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/idr.h>
  44 #include <linux/workqueue.h>
  45 #include <uapi/linux/if_ether.h>
  46 #include <rdma/ib_pack.h>
  47 #include <rdma/ib_cache.h>
  48 #include <rdma/rdma_netlink.h>
  49 #include <net/netlink.h>
  50 #include <uapi/rdma/ib_user_sa.h>
  51 #include <rdma/ib_marshall.h>
  52 #include <rdma/ib_addr.h>
  53 #include "sa.h"
  54 #include "core_priv.h"
  55
  56 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN             100
  57 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT         2000
  58 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX             200000
  59 #define IB_SA_CPI_MAX_RETRY_CNT                 3
  60 #define IB_SA_CPI_RETRY_WAIT                    1000 /*msecs */
  61 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
  62
  63 struct ib_sa_sm_ah {
  64         struct ib_ah        *ah;
  65         struct kref          ref;
  66         u16                  pkey_index;
  67         u8                   src_path_mask;
  68 };
  69
  70 enum rdma_class_port_info_type {
  71         RDMA_CLASS_PORT_INFO_IB,
  72         RDMA_CLASS_PORT_INFO_OPA
  73 };
  74
  75 struct rdma_class_port_info {
  76         enum rdma_class_port_info_type type;
  77         union {
  78                 struct ib_class_port_info ib;
  79                 struct opa_class_port_info opa;
  80         };
  81 };
  82
  83 struct ib_sa_classport_cache {
  84         bool valid;
  85         int retry_cnt;
  86         struct rdma_class_port_info data;
  87 };
  88
  89 struct ib_sa_port {
  90         struct ib_mad_agent *agent;
  91         struct ib_sa_sm_ah  *sm_ah;
  92         struct work_struct   update_task;
  93         struct ib_sa_classport_cache classport_info;
  94         struct delayed_work ib_cpi_work;
  95         spinlock_t                   classport_lock; /* protects class port info set */
  96         spinlock_t           ah_lock;
  97         u8                   port_num;
  98 };
  99
 100 struct ib_sa_device {
 101         int                     start_port, end_port;
 102         struct ib_event_handler event_handler;
 103         struct ib_sa_port port[0];
 104 };
 105
 106 struct ib_sa_query {
 107         void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
 108         void (*release)(struct ib_sa_query *);
 109         struct ib_sa_client    *client;
 110         struct ib_sa_port      *port;
 111         struct ib_mad_send_buf *mad_buf;
 112         struct ib_sa_sm_ah     *sm_ah;
 113         int                     id;
 114         u32                     flags;
 115         struct list_head        list; /* Local svc request list */
 116         u32                     seq; /* Local svc request sequence number */
 117         unsigned long           timeout; /* Local svc timeout */
 118         u8                      path_use; /* How will the pathrecord be used */
 119 };
 120
 121 #define IB_SA_ENABLE_LOCAL_SERVICE      0x00000001
 122 #define IB_SA_CANCEL                    0x00000002
 123 #define IB_SA_QUERY_OPA                 0x00000004
 124
 125 struct ib_sa_service_query {
 126         void (*callback)(int, struct ib_sa_service_rec *, void *);
 127         void *context;
 128         struct ib_sa_query sa_query;
 129 };
 130
 131 struct ib_sa_path_query {
 132         void (*callback)(int, struct ib_sa_path_rec *, void *);
 133         void *context;
 134         struct ib_sa_query sa_query;
 135 };
 136
 137 struct ib_sa_guidinfo_query {
 138         void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
 139         void *context;
 140         struct ib_sa_query sa_query;
 141 };
 142
 143 struct ib_sa_classport_info_query {
 144         void (*callback)(void *);
 145         void *context;
 146         struct ib_sa_query sa_query;
 147 };
 148
 149 struct ib_sa_mcmember_query {
 150         void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
 151         void *context;
 152         struct ib_sa_query sa_query;
 153 };
 154
 155 static LIST_HEAD(ib_nl_request_list);
 156 static DEFINE_SPINLOCK(ib_nl_request_lock);
 157 static atomic_t ib_nl_sa_request_seq;
 158 static struct workqueue_struct *ib_nl_wq;
 159 static struct delayed_work ib_nl_timed_work;
 160 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
 161         [LS_NLA_TYPE_PATH_RECORD]       = {.type = NLA_BINARY,
 162                 .len = sizeof(struct ib_path_rec_data)},
 163         [LS_NLA_TYPE_TIMEOUT]           = {.type = NLA_U32},
 164         [LS_NLA_TYPE_SERVICE_ID]        = {.type = NLA_U64},
 165         [LS_NLA_TYPE_DGID]              = {.type = NLA_BINARY,
 166                 .len = sizeof(struct rdma_nla_ls_gid)},
 167         [LS_NLA_TYPE_SGID]              = {.type = NLA_BINARY,
 168                 .len = sizeof(struct rdma_nla_ls_gid)},
 169         [LS_NLA_TYPE_TCLASS]            = {.type = NLA_U8},
 170         [LS_NLA_TYPE_PKEY]              = {.type = NLA_U16},
 171         [LS_NLA_TYPE_QOS_CLASS]         = {.type = NLA_U16},
 172 };
 173
 174
 175 static void ib_sa_add_one(struct ib_device *device);
 176 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
 177
 178 static struct ib_client sa_client = {
 179         .name   = "sa",
 180         .add    = ib_sa_add_one,
 181         .remove = ib_sa_remove_one
 182 };
 183
 184 static DEFINE_SPINLOCK(idr_lock);
 185 static DEFINE_IDR(query_idr);
 186
 187 static DEFINE_SPINLOCK(tid_lock);
 188 static u32 tid;
 189
 190 #define PATH_REC_FIELD(field) \
 191         .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),          \
 192         .struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,     \
 193         .field_name          = "sa_path_rec:" #field
 194
 195 static const struct ib_field path_rec_table[] = {
 196         { PATH_REC_FIELD(service_id),
 197           .offset_words = 0,
 198           .offset_bits  = 0,
 199           .size_bits    = 64 },
 200         { PATH_REC_FIELD(dgid),
 201           .offset_words = 2,
 202           .offset_bits  = 0,
 203           .size_bits    = 128 },
 204         { PATH_REC_FIELD(sgid),
 205           .offset_words = 6,
 206           .offset_bits  = 0,
 207           .size_bits    = 128 },
 208         { PATH_REC_FIELD(dlid),
 209           .offset_words = 10,
 210           .offset_bits  = 0,
 211           .size_bits    = 16 },
 212         { PATH_REC_FIELD(slid),
 213           .offset_words = 10,
 214           .offset_bits  = 16,
 215           .size_bits    = 16 },
 216         { PATH_REC_FIELD(raw_traffic),
 217           .offset_words = 11,
 218           .offset_bits  = 0,
 219           .size_bits    = 1 },
 220         { RESERVED,
 221           .offset_words = 11,
 222           .offset_bits  = 1,
 223           .size_bits    = 3 },
 224         { PATH_REC_FIELD(flow_label),
 225           .offset_words = 11,
 226           .offset_bits  = 4,
 227           .size_bits    = 20 },
 228         { PATH_REC_FIELD(hop_limit),
 229           .offset_words = 11,
 230           .offset_bits  = 24,
 231           .size_bits    = 8 },
 232         { PATH_REC_FIELD(traffic_class),
 233           .offset_words = 12,
 234           .offset_bits  = 0,
 235           .size_bits    = 8 },
 236         { PATH_REC_FIELD(reversible),
 237           .offset_words = 12,
 238           .offset_bits  = 8,
 239           .size_bits    = 1 },
 240         { PATH_REC_FIELD(numb_path),
 241           .offset_words = 12,
 242           .offset_bits  = 9,
 243           .size_bits    = 7 },
 244         { PATH_REC_FIELD(pkey),
 245           .offset_words = 12,
 246           .offset_bits  = 16,
 247           .size_bits    = 16 },
 248         { PATH_REC_FIELD(qos_class),
 249           .offset_words = 13,
 250           .offset_bits  = 0,
 251           .size_bits    = 12 },
 252         { PATH_REC_FIELD(sl),
 253           .offset_words = 13,
 254           .offset_bits  = 12,
 255           .size_bits    = 4 },
 256         { PATH_REC_FIELD(mtu_selector),
 257           .offset_words = 13,
 258           .offset_bits  = 16,
 259           .size_bits    = 2 },
 260         { PATH_REC_FIELD(mtu),
 261           .offset_words = 13,
 262           .offset_bits  = 18,
 263           .size_bits    = 6 },
 264         { PATH_REC_FIELD(rate_selector),
 265           .offset_words = 13,
 266           .offset_bits  = 24,
 267           .size_bits    = 2 },
 268         { PATH_REC_FIELD(rate),
 269           .offset_words = 13,
 270           .offset_bits  = 26,
 271           .size_bits    = 6 },
 272         { PATH_REC_FIELD(packet_life_time_selector),
 273           .offset_words = 14,
 274           .offset_bits  = 0,
 275           .size_bits    = 2 },
 276         { PATH_REC_FIELD(packet_life_time),
 277           .offset_words = 14,
 278           .offset_bits  = 2,
 279           .size_bits    = 6 },
 280         { PATH_REC_FIELD(preference),
 281           .offset_words = 14,
 282           .offset_bits  = 8,
 283           .size_bits    = 8 },
 284         { RESERVED,
 285           .offset_words = 14,
 286           .offset_bits  = 16,
 287           .size_bits    = 48 },
 288 };
 289
 290 #define MCMEMBER_REC_FIELD(field) \
 291         .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),      \
 292         .struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \
 293         .field_name          = "sa_mcmember_rec:" #field
 294
 295 static const struct ib_field mcmember_rec_table[] = {
 296         { MCMEMBER_REC_FIELD(mgid),
 297           .offset_words = 0,
 298           .offset_bits  = 0,
 299           .size_bits    = 128 },
 300         { MCMEMBER_REC_FIELD(port_gid),
 301           .offset_words = 4,
 302           .offset_bits  = 0,
 303           .size_bits    = 128 },
 304         { MCMEMBER_REC_FIELD(qkey),
 305           .offset_words = 8,
 306           .offset_bits  = 0,
 307           .size_bits    = 32 },
 308         { MCMEMBER_REC_FIELD(mlid),
 309           .offset_words = 9,
 310           .offset_bits  = 0,
 311           .size_bits    = 16 },
 312         { MCMEMBER_REC_FIELD(mtu_selector),
 313           .offset_words = 9,
 314           .offset_bits  = 16,
 315           .size_bits    = 2 },
 316         { MCMEMBER_REC_FIELD(mtu),
 317           .offset_words = 9,
 318           .offset_bits  = 18,
 319           .size_bits    = 6 },
 320         { MCMEMBER_REC_FIELD(traffic_class),
 321           .offset_words = 9,
 322           .offset_bits  = 24,
 323           .size_bits    = 8 },
 324         { MCMEMBER_REC_FIELD(pkey),
 325           .offset_words = 10,
 326           .offset_bits  = 0,
 327           .size_bits    = 16 },
 328         { MCMEMBER_REC_FIELD(rate_selector),
 329           .offset_words = 10,
 330           .offset_bits  = 16,
 331           .size_bits    = 2 },
 332         { MCMEMBER_REC_FIELD(rate),
 333           .offset_words = 10,
 334           .offset_bits  = 18,
 335           .size_bits    = 6 },
 336         { MCMEMBER_REC_FIELD(packet_life_time_selector),
 337           .offset_words = 10,
 338           .offset_bits  = 24,
 339           .size_bits    = 2 },
 340         { MCMEMBER_REC_FIELD(packet_life_time),
 341           .offset_words = 10,
 342           .offset_bits  = 26,
 343           .size_bits    = 6 },
 344         { MCMEMBER_REC_FIELD(sl),
 345           .offset_words = 11,
 346           .offset_bits  = 0,
 347           .size_bits    = 4 },
 348         { MCMEMBER_REC_FIELD(flow_label),
 349           .offset_words = 11,
 350           .offset_bits  = 4,
 351           .size_bits    = 20 },
 352         { MCMEMBER_REC_FIELD(hop_limit),
 353           .offset_words = 11,
 354           .offset_bits  = 24,
 355           .size_bits    = 8 },
 356         { MCMEMBER_REC_FIELD(scope),
 357           .offset_words = 12,
 358           .offset_bits  = 0,
 359           .size_bits    = 4 },
 360         { MCMEMBER_REC_FIELD(join_state),
 361           .offset_words = 12,
 362           .offset_bits  = 4,
 363           .size_bits    = 4 },
 364         { MCMEMBER_REC_FIELD(proxy_join),
 365           .offset_words = 12,
 366           .offset_bits  = 8,
 367           .size_bits    = 1 },
 368         { RESERVED,
 369           .offset_words = 12,
 370           .offset_bits  = 9,
 371           .size_bits    = 23 },
 372 };
 373
 374 #define SERVICE_REC_FIELD(field) \
 375         .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),       \
 376         .struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,  \
 377         .field_name          = "sa_service_rec:" #field
 378
 379 static const struct ib_field service_rec_table[] = {
 380         { SERVICE_REC_FIELD(id),
 381           .offset_words = 0,
 382           .offset_bits  = 0,
 383           .size_bits    = 64 },
 384         { SERVICE_REC_FIELD(gid),
 385           .offset_words = 2,
 386           .offset_bits  = 0,
 387           .size_bits    = 128 },
 388         { SERVICE_REC_FIELD(pkey),
 389           .offset_words = 6,
 390           .offset_bits  = 0,
 391           .size_bits    = 16 },
 392         { SERVICE_REC_FIELD(lease),
 393           .offset_words = 7,
 394           .offset_bits  = 0,
 395           .size_bits    = 32 },
 396         { SERVICE_REC_FIELD(key),
 397           .offset_words = 8,
 398           .offset_bits  = 0,
 399           .size_bits    = 128 },
 400         { SERVICE_REC_FIELD(name),
 401           .offset_words = 12,
 402           .offset_bits  = 0,
 403           .size_bits    = 64*8 },
 404         { SERVICE_REC_FIELD(data8),
 405           .offset_words = 28,
 406           .offset_bits  = 0,
 407           .size_bits    = 16*8 },
 408         { SERVICE_REC_FIELD(data16),
 409           .offset_words = 32,
 410           .offset_bits  = 0,
 411           .size_bits    = 8*16 },
 412         { SERVICE_REC_FIELD(data32),
 413           .offset_words = 36,
 414           .offset_bits  = 0,
 415           .size_bits    = 4*32 },
 416         { SERVICE_REC_FIELD(data64),
 417           .offset_words = 40,
 418           .offset_bits  = 0,
 419           .size_bits    = 2*64 },
 420 };
 421
 422 #define CLASSPORTINFO_REC_FIELD(field) \
 423         .struct_offset_bytes = offsetof(struct ib_class_port_info, field),      \
 424         .struct_size_bytes   = sizeof((struct ib_class_port_info *)0)->field,   \
 425         .field_name          = "ib_class_port_info:" #field
 426
 427 static const struct ib_field ib_classport_info_rec_table[] = {
 428         { CLASSPORTINFO_REC_FIELD(base_version),
 429           .offset_words = 0,
 430           .offset_bits  = 0,
 431           .size_bits    = 8 },
 432         { CLASSPORTINFO_REC_FIELD(class_version),
 433           .offset_words = 0,
 434           .offset_bits  = 8,
 435           .size_bits    = 8 },
 436         { CLASSPORTINFO_REC_FIELD(capability_mask),
 437           .offset_words = 0,
 438           .offset_bits  = 16,
 439           .size_bits    = 16 },
 440         { CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
 441           .offset_words = 1,
 442           .offset_bits  = 0,
 443           .size_bits    = 32 },
 444         { CLASSPORTINFO_REC_FIELD(redirect_gid),
 445           .offset_words = 2,
 446           .offset_bits  = 0,
 447           .size_bits    = 128 },
 448         { CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
 449           .offset_words = 6,
 450           .offset_bits  = 0,
 451           .size_bits    = 32 },
 452         { CLASSPORTINFO_REC_FIELD(redirect_lid),
 453           .offset_words = 7,
 454           .offset_bits  = 0,
 455           .size_bits    = 16 },
 456         { CLASSPORTINFO_REC_FIELD(redirect_pkey),
 457           .offset_words = 7,
 458           .offset_bits  = 16,
 459           .size_bits    = 16 },
 460
 461         { CLASSPORTINFO_REC_FIELD(redirect_qp),
 462           .offset_words = 8,
 463           .offset_bits  = 0,
 464           .size_bits    = 32 },
 465         { CLASSPORTINFO_REC_FIELD(redirect_qkey),
 466           .offset_words = 9,
 467           .offset_bits  = 0,
 468           .size_bits    = 32 },
 469
 470         { CLASSPORTINFO_REC_FIELD(trap_gid),
 471           .offset_words = 10,
 472           .offset_bits  = 0,
 473           .size_bits    = 128 },
 474         { CLASSPORTINFO_REC_FIELD(trap_tcslfl),
 475           .offset_words = 14,
 476           .offset_bits  = 0,
 477           .size_bits    = 32 },
 478
 479         { CLASSPORTINFO_REC_FIELD(trap_lid),
 480           .offset_words = 15,
 481           .offset_bits  = 0,
 482           .size_bits    = 16 },
 483         { CLASSPORTINFO_REC_FIELD(trap_pkey),
 484           .offset_words = 15,
 485           .offset_bits  = 16,
 486           .size_bits    = 16 },
 487
 488         { CLASSPORTINFO_REC_FIELD(trap_hlqp),
 489           .offset_words = 16,
 490           .offset_bits  = 0,
 491           .size_bits    = 32 },
 492         { CLASSPORTINFO_REC_FIELD(trap_qkey),
 493           .offset_words = 17,
 494           .offset_bits  = 0,
 495           .size_bits    = 32 },
 496 };
 497
 498 #define OPA_CLASSPORTINFO_REC_FIELD(field) \
 499         .struct_offset_bytes =\
 500                 offsetof(struct opa_class_port_info, field),    \
 501         .struct_size_bytes   = \
 502                 sizeof((struct opa_class_port_info *)0)->field, \
 503         .field_name          = "opa_class_port_info:" #field
 504
 505 static const struct ib_field opa_classport_info_rec_table[] = {
 506         { OPA_CLASSPORTINFO_REC_FIELD(base_version),
 507           .offset_words = 0,
 508           .offset_bits  = 0,
 509           .size_bits    = 8 },
 510         { OPA_CLASSPORTINFO_REC_FIELD(class_version),
 511           .offset_words = 0,
 512           .offset_bits  = 8,
 513           .size_bits    = 8 },
 514         { OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
 515           .offset_words = 0,
 516           .offset_bits  = 16,
 517           .size_bits    = 16 },
 518         { OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
 519           .offset_words = 1,
 520           .offset_bits  = 0,
 521           .size_bits    = 32 },
 522         { OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
 523           .offset_words = 2,
 524           .offset_bits  = 0,
 525           .size_bits    = 128 },
 526         { OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
 527           .offset_words = 6,
 528           .offset_bits  = 0,
 529           .size_bits    = 32 },
 530         { OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
 531           .offset_words = 7,
 532           .offset_bits  = 0,
 533           .size_bits    = 32 },
 534         { OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
 535           .offset_words = 8,
 536           .offset_bits  = 0,
 537           .size_bits    = 32 },
 538         { OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
 539           .offset_words = 9,
 540           .offset_bits  = 0,
 541           .size_bits    = 32 },
 542         { OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
 543           .offset_words = 10,
 544           .offset_bits  = 0,
 545           .size_bits    = 128 },
 546         { OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
 547           .offset_words = 14,
 548           .offset_bits  = 0,
 549           .size_bits    = 32 },
 550         { OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
 551           .offset_words = 15,
 552           .offset_bits  = 0,
 553           .size_bits    = 32 },
 554         { OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
 555           .offset_words = 16,
 556           .offset_bits  = 0,
 557           .size_bits    = 32 },
 558         { OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
 559           .offset_words = 17,
 560           .offset_bits  = 0,
 561           .size_bits    = 32 },
 562         { OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
 563           .offset_words = 18,
 564           .offset_bits  = 0,
 565           .size_bits    = 16 },
 566         { OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
 567           .offset_words = 18,
 568           .offset_bits  = 16,
 569           .size_bits    = 16 },
 570         { OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
 571           .offset_words = 19,
 572           .offset_bits  = 0,
 573           .size_bits    = 8 },
 574         { RESERVED,
 575           .offset_words = 19,
 576           .offset_bits  = 8,
 577           .size_bits    = 24 },
 578 };
 579
 580 #define GUIDINFO_REC_FIELD(field) \
 581         .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),      \
 582         .struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,  \
 583         .field_name          = "sa_guidinfo_rec:" #field
 584
 585 static const struct ib_field guidinfo_rec_table[] = {
 586         { GUIDINFO_REC_FIELD(lid),
 587           .offset_words = 0,
 588           .offset_bits  = 0,
 589           .size_bits    = 16 },
 590         { GUIDINFO_REC_FIELD(block_num),
 591           .offset_words = 0,
 592           .offset_bits  = 16,
 593           .size_bits    = 8 },
 594         { GUIDINFO_REC_FIELD(res1),
 595           .offset_words = 0,
 596           .offset_bits  = 24,
 597           .size_bits    = 8 },
 598         { GUIDINFO_REC_FIELD(res2),
 599           .offset_words = 1,
 600           .offset_bits  = 0,
 601           .size_bits    = 32 },
 602         { GUIDINFO_REC_FIELD(guid_info_list),
 603           .offset_words = 2,
 604           .offset_bits  = 0,
 605           .size_bits    = 512 },
 606 };
 607
 608 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
 609 {
 610         query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
 611 }
 612
 613 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
 614 {
 615         return (query->flags & IB_SA_CANCEL);
 616 }
 617
 618 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
 619                                      struct ib_sa_query *query)
 620 {
 621         struct ib_sa_path_rec *sa_rec = query->mad_buf->context[1];
 622         struct ib_sa_mad *mad = query->mad_buf->mad;
 623         ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
 624         u16 val16;
 625         u64 val64;
 626         struct rdma_ls_resolve_header *header;
 627
 628         query->mad_buf->context[1] = NULL;
 629
 630         /* Construct the family header first */
 631         header = (struct rdma_ls_resolve_header *)
 632                 skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 633         memcpy(header->device_name, query->port->agent->device->name,
 634                LS_DEVICE_NAME_MAX);
 635         header->port_num = query->port->port_num;
 636
 637         if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
 638             sa_rec->reversible != 0)
 639                 query->path_use = LS_RESOLVE_PATH_USE_GMP;
 640         else
 641                 query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
 642         header->path_use = query->path_use;
 643
 644         /* Now build the attributes */
 645         if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
 646                 val64 = be64_to_cpu(sa_rec->service_id);
 647                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
 648                         sizeof(val64), &val64);
 649         }
 650         if (comp_mask & IB_SA_PATH_REC_DGID)
 651                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
 652                         sizeof(sa_rec->dgid), &sa_rec->dgid);
 653         if (comp_mask & IB_SA_PATH_REC_SGID)
 654                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
 655                         sizeof(sa_rec->sgid), &sa_rec->sgid);
 656         if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
 657                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
 658                         sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
 659
 660         if (comp_mask & IB_SA_PATH_REC_PKEY) {
 661                 val16 = be16_to_cpu(sa_rec->pkey);
 662                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
 663                         sizeof(val16), &val16);
 664         }
 665         if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
 666                 val16 = be16_to_cpu(sa_rec->qos_class);
 667                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
 668                         sizeof(val16), &val16);
 669         }
 670 }
 671
 672 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
 673 {
 674         int len = 0;
 675
 676         if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
 677                 len += nla_total_size(sizeof(u64));
 678         if (comp_mask & IB_SA_PATH_REC_DGID)
 679                 len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
 680         if (comp_mask & IB_SA_PATH_REC_SGID)
 681                 len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
 682         if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
 683                 len += nla_total_size(sizeof(u8));
 684         if (comp_mask & IB_SA_PATH_REC_PKEY)
 685                 len += nla_total_size(sizeof(u16));
 686         if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
 687                 len += nla_total_size(sizeof(u16));
 688
 689         /*
 690          * Make sure that at least some of the required comp_mask bits are
 691          * set.
 692          */
 693         if (WARN_ON(len == 0))
 694                 return len;
 695
 696         /* Add the family header */
 697         len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
 698
 699         return len;
 700 }
 701
 702 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
 703 {
 704         struct sk_buff *skb = NULL;
 705         struct nlmsghdr *nlh;
 706         void *data;
 707         int ret = 0;
 708         struct ib_sa_mad *mad;
 709         int len;
 710
 711         mad = query->mad_buf->mad;
 712         len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
 713         if (len <= 0)
 714                 return -EMSGSIZE;
 715
 716         skb = nlmsg_new(len, gfp_mask);
 717         if (!skb)
 718                 return -ENOMEM;
 719
 720         /* Put nlmsg header only for now */
 721         data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
 722                             RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
 723         if (!data) {
 724                 nlmsg_free(skb);
 725                 return -EMSGSIZE;
 726         }
 727
 728         /* Add attributes */
 729         ib_nl_set_path_rec_attrs(skb, query);
 730
 731         /* Repair the nlmsg header length */
 732         nlmsg_end(skb, nlh);
 733
 734         ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, gfp_mask);
 735         if (!ret)
 736                 ret = len;
 737         else
 738                 ret = 0;
 739
 740         return ret;
 741 }
 742
 743 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
 744 {
 745         unsigned long flags;
 746         unsigned long delay;
 747         int ret;
 748
 749         INIT_LIST_HEAD(&query->list);
 750         query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
 751
 752         /* Put the request on the list first.*/
 753         spin_lock_irqsave(&ib_nl_request_lock, flags);
 754         delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
 755         query->timeout = delay + jiffies;
 756         list_add_tail(&query->list, &ib_nl_request_list);
 757         /* Start the timeout if this is the only request */
 758         if (ib_nl_request_list.next == &query->list)
 759                 queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
 760         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 761
 762         ret = ib_nl_send_msg(query, gfp_mask);
 763         if (ret <= 0) {
 764                 ret = -EIO;
 765                 /* Remove the request */
 766                 spin_lock_irqsave(&ib_nl_request_lock, flags);
 767                 list_del(&query->list);
 768                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 769         } else {
 770                 ret = 0;
 771         }
 772
 773         return ret;
 774 }
 775
 776 static int ib_nl_cancel_request(struct ib_sa_query *query)
 777 {
 778         unsigned long flags;
 779         struct ib_sa_query *wait_query;
 780         int found = 0;
 781
 782         spin_lock_irqsave(&ib_nl_request_lock, flags);
 783         list_for_each_entry(wait_query, &ib_nl_request_list, list) {
 784                 /* Let the timeout to take care of the callback */
 785                 if (query == wait_query) {
 786                         query->flags |= IB_SA_CANCEL;
 787                         query->timeout = jiffies;
 788                         list_move(&query->list, &ib_nl_request_list);
 789                         found = 1;
 790                         mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
 791                         break;
 792                 }
 793         }
 794         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 795
 796         return found;
 797 }
 798
 799 static void send_handler(struct ib_mad_agent *agent,
 800                          struct ib_mad_send_wc *mad_send_wc);
 801
 802 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
 803                                            const struct nlmsghdr *nlh)
 804 {
 805         struct ib_mad_send_wc mad_send_wc;
 806         struct ib_sa_mad *mad = NULL;
 807         const struct nlattr *head, *curr;
 808         struct ib_path_rec_data  *rec;
 809         int len, rem;
 810         u32 mask = 0;
 811         int status = -EIO;
 812
 813         if (query->callback) {
 814                 head = (const struct nlattr *) nlmsg_data(nlh);
 815                 len = nlmsg_len(nlh);
 816                 switch (query->path_use) {
 817                 case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
 818                         mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
 819                         break;
 820
 821                 case LS_RESOLVE_PATH_USE_ALL:
 822                 case LS_RESOLVE_PATH_USE_GMP:
 823                 default:
 824                         mask = IB_PATH_PRIMARY | IB_PATH_GMP |
 825                                 IB_PATH_BIDIRECTIONAL;
 826                         break;
 827                 }
 828                 nla_for_each_attr(curr, head, len, rem) {
 829                         if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
 830                                 rec = nla_data(curr);
 831                                 /*
 832                                  * Get the first one. In the future, we may
 833                                  * need to get up to 6 pathrecords.
 834                                  */
 835                                 if ((rec->flags & mask) == mask) {
 836                                         mad = query->mad_buf->mad;
 837                                         mad->mad_hdr.method |=
 838                                                 IB_MGMT_METHOD_RESP;
 839                                         memcpy(mad->data, rec->path_rec,
 840                                                sizeof(rec->path_rec));
 841                                         status = 0;
 842                                         break;
 843                                 }
 844                         }
 845                 }
 846                 query->callback(query, status, mad);
 847         }
 848
 849         mad_send_wc.send_buf = query->mad_buf;
 850         mad_send_wc.status = IB_WC_SUCCESS;
 851         send_handler(query->mad_buf->mad_agent, &mad_send_wc);
 852 }
 853
 854 static void ib_nl_request_timeout(struct work_struct *work)
 855 {
 856         unsigned long flags;
 857         struct ib_sa_query *query;
 858         unsigned long delay;
 859         struct ib_mad_send_wc mad_send_wc;
 860         int ret;
 861
 862         spin_lock_irqsave(&ib_nl_request_lock, flags);
 863         while (!list_empty(&ib_nl_request_list)) {
 864                 query = list_entry(ib_nl_request_list.next,
 865                                    struct ib_sa_query, list);
 866
 867                 if (time_after(query->timeout, jiffies)) {
 868                         delay = query->timeout - jiffies;
 869                         if ((long)delay <= 0)
 870                                 delay = 1;
 871                         queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
 872                         break;
 873                 }
 874
 875                 list_del(&query->list);
 876                 ib_sa_disable_local_svc(query);
 877                 /* Hold the lock to protect against query cancellation */
 878                 if (ib_sa_query_cancelled(query))
 879                         ret = -1;
 880                 else
 881                         ret = ib_post_send_mad(query->mad_buf, NULL);
 882                 if (ret) {
 883                         mad_send_wc.send_buf = query->mad_buf;
 884                         mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
 885                         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 886                         send_handler(query->port->agent, &mad_send_wc);
 887                         spin_lock_irqsave(&ib_nl_request_lock, flags);
 888                 }
 889         }
 890         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 891 }
 892
 893 int ib_nl_handle_set_timeout(struct sk_buff *skb,
 894                              struct netlink_callback *cb)
 895 {
 896         const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
 897         int timeout, delta, abs_delta;
 898         const struct nlattr *attr;
 899         unsigned long flags;
 900         struct ib_sa_query *query;
 901         long delay = 0;
 902         struct nlattr *tb[LS_NLA_TYPE_MAX];
 903         int ret;
 904
 905         if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
 906             !(NETLINK_CB(skb).sk) ||
 907             !netlink_capable(skb, CAP_NET_ADMIN))
 908                 return -EPERM;
 909
 910         ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
 911                         nlmsg_len(nlh), ib_nl_policy, NULL);
 912         attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
 913         if (ret || !attr)
 914                 goto settimeout_out;
 915
 916         timeout = *(int *) nla_data(attr);
 917         if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
 918                 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
 919         if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
 920                 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
 921
 922         delta = timeout - sa_local_svc_timeout_ms;
 923         if (delta < 0)
 924                 abs_delta = -delta;
 925         else
 926                 abs_delta = delta;
 927
 928         if (delta != 0) {
 929                 spin_lock_irqsave(&ib_nl_request_lock, flags);
 930                 sa_local_svc_timeout_ms = timeout;
 931                 list_for_each_entry(query, &ib_nl_request_list, list) {
 932                         if (delta < 0 && abs_delta > query->timeout)
 933                                 query->timeout = 0;
 934                         else
 935                                 query->timeout += delta;
 936
 937                         /* Get the new delay from the first entry */
 938                         if (!delay) {
 939                                 delay = query->timeout - jiffies;
 940                                 if (delay <= 0)
 941                                         delay = 1;
 942                         }
 943                 }
 944                 if (delay)
 945                         mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
 946                                          (unsigned long)delay);
 947                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 948         }
 949
 950 settimeout_out:
 951         return skb->len;
 952 }
 953
 954 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
 955 {
 956         struct nlattr *tb[LS_NLA_TYPE_MAX];
 957         int ret;
 958
 959         if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
 960                 return 0;
 961
 962         ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
 963                         nlmsg_len(nlh), ib_nl_policy, NULL);
 964         if (ret)
 965                 return 0;
 966
 967         return 1;
 968 }
 969
 970 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
 971                               struct netlink_callback *cb)
 972 {
 973         const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
 974         unsigned long flags;
 975         struct ib_sa_query *query;
 976         struct ib_mad_send_buf *send_buf;
 977         struct ib_mad_send_wc mad_send_wc;
 978         int found = 0;
 979         int ret;
 980
 981         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
 982             !(NETLINK_CB(skb).sk) ||
 983             !netlink_capable(skb, CAP_NET_ADMIN))
 984                 return -EPERM;
 985
 986         spin_lock_irqsave(&ib_nl_request_lock, flags);
 987         list_for_each_entry(query, &ib_nl_request_list, list) {
 988                 /*
 989                  * If the query is cancelled, let the timeout routine
 990                  * take care of it.
 991                  */
 992                 if (nlh->nlmsg_seq == query->seq) {
 993                         found = !ib_sa_query_cancelled(query);
 994                         if (found)
 995                                 list_del(&query->list);
 996                         break;
 997                 }
 998         }
 999
1000         if (!found) {
1001                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1002                 goto resp_out;
1003         }
1004
1005         send_buf = query->mad_buf;
1006
1007         if (!ib_nl_is_good_resolve_resp(nlh)) {
1008                 /* if the result is a failure, send out the packet via IB */
1009                 ib_sa_disable_local_svc(query);
1010                 ret = ib_post_send_mad(query->mad_buf, NULL);
1011                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1012                 if (ret) {
1013                         mad_send_wc.send_buf = send_buf;
1014                         mad_send_wc.status = IB_WC_GENERAL_ERR;
1015                         send_handler(query->port->agent, &mad_send_wc);
1016                 }
1017         } else {
1018                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1019                 ib_nl_process_good_resolve_rsp(query, nlh);
1020         }
1021
1022 resp_out:
1023         return skb->len;
1024 }
1025
1026 static void free_sm_ah(struct kref *kref)
1027 {
1028         struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1029
1030         ib_destroy_ah(sm_ah->ah);
1031         kfree(sm_ah);
1032 }
1033
1034 void ib_sa_register_client(struct ib_sa_client *client)
1035 {
1036         atomic_set(&client->users, 1);
1037         init_completion(&client->comp);
1038 }
1039 EXPORT_SYMBOL(ib_sa_register_client);
1040
1041 void ib_sa_unregister_client(struct ib_sa_client *client)
1042 {
1043         ib_sa_client_put(client);
1044         wait_for_completion(&client->comp);
1045 }
1046 EXPORT_SYMBOL(ib_sa_unregister_client);
1047
1048 /**
1049  * ib_sa_cancel_query - try to cancel an SA query
1050  * @id:ID of query to cancel
1051  * @query:query pointer to cancel
1052  *
1053  * Try to cancel an SA query.  If the id and query don't match up or
1054  * the query has already completed, nothing is done.  Otherwise the
1055  * query is canceled and will complete with a status of -EINTR.
1056  */
1057 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1058 {
1059         unsigned long flags;
1060         struct ib_mad_agent *agent;
1061         struct ib_mad_send_buf *mad_buf;
1062
1063         spin_lock_irqsave(&idr_lock, flags);
1064         if (idr_find(&query_idr, id) != query) {
1065                 spin_unlock_irqrestore(&idr_lock, flags);
1066                 return;
1067         }
1068         agent = query->port->agent;
1069         mad_buf = query->mad_buf;
1070         spin_unlock_irqrestore(&idr_lock, flags);
1071
1072         /*
1073          * If the query is still on the netlink request list, schedule
1074          * it to be cancelled by the timeout routine. Otherwise, it has been
1075          * sent to the MAD layer and has to be cancelled from there.
1076          */
1077         if (!ib_nl_cancel_request(query))
1078                 ib_cancel_mad(agent, mad_buf);
1079 }
1080 EXPORT_SYMBOL(ib_sa_cancel_query);
1081
1082 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1083 {
1084         struct ib_sa_device *sa_dev;
1085         struct ib_sa_port   *port;
1086         unsigned long flags;
1087         u8 src_path_mask;
1088
1089         sa_dev = ib_get_client_data(device, &sa_client);
1090         if (!sa_dev)
1091                 return 0x7f;
1092
1093         port  = &sa_dev->port[port_num - sa_dev->start_port];
1094         spin_lock_irqsave(&port->ah_lock, flags);
1095         src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1096         spin_unlock_irqrestore(&port->ah_lock, flags);
1097
1098         return src_path_mask;
1099 }
1100
1101 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
1102                          struct ib_sa_path_rec *rec,
1103                          struct rdma_ah_attr *ah_attr)
1104 {
1105         int ret;
1106         u16 gid_index;
1107         int use_roce;
1108         struct net_device *ndev = NULL;
1109
1110         memset(ah_attr, 0, sizeof *ah_attr);
1111         ah_attr->dlid = be16_to_cpu(rec->dlid);
1112         ah_attr->sl = rec->sl;
1113         ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
1114                                  get_src_path_mask(device, port_num);
1115         ah_attr->port_num = port_num;
1116         ah_attr->static_rate = rec->rate;
1117
1118         use_roce = rdma_cap_eth_ah(device, port_num);
1119
1120         if (use_roce) {
1121                 struct net_device *idev;
1122                 struct net_device *resolved_dev;
1123                 struct rdma_dev_addr dev_addr = {.bound_dev_if = rec->ifindex,
1124                                                  .net = rec->net ? rec->net :
1125                                                          &init_net};
1126                 union {
1127                         struct sockaddr     _sockaddr;
1128                         struct sockaddr_in  _sockaddr_in;
1129                         struct sockaddr_in6 _sockaddr_in6;
1130                 } sgid_addr, dgid_addr;
1131
1132                 if (!device->get_netdev)
1133                         return -EOPNOTSUPP;
1134
1135                 rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1136                 rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1137
1138                 /* validate the route */
1139                 ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1140                                             &dgid_addr._sockaddr, &dev_addr);
1141                 if (ret)
1142                         return ret;
1143
1144                 if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1145                      dev_addr.network == RDMA_NETWORK_IPV6) &&
1146                     rec->gid_type != IB_GID_TYPE_ROCE_UDP_ENCAP)
1147                         return -EINVAL;
1148
1149                 idev = device->get_netdev(device, port_num);
1150                 if (!idev)
1151                         return -ENODEV;
1152
1153                 resolved_dev = dev_get_by_index(dev_addr.net,
1154                                                 dev_addr.bound_dev_if);
1155                 if (resolved_dev->flags & IFF_LOOPBACK) {
1156                         dev_put(resolved_dev);
1157                         resolved_dev = idev;
1158                         dev_hold(resolved_dev);
1159                 }
1160                 ndev = ib_get_ndev_from_path(rec);
1161                 rcu_read_lock();
1162                 if ((ndev && ndev != resolved_dev) ||
1163                     (resolved_dev != idev &&
1164                      !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1165                         ret = -EHOSTUNREACH;
1166                 rcu_read_unlock();
1167                 dev_put(idev);
1168                 dev_put(resolved_dev);
1169                 if (ret) {
1170                         if (ndev)
1171                                 dev_put(ndev);
1172                         return ret;
1173                 }
1174         }
1175
1176         if (rec->hop_limit > 0 || use_roce) {
1177                 ah_attr->ah_flags = IB_AH_GRH;
1178                 ah_attr->grh.dgid = rec->dgid;
1179
1180                 ret = ib_find_cached_gid_by_port(device, &rec->sgid,
1181                                                  rec->gid_type, port_num, ndev,
1182                                                  &gid_index);
1183                 if (ret) {
1184                         if (ndev)
1185                                 dev_put(ndev);
1186                         return ret;
1187                 }
1188
1189                 ah_attr->grh.sgid_index    = gid_index;
1190                 ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
1191                 ah_attr->grh.hop_limit     = rec->hop_limit;
1192                 ah_attr->grh.traffic_class = rec->traffic_class;
1193                 if (ndev)
1194                         dev_put(ndev);
1195         }
1196
1197         if (use_roce)
1198                 memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
1199
1200         return 0;
1201 }
1202 EXPORT_SYMBOL(ib_init_ah_from_path);
1203
1204 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1205 {
1206         unsigned long flags;
1207
1208         spin_lock_irqsave(&query->port->ah_lock, flags);
1209         if (!query->port->sm_ah) {
1210                 spin_unlock_irqrestore(&query->port->ah_lock, flags);
1211                 return -EAGAIN;
1212         }
1213         kref_get(&query->port->sm_ah->ref);
1214         query->sm_ah = query->port->sm_ah;
1215         spin_unlock_irqrestore(&query->port->ah_lock, flags);
1216
1217         query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1218                                             query->sm_ah->pkey_index,
1219                                             0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1220                                             gfp_mask,
1221                                             ((query->flags & IB_SA_QUERY_OPA) ?
1222                                              OPA_MGMT_BASE_VERSION :
1223                                              IB_MGMT_BASE_VERSION));
1224         if (IS_ERR(query->mad_buf)) {
1225                 kref_put(&query->sm_ah->ref, free_sm_ah);
1226                 return -ENOMEM;
1227         }
1228
1229         query->mad_buf->ah = query->sm_ah->ah;
1230
1231         return 0;
1232 }
1233
1234 static void free_mad(struct ib_sa_query *query)
1235 {
1236         ib_free_send_mad(query->mad_buf);
1237         kref_put(&query->sm_ah->ref, free_sm_ah);
1238 }
1239
1240 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1241 {
1242         struct ib_sa_mad *mad = query->mad_buf->mad;
1243         unsigned long flags;
1244
1245         memset(mad, 0, sizeof *mad);
1246
1247         if (query->flags & IB_SA_QUERY_OPA) {
1248                 mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1249                 mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1250         } else {
1251                 mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1252                 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1253         }
1254         mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1255         spin_lock_irqsave(&tid_lock, flags);
1256         mad->mad_hdr.tid           =
1257                 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1258         spin_unlock_irqrestore(&tid_lock, flags);
1259 }
1260
1261 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1262 {
1263         bool preload = gfpflags_allow_blocking(gfp_mask);
1264         unsigned long flags;
1265         int ret, id;
1266
1267         if (preload)
1268                 idr_preload(gfp_mask);
1269         spin_lock_irqsave(&idr_lock, flags);
1270
1271         id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1272
1273         spin_unlock_irqrestore(&idr_lock, flags);
1274         if (preload)
1275                 idr_preload_end();
1276         if (id < 0)
1277                 return id;
1278
1279         query->mad_buf->timeout_ms  = timeout_ms;
1280         query->mad_buf->context[0] = query;
1281         query->id = id;
1282
1283         if (query->flags & IB_SA_ENABLE_LOCAL_SERVICE) {
1284                 if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1285                         if (!ib_nl_make_request(query, gfp_mask))
1286                                 return id;
1287                 }
1288                 ib_sa_disable_local_svc(query);
1289         }
1290
1291         ret = ib_post_send_mad(query->mad_buf, NULL);
1292         if (ret) {
1293                 spin_lock_irqsave(&idr_lock, flags);
1294                 idr_remove(&query_idr, id);
1295                 spin_unlock_irqrestore(&idr_lock, flags);
1296         }
1297
1298         /*
1299          * It's not safe to dereference query any more, because the
1300          * send may already have completed and freed the query in
1301          * another context.
1302          */
1303         return ret ? ret : id;
1304 }
1305
1306 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
1307 {
1308         ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1309 }
1310 EXPORT_SYMBOL(ib_sa_unpack_path);
1311
1312 void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
1313 {
1314         ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1315 }
1316 EXPORT_SYMBOL(ib_sa_pack_path);
1317
1318 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1319                                     int status,
1320                                     struct ib_sa_mad *mad)
1321 {
1322         struct ib_sa_path_query *query =
1323                 container_of(sa_query, struct ib_sa_path_query, sa_query);
1324
1325         if (mad) {
1326                 struct ib_sa_path_rec rec;
1327
1328                 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1329                           mad->data, &rec);
1330                 rec.net = NULL;
1331                 rec.ifindex = 0;
1332                 rec.gid_type = IB_GID_TYPE_IB;
1333                 eth_zero_addr(rec.dmac);
1334                 query->callback(status, &rec, query->context);
1335         } else
1336                 query->callback(status, NULL, query->context);
1337 }
1338
1339 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1340 {
1341         kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
1342 }
1343
1344 /**
1345  * ib_sa_path_rec_get - Start a Path get query
1346  * @client:SA client
1347  * @device:device to send query on
1348  * @port_num: port number to send query on
1349  * @rec:Path Record to send in query
1350  * @comp_mask:component mask to send in query
1351  * @timeout_ms:time to wait for response
1352  * @gfp_mask:GFP mask to use for internal allocations
1353  * @callback:function called when query completes, times out or is
1354  * canceled
1355  * @context:opaque user context passed to callback
1356  * @sa_query:query context, used to cancel query
1357  *
1358  * Send a Path Record Get query to the SA to look up a path.  The
1359  * callback function will be called when the query completes (or
1360  * fails); status is 0 for a successful response, -EINTR if the query
1361  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1362  * occurred sending the query.  The resp parameter of the callback is
1363  * only valid if status is 0.
1364  *
1365  * If the return value of ib_sa_path_rec_get() is negative, it is an
1366  * error code.  Otherwise it is a query ID that can be used to cancel
1367  * the query.
1368  */
1369 int ib_sa_path_rec_get(struct ib_sa_client *client,
1370                        struct ib_device *device, u8 port_num,
1371                        struct ib_sa_path_rec *rec,
1372                        ib_sa_comp_mask comp_mask,
1373                        int timeout_ms, gfp_t gfp_mask,
1374                        void (*callback)(int status,
1375                                         struct ib_sa_path_rec *resp,
1376                                         void *context),
1377                        void *context,
1378                        struct ib_sa_query **sa_query)
1379 {
1380         struct ib_sa_path_query *query;
1381         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1382         struct ib_sa_port   *port;
1383         struct ib_mad_agent *agent;
1384         struct ib_sa_mad *mad;
1385         int ret;
1386
1387         if (!sa_dev)
1388                 return -ENODEV;
1389
1390         port  = &sa_dev->port[port_num - sa_dev->start_port];
1391         agent = port->agent;
1392
1393         query = kzalloc(sizeof(*query), gfp_mask);
1394         if (!query)
1395                 return -ENOMEM;
1396
1397         query->sa_query.port     = port;
1398         ret = alloc_mad(&query->sa_query, gfp_mask);
1399         if (ret)
1400                 goto err1;
1401
1402         ib_sa_client_get(client);
1403         query->sa_query.client = client;
1404         query->callback        = callback;
1405         query->context         = context;
1406
1407         mad = query->sa_query.mad_buf->mad;
1408         init_mad(&query->sa_query, agent);
1409
1410         query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1411         query->sa_query.release  = ib_sa_path_rec_release;
1412         mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1413         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1414         mad->sa_hdr.comp_mask    = comp_mask;
1415
1416         ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
1417
1418         *sa_query = &query->sa_query;
1419
1420         query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1421         query->sa_query.mad_buf->context[1] = rec;
1422
1423         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1424         if (ret < 0)
1425                 goto err2;
1426
1427         return ret;
1428
1429 err2:
1430         *sa_query = NULL;
1431         ib_sa_client_put(query->sa_query.client);
1432         free_mad(&query->sa_query);
1433
1434 err1:
1435         kfree(query);
1436         return ret;
1437 }
1438 EXPORT_SYMBOL(ib_sa_path_rec_get);
1439
1440 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1441                                     int status,
1442                                     struct ib_sa_mad *mad)
1443 {
1444         struct ib_sa_service_query *query =
1445                 container_of(sa_query, struct ib_sa_service_query, sa_query);
1446
1447         if (mad) {
1448                 struct ib_sa_service_rec rec;
1449
1450                 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1451                           mad->data, &rec);
1452                 query->callback(status, &rec, query->context);
1453         } else
1454                 query->callback(status, NULL, query->context);
1455 }
1456
1457 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1458 {
1459         kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1460 }
1461
1462 /**
1463  * ib_sa_service_rec_query - Start Service Record operation
1464  * @client:SA client
1465  * @device:device to send request on
1466  * @port_num: port number to send request on
1467  * @method:SA method - should be get, set, or delete
1468  * @rec:Service Record to send in request
1469  * @comp_mask:component mask to send in request
1470  * @timeout_ms:time to wait for response
1471  * @gfp_mask:GFP mask to use for internal allocations
1472  * @callback:function called when request completes, times out or is
1473  * canceled
1474  * @context:opaque user context passed to callback
1475  * @sa_query:request context, used to cancel request
1476  *
1477  * Send a Service Record set/get/delete to the SA to register,
1478  * unregister or query a service record.
1479  * The callback function will be called when the request completes (or
1480  * fails); status is 0 for a successful response, -EINTR if the query
1481  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1482  * occurred sending the query.  The resp parameter of the callback is
1483  * only valid if status is 0.
1484  *
1485  * If the return value of ib_sa_service_rec_query() is negative, it is an
1486  * error code.  Otherwise it is a request ID that can be used to cancel
1487  * the query.
1488  */
1489 int ib_sa_service_rec_query(struct ib_sa_client *client,
1490                             struct ib_device *device, u8 port_num, u8 method,
1491                             struct ib_sa_service_rec *rec,
1492                             ib_sa_comp_mask comp_mask,
1493                             int timeout_ms, gfp_t gfp_mask,
1494                             void (*callback)(int status,
1495                                              struct ib_sa_service_rec *resp,
1496                                              void *context),
1497                             void *context,
1498                             struct ib_sa_query **sa_query)
1499 {
1500         struct ib_sa_service_query *query;
1501         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1502         struct ib_sa_port   *port;
1503         struct ib_mad_agent *agent;
1504         struct ib_sa_mad *mad;
1505         int ret;
1506
1507         if (!sa_dev)
1508                 return -ENODEV;
1509
1510         port  = &sa_dev->port[port_num - sa_dev->start_port];
1511         agent = port->agent;
1512
1513         if (method != IB_MGMT_METHOD_GET &&
1514             method != IB_MGMT_METHOD_SET &&
1515             method != IB_SA_METHOD_DELETE)
1516                 return -EINVAL;
1517
1518         query = kzalloc(sizeof(*query), gfp_mask);
1519         if (!query)
1520                 return -ENOMEM;
1521
1522         query->sa_query.port     = port;
1523         ret = alloc_mad(&query->sa_query, gfp_mask);
1524         if (ret)
1525                 goto err1;
1526
1527         ib_sa_client_get(client);
1528         query->sa_query.client = client;
1529         query->callback        = callback;
1530         query->context         = context;
1531
1532         mad = query->sa_query.mad_buf->mad;
1533         init_mad(&query->sa_query, agent);
1534
1535         query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1536         query->sa_query.release  = ib_sa_service_rec_release;
1537         mad->mad_hdr.method      = method;
1538         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1539         mad->sa_hdr.comp_mask    = comp_mask;
1540
1541         ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1542                 rec, mad->data);
1543
1544         *sa_query = &query->sa_query;
1545
1546         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1547         if (ret < 0)
1548                 goto err2;
1549
1550         return ret;
1551
1552 err2:
1553         *sa_query = NULL;
1554         ib_sa_client_put(query->sa_query.client);
1555         free_mad(&query->sa_query);
1556
1557 err1:
1558         kfree(query);
1559         return ret;
1560 }
1561 EXPORT_SYMBOL(ib_sa_service_rec_query);
1562
1563 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1564                                         int status,
1565                                         struct ib_sa_mad *mad)
1566 {
1567         struct ib_sa_mcmember_query *query =
1568                 container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1569
1570         if (mad) {
1571                 struct ib_sa_mcmember_rec rec;
1572
1573                 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1574                           mad->data, &rec);
1575                 query->callback(status, &rec, query->context);
1576         } else
1577                 query->callback(status, NULL, query->context);
1578 }
1579
1580 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1581 {
1582         kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1583 }
1584
1585 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1586                              struct ib_device *device, u8 port_num,
1587                              u8 method,
1588                              struct ib_sa_mcmember_rec *rec,
1589                              ib_sa_comp_mask comp_mask,
1590                              int timeout_ms, gfp_t gfp_mask,
1591                              void (*callback)(int status,
1592                                               struct ib_sa_mcmember_rec *resp,
1593                                               void *context),
1594                              void *context,
1595                              struct ib_sa_query **sa_query)
1596 {
1597         struct ib_sa_mcmember_query *query;
1598         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1599         struct ib_sa_port   *port;
1600         struct ib_mad_agent *agent;
1601         struct ib_sa_mad *mad;
1602         int ret;
1603
1604         if (!sa_dev)
1605                 return -ENODEV;
1606
1607         port  = &sa_dev->port[port_num - sa_dev->start_port];
1608         agent = port->agent;
1609
1610         query = kzalloc(sizeof(*query), gfp_mask);
1611         if (!query)
1612                 return -ENOMEM;
1613
1614         query->sa_query.port     = port;
1615         ret = alloc_mad(&query->sa_query, gfp_mask);
1616         if (ret)
1617                 goto err1;
1618
1619         ib_sa_client_get(client);
1620         query->sa_query.client = client;
1621         query->callback        = callback;
1622         query->context         = context;
1623
1624         mad = query->sa_query.mad_buf->mad;
1625         init_mad(&query->sa_query, agent);
1626
1627         query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1628         query->sa_query.release  = ib_sa_mcmember_rec_release;
1629         mad->mad_hdr.method      = method;
1630         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1631         mad->sa_hdr.comp_mask    = comp_mask;
1632
1633         ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1634                 rec, mad->data);
1635
1636         *sa_query = &query->sa_query;
1637
1638         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1639         if (ret < 0)
1640                 goto err2;
1641
1642         return ret;
1643
1644 err2:
1645         *sa_query = NULL;
1646         ib_sa_client_put(query->sa_query.client);
1647         free_mad(&query->sa_query);
1648
1649 err1:
1650         kfree(query);
1651         return ret;
1652 }
1653
1654 /* Support GuidInfoRecord */
1655 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1656                                         int status,
1657                                         struct ib_sa_mad *mad)
1658 {
1659         struct ib_sa_guidinfo_query *query =
1660                 container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1661
1662         if (mad) {
1663                 struct ib_sa_guidinfo_rec rec;
1664
1665                 ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1666                           mad->data, &rec);
1667                 query->callback(status, &rec, query->context);
1668         } else
1669                 query->callback(status, NULL, query->context);
1670 }
1671
1672 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1673 {
1674         kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1675 }
1676
1677 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1678                               struct ib_device *device, u8 port_num,
1679                               struct ib_sa_guidinfo_rec *rec,
1680                               ib_sa_comp_mask comp_mask, u8 method,
1681                               int timeout_ms, gfp_t gfp_mask,
1682                               void (*callback)(int status,
1683                                                struct ib_sa_guidinfo_rec *resp,
1684                                                void *context),
1685                               void *context,
1686                               struct ib_sa_query **sa_query)
1687 {
1688         struct ib_sa_guidinfo_query *query;
1689         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1690         struct ib_sa_port *port;
1691         struct ib_mad_agent *agent;
1692         struct ib_sa_mad *mad;
1693         int ret;
1694
1695         if (!sa_dev)
1696                 return -ENODEV;
1697
1698         if (method != IB_MGMT_METHOD_GET &&
1699             method != IB_MGMT_METHOD_SET &&
1700             method != IB_SA_METHOD_DELETE) {
1701                 return -EINVAL;
1702         }
1703
1704         port  = &sa_dev->port[port_num - sa_dev->start_port];
1705         agent = port->agent;
1706
1707         query = kzalloc(sizeof(*query), gfp_mask);
1708         if (!query)
1709                 return -ENOMEM;
1710
1711         query->sa_query.port = port;
1712         ret = alloc_mad(&query->sa_query, gfp_mask);
1713         if (ret)
1714                 goto err1;
1715
1716         ib_sa_client_get(client);
1717         query->sa_query.client = client;
1718         query->callback        = callback;
1719         query->context         = context;
1720
1721         mad = query->sa_query.mad_buf->mad;
1722         init_mad(&query->sa_query, agent);
1723
1724         query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1725         query->sa_query.release  = ib_sa_guidinfo_rec_release;
1726
1727         mad->mad_hdr.method      = method;
1728         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1729         mad->sa_hdr.comp_mask    = comp_mask;
1730
1731         ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1732                 mad->data);
1733
1734         *sa_query = &query->sa_query;
1735
1736         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1737         if (ret < 0)
1738                 goto err2;
1739
1740         return ret;
1741
1742 err2:
1743         *sa_query = NULL;
1744         ib_sa_client_put(query->sa_query.client);
1745         free_mad(&query->sa_query);
1746
1747 err1:
1748         kfree(query);
1749         return ret;
1750 }
1751 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1752
1753 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
1754                                     struct ib_device *device,
1755                                     u8 port_num)
1756 {
1757         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1758         struct ib_sa_port *port;
1759         bool ret = false;
1760         unsigned long flags;
1761
1762         if (!sa_dev)
1763                 return ret;
1764
1765         port  = &sa_dev->port[port_num - sa_dev->start_port];
1766
1767         spin_lock_irqsave(&port->classport_lock, flags);
1768         if ((port->classport_info.valid) &&
1769             (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
1770                 ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
1771                         & IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
1772         spin_unlock_irqrestore(&port->classport_lock, flags);
1773         return ret;
1774 }
1775 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
1776
1777 struct ib_classport_info_context {
1778         struct completion       done;
1779         struct ib_sa_query      *sa_query;
1780 };
1781
1782 static void ib_classportinfo_cb(void *context)
1783 {
1784         struct ib_classport_info_context *cb_ctx = context;
1785
1786         complete(&cb_ctx->done);
1787 }
1788
1789 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
1790                                               int status,
1791                                               struct ib_sa_mad *mad)
1792 {
1793         unsigned long flags;
1794         struct ib_sa_classport_info_query *query =
1795                 container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
1796         struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
1797
1798         if (mad) {
1799                 if (sa_query->flags & IB_SA_QUERY_OPA) {
1800                         struct opa_class_port_info rec;
1801
1802                         ib_unpack(opa_classport_info_rec_table,
1803                                   ARRAY_SIZE(opa_classport_info_rec_table),
1804                                   mad->data, &rec);
1805
1806                         spin_lock_irqsave(&sa_query->port->classport_lock,
1807                                           flags);
1808                         if (!status && !info->valid) {
1809                                 memcpy(&info->data.opa, &rec,
1810                                        sizeof(info->data.opa));
1811
1812                                 info->valid = true;
1813                                 info->data.type = RDMA_CLASS_PORT_INFO_OPA;
1814                         }
1815                         spin_unlock_irqrestore(&sa_query->port->classport_lock,
1816                                                flags);
1817
1818                 } else {
1819                         struct ib_class_port_info rec;
1820
1821                         ib_unpack(ib_classport_info_rec_table,
1822                                   ARRAY_SIZE(ib_classport_info_rec_table),
1823                                   mad->data, &rec);
1824
1825                         spin_lock_irqsave(&sa_query->port->classport_lock,
1826                                           flags);
1827                         if (!status && !info->valid) {
1828                                 memcpy(&info->data.ib, &rec,
1829                                        sizeof(info->data.ib));
1830
1831                                 info->valid = true;
1832                                 info->data.type = RDMA_CLASS_PORT_INFO_IB;
1833                         }
1834                         spin_unlock_irqrestore(&sa_query->port->classport_lock,
1835                                                flags);
1836                 }
1837         }
1838         query->callback(query->context);
1839 }
1840
1841 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
1842 {
1843         kfree(container_of(sa_query, struct ib_sa_classport_info_query,
1844                            sa_query));
1845 }
1846
1847 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
1848                                           int timeout_ms,
1849                                           void (*callback)(void *context),
1850                                           void *context,
1851                                           struct ib_sa_query **sa_query)
1852 {
1853         struct ib_mad_agent *agent;
1854         struct ib_sa_classport_info_query *query;
1855         struct ib_sa_mad *mad;
1856         gfp_t gfp_mask = GFP_KERNEL;
1857         int ret;
1858
1859         agent = port->agent;
1860
1861         query = kzalloc(sizeof(*query), gfp_mask);
1862         if (!query)
1863                 return -ENOMEM;
1864
1865         query->sa_query.port = port;
1866         query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
1867                                                  port->port_num) ?
1868                                  IB_SA_QUERY_OPA : 0;
1869         ret = alloc_mad(&query->sa_query, gfp_mask);
1870         if (ret)
1871                 goto err_free;
1872
1873         query->callback = callback;
1874         query->context = context;
1875
1876         mad = query->sa_query.mad_buf->mad;
1877         init_mad(&query->sa_query, agent);
1878
1879         query->sa_query.callback = ib_sa_classport_info_rec_callback;
1880         query->sa_query.release  = ib_sa_classport_info_rec_release;
1881         mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1882         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
1883         mad->sa_hdr.comp_mask    = 0;
1884         *sa_query = &query->sa_query;
1885
1886         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1887         if (ret < 0)
1888                 goto err_free_mad;
1889
1890         return ret;
1891
1892 err_free_mad:
1893         *sa_query = NULL;
1894         free_mad(&query->sa_query);
1895
1896 err_free:
1897         kfree(query);
1898         return ret;
1899 }
1900
1901 static void update_ib_cpi(struct work_struct *work)
1902 {
1903         struct ib_sa_port *port =
1904                 container_of(work, struct ib_sa_port, ib_cpi_work.work);
1905         struct ib_classport_info_context *cb_context;
1906         unsigned long flags;
1907         int ret;
1908
1909         /* If the classport info is valid, nothing
1910          * to do here.
1911          */
1912         spin_lock_irqsave(&port->classport_lock, flags);
1913         if (port->classport_info.valid) {
1914                 spin_unlock_irqrestore(&port->classport_lock, flags);
1915                 return;
1916         }
1917         spin_unlock_irqrestore(&port->classport_lock, flags);
1918
1919         cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
1920         if (!cb_context)
1921                 goto err_nomem;
1922
1923         init_completion(&cb_context->done);
1924
1925         ret = ib_sa_classport_info_rec_query(port, 3000,
1926                                              ib_classportinfo_cb, cb_context,
1927                                              &cb_context->sa_query);
1928         if (ret < 0)
1929                 goto free_cb_err;
1930         wait_for_completion(&cb_context->done);
1931 free_cb_err:
1932         kfree(cb_context);
1933         spin_lock_irqsave(&port->classport_lock, flags);
1934
1935         /* If the classport info is still not valid, the query should have
1936          * failed for some reason. Retry issuing the query
1937          */
1938         if (!port->classport_info.valid) {
1939                 port->classport_info.retry_cnt++;
1940                 if (port->classport_info.retry_cnt <=
1941                     IB_SA_CPI_MAX_RETRY_CNT) {
1942                         unsigned long delay =
1943                                 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
1944
1945                         queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
1946                 }
1947         }
1948         spin_unlock_irqrestore(&port->classport_lock, flags);
1949
1950 err_nomem:
1951         return;
1952 }
1953
1954 static void send_handler(struct ib_mad_agent *agent,
1955                          struct ib_mad_send_wc *mad_send_wc)
1956 {
1957         struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1958         unsigned long flags;
1959
1960         if (query->callback)
1961                 switch (mad_send_wc->status) {
1962                 case IB_WC_SUCCESS:
1963                         /* No callback -- already got recv */
1964                         break;
1965                 case IB_WC_RESP_TIMEOUT_ERR:
1966                         query->callback(query, -ETIMEDOUT, NULL);
1967                         break;
1968                 case IB_WC_WR_FLUSH_ERR:
1969                         query->callback(query, -EINTR, NULL);
1970                         break;
1971                 default:
1972                         query->callback(query, -EIO, NULL);
1973                         break;
1974                 }
1975
1976         spin_lock_irqsave(&idr_lock, flags);
1977         idr_remove(&query_idr, query->id);
1978         spin_unlock_irqrestore(&idr_lock, flags);
1979
1980         free_mad(query);
1981         if (query->client)
1982                 ib_sa_client_put(query->client);
1983         query->release(query);
1984 }
1985
1986 static void recv_handler(struct ib_mad_agent *mad_agent,
1987                          struct ib_mad_send_buf *send_buf,
1988                          struct ib_mad_recv_wc *mad_recv_wc)
1989 {
1990         struct ib_sa_query *query;
1991
1992         if (!send_buf)
1993                 return;
1994
1995         query = send_buf->context[0];
1996         if (query->callback) {
1997                 if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1998                         query->callback(query,
1999                                         mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2000                                         -EINVAL : 0,
2001                                         (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2002                 else
2003                         query->callback(query, -EIO, NULL);
2004         }
2005
2006         ib_free_recv_mad(mad_recv_wc);
2007 }
2008
2009 static void update_sm_ah(struct work_struct *work)
2010 {
2011         struct ib_sa_port *port =
2012                 container_of(work, struct ib_sa_port, update_task);
2013         struct ib_sa_sm_ah *new_ah;
2014         struct ib_port_attr port_attr;
2015         struct rdma_ah_attr   ah_attr;
2016
2017         if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2018                 pr_warn("Couldn't query port\n");
2019                 return;
2020         }
2021
2022         new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2023         if (!new_ah)
2024                 return;
2025
2026         kref_init(&new_ah->ref);
2027         new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2028
2029         new_ah->pkey_index = 0;
2030         if (ib_find_pkey(port->agent->device, port->port_num,
2031                          IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2032                 pr_err("Couldn't find index for default PKey\n");
2033
2034         memset(&ah_attr, 0, sizeof(ah_attr));
2035         ah_attr.dlid     = port_attr.sm_lid;
2036         ah_attr.sl       = port_attr.sm_sl;
2037         ah_attr.port_num = port->port_num;
2038         if (port_attr.grh_required) {
2039                 ah_attr.ah_flags = IB_AH_GRH;
2040                 ah_attr.grh.dgid.global.subnet_prefix =
2041                         cpu_to_be64(port_attr.subnet_prefix);
2042                 ah_attr.grh.dgid.global.interface_id =
2043                         cpu_to_be64(IB_SA_WELL_KNOWN_GUID);
2044         }
2045
2046         new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr);
2047         if (IS_ERR(new_ah->ah)) {
2048                 pr_warn("Couldn't create new SM AH\n");
2049                 kfree(new_ah);
2050                 return;
2051         }
2052
2053         spin_lock_irq(&port->ah_lock);
2054         if (port->sm_ah)
2055                 kref_put(&port->sm_ah->ref, free_sm_ah);
2056         port->sm_ah = new_ah;
2057         spin_unlock_irq(&port->ah_lock);
2058 }
2059
2060 static void ib_sa_event(struct ib_event_handler *handler,
2061                         struct ib_event *event)
2062 {
2063         if (event->event == IB_EVENT_PORT_ERR    ||
2064             event->event == IB_EVENT_PORT_ACTIVE ||
2065             event->event == IB_EVENT_LID_CHANGE  ||
2066             event->event == IB_EVENT_PKEY_CHANGE ||
2067             event->event == IB_EVENT_SM_CHANGE   ||
2068             event->event == IB_EVENT_CLIENT_REREGISTER) {
2069                 unsigned long flags;
2070                 struct ib_sa_device *sa_dev =
2071                         container_of(handler, typeof(*sa_dev), event_handler);
2072                 u8 port_num = event->element.port_num - sa_dev->start_port;
2073                 struct ib_sa_port *port = &sa_dev->port[port_num];
2074
2075                 if (!rdma_cap_ib_sa(handler->device, port->port_num))
2076                         return;
2077
2078                 spin_lock_irqsave(&port->ah_lock, flags);
2079                 if (port->sm_ah)
2080                         kref_put(&port->sm_ah->ref, free_sm_ah);
2081                 port->sm_ah = NULL;
2082                 spin_unlock_irqrestore(&port->ah_lock, flags);
2083
2084                 if (event->event == IB_EVENT_SM_CHANGE ||
2085                     event->event == IB_EVENT_CLIENT_REREGISTER ||
2086                     event->event == IB_EVENT_LID_CHANGE ||
2087                     event->event == IB_EVENT_PORT_ACTIVE) {
2088                         unsigned long delay =
2089                                 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2090
2091                         spin_lock_irqsave(&port->classport_lock, flags);
2092                         port->classport_info.valid = false;
2093                         port->classport_info.retry_cnt = 0;
2094                         spin_unlock_irqrestore(&port->classport_lock, flags);
2095                         queue_delayed_work(ib_wq,
2096                                            &port->ib_cpi_work, delay);
2097                 }
2098                 queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2099         }
2100 }
2101
2102 static void ib_sa_add_one(struct ib_device *device)
2103 {
2104         struct ib_sa_device *sa_dev;
2105         int s, e, i;
2106         int count = 0;
2107
2108         s = rdma_start_port(device);
2109         e = rdma_end_port(device);
2110
2111         sa_dev = kzalloc(sizeof *sa_dev +
2112                          (e - s + 1) * sizeof (struct ib_sa_port),
2113                          GFP_KERNEL);
2114         if (!sa_dev)
2115                 return;
2116
2117         sa_dev->start_port = s;
2118         sa_dev->end_port   = e;
2119
2120         for (i = 0; i <= e - s; ++i) {
2121                 spin_lock_init(&sa_dev->port[i].ah_lock);
2122                 if (!rdma_cap_ib_sa(device, i + 1))
2123                         continue;
2124
2125                 sa_dev->port[i].sm_ah    = NULL;
2126                 sa_dev->port[i].port_num = i + s;
2127
2128                 spin_lock_init(&sa_dev->port[i].classport_lock);
2129                 sa_dev->port[i].classport_info.valid = false;
2130
2131                 sa_dev->port[i].agent =
2132                         ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2133                                               NULL, 0, send_handler,
2134                                               recv_handler, sa_dev, 0);
2135                 if (IS_ERR(sa_dev->port[i].agent))
2136                         goto err;
2137
2138                 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2139                 INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2140                                   update_ib_cpi);
2141
2142                 count++;
2143         }
2144
2145         if (!count)
2146                 goto free;
2147
2148         ib_set_client_data(device, &sa_client, sa_dev);
2149
2150         /*
2151          * We register our event handler after everything is set up,
2152          * and then update our cached info after the event handler is
2153          * registered to avoid any problems if a port changes state
2154          * during our initialization.
2155          */
2156
2157         INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2158         if (ib_register_event_handler(&sa_dev->event_handler))
2159                 goto err;
2160
2161         for (i = 0; i <= e - s; ++i) {
2162                 if (rdma_cap_ib_sa(device, i + 1))
2163                         update_sm_ah(&sa_dev->port[i].update_task);
2164         }
2165
2166         return;
2167
2168 err:
2169         while (--i >= 0) {
2170                 if (rdma_cap_ib_sa(device, i + 1))
2171                         ib_unregister_mad_agent(sa_dev->port[i].agent);
2172         }
2173 free:
2174         kfree(sa_dev);
2175         return;
2176 }
2177
2178 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2179 {
2180         struct ib_sa_device *sa_dev = client_data;
2181         int i;
2182
2183         if (!sa_dev)
2184                 return;
2185
2186         ib_unregister_event_handler(&sa_dev->event_handler);
2187         flush_workqueue(ib_wq);
2188
2189         for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2190                 if (rdma_cap_ib_sa(device, i + 1)) {
2191                         cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2192                         ib_unregister_mad_agent(sa_dev->port[i].agent);
2193                         if (sa_dev->port[i].sm_ah)
2194                                 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2195                 }
2196
2197         }
2198
2199         kfree(sa_dev);
2200 }
2201
2202 int ib_sa_init(void)
2203 {
2204         int ret;
2205
2206         get_random_bytes(&tid, sizeof tid);
2207
2208         atomic_set(&ib_nl_sa_request_seq, 0);
2209
2210         ret = ib_register_client(&sa_client);
2211         if (ret) {
2212                 pr_err("Couldn't register ib_sa client\n");
2213                 goto err1;
2214         }
2215
2216         ret = mcast_init();
2217         if (ret) {
2218                 pr_err("Couldn't initialize multicast handling\n");
2219                 goto err2;
2220         }
2221
2222         ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2223         if (!ib_nl_wq) {
2224                 ret = -ENOMEM;
2225                 goto err3;
2226         }
2227
2228         INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2229
2230         return 0;
2231
2232 err3:
2233         mcast_cleanup();
2234 err2:
2235         ib_unregister_client(&sa_client);
2236 err1:
2237         return ret;
2238 }
2239
2240 void ib_sa_cleanup(void)
2241 {
2242         cancel_delayed_work(&ib_nl_timed_work);
2243         flush_workqueue(ib_nl_wq);
2244         destroy_workqueue(ib_nl_wq);
2245         mcast_cleanup();
2246         ib_unregister_client(&sa_client);
2247         idr_destroy(&query_idr);
2248 }