drivers/infiniband/core/addr.c

   1 /*
   2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
   3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
   4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
   5  * Copyright (c) 2005 Intel Corporation.  All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/mutex.h>
  37 #include <linux/inetdevice.h>
  38 #include <linux/slab.h>
  39 #include <linux/workqueue.h>
  40 #include <linux/module.h>
  41 #include <net/arp.h>
  42 #include <net/neighbour.h>
  43 #include <net/route.h>
  44 #include <net/netevent.h>
  45 #include <net/addrconf.h>
  46 #include <net/ip6_route.h>
  47 #include <rdma/ib_addr.h>
  48 #include <rdma/ib.h>
  49 #include <rdma/rdma_netlink.h>
  50 #include <net/netlink.h>
  51
  52 #include "core_priv.h"
  53
  54 struct addr_req {
  55         struct list_head list;
  56         struct sockaddr_storage src_addr;
  57         struct sockaddr_storage dst_addr;
  58         struct rdma_dev_addr *addr;
  59         struct rdma_addr_client *client;
  60         void *context;
  61         void (*callback)(int status, struct sockaddr *src_addr,
  62                          struct rdma_dev_addr *addr, void *context);
  63         unsigned long timeout;
  64         int status;
  65         u32 seq;
  66 };
  67
  68 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
  69
  70 static void process_req(struct work_struct *work);
  71
  72 static DEFINE_MUTEX(lock);
  73 static LIST_HEAD(req_list);
  74 static DECLARE_DELAYED_WORK(work, process_req);
  75 static struct workqueue_struct *addr_wq;
  76
  77 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
  78         [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
  79                 .len = sizeof(struct rdma_nla_ls_gid)},
  80 };
  81
  82 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
  83 {
  84         struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
  85         int ret;
  86
  87         if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
  88                 return false;
  89
  90         ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
  91                         nlmsg_len(nlh), ib_nl_addr_policy, NULL);
  92         if (ret)
  93                 return false;
  94
  95         return true;
  96 }
  97
  98 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
  99 {
 100         const struct nlattr *head, *curr;
 101         union ib_gid gid;
 102         struct addr_req *req;
 103         int len, rem;
 104         int found = 0;
 105
 106         head = (const struct nlattr *)nlmsg_data(nlh);
 107         len = nlmsg_len(nlh);
 108
 109         nla_for_each_attr(curr, head, len, rem) {
 110                 if (curr->nla_type == LS_NLA_TYPE_DGID)
 111                         memcpy(&gid, nla_data(curr), nla_len(curr));
 112         }
 113
 114         mutex_lock(&lock);
 115         list_for_each_entry(req, &req_list, list) {
 116                 if (nlh->nlmsg_seq != req->seq)
 117                         continue;
 118                 /* We set the DGID part, the rest was set earlier */
 119                 rdma_addr_set_dgid(req->addr, &gid);
 120                 req->status = 0;
 121                 found = 1;
 122                 break;
 123         }
 124         mutex_unlock(&lock);
 125
 126         if (!found)
 127                 pr_info("Couldn't find request waiting for DGID: %pI6\n",
 128                         &gid);
 129 }
 130
 131 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
 132                              struct netlink_callback *cb)
 133 {
 134         const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
 135
 136         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
 137             !(NETLINK_CB(skb).sk) ||
 138             !netlink_capable(skb, CAP_NET_ADMIN))
 139                 return -EPERM;
 140
 141         if (ib_nl_is_good_ip_resp(nlh))
 142                 ib_nl_process_good_ip_rsep(nlh);
 143
 144         return skb->len;
 145 }
 146
 147 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
 148                              const void *daddr,
 149                              u32 seq, u16 family)
 150 {
 151         struct sk_buff *skb = NULL;
 152         struct nlmsghdr *nlh;
 153         struct rdma_ls_ip_resolve_header *header;
 154         void *data;
 155         size_t size;
 156         int attrtype;
 157         int len;
 158
 159         if (family == AF_INET) {
 160                 size = sizeof(struct in_addr);
 161                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
 162         } else {
 163                 size = sizeof(struct in6_addr);
 164                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
 165         }
 166
 167         len = nla_total_size(sizeof(size));
 168         len += NLMSG_ALIGN(sizeof(*header));
 169
 170         skb = nlmsg_new(len, GFP_KERNEL);
 171         if (!skb)
 172                 return -ENOMEM;
 173
 174         data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
 175                             RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
 176         if (!data) {
 177                 nlmsg_free(skb);
 178                 return -ENODATA;
 179         }
 180
 181         /* Construct the family header first */
 182         header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 183         header->ifindex = dev_addr->bound_dev_if;
 184         nla_put(skb, attrtype, size, daddr);
 185
 186         /* Repair the nlmsg header length */
 187         nlmsg_end(skb, nlh);
 188         ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, GFP_KERNEL);
 189
 190         /* Make the request retry, so when we get the response from userspace
 191          * we will have something.
 192          */
 193         return -ENODATA;
 194 }
 195
 196 int rdma_addr_size(struct sockaddr *addr)
 197 {
 198         switch (addr->sa_family) {
 199         case AF_INET:
 200                 return sizeof(struct sockaddr_in);
 201         case AF_INET6:
 202                 return sizeof(struct sockaddr_in6);
 203         case AF_IB:
 204                 return sizeof(struct sockaddr_ib);
 205         default:
 206                 return 0;
 207         }
 208 }
 209 EXPORT_SYMBOL(rdma_addr_size);
 210
 211 static struct rdma_addr_client self;
 212
 213 void rdma_addr_register_client(struct rdma_addr_client *client)
 214 {
 215         atomic_set(&client->refcount, 1);
 216         init_completion(&client->comp);
 217 }
 218 EXPORT_SYMBOL(rdma_addr_register_client);
 219
 220 static inline void put_client(struct rdma_addr_client *client)
 221 {
 222         if (atomic_dec_and_test(&client->refcount))
 223                 complete(&client->comp);
 224 }
 225
 226 void rdma_addr_unregister_client(struct rdma_addr_client *client)
 227 {
 228         put_client(client);
 229         wait_for_completion(&client->comp);
 230 }
 231 EXPORT_SYMBOL(rdma_addr_unregister_client);
 232
 233 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
 234                      const unsigned char *dst_dev_addr)
 235 {
 236         dev_addr->dev_type = dev->type;
 237         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 238         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
 239         if (dst_dev_addr)
 240                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 241         dev_addr->bound_dev_if = dev->ifindex;
 242         return 0;
 243 }
 244 EXPORT_SYMBOL(rdma_copy_addr);
 245
 246 int rdma_translate_ip(const struct sockaddr *addr,
 247                       struct rdma_dev_addr *dev_addr,
 248                       u16 *vlan_id)
 249 {
 250         struct net_device *dev;
 251         int ret = -EADDRNOTAVAIL;
 252
 253         if (dev_addr->bound_dev_if) {
 254                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 255                 if (!dev)
 256                         return -ENODEV;
 257                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 258                 dev_put(dev);
 259                 return ret;
 260         }
 261
 262         switch (addr->sa_family) {
 263         case AF_INET:
 264                 dev = ip_dev_find(dev_addr->net,
 265                         ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
 266
 267                 if (!dev)
 268                         return ret;
 269
 270                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 271                 if (vlan_id)
 272                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 273                 dev_put(dev);
 274                 break;
 275 #if IS_ENABLED(CONFIG_IPV6)
 276         case AF_INET6:
 277                 rcu_read_lock();
 278                 for_each_netdev_rcu(dev_addr->net, dev) {
 279                         if (ipv6_chk_addr(dev_addr->net,
 280                                           &((const struct sockaddr_in6 *)addr)->sin6_addr,
 281                                           dev, 1)) {
 282                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 283                                 if (vlan_id)
 284                                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 285                                 break;
 286                         }
 287                 }
 288                 rcu_read_unlock();
 289                 break;
 290 #endif
 291         }
 292         return ret;
 293 }
 294 EXPORT_SYMBOL(rdma_translate_ip);
 295
 296 static void set_timeout(unsigned long time)
 297 {
 298         unsigned long delay;
 299
 300         delay = time - jiffies;
 301         if ((long)delay < 0)
 302                 delay = 0;
 303
 304         mod_delayed_work(addr_wq, &work, delay);
 305 }
 306
 307 static void queue_req(struct addr_req *req)
 308 {
 309         struct addr_req *temp_req;
 310
 311         mutex_lock(&lock);
 312         list_for_each_entry_reverse(temp_req, &req_list, list) {
 313                 if (time_after_eq(req->timeout, temp_req->timeout))
 314                         break;
 315         }
 316
 317         list_add(&req->list, &temp_req->list);
 318
 319         if (req_list.next == &req->list)
 320                 set_timeout(req->timeout);
 321         mutex_unlock(&lock);
 322 }
 323
 324 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 325                           const void *daddr, u32 seq, u16 family)
 326 {
 327         if (ibnl_chk_listeners(RDMA_NL_GROUP_LS))
 328                 return -EADDRNOTAVAIL;
 329
 330         /* We fill in what we can, the response will fill the rest */
 331         rdma_copy_addr(dev_addr, dst->dev, NULL);
 332         return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
 333 }
 334
 335 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 336                         const void *daddr)
 337 {
 338         struct neighbour *n;
 339         int ret;
 340
 341         n = dst_neigh_lookup(dst, daddr);
 342
 343         rcu_read_lock();
 344         if (!n || !(n->nud_state & NUD_VALID)) {
 345                 if (n)
 346                         neigh_event_send(n, NULL);
 347                 ret = -ENODATA;
 348         } else {
 349                 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
 350         }
 351         rcu_read_unlock();
 352
 353         if (n)
 354                 neigh_release(n);
 355
 356         return ret;
 357 }
 358
 359 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
 360 {
 361         struct rtable *rt;
 362         struct rt6_info *rt6;
 363
 364         if (family == AF_INET) {
 365                 rt = container_of(dst, struct rtable, dst);
 366                 return rt->rt_uses_gateway;
 367         }
 368
 369         rt6 = container_of(dst, struct rt6_info, dst);
 370         return rt6->rt6i_flags & RTF_GATEWAY;
 371 }
 372
 373 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 374                     const struct sockaddr *dst_in, u32 seq)
 375 {
 376         const struct sockaddr_in *dst_in4 =
 377                 (const struct sockaddr_in *)dst_in;
 378         const struct sockaddr_in6 *dst_in6 =
 379                 (const struct sockaddr_in6 *)dst_in;
 380         const void *daddr = (dst_in->sa_family == AF_INET) ?
 381                 (const void *)&dst_in4->sin_addr.s_addr :
 382                 (const void *)&dst_in6->sin6_addr;
 383         sa_family_t family = dst_in->sa_family;
 384
 385         /* Gateway + ARPHRD_INFINIBAND -> IB router */
 386         if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
 387                 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
 388         else
 389                 return dst_fetch_ha(dst, dev_addr, daddr);
 390 }
 391
 392 static int addr4_resolve(struct sockaddr_in *src_in,
 393                          const struct sockaddr_in *dst_in,
 394                          struct rdma_dev_addr *addr,
 395                          struct rtable **prt)
 396 {
 397         __be32 src_ip = src_in->sin_addr.s_addr;
 398         __be32 dst_ip = dst_in->sin_addr.s_addr;
 399         struct rtable *rt;
 400         struct flowi4 fl4;
 401         int ret;
 402
 403         memset(&fl4, 0, sizeof(fl4));
 404         fl4.daddr = dst_ip;
 405         fl4.saddr = src_ip;
 406         fl4.flowi4_oif = addr->bound_dev_if;
 407         rt = ip_route_output_key(addr->net, &fl4);
 408         if (IS_ERR(rt)) {
 409                 ret = PTR_ERR(rt);
 410                 goto out;
 411         }
 412         src_in->sin_family = AF_INET;
 413         src_in->sin_addr.s_addr = fl4.saddr;
 414
 415         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 416          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 417          * type accordingly.
 418          */
 419         if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
 420                 addr->network = RDMA_NETWORK_IPV4;
 421
 422         addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
 423
 424         *prt = rt;
 425         return 0;
 426 out:
 427         return ret;
 428 }
 429
 430 #if IS_ENABLED(CONFIG_IPV6)
 431 static int addr6_resolve(struct sockaddr_in6 *src_in,
 432                          const struct sockaddr_in6 *dst_in,
 433                          struct rdma_dev_addr *addr,
 434                          struct dst_entry **pdst)
 435 {
 436         struct flowi6 fl6;
 437         struct dst_entry *dst;
 438         struct rt6_info *rt;
 439         int ret;
 440
 441         memset(&fl6, 0, sizeof fl6);
 442         fl6.daddr = dst_in->sin6_addr;
 443         fl6.saddr = src_in->sin6_addr;
 444         fl6.flowi6_oif = addr->bound_dev_if;
 445
 446         ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
 447         if (ret < 0)
 448                 return ret;
 449
 450         rt = (struct rt6_info *)dst;
 451         if (ipv6_addr_any(&fl6.saddr)) {
 452                 ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev,
 453                                          &fl6.daddr, 0, &fl6.saddr);
 454                 if (ret)
 455                         goto put;
 456
 457                 src_in->sin6_family = AF_INET6;
 458                 src_in->sin6_addr = fl6.saddr;
 459         }
 460
 461         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 462          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 463          * type accordingly.
 464          */
 465         if (rt->rt6i_flags & RTF_GATEWAY &&
 466             ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
 467                 addr->network = RDMA_NETWORK_IPV6;
 468
 469         addr->hoplimit = ip6_dst_hoplimit(dst);
 470
 471         *pdst = dst;
 472         return 0;
 473 put:
 474         dst_release(dst);
 475         return ret;
 476 }
 477 #else
 478 static int addr6_resolve(struct sockaddr_in6 *src_in,
 479                          const struct sockaddr_in6 *dst_in,
 480                          struct rdma_dev_addr *addr,
 481                          struct dst_entry **pdst)
 482 {
 483         return -EADDRNOTAVAIL;
 484 }
 485 #endif
 486
 487 static int addr_resolve_neigh(struct dst_entry *dst,
 488                               const struct sockaddr *dst_in,
 489                               struct rdma_dev_addr *addr,
 490                               u32 seq)
 491 {
 492         if (dst->dev->flags & IFF_LOOPBACK) {
 493                 int ret;
 494
 495                 ret = rdma_translate_ip(dst_in, addr, NULL);
 496                 if (!ret)
 497                         memcpy(addr->dst_dev_addr, addr->src_dev_addr,
 498                                MAX_ADDR_LEN);
 499
 500                 return ret;
 501         }
 502
 503         /* If the device doesn't do ARP internally */
 504         if (!(dst->dev->flags & IFF_NOARP))
 505                 return fetch_ha(dst, addr, dst_in, seq);
 506
 507         return rdma_copy_addr(addr, dst->dev, NULL);
 508 }
 509
 510 static int addr_resolve(struct sockaddr *src_in,
 511                         const struct sockaddr *dst_in,
 512                         struct rdma_dev_addr *addr,
 513                         bool resolve_neigh,
 514                         u32 seq)
 515 {
 516         struct net_device *ndev;
 517         struct dst_entry *dst;
 518         int ret;
 519
 520         if (src_in->sa_family == AF_INET) {
 521                 struct rtable *rt = NULL;
 522                 const struct sockaddr_in *dst_in4 =
 523                         (const struct sockaddr_in *)dst_in;
 524
 525                 ret = addr4_resolve((struct sockaddr_in *)src_in,
 526                                     dst_in4, addr, &rt);
 527                 if (ret)
 528                         return ret;
 529
 530                 if (resolve_neigh)
 531                         ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
 532
 533                 ndev = rt->dst.dev;
 534                 dev_hold(ndev);
 535
 536                 ip_rt_put(rt);
 537         } else {
 538                 const struct sockaddr_in6 *dst_in6 =
 539                         (const struct sockaddr_in6 *)dst_in;
 540
 541                 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
 542                                     dst_in6, addr,
 543                                     &dst);
 544                 if (ret)
 545                         return ret;
 546
 547                 if (resolve_neigh)
 548                         ret = addr_resolve_neigh(dst, dst_in, addr, seq);
 549
 550                 ndev = dst->dev;
 551                 dev_hold(ndev);
 552
 553                 dst_release(dst);
 554         }
 555
 556         addr->bound_dev_if = ndev->ifindex;
 557         addr->net = dev_net(ndev);
 558         dev_put(ndev);
 559
 560         return ret;
 561 }
 562
 563 static void process_req(struct work_struct *work)
 564 {
 565         struct addr_req *req, *temp_req;
 566         struct sockaddr *src_in, *dst_in;
 567         struct list_head done_list;
 568
 569         INIT_LIST_HEAD(&done_list);
 570
 571         mutex_lock(&lock);
 572         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 573                 if (req->status == -ENODATA) {
 574                         src_in = (struct sockaddr *) &req->src_addr;
 575                         dst_in = (struct sockaddr *) &req->dst_addr;
 576                         req->status = addr_resolve(src_in, dst_in, req->addr,
 577                                                    true, req->seq);
 578                         if (req->status && time_after_eq(jiffies, req->timeout))
 579                                 req->status = -ETIMEDOUT;
 580                         else if (req->status == -ENODATA)
 581                                 continue;
 582                 }
 583                 list_move_tail(&req->list, &done_list);
 584         }
 585
 586         if (!list_empty(&req_list)) {
 587                 req = list_entry(req_list.next, struct addr_req, list);
 588                 set_timeout(req->timeout);
 589         }
 590         mutex_unlock(&lock);
 591
 592         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 593                 list_del(&req->list);
 594                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 595                         req->addr, req->context);
 596                 put_client(req->client);
 597                 kfree(req);
 598         }
 599 }
 600
 601 int rdma_resolve_ip(struct rdma_addr_client *client,
 602                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 603                     struct rdma_dev_addr *addr, int timeout_ms,
 604                     void (*callback)(int status, struct sockaddr *src_addr,
 605                                      struct rdma_dev_addr *addr, void *context),
 606                     void *context)
 607 {
 608         struct sockaddr *src_in, *dst_in;
 609         struct addr_req *req;
 610         int ret = 0;
 611
 612         req = kzalloc(sizeof *req, GFP_KERNEL);
 613         if (!req)
 614                 return -ENOMEM;
 615
 616         src_in = (struct sockaddr *) &req->src_addr;
 617         dst_in = (struct sockaddr *) &req->dst_addr;
 618
 619         if (src_addr) {
 620                 if (src_addr->sa_family != dst_addr->sa_family) {
 621                         ret = -EINVAL;
 622                         goto err;
 623                 }
 624
 625                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 626         } else {
 627                 src_in->sa_family = dst_addr->sa_family;
 628         }
 629
 630         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 631         req->addr = addr;
 632         req->callback = callback;
 633         req->context = context;
 634         req->client = client;
 635         atomic_inc(&client->refcount);
 636         req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 637
 638         req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
 639         switch (req->status) {
 640         case 0:
 641                 req->timeout = jiffies;
 642                 queue_req(req);
 643                 break;
 644         case -ENODATA:
 645                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 646                 queue_req(req);
 647                 break;
 648         default:
 649                 ret = req->status;
 650                 atomic_dec(&client->refcount);
 651                 goto err;
 652         }
 653         return ret;
 654 err:
 655         kfree(req);
 656         return ret;
 657 }
 658 EXPORT_SYMBOL(rdma_resolve_ip);
 659
 660 int rdma_resolve_ip_route(struct sockaddr *src_addr,
 661                           const struct sockaddr *dst_addr,
 662                           struct rdma_dev_addr *addr)
 663 {
 664         struct sockaddr_storage ssrc_addr = {};
 665         struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
 666
 667         if (src_addr) {
 668                 if (src_addr->sa_family != dst_addr->sa_family)
 669                         return -EINVAL;
 670
 671                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 672         } else {
 673                 src_in->sa_family = dst_addr->sa_family;
 674         }
 675
 676         return addr_resolve(src_in, dst_addr, addr, false, 0);
 677 }
 678 EXPORT_SYMBOL(rdma_resolve_ip_route);
 679
 680 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 681 {
 682         struct addr_req *req, *temp_req;
 683
 684         mutex_lock(&lock);
 685         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 686                 if (req->addr == addr) {
 687                         req->status = -ECANCELED;
 688                         req->timeout = jiffies;
 689                         list_move(&req->list, &req_list);
 690                         set_timeout(req->timeout);
 691                         break;
 692                 }
 693         }
 694         mutex_unlock(&lock);
 695 }
 696 EXPORT_SYMBOL(rdma_addr_cancel);
 697
 698 struct resolve_cb_context {
 699         struct rdma_dev_addr *addr;
 700         struct completion comp;
 701         int status;
 702 };
 703
 704 static void resolve_cb(int status, struct sockaddr *src_addr,
 705              struct rdma_dev_addr *addr, void *context)
 706 {
 707         if (!status)
 708                 memcpy(((struct resolve_cb_context *)context)->addr,
 709                        addr, sizeof(struct rdma_dev_addr));
 710         ((struct resolve_cb_context *)context)->status = status;
 711         complete(&((struct resolve_cb_context *)context)->comp);
 712 }
 713
 714 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 715                                  const union ib_gid *dgid,
 716                                  u8 *dmac, u16 *vlan_id, int *if_index,
 717                                  int *hoplimit)
 718 {
 719         int ret = 0;
 720         struct rdma_dev_addr dev_addr;
 721         struct resolve_cb_context ctx;
 722         struct net_device *dev;
 723
 724         union {
 725                 struct sockaddr     _sockaddr;
 726                 struct sockaddr_in  _sockaddr_in;
 727                 struct sockaddr_in6 _sockaddr_in6;
 728         } sgid_addr, dgid_addr;
 729
 730
 731         rdma_gid2ip(&sgid_addr._sockaddr, sgid);
 732         rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 733
 734         memset(&dev_addr, 0, sizeof(dev_addr));
 735         if (if_index)
 736                 dev_addr.bound_dev_if = *if_index;
 737         dev_addr.net = &init_net;
 738
 739         ctx.addr = &dev_addr;
 740         init_completion(&ctx.comp);
 741         ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
 742                         &dev_addr, 1000, resolve_cb, &ctx);
 743         if (ret)
 744                 return ret;
 745
 746         wait_for_completion(&ctx.comp);
 747
 748         ret = ctx.status;
 749         if (ret)
 750                 return ret;
 751
 752         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 753         dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
 754         if (!dev)
 755                 return -ENODEV;
 756         if (if_index)
 757                 *if_index = dev_addr.bound_dev_if;
 758         if (vlan_id)
 759                 *vlan_id = rdma_vlan_dev_vlan_id(dev);
 760         if (hoplimit)
 761                 *hoplimit = dev_addr.hoplimit;
 762         dev_put(dev);
 763         return ret;
 764 }
 765 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
 766
 767 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
 768 {
 769         int ret = 0;
 770         struct rdma_dev_addr dev_addr;
 771         union {
 772                 struct sockaddr     _sockaddr;
 773                 struct sockaddr_in  _sockaddr_in;
 774                 struct sockaddr_in6 _sockaddr_in6;
 775         } gid_addr;
 776
 777         rdma_gid2ip(&gid_addr._sockaddr, sgid);
 778
 779         memset(&dev_addr, 0, sizeof(dev_addr));
 780         dev_addr.net = &init_net;
 781         ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
 782         if (ret)
 783                 return ret;
 784
 785         memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
 786         return ret;
 787 }
 788 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
 789
 790 static int netevent_callback(struct notifier_block *self, unsigned long event,
 791         void *ctx)
 792 {
 793         if (event == NETEVENT_NEIGH_UPDATE) {
 794                 struct neighbour *neigh = ctx;
 795
 796                 if (neigh->nud_state & NUD_VALID) {
 797                         set_timeout(jiffies);
 798                 }
 799         }
 800         return 0;
 801 }
 802
 803 static struct notifier_block nb = {
 804         .notifier_call = netevent_callback
 805 };
 806
 807 int addr_init(void)
 808 {
 809         addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
 810         if (!addr_wq)
 811                 return -ENOMEM;
 812
 813         register_netevent_notifier(&nb);
 814         rdma_addr_register_client(&self);
 815
 816         return 0;
 817 }
 818
 819 void addr_cleanup(void)
 820 {
 821         rdma_addr_unregister_client(&self);
 822         unregister_netevent_notifier(&nb);
 823         destroy_workqueue(addr_wq);
 824 }