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 = (struct rdma_ls_ip_resolve_header *)
 183                 skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 184         header->ifindex = dev_addr->bound_dev_if;
 185         nla_put(skb, attrtype, size, daddr);
 186
 187         /* Repair the nlmsg header length */
 188         nlmsg_end(skb, nlh);
 189         ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, GFP_KERNEL);
 190
 191         /* Make the request retry, so when we get the response from userspace
 192          * we will have something.
 193          */
 194         return -ENODATA;
 195 }
 196
 197 int rdma_addr_size(struct sockaddr *addr)
 198 {
 199         switch (addr->sa_family) {
 200         case AF_INET:
 201                 return sizeof(struct sockaddr_in);
 202         case AF_INET6:
 203                 return sizeof(struct sockaddr_in6);
 204         case AF_IB:
 205                 return sizeof(struct sockaddr_ib);
 206         default:
 207                 return 0;
 208         }
 209 }
 210 EXPORT_SYMBOL(rdma_addr_size);
 211
 212 static struct rdma_addr_client self;
 213
 214 void rdma_addr_register_client(struct rdma_addr_client *client)
 215 {
 216         atomic_set(&client->refcount, 1);
 217         init_completion(&client->comp);
 218 }
 219 EXPORT_SYMBOL(rdma_addr_register_client);
 220
 221 static inline void put_client(struct rdma_addr_client *client)
 222 {
 223         if (atomic_dec_and_test(&client->refcount))
 224                 complete(&client->comp);
 225 }
 226
 227 void rdma_addr_unregister_client(struct rdma_addr_client *client)
 228 {
 229         put_client(client);
 230         wait_for_completion(&client->comp);
 231 }
 232 EXPORT_SYMBOL(rdma_addr_unregister_client);
 233
 234 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
 235                      const unsigned char *dst_dev_addr)
 236 {
 237         dev_addr->dev_type = dev->type;
 238         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 239         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
 240         if (dst_dev_addr)
 241                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 242         dev_addr->bound_dev_if = dev->ifindex;
 243         return 0;
 244 }
 245 EXPORT_SYMBOL(rdma_copy_addr);
 246
 247 int rdma_translate_ip(const struct sockaddr *addr,
 248                       struct rdma_dev_addr *dev_addr,
 249                       u16 *vlan_id)
 250 {
 251         struct net_device *dev;
 252         int ret = -EADDRNOTAVAIL;
 253
 254         if (dev_addr->bound_dev_if) {
 255                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 256                 if (!dev)
 257                         return -ENODEV;
 258                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 259                 dev_put(dev);
 260                 return ret;
 261         }
 262
 263         switch (addr->sa_family) {
 264         case AF_INET:
 265                 dev = ip_dev_find(dev_addr->net,
 266                         ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
 267
 268                 if (!dev)
 269                         return ret;
 270
 271                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 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                                 if (vlan_id)
 285                                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 286                                 break;
 287                         }
 288                 }
 289                 rcu_read_unlock();
 290                 break;
 291 #endif
 292         }
 293         return ret;
 294 }
 295 EXPORT_SYMBOL(rdma_translate_ip);
 296
 297 static void set_timeout(unsigned long time)
 298 {
 299         unsigned long delay;
 300
 301         delay = time - jiffies;
 302         if ((long)delay < 0)
 303                 delay = 0;
 304
 305         mod_delayed_work(addr_wq, &work, delay);
 306 }
 307
 308 static void queue_req(struct addr_req *req)
 309 {
 310         struct addr_req *temp_req;
 311
 312         mutex_lock(&lock);
 313         list_for_each_entry_reverse(temp_req, &req_list, list) {
 314                 if (time_after_eq(req->timeout, temp_req->timeout))
 315                         break;
 316         }
 317
 318         list_add(&req->list, &temp_req->list);
 319
 320         if (req_list.next == &req->list)
 321                 set_timeout(req->timeout);
 322         mutex_unlock(&lock);
 323 }
 324
 325 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 326                           const void *daddr, u32 seq, u16 family)
 327 {
 328         if (ibnl_chk_listeners(RDMA_NL_GROUP_LS))
 329                 return -EADDRNOTAVAIL;
 330
 331         /* We fill in what we can, the response will fill the rest */
 332         rdma_copy_addr(dev_addr, dst->dev, NULL);
 333         return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
 334 }
 335
 336 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 337                         const void *daddr)
 338 {
 339         struct neighbour *n;
 340         int ret;
 341
 342         n = dst_neigh_lookup(dst, daddr);
 343
 344         rcu_read_lock();
 345         if (!n || !(n->nud_state & NUD_VALID)) {
 346                 if (n)
 347                         neigh_event_send(n, NULL);
 348                 ret = -ENODATA;
 349         } else {
 350                 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
 351         }
 352         rcu_read_unlock();
 353
 354         if (n)
 355                 neigh_release(n);
 356
 357         return ret;
 358 }
 359
 360 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
 361 {
 362         struct rtable *rt;
 363         struct rt6_info *rt6;
 364
 365         if (family == AF_INET) {
 366                 rt = container_of(dst, struct rtable, dst);
 367                 return rt->rt_uses_gateway;
 368         }
 369
 370         rt6 = container_of(dst, struct rt6_info, dst);
 371         return rt6->rt6i_flags & RTF_GATEWAY;
 372 }
 373
 374 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 375                     const struct sockaddr *dst_in, u32 seq)
 376 {
 377         const struct sockaddr_in *dst_in4 =
 378                 (const struct sockaddr_in *)dst_in;
 379         const struct sockaddr_in6 *dst_in6 =
 380                 (const struct sockaddr_in6 *)dst_in;
 381         const void *daddr = (dst_in->sa_family == AF_INET) ?
 382                 (const void *)&dst_in4->sin_addr.s_addr :
 383                 (const void *)&dst_in6->sin6_addr;
 384         sa_family_t family = dst_in->sa_family;
 385
 386         /* Gateway + ARPHRD_INFINIBAND -> IB router */
 387         if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
 388                 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
 389         else
 390                 return dst_fetch_ha(dst, dev_addr, daddr);
 391 }
 392
 393 static int addr4_resolve(struct sockaddr_in *src_in,
 394                          const struct sockaddr_in *dst_in,
 395                          struct rdma_dev_addr *addr,
 396                          struct rtable **prt)
 397 {
 398         __be32 src_ip = src_in->sin_addr.s_addr;
 399         __be32 dst_ip = dst_in->sin_addr.s_addr;
 400         struct rtable *rt;
 401         struct flowi4 fl4;
 402         int ret;
 403
 404         memset(&fl4, 0, sizeof(fl4));
 405         fl4.daddr = dst_ip;
 406         fl4.saddr = src_ip;
 407         fl4.flowi4_oif = addr->bound_dev_if;
 408         rt = ip_route_output_key(addr->net, &fl4);
 409         if (IS_ERR(rt)) {
 410                 ret = PTR_ERR(rt);
 411                 goto out;
 412         }
 413         src_in->sin_family = AF_INET;
 414         src_in->sin_addr.s_addr = fl4.saddr;
 415
 416         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 417          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 418          * type accordingly.
 419          */
 420         if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
 421                 addr->network = RDMA_NETWORK_IPV4;
 422
 423         addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
 424
 425         *prt = rt;
 426         return 0;
 427 out:
 428         return ret;
 429 }
 430
 431 #if IS_ENABLED(CONFIG_IPV6)
 432 static int addr6_resolve(struct sockaddr_in6 *src_in,
 433                          const struct sockaddr_in6 *dst_in,
 434                          struct rdma_dev_addr *addr,
 435                          struct dst_entry **pdst)
 436 {
 437         struct flowi6 fl6;
 438         struct dst_entry *dst;
 439         struct rt6_info *rt;
 440         int ret;
 441
 442         memset(&fl6, 0, sizeof fl6);
 443         fl6.daddr = dst_in->sin6_addr;
 444         fl6.saddr = src_in->sin6_addr;
 445         fl6.flowi6_oif = addr->bound_dev_if;
 446
 447         ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
 448         if (ret < 0)
 449                 return ret;
 450
 451         rt = (struct rt6_info *)dst;
 452         if (ipv6_addr_any(&src_in->sin6_addr)) {
 453                 src_in->sin6_family = AF_INET6;
 454                 src_in->sin6_addr = fl6.saddr;
 455         }
 456
 457         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 458          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 459          * type accordingly.
 460          */
 461         if (rt->rt6i_flags & RTF_GATEWAY &&
 462             ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
 463                 addr->network = RDMA_NETWORK_IPV6;
 464
 465         addr->hoplimit = ip6_dst_hoplimit(dst);
 466
 467         *pdst = dst;
 468         return 0;
 469 }
 470 #else
 471 static int addr6_resolve(struct sockaddr_in6 *src_in,
 472                          const struct sockaddr_in6 *dst_in,
 473                          struct rdma_dev_addr *addr,
 474                          struct dst_entry **pdst)
 475 {
 476         return -EADDRNOTAVAIL;
 477 }
 478 #endif
 479
 480 static int addr_resolve_neigh(struct dst_entry *dst,
 481                               const struct sockaddr *dst_in,
 482                               struct rdma_dev_addr *addr,
 483                               u32 seq)
 484 {
 485         if (dst->dev->flags & IFF_LOOPBACK) {
 486                 int ret;
 487
 488                 ret = rdma_translate_ip(dst_in, addr, NULL);
 489                 if (!ret)
 490                         memcpy(addr->dst_dev_addr, addr->src_dev_addr,
 491                                MAX_ADDR_LEN);
 492
 493                 return ret;
 494         }
 495
 496         /* If the device doesn't do ARP internally */
 497         if (!(dst->dev->flags & IFF_NOARP))
 498                 return fetch_ha(dst, addr, dst_in, seq);
 499
 500         return rdma_copy_addr(addr, dst->dev, NULL);
 501 }
 502
 503 static int addr_resolve(struct sockaddr *src_in,
 504                         const struct sockaddr *dst_in,
 505                         struct rdma_dev_addr *addr,
 506                         bool resolve_neigh,
 507                         u32 seq)
 508 {
 509         struct net_device *ndev;
 510         struct dst_entry *dst;
 511         int ret;
 512
 513         if (src_in->sa_family == AF_INET) {
 514                 struct rtable *rt = NULL;
 515                 const struct sockaddr_in *dst_in4 =
 516                         (const struct sockaddr_in *)dst_in;
 517
 518                 ret = addr4_resolve((struct sockaddr_in *)src_in,
 519                                     dst_in4, addr, &rt);
 520                 if (ret)
 521                         return ret;
 522
 523                 if (resolve_neigh)
 524                         ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
 525
 526                 ndev = rt->dst.dev;
 527                 dev_hold(ndev);
 528
 529                 ip_rt_put(rt);
 530         } else {
 531                 const struct sockaddr_in6 *dst_in6 =
 532                         (const struct sockaddr_in6 *)dst_in;
 533
 534                 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
 535                                     dst_in6, addr,
 536                                     &dst);
 537                 if (ret)
 538                         return ret;
 539
 540                 if (resolve_neigh)
 541                         ret = addr_resolve_neigh(dst, dst_in, addr, seq);
 542
 543                 ndev = dst->dev;
 544                 dev_hold(ndev);
 545
 546                 dst_release(dst);
 547         }
 548
 549         addr->bound_dev_if = ndev->ifindex;
 550         addr->net = dev_net(ndev);
 551         dev_put(ndev);
 552
 553         return ret;
 554 }
 555
 556 static void process_req(struct work_struct *work)
 557 {
 558         struct addr_req *req, *temp_req;
 559         struct sockaddr *src_in, *dst_in;
 560         struct list_head done_list;
 561
 562         INIT_LIST_HEAD(&done_list);
 563
 564         mutex_lock(&lock);
 565         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 566                 if (req->status == -ENODATA) {
 567                         src_in = (struct sockaddr *) &req->src_addr;
 568                         dst_in = (struct sockaddr *) &req->dst_addr;
 569                         req->status = addr_resolve(src_in, dst_in, req->addr,
 570                                                    true, req->seq);
 571                         if (req->status && time_after_eq(jiffies, req->timeout))
 572                                 req->status = -ETIMEDOUT;
 573                         else if (req->status == -ENODATA)
 574                                 continue;
 575                 }
 576                 list_move_tail(&req->list, &done_list);
 577         }
 578
 579         if (!list_empty(&req_list)) {
 580                 req = list_entry(req_list.next, struct addr_req, list);
 581                 set_timeout(req->timeout);
 582         }
 583         mutex_unlock(&lock);
 584
 585         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 586                 list_del(&req->list);
 587                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 588                         req->addr, req->context);
 589                 put_client(req->client);
 590                 kfree(req);
 591         }
 592 }
 593
 594 int rdma_resolve_ip(struct rdma_addr_client *client,
 595                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 596                     struct rdma_dev_addr *addr, int timeout_ms,
 597                     void (*callback)(int status, struct sockaddr *src_addr,
 598                                      struct rdma_dev_addr *addr, void *context),
 599                     void *context)
 600 {
 601         struct sockaddr *src_in, *dst_in;
 602         struct addr_req *req;
 603         int ret = 0;
 604
 605         req = kzalloc(sizeof *req, GFP_KERNEL);
 606         if (!req)
 607                 return -ENOMEM;
 608
 609         src_in = (struct sockaddr *) &req->src_addr;
 610         dst_in = (struct sockaddr *) &req->dst_addr;
 611
 612         if (src_addr) {
 613                 if (src_addr->sa_family != dst_addr->sa_family) {
 614                         ret = -EINVAL;
 615                         goto err;
 616                 }
 617
 618                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 619         } else {
 620                 src_in->sa_family = dst_addr->sa_family;
 621         }
 622
 623         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 624         req->addr = addr;
 625         req->callback = callback;
 626         req->context = context;
 627         req->client = client;
 628         atomic_inc(&client->refcount);
 629         req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 630
 631         req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
 632         switch (req->status) {
 633         case 0:
 634                 req->timeout = jiffies;
 635                 queue_req(req);
 636                 break;
 637         case -ENODATA:
 638                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 639                 queue_req(req);
 640                 break;
 641         default:
 642                 ret = req->status;
 643                 atomic_dec(&client->refcount);
 644                 goto err;
 645         }
 646         return ret;
 647 err:
 648         kfree(req);
 649         return ret;
 650 }
 651 EXPORT_SYMBOL(rdma_resolve_ip);
 652
 653 int rdma_resolve_ip_route(struct sockaddr *src_addr,
 654                           const struct sockaddr *dst_addr,
 655                           struct rdma_dev_addr *addr)
 656 {
 657         struct sockaddr_storage ssrc_addr = {};
 658         struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
 659
 660         if (src_addr) {
 661                 if (src_addr->sa_family != dst_addr->sa_family)
 662                         return -EINVAL;
 663
 664                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 665         } else {
 666                 src_in->sa_family = dst_addr->sa_family;
 667         }
 668
 669         return addr_resolve(src_in, dst_addr, addr, false, 0);
 670 }
 671 EXPORT_SYMBOL(rdma_resolve_ip_route);
 672
 673 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 674 {
 675         struct addr_req *req, *temp_req;
 676
 677         mutex_lock(&lock);
 678         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 679                 if (req->addr == addr) {
 680                         req->status = -ECANCELED;
 681                         req->timeout = jiffies;
 682                         list_move(&req->list, &req_list);
 683                         set_timeout(req->timeout);
 684                         break;
 685                 }
 686         }
 687         mutex_unlock(&lock);
 688 }
 689 EXPORT_SYMBOL(rdma_addr_cancel);
 690
 691 struct resolve_cb_context {
 692         struct rdma_dev_addr *addr;
 693         struct completion comp;
 694         int status;
 695 };
 696
 697 static void resolve_cb(int status, struct sockaddr *src_addr,
 698              struct rdma_dev_addr *addr, void *context)
 699 {
 700         if (!status)
 701                 memcpy(((struct resolve_cb_context *)context)->addr,
 702                        addr, sizeof(struct rdma_dev_addr));
 703         ((struct resolve_cb_context *)context)->status = status;
 704         complete(&((struct resolve_cb_context *)context)->comp);
 705 }
 706
 707 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 708                                  const union ib_gid *dgid,
 709                                  u8 *dmac, u16 *vlan_id, int *if_index,
 710                                  int *hoplimit)
 711 {
 712         int ret = 0;
 713         struct rdma_dev_addr dev_addr;
 714         struct resolve_cb_context ctx;
 715         struct net_device *dev;
 716
 717         union {
 718                 struct sockaddr     _sockaddr;
 719                 struct sockaddr_in  _sockaddr_in;
 720                 struct sockaddr_in6 _sockaddr_in6;
 721         } sgid_addr, dgid_addr;
 722
 723
 724         rdma_gid2ip(&sgid_addr._sockaddr, sgid);
 725         rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 726
 727         memset(&dev_addr, 0, sizeof(dev_addr));
 728         if (if_index)
 729                 dev_addr.bound_dev_if = *if_index;
 730         dev_addr.net = &init_net;
 731
 732         ctx.addr = &dev_addr;
 733         init_completion(&ctx.comp);
 734         ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
 735                         &dev_addr, 1000, resolve_cb, &ctx);
 736         if (ret)
 737                 return ret;
 738
 739         wait_for_completion(&ctx.comp);
 740
 741         ret = ctx.status;
 742         if (ret)
 743                 return ret;
 744
 745         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 746         dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
 747         if (!dev)
 748                 return -ENODEV;
 749         if (if_index)
 750                 *if_index = dev_addr.bound_dev_if;
 751         if (vlan_id)
 752                 *vlan_id = rdma_vlan_dev_vlan_id(dev);
 753         if (hoplimit)
 754                 *hoplimit = dev_addr.hoplimit;
 755         dev_put(dev);
 756         return ret;
 757 }
 758 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
 759
 760 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
 761 {
 762         int ret = 0;
 763         struct rdma_dev_addr dev_addr;
 764         union {
 765                 struct sockaddr     _sockaddr;
 766                 struct sockaddr_in  _sockaddr_in;
 767                 struct sockaddr_in6 _sockaddr_in6;
 768         } gid_addr;
 769
 770         rdma_gid2ip(&gid_addr._sockaddr, sgid);
 771
 772         memset(&dev_addr, 0, sizeof(dev_addr));
 773         dev_addr.net = &init_net;
 774         ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
 775         if (ret)
 776                 return ret;
 777
 778         memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
 779         return ret;
 780 }
 781 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
 782
 783 static int netevent_callback(struct notifier_block *self, unsigned long event,
 784         void *ctx)
 785 {
 786         if (event == NETEVENT_NEIGH_UPDATE) {
 787                 struct neighbour *neigh = ctx;
 788
 789                 if (neigh->nud_state & NUD_VALID) {
 790                         set_timeout(jiffies);
 791                 }
 792         }
 793         return 0;
 794 }
 795
 796 static struct notifier_block nb = {
 797         .notifier_call = netevent_callback
 798 };
 799
 800 int addr_init(void)
 801 {
 802         addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
 803         if (!addr_wq)
 804                 return -ENOMEM;
 805
 806         register_netevent_notifier(&nb);
 807         rdma_addr_register_client(&self);
 808
 809         return 0;
 810 }
 811
 812 void addr_cleanup(void)
 813 {
 814         rdma_addr_unregister_client(&self);
 815         unregister_netevent_notifier(&nb);
 816         destroy_workqueue(addr_wq);
 817 }