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 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN		100
57 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT		2000
58 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX		200000
59 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
60 
61 struct ib_sa_sm_ah {
62 	struct ib_ah        *ah;
63 	struct kref          ref;
64 	u16		     pkey_index;
65 	u8		     src_path_mask;
66 };
67 
68 struct ib_sa_port {
69 	struct ib_mad_agent *agent;
70 	struct ib_sa_sm_ah  *sm_ah;
71 	struct work_struct   update_task;
72 	spinlock_t           ah_lock;
73 	u8                   port_num;
74 };
75 
76 struct ib_sa_device {
77 	int                     start_port, end_port;
78 	struct ib_event_handler event_handler;
79 	struct ib_sa_port port[0];
80 };
81 
82 struct ib_sa_query {
83 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
84 	void (*release)(struct ib_sa_query *);
85 	struct ib_sa_client    *client;
86 	struct ib_sa_port      *port;
87 	struct ib_mad_send_buf *mad_buf;
88 	struct ib_sa_sm_ah     *sm_ah;
89 	int			id;
90 	u32			flags;
91 	struct list_head	list; /* Local svc request list */
92 	u32			seq; /* Local svc request sequence number */
93 	unsigned long		timeout; /* Local svc timeout */
94 	u8			path_use; /* How will the pathrecord be used */
95 };
96 
97 #define IB_SA_ENABLE_LOCAL_SERVICE	0x00000001
98 #define IB_SA_CANCEL			0x00000002
99 
100 struct ib_sa_service_query {
101 	void (*callback)(int, struct ib_sa_service_rec *, void *);
102 	void *context;
103 	struct ib_sa_query sa_query;
104 };
105 
106 struct ib_sa_path_query {
107 	void (*callback)(int, struct ib_sa_path_rec *, void *);
108 	void *context;
109 	struct ib_sa_query sa_query;
110 };
111 
112 struct ib_sa_guidinfo_query {
113 	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
114 	void *context;
115 	struct ib_sa_query sa_query;
116 };
117 
118 struct ib_sa_classport_info_query {
119 	void (*callback)(int, struct ib_class_port_info *, void *);
120 	void *context;
121 	struct ib_sa_query sa_query;
122 };
123 
124 struct ib_sa_mcmember_query {
125 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
126 	void *context;
127 	struct ib_sa_query sa_query;
128 };
129 
130 static LIST_HEAD(ib_nl_request_list);
131 static DEFINE_SPINLOCK(ib_nl_request_lock);
132 static atomic_t ib_nl_sa_request_seq;
133 static struct workqueue_struct *ib_nl_wq;
134 static struct delayed_work ib_nl_timed_work;
135 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
136 	[LS_NLA_TYPE_PATH_RECORD]	= {.type = NLA_BINARY,
137 		.len = sizeof(struct ib_path_rec_data)},
138 	[LS_NLA_TYPE_TIMEOUT]		= {.type = NLA_U32},
139 	[LS_NLA_TYPE_SERVICE_ID]	= {.type = NLA_U64},
140 	[LS_NLA_TYPE_DGID]		= {.type = NLA_BINARY,
141 		.len = sizeof(struct rdma_nla_ls_gid)},
142 	[LS_NLA_TYPE_SGID]		= {.type = NLA_BINARY,
143 		.len = sizeof(struct rdma_nla_ls_gid)},
144 	[LS_NLA_TYPE_TCLASS]		= {.type = NLA_U8},
145 	[LS_NLA_TYPE_PKEY]		= {.type = NLA_U16},
146 	[LS_NLA_TYPE_QOS_CLASS]		= {.type = NLA_U16},
147 };
148 
149 
150 static void ib_sa_add_one(struct ib_device *device);
151 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
152 
153 static struct ib_client sa_client = {
154 	.name   = "sa",
155 	.add    = ib_sa_add_one,
156 	.remove = ib_sa_remove_one
157 };
158 
159 static DEFINE_SPINLOCK(idr_lock);
160 static DEFINE_IDR(query_idr);
161 
162 static DEFINE_SPINLOCK(tid_lock);
163 static u32 tid;
164 
165 #define PATH_REC_FIELD(field) \
166 	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
167 	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
168 	.field_name          = "sa_path_rec:" #field
169 
170 static const struct ib_field path_rec_table[] = {
171 	{ PATH_REC_FIELD(service_id),
172 	  .offset_words = 0,
173 	  .offset_bits  = 0,
174 	  .size_bits    = 64 },
175 	{ PATH_REC_FIELD(dgid),
176 	  .offset_words = 2,
177 	  .offset_bits  = 0,
178 	  .size_bits    = 128 },
179 	{ PATH_REC_FIELD(sgid),
180 	  .offset_words = 6,
181 	  .offset_bits  = 0,
182 	  .size_bits    = 128 },
183 	{ PATH_REC_FIELD(dlid),
184 	  .offset_words = 10,
185 	  .offset_bits  = 0,
186 	  .size_bits    = 16 },
187 	{ PATH_REC_FIELD(slid),
188 	  .offset_words = 10,
189 	  .offset_bits  = 16,
190 	  .size_bits    = 16 },
191 	{ PATH_REC_FIELD(raw_traffic),
192 	  .offset_words = 11,
193 	  .offset_bits  = 0,
194 	  .size_bits    = 1 },
195 	{ RESERVED,
196 	  .offset_words = 11,
197 	  .offset_bits  = 1,
198 	  .size_bits    = 3 },
199 	{ PATH_REC_FIELD(flow_label),
200 	  .offset_words = 11,
201 	  .offset_bits  = 4,
202 	  .size_bits    = 20 },
203 	{ PATH_REC_FIELD(hop_limit),
204 	  .offset_words = 11,
205 	  .offset_bits  = 24,
206 	  .size_bits    = 8 },
207 	{ PATH_REC_FIELD(traffic_class),
208 	  .offset_words = 12,
209 	  .offset_bits  = 0,
210 	  .size_bits    = 8 },
211 	{ PATH_REC_FIELD(reversible),
212 	  .offset_words = 12,
213 	  .offset_bits  = 8,
214 	  .size_bits    = 1 },
215 	{ PATH_REC_FIELD(numb_path),
216 	  .offset_words = 12,
217 	  .offset_bits  = 9,
218 	  .size_bits    = 7 },
219 	{ PATH_REC_FIELD(pkey),
220 	  .offset_words = 12,
221 	  .offset_bits  = 16,
222 	  .size_bits    = 16 },
223 	{ PATH_REC_FIELD(qos_class),
224 	  .offset_words = 13,
225 	  .offset_bits  = 0,
226 	  .size_bits    = 12 },
227 	{ PATH_REC_FIELD(sl),
228 	  .offset_words = 13,
229 	  .offset_bits  = 12,
230 	  .size_bits    = 4 },
231 	{ PATH_REC_FIELD(mtu_selector),
232 	  .offset_words = 13,
233 	  .offset_bits  = 16,
234 	  .size_bits    = 2 },
235 	{ PATH_REC_FIELD(mtu),
236 	  .offset_words = 13,
237 	  .offset_bits  = 18,
238 	  .size_bits    = 6 },
239 	{ PATH_REC_FIELD(rate_selector),
240 	  .offset_words = 13,
241 	  .offset_bits  = 24,
242 	  .size_bits    = 2 },
243 	{ PATH_REC_FIELD(rate),
244 	  .offset_words = 13,
245 	  .offset_bits  = 26,
246 	  .size_bits    = 6 },
247 	{ PATH_REC_FIELD(packet_life_time_selector),
248 	  .offset_words = 14,
249 	  .offset_bits  = 0,
250 	  .size_bits    = 2 },
251 	{ PATH_REC_FIELD(packet_life_time),
252 	  .offset_words = 14,
253 	  .offset_bits  = 2,
254 	  .size_bits    = 6 },
255 	{ PATH_REC_FIELD(preference),
256 	  .offset_words = 14,
257 	  .offset_bits  = 8,
258 	  .size_bits    = 8 },
259 	{ RESERVED,
260 	  .offset_words = 14,
261 	  .offset_bits  = 16,
262 	  .size_bits    = 48 },
263 };
264 
265 #define MCMEMBER_REC_FIELD(field) \
266 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
267 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
268 	.field_name          = "sa_mcmember_rec:" #field
269 
270 static const struct ib_field mcmember_rec_table[] = {
271 	{ MCMEMBER_REC_FIELD(mgid),
272 	  .offset_words = 0,
273 	  .offset_bits  = 0,
274 	  .size_bits    = 128 },
275 	{ MCMEMBER_REC_FIELD(port_gid),
276 	  .offset_words = 4,
277 	  .offset_bits  = 0,
278 	  .size_bits    = 128 },
279 	{ MCMEMBER_REC_FIELD(qkey),
280 	  .offset_words = 8,
281 	  .offset_bits  = 0,
282 	  .size_bits    = 32 },
283 	{ MCMEMBER_REC_FIELD(mlid),
284 	  .offset_words = 9,
285 	  .offset_bits  = 0,
286 	  .size_bits    = 16 },
287 	{ MCMEMBER_REC_FIELD(mtu_selector),
288 	  .offset_words = 9,
289 	  .offset_bits  = 16,
290 	  .size_bits    = 2 },
291 	{ MCMEMBER_REC_FIELD(mtu),
292 	  .offset_words = 9,
293 	  .offset_bits  = 18,
294 	  .size_bits    = 6 },
295 	{ MCMEMBER_REC_FIELD(traffic_class),
296 	  .offset_words = 9,
297 	  .offset_bits  = 24,
298 	  .size_bits    = 8 },
299 	{ MCMEMBER_REC_FIELD(pkey),
300 	  .offset_words = 10,
301 	  .offset_bits  = 0,
302 	  .size_bits    = 16 },
303 	{ MCMEMBER_REC_FIELD(rate_selector),
304 	  .offset_words = 10,
305 	  .offset_bits  = 16,
306 	  .size_bits    = 2 },
307 	{ MCMEMBER_REC_FIELD(rate),
308 	  .offset_words = 10,
309 	  .offset_bits  = 18,
310 	  .size_bits    = 6 },
311 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
312 	  .offset_words = 10,
313 	  .offset_bits  = 24,
314 	  .size_bits    = 2 },
315 	{ MCMEMBER_REC_FIELD(packet_life_time),
316 	  .offset_words = 10,
317 	  .offset_bits  = 26,
318 	  .size_bits    = 6 },
319 	{ MCMEMBER_REC_FIELD(sl),
320 	  .offset_words = 11,
321 	  .offset_bits  = 0,
322 	  .size_bits    = 4 },
323 	{ MCMEMBER_REC_FIELD(flow_label),
324 	  .offset_words = 11,
325 	  .offset_bits  = 4,
326 	  .size_bits    = 20 },
327 	{ MCMEMBER_REC_FIELD(hop_limit),
328 	  .offset_words = 11,
329 	  .offset_bits  = 24,
330 	  .size_bits    = 8 },
331 	{ MCMEMBER_REC_FIELD(scope),
332 	  .offset_words = 12,
333 	  .offset_bits  = 0,
334 	  .size_bits    = 4 },
335 	{ MCMEMBER_REC_FIELD(join_state),
336 	  .offset_words = 12,
337 	  .offset_bits  = 4,
338 	  .size_bits    = 4 },
339 	{ MCMEMBER_REC_FIELD(proxy_join),
340 	  .offset_words = 12,
341 	  .offset_bits  = 8,
342 	  .size_bits    = 1 },
343 	{ RESERVED,
344 	  .offset_words = 12,
345 	  .offset_bits  = 9,
346 	  .size_bits    = 23 },
347 };
348 
349 #define SERVICE_REC_FIELD(field) \
350 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
351 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
352 	.field_name          = "sa_service_rec:" #field
353 
354 static const struct ib_field service_rec_table[] = {
355 	{ SERVICE_REC_FIELD(id),
356 	  .offset_words = 0,
357 	  .offset_bits  = 0,
358 	  .size_bits    = 64 },
359 	{ SERVICE_REC_FIELD(gid),
360 	  .offset_words = 2,
361 	  .offset_bits  = 0,
362 	  .size_bits    = 128 },
363 	{ SERVICE_REC_FIELD(pkey),
364 	  .offset_words = 6,
365 	  .offset_bits  = 0,
366 	  .size_bits    = 16 },
367 	{ SERVICE_REC_FIELD(lease),
368 	  .offset_words = 7,
369 	  .offset_bits  = 0,
370 	  .size_bits    = 32 },
371 	{ SERVICE_REC_FIELD(key),
372 	  .offset_words = 8,
373 	  .offset_bits  = 0,
374 	  .size_bits    = 128 },
375 	{ SERVICE_REC_FIELD(name),
376 	  .offset_words = 12,
377 	  .offset_bits  = 0,
378 	  .size_bits    = 64*8 },
379 	{ SERVICE_REC_FIELD(data8),
380 	  .offset_words = 28,
381 	  .offset_bits  = 0,
382 	  .size_bits    = 16*8 },
383 	{ SERVICE_REC_FIELD(data16),
384 	  .offset_words = 32,
385 	  .offset_bits  = 0,
386 	  .size_bits    = 8*16 },
387 	{ SERVICE_REC_FIELD(data32),
388 	  .offset_words = 36,
389 	  .offset_bits  = 0,
390 	  .size_bits    = 4*32 },
391 	{ SERVICE_REC_FIELD(data64),
392 	  .offset_words = 40,
393 	  .offset_bits  = 0,
394 	  .size_bits    = 2*64 },
395 };
396 
397 #define CLASSPORTINFO_REC_FIELD(field) \
398 	.struct_offset_bytes = offsetof(struct ib_class_port_info, field),	\
399 	.struct_size_bytes   = sizeof((struct ib_class_port_info *)0)->field,	\
400 	.field_name          = "ib_class_port_info:" #field
401 
402 static const struct ib_field classport_info_rec_table[] = {
403 	{ CLASSPORTINFO_REC_FIELD(base_version),
404 	  .offset_words = 0,
405 	  .offset_bits  = 0,
406 	  .size_bits    = 8 },
407 	{ CLASSPORTINFO_REC_FIELD(class_version),
408 	  .offset_words = 0,
409 	  .offset_bits  = 8,
410 	  .size_bits    = 8 },
411 	{ CLASSPORTINFO_REC_FIELD(capability_mask),
412 	  .offset_words = 0,
413 	  .offset_bits  = 16,
414 	  .size_bits    = 16 },
415 	{ CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
416 	  .offset_words = 1,
417 	  .offset_bits  = 0,
418 	  .size_bits    = 32 },
419 	{ CLASSPORTINFO_REC_FIELD(redirect_gid),
420 	  .offset_words = 2,
421 	  .offset_bits  = 0,
422 	  .size_bits    = 128 },
423 	{ CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
424 	  .offset_words = 6,
425 	  .offset_bits  = 0,
426 	  .size_bits    = 32 },
427 	{ CLASSPORTINFO_REC_FIELD(redirect_lid),
428 	  .offset_words = 7,
429 	  .offset_bits  = 0,
430 	  .size_bits    = 16 },
431 	{ CLASSPORTINFO_REC_FIELD(redirect_pkey),
432 	  .offset_words = 7,
433 	  .offset_bits  = 16,
434 	  .size_bits    = 16 },
435 
436 	{ CLASSPORTINFO_REC_FIELD(redirect_qp),
437 	  .offset_words = 8,
438 	  .offset_bits  = 0,
439 	  .size_bits    = 32 },
440 	{ CLASSPORTINFO_REC_FIELD(redirect_qkey),
441 	  .offset_words = 9,
442 	  .offset_bits  = 0,
443 	  .size_bits    = 32 },
444 
445 	{ CLASSPORTINFO_REC_FIELD(trap_gid),
446 	  .offset_words = 10,
447 	  .offset_bits  = 0,
448 	  .size_bits    = 128 },
449 	{ CLASSPORTINFO_REC_FIELD(trap_tcslfl),
450 	  .offset_words = 14,
451 	  .offset_bits  = 0,
452 	  .size_bits    = 32 },
453 
454 	{ CLASSPORTINFO_REC_FIELD(trap_lid),
455 	  .offset_words = 15,
456 	  .offset_bits  = 0,
457 	  .size_bits    = 16 },
458 	{ CLASSPORTINFO_REC_FIELD(trap_pkey),
459 	  .offset_words = 15,
460 	  .offset_bits  = 16,
461 	  .size_bits    = 16 },
462 
463 	{ CLASSPORTINFO_REC_FIELD(trap_hlqp),
464 	  .offset_words = 16,
465 	  .offset_bits  = 0,
466 	  .size_bits    = 32 },
467 	{ CLASSPORTINFO_REC_FIELD(trap_qkey),
468 	  .offset_words = 17,
469 	  .offset_bits  = 0,
470 	  .size_bits    = 32 },
471 };
472 
473 #define GUIDINFO_REC_FIELD(field) \
474 	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
475 	.struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,	\
476 	.field_name          = "sa_guidinfo_rec:" #field
477 
478 static const struct ib_field guidinfo_rec_table[] = {
479 	{ GUIDINFO_REC_FIELD(lid),
480 	  .offset_words = 0,
481 	  .offset_bits  = 0,
482 	  .size_bits    = 16 },
483 	{ GUIDINFO_REC_FIELD(block_num),
484 	  .offset_words = 0,
485 	  .offset_bits  = 16,
486 	  .size_bits    = 8 },
487 	{ GUIDINFO_REC_FIELD(res1),
488 	  .offset_words = 0,
489 	  .offset_bits  = 24,
490 	  .size_bits    = 8 },
491 	{ GUIDINFO_REC_FIELD(res2),
492 	  .offset_words = 1,
493 	  .offset_bits  = 0,
494 	  .size_bits    = 32 },
495 	{ GUIDINFO_REC_FIELD(guid_info_list),
496 	  .offset_words = 2,
497 	  .offset_bits  = 0,
498 	  .size_bits    = 512 },
499 };
500 
501 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
502 {
503 	query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
504 }
505 
506 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
507 {
508 	return (query->flags & IB_SA_CANCEL);
509 }
510 
511 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
512 				     struct ib_sa_query *query)
513 {
514 	struct ib_sa_path_rec *sa_rec = query->mad_buf->context[1];
515 	struct ib_sa_mad *mad = query->mad_buf->mad;
516 	ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
517 	u16 val16;
518 	u64 val64;
519 	struct rdma_ls_resolve_header *header;
520 
521 	query->mad_buf->context[1] = NULL;
522 
523 	/* Construct the family header first */
524 	header = (struct rdma_ls_resolve_header *)
525 		skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
526 	memcpy(header->device_name, query->port->agent->device->name,
527 	       LS_DEVICE_NAME_MAX);
528 	header->port_num = query->port->port_num;
529 
530 	if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
531 	    sa_rec->reversible != 0)
532 		query->path_use = LS_RESOLVE_PATH_USE_GMP;
533 	else
534 		query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
535 	header->path_use = query->path_use;
536 
537 	/* Now build the attributes */
538 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
539 		val64 = be64_to_cpu(sa_rec->service_id);
540 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
541 			sizeof(val64), &val64);
542 	}
543 	if (comp_mask & IB_SA_PATH_REC_DGID)
544 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
545 			sizeof(sa_rec->dgid), &sa_rec->dgid);
546 	if (comp_mask & IB_SA_PATH_REC_SGID)
547 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
548 			sizeof(sa_rec->sgid), &sa_rec->sgid);
549 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
550 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
551 			sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
552 
553 	if (comp_mask & IB_SA_PATH_REC_PKEY) {
554 		val16 = be16_to_cpu(sa_rec->pkey);
555 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
556 			sizeof(val16), &val16);
557 	}
558 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
559 		val16 = be16_to_cpu(sa_rec->qos_class);
560 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
561 			sizeof(val16), &val16);
562 	}
563 }
564 
565 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
566 {
567 	int len = 0;
568 
569 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
570 		len += nla_total_size(sizeof(u64));
571 	if (comp_mask & IB_SA_PATH_REC_DGID)
572 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
573 	if (comp_mask & IB_SA_PATH_REC_SGID)
574 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
575 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
576 		len += nla_total_size(sizeof(u8));
577 	if (comp_mask & IB_SA_PATH_REC_PKEY)
578 		len += nla_total_size(sizeof(u16));
579 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
580 		len += nla_total_size(sizeof(u16));
581 
582 	/*
583 	 * Make sure that at least some of the required comp_mask bits are
584 	 * set.
585 	 */
586 	if (WARN_ON(len == 0))
587 		return len;
588 
589 	/* Add the family header */
590 	len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
591 
592 	return len;
593 }
594 
595 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
596 {
597 	struct sk_buff *skb = NULL;
598 	struct nlmsghdr *nlh;
599 	void *data;
600 	int ret = 0;
601 	struct ib_sa_mad *mad;
602 	int len;
603 
604 	mad = query->mad_buf->mad;
605 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
606 	if (len <= 0)
607 		return -EMSGSIZE;
608 
609 	skb = nlmsg_new(len, gfp_mask);
610 	if (!skb)
611 		return -ENOMEM;
612 
613 	/* Put nlmsg header only for now */
614 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
615 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
616 	if (!data) {
617 		nlmsg_free(skb);
618 		return -EMSGSIZE;
619 	}
620 
621 	/* Add attributes */
622 	ib_nl_set_path_rec_attrs(skb, query);
623 
624 	/* Repair the nlmsg header length */
625 	nlmsg_end(skb, nlh);
626 
627 	ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, gfp_mask);
628 	if (!ret)
629 		ret = len;
630 	else
631 		ret = 0;
632 
633 	return ret;
634 }
635 
636 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
637 {
638 	unsigned long flags;
639 	unsigned long delay;
640 	int ret;
641 
642 	INIT_LIST_HEAD(&query->list);
643 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
644 
645 	/* Put the request on the list first.*/
646 	spin_lock_irqsave(&ib_nl_request_lock, flags);
647 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
648 	query->timeout = delay + jiffies;
649 	list_add_tail(&query->list, &ib_nl_request_list);
650 	/* Start the timeout if this is the only request */
651 	if (ib_nl_request_list.next == &query->list)
652 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
653 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
654 
655 	ret = ib_nl_send_msg(query, gfp_mask);
656 	if (ret <= 0) {
657 		ret = -EIO;
658 		/* Remove the request */
659 		spin_lock_irqsave(&ib_nl_request_lock, flags);
660 		list_del(&query->list);
661 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
662 	} else {
663 		ret = 0;
664 	}
665 
666 	return ret;
667 }
668 
669 static int ib_nl_cancel_request(struct ib_sa_query *query)
670 {
671 	unsigned long flags;
672 	struct ib_sa_query *wait_query;
673 	int found = 0;
674 
675 	spin_lock_irqsave(&ib_nl_request_lock, flags);
676 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
677 		/* Let the timeout to take care of the callback */
678 		if (query == wait_query) {
679 			query->flags |= IB_SA_CANCEL;
680 			query->timeout = jiffies;
681 			list_move(&query->list, &ib_nl_request_list);
682 			found = 1;
683 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
684 			break;
685 		}
686 	}
687 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
688 
689 	return found;
690 }
691 
692 static void send_handler(struct ib_mad_agent *agent,
693 			 struct ib_mad_send_wc *mad_send_wc);
694 
695 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
696 					   const struct nlmsghdr *nlh)
697 {
698 	struct ib_mad_send_wc mad_send_wc;
699 	struct ib_sa_mad *mad = NULL;
700 	const struct nlattr *head, *curr;
701 	struct ib_path_rec_data  *rec;
702 	int len, rem;
703 	u32 mask = 0;
704 	int status = -EIO;
705 
706 	if (query->callback) {
707 		head = (const struct nlattr *) nlmsg_data(nlh);
708 		len = nlmsg_len(nlh);
709 		switch (query->path_use) {
710 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
711 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
712 			break;
713 
714 		case LS_RESOLVE_PATH_USE_ALL:
715 		case LS_RESOLVE_PATH_USE_GMP:
716 		default:
717 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
718 				IB_PATH_BIDIRECTIONAL;
719 			break;
720 		}
721 		nla_for_each_attr(curr, head, len, rem) {
722 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
723 				rec = nla_data(curr);
724 				/*
725 				 * Get the first one. In the future, we may
726 				 * need to get up to 6 pathrecords.
727 				 */
728 				if ((rec->flags & mask) == mask) {
729 					mad = query->mad_buf->mad;
730 					mad->mad_hdr.method |=
731 						IB_MGMT_METHOD_RESP;
732 					memcpy(mad->data, rec->path_rec,
733 					       sizeof(rec->path_rec));
734 					status = 0;
735 					break;
736 				}
737 			}
738 		}
739 		query->callback(query, status, mad);
740 	}
741 
742 	mad_send_wc.send_buf = query->mad_buf;
743 	mad_send_wc.status = IB_WC_SUCCESS;
744 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
745 }
746 
747 static void ib_nl_request_timeout(struct work_struct *work)
748 {
749 	unsigned long flags;
750 	struct ib_sa_query *query;
751 	unsigned long delay;
752 	struct ib_mad_send_wc mad_send_wc;
753 	int ret;
754 
755 	spin_lock_irqsave(&ib_nl_request_lock, flags);
756 	while (!list_empty(&ib_nl_request_list)) {
757 		query = list_entry(ib_nl_request_list.next,
758 				   struct ib_sa_query, list);
759 
760 		if (time_after(query->timeout, jiffies)) {
761 			delay = query->timeout - jiffies;
762 			if ((long)delay <= 0)
763 				delay = 1;
764 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
765 			break;
766 		}
767 
768 		list_del(&query->list);
769 		ib_sa_disable_local_svc(query);
770 		/* Hold the lock to protect against query cancellation */
771 		if (ib_sa_query_cancelled(query))
772 			ret = -1;
773 		else
774 			ret = ib_post_send_mad(query->mad_buf, NULL);
775 		if (ret) {
776 			mad_send_wc.send_buf = query->mad_buf;
777 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
778 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
779 			send_handler(query->port->agent, &mad_send_wc);
780 			spin_lock_irqsave(&ib_nl_request_lock, flags);
781 		}
782 	}
783 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
784 }
785 
786 int ib_nl_handle_set_timeout(struct sk_buff *skb,
787 			     struct netlink_callback *cb)
788 {
789 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
790 	int timeout, delta, abs_delta;
791 	const struct nlattr *attr;
792 	unsigned long flags;
793 	struct ib_sa_query *query;
794 	long delay = 0;
795 	struct nlattr *tb[LS_NLA_TYPE_MAX];
796 	int ret;
797 
798 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
799 	    !(NETLINK_CB(skb).sk) ||
800 	    !netlink_capable(skb, CAP_NET_ADMIN))
801 		return -EPERM;
802 
803 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
804 			nlmsg_len(nlh), ib_nl_policy);
805 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
806 	if (ret || !attr)
807 		goto settimeout_out;
808 
809 	timeout = *(int *) nla_data(attr);
810 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
811 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
812 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
813 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
814 
815 	delta = timeout - sa_local_svc_timeout_ms;
816 	if (delta < 0)
817 		abs_delta = -delta;
818 	else
819 		abs_delta = delta;
820 
821 	if (delta != 0) {
822 		spin_lock_irqsave(&ib_nl_request_lock, flags);
823 		sa_local_svc_timeout_ms = timeout;
824 		list_for_each_entry(query, &ib_nl_request_list, list) {
825 			if (delta < 0 && abs_delta > query->timeout)
826 				query->timeout = 0;
827 			else
828 				query->timeout += delta;
829 
830 			/* Get the new delay from the first entry */
831 			if (!delay) {
832 				delay = query->timeout - jiffies;
833 				if (delay <= 0)
834 					delay = 1;
835 			}
836 		}
837 		if (delay)
838 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
839 					 (unsigned long)delay);
840 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
841 	}
842 
843 settimeout_out:
844 	return skb->len;
845 }
846 
847 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
848 {
849 	struct nlattr *tb[LS_NLA_TYPE_MAX];
850 	int ret;
851 
852 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
853 		return 0;
854 
855 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
856 			nlmsg_len(nlh), ib_nl_policy);
857 	if (ret)
858 		return 0;
859 
860 	return 1;
861 }
862 
863 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
864 			      struct netlink_callback *cb)
865 {
866 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
867 	unsigned long flags;
868 	struct ib_sa_query *query;
869 	struct ib_mad_send_buf *send_buf;
870 	struct ib_mad_send_wc mad_send_wc;
871 	int found = 0;
872 	int ret;
873 
874 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
875 	    !(NETLINK_CB(skb).sk) ||
876 	    !netlink_capable(skb, CAP_NET_ADMIN))
877 		return -EPERM;
878 
879 	spin_lock_irqsave(&ib_nl_request_lock, flags);
880 	list_for_each_entry(query, &ib_nl_request_list, list) {
881 		/*
882 		 * If the query is cancelled, let the timeout routine
883 		 * take care of it.
884 		 */
885 		if (nlh->nlmsg_seq == query->seq) {
886 			found = !ib_sa_query_cancelled(query);
887 			if (found)
888 				list_del(&query->list);
889 			break;
890 		}
891 	}
892 
893 	if (!found) {
894 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
895 		goto resp_out;
896 	}
897 
898 	send_buf = query->mad_buf;
899 
900 	if (!ib_nl_is_good_resolve_resp(nlh)) {
901 		/* if the result is a failure, send out the packet via IB */
902 		ib_sa_disable_local_svc(query);
903 		ret = ib_post_send_mad(query->mad_buf, NULL);
904 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
905 		if (ret) {
906 			mad_send_wc.send_buf = send_buf;
907 			mad_send_wc.status = IB_WC_GENERAL_ERR;
908 			send_handler(query->port->agent, &mad_send_wc);
909 		}
910 	} else {
911 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
912 		ib_nl_process_good_resolve_rsp(query, nlh);
913 	}
914 
915 resp_out:
916 	return skb->len;
917 }
918 
919 static void free_sm_ah(struct kref *kref)
920 {
921 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
922 
923 	ib_destroy_ah(sm_ah->ah);
924 	kfree(sm_ah);
925 }
926 
927 static void update_sm_ah(struct work_struct *work)
928 {
929 	struct ib_sa_port *port =
930 		container_of(work, struct ib_sa_port, update_task);
931 	struct ib_sa_sm_ah *new_ah;
932 	struct ib_port_attr port_attr;
933 	struct ib_ah_attr   ah_attr;
934 
935 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
936 		pr_warn("Couldn't query port\n");
937 		return;
938 	}
939 
940 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
941 	if (!new_ah) {
942 		return;
943 	}
944 
945 	kref_init(&new_ah->ref);
946 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
947 
948 	new_ah->pkey_index = 0;
949 	if (ib_find_pkey(port->agent->device, port->port_num,
950 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
951 		pr_err("Couldn't find index for default PKey\n");
952 
953 	memset(&ah_attr, 0, sizeof ah_attr);
954 	ah_attr.dlid     = port_attr.sm_lid;
955 	ah_attr.sl       = port_attr.sm_sl;
956 	ah_attr.port_num = port->port_num;
957 	if (port_attr.grh_required) {
958 		ah_attr.ah_flags = IB_AH_GRH;
959 		ah_attr.grh.dgid.global.subnet_prefix = cpu_to_be64(port_attr.subnet_prefix);
960 		ah_attr.grh.dgid.global.interface_id = cpu_to_be64(IB_SA_WELL_KNOWN_GUID);
961 	}
962 
963 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
964 	if (IS_ERR(new_ah->ah)) {
965 		pr_warn("Couldn't create new SM AH\n");
966 		kfree(new_ah);
967 		return;
968 	}
969 
970 	spin_lock_irq(&port->ah_lock);
971 	if (port->sm_ah)
972 		kref_put(&port->sm_ah->ref, free_sm_ah);
973 	port->sm_ah = new_ah;
974 	spin_unlock_irq(&port->ah_lock);
975 
976 }
977 
978 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
979 {
980 	if (event->event == IB_EVENT_PORT_ERR    ||
981 	    event->event == IB_EVENT_PORT_ACTIVE ||
982 	    event->event == IB_EVENT_LID_CHANGE  ||
983 	    event->event == IB_EVENT_PKEY_CHANGE ||
984 	    event->event == IB_EVENT_SM_CHANGE   ||
985 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
986 		unsigned long flags;
987 		struct ib_sa_device *sa_dev =
988 			container_of(handler, typeof(*sa_dev), event_handler);
989 		struct ib_sa_port *port =
990 			&sa_dev->port[event->element.port_num - sa_dev->start_port];
991 
992 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
993 			return;
994 
995 		spin_lock_irqsave(&port->ah_lock, flags);
996 		if (port->sm_ah)
997 			kref_put(&port->sm_ah->ref, free_sm_ah);
998 		port->sm_ah = NULL;
999 		spin_unlock_irqrestore(&port->ah_lock, flags);
1000 
1001 		queue_work(ib_wq, &sa_dev->port[event->element.port_num -
1002 					    sa_dev->start_port].update_task);
1003 	}
1004 }
1005 
1006 void ib_sa_register_client(struct ib_sa_client *client)
1007 {
1008 	atomic_set(&client->users, 1);
1009 	init_completion(&client->comp);
1010 }
1011 EXPORT_SYMBOL(ib_sa_register_client);
1012 
1013 void ib_sa_unregister_client(struct ib_sa_client *client)
1014 {
1015 	ib_sa_client_put(client);
1016 	wait_for_completion(&client->comp);
1017 }
1018 EXPORT_SYMBOL(ib_sa_unregister_client);
1019 
1020 /**
1021  * ib_sa_cancel_query - try to cancel an SA query
1022  * @id:ID of query to cancel
1023  * @query:query pointer to cancel
1024  *
1025  * Try to cancel an SA query.  If the id and query don't match up or
1026  * the query has already completed, nothing is done.  Otherwise the
1027  * query is canceled and will complete with a status of -EINTR.
1028  */
1029 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1030 {
1031 	unsigned long flags;
1032 	struct ib_mad_agent *agent;
1033 	struct ib_mad_send_buf *mad_buf;
1034 
1035 	spin_lock_irqsave(&idr_lock, flags);
1036 	if (idr_find(&query_idr, id) != query) {
1037 		spin_unlock_irqrestore(&idr_lock, flags);
1038 		return;
1039 	}
1040 	agent = query->port->agent;
1041 	mad_buf = query->mad_buf;
1042 	spin_unlock_irqrestore(&idr_lock, flags);
1043 
1044 	/*
1045 	 * If the query is still on the netlink request list, schedule
1046 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1047 	 * sent to the MAD layer and has to be cancelled from there.
1048 	 */
1049 	if (!ib_nl_cancel_request(query))
1050 		ib_cancel_mad(agent, mad_buf);
1051 }
1052 EXPORT_SYMBOL(ib_sa_cancel_query);
1053 
1054 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1055 {
1056 	struct ib_sa_device *sa_dev;
1057 	struct ib_sa_port   *port;
1058 	unsigned long flags;
1059 	u8 src_path_mask;
1060 
1061 	sa_dev = ib_get_client_data(device, &sa_client);
1062 	if (!sa_dev)
1063 		return 0x7f;
1064 
1065 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1066 	spin_lock_irqsave(&port->ah_lock, flags);
1067 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1068 	spin_unlock_irqrestore(&port->ah_lock, flags);
1069 
1070 	return src_path_mask;
1071 }
1072 
1073 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
1074 			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
1075 {
1076 	int ret;
1077 	u16 gid_index;
1078 	int use_roce;
1079 	struct net_device *ndev = NULL;
1080 
1081 	memset(ah_attr, 0, sizeof *ah_attr);
1082 	ah_attr->dlid = be16_to_cpu(rec->dlid);
1083 	ah_attr->sl = rec->sl;
1084 	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
1085 				 get_src_path_mask(device, port_num);
1086 	ah_attr->port_num = port_num;
1087 	ah_attr->static_rate = rec->rate;
1088 
1089 	use_roce = rdma_cap_eth_ah(device, port_num);
1090 
1091 	if (use_roce) {
1092 		struct net_device *idev;
1093 		struct net_device *resolved_dev;
1094 		struct rdma_dev_addr dev_addr = {.bound_dev_if = rec->ifindex,
1095 						 .net = rec->net ? rec->net :
1096 							 &init_net};
1097 		union {
1098 			struct sockaddr     _sockaddr;
1099 			struct sockaddr_in  _sockaddr_in;
1100 			struct sockaddr_in6 _sockaddr_in6;
1101 		} sgid_addr, dgid_addr;
1102 
1103 		if (!device->get_netdev)
1104 			return -EOPNOTSUPP;
1105 
1106 		rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1107 		rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1108 
1109 		/* validate the route */
1110 		ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1111 					    &dgid_addr._sockaddr, &dev_addr);
1112 		if (ret)
1113 			return ret;
1114 
1115 		if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1116 		     dev_addr.network == RDMA_NETWORK_IPV6) &&
1117 		    rec->gid_type != IB_GID_TYPE_ROCE_UDP_ENCAP)
1118 			return -EINVAL;
1119 
1120 		idev = device->get_netdev(device, port_num);
1121 		if (!idev)
1122 			return -ENODEV;
1123 
1124 		resolved_dev = dev_get_by_index(dev_addr.net,
1125 						dev_addr.bound_dev_if);
1126 		if (resolved_dev->flags & IFF_LOOPBACK) {
1127 			dev_put(resolved_dev);
1128 			resolved_dev = idev;
1129 			dev_hold(resolved_dev);
1130 		}
1131 		ndev = ib_get_ndev_from_path(rec);
1132 		rcu_read_lock();
1133 		if ((ndev && ndev != resolved_dev) ||
1134 		    (resolved_dev != idev &&
1135 		     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1136 			ret = -EHOSTUNREACH;
1137 		rcu_read_unlock();
1138 		dev_put(idev);
1139 		dev_put(resolved_dev);
1140 		if (ret) {
1141 			if (ndev)
1142 				dev_put(ndev);
1143 			return ret;
1144 		}
1145 	}
1146 
1147 	if (rec->hop_limit > 0 || use_roce) {
1148 		ah_attr->ah_flags = IB_AH_GRH;
1149 		ah_attr->grh.dgid = rec->dgid;
1150 
1151 		ret = ib_find_cached_gid_by_port(device, &rec->sgid,
1152 						 rec->gid_type, port_num, ndev,
1153 						 &gid_index);
1154 		if (ret) {
1155 			if (ndev)
1156 				dev_put(ndev);
1157 			return ret;
1158 		}
1159 
1160 		ah_attr->grh.sgid_index    = gid_index;
1161 		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
1162 		ah_attr->grh.hop_limit     = rec->hop_limit;
1163 		ah_attr->grh.traffic_class = rec->traffic_class;
1164 		if (ndev)
1165 			dev_put(ndev);
1166 	}
1167 
1168 	if (use_roce)
1169 		memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
1170 
1171 	return 0;
1172 }
1173 EXPORT_SYMBOL(ib_init_ah_from_path);
1174 
1175 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1176 {
1177 	unsigned long flags;
1178 
1179 	spin_lock_irqsave(&query->port->ah_lock, flags);
1180 	if (!query->port->sm_ah) {
1181 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1182 		return -EAGAIN;
1183 	}
1184 	kref_get(&query->port->sm_ah->ref);
1185 	query->sm_ah = query->port->sm_ah;
1186 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1187 
1188 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1189 					    query->sm_ah->pkey_index,
1190 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1191 					    gfp_mask,
1192 					    IB_MGMT_BASE_VERSION);
1193 	if (IS_ERR(query->mad_buf)) {
1194 		kref_put(&query->sm_ah->ref, free_sm_ah);
1195 		return -ENOMEM;
1196 	}
1197 
1198 	query->mad_buf->ah = query->sm_ah->ah;
1199 
1200 	return 0;
1201 }
1202 
1203 static void free_mad(struct ib_sa_query *query)
1204 {
1205 	ib_free_send_mad(query->mad_buf);
1206 	kref_put(&query->sm_ah->ref, free_sm_ah);
1207 }
1208 
1209 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
1210 {
1211 	unsigned long flags;
1212 
1213 	memset(mad, 0, sizeof *mad);
1214 
1215 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1216 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1217 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1218 
1219 	spin_lock_irqsave(&tid_lock, flags);
1220 	mad->mad_hdr.tid           =
1221 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1222 	spin_unlock_irqrestore(&tid_lock, flags);
1223 }
1224 
1225 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1226 {
1227 	bool preload = gfpflags_allow_blocking(gfp_mask);
1228 	unsigned long flags;
1229 	int ret, id;
1230 
1231 	if (preload)
1232 		idr_preload(gfp_mask);
1233 	spin_lock_irqsave(&idr_lock, flags);
1234 
1235 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1236 
1237 	spin_unlock_irqrestore(&idr_lock, flags);
1238 	if (preload)
1239 		idr_preload_end();
1240 	if (id < 0)
1241 		return id;
1242 
1243 	query->mad_buf->timeout_ms  = timeout_ms;
1244 	query->mad_buf->context[0] = query;
1245 	query->id = id;
1246 
1247 	if (query->flags & IB_SA_ENABLE_LOCAL_SERVICE) {
1248 		if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1249 			if (!ib_nl_make_request(query, gfp_mask))
1250 				return id;
1251 		}
1252 		ib_sa_disable_local_svc(query);
1253 	}
1254 
1255 	ret = ib_post_send_mad(query->mad_buf, NULL);
1256 	if (ret) {
1257 		spin_lock_irqsave(&idr_lock, flags);
1258 		idr_remove(&query_idr, id);
1259 		spin_unlock_irqrestore(&idr_lock, flags);
1260 	}
1261 
1262 	/*
1263 	 * It's not safe to dereference query any more, because the
1264 	 * send may already have completed and freed the query in
1265 	 * another context.
1266 	 */
1267 	return ret ? ret : id;
1268 }
1269 
1270 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
1271 {
1272 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1273 }
1274 EXPORT_SYMBOL(ib_sa_unpack_path);
1275 
1276 void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
1277 {
1278 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1279 }
1280 EXPORT_SYMBOL(ib_sa_pack_path);
1281 
1282 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1283 				    int status,
1284 				    struct ib_sa_mad *mad)
1285 {
1286 	struct ib_sa_path_query *query =
1287 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1288 
1289 	if (mad) {
1290 		struct ib_sa_path_rec rec;
1291 
1292 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1293 			  mad->data, &rec);
1294 		rec.net = NULL;
1295 		rec.ifindex = 0;
1296 		rec.gid_type = IB_GID_TYPE_IB;
1297 		eth_zero_addr(rec.dmac);
1298 		query->callback(status, &rec, query->context);
1299 	} else
1300 		query->callback(status, NULL, query->context);
1301 }
1302 
1303 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1304 {
1305 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
1306 }
1307 
1308 /**
1309  * ib_sa_path_rec_get - Start a Path get query
1310  * @client:SA client
1311  * @device:device to send query on
1312  * @port_num: port number to send query on
1313  * @rec:Path Record to send in query
1314  * @comp_mask:component mask to send in query
1315  * @timeout_ms:time to wait for response
1316  * @gfp_mask:GFP mask to use for internal allocations
1317  * @callback:function called when query completes, times out or is
1318  * canceled
1319  * @context:opaque user context passed to callback
1320  * @sa_query:query context, used to cancel query
1321  *
1322  * Send a Path Record Get query to the SA to look up a path.  The
1323  * callback function will be called when the query completes (or
1324  * fails); status is 0 for a successful response, -EINTR if the query
1325  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1326  * occurred sending the query.  The resp parameter of the callback is
1327  * only valid if status is 0.
1328  *
1329  * If the return value of ib_sa_path_rec_get() is negative, it is an
1330  * error code.  Otherwise it is a query ID that can be used to cancel
1331  * the query.
1332  */
1333 int ib_sa_path_rec_get(struct ib_sa_client *client,
1334 		       struct ib_device *device, u8 port_num,
1335 		       struct ib_sa_path_rec *rec,
1336 		       ib_sa_comp_mask comp_mask,
1337 		       int timeout_ms, gfp_t gfp_mask,
1338 		       void (*callback)(int status,
1339 					struct ib_sa_path_rec *resp,
1340 					void *context),
1341 		       void *context,
1342 		       struct ib_sa_query **sa_query)
1343 {
1344 	struct ib_sa_path_query *query;
1345 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1346 	struct ib_sa_port   *port;
1347 	struct ib_mad_agent *agent;
1348 	struct ib_sa_mad *mad;
1349 	int ret;
1350 
1351 	if (!sa_dev)
1352 		return -ENODEV;
1353 
1354 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1355 	agent = port->agent;
1356 
1357 	query = kzalloc(sizeof(*query), gfp_mask);
1358 	if (!query)
1359 		return -ENOMEM;
1360 
1361 	query->sa_query.port     = port;
1362 	ret = alloc_mad(&query->sa_query, gfp_mask);
1363 	if (ret)
1364 		goto err1;
1365 
1366 	ib_sa_client_get(client);
1367 	query->sa_query.client = client;
1368 	query->callback        = callback;
1369 	query->context         = context;
1370 
1371 	mad = query->sa_query.mad_buf->mad;
1372 	init_mad(mad, agent);
1373 
1374 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1375 	query->sa_query.release  = ib_sa_path_rec_release;
1376 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1377 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1378 	mad->sa_hdr.comp_mask	 = comp_mask;
1379 
1380 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
1381 
1382 	*sa_query = &query->sa_query;
1383 
1384 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1385 	query->sa_query.mad_buf->context[1] = rec;
1386 
1387 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1388 	if (ret < 0)
1389 		goto err2;
1390 
1391 	return ret;
1392 
1393 err2:
1394 	*sa_query = NULL;
1395 	ib_sa_client_put(query->sa_query.client);
1396 	free_mad(&query->sa_query);
1397 
1398 err1:
1399 	kfree(query);
1400 	return ret;
1401 }
1402 EXPORT_SYMBOL(ib_sa_path_rec_get);
1403 
1404 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1405 				    int status,
1406 				    struct ib_sa_mad *mad)
1407 {
1408 	struct ib_sa_service_query *query =
1409 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1410 
1411 	if (mad) {
1412 		struct ib_sa_service_rec rec;
1413 
1414 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1415 			  mad->data, &rec);
1416 		query->callback(status, &rec, query->context);
1417 	} else
1418 		query->callback(status, NULL, query->context);
1419 }
1420 
1421 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1422 {
1423 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1424 }
1425 
1426 /**
1427  * ib_sa_service_rec_query - Start Service Record operation
1428  * @client:SA client
1429  * @device:device to send request on
1430  * @port_num: port number to send request on
1431  * @method:SA method - should be get, set, or delete
1432  * @rec:Service Record to send in request
1433  * @comp_mask:component mask to send in request
1434  * @timeout_ms:time to wait for response
1435  * @gfp_mask:GFP mask to use for internal allocations
1436  * @callback:function called when request completes, times out or is
1437  * canceled
1438  * @context:opaque user context passed to callback
1439  * @sa_query:request context, used to cancel request
1440  *
1441  * Send a Service Record set/get/delete to the SA to register,
1442  * unregister or query a service record.
1443  * The callback function will be called when the request completes (or
1444  * fails); status is 0 for a successful response, -EINTR if the query
1445  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1446  * occurred sending the query.  The resp parameter of the callback is
1447  * only valid if status is 0.
1448  *
1449  * If the return value of ib_sa_service_rec_query() is negative, it is an
1450  * error code.  Otherwise it is a request ID that can be used to cancel
1451  * the query.
1452  */
1453 int ib_sa_service_rec_query(struct ib_sa_client *client,
1454 			    struct ib_device *device, u8 port_num, u8 method,
1455 			    struct ib_sa_service_rec *rec,
1456 			    ib_sa_comp_mask comp_mask,
1457 			    int timeout_ms, gfp_t gfp_mask,
1458 			    void (*callback)(int status,
1459 					     struct ib_sa_service_rec *resp,
1460 					     void *context),
1461 			    void *context,
1462 			    struct ib_sa_query **sa_query)
1463 {
1464 	struct ib_sa_service_query *query;
1465 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1466 	struct ib_sa_port   *port;
1467 	struct ib_mad_agent *agent;
1468 	struct ib_sa_mad *mad;
1469 	int ret;
1470 
1471 	if (!sa_dev)
1472 		return -ENODEV;
1473 
1474 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1475 	agent = port->agent;
1476 
1477 	if (method != IB_MGMT_METHOD_GET &&
1478 	    method != IB_MGMT_METHOD_SET &&
1479 	    method != IB_SA_METHOD_DELETE)
1480 		return -EINVAL;
1481 
1482 	query = kzalloc(sizeof(*query), gfp_mask);
1483 	if (!query)
1484 		return -ENOMEM;
1485 
1486 	query->sa_query.port     = port;
1487 	ret = alloc_mad(&query->sa_query, gfp_mask);
1488 	if (ret)
1489 		goto err1;
1490 
1491 	ib_sa_client_get(client);
1492 	query->sa_query.client = client;
1493 	query->callback        = callback;
1494 	query->context         = context;
1495 
1496 	mad = query->sa_query.mad_buf->mad;
1497 	init_mad(mad, agent);
1498 
1499 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1500 	query->sa_query.release  = ib_sa_service_rec_release;
1501 	mad->mad_hdr.method	 = method;
1502 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1503 	mad->sa_hdr.comp_mask	 = comp_mask;
1504 
1505 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1506 		rec, mad->data);
1507 
1508 	*sa_query = &query->sa_query;
1509 
1510 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1511 	if (ret < 0)
1512 		goto err2;
1513 
1514 	return ret;
1515 
1516 err2:
1517 	*sa_query = NULL;
1518 	ib_sa_client_put(query->sa_query.client);
1519 	free_mad(&query->sa_query);
1520 
1521 err1:
1522 	kfree(query);
1523 	return ret;
1524 }
1525 EXPORT_SYMBOL(ib_sa_service_rec_query);
1526 
1527 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1528 					int status,
1529 					struct ib_sa_mad *mad)
1530 {
1531 	struct ib_sa_mcmember_query *query =
1532 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1533 
1534 	if (mad) {
1535 		struct ib_sa_mcmember_rec rec;
1536 
1537 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1538 			  mad->data, &rec);
1539 		query->callback(status, &rec, query->context);
1540 	} else
1541 		query->callback(status, NULL, query->context);
1542 }
1543 
1544 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1545 {
1546 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1547 }
1548 
1549 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1550 			     struct ib_device *device, u8 port_num,
1551 			     u8 method,
1552 			     struct ib_sa_mcmember_rec *rec,
1553 			     ib_sa_comp_mask comp_mask,
1554 			     int timeout_ms, gfp_t gfp_mask,
1555 			     void (*callback)(int status,
1556 					      struct ib_sa_mcmember_rec *resp,
1557 					      void *context),
1558 			     void *context,
1559 			     struct ib_sa_query **sa_query)
1560 {
1561 	struct ib_sa_mcmember_query *query;
1562 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1563 	struct ib_sa_port   *port;
1564 	struct ib_mad_agent *agent;
1565 	struct ib_sa_mad *mad;
1566 	int ret;
1567 
1568 	if (!sa_dev)
1569 		return -ENODEV;
1570 
1571 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1572 	agent = port->agent;
1573 
1574 	query = kzalloc(sizeof(*query), gfp_mask);
1575 	if (!query)
1576 		return -ENOMEM;
1577 
1578 	query->sa_query.port     = port;
1579 	ret = alloc_mad(&query->sa_query, gfp_mask);
1580 	if (ret)
1581 		goto err1;
1582 
1583 	ib_sa_client_get(client);
1584 	query->sa_query.client = client;
1585 	query->callback        = callback;
1586 	query->context         = context;
1587 
1588 	mad = query->sa_query.mad_buf->mad;
1589 	init_mad(mad, agent);
1590 
1591 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1592 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1593 	mad->mad_hdr.method	 = method;
1594 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1595 	mad->sa_hdr.comp_mask	 = comp_mask;
1596 
1597 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1598 		rec, mad->data);
1599 
1600 	*sa_query = &query->sa_query;
1601 
1602 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1603 	if (ret < 0)
1604 		goto err2;
1605 
1606 	return ret;
1607 
1608 err2:
1609 	*sa_query = NULL;
1610 	ib_sa_client_put(query->sa_query.client);
1611 	free_mad(&query->sa_query);
1612 
1613 err1:
1614 	kfree(query);
1615 	return ret;
1616 }
1617 
1618 /* Support GuidInfoRecord */
1619 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1620 					int status,
1621 					struct ib_sa_mad *mad)
1622 {
1623 	struct ib_sa_guidinfo_query *query =
1624 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1625 
1626 	if (mad) {
1627 		struct ib_sa_guidinfo_rec rec;
1628 
1629 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1630 			  mad->data, &rec);
1631 		query->callback(status, &rec, query->context);
1632 	} else
1633 		query->callback(status, NULL, query->context);
1634 }
1635 
1636 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1637 {
1638 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1639 }
1640 
1641 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1642 			      struct ib_device *device, u8 port_num,
1643 			      struct ib_sa_guidinfo_rec *rec,
1644 			      ib_sa_comp_mask comp_mask, u8 method,
1645 			      int timeout_ms, gfp_t gfp_mask,
1646 			      void (*callback)(int status,
1647 					       struct ib_sa_guidinfo_rec *resp,
1648 					       void *context),
1649 			      void *context,
1650 			      struct ib_sa_query **sa_query)
1651 {
1652 	struct ib_sa_guidinfo_query *query;
1653 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1654 	struct ib_sa_port *port;
1655 	struct ib_mad_agent *agent;
1656 	struct ib_sa_mad *mad;
1657 	int ret;
1658 
1659 	if (!sa_dev)
1660 		return -ENODEV;
1661 
1662 	if (method != IB_MGMT_METHOD_GET &&
1663 	    method != IB_MGMT_METHOD_SET &&
1664 	    method != IB_SA_METHOD_DELETE) {
1665 		return -EINVAL;
1666 	}
1667 
1668 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1669 	agent = port->agent;
1670 
1671 	query = kzalloc(sizeof(*query), gfp_mask);
1672 	if (!query)
1673 		return -ENOMEM;
1674 
1675 	query->sa_query.port = port;
1676 	ret = alloc_mad(&query->sa_query, gfp_mask);
1677 	if (ret)
1678 		goto err1;
1679 
1680 	ib_sa_client_get(client);
1681 	query->sa_query.client = client;
1682 	query->callback        = callback;
1683 	query->context         = context;
1684 
1685 	mad = query->sa_query.mad_buf->mad;
1686 	init_mad(mad, agent);
1687 
1688 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1689 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1690 
1691 	mad->mad_hdr.method	 = method;
1692 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1693 	mad->sa_hdr.comp_mask	 = comp_mask;
1694 
1695 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1696 		mad->data);
1697 
1698 	*sa_query = &query->sa_query;
1699 
1700 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1701 	if (ret < 0)
1702 		goto err2;
1703 
1704 	return ret;
1705 
1706 err2:
1707 	*sa_query = NULL;
1708 	ib_sa_client_put(query->sa_query.client);
1709 	free_mad(&query->sa_query);
1710 
1711 err1:
1712 	kfree(query);
1713 	return ret;
1714 }
1715 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1716 
1717 /* Support get SA ClassPortInfo */
1718 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
1719 					      int status,
1720 					      struct ib_sa_mad *mad)
1721 {
1722 	struct ib_sa_classport_info_query *query =
1723 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
1724 
1725 	if (mad) {
1726 		struct ib_class_port_info rec;
1727 
1728 		ib_unpack(classport_info_rec_table,
1729 			  ARRAY_SIZE(classport_info_rec_table),
1730 			  mad->data, &rec);
1731 		query->callback(status, &rec, query->context);
1732 	} else {
1733 		query->callback(status, NULL, query->context);
1734 	}
1735 }
1736 
1737 static void ib_sa_portclass_info_rec_release(struct ib_sa_query *sa_query)
1738 {
1739 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
1740 			   sa_query));
1741 }
1742 
1743 int ib_sa_classport_info_rec_query(struct ib_sa_client *client,
1744 				   struct ib_device *device, u8 port_num,
1745 				   int timeout_ms, gfp_t gfp_mask,
1746 				   void (*callback)(int status,
1747 						    struct ib_class_port_info *resp,
1748 						    void *context),
1749 				   void *context,
1750 				   struct ib_sa_query **sa_query)
1751 {
1752 	struct ib_sa_classport_info_query *query;
1753 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1754 	struct ib_sa_port *port;
1755 	struct ib_mad_agent *agent;
1756 	struct ib_sa_mad *mad;
1757 	int ret;
1758 
1759 	if (!sa_dev)
1760 		return -ENODEV;
1761 
1762 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1763 	agent = port->agent;
1764 
1765 	query = kzalloc(sizeof(*query), gfp_mask);
1766 	if (!query)
1767 		return -ENOMEM;
1768 
1769 	query->sa_query.port = port;
1770 	ret = alloc_mad(&query->sa_query, gfp_mask);
1771 	if (ret)
1772 		goto err1;
1773 
1774 	ib_sa_client_get(client);
1775 	query->sa_query.client = client;
1776 	query->callback        = callback;
1777 	query->context         = context;
1778 
1779 	mad = query->sa_query.mad_buf->mad;
1780 	init_mad(mad, agent);
1781 
1782 	query->sa_query.callback = callback ? ib_sa_classport_info_rec_callback : NULL;
1783 
1784 	query->sa_query.release  = ib_sa_portclass_info_rec_release;
1785 	/* support GET only */
1786 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1787 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
1788 	mad->sa_hdr.comp_mask	 = 0;
1789 	*sa_query = &query->sa_query;
1790 
1791 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1792 	if (ret < 0)
1793 		goto err2;
1794 
1795 	return ret;
1796 
1797 err2:
1798 	*sa_query = NULL;
1799 	ib_sa_client_put(query->sa_query.client);
1800 	free_mad(&query->sa_query);
1801 
1802 err1:
1803 	kfree(query);
1804 	return ret;
1805 }
1806 EXPORT_SYMBOL(ib_sa_classport_info_rec_query);
1807 
1808 static void send_handler(struct ib_mad_agent *agent,
1809 			 struct ib_mad_send_wc *mad_send_wc)
1810 {
1811 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1812 	unsigned long flags;
1813 
1814 	if (query->callback)
1815 		switch (mad_send_wc->status) {
1816 		case IB_WC_SUCCESS:
1817 			/* No callback -- already got recv */
1818 			break;
1819 		case IB_WC_RESP_TIMEOUT_ERR:
1820 			query->callback(query, -ETIMEDOUT, NULL);
1821 			break;
1822 		case IB_WC_WR_FLUSH_ERR:
1823 			query->callback(query, -EINTR, NULL);
1824 			break;
1825 		default:
1826 			query->callback(query, -EIO, NULL);
1827 			break;
1828 		}
1829 
1830 	spin_lock_irqsave(&idr_lock, flags);
1831 	idr_remove(&query_idr, query->id);
1832 	spin_unlock_irqrestore(&idr_lock, flags);
1833 
1834 	free_mad(query);
1835 	ib_sa_client_put(query->client);
1836 	query->release(query);
1837 }
1838 
1839 static void recv_handler(struct ib_mad_agent *mad_agent,
1840 			 struct ib_mad_send_buf *send_buf,
1841 			 struct ib_mad_recv_wc *mad_recv_wc)
1842 {
1843 	struct ib_sa_query *query;
1844 
1845 	if (!send_buf)
1846 		return;
1847 
1848 	query = send_buf->context[0];
1849 	if (query->callback) {
1850 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1851 			query->callback(query,
1852 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1853 					-EINVAL : 0,
1854 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
1855 		else
1856 			query->callback(query, -EIO, NULL);
1857 	}
1858 
1859 	ib_free_recv_mad(mad_recv_wc);
1860 }
1861 
1862 static void ib_sa_add_one(struct ib_device *device)
1863 {
1864 	struct ib_sa_device *sa_dev;
1865 	int s, e, i;
1866 	int count = 0;
1867 
1868 	s = rdma_start_port(device);
1869 	e = rdma_end_port(device);
1870 
1871 	sa_dev = kzalloc(sizeof *sa_dev +
1872 			 (e - s + 1) * sizeof (struct ib_sa_port),
1873 			 GFP_KERNEL);
1874 	if (!sa_dev)
1875 		return;
1876 
1877 	sa_dev->start_port = s;
1878 	sa_dev->end_port   = e;
1879 
1880 	for (i = 0; i <= e - s; ++i) {
1881 		spin_lock_init(&sa_dev->port[i].ah_lock);
1882 		if (!rdma_cap_ib_sa(device, i + 1))
1883 			continue;
1884 
1885 		sa_dev->port[i].sm_ah    = NULL;
1886 		sa_dev->port[i].port_num = i + s;
1887 
1888 		sa_dev->port[i].agent =
1889 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1890 					      NULL, 0, send_handler,
1891 					      recv_handler, sa_dev, 0);
1892 		if (IS_ERR(sa_dev->port[i].agent))
1893 			goto err;
1894 
1895 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1896 
1897 		count++;
1898 	}
1899 
1900 	if (!count)
1901 		goto free;
1902 
1903 	ib_set_client_data(device, &sa_client, sa_dev);
1904 
1905 	/*
1906 	 * We register our event handler after everything is set up,
1907 	 * and then update our cached info after the event handler is
1908 	 * registered to avoid any problems if a port changes state
1909 	 * during our initialization.
1910 	 */
1911 
1912 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1913 	if (ib_register_event_handler(&sa_dev->event_handler))
1914 		goto err;
1915 
1916 	for (i = 0; i <= e - s; ++i) {
1917 		if (rdma_cap_ib_sa(device, i + 1))
1918 			update_sm_ah(&sa_dev->port[i].update_task);
1919 	}
1920 
1921 	return;
1922 
1923 err:
1924 	while (--i >= 0) {
1925 		if (rdma_cap_ib_sa(device, i + 1))
1926 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1927 	}
1928 free:
1929 	kfree(sa_dev);
1930 	return;
1931 }
1932 
1933 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
1934 {
1935 	struct ib_sa_device *sa_dev = client_data;
1936 	int i;
1937 
1938 	if (!sa_dev)
1939 		return;
1940 
1941 	ib_unregister_event_handler(&sa_dev->event_handler);
1942 
1943 	flush_workqueue(ib_wq);
1944 
1945 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1946 		if (rdma_cap_ib_sa(device, i + 1)) {
1947 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1948 			if (sa_dev->port[i].sm_ah)
1949 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1950 		}
1951 
1952 	}
1953 
1954 	kfree(sa_dev);
1955 }
1956 
1957 int ib_sa_init(void)
1958 {
1959 	int ret;
1960 
1961 	get_random_bytes(&tid, sizeof tid);
1962 
1963 	atomic_set(&ib_nl_sa_request_seq, 0);
1964 
1965 	ret = ib_register_client(&sa_client);
1966 	if (ret) {
1967 		pr_err("Couldn't register ib_sa client\n");
1968 		goto err1;
1969 	}
1970 
1971 	ret = mcast_init();
1972 	if (ret) {
1973 		pr_err("Couldn't initialize multicast handling\n");
1974 		goto err2;
1975 	}
1976 
1977 	ib_nl_wq = create_singlethread_workqueue("ib_nl_sa_wq");
1978 	if (!ib_nl_wq) {
1979 		ret = -ENOMEM;
1980 		goto err3;
1981 	}
1982 
1983 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
1984 
1985 	return 0;
1986 
1987 err3:
1988 	mcast_cleanup();
1989 err2:
1990 	ib_unregister_client(&sa_client);
1991 err1:
1992 	return ret;
1993 }
1994 
1995 void ib_sa_cleanup(void)
1996 {
1997 	cancel_delayed_work(&ib_nl_timed_work);
1998 	flush_workqueue(ib_nl_wq);
1999 	destroy_workqueue(ib_nl_wq);
2000 	mcast_cleanup();
2001 	ib_unregister_client(&sa_client);
2002 	idr_destroy(&query_idr);
2003 }
2004