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 addr->hoplimit = ip4_dst_hoplimit(&rt->dst); 256 257 *prt = rt; 258 return 0; 259 out: 260 return ret; 261 } 262 263 #if IS_ENABLED(CONFIG_IPV6) 264 static int addr6_resolve(struct sockaddr_in6 *src_in, 265 const struct sockaddr_in6 *dst_in, 266 struct rdma_dev_addr *addr, 267 struct dst_entry **pdst) 268 { 269 struct flowi6 fl6; 270 struct dst_entry *dst; 271 struct rt6_info *rt; 272 int ret; 273 274 memset(&fl6, 0, sizeof fl6); 275 fl6.daddr = dst_in->sin6_addr; 276 fl6.saddr = src_in->sin6_addr; 277 fl6.flowi6_oif = addr->bound_dev_if; 278 279 dst = ip6_route_output(addr->net, NULL, &fl6); 280 if ((ret = dst->error)) 281 goto put; 282 283 rt = (struct rt6_info *)dst; 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 there's a gateway, we're definitely in RoCE v2 (as RoCE v1 isn't 295 * routable) and we could set the network type accordingly. 296 */ 297 if (rt->rt6i_flags & RTF_GATEWAY) 298 addr->network = RDMA_NETWORK_IPV6; 299 300 addr->hoplimit = ip6_dst_hoplimit(dst); 301 302 *pdst = dst; 303 return 0; 304 put: 305 dst_release(dst); 306 return ret; 307 } 308 #else 309 static int addr6_resolve(struct sockaddr_in6 *src_in, 310 const struct sockaddr_in6 *dst_in, 311 struct rdma_dev_addr *addr, 312 struct dst_entry **pdst) 313 { 314 return -EADDRNOTAVAIL; 315 } 316 #endif 317 318 static int addr_resolve_neigh(struct dst_entry *dst, 319 const struct sockaddr *dst_in, 320 struct rdma_dev_addr *addr) 321 { 322 if (dst->dev->flags & IFF_LOOPBACK) { 323 int ret; 324 325 ret = rdma_translate_ip(dst_in, addr, NULL); 326 if (!ret) 327 memcpy(addr->dst_dev_addr, addr->src_dev_addr, 328 MAX_ADDR_LEN); 329 330 return ret; 331 } 332 333 /* If the device doesn't do ARP internally */ 334 if (!(dst->dev->flags & IFF_NOARP)) { 335 const struct sockaddr_in *dst_in4 = 336 (const struct sockaddr_in *)dst_in; 337 const struct sockaddr_in6 *dst_in6 = 338 (const struct sockaddr_in6 *)dst_in; 339 340 return dst_fetch_ha(dst, addr, 341 dst_in->sa_family == AF_INET ? 342 (const void *)&dst_in4->sin_addr.s_addr : 343 (const void *)&dst_in6->sin6_addr); 344 } 345 346 return rdma_copy_addr(addr, dst->dev, NULL); 347 } 348 349 static int addr_resolve(struct sockaddr *src_in, 350 const struct sockaddr *dst_in, 351 struct rdma_dev_addr *addr, 352 bool resolve_neigh) 353 { 354 struct net_device *ndev; 355 struct dst_entry *dst; 356 int ret; 357 358 if (src_in->sa_family == AF_INET) { 359 struct rtable *rt = NULL; 360 const struct sockaddr_in *dst_in4 = 361 (const struct sockaddr_in *)dst_in; 362 363 ret = addr4_resolve((struct sockaddr_in *)src_in, 364 dst_in4, addr, &rt); 365 if (ret) 366 return ret; 367 368 if (resolve_neigh) 369 ret = addr_resolve_neigh(&rt->dst, dst_in, addr); 370 371 ndev = rt->dst.dev; 372 dev_hold(ndev); 373 374 ip_rt_put(rt); 375 } else { 376 const struct sockaddr_in6 *dst_in6 = 377 (const struct sockaddr_in6 *)dst_in; 378 379 ret = addr6_resolve((struct sockaddr_in6 *)src_in, 380 dst_in6, addr, 381 &dst); 382 if (ret) 383 return ret; 384 385 if (resolve_neigh) 386 ret = addr_resolve_neigh(dst, dst_in, addr); 387 388 ndev = dst->dev; 389 dev_hold(ndev); 390 391 dst_release(dst); 392 } 393 394 addr->bound_dev_if = ndev->ifindex; 395 addr->net = dev_net(ndev); 396 dev_put(ndev); 397 398 return ret; 399 } 400 401 static void process_req(struct work_struct *work) 402 { 403 struct addr_req *req, *temp_req; 404 struct sockaddr *src_in, *dst_in; 405 struct list_head done_list; 406 407 INIT_LIST_HEAD(&done_list); 408 409 mutex_lock(&lock); 410 list_for_each_entry_safe(req, temp_req, &req_list, list) { 411 if (req->status == -ENODATA) { 412 src_in = (struct sockaddr *) &req->src_addr; 413 dst_in = (struct sockaddr *) &req->dst_addr; 414 req->status = addr_resolve(src_in, dst_in, req->addr, 415 true); 416 if (req->status && time_after_eq(jiffies, req->timeout)) 417 req->status = -ETIMEDOUT; 418 else if (req->status == -ENODATA) 419 continue; 420 } 421 list_move_tail(&req->list, &done_list); 422 } 423 424 if (!list_empty(&req_list)) { 425 req = list_entry(req_list.next, struct addr_req, list); 426 set_timeout(req->timeout); 427 } 428 mutex_unlock(&lock); 429 430 list_for_each_entry_safe(req, temp_req, &done_list, list) { 431 list_del(&req->list); 432 req->callback(req->status, (struct sockaddr *) &req->src_addr, 433 req->addr, req->context); 434 put_client(req->client); 435 kfree(req); 436 } 437 } 438 439 int rdma_resolve_ip(struct rdma_addr_client *client, 440 struct sockaddr *src_addr, struct sockaddr *dst_addr, 441 struct rdma_dev_addr *addr, int timeout_ms, 442 void (*callback)(int status, struct sockaddr *src_addr, 443 struct rdma_dev_addr *addr, void *context), 444 void *context) 445 { 446 struct sockaddr *src_in, *dst_in; 447 struct addr_req *req; 448 int ret = 0; 449 450 req = kzalloc(sizeof *req, GFP_KERNEL); 451 if (!req) 452 return -ENOMEM; 453 454 src_in = (struct sockaddr *) &req->src_addr; 455 dst_in = (struct sockaddr *) &req->dst_addr; 456 457 if (src_addr) { 458 if (src_addr->sa_family != dst_addr->sa_family) { 459 ret = -EINVAL; 460 goto err; 461 } 462 463 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 464 } else { 465 src_in->sa_family = dst_addr->sa_family; 466 } 467 468 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr)); 469 req->addr = addr; 470 req->callback = callback; 471 req->context = context; 472 req->client = client; 473 atomic_inc(&client->refcount); 474 475 req->status = addr_resolve(src_in, dst_in, addr, true); 476 switch (req->status) { 477 case 0: 478 req->timeout = jiffies; 479 queue_req(req); 480 break; 481 case -ENODATA: 482 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies; 483 queue_req(req); 484 break; 485 default: 486 ret = req->status; 487 atomic_dec(&client->refcount); 488 goto err; 489 } 490 return ret; 491 err: 492 kfree(req); 493 return ret; 494 } 495 EXPORT_SYMBOL(rdma_resolve_ip); 496 497 int rdma_resolve_ip_route(struct sockaddr *src_addr, 498 const struct sockaddr *dst_addr, 499 struct rdma_dev_addr *addr) 500 { 501 struct sockaddr_storage ssrc_addr = {}; 502 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr; 503 504 if (src_addr) { 505 if (src_addr->sa_family != dst_addr->sa_family) 506 return -EINVAL; 507 508 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 509 } else { 510 src_in->sa_family = dst_addr->sa_family; 511 } 512 513 return addr_resolve(src_in, dst_addr, addr, false); 514 } 515 EXPORT_SYMBOL(rdma_resolve_ip_route); 516 517 void rdma_addr_cancel(struct rdma_dev_addr *addr) 518 { 519 struct addr_req *req, *temp_req; 520 521 mutex_lock(&lock); 522 list_for_each_entry_safe(req, temp_req, &req_list, list) { 523 if (req->addr == addr) { 524 req->status = -ECANCELED; 525 req->timeout = jiffies; 526 list_move(&req->list, &req_list); 527 set_timeout(req->timeout); 528 break; 529 } 530 } 531 mutex_unlock(&lock); 532 } 533 EXPORT_SYMBOL(rdma_addr_cancel); 534 535 struct resolve_cb_context { 536 struct rdma_dev_addr *addr; 537 struct completion comp; 538 }; 539 540 static void resolve_cb(int status, struct sockaddr *src_addr, 541 struct rdma_dev_addr *addr, void *context) 542 { 543 memcpy(((struct resolve_cb_context *)context)->addr, addr, sizeof(struct 544 rdma_dev_addr)); 545 complete(&((struct resolve_cb_context *)context)->comp); 546 } 547 548 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid, 549 const union ib_gid *dgid, 550 u8 *dmac, u16 *vlan_id, int *if_index, 551 int *hoplimit) 552 { 553 int ret = 0; 554 struct rdma_dev_addr dev_addr; 555 struct resolve_cb_context ctx; 556 struct net_device *dev; 557 558 union { 559 struct sockaddr _sockaddr; 560 struct sockaddr_in _sockaddr_in; 561 struct sockaddr_in6 _sockaddr_in6; 562 } sgid_addr, dgid_addr; 563 564 565 rdma_gid2ip(&sgid_addr._sockaddr, sgid); 566 rdma_gid2ip(&dgid_addr._sockaddr, dgid); 567 568 memset(&dev_addr, 0, sizeof(dev_addr)); 569 if (if_index) 570 dev_addr.bound_dev_if = *if_index; 571 dev_addr.net = &init_net; 572 573 ctx.addr = &dev_addr; 574 init_completion(&ctx.comp); 575 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr, 576 &dev_addr, 1000, resolve_cb, &ctx); 577 if (ret) 578 return ret; 579 580 wait_for_completion(&ctx.comp); 581 582 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN); 583 dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if); 584 if (!dev) 585 return -ENODEV; 586 if (if_index) 587 *if_index = dev_addr.bound_dev_if; 588 if (vlan_id) 589 *vlan_id = rdma_vlan_dev_vlan_id(dev); 590 if (hoplimit) 591 *hoplimit = dev_addr.hoplimit; 592 dev_put(dev); 593 return ret; 594 } 595 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh); 596 597 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id) 598 { 599 int ret = 0; 600 struct rdma_dev_addr dev_addr; 601 union { 602 struct sockaddr _sockaddr; 603 struct sockaddr_in _sockaddr_in; 604 struct sockaddr_in6 _sockaddr_in6; 605 } gid_addr; 606 607 rdma_gid2ip(&gid_addr._sockaddr, sgid); 608 609 memset(&dev_addr, 0, sizeof(dev_addr)); 610 dev_addr.net = &init_net; 611 ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id); 612 if (ret) 613 return ret; 614 615 memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN); 616 return ret; 617 } 618 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid); 619 620 static int netevent_callback(struct notifier_block *self, unsigned long event, 621 void *ctx) 622 { 623 if (event == NETEVENT_NEIGH_UPDATE) { 624 struct neighbour *neigh = ctx; 625 626 if (neigh->nud_state & NUD_VALID) { 627 set_timeout(jiffies); 628 } 629 } 630 return 0; 631 } 632 633 static struct notifier_block nb = { 634 .notifier_call = netevent_callback 635 }; 636 637 static int __init addr_init(void) 638 { 639 addr_wq = create_singlethread_workqueue("ib_addr"); 640 if (!addr_wq) 641 return -ENOMEM; 642 643 register_netevent_notifier(&nb); 644 rdma_addr_register_client(&self); 645 return 0; 646 } 647 648 static void __exit addr_cleanup(void) 649 { 650 rdma_addr_unregister_client(&self); 651 unregister_netevent_notifier(&nb); 652 destroy_workqueue(addr_wq); 653 } 654 655 module_init(addr_init); 656 module_exit(addr_cleanup); 657