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         struct delayed_work work;
  65         int status;
  66         u32 seq;
  67 };
  68
  69 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
  70
  71 static void process_req(struct work_struct *work);
  72
  73 static DEFINE_MUTEX(lock);
  74 static LIST_HEAD(req_list);
  75 static DECLARE_DELAYED_WORK(work, process_req);
  76 static struct workqueue_struct *addr_wq;
  77
  78 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
  79         [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
  80                 .len = sizeof(struct rdma_nla_ls_gid)},
  81 };
  82
  83 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
  84 {
  85         struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
  86         int ret;
  87
  88         if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
  89                 return false;
  90
  91         ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
  92                         nlmsg_len(nlh), ib_nl_addr_policy, NULL);
  93         if (ret)
  94                 return false;
  95
  96         return true;
  97 }
  98
  99 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
 100 {
 101         const struct nlattr *head, *curr;
 102         union ib_gid gid;
 103         struct addr_req *req;
 104         int len, rem;
 105         int found = 0;
 106
 107         head = (const struct nlattr *)nlmsg_data(nlh);
 108         len = nlmsg_len(nlh);
 109
 110         nla_for_each_attr(curr, head, len, rem) {
 111                 if (curr->nla_type == LS_NLA_TYPE_DGID)
 112                         memcpy(&gid, nla_data(curr), nla_len(curr));
 113         }
 114
 115         mutex_lock(&lock);
 116         list_for_each_entry(req, &req_list, list) {
 117                 if (nlh->nlmsg_seq != req->seq)
 118                         continue;
 119                 /* We set the DGID part, the rest was set earlier */
 120                 rdma_addr_set_dgid(req->addr, &gid);
 121                 req->status = 0;
 122                 found = 1;
 123                 break;
 124         }
 125         mutex_unlock(&lock);
 126
 127         if (!found)
 128                 pr_info("Couldn't find request waiting for DGID: %pI6\n",
 129                         &gid);
 130 }
 131
 132 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
 133                              struct nlmsghdr *nlh,
 134                              struct netlink_ext_ack *extack)
 135 {
 136         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
 137             !(NETLINK_CB(skb).sk))
 138                 return -EPERM;
 139
 140         if (ib_nl_is_good_ip_resp(nlh))
 141                 ib_nl_process_good_ip_rsep(nlh);
 142
 143         return skb->len;
 144 }
 145
 146 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
 147                              const void *daddr,
 148                              u32 seq, u16 family)
 149 {
 150         struct sk_buff *skb = NULL;
 151         struct nlmsghdr *nlh;
 152         struct rdma_ls_ip_resolve_header *header;
 153         void *data;
 154         size_t size;
 155         int attrtype;
 156         int len;
 157
 158         if (family == AF_INET) {
 159                 size = sizeof(struct in_addr);
 160                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
 161         } else {
 162                 size = sizeof(struct in6_addr);
 163                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
 164         }
 165
 166         len = nla_total_size(sizeof(size));
 167         len += NLMSG_ALIGN(sizeof(*header));
 168
 169         skb = nlmsg_new(len, GFP_KERNEL);
 170         if (!skb)
 171                 return -ENOMEM;
 172
 173         data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
 174                             RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
 175         if (!data) {
 176                 nlmsg_free(skb);
 177                 return -ENODATA;
 178         }
 179
 180         /* Construct the family header first */
 181         header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 182         header->ifindex = dev_addr->bound_dev_if;
 183         nla_put(skb, attrtype, size, daddr);
 184
 185         /* Repair the nlmsg header length */
 186         nlmsg_end(skb, nlh);
 187         rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
 188
 189         /* Make the request retry, so when we get the response from userspace
 190          * we will have something.
 191          */
 192         return -ENODATA;
 193 }
 194
 195 int rdma_addr_size(struct sockaddr *addr)
 196 {
 197         switch (addr->sa_family) {
 198         case AF_INET:
 199                 return sizeof(struct sockaddr_in);
 200         case AF_INET6:
 201                 return sizeof(struct sockaddr_in6);
 202         case AF_IB:
 203                 return sizeof(struct sockaddr_ib);
 204         default:
 205                 return 0;
 206         }
 207 }
 208 EXPORT_SYMBOL(rdma_addr_size);
 209
 210 static struct rdma_addr_client self;
 211
 212 void rdma_addr_register_client(struct rdma_addr_client *client)
 213 {
 214         atomic_set(&client->refcount, 1);
 215         init_completion(&client->comp);
 216 }
 217 EXPORT_SYMBOL(rdma_addr_register_client);
 218
 219 static inline void put_client(struct rdma_addr_client *client)
 220 {
 221         if (atomic_dec_and_test(&client->refcount))
 222                 complete(&client->comp);
 223 }
 224
 225 void rdma_addr_unregister_client(struct rdma_addr_client *client)
 226 {
 227         put_client(client);
 228         wait_for_completion(&client->comp);
 229 }
 230 EXPORT_SYMBOL(rdma_addr_unregister_client);
 231
 232 void rdma_copy_addr(struct rdma_dev_addr *dev_addr,
 233                     const 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 }
 243 EXPORT_SYMBOL(rdma_copy_addr);
 244
 245 int rdma_translate_ip(const struct sockaddr *addr,
 246                       struct rdma_dev_addr *dev_addr)
 247 {
 248         struct net_device *dev;
 249
 250         if (dev_addr->bound_dev_if) {
 251                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 252                 if (!dev)
 253                         return -ENODEV;
 254                 rdma_copy_addr(dev_addr, dev, NULL);
 255                 dev_put(dev);
 256                 return 0;
 257         }
 258
 259         switch (addr->sa_family) {
 260         case AF_INET:
 261                 dev = ip_dev_find(dev_addr->net,
 262                         ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
 263
 264                 if (!dev)
 265                         return -EADDRNOTAVAIL;
 266
 267                 rdma_copy_addr(dev_addr, dev, NULL);
 268                 dev_put(dev);
 269                 break;
 270 #if IS_ENABLED(CONFIG_IPV6)
 271         case AF_INET6:
 272                 rcu_read_lock();
 273                 for_each_netdev_rcu(dev_addr->net, dev) {
 274                         if (ipv6_chk_addr(dev_addr->net,
 275                                           &((const struct sockaddr_in6 *)addr)->sin6_addr,
 276                                           dev, 1)) {
 277                                 rdma_copy_addr(dev_addr, dev, NULL);
 278                                 break;
 279                         }
 280                 }
 281                 rcu_read_unlock();
 282                 break;
 283 #endif
 284         }
 285         return 0;
 286 }
 287 EXPORT_SYMBOL(rdma_translate_ip);
 288
 289 static void set_timeout(struct delayed_work *delayed_work, unsigned long time)
 290 {
 291         unsigned long delay;
 292
 293         delay = time - jiffies;
 294         if ((long)delay < 0)
 295                 delay = 0;
 296
 297         mod_delayed_work(addr_wq, delayed_work, delay);
 298 }
 299
 300 static void queue_req(struct addr_req *req)
 301 {
 302         struct addr_req *temp_req;
 303
 304         mutex_lock(&lock);
 305         list_for_each_entry_reverse(temp_req, &req_list, list) {
 306                 if (time_after_eq(req->timeout, temp_req->timeout))
 307                         break;
 308         }
 309
 310         list_add(&req->list, &temp_req->list);
 311
 312         set_timeout(&req->work, req->timeout);
 313         mutex_unlock(&lock);
 314 }
 315
 316 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 317                           const void *daddr, u32 seq, u16 family)
 318 {
 319         if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
 320                 return -EADDRNOTAVAIL;
 321
 322         /* We fill in what we can, the response will fill the rest */
 323         rdma_copy_addr(dev_addr, dst->dev, NULL);
 324         return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
 325 }
 326
 327 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 328                         const void *daddr)
 329 {
 330         struct neighbour *n;
 331         int ret = 0;
 332
 333         n = dst_neigh_lookup(dst, daddr);
 334
 335         rcu_read_lock();
 336         if (!n || !(n->nud_state & NUD_VALID)) {
 337                 if (n)
 338                         neigh_event_send(n, NULL);
 339                 ret = -ENODATA;
 340         } else {
 341                 rdma_copy_addr(dev_addr, dst->dev, n->ha);
 342         }
 343         rcu_read_unlock();
 344
 345         if (n)
 346                 neigh_release(n);
 347
 348         return ret;
 349 }
 350
 351 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
 352 {
 353         struct rtable *rt;
 354         struct rt6_info *rt6;
 355
 356         if (family == AF_INET) {
 357                 rt = container_of(dst, struct rtable, dst);
 358                 return rt->rt_uses_gateway;
 359         }
 360
 361         rt6 = container_of(dst, struct rt6_info, dst);
 362         return rt6->rt6i_flags & RTF_GATEWAY;
 363 }
 364
 365 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 366                     const struct sockaddr *dst_in, u32 seq)
 367 {
 368         const struct sockaddr_in *dst_in4 =
 369                 (const struct sockaddr_in *)dst_in;
 370         const struct sockaddr_in6 *dst_in6 =
 371                 (const struct sockaddr_in6 *)dst_in;
 372         const void *daddr = (dst_in->sa_family == AF_INET) ?
 373                 (const void *)&dst_in4->sin_addr.s_addr :
 374                 (const void *)&dst_in6->sin6_addr;
 375         sa_family_t family = dst_in->sa_family;
 376
 377         /* Gateway + ARPHRD_INFINIBAND -> IB router */
 378         if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
 379                 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
 380         else
 381                 return dst_fetch_ha(dst, dev_addr, daddr);
 382 }
 383
 384 static int addr4_resolve(struct sockaddr_in *src_in,
 385                          const struct sockaddr_in *dst_in,
 386                          struct rdma_dev_addr *addr,
 387                          struct rtable **prt)
 388 {
 389         __be32 src_ip = src_in->sin_addr.s_addr;
 390         __be32 dst_ip = dst_in->sin_addr.s_addr;
 391         struct rtable *rt;
 392         struct flowi4 fl4;
 393         int ret;
 394
 395         memset(&fl4, 0, sizeof(fl4));
 396         fl4.daddr = dst_ip;
 397         fl4.saddr = src_ip;
 398         fl4.flowi4_oif = addr->bound_dev_if;
 399         rt = ip_route_output_key(addr->net, &fl4);
 400         ret = PTR_ERR_OR_ZERO(rt);
 401         if (ret)
 402                 return ret;
 403
 404         src_in->sin_family = AF_INET;
 405         src_in->sin_addr.s_addr = fl4.saddr;
 406
 407         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 408          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 409          * type accordingly.
 410          */
 411         if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
 412                 addr->network = RDMA_NETWORK_IPV4;
 413
 414         addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
 415
 416         *prt = rt;
 417         return 0;
 418 }
 419
 420 #if IS_ENABLED(CONFIG_IPV6)
 421 static int addr6_resolve(struct sockaddr_in6 *src_in,
 422                          const struct sockaddr_in6 *dst_in,
 423                          struct rdma_dev_addr *addr,
 424                          struct dst_entry **pdst)
 425 {
 426         struct flowi6 fl6;
 427         struct dst_entry *dst;
 428         struct rt6_info *rt;
 429         int ret;
 430
 431         memset(&fl6, 0, sizeof fl6);
 432         fl6.daddr = dst_in->sin6_addr;
 433         fl6.saddr = src_in->sin6_addr;
 434         fl6.flowi6_oif = addr->bound_dev_if;
 435
 436         ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
 437         if (ret < 0)
 438                 return ret;
 439
 440         rt = (struct rt6_info *)dst;
 441         if (ipv6_addr_any(&src_in->sin6_addr)) {
 442                 src_in->sin6_family = AF_INET6;
 443                 src_in->sin6_addr = fl6.saddr;
 444         }
 445
 446         /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
 447          * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
 448          * type accordingly.
 449          */
 450         if (rt->rt6i_flags & RTF_GATEWAY &&
 451             ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
 452                 addr->network = RDMA_NETWORK_IPV6;
 453
 454         addr->hoplimit = ip6_dst_hoplimit(dst);
 455
 456         *pdst = dst;
 457         return 0;
 458 }
 459 #else
 460 static int addr6_resolve(struct sockaddr_in6 *src_in,
 461                          const struct sockaddr_in6 *dst_in,
 462                          struct rdma_dev_addr *addr,
 463                          struct dst_entry **pdst)
 464 {
 465         return -EADDRNOTAVAIL;
 466 }
 467 #endif
 468
 469 static int addr_resolve_neigh(struct dst_entry *dst,
 470                               const struct sockaddr *dst_in,
 471                               struct rdma_dev_addr *addr,
 472                               u32 seq)
 473 {
 474         if (dst->dev->flags & IFF_LOOPBACK) {
 475                 int ret;
 476
 477                 ret = rdma_translate_ip(dst_in, addr);
 478                 if (!ret)
 479                         memcpy(addr->dst_dev_addr, addr->src_dev_addr,
 480                                MAX_ADDR_LEN);
 481
 482                 return ret;
 483         }
 484
 485         /* If the device doesn't do ARP internally */
 486         if (!(dst->dev->flags & IFF_NOARP))
 487                 return fetch_ha(dst, addr, dst_in, seq);
 488
 489         rdma_copy_addr(addr, dst->dev, NULL);
 490
 491         return 0;
 492 }
 493
 494 static int addr_resolve(struct sockaddr *src_in,
 495                         const struct sockaddr *dst_in,
 496                         struct rdma_dev_addr *addr,
 497                         bool resolve_neigh,
 498                         u32 seq)
 499 {
 500         struct net_device *ndev;
 501         struct dst_entry *dst;
 502         int ret;
 503
 504         if (!addr->net) {
 505                 pr_warn_ratelimited("%s: missing namespace\n", __func__);
 506                 return -EINVAL;
 507         }
 508
 509         if (src_in->sa_family == AF_INET) {
 510                 struct rtable *rt = NULL;
 511                 const struct sockaddr_in *dst_in4 =
 512                         (const struct sockaddr_in *)dst_in;
 513
 514                 ret = addr4_resolve((struct sockaddr_in *)src_in,
 515                                     dst_in4, addr, &rt);
 516                 if (ret)
 517                         return ret;
 518
 519                 if (resolve_neigh)
 520                         ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
 521
 522                 if (addr->bound_dev_if) {
 523                         ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
 524                 } else {
 525                         ndev = rt->dst.dev;
 526                         dev_hold(ndev);
 527                 }
 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                 if (addr->bound_dev_if) {
 544                         ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
 545                 } else {
 546                         ndev = dst->dev;
 547                         dev_hold(ndev);
 548                 }
 549
 550                 dst_release(dst);
 551         }
 552
 553         if (ndev->flags & IFF_LOOPBACK) {
 554                 ret = rdma_translate_ip(dst_in, addr);
 555                 /*
 556                  * Put the loopback device and get the translated
 557                  * device instead.
 558                  */
 559                 dev_put(ndev);
 560                 ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
 561         } else {
 562                 addr->bound_dev_if = ndev->ifindex;
 563         }
 564         dev_put(ndev);
 565
 566         return ret;
 567 }
 568
 569 static void process_one_req(struct work_struct *_work)
 570 {
 571         struct addr_req *req;
 572         struct sockaddr *src_in, *dst_in;
 573
 574         mutex_lock(&lock);
 575         req = container_of(_work, struct addr_req, work.work);
 576
 577         if (req->status == -ENODATA) {
 578                 src_in = (struct sockaddr *)&req->src_addr;
 579                 dst_in = (struct sockaddr *)&req->dst_addr;
 580                 req->status = addr_resolve(src_in, dst_in, req->addr,
 581                                            true, req->seq);
 582                 if (req->status && time_after_eq(jiffies, req->timeout)) {
 583                         req->status = -ETIMEDOUT;
 584                 } else if (req->status == -ENODATA) {
 585                         /* requeue the work for retrying again */
 586                         set_timeout(&req->work, req->timeout);
 587                         mutex_unlock(&lock);
 588                         return;
 589                 }
 590         }
 591         list_del(&req->list);
 592         mutex_unlock(&lock);
 593
 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 static void process_req(struct work_struct *work)
 601 {
 602         struct addr_req *req, *temp_req;
 603         struct sockaddr *src_in, *dst_in;
 604         struct list_head done_list;
 605
 606         INIT_LIST_HEAD(&done_list);
 607
 608         mutex_lock(&lock);
 609         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 610                 if (req->status == -ENODATA) {
 611                         src_in = (struct sockaddr *) &req->src_addr;
 612                         dst_in = (struct sockaddr *) &req->dst_addr;
 613                         req->status = addr_resolve(src_in, dst_in, req->addr,
 614                                                    true, req->seq);
 615                         if (req->status && time_after_eq(jiffies, req->timeout))
 616                                 req->status = -ETIMEDOUT;
 617                         else if (req->status == -ENODATA) {
 618                                 set_timeout(&req->work, req->timeout);
 619                                 continue;
 620                         }
 621                 }
 622                 list_move_tail(&req->list, &done_list);
 623         }
 624
 625         mutex_unlock(&lock);
 626
 627         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 628                 list_del(&req->list);
 629                 /* It is safe to cancel other work items from this work item
 630                  * because at a time there can be only one work item running
 631                  * with this single threaded work queue.
 632                  */
 633                 cancel_delayed_work(&req->work);
 634                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 635                         req->addr, req->context);
 636                 put_client(req->client);
 637                 kfree(req);
 638         }
 639 }
 640
 641 int rdma_resolve_ip(struct rdma_addr_client *client,
 642                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 643                     struct rdma_dev_addr *addr, int timeout_ms,
 644                     void (*callback)(int status, struct sockaddr *src_addr,
 645                                      struct rdma_dev_addr *addr, void *context),
 646                     void *context)
 647 {
 648         struct sockaddr *src_in, *dst_in;
 649         struct addr_req *req;
 650         int ret = 0;
 651
 652         req = kzalloc(sizeof *req, GFP_KERNEL);
 653         if (!req)
 654                 return -ENOMEM;
 655
 656         src_in = (struct sockaddr *) &req->src_addr;
 657         dst_in = (struct sockaddr *) &req->dst_addr;
 658
 659         if (src_addr) {
 660                 if (src_addr->sa_family != dst_addr->sa_family) {
 661                         ret = -EINVAL;
 662                         goto err;
 663                 }
 664
 665                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 666         } else {
 667                 src_in->sa_family = dst_addr->sa_family;
 668         }
 669
 670         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 671         req->addr = addr;
 672         req->callback = callback;
 673         req->context = context;
 674         req->client = client;
 675         atomic_inc(&client->refcount);
 676         INIT_DELAYED_WORK(&req->work, process_one_req);
 677         req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 678
 679         req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
 680         switch (req->status) {
 681         case 0:
 682                 req->timeout = jiffies;
 683                 queue_req(req);
 684                 break;
 685         case -ENODATA:
 686                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 687                 queue_req(req);
 688                 break;
 689         default:
 690                 ret = req->status;
 691                 atomic_dec(&client->refcount);
 692                 goto err;
 693         }
 694         return ret;
 695 err:
 696         kfree(req);
 697         return ret;
 698 }
 699 EXPORT_SYMBOL(rdma_resolve_ip);
 700
 701 int rdma_resolve_ip_route(struct sockaddr *src_addr,
 702                           const struct sockaddr *dst_addr,
 703                           struct rdma_dev_addr *addr)
 704 {
 705         struct sockaddr_storage ssrc_addr = {};
 706         struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
 707
 708         if (src_addr) {
 709                 if (src_addr->sa_family != dst_addr->sa_family)
 710                         return -EINVAL;
 711
 712                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 713         } else {
 714                 src_in->sa_family = dst_addr->sa_family;
 715         }
 716
 717         return addr_resolve(src_in, dst_addr, addr, false, 0);
 718 }
 719 EXPORT_SYMBOL(rdma_resolve_ip_route);
 720
 721 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 722 {
 723         struct addr_req *req, *temp_req;
 724
 725         mutex_lock(&lock);
 726         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 727                 if (req->addr == addr) {
 728                         req->status = -ECANCELED;
 729                         req->timeout = jiffies;
 730                         list_move(&req->list, &req_list);
 731                         set_timeout(&req->work, req->timeout);
 732                         break;
 733                 }
 734         }
 735         mutex_unlock(&lock);
 736 }
 737 EXPORT_SYMBOL(rdma_addr_cancel);
 738
 739 struct resolve_cb_context {
 740         struct completion comp;
 741         int status;
 742 };
 743
 744 static void resolve_cb(int status, struct sockaddr *src_addr,
 745              struct rdma_dev_addr *addr, void *context)
 746 {
 747         ((struct resolve_cb_context *)context)->status = status;
 748         complete(&((struct resolve_cb_context *)context)->comp);
 749 }
 750
 751 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 752                                  const union ib_gid *dgid,
 753                                  u8 *dmac, const struct net_device *ndev,
 754                                  int *hoplimit)
 755 {
 756         struct rdma_dev_addr dev_addr;
 757         struct resolve_cb_context ctx;
 758         union {
 759                 struct sockaddr     _sockaddr;
 760                 struct sockaddr_in  _sockaddr_in;
 761                 struct sockaddr_in6 _sockaddr_in6;
 762         } sgid_addr, dgid_addr;
 763         int ret;
 764
 765         rdma_gid2ip(&sgid_addr._sockaddr, sgid);
 766         rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 767
 768         memset(&dev_addr, 0, sizeof(dev_addr));
 769         dev_addr.bound_dev_if = ndev->ifindex;
 770         dev_addr.net = &init_net;
 771
 772         init_completion(&ctx.comp);
 773         ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
 774                         &dev_addr, 1000, resolve_cb, &ctx);
 775         if (ret)
 776                 return ret;
 777
 778         wait_for_completion(&ctx.comp);
 779
 780         ret = ctx.status;
 781         if (ret)
 782                 return ret;
 783
 784         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 785         *hoplimit = dev_addr.hoplimit;
 786         return 0;
 787 }
 788
 789 static int netevent_callback(struct notifier_block *self, unsigned long event,
 790         void *ctx)
 791 {
 792         if (event == NETEVENT_NEIGH_UPDATE) {
 793                 struct neighbour *neigh = ctx;
 794
 795                 if (neigh->nud_state & NUD_VALID)
 796                         set_timeout(&work, jiffies);
 797         }
 798         return 0;
 799 }
 800
 801 static struct notifier_block nb = {
 802         .notifier_call = netevent_callback
 803 };
 804
 805 int addr_init(void)
 806 {
 807         addr_wq = alloc_ordered_workqueue("ib_addr", 0);
 808         if (!addr_wq)
 809                 return -ENOMEM;
 810
 811         register_netevent_notifier(&nb);
 812         rdma_addr_register_client(&self);
 813
 814         return 0;
 815 }
 816
 817 void addr_cleanup(void)
 818 {
 819         rdma_addr_unregister_client(&self);
 820         unregister_netevent_notifier(&nb);
 821         destroy_workqueue(addr_wq);
 822 }