xref: /linux/drivers/infiniband/core/sa_query.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 "sa.h"
53 
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
56 MODULE_LICENSE("Dual BSD/GPL");
57 
58 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN		100
59 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT		2000
60 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX		200000
61 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
62 
63 struct ib_sa_sm_ah {
64 	struct ib_ah        *ah;
65 	struct kref          ref;
66 	u16		     pkey_index;
67 	u8		     src_path_mask;
68 };
69 
70 struct ib_sa_port {
71 	struct ib_mad_agent *agent;
72 	struct ib_sa_sm_ah  *sm_ah;
73 	struct work_struct   update_task;
74 	spinlock_t           ah_lock;
75 	u8                   port_num;
76 };
77 
78 struct ib_sa_device {
79 	int                     start_port, end_port;
80 	struct ib_event_handler event_handler;
81 	struct ib_sa_port port[0];
82 };
83 
84 struct ib_sa_query {
85 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
86 	void (*release)(struct ib_sa_query *);
87 	struct ib_sa_client    *client;
88 	struct ib_sa_port      *port;
89 	struct ib_mad_send_buf *mad_buf;
90 	struct ib_sa_sm_ah     *sm_ah;
91 	int			id;
92 	u32			flags;
93 	struct list_head	list; /* Local svc request list */
94 	u32			seq; /* Local svc request sequence number */
95 	unsigned long		timeout; /* Local svc timeout */
96 	u8			path_use; /* How will the pathrecord be used */
97 };
98 
99 #define IB_SA_ENABLE_LOCAL_SERVICE	0x00000001
100 #define IB_SA_CANCEL			0x00000002
101 
102 struct ib_sa_service_query {
103 	void (*callback)(int, struct ib_sa_service_rec *, void *);
104 	void *context;
105 	struct ib_sa_query sa_query;
106 };
107 
108 struct ib_sa_path_query {
109 	void (*callback)(int, struct ib_sa_path_rec *, void *);
110 	void *context;
111 	struct ib_sa_query sa_query;
112 };
113 
114 struct ib_sa_guidinfo_query {
115 	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
116 	void *context;
117 	struct ib_sa_query sa_query;
118 };
119 
120 struct ib_sa_mcmember_query {
121 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
122 	void *context;
123 	struct ib_sa_query sa_query;
124 };
125 
126 static LIST_HEAD(ib_nl_request_list);
127 static DEFINE_SPINLOCK(ib_nl_request_lock);
128 static atomic_t ib_nl_sa_request_seq;
129 static struct workqueue_struct *ib_nl_wq;
130 static struct delayed_work ib_nl_timed_work;
131 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
132 	[LS_NLA_TYPE_PATH_RECORD]	= {.type = NLA_BINARY,
133 		.len = sizeof(struct ib_path_rec_data)},
134 	[LS_NLA_TYPE_TIMEOUT]		= {.type = NLA_U32},
135 	[LS_NLA_TYPE_SERVICE_ID]	= {.type = NLA_U64},
136 	[LS_NLA_TYPE_DGID]		= {.type = NLA_BINARY,
137 		.len = sizeof(struct rdma_nla_ls_gid)},
138 	[LS_NLA_TYPE_SGID]		= {.type = NLA_BINARY,
139 		.len = sizeof(struct rdma_nla_ls_gid)},
140 	[LS_NLA_TYPE_TCLASS]		= {.type = NLA_U8},
141 	[LS_NLA_TYPE_PKEY]		= {.type = NLA_U16},
142 	[LS_NLA_TYPE_QOS_CLASS]		= {.type = NLA_U16},
143 };
144 
145 
146 static void ib_sa_add_one(struct ib_device *device);
147 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
148 
149 static struct ib_client sa_client = {
150 	.name   = "sa",
151 	.add    = ib_sa_add_one,
152 	.remove = ib_sa_remove_one
153 };
154 
155 static DEFINE_SPINLOCK(idr_lock);
156 static DEFINE_IDR(query_idr);
157 
158 static DEFINE_SPINLOCK(tid_lock);
159 static u32 tid;
160 
161 #define PATH_REC_FIELD(field) \
162 	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
163 	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
164 	.field_name          = "sa_path_rec:" #field
165 
166 static const struct ib_field path_rec_table[] = {
167 	{ PATH_REC_FIELD(service_id),
168 	  .offset_words = 0,
169 	  .offset_bits  = 0,
170 	  .size_bits    = 64 },
171 	{ PATH_REC_FIELD(dgid),
172 	  .offset_words = 2,
173 	  .offset_bits  = 0,
174 	  .size_bits    = 128 },
175 	{ PATH_REC_FIELD(sgid),
176 	  .offset_words = 6,
177 	  .offset_bits  = 0,
178 	  .size_bits    = 128 },
179 	{ PATH_REC_FIELD(dlid),
180 	  .offset_words = 10,
181 	  .offset_bits  = 0,
182 	  .size_bits    = 16 },
183 	{ PATH_REC_FIELD(slid),
184 	  .offset_words = 10,
185 	  .offset_bits  = 16,
186 	  .size_bits    = 16 },
187 	{ PATH_REC_FIELD(raw_traffic),
188 	  .offset_words = 11,
189 	  .offset_bits  = 0,
190 	  .size_bits    = 1 },
191 	{ RESERVED,
192 	  .offset_words = 11,
193 	  .offset_bits  = 1,
194 	  .size_bits    = 3 },
195 	{ PATH_REC_FIELD(flow_label),
196 	  .offset_words = 11,
197 	  .offset_bits  = 4,
198 	  .size_bits    = 20 },
199 	{ PATH_REC_FIELD(hop_limit),
200 	  .offset_words = 11,
201 	  .offset_bits  = 24,
202 	  .size_bits    = 8 },
203 	{ PATH_REC_FIELD(traffic_class),
204 	  .offset_words = 12,
205 	  .offset_bits  = 0,
206 	  .size_bits    = 8 },
207 	{ PATH_REC_FIELD(reversible),
208 	  .offset_words = 12,
209 	  .offset_bits  = 8,
210 	  .size_bits    = 1 },
211 	{ PATH_REC_FIELD(numb_path),
212 	  .offset_words = 12,
213 	  .offset_bits  = 9,
214 	  .size_bits    = 7 },
215 	{ PATH_REC_FIELD(pkey),
216 	  .offset_words = 12,
217 	  .offset_bits  = 16,
218 	  .size_bits    = 16 },
219 	{ PATH_REC_FIELD(qos_class),
220 	  .offset_words = 13,
221 	  .offset_bits  = 0,
222 	  .size_bits    = 12 },
223 	{ PATH_REC_FIELD(sl),
224 	  .offset_words = 13,
225 	  .offset_bits  = 12,
226 	  .size_bits    = 4 },
227 	{ PATH_REC_FIELD(mtu_selector),
228 	  .offset_words = 13,
229 	  .offset_bits  = 16,
230 	  .size_bits    = 2 },
231 	{ PATH_REC_FIELD(mtu),
232 	  .offset_words = 13,
233 	  .offset_bits  = 18,
234 	  .size_bits    = 6 },
235 	{ PATH_REC_FIELD(rate_selector),
236 	  .offset_words = 13,
237 	  .offset_bits  = 24,
238 	  .size_bits    = 2 },
239 	{ PATH_REC_FIELD(rate),
240 	  .offset_words = 13,
241 	  .offset_bits  = 26,
242 	  .size_bits    = 6 },
243 	{ PATH_REC_FIELD(packet_life_time_selector),
244 	  .offset_words = 14,
245 	  .offset_bits  = 0,
246 	  .size_bits    = 2 },
247 	{ PATH_REC_FIELD(packet_life_time),
248 	  .offset_words = 14,
249 	  .offset_bits  = 2,
250 	  .size_bits    = 6 },
251 	{ PATH_REC_FIELD(preference),
252 	  .offset_words = 14,
253 	  .offset_bits  = 8,
254 	  .size_bits    = 8 },
255 	{ RESERVED,
256 	  .offset_words = 14,
257 	  .offset_bits  = 16,
258 	  .size_bits    = 48 },
259 };
260 
261 #define MCMEMBER_REC_FIELD(field) \
262 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
263 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
264 	.field_name          = "sa_mcmember_rec:" #field
265 
266 static const struct ib_field mcmember_rec_table[] = {
267 	{ MCMEMBER_REC_FIELD(mgid),
268 	  .offset_words = 0,
269 	  .offset_bits  = 0,
270 	  .size_bits    = 128 },
271 	{ MCMEMBER_REC_FIELD(port_gid),
272 	  .offset_words = 4,
273 	  .offset_bits  = 0,
274 	  .size_bits    = 128 },
275 	{ MCMEMBER_REC_FIELD(qkey),
276 	  .offset_words = 8,
277 	  .offset_bits  = 0,
278 	  .size_bits    = 32 },
279 	{ MCMEMBER_REC_FIELD(mlid),
280 	  .offset_words = 9,
281 	  .offset_bits  = 0,
282 	  .size_bits    = 16 },
283 	{ MCMEMBER_REC_FIELD(mtu_selector),
284 	  .offset_words = 9,
285 	  .offset_bits  = 16,
286 	  .size_bits    = 2 },
287 	{ MCMEMBER_REC_FIELD(mtu),
288 	  .offset_words = 9,
289 	  .offset_bits  = 18,
290 	  .size_bits    = 6 },
291 	{ MCMEMBER_REC_FIELD(traffic_class),
292 	  .offset_words = 9,
293 	  .offset_bits  = 24,
294 	  .size_bits    = 8 },
295 	{ MCMEMBER_REC_FIELD(pkey),
296 	  .offset_words = 10,
297 	  .offset_bits  = 0,
298 	  .size_bits    = 16 },
299 	{ MCMEMBER_REC_FIELD(rate_selector),
300 	  .offset_words = 10,
301 	  .offset_bits  = 16,
302 	  .size_bits    = 2 },
303 	{ MCMEMBER_REC_FIELD(rate),
304 	  .offset_words = 10,
305 	  .offset_bits  = 18,
306 	  .size_bits    = 6 },
307 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
308 	  .offset_words = 10,
309 	  .offset_bits  = 24,
310 	  .size_bits    = 2 },
311 	{ MCMEMBER_REC_FIELD(packet_life_time),
312 	  .offset_words = 10,
313 	  .offset_bits  = 26,
314 	  .size_bits    = 6 },
315 	{ MCMEMBER_REC_FIELD(sl),
316 	  .offset_words = 11,
317 	  .offset_bits  = 0,
318 	  .size_bits    = 4 },
319 	{ MCMEMBER_REC_FIELD(flow_label),
320 	  .offset_words = 11,
321 	  .offset_bits  = 4,
322 	  .size_bits    = 20 },
323 	{ MCMEMBER_REC_FIELD(hop_limit),
324 	  .offset_words = 11,
325 	  .offset_bits  = 24,
326 	  .size_bits    = 8 },
327 	{ MCMEMBER_REC_FIELD(scope),
328 	  .offset_words = 12,
329 	  .offset_bits  = 0,
330 	  .size_bits    = 4 },
331 	{ MCMEMBER_REC_FIELD(join_state),
332 	  .offset_words = 12,
333 	  .offset_bits  = 4,
334 	  .size_bits    = 4 },
335 	{ MCMEMBER_REC_FIELD(proxy_join),
336 	  .offset_words = 12,
337 	  .offset_bits  = 8,
338 	  .size_bits    = 1 },
339 	{ RESERVED,
340 	  .offset_words = 12,
341 	  .offset_bits  = 9,
342 	  .size_bits    = 23 },
343 };
344 
345 #define SERVICE_REC_FIELD(field) \
346 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
347 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
348 	.field_name          = "sa_service_rec:" #field
349 
350 static const struct ib_field service_rec_table[] = {
351 	{ SERVICE_REC_FIELD(id),
352 	  .offset_words = 0,
353 	  .offset_bits  = 0,
354 	  .size_bits    = 64 },
355 	{ SERVICE_REC_FIELD(gid),
356 	  .offset_words = 2,
357 	  .offset_bits  = 0,
358 	  .size_bits    = 128 },
359 	{ SERVICE_REC_FIELD(pkey),
360 	  .offset_words = 6,
361 	  .offset_bits  = 0,
362 	  .size_bits    = 16 },
363 	{ SERVICE_REC_FIELD(lease),
364 	  .offset_words = 7,
365 	  .offset_bits  = 0,
366 	  .size_bits    = 32 },
367 	{ SERVICE_REC_FIELD(key),
368 	  .offset_words = 8,
369 	  .offset_bits  = 0,
370 	  .size_bits    = 128 },
371 	{ SERVICE_REC_FIELD(name),
372 	  .offset_words = 12,
373 	  .offset_bits  = 0,
374 	  .size_bits    = 64*8 },
375 	{ SERVICE_REC_FIELD(data8),
376 	  .offset_words = 28,
377 	  .offset_bits  = 0,
378 	  .size_bits    = 16*8 },
379 	{ SERVICE_REC_FIELD(data16),
380 	  .offset_words = 32,
381 	  .offset_bits  = 0,
382 	  .size_bits    = 8*16 },
383 	{ SERVICE_REC_FIELD(data32),
384 	  .offset_words = 36,
385 	  .offset_bits  = 0,
386 	  .size_bits    = 4*32 },
387 	{ SERVICE_REC_FIELD(data64),
388 	  .offset_words = 40,
389 	  .offset_bits  = 0,
390 	  .size_bits    = 2*64 },
391 };
392 
393 #define GUIDINFO_REC_FIELD(field) \
394 	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
395 	.struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,	\
396 	.field_name          = "sa_guidinfo_rec:" #field
397 
398 static const struct ib_field guidinfo_rec_table[] = {
399 	{ GUIDINFO_REC_FIELD(lid),
400 	  .offset_words = 0,
401 	  .offset_bits  = 0,
402 	  .size_bits    = 16 },
403 	{ GUIDINFO_REC_FIELD(block_num),
404 	  .offset_words = 0,
405 	  .offset_bits  = 16,
406 	  .size_bits    = 8 },
407 	{ GUIDINFO_REC_FIELD(res1),
408 	  .offset_words = 0,
409 	  .offset_bits  = 24,
410 	  .size_bits    = 8 },
411 	{ GUIDINFO_REC_FIELD(res2),
412 	  .offset_words = 1,
413 	  .offset_bits  = 0,
414 	  .size_bits    = 32 },
415 	{ GUIDINFO_REC_FIELD(guid_info_list),
416 	  .offset_words = 2,
417 	  .offset_bits  = 0,
418 	  .size_bits    = 512 },
419 };
420 
421 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
422 {
423 	query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
424 }
425 
426 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
427 {
428 	return (query->flags & IB_SA_CANCEL);
429 }
430 
431 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
432 				     struct ib_sa_query *query)
433 {
434 	struct ib_sa_path_rec *sa_rec = query->mad_buf->context[1];
435 	struct ib_sa_mad *mad = query->mad_buf->mad;
436 	ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
437 	u16 val16;
438 	u64 val64;
439 	struct rdma_ls_resolve_header *header;
440 
441 	query->mad_buf->context[1] = NULL;
442 
443 	/* Construct the family header first */
444 	header = (struct rdma_ls_resolve_header *)
445 		skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
446 	memcpy(header->device_name, query->port->agent->device->name,
447 	       LS_DEVICE_NAME_MAX);
448 	header->port_num = query->port->port_num;
449 
450 	if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
451 	    sa_rec->reversible != 0)
452 		query->path_use = LS_RESOLVE_PATH_USE_GMP;
453 	else
454 		query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
455 	header->path_use = query->path_use;
456 
457 	/* Now build the attributes */
458 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
459 		val64 = be64_to_cpu(sa_rec->service_id);
460 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
461 			sizeof(val64), &val64);
462 	}
463 	if (comp_mask & IB_SA_PATH_REC_DGID)
464 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
465 			sizeof(sa_rec->dgid), &sa_rec->dgid);
466 	if (comp_mask & IB_SA_PATH_REC_SGID)
467 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
468 			sizeof(sa_rec->sgid), &sa_rec->sgid);
469 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
470 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
471 			sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
472 
473 	if (comp_mask & IB_SA_PATH_REC_PKEY) {
474 		val16 = be16_to_cpu(sa_rec->pkey);
475 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
476 			sizeof(val16), &val16);
477 	}
478 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
479 		val16 = be16_to_cpu(sa_rec->qos_class);
480 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
481 			sizeof(val16), &val16);
482 	}
483 }
484 
485 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
486 {
487 	int len = 0;
488 
489 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
490 		len += nla_total_size(sizeof(u64));
491 	if (comp_mask & IB_SA_PATH_REC_DGID)
492 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
493 	if (comp_mask & IB_SA_PATH_REC_SGID)
494 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
495 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
496 		len += nla_total_size(sizeof(u8));
497 	if (comp_mask & IB_SA_PATH_REC_PKEY)
498 		len += nla_total_size(sizeof(u16));
499 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
500 		len += nla_total_size(sizeof(u16));
501 
502 	/*
503 	 * Make sure that at least some of the required comp_mask bits are
504 	 * set.
505 	 */
506 	if (WARN_ON(len == 0))
507 		return len;
508 
509 	/* Add the family header */
510 	len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
511 
512 	return len;
513 }
514 
515 static int ib_nl_send_msg(struct ib_sa_query *query)
516 {
517 	struct sk_buff *skb = NULL;
518 	struct nlmsghdr *nlh;
519 	void *data;
520 	int ret = 0;
521 	struct ib_sa_mad *mad;
522 	int len;
523 
524 	mad = query->mad_buf->mad;
525 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
526 	if (len <= 0)
527 		return -EMSGSIZE;
528 
529 	skb = nlmsg_new(len, GFP_KERNEL);
530 	if (!skb)
531 		return -ENOMEM;
532 
533 	/* Put nlmsg header only for now */
534 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
535 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
536 	if (!data) {
537 		kfree_skb(skb);
538 		return -EMSGSIZE;
539 	}
540 
541 	/* Add attributes */
542 	ib_nl_set_path_rec_attrs(skb, query);
543 
544 	/* Repair the nlmsg header length */
545 	nlmsg_end(skb, nlh);
546 
547 	ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, GFP_KERNEL);
548 	if (!ret)
549 		ret = len;
550 	else
551 		ret = 0;
552 
553 	return ret;
554 }
555 
556 static int ib_nl_make_request(struct ib_sa_query *query)
557 {
558 	unsigned long flags;
559 	unsigned long delay;
560 	int ret;
561 
562 	INIT_LIST_HEAD(&query->list);
563 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
564 
565 	spin_lock_irqsave(&ib_nl_request_lock, flags);
566 	ret = ib_nl_send_msg(query);
567 	if (ret <= 0) {
568 		ret = -EIO;
569 		goto request_out;
570 	} else {
571 		ret = 0;
572 	}
573 
574 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
575 	query->timeout = delay + jiffies;
576 	list_add_tail(&query->list, &ib_nl_request_list);
577 	/* Start the timeout if this is the only request */
578 	if (ib_nl_request_list.next == &query->list)
579 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
580 
581 request_out:
582 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
583 
584 	return ret;
585 }
586 
587 static int ib_nl_cancel_request(struct ib_sa_query *query)
588 {
589 	unsigned long flags;
590 	struct ib_sa_query *wait_query;
591 	int found = 0;
592 
593 	spin_lock_irqsave(&ib_nl_request_lock, flags);
594 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
595 		/* Let the timeout to take care of the callback */
596 		if (query == wait_query) {
597 			query->flags |= IB_SA_CANCEL;
598 			query->timeout = jiffies;
599 			list_move(&query->list, &ib_nl_request_list);
600 			found = 1;
601 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
602 			break;
603 		}
604 	}
605 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
606 
607 	return found;
608 }
609 
610 static void send_handler(struct ib_mad_agent *agent,
611 			 struct ib_mad_send_wc *mad_send_wc);
612 
613 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
614 					   const struct nlmsghdr *nlh)
615 {
616 	struct ib_mad_send_wc mad_send_wc;
617 	struct ib_sa_mad *mad = NULL;
618 	const struct nlattr *head, *curr;
619 	struct ib_path_rec_data  *rec;
620 	int len, rem;
621 	u32 mask = 0;
622 	int status = -EIO;
623 
624 	if (query->callback) {
625 		head = (const struct nlattr *) nlmsg_data(nlh);
626 		len = nlmsg_len(nlh);
627 		switch (query->path_use) {
628 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
629 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
630 			break;
631 
632 		case LS_RESOLVE_PATH_USE_ALL:
633 		case LS_RESOLVE_PATH_USE_GMP:
634 		default:
635 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
636 				IB_PATH_BIDIRECTIONAL;
637 			break;
638 		}
639 		nla_for_each_attr(curr, head, len, rem) {
640 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
641 				rec = nla_data(curr);
642 				/*
643 				 * Get the first one. In the future, we may
644 				 * need to get up to 6 pathrecords.
645 				 */
646 				if ((rec->flags & mask) == mask) {
647 					mad = query->mad_buf->mad;
648 					mad->mad_hdr.method |=
649 						IB_MGMT_METHOD_RESP;
650 					memcpy(mad->data, rec->path_rec,
651 					       sizeof(rec->path_rec));
652 					status = 0;
653 					break;
654 				}
655 			}
656 		}
657 		query->callback(query, status, mad);
658 	}
659 
660 	mad_send_wc.send_buf = query->mad_buf;
661 	mad_send_wc.status = IB_WC_SUCCESS;
662 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
663 }
664 
665 static void ib_nl_request_timeout(struct work_struct *work)
666 {
667 	unsigned long flags;
668 	struct ib_sa_query *query;
669 	unsigned long delay;
670 	struct ib_mad_send_wc mad_send_wc;
671 	int ret;
672 
673 	spin_lock_irqsave(&ib_nl_request_lock, flags);
674 	while (!list_empty(&ib_nl_request_list)) {
675 		query = list_entry(ib_nl_request_list.next,
676 				   struct ib_sa_query, list);
677 
678 		if (time_after(query->timeout, jiffies)) {
679 			delay = query->timeout - jiffies;
680 			if ((long)delay <= 0)
681 				delay = 1;
682 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
683 			break;
684 		}
685 
686 		list_del(&query->list);
687 		ib_sa_disable_local_svc(query);
688 		/* Hold the lock to protect against query cancellation */
689 		if (ib_sa_query_cancelled(query))
690 			ret = -1;
691 		else
692 			ret = ib_post_send_mad(query->mad_buf, NULL);
693 		if (ret) {
694 			mad_send_wc.send_buf = query->mad_buf;
695 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
696 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
697 			send_handler(query->port->agent, &mad_send_wc);
698 			spin_lock_irqsave(&ib_nl_request_lock, flags);
699 		}
700 	}
701 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
702 }
703 
704 static int ib_nl_handle_set_timeout(struct sk_buff *skb,
705 				    struct netlink_callback *cb)
706 {
707 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
708 	int timeout, delta, abs_delta;
709 	const struct nlattr *attr;
710 	unsigned long flags;
711 	struct ib_sa_query *query;
712 	long delay = 0;
713 	struct nlattr *tb[LS_NLA_TYPE_MAX];
714 	int ret;
715 
716 	if (!netlink_capable(skb, CAP_NET_ADMIN))
717 		return -EPERM;
718 
719 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
720 			nlmsg_len(nlh), ib_nl_policy);
721 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
722 	if (ret || !attr)
723 		goto settimeout_out;
724 
725 	timeout = *(int *) nla_data(attr);
726 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
727 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
728 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
729 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
730 
731 	delta = timeout - sa_local_svc_timeout_ms;
732 	if (delta < 0)
733 		abs_delta = -delta;
734 	else
735 		abs_delta = delta;
736 
737 	if (delta != 0) {
738 		spin_lock_irqsave(&ib_nl_request_lock, flags);
739 		sa_local_svc_timeout_ms = timeout;
740 		list_for_each_entry(query, &ib_nl_request_list, list) {
741 			if (delta < 0 && abs_delta > query->timeout)
742 				query->timeout = 0;
743 			else
744 				query->timeout += delta;
745 
746 			/* Get the new delay from the first entry */
747 			if (!delay) {
748 				delay = query->timeout - jiffies;
749 				if (delay <= 0)
750 					delay = 1;
751 			}
752 		}
753 		if (delay)
754 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
755 					 (unsigned long)delay);
756 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
757 	}
758 
759 settimeout_out:
760 	return skb->len;
761 }
762 
763 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
764 {
765 	struct nlattr *tb[LS_NLA_TYPE_MAX];
766 	int ret;
767 
768 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
769 		return 0;
770 
771 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
772 			nlmsg_len(nlh), ib_nl_policy);
773 	if (ret)
774 		return 0;
775 
776 	return 1;
777 }
778 
779 static int ib_nl_handle_resolve_resp(struct sk_buff *skb,
780 				     struct netlink_callback *cb)
781 {
782 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
783 	unsigned long flags;
784 	struct ib_sa_query *query;
785 	struct ib_mad_send_buf *send_buf;
786 	struct ib_mad_send_wc mad_send_wc;
787 	int found = 0;
788 	int ret;
789 
790 	if (!netlink_capable(skb, CAP_NET_ADMIN))
791 		return -EPERM;
792 
793 	spin_lock_irqsave(&ib_nl_request_lock, flags);
794 	list_for_each_entry(query, &ib_nl_request_list, list) {
795 		/*
796 		 * If the query is cancelled, let the timeout routine
797 		 * take care of it.
798 		 */
799 		if (nlh->nlmsg_seq == query->seq) {
800 			found = !ib_sa_query_cancelled(query);
801 			if (found)
802 				list_del(&query->list);
803 			break;
804 		}
805 	}
806 
807 	if (!found) {
808 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
809 		goto resp_out;
810 	}
811 
812 	send_buf = query->mad_buf;
813 
814 	if (!ib_nl_is_good_resolve_resp(nlh)) {
815 		/* if the result is a failure, send out the packet via IB */
816 		ib_sa_disable_local_svc(query);
817 		ret = ib_post_send_mad(query->mad_buf, NULL);
818 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
819 		if (ret) {
820 			mad_send_wc.send_buf = send_buf;
821 			mad_send_wc.status = IB_WC_GENERAL_ERR;
822 			send_handler(query->port->agent, &mad_send_wc);
823 		}
824 	} else {
825 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
826 		ib_nl_process_good_resolve_rsp(query, nlh);
827 	}
828 
829 resp_out:
830 	return skb->len;
831 }
832 
833 static struct ibnl_client_cbs ib_sa_cb_table[] = {
834 	[RDMA_NL_LS_OP_RESOLVE] = {
835 		.dump = ib_nl_handle_resolve_resp,
836 		.module = THIS_MODULE },
837 	[RDMA_NL_LS_OP_SET_TIMEOUT] = {
838 		.dump = ib_nl_handle_set_timeout,
839 		.module = THIS_MODULE },
840 };
841 
842 static void free_sm_ah(struct kref *kref)
843 {
844 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
845 
846 	ib_destroy_ah(sm_ah->ah);
847 	kfree(sm_ah);
848 }
849 
850 static void update_sm_ah(struct work_struct *work)
851 {
852 	struct ib_sa_port *port =
853 		container_of(work, struct ib_sa_port, update_task);
854 	struct ib_sa_sm_ah *new_ah;
855 	struct ib_port_attr port_attr;
856 	struct ib_ah_attr   ah_attr;
857 
858 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
859 		printk(KERN_WARNING "Couldn't query port\n");
860 		return;
861 	}
862 
863 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
864 	if (!new_ah) {
865 		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
866 		return;
867 	}
868 
869 	kref_init(&new_ah->ref);
870 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
871 
872 	new_ah->pkey_index = 0;
873 	if (ib_find_pkey(port->agent->device, port->port_num,
874 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
875 		printk(KERN_ERR "Couldn't find index for default PKey\n");
876 
877 	memset(&ah_attr, 0, sizeof ah_attr);
878 	ah_attr.dlid     = port_attr.sm_lid;
879 	ah_attr.sl       = port_attr.sm_sl;
880 	ah_attr.port_num = port->port_num;
881 
882 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
883 	if (IS_ERR(new_ah->ah)) {
884 		printk(KERN_WARNING "Couldn't create new SM AH\n");
885 		kfree(new_ah);
886 		return;
887 	}
888 
889 	spin_lock_irq(&port->ah_lock);
890 	if (port->sm_ah)
891 		kref_put(&port->sm_ah->ref, free_sm_ah);
892 	port->sm_ah = new_ah;
893 	spin_unlock_irq(&port->ah_lock);
894 
895 }
896 
897 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
898 {
899 	if (event->event == IB_EVENT_PORT_ERR    ||
900 	    event->event == IB_EVENT_PORT_ACTIVE ||
901 	    event->event == IB_EVENT_LID_CHANGE  ||
902 	    event->event == IB_EVENT_PKEY_CHANGE ||
903 	    event->event == IB_EVENT_SM_CHANGE   ||
904 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
905 		unsigned long flags;
906 		struct ib_sa_device *sa_dev =
907 			container_of(handler, typeof(*sa_dev), event_handler);
908 		struct ib_sa_port *port =
909 			&sa_dev->port[event->element.port_num - sa_dev->start_port];
910 
911 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
912 			return;
913 
914 		spin_lock_irqsave(&port->ah_lock, flags);
915 		if (port->sm_ah)
916 			kref_put(&port->sm_ah->ref, free_sm_ah);
917 		port->sm_ah = NULL;
918 		spin_unlock_irqrestore(&port->ah_lock, flags);
919 
920 		queue_work(ib_wq, &sa_dev->port[event->element.port_num -
921 					    sa_dev->start_port].update_task);
922 	}
923 }
924 
925 void ib_sa_register_client(struct ib_sa_client *client)
926 {
927 	atomic_set(&client->users, 1);
928 	init_completion(&client->comp);
929 }
930 EXPORT_SYMBOL(ib_sa_register_client);
931 
932 void ib_sa_unregister_client(struct ib_sa_client *client)
933 {
934 	ib_sa_client_put(client);
935 	wait_for_completion(&client->comp);
936 }
937 EXPORT_SYMBOL(ib_sa_unregister_client);
938 
939 /**
940  * ib_sa_cancel_query - try to cancel an SA query
941  * @id:ID of query to cancel
942  * @query:query pointer to cancel
943  *
944  * Try to cancel an SA query.  If the id and query don't match up or
945  * the query has already completed, nothing is done.  Otherwise the
946  * query is canceled and will complete with a status of -EINTR.
947  */
948 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
949 {
950 	unsigned long flags;
951 	struct ib_mad_agent *agent;
952 	struct ib_mad_send_buf *mad_buf;
953 
954 	spin_lock_irqsave(&idr_lock, flags);
955 	if (idr_find(&query_idr, id) != query) {
956 		spin_unlock_irqrestore(&idr_lock, flags);
957 		return;
958 	}
959 	agent = query->port->agent;
960 	mad_buf = query->mad_buf;
961 	spin_unlock_irqrestore(&idr_lock, flags);
962 
963 	/*
964 	 * If the query is still on the netlink request list, schedule
965 	 * it to be cancelled by the timeout routine. Otherwise, it has been
966 	 * sent to the MAD layer and has to be cancelled from there.
967 	 */
968 	if (!ib_nl_cancel_request(query))
969 		ib_cancel_mad(agent, mad_buf);
970 }
971 EXPORT_SYMBOL(ib_sa_cancel_query);
972 
973 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
974 {
975 	struct ib_sa_device *sa_dev;
976 	struct ib_sa_port   *port;
977 	unsigned long flags;
978 	u8 src_path_mask;
979 
980 	sa_dev = ib_get_client_data(device, &sa_client);
981 	if (!sa_dev)
982 		return 0x7f;
983 
984 	port  = &sa_dev->port[port_num - sa_dev->start_port];
985 	spin_lock_irqsave(&port->ah_lock, flags);
986 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
987 	spin_unlock_irqrestore(&port->ah_lock, flags);
988 
989 	return src_path_mask;
990 }
991 
992 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
993 			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
994 {
995 	int ret;
996 	u16 gid_index;
997 	int force_grh;
998 
999 	memset(ah_attr, 0, sizeof *ah_attr);
1000 	ah_attr->dlid = be16_to_cpu(rec->dlid);
1001 	ah_attr->sl = rec->sl;
1002 	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
1003 				 get_src_path_mask(device, port_num);
1004 	ah_attr->port_num = port_num;
1005 	ah_attr->static_rate = rec->rate;
1006 
1007 	force_grh = rdma_cap_eth_ah(device, port_num);
1008 
1009 	if (rec->hop_limit > 1 || force_grh) {
1010 		ah_attr->ah_flags = IB_AH_GRH;
1011 		ah_attr->grh.dgid = rec->dgid;
1012 
1013 		ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
1014 					 &gid_index);
1015 		if (ret)
1016 			return ret;
1017 
1018 		ah_attr->grh.sgid_index    = gid_index;
1019 		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
1020 		ah_attr->grh.hop_limit     = rec->hop_limit;
1021 		ah_attr->grh.traffic_class = rec->traffic_class;
1022 	}
1023 	if (force_grh) {
1024 		memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
1025 		ah_attr->vlan_id = rec->vlan_id;
1026 	} else {
1027 		ah_attr->vlan_id = 0xffff;
1028 	}
1029 
1030 	return 0;
1031 }
1032 EXPORT_SYMBOL(ib_init_ah_from_path);
1033 
1034 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1035 {
1036 	unsigned long flags;
1037 
1038 	spin_lock_irqsave(&query->port->ah_lock, flags);
1039 	if (!query->port->sm_ah) {
1040 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1041 		return -EAGAIN;
1042 	}
1043 	kref_get(&query->port->sm_ah->ref);
1044 	query->sm_ah = query->port->sm_ah;
1045 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1046 
1047 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1048 					    query->sm_ah->pkey_index,
1049 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1050 					    gfp_mask,
1051 					    IB_MGMT_BASE_VERSION);
1052 	if (IS_ERR(query->mad_buf)) {
1053 		kref_put(&query->sm_ah->ref, free_sm_ah);
1054 		return -ENOMEM;
1055 	}
1056 
1057 	query->mad_buf->ah = query->sm_ah->ah;
1058 
1059 	return 0;
1060 }
1061 
1062 static void free_mad(struct ib_sa_query *query)
1063 {
1064 	ib_free_send_mad(query->mad_buf);
1065 	kref_put(&query->sm_ah->ref, free_sm_ah);
1066 }
1067 
1068 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
1069 {
1070 	unsigned long flags;
1071 
1072 	memset(mad, 0, sizeof *mad);
1073 
1074 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1075 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1076 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1077 
1078 	spin_lock_irqsave(&tid_lock, flags);
1079 	mad->mad_hdr.tid           =
1080 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1081 	spin_unlock_irqrestore(&tid_lock, flags);
1082 }
1083 
1084 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1085 {
1086 	bool preload = !!(gfp_mask & __GFP_WAIT);
1087 	unsigned long flags;
1088 	int ret, id;
1089 
1090 	if (preload)
1091 		idr_preload(gfp_mask);
1092 	spin_lock_irqsave(&idr_lock, flags);
1093 
1094 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1095 
1096 	spin_unlock_irqrestore(&idr_lock, flags);
1097 	if (preload)
1098 		idr_preload_end();
1099 	if (id < 0)
1100 		return id;
1101 
1102 	query->mad_buf->timeout_ms  = timeout_ms;
1103 	query->mad_buf->context[0] = query;
1104 	query->id = id;
1105 
1106 	if (query->flags & IB_SA_ENABLE_LOCAL_SERVICE) {
1107 		if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1108 			if (!ib_nl_make_request(query))
1109 				return id;
1110 		}
1111 		ib_sa_disable_local_svc(query);
1112 	}
1113 
1114 	ret = ib_post_send_mad(query->mad_buf, NULL);
1115 	if (ret) {
1116 		spin_lock_irqsave(&idr_lock, flags);
1117 		idr_remove(&query_idr, id);
1118 		spin_unlock_irqrestore(&idr_lock, flags);
1119 	}
1120 
1121 	/*
1122 	 * It's not safe to dereference query any more, because the
1123 	 * send may already have completed and freed the query in
1124 	 * another context.
1125 	 */
1126 	return ret ? ret : id;
1127 }
1128 
1129 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
1130 {
1131 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1132 }
1133 EXPORT_SYMBOL(ib_sa_unpack_path);
1134 
1135 void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
1136 {
1137 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1138 }
1139 EXPORT_SYMBOL(ib_sa_pack_path);
1140 
1141 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1142 				    int status,
1143 				    struct ib_sa_mad *mad)
1144 {
1145 	struct ib_sa_path_query *query =
1146 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1147 
1148 	if (mad) {
1149 		struct ib_sa_path_rec rec;
1150 
1151 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1152 			  mad->data, &rec);
1153 		rec.vlan_id = 0xffff;
1154 		memset(rec.dmac, 0, ETH_ALEN);
1155 		memset(rec.smac, 0, ETH_ALEN);
1156 		query->callback(status, &rec, query->context);
1157 	} else
1158 		query->callback(status, NULL, query->context);
1159 }
1160 
1161 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1162 {
1163 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
1164 }
1165 
1166 /**
1167  * ib_sa_path_rec_get - Start a Path get query
1168  * @client:SA client
1169  * @device:device to send query on
1170  * @port_num: port number to send query on
1171  * @rec:Path Record to send in query
1172  * @comp_mask:component mask to send in query
1173  * @timeout_ms:time to wait for response
1174  * @gfp_mask:GFP mask to use for internal allocations
1175  * @callback:function called when query completes, times out or is
1176  * canceled
1177  * @context:opaque user context passed to callback
1178  * @sa_query:query context, used to cancel query
1179  *
1180  * Send a Path Record Get query to the SA to look up a path.  The
1181  * callback function will be called when the query completes (or
1182  * fails); status is 0 for a successful response, -EINTR if the query
1183  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1184  * occurred sending the query.  The resp parameter of the callback is
1185  * only valid if status is 0.
1186  *
1187  * If the return value of ib_sa_path_rec_get() is negative, it is an
1188  * error code.  Otherwise it is a query ID that can be used to cancel
1189  * the query.
1190  */
1191 int ib_sa_path_rec_get(struct ib_sa_client *client,
1192 		       struct ib_device *device, u8 port_num,
1193 		       struct ib_sa_path_rec *rec,
1194 		       ib_sa_comp_mask comp_mask,
1195 		       int timeout_ms, gfp_t gfp_mask,
1196 		       void (*callback)(int status,
1197 					struct ib_sa_path_rec *resp,
1198 					void *context),
1199 		       void *context,
1200 		       struct ib_sa_query **sa_query)
1201 {
1202 	struct ib_sa_path_query *query;
1203 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1204 	struct ib_sa_port   *port;
1205 	struct ib_mad_agent *agent;
1206 	struct ib_sa_mad *mad;
1207 	int ret;
1208 
1209 	if (!sa_dev)
1210 		return -ENODEV;
1211 
1212 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1213 	agent = port->agent;
1214 
1215 	query = kzalloc(sizeof(*query), gfp_mask);
1216 	if (!query)
1217 		return -ENOMEM;
1218 
1219 	query->sa_query.port     = port;
1220 	ret = alloc_mad(&query->sa_query, gfp_mask);
1221 	if (ret)
1222 		goto err1;
1223 
1224 	ib_sa_client_get(client);
1225 	query->sa_query.client = client;
1226 	query->callback        = callback;
1227 	query->context         = context;
1228 
1229 	mad = query->sa_query.mad_buf->mad;
1230 	init_mad(mad, agent);
1231 
1232 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1233 	query->sa_query.release  = ib_sa_path_rec_release;
1234 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1235 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1236 	mad->sa_hdr.comp_mask	 = comp_mask;
1237 
1238 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
1239 
1240 	*sa_query = &query->sa_query;
1241 
1242 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1243 	query->sa_query.mad_buf->context[1] = rec;
1244 
1245 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1246 	if (ret < 0)
1247 		goto err2;
1248 
1249 	return ret;
1250 
1251 err2:
1252 	*sa_query = NULL;
1253 	ib_sa_client_put(query->sa_query.client);
1254 	free_mad(&query->sa_query);
1255 
1256 err1:
1257 	kfree(query);
1258 	return ret;
1259 }
1260 EXPORT_SYMBOL(ib_sa_path_rec_get);
1261 
1262 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1263 				    int status,
1264 				    struct ib_sa_mad *mad)
1265 {
1266 	struct ib_sa_service_query *query =
1267 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1268 
1269 	if (mad) {
1270 		struct ib_sa_service_rec rec;
1271 
1272 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1273 			  mad->data, &rec);
1274 		query->callback(status, &rec, query->context);
1275 	} else
1276 		query->callback(status, NULL, query->context);
1277 }
1278 
1279 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1280 {
1281 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1282 }
1283 
1284 /**
1285  * ib_sa_service_rec_query - Start Service Record operation
1286  * @client:SA client
1287  * @device:device to send request on
1288  * @port_num: port number to send request on
1289  * @method:SA method - should be get, set, or delete
1290  * @rec:Service Record to send in request
1291  * @comp_mask:component mask to send in request
1292  * @timeout_ms:time to wait for response
1293  * @gfp_mask:GFP mask to use for internal allocations
1294  * @callback:function called when request completes, times out or is
1295  * canceled
1296  * @context:opaque user context passed to callback
1297  * @sa_query:request context, used to cancel request
1298  *
1299  * Send a Service Record set/get/delete to the SA to register,
1300  * unregister or query a service record.
1301  * The callback function will be called when the request completes (or
1302  * fails); status is 0 for a successful response, -EINTR if the query
1303  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1304  * occurred sending the query.  The resp parameter of the callback is
1305  * only valid if status is 0.
1306  *
1307  * If the return value of ib_sa_service_rec_query() is negative, it is an
1308  * error code.  Otherwise it is a request ID that can be used to cancel
1309  * the query.
1310  */
1311 int ib_sa_service_rec_query(struct ib_sa_client *client,
1312 			    struct ib_device *device, u8 port_num, u8 method,
1313 			    struct ib_sa_service_rec *rec,
1314 			    ib_sa_comp_mask comp_mask,
1315 			    int timeout_ms, gfp_t gfp_mask,
1316 			    void (*callback)(int status,
1317 					     struct ib_sa_service_rec *resp,
1318 					     void *context),
1319 			    void *context,
1320 			    struct ib_sa_query **sa_query)
1321 {
1322 	struct ib_sa_service_query *query;
1323 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1324 	struct ib_sa_port   *port;
1325 	struct ib_mad_agent *agent;
1326 	struct ib_sa_mad *mad;
1327 	int ret;
1328 
1329 	if (!sa_dev)
1330 		return -ENODEV;
1331 
1332 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1333 	agent = port->agent;
1334 
1335 	if (method != IB_MGMT_METHOD_GET &&
1336 	    method != IB_MGMT_METHOD_SET &&
1337 	    method != IB_SA_METHOD_DELETE)
1338 		return -EINVAL;
1339 
1340 	query = kzalloc(sizeof(*query), gfp_mask);
1341 	if (!query)
1342 		return -ENOMEM;
1343 
1344 	query->sa_query.port     = port;
1345 	ret = alloc_mad(&query->sa_query, gfp_mask);
1346 	if (ret)
1347 		goto err1;
1348 
1349 	ib_sa_client_get(client);
1350 	query->sa_query.client = client;
1351 	query->callback        = callback;
1352 	query->context         = context;
1353 
1354 	mad = query->sa_query.mad_buf->mad;
1355 	init_mad(mad, agent);
1356 
1357 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1358 	query->sa_query.release  = ib_sa_service_rec_release;
1359 	mad->mad_hdr.method	 = method;
1360 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1361 	mad->sa_hdr.comp_mask	 = comp_mask;
1362 
1363 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1364 		rec, mad->data);
1365 
1366 	*sa_query = &query->sa_query;
1367 
1368 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1369 	if (ret < 0)
1370 		goto err2;
1371 
1372 	return ret;
1373 
1374 err2:
1375 	*sa_query = NULL;
1376 	ib_sa_client_put(query->sa_query.client);
1377 	free_mad(&query->sa_query);
1378 
1379 err1:
1380 	kfree(query);
1381 	return ret;
1382 }
1383 EXPORT_SYMBOL(ib_sa_service_rec_query);
1384 
1385 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1386 					int status,
1387 					struct ib_sa_mad *mad)
1388 {
1389 	struct ib_sa_mcmember_query *query =
1390 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1391 
1392 	if (mad) {
1393 		struct ib_sa_mcmember_rec rec;
1394 
1395 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1396 			  mad->data, &rec);
1397 		query->callback(status, &rec, query->context);
1398 	} else
1399 		query->callback(status, NULL, query->context);
1400 }
1401 
1402 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1403 {
1404 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1405 }
1406 
1407 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1408 			     struct ib_device *device, u8 port_num,
1409 			     u8 method,
1410 			     struct ib_sa_mcmember_rec *rec,
1411 			     ib_sa_comp_mask comp_mask,
1412 			     int timeout_ms, gfp_t gfp_mask,
1413 			     void (*callback)(int status,
1414 					      struct ib_sa_mcmember_rec *resp,
1415 					      void *context),
1416 			     void *context,
1417 			     struct ib_sa_query **sa_query)
1418 {
1419 	struct ib_sa_mcmember_query *query;
1420 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1421 	struct ib_sa_port   *port;
1422 	struct ib_mad_agent *agent;
1423 	struct ib_sa_mad *mad;
1424 	int ret;
1425 
1426 	if (!sa_dev)
1427 		return -ENODEV;
1428 
1429 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1430 	agent = port->agent;
1431 
1432 	query = kzalloc(sizeof(*query), gfp_mask);
1433 	if (!query)
1434 		return -ENOMEM;
1435 
1436 	query->sa_query.port     = port;
1437 	ret = alloc_mad(&query->sa_query, gfp_mask);
1438 	if (ret)
1439 		goto err1;
1440 
1441 	ib_sa_client_get(client);
1442 	query->sa_query.client = client;
1443 	query->callback        = callback;
1444 	query->context         = context;
1445 
1446 	mad = query->sa_query.mad_buf->mad;
1447 	init_mad(mad, agent);
1448 
1449 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1450 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1451 	mad->mad_hdr.method	 = method;
1452 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1453 	mad->sa_hdr.comp_mask	 = comp_mask;
1454 
1455 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1456 		rec, mad->data);
1457 
1458 	*sa_query = &query->sa_query;
1459 
1460 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1461 	if (ret < 0)
1462 		goto err2;
1463 
1464 	return ret;
1465 
1466 err2:
1467 	*sa_query = NULL;
1468 	ib_sa_client_put(query->sa_query.client);
1469 	free_mad(&query->sa_query);
1470 
1471 err1:
1472 	kfree(query);
1473 	return ret;
1474 }
1475 
1476 /* Support GuidInfoRecord */
1477 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1478 					int status,
1479 					struct ib_sa_mad *mad)
1480 {
1481 	struct ib_sa_guidinfo_query *query =
1482 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1483 
1484 	if (mad) {
1485 		struct ib_sa_guidinfo_rec rec;
1486 
1487 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1488 			  mad->data, &rec);
1489 		query->callback(status, &rec, query->context);
1490 	} else
1491 		query->callback(status, NULL, query->context);
1492 }
1493 
1494 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1495 {
1496 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1497 }
1498 
1499 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1500 			      struct ib_device *device, u8 port_num,
1501 			      struct ib_sa_guidinfo_rec *rec,
1502 			      ib_sa_comp_mask comp_mask, u8 method,
1503 			      int timeout_ms, gfp_t gfp_mask,
1504 			      void (*callback)(int status,
1505 					       struct ib_sa_guidinfo_rec *resp,
1506 					       void *context),
1507 			      void *context,
1508 			      struct ib_sa_query **sa_query)
1509 {
1510 	struct ib_sa_guidinfo_query *query;
1511 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1512 	struct ib_sa_port *port;
1513 	struct ib_mad_agent *agent;
1514 	struct ib_sa_mad *mad;
1515 	int ret;
1516 
1517 	if (!sa_dev)
1518 		return -ENODEV;
1519 
1520 	if (method != IB_MGMT_METHOD_GET &&
1521 	    method != IB_MGMT_METHOD_SET &&
1522 	    method != IB_SA_METHOD_DELETE) {
1523 		return -EINVAL;
1524 	}
1525 
1526 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1527 	agent = port->agent;
1528 
1529 	query = kzalloc(sizeof(*query), gfp_mask);
1530 	if (!query)
1531 		return -ENOMEM;
1532 
1533 	query->sa_query.port = port;
1534 	ret = alloc_mad(&query->sa_query, gfp_mask);
1535 	if (ret)
1536 		goto err1;
1537 
1538 	ib_sa_client_get(client);
1539 	query->sa_query.client = client;
1540 	query->callback        = callback;
1541 	query->context         = context;
1542 
1543 	mad = query->sa_query.mad_buf->mad;
1544 	init_mad(mad, agent);
1545 
1546 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1547 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1548 
1549 	mad->mad_hdr.method	 = method;
1550 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1551 	mad->sa_hdr.comp_mask	 = comp_mask;
1552 
1553 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1554 		mad->data);
1555 
1556 	*sa_query = &query->sa_query;
1557 
1558 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1559 	if (ret < 0)
1560 		goto err2;
1561 
1562 	return ret;
1563 
1564 err2:
1565 	*sa_query = NULL;
1566 	ib_sa_client_put(query->sa_query.client);
1567 	free_mad(&query->sa_query);
1568 
1569 err1:
1570 	kfree(query);
1571 	return ret;
1572 }
1573 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1574 
1575 static void send_handler(struct ib_mad_agent *agent,
1576 			 struct ib_mad_send_wc *mad_send_wc)
1577 {
1578 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1579 	unsigned long flags;
1580 
1581 	if (query->callback)
1582 		switch (mad_send_wc->status) {
1583 		case IB_WC_SUCCESS:
1584 			/* No callback -- already got recv */
1585 			break;
1586 		case IB_WC_RESP_TIMEOUT_ERR:
1587 			query->callback(query, -ETIMEDOUT, NULL);
1588 			break;
1589 		case IB_WC_WR_FLUSH_ERR:
1590 			query->callback(query, -EINTR, NULL);
1591 			break;
1592 		default:
1593 			query->callback(query, -EIO, NULL);
1594 			break;
1595 		}
1596 
1597 	spin_lock_irqsave(&idr_lock, flags);
1598 	idr_remove(&query_idr, query->id);
1599 	spin_unlock_irqrestore(&idr_lock, flags);
1600 
1601 	free_mad(query);
1602 	ib_sa_client_put(query->client);
1603 	query->release(query);
1604 }
1605 
1606 static void recv_handler(struct ib_mad_agent *mad_agent,
1607 			 struct ib_mad_recv_wc *mad_recv_wc)
1608 {
1609 	struct ib_sa_query *query;
1610 	struct ib_mad_send_buf *mad_buf;
1611 
1612 	mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
1613 	query = mad_buf->context[0];
1614 
1615 	if (query->callback) {
1616 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1617 			query->callback(query,
1618 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1619 					-EINVAL : 0,
1620 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
1621 		else
1622 			query->callback(query, -EIO, NULL);
1623 	}
1624 
1625 	ib_free_recv_mad(mad_recv_wc);
1626 }
1627 
1628 static void ib_sa_add_one(struct ib_device *device)
1629 {
1630 	struct ib_sa_device *sa_dev;
1631 	int s, e, i;
1632 	int count = 0;
1633 
1634 	s = rdma_start_port(device);
1635 	e = rdma_end_port(device);
1636 
1637 	sa_dev = kzalloc(sizeof *sa_dev +
1638 			 (e - s + 1) * sizeof (struct ib_sa_port),
1639 			 GFP_KERNEL);
1640 	if (!sa_dev)
1641 		return;
1642 
1643 	sa_dev->start_port = s;
1644 	sa_dev->end_port   = e;
1645 
1646 	for (i = 0; i <= e - s; ++i) {
1647 		spin_lock_init(&sa_dev->port[i].ah_lock);
1648 		if (!rdma_cap_ib_sa(device, i + 1))
1649 			continue;
1650 
1651 		sa_dev->port[i].sm_ah    = NULL;
1652 		sa_dev->port[i].port_num = i + s;
1653 
1654 		sa_dev->port[i].agent =
1655 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1656 					      NULL, 0, send_handler,
1657 					      recv_handler, sa_dev, 0);
1658 		if (IS_ERR(sa_dev->port[i].agent))
1659 			goto err;
1660 
1661 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1662 
1663 		count++;
1664 	}
1665 
1666 	if (!count)
1667 		goto free;
1668 
1669 	ib_set_client_data(device, &sa_client, sa_dev);
1670 
1671 	/*
1672 	 * We register our event handler after everything is set up,
1673 	 * and then update our cached info after the event handler is
1674 	 * registered to avoid any problems if a port changes state
1675 	 * during our initialization.
1676 	 */
1677 
1678 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1679 	if (ib_register_event_handler(&sa_dev->event_handler))
1680 		goto err;
1681 
1682 	for (i = 0; i <= e - s; ++i) {
1683 		if (rdma_cap_ib_sa(device, i + 1))
1684 			update_sm_ah(&sa_dev->port[i].update_task);
1685 	}
1686 
1687 	return;
1688 
1689 err:
1690 	while (--i >= 0) {
1691 		if (rdma_cap_ib_sa(device, i + 1))
1692 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1693 	}
1694 free:
1695 	kfree(sa_dev);
1696 	return;
1697 }
1698 
1699 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
1700 {
1701 	struct ib_sa_device *sa_dev = client_data;
1702 	int i;
1703 
1704 	if (!sa_dev)
1705 		return;
1706 
1707 	ib_unregister_event_handler(&sa_dev->event_handler);
1708 
1709 	flush_workqueue(ib_wq);
1710 
1711 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1712 		if (rdma_cap_ib_sa(device, i + 1)) {
1713 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1714 			if (sa_dev->port[i].sm_ah)
1715 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1716 		}
1717 
1718 	}
1719 
1720 	kfree(sa_dev);
1721 }
1722 
1723 static int __init ib_sa_init(void)
1724 {
1725 	int ret;
1726 
1727 	get_random_bytes(&tid, sizeof tid);
1728 
1729 	atomic_set(&ib_nl_sa_request_seq, 0);
1730 
1731 	ret = ib_register_client(&sa_client);
1732 	if (ret) {
1733 		printk(KERN_ERR "Couldn't register ib_sa client\n");
1734 		goto err1;
1735 	}
1736 
1737 	ret = mcast_init();
1738 	if (ret) {
1739 		printk(KERN_ERR "Couldn't initialize multicast handling\n");
1740 		goto err2;
1741 	}
1742 
1743 	ib_nl_wq = create_singlethread_workqueue("ib_nl_sa_wq");
1744 	if (!ib_nl_wq) {
1745 		ret = -ENOMEM;
1746 		goto err3;
1747 	}
1748 
1749 	if (ibnl_add_client(RDMA_NL_LS, RDMA_NL_LS_NUM_OPS,
1750 			    ib_sa_cb_table)) {
1751 		pr_err("Failed to add netlink callback\n");
1752 		ret = -EINVAL;
1753 		goto err4;
1754 	}
1755 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
1756 
1757 	return 0;
1758 err4:
1759 	destroy_workqueue(ib_nl_wq);
1760 err3:
1761 	mcast_cleanup();
1762 err2:
1763 	ib_unregister_client(&sa_client);
1764 err1:
1765 	return ret;
1766 }
1767 
1768 static void __exit ib_sa_cleanup(void)
1769 {
1770 	ibnl_remove_client(RDMA_NL_LS);
1771 	cancel_delayed_work(&ib_nl_timed_work);
1772 	flush_workqueue(ib_nl_wq);
1773 	destroy_workqueue(ib_nl_wq);
1774 	mcast_cleanup();
1775 	ib_unregister_client(&sa_client);
1776 	idr_destroy(&query_idr);
1777 }
1778 
1779 module_init(ib_sa_init);
1780 module_exit(ib_sa_cleanup);
1781