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  * $Id: sa_query.c 2811 2005-07-06 18:11:43Z halr $
35  */
36 
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/err.h>
40 #include <linux/random.h>
41 #include <linux/spinlock.h>
42 #include <linux/slab.h>
43 #include <linux/dma-mapping.h>
44 #include <linux/kref.h>
45 #include <linux/idr.h>
46 #include <linux/workqueue.h>
47 
48 #include <rdma/ib_pack.h>
49 #include <rdma/ib_cache.h>
50 #include "sa.h"
51 
52 MODULE_AUTHOR("Roland Dreier");
53 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
54 MODULE_LICENSE("Dual BSD/GPL");
55 
56 struct ib_sa_sm_ah {
57 	struct ib_ah        *ah;
58 	struct kref          ref;
59 	u8		     src_path_mask;
60 };
61 
62 struct ib_sa_port {
63 	struct ib_mad_agent *agent;
64 	struct ib_sa_sm_ah  *sm_ah;
65 	struct work_struct   update_task;
66 	spinlock_t           ah_lock;
67 	u8                   port_num;
68 };
69 
70 struct ib_sa_device {
71 	int                     start_port, end_port;
72 	struct ib_event_handler event_handler;
73 	struct ib_sa_port port[0];
74 };
75 
76 struct ib_sa_query {
77 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
78 	void (*release)(struct ib_sa_query *);
79 	struct ib_sa_client    *client;
80 	struct ib_sa_port      *port;
81 	struct ib_mad_send_buf *mad_buf;
82 	struct ib_sa_sm_ah     *sm_ah;
83 	int			id;
84 };
85 
86 struct ib_sa_service_query {
87 	void (*callback)(int, struct ib_sa_service_rec *, void *);
88 	void *context;
89 	struct ib_sa_query sa_query;
90 };
91 
92 struct ib_sa_path_query {
93 	void (*callback)(int, struct ib_sa_path_rec *, void *);
94 	void *context;
95 	struct ib_sa_query sa_query;
96 };
97 
98 struct ib_sa_mcmember_query {
99 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
100 	void *context;
101 	struct ib_sa_query sa_query;
102 };
103 
104 static void ib_sa_add_one(struct ib_device *device);
105 static void ib_sa_remove_one(struct ib_device *device);
106 
107 static struct ib_client sa_client = {
108 	.name   = "sa",
109 	.add    = ib_sa_add_one,
110 	.remove = ib_sa_remove_one
111 };
112 
113 static spinlock_t idr_lock;
114 static DEFINE_IDR(query_idr);
115 
116 static spinlock_t tid_lock;
117 static u32 tid;
118 
119 #define PATH_REC_FIELD(field) \
120 	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
121 	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
122 	.field_name          = "sa_path_rec:" #field
123 
124 static const struct ib_field path_rec_table[] = {
125 	{ RESERVED,
126 	  .offset_words = 0,
127 	  .offset_bits  = 0,
128 	  .size_bits    = 32 },
129 	{ RESERVED,
130 	  .offset_words = 1,
131 	  .offset_bits  = 0,
132 	  .size_bits    = 32 },
133 	{ PATH_REC_FIELD(dgid),
134 	  .offset_words = 2,
135 	  .offset_bits  = 0,
136 	  .size_bits    = 128 },
137 	{ PATH_REC_FIELD(sgid),
138 	  .offset_words = 6,
139 	  .offset_bits  = 0,
140 	  .size_bits    = 128 },
141 	{ PATH_REC_FIELD(dlid),
142 	  .offset_words = 10,
143 	  .offset_bits  = 0,
144 	  .size_bits    = 16 },
145 	{ PATH_REC_FIELD(slid),
146 	  .offset_words = 10,
147 	  .offset_bits  = 16,
148 	  .size_bits    = 16 },
149 	{ PATH_REC_FIELD(raw_traffic),
150 	  .offset_words = 11,
151 	  .offset_bits  = 0,
152 	  .size_bits    = 1 },
153 	{ RESERVED,
154 	  .offset_words = 11,
155 	  .offset_bits  = 1,
156 	  .size_bits    = 3 },
157 	{ PATH_REC_FIELD(flow_label),
158 	  .offset_words = 11,
159 	  .offset_bits  = 4,
160 	  .size_bits    = 20 },
161 	{ PATH_REC_FIELD(hop_limit),
162 	  .offset_words = 11,
163 	  .offset_bits  = 24,
164 	  .size_bits    = 8 },
165 	{ PATH_REC_FIELD(traffic_class),
166 	  .offset_words = 12,
167 	  .offset_bits  = 0,
168 	  .size_bits    = 8 },
169 	{ PATH_REC_FIELD(reversible),
170 	  .offset_words = 12,
171 	  .offset_bits  = 8,
172 	  .size_bits    = 1 },
173 	{ PATH_REC_FIELD(numb_path),
174 	  .offset_words = 12,
175 	  .offset_bits  = 9,
176 	  .size_bits    = 7 },
177 	{ PATH_REC_FIELD(pkey),
178 	  .offset_words = 12,
179 	  .offset_bits  = 16,
180 	  .size_bits    = 16 },
181 	{ RESERVED,
182 	  .offset_words = 13,
183 	  .offset_bits  = 0,
184 	  .size_bits    = 12 },
185 	{ PATH_REC_FIELD(sl),
186 	  .offset_words = 13,
187 	  .offset_bits  = 12,
188 	  .size_bits    = 4 },
189 	{ PATH_REC_FIELD(mtu_selector),
190 	  .offset_words = 13,
191 	  .offset_bits  = 16,
192 	  .size_bits    = 2 },
193 	{ PATH_REC_FIELD(mtu),
194 	  .offset_words = 13,
195 	  .offset_bits  = 18,
196 	  .size_bits    = 6 },
197 	{ PATH_REC_FIELD(rate_selector),
198 	  .offset_words = 13,
199 	  .offset_bits  = 24,
200 	  .size_bits    = 2 },
201 	{ PATH_REC_FIELD(rate),
202 	  .offset_words = 13,
203 	  .offset_bits  = 26,
204 	  .size_bits    = 6 },
205 	{ PATH_REC_FIELD(packet_life_time_selector),
206 	  .offset_words = 14,
207 	  .offset_bits  = 0,
208 	  .size_bits    = 2 },
209 	{ PATH_REC_FIELD(packet_life_time),
210 	  .offset_words = 14,
211 	  .offset_bits  = 2,
212 	  .size_bits    = 6 },
213 	{ PATH_REC_FIELD(preference),
214 	  .offset_words = 14,
215 	  .offset_bits  = 8,
216 	  .size_bits    = 8 },
217 	{ RESERVED,
218 	  .offset_words = 14,
219 	  .offset_bits  = 16,
220 	  .size_bits    = 48 },
221 };
222 
223 #define MCMEMBER_REC_FIELD(field) \
224 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
225 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
226 	.field_name          = "sa_mcmember_rec:" #field
227 
228 static const struct ib_field mcmember_rec_table[] = {
229 	{ MCMEMBER_REC_FIELD(mgid),
230 	  .offset_words = 0,
231 	  .offset_bits  = 0,
232 	  .size_bits    = 128 },
233 	{ MCMEMBER_REC_FIELD(port_gid),
234 	  .offset_words = 4,
235 	  .offset_bits  = 0,
236 	  .size_bits    = 128 },
237 	{ MCMEMBER_REC_FIELD(qkey),
238 	  .offset_words = 8,
239 	  .offset_bits  = 0,
240 	  .size_bits    = 32 },
241 	{ MCMEMBER_REC_FIELD(mlid),
242 	  .offset_words = 9,
243 	  .offset_bits  = 0,
244 	  .size_bits    = 16 },
245 	{ MCMEMBER_REC_FIELD(mtu_selector),
246 	  .offset_words = 9,
247 	  .offset_bits  = 16,
248 	  .size_bits    = 2 },
249 	{ MCMEMBER_REC_FIELD(mtu),
250 	  .offset_words = 9,
251 	  .offset_bits  = 18,
252 	  .size_bits    = 6 },
253 	{ MCMEMBER_REC_FIELD(traffic_class),
254 	  .offset_words = 9,
255 	  .offset_bits  = 24,
256 	  .size_bits    = 8 },
257 	{ MCMEMBER_REC_FIELD(pkey),
258 	  .offset_words = 10,
259 	  .offset_bits  = 0,
260 	  .size_bits    = 16 },
261 	{ MCMEMBER_REC_FIELD(rate_selector),
262 	  .offset_words = 10,
263 	  .offset_bits  = 16,
264 	  .size_bits    = 2 },
265 	{ MCMEMBER_REC_FIELD(rate),
266 	  .offset_words = 10,
267 	  .offset_bits  = 18,
268 	  .size_bits    = 6 },
269 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
270 	  .offset_words = 10,
271 	  .offset_bits  = 24,
272 	  .size_bits    = 2 },
273 	{ MCMEMBER_REC_FIELD(packet_life_time),
274 	  .offset_words = 10,
275 	  .offset_bits  = 26,
276 	  .size_bits    = 6 },
277 	{ MCMEMBER_REC_FIELD(sl),
278 	  .offset_words = 11,
279 	  .offset_bits  = 0,
280 	  .size_bits    = 4 },
281 	{ MCMEMBER_REC_FIELD(flow_label),
282 	  .offset_words = 11,
283 	  .offset_bits  = 4,
284 	  .size_bits    = 20 },
285 	{ MCMEMBER_REC_FIELD(hop_limit),
286 	  .offset_words = 11,
287 	  .offset_bits  = 24,
288 	  .size_bits    = 8 },
289 	{ MCMEMBER_REC_FIELD(scope),
290 	  .offset_words = 12,
291 	  .offset_bits  = 0,
292 	  .size_bits    = 4 },
293 	{ MCMEMBER_REC_FIELD(join_state),
294 	  .offset_words = 12,
295 	  .offset_bits  = 4,
296 	  .size_bits    = 4 },
297 	{ MCMEMBER_REC_FIELD(proxy_join),
298 	  .offset_words = 12,
299 	  .offset_bits  = 8,
300 	  .size_bits    = 1 },
301 	{ RESERVED,
302 	  .offset_words = 12,
303 	  .offset_bits  = 9,
304 	  .size_bits    = 23 },
305 };
306 
307 #define SERVICE_REC_FIELD(field) \
308 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
309 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
310 	.field_name          = "sa_service_rec:" #field
311 
312 static const struct ib_field service_rec_table[] = {
313 	{ SERVICE_REC_FIELD(id),
314 	  .offset_words = 0,
315 	  .offset_bits  = 0,
316 	  .size_bits    = 64 },
317 	{ SERVICE_REC_FIELD(gid),
318 	  .offset_words = 2,
319 	  .offset_bits  = 0,
320 	  .size_bits    = 128 },
321 	{ SERVICE_REC_FIELD(pkey),
322 	  .offset_words = 6,
323 	  .offset_bits  = 0,
324 	  .size_bits    = 16 },
325 	{ SERVICE_REC_FIELD(lease),
326 	  .offset_words = 7,
327 	  .offset_bits  = 0,
328 	  .size_bits    = 32 },
329 	{ SERVICE_REC_FIELD(key),
330 	  .offset_words = 8,
331 	  .offset_bits  = 0,
332 	  .size_bits    = 128 },
333 	{ SERVICE_REC_FIELD(name),
334 	  .offset_words = 12,
335 	  .offset_bits  = 0,
336 	  .size_bits    = 64*8 },
337 	{ SERVICE_REC_FIELD(data8),
338 	  .offset_words = 28,
339 	  .offset_bits  = 0,
340 	  .size_bits    = 16*8 },
341 	{ SERVICE_REC_FIELD(data16),
342 	  .offset_words = 32,
343 	  .offset_bits  = 0,
344 	  .size_bits    = 8*16 },
345 	{ SERVICE_REC_FIELD(data32),
346 	  .offset_words = 36,
347 	  .offset_bits  = 0,
348 	  .size_bits    = 4*32 },
349 	{ SERVICE_REC_FIELD(data64),
350 	  .offset_words = 40,
351 	  .offset_bits  = 0,
352 	  .size_bits    = 2*64 },
353 };
354 
355 static void free_sm_ah(struct kref *kref)
356 {
357 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
358 
359 	ib_destroy_ah(sm_ah->ah);
360 	kfree(sm_ah);
361 }
362 
363 static void update_sm_ah(struct work_struct *work)
364 {
365 	struct ib_sa_port *port =
366 		container_of(work, struct ib_sa_port, update_task);
367 	struct ib_sa_sm_ah *new_ah, *old_ah;
368 	struct ib_port_attr port_attr;
369 	struct ib_ah_attr   ah_attr;
370 
371 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
372 		printk(KERN_WARNING "Couldn't query port\n");
373 		return;
374 	}
375 
376 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
377 	if (!new_ah) {
378 		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
379 		return;
380 	}
381 
382 	kref_init(&new_ah->ref);
383 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
384 
385 	memset(&ah_attr, 0, sizeof ah_attr);
386 	ah_attr.dlid     = port_attr.sm_lid;
387 	ah_attr.sl       = port_attr.sm_sl;
388 	ah_attr.port_num = port->port_num;
389 
390 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
391 	if (IS_ERR(new_ah->ah)) {
392 		printk(KERN_WARNING "Couldn't create new SM AH\n");
393 		kfree(new_ah);
394 		return;
395 	}
396 
397 	spin_lock_irq(&port->ah_lock);
398 	old_ah = port->sm_ah;
399 	port->sm_ah = new_ah;
400 	spin_unlock_irq(&port->ah_lock);
401 
402 	if (old_ah)
403 		kref_put(&old_ah->ref, free_sm_ah);
404 }
405 
406 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
407 {
408 	if (event->event == IB_EVENT_PORT_ERR    ||
409 	    event->event == IB_EVENT_PORT_ACTIVE ||
410 	    event->event == IB_EVENT_LID_CHANGE  ||
411 	    event->event == IB_EVENT_PKEY_CHANGE ||
412 	    event->event == IB_EVENT_SM_CHANGE   ||
413 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
414 		struct ib_sa_device *sa_dev;
415 		sa_dev = container_of(handler, typeof(*sa_dev), event_handler);
416 
417 		schedule_work(&sa_dev->port[event->element.port_num -
418 					    sa_dev->start_port].update_task);
419 	}
420 }
421 
422 void ib_sa_register_client(struct ib_sa_client *client)
423 {
424 	atomic_set(&client->users, 1);
425 	init_completion(&client->comp);
426 }
427 EXPORT_SYMBOL(ib_sa_register_client);
428 
429 void ib_sa_unregister_client(struct ib_sa_client *client)
430 {
431 	ib_sa_client_put(client);
432 	wait_for_completion(&client->comp);
433 }
434 EXPORT_SYMBOL(ib_sa_unregister_client);
435 
436 /**
437  * ib_sa_cancel_query - try to cancel an SA query
438  * @id:ID of query to cancel
439  * @query:query pointer to cancel
440  *
441  * Try to cancel an SA query.  If the id and query don't match up or
442  * the query has already completed, nothing is done.  Otherwise the
443  * query is canceled and will complete with a status of -EINTR.
444  */
445 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
446 {
447 	unsigned long flags;
448 	struct ib_mad_agent *agent;
449 	struct ib_mad_send_buf *mad_buf;
450 
451 	spin_lock_irqsave(&idr_lock, flags);
452 	if (idr_find(&query_idr, id) != query) {
453 		spin_unlock_irqrestore(&idr_lock, flags);
454 		return;
455 	}
456 	agent = query->port->agent;
457 	mad_buf = query->mad_buf;
458 	spin_unlock_irqrestore(&idr_lock, flags);
459 
460 	ib_cancel_mad(agent, mad_buf);
461 }
462 EXPORT_SYMBOL(ib_sa_cancel_query);
463 
464 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
465 {
466 	struct ib_sa_device *sa_dev;
467 	struct ib_sa_port   *port;
468 	unsigned long flags;
469 	u8 src_path_mask;
470 
471 	sa_dev = ib_get_client_data(device, &sa_client);
472 	if (!sa_dev)
473 		return 0x7f;
474 
475 	port  = &sa_dev->port[port_num - sa_dev->start_port];
476 	spin_lock_irqsave(&port->ah_lock, flags);
477 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
478 	spin_unlock_irqrestore(&port->ah_lock, flags);
479 
480 	return src_path_mask;
481 }
482 
483 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
484 			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
485 {
486 	int ret;
487 	u16 gid_index;
488 
489 	memset(ah_attr, 0, sizeof *ah_attr);
490 	ah_attr->dlid = be16_to_cpu(rec->dlid);
491 	ah_attr->sl = rec->sl;
492 	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
493 				 get_src_path_mask(device, port_num);
494 	ah_attr->port_num = port_num;
495 	ah_attr->static_rate = rec->rate;
496 
497 	if (rec->hop_limit > 1) {
498 		ah_attr->ah_flags = IB_AH_GRH;
499 		ah_attr->grh.dgid = rec->dgid;
500 
501 		ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
502 					 &gid_index);
503 		if (ret)
504 			return ret;
505 
506 		ah_attr->grh.sgid_index    = gid_index;
507 		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
508 		ah_attr->grh.hop_limit     = rec->hop_limit;
509 		ah_attr->grh.traffic_class = rec->traffic_class;
510 	}
511 	return 0;
512 }
513 EXPORT_SYMBOL(ib_init_ah_from_path);
514 
515 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
516 {
517 	unsigned long flags;
518 
519 	memset(mad, 0, sizeof *mad);
520 
521 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
522 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
523 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
524 
525 	spin_lock_irqsave(&tid_lock, flags);
526 	mad->mad_hdr.tid           =
527 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
528 	spin_unlock_irqrestore(&tid_lock, flags);
529 }
530 
531 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
532 {
533 	unsigned long flags;
534 	int ret, id;
535 
536 retry:
537 	if (!idr_pre_get(&query_idr, gfp_mask))
538 		return -ENOMEM;
539 	spin_lock_irqsave(&idr_lock, flags);
540 	ret = idr_get_new(&query_idr, query, &id);
541 	spin_unlock_irqrestore(&idr_lock, flags);
542 	if (ret == -EAGAIN)
543 		goto retry;
544 	if (ret)
545 		return ret;
546 
547 	query->mad_buf->timeout_ms  = timeout_ms;
548 	query->mad_buf->context[0] = query;
549 	query->id = id;
550 
551 	spin_lock_irqsave(&query->port->ah_lock, flags);
552 	kref_get(&query->port->sm_ah->ref);
553 	query->sm_ah = query->port->sm_ah;
554 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
555 
556 	query->mad_buf->ah = query->sm_ah->ah;
557 
558 	ret = ib_post_send_mad(query->mad_buf, NULL);
559 	if (ret) {
560 		spin_lock_irqsave(&idr_lock, flags);
561 		idr_remove(&query_idr, id);
562 		spin_unlock_irqrestore(&idr_lock, flags);
563 
564 		kref_put(&query->sm_ah->ref, free_sm_ah);
565 	}
566 
567 	/*
568 	 * It's not safe to dereference query any more, because the
569 	 * send may already have completed and freed the query in
570 	 * another context.
571 	 */
572 	return ret ? ret : id;
573 }
574 
575 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
576 				    int status,
577 				    struct ib_sa_mad *mad)
578 {
579 	struct ib_sa_path_query *query =
580 		container_of(sa_query, struct ib_sa_path_query, sa_query);
581 
582 	if (mad) {
583 		struct ib_sa_path_rec rec;
584 
585 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
586 			  mad->data, &rec);
587 		query->callback(status, &rec, query->context);
588 	} else
589 		query->callback(status, NULL, query->context);
590 }
591 
592 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
593 {
594 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
595 }
596 
597 /**
598  * ib_sa_path_rec_get - Start a Path get query
599  * @client:SA client
600  * @device:device to send query on
601  * @port_num: port number to send query on
602  * @rec:Path Record to send in query
603  * @comp_mask:component mask to send in query
604  * @timeout_ms:time to wait for response
605  * @gfp_mask:GFP mask to use for internal allocations
606  * @callback:function called when query completes, times out or is
607  * canceled
608  * @context:opaque user context passed to callback
609  * @sa_query:query context, used to cancel query
610  *
611  * Send a Path Record Get query to the SA to look up a path.  The
612  * callback function will be called when the query completes (or
613  * fails); status is 0 for a successful response, -EINTR if the query
614  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
615  * occurred sending the query.  The resp parameter of the callback is
616  * only valid if status is 0.
617  *
618  * If the return value of ib_sa_path_rec_get() is negative, it is an
619  * error code.  Otherwise it is a query ID that can be used to cancel
620  * the query.
621  */
622 int ib_sa_path_rec_get(struct ib_sa_client *client,
623 		       struct ib_device *device, u8 port_num,
624 		       struct ib_sa_path_rec *rec,
625 		       ib_sa_comp_mask comp_mask,
626 		       int timeout_ms, gfp_t gfp_mask,
627 		       void (*callback)(int status,
628 					struct ib_sa_path_rec *resp,
629 					void *context),
630 		       void *context,
631 		       struct ib_sa_query **sa_query)
632 {
633 	struct ib_sa_path_query *query;
634 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
635 	struct ib_sa_port   *port;
636 	struct ib_mad_agent *agent;
637 	struct ib_sa_mad *mad;
638 	int ret;
639 
640 	if (!sa_dev)
641 		return -ENODEV;
642 
643 	port  = &sa_dev->port[port_num - sa_dev->start_port];
644 	agent = port->agent;
645 
646 	query = kmalloc(sizeof *query, gfp_mask);
647 	if (!query)
648 		return -ENOMEM;
649 
650 	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
651 						     0, IB_MGMT_SA_HDR,
652 						     IB_MGMT_SA_DATA, gfp_mask);
653 	if (!query->sa_query.mad_buf) {
654 		ret = -ENOMEM;
655 		goto err1;
656 	}
657 
658 	ib_sa_client_get(client);
659 	query->sa_query.client = client;
660 	query->callback        = callback;
661 	query->context         = context;
662 
663 	mad = query->sa_query.mad_buf->mad;
664 	init_mad(mad, agent);
665 
666 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
667 	query->sa_query.release  = ib_sa_path_rec_release;
668 	query->sa_query.port     = port;
669 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
670 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
671 	mad->sa_hdr.comp_mask	 = comp_mask;
672 
673 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
674 
675 	*sa_query = &query->sa_query;
676 
677 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
678 	if (ret < 0)
679 		goto err2;
680 
681 	return ret;
682 
683 err2:
684 	*sa_query = NULL;
685 	ib_sa_client_put(query->sa_query.client);
686 	ib_free_send_mad(query->sa_query.mad_buf);
687 
688 err1:
689 	kfree(query);
690 	return ret;
691 }
692 EXPORT_SYMBOL(ib_sa_path_rec_get);
693 
694 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
695 				    int status,
696 				    struct ib_sa_mad *mad)
697 {
698 	struct ib_sa_service_query *query =
699 		container_of(sa_query, struct ib_sa_service_query, sa_query);
700 
701 	if (mad) {
702 		struct ib_sa_service_rec rec;
703 
704 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
705 			  mad->data, &rec);
706 		query->callback(status, &rec, query->context);
707 	} else
708 		query->callback(status, NULL, query->context);
709 }
710 
711 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
712 {
713 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
714 }
715 
716 /**
717  * ib_sa_service_rec_query - Start Service Record operation
718  * @client:SA client
719  * @device:device to send request on
720  * @port_num: port number to send request on
721  * @method:SA method - should be get, set, or delete
722  * @rec:Service Record to send in request
723  * @comp_mask:component mask to send in request
724  * @timeout_ms:time to wait for response
725  * @gfp_mask:GFP mask to use for internal allocations
726  * @callback:function called when request completes, times out or is
727  * canceled
728  * @context:opaque user context passed to callback
729  * @sa_query:request context, used to cancel request
730  *
731  * Send a Service Record set/get/delete to the SA to register,
732  * unregister or query a service record.
733  * The callback function will be called when the request completes (or
734  * fails); status is 0 for a successful response, -EINTR if the query
735  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
736  * occurred sending the query.  The resp parameter of the callback is
737  * only valid if status is 0.
738  *
739  * If the return value of ib_sa_service_rec_query() is negative, it is an
740  * error code.  Otherwise it is a request ID that can be used to cancel
741  * the query.
742  */
743 int ib_sa_service_rec_query(struct ib_sa_client *client,
744 			    struct ib_device *device, u8 port_num, u8 method,
745 			    struct ib_sa_service_rec *rec,
746 			    ib_sa_comp_mask comp_mask,
747 			    int timeout_ms, gfp_t gfp_mask,
748 			    void (*callback)(int status,
749 					     struct ib_sa_service_rec *resp,
750 					     void *context),
751 			    void *context,
752 			    struct ib_sa_query **sa_query)
753 {
754 	struct ib_sa_service_query *query;
755 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
756 	struct ib_sa_port   *port;
757 	struct ib_mad_agent *agent;
758 	struct ib_sa_mad *mad;
759 	int ret;
760 
761 	if (!sa_dev)
762 		return -ENODEV;
763 
764 	port  = &sa_dev->port[port_num - sa_dev->start_port];
765 	agent = port->agent;
766 
767 	if (method != IB_MGMT_METHOD_GET &&
768 	    method != IB_MGMT_METHOD_SET &&
769 	    method != IB_SA_METHOD_DELETE)
770 		return -EINVAL;
771 
772 	query = kmalloc(sizeof *query, gfp_mask);
773 	if (!query)
774 		return -ENOMEM;
775 
776 	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
777 						     0, IB_MGMT_SA_HDR,
778 						     IB_MGMT_SA_DATA, gfp_mask);
779 	if (!query->sa_query.mad_buf) {
780 		ret = -ENOMEM;
781 		goto err1;
782 	}
783 
784 	ib_sa_client_get(client);
785 	query->sa_query.client = client;
786 	query->callback        = callback;
787 	query->context         = context;
788 
789 	mad = query->sa_query.mad_buf->mad;
790 	init_mad(mad, agent);
791 
792 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
793 	query->sa_query.release  = ib_sa_service_rec_release;
794 	query->sa_query.port     = port;
795 	mad->mad_hdr.method	 = method;
796 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
797 	mad->sa_hdr.comp_mask	 = comp_mask;
798 
799 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
800 		rec, mad->data);
801 
802 	*sa_query = &query->sa_query;
803 
804 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
805 	if (ret < 0)
806 		goto err2;
807 
808 	return ret;
809 
810 err2:
811 	*sa_query = NULL;
812 	ib_sa_client_put(query->sa_query.client);
813 	ib_free_send_mad(query->sa_query.mad_buf);
814 
815 err1:
816 	kfree(query);
817 	return ret;
818 }
819 EXPORT_SYMBOL(ib_sa_service_rec_query);
820 
821 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
822 					int status,
823 					struct ib_sa_mad *mad)
824 {
825 	struct ib_sa_mcmember_query *query =
826 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
827 
828 	if (mad) {
829 		struct ib_sa_mcmember_rec rec;
830 
831 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
832 			  mad->data, &rec);
833 		query->callback(status, &rec, query->context);
834 	} else
835 		query->callback(status, NULL, query->context);
836 }
837 
838 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
839 {
840 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
841 }
842 
843 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
844 			     struct ib_device *device, u8 port_num,
845 			     u8 method,
846 			     struct ib_sa_mcmember_rec *rec,
847 			     ib_sa_comp_mask comp_mask,
848 			     int timeout_ms, gfp_t gfp_mask,
849 			     void (*callback)(int status,
850 					      struct ib_sa_mcmember_rec *resp,
851 					      void *context),
852 			     void *context,
853 			     struct ib_sa_query **sa_query)
854 {
855 	struct ib_sa_mcmember_query *query;
856 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
857 	struct ib_sa_port   *port;
858 	struct ib_mad_agent *agent;
859 	struct ib_sa_mad *mad;
860 	int ret;
861 
862 	if (!sa_dev)
863 		return -ENODEV;
864 
865 	port  = &sa_dev->port[port_num - sa_dev->start_port];
866 	agent = port->agent;
867 
868 	query = kmalloc(sizeof *query, gfp_mask);
869 	if (!query)
870 		return -ENOMEM;
871 
872 	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
873 						     0, IB_MGMT_SA_HDR,
874 						     IB_MGMT_SA_DATA, gfp_mask);
875 	if (!query->sa_query.mad_buf) {
876 		ret = -ENOMEM;
877 		goto err1;
878 	}
879 
880 	ib_sa_client_get(client);
881 	query->sa_query.client = client;
882 	query->callback        = callback;
883 	query->context         = context;
884 
885 	mad = query->sa_query.mad_buf->mad;
886 	init_mad(mad, agent);
887 
888 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
889 	query->sa_query.release  = ib_sa_mcmember_rec_release;
890 	query->sa_query.port     = port;
891 	mad->mad_hdr.method	 = method;
892 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
893 	mad->sa_hdr.comp_mask	 = comp_mask;
894 
895 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
896 		rec, mad->data);
897 
898 	*sa_query = &query->sa_query;
899 
900 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
901 	if (ret < 0)
902 		goto err2;
903 
904 	return ret;
905 
906 err2:
907 	*sa_query = NULL;
908 	ib_sa_client_put(query->sa_query.client);
909 	ib_free_send_mad(query->sa_query.mad_buf);
910 
911 err1:
912 	kfree(query);
913 	return ret;
914 }
915 
916 static void send_handler(struct ib_mad_agent *agent,
917 			 struct ib_mad_send_wc *mad_send_wc)
918 {
919 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
920 	unsigned long flags;
921 
922 	if (query->callback)
923 		switch (mad_send_wc->status) {
924 		case IB_WC_SUCCESS:
925 			/* No callback -- already got recv */
926 			break;
927 		case IB_WC_RESP_TIMEOUT_ERR:
928 			query->callback(query, -ETIMEDOUT, NULL);
929 			break;
930 		case IB_WC_WR_FLUSH_ERR:
931 			query->callback(query, -EINTR, NULL);
932 			break;
933 		default:
934 			query->callback(query, -EIO, NULL);
935 			break;
936 		}
937 
938 	spin_lock_irqsave(&idr_lock, flags);
939 	idr_remove(&query_idr, query->id);
940 	spin_unlock_irqrestore(&idr_lock, flags);
941 
942 	ib_free_send_mad(mad_send_wc->send_buf);
943 	kref_put(&query->sm_ah->ref, free_sm_ah);
944 	ib_sa_client_put(query->client);
945 	query->release(query);
946 }
947 
948 static void recv_handler(struct ib_mad_agent *mad_agent,
949 			 struct ib_mad_recv_wc *mad_recv_wc)
950 {
951 	struct ib_sa_query *query;
952 	struct ib_mad_send_buf *mad_buf;
953 
954 	mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
955 	query = mad_buf->context[0];
956 
957 	if (query->callback) {
958 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
959 			query->callback(query,
960 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
961 					-EINVAL : 0,
962 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
963 		else
964 			query->callback(query, -EIO, NULL);
965 	}
966 
967 	ib_free_recv_mad(mad_recv_wc);
968 }
969 
970 static void ib_sa_add_one(struct ib_device *device)
971 {
972 	struct ib_sa_device *sa_dev;
973 	int s, e, i;
974 
975 	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
976 		return;
977 
978 	if (device->node_type == RDMA_NODE_IB_SWITCH)
979 		s = e = 0;
980 	else {
981 		s = 1;
982 		e = device->phys_port_cnt;
983 	}
984 
985 	sa_dev = kmalloc(sizeof *sa_dev +
986 			 (e - s + 1) * sizeof (struct ib_sa_port),
987 			 GFP_KERNEL);
988 	if (!sa_dev)
989 		return;
990 
991 	sa_dev->start_port = s;
992 	sa_dev->end_port   = e;
993 
994 	for (i = 0; i <= e - s; ++i) {
995 		sa_dev->port[i].sm_ah    = NULL;
996 		sa_dev->port[i].port_num = i + s;
997 		spin_lock_init(&sa_dev->port[i].ah_lock);
998 
999 		sa_dev->port[i].agent =
1000 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1001 					      NULL, 0, send_handler,
1002 					      recv_handler, sa_dev);
1003 		if (IS_ERR(sa_dev->port[i].agent))
1004 			goto err;
1005 
1006 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1007 	}
1008 
1009 	ib_set_client_data(device, &sa_client, sa_dev);
1010 
1011 	/*
1012 	 * We register our event handler after everything is set up,
1013 	 * and then update our cached info after the event handler is
1014 	 * registered to avoid any problems if a port changes state
1015 	 * during our initialization.
1016 	 */
1017 
1018 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1019 	if (ib_register_event_handler(&sa_dev->event_handler))
1020 		goto err;
1021 
1022 	for (i = 0; i <= e - s; ++i)
1023 		update_sm_ah(&sa_dev->port[i].update_task);
1024 
1025 	return;
1026 
1027 err:
1028 	while (--i >= 0)
1029 		ib_unregister_mad_agent(sa_dev->port[i].agent);
1030 
1031 	kfree(sa_dev);
1032 
1033 	return;
1034 }
1035 
1036 static void ib_sa_remove_one(struct ib_device *device)
1037 {
1038 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1039 	int i;
1040 
1041 	if (!sa_dev)
1042 		return;
1043 
1044 	ib_unregister_event_handler(&sa_dev->event_handler);
1045 
1046 	flush_scheduled_work();
1047 
1048 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1049 		ib_unregister_mad_agent(sa_dev->port[i].agent);
1050 		kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1051 	}
1052 
1053 	kfree(sa_dev);
1054 }
1055 
1056 static int __init ib_sa_init(void)
1057 {
1058 	int ret;
1059 
1060 	spin_lock_init(&idr_lock);
1061 	spin_lock_init(&tid_lock);
1062 
1063 	get_random_bytes(&tid, sizeof tid);
1064 
1065 	ret = ib_register_client(&sa_client);
1066 	if (ret) {
1067 		printk(KERN_ERR "Couldn't register ib_sa client\n");
1068 		goto err1;
1069 	}
1070 
1071 	ret = mcast_init();
1072 	if (ret) {
1073 		printk(KERN_ERR "Couldn't initialize multicast handling\n");
1074 		goto err2;
1075 	}
1076 
1077 	return 0;
1078 err2:
1079 	ib_unregister_client(&sa_client);
1080 err1:
1081 	return ret;
1082 }
1083 
1084 static void __exit ib_sa_cleanup(void)
1085 {
1086 	mcast_cleanup();
1087 	ib_unregister_client(&sa_client);
1088 	idr_destroy(&query_idr);
1089 }
1090 
1091 module_init(ib_sa_init);
1092 module_exit(ib_sa_cleanup);
1093