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/ipv6_stubs.h>
  46 #include <net/ip6_route.h>
  47 #include <rdma/ib_addr.h>
  48 #include <rdma/ib_cache.h>
  49 #include <rdma/ib_sa.h>
  50 #include <rdma/ib.h>
  51 #include <rdma/rdma_netlink.h>
  52 #include <net/netlink.h>
  53
  54 #include "core_priv.h"
  55
  56 struct addr_req {
  57         struct list_head list;
  58         struct sockaddr_storage src_addr;
  59         struct sockaddr_storage dst_addr;
  60         struct rdma_dev_addr *addr;
  61         void *context;
  62         void (*callback)(int status, struct sockaddr *src_addr,
  63                          struct rdma_dev_addr *addr, void *context);
  64         unsigned long timeout;
  65         struct delayed_work work;
  66         bool resolve_by_gid_attr;       /* Consider gid attr in resolve phase */
  67         int status;
  68         u32 seq;
  69 };
  70
  71 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
  72
  73 static DEFINE_SPINLOCK(lock);
  74 static LIST_HEAD(req_list);
  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_deprecated(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         spin_lock_bh(&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         spin_unlock_bh(&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 nlmsghdr *nlh,
 133                              struct netlink_ext_ack *extack)
 134 {
 135         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
 136             !(NETLINK_CB(skb).sk))
 137                 return -EPERM;
 138
 139         if (ib_nl_is_good_ip_resp(nlh))
 140                 ib_nl_process_good_ip_rsep(nlh);
 141
 142         return 0;
 143 }
 144
 145 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
 146                              const void *daddr,
 147                              u32 seq, u16 family)
 148 {
 149         struct sk_buff *skb = NULL;
 150         struct nlmsghdr *nlh;
 151         struct rdma_ls_ip_resolve_header *header;
 152         void *data;
 153         size_t size;
 154         int attrtype;
 155         int len;
 156
 157         if (family == AF_INET) {
 158                 size = sizeof(struct in_addr);
 159                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
 160         } else {
 161                 size = sizeof(struct in6_addr);
 162                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
 163         }
 164
 165         len = nla_total_size(sizeof(size));
 166         len += NLMSG_ALIGN(sizeof(*header));
 167
 168         skb = nlmsg_new(len, GFP_KERNEL);
 169         if (!skb)
 170                 return -ENOMEM;
 171
 172         data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
 173                             RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
 174         if (!data) {
 175                 nlmsg_free(skb);
 176                 return -ENODATA;
 177         }
 178
 179         /* Construct the family header first */
 180         header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 181         header->ifindex = dev_addr->bound_dev_if;
 182         nla_put(skb, attrtype, size, daddr);
 183
 184         /* Repair the nlmsg header length */
 185         nlmsg_end(skb, nlh);
 186         rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
 187
 188         /* Make the request retry, so when we get the response from userspace
 189          * we will have something.
 190          */
 191         return -ENODATA;
 192 }
 193
 194 int rdma_addr_size(const struct sockaddr *addr)
 195 {
 196         switch (addr->sa_family) {
 197         case AF_INET:
 198                 return sizeof(struct sockaddr_in);
 199         case AF_INET6:
 200                 return sizeof(struct sockaddr_in6);
 201         case AF_IB:
 202                 return sizeof(struct sockaddr_ib);
 203         default:
 204                 return 0;
 205         }
 206 }
 207 EXPORT_SYMBOL(rdma_addr_size);
 208
 209 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
 210 {
 211         int ret = rdma_addr_size((struct sockaddr *) addr);
 212
 213         return ret <= sizeof(*addr) ? ret : 0;
 214 }
 215 EXPORT_SYMBOL(rdma_addr_size_in6);
 216
 217 int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
 218 {
 219         int ret = rdma_addr_size((struct sockaddr *) addr);
 220
 221         return ret <= sizeof(*addr) ? ret : 0;
 222 }
 223 EXPORT_SYMBOL(rdma_addr_size_kss);
 224
 225 /**
 226  * rdma_copy_src_l2_addr - Copy netdevice source addresses
 227  * @dev_addr:   Destination address pointer where to copy the addresses
 228  * @dev:        Netdevice whose source addresses to copy
 229  *
 230  * rdma_copy_src_l2_addr() copies source addresses from the specified netdevice.
 231  * This includes unicast address, broadcast address, device type and
 232  * interface index.
 233  */
 234 void rdma_copy_src_l2_addr(struct rdma_dev_addr *dev_addr,
 235                            const struct net_device *dev)
 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         dev_addr->bound_dev_if = dev->ifindex;
 241 }
 242 EXPORT_SYMBOL(rdma_copy_src_l2_addr);
 243
 244 static struct net_device *
 245 rdma_find_ndev_for_src_ip_rcu(struct net *net, const struct sockaddr *src_in)
 246 {
 247         struct net_device *dev = NULL;
 248         int ret = -EADDRNOTAVAIL;
 249
 250         switch (src_in->sa_family) {
 251         case AF_INET:
 252                 dev = __ip_dev_find(net,
 253                                     ((const struct sockaddr_in *)src_in)->sin_addr.s_addr,
 254                                     false);
 255                 if (dev)
 256                         ret = 0;
 257                 break;
 258 #if IS_ENABLED(CONFIG_IPV6)
 259         case AF_INET6:
 260                 for_each_netdev_rcu(net, dev) {
 261                         if (ipv6_chk_addr(net,
 262                                           &((const struct sockaddr_in6 *)src_in)->sin6_addr,
 263                                           dev, 1)) {
 264                                 ret = 0;
 265                                 break;
 266                         }
 267                 }
 268                 break;
 269 #endif
 270         }
 271         return ret ? ERR_PTR(ret) : dev;
 272 }
 273
 274 int rdma_translate_ip(const struct sockaddr *addr,
 275                       struct rdma_dev_addr *dev_addr)
 276 {
 277         struct net_device *dev;
 278
 279         if (dev_addr->bound_dev_if) {
 280                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 281                 if (!dev)
 282                         return -ENODEV;
 283                 rdma_copy_src_l2_addr(dev_addr, dev);
 284                 dev_put(dev);
 285                 return 0;
 286         }
 287
 288         rcu_read_lock();
 289         dev = rdma_find_ndev_for_src_ip_rcu(dev_addr->net, addr);
 290         if (!IS_ERR(dev))
 291                 rdma_copy_src_l2_addr(dev_addr, dev);
 292         rcu_read_unlock();
 293         return PTR_ERR_OR_ZERO(dev);
 294 }
 295 EXPORT_SYMBOL(rdma_translate_ip);
 296
 297 static void set_timeout(struct addr_req *req, 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, &req->work, delay);
 306 }
 307
 308 static void queue_req(struct addr_req *req)
 309 {
 310         spin_lock_bh(&lock);
 311         list_add_tail(&req->list, &req_list);
 312         set_timeout(req, req->timeout);
 313         spin_unlock_bh(&lock);
 314 }
 315
 316 static int ib_nl_fetch_ha(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         return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
 323 }
 324
 325 static int dst_fetch_ha(const struct dst_entry *dst,
 326                         struct rdma_dev_addr *dev_addr,
 327                         const void *daddr)
 328 {
 329         struct neighbour *n;
 330         int ret = 0;
 331
 332         n = dst_neigh_lookup(dst, daddr);
 333         if (!n)
 334                 return -ENODATA;
 335
 336         if (!(n->nud_state & NUD_VALID)) {
 337                 neigh_event_send(n, NULL);
 338                 ret = -ENODATA;
 339         } else {
 340                 neigh_ha_snapshot(dev_addr->dst_dev_addr, n, dst->dev);
 341         }
 342
 343         neigh_release(n);
 344
 345         return ret;
 346 }
 347
 348 static bool has_gateway(const struct dst_entry *dst, sa_family_t family)
 349 {
 350         struct rtable *rt;
 351         struct rt6_info *rt6;
 352
 353         if (family == AF_INET) {
 354                 rt = container_of(dst, struct rtable, dst);
 355                 return rt->rt_uses_gateway;
 356         }
 357
 358         rt6 = container_of(dst, struct rt6_info, dst);
 359         return rt6->rt6i_flags & RTF_GATEWAY;
 360 }
 361
 362 static int fetch_ha(const struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 363                     const struct sockaddr *dst_in, u32 seq)
 364 {
 365         const struct sockaddr_in *dst_in4 =
 366                 (const struct sockaddr_in *)dst_in;
 367         const struct sockaddr_in6 *dst_in6 =
 368                 (const struct sockaddr_in6 *)dst_in;
 369         const void *daddr = (dst_in->sa_family == AF_INET) ?
 370                 (const void *)&dst_in4->sin_addr.s_addr :
 371                 (const void *)&dst_in6->sin6_addr;
 372         sa_family_t family = dst_in->sa_family;
 373
 374         might_sleep();
 375
 376         /* If we have a gateway in IB mode then it must be an IB network */
 377         if (has_gateway(dst, family) && dev_addr->network == RDMA_NETWORK_IB)
 378                 return ib_nl_fetch_ha(dev_addr, daddr, seq, family);
 379         else
 380                 return dst_fetch_ha(dst, dev_addr, daddr);
 381 }
 382
 383 static int addr4_resolve(struct sockaddr *src_sock,
 384                          const struct sockaddr *dst_sock,
 385                          struct rdma_dev_addr *addr,
 386                          struct rtable **prt)
 387 {
 388         struct sockaddr_in *src_in = (struct sockaddr_in *)src_sock;
 389         const struct sockaddr_in *dst_in =
 390                         (const struct sockaddr_in *)dst_sock;
 391
 392         __be32 src_ip = src_in->sin_addr.s_addr;
 393         __be32 dst_ip = dst_in->sin_addr.s_addr;
 394         struct rtable *rt;
 395         struct flowi4 fl4;
 396         int ret;
 397
 398         memset(&fl4, 0, sizeof(fl4));
 399         fl4.daddr = dst_ip;
 400         fl4.saddr = src_ip;
 401         fl4.flowi4_oif = addr->bound_dev_if;
 402         rt = ip_route_output_key(addr->net, &fl4);
 403         ret = PTR_ERR_OR_ZERO(rt);
 404         if (ret)
 405                 return ret;
 406
 407         src_in->sin_addr.s_addr = fl4.saddr;
 408
 409         addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
 410
 411         *prt = rt;
 412         return 0;
 413 }
 414
 415 #if IS_ENABLED(CONFIG_IPV6)
 416 static int addr6_resolve(struct sockaddr *src_sock,
 417                          const struct sockaddr *dst_sock,
 418                          struct rdma_dev_addr *addr,
 419                          struct dst_entry **pdst)
 420 {
 421         struct sockaddr_in6 *src_in = (struct sockaddr_in6 *)src_sock;
 422         const struct sockaddr_in6 *dst_in =
 423                                 (const struct sockaddr_in6 *)dst_sock;
 424         struct flowi6 fl6;
 425         struct dst_entry *dst;
 426
 427         memset(&fl6, 0, sizeof fl6);
 428         fl6.daddr = dst_in->sin6_addr;
 429         fl6.saddr = src_in->sin6_addr;
 430         fl6.flowi6_oif = addr->bound_dev_if;
 431
 432         dst = ipv6_stub->ipv6_dst_lookup_flow(addr->net, NULL, &fl6, NULL);
 433         if (IS_ERR(dst))
 434                 return PTR_ERR(dst);
 435
 436         if (ipv6_addr_any(&src_in->sin6_addr))
 437                 src_in->sin6_addr = fl6.saddr;
 438
 439         addr->hoplimit = ip6_dst_hoplimit(dst);
 440
 441         *pdst = dst;
 442         return 0;
 443 }
 444 #else
 445 static int addr6_resolve(struct sockaddr *src_sock,
 446                          const struct sockaddr *dst_sock,
 447                          struct rdma_dev_addr *addr,
 448                          struct dst_entry **pdst)
 449 {
 450         return -EADDRNOTAVAIL;
 451 }
 452 #endif
 453
 454 static int addr_resolve_neigh(const struct dst_entry *dst,
 455                               const struct sockaddr *dst_in,
 456                               struct rdma_dev_addr *addr,
 457                               unsigned int ndev_flags,
 458                               u32 seq)
 459 {
 460         int ret = 0;
 461
 462         if (ndev_flags & IFF_LOOPBACK) {
 463                 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 464         } else {
 465                 if (!(ndev_flags & IFF_NOARP)) {
 466                         /* If the device doesn't do ARP internally */
 467                         ret = fetch_ha(dst, addr, dst_in, seq);
 468                 }
 469         }
 470         return ret;
 471 }
 472
 473 static int copy_src_l2_addr(struct rdma_dev_addr *dev_addr,
 474                             const struct sockaddr *dst_in,
 475                             const struct dst_entry *dst,
 476                             const struct net_device *ndev)
 477 {
 478         int ret = 0;
 479
 480         if (dst->dev->flags & IFF_LOOPBACK)
 481                 ret = rdma_translate_ip(dst_in, dev_addr);
 482         else
 483                 rdma_copy_src_l2_addr(dev_addr, dst->dev);
 484
 485         /*
 486          * If there's a gateway and type of device not ARPHRD_INFINIBAND,
 487          * we're definitely in RoCE v2 (as RoCE v1 isn't routable) set the
 488          * network type accordingly.
 489          */
 490         if (has_gateway(dst, dst_in->sa_family) &&
 491             ndev->type != ARPHRD_INFINIBAND)
 492                 dev_addr->network = dst_in->sa_family == AF_INET ?
 493                                                 RDMA_NETWORK_IPV4 :
 494                                                 RDMA_NETWORK_IPV6;
 495         else
 496                 dev_addr->network = RDMA_NETWORK_IB;
 497
 498         return ret;
 499 }
 500
 501 static int rdma_set_src_addr_rcu(struct rdma_dev_addr *dev_addr,
 502                                  unsigned int *ndev_flags,
 503                                  const struct sockaddr *dst_in,
 504                                  const struct dst_entry *dst)
 505 {
 506         struct net_device *ndev = READ_ONCE(dst->dev);
 507
 508         *ndev_flags = ndev->flags;
 509         /* A physical device must be the RDMA device to use */
 510         if (ndev->flags & IFF_LOOPBACK) {
 511                 /*
 512                  * RDMA (IB/RoCE, iWarp) doesn't run on lo interface or
 513                  * loopback IP address. So if route is resolved to loopback
 514                  * interface, translate that to a real ndev based on non
 515                  * loopback IP address.
 516                  */
 517                 ndev = rdma_find_ndev_for_src_ip_rcu(dev_net(ndev), dst_in);
 518                 if (IS_ERR(ndev))
 519                         return -ENODEV;
 520         }
 521
 522         return copy_src_l2_addr(dev_addr, dst_in, dst, ndev);
 523 }
 524
 525 static int set_addr_netns_by_gid_rcu(struct rdma_dev_addr *addr)
 526 {
 527         struct net_device *ndev;
 528
 529         ndev = rdma_read_gid_attr_ndev_rcu(addr->sgid_attr);
 530         if (IS_ERR(ndev))
 531                 return PTR_ERR(ndev);
 532
 533         /*
 534          * Since we are holding the rcu, reading net and ifindex
 535          * are safe without any additional reference; because
 536          * change_net_namespace() in net/core/dev.c does rcu sync
 537          * after it changes the state to IFF_DOWN and before
 538          * updating netdev fields {net, ifindex}.
 539          */
 540         addr->net = dev_net(ndev);
 541         addr->bound_dev_if = ndev->ifindex;
 542         return 0;
 543 }
 544
 545 static void rdma_addr_set_net_defaults(struct rdma_dev_addr *addr)
 546 {
 547         addr->net = &init_net;
 548         addr->bound_dev_if = 0;
 549 }
 550
 551 static int addr_resolve(struct sockaddr *src_in,
 552                         const struct sockaddr *dst_in,
 553                         struct rdma_dev_addr *addr,
 554                         bool resolve_neigh,
 555                         bool resolve_by_gid_attr,
 556                         u32 seq)
 557 {
 558         struct dst_entry *dst = NULL;
 559         unsigned int ndev_flags = 0;
 560         struct rtable *rt = NULL;
 561         int ret;
 562
 563         if (!addr->net) {
 564                 pr_warn_ratelimited("%s: missing namespace\n", __func__);
 565                 return -EINVAL;
 566         }
 567
 568         rcu_read_lock();
 569         if (resolve_by_gid_attr) {
 570                 if (!addr->sgid_attr) {
 571                         rcu_read_unlock();
 572                         pr_warn_ratelimited("%s: missing gid_attr\n", __func__);
 573                         return -EINVAL;
 574                 }
 575                 /*
 576                  * If the request is for a specific gid attribute of the
 577                  * rdma_dev_addr, derive net from the netdevice of the
 578                  * GID attribute.
 579                  */
 580                 ret = set_addr_netns_by_gid_rcu(addr);
 581                 if (ret) {
 582                         rcu_read_unlock();
 583                         return ret;
 584                 }
 585         }
 586         if (src_in->sa_family == AF_INET) {
 587                 ret = addr4_resolve(src_in, dst_in, addr, &rt);
 588                 dst = &rt->dst;
 589         } else {
 590                 ret = addr6_resolve(src_in, dst_in, addr, &dst);
 591         }
 592         if (ret) {
 593                 rcu_read_unlock();
 594                 goto done;
 595         }
 596         ret = rdma_set_src_addr_rcu(addr, &ndev_flags, dst_in, dst);
 597         rcu_read_unlock();
 598
 599         /*
 600          * Resolve neighbor destination address if requested and
 601          * only if src addr translation didn't fail.
 602          */
 603         if (!ret && resolve_neigh)
 604                 ret = addr_resolve_neigh(dst, dst_in, addr, ndev_flags, seq);
 605
 606         if (src_in->sa_family == AF_INET)
 607                 ip_rt_put(rt);
 608         else
 609                 dst_release(dst);
 610 done:
 611         /*
 612          * Clear the addr net to go back to its original state, only if it was
 613          * derived from GID attribute in this context.
 614          */
 615         if (resolve_by_gid_attr)
 616                 rdma_addr_set_net_defaults(addr);
 617         return ret;
 618 }
 619
 620 static void process_one_req(struct work_struct *_work)
 621 {
 622         struct addr_req *req;
 623         struct sockaddr *src_in, *dst_in;
 624
 625         req = container_of(_work, struct addr_req, work.work);
 626
 627         if (req->status == -ENODATA) {
 628                 src_in = (struct sockaddr *)&req->src_addr;
 629                 dst_in = (struct sockaddr *)&req->dst_addr;
 630                 req->status = addr_resolve(src_in, dst_in, req->addr,
 631                                            true, req->resolve_by_gid_attr,
 632                                            req->seq);
 633                 if (req->status && time_after_eq(jiffies, req->timeout)) {
 634                         req->status = -ETIMEDOUT;
 635                 } else if (req->status == -ENODATA) {
 636                         /* requeue the work for retrying again */
 637                         spin_lock_bh(&lock);
 638                         if (!list_empty(&req->list))
 639                                 set_timeout(req, req->timeout);
 640                         spin_unlock_bh(&lock);
 641                         return;
 642                 }
 643         }
 644
 645         req->callback(req->status, (struct sockaddr *)&req->src_addr,
 646                 req->addr, req->context);
 647         req->callback = NULL;
 648
 649         spin_lock_bh(&lock);
 650         /*
 651          * Although the work will normally have been canceled by the workqueue,
 652          * it can still be requeued as long as it is on the req_list.
 653          */
 654         cancel_delayed_work(&req->work);
 655         if (!list_empty(&req->list)) {
 656                 list_del_init(&req->list);
 657                 kfree(req);
 658         }
 659         spin_unlock_bh(&lock);
 660 }
 661
 662 int rdma_resolve_ip(struct sockaddr *src_addr, const struct sockaddr *dst_addr,
 663                     struct rdma_dev_addr *addr, unsigned long timeout_ms,
 664                     void (*callback)(int status, struct sockaddr *src_addr,
 665                                      struct rdma_dev_addr *addr, void *context),
 666                     bool resolve_by_gid_attr, void *context)
 667 {
 668         struct sockaddr *src_in, *dst_in;
 669         struct addr_req *req;
 670         int ret = 0;
 671
 672         req = kzalloc(sizeof *req, GFP_KERNEL);
 673         if (!req)
 674                 return -ENOMEM;
 675
 676         src_in = (struct sockaddr *) &req->src_addr;
 677         dst_in = (struct sockaddr *) &req->dst_addr;
 678
 679         if (src_addr) {
 680                 if (src_addr->sa_family != dst_addr->sa_family) {
 681                         ret = -EINVAL;
 682                         goto err;
 683                 }
 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         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 691         req->addr = addr;
 692         req->callback = callback;
 693         req->context = context;
 694         req->resolve_by_gid_attr = resolve_by_gid_attr;
 695         INIT_DELAYED_WORK(&req->work, process_one_req);
 696         req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 697
 698         req->status = addr_resolve(src_in, dst_in, addr, true,
 699                                    req->resolve_by_gid_attr, req->seq);
 700         switch (req->status) {
 701         case 0:
 702                 req->timeout = jiffies;
 703                 queue_req(req);
 704                 break;
 705         case -ENODATA:
 706                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 707                 queue_req(req);
 708                 break;
 709         default:
 710                 ret = req->status;
 711                 goto err;
 712         }
 713         return ret;
 714 err:
 715         kfree(req);
 716         return ret;
 717 }
 718 EXPORT_SYMBOL(rdma_resolve_ip);
 719
 720 int roce_resolve_route_from_path(struct sa_path_rec *rec,
 721                                  const struct ib_gid_attr *attr)
 722 {
 723         union {
 724                 struct sockaddr     _sockaddr;
 725                 struct sockaddr_in  _sockaddr_in;
 726                 struct sockaddr_in6 _sockaddr_in6;
 727         } sgid, dgid;
 728         struct rdma_dev_addr dev_addr = {};
 729         int ret;
 730
 731         might_sleep();
 732
 733         if (rec->roce.route_resolved)
 734                 return 0;
 735
 736         rdma_gid2ip((struct sockaddr *)&sgid, &rec->sgid);
 737         rdma_gid2ip((struct sockaddr *)&dgid, &rec->dgid);
 738
 739         if (sgid._sockaddr.sa_family != dgid._sockaddr.sa_family)
 740                 return -EINVAL;
 741
 742         if (!attr || !attr->ndev)
 743                 return -EINVAL;
 744
 745         dev_addr.net = &init_net;
 746         dev_addr.sgid_attr = attr;
 747
 748         ret = addr_resolve((struct sockaddr *)&sgid, (struct sockaddr *)&dgid,
 749                            &dev_addr, false, true, 0);
 750         if (ret)
 751                 return ret;
 752
 753         if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
 754              dev_addr.network == RDMA_NETWORK_IPV6) &&
 755             rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
 756                 return -EINVAL;
 757
 758         rec->roce.route_resolved = true;
 759         return 0;
 760 }
 761
 762 /**
 763  * rdma_addr_cancel - Cancel resolve ip request
 764  * @addr:       Pointer to address structure given previously
 765  *              during rdma_resolve_ip().
 766  * rdma_addr_cancel() is synchronous function which cancels any pending
 767  * request if there is any.
 768  */
 769 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 770 {
 771         struct addr_req *req, *temp_req;
 772         struct addr_req *found = NULL;
 773
 774         spin_lock_bh(&lock);
 775         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 776                 if (req->addr == addr) {
 777                         /*
 778                          * Removing from the list means we take ownership of
 779                          * the req
 780                          */
 781                         list_del_init(&req->list);
 782                         found = req;
 783                         break;
 784                 }
 785         }
 786         spin_unlock_bh(&lock);
 787
 788         if (!found)
 789                 return;
 790
 791         /*
 792          * sync canceling the work after removing it from the req_list
 793          * guarentees no work is running and none will be started.
 794          */
 795         cancel_delayed_work_sync(&found->work);
 796         kfree(found);
 797 }
 798 EXPORT_SYMBOL(rdma_addr_cancel);
 799
 800 struct resolve_cb_context {
 801         struct completion comp;
 802         int status;
 803 };
 804
 805 static void resolve_cb(int status, struct sockaddr *src_addr,
 806              struct rdma_dev_addr *addr, void *context)
 807 {
 808         ((struct resolve_cb_context *)context)->status = status;
 809         complete(&((struct resolve_cb_context *)context)->comp);
 810 }
 811
 812 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 813                                  const union ib_gid *dgid,
 814                                  u8 *dmac, const struct ib_gid_attr *sgid_attr,
 815                                  int *hoplimit)
 816 {
 817         struct rdma_dev_addr dev_addr;
 818         struct resolve_cb_context ctx;
 819         union {
 820                 struct sockaddr_in  _sockaddr_in;
 821                 struct sockaddr_in6 _sockaddr_in6;
 822         } sgid_addr, dgid_addr;
 823         int ret;
 824
 825         rdma_gid2ip((struct sockaddr *)&sgid_addr, sgid);
 826         rdma_gid2ip((struct sockaddr *)&dgid_addr, dgid);
 827
 828         memset(&dev_addr, 0, sizeof(dev_addr));
 829         dev_addr.net = &init_net;
 830         dev_addr.sgid_attr = sgid_attr;
 831
 832         init_completion(&ctx.comp);
 833         ret = rdma_resolve_ip((struct sockaddr *)&sgid_addr,
 834                               (struct sockaddr *)&dgid_addr, &dev_addr, 1000,
 835                               resolve_cb, true, &ctx);
 836         if (ret)
 837                 return ret;
 838
 839         wait_for_completion(&ctx.comp);
 840
 841         ret = ctx.status;
 842         if (ret)
 843                 return ret;
 844
 845         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 846         *hoplimit = dev_addr.hoplimit;
 847         return 0;
 848 }
 849
 850 static int netevent_callback(struct notifier_block *self, unsigned long event,
 851         void *ctx)
 852 {
 853         struct addr_req *req;
 854
 855         if (event == NETEVENT_NEIGH_UPDATE) {
 856                 struct neighbour *neigh = ctx;
 857
 858                 if (neigh->nud_state & NUD_VALID) {
 859                         spin_lock_bh(&lock);
 860                         list_for_each_entry(req, &req_list, list)
 861                                 set_timeout(req, jiffies);
 862                         spin_unlock_bh(&lock);
 863                 }
 864         }
 865         return 0;
 866 }
 867
 868 static struct notifier_block nb = {
 869         .notifier_call = netevent_callback
 870 };
 871
 872 int addr_init(void)
 873 {
 874         addr_wq = alloc_ordered_workqueue("ib_addr", 0);
 875         if (!addr_wq)
 876                 return -ENOMEM;
 877
 878         register_netevent_notifier(&nb);
 879
 880         return 0;
 881 }
 882
 883 void addr_cleanup(void)
 884 {
 885         unregister_netevent_notifier(&nb);
 886         destroy_workqueue(addr_wq);
 887         WARN_ON(!list_empty(&req_list));
 888 }