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 <linux/if_arp.h>
  40 #include <net/arp.h>
  41 #include <net/neighbour.h>
  42 #include <net/route.h>
  43 #include <net/netevent.h>
  44 #include <net/addrconf.h>
  45 #include <net/ip6_route.h>
  46 #include <rdma/ib_addr.h>
  47
  48 MODULE_AUTHOR("Sean Hefty");
  49 MODULE_DESCRIPTION("IB Address Translation");
  50 MODULE_LICENSE("Dual BSD/GPL");
  51
  52 struct addr_req {
  53         struct list_head list;
  54         struct sockaddr_storage src_addr;
  55         struct sockaddr_storage dst_addr;
  56         struct rdma_dev_addr *addr;
  57         struct rdma_addr_client *client;
  58         void *context;
  59         void (*callback)(int status, struct sockaddr *src_addr,
  60                          struct rdma_dev_addr *addr, void *context);
  61         unsigned long timeout;
  62         int status;
  63 };
  64
  65 static void process_req(struct work_struct *work);
  66
  67 static DEFINE_MUTEX(lock);
  68 static LIST_HEAD(req_list);
  69 static DECLARE_DELAYED_WORK(work, process_req);
  70 static struct workqueue_struct *addr_wq;
  71
  72 void rdma_addr_register_client(struct rdma_addr_client *client)
  73 {
  74         atomic_set(&client->refcount, 1);
  75         init_completion(&client->comp);
  76 }
  77 EXPORT_SYMBOL(rdma_addr_register_client);
  78
  79 static inline void put_client(struct rdma_addr_client *client)
  80 {
  81         if (atomic_dec_and_test(&client->refcount))
  82                 complete(&client->comp);
  83 }
  84
  85 void rdma_addr_unregister_client(struct rdma_addr_client *client)
  86 {
  87         put_client(client);
  88         wait_for_completion(&client->comp);
  89 }
  90 EXPORT_SYMBOL(rdma_addr_unregister_client);
  91
  92 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
  93                      const unsigned char *dst_dev_addr)
  94 {
  95         switch (dev->type) {
  96         case ARPHRD_INFINIBAND:
  97                 dev_addr->dev_type = RDMA_NODE_IB_CA;
  98                 break;
  99         case ARPHRD_ETHER:
 100                 dev_addr->dev_type = RDMA_NODE_RNIC;
 101                 break;
 102         default:
 103                 return -EADDRNOTAVAIL;
 104         }
 105
 106         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 107         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
 108         if (dst_dev_addr)
 109                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 110         dev_addr->src_dev = dev;
 111         return 0;
 112 }
 113 EXPORT_SYMBOL(rdma_copy_addr);
 114
 115 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 116 {
 117         struct net_device *dev;
 118         int ret = -EADDRNOTAVAIL;
 119
 120         switch (addr->sa_family) {
 121         case AF_INET:
 122                 dev = ip_dev_find(&init_net,
 123                         ((struct sockaddr_in *) addr)->sin_addr.s_addr);
 124
 125                 if (!dev)
 126                         return ret;
 127
 128                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 129                 dev_put(dev);
 130                 break;
 131
 132 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 133         case AF_INET6:
 134                 read_lock(&dev_base_lock);
 135                 for_each_netdev(&init_net, dev) {
 136                         if (ipv6_chk_addr(&init_net,
 137                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
 138                                           dev, 1)) {
 139                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 140                                 break;
 141                         }
 142                 }
 143                 read_unlock(&dev_base_lock);
 144                 break;
 145 #endif
 146         }
 147         return ret;
 148 }
 149 EXPORT_SYMBOL(rdma_translate_ip);
 150
 151 static void set_timeout(unsigned long time)
 152 {
 153         unsigned long delay;
 154
 155         cancel_delayed_work(&work);
 156
 157         delay = time - jiffies;
 158         if ((long)delay <= 0)
 159                 delay = 1;
 160
 161         queue_delayed_work(addr_wq, &work, delay);
 162 }
 163
 164 static void queue_req(struct addr_req *req)
 165 {
 166         struct addr_req *temp_req;
 167
 168         mutex_lock(&lock);
 169         list_for_each_entry_reverse(temp_req, &req_list, list) {
 170                 if (time_after_eq(req->timeout, temp_req->timeout))
 171                         break;
 172         }
 173
 174         list_add(&req->list, &temp_req->list);
 175
 176         if (req_list.next == &req->list)
 177                 set_timeout(req->timeout);
 178         mutex_unlock(&lock);
 179 }
 180
 181 static void addr_send_arp(struct sockaddr *dst_in)
 182 {
 183         struct rtable *rt;
 184         struct flowi fl;
 185
 186         memset(&fl, 0, sizeof fl);
 187
 188         switch (dst_in->sa_family) {
 189         case AF_INET:
 190                 fl.nl_u.ip4_u.daddr =
 191                         ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
 192
 193                 if (ip_route_output_key(&init_net, &rt, &fl))
 194                         return;
 195
 196                 neigh_event_send(rt->u.dst.neighbour, NULL);
 197                 ip_rt_put(rt);
 198                 break;
 199
 200 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 201         case AF_INET6:
 202         {
 203                 struct dst_entry *dst;
 204
 205                 fl.nl_u.ip6_u.daddr =
 206                         ((struct sockaddr_in6 *) dst_in)->sin6_addr;
 207
 208                 dst = ip6_route_output(&init_net, NULL, &fl);
 209                 if (!dst)
 210                         return;
 211
 212                 neigh_event_send(dst->neighbour, NULL);
 213                 dst_release(dst);
 214                 break;
 215         }
 216 #endif
 217         }
 218 }
 219
 220 static int addr4_resolve_remote(struct sockaddr_in *src_in,
 221                                struct sockaddr_in *dst_in,
 222                                struct rdma_dev_addr *addr)
 223 {
 224         __be32 src_ip = src_in->sin_addr.s_addr;
 225         __be32 dst_ip = dst_in->sin_addr.s_addr;
 226         struct flowi fl;
 227         struct rtable *rt;
 228         struct neighbour *neigh;
 229         int ret;
 230
 231         memset(&fl, 0, sizeof fl);
 232         fl.nl_u.ip4_u.daddr = dst_ip;
 233         fl.nl_u.ip4_u.saddr = src_ip;
 234         ret = ip_route_output_key(&init_net, &rt, &fl);
 235         if (ret)
 236                 goto out;
 237
 238         /* If the device does ARP internally, return 'done' */
 239         if (rt->idev->dev->flags & IFF_NOARP) {
 240                 rdma_copy_addr(addr, rt->idev->dev, NULL);
 241                 goto put;
 242         }
 243
 244         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
 245         if (!neigh) {
 246                 ret = -ENODATA;
 247                 goto put;
 248         }
 249
 250         if (!(neigh->nud_state & NUD_VALID)) {
 251                 ret = -ENODATA;
 252                 goto release;
 253         }
 254
 255         if (!src_ip) {
 256                 src_in->sin_family = dst_in->sin_family;
 257                 src_in->sin_addr.s_addr = rt->rt_src;
 258         }
 259
 260         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
 261 release:
 262         neigh_release(neigh);
 263 put:
 264         ip_rt_put(rt);
 265 out:
 266         return ret;
 267 }
 268
 269 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 270 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
 271                                struct sockaddr_in6 *dst_in,
 272                                struct rdma_dev_addr *addr)
 273 {
 274         struct flowi fl;
 275         struct neighbour *neigh;
 276         struct dst_entry *dst;
 277         int ret = -ENODATA;
 278
 279         memset(&fl, 0, sizeof fl);
 280         fl.nl_u.ip6_u.daddr = dst_in->sin6_addr;
 281         fl.nl_u.ip6_u.saddr = src_in->sin6_addr;
 282
 283         dst = ip6_route_output(&init_net, NULL, &fl);
 284         if (!dst)
 285                 return ret;
 286
 287         if (dst->dev->flags & IFF_NOARP) {
 288                 ret = rdma_copy_addr(addr, dst->dev, NULL);
 289         } else {
 290                 neigh = dst->neighbour;
 291                 if (neigh && (neigh->nud_state & NUD_VALID))
 292                         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
 293         }
 294
 295         dst_release(dst);
 296         return ret;
 297 }
 298 #else
 299 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
 300                                struct sockaddr_in6 *dst_in,
 301                                struct rdma_dev_addr *addr)
 302 {
 303         return -EADDRNOTAVAIL;
 304 }
 305 #endif
 306
 307 static int addr_resolve_remote(struct sockaddr *src_in,
 308                                 struct sockaddr *dst_in,
 309                                 struct rdma_dev_addr *addr)
 310 {
 311         if (src_in->sa_family == AF_INET) {
 312                 return addr4_resolve_remote((struct sockaddr_in *) src_in,
 313                         (struct sockaddr_in *) dst_in, addr);
 314         } else
 315                 return addr6_resolve_remote((struct sockaddr_in6 *) src_in,
 316                         (struct sockaddr_in6 *) dst_in, addr);
 317 }
 318
 319 static void process_req(struct work_struct *work)
 320 {
 321         struct addr_req *req, *temp_req;
 322         struct sockaddr *src_in, *dst_in;
 323         struct list_head done_list;
 324
 325         INIT_LIST_HEAD(&done_list);
 326
 327         mutex_lock(&lock);
 328         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 329                 if (req->status == -ENODATA) {
 330                         src_in = (struct sockaddr *) &req->src_addr;
 331                         dst_in = (struct sockaddr *) &req->dst_addr;
 332                         req->status = addr_resolve_remote(src_in, dst_in,
 333                                                           req->addr);
 334                         if (req->status && time_after_eq(jiffies, req->timeout))
 335                                 req->status = -ETIMEDOUT;
 336                         else if (req->status == -ENODATA)
 337                                 continue;
 338                 }
 339                 list_move_tail(&req->list, &done_list);
 340         }
 341
 342         if (!list_empty(&req_list)) {
 343                 req = list_entry(req_list.next, struct addr_req, list);
 344                 set_timeout(req->timeout);
 345         }
 346         mutex_unlock(&lock);
 347
 348         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 349                 list_del(&req->list);
 350                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 351                         req->addr, req->context);
 352                 put_client(req->client);
 353                 kfree(req);
 354         }
 355 }
 356
 357 static int addr_resolve_local(struct sockaddr *src_in,
 358                               struct sockaddr *dst_in,
 359                               struct rdma_dev_addr *addr)
 360 {
 361         struct net_device *dev;
 362         int ret;
 363
 364         switch (dst_in->sa_family) {
 365         case AF_INET:
 366         {
 367                 __be32 src_ip = ((struct sockaddr_in *) src_in)->sin_addr.s_addr;
 368                 __be32 dst_ip = ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
 369
 370                 dev = ip_dev_find(&init_net, dst_ip);
 371                 if (!dev)
 372                         return -EADDRNOTAVAIL;
 373
 374                 if (ipv4_is_zeronet(src_ip)) {
 375                         src_in->sa_family = dst_in->sa_family;
 376                         ((struct sockaddr_in *) src_in)->sin_addr.s_addr = dst_ip;
 377                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
 378                 } else if (ipv4_is_loopback(src_ip)) {
 379                         ret = rdma_translate_ip(dst_in, addr);
 380                         if (!ret)
 381                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 382                 } else {
 383                         ret = rdma_translate_ip(src_in, addr);
 384                         if (!ret)
 385                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 386                 }
 387                 dev_put(dev);
 388                 break;
 389         }
 390
 391 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 392         case AF_INET6:
 393         {
 394                 struct in6_addr *a;
 395
 396                 read_lock(&dev_base_lock);
 397                 for_each_netdev(&init_net, dev)
 398                         if (ipv6_chk_addr(&init_net,
 399                                           &((struct sockaddr_in6 *) dst_in)->sin6_addr,
 400                                           dev, 1))
 401                                 break;
 402
 403                 if (!dev) {
 404                         read_unlock(&dev_base_lock);
 405                         return -EADDRNOTAVAIL;
 406                 }
 407
 408                 a = &((struct sockaddr_in6 *) src_in)->sin6_addr;
 409
 410                 if (ipv6_addr_any(a)) {
 411                         src_in->sa_family = dst_in->sa_family;
 412                         ((struct sockaddr_in6 *) src_in)->sin6_addr =
 413                                 ((struct sockaddr_in6 *) dst_in)->sin6_addr;
 414                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
 415                 } else if (ipv6_addr_loopback(a)) {
 416                         ret = rdma_translate_ip(dst_in, addr);
 417                         if (!ret)
 418                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 419                 } else  {
 420                         ret = rdma_translate_ip(src_in, addr);
 421                         if (!ret)
 422                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 423                 }
 424                 read_unlock(&dev_base_lock);
 425                 break;
 426         }
 427 #endif
 428
 429         default:
 430                 ret = -EADDRNOTAVAIL;
 431                 break;
 432         }
 433
 434         return ret;
 435 }
 436
 437 int rdma_resolve_ip(struct rdma_addr_client *client,
 438                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 439                     struct rdma_dev_addr *addr, int timeout_ms,
 440                     void (*callback)(int status, struct sockaddr *src_addr,
 441                                      struct rdma_dev_addr *addr, void *context),
 442                     void *context)
 443 {
 444         struct sockaddr *src_in, *dst_in;
 445         struct addr_req *req;
 446         int ret = 0;
 447
 448         req = kzalloc(sizeof *req, GFP_KERNEL);
 449         if (!req)
 450                 return -ENOMEM;
 451
 452         if (src_addr)
 453                 memcpy(&req->src_addr, src_addr, ip_addr_size(src_addr));
 454         memcpy(&req->dst_addr, dst_addr, ip_addr_size(dst_addr));
 455         req->addr = addr;
 456         req->callback = callback;
 457         req->context = context;
 458         req->client = client;
 459         atomic_inc(&client->refcount);
 460
 461         src_in = (struct sockaddr *) &req->src_addr;
 462         dst_in = (struct sockaddr *) &req->dst_addr;
 463
 464         req->status = addr_resolve_local(src_in, dst_in, addr);
 465         if (req->status == -EADDRNOTAVAIL)
 466                 req->status = addr_resolve_remote(src_in, dst_in, addr);
 467
 468         switch (req->status) {
 469         case 0:
 470                 req->timeout = jiffies;
 471                 queue_req(req);
 472                 break;
 473         case -ENODATA:
 474                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 475                 queue_req(req);
 476                 addr_send_arp(dst_in);
 477                 break;
 478         default:
 479                 ret = req->status;
 480                 atomic_dec(&client->refcount);
 481                 kfree(req);
 482                 break;
 483         }
 484         return ret;
 485 }
 486 EXPORT_SYMBOL(rdma_resolve_ip);
 487
 488 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 489 {
 490         struct addr_req *req, *temp_req;
 491
 492         mutex_lock(&lock);
 493         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 494                 if (req->addr == addr) {
 495                         req->status = -ECANCELED;
 496                         req->timeout = jiffies;
 497                         list_move(&req->list, &req_list);
 498                         set_timeout(req->timeout);
 499                         break;
 500                 }
 501         }
 502         mutex_unlock(&lock);
 503 }
 504 EXPORT_SYMBOL(rdma_addr_cancel);
 505
 506 static int netevent_callback(struct notifier_block *self, unsigned long event,
 507         void *ctx)
 508 {
 509         if (event == NETEVENT_NEIGH_UPDATE) {
 510                 struct neighbour *neigh = ctx;
 511
 512                 if (neigh->nud_state & NUD_VALID) {
 513                         set_timeout(jiffies);
 514                 }
 515         }
 516         return 0;
 517 }
 518
 519 static struct notifier_block nb = {
 520         .notifier_call = netevent_callback
 521 };
 522
 523 static int __init addr_init(void)
 524 {
 525         addr_wq = create_singlethread_workqueue("ib_addr");
 526         if (!addr_wq)
 527                 return -ENOMEM;
 528
 529         register_netevent_notifier(&nb);
 530         return 0;
 531 }
 532
 533 static void __exit addr_cleanup(void)
 534 {
 535         unregister_netevent_notifier(&nb);
 536         destroy_workqueue(addr_wq);
 537 }
 538
 539 module_init(addr_init);
 540 module_exit(addr_cleanup);