linux/drivers/infiniband/core/sa_query.c
<<
>>
Prefs
   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
  56MODULE_AUTHOR("Roland Dreier");
  57MODULE_DESCRIPTION("InfiniBand subnet administration query support");
  58MODULE_LICENSE("Dual BSD/GPL");
  59
  60#define IB_SA_LOCAL_SVC_TIMEOUT_MIN             100
  61#define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT         2000
  62#define IB_SA_LOCAL_SVC_TIMEOUT_MAX             200000
  63static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
  64
  65struct ib_sa_sm_ah {
  66        struct ib_ah        *ah;
  67        struct kref          ref;
  68        u16                  pkey_index;
  69        u8                   src_path_mask;
  70};
  71
  72struct ib_sa_port {
  73        struct ib_mad_agent *agent;
  74        struct ib_sa_sm_ah  *sm_ah;
  75        struct work_struct   update_task;
  76        spinlock_t           ah_lock;
  77        u8                   port_num;
  78};
  79
  80struct ib_sa_device {
  81        int                     start_port, end_port;
  82        struct ib_event_handler event_handler;
  83        struct ib_sa_port port[0];
  84};
  85
  86struct ib_sa_query {
  87        void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
  88        void (*release)(struct ib_sa_query *);
  89        struct ib_sa_client    *client;
  90        struct ib_sa_port      *port;
  91        struct ib_mad_send_buf *mad_buf;
  92        struct ib_sa_sm_ah     *sm_ah;
  93        int                     id;
  94        u32                     flags;
  95        struct list_head        list; /* Local svc request list */
  96        u32                     seq; /* Local svc request sequence number */
  97        unsigned long           timeout; /* Local svc timeout */
  98        u8                      path_use; /* How will the pathrecord be used */
  99};
 100
 101#define IB_SA_ENABLE_LOCAL_SERVICE      0x00000001
 102#define IB_SA_CANCEL                    0x00000002
 103
 104struct ib_sa_service_query {
 105        void (*callback)(int, struct ib_sa_service_rec *, void *);
 106        void *context;
 107        struct ib_sa_query sa_query;
 108};
 109
 110struct ib_sa_path_query {
 111        void (*callback)(int, struct ib_sa_path_rec *, void *);
 112        void *context;
 113        struct ib_sa_query sa_query;
 114};
 115
 116struct ib_sa_guidinfo_query {
 117        void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
 118        void *context;
 119        struct ib_sa_query sa_query;
 120};
 121
 122struct ib_sa_mcmember_query {
 123        void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
 124        void *context;
 125        struct ib_sa_query sa_query;
 126};
 127
 128static LIST_HEAD(ib_nl_request_list);
 129static DEFINE_SPINLOCK(ib_nl_request_lock);
 130static atomic_t ib_nl_sa_request_seq;
 131static struct workqueue_struct *ib_nl_wq;
 132static struct delayed_work ib_nl_timed_work;
 133static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
 134        [LS_NLA_TYPE_PATH_RECORD]       = {.type = NLA_BINARY,
 135                .len = sizeof(struct ib_path_rec_data)},
 136        [LS_NLA_TYPE_TIMEOUT]           = {.type = NLA_U32},
 137        [LS_NLA_TYPE_SERVICE_ID]        = {.type = NLA_U64},
 138        [LS_NLA_TYPE_DGID]              = {.type = NLA_BINARY,
 139                .len = sizeof(struct rdma_nla_ls_gid)},
 140        [LS_NLA_TYPE_SGID]              = {.type = NLA_BINARY,
 141                .len = sizeof(struct rdma_nla_ls_gid)},
 142        [LS_NLA_TYPE_TCLASS]            = {.type = NLA_U8},
 143        [LS_NLA_TYPE_PKEY]              = {.type = NLA_U16},
 144        [LS_NLA_TYPE_QOS_CLASS]         = {.type = NLA_U16},
 145};
 146
 147
 148static void ib_sa_add_one(struct ib_device *device);
 149static void ib_sa_remove_one(struct ib_device *device, void *client_data);
 150
 151static struct ib_client sa_client = {
 152        .name   = "sa",
 153        .add    = ib_sa_add_one,
 154        .remove = ib_sa_remove_one
 155};
 156
 157static DEFINE_SPINLOCK(idr_lock);
 158static DEFINE_IDR(query_idr);
 159
 160static DEFINE_SPINLOCK(tid_lock);
 161static u32 tid;
 162
 163#define PATH_REC_FIELD(field) \
 164        .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),          \
 165        .struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,     \
 166        .field_name          = "sa_path_rec:" #field
 167
 168static const struct ib_field path_rec_table[] = {
 169        { PATH_REC_FIELD(service_id),
 170          .offset_words = 0,
 171          .offset_bits  = 0,
 172          .size_bits    = 64 },
 173        { PATH_REC_FIELD(dgid),
 174          .offset_words = 2,
 175          .offset_bits  = 0,
 176          .size_bits    = 128 },
 177        { PATH_REC_FIELD(sgid),
 178          .offset_words = 6,
 179          .offset_bits  = 0,
 180          .size_bits    = 128 },
 181        { PATH_REC_FIELD(dlid),
 182          .offset_words = 10,
 183          .offset_bits  = 0,
 184          .size_bits    = 16 },
 185        { PATH_REC_FIELD(slid),
 186          .offset_words = 10,
 187          .offset_bits  = 16,
 188          .size_bits    = 16 },
 189        { PATH_REC_FIELD(raw_traffic),
 190          .offset_words = 11,
 191          .offset_bits  = 0,
 192          .size_bits    = 1 },
 193        { RESERVED,
 194          .offset_words = 11,
 195          .offset_bits  = 1,
 196          .size_bits    = 3 },
 197        { PATH_REC_FIELD(flow_label),
 198          .offset_words = 11,
 199          .offset_bits  = 4,
 200          .size_bits    = 20 },
 201        { PATH_REC_FIELD(hop_limit),
 202          .offset_words = 11,
 203          .offset_bits  = 24,
 204          .size_bits    = 8 },
 205        { PATH_REC_FIELD(traffic_class),
 206          .offset_words = 12,
 207          .offset_bits  = 0,
 208          .size_bits    = 8 },
 209        { PATH_REC_FIELD(reversible),
 210          .offset_words = 12,
 211          .offset_bits  = 8,
 212          .size_bits    = 1 },
 213        { PATH_REC_FIELD(numb_path),
 214          .offset_words = 12,
 215          .offset_bits  = 9,
 216          .size_bits    = 7 },
 217        { PATH_REC_FIELD(pkey),
 218          .offset_words = 12,
 219          .offset_bits  = 16,
 220          .size_bits    = 16 },
 221        { PATH_REC_FIELD(qos_class),
 222          .offset_words = 13,
 223          .offset_bits  = 0,
 224          .size_bits    = 12 },
 225        { PATH_REC_FIELD(sl),
 226          .offset_words = 13,
 227          .offset_bits  = 12,
 228          .size_bits    = 4 },
 229        { PATH_REC_FIELD(mtu_selector),
 230          .offset_words = 13,
 231          .offset_bits  = 16,
 232          .size_bits    = 2 },
 233        { PATH_REC_FIELD(mtu),
 234          .offset_words = 13,
 235          .offset_bits  = 18,
 236          .size_bits    = 6 },
 237        { PATH_REC_FIELD(rate_selector),
 238          .offset_words = 13,
 239          .offset_bits  = 24,
 240          .size_bits    = 2 },
 241        { PATH_REC_FIELD(rate),
 242          .offset_words = 13,
 243          .offset_bits  = 26,
 244          .size_bits    = 6 },
 245        { PATH_REC_FIELD(packet_life_time_selector),
 246          .offset_words = 14,
 247          .offset_bits  = 0,
 248          .size_bits    = 2 },
 249        { PATH_REC_FIELD(packet_life_time),
 250          .offset_words = 14,
 251          .offset_bits  = 2,
 252          .size_bits    = 6 },
 253        { PATH_REC_FIELD(preference),
 254          .offset_words = 14,
 255          .offset_bits  = 8,
 256          .size_bits    = 8 },
 257        { RESERVED,
 258          .offset_words = 14,
 259          .offset_bits  = 16,
 260          .size_bits    = 48 },
 261};
 262
 263#define MCMEMBER_REC_FIELD(field) \
 264        .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),      \
 265        .struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \
 266        .field_name          = "sa_mcmember_rec:" #field
 267
 268static const struct ib_field mcmember_rec_table[] = {
 269        { MCMEMBER_REC_FIELD(mgid),
 270          .offset_words = 0,
 271          .offset_bits  = 0,
 272          .size_bits    = 128 },
 273        { MCMEMBER_REC_FIELD(port_gid),
 274          .offset_words = 4,
 275          .offset_bits  = 0,
 276          .size_bits    = 128 },
 277        { MCMEMBER_REC_FIELD(qkey),
 278          .offset_words = 8,
 279          .offset_bits  = 0,
 280          .size_bits    = 32 },
 281        { MCMEMBER_REC_FIELD(mlid),
 282          .offset_words = 9,
 283          .offset_bits  = 0,
 284          .size_bits    = 16 },
 285        { MCMEMBER_REC_FIELD(mtu_selector),
 286          .offset_words = 9,
 287          .offset_bits  = 16,
 288          .size_bits    = 2 },
 289        { MCMEMBER_REC_FIELD(mtu),
 290          .offset_words = 9,
 291          .offset_bits  = 18,
 292          .size_bits    = 6 },
 293        { MCMEMBER_REC_FIELD(traffic_class),
 294          .offset_words = 9,
 295          .offset_bits  = 24,
 296          .size_bits    = 8 },
 297        { MCMEMBER_REC_FIELD(pkey),
 298          .offset_words = 10,
 299          .offset_bits  = 0,
 300          .size_bits    = 16 },
 301        { MCMEMBER_REC_FIELD(rate_selector),
 302          .offset_words = 10,
 303          .offset_bits  = 16,
 304          .size_bits    = 2 },
 305        { MCMEMBER_REC_FIELD(rate),
 306          .offset_words = 10,
 307          .offset_bits  = 18,
 308          .size_bits    = 6 },
 309        { MCMEMBER_REC_FIELD(packet_life_time_selector),
 310          .offset_words = 10,
 311          .offset_bits  = 24,
 312          .size_bits    = 2 },
 313        { MCMEMBER_REC_FIELD(packet_life_time),
 314          .offset_words = 10,
 315          .offset_bits  = 26,
 316          .size_bits    = 6 },
 317        { MCMEMBER_REC_FIELD(sl),
 318          .offset_words = 11,
 319          .offset_bits  = 0,
 320          .size_bits    = 4 },
 321        { MCMEMBER_REC_FIELD(flow_label),
 322          .offset_words = 11,
 323          .offset_bits  = 4,
 324          .size_bits    = 20 },
 325        { MCMEMBER_REC_FIELD(hop_limit),
 326          .offset_words = 11,
 327          .offset_bits  = 24,
 328          .size_bits    = 8 },
 329        { MCMEMBER_REC_FIELD(scope),
 330          .offset_words = 12,
 331          .offset_bits  = 0,
 332          .size_bits    = 4 },
 333        { MCMEMBER_REC_FIELD(join_state),
 334          .offset_words = 12,
 335          .offset_bits  = 4,
 336          .size_bits    = 4 },
 337        { MCMEMBER_REC_FIELD(proxy_join),
 338          .offset_words = 12,
 339          .offset_bits  = 8,
 340          .size_bits    = 1 },
 341        { RESERVED,
 342          .offset_words = 12,
 343          .offset_bits  = 9,
 344          .size_bits    = 23 },
 345};
 346
 347#define SERVICE_REC_FIELD(field) \
 348        .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),       \
 349        .struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,  \
 350        .field_name          = "sa_service_rec:" #field
 351
 352static const struct ib_field service_rec_table[] = {
 353        { SERVICE_REC_FIELD(id),
 354          .offset_words = 0,
 355          .offset_bits  = 0,
 356          .size_bits    = 64 },
 357        { SERVICE_REC_FIELD(gid),
 358          .offset_words = 2,
 359          .offset_bits  = 0,
 360          .size_bits    = 128 },
 361        { SERVICE_REC_FIELD(pkey),
 362          .offset_words = 6,
 363          .offset_bits  = 0,
 364          .size_bits    = 16 },
 365        { SERVICE_REC_FIELD(lease),
 366          .offset_words = 7,
 367          .offset_bits  = 0,
 368          .size_bits    = 32 },
 369        { SERVICE_REC_FIELD(key),
 370          .offset_words = 8,
 371          .offset_bits  = 0,
 372          .size_bits    = 128 },
 373        { SERVICE_REC_FIELD(name),
 374          .offset_words = 12,
 375          .offset_bits  = 0,
 376          .size_bits    = 64*8 },
 377        { SERVICE_REC_FIELD(data8),
 378          .offset_words = 28,
 379          .offset_bits  = 0,
 380          .size_bits    = 16*8 },
 381        { SERVICE_REC_FIELD(data16),
 382          .offset_words = 32,
 383          .offset_bits  = 0,
 384          .size_bits    = 8*16 },
 385        { SERVICE_REC_FIELD(data32),
 386          .offset_words = 36,
 387          .offset_bits  = 0,
 388          .size_bits    = 4*32 },
 389        { SERVICE_REC_FIELD(data64),
 390          .offset_words = 40,
 391          .offset_bits  = 0,
 392          .size_bits    = 2*64 },
 393};
 394
 395#define GUIDINFO_REC_FIELD(field) \
 396        .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),      \
 397        .struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,  \
 398        .field_name          = "sa_guidinfo_rec:" #field
 399
 400static const struct ib_field guidinfo_rec_table[] = {
 401        { GUIDINFO_REC_FIELD(lid),
 402          .offset_words = 0,
 403          .offset_bits  = 0,
 404          .size_bits    = 16 },
 405        { GUIDINFO_REC_FIELD(block_num),
 406          .offset_words = 0,
 407          .offset_bits  = 16,
 408          .size_bits    = 8 },
 409        { GUIDINFO_REC_FIELD(res1),
 410          .offset_words = 0,
 411          .offset_bits  = 24,
 412          .size_bits    = 8 },
 413        { GUIDINFO_REC_FIELD(res2),
 414          .offset_words = 1,
 415          .offset_bits  = 0,
 416          .size_bits    = 32 },
 417        { GUIDINFO_REC_FIELD(guid_info_list),
 418          .offset_words = 2,
 419          .offset_bits  = 0,
 420          .size_bits    = 512 },
 421};
 422
 423static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
 424{
 425        query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
 426}
 427
 428static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
 429{
 430        return (query->flags & IB_SA_CANCEL);
 431}
 432
 433static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
 434                                     struct ib_sa_query *query)
 435{
 436        struct ib_sa_path_rec *sa_rec = query->mad_buf->context[1];
 437        struct ib_sa_mad *mad = query->mad_buf->mad;
 438        ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
 439        u16 val16;
 440        u64 val64;
 441        struct rdma_ls_resolve_header *header;
 442
 443        query->mad_buf->context[1] = NULL;
 444
 445        /* Construct the family header first */
 446        header = (struct rdma_ls_resolve_header *)
 447                skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 448        memcpy(header->device_name, query->port->agent->device->name,
 449               LS_DEVICE_NAME_MAX);
 450        header->port_num = query->port->port_num;
 451
 452        if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
 453            sa_rec->reversible != 0)
 454                query->path_use = LS_RESOLVE_PATH_USE_GMP;
 455        else
 456                query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
 457        header->path_use = query->path_use;
 458
 459        /* Now build the attributes */
 460        if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
 461                val64 = be64_to_cpu(sa_rec->service_id);
 462                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
 463                        sizeof(val64), &val64);
 464        }
 465        if (comp_mask & IB_SA_PATH_REC_DGID)
 466                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
 467                        sizeof(sa_rec->dgid), &sa_rec->dgid);
 468        if (comp_mask & IB_SA_PATH_REC_SGID)
 469                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
 470                        sizeof(sa_rec->sgid), &sa_rec->sgid);
 471        if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
 472                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
 473                        sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
 474
 475        if (comp_mask & IB_SA_PATH_REC_PKEY) {
 476                val16 = be16_to_cpu(sa_rec->pkey);
 477                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
 478                        sizeof(val16), &val16);
 479        }
 480        if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
 481                val16 = be16_to_cpu(sa_rec->qos_class);
 482                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
 483                        sizeof(val16), &val16);
 484        }
 485}
 486
 487static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
 488{
 489        int len = 0;
 490
 491        if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
 492                len += nla_total_size(sizeof(u64));
 493        if (comp_mask & IB_SA_PATH_REC_DGID)
 494                len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
 495        if (comp_mask & IB_SA_PATH_REC_SGID)
 496                len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
 497        if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
 498                len += nla_total_size(sizeof(u8));
 499        if (comp_mask & IB_SA_PATH_REC_PKEY)
 500                len += nla_total_size(sizeof(u16));
 501        if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
 502                len += nla_total_size(sizeof(u16));
 503
 504        /*
 505         * Make sure that at least some of the required comp_mask bits are
 506         * set.
 507         */
 508        if (WARN_ON(len == 0))
 509                return len;
 510
 511        /* Add the family header */
 512        len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
 513
 514        return len;
 515}
 516
 517static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
 518{
 519        struct sk_buff *skb = NULL;
 520        struct nlmsghdr *nlh;
 521        void *data;
 522        int ret = 0;
 523        struct ib_sa_mad *mad;
 524        int len;
 525
 526        mad = query->mad_buf->mad;
 527        len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
 528        if (len <= 0)
 529                return -EMSGSIZE;
 530
 531        skb = nlmsg_new(len, gfp_mask);
 532        if (!skb)
 533                return -ENOMEM;
 534
 535        /* Put nlmsg header only for now */
 536        data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
 537                            RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
 538        if (!data) {
 539                kfree_skb(skb);
 540                return -EMSGSIZE;
 541        }
 542
 543        /* Add attributes */
 544        ib_nl_set_path_rec_attrs(skb, query);
 545
 546        /* Repair the nlmsg header length */
 547        nlmsg_end(skb, nlh);
 548
 549        ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, gfp_mask);
 550        if (!ret)
 551                ret = len;
 552        else
 553                ret = 0;
 554
 555        return ret;
 556}
 557
 558static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
 559{
 560        unsigned long flags;
 561        unsigned long delay;
 562        int ret;
 563
 564        INIT_LIST_HEAD(&query->list);
 565        query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
 566
 567        /* Put the request on the list first.*/
 568        spin_lock_irqsave(&ib_nl_request_lock, flags);
 569        delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
 570        query->timeout = delay + jiffies;
 571        list_add_tail(&query->list, &ib_nl_request_list);
 572        /* Start the timeout if this is the only request */
 573        if (ib_nl_request_list.next == &query->list)
 574                queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
 575        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 576
 577        ret = ib_nl_send_msg(query, gfp_mask);
 578        if (ret <= 0) {
 579                ret = -EIO;
 580                /* Remove the request */
 581                spin_lock_irqsave(&ib_nl_request_lock, flags);
 582                list_del(&query->list);
 583                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 584        } else {
 585                ret = 0;
 586        }
 587
 588        return ret;
 589}
 590
 591static int ib_nl_cancel_request(struct ib_sa_query *query)
 592{
 593        unsigned long flags;
 594        struct ib_sa_query *wait_query;
 595        int found = 0;
 596
 597        spin_lock_irqsave(&ib_nl_request_lock, flags);
 598        list_for_each_entry(wait_query, &ib_nl_request_list, list) {
 599                /* Let the timeout to take care of the callback */
 600                if (query == wait_query) {
 601                        query->flags |= IB_SA_CANCEL;
 602                        query->timeout = jiffies;
 603                        list_move(&query->list, &ib_nl_request_list);
 604                        found = 1;
 605                        mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
 606                        break;
 607                }
 608        }
 609        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 610
 611        return found;
 612}
 613
 614static void send_handler(struct ib_mad_agent *agent,
 615                         struct ib_mad_send_wc *mad_send_wc);
 616
 617static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
 618                                           const struct nlmsghdr *nlh)
 619{
 620        struct ib_mad_send_wc mad_send_wc;
 621        struct ib_sa_mad *mad = NULL;
 622        const struct nlattr *head, *curr;
 623        struct ib_path_rec_data  *rec;
 624        int len, rem;
 625        u32 mask = 0;
 626        int status = -EIO;
 627
 628        if (query->callback) {
 629                head = (const struct nlattr *) nlmsg_data(nlh);
 630                len = nlmsg_len(nlh);
 631                switch (query->path_use) {
 632                case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
 633                        mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
 634                        break;
 635
 636                case LS_RESOLVE_PATH_USE_ALL:
 637                case LS_RESOLVE_PATH_USE_GMP:
 638                default:
 639                        mask = IB_PATH_PRIMARY | IB_PATH_GMP |
 640                                IB_PATH_BIDIRECTIONAL;
 641                        break;
 642                }
 643                nla_for_each_attr(curr, head, len, rem) {
 644                        if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
 645                                rec = nla_data(curr);
 646                                /*
 647                                 * Get the first one. In the future, we may
 648                                 * need to get up to 6 pathrecords.
 649                                 */
 650                                if ((rec->flags & mask) == mask) {
 651                                        mad = query->mad_buf->mad;
 652                                        mad->mad_hdr.method |=
 653                                                IB_MGMT_METHOD_RESP;
 654                                        memcpy(mad->data, rec->path_rec,
 655                                               sizeof(rec->path_rec));
 656                                        status = 0;
 657                                        break;
 658                                }
 659                        }
 660                }
 661                query->callback(query, status, mad);
 662        }
 663
 664        mad_send_wc.send_buf = query->mad_buf;
 665        mad_send_wc.status = IB_WC_SUCCESS;
 666        send_handler(query->mad_buf->mad_agent, &mad_send_wc);
 667}
 668
 669static void ib_nl_request_timeout(struct work_struct *work)
 670{
 671        unsigned long flags;
 672        struct ib_sa_query *query;
 673        unsigned long delay;
 674        struct ib_mad_send_wc mad_send_wc;
 675        int ret;
 676
 677        spin_lock_irqsave(&ib_nl_request_lock, flags);
 678        while (!list_empty(&ib_nl_request_list)) {
 679                query = list_entry(ib_nl_request_list.next,
 680                                   struct ib_sa_query, list);
 681
 682                if (time_after(query->timeout, jiffies)) {
 683                        delay = query->timeout - jiffies;
 684                        if ((long)delay <= 0)
 685                                delay = 1;
 686                        queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
 687                        break;
 688                }
 689
 690                list_del(&query->list);
 691                ib_sa_disable_local_svc(query);
 692                /* Hold the lock to protect against query cancellation */
 693                if (ib_sa_query_cancelled(query))
 694                        ret = -1;
 695                else
 696                        ret = ib_post_send_mad(query->mad_buf, NULL);
 697                if (ret) {
 698                        mad_send_wc.send_buf = query->mad_buf;
 699                        mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
 700                        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 701                        send_handler(query->port->agent, &mad_send_wc);
 702                        spin_lock_irqsave(&ib_nl_request_lock, flags);
 703                }
 704        }
 705        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 706}
 707
 708static int ib_nl_handle_set_timeout(struct sk_buff *skb,
 709                                    struct netlink_callback *cb)
 710{
 711        const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
 712        int timeout, delta, abs_delta;
 713        const struct nlattr *attr;
 714        unsigned long flags;
 715        struct ib_sa_query *query;
 716        long delay = 0;
 717        struct nlattr *tb[LS_NLA_TYPE_MAX];
 718        int ret;
 719
 720        if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
 721            !(NETLINK_CB(skb).sk) ||
 722            !netlink_capable(skb, CAP_NET_ADMIN))
 723                return -EPERM;
 724
 725        ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
 726                        nlmsg_len(nlh), ib_nl_policy);
 727        attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
 728        if (ret || !attr)
 729                goto settimeout_out;
 730
 731        timeout = *(int *) nla_data(attr);
 732        if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
 733                timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
 734        if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
 735                timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
 736
 737        delta = timeout - sa_local_svc_timeout_ms;
 738        if (delta < 0)
 739                abs_delta = -delta;
 740        else
 741                abs_delta = delta;
 742
 743        if (delta != 0) {
 744                spin_lock_irqsave(&ib_nl_request_lock, flags);
 745                sa_local_svc_timeout_ms = timeout;
 746                list_for_each_entry(query, &ib_nl_request_list, list) {
 747                        if (delta < 0 && abs_delta > query->timeout)
 748                                query->timeout = 0;
 749                        else
 750                                query->timeout += delta;
 751
 752                        /* Get the new delay from the first entry */
 753                        if (!delay) {
 754                                delay = query->timeout - jiffies;
 755                                if (delay <= 0)
 756                                        delay = 1;
 757                        }
 758                }
 759                if (delay)
 760                        mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
 761                                         (unsigned long)delay);
 762                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 763        }
 764
 765settimeout_out:
 766        return skb->len;
 767}
 768
 769static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
 770{
 771        struct nlattr *tb[LS_NLA_TYPE_MAX];
 772        int ret;
 773
 774        if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
 775                return 0;
 776
 777        ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
 778                        nlmsg_len(nlh), ib_nl_policy);
 779        if (ret)
 780                return 0;
 781
 782        return 1;
 783}
 784
 785static int ib_nl_handle_resolve_resp(struct sk_buff *skb,
 786                                     struct netlink_callback *cb)
 787{
 788        const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
 789        unsigned long flags;
 790        struct ib_sa_query *query;
 791        struct ib_mad_send_buf *send_buf;
 792        struct ib_mad_send_wc mad_send_wc;
 793        int found = 0;
 794        int ret;
 795
 796        if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
 797            !(NETLINK_CB(skb).sk) ||
 798            !netlink_capable(skb, CAP_NET_ADMIN))
 799                return -EPERM;
 800
 801        spin_lock_irqsave(&ib_nl_request_lock, flags);
 802        list_for_each_entry(query, &ib_nl_request_list, list) {
 803                /*
 804                 * If the query is cancelled, let the timeout routine
 805                 * take care of it.
 806                 */
 807                if (nlh->nlmsg_seq == query->seq) {
 808                        found = !ib_sa_query_cancelled(query);
 809                        if (found)
 810                                list_del(&query->list);
 811                        break;
 812                }
 813        }
 814
 815        if (!found) {
 816                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 817                goto resp_out;
 818        }
 819
 820        send_buf = query->mad_buf;
 821
 822        if (!ib_nl_is_good_resolve_resp(nlh)) {
 823                /* if the result is a failure, send out the packet via IB */
 824                ib_sa_disable_local_svc(query);
 825                ret = ib_post_send_mad(query->mad_buf, NULL);
 826                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 827                if (ret) {
 828                        mad_send_wc.send_buf = send_buf;
 829                        mad_send_wc.status = IB_WC_GENERAL_ERR;
 830                        send_handler(query->port->agent, &mad_send_wc);
 831                }
 832        } else {
 833                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 834                ib_nl_process_good_resolve_rsp(query, nlh);
 835        }
 836
 837resp_out:
 838        return skb->len;
 839}
 840
 841static struct ibnl_client_cbs ib_sa_cb_table[] = {
 842        [RDMA_NL_LS_OP_RESOLVE] = {
 843                .dump = ib_nl_handle_resolve_resp,
 844                .module = THIS_MODULE },
 845        [RDMA_NL_LS_OP_SET_TIMEOUT] = {
 846                .dump = ib_nl_handle_set_timeout,
 847                .module = THIS_MODULE },
 848};
 849
 850static void free_sm_ah(struct kref *kref)
 851{
 852        struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
 853
 854        ib_destroy_ah(sm_ah->ah);
 855        kfree(sm_ah);
 856}
 857
 858static void update_sm_ah(struct work_struct *work)
 859{
 860        struct ib_sa_port *port =
 861                container_of(work, struct ib_sa_port, update_task);
 862        struct ib_sa_sm_ah *new_ah;
 863        struct ib_port_attr port_attr;
 864        struct ib_ah_attr   ah_attr;
 865
 866        if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
 867                pr_warn("Couldn't query port\n");
 868                return;
 869        }
 870
 871        new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
 872        if (!new_ah) {
 873                return;
 874        }
 875
 876        kref_init(&new_ah->ref);
 877        new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
 878
 879        new_ah->pkey_index = 0;
 880        if (ib_find_pkey(port->agent->device, port->port_num,
 881                         IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
 882                pr_err("Couldn't find index for default PKey\n");
 883
 884        memset(&ah_attr, 0, sizeof ah_attr);
 885        ah_attr.dlid     = port_attr.sm_lid;
 886        ah_attr.sl       = port_attr.sm_sl;
 887        ah_attr.port_num = port->port_num;
 888        if (port_attr.grh_required) {
 889                ah_attr.ah_flags = IB_AH_GRH;
 890                ah_attr.grh.dgid.global.subnet_prefix = cpu_to_be64(port_attr.subnet_prefix);
 891                ah_attr.grh.dgid.global.interface_id = cpu_to_be64(IB_SA_WELL_KNOWN_GUID);
 892        }
 893
 894        new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
 895        if (IS_ERR(new_ah->ah)) {
 896                pr_warn("Couldn't create new SM AH\n");
 897                kfree(new_ah);
 898                return;
 899        }
 900
 901        spin_lock_irq(&port->ah_lock);
 902        if (port->sm_ah)
 903                kref_put(&port->sm_ah->ref, free_sm_ah);
 904        port->sm_ah = new_ah;
 905        spin_unlock_irq(&port->ah_lock);
 906
 907}
 908
 909static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
 910{
 911        if (event->event == IB_EVENT_PORT_ERR    ||
 912            event->event == IB_EVENT_PORT_ACTIVE ||
 913            event->event == IB_EVENT_LID_CHANGE  ||
 914            event->event == IB_EVENT_PKEY_CHANGE ||
 915            event->event == IB_EVENT_SM_CHANGE   ||
 916            event->event == IB_EVENT_CLIENT_REREGISTER) {
 917                unsigned long flags;
 918                struct ib_sa_device *sa_dev =
 919                        container_of(handler, typeof(*sa_dev), event_handler);
 920                struct ib_sa_port *port =
 921                        &sa_dev->port[event->element.port_num - sa_dev->start_port];
 922
 923                if (!rdma_cap_ib_sa(handler->device, port->port_num))
 924                        return;
 925
 926                spin_lock_irqsave(&port->ah_lock, flags);
 927                if (port->sm_ah)
 928                        kref_put(&port->sm_ah->ref, free_sm_ah);
 929                port->sm_ah = NULL;
 930                spin_unlock_irqrestore(&port->ah_lock, flags);
 931
 932                queue_work(ib_wq, &sa_dev->port[event->element.port_num -
 933                                            sa_dev->start_port].update_task);
 934        }
 935}
 936
 937void ib_sa_register_client(struct ib_sa_client *client)
 938{
 939        atomic_set(&client->users, 1);
 940        init_completion(&client->comp);
 941}
 942EXPORT_SYMBOL(ib_sa_register_client);
 943
 944void ib_sa_unregister_client(struct ib_sa_client *client)
 945{
 946        ib_sa_client_put(client);
 947        wait_for_completion(&client->comp);
 948}
 949EXPORT_SYMBOL(ib_sa_unregister_client);
 950
 951/**
 952 * ib_sa_cancel_query - try to cancel an SA query
 953 * @id:ID of query to cancel
 954 * @query:query pointer to cancel
 955 *
 956 * Try to cancel an SA query.  If the id and query don't match up or
 957 * the query has already completed, nothing is done.  Otherwise the
 958 * query is canceled and will complete with a status of -EINTR.
 959 */
 960void ib_sa_cancel_query(int id, struct ib_sa_query *query)
 961{
 962        unsigned long flags;
 963        struct ib_mad_agent *agent;
 964        struct ib_mad_send_buf *mad_buf;
 965
 966        spin_lock_irqsave(&idr_lock, flags);
 967        if (idr_find(&query_idr, id) != query) {
 968                spin_unlock_irqrestore(&idr_lock, flags);
 969                return;
 970        }
 971        agent = query->port->agent;
 972        mad_buf = query->mad_buf;
 973        spin_unlock_irqrestore(&idr_lock, flags);
 974
 975        /*
 976         * If the query is still on the netlink request list, schedule
 977         * it to be cancelled by the timeout routine. Otherwise, it has been
 978         * sent to the MAD layer and has to be cancelled from there.
 979         */
 980        if (!ib_nl_cancel_request(query))
 981                ib_cancel_mad(agent, mad_buf);
 982}
 983EXPORT_SYMBOL(ib_sa_cancel_query);
 984
 985static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
 986{
 987        struct ib_sa_device *sa_dev;
 988        struct ib_sa_port   *port;
 989        unsigned long flags;
 990        u8 src_path_mask;
 991
 992        sa_dev = ib_get_client_data(device, &sa_client);
 993        if (!sa_dev)
 994                return 0x7f;
 995
 996        port  = &sa_dev->port[port_num - sa_dev->start_port];
 997        spin_lock_irqsave(&port->ah_lock, flags);
 998        src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
 999        spin_unlock_irqrestore(&port->ah_lock, flags);
1000
1001        return src_path_mask;
1002}
1003
1004int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
1005                         struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
1006{
1007        int ret;
1008        u16 gid_index;
1009        int use_roce;
1010        struct net_device *ndev = NULL;
1011
1012        memset(ah_attr, 0, sizeof *ah_attr);
1013        ah_attr->dlid = be16_to_cpu(rec->dlid);
1014        ah_attr->sl = rec->sl;
1015        ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
1016                                 get_src_path_mask(device, port_num);
1017        ah_attr->port_num = port_num;
1018        ah_attr->static_rate = rec->rate;
1019
1020        use_roce = rdma_cap_eth_ah(device, port_num);
1021
1022        if (use_roce) {
1023                struct net_device *idev;
1024                struct net_device *resolved_dev;
1025                struct rdma_dev_addr dev_addr = {.bound_dev_if = rec->ifindex,
1026                                                 .net = rec->net ? rec->net :
1027                                                         &init_net};
1028                union {
1029                        struct sockaddr     _sockaddr;
1030                        struct sockaddr_in  _sockaddr_in;
1031                        struct sockaddr_in6 _sockaddr_in6;
1032                } sgid_addr, dgid_addr;
1033
1034                if (!device->get_netdev)
1035                        return -EOPNOTSUPP;
1036
1037                rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1038                rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1039
1040                /* validate the route */
1041                ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1042                                            &dgid_addr._sockaddr, &dev_addr);
1043                if (ret)
1044                        return ret;
1045
1046                if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1047                     dev_addr.network == RDMA_NETWORK_IPV6) &&
1048                    rec->gid_type != IB_GID_TYPE_ROCE_UDP_ENCAP)
1049                        return -EINVAL;
1050
1051                idev = device->get_netdev(device, port_num);
1052                if (!idev)
1053                        return -ENODEV;
1054
1055                resolved_dev = dev_get_by_index(dev_addr.net,
1056                                                dev_addr.bound_dev_if);
1057                if (resolved_dev->flags & IFF_LOOPBACK) {
1058                        dev_put(resolved_dev);
1059                        resolved_dev = idev;
1060                        dev_hold(resolved_dev);
1061                }
1062                ndev = ib_get_ndev_from_path(rec);
1063                rcu_read_lock();
1064                if ((ndev && ndev != resolved_dev) ||
1065                    (resolved_dev != idev &&
1066                     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1067                        ret = -EHOSTUNREACH;
1068                rcu_read_unlock();
1069                dev_put(idev);
1070                dev_put(resolved_dev);
1071                if (ret) {
1072                        if (ndev)
1073                                dev_put(ndev);
1074                        return ret;
1075                }
1076        }
1077
1078        if (rec->hop_limit > 0 || use_roce) {
1079                ah_attr->ah_flags = IB_AH_GRH;
1080                ah_attr->grh.dgid = rec->dgid;
1081
1082                ret = ib_find_cached_gid_by_port(device, &rec->sgid,
1083                                                 rec->gid_type, port_num, ndev,
1084                                                 &gid_index);
1085                if (ret) {
1086                        if (ndev)
1087                                dev_put(ndev);
1088                        return ret;
1089                }
1090
1091                ah_attr->grh.sgid_index    = gid_index;
1092                ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
1093                ah_attr->grh.hop_limit     = rec->hop_limit;
1094                ah_attr->grh.traffic_class = rec->traffic_class;
1095                if (ndev)
1096                        dev_put(ndev);
1097        }
1098
1099        if (use_roce)
1100                memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
1101
1102        return 0;
1103}
1104EXPORT_SYMBOL(ib_init_ah_from_path);
1105
1106static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1107{
1108        unsigned long flags;
1109
1110        spin_lock_irqsave(&query->port->ah_lock, flags);
1111        if (!query->port->sm_ah) {
1112                spin_unlock_irqrestore(&query->port->ah_lock, flags);
1113                return -EAGAIN;
1114        }
1115        kref_get(&query->port->sm_ah->ref);
1116        query->sm_ah = query->port->sm_ah;
1117        spin_unlock_irqrestore(&query->port->ah_lock, flags);
1118
1119        query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1120                                            query->sm_ah->pkey_index,
1121                                            0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1122                                            gfp_mask,
1123                                            IB_MGMT_BASE_VERSION);
1124        if (IS_ERR(query->mad_buf)) {
1125                kref_put(&query->sm_ah->ref, free_sm_ah);
1126                return -ENOMEM;
1127        }
1128
1129        query->mad_buf->ah = query->sm_ah->ah;
1130
1131        return 0;
1132}
1133
1134static void free_mad(struct ib_sa_query *query)
1135{
1136        ib_free_send_mad(query->mad_buf);
1137        kref_put(&query->sm_ah->ref, free_sm_ah);
1138}
1139
1140static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
1141{
1142        unsigned long flags;
1143
1144        memset(mad, 0, sizeof *mad);
1145
1146        mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1147        mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1148        mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1149
1150        spin_lock_irqsave(&tid_lock, flags);
1151        mad->mad_hdr.tid           =
1152                cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1153        spin_unlock_irqrestore(&tid_lock, flags);
1154}
1155
1156static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1157{
1158        bool preload = gfpflags_allow_blocking(gfp_mask);
1159        unsigned long flags;
1160        int ret, id;
1161
1162        if (preload)
1163                idr_preload(gfp_mask);
1164        spin_lock_irqsave(&idr_lock, flags);
1165
1166        id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1167
1168        spin_unlock_irqrestore(&idr_lock, flags);
1169        if (preload)
1170                idr_preload_end();
1171        if (id < 0)
1172                return id;
1173
1174        query->mad_buf->timeout_ms  = timeout_ms;
1175        query->mad_buf->context[0] = query;
1176        query->id = id;
1177
1178        if (query->flags & IB_SA_ENABLE_LOCAL_SERVICE) {
1179                if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1180                        if (!ib_nl_make_request(query, gfp_mask))
1181                                return id;
1182                }
1183                ib_sa_disable_local_svc(query);
1184        }
1185
1186        ret = ib_post_send_mad(query->mad_buf, NULL);
1187        if (ret) {
1188                spin_lock_irqsave(&idr_lock, flags);
1189                idr_remove(&query_idr, id);
1190                spin_unlock_irqrestore(&idr_lock, flags);
1191        }
1192
1193        /*
1194         * It's not safe to dereference query any more, because the
1195         * send may already have completed and freed the query in
1196         * another context.
1197         */
1198        return ret ? ret : id;
1199}
1200
1201void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
1202{
1203        ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1204}
1205EXPORT_SYMBOL(ib_sa_unpack_path);
1206
1207void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
1208{
1209        ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1210}
1211EXPORT_SYMBOL(ib_sa_pack_path);
1212
1213static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1214                                    int status,
1215                                    struct ib_sa_mad *mad)
1216{
1217        struct ib_sa_path_query *query =
1218                container_of(sa_query, struct ib_sa_path_query, sa_query);
1219
1220        if (mad) {
1221                struct ib_sa_path_rec rec;
1222
1223                ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1224                          mad->data, &rec);
1225                rec.net = NULL;
1226                rec.ifindex = 0;
1227                rec.gid_type = IB_GID_TYPE_IB;
1228                eth_zero_addr(rec.dmac);
1229                query->callback(status, &rec, query->context);
1230        } else
1231                query->callback(status, NULL, query->context);
1232}
1233
1234static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1235{
1236        kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
1237}
1238
1239/**
1240 * ib_sa_path_rec_get - Start a Path get query
1241 * @client:SA client
1242 * @device:device to send query on
1243 * @port_num: port number to send query on
1244 * @rec:Path Record to send in query
1245 * @comp_mask:component mask to send in query
1246 * @timeout_ms:time to wait for response
1247 * @gfp_mask:GFP mask to use for internal allocations
1248 * @callback:function called when query completes, times out or is
1249 * canceled
1250 * @context:opaque user context passed to callback
1251 * @sa_query:query context, used to cancel query
1252 *
1253 * Send a Path Record Get query to the SA to look up a path.  The
1254 * callback function will be called when the query completes (or
1255 * fails); status is 0 for a successful response, -EINTR if the query
1256 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1257 * occurred sending the query.  The resp parameter of the callback is
1258 * only valid if status is 0.
1259 *
1260 * If the return value of ib_sa_path_rec_get() is negative, it is an
1261 * error code.  Otherwise it is a query ID that can be used to cancel
1262 * the query.
1263 */
1264int ib_sa_path_rec_get(struct ib_sa_client *client,
1265                       struct ib_device *device, u8 port_num,
1266                       struct ib_sa_path_rec *rec,
1267                       ib_sa_comp_mask comp_mask,
1268                       int timeout_ms, gfp_t gfp_mask,
1269                       void (*callback)(int status,
1270                                        struct ib_sa_path_rec *resp,
1271                                        void *context),
1272                       void *context,
1273                       struct ib_sa_query **sa_query)
1274{
1275        struct ib_sa_path_query *query;
1276        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1277        struct ib_sa_port   *port;
1278        struct ib_mad_agent *agent;
1279        struct ib_sa_mad *mad;
1280        int ret;
1281
1282        if (!sa_dev)
1283                return -ENODEV;
1284
1285        port  = &sa_dev->port[port_num - sa_dev->start_port];
1286        agent = port->agent;
1287
1288        query = kzalloc(sizeof(*query), gfp_mask);
1289        if (!query)
1290                return -ENOMEM;
1291
1292        query->sa_query.port     = port;
1293        ret = alloc_mad(&query->sa_query, gfp_mask);
1294        if (ret)
1295                goto err1;
1296
1297        ib_sa_client_get(client);
1298        query->sa_query.client = client;
1299        query->callback        = callback;
1300        query->context         = context;
1301
1302        mad = query->sa_query.mad_buf->mad;
1303        init_mad(mad, agent);
1304
1305        query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1306        query->sa_query.release  = ib_sa_path_rec_release;
1307        mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1308        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1309        mad->sa_hdr.comp_mask    = comp_mask;
1310
1311        ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
1312
1313        *sa_query = &query->sa_query;
1314
1315        query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1316        query->sa_query.mad_buf->context[1] = rec;
1317
1318        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1319        if (ret < 0)
1320                goto err2;
1321
1322        return ret;
1323
1324err2:
1325        *sa_query = NULL;
1326        ib_sa_client_put(query->sa_query.client);
1327        free_mad(&query->sa_query);
1328
1329err1:
1330        kfree(query);
1331        return ret;
1332}
1333EXPORT_SYMBOL(ib_sa_path_rec_get);
1334
1335static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1336                                    int status,
1337                                    struct ib_sa_mad *mad)
1338{
1339        struct ib_sa_service_query *query =
1340                container_of(sa_query, struct ib_sa_service_query, sa_query);
1341
1342        if (mad) {
1343                struct ib_sa_service_rec rec;
1344
1345                ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1346                          mad->data, &rec);
1347                query->callback(status, &rec, query->context);
1348        } else
1349                query->callback(status, NULL, query->context);
1350}
1351
1352static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1353{
1354        kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1355}
1356
1357/**
1358 * ib_sa_service_rec_query - Start Service Record operation
1359 * @client:SA client
1360 * @device:device to send request on
1361 * @port_num: port number to send request on
1362 * @method:SA method - should be get, set, or delete
1363 * @rec:Service Record to send in request
1364 * @comp_mask:component mask to send in request
1365 * @timeout_ms:time to wait for response
1366 * @gfp_mask:GFP mask to use for internal allocations
1367 * @callback:function called when request completes, times out or is
1368 * canceled
1369 * @context:opaque user context passed to callback
1370 * @sa_query:request context, used to cancel request
1371 *
1372 * Send a Service Record set/get/delete to the SA to register,
1373 * unregister or query a service record.
1374 * The callback function will be called when the request completes (or
1375 * fails); status is 0 for a successful response, -EINTR if the query
1376 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1377 * occurred sending the query.  The resp parameter of the callback is
1378 * only valid if status is 0.
1379 *
1380 * If the return value of ib_sa_service_rec_query() is negative, it is an
1381 * error code.  Otherwise it is a request ID that can be used to cancel
1382 * the query.
1383 */
1384int ib_sa_service_rec_query(struct ib_sa_client *client,
1385                            struct ib_device *device, u8 port_num, u8 method,
1386                            struct ib_sa_service_rec *rec,
1387                            ib_sa_comp_mask comp_mask,
1388                            int timeout_ms, gfp_t gfp_mask,
1389                            void (*callback)(int status,
1390                                             struct ib_sa_service_rec *resp,
1391                                             void *context),
1392                            void *context,
1393                            struct ib_sa_query **sa_query)
1394{
1395        struct ib_sa_service_query *query;
1396        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1397        struct ib_sa_port   *port;
1398        struct ib_mad_agent *agent;
1399        struct ib_sa_mad *mad;
1400        int ret;
1401
1402        if (!sa_dev)
1403                return -ENODEV;
1404
1405        port  = &sa_dev->port[port_num - sa_dev->start_port];
1406        agent = port->agent;
1407
1408        if (method != IB_MGMT_METHOD_GET &&
1409            method != IB_MGMT_METHOD_SET &&
1410            method != IB_SA_METHOD_DELETE)
1411                return -EINVAL;
1412
1413        query = kzalloc(sizeof(*query), gfp_mask);
1414        if (!query)
1415                return -ENOMEM;
1416
1417        query->sa_query.port     = port;
1418        ret = alloc_mad(&query->sa_query, gfp_mask);
1419        if (ret)
1420                goto err1;
1421
1422        ib_sa_client_get(client);
1423        query->sa_query.client = client;
1424        query->callback        = callback;
1425        query->context         = context;
1426
1427        mad = query->sa_query.mad_buf->mad;
1428        init_mad(mad, agent);
1429
1430        query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1431        query->sa_query.release  = ib_sa_service_rec_release;
1432        mad->mad_hdr.method      = method;
1433        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1434        mad->sa_hdr.comp_mask    = comp_mask;
1435
1436        ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1437                rec, mad->data);
1438
1439        *sa_query = &query->sa_query;
1440
1441        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1442        if (ret < 0)
1443                goto err2;
1444
1445        return ret;
1446
1447err2:
1448        *sa_query = NULL;
1449        ib_sa_client_put(query->sa_query.client);
1450        free_mad(&query->sa_query);
1451
1452err1:
1453        kfree(query);
1454        return ret;
1455}
1456EXPORT_SYMBOL(ib_sa_service_rec_query);
1457
1458static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1459                                        int status,
1460                                        struct ib_sa_mad *mad)
1461{
1462        struct ib_sa_mcmember_query *query =
1463                container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1464
1465        if (mad) {
1466                struct ib_sa_mcmember_rec rec;
1467
1468                ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1469                          mad->data, &rec);
1470                query->callback(status, &rec, query->context);
1471        } else
1472                query->callback(status, NULL, query->context);
1473}
1474
1475static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1476{
1477        kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1478}
1479
1480int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1481                             struct ib_device *device, u8 port_num,
1482                             u8 method,
1483                             struct ib_sa_mcmember_rec *rec,
1484                             ib_sa_comp_mask comp_mask,
1485                             int timeout_ms, gfp_t gfp_mask,
1486                             void (*callback)(int status,
1487                                              struct ib_sa_mcmember_rec *resp,
1488                                              void *context),
1489                             void *context,
1490                             struct ib_sa_query **sa_query)
1491{
1492        struct ib_sa_mcmember_query *query;
1493        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1494        struct ib_sa_port   *port;
1495        struct ib_mad_agent *agent;
1496        struct ib_sa_mad *mad;
1497        int ret;
1498
1499        if (!sa_dev)
1500                return -ENODEV;
1501
1502        port  = &sa_dev->port[port_num - sa_dev->start_port];
1503        agent = port->agent;
1504
1505        query = kzalloc(sizeof(*query), gfp_mask);
1506        if (!query)
1507                return -ENOMEM;
1508
1509        query->sa_query.port     = port;
1510        ret = alloc_mad(&query->sa_query, gfp_mask);
1511        if (ret)
1512                goto err1;
1513
1514        ib_sa_client_get(client);
1515        query->sa_query.client = client;
1516        query->callback        = callback;
1517        query->context         = context;
1518
1519        mad = query->sa_query.mad_buf->mad;
1520        init_mad(mad, agent);
1521
1522        query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1523        query->sa_query.release  = ib_sa_mcmember_rec_release;
1524        mad->mad_hdr.method      = method;
1525        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1526        mad->sa_hdr.comp_mask    = comp_mask;
1527
1528        ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1529                rec, mad->data);
1530
1531        *sa_query = &query->sa_query;
1532
1533        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1534        if (ret < 0)
1535                goto err2;
1536
1537        return ret;
1538
1539err2:
1540        *sa_query = NULL;
1541        ib_sa_client_put(query->sa_query.client);
1542        free_mad(&query->sa_query);
1543
1544err1:
1545        kfree(query);
1546        return ret;
1547}
1548
1549/* Support GuidInfoRecord */
1550static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1551                                        int status,
1552                                        struct ib_sa_mad *mad)
1553{
1554        struct ib_sa_guidinfo_query *query =
1555                container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1556
1557        if (mad) {
1558                struct ib_sa_guidinfo_rec rec;
1559
1560                ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1561                          mad->data, &rec);
1562                query->callback(status, &rec, query->context);
1563        } else
1564                query->callback(status, NULL, query->context);
1565}
1566
1567static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1568{
1569        kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1570}
1571
1572int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1573                              struct ib_device *device, u8 port_num,
1574                              struct ib_sa_guidinfo_rec *rec,
1575                              ib_sa_comp_mask comp_mask, u8 method,
1576                              int timeout_ms, gfp_t gfp_mask,
1577                              void (*callback)(int status,
1578                                               struct ib_sa_guidinfo_rec *resp,
1579                                               void *context),
1580                              void *context,
1581                              struct ib_sa_query **sa_query)
1582{
1583        struct ib_sa_guidinfo_query *query;
1584        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1585        struct ib_sa_port *port;
1586        struct ib_mad_agent *agent;
1587        struct ib_sa_mad *mad;
1588        int ret;
1589
1590        if (!sa_dev)
1591                return -ENODEV;
1592
1593        if (method != IB_MGMT_METHOD_GET &&
1594            method != IB_MGMT_METHOD_SET &&
1595            method != IB_SA_METHOD_DELETE) {
1596                return -EINVAL;
1597        }
1598
1599        port  = &sa_dev->port[port_num - sa_dev->start_port];
1600        agent = port->agent;
1601
1602        query = kzalloc(sizeof(*query), gfp_mask);
1603        if (!query)
1604                return -ENOMEM;
1605
1606        query->sa_query.port = port;
1607        ret = alloc_mad(&query->sa_query, gfp_mask);
1608        if (ret)
1609                goto err1;
1610
1611        ib_sa_client_get(client);
1612        query->sa_query.client = client;
1613        query->callback        = callback;
1614        query->context         = context;
1615
1616        mad = query->sa_query.mad_buf->mad;
1617        init_mad(mad, agent);
1618
1619        query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1620        query->sa_query.release  = ib_sa_guidinfo_rec_release;
1621
1622        mad->mad_hdr.method      = method;
1623        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1624        mad->sa_hdr.comp_mask    = comp_mask;
1625
1626        ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1627                mad->data);
1628
1629        *sa_query = &query->sa_query;
1630
1631        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1632        if (ret < 0)
1633                goto err2;
1634
1635        return ret;
1636
1637err2:
1638        *sa_query = NULL;
1639        ib_sa_client_put(query->sa_query.client);
1640        free_mad(&query->sa_query);
1641
1642err1:
1643        kfree(query);
1644        return ret;
1645}
1646EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1647
1648static void send_handler(struct ib_mad_agent *agent,
1649                         struct ib_mad_send_wc *mad_send_wc)
1650{
1651        struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1652        unsigned long flags;
1653
1654        if (query->callback)
1655                switch (mad_send_wc->status) {
1656                case IB_WC_SUCCESS:
1657                        /* No callback -- already got recv */
1658                        break;
1659                case IB_WC_RESP_TIMEOUT_ERR:
1660                        query->callback(query, -ETIMEDOUT, NULL);
1661                        break;
1662                case IB_WC_WR_FLUSH_ERR:
1663                        query->callback(query, -EINTR, NULL);
1664                        break;
1665                default:
1666                        query->callback(query, -EIO, NULL);
1667                        break;
1668                }
1669
1670        spin_lock_irqsave(&idr_lock, flags);
1671        idr_remove(&query_idr, query->id);
1672        spin_unlock_irqrestore(&idr_lock, flags);
1673
1674        free_mad(query);
1675        ib_sa_client_put(query->client);
1676        query->release(query);
1677}
1678
1679static void recv_handler(struct ib_mad_agent *mad_agent,
1680                         struct ib_mad_send_buf *send_buf,
1681                         struct ib_mad_recv_wc *mad_recv_wc)
1682{
1683        struct ib_sa_query *query;
1684
1685        if (!send_buf)
1686                return;
1687
1688        query = send_buf->context[0];
1689        if (query->callback) {
1690                if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1691                        query->callback(query,
1692                                        mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1693                                        -EINVAL : 0,
1694                                        (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
1695                else
1696                        query->callback(query, -EIO, NULL);
1697        }
1698
1699        ib_free_recv_mad(mad_recv_wc);
1700}
1701
1702static void ib_sa_add_one(struct ib_device *device)
1703{
1704        struct ib_sa_device *sa_dev;
1705        int s, e, i;
1706        int count = 0;
1707
1708        s = rdma_start_port(device);
1709        e = rdma_end_port(device);
1710
1711        sa_dev = kzalloc(sizeof *sa_dev +
1712                         (e - s + 1) * sizeof (struct ib_sa_port),
1713                         GFP_KERNEL);
1714        if (!sa_dev)
1715                return;
1716
1717        sa_dev->start_port = s;
1718        sa_dev->end_port   = e;
1719
1720        for (i = 0; i <= e - s; ++i) {
1721                spin_lock_init(&sa_dev->port[i].ah_lock);
1722                if (!rdma_cap_ib_sa(device, i + 1))
1723                        continue;
1724
1725                sa_dev->port[i].sm_ah    = NULL;
1726                sa_dev->port[i].port_num = i + s;
1727
1728                sa_dev->port[i].agent =
1729                        ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1730                                              NULL, 0, send_handler,
1731                                              recv_handler, sa_dev, 0);
1732                if (IS_ERR(sa_dev->port[i].agent))
1733                        goto err;
1734
1735                INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1736
1737                count++;
1738        }
1739
1740        if (!count)
1741                goto free;
1742
1743        ib_set_client_data(device, &sa_client, sa_dev);
1744
1745        /*
1746         * We register our event handler after everything is set up,
1747         * and then update our cached info after the event handler is
1748         * registered to avoid any problems if a port changes state
1749         * during our initialization.
1750         */
1751
1752        INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1753        if (ib_register_event_handler(&sa_dev->event_handler))
1754                goto err;
1755
1756        for (i = 0; i <= e - s; ++i) {
1757                if (rdma_cap_ib_sa(device, i + 1))
1758                        update_sm_ah(&sa_dev->port[i].update_task);
1759        }
1760
1761        return;
1762
1763err:
1764        while (--i >= 0) {
1765                if (rdma_cap_ib_sa(device, i + 1))
1766                        ib_unregister_mad_agent(sa_dev->port[i].agent);
1767        }
1768free:
1769        kfree(sa_dev);
1770        return;
1771}
1772
1773static void ib_sa_remove_one(struct ib_device *device, void *client_data)
1774{
1775        struct ib_sa_device *sa_dev = client_data;
1776        int i;
1777
1778        if (!sa_dev)
1779                return;
1780
1781        ib_unregister_event_handler(&sa_dev->event_handler);
1782
1783        flush_workqueue(ib_wq);
1784
1785        for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1786                if (rdma_cap_ib_sa(device, i + 1)) {
1787                        ib_unregister_mad_agent(sa_dev->port[i].agent);
1788                        if (sa_dev->port[i].sm_ah)
1789                                kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1790                }
1791
1792        }
1793
1794        kfree(sa_dev);
1795}
1796
1797static int __init ib_sa_init(void)
1798{
1799        int ret;
1800
1801        get_random_bytes(&tid, sizeof tid);
1802
1803        atomic_set(&ib_nl_sa_request_seq, 0);
1804
1805        ret = ib_register_client(&sa_client);
1806        if (ret) {
1807                pr_err("Couldn't register ib_sa client\n");
1808                goto err1;
1809        }
1810
1811        ret = mcast_init();
1812        if (ret) {
1813                pr_err("Couldn't initialize multicast handling\n");
1814                goto err2;
1815        }
1816
1817        ib_nl_wq = create_singlethread_workqueue("ib_nl_sa_wq");
1818        if (!ib_nl_wq) {
1819                ret = -ENOMEM;
1820                goto err3;
1821        }
1822
1823        if (ibnl_add_client(RDMA_NL_LS, RDMA_NL_LS_NUM_OPS,
1824                            ib_sa_cb_table)) {
1825                pr_err("Failed to add netlink callback\n");
1826                ret = -EINVAL;
1827                goto err4;
1828        }
1829        INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
1830
1831        return 0;
1832err4:
1833        destroy_workqueue(ib_nl_wq);
1834err3:
1835        mcast_cleanup();
1836err2:
1837        ib_unregister_client(&sa_client);
1838err1:
1839        return ret;
1840}
1841
1842static void __exit ib_sa_cleanup(void)
1843{
1844        ibnl_remove_client(RDMA_NL_LS);
1845        cancel_delayed_work(&ib_nl_timed_work);
1846        flush_workqueue(ib_nl_wq);
1847        destroy_workqueue(ib_nl_wq);
1848        mcast_cleanup();
1849        ib_unregister_client(&sa_client);
1850        idr_destroy(&query_idr);
1851}
1852
1853module_init(ib_sa_init);
1854module_exit(ib_sa_cleanup);
1855