]> Git Repo - linux.git/blob - drivers/infiniband/core/sa_query.c
quota: Wire up quotactl_fd syscall
[linux.git] / drivers / infiniband / core / sa_query.c
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/xarray.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 <rdma/opa_addr.h>
54 #include "sa.h"
55 #include "core_priv.h"
56
57 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN             100
58 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT         2000
59 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX             200000
60 #define IB_SA_CPI_MAX_RETRY_CNT                 3
61 #define IB_SA_CPI_RETRY_WAIT                    1000 /*msecs */
62 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
63
64 struct ib_sa_sm_ah {
65         struct ib_ah        *ah;
66         struct kref          ref;
67         u16                  pkey_index;
68         u8                   src_path_mask;
69 };
70
71 enum rdma_class_port_info_type {
72         RDMA_CLASS_PORT_INFO_IB,
73         RDMA_CLASS_PORT_INFO_OPA
74 };
75
76 struct rdma_class_port_info {
77         enum rdma_class_port_info_type type;
78         union {
79                 struct ib_class_port_info ib;
80                 struct opa_class_port_info opa;
81         };
82 };
83
84 struct ib_sa_classport_cache {
85         bool valid;
86         int retry_cnt;
87         struct rdma_class_port_info data;
88 };
89
90 struct ib_sa_port {
91         struct ib_mad_agent *agent;
92         struct ib_sa_sm_ah  *sm_ah;
93         struct work_struct   update_task;
94         struct ib_sa_classport_cache classport_info;
95         struct delayed_work ib_cpi_work;
96         spinlock_t                   classport_lock; /* protects class port info set */
97         spinlock_t           ah_lock;
98         u32                  port_num;
99 };
100
101 struct ib_sa_device {
102         int                     start_port, end_port;
103         struct ib_event_handler event_handler;
104         struct ib_sa_port port[];
105 };
106
107 struct ib_sa_query {
108         void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
109         void (*release)(struct ib_sa_query *);
110         struct ib_sa_client    *client;
111         struct ib_sa_port      *port;
112         struct ib_mad_send_buf *mad_buf;
113         struct ib_sa_sm_ah     *sm_ah;
114         int                     id;
115         u32                     flags;
116         struct list_head        list; /* Local svc request list */
117         u32                     seq; /* Local svc request sequence number */
118         unsigned long           timeout; /* Local svc timeout */
119         u8                      path_use; /* How will the pathrecord be used */
120 };
121
122 #define IB_SA_ENABLE_LOCAL_SERVICE      0x00000001
123 #define IB_SA_CANCEL                    0x00000002
124 #define IB_SA_QUERY_OPA                 0x00000004
125
126 struct ib_sa_service_query {
127         void (*callback)(int, struct ib_sa_service_rec *, void *);
128         void *context;
129         struct ib_sa_query sa_query;
130 };
131
132 struct ib_sa_path_query {
133         void (*callback)(int, struct sa_path_rec *, void *);
134         void *context;
135         struct ib_sa_query sa_query;
136         struct sa_path_rec *conv_pr;
137 };
138
139 struct ib_sa_guidinfo_query {
140         void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
141         void *context;
142         struct ib_sa_query sa_query;
143 };
144
145 struct ib_sa_classport_info_query {
146         void (*callback)(void *);
147         void *context;
148         struct ib_sa_query sa_query;
149 };
150
151 struct ib_sa_mcmember_query {
152         void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
153         void *context;
154         struct ib_sa_query sa_query;
155 };
156
157 static LIST_HEAD(ib_nl_request_list);
158 static DEFINE_SPINLOCK(ib_nl_request_lock);
159 static atomic_t ib_nl_sa_request_seq;
160 static struct workqueue_struct *ib_nl_wq;
161 static struct delayed_work ib_nl_timed_work;
162 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
163         [LS_NLA_TYPE_PATH_RECORD]       = {.type = NLA_BINARY,
164                 .len = sizeof(struct ib_path_rec_data)},
165         [LS_NLA_TYPE_TIMEOUT]           = {.type = NLA_U32},
166         [LS_NLA_TYPE_SERVICE_ID]        = {.type = NLA_U64},
167         [LS_NLA_TYPE_DGID]              = {.type = NLA_BINARY,
168                 .len = sizeof(struct rdma_nla_ls_gid)},
169         [LS_NLA_TYPE_SGID]              = {.type = NLA_BINARY,
170                 .len = sizeof(struct rdma_nla_ls_gid)},
171         [LS_NLA_TYPE_TCLASS]            = {.type = NLA_U8},
172         [LS_NLA_TYPE_PKEY]              = {.type = NLA_U16},
173         [LS_NLA_TYPE_QOS_CLASS]         = {.type = NLA_U16},
174 };
175
176
177 static int ib_sa_add_one(struct ib_device *device);
178 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
179
180 static struct ib_client sa_client = {
181         .name   = "sa",
182         .add    = ib_sa_add_one,
183         .remove = ib_sa_remove_one
184 };
185
186 static DEFINE_XARRAY_FLAGS(queries, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ);
187
188 static DEFINE_SPINLOCK(tid_lock);
189 static u32 tid;
190
191 #define PATH_REC_FIELD(field) \
192         .struct_offset_bytes = offsetof(struct sa_path_rec, field),     \
193         .struct_size_bytes   = sizeof_field(struct sa_path_rec, field), \
194         .field_name          = "sa_path_rec:" #field
195
196 static const struct ib_field path_rec_table[] = {
197         { PATH_REC_FIELD(service_id),
198           .offset_words = 0,
199           .offset_bits  = 0,
200           .size_bits    = 64 },
201         { PATH_REC_FIELD(dgid),
202           .offset_words = 2,
203           .offset_bits  = 0,
204           .size_bits    = 128 },
205         { PATH_REC_FIELD(sgid),
206           .offset_words = 6,
207           .offset_bits  = 0,
208           .size_bits    = 128 },
209         { PATH_REC_FIELD(ib.dlid),
210           .offset_words = 10,
211           .offset_bits  = 0,
212           .size_bits    = 16 },
213         { PATH_REC_FIELD(ib.slid),
214           .offset_words = 10,
215           .offset_bits  = 16,
216           .size_bits    = 16 },
217         { PATH_REC_FIELD(ib.raw_traffic),
218           .offset_words = 11,
219           .offset_bits  = 0,
220           .size_bits    = 1 },
221         { RESERVED,
222           .offset_words = 11,
223           .offset_bits  = 1,
224           .size_bits    = 3 },
225         { PATH_REC_FIELD(flow_label),
226           .offset_words = 11,
227           .offset_bits  = 4,
228           .size_bits    = 20 },
229         { PATH_REC_FIELD(hop_limit),
230           .offset_words = 11,
231           .offset_bits  = 24,
232           .size_bits    = 8 },
233         { PATH_REC_FIELD(traffic_class),
234           .offset_words = 12,
235           .offset_bits  = 0,
236           .size_bits    = 8 },
237         { PATH_REC_FIELD(reversible),
238           .offset_words = 12,
239           .offset_bits  = 8,
240           .size_bits    = 1 },
241         { PATH_REC_FIELD(numb_path),
242           .offset_words = 12,
243           .offset_bits  = 9,
244           .size_bits    = 7 },
245         { PATH_REC_FIELD(pkey),
246           .offset_words = 12,
247           .offset_bits  = 16,
248           .size_bits    = 16 },
249         { PATH_REC_FIELD(qos_class),
250           .offset_words = 13,
251           .offset_bits  = 0,
252           .size_bits    = 12 },
253         { PATH_REC_FIELD(sl),
254           .offset_words = 13,
255           .offset_bits  = 12,
256           .size_bits    = 4 },
257         { PATH_REC_FIELD(mtu_selector),
258           .offset_words = 13,
259           .offset_bits  = 16,
260           .size_bits    = 2 },
261         { PATH_REC_FIELD(mtu),
262           .offset_words = 13,
263           .offset_bits  = 18,
264           .size_bits    = 6 },
265         { PATH_REC_FIELD(rate_selector),
266           .offset_words = 13,
267           .offset_bits  = 24,
268           .size_bits    = 2 },
269         { PATH_REC_FIELD(rate),
270           .offset_words = 13,
271           .offset_bits  = 26,
272           .size_bits    = 6 },
273         { PATH_REC_FIELD(packet_life_time_selector),
274           .offset_words = 14,
275           .offset_bits  = 0,
276           .size_bits    = 2 },
277         { PATH_REC_FIELD(packet_life_time),
278           .offset_words = 14,
279           .offset_bits  = 2,
280           .size_bits    = 6 },
281         { PATH_REC_FIELD(preference),
282           .offset_words = 14,
283           .offset_bits  = 8,
284           .size_bits    = 8 },
285         { RESERVED,
286           .offset_words = 14,
287           .offset_bits  = 16,
288           .size_bits    = 48 },
289 };
290
291 #define OPA_PATH_REC_FIELD(field) \
292         .struct_offset_bytes = \
293                 offsetof(struct sa_path_rec, field), \
294         .struct_size_bytes   = \
295                 sizeof_field(struct sa_path_rec, field),        \
296         .field_name          = "sa_path_rec:" #field
297
298 static const struct ib_field opa_path_rec_table[] = {
299         { OPA_PATH_REC_FIELD(service_id),
300           .offset_words = 0,
301           .offset_bits  = 0,
302           .size_bits    = 64 },
303         { OPA_PATH_REC_FIELD(dgid),
304           .offset_words = 2,
305           .offset_bits  = 0,
306           .size_bits    = 128 },
307         { OPA_PATH_REC_FIELD(sgid),
308           .offset_words = 6,
309           .offset_bits  = 0,
310           .size_bits    = 128 },
311         { OPA_PATH_REC_FIELD(opa.dlid),
312           .offset_words = 10,
313           .offset_bits  = 0,
314           .size_bits    = 32 },
315         { OPA_PATH_REC_FIELD(opa.slid),
316           .offset_words = 11,
317           .offset_bits  = 0,
318           .size_bits    = 32 },
319         { OPA_PATH_REC_FIELD(opa.raw_traffic),
320           .offset_words = 12,
321           .offset_bits  = 0,
322           .size_bits    = 1 },
323         { RESERVED,
324           .offset_words = 12,
325           .offset_bits  = 1,
326           .size_bits    = 3 },
327         { OPA_PATH_REC_FIELD(flow_label),
328           .offset_words = 12,
329           .offset_bits  = 4,
330           .size_bits    = 20 },
331         { OPA_PATH_REC_FIELD(hop_limit),
332           .offset_words = 12,
333           .offset_bits  = 24,
334           .size_bits    = 8 },
335         { OPA_PATH_REC_FIELD(traffic_class),
336           .offset_words = 13,
337           .offset_bits  = 0,
338           .size_bits    = 8 },
339         { OPA_PATH_REC_FIELD(reversible),
340           .offset_words = 13,
341           .offset_bits  = 8,
342           .size_bits    = 1 },
343         { OPA_PATH_REC_FIELD(numb_path),
344           .offset_words = 13,
345           .offset_bits  = 9,
346           .size_bits    = 7 },
347         { OPA_PATH_REC_FIELD(pkey),
348           .offset_words = 13,
349           .offset_bits  = 16,
350           .size_bits    = 16 },
351         { OPA_PATH_REC_FIELD(opa.l2_8B),
352           .offset_words = 14,
353           .offset_bits  = 0,
354           .size_bits    = 1 },
355         { OPA_PATH_REC_FIELD(opa.l2_10B),
356           .offset_words = 14,
357           .offset_bits  = 1,
358           .size_bits    = 1 },
359         { OPA_PATH_REC_FIELD(opa.l2_9B),
360           .offset_words = 14,
361           .offset_bits  = 2,
362           .size_bits    = 1 },
363         { OPA_PATH_REC_FIELD(opa.l2_16B),
364           .offset_words = 14,
365           .offset_bits  = 3,
366           .size_bits    = 1 },
367         { RESERVED,
368           .offset_words = 14,
369           .offset_bits  = 4,
370           .size_bits    = 2 },
371         { OPA_PATH_REC_FIELD(opa.qos_type),
372           .offset_words = 14,
373           .offset_bits  = 6,
374           .size_bits    = 2 },
375         { OPA_PATH_REC_FIELD(opa.qos_priority),
376           .offset_words = 14,
377           .offset_bits  = 8,
378           .size_bits    = 8 },
379         { RESERVED,
380           .offset_words = 14,
381           .offset_bits  = 16,
382           .size_bits    = 3 },
383         { OPA_PATH_REC_FIELD(sl),
384           .offset_words = 14,
385           .offset_bits  = 19,
386           .size_bits    = 5 },
387         { RESERVED,
388           .offset_words = 14,
389           .offset_bits  = 24,
390           .size_bits    = 8 },
391         { OPA_PATH_REC_FIELD(mtu_selector),
392           .offset_words = 15,
393           .offset_bits  = 0,
394           .size_bits    = 2 },
395         { OPA_PATH_REC_FIELD(mtu),
396           .offset_words = 15,
397           .offset_bits  = 2,
398           .size_bits    = 6 },
399         { OPA_PATH_REC_FIELD(rate_selector),
400           .offset_words = 15,
401           .offset_bits  = 8,
402           .size_bits    = 2 },
403         { OPA_PATH_REC_FIELD(rate),
404           .offset_words = 15,
405           .offset_bits  = 10,
406           .size_bits    = 6 },
407         { OPA_PATH_REC_FIELD(packet_life_time_selector),
408           .offset_words = 15,
409           .offset_bits  = 16,
410           .size_bits    = 2 },
411         { OPA_PATH_REC_FIELD(packet_life_time),
412           .offset_words = 15,
413           .offset_bits  = 18,
414           .size_bits    = 6 },
415         { OPA_PATH_REC_FIELD(preference),
416           .offset_words = 15,
417           .offset_bits  = 24,
418           .size_bits    = 8 },
419 };
420
421 #define MCMEMBER_REC_FIELD(field) \
422         .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),      \
423         .struct_size_bytes   = sizeof_field(struct ib_sa_mcmember_rec, field),  \
424         .field_name          = "sa_mcmember_rec:" #field
425
426 static const struct ib_field mcmember_rec_table[] = {
427         { MCMEMBER_REC_FIELD(mgid),
428           .offset_words = 0,
429           .offset_bits  = 0,
430           .size_bits    = 128 },
431         { MCMEMBER_REC_FIELD(port_gid),
432           .offset_words = 4,
433           .offset_bits  = 0,
434           .size_bits    = 128 },
435         { MCMEMBER_REC_FIELD(qkey),
436           .offset_words = 8,
437           .offset_bits  = 0,
438           .size_bits    = 32 },
439         { MCMEMBER_REC_FIELD(mlid),
440           .offset_words = 9,
441           .offset_bits  = 0,
442           .size_bits    = 16 },
443         { MCMEMBER_REC_FIELD(mtu_selector),
444           .offset_words = 9,
445           .offset_bits  = 16,
446           .size_bits    = 2 },
447         { MCMEMBER_REC_FIELD(mtu),
448           .offset_words = 9,
449           .offset_bits  = 18,
450           .size_bits    = 6 },
451         { MCMEMBER_REC_FIELD(traffic_class),
452           .offset_words = 9,
453           .offset_bits  = 24,
454           .size_bits    = 8 },
455         { MCMEMBER_REC_FIELD(pkey),
456           .offset_words = 10,
457           .offset_bits  = 0,
458           .size_bits    = 16 },
459         { MCMEMBER_REC_FIELD(rate_selector),
460           .offset_words = 10,
461           .offset_bits  = 16,
462           .size_bits    = 2 },
463         { MCMEMBER_REC_FIELD(rate),
464           .offset_words = 10,
465           .offset_bits  = 18,
466           .size_bits    = 6 },
467         { MCMEMBER_REC_FIELD(packet_life_time_selector),
468           .offset_words = 10,
469           .offset_bits  = 24,
470           .size_bits    = 2 },
471         { MCMEMBER_REC_FIELD(packet_life_time),
472           .offset_words = 10,
473           .offset_bits  = 26,
474           .size_bits    = 6 },
475         { MCMEMBER_REC_FIELD(sl),
476           .offset_words = 11,
477           .offset_bits  = 0,
478           .size_bits    = 4 },
479         { MCMEMBER_REC_FIELD(flow_label),
480           .offset_words = 11,
481           .offset_bits  = 4,
482           .size_bits    = 20 },
483         { MCMEMBER_REC_FIELD(hop_limit),
484           .offset_words = 11,
485           .offset_bits  = 24,
486           .size_bits    = 8 },
487         { MCMEMBER_REC_FIELD(scope),
488           .offset_words = 12,
489           .offset_bits  = 0,
490           .size_bits    = 4 },
491         { MCMEMBER_REC_FIELD(join_state),
492           .offset_words = 12,
493           .offset_bits  = 4,
494           .size_bits    = 4 },
495         { MCMEMBER_REC_FIELD(proxy_join),
496           .offset_words = 12,
497           .offset_bits  = 8,
498           .size_bits    = 1 },
499         { RESERVED,
500           .offset_words = 12,
501           .offset_bits  = 9,
502           .size_bits    = 23 },
503 };
504
505 #define SERVICE_REC_FIELD(field) \
506         .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),       \
507         .struct_size_bytes   = sizeof_field(struct ib_sa_service_rec, field),   \
508         .field_name          = "sa_service_rec:" #field
509
510 static const struct ib_field service_rec_table[] = {
511         { SERVICE_REC_FIELD(id),
512           .offset_words = 0,
513           .offset_bits  = 0,
514           .size_bits    = 64 },
515         { SERVICE_REC_FIELD(gid),
516           .offset_words = 2,
517           .offset_bits  = 0,
518           .size_bits    = 128 },
519         { SERVICE_REC_FIELD(pkey),
520           .offset_words = 6,
521           .offset_bits  = 0,
522           .size_bits    = 16 },
523         { SERVICE_REC_FIELD(lease),
524           .offset_words = 7,
525           .offset_bits  = 0,
526           .size_bits    = 32 },
527         { SERVICE_REC_FIELD(key),
528           .offset_words = 8,
529           .offset_bits  = 0,
530           .size_bits    = 128 },
531         { SERVICE_REC_FIELD(name),
532           .offset_words = 12,
533           .offset_bits  = 0,
534           .size_bits    = 64*8 },
535         { SERVICE_REC_FIELD(data8),
536           .offset_words = 28,
537           .offset_bits  = 0,
538           .size_bits    = 16*8 },
539         { SERVICE_REC_FIELD(data16),
540           .offset_words = 32,
541           .offset_bits  = 0,
542           .size_bits    = 8*16 },
543         { SERVICE_REC_FIELD(data32),
544           .offset_words = 36,
545           .offset_bits  = 0,
546           .size_bits    = 4*32 },
547         { SERVICE_REC_FIELD(data64),
548           .offset_words = 40,
549           .offset_bits  = 0,
550           .size_bits    = 2*64 },
551 };
552
553 #define CLASSPORTINFO_REC_FIELD(field) \
554         .struct_offset_bytes = offsetof(struct ib_class_port_info, field),      \
555         .struct_size_bytes   = sizeof_field(struct ib_class_port_info, field),  \
556         .field_name          = "ib_class_port_info:" #field
557
558 static const struct ib_field ib_classport_info_rec_table[] = {
559         { CLASSPORTINFO_REC_FIELD(base_version),
560           .offset_words = 0,
561           .offset_bits  = 0,
562           .size_bits    = 8 },
563         { CLASSPORTINFO_REC_FIELD(class_version),
564           .offset_words = 0,
565           .offset_bits  = 8,
566           .size_bits    = 8 },
567         { CLASSPORTINFO_REC_FIELD(capability_mask),
568           .offset_words = 0,
569           .offset_bits  = 16,
570           .size_bits    = 16 },
571         { CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
572           .offset_words = 1,
573           .offset_bits  = 0,
574           .size_bits    = 32 },
575         { CLASSPORTINFO_REC_FIELD(redirect_gid),
576           .offset_words = 2,
577           .offset_bits  = 0,
578           .size_bits    = 128 },
579         { CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
580           .offset_words = 6,
581           .offset_bits  = 0,
582           .size_bits    = 32 },
583         { CLASSPORTINFO_REC_FIELD(redirect_lid),
584           .offset_words = 7,
585           .offset_bits  = 0,
586           .size_bits    = 16 },
587         { CLASSPORTINFO_REC_FIELD(redirect_pkey),
588           .offset_words = 7,
589           .offset_bits  = 16,
590           .size_bits    = 16 },
591
592         { CLASSPORTINFO_REC_FIELD(redirect_qp),
593           .offset_words = 8,
594           .offset_bits  = 0,
595           .size_bits    = 32 },
596         { CLASSPORTINFO_REC_FIELD(redirect_qkey),
597           .offset_words = 9,
598           .offset_bits  = 0,
599           .size_bits    = 32 },
600
601         { CLASSPORTINFO_REC_FIELD(trap_gid),
602           .offset_words = 10,
603           .offset_bits  = 0,
604           .size_bits    = 128 },
605         { CLASSPORTINFO_REC_FIELD(trap_tcslfl),
606           .offset_words = 14,
607           .offset_bits  = 0,
608           .size_bits    = 32 },
609
610         { CLASSPORTINFO_REC_FIELD(trap_lid),
611           .offset_words = 15,
612           .offset_bits  = 0,
613           .size_bits    = 16 },
614         { CLASSPORTINFO_REC_FIELD(trap_pkey),
615           .offset_words = 15,
616           .offset_bits  = 16,
617           .size_bits    = 16 },
618
619         { CLASSPORTINFO_REC_FIELD(trap_hlqp),
620           .offset_words = 16,
621           .offset_bits  = 0,
622           .size_bits    = 32 },
623         { CLASSPORTINFO_REC_FIELD(trap_qkey),
624           .offset_words = 17,
625           .offset_bits  = 0,
626           .size_bits    = 32 },
627 };
628
629 #define OPA_CLASSPORTINFO_REC_FIELD(field) \
630         .struct_offset_bytes =\
631                 offsetof(struct opa_class_port_info, field),    \
632         .struct_size_bytes   = \
633                 sizeof_field(struct opa_class_port_info, field),        \
634         .field_name          = "opa_class_port_info:" #field
635
636 static const struct ib_field opa_classport_info_rec_table[] = {
637         { OPA_CLASSPORTINFO_REC_FIELD(base_version),
638           .offset_words = 0,
639           .offset_bits  = 0,
640           .size_bits    = 8 },
641         { OPA_CLASSPORTINFO_REC_FIELD(class_version),
642           .offset_words = 0,
643           .offset_bits  = 8,
644           .size_bits    = 8 },
645         { OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
646           .offset_words = 0,
647           .offset_bits  = 16,
648           .size_bits    = 16 },
649         { OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
650           .offset_words = 1,
651           .offset_bits  = 0,
652           .size_bits    = 32 },
653         { OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
654           .offset_words = 2,
655           .offset_bits  = 0,
656           .size_bits    = 128 },
657         { OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
658           .offset_words = 6,
659           .offset_bits  = 0,
660           .size_bits    = 32 },
661         { OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
662           .offset_words = 7,
663           .offset_bits  = 0,
664           .size_bits    = 32 },
665         { OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
666           .offset_words = 8,
667           .offset_bits  = 0,
668           .size_bits    = 32 },
669         { OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
670           .offset_words = 9,
671           .offset_bits  = 0,
672           .size_bits    = 32 },
673         { OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
674           .offset_words = 10,
675           .offset_bits  = 0,
676           .size_bits    = 128 },
677         { OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
678           .offset_words = 14,
679           .offset_bits  = 0,
680           .size_bits    = 32 },
681         { OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
682           .offset_words = 15,
683           .offset_bits  = 0,
684           .size_bits    = 32 },
685         { OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
686           .offset_words = 16,
687           .offset_bits  = 0,
688           .size_bits    = 32 },
689         { OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
690           .offset_words = 17,
691           .offset_bits  = 0,
692           .size_bits    = 32 },
693         { OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
694           .offset_words = 18,
695           .offset_bits  = 0,
696           .size_bits    = 16 },
697         { OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
698           .offset_words = 18,
699           .offset_bits  = 16,
700           .size_bits    = 16 },
701         { OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
702           .offset_words = 19,
703           .offset_bits  = 0,
704           .size_bits    = 8 },
705         { RESERVED,
706           .offset_words = 19,
707           .offset_bits  = 8,
708           .size_bits    = 24 },
709 };
710
711 #define GUIDINFO_REC_FIELD(field) \
712         .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),      \
713         .struct_size_bytes   = sizeof_field(struct ib_sa_guidinfo_rec, field),  \
714         .field_name          = "sa_guidinfo_rec:" #field
715
716 static const struct ib_field guidinfo_rec_table[] = {
717         { GUIDINFO_REC_FIELD(lid),
718           .offset_words = 0,
719           .offset_bits  = 0,
720           .size_bits    = 16 },
721         { GUIDINFO_REC_FIELD(block_num),
722           .offset_words = 0,
723           .offset_bits  = 16,
724           .size_bits    = 8 },
725         { GUIDINFO_REC_FIELD(res1),
726           .offset_words = 0,
727           .offset_bits  = 24,
728           .size_bits    = 8 },
729         { GUIDINFO_REC_FIELD(res2),
730           .offset_words = 1,
731           .offset_bits  = 0,
732           .size_bits    = 32 },
733         { GUIDINFO_REC_FIELD(guid_info_list),
734           .offset_words = 2,
735           .offset_bits  = 0,
736           .size_bits    = 512 },
737 };
738
739 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
740 {
741         query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
742 }
743
744 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
745 {
746         return (query->flags & IB_SA_CANCEL);
747 }
748
749 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
750                                      struct ib_sa_query *query)
751 {
752         struct sa_path_rec *sa_rec = query->mad_buf->context[1];
753         struct ib_sa_mad *mad = query->mad_buf->mad;
754         ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
755         u16 val16;
756         u64 val64;
757         struct rdma_ls_resolve_header *header;
758
759         query->mad_buf->context[1] = NULL;
760
761         /* Construct the family header first */
762         header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
763         memcpy(header->device_name, dev_name(&query->port->agent->device->dev),
764                LS_DEVICE_NAME_MAX);
765         header->port_num = query->port->port_num;
766
767         if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
768             sa_rec->reversible != 0)
769                 query->path_use = LS_RESOLVE_PATH_USE_GMP;
770         else
771                 query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
772         header->path_use = query->path_use;
773
774         /* Now build the attributes */
775         if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
776                 val64 = be64_to_cpu(sa_rec->service_id);
777                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
778                         sizeof(val64), &val64);
779         }
780         if (comp_mask & IB_SA_PATH_REC_DGID)
781                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
782                         sizeof(sa_rec->dgid), &sa_rec->dgid);
783         if (comp_mask & IB_SA_PATH_REC_SGID)
784                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
785                         sizeof(sa_rec->sgid), &sa_rec->sgid);
786         if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
787                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
788                         sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
789
790         if (comp_mask & IB_SA_PATH_REC_PKEY) {
791                 val16 = be16_to_cpu(sa_rec->pkey);
792                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
793                         sizeof(val16), &val16);
794         }
795         if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
796                 val16 = be16_to_cpu(sa_rec->qos_class);
797                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
798                         sizeof(val16), &val16);
799         }
800 }
801
802 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
803 {
804         int len = 0;
805
806         if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
807                 len += nla_total_size(sizeof(u64));
808         if (comp_mask & IB_SA_PATH_REC_DGID)
809                 len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
810         if (comp_mask & IB_SA_PATH_REC_SGID)
811                 len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
812         if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
813                 len += nla_total_size(sizeof(u8));
814         if (comp_mask & IB_SA_PATH_REC_PKEY)
815                 len += nla_total_size(sizeof(u16));
816         if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
817                 len += nla_total_size(sizeof(u16));
818
819         /*
820          * Make sure that at least some of the required comp_mask bits are
821          * set.
822          */
823         if (WARN_ON(len == 0))
824                 return len;
825
826         /* Add the family header */
827         len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
828
829         return len;
830 }
831
832 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
833 {
834         struct sk_buff *skb = NULL;
835         struct nlmsghdr *nlh;
836         void *data;
837         struct ib_sa_mad *mad;
838         int len;
839         unsigned long flags;
840         unsigned long delay;
841         gfp_t gfp_flag;
842         int ret;
843
844         INIT_LIST_HEAD(&query->list);
845         query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
846
847         mad = query->mad_buf->mad;
848         len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
849         if (len <= 0)
850                 return -EMSGSIZE;
851
852         skb = nlmsg_new(len, gfp_mask);
853         if (!skb)
854                 return -ENOMEM;
855
856         /* Put nlmsg header only for now */
857         data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
858                             RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
859         if (!data) {
860                 nlmsg_free(skb);
861                 return -EMSGSIZE;
862         }
863
864         /* Add attributes */
865         ib_nl_set_path_rec_attrs(skb, query);
866
867         /* Repair the nlmsg header length */
868         nlmsg_end(skb, nlh);
869
870         gfp_flag = ((gfp_mask & GFP_ATOMIC) == GFP_ATOMIC) ? GFP_ATOMIC :
871                 GFP_NOWAIT;
872
873         spin_lock_irqsave(&ib_nl_request_lock, flags);
874         ret = rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, gfp_flag);
875
876         if (ret)
877                 goto out;
878
879         /* Put the request on the list.*/
880         delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
881         query->timeout = delay + jiffies;
882         list_add_tail(&query->list, &ib_nl_request_list);
883         /* Start the timeout if this is the only request */
884         if (ib_nl_request_list.next == &query->list)
885                 queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
886
887 out:
888         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
889
890         return ret;
891 }
892
893 static int ib_nl_cancel_request(struct ib_sa_query *query)
894 {
895         unsigned long flags;
896         struct ib_sa_query *wait_query;
897         int found = 0;
898
899         spin_lock_irqsave(&ib_nl_request_lock, flags);
900         list_for_each_entry(wait_query, &ib_nl_request_list, list) {
901                 /* Let the timeout to take care of the callback */
902                 if (query == wait_query) {
903                         query->flags |= IB_SA_CANCEL;
904                         query->timeout = jiffies;
905                         list_move(&query->list, &ib_nl_request_list);
906                         found = 1;
907                         mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
908                         break;
909                 }
910         }
911         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
912
913         return found;
914 }
915
916 static void send_handler(struct ib_mad_agent *agent,
917                          struct ib_mad_send_wc *mad_send_wc);
918
919 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
920                                            const struct nlmsghdr *nlh)
921 {
922         struct ib_mad_send_wc mad_send_wc;
923         struct ib_sa_mad *mad = NULL;
924         const struct nlattr *head, *curr;
925         struct ib_path_rec_data  *rec;
926         int len, rem;
927         u32 mask = 0;
928         int status = -EIO;
929
930         if (query->callback) {
931                 head = (const struct nlattr *) nlmsg_data(nlh);
932                 len = nlmsg_len(nlh);
933                 switch (query->path_use) {
934                 case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
935                         mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
936                         break;
937
938                 case LS_RESOLVE_PATH_USE_ALL:
939                 case LS_RESOLVE_PATH_USE_GMP:
940                 default:
941                         mask = IB_PATH_PRIMARY | IB_PATH_GMP |
942                                 IB_PATH_BIDIRECTIONAL;
943                         break;
944                 }
945                 nla_for_each_attr(curr, head, len, rem) {
946                         if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
947                                 rec = nla_data(curr);
948                                 /*
949                                  * Get the first one. In the future, we may
950                                  * need to get up to 6 pathrecords.
951                                  */
952                                 if ((rec->flags & mask) == mask) {
953                                         mad = query->mad_buf->mad;
954                                         mad->mad_hdr.method |=
955                                                 IB_MGMT_METHOD_RESP;
956                                         memcpy(mad->data, rec->path_rec,
957                                                sizeof(rec->path_rec));
958                                         status = 0;
959                                         break;
960                                 }
961                         }
962                 }
963                 query->callback(query, status, mad);
964         }
965
966         mad_send_wc.send_buf = query->mad_buf;
967         mad_send_wc.status = IB_WC_SUCCESS;
968         send_handler(query->mad_buf->mad_agent, &mad_send_wc);
969 }
970
971 static void ib_nl_request_timeout(struct work_struct *work)
972 {
973         unsigned long flags;
974         struct ib_sa_query *query;
975         unsigned long delay;
976         struct ib_mad_send_wc mad_send_wc;
977         int ret;
978
979         spin_lock_irqsave(&ib_nl_request_lock, flags);
980         while (!list_empty(&ib_nl_request_list)) {
981                 query = list_entry(ib_nl_request_list.next,
982                                    struct ib_sa_query, list);
983
984                 if (time_after(query->timeout, jiffies)) {
985                         delay = query->timeout - jiffies;
986                         if ((long)delay <= 0)
987                                 delay = 1;
988                         queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
989                         break;
990                 }
991
992                 list_del(&query->list);
993                 ib_sa_disable_local_svc(query);
994                 /* Hold the lock to protect against query cancellation */
995                 if (ib_sa_query_cancelled(query))
996                         ret = -1;
997                 else
998                         ret = ib_post_send_mad(query->mad_buf, NULL);
999                 if (ret) {
1000                         mad_send_wc.send_buf = query->mad_buf;
1001                         mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1002                         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1003                         send_handler(query->port->agent, &mad_send_wc);
1004                         spin_lock_irqsave(&ib_nl_request_lock, flags);
1005                 }
1006         }
1007         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1008 }
1009
1010 int ib_nl_handle_set_timeout(struct sk_buff *skb,
1011                              struct nlmsghdr *nlh,
1012                              struct netlink_ext_ack *extack)
1013 {
1014         int timeout, delta, abs_delta;
1015         const struct nlattr *attr;
1016         unsigned long flags;
1017         struct ib_sa_query *query;
1018         long delay = 0;
1019         struct nlattr *tb[LS_NLA_TYPE_MAX];
1020         int ret;
1021
1022         if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1023             !(NETLINK_CB(skb).sk))
1024                 return -EPERM;
1025
1026         ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1027                                    nlmsg_len(nlh), ib_nl_policy, NULL);
1028         attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1029         if (ret || !attr)
1030                 goto settimeout_out;
1031
1032         timeout = *(int *) nla_data(attr);
1033         if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1034                 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1035         if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1036                 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1037
1038         delta = timeout - sa_local_svc_timeout_ms;
1039         if (delta < 0)
1040                 abs_delta = -delta;
1041         else
1042                 abs_delta = delta;
1043
1044         if (delta != 0) {
1045                 spin_lock_irqsave(&ib_nl_request_lock, flags);
1046                 sa_local_svc_timeout_ms = timeout;
1047                 list_for_each_entry(query, &ib_nl_request_list, list) {
1048                         if (delta < 0 && abs_delta > query->timeout)
1049                                 query->timeout = 0;
1050                         else
1051                                 query->timeout += delta;
1052
1053                         /* Get the new delay from the first entry */
1054                         if (!delay) {
1055                                 delay = query->timeout - jiffies;
1056                                 if (delay <= 0)
1057                                         delay = 1;
1058                         }
1059                 }
1060                 if (delay)
1061                         mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1062                                          (unsigned long)delay);
1063                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1064         }
1065
1066 settimeout_out:
1067         return 0;
1068 }
1069
1070 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1071 {
1072         struct nlattr *tb[LS_NLA_TYPE_MAX];
1073         int ret;
1074
1075         if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1076                 return 0;
1077
1078         ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1079                                    nlmsg_len(nlh), ib_nl_policy, NULL);
1080         if (ret)
1081                 return 0;
1082
1083         return 1;
1084 }
1085
1086 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1087                               struct nlmsghdr *nlh,
1088                               struct netlink_ext_ack *extack)
1089 {
1090         unsigned long flags;
1091         struct ib_sa_query *query;
1092         struct ib_mad_send_buf *send_buf;
1093         struct ib_mad_send_wc mad_send_wc;
1094         int found = 0;
1095         int ret;
1096
1097         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1098             !(NETLINK_CB(skb).sk))
1099                 return -EPERM;
1100
1101         spin_lock_irqsave(&ib_nl_request_lock, flags);
1102         list_for_each_entry(query, &ib_nl_request_list, list) {
1103                 /*
1104                  * If the query is cancelled, let the timeout routine
1105                  * take care of it.
1106                  */
1107                 if (nlh->nlmsg_seq == query->seq) {
1108                         found = !ib_sa_query_cancelled(query);
1109                         if (found)
1110                                 list_del(&query->list);
1111                         break;
1112                 }
1113         }
1114
1115         if (!found) {
1116                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1117                 goto resp_out;
1118         }
1119
1120         send_buf = query->mad_buf;
1121
1122         if (!ib_nl_is_good_resolve_resp(nlh)) {
1123                 /* if the result is a failure, send out the packet via IB */
1124                 ib_sa_disable_local_svc(query);
1125                 ret = ib_post_send_mad(query->mad_buf, NULL);
1126                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1127                 if (ret) {
1128                         mad_send_wc.send_buf = send_buf;
1129                         mad_send_wc.status = IB_WC_GENERAL_ERR;
1130                         send_handler(query->port->agent, &mad_send_wc);
1131                 }
1132         } else {
1133                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1134                 ib_nl_process_good_resolve_rsp(query, nlh);
1135         }
1136
1137 resp_out:
1138         return 0;
1139 }
1140
1141 static void free_sm_ah(struct kref *kref)
1142 {
1143         struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1144
1145         rdma_destroy_ah(sm_ah->ah, 0);
1146         kfree(sm_ah);
1147 }
1148
1149 void ib_sa_register_client(struct ib_sa_client *client)
1150 {
1151         atomic_set(&client->users, 1);
1152         init_completion(&client->comp);
1153 }
1154 EXPORT_SYMBOL(ib_sa_register_client);
1155
1156 void ib_sa_unregister_client(struct ib_sa_client *client)
1157 {
1158         ib_sa_client_put(client);
1159         wait_for_completion(&client->comp);
1160 }
1161 EXPORT_SYMBOL(ib_sa_unregister_client);
1162
1163 /**
1164  * ib_sa_cancel_query - try to cancel an SA query
1165  * @id:ID of query to cancel
1166  * @query:query pointer to cancel
1167  *
1168  * Try to cancel an SA query.  If the id and query don't match up or
1169  * the query has already completed, nothing is done.  Otherwise the
1170  * query is canceled and will complete with a status of -EINTR.
1171  */
1172 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1173 {
1174         unsigned long flags;
1175         struct ib_mad_agent *agent;
1176         struct ib_mad_send_buf *mad_buf;
1177
1178         xa_lock_irqsave(&queries, flags);
1179         if (xa_load(&queries, id) != query) {
1180                 xa_unlock_irqrestore(&queries, flags);
1181                 return;
1182         }
1183         agent = query->port->agent;
1184         mad_buf = query->mad_buf;
1185         xa_unlock_irqrestore(&queries, flags);
1186
1187         /*
1188          * If the query is still on the netlink request list, schedule
1189          * it to be cancelled by the timeout routine. Otherwise, it has been
1190          * sent to the MAD layer and has to be cancelled from there.
1191          */
1192         if (!ib_nl_cancel_request(query))
1193                 ib_cancel_mad(agent, mad_buf);
1194 }
1195 EXPORT_SYMBOL(ib_sa_cancel_query);
1196
1197 static u8 get_src_path_mask(struct ib_device *device, u32 port_num)
1198 {
1199         struct ib_sa_device *sa_dev;
1200         struct ib_sa_port   *port;
1201         unsigned long flags;
1202         u8 src_path_mask;
1203
1204         sa_dev = ib_get_client_data(device, &sa_client);
1205         if (!sa_dev)
1206                 return 0x7f;
1207
1208         port  = &sa_dev->port[port_num - sa_dev->start_port];
1209         spin_lock_irqsave(&port->ah_lock, flags);
1210         src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1211         spin_unlock_irqrestore(&port->ah_lock, flags);
1212
1213         return src_path_mask;
1214 }
1215
1216 static int init_ah_attr_grh_fields(struct ib_device *device, u32 port_num,
1217                                    struct sa_path_rec *rec,
1218                                    struct rdma_ah_attr *ah_attr,
1219                                    const struct ib_gid_attr *gid_attr)
1220 {
1221         enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1222
1223         if (!gid_attr) {
1224                 gid_attr = rdma_find_gid_by_port(device, &rec->sgid, type,
1225                                                  port_num, NULL);
1226                 if (IS_ERR(gid_attr))
1227                         return PTR_ERR(gid_attr);
1228         } else
1229                 rdma_hold_gid_attr(gid_attr);
1230
1231         rdma_move_grh_sgid_attr(ah_attr, &rec->dgid,
1232                                 be32_to_cpu(rec->flow_label),
1233                                 rec->hop_limit, rec->traffic_class,
1234                                 gid_attr);
1235         return 0;
1236 }
1237
1238 /**
1239  * ib_init_ah_attr_from_path - Initialize address handle attributes based on
1240  *   an SA path record.
1241  * @device: Device associated ah attributes initialization.
1242  * @port_num: Port on the specified device.
1243  * @rec: path record entry to use for ah attributes initialization.
1244  * @ah_attr: address handle attributes to initialization from path record.
1245  * @gid_attr: SGID attribute to consider during initialization.
1246  *
1247  * When ib_init_ah_attr_from_path() returns success,
1248  * (a) for IB link layer it optionally contains a reference to SGID attribute
1249  * when GRH is present for IB link layer.
1250  * (b) for RoCE link layer it contains a reference to SGID attribute.
1251  * User must invoke rdma_destroy_ah_attr() to release reference to SGID
1252  * attributes which are initialized using ib_init_ah_attr_from_path().
1253  */
1254 int ib_init_ah_attr_from_path(struct ib_device *device, u32 port_num,
1255                               struct sa_path_rec *rec,
1256                               struct rdma_ah_attr *ah_attr,
1257                               const struct ib_gid_attr *gid_attr)
1258 {
1259         int ret = 0;
1260
1261         memset(ah_attr, 0, sizeof(*ah_attr));
1262         ah_attr->type = rdma_ah_find_type(device, port_num);
1263         rdma_ah_set_sl(ah_attr, rec->sl);
1264         rdma_ah_set_port_num(ah_attr, port_num);
1265         rdma_ah_set_static_rate(ah_attr, rec->rate);
1266
1267         if (sa_path_is_roce(rec)) {
1268                 ret = roce_resolve_route_from_path(rec, gid_attr);
1269                 if (ret)
1270                         return ret;
1271
1272                 memcpy(ah_attr->roce.dmac, sa_path_get_dmac(rec), ETH_ALEN);
1273         } else {
1274                 rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1275                 if (sa_path_is_opa(rec) &&
1276                     rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))
1277                         rdma_ah_set_make_grd(ah_attr, true);
1278
1279                 rdma_ah_set_path_bits(ah_attr,
1280                                       be32_to_cpu(sa_path_get_slid(rec)) &
1281                                       get_src_path_mask(device, port_num));
1282         }
1283
1284         if (rec->hop_limit > 0 || sa_path_is_roce(rec))
1285                 ret = init_ah_attr_grh_fields(device, port_num,
1286                                               rec, ah_attr, gid_attr);
1287         return ret;
1288 }
1289 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1290
1291 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1292 {
1293         struct rdma_ah_attr ah_attr;
1294         unsigned long flags;
1295
1296         spin_lock_irqsave(&query->port->ah_lock, flags);
1297         if (!query->port->sm_ah) {
1298                 spin_unlock_irqrestore(&query->port->ah_lock, flags);
1299                 return -EAGAIN;
1300         }
1301         kref_get(&query->port->sm_ah->ref);
1302         query->sm_ah = query->port->sm_ah;
1303         spin_unlock_irqrestore(&query->port->ah_lock, flags);
1304
1305         /*
1306          * Always check if sm_ah has valid dlid assigned,
1307          * before querying for class port info
1308          */
1309         if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1310             !rdma_is_valid_unicast_lid(&ah_attr)) {
1311                 kref_put(&query->sm_ah->ref, free_sm_ah);
1312                 return -EAGAIN;
1313         }
1314         query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1315                                             query->sm_ah->pkey_index,
1316                                             0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1317                                             gfp_mask,
1318                                             ((query->flags & IB_SA_QUERY_OPA) ?
1319                                              OPA_MGMT_BASE_VERSION :
1320                                              IB_MGMT_BASE_VERSION));
1321         if (IS_ERR(query->mad_buf)) {
1322                 kref_put(&query->sm_ah->ref, free_sm_ah);
1323                 return -ENOMEM;
1324         }
1325
1326         query->mad_buf->ah = query->sm_ah->ah;
1327
1328         return 0;
1329 }
1330
1331 static void free_mad(struct ib_sa_query *query)
1332 {
1333         ib_free_send_mad(query->mad_buf);
1334         kref_put(&query->sm_ah->ref, free_sm_ah);
1335 }
1336
1337 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1338 {
1339         struct ib_sa_mad *mad = query->mad_buf->mad;
1340         unsigned long flags;
1341
1342         memset(mad, 0, sizeof *mad);
1343
1344         if (query->flags & IB_SA_QUERY_OPA) {
1345                 mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1346                 mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1347         } else {
1348                 mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1349                 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1350         }
1351         mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1352         spin_lock_irqsave(&tid_lock, flags);
1353         mad->mad_hdr.tid           =
1354                 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1355         spin_unlock_irqrestore(&tid_lock, flags);
1356 }
1357
1358 static int send_mad(struct ib_sa_query *query, unsigned long timeout_ms,
1359                     gfp_t gfp_mask)
1360 {
1361         unsigned long flags;
1362         int ret, id;
1363
1364         xa_lock_irqsave(&queries, flags);
1365         ret = __xa_alloc(&queries, &id, query, xa_limit_32b, gfp_mask);
1366         xa_unlock_irqrestore(&queries, flags);
1367         if (ret < 0)
1368                 return ret;
1369
1370         query->mad_buf->timeout_ms  = timeout_ms;
1371         query->mad_buf->context[0] = query;
1372         query->id = id;
1373
1374         if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1375             (!(query->flags & IB_SA_QUERY_OPA))) {
1376                 if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1377                         if (!ib_nl_make_request(query, gfp_mask))
1378                                 return id;
1379                 }
1380                 ib_sa_disable_local_svc(query);
1381         }
1382
1383         ret = ib_post_send_mad(query->mad_buf, NULL);
1384         if (ret) {
1385                 xa_lock_irqsave(&queries, flags);
1386                 __xa_erase(&queries, id);
1387                 xa_unlock_irqrestore(&queries, flags);
1388         }
1389
1390         /*
1391          * It's not safe to dereference query any more, because the
1392          * send may already have completed and freed the query in
1393          * another context.
1394          */
1395         return ret ? ret : id;
1396 }
1397
1398 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1399 {
1400         ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1401 }
1402 EXPORT_SYMBOL(ib_sa_unpack_path);
1403
1404 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1405 {
1406         ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1407 }
1408 EXPORT_SYMBOL(ib_sa_pack_path);
1409
1410 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1411                                          struct ib_sa_device *sa_dev,
1412                                          u32 port_num)
1413 {
1414         struct ib_sa_port *port;
1415         unsigned long flags;
1416         bool ret = false;
1417
1418         port = &sa_dev->port[port_num - sa_dev->start_port];
1419         spin_lock_irqsave(&port->classport_lock, flags);
1420         if (!port->classport_info.valid)
1421                 goto ret;
1422
1423         if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1424                 ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1425                         OPA_CLASS_PORT_INFO_PR_SUPPORT;
1426 ret:
1427         spin_unlock_irqrestore(&port->classport_lock, flags);
1428         return ret;
1429 }
1430
1431 enum opa_pr_supported {
1432         PR_NOT_SUPPORTED,
1433         PR_OPA_SUPPORTED,
1434         PR_IB_SUPPORTED
1435 };
1436
1437 /*
1438  * opa_pr_query_possible - Check if current PR query can be an OPA query.
1439  *
1440  * Retuns PR_NOT_SUPPORTED if a path record query is not
1441  * possible, PR_OPA_SUPPORTED if an OPA path record query
1442  * is possible and PR_IB_SUPPORTED if an IB path record
1443  * query is possible.
1444  */
1445 static int opa_pr_query_possible(struct ib_sa_client *client,
1446                                  struct ib_sa_device *sa_dev,
1447                                  struct ib_device *device, u32 port_num,
1448                                  struct sa_path_rec *rec)
1449 {
1450         struct ib_port_attr port_attr;
1451
1452         if (ib_query_port(device, port_num, &port_attr))
1453                 return PR_NOT_SUPPORTED;
1454
1455         if (ib_sa_opa_pathrecord_support(client, sa_dev, port_num))
1456                 return PR_OPA_SUPPORTED;
1457
1458         if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1459                 return PR_NOT_SUPPORTED;
1460         else
1461                 return PR_IB_SUPPORTED;
1462 }
1463
1464 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1465                                     int status,
1466                                     struct ib_sa_mad *mad)
1467 {
1468         struct ib_sa_path_query *query =
1469                 container_of(sa_query, struct ib_sa_path_query, sa_query);
1470
1471         if (mad) {
1472                 struct sa_path_rec rec;
1473
1474                 if (sa_query->flags & IB_SA_QUERY_OPA) {
1475                         ib_unpack(opa_path_rec_table,
1476                                   ARRAY_SIZE(opa_path_rec_table),
1477                                   mad->data, &rec);
1478                         rec.rec_type = SA_PATH_REC_TYPE_OPA;
1479                         query->callback(status, &rec, query->context);
1480                 } else {
1481                         ib_unpack(path_rec_table,
1482                                   ARRAY_SIZE(path_rec_table),
1483                                   mad->data, &rec);
1484                         rec.rec_type = SA_PATH_REC_TYPE_IB;
1485                         sa_path_set_dmac_zero(&rec);
1486
1487                         if (query->conv_pr) {
1488                                 struct sa_path_rec opa;
1489
1490                                 memset(&opa, 0, sizeof(struct sa_path_rec));
1491                                 sa_convert_path_ib_to_opa(&opa, &rec);
1492                                 query->callback(status, &opa, query->context);
1493                         } else {
1494                                 query->callback(status, &rec, query->context);
1495                         }
1496                 }
1497         } else
1498                 query->callback(status, NULL, query->context);
1499 }
1500
1501 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1502 {
1503         struct ib_sa_path_query *query =
1504                 container_of(sa_query, struct ib_sa_path_query, sa_query);
1505
1506         kfree(query->conv_pr);
1507         kfree(query);
1508 }
1509
1510 /**
1511  * ib_sa_path_rec_get - Start a Path get query
1512  * @client:SA client
1513  * @device:device to send query on
1514  * @port_num: port number to send query on
1515  * @rec:Path Record to send in query
1516  * @comp_mask:component mask to send in query
1517  * @timeout_ms:time to wait for response
1518  * @gfp_mask:GFP mask to use for internal allocations
1519  * @callback:function called when query completes, times out or is
1520  * canceled
1521  * @context:opaque user context passed to callback
1522  * @sa_query:query context, used to cancel query
1523  *
1524  * Send a Path Record Get query to the SA to look up a path.  The
1525  * callback function will be called when the query completes (or
1526  * fails); status is 0 for a successful response, -EINTR if the query
1527  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1528  * occurred sending the query.  The resp parameter of the callback is
1529  * only valid if status is 0.
1530  *
1531  * If the return value of ib_sa_path_rec_get() is negative, it is an
1532  * error code.  Otherwise it is a query ID that can be used to cancel
1533  * the query.
1534  */
1535 int ib_sa_path_rec_get(struct ib_sa_client *client,
1536                        struct ib_device *device, u32 port_num,
1537                        struct sa_path_rec *rec,
1538                        ib_sa_comp_mask comp_mask,
1539                        unsigned long timeout_ms, gfp_t gfp_mask,
1540                        void (*callback)(int status,
1541                                         struct sa_path_rec *resp,
1542                                         void *context),
1543                        void *context,
1544                        struct ib_sa_query **sa_query)
1545 {
1546         struct ib_sa_path_query *query;
1547         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1548         struct ib_sa_port   *port;
1549         struct ib_mad_agent *agent;
1550         struct ib_sa_mad *mad;
1551         enum opa_pr_supported status;
1552         int ret;
1553
1554         if (!sa_dev)
1555                 return -ENODEV;
1556
1557         if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1558             (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1559                 return -EINVAL;
1560
1561         port  = &sa_dev->port[port_num - sa_dev->start_port];
1562         agent = port->agent;
1563
1564         query = kzalloc(sizeof(*query), gfp_mask);
1565         if (!query)
1566                 return -ENOMEM;
1567
1568         query->sa_query.port     = port;
1569         if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1570                 status = opa_pr_query_possible(client, sa_dev, device, port_num,
1571                                                rec);
1572                 if (status == PR_NOT_SUPPORTED) {
1573                         ret = -EINVAL;
1574                         goto err1;
1575                 } else if (status == PR_OPA_SUPPORTED) {
1576                         query->sa_query.flags |= IB_SA_QUERY_OPA;
1577                 } else {
1578                         query->conv_pr =
1579                                 kmalloc(sizeof(*query->conv_pr), gfp_mask);
1580                         if (!query->conv_pr) {
1581                                 ret = -ENOMEM;
1582                                 goto err1;
1583                         }
1584                 }
1585         }
1586
1587         ret = alloc_mad(&query->sa_query, gfp_mask);
1588         if (ret)
1589                 goto err2;
1590
1591         ib_sa_client_get(client);
1592         query->sa_query.client = client;
1593         query->callback        = callback;
1594         query->context         = context;
1595
1596         mad = query->sa_query.mad_buf->mad;
1597         init_mad(&query->sa_query, agent);
1598
1599         query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1600         query->sa_query.release  = ib_sa_path_rec_release;
1601         mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1602         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1603         mad->sa_hdr.comp_mask    = comp_mask;
1604
1605         if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1606                 ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1607                         rec, mad->data);
1608         } else if (query->conv_pr) {
1609                 sa_convert_path_opa_to_ib(query->conv_pr, rec);
1610                 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1611                         query->conv_pr, mad->data);
1612         } else {
1613                 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1614                         rec, mad->data);
1615         }
1616
1617         *sa_query = &query->sa_query;
1618
1619         query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1620         query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1621                                                 query->conv_pr : rec;
1622
1623         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1624         if (ret < 0)
1625                 goto err3;
1626
1627         return ret;
1628
1629 err3:
1630         *sa_query = NULL;
1631         ib_sa_client_put(query->sa_query.client);
1632         free_mad(&query->sa_query);
1633 err2:
1634         kfree(query->conv_pr);
1635 err1:
1636         kfree(query);
1637         return ret;
1638 }
1639 EXPORT_SYMBOL(ib_sa_path_rec_get);
1640
1641 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1642                                     int status,
1643                                     struct ib_sa_mad *mad)
1644 {
1645         struct ib_sa_service_query *query =
1646                 container_of(sa_query, struct ib_sa_service_query, sa_query);
1647
1648         if (mad) {
1649                 struct ib_sa_service_rec rec;
1650
1651                 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1652                           mad->data, &rec);
1653                 query->callback(status, &rec, query->context);
1654         } else
1655                 query->callback(status, NULL, query->context);
1656 }
1657
1658 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1659 {
1660         kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1661 }
1662
1663 /**
1664  * ib_sa_service_rec_query - Start Service Record operation
1665  * @client:SA client
1666  * @device:device to send request on
1667  * @port_num: port number to send request on
1668  * @method:SA method - should be get, set, or delete
1669  * @rec:Service Record to send in request
1670  * @comp_mask:component mask to send in request
1671  * @timeout_ms:time to wait for response
1672  * @gfp_mask:GFP mask to use for internal allocations
1673  * @callback:function called when request completes, times out or is
1674  * canceled
1675  * @context:opaque user context passed to callback
1676  * @sa_query:request context, used to cancel request
1677  *
1678  * Send a Service Record set/get/delete to the SA to register,
1679  * unregister or query a service record.
1680  * The callback function will be called when the request completes (or
1681  * fails); status is 0 for a successful response, -EINTR if the query
1682  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1683  * occurred sending the query.  The resp parameter of the callback is
1684  * only valid if status is 0.
1685  *
1686  * If the return value of ib_sa_service_rec_query() is negative, it is an
1687  * error code.  Otherwise it is a request ID that can be used to cancel
1688  * the query.
1689  */
1690 int ib_sa_service_rec_query(struct ib_sa_client *client,
1691                             struct ib_device *device, u32 port_num, u8 method,
1692                             struct ib_sa_service_rec *rec,
1693                             ib_sa_comp_mask comp_mask,
1694                             unsigned long timeout_ms, gfp_t gfp_mask,
1695                             void (*callback)(int status,
1696                                              struct ib_sa_service_rec *resp,
1697                                              void *context),
1698                             void *context,
1699                             struct ib_sa_query **sa_query)
1700 {
1701         struct ib_sa_service_query *query;
1702         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1703         struct ib_sa_port   *port;
1704         struct ib_mad_agent *agent;
1705         struct ib_sa_mad *mad;
1706         int ret;
1707
1708         if (!sa_dev)
1709                 return -ENODEV;
1710
1711         port  = &sa_dev->port[port_num - sa_dev->start_port];
1712         agent = port->agent;
1713
1714         if (method != IB_MGMT_METHOD_GET &&
1715             method != IB_MGMT_METHOD_SET &&
1716             method != IB_SA_METHOD_DELETE)
1717                 return -EINVAL;
1718
1719         query = kzalloc(sizeof(*query), gfp_mask);
1720         if (!query)
1721                 return -ENOMEM;
1722
1723         query->sa_query.port     = port;
1724         ret = alloc_mad(&query->sa_query, gfp_mask);
1725         if (ret)
1726                 goto err1;
1727
1728         ib_sa_client_get(client);
1729         query->sa_query.client = client;
1730         query->callback        = callback;
1731         query->context         = context;
1732
1733         mad = query->sa_query.mad_buf->mad;
1734         init_mad(&query->sa_query, agent);
1735
1736         query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1737         query->sa_query.release  = ib_sa_service_rec_release;
1738         mad->mad_hdr.method      = method;
1739         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1740         mad->sa_hdr.comp_mask    = comp_mask;
1741
1742         ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1743                 rec, mad->data);
1744
1745         *sa_query = &query->sa_query;
1746
1747         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1748         if (ret < 0)
1749                 goto err2;
1750
1751         return ret;
1752
1753 err2:
1754         *sa_query = NULL;
1755         ib_sa_client_put(query->sa_query.client);
1756         free_mad(&query->sa_query);
1757
1758 err1:
1759         kfree(query);
1760         return ret;
1761 }
1762 EXPORT_SYMBOL(ib_sa_service_rec_query);
1763
1764 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1765                                         int status,
1766                                         struct ib_sa_mad *mad)
1767 {
1768         struct ib_sa_mcmember_query *query =
1769                 container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1770
1771         if (mad) {
1772                 struct ib_sa_mcmember_rec rec;
1773
1774                 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1775                           mad->data, &rec);
1776                 query->callback(status, &rec, query->context);
1777         } else
1778                 query->callback(status, NULL, query->context);
1779 }
1780
1781 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1782 {
1783         kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1784 }
1785
1786 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1787                              struct ib_device *device, u32 port_num,
1788                              u8 method,
1789                              struct ib_sa_mcmember_rec *rec,
1790                              ib_sa_comp_mask comp_mask,
1791                              unsigned long timeout_ms, gfp_t gfp_mask,
1792                              void (*callback)(int status,
1793                                               struct ib_sa_mcmember_rec *resp,
1794                                               void *context),
1795                              void *context,
1796                              struct ib_sa_query **sa_query)
1797 {
1798         struct ib_sa_mcmember_query *query;
1799         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1800         struct ib_sa_port   *port;
1801         struct ib_mad_agent *agent;
1802         struct ib_sa_mad *mad;
1803         int ret;
1804
1805         if (!sa_dev)
1806                 return -ENODEV;
1807
1808         port  = &sa_dev->port[port_num - sa_dev->start_port];
1809         agent = port->agent;
1810
1811         query = kzalloc(sizeof(*query), gfp_mask);
1812         if (!query)
1813                 return -ENOMEM;
1814
1815         query->sa_query.port     = port;
1816         ret = alloc_mad(&query->sa_query, gfp_mask);
1817         if (ret)
1818                 goto err1;
1819
1820         ib_sa_client_get(client);
1821         query->sa_query.client = client;
1822         query->callback        = callback;
1823         query->context         = context;
1824
1825         mad = query->sa_query.mad_buf->mad;
1826         init_mad(&query->sa_query, agent);
1827
1828         query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1829         query->sa_query.release  = ib_sa_mcmember_rec_release;
1830         mad->mad_hdr.method      = method;
1831         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1832         mad->sa_hdr.comp_mask    = comp_mask;
1833
1834         ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1835                 rec, mad->data);
1836
1837         *sa_query = &query->sa_query;
1838
1839         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1840         if (ret < 0)
1841                 goto err2;
1842
1843         return ret;
1844
1845 err2:
1846         *sa_query = NULL;
1847         ib_sa_client_put(query->sa_query.client);
1848         free_mad(&query->sa_query);
1849
1850 err1:
1851         kfree(query);
1852         return ret;
1853 }
1854
1855 /* Support GuidInfoRecord */
1856 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1857                                         int status,
1858                                         struct ib_sa_mad *mad)
1859 {
1860         struct ib_sa_guidinfo_query *query =
1861                 container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1862
1863         if (mad) {
1864                 struct ib_sa_guidinfo_rec rec;
1865
1866                 ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1867                           mad->data, &rec);
1868                 query->callback(status, &rec, query->context);
1869         } else
1870                 query->callback(status, NULL, query->context);
1871 }
1872
1873 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1874 {
1875         kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1876 }
1877
1878 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1879                               struct ib_device *device, u32 port_num,
1880                               struct ib_sa_guidinfo_rec *rec,
1881                               ib_sa_comp_mask comp_mask, u8 method,
1882                               unsigned long timeout_ms, gfp_t gfp_mask,
1883                               void (*callback)(int status,
1884                                                struct ib_sa_guidinfo_rec *resp,
1885                                                void *context),
1886                               void *context,
1887                               struct ib_sa_query **sa_query)
1888 {
1889         struct ib_sa_guidinfo_query *query;
1890         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1891         struct ib_sa_port *port;
1892         struct ib_mad_agent *agent;
1893         struct ib_sa_mad *mad;
1894         int ret;
1895
1896         if (!sa_dev)
1897                 return -ENODEV;
1898
1899         if (method != IB_MGMT_METHOD_GET &&
1900             method != IB_MGMT_METHOD_SET &&
1901             method != IB_SA_METHOD_DELETE) {
1902                 return -EINVAL;
1903         }
1904
1905         port  = &sa_dev->port[port_num - sa_dev->start_port];
1906         agent = port->agent;
1907
1908         query = kzalloc(sizeof(*query), gfp_mask);
1909         if (!query)
1910                 return -ENOMEM;
1911
1912         query->sa_query.port = port;
1913         ret = alloc_mad(&query->sa_query, gfp_mask);
1914         if (ret)
1915                 goto err1;
1916
1917         ib_sa_client_get(client);
1918         query->sa_query.client = client;
1919         query->callback        = callback;
1920         query->context         = context;
1921
1922         mad = query->sa_query.mad_buf->mad;
1923         init_mad(&query->sa_query, agent);
1924
1925         query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1926         query->sa_query.release  = ib_sa_guidinfo_rec_release;
1927
1928         mad->mad_hdr.method      = method;
1929         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1930         mad->sa_hdr.comp_mask    = comp_mask;
1931
1932         ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1933                 mad->data);
1934
1935         *sa_query = &query->sa_query;
1936
1937         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1938         if (ret < 0)
1939                 goto err2;
1940
1941         return ret;
1942
1943 err2:
1944         *sa_query = NULL;
1945         ib_sa_client_put(query->sa_query.client);
1946         free_mad(&query->sa_query);
1947
1948 err1:
1949         kfree(query);
1950         return ret;
1951 }
1952 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1953
1954 struct ib_classport_info_context {
1955         struct completion       done;
1956         struct ib_sa_query      *sa_query;
1957 };
1958
1959 static void ib_classportinfo_cb(void *context)
1960 {
1961         struct ib_classport_info_context *cb_ctx = context;
1962
1963         complete(&cb_ctx->done);
1964 }
1965
1966 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
1967                                               int status,
1968                                               struct ib_sa_mad *mad)
1969 {
1970         unsigned long flags;
1971         struct ib_sa_classport_info_query *query =
1972                 container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
1973         struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
1974
1975         if (mad) {
1976                 if (sa_query->flags & IB_SA_QUERY_OPA) {
1977                         struct opa_class_port_info rec;
1978
1979                         ib_unpack(opa_classport_info_rec_table,
1980                                   ARRAY_SIZE(opa_classport_info_rec_table),
1981                                   mad->data, &rec);
1982
1983                         spin_lock_irqsave(&sa_query->port->classport_lock,
1984                                           flags);
1985                         if (!status && !info->valid) {
1986                                 memcpy(&info->data.opa, &rec,
1987                                        sizeof(info->data.opa));
1988
1989                                 info->valid = true;
1990                                 info->data.type = RDMA_CLASS_PORT_INFO_OPA;
1991                         }
1992                         spin_unlock_irqrestore(&sa_query->port->classport_lock,
1993                                                flags);
1994
1995                 } else {
1996                         struct ib_class_port_info rec;
1997
1998                         ib_unpack(ib_classport_info_rec_table,
1999                                   ARRAY_SIZE(ib_classport_info_rec_table),
2000                                   mad->data, &rec);
2001
2002                         spin_lock_irqsave(&sa_query->port->classport_lock,
2003                                           flags);
2004                         if (!status && !info->valid) {
2005                                 memcpy(&info->data.ib, &rec,
2006                                        sizeof(info->data.ib));
2007
2008                                 info->valid = true;
2009                                 info->data.type = RDMA_CLASS_PORT_INFO_IB;
2010                         }
2011                         spin_unlock_irqrestore(&sa_query->port->classport_lock,
2012                                                flags);
2013                 }
2014         }
2015         query->callback(query->context);
2016 }
2017
2018 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2019 {
2020         kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2021                            sa_query));
2022 }
2023
2024 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2025                                           unsigned long timeout_ms,
2026                                           void (*callback)(void *context),
2027                                           void *context,
2028                                           struct ib_sa_query **sa_query)
2029 {
2030         struct ib_mad_agent *agent;
2031         struct ib_sa_classport_info_query *query;
2032         struct ib_sa_mad *mad;
2033         gfp_t gfp_mask = GFP_KERNEL;
2034         int ret;
2035
2036         agent = port->agent;
2037
2038         query = kzalloc(sizeof(*query), gfp_mask);
2039         if (!query)
2040                 return -ENOMEM;
2041
2042         query->sa_query.port = port;
2043         query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2044                                                  port->port_num) ?
2045                                  IB_SA_QUERY_OPA : 0;
2046         ret = alloc_mad(&query->sa_query, gfp_mask);
2047         if (ret)
2048                 goto err_free;
2049
2050         query->callback = callback;
2051         query->context = context;
2052
2053         mad = query->sa_query.mad_buf->mad;
2054         init_mad(&query->sa_query, agent);
2055
2056         query->sa_query.callback = ib_sa_classport_info_rec_callback;
2057         query->sa_query.release  = ib_sa_classport_info_rec_release;
2058         mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
2059         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2060         mad->sa_hdr.comp_mask    = 0;
2061         *sa_query = &query->sa_query;
2062
2063         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2064         if (ret < 0)
2065                 goto err_free_mad;
2066
2067         return ret;
2068
2069 err_free_mad:
2070         *sa_query = NULL;
2071         free_mad(&query->sa_query);
2072
2073 err_free:
2074         kfree(query);
2075         return ret;
2076 }
2077
2078 static void update_ib_cpi(struct work_struct *work)
2079 {
2080         struct ib_sa_port *port =
2081                 container_of(work, struct ib_sa_port, ib_cpi_work.work);
2082         struct ib_classport_info_context *cb_context;
2083         unsigned long flags;
2084         int ret;
2085
2086         /* If the classport info is valid, nothing
2087          * to do here.
2088          */
2089         spin_lock_irqsave(&port->classport_lock, flags);
2090         if (port->classport_info.valid) {
2091                 spin_unlock_irqrestore(&port->classport_lock, flags);
2092                 return;
2093         }
2094         spin_unlock_irqrestore(&port->classport_lock, flags);
2095
2096         cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2097         if (!cb_context)
2098                 goto err_nomem;
2099
2100         init_completion(&cb_context->done);
2101
2102         ret = ib_sa_classport_info_rec_query(port, 3000,
2103                                              ib_classportinfo_cb, cb_context,
2104                                              &cb_context->sa_query);
2105         if (ret < 0)
2106                 goto free_cb_err;
2107         wait_for_completion(&cb_context->done);
2108 free_cb_err:
2109         kfree(cb_context);
2110         spin_lock_irqsave(&port->classport_lock, flags);
2111
2112         /* If the classport info is still not valid, the query should have
2113          * failed for some reason. Retry issuing the query
2114          */
2115         if (!port->classport_info.valid) {
2116                 port->classport_info.retry_cnt++;
2117                 if (port->classport_info.retry_cnt <=
2118                     IB_SA_CPI_MAX_RETRY_CNT) {
2119                         unsigned long delay =
2120                                 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2121
2122                         queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2123                 }
2124         }
2125         spin_unlock_irqrestore(&port->classport_lock, flags);
2126
2127 err_nomem:
2128         return;
2129 }
2130
2131 static void send_handler(struct ib_mad_agent *agent,
2132                          struct ib_mad_send_wc *mad_send_wc)
2133 {
2134         struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2135         unsigned long flags;
2136
2137         if (query->callback)
2138                 switch (mad_send_wc->status) {
2139                 case IB_WC_SUCCESS:
2140                         /* No callback -- already got recv */
2141                         break;
2142                 case IB_WC_RESP_TIMEOUT_ERR:
2143                         query->callback(query, -ETIMEDOUT, NULL);
2144                         break;
2145                 case IB_WC_WR_FLUSH_ERR:
2146                         query->callback(query, -EINTR, NULL);
2147                         break;
2148                 default:
2149                         query->callback(query, -EIO, NULL);
2150                         break;
2151                 }
2152
2153         xa_lock_irqsave(&queries, flags);
2154         __xa_erase(&queries, query->id);
2155         xa_unlock_irqrestore(&queries, flags);
2156
2157         free_mad(query);
2158         if (query->client)
2159                 ib_sa_client_put(query->client);
2160         query->release(query);
2161 }
2162
2163 static void recv_handler(struct ib_mad_agent *mad_agent,
2164                          struct ib_mad_send_buf *send_buf,
2165                          struct ib_mad_recv_wc *mad_recv_wc)
2166 {
2167         struct ib_sa_query *query;
2168
2169         if (!send_buf)
2170                 return;
2171
2172         query = send_buf->context[0];
2173         if (query->callback) {
2174                 if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2175                         query->callback(query,
2176                                         mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2177                                         -EINVAL : 0,
2178                                         (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2179                 else
2180                         query->callback(query, -EIO, NULL);
2181         }
2182
2183         ib_free_recv_mad(mad_recv_wc);
2184 }
2185
2186 static void update_sm_ah(struct work_struct *work)
2187 {
2188         struct ib_sa_port *port =
2189                 container_of(work, struct ib_sa_port, update_task);
2190         struct ib_sa_sm_ah *new_ah;
2191         struct ib_port_attr port_attr;
2192         struct rdma_ah_attr   ah_attr;
2193         bool grh_required;
2194
2195         if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2196                 pr_warn("Couldn't query port\n");
2197                 return;
2198         }
2199
2200         new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2201         if (!new_ah)
2202                 return;
2203
2204         kref_init(&new_ah->ref);
2205         new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2206
2207         new_ah->pkey_index = 0;
2208         if (ib_find_pkey(port->agent->device, port->port_num,
2209                          IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2210                 pr_err("Couldn't find index for default PKey\n");
2211
2212         memset(&ah_attr, 0, sizeof(ah_attr));
2213         ah_attr.type = rdma_ah_find_type(port->agent->device,
2214                                          port->port_num);
2215         rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2216         rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2217         rdma_ah_set_port_num(&ah_attr, port->port_num);
2218
2219         grh_required = rdma_is_grh_required(port->agent->device,
2220                                             port->port_num);
2221
2222         /*
2223          * The OPA sm_lid of 0xFFFF needs special handling so that it can be
2224          * differentiated from a permissive LID of 0xFFFF.  We set the
2225          * grh_required flag here so the SA can program the DGID in the
2226          * address handle appropriately
2227          */
2228         if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA &&
2229             (grh_required ||
2230              port_attr.sm_lid == be16_to_cpu(IB_LID_PERMISSIVE)))
2231                 rdma_ah_set_make_grd(&ah_attr, true);
2232
2233         if (ah_attr.type == RDMA_AH_ATTR_TYPE_IB && grh_required) {
2234                 rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2235                 rdma_ah_set_subnet_prefix(&ah_attr,
2236                                           cpu_to_be64(port_attr.subnet_prefix));
2237                 rdma_ah_set_interface_id(&ah_attr,
2238                                          cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2239         }
2240
2241         new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr,
2242                                     RDMA_CREATE_AH_SLEEPABLE);
2243         if (IS_ERR(new_ah->ah)) {
2244                 pr_warn("Couldn't create new SM AH\n");
2245                 kfree(new_ah);
2246                 return;
2247         }
2248
2249         spin_lock_irq(&port->ah_lock);
2250         if (port->sm_ah)
2251                 kref_put(&port->sm_ah->ref, free_sm_ah);
2252         port->sm_ah = new_ah;
2253         spin_unlock_irq(&port->ah_lock);
2254 }
2255
2256 static void ib_sa_event(struct ib_event_handler *handler,
2257                         struct ib_event *event)
2258 {
2259         if (event->event == IB_EVENT_PORT_ERR    ||
2260             event->event == IB_EVENT_PORT_ACTIVE ||
2261             event->event == IB_EVENT_LID_CHANGE  ||
2262             event->event == IB_EVENT_PKEY_CHANGE ||
2263             event->event == IB_EVENT_SM_CHANGE   ||
2264             event->event == IB_EVENT_CLIENT_REREGISTER) {
2265                 unsigned long flags;
2266                 struct ib_sa_device *sa_dev =
2267                         container_of(handler, typeof(*sa_dev), event_handler);
2268                 u32 port_num = event->element.port_num - sa_dev->start_port;
2269                 struct ib_sa_port *port = &sa_dev->port[port_num];
2270
2271                 if (!rdma_cap_ib_sa(handler->device, port->port_num))
2272                         return;
2273
2274                 spin_lock_irqsave(&port->ah_lock, flags);
2275                 if (port->sm_ah)
2276                         kref_put(&port->sm_ah->ref, free_sm_ah);
2277                 port->sm_ah = NULL;
2278                 spin_unlock_irqrestore(&port->ah_lock, flags);
2279
2280                 if (event->event == IB_EVENT_SM_CHANGE ||
2281                     event->event == IB_EVENT_CLIENT_REREGISTER ||
2282                     event->event == IB_EVENT_LID_CHANGE ||
2283                     event->event == IB_EVENT_PORT_ACTIVE) {
2284                         unsigned long delay =
2285                                 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2286
2287                         spin_lock_irqsave(&port->classport_lock, flags);
2288                         port->classport_info.valid = false;
2289                         port->classport_info.retry_cnt = 0;
2290                         spin_unlock_irqrestore(&port->classport_lock, flags);
2291                         queue_delayed_work(ib_wq,
2292                                            &port->ib_cpi_work, delay);
2293                 }
2294                 queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2295         }
2296 }
2297
2298 static int ib_sa_add_one(struct ib_device *device)
2299 {
2300         struct ib_sa_device *sa_dev;
2301         int s, e, i;
2302         int count = 0;
2303         int ret;
2304
2305         s = rdma_start_port(device);
2306         e = rdma_end_port(device);
2307
2308         sa_dev = kzalloc(struct_size(sa_dev, port, e - s + 1), GFP_KERNEL);
2309         if (!sa_dev)
2310                 return -ENOMEM;
2311
2312         sa_dev->start_port = s;
2313         sa_dev->end_port   = e;
2314
2315         for (i = 0; i <= e - s; ++i) {
2316                 spin_lock_init(&sa_dev->port[i].ah_lock);
2317                 if (!rdma_cap_ib_sa(device, i + 1))
2318                         continue;
2319
2320                 sa_dev->port[i].sm_ah    = NULL;
2321                 sa_dev->port[i].port_num = i + s;
2322
2323                 spin_lock_init(&sa_dev->port[i].classport_lock);
2324                 sa_dev->port[i].classport_info.valid = false;
2325
2326                 sa_dev->port[i].agent =
2327                         ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2328                                               NULL, 0, send_handler,
2329                                               recv_handler, sa_dev, 0);
2330                 if (IS_ERR(sa_dev->port[i].agent)) {
2331                         ret = PTR_ERR(sa_dev->port[i].agent);
2332                         goto err;
2333                 }
2334
2335                 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2336                 INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2337                                   update_ib_cpi);
2338
2339                 count++;
2340         }
2341
2342         if (!count) {
2343                 ret = -EOPNOTSUPP;
2344                 goto free;
2345         }
2346
2347         ib_set_client_data(device, &sa_client, sa_dev);
2348
2349         /*
2350          * We register our event handler after everything is set up,
2351          * and then update our cached info after the event handler is
2352          * registered to avoid any problems if a port changes state
2353          * during our initialization.
2354          */
2355
2356         INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2357         ib_register_event_handler(&sa_dev->event_handler);
2358
2359         for (i = 0; i <= e - s; ++i) {
2360                 if (rdma_cap_ib_sa(device, i + 1))
2361                         update_sm_ah(&sa_dev->port[i].update_task);
2362         }
2363
2364         return 0;
2365
2366 err:
2367         while (--i >= 0) {
2368                 if (rdma_cap_ib_sa(device, i + 1))
2369                         ib_unregister_mad_agent(sa_dev->port[i].agent);
2370         }
2371 free:
2372         kfree(sa_dev);
2373         return ret;
2374 }
2375
2376 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2377 {
2378         struct ib_sa_device *sa_dev = client_data;
2379         int i;
2380
2381         ib_unregister_event_handler(&sa_dev->event_handler);
2382         flush_workqueue(ib_wq);
2383
2384         for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2385                 if (rdma_cap_ib_sa(device, i + 1)) {
2386                         cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2387                         ib_unregister_mad_agent(sa_dev->port[i].agent);
2388                         if (sa_dev->port[i].sm_ah)
2389                                 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2390                 }
2391
2392         }
2393
2394         kfree(sa_dev);
2395 }
2396
2397 int ib_sa_init(void)
2398 {
2399         int ret;
2400
2401         get_random_bytes(&tid, sizeof tid);
2402
2403         atomic_set(&ib_nl_sa_request_seq, 0);
2404
2405         ret = ib_register_client(&sa_client);
2406         if (ret) {
2407                 pr_err("Couldn't register ib_sa client\n");
2408                 goto err1;
2409         }
2410
2411         ret = mcast_init();
2412         if (ret) {
2413                 pr_err("Couldn't initialize multicast handling\n");
2414                 goto err2;
2415         }
2416
2417         ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2418         if (!ib_nl_wq) {
2419                 ret = -ENOMEM;
2420                 goto err3;
2421         }
2422
2423         INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2424
2425         return 0;
2426
2427 err3:
2428         mcast_cleanup();
2429 err2:
2430         ib_unregister_client(&sa_client);
2431 err1:
2432         return ret;
2433 }
2434
2435 void ib_sa_cleanup(void)
2436 {
2437         cancel_delayed_work(&ib_nl_timed_work);
2438         flush_workqueue(ib_nl_wq);
2439         destroy_workqueue(ib_nl_wq);
2440         mcast_cleanup();
2441         ib_unregister_client(&sa_client);
2442         WARN_ON(!xa_empty(&queries));
2443 }
This page took 0.179185 seconds and 4 git commands to generate.