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