git.oblomov.eu Git - linux-2.6/blob - 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         case AF_INET6:
 132                 for_each_netdev(&init_net, dev) {
 133                         if (ipv6_chk_addr(&init_net,
 134                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
 135                                           dev, 1)) {
 136                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 137                                 break;
 138                         }
 139                 }
 140                 break;
 141         default:
 142                 break;
 143         }
 144         return ret;
 145 }
 146 EXPORT_SYMBOL(rdma_translate_ip);
 147
 148 static void set_timeout(unsigned long time)
 149 {
 150         unsigned long delay;
 151
 152         cancel_delayed_work(&work);
 153
 154         delay = time - jiffies;
 155         if ((long)delay <= 0)
 156                 delay = 1;
 157
 158         queue_delayed_work(addr_wq, &work, delay);
 159 }
 160
 161 static void queue_req(struct addr_req *req)
 162 {
 163         struct addr_req *temp_req;
 164
 165         mutex_lock(&lock);
 166         list_for_each_entry_reverse(temp_req, &req_list, list) {
 167                 if (time_after_eq(req->timeout, temp_req->timeout))
 168                         break;
 169         }
 170
 171         list_add(&req->list, &temp_req->list);
 172
 173         if (req_list.next == &req->list)
 174                 set_timeout(req->timeout);
 175         mutex_unlock(&lock);
 176 }
 177
 178 static void addr_send_arp(struct sockaddr *dst_in)
 179 {
 180         struct rtable *rt;
 181         struct flowi fl;
 182         struct dst_entry *dst;
 183
 184         memset(&fl, 0, sizeof fl);
 185         if (dst_in->sa_family == AF_INET)  {
 186                 fl.nl_u.ip4_u.daddr =
 187                         ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
 188
 189                 if (ip_route_output_key(&init_net, &rt, &fl))
 190                         return;
 191
 192                 neigh_event_send(rt->u.dst.neighbour, NULL);
 193                 ip_rt_put(rt);
 194
 195         } else {
 196                 fl.nl_u.ip6_u.daddr =
 197                         ((struct sockaddr_in6 *) dst_in)->sin6_addr;
 198
 199                 dst = ip6_route_output(&init_net, NULL, &fl);
 200                 if (!dst)
 201                         return;
 202
 203                 neigh_event_send(dst->neighbour, NULL);
 204                 dst_release(dst);
 205         }
 206 }
 207
 208 static int addr4_resolve_remote(struct sockaddr_in *src_in,
 209                                struct sockaddr_in *dst_in,
 210                                struct rdma_dev_addr *addr)
 211 {
 212         __be32 src_ip = src_in->sin_addr.s_addr;
 213         __be32 dst_ip = dst_in->sin_addr.s_addr;
 214         struct flowi fl;
 215         struct rtable *rt;
 216         struct neighbour *neigh;
 217         int ret;
 218
 219         memset(&fl, 0, sizeof fl);
 220         fl.nl_u.ip4_u.daddr = dst_ip;
 221         fl.nl_u.ip4_u.saddr = src_ip;
 222         ret = ip_route_output_key(&init_net, &rt, &fl);
 223         if (ret)
 224                 goto out;
 225
 226         /* If the device does ARP internally, return 'done' */
 227         if (rt->idev->dev->flags & IFF_NOARP) {
 228                 rdma_copy_addr(addr, rt->idev->dev, NULL);
 229                 goto put;
 230         }
 231
 232         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
 233         if (!neigh) {
 234                 ret = -ENODATA;
 235                 goto put;
 236         }
 237
 238         if (!(neigh->nud_state & NUD_VALID)) {
 239                 ret = -ENODATA;
 240                 goto release;
 241         }
 242
 243         if (!src_ip) {
 244                 src_in->sin_family = dst_in->sin_family;
 245                 src_in->sin_addr.s_addr = rt->rt_src;
 246         }
 247
 248         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
 249 release:
 250         neigh_release(neigh);
 251 put:
 252         ip_rt_put(rt);
 253 out:
 254         return ret;
 255 }
 256
 257 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
 258                                struct sockaddr_in6 *dst_in,
 259                                struct rdma_dev_addr *addr)
 260 {
 261         struct flowi fl;
 262         struct neighbour *neigh;
 263         struct dst_entry *dst;
 264         int ret = -ENODATA;
 265
 266         memset(&fl, 0, sizeof fl);
 267         fl.nl_u.ip6_u.daddr = dst_in->sin6_addr;
 268         fl.nl_u.ip6_u.saddr = src_in->sin6_addr;
 269
 270         dst = ip6_route_output(&init_net, NULL, &fl);
 271         if (!dst)
 272                 return ret;
 273
 274         if (dst->dev->flags & IFF_NOARP) {
 275                 ret = rdma_copy_addr(addr, dst->dev, NULL);
 276         } else {
 277                 neigh = dst->neighbour;
 278                 if (neigh && (neigh->nud_state & NUD_VALID))
 279                         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
 280         }
 281
 282         dst_release(dst);
 283         return ret;
 284 }
 285
 286 static int addr_resolve_remote(struct sockaddr *src_in,
 287                                 struct sockaddr *dst_in,
 288                                 struct rdma_dev_addr *addr)
 289 {
 290         if (src_in->sa_family == AF_INET) {
 291                 return addr4_resolve_remote((struct sockaddr_in *) src_in,
 292                         (struct sockaddr_in *) dst_in, addr);
 293         } else
 294                 return addr6_resolve_remote((struct sockaddr_in6 *) src_in,
 295                         (struct sockaddr_in6 *) dst_in, addr);
 296 }
 297
 298 static void process_req(struct work_struct *work)
 299 {
 300         struct addr_req *req, *temp_req;
 301         struct sockaddr *src_in, *dst_in;
 302         struct list_head done_list;
 303
 304         INIT_LIST_HEAD(&done_list);
 305
 306         mutex_lock(&lock);
 307         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 308                 if (req->status == -ENODATA) {
 309                         src_in = (struct sockaddr *) &req->src_addr;
 310                         dst_in = (struct sockaddr *) &req->dst_addr;
 311                         req->status = addr_resolve_remote(src_in, dst_in,
 312                                                           req->addr);
 313                         if (req->status && time_after_eq(jiffies, req->timeout))
 314                                 req->status = -ETIMEDOUT;
 315                         else if (req->status == -ENODATA)
 316                                 continue;
 317                 }
 318                 list_move_tail(&req->list, &done_list);
 319         }
 320
 321         if (!list_empty(&req_list)) {
 322                 req = list_entry(req_list.next, struct addr_req, list);
 323                 set_timeout(req->timeout);
 324         }
 325         mutex_unlock(&lock);
 326
 327         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 328                 list_del(&req->list);
 329                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 330                         req->addr, req->context);
 331                 put_client(req->client);
 332                 kfree(req);
 333         }
 334 }
 335
 336 static int addr_resolve_local(struct sockaddr *src_in,
 337                               struct sockaddr *dst_in,
 338                               struct rdma_dev_addr *addr)
 339 {
 340         struct net_device *dev;
 341         int ret;
 342
 343         if (dst_in->sa_family == AF_INET) {
 344                 __be32 src_ip = ((struct sockaddr_in *) src_in)->sin_addr.s_addr;
 345                 __be32 dst_ip = ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
 346
 347                 dev = ip_dev_find(&init_net, dst_ip);
 348                 if (!dev)
 349                         return -EADDRNOTAVAIL;
 350
 351                 if (ipv4_is_zeronet(src_ip)) {
 352                         src_in->sa_family = dst_in->sa_family;
 353                         ((struct sockaddr_in *) src_in)->sin_addr.s_addr = dst_ip;
 354                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
 355                 } else if (ipv4_is_loopback(src_ip)) {
 356                         ret = rdma_translate_ip(dst_in, addr);
 357                         if (!ret)
 358                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 359                 } else {
 360                         ret = rdma_translate_ip(src_in, addr);
 361                         if (!ret)
 362                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 363                 }
 364                 dev_put(dev);
 365         } else {
 366                 struct in6_addr *a;
 367
 368                 for_each_netdev(&init_net, dev)
 369                         if (ipv6_chk_addr(&init_net,
 370                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
 371                                           dev, 1))
 372                                 break;
 373
 374                 if (!dev)
 375                         return -EADDRNOTAVAIL;
 376
 377                 a = &((struct sockaddr_in6 *) src_in)->sin6_addr;
 378
 379                 if (ipv6_addr_any(a)) {
 380                         src_in->sa_family = dst_in->sa_family;
 381                         ((struct sockaddr_in6 *) src_in)->sin6_addr =
 382                                 ((struct sockaddr_in6 *) dst_in)->sin6_addr;
 383                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
 384                 } else if (ipv6_addr_loopback(a)) {
 385                         ret = rdma_translate_ip(dst_in, addr);
 386                         if (!ret)
 387                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 388                 } else  {
 389                         ret = rdma_translate_ip(src_in, addr);
 390                         if (!ret)
 391                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 392                 }
 393         }
 394
 395         return ret;
 396 }
 397
 398 int rdma_resolve_ip(struct rdma_addr_client *client,
 399                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 400                     struct rdma_dev_addr *addr, int timeout_ms,
 401                     void (*callback)(int status, struct sockaddr *src_addr,
 402                                      struct rdma_dev_addr *addr, void *context),
 403                     void *context)
 404 {
 405         struct sockaddr *src_in, *dst_in;
 406         struct addr_req *req;
 407         int ret = 0;
 408
 409         req = kzalloc(sizeof *req, GFP_KERNEL);
 410         if (!req)
 411                 return -ENOMEM;
 412
 413         if (src_addr)
 414                 memcpy(&req->src_addr, src_addr, ip_addr_size(src_addr));
 415         memcpy(&req->dst_addr, dst_addr, ip_addr_size(dst_addr));
 416         req->addr = addr;
 417         req->callback = callback;
 418         req->context = context;
 419         req->client = client;
 420         atomic_inc(&client->refcount);
 421
 422         src_in = (struct sockaddr *) &req->src_addr;
 423         dst_in = (struct sockaddr *) &req->dst_addr;
 424
 425         req->status = addr_resolve_local(src_in, dst_in, addr);
 426         if (req->status == -EADDRNOTAVAIL)
 427                 req->status = addr_resolve_remote(src_in, dst_in, addr);
 428
 429         switch (req->status) {
 430         case 0:
 431                 req->timeout = jiffies;
 432                 queue_req(req);
 433                 break;
 434         case -ENODATA:
 435                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 436                 queue_req(req);
 437                 addr_send_arp(dst_in);
 438                 break;
 439         default:
 440                 ret = req->status;
 441                 atomic_dec(&client->refcount);
 442                 kfree(req);
 443                 break;
 444         }
 445         return ret;
 446 }
 447 EXPORT_SYMBOL(rdma_resolve_ip);
 448
 449 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 450 {
 451         struct addr_req *req, *temp_req;
 452
 453         mutex_lock(&lock);
 454         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 455                 if (req->addr == addr) {
 456                         req->status = -ECANCELED;
 457                         req->timeout = jiffies;
 458                         list_move(&req->list, &req_list);
 459                         set_timeout(req->timeout);
 460                         break;
 461                 }
 462         }
 463         mutex_unlock(&lock);
 464 }
 465 EXPORT_SYMBOL(rdma_addr_cancel);
 466
 467 static int netevent_callback(struct notifier_block *self, unsigned long event,
 468         void *ctx)
 469 {
 470         if (event == NETEVENT_NEIGH_UPDATE) {
 471                 struct neighbour *neigh = ctx;
 472
 473                 if (neigh->nud_state & NUD_VALID) {
 474                         set_timeout(jiffies);
 475                 }
 476         }
 477         return 0;
 478 }
 479
 480 static struct notifier_block nb = {
 481         .notifier_call = netevent_callback
 482 };
 483
 484 static int addr_init(void)
 485 {
 486         addr_wq = create_singlethread_workqueue("ib_addr");
 487         if (!addr_wq)
 488                 return -ENOMEM;
 489
 490         register_netevent_notifier(&nb);
 491         return 0;
 492 }
 493
 494 static void addr_cleanup(void)
 495 {
 496         unregister_netevent_notifier(&nb);
 497         destroy_workqueue(addr_wq);
 498 }
 499
 500 module_init(addr_init);
 501 module_exit(addr_cleanup);