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(const struct sockaddr *addr, 125 struct rdma_dev_addr *dev_addr, 126 u16 *vlan_id) 127 { 128 struct net_device *dev; 129 int ret = -EADDRNOTAVAIL; 130 131 if (dev_addr->bound_dev_if) { 132 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if); 133 if (!dev) 134 return -ENODEV; 135 ret = rdma_copy_addr(dev_addr, dev, NULL); 136 dev_put(dev); 137 return ret; 138 } 139 140 switch (addr->sa_family) { 141 case AF_INET: 142 dev = ip_dev_find(dev_addr->net, 143 ((const struct sockaddr_in *)addr)->sin_addr.s_addr); 144 145 if (!dev) 146 return ret; 147 148 ret = rdma_copy_addr(dev_addr, dev, NULL); 149 if (vlan_id) 150 *vlan_id = rdma_vlan_dev_vlan_id(dev); 151 dev_put(dev); 152 break; 153 #if IS_ENABLED(CONFIG_IPV6) 154 case AF_INET6: 155 rcu_read_lock(); 156 for_each_netdev_rcu(dev_addr->net, dev) { 157 if (ipv6_chk_addr(dev_addr->net, 158 &((const struct sockaddr_in6 *)addr)->sin6_addr, 159 dev, 1)) { 160 ret = rdma_copy_addr(dev_addr, dev, NULL); 161 if (vlan_id) 162 *vlan_id = rdma_vlan_dev_vlan_id(dev); 163 break; 164 } 165 } 166 rcu_read_unlock(); 167 break; 168 #endif 169 } 170 return ret; 171 } 172 EXPORT_SYMBOL(rdma_translate_ip); 173 174 static void set_timeout(unsigned long time) 175 { 176 unsigned long delay; 177 178 delay = time - jiffies; 179 if ((long)delay < 0) 180 delay = 0; 181 182 mod_delayed_work(addr_wq, &work, delay); 183 } 184 185 static void queue_req(struct addr_req *req) 186 { 187 struct addr_req *temp_req; 188 189 mutex_lock(&lock); 190 list_for_each_entry_reverse(temp_req, &req_list, list) { 191 if (time_after_eq(req->timeout, temp_req->timeout)) 192 break; 193 } 194 195 list_add(&req->list, &temp_req->list); 196 197 if (req_list.next == &req->list) 198 set_timeout(req->timeout); 199 mutex_unlock(&lock); 200 } 201 202 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr, 203 const void *daddr) 204 { 205 struct neighbour *n; 206 int ret; 207 208 n = dst_neigh_lookup(dst, daddr); 209 210 rcu_read_lock(); 211 if (!n || !(n->nud_state & NUD_VALID)) { 212 if (n) 213 neigh_event_send(n, NULL); 214 ret = -ENODATA; 215 } else { 216 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha); 217 } 218 rcu_read_unlock(); 219 220 if (n) 221 neigh_release(n); 222 223 return ret; 224 } 225 226 static int addr4_resolve(struct sockaddr_in *src_in, 227 const struct sockaddr_in *dst_in, 228 struct rdma_dev_addr *addr, 229 struct rtable **prt) 230 { 231 __be32 src_ip = src_in->sin_addr.s_addr; 232 __be32 dst_ip = dst_in->sin_addr.s_addr; 233 struct rtable *rt; 234 struct flowi4 fl4; 235 int ret; 236 237 memset(&fl4, 0, sizeof(fl4)); 238 fl4.daddr = dst_ip; 239 fl4.saddr = src_ip; 240 fl4.flowi4_oif = addr->bound_dev_if; 241 rt = ip_route_output_key(addr->net, &fl4); 242 if (IS_ERR(rt)) { 243 ret = PTR_ERR(rt); 244 goto out; 245 } 246 src_in->sin_family = AF_INET; 247 src_in->sin_addr.s_addr = fl4.saddr; 248 249 /* If there's a gateway, we're definitely in RoCE v2 (as RoCE v1 isn't 250 * routable) and we could set the network type accordingly. 251 */ 252 if (rt->rt_uses_gateway) 253 addr->network = RDMA_NETWORK_IPV4; 254 255 *prt = rt; 256 return 0; 257 out: 258 return ret; 259 } 260 261 #if IS_ENABLED(CONFIG_IPV6) 262 static int addr6_resolve(struct sockaddr_in6 *src_in, 263 const struct sockaddr_in6 *dst_in, 264 struct rdma_dev_addr *addr, 265 struct dst_entry **pdst) 266 { 267 struct flowi6 fl6; 268 struct dst_entry *dst; 269 struct rt6_info *rt; 270 int ret; 271 272 memset(&fl6, 0, sizeof fl6); 273 fl6.daddr = dst_in->sin6_addr; 274 fl6.saddr = src_in->sin6_addr; 275 fl6.flowi6_oif = addr->bound_dev_if; 276 277 dst = ip6_route_output(addr->net, NULL, &fl6); 278 if ((ret = dst->error)) 279 goto put; 280 281 rt = (struct rt6_info *)dst; 282 if (ipv6_addr_any(&fl6.saddr)) { 283 ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev, 284 &fl6.daddr, 0, &fl6.saddr); 285 if (ret) 286 goto put; 287 288 src_in->sin6_family = AF_INET6; 289 src_in->sin6_addr = fl6.saddr; 290 } 291 292 /* If there's a gateway, we're definitely in RoCE v2 (as RoCE v1 isn't 293 * routable) and we could set the network type accordingly. 294 */ 295 if (rt->rt6i_flags & RTF_GATEWAY) 296 addr->network = RDMA_NETWORK_IPV6; 297 298 *pdst = dst; 299 return 0; 300 put: 301 dst_release(dst); 302 return ret; 303 } 304 #else 305 static int addr6_resolve(struct sockaddr_in6 *src_in, 306 const struct sockaddr_in6 *dst_in, 307 struct rdma_dev_addr *addr, 308 struct dst_entry **pdst) 309 { 310 return -EADDRNOTAVAIL; 311 } 312 #endif 313 314 static int addr_resolve_neigh(struct dst_entry *dst, 315 const struct sockaddr *dst_in, 316 struct rdma_dev_addr *addr) 317 { 318 if (dst->dev->flags & IFF_LOOPBACK) { 319 int ret; 320 321 ret = rdma_translate_ip(dst_in, addr, NULL); 322 if (!ret) 323 memcpy(addr->dst_dev_addr, addr->src_dev_addr, 324 MAX_ADDR_LEN); 325 326 return ret; 327 } 328 329 /* If the device doesn't do ARP internally */ 330 if (!(dst->dev->flags & IFF_NOARP)) { 331 const struct sockaddr_in *dst_in4 = 332 (const struct sockaddr_in *)dst_in; 333 const struct sockaddr_in6 *dst_in6 = 334 (const struct sockaddr_in6 *)dst_in; 335 336 return dst_fetch_ha(dst, addr, 337 dst_in->sa_family == AF_INET ? 338 (const void *)&dst_in4->sin_addr.s_addr : 339 (const void *)&dst_in6->sin6_addr); 340 } 341 342 return rdma_copy_addr(addr, dst->dev, NULL); 343 } 344 345 static int addr_resolve(struct sockaddr *src_in, 346 const struct sockaddr *dst_in, 347 struct rdma_dev_addr *addr, 348 bool resolve_neigh) 349 { 350 struct net_device *ndev; 351 struct dst_entry *dst; 352 int ret; 353 354 if (src_in->sa_family == AF_INET) { 355 struct rtable *rt = NULL; 356 const struct sockaddr_in *dst_in4 = 357 (const struct sockaddr_in *)dst_in; 358 359 ret = addr4_resolve((struct sockaddr_in *)src_in, 360 dst_in4, addr, &rt); 361 if (ret) 362 return ret; 363 364 if (resolve_neigh) 365 ret = addr_resolve_neigh(&rt->dst, dst_in, addr); 366 367 ndev = rt->dst.dev; 368 dev_hold(ndev); 369 370 ip_rt_put(rt); 371 } else { 372 const struct sockaddr_in6 *dst_in6 = 373 (const struct sockaddr_in6 *)dst_in; 374 375 ret = addr6_resolve((struct sockaddr_in6 *)src_in, 376 dst_in6, addr, 377 &dst); 378 if (ret) 379 return ret; 380 381 if (resolve_neigh) 382 ret = addr_resolve_neigh(dst, dst_in, addr); 383 384 ndev = dst->dev; 385 dev_hold(ndev); 386 387 dst_release(dst); 388 } 389 390 addr->bound_dev_if = ndev->ifindex; 391 addr->net = dev_net(ndev); 392 dev_put(ndev); 393 394 return ret; 395 } 396 397 static void process_req(struct work_struct *work) 398 { 399 struct addr_req *req, *temp_req; 400 struct sockaddr *src_in, *dst_in; 401 struct list_head done_list; 402 403 INIT_LIST_HEAD(&done_list); 404 405 mutex_lock(&lock); 406 list_for_each_entry_safe(req, temp_req, &req_list, list) { 407 if (req->status == -ENODATA) { 408 src_in = (struct sockaddr *) &req->src_addr; 409 dst_in = (struct sockaddr *) &req->dst_addr; 410 req->status = addr_resolve(src_in, dst_in, req->addr, 411 true); 412 if (req->status && time_after_eq(jiffies, req->timeout)) 413 req->status = -ETIMEDOUT; 414 else if (req->status == -ENODATA) 415 continue; 416 } 417 list_move_tail(&req->list, &done_list); 418 } 419 420 if (!list_empty(&req_list)) { 421 req = list_entry(req_list.next, struct addr_req, list); 422 set_timeout(req->timeout); 423 } 424 mutex_unlock(&lock); 425 426 list_for_each_entry_safe(req, temp_req, &done_list, list) { 427 list_del(&req->list); 428 req->callback(req->status, (struct sockaddr *) &req->src_addr, 429 req->addr, req->context); 430 put_client(req->client); 431 kfree(req); 432 } 433 } 434 435 int rdma_resolve_ip(struct rdma_addr_client *client, 436 struct sockaddr *src_addr, struct sockaddr *dst_addr, 437 struct rdma_dev_addr *addr, int timeout_ms, 438 void (*callback)(int status, struct sockaddr *src_addr, 439 struct rdma_dev_addr *addr, void *context), 440 void *context) 441 { 442 struct sockaddr *src_in, *dst_in; 443 struct addr_req *req; 444 int ret = 0; 445 446 req = kzalloc(sizeof *req, GFP_KERNEL); 447 if (!req) 448 return -ENOMEM; 449 450 src_in = (struct sockaddr *) &req->src_addr; 451 dst_in = (struct sockaddr *) &req->dst_addr; 452 453 if (src_addr) { 454 if (src_addr->sa_family != dst_addr->sa_family) { 455 ret = -EINVAL; 456 goto err; 457 } 458 459 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 460 } else { 461 src_in->sa_family = dst_addr->sa_family; 462 } 463 464 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr)); 465 req->addr = addr; 466 req->callback = callback; 467 req->context = context; 468 req->client = client; 469 atomic_inc(&client->refcount); 470 471 req->status = addr_resolve(src_in, dst_in, addr, true); 472 switch (req->status) { 473 case 0: 474 req->timeout = jiffies; 475 queue_req(req); 476 break; 477 case -ENODATA: 478 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies; 479 queue_req(req); 480 break; 481 default: 482 ret = req->status; 483 atomic_dec(&client->refcount); 484 goto err; 485 } 486 return ret; 487 err: 488 kfree(req); 489 return ret; 490 } 491 EXPORT_SYMBOL(rdma_resolve_ip); 492 493 int rdma_resolve_ip_route(struct sockaddr *src_addr, 494 const struct sockaddr *dst_addr, 495 struct rdma_dev_addr *addr) 496 { 497 struct sockaddr_storage ssrc_addr = {}; 498 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr; 499 500 if (src_addr->sa_family != dst_addr->sa_family) 501 return -EINVAL; 502 503 if (src_addr) 504 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 505 else 506 src_in->sa_family = dst_addr->sa_family; 507 508 return addr_resolve(src_in, dst_addr, addr, false); 509 } 510 EXPORT_SYMBOL(rdma_resolve_ip_route); 511 512 void rdma_addr_cancel(struct rdma_dev_addr *addr) 513 { 514 struct addr_req *req, *temp_req; 515 516 mutex_lock(&lock); 517 list_for_each_entry_safe(req, temp_req, &req_list, list) { 518 if (req->addr == addr) { 519 req->status = -ECANCELED; 520 req->timeout = jiffies; 521 list_move(&req->list, &req_list); 522 set_timeout(req->timeout); 523 break; 524 } 525 } 526 mutex_unlock(&lock); 527 } 528 EXPORT_SYMBOL(rdma_addr_cancel); 529 530 struct resolve_cb_context { 531 struct rdma_dev_addr *addr; 532 struct completion comp; 533 }; 534 535 static void resolve_cb(int status, struct sockaddr *src_addr, 536 struct rdma_dev_addr *addr, void *context) 537 { 538 memcpy(((struct resolve_cb_context *)context)->addr, addr, sizeof(struct 539 rdma_dev_addr)); 540 complete(&((struct resolve_cb_context *)context)->comp); 541 } 542 543 int rdma_addr_find_dmac_by_grh(const union ib_gid *sgid, const union ib_gid *dgid, 544 u8 *dmac, u16 *vlan_id, int *if_index) 545 { 546 int ret = 0; 547 struct rdma_dev_addr dev_addr; 548 struct resolve_cb_context ctx; 549 struct net_device *dev; 550 551 union { 552 struct sockaddr _sockaddr; 553 struct sockaddr_in _sockaddr_in; 554 struct sockaddr_in6 _sockaddr_in6; 555 } sgid_addr, dgid_addr; 556 557 558 rdma_gid2ip(&sgid_addr._sockaddr, sgid); 559 rdma_gid2ip(&dgid_addr._sockaddr, dgid); 560 561 memset(&dev_addr, 0, sizeof(dev_addr)); 562 if (if_index) 563 dev_addr.bound_dev_if = *if_index; 564 dev_addr.net = &init_net; 565 566 ctx.addr = &dev_addr; 567 init_completion(&ctx.comp); 568 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr, 569 &dev_addr, 1000, resolve_cb, &ctx); 570 if (ret) 571 return ret; 572 573 wait_for_completion(&ctx.comp); 574 575 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN); 576 dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if); 577 if (!dev) 578 return -ENODEV; 579 if (if_index) 580 *if_index = dev_addr.bound_dev_if; 581 if (vlan_id) 582 *vlan_id = rdma_vlan_dev_vlan_id(dev); 583 dev_put(dev); 584 return ret; 585 } 586 EXPORT_SYMBOL(rdma_addr_find_dmac_by_grh); 587 588 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id) 589 { 590 int ret = 0; 591 struct rdma_dev_addr dev_addr; 592 union { 593 struct sockaddr _sockaddr; 594 struct sockaddr_in _sockaddr_in; 595 struct sockaddr_in6 _sockaddr_in6; 596 } gid_addr; 597 598 rdma_gid2ip(&gid_addr._sockaddr, sgid); 599 600 memset(&dev_addr, 0, sizeof(dev_addr)); 601 dev_addr.net = &init_net; 602 ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id); 603 if (ret) 604 return ret; 605 606 memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN); 607 return ret; 608 } 609 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid); 610 611 static int netevent_callback(struct notifier_block *self, unsigned long event, 612 void *ctx) 613 { 614 if (event == NETEVENT_NEIGH_UPDATE) { 615 struct neighbour *neigh = ctx; 616 617 if (neigh->nud_state & NUD_VALID) { 618 set_timeout(jiffies); 619 } 620 } 621 return 0; 622 } 623 624 static struct notifier_block nb = { 625 .notifier_call = netevent_callback 626 }; 627 628 static int __init addr_init(void) 629 { 630 addr_wq = create_singlethread_workqueue("ib_addr"); 631 if (!addr_wq) 632 return -ENOMEM; 633 634 register_netevent_notifier(&nb); 635 rdma_addr_register_client(&self); 636 return 0; 637 } 638 639 static void __exit addr_cleanup(void) 640 { 641 rdma_addr_unregister_client(&self); 642 unregister_netevent_notifier(&nb); 643 destroy_workqueue(addr_wq); 644 } 645 646 module_init(addr_init); 647 module_exit(addr_cleanup); 648