]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - drivers/infiniband/core/cma.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: Ubuntu-4.10.0-37.41
[mirror_ubuntu-zesty-kernel.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>
bee3c3c9 41#include <linux/igmp.h>
e51060f0 42#include <linux/idr.h>
07ebafba 43#include <linux/inetdevice.h>
5a0e3ad6 44#include <linux/slab.h>
e4dd23d7 45#include <linux/module.h>
366cddb4 46#include <net/route.h>
e51060f0 47
4be74b42
HE		 48#include <net/net_namespace.h>
49#include <net/netns/generic.h>
e51060f0 50#include <net/tcp.h>
1f5175ad 51#include <net/ipv6.h>
f887f2ac
HE		 52#include <net/ip_fib.h>
53#include <net/ip6_route.h>
e51060f0
SH		 54
55#include <rdma/rdma_cm.h>
56#include <rdma/rdma_cm_ib.h>
753f618a 57#include <rdma/rdma_netlink.h>
2e2d190c 58#include <rdma/ib.h>
e51060f0
SH		 59#include <rdma/ib_cache.h>
60#include <rdma/ib_cm.h>
61#include <rdma/ib_sa.h>
07ebafba 62#include <rdma/iw_cm.h>
e51060f0 63
218a773f
MB		 64#include "core_priv.h"
65
e51060f0
SH		 66MODULE_AUTHOR("Sean Hefty");
67MODULE_DESCRIPTION("Generic RDMA CM Agent");
68MODULE_LICENSE("Dual BSD/GPL");
69
70#define CMA_CM_RESPONSE_TIMEOUT 20
ab15c95a 71#define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
d5bb7599 72#define CMA_MAX_CM_RETRIES 15
dcb3f974 73#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
3c86aa70 74#define CMA_IBOE_PACKET_LIFETIME 18
e51060f0 75
2b1b5b60
SG		 76static const char * const cma_events[] = {
77 [RDMA_CM_EVENT_ADDR_RESOLVED] = "address resolved",
78 [RDMA_CM_EVENT_ADDR_ERROR] = "address error",
79 [RDMA_CM_EVENT_ROUTE_RESOLVED] = "route resolved ",
80 [RDMA_CM_EVENT_ROUTE_ERROR] = "route error",
81 [RDMA_CM_EVENT_CONNECT_REQUEST] = "connect request",
82 [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
83 [RDMA_CM_EVENT_CONNECT_ERROR] = "connect error",
84 [RDMA_CM_EVENT_UNREACHABLE] = "unreachable",
85 [RDMA_CM_EVENT_REJECTED] = "rejected",
86 [RDMA_CM_EVENT_ESTABLISHED] = "established",
87 [RDMA_CM_EVENT_DISCONNECTED] = "disconnected",
88 [RDMA_CM_EVENT_DEVICE_REMOVAL] = "device removal",
89 [RDMA_CM_EVENT_MULTICAST_JOIN] = "multicast join",
90 [RDMA_CM_EVENT_MULTICAST_ERROR] = "multicast error",
91 [RDMA_CM_EVENT_ADDR_CHANGE] = "address change",
92 [RDMA_CM_EVENT_TIMEWAIT_EXIT] = "timewait exit",
93};
94
db7489e0 95const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
2b1b5b60
SG		 96{
97 size_t index = event;
98
99 return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
100 cma_events[index] : "unrecognized event";
101}
102EXPORT_SYMBOL(rdma_event_msg);
103
77a5db13
SW		 104const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
105 int reason)
106{
107 if (rdma_ib_or_roce(id->device, id->port_num))
108 return ibcm_reject_msg(reason);
109
110 if (rdma_protocol_iwarp(id->device, id->port_num))
111 return iwcm_reject_msg(reason);
112
113 WARN_ON_ONCE(1);
114 return "unrecognized transport";
115}
116EXPORT_SYMBOL(rdma_reject_msg);
117
5042a73d
SW		 118bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
119{
120 if (rdma_ib_or_roce(id->device, id->port_num))
121 return reason == IB_CM_REJ_CONSUMER_DEFINED;
122
123 if (rdma_protocol_iwarp(id->device, id->port_num))
124 return reason == -ECONNREFUSED;
125
126 WARN_ON_ONCE(1);
127 return false;
128}
129EXPORT_SYMBOL(rdma_is_consumer_reject);
130
5f244104
SW		 131const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
132 struct rdma_cm_event *ev, u8 *data_len)
133{
134 const void *p;
135
136 if (rdma_is_consumer_reject(id, ev->status)) {
137 *data_len = ev->param.conn.private_data_len;
138 p = ev->param.conn.private_data;
139 } else {
140 *data_len = 0;
141 p = NULL;
142 }
143 return p;
144}
145EXPORT_SYMBOL(rdma_consumer_reject_data);
146
e51060f0 147static void cma_add_one(struct ib_device *device);
7c1eb45a 148static void cma_remove_one(struct ib_device *device, void *client_data);
e51060f0
SH		 149
150static struct ib_client cma_client = {
151 .name = "cma",
152 .add = cma_add_one,
153 .remove = cma_remove_one
154};
155
c1a0b23b 156static struct ib_sa_client sa_client;
7a118df3 157static struct rdma_addr_client addr_client;
e51060f0
SH		 158static LIST_HEAD(dev_list);
159static LIST_HEAD(listen_any_list);
160static DEFINE_MUTEX(lock);
161static struct workqueue_struct *cma_wq;
c7d03a00 162static unsigned int cma_pernet_id;
e51060f0 163
4be74b42
HE		 164struct cma_pernet {
165 struct idr tcp_ps;
166 struct idr udp_ps;
167 struct idr ipoib_ps;
168 struct idr ib_ps;
169};
170
171static struct cma_pernet *cma_pernet(struct net *net)
172{
173 return net_generic(net, cma_pernet_id);
174}
175
176static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
aac978e1 177{
4be74b42
HE		 178 struct cma_pernet *pernet = cma_pernet(net);
179
aac978e1
HE		 180 switch (ps) {
181 case RDMA_PS_TCP:
4be74b42 182 return &pernet->tcp_ps;
aac978e1 183 case RDMA_PS_UDP:
4be74b42 184 return &pernet->udp_ps;
aac978e1 185 case RDMA_PS_IPOIB:
4be74b42 186 return &pernet->ipoib_ps;
aac978e1 187 case RDMA_PS_IB:
4be74b42 188 return &pernet->ib_ps;
aac978e1
HE		 189 default:
190 return NULL;
191 }
192}
193
e51060f0
SH		 194struct cma_device {
195 struct list_head list;
196 struct ib_device *device;
e51060f0
SH		 197 struct completion comp;
198 atomic_t refcount;
199 struct list_head id_list;
045959db 200 enum ib_gid_type *default_gid_type;
e51060f0
SH		 201};
202
e51060f0 203struct rdma_bind_list {
aac978e1 204 enum rdma_port_space ps;
e51060f0
SH		 205 struct hlist_head owners;
206 unsigned short port;
207};
208
ab15c95a
AV		 209struct class_port_info_context {
210 struct ib_class_port_info *class_port_info;
211 struct ib_device *device;
212 struct completion done;
213 struct ib_sa_query *sa_query;
214 u8 port_num;
215};
216
4be74b42 217static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
aac978e1
HE		 218 struct rdma_bind_list *bind_list, int snum)
219{
4be74b42 220 struct idr *idr = cma_pernet_idr(net, ps);
aac978e1
HE		 221
222 return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
223}
224
4be74b42
HE		 225static struct rdma_bind_list *cma_ps_find(struct net *net,
226 enum rdma_port_space ps, int snum)
aac978e1 227{
4be74b42 228 struct idr *idr = cma_pernet_idr(net, ps);
aac978e1
HE		 229
230 return idr_find(idr, snum);
231}
232
4be74b42 233static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
aac978e1 234{
4be74b42 235 struct idr *idr = cma_pernet_idr(net, ps);
aac978e1
HE		 236
237 idr_remove(idr, snum);
238}
239
68602120
SH		 240enum {
241 CMA_OPTION_AFONLY,
242};
243
218a773f
MB		 244void cma_ref_dev(struct cma_device *cma_dev)
245{
246 atomic_inc(&cma_dev->refcount);
247}
248
045959db
MB		 249struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter filter,
250 void *cookie)
251{
252 struct cma_device *cma_dev;
253 struct cma_device *found_cma_dev = NULL;
254
255 mutex_lock(&lock);
256
257 list_for_each_entry(cma_dev, &dev_list, list)
258 if (filter(cma_dev->device, cookie)) {
259 found_cma_dev = cma_dev;
260 break;
261 }
262
263 if (found_cma_dev)
264 cma_ref_dev(found_cma_dev);
265 mutex_unlock(&lock);
266 return found_cma_dev;
267}
268
269int cma_get_default_gid_type(struct cma_device *cma_dev,
270 unsigned int port)
271{
272 if (port < rdma_start_port(cma_dev->device) ||
273 port > rdma_end_port(cma_dev->device))
274 return -EINVAL;
275
276 return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
277}
278
279int cma_set_default_gid_type(struct cma_device *cma_dev,
280 unsigned int port,
281 enum ib_gid_type default_gid_type)
282{
283 unsigned long supported_gids;
284
285 if (port < rdma_start_port(cma_dev->device) ||
286 port > rdma_end_port(cma_dev->device))
287 return -EINVAL;
288
289 supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
290
291 if (!(supported_gids & 1 << default_gid_type))
292 return -EINVAL;
293
294 cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
295 default_gid_type;
296
297 return 0;
298}
299
300struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
301{
302 return cma_dev->device;
303}
304
e51060f0
SH		 305/*
306 * Device removal can occur at anytime, so we need extra handling to
307 * serialize notifying the user of device removal with other callbacks.
308 * We do this by disabling removal notification while a callback is in process,
309 * and reporting it after the callback completes.
310 */
311struct rdma_id_private {
312 struct rdma_cm_id id;
313
314 struct rdma_bind_list *bind_list;
315 struct hlist_node node;
d02d1f53
SH		 316 struct list_head list; /* listen_any_list or cma_device.list */
317 struct list_head listen_list; /* per device listens */
e51060f0 318 struct cma_device *cma_dev;
c8f6a362 319 struct list_head mc_list;
e51060f0 320
d02d1f53 321 int internal_id;
550e5ca7 322 enum rdma_cm_state state;
e51060f0 323 spinlock_t lock;
c5483388
SH		 324 struct mutex qp_mutex;
325
e51060f0
SH		 326 struct completion comp;
327 atomic_t refcount;
de910bd9 328 struct mutex handler_mutex;
e51060f0
SH		 329
330 int backlog;
331 int timeout_ms;
332 struct ib_sa_query *query;
333 int query_id;
334 union {
335 struct ib_cm_id *ib;
07ebafba 336 struct iw_cm_id *iw;
e51060f0
SH		 337 } cm_id;
338
339 u32 seq_num;
c8f6a362 340 u32 qkey;
e51060f0 341 u32 qp_num;
83e9502d 342 pid_t owner;
68602120 343 u32 options;
e51060f0 344 u8 srq;
a81c994d 345 u8 tos;
a9bb7912 346 u8 reuseaddr;
5b0ec991 347 u8 afonly;
cb57bb84 348 enum ib_gid_type gid_type;
e51060f0
SH		 349};
350
c8f6a362
SH		 351struct cma_multicast {
352 struct rdma_id_private *id_priv;
353 union {
354 struct ib_sa_multicast *ib;
355 } multicast;
356 struct list_head list;
357 void *context;
3f446754 358 struct sockaddr_storage addr;
3c86aa70 359 struct kref mcref;
bee3c3c9 360 bool igmp_joined;
ab15c95a 361 u8 join_state;
c8f6a362
SH		 362};
363
e51060f0
SH		 364struct cma_work {
365 struct work_struct work;
366 struct rdma_id_private *id;
550e5ca7
NM		 367 enum rdma_cm_state old_state;
368 enum rdma_cm_state new_state;
e51060f0
SH		 369 struct rdma_cm_event event;
370};
371
dd5bdff8
OG		 372struct cma_ndev_work {
373 struct work_struct work;
374 struct rdma_id_private *id;
375 struct rdma_cm_event event;
376};
377
3c86aa70
EC		 378struct iboe_mcast_work {
379 struct work_struct work;
380 struct rdma_id_private *id;
381 struct cma_multicast *mc;
382};
383
e51060f0
SH		 384union cma_ip_addr {
385 struct in6_addr ip6;
386 struct {
1b90c137
AV		 387 __be32 pad[3];
388 __be32 addr;
e51060f0
SH		 389 } ip4;
390};
391
392struct cma_hdr {
393 u8 cma_version;
394 u8 ip_version; /* IP version: 7:4 */
1b90c137 395 __be16 port;
e51060f0
SH		 396 union cma_ip_addr src_addr;
397 union cma_ip_addr dst_addr;
398};
399
e51060f0 400#define CMA_VERSION 0x00
e51060f0 401
4c21b5bc
HE		 402struct cma_req_info {
403 struct ib_device *device;
404 int port;
405 union ib_gid local_gid;
406 __be64 service_id;
407 u16 pkey;
408 bool has_gid:1;
409};
410
550e5ca7 411static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
e51060f0
SH		 412{
413 unsigned long flags;
414 int ret;
415
416 spin_lock_irqsave(&id_priv->lock, flags);
417 ret = (id_priv->state == comp);
418 spin_unlock_irqrestore(&id_priv->lock, flags);
419 return ret;
420}
421
422static int cma_comp_exch(struct rdma_id_private *id_priv,
550e5ca7 423 enum rdma_cm_state comp, enum rdma_cm_state exch)
e51060f0
SH		 424{
425 unsigned long flags;
426 int ret;
427
428 spin_lock_irqsave(&id_priv->lock, flags);
429 if ((ret = (id_priv->state == comp)))
430 id_priv->state = exch;
431 spin_unlock_irqrestore(&id_priv->lock, flags);
432 return ret;
433}
434
550e5ca7
NM		 435static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
436 enum rdma_cm_state exch)
e51060f0
SH		 437{
438 unsigned long flags;
550e5ca7 439 enum rdma_cm_state old;
e51060f0
SH		 440
441 spin_lock_irqsave(&id_priv->lock, flags);
442 old = id_priv->state;
443 id_priv->state = exch;
444 spin_unlock_irqrestore(&id_priv->lock, flags);
445 return old;
446}
447
4c21b5bc 448static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
e51060f0
SH		 449{
450 return hdr->ip_version >> 4;
451}
452
453static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
454{
455 hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
456}
457
bee3c3c9 458static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
e51060f0 459{
bee3c3c9
MS		 460 struct in_device *in_dev = NULL;
461
462 if (ndev) {
463 rtnl_lock();
464 in_dev = __in_dev_get_rtnl(ndev);
465 if (in_dev) {
466 if (join)
467 ip_mc_inc_group(in_dev,
468 *(__be32 *)(mgid->raw + 12));
469 else
470 ip_mc_dec_group(in_dev,
471 *(__be32 *)(mgid->raw + 12));
472 }
473 rtnl_unlock();
474 }
475 return (in_dev) ? 0 : -ENODEV;
476}
477
045959db
MB		 478static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
479 struct cma_device *cma_dev)
e51060f0 480{
218a773f 481 cma_ref_dev(cma_dev);
e51060f0 482 id_priv->cma_dev = cma_dev;
045959db 483 id_priv->gid_type = 0;
e51060f0 484 id_priv->id.device = cma_dev->device;
3c86aa70
EC		 485 id_priv->id.route.addr.dev_addr.transport =
486 rdma_node_get_transport(cma_dev->device->node_type);
e51060f0
SH		 487 list_add_tail(&id_priv->list, &cma_dev->id_list);
488}
489
045959db
MB		 490static void cma_attach_to_dev(struct rdma_id_private *id_priv,
491 struct cma_device *cma_dev)
492{
493 _cma_attach_to_dev(id_priv, cma_dev);
494 id_priv->gid_type =
495 cma_dev->default_gid_type[id_priv->id.port_num -
496 rdma_start_port(cma_dev->device)];
497}
498
218a773f 499void cma_deref_dev(struct cma_device *cma_dev)
e51060f0
SH		 500{
501 if (atomic_dec_and_test(&cma_dev->refcount))
502 complete(&cma_dev->comp);
503}
504
3c86aa70
EC		 505static inline void release_mc(struct kref *kref)
506{
507 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
508
509 kfree(mc->multicast.ib);
510 kfree(mc);
511}
512
a396d43a 513static void cma_release_dev(struct rdma_id_private *id_priv)
e51060f0 514{
a396d43a 515 mutex_lock(&lock);
e51060f0
SH		 516 list_del(&id_priv->list);
517 cma_deref_dev(id_priv->cma_dev);
518 id_priv->cma_dev = NULL;
a396d43a 519 mutex_unlock(&lock);
e51060f0
SH		 520}
521
f4753834
SH		 522static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
523{
524 return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
525}
526
527static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
528{
529 return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
530}
531
532static inline unsigned short cma_family(struct rdma_id_private *id_priv)
533{
534 return id_priv->id.route.addr.src_addr.ss_family;
535}
536
5c438135 537static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
c8f6a362
SH		 538{
539 struct ib_sa_mcmember_rec rec;
540 int ret = 0;
541
5c438135
SH		 542 if (id_priv->qkey) {
543 if (qkey && id_priv->qkey != qkey)
544 return -EINVAL;
d2ca39f2 545 return 0;
5c438135
SH		 546 }
547
548 if (qkey) {
549 id_priv->qkey = qkey;
550 return 0;
551 }
d2ca39f2
YE		 552
553 switch (id_priv->id.ps) {
c8f6a362 554 case RDMA_PS_UDP:
5c438135 555 case RDMA_PS_IB:
d2ca39f2 556 id_priv->qkey = RDMA_UDP_QKEY;
c8f6a362
SH		 557 break;
558 case RDMA_PS_IPOIB:
d2ca39f2
YE		 559 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
560 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
561 id_priv->id.port_num, &rec.mgid,
562 &rec);
563 if (!ret)
564 id_priv->qkey = be32_to_cpu(rec.qkey);
c8f6a362
SH		 565 break;
566 default:
567 break;
568 }
569 return ret;
570}
571
680f920a
SH		 572static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
573{
574 dev_addr->dev_type = ARPHRD_INFINIBAND;
575 rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
576 ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
577}
578
579static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
580{
581 int ret;
582
583 if (addr->sa_family != AF_IB) {
dd5f03be 584 ret = rdma_translate_ip(addr, dev_addr, NULL);
680f920a
SH		 585 } else {
586 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
587 ret = 0;
588 }
589
590 return ret;
591}
592
7c11147d 593static inline int cma_validate_port(struct ib_device *device, u8 port,
045959db 594 enum ib_gid_type gid_type,
abae1b71
MB		 595 union ib_gid *gid, int dev_type,
596 int bound_if_index)
7c11147d 597{
7c11147d 598 int ret = -ENODEV;
abae1b71 599 struct net_device *ndev = NULL;
7c11147d
MW		 600
601 if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
602 return ret;
603
604 if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
605 return ret;
606
64936773 607 if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
abae1b71 608 ndev = dev_get_by_index(&init_net, bound_if_index);
20029832
MB		 609 if (ndev && ndev->flags & IFF_LOOPBACK) {
610 pr_info("detected loopback device\n");
611 dev_put(ndev);
abae1b71 612
20029832
MB		 613 if (!device->get_netdev)
614 return -EOPNOTSUPP;
615
616 ndev = device->get_netdev(device, port);
617 if (!ndev)
618 return -ENODEV;
619 }
045959db
MB		 620 } else {
621 gid_type = IB_GID_TYPE_IB;
20029832 622 }
abae1b71 623
045959db 624 ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
b39ffa1d 625 ndev, NULL);
abae1b71
MB		 626
627 if (ndev)
628 dev_put(ndev);
7c11147d
MW		 629
630 return ret;
631}
632
be9130cc
DL		 633static int cma_acquire_dev(struct rdma_id_private *id_priv,
634 struct rdma_id_private *listen_id_priv)
e51060f0 635{
c8f6a362 636 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
e51060f0 637 struct cma_device *cma_dev;
7c11147d 638 union ib_gid gid, iboe_gid, *gidp;
e51060f0 639 int ret = -ENODEV;
7c11147d 640 u8 port;
e51060f0 641
7c11147d 642 if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
2efdd6a0
MS		 643 id_priv->id.ps == RDMA_PS_IPOIB)
644 return -EINVAL;
645
a396d43a 646 mutex_lock(&lock);
7b85627b
MS		 647 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
648 &iboe_gid);
649
3c86aa70
EC		 650 memcpy(&gid, dev_addr->src_dev_addr +
651 rdma_addr_gid_offset(dev_addr), sizeof gid);
7c11147d
MW		 652
653 if (listen_id_priv) {
be9130cc
DL		 654 cma_dev = listen_id_priv->cma_dev;
655 port = listen_id_priv->id.port_num;
5d9fb044 656 gidp = rdma_protocol_roce(cma_dev->device, port) ?
7c11147d 657 &iboe_gid : &gid;
be9130cc 658
045959db
MB		 659 ret = cma_validate_port(cma_dev->device, port,
660 rdma_protocol_ib(cma_dev->device, port) ?
661 IB_GID_TYPE_IB :
662 listen_id_priv->gid_type, gidp,
abae1b71
MB		 663 dev_addr->dev_type,
664 dev_addr->bound_dev_if);
7c11147d
MW		 665 if (!ret) {
666 id_priv->id.port_num = port;
be9130cc
DL		 667 goto out;
668 }
669 }
7c11147d 670
e51060f0 671 list_for_each_entry(cma_dev, &dev_list, list) {
3c86aa70 672 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
be9130cc
DL		 673 if (listen_id_priv &&
674 listen_id_priv->cma_dev == cma_dev &&
675 listen_id_priv->id.port_num == port)
676 continue;
7c11147d 677
5d9fb044 678 gidp = rdma_protocol_roce(cma_dev->device, port) ?
7c11147d
MW		 679 &iboe_gid : &gid;
680
045959db
MB		 681 ret = cma_validate_port(cma_dev->device, port,
682 rdma_protocol_ib(cma_dev->device, port) ?
683 IB_GID_TYPE_IB :
684 cma_dev->default_gid_type[port - 1],
685 gidp, dev_addr->dev_type,
abae1b71 686 dev_addr->bound_dev_if);
7c11147d
MW		 687 if (!ret) {
688 id_priv->id.port_num = port;
689 goto out;
3c86aa70 690 }
e51060f0
SH		 691 }
692 }
3c86aa70
EC		 693
694out:
695 if (!ret)
696 cma_attach_to_dev(id_priv, cma_dev);
697
a396d43a 698 mutex_unlock(&lock);
e51060f0
SH		 699 return ret;
700}
701
f17df3b0
SH		 702/*
703 * Select the source IB device and address to reach the destination IB address.
704 */
705static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
706{
707 struct cma_device *cma_dev, *cur_dev;
708 struct sockaddr_ib *addr;
709 union ib_gid gid, sgid, *dgid;
710 u16 pkey, index;
8fb488d7 711 u8 p;
f17df3b0
SH		 712 int i;
713
714 cma_dev = NULL;
715 addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
716 dgid = (union ib_gid *) &addr->sib_addr;
717 pkey = ntohs(addr->sib_pkey);
718
719 list_for_each_entry(cur_dev, &dev_list, list) {
f17df3b0 720 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
30a74ef4 721 if (!rdma_cap_af_ib(cur_dev->device, p))
fef60902
MW		 722 continue;
723
f17df3b0
SH		 724 if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
725 continue;
726
55ee3ab2
MB		 727 for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
728 &gid, NULL);
729 i++) {
f17df3b0
SH		 730 if (!memcmp(&gid, dgid, sizeof(gid))) {
731 cma_dev = cur_dev;
732 sgid = gid;
8fb488d7 733 id_priv->id.port_num = p;
f17df3b0
SH		 734 goto found;
735 }
736
737 if (!cma_dev && (gid.global.subnet_prefix ==
738 dgid->global.subnet_prefix)) {
739 cma_dev = cur_dev;
740 sgid = gid;
8fb488d7 741 id_priv->id.port_num = p;
f17df3b0
SH		 742 }
743 }
744 }
745 }
746
747 if (!cma_dev)
748 return -ENODEV;
749
750found:
751 cma_attach_to_dev(id_priv, cma_dev);
f17df3b0
SH		 752 addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
753 memcpy(&addr->sib_addr, &sgid, sizeof sgid);
754 cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
755 return 0;
756}
757
e51060f0
SH		 758static void cma_deref_id(struct rdma_id_private *id_priv)
759{
760 if (atomic_dec_and_test(&id_priv->refcount))
761 complete(&id_priv->comp);
762}
763
fa20105e
GS		 764struct rdma_cm_id *rdma_create_id(struct net *net,
765 rdma_cm_event_handler event_handler,
b26f9b99
SH		 766 void *context, enum rdma_port_space ps,
767 enum ib_qp_type qp_type)
e51060f0
SH		 768{
769 struct rdma_id_private *id_priv;
770
771 id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
772 if (!id_priv)
773 return ERR_PTR(-ENOMEM);
774
83e9502d 775 id_priv->owner = task_pid_nr(current);
550e5ca7 776 id_priv->state = RDMA_CM_IDLE;
e51060f0
SH		 777 id_priv->id.context = context;
778 id_priv->id.event_handler = event_handler;
779 id_priv->id.ps = ps;
b26f9b99 780 id_priv->id.qp_type = qp_type;
e51060f0 781 spin_lock_init(&id_priv->lock);
c5483388 782 mutex_init(&id_priv->qp_mutex);
e51060f0
SH		 783 init_completion(&id_priv->comp);
784 atomic_set(&id_priv->refcount, 1);
de910bd9 785 mutex_init(&id_priv->handler_mutex);
e51060f0 786 INIT_LIST_HEAD(&id_priv->listen_list);
c8f6a362 787 INIT_LIST_HEAD(&id_priv->mc_list);
e51060f0 788 get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
fa20105e 789 id_priv->id.route.addr.dev_addr.net = get_net(net);
e51060f0
SH		 790
791 return &id_priv->id;
792}
793EXPORT_SYMBOL(rdma_create_id);
794
c8f6a362 795static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
e51060f0
SH		 796{
797 struct ib_qp_attr qp_attr;
c8f6a362 798 int qp_attr_mask, ret;
e51060f0 799
c8f6a362
SH		 800 qp_attr.qp_state = IB_QPS_INIT;
801 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0
SH		 802 if (ret)
803 return ret;
804
c8f6a362
SH		 805 ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
806 if (ret)
807 return ret;
808
809 qp_attr.qp_state = IB_QPS_RTR;
810 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
811 if (ret)
812 return ret;
813
814 qp_attr.qp_state = IB_QPS_RTS;
815 qp_attr.sq_psn = 0;
816 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
817
818 return ret;
e51060f0
SH		 819}
820
c8f6a362 821static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
07ebafba
TT		 822{
823 struct ib_qp_attr qp_attr;
c8f6a362 824 int qp_attr_mask, ret;
07ebafba
TT		 825
826 qp_attr.qp_state = IB_QPS_INIT;
c8f6a362
SH		 827 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
828 if (ret)
829 return ret;
07ebafba 830
c8f6a362 831 return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
07ebafba
TT		 832}
833
e51060f0
SH		 834int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
835 struct ib_qp_init_attr *qp_init_attr)
836{
837 struct rdma_id_private *id_priv;
838 struct ib_qp *qp;
839 int ret;
840
841 id_priv = container_of(id, struct rdma_id_private, id);
842 if (id->device != pd->device)
843 return -EINVAL;
844
0691a286 845 qp_init_attr->port_num = id->port_num;
e51060f0
SH		 846 qp = ib_create_qp(pd, qp_init_attr);
847 if (IS_ERR(qp))
848 return PTR_ERR(qp);
849
b26f9b99 850 if (id->qp_type == IB_QPT_UD)
c8f6a362
SH		 851 ret = cma_init_ud_qp(id_priv, qp);
852 else
853 ret = cma_init_conn_qp(id_priv, qp);
e51060f0
SH		 854 if (ret)
855 goto err;
856
857 id->qp = qp;
858 id_priv->qp_num = qp->qp_num;
e51060f0
SH		 859 id_priv->srq = (qp->srq != NULL);
860 return 0;
861err:
862 ib_destroy_qp(qp);
863 return ret;
864}
865EXPORT_SYMBOL(rdma_create_qp);
866
867void rdma_destroy_qp(struct rdma_cm_id *id)
868{
c5483388
SH		 869 struct rdma_id_private *id_priv;
870
871 id_priv = container_of(id, struct rdma_id_private, id);
872 mutex_lock(&id_priv->qp_mutex);
873 ib_destroy_qp(id_priv->id.qp);
874 id_priv->id.qp = NULL;
875 mutex_unlock(&id_priv->qp_mutex);
e51060f0
SH		 876}
877EXPORT_SYMBOL(rdma_destroy_qp);
878
5851bb89
SH		 879static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
880 struct rdma_conn_param *conn_param)
e51060f0
SH		 881{
882 struct ib_qp_attr qp_attr;
883 int qp_attr_mask, ret;
dd5f03be 884 union ib_gid sgid;
e51060f0 885
c5483388
SH		 886 mutex_lock(&id_priv->qp_mutex);
887 if (!id_priv->id.qp) {
888 ret = 0;
889 goto out;
890 }
e51060f0
SH		 891
892 /* Need to update QP attributes from default values. */
893 qp_attr.qp_state = IB_QPS_INIT;
c5483388 894 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 895 if (ret)
c5483388 896 goto out;
e51060f0 897
c5483388 898 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
e51060f0 899 if (ret)
c5483388 900 goto out;
e51060f0
SH		 901
902 qp_attr.qp_state = IB_QPS_RTR;
c5483388 903 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 904 if (ret)
c5483388 905 goto out;
e51060f0 906
dd5f03be 907 ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
55ee3ab2 908 qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
dd5f03be
MB		 909 if (ret)
910 goto out;
911
fef60902
MW		 912 BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
913
5851bb89
SH		 914 if (conn_param)
915 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
c5483388
SH		 916 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
917out:
918 mutex_unlock(&id_priv->qp_mutex);
919 return ret;
e51060f0
SH		 920}
921
5851bb89
SH		 922static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
923 struct rdma_conn_param *conn_param)
e51060f0
SH		 924{
925 struct ib_qp_attr qp_attr;
926 int qp_attr_mask, ret;
927
c5483388
SH		 928 mutex_lock(&id_priv->qp_mutex);
929 if (!id_priv->id.qp) {
930 ret = 0;
931 goto out;
932 }
e51060f0
SH		 933
934 qp_attr.qp_state = IB_QPS_RTS;
c5483388 935 ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
e51060f0 936 if (ret)
c5483388 937 goto out;
e51060f0 938
5851bb89
SH		 939 if (conn_param)
940 qp_attr.max_rd_atomic = conn_param->initiator_depth;
c5483388
SH		 941 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
942out:
943 mutex_unlock(&id_priv->qp_mutex);
944 return ret;
e51060f0
SH		 945}
946
c5483388 947static int cma_modify_qp_err(struct rdma_id_private *id_priv)
e51060f0
SH		 948{
949 struct ib_qp_attr qp_attr;
c5483388 950 int ret;
e51060f0 951
c5483388
SH		 952 mutex_lock(&id_priv->qp_mutex);
953 if (!id_priv->id.qp) {
954 ret = 0;
955 goto out;
956 }
e51060f0
SH		 957
958 qp_attr.qp_state = IB_QPS_ERR;
c5483388
SH		 959 ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
960out:
961 mutex_unlock(&id_priv->qp_mutex);
962 return ret;
e51060f0
SH		 963}
964
c8f6a362
SH		 965static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
966 struct ib_qp_attr *qp_attr, int *qp_attr_mask)
967{
968 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
969 int ret;
3c86aa70
EC		 970 u16 pkey;
971
227128fc 972 if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
3c86aa70 973 pkey = 0xffff;
fef60902
MW		 974 else
975 pkey = ib_addr_get_pkey(dev_addr);
c8f6a362
SH		 976
977 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
3c86aa70 978 pkey, &qp_attr->pkey_index);
c8f6a362
SH		 979 if (ret)
980 return ret;
981
982 qp_attr->port_num = id_priv->id.port_num;
983 *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
984
b26f9b99 985 if (id_priv->id.qp_type == IB_QPT_UD) {
5c438135 986 ret = cma_set_qkey(id_priv, 0);
d2ca39f2
YE		 987 if (ret)
988 return ret;
989
c8f6a362
SH		 990 qp_attr->qkey = id_priv->qkey;
991 *qp_attr_mask |= IB_QP_QKEY;
992 } else {
993 qp_attr->qp_access_flags = 0;
994 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
995 }
996 return 0;
997}
998
e51060f0
SH		 999int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1000 int *qp_attr_mask)
1001{
1002 struct rdma_id_private *id_priv;
c8f6a362 1003 int ret = 0;
e51060f0
SH		 1004
1005 id_priv = container_of(id, struct rdma_id_private, id);
72219cea 1006 if (rdma_cap_ib_cm(id->device, id->port_num)) {
b26f9b99 1007 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
c8f6a362
SH		 1008 ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1009 else
1010 ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1011 qp_attr_mask);
dd5f03be 1012
e51060f0
SH		 1013 if (qp_attr->qp_state == IB_QPS_RTR)
1014 qp_attr->rq_psn = id_priv->seq_num;
04215330 1015 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
c8f6a362 1016 if (!id_priv->cm_id.iw) {
8f076531 1017 qp_attr->qp_access_flags = 0;
c8f6a362
SH		 1018 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1019 } else
1020 ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1021 qp_attr_mask);
21655afc 1022 } else
e51060f0 1023 ret = -ENOSYS;
e51060f0
SH		 1024
1025 return ret;
1026}
1027EXPORT_SYMBOL(rdma_init_qp_attr);
1028
1029static inline int cma_zero_addr(struct sockaddr *addr)
1030{
2e2d190c
SH		 1031 switch (addr->sa_family) {
1032 case AF_INET:
1033 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1034 case AF_INET6:
1035 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1036 case AF_IB:
1037 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1038 default:
1039 return 0;
e51060f0
SH		 1040 }
1041}
1042
1043static inline int cma_loopback_addr(struct sockaddr *addr)
1044{
2e2d190c
SH		 1045 switch (addr->sa_family) {
1046 case AF_INET:
1047 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1048 case AF_INET6:
1049 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1050 case AF_IB:
1051 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1052 default:
1053 return 0;
1054 }
e51060f0
SH		 1055}
1056
1057static inline int cma_any_addr(struct sockaddr *addr)
1058{
1059 return cma_zero_addr(addr) || cma_loopback_addr(addr);
1060}
1061
43b752da
HS		 1062static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1063{
1064 if (src->sa_family != dst->sa_family)
1065 return -1;
1066
1067 switch (src->sa_family) {
1068 case AF_INET:
1069 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1070 ((struct sockaddr_in *) dst)->sin_addr.s_addr;
2e2d190c 1071 case AF_INET6:
43b752da
HS		 1072 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1073 &((struct sockaddr_in6 *) dst)->sin6_addr);
2e2d190c
SH		 1074 default:
1075 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1076 &((struct sockaddr_ib *) dst)->sib_addr);
43b752da
HS		 1077 }
1078}
1079
58afdcb7 1080static __be16 cma_port(struct sockaddr *addr)
628e5f6d 1081{
58afdcb7
SH		 1082 struct sockaddr_ib *sib;
1083
1084 switch (addr->sa_family) {
1085 case AF_INET:
628e5f6d 1086 return ((struct sockaddr_in *) addr)->sin_port;
58afdcb7 1087 case AF_INET6:
628e5f6d 1088 return ((struct sockaddr_in6 *) addr)->sin6_port;
58afdcb7
SH		 1089 case AF_IB:
1090 sib = (struct sockaddr_ib *) addr;
1091 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1092 be64_to_cpu(sib->sib_sid_mask)));
1093 default:
1094 return 0;
1095 }
628e5f6d
SH		 1096}
1097
e51060f0
SH		 1098static inline int cma_any_port(struct sockaddr *addr)
1099{
628e5f6d 1100 return !cma_port(addr);
e51060f0
SH		 1101}
1102
0c505f70
HE		 1103static void cma_save_ib_info(struct sockaddr *src_addr,
1104 struct sockaddr *dst_addr,
1105 struct rdma_cm_id *listen_id,
fbaa1a6d 1106 struct ib_sa_path_rec *path)
e51060f0 1107{
fbaa1a6d
SH		 1108 struct sockaddr_ib *listen_ib, *ib;
1109
1110 listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
0c505f70
HE		 1111 if (src_addr) {
1112 ib = (struct sockaddr_ib *)src_addr;
1113 ib->sib_family = AF_IB;
1114 if (path) {
1115 ib->sib_pkey = path->pkey;
1116 ib->sib_flowinfo = path->flow_label;
1117 memcpy(&ib->sib_addr, &path->sgid, 16);
1118 ib->sib_sid = path->service_id;
1119 ib->sib_scope_id = 0;
1120 } else {
1121 ib->sib_pkey = listen_ib->sib_pkey;
1122 ib->sib_flowinfo = listen_ib->sib_flowinfo;
1123 ib->sib_addr = listen_ib->sib_addr;
1124 ib->sib_sid = listen_ib->sib_sid;
1125 ib->sib_scope_id = listen_ib->sib_scope_id;
1126 }
1127 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
c07678bb 1128 }
0c505f70
HE		 1129 if (dst_addr) {
1130 ib = (struct sockaddr_ib *)dst_addr;
1131 ib->sib_family = AF_IB;
1132 if (path) {
1133 ib->sib_pkey = path->pkey;
1134 ib->sib_flowinfo = path->flow_label;
1135 memcpy(&ib->sib_addr, &path->dgid, 16);
1136 }
c07678bb 1137 }
fbaa1a6d 1138}
e51060f0 1139
c50e90d0
AB		 1140static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1141 struct sockaddr_in *dst_addr,
0c505f70
HE		 1142 struct cma_hdr *hdr,
1143 __be16 local_port)
fbaa1a6d 1144{
0c505f70 1145 if (src_addr) {
c50e90d0
AB		 1146 *src_addr = (struct sockaddr_in) {
1147 .sin_family = AF_INET,
1148 .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1149 .sin_port = local_port,
1150 };
0c505f70 1151 }
fbaa1a6d 1152
0c505f70 1153 if (dst_addr) {
c50e90d0
AB		 1154 *dst_addr = (struct sockaddr_in) {
1155 .sin_family = AF_INET,
1156 .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1157 .sin_port = hdr->port,
1158 };
0c505f70 1159 }
e51060f0
SH		 1160}
1161
c50e90d0
AB		 1162static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1163 struct sockaddr_in6 *dst_addr,
0c505f70
HE		 1164 struct cma_hdr *hdr,
1165 __be16 local_port)
e51060f0 1166{
0c505f70 1167 if (src_addr) {
c50e90d0
AB		 1168 *src_addr = (struct sockaddr_in6) {
1169 .sin6_family = AF_INET6,
1170 .sin6_addr = hdr->dst_addr.ip6,
1171 .sin6_port = local_port,
1172 };
0c505f70 1173 }
fbaa1a6d 1174
0c505f70 1175 if (dst_addr) {
c50e90d0
AB		 1176 *dst_addr = (struct sockaddr_in6) {
1177 .sin6_family = AF_INET6,
1178 .sin6_addr = hdr->src_addr.ip6,
1179 .sin6_port = hdr->port,
1180 };
0c505f70 1181 }
fbaa1a6d
SH		 1182}
1183
0c505f70 1184static u16 cma_port_from_service_id(__be64 service_id)
fbaa1a6d 1185{
0c505f70
HE		 1186 return (u16)be64_to_cpu(service_id);
1187}
fbaa1a6d 1188
0c505f70
HE		 1189static int cma_save_ip_info(struct sockaddr *src_addr,
1190 struct sockaddr *dst_addr,
1191 struct ib_cm_event *ib_event,
1192 __be64 service_id)
1193{
1194 struct cma_hdr *hdr;
1195 __be16 port;
fbaa1a6d
SH		 1196
1197 hdr = ib_event->private_data;
1198 if (hdr->cma_version != CMA_VERSION)
1199 return -EINVAL;
1200
0c505f70
HE		 1201 port = htons(cma_port_from_service_id(service_id));
1202
fbaa1a6d 1203 switch (cma_get_ip_ver(hdr)) {
e51060f0 1204 case 4:
c50e90d0
AB		 1205 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1206 (struct sockaddr_in *)dst_addr, hdr, port);
e51060f0
SH		 1207 break;
1208 case 6:
c50e90d0
AB		 1209 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1210 (struct sockaddr_in6 *)dst_addr, hdr, port);
e51060f0
SH		 1211 break;
1212 default:
4c21b5bc 1213 return -EAFNOSUPPORT;
e51060f0 1214 }
0c505f70 1215
fbaa1a6d 1216 return 0;
e51060f0
SH		 1217}
1218
0c505f70
HE		 1219static int cma_save_net_info(struct sockaddr *src_addr,
1220 struct sockaddr *dst_addr,
1221 struct rdma_cm_id *listen_id,
1222 struct ib_cm_event *ib_event,
1223 sa_family_t sa_family, __be64 service_id)
1224{
1225 if (sa_family == AF_IB) {
1226 if (ib_event->event == IB_CM_REQ_RECEIVED)
1227 cma_save_ib_info(src_addr, dst_addr, listen_id,
1228 ib_event->param.req_rcvd.primary_path);
1229 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1230 cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1231 return 0;
1232 }
1233
1234 return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1235}
1236
4c21b5bc
HE		 1237static int cma_save_req_info(const struct ib_cm_event *ib_event,
1238 struct cma_req_info *req)
1239{
1240 const struct ib_cm_req_event_param *req_param =
1241 &ib_event->param.req_rcvd;
1242 const struct ib_cm_sidr_req_event_param *sidr_param =
1243 &ib_event->param.sidr_req_rcvd;
1244
1245 switch (ib_event->event) {
1246 case IB_CM_REQ_RECEIVED:
1247 req->device = req_param->listen_id->device;
1248 req->port = req_param->port;
1249 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1250 sizeof(req->local_gid));
1251 req->has_gid = true;
1252 req->service_id = req_param->primary_path->service_id;
ab3964ad 1253 req->pkey = be16_to_cpu(req_param->primary_path->pkey);
84424a7f
HE		 1254 if (req->pkey != req_param->bth_pkey)
1255 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1256 "RDMA CMA: in the future this may cause the request to be dropped\n",
1257 req_param->bth_pkey, req->pkey);
4c21b5bc
HE		 1258 break;
1259 case IB_CM_SIDR_REQ_RECEIVED:
1260 req->device = sidr_param->listen_id->device;
1261 req->port = sidr_param->port;
1262 req->has_gid = false;
1263 req->service_id = sidr_param->service_id;
ab3964ad 1264 req->pkey = sidr_param->pkey;
84424a7f
HE		 1265 if (req->pkey != sidr_param->bth_pkey)
1266 pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1267 "RDMA CMA: in the future this may cause the request to be dropped\n",
1268 sidr_param->bth_pkey, req->pkey);
4c21b5bc
HE		 1269 break;
1270 default:
1271 return -EINVAL;
1272 }
1273
1274 return 0;
1275}
1276
f887f2ac
HE		 1277static bool validate_ipv4_net_dev(struct net_device *net_dev,
1278 const struct sockaddr_in *dst_addr,
1279 const struct sockaddr_in *src_addr)
1280{
1281 __be32 daddr = dst_addr->sin_addr.s_addr,
1282 saddr = src_addr->sin_addr.s_addr;
1283 struct fib_result res;
1284 struct flowi4 fl4;
1285 int err;
1286 bool ret;
1287
1288 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1289 ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1290 ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1291 ipv4_is_loopback(saddr))
1292 return false;
1293
1294 memset(&fl4, 0, sizeof(fl4));
1295 fl4.flowi4_iif = net_dev->ifindex;
1296 fl4.daddr = daddr;
1297 fl4.saddr = saddr;
1298
1299 rcu_read_lock();
1300 err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
d3632493 1301 ret = err == 0 && FIB_RES_DEV(res) == net_dev;
f887f2ac
HE		 1302 rcu_read_unlock();
1303
1304 return ret;
1305}
1306
1307static bool validate_ipv6_net_dev(struct net_device *net_dev,
1308 const struct sockaddr_in6 *dst_addr,
1309 const struct sockaddr_in6 *src_addr)
1310{
1311#if IS_ENABLED(CONFIG_IPV6)
1312 const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1313 IPV6_ADDR_LINKLOCAL;
1314 struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1315 &src_addr->sin6_addr, net_dev->ifindex,
1316 strict);
1317 bool ret;
1318
1319 if (!rt)
1320 return false;
1321
1322 ret = rt->rt6i_idev->dev == net_dev;
1323 ip6_rt_put(rt);
1324
1325 return ret;
1326#else
1327 return false;
1328#endif
1329}
1330
1331static bool validate_net_dev(struct net_device *net_dev,
1332 const struct sockaddr *daddr,
1333 const struct sockaddr *saddr)
1334{
1335 const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1336 const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1337 const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1338 const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1339
1340 switch (daddr->sa_family) {
1341 case AF_INET:
1342 return saddr->sa_family == AF_INET &&
1343 validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1344
1345 case AF_INET6:
1346 return saddr->sa_family == AF_INET6 &&
1347 validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1348
1349 default:
1350 return false;
1351 }
1352}
1353
4c21b5bc
HE		 1354static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1355 const struct cma_req_info *req)
1356{
f887f2ac
HE		 1357 struct sockaddr_storage listen_addr_storage, src_addr_storage;
1358 struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1359 *src_addr = (struct sockaddr *)&src_addr_storage;
4c21b5bc
HE		 1360 struct net_device *net_dev;
1361 const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1362 int err;
1363
f887f2ac
HE		 1364 err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1365 req->service_id);
4c21b5bc
HE		 1366 if (err)
1367 return ERR_PTR(err);
1368
1369 net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1370 gid, listen_addr);
1371 if (!net_dev)
1372 return ERR_PTR(-ENODEV);
1373
f887f2ac
HE		 1374 if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1375 dev_put(net_dev);
1376 return ERR_PTR(-EHOSTUNREACH);
1377 }
1378
4c21b5bc
HE		 1379 return net_dev;
1380}
1381
1382static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1383{
1384 return (be64_to_cpu(service_id) >> 16) & 0xffff;
1385}
1386
1387static bool cma_match_private_data(struct rdma_id_private *id_priv,
1388 const struct cma_hdr *hdr)
1389{
1390 struct sockaddr *addr = cma_src_addr(id_priv);
1391 __be32 ip4_addr;
1392 struct in6_addr ip6_addr;
1393
1394 if (cma_any_addr(addr) && !id_priv->afonly)
1395 return true;
1396
1397 switch (addr->sa_family) {
1398 case AF_INET:
1399 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1400 if (cma_get_ip_ver(hdr) != 4)
1401 return false;
1402 if (!cma_any_addr(addr) &&
1403 hdr->dst_addr.ip4.addr != ip4_addr)
1404 return false;
1405 break;
1406 case AF_INET6:
1407 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1408 if (cma_get_ip_ver(hdr) != 6)
1409 return false;
1410 if (!cma_any_addr(addr) &&
1411 memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1412 return false;
1413 break;
1414 case AF_IB:
1415 return true;
1416 default:
1417 return false;
1418 }
1419
1420 return true;
1421}
1422
b8cab5da
HE		 1423static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1424{
1425 enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1426 enum rdma_transport_type transport =
1427 rdma_node_get_transport(device->node_type);
1428
1429 return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1430}
1431
1432static bool cma_protocol_roce(const struct rdma_cm_id *id)
1433{
1434 struct ib_device *device = id->device;
1435 const int port_num = id->port_num ?: rdma_start_port(device);
1436
1437 return cma_protocol_roce_dev_port(device, port_num);
1438}
1439
fac51590
MB		 1440static bool cma_match_net_dev(const struct rdma_cm_id *id,
1441 const struct net_device *net_dev,
1442 u8 port_num)
4c21b5bc 1443{
fac51590 1444 const struct rdma_addr *addr = &id->route.addr;
4c21b5bc
HE		 1445
1446 if (!net_dev)
b8cab5da 1447 /* This request is an AF_IB request or a RoCE request */
fac51590
MB		 1448 return (!id->port_num || id->port_num == port_num) &&
1449 (addr->src_addr.ss_family == AF_IB ||
1450 cma_protocol_roce_dev_port(id->device, port_num));
4c21b5bc
HE		 1451
1452 return !addr->dev_addr.bound_dev_if ||
fa20105e 1453 (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
4c21b5bc
HE		 1454 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1455}
1456
1457static struct rdma_id_private *cma_find_listener(
1458 const struct rdma_bind_list *bind_list,
1459 const struct ib_cm_id *cm_id,
1460 const struct ib_cm_event *ib_event,
1461 const struct cma_req_info *req,
1462 const struct net_device *net_dev)
1463{
1464 struct rdma_id_private *id_priv, *id_priv_dev;
1465
1466 if (!bind_list)
1467 return ERR_PTR(-EINVAL);
1468
1469 hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1470 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1471 if (id_priv->id.device == cm_id->device &&
fac51590 1472 cma_match_net_dev(&id_priv->id, net_dev, req->port))
4c21b5bc
HE		 1473 return id_priv;
1474 list_for_each_entry(id_priv_dev,
1475 &id_priv->listen_list,
1476 listen_list) {
1477 if (id_priv_dev->id.device == cm_id->device &&
fac51590 1478 cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
4c21b5bc
HE		 1479 return id_priv_dev;
1480 }
1481 }
1482 }
1483
1484 return ERR_PTR(-EINVAL);
1485}
1486
1487static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
0b3ca768
HE		 1488 struct ib_cm_event *ib_event,
1489 struct net_device **net_dev)
4c21b5bc
HE		 1490{
1491 struct cma_req_info req;
1492 struct rdma_bind_list *bind_list;
1493 struct rdma_id_private *id_priv;
4c21b5bc
HE		 1494 int err;
1495
1496 err = cma_save_req_info(ib_event, &req);
1497 if (err)
1498 return ERR_PTR(err);
1499
0b3ca768
HE		 1500 *net_dev = cma_get_net_dev(ib_event, &req);
1501 if (IS_ERR(*net_dev)) {
1502 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
4c21b5bc 1503 /* Assuming the protocol is AF_IB */
0b3ca768 1504 *net_dev = NULL;
b8cab5da
HE		 1505 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1506 /* TODO find the net dev matching the request parameters
1507 * through the RoCE GID table */
1508 *net_dev = NULL;
4c21b5bc 1509 } else {
0b3ca768 1510 return ERR_CAST(*net_dev);
4c21b5bc
HE		 1511 }
1512 }
1513
fa20105e 1514 bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
4be74b42 1515 rdma_ps_from_service_id(req.service_id),
4c21b5bc 1516 cma_port_from_service_id(req.service_id));
0b3ca768 1517 id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
b3b51f9f 1518 if (IS_ERR(id_priv) && *net_dev) {
be688195
HE		 1519 dev_put(*net_dev);
1520 *net_dev = NULL;
1521 }
4c21b5bc
HE		 1522
1523 return id_priv;
1524}
1525
e8160e15 1526static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
e51060f0 1527{
e8160e15 1528 return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
e51060f0
SH		 1529}
1530
e51060f0
SH		 1531static void cma_cancel_route(struct rdma_id_private *id_priv)
1532{
fe53ba2f 1533 if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
e51060f0
SH		 1534 if (id_priv->query)
1535 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
e51060f0
SH		 1536 }
1537}
1538
e51060f0
SH		 1539static void cma_cancel_listens(struct rdma_id_private *id_priv)
1540{
1541 struct rdma_id_private *dev_id_priv;
1542
d02d1f53
SH		 1543 /*
1544 * Remove from listen_any_list to prevent added devices from spawning
1545 * additional listen requests.
1546 */
e51060f0
SH		 1547 mutex_lock(&lock);
1548 list_del(&id_priv->list);
1549
1550 while (!list_empty(&id_priv->listen_list)) {
1551 dev_id_priv = list_entry(id_priv->listen_list.next,
1552 struct rdma_id_private, listen_list);
d02d1f53
SH		 1553 /* sync with device removal to avoid duplicate destruction */
1554 list_del_init(&dev_id_priv->list);
1555 list_del(&dev_id_priv->listen_list);
1556 mutex_unlock(&lock);
1557
1558 rdma_destroy_id(&dev_id_priv->id);
1559 mutex_lock(&lock);
e51060f0
SH		 1560 }
1561 mutex_unlock(&lock);
1562}
1563
1564static void cma_cancel_operation(struct rdma_id_private *id_priv,
550e5ca7 1565 enum rdma_cm_state state)
e51060f0
SH		 1566{
1567 switch (state) {
550e5ca7 1568 case RDMA_CM_ADDR_QUERY:
e51060f0
SH		 1569 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1570 break;
550e5ca7 1571 case RDMA_CM_ROUTE_QUERY:
e51060f0
SH		 1572 cma_cancel_route(id_priv);
1573 break;
550e5ca7 1574 case RDMA_CM_LISTEN:
f4753834 1575 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
e51060f0
SH		 1576 cma_cancel_listens(id_priv);
1577 break;
1578 default:
1579 break;
1580 }
1581}
1582
1583static void cma_release_port(struct rdma_id_private *id_priv)
1584{
1585 struct rdma_bind_list *bind_list = id_priv->bind_list;
fa20105e 1586 struct net *net = id_priv->id.route.addr.dev_addr.net;
e51060f0
SH		 1587
1588 if (!bind_list)
1589 return;
1590
1591 mutex_lock(&lock);
1592 hlist_del(&id_priv->node);
1593 if (hlist_empty(&bind_list->owners)) {
fa20105e 1594 cma_ps_remove(net, bind_list->ps, bind_list->port);
e51060f0
SH		 1595 kfree(bind_list);
1596 }
1597 mutex_unlock(&lock);
1598}
1599
c8f6a362
SH		 1600static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1601{
1602 struct cma_multicast *mc;
1603
1604 while (!list_empty(&id_priv->mc_list)) {
1605 mc = container_of(id_priv->mc_list.next,
1606 struct cma_multicast, list);
1607 list_del(&mc->list);
a31ad3b0 1608 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
5c9a5282 1609 id_priv->id.port_num)) {
3c86aa70
EC		 1610 ib_sa_free_multicast(mc->multicast.ib);
1611 kfree(mc);
bee3c3c9
MS		 1612 } else {
1613 if (mc->igmp_joined) {
1614 struct rdma_dev_addr *dev_addr =
1615 &id_priv->id.route.addr.dev_addr;
1616 struct net_device *ndev = NULL;
1617
1618 if (dev_addr->bound_dev_if)
1619 ndev = dev_get_by_index(&init_net,
1620 dev_addr->bound_dev_if);
1621 if (ndev) {
1622 cma_igmp_send(ndev,
1623 &mc->multicast.ib->rec.mgid,
1624 false);
1625 dev_put(ndev);
1626 }
1627 }
3c86aa70 1628 kref_put(&mc->mcref, release_mc);
bee3c3c9 1629 }
c8f6a362
SH		 1630 }
1631}
1632
e51060f0
SH		 1633void rdma_destroy_id(struct rdma_cm_id *id)
1634{
1635 struct rdma_id_private *id_priv;
550e5ca7 1636 enum rdma_cm_state state;
e51060f0
SH		 1637
1638 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 1639 state = cma_exch(id_priv, RDMA_CM_DESTROYING);
e51060f0
SH		 1640 cma_cancel_operation(id_priv, state);
1641
a396d43a
SH		 1642 /*
1643 * Wait for any active callback to finish. New callbacks will find
1644 * the id_priv state set to destroying and abort.
1645 */
1646 mutex_lock(&id_priv->handler_mutex);
1647 mutex_unlock(&id_priv->handler_mutex);
1648
e51060f0 1649 if (id_priv->cma_dev) {
72219cea 1650 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
0c9361fc 1651 if (id_priv->cm_id.ib)
e51060f0 1652 ib_destroy_cm_id(id_priv->cm_id.ib);
04215330 1653 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
0c9361fc 1654 if (id_priv->cm_id.iw)
07ebafba 1655 iw_destroy_cm_id(id_priv->cm_id.iw);
e51060f0 1656 }
c8f6a362 1657 cma_leave_mc_groups(id_priv);
a396d43a 1658 cma_release_dev(id_priv);
e51060f0
SH		 1659 }
1660
1661 cma_release_port(id_priv);
1662 cma_deref_id(id_priv);
1663 wait_for_completion(&id_priv->comp);
1664
d02d1f53
SH		 1665 if (id_priv->internal_id)
1666 cma_deref_id(id_priv->id.context);
1667
e51060f0 1668 kfree(id_priv->id.route.path_rec);
fa20105e 1669 put_net(id_priv->id.route.addr.dev_addr.net);
e51060f0
SH		 1670 kfree(id_priv);
1671}
1672EXPORT_SYMBOL(rdma_destroy_id);
1673
1674static int cma_rep_recv(struct rdma_id_private *id_priv)
1675{
1676 int ret;
1677
5851bb89 1678 ret = cma_modify_qp_rtr(id_priv, NULL);
e51060f0
SH		 1679 if (ret)
1680 goto reject;
1681
5851bb89 1682 ret = cma_modify_qp_rts(id_priv, NULL);
e51060f0
SH		 1683 if (ret)
1684 goto reject;
1685
1686 ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1687 if (ret)
1688 goto reject;
1689
1690 return 0;
1691reject:
c5483388 1692 cma_modify_qp_err(id_priv);
e51060f0
SH		 1693 ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1694 NULL, 0, NULL, 0);
1695 return ret;
1696}
1697
a1b1b61f
SH		 1698static void cma_set_rep_event_data(struct rdma_cm_event *event,
1699 struct ib_cm_rep_event_param *rep_data,
1700 void *private_data)
1701{
1702 event->param.conn.private_data = private_data;
1703 event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1704 event->param.conn.responder_resources = rep_data->responder_resources;
1705 event->param.conn.initiator_depth = rep_data->initiator_depth;
1706 event->param.conn.flow_control = rep_data->flow_control;
1707 event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1708 event->param.conn.srq = rep_data->srq;
1709 event->param.conn.qp_num = rep_data->remote_qpn;
1710}
1711
e51060f0
SH		 1712static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1713{
1714 struct rdma_id_private *id_priv = cm_id->context;
a1b1b61f
SH		 1715 struct rdma_cm_event event;
1716 int ret = 0;
e51060f0 1717
37e07cda 1718 mutex_lock(&id_priv->handler_mutex);
38ca83a5 1719 if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
37e07cda 1720 id_priv->state != RDMA_CM_CONNECT) ||
38ca83a5 1721 (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
37e07cda
BVA		 1722 id_priv->state != RDMA_CM_DISCONNECT))
1723 goto out;
e51060f0 1724
a1b1b61f 1725 memset(&event, 0, sizeof event);
e51060f0
SH		 1726 switch (ib_event->event) {
1727 case IB_CM_REQ_ERROR:
1728 case IB_CM_REP_ERROR:
a1b1b61f
SH		 1729 event.event = RDMA_CM_EVENT_UNREACHABLE;
1730 event.status = -ETIMEDOUT;
e51060f0
SH		 1731 break;
1732 case IB_CM_REP_RECEIVED:
01602f11 1733 if (id_priv->id.qp) {
a1b1b61f
SH		 1734 event.status = cma_rep_recv(id_priv);
1735 event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1736 RDMA_CM_EVENT_ESTABLISHED;
01602f11 1737 } else {
a1b1b61f 1738 event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
01602f11 1739 }
a1b1b61f
SH		 1740 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1741 ib_event->private_data);
e51060f0
SH		 1742 break;
1743 case IB_CM_RTU_RECEIVED:
0fe313b0
SH		 1744 case IB_CM_USER_ESTABLISHED:
1745 event.event = RDMA_CM_EVENT_ESTABLISHED;
e51060f0
SH		 1746 break;
1747 case IB_CM_DREQ_ERROR:
a1b1b61f 1748 event.status = -ETIMEDOUT; /* fall through */
e51060f0
SH		 1749 case IB_CM_DREQ_RECEIVED:
1750 case IB_CM_DREP_RECEIVED:
550e5ca7
NM		 1751 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1752 RDMA_CM_DISCONNECT))
e51060f0 1753 goto out;
a1b1b61f 1754 event.event = RDMA_CM_EVENT_DISCONNECTED;
e51060f0
SH		 1755 break;
1756 case IB_CM_TIMEWAIT_EXIT:
38ca83a5
AV		 1757 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1758 break;
e51060f0
SH		 1759 case IB_CM_MRA_RECEIVED:
1760 /* ignore event */
1761 goto out;
1762 case IB_CM_REJ_RECEIVED:
c5483388 1763 cma_modify_qp_err(id_priv);
a1b1b61f
SH		 1764 event.status = ib_event->param.rej_rcvd.reason;
1765 event.event = RDMA_CM_EVENT_REJECTED;
1766 event.param.conn.private_data = ib_event->private_data;
1767 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
e51060f0
SH		 1768 break;
1769 default:
aba25a3e 1770 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
e51060f0
SH		 1771 ib_event->event);
1772 goto out;
1773 }
1774
a1b1b61f 1775 ret = id_priv->id.event_handler(&id_priv->id, &event);
e51060f0
SH		 1776 if (ret) {
1777 /* Destroy the CM ID by returning a non-zero value. */
1778 id_priv->cm_id.ib = NULL;
550e5ca7 1779 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 1780 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1781 rdma_destroy_id(&id_priv->id);
1782 return ret;
1783 }
1784out:
de910bd9 1785 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 1786 return ret;
1787}
1788
628e5f6d 1789static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
0b3ca768
HE		 1790 struct ib_cm_event *ib_event,
1791 struct net_device *net_dev)
e51060f0
SH		 1792{
1793 struct rdma_id_private *id_priv;
1794 struct rdma_cm_id *id;
1795 struct rdma_route *rt;
0c505f70
HE		 1796 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1797 const __be64 service_id =
1798 ib_event->param.req_rcvd.primary_path->service_id;
64c5e613 1799 int ret;
e51060f0 1800
fa20105e
GS		 1801 id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1802 listen_id->event_handler, listen_id->context,
b26f9b99 1803 listen_id->ps, ib_event->param.req_rcvd.qp_type);
e51060f0 1804 if (IS_ERR(id))
0c9361fc 1805 return NULL;
3f168d2b 1806
f4753834 1807 id_priv = container_of(id, struct rdma_id_private, id);
0c505f70
HE		 1808 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1809 (struct sockaddr *)&id->route.addr.dst_addr,
1810 listen_id, ib_event, ss_family, service_id))
fbaa1a6d 1811 goto err;
e51060f0
SH		 1812
1813 rt = &id->route;
1814 rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
3f168d2b
KK		 1815 rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1816 GFP_KERNEL);
e51060f0 1817 if (!rt->path_rec)
0c9361fc 1818 goto err;
e51060f0 1819
e51060f0
SH		 1820 rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1821 if (rt->num_paths == 2)
1822 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1823
0b3ca768
HE		 1824 if (net_dev) {
1825 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
6f8372b6 1826 if (ret)
0c9361fc 1827 goto err;
0b3ca768 1828 } else {
b8cab5da
HE		 1829 if (!cma_protocol_roce(listen_id) &&
1830 cma_any_addr(cma_src_addr(id_priv))) {
1831 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1832 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1833 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1834 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1835 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1836 if (ret)
1837 goto err;
1838 }
6f8372b6
SH		 1839 }
1840 rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
e51060f0 1841
550e5ca7 1842 id_priv->state = RDMA_CM_CONNECT;
e51060f0 1843 return id_priv;
3f168d2b 1844
3f168d2b 1845err:
0c9361fc 1846 rdma_destroy_id(id);
e51060f0
SH		 1847 return NULL;
1848}
1849
628e5f6d 1850static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
0b3ca768
HE		 1851 struct ib_cm_event *ib_event,
1852 struct net_device *net_dev)
628e5f6d
SH		 1853{
1854 struct rdma_id_private *id_priv;
1855 struct rdma_cm_id *id;
0c505f70 1856 const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
fa20105e 1857 struct net *net = listen_id->route.addr.dev_addr.net;
628e5f6d
SH		 1858 int ret;
1859
fa20105e 1860 id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
b26f9b99 1861 listen_id->ps, IB_QPT_UD);
628e5f6d
SH		 1862 if (IS_ERR(id))
1863 return NULL;
1864
f4753834 1865 id_priv = container_of(id, struct rdma_id_private, id);
0c505f70
HE		 1866 if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1867 (struct sockaddr *)&id->route.addr.dst_addr,
1868 listen_id, ib_event, ss_family,
1869 ib_event->param.sidr_req_rcvd.service_id))
628e5f6d
SH		 1870 goto err;
1871
0b3ca768
HE		 1872 if (net_dev) {
1873 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
6f8372b6
SH		 1874 if (ret)
1875 goto err;
0b3ca768 1876 } else {
b8cab5da
HE		 1877 if (!cma_any_addr(cma_src_addr(id_priv))) {
1878 ret = cma_translate_addr(cma_src_addr(id_priv),
1879 &id->route.addr.dev_addr);
1880 if (ret)
1881 goto err;
1882 }
6f8372b6 1883 }
628e5f6d 1884
550e5ca7 1885 id_priv->state = RDMA_CM_CONNECT;
628e5f6d
SH		 1886 return id_priv;
1887err:
1888 rdma_destroy_id(id);
1889 return NULL;
1890}
1891
a1b1b61f
SH		 1892static void cma_set_req_event_data(struct rdma_cm_event *event,
1893 struct ib_cm_req_event_param *req_data,
1894 void *private_data, int offset)
1895{
1896 event->param.conn.private_data = private_data + offset;
1897 event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1898 event->param.conn.responder_resources = req_data->responder_resources;
1899 event->param.conn.initiator_depth = req_data->initiator_depth;
1900 event->param.conn.flow_control = req_data->flow_control;
1901 event->param.conn.retry_count = req_data->retry_count;
1902 event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1903 event->param.conn.srq = req_data->srq;
1904 event->param.conn.qp_num = req_data->remote_qpn;
1905}
1906
9595480c
HS		 1907static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1908{
4dd81e89 1909 return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
9595480c
HS		 1910 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1911 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1912 (id->qp_type == IB_QPT_UD)) ||
1913 (!id->qp_type));
1914}
1915
e51060f0
SH		 1916static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1917{
37e07cda 1918 struct rdma_id_private *listen_id, *conn_id = NULL;
a1b1b61f 1919 struct rdma_cm_event event;
0b3ca768 1920 struct net_device *net_dev;
e51060f0
SH		 1921 int offset, ret;
1922
0b3ca768 1923 listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
4c21b5bc
HE		 1924 if (IS_ERR(listen_id))
1925 return PTR_ERR(listen_id);
1926
0b3ca768
HE		 1927 if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1928 ret = -EINVAL;
1929 goto net_dev_put;
1930 }
9595480c 1931
37e07cda
BVA		 1932 mutex_lock(&listen_id->handler_mutex);
1933 if (listen_id->state != RDMA_CM_LISTEN) {
0b3ca768 1934 ret = -ECONNABORTED;
37e07cda 1935 goto err1;
0b3ca768 1936 }
e51060f0 1937
628e5f6d 1938 memset(&event, 0, sizeof event);
e8160e15 1939 offset = cma_user_data_offset(listen_id);
628e5f6d 1940 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
9595480c 1941 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
0b3ca768 1942 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
628e5f6d
SH		 1943 event.param.ud.private_data = ib_event->private_data + offset;
1944 event.param.ud.private_data_len =
1945 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1946 } else {
0b3ca768 1947 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
628e5f6d
SH		 1948 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1949 ib_event->private_data, offset);
1950 }
e51060f0
SH		 1951 if (!conn_id) {
1952 ret = -ENOMEM;
b6cec8aa 1953 goto err1;
e51060f0
SH		 1954 }
1955
de910bd9 1956 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
be9130cc 1957 ret = cma_acquire_dev(conn_id, listen_id);
a1a733f6 1958 if (ret)
b6cec8aa 1959 goto err2;
e51060f0
SH		 1960
1961 conn_id->cm_id.ib = cm_id;
1962 cm_id->context = conn_id;
1963 cm_id->cm_handler = cma_ib_handler;
1964
25ae21a1
SH		 1965 /*
1966 * Protect against the user destroying conn_id from another thread
1967 * until we're done accessing it.
1968 */
1969 atomic_inc(&conn_id->refcount);
a1b1b61f 1970 ret = conn_id->id.event_handler(&conn_id->id, &event);
b6cec8aa
SH		 1971 if (ret)
1972 goto err3;
b6cec8aa
SH		 1973 /*
1974 * Acquire mutex to prevent user executing rdma_destroy_id()
1975 * while we're accessing the cm_id.
1976 */
1977 mutex_lock(&lock);
dd5f03be
MB		 1978 if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1979 (conn_id->id.qp_type != IB_QPT_UD))
b6cec8aa
SH		 1980 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1981 mutex_unlock(&lock);
1982 mutex_unlock(&conn_id->handler_mutex);
1983 mutex_unlock(&listen_id->handler_mutex);
25ae21a1 1984 cma_deref_id(conn_id);
0b3ca768
HE		 1985 if (net_dev)
1986 dev_put(net_dev);
b6cec8aa 1987 return 0;
a1a733f6 1988
b6cec8aa
SH		 1989err3:
1990 cma_deref_id(conn_id);
a1a733f6
KK		 1991 /* Destroy the CM ID by returning a non-zero value. */
1992 conn_id->cm_id.ib = NULL;
b6cec8aa 1993err2:
550e5ca7 1994 cma_exch(conn_id, RDMA_CM_DESTROYING);
de910bd9 1995 mutex_unlock(&conn_id->handler_mutex);
b6cec8aa 1996err1:
de910bd9 1997 mutex_unlock(&listen_id->handler_mutex);
b6cec8aa
SH		 1998 if (conn_id)
1999 rdma_destroy_id(&conn_id->id);
0b3ca768
HE		 2000
2001net_dev_put:
2002 if (net_dev)
2003 dev_put(net_dev);
2004
e51060f0
SH		 2005 return ret;
2006}
2007
cf53936f 2008__be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
e51060f0 2009{
496ce3ce
SH		 2010 if (addr->sa_family == AF_IB)
2011 return ((struct sockaddr_ib *) addr)->sib_sid;
2012
cf53936f 2013 return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
e51060f0 2014}
cf53936f 2015EXPORT_SYMBOL(rdma_get_service_id);
e51060f0 2016
07ebafba
TT		 2017static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2018{
2019 struct rdma_id_private *id_priv = iw_id->context;
a1b1b61f 2020 struct rdma_cm_event event;
07ebafba 2021 int ret = 0;
24d44a39
SW		 2022 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2023 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
07ebafba 2024
37e07cda
BVA		 2025 mutex_lock(&id_priv->handler_mutex);
2026 if (id_priv->state != RDMA_CM_CONNECT)
2027 goto out;
07ebafba 2028
be65f086 2029 memset(&event, 0, sizeof event);
07ebafba
TT		 2030 switch (iw_event->event) {
2031 case IW_CM_EVENT_CLOSE:
a1b1b61f 2032 event.event = RDMA_CM_EVENT_DISCONNECTED;
07ebafba
TT		 2033 break;
2034 case IW_CM_EVENT_CONNECT_REPLY:
24d44a39
SW		 2035 memcpy(cma_src_addr(id_priv), laddr,
2036 rdma_addr_size(laddr));
2037 memcpy(cma_dst_addr(id_priv), raddr,
2038 rdma_addr_size(raddr));
881a045f
SW		 2039 switch (iw_event->status) {
2040 case 0:
a1b1b61f 2041 event.event = RDMA_CM_EVENT_ESTABLISHED;
3ebeebc3
KS		 2042 event.param.conn.initiator_depth = iw_event->ird;
2043 event.param.conn.responder_resources = iw_event->ord;
881a045f
SW		 2044 break;
2045 case -ECONNRESET:
2046 case -ECONNREFUSED:
2047 event.event = RDMA_CM_EVENT_REJECTED;
2048 break;
2049 case -ETIMEDOUT:
2050 event.event = RDMA_CM_EVENT_UNREACHABLE;
2051 break;
2052 default:
2053 event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2054 break;
2055 }
07ebafba
TT		 2056 break;
2057 case IW_CM_EVENT_ESTABLISHED:
a1b1b61f 2058 event.event = RDMA_CM_EVENT_ESTABLISHED;
3ebeebc3
KS		 2059 event.param.conn.initiator_depth = iw_event->ird;
2060 event.param.conn.responder_resources = iw_event->ord;
07ebafba
TT		 2061 break;
2062 default:
2063 BUG_ON(1);
2064 }
2065
a1b1b61f
SH		 2066 event.status = iw_event->status;
2067 event.param.conn.private_data = iw_event->private_data;
2068 event.param.conn.private_data_len = iw_event->private_data_len;
2069 ret = id_priv->id.event_handler(&id_priv->id, &event);
07ebafba
TT		 2070 if (ret) {
2071 /* Destroy the CM ID by returning a non-zero value. */
2072 id_priv->cm_id.iw = NULL;
550e5ca7 2073 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 2074 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 2075 rdma_destroy_id(&id_priv->id);
2076 return ret;
2077 }
2078
37e07cda 2079out:
de910bd9 2080 mutex_unlock(&id_priv->handler_mutex);
07ebafba
TT		 2081 return ret;
2082}
2083
2084static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2085 struct iw_cm_event *iw_event)
2086{
2087 struct rdma_cm_id *new_cm_id;
2088 struct rdma_id_private *listen_id, *conn_id;
a1b1b61f 2089 struct rdma_cm_event event;
37e07cda 2090 int ret = -ECONNABORTED;
24d44a39
SW		 2091 struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2092 struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
07ebafba
TT		 2093
2094 listen_id = cm_id->context;
37e07cda
BVA		 2095
2096 mutex_lock(&listen_id->handler_mutex);
2097 if (listen_id->state != RDMA_CM_LISTEN)
2098 goto out;
07ebafba
TT		 2099
2100 /* Create a new RDMA id for the new IW CM ID */
fa20105e
GS		 2101 new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2102 listen_id->id.event_handler,
07ebafba 2103 listen_id->id.context,
b26f9b99 2104 RDMA_PS_TCP, IB_QPT_RC);
10f32065 2105 if (IS_ERR(new_cm_id)) {
07ebafba
TT		 2106 ret = -ENOMEM;
2107 goto out;
2108 }
2109 conn_id = container_of(new_cm_id, struct rdma_id_private, id);
de910bd9 2110 mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
550e5ca7 2111 conn_id->state = RDMA_CM_CONNECT;
07ebafba 2112
dd5f03be 2113 ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
07ebafba 2114 if (ret) {
de910bd9 2115 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 2116 rdma_destroy_id(new_cm_id);
2117 goto out;
2118 }
2119
be9130cc 2120 ret = cma_acquire_dev(conn_id, listen_id);
07ebafba 2121 if (ret) {
de910bd9 2122 mutex_unlock(&conn_id->handler_mutex);
07ebafba
TT		 2123 rdma_destroy_id(new_cm_id);
2124 goto out;
2125 }
2126
2127 conn_id->cm_id.iw = cm_id;
2128 cm_id->context = conn_id;
2129 cm_id->cm_handler = cma_iw_handler;
2130
24d44a39
SW		 2131 memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2132 memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
07ebafba 2133
a1b1b61f
SH		 2134 memset(&event, 0, sizeof event);
2135 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2136 event.param.conn.private_data = iw_event->private_data;
2137 event.param.conn.private_data_len = iw_event->private_data_len;
3ebeebc3
KS		 2138 event.param.conn.initiator_depth = iw_event->ird;
2139 event.param.conn.responder_resources = iw_event->ord;
25ae21a1
SH		 2140
2141 /*
2142 * Protect against the user destroying conn_id from another thread
2143 * until we're done accessing it.
2144 */
2145 atomic_inc(&conn_id->refcount);
a1b1b61f 2146 ret = conn_id->id.event_handler(&conn_id->id, &event);
07ebafba
TT		 2147 if (ret) {
2148 /* User wants to destroy the CM ID */
2149 conn_id->cm_id.iw = NULL;
550e5ca7 2150 cma_exch(conn_id, RDMA_CM_DESTROYING);
de910bd9 2151 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 2152 cma_deref_id(conn_id);
07ebafba 2153 rdma_destroy_id(&conn_id->id);
de910bd9 2154 goto out;
07ebafba
TT		 2155 }
2156
de910bd9 2157 mutex_unlock(&conn_id->handler_mutex);
25ae21a1 2158 cma_deref_id(conn_id);
de910bd9 2159
07ebafba 2160out:
de910bd9 2161 mutex_unlock(&listen_id->handler_mutex);
07ebafba
TT		 2162 return ret;
2163}
2164
e51060f0
SH		 2165static int cma_ib_listen(struct rdma_id_private *id_priv)
2166{
e51060f0 2167 struct sockaddr *addr;
0c9361fc 2168 struct ib_cm_id *id;
e51060f0 2169 __be64 svc_id;
e51060f0 2170
51efe394
HE		 2171 addr = cma_src_addr(id_priv);
2172 svc_id = rdma_get_service_id(&id_priv->id, addr);
2173 id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
0c9361fc
JM		 2174 if (IS_ERR(id))
2175 return PTR_ERR(id);
0c9361fc 2176 id_priv->cm_id.ib = id;
e51060f0 2177
51efe394 2178 return 0;
e51060f0
SH		 2179}
2180
07ebafba
TT		 2181static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2182{
2183 int ret;
0c9361fc
JM		 2184 struct iw_cm_id *id;
2185
2186 id = iw_create_cm_id(id_priv->id.device,
2187 iw_conn_req_handler,
2188 id_priv);
2189 if (IS_ERR(id))
2190 return PTR_ERR(id);
07ebafba 2191
68cdba06 2192 id->tos = id_priv->tos;
0c9361fc 2193 id_priv->cm_id.iw = id;
07ebafba 2194
24d44a39
SW		 2195 memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2196 rdma_addr_size(cma_src_addr(id_priv)));
07ebafba
TT		 2197
2198 ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2199
2200 if (ret) {
2201 iw_destroy_cm_id(id_priv->cm_id.iw);
2202 id_priv->cm_id.iw = NULL;
2203 }
2204
2205 return ret;
2206}
2207
e51060f0
SH		 2208static int cma_listen_handler(struct rdma_cm_id *id,
2209 struct rdma_cm_event *event)
2210{
2211 struct rdma_id_private *id_priv = id->context;
2212
2213 id->context = id_priv->id.context;
2214 id->event_handler = id_priv->id.event_handler;
2215 return id_priv->id.event_handler(id, event);
2216}
2217
2218static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2219 struct cma_device *cma_dev)
2220{
2221 struct rdma_id_private *dev_id_priv;
2222 struct rdma_cm_id *id;
fa20105e 2223 struct net *net = id_priv->id.route.addr.dev_addr.net;
e51060f0
SH		 2224 int ret;
2225
72219cea 2226 if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
94d0c939
SH		 2227 return;
2228
fa20105e 2229 id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
b26f9b99 2230 id_priv->id.qp_type);
e51060f0
SH		 2231 if (IS_ERR(id))
2232 return;
2233
2234 dev_id_priv = container_of(id, struct rdma_id_private, id);
2235
550e5ca7 2236 dev_id_priv->state = RDMA_CM_ADDR_BOUND;
f4753834
SH		 2237 memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2238 rdma_addr_size(cma_src_addr(id_priv)));
e51060f0 2239
045959db 2240 _cma_attach_to_dev(dev_id_priv, cma_dev);
e51060f0 2241 list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
d02d1f53
SH		 2242 atomic_inc(&id_priv->refcount);
2243 dev_id_priv->internal_id = 1;
5b0ec991 2244 dev_id_priv->afonly = id_priv->afonly;
e51060f0
SH		 2245
2246 ret = rdma_listen(id, id_priv->backlog);
2247 if (ret)
aba25a3e
PP		 2248 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2249 ret, cma_dev->device->name);
e51060f0
SH		 2250}
2251
2252static void cma_listen_on_all(struct rdma_id_private *id_priv)
2253{
2254 struct cma_device *cma_dev;
2255
2256 mutex_lock(&lock);
2257 list_add_tail(&id_priv->list, &listen_any_list);
2258 list_for_each_entry(cma_dev, &dev_list, list)
2259 cma_listen_on_dev(id_priv, cma_dev);
2260 mutex_unlock(&lock);
2261}
2262
a81c994d
SH		 2263void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2264{
2265 struct rdma_id_private *id_priv;
2266
2267 id_priv = container_of(id, struct rdma_id_private, id);
2268 id_priv->tos = (u8) tos;
2269}
2270EXPORT_SYMBOL(rdma_set_service_type);
2271
e51060f0
SH		 2272static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2273 void *context)
2274{
2275 struct cma_work *work = context;
2276 struct rdma_route *route;
2277
2278 route = &work->id->id.route;
2279
2280 if (!status) {
2281 route->num_paths = 1;
2282 *route->path_rec = *path_rec;
2283 } else {
550e5ca7
NM		 2284 work->old_state = RDMA_CM_ROUTE_QUERY;
2285 work->new_state = RDMA_CM_ADDR_RESOLVED;
e51060f0 2286 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
8f0472d3 2287 work->event.status = status;
e51060f0
SH		 2288 }
2289
2290 queue_work(cma_wq, &work->work);
2291}
2292
2293static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2294 struct cma_work *work)
2295{
f4753834 2296 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
e51060f0 2297 struct ib_sa_path_rec path_rec;
a81c994d
SH		 2298 ib_sa_comp_mask comp_mask;
2299 struct sockaddr_in6 *sin6;
f68194ca 2300 struct sockaddr_ib *sib;
e51060f0
SH		 2301
2302 memset(&path_rec, 0, sizeof path_rec);
f4753834
SH		 2303 rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2304 rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2305 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
e51060f0 2306 path_rec.numb_path = 1;
962063e6 2307 path_rec.reversible = 1;
cf53936f 2308 path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
a81c994d
SH		 2309
2310 comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2311 IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2312 IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2313
f68194ca
SH		 2314 switch (cma_family(id_priv)) {
2315 case AF_INET:
a81c994d
SH		 2316 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2317 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
f68194ca
SH		 2318 break;
2319 case AF_INET6:
f4753834 2320 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
a81c994d
SH		 2321 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2322 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
f68194ca
SH		 2323 break;
2324 case AF_IB:
2325 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2326 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2327 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2328 break;
a81c994d 2329 }
e51060f0 2330
c1a0b23b 2331 id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
a81c994d
SH		 2332 id_priv->id.port_num, &path_rec,
2333 comp_mask, timeout_ms,
2334 GFP_KERNEL, cma_query_handler,
2335 work, &id_priv->query);
e51060f0
SH		 2336
2337 return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2338}
2339
c4028958 2340static void cma_work_handler(struct work_struct *_work)
e51060f0 2341{
c4028958 2342 struct cma_work *work = container_of(_work, struct cma_work, work);
e51060f0
SH		 2343 struct rdma_id_private *id_priv = work->id;
2344 int destroy = 0;
2345
de910bd9 2346 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 2347 if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2348 goto out;
2349
2350 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
550e5ca7 2351 cma_exch(id_priv, RDMA_CM_DESTROYING);
e51060f0
SH		 2352 destroy = 1;
2353 }
2354out:
de910bd9 2355 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 2356 cma_deref_id(id_priv);
2357 if (destroy)
2358 rdma_destroy_id(&id_priv->id);
2359 kfree(work);
2360}
2361
dd5bdff8
OG		 2362static void cma_ndev_work_handler(struct work_struct *_work)
2363{
2364 struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2365 struct rdma_id_private *id_priv = work->id;
2366 int destroy = 0;
2367
2368 mutex_lock(&id_priv->handler_mutex);
550e5ca7
NM		 2369 if (id_priv->state == RDMA_CM_DESTROYING ||
2370 id_priv->state == RDMA_CM_DEVICE_REMOVAL)
dd5bdff8
OG		 2371 goto out;
2372
2373 if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
550e5ca7 2374 cma_exch(id_priv, RDMA_CM_DESTROYING);
dd5bdff8
OG		 2375 destroy = 1;
2376 }
2377
2378out:
2379 mutex_unlock(&id_priv->handler_mutex);
2380 cma_deref_id(id_priv);
2381 if (destroy)
2382 rdma_destroy_id(&id_priv->id);
2383 kfree(work);
2384}
2385
e51060f0
SH		 2386static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2387{
2388 struct rdma_route *route = &id_priv->id.route;
2389 struct cma_work *work;
2390 int ret;
2391
2392 work = kzalloc(sizeof *work, GFP_KERNEL);
2393 if (!work)
2394 return -ENOMEM;
2395
2396 work->id = id_priv;
c4028958 2397 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 2398 work->old_state = RDMA_CM_ROUTE_QUERY;
2399 work->new_state = RDMA_CM_ROUTE_RESOLVED;
e51060f0
SH		 2400 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2401
2402 route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2403 if (!route->path_rec) {
2404 ret = -ENOMEM;
2405 goto err1;
2406 }
2407
2408 ret = cma_query_ib_route(id_priv, timeout_ms, work);
2409 if (ret)
2410 goto err2;
2411
2412 return 0;
2413err2:
2414 kfree(route->path_rec);
2415 route->path_rec = NULL;
2416err1:
2417 kfree(work);
2418 return ret;
2419}
2420
2421int rdma_set_ib_paths(struct rdma_cm_id *id,
2422 struct ib_sa_path_rec *path_rec, int num_paths)
2423{
2424 struct rdma_id_private *id_priv;
2425 int ret;
2426
2427 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7
NM		 2428 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2429 RDMA_CM_ROUTE_RESOLVED))
e51060f0
SH		 2430 return -EINVAL;
2431
9893e742
JL		 2432 id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2433 GFP_KERNEL);
e51060f0
SH		 2434 if (!id->route.path_rec) {
2435 ret = -ENOMEM;
2436 goto err;
2437 }
2438
ae2d9293 2439 id->route.num_paths = num_paths;
e51060f0
SH		 2440 return 0;
2441err:
550e5ca7 2442 cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
e51060f0
SH		 2443 return ret;
2444}
2445EXPORT_SYMBOL(rdma_set_ib_paths);
2446
07ebafba
TT		 2447static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2448{
2449 struct cma_work *work;
2450
2451 work = kzalloc(sizeof *work, GFP_KERNEL);
2452 if (!work)
2453 return -ENOMEM;
2454
2455 work->id = id_priv;
c4028958 2456 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 2457 work->old_state = RDMA_CM_ROUTE_QUERY;
2458 work->new_state = RDMA_CM_ROUTE_RESOLVED;
07ebafba
TT		 2459 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2460 queue_work(cma_wq, &work->work);
2461 return 0;
2462}
2463
eb072c4b
EP		 2464static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2465{
2466 int prio;
2467 struct net_device *dev;
2468
2469 prio = rt_tos2priority(tos);
2470 dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2471 vlan_dev_real_dev(ndev) : ndev;
2472
2473 if (dev->num_tc)
2474 return netdev_get_prio_tc_map(dev, prio);
2475
2476#if IS_ENABLED(CONFIG_VLAN_8021Q)
2477 if (ndev->priv_flags & IFF_802_1Q_VLAN)
2478 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2479 VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2480#endif
2481 return 0;
2482}
2483
aeb76df4
LR		 2484static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2485 unsigned long supported_gids,
2486 enum ib_gid_type default_gid)
2487{
2488 if ((network_type == RDMA_NETWORK_IPV4 ||
2489 network_type == RDMA_NETWORK_IPV6) &&
2490 test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2491 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2492
2493 return default_gid;
2494}
2495
3c86aa70
EC		 2496static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2497{
2498 struct rdma_route *route = &id_priv->id.route;
2499 struct rdma_addr *addr = &route->addr;
2500 struct cma_work *work;
2501 int ret;
3c86aa70 2502 struct net_device *ndev = NULL;
dd5f03be 2503
3c86aa70 2504
3c86aa70
EC		 2505 work = kzalloc(sizeof *work, GFP_KERNEL);
2506 if (!work)
2507 return -ENOMEM;
2508
2509 work->id = id_priv;
2510 INIT_WORK(&work->work, cma_work_handler);
2511
2512 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2513 if (!route->path_rec) {
2514 ret = -ENOMEM;
2515 goto err1;
2516 }
2517
2518 route->num_paths = 1;
2519
abae1b71 2520 if (addr->dev_addr.bound_dev_if) {
aeb76df4
LR		 2521 unsigned long supported_gids;
2522
3c86aa70 2523 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
23d70503
WY		 2524 if (!ndev) {
2525 ret = -ENODEV;
2526 goto err2;
2527 }
20029832
MB		 2528
2529 if (ndev->flags & IFF_LOOPBACK) {
2530 dev_put(ndev);
23d70503
WY		 2531 if (!id_priv->id.device->get_netdev) {
2532 ret = -EOPNOTSUPP;
2533 goto err2;
2534 }
20029832
MB		 2535
2536 ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2537 id_priv->id.port_num);
23d70503
WY		 2538 if (!ndev) {
2539 ret = -ENODEV;
2540 goto err2;
2541 }
20029832
MB		 2542 }
2543
abae1b71 2544 route->path_rec->net = &init_net;
20029832 2545 route->path_rec->ifindex = ndev->ifindex;
aeb76df4
LR		 2546 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2547 id_priv->id.port_num);
2548 route->path_rec->gid_type =
2549 cma_route_gid_type(addr->dev_addr.network,
2550 supported_gids,
2551 id_priv->gid_type);
abae1b71 2552 }
3c86aa70
EC		 2553 if (!ndev) {
2554 ret = -ENODEV;
2555 goto err2;
2556 }
2557
dd5f03be 2558 memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
af7bd463 2559
7b85627b
MS		 2560 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2561 &route->path_rec->sgid);
2562 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2563 &route->path_rec->dgid);
af7bd463 2564
c865f246 2565 /* Use the hint from IP Stack to select GID Type */
c3efe750
MB		 2566 if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2567 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2568 if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
c865f246 2569 /* TODO: get the hoplimit from the inet/inet6 device */
c3efe750
MB		 2570 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2571 else
c865f246 2572 route->path_rec->hop_limit = 1;
af7bd463
EC		 2573 route->path_rec->reversible = 1;
2574 route->path_rec->pkey = cpu_to_be16(0xffff);
2575 route->path_rec->mtu_selector = IB_SA_EQ;
eb072c4b 2576 route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
3c86aa70
EC		 2577 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2578 route->path_rec->rate_selector = IB_SA_EQ;
2579 route->path_rec->rate = iboe_get_rate(ndev);
2580 dev_put(ndev);
2581 route->path_rec->packet_life_time_selector = IB_SA_EQ;
2582 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2583 if (!route->path_rec->mtu) {
2584 ret = -EINVAL;
2585 goto err2;
2586 }
2587
550e5ca7
NM		 2588 work->old_state = RDMA_CM_ROUTE_QUERY;
2589 work->new_state = RDMA_CM_ROUTE_RESOLVED;
3c86aa70
EC		 2590 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2591 work->event.status = 0;
2592
2593 queue_work(cma_wq, &work->work);
2594
2595 return 0;
2596
2597err2:
2598 kfree(route->path_rec);
2599 route->path_rec = NULL;
2600err1:
2601 kfree(work);
2602 return ret;
2603}
2604
e51060f0
SH		 2605int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2606{
2607 struct rdma_id_private *id_priv;
2608 int ret;
2609
2610 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2611 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
e51060f0
SH		 2612 return -EINVAL;
2613
2614 atomic_inc(&id_priv->refcount);
fe53ba2f 2615 if (rdma_cap_ib_sa(id->device, id->port_num))
c72f2189 2616 ret = cma_resolve_ib_route(id_priv, timeout_ms);
5d9fb044 2617 else if (rdma_protocol_roce(id->device, id->port_num))
c72f2189
MW		 2618 ret = cma_resolve_iboe_route(id_priv);
2619 else if (rdma_protocol_iwarp(id->device, id->port_num))
07ebafba 2620 ret = cma_resolve_iw_route(id_priv, timeout_ms);
c72f2189 2621 else
e51060f0 2622 ret = -ENOSYS;
c72f2189 2623
e51060f0
SH		 2624 if (ret)
2625 goto err;
2626
2627 return 0;
2628err:
550e5ca7 2629 cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
e51060f0
SH		 2630 cma_deref_id(id_priv);
2631 return ret;
2632}
2633EXPORT_SYMBOL(rdma_resolve_route);
2634
6a3e362d
SH		 2635static void cma_set_loopback(struct sockaddr *addr)
2636{
2637 switch (addr->sa_family) {
2638 case AF_INET:
2639 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2640 break;
2641 case AF_INET6:
2642 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2643 0, 0, 0, htonl(1));
2644 break;
2645 default:
2646 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2647 0, 0, 0, htonl(1));
2648 break;
2649 }
2650}
2651
e51060f0
SH		 2652static int cma_bind_loopback(struct rdma_id_private *id_priv)
2653{
b0569e40 2654 struct cma_device *cma_dev, *cur_dev;
e51060f0 2655 struct ib_port_attr port_attr;
f0ee3404 2656 union ib_gid gid;
e51060f0
SH		 2657 u16 pkey;
2658 int ret;
2659 u8 p;
2660
b0569e40 2661 cma_dev = NULL;
e51060f0 2662 mutex_lock(&lock);
b0569e40
SH		 2663 list_for_each_entry(cur_dev, &dev_list, list) {
2664 if (cma_family(id_priv) == AF_IB &&
72219cea 2665 !rdma_cap_ib_cm(cur_dev->device, 1))
b0569e40
SH		 2666 continue;
2667
2668 if (!cma_dev)
2669 cma_dev = cur_dev;
2670
2671 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2672 if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2673 port_attr.state == IB_PORT_ACTIVE) {
2674 cma_dev = cur_dev;
2675 goto port_found;
2676 }
2677 }
2678 }
2679
2680 if (!cma_dev) {
e82153b5
KK		 2681 ret = -ENODEV;
2682 goto out;
2683 }
e51060f0 2684
e82153b5 2685 p = 1;
e51060f0
SH		 2686
2687port_found:
55ee3ab2 2688 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
e51060f0
SH		 2689 if (ret)
2690 goto out;
2691
2692 ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2693 if (ret)
2694 goto out;
2695
6f8372b6 2696 id_priv->id.route.addr.dev_addr.dev_type =
21655afc 2697 (rdma_protocol_ib(cma_dev->device, p)) ?
6f8372b6
SH		 2698 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2699
2700 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0
SH		 2701 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2702 id_priv->id.port_num = p;
2703 cma_attach_to_dev(id_priv, cma_dev);
f4753834 2704 cma_set_loopback(cma_src_addr(id_priv));
e51060f0
SH		 2705out:
2706 mutex_unlock(&lock);
2707 return ret;
2708}
2709
2710static void addr_handler(int status, struct sockaddr *src_addr,
2711 struct rdma_dev_addr *dev_addr, void *context)
2712{
2713 struct rdma_id_private *id_priv = context;
a1b1b61f 2714 struct rdma_cm_event event;
e51060f0 2715
a1b1b61f 2716 memset(&event, 0, sizeof event);
de910bd9 2717 mutex_lock(&id_priv->handler_mutex);
550e5ca7
NM		 2718 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2719 RDMA_CM_ADDR_RESOLVED))
61a73c70 2720 goto out;
61a73c70 2721
7b85627b 2722 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
61a73c70 2723 if (!status && !id_priv->cma_dev)
be9130cc 2724 status = cma_acquire_dev(id_priv, NULL);
e51060f0
SH		 2725
2726 if (status) {
550e5ca7
NM		 2727 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2728 RDMA_CM_ADDR_BOUND))
e51060f0 2729 goto out;
a1b1b61f
SH		 2730 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2731 event.status = status;
7b85627b 2732 } else
a1b1b61f 2733 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
e51060f0 2734
a1b1b61f 2735 if (id_priv->id.event_handler(&id_priv->id, &event)) {
550e5ca7 2736 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 2737 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 2738 cma_deref_id(id_priv);
2739 rdma_destroy_id(&id_priv->id);
2740 return;
2741 }
2742out:
de910bd9 2743 mutex_unlock(&id_priv->handler_mutex);
e51060f0
SH		 2744 cma_deref_id(id_priv);
2745}
2746
2747static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2748{
2749 struct cma_work *work;
f0ee3404 2750 union ib_gid gid;
e51060f0
SH		 2751 int ret;
2752
2753 work = kzalloc(sizeof *work, GFP_KERNEL);
2754 if (!work)
2755 return -ENOMEM;
2756
2757 if (!id_priv->cma_dev) {
2758 ret = cma_bind_loopback(id_priv);
2759 if (ret)
2760 goto err;
2761 }
2762
6f8372b6
SH		 2763 rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2764 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
e51060f0 2765
e51060f0 2766 work->id = id_priv;
c4028958 2767 INIT_WORK(&work->work, cma_work_handler);
550e5ca7
NM		 2768 work->old_state = RDMA_CM_ADDR_QUERY;
2769 work->new_state = RDMA_CM_ADDR_RESOLVED;
e51060f0
SH		 2770 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2771 queue_work(cma_wq, &work->work);
2772 return 0;
2773err:
2774 kfree(work);
2775 return ret;
2776}
2777
f17df3b0
SH		 2778static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2779{
2780 struct cma_work *work;
2781 int ret;
2782
2783 work = kzalloc(sizeof *work, GFP_KERNEL);
2784 if (!work)
2785 return -ENOMEM;
2786
2787 if (!id_priv->cma_dev) {
2788 ret = cma_resolve_ib_dev(id_priv);
2789 if (ret)
2790 goto err;
2791 }
2792
2793 rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2794 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2795
2796 work->id = id_priv;
2797 INIT_WORK(&work->work, cma_work_handler);
2798 work->old_state = RDMA_CM_ADDR_QUERY;
2799 work->new_state = RDMA_CM_ADDR_RESOLVED;
2800 work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2801 queue_work(cma_wq, &work->work);
2802 return 0;
2803err:
2804 kfree(work);
2805 return ret;
2806}
2807
e51060f0
SH		 2808static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2809 struct sockaddr *dst_addr)
2810{
d14714df
SH		 2811 if (!src_addr || !src_addr->sa_family) {
2812 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
f17df3b0 2813 src_addr->sa_family = dst_addr->sa_family;
b4cfe397
JM		 2814 if (IS_ENABLED(CONFIG_IPV6) &&
2815 dst_addr->sa_family == AF_INET6) {
6c26a771
SB		 2816 struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2817 struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2818 src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2819 if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2820 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
f17df3b0
SH		 2821 } else if (dst_addr->sa_family == AF_IB) {
2822 ((struct sockaddr_ib *) src_addr)->sib_pkey =
2823 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
d14714df
SH		 2824 }
2825 }
2826 return rdma_bind_addr(id, src_addr);
e51060f0
SH		 2827}
2828
2829int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2830 struct sockaddr *dst_addr, int timeout_ms)
2831{
2832 struct rdma_id_private *id_priv;
2833 int ret;
2834
2835 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 2836 if (id_priv->state == RDMA_CM_IDLE) {
e51060f0
SH		 2837 ret = cma_bind_addr(id, src_addr, dst_addr);
2838 if (ret)
2839 return ret;
2840 }
2841
4ae7152e
SH		 2842 if (cma_family(id_priv) != dst_addr->sa_family)
2843 return -EINVAL;
2844
550e5ca7 2845 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
e51060f0
SH		 2846 return -EINVAL;
2847
2848 atomic_inc(&id_priv->refcount);
f4753834 2849 memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
f17df3b0 2850 if (cma_any_addr(dst_addr)) {
e51060f0 2851 ret = cma_resolve_loopback(id_priv);
f17df3b0
SH		 2852 } else {
2853 if (dst_addr->sa_family == AF_IB) {
2854 ret = cma_resolve_ib_addr(id_priv);
2855 } else {
2856 ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2857 dst_addr, &id->route.addr.dev_addr,
2858 timeout_ms, addr_handler, id_priv);
2859 }
2860 }
e51060f0
SH		 2861 if (ret)
2862 goto err;
2863
2864 return 0;
2865err:
550e5ca7 2866 cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
e51060f0
SH		 2867 cma_deref_id(id_priv);
2868 return ret;
2869}
2870EXPORT_SYMBOL(rdma_resolve_addr);
2871
a9bb7912
HS		 2872int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2873{
2874 struct rdma_id_private *id_priv;
2875 unsigned long flags;
2876 int ret;
2877
2878 id_priv = container_of(id, struct rdma_id_private, id);
2879 spin_lock_irqsave(&id_priv->lock, flags);
c8dea2f9 2880 if (reuse || id_priv->state == RDMA_CM_IDLE) {
a9bb7912
HS		 2881 id_priv->reuseaddr = reuse;
2882 ret = 0;
2883 } else {
2884 ret = -EINVAL;
2885 }
2886 spin_unlock_irqrestore(&id_priv->lock, flags);
2887 return ret;
2888}
2889EXPORT_SYMBOL(rdma_set_reuseaddr);
2890
68602120
SH		 2891int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2892{
2893 struct rdma_id_private *id_priv;
2894 unsigned long flags;
2895 int ret;
2896
2897 id_priv = container_of(id, struct rdma_id_private, id);
2898 spin_lock_irqsave(&id_priv->lock, flags);
2899 if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2900 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2901 id_priv->afonly = afonly;
2902 ret = 0;
2903 } else {
2904 ret = -EINVAL;
2905 }
2906 spin_unlock_irqrestore(&id_priv->lock, flags);
2907 return ret;
2908}
2909EXPORT_SYMBOL(rdma_set_afonly);
2910
e51060f0
SH		 2911static void cma_bind_port(struct rdma_bind_list *bind_list,
2912 struct rdma_id_private *id_priv)
2913{
58afdcb7
SH		 2914 struct sockaddr *addr;
2915 struct sockaddr_ib *sib;
2916 u64 sid, mask;
2917 __be16 port;
e51060f0 2918
f4753834 2919 addr = cma_src_addr(id_priv);
58afdcb7
SH		 2920 port = htons(bind_list->port);
2921
2922 switch (addr->sa_family) {
2923 case AF_INET:
2924 ((struct sockaddr_in *) addr)->sin_port = port;
2925 break;
2926 case AF_INET6:
2927 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2928 break;
2929 case AF_IB:
2930 sib = (struct sockaddr_ib *) addr;
2931 sid = be64_to_cpu(sib->sib_sid);
2932 mask = be64_to_cpu(sib->sib_sid_mask);
2933 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2934 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2935 break;
2936 }
e51060f0
SH		 2937 id_priv->bind_list = bind_list;
2938 hlist_add_head(&id_priv->node, &bind_list->owners);
2939}
2940
aac978e1
HE		 2941static int cma_alloc_port(enum rdma_port_space ps,
2942 struct rdma_id_private *id_priv, unsigned short snum)
e51060f0
SH		 2943{
2944 struct rdma_bind_list *bind_list;
3b069c5d 2945 int ret;
e51060f0 2946
cb164b8c 2947 bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
e51060f0
SH		 2948 if (!bind_list)
2949 return -ENOMEM;
2950
fa20105e
GS		 2951 ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2952 snum);
3b069c5d
TH		 2953 if (ret < 0)
2954 goto err;
aedec080
SH		 2955
2956 bind_list->ps = ps;
3b069c5d 2957 bind_list->port = (unsigned short)ret;
aedec080
SH		 2958 cma_bind_port(bind_list, id_priv);
2959 return 0;
3b069c5d 2960err:
aedec080 2961 kfree(bind_list);
3b069c5d 2962 return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
aedec080 2963}
e51060f0 2964
aac978e1
HE		 2965static int cma_alloc_any_port(enum rdma_port_space ps,
2966 struct rdma_id_private *id_priv)
aedec080 2967{
5d7220e8
TH		 2968 static unsigned int last_used_port;
2969 int low, high, remaining;
2970 unsigned int rover;
fa20105e 2971 struct net *net = id_priv->id.route.addr.dev_addr.net;
e51060f0 2972
fa20105e 2973 inet_get_local_port_range(net, &low, &high);
5d7220e8 2974 remaining = (high - low) + 1;
63862b5b 2975 rover = prandom_u32() % remaining + low;
5d7220e8
TH		 2976retry:
2977 if (last_used_port != rover &&
fa20105e 2978 !cma_ps_find(net, ps, (unsigned short)rover)) {
5d7220e8
TH		 2979 int ret = cma_alloc_port(ps, id_priv, rover);
2980 /*
2981 * Remember previously used port number in order to avoid
2982 * re-using same port immediately after it is closed.
2983 */
2984 if (!ret)
2985 last_used_port = rover;
2986 if (ret != -EADDRNOTAVAIL)
2987 return ret;
e51060f0 2988 }
5d7220e8
TH		 2989 if (--remaining) {
2990 rover++;
2991 if ((rover < low) || (rover > high))
2992 rover = low;
2993 goto retry;
2994 }
2995 return -EADDRNOTAVAIL;
e51060f0
SH		 2996}
2997
a9bb7912
HS		 2998/*
2999 * Check that the requested port is available. This is called when trying to
3000 * bind to a specific port, or when trying to listen on a bound port. In
3001 * the latter case, the provided id_priv may already be on the bind_list, but
3002 * we still need to check that it's okay to start listening.
3003 */
3004static int cma_check_port(struct rdma_bind_list *bind_list,
3005 struct rdma_id_private *id_priv, uint8_t reuseaddr)
e51060f0
SH		 3006{
3007 struct rdma_id_private *cur_id;
43b752da 3008 struct sockaddr *addr, *cur_addr;
e51060f0 3009
f4753834 3010 addr = cma_src_addr(id_priv);
b67bfe0d 3011 hlist_for_each_entry(cur_id, &bind_list->owners, node) {
a9bb7912
HS		 3012 if (id_priv == cur_id)
3013 continue;
3cd96564 3014
5b0ec991
SH		 3015 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3016 cur_id->reuseaddr)
3017 continue;
e51060f0 3018
f4753834 3019 cur_addr = cma_src_addr(cur_id);
5b0ec991
SH		 3020 if (id_priv->afonly && cur_id->afonly &&
3021 (addr->sa_family != cur_addr->sa_family))
3022 continue;
3023
3024 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3025 return -EADDRNOTAVAIL;
3026
3027 if (!cma_addr_cmp(addr, cur_addr))
3028 return -EADDRINUSE;
a9bb7912 3029 }
e51060f0
SH		 3030 return 0;
3031}
3032
aac978e1
HE		 3033static int cma_use_port(enum rdma_port_space ps,
3034 struct rdma_id_private *id_priv)
a9bb7912
HS		 3035{
3036 struct rdma_bind_list *bind_list;
3037 unsigned short snum;
3038 int ret;
3039
f4753834 3040 snum = ntohs(cma_port(cma_src_addr(id_priv)));
a9bb7912
HS		 3041 if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3042 return -EACCES;
3043
fa20105e 3044 bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
a9bb7912
HS		 3045 if (!bind_list) {
3046 ret = cma_alloc_port(ps, id_priv, snum);
3047 } else {
3048 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3049 if (!ret)
3050 cma_bind_port(bind_list, id_priv);
3051 }
3052 return ret;
3053}
3054
3055static int cma_bind_listen(struct rdma_id_private *id_priv)
3056{
3057 struct rdma_bind_list *bind_list = id_priv->bind_list;
3058 int ret = 0;
3059
3060 mutex_lock(&lock);
3061 if (bind_list->owners.first->next)
3062 ret = cma_check_port(bind_list, id_priv, 0);
3063 mutex_unlock(&lock);
3064 return ret;
3065}
3066
aac978e1
HE		 3067static enum rdma_port_space cma_select_inet_ps(
3068 struct rdma_id_private *id_priv)
e51060f0 3069{
e51060f0 3070 switch (id_priv->id.ps) {
e51060f0 3071 case RDMA_PS_TCP:
628e5f6d 3072 case RDMA_PS_UDP:
c8f6a362 3073 case RDMA_PS_IPOIB:
2d2e9415 3074 case RDMA_PS_IB:
aac978e1 3075 return id_priv->id.ps;
e51060f0 3076 default:
aac978e1
HE		 3077
3078 return 0;
58afdcb7
SH		 3079 }
3080}
3081
aac978e1 3082static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
58afdcb7 3083{
aac978e1 3084 enum rdma_port_space ps = 0;
58afdcb7
SH		 3085 struct sockaddr_ib *sib;
3086 u64 sid_ps, mask, sid;
3087
f4753834 3088 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
58afdcb7
SH		 3089 mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3090 sid = be64_to_cpu(sib->sib_sid) & mask;
3091
3092 if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3093 sid_ps = RDMA_IB_IP_PS_IB;
aac978e1 3094 ps = RDMA_PS_IB;
58afdcb7
SH		 3095 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3096 (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3097 sid_ps = RDMA_IB_IP_PS_TCP;
aac978e1 3098 ps = RDMA_PS_TCP;
58afdcb7
SH		 3099 } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3100 (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3101 sid_ps = RDMA_IB_IP_PS_UDP;
aac978e1 3102 ps = RDMA_PS_UDP;
e51060f0
SH		 3103 }
3104
58afdcb7
SH		 3105 if (ps) {
3106 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3107 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3108 be64_to_cpu(sib->sib_sid_mask));
3109 }
3110 return ps;
3111}
3112
3113static int cma_get_port(struct rdma_id_private *id_priv)
3114{
aac978e1 3115 enum rdma_port_space ps;
58afdcb7
SH		 3116 int ret;
3117
f4753834 3118 if (cma_family(id_priv) != AF_IB)
58afdcb7
SH		 3119 ps = cma_select_inet_ps(id_priv);
3120 else
3121 ps = cma_select_ib_ps(id_priv);
3122 if (!ps)
3123 return -EPROTONOSUPPORT;
3124
e51060f0 3125 mutex_lock(&lock);
f4753834 3126 if (cma_any_port(cma_src_addr(id_priv)))
aedec080 3127 ret = cma_alloc_any_port(ps, id_priv);
e51060f0
SH		 3128 else
3129 ret = cma_use_port(ps, id_priv);
3130 mutex_unlock(&lock);
3131
3132 return ret;
3133}
3134
d14714df
SH		 3135static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3136 struct sockaddr *addr)
3137{
d90f9b35 3138#if IS_ENABLED(CONFIG_IPV6)
d14714df
SH		 3139 struct sockaddr_in6 *sin6;
3140
3141 if (addr->sa_family != AF_INET6)
3142 return 0;
3143
3144 sin6 = (struct sockaddr_in6 *) addr;
5462eddd
SK		 3145
3146 if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3147 return 0;
3148
3149 if (!sin6->sin6_scope_id)
d14714df
SH		 3150 return -EINVAL;
3151
3152 dev_addr->bound_dev_if = sin6->sin6_scope_id;
3153#endif
3154 return 0;
3155}
3156
a9bb7912
HS		 3157int rdma_listen(struct rdma_cm_id *id, int backlog)
3158{
3159 struct rdma_id_private *id_priv;
3160 int ret;
3161
3162 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 3163 if (id_priv->state == RDMA_CM_IDLE) {
f4753834
SH		 3164 id->route.addr.src_addr.ss_family = AF_INET;
3165 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
a9bb7912
HS		 3166 if (ret)
3167 return ret;
3168 }
3169
550e5ca7 3170 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
a9bb7912
HS		 3171 return -EINVAL;
3172
3173 if (id_priv->reuseaddr) {
3174 ret = cma_bind_listen(id_priv);
3175 if (ret)
3176 goto err;
3177 }
3178
3179 id_priv->backlog = backlog;
3180 if (id->device) {
72219cea 3181 if (rdma_cap_ib_cm(id->device, 1)) {
a9bb7912
HS		 3182 ret = cma_ib_listen(id_priv);
3183 if (ret)
3184 goto err;
04215330 3185 } else if (rdma_cap_iw_cm(id->device, 1)) {
a9bb7912
HS		 3186 ret = cma_iw_listen(id_priv, backlog);
3187 if (ret)
3188 goto err;
21655afc 3189 } else {
a9bb7912
HS		 3190 ret = -ENOSYS;
3191 goto err;
3192 }
3193 } else
3194 cma_listen_on_all(id_priv);
3195
3196 return 0;
3197err:
3198 id_priv->backlog = 0;
550e5ca7 3199 cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
a9bb7912
HS		 3200 return ret;
3201}
3202EXPORT_SYMBOL(rdma_listen);
3203
e51060f0
SH		 3204int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3205{
3206 struct rdma_id_private *id_priv;
3207 int ret;
3208
680f920a
SH		 3209 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3210 addr->sa_family != AF_IB)
e51060f0
SH		 3211 return -EAFNOSUPPORT;
3212
3213 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 3214 if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
e51060f0
SH		 3215 return -EINVAL;
3216
d14714df
SH		 3217 ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3218 if (ret)
3219 goto err1;
3220
7b85627b 3221 memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
8523c048 3222 if (!cma_any_addr(addr)) {
680f920a 3223 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
e51060f0 3224 if (ret)
255d0c14
KK		 3225 goto err1;
3226
be9130cc 3227 ret = cma_acquire_dev(id_priv, NULL);
255d0c14
KK		 3228 if (ret)
3229 goto err1;
e51060f0
SH		 3230 }
3231
68602120
SH		 3232 if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3233 if (addr->sa_family == AF_INET)
3234 id_priv->afonly = 1;
5b0ec991 3235#if IS_ENABLED(CONFIG_IPV6)
fa20105e
GS		 3236 else if (addr->sa_family == AF_INET6) {
3237 struct net *net = id_priv->id.route.addr.dev_addr.net;
3238
3239 id_priv->afonly = net->ipv6.sysctl.bindv6only;
3240 }
5b0ec991 3241#endif
68602120 3242 }
e51060f0
SH		 3243 ret = cma_get_port(id_priv);
3244 if (ret)
255d0c14 3245 goto err2;
e51060f0
SH		 3246
3247 return 0;
255d0c14 3248err2:
a396d43a
SH		 3249 if (id_priv->cma_dev)
3250 cma_release_dev(id_priv);
255d0c14 3251err1:
550e5ca7 3252 cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
e51060f0
SH		 3253 return ret;
3254}
3255EXPORT_SYMBOL(rdma_bind_addr);
3256
f4753834 3257static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
e51060f0 3258{
e51060f0 3259 struct cma_hdr *cma_hdr;
e51060f0 3260
01602f11
SH		 3261 cma_hdr = hdr;
3262 cma_hdr->cma_version = CMA_VERSION;
f4753834 3263 if (cma_family(id_priv) == AF_INET) {
1f5175ad
AS		 3264 struct sockaddr_in *src4, *dst4;
3265
f4753834
SH		 3266 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3267 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
1f5175ad 3268
01602f11
SH		 3269 cma_set_ip_ver(cma_hdr, 4);
3270 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3271 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3272 cma_hdr->port = src4->sin_port;
e8160e15 3273 } else if (cma_family(id_priv) == AF_INET6) {
1f5175ad
AS		 3274 struct sockaddr_in6 *src6, *dst6;
3275
f4753834
SH		 3276 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3277 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
1f5175ad 3278
01602f11
SH		 3279 cma_set_ip_ver(cma_hdr, 6);
3280 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3281 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3282 cma_hdr->port = src6->sin6_port;
e51060f0
SH		 3283 }
3284 return 0;
3285}
3286
628e5f6d
SH		 3287static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3288 struct ib_cm_event *ib_event)
3289{
3290 struct rdma_id_private *id_priv = cm_id->context;
3291 struct rdma_cm_event event;
3292 struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3293 int ret = 0;
3294
37e07cda
BVA		 3295 mutex_lock(&id_priv->handler_mutex);
3296 if (id_priv->state != RDMA_CM_CONNECT)
3297 goto out;
628e5f6d 3298
8aa08602 3299 memset(&event, 0, sizeof event);
628e5f6d
SH		 3300 switch (ib_event->event) {
3301 case IB_CM_SIDR_REQ_ERROR:
3302 event.event = RDMA_CM_EVENT_UNREACHABLE;
3303 event.status = -ETIMEDOUT;
3304 break;
3305 case IB_CM_SIDR_REP_RECEIVED:
3306 event.param.ud.private_data = ib_event->private_data;
3307 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3308 if (rep->status != IB_SIDR_SUCCESS) {
3309 event.event = RDMA_CM_EVENT_UNREACHABLE;
3310 event.status = ib_event->param.sidr_rep_rcvd.status;
3311 break;
3312 }
5c438135 3313 ret = cma_set_qkey(id_priv, rep->qkey);
d2ca39f2
YE		 3314 if (ret) {
3315 event.event = RDMA_CM_EVENT_ADDR_ERROR;
5c438135 3316 event.status = ret;
628e5f6d
SH		 3317 break;
3318 }
3319 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3320 id_priv->id.route.path_rec,
3321 &event.param.ud.ah_attr);
3322 event.param.ud.qp_num = rep->qpn;
3323 event.param.ud.qkey = rep->qkey;
3324 event.event = RDMA_CM_EVENT_ESTABLISHED;
3325 event.status = 0;
3326 break;
3327 default:
aba25a3e 3328 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
628e5f6d
SH		 3329 ib_event->event);
3330 goto out;
3331 }
3332
3333 ret = id_priv->id.event_handler(&id_priv->id, &event);
3334 if (ret) {
3335 /* Destroy the CM ID by returning a non-zero value. */
3336 id_priv->cm_id.ib = NULL;
550e5ca7 3337 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 3338 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 3339 rdma_destroy_id(&id_priv->id);
3340 return ret;
3341 }
3342out:
de910bd9 3343 mutex_unlock(&id_priv->handler_mutex);
628e5f6d
SH		 3344 return ret;
3345}
3346
3347static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3348 struct rdma_conn_param *conn_param)
3349{
3350 struct ib_cm_sidr_req_param req;
0c9361fc 3351 struct ib_cm_id *id;
e511d1ae 3352 void *private_data;
e8160e15 3353 int offset, ret;
628e5f6d 3354
e511d1ae 3355 memset(&req, 0, sizeof req);
e8160e15
SH		 3356 offset = cma_user_data_offset(id_priv);
3357 req.private_data_len = offset + conn_param->private_data_len;
04ded167
SH		 3358 if (req.private_data_len < conn_param->private_data_len)
3359 return -EINVAL;
3360
e8160e15 3361 if (req.private_data_len) {
e511d1ae
SH		 3362 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3363 if (!private_data)
e8160e15
SH		 3364 return -ENOMEM;
3365 } else {
e511d1ae 3366 private_data = NULL;
e8160e15 3367 }
628e5f6d
SH		 3368
3369 if (conn_param->private_data && conn_param->private_data_len)
e511d1ae
SH		 3370 memcpy(private_data + offset, conn_param->private_data,
3371 conn_param->private_data_len);
628e5f6d 3372
e511d1ae
SH		 3373 if (private_data) {
3374 ret = cma_format_hdr(private_data, id_priv);
e8160e15
SH		 3375 if (ret)
3376 goto out;
e511d1ae 3377 req.private_data = private_data;
e8160e15 3378 }
628e5f6d 3379
0c9361fc
JM		 3380 id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3381 id_priv);
3382 if (IS_ERR(id)) {
3383 ret = PTR_ERR(id);
628e5f6d
SH		 3384 goto out;
3385 }
0c9361fc 3386 id_priv->cm_id.ib = id;
628e5f6d 3387
f4753834 3388 req.path = id_priv->id.route.path_rec;
cf53936f 3389 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
628e5f6d
SH		 3390 req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3391 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3392
3393 ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3394 if (ret) {
3395 ib_destroy_cm_id(id_priv->cm_id.ib);
3396 id_priv->cm_id.ib = NULL;
3397 }
3398out:
e511d1ae 3399 kfree(private_data);
628e5f6d
SH		 3400 return ret;
3401}
3402
e51060f0
SH		 3403static int cma_connect_ib(struct rdma_id_private *id_priv,
3404 struct rdma_conn_param *conn_param)
3405{
3406 struct ib_cm_req_param req;
3407 struct rdma_route *route;
3408 void *private_data;
0c9361fc 3409 struct ib_cm_id *id;
e51060f0
SH		 3410 int offset, ret;
3411
3412 memset(&req, 0, sizeof req);
e8160e15 3413 offset = cma_user_data_offset(id_priv);
e51060f0 3414 req.private_data_len = offset + conn_param->private_data_len;
04ded167
SH		 3415 if (req.private_data_len < conn_param->private_data_len)
3416 return -EINVAL;
3417
e8160e15
SH		 3418 if (req.private_data_len) {
3419 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3420 if (!private_data)
3421 return -ENOMEM;
3422 } else {
3423 private_data = NULL;
3424 }
e51060f0
SH		 3425
3426 if (conn_param->private_data && conn_param->private_data_len)
3427 memcpy(private_data + offset, conn_param->private_data,
3428 conn_param->private_data_len);
3429
0c9361fc
JM		 3430 id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3431 if (IS_ERR(id)) {
3432 ret = PTR_ERR(id);
e51060f0
SH		 3433 goto out;
3434 }
0c9361fc 3435 id_priv->cm_id.ib = id;
e51060f0
SH		 3436
3437 route = &id_priv->id.route;
e8160e15
SH		 3438 if (private_data) {
3439 ret = cma_format_hdr(private_data, id_priv);
3440 if (ret)
3441 goto out;
3442 req.private_data = private_data;
3443 }
e51060f0
SH		 3444
3445 req.primary_path = &route->path_rec[0];
3446 if (route->num_paths == 2)
3447 req.alternate_path = &route->path_rec[1];
3448
cf53936f 3449 req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
e51060f0 3450 req.qp_num = id_priv->qp_num;
18c441a6 3451 req.qp_type = id_priv->id.qp_type;
e51060f0
SH		 3452 req.starting_psn = id_priv->seq_num;
3453 req.responder_resources = conn_param->responder_resources;
3454 req.initiator_depth = conn_param->initiator_depth;
3455 req.flow_control = conn_param->flow_control;
4ede178a
SH		 3456 req.retry_count = min_t(u8, 7, conn_param->retry_count);
3457 req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
e51060f0
SH		 3458 req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3459 req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3460 req.max_cm_retries = CMA_MAX_CM_RETRIES;
3461 req.srq = id_priv->srq ? 1 : 0;
3462
3463 ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3464out:
0c9361fc
JM		 3465 if (ret && !IS_ERR(id)) {
3466 ib_destroy_cm_id(id);
675a027c
KK		 3467 id_priv->cm_id.ib = NULL;
3468 }
3469
e51060f0
SH		 3470 kfree(private_data);
3471 return ret;
3472}
3473
07ebafba
TT		 3474static int cma_connect_iw(struct rdma_id_private *id_priv,
3475 struct rdma_conn_param *conn_param)
3476{
3477 struct iw_cm_id *cm_id;
07ebafba
TT		 3478 int ret;
3479 struct iw_cm_conn_param iw_param;
3480
3481 cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
0c9361fc
JM		 3482 if (IS_ERR(cm_id))
3483 return PTR_ERR(cm_id);
07ebafba 3484
68cdba06 3485 cm_id->tos = id_priv->tos;
07ebafba
TT		 3486 id_priv->cm_id.iw = cm_id;
3487
24d44a39
SW		 3488 memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3489 rdma_addr_size(cma_src_addr(id_priv)));
3490 memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3491 rdma_addr_size(cma_dst_addr(id_priv)));
07ebafba 3492
5851bb89 3493 ret = cma_modify_qp_rtr(id_priv, conn_param);
675a027c
KK		 3494 if (ret)
3495 goto out;
07ebafba 3496
f45ee80e
HS		 3497 if (conn_param) {
3498 iw_param.ord = conn_param->initiator_depth;
3499 iw_param.ird = conn_param->responder_resources;
3500 iw_param.private_data = conn_param->private_data;
3501 iw_param.private_data_len = conn_param->private_data_len;
3502 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3503 } else {
3504 memset(&iw_param, 0, sizeof iw_param);
07ebafba 3505 iw_param.qpn = id_priv->qp_num;
f45ee80e 3506 }
07ebafba
TT		 3507 ret = iw_cm_connect(cm_id, &iw_param);
3508out:
0c9361fc 3509 if (ret) {
675a027c
KK		 3510 iw_destroy_cm_id(cm_id);
3511 id_priv->cm_id.iw = NULL;
3512 }
07ebafba
TT		 3513 return ret;
3514}
3515
e51060f0
SH		 3516int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3517{
3518 struct rdma_id_private *id_priv;
3519 int ret;
3520
3521 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7 3522 if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
e51060f0
SH		 3523 return -EINVAL;
3524
3525 if (!id->qp) {
3526 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 3527 id_priv->srq = conn_param->srq;
3528 }
3529
72219cea 3530 if (rdma_cap_ib_cm(id->device, id->port_num)) {
b26f9b99 3531 if (id->qp_type == IB_QPT_UD)
628e5f6d
SH		 3532 ret = cma_resolve_ib_udp(id_priv, conn_param);
3533 else
3534 ret = cma_connect_ib(id_priv, conn_param);
04215330 3535 } else if (rdma_cap_iw_cm(id->device, id->port_num))
07ebafba 3536 ret = cma_connect_iw(id_priv, conn_param);
21655afc 3537 else
e51060f0 3538 ret = -ENOSYS;
e51060f0
SH		 3539 if (ret)
3540 goto err;
3541
3542 return 0;
3543err:
550e5ca7 3544 cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
e51060f0
SH		 3545 return ret;
3546}
3547EXPORT_SYMBOL(rdma_connect);
3548
3549static int cma_accept_ib(struct rdma_id_private *id_priv,
3550 struct rdma_conn_param *conn_param)
3551{
3552 struct ib_cm_rep_param rep;
5851bb89 3553 int ret;
0fe313b0 3554
5851bb89
SH		 3555 ret = cma_modify_qp_rtr(id_priv, conn_param);
3556 if (ret)
3557 goto out;
0fe313b0 3558
5851bb89
SH		 3559 ret = cma_modify_qp_rts(id_priv, conn_param);
3560 if (ret)
3561 goto out;
e51060f0
SH		 3562
3563 memset(&rep, 0, sizeof rep);
3564 rep.qp_num = id_priv->qp_num;
3565 rep.starting_psn = id_priv->seq_num;
3566 rep.private_data = conn_param->private_data;
3567 rep.private_data_len = conn_param->private_data_len;
3568 rep.responder_resources = conn_param->responder_resources;
3569 rep.initiator_depth = conn_param->initiator_depth;
e51060f0
SH		 3570 rep.failover_accepted = 0;
3571 rep.flow_control = conn_param->flow_control;
4ede178a 3572 rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
e51060f0
SH		 3573 rep.srq = id_priv->srq ? 1 : 0;
3574
0fe313b0
SH		 3575 ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3576out:
3577 return ret;
e51060f0
SH		 3578}
3579
07ebafba
TT		 3580static int cma_accept_iw(struct rdma_id_private *id_priv,
3581 struct rdma_conn_param *conn_param)
3582{
3583 struct iw_cm_conn_param iw_param;
3584 int ret;
3585
f6506550
SW		 3586 if (!conn_param)
3587 return -EINVAL;
3588
5851bb89 3589 ret = cma_modify_qp_rtr(id_priv, conn_param);
07ebafba
TT		 3590 if (ret)
3591 return ret;
3592
3593 iw_param.ord = conn_param->initiator_depth;
3594 iw_param.ird = conn_param->responder_resources;
3595 iw_param.private_data = conn_param->private_data;
3596 iw_param.private_data_len = conn_param->private_data_len;
3597 if (id_priv->id.qp) {
3598 iw_param.qpn = id_priv->qp_num;
3599 } else
3600 iw_param.qpn = conn_param->qp_num;
3601
3602 return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3603}
3604
628e5f6d 3605static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
5c438135 3606 enum ib_cm_sidr_status status, u32 qkey,
628e5f6d
SH		 3607 const void *private_data, int private_data_len)
3608{
3609 struct ib_cm_sidr_rep_param rep;
d2ca39f2 3610 int ret;
628e5f6d
SH		 3611
3612 memset(&rep, 0, sizeof rep);
3613 rep.status = status;
3614 if (status == IB_SIDR_SUCCESS) {
5c438135 3615 ret = cma_set_qkey(id_priv, qkey);
d2ca39f2
YE		 3616 if (ret)
3617 return ret;
628e5f6d 3618 rep.qp_num = id_priv->qp_num;
c8f6a362 3619 rep.qkey = id_priv->qkey;
628e5f6d
SH		 3620 }
3621 rep.private_data = private_data;
3622 rep.private_data_len = private_data_len;
3623
3624 return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3625}
3626
e51060f0
SH		 3627int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3628{
3629 struct rdma_id_private *id_priv;
3630 int ret;
3631
3632 id_priv = container_of(id, struct rdma_id_private, id);
83e9502d
NM		 3633
3634 id_priv->owner = task_pid_nr(current);
3635
550e5ca7 3636 if (!cma_comp(id_priv, RDMA_CM_CONNECT))
e51060f0
SH		 3637 return -EINVAL;
3638
3639 if (!id->qp && conn_param) {
3640 id_priv->qp_num = conn_param->qp_num;
e51060f0
SH		 3641 id_priv->srq = conn_param->srq;
3642 }
3643
72219cea 3644 if (rdma_cap_ib_cm(id->device, id->port_num)) {
f45ee80e
HS		 3645 if (id->qp_type == IB_QPT_UD) {
3646 if (conn_param)
3647 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
5c438135 3648 conn_param->qkey,
f45ee80e
HS		 3649 conn_param->private_data,
3650 conn_param->private_data_len);
3651 else
3652 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
5c438135 3653 0, NULL, 0);
f45ee80e
HS		 3654 } else {
3655 if (conn_param)
3656 ret = cma_accept_ib(id_priv, conn_param);
3657 else
3658 ret = cma_rep_recv(id_priv);
3659 }
04215330 3660 } else if (rdma_cap_iw_cm(id->device, id->port_num))
07ebafba 3661 ret = cma_accept_iw(id_priv, conn_param);
21655afc 3662 else
e51060f0 3663 ret = -ENOSYS;
e51060f0
SH		 3664
3665 if (ret)
3666 goto reject;
3667
3668 return 0;
3669reject:
c5483388 3670 cma_modify_qp_err(id_priv);
e51060f0
SH		 3671 rdma_reject(id, NULL, 0);
3672 return ret;
3673}
3674EXPORT_SYMBOL(rdma_accept);
3675
0fe313b0
SH		 3676int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3677{
3678 struct rdma_id_private *id_priv;
3679 int ret;
3680
3681 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 3682 if (!id_priv->cm_id.ib)
0fe313b0
SH		 3683 return -EINVAL;
3684
3685 switch (id->device->node_type) {
3686 case RDMA_NODE_IB_CA:
3687 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3688 break;
3689 default:
3690 ret = 0;
3691 break;
3692 }
3693 return ret;
3694}
3695EXPORT_SYMBOL(rdma_notify);
3696
e51060f0
SH		 3697int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3698 u8 private_data_len)
3699{
3700 struct rdma_id_private *id_priv;
3701 int ret;
3702
3703 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 3704 if (!id_priv->cm_id.ib)
e51060f0
SH		 3705 return -EINVAL;
3706
72219cea 3707 if (rdma_cap_ib_cm(id->device, id->port_num)) {
b26f9b99 3708 if (id->qp_type == IB_QPT_UD)
5c438135 3709 ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
628e5f6d
SH		 3710 private_data, private_data_len);
3711 else
3712 ret = ib_send_cm_rej(id_priv->cm_id.ib,
3713 IB_CM_REJ_CONSUMER_DEFINED, NULL,
3714 0, private_data, private_data_len);
04215330 3715 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
07ebafba
TT		 3716 ret = iw_cm_reject(id_priv->cm_id.iw,
3717 private_data, private_data_len);
21655afc 3718 } else
e51060f0 3719 ret = -ENOSYS;
21655afc 3720
e51060f0
SH		 3721 return ret;
3722}
3723EXPORT_SYMBOL(rdma_reject);
3724
3725int rdma_disconnect(struct rdma_cm_id *id)
3726{
3727 struct rdma_id_private *id_priv;
3728 int ret;
3729
3730 id_priv = container_of(id, struct rdma_id_private, id);
0c9361fc 3731 if (!id_priv->cm_id.ib)
e51060f0
SH		 3732 return -EINVAL;
3733
72219cea 3734 if (rdma_cap_ib_cm(id->device, id->port_num)) {
c5483388 3735 ret = cma_modify_qp_err(id_priv);
07ebafba
TT		 3736 if (ret)
3737 goto out;
e51060f0
SH		 3738 /* Initiate or respond to a disconnect. */
3739 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3740 ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
04215330 3741 } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
07ebafba 3742 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
21655afc 3743 } else
07ebafba 3744 ret = -EINVAL;
21655afc 3745
e51060f0
SH		 3746out:
3747 return ret;
3748}
3749EXPORT_SYMBOL(rdma_disconnect);
3750
c8f6a362
SH		 3751static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3752{
3753 struct rdma_id_private *id_priv;
3754 struct cma_multicast *mc = multicast->context;
3755 struct rdma_cm_event event;
37e07cda 3756 int ret = 0;
c8f6a362
SH		 3757
3758 id_priv = mc->id_priv;
37e07cda
BVA		 3759 mutex_lock(&id_priv->handler_mutex);
3760 if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3761 id_priv->state != RDMA_CM_ADDR_RESOLVED)
3762 goto out;
c8f6a362 3763
5c438135
SH		 3764 if (!status)
3765 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
c5483388 3766 mutex_lock(&id_priv->qp_mutex);
c8f6a362
SH		 3767 if (!status && id_priv->id.qp)
3768 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
46ea5061 3769 be16_to_cpu(multicast->rec.mlid));
c5483388 3770 mutex_unlock(&id_priv->qp_mutex);
c8f6a362
SH		 3771
3772 memset(&event, 0, sizeof event);
3773 event.status = status;
3774 event.param.ud.private_data = mc->context;
3775 if (!status) {
bee3c3c9
MS		 3776 struct rdma_dev_addr *dev_addr =
3777 &id_priv->id.route.addr.dev_addr;
3778 struct net_device *ndev =
3779 dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3780 enum ib_gid_type gid_type =
3781 id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3782 rdma_start_port(id_priv->cma_dev->device)];
3783
c8f6a362
SH		 3784 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3785 ib_init_ah_from_mcmember(id_priv->id.device,
3786 id_priv->id.port_num, &multicast->rec,
bee3c3c9 3787 ndev, gid_type,
c8f6a362
SH		 3788 &event.param.ud.ah_attr);
3789 event.param.ud.qp_num = 0xFFFFFF;
3790 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
bee3c3c9
MS		 3791 if (ndev)
3792 dev_put(ndev);
c8f6a362
SH		 3793 } else
3794 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3795
3796 ret = id_priv->id.event_handler(&id_priv->id, &event);
3797 if (ret) {
550e5ca7 3798 cma_exch(id_priv, RDMA_CM_DESTROYING);
de910bd9 3799 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 3800 rdma_destroy_id(&id_priv->id);
3801 return 0;
3802 }
8aa08602 3803
37e07cda 3804out:
de910bd9 3805 mutex_unlock(&id_priv->handler_mutex);
c8f6a362
SH		 3806 return 0;
3807}
3808
3809static void cma_set_mgid(struct rdma_id_private *id_priv,
3810 struct sockaddr *addr, union ib_gid *mgid)
3811{
3812 unsigned char mc_map[MAX_ADDR_LEN];
3813 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3814 struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3815 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3816
3817 if (cma_any_addr(addr)) {
3818 memset(mgid, 0, sizeof *mgid);
3819 } else if ((addr->sa_family == AF_INET6) &&
1c9b2819 3820 ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
c8f6a362
SH		 3821 0xFF10A01B)) {
3822 /* IPv6 address is an SA assigned MGID. */
3823 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
5bc2b7b3
SH		 3824 } else if (addr->sa_family == AF_IB) {
3825 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
e2e62697
JG		 3826 } else if ((addr->sa_family == AF_INET6)) {
3827 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3828 if (id_priv->id.ps == RDMA_PS_UDP)
3829 mc_map[7] = 0x01; /* Use RDMA CM signature */
3830 *mgid = *(union ib_gid *) (mc_map + 4);
c8f6a362 3831 } else {
a9e527e3 3832 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
c8f6a362
SH		 3833 if (id_priv->id.ps == RDMA_PS_UDP)
3834 mc_map[7] = 0x01; /* Use RDMA CM signature */
c8f6a362
SH		 3835 *mgid = *(union ib_gid *) (mc_map + 4);
3836 }
3837}
3838
ab15c95a
AV		 3839static void cma_query_sa_classport_info_cb(int status,
3840 struct ib_class_port_info *rec,
3841 void *context)
3842{
3843 struct class_port_info_context *cb_ctx = context;
3844
3845 WARN_ON(!context);
3846
3847 if (status || !rec) {
3848 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3849 cb_ctx->device->name, cb_ctx->port_num, status);
3850 goto out;
3851 }
3852
3853 memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3854
3855out:
3856 complete(&cb_ctx->done);
3857}
3858
3859static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3860 struct ib_class_port_info *class_port_info)
3861{
3862 struct class_port_info_context *cb_ctx;
3863 int ret;
3864
3865 cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3866 if (!cb_ctx)
3867 return -ENOMEM;
3868
3869 cb_ctx->device = device;
3870 cb_ctx->class_port_info = class_port_info;
3871 cb_ctx->port_num = port_num;
3872 init_completion(&cb_ctx->done);
3873
3874 ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3875 CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3876 GFP_KERNEL, cma_query_sa_classport_info_cb,
3877 cb_ctx, &cb_ctx->sa_query);
3878 if (ret < 0) {
3879 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3880 device->name, port_num, ret);
3881 goto out;
3882 }
3883
3884 wait_for_completion(&cb_ctx->done);
3885
3886out:
3887 kfree(cb_ctx);
3888 return ret;
3889}
3890
c8f6a362
SH		 3891static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3892 struct cma_multicast *mc)
3893{
3894 struct ib_sa_mcmember_rec rec;
ab15c95a 3895 struct ib_class_port_info class_port_info;
c8f6a362
SH		 3896 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3897 ib_sa_comp_mask comp_mask;
3898 int ret;
3899
3900 ib_addr_get_mgid(dev_addr, &rec.mgid);
3901 ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3902 &rec.mgid, &rec);
3903 if (ret)
3904 return ret;
3905
5bc2b7b3
SH		 3906 ret = cma_set_qkey(id_priv, 0);
3907 if (ret)
3908 return ret;
3909
3f446754 3910 cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
5bc2b7b3 3911 rec.qkey = cpu_to_be32(id_priv->qkey);
6f8372b6 3912 rdma_addr_get_sgid(dev_addr, &rec.port_gid);
c8f6a362 3913 rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
ab15c95a
AV		 3914 rec.join_state = mc->join_state;
3915
3916 if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3917 ret = cma_query_sa_classport_info(id_priv->id.device,
3918 id_priv->id.port_num,
3919 &class_port_info);
3920
3921 if (ret)
3922 return ret;
3923
3924 if (!(ib_get_cpi_capmask2(&class_port_info) &
3925 IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3926 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3927 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3928 id_priv->id.device->name, id_priv->id.port_num);
3929 return -EOPNOTSUPP;
3930 }
3931 }
c8f6a362
SH		 3932
3933 comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3934 IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3935 IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3936 IB_SA_MCMEMBER_REC_FLOW_LABEL |
3937 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3938
84adeee9
YE		 3939 if (id_priv->id.ps == RDMA_PS_IPOIB)
3940 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
2a22fb8c
DB		 3941 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3942 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3943 IB_SA_MCMEMBER_REC_MTU |
3944 IB_SA_MCMEMBER_REC_HOP_LIMIT;
84adeee9 3945
c8f6a362
SH		 3946 mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3947 id_priv->id.port_num, &rec,
3948 comp_mask, GFP_KERNEL,
3949 cma_ib_mc_handler, mc);
8c6ffba0 3950 return PTR_ERR_OR_ZERO(mc->multicast.ib);
c8f6a362
SH		 3951}
3952
3c86aa70
EC		 3953static void iboe_mcast_work_handler(struct work_struct *work)
3954{
3955 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3956 struct cma_multicast *mc = mw->mc;
3957 struct ib_sa_multicast *m = mc->multicast.ib;
3958
3959 mc->multicast.ib->context = mc;
3960 cma_ib_mc_handler(0, m);
3961 kref_put(&mc->mcref, release_mc);
3962 kfree(mw);
3963}
3964
3965static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3966{
3967 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3968 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3969
3970 if (cma_any_addr(addr)) {
3971 memset(mgid, 0, sizeof *mgid);
3972 } else if (addr->sa_family == AF_INET6) {
3973 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3974 } else {
3975 mgid->raw[0] = 0xff;
3976 mgid->raw[1] = 0x0e;
3977 mgid->raw[2] = 0;
3978 mgid->raw[3] = 0;
3979 mgid->raw[4] = 0;
3980 mgid->raw[5] = 0;
3981 mgid->raw[6] = 0;
3982 mgid->raw[7] = 0;
3983 mgid->raw[8] = 0;
3984 mgid->raw[9] = 0;
3985 mgid->raw[10] = 0xff;
3986 mgid->raw[11] = 0xff;
3987 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3988 }
3989}
3990
3991static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3992 struct cma_multicast *mc)
3993{
3994 struct iboe_mcast_work *work;
3995 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
bee3c3c9 3996 int err = 0;
3c86aa70
EC		 3997 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3998 struct net_device *ndev = NULL;
bee3c3c9 3999 enum ib_gid_type gid_type;
ab15c95a
AV		 4000 bool send_only;
4001
4002 send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
3c86aa70
EC		 4003
4004 if (cma_zero_addr((struct sockaddr *)&mc->addr))
4005 return -EINVAL;
4006
4007 work = kzalloc(sizeof *work, GFP_KERNEL);
4008 if (!work)
4009 return -ENOMEM;
4010
4011 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4012 if (!mc->multicast.ib) {
4013 err = -ENOMEM;
4014 goto out1;
4015 }
4016
4017 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4018
4019 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4020 if (id_priv->id.ps == RDMA_PS_UDP)
4021 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4022
4023 if (dev_addr->bound_dev_if)
4024 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4025 if (!ndev) {
4026 err = -ENODEV;
4027 goto out2;
4028 }
4029 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4030 mc->multicast.ib->rec.hop_limit = 1;
4031 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
bee3c3c9
MS		 4032
4033 gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4034 rdma_start_port(id_priv->cma_dev->device)];
4035 if (addr->sa_family == AF_INET) {
c65f6c5a
AV		 4036 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4037 mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
ab15c95a
AV		 4038 if (!send_only) {
4039 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4040 true);
4041 if (!err)
4042 mc->igmp_joined = true;
4043 }
bee3c3c9
MS		 4044 }
4045 } else {
4046 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4047 err = -ENOTSUPP;
4048 }
3c86aa70 4049 dev_put(ndev);
bee3c3c9
MS		 4050 if (err || !mc->multicast.ib->rec.mtu) {
4051 if (!err)
4052 err = -EINVAL;
3c86aa70
EC		 4053 goto out2;
4054 }
7b85627b
MS		 4055 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4056 &mc->multicast.ib->rec.port_gid);
3c86aa70
EC		 4057 work->id = id_priv;
4058 work->mc = mc;
4059 INIT_WORK(&work->work, iboe_mcast_work_handler);
4060 kref_get(&mc->mcref);
4061 queue_work(cma_wq, &work->work);
4062
4063 return 0;
4064
4065out2:
4066 kfree(mc->multicast.ib);
4067out1:
4068 kfree(work);
4069 return err;
4070}
4071
c8f6a362 4072int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
ab15c95a 4073 u8 join_state, void *context)
c8f6a362
SH		 4074{
4075 struct rdma_id_private *id_priv;
4076 struct cma_multicast *mc;
4077 int ret;
4078
4079 id_priv = container_of(id, struct rdma_id_private, id);
550e5ca7
NM		 4080 if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4081 !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
c8f6a362
SH		 4082 return -EINVAL;
4083
4084 mc = kmalloc(sizeof *mc, GFP_KERNEL);
4085 if (!mc)
4086 return -ENOMEM;
4087
ef560861 4088 memcpy(&mc->addr, addr, rdma_addr_size(addr));
c8f6a362
SH		 4089 mc->context = context;
4090 mc->id_priv = id_priv;
bee3c3c9 4091 mc->igmp_joined = false;
ab15c95a 4092 mc->join_state = join_state;
c8f6a362
SH		 4093 spin_lock(&id_priv->lock);
4094 list_add(&mc->list, &id_priv->mc_list);
4095 spin_unlock(&id_priv->lock);
4096
5d9fb044 4097 if (rdma_protocol_roce(id->device, id->port_num)) {
5c9a5282
MW		 4098 kref_init(&mc->mcref);
4099 ret = cma_iboe_join_multicast(id_priv, mc);
a31ad3b0 4100 } else if (rdma_cap_ib_mcast(id->device, id->port_num))
5c9a5282
MW		 4101 ret = cma_join_ib_multicast(id_priv, mc);
4102 else
c8f6a362 4103 ret = -ENOSYS;
c8f6a362
SH		 4104
4105 if (ret) {
4106 spin_lock_irq(&id_priv->lock);
4107 list_del(&mc->list);
4108 spin_unlock_irq(&id_priv->lock);
4109 kfree(mc);
4110 }
4111 return ret;
4112}
4113EXPORT_SYMBOL(rdma_join_multicast);
4114
4115void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4116{
4117 struct rdma_id_private *id_priv;
4118 struct cma_multicast *mc;
4119
4120 id_priv = container_of(id, struct rdma_id_private, id);
4121 spin_lock_irq(&id_priv->lock);
4122 list_for_each_entry(mc, &id_priv->mc_list, list) {
ef560861 4123 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
c8f6a362
SH		 4124 list_del(&mc->list);
4125 spin_unlock_irq(&id_priv->lock);
4126
4127 if (id->qp)
4128 ib_detach_mcast(id->qp,
4129 &mc->multicast.ib->rec.mgid,
46ea5061 4130 be16_to_cpu(mc->multicast.ib->rec.mlid));
5c9a5282
MW		 4131
4132 BUG_ON(id_priv->cma_dev->device != id->device);
4133
a31ad3b0 4134 if (rdma_cap_ib_mcast(id->device, id->port_num)) {
5c9a5282
MW		 4135 ib_sa_free_multicast(mc->multicast.ib);
4136 kfree(mc);
bee3c3c9
MS		 4137 } else if (rdma_protocol_roce(id->device, id->port_num)) {
4138 if (mc->igmp_joined) {
4139 struct rdma_dev_addr *dev_addr =
4140 &id->route.addr.dev_addr;
4141 struct net_device *ndev = NULL;
4142
4143 if (dev_addr->bound_dev_if)
4144 ndev = dev_get_by_index(&init_net,
4145 dev_addr->bound_dev_if);
4146 if (ndev) {
4147 cma_igmp_send(ndev,
4148 &mc->multicast.ib->rec.mgid,
4149 false);
4150 dev_put(ndev);
4151 }
4152 mc->igmp_joined = false;
4153 }
5c9a5282 4154 kref_put(&mc->mcref, release_mc);
bee3c3c9 4155 }
c8f6a362
SH		 4156 return;
4157 }
4158 }
4159 spin_unlock_irq(&id_priv->lock);
4160}
4161EXPORT_SYMBOL(rdma_leave_multicast);
4162
dd5bdff8
OG		 4163static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4164{
4165 struct rdma_dev_addr *dev_addr;
4166 struct cma_ndev_work *work;
4167
4168 dev_addr = &id_priv->id.route.addr.dev_addr;
4169
6266ed6e 4170 if ((dev_addr->bound_dev_if == ndev->ifindex) &&
fa20105e 4171 (net_eq(dev_net(ndev), dev_addr->net)) &&
dd5bdff8 4172 memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
aba25a3e
PP		 4173 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4174 ndev->name, &id_priv->id);
dd5bdff8
OG		 4175 work = kzalloc(sizeof *work, GFP_KERNEL);
4176 if (!work)
4177 return -ENOMEM;
4178
4179 INIT_WORK(&work->work, cma_ndev_work_handler);
4180 work->id = id_priv;
4181 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4182 atomic_inc(&id_priv->refcount);
4183 queue_work(cma_wq, &work->work);
4184 }
4185
4186 return 0;
4187}
4188
4189static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
351638e7 4190 void *ptr)
dd5bdff8 4191{
351638e7 4192 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
dd5bdff8
OG		 4193 struct cma_device *cma_dev;
4194 struct rdma_id_private *id_priv;
4195 int ret = NOTIFY_DONE;
4196
dd5bdff8
OG		 4197 if (event != NETDEV_BONDING_FAILOVER)
4198 return NOTIFY_DONE;
4199
4200 if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4201 return NOTIFY_DONE;
4202
4203 mutex_lock(&lock);
4204 list_for_each_entry(cma_dev, &dev_list, list)
4205 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4206 ret = cma_netdev_change(ndev, id_priv);
4207 if (ret)
4208 goto out;
4209 }
4210
4211out:
4212 mutex_unlock(&lock);
4213 return ret;
4214}
4215
4216static struct notifier_block cma_nb = {
4217 .notifier_call = cma_netdev_callback
4218};
4219
e51060f0
SH		 4220static void cma_add_one(struct ib_device *device)
4221{
4222 struct cma_device *cma_dev;
4223 struct rdma_id_private *id_priv;
045959db
MB		 4224 unsigned int i;
4225 unsigned long supported_gids = 0;
e51060f0
SH		 4226
4227 cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4228 if (!cma_dev)
4229 return;
4230
4231 cma_dev->device = device;
045959db
MB		 4232 cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4233 sizeof(*cma_dev->default_gid_type),
4234 GFP_KERNEL);
4235 if (!cma_dev->default_gid_type) {
4236 kfree(cma_dev);
4237 return;
4238 }
4239 for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4240 supported_gids = roce_gid_type_mask_support(device, i);
4241 WARN_ON(!supported_gids);
4242 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4243 find_first_bit(&supported_gids, BITS_PER_LONG);
4244 }
e51060f0
SH		 4245
4246 init_completion(&cma_dev->comp);
4247 atomic_set(&cma_dev->refcount, 1);
4248 INIT_LIST_HEAD(&cma_dev->id_list);
4249 ib_set_client_data(device, &cma_client, cma_dev);
4250
4251 mutex_lock(&lock);
4252 list_add_tail(&cma_dev->list, &dev_list);
4253 list_for_each_entry(id_priv, &listen_any_list, list)
4254 cma_listen_on_dev(id_priv, cma_dev);
4255 mutex_unlock(&lock);
e51060f0
SH		 4256}
4257
4258static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4259{
a1b1b61f 4260 struct rdma_cm_event event;
550e5ca7 4261 enum rdma_cm_state state;
de910bd9 4262 int ret = 0;
e51060f0
SH		 4263
4264 /* Record that we want to remove the device */
550e5ca7
NM		 4265 state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4266 if (state == RDMA_CM_DESTROYING)
e51060f0
SH		 4267 return 0;
4268
4269 cma_cancel_operation(id_priv, state);
de910bd9 4270 mutex_lock(&id_priv->handler_mutex);
e51060f0
SH		 4271
4272 /* Check for destruction from another callback. */
550e5ca7 4273 if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
de910bd9 4274 goto out;
e51060f0 4275
a1b1b61f
SH		 4276 memset(&event, 0, sizeof event);
4277 event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
de910bd9
OG		 4278 ret = id_priv->id.event_handler(&id_priv->id, &event);
4279out:
4280 mutex_unlock(&id_priv->handler_mutex);
4281 return ret;
e51060f0
SH		 4282}
4283
4284static void cma_process_remove(struct cma_device *cma_dev)
4285{
e51060f0
SH		 4286 struct rdma_id_private *id_priv;
4287 int ret;
4288
e51060f0
SH		 4289 mutex_lock(&lock);
4290 while (!list_empty(&cma_dev->id_list)) {
4291 id_priv = list_entry(cma_dev->id_list.next,
4292 struct rdma_id_private, list);
4293
d02d1f53 4294 list_del(&id_priv->listen_list);
94de178a 4295 list_del_init(&id_priv->list);
e51060f0
SH		 4296 atomic_inc(&id_priv->refcount);
4297 mutex_unlock(&lock);
4298
d02d1f53 4299 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
e51060f0
SH		 4300 cma_deref_id(id_priv);
4301 if (ret)
4302 rdma_destroy_id(&id_priv->id);
4303
4304 mutex_lock(&lock);
4305 }
4306 mutex_unlock(&lock);
4307
4308 cma_deref_dev(cma_dev);
4309 wait_for_completion(&cma_dev->comp);
4310}
4311
7c1eb45a 4312static void cma_remove_one(struct ib_device *device, void *client_data)
e51060f0 4313{
7c1eb45a 4314 struct cma_device *cma_dev = client_data;
e51060f0 4315
e51060f0
SH		 4316 if (!cma_dev)
4317 return;
4318
4319 mutex_lock(&lock);
4320 list_del(&cma_dev->list);
4321 mutex_unlock(&lock);
4322
4323 cma_process_remove(cma_dev);
045959db 4324 kfree(cma_dev->default_gid_type);
e51060f0
SH		 4325 kfree(cma_dev);
4326}
4327
753f618a
NM		 4328static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4329{
4330 struct nlmsghdr *nlh;
4331 struct rdma_cm_id_stats *id_stats;
4332 struct rdma_id_private *id_priv;
4333 struct rdma_cm_id *id = NULL;
4334 struct cma_device *cma_dev;
4335 int i_dev = 0, i_id = 0;
4336
4337 /*
4338 * We export all of the IDs as a sequence of messages. Each
4339 * ID gets its own netlink message.
4340 */
4341 mutex_lock(&lock);
4342
4343 list_for_each_entry(cma_dev, &dev_list, list) {
4344 if (i_dev < cb->args[0]) {
4345 i_dev++;
4346 continue;
4347 }
4348
4349 i_id = 0;
4350 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4351 if (i_id < cb->args[1]) {
4352 i_id++;
4353 continue;
4354 }
4355
4356 id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4357 sizeof *id_stats, RDMA_NL_RDMA_CM,
30dc5e63
TN		 4358 RDMA_NL_RDMA_CM_ID_STATS,
4359 NLM_F_MULTI);
753f618a
NM		 4360 if (!id_stats)
4361 goto out;
4362
4363 memset(id_stats, 0, sizeof *id_stats);
4364 id = &id_priv->id;
4365 id_stats->node_type = id->route.addr.dev_addr.dev_type;
4366 id_stats->port_num = id->port_num;
4367 id_stats->bound_dev_if =
4368 id->route.addr.dev_addr.bound_dev_if;
4369
ce117ffa
SH		 4370 if (ibnl_put_attr(skb, nlh,
4371 rdma_addr_size(cma_src_addr(id_priv)),
4372 cma_src_addr(id_priv),
4373 RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4374 goto out;
4375 if (ibnl_put_attr(skb, nlh,
4376 rdma_addr_size(cma_src_addr(id_priv)),
4377 cma_dst_addr(id_priv),
4378 RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4379 goto out;
753f618a 4380
83e9502d 4381 id_stats->pid = id_priv->owner;
753f618a
NM		 4382 id_stats->port_space = id->ps;
4383 id_stats->cm_state = id_priv->state;
4384 id_stats->qp_num = id_priv->qp_num;
4385 id_stats->qp_type = id->qp_type;
4386
4387 i_id++;
4388 }
4389
4390 cb->args[1] = 0;
4391 i_dev++;
4392 }
4393
4394out:
4395 mutex_unlock(&lock);
4396 cb->args[0] = i_dev;
4397 cb->args[1] = i_id;
4398
4399 return skb->len;
4400}
4401
4402static const struct ibnl_client_cbs cma_cb_table[] = {
809d5fc9
G		 4403 [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4404 .module = THIS_MODULE },
753f618a
NM		 4405};
4406
4be74b42
HE		 4407static int cma_init_net(struct net *net)
4408{
4409 struct cma_pernet *pernet = cma_pernet(net);
4410
4411 idr_init(&pernet->tcp_ps);
4412 idr_init(&pernet->udp_ps);
4413 idr_init(&pernet->ipoib_ps);
4414 idr_init(&pernet->ib_ps);
4415
4416 return 0;
4417}
4418
4419static void cma_exit_net(struct net *net)
4420{
4421 struct cma_pernet *pernet = cma_pernet(net);
4422
4423 idr_destroy(&pernet->tcp_ps);
4424 idr_destroy(&pernet->udp_ps);
4425 idr_destroy(&pernet->ipoib_ps);
4426 idr_destroy(&pernet->ib_ps);
4427}
4428
4429static struct pernet_operations cma_pernet_operations = {
4430 .init = cma_init_net,
4431 .exit = cma_exit_net,
4432 .id = &cma_pernet_id,
4433 .size = sizeof(struct cma_pernet),
4434};
4435
716abb1f 4436static int __init cma_init(void)
e51060f0 4437{
5d7220e8 4438 int ret;
227b60f5 4439
dee9acbb 4440 cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
e51060f0
SH		 4441 if (!cma_wq)
4442 return -ENOMEM;
4443
4be74b42
HE		 4444 ret = register_pernet_subsys(&cma_pernet_operations);
4445 if (ret)
4446 goto err_wq;
4447
c1a0b23b 4448 ib_sa_register_client(&sa_client);
7a118df3 4449 rdma_addr_register_client(&addr_client);
dd5bdff8 4450 register_netdevice_notifier(&cma_nb);
c1a0b23b 4451
e51060f0
SH		 4452 ret = ib_register_client(&cma_client);
4453 if (ret)
4454 goto err;
753f618a 4455
2fa2d4fb
MB		 4456 if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4457 cma_cb_table))
aba25a3e 4458 pr_warn("RDMA CMA: failed to add netlink callback\n");
045959db 4459 cma_configfs_init();
753f618a 4460
e51060f0
SH		 4461 return 0;
4462
4463err:
dd5bdff8 4464 unregister_netdevice_notifier(&cma_nb);
7a118df3 4465 rdma_addr_unregister_client(&addr_client);
c1a0b23b 4466 ib_sa_unregister_client(&sa_client);
4be74b42 4467err_wq:
e51060f0
SH		 4468 destroy_workqueue(cma_wq);
4469 return ret;
4470}
4471
716abb1f 4472static void __exit cma_cleanup(void)
e51060f0 4473{
045959db 4474 cma_configfs_exit();
753f618a 4475 ibnl_remove_client(RDMA_NL_RDMA_CM);
e51060f0 4476 ib_unregister_client(&cma_client);
dd5bdff8 4477 unregister_netdevice_notifier(&cma_nb);
7a118df3 4478 rdma_addr_unregister_client(&addr_client);
c1a0b23b 4479 ib_sa_unregister_client(&sa_client);
4be74b42 4480 unregister_pernet_subsys(&cma_pernet_operations);
e51060f0 4481 destroy_workqueue(cma_wq);
e51060f0
SH		 4482}
4483
4484module_init(cma_init);
4485module_exit(cma_cleanup);
ubuntu-zesty-kernel mirror