projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/cm: Cancel pending LAP message when exiting IB_CM_ESTABLISH state
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / infiniband / core / cma.c
CommitLineData
e51060f0
SH		 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 *
a9474917
SH		 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:
e51060f0 12 *
a9474917
SH		 13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
e51060f0 16 *
a9474917
SH		 17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
e51060f0 20 *
a9474917
SH		 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.
e51060f0 25 *
a9474917
SH		 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.
e51060f0
SH		 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>
07ebafba 42#include <linux/inetdevice.h>
5a0e3ad6 43#include <linux/slab.h>
e51060f0
SH		 44
45#include <net/tcp.h>
1f5175ad 46#include <net/ipv6.h>
e51060f0
SH		 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>
07ebafba 53#include <rdma/iw_cm.h>
e51060f0
SH		 54
55MODULE_AUTHOR("Sean Hefty");
56MODULE_DESCRIPTION("Generic RDMA CM Agent");
57MODULE_LICENSE("Dual BSD/GPL");
58
59#define CMA_CM_RESPONSE_TIMEOUT 20
d5bb7599 60#define CMA_MAX_CM_RETRIES 15
dcb3f974 61#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
3c86aa70 62#define CMA_IBOE_PACKET_LIFETIME 18
e51060f0
SH		 63
64static void cma_add_one(struct ib_device *device);
65static void cma_remove_one(struct ib_device *device);
66
67static struct ib_client cma_client = {
68 .name = "cma",
69 .add = cma_add_one,
70 .remove = cma_remove_one
71};
72
c1a0b23b 73static struct ib_sa_client sa_client;
7a118df3 74static struct rdma_addr_client addr_client;
e51060f0
SH		 75static LIST_HEAD(dev_list);
76static LIST_HEAD(listen_any_list);
77static DEFINE_MUTEX(lock);
78static struct workqueue_struct *cma_wq;
79static DEFINE_IDR(sdp_ps);
80static DEFINE_IDR(tcp_ps);
628e5f6d 81static DEFINE_IDR(udp_ps);
c8f6a362 82static DEFINE_IDR(ipoib_ps);
e51060f0
SH		 83
84struct cma_device {
85 struct list_head list;
86 struct ib_device *device;
e51060f0
SH		 87 struct completion comp;
88 atomic_t refcount;
89 struct list_head id_list;
90};
91
92enum 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
106struct 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 */
118struct rdma_id_private {
119 struct rdma_cm_id id;
120
121 struct rdma_bind_list *bind_list;
122 struct hlist_node node;
d02d1f53
SH		 123 struct list_head list; /* listen_any_list or cma_device.list */
124 struct list_head listen_list; /* per device listens */
e51060f0 125 struct cma_device *cma_dev;
c8f6a362 126 struct list_head mc_list;
e51060f0 127
d02d1f53 128 int internal_id;
e51060f0
SH		 129 enum cma_state state;
130 spinlock_t lock;
c5483388
SH		 131 struct mutex qp_mutex;
132
e51060f0
SH		 133 struct completion comp;
134 atomic_t refcount;
de910bd9 135 struct mutex handler_mutex;
e51060f0
SH		 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;
07ebafba 143 struct iw_cm_id *iw;
e51060f0
SH		 144 } cm_id;
145
146 u32 seq_num;
c8f6a362 147 u32 qkey;
e51060f0 148 u32 qp_num;
e51060f0 149 u8 srq;
a81c994d 150 u8 tos;
e51060f0
SH		 151};
152
c8f6a362
SH		 153struct 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;
3f446754 160 struct sockaddr_storage addr;
3c86aa70 161 struct kref mcref;
c8f6a362
SH		 162};
163
e51060f0
SH		 164struct cma_work {
165 struct work_struct work;
166 struct rdma_id_private *id;
167 enum cma_state old_state;
168 enum cma_state new_state;
169 struct rdma_cm_event event;
170};
171
dd5bdff8
OG		 172struct cma_ndev_work {
173 struct work_struct work;
174 struct rdma_id_private *id;
175 struct rdma_cm_event event;
176};
177
3c86aa70
EC		 178struct iboe_mcast_work {
179 struct work_struct work;
180 struct rdma_id_private *id;
181 struct cma_multicast *mc;
182};
183
e51060f0
SH		 184union cma_ip_addr {
185 struct in6_addr ip6;
186 struct {
1b90c137
AV		 187 __be32 pad[3];
188 __be32 addr;
e51060f0
SH		 189 } ip4;
190};
191
192struct cma_hdr {
193 u8 cma_version;
194 u8 ip_version; /* IP version: 7:4 */
1b90c137 195 __be16 port;
e51060f0
SH		 196 union cma_ip_addr src_addr;
197 union cma_ip_addr dst_addr;
198};
199
200struct sdp_hh {
201 u8 bsdh[16];
202 u8 sdp_version; /* Major version: 7:4 */
203 u8 ip_version; /* IP version: 7:4 */
204 u8 sdp_specific1[10];
1b90c137
AV		 205 __be16 port;
206 __be16 sdp_specific2;
e51060f0
SH		 207 union cma_ip_addr src_addr;
208 union cma_ip_addr dst_addr;
209};
210
211struct sdp_hah {
212 u8 bsdh[16];
213 u8 sdp_version;
214};
215
216#define CMA_VERSION 0x00
217#define SDP_MAJ_VERSION 0x2
218
219static int cma_comp(struct rdma_id_private *id_priv, enum cma_state comp)
220{
221 unsigned long flags;
222 int ret;
223
224 spin_lock_irqsave(&id_priv->lock, flags);
225 ret = (id_priv->state == comp);
226 spin_unlock_irqrestore(&id_priv->lock, flags);
227 return ret;
228}
229
230static int cma_comp_exch(struct rdma_id_private *id_priv,
231 enum cma_state comp, enum cma_state exch)
232{
233 unsigned long flags;
234 int ret;
235
236 spin_lock_irqsave(&id_priv->lock, flags);
237 if ((ret = (id_priv->state == comp)))
238 id_priv->state = exch;
239 spin_unlock_irqrestore(&id_priv->lock, flags);
240 return ret;
241}
242
243static enum cma_state cma_exch(struct rdma_id_private *id_priv,
244 enum cma_state exch)
245{
246 unsigned long flags;
247 enum cma_state old;
248
249 spin_lock_irqsave(&id_priv->lock, flags);
250 old = id_priv->state;
251 id_priv->state = exch;
252 spin_unlock_irqrestore(&id_priv->lock, flags);
253 return old;
254}
255
256static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
257{
258 return hdr->ip_version >> 4;
259}
260
261static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
262{
263 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
264}
265
266static inline u8 sdp_get_majv(u8 sdp_version)
267{
268 return sdp_version >> 4;
269}
270
271static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
272{
273 return hh->ip_version >> 4;
274}
275
276static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
277{
278 hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
279}
280
c8f6a362
SH		 281static inline int cma_is_ud_ps(enum rdma_port_space ps)
282{
283 return (ps == RDMA_PS_UDP || ps == RDMA_PS_IPOIB);
284}
285
e51060f0
SH		 286static void cma_attach_to_dev(struct rdma_id_private *id_priv,
287 struct cma_device *cma_dev)
288{
289 atomic_inc(&cma_dev->refcount);
290 id_priv->cma_dev = cma_dev;
291 id_priv->id.device = cma_dev->device;
3c86aa70
EC		 292 id_priv->id.route.addr.dev_addr.transport =
293 rdma_node_get_transport(cma_dev->device->node_type);
e51060f0
SH		 294 list_add_tail(&id_priv->list, &cma_dev->id_list);
295}
296
297static inline void cma_deref_dev(struct cma_device *cma_dev)
298{
299 if (atomic_dec_and_test(&cma_dev->refcount))
300 complete(&cma_dev->comp);
301}
302
3c86aa70
EC		 303static inline void release_mc(struct kref *kref)
304{
305 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
306
307 kfree(mc->multicast.ib);
308 kfree(mc);
309}
310
e51060f0
SH		 311static void cma_detach_from_dev(struct rdma_id_private *id_priv)
312{
313 list_del(&id_priv->list);
314 cma_deref_dev(id_priv->cma_dev);
315 id_priv->cma_dev = NULL;
316}
317
d2ca39f2 318static int cma_set_qkey(struct rdma_id_private *id_priv)
c8f6a362
SH		 319{
320 struct ib_sa_mcmember_rec rec;
321 int ret = 0;
322
d2ca39f2
YE		 323 if (id_priv->qkey)
324 return 0;
325
326 switch (id_priv->id.ps) {
c8f6a362 327 case RDMA_PS_UDP:
d2ca39f2 328 id_priv->qkey = RDMA_UDP_QKEY;
c8f6a362
SH		 329 break;
330 case RDMA_PS_IPOIB:
d2ca39f2
YE		 331 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
332 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
333 id_priv->id.port_num, &rec.mgid,
334 &rec);
335 if (!ret)
336 id_priv->qkey = be32_to_cpu(rec.qkey);
c8f6a362
SH		 337 break;
338 default:
339 break;
340 }
341 return ret;
342}
343
3c86aa70
EC		 344static int find_gid_port(struct ib_device *device, union ib_gid *gid, u8 port_num)
345{
346 int i;
347 int err;
348 struct ib_port_attr props;
349 union ib_gid tmp;
350
351 err = ib_query_port(device, port_num, &props);
352 if (err)
353 return 1;
354
355 for (i = 0; i < props.gid_tbl_len; ++i) {
356 err = ib_query_gid(device, port_num, i, &tmp);
357 if (err)
358 return 1;
359 if (!memcmp(&tmp, gid, sizeof tmp))
360 return 0;
361 }
362
363 return -EAGAIN;
364}
365
07ebafba 366static int cma_acquire_dev(struct rdma_id_private *id_priv)
e51060f0 367{
c8f6a362 368 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
e51060f0 369 struct cma_device *cma_dev;
3c86aa70 370 union ib_gid gid, iboe_gid;
e51060f0 371 int ret = -ENODEV;
3c86aa70
EC		 372 u8 port;
373 enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
374 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
e51060f0 375
3c86aa70
EC		 376 iboe_addr_get_sgid(dev_addr, &iboe_gid);
377 memcpy(&gid, dev_addr->src_dev_addr +
378 rdma_addr_gid_offset(dev_addr), sizeof gid);
e51060f0 379 list_for_each_entry(cma_dev, &dev_list, list) {
3c86aa70
EC		 380 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
381 if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
382 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
383 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
384 ret = find_gid_port(cma_dev->device, &iboe_gid, port);
385 else
386 ret = find_gid_port(cma_dev->device, &gid, port);
387
388 if (!ret) {
389 id_priv->id.port_num = port;
390 goto out;
391 } else if (ret == 1)
392 break;
393 }
e51060f0
SH		 394 }
395 }
3c86aa70
EC		 396
397out:
398 if (!ret)
399 cma_attach_to_dev(id_priv, cma_dev);
400
e51060f0
SH		 401 return ret;
402}
403
e51060f0
SH		 404static void cma_deref_id(struct rdma_id_private *id_priv)
405{
406 if (atomic_dec_and_test(&id_priv->refcount))
407 complete(&id_priv->comp);
408}
409
de910bd9 410static int cma_disable_callback(struct rdma_id_private *id_priv,
8aa08602
SH		 411 enum cma_state state)
412{
de910bd9
OG		 413 mutex_lock(&id_priv->handler_mutex);
414 if (id_priv->state != state) {
415 mutex_unlock(&id_priv->handler_mutex);
416 return -EINVAL;
417 }
418 return 0;
e51060f0
SH		 419}
420
6c719f5c
SH		 421static int cma_has_cm_dev(struct rdma_id_private *id_priv)
422{
423 return (id_priv->id.device && id_priv->cm_id.ib);
424}
425
e51060f0
SH		 426struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
427 void *context, enum rdma_port_space ps)
428{
429 struct rdma_id_private *id_priv;
430
431 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
432 if (!id_priv)
433 return ERR_PTR(-ENOMEM);
434
435 id_priv->state = CMA_IDLE;
436 id_priv->id.context = context;
437 id_priv->id.event_handler = event_handler;
438 id_priv->id.ps = ps;
439 spin_lock_init(&id_priv->lock);
c5483388 440 mutex_init(&id_priv->qp_mutex);
e51060f0
SH		 441 init_completion(&id_priv->comp);
442 atomic_set(&id_priv->refcount, 1);
de910bd9 443 mutex_init(&id_priv->handler_mutex);
e51060f0 444 INIT_LIST_HEAD(&id_priv->listen_list);
c8f6a362 445 INIT_LIST_HEAD(&id_priv->mc_list);
e51060f0
SH		 446 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
447
448 return &id_priv->id;
449}
450EXPORT_SYMBOL(rdma_create_id);
451
c8f6a362 452static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
e51060f0
SH		 453{
454 struct ib_qp_attr qp_attr;
c8f6a362 455 int qp_attr_mask, ret;
e51060f0 456
c8f6a362
SH		 457 qp_attr.qp_state = IB_QPS_INIT;
458 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0
SH		 459 if (ret)
460 return ret;
461
c8f6a362
SH		 462 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
463 if (ret)
464 return ret;
465
466 qp_attr.qp_state = IB_QPS_RTR;
467 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
468 if (ret)
469 return ret;
470
471 qp_attr.qp_state = IB_QPS_RTS;
472 qp_attr.sq_psn = 0;
473 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
474
475 return ret;
e51060f0
SH		 476}
477
c8f6a362 478static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
07ebafba
TT		 479{
480 struct ib_qp_attr qp_attr;
c8f6a362 481 int qp_attr_mask, ret;
07ebafba
TT		 482
483 qp_attr.qp_state = IB_QPS_INIT;
c8f6a362
SH		 484 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
485 if (ret)
486 return ret;
07ebafba 487
c8f6a362 488 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
07ebafba
TT		 489}
490
e51060f0
SH		 491int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
492 struct ib_qp_init_attr *qp_init_attr)
493{
494 struct rdma_id_private *id_priv;
495 struct ib_qp *qp;
496 int ret;
497
498 id_priv = container_of(id, struct rdma_id_private, id);
499 if (id->device != pd->device)
500 return -EINVAL;
501
502 qp = ib_create_qp(pd, qp_init_attr);
503 if (IS_ERR(qp))
504 return PTR_ERR(qp);
505
c8f6a362
SH		 506 if (cma_is_ud_ps(id_priv->id.ps))
507 ret = cma_init_ud_qp(id_priv, qp);
508 else
509 ret = cma_init_conn_qp(id_priv, qp);
e51060f0
SH		 510 if (ret)
511 goto err;
512
513 id->qp = qp;
514 id_priv->qp_num = qp->qp_num;
e51060f0
SH		 515 id_priv->srq = (qp->srq != NULL);
516 return 0;
517err:
518 ib_destroy_qp(qp);
519 return ret;
520}
521EXPORT_SYMBOL(rdma_create_qp);
522
523void rdma_destroy_qp(struct rdma_cm_id *id)
524{
c5483388
SH		 525 struct rdma_id_private *id_priv;
526
527 id_priv = container_of(id, struct rdma_id_private, id);
528 mutex_lock(&id_priv->qp_mutex);
529 ib_destroy_qp(id_priv->id.qp);
530 id_priv->id.qp = NULL;
531 mutex_unlock(&id_priv->qp_mutex);
e51060f0
SH		 532}
533EXPORT_SYMBOL(rdma_destroy_qp);
534
5851bb89
SH		 535static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
536 struct rdma_conn_param *conn_param)
e51060f0
SH		 537{
538 struct ib_qp_attr qp_attr;
539 int qp_attr_mask, ret;
540
c5483388
SH		 541 mutex_lock(&id_priv->qp_mutex);
542 if (!id_priv->id.qp) {
543 ret = 0;
544 goto out;
545 }
e51060f0
SH		 546
547 /* Need to update QP attributes from default values. */
548 qp_attr.qp_state = IB_QPS_INIT;
c5483388 549 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 550 if (ret)
c5483388 551 goto out;
e51060f0 552
c5483388 553 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
e51060f0 554 if (ret)
c5483388 555 goto out;
e51060f0
SH		 556
557 qp_attr.qp_state = IB_QPS_RTR;
c5483388 558 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 559 if (ret)
c5483388 560 goto out;
e51060f0 561
5851bb89
SH		 562 if (conn_param)
563 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
c5483388
SH		 564 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
565out:
566 mutex_unlock(&id_priv->qp_mutex);
567 return ret;
e51060f0
SH		 568}
569
5851bb89
SH		 570static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
571 struct rdma_conn_param *conn_param)
e51060f0
SH		 572{
573 struct ib_qp_attr qp_attr;
574 int qp_attr_mask, ret;
575
c5483388
SH		 576 mutex_lock(&id_priv->qp_mutex);
577 if (!id_priv->id.qp) {
578 ret = 0;
579 goto out;
580 }
e51060f0
SH		 581
582 qp_attr.qp_state = IB_QPS_RTS;
c5483388 583 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 584 if (ret)
c5483388 585 goto out;
e51060f0 586
5851bb89
SH		 587 if (conn_param)
588 qp_attr.max_rd_atomic = conn_param->initiator_depth;
c5483388
SH		 589 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
590out:
591 mutex_unlock(&id_priv->qp_mutex);
592 return ret;
e51060f0
SH		 593}
594
c5483388 595static int cma_modify_qp_err(struct rdma_id_private *id_priv)
e51060f0
SH		 596{
597 struct ib_qp_attr qp_attr;
c5483388 598 int ret;
e51060f0 599
c5483388
SH		 600 mutex_lock(&id_priv->qp_mutex);
601 if (!id_priv->id.qp) {
602 ret = 0;
603 goto out;
604 }
e51060f0
SH		 605
606 qp_attr.qp_state = IB_QPS_ERR;
c5483388
SH		 607 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
608out:
609 mutex_unlock(&id_priv->qp_mutex);
610 return ret;
e51060f0
SH		 611}
612
c8f6a362
SH		 613static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
614 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
615{
616 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
617 int ret;
3c86aa70
EC		 618 u16 pkey;
619
620 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
621 IB_LINK_LAYER_INFINIBAND)
622 pkey = ib_addr_get_pkey(dev_addr);
623 else
624 pkey = 0xffff;
c8f6a362
SH		 625
626 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
3c86aa70 627 pkey, &qp_attr->pkey_index);
c8f6a362
SH		 628 if (ret)
629 return ret;
630
631 qp_attr->port_num = id_priv->id.port_num;
632 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
633
634 if (cma_is_ud_ps(id_priv->id.ps)) {
d2ca39f2
YE		 635 ret = cma_set_qkey(id_priv);
636 if (ret)
637 return ret;
638
c8f6a362
SH		 639 qp_attr->qkey = id_priv->qkey;
640 *qp_attr_mask |= IB_QP_QKEY;
641 } else {
642 qp_attr->qp_access_flags = 0;
643 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
644 }
645 return 0;
646}
647
e51060f0
SH		 648int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
649 int *qp_attr_mask)
650{
651 struct rdma_id_private *id_priv;
c8f6a362 652 int ret = 0;
e51060f0
SH		 653
654 id_priv = container_of(id, struct rdma_id_private, id);
07ebafba
TT		 655 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
656 case RDMA_TRANSPORT_IB:
c8f6a362
SH		 657 if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps))
658 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
659 else
660 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
661 qp_attr_mask);
e51060f0
SH		 662 if (qp_attr->qp_state == IB_QPS_RTR)
663 qp_attr->rq_psn = id_priv->seq_num;
664 break;
07ebafba 665 case RDMA_TRANSPORT_IWARP:
c8f6a362 666 if (!id_priv->cm_id.iw) {
8f076531 667 qp_attr->qp_access_flags = 0;
c8f6a362
SH		 668 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
669 } else
670 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
671 qp_attr_mask);
07ebafba 672 break;
e51060f0
SH		 673 default:
674 ret = -ENOSYS;
675 break;
676 }
677
678 return ret;
679}
680EXPORT_SYMBOL(rdma_init_qp_attr);
681
682static inline int cma_zero_addr(struct sockaddr *addr)
683{
684 struct in6_addr *ip6;
685
686 if (addr->sa_family == AF_INET)
6360a02a
JP		 687 return ipv4_is_zeronet(
688 ((struct sockaddr_in *)addr)->sin_addr.s_addr);
e51060f0
SH		 689 else {
690 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
691 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
5fd571cb 692 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
e51060f0
SH		 693 }
694}
695
696static inline int cma_loopback_addr(struct sockaddr *addr)
697{
1f5175ad
AS		 698 if (addr->sa_family == AF_INET)
699 return ipv4_is_loopback(
700 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
701 else
702 return ipv6_addr_loopback(
703 &((struct sockaddr_in6 *) addr)->sin6_addr);
e51060f0
SH		 704}
705
706static inline int cma_any_addr(struct sockaddr *addr)
707{
708 return cma_zero_addr(addr) || cma_loopback_addr(addr);
709}
710
628e5f6d
SH		 711static inline __be16 cma_port(struct sockaddr *addr)
712{
713 if (addr->sa_family == AF_INET)
714 return ((struct sockaddr_in *) addr)->sin_port;
715 else
716 return ((struct sockaddr_in6 *) addr)->sin6_port;
717}
718
e51060f0
SH		 719static inline int cma_any_port(struct sockaddr *addr)
720{
628e5f6d 721 return !cma_port(addr);
e51060f0
SH		 722}
723
724static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
1b90c137 725 u8 *ip_ver, __be16 *port,
e51060f0
SH		 726 union cma_ip_addr **src, union cma_ip_addr **dst)
727{
728 switch (ps) {
729 case RDMA_PS_SDP:
730 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
731 SDP_MAJ_VERSION)
732 return -EINVAL;
733
734 *ip_ver = sdp_get_ip_ver(hdr);
735 *port = ((struct sdp_hh *) hdr)->port;
736 *src = &((struct sdp_hh *) hdr)->src_addr;
737 *dst = &((struct sdp_hh *) hdr)->dst_addr;
738 break;
739 default:
740 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
741 return -EINVAL;
742
743 *ip_ver = cma_get_ip_ver(hdr);
744 *port = ((struct cma_hdr *) hdr)->port;
745 *src = &((struct cma_hdr *) hdr)->src_addr;
746 *dst = &((struct cma_hdr *) hdr)->dst_addr;
747 break;
748 }
749
750 if (*ip_ver != 4 && *ip_ver != 6)
751 return -EINVAL;
752 return 0;
753}
754
755static void cma_save_net_info(struct rdma_addr *addr,
756 struct rdma_addr *listen_addr,
1b90c137 757 u8 ip_ver, __be16 port,
e51060f0
SH		 758 union cma_ip_addr *src, union cma_ip_addr *dst)
759{
760 struct sockaddr_in *listen4, *ip4;
761 struct sockaddr_in6 *listen6, *ip6;
762
763 switch (ip_ver) {
764 case 4:
765 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
766 ip4 = (struct sockaddr_in *) &addr->src_addr;
767 ip4->sin_family = listen4->sin_family;
768 ip4->sin_addr.s_addr = dst->ip4.addr;
769 ip4->sin_port = listen4->sin_port;
770
771 ip4 = (struct sockaddr_in *) &addr->dst_addr;
772 ip4->sin_family = listen4->sin_family;
773 ip4->sin_addr.s_addr = src->ip4.addr;
774 ip4->sin_port = port;
775 break;
776 case 6:
777 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
778 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
779 ip6->sin6_family = listen6->sin6_family;
780 ip6->sin6_addr = dst->ip6;
781 ip6->sin6_port = listen6->sin6_port;
782
783 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
784 ip6->sin6_family = listen6->sin6_family;
785 ip6->sin6_addr = src->ip6;
786 ip6->sin6_port = port;
787 break;
788 default:
789 break;
790 }
791}
792
793static inline int cma_user_data_offset(enum rdma_port_space ps)
794{
795 switch (ps) {
796 case RDMA_PS_SDP:
797 return 0;
798 default:
799 return sizeof(struct cma_hdr);
800 }
801}
802
e51060f0
SH		 803static void cma_cancel_route(struct rdma_id_private *id_priv)
804{
3c86aa70
EC		 805 switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
806 case IB_LINK_LAYER_INFINIBAND:
e51060f0
SH		 807 if (id_priv->query)
808 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
809 break;
810 default:
811 break;
812 }
813}
814
e51060f0
SH		 815static void cma_cancel_listens(struct rdma_id_private *id_priv)
816{
817 struct rdma_id_private *dev_id_priv;
818
d02d1f53
SH		 819 /*
820 * Remove from listen_any_list to prevent added devices from spawning
821 * additional listen requests.
822 */
e51060f0
SH		 823 mutex_lock(&lock);
824 list_del(&id_priv->list);
825
826 while (!list_empty(&id_priv->listen_list)) {
827 dev_id_priv = list_entry(id_priv->listen_list.next,
828 struct rdma_id_private, listen_list);
d02d1f53
SH		 829 /* sync with device removal to avoid duplicate destruction */
830 list_del_init(&dev_id_priv->list);
831 list_del(&dev_id_priv->listen_list);
832 mutex_unlock(&lock);
833
834 rdma_destroy_id(&dev_id_priv->id);
835 mutex_lock(&lock);
e51060f0
SH		 836 }
837 mutex_unlock(&lock);
838}
839
840static void cma_cancel_operation(struct rdma_id_private *id_priv,
841 enum cma_state state)
842{
843 switch (state) {
844 case CMA_ADDR_QUERY:
845 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
846 break;
847 case CMA_ROUTE_QUERY:
848 cma_cancel_route(id_priv);
849 break;
850 case CMA_LISTEN:
3f446754
RD		 851 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr)
852 && !id_priv->cma_dev)
e51060f0
SH		 853 cma_cancel_listens(id_priv);
854 break;
855 default:
856 break;
857 }
858}
859
860static void cma_release_port(struct rdma_id_private *id_priv)
861{
862 struct rdma_bind_list *bind_list = id_priv->bind_list;
863
864 if (!bind_list)
865 return;
866
867 mutex_lock(&lock);
868 hlist_del(&id_priv->node);
869 if (hlist_empty(&bind_list->owners)) {
870 idr_remove(bind_list->ps, bind_list->port);
871 kfree(bind_list);
872 }
873 mutex_unlock(&lock);
874}
875
c8f6a362
SH		 876static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
877{
878 struct cma_multicast *mc;
879
880 while (!list_empty(&id_priv->mc_list)) {
881 mc = container_of(id_priv->mc_list.next,
882 struct cma_multicast, list);
883 list_del(&mc->list);
3c86aa70
EC		 884 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
885 case IB_LINK_LAYER_INFINIBAND:
886 ib_sa_free_multicast(mc->multicast.ib);
887 kfree(mc);
888 break;
889 case IB_LINK_LAYER_ETHERNET:
890 kref_put(&mc->mcref, release_mc);
891 break;
892 default:
893 break;
894 }
c8f6a362
SH		 895 }
896}
897
e51060f0
SH		 898void rdma_destroy_id(struct rdma_cm_id *id)
899{
900 struct rdma_id_private *id_priv;
901 enum cma_state state;
902
903 id_priv = container_of(id, struct rdma_id_private, id);
904 state = cma_exch(id_priv, CMA_DESTROYING);
905 cma_cancel_operation(id_priv, state);
906
61a73c70 907 mutex_lock(&lock);
e51060f0 908 if (id_priv->cma_dev) {
61a73c70 909 mutex_unlock(&lock);
3c86aa70 910 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
07ebafba 911 case RDMA_TRANSPORT_IB:
3cd96564 912 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib))
e51060f0
SH		 913 ib_destroy_cm_id(id_priv->cm_id.ib);
914 break;
07ebafba
TT		 915 case RDMA_TRANSPORT_IWARP:
916 if (id_priv->cm_id.iw && !IS_ERR(id_priv->cm_id.iw))
917 iw_destroy_cm_id(id_priv->cm_id.iw);
918 break;
e51060f0
SH		 919 default:
920 break;
921 }
c8f6a362 922 cma_leave_mc_groups(id_priv);
3cd96564 923 mutex_lock(&lock);
e51060f0 924 cma_detach_from_dev(id_priv);
e51060f0 925 }
61a73c70 926 mutex_unlock(&lock);
e51060f0
SH		 927
928 cma_release_port(id_priv);
929 cma_deref_id(id_priv);
930 wait_for_completion(&id_priv->comp);
931
d02d1f53
SH		 932 if (id_priv->internal_id)
933 cma_deref_id(id_priv->id.context);
934
e51060f0
SH		 935 kfree(id_priv->id.route.path_rec);
936 kfree(id_priv);
937}
938EXPORT_SYMBOL(rdma_destroy_id);
939
940static int cma_rep_recv(struct rdma_id_private *id_priv)
941{
942 int ret;
943
5851bb89 944 ret = cma_modify_qp_rtr(id_priv, NULL);
e51060f0
SH		 945 if (ret)
946 goto reject;
947
5851bb89 948 ret = cma_modify_qp_rts(id_priv, NULL);
e51060f0
SH		 949 if (ret)
950 goto reject;
951
952 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
953 if (ret)
954 goto reject;
955
956 return 0;
957reject:
c5483388 958 cma_modify_qp_err(id_priv);
e51060f0
SH		 959 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
960 NULL, 0, NULL, 0);
961 return ret;
962}
963
964static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
965{
966 if (id_priv->id.ps == RDMA_PS_SDP &&
967 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
968 SDP_MAJ_VERSION)
969 return -EINVAL;
970
971 return 0;
972}
973
a1b1b61f
SH		 974static void cma_set_rep_event_data(struct rdma_cm_event *event,
975 struct ib_cm_rep_event_param *rep_data,
976 void *private_data)
977{
978 event->param.conn.private_data = private_data;
979 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
980 event->param.conn.responder_resources = rep_data->responder_resources;
981 event->param.conn.initiator_depth = rep_data->initiator_depth;
982 event->param.conn.flow_control = rep_data->flow_control;
983 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
984 event->param.conn.srq = rep_data->srq;
985 event->param.conn.qp_num = rep_data->remote_qpn;
986}
987
e51060f0
SH		 988static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
989{
990 struct rdma_id_private *id_priv = cm_id->context;
a1b1b61f
SH		 991 struct rdma_cm_event event;
992 int ret = 0;
e51060f0 993
38ca83a5
AV		 994 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
995 cma_disable_callback(id_priv, CMA_CONNECT)) ||
996 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
997 cma_disable_callback(id_priv, CMA_DISCONNECT)))
8aa08602 998 return 0;
e51060f0 999
a1b1b61f 1000 memset(&event, 0, sizeof event);
e51060f0
SH		 1001 switch (ib_event->event) {
1002 case IB_CM_REQ_ERROR:
1003 case IB_CM_REP_ERROR:
a1b1b61f
SH		 1004 event.event = RDMA_CM_EVENT_UNREACHABLE;
1005 event.status = -ETIMEDOUT;
e51060f0
SH		 1006 break;
1007 case IB_CM_REP_RECEIVED:
a1b1b61f
SH		 1008 event.status = cma_verify_rep(id_priv, ib_event->private_data);
1009 if (event.status)
1010 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
e51060f0 1011 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
a1b1b61f
SH		 1012 event.status = cma_rep_recv(id_priv);
1013 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1014 RDMA_CM_EVENT_ESTABLISHED;
e51060f0 1015 } else
a1b1b61f
SH		 1016 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1017 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1018 ib_event->private_data);
e51060f0
SH		 1019 break;
1020 case IB_CM_RTU_RECEIVED:
0fe313b0
SH		 1021 case IB_CM_USER_ESTABLISHED:
1022 event.event = RDMA_CM_EVENT_ESTABLISHED;
e51060f0
SH		 1023 break;
1024 case IB_CM_DREQ_ERROR:
a1b1b61f 1025 event.status = -ETIMEDOUT; /* fall through */
e51060f0
SH		 1026 case IB_CM_DREQ_RECEIVED:
1027 case IB_CM_DREP_RECEIVED:
1028 if (!cma_comp_exch(id_priv, CMA_CONNECT, CMA_DISCONNECT))
1029 goto out;
a1b1b61f 1030 event.event = RDMA_CM_EVENT_DISCONNECTED;
e51060f0
SH		 1031 break;
1032 case IB_CM_TIMEWAIT_EXIT:
38ca83a5
AV		 1033 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1034 break;
e51060f0
SH		 1035 case IB_CM_MRA_RECEIVED:
1036 /* ignore event */
1037 goto out;
1038 case IB_CM_REJ_RECEIVED:
c5483388 1039 cma_modify_qp_err(id_priv);
a1b1b61f
SH		 1040 event.status = ib_event->param.rej_rcvd.reason;
1041 event.event = RDMA_CM_EVENT_REJECTED;
1042 event.param.conn.private_data = ib_event->private_data;
1043 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
e51060f0
SH		 1044 break;
1045 default:
468f2239 1046 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
e51060f0
SH		 1047 ib_event->event);
1048 goto out;
1049 }
1050
a1b1b61f 1051 ret = id_priv->id.event_handler(&id_priv->id, &event);
e51060f0
SH		 1052 if (ret) {
1053 /* Destroy the CM ID by returning a non-zero value. */
1054 id_priv->cm_id.ib = NULL;
1055 cma_exch(id_priv, CMA_DESTROYING);
de910bd9 1056 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1057 rdma_destroy_id(&id_priv->id);
1058 return ret;
1059 }
1060out:
de910bd9 1061 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1062 return ret;
1063}
1064
628e5f6d
SH		 1065static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1066 struct ib_cm_event *ib_event)
e51060f0
SH		 1067{
1068 struct rdma_id_private *id_priv;
1069 struct rdma_cm_id *id;
1070 struct rdma_route *rt;
1071 union cma_ip_addr *src, *dst;
1b90c137 1072 __be16 port;
e51060f0 1073 u8 ip_ver;
64c5e613 1074 int ret;
e51060f0 1075
3f168d2b
KK		 1076 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1077 &ip_ver, &port, &src, &dst))
1078 goto err;
1079
e51060f0
SH		 1080 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1081 listen_id->ps);
1082 if (IS_ERR(id))
3f168d2b
KK		 1083 goto err;
1084
1085 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1086 ip_ver, port, src, dst);
e51060f0
SH		 1087
1088 rt = &id->route;
1089 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
3f168d2b
KK		 1090 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1091 GFP_KERNEL);
e51060f0 1092 if (!rt->path_rec)
3f168d2b 1093 goto destroy_id;
e51060f0 1094
e51060f0
SH		 1095 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1096 if (rt->num_paths == 2)
1097 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1098
6f8372b6
SH		 1099 if (cma_any_addr((struct sockaddr *) &rt->addr.src_addr)) {
1100 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1101 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1102 ib_addr_set_pkey(&rt->addr.dev_addr, rt->path_rec[0].pkey);
1103 } else {
1104 ret = rdma_translate_ip((struct sockaddr *) &rt->addr.src_addr,
1105 &rt->addr.dev_addr);
1106 if (ret)
1107 goto destroy_id;
1108 }
1109 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
e51060f0
SH		 1110
1111 id_priv = container_of(id, struct rdma_id_private, id);
1112 id_priv->state = CMA_CONNECT;
1113 return id_priv;
3f168d2b
KK		 1114
1115destroy_id:
e51060f0 1116 rdma_destroy_id(id);
3f168d2b 1117err:
e51060f0
SH		 1118 return NULL;
1119}
1120
628e5f6d
SH		 1121static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1122 struct ib_cm_event *ib_event)
1123{
1124 struct rdma_id_private *id_priv;
1125 struct rdma_cm_id *id;
1126 union cma_ip_addr *src, *dst;
1b90c137 1127 __be16 port;
628e5f6d
SH		 1128 u8 ip_ver;
1129 int ret;
1130
1131 id = rdma_create_id(listen_id->event_handler, listen_id->context,
1132 listen_id->ps);
1133 if (IS_ERR(id))
1134 return NULL;
1135
1136
1137 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1138 &ip_ver, &port, &src, &dst))
1139 goto err;
1140
1141 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1142 ip_ver, port, src, dst);
1143
6f8372b6
SH		 1144 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1145 ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr,
1146 &id->route.addr.dev_addr);
1147 if (ret)
1148 goto err;
1149 }
628e5f6d
SH		 1150
1151 id_priv = container_of(id, struct rdma_id_private, id);
1152 id_priv->state = CMA_CONNECT;
1153 return id_priv;
1154err:
1155 rdma_destroy_id(id);
1156 return NULL;
1157}
1158
a1b1b61f
SH		 1159static void cma_set_req_event_data(struct rdma_cm_event *event,
1160 struct ib_cm_req_event_param *req_data,
1161 void *private_data, int offset)
1162{
1163 event->param.conn.private_data = private_data + offset;
1164 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1165 event->param.conn.responder_resources = req_data->responder_resources;
1166 event->param.conn.initiator_depth = req_data->initiator_depth;
1167 event->param.conn.flow_control = req_data->flow_control;
1168 event->param.conn.retry_count = req_data->retry_count;
1169 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1170 event->param.conn.srq = req_data->srq;
1171 event->param.conn.qp_num = req_data->remote_qpn;
1172}
1173
e51060f0
SH		 1174static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1175{
1176 struct rdma_id_private *listen_id, *conn_id;
a1b1b61f 1177 struct rdma_cm_event event;
e51060f0
SH		 1178 int offset, ret;
1179
1180 listen_id = cm_id->context;
de910bd9 1181 if (cma_disable_callback(listen_id, CMA_LISTEN))
8aa08602 1182 return -ECONNABORTED;
e51060f0 1183
628e5f6d
SH		 1184 memset(&event, 0, sizeof event);
1185 offset = cma_user_data_offset(listen_id->id.ps);
1186 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
c8f6a362 1187 if (cma_is_ud_ps(listen_id->id.ps)) {
628e5f6d
SH		 1188 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1189 event.param.ud.private_data = ib_event->private_data + offset;
1190 event.param.ud.private_data_len =
1191 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1192 } else {
1193 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1194 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1195 ib_event->private_data, offset);
1196 }
e51060f0
SH		 1197 if (!conn_id) {
1198 ret = -ENOMEM;
1199 goto out;
1200 }
1201
de910bd9 1202 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
61a73c70 1203 mutex_lock(&lock);
07ebafba 1204 ret = cma_acquire_dev(conn_id);
61a73c70 1205 mutex_unlock(&lock);
a1a733f6
KK		 1206 if (ret)
1207 goto release_conn_id;
e51060f0
SH		 1208
1209 conn_id->cm_id.ib = cm_id;
1210 cm_id->context = conn_id;
1211 cm_id->cm_handler = cma_ib_handler;
1212
25ae21a1
SH		 1213 /*
1214 * Protect against the user destroying conn_id from another thread
1215 * until we're done accessing it.
1216 */
1217 atomic_inc(&conn_id->refcount);
a1b1b61f 1218 ret = conn_id->id.event_handler(&conn_id->id, &event);
45d9478d 1219 if (!ret) {
ead595ae
SH		 1220 /*
1221 * Acquire mutex to prevent user executing rdma_destroy_id()
1222 * while we're accessing the cm_id.
1223 */
1224 mutex_lock(&lock);
1225 if (cma_comp(conn_id, CMA_CONNECT) &&
1226 !cma_is_ud_ps(conn_id->id.ps))
1227 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1228 mutex_unlock(&lock);
de910bd9 1229 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1230 cma_deref_id(conn_id);
a1a733f6 1231 goto out;
45d9478d 1232 }
25ae21a1 1233 cma_deref_id(conn_id);
a1a733f6
KK		 1234
1235 /* Destroy the CM ID by returning a non-zero value. */
1236 conn_id->cm_id.ib = NULL;
1237
1238release_conn_id:
1239 cma_exch(conn_id, CMA_DESTROYING);
de910bd9 1240 mutex_unlock(&conn_id->handler_mutex);
a1a733f6
KK		 1241 rdma_destroy_id(&conn_id->id);
1242
e51060f0 1243out:
de910bd9 1244 mutex_unlock(&listen_id->handler_mutex);
e51060f0
SH		 1245 return ret;
1246}
1247
1248static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1249{
628e5f6d 1250 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
e51060f0
SH		 1251}
1252
1253static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1254 struct ib_cm_compare_data *compare)
1255{
1256 struct cma_hdr *cma_data, *cma_mask;
1257 struct sdp_hh *sdp_data, *sdp_mask;
1b90c137 1258 __be32 ip4_addr;
e51060f0
SH		 1259 struct in6_addr ip6_addr;
1260
1261 memset(compare, 0, sizeof *compare);
1262 cma_data = (void *) compare->data;
1263 cma_mask = (void *) compare->mask;
1264 sdp_data = (void *) compare->data;
1265 sdp_mask = (void *) compare->mask;
1266
1267 switch (addr->sa_family) {
1268 case AF_INET:
1269 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1270 if (ps == RDMA_PS_SDP) {
1271 sdp_set_ip_ver(sdp_data, 4);
1272 sdp_set_ip_ver(sdp_mask, 0xF);
1273 sdp_data->dst_addr.ip4.addr = ip4_addr;
1b90c137 1274 sdp_mask->dst_addr.ip4.addr = htonl(~0);
e51060f0
SH		 1275 } else {
1276 cma_set_ip_ver(cma_data, 4);
1277 cma_set_ip_ver(cma_mask, 0xF);
1278 cma_data->dst_addr.ip4.addr = ip4_addr;
1b90c137 1279 cma_mask->dst_addr.ip4.addr = htonl(~0);
e51060f0
SH		 1280 }
1281 break;
1282 case AF_INET6:
1283 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1284 if (ps == RDMA_PS_SDP) {
1285 sdp_set_ip_ver(sdp_data, 6);
1286 sdp_set_ip_ver(sdp_mask, 0xF);
1287 sdp_data->dst_addr.ip6 = ip6_addr;
1288 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1289 sizeof sdp_mask->dst_addr.ip6);
1290 } else {
1291 cma_set_ip_ver(cma_data, 6);
1292 cma_set_ip_ver(cma_mask, 0xF);
1293 cma_data->dst_addr.ip6 = ip6_addr;
1294 memset(&cma_mask->dst_addr.ip6, 0xFF,
1295 sizeof cma_mask->dst_addr.ip6);
1296 }
1297 break;
1298 default:
1299 break;
1300 }
1301}
1302
07ebafba
TT		 1303static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1304{
1305 struct rdma_id_private *id_priv = iw_id->context;
a1b1b61f 1306 struct rdma_cm_event event;
07ebafba
TT		 1307 struct sockaddr_in *sin;
1308 int ret = 0;
1309
de910bd9 1310 if (cma_disable_callback(id_priv, CMA_CONNECT))
be65f086 1311 return 0;
07ebafba 1312
be65f086 1313 memset(&event, 0, sizeof event);
07ebafba
TT		 1314 switch (iw_event->event) {
1315 case IW_CM_EVENT_CLOSE:
a1b1b61f 1316 event.event = RDMA_CM_EVENT_DISCONNECTED;
07ebafba
TT		 1317 break;
1318 case IW_CM_EVENT_CONNECT_REPLY:
1319 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1320 *sin = iw_event->local_addr;
1321 sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
1322 *sin = iw_event->remote_addr;
881a045f
SW		 1323 switch (iw_event->status) {
1324 case 0:
a1b1b61f 1325 event.event = RDMA_CM_EVENT_ESTABLISHED;
881a045f
SW		 1326 break;
1327 case -ECONNRESET:
1328 case -ECONNREFUSED:
1329 event.event = RDMA_CM_EVENT_REJECTED;
1330 break;
1331 case -ETIMEDOUT:
1332 event.event = RDMA_CM_EVENT_UNREACHABLE;
1333 break;
1334 default:
1335 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1336 break;
1337 }
07ebafba
TT		 1338 break;
1339 case IW_CM_EVENT_ESTABLISHED:
a1b1b61f 1340 event.event = RDMA_CM_EVENT_ESTABLISHED;
07ebafba
TT		 1341 break;
1342 default:
1343 BUG_ON(1);
1344 }
1345
a1b1b61f
SH		 1346 event.status = iw_event->status;
1347 event.param.conn.private_data = iw_event->private_data;
1348 event.param.conn.private_data_len = iw_event->private_data_len;
1349 ret = id_priv->id.event_handler(&id_priv->id, &event);
07ebafba
TT		 1350 if (ret) {
1351 /* Destroy the CM ID by returning a non-zero value. */
1352 id_priv->cm_id.iw = NULL;
1353 cma_exch(id_priv, CMA_DESTROYING);
de910bd9 1354 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 1355 rdma_destroy_id(&id_priv->id);
1356 return ret;
1357 }
1358
de910bd9 1359 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 1360 return ret;
1361}
1362
1363static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1364 struct iw_cm_event *iw_event)
1365{
1366 struct rdma_cm_id *new_cm_id;
1367 struct rdma_id_private *listen_id, *conn_id;
1368 struct sockaddr_in *sin;
1369 struct net_device *dev = NULL;
a1b1b61f 1370 struct rdma_cm_event event;
07ebafba 1371 int ret;
8d8293cf 1372 struct ib_device_attr attr;
07ebafba
TT		 1373
1374 listen_id = cm_id->context;
de910bd9 1375 if (cma_disable_callback(listen_id, CMA_LISTEN))
8aa08602 1376 return -ECONNABORTED;
07ebafba
TT		 1377
1378 /* Create a new RDMA id for the new IW CM ID */
1379 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1380 listen_id->id.context,
1381 RDMA_PS_TCP);
10f32065 1382 if (IS_ERR(new_cm_id)) {
07ebafba
TT		 1383 ret = -ENOMEM;
1384 goto out;
1385 }
1386 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
de910bd9 1387 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
07ebafba
TT		 1388 conn_id->state = CMA_CONNECT;
1389
1ab35276 1390 dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
07ebafba
TT		 1391 if (!dev) {
1392 ret = -EADDRNOTAVAIL;
de910bd9 1393 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1394 rdma_destroy_id(new_cm_id);
1395 goto out;
1396 }
1397 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1398 if (ret) {
de910bd9 1399 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1400 rdma_destroy_id(new_cm_id);
1401 goto out;
1402 }
1403
61a73c70 1404 mutex_lock(&lock);
07ebafba 1405 ret = cma_acquire_dev(conn_id);
61a73c70 1406 mutex_unlock(&lock);
07ebafba 1407 if (ret) {
de910bd9 1408 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1409 rdma_destroy_id(new_cm_id);
1410 goto out;
1411 }
1412
1413 conn_id->cm_id.iw = cm_id;
1414 cm_id->context = conn_id;
1415 cm_id->cm_handler = cma_iw_handler;
1416
1417 sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
1418 *sin = iw_event->local_addr;
1419 sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
1420 *sin = iw_event->remote_addr;
1421
8d8293cf
SW		 1422 ret = ib_query_device(conn_id->id.device, &attr);
1423 if (ret) {
de910bd9 1424 mutex_unlock(&conn_id->handler_mutex);
8d8293cf
SW		 1425 rdma_destroy_id(new_cm_id);
1426 goto out;
1427 }
1428
a1b1b61f
SH		 1429 memset(&event, 0, sizeof event);
1430 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1431 event.param.conn.private_data = iw_event->private_data;
1432 event.param.conn.private_data_len = iw_event->private_data_len;
8d8293cf
SW		 1433 event.param.conn.initiator_depth = attr.max_qp_init_rd_atom;
1434 event.param.conn.responder_resources = attr.max_qp_rd_atom;
25ae21a1
SH		 1435
1436 /*
1437 * Protect against the user destroying conn_id from another thread
1438 * until we're done accessing it.
1439 */
1440 atomic_inc(&conn_id->refcount);
a1b1b61f 1441 ret = conn_id->id.event_handler(&conn_id->id, &event);
07ebafba
TT		 1442 if (ret) {
1443 /* User wants to destroy the CM ID */
1444 conn_id->cm_id.iw = NULL;
1445 cma_exch(conn_id, CMA_DESTROYING);
de910bd9 1446 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1447 cma_deref_id(conn_id);
07ebafba 1448 rdma_destroy_id(&conn_id->id);
de910bd9 1449 goto out;
07ebafba
TT		 1450 }
1451
de910bd9 1452 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1453 cma_deref_id(conn_id);
de910bd9 1454
07ebafba
TT		 1455out:
1456 if (dev)
1457 dev_put(dev);
de910bd9 1458 mutex_unlock(&listen_id->handler_mutex);
07ebafba
TT		 1459 return ret;
1460}
1461
e51060f0
SH		 1462static int cma_ib_listen(struct rdma_id_private *id_priv)
1463{
1464 struct ib_cm_compare_data compare_data;
1465 struct sockaddr *addr;
1466 __be64 svc_id;
1467 int ret;
1468
1469 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_req_handler,
1470 id_priv);
1471 if (IS_ERR(id_priv->cm_id.ib))
1472 return PTR_ERR(id_priv->cm_id.ib);
1473
3f446754 1474 addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
e51060f0
SH		 1475 svc_id = cma_get_service_id(id_priv->id.ps, addr);
1476 if (cma_any_addr(addr))
1477 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1478 else {
1479 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1480 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1481 }
1482
1483 if (ret) {
1484 ib_destroy_cm_id(id_priv->cm_id.ib);
1485 id_priv->cm_id.ib = NULL;
1486 }
1487
1488 return ret;
1489}
1490
07ebafba
TT		 1491static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1492{
1493 int ret;
1494 struct sockaddr_in *sin;
1495
1496 id_priv->cm_id.iw = iw_create_cm_id(id_priv->id.device,
1497 iw_conn_req_handler,
1498 id_priv);
1499 if (IS_ERR(id_priv->cm_id.iw))
1500 return PTR_ERR(id_priv->cm_id.iw);
1501
1502 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1503 id_priv->cm_id.iw->local_addr = *sin;
1504
1505 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1506
1507 if (ret) {
1508 iw_destroy_cm_id(id_priv->cm_id.iw);
1509 id_priv->cm_id.iw = NULL;
1510 }
1511
1512 return ret;
1513}
1514
e51060f0
SH		 1515static int cma_listen_handler(struct rdma_cm_id *id,
1516 struct rdma_cm_event *event)
1517{
1518 struct rdma_id_private *id_priv = id->context;
1519
1520 id->context = id_priv->id.context;
1521 id->event_handler = id_priv->id.event_handler;
1522 return id_priv->id.event_handler(id, event);
1523}
1524
1525static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1526 struct cma_device *cma_dev)
1527{
1528 struct rdma_id_private *dev_id_priv;
1529 struct rdma_cm_id *id;
1530 int ret;
1531
1532 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps);
1533 if (IS_ERR(id))
1534 return;
1535
1536 dev_id_priv = container_of(id, struct rdma_id_private, id);
1537
1538 dev_id_priv->state = CMA_ADDR_BOUND;
1539 memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
3f446754 1540 ip_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
e51060f0
SH		 1541
1542 cma_attach_to_dev(dev_id_priv, cma_dev);
1543 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
d02d1f53
SH		 1544 atomic_inc(&id_priv->refcount);
1545 dev_id_priv->internal_id = 1;
e51060f0
SH		 1546
1547 ret = rdma_listen(id, id_priv->backlog);
1548 if (ret)
d02d1f53 1549 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
468f2239 1550 "listening on device %s\n", ret, cma_dev->device->name);
e51060f0
SH		 1551}
1552
1553static void cma_listen_on_all(struct rdma_id_private *id_priv)
1554{
1555 struct cma_device *cma_dev;
1556
1557 mutex_lock(&lock);
1558 list_add_tail(&id_priv->list, &listen_any_list);
1559 list_for_each_entry(cma_dev, &dev_list, list)
1560 cma_listen_on_dev(id_priv, cma_dev);
1561 mutex_unlock(&lock);
1562}
1563
e51060f0
SH		 1564int rdma_listen(struct rdma_cm_id *id, int backlog)
1565{
1566 struct rdma_id_private *id_priv;
1567 int ret;
1568
1569 id_priv = container_of(id, struct rdma_id_private, id);
1570 if (id_priv->state == CMA_IDLE) {
d14714df
SH		 1571 ((struct sockaddr *) &id->route.addr.src_addr)->sa_family = AF_INET;
1572 ret = rdma_bind_addr(id, (struct sockaddr *) &id->route.addr.src_addr);
e51060f0
SH		 1573 if (ret)
1574 return ret;
1575 }
1576
1577 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
1578 return -EINVAL;
1579
1580 id_priv->backlog = backlog;
1581 if (id->device) {
07ebafba
TT		 1582 switch (rdma_node_get_transport(id->device->node_type)) {
1583 case RDMA_TRANSPORT_IB:
e51060f0
SH		 1584 ret = cma_ib_listen(id_priv);
1585 if (ret)
1586 goto err;
1587 break;
07ebafba
TT		 1588 case RDMA_TRANSPORT_IWARP:
1589 ret = cma_iw_listen(id_priv, backlog);
1590 if (ret)
1591 goto err;
1592 break;
e51060f0
SH		 1593 default:
1594 ret = -ENOSYS;
1595 goto err;
1596 }
1597 } else
1598 cma_listen_on_all(id_priv);
1599
1600 return 0;
1601err:
1602 id_priv->backlog = 0;
1603 cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
1604 return ret;
1605}
1606EXPORT_SYMBOL(rdma_listen);
1607
a81c994d
SH		 1608void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1609{
1610 struct rdma_id_private *id_priv;
1611
1612 id_priv = container_of(id, struct rdma_id_private, id);
1613 id_priv->tos = (u8) tos;
1614}
1615EXPORT_SYMBOL(rdma_set_service_type);
1616
e51060f0
SH		 1617static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1618 void *context)
1619{
1620 struct cma_work *work = context;
1621 struct rdma_route *route;
1622
1623 route = &work->id->id.route;
1624
1625 if (!status) {
1626 route->num_paths = 1;
1627 *route->path_rec = *path_rec;
1628 } else {
1629 work->old_state = CMA_ROUTE_QUERY;
1630 work->new_state = CMA_ADDR_RESOLVED;
1631 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
8f0472d3 1632 work->event.status = status;
e51060f0
SH		 1633 }
1634
1635 queue_work(cma_wq, &work->work);
1636}
1637
1638static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1639 struct cma_work *work)
1640{
a81c994d 1641 struct rdma_addr *addr = &id_priv->id.route.addr;
e51060f0 1642 struct ib_sa_path_rec path_rec;
a81c994d
SH		 1643 ib_sa_comp_mask comp_mask;
1644 struct sockaddr_in6 *sin6;
e51060f0
SH		 1645
1646 memset(&path_rec, 0, sizeof path_rec);
6f8372b6
SH		 1647 rdma_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
1648 rdma_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
a81c994d 1649 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
e51060f0 1650 path_rec.numb_path = 1;
962063e6 1651 path_rec.reversible = 1;
3f446754
RD		 1652 path_rec.service_id = cma_get_service_id(id_priv->id.ps,
1653 (struct sockaddr *) &addr->dst_addr);
a81c994d
SH		 1654
1655 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1656 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1657 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1658
3f446754 1659 if (addr->src_addr.ss_family == AF_INET) {
a81c994d
SH		 1660 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1661 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1662 } else {
1663 sin6 = (struct sockaddr_in6 *) &addr->src_addr;
1664 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1665 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1666 }
e51060f0 1667
c1a0b23b 1668 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
a81c994d
SH		 1669 id_priv->id.port_num, &path_rec,
1670 comp_mask, timeout_ms,
1671 GFP_KERNEL, cma_query_handler,
1672 work, &id_priv->query);
e51060f0
SH		 1673
1674 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1675}
1676
c4028958 1677static void cma_work_handler(struct work_struct *_work)
e51060f0 1678{
c4028958 1679 struct cma_work *work = container_of(_work, struct cma_work, work);
e51060f0
SH		 1680 struct rdma_id_private *id_priv = work->id;
1681 int destroy = 0;
1682
de910bd9 1683 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 1684 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1685 goto out;
1686
1687 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1688 cma_exch(id_priv, CMA_DESTROYING);
1689 destroy = 1;
1690 }
1691out:
de910bd9 1692 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1693 cma_deref_id(id_priv);
1694 if (destroy)
1695 rdma_destroy_id(&id_priv->id);
1696 kfree(work);
1697}
1698
dd5bdff8
OG		 1699static void cma_ndev_work_handler(struct work_struct *_work)
1700{
1701 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1702 struct rdma_id_private *id_priv = work->id;
1703 int destroy = 0;
1704
1705 mutex_lock(&id_priv->handler_mutex);
1706 if (id_priv->state == CMA_DESTROYING ||
1707 id_priv->state == CMA_DEVICE_REMOVAL)
1708 goto out;
1709
1710 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1711 cma_exch(id_priv, CMA_DESTROYING);
1712 destroy = 1;
1713 }
1714
1715out:
1716 mutex_unlock(&id_priv->handler_mutex);
1717 cma_deref_id(id_priv);
1718 if (destroy)
1719 rdma_destroy_id(&id_priv->id);
1720 kfree(work);
1721}
1722
e51060f0
SH		 1723static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1724{
1725 struct rdma_route *route = &id_priv->id.route;
1726 struct cma_work *work;
1727 int ret;
1728
1729 work = kzalloc(sizeof *work, GFP_KERNEL);
1730 if (!work)
1731 return -ENOMEM;
1732
1733 work->id = id_priv;
c4028958 1734 INIT_WORK(&work->work, cma_work_handler);
e51060f0
SH		 1735 work->old_state = CMA_ROUTE_QUERY;
1736 work->new_state = CMA_ROUTE_RESOLVED;
1737 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1738
1739 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1740 if (!route->path_rec) {
1741 ret = -ENOMEM;
1742 goto err1;
1743 }
1744
1745 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1746 if (ret)
1747 goto err2;
1748
1749 return 0;
1750err2:
1751 kfree(route->path_rec);
1752 route->path_rec = NULL;
1753err1:
1754 kfree(work);
1755 return ret;
1756}
1757
1758int rdma_set_ib_paths(struct rdma_cm_id *id,
1759 struct ib_sa_path_rec *path_rec, int num_paths)
1760{
1761 struct rdma_id_private *id_priv;
1762 int ret;
1763
1764 id_priv = container_of(id, struct rdma_id_private, id);
1765 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_RESOLVED))
1766 return -EINVAL;
1767
9893e742
JL		 1768 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1769 GFP_KERNEL);
e51060f0
SH		 1770 if (!id->route.path_rec) {
1771 ret = -ENOMEM;
1772 goto err;
1773 }
1774
ae2d9293 1775 id->route.num_paths = num_paths;
e51060f0
SH		 1776 return 0;
1777err:
1778 cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_ADDR_RESOLVED);
1779 return ret;
1780}
1781EXPORT_SYMBOL(rdma_set_ib_paths);
1782
07ebafba
TT		 1783static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1784{
1785 struct cma_work *work;
1786
1787 work = kzalloc(sizeof *work, GFP_KERNEL);
1788 if (!work)
1789 return -ENOMEM;
1790
1791 work->id = id_priv;
c4028958 1792 INIT_WORK(&work->work, cma_work_handler);
07ebafba
TT		 1793 work->old_state = CMA_ROUTE_QUERY;
1794 work->new_state = CMA_ROUTE_RESOLVED;
1795 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1796 queue_work(cma_wq, &work->work);
1797 return 0;
1798}
1799
3c86aa70
EC		 1800static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1801{
1802 struct rdma_route *route = &id_priv->id.route;
1803 struct rdma_addr *addr = &route->addr;
1804 struct cma_work *work;
1805 int ret;
1806 struct sockaddr_in *src_addr = (struct sockaddr_in *)&route->addr.src_addr;
1807 struct sockaddr_in *dst_addr = (struct sockaddr_in *)&route->addr.dst_addr;
1808 struct net_device *ndev = NULL;
af7bd463 1809 u16 vid;
3c86aa70
EC		 1810
1811 if (src_addr->sin_family != dst_addr->sin_family)
1812 return -EINVAL;
1813
1814 work = kzalloc(sizeof *work, GFP_KERNEL);
1815 if (!work)
1816 return -ENOMEM;
1817
1818 work->id = id_priv;
1819 INIT_WORK(&work->work, cma_work_handler);
1820
1821 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1822 if (!route->path_rec) {
1823 ret = -ENOMEM;
1824 goto err1;
1825 }
1826
1827 route->num_paths = 1;
1828
3c86aa70
EC		 1829 if (addr->dev_addr.bound_dev_if)
1830 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1831 if (!ndev) {
1832 ret = -ENODEV;
1833 goto err2;
1834 }
1835
af7bd463
EC		 1836 vid = rdma_vlan_dev_vlan_id(ndev);
1837
1838 iboe_mac_vlan_to_ll(&route->path_rec->sgid, addr->dev_addr.src_dev_addr, vid);
1839 iboe_mac_vlan_to_ll(&route->path_rec->dgid, addr->dev_addr.dst_dev_addr, vid);
1840
1841 route->path_rec->hop_limit = 1;
1842 route->path_rec->reversible = 1;
1843 route->path_rec->pkey = cpu_to_be16(0xffff);
1844 route->path_rec->mtu_selector = IB_SA_EQ;
1845 route->path_rec->sl = id_priv->tos >> 5;
1846
3c86aa70
EC		 1847 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1848 route->path_rec->rate_selector = IB_SA_EQ;
1849 route->path_rec->rate = iboe_get_rate(ndev);
1850 dev_put(ndev);
1851 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1852 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1853 if (!route->path_rec->mtu) {
1854 ret = -EINVAL;
1855 goto err2;
1856 }
1857
1858 work->old_state = CMA_ROUTE_QUERY;
1859 work->new_state = CMA_ROUTE_RESOLVED;
1860 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1861 work->event.status = 0;
1862
1863 queue_work(cma_wq, &work->work);
1864
1865 return 0;
1866
1867err2:
1868 kfree(route->path_rec);
1869 route->path_rec = NULL;
1870err1:
1871 kfree(work);
1872 return ret;
1873}
1874
e51060f0
SH		 1875int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1876{
1877 struct rdma_id_private *id_priv;
1878 int ret;
1879
1880 id_priv = container_of(id, struct rdma_id_private, id);
1881 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ROUTE_QUERY))
1882 return -EINVAL;
1883
1884 atomic_inc(&id_priv->refcount);
07ebafba
TT		 1885 switch (rdma_node_get_transport(id->device->node_type)) {
1886 case RDMA_TRANSPORT_IB:
3c86aa70
EC		 1887 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1888 case IB_LINK_LAYER_INFINIBAND:
1889 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1890 break;
1891 case IB_LINK_LAYER_ETHERNET:
1892 ret = cma_resolve_iboe_route(id_priv);
1893 break;
1894 default:
1895 ret = -ENOSYS;
1896 }
e51060f0 1897 break;
07ebafba
TT		 1898 case RDMA_TRANSPORT_IWARP:
1899 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1900 break;
e51060f0
SH		 1901 default:
1902 ret = -ENOSYS;
1903 break;
1904 }
1905 if (ret)
1906 goto err;
1907
1908 return 0;
1909err:
1910 cma_comp_exch(id_priv, CMA_ROUTE_QUERY, CMA_ADDR_RESOLVED);
1911 cma_deref_id(id_priv);
1912 return ret;
1913}
1914EXPORT_SYMBOL(rdma_resolve_route);
1915
1916static int cma_bind_loopback(struct rdma_id_private *id_priv)
1917{
1918 struct cma_device *cma_dev;
1919 struct ib_port_attr port_attr;
f0ee3404 1920 union ib_gid gid;
e51060f0
SH		 1921 u16 pkey;
1922 int ret;
1923 u8 p;
1924
1925 mutex_lock(&lock);
e82153b5
KK		 1926 if (list_empty(&dev_list)) {
1927 ret = -ENODEV;
1928 goto out;
1929 }
e51060f0
SH		 1930 list_for_each_entry(cma_dev, &dev_list, list)
1931 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
e82153b5 1932 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
e51060f0
SH		 1933 port_attr.state == IB_PORT_ACTIVE)
1934 goto port_found;
1935
e82153b5
KK		 1936 p = 1;
1937 cma_dev = list_entry(dev_list.next, struct cma_device, list);
e51060f0
SH		 1938
1939port_found:
f0ee3404 1940 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
e51060f0
SH		 1941 if (ret)
1942 goto out;
1943
1944 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
1945 if (ret)
1946 goto out;
1947
6f8372b6 1948 id_priv->id.route.addr.dev_addr.dev_type =
3c86aa70 1949 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
6f8372b6
SH		 1950 ARPHRD_INFINIBAND : ARPHRD_ETHER;
1951
1952 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0
SH		 1953 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1954 id_priv->id.port_num = p;
1955 cma_attach_to_dev(id_priv, cma_dev);
1956out:
1957 mutex_unlock(&lock);
1958 return ret;
1959}
1960
1961static void addr_handler(int status, struct sockaddr *src_addr,
1962 struct rdma_dev_addr *dev_addr, void *context)
1963{
1964 struct rdma_id_private *id_priv = context;
a1b1b61f 1965 struct rdma_cm_event event;
e51060f0 1966
a1b1b61f 1967 memset(&event, 0, sizeof event);
de910bd9 1968 mutex_lock(&id_priv->handler_mutex);
61a73c70
SH		 1969
1970 /*
1971 * Grab mutex to block rdma_destroy_id() from removing the device while
1972 * we're trying to acquire it.
1973 */
1974 mutex_lock(&lock);
1975 if (!cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_RESOLVED)) {
1976 mutex_unlock(&lock);
1977 goto out;
1978 }
1979
1980 if (!status && !id_priv->cma_dev)
e51060f0 1981 status = cma_acquire_dev(id_priv);
61a73c70 1982 mutex_unlock(&lock);
e51060f0
SH		 1983
1984 if (status) {
61a73c70 1985 if (!cma_comp_exch(id_priv, CMA_ADDR_RESOLVED, CMA_ADDR_BOUND))
e51060f0 1986 goto out;
a1b1b61f
SH		 1987 event.event = RDMA_CM_EVENT_ADDR_ERROR;
1988 event.status = status;
e51060f0 1989 } else {
e51060f0
SH		 1990 memcpy(&id_priv->id.route.addr.src_addr, src_addr,
1991 ip_addr_size(src_addr));
a1b1b61f 1992 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
e51060f0
SH		 1993 }
1994
a1b1b61f 1995 if (id_priv->id.event_handler(&id_priv->id, &event)) {
e51060f0 1996 cma_exch(id_priv, CMA_DESTROYING);
de910bd9 1997 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1998 cma_deref_id(id_priv);
1999 rdma_destroy_id(&id_priv->id);
2000 return;
2001 }
2002out:
de910bd9 2003 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 2004 cma_deref_id(id_priv);
2005}
2006
2007static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2008{
2009 struct cma_work *work;
6f8372b6 2010 struct sockaddr *src, *dst;
f0ee3404 2011 union ib_gid gid;
e51060f0
SH		 2012 int ret;
2013
2014 work = kzalloc(sizeof *work, GFP_KERNEL);
2015 if (!work)
2016 return -ENOMEM;
2017
2018 if (!id_priv->cma_dev) {
2019 ret = cma_bind_loopback(id_priv);
2020 if (ret)
2021 goto err;
2022 }
2023
6f8372b6
SH		 2024 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2025 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0 2026
6f8372b6
SH		 2027 src = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
2028 if (cma_zero_addr(src)) {
2029 dst = (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
2030 if ((src->sa_family = dst->sa_family) == AF_INET) {
2031 ((struct sockaddr_in *) src)->sin_addr.s_addr =
2032 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
2033 } else {
2034 ipv6_addr_copy(&((struct sockaddr_in6 *) src)->sin6_addr,
2035 &((struct sockaddr_in6 *) dst)->sin6_addr);
2036 }
e51060f0
SH		 2037 }
2038
2039 work->id = id_priv;
c4028958 2040 INIT_WORK(&work->work, cma_work_handler);
e51060f0
SH		 2041 work->old_state = CMA_ADDR_QUERY;
2042 work->new_state = CMA_ADDR_RESOLVED;
2043 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2044 queue_work(cma_wq, &work->work);
2045 return 0;
2046err:
2047 kfree(work);
2048 return ret;
2049}
2050
2051static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2052 struct sockaddr *dst_addr)
2053{
d14714df
SH		 2054 if (!src_addr || !src_addr->sa_family) {
2055 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2056 if ((src_addr->sa_family = dst_addr->sa_family) == AF_INET6) {
2057 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2058 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2059 }
2060 }
2061 return rdma_bind_addr(id, src_addr);
e51060f0
SH		 2062}
2063
2064int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2065 struct sockaddr *dst_addr, int timeout_ms)
2066{
2067 struct rdma_id_private *id_priv;
2068 int ret;
2069
2070 id_priv = container_of(id, struct rdma_id_private, id);
2071 if (id_priv->state == CMA_IDLE) {
2072 ret = cma_bind_addr(id, src_addr, dst_addr);
2073 if (ret)
2074 return ret;
2075 }
2076
2077 if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_ADDR_QUERY))
2078 return -EINVAL;
2079
2080 atomic_inc(&id_priv->refcount);
2081 memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
2082 if (cma_any_addr(dst_addr))
2083 ret = cma_resolve_loopback(id_priv);
2084 else
3f446754 2085 ret = rdma_resolve_ip(&addr_client, (struct sockaddr *) &id->route.addr.src_addr,
7a118df3 2086 dst_addr, &id->route.addr.dev_addr,
e51060f0
SH		 2087 timeout_ms, addr_handler, id_priv);
2088 if (ret)
2089 goto err;
2090
2091 return 0;
2092err:
2093 cma_comp_exch(id_priv, CMA_ADDR_QUERY, CMA_ADDR_BOUND);
2094 cma_deref_id(id_priv);
2095 return ret;
2096}
2097EXPORT_SYMBOL(rdma_resolve_addr);
2098
2099static void cma_bind_port(struct rdma_bind_list *bind_list,
2100 struct rdma_id_private *id_priv)
2101{
2102 struct sockaddr_in *sin;
2103
2104 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
2105 sin->sin_port = htons(bind_list->port);
2106 id_priv->bind_list = bind_list;
2107 hlist_add_head(&id_priv->node, &bind_list->owners);
2108}
2109
2110static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2111 unsigned short snum)
2112{
2113 struct rdma_bind_list *bind_list;
aedec080 2114 int port, ret;
e51060f0 2115
cb164b8c 2116 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
e51060f0
SH		 2117 if (!bind_list)
2118 return -ENOMEM;
2119
aedec080
SH		 2120 do {
2121 ret = idr_get_new_above(ps, bind_list, snum, &port);
2122 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
2123
2124 if (ret)
2125 goto err1;
2126
2127 if (port != snum) {
2128 ret = -EADDRNOTAVAIL;
2129 goto err2;
2130 }
2131
2132 bind_list->ps = ps;
2133 bind_list->port = (unsigned short) port;
2134 cma_bind_port(bind_list, id_priv);
2135 return 0;
2136err2:
2137 idr_remove(ps, port);
2138err1:
2139 kfree(bind_list);
2140 return ret;
2141}
e51060f0 2142
aedec080
SH		 2143static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2144{
5d7220e8
TH		 2145 static unsigned int last_used_port;
2146 int low, high, remaining;
2147 unsigned int rover;
e51060f0 2148
227b60f5 2149 inet_get_local_port_range(&low, &high);
5d7220e8
TH		 2150 remaining = (high - low) + 1;
2151 rover = net_random() % remaining + low;
2152retry:
2153 if (last_used_port != rover &&
2154 !idr_find(ps, (unsigned short) rover)) {
2155 int ret = cma_alloc_port(ps, id_priv, rover);
2156 /*
2157 * Remember previously used port number in order to avoid
2158 * re-using same port immediately after it is closed.
2159 */
2160 if (!ret)
2161 last_used_port = rover;
2162 if (ret != -EADDRNOTAVAIL)
2163 return ret;
e51060f0 2164 }
5d7220e8
TH		 2165 if (--remaining) {
2166 rover++;
2167 if ((rover < low) || (rover > high))
2168 rover = low;
2169 goto retry;
2170 }
2171 return -EADDRNOTAVAIL;
e51060f0
SH		 2172}
2173
2174static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2175{
2176 struct rdma_id_private *cur_id;
2177 struct sockaddr_in *sin, *cur_sin;
2178 struct rdma_bind_list *bind_list;
2179 struct hlist_node *node;
2180 unsigned short snum;
2181
2182 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
2183 snum = ntohs(sin->sin_port);
2184 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2185 return -EACCES;
2186
2187 bind_list = idr_find(ps, snum);
2188 if (!bind_list)
2189 return cma_alloc_port(ps, id_priv, snum);
2190
2191 /*
2192 * We don't support binding to any address if anyone is bound to
2193 * a specific address on the same port.
2194 */
3f446754 2195 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr))
e51060f0
SH		 2196 return -EADDRNOTAVAIL;
2197
2198 hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
3f446754 2199 if (cma_any_addr((struct sockaddr *) &cur_id->id.route.addr.src_addr))
e51060f0 2200 return -EADDRNOTAVAIL;
3cd96564 2201
e51060f0
SH		 2202 cur_sin = (struct sockaddr_in *) &cur_id->id.route.addr.src_addr;
2203 if (sin->sin_addr.s_addr == cur_sin->sin_addr.s_addr)
2204 return -EADDRINUSE;
2205 }
2206
2207 cma_bind_port(bind_list, id_priv);
2208 return 0;
2209}
2210
2211static int cma_get_port(struct rdma_id_private *id_priv)
2212{
2213 struct idr *ps;
2214 int ret;
2215
2216 switch (id_priv->id.ps) {
2217 case RDMA_PS_SDP:
2218 ps = &sdp_ps;
2219 break;
2220 case RDMA_PS_TCP:
2221 ps = &tcp_ps;
2222 break;
628e5f6d
SH		 2223 case RDMA_PS_UDP:
2224 ps = &udp_ps;
2225 break;
c8f6a362
SH		 2226 case RDMA_PS_IPOIB:
2227 ps = &ipoib_ps;
2228 break;
e51060f0
SH		 2229 default:
2230 return -EPROTONOSUPPORT;
2231 }
2232
2233 mutex_lock(&lock);
3f446754 2234 if (cma_any_port((struct sockaddr *) &id_priv->id.route.addr.src_addr))
aedec080 2235 ret = cma_alloc_any_port(ps, id_priv);
e51060f0
SH		 2236 else
2237 ret = cma_use_port(ps, id_priv);
2238 mutex_unlock(&lock);
2239
2240 return ret;
2241}
2242
d14714df
SH		 2243static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2244 struct sockaddr *addr)
2245{
fd4582a3 2246#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
d14714df
SH		 2247 struct sockaddr_in6 *sin6;
2248
2249 if (addr->sa_family != AF_INET6)
2250 return 0;
2251
2252 sin6 = (struct sockaddr_in6 *) addr;
2253 if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) &&
2254 !sin6->sin6_scope_id)
2255 return -EINVAL;
2256
2257 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2258#endif
2259 return 0;
2260}
2261
e51060f0
SH		 2262int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2263{
2264 struct rdma_id_private *id_priv;
2265 int ret;
2266
1f5175ad 2267 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
e51060f0
SH		 2268 return -EAFNOSUPPORT;
2269
2270 id_priv = container_of(id, struct rdma_id_private, id);
2271 if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND))
2272 return -EINVAL;
2273
d14714df
SH		 2274 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2275 if (ret)
2276 goto err1;
2277
8523c048 2278 if (!cma_any_addr(addr)) {
e51060f0 2279 ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
e51060f0 2280 if (ret)
255d0c14
KK		 2281 goto err1;
2282
2283 mutex_lock(&lock);
2284 ret = cma_acquire_dev(id_priv);
2285 mutex_unlock(&lock);
2286 if (ret)
2287 goto err1;
e51060f0
SH		 2288 }
2289
2290 memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
2291 ret = cma_get_port(id_priv);
2292 if (ret)
255d0c14 2293 goto err2;
e51060f0
SH		 2294
2295 return 0;
255d0c14 2296err2:
6f8372b6 2297 if (id_priv->cma_dev) {
255d0c14
KK		 2298 mutex_lock(&lock);
2299 cma_detach_from_dev(id_priv);
2300 mutex_unlock(&lock);
2301 }
2302err1:
e51060f0
SH		 2303 cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_IDLE);
2304 return ret;
2305}
2306EXPORT_SYMBOL(rdma_bind_addr);
2307
2308static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
2309 struct rdma_route *route)
2310{
e51060f0
SH		 2311 struct cma_hdr *cma_hdr;
2312 struct sdp_hh *sdp_hdr;
2313
1f5175ad
AS		 2314 if (route->addr.src_addr.ss_family == AF_INET) {
2315 struct sockaddr_in *src4, *dst4;
2316
2317 src4 = (struct sockaddr_in *) &route->addr.src_addr;
2318 dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
2319
2320 switch (ps) {
2321 case RDMA_PS_SDP:
2322 sdp_hdr = hdr;
2323 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2324 return -EINVAL;
2325 sdp_set_ip_ver(sdp_hdr, 4);
2326 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2327 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2328 sdp_hdr->port = src4->sin_port;
2329 break;
2330 default:
2331 cma_hdr = hdr;
2332 cma_hdr->cma_version = CMA_VERSION;
2333 cma_set_ip_ver(cma_hdr, 4);
2334 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2335 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2336 cma_hdr->port = src4->sin_port;
2337 break;
2338 }
2339 } else {
2340 struct sockaddr_in6 *src6, *dst6;
2341
2342 src6 = (struct sockaddr_in6 *) &route->addr.src_addr;
2343 dst6 = (struct sockaddr_in6 *) &route->addr.dst_addr;
2344
2345 switch (ps) {
2346 case RDMA_PS_SDP:
2347 sdp_hdr = hdr;
2348 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2349 return -EINVAL;
2350 sdp_set_ip_ver(sdp_hdr, 6);
2351 sdp_hdr->src_addr.ip6 = src6->sin6_addr;
2352 sdp_hdr->dst_addr.ip6 = dst6->sin6_addr;
2353 sdp_hdr->port = src6->sin6_port;
2354 break;
2355 default:
2356 cma_hdr = hdr;
2357 cma_hdr->cma_version = CMA_VERSION;
2358 cma_set_ip_ver(cma_hdr, 6);
2359 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2360 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2361 cma_hdr->port = src6->sin6_port;
2362 break;
2363 }
e51060f0
SH		 2364 }
2365 return 0;
2366}
2367
628e5f6d
SH		 2368static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2369 struct ib_cm_event *ib_event)
2370{
2371 struct rdma_id_private *id_priv = cm_id->context;
2372 struct rdma_cm_event event;
2373 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2374 int ret = 0;
2375
de910bd9 2376 if (cma_disable_callback(id_priv, CMA_CONNECT))
8aa08602 2377 return 0;
628e5f6d 2378
8aa08602 2379 memset(&event, 0, sizeof event);
628e5f6d
SH		 2380 switch (ib_event->event) {
2381 case IB_CM_SIDR_REQ_ERROR:
2382 event.event = RDMA_CM_EVENT_UNREACHABLE;
2383 event.status = -ETIMEDOUT;
2384 break;
2385 case IB_CM_SIDR_REP_RECEIVED:
2386 event.param.ud.private_data = ib_event->private_data;
2387 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2388 if (rep->status != IB_SIDR_SUCCESS) {
2389 event.event = RDMA_CM_EVENT_UNREACHABLE;
2390 event.status = ib_event->param.sidr_rep_rcvd.status;
2391 break;
2392 }
d2ca39f2
YE		 2393 ret = cma_set_qkey(id_priv);
2394 if (ret) {
2395 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2396 event.status = -EINVAL;
2397 break;
2398 }
c8f6a362 2399 if (id_priv->qkey != rep->qkey) {
628e5f6d
SH		 2400 event.event = RDMA_CM_EVENT_UNREACHABLE;
2401 event.status = -EINVAL;
2402 break;
2403 }
2404 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2405 id_priv->id.route.path_rec,
2406 &event.param.ud.ah_attr);
2407 event.param.ud.qp_num = rep->qpn;
2408 event.param.ud.qkey = rep->qkey;
2409 event.event = RDMA_CM_EVENT_ESTABLISHED;
2410 event.status = 0;
2411 break;
2412 default:
468f2239 2413 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
628e5f6d
SH		 2414 ib_event->event);
2415 goto out;
2416 }
2417
2418 ret = id_priv->id.event_handler(&id_priv->id, &event);
2419 if (ret) {
2420 /* Destroy the CM ID by returning a non-zero value. */
2421 id_priv->cm_id.ib = NULL;
2422 cma_exch(id_priv, CMA_DESTROYING);
de910bd9 2423 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 2424 rdma_destroy_id(&id_priv->id);
2425 return ret;
2426 }
2427out:
de910bd9 2428 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 2429 return ret;
2430}
2431
2432static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2433 struct rdma_conn_param *conn_param)
2434{
2435 struct ib_cm_sidr_req_param req;
2436 struct rdma_route *route;
2437 int ret;
2438
2439 req.private_data_len = sizeof(struct cma_hdr) +
2440 conn_param->private_data_len;
2441 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2442 if (!req.private_data)
2443 return -ENOMEM;
2444
2445 if (conn_param->private_data && conn_param->private_data_len)
2446 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2447 conn_param->private_data, conn_param->private_data_len);
2448
2449 route = &id_priv->id.route;
2450 ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
2451 if (ret)
2452 goto out;
2453
2454 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device,
2455 cma_sidr_rep_handler, id_priv);
2456 if (IS_ERR(id_priv->cm_id.ib)) {
2457 ret = PTR_ERR(id_priv->cm_id.ib);
2458 goto out;
2459 }
2460
2461 req.path = route->path_rec;
2462 req.service_id = cma_get_service_id(id_priv->id.ps,
3f446754 2463 (struct sockaddr *) &route->addr.dst_addr);
628e5f6d
SH		 2464 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2465 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2466
2467 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2468 if (ret) {
2469 ib_destroy_cm_id(id_priv->cm_id.ib);
2470 id_priv->cm_id.ib = NULL;
2471 }
2472out:
2473 kfree(req.private_data);
2474 return ret;
2475}
2476
e51060f0
SH		 2477static int cma_connect_ib(struct rdma_id_private *id_priv,
2478 struct rdma_conn_param *conn_param)
2479{
2480 struct ib_cm_req_param req;
2481 struct rdma_route *route;
2482 void *private_data;
2483 int offset, ret;
2484
2485 memset(&req, 0, sizeof req);
2486 offset = cma_user_data_offset(id_priv->id.ps);
2487 req.private_data_len = offset + conn_param->private_data_len;
2488 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2489 if (!private_data)
2490 return -ENOMEM;
2491
2492 if (conn_param->private_data && conn_param->private_data_len)
2493 memcpy(private_data + offset, conn_param->private_data,
2494 conn_param->private_data_len);
2495
2496 id_priv->cm_id.ib = ib_create_cm_id(id_priv->id.device, cma_ib_handler,
2497 id_priv);
2498 if (IS_ERR(id_priv->cm_id.ib)) {
2499 ret = PTR_ERR(id_priv->cm_id.ib);
2500 goto out;
2501 }
2502
2503 route = &id_priv->id.route;
2504 ret = cma_format_hdr(private_data, id_priv->id.ps, route);
2505 if (ret)
2506 goto out;
2507 req.private_data = private_data;
2508
2509 req.primary_path = &route->path_rec[0];
2510 if (route->num_paths == 2)
2511 req.alternate_path = &route->path_rec[1];
2512
2513 req.service_id = cma_get_service_id(id_priv->id.ps,
3f446754 2514 (struct sockaddr *) &route->addr.dst_addr);
e51060f0 2515 req.qp_num = id_priv->qp_num;
9b2e9c0c 2516 req.qp_type = IB_QPT_RC;
e51060f0
SH		 2517 req.starting_psn = id_priv->seq_num;
2518 req.responder_resources = conn_param->responder_resources;
2519 req.initiator_depth = conn_param->initiator_depth;
2520 req.flow_control = conn_param->flow_control;
2521 req.retry_count = conn_param->retry_count;
2522 req.rnr_retry_count = conn_param->rnr_retry_count;
2523 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2524 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2525 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2526 req.srq = id_priv->srq ? 1 : 0;
2527
2528 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2529out:
675a027c
KK		 2530 if (ret && !IS_ERR(id_priv->cm_id.ib)) {
2531 ib_destroy_cm_id(id_priv->cm_id.ib);
2532 id_priv->cm_id.ib = NULL;
2533 }
2534
e51060f0
SH		 2535 kfree(private_data);
2536 return ret;
2537}
2538
07ebafba
TT		 2539static int cma_connect_iw(struct rdma_id_private *id_priv,
2540 struct rdma_conn_param *conn_param)
2541{
2542 struct iw_cm_id *cm_id;
2543 struct sockaddr_in* sin;
2544 int ret;
2545 struct iw_cm_conn_param iw_param;
2546
2547 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2548 if (IS_ERR(cm_id)) {
2549 ret = PTR_ERR(cm_id);
2550 goto out;
2551 }
2552
2553 id_priv->cm_id.iw = cm_id;
2554
2555 sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
2556 cm_id->local_addr = *sin;
2557
2558 sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
2559 cm_id->remote_addr = *sin;
2560
5851bb89 2561 ret = cma_modify_qp_rtr(id_priv, conn_param);
675a027c
KK		 2562 if (ret)
2563 goto out;
07ebafba
TT		 2564
2565 iw_param.ord = conn_param->initiator_depth;
2566 iw_param.ird = conn_param->responder_resources;
2567 iw_param.private_data = conn_param->private_data;
2568 iw_param.private_data_len = conn_param->private_data_len;
2569 if (id_priv->id.qp)
2570 iw_param.qpn = id_priv->qp_num;
2571 else
2572 iw_param.qpn = conn_param->qp_num;
2573 ret = iw_cm_connect(cm_id, &iw_param);
2574out:
675a027c
KK		 2575 if (ret && !IS_ERR(cm_id)) {
2576 iw_destroy_cm_id(cm_id);
2577 id_priv->cm_id.iw = NULL;
2578 }
07ebafba
TT		 2579 return ret;
2580}
2581
e51060f0
SH		 2582int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2583{
2584 struct rdma_id_private *id_priv;
2585 int ret;
2586
2587 id_priv = container_of(id, struct rdma_id_private, id);
2588 if (!cma_comp_exch(id_priv, CMA_ROUTE_RESOLVED, CMA_CONNECT))
2589 return -EINVAL;
2590
2591 if (!id->qp) {
2592 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 2593 id_priv->srq = conn_param->srq;
2594 }
2595
07ebafba
TT		 2596 switch (rdma_node_get_transport(id->device->node_type)) {
2597 case RDMA_TRANSPORT_IB:
c8f6a362 2598 if (cma_is_ud_ps(id->ps))
628e5f6d
SH		 2599 ret = cma_resolve_ib_udp(id_priv, conn_param);
2600 else
2601 ret = cma_connect_ib(id_priv, conn_param);
e51060f0 2602 break;
07ebafba
TT		 2603 case RDMA_TRANSPORT_IWARP:
2604 ret = cma_connect_iw(id_priv, conn_param);
2605 break;
e51060f0
SH		 2606 default:
2607 ret = -ENOSYS;
2608 break;
2609 }
2610 if (ret)
2611 goto err;
2612
2613 return 0;
2614err:
2615 cma_comp_exch(id_priv, CMA_CONNECT, CMA_ROUTE_RESOLVED);
2616 return ret;
2617}
2618EXPORT_SYMBOL(rdma_connect);
2619
2620static int cma_accept_ib(struct rdma_id_private *id_priv,
2621 struct rdma_conn_param *conn_param)
2622{
2623 struct ib_cm_rep_param rep;
5851bb89 2624 int ret;
0fe313b0 2625
5851bb89
SH		 2626 ret = cma_modify_qp_rtr(id_priv, conn_param);
2627 if (ret)
2628 goto out;
0fe313b0 2629
5851bb89
SH		 2630 ret = cma_modify_qp_rts(id_priv, conn_param);
2631 if (ret)
2632 goto out;
e51060f0
SH		 2633
2634 memset(&rep, 0, sizeof rep);
2635 rep.qp_num = id_priv->qp_num;
2636 rep.starting_psn = id_priv->seq_num;
2637 rep.private_data = conn_param->private_data;
2638 rep.private_data_len = conn_param->private_data_len;
2639 rep.responder_resources = conn_param->responder_resources;
2640 rep.initiator_depth = conn_param->initiator_depth;
e51060f0
SH		 2641 rep.failover_accepted = 0;
2642 rep.flow_control = conn_param->flow_control;
2643 rep.rnr_retry_count = conn_param->rnr_retry_count;
2644 rep.srq = id_priv->srq ? 1 : 0;
2645
0fe313b0
SH		 2646 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2647out:
2648 return ret;
e51060f0
SH		 2649}
2650
07ebafba
TT		 2651static int cma_accept_iw(struct rdma_id_private *id_priv,
2652 struct rdma_conn_param *conn_param)
2653{
2654 struct iw_cm_conn_param iw_param;
2655 int ret;
2656
5851bb89 2657 ret = cma_modify_qp_rtr(id_priv, conn_param);
07ebafba
TT		 2658 if (ret)
2659 return ret;
2660
2661 iw_param.ord = conn_param->initiator_depth;
2662 iw_param.ird = conn_param->responder_resources;
2663 iw_param.private_data = conn_param->private_data;
2664 iw_param.private_data_len = conn_param->private_data_len;
2665 if (id_priv->id.qp) {
2666 iw_param.qpn = id_priv->qp_num;
2667 } else
2668 iw_param.qpn = conn_param->qp_num;
2669
2670 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2671}
2672
628e5f6d
SH		 2673static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2674 enum ib_cm_sidr_status status,
2675 const void *private_data, int private_data_len)
2676{
2677 struct ib_cm_sidr_rep_param rep;
d2ca39f2 2678 int ret;
628e5f6d
SH		 2679
2680 memset(&rep, 0, sizeof rep);
2681 rep.status = status;
2682 if (status == IB_SIDR_SUCCESS) {
d2ca39f2
YE		 2683 ret = cma_set_qkey(id_priv);
2684 if (ret)
2685 return ret;
628e5f6d 2686 rep.qp_num = id_priv->qp_num;
c8f6a362 2687 rep.qkey = id_priv->qkey;
628e5f6d
SH		 2688 }
2689 rep.private_data = private_data;
2690 rep.private_data_len = private_data_len;
2691
2692 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2693}
2694
e51060f0
SH		 2695int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2696{
2697 struct rdma_id_private *id_priv;
2698 int ret;
2699
2700 id_priv = container_of(id, struct rdma_id_private, id);
2701 if (!cma_comp(id_priv, CMA_CONNECT))
2702 return -EINVAL;
2703
2704 if (!id->qp && conn_param) {
2705 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 2706 id_priv->srq = conn_param->srq;
2707 }
2708
07ebafba
TT		 2709 switch (rdma_node_get_transport(id->device->node_type)) {
2710 case RDMA_TRANSPORT_IB:
c8f6a362 2711 if (cma_is_ud_ps(id->ps))
628e5f6d
SH		 2712 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2713 conn_param->private_data,
2714 conn_param->private_data_len);
2715 else if (conn_param)
e51060f0
SH		 2716 ret = cma_accept_ib(id_priv, conn_param);
2717 else
2718 ret = cma_rep_recv(id_priv);
2719 break;
07ebafba
TT		 2720 case RDMA_TRANSPORT_IWARP:
2721 ret = cma_accept_iw(id_priv, conn_param);
2722 break;
e51060f0
SH		 2723 default:
2724 ret = -ENOSYS;
2725 break;
2726 }
2727
2728 if (ret)
2729 goto reject;
2730
2731 return 0;
2732reject:
c5483388 2733 cma_modify_qp_err(id_priv);
e51060f0
SH		 2734 rdma_reject(id, NULL, 0);
2735 return ret;
2736}
2737EXPORT_SYMBOL(rdma_accept);
2738
0fe313b0
SH		 2739int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2740{
2741 struct rdma_id_private *id_priv;
2742 int ret;
2743
2744 id_priv = container_of(id, struct rdma_id_private, id);
6c719f5c 2745 if (!cma_has_cm_dev(id_priv))
0fe313b0
SH		 2746 return -EINVAL;
2747
2748 switch (id->device->node_type) {
2749 case RDMA_NODE_IB_CA:
2750 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2751 break;
2752 default:
2753 ret = 0;
2754 break;
2755 }
2756 return ret;
2757}
2758EXPORT_SYMBOL(rdma_notify);
2759
e51060f0
SH		 2760int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2761 u8 private_data_len)
2762{
2763 struct rdma_id_private *id_priv;
2764 int ret;
2765
2766 id_priv = container_of(id, struct rdma_id_private, id);
6c719f5c 2767 if (!cma_has_cm_dev(id_priv))
e51060f0
SH		 2768 return -EINVAL;
2769
07ebafba
TT		 2770 switch (rdma_node_get_transport(id->device->node_type)) {
2771 case RDMA_TRANSPORT_IB:
c8f6a362 2772 if (cma_is_ud_ps(id->ps))
628e5f6d
SH		 2773 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
2774 private_data, private_data_len);
2775 else
2776 ret = ib_send_cm_rej(id_priv->cm_id.ib,
2777 IB_CM_REJ_CONSUMER_DEFINED, NULL,
2778 0, private_data, private_data_len);
e51060f0 2779 break;
07ebafba
TT		 2780 case RDMA_TRANSPORT_IWARP:
2781 ret = iw_cm_reject(id_priv->cm_id.iw,
2782 private_data, private_data_len);
2783 break;
e51060f0
SH		 2784 default:
2785 ret = -ENOSYS;
2786 break;
2787 }
2788 return ret;
2789}
2790EXPORT_SYMBOL(rdma_reject);
2791
2792int rdma_disconnect(struct rdma_cm_id *id)
2793{
2794 struct rdma_id_private *id_priv;
2795 int ret;
2796
2797 id_priv = container_of(id, struct rdma_id_private, id);
6c719f5c 2798 if (!cma_has_cm_dev(id_priv))
e51060f0
SH		 2799 return -EINVAL;
2800
07ebafba
TT		 2801 switch (rdma_node_get_transport(id->device->node_type)) {
2802 case RDMA_TRANSPORT_IB:
c5483388 2803 ret = cma_modify_qp_err(id_priv);
07ebafba
TT		 2804 if (ret)
2805 goto out;
e51060f0
SH		 2806 /* Initiate or respond to a disconnect. */
2807 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
2808 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
2809 break;
07ebafba
TT		 2810 case RDMA_TRANSPORT_IWARP:
2811 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
2812 break;
e51060f0 2813 default:
07ebafba 2814 ret = -EINVAL;
e51060f0
SH		 2815 break;
2816 }
2817out:
2818 return ret;
2819}
2820EXPORT_SYMBOL(rdma_disconnect);
2821
c8f6a362
SH		 2822static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
2823{
2824 struct rdma_id_private *id_priv;
2825 struct cma_multicast *mc = multicast->context;
2826 struct rdma_cm_event event;
2827 int ret;
2828
2829 id_priv = mc->id_priv;
de910bd9
OG		 2830 if (cma_disable_callback(id_priv, CMA_ADDR_BOUND) &&
2831 cma_disable_callback(id_priv, CMA_ADDR_RESOLVED))
8aa08602 2832 return 0;
c8f6a362 2833
c5483388 2834 mutex_lock(&id_priv->qp_mutex);
c8f6a362
SH		 2835 if (!status && id_priv->id.qp)
2836 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
2837 multicast->rec.mlid);
c5483388 2838 mutex_unlock(&id_priv->qp_mutex);
c8f6a362
SH		 2839
2840 memset(&event, 0, sizeof event);
2841 event.status = status;
2842 event.param.ud.private_data = mc->context;
2843 if (!status) {
2844 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
2845 ib_init_ah_from_mcmember(id_priv->id.device,
2846 id_priv->id.port_num, &multicast->rec,
2847 &event.param.ud.ah_attr);
2848 event.param.ud.qp_num = 0xFFFFFF;
2849 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
2850 } else
2851 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
2852
2853 ret = id_priv->id.event_handler(&id_priv->id, &event);
2854 if (ret) {
2855 cma_exch(id_priv, CMA_DESTROYING);
de910bd9 2856 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 2857 rdma_destroy_id(&id_priv->id);
2858 return 0;
2859 }
8aa08602 2860
de910bd9 2861 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 2862 return 0;
2863}
2864
2865static void cma_set_mgid(struct rdma_id_private *id_priv,
2866 struct sockaddr *addr, union ib_gid *mgid)
2867{
2868 unsigned char mc_map[MAX_ADDR_LEN];
2869 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2870 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
2871 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
2872
2873 if (cma_any_addr(addr)) {
2874 memset(mgid, 0, sizeof *mgid);
2875 } else if ((addr->sa_family == AF_INET6) &&
1c9b2819 2876 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
c8f6a362
SH		 2877 0xFF10A01B)) {
2878 /* IPv6 address is an SA assigned MGID. */
2879 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
e2e62697
JG		 2880 } else if ((addr->sa_family == AF_INET6)) {
2881 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
2882 if (id_priv->id.ps == RDMA_PS_UDP)
2883 mc_map[7] = 0x01; /* Use RDMA CM signature */
2884 *mgid = *(union ib_gid *) (mc_map + 4);
c8f6a362 2885 } else {
a9e527e3 2886 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
c8f6a362
SH		 2887 if (id_priv->id.ps == RDMA_PS_UDP)
2888 mc_map[7] = 0x01; /* Use RDMA CM signature */
c8f6a362
SH		 2889 *mgid = *(union ib_gid *) (mc_map + 4);
2890 }
2891}
2892
2893static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
2894 struct cma_multicast *mc)
2895{
2896 struct ib_sa_mcmember_rec rec;
2897 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2898 ib_sa_comp_mask comp_mask;
2899 int ret;
2900
2901 ib_addr_get_mgid(dev_addr, &rec.mgid);
2902 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
2903 &rec.mgid, &rec);
2904 if (ret)
2905 return ret;
2906
3f446754 2907 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
c8f6a362
SH		 2908 if (id_priv->id.ps == RDMA_PS_UDP)
2909 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
6f8372b6 2910 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
c8f6a362
SH		 2911 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2912 rec.join_state = 1;
2913
2914 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
2915 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
2916 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
2917 IB_SA_MCMEMBER_REC_FLOW_LABEL |
2918 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
2919
84adeee9
YE		 2920 if (id_priv->id.ps == RDMA_PS_IPOIB)
2921 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
2922 IB_SA_MCMEMBER_REC_RATE_SELECTOR;
2923
c8f6a362
SH		 2924 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
2925 id_priv->id.port_num, &rec,
2926 comp_mask, GFP_KERNEL,
2927 cma_ib_mc_handler, mc);
2928 if (IS_ERR(mc->multicast.ib))
2929 return PTR_ERR(mc->multicast.ib);
2930
2931 return 0;
2932}
2933
3c86aa70
EC		 2934static void iboe_mcast_work_handler(struct work_struct *work)
2935{
2936 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
2937 struct cma_multicast *mc = mw->mc;
2938 struct ib_sa_multicast *m = mc->multicast.ib;
2939
2940 mc->multicast.ib->context = mc;
2941 cma_ib_mc_handler(0, m);
2942 kref_put(&mc->mcref, release_mc);
2943 kfree(mw);
2944}
2945
2946static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
2947{
2948 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
2949 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
2950
2951 if (cma_any_addr(addr)) {
2952 memset(mgid, 0, sizeof *mgid);
2953 } else if (addr->sa_family == AF_INET6) {
2954 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
2955 } else {
2956 mgid->raw[0] = 0xff;
2957 mgid->raw[1] = 0x0e;
2958 mgid->raw[2] = 0;
2959 mgid->raw[3] = 0;
2960 mgid->raw[4] = 0;
2961 mgid->raw[5] = 0;
2962 mgid->raw[6] = 0;
2963 mgid->raw[7] = 0;
2964 mgid->raw[8] = 0;
2965 mgid->raw[9] = 0;
2966 mgid->raw[10] = 0xff;
2967 mgid->raw[11] = 0xff;
2968 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
2969 }
2970}
2971
2972static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
2973 struct cma_multicast *mc)
2974{
2975 struct iboe_mcast_work *work;
2976 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2977 int err;
2978 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
2979 struct net_device *ndev = NULL;
2980
2981 if (cma_zero_addr((struct sockaddr *)&mc->addr))
2982 return -EINVAL;
2983
2984 work = kzalloc(sizeof *work, GFP_KERNEL);
2985 if (!work)
2986 return -ENOMEM;
2987
2988 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
2989 if (!mc->multicast.ib) {
2990 err = -ENOMEM;
2991 goto out1;
2992 }
2993
2994 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
2995
2996 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
2997 if (id_priv->id.ps == RDMA_PS_UDP)
2998 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
2999
3000 if (dev_addr->bound_dev_if)
3001 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3002 if (!ndev) {
3003 err = -ENODEV;
3004 goto out2;
3005 }
3006 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3007 mc->multicast.ib->rec.hop_limit = 1;
3008 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3009 dev_put(ndev);
3010 if (!mc->multicast.ib->rec.mtu) {
3011 err = -EINVAL;
3012 goto out2;
3013 }
3014 iboe_addr_get_sgid(dev_addr, &mc->multicast.ib->rec.port_gid);
3015 work->id = id_priv;
3016 work->mc = mc;
3017 INIT_WORK(&work->work, iboe_mcast_work_handler);
3018 kref_get(&mc->mcref);
3019 queue_work(cma_wq, &work->work);
3020
3021 return 0;
3022
3023out2:
3024 kfree(mc->multicast.ib);
3025out1:
3026 kfree(work);
3027 return err;
3028}
3029
c8f6a362
SH		 3030int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3031 void *context)
3032{
3033 struct rdma_id_private *id_priv;
3034 struct cma_multicast *mc;
3035 int ret;
3036
3037 id_priv = container_of(id, struct rdma_id_private, id);
3038 if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
3039 !cma_comp(id_priv, CMA_ADDR_RESOLVED))
3040 return -EINVAL;
3041
3042 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3043 if (!mc)
3044 return -ENOMEM;
3045
3046 memcpy(&mc->addr, addr, ip_addr_size(addr));
3047 mc->context = context;
3048 mc->id_priv = id_priv;
3049
3050 spin_lock(&id_priv->lock);
3051 list_add(&mc->list, &id_priv->mc_list);
3052 spin_unlock(&id_priv->lock);
3053
3054 switch (rdma_node_get_transport(id->device->node_type)) {
3055 case RDMA_TRANSPORT_IB:
3c86aa70
EC		 3056 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3057 case IB_LINK_LAYER_INFINIBAND:
3058 ret = cma_join_ib_multicast(id_priv, mc);
3059 break;
3060 case IB_LINK_LAYER_ETHERNET:
3061 kref_init(&mc->mcref);
3062 ret = cma_iboe_join_multicast(id_priv, mc);
3063 break;
3064 default:
3065 ret = -EINVAL;
3066 }
c8f6a362
SH		 3067 break;
3068 default:
3069 ret = -ENOSYS;
3070 break;
3071 }
3072
3073 if (ret) {
3074 spin_lock_irq(&id_priv->lock);
3075 list_del(&mc->list);
3076 spin_unlock_irq(&id_priv->lock);
3077 kfree(mc);
3078 }
3079 return ret;
3080}
3081EXPORT_SYMBOL(rdma_join_multicast);
3082
3083void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3084{
3085 struct rdma_id_private *id_priv;
3086 struct cma_multicast *mc;
3087
3088 id_priv = container_of(id, struct rdma_id_private, id);
3089 spin_lock_irq(&id_priv->lock);
3090 list_for_each_entry(mc, &id_priv->mc_list, list) {
3091 if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
3092 list_del(&mc->list);
3093 spin_unlock_irq(&id_priv->lock);
3094
3095 if (id->qp)
3096 ib_detach_mcast(id->qp,
3097 &mc->multicast.ib->rec.mgid,
3098 mc->multicast.ib->rec.mlid);
3c86aa70
EC		 3099 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3100 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3101 case IB_LINK_LAYER_INFINIBAND:
3102 ib_sa_free_multicast(mc->multicast.ib);
3103 kfree(mc);
3104 break;
3105 case IB_LINK_LAYER_ETHERNET:
3106 kref_put(&mc->mcref, release_mc);
3107 break;
3108 default:
3109 break;
3110 }
3111 }
c8f6a362
SH		 3112 return;
3113 }
3114 }
3115 spin_unlock_irq(&id_priv->lock);
3116}
3117EXPORT_SYMBOL(rdma_leave_multicast);
3118
dd5bdff8
OG		 3119static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3120{
3121 struct rdma_dev_addr *dev_addr;
3122 struct cma_ndev_work *work;
3123
3124 dev_addr = &id_priv->id.route.addr.dev_addr;
3125
6266ed6e 3126 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
dd5bdff8
OG		 3127 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3128 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3129 ndev->name, &id_priv->id);
3130 work = kzalloc(sizeof *work, GFP_KERNEL);
3131 if (!work)
3132 return -ENOMEM;
3133
3134 INIT_WORK(&work->work, cma_ndev_work_handler);
3135 work->id = id_priv;
3136 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3137 atomic_inc(&id_priv->refcount);
3138 queue_work(cma_wq, &work->work);
3139 }
3140
3141 return 0;
3142}
3143
3144static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3145 void *ctx)
3146{
3147 struct net_device *ndev = (struct net_device *)ctx;
3148 struct cma_device *cma_dev;
3149 struct rdma_id_private *id_priv;
3150 int ret = NOTIFY_DONE;
3151
3152 if (dev_net(ndev) != &init_net)
3153 return NOTIFY_DONE;
3154
3155 if (event != NETDEV_BONDING_FAILOVER)
3156 return NOTIFY_DONE;
3157
3158 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3159 return NOTIFY_DONE;
3160
3161 mutex_lock(&lock);
3162 list_for_each_entry(cma_dev, &dev_list, list)
3163 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3164 ret = cma_netdev_change(ndev, id_priv);
3165 if (ret)
3166 goto out;
3167 }
3168
3169out:
3170 mutex_unlock(&lock);
3171 return ret;
3172}
3173
3174static struct notifier_block cma_nb = {
3175 .notifier_call = cma_netdev_callback
3176};
3177
e51060f0
SH		 3178static void cma_add_one(struct ib_device *device)
3179{
3180 struct cma_device *cma_dev;
3181 struct rdma_id_private *id_priv;
3182
3183 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3184 if (!cma_dev)
3185 return;
3186
3187 cma_dev->device = device;
e51060f0
SH		 3188
3189 init_completion(&cma_dev->comp);
3190 atomic_set(&cma_dev->refcount, 1);
3191 INIT_LIST_HEAD(&cma_dev->id_list);
3192 ib_set_client_data(device, &cma_client, cma_dev);
3193
3194 mutex_lock(&lock);
3195 list_add_tail(&cma_dev->list, &dev_list);
3196 list_for_each_entry(id_priv, &listen_any_list, list)
3197 cma_listen_on_dev(id_priv, cma_dev);
3198 mutex_unlock(&lock);
e51060f0
SH		 3199}
3200
3201static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3202{
a1b1b61f 3203 struct rdma_cm_event event;
e51060f0 3204 enum cma_state state;
de910bd9 3205 int ret = 0;
e51060f0
SH		 3206
3207 /* Record that we want to remove the device */
3208 state = cma_exch(id_priv, CMA_DEVICE_REMOVAL);
3209 if (state == CMA_DESTROYING)
3210 return 0;
3211
3212 cma_cancel_operation(id_priv, state);
de910bd9 3213 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 3214
3215 /* Check for destruction from another callback. */
3216 if (!cma_comp(id_priv, CMA_DEVICE_REMOVAL))
de910bd9 3217 goto out;
e51060f0 3218
a1b1b61f
SH		 3219 memset(&event, 0, sizeof event);
3220 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
de910bd9
OG		 3221 ret = id_priv->id.event_handler(&id_priv->id, &event);
3222out:
3223 mutex_unlock(&id_priv->handler_mutex);
3224 return ret;
e51060f0
SH		 3225}
3226
3227static void cma_process_remove(struct cma_device *cma_dev)
3228{
e51060f0
SH		 3229 struct rdma_id_private *id_priv;
3230 int ret;
3231
e51060f0
SH		 3232 mutex_lock(&lock);
3233 while (!list_empty(&cma_dev->id_list)) {
3234 id_priv = list_entry(cma_dev->id_list.next,
3235 struct rdma_id_private, list);
3236
d02d1f53 3237 list_del(&id_priv->listen_list);
94de178a 3238 list_del_init(&id_priv->list);
e51060f0
SH		 3239 atomic_inc(&id_priv->refcount);
3240 mutex_unlock(&lock);
3241
d02d1f53 3242 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
e51060f0
SH		 3243 cma_deref_id(id_priv);
3244 if (ret)
3245 rdma_destroy_id(&id_priv->id);
3246
3247 mutex_lock(&lock);
3248 }
3249 mutex_unlock(&lock);
3250
3251 cma_deref_dev(cma_dev);
3252 wait_for_completion(&cma_dev->comp);
3253}
3254
3255static void cma_remove_one(struct ib_device *device)
3256{
3257 struct cma_device *cma_dev;
3258
3259 cma_dev = ib_get_client_data(device, &cma_client);
3260 if (!cma_dev)
3261 return;
3262
3263 mutex_lock(&lock);
3264 list_del(&cma_dev->list);
3265 mutex_unlock(&lock);
3266
3267 cma_process_remove(cma_dev);
3268 kfree(cma_dev);
3269}
3270
716abb1f 3271static int __init cma_init(void)
e51060f0 3272{
5d7220e8 3273 int ret;
227b60f5 3274
c7f743a6 3275 cma_wq = create_singlethread_workqueue("rdma_cm");
e51060f0
SH		 3276 if (!cma_wq)
3277 return -ENOMEM;
3278
c1a0b23b 3279 ib_sa_register_client(&sa_client);
7a118df3 3280 rdma_addr_register_client(&addr_client);
dd5bdff8 3281 register_netdevice_notifier(&cma_nb);
c1a0b23b 3282
e51060f0
SH		 3283 ret = ib_register_client(&cma_client);
3284 if (ret)
3285 goto err;
3286 return 0;
3287
3288err:
dd5bdff8 3289 unregister_netdevice_notifier(&cma_nb);
7a118df3 3290 rdma_addr_unregister_client(&addr_client);
c1a0b23b 3291 ib_sa_unregister_client(&sa_client);
e51060f0
SH		 3292 destroy_workqueue(cma_wq);
3293 return ret;
3294}
3295
716abb1f 3296static void __exit cma_cleanup(void)
e51060f0
SH		 3297{
3298 ib_unregister_client(&cma_client);
dd5bdff8 3299 unregister_netdevice_notifier(&cma_nb);
7a118df3 3300 rdma_addr_unregister_client(&addr_client);
c1a0b23b 3301 ib_sa_unregister_client(&sa_client);
e51060f0
SH		 3302 destroy_workqueue(cma_wq);
3303 idr_destroy(&sdp_ps);
3304 idr_destroy(&tcp_ps);
628e5f6d 3305 idr_destroy(&udp_ps);
c8f6a362 3306 idr_destroy(&ipoib_ps);
e51060f0
SH		 3307}
3308
3309module_init(cma_init);
3310module_exit(cma_cleanup);
kernel source for telekom puls (alcatel/mediatek)