sys/ofed/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/workqueue.h>
  39 #include <net/arp.h>
  40 #include <net/neighbour.h>
  41 #include <net/route.h>
  42 #include <net/netevent.h>
  43 #include <net/addrconf.h>
  44 #include <net/ip6_route.h>
  45 #include <rdma/ib_addr.h>
  46
  47 MODULE_AUTHOR("Sean Hefty");
  48 MODULE_DESCRIPTION("IB Address Translation");
  49 MODULE_LICENSE("Dual BSD/GPL");
  50
  51 struct addr_req {
  52         struct list_head list;
  53         struct sockaddr_storage src_addr;
  54         struct sockaddr_storage dst_addr;
  55         struct rdma_dev_addr *addr;
  56         struct rdma_addr_client *client;
  57         void *context;
  58         void (*callback)(int status, struct sockaddr *src_addr,
  59                          struct rdma_dev_addr *addr, void *context);
  60         unsigned long timeout;
  61         int status;
  62 };
  63
  64 static void process_req(struct work_struct *work);
  65
  66 static DEFINE_MUTEX(lock);
  67 static LIST_HEAD(req_list);
  68 static struct delayed_work work;
  69 static struct workqueue_struct *addr_wq;
  70
  71 void rdma_addr_register_client(struct rdma_addr_client *client)
  72 {
  73         atomic_set(&client->refcount, 1);
  74         init_completion(&client->comp);
  75 }
  76 EXPORT_SYMBOL(rdma_addr_register_client);
  77
  78 static inline void put_client(struct rdma_addr_client *client)
  79 {
  80         if (atomic_dec_and_test(&client->refcount))
  81                 complete(&client->comp);
  82 }
  83
  84 void rdma_addr_unregister_client(struct rdma_addr_client *client)
  85 {
  86         put_client(client);
  87         wait_for_completion(&client->comp);
  88 }
  89 EXPORT_SYMBOL(rdma_addr_unregister_client);
  90
  91 #ifdef __linux__
  92 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
  93                      const unsigned char *dst_dev_addr)
  94 {
  95         dev_addr->dev_type = dev->type;
  96         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
  97         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
  98         if (dst_dev_addr)
  99                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 100         dev_addr->bound_dev_if = dev->ifindex;
 101         return 0;
 102 }
 103 #else
 104 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct ifnet *dev,
 105                      const unsigned char *dst_dev_addr)
 106 {
 107         if (dev->if_type == IFT_INFINIBAND)
 108                 dev_addr->dev_type = ARPHRD_INFINIBAND;
 109         else if (dev->if_type == IFT_ETHER)
 110                 dev_addr->dev_type = ARPHRD_ETHER;
 111         else
 112                 dev_addr->dev_type = 0;
 113         memcpy(dev_addr->src_dev_addr, IF_LLADDR(dev), dev->if_addrlen);
 114         memcpy(dev_addr->broadcast, __DECONST(char *, dev->if_broadcastaddr),
 115             dev->if_addrlen);
 116         if (dst_dev_addr)
 117                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, dev->if_addrlen);
 118         dev_addr->bound_dev_if = dev->if_index;
 119         return 0;
 120 }
 121 #endif
 122 EXPORT_SYMBOL(rdma_copy_addr);
 123
 124 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 125 {
 126         struct net_device *dev;
 127         int ret = -EADDRNOTAVAIL;
 128
 129         if (dev_addr->bound_dev_if) {
 130                 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
 131                 if (!dev)
 132                         return -ENODEV;
 133                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 134                 dev_put(dev);
 135                 return ret;
 136         }
 137
 138         switch (addr->sa_family) {
 139 #ifdef INET
 140         case AF_INET:
 141                 dev = ip_dev_find(NULL,
 142                         ((struct sockaddr_in *) addr)->sin_addr.s_addr);
 143
 144                 if (!dev)
 145                         return ret;
 146
 147                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 148                 dev_put(dev);
 149                 break;
 150 #endif
 151
 152 #if defined(INET6)
 153         case AF_INET6:
 154 #ifdef __linux__
 155                 read_lock(&dev_base_lock);
 156                 for_each_netdev(&init_net, dev) {
 157                         if (ipv6_chk_addr(&init_net,
 158                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
 159                                           dev, 1)) {
 160                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 161                                 break;
 162                         }
 163                 }
 164                 read_unlock(&dev_base_lock);
 165 #else
 166                 {
 167                         struct sockaddr_in6 *sin6;
 168                         struct ifaddr *ifa;
 169                         in_port_t port;
 170
 171                         sin6 = (struct sockaddr_in6 *)addr;
 172                         port = sin6->sin6_port;
 173                         sin6->sin6_port = 0;
 174                         ifa = ifa_ifwithaddr(addr);
 175                         sin6->sin6_port = port;
 176                         if (ifa == NULL) {
 177                                 ret = -ENODEV;
 178                                 break;
 179                         }
 180                         ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
 181                         ifa_free(ifa);
 182                         break;
 183                 }
 184 #endif
 185                 break;
 186 #endif
 187         }
 188         return ret;
 189 }
 190 EXPORT_SYMBOL(rdma_translate_ip);
 191
 192 static void set_timeout(unsigned long time)
 193 {
 194         unsigned long delay;
 195
 196         cancel_delayed_work(&work);
 197
 198         delay = time - jiffies;
 199         if ((long)delay <= 0)
 200                 delay = 1;
 201
 202         queue_delayed_work(addr_wq, &work, delay);
 203 }
 204
 205 static void queue_req(struct addr_req *req)
 206 {
 207         struct addr_req *temp_req;
 208
 209         mutex_lock(&lock);
 210         list_for_each_entry_reverse(temp_req, &req_list, list) {
 211                 if (time_after_eq(req->timeout, temp_req->timeout))
 212                         break;
 213         }
 214
 215         list_add(&req->list, &temp_req->list);
 216
 217         if (req_list.next == &req->list)
 218                 set_timeout(req->timeout);
 219         mutex_unlock(&lock);
 220 }
 221
 222 #ifdef __linux__
 223 static int addr4_resolve(struct sockaddr_in *src_in,
 224                          struct sockaddr_in *dst_in,
 225                          struct rdma_dev_addr *addr)
 226 {
 227         __be32 src_ip = src_in->sin_addr.s_addr;
 228         __be32 dst_ip = dst_in->sin_addr.s_addr;
 229         struct flowi fl;
 230         struct rtable *rt;
 231         struct neighbour *neigh;
 232         int ret;
 233
 234         memset(&fl, 0, sizeof fl);
 235         fl.nl_u.ip4_u.daddr = dst_ip;
 236         fl.nl_u.ip4_u.saddr = src_ip;
 237         fl.oif = addr->bound_dev_if;
 238
 239         ret = ip_route_output_key(&init_net, &rt, &fl);
 240         if (ret)
 241                 goto out;
 242
 243         src_in->sin_family = AF_INET;
 244         src_in->sin_addr.s_addr = rt->rt_src;
 245
 246         if (rt->idev->dev->flags & IFF_LOOPBACK) {
 247                 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
 248                 if (!ret)
 249                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 250                 goto put;
 251         }
 252
 253         /* If the device does ARP internally, return 'done' */
 254         if (rt->idev->dev->flags & IFF_NOARP) {
 255                 rdma_copy_addr(addr, rt->idev->dev, NULL);
 256                 goto put;
 257         }
 258
 259         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
 260         if (!neigh || !(neigh->nud_state & NUD_VALID)) {
 261                 neigh_event_send(rt->u.dst.neighbour, NULL);
 262                 ret = -ENODATA;
 263                 if (neigh)
 264                         goto release;
 265                 goto put;
 266         }
 267
 268         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
 269 release:
 270         neigh_release(neigh);
 271 put:
 272         ip_rt_put(rt);
 273 out:
 274         return ret;
 275 }
 276
 277 #if defined(INET6)
 278 static int addr6_resolve(struct sockaddr_in6 *src_in,
 279                          struct sockaddr_in6 *dst_in,
 280                          struct rdma_dev_addr *addr)
 281 {
 282         struct flowi fl;
 283         struct neighbour *neigh;
 284         struct dst_entry *dst;
 285         int ret;
 286
 287         memset(&fl, 0, sizeof fl);
 288         ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
 289         ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
 290         fl.oif = addr->bound_dev_if;
 291
 292         dst = ip6_route_output(&init_net, NULL, &fl);
 293         if ((ret = dst->error))
 294                 goto put;
 295
 296         if (ipv6_addr_any(&fl.fl6_src)) {
 297                 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
 298                                          &fl.fl6_dst, 0, &fl.fl6_src);
 299                 if (ret)
 300                         goto put;
 301
 302                 src_in->sin6_family = AF_INET6;
 303                 ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
 304         }
 305
 306         if (dst->dev->flags & IFF_LOOPBACK) {
 307                 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
 308                 if (!ret)
 309                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 310                 goto put;
 311         }
 312
 313         /* If the device does ARP internally, return 'done' */
 314         if (dst->dev->flags & IFF_NOARP) {
 315                 ret = rdma_copy_addr(addr, dst->dev, NULL);
 316                 goto put;
 317         }
 318
 319         neigh = dst->neighbour;
 320         if (!neigh || !(neigh->nud_state & NUD_VALID)) {
 321                 neigh_event_send(dst->neighbour, NULL);
 322                 ret = -ENODATA;
 323                 goto put;
 324         }
 325
 326         ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
 327 put:
 328         dst_release(dst);
 329         return ret;
 330 }
 331 #else
 332 static int addr6_resolve(struct sockaddr_in6 *src_in,
 333                          struct sockaddr_in6 *dst_in,
 334                          struct rdma_dev_addr *addr)
 335 {
 336         return -EADDRNOTAVAIL;
 337 }
 338 #endif
 339
 340 #else
 341 #include <netinet/if_ether.h>
 342
 343 static int addr_resolve(struct sockaddr *src_in,
 344                         struct sockaddr *dst_in,
 345                         struct rdma_dev_addr *addr)
 346 {
 347         struct sockaddr_in *sin;
 348         struct sockaddr_in6 *sin6;
 349         struct ifaddr *ifa;
 350         struct ifnet *ifp;
 351 #if defined(INET) || defined(INET6)
 352         struct llentry *lle;
 353 #endif
 354         struct rtentry *rte;
 355         in_port_t port;
 356         u_char edst[MAX_ADDR_LEN];
 357         int multi;
 358         int bcast;
 359         int error = 0;
 360
 361         /*
 362          * Determine whether the address is unicast, multicast, or broadcast
 363          * and whether the source interface is valid.
 364          */
 365         multi = 0;
 366         bcast = 0;
 367         sin = NULL;
 368         sin6 = NULL;
 369         ifp = NULL;
 370         rte = NULL;
 371         switch (dst_in->sa_family) {
 372 #ifdef INET
 373         case AF_INET:
 374                 sin = (struct sockaddr_in *)dst_in;
 375                 if (sin->sin_addr.s_addr == INADDR_BROADCAST)
 376                         bcast = 1;
 377                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
 378                         multi = 1;
 379                 sin = (struct sockaddr_in *)src_in;
 380                 if (sin->sin_addr.s_addr != INADDR_ANY) {
 381                         /*
 382                          * Address comparison fails if the port is set
 383                          * cache it here to be restored later.
 384                          */
 385                         port = sin->sin_port;
 386                         sin->sin_port = 0;
 387                         memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
 388                 } else
 389                         src_in = NULL;
 390                 break;
 391 #endif
 392 #ifdef INET6
 393         case AF_INET6:
 394                 sin6 = (struct sockaddr_in6 *)dst_in;
 395                 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 396                         multi = 1;
 397                 sin6 = (struct sockaddr_in6 *)src_in;
 398                 if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
 399                         port = sin6->sin6_port;
 400                         sin6->sin6_port = 0;
 401                 } else
 402                         src_in = NULL;
 403                 break;
 404 #endif
 405         default:
 406                 return -EINVAL;
 407         }
 408         /*
 409          * If we have a source address to use look it up first and verify
 410          * that it is a local interface.
 411          */
 412         if (src_in) {
 413                 ifa = ifa_ifwithaddr(src_in);
 414                 if (sin)
 415                         sin->sin_port = port;
 416                 if (sin6)
 417                         sin6->sin6_port = port;
 418                 if (ifa == NULL)
 419                         return -ENETUNREACH;
 420                 ifp = ifa->ifa_ifp;
 421                 ifa_free(ifa);
 422                 if (bcast || multi)
 423                         goto mcast;
 424         }
 425         /*
 426          * Make sure the route exists and has a valid link.
 427          */
 428         rte = rtalloc1(dst_in, 1, 0);
 429         if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
 430                 if (rte)
 431                         RTFREE_LOCKED(rte);
 432                 return -EHOSTUNREACH;
 433         }
 434         /*
 435          * If it's not multicast or broadcast and the route doesn't match the
 436          * requested interface return unreachable.  Otherwise fetch the
 437          * correct interface pointer and unlock the route.
 438          */
 439         if (multi || bcast) {
 440                 if (ifp == NULL)
 441                         ifp = rte->rt_ifp;
 442                 RTFREE_LOCKED(rte);
 443         } else if (ifp && ifp != rte->rt_ifp) {
 444                 RTFREE_LOCKED(rte);
 445                 return -ENETUNREACH;
 446         } else {
 447                 if (ifp == NULL)
 448                         ifp = rte->rt_ifp;
 449                 RT_UNLOCK(rte);
 450         }
 451 mcast:
 452         if (bcast)
 453                 return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
 454         if (multi) {
 455                 struct sockaddr *llsa;
 456
 457                 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
 458                 if (error)
 459                         return -error;
 460                 error = rdma_copy_addr(addr, ifp,
 461                     LLADDR((struct sockaddr_dl *)llsa));
 462                 free(llsa, M_IFMADDR);
 463                 return error;
 464         }
 465         /*
 466          * Resolve the link local address.
 467          */
 468         switch (dst_in->sa_family) {
 469 #ifdef INET
 470         case AF_INET:
 471                 error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
 472                 break;
 473 #endif
 474 #ifdef INET6
 475         case AF_INET6:
 476                 error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
 477                 break;
 478 #endif
 479         default:
 480                 /* XXX: Shouldn't happen. */
 481                 error = -EINVAL;
 482         }
 483         RTFREE(rte);
 484         if (error == 0)
 485                 return rdma_copy_addr(addr, ifp, edst);
 486         if (error == EWOULDBLOCK)
 487                 return -ENODATA;
 488         return -error;
 489 }
 490
 491 #endif
 492
 493 static void process_req(struct work_struct *work)
 494 {
 495         struct addr_req *req, *temp_req;
 496         struct sockaddr *src_in, *dst_in;
 497         struct list_head done_list;
 498
 499         INIT_LIST_HEAD(&done_list);
 500
 501         mutex_lock(&lock);
 502         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 503                 if (req->status == -ENODATA) {
 504                         src_in = (struct sockaddr *) &req->src_addr;
 505                         dst_in = (struct sockaddr *) &req->dst_addr;
 506                         req->status = addr_resolve(src_in, dst_in, req->addr);
 507                         if (req->status && time_after_eq(jiffies, req->timeout))
 508                                 req->status = -ETIMEDOUT;
 509                         else if (req->status == -ENODATA)
 510                                 continue;
 511                 }
 512                 list_move_tail(&req->list, &done_list);
 513         }
 514
 515         if (!list_empty(&req_list)) {
 516                 req = list_entry(req_list.next, struct addr_req, list);
 517                 set_timeout(req->timeout);
 518         }
 519         mutex_unlock(&lock);
 520
 521         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 522                 list_del(&req->list);
 523                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 524                         req->addr, req->context);
 525                 put_client(req->client);
 526                 kfree(req);
 527         }
 528 }
 529
 530 int rdma_resolve_ip(struct rdma_addr_client *client,
 531                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 532                     struct rdma_dev_addr *addr, int timeout_ms,
 533                     void (*callback)(int status, struct sockaddr *src_addr,
 534                                      struct rdma_dev_addr *addr, void *context),
 535                     void *context)
 536 {
 537         struct sockaddr *src_in, *dst_in;
 538         struct addr_req *req;
 539         int ret = 0;
 540
 541         req = kzalloc(sizeof *req, GFP_KERNEL);
 542         if (!req)
 543                 return -ENOMEM;
 544
 545         src_in = (struct sockaddr *) &req->src_addr;
 546         dst_in = (struct sockaddr *) &req->dst_addr;
 547
 548         if (src_addr) {
 549                 if (src_addr->sa_family != dst_addr->sa_family) {
 550                         ret = -EINVAL;
 551                         goto err;
 552                 }
 553
 554                 memcpy(src_in, src_addr, ip_addr_size(src_addr));
 555         } else {
 556                 src_in->sa_family = dst_addr->sa_family;
 557         }
 558
 559         memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
 560         req->addr = addr;
 561         req->callback = callback;
 562         req->context = context;
 563         req->client = client;
 564         atomic_inc(&client->refcount);
 565
 566         req->status = addr_resolve(src_in, dst_in, addr);
 567         switch (req->status) {
 568         case 0:
 569                 req->timeout = jiffies;
 570                 queue_req(req);
 571                 break;
 572         case -ENODATA:
 573                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 574                 queue_req(req);
 575                 break;
 576         default:
 577                 ret = req->status;
 578                 atomic_dec(&client->refcount);
 579                 goto err;
 580         }
 581         return ret;
 582 err:
 583         kfree(req);
 584         return ret;
 585 }
 586 EXPORT_SYMBOL(rdma_resolve_ip);
 587
 588 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 589 {
 590         struct addr_req *req, *temp_req;
 591
 592         mutex_lock(&lock);
 593         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 594                 if (req->addr == addr) {
 595                         req->status = -ECANCELED;
 596                         req->timeout = jiffies;
 597                         list_move(&req->list, &req_list);
 598                         set_timeout(req->timeout);
 599                         break;
 600                 }
 601         }
 602         mutex_unlock(&lock);
 603 }
 604 EXPORT_SYMBOL(rdma_addr_cancel);
 605
 606 static int netevent_callback(struct notifier_block *self, unsigned long event,
 607         void *ctx)
 608 {
 609         if (event == NETEVENT_NEIGH_UPDATE) {
 610 #ifdef __linux__
 611                 struct neighbour *neigh = ctx;
 612
 613                 if (neigh->nud_state & NUD_VALID) {
 614                         set_timeout(jiffies);
 615                 }
 616 #else
 617                 set_timeout(jiffies);
 618 #endif
 619         }
 620         return 0;
 621 }
 622
 623 static struct notifier_block nb = {
 624         .notifier_call = netevent_callback
 625 };
 626
 627 static int addr_init(void)
 628 {
 629         INIT_DELAYED_WORK(&work, process_req);
 630         addr_wq = create_singlethread_workqueue("ib_addr");
 631         if (!addr_wq)
 632                 return -ENOMEM;
 633
 634         register_netevent_notifier(&nb);
 635         return 0;
 636 }
 637
 638 static void addr_cleanup(void)
 639 {
 640         unregister_netevent_notifier(&nb);
 641         destroy_workqueue(addr_wq);
 642 }
 643
 644 module_init(addr_init);
 645 module_exit(addr_cleanup);