projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
RDMA/cma: Add helper functions to return id address information
[linux-2.6-block.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>
e4dd23d7 44#include <linux/module.h>
366cddb4 45#include <net/route.h>
e51060f0
SH		 46
47#include <net/tcp.h>
1f5175ad 48#include <net/ipv6.h>
e51060f0
SH		 49
50#include <rdma/rdma_cm.h>
51#include <rdma/rdma_cm_ib.h>
753f618a 52#include <rdma/rdma_netlink.h>
2e2d190c 53#include <rdma/ib.h>
e51060f0
SH		 54#include <rdma/ib_cache.h>
55#include <rdma/ib_cm.h>
56#include <rdma/ib_sa.h>
07ebafba 57#include <rdma/iw_cm.h>
e51060f0
SH		 58
59MODULE_AUTHOR("Sean Hefty");
60MODULE_DESCRIPTION("Generic RDMA CM Agent");
61MODULE_LICENSE("Dual BSD/GPL");
62
63#define CMA_CM_RESPONSE_TIMEOUT 20
d5bb7599 64#define CMA_MAX_CM_RETRIES 15
dcb3f974 65#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
3c86aa70 66#define CMA_IBOE_PACKET_LIFETIME 18
e51060f0
SH		 67
68static void cma_add_one(struct ib_device *device);
69static void cma_remove_one(struct ib_device *device);
70
71static struct ib_client cma_client = {
72 .name = "cma",
73 .add = cma_add_one,
74 .remove = cma_remove_one
75};
76
c1a0b23b 77static struct ib_sa_client sa_client;
7a118df3 78static struct rdma_addr_client addr_client;
e51060f0
SH		 79static LIST_HEAD(dev_list);
80static LIST_HEAD(listen_any_list);
81static DEFINE_MUTEX(lock);
82static struct workqueue_struct *cma_wq;
83static DEFINE_IDR(sdp_ps);
84static DEFINE_IDR(tcp_ps);
628e5f6d 85static DEFINE_IDR(udp_ps);
c8f6a362 86static DEFINE_IDR(ipoib_ps);
2d2e9415 87static DEFINE_IDR(ib_ps);
e51060f0
SH		 88
89struct cma_device {
90 struct list_head list;
91 struct ib_device *device;
e51060f0
SH		 92 struct completion comp;
93 atomic_t refcount;
94 struct list_head id_list;
95};
96
e51060f0
SH		 97struct rdma_bind_list {
98 struct idr *ps;
99 struct hlist_head owners;
100 unsigned short port;
101};
102
68602120
SH		 103enum {
104 CMA_OPTION_AFONLY,
105};
106
e51060f0
SH		 107/*
108 * Device removal can occur at anytime, so we need extra handling to
109 * serialize notifying the user of device removal with other callbacks.
110 * We do this by disabling removal notification while a callback is in process,
111 * and reporting it after the callback completes.
112 */
113struct rdma_id_private {
114 struct rdma_cm_id id;
115
116 struct rdma_bind_list *bind_list;
117 struct hlist_node node;
d02d1f53
SH		 118 struct list_head list; /* listen_any_list or cma_device.list */
119 struct list_head listen_list; /* per device listens */
e51060f0 120 struct cma_device *cma_dev;
c8f6a362 121 struct list_head mc_list;
e51060f0 122
d02d1f53 123 int internal_id;
550e5ca7 124 enum rdma_cm_state state;
e51060f0 125 spinlock_t lock;
c5483388
SH		 126 struct mutex qp_mutex;
127
e51060f0
SH		 128 struct completion comp;
129 atomic_t refcount;
de910bd9 130 struct mutex handler_mutex;
e51060f0
SH		 131
132 int backlog;
133 int timeout_ms;
134 struct ib_sa_query *query;
135 int query_id;
136 union {
137 struct ib_cm_id *ib;
07ebafba 138 struct iw_cm_id *iw;
e51060f0
SH		 139 } cm_id;
140
141 u32 seq_num;
c8f6a362 142 u32 qkey;
e51060f0 143 u32 qp_num;
83e9502d 144 pid_t owner;
68602120 145 u32 options;
e51060f0 146 u8 srq;
a81c994d 147 u8 tos;
a9bb7912 148 u8 reuseaddr;
5b0ec991 149 u8 afonly;
e51060f0
SH		 150};
151
c8f6a362
SH		 152struct cma_multicast {
153 struct rdma_id_private *id_priv;
154 union {
155 struct ib_sa_multicast *ib;
156 } multicast;
157 struct list_head list;
158 void *context;
3f446754 159 struct sockaddr_storage addr;
3c86aa70 160 struct kref mcref;
c8f6a362
SH		 161};
162
e51060f0
SH		 163struct cma_work {
164 struct work_struct work;
165 struct rdma_id_private *id;
550e5ca7
NM		 166 enum rdma_cm_state old_state;
167 enum rdma_cm_state new_state;
e51060f0
SH		 168 struct rdma_cm_event event;
169};
170
dd5bdff8
OG		 171struct cma_ndev_work {
172 struct work_struct work;
173 struct rdma_id_private *id;
174 struct rdma_cm_event event;
175};
176
3c86aa70
EC		 177struct iboe_mcast_work {
178 struct work_struct work;
179 struct rdma_id_private *id;
180 struct cma_multicast *mc;
181};
182
e51060f0
SH		 183union cma_ip_addr {
184 struct in6_addr ip6;
185 struct {
1b90c137
AV		 186 __be32 pad[3];
187 __be32 addr;
e51060f0
SH		 188 } ip4;
189};
190
191struct cma_hdr {
192 u8 cma_version;
193 u8 ip_version; /* IP version: 7:4 */
1b90c137 194 __be16 port;
e51060f0
SH		 195 union cma_ip_addr src_addr;
196 union cma_ip_addr dst_addr;
197};
198
199struct sdp_hh {
200 u8 bsdh[16];
201 u8 sdp_version; /* Major version: 7:4 */
202 u8 ip_version; /* IP version: 7:4 */
203 u8 sdp_specific1[10];
1b90c137
AV		 204 __be16 port;
205 __be16 sdp_specific2;
e51060f0
SH		 206 union cma_ip_addr src_addr;
207 union cma_ip_addr dst_addr;
208};
209
210struct sdp_hah {
211 u8 bsdh[16];
212 u8 sdp_version;
213};
214
215#define CMA_VERSION 0x00
216#define SDP_MAJ_VERSION 0x2
217
550e5ca7 218static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
e51060f0
SH		 219{
220 unsigned long flags;
221 int ret;
222
223 spin_lock_irqsave(&id_priv->lock, flags);
224 ret = (id_priv->state == comp);
225 spin_unlock_irqrestore(&id_priv->lock, flags);
226 return ret;
227}
228
229static int cma_comp_exch(struct rdma_id_private *id_priv,
550e5ca7 230 enum rdma_cm_state comp, enum rdma_cm_state exch)
e51060f0
SH		 231{
232 unsigned long flags;
233 int ret;
234
235 spin_lock_irqsave(&id_priv->lock, flags);
236 if ((ret = (id_priv->state == comp)))
237 id_priv->state = exch;
238 spin_unlock_irqrestore(&id_priv->lock, flags);
239 return ret;
240}
241
550e5ca7
NM		 242static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
243 enum rdma_cm_state exch)
e51060f0
SH		 244{
245 unsigned long flags;
550e5ca7 246 enum rdma_cm_state old;
e51060f0
SH		 247
248 spin_lock_irqsave(&id_priv->lock, flags);
249 old = id_priv->state;
250 id_priv->state = exch;
251 spin_unlock_irqrestore(&id_priv->lock, flags);
252 return old;
253}
254
255static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
256{
257 return hdr->ip_version >> 4;
258}
259
260static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
261{
262 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
263}
264
265static inline u8 sdp_get_majv(u8 sdp_version)
266{
267 return sdp_version >> 4;
268}
269
270static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
271{
272 return hh->ip_version >> 4;
273}
274
275static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
276{
277 hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
278}
279
280static void cma_attach_to_dev(struct rdma_id_private *id_priv,
281 struct cma_device *cma_dev)
282{
283 atomic_inc(&cma_dev->refcount);
284 id_priv->cma_dev = cma_dev;
285 id_priv->id.device = cma_dev->device;
3c86aa70
EC		 286 id_priv->id.route.addr.dev_addr.transport =
287 rdma_node_get_transport(cma_dev->device->node_type);
e51060f0
SH		 288 list_add_tail(&id_priv->list, &cma_dev->id_list);
289}
290
291static inline void cma_deref_dev(struct cma_device *cma_dev)
292{
293 if (atomic_dec_and_test(&cma_dev->refcount))
294 complete(&cma_dev->comp);
295}
296
3c86aa70
EC		 297static inline void release_mc(struct kref *kref)
298{
299 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
300
301 kfree(mc->multicast.ib);
302 kfree(mc);
303}
304
a396d43a 305static void cma_release_dev(struct rdma_id_private *id_priv)
e51060f0 306{
a396d43a 307 mutex_lock(&lock);
e51060f0
SH		 308 list_del(&id_priv->list);
309 cma_deref_dev(id_priv->cma_dev);
310 id_priv->cma_dev = NULL;
a396d43a 311 mutex_unlock(&lock);
e51060f0
SH		 312}
313
f4753834
SH		 314static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
315{
316 return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
317}
318
319static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
320{
321 return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
322}
323
324static inline unsigned short cma_family(struct rdma_id_private *id_priv)
325{
326 return id_priv->id.route.addr.src_addr.ss_family;
327}
328
d2ca39f2 329static int cma_set_qkey(struct rdma_id_private *id_priv)
c8f6a362
SH		 330{
331 struct ib_sa_mcmember_rec rec;
332 int ret = 0;
333
d2ca39f2
YE		 334 if (id_priv->qkey)
335 return 0;
336
337 switch (id_priv->id.ps) {
c8f6a362 338 case RDMA_PS_UDP:
d2ca39f2 339 id_priv->qkey = RDMA_UDP_QKEY;
c8f6a362
SH		 340 break;
341 case RDMA_PS_IPOIB:
d2ca39f2
YE		 342 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
343 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
344 id_priv->id.port_num, &rec.mgid,
345 &rec);
346 if (!ret)
347 id_priv->qkey = be32_to_cpu(rec.qkey);
c8f6a362
SH		 348 break;
349 default:
350 break;
351 }
352 return ret;
353}
354
3c86aa70
EC		 355static int find_gid_port(struct ib_device *device, union ib_gid *gid, u8 port_num)
356{
357 int i;
358 int err;
359 struct ib_port_attr props;
360 union ib_gid tmp;
361
362 err = ib_query_port(device, port_num, &props);
363 if (err)
63f05be2 364 return err;
3c86aa70
EC		 365
366 for (i = 0; i < props.gid_tbl_len; ++i) {
367 err = ib_query_gid(device, port_num, i, &tmp);
368 if (err)
63f05be2 369 return err;
3c86aa70
EC		 370 if (!memcmp(&tmp, gid, sizeof tmp))
371 return 0;
372 }
373
63f05be2 374 return -EADDRNOTAVAIL;
3c86aa70
EC		 375}
376
680f920a
SH		 377static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
378{
379 dev_addr->dev_type = ARPHRD_INFINIBAND;
380 rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
381 ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
382}
383
384static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
385{
386 int ret;
387
388 if (addr->sa_family != AF_IB) {
389 ret = rdma_translate_ip(addr, dev_addr);
390 } else {
391 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
392 ret = 0;
393 }
394
395 return ret;
396}
397
07ebafba 398static int cma_acquire_dev(struct rdma_id_private *id_priv)
e51060f0 399{
c8f6a362 400 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
e51060f0 401 struct cma_device *cma_dev;
3c86aa70 402 union ib_gid gid, iboe_gid;
e51060f0 403 int ret = -ENODEV;
3c86aa70
EC		 404 u8 port;
405 enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
406 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
e51060f0 407
2efdd6a0
MS		 408 if (dev_ll != IB_LINK_LAYER_INFINIBAND &&
409 id_priv->id.ps == RDMA_PS_IPOIB)
410 return -EINVAL;
411
a396d43a 412 mutex_lock(&lock);
3c86aa70
EC		 413 iboe_addr_get_sgid(dev_addr, &iboe_gid);
414 memcpy(&gid, dev_addr->src_dev_addr +
415 rdma_addr_gid_offset(dev_addr), sizeof gid);
e51060f0 416 list_for_each_entry(cma_dev, &dev_list, list) {
3c86aa70
EC		 417 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
418 if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
419 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
420 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
421 ret = find_gid_port(cma_dev->device, &iboe_gid, port);
422 else
423 ret = find_gid_port(cma_dev->device, &gid, port);
424
425 if (!ret) {
426 id_priv->id.port_num = port;
427 goto out;
63f05be2 428 }
3c86aa70 429 }
e51060f0
SH		 430 }
431 }
3c86aa70
EC		 432
433out:
434 if (!ret)
435 cma_attach_to_dev(id_priv, cma_dev);
436
a396d43a 437 mutex_unlock(&lock);
e51060f0
SH		 438 return ret;
439}
440
e51060f0
SH		 441static void cma_deref_id(struct rdma_id_private *id_priv)
442{
443 if (atomic_dec_and_test(&id_priv->refcount))
444 complete(&id_priv->comp);
445}
446
de910bd9 447static int cma_disable_callback(struct rdma_id_private *id_priv,
550e5ca7 448 enum rdma_cm_state state)
8aa08602 449{
de910bd9
OG		 450 mutex_lock(&id_priv->handler_mutex);
451 if (id_priv->state != state) {
452 mutex_unlock(&id_priv->handler_mutex);
453 return -EINVAL;
454 }
455 return 0;
e51060f0
SH		 456}
457
458struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
b26f9b99
SH		 459 void *context, enum rdma_port_space ps,
460 enum ib_qp_type qp_type)
e51060f0
SH		 461{
462 struct rdma_id_private *id_priv;
463
464 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
465 if (!id_priv)
466 return ERR_PTR(-ENOMEM);
467
83e9502d 468 id_priv->owner = task_pid_nr(current);
550e5ca7 469 id_priv->state = RDMA_CM_IDLE;
e51060f0
SH		 470 id_priv->id.context = context;
471 id_priv->id.event_handler = event_handler;
472 id_priv->id.ps = ps;
b26f9b99 473 id_priv->id.qp_type = qp_type;
e51060f0 474 spin_lock_init(&id_priv->lock);
c5483388 475 mutex_init(&id_priv->qp_mutex);
e51060f0
SH		 476 init_completion(&id_priv->comp);
477 atomic_set(&id_priv->refcount, 1);
de910bd9 478 mutex_init(&id_priv->handler_mutex);
e51060f0 479 INIT_LIST_HEAD(&id_priv->listen_list);
c8f6a362 480 INIT_LIST_HEAD(&id_priv->mc_list);
e51060f0
SH		 481 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
482
483 return &id_priv->id;
484}
485EXPORT_SYMBOL(rdma_create_id);
486
c8f6a362 487static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
e51060f0
SH		 488{
489 struct ib_qp_attr qp_attr;
c8f6a362 490 int qp_attr_mask, ret;
e51060f0 491
c8f6a362
SH		 492 qp_attr.qp_state = IB_QPS_INIT;
493 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0
SH		 494 if (ret)
495 return ret;
496
c8f6a362
SH		 497 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
498 if (ret)
499 return ret;
500
501 qp_attr.qp_state = IB_QPS_RTR;
502 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
503 if (ret)
504 return ret;
505
506 qp_attr.qp_state = IB_QPS_RTS;
507 qp_attr.sq_psn = 0;
508 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
509
510 return ret;
e51060f0
SH		 511}
512
c8f6a362 513static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
07ebafba
TT		 514{
515 struct ib_qp_attr qp_attr;
c8f6a362 516 int qp_attr_mask, ret;
07ebafba
TT		 517
518 qp_attr.qp_state = IB_QPS_INIT;
c8f6a362
SH		 519 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
520 if (ret)
521 return ret;
07ebafba 522
c8f6a362 523 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
07ebafba
TT		 524}
525
e51060f0
SH		 526int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
527 struct ib_qp_init_attr *qp_init_attr)
528{
529 struct rdma_id_private *id_priv;
530 struct ib_qp *qp;
531 int ret;
532
533 id_priv = container_of(id, struct rdma_id_private, id);
534 if (id->device != pd->device)
535 return -EINVAL;
536
537 qp = ib_create_qp(pd, qp_init_attr);
538 if (IS_ERR(qp))
539 return PTR_ERR(qp);
540
b26f9b99 541 if (id->qp_type == IB_QPT_UD)
c8f6a362
SH		 542 ret = cma_init_ud_qp(id_priv, qp);
543 else
544 ret = cma_init_conn_qp(id_priv, qp);
e51060f0
SH		 545 if (ret)
546 goto err;
547
548 id->qp = qp;
549 id_priv->qp_num = qp->qp_num;
e51060f0
SH		 550 id_priv->srq = (qp->srq != NULL);
551 return 0;
552err:
553 ib_destroy_qp(qp);
554 return ret;
555}
556EXPORT_SYMBOL(rdma_create_qp);
557
558void rdma_destroy_qp(struct rdma_cm_id *id)
559{
c5483388
SH		 560 struct rdma_id_private *id_priv;
561
562 id_priv = container_of(id, struct rdma_id_private, id);
563 mutex_lock(&id_priv->qp_mutex);
564 ib_destroy_qp(id_priv->id.qp);
565 id_priv->id.qp = NULL;
566 mutex_unlock(&id_priv->qp_mutex);
e51060f0
SH		 567}
568EXPORT_SYMBOL(rdma_destroy_qp);
569
5851bb89
SH		 570static int cma_modify_qp_rtr(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 /* Need to update QP attributes from default values. */
583 qp_attr.qp_state = IB_QPS_INIT;
c5483388 584 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 585 if (ret)
c5483388 586 goto out;
e51060f0 587
c5483388 588 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
e51060f0 589 if (ret)
c5483388 590 goto out;
e51060f0
SH		 591
592 qp_attr.qp_state = IB_QPS_RTR;
c5483388 593 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 594 if (ret)
c5483388 595 goto out;
e51060f0 596
5851bb89
SH		 597 if (conn_param)
598 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
c5483388
SH		 599 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
600out:
601 mutex_unlock(&id_priv->qp_mutex);
602 return ret;
e51060f0
SH		 603}
604
5851bb89
SH		 605static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
606 struct rdma_conn_param *conn_param)
e51060f0
SH		 607{
608 struct ib_qp_attr qp_attr;
609 int qp_attr_mask, ret;
610
c5483388
SH		 611 mutex_lock(&id_priv->qp_mutex);
612 if (!id_priv->id.qp) {
613 ret = 0;
614 goto out;
615 }
e51060f0
SH		 616
617 qp_attr.qp_state = IB_QPS_RTS;
c5483388 618 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 619 if (ret)
c5483388 620 goto out;
e51060f0 621
5851bb89
SH		 622 if (conn_param)
623 qp_attr.max_rd_atomic = conn_param->initiator_depth;
c5483388
SH		 624 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
625out:
626 mutex_unlock(&id_priv->qp_mutex);
627 return ret;
e51060f0
SH		 628}
629
c5483388 630static int cma_modify_qp_err(struct rdma_id_private *id_priv)
e51060f0
SH		 631{
632 struct ib_qp_attr qp_attr;
c5483388 633 int ret;
e51060f0 634
c5483388
SH		 635 mutex_lock(&id_priv->qp_mutex);
636 if (!id_priv->id.qp) {
637 ret = 0;
638 goto out;
639 }
e51060f0
SH		 640
641 qp_attr.qp_state = IB_QPS_ERR;
c5483388
SH		 642 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
643out:
644 mutex_unlock(&id_priv->qp_mutex);
645 return ret;
e51060f0
SH		 646}
647
c8f6a362
SH		 648static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
649 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
650{
651 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
652 int ret;
3c86aa70
EC		 653 u16 pkey;
654
655 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
656 IB_LINK_LAYER_INFINIBAND)
657 pkey = ib_addr_get_pkey(dev_addr);
658 else
659 pkey = 0xffff;
c8f6a362
SH		 660
661 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
3c86aa70 662 pkey, &qp_attr->pkey_index);
c8f6a362
SH		 663 if (ret)
664 return ret;
665
666 qp_attr->port_num = id_priv->id.port_num;
667 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
668
b26f9b99 669 if (id_priv->id.qp_type == IB_QPT_UD) {
d2ca39f2
YE		 670 ret = cma_set_qkey(id_priv);
671 if (ret)
672 return ret;
673
c8f6a362
SH		 674 qp_attr->qkey = id_priv->qkey;
675 *qp_attr_mask |= IB_QP_QKEY;
676 } else {
677 qp_attr->qp_access_flags = 0;
678 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
679 }
680 return 0;
681}
682
e51060f0
SH		 683int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
684 int *qp_attr_mask)
685{
686 struct rdma_id_private *id_priv;
c8f6a362 687 int ret = 0;
e51060f0
SH		 688
689 id_priv = container_of(id, struct rdma_id_private, id);
07ebafba
TT		 690 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
691 case RDMA_TRANSPORT_IB:
b26f9b99 692 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
c8f6a362
SH		 693 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
694 else
695 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
696 qp_attr_mask);
e51060f0
SH		 697 if (qp_attr->qp_state == IB_QPS_RTR)
698 qp_attr->rq_psn = id_priv->seq_num;
699 break;
07ebafba 700 case RDMA_TRANSPORT_IWARP:
c8f6a362 701 if (!id_priv->cm_id.iw) {
8f076531 702 qp_attr->qp_access_flags = 0;
c8f6a362
SH		 703 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
704 } else
705 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
706 qp_attr_mask);
07ebafba 707 break;
e51060f0
SH		 708 default:
709 ret = -ENOSYS;
710 break;
711 }
712
713 return ret;
714}
715EXPORT_SYMBOL(rdma_init_qp_attr);
716
717static inline int cma_zero_addr(struct sockaddr *addr)
718{
2e2d190c
SH		 719 switch (addr->sa_family) {
720 case AF_INET:
721 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
722 case AF_INET6:
723 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
724 case AF_IB:
725 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
726 default:
727 return 0;
e51060f0
SH		 728 }
729}
730
731static inline int cma_loopback_addr(struct sockaddr *addr)
732{
2e2d190c
SH		 733 switch (addr->sa_family) {
734 case AF_INET:
735 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
736 case AF_INET6:
737 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
738 case AF_IB:
739 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
740 default:
741 return 0;
742 }
e51060f0
SH		 743}
744
745static inline int cma_any_addr(struct sockaddr *addr)
746{
747 return cma_zero_addr(addr) || cma_loopback_addr(addr);
748}
749
43b752da
HS		 750static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
751{
752 if (src->sa_family != dst->sa_family)
753 return -1;
754
755 switch (src->sa_family) {
756 case AF_INET:
757 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
758 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
2e2d190c 759 case AF_INET6:
43b752da
HS		 760 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
761 &((struct sockaddr_in6 *) dst)->sin6_addr);
2e2d190c
SH		 762 default:
763 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
764 &((struct sockaddr_ib *) dst)->sib_addr);
43b752da
HS		 765 }
766}
767
58afdcb7 768static __be16 cma_port(struct sockaddr *addr)
628e5f6d 769{
58afdcb7
SH		 770 struct sockaddr_ib *sib;
771
772 switch (addr->sa_family) {
773 case AF_INET:
628e5f6d 774 return ((struct sockaddr_in *) addr)->sin_port;
58afdcb7 775 case AF_INET6:
628e5f6d 776 return ((struct sockaddr_in6 *) addr)->sin6_port;
58afdcb7
SH		 777 case AF_IB:
778 sib = (struct sockaddr_ib *) addr;
779 return htons((u16) (be64_to_cpu(sib->sib_sid) &
780 be64_to_cpu(sib->sib_sid_mask)));
781 default:
782 return 0;
783 }
628e5f6d
SH		 784}
785
e51060f0
SH		 786static inline int cma_any_port(struct sockaddr *addr)
787{
628e5f6d 788 return !cma_port(addr);
e51060f0
SH		 789}
790
791static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
1b90c137 792 u8 *ip_ver, __be16 *port,
e51060f0
SH		 793 union cma_ip_addr **src, union cma_ip_addr **dst)
794{
795 switch (ps) {
796 case RDMA_PS_SDP:
797 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
798 SDP_MAJ_VERSION)
799 return -EINVAL;
800
801 *ip_ver = sdp_get_ip_ver(hdr);
802 *port = ((struct sdp_hh *) hdr)->port;
803 *src = &((struct sdp_hh *) hdr)->src_addr;
804 *dst = &((struct sdp_hh *) hdr)->dst_addr;
805 break;
806 default:
807 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
808 return -EINVAL;
809
810 *ip_ver = cma_get_ip_ver(hdr);
811 *port = ((struct cma_hdr *) hdr)->port;
812 *src = &((struct cma_hdr *) hdr)->src_addr;
813 *dst = &((struct cma_hdr *) hdr)->dst_addr;
814 break;
815 }
816
817 if (*ip_ver != 4 && *ip_ver != 6)
818 return -EINVAL;
819 return 0;
820}
821
822static void cma_save_net_info(struct rdma_addr *addr,
823 struct rdma_addr *listen_addr,
1b90c137 824 u8 ip_ver, __be16 port,
e51060f0
SH		 825 union cma_ip_addr *src, union cma_ip_addr *dst)
826{
827 struct sockaddr_in *listen4, *ip4;
828 struct sockaddr_in6 *listen6, *ip6;
829
830 switch (ip_ver) {
831 case 4:
832 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
833 ip4 = (struct sockaddr_in *) &addr->src_addr;
834 ip4->sin_family = listen4->sin_family;
835 ip4->sin_addr.s_addr = dst->ip4.addr;
836 ip4->sin_port = listen4->sin_port;
837
838 ip4 = (struct sockaddr_in *) &addr->dst_addr;
839 ip4->sin_family = listen4->sin_family;
840 ip4->sin_addr.s_addr = src->ip4.addr;
841 ip4->sin_port = port;
842 break;
843 case 6:
844 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
845 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
846 ip6->sin6_family = listen6->sin6_family;
847 ip6->sin6_addr = dst->ip6;
848 ip6->sin6_port = listen6->sin6_port;
849
850 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
851 ip6->sin6_family = listen6->sin6_family;
852 ip6->sin6_addr = src->ip6;
853 ip6->sin6_port = port;
854 break;
855 default:
856 break;
857 }
858}
859
860static inline int cma_user_data_offset(enum rdma_port_space ps)
861{
862 switch (ps) {
863 case RDMA_PS_SDP:
864 return 0;
865 default:
866 return sizeof(struct cma_hdr);
867 }
868}
869
e51060f0
SH		 870static void cma_cancel_route(struct rdma_id_private *id_priv)
871{
3c86aa70
EC		 872 switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
873 case IB_LINK_LAYER_INFINIBAND:
e51060f0
SH		 874 if (id_priv->query)
875 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
876 break;
877 default:
878 break;
879 }
880}
881
e51060f0
SH		 882static void cma_cancel_listens(struct rdma_id_private *id_priv)
883{
884 struct rdma_id_private *dev_id_priv;
885
d02d1f53
SH		 886 /*
887 * Remove from listen_any_list to prevent added devices from spawning
888 * additional listen requests.
889 */
e51060f0
SH		 890 mutex_lock(&lock);
891 list_del(&id_priv->list);
892
893 while (!list_empty(&id_priv->listen_list)) {
894 dev_id_priv = list_entry(id_priv->listen_list.next,
895 struct rdma_id_private, listen_list);
d02d1f53
SH		 896 /* sync with device removal to avoid duplicate destruction */
897 list_del_init(&dev_id_priv->list);
898 list_del(&dev_id_priv->listen_list);
899 mutex_unlock(&lock);
900
901 rdma_destroy_id(&dev_id_priv->id);
902 mutex_lock(&lock);
e51060f0
SH		 903 }
904 mutex_unlock(&lock);
905}
906
907static void cma_cancel_operation(struct rdma_id_private *id_priv,
550e5ca7 908 enum rdma_cm_state state)
e51060f0
SH		 909{
910 switch (state) {
550e5ca7 911 case RDMA_CM_ADDR_QUERY:
e51060f0
SH		 912 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
913 break;
550e5ca7 914 case RDMA_CM_ROUTE_QUERY:
e51060f0
SH		 915 cma_cancel_route(id_priv);
916 break;
550e5ca7 917 case RDMA_CM_LISTEN:
f4753834 918 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
e51060f0
SH		 919 cma_cancel_listens(id_priv);
920 break;
921 default:
922 break;
923 }
924}
925
926static void cma_release_port(struct rdma_id_private *id_priv)
927{
928 struct rdma_bind_list *bind_list = id_priv->bind_list;
929
930 if (!bind_list)
931 return;
932
933 mutex_lock(&lock);
934 hlist_del(&id_priv->node);
935 if (hlist_empty(&bind_list->owners)) {
936 idr_remove(bind_list->ps, bind_list->port);
937 kfree(bind_list);
938 }
939 mutex_unlock(&lock);
940}
941
c8f6a362
SH		 942static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
943{
944 struct cma_multicast *mc;
945
946 while (!list_empty(&id_priv->mc_list)) {
947 mc = container_of(id_priv->mc_list.next,
948 struct cma_multicast, list);
949 list_del(&mc->list);
3c86aa70
EC		 950 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
951 case IB_LINK_LAYER_INFINIBAND:
952 ib_sa_free_multicast(mc->multicast.ib);
953 kfree(mc);
954 break;
955 case IB_LINK_LAYER_ETHERNET:
956 kref_put(&mc->mcref, release_mc);
957 break;
958 default:
959 break;
960 }
c8f6a362
SH		 961 }
962}
963
e51060f0
SH		 964void rdma_destroy_id(struct rdma_cm_id *id)
965{
966 struct rdma_id_private *id_priv;
550e5ca7 967 enum rdma_cm_state state;
e51060f0
SH		 968
969 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 970 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
e51060f0
SH		 971 cma_cancel_operation(id_priv, state);
972
a396d43a
SH		 973 /*
974 * Wait for any active callback to finish. New callbacks will find
975 * the id_priv state set to destroying and abort.
976 */
977 mutex_lock(&id_priv->handler_mutex);
978 mutex_unlock(&id_priv->handler_mutex);
979
e51060f0 980 if (id_priv->cma_dev) {
3c86aa70 981 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
07ebafba 982 case RDMA_TRANSPORT_IB:
0c9361fc 983 if (id_priv->cm_id.ib)
e51060f0
SH		 984 ib_destroy_cm_id(id_priv->cm_id.ib);
985 break;
07ebafba 986 case RDMA_TRANSPORT_IWARP:
0c9361fc 987 if (id_priv->cm_id.iw)
07ebafba
TT		 988 iw_destroy_cm_id(id_priv->cm_id.iw);
989 break;
e51060f0
SH		 990 default:
991 break;
992 }
c8f6a362 993 cma_leave_mc_groups(id_priv);
a396d43a 994 cma_release_dev(id_priv);
e51060f0
SH		 995 }
996
997 cma_release_port(id_priv);
998 cma_deref_id(id_priv);
999 wait_for_completion(&id_priv->comp);
1000
d02d1f53
SH		 1001 if (id_priv->internal_id)
1002 cma_deref_id(id_priv->id.context);
1003
e51060f0
SH		 1004 kfree(id_priv->id.route.path_rec);
1005 kfree(id_priv);
1006}
1007EXPORT_SYMBOL(rdma_destroy_id);
1008
1009static int cma_rep_recv(struct rdma_id_private *id_priv)
1010{
1011 int ret;
1012
5851bb89 1013 ret = cma_modify_qp_rtr(id_priv, NULL);
e51060f0
SH		 1014 if (ret)
1015 goto reject;
1016
5851bb89 1017 ret = cma_modify_qp_rts(id_priv, NULL);
e51060f0
SH		 1018 if (ret)
1019 goto reject;
1020
1021 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1022 if (ret)
1023 goto reject;
1024
1025 return 0;
1026reject:
c5483388 1027 cma_modify_qp_err(id_priv);
e51060f0
SH		 1028 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1029 NULL, 0, NULL, 0);
1030 return ret;
1031}
1032
1033static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
1034{
1035 if (id_priv->id.ps == RDMA_PS_SDP &&
1036 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
1037 SDP_MAJ_VERSION)
1038 return -EINVAL;
1039
1040 return 0;
1041}
1042
a1b1b61f
SH		 1043static void cma_set_rep_event_data(struct rdma_cm_event *event,
1044 struct ib_cm_rep_event_param *rep_data,
1045 void *private_data)
1046{
1047 event->param.conn.private_data = private_data;
1048 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1049 event->param.conn.responder_resources = rep_data->responder_resources;
1050 event->param.conn.initiator_depth = rep_data->initiator_depth;
1051 event->param.conn.flow_control = rep_data->flow_control;
1052 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1053 event->param.conn.srq = rep_data->srq;
1054 event->param.conn.qp_num = rep_data->remote_qpn;
1055}
1056
e51060f0
SH		 1057static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1058{
1059 struct rdma_id_private *id_priv = cm_id->context;
a1b1b61f
SH		 1060 struct rdma_cm_event event;
1061 int ret = 0;
e51060f0 1062
38ca83a5 1063 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
550e5ca7 1064 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
38ca83a5 1065 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
550e5ca7 1066 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
8aa08602 1067 return 0;
e51060f0 1068
a1b1b61f 1069 memset(&event, 0, sizeof event);
e51060f0
SH		 1070 switch (ib_event->event) {
1071 case IB_CM_REQ_ERROR:
1072 case IB_CM_REP_ERROR:
a1b1b61f
SH		 1073 event.event = RDMA_CM_EVENT_UNREACHABLE;
1074 event.status = -ETIMEDOUT;
e51060f0
SH		 1075 break;
1076 case IB_CM_REP_RECEIVED:
a1b1b61f
SH		 1077 event.status = cma_verify_rep(id_priv, ib_event->private_data);
1078 if (event.status)
1079 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
e51060f0 1080 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
a1b1b61f
SH		 1081 event.status = cma_rep_recv(id_priv);
1082 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1083 RDMA_CM_EVENT_ESTABLISHED;
e51060f0 1084 } else
a1b1b61f
SH		 1085 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1086 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1087 ib_event->private_data);
e51060f0
SH		 1088 break;
1089 case IB_CM_RTU_RECEIVED:
0fe313b0
SH		 1090 case IB_CM_USER_ESTABLISHED:
1091 event.event = RDMA_CM_EVENT_ESTABLISHED;
e51060f0
SH		 1092 break;
1093 case IB_CM_DREQ_ERROR:
a1b1b61f 1094 event.status = -ETIMEDOUT; /* fall through */
e51060f0
SH		 1095 case IB_CM_DREQ_RECEIVED:
1096 case IB_CM_DREP_RECEIVED:
550e5ca7
NM		 1097 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1098 RDMA_CM_DISCONNECT))
e51060f0 1099 goto out;
a1b1b61f 1100 event.event = RDMA_CM_EVENT_DISCONNECTED;
e51060f0
SH		 1101 break;
1102 case IB_CM_TIMEWAIT_EXIT:
38ca83a5
AV		 1103 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1104 break;
e51060f0
SH		 1105 case IB_CM_MRA_RECEIVED:
1106 /* ignore event */
1107 goto out;
1108 case IB_CM_REJ_RECEIVED:
c5483388 1109 cma_modify_qp_err(id_priv);
a1b1b61f
SH		 1110 event.status = ib_event->param.rej_rcvd.reason;
1111 event.event = RDMA_CM_EVENT_REJECTED;
1112 event.param.conn.private_data = ib_event->private_data;
1113 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
e51060f0
SH		 1114 break;
1115 default:
468f2239 1116 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
e51060f0
SH		 1117 ib_event->event);
1118 goto out;
1119 }
1120
a1b1b61f 1121 ret = id_priv->id.event_handler(&id_priv->id, &event);
e51060f0
SH		 1122 if (ret) {
1123 /* Destroy the CM ID by returning a non-zero value. */
1124 id_priv->cm_id.ib = NULL;
550e5ca7 1125 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 1126 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1127 rdma_destroy_id(&id_priv->id);
1128 return ret;
1129 }
1130out:
de910bd9 1131 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1132 return ret;
1133}
1134
628e5f6d
SH		 1135static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1136 struct ib_cm_event *ib_event)
e51060f0
SH		 1137{
1138 struct rdma_id_private *id_priv;
1139 struct rdma_cm_id *id;
1140 struct rdma_route *rt;
1141 union cma_ip_addr *src, *dst;
1b90c137 1142 __be16 port;
e51060f0 1143 u8 ip_ver;
64c5e613 1144 int ret;
e51060f0 1145
3f168d2b
KK		 1146 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1147 &ip_ver, &port, &src, &dst))
0c9361fc 1148 return NULL;
3f168d2b 1149
e51060f0 1150 id = rdma_create_id(listen_id->event_handler, listen_id->context,
b26f9b99 1151 listen_id->ps, ib_event->param.req_rcvd.qp_type);
e51060f0 1152 if (IS_ERR(id))
0c9361fc 1153 return NULL;
3f168d2b 1154
f4753834 1155 id_priv = container_of(id, struct rdma_id_private, id);
3f168d2b
KK		 1156 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1157 ip_ver, port, src, dst);
e51060f0
SH		 1158
1159 rt = &id->route;
1160 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
3f168d2b
KK		 1161 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1162 GFP_KERNEL);
e51060f0 1163 if (!rt->path_rec)
0c9361fc 1164 goto err;
e51060f0 1165
e51060f0
SH		 1166 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1167 if (rt->num_paths == 2)
1168 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1169
f4753834 1170 if (cma_any_addr(cma_src_addr(id_priv))) {
6f8372b6
SH		 1171 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1172 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
46ea5061 1173 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
6f8372b6 1174 } else {
f4753834 1175 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
6f8372b6 1176 if (ret)
0c9361fc 1177 goto err;
6f8372b6
SH		 1178 }
1179 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
e51060f0 1180
550e5ca7 1181 id_priv->state = RDMA_CM_CONNECT;
e51060f0 1182 return id_priv;
3f168d2b 1183
3f168d2b 1184err:
0c9361fc 1185 rdma_destroy_id(id);
e51060f0
SH		 1186 return NULL;
1187}
1188
628e5f6d
SH		 1189static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1190 struct ib_cm_event *ib_event)
1191{
1192 struct rdma_id_private *id_priv;
1193 struct rdma_cm_id *id;
1194 union cma_ip_addr *src, *dst;
1b90c137 1195 __be16 port;
628e5f6d
SH		 1196 u8 ip_ver;
1197 int ret;
1198
1199 id = rdma_create_id(listen_id->event_handler, listen_id->context,
b26f9b99 1200 listen_id->ps, IB_QPT_UD);
628e5f6d
SH		 1201 if (IS_ERR(id))
1202 return NULL;
1203
f4753834 1204 id_priv = container_of(id, struct rdma_id_private, id);
628e5f6d
SH		 1205 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1206 &ip_ver, &port, &src, &dst))
1207 goto err;
1208
1209 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1210 ip_ver, port, src, dst);
1211
6f8372b6 1212 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
f4753834 1213 ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
6f8372b6
SH		 1214 if (ret)
1215 goto err;
1216 }
628e5f6d 1217
550e5ca7 1218 id_priv->state = RDMA_CM_CONNECT;
628e5f6d
SH		 1219 return id_priv;
1220err:
1221 rdma_destroy_id(id);
1222 return NULL;
1223}
1224
a1b1b61f
SH		 1225static void cma_set_req_event_data(struct rdma_cm_event *event,
1226 struct ib_cm_req_event_param *req_data,
1227 void *private_data, int offset)
1228{
1229 event->param.conn.private_data = private_data + offset;
1230 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1231 event->param.conn.responder_resources = req_data->responder_resources;
1232 event->param.conn.initiator_depth = req_data->initiator_depth;
1233 event->param.conn.flow_control = req_data->flow_control;
1234 event->param.conn.retry_count = req_data->retry_count;
1235 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1236 event->param.conn.srq = req_data->srq;
1237 event->param.conn.qp_num = req_data->remote_qpn;
1238}
1239
9595480c
HS		 1240static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1241{
4dd81e89 1242 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
9595480c
HS		 1243 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1244 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1245 (id->qp_type == IB_QPT_UD)) ||
1246 (!id->qp_type));
1247}
1248
e51060f0
SH		 1249static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1250{
1251 struct rdma_id_private *listen_id, *conn_id;
a1b1b61f 1252 struct rdma_cm_event event;
e51060f0
SH		 1253 int offset, ret;
1254
1255 listen_id = cm_id->context;
9595480c
HS		 1256 if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1257 return -EINVAL;
1258
550e5ca7 1259 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
8aa08602 1260 return -ECONNABORTED;
e51060f0 1261
628e5f6d
SH		 1262 memset(&event, 0, sizeof event);
1263 offset = cma_user_data_offset(listen_id->id.ps);
1264 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
9595480c 1265 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
628e5f6d
SH		 1266 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1267 event.param.ud.private_data = ib_event->private_data + offset;
1268 event.param.ud.private_data_len =
1269 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1270 } else {
1271 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1272 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1273 ib_event->private_data, offset);
1274 }
e51060f0
SH		 1275 if (!conn_id) {
1276 ret = -ENOMEM;
b6cec8aa 1277 goto err1;
e51060f0
SH		 1278 }
1279
de910bd9 1280 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
07ebafba 1281 ret = cma_acquire_dev(conn_id);
a1a733f6 1282 if (ret)
b6cec8aa 1283 goto err2;
e51060f0
SH		 1284
1285 conn_id->cm_id.ib = cm_id;
1286 cm_id->context = conn_id;
1287 cm_id->cm_handler = cma_ib_handler;
1288
25ae21a1
SH		 1289 /*
1290 * Protect against the user destroying conn_id from another thread
1291 * until we're done accessing it.
1292 */
1293 atomic_inc(&conn_id->refcount);
a1b1b61f 1294 ret = conn_id->id.event_handler(&conn_id->id, &event);
b6cec8aa
SH		 1295 if (ret)
1296 goto err3;
1297
1298 /*
1299 * Acquire mutex to prevent user executing rdma_destroy_id()
1300 * while we're accessing the cm_id.
1301 */
1302 mutex_lock(&lock);
1303 if (cma_comp(conn_id, RDMA_CM_CONNECT) && (conn_id->id.qp_type != IB_QPT_UD))
1304 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1305 mutex_unlock(&lock);
1306 mutex_unlock(&conn_id->handler_mutex);
1307 mutex_unlock(&listen_id->handler_mutex);
25ae21a1 1308 cma_deref_id(conn_id);
b6cec8aa 1309 return 0;
a1a733f6 1310
b6cec8aa
SH		 1311err3:
1312 cma_deref_id(conn_id);
a1a733f6
KK		 1313 /* Destroy the CM ID by returning a non-zero value. */
1314 conn_id->cm_id.ib = NULL;
b6cec8aa 1315err2:
550e5ca7 1316 cma_exch(conn_id, RDMA_CM_DESTROYING);
de910bd9 1317 mutex_unlock(&conn_id->handler_mutex);
b6cec8aa 1318err1:
de910bd9 1319 mutex_unlock(&listen_id->handler_mutex);
b6cec8aa
SH		 1320 if (conn_id)
1321 rdma_destroy_id(&conn_id->id);
e51060f0
SH		 1322 return ret;
1323}
1324
1325static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1326{
628e5f6d 1327 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
e51060f0
SH		 1328}
1329
1330static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1331 struct ib_cm_compare_data *compare)
1332{
1333 struct cma_hdr *cma_data, *cma_mask;
1334 struct sdp_hh *sdp_data, *sdp_mask;
1b90c137 1335 __be32 ip4_addr;
e51060f0
SH		 1336 struct in6_addr ip6_addr;
1337
1338 memset(compare, 0, sizeof *compare);
1339 cma_data = (void *) compare->data;
1340 cma_mask = (void *) compare->mask;
1341 sdp_data = (void *) compare->data;
1342 sdp_mask = (void *) compare->mask;
1343
1344 switch (addr->sa_family) {
1345 case AF_INET:
1346 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1347 if (ps == RDMA_PS_SDP) {
1348 sdp_set_ip_ver(sdp_data, 4);
1349 sdp_set_ip_ver(sdp_mask, 0xF);
1350 sdp_data->dst_addr.ip4.addr = ip4_addr;
1b90c137 1351 sdp_mask->dst_addr.ip4.addr = htonl(~0);
e51060f0
SH		 1352 } else {
1353 cma_set_ip_ver(cma_data, 4);
1354 cma_set_ip_ver(cma_mask, 0xF);
406b6a25
SH		 1355 if (!cma_any_addr(addr)) {
1356 cma_data->dst_addr.ip4.addr = ip4_addr;
1357 cma_mask->dst_addr.ip4.addr = htonl(~0);
1358 }
e51060f0
SH		 1359 }
1360 break;
1361 case AF_INET6:
1362 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1363 if (ps == RDMA_PS_SDP) {
1364 sdp_set_ip_ver(sdp_data, 6);
1365 sdp_set_ip_ver(sdp_mask, 0xF);
1366 sdp_data->dst_addr.ip6 = ip6_addr;
1367 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1368 sizeof sdp_mask->dst_addr.ip6);
1369 } else {
1370 cma_set_ip_ver(cma_data, 6);
1371 cma_set_ip_ver(cma_mask, 0xF);
406b6a25
SH		 1372 if (!cma_any_addr(addr)) {
1373 cma_data->dst_addr.ip6 = ip6_addr;
1374 memset(&cma_mask->dst_addr.ip6, 0xFF,
1375 sizeof cma_mask->dst_addr.ip6);
1376 }
e51060f0
SH		 1377 }
1378 break;
1379 default:
1380 break;
1381 }
1382}
1383
07ebafba
TT		 1384static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1385{
1386 struct rdma_id_private *id_priv = iw_id->context;
a1b1b61f 1387 struct rdma_cm_event event;
07ebafba
TT		 1388 struct sockaddr_in *sin;
1389 int ret = 0;
1390
550e5ca7 1391 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
be65f086 1392 return 0;
07ebafba 1393
be65f086 1394 memset(&event, 0, sizeof event);
07ebafba
TT		 1395 switch (iw_event->event) {
1396 case IW_CM_EVENT_CLOSE:
a1b1b61f 1397 event.event = RDMA_CM_EVENT_DISCONNECTED;
07ebafba
TT		 1398 break;
1399 case IW_CM_EVENT_CONNECT_REPLY:
f4753834 1400 sin = (struct sockaddr_in *) cma_src_addr(id_priv);
07ebafba 1401 *sin = iw_event->local_addr;
f4753834 1402 sin = (struct sockaddr_in *) cma_dst_addr(id_priv);
07ebafba 1403 *sin = iw_event->remote_addr;
881a045f
SW		 1404 switch (iw_event->status) {
1405 case 0:
a1b1b61f 1406 event.event = RDMA_CM_EVENT_ESTABLISHED;
3ebeebc3
KS		 1407 event.param.conn.initiator_depth = iw_event->ird;
1408 event.param.conn.responder_resources = iw_event->ord;
881a045f
SW		 1409 break;
1410 case -ECONNRESET:
1411 case -ECONNREFUSED:
1412 event.event = RDMA_CM_EVENT_REJECTED;
1413 break;
1414 case -ETIMEDOUT:
1415 event.event = RDMA_CM_EVENT_UNREACHABLE;
1416 break;
1417 default:
1418 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1419 break;
1420 }
07ebafba
TT		 1421 break;
1422 case IW_CM_EVENT_ESTABLISHED:
a1b1b61f 1423 event.event = RDMA_CM_EVENT_ESTABLISHED;
3ebeebc3
KS		 1424 event.param.conn.initiator_depth = iw_event->ird;
1425 event.param.conn.responder_resources = iw_event->ord;
07ebafba
TT		 1426 break;
1427 default:
1428 BUG_ON(1);
1429 }
1430
a1b1b61f
SH		 1431 event.status = iw_event->status;
1432 event.param.conn.private_data = iw_event->private_data;
1433 event.param.conn.private_data_len = iw_event->private_data_len;
1434 ret = id_priv->id.event_handler(&id_priv->id, &event);
07ebafba
TT		 1435 if (ret) {
1436 /* Destroy the CM ID by returning a non-zero value. */
1437 id_priv->cm_id.iw = NULL;
550e5ca7 1438 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 1439 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 1440 rdma_destroy_id(&id_priv->id);
1441 return ret;
1442 }
1443
de910bd9 1444 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 1445 return ret;
1446}
1447
1448static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1449 struct iw_cm_event *iw_event)
1450{
1451 struct rdma_cm_id *new_cm_id;
1452 struct rdma_id_private *listen_id, *conn_id;
1453 struct sockaddr_in *sin;
1454 struct net_device *dev = NULL;
a1b1b61f 1455 struct rdma_cm_event event;
07ebafba 1456 int ret;
8d8293cf 1457 struct ib_device_attr attr;
07ebafba
TT		 1458
1459 listen_id = cm_id->context;
550e5ca7 1460 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
8aa08602 1461 return -ECONNABORTED;
07ebafba
TT		 1462
1463 /* Create a new RDMA id for the new IW CM ID */
1464 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1465 listen_id->id.context,
b26f9b99 1466 RDMA_PS_TCP, IB_QPT_RC);
10f32065 1467 if (IS_ERR(new_cm_id)) {
07ebafba
TT		 1468 ret = -ENOMEM;
1469 goto out;
1470 }
1471 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
de910bd9 1472 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
550e5ca7 1473 conn_id->state = RDMA_CM_CONNECT;
07ebafba 1474
1ab35276 1475 dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
07ebafba
TT		 1476 if (!dev) {
1477 ret = -EADDRNOTAVAIL;
de910bd9 1478 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1479 rdma_destroy_id(new_cm_id);
1480 goto out;
1481 }
1482 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1483 if (ret) {
de910bd9 1484 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1485 rdma_destroy_id(new_cm_id);
1486 goto out;
1487 }
1488
1489 ret = cma_acquire_dev(conn_id);
1490 if (ret) {
de910bd9 1491 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1492 rdma_destroy_id(new_cm_id);
1493 goto out;
1494 }
1495
1496 conn_id->cm_id.iw = cm_id;
1497 cm_id->context = conn_id;
1498 cm_id->cm_handler = cma_iw_handler;
1499
f4753834 1500 sin = (struct sockaddr_in *) cma_src_addr(conn_id);
07ebafba 1501 *sin = iw_event->local_addr;
f4753834 1502 sin = (struct sockaddr_in *) cma_dst_addr(conn_id);
07ebafba
TT		 1503 *sin = iw_event->remote_addr;
1504
8d8293cf
SW		 1505 ret = ib_query_device(conn_id->id.device, &attr);
1506 if (ret) {
de910bd9 1507 mutex_unlock(&conn_id->handler_mutex);
8d8293cf
SW		 1508 rdma_destroy_id(new_cm_id);
1509 goto out;
1510 }
1511
a1b1b61f
SH		 1512 memset(&event, 0, sizeof event);
1513 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1514 event.param.conn.private_data = iw_event->private_data;
1515 event.param.conn.private_data_len = iw_event->private_data_len;
3ebeebc3
KS		 1516 event.param.conn.initiator_depth = iw_event->ird;
1517 event.param.conn.responder_resources = iw_event->ord;
25ae21a1
SH		 1518
1519 /*
1520 * Protect against the user destroying conn_id from another thread
1521 * until we're done accessing it.
1522 */
1523 atomic_inc(&conn_id->refcount);
a1b1b61f 1524 ret = conn_id->id.event_handler(&conn_id->id, &event);
07ebafba
TT		 1525 if (ret) {
1526 /* User wants to destroy the CM ID */
1527 conn_id->cm_id.iw = NULL;
550e5ca7 1528 cma_exch(conn_id, RDMA_CM_DESTROYING);
de910bd9 1529 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1530 cma_deref_id(conn_id);
07ebafba 1531 rdma_destroy_id(&conn_id->id);
de910bd9 1532 goto out;
07ebafba
TT		 1533 }
1534
de910bd9 1535 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1536 cma_deref_id(conn_id);
de910bd9 1537
07ebafba
TT		 1538out:
1539 if (dev)
1540 dev_put(dev);
de910bd9 1541 mutex_unlock(&listen_id->handler_mutex);
07ebafba
TT		 1542 return ret;
1543}
1544
e51060f0
SH		 1545static int cma_ib_listen(struct rdma_id_private *id_priv)
1546{
1547 struct ib_cm_compare_data compare_data;
1548 struct sockaddr *addr;
0c9361fc 1549 struct ib_cm_id *id;
e51060f0
SH		 1550 __be64 svc_id;
1551 int ret;
1552
0c9361fc
JM		 1553 id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1554 if (IS_ERR(id))
1555 return PTR_ERR(id);
1556
1557 id_priv->cm_id.ib = id;
e51060f0 1558
f4753834 1559 addr = cma_src_addr(id_priv);
e51060f0 1560 svc_id = cma_get_service_id(id_priv->id.ps, addr);
406b6a25 1561 if (cma_any_addr(addr) && !id_priv->afonly)
e51060f0
SH		 1562 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1563 else {
1564 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1565 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1566 }
1567
1568 if (ret) {
1569 ib_destroy_cm_id(id_priv->cm_id.ib);
1570 id_priv->cm_id.ib = NULL;
1571 }
1572
1573 return ret;
1574}
1575
07ebafba
TT		 1576static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1577{
1578 int ret;
1579 struct sockaddr_in *sin;
0c9361fc
JM		 1580 struct iw_cm_id *id;
1581
1582 id = iw_create_cm_id(id_priv->id.device,
1583 iw_conn_req_handler,
1584 id_priv);
1585 if (IS_ERR(id))
1586 return PTR_ERR(id);
07ebafba 1587
0c9361fc 1588 id_priv->cm_id.iw = id;
07ebafba 1589
f4753834 1590 sin = (struct sockaddr_in *) cma_src_addr(id_priv);
07ebafba
TT		 1591 id_priv->cm_id.iw->local_addr = *sin;
1592
1593 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1594
1595 if (ret) {
1596 iw_destroy_cm_id(id_priv->cm_id.iw);
1597 id_priv->cm_id.iw = NULL;
1598 }
1599
1600 return ret;
1601}
1602
e51060f0
SH		 1603static int cma_listen_handler(struct rdma_cm_id *id,
1604 struct rdma_cm_event *event)
1605{
1606 struct rdma_id_private *id_priv = id->context;
1607
1608 id->context = id_priv->id.context;
1609 id->event_handler = id_priv->id.event_handler;
1610 return id_priv->id.event_handler(id, event);
1611}
1612
1613static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1614 struct cma_device *cma_dev)
1615{
1616 struct rdma_id_private *dev_id_priv;
1617 struct rdma_cm_id *id;
1618 int ret;
1619
b26f9b99
SH		 1620 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1621 id_priv->id.qp_type);
e51060f0
SH		 1622 if (IS_ERR(id))
1623 return;
1624
1625 dev_id_priv = container_of(id, struct rdma_id_private, id);
1626
550e5ca7 1627 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
f4753834
SH		 1628 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1629 rdma_addr_size(cma_src_addr(id_priv)));
e51060f0
SH		 1630
1631 cma_attach_to_dev(dev_id_priv, cma_dev);
1632 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
d02d1f53
SH		 1633 atomic_inc(&id_priv->refcount);
1634 dev_id_priv->internal_id = 1;
5b0ec991 1635 dev_id_priv->afonly = id_priv->afonly;
e51060f0
SH		 1636
1637 ret = rdma_listen(id, id_priv->backlog);
1638 if (ret)
d02d1f53 1639 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
468f2239 1640 "listening on device %s\n", ret, cma_dev->device->name);
e51060f0
SH		 1641}
1642
1643static void cma_listen_on_all(struct rdma_id_private *id_priv)
1644{
1645 struct cma_device *cma_dev;
1646
1647 mutex_lock(&lock);
1648 list_add_tail(&id_priv->list, &listen_any_list);
1649 list_for_each_entry(cma_dev, &dev_list, list)
1650 cma_listen_on_dev(id_priv, cma_dev);
1651 mutex_unlock(&lock);
1652}
1653
a81c994d
SH		 1654void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1655{
1656 struct rdma_id_private *id_priv;
1657
1658 id_priv = container_of(id, struct rdma_id_private, id);
1659 id_priv->tos = (u8) tos;
1660}
1661EXPORT_SYMBOL(rdma_set_service_type);
1662
e51060f0
SH		 1663static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1664 void *context)
1665{
1666 struct cma_work *work = context;
1667 struct rdma_route *route;
1668
1669 route = &work->id->id.route;
1670
1671 if (!status) {
1672 route->num_paths = 1;
1673 *route->path_rec = *path_rec;
1674 } else {
550e5ca7
NM		 1675 work->old_state = RDMA_CM_ROUTE_QUERY;
1676 work->new_state = RDMA_CM_ADDR_RESOLVED;
e51060f0 1677 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
8f0472d3 1678 work->event.status = status;
e51060f0
SH		 1679 }
1680
1681 queue_work(cma_wq, &work->work);
1682}
1683
1684static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1685 struct cma_work *work)
1686{
f4753834 1687 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
e51060f0 1688 struct ib_sa_path_rec path_rec;
a81c994d
SH		 1689 ib_sa_comp_mask comp_mask;
1690 struct sockaddr_in6 *sin6;
e51060f0
SH		 1691
1692 memset(&path_rec, 0, sizeof path_rec);
f4753834
SH		 1693 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1694 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1695 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
e51060f0 1696 path_rec.numb_path = 1;
962063e6 1697 path_rec.reversible = 1;
f4753834 1698 path_rec.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
a81c994d
SH		 1699
1700 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1701 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1702 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1703
f4753834 1704 if (cma_family(id_priv) == AF_INET) {
a81c994d
SH		 1705 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1706 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1707 } else {
f4753834 1708 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
a81c994d
SH		 1709 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1710 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1711 }
e51060f0 1712
c1a0b23b 1713 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
a81c994d
SH		 1714 id_priv->id.port_num, &path_rec,
1715 comp_mask, timeout_ms,
1716 GFP_KERNEL, cma_query_handler,
1717 work, &id_priv->query);
e51060f0
SH		 1718
1719 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1720}
1721
c4028958 1722static void cma_work_handler(struct work_struct *_work)
e51060f0 1723{
c4028958 1724 struct cma_work *work = container_of(_work, struct cma_work, work);
e51060f0
SH		 1725 struct rdma_id_private *id_priv = work->id;
1726 int destroy = 0;
1727
de910bd9 1728 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 1729 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1730 goto out;
1731
1732 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
550e5ca7 1733 cma_exch(id_priv, RDMA_CM_DESTROYING);
e51060f0
SH		 1734 destroy = 1;
1735 }
1736out:
de910bd9 1737 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1738 cma_deref_id(id_priv);
1739 if (destroy)
1740 rdma_destroy_id(&id_priv->id);
1741 kfree(work);
1742}
1743
dd5bdff8
OG		 1744static void cma_ndev_work_handler(struct work_struct *_work)
1745{
1746 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1747 struct rdma_id_private *id_priv = work->id;
1748 int destroy = 0;
1749
1750 mutex_lock(&id_priv->handler_mutex);
550e5ca7
NM		 1751 if (id_priv->state == RDMA_CM_DESTROYING ||
1752 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
dd5bdff8
OG		 1753 goto out;
1754
1755 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
550e5ca7 1756 cma_exch(id_priv, RDMA_CM_DESTROYING);
dd5bdff8
OG		 1757 destroy = 1;
1758 }
1759
1760out:
1761 mutex_unlock(&id_priv->handler_mutex);
1762 cma_deref_id(id_priv);
1763 if (destroy)
1764 rdma_destroy_id(&id_priv->id);
1765 kfree(work);
1766}
1767
e51060f0
SH		 1768static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1769{
1770 struct rdma_route *route = &id_priv->id.route;
1771 struct cma_work *work;
1772 int ret;
1773
1774 work = kzalloc(sizeof *work, GFP_KERNEL);
1775 if (!work)
1776 return -ENOMEM;
1777
1778 work->id = id_priv;
c4028958 1779 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 1780 work->old_state = RDMA_CM_ROUTE_QUERY;
1781 work->new_state = RDMA_CM_ROUTE_RESOLVED;
e51060f0
SH		 1782 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1783
1784 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1785 if (!route->path_rec) {
1786 ret = -ENOMEM;
1787 goto err1;
1788 }
1789
1790 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1791 if (ret)
1792 goto err2;
1793
1794 return 0;
1795err2:
1796 kfree(route->path_rec);
1797 route->path_rec = NULL;
1798err1:
1799 kfree(work);
1800 return ret;
1801}
1802
1803int rdma_set_ib_paths(struct rdma_cm_id *id,
1804 struct ib_sa_path_rec *path_rec, int num_paths)
1805{
1806 struct rdma_id_private *id_priv;
1807 int ret;
1808
1809 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7
NM		 1810 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1811 RDMA_CM_ROUTE_RESOLVED))
e51060f0
SH		 1812 return -EINVAL;
1813
9893e742
JL		 1814 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1815 GFP_KERNEL);
e51060f0
SH		 1816 if (!id->route.path_rec) {
1817 ret = -ENOMEM;
1818 goto err;
1819 }
1820
ae2d9293 1821 id->route.num_paths = num_paths;
e51060f0
SH		 1822 return 0;
1823err:
550e5ca7 1824 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
e51060f0
SH		 1825 return ret;
1826}
1827EXPORT_SYMBOL(rdma_set_ib_paths);
1828
07ebafba
TT		 1829static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1830{
1831 struct cma_work *work;
1832
1833 work = kzalloc(sizeof *work, GFP_KERNEL);
1834 if (!work)
1835 return -ENOMEM;
1836
1837 work->id = id_priv;
c4028958 1838 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 1839 work->old_state = RDMA_CM_ROUTE_QUERY;
1840 work->new_state = RDMA_CM_ROUTE_RESOLVED;
07ebafba
TT		 1841 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1842 queue_work(cma_wq, &work->work);
1843 return 0;
1844}
1845
3c86aa70
EC		 1846static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1847{
1848 struct rdma_route *route = &id_priv->id.route;
1849 struct rdma_addr *addr = &route->addr;
1850 struct cma_work *work;
1851 int ret;
1852 struct sockaddr_in *src_addr = (struct sockaddr_in *)&route->addr.src_addr;
1853 struct sockaddr_in *dst_addr = (struct sockaddr_in *)&route->addr.dst_addr;
1854 struct net_device *ndev = NULL;
af7bd463 1855 u16 vid;
3c86aa70
EC		 1856
1857 if (src_addr->sin_family != dst_addr->sin_family)
1858 return -EINVAL;
1859
1860 work = kzalloc(sizeof *work, GFP_KERNEL);
1861 if (!work)
1862 return -ENOMEM;
1863
1864 work->id = id_priv;
1865 INIT_WORK(&work->work, cma_work_handler);
1866
1867 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1868 if (!route->path_rec) {
1869 ret = -ENOMEM;
1870 goto err1;
1871 }
1872
1873 route->num_paths = 1;
1874
3c86aa70
EC		 1875 if (addr->dev_addr.bound_dev_if)
1876 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1877 if (!ndev) {
1878 ret = -ENODEV;
1879 goto err2;
1880 }
1881
af7bd463
EC		 1882 vid = rdma_vlan_dev_vlan_id(ndev);
1883
1884 iboe_mac_vlan_to_ll(&route->path_rec->sgid, addr->dev_addr.src_dev_addr, vid);
1885 iboe_mac_vlan_to_ll(&route->path_rec->dgid, addr->dev_addr.dst_dev_addr, vid);
1886
1887 route->path_rec->hop_limit = 1;
1888 route->path_rec->reversible = 1;
1889 route->path_rec->pkey = cpu_to_be16(0xffff);
1890 route->path_rec->mtu_selector = IB_SA_EQ;
366cddb4
AV		 1891 route->path_rec->sl = netdev_get_prio_tc_map(
1892 ndev->priv_flags & IFF_802_1Q_VLAN ?
1893 vlan_dev_real_dev(ndev) : ndev,
1894 rt_tos2priority(id_priv->tos));
af7bd463 1895
3c86aa70
EC		 1896 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1897 route->path_rec->rate_selector = IB_SA_EQ;
1898 route->path_rec->rate = iboe_get_rate(ndev);
1899 dev_put(ndev);
1900 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1901 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1902 if (!route->path_rec->mtu) {
1903 ret = -EINVAL;
1904 goto err2;
1905 }
1906
550e5ca7
NM		 1907 work->old_state = RDMA_CM_ROUTE_QUERY;
1908 work->new_state = RDMA_CM_ROUTE_RESOLVED;
3c86aa70
EC		 1909 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1910 work->event.status = 0;
1911
1912 queue_work(cma_wq, &work->work);
1913
1914 return 0;
1915
1916err2:
1917 kfree(route->path_rec);
1918 route->path_rec = NULL;
1919err1:
1920 kfree(work);
1921 return ret;
1922}
1923
e51060f0
SH		 1924int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1925{
1926 struct rdma_id_private *id_priv;
1927 int ret;
1928
1929 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 1930 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
e51060f0
SH		 1931 return -EINVAL;
1932
1933 atomic_inc(&id_priv->refcount);
07ebafba
TT		 1934 switch (rdma_node_get_transport(id->device->node_type)) {
1935 case RDMA_TRANSPORT_IB:
3c86aa70
EC		 1936 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1937 case IB_LINK_LAYER_INFINIBAND:
1938 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1939 break;
1940 case IB_LINK_LAYER_ETHERNET:
1941 ret = cma_resolve_iboe_route(id_priv);
1942 break;
1943 default:
1944 ret = -ENOSYS;
1945 }
e51060f0 1946 break;
07ebafba
TT		 1947 case RDMA_TRANSPORT_IWARP:
1948 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1949 break;
e51060f0
SH		 1950 default:
1951 ret = -ENOSYS;
1952 break;
1953 }
1954 if (ret)
1955 goto err;
1956
1957 return 0;
1958err:
550e5ca7 1959 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
e51060f0
SH		 1960 cma_deref_id(id_priv);
1961 return ret;
1962}
1963EXPORT_SYMBOL(rdma_resolve_route);
1964
6a3e362d
SH		 1965static void cma_set_loopback(struct sockaddr *addr)
1966{
1967 switch (addr->sa_family) {
1968 case AF_INET:
1969 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1970 break;
1971 case AF_INET6:
1972 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
1973 0, 0, 0, htonl(1));
1974 break;
1975 default:
1976 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
1977 0, 0, 0, htonl(1));
1978 break;
1979 }
1980}
1981
e51060f0
SH		 1982static int cma_bind_loopback(struct rdma_id_private *id_priv)
1983{
1984 struct cma_device *cma_dev;
1985 struct ib_port_attr port_attr;
f0ee3404 1986 union ib_gid gid;
e51060f0
SH		 1987 u16 pkey;
1988 int ret;
1989 u8 p;
1990
1991 mutex_lock(&lock);
e82153b5
KK		 1992 if (list_empty(&dev_list)) {
1993 ret = -ENODEV;
1994 goto out;
1995 }
e51060f0
SH		 1996 list_for_each_entry(cma_dev, &dev_list, list)
1997 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
e82153b5 1998 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
e51060f0
SH		 1999 port_attr.state == IB_PORT_ACTIVE)
2000 goto port_found;
2001
e82153b5
KK		 2002 p = 1;
2003 cma_dev = list_entry(dev_list.next, struct cma_device, list);
e51060f0
SH		 2004
2005port_found:
f0ee3404 2006 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
e51060f0
SH		 2007 if (ret)
2008 goto out;
2009
2010 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2011 if (ret)
2012 goto out;
2013
6f8372b6 2014 id_priv->id.route.addr.dev_addr.dev_type =
3c86aa70 2015 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
6f8372b6
SH		 2016 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2017
2018 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0
SH		 2019 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2020 id_priv->id.port_num = p;
2021 cma_attach_to_dev(id_priv, cma_dev);
f4753834 2022 cma_set_loopback(cma_src_addr(id_priv));
e51060f0
SH		 2023out:
2024 mutex_unlock(&lock);
2025 return ret;
2026}
2027
2028static void addr_handler(int status, struct sockaddr *src_addr,
2029 struct rdma_dev_addr *dev_addr, void *context)
2030{
2031 struct rdma_id_private *id_priv = context;
a1b1b61f 2032 struct rdma_cm_event event;
e51060f0 2033
a1b1b61f 2034 memset(&event, 0, sizeof event);
de910bd9 2035 mutex_lock(&id_priv->handler_mutex);
550e5ca7
NM		 2036 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2037 RDMA_CM_ADDR_RESOLVED))
61a73c70 2038 goto out;
61a73c70
SH		 2039
2040 if (!status && !id_priv->cma_dev)
e51060f0
SH		 2041 status = cma_acquire_dev(id_priv);
2042
2043 if (status) {
550e5ca7
NM		 2044 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2045 RDMA_CM_ADDR_BOUND))
e51060f0 2046 goto out;
a1b1b61f
SH		 2047 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2048 event.status = status;
e51060f0 2049 } else {
f4753834 2050 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
a1b1b61f 2051 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
e51060f0
SH		 2052 }
2053
a1b1b61f 2054 if (id_priv->id.event_handler(&id_priv->id, &event)) {
550e5ca7 2055 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 2056 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 2057 cma_deref_id(id_priv);
2058 rdma_destroy_id(&id_priv->id);
2059 return;
2060 }
2061out:
de910bd9 2062 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 2063 cma_deref_id(id_priv);
2064}
2065
2066static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2067{
2068 struct cma_work *work;
f0ee3404 2069 union ib_gid gid;
e51060f0
SH		 2070 int ret;
2071
2072 work = kzalloc(sizeof *work, GFP_KERNEL);
2073 if (!work)
2074 return -ENOMEM;
2075
2076 if (!id_priv->cma_dev) {
2077 ret = cma_bind_loopback(id_priv);
2078 if (ret)
2079 goto err;
2080 }
2081
6f8372b6
SH		 2082 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2083 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0 2084
e51060f0 2085 work->id = id_priv;
c4028958 2086 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 2087 work->old_state = RDMA_CM_ADDR_QUERY;
2088 work->new_state = RDMA_CM_ADDR_RESOLVED;
e51060f0
SH		 2089 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2090 queue_work(cma_wq, &work->work);
2091 return 0;
2092err:
2093 kfree(work);
2094 return ret;
2095}
2096
2097static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2098 struct sockaddr *dst_addr)
2099{
d14714df
SH		 2100 if (!src_addr || !src_addr->sa_family) {
2101 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2102 if ((src_addr->sa_family = dst_addr->sa_family) == AF_INET6) {
2103 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2104 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2105 }
2106 }
2107 return rdma_bind_addr(id, src_addr);
e51060f0
SH		 2108}
2109
2110int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2111 struct sockaddr *dst_addr, int timeout_ms)
2112{
2113 struct rdma_id_private *id_priv;
2114 int ret;
2115
2116 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2117 if (id_priv->state == RDMA_CM_IDLE) {
e51060f0
SH		 2118 ret = cma_bind_addr(id, src_addr, dst_addr);
2119 if (ret)
2120 return ret;
2121 }
2122
550e5ca7 2123 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
e51060f0
SH		 2124 return -EINVAL;
2125
2126 atomic_inc(&id_priv->refcount);
f4753834 2127 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
e51060f0
SH		 2128 if (cma_any_addr(dst_addr))
2129 ret = cma_resolve_loopback(id_priv);
2130 else
f4753834 2131 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
7a118df3 2132 dst_addr, &id->route.addr.dev_addr,
e51060f0
SH		 2133 timeout_ms, addr_handler, id_priv);
2134 if (ret)
2135 goto err;
2136
2137 return 0;
2138err:
550e5ca7 2139 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
e51060f0
SH		 2140 cma_deref_id(id_priv);
2141 return ret;
2142}
2143EXPORT_SYMBOL(rdma_resolve_addr);
2144
a9bb7912
HS		 2145int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2146{
2147 struct rdma_id_private *id_priv;
2148 unsigned long flags;
2149 int ret;
2150
2151 id_priv = container_of(id, struct rdma_id_private, id);
2152 spin_lock_irqsave(&id_priv->lock, flags);
c8dea2f9 2153 if (reuse || id_priv->state == RDMA_CM_IDLE) {
a9bb7912
HS		 2154 id_priv->reuseaddr = reuse;
2155 ret = 0;
2156 } else {
2157 ret = -EINVAL;
2158 }
2159 spin_unlock_irqrestore(&id_priv->lock, flags);
2160 return ret;
2161}
2162EXPORT_SYMBOL(rdma_set_reuseaddr);
2163
68602120
SH		 2164int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2165{
2166 struct rdma_id_private *id_priv;
2167 unsigned long flags;
2168 int ret;
2169
2170 id_priv = container_of(id, struct rdma_id_private, id);
2171 spin_lock_irqsave(&id_priv->lock, flags);
2172 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2173 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2174 id_priv->afonly = afonly;
2175 ret = 0;
2176 } else {
2177 ret = -EINVAL;
2178 }
2179 spin_unlock_irqrestore(&id_priv->lock, flags);
2180 return ret;
2181}
2182EXPORT_SYMBOL(rdma_set_afonly);
2183
e51060f0
SH		 2184static void cma_bind_port(struct rdma_bind_list *bind_list,
2185 struct rdma_id_private *id_priv)
2186{
58afdcb7
SH		 2187 struct sockaddr *addr;
2188 struct sockaddr_ib *sib;
2189 u64 sid, mask;
2190 __be16 port;
e51060f0 2191
f4753834 2192 addr = cma_src_addr(id_priv);
58afdcb7
SH		 2193 port = htons(bind_list->port);
2194
2195 switch (addr->sa_family) {
2196 case AF_INET:
2197 ((struct sockaddr_in *) addr)->sin_port = port;
2198 break;
2199 case AF_INET6:
2200 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2201 break;
2202 case AF_IB:
2203 sib = (struct sockaddr_ib *) addr;
2204 sid = be64_to_cpu(sib->sib_sid);
2205 mask = be64_to_cpu(sib->sib_sid_mask);
2206 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2207 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2208 break;
2209 }
e51060f0
SH		 2210 id_priv->bind_list = bind_list;
2211 hlist_add_head(&id_priv->node, &bind_list->owners);
2212}
2213
2214static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2215 unsigned short snum)
2216{
2217 struct rdma_bind_list *bind_list;
3b069c5d 2218 int ret;
e51060f0 2219
cb164b8c 2220 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
e51060f0
SH		 2221 if (!bind_list)
2222 return -ENOMEM;
2223
3b069c5d
TH		 2224 ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2225 if (ret < 0)
2226 goto err;
aedec080
SH		 2227
2228 bind_list->ps = ps;
3b069c5d 2229 bind_list->port = (unsigned short)ret;
aedec080
SH		 2230 cma_bind_port(bind_list, id_priv);
2231 return 0;
3b069c5d 2232err:
aedec080 2233 kfree(bind_list);
3b069c5d 2234 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
aedec080 2235}
e51060f0 2236
aedec080
SH		 2237static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2238{
5d7220e8
TH		 2239 static unsigned int last_used_port;
2240 int low, high, remaining;
2241 unsigned int rover;
e51060f0 2242
227b60f5 2243 inet_get_local_port_range(&low, &high);
5d7220e8
TH		 2244 remaining = (high - low) + 1;
2245 rover = net_random() % remaining + low;
2246retry:
2247 if (last_used_port != rover &&
2248 !idr_find(ps, (unsigned short) rover)) {
2249 int ret = cma_alloc_port(ps, id_priv, rover);
2250 /*
2251 * Remember previously used port number in order to avoid
2252 * re-using same port immediately after it is closed.
2253 */
2254 if (!ret)
2255 last_used_port = rover;
2256 if (ret != -EADDRNOTAVAIL)
2257 return ret;
e51060f0 2258 }
5d7220e8
TH		 2259 if (--remaining) {
2260 rover++;
2261 if ((rover < low) || (rover > high))
2262 rover = low;
2263 goto retry;
2264 }
2265 return -EADDRNOTAVAIL;
e51060f0
SH		 2266}
2267
a9bb7912
HS		 2268/*
2269 * Check that the requested port is available. This is called when trying to
2270 * bind to a specific port, or when trying to listen on a bound port. In
2271 * the latter case, the provided id_priv may already be on the bind_list, but
2272 * we still need to check that it's okay to start listening.
2273 */
2274static int cma_check_port(struct rdma_bind_list *bind_list,
2275 struct rdma_id_private *id_priv, uint8_t reuseaddr)
e51060f0
SH		 2276{
2277 struct rdma_id_private *cur_id;
43b752da 2278 struct sockaddr *addr, *cur_addr;
e51060f0 2279
f4753834 2280 addr = cma_src_addr(id_priv);
b67bfe0d 2281 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
a9bb7912
HS		 2282 if (id_priv == cur_id)
2283 continue;
3cd96564 2284
5b0ec991
SH		 2285 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2286 cur_id->reuseaddr)
2287 continue;
e51060f0 2288
f4753834 2289 cur_addr = cma_src_addr(cur_id);
5b0ec991
SH		 2290 if (id_priv->afonly && cur_id->afonly &&
2291 (addr->sa_family != cur_addr->sa_family))
2292 continue;
2293
2294 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2295 return -EADDRNOTAVAIL;
2296
2297 if (!cma_addr_cmp(addr, cur_addr))
2298 return -EADDRINUSE;
a9bb7912 2299 }
e51060f0
SH		 2300 return 0;
2301}
2302
a9bb7912
HS		 2303static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2304{
2305 struct rdma_bind_list *bind_list;
2306 unsigned short snum;
2307 int ret;
2308
f4753834 2309 snum = ntohs(cma_port(cma_src_addr(id_priv)));
a9bb7912
HS		 2310 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2311 return -EACCES;
2312
2313 bind_list = idr_find(ps, snum);
2314 if (!bind_list) {
2315 ret = cma_alloc_port(ps, id_priv, snum);
2316 } else {
2317 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2318 if (!ret)
2319 cma_bind_port(bind_list, id_priv);
2320 }
2321 return ret;
2322}
2323
2324static int cma_bind_listen(struct rdma_id_private *id_priv)
2325{
2326 struct rdma_bind_list *bind_list = id_priv->bind_list;
2327 int ret = 0;
2328
2329 mutex_lock(&lock);
2330 if (bind_list->owners.first->next)
2331 ret = cma_check_port(bind_list, id_priv, 0);
2332 mutex_unlock(&lock);
2333 return ret;
2334}
2335
58afdcb7 2336static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
e51060f0 2337{
e51060f0
SH		 2338 switch (id_priv->id.ps) {
2339 case RDMA_PS_SDP:
58afdcb7 2340 return &sdp_ps;
e51060f0 2341 case RDMA_PS_TCP:
58afdcb7 2342 return &tcp_ps;
628e5f6d 2343 case RDMA_PS_UDP:
58afdcb7 2344 return &udp_ps;
c8f6a362 2345 case RDMA_PS_IPOIB:
58afdcb7 2346 return &ipoib_ps;
2d2e9415 2347 case RDMA_PS_IB:
58afdcb7 2348 return &ib_ps;
e51060f0 2349 default:
58afdcb7
SH		 2350 return NULL;
2351 }
2352}
2353
2354static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2355{
2356 struct idr *ps = NULL;
2357 struct sockaddr_ib *sib;
2358 u64 sid_ps, mask, sid;
2359
f4753834 2360 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
58afdcb7
SH		 2361 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2362 sid = be64_to_cpu(sib->sib_sid) & mask;
2363
2364 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2365 sid_ps = RDMA_IB_IP_PS_IB;
2366 ps = &ib_ps;
2367 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2368 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2369 sid_ps = RDMA_IB_IP_PS_TCP;
2370 ps = &tcp_ps;
2371 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2372 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2373 sid_ps = RDMA_IB_IP_PS_UDP;
2374 ps = &udp_ps;
e51060f0
SH		 2375 }
2376
58afdcb7
SH		 2377 if (ps) {
2378 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2379 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2380 be64_to_cpu(sib->sib_sid_mask));
2381 }
2382 return ps;
2383}
2384
2385static int cma_get_port(struct rdma_id_private *id_priv)
2386{
2387 struct idr *ps;
2388 int ret;
2389
f4753834 2390 if (cma_family(id_priv) != AF_IB)
58afdcb7
SH		 2391 ps = cma_select_inet_ps(id_priv);
2392 else
2393 ps = cma_select_ib_ps(id_priv);
2394 if (!ps)
2395 return -EPROTONOSUPPORT;
2396
e51060f0 2397 mutex_lock(&lock);
f4753834 2398 if (cma_any_port(cma_src_addr(id_priv)))
aedec080 2399 ret = cma_alloc_any_port(ps, id_priv);
e51060f0
SH		 2400 else
2401 ret = cma_use_port(ps, id_priv);
2402 mutex_unlock(&lock);
2403
2404 return ret;
2405}
2406
d14714df
SH		 2407static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2408 struct sockaddr *addr)
2409{
d90f9b35 2410#if IS_ENABLED(CONFIG_IPV6)
d14714df
SH		 2411 struct sockaddr_in6 *sin6;
2412
2413 if (addr->sa_family != AF_INET6)
2414 return 0;
2415
2416 sin6 = (struct sockaddr_in6 *) addr;
2417 if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) &&
2418 !sin6->sin6_scope_id)
2419 return -EINVAL;
2420
2421 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2422#endif
2423 return 0;
2424}
2425
a9bb7912
HS		 2426int rdma_listen(struct rdma_cm_id *id, int backlog)
2427{
2428 struct rdma_id_private *id_priv;
2429 int ret;
2430
2431 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2432 if (id_priv->state == RDMA_CM_IDLE) {
f4753834
SH		 2433 id->route.addr.src_addr.ss_family = AF_INET;
2434 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
a9bb7912
HS		 2435 if (ret)
2436 return ret;
2437 }
2438
550e5ca7 2439 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
a9bb7912
HS		 2440 return -EINVAL;
2441
2442 if (id_priv->reuseaddr) {
2443 ret = cma_bind_listen(id_priv);
2444 if (ret)
2445 goto err;
2446 }
2447
2448 id_priv->backlog = backlog;
2449 if (id->device) {
2450 switch (rdma_node_get_transport(id->device->node_type)) {
2451 case RDMA_TRANSPORT_IB:
2452 ret = cma_ib_listen(id_priv);
2453 if (ret)
2454 goto err;
2455 break;
2456 case RDMA_TRANSPORT_IWARP:
2457 ret = cma_iw_listen(id_priv, backlog);
2458 if (ret)
2459 goto err;
2460 break;
2461 default:
2462 ret = -ENOSYS;
2463 goto err;
2464 }
2465 } else
2466 cma_listen_on_all(id_priv);
2467
2468 return 0;
2469err:
2470 id_priv->backlog = 0;
550e5ca7 2471 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
a9bb7912
HS		 2472 return ret;
2473}
2474EXPORT_SYMBOL(rdma_listen);
2475
e51060f0
SH		 2476int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2477{
2478 struct rdma_id_private *id_priv;
2479 int ret;
2480
680f920a
SH		 2481 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2482 addr->sa_family != AF_IB)
e51060f0
SH		 2483 return -EAFNOSUPPORT;
2484
2485 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2486 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
e51060f0
SH		 2487 return -EINVAL;
2488
d14714df
SH		 2489 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2490 if (ret)
2491 goto err1;
2492
8523c048 2493 if (!cma_any_addr(addr)) {
680f920a 2494 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
e51060f0 2495 if (ret)
255d0c14
KK		 2496 goto err1;
2497
255d0c14 2498 ret = cma_acquire_dev(id_priv);
255d0c14
KK		 2499 if (ret)
2500 goto err1;
e51060f0
SH		 2501 }
2502
f4753834 2503 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
68602120
SH		 2504 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2505 if (addr->sa_family == AF_INET)
2506 id_priv->afonly = 1;
5b0ec991 2507#if IS_ENABLED(CONFIG_IPV6)
68602120
SH		 2508 else if (addr->sa_family == AF_INET6)
2509 id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
5b0ec991 2510#endif
68602120 2511 }
e51060f0
SH		 2512 ret = cma_get_port(id_priv);
2513 if (ret)
255d0c14 2514 goto err2;
e51060f0
SH		 2515
2516 return 0;
255d0c14 2517err2:
a396d43a
SH		 2518 if (id_priv->cma_dev)
2519 cma_release_dev(id_priv);
255d0c14 2520err1:
550e5ca7 2521 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
e51060f0
SH		 2522 return ret;
2523}
2524EXPORT_SYMBOL(rdma_bind_addr);
2525
f4753834 2526static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
e51060f0 2527{
e51060f0
SH		 2528 struct cma_hdr *cma_hdr;
2529 struct sdp_hh *sdp_hdr;
2530
f4753834 2531 if (cma_family(id_priv) == AF_INET) {
1f5175ad
AS		 2532 struct sockaddr_in *src4, *dst4;
2533
f4753834
SH		 2534 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2535 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
1f5175ad 2536
f4753834 2537 switch (id_priv->id.ps) {
1f5175ad
AS		 2538 case RDMA_PS_SDP:
2539 sdp_hdr = hdr;
2540 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2541 return -EINVAL;
2542 sdp_set_ip_ver(sdp_hdr, 4);
2543 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2544 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2545 sdp_hdr->port = src4->sin_port;
2546 break;
2547 default:
2548 cma_hdr = hdr;
2549 cma_hdr->cma_version = CMA_VERSION;
2550 cma_set_ip_ver(cma_hdr, 4);
2551 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2552 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2553 cma_hdr->port = src4->sin_port;
2554 break;
2555 }
2556 } else {
2557 struct sockaddr_in6 *src6, *dst6;
2558
f4753834
SH		 2559 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2560 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
1f5175ad 2561
f4753834 2562 switch (id_priv->id.ps) {
1f5175ad
AS		 2563 case RDMA_PS_SDP:
2564 sdp_hdr = hdr;
2565 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2566 return -EINVAL;
2567 sdp_set_ip_ver(sdp_hdr, 6);
2568 sdp_hdr->src_addr.ip6 = src6->sin6_addr;
2569 sdp_hdr->dst_addr.ip6 = dst6->sin6_addr;
2570 sdp_hdr->port = src6->sin6_port;
2571 break;
2572 default:
2573 cma_hdr = hdr;
2574 cma_hdr->cma_version = CMA_VERSION;
2575 cma_set_ip_ver(cma_hdr, 6);
2576 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2577 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2578 cma_hdr->port = src6->sin6_port;
2579 break;
2580 }
e51060f0
SH		 2581 }
2582 return 0;
2583}
2584
628e5f6d
SH		 2585static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2586 struct ib_cm_event *ib_event)
2587{
2588 struct rdma_id_private *id_priv = cm_id->context;
2589 struct rdma_cm_event event;
2590 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2591 int ret = 0;
2592
550e5ca7 2593 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
8aa08602 2594 return 0;
628e5f6d 2595
8aa08602 2596 memset(&event, 0, sizeof event);
628e5f6d
SH		 2597 switch (ib_event->event) {
2598 case IB_CM_SIDR_REQ_ERROR:
2599 event.event = RDMA_CM_EVENT_UNREACHABLE;
2600 event.status = -ETIMEDOUT;
2601 break;
2602 case IB_CM_SIDR_REP_RECEIVED:
2603 event.param.ud.private_data = ib_event->private_data;
2604 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2605 if (rep->status != IB_SIDR_SUCCESS) {
2606 event.event = RDMA_CM_EVENT_UNREACHABLE;
2607 event.status = ib_event->param.sidr_rep_rcvd.status;
2608 break;
2609 }
d2ca39f2
YE		 2610 ret = cma_set_qkey(id_priv);
2611 if (ret) {
2612 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2613 event.status = -EINVAL;
2614 break;
2615 }
c8f6a362 2616 if (id_priv->qkey != rep->qkey) {
628e5f6d
SH		 2617 event.event = RDMA_CM_EVENT_UNREACHABLE;
2618 event.status = -EINVAL;
2619 break;
2620 }
2621 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2622 id_priv->id.route.path_rec,
2623 &event.param.ud.ah_attr);
2624 event.param.ud.qp_num = rep->qpn;
2625 event.param.ud.qkey = rep->qkey;
2626 event.event = RDMA_CM_EVENT_ESTABLISHED;
2627 event.status = 0;
2628 break;
2629 default:
468f2239 2630 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
628e5f6d
SH		 2631 ib_event->event);
2632 goto out;
2633 }
2634
2635 ret = id_priv->id.event_handler(&id_priv->id, &event);
2636 if (ret) {
2637 /* Destroy the CM ID by returning a non-zero value. */
2638 id_priv->cm_id.ib = NULL;
550e5ca7 2639 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 2640 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 2641 rdma_destroy_id(&id_priv->id);
2642 return ret;
2643 }
2644out:
de910bd9 2645 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 2646 return ret;
2647}
2648
2649static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2650 struct rdma_conn_param *conn_param)
2651{
2652 struct ib_cm_sidr_req_param req;
0c9361fc 2653 struct ib_cm_id *id;
628e5f6d
SH		 2654 int ret;
2655
2656 req.private_data_len = sizeof(struct cma_hdr) +
2657 conn_param->private_data_len;
04ded167
SH		 2658 if (req.private_data_len < conn_param->private_data_len)
2659 return -EINVAL;
2660
628e5f6d
SH		 2661 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2662 if (!req.private_data)
2663 return -ENOMEM;
2664
2665 if (conn_param->private_data && conn_param->private_data_len)
2666 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2667 conn_param->private_data, conn_param->private_data_len);
2668
f4753834 2669 ret = cma_format_hdr((void *) req.private_data, id_priv);
628e5f6d
SH		 2670 if (ret)
2671 goto out;
2672
0c9361fc
JM		 2673 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2674 id_priv);
2675 if (IS_ERR(id)) {
2676 ret = PTR_ERR(id);
628e5f6d
SH		 2677 goto out;
2678 }
0c9361fc 2679 id_priv->cm_id.ib = id;
628e5f6d 2680
f4753834
SH		 2681 req.path = id_priv->id.route.path_rec;
2682 req.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
628e5f6d
SH		 2683 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2684 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2685
2686 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2687 if (ret) {
2688 ib_destroy_cm_id(id_priv->cm_id.ib);
2689 id_priv->cm_id.ib = NULL;
2690 }
2691out:
2692 kfree(req.private_data);
2693 return ret;
2694}
2695
e51060f0
SH		 2696static int cma_connect_ib(struct rdma_id_private *id_priv,
2697 struct rdma_conn_param *conn_param)
2698{
2699 struct ib_cm_req_param req;
2700 struct rdma_route *route;
2701 void *private_data;
0c9361fc 2702 struct ib_cm_id *id;
e51060f0
SH		 2703 int offset, ret;
2704
2705 memset(&req, 0, sizeof req);
2706 offset = cma_user_data_offset(id_priv->id.ps);
2707 req.private_data_len = offset + conn_param->private_data_len;
04ded167
SH		 2708 if (req.private_data_len < conn_param->private_data_len)
2709 return -EINVAL;
2710
e51060f0
SH		 2711 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2712 if (!private_data)
2713 return -ENOMEM;
2714
2715 if (conn_param->private_data && conn_param->private_data_len)
2716 memcpy(private_data + offset, conn_param->private_data,
2717 conn_param->private_data_len);
2718
0c9361fc
JM		 2719 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2720 if (IS_ERR(id)) {
2721 ret = PTR_ERR(id);
e51060f0
SH		 2722 goto out;
2723 }
0c9361fc 2724 id_priv->cm_id.ib = id;
e51060f0
SH		 2725
2726 route = &id_priv->id.route;
f4753834 2727 ret = cma_format_hdr(private_data, id_priv);
e51060f0
SH		 2728 if (ret)
2729 goto out;
2730 req.private_data = private_data;
2731
2732 req.primary_path = &route->path_rec[0];
2733 if (route->num_paths == 2)
2734 req.alternate_path = &route->path_rec[1];
2735
f4753834 2736 req.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
e51060f0 2737 req.qp_num = id_priv->qp_num;
18c441a6 2738 req.qp_type = id_priv->id.qp_type;
e51060f0
SH		 2739 req.starting_psn = id_priv->seq_num;
2740 req.responder_resources = conn_param->responder_resources;
2741 req.initiator_depth = conn_param->initiator_depth;
2742 req.flow_control = conn_param->flow_control;
4ede178a
SH		 2743 req.retry_count = min_t(u8, 7, conn_param->retry_count);
2744 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
e51060f0
SH		 2745 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2746 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2747 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2748 req.srq = id_priv->srq ? 1 : 0;
2749
2750 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2751out:
0c9361fc
JM		 2752 if (ret && !IS_ERR(id)) {
2753 ib_destroy_cm_id(id);
675a027c
KK		 2754 id_priv->cm_id.ib = NULL;
2755 }
2756
e51060f0
SH		 2757 kfree(private_data);
2758 return ret;
2759}
2760
07ebafba
TT		 2761static int cma_connect_iw(struct rdma_id_private *id_priv,
2762 struct rdma_conn_param *conn_param)
2763{
2764 struct iw_cm_id *cm_id;
2765 struct sockaddr_in* sin;
2766 int ret;
2767 struct iw_cm_conn_param iw_param;
2768
2769 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
0c9361fc
JM		 2770 if (IS_ERR(cm_id))
2771 return PTR_ERR(cm_id);
07ebafba
TT		 2772
2773 id_priv->cm_id.iw = cm_id;
2774
f4753834 2775 sin = (struct sockaddr_in *) cma_src_addr(id_priv);
07ebafba
TT		 2776 cm_id->local_addr = *sin;
2777
f4753834 2778 sin = (struct sockaddr_in *) cma_dst_addr(id_priv);
07ebafba
TT		 2779 cm_id->remote_addr = *sin;
2780
5851bb89 2781 ret = cma_modify_qp_rtr(id_priv, conn_param);
675a027c
KK		 2782 if (ret)
2783 goto out;
07ebafba 2784
f45ee80e
HS		 2785 if (conn_param) {
2786 iw_param.ord = conn_param->initiator_depth;
2787 iw_param.ird = conn_param->responder_resources;
2788 iw_param.private_data = conn_param->private_data;
2789 iw_param.private_data_len = conn_param->private_data_len;
2790 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2791 } else {
2792 memset(&iw_param, 0, sizeof iw_param);
07ebafba 2793 iw_param.qpn = id_priv->qp_num;
f45ee80e 2794 }
07ebafba
TT		 2795 ret = iw_cm_connect(cm_id, &iw_param);
2796out:
0c9361fc 2797 if (ret) {
675a027c
KK		 2798 iw_destroy_cm_id(cm_id);
2799 id_priv->cm_id.iw = NULL;
2800 }
07ebafba
TT		 2801 return ret;
2802}
2803
e51060f0
SH		 2804int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2805{
2806 struct rdma_id_private *id_priv;
2807 int ret;
2808
2809 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2810 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
e51060f0
SH		 2811 return -EINVAL;
2812
2813 if (!id->qp) {
2814 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 2815 id_priv->srq = conn_param->srq;
2816 }
2817
07ebafba
TT		 2818 switch (rdma_node_get_transport(id->device->node_type)) {
2819 case RDMA_TRANSPORT_IB:
b26f9b99 2820 if (id->qp_type == IB_QPT_UD)
628e5f6d
SH		 2821 ret = cma_resolve_ib_udp(id_priv, conn_param);
2822 else
2823 ret = cma_connect_ib(id_priv, conn_param);
e51060f0 2824 break;
07ebafba
TT		 2825 case RDMA_TRANSPORT_IWARP:
2826 ret = cma_connect_iw(id_priv, conn_param);
2827 break;
e51060f0
SH		 2828 default:
2829 ret = -ENOSYS;
2830 break;
2831 }
2832 if (ret)
2833 goto err;
2834
2835 return 0;
2836err:
550e5ca7 2837 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
e51060f0
SH		 2838 return ret;
2839}
2840EXPORT_SYMBOL(rdma_connect);
2841
2842static int cma_accept_ib(struct rdma_id_private *id_priv,
2843 struct rdma_conn_param *conn_param)
2844{
2845 struct ib_cm_rep_param rep;
5851bb89 2846 int ret;
0fe313b0 2847
5851bb89
SH		 2848 ret = cma_modify_qp_rtr(id_priv, conn_param);
2849 if (ret)
2850 goto out;
0fe313b0 2851
5851bb89
SH		 2852 ret = cma_modify_qp_rts(id_priv, conn_param);
2853 if (ret)
2854 goto out;
e51060f0
SH		 2855
2856 memset(&rep, 0, sizeof rep);
2857 rep.qp_num = id_priv->qp_num;
2858 rep.starting_psn = id_priv->seq_num;
2859 rep.private_data = conn_param->private_data;
2860 rep.private_data_len = conn_param->private_data_len;
2861 rep.responder_resources = conn_param->responder_resources;
2862 rep.initiator_depth = conn_param->initiator_depth;
e51060f0
SH		 2863 rep.failover_accepted = 0;
2864 rep.flow_control = conn_param->flow_control;
4ede178a 2865 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
e51060f0
SH		 2866 rep.srq = id_priv->srq ? 1 : 0;
2867
0fe313b0
SH		 2868 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2869out:
2870 return ret;
e51060f0
SH		 2871}
2872
07ebafba
TT		 2873static int cma_accept_iw(struct rdma_id_private *id_priv,
2874 struct rdma_conn_param *conn_param)
2875{
2876 struct iw_cm_conn_param iw_param;
2877 int ret;
2878
5851bb89 2879 ret = cma_modify_qp_rtr(id_priv, conn_param);
07ebafba
TT		 2880 if (ret)
2881 return ret;
2882
2883 iw_param.ord = conn_param->initiator_depth;
2884 iw_param.ird = conn_param->responder_resources;
2885 iw_param.private_data = conn_param->private_data;
2886 iw_param.private_data_len = conn_param->private_data_len;
2887 if (id_priv->id.qp) {
2888 iw_param.qpn = id_priv->qp_num;
2889 } else
2890 iw_param.qpn = conn_param->qp_num;
2891
2892 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2893}
2894
628e5f6d
SH		 2895static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2896 enum ib_cm_sidr_status status,
2897 const void *private_data, int private_data_len)
2898{
2899 struct ib_cm_sidr_rep_param rep;
d2ca39f2 2900 int ret;
628e5f6d
SH		 2901
2902 memset(&rep, 0, sizeof rep);
2903 rep.status = status;
2904 if (status == IB_SIDR_SUCCESS) {
d2ca39f2
YE		 2905 ret = cma_set_qkey(id_priv);
2906 if (ret)
2907 return ret;
628e5f6d 2908 rep.qp_num = id_priv->qp_num;
c8f6a362 2909 rep.qkey = id_priv->qkey;
628e5f6d
SH		 2910 }
2911 rep.private_data = private_data;
2912 rep.private_data_len = private_data_len;
2913
2914 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2915}
2916
e51060f0
SH		 2917int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2918{
2919 struct rdma_id_private *id_priv;
2920 int ret;
2921
2922 id_priv = container_of(id, struct rdma_id_private, id);
83e9502d
NM		 2923
2924 id_priv->owner = task_pid_nr(current);
2925
550e5ca7 2926 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
e51060f0
SH		 2927 return -EINVAL;
2928
2929 if (!id->qp && conn_param) {
2930 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 2931 id_priv->srq = conn_param->srq;
2932 }
2933
07ebafba
TT		 2934 switch (rdma_node_get_transport(id->device->node_type)) {
2935 case RDMA_TRANSPORT_IB:
f45ee80e
HS		 2936 if (id->qp_type == IB_QPT_UD) {
2937 if (conn_param)
2938 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2939 conn_param->private_data,
2940 conn_param->private_data_len);
2941 else
2942 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2943 NULL, 0);
2944 } else {
2945 if (conn_param)
2946 ret = cma_accept_ib(id_priv, conn_param);
2947 else
2948 ret = cma_rep_recv(id_priv);
2949 }
e51060f0 2950 break;
07ebafba
TT		 2951 case RDMA_TRANSPORT_IWARP:
2952 ret = cma_accept_iw(id_priv, conn_param);
2953 break;
e51060f0
SH		 2954 default:
2955 ret = -ENOSYS;
2956 break;
2957 }
2958
2959 if (ret)
2960 goto reject;
2961
2962 return 0;
2963reject:
c5483388 2964 cma_modify_qp_err(id_priv);
e51060f0
SH		 2965 rdma_reject(id, NULL, 0);
2966 return ret;
2967}
2968EXPORT_SYMBOL(rdma_accept);
2969
0fe313b0
SH		 2970int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2971{
2972 struct rdma_id_private *id_priv;
2973 int ret;
2974
2975 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 2976 if (!id_priv->cm_id.ib)
0fe313b0
SH		 2977 return -EINVAL;
2978
2979 switch (id->device->node_type) {
2980 case RDMA_NODE_IB_CA:
2981 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2982 break;
2983 default:
2984 ret = 0;
2985 break;
2986 }
2987 return ret;
2988}
2989EXPORT_SYMBOL(rdma_notify);
2990
e51060f0
SH		 2991int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2992 u8 private_data_len)
2993{
2994 struct rdma_id_private *id_priv;
2995 int ret;
2996
2997 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 2998 if (!id_priv->cm_id.ib)
e51060f0
SH		 2999 return -EINVAL;
3000
07ebafba
TT		 3001 switch (rdma_node_get_transport(id->device->node_type)) {
3002 case RDMA_TRANSPORT_IB:
b26f9b99 3003 if (id->qp_type == IB_QPT_UD)
628e5f6d
SH		 3004 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
3005 private_data, private_data_len);
3006 else
3007 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3008 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3009 0, private_data, private_data_len);
e51060f0 3010 break;
07ebafba
TT		 3011 case RDMA_TRANSPORT_IWARP:
3012 ret = iw_cm_reject(id_priv->cm_id.iw,
3013 private_data, private_data_len);
3014 break;
e51060f0
SH		 3015 default:
3016 ret = -ENOSYS;
3017 break;
3018 }
3019 return ret;
3020}
3021EXPORT_SYMBOL(rdma_reject);
3022
3023int rdma_disconnect(struct rdma_cm_id *id)
3024{
3025 struct rdma_id_private *id_priv;
3026 int ret;
3027
3028 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 3029 if (!id_priv->cm_id.ib)
e51060f0
SH		 3030 return -EINVAL;
3031
07ebafba
TT		 3032 switch (rdma_node_get_transport(id->device->node_type)) {
3033 case RDMA_TRANSPORT_IB:
c5483388 3034 ret = cma_modify_qp_err(id_priv);
07ebafba
TT		 3035 if (ret)
3036 goto out;
e51060f0
SH		 3037 /* Initiate or respond to a disconnect. */
3038 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3039 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3040 break;
07ebafba
TT		 3041 case RDMA_TRANSPORT_IWARP:
3042 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3043 break;
e51060f0 3044 default:
07ebafba 3045 ret = -EINVAL;
e51060f0
SH		 3046 break;
3047 }
3048out:
3049 return ret;
3050}
3051EXPORT_SYMBOL(rdma_disconnect);
3052
c8f6a362
SH		 3053static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3054{
3055 struct rdma_id_private *id_priv;
3056 struct cma_multicast *mc = multicast->context;
3057 struct rdma_cm_event event;
3058 int ret;
3059
3060 id_priv = mc->id_priv;
550e5ca7
NM		 3061 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3062 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
8aa08602 3063 return 0;
c8f6a362 3064
c5483388 3065 mutex_lock(&id_priv->qp_mutex);
c8f6a362
SH		 3066 if (!status && id_priv->id.qp)
3067 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
46ea5061 3068 be16_to_cpu(multicast->rec.mlid));
c5483388 3069 mutex_unlock(&id_priv->qp_mutex);
c8f6a362
SH		 3070
3071 memset(&event, 0, sizeof event);
3072 event.status = status;
3073 event.param.ud.private_data = mc->context;
3074 if (!status) {
3075 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3076 ib_init_ah_from_mcmember(id_priv->id.device,
3077 id_priv->id.port_num, &multicast->rec,
3078 &event.param.ud.ah_attr);
3079 event.param.ud.qp_num = 0xFFFFFF;
3080 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3081 } else
3082 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3083
3084 ret = id_priv->id.event_handler(&id_priv->id, &event);
3085 if (ret) {
550e5ca7 3086 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 3087 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 3088 rdma_destroy_id(&id_priv->id);
3089 return 0;
3090 }
8aa08602 3091
de910bd9 3092 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 3093 return 0;
3094}
3095
3096static void cma_set_mgid(struct rdma_id_private *id_priv,
3097 struct sockaddr *addr, union ib_gid *mgid)
3098{
3099 unsigned char mc_map[MAX_ADDR_LEN];
3100 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3101 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3102 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3103
3104 if (cma_any_addr(addr)) {
3105 memset(mgid, 0, sizeof *mgid);
3106 } else if ((addr->sa_family == AF_INET6) &&
1c9b2819 3107 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
c8f6a362
SH		 3108 0xFF10A01B)) {
3109 /* IPv6 address is an SA assigned MGID. */
3110 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
e2e62697
JG		 3111 } else if ((addr->sa_family == AF_INET6)) {
3112 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3113 if (id_priv->id.ps == RDMA_PS_UDP)
3114 mc_map[7] = 0x01; /* Use RDMA CM signature */
3115 *mgid = *(union ib_gid *) (mc_map + 4);
c8f6a362 3116 } else {
a9e527e3 3117 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
c8f6a362
SH		 3118 if (id_priv->id.ps == RDMA_PS_UDP)
3119 mc_map[7] = 0x01; /* Use RDMA CM signature */
c8f6a362
SH		 3120 *mgid = *(union ib_gid *) (mc_map + 4);
3121 }
3122}
3123
3124static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3125 struct cma_multicast *mc)
3126{
3127 struct ib_sa_mcmember_rec rec;
3128 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3129 ib_sa_comp_mask comp_mask;
3130 int ret;
3131
3132 ib_addr_get_mgid(dev_addr, &rec.mgid);
3133 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3134 &rec.mgid, &rec);
3135 if (ret)
3136 return ret;
3137
3f446754 3138 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
c8f6a362
SH		 3139 if (id_priv->id.ps == RDMA_PS_UDP)
3140 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
6f8372b6 3141 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
c8f6a362
SH		 3142 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3143 rec.join_state = 1;
3144
3145 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3146 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3147 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3148 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3149 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3150
84adeee9
YE		 3151 if (id_priv->id.ps == RDMA_PS_IPOIB)
3152 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
2a22fb8c
DB		 3153 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3154 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3155 IB_SA_MCMEMBER_REC_MTU |
3156 IB_SA_MCMEMBER_REC_HOP_LIMIT;
84adeee9 3157
c8f6a362
SH		 3158 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3159 id_priv->id.port_num, &rec,
3160 comp_mask, GFP_KERNEL,
3161 cma_ib_mc_handler, mc);
4e289045 3162 return PTR_RET(mc->multicast.ib);
c8f6a362
SH		 3163}
3164
3c86aa70
EC		 3165static void iboe_mcast_work_handler(struct work_struct *work)
3166{
3167 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3168 struct cma_multicast *mc = mw->mc;
3169 struct ib_sa_multicast *m = mc->multicast.ib;
3170
3171 mc->multicast.ib->context = mc;
3172 cma_ib_mc_handler(0, m);
3173 kref_put(&mc->mcref, release_mc);
3174 kfree(mw);
3175}
3176
3177static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3178{
3179 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3180 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3181
3182 if (cma_any_addr(addr)) {
3183 memset(mgid, 0, sizeof *mgid);
3184 } else if (addr->sa_family == AF_INET6) {
3185 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3186 } else {
3187 mgid->raw[0] = 0xff;
3188 mgid->raw[1] = 0x0e;
3189 mgid->raw[2] = 0;
3190 mgid->raw[3] = 0;
3191 mgid->raw[4] = 0;
3192 mgid->raw[5] = 0;
3193 mgid->raw[6] = 0;
3194 mgid->raw[7] = 0;
3195 mgid->raw[8] = 0;
3196 mgid->raw[9] = 0;
3197 mgid->raw[10] = 0xff;
3198 mgid->raw[11] = 0xff;
3199 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3200 }
3201}
3202
3203static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3204 struct cma_multicast *mc)
3205{
3206 struct iboe_mcast_work *work;
3207 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3208 int err;
3209 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3210 struct net_device *ndev = NULL;
3211
3212 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3213 return -EINVAL;
3214
3215 work = kzalloc(sizeof *work, GFP_KERNEL);
3216 if (!work)
3217 return -ENOMEM;
3218
3219 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3220 if (!mc->multicast.ib) {
3221 err = -ENOMEM;
3222 goto out1;
3223 }
3224
3225 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3226
3227 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3228 if (id_priv->id.ps == RDMA_PS_UDP)
3229 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3230
3231 if (dev_addr->bound_dev_if)
3232 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3233 if (!ndev) {
3234 err = -ENODEV;
3235 goto out2;
3236 }
3237 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3238 mc->multicast.ib->rec.hop_limit = 1;
3239 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3240 dev_put(ndev);
3241 if (!mc->multicast.ib->rec.mtu) {
3242 err = -EINVAL;
3243 goto out2;
3244 }
3245 iboe_addr_get_sgid(dev_addr, &mc->multicast.ib->rec.port_gid);
3246 work->id = id_priv;
3247 work->mc = mc;
3248 INIT_WORK(&work->work, iboe_mcast_work_handler);
3249 kref_get(&mc->mcref);
3250 queue_work(cma_wq, &work->work);
3251
3252 return 0;
3253
3254out2:
3255 kfree(mc->multicast.ib);
3256out1:
3257 kfree(work);
3258 return err;
3259}
3260
c8f6a362
SH		 3261int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3262 void *context)
3263{
3264 struct rdma_id_private *id_priv;
3265 struct cma_multicast *mc;
3266 int ret;
3267
3268 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7
NM		 3269 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3270 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
c8f6a362
SH		 3271 return -EINVAL;
3272
3273 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3274 if (!mc)
3275 return -ENOMEM;
3276
ef560861 3277 memcpy(&mc->addr, addr, rdma_addr_size(addr));
c8f6a362
SH		 3278 mc->context = context;
3279 mc->id_priv = id_priv;
3280
3281 spin_lock(&id_priv->lock);
3282 list_add(&mc->list, &id_priv->mc_list);
3283 spin_unlock(&id_priv->lock);
3284
3285 switch (rdma_node_get_transport(id->device->node_type)) {
3286 case RDMA_TRANSPORT_IB:
3c86aa70
EC		 3287 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3288 case IB_LINK_LAYER_INFINIBAND:
3289 ret = cma_join_ib_multicast(id_priv, mc);
3290 break;
3291 case IB_LINK_LAYER_ETHERNET:
3292 kref_init(&mc->mcref);
3293 ret = cma_iboe_join_multicast(id_priv, mc);
3294 break;
3295 default:
3296 ret = -EINVAL;
3297 }
c8f6a362
SH		 3298 break;
3299 default:
3300 ret = -ENOSYS;
3301 break;
3302 }
3303
3304 if (ret) {
3305 spin_lock_irq(&id_priv->lock);
3306 list_del(&mc->list);
3307 spin_unlock_irq(&id_priv->lock);
3308 kfree(mc);
3309 }
3310 return ret;
3311}
3312EXPORT_SYMBOL(rdma_join_multicast);
3313
3314void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3315{
3316 struct rdma_id_private *id_priv;
3317 struct cma_multicast *mc;
3318
3319 id_priv = container_of(id, struct rdma_id_private, id);
3320 spin_lock_irq(&id_priv->lock);
3321 list_for_each_entry(mc, &id_priv->mc_list, list) {
ef560861 3322 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
c8f6a362
SH		 3323 list_del(&mc->list);
3324 spin_unlock_irq(&id_priv->lock);
3325
3326 if (id->qp)
3327 ib_detach_mcast(id->qp,
3328 &mc->multicast.ib->rec.mgid,
46ea5061 3329 be16_to_cpu(mc->multicast.ib->rec.mlid));
3c86aa70
EC		 3330 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3331 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3332 case IB_LINK_LAYER_INFINIBAND:
3333 ib_sa_free_multicast(mc->multicast.ib);
3334 kfree(mc);
3335 break;
3336 case IB_LINK_LAYER_ETHERNET:
3337 kref_put(&mc->mcref, release_mc);
3338 break;
3339 default:
3340 break;
3341 }
3342 }
c8f6a362
SH		 3343 return;
3344 }
3345 }
3346 spin_unlock_irq(&id_priv->lock);
3347}
3348EXPORT_SYMBOL(rdma_leave_multicast);
3349
dd5bdff8
OG		 3350static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3351{
3352 struct rdma_dev_addr *dev_addr;
3353 struct cma_ndev_work *work;
3354
3355 dev_addr = &id_priv->id.route.addr.dev_addr;
3356
6266ed6e 3357 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
dd5bdff8
OG		 3358 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3359 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3360 ndev->name, &id_priv->id);
3361 work = kzalloc(sizeof *work, GFP_KERNEL);
3362 if (!work)
3363 return -ENOMEM;
3364
3365 INIT_WORK(&work->work, cma_ndev_work_handler);
3366 work->id = id_priv;
3367 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3368 atomic_inc(&id_priv->refcount);
3369 queue_work(cma_wq, &work->work);
3370 }
3371
3372 return 0;
3373}
3374
3375static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
351638e7 3376 void *ptr)
dd5bdff8 3377{
351638e7 3378 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
dd5bdff8
OG		 3379 struct cma_device *cma_dev;
3380 struct rdma_id_private *id_priv;
3381 int ret = NOTIFY_DONE;
3382
3383 if (dev_net(ndev) != &init_net)
3384 return NOTIFY_DONE;
3385
3386 if (event != NETDEV_BONDING_FAILOVER)
3387 return NOTIFY_DONE;
3388
3389 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3390 return NOTIFY_DONE;
3391
3392 mutex_lock(&lock);
3393 list_for_each_entry(cma_dev, &dev_list, list)
3394 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3395 ret = cma_netdev_change(ndev, id_priv);
3396 if (ret)
3397 goto out;
3398 }
3399
3400out:
3401 mutex_unlock(&lock);
3402 return ret;
3403}
3404
3405static struct notifier_block cma_nb = {
3406 .notifier_call = cma_netdev_callback
3407};
3408
e51060f0
SH		 3409static void cma_add_one(struct ib_device *device)
3410{
3411 struct cma_device *cma_dev;
3412 struct rdma_id_private *id_priv;
3413
3414 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3415 if (!cma_dev)
3416 return;
3417
3418 cma_dev->device = device;
e51060f0
SH		 3419
3420 init_completion(&cma_dev->comp);
3421 atomic_set(&cma_dev->refcount, 1);
3422 INIT_LIST_HEAD(&cma_dev->id_list);
3423 ib_set_client_data(device, &cma_client, cma_dev);
3424
3425 mutex_lock(&lock);
3426 list_add_tail(&cma_dev->list, &dev_list);
3427 list_for_each_entry(id_priv, &listen_any_list, list)
3428 cma_listen_on_dev(id_priv, cma_dev);
3429 mutex_unlock(&lock);
e51060f0
SH		 3430}
3431
3432static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3433{
a1b1b61f 3434 struct rdma_cm_event event;
550e5ca7 3435 enum rdma_cm_state state;
de910bd9 3436 int ret = 0;
e51060f0
SH		 3437
3438 /* Record that we want to remove the device */
550e5ca7
NM		 3439 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3440 if (state == RDMA_CM_DESTROYING)
e51060f0
SH		 3441 return 0;
3442
3443 cma_cancel_operation(id_priv, state);
de910bd9 3444 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 3445
3446 /* Check for destruction from another callback. */
550e5ca7 3447 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
de910bd9 3448 goto out;
e51060f0 3449
a1b1b61f
SH		 3450 memset(&event, 0, sizeof event);
3451 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
de910bd9
OG		 3452 ret = id_priv->id.event_handler(&id_priv->id, &event);
3453out:
3454 mutex_unlock(&id_priv->handler_mutex);
3455 return ret;
e51060f0
SH		 3456}
3457
3458static void cma_process_remove(struct cma_device *cma_dev)
3459{
e51060f0
SH		 3460 struct rdma_id_private *id_priv;
3461 int ret;
3462
e51060f0
SH		 3463 mutex_lock(&lock);
3464 while (!list_empty(&cma_dev->id_list)) {
3465 id_priv = list_entry(cma_dev->id_list.next,
3466 struct rdma_id_private, list);
3467
d02d1f53 3468 list_del(&id_priv->listen_list);
94de178a 3469 list_del_init(&id_priv->list);
e51060f0
SH		 3470 atomic_inc(&id_priv->refcount);
3471 mutex_unlock(&lock);
3472
d02d1f53 3473 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
e51060f0
SH		 3474 cma_deref_id(id_priv);
3475 if (ret)
3476 rdma_destroy_id(&id_priv->id);
3477
3478 mutex_lock(&lock);
3479 }
3480 mutex_unlock(&lock);
3481
3482 cma_deref_dev(cma_dev);
3483 wait_for_completion(&cma_dev->comp);
3484}
3485
3486static void cma_remove_one(struct ib_device *device)
3487{
3488 struct cma_device *cma_dev;
3489
3490 cma_dev = ib_get_client_data(device, &cma_client);
3491 if (!cma_dev)
3492 return;
3493
3494 mutex_lock(&lock);
3495 list_del(&cma_dev->list);
3496 mutex_unlock(&lock);
3497
3498 cma_process_remove(cma_dev);
3499 kfree(cma_dev);
3500}
3501
753f618a
NM		 3502static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3503{
3504 struct nlmsghdr *nlh;
3505 struct rdma_cm_id_stats *id_stats;
3506 struct rdma_id_private *id_priv;
3507 struct rdma_cm_id *id = NULL;
3508 struct cma_device *cma_dev;
3509 int i_dev = 0, i_id = 0;
3510
3511 /*
3512 * We export all of the IDs as a sequence of messages. Each
3513 * ID gets its own netlink message.
3514 */
3515 mutex_lock(&lock);
3516
3517 list_for_each_entry(cma_dev, &dev_list, list) {
3518 if (i_dev < cb->args[0]) {
3519 i_dev++;
3520 continue;
3521 }
3522
3523 i_id = 0;
3524 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3525 if (i_id < cb->args[1]) {
3526 i_id++;
3527 continue;
3528 }
3529
3530 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3531 sizeof *id_stats, RDMA_NL_RDMA_CM,
3532 RDMA_NL_RDMA_CM_ID_STATS);
3533 if (!id_stats)
3534 goto out;
3535
3536 memset(id_stats, 0, sizeof *id_stats);
3537 id = &id_priv->id;
3538 id_stats->node_type = id->route.addr.dev_addr.dev_type;
3539 id_stats->port_num = id->port_num;
3540 id_stats->bound_dev_if =
3541 id->route.addr.dev_addr.bound_dev_if;
3542
f4753834 3543 if (cma_family(id_priv) == AF_INET) {
753f618a
NM		 3544 if (ibnl_put_attr(skb, nlh,
3545 sizeof(struct sockaddr_in),
f4753834 3546 cma_src_addr(id_priv),
753f618a
NM		 3547 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
3548 goto out;
3549 }
3550 if (ibnl_put_attr(skb, nlh,
3551 sizeof(struct sockaddr_in),
f4753834 3552 cma_dst_addr(id_priv),
753f618a
NM		 3553 RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
3554 goto out;
3555 }
f4753834 3556 } else if (cma_family(id_priv) == AF_INET6) {
753f618a
NM		 3557 if (ibnl_put_attr(skb, nlh,
3558 sizeof(struct sockaddr_in6),
f4753834 3559 cma_src_addr(id_priv),
753f618a
NM		 3560 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
3561 goto out;
3562 }
3563 if (ibnl_put_attr(skb, nlh,
3564 sizeof(struct sockaddr_in6),
f4753834 3565 cma_dst_addr(id_priv),
753f618a
NM		 3566 RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
3567 goto out;
3568 }
3569 }
3570
83e9502d 3571 id_stats->pid = id_priv->owner;
753f618a
NM		 3572 id_stats->port_space = id->ps;
3573 id_stats->cm_state = id_priv->state;
3574 id_stats->qp_num = id_priv->qp_num;
3575 id_stats->qp_type = id->qp_type;
3576
3577 i_id++;
3578 }
3579
3580 cb->args[1] = 0;
3581 i_dev++;
3582 }
3583
3584out:
3585 mutex_unlock(&lock);
3586 cb->args[0] = i_dev;
3587 cb->args[1] = i_id;
3588
3589 return skb->len;
3590}
3591
3592static const struct ibnl_client_cbs cma_cb_table[] = {
809d5fc9
G		 3593 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3594 .module = THIS_MODULE },
753f618a
NM		 3595};
3596
716abb1f 3597static int __init cma_init(void)
e51060f0 3598{
5d7220e8 3599 int ret;
227b60f5 3600
c7f743a6 3601 cma_wq = create_singlethread_workqueue("rdma_cm");
e51060f0
SH		 3602 if (!cma_wq)
3603 return -ENOMEM;
3604
c1a0b23b 3605 ib_sa_register_client(&sa_client);
7a118df3 3606 rdma_addr_register_client(&addr_client);
dd5bdff8 3607 register_netdevice_notifier(&cma_nb);
c1a0b23b 3608
e51060f0
SH		 3609 ret = ib_register_client(&cma_client);
3610 if (ret)
3611 goto err;
753f618a
NM		 3612
3613 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3614 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3615
e51060f0
SH		 3616 return 0;
3617
3618err:
dd5bdff8 3619 unregister_netdevice_notifier(&cma_nb);
7a118df3 3620 rdma_addr_unregister_client(&addr_client);
c1a0b23b 3621 ib_sa_unregister_client(&sa_client);
e51060f0
SH		 3622 destroy_workqueue(cma_wq);
3623 return ret;
3624}
3625
716abb1f 3626static void __exit cma_cleanup(void)
e51060f0 3627{
753f618a 3628 ibnl_remove_client(RDMA_NL_RDMA_CM);
e51060f0 3629 ib_unregister_client(&cma_client);
dd5bdff8 3630 unregister_netdevice_notifier(&cma_nb);
7a118df3 3631 rdma_addr_unregister_client(&addr_client);
c1a0b23b 3632 ib_sa_unregister_client(&sa_client);
e51060f0
SH		 3633 destroy_workqueue(cma_wq);
3634 idr_destroy(&sdp_ps);
3635 idr_destroy(&tcp_ps);
628e5f6d 3636 idr_destroy(&udp_ps);
c8f6a362 3637 idr_destroy(&ipoib_ps);
2d2e9415 3638 idr_destroy(&ib_ps);
e51060f0
SH		 3639}
3640
3641module_init(cma_init);
3642module_exit(cma_cleanup);
Linux 6.x block layer and io_uring tree(s)