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