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                 dev_addr->bound_dev_if = dev->ifindex;
 272                 if (vlan_id)
 273                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 274                 dev_put(dev);
 275                 break;
 276 #if IS_ENABLED(CONFIG_IPV6)
 277         case AF_INET6:
 278                 rcu_read_lock();
 279                 for_each_netdev_rcu(dev_addr->net, dev) {
 280                         if (ipv6_chk_addr(dev_addr->net,
 281                                           &((const struct sockaddr_in6 *)addr)->sin6_addr,
 282                                           dev, 1)) {
 283                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 284                                 dev_addr->bound_dev_if = dev->ifindex;
 285                                 if (vlan_id)
 286                                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 287                                 break;
 288                         }
 289                 }
 290                 rcu_read_unlock();
 291                 break;
 292 #endif
 293         }
 294         return ret;
 295 }
 296 EXPORT_SYMBOL(rdma_translate_ip);
 297
 298 static void set_timeout(unsigned long time)
 299 {
 300         unsigned long delay;
 301
 302         delay = time - jiffies;
 303         if ((long)delay < 0)
 304                 delay = 0;
 305
 306         mod_delayed_work(addr_wq, &work, delay);
 307 }
 308
 309 static void queue_req(struct addr_req *req)
 310 {
 311         struct addr_req *temp_req;
 312
 313         mutex_lock(&lock);
 314         list_for_each_entry_reverse(temp_req, &req_list, list) {
 315                 if (time_after_eq(req->timeout, temp_req->timeout))
 316                         break;
 317         }
 318
 319         list_add(&req->list, &temp_req->list);
 320
 321         if (req_list.next == &req->list)
 322                 set_timeout(req->timeout);
 323         mutex_unlock(&lock);
 324 }
 325
 326 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 327                           const void *daddr, u32 seq, u16 family)
 328 {
 329         if (ibnl_chk_listeners(RDMA_NL_GROUP_LS))
 330                 return -EADDRNOTAVAIL;
 331
 332         /* We fill in what we can, the response will fill the rest */
 333         rdma_copy_addr(dev_addr, dst->dev, NULL);
 334         return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
 335 }
 336
 337 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 338                         const void *daddr)
 339 {
 340         struct neighbour *n;
 341         int ret;
 342
 343         n = dst_neigh_lookup(dst, daddr);
 344
 345         rcu_read_lock();
 346         if (!n || !(n->nud_state & NUD_VALID)) {
 347                 if (n)
 348                         neigh_event_send(n, NULL);
 349                 ret = -ENODATA;
 350         } else {
 351                 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
 352         }
 353         rcu_read_unlock();
 354
 355         if (n)
 356                 neigh_release(n);
 357
 358         return ret;
 359 }
 360
 361 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
 362 {
 363         struct rtable *rt;
 364         struct rt6_info *rt6;
 365
 366         if (family == AF_INET) {
 367                 rt = container_of(dst, struct rtable, dst);
 368                 return rt->rt_uses_gateway;
 369         }
 370
 371         rt6 = container_of(dst, struct rt6_info, dst);
 372         return rt6->rt6i_flags & RTF_GATEWAY;
 373 }
 374
 375 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 376                     const struct sockaddr *dst_in, u32 seq)
 377 {
 378         const struct sockaddr_in *dst_in4 =
 379                 (const struct sockaddr_in *)dst_in;
 380         const struct sockaddr_in6 *dst_in6 =
 381                 (const struct sockaddr_in6 *)dst_in;
 382         const void *daddr = (dst_in->sa_family == AF_INET) ?
 383                 (const void *)&dst_in4->sin_addr.s_addr :
 384                 (const void *)&dst_in6->sin6_addr;
 385         sa_family_t family = dst_in->sa_family;
 386
 387         /* Gateway + ARPHRD_INFINIBAND -> IB router */
 388         if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
 389                 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
 390         else
 391                 return dst_fetch_ha(dst, dev_addr, daddr);
 392 }
 393
 394 static int addr4_resolve(struct sockaddr_in *src_in,
 395                          const struct sockaddr_in *dst_in,
 396                          struct rdma_dev_addr *addr,
 397                          struct rtable **prt)
 398 {
 399         __be32 src_ip = src_in->sin_addr.s_addr;
 400         __be32 dst_ip = dst_in->sin_addr.s_addr;
 401         struct rtable *rt;
 402         struct flowi4 fl4;
 403         int ret;
 404
 405         memset(&fl4, 0, sizeof(fl4));
 406         fl4.daddr = dst_ip;
 407         fl4.saddr = src_ip;
 408         fl4.flowi4_oif = addr->bound_dev_if;
 409         rt = ip_route_output_key(addr->net, &fl4);
 410         ret = PTR_ERR_OR_ZERO(rt);
 411         if (ret)
 412                 return ret;
 413
 414         src_in->sin_family = AF_INET;
 415         src_in->sin_addr.s_addr = fl4.saddr;
 416
 417         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 418          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 419          * type accordingly.
 420          */
 421         if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
 422                 addr->network = RDMA_NETWORK_IPV4;
 423
 424         addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
 425
 426         *prt = rt;
 427         return 0;
 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(&src_in->sin6_addr)) {
 452                 src_in->sin6_family = AF_INET6;
 453                 src_in->sin6_addr = fl6.saddr;
 454         }
 455
 456         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 457          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 458          * type accordingly.
 459          */
 460         if (rt->rt6i_flags & RTF_GATEWAY &&
 461             ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
 462                 addr->network = RDMA_NETWORK_IPV6;
 463
 464         addr->hoplimit = ip6_dst_hoplimit(dst);
 465
 466         *pdst = dst;
 467         return 0;
 468 }
 469 #else
 470 static int addr6_resolve(struct sockaddr_in6 *src_in,
 471                          const struct sockaddr_in6 *dst_in,
 472                          struct rdma_dev_addr *addr,
 473                          struct dst_entry **pdst)
 474 {
 475         return -EADDRNOTAVAIL;
 476 }
 477 #endif
 478
 479 static int addr_resolve_neigh(struct dst_entry *dst,
 480                               const struct sockaddr *dst_in,
 481                               struct rdma_dev_addr *addr,
 482                               u32 seq)
 483 {
 484         if (dst->dev->flags & IFF_LOOPBACK) {
 485                 int ret;
 486
 487                 ret = rdma_translate_ip(dst_in, addr, NULL);
 488                 if (!ret)
 489                         memcpy(addr->dst_dev_addr, addr->src_dev_addr,
 490                                MAX_ADDR_LEN);
 491
 492                 return ret;
 493         }
 494
 495         /* If the device doesn't do ARP internally */
 496         if (!(dst->dev->flags & IFF_NOARP))
 497                 return fetch_ha(dst, addr, dst_in, seq);
 498
 499         return rdma_copy_addr(addr, dst->dev, NULL);
 500 }
 501
 502 static int addr_resolve(struct sockaddr *src_in,
 503                         const struct sockaddr *dst_in,
 504                         struct rdma_dev_addr *addr,
 505                         bool resolve_neigh,
 506                         u32 seq)
 507 {
 508         struct net_device *ndev;
 509         struct dst_entry *dst;
 510         int ret;
 511
 512         if (!addr->net) {
 513                 pr_warn_ratelimited("%s: missing namespace\n", __func__);
 514                 return -EINVAL;
 515         }
 516
 517         if (src_in->sa_family == AF_INET) {
 518                 struct rtable *rt = NULL;
 519                 const struct sockaddr_in *dst_in4 =
 520                         (const struct sockaddr_in *)dst_in;
 521
 522                 ret = addr4_resolve((struct sockaddr_in *)src_in,
 523                                     dst_in4, addr, &rt);
 524                 if (ret)
 525                         return ret;
 526
 527                 if (resolve_neigh)
 528                         ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
 529
 530                 if (addr->bound_dev_if) {
 531                         ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
 532                 } else {
 533                         ndev = rt->dst.dev;
 534                         dev_hold(ndev);
 535                 }
 536
 537                 ip_rt_put(rt);
 538         } else {
 539                 const struct sockaddr_in6 *dst_in6 =
 540                         (const struct sockaddr_in6 *)dst_in;
 541
 542                 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
 543                                     dst_in6, addr,
 544                                     &dst);
 545                 if (ret)
 546                         return ret;
 547
 548                 if (resolve_neigh)
 549                         ret = addr_resolve_neigh(dst, dst_in, addr, seq);
 550
 551                 if (addr->bound_dev_if) {
 552                         ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
 553                 } else {
 554                         ndev = dst->dev;
 555                         dev_hold(ndev);
 556                 }
 557
 558                 dst_release(dst);
 559         }
 560
 561         if (ndev->flags & IFF_LOOPBACK) {
 562                 ret = rdma_translate_ip(dst_in, addr, NULL);
 563                 /*
 564                  * Put the loopback device and get the translated
 565                  * device instead.
 566                  */
 567                 dev_put(ndev);
 568                 ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
 569         } else {
 570                 addr->bound_dev_if = ndev->ifindex;
 571         }
 572         dev_put(ndev);
 573
 574         return ret;
 575 }
 576
 577 static void process_req(struct work_struct *work)
 578 {
 579         struct addr_req *req, *temp_req;
 580         struct sockaddr *src_in, *dst_in;
 581         struct list_head done_list;
 582
 583         INIT_LIST_HEAD(&done_list);
 584
 585         mutex_lock(&lock);
 586         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 587                 if (req->status == -ENODATA) {
 588                         src_in = (struct sockaddr *) &req->src_addr;
 589                         dst_in = (struct sockaddr *) &req->dst_addr;
 590                         req->status = addr_resolve(src_in, dst_in, req->addr,
 591                                                    true, req->seq);
 592                         if (req->status && time_after_eq(jiffies, req->timeout))
 593                                 req->status = -ETIMEDOUT;
 594                         else if (req->status == -ENODATA)
 595                                 continue;
 596                 }
 597                 list_move_tail(&req->list, &done_list);
 598         }
 599
 600         if (!list_empty(&req_list)) {
 601                 req = list_entry(req_list.next, struct addr_req, list);
 602                 set_timeout(req->timeout);
 603         }
 604         mutex_unlock(&lock);
 605
 606         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 607                 list_del(&req->list);
 608                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 609                         req->addr, req->context);
 610                 put_client(req->client);
 611                 kfree(req);
 612         }
 613 }
 614
 615 int rdma_resolve_ip(struct rdma_addr_client *client,
 616                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 617                     struct rdma_dev_addr *addr, int timeout_ms,
 618                     void (*callback)(int status, struct sockaddr *src_addr,
 619                                      struct rdma_dev_addr *addr, void *context),
 620                     void *context)
 621 {
 622         struct sockaddr *src_in, *dst_in;
 623         struct addr_req *req;
 624         int ret = 0;
 625
 626         req = kzalloc(sizeof *req, GFP_KERNEL);
 627         if (!req)
 628                 return -ENOMEM;
 629
 630         src_in = (struct sockaddr *) &req->src_addr;
 631         dst_in = (struct sockaddr *) &req->dst_addr;
 632
 633         if (src_addr) {
 634                 if (src_addr->sa_family != dst_addr->sa_family) {
 635                         ret = -EINVAL;
 636                         goto err;
 637                 }
 638
 639                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 640         } else {
 641                 src_in->sa_family = dst_addr->sa_family;
 642         }
 643
 644         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 645         req->addr = addr;
 646         req->callback = callback;
 647         req->context = context;
 648         req->client = client;
 649         atomic_inc(&client->refcount);
 650         req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 651
 652         req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
 653         switch (req->status) {
 654         case 0:
 655                 req->timeout = jiffies;
 656                 queue_req(req);
 657                 break;
 658         case -ENODATA:
 659                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 660                 queue_req(req);
 661                 break;
 662         default:
 663                 ret = req->status;
 664                 atomic_dec(&client->refcount);
 665                 goto err;
 666         }
 667         return ret;
 668 err:
 669         kfree(req);
 670         return ret;
 671 }
 672 EXPORT_SYMBOL(rdma_resolve_ip);
 673
 674 int rdma_resolve_ip_route(struct sockaddr *src_addr,
 675                           const struct sockaddr *dst_addr,
 676                           struct rdma_dev_addr *addr)
 677 {
 678         struct sockaddr_storage ssrc_addr = {};
 679         struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
 680
 681         if (src_addr) {
 682                 if (src_addr->sa_family != dst_addr->sa_family)
 683                         return -EINVAL;
 684
 685                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 686         } else {
 687                 src_in->sa_family = dst_addr->sa_family;
 688         }
 689
 690         return addr_resolve(src_in, dst_addr, addr, false, 0);
 691 }
 692 EXPORT_SYMBOL(rdma_resolve_ip_route);
 693
 694 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 695 {
 696         struct addr_req *req, *temp_req;
 697
 698         mutex_lock(&lock);
 699         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 700                 if (req->addr == addr) {
 701                         req->status = -ECANCELED;
 702                         req->timeout = jiffies;
 703                         list_move(&req->list, &req_list);
 704                         set_timeout(req->timeout);
 705                         break;
 706                 }
 707         }
 708         mutex_unlock(&lock);
 709 }
 710 EXPORT_SYMBOL(rdma_addr_cancel);
 711
 712 struct resolve_cb_context {
 713         struct rdma_dev_addr *addr;
 714         struct completion comp;
 715         int status;
 716 };
 717
 718 static void resolve_cb(int status, struct sockaddr *src_addr,
 719              struct rdma_dev_addr *addr, void *context)
 720 {
 721         if (!status)
 722                 memcpy(((struct resolve_cb_context *)context)->addr,
 723                        addr, sizeof(struct rdma_dev_addr));
 724         ((struct resolve_cb_context *)context)->status = status;
 725         complete(&((struct resolve_cb_context *)context)->comp);
 726 }
 727
 728 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 729                                  const union ib_gid *dgid,
 730                                  u8 *dmac, u16 *vlan_id, int *if_index,
 731                                  int *hoplimit)
 732 {
 733         int ret = 0;
 734         struct rdma_dev_addr dev_addr;
 735         struct resolve_cb_context ctx;
 736         struct net_device *dev;
 737
 738         union {
 739                 struct sockaddr     _sockaddr;
 740                 struct sockaddr_in  _sockaddr_in;
 741                 struct sockaddr_in6 _sockaddr_in6;
 742         } sgid_addr, dgid_addr;
 743
 744
 745         rdma_gid2ip(&sgid_addr._sockaddr, sgid);
 746         rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 747
 748         memset(&dev_addr, 0, sizeof(dev_addr));
 749         if (if_index)
 750                 dev_addr.bound_dev_if = *if_index;
 751         dev_addr.net = &init_net;
 752
 753         ctx.addr = &dev_addr;
 754         init_completion(&ctx.comp);
 755         ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
 756                         &dev_addr, 1000, resolve_cb, &ctx);
 757         if (ret)
 758                 return ret;
 759
 760         wait_for_completion(&ctx.comp);
 761
 762         ret = ctx.status;
 763         if (ret)
 764                 return ret;
 765
 766         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 767         dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
 768         if (!dev)
 769                 return -ENODEV;
 770         if (if_index)
 771                 *if_index = dev_addr.bound_dev_if;
 772         if (vlan_id)
 773                 *vlan_id = rdma_vlan_dev_vlan_id(dev);
 774         if (hoplimit)
 775                 *hoplimit = dev_addr.hoplimit;
 776         dev_put(dev);
 777         return ret;
 778 }
 779 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
 780
 781 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
 782 {
 783         int ret = 0;
 784         struct rdma_dev_addr dev_addr;
 785         union {
 786                 struct sockaddr     _sockaddr;
 787                 struct sockaddr_in  _sockaddr_in;
 788                 struct sockaddr_in6 _sockaddr_in6;
 789         } gid_addr;
 790
 791         rdma_gid2ip(&gid_addr._sockaddr, sgid);
 792
 793         memset(&dev_addr, 0, sizeof(dev_addr));
 794         dev_addr.net = &init_net;
 795         ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
 796         if (ret)
 797                 return ret;
 798
 799         memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
 800         return ret;
 801 }
 802 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
 803
 804 static int netevent_callback(struct notifier_block *self, unsigned long event,
 805         void *ctx)
 806 {
 807         if (event == NETEVENT_NEIGH_UPDATE) {
 808                 struct neighbour *neigh = ctx;
 809
 810                 if (neigh->nud_state & NUD_VALID) {
 811                         set_timeout(jiffies);
 812                 }
 813         }
 814         return 0;
 815 }
 816
 817 static struct notifier_block nb = {
 818         .notifier_call = netevent_callback
 819 };
 820
 821 int addr_init(void)
 822 {
 823         addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
 824         if (!addr_wq)
 825                 return -ENOMEM;
 826
 827         register_netevent_notifier(&nb);
 828         rdma_addr_register_client(&self);
 829
 830         return 0;
 831 }
 832
 833 void addr_cleanup(void)
 834 {
 835         rdma_addr_unregister_client(&self);
 836         unregister_netevent_notifier(&nb);
 837         destroy_workqueue(addr_wq);
 838 }