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 static struct rdma_addr_client self;
  90
  91 void rdma_addr_register_client(struct rdma_addr_client *client)
  92 {
  93         atomic_set(&client->refcount, 1);
  94         init_completion(&client->comp);
  95 }
  96 EXPORT_SYMBOL(rdma_addr_register_client);
  97
  98 static inline void put_client(struct rdma_addr_client *client)
  99 {
 100         if (atomic_dec_and_test(&client->refcount))
 101                 complete(&client->comp);
 102 }
 103
 104 void rdma_addr_unregister_client(struct rdma_addr_client *client)
 105 {
 106         put_client(client);
 107         wait_for_completion(&client->comp);
 108 }
 109 EXPORT_SYMBOL(rdma_addr_unregister_client);
 110
 111 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
 112                      const unsigned char *dst_dev_addr)
 113 {
 114         dev_addr->dev_type = dev->type;
 115         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 116         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
 117         if (dst_dev_addr)
 118                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 119         dev_addr->bound_dev_if = dev->ifindex;
 120         return 0;
 121 }
 122 EXPORT_SYMBOL(rdma_copy_addr);
 123
 124 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr,
 125                       u16 *vlan_id)
 126 {
 127         struct net_device *dev;
 128         int ret = -EADDRNOTAVAIL;
 129
 130         if (dev_addr->bound_dev_if) {
 131                 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
 132                 if (!dev)
 133                         return -ENODEV;
 134                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 135                 dev_put(dev);
 136                 return ret;
 137         }
 138
 139         switch (addr->sa_family) {
 140         case AF_INET:
 141                 dev = ip_dev_find(dev_addr->net,
 142                         ((struct sockaddr_in *) addr)->sin_addr.s_addr);
 143
 144                 if (!dev)
 145                         return ret;
 146
 147                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 148                 if (vlan_id)
 149                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 150                 dev_put(dev);
 151                 break;
 152 #if IS_ENABLED(CONFIG_IPV6)
 153         case AF_INET6:
 154                 rcu_read_lock();
 155                 for_each_netdev_rcu(dev_addr->net, dev) {
 156                         if (ipv6_chk_addr(dev_addr->net,
 157                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
 158                                           dev, 1)) {
 159                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
 160                                 if (vlan_id)
 161                                         *vlan_id = rdma_vlan_dev_vlan_id(dev);
 162                                 break;
 163                         }
 164                 }
 165                 rcu_read_unlock();
 166                 break;
 167 #endif
 168         }
 169         return ret;
 170 }
 171 EXPORT_SYMBOL(rdma_translate_ip);
 172
 173 static void set_timeout(unsigned long time)
 174 {
 175         unsigned long delay;
 176
 177         delay = time - jiffies;
 178         if ((long)delay < 0)
 179                 delay = 0;
 180
 181         mod_delayed_work(addr_wq, &work, delay);
 182 }
 183
 184 static void queue_req(struct addr_req *req)
 185 {
 186         struct addr_req *temp_req;
 187
 188         mutex_lock(&lock);
 189         list_for_each_entry_reverse(temp_req, &req_list, list) {
 190                 if (time_after_eq(req->timeout, temp_req->timeout))
 191                         break;
 192         }
 193
 194         list_add(&req->list, &temp_req->list);
 195
 196         if (req_list.next == &req->list)
 197                 set_timeout(req->timeout);
 198         mutex_unlock(&lock);
 199 }
 200
 201 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr, void *daddr)
 202 {
 203         struct neighbour *n;
 204         int ret;
 205
 206         n = dst_neigh_lookup(dst, daddr);
 207
 208         rcu_read_lock();
 209         if (!n || !(n->nud_state & NUD_VALID)) {
 210                 if (n)
 211                         neigh_event_send(n, NULL);
 212                 ret = -ENODATA;
 213         } else {
 214                 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
 215         }
 216         rcu_read_unlock();
 217
 218         if (n)
 219                 neigh_release(n);
 220
 221         return ret;
 222 }
 223
 224 static int addr4_resolve(struct sockaddr_in *src_in,
 225                          struct sockaddr_in *dst_in,
 226                          struct rdma_dev_addr *addr)
 227 {
 228         __be32 src_ip = src_in->sin_addr.s_addr;
 229         __be32 dst_ip = dst_in->sin_addr.s_addr;
 230         struct rtable *rt;
 231         struct flowi4 fl4;
 232         int ret;
 233
 234         memset(&fl4, 0, sizeof(fl4));
 235         fl4.daddr = dst_ip;
 236         fl4.saddr = src_ip;
 237         fl4.flowi4_oif = addr->bound_dev_if;
 238         rt = ip_route_output_key(addr->net, &fl4);
 239         if (IS_ERR(rt)) {
 240                 ret = PTR_ERR(rt);
 241                 goto out;
 242         }
 243         src_in->sin_family = AF_INET;
 244         src_in->sin_addr.s_addr = fl4.saddr;
 245
 246         if (rt->dst.dev->flags & IFF_LOOPBACK) {
 247                 ret = rdma_translate_ip((struct sockaddr *)dst_in, addr, NULL);
 248                 if (!ret)
 249                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 250                 goto put;
 251         }
 252
 253         /* If the device does ARP internally, return 'done' */
 254         if (rt->dst.dev->flags & IFF_NOARP) {
 255                 ret = rdma_copy_addr(addr, rt->dst.dev, NULL);
 256                 goto put;
 257         }
 258
 259         ret = dst_fetch_ha(&rt->dst, addr, &fl4.daddr);
 260 put:
 261         ip_rt_put(rt);
 262 out:
 263         return ret;
 264 }
 265
 266 #if IS_ENABLED(CONFIG_IPV6)
 267 static int addr6_resolve(struct sockaddr_in6 *src_in,
 268                          struct sockaddr_in6 *dst_in,
 269                          struct rdma_dev_addr *addr)
 270 {
 271         struct flowi6 fl6;
 272         struct dst_entry *dst;
 273         int ret;
 274
 275         memset(&fl6, 0, sizeof fl6);
 276         fl6.daddr = dst_in->sin6_addr;
 277         fl6.saddr = src_in->sin6_addr;
 278         fl6.flowi6_oif = addr->bound_dev_if;
 279
 280         dst = ip6_route_output(addr->net, NULL, &fl6);
 281         if ((ret = dst->error))
 282                 goto put;
 283
 284         if (ipv6_addr_any(&fl6.saddr)) {
 285                 ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev,
 286                                          &fl6.daddr, 0, &fl6.saddr);
 287                 if (ret)
 288                         goto put;
 289
 290                 src_in->sin6_family = AF_INET6;
 291                 src_in->sin6_addr = fl6.saddr;
 292         }
 293
 294         if (dst->dev->flags & IFF_LOOPBACK) {
 295                 ret = rdma_translate_ip((struct sockaddr *)dst_in, addr, NULL);
 296                 if (!ret)
 297                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
 298                 goto put;
 299         }
 300
 301         /* If the device does ARP internally, return 'done' */
 302         if (dst->dev->flags & IFF_NOARP) {
 303                 ret = rdma_copy_addr(addr, dst->dev, NULL);
 304                 goto put;
 305         }
 306
 307         ret = dst_fetch_ha(dst, addr, &fl6.daddr);
 308 put:
 309         dst_release(dst);
 310         return ret;
 311 }
 312 #else
 313 static int addr6_resolve(struct sockaddr_in6 *src_in,
 314                          struct sockaddr_in6 *dst_in,
 315                          struct rdma_dev_addr *addr)
 316 {
 317         return -EADDRNOTAVAIL;
 318 }
 319 #endif
 320
 321 static int addr_resolve(struct sockaddr *src_in,
 322                         struct sockaddr *dst_in,
 323                         struct rdma_dev_addr *addr)
 324 {
 325         if (src_in->sa_family == AF_INET) {
 326                 return addr4_resolve((struct sockaddr_in *) src_in,
 327                         (struct sockaddr_in *) dst_in, addr);
 328         } else
 329                 return addr6_resolve((struct sockaddr_in6 *) src_in,
 330                         (struct sockaddr_in6 *) dst_in, addr);
 331 }
 332
 333 static void process_req(struct work_struct *work)
 334 {
 335         struct addr_req *req, *temp_req;
 336         struct sockaddr *src_in, *dst_in;
 337         struct list_head done_list;
 338
 339         INIT_LIST_HEAD(&done_list);
 340
 341         mutex_lock(&lock);
 342         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 343                 if (req->status == -ENODATA) {
 344                         src_in = (struct sockaddr *) &req->src_addr;
 345                         dst_in = (struct sockaddr *) &req->dst_addr;
 346                         req->status = addr_resolve(src_in, dst_in, req->addr);
 347                         if (req->status && time_after_eq(jiffies, req->timeout))
 348                                 req->status = -ETIMEDOUT;
 349                         else if (req->status == -ENODATA)
 350                                 continue;
 351                 }
 352                 list_move_tail(&req->list, &done_list);
 353         }
 354
 355         if (!list_empty(&req_list)) {
 356                 req = list_entry(req_list.next, struct addr_req, list);
 357                 set_timeout(req->timeout);
 358         }
 359         mutex_unlock(&lock);
 360
 361         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 362                 list_del(&req->list);
 363                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
 364                         req->addr, req->context);
 365                 put_client(req->client);
 366                 kfree(req);
 367         }
 368 }
 369
 370 int rdma_resolve_ip(struct rdma_addr_client *client,
 371                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 372                     struct rdma_dev_addr *addr, int timeout_ms,
 373                     void (*callback)(int status, struct sockaddr *src_addr,
 374                                      struct rdma_dev_addr *addr, void *context),
 375                     void *context)
 376 {
 377         struct sockaddr *src_in, *dst_in;
 378         struct addr_req *req;
 379         int ret = 0;
 380
 381         req = kzalloc(sizeof *req, GFP_KERNEL);
 382         if (!req)
 383                 return -ENOMEM;
 384
 385         src_in = (struct sockaddr *) &req->src_addr;
 386         dst_in = (struct sockaddr *) &req->dst_addr;
 387
 388         if (src_addr) {
 389                 if (src_addr->sa_family != dst_addr->sa_family) {
 390                         ret = -EINVAL;
 391                         goto err;
 392                 }
 393
 394                 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
 395         } else {
 396                 src_in->sa_family = dst_addr->sa_family;
 397         }
 398
 399         memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
 400         req->addr = addr;
 401         req->callback = callback;
 402         req->context = context;
 403         req->client = client;
 404         atomic_inc(&client->refcount);
 405
 406         req->status = addr_resolve(src_in, dst_in, addr);
 407         switch (req->status) {
 408         case 0:
 409                 req->timeout = jiffies;
 410                 queue_req(req);
 411                 break;
 412         case -ENODATA:
 413                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 414                 queue_req(req);
 415                 break;
 416         default:
 417                 ret = req->status;
 418                 atomic_dec(&client->refcount);
 419                 goto err;
 420         }
 421         return ret;
 422 err:
 423         kfree(req);
 424         return ret;
 425 }
 426 EXPORT_SYMBOL(rdma_resolve_ip);
 427
 428 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 429 {
 430         struct addr_req *req, *temp_req;
 431
 432         mutex_lock(&lock);
 433         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 434                 if (req->addr == addr) {
 435                         req->status = -ECANCELED;
 436                         req->timeout = jiffies;
 437                         list_move(&req->list, &req_list);
 438                         set_timeout(req->timeout);
 439                         break;
 440                 }
 441         }
 442         mutex_unlock(&lock);
 443 }
 444 EXPORT_SYMBOL(rdma_addr_cancel);
 445
 446 struct resolve_cb_context {
 447         struct rdma_dev_addr *addr;
 448         struct completion comp;
 449 };
 450
 451 static void resolve_cb(int status, struct sockaddr *src_addr,
 452              struct rdma_dev_addr *addr, void *context)
 453 {
 454         memcpy(((struct resolve_cb_context *)context)->addr, addr, sizeof(struct
 455                                 rdma_dev_addr));
 456         complete(&((struct resolve_cb_context *)context)->comp);
 457 }
 458
 459 int rdma_addr_find_dmac_by_grh(const union ib_gid *sgid, const union ib_gid *dgid,
 460                                u8 *dmac, u16 *vlan_id, int if_index)
 461 {
 462         int ret = 0;
 463         struct rdma_dev_addr dev_addr;
 464         struct resolve_cb_context ctx;
 465         struct net_device *dev;
 466
 467         union {
 468                 struct sockaddr     _sockaddr;
 469                 struct sockaddr_in  _sockaddr_in;
 470                 struct sockaddr_in6 _sockaddr_in6;
 471         } sgid_addr, dgid_addr;
 472
 473
 474         rdma_gid2ip(&sgid_addr._sockaddr, sgid);
 475         rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 476
 477         memset(&dev_addr, 0, sizeof(dev_addr));
 478         dev_addr.bound_dev_if = if_index;
 479         dev_addr.net = &init_net;
 480
 481         ctx.addr = &dev_addr;
 482         init_completion(&ctx.comp);
 483         ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
 484                         &dev_addr, 1000, resolve_cb, &ctx);
 485         if (ret)
 486                 return ret;
 487
 488         wait_for_completion(&ctx.comp);
 489
 490         memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
 491         dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
 492         if (!dev)
 493                 return -ENODEV;
 494         if (vlan_id)
 495                 *vlan_id = rdma_vlan_dev_vlan_id(dev);
 496         dev_put(dev);
 497         return ret;
 498 }
 499 EXPORT_SYMBOL(rdma_addr_find_dmac_by_grh);
 500
 501 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
 502 {
 503         int ret = 0;
 504         struct rdma_dev_addr dev_addr;
 505         union {
 506                 struct sockaddr     _sockaddr;
 507                 struct sockaddr_in  _sockaddr_in;
 508                 struct sockaddr_in6 _sockaddr_in6;
 509         } gid_addr;
 510
 511         rdma_gid2ip(&gid_addr._sockaddr, sgid);
 512
 513         memset(&dev_addr, 0, sizeof(dev_addr));
 514         dev_addr.net = &init_net;
 515         ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
 516         if (ret)
 517                 return ret;
 518
 519         memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
 520         return ret;
 521 }
 522 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
 523
 524 static int netevent_callback(struct notifier_block *self, unsigned long event,
 525         void *ctx)
 526 {
 527         if (event == NETEVENT_NEIGH_UPDATE) {
 528                 struct neighbour *neigh = ctx;
 529
 530                 if (neigh->nud_state & NUD_VALID) {
 531                         set_timeout(jiffies);
 532                 }
 533         }
 534         return 0;
 535 }
 536
 537 static struct notifier_block nb = {
 538         .notifier_call = netevent_callback
 539 };
 540
 541 static int __init addr_init(void)
 542 {
 543         addr_wq = create_singlethread_workqueue("ib_addr");
 544         if (!addr_wq)
 545                 return -ENOMEM;
 546
 547         register_netevent_notifier(&nb);
 548         rdma_addr_register_client(&self);
 549         return 0;
 550 }
 551
 552 static void __exit addr_cleanup(void)
 553 {
 554         rdma_addr_unregister_client(&self);
 555         unregister_netevent_notifier(&nb);
 556         destroy_workqueue(addr_wq);
 557 }
 558
 559 module_init(addr_init);
 560 module_exit(addr_cleanup);