cma_exch(id_priv, CMA_DESTROYING);
983 mutex_unlock(&id_priv->handler_mutex);
984 rdma_destroy_id(&id_priv->id);
985 return ret;
986 }
987 out:
988 mutex_unlock(&id_priv->handler_mutex);
989 return ret;
990 }
991
992 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
993 struct ib_cm_event *ib_event)
994 {
995 struct rdma_id_private *id_priv;
996 struct rdma_cm_id *id;
997 struct rdma_route *rt;
998 union cma_ip_addr *src, *dst;
999 __be16 port;
1000 u8 ip_ver;
1001 int ret;
1002
1003 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1004 &ip_ver, &port, &src, &dst))
1005 goto err;
1006
1007 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1008 listen_id->ps);
1009 if (IS_ERR(id))
1010 goto err;
1011
1012 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1013 ip_ver, port, src, dst);
1014
1015 rt = &id->route;
1016 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1017 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1018 GFP_KERNEL);
1019 if (!rt->path_rec)
1020 goto destroy_id;
1021
1022 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1023 if (rt->num_paths == 2)
1024 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1025
1026 if (cma_any_addr((struct sockaddr *) &rt->addr.src_addr)) {
1027 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1028 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1029 ib_addr_set_pkey(&rt->addr.dev_addr, rt->path_rec[0].pkey);
1030 } else {
1031 ret = rdma_translate_ip((struct sockaddr *) &rt->addr.src_addr,
1032 &rt->addr.dev_addr);
1033 if (ret)
1034 goto destroy_id;
1035 }
1036 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1037
1038 id_priv = container_of(id, struct rdma_id_private, id);
1039 id_priv->state = CMA_CONNECT;
1040 return id_priv;
1041
1042 destroy_id:
1043 rdma_destroy_id(id);
1044 err:
1045 return NULL;
1046 }
1047
1048 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1049 struct ib_cm_event *ib_event)
1050 {
1051 struct rdma_id_private *id_priv;
1052 struct rdma_cm_id *id;
1053 union cma_ip_addr *src, *dst;
1054 __be16 port;
1055 u8 ip_ver;
1056 int ret;
1057
1058 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1059 listen_id->ps);
1060 if (IS_ERR(id))
1061 return NULL;
1062
1063
1064 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1065 &ip_ver, &port, &src, &dst))
1066 goto err;
1067
1068 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1069 ip_ver, port, src, dst);
1070
1071 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1072 ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr,
1073 &id->route.addr.dev_addr);
1074 if (ret)
1075 goto err;
1076 }
1077
1078 id_priv = container_of(id, struct rdma_id_private, id);
1079 id_priv->state = CMA_CONNECT;
1080 return id_priv;
1081 err:
1082 rdma_destroy_id(id);
1083 return NULL;
1084 }
1085
1086 static void cma_set_req_event_data(struct rdma_cm_event *event,
1087 struct ib_cm_req_event_param *req_data,
1088 void *private_data, int offset)
1089 {
1090 event->param.conn.private_data = private_data + offset;
1091 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1092 event->param.conn.responder_resources = req_data->responder_resources;
1093 event->param.conn.initiator_depth = req_data->initiator_depth;
1094 event->param.conn.flow_control = req_data->flow_control;
1095 event->param.conn.retry_count = req_data->retry_count;
1096 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1097 event->param.conn.srq = req_data->srq;
1098 event->param.conn.qp_num = req_data->remote_qpn;
1099 }
1100
1101 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1102 {
1103 struct rdma_id_private *listen_id, *conn_id;
1104 struct rdma_cm_event event;
1105 int offset, ret;
1106
1107 listen_id = cm_id->context;
1108 if (cma_disable_callback(listen_id, CMA_LISTEN))
1109 return -ECONNABORTED;
1110
1111 memset(&event, 0, sizeof event);
1112 offset = cma_user_data_offset(listen_id->id.ps);
1113 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1114 if (cma_is_ud_ps(listen_id->id.ps)) {
1115 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1116 event.param.ud.private_data = ib_event->private_data + offset;
1117 event.param.ud.private_data_len =
1118 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1119 } else {
1120 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1121 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1122 ib_event->private_data, offset);
1123 }
1124 if (!conn_id) {
1125 ret = -ENOMEM;
1126 goto out;
1127 }
1128
1129 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1130 mutex_lock(&lock);
1131 ret = cma_acquire_dev(conn_id);
1132 mutex_unlock(&lock);
1133 if (ret)
1134 goto release_conn_id;
1135
1136 conn_id->cm_id.ib = cm_id;
1137 cm_id->context = conn_id;
1138 cm_id->cm_handler = cma_ib_handler;
1139
1140 ret = conn_id->id.event_handler(&conn_id->id, &event);
1141 if (!ret) {
1142 /*
1143 * Acquire mutex to prevent user executing rdma_destroy_id()
1144 * while we're accessing the cm_id.
1145 */
1146 mutex_lock(&lock);
1147 if (cma_comp(conn_id, CMA_CONNECT) &&
1148 !cma_is_ud_ps(conn_id->id.ps))
1149 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1150 mutex_unlock(&lock);
1151 mutex_unlock(&conn_id->handler_mutex);
1152 goto out;
1153 }
1154
1155 /* Destroy the CM ID by returning a non-zero value. */
1156 conn_id->cm_id.ib = NULL;
1157
1158 release_conn_id:
1159 cma_exch(conn_id, CMA_DESTROYING);
1160 mutex_unlock(&conn_id->handler_mutex);
1161 rdma_destroy_id(&conn_id->id);
1162
1163 out:
1164 mutex_unlock(&listen_id->handler_mutex);
1165 return ret;
1166 }
1167
1168 static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1169 {
1170 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
1171 }
1172
1173 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1174 struct ib_cm_compare_data *compare)
1175 {
1176 struct cma_hdr *cma_data, *cma_mask;
1177 struct sdp_hh *sdp_data, *sdp_mask;
1178 __be32 ip4_addr;
1179 struct in6_addr ip6_addr;
1180
1181 memset(compare, 0, sizeof *compare);
1182 cma_data = (void *) compare->data;
1183 cma_mask = (void *) compare->mask;
1184 sdp_data = (void *) compare->data;
1185 sdp_mask = (void *) compare->mask;
1186
1187 switch (addr->sa_family) {
1188 case AF_INET:
1189 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1190 if (ps == RDMA_PS_SDP) {
1191 sdp_set_ip_ver(sdp_data, 4);
1192 sdp_set_ip_ver(sdp_mask, 0xF);
1193 sdp_data->dst_addr.ip4.addr = ip4_addr;
1194 sdp_mask->dst_addr.ip4.addr = htonl(~0);
1195 } else {
1196 cma_set_ip_ver(cma_data, 4);
1197 cma_set_ip_ver(cma_mask, 0xF);
1198 cma_data->dst_addr.ip4.addr = ip4_addr;
1199 cma_mask->dst_addr.ip4.addr = htonl(~0);
1200 }
1201 break;
1202 case AF_INET6:
1203 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1204 if (ps == RDMA_PS_SDP) {
1205 sdp_set_ip_ver(sdp_data, 6);
1206 sdp_set_ip_ver(sdp_mask, 0xF);
1207 sdp_data->dst_addr.ip6 = ip6_addr;
1208 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1209 sizeof sdp_mask->dst_addr.ip6);
1210 } else {
1211 cma_set_ip_ver(cma_data, 6);
1212 cma_set_ip_ver(cma_mask, 0xF);
1213 cma_data->dst_addr.ip6 = ip6_addr;
1214 memset(&cma_mask->dst_addr.ip6, 0xFF,
1215 sizeof cma_mask->dst_addr.ip6);
1216 }
1217 break;
1218 default:
1219 break;
1220 }
1221 }
1222
1223 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1224 {
1225 struct rdma_id_private *id_priv = iw_id->context;
1226 struct rdma_cm_event event;
1227 struct sockaddr_in *sin;
1228 int ret = 0;
1229
1230 if (cma_disable_callback(id_priv, CMA_CONNECT))
1231 return 0;
1232
1233 memset(&event, 0, sizeof event);
1234 switch (iw_event->event) {
1235 case IW_CM_EVENT_CLOSE:
1236 event.event = RDMA_CM_EVENT_DISCONNECTED;
1237 break;
1238 case IW_CM_EVENT_CONNECT_REPLY:
1239 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1240 *sin = iw_event->local_addr;
1241 sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
1242 *sin = iw_event->remote_addr;
1243 switch (iw_event->status) {
1244 case 0:
1245 event.event = RDMA_CM_EVENT_ESTABLISHED;
1246 break;
1247 case -ECONNRESET:
1248 case -ECONNREFUSED:
1249 event.event = RDMA_CM_EVENT_REJECTED;
1250 break;
1251 case -ETIMEDOUT:
1252 event.event = RDMA_CM_EVENT_UNREACHABLE;
1253 break;
1254 default:
1255 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1256 break;
1257 }
1258 break;
1259 case IW_CM_EVENT_ESTABLISHED:
1260 event.event = RDMA_CM_EVENT_ESTABLISHED;
1261 break;
1262 default:
1263 BUG_ON(1);
1264 }
1265
1266 event.status = iw_event->status;
1267 event.param.conn.private_data = iw_event->private_data;
1268 event.param.conn.private_data_len = iw_event->private_data_len;
1269 ret = id_priv->id.event_handler(&id_priv->id, &event);
1270 if (ret) {
1271 /* Destroy the CM ID by returning a non-zero value. */
1272 id_priv->cm_id.iw = NULL;
1273 cma_exch(id_priv, CMA_DESTROYING);
1274 mutex_unlock(&id_priv->handler_mutex);
1275 rdma_destroy_id(&id_priv->id);
1276 return ret;
1277 }
1278
1279 mutex_unlock(&id_priv->handler_mutex);
1280 return ret;
1281 }
1282
1283 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1284 struct iw_cm_event *iw_event)
1285 {
1286 struct rdma_cm_id *new_cm_id;
1287 struct rdma_id_private *listen_id, *conn_id;
1288 struct sockaddr_in *sin;
1289 struct net_device *dev = NULL;
1290 struct rdma_cm_event event;
1291 int ret;
1292 struct ib_device_attr attr;
1293
1294 listen_id = cm_id->context;
1295 if (cma_disable_callback(listen_id, CMA_LISTEN))
1296 return -ECONNABORTED;
1297
1298 /* Create a new RDMA id for the new IW CM ID */
1299 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1300 listen_id->id.context,
1301 RDMA_PS_TCP);
1302 if (IS_ERR(new_cm_id)) {
1303 ret = -ENOMEM;
1304 goto out;
1305 }
1306 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1307 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1308 conn_id->state = CMA_CONNECT;
1309
1310 dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
1311 if (!dev) {
1312 ret = -EADDRNOTAVAIL;
1313 mutex_unlock(&conn_id->handler_mutex);
1314 rdma_destroy_id(new_cm_id);
1315 goto out;
1316 }
1317 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1318 if (ret) {
1319 mutex_unlock(&conn_id->handler_mutex);
1320 rdma_destroy_id(new_cm_id);
1321 goto out;
1322 }
1323
1324 mutex_lock(&lock);
1325 ret = cma_acquire_dev(conn_id);
1326 mutex_unlock(&lock);
1327 if (ret) {
1328 mutex_unlock(&conn_id->handler_mutex);
1329 rdma_destroy_id(new_cm_id);
1330 goto out;
1331 }
1332
1333 conn_id->cm_id.iw = cm_id;
1334 cm_id->context = conn_id;
1335 cm_id->cm_handler = cma_iw_handler;
1336
1337 sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
1338 *sin = iw_event->local_addr;
1339 sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
1340 *sin = iw_event->remote_addr;
1341
1342 ret = ib_query_device(conn_id->id.device, &attr);
1343 if (ret) {
1344 mutex_unlock(&conn_id->handler_mutex);
1345 rdma_destroy_id(new_cm_id);
1346 goto out;
1347 }
1348
1349 memset(&event, 0, sizeof event);
1350 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1351 event.param.conn.private_data = iw_event->private_data;
1352 event.param.conn.private_data_len = iw_event->private_data_len;
1353 event.param.conn.initiator_depth = attr.max_qp_init_rd_atom;
1354 event.param.conn.responder_resources = attr.max_qp_rd_atom;
1355 ret = conn_id->id.event_handler(&conn_id->id, &event);
1356 if (ret) {
1357 /* User wants to destroy the CM ID */
1358 conn_id->cm_id.iw = NULL;
1359 cma_exch(conn_id, CMA_DESTROYING);
1360 mutex_unlock(&conn_id->handler_mutex);
1361 rdma_destroy_id(&conn_id->id);
1362 goto out;
1363 }
1364
1365 mutex_unlock(&conn_id->handler_mutex);
1366
1367 out:
1368 if (dev)
1369 dev_put(dev);
1370 mutex_unlock(&listen_id->handler_mutex);
1371 return ret;
1372 }
1373
1374 static int cma_ib_listen(struct rdma_id_private *id_priv)
1375 {
1376 struct ib_cm_compare_data compare_data;
1377 struct sockaddr *addr;
1378 __be64 svc_id;
1379 int ret;
1380
1381 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_req_handler,
1382 id_priv);
1383 if (IS_ERR(id_priv->cm_id.ib))
1384 return PTR_ERR(id_priv->cm_id.ib);
1385
1386 addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
1387 svc_id = cma_get_service_id(id_priv->id.ps, addr);
1388 if (cma_any_addr(addr))
1389 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1390 else {
1391 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1392 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1393 }
1394
1395 if (ret) {
1396 ib_destroy_cm_id(id_priv->cm_id.ib);
1397 id_priv->cm_id.ib = NULL;
1398 }
1399
1400 return ret;
1401 }
1402
1403 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1404 {
1405 int ret;
1406 struct sockaddr_in *sin;
1407
1408 id_priv->cm_id.iw = iw_create_cm_id(id_priv->id.device,
1409 iw_conn_req_handler,
1410 id_priv);
1411 if (IS_ERR(id_priv->cm_id.iw))
1412 return PTR_ERR(id_priv->cm_id.iw);
1413
1414 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1415 id_priv->cm_id.iw->local_addr = *sin;
1416
1417 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1418
1419 if (ret) {
1420 iw_destroy_cm_id(id_priv->cm_id.iw);
1421 id_priv->cm_id.iw = NULL;
1422 }
1423
1424 return ret;
1425 }
1426
1427 static int cma_listen_handler(struct rdma_cm_id *id,
1428 struct rdma_cm_event *event)
1429 {
1430 struct rdma_id_private *id_priv = id->context;
1431
1432 id->context = id_priv->id.context;
1433 id->event_handler = id_priv->id.event_handler;
1434 return id_priv->id.event_handler(id, event);
1435 }
1436
1437 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1438 struct cma_device *cma_dev)
1439 {
1440 struct rdma_id_private *dev_id_priv;
1441 struct rdma_cm_id *id;
1442 int ret;
1443
1444 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps);
1445 if (IS_ERR(id))
1446 return;
1447
1448 dev_id_priv = container_of(id, struct rdma_id_private, id);
1449
1450 dev_id_priv->state = CMA_ADDR_BOUND;
1451 memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
1452 ip_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
1453
1454 cma_attach_to_dev(dev_id_priv, cma_dev);
1455 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1456 atomic_inc(&id_priv->refcount);
1457 dev_id_priv->internal_id = 1;
1458
1459 ret = rdma_listen(id, id_priv->backlog);
1460 if (ret)
1461 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1462 "listening on device %s\n", ret, cma_dev->device->name);
1463 }
1464
1465 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1466 {
1467 struct cma_device *cma_dev;
1468
1469 mutex_lock(&lock);
1470 list_add_tail(&id_priv->list, &listen_any_list);
1471 list_for_each_entry(cma_dev, &dev_list, list)
1472 cma_listen_on_dev(id_priv, cma_dev);
1473 mutex_unlock(&lock);
1474 }
1475
1476 int rdma_listen(struct rdma_cm_id *id, int backlog)
1477 {
1478 struct rdma_id_private *id_priv;
1479 int ret;
1480
1481 id_priv = container_of(id, struct rdma_id_private, id);
1482 if (id_priv->state == CMA_IDLE) {
1483 ((struct sockaddr *) &id->route.addr.src_addr)->sa_family = AF_INET;
1484 ret = rdma_bind_addr(id, (struct sockaddr *) &id->route.addr.src_addr);
1485 if (ret)
1486 return ret;
1487 }
1488
1489 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
1490 return -EINVAL;
1491
1492 id_priv->backlog = backlog;
1493 if (id->device) {
1494 switch (rdma_node_get_transport(id->device->node_type)) {
1495 case RDMA_TRANSPORT_IB:
1496 ret = cma_ib_listen(id_priv);
1497 if (ret)
1498 goto err;
1499 break;
1500 case RDMA_TRANSPORT_IWARP:
1501 ret = cma_iw_listen(id_priv, backlog);
1502 if (ret)
1503 goto err;
1504 break;
1505 default:
1506 ret = -ENOSYS;
1507 goto err;
1508 }
1509 } else
1510 cma_listen_on_all(id_priv);
1511
1512 return 0;
1513 err:
1514 id_priv->backlog = 0;
1515 cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
1516 return ret;
1517 }
1518 EXPORT_SYMBOL(rdma_listen);
1519
1520 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1521 {
1522 struct rdma_id_private *id_priv;
1523
1524 id_priv = container_of(id, struct rdma_id_private, id);
1525 id_priv->tos = (u8) tos;
1526 }
1527 EXPORT_SYMBOL(rdma_set_service_type);
1528
1529 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1530 void *context)
1531 {
1532 struct cma_work *work = context;
1533 struct rdma_route *route;
1534
1535 route = &work->id->id.route;
1536
1537 if (!status) {
1538 route->num_paths = 1;
1539 *route->path_rec = *path_rec;
1540 } else {
1541 work->old_state = CMA_ROUTE_QUERY;
1542 work->new_state = CMA_ADDR_RESOLVED;
1543 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1544 work->event.status = status;
1545 }
1546
1547 queue_work(cma_wq, &work->work);
1548 }
1549
1550 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1551 struct cma_work *work)
1552 {
1553 struct rdma_addr *addr = &id_priv->id.route.addr;
1554 struct ib_sa_path_rec path_rec;
1555 ib_sa_comp_mask comp_mask;
1556 struct sockaddr_in6 *sin6;
1557
1558 memset(&path_rec, 0, sizeof path_rec);
1559 rdma_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
1560 rdma_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
1561 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
1562 path_rec.numb_path = 1;
1563 path_rec.reversible = 1;
1564 path_rec.service_id = cma_get_service_id(id_priv->id.ps,
1565 (struct sockaddr *) &addr->dst_addr);
1566
1567 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1568 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1569 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1570
1571 if (addr->src_addr.ss_family == AF_INET) {
1572 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1573 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1574 } else {
1575 sin6 = (struct sockaddr_in6 *) &addr->src_addr;
1576 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1577 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1578 }
1579
1580 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1581 id_priv->id.port_num, &path_rec,
1582 comp_mask, timeout_ms,
1583 GFP_KERNEL, cma_query_handler,
1584 work, &id_priv->query);
1585
1586 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1587 }
1588
1589 static void cma_work_handler(struct work_struct *_work)
1590 {
1591 struct cma_work *work = container_of(_work, struct cma_work, work);
1592 struct rdma_id_private *id_priv = work->id;
1593 int destroy = 0;
1594
1595 mutex_lock(&id_priv->handler_mutex);
1596 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1597 goto out;
1598
1599 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1600 cma_exch(id_priv, CMA_DESTROYING);
1601 destroy = 1;
1602 }
1603 out:
1604 mutex_unlock(&id_priv->handler_mutex);
1605 cma_deref_id(id_priv);
1606 if (destroy)
1607 rdma_destroy_id(&id_priv->id);
1608 kfree(work);
1609 }
1610
1611 static void cma_ndev_work_handler(struct work_struct *_work)
1612 {
1613 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1614 struct rdma_id_private *id_priv = work->id;
1615 int destroy = 0;
1616
1617 mutex_lock(&id_priv->handler_mutex);
1618 if (id_priv->state == CMA_DESTROYING ||
1619 id_priv->state == CMA_DEVICE_REMOVAL)
1620 goto out;
1621
1622 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1623 cma_exch(id_priv, CMA_DESTROYING);
1624 destroy = 1;
1625 }
1626
1627 out:
1628 mutex_unlock(&id_priv->handler_mutex);
1629 cma_deref_id(id_priv);
1630 if (destroy)
1631 rdma_destroy_id(&id_priv->id);
1632 kfree(work);
1633 }
1634
1635 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1636 {
1637 struct rdma_route *route = &id_priv->id.route;
1638 struct cma_work *work;
1639 int ret;
1640
1641 work = kzalloc(sizeof *work, GFP_KERNEL);
1642 if (!work)
1643 return -ENOMEM;
1644
1645 work->id = id_priv;
1646 INIT_WORK(&work->work, cma_work_handler);
1647 work->old_state = CMA_ROUTE_QUERY;
1648 work->new_state = CMA_ROUTE_RESOLVED;
1649 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1650
1651 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1652 if (!route->path_rec) {
1653 ret = -ENOMEM;
1654 goto err1;
1655 }
1656
1657 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1658 if (ret)
1659 goto err2;
1660
1661 return 0;
1662 err2:
1663 kfree(route->path_rec);
1664 route->path_rec = NULL;
1665 err1:
1666 kfree(work);
1667 return ret;
1668 }
1669
1670 int rdma_set_ib_paths(struct rdma_cm_id *id,
1671 struct ib_sa_path_rec *path_rec, int num_paths)
1672 {
1673 struct rdma_id_private *id_priv;
1674 int ret;
1675
1676 id_priv = container_of(id, struct rdma_id_private, id);
1677 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_RESOLVED))
1678 return -EINVAL;
1679
1680 id->route.path_rec = kmalloc(sizeof *path_rec * num_paths, GFP_KERNEL);
1681 if (!id->route.path_rec) {
1682 ret = -ENOMEM;
1683 goto err;
1684 }
1685
1686 memcpy(id->route.path_rec, path_rec, sizeof *path_rec * num_paths);
1687 return 0;
1688 err:
1689 cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_ADDR_RESOLVED);
1690 return ret;
1691 }
1692 EXPORT_SYMBOL(rdma_set_ib_paths);
1693
1694 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1695 {
1696 struct cma_work *work;
1697
1698 work = kzalloc(sizeof *work, GFP_KERNEL);
1699 if (!work)
1700 return -ENOMEM;
1701
1702 work->id = id_priv;
1703 INIT_WORK(&work->work, cma_work_handler);
1704 work->old_state = CMA_ROUTE_QUERY;
1705 work->new_state = CMA_ROUTE_RESOLVED;
1706 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1707 queue_work(cma_wq, &work->work);
1708 return 0;
1709 }
1710
1711 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1712 {
1713 struct rdma_id_private *id_priv;
1714 int ret;
1715
1716 id_priv = container_of(id, struct rdma_id_private, id);
1717 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_QUERY))
1718 return -EINVAL;
1719
1720 atomic_inc(&id_priv->refcount);
1721 switch (rdma_node_get_transport(id->device->node_type)) {
1722 case RDMA_TRANSPORT_IB:
1723 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1724 break;
1725 case RDMA_TRANSPORT_IWARP:
1726 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1727 break;
1728 default:
1729 ret = -ENOSYS;
1730 break;
1731 }
1732 if (ret)
1733 goto err;
1734
1735 return 0;
1736 err:
1737 cma_comp_exch(id_priv, CMA_ROUTE_QUERY, CMA_ADDR_RESOLVED);
1738 cma_deref_id(id_priv);
1739 return ret;
1740 }
1741 EXPORT_SYMBOL(rdma_resolve_route);
1742
1743 static int cma_bind_loopback(struct rdma_id_private *id_priv)
1744 {
1745 struct cma_device *cma_dev;
1746 struct ib_port_attr port_attr;
1747 union ib_gid gid;
1748 u16 pkey;
1749 int ret;
1750 u8 p;
1751
1752 mutex_lock(&lock);
1753 if (list_empty(&dev_list)) {
1754 ret = -ENODEV;
1755 goto out;
1756 }
1757 list_for_each_entry(cma_dev, &dev_list, list)
1758 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
1759 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
1760 port_attr.state == IB_PORT_ACTIVE)
1761 goto port_found;
1762
1763 p = 1;
1764 cma_dev = list_entry(dev_list.next, struct cma_device, list);
1765
1766 port_found:
1767 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
1768 if (ret)
1769 goto out;
1770
1771 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
1772 if (ret)
1773 goto out;
1774
1775 id_priv->id.route.addr.dev_addr.dev_type =
1776 (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB) ?
1777 ARPHRD_INFINIBAND : ARPHRD_ETHER;
1778
1779 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1780 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1781 id_priv->id.port_num = p;
1782 cma_attach_to_dev(id_priv, cma_dev);
1783 out:
1784 mutex_unlock(&lock);
1785 return ret;
1786 }
1787
1788 static void addr_handler(int status, struct sockaddr *src_addr,
1789 struct rdma_dev_addr *dev_addr, void *context)
1790 {
1791 struct rdma_id_private *id_priv = context;
1792 struct rdma_cm_event event;
1793
1794 memset(&event, 0, sizeof event);
1795 mutex_lock(&id_priv->handler_mutex);
1796
1797 /*
1798 * Grab mutex to block rdma_destroy_id() from removing the device while
1799 * we're trying to acquire it.
1800 */
1801 mutex_lock(&lock);
1802 if (!cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_RESOLVED)) {
1803 mutex_unlock(&lock);
1804 goto out;
1805 }
1806
1807 if (!status && !id_priv->cma_dev)
1808 status = cma_acquire_dev(id_priv);
1809 mutex_unlock(&lock);
1810
1811 if (status) {
1812 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ADDR_BOUND))
1813 goto out;
1814 event.event = RDMA_CM_EVENT_ADDR_ERROR;
1815 event.status = status;
1816 } else {
1817 memcpy(&id_priv->id.route.addr.src_addr, src_addr,
1818 ip_addr_size(src_addr));
1819 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1820 }
1821
1822 if (id_priv->id.event_handler(&id_priv->id, &event)) {
1823 cma_exch(id_priv, CMA_DESTROYING);
1824 mutex_unlock(&id_priv->handler_mutex);
1825 cma_deref_id(id_priv);
1826 rdma_destroy_id(&id_priv->id);
1827 return;
1828 }
1829 out:
1830 mutex_unlock(&id_priv->handler_mutex);
1831 cma_deref_id(id_priv);
1832 }
1833
1834 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1835 {
1836 struct cma_work *work;
1837 struct sockaddr *src, *dst;
1838 union ib_gid gid;
1839 int ret;
1840
1841 work = kzalloc(sizeof *work, GFP_KERNEL);
1842 if (!work)
1843 return -ENOMEM;
1844
1845 if (!id_priv->cma_dev) {
1846 ret = cma_bind_loopback(id_priv);
1847 if (ret)
1848 goto err;
1849 }
1850
1851 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1852 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
1853
1854 src = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
1855 if (cma_zero_addr(src)) {
1856 dst = (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
1857 if ((src->sa_family = dst->sa_family) == AF_INET) {
1858 ((struct sockaddr_in *) src)->sin_addr.s_addr =
1859 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1860 } else {
1861 ipv6_addr_copy(&((struct sockaddr_in6 *) src)->sin6_addr,
1862 &((struct sockaddr_in6 *) dst)->sin6_addr);
1863 }
1864 }
1865
1866 work->id = id_priv;
1867 INIT_WORK(&work->work, cma_work_handler);
1868 work->old_state = CMA_ADDR_QUERY;
1869 work->new_state = CMA_ADDR_RESOLVED;
1870 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
1871 queue_work(cma_wq, &work->work);
1872 return 0;
1873 err:
1874 kfree(work);
1875 return ret;
1876 }
1877
1878 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1879 struct sockaddr *dst_addr)
1880 {
1881 if (!src_addr || !src_addr->sa_family) {
1882 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
1883 if ((src_addr->sa_family = dst_addr->sa_family) == AF_INET6) {
1884 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
1885 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
1886 }
1887 }
1888 return rdma_bind_addr(id, src_addr);
1889 }
1890
1891 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
1892 struct sockaddr *dst_addr, int timeout_ms)
1893 {
1894 struct rdma_id_private *id_priv;
1895 int ret;
1896
1897 id_priv = container_of(id, struct rdma_id_private, id);
1898 if (id_priv->state == CMA_IDLE) {
1899 ret = cma_bind_addr(id, src_addr, dst_addr);
1900 if (ret)
1901 return ret;
1902 }
1903
1904 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_ADDR_QUERY))
1905 return -EINVAL;
1906
1907 atomic_inc(&id_priv->refcount);
1908 memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
1909 if (cma_any_addr(dst_addr))
1910 ret = cma_resolve_loopback(id_priv);
1911 else
1912 ret = rdma_resolve_ip(&addr_client, (struct sockaddr *) &id->route.addr.src_addr,
1913 dst_addr, &id->route.addr.dev_addr,
1914 timeout_ms, addr_handler, id_priv);
1915 if (ret)
1916 goto err;
1917
1918 return 0;
1919 err:
1920 cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_BOUND);
1921 cma_deref_id(id_priv);
1922 return ret;
1923 }
1924 EXPORT_SYMBOL(rdma_resolve_addr);
1925
1926 static void cma_bind_port(struct rdma_bind_list *bind_list,
1927 struct rdma_id_private *id_priv)
1928 {
1929 struct sockaddr_in *sin;
1930
1931 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1932 sin->sin_port = htons(bind_list->port);
1933 id_priv->bind_list = bind_list;
1934 hlist_add_head(&id_priv->node, &bind_list->owners);
1935 }
1936
1937 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
1938 unsigned short snum)
1939 {
1940 struct rdma_bind_list *bind_list;
1941 int port, ret;
1942
1943 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1944 if (!bind_list)
1945 return -ENOMEM;
1946
1947 do {
1948 ret = idr_get_new_above(ps, bind_list, snum, &port);
1949 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1950
1951 if (ret)
1952 goto err1;
1953
1954 if (port != snum) {
1955 ret = -EADDRNOTAVAIL;
1956 goto err2;
1957 }
1958
1959 bind_list->ps = ps;
1960 bind_list->port = (unsigned short) port;
1961 cma_bind_port(bind_list, id_priv);
1962 return 0;
1963 err2:
1964 idr_remove(ps, port);
1965 err1:
1966 kfree(bind_list);
1967 return ret;
1968 }
1969
1970 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
1971 {
1972 struct rdma_bind_list *bind_list;
1973 int port, ret, low, high;
1974
1975 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
1976 if (!bind_list)
1977 return -ENOMEM;
1978
1979 retry:
1980 /* FIXME: add proper port randomization per like inet_csk_get_port */
1981 do {
1982 ret = idr_get_new_above(ps, bind_list, next_port, &port);
1983 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
1984
1985 if (ret)
1986 goto err1;
1987
1988 inet_get_local_port_range(&low, &high);
1989 if (port > high) {
1990 if (next_port != low) {
1991 idr_remove(ps, port);
1992 next_port = low;
1993 goto retry;
1994 }
1995 ret = -EADDRNOTAVAIL;
1996 goto err2;
1997 }
1998
1999 if (port == high)
2000 next_port = low;
2001 else
2002 next_port = port + 1;
2003
2004 bind_list->ps = ps;
2005 bind_list->port = (unsigned short) port;
2006 cma_bind_port(bind_list, id_priv);
2007 return 0;
2008 err2:
2009 idr_remove(ps, port);
2010 err1:
2011 kfree(bind_list);
2012 return ret;
2013 }
2014
2015 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2016 {
2017 struct rdma_id_private *cur_id;
2018 struct sockaddr_in *sin, *cur_sin;
2019 struct rdma_bind_list *bind_list;
2020 struct hlist_node *node;
2021 unsigned short snum;
2022
2023 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
2024 snum = ntohs(sin->sin_port);
2025 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2026 return -EACCES;
2027
2028 bind_list = idr_find(ps, snum);
2029 if (!bind_list)
2030 return cma_alloc_port(ps, id_priv, snum);
2031
2032 /*
2033 * We don't support binding to any address if anyone is bound to
2034 * a specific address on the same port.
2035 */
2036 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr))
2037 return -EADDRNOTAVAIL;
2038
2039 hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
2040 if (cma_any_addr((struct sockaddr *) &cur_id->id.route.addr.src_addr))
2041 return -EADDRNOTAVAIL;
2042
2043 cur_sin = (struct sockaddr_in *) &cur_id->id.route.addr.src_addr;
2044 if (sin->sin_addr.s_addr == cur_sin->sin_addr.s_addr)
2045 return -EADDRINUSE;
2046 }
2047
2048 cma_bind_port(bind_list, id_priv);
2049 return 0;
2050 }
2051
2052 static int cma_get_port(struct rdma_id_private *id_priv)
2053 {
2054 struct idr *ps;
2055 int ret;
2056
2057 switch (id_priv->id.ps) {
2058 case RDMA_PS_SDP:
2059 ps = &sdp_ps;
2060 break;
2061 case RDMA_PS_TCP:
2062 ps = &tcp_ps;
2063 break;
2064 case RDMA_PS_UDP:
2065 ps = &udp_ps;
2066 break;
2067 case RDMA_PS_IPOIB:
2068 ps = &ipoib_ps;
2069 break;
2070 default:
2071 return -EPROTONOSUPPORT;
2072 }
2073
2074 mutex_lock(&lock);
2075 if (cma_any_port((struct sockaddr *) &id_priv->id.route.addr.src_addr))
2076 ret = cma_alloc_any_port(ps, id_priv);
2077 else
2078 ret = cma_use_port(ps, id_priv);
2079 mutex_unlock(&lock);
2080
2081 return ret;
2082 }
2083
2084 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2085 struct sockaddr *addr)
2086 {
2087 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
2088 struct sockaddr_in6 *sin6;
2089
2090 if (addr->sa_family != AF_INET6)
2091 return 0;
2092
2093 sin6 = (struct sockaddr_in6 *) addr;
2094 if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) &&
2095 !sin6->sin6_scope_id)
2096 return -EINVAL;
2097
2098 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2099 #endif
2100 return 0;
2101 }
2102
2103 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2104 {
2105 struct rdma_id_private *id_priv;
2106 int ret;
2107
2108 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
2109 return -EAFNOSUPPORT;
2110
2111 id_priv = container_of(id, struct rdma_id_private, id);
2112 if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND))
2113 return -EINVAL;
2114
2115 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2116 if (ret)
2117 goto err1;
2118
2119 if (!cma_any_addr(addr)) {
2120 ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
2121 if (ret)
2122 goto err1;
2123
2124 mutex_lock(&lock);
2125 ret = cma_acquire_dev(id_priv);
2126 mutex_unlock(&lock);
2127 if (ret)
2128 goto err1;
2129 }
2130
2131 memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
2132 ret = cma_get_port(id_priv);
2133 if (ret)
2134 goto err2;
2135
2136 return 0;
2137 err2:
2138 if (id_priv->cma_dev) {
2139 mutex_lock(&lock);
2140 cma_detach_from_dev(id_priv);
2141 mutex_unlock(&lock);
2142 }
2143 err1:
2144 cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_IDLE);
2145 return ret;
2146 }
2147 EXPORT_SYMBOL(rdma_bind_addr);
2148
2149 static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
2150 struct rdma_route *route)
2151 {
2152 struct cma_hdr *cma_hdr;
2153 struct sdp_hh *sdp_hdr;
2154
2155 if (route->addr.src_addr.ss_family == AF_INET) {
2156 struct sockaddr_in *src4, *dst4;
2157
2158 src4 = (struct sockaddr_in *) &route->addr.src_addr;
2159 dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
2160
2161 switch (ps) {
2162 case RDMA_PS_SDP:
2163 sdp_hdr = hdr;
2164 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2165 return -EINVAL;
2166 sdp_set_ip_ver(sdp_hdr, 4);
2167 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2168 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2169 sdp_hdr->port = src4->sin_port;
2170 break;
2171 default:
2172 cma_hdr = hdr;
2173 cma_hdr->cma_version = CMA_VERSION;
2174 cma_set_ip_ver(cma_hdr, 4);
2175 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2176 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2177 cma_hdr->port = src4->sin_port;
2178 break;
2179 }
2180 } else {
2181 struct sockaddr_in6 *src6, *dst6;
2182
2183 src6 = (struct sockaddr_in6 *) &route->addr.src_addr;
2184 dst6 = (struct sockaddr_in6 *) &route->addr.dst_addr;
2185
2186 switch (ps) {
2187 case RDMA_PS_SDP:
2188 sdp_hdr = hdr;
2189 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2190 return -EINVAL;
2191 sdp_set_ip_ver(sdp_hdr, 6);
2192 sdp_hdr->src_addr.ip6 = src6->sin6_addr;
2193 sdp_hdr->dst_addr.ip6 = dst6->sin6_addr;
2194 sdp_hdr->port = src6->sin6_port;
2195 break;
2196 default:
2197 cma_hdr = hdr;
2198 cma_hdr->cma_version = CMA_VERSION;
2199 cma_set_ip_ver(cma_hdr, 6);
2200 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2201 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2202 cma_hdr->port = src6->sin6_port;
2203 break;
2204 }
2205 }
2206 return 0;
2207 }
2208
2209 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2210 struct ib_cm_event *ib_event)
2211 {
2212 struct rdma_id_private *id_priv = cm_id->context;
2213 struct rdma_cm_event event;
2214 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2215 int ret = 0;
2216
2217 if (cma_disable_callback(id_priv, CMA_CONNECT))
2218 return 0;
2219
2220 memset(&event, 0, sizeof event);
2221 switch (ib_event->event) {
2222 case IB_CM_SIDR_REQ_ERROR:
2223 event.event = RDMA_CM_EVENT_UNREACHABLE;
2224 event.status = -ETIMEDOUT;
2225 break;
2226 case IB_CM_SIDR_REP_RECEIVED:
2227 event.param.ud.private_data = ib_event->private_data;
2228 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2229 if (rep->status != IB_SIDR_SUCCESS) {
2230 event.event = RDMA_CM_EVENT_UNREACHABLE;
2231 event.status = ib_event->param.sidr_rep_rcvd.status;
2232 break;
2233 }
2234 ret = cma_set_qkey(id_priv);
2235 if (ret) {
2236 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2237 event.status = -EINVAL;
2238 break;
2239 }
2240 if (id_priv->qkey != rep->qkey) {
2241 event.event = RDMA_CM_EVENT_UNREACHABLE;
2242 event.status = -EINVAL;
2243 break;
2244 }
2245 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2246 id_priv->id.route.path_rec,
2247 &event.param.ud.ah_attr);
2248 event.param.ud.qp_num = rep->qpn;
2249 event.param.ud.qkey = rep->qkey;
2250 event.event = RDMA_CM_EVENT_ESTABLISHED;
2251 event.status = 0;
2252 break;
2253 default:
2254 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2255 ib_event->event);
2256 goto out;
2257 }
2258
2259 ret = id_priv->id.event_handler(&id_priv->id, &event);
2260 if (ret) {
2261 /* Destroy the CM ID by returning a non-zero value. */
2262 id_priv->cm_id.ib = NULL;
2263 cma_exch(id_priv, CMA_DESTROYING);
2264 mutex_unlock(&id_priv->handler_mutex);
2265 rdma_destroy_id(&id_priv->id);
2266 return ret;
2267 }
2268 out:
2269 mutex_unlock(&id_priv->handler_mutex);
2270 return ret;
2271 }
2272
2273 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2274 struct rdma_conn_param *conn_param)
2275 {
2276 struct ib_cm_sidr_req_param req;
2277 struct rdma_route *route;
2278 int ret;
2279
2280 req.private_data_len = sizeof(struct cma_hdr) +
2281 conn_param->private_data_len;
2282 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2283 if (!req.private_data)
2284 return -ENOMEM;
2285
2286 if (conn_param->private_data && conn_param->private_data_len)
2287 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2288 conn_param->private_data, conn_param->private_data_len);
2289
2290 route = &id_priv->id.route;
2291 ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
2292 if (ret)
2293 goto out;
2294
2295 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device,
2296 cma_sidr_rep_handler, id_priv);
2297 if (IS_ERR(id_priv->cm_id.ib)) {
2298 ret = PTR_ERR(id_priv->cm_id.ib);
2299 goto out;
2300 }
2301
2302 req.path = route->path_rec;
2303 req.service_id = cma_get_service_id(id_priv->id.ps,
2304 (struct sockaddr *) &route->addr.dst_addr);
2305 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2306 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2307
2308 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2309 if (ret) {
2310 ib_destroy_cm_id(id_priv->cm_id.ib);
2311 id_priv->cm_id.ib = NULL;
2312 }
2313 out:
2314 kfree(req.private_data);
2315 return ret;
2316 }
2317
2318 static int cma_connect_ib(struct rdma_id_private *id_priv,
2319 struct rdma_conn_param *conn_param)
2320 {
2321 struct ib_cm_req_param req;
2322 struct rdma_route *route;
2323 void *private_data;
2324 int offset, ret;
2325
2326 memset(&req, 0, sizeof req);
2327 offset = cma_user_data_offset(id_priv->id.ps);
2328 req.private_data_len = offset + conn_param->private_data_len;
2329 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2330 if (!private_data)
2331 return -ENOMEM;
2332
2333 if (conn_param->private_data && conn_param->private_data_len)
2334 memcpy(private_data + offset, conn_param->private_data,
2335 conn_param->private_data_len);
2336
2337 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_ib_handler,
2338 id_priv);
2339 if (IS_ERR(id_priv->cm_id.ib)) {
2340 ret = PTR_ERR(id_priv->cm_id.ib);
2341 goto out;
2342 }
2343
2344 route = &id_priv->id.route;
2345 ret = cma_format_hdr(private_data, id_priv->id.ps, route);
2346 if (ret)
2347 goto out;
2348 req.private_data = private_data;
2349
2350 req.primary_path = &route->path_rec[0];
2351 if (route->num_paths == 2)
2352 req.alternate_path = &route->path_rec[1];
2353
2354 req.service_id = cma_get_service_id(id_priv->id.ps,
2355 (struct sockaddr *) &route->addr.dst_addr);
2356 req.qp_num = id_priv->qp_num;
2357 req.qp_type = IB_QPT_RC;
2358 req.starting_psn = id_priv->seq_num;
2359 req.responder_resources = conn_param->responder_resources;
2360 req.initiator_depth = conn_param->initiator_depth;
2361 req.flow_control = conn_param->flow_control;
2362 req.retry_count = conn_param->retry_count;
2363 req.rnr_retry_count = conn_param->rnr_retry_count;
2364 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2365 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2366 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2367 req.srq = id_priv->srq ? 1 : 0;
2368
2369 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2370 out:
2371 if (ret && !IS_ERR(id_priv->cm_id.ib)) {
2372 ib_destroy_cm_id(id_priv->cm_id.ib);
2373 id_priv->cm_id.ib = NULL;
2374 }
2375
2376 kfree(private_data);
2377 return ret;
2378 }
2379
2380 static int cma_connect_iw(struct rdma_id_private *id_priv,
2381 struct rdma_conn_param *conn_param)
2382 {
2383 struct iw_cm_id *cm_id;
2384 struct sockaddr_in* sin;
2385 int ret;
2386 struct iw_cm_conn_param iw_param;
2387
2388 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2389 if (IS_ERR(cm_id)) {
2390 ret = PTR_ERR(cm_id);
2391 goto out;
2392 }
2393
2394 id_priv->cm_id.iw = cm_id;
2395
2396 sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
2397 cm_id->local_addr = *sin;
2398
2399 sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
2400 cm_id->remote_addr = *sin;
2401
2402 ret = cma_modify_qp_rtr(id_priv, conn_param);
2403 if (ret)
2404 goto out;
2405
2406 iw_param.ord = conn_param->initiator_depth;
2407 iw_param.ird = conn_param->responder_resources;
2408 iw_param.private_data = conn_param->private_data;
2409 iw_param.private_data_len = conn_param->private_data_len;
2410 if (id_priv->id.qp)
2411 iw_param.qpn = id_priv->qp_num;
2412 else
2413 iw_param.qpn = conn_param->qp_num;
2414 ret = iw_cm_connect(cm_id, &iw_param);
2415 out:
2416 if (ret && !IS_ERR(cm_id)) {
2417 iw_destroy_cm_id(cm_id);
2418 id_priv->cm_id.iw = NULL;
2419 }
2420 return ret;
2421 }
2422
2423 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2424 {
2425 struct rdma_id_private *id_priv;
2426 int ret;
2427
2428 id_priv = container_of(id, struct rdma_id_private, id);
2429 if (!cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_CONNECT))
2430 return -EINVAL;
2431
2432 if (!id->qp) {
2433 id_priv->qp_num = conn_param->qp_num;
2434 id_priv->srq = conn_param->srq;
2435 }
2436
2437 switch (rdma_node_get_transport(id->device->node_type)) {
2438 case RDMA_TRANSPORT_IB:
2439 if (cma_is_ud_ps(id->ps))
2440 ret = cma_resolve_ib_udp(id_priv, conn_param);
2441 else
2442 ret = cma_connect_ib(id_priv, conn_param);
2443 break;
2444 case RDMA_TRANSPORT_IWARP:
2445 ret = cma_connect_iw(id_priv, conn_param);
2446 break;
2447 default:
2448 ret = -ENOSYS;
2449 break;
2450 }
2451 if (ret)
2452 goto err;
2453
2454 return 0;
2455 err:
2456 cma_comp_exch(id_priv, CMA_CONNECT, CMA_ROUTE_RESOLVED);
2457 return ret;
2458 }
2459 EXPORT_SYMBOL(rdma_connect);
2460
2461 static int cma_accept_ib(struct rdma_id_private *id_priv,
2462 struct rdma_conn_param *conn_param)
2463 {
2464 struct ib_cm_rep_param rep;
2465 int ret;
2466
2467 ret = cma_modify_qp_rtr(id_priv, conn_param);
2468 if (ret)
2469 goto out;
2470
2471 ret = cma_modify_qp_rts(id_priv, conn_param);
2472 if (ret)
2473 goto out;
2474
2475 memset(&rep, 0, sizeof rep);
2476 rep.qp_num = id_priv->qp_num;
2477 rep.starting_psn = id_priv->seq_num;
2478 rep.private_data = conn_param->private_data;
2479 rep.private_data_len = conn_param->private_data_len;
2480 rep.responder_resources = conn_param->responder_resources;
2481 rep.initiator_depth = conn_param->initiator_depth;
2482 rep.failover_accepted = 0;
2483 rep.flow_control = conn_param->flow_control;
2484 rep.rnr_retry_count = conn_param->rnr_retry_count;
2485 rep.srq = id_priv->srq ? 1 : 0;
2486
2487 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2488 out:
2489 return ret;
2490 }
2491
2492 static int cma_accept_iw(struct rdma_id_private *id_priv,
2493 struct rdma_conn_param *conn_param)
2494 {
2495 struct iw_cm_conn_param iw_param;
2496 int ret;
2497
2498 ret = cma_modify_qp_rtr(id_priv, conn_param);
2499 if (ret)
2500 return ret;
2501
2502 iw_param.ord = conn_param->initiator_depth;
2503 iw_param.ird = conn_param->responder_resources;
2504 iw_param.private_data = conn_param->private_data;
2505 iw_param.private_data_len = conn_param->private_data_len;
2506 if (id_priv->id.qp) {
2507 iw_param.qpn = id_priv->qp_num;
2508 } else
2509 iw_param.qpn = conn_param->qp_num;
2510
2511 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2512 }
2513
2514 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2515 enum ib_cm_sidr_status status,
2516 const void *private_data, int private_data_len)
2517 {
2518 struct ib_cm_sidr_rep_param rep;
2519 int ret;
2520
2521 memset(&rep, 0, sizeof rep);
2522 rep.status = status;
2523 if (status == IB_SIDR_SUCCESS) {
2524 ret = cma_set_qkey(id_priv);
2525 if (ret)
2526 return ret;
2527 rep.qp_num = id_priv->qp_num;
2528 rep.qkey = id_priv->qkey;
2529 }
2530 rep.private_data = private_data;
2531 rep.private_data_len = private_data_len;
2532
2533 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2534 }
2535
2536 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2537 {
2538 struct rdma_id_private *id_priv;
2539 int ret;
2540
2541 id_priv = container_of(id, struct rdma_id_private, id);
2542 if (!cma_comp(id_priv, CMA_CONNECT))
2543 return -EINVAL;
2544
2545 if (!id->qp && conn_param) {
2546 id_priv->qp_num = conn_param->qp_num;
2547 id_priv->srq = conn_param->srq;
2548 }
2549
2550 switch (rdma_node_get_transport(id->device->node_type)) {
2551 case RDMA_TRANSPORT_IB:
2552 if (cma_is_ud_ps(id->ps))
2553 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2554 conn_param->private_data,
2555 conn_param->private_data_len);
2556 else if (conn_param)
2557 ret = cma_accept_ib(id_priv, conn_param);
2558 else
2559 ret = cma_rep_recv(id_priv);
2560 break;
2561 case RDMA_TRANSPORT_IWARP:
2562 ret = cma_accept_iw(id_priv, conn_param);
2563 break;
2564 default:
2565 ret = -ENOSYS;
2566 break;
2567 }
2568
2569 if (ret)
2570 goto reject;
2571
2572 return 0;
2573 reject:
2574 cma_modify_qp_err(id_priv);
2575 rdma_reject(id, NULL, 0);
2576 return ret;
2577 }
2578 EXPORT_SYMBOL(rdma_accept);
2579
2580 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2581 {
2582 struct rdma_id_private *id_priv;
2583 int ret;
2584
2585 id_priv = container_of(id, struct rdma_id_private, id);
2586 if (!cma_has_cm_dev(id_priv))
2587 return -EINVAL;
2588
2589 switch (id->device->node_type) {
2590 case RDMA_NODE_IB_CA:
2591 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2592 break;
2593 default:
2594 ret = 0;
2595 break;
2596 }
2597 return ret;
2598 }
2599 EXPORT_SYMBOL(rdma_notify);
2600
2601 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2602 u8 private_data_len)
2603 {
2604 struct rdma_id_private *id_priv;
2605 int ret;
2606
2607 id_priv = container_of(id, struct rdma_id_private, id);
2608 if (!cma_has_cm_dev(id_priv))
2609 return -EINVAL;
2610
2611 switch (rdma_node_get_transport(id->device->node_type)) {
2612 case RDMA_TRANSPORT_IB:
2613 if (cma_is_ud_ps(id->ps))
2614 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
2615 private_data, private_data_len);
2616 else
2617 ret = ib_send_cm_rej(id_priv->cm_id.ib,
2618 IB_CM_REJ_CONSUMER_DEFINED, NULL,
2619 0, private_data, private_data_len);
2620 break;
2621 case RDMA_TRANSPORT_IWARP:
2622 ret = iw_cm_reject(id_priv->cm_id.iw,
2623 private_data, private_data_len);
2624 break;
2625 default:
2626 ret = -ENOSYS;
2627 break;
2628 }
2629 return ret;
2630 }
2631 EXPORT_SYMBOL(rdma_reject);
2632
2633 int rdma_disconnect(struct rdma_cm_id *id)
2634 {
2635 struct rdma_id_private *id_priv;
2636 int ret;
2637
2638 id_priv = container_of(id, struct rdma_id_private, id);
2639 if (!cma_has_cm_dev(id_priv))
2640 return -EINVAL;
2641
2642 switch (rdma_node_get_transport(id->device->node_type)) {
2643 case RDMA_TRANSPORT_IB:
2644 ret = cma_modify_qp_err(id_priv);
2645 if (ret)
2646 goto out;
2647 /* Initiate or respond to a disconnect. */
2648 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
2649 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
2650 break;
2651 case RDMA_TRANSPORT_IWARP:
2652 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
2653 break;
2654 default:
2655 ret = -EINVAL;
2656 break;
2657 }
2658 out:
2659 return ret;
2660 }
2661 EXPORT_SYMBOL(rdma_disconnect);
2662
2663 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
2664 {
2665 struct rdma_id_private *id_priv;
2666 struct cma_multicast *mc = multicast->context;
2667 struct rdma_cm_event event;
2668 int ret;
2669
2670 id_priv = mc->id_priv;
2671 if (cma_disable_callback(id_priv, CMA_ADDR_BOUND) &&
2672 cma_disable_callback(id_priv, CMA_ADDR_RESOLVED))
2673 return 0;
2674
2675 mutex_lock(&id_priv->qp_mutex);
2676 if (!status && id_priv->id.qp)
2677 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
2678 multicast->rec.mlid);
2679 mutex_unlock(&id_priv->qp_mutex);
2680
2681 memset(&event, 0, sizeof event);
2682 event.status = status;
2683 event.param.ud.private_data = mc->context;
2684 if (!status) {
2685 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
2686 ib_init_ah_from_mcmember(id_priv->id.device,
2687 id_priv->id.port_num, &multicast->rec,
2688 &event.param.ud.ah_attr);
2689 event.param.ud.qp_num = 0xFFFFFF;
2690 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
2691 } else
2692 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
2693
2694 ret = id_priv->id.event_handler(&id_priv->id, &event);
2695 if (ret) {
2696 cma_exch(id_priv, CMA_DESTROYING);
2697 mutex_unlock(&id_priv->handler_mutex);
2698 rdma_destroy_id(&id_priv->id);
2699 return 0;
2700 }
2701
2702 mutex_unlock(&id_priv->handler_mutex);
2703 return 0;
2704 }
2705
2706 static void cma_set_mgid(struct rdma_id_private *id_priv,
2707 struct sockaddr *addr, union ib_gid *mgid)
2708 {
2709 unsigned char mc_map[MAX_ADDR_LEN];
2710 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2711 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
2712 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
2713
2714 if (cma_any_addr(addr)) {
2715 memset(mgid, 0, sizeof *mgid);
2716 } else if ((addr->sa_family == AF_INET6) &&
2717 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
2718 0xFF10A01B)) {
2719 /* IPv6 address is an SA assigned MGID. */
2720 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
2721 } else if ((addr->sa_family == AF_INET6)) {
2722 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
2723 if (id_priv->id.ps == RDMA_PS_UDP)
2724 mc_map[7] = 0x01; /* Use RDMA CM signature */
2725 *mgid = *(union ib_gid *) (mc_map + 4);
2726 } else {
2727 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
2728 if (id_priv->id.ps == RDMA_PS_UDP)
2729 mc_map[7] = 0x01; /* Use RDMA CM signature */
2730 *mgid = *(union ib_gid *) (mc_map + 4);
2731 }
2732 }
2733
2734 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
2735 struct cma_multicast *mc)
2736 {
2737 struct ib_sa_mcmember_rec rec;
2738 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2739 ib_sa_comp_mask comp_mask;
2740 int ret;
2741
2742 ib_addr_get_mgid(dev_addr, &rec.mgid);
2743 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
2744 &rec.mgid, &rec);
2745 if (ret)
2746 return ret;
2747
2748 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
2749 if (id_priv->id.ps == RDMA_PS_UDP)
2750 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
2751 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
2752 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2753 rec.join_state = 1;
2754
2755 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
2756 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
2757 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
2758 IB_SA_MCMEMBER_REC_FLOW_LABEL |
2759 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
2760
2761 if (id_priv->id.ps == RDMA_PS_IPOIB)
2762 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
2763 IB_SA_MCMEMBER_REC_RATE_SELECTOR;
2764
2765 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
2766 id_priv->id.port_num, &rec,
2767 comp_mask, GFP_KERNEL,
2768 cma_ib_mc_handler, mc);
2769 if (IS_ERR(mc->multicast.ib))
2770 return PTR_ERR(mc->multicast.ib);
2771
2772 return 0;
2773 }
2774
2775 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
2776 void *context)
2777 {
2778 struct rdma_id_private *id_priv;
2779 struct cma_multicast *mc;
2780 int ret;
2781
2782 id_priv = container_of(id, struct rdma_id_private, id);
2783 if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
2784 !cma_comp(id_priv, CMA_ADDR_RESOLVED))
2785 return -EINVAL;
2786
2787 mc = kmalloc(sizeof *mc, GFP_KERNEL);
2788 if (!mc)
2789 return -ENOMEM;
2790
2791 memcpy(&mc->addr, addr, ip_addr_size(addr));
2792 mc->context = context;
2793 mc->id_priv = id_priv;
2794
2795 spin_lock(&id_priv->lock);
2796 list_add(&mc->list, &id_priv->mc_list);
2797 spin_unlock(&id_priv->lock);
2798
2799 switch (rdma_node_get_transport(id->device->node_type)) {
2800 case RDMA_TRANSPORT_IB:
2801 ret = cma_join_ib_multicast(id_priv, mc);
2802 break;
2803 default:
2804 ret = -ENOSYS;
2805 break;
2806 }
2807
2808 if (ret) {
2809 spin_lock_irq(&id_priv->lock);
2810 list_del(&mc->list);
2811 spin_unlock_irq(&id_priv->lock);
2812 kfree(mc);
2813 }
2814 return ret;
2815 }
2816 EXPORT_SYMBOL(rdma_join_multicast);
2817
2818 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
2819 {
2820 struct rdma_id_private *id_priv;
2821 struct cma_multicast *mc;
2822
2823 id_priv = container_of(id, struct rdma_id_private, id);
2824 spin_lock_irq(&id_priv->lock);
2825 list_for_each_entry(mc, &id_priv->mc_list, list) {
2826 if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
2827 list_del(&mc->list);
2828 spin_unlock_irq(&id_priv->lock);
2829
2830 if (id->qp)
2831 ib_detach_mcast(id->qp,
2832 &mc->multicast.ib->rec.mgid,
2833 mc->multicast.ib->rec.mlid);
2834 ib_sa_free_multicast(mc->multicast.ib);
2835 kfree(mc);
2836 return;
2837 }
2838 }
2839 spin_unlock_irq(&id_priv->lock);
2840 }
2841 EXPORT_SYMBOL(rdma_leave_multicast);
2842
2843 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
2844 {
2845 struct rdma_dev_addr *dev_addr;
2846 struct cma_ndev_work *work;
2847
2848 dev_addr = &id_priv->id.route.addr.dev_addr;
2849
2850 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
2851 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
2852 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
2853 ndev->name, &id_priv->id);
2854 work = kzalloc(sizeof *work, GFP_KERNEL);
2855 if (!work)
2856 return -ENOMEM;
2857
2858 INIT_WORK(&work->work, cma_ndev_work_handler);
2859 work->id = id_priv;
2860 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
2861 atomic_inc(&id_priv->refcount);
2862 queue_work(cma_wq, &work->work);
2863 }
2864
2865 return 0;
2866 }
2867
2868 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
2869 void *ctx)
2870 {
2871 struct net_device *ndev = (struct net_device *)ctx;
2872 struct cma_device *cma_dev;
2873 struct rdma_id_private *id_priv;
2874 int ret = NOTIFY_DONE;
2875
2876 if (dev_net(ndev) != &init_net)
2877 return NOTIFY_DONE;
2878
2879 if (event != NETDEV_BONDING_FAILOVER)
2880 return NOTIFY_DONE;
2881
2882 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
2883 return NOTIFY_DONE;
2884
2885 mutex_lock(&lock);
2886 list_for_each_entry(cma_dev, &dev_list, list)
2887 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
2888 ret = cma_netdev_change(ndev, id_priv);
2889 if (ret)
2890 goto out;
2891 }
2892
2893 out:
2894 mutex_unlock(&lock);
2895 return ret;
2896 }
2897
2898 static struct notifier_block cma_nb = {
2899 .notifier_call = cma_netdev_callback
2900 };
2901
2902 static void cma_add_one(struct ib_device *device)
2903 {
2904 struct cma_device *cma_dev;
2905 struct rdma_id_private *id_priv;
2906
2907 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
2908 if (!cma_dev)
2909 return;
2910
2911 cma_dev->device = device;
2912
2913 init_completion(&cma_dev->comp);
2914 atomic_set(&cma_dev->refcount, 1);
2915 INIT_LIST_HEAD(&cma_dev->id_list);
2916 ib_set_client_data(device, &cma_client, cma_dev);
2917
2918 mutex_lock(&lock);
2919 list_add_tail(&cma_dev->list, &dev_list);
2920 list_for_each_entry(id_priv, &listen_any_list, list)
2921 cma_listen_on_dev(id_priv, cma_dev);
2922 mutex_unlock(&lock);
2923 }
2924
2925 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
2926 {
2927 struct rdma_cm_event event;
2928 enum cma_state state;
2929 int ret = 0;
2930
2931 /* Record that we want to remove the device */
2932 state = cma_exch(id_priv, CMA_DEVICE_REMOVAL);
2933 if (state == CMA_DESTROYING)
2934 return 0;
2935
2936 cma_cancel_operation(id_priv, state);
2937 mutex_lock(&id_priv->handler_mutex);
2938
2939 /* Check for destruction from another callback. */
2940 if (!cma_comp(id_priv, CMA_DEVICE_REMOVAL))
2941 goto out;
2942
2943 memset(&event, 0, sizeof event);
2944 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
2945 ret = id_priv->id.event_handler(&id_priv->id, &event);
2946 out:
2947 mutex_unlock(&id_priv->handler_mutex);
2948 return ret;
2949 }
2950
2951 static void cma_process_remove(struct cma_device *cma_dev)
2952 {
2953 struct rdma_id_private *id_priv;
2954 int ret;
2955
2956 mutex_lock(&lock);
2957 while (!list_empty(&cma_dev->id_list)) {
2958 id_priv = list_entry(cma_dev->id_list.next,
2959 struct rdma_id_private, list);
2960
2961 list_del(&id_priv->listen_list);
2962 list_del_init(&id_priv->list);
2963 atomic_inc(&id_priv->refcount);
2964 mutex_unlock(&lock);
2965
2966 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
2967 cma_deref_id(id_priv);
2968 if (ret)
2969 rdma_destroy_id(&id_priv->id);
2970
2971 mutex_lock(&lock);
2972 }
2973 mutex_unlock(&lock);
2974
2975 cma_deref_dev(cma_dev);
2976 wait_for_completion(&cma_dev->comp);
2977 }
2978
2979 static void cma_remove_one(struct ib_device *device)
2980 {
2981 struct cma_device *cma_dev;
2982
2983 cma_dev = ib_get_client_data(device, &cma_client);
2984 if (!cma_dev)
2985 return;
2986
2987 mutex_lock(&lock);
2988 list_del(&cma_dev->list);
2989 mutex_unlock(&lock);
2990
2991 cma_process_remove(cma_dev);
2992 kfree(cma_dev);
2993 }
2994
2995 static int __init cma_init(void)
2996 {
2997 int ret, low, high, remaining;
2998
2999 get_random_bytes(&next_port, sizeof next_port);
3000 inet_get_local_port_range(&low, &high);
3001 remaining = (high - low) + 1;
3002 next_port = ((unsigned int) next_port % remaining) + low;
3003
3004 cma_wq = create_singlethread_workqueue("rdma_cm");
3005 if (!cma_wq)
3006 return -ENOMEM;
3007
3008 ib_sa_register_client(&sa_client);
3009 rdma_addr_register_client(&addr_client);
3010 register_netdevice_notifier(&cma_nb);
3011
3012 ret = ib_register_client(&cma_client);
3013 if (ret)
3014 goto err;
3015 return 0;
3016
3017 err:
3018 unregister_netdevice_notifier(&cma_nb);
3019 rdma_addr_unregister_client(&addr_client);
3020 ib_sa_unregister_client(&sa_client);
3021 destroy_workqueue(cma_wq);
3022 return ret;
3023 }
3024
3025 static void __exit cma_cleanup(void)
3026 {
3027 ib_unregister_client(&cma_client);
3028 unregister_netdevice_notifier(&cma_nb);
3029 rdma_addr_unregister_client(&addr_client);
3030 ib_sa_unregister_client(&sa_client);
3031 destroy_workqueue(cma_wq);
3032 idr_destroy(&sdp_ps);
3033 idr_destroy(&tcp_ps);
3034 idr_destroy(&udp_ps);
3035 idr_destroy(&ipoib_ps);
3036 }
3037
3038 module_init(cma_init);
3039 module_exit(cma_cleanup);
ubuntu-bionic-kernel mirror