projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
RDMA/cma: Bind to a specific address family
[deliverable/linux.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>
e51060f0
SH		 53#include <rdma/ib_cache.h>
54#include <rdma/ib_cm.h>
55#include <rdma/ib_sa.h>
07ebafba 56#include <rdma/iw_cm.h>
e51060f0
SH		 57
58MODULE_AUTHOR("Sean Hefty");
59MODULE_DESCRIPTION("Generic RDMA CM Agent");
60MODULE_LICENSE("Dual BSD/GPL");
61
62#define CMA_CM_RESPONSE_TIMEOUT 20
d5bb7599 63#define CMA_MAX_CM_RETRIES 15
dcb3f974 64#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
3c86aa70 65#define CMA_IBOE_PACKET_LIFETIME 18
e51060f0
SH		 66
67static void cma_add_one(struct ib_device *device);
68static void cma_remove_one(struct ib_device *device);
69
70static struct ib_client cma_client = {
71 .name = "cma",
72 .add = cma_add_one,
73 .remove = cma_remove_one
74};
75
c1a0b23b 76static struct ib_sa_client sa_client;
7a118df3 77static struct rdma_addr_client addr_client;
e51060f0
SH		 78static LIST_HEAD(dev_list);
79static LIST_HEAD(listen_any_list);
80static DEFINE_MUTEX(lock);
81static struct workqueue_struct *cma_wq;
82static DEFINE_IDR(sdp_ps);
83static DEFINE_IDR(tcp_ps);
628e5f6d 84static DEFINE_IDR(udp_ps);
c8f6a362 85static DEFINE_IDR(ipoib_ps);
2d2e9415 86static DEFINE_IDR(ib_ps);
e51060f0
SH		 87
88struct cma_device {
89 struct list_head list;
90 struct ib_device *device;
e51060f0
SH		 91 struct completion comp;
92 atomic_t refcount;
93 struct list_head id_list;
94};
95
e51060f0
SH		 96struct rdma_bind_list {
97 struct idr *ps;
98 struct hlist_head owners;
99 unsigned short port;
100};
101
102/*
103 * Device removal can occur at anytime, so we need extra handling to
104 * serialize notifying the user of device removal with other callbacks.
105 * We do this by disabling removal notification while a callback is in process,
106 * and reporting it after the callback completes.
107 */
108struct rdma_id_private {
109 struct rdma_cm_id id;
110
111 struct rdma_bind_list *bind_list;
112 struct hlist_node node;
d02d1f53
SH		 113 struct list_head list; /* listen_any_list or cma_device.list */
114 struct list_head listen_list; /* per device listens */
e51060f0 115 struct cma_device *cma_dev;
c8f6a362 116 struct list_head mc_list;
e51060f0 117
d02d1f53 118 int internal_id;
550e5ca7 119 enum rdma_cm_state state;
e51060f0 120 spinlock_t lock;
c5483388
SH		 121 struct mutex qp_mutex;
122
e51060f0
SH		 123 struct completion comp;
124 atomic_t refcount;
de910bd9 125 struct mutex handler_mutex;
e51060f0
SH		 126
127 int backlog;
128 int timeout_ms;
129 struct ib_sa_query *query;
130 int query_id;
131 union {
132 struct ib_cm_id *ib;
07ebafba 133 struct iw_cm_id *iw;
e51060f0
SH		 134 } cm_id;
135
136 u32 seq_num;
c8f6a362 137 u32 qkey;
e51060f0 138 u32 qp_num;
83e9502d 139 pid_t owner;
e51060f0 140 u8 srq;
a81c994d 141 u8 tos;
a9bb7912 142 u8 reuseaddr;
5b0ec991 143 u8 afonly;
e51060f0
SH		 144};
145
c8f6a362
SH		 146struct cma_multicast {
147 struct rdma_id_private *id_priv;
148 union {
149 struct ib_sa_multicast *ib;
150 } multicast;
151 struct list_head list;
152 void *context;
3f446754 153 struct sockaddr_storage addr;
3c86aa70 154 struct kref mcref;
c8f6a362
SH		 155};
156
e51060f0
SH		 157struct cma_work {
158 struct work_struct work;
159 struct rdma_id_private *id;
550e5ca7
NM		 160 enum rdma_cm_state old_state;
161 enum rdma_cm_state new_state;
e51060f0
SH		 162 struct rdma_cm_event event;
163};
164
dd5bdff8
OG		 165struct cma_ndev_work {
166 struct work_struct work;
167 struct rdma_id_private *id;
168 struct rdma_cm_event event;
169};
170
3c86aa70
EC		 171struct iboe_mcast_work {
172 struct work_struct work;
173 struct rdma_id_private *id;
174 struct cma_multicast *mc;
175};
176
e51060f0
SH		 177union cma_ip_addr {
178 struct in6_addr ip6;
179 struct {
1b90c137
AV		 180 __be32 pad[3];
181 __be32 addr;
e51060f0
SH		 182 } ip4;
183};
184
185struct cma_hdr {
186 u8 cma_version;
187 u8 ip_version; /* IP version: 7:4 */
1b90c137 188 __be16 port;
e51060f0
SH		 189 union cma_ip_addr src_addr;
190 union cma_ip_addr dst_addr;
191};
192
193struct sdp_hh {
194 u8 bsdh[16];
195 u8 sdp_version; /* Major version: 7:4 */
196 u8 ip_version; /* IP version: 7:4 */
197 u8 sdp_specific1[10];
1b90c137
AV		 198 __be16 port;
199 __be16 sdp_specific2;
e51060f0
SH		 200 union cma_ip_addr src_addr;
201 union cma_ip_addr dst_addr;
202};
203
204struct sdp_hah {
205 u8 bsdh[16];
206 u8 sdp_version;
207};
208
209#define CMA_VERSION 0x00
210#define SDP_MAJ_VERSION 0x2
211
550e5ca7 212static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
e51060f0
SH		 213{
214 unsigned long flags;
215 int ret;
216
217 spin_lock_irqsave(&id_priv->lock, flags);
218 ret = (id_priv->state == comp);
219 spin_unlock_irqrestore(&id_priv->lock, flags);
220 return ret;
221}
222
223static int cma_comp_exch(struct rdma_id_private *id_priv,
550e5ca7 224 enum rdma_cm_state comp, enum rdma_cm_state exch)
e51060f0
SH		 225{
226 unsigned long flags;
227 int ret;
228
229 spin_lock_irqsave(&id_priv->lock, flags);
230 if ((ret = (id_priv->state == comp)))
231 id_priv->state = exch;
232 spin_unlock_irqrestore(&id_priv->lock, flags);
233 return ret;
234}
235
550e5ca7
NM		 236static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
237 enum rdma_cm_state exch)
e51060f0
SH		 238{
239 unsigned long flags;
550e5ca7 240 enum rdma_cm_state old;
e51060f0
SH		 241
242 spin_lock_irqsave(&id_priv->lock, flags);
243 old = id_priv->state;
244 id_priv->state = exch;
245 spin_unlock_irqrestore(&id_priv->lock, flags);
246 return old;
247}
248
249static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
250{
251 return hdr->ip_version >> 4;
252}
253
254static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
255{
256 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
257}
258
259static inline u8 sdp_get_majv(u8 sdp_version)
260{
261 return sdp_version >> 4;
262}
263
264static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
265{
266 return hh->ip_version >> 4;
267}
268
269static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
270{
271 hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
272}
273
274static void cma_attach_to_dev(struct rdma_id_private *id_priv,
275 struct cma_device *cma_dev)
276{
277 atomic_inc(&cma_dev->refcount);
278 id_priv->cma_dev = cma_dev;
279 id_priv->id.device = cma_dev->device;
3c86aa70
EC		 280 id_priv->id.route.addr.dev_addr.transport =
281 rdma_node_get_transport(cma_dev->device->node_type);
e51060f0
SH		 282 list_add_tail(&id_priv->list, &cma_dev->id_list);
283}
284
285static inline void cma_deref_dev(struct cma_device *cma_dev)
286{
287 if (atomic_dec_and_test(&cma_dev->refcount))
288 complete(&cma_dev->comp);
289}
290
3c86aa70
EC		 291static inline void release_mc(struct kref *kref)
292{
293 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
294
295 kfree(mc->multicast.ib);
296 kfree(mc);
297}
298
a396d43a 299static void cma_release_dev(struct rdma_id_private *id_priv)
e51060f0 300{
a396d43a 301 mutex_lock(&lock);
e51060f0
SH		 302 list_del(&id_priv->list);
303 cma_deref_dev(id_priv->cma_dev);
304 id_priv->cma_dev = NULL;
a396d43a 305 mutex_unlock(&lock);
e51060f0
SH		 306}
307
d2ca39f2 308static int cma_set_qkey(struct rdma_id_private *id_priv)
c8f6a362
SH		 309{
310 struct ib_sa_mcmember_rec rec;
311 int ret = 0;
312
d2ca39f2
YE		 313 if (id_priv->qkey)
314 return 0;
315
316 switch (id_priv->id.ps) {
c8f6a362 317 case RDMA_PS_UDP:
d2ca39f2 318 id_priv->qkey = RDMA_UDP_QKEY;
c8f6a362
SH		 319 break;
320 case RDMA_PS_IPOIB:
d2ca39f2
YE		 321 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
322 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
323 id_priv->id.port_num, &rec.mgid,
324 &rec);
325 if (!ret)
326 id_priv->qkey = be32_to_cpu(rec.qkey);
c8f6a362
SH		 327 break;
328 default:
329 break;
330 }
331 return ret;
332}
333
3c86aa70
EC		 334static int find_gid_port(struct ib_device *device, union ib_gid *gid, u8 port_num)
335{
336 int i;
337 int err;
338 struct ib_port_attr props;
339 union ib_gid tmp;
340
341 err = ib_query_port(device, port_num, &props);
342 if (err)
343 return 1;
344
345 for (i = 0; i < props.gid_tbl_len; ++i) {
346 err = ib_query_gid(device, port_num, i, &tmp);
347 if (err)
348 return 1;
349 if (!memcmp(&tmp, gid, sizeof tmp))
350 return 0;
351 }
352
353 return -EAGAIN;
354}
355
07ebafba 356static int cma_acquire_dev(struct rdma_id_private *id_priv)
e51060f0 357{
c8f6a362 358 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
e51060f0 359 struct cma_device *cma_dev;
3c86aa70 360 union ib_gid gid, iboe_gid;
e51060f0 361 int ret = -ENODEV;
3c86aa70
EC		 362 u8 port;
363 enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
364 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
e51060f0 365
2efdd6a0
MS		 366 if (dev_ll != IB_LINK_LAYER_INFINIBAND &&
367 id_priv->id.ps == RDMA_PS_IPOIB)
368 return -EINVAL;
369
a396d43a 370 mutex_lock(&lock);
3c86aa70
EC		 371 iboe_addr_get_sgid(dev_addr, &iboe_gid);
372 memcpy(&gid, dev_addr->src_dev_addr +
373 rdma_addr_gid_offset(dev_addr), sizeof gid);
e51060f0 374 list_for_each_entry(cma_dev, &dev_list, list) {
3c86aa70
EC		 375 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
376 if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
377 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
378 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
379 ret = find_gid_port(cma_dev->device, &iboe_gid, port);
380 else
381 ret = find_gid_port(cma_dev->device, &gid, port);
382
383 if (!ret) {
384 id_priv->id.port_num = port;
385 goto out;
386 } else if (ret == 1)
387 break;
388 }
e51060f0
SH		 389 }
390 }
3c86aa70
EC		 391
392out:
393 if (!ret)
394 cma_attach_to_dev(id_priv, cma_dev);
395
a396d43a 396 mutex_unlock(&lock);
e51060f0
SH		 397 return ret;
398}
399
e51060f0
SH		 400static void cma_deref_id(struct rdma_id_private *id_priv)
401{
402 if (atomic_dec_and_test(&id_priv->refcount))
403 complete(&id_priv->comp);
404}
405
de910bd9 406static int cma_disable_callback(struct rdma_id_private *id_priv,
550e5ca7 407 enum rdma_cm_state state)
8aa08602 408{
de910bd9
OG		 409 mutex_lock(&id_priv->handler_mutex);
410 if (id_priv->state != state) {
411 mutex_unlock(&id_priv->handler_mutex);
412 return -EINVAL;
413 }
414 return 0;
e51060f0
SH		 415}
416
417struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
b26f9b99
SH		 418 void *context, enum rdma_port_space ps,
419 enum ib_qp_type qp_type)
e51060f0
SH		 420{
421 struct rdma_id_private *id_priv;
422
423 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
424 if (!id_priv)
425 return ERR_PTR(-ENOMEM);
426
83e9502d 427 id_priv->owner = task_pid_nr(current);
550e5ca7 428 id_priv->state = RDMA_CM_IDLE;
e51060f0
SH		 429 id_priv->id.context = context;
430 id_priv->id.event_handler = event_handler;
431 id_priv->id.ps = ps;
b26f9b99 432 id_priv->id.qp_type = qp_type;
e51060f0 433 spin_lock_init(&id_priv->lock);
c5483388 434 mutex_init(&id_priv->qp_mutex);
e51060f0
SH		 435 init_completion(&id_priv->comp);
436 atomic_set(&id_priv->refcount, 1);
de910bd9 437 mutex_init(&id_priv->handler_mutex);
e51060f0 438 INIT_LIST_HEAD(&id_priv->listen_list);
c8f6a362 439 INIT_LIST_HEAD(&id_priv->mc_list);
e51060f0
SH		 440 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
441
442 return &id_priv->id;
443}
444EXPORT_SYMBOL(rdma_create_id);
445
c8f6a362 446static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
e51060f0
SH		 447{
448 struct ib_qp_attr qp_attr;
c8f6a362 449 int qp_attr_mask, ret;
e51060f0 450
c8f6a362
SH		 451 qp_attr.qp_state = IB_QPS_INIT;
452 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0
SH		 453 if (ret)
454 return ret;
455
c8f6a362
SH		 456 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
457 if (ret)
458 return ret;
459
460 qp_attr.qp_state = IB_QPS_RTR;
461 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
462 if (ret)
463 return ret;
464
465 qp_attr.qp_state = IB_QPS_RTS;
466 qp_attr.sq_psn = 0;
467 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
468
469 return ret;
e51060f0
SH		 470}
471
c8f6a362 472static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
07ebafba
TT		 473{
474 struct ib_qp_attr qp_attr;
c8f6a362 475 int qp_attr_mask, ret;
07ebafba
TT		 476
477 qp_attr.qp_state = IB_QPS_INIT;
c8f6a362
SH		 478 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
479 if (ret)
480 return ret;
07ebafba 481
c8f6a362 482 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
07ebafba
TT		 483}
484
e51060f0
SH		 485int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
486 struct ib_qp_init_attr *qp_init_attr)
487{
488 struct rdma_id_private *id_priv;
489 struct ib_qp *qp;
490 int ret;
491
492 id_priv = container_of(id, struct rdma_id_private, id);
493 if (id->device != pd->device)
494 return -EINVAL;
495
496 qp = ib_create_qp(pd, qp_init_attr);
497 if (IS_ERR(qp))
498 return PTR_ERR(qp);
499
b26f9b99 500 if (id->qp_type == IB_QPT_UD)
c8f6a362
SH		 501 ret = cma_init_ud_qp(id_priv, qp);
502 else
503 ret = cma_init_conn_qp(id_priv, qp);
e51060f0
SH		 504 if (ret)
505 goto err;
506
507 id->qp = qp;
508 id_priv->qp_num = qp->qp_num;
e51060f0
SH		 509 id_priv->srq = (qp->srq != NULL);
510 return 0;
511err:
512 ib_destroy_qp(qp);
513 return ret;
514}
515EXPORT_SYMBOL(rdma_create_qp);
516
517void rdma_destroy_qp(struct rdma_cm_id *id)
518{
c5483388
SH		 519 struct rdma_id_private *id_priv;
520
521 id_priv = container_of(id, struct rdma_id_private, id);
522 mutex_lock(&id_priv->qp_mutex);
523 ib_destroy_qp(id_priv->id.qp);
524 id_priv->id.qp = NULL;
525 mutex_unlock(&id_priv->qp_mutex);
e51060f0
SH		 526}
527EXPORT_SYMBOL(rdma_destroy_qp);
528
5851bb89
SH		 529static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
530 struct rdma_conn_param *conn_param)
e51060f0
SH		 531{
532 struct ib_qp_attr qp_attr;
533 int qp_attr_mask, ret;
534
c5483388
SH		 535 mutex_lock(&id_priv->qp_mutex);
536 if (!id_priv->id.qp) {
537 ret = 0;
538 goto out;
539 }
e51060f0
SH		 540
541 /* Need to update QP attributes from default values. */
542 qp_attr.qp_state = IB_QPS_INIT;
c5483388 543 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 544 if (ret)
c5483388 545 goto out;
e51060f0 546
c5483388 547 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
e51060f0 548 if (ret)
c5483388 549 goto out;
e51060f0
SH		 550
551 qp_attr.qp_state = IB_QPS_RTR;
c5483388 552 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 553 if (ret)
c5483388 554 goto out;
e51060f0 555
5851bb89
SH		 556 if (conn_param)
557 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
c5483388
SH		 558 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
559out:
560 mutex_unlock(&id_priv->qp_mutex);
561 return ret;
e51060f0
SH		 562}
563
5851bb89
SH		 564static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
565 struct rdma_conn_param *conn_param)
e51060f0
SH		 566{
567 struct ib_qp_attr qp_attr;
568 int qp_attr_mask, ret;
569
c5483388
SH		 570 mutex_lock(&id_priv->qp_mutex);
571 if (!id_priv->id.qp) {
572 ret = 0;
573 goto out;
574 }
e51060f0
SH		 575
576 qp_attr.qp_state = IB_QPS_RTS;
c5483388 577 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 578 if (ret)
c5483388 579 goto out;
e51060f0 580
5851bb89
SH		 581 if (conn_param)
582 qp_attr.max_rd_atomic = conn_param->initiator_depth;
c5483388
SH		 583 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
584out:
585 mutex_unlock(&id_priv->qp_mutex);
586 return ret;
e51060f0
SH		 587}
588
c5483388 589static int cma_modify_qp_err(struct rdma_id_private *id_priv)
e51060f0
SH		 590{
591 struct ib_qp_attr qp_attr;
c5483388 592 int ret;
e51060f0 593
c5483388
SH		 594 mutex_lock(&id_priv->qp_mutex);
595 if (!id_priv->id.qp) {
596 ret = 0;
597 goto out;
598 }
e51060f0
SH		 599
600 qp_attr.qp_state = IB_QPS_ERR;
c5483388
SH		 601 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
602out:
603 mutex_unlock(&id_priv->qp_mutex);
604 return ret;
e51060f0
SH		 605}
606
c8f6a362
SH		 607static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
608 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
609{
610 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
611 int ret;
3c86aa70
EC		 612 u16 pkey;
613
614 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
615 IB_LINK_LAYER_INFINIBAND)
616 pkey = ib_addr_get_pkey(dev_addr);
617 else
618 pkey = 0xffff;
c8f6a362
SH		 619
620 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
3c86aa70 621 pkey, &qp_attr->pkey_index);
c8f6a362
SH		 622 if (ret)
623 return ret;
624
625 qp_attr->port_num = id_priv->id.port_num;
626 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
627
b26f9b99 628 if (id_priv->id.qp_type == IB_QPT_UD) {
d2ca39f2
YE		 629 ret = cma_set_qkey(id_priv);
630 if (ret)
631 return ret;
632
c8f6a362
SH		 633 qp_attr->qkey = id_priv->qkey;
634 *qp_attr_mask |= IB_QP_QKEY;
635 } else {
636 qp_attr->qp_access_flags = 0;
637 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
638 }
639 return 0;
640}
641
e51060f0
SH		 642int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
643 int *qp_attr_mask)
644{
645 struct rdma_id_private *id_priv;
c8f6a362 646 int ret = 0;
e51060f0
SH		 647
648 id_priv = container_of(id, struct rdma_id_private, id);
07ebafba
TT		 649 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
650 case RDMA_TRANSPORT_IB:
b26f9b99 651 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
c8f6a362
SH		 652 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
653 else
654 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
655 qp_attr_mask);
e51060f0
SH		 656 if (qp_attr->qp_state == IB_QPS_RTR)
657 qp_attr->rq_psn = id_priv->seq_num;
658 break;
07ebafba 659 case RDMA_TRANSPORT_IWARP:
c8f6a362 660 if (!id_priv->cm_id.iw) {
8f076531 661 qp_attr->qp_access_flags = 0;
c8f6a362
SH		 662 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
663 } else
664 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
665 qp_attr_mask);
07ebafba 666 break;
e51060f0
SH		 667 default:
668 ret = -ENOSYS;
669 break;
670 }
671
672 return ret;
673}
674EXPORT_SYMBOL(rdma_init_qp_attr);
675
676static inline int cma_zero_addr(struct sockaddr *addr)
677{
678 struct in6_addr *ip6;
679
680 if (addr->sa_family == AF_INET)
6360a02a
JP		 681 return ipv4_is_zeronet(
682 ((struct sockaddr_in *)addr)->sin_addr.s_addr);
e51060f0
SH		 683 else {
684 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
685 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
5fd571cb 686 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
e51060f0
SH		 687 }
688}
689
690static inline int cma_loopback_addr(struct sockaddr *addr)
691{
1f5175ad
AS		 692 if (addr->sa_family == AF_INET)
693 return ipv4_is_loopback(
694 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
695 else
696 return ipv6_addr_loopback(
697 &((struct sockaddr_in6 *) addr)->sin6_addr);
e51060f0
SH		 698}
699
700static inline int cma_any_addr(struct sockaddr *addr)
701{
702 return cma_zero_addr(addr) || cma_loopback_addr(addr);
703}
704
43b752da
HS		 705static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
706{
707 if (src->sa_family != dst->sa_family)
708 return -1;
709
710 switch (src->sa_family) {
711 case AF_INET:
712 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
713 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
714 default:
715 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
716 &((struct sockaddr_in6 *) dst)->sin6_addr);
717 }
718}
719
628e5f6d
SH		 720static inline __be16 cma_port(struct sockaddr *addr)
721{
722 if (addr->sa_family == AF_INET)
723 return ((struct sockaddr_in *) addr)->sin_port;
724 else
725 return ((struct sockaddr_in6 *) addr)->sin6_port;
726}
727
e51060f0
SH		 728static inline int cma_any_port(struct sockaddr *addr)
729{
628e5f6d 730 return !cma_port(addr);
e51060f0
SH		 731}
732
733static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
1b90c137 734 u8 *ip_ver, __be16 *port,
e51060f0
SH		 735 union cma_ip_addr **src, union cma_ip_addr **dst)
736{
737 switch (ps) {
738 case RDMA_PS_SDP:
739 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
740 SDP_MAJ_VERSION)
741 return -EINVAL;
742
743 *ip_ver = sdp_get_ip_ver(hdr);
744 *port = ((struct sdp_hh *) hdr)->port;
745 *src = &((struct sdp_hh *) hdr)->src_addr;
746 *dst = &((struct sdp_hh *) hdr)->dst_addr;
747 break;
748 default:
749 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
750 return -EINVAL;
751
752 *ip_ver = cma_get_ip_ver(hdr);
753 *port = ((struct cma_hdr *) hdr)->port;
754 *src = &((struct cma_hdr *) hdr)->src_addr;
755 *dst = &((struct cma_hdr *) hdr)->dst_addr;
756 break;
757 }
758
759 if (*ip_ver != 4 && *ip_ver != 6)
760 return -EINVAL;
761 return 0;
762}
763
764static void cma_save_net_info(struct rdma_addr *addr,
765 struct rdma_addr *listen_addr,
1b90c137 766 u8 ip_ver, __be16 port,
e51060f0
SH		 767 union cma_ip_addr *src, union cma_ip_addr *dst)
768{
769 struct sockaddr_in *listen4, *ip4;
770 struct sockaddr_in6 *listen6, *ip6;
771
772 switch (ip_ver) {
773 case 4:
774 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
775 ip4 = (struct sockaddr_in *) &addr->src_addr;
776 ip4->sin_family = listen4->sin_family;
777 ip4->sin_addr.s_addr = dst->ip4.addr;
778 ip4->sin_port = listen4->sin_port;
779
780 ip4 = (struct sockaddr_in *) &addr->dst_addr;
781 ip4->sin_family = listen4->sin_family;
782 ip4->sin_addr.s_addr = src->ip4.addr;
783 ip4->sin_port = port;
784 break;
785 case 6:
786 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
787 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
788 ip6->sin6_family = listen6->sin6_family;
789 ip6->sin6_addr = dst->ip6;
790 ip6->sin6_port = listen6->sin6_port;
791
792 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
793 ip6->sin6_family = listen6->sin6_family;
794 ip6->sin6_addr = src->ip6;
795 ip6->sin6_port = port;
796 break;
797 default:
798 break;
799 }
800}
801
802static inline int cma_user_data_offset(enum rdma_port_space ps)
803{
804 switch (ps) {
805 case RDMA_PS_SDP:
806 return 0;
807 default:
808 return sizeof(struct cma_hdr);
809 }
810}
811
e51060f0
SH		 812static void cma_cancel_route(struct rdma_id_private *id_priv)
813{
3c86aa70
EC		 814 switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
815 case IB_LINK_LAYER_INFINIBAND:
e51060f0
SH		 816 if (id_priv->query)
817 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
818 break;
819 default:
820 break;
821 }
822}
823
e51060f0
SH		 824static void cma_cancel_listens(struct rdma_id_private *id_priv)
825{
826 struct rdma_id_private *dev_id_priv;
827
d02d1f53
SH		 828 /*
829 * Remove from listen_any_list to prevent added devices from spawning
830 * additional listen requests.
831 */
e51060f0
SH		 832 mutex_lock(&lock);
833 list_del(&id_priv->list);
834
835 while (!list_empty(&id_priv->listen_list)) {
836 dev_id_priv = list_entry(id_priv->listen_list.next,
837 struct rdma_id_private, listen_list);
d02d1f53
SH		 838 /* sync with device removal to avoid duplicate destruction */
839 list_del_init(&dev_id_priv->list);
840 list_del(&dev_id_priv->listen_list);
841 mutex_unlock(&lock);
842
843 rdma_destroy_id(&dev_id_priv->id);
844 mutex_lock(&lock);
e51060f0
SH		 845 }
846 mutex_unlock(&lock);
847}
848
849static void cma_cancel_operation(struct rdma_id_private *id_priv,
550e5ca7 850 enum rdma_cm_state state)
e51060f0
SH		 851{
852 switch (state) {
550e5ca7 853 case RDMA_CM_ADDR_QUERY:
e51060f0
SH		 854 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
855 break;
550e5ca7 856 case RDMA_CM_ROUTE_QUERY:
e51060f0
SH		 857 cma_cancel_route(id_priv);
858 break;
550e5ca7 859 case RDMA_CM_LISTEN:
3f446754
RD		 860 if (cma_any_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr)
861 && !id_priv->cma_dev)
e51060f0
SH		 862 cma_cancel_listens(id_priv);
863 break;
864 default:
865 break;
866 }
867}
868
869static void cma_release_port(struct rdma_id_private *id_priv)
870{
871 struct rdma_bind_list *bind_list = id_priv->bind_list;
872
873 if (!bind_list)
874 return;
875
876 mutex_lock(&lock);
877 hlist_del(&id_priv->node);
878 if (hlist_empty(&bind_list->owners)) {
879 idr_remove(bind_list->ps, bind_list->port);
880 kfree(bind_list);
881 }
882 mutex_unlock(&lock);
883}
884
c8f6a362
SH		 885static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
886{
887 struct cma_multicast *mc;
888
889 while (!list_empty(&id_priv->mc_list)) {
890 mc = container_of(id_priv->mc_list.next,
891 struct cma_multicast, list);
892 list_del(&mc->list);
3c86aa70
EC		 893 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
894 case IB_LINK_LAYER_INFINIBAND:
895 ib_sa_free_multicast(mc->multicast.ib);
896 kfree(mc);
897 break;
898 case IB_LINK_LAYER_ETHERNET:
899 kref_put(&mc->mcref, release_mc);
900 break;
901 default:
902 break;
903 }
c8f6a362
SH		 904 }
905}
906
e51060f0
SH		 907void rdma_destroy_id(struct rdma_cm_id *id)
908{
909 struct rdma_id_private *id_priv;
550e5ca7 910 enum rdma_cm_state state;
e51060f0
SH		 911
912 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 913 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
e51060f0
SH		 914 cma_cancel_operation(id_priv, state);
915
a396d43a
SH		 916 /*
917 * Wait for any active callback to finish. New callbacks will find
918 * the id_priv state set to destroying and abort.
919 */
920 mutex_lock(&id_priv->handler_mutex);
921 mutex_unlock(&id_priv->handler_mutex);
922
e51060f0 923 if (id_priv->cma_dev) {
3c86aa70 924 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
07ebafba 925 case RDMA_TRANSPORT_IB:
0c9361fc 926 if (id_priv->cm_id.ib)
e51060f0
SH		 927 ib_destroy_cm_id(id_priv->cm_id.ib);
928 break;
07ebafba 929 case RDMA_TRANSPORT_IWARP:
0c9361fc 930 if (id_priv->cm_id.iw)
07ebafba
TT		 931 iw_destroy_cm_id(id_priv->cm_id.iw);
932 break;
e51060f0
SH		 933 default:
934 break;
935 }
c8f6a362 936 cma_leave_mc_groups(id_priv);
a396d43a 937 cma_release_dev(id_priv);
e51060f0
SH		 938 }
939
940 cma_release_port(id_priv);
941 cma_deref_id(id_priv);
942 wait_for_completion(&id_priv->comp);
943
d02d1f53
SH		 944 if (id_priv->internal_id)
945 cma_deref_id(id_priv->id.context);
946
e51060f0
SH		 947 kfree(id_priv->id.route.path_rec);
948 kfree(id_priv);
949}
950EXPORT_SYMBOL(rdma_destroy_id);
951
952static int cma_rep_recv(struct rdma_id_private *id_priv)
953{
954 int ret;
955
5851bb89 956 ret = cma_modify_qp_rtr(id_priv, NULL);
e51060f0
SH		 957 if (ret)
958 goto reject;
959
5851bb89 960 ret = cma_modify_qp_rts(id_priv, NULL);
e51060f0
SH		 961 if (ret)
962 goto reject;
963
964 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
965 if (ret)
966 goto reject;
967
968 return 0;
969reject:
c5483388 970 cma_modify_qp_err(id_priv);
e51060f0
SH		 971 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
972 NULL, 0, NULL, 0);
973 return ret;
974}
975
976static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
977{
978 if (id_priv->id.ps == RDMA_PS_SDP &&
979 sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
980 SDP_MAJ_VERSION)
981 return -EINVAL;
982
983 return 0;
984}
985
a1b1b61f
SH		 986static void cma_set_rep_event_data(struct rdma_cm_event *event,
987 struct ib_cm_rep_event_param *rep_data,
988 void *private_data)
989{
990 event->param.conn.private_data = private_data;
991 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
992 event->param.conn.responder_resources = rep_data->responder_resources;
993 event->param.conn.initiator_depth = rep_data->initiator_depth;
994 event->param.conn.flow_control = rep_data->flow_control;
995 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
996 event->param.conn.srq = rep_data->srq;
997 event->param.conn.qp_num = rep_data->remote_qpn;
998}
999
e51060f0
SH		 1000static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1001{
1002 struct rdma_id_private *id_priv = cm_id->context;
a1b1b61f
SH		 1003 struct rdma_cm_event event;
1004 int ret = 0;
e51060f0 1005
38ca83a5 1006 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
550e5ca7 1007 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
38ca83a5 1008 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
550e5ca7 1009 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
8aa08602 1010 return 0;
e51060f0 1011
a1b1b61f 1012 memset(&event, 0, sizeof event);
e51060f0
SH		 1013 switch (ib_event->event) {
1014 case IB_CM_REQ_ERROR:
1015 case IB_CM_REP_ERROR:
a1b1b61f
SH		 1016 event.event = RDMA_CM_EVENT_UNREACHABLE;
1017 event.status = -ETIMEDOUT;
e51060f0
SH		 1018 break;
1019 case IB_CM_REP_RECEIVED:
a1b1b61f
SH		 1020 event.status = cma_verify_rep(id_priv, ib_event->private_data);
1021 if (event.status)
1022 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
e51060f0 1023 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
a1b1b61f
SH		 1024 event.status = cma_rep_recv(id_priv);
1025 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1026 RDMA_CM_EVENT_ESTABLISHED;
e51060f0 1027 } else
a1b1b61f
SH		 1028 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1029 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1030 ib_event->private_data);
e51060f0
SH		 1031 break;
1032 case IB_CM_RTU_RECEIVED:
0fe313b0
SH		 1033 case IB_CM_USER_ESTABLISHED:
1034 event.event = RDMA_CM_EVENT_ESTABLISHED;
e51060f0
SH		 1035 break;
1036 case IB_CM_DREQ_ERROR:
a1b1b61f 1037 event.status = -ETIMEDOUT; /* fall through */
e51060f0
SH		 1038 case IB_CM_DREQ_RECEIVED:
1039 case IB_CM_DREP_RECEIVED:
550e5ca7
NM		 1040 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1041 RDMA_CM_DISCONNECT))
e51060f0 1042 goto out;
a1b1b61f 1043 event.event = RDMA_CM_EVENT_DISCONNECTED;
e51060f0
SH		 1044 break;
1045 case IB_CM_TIMEWAIT_EXIT:
38ca83a5
AV		 1046 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1047 break;
e51060f0
SH		 1048 case IB_CM_MRA_RECEIVED:
1049 /* ignore event */
1050 goto out;
1051 case IB_CM_REJ_RECEIVED:
c5483388 1052 cma_modify_qp_err(id_priv);
a1b1b61f
SH		 1053 event.status = ib_event->param.rej_rcvd.reason;
1054 event.event = RDMA_CM_EVENT_REJECTED;
1055 event.param.conn.private_data = ib_event->private_data;
1056 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
e51060f0
SH		 1057 break;
1058 default:
468f2239 1059 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
e51060f0
SH		 1060 ib_event->event);
1061 goto out;
1062 }
1063
a1b1b61f 1064 ret = id_priv->id.event_handler(&id_priv->id, &event);
e51060f0
SH		 1065 if (ret) {
1066 /* Destroy the CM ID by returning a non-zero value. */
1067 id_priv->cm_id.ib = NULL;
550e5ca7 1068 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 1069 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1070 rdma_destroy_id(&id_priv->id);
1071 return ret;
1072 }
1073out:
de910bd9 1074 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1075 return ret;
1076}
1077
628e5f6d
SH		 1078static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1079 struct ib_cm_event *ib_event)
e51060f0
SH		 1080{
1081 struct rdma_id_private *id_priv;
1082 struct rdma_cm_id *id;
1083 struct rdma_route *rt;
1084 union cma_ip_addr *src, *dst;
1b90c137 1085 __be16 port;
e51060f0 1086 u8 ip_ver;
64c5e613 1087 int ret;
e51060f0 1088
3f168d2b
KK		 1089 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1090 &ip_ver, &port, &src, &dst))
0c9361fc 1091 return NULL;
3f168d2b 1092
e51060f0 1093 id = rdma_create_id(listen_id->event_handler, listen_id->context,
b26f9b99 1094 listen_id->ps, ib_event->param.req_rcvd.qp_type);
e51060f0 1095 if (IS_ERR(id))
0c9361fc 1096 return NULL;
3f168d2b
KK		 1097
1098 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1099 ip_ver, port, src, dst);
e51060f0
SH		 1100
1101 rt = &id->route;
1102 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
3f168d2b
KK		 1103 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1104 GFP_KERNEL);
e51060f0 1105 if (!rt->path_rec)
0c9361fc 1106 goto err;
e51060f0 1107
e51060f0
SH		 1108 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1109 if (rt->num_paths == 2)
1110 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1111
6f8372b6
SH		 1112 if (cma_any_addr((struct sockaddr *) &rt->addr.src_addr)) {
1113 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1114 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
46ea5061 1115 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
6f8372b6
SH		 1116 } else {
1117 ret = rdma_translate_ip((struct sockaddr *) &rt->addr.src_addr,
1118 &rt->addr.dev_addr);
1119 if (ret)
0c9361fc 1120 goto err;
6f8372b6
SH		 1121 }
1122 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
e51060f0
SH		 1123
1124 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 1125 id_priv->state = RDMA_CM_CONNECT;
e51060f0 1126 return id_priv;
3f168d2b 1127
3f168d2b 1128err:
0c9361fc 1129 rdma_destroy_id(id);
e51060f0
SH		 1130 return NULL;
1131}
1132
628e5f6d
SH		 1133static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1134 struct ib_cm_event *ib_event)
1135{
1136 struct rdma_id_private *id_priv;
1137 struct rdma_cm_id *id;
1138 union cma_ip_addr *src, *dst;
1b90c137 1139 __be16 port;
628e5f6d
SH		 1140 u8 ip_ver;
1141 int ret;
1142
1143 id = rdma_create_id(listen_id->event_handler, listen_id->context,
b26f9b99 1144 listen_id->ps, IB_QPT_UD);
628e5f6d
SH		 1145 if (IS_ERR(id))
1146 return NULL;
1147
1148
1149 if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1150 &ip_ver, &port, &src, &dst))
1151 goto err;
1152
1153 cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1154 ip_ver, port, src, dst);
1155
6f8372b6
SH		 1156 if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1157 ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr,
1158 &id->route.addr.dev_addr);
1159 if (ret)
1160 goto err;
1161 }
628e5f6d
SH		 1162
1163 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 1164 id_priv->state = RDMA_CM_CONNECT;
628e5f6d
SH		 1165 return id_priv;
1166err:
1167 rdma_destroy_id(id);
1168 return NULL;
1169}
1170
a1b1b61f
SH		 1171static void cma_set_req_event_data(struct rdma_cm_event *event,
1172 struct ib_cm_req_event_param *req_data,
1173 void *private_data, int offset)
1174{
1175 event->param.conn.private_data = private_data + offset;
1176 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1177 event->param.conn.responder_resources = req_data->responder_resources;
1178 event->param.conn.initiator_depth = req_data->initiator_depth;
1179 event->param.conn.flow_control = req_data->flow_control;
1180 event->param.conn.retry_count = req_data->retry_count;
1181 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1182 event->param.conn.srq = req_data->srq;
1183 event->param.conn.qp_num = req_data->remote_qpn;
1184}
1185
9595480c
HS		 1186static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1187{
4dd81e89 1188 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
9595480c
HS		 1189 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1190 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1191 (id->qp_type == IB_QPT_UD)) ||
1192 (!id->qp_type));
1193}
1194
e51060f0
SH		 1195static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1196{
1197 struct rdma_id_private *listen_id, *conn_id;
a1b1b61f 1198 struct rdma_cm_event event;
e51060f0
SH		 1199 int offset, ret;
1200
1201 listen_id = cm_id->context;
9595480c
HS		 1202 if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1203 return -EINVAL;
1204
550e5ca7 1205 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
8aa08602 1206 return -ECONNABORTED;
e51060f0 1207
628e5f6d
SH		 1208 memset(&event, 0, sizeof event);
1209 offset = cma_user_data_offset(listen_id->id.ps);
1210 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
9595480c 1211 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
628e5f6d
SH		 1212 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1213 event.param.ud.private_data = ib_event->private_data + offset;
1214 event.param.ud.private_data_len =
1215 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1216 } else {
1217 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1218 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1219 ib_event->private_data, offset);
1220 }
e51060f0
SH		 1221 if (!conn_id) {
1222 ret = -ENOMEM;
b6cec8aa 1223 goto err1;
e51060f0
SH		 1224 }
1225
de910bd9 1226 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
07ebafba 1227 ret = cma_acquire_dev(conn_id);
a1a733f6 1228 if (ret)
b6cec8aa 1229 goto err2;
e51060f0
SH		 1230
1231 conn_id->cm_id.ib = cm_id;
1232 cm_id->context = conn_id;
1233 cm_id->cm_handler = cma_ib_handler;
1234
25ae21a1
SH		 1235 /*
1236 * Protect against the user destroying conn_id from another thread
1237 * until we're done accessing it.
1238 */
1239 atomic_inc(&conn_id->refcount);
a1b1b61f 1240 ret = conn_id->id.event_handler(&conn_id->id, &event);
b6cec8aa
SH		 1241 if (ret)
1242 goto err3;
1243
1244 /*
1245 * Acquire mutex to prevent user executing rdma_destroy_id()
1246 * while we're accessing the cm_id.
1247 */
1248 mutex_lock(&lock);
1249 if (cma_comp(conn_id, RDMA_CM_CONNECT) && (conn_id->id.qp_type != IB_QPT_UD))
1250 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1251 mutex_unlock(&lock);
1252 mutex_unlock(&conn_id->handler_mutex);
1253 mutex_unlock(&listen_id->handler_mutex);
25ae21a1 1254 cma_deref_id(conn_id);
b6cec8aa 1255 return 0;
a1a733f6 1256
b6cec8aa
SH		 1257err3:
1258 cma_deref_id(conn_id);
a1a733f6
KK		 1259 /* Destroy the CM ID by returning a non-zero value. */
1260 conn_id->cm_id.ib = NULL;
b6cec8aa 1261err2:
550e5ca7 1262 cma_exch(conn_id, RDMA_CM_DESTROYING);
de910bd9 1263 mutex_unlock(&conn_id->handler_mutex);
b6cec8aa 1264err1:
de910bd9 1265 mutex_unlock(&listen_id->handler_mutex);
b6cec8aa
SH		 1266 if (conn_id)
1267 rdma_destroy_id(&conn_id->id);
e51060f0
SH		 1268 return ret;
1269}
1270
1271static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1272{
628e5f6d 1273 return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
e51060f0
SH		 1274}
1275
1276static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1277 struct ib_cm_compare_data *compare)
1278{
1279 struct cma_hdr *cma_data, *cma_mask;
1280 struct sdp_hh *sdp_data, *sdp_mask;
1b90c137 1281 __be32 ip4_addr;
e51060f0
SH		 1282 struct in6_addr ip6_addr;
1283
1284 memset(compare, 0, sizeof *compare);
1285 cma_data = (void *) compare->data;
1286 cma_mask = (void *) compare->mask;
1287 sdp_data = (void *) compare->data;
1288 sdp_mask = (void *) compare->mask;
1289
1290 switch (addr->sa_family) {
1291 case AF_INET:
1292 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1293 if (ps == RDMA_PS_SDP) {
1294 sdp_set_ip_ver(sdp_data, 4);
1295 sdp_set_ip_ver(sdp_mask, 0xF);
1296 sdp_data->dst_addr.ip4.addr = ip4_addr;
1b90c137 1297 sdp_mask->dst_addr.ip4.addr = htonl(~0);
e51060f0
SH		 1298 } else {
1299 cma_set_ip_ver(cma_data, 4);
1300 cma_set_ip_ver(cma_mask, 0xF);
1301 cma_data->dst_addr.ip4.addr = ip4_addr;
1b90c137 1302 cma_mask->dst_addr.ip4.addr = htonl(~0);
e51060f0
SH		 1303 }
1304 break;
1305 case AF_INET6:
1306 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1307 if (ps == RDMA_PS_SDP) {
1308 sdp_set_ip_ver(sdp_data, 6);
1309 sdp_set_ip_ver(sdp_mask, 0xF);
1310 sdp_data->dst_addr.ip6 = ip6_addr;
1311 memset(&sdp_mask->dst_addr.ip6, 0xFF,
1312 sizeof sdp_mask->dst_addr.ip6);
1313 } else {
1314 cma_set_ip_ver(cma_data, 6);
1315 cma_set_ip_ver(cma_mask, 0xF);
1316 cma_data->dst_addr.ip6 = ip6_addr;
1317 memset(&cma_mask->dst_addr.ip6, 0xFF,
1318 sizeof cma_mask->dst_addr.ip6);
1319 }
1320 break;
1321 default:
1322 break;
1323 }
1324}
1325
07ebafba
TT		 1326static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1327{
1328 struct rdma_id_private *id_priv = iw_id->context;
a1b1b61f 1329 struct rdma_cm_event event;
07ebafba
TT		 1330 struct sockaddr_in *sin;
1331 int ret = 0;
1332
550e5ca7 1333 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
be65f086 1334 return 0;
07ebafba 1335
be65f086 1336 memset(&event, 0, sizeof event);
07ebafba
TT		 1337 switch (iw_event->event) {
1338 case IW_CM_EVENT_CLOSE:
a1b1b61f 1339 event.event = RDMA_CM_EVENT_DISCONNECTED;
07ebafba
TT		 1340 break;
1341 case IW_CM_EVENT_CONNECT_REPLY:
1342 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1343 *sin = iw_event->local_addr;
1344 sin = (struct sockaddr_in *) &id_priv->id.route.addr.dst_addr;
1345 *sin = iw_event->remote_addr;
881a045f
SW		 1346 switch (iw_event->status) {
1347 case 0:
a1b1b61f 1348 event.event = RDMA_CM_EVENT_ESTABLISHED;
3ebeebc3
KS		 1349 event.param.conn.initiator_depth = iw_event->ird;
1350 event.param.conn.responder_resources = iw_event->ord;
881a045f
SW		 1351 break;
1352 case -ECONNRESET:
1353 case -ECONNREFUSED:
1354 event.event = RDMA_CM_EVENT_REJECTED;
1355 break;
1356 case -ETIMEDOUT:
1357 event.event = RDMA_CM_EVENT_UNREACHABLE;
1358 break;
1359 default:
1360 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1361 break;
1362 }
07ebafba
TT		 1363 break;
1364 case IW_CM_EVENT_ESTABLISHED:
a1b1b61f 1365 event.event = RDMA_CM_EVENT_ESTABLISHED;
3ebeebc3
KS		 1366 event.param.conn.initiator_depth = iw_event->ird;
1367 event.param.conn.responder_resources = iw_event->ord;
07ebafba
TT		 1368 break;
1369 default:
1370 BUG_ON(1);
1371 }
1372
a1b1b61f
SH		 1373 event.status = iw_event->status;
1374 event.param.conn.private_data = iw_event->private_data;
1375 event.param.conn.private_data_len = iw_event->private_data_len;
1376 ret = id_priv->id.event_handler(&id_priv->id, &event);
07ebafba
TT		 1377 if (ret) {
1378 /* Destroy the CM ID by returning a non-zero value. */
1379 id_priv->cm_id.iw = NULL;
550e5ca7 1380 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 1381 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 1382 rdma_destroy_id(&id_priv->id);
1383 return ret;
1384 }
1385
de910bd9 1386 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 1387 return ret;
1388}
1389
1390static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1391 struct iw_cm_event *iw_event)
1392{
1393 struct rdma_cm_id *new_cm_id;
1394 struct rdma_id_private *listen_id, *conn_id;
1395 struct sockaddr_in *sin;
1396 struct net_device *dev = NULL;
a1b1b61f 1397 struct rdma_cm_event event;
07ebafba 1398 int ret;
8d8293cf 1399 struct ib_device_attr attr;
07ebafba
TT		 1400
1401 listen_id = cm_id->context;
550e5ca7 1402 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
8aa08602 1403 return -ECONNABORTED;
07ebafba
TT		 1404
1405 /* Create a new RDMA id for the new IW CM ID */
1406 new_cm_id = rdma_create_id(listen_id->id.event_handler,
1407 listen_id->id.context,
b26f9b99 1408 RDMA_PS_TCP, IB_QPT_RC);
10f32065 1409 if (IS_ERR(new_cm_id)) {
07ebafba
TT		 1410 ret = -ENOMEM;
1411 goto out;
1412 }
1413 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
de910bd9 1414 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
550e5ca7 1415 conn_id->state = RDMA_CM_CONNECT;
07ebafba 1416
1ab35276 1417 dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
07ebafba
TT		 1418 if (!dev) {
1419 ret = -EADDRNOTAVAIL;
de910bd9 1420 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1421 rdma_destroy_id(new_cm_id);
1422 goto out;
1423 }
1424 ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1425 if (ret) {
de910bd9 1426 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1427 rdma_destroy_id(new_cm_id);
1428 goto out;
1429 }
1430
1431 ret = cma_acquire_dev(conn_id);
1432 if (ret) {
de910bd9 1433 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 1434 rdma_destroy_id(new_cm_id);
1435 goto out;
1436 }
1437
1438 conn_id->cm_id.iw = cm_id;
1439 cm_id->context = conn_id;
1440 cm_id->cm_handler = cma_iw_handler;
1441
1442 sin = (struct sockaddr_in *) &new_cm_id->route.addr.src_addr;
1443 *sin = iw_event->local_addr;
1444 sin = (struct sockaddr_in *) &new_cm_id->route.addr.dst_addr;
1445 *sin = iw_event->remote_addr;
1446
8d8293cf
SW		 1447 ret = ib_query_device(conn_id->id.device, &attr);
1448 if (ret) {
de910bd9 1449 mutex_unlock(&conn_id->handler_mutex);
8d8293cf
SW		 1450 rdma_destroy_id(new_cm_id);
1451 goto out;
1452 }
1453
a1b1b61f
SH		 1454 memset(&event, 0, sizeof event);
1455 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1456 event.param.conn.private_data = iw_event->private_data;
1457 event.param.conn.private_data_len = iw_event->private_data_len;
3ebeebc3
KS		 1458 event.param.conn.initiator_depth = iw_event->ird;
1459 event.param.conn.responder_resources = iw_event->ord;
25ae21a1
SH		 1460
1461 /*
1462 * Protect against the user destroying conn_id from another thread
1463 * until we're done accessing it.
1464 */
1465 atomic_inc(&conn_id->refcount);
a1b1b61f 1466 ret = conn_id->id.event_handler(&conn_id->id, &event);
07ebafba
TT		 1467 if (ret) {
1468 /* User wants to destroy the CM ID */
1469 conn_id->cm_id.iw = NULL;
550e5ca7 1470 cma_exch(conn_id, RDMA_CM_DESTROYING);
de910bd9 1471 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1472 cma_deref_id(conn_id);
07ebafba 1473 rdma_destroy_id(&conn_id->id);
de910bd9 1474 goto out;
07ebafba
TT		 1475 }
1476
de910bd9 1477 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 1478 cma_deref_id(conn_id);
de910bd9 1479
07ebafba
TT		 1480out:
1481 if (dev)
1482 dev_put(dev);
de910bd9 1483 mutex_unlock(&listen_id->handler_mutex);
07ebafba
TT		 1484 return ret;
1485}
1486
e51060f0
SH		 1487static int cma_ib_listen(struct rdma_id_private *id_priv)
1488{
1489 struct ib_cm_compare_data compare_data;
1490 struct sockaddr *addr;
0c9361fc 1491 struct ib_cm_id *id;
e51060f0
SH		 1492 __be64 svc_id;
1493 int ret;
1494
0c9361fc
JM		 1495 id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1496 if (IS_ERR(id))
1497 return PTR_ERR(id);
1498
1499 id_priv->cm_id.ib = id;
e51060f0 1500
3f446754 1501 addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
e51060f0
SH		 1502 svc_id = cma_get_service_id(id_priv->id.ps, addr);
1503 if (cma_any_addr(addr))
1504 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1505 else {
1506 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1507 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1508 }
1509
1510 if (ret) {
1511 ib_destroy_cm_id(id_priv->cm_id.ib);
1512 id_priv->cm_id.ib = NULL;
1513 }
1514
1515 return ret;
1516}
1517
07ebafba
TT		 1518static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1519{
1520 int ret;
1521 struct sockaddr_in *sin;
0c9361fc
JM		 1522 struct iw_cm_id *id;
1523
1524 id = iw_create_cm_id(id_priv->id.device,
1525 iw_conn_req_handler,
1526 id_priv);
1527 if (IS_ERR(id))
1528 return PTR_ERR(id);
07ebafba 1529
0c9361fc 1530 id_priv->cm_id.iw = id;
07ebafba
TT		 1531
1532 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
1533 id_priv->cm_id.iw->local_addr = *sin;
1534
1535 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1536
1537 if (ret) {
1538 iw_destroy_cm_id(id_priv->cm_id.iw);
1539 id_priv->cm_id.iw = NULL;
1540 }
1541
1542 return ret;
1543}
1544
e51060f0
SH		 1545static int cma_listen_handler(struct rdma_cm_id *id,
1546 struct rdma_cm_event *event)
1547{
1548 struct rdma_id_private *id_priv = id->context;
1549
1550 id->context = id_priv->id.context;
1551 id->event_handler = id_priv->id.event_handler;
1552 return id_priv->id.event_handler(id, event);
1553}
1554
1555static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1556 struct cma_device *cma_dev)
1557{
1558 struct rdma_id_private *dev_id_priv;
1559 struct rdma_cm_id *id;
1560 int ret;
1561
b26f9b99
SH		 1562 id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1563 id_priv->id.qp_type);
e51060f0
SH		 1564 if (IS_ERR(id))
1565 return;
1566
1567 dev_id_priv = container_of(id, struct rdma_id_private, id);
1568
550e5ca7 1569 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
e51060f0 1570 memcpy(&id->route.addr.src_addr, &id_priv->id.route.addr.src_addr,
3f446754 1571 ip_addr_size((struct sockaddr *) &id_priv->id.route.addr.src_addr));
e51060f0
SH		 1572
1573 cma_attach_to_dev(dev_id_priv, cma_dev);
1574 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
d02d1f53
SH		 1575 atomic_inc(&id_priv->refcount);
1576 dev_id_priv->internal_id = 1;
5b0ec991 1577 dev_id_priv->afonly = id_priv->afonly;
e51060f0
SH		 1578
1579 ret = rdma_listen(id, id_priv->backlog);
1580 if (ret)
d02d1f53 1581 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
468f2239 1582 "listening on device %s\n", ret, cma_dev->device->name);
e51060f0
SH		 1583}
1584
1585static void cma_listen_on_all(struct rdma_id_private *id_priv)
1586{
1587 struct cma_device *cma_dev;
1588
1589 mutex_lock(&lock);
1590 list_add_tail(&id_priv->list, &listen_any_list);
1591 list_for_each_entry(cma_dev, &dev_list, list)
1592 cma_listen_on_dev(id_priv, cma_dev);
1593 mutex_unlock(&lock);
1594}
1595
a81c994d
SH		 1596void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1597{
1598 struct rdma_id_private *id_priv;
1599
1600 id_priv = container_of(id, struct rdma_id_private, id);
1601 id_priv->tos = (u8) tos;
1602}
1603EXPORT_SYMBOL(rdma_set_service_type);
1604
e51060f0
SH		 1605static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1606 void *context)
1607{
1608 struct cma_work *work = context;
1609 struct rdma_route *route;
1610
1611 route = &work->id->id.route;
1612
1613 if (!status) {
1614 route->num_paths = 1;
1615 *route->path_rec = *path_rec;
1616 } else {
550e5ca7
NM		 1617 work->old_state = RDMA_CM_ROUTE_QUERY;
1618 work->new_state = RDMA_CM_ADDR_RESOLVED;
e51060f0 1619 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
8f0472d3 1620 work->event.status = status;
e51060f0
SH		 1621 }
1622
1623 queue_work(cma_wq, &work->work);
1624}
1625
1626static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1627 struct cma_work *work)
1628{
a81c994d 1629 struct rdma_addr *addr = &id_priv->id.route.addr;
e51060f0 1630 struct ib_sa_path_rec path_rec;
a81c994d
SH		 1631 ib_sa_comp_mask comp_mask;
1632 struct sockaddr_in6 *sin6;
e51060f0
SH		 1633
1634 memset(&path_rec, 0, sizeof path_rec);
6f8372b6
SH		 1635 rdma_addr_get_sgid(&addr->dev_addr, &path_rec.sgid);
1636 rdma_addr_get_dgid(&addr->dev_addr, &path_rec.dgid);
a81c994d 1637 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr));
e51060f0 1638 path_rec.numb_path = 1;
962063e6 1639 path_rec.reversible = 1;
3f446754
RD		 1640 path_rec.service_id = cma_get_service_id(id_priv->id.ps,
1641 (struct sockaddr *) &addr->dst_addr);
a81c994d
SH		 1642
1643 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1644 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1645 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1646
3f446754 1647 if (addr->src_addr.ss_family == AF_INET) {
a81c994d
SH		 1648 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1649 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1650 } else {
1651 sin6 = (struct sockaddr_in6 *) &addr->src_addr;
1652 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1653 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1654 }
e51060f0 1655
c1a0b23b 1656 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
a81c994d
SH		 1657 id_priv->id.port_num, &path_rec,
1658 comp_mask, timeout_ms,
1659 GFP_KERNEL, cma_query_handler,
1660 work, &id_priv->query);
e51060f0
SH		 1661
1662 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1663}
1664
c4028958 1665static void cma_work_handler(struct work_struct *_work)
e51060f0 1666{
c4028958 1667 struct cma_work *work = container_of(_work, struct cma_work, work);
e51060f0
SH		 1668 struct rdma_id_private *id_priv = work->id;
1669 int destroy = 0;
1670
de910bd9 1671 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 1672 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1673 goto out;
1674
1675 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
550e5ca7 1676 cma_exch(id_priv, RDMA_CM_DESTROYING);
e51060f0
SH		 1677 destroy = 1;
1678 }
1679out:
de910bd9 1680 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1681 cma_deref_id(id_priv);
1682 if (destroy)
1683 rdma_destroy_id(&id_priv->id);
1684 kfree(work);
1685}
1686
dd5bdff8
OG		 1687static void cma_ndev_work_handler(struct work_struct *_work)
1688{
1689 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1690 struct rdma_id_private *id_priv = work->id;
1691 int destroy = 0;
1692
1693 mutex_lock(&id_priv->handler_mutex);
550e5ca7
NM		 1694 if (id_priv->state == RDMA_CM_DESTROYING ||
1695 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
dd5bdff8
OG		 1696 goto out;
1697
1698 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
550e5ca7 1699 cma_exch(id_priv, RDMA_CM_DESTROYING);
dd5bdff8
OG		 1700 destroy = 1;
1701 }
1702
1703out:
1704 mutex_unlock(&id_priv->handler_mutex);
1705 cma_deref_id(id_priv);
1706 if (destroy)
1707 rdma_destroy_id(&id_priv->id);
1708 kfree(work);
1709}
1710
e51060f0
SH		 1711static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1712{
1713 struct rdma_route *route = &id_priv->id.route;
1714 struct cma_work *work;
1715 int ret;
1716
1717 work = kzalloc(sizeof *work, GFP_KERNEL);
1718 if (!work)
1719 return -ENOMEM;
1720
1721 work->id = id_priv;
c4028958 1722 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 1723 work->old_state = RDMA_CM_ROUTE_QUERY;
1724 work->new_state = RDMA_CM_ROUTE_RESOLVED;
e51060f0
SH		 1725 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1726
1727 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1728 if (!route->path_rec) {
1729 ret = -ENOMEM;
1730 goto err1;
1731 }
1732
1733 ret = cma_query_ib_route(id_priv, timeout_ms, work);
1734 if (ret)
1735 goto err2;
1736
1737 return 0;
1738err2:
1739 kfree(route->path_rec);
1740 route->path_rec = NULL;
1741err1:
1742 kfree(work);
1743 return ret;
1744}
1745
1746int rdma_set_ib_paths(struct rdma_cm_id *id,
1747 struct ib_sa_path_rec *path_rec, int num_paths)
1748{
1749 struct rdma_id_private *id_priv;
1750 int ret;
1751
1752 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7
NM		 1753 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1754 RDMA_CM_ROUTE_RESOLVED))
e51060f0
SH		 1755 return -EINVAL;
1756
9893e742
JL		 1757 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1758 GFP_KERNEL);
e51060f0
SH		 1759 if (!id->route.path_rec) {
1760 ret = -ENOMEM;
1761 goto err;
1762 }
1763
ae2d9293 1764 id->route.num_paths = num_paths;
e51060f0
SH		 1765 return 0;
1766err:
550e5ca7 1767 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
e51060f0
SH		 1768 return ret;
1769}
1770EXPORT_SYMBOL(rdma_set_ib_paths);
1771
07ebafba
TT		 1772static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1773{
1774 struct cma_work *work;
1775
1776 work = kzalloc(sizeof *work, GFP_KERNEL);
1777 if (!work)
1778 return -ENOMEM;
1779
1780 work->id = id_priv;
c4028958 1781 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 1782 work->old_state = RDMA_CM_ROUTE_QUERY;
1783 work->new_state = RDMA_CM_ROUTE_RESOLVED;
07ebafba
TT		 1784 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1785 queue_work(cma_wq, &work->work);
1786 return 0;
1787}
1788
3c86aa70
EC		 1789static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1790{
1791 struct rdma_route *route = &id_priv->id.route;
1792 struct rdma_addr *addr = &route->addr;
1793 struct cma_work *work;
1794 int ret;
1795 struct sockaddr_in *src_addr = (struct sockaddr_in *)&route->addr.src_addr;
1796 struct sockaddr_in *dst_addr = (struct sockaddr_in *)&route->addr.dst_addr;
1797 struct net_device *ndev = NULL;
af7bd463 1798 u16 vid;
3c86aa70
EC		 1799
1800 if (src_addr->sin_family != dst_addr->sin_family)
1801 return -EINVAL;
1802
1803 work = kzalloc(sizeof *work, GFP_KERNEL);
1804 if (!work)
1805 return -ENOMEM;
1806
1807 work->id = id_priv;
1808 INIT_WORK(&work->work, cma_work_handler);
1809
1810 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1811 if (!route->path_rec) {
1812 ret = -ENOMEM;
1813 goto err1;
1814 }
1815
1816 route->num_paths = 1;
1817
3c86aa70
EC		 1818 if (addr->dev_addr.bound_dev_if)
1819 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1820 if (!ndev) {
1821 ret = -ENODEV;
1822 goto err2;
1823 }
1824
af7bd463
EC		 1825 vid = rdma_vlan_dev_vlan_id(ndev);
1826
1827 iboe_mac_vlan_to_ll(&route->path_rec->sgid, addr->dev_addr.src_dev_addr, vid);
1828 iboe_mac_vlan_to_ll(&route->path_rec->dgid, addr->dev_addr.dst_dev_addr, vid);
1829
1830 route->path_rec->hop_limit = 1;
1831 route->path_rec->reversible = 1;
1832 route->path_rec->pkey = cpu_to_be16(0xffff);
1833 route->path_rec->mtu_selector = IB_SA_EQ;
366cddb4
AV		 1834 route->path_rec->sl = netdev_get_prio_tc_map(
1835 ndev->priv_flags & IFF_802_1Q_VLAN ?
1836 vlan_dev_real_dev(ndev) : ndev,
1837 rt_tos2priority(id_priv->tos));
af7bd463 1838
3c86aa70
EC		 1839 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1840 route->path_rec->rate_selector = IB_SA_EQ;
1841 route->path_rec->rate = iboe_get_rate(ndev);
1842 dev_put(ndev);
1843 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1844 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1845 if (!route->path_rec->mtu) {
1846 ret = -EINVAL;
1847 goto err2;
1848 }
1849
550e5ca7
NM		 1850 work->old_state = RDMA_CM_ROUTE_QUERY;
1851 work->new_state = RDMA_CM_ROUTE_RESOLVED;
3c86aa70
EC		 1852 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1853 work->event.status = 0;
1854
1855 queue_work(cma_wq, &work->work);
1856
1857 return 0;
1858
1859err2:
1860 kfree(route->path_rec);
1861 route->path_rec = NULL;
1862err1:
1863 kfree(work);
1864 return ret;
1865}
1866
e51060f0
SH		 1867int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1868{
1869 struct rdma_id_private *id_priv;
1870 int ret;
1871
1872 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 1873 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
e51060f0
SH		 1874 return -EINVAL;
1875
1876 atomic_inc(&id_priv->refcount);
07ebafba
TT		 1877 switch (rdma_node_get_transport(id->device->node_type)) {
1878 case RDMA_TRANSPORT_IB:
3c86aa70
EC		 1879 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1880 case IB_LINK_LAYER_INFINIBAND:
1881 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1882 break;
1883 case IB_LINK_LAYER_ETHERNET:
1884 ret = cma_resolve_iboe_route(id_priv);
1885 break;
1886 default:
1887 ret = -ENOSYS;
1888 }
e51060f0 1889 break;
07ebafba
TT		 1890 case RDMA_TRANSPORT_IWARP:
1891 ret = cma_resolve_iw_route(id_priv, timeout_ms);
1892 break;
e51060f0
SH		 1893 default:
1894 ret = -ENOSYS;
1895 break;
1896 }
1897 if (ret)
1898 goto err;
1899
1900 return 0;
1901err:
550e5ca7 1902 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
e51060f0
SH		 1903 cma_deref_id(id_priv);
1904 return ret;
1905}
1906EXPORT_SYMBOL(rdma_resolve_route);
1907
1908static int cma_bind_loopback(struct rdma_id_private *id_priv)
1909{
1910 struct cma_device *cma_dev;
1911 struct ib_port_attr port_attr;
f0ee3404 1912 union ib_gid gid;
e51060f0
SH		 1913 u16 pkey;
1914 int ret;
1915 u8 p;
1916
1917 mutex_lock(&lock);
e82153b5
KK		 1918 if (list_empty(&dev_list)) {
1919 ret = -ENODEV;
1920 goto out;
1921 }
e51060f0
SH		 1922 list_for_each_entry(cma_dev, &dev_list, list)
1923 for (p = 1; p <= cma_dev->device->phys_port_cnt; ++p)
e82153b5 1924 if (!ib_query_port(cma_dev->device, p, &port_attr) &&
e51060f0
SH		 1925 port_attr.state == IB_PORT_ACTIVE)
1926 goto port_found;
1927
e82153b5
KK		 1928 p = 1;
1929 cma_dev = list_entry(dev_list.next, struct cma_device, list);
e51060f0
SH		 1930
1931port_found:
f0ee3404 1932 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
e51060f0
SH		 1933 if (ret)
1934 goto out;
1935
1936 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
1937 if (ret)
1938 goto out;
1939
6f8372b6 1940 id_priv->id.route.addr.dev_addr.dev_type =
3c86aa70 1941 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
6f8372b6
SH		 1942 ARPHRD_INFINIBAND : ARPHRD_ETHER;
1943
1944 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0
SH		 1945 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1946 id_priv->id.port_num = p;
1947 cma_attach_to_dev(id_priv, cma_dev);
1948out:
1949 mutex_unlock(&lock);
1950 return ret;
1951}
1952
1953static void addr_handler(int status, struct sockaddr *src_addr,
1954 struct rdma_dev_addr *dev_addr, void *context)
1955{
1956 struct rdma_id_private *id_priv = context;
a1b1b61f 1957 struct rdma_cm_event event;
e51060f0 1958
a1b1b61f 1959 memset(&event, 0, sizeof event);
de910bd9 1960 mutex_lock(&id_priv->handler_mutex);
550e5ca7
NM		 1961 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
1962 RDMA_CM_ADDR_RESOLVED))
61a73c70 1963 goto out;
61a73c70
SH		 1964
1965 if (!status && !id_priv->cma_dev)
e51060f0
SH		 1966 status = cma_acquire_dev(id_priv);
1967
1968 if (status) {
550e5ca7
NM		 1969 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1970 RDMA_CM_ADDR_BOUND))
e51060f0 1971 goto out;
a1b1b61f
SH		 1972 event.event = RDMA_CM_EVENT_ADDR_ERROR;
1973 event.status = status;
e51060f0 1974 } else {
e51060f0
SH		 1975 memcpy(&id_priv->id.route.addr.src_addr, src_addr,
1976 ip_addr_size(src_addr));
a1b1b61f 1977 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
e51060f0
SH		 1978 }
1979
a1b1b61f 1980 if (id_priv->id.event_handler(&id_priv->id, &event)) {
550e5ca7 1981 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 1982 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1983 cma_deref_id(id_priv);
1984 rdma_destroy_id(&id_priv->id);
1985 return;
1986 }
1987out:
de910bd9 1988 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1989 cma_deref_id(id_priv);
1990}
1991
1992static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1993{
1994 struct cma_work *work;
6f8372b6 1995 struct sockaddr *src, *dst;
f0ee3404 1996 union ib_gid gid;
e51060f0
SH		 1997 int ret;
1998
1999 work = kzalloc(sizeof *work, GFP_KERNEL);
2000 if (!work)
2001 return -ENOMEM;
2002
2003 if (!id_priv->cma_dev) {
2004 ret = cma_bind_loopback(id_priv);
2005 if (ret)
2006 goto err;
2007 }
2008
6f8372b6
SH		 2009 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2010 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0 2011
6f8372b6
SH		 2012 src = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
2013 if (cma_zero_addr(src)) {
2014 dst = (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
2015 if ((src->sa_family = dst->sa_family) == AF_INET) {
4e3fd7a0
AD		 2016 ((struct sockaddr_in *)src)->sin_addr =
2017 ((struct sockaddr_in *)dst)->sin_addr;
6f8372b6 2018 } else {
4e3fd7a0
AD		 2019 ((struct sockaddr_in6 *)src)->sin6_addr =
2020 ((struct sockaddr_in6 *)dst)->sin6_addr;
6f8372b6 2021 }
e51060f0
SH		 2022 }
2023
2024 work->id = id_priv;
c4028958 2025 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 2026 work->old_state = RDMA_CM_ADDR_QUERY;
2027 work->new_state = RDMA_CM_ADDR_RESOLVED;
e51060f0
SH		 2028 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2029 queue_work(cma_wq, &work->work);
2030 return 0;
2031err:
2032 kfree(work);
2033 return ret;
2034}
2035
2036static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2037 struct sockaddr *dst_addr)
2038{
d14714df
SH		 2039 if (!src_addr || !src_addr->sa_family) {
2040 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2041 if ((src_addr->sa_family = dst_addr->sa_family) == AF_INET6) {
2042 ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2043 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2044 }
2045 }
2046 return rdma_bind_addr(id, src_addr);
e51060f0
SH		 2047}
2048
2049int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2050 struct sockaddr *dst_addr, int timeout_ms)
2051{
2052 struct rdma_id_private *id_priv;
2053 int ret;
2054
2055 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2056 if (id_priv->state == RDMA_CM_IDLE) {
e51060f0
SH		 2057 ret = cma_bind_addr(id, src_addr, dst_addr);
2058 if (ret)
2059 return ret;
2060 }
2061
550e5ca7 2062 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
e51060f0
SH		 2063 return -EINVAL;
2064
2065 atomic_inc(&id_priv->refcount);
2066 memcpy(&id->route.addr.dst_addr, dst_addr, ip_addr_size(dst_addr));
2067 if (cma_any_addr(dst_addr))
2068 ret = cma_resolve_loopback(id_priv);
2069 else
3f446754 2070 ret = rdma_resolve_ip(&addr_client, (struct sockaddr *) &id->route.addr.src_addr,
7a118df3 2071 dst_addr, &id->route.addr.dev_addr,
e51060f0
SH		 2072 timeout_ms, addr_handler, id_priv);
2073 if (ret)
2074 goto err;
2075
2076 return 0;
2077err:
550e5ca7 2078 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
e51060f0
SH		 2079 cma_deref_id(id_priv);
2080 return ret;
2081}
2082EXPORT_SYMBOL(rdma_resolve_addr);
2083
a9bb7912
HS		 2084int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2085{
2086 struct rdma_id_private *id_priv;
2087 unsigned long flags;
2088 int ret;
2089
2090 id_priv = container_of(id, struct rdma_id_private, id);
2091 spin_lock_irqsave(&id_priv->lock, flags);
550e5ca7 2092 if (id_priv->state == RDMA_CM_IDLE) {
a9bb7912
HS		 2093 id_priv->reuseaddr = reuse;
2094 ret = 0;
2095 } else {
2096 ret = -EINVAL;
2097 }
2098 spin_unlock_irqrestore(&id_priv->lock, flags);
2099 return ret;
2100}
2101EXPORT_SYMBOL(rdma_set_reuseaddr);
2102
e51060f0
SH		 2103static void cma_bind_port(struct rdma_bind_list *bind_list,
2104 struct rdma_id_private *id_priv)
2105{
2106 struct sockaddr_in *sin;
2107
2108 sin = (struct sockaddr_in *) &id_priv->id.route.addr.src_addr;
2109 sin->sin_port = htons(bind_list->port);
2110 id_priv->bind_list = bind_list;
2111 hlist_add_head(&id_priv->node, &bind_list->owners);
2112}
2113
2114static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2115 unsigned short snum)
2116{
2117 struct rdma_bind_list *bind_list;
aedec080 2118 int port, ret;
e51060f0 2119
cb164b8c 2120 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
e51060f0
SH		 2121 if (!bind_list)
2122 return -ENOMEM;
2123
aedec080
SH		 2124 do {
2125 ret = idr_get_new_above(ps, bind_list, snum, &port);
2126 } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
2127
2128 if (ret)
2129 goto err1;
2130
2131 if (port != snum) {
2132 ret = -EADDRNOTAVAIL;
2133 goto err2;
2134 }
2135
2136 bind_list->ps = ps;
2137 bind_list->port = (unsigned short) port;
2138 cma_bind_port(bind_list, id_priv);
2139 return 0;
2140err2:
2141 idr_remove(ps, port);
2142err1:
2143 kfree(bind_list);
2144 return ret;
2145}
e51060f0 2146
aedec080
SH		 2147static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2148{
5d7220e8
TH		 2149 static unsigned int last_used_port;
2150 int low, high, remaining;
2151 unsigned int rover;
e51060f0 2152
227b60f5 2153 inet_get_local_port_range(&low, &high);
5d7220e8
TH		 2154 remaining = (high - low) + 1;
2155 rover = net_random() % remaining + low;
2156retry:
2157 if (last_used_port != rover &&
2158 !idr_find(ps, (unsigned short) rover)) {
2159 int ret = cma_alloc_port(ps, id_priv, rover);
2160 /*
2161 * Remember previously used port number in order to avoid
2162 * re-using same port immediately after it is closed.
2163 */
2164 if (!ret)
2165 last_used_port = rover;
2166 if (ret != -EADDRNOTAVAIL)
2167 return ret;
e51060f0 2168 }
5d7220e8
TH		 2169 if (--remaining) {
2170 rover++;
2171 if ((rover < low) || (rover > high))
2172 rover = low;
2173 goto retry;
2174 }
2175 return -EADDRNOTAVAIL;
e51060f0
SH		 2176}
2177
a9bb7912
HS		 2178/*
2179 * Check that the requested port is available. This is called when trying to
2180 * bind to a specific port, or when trying to listen on a bound port. In
2181 * the latter case, the provided id_priv may already be on the bind_list, but
2182 * we still need to check that it's okay to start listening.
2183 */
2184static int cma_check_port(struct rdma_bind_list *bind_list,
2185 struct rdma_id_private *id_priv, uint8_t reuseaddr)
e51060f0
SH		 2186{
2187 struct rdma_id_private *cur_id;
43b752da 2188 struct sockaddr *addr, *cur_addr;
e51060f0 2189 struct hlist_node *node;
e51060f0 2190
43b752da 2191 addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
e51060f0 2192 hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
a9bb7912
HS		 2193 if (id_priv == cur_id)
2194 continue;
3cd96564 2195
5b0ec991
SH		 2196 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2197 cur_id->reuseaddr)
2198 continue;
e51060f0 2199
5b0ec991
SH		 2200 cur_addr = (struct sockaddr *) &cur_id->id.route.addr.src_addr;
2201 if (id_priv->afonly && cur_id->afonly &&
2202 (addr->sa_family != cur_addr->sa_family))
2203 continue;
2204
2205 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2206 return -EADDRNOTAVAIL;
2207
2208 if (!cma_addr_cmp(addr, cur_addr))
2209 return -EADDRINUSE;
a9bb7912 2210 }
e51060f0
SH		 2211 return 0;
2212}
2213
a9bb7912
HS		 2214static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2215{
2216 struct rdma_bind_list *bind_list;
2217 unsigned short snum;
2218 int ret;
2219
2220 snum = ntohs(cma_port((struct sockaddr *) &id_priv->id.route.addr.src_addr));
2221 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2222 return -EACCES;
2223
2224 bind_list = idr_find(ps, snum);
2225 if (!bind_list) {
2226 ret = cma_alloc_port(ps, id_priv, snum);
2227 } else {
2228 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2229 if (!ret)
2230 cma_bind_port(bind_list, id_priv);
2231 }
2232 return ret;
2233}
2234
2235static int cma_bind_listen(struct rdma_id_private *id_priv)
2236{
2237 struct rdma_bind_list *bind_list = id_priv->bind_list;
2238 int ret = 0;
2239
2240 mutex_lock(&lock);
2241 if (bind_list->owners.first->next)
2242 ret = cma_check_port(bind_list, id_priv, 0);
2243 mutex_unlock(&lock);
2244 return ret;
2245}
2246
e51060f0
SH		 2247static int cma_get_port(struct rdma_id_private *id_priv)
2248{
2249 struct idr *ps;
2250 int ret;
2251
2252 switch (id_priv->id.ps) {
2253 case RDMA_PS_SDP:
2254 ps = &sdp_ps;
2255 break;
2256 case RDMA_PS_TCP:
2257 ps = &tcp_ps;
2258 break;
628e5f6d
SH		 2259 case RDMA_PS_UDP:
2260 ps = &udp_ps;
2261 break;
c8f6a362
SH		 2262 case RDMA_PS_IPOIB:
2263 ps = &ipoib_ps;
2264 break;
2d2e9415
SH		 2265 case RDMA_PS_IB:
2266 ps = &ib_ps;
2267 break;
e51060f0
SH		 2268 default:
2269 return -EPROTONOSUPPORT;
2270 }
2271
2272 mutex_lock(&lock);
3f446754 2273 if (cma_any_port((struct sockaddr *) &id_priv->id.route.addr.src_addr))
aedec080 2274 ret = cma_alloc_any_port(ps, id_priv);
e51060f0
SH		 2275 else
2276 ret = cma_use_port(ps, id_priv);
2277 mutex_unlock(&lock);
2278
2279 return ret;
2280}
2281
d14714df
SH		 2282static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2283 struct sockaddr *addr)
2284{
fd4582a3 2285#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
d14714df
SH		 2286 struct sockaddr_in6 *sin6;
2287
2288 if (addr->sa_family != AF_INET6)
2289 return 0;
2290
2291 sin6 = (struct sockaddr_in6 *) addr;
2292 if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) &&
2293 !sin6->sin6_scope_id)
2294 return -EINVAL;
2295
2296 dev_addr->bound_dev_if = sin6->sin6_scope_id;
2297#endif
2298 return 0;
2299}
2300
a9bb7912
HS		 2301int rdma_listen(struct rdma_cm_id *id, int backlog)
2302{
2303 struct rdma_id_private *id_priv;
2304 int ret;
2305
2306 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2307 if (id_priv->state == RDMA_CM_IDLE) {
a9bb7912
HS		 2308 ((struct sockaddr *) &id->route.addr.src_addr)->sa_family = AF_INET;
2309 ret = rdma_bind_addr(id, (struct sockaddr *) &id->route.addr.src_addr);
2310 if (ret)
2311 return ret;
2312 }
2313
550e5ca7 2314 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
a9bb7912
HS		 2315 return -EINVAL;
2316
2317 if (id_priv->reuseaddr) {
2318 ret = cma_bind_listen(id_priv);
2319 if (ret)
2320 goto err;
2321 }
2322
2323 id_priv->backlog = backlog;
2324 if (id->device) {
2325 switch (rdma_node_get_transport(id->device->node_type)) {
2326 case RDMA_TRANSPORT_IB:
2327 ret = cma_ib_listen(id_priv);
2328 if (ret)
2329 goto err;
2330 break;
2331 case RDMA_TRANSPORT_IWARP:
2332 ret = cma_iw_listen(id_priv, backlog);
2333 if (ret)
2334 goto err;
2335 break;
2336 default:
2337 ret = -ENOSYS;
2338 goto err;
2339 }
2340 } else
2341 cma_listen_on_all(id_priv);
2342
2343 return 0;
2344err:
2345 id_priv->backlog = 0;
550e5ca7 2346 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
a9bb7912
HS		 2347 return ret;
2348}
2349EXPORT_SYMBOL(rdma_listen);
2350
e51060f0
SH		 2351int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2352{
2353 struct rdma_id_private *id_priv;
2354 int ret;
2355
1f5175ad 2356 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
e51060f0
SH		 2357 return -EAFNOSUPPORT;
2358
2359 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2360 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
e51060f0
SH		 2361 return -EINVAL;
2362
d14714df
SH		 2363 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2364 if (ret)
2365 goto err1;
2366
8523c048 2367 if (!cma_any_addr(addr)) {
e51060f0 2368 ret = rdma_translate_ip(addr, &id->route.addr.dev_addr);
e51060f0 2369 if (ret)
255d0c14
KK		 2370 goto err1;
2371
255d0c14 2372 ret = cma_acquire_dev(id_priv);
255d0c14
KK		 2373 if (ret)
2374 goto err1;
e51060f0
SH		 2375 }
2376
2377 memcpy(&id->route.addr.src_addr, addr, ip_addr_size(addr));
5b0ec991
SH		 2378 if (addr->sa_family == AF_INET)
2379 id_priv->afonly = 1;
2380#if IS_ENABLED(CONFIG_IPV6)
2381 else if (addr->sa_family == AF_INET6)
2382 id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2383#endif
e51060f0
SH		 2384 ret = cma_get_port(id_priv);
2385 if (ret)
255d0c14 2386 goto err2;
e51060f0
SH		 2387
2388 return 0;
255d0c14 2389err2:
a396d43a
SH		 2390 if (id_priv->cma_dev)
2391 cma_release_dev(id_priv);
255d0c14 2392err1:
550e5ca7 2393 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
e51060f0
SH		 2394 return ret;
2395}
2396EXPORT_SYMBOL(rdma_bind_addr);
2397
2398static int cma_format_hdr(void *hdr, enum rdma_port_space ps,
2399 struct rdma_route *route)
2400{
e51060f0
SH		 2401 struct cma_hdr *cma_hdr;
2402 struct sdp_hh *sdp_hdr;
2403
1f5175ad
AS		 2404 if (route->addr.src_addr.ss_family == AF_INET) {
2405 struct sockaddr_in *src4, *dst4;
2406
2407 src4 = (struct sockaddr_in *) &route->addr.src_addr;
2408 dst4 = (struct sockaddr_in *) &route->addr.dst_addr;
2409
2410 switch (ps) {
2411 case RDMA_PS_SDP:
2412 sdp_hdr = hdr;
2413 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2414 return -EINVAL;
2415 sdp_set_ip_ver(sdp_hdr, 4);
2416 sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2417 sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2418 sdp_hdr->port = src4->sin_port;
2419 break;
2420 default:
2421 cma_hdr = hdr;
2422 cma_hdr->cma_version = CMA_VERSION;
2423 cma_set_ip_ver(cma_hdr, 4);
2424 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2425 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2426 cma_hdr->port = src4->sin_port;
2427 break;
2428 }
2429 } else {
2430 struct sockaddr_in6 *src6, *dst6;
2431
2432 src6 = (struct sockaddr_in6 *) &route->addr.src_addr;
2433 dst6 = (struct sockaddr_in6 *) &route->addr.dst_addr;
2434
2435 switch (ps) {
2436 case RDMA_PS_SDP:
2437 sdp_hdr = hdr;
2438 if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2439 return -EINVAL;
2440 sdp_set_ip_ver(sdp_hdr, 6);
2441 sdp_hdr->src_addr.ip6 = src6->sin6_addr;
2442 sdp_hdr->dst_addr.ip6 = dst6->sin6_addr;
2443 sdp_hdr->port = src6->sin6_port;
2444 break;
2445 default:
2446 cma_hdr = hdr;
2447 cma_hdr->cma_version = CMA_VERSION;
2448 cma_set_ip_ver(cma_hdr, 6);
2449 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2450 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2451 cma_hdr->port = src6->sin6_port;
2452 break;
2453 }
e51060f0
SH		 2454 }
2455 return 0;
2456}
2457
628e5f6d
SH		 2458static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2459 struct ib_cm_event *ib_event)
2460{
2461 struct rdma_id_private *id_priv = cm_id->context;
2462 struct rdma_cm_event event;
2463 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2464 int ret = 0;
2465
550e5ca7 2466 if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
8aa08602 2467 return 0;
628e5f6d 2468
8aa08602 2469 memset(&event, 0, sizeof event);
628e5f6d
SH		 2470 switch (ib_event->event) {
2471 case IB_CM_SIDR_REQ_ERROR:
2472 event.event = RDMA_CM_EVENT_UNREACHABLE;
2473 event.status = -ETIMEDOUT;
2474 break;
2475 case IB_CM_SIDR_REP_RECEIVED:
2476 event.param.ud.private_data = ib_event->private_data;
2477 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2478 if (rep->status != IB_SIDR_SUCCESS) {
2479 event.event = RDMA_CM_EVENT_UNREACHABLE;
2480 event.status = ib_event->param.sidr_rep_rcvd.status;
2481 break;
2482 }
d2ca39f2
YE		 2483 ret = cma_set_qkey(id_priv);
2484 if (ret) {
2485 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2486 event.status = -EINVAL;
2487 break;
2488 }
c8f6a362 2489 if (id_priv->qkey != rep->qkey) {
628e5f6d
SH		 2490 event.event = RDMA_CM_EVENT_UNREACHABLE;
2491 event.status = -EINVAL;
2492 break;
2493 }
2494 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2495 id_priv->id.route.path_rec,
2496 &event.param.ud.ah_attr);
2497 event.param.ud.qp_num = rep->qpn;
2498 event.param.ud.qkey = rep->qkey;
2499 event.event = RDMA_CM_EVENT_ESTABLISHED;
2500 event.status = 0;
2501 break;
2502 default:
468f2239 2503 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
628e5f6d
SH		 2504 ib_event->event);
2505 goto out;
2506 }
2507
2508 ret = id_priv->id.event_handler(&id_priv->id, &event);
2509 if (ret) {
2510 /* Destroy the CM ID by returning a non-zero value. */
2511 id_priv->cm_id.ib = NULL;
550e5ca7 2512 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 2513 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 2514 rdma_destroy_id(&id_priv->id);
2515 return ret;
2516 }
2517out:
de910bd9 2518 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 2519 return ret;
2520}
2521
2522static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2523 struct rdma_conn_param *conn_param)
2524{
2525 struct ib_cm_sidr_req_param req;
2526 struct rdma_route *route;
0c9361fc 2527 struct ib_cm_id *id;
628e5f6d
SH		 2528 int ret;
2529
2530 req.private_data_len = sizeof(struct cma_hdr) +
2531 conn_param->private_data_len;
04ded167
SH		 2532 if (req.private_data_len < conn_param->private_data_len)
2533 return -EINVAL;
2534
628e5f6d
SH		 2535 req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2536 if (!req.private_data)
2537 return -ENOMEM;
2538
2539 if (conn_param->private_data && conn_param->private_data_len)
2540 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2541 conn_param->private_data, conn_param->private_data_len);
2542
2543 route = &id_priv->id.route;
2544 ret = cma_format_hdr((void *) req.private_data, id_priv->id.ps, route);
2545 if (ret)
2546 goto out;
2547
0c9361fc
JM		 2548 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2549 id_priv);
2550 if (IS_ERR(id)) {
2551 ret = PTR_ERR(id);
628e5f6d
SH		 2552 goto out;
2553 }
0c9361fc 2554 id_priv->cm_id.ib = id;
628e5f6d
SH		 2555
2556 req.path = route->path_rec;
2557 req.service_id = cma_get_service_id(id_priv->id.ps,
3f446754 2558 (struct sockaddr *) &route->addr.dst_addr);
628e5f6d
SH		 2559 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2560 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2561
2562 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2563 if (ret) {
2564 ib_destroy_cm_id(id_priv->cm_id.ib);
2565 id_priv->cm_id.ib = NULL;
2566 }
2567out:
2568 kfree(req.private_data);
2569 return ret;
2570}
2571
e51060f0
SH		 2572static int cma_connect_ib(struct rdma_id_private *id_priv,
2573 struct rdma_conn_param *conn_param)
2574{
2575 struct ib_cm_req_param req;
2576 struct rdma_route *route;
2577 void *private_data;
0c9361fc 2578 struct ib_cm_id *id;
e51060f0
SH		 2579 int offset, ret;
2580
2581 memset(&req, 0, sizeof req);
2582 offset = cma_user_data_offset(id_priv->id.ps);
2583 req.private_data_len = offset + conn_param->private_data_len;
04ded167
SH		 2584 if (req.private_data_len < conn_param->private_data_len)
2585 return -EINVAL;
2586
e51060f0
SH		 2587 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2588 if (!private_data)
2589 return -ENOMEM;
2590
2591 if (conn_param->private_data && conn_param->private_data_len)
2592 memcpy(private_data + offset, conn_param->private_data,
2593 conn_param->private_data_len);
2594
0c9361fc
JM		 2595 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2596 if (IS_ERR(id)) {
2597 ret = PTR_ERR(id);
e51060f0
SH		 2598 goto out;
2599 }
0c9361fc 2600 id_priv->cm_id.ib = id;
e51060f0
SH		 2601
2602 route = &id_priv->id.route;
2603 ret = cma_format_hdr(private_data, id_priv->id.ps, route);
2604 if (ret)
2605 goto out;
2606 req.private_data = private_data;
2607
2608 req.primary_path = &route->path_rec[0];
2609 if (route->num_paths == 2)
2610 req.alternate_path = &route->path_rec[1];
2611
2612 req.service_id = cma_get_service_id(id_priv->id.ps,
3f446754 2613 (struct sockaddr *) &route->addr.dst_addr);
e51060f0 2614 req.qp_num = id_priv->qp_num;
18c441a6 2615 req.qp_type = id_priv->id.qp_type;
e51060f0
SH		 2616 req.starting_psn = id_priv->seq_num;
2617 req.responder_resources = conn_param->responder_resources;
2618 req.initiator_depth = conn_param->initiator_depth;
2619 req.flow_control = conn_param->flow_control;
2620 req.retry_count = conn_param->retry_count;
2621 req.rnr_retry_count = conn_param->rnr_retry_count;
2622 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2623 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2624 req.max_cm_retries = CMA_MAX_CM_RETRIES;
2625 req.srq = id_priv->srq ? 1 : 0;
2626
2627 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2628out:
0c9361fc
JM		 2629 if (ret && !IS_ERR(id)) {
2630 ib_destroy_cm_id(id);
675a027c
KK		 2631 id_priv->cm_id.ib = NULL;
2632 }
2633
e51060f0
SH		 2634 kfree(private_data);
2635 return ret;
2636}
2637
07ebafba
TT		 2638static int cma_connect_iw(struct rdma_id_private *id_priv,
2639 struct rdma_conn_param *conn_param)
2640{
2641 struct iw_cm_id *cm_id;
2642 struct sockaddr_in* sin;
2643 int ret;
2644 struct iw_cm_conn_param iw_param;
2645
2646 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
0c9361fc
JM		 2647 if (IS_ERR(cm_id))
2648 return PTR_ERR(cm_id);
07ebafba
TT		 2649
2650 id_priv->cm_id.iw = cm_id;
2651
2652 sin = (struct sockaddr_in*) &id_priv->id.route.addr.src_addr;
2653 cm_id->local_addr = *sin;
2654
2655 sin = (struct sockaddr_in*) &id_priv->id.route.addr.dst_addr;
2656 cm_id->remote_addr = *sin;
2657
5851bb89 2658 ret = cma_modify_qp_rtr(id_priv, conn_param);
675a027c
KK		 2659 if (ret)
2660 goto out;
07ebafba 2661
f45ee80e
HS		 2662 if (conn_param) {
2663 iw_param.ord = conn_param->initiator_depth;
2664 iw_param.ird = conn_param->responder_resources;
2665 iw_param.private_data = conn_param->private_data;
2666 iw_param.private_data_len = conn_param->private_data_len;
2667 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2668 } else {
2669 memset(&iw_param, 0, sizeof iw_param);
07ebafba 2670 iw_param.qpn = id_priv->qp_num;
f45ee80e 2671 }
07ebafba
TT		 2672 ret = iw_cm_connect(cm_id, &iw_param);
2673out:
0c9361fc 2674 if (ret) {
675a027c
KK		 2675 iw_destroy_cm_id(cm_id);
2676 id_priv->cm_id.iw = NULL;
2677 }
07ebafba
TT		 2678 return ret;
2679}
2680
e51060f0
SH		 2681int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2682{
2683 struct rdma_id_private *id_priv;
2684 int ret;
2685
2686 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2687 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
e51060f0
SH		 2688 return -EINVAL;
2689
2690 if (!id->qp) {
2691 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 2692 id_priv->srq = conn_param->srq;
2693 }
2694
07ebafba
TT		 2695 switch (rdma_node_get_transport(id->device->node_type)) {
2696 case RDMA_TRANSPORT_IB:
b26f9b99 2697 if (id->qp_type == IB_QPT_UD)
628e5f6d
SH		 2698 ret = cma_resolve_ib_udp(id_priv, conn_param);
2699 else
2700 ret = cma_connect_ib(id_priv, conn_param);
e51060f0 2701 break;
07ebafba
TT		 2702 case RDMA_TRANSPORT_IWARP:
2703 ret = cma_connect_iw(id_priv, conn_param);
2704 break;
e51060f0
SH		 2705 default:
2706 ret = -ENOSYS;
2707 break;
2708 }
2709 if (ret)
2710 goto err;
2711
2712 return 0;
2713err:
550e5ca7 2714 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
e51060f0
SH		 2715 return ret;
2716}
2717EXPORT_SYMBOL(rdma_connect);
2718
2719static int cma_accept_ib(struct rdma_id_private *id_priv,
2720 struct rdma_conn_param *conn_param)
2721{
2722 struct ib_cm_rep_param rep;
5851bb89 2723 int ret;
0fe313b0 2724
5851bb89
SH		 2725 ret = cma_modify_qp_rtr(id_priv, conn_param);
2726 if (ret)
2727 goto out;
0fe313b0 2728
5851bb89
SH		 2729 ret = cma_modify_qp_rts(id_priv, conn_param);
2730 if (ret)
2731 goto out;
e51060f0
SH		 2732
2733 memset(&rep, 0, sizeof rep);
2734 rep.qp_num = id_priv->qp_num;
2735 rep.starting_psn = id_priv->seq_num;
2736 rep.private_data = conn_param->private_data;
2737 rep.private_data_len = conn_param->private_data_len;
2738 rep.responder_resources = conn_param->responder_resources;
2739 rep.initiator_depth = conn_param->initiator_depth;
e51060f0
SH		 2740 rep.failover_accepted = 0;
2741 rep.flow_control = conn_param->flow_control;
2742 rep.rnr_retry_count = conn_param->rnr_retry_count;
2743 rep.srq = id_priv->srq ? 1 : 0;
2744
0fe313b0
SH		 2745 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2746out:
2747 return ret;
e51060f0
SH		 2748}
2749
07ebafba
TT		 2750static int cma_accept_iw(struct rdma_id_private *id_priv,
2751 struct rdma_conn_param *conn_param)
2752{
2753 struct iw_cm_conn_param iw_param;
2754 int ret;
2755
5851bb89 2756 ret = cma_modify_qp_rtr(id_priv, conn_param);
07ebafba
TT		 2757 if (ret)
2758 return ret;
2759
2760 iw_param.ord = conn_param->initiator_depth;
2761 iw_param.ird = conn_param->responder_resources;
2762 iw_param.private_data = conn_param->private_data;
2763 iw_param.private_data_len = conn_param->private_data_len;
2764 if (id_priv->id.qp) {
2765 iw_param.qpn = id_priv->qp_num;
2766 } else
2767 iw_param.qpn = conn_param->qp_num;
2768
2769 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2770}
2771
628e5f6d
SH		 2772static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2773 enum ib_cm_sidr_status status,
2774 const void *private_data, int private_data_len)
2775{
2776 struct ib_cm_sidr_rep_param rep;
d2ca39f2 2777 int ret;
628e5f6d
SH		 2778
2779 memset(&rep, 0, sizeof rep);
2780 rep.status = status;
2781 if (status == IB_SIDR_SUCCESS) {
d2ca39f2
YE		 2782 ret = cma_set_qkey(id_priv);
2783 if (ret)
2784 return ret;
628e5f6d 2785 rep.qp_num = id_priv->qp_num;
c8f6a362 2786 rep.qkey = id_priv->qkey;
628e5f6d
SH		 2787 }
2788 rep.private_data = private_data;
2789 rep.private_data_len = private_data_len;
2790
2791 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2792}
2793
e51060f0
SH		 2794int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2795{
2796 struct rdma_id_private *id_priv;
2797 int ret;
2798
2799 id_priv = container_of(id, struct rdma_id_private, id);
83e9502d
NM		 2800
2801 id_priv->owner = task_pid_nr(current);
2802
550e5ca7 2803 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
e51060f0
SH		 2804 return -EINVAL;
2805
2806 if (!id->qp && conn_param) {
2807 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 2808 id_priv->srq = conn_param->srq;
2809 }
2810
07ebafba
TT		 2811 switch (rdma_node_get_transport(id->device->node_type)) {
2812 case RDMA_TRANSPORT_IB:
f45ee80e
HS		 2813 if (id->qp_type == IB_QPT_UD) {
2814 if (conn_param)
2815 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2816 conn_param->private_data,
2817 conn_param->private_data_len);
2818 else
2819 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2820 NULL, 0);
2821 } else {
2822 if (conn_param)
2823 ret = cma_accept_ib(id_priv, conn_param);
2824 else
2825 ret = cma_rep_recv(id_priv);
2826 }
e51060f0 2827 break;
07ebafba
TT		 2828 case RDMA_TRANSPORT_IWARP:
2829 ret = cma_accept_iw(id_priv, conn_param);
2830 break;
e51060f0
SH		 2831 default:
2832 ret = -ENOSYS;
2833 break;
2834 }
2835
2836 if (ret)
2837 goto reject;
2838
2839 return 0;
2840reject:
c5483388 2841 cma_modify_qp_err(id_priv);
e51060f0
SH		 2842 rdma_reject(id, NULL, 0);
2843 return ret;
2844}
2845EXPORT_SYMBOL(rdma_accept);
2846
0fe313b0
SH		 2847int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
2848{
2849 struct rdma_id_private *id_priv;
2850 int ret;
2851
2852 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 2853 if (!id_priv->cm_id.ib)
0fe313b0
SH		 2854 return -EINVAL;
2855
2856 switch (id->device->node_type) {
2857 case RDMA_NODE_IB_CA:
2858 ret = ib_cm_notify(id_priv->cm_id.ib, event);
2859 break;
2860 default:
2861 ret = 0;
2862 break;
2863 }
2864 return ret;
2865}
2866EXPORT_SYMBOL(rdma_notify);
2867
e51060f0
SH		 2868int rdma_reject(struct rdma_cm_id *id, const void *private_data,
2869 u8 private_data_len)
2870{
2871 struct rdma_id_private *id_priv;
2872 int ret;
2873
2874 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 2875 if (!id_priv->cm_id.ib)
e51060f0
SH		 2876 return -EINVAL;
2877
07ebafba
TT		 2878 switch (rdma_node_get_transport(id->device->node_type)) {
2879 case RDMA_TRANSPORT_IB:
b26f9b99 2880 if (id->qp_type == IB_QPT_UD)
628e5f6d
SH		 2881 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
2882 private_data, private_data_len);
2883 else
2884 ret = ib_send_cm_rej(id_priv->cm_id.ib,
2885 IB_CM_REJ_CONSUMER_DEFINED, NULL,
2886 0, private_data, private_data_len);
e51060f0 2887 break;
07ebafba
TT		 2888 case RDMA_TRANSPORT_IWARP:
2889 ret = iw_cm_reject(id_priv->cm_id.iw,
2890 private_data, private_data_len);
2891 break;
e51060f0
SH		 2892 default:
2893 ret = -ENOSYS;
2894 break;
2895 }
2896 return ret;
2897}
2898EXPORT_SYMBOL(rdma_reject);
2899
2900int rdma_disconnect(struct rdma_cm_id *id)
2901{
2902 struct rdma_id_private *id_priv;
2903 int ret;
2904
2905 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 2906 if (!id_priv->cm_id.ib)
e51060f0
SH		 2907 return -EINVAL;
2908
07ebafba
TT		 2909 switch (rdma_node_get_transport(id->device->node_type)) {
2910 case RDMA_TRANSPORT_IB:
c5483388 2911 ret = cma_modify_qp_err(id_priv);
07ebafba
TT		 2912 if (ret)
2913 goto out;
e51060f0
SH		 2914 /* Initiate or respond to a disconnect. */
2915 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
2916 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
2917 break;
07ebafba
TT		 2918 case RDMA_TRANSPORT_IWARP:
2919 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
2920 break;
e51060f0 2921 default:
07ebafba 2922 ret = -EINVAL;
e51060f0
SH		 2923 break;
2924 }
2925out:
2926 return ret;
2927}
2928EXPORT_SYMBOL(rdma_disconnect);
2929
c8f6a362
SH		 2930static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
2931{
2932 struct rdma_id_private *id_priv;
2933 struct cma_multicast *mc = multicast->context;
2934 struct rdma_cm_event event;
2935 int ret;
2936
2937 id_priv = mc->id_priv;
550e5ca7
NM		 2938 if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
2939 cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
8aa08602 2940 return 0;
c8f6a362 2941
c5483388 2942 mutex_lock(&id_priv->qp_mutex);
c8f6a362
SH		 2943 if (!status && id_priv->id.qp)
2944 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
46ea5061 2945 be16_to_cpu(multicast->rec.mlid));
c5483388 2946 mutex_unlock(&id_priv->qp_mutex);
c8f6a362
SH		 2947
2948 memset(&event, 0, sizeof event);
2949 event.status = status;
2950 event.param.ud.private_data = mc->context;
2951 if (!status) {
2952 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
2953 ib_init_ah_from_mcmember(id_priv->id.device,
2954 id_priv->id.port_num, &multicast->rec,
2955 &event.param.ud.ah_attr);
2956 event.param.ud.qp_num = 0xFFFFFF;
2957 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
2958 } else
2959 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
2960
2961 ret = id_priv->id.event_handler(&id_priv->id, &event);
2962 if (ret) {
550e5ca7 2963 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 2964 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 2965 rdma_destroy_id(&id_priv->id);
2966 return 0;
2967 }
8aa08602 2968
de910bd9 2969 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 2970 return 0;
2971}
2972
2973static void cma_set_mgid(struct rdma_id_private *id_priv,
2974 struct sockaddr *addr, union ib_gid *mgid)
2975{
2976 unsigned char mc_map[MAX_ADDR_LEN];
2977 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2978 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
2979 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
2980
2981 if (cma_any_addr(addr)) {
2982 memset(mgid, 0, sizeof *mgid);
2983 } else if ((addr->sa_family == AF_INET6) &&
1c9b2819 2984 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
c8f6a362
SH		 2985 0xFF10A01B)) {
2986 /* IPv6 address is an SA assigned MGID. */
2987 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
e2e62697
JG		 2988 } else if ((addr->sa_family == AF_INET6)) {
2989 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
2990 if (id_priv->id.ps == RDMA_PS_UDP)
2991 mc_map[7] = 0x01; /* Use RDMA CM signature */
2992 *mgid = *(union ib_gid *) (mc_map + 4);
c8f6a362 2993 } else {
a9e527e3 2994 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
c8f6a362
SH		 2995 if (id_priv->id.ps == RDMA_PS_UDP)
2996 mc_map[7] = 0x01; /* Use RDMA CM signature */
c8f6a362
SH		 2997 *mgid = *(union ib_gid *) (mc_map + 4);
2998 }
2999}
3000
3001static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3002 struct cma_multicast *mc)
3003{
3004 struct ib_sa_mcmember_rec rec;
3005 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3006 ib_sa_comp_mask comp_mask;
3007 int ret;
3008
3009 ib_addr_get_mgid(dev_addr, &rec.mgid);
3010 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3011 &rec.mgid, &rec);
3012 if (ret)
3013 return ret;
3014
3f446754 3015 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
c8f6a362
SH		 3016 if (id_priv->id.ps == RDMA_PS_UDP)
3017 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
6f8372b6 3018 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
c8f6a362
SH		 3019 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3020 rec.join_state = 1;
3021
3022 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3023 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3024 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3025 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3026 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3027
84adeee9
YE		 3028 if (id_priv->id.ps == RDMA_PS_IPOIB)
3029 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3030 IB_SA_MCMEMBER_REC_RATE_SELECTOR;
3031
c8f6a362
SH		 3032 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3033 id_priv->id.port_num, &rec,
3034 comp_mask, GFP_KERNEL,
3035 cma_ib_mc_handler, mc);
3036 if (IS_ERR(mc->multicast.ib))
3037 return PTR_ERR(mc->multicast.ib);
3038
3039 return 0;
3040}
3041
3c86aa70
EC		 3042static void iboe_mcast_work_handler(struct work_struct *work)
3043{
3044 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3045 struct cma_multicast *mc = mw->mc;
3046 struct ib_sa_multicast *m = mc->multicast.ib;
3047
3048 mc->multicast.ib->context = mc;
3049 cma_ib_mc_handler(0, m);
3050 kref_put(&mc->mcref, release_mc);
3051 kfree(mw);
3052}
3053
3054static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3055{
3056 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3057 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3058
3059 if (cma_any_addr(addr)) {
3060 memset(mgid, 0, sizeof *mgid);
3061 } else if (addr->sa_family == AF_INET6) {
3062 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3063 } else {
3064 mgid->raw[0] = 0xff;
3065 mgid->raw[1] = 0x0e;
3066 mgid->raw[2] = 0;
3067 mgid->raw[3] = 0;
3068 mgid->raw[4] = 0;
3069 mgid->raw[5] = 0;
3070 mgid->raw[6] = 0;
3071 mgid->raw[7] = 0;
3072 mgid->raw[8] = 0;
3073 mgid->raw[9] = 0;
3074 mgid->raw[10] = 0xff;
3075 mgid->raw[11] = 0xff;
3076 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3077 }
3078}
3079
3080static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3081 struct cma_multicast *mc)
3082{
3083 struct iboe_mcast_work *work;
3084 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3085 int err;
3086 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3087 struct net_device *ndev = NULL;
3088
3089 if (cma_zero_addr((struct sockaddr *)&mc->addr))
3090 return -EINVAL;
3091
3092 work = kzalloc(sizeof *work, GFP_KERNEL);
3093 if (!work)
3094 return -ENOMEM;
3095
3096 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3097 if (!mc->multicast.ib) {
3098 err = -ENOMEM;
3099 goto out1;
3100 }
3101
3102 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3103
3104 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3105 if (id_priv->id.ps == RDMA_PS_UDP)
3106 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3107
3108 if (dev_addr->bound_dev_if)
3109 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3110 if (!ndev) {
3111 err = -ENODEV;
3112 goto out2;
3113 }
3114 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3115 mc->multicast.ib->rec.hop_limit = 1;
3116 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3117 dev_put(ndev);
3118 if (!mc->multicast.ib->rec.mtu) {
3119 err = -EINVAL;
3120 goto out2;
3121 }
3122 iboe_addr_get_sgid(dev_addr, &mc->multicast.ib->rec.port_gid);
3123 work->id = id_priv;
3124 work->mc = mc;
3125 INIT_WORK(&work->work, iboe_mcast_work_handler);
3126 kref_get(&mc->mcref);
3127 queue_work(cma_wq, &work->work);
3128
3129 return 0;
3130
3131out2:
3132 kfree(mc->multicast.ib);
3133out1:
3134 kfree(work);
3135 return err;
3136}
3137
c8f6a362
SH		 3138int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3139 void *context)
3140{
3141 struct rdma_id_private *id_priv;
3142 struct cma_multicast *mc;
3143 int ret;
3144
3145 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7
NM		 3146 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3147 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
c8f6a362
SH		 3148 return -EINVAL;
3149
3150 mc = kmalloc(sizeof *mc, GFP_KERNEL);
3151 if (!mc)
3152 return -ENOMEM;
3153
3154 memcpy(&mc->addr, addr, ip_addr_size(addr));
3155 mc->context = context;
3156 mc->id_priv = id_priv;
3157
3158 spin_lock(&id_priv->lock);
3159 list_add(&mc->list, &id_priv->mc_list);
3160 spin_unlock(&id_priv->lock);
3161
3162 switch (rdma_node_get_transport(id->device->node_type)) {
3163 case RDMA_TRANSPORT_IB:
3c86aa70
EC		 3164 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3165 case IB_LINK_LAYER_INFINIBAND:
3166 ret = cma_join_ib_multicast(id_priv, mc);
3167 break;
3168 case IB_LINK_LAYER_ETHERNET:
3169 kref_init(&mc->mcref);
3170 ret = cma_iboe_join_multicast(id_priv, mc);
3171 break;
3172 default:
3173 ret = -EINVAL;
3174 }
c8f6a362
SH		 3175 break;
3176 default:
3177 ret = -ENOSYS;
3178 break;
3179 }
3180
3181 if (ret) {
3182 spin_lock_irq(&id_priv->lock);
3183 list_del(&mc->list);
3184 spin_unlock_irq(&id_priv->lock);
3185 kfree(mc);
3186 }
3187 return ret;
3188}
3189EXPORT_SYMBOL(rdma_join_multicast);
3190
3191void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3192{
3193 struct rdma_id_private *id_priv;
3194 struct cma_multicast *mc;
3195
3196 id_priv = container_of(id, struct rdma_id_private, id);
3197 spin_lock_irq(&id_priv->lock);
3198 list_for_each_entry(mc, &id_priv->mc_list, list) {
3199 if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
3200 list_del(&mc->list);
3201 spin_unlock_irq(&id_priv->lock);
3202
3203 if (id->qp)
3204 ib_detach_mcast(id->qp,
3205 &mc->multicast.ib->rec.mgid,
46ea5061 3206 be16_to_cpu(mc->multicast.ib->rec.mlid));
3c86aa70
EC		 3207 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3208 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3209 case IB_LINK_LAYER_INFINIBAND:
3210 ib_sa_free_multicast(mc->multicast.ib);
3211 kfree(mc);
3212 break;
3213 case IB_LINK_LAYER_ETHERNET:
3214 kref_put(&mc->mcref, release_mc);
3215 break;
3216 default:
3217 break;
3218 }
3219 }
c8f6a362
SH		 3220 return;
3221 }
3222 }
3223 spin_unlock_irq(&id_priv->lock);
3224}
3225EXPORT_SYMBOL(rdma_leave_multicast);
3226
dd5bdff8
OG		 3227static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3228{
3229 struct rdma_dev_addr *dev_addr;
3230 struct cma_ndev_work *work;
3231
3232 dev_addr = &id_priv->id.route.addr.dev_addr;
3233
6266ed6e 3234 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
dd5bdff8
OG		 3235 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3236 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3237 ndev->name, &id_priv->id);
3238 work = kzalloc(sizeof *work, GFP_KERNEL);
3239 if (!work)
3240 return -ENOMEM;
3241
3242 INIT_WORK(&work->work, cma_ndev_work_handler);
3243 work->id = id_priv;
3244 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3245 atomic_inc(&id_priv->refcount);
3246 queue_work(cma_wq, &work->work);
3247 }
3248
3249 return 0;
3250}
3251
3252static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3253 void *ctx)
3254{
3255 struct net_device *ndev = (struct net_device *)ctx;
3256 struct cma_device *cma_dev;
3257 struct rdma_id_private *id_priv;
3258 int ret = NOTIFY_DONE;
3259
3260 if (dev_net(ndev) != &init_net)
3261 return NOTIFY_DONE;
3262
3263 if (event != NETDEV_BONDING_FAILOVER)
3264 return NOTIFY_DONE;
3265
3266 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3267 return NOTIFY_DONE;
3268
3269 mutex_lock(&lock);
3270 list_for_each_entry(cma_dev, &dev_list, list)
3271 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3272 ret = cma_netdev_change(ndev, id_priv);
3273 if (ret)
3274 goto out;
3275 }
3276
3277out:
3278 mutex_unlock(&lock);
3279 return ret;
3280}
3281
3282static struct notifier_block cma_nb = {
3283 .notifier_call = cma_netdev_callback
3284};
3285
e51060f0
SH		 3286static void cma_add_one(struct ib_device *device)
3287{
3288 struct cma_device *cma_dev;
3289 struct rdma_id_private *id_priv;
3290
3291 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3292 if (!cma_dev)
3293 return;
3294
3295 cma_dev->device = device;
e51060f0
SH		 3296
3297 init_completion(&cma_dev->comp);
3298 atomic_set(&cma_dev->refcount, 1);
3299 INIT_LIST_HEAD(&cma_dev->id_list);
3300 ib_set_client_data(device, &cma_client, cma_dev);
3301
3302 mutex_lock(&lock);
3303 list_add_tail(&cma_dev->list, &dev_list);
3304 list_for_each_entry(id_priv, &listen_any_list, list)
3305 cma_listen_on_dev(id_priv, cma_dev);
3306 mutex_unlock(&lock);
e51060f0
SH		 3307}
3308
3309static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3310{
a1b1b61f 3311 struct rdma_cm_event event;
550e5ca7 3312 enum rdma_cm_state state;
de910bd9 3313 int ret = 0;
e51060f0
SH		 3314
3315 /* Record that we want to remove the device */
550e5ca7
NM		 3316 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3317 if (state == RDMA_CM_DESTROYING)
e51060f0
SH		 3318 return 0;
3319
3320 cma_cancel_operation(id_priv, state);
de910bd9 3321 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 3322
3323 /* Check for destruction from another callback. */
550e5ca7 3324 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
de910bd9 3325 goto out;
e51060f0 3326
a1b1b61f
SH		 3327 memset(&event, 0, sizeof event);
3328 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
de910bd9
OG		 3329 ret = id_priv->id.event_handler(&id_priv->id, &event);
3330out:
3331 mutex_unlock(&id_priv->handler_mutex);
3332 return ret;
e51060f0
SH		 3333}
3334
3335static void cma_process_remove(struct cma_device *cma_dev)
3336{
e51060f0
SH		 3337 struct rdma_id_private *id_priv;
3338 int ret;
3339
e51060f0
SH		 3340 mutex_lock(&lock);
3341 while (!list_empty(&cma_dev->id_list)) {
3342 id_priv = list_entry(cma_dev->id_list.next,
3343 struct rdma_id_private, list);
3344
d02d1f53 3345 list_del(&id_priv->listen_list);
94de178a 3346 list_del_init(&id_priv->list);
e51060f0
SH		 3347 atomic_inc(&id_priv->refcount);
3348 mutex_unlock(&lock);
3349
d02d1f53 3350 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
e51060f0
SH		 3351 cma_deref_id(id_priv);
3352 if (ret)
3353 rdma_destroy_id(&id_priv->id);
3354
3355 mutex_lock(&lock);
3356 }
3357 mutex_unlock(&lock);
3358
3359 cma_deref_dev(cma_dev);
3360 wait_for_completion(&cma_dev->comp);
3361}
3362
3363static void cma_remove_one(struct ib_device *device)
3364{
3365 struct cma_device *cma_dev;
3366
3367 cma_dev = ib_get_client_data(device, &cma_client);
3368 if (!cma_dev)
3369 return;
3370
3371 mutex_lock(&lock);
3372 list_del(&cma_dev->list);
3373 mutex_unlock(&lock);
3374
3375 cma_process_remove(cma_dev);
3376 kfree(cma_dev);
3377}
3378
753f618a
NM		 3379static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3380{
3381 struct nlmsghdr *nlh;
3382 struct rdma_cm_id_stats *id_stats;
3383 struct rdma_id_private *id_priv;
3384 struct rdma_cm_id *id = NULL;
3385 struct cma_device *cma_dev;
3386 int i_dev = 0, i_id = 0;
3387
3388 /*
3389 * We export all of the IDs as a sequence of messages. Each
3390 * ID gets its own netlink message.
3391 */
3392 mutex_lock(&lock);
3393
3394 list_for_each_entry(cma_dev, &dev_list, list) {
3395 if (i_dev < cb->args[0]) {
3396 i_dev++;
3397 continue;
3398 }
3399
3400 i_id = 0;
3401 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3402 if (i_id < cb->args[1]) {
3403 i_id++;
3404 continue;
3405 }
3406
3407 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3408 sizeof *id_stats, RDMA_NL_RDMA_CM,
3409 RDMA_NL_RDMA_CM_ID_STATS);
3410 if (!id_stats)
3411 goto out;
3412
3413 memset(id_stats, 0, sizeof *id_stats);
3414 id = &id_priv->id;
3415 id_stats->node_type = id->route.addr.dev_addr.dev_type;
3416 id_stats->port_num = id->port_num;
3417 id_stats->bound_dev_if =
3418 id->route.addr.dev_addr.bound_dev_if;
3419
3420 if (id->route.addr.src_addr.ss_family == AF_INET) {
3421 if (ibnl_put_attr(skb, nlh,
3422 sizeof(struct sockaddr_in),
3423 &id->route.addr.src_addr,
3424 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
3425 goto out;
3426 }
3427 if (ibnl_put_attr(skb, nlh,
3428 sizeof(struct sockaddr_in),
3429 &id->route.addr.dst_addr,
3430 RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
3431 goto out;
3432 }
3433 } else if (id->route.addr.src_addr.ss_family == AF_INET6) {
3434 if (ibnl_put_attr(skb, nlh,
3435 sizeof(struct sockaddr_in6),
3436 &id->route.addr.src_addr,
3437 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
3438 goto out;
3439 }
3440 if (ibnl_put_attr(skb, nlh,
3441 sizeof(struct sockaddr_in6),
3442 &id->route.addr.dst_addr,
3443 RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
3444 goto out;
3445 }
3446 }
3447
83e9502d 3448 id_stats->pid = id_priv->owner;
753f618a
NM		 3449 id_stats->port_space = id->ps;
3450 id_stats->cm_state = id_priv->state;
3451 id_stats->qp_num = id_priv->qp_num;
3452 id_stats->qp_type = id->qp_type;
3453
3454 i_id++;
3455 }
3456
3457 cb->args[1] = 0;
3458 i_dev++;
3459 }
3460
3461out:
3462 mutex_unlock(&lock);
3463 cb->args[0] = i_dev;
3464 cb->args[1] = i_id;
3465
3466 return skb->len;
3467}
3468
3469static const struct ibnl_client_cbs cma_cb_table[] = {
3470 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats },
3471};
3472
716abb1f 3473static int __init cma_init(void)
e51060f0 3474{
5d7220e8 3475 int ret;
227b60f5 3476
c7f743a6 3477 cma_wq = create_singlethread_workqueue("rdma_cm");
e51060f0
SH		 3478 if (!cma_wq)
3479 return -ENOMEM;
3480
c1a0b23b 3481 ib_sa_register_client(&sa_client);
7a118df3 3482 rdma_addr_register_client(&addr_client);
dd5bdff8 3483 register_netdevice_notifier(&cma_nb);
c1a0b23b 3484
e51060f0
SH		 3485 ret = ib_register_client(&cma_client);
3486 if (ret)
3487 goto err;
753f618a
NM		 3488
3489 if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3490 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3491
e51060f0
SH		 3492 return 0;
3493
3494err:
dd5bdff8 3495 unregister_netdevice_notifier(&cma_nb);
7a118df3 3496 rdma_addr_unregister_client(&addr_client);
c1a0b23b 3497 ib_sa_unregister_client(&sa_client);
e51060f0
SH		 3498 destroy_workqueue(cma_wq);
3499 return ret;
3500}
3501
716abb1f 3502static void __exit cma_cleanup(void)
e51060f0 3503{
753f618a 3504 ibnl_remove_client(RDMA_NL_RDMA_CM);
e51060f0 3505 ib_unregister_client(&cma_client);
dd5bdff8 3506 unregister_netdevice_notifier(&cma_nb);
7a118df3 3507 rdma_addr_unregister_client(&addr_client);
c1a0b23b 3508 ib_sa_unregister_client(&sa_client);
e51060f0
SH		 3509 destroy_workqueue(cma_wq);
3510 idr_destroy(&sdp_ps);
3511 idr_destroy(&tcp_ps);
628e5f6d 3512 idr_destroy(&udp_ps);
c8f6a362 3513 idr_destroy(&ipoib_ps);
2d2e9415 3514 idr_destroy(&ib_ps);
e51060f0
SH		 3515}
3516
3517module_init(cma_init);
3518module_exit(cma_cleanup);
Linux kernel - Deliverables
This page took 1.217111 seconds and 5 git commands to generate.