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 #include <rdma/ib.h>
  49
  50 MODULE_AUTHOR("Sean Hefty");
  51 MODULE_DESCRIPTION("IB Address Translation");
  52 MODULE_LICENSE("Dual BSD/GPL");
  53
  54 struct addr_req {
  55         struct list_head list;
  56         struct sockaddr_storage src_addr;
  57         struct sockaddr_storage dst_addr;
  58         struct rdma_dev_addr *addr;
  59         struct rdma_addr_client *client;
  60         void *context;
  61         void (*callback)(int status, struct sockaddr *src_addr,
  62                          struct rdma_dev_addr *addr, void *context);
  63         unsigned long timeout;
  64         int status;
  65 };
  66
  67 static void process_req(struct work_struct *work);
  68
  69 static DEFINE_MUTEX(lock);
  70 static LIST_HEAD(req_list);
  71 static DECLARE_DELAYED_WORK(work, process_req);
  72 static struct workqueue_struct *addr_wq;
  73
  74 int rdma_addr_size(struct sockaddr *addr)
  75 {
  76         switch (addr->sa_family) {
  77         case AF_INET:
  78                 return sizeof(struct sockaddr_in);
  79         case AF_INET6:
  80                 return sizeof(struct sockaddr_in6);
  81         case AF_IB:
  82                 return sizeof(struct sockaddr_ib);
  83         default:
  84                 return 0;
  85         }
  86 }
  87 EXPORT_SYMBOL(rdma_addr_size);
  88
  89 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
  90 {
  91         int ret = rdma_addr_size((struct sockaddr *) addr);
  92
  93         return ret <= sizeof(*addr) ? ret : 0;
  94 }
  95 EXPORT_SYMBOL(rdma_addr_size_in6);
  96
  97 int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
  98 {
  99         int ret = rdma_addr_size((struct sockaddr *) addr);
 100
 101         return ret <= sizeof(*addr) ? ret : 0;
 102 }
 103 EXPORT_SYMBOL(rdma_addr_size_kss);
 104
 105 static struct rdma_addr_client self;
 106
 107 void rdma_addr_register_client(struct rdma_addr_client *client)
 108 {
 109         atomic_set(&client->refcount, 1);
 110         init_completion(&client->comp);
 111 }
 112 EXPORT_SYMBOL(rdma_addr_register_client);
 113
 114 static inline void put_client(struct rdma_addr_client *client)
 115 {
 116         if (atomic_dec_and_test(&client->refcount))
 117                 complete(&client->comp);
 118 }
 119
 120 void rdma_addr_unregister_client(struct rdma_addr_client *client)
 121 {
 122         put_client(client);
 123         wait_for_completion(&client->comp);
 124 }
 125 EXPORT_SYMBOL(rdma_addr_unregister_client);
 126
 127 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
 128                      const unsigned char *dst_dev_addr)
 129 {
 130         dev_addr->dev_type = dev->type;
 131         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 132         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
 133         if (dst_dev_addr)
 134                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 135         dev_addr->bound_dev_if = dev->ifindex;
 136         return 0;
 137 }
 138 EXPORT_SYMBOL(rdma_copy_addr);
 139
 140 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr,
 141                       u16 *vlan_id)
 142 {
 143         struct net_device *dev;
 144         int ret = -EADDRNOTAVAIL;
 145
 146         if (dev_addr->bound_dev_if) {
 147                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 148                 if (!dev)
 149                         return -ENODEV;
 150                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 151                 dev_put(dev);
 152                 return ret;
 153         }
 154
 155         switch (addr->sa_family) {
 156         case AF_INET:
 157                 dev = ip_dev_find(dev_addr->net,
 158                         ((struct sockaddr_in *) addr)->sin_addr.s_addr);
 159
 160                 if (!dev)
 161                         return ret;
 162
 163                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 164                 if (vlan_id)
 165                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 166                 dev_put(dev);
 167                 break;
 168 #if IS_ENABLED(CONFIG_IPV6)
 169         case AF_INET6:
 170                 rcu_read_lock();
 171                 for_each_netdev_rcu(dev_addr->net, dev) {
 172                         if (ipv6_chk_addr(dev_addr->net,
 173                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
 174                                           dev, 1)) {
 175                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 176                                 if (vlan_id)
 177                                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 178                                 break;
 179                         }
 180                 }
 181                 rcu_read_unlock();
 182                 break;
 183 #endif
 184         }
 185         return ret;
 186 }
 187 EXPORT_SYMBOL(rdma_translate_ip);
 188
 189 static void set_timeout(unsigned long time)
 190 {
 191         unsigned long delay;
 192
 193         delay = time - jiffies;
 194         if ((long)delay < 0)
 195                 delay = 0;
 196
 197         mod_delayed_work(addr_wq, &work, delay);
 198 }
 199
 200 static void queue_req(struct addr_req *req)
 201 {
 202         struct addr_req *temp_req;
 203
 204         mutex_lock(&lock);
 205         list_for_each_entry_reverse(temp_req, &req_list, list) {
 206                 if (time_after_eq(req->timeout, temp_req->timeout))
 207                         break;
 208         }
 209
 210         list_add(&req->list, &temp_req->list);
 211
 212         if (req_list.next == &req->list)
 213                 set_timeout(req->timeout);
 214         mutex_unlock(&lock);
 215 }
 216
 217 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr, void *daddr)
 218 {
 219         struct neighbour *n;
 220         int ret;
 221
 222         n = dst_neigh_lookup(dst, daddr);
 223
 224         rcu_read_lock();
 225         if (!n || !(n->nud_state & NUD_VALID)) {
 226                 if (n)
 227                         neigh_event_send(n, NULL);
 228                 ret = -ENODATA;
 229         } else {
 230                 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
 231         }
 232         rcu_read_unlock();
 233
 234         if (n)
 235                 neigh_release(n);
 236
 237         return ret;
 238 }
 239
 240 static int addr4_resolve(struct sockaddr_in *src_in,
 241                          struct sockaddr_in *dst_in,
 242                          struct rdma_dev_addr *addr)
 243 {
 244         __be32 src_ip = src_in->sin_addr.s_addr;
 245         __be32 dst_ip = dst_in->sin_addr.s_addr;
 246         struct rtable *rt;
 247         struct flowi4 fl4;
 248         int ret;
 249
 250         memset(&fl4, 0, sizeof(fl4));
 251         fl4.daddr = dst_ip;
 252         fl4.saddr = src_ip;
 253         fl4.flowi4_oif = addr->bound_dev_if;
 254         rt = ip_route_output_key(addr->net, &fl4);
 255         if (IS_ERR(rt)) {
 256                 ret = PTR_ERR(rt);
 257                 goto out;
 258         }
 259         src_in->sin_family = AF_INET;
 260         src_in->sin_addr.s_addr = fl4.saddr;
 261
 262         if (rt->dst.dev->flags & IFF_LOOPBACK) {
 263                 ret = rdma_translate_ip((struct sockaddr *)dst_in, addr, NULL);
 264                 if (!ret)
 265                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 266                 goto put;
 267         }
 268
 269         /* If the device does ARP internally, return 'done' */
 270         if (rt->dst.dev->flags & IFF_NOARP) {
 271                 ret = rdma_copy_addr(addr, rt->dst.dev, NULL);
 272                 goto put;
 273         }
 274
 275         ret = dst_fetch_ha(&rt->dst, addr, &fl4.daddr);
 276 put:
 277         ip_rt_put(rt);
 278 out:
 279         return ret;
 280 }
 281
 282 #if IS_ENABLED(CONFIG_IPV6)
 283 static int addr6_resolve(struct sockaddr_in6 *src_in,
 284                          struct sockaddr_in6 *dst_in,
 285                          struct rdma_dev_addr *addr)
 286 {
 287         struct flowi6 fl6;
 288         struct dst_entry *dst;
 289         int ret;
 290
 291         memset(&fl6, 0, sizeof fl6);
 292         fl6.daddr = dst_in->sin6_addr;
 293         fl6.saddr = src_in->sin6_addr;
 294         fl6.flowi6_oif = addr->bound_dev_if;
 295
 296         dst = ipv6_stub->ipv6_dst_lookup_flow(addr->net, NULL, &fl6, NULL);
 297         if (IS_ERR(dst))
 298                 return PTR_ERR(dst);
 299
 300         if (ipv6_addr_any(&fl6.saddr)) {
 301                 ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev,
 302                                          &fl6.daddr, 0, &fl6.saddr);
 303                 if (ret)
 304                         goto put;
 305
 306                 src_in->sin6_family = AF_INET6;
 307                 src_in->sin6_addr = fl6.saddr;
 308         }
 309
 310         if (dst->dev->flags & IFF_LOOPBACK) {
 311                 ret = rdma_translate_ip((struct sockaddr *)dst_in, addr, NULL);
 312                 if (!ret)
 313                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 314                 goto put;
 315         }
 316
 317         /* If the device does ARP internally, return 'done' */
 318         if (dst->dev->flags & IFF_NOARP) {
 319                 ret = rdma_copy_addr(addr, dst->dev, NULL);
 320                 goto put;
 321         }
 322
 323         ret = dst_fetch_ha(dst, addr, &fl6.daddr);
 324 put:
 325         dst_release(dst);
 326         return ret;
 327 }
 328 #else
 329 static int addr6_resolve(struct sockaddr_in6 *src_in,
 330                          struct sockaddr_in6 *dst_in,
 331                          struct rdma_dev_addr *addr)
 332 {
 333         return -EADDRNOTAVAIL;
 334 }
 335 #endif
 336
 337 static int addr_resolve(struct sockaddr *src_in,
 338                         struct sockaddr *dst_in,
 339                         struct rdma_dev_addr *addr)
 340 {
 341         if (src_in->sa_family == AF_INET) {
 342                 return addr4_resolve((struct sockaddr_in *) src_in,
 343                         (struct sockaddr_in *) dst_in, addr);
 344         } else
 345                 return addr6_resolve((struct sockaddr_in6 *) src_in,
 346                         (struct sockaddr_in6 *) dst_in, addr);
 347 }
 348
 349 static void process_req(struct work_struct *work)
 350 {
 351         struct addr_req *req, *temp_req;
 352         struct sockaddr *src_in, *dst_in;
 353         struct list_head done_list;
 354
 355         INIT_LIST_HEAD(&done_list);
 356
 357         mutex_lock(&lock);
 358         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 359                 if (req->status == -ENODATA) {
 360                         src_in = (struct sockaddr *) &req->src_addr;
 361                         dst_in = (struct sockaddr *) &req->dst_addr;
 362                         req->status = addr_resolve(src_in, dst_in, req->addr);
 363                         if (req->status && time_after_eq(jiffies, req->timeout))
 364                                 req->status = -ETIMEDOUT;
 365                         else if (req->status == -ENODATA)
 366                                 continue;
 367                 }
 368                 list_move_tail(&req->list, &done_list);
 369         }
 370
 371         if (!list_empty(&req_list)) {
 372                 req = list_entry(req_list.next, struct addr_req, list);
 373                 set_timeout(req->timeout);
 374         }
 375         mutex_unlock(&lock);
 376
 377         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 378                 list_del(&req->list);
 379                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 380                         req->addr, req->context);
 381                 put_client(req->client);
 382                 kfree(req);
 383         }
 384 }
 385
 386 int rdma_resolve_ip(struct rdma_addr_client *client,
 387                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 388                     struct rdma_dev_addr *addr, int timeout_ms,
 389                     void (*callback)(int status, struct sockaddr *src_addr,
 390                                      struct rdma_dev_addr *addr, void *context),
 391                     void *context)
 392 {
 393         struct sockaddr *src_in, *dst_in;
 394         struct addr_req *req;
 395         int ret = 0;
 396
 397         req = kzalloc(sizeof *req, GFP_KERNEL);
 398         if (!req)
 399                 return -ENOMEM;
 400
 401         src_in = (struct sockaddr *) &req->src_addr;
 402         dst_in = (struct sockaddr *) &req->dst_addr;
 403
 404         if (src_addr) {
 405                 if (src_addr->sa_family != dst_addr->sa_family) {
 406                         ret = -EINVAL;
 407                         goto err;
 408                 }
 409
 410                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 411         } else {
 412                 src_in->sa_family = dst_addr->sa_family;
 413         }
 414
 415         memcpy(dst_in, dst_addr, rdma_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         req->status = addr_resolve(src_in, dst_in, addr);
 423         switch (req->status) {
 424         case 0:
 425                 req->timeout = jiffies;
 426                 queue_req(req);
 427                 break;
 428         case -ENODATA:
 429                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 430                 queue_req(req);
 431                 break;
 432         default:
 433                 ret = req->status;
 434                 atomic_dec(&client->refcount);
 435                 goto err;
 436         }
 437         return ret;
 438 err:
 439         kfree(req);
 440         return ret;
 441 }
 442 EXPORT_SYMBOL(rdma_resolve_ip);
 443
 444 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 445 {
 446         struct addr_req *req, *temp_req;
 447
 448         mutex_lock(&lock);
 449         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 450                 if (req->addr == addr) {
 451                         req->status = -ECANCELED;
 452                         req->timeout = jiffies;
 453                         list_move(&req->list, &req_list);
 454                         set_timeout(req->timeout);
 455                         break;
 456                 }
 457         }
 458         mutex_unlock(&lock);
 459 }
 460 EXPORT_SYMBOL(rdma_addr_cancel);
 461
 462 struct resolve_cb_context {
 463         struct rdma_dev_addr *addr;
 464         struct completion comp;
 465 };
 466
 467 static void resolve_cb(int status, struct sockaddr *src_addr,
 468              struct rdma_dev_addr *addr, void *context)
 469 {
 470         memcpy(((struct resolve_cb_context *)context)->addr, addr, sizeof(struct
 471                                 rdma_dev_addr));
 472         complete(&((struct resolve_cb_context *)context)->comp);
 473 }
 474
 475 int rdma_addr_find_dmac_by_grh(const union ib_gid *sgid, const union ib_gid *dgid,
 476                                u8 *dmac, u16 *vlan_id, int if_index)
 477 {
 478         int ret = 0;
 479         struct rdma_dev_addr dev_addr;
 480         struct resolve_cb_context ctx;
 481         struct net_device *dev;
 482
 483         union {
 484                 struct sockaddr_in  _sockaddr_in;
 485                 struct sockaddr_in6 _sockaddr_in6;
 486         } sgid_addr, dgid_addr;
 487
 488
 489         rdma_gid2ip((struct sockaddr *)&sgid_addr, sgid);
 490         rdma_gid2ip((struct sockaddr *)&dgid_addr, dgid);
 491
 492         memset(&dev_addr, 0, sizeof(dev_addr));
 493         dev_addr.bound_dev_if = if_index;
 494         dev_addr.net = &init_net;
 495
 496         ctx.addr = &dev_addr;
 497         init_completion(&ctx.comp);
 498         ret = rdma_resolve_ip(&self, (struct sockaddr *)&sgid_addr,
 499                               (struct sockaddr *)&dgid_addr, &dev_addr, 1000,
 500                               resolve_cb, &ctx);
 501         if (ret)
 502                 return ret;
 503
 504         wait_for_completion(&ctx.comp);
 505
 506         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 507         dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
 508         if (!dev)
 509                 return -ENODEV;
 510         if (vlan_id)
 511                 *vlan_id = rdma_vlan_dev_vlan_id(dev);
 512         dev_put(dev);
 513         return ret;
 514 }
 515 EXPORT_SYMBOL(rdma_addr_find_dmac_by_grh);
 516
 517 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
 518 {
 519         int ret = 0;
 520         struct rdma_dev_addr dev_addr;
 521         union {
 522                 struct sockaddr_in  _sockaddr_in;
 523                 struct sockaddr_in6 _sockaddr_in6;
 524         } gid_addr;
 525
 526         rdma_gid2ip((struct sockaddr *)&gid_addr, sgid);
 527
 528         memset(&dev_addr, 0, sizeof(dev_addr));
 529         dev_addr.net = &init_net;
 530         ret = rdma_translate_ip((struct sockaddr *)&gid_addr, &dev_addr, vlan_id);
 531         if (ret)
 532                 return ret;
 533
 534         memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
 535         return ret;
 536 }
 537 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
 538
 539 static int netevent_callback(struct notifier_block *self, unsigned long event,
 540         void *ctx)
 541 {
 542         if (event == NETEVENT_NEIGH_UPDATE) {
 543                 struct neighbour *neigh = ctx;
 544
 545                 if (neigh->nud_state & NUD_VALID) {
 546                         set_timeout(jiffies);
 547                 }
 548         }
 549         return 0;
 550 }
 551
 552 static struct notifier_block nb = {
 553         .notifier_call = netevent_callback
 554 };
 555
 556 static int __init addr_init(void)
 557 {
 558         addr_wq = create_singlethread_workqueue("ib_addr");
 559         if (!addr_wq)
 560                 return -ENOMEM;
 561
 562         register_netevent_notifier(&nb);
 563         rdma_addr_register_client(&self);
 564         return 0;
 565 }
 566
 567 static void __exit addr_cleanup(void)
 568 {
 569         rdma_addr_unregister_client(&self);
 570         unregister_netevent_notifier(&nb);
 571         destroy_workqueue(addr_wq);
 572 }
 573
 574 module_init(addr_init);
 575 module_exit(addr_cleanup);