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/addrconf.h>
  46 #include <net/ip6_route.h>
  47 #include <rdma/ib_addr.h>
  48
  49 MODULE_AUTHOR("Sean Hefty");
  50 MODULE_DESCRIPTION("IB Address Translation");
  51 MODULE_LICENSE("Dual BSD/GPL");
  52
  53 struct addr_req {
  54         struct list_head list;
  55         struct sockaddr_storage src_addr;
  56         struct sockaddr_storage dst_addr;
  57         struct rdma_dev_addr *addr;
  58         struct rdma_addr_client *client;
  59         void *context;
  60         void (*callback)(int status, struct sockaddr *src_addr,
  61                          struct rdma_dev_addr *addr, void *context);
  62         unsigned long timeout;
  63         int status;
  64 };
  65
  66 static void process_req(struct work_struct *work);
  67
  68 static DEFINE_MUTEX(lock);
  69 static LIST_HEAD(req_list);
  70 static DECLARE_DELAYED_WORK(work, process_req);
  71 static struct workqueue_struct *addr_wq;
  72
  73 void rdma_addr_register_client(struct rdma_addr_client *client)
  74 {
  75         atomic_set(&client->refcount, 1);
  76         init_completion(&client->comp);
  77 }
  78 EXPORT_SYMBOL(rdma_addr_register_client);
  79
  80 static inline void put_client(struct rdma_addr_client *client)
  81 {
  82         if (atomic_dec_and_test(&client->refcount))
  83                 complete(&client->comp);
  84 }
  85
  86 void rdma_addr_unregister_client(struct rdma_addr_client *client)
  87 {
  88         put_client(client);
  89         wait_for_completion(&client->comp);
  90 }
  91 EXPORT_SYMBOL(rdma_addr_unregister_client);
  92
  93 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
  94                      const unsigned char *dst_dev_addr)
  95 {
  96         dev_addr->dev_type = dev->type;
  97         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
  98         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
  99         if (dst_dev_addr)
 100                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 101         dev_addr->bound_dev_if = dev->ifindex;
 102         return 0;
 103 }
 104 EXPORT_SYMBOL(rdma_copy_addr);
 105
 106 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 107 {
 108         struct net_device *dev;
 109         int ret = -EADDRNOTAVAIL;
 110
 111         if (dev_addr->bound_dev_if) {
 112                 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
 113                 if (!dev)
 114                         return -ENODEV;
 115                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 116                 dev_put(dev);
 117                 return ret;
 118         }
 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                 rcu_read_lock();
 135                 for_each_netdev_rcu(&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                 rcu_read_unlock();
 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 int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr, void *daddr)
 182 {
 183         struct neighbour *n;
 184         int ret;
 185
 186         n = dst_neigh_lookup(dst, daddr);
 187
 188         rcu_read_lock();
 189         if (!n || !(n->nud_state & NUD_VALID)) {
 190                 if (n)
 191                         neigh_event_send(n, NULL);
 192                 ret = -ENODATA;
 193         } else {
 194                 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
 195         }
 196         rcu_read_unlock();
 197
 198         if (n)
 199                 neigh_release(n);
 200
 201         return ret;
 202 }
 203
 204 static int addr4_resolve(struct sockaddr_in *src_in,
 205                          struct sockaddr_in *dst_in,
 206                          struct rdma_dev_addr *addr)
 207 {
 208         __be32 src_ip = src_in->sin_addr.s_addr;
 209         __be32 dst_ip = dst_in->sin_addr.s_addr;
 210         struct rtable *rt;
 211         struct flowi4 fl4;
 212         int ret;
 213
 214         memset(&fl4, 0, sizeof(fl4));
 215         fl4.daddr = dst_ip;
 216         fl4.saddr = src_ip;
 217         fl4.flowi4_oif = addr->bound_dev_if;
 218         rt = ip_route_output_key(&init_net, &fl4);
 219         if (IS_ERR(rt)) {
 220                 ret = PTR_ERR(rt);
 221                 goto out;
 222         }
 223         src_in->sin_family = AF_INET;
 224         src_in->sin_addr.s_addr = fl4.saddr;
 225
 226         if (rt->dst.dev->flags & IFF_LOOPBACK) {
 227                 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
 228                 if (!ret)
 229                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 230                 goto put;
 231         }
 232
 233         /* If the device does ARP internally, return 'done' */
 234         if (rt->dst.dev->flags & IFF_NOARP) {
 235                 ret = rdma_copy_addr(addr, rt->dst.dev, NULL);
 236                 goto put;
 237         }
 238
 239         ret = dst_fetch_ha(&rt->dst, addr, &fl4.daddr);
 240 put:
 241         ip_rt_put(rt);
 242 out:
 243         return ret;
 244 }
 245
 246 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 247 static int addr6_resolve(struct sockaddr_in6 *src_in,
 248                          struct sockaddr_in6 *dst_in,
 249                          struct rdma_dev_addr *addr)
 250 {
 251         struct flowi6 fl6;
 252         struct dst_entry *dst;
 253         int ret;
 254
 255         memset(&fl6, 0, sizeof fl6);
 256         fl6.daddr = dst_in->sin6_addr;
 257         fl6.saddr = src_in->sin6_addr;
 258         fl6.flowi6_oif = addr->bound_dev_if;
 259
 260         dst = ip6_route_output(&init_net, NULL, &fl6);
 261         if ((ret = dst->error))
 262                 goto put;
 263
 264         if (ipv6_addr_any(&fl6.saddr)) {
 265                 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
 266                                          &fl6.daddr, 0, &fl6.saddr);
 267                 if (ret)
 268                         goto put;
 269
 270                 src_in->sin6_family = AF_INET6;
 271                 src_in->sin6_addr = fl6.saddr;
 272         }
 273
 274         if (dst->dev->flags & IFF_LOOPBACK) {
 275                 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
 276                 if (!ret)
 277                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 278                 goto put;
 279         }
 280
 281         /* If the device does ARP internally, return 'done' */
 282         if (dst->dev->flags & IFF_NOARP) {
 283                 ret = rdma_copy_addr(addr, dst->dev, NULL);
 284                 goto put;
 285         }
 286
 287         ret = dst_fetch_ha(dst, addr, &fl6.daddr);
 288 put:
 289         dst_release(dst);
 290         return ret;
 291 }
 292 #else
 293 static int addr6_resolve(struct sockaddr_in6 *src_in,
 294                          struct sockaddr_in6 *dst_in,
 295                          struct rdma_dev_addr *addr)
 296 {
 297         return -EADDRNOTAVAIL;
 298 }
 299 #endif
 300
 301 static int addr_resolve(struct sockaddr *src_in,
 302                         struct sockaddr *dst_in,
 303                         struct rdma_dev_addr *addr)
 304 {
 305         if (src_in->sa_family == AF_INET) {
 306                 return addr4_resolve((struct sockaddr_in *) src_in,
 307                         (struct sockaddr_in *) dst_in, addr);
 308         } else
 309                 return addr6_resolve((struct sockaddr_in6 *) src_in,
 310                         (struct sockaddr_in6 *) dst_in, addr);
 311 }
 312
 313 static void process_req(struct work_struct *work)
 314 {
 315         struct addr_req *req, *temp_req;
 316         struct sockaddr *src_in, *dst_in;
 317         struct list_head done_list;
 318
 319         INIT_LIST_HEAD(&done_list);
 320
 321         mutex_lock(&lock);
 322         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 323                 if (req->status == -ENODATA) {
 324                         src_in = (struct sockaddr *) &req->src_addr;
 325                         dst_in = (struct sockaddr *) &req->dst_addr;
 326                         req->status = addr_resolve(src_in, dst_in, req->addr);
 327                         if (req->status && time_after_eq(jiffies, req->timeout))
 328                                 req->status = -ETIMEDOUT;
 329                         else if (req->status == -ENODATA)
 330                                 continue;
 331                 }
 332                 list_move_tail(&req->list, &done_list);
 333         }
 334
 335         if (!list_empty(&req_list)) {
 336                 req = list_entry(req_list.next, struct addr_req, list);
 337                 set_timeout(req->timeout);
 338         }
 339         mutex_unlock(&lock);
 340
 341         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 342                 list_del(&req->list);
 343                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 344                         req->addr, req->context);
 345                 put_client(req->client);
 346                 kfree(req);
 347         }
 348 }
 349
 350 int rdma_resolve_ip(struct rdma_addr_client *client,
 351                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 352                     struct rdma_dev_addr *addr, int timeout_ms,
 353                     void (*callback)(int status, struct sockaddr *src_addr,
 354                                      struct rdma_dev_addr *addr, void *context),
 355                     void *context)
 356 {
 357         struct sockaddr *src_in, *dst_in;
 358         struct addr_req *req;
 359         int ret = 0;
 360
 361         req = kzalloc(sizeof *req, GFP_KERNEL);
 362         if (!req)
 363                 return -ENOMEM;
 364
 365         src_in = (struct sockaddr *) &req->src_addr;
 366         dst_in = (struct sockaddr *) &req->dst_addr;
 367
 368         if (src_addr) {
 369                 if (src_addr->sa_family != dst_addr->sa_family) {
 370                         ret = -EINVAL;
 371                         goto err;
 372                 }
 373
 374                 memcpy(src_in, src_addr, ip_addr_size(src_addr));
 375         } else {
 376                 src_in->sa_family = dst_addr->sa_family;
 377         }
 378
 379         memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
 380         req->addr = addr;
 381         req->callback = callback;
 382         req->context = context;
 383         req->client = client;
 384         atomic_inc(&client->refcount);
 385
 386         req->status = addr_resolve(src_in, dst_in, addr);
 387         switch (req->status) {
 388         case 0:
 389                 req->timeout = jiffies;
 390                 queue_req(req);
 391                 break;
 392         case -ENODATA:
 393                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 394                 queue_req(req);
 395                 break;
 396         default:
 397                 ret = req->status;
 398                 atomic_dec(&client->refcount);
 399                 goto err;
 400         }
 401         return ret;
 402 err:
 403         kfree(req);
 404         return ret;
 405 }
 406 EXPORT_SYMBOL(rdma_resolve_ip);
 407
 408 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 409 {
 410         struct addr_req *req, *temp_req;
 411
 412         mutex_lock(&lock);
 413         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 414                 if (req->addr == addr) {
 415                         req->status = -ECANCELED;
 416                         req->timeout = jiffies;
 417                         list_move(&req->list, &req_list);
 418                         set_timeout(req->timeout);
 419                         break;
 420                 }
 421         }
 422         mutex_unlock(&lock);
 423 }
 424 EXPORT_SYMBOL(rdma_addr_cancel);
 425
 426 static int netevent_callback(struct notifier_block *self, unsigned long event,
 427         void *ctx)
 428 {
 429         if (event == NETEVENT_NEIGH_UPDATE) {
 430                 struct neighbour *neigh = ctx;
 431
 432                 if (neigh->nud_state & NUD_VALID) {
 433                         set_timeout(jiffies);
 434                 }
 435         }
 436         return 0;
 437 }
 438
 439 static struct notifier_block nb = {
 440         .notifier_call = netevent_callback
 441 };
 442
 443 static int __init addr_init(void)
 444 {
 445         addr_wq = create_singlethread_workqueue("ib_addr");
 446         if (!addr_wq)
 447                 return -ENOMEM;
 448
 449         register_netevent_notifier(&nb);
 450         return 0;
 451 }
 452
 453 static void __exit addr_cleanup(void)
 454 {
 455         unregister_netevent_notifier(&nb);
 456         destroy_workqueue(addr_wq);
 457 }
 458
 459 module_init(addr_init);
 460 module_exit(addr_cleanup);