xref: /linux/drivers/infiniband/core/sa_query.c (revision a8fe58cec351c25e09c393bf46117c0c47b5a17c)
1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
4  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/err.h>
38 #include <linux/random.h>
39 #include <linux/spinlock.h>
40 #include <linux/slab.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kref.h>
43 #include <linux/idr.h>
44 #include <linux/workqueue.h>
45 #include <uapi/linux/if_ether.h>
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_cache.h>
48 #include <rdma/rdma_netlink.h>
49 #include <net/netlink.h>
50 #include <uapi/rdma/ib_user_sa.h>
51 #include <rdma/ib_marshall.h>
52 #include <rdma/ib_addr.h>
53 #include "sa.h"
54 #include "core_priv.h"
55 
56 MODULE_AUTHOR("Roland Dreier");
57 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
58 MODULE_LICENSE("Dual BSD/GPL");
59 
60 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN		100
61 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT		2000
62 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX		200000
63 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
64 
65 struct ib_sa_sm_ah {
66 	struct ib_ah        *ah;
67 	struct kref          ref;
68 	u16		     pkey_index;
69 	u8		     src_path_mask;
70 };
71 
72 struct ib_sa_port {
73 	struct ib_mad_agent *agent;
74 	struct ib_sa_sm_ah  *sm_ah;
75 	struct work_struct   update_task;
76 	spinlock_t           ah_lock;
77 	u8                   port_num;
78 };
79 
80 struct ib_sa_device {
81 	int                     start_port, end_port;
82 	struct ib_event_handler event_handler;
83 	struct ib_sa_port port[0];
84 };
85 
86 struct ib_sa_query {
87 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
88 	void (*release)(struct ib_sa_query *);
89 	struct ib_sa_client    *client;
90 	struct ib_sa_port      *port;
91 	struct ib_mad_send_buf *mad_buf;
92 	struct ib_sa_sm_ah     *sm_ah;
93 	int			id;
94 	u32			flags;
95 	struct list_head	list; /* Local svc request list */
96 	u32			seq; /* Local svc request sequence number */
97 	unsigned long		timeout; /* Local svc timeout */
98 	u8			path_use; /* How will the pathrecord be used */
99 };
100 
101 #define IB_SA_ENABLE_LOCAL_SERVICE	0x00000001
102 #define IB_SA_CANCEL			0x00000002
103 
104 struct ib_sa_service_query {
105 	void (*callback)(int, struct ib_sa_service_rec *, void *);
106 	void *context;
107 	struct ib_sa_query sa_query;
108 };
109 
110 struct ib_sa_path_query {
111 	void (*callback)(int, struct ib_sa_path_rec *, void *);
112 	void *context;
113 	struct ib_sa_query sa_query;
114 };
115 
116 struct ib_sa_guidinfo_query {
117 	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
118 	void *context;
119 	struct ib_sa_query sa_query;
120 };
121 
122 struct ib_sa_mcmember_query {
123 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
124 	void *context;
125 	struct ib_sa_query sa_query;
126 };
127 
128 static LIST_HEAD(ib_nl_request_list);
129 static DEFINE_SPINLOCK(ib_nl_request_lock);
130 static atomic_t ib_nl_sa_request_seq;
131 static struct workqueue_struct *ib_nl_wq;
132 static struct delayed_work ib_nl_timed_work;
133 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
134 	[LS_NLA_TYPE_PATH_RECORD]	= {.type = NLA_BINARY,
135 		.len = sizeof(struct ib_path_rec_data)},
136 	[LS_NLA_TYPE_TIMEOUT]		= {.type = NLA_U32},
137 	[LS_NLA_TYPE_SERVICE_ID]	= {.type = NLA_U64},
138 	[LS_NLA_TYPE_DGID]		= {.type = NLA_BINARY,
139 		.len = sizeof(struct rdma_nla_ls_gid)},
140 	[LS_NLA_TYPE_SGID]		= {.type = NLA_BINARY,
141 		.len = sizeof(struct rdma_nla_ls_gid)},
142 	[LS_NLA_TYPE_TCLASS]		= {.type = NLA_U8},
143 	[LS_NLA_TYPE_PKEY]		= {.type = NLA_U16},
144 	[LS_NLA_TYPE_QOS_CLASS]		= {.type = NLA_U16},
145 };
146 
147 
148 static void ib_sa_add_one(struct ib_device *device);
149 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
150 
151 static struct ib_client sa_client = {
152 	.name   = "sa",
153 	.add    = ib_sa_add_one,
154 	.remove = ib_sa_remove_one
155 };
156 
157 static DEFINE_SPINLOCK(idr_lock);
158 static DEFINE_IDR(query_idr);
159 
160 static DEFINE_SPINLOCK(tid_lock);
161 static u32 tid;
162 
163 #define PATH_REC_FIELD(field) \
164 	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
165 	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
166 	.field_name          = "sa_path_rec:" #field
167 
168 static const struct ib_field path_rec_table[] = {
169 	{ PATH_REC_FIELD(service_id),
170 	  .offset_words = 0,
171 	  .offset_bits  = 0,
172 	  .size_bits    = 64 },
173 	{ PATH_REC_FIELD(dgid),
174 	  .offset_words = 2,
175 	  .offset_bits  = 0,
176 	  .size_bits    = 128 },
177 	{ PATH_REC_FIELD(sgid),
178 	  .offset_words = 6,
179 	  .offset_bits  = 0,
180 	  .size_bits    = 128 },
181 	{ PATH_REC_FIELD(dlid),
182 	  .offset_words = 10,
183 	  .offset_bits  = 0,
184 	  .size_bits    = 16 },
185 	{ PATH_REC_FIELD(slid),
186 	  .offset_words = 10,
187 	  .offset_bits  = 16,
188 	  .size_bits    = 16 },
189 	{ PATH_REC_FIELD(raw_traffic),
190 	  .offset_words = 11,
191 	  .offset_bits  = 0,
192 	  .size_bits    = 1 },
193 	{ RESERVED,
194 	  .offset_words = 11,
195 	  .offset_bits  = 1,
196 	  .size_bits    = 3 },
197 	{ PATH_REC_FIELD(flow_label),
198 	  .offset_words = 11,
199 	  .offset_bits  = 4,
200 	  .size_bits    = 20 },
201 	{ PATH_REC_FIELD(hop_limit),
202 	  .offset_words = 11,
203 	  .offset_bits  = 24,
204 	  .size_bits    = 8 },
205 	{ PATH_REC_FIELD(traffic_class),
206 	  .offset_words = 12,
207 	  .offset_bits  = 0,
208 	  .size_bits    = 8 },
209 	{ PATH_REC_FIELD(reversible),
210 	  .offset_words = 12,
211 	  .offset_bits  = 8,
212 	  .size_bits    = 1 },
213 	{ PATH_REC_FIELD(numb_path),
214 	  .offset_words = 12,
215 	  .offset_bits  = 9,
216 	  .size_bits    = 7 },
217 	{ PATH_REC_FIELD(pkey),
218 	  .offset_words = 12,
219 	  .offset_bits  = 16,
220 	  .size_bits    = 16 },
221 	{ PATH_REC_FIELD(qos_class),
222 	  .offset_words = 13,
223 	  .offset_bits  = 0,
224 	  .size_bits    = 12 },
225 	{ PATH_REC_FIELD(sl),
226 	  .offset_words = 13,
227 	  .offset_bits  = 12,
228 	  .size_bits    = 4 },
229 	{ PATH_REC_FIELD(mtu_selector),
230 	  .offset_words = 13,
231 	  .offset_bits  = 16,
232 	  .size_bits    = 2 },
233 	{ PATH_REC_FIELD(mtu),
234 	  .offset_words = 13,
235 	  .offset_bits  = 18,
236 	  .size_bits    = 6 },
237 	{ PATH_REC_FIELD(rate_selector),
238 	  .offset_words = 13,
239 	  .offset_bits  = 24,
240 	  .size_bits    = 2 },
241 	{ PATH_REC_FIELD(rate),
242 	  .offset_words = 13,
243 	  .offset_bits  = 26,
244 	  .size_bits    = 6 },
245 	{ PATH_REC_FIELD(packet_life_time_selector),
246 	  .offset_words = 14,
247 	  .offset_bits  = 0,
248 	  .size_bits    = 2 },
249 	{ PATH_REC_FIELD(packet_life_time),
250 	  .offset_words = 14,
251 	  .offset_bits  = 2,
252 	  .size_bits    = 6 },
253 	{ PATH_REC_FIELD(preference),
254 	  .offset_words = 14,
255 	  .offset_bits  = 8,
256 	  .size_bits    = 8 },
257 	{ RESERVED,
258 	  .offset_words = 14,
259 	  .offset_bits  = 16,
260 	  .size_bits    = 48 },
261 };
262 
263 #define MCMEMBER_REC_FIELD(field) \
264 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
265 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
266 	.field_name          = "sa_mcmember_rec:" #field
267 
268 static const struct ib_field mcmember_rec_table[] = {
269 	{ MCMEMBER_REC_FIELD(mgid),
270 	  .offset_words = 0,
271 	  .offset_bits  = 0,
272 	  .size_bits    = 128 },
273 	{ MCMEMBER_REC_FIELD(port_gid),
274 	  .offset_words = 4,
275 	  .offset_bits  = 0,
276 	  .size_bits    = 128 },
277 	{ MCMEMBER_REC_FIELD(qkey),
278 	  .offset_words = 8,
279 	  .offset_bits  = 0,
280 	  .size_bits    = 32 },
281 	{ MCMEMBER_REC_FIELD(mlid),
282 	  .offset_words = 9,
283 	  .offset_bits  = 0,
284 	  .size_bits    = 16 },
285 	{ MCMEMBER_REC_FIELD(mtu_selector),
286 	  .offset_words = 9,
287 	  .offset_bits  = 16,
288 	  .size_bits    = 2 },
289 	{ MCMEMBER_REC_FIELD(mtu),
290 	  .offset_words = 9,
291 	  .offset_bits  = 18,
292 	  .size_bits    = 6 },
293 	{ MCMEMBER_REC_FIELD(traffic_class),
294 	  .offset_words = 9,
295 	  .offset_bits  = 24,
296 	  .size_bits    = 8 },
297 	{ MCMEMBER_REC_FIELD(pkey),
298 	  .offset_words = 10,
299 	  .offset_bits  = 0,
300 	  .size_bits    = 16 },
301 	{ MCMEMBER_REC_FIELD(rate_selector),
302 	  .offset_words = 10,
303 	  .offset_bits  = 16,
304 	  .size_bits    = 2 },
305 	{ MCMEMBER_REC_FIELD(rate),
306 	  .offset_words = 10,
307 	  .offset_bits  = 18,
308 	  .size_bits    = 6 },
309 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
310 	  .offset_words = 10,
311 	  .offset_bits  = 24,
312 	  .size_bits    = 2 },
313 	{ MCMEMBER_REC_FIELD(packet_life_time),
314 	  .offset_words = 10,
315 	  .offset_bits  = 26,
316 	  .size_bits    = 6 },
317 	{ MCMEMBER_REC_FIELD(sl),
318 	  .offset_words = 11,
319 	  .offset_bits  = 0,
320 	  .size_bits    = 4 },
321 	{ MCMEMBER_REC_FIELD(flow_label),
322 	  .offset_words = 11,
323 	  .offset_bits  = 4,
324 	  .size_bits    = 20 },
325 	{ MCMEMBER_REC_FIELD(hop_limit),
326 	  .offset_words = 11,
327 	  .offset_bits  = 24,
328 	  .size_bits    = 8 },
329 	{ MCMEMBER_REC_FIELD(scope),
330 	  .offset_words = 12,
331 	  .offset_bits  = 0,
332 	  .size_bits    = 4 },
333 	{ MCMEMBER_REC_FIELD(join_state),
334 	  .offset_words = 12,
335 	  .offset_bits  = 4,
336 	  .size_bits    = 4 },
337 	{ MCMEMBER_REC_FIELD(proxy_join),
338 	  .offset_words = 12,
339 	  .offset_bits  = 8,
340 	  .size_bits    = 1 },
341 	{ RESERVED,
342 	  .offset_words = 12,
343 	  .offset_bits  = 9,
344 	  .size_bits    = 23 },
345 };
346 
347 #define SERVICE_REC_FIELD(field) \
348 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
349 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
350 	.field_name          = "sa_service_rec:" #field
351 
352 static const struct ib_field service_rec_table[] = {
353 	{ SERVICE_REC_FIELD(id),
354 	  .offset_words = 0,
355 	  .offset_bits  = 0,
356 	  .size_bits    = 64 },
357 	{ SERVICE_REC_FIELD(gid),
358 	  .offset_words = 2,
359 	  .offset_bits  = 0,
360 	  .size_bits    = 128 },
361 	{ SERVICE_REC_FIELD(pkey),
362 	  .offset_words = 6,
363 	  .offset_bits  = 0,
364 	  .size_bits    = 16 },
365 	{ SERVICE_REC_FIELD(lease),
366 	  .offset_words = 7,
367 	  .offset_bits  = 0,
368 	  .size_bits    = 32 },
369 	{ SERVICE_REC_FIELD(key),
370 	  .offset_words = 8,
371 	  .offset_bits  = 0,
372 	  .size_bits    = 128 },
373 	{ SERVICE_REC_FIELD(name),
374 	  .offset_words = 12,
375 	  .offset_bits  = 0,
376 	  .size_bits    = 64*8 },
377 	{ SERVICE_REC_FIELD(data8),
378 	  .offset_words = 28,
379 	  .offset_bits  = 0,
380 	  .size_bits    = 16*8 },
381 	{ SERVICE_REC_FIELD(data16),
382 	  .offset_words = 32,
383 	  .offset_bits  = 0,
384 	  .size_bits    = 8*16 },
385 	{ SERVICE_REC_FIELD(data32),
386 	  .offset_words = 36,
387 	  .offset_bits  = 0,
388 	  .size_bits    = 4*32 },
389 	{ SERVICE_REC_FIELD(data64),
390 	  .offset_words = 40,
391 	  .offset_bits  = 0,
392 	  .size_bits    = 2*64 },
393 };
394 
395 #define GUIDINFO_REC_FIELD(field) \
396 	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
397 	.struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,	\
398 	.field_name          = "sa_guidinfo_rec:" #field
399 
400 static const struct ib_field guidinfo_rec_table[] = {
401 	{ GUIDINFO_REC_FIELD(lid),
402 	  .offset_words = 0,
403 	  .offset_bits  = 0,
404 	  .size_bits    = 16 },
405 	{ GUIDINFO_REC_FIELD(block_num),
406 	  .offset_words = 0,
407 	  .offset_bits  = 16,
408 	  .size_bits    = 8 },
409 	{ GUIDINFO_REC_FIELD(res1),
410 	  .offset_words = 0,
411 	  .offset_bits  = 24,
412 	  .size_bits    = 8 },
413 	{ GUIDINFO_REC_FIELD(res2),
414 	  .offset_words = 1,
415 	  .offset_bits  = 0,
416 	  .size_bits    = 32 },
417 	{ GUIDINFO_REC_FIELD(guid_info_list),
418 	  .offset_words = 2,
419 	  .offset_bits  = 0,
420 	  .size_bits    = 512 },
421 };
422 
423 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
424 {
425 	query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
426 }
427 
428 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
429 {
430 	return (query->flags & IB_SA_CANCEL);
431 }
432 
433 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
434 				     struct ib_sa_query *query)
435 {
436 	struct ib_sa_path_rec *sa_rec = query->mad_buf->context[1];
437 	struct ib_sa_mad *mad = query->mad_buf->mad;
438 	ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
439 	u16 val16;
440 	u64 val64;
441 	struct rdma_ls_resolve_header *header;
442 
443 	query->mad_buf->context[1] = NULL;
444 
445 	/* Construct the family header first */
446 	header = (struct rdma_ls_resolve_header *)
447 		skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
448 	memcpy(header->device_name, query->port->agent->device->name,
449 	       LS_DEVICE_NAME_MAX);
450 	header->port_num = query->port->port_num;
451 
452 	if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
453 	    sa_rec->reversible != 0)
454 		query->path_use = LS_RESOLVE_PATH_USE_GMP;
455 	else
456 		query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
457 	header->path_use = query->path_use;
458 
459 	/* Now build the attributes */
460 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
461 		val64 = be64_to_cpu(sa_rec->service_id);
462 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
463 			sizeof(val64), &val64);
464 	}
465 	if (comp_mask & IB_SA_PATH_REC_DGID)
466 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
467 			sizeof(sa_rec->dgid), &sa_rec->dgid);
468 	if (comp_mask & IB_SA_PATH_REC_SGID)
469 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
470 			sizeof(sa_rec->sgid), &sa_rec->sgid);
471 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
472 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
473 			sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
474 
475 	if (comp_mask & IB_SA_PATH_REC_PKEY) {
476 		val16 = be16_to_cpu(sa_rec->pkey);
477 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
478 			sizeof(val16), &val16);
479 	}
480 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
481 		val16 = be16_to_cpu(sa_rec->qos_class);
482 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
483 			sizeof(val16), &val16);
484 	}
485 }
486 
487 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
488 {
489 	int len = 0;
490 
491 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
492 		len += nla_total_size(sizeof(u64));
493 	if (comp_mask & IB_SA_PATH_REC_DGID)
494 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
495 	if (comp_mask & IB_SA_PATH_REC_SGID)
496 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
497 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
498 		len += nla_total_size(sizeof(u8));
499 	if (comp_mask & IB_SA_PATH_REC_PKEY)
500 		len += nla_total_size(sizeof(u16));
501 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
502 		len += nla_total_size(sizeof(u16));
503 
504 	/*
505 	 * Make sure that at least some of the required comp_mask bits are
506 	 * set.
507 	 */
508 	if (WARN_ON(len == 0))
509 		return len;
510 
511 	/* Add the family header */
512 	len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
513 
514 	return len;
515 }
516 
517 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
518 {
519 	struct sk_buff *skb = NULL;
520 	struct nlmsghdr *nlh;
521 	void *data;
522 	int ret = 0;
523 	struct ib_sa_mad *mad;
524 	int len;
525 
526 	mad = query->mad_buf->mad;
527 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
528 	if (len <= 0)
529 		return -EMSGSIZE;
530 
531 	skb = nlmsg_new(len, gfp_mask);
532 	if (!skb)
533 		return -ENOMEM;
534 
535 	/* Put nlmsg header only for now */
536 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
537 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
538 	if (!data) {
539 		kfree_skb(skb);
540 		return -EMSGSIZE;
541 	}
542 
543 	/* Add attributes */
544 	ib_nl_set_path_rec_attrs(skb, query);
545 
546 	/* Repair the nlmsg header length */
547 	nlmsg_end(skb, nlh);
548 
549 	ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, gfp_mask);
550 	if (!ret)
551 		ret = len;
552 	else
553 		ret = 0;
554 
555 	return ret;
556 }
557 
558 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
559 {
560 	unsigned long flags;
561 	unsigned long delay;
562 	int ret;
563 
564 	INIT_LIST_HEAD(&query->list);
565 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
566 
567 	/* Put the request on the list first.*/
568 	spin_lock_irqsave(&ib_nl_request_lock, flags);
569 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
570 	query->timeout = delay + jiffies;
571 	list_add_tail(&query->list, &ib_nl_request_list);
572 	/* Start the timeout if this is the only request */
573 	if (ib_nl_request_list.next == &query->list)
574 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
575 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
576 
577 	ret = ib_nl_send_msg(query, gfp_mask);
578 	if (ret <= 0) {
579 		ret = -EIO;
580 		/* Remove the request */
581 		spin_lock_irqsave(&ib_nl_request_lock, flags);
582 		list_del(&query->list);
583 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
584 	} else {
585 		ret = 0;
586 	}
587 
588 	return ret;
589 }
590 
591 static int ib_nl_cancel_request(struct ib_sa_query *query)
592 {
593 	unsigned long flags;
594 	struct ib_sa_query *wait_query;
595 	int found = 0;
596 
597 	spin_lock_irqsave(&ib_nl_request_lock, flags);
598 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
599 		/* Let the timeout to take care of the callback */
600 		if (query == wait_query) {
601 			query->flags |= IB_SA_CANCEL;
602 			query->timeout = jiffies;
603 			list_move(&query->list, &ib_nl_request_list);
604 			found = 1;
605 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
606 			break;
607 		}
608 	}
609 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
610 
611 	return found;
612 }
613 
614 static void send_handler(struct ib_mad_agent *agent,
615 			 struct ib_mad_send_wc *mad_send_wc);
616 
617 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
618 					   const struct nlmsghdr *nlh)
619 {
620 	struct ib_mad_send_wc mad_send_wc;
621 	struct ib_sa_mad *mad = NULL;
622 	const struct nlattr *head, *curr;
623 	struct ib_path_rec_data  *rec;
624 	int len, rem;
625 	u32 mask = 0;
626 	int status = -EIO;
627 
628 	if (query->callback) {
629 		head = (const struct nlattr *) nlmsg_data(nlh);
630 		len = nlmsg_len(nlh);
631 		switch (query->path_use) {
632 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
633 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
634 			break;
635 
636 		case LS_RESOLVE_PATH_USE_ALL:
637 		case LS_RESOLVE_PATH_USE_GMP:
638 		default:
639 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
640 				IB_PATH_BIDIRECTIONAL;
641 			break;
642 		}
643 		nla_for_each_attr(curr, head, len, rem) {
644 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
645 				rec = nla_data(curr);
646 				/*
647 				 * Get the first one. In the future, we may
648 				 * need to get up to 6 pathrecords.
649 				 */
650 				if ((rec->flags & mask) == mask) {
651 					mad = query->mad_buf->mad;
652 					mad->mad_hdr.method |=
653 						IB_MGMT_METHOD_RESP;
654 					memcpy(mad->data, rec->path_rec,
655 					       sizeof(rec->path_rec));
656 					status = 0;
657 					break;
658 				}
659 			}
660 		}
661 		query->callback(query, status, mad);
662 	}
663 
664 	mad_send_wc.send_buf = query->mad_buf;
665 	mad_send_wc.status = IB_WC_SUCCESS;
666 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
667 }
668 
669 static void ib_nl_request_timeout(struct work_struct *work)
670 {
671 	unsigned long flags;
672 	struct ib_sa_query *query;
673 	unsigned long delay;
674 	struct ib_mad_send_wc mad_send_wc;
675 	int ret;
676 
677 	spin_lock_irqsave(&ib_nl_request_lock, flags);
678 	while (!list_empty(&ib_nl_request_list)) {
679 		query = list_entry(ib_nl_request_list.next,
680 				   struct ib_sa_query, list);
681 
682 		if (time_after(query->timeout, jiffies)) {
683 			delay = query->timeout - jiffies;
684 			if ((long)delay <= 0)
685 				delay = 1;
686 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
687 			break;
688 		}
689 
690 		list_del(&query->list);
691 		ib_sa_disable_local_svc(query);
692 		/* Hold the lock to protect against query cancellation */
693 		if (ib_sa_query_cancelled(query))
694 			ret = -1;
695 		else
696 			ret = ib_post_send_mad(query->mad_buf, NULL);
697 		if (ret) {
698 			mad_send_wc.send_buf = query->mad_buf;
699 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
700 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
701 			send_handler(query->port->agent, &mad_send_wc);
702 			spin_lock_irqsave(&ib_nl_request_lock, flags);
703 		}
704 	}
705 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
706 }
707 
708 static int ib_nl_handle_set_timeout(struct sk_buff *skb,
709 				    struct netlink_callback *cb)
710 {
711 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
712 	int timeout, delta, abs_delta;
713 	const struct nlattr *attr;
714 	unsigned long flags;
715 	struct ib_sa_query *query;
716 	long delay = 0;
717 	struct nlattr *tb[LS_NLA_TYPE_MAX];
718 	int ret;
719 
720 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
721 	    !(NETLINK_CB(skb).sk) ||
722 	    !netlink_capable(skb, CAP_NET_ADMIN))
723 		return -EPERM;
724 
725 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
726 			nlmsg_len(nlh), ib_nl_policy);
727 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
728 	if (ret || !attr)
729 		goto settimeout_out;
730 
731 	timeout = *(int *) nla_data(attr);
732 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
733 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
734 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
735 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
736 
737 	delta = timeout - sa_local_svc_timeout_ms;
738 	if (delta < 0)
739 		abs_delta = -delta;
740 	else
741 		abs_delta = delta;
742 
743 	if (delta != 0) {
744 		spin_lock_irqsave(&ib_nl_request_lock, flags);
745 		sa_local_svc_timeout_ms = timeout;
746 		list_for_each_entry(query, &ib_nl_request_list, list) {
747 			if (delta < 0 && abs_delta > query->timeout)
748 				query->timeout = 0;
749 			else
750 				query->timeout += delta;
751 
752 			/* Get the new delay from the first entry */
753 			if (!delay) {
754 				delay = query->timeout - jiffies;
755 				if (delay <= 0)
756 					delay = 1;
757 			}
758 		}
759 		if (delay)
760 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
761 					 (unsigned long)delay);
762 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
763 	}
764 
765 settimeout_out:
766 	return skb->len;
767 }
768 
769 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
770 {
771 	struct nlattr *tb[LS_NLA_TYPE_MAX];
772 	int ret;
773 
774 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
775 		return 0;
776 
777 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
778 			nlmsg_len(nlh), ib_nl_policy);
779 	if (ret)
780 		return 0;
781 
782 	return 1;
783 }
784 
785 static int ib_nl_handle_resolve_resp(struct sk_buff *skb,
786 				     struct netlink_callback *cb)
787 {
788 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
789 	unsigned long flags;
790 	struct ib_sa_query *query;
791 	struct ib_mad_send_buf *send_buf;
792 	struct ib_mad_send_wc mad_send_wc;
793 	int found = 0;
794 	int ret;
795 
796 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
797 	    !(NETLINK_CB(skb).sk) ||
798 	    !netlink_capable(skb, CAP_NET_ADMIN))
799 		return -EPERM;
800 
801 	spin_lock_irqsave(&ib_nl_request_lock, flags);
802 	list_for_each_entry(query, &ib_nl_request_list, list) {
803 		/*
804 		 * If the query is cancelled, let the timeout routine
805 		 * take care of it.
806 		 */
807 		if (nlh->nlmsg_seq == query->seq) {
808 			found = !ib_sa_query_cancelled(query);
809 			if (found)
810 				list_del(&query->list);
811 			break;
812 		}
813 	}
814 
815 	if (!found) {
816 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
817 		goto resp_out;
818 	}
819 
820 	send_buf = query->mad_buf;
821 
822 	if (!ib_nl_is_good_resolve_resp(nlh)) {
823 		/* if the result is a failure, send out the packet via IB */
824 		ib_sa_disable_local_svc(query);
825 		ret = ib_post_send_mad(query->mad_buf, NULL);
826 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
827 		if (ret) {
828 			mad_send_wc.send_buf = send_buf;
829 			mad_send_wc.status = IB_WC_GENERAL_ERR;
830 			send_handler(query->port->agent, &mad_send_wc);
831 		}
832 	} else {
833 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
834 		ib_nl_process_good_resolve_rsp(query, nlh);
835 	}
836 
837 resp_out:
838 	return skb->len;
839 }
840 
841 static struct ibnl_client_cbs ib_sa_cb_table[] = {
842 	[RDMA_NL_LS_OP_RESOLVE] = {
843 		.dump = ib_nl_handle_resolve_resp,
844 		.module = THIS_MODULE },
845 	[RDMA_NL_LS_OP_SET_TIMEOUT] = {
846 		.dump = ib_nl_handle_set_timeout,
847 		.module = THIS_MODULE },
848 };
849 
850 static void free_sm_ah(struct kref *kref)
851 {
852 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
853 
854 	ib_destroy_ah(sm_ah->ah);
855 	kfree(sm_ah);
856 }
857 
858 static void update_sm_ah(struct work_struct *work)
859 {
860 	struct ib_sa_port *port =
861 		container_of(work, struct ib_sa_port, update_task);
862 	struct ib_sa_sm_ah *new_ah;
863 	struct ib_port_attr port_attr;
864 	struct ib_ah_attr   ah_attr;
865 
866 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
867 		printk(KERN_WARNING "Couldn't query port\n");
868 		return;
869 	}
870 
871 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
872 	if (!new_ah) {
873 		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
874 		return;
875 	}
876 
877 	kref_init(&new_ah->ref);
878 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
879 
880 	new_ah->pkey_index = 0;
881 	if (ib_find_pkey(port->agent->device, port->port_num,
882 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
883 		printk(KERN_ERR "Couldn't find index for default PKey\n");
884 
885 	memset(&ah_attr, 0, sizeof ah_attr);
886 	ah_attr.dlid     = port_attr.sm_lid;
887 	ah_attr.sl       = port_attr.sm_sl;
888 	ah_attr.port_num = port->port_num;
889 
890 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
891 	if (IS_ERR(new_ah->ah)) {
892 		printk(KERN_WARNING "Couldn't create new SM AH\n");
893 		kfree(new_ah);
894 		return;
895 	}
896 
897 	spin_lock_irq(&port->ah_lock);
898 	if (port->sm_ah)
899 		kref_put(&port->sm_ah->ref, free_sm_ah);
900 	port->sm_ah = new_ah;
901 	spin_unlock_irq(&port->ah_lock);
902 
903 }
904 
905 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
906 {
907 	if (event->event == IB_EVENT_PORT_ERR    ||
908 	    event->event == IB_EVENT_PORT_ACTIVE ||
909 	    event->event == IB_EVENT_LID_CHANGE  ||
910 	    event->event == IB_EVENT_PKEY_CHANGE ||
911 	    event->event == IB_EVENT_SM_CHANGE   ||
912 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
913 		unsigned long flags;
914 		struct ib_sa_device *sa_dev =
915 			container_of(handler, typeof(*sa_dev), event_handler);
916 		struct ib_sa_port *port =
917 			&sa_dev->port[event->element.port_num - sa_dev->start_port];
918 
919 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
920 			return;
921 
922 		spin_lock_irqsave(&port->ah_lock, flags);
923 		if (port->sm_ah)
924 			kref_put(&port->sm_ah->ref, free_sm_ah);
925 		port->sm_ah = NULL;
926 		spin_unlock_irqrestore(&port->ah_lock, flags);
927 
928 		queue_work(ib_wq, &sa_dev->port[event->element.port_num -
929 					    sa_dev->start_port].update_task);
930 	}
931 }
932 
933 void ib_sa_register_client(struct ib_sa_client *client)
934 {
935 	atomic_set(&client->users, 1);
936 	init_completion(&client->comp);
937 }
938 EXPORT_SYMBOL(ib_sa_register_client);
939 
940 void ib_sa_unregister_client(struct ib_sa_client *client)
941 {
942 	ib_sa_client_put(client);
943 	wait_for_completion(&client->comp);
944 }
945 EXPORT_SYMBOL(ib_sa_unregister_client);
946 
947 /**
948  * ib_sa_cancel_query - try to cancel an SA query
949  * @id:ID of query to cancel
950  * @query:query pointer to cancel
951  *
952  * Try to cancel an SA query.  If the id and query don't match up or
953  * the query has already completed, nothing is done.  Otherwise the
954  * query is canceled and will complete with a status of -EINTR.
955  */
956 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
957 {
958 	unsigned long flags;
959 	struct ib_mad_agent *agent;
960 	struct ib_mad_send_buf *mad_buf;
961 
962 	spin_lock_irqsave(&idr_lock, flags);
963 	if (idr_find(&query_idr, id) != query) {
964 		spin_unlock_irqrestore(&idr_lock, flags);
965 		return;
966 	}
967 	agent = query->port->agent;
968 	mad_buf = query->mad_buf;
969 	spin_unlock_irqrestore(&idr_lock, flags);
970 
971 	/*
972 	 * If the query is still on the netlink request list, schedule
973 	 * it to be cancelled by the timeout routine. Otherwise, it has been
974 	 * sent to the MAD layer and has to be cancelled from there.
975 	 */
976 	if (!ib_nl_cancel_request(query))
977 		ib_cancel_mad(agent, mad_buf);
978 }
979 EXPORT_SYMBOL(ib_sa_cancel_query);
980 
981 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
982 {
983 	struct ib_sa_device *sa_dev;
984 	struct ib_sa_port   *port;
985 	unsigned long flags;
986 	u8 src_path_mask;
987 
988 	sa_dev = ib_get_client_data(device, &sa_client);
989 	if (!sa_dev)
990 		return 0x7f;
991 
992 	port  = &sa_dev->port[port_num - sa_dev->start_port];
993 	spin_lock_irqsave(&port->ah_lock, flags);
994 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
995 	spin_unlock_irqrestore(&port->ah_lock, flags);
996 
997 	return src_path_mask;
998 }
999 
1000 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
1001 			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
1002 {
1003 	int ret;
1004 	u16 gid_index;
1005 	int use_roce;
1006 	struct net_device *ndev = NULL;
1007 
1008 	memset(ah_attr, 0, sizeof *ah_attr);
1009 	ah_attr->dlid = be16_to_cpu(rec->dlid);
1010 	ah_attr->sl = rec->sl;
1011 	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
1012 				 get_src_path_mask(device, port_num);
1013 	ah_attr->port_num = port_num;
1014 	ah_attr->static_rate = rec->rate;
1015 
1016 	use_roce = rdma_cap_eth_ah(device, port_num);
1017 
1018 	if (use_roce) {
1019 		struct net_device *idev;
1020 		struct net_device *resolved_dev;
1021 		struct rdma_dev_addr dev_addr = {.bound_dev_if = rec->ifindex,
1022 						 .net = rec->net ? rec->net :
1023 							 &init_net};
1024 		union {
1025 			struct sockaddr     _sockaddr;
1026 			struct sockaddr_in  _sockaddr_in;
1027 			struct sockaddr_in6 _sockaddr_in6;
1028 		} sgid_addr, dgid_addr;
1029 
1030 		if (!device->get_netdev)
1031 			return -EOPNOTSUPP;
1032 
1033 		rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1034 		rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1035 
1036 		/* validate the route */
1037 		ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1038 					    &dgid_addr._sockaddr, &dev_addr);
1039 		if (ret)
1040 			return ret;
1041 
1042 		if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1043 		     dev_addr.network == RDMA_NETWORK_IPV6) &&
1044 		    rec->gid_type != IB_GID_TYPE_ROCE_UDP_ENCAP)
1045 			return -EINVAL;
1046 
1047 		idev = device->get_netdev(device, port_num);
1048 		if (!idev)
1049 			return -ENODEV;
1050 
1051 		resolved_dev = dev_get_by_index(dev_addr.net,
1052 						dev_addr.bound_dev_if);
1053 		if (resolved_dev->flags & IFF_LOOPBACK) {
1054 			dev_put(resolved_dev);
1055 			resolved_dev = idev;
1056 			dev_hold(resolved_dev);
1057 		}
1058 		ndev = ib_get_ndev_from_path(rec);
1059 		rcu_read_lock();
1060 		if ((ndev && ndev != resolved_dev) ||
1061 		    (resolved_dev != idev &&
1062 		     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1063 			ret = -EHOSTUNREACH;
1064 		rcu_read_unlock();
1065 		dev_put(idev);
1066 		dev_put(resolved_dev);
1067 		if (ret) {
1068 			if (ndev)
1069 				dev_put(ndev);
1070 			return ret;
1071 		}
1072 	}
1073 
1074 	if (rec->hop_limit > 1 || use_roce) {
1075 		ah_attr->ah_flags = IB_AH_GRH;
1076 		ah_attr->grh.dgid = rec->dgid;
1077 
1078 		ret = ib_find_cached_gid_by_port(device, &rec->sgid,
1079 						 rec->gid_type, port_num, ndev,
1080 						 &gid_index);
1081 		if (ret) {
1082 			if (ndev)
1083 				dev_put(ndev);
1084 			return ret;
1085 		}
1086 
1087 		ah_attr->grh.sgid_index    = gid_index;
1088 		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
1089 		ah_attr->grh.hop_limit     = rec->hop_limit;
1090 		ah_attr->grh.traffic_class = rec->traffic_class;
1091 		if (ndev)
1092 			dev_put(ndev);
1093 	}
1094 
1095 	if (use_roce)
1096 		memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
1097 
1098 	return 0;
1099 }
1100 EXPORT_SYMBOL(ib_init_ah_from_path);
1101 
1102 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1103 {
1104 	unsigned long flags;
1105 
1106 	spin_lock_irqsave(&query->port->ah_lock, flags);
1107 	if (!query->port->sm_ah) {
1108 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1109 		return -EAGAIN;
1110 	}
1111 	kref_get(&query->port->sm_ah->ref);
1112 	query->sm_ah = query->port->sm_ah;
1113 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1114 
1115 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1116 					    query->sm_ah->pkey_index,
1117 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1118 					    gfp_mask,
1119 					    IB_MGMT_BASE_VERSION);
1120 	if (IS_ERR(query->mad_buf)) {
1121 		kref_put(&query->sm_ah->ref, free_sm_ah);
1122 		return -ENOMEM;
1123 	}
1124 
1125 	query->mad_buf->ah = query->sm_ah->ah;
1126 
1127 	return 0;
1128 }
1129 
1130 static void free_mad(struct ib_sa_query *query)
1131 {
1132 	ib_free_send_mad(query->mad_buf);
1133 	kref_put(&query->sm_ah->ref, free_sm_ah);
1134 }
1135 
1136 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
1137 {
1138 	unsigned long flags;
1139 
1140 	memset(mad, 0, sizeof *mad);
1141 
1142 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1143 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1144 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1145 
1146 	spin_lock_irqsave(&tid_lock, flags);
1147 	mad->mad_hdr.tid           =
1148 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1149 	spin_unlock_irqrestore(&tid_lock, flags);
1150 }
1151 
1152 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1153 {
1154 	bool preload = gfpflags_allow_blocking(gfp_mask);
1155 	unsigned long flags;
1156 	int ret, id;
1157 
1158 	if (preload)
1159 		idr_preload(gfp_mask);
1160 	spin_lock_irqsave(&idr_lock, flags);
1161 
1162 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1163 
1164 	spin_unlock_irqrestore(&idr_lock, flags);
1165 	if (preload)
1166 		idr_preload_end();
1167 	if (id < 0)
1168 		return id;
1169 
1170 	query->mad_buf->timeout_ms  = timeout_ms;
1171 	query->mad_buf->context[0] = query;
1172 	query->id = id;
1173 
1174 	if (query->flags & IB_SA_ENABLE_LOCAL_SERVICE) {
1175 		if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1176 			if (!ib_nl_make_request(query, gfp_mask))
1177 				return id;
1178 		}
1179 		ib_sa_disable_local_svc(query);
1180 	}
1181 
1182 	ret = ib_post_send_mad(query->mad_buf, NULL);
1183 	if (ret) {
1184 		spin_lock_irqsave(&idr_lock, flags);
1185 		idr_remove(&query_idr, id);
1186 		spin_unlock_irqrestore(&idr_lock, flags);
1187 	}
1188 
1189 	/*
1190 	 * It's not safe to dereference query any more, because the
1191 	 * send may already have completed and freed the query in
1192 	 * another context.
1193 	 */
1194 	return ret ? ret : id;
1195 }
1196 
1197 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
1198 {
1199 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1200 }
1201 EXPORT_SYMBOL(ib_sa_unpack_path);
1202 
1203 void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
1204 {
1205 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1206 }
1207 EXPORT_SYMBOL(ib_sa_pack_path);
1208 
1209 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1210 				    int status,
1211 				    struct ib_sa_mad *mad)
1212 {
1213 	struct ib_sa_path_query *query =
1214 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1215 
1216 	if (mad) {
1217 		struct ib_sa_path_rec rec;
1218 
1219 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1220 			  mad->data, &rec);
1221 		rec.net = NULL;
1222 		rec.ifindex = 0;
1223 		rec.gid_type = IB_GID_TYPE_IB;
1224 		memset(rec.dmac, 0, ETH_ALEN);
1225 		query->callback(status, &rec, query->context);
1226 	} else
1227 		query->callback(status, NULL, query->context);
1228 }
1229 
1230 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1231 {
1232 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
1233 }
1234 
1235 /**
1236  * ib_sa_path_rec_get - Start a Path get query
1237  * @client:SA client
1238  * @device:device to send query on
1239  * @port_num: port number to send query on
1240  * @rec:Path Record to send in query
1241  * @comp_mask:component mask to send in query
1242  * @timeout_ms:time to wait for response
1243  * @gfp_mask:GFP mask to use for internal allocations
1244  * @callback:function called when query completes, times out or is
1245  * canceled
1246  * @context:opaque user context passed to callback
1247  * @sa_query:query context, used to cancel query
1248  *
1249  * Send a Path Record Get query to the SA to look up a path.  The
1250  * callback function will be called when the query completes (or
1251  * fails); status is 0 for a successful response, -EINTR if the query
1252  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1253  * occurred sending the query.  The resp parameter of the callback is
1254  * only valid if status is 0.
1255  *
1256  * If the return value of ib_sa_path_rec_get() is negative, it is an
1257  * error code.  Otherwise it is a query ID that can be used to cancel
1258  * the query.
1259  */
1260 int ib_sa_path_rec_get(struct ib_sa_client *client,
1261 		       struct ib_device *device, u8 port_num,
1262 		       struct ib_sa_path_rec *rec,
1263 		       ib_sa_comp_mask comp_mask,
1264 		       int timeout_ms, gfp_t gfp_mask,
1265 		       void (*callback)(int status,
1266 					struct ib_sa_path_rec *resp,
1267 					void *context),
1268 		       void *context,
1269 		       struct ib_sa_query **sa_query)
1270 {
1271 	struct ib_sa_path_query *query;
1272 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1273 	struct ib_sa_port   *port;
1274 	struct ib_mad_agent *agent;
1275 	struct ib_sa_mad *mad;
1276 	int ret;
1277 
1278 	if (!sa_dev)
1279 		return -ENODEV;
1280 
1281 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1282 	agent = port->agent;
1283 
1284 	query = kzalloc(sizeof(*query), gfp_mask);
1285 	if (!query)
1286 		return -ENOMEM;
1287 
1288 	query->sa_query.port     = port;
1289 	ret = alloc_mad(&query->sa_query, gfp_mask);
1290 	if (ret)
1291 		goto err1;
1292 
1293 	ib_sa_client_get(client);
1294 	query->sa_query.client = client;
1295 	query->callback        = callback;
1296 	query->context         = context;
1297 
1298 	mad = query->sa_query.mad_buf->mad;
1299 	init_mad(mad, agent);
1300 
1301 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1302 	query->sa_query.release  = ib_sa_path_rec_release;
1303 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1304 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1305 	mad->sa_hdr.comp_mask	 = comp_mask;
1306 
1307 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
1308 
1309 	*sa_query = &query->sa_query;
1310 
1311 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1312 	query->sa_query.mad_buf->context[1] = rec;
1313 
1314 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1315 	if (ret < 0)
1316 		goto err2;
1317 
1318 	return ret;
1319 
1320 err2:
1321 	*sa_query = NULL;
1322 	ib_sa_client_put(query->sa_query.client);
1323 	free_mad(&query->sa_query);
1324 
1325 err1:
1326 	kfree(query);
1327 	return ret;
1328 }
1329 EXPORT_SYMBOL(ib_sa_path_rec_get);
1330 
1331 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1332 				    int status,
1333 				    struct ib_sa_mad *mad)
1334 {
1335 	struct ib_sa_service_query *query =
1336 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1337 
1338 	if (mad) {
1339 		struct ib_sa_service_rec rec;
1340 
1341 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1342 			  mad->data, &rec);
1343 		query->callback(status, &rec, query->context);
1344 	} else
1345 		query->callback(status, NULL, query->context);
1346 }
1347 
1348 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1349 {
1350 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1351 }
1352 
1353 /**
1354  * ib_sa_service_rec_query - Start Service Record operation
1355  * @client:SA client
1356  * @device:device to send request on
1357  * @port_num: port number to send request on
1358  * @method:SA method - should be get, set, or delete
1359  * @rec:Service Record to send in request
1360  * @comp_mask:component mask to send in request
1361  * @timeout_ms:time to wait for response
1362  * @gfp_mask:GFP mask to use for internal allocations
1363  * @callback:function called when request completes, times out or is
1364  * canceled
1365  * @context:opaque user context passed to callback
1366  * @sa_query:request context, used to cancel request
1367  *
1368  * Send a Service Record set/get/delete to the SA to register,
1369  * unregister or query a service record.
1370  * The callback function will be called when the request completes (or
1371  * fails); status is 0 for a successful response, -EINTR if the query
1372  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1373  * occurred sending the query.  The resp parameter of the callback is
1374  * only valid if status is 0.
1375  *
1376  * If the return value of ib_sa_service_rec_query() is negative, it is an
1377  * error code.  Otherwise it is a request ID that can be used to cancel
1378  * the query.
1379  */
1380 int ib_sa_service_rec_query(struct ib_sa_client *client,
1381 			    struct ib_device *device, u8 port_num, u8 method,
1382 			    struct ib_sa_service_rec *rec,
1383 			    ib_sa_comp_mask comp_mask,
1384 			    int timeout_ms, gfp_t gfp_mask,
1385 			    void (*callback)(int status,
1386 					     struct ib_sa_service_rec *resp,
1387 					     void *context),
1388 			    void *context,
1389 			    struct ib_sa_query **sa_query)
1390 {
1391 	struct ib_sa_service_query *query;
1392 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1393 	struct ib_sa_port   *port;
1394 	struct ib_mad_agent *agent;
1395 	struct ib_sa_mad *mad;
1396 	int ret;
1397 
1398 	if (!sa_dev)
1399 		return -ENODEV;
1400 
1401 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1402 	agent = port->agent;
1403 
1404 	if (method != IB_MGMT_METHOD_GET &&
1405 	    method != IB_MGMT_METHOD_SET &&
1406 	    method != IB_SA_METHOD_DELETE)
1407 		return -EINVAL;
1408 
1409 	query = kzalloc(sizeof(*query), gfp_mask);
1410 	if (!query)
1411 		return -ENOMEM;
1412 
1413 	query->sa_query.port     = port;
1414 	ret = alloc_mad(&query->sa_query, gfp_mask);
1415 	if (ret)
1416 		goto err1;
1417 
1418 	ib_sa_client_get(client);
1419 	query->sa_query.client = client;
1420 	query->callback        = callback;
1421 	query->context         = context;
1422 
1423 	mad = query->sa_query.mad_buf->mad;
1424 	init_mad(mad, agent);
1425 
1426 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1427 	query->sa_query.release  = ib_sa_service_rec_release;
1428 	mad->mad_hdr.method	 = method;
1429 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1430 	mad->sa_hdr.comp_mask	 = comp_mask;
1431 
1432 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1433 		rec, mad->data);
1434 
1435 	*sa_query = &query->sa_query;
1436 
1437 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1438 	if (ret < 0)
1439 		goto err2;
1440 
1441 	return ret;
1442 
1443 err2:
1444 	*sa_query = NULL;
1445 	ib_sa_client_put(query->sa_query.client);
1446 	free_mad(&query->sa_query);
1447 
1448 err1:
1449 	kfree(query);
1450 	return ret;
1451 }
1452 EXPORT_SYMBOL(ib_sa_service_rec_query);
1453 
1454 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1455 					int status,
1456 					struct ib_sa_mad *mad)
1457 {
1458 	struct ib_sa_mcmember_query *query =
1459 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1460 
1461 	if (mad) {
1462 		struct ib_sa_mcmember_rec rec;
1463 
1464 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1465 			  mad->data, &rec);
1466 		query->callback(status, &rec, query->context);
1467 	} else
1468 		query->callback(status, NULL, query->context);
1469 }
1470 
1471 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1472 {
1473 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1474 }
1475 
1476 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1477 			     struct ib_device *device, u8 port_num,
1478 			     u8 method,
1479 			     struct ib_sa_mcmember_rec *rec,
1480 			     ib_sa_comp_mask comp_mask,
1481 			     int timeout_ms, gfp_t gfp_mask,
1482 			     void (*callback)(int status,
1483 					      struct ib_sa_mcmember_rec *resp,
1484 					      void *context),
1485 			     void *context,
1486 			     struct ib_sa_query **sa_query)
1487 {
1488 	struct ib_sa_mcmember_query *query;
1489 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1490 	struct ib_sa_port   *port;
1491 	struct ib_mad_agent *agent;
1492 	struct ib_sa_mad *mad;
1493 	int ret;
1494 
1495 	if (!sa_dev)
1496 		return -ENODEV;
1497 
1498 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1499 	agent = port->agent;
1500 
1501 	query = kzalloc(sizeof(*query), gfp_mask);
1502 	if (!query)
1503 		return -ENOMEM;
1504 
1505 	query->sa_query.port     = port;
1506 	ret = alloc_mad(&query->sa_query, gfp_mask);
1507 	if (ret)
1508 		goto err1;
1509 
1510 	ib_sa_client_get(client);
1511 	query->sa_query.client = client;
1512 	query->callback        = callback;
1513 	query->context         = context;
1514 
1515 	mad = query->sa_query.mad_buf->mad;
1516 	init_mad(mad, agent);
1517 
1518 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1519 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1520 	mad->mad_hdr.method	 = method;
1521 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1522 	mad->sa_hdr.comp_mask	 = comp_mask;
1523 
1524 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1525 		rec, mad->data);
1526 
1527 	*sa_query = &query->sa_query;
1528 
1529 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1530 	if (ret < 0)
1531 		goto err2;
1532 
1533 	return ret;
1534 
1535 err2:
1536 	*sa_query = NULL;
1537 	ib_sa_client_put(query->sa_query.client);
1538 	free_mad(&query->sa_query);
1539 
1540 err1:
1541 	kfree(query);
1542 	return ret;
1543 }
1544 
1545 /* Support GuidInfoRecord */
1546 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1547 					int status,
1548 					struct ib_sa_mad *mad)
1549 {
1550 	struct ib_sa_guidinfo_query *query =
1551 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1552 
1553 	if (mad) {
1554 		struct ib_sa_guidinfo_rec rec;
1555 
1556 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1557 			  mad->data, &rec);
1558 		query->callback(status, &rec, query->context);
1559 	} else
1560 		query->callback(status, NULL, query->context);
1561 }
1562 
1563 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1564 {
1565 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1566 }
1567 
1568 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1569 			      struct ib_device *device, u8 port_num,
1570 			      struct ib_sa_guidinfo_rec *rec,
1571 			      ib_sa_comp_mask comp_mask, u8 method,
1572 			      int timeout_ms, gfp_t gfp_mask,
1573 			      void (*callback)(int status,
1574 					       struct ib_sa_guidinfo_rec *resp,
1575 					       void *context),
1576 			      void *context,
1577 			      struct ib_sa_query **sa_query)
1578 {
1579 	struct ib_sa_guidinfo_query *query;
1580 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1581 	struct ib_sa_port *port;
1582 	struct ib_mad_agent *agent;
1583 	struct ib_sa_mad *mad;
1584 	int ret;
1585 
1586 	if (!sa_dev)
1587 		return -ENODEV;
1588 
1589 	if (method != IB_MGMT_METHOD_GET &&
1590 	    method != IB_MGMT_METHOD_SET &&
1591 	    method != IB_SA_METHOD_DELETE) {
1592 		return -EINVAL;
1593 	}
1594 
1595 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1596 	agent = port->agent;
1597 
1598 	query = kzalloc(sizeof(*query), gfp_mask);
1599 	if (!query)
1600 		return -ENOMEM;
1601 
1602 	query->sa_query.port = port;
1603 	ret = alloc_mad(&query->sa_query, gfp_mask);
1604 	if (ret)
1605 		goto err1;
1606 
1607 	ib_sa_client_get(client);
1608 	query->sa_query.client = client;
1609 	query->callback        = callback;
1610 	query->context         = context;
1611 
1612 	mad = query->sa_query.mad_buf->mad;
1613 	init_mad(mad, agent);
1614 
1615 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1616 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1617 
1618 	mad->mad_hdr.method	 = method;
1619 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1620 	mad->sa_hdr.comp_mask	 = comp_mask;
1621 
1622 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1623 		mad->data);
1624 
1625 	*sa_query = &query->sa_query;
1626 
1627 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1628 	if (ret < 0)
1629 		goto err2;
1630 
1631 	return ret;
1632 
1633 err2:
1634 	*sa_query = NULL;
1635 	ib_sa_client_put(query->sa_query.client);
1636 	free_mad(&query->sa_query);
1637 
1638 err1:
1639 	kfree(query);
1640 	return ret;
1641 }
1642 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1643 
1644 static void send_handler(struct ib_mad_agent *agent,
1645 			 struct ib_mad_send_wc *mad_send_wc)
1646 {
1647 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1648 	unsigned long flags;
1649 
1650 	if (query->callback)
1651 		switch (mad_send_wc->status) {
1652 		case IB_WC_SUCCESS:
1653 			/* No callback -- already got recv */
1654 			break;
1655 		case IB_WC_RESP_TIMEOUT_ERR:
1656 			query->callback(query, -ETIMEDOUT, NULL);
1657 			break;
1658 		case IB_WC_WR_FLUSH_ERR:
1659 			query->callback(query, -EINTR, NULL);
1660 			break;
1661 		default:
1662 			query->callback(query, -EIO, NULL);
1663 			break;
1664 		}
1665 
1666 	spin_lock_irqsave(&idr_lock, flags);
1667 	idr_remove(&query_idr, query->id);
1668 	spin_unlock_irqrestore(&idr_lock, flags);
1669 
1670 	free_mad(query);
1671 	ib_sa_client_put(query->client);
1672 	query->release(query);
1673 }
1674 
1675 static void recv_handler(struct ib_mad_agent *mad_agent,
1676 			 struct ib_mad_send_buf *send_buf,
1677 			 struct ib_mad_recv_wc *mad_recv_wc)
1678 {
1679 	struct ib_sa_query *query;
1680 
1681 	if (!send_buf)
1682 		return;
1683 
1684 	query = send_buf->context[0];
1685 	if (query->callback) {
1686 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1687 			query->callback(query,
1688 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1689 					-EINVAL : 0,
1690 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
1691 		else
1692 			query->callback(query, -EIO, NULL);
1693 	}
1694 
1695 	ib_free_recv_mad(mad_recv_wc);
1696 }
1697 
1698 static void ib_sa_add_one(struct ib_device *device)
1699 {
1700 	struct ib_sa_device *sa_dev;
1701 	int s, e, i;
1702 	int count = 0;
1703 
1704 	s = rdma_start_port(device);
1705 	e = rdma_end_port(device);
1706 
1707 	sa_dev = kzalloc(sizeof *sa_dev +
1708 			 (e - s + 1) * sizeof (struct ib_sa_port),
1709 			 GFP_KERNEL);
1710 	if (!sa_dev)
1711 		return;
1712 
1713 	sa_dev->start_port = s;
1714 	sa_dev->end_port   = e;
1715 
1716 	for (i = 0; i <= e - s; ++i) {
1717 		spin_lock_init(&sa_dev->port[i].ah_lock);
1718 		if (!rdma_cap_ib_sa(device, i + 1))
1719 			continue;
1720 
1721 		sa_dev->port[i].sm_ah    = NULL;
1722 		sa_dev->port[i].port_num = i + s;
1723 
1724 		sa_dev->port[i].agent =
1725 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1726 					      NULL, 0, send_handler,
1727 					      recv_handler, sa_dev, 0);
1728 		if (IS_ERR(sa_dev->port[i].agent))
1729 			goto err;
1730 
1731 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1732 
1733 		count++;
1734 	}
1735 
1736 	if (!count)
1737 		goto free;
1738 
1739 	ib_set_client_data(device, &sa_client, sa_dev);
1740 
1741 	/*
1742 	 * We register our event handler after everything is set up,
1743 	 * and then update our cached info after the event handler is
1744 	 * registered to avoid any problems if a port changes state
1745 	 * during our initialization.
1746 	 */
1747 
1748 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1749 	if (ib_register_event_handler(&sa_dev->event_handler))
1750 		goto err;
1751 
1752 	for (i = 0; i <= e - s; ++i) {
1753 		if (rdma_cap_ib_sa(device, i + 1))
1754 			update_sm_ah(&sa_dev->port[i].update_task);
1755 	}
1756 
1757 	return;
1758 
1759 err:
1760 	while (--i >= 0) {
1761 		if (rdma_cap_ib_sa(device, i + 1))
1762 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1763 	}
1764 free:
1765 	kfree(sa_dev);
1766 	return;
1767 }
1768 
1769 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
1770 {
1771 	struct ib_sa_device *sa_dev = client_data;
1772 	int i;
1773 
1774 	if (!sa_dev)
1775 		return;
1776 
1777 	ib_unregister_event_handler(&sa_dev->event_handler);
1778 
1779 	flush_workqueue(ib_wq);
1780 
1781 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1782 		if (rdma_cap_ib_sa(device, i + 1)) {
1783 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1784 			if (sa_dev->port[i].sm_ah)
1785 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1786 		}
1787 
1788 	}
1789 
1790 	kfree(sa_dev);
1791 }
1792 
1793 static int __init ib_sa_init(void)
1794 {
1795 	int ret;
1796 
1797 	get_random_bytes(&tid, sizeof tid);
1798 
1799 	atomic_set(&ib_nl_sa_request_seq, 0);
1800 
1801 	ret = ib_register_client(&sa_client);
1802 	if (ret) {
1803 		printk(KERN_ERR "Couldn't register ib_sa client\n");
1804 		goto err1;
1805 	}
1806 
1807 	ret = mcast_init();
1808 	if (ret) {
1809 		printk(KERN_ERR "Couldn't initialize multicast handling\n");
1810 		goto err2;
1811 	}
1812 
1813 	ib_nl_wq = create_singlethread_workqueue("ib_nl_sa_wq");
1814 	if (!ib_nl_wq) {
1815 		ret = -ENOMEM;
1816 		goto err3;
1817 	}
1818 
1819 	if (ibnl_add_client(RDMA_NL_LS, RDMA_NL_LS_NUM_OPS,
1820 			    ib_sa_cb_table)) {
1821 		pr_err("Failed to add netlink callback\n");
1822 		ret = -EINVAL;
1823 		goto err4;
1824 	}
1825 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
1826 
1827 	return 0;
1828 err4:
1829 	destroy_workqueue(ib_nl_wq);
1830 err3:
1831 	mcast_cleanup();
1832 err2:
1833 	ib_unregister_client(&sa_client);
1834 err1:
1835 	return ret;
1836 }
1837 
1838 static void __exit ib_sa_cleanup(void)
1839 {
1840 	ibnl_remove_client(RDMA_NL_LS);
1841 	cancel_delayed_work(&ib_nl_timed_work);
1842 	flush_workqueue(ib_nl_wq);
1843 	destroy_workqueue(ib_nl_wq);
1844 	mcast_cleanup();
1845 	ib_unregister_client(&sa_client);
1846 	idr_destroy(&query_idr);
1847 }
1848 
1849 module_init(ib_sa_init);
1850 module_exit(ib_sa_cleanup);
1851