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_sa.h>
  49 #include <rdma/ib.h>
  50 #include <rdma/rdma_netlink.h>
  51 #include <net/netlink.h>
  52
  53 #include "core_priv.h"
  54
  55 struct addr_req {
  56         struct list_head list;
  57         struct sockaddr_storage src_addr;
  58         struct sockaddr_storage dst_addr;
  59         struct rdma_dev_addr *addr;
  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         bool resolve_by_gid_attr;       /* Consider gid attr in resolve phase */
  66         int status;
  67         u32 seq;
  68 };
  69
  70 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
  71
  72 static DEFINE_SPINLOCK(lock);
  73 static LIST_HEAD(req_list);
  74 static struct workqueue_struct *addr_wq;
  75
  76 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
  77         [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
  78                 .len = sizeof(struct rdma_nla_ls_gid)},
  79 };
  80
  81 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
  82 {
  83         struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
  84         int ret;
  85
  86         if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
  87                 return false;
  88
  89         ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
  90                                    nlmsg_len(nlh), ib_nl_addr_policy, NULL);
  91         if (ret)
  92                 return false;
  93
  94         return true;
  95 }
  96
  97 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
  98 {
  99         const struct nlattr *head, *curr;
 100         union ib_gid gid;
 101         struct addr_req *req;
 102         int len, rem;
 103         int found = 0;
 104
 105         head = (const struct nlattr *)nlmsg_data(nlh);
 106         len = nlmsg_len(nlh);
 107
 108         nla_for_each_attr(curr, head, len, rem) {
 109                 if (curr->nla_type == LS_NLA_TYPE_DGID)
 110                         memcpy(&gid, nla_data(curr), nla_len(curr));
 111         }
 112
 113         spin_lock_bh(&lock);
 114         list_for_each_entry(req, &req_list, list) {
 115                 if (nlh->nlmsg_seq != req->seq)
 116                         continue;
 117                 /* We set the DGID part, the rest was set earlier */
 118                 rdma_addr_set_dgid(req->addr, &gid);
 119                 req->status = 0;
 120                 found = 1;
 121                 break;
 122         }
 123         spin_unlock_bh(&lock);
 124
 125         if (!found)
 126                 pr_info("Couldn't find request waiting for DGID: %pI6\n",
 127                         &gid);
 128 }
 129
 130 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
 131                              struct nlmsghdr *nlh,
 132                              struct netlink_ext_ack *extack)
 133 {
 134         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
 135             !(NETLINK_CB(skb).sk))
 136                 return -EPERM;
 137
 138         if (ib_nl_is_good_ip_resp(nlh))
 139                 ib_nl_process_good_ip_rsep(nlh);
 140
 141         return skb->len;
 142 }
 143
 144 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
 145                              const void *daddr,
 146                              u32 seq, u16 family)
 147 {
 148         struct sk_buff *skb = NULL;
 149         struct nlmsghdr *nlh;
 150         struct rdma_ls_ip_resolve_header *header;
 151         void *data;
 152         size_t size;
 153         int attrtype;
 154         int len;
 155
 156         if (family == AF_INET) {
 157                 size = sizeof(struct in_addr);
 158                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
 159         } else {
 160                 size = sizeof(struct in6_addr);
 161                 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
 162         }
 163
 164         len = nla_total_size(sizeof(size));
 165         len += NLMSG_ALIGN(sizeof(*header));
 166
 167         skb = nlmsg_new(len, GFP_KERNEL);
 168         if (!skb)
 169                 return -ENOMEM;
 170
 171         data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
 172                             RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
 173         if (!data) {
 174                 nlmsg_free(skb);
 175                 return -ENODATA;
 176         }
 177
 178         /* Construct the family header first */
 179         header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 180         header->ifindex = dev_addr->bound_dev_if;
 181         nla_put(skb, attrtype, size, daddr);
 182
 183         /* Repair the nlmsg header length */
 184         nlmsg_end(skb, nlh);
 185         rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
 186
 187         /* Make the request retry, so when we get the response from userspace
 188          * we will have something.
 189          */
 190         return -ENODATA;
 191 }
 192
 193 int rdma_addr_size(const struct sockaddr *addr)
 194 {
 195         switch (addr->sa_family) {
 196         case AF_INET:
 197                 return sizeof(struct sockaddr_in);
 198         case AF_INET6:
 199                 return sizeof(struct sockaddr_in6);
 200         case AF_IB:
 201                 return sizeof(struct sockaddr_ib);
 202         default:
 203                 return 0;
 204         }
 205 }
 206 EXPORT_SYMBOL(rdma_addr_size);
 207
 208 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
 209 {
 210         int ret = rdma_addr_size((struct sockaddr *) addr);
 211
 212         return ret <= sizeof(*addr) ? ret : 0;
 213 }
 214 EXPORT_SYMBOL(rdma_addr_size_in6);
 215
 216 int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
 217 {
 218         int ret = rdma_addr_size((struct sockaddr *) addr);
 219
 220         return ret <= sizeof(*addr) ? ret : 0;
 221 }
 222 EXPORT_SYMBOL(rdma_addr_size_kss);
 223
 224 /**
 225  * rdma_copy_src_l2_addr - Copy netdevice source addresses
 226  * @dev_addr:   Destination address pointer where to copy the addresses
 227  * @dev:        Netdevice whose source addresses to copy
 228  *
 229  * rdma_copy_src_l2_addr() copies source addresses from the specified netdevice.
 230  * This includes unicast address, broadcast address, device type and
 231  * interface index.
 232  */
 233 void rdma_copy_src_l2_addr(struct rdma_dev_addr *dev_addr,
 234                            const struct net_device *dev)
 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         dev_addr->bound_dev_if = dev->ifindex;
 240 }
 241 EXPORT_SYMBOL(rdma_copy_src_l2_addr);
 242
 243 static struct net_device *
 244 rdma_find_ndev_for_src_ip_rcu(struct net *net, const struct sockaddr *src_in)
 245 {
 246         struct net_device *dev = NULL;
 247         int ret = -EADDRNOTAVAIL;
 248
 249         switch (src_in->sa_family) {
 250         case AF_INET:
 251                 dev = __ip_dev_find(net,
 252                                     ((const struct sockaddr_in *)src_in)->sin_addr.s_addr,
 253                                     false);
 254                 if (dev)
 255                         ret = 0;
 256                 break;
 257 #if IS_ENABLED(CONFIG_IPV6)
 258         case AF_INET6:
 259                 for_each_netdev_rcu(net, dev) {
 260                         if (ipv6_chk_addr(net,
 261                                           &((const struct sockaddr_in6 *)src_in)->sin6_addr,
 262                                           dev, 1)) {
 263                                 ret = 0;
 264                                 break;
 265                         }
 266                 }
 267                 break;
 268 #endif
 269         }
 270         return ret ? ERR_PTR(ret) : dev;
 271 }
 272
 273 int rdma_translate_ip(const struct sockaddr *addr,
 274                       struct rdma_dev_addr *dev_addr)
 275 {
 276         struct net_device *dev;
 277
 278         if (dev_addr->bound_dev_if) {
 279                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 280                 if (!dev)
 281                         return -ENODEV;
 282                 rdma_copy_src_l2_addr(dev_addr, dev);
 283                 dev_put(dev);
 284                 return 0;
 285         }
 286
 287         rcu_read_lock();
 288         dev = rdma_find_ndev_for_src_ip_rcu(dev_addr->net, addr);
 289         if (!IS_ERR(dev))
 290                 rdma_copy_src_l2_addr(dev_addr, dev);
 291         rcu_read_unlock();
 292         return PTR_ERR_OR_ZERO(dev);
 293 }
 294 EXPORT_SYMBOL(rdma_translate_ip);
 295
 296 static void set_timeout(struct addr_req *req, unsigned long time)
 297 {
 298         unsigned long delay;
 299
 300         delay = time - jiffies;
 301         if ((long)delay < 0)
 302                 delay = 0;
 303
 304         mod_delayed_work(addr_wq, &req->work, delay);
 305 }
 306
 307 static void queue_req(struct addr_req *req)
 308 {
 309         spin_lock_bh(&lock);
 310         list_add_tail(&req->list, &req_list);
 311         set_timeout(req, req->timeout);
 312         spin_unlock_bh(&lock);
 313 }
 314
 315 static int ib_nl_fetch_ha(struct rdma_dev_addr *dev_addr,
 316                           const void *daddr, u32 seq, u16 family)
 317 {
 318         if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
 319                 return -EADDRNOTAVAIL;
 320
 321         return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
 322 }
 323
 324 static int dst_fetch_ha(const struct dst_entry *dst,
 325                         struct rdma_dev_addr *dev_addr,
 326                         const void *daddr)
 327 {
 328         struct neighbour *n;
 329         int ret = 0;
 330
 331         n = dst_neigh_lookup(dst, daddr);
 332         if (!n)
 333                 return -ENODATA;
 334
 335         if (!(n->nud_state & NUD_VALID)) {
 336                 neigh_event_send(n, NULL);
 337                 ret = -ENODATA;
 338         } else {
 339                 memcpy(dev_addr->dst_dev_addr, n->ha, MAX_ADDR_LEN);
 340         }
 341
 342         neigh_release(n);
 343
 344         return ret;
 345 }
 346
 347 static bool has_gateway(const struct dst_entry *dst, sa_family_t family)
 348 {
 349         struct rtable *rt;
 350         struct rt6_info *rt6;
 351
 352         if (family == AF_INET) {
 353                 rt = container_of(dst, struct rtable, dst);
 354                 return rt->rt_gw_family == AF_INET;
 355         }
 356
 357         rt6 = container_of(dst, struct rt6_info, dst);
 358         return rt6->rt6i_flags & RTF_GATEWAY;
 359 }
 360
 361 static int fetch_ha(const struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
 362                     const struct sockaddr *dst_in, u32 seq)
 363 {
 364         const struct sockaddr_in *dst_in4 =
 365                 (const struct sockaddr_in *)dst_in;
 366         const struct sockaddr_in6 *dst_in6 =
 367                 (const struct sockaddr_in6 *)dst_in;
 368         const void *daddr = (dst_in->sa_family == AF_INET) ?
 369                 (const void *)&dst_in4->sin_addr.s_addr :
 370                 (const void *)&dst_in6->sin6_addr;
 371         sa_family_t family = dst_in->sa_family;
 372
 373         /* If we have a gateway in IB mode then it must be an IB network */
 374         if (has_gateway(dst, family) && dev_addr->network == RDMA_NETWORK_IB)
 375                 return ib_nl_fetch_ha(dev_addr, daddr, seq, family);
 376         else
 377                 return dst_fetch_ha(dst, dev_addr, daddr);
 378 }
 379
 380 static int addr4_resolve(struct sockaddr *src_sock,
 381                          const struct sockaddr *dst_sock,
 382                          struct rdma_dev_addr *addr,
 383                          struct rtable **prt)
 384 {
 385         struct sockaddr_in *src_in = (struct sockaddr_in *)src_sock;
 386         const struct sockaddr_in *dst_in =
 387                         (const struct sockaddr_in *)dst_sock;
 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_addr.s_addr = fl4.saddr;
 405
 406         addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
 407
 408         *prt = rt;
 409         return 0;
 410 }
 411
 412 #if IS_ENABLED(CONFIG_IPV6)
 413 static int addr6_resolve(struct sockaddr *src_sock,
 414                          const struct sockaddr *dst_sock,
 415                          struct rdma_dev_addr *addr,
 416                          struct dst_entry **pdst)
 417 {
 418         struct sockaddr_in6 *src_in = (struct sockaddr_in6 *)src_sock;
 419         const struct sockaddr_in6 *dst_in =
 420                                 (const struct sockaddr_in6 *)dst_sock;
 421         struct flowi6 fl6;
 422         struct dst_entry *dst;
 423         int ret;
 424
 425         memset(&fl6, 0, sizeof fl6);
 426         fl6.daddr = dst_in->sin6_addr;
 427         fl6.saddr = src_in->sin6_addr;
 428         fl6.flowi6_oif = addr->bound_dev_if;
 429
 430         ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
 431         if (ret < 0)
 432                 return ret;
 433
 434         if (ipv6_addr_any(&src_in->sin6_addr))
 435                 src_in->sin6_addr = fl6.saddr;
 436
 437         addr->hoplimit = ip6_dst_hoplimit(dst);
 438
 439         *pdst = dst;
 440         return 0;
 441 }
 442 #else
 443 static int addr6_resolve(struct sockaddr *src_sock,
 444                          const struct sockaddr *dst_sock,
 445                          struct rdma_dev_addr *addr,
 446                          struct dst_entry **pdst)
 447 {
 448         return -EADDRNOTAVAIL;
 449 }
 450 #endif
 451
 452 static int addr_resolve_neigh(const struct dst_entry *dst,
 453                               const struct sockaddr *dst_in,
 454                               struct rdma_dev_addr *addr,
 455                               unsigned int ndev_flags,
 456                               u32 seq)
 457 {
 458         int ret = 0;
 459
 460         if (ndev_flags & IFF_LOOPBACK) {
 461                 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 462         } else {
 463                 if (!(ndev_flags & IFF_NOARP)) {
 464                         /* If the device doesn't do ARP internally */
 465                         ret = fetch_ha(dst, addr, dst_in, seq);
 466                 }
 467         }
 468         return ret;
 469 }
 470
 471 static int copy_src_l2_addr(struct rdma_dev_addr *dev_addr,
 472                             const struct sockaddr *dst_in,
 473                             const struct dst_entry *dst,
 474                             const struct net_device *ndev)
 475 {
 476         int ret = 0;
 477
 478         if (dst->dev->flags & IFF_LOOPBACK)
 479                 ret = rdma_translate_ip(dst_in, dev_addr);
 480         else
 481                 rdma_copy_src_l2_addr(dev_addr, dst->dev);
 482
 483         /*
 484          * If there's a gateway and type of device not ARPHRD_INFINIBAND,
 485          * we're definitely in RoCE v2 (as RoCE v1 isn't routable) set the
 486          * network type accordingly.
 487          */
 488         if (has_gateway(dst, dst_in->sa_family) &&
 489             ndev->type != ARPHRD_INFINIBAND)
 490                 dev_addr->network = dst_in->sa_family == AF_INET ?
 491                                                 RDMA_NETWORK_IPV4 :
 492                                                 RDMA_NETWORK_IPV6;
 493         else
 494                 dev_addr->network = RDMA_NETWORK_IB;
 495
 496         return ret;
 497 }
 498
 499 static int rdma_set_src_addr_rcu(struct rdma_dev_addr *dev_addr,
 500                                  unsigned int *ndev_flags,
 501                                  const struct sockaddr *dst_in,
 502                                  const struct dst_entry *dst)
 503 {
 504         struct net_device *ndev = READ_ONCE(dst->dev);
 505
 506         *ndev_flags = ndev->flags;
 507         /* A physical device must be the RDMA device to use */
 508         if (ndev->flags & IFF_LOOPBACK) {
 509                 /*
 510                  * RDMA (IB/RoCE, iWarp) doesn't run on lo interface or
 511                  * loopback IP address. So if route is resolved to loopback
 512                  * interface, translate that to a real ndev based on non
 513                  * loopback IP address.
 514                  */
 515                 ndev = rdma_find_ndev_for_src_ip_rcu(dev_net(ndev), dst_in);
 516                 if (IS_ERR(ndev))
 517                         return -ENODEV;
 518         }
 519
 520         return copy_src_l2_addr(dev_addr, dst_in, dst, ndev);
 521 }
 522
 523 static int set_addr_netns_by_gid_rcu(struct rdma_dev_addr *addr)
 524 {
 525         struct net_device *ndev;
 526
 527         ndev = rdma_read_gid_attr_ndev_rcu(addr->sgid_attr);
 528         if (IS_ERR(ndev))
 529                 return PTR_ERR(ndev);
 530
 531         /*
 532          * Since we are holding the rcu, reading net and ifindex
 533          * are safe without any additional reference; because
 534          * change_net_namespace() in net/core/dev.c does rcu sync
 535          * after it changes the state to IFF_DOWN and before
 536          * updating netdev fields {net, ifindex}.
 537          */
 538         addr->net = dev_net(ndev);
 539         addr->bound_dev_if = ndev->ifindex;
 540         return 0;
 541 }
 542
 543 static void rdma_addr_set_net_defaults(struct rdma_dev_addr *addr)
 544 {
 545         addr->net = &init_net;
 546         addr->bound_dev_if = 0;
 547 }
 548
 549 static int addr_resolve(struct sockaddr *src_in,
 550                         const struct sockaddr *dst_in,
 551                         struct rdma_dev_addr *addr,
 552                         bool resolve_neigh,
 553                         bool resolve_by_gid_attr,
 554                         u32 seq)
 555 {
 556         struct dst_entry *dst = NULL;
 557         unsigned int ndev_flags = 0;
 558         struct rtable *rt = NULL;
 559         int ret;
 560
 561         if (!addr->net) {
 562                 pr_warn_ratelimited("%s: missing namespace\n", __func__);
 563                 return -EINVAL;
 564         }
 565
 566         rcu_read_lock();
 567         if (resolve_by_gid_attr) {
 568                 if (!addr->sgid_attr) {
 569                         rcu_read_unlock();
 570                         pr_warn_ratelimited("%s: missing gid_attr\n", __func__);
 571                         return -EINVAL;
 572                 }
 573                 /*
 574                  * If the request is for a specific gid attribute of the
 575                  * rdma_dev_addr, derive net from the netdevice of the
 576                  * GID attribute.
 577                  */
 578                 ret = set_addr_netns_by_gid_rcu(addr);
 579                 if (ret) {
 580                         rcu_read_unlock();
 581                         return ret;
 582                 }
 583         }
 584         if (src_in->sa_family == AF_INET) {
 585                 ret = addr4_resolve(src_in, dst_in, addr, &rt);
 586                 dst = &rt->dst;
 587         } else {
 588                 ret = addr6_resolve(src_in, dst_in, addr, &dst);
 589         }
 590         if (ret) {
 591                 rcu_read_unlock();
 592                 goto done;
 593         }
 594         ret = rdma_set_src_addr_rcu(addr, &ndev_flags, dst_in, dst);
 595         rcu_read_unlock();
 596
 597         /*
 598          * Resolve neighbor destination address if requested and
 599          * only if src addr translation didn't fail.
 600          */
 601         if (!ret && resolve_neigh)
 602                 ret = addr_resolve_neigh(dst, dst_in, addr, ndev_flags, seq);
 603
 604         if (src_in->sa_family == AF_INET)
 605                 ip_rt_put(rt);
 606         else
 607                 dst_release(dst);
 608 done:
 609         /*
 610          * Clear the addr net to go back to its original state, only if it was
 611          * derived from GID attribute in this context.
 612          */
 613         if (resolve_by_gid_attr)
 614                 rdma_addr_set_net_defaults(addr);
 615         return ret;
 616 }
 617
 618 static void process_one_req(struct work_struct *_work)
 619 {
 620         struct addr_req *req;
 621         struct sockaddr *src_in, *dst_in;
 622
 623         req = container_of(_work, struct addr_req, work.work);
 624
 625         if (req->status == -ENODATA) {
 626                 src_in = (struct sockaddr *)&req->src_addr;
 627                 dst_in = (struct sockaddr *)&req->dst_addr;
 628                 req->status = addr_resolve(src_in, dst_in, req->addr,
 629                                            true, req->resolve_by_gid_attr,
 630                                            req->seq);
 631                 if (req->status && time_after_eq(jiffies, req->timeout)) {
 632                         req->status = -ETIMEDOUT;
 633                 } else if (req->status == -ENODATA) {
 634                         /* requeue the work for retrying again */
 635                         spin_lock_bh(&lock);
 636                         if (!list_empty(&req->list))
 637                                 set_timeout(req, req->timeout);
 638                         spin_unlock_bh(&lock);
 639                         return;
 640                 }
 641         }
 642
 643         req->callback(req->status, (struct sockaddr *)&req->src_addr,
 644                 req->addr, req->context);
 645         req->callback = NULL;
 646
 647         spin_lock_bh(&lock);
 648         if (!list_empty(&req->list)) {
 649                 /*
 650                  * Although the work will normally have been canceled by the
 651                  * workqueue, it can still be requeued as long as it is on the
 652                  * req_list.
 653                  */
 654                 cancel_delayed_work(&req->work);
 655                 list_del_init(&req->list);
 656                 kfree(req);
 657         }
 658         spin_unlock_bh(&lock);
 659 }
 660
 661 int rdma_resolve_ip(struct sockaddr *src_addr, const struct sockaddr *dst_addr,
 662                     struct rdma_dev_addr *addr, unsigned long timeout_ms,
 663                     void (*callback)(int status, struct sockaddr *src_addr,
 664                                      struct rdma_dev_addr *addr, void *context),
 665                     bool resolve_by_gid_attr, void *context)
 666 {
 667         struct sockaddr *src_in, *dst_in;
 668         struct addr_req *req;
 669         int ret = 0;
 670
 671         req = kzalloc(sizeof *req, GFP_KERNEL);
 672         if (!req)
 673                 return -ENOMEM;
 674
 675         src_in = (struct sockaddr *) &req->src_addr;
 676         dst_in = (struct sockaddr *) &req->dst_addr;
 677
 678         if (src_addr) {
 679                 if (src_addr->sa_family != dst_addr->sa_family) {
 680                         ret = -EINVAL;
 681                         goto err;
 682                 }
 683
 684                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 685         } else {
 686                 src_in->sa_family = dst_addr->sa_family;
 687         }
 688
 689         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 690         req->addr = addr;
 691         req->callback = callback;
 692         req->context = context;
 693         req->resolve_by_gid_attr = resolve_by_gid_attr;
 694         INIT_DELAYED_WORK(&req->work, process_one_req);
 695         req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 696
 697         req->status = addr_resolve(src_in, dst_in, addr, true,
 698                                    req->resolve_by_gid_attr, req->seq);
 699         switch (req->status) {
 700         case 0:
 701                 req->timeout = jiffies;
 702                 queue_req(req);
 703                 break;
 704         case -ENODATA:
 705                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 706                 queue_req(req);
 707                 break;
 708         default:
 709                 ret = req->status;
 710                 goto err;
 711         }
 712         return ret;
 713 err:
 714         kfree(req);
 715         return ret;
 716 }
 717 EXPORT_SYMBOL(rdma_resolve_ip);
 718
 719 int roce_resolve_route_from_path(struct sa_path_rec *rec,
 720                                  const struct ib_gid_attr *attr)
 721 {
 722         union {
 723                 struct sockaddr     _sockaddr;
 724                 struct sockaddr_in  _sockaddr_in;
 725                 struct sockaddr_in6 _sockaddr_in6;
 726         } sgid, dgid;
 727         struct rdma_dev_addr dev_addr = {};
 728         int ret;
 729
 730         if (rec->roce.route_resolved)
 731                 return 0;
 732
 733         rdma_gid2ip(&sgid._sockaddr, &rec->sgid);
 734         rdma_gid2ip(&dgid._sockaddr, &rec->dgid);
 735
 736         if (sgid._sockaddr.sa_family != dgid._sockaddr.sa_family)
 737                 return -EINVAL;
 738
 739         if (!attr || !attr->ndev)
 740                 return -EINVAL;
 741
 742         dev_addr.net = &init_net;
 743         dev_addr.sgid_attr = attr;
 744
 745         ret = addr_resolve(&sgid._sockaddr, &dgid._sockaddr,
 746                            &dev_addr, false, true, 0);
 747         if (ret)
 748                 return ret;
 749
 750         if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
 751              dev_addr.network == RDMA_NETWORK_IPV6) &&
 752             rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
 753                 return -EINVAL;
 754
 755         rec->roce.route_resolved = true;
 756         return 0;
 757 }
 758
 759 /**
 760  * rdma_addr_cancel - Cancel resolve ip request
 761  * @addr:       Pointer to address structure given previously
 762  *              during rdma_resolve_ip().
 763  * rdma_addr_cancel() is synchronous function which cancels any pending
 764  * request if there is any.
 765  */
 766 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 767 {
 768         struct addr_req *req, *temp_req;
 769         struct addr_req *found = NULL;
 770
 771         spin_lock_bh(&lock);
 772         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 773                 if (req->addr == addr) {
 774                         /*
 775                          * Removing from the list means we take ownership of
 776                          * the req
 777                          */
 778                         list_del_init(&req->list);
 779                         found = req;
 780                         break;
 781                 }
 782         }
 783         spin_unlock_bh(&lock);
 784
 785         if (!found)
 786                 return;
 787
 788         /*
 789          * sync canceling the work after removing it from the req_list
 790          * guarentees no work is running and none will be started.
 791          */
 792         cancel_delayed_work_sync(&found->work);
 793         kfree(found);
 794 }
 795 EXPORT_SYMBOL(rdma_addr_cancel);
 796
 797 struct resolve_cb_context {
 798         struct completion comp;
 799         int status;
 800 };
 801
 802 static void resolve_cb(int status, struct sockaddr *src_addr,
 803              struct rdma_dev_addr *addr, void *context)
 804 {
 805         ((struct resolve_cb_context *)context)->status = status;
 806         complete(&((struct resolve_cb_context *)context)->comp);
 807 }
 808
 809 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 810                                  const union ib_gid *dgid,
 811                                  u8 *dmac, const struct ib_gid_attr *sgid_attr,
 812                                  int *hoplimit)
 813 {
 814         struct rdma_dev_addr dev_addr;
 815         struct resolve_cb_context ctx;
 816         union {
 817                 struct sockaddr     _sockaddr;
 818                 struct sockaddr_in  _sockaddr_in;
 819                 struct sockaddr_in6 _sockaddr_in6;
 820         } sgid_addr, dgid_addr;
 821         int ret;
 822
 823         rdma_gid2ip(&sgid_addr._sockaddr, sgid);
 824         rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 825
 826         memset(&dev_addr, 0, sizeof(dev_addr));
 827         dev_addr.net = &init_net;
 828         dev_addr.sgid_attr = sgid_attr;
 829
 830         init_completion(&ctx.comp);
 831         ret = rdma_resolve_ip(&sgid_addr._sockaddr, &dgid_addr._sockaddr,
 832                               &dev_addr, 1000, resolve_cb, true, &ctx);
 833         if (ret)
 834                 return ret;
 835
 836         wait_for_completion(&ctx.comp);
 837
 838         ret = ctx.status;
 839         if (ret)
 840                 return ret;
 841
 842         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 843         *hoplimit = dev_addr.hoplimit;
 844         return 0;
 845 }
 846
 847 static int netevent_callback(struct notifier_block *self, unsigned long event,
 848         void *ctx)
 849 {
 850         struct addr_req *req;
 851
 852         if (event == NETEVENT_NEIGH_UPDATE) {
 853                 struct neighbour *neigh = ctx;
 854
 855                 if (neigh->nud_state & NUD_VALID) {
 856                         spin_lock_bh(&lock);
 857                         list_for_each_entry(req, &req_list, list)
 858                                 set_timeout(req, jiffies);
 859                         spin_unlock_bh(&lock);
 860                 }
 861         }
 862         return 0;
 863 }
 864
 865 static struct notifier_block nb = {
 866         .notifier_call = netevent_callback
 867 };
 868
 869 int addr_init(void)
 870 {
 871         addr_wq = alloc_ordered_workqueue("ib_addr", 0);
 872         if (!addr_wq)
 873                 return -ENOMEM;
 874
 875         register_netevent_notifier(&nb);
 876
 877         return 0;
 878 }
 879
 880 void addr_cleanup(void)
 881 {
 882         unregister_netevent_notifier(&nb);
 883         destroy_workqueue(addr_wq);
 884         WARN_ON(!list_empty(&req_list));
 885 }