1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0 3 * 4 * Copyright (c) 2005 Voltaire Inc. All rights reserved. 5 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved. 6 * Copyright (c) 1999-2019, Mellanox Technologies, Inc. All rights reserved. 7 * Copyright (c) 2005 Intel Corporation. All rights reserved. 8 * 9 * This software is available to you under a choice of one of two 10 * licenses. You may choose to be licensed under the terms of the GNU 11 * General Public License (GPL) Version 2, available from the file 12 * COPYING in the main directory of this source tree, or the 13 * OpenIB.org BSD license below: 14 * 15 * Redistribution and use in source and binary forms, with or 16 * without modification, are permitted provided that the following 17 * conditions are met: 18 * 19 * - Redistributions of source code must retain the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer. 22 * 23 * - Redistributions in binary form must reproduce the above 24 * copyright notice, this list of conditions and the following 25 * disclaimer in the documentation and/or other materials 26 * provided with the distribution. 27 * 28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 35 * SOFTWARE. 36 */ 37 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 40 41 #include <linux/mutex.h> 42 #include <linux/inetdevice.h> 43 #include <linux/slab.h> 44 #include <linux/workqueue.h> 45 #include <linux/module.h> 46 #include <net/route.h> 47 #include <net/route/nhop.h> 48 #include <net/netevent.h> 49 #include <rdma/ib_addr.h> 50 #include <rdma/ib.h> 51 52 #include <netinet/in_fib.h> 53 #include <netinet/if_ether.h> 54 #include <netinet/ip_var.h> 55 #include <netinet6/scope6_var.h> 56 #include <netinet6/in6_pcb.h> 57 #include <netinet6/in6_fib.h> 58 59 #include "core_priv.h" 60 61 struct addr_req { 62 struct list_head list; 63 struct sockaddr_storage src_addr; 64 struct sockaddr_storage dst_addr; 65 struct rdma_dev_addr *addr; 66 struct rdma_addr_client *client; 67 void *context; 68 void (*callback)(int status, struct sockaddr *src_addr, 69 struct rdma_dev_addr *addr, void *context); 70 int timeout; 71 int status; 72 }; 73 74 static void process_req(struct work_struct *work); 75 76 static DEFINE_MUTEX(lock); 77 static LIST_HEAD(req_list); 78 static DECLARE_DELAYED_WORK(work, process_req); 79 static struct workqueue_struct *addr_wq; 80 81 int rdma_addr_size(struct sockaddr *addr) 82 { 83 switch (addr->sa_family) { 84 case AF_INET: 85 return sizeof(struct sockaddr_in); 86 case AF_INET6: 87 return sizeof(struct sockaddr_in6); 88 case AF_IB: 89 return sizeof(struct sockaddr_ib); 90 default: 91 return 0; 92 } 93 } 94 EXPORT_SYMBOL(rdma_addr_size); 95 96 int rdma_addr_size_in6(struct sockaddr_in6 *addr) 97 { 98 int ret = rdma_addr_size((struct sockaddr *) addr); 99 100 return ret <= sizeof(*addr) ? ret : 0; 101 } 102 EXPORT_SYMBOL(rdma_addr_size_in6); 103 104 int rdma_addr_size_kss(struct sockaddr_storage *addr) 105 { 106 int ret = rdma_addr_size((struct sockaddr *) addr); 107 108 return ret <= sizeof(*addr) ? ret : 0; 109 } 110 EXPORT_SYMBOL(rdma_addr_size_kss); 111 112 static struct rdma_addr_client self; 113 114 void rdma_addr_register_client(struct rdma_addr_client *client) 115 { 116 atomic_set(&client->refcount, 1); 117 init_completion(&client->comp); 118 } 119 EXPORT_SYMBOL(rdma_addr_register_client); 120 121 static inline void put_client(struct rdma_addr_client *client) 122 { 123 if (atomic_dec_and_test(&client->refcount)) 124 complete(&client->comp); 125 } 126 127 void rdma_addr_unregister_client(struct rdma_addr_client *client) 128 { 129 put_client(client); 130 wait_for_completion(&client->comp); 131 } 132 EXPORT_SYMBOL(rdma_addr_unregister_client); 133 134 static inline void 135 rdma_copy_addr_sub(u8 *dst, const u8 *src, unsigned min, unsigned max) 136 { 137 if (min > max) 138 min = max; 139 memcpy(dst, src, min); 140 memset(dst + min, 0, max - min); 141 } 142 143 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev, 144 const unsigned char *dst_dev_addr) 145 { 146 /* check for loopback device */ 147 if (dev->if_flags & IFF_LOOPBACK) { 148 dev_addr->dev_type = ARPHRD_ETHER; 149 memset(dev_addr->src_dev_addr, 0, MAX_ADDR_LEN); 150 memset(dev_addr->broadcast, 0, MAX_ADDR_LEN); 151 memset(dev_addr->dst_dev_addr, 0, MAX_ADDR_LEN); 152 dev_addr->bound_dev_if = dev->if_index; 153 return (0); 154 } else if (dev->if_type == IFT_INFINIBAND) 155 dev_addr->dev_type = ARPHRD_INFINIBAND; 156 else if (dev->if_type == IFT_ETHER) 157 dev_addr->dev_type = ARPHRD_ETHER; 158 else 159 dev_addr->dev_type = 0; 160 rdma_copy_addr_sub(dev_addr->src_dev_addr, IF_LLADDR(dev), 161 dev->if_addrlen, MAX_ADDR_LEN); 162 rdma_copy_addr_sub(dev_addr->broadcast, dev->if_broadcastaddr, 163 dev->if_addrlen, MAX_ADDR_LEN); 164 if (dst_dev_addr != NULL) { 165 rdma_copy_addr_sub(dev_addr->dst_dev_addr, dst_dev_addr, 166 dev->if_addrlen, MAX_ADDR_LEN); 167 } 168 dev_addr->bound_dev_if = dev->if_index; 169 return 0; 170 } 171 EXPORT_SYMBOL(rdma_copy_addr); 172 173 int rdma_translate_ip(const struct sockaddr *addr, 174 struct rdma_dev_addr *dev_addr) 175 { 176 struct net_device *dev; 177 int ret; 178 179 if (dev_addr->bound_dev_if) { 180 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if); 181 } else switch (addr->sa_family) { 182 #ifdef INET 183 case AF_INET: 184 dev = ip_dev_find(dev_addr->net, 185 ((const struct sockaddr_in *)addr)->sin_addr.s_addr); 186 break; 187 #endif 188 #ifdef INET6 189 case AF_INET6: 190 dev = ip6_dev_find(dev_addr->net, 191 ((const struct sockaddr_in6 *)addr)->sin6_addr, 0); 192 break; 193 #endif 194 default: 195 dev = NULL; 196 break; 197 } 198 199 if (dev != NULL) { 200 /* disallow connections through 127.0.0.1 itself */ 201 if (dev->if_flags & IFF_LOOPBACK) 202 ret = -EINVAL; 203 else 204 ret = rdma_copy_addr(dev_addr, dev, NULL); 205 dev_put(dev); 206 } else { 207 ret = -ENODEV; 208 } 209 return ret; 210 } 211 EXPORT_SYMBOL(rdma_translate_ip); 212 213 static void set_timeout(int time) 214 { 215 int delay; /* under FreeBSD ticks are 32-bit */ 216 217 delay = time - jiffies; 218 if (delay <= 0) 219 delay = 1; 220 else if (delay > hz) 221 delay = hz; 222 223 mod_delayed_work(addr_wq, &work, delay); 224 } 225 226 static void queue_req(struct addr_req *req) 227 { 228 struct addr_req *temp_req; 229 230 mutex_lock(&lock); 231 list_for_each_entry_reverse(temp_req, &req_list, list) { 232 if (time_after_eq(req->timeout, temp_req->timeout)) 233 break; 234 } 235 236 list_add(&req->list, &temp_req->list); 237 238 if (req_list.next == &req->list) 239 set_timeout(req->timeout); 240 mutex_unlock(&lock); 241 } 242 243 #if defined(INET) || defined(INET6) 244 static int addr_resolve_multi(u8 *edst, struct ifnet *ifp, struct sockaddr *dst_in) 245 { 246 struct sockaddr *llsa; 247 struct sockaddr_dl sdl; 248 int error; 249 250 sdl.sdl_len = sizeof(sdl); 251 llsa = (struct sockaddr *)&sdl; 252 253 if (ifp->if_resolvemulti == NULL) { 254 error = EOPNOTSUPP; 255 } else { 256 error = ifp->if_resolvemulti(ifp, &llsa, dst_in); 257 if (error == 0) { 258 rdma_copy_addr_sub(edst, LLADDR((struct sockaddr_dl *)llsa), 259 ifp->if_addrlen, MAX_ADDR_LEN); 260 } 261 } 262 return (error); 263 } 264 #endif 265 266 #ifdef INET 267 static int addr4_resolve(struct sockaddr_in *src_in, 268 const struct sockaddr_in *dst_in, 269 struct rdma_dev_addr *addr, 270 u8 *edst, 271 struct ifnet **ifpp) 272 { 273 enum { 274 ADDR_VALID = 0, 275 ADDR_SRC_ANY = 1, 276 ADDR_DST_ANY = 2, 277 }; 278 struct sockaddr_in dst_tmp = *dst_in; 279 in_port_t src_port; 280 struct sockaddr *saddr = NULL; 281 struct nhop_object *nh; 282 struct ifnet *ifp; 283 int error; 284 int type; 285 286 NET_EPOCH_ASSERT(); 287 288 /* set VNET, if any */ 289 CURVNET_SET(addr->net); 290 291 /* set default TTL limit */ 292 addr->hoplimit = V_ip_defttl; 293 294 type = ADDR_VALID; 295 if (src_in->sin_addr.s_addr == INADDR_ANY) 296 type |= ADDR_SRC_ANY; 297 if (dst_tmp.sin_addr.s_addr == INADDR_ANY) 298 type |= ADDR_DST_ANY; 299 300 /* 301 * Make sure the socket address length field is set. 302 */ 303 dst_tmp.sin_len = sizeof(dst_tmp); 304 305 /* Step 1 - lookup destination route if any */ 306 switch (type) { 307 case ADDR_VALID: 308 case ADDR_SRC_ANY: 309 /* regular destination route lookup */ 310 nh = fib4_lookup(RT_DEFAULT_FIB, dst_tmp.sin_addr,0,NHR_NONE,0); 311 if (nh == NULL) { 312 error = EHOSTUNREACH; 313 goto done; 314 } 315 break; 316 default: 317 error = ENETUNREACH; 318 goto done; 319 } 320 321 /* Step 2 - find outgoing network interface */ 322 switch (type) { 323 case ADDR_VALID: 324 /* get source interface */ 325 if (addr->bound_dev_if != 0) { 326 ifp = dev_get_by_index(addr->net, addr->bound_dev_if); 327 } else { 328 ifp = ip_dev_find(addr->net, src_in->sin_addr.s_addr); 329 } 330 331 /* check source interface */ 332 if (ifp == NULL) { 333 error = ENETUNREACH; 334 goto done; 335 } else if (ifp->if_flags & IFF_LOOPBACK) { 336 /* 337 * Source address cannot be a loopback device. 338 */ 339 error = EHOSTUNREACH; 340 goto error_put_ifp; 341 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) { 342 if (memcmp(&src_in->sin_addr, &dst_in->sin_addr, 343 sizeof(src_in->sin_addr))) { 344 /* 345 * Destination is loopback, but source 346 * and destination address is not the 347 * same. 348 */ 349 error = EHOSTUNREACH; 350 goto error_put_ifp; 351 } 352 /* get destination network interface from route */ 353 dev_put(ifp); 354 ifp = nh->nh_ifp; 355 dev_hold(ifp); 356 } else if (ifp != nh->nh_ifp) { 357 /* 358 * Source and destination interfaces are 359 * different. 360 */ 361 error = ENETUNREACH; 362 goto error_put_ifp; 363 } 364 break; 365 case ADDR_SRC_ANY: 366 /* check for loopback device */ 367 if (nh->nh_ifp->if_flags & IFF_LOOPBACK) 368 saddr = (struct sockaddr *)&dst_tmp; 369 else 370 saddr = nh->nh_ifa->ifa_addr; 371 372 /* get destination network interface from route */ 373 ifp = nh->nh_ifp; 374 dev_hold(ifp); 375 break; 376 default: 377 break; 378 } 379 380 /* 381 * Step 3 - resolve destination MAC address 382 */ 383 if (dst_tmp.sin_addr.s_addr == INADDR_BROADCAST) { 384 rdma_copy_addr_sub(edst, ifp->if_broadcastaddr, 385 ifp->if_addrlen, MAX_ADDR_LEN); 386 error = 0; 387 } else if (IN_MULTICAST(ntohl(dst_tmp.sin_addr.s_addr))) { 388 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0; 389 error = addr_resolve_multi(edst, ifp, (struct sockaddr *)&dst_tmp); 390 if (error != 0) 391 goto error_put_ifp; 392 else if (is_gw) 393 addr->network = RDMA_NETWORK_IPV4; 394 } else if (ifp->if_flags & IFF_LOOPBACK) { 395 memset(edst, 0, MAX_ADDR_LEN); 396 error = 0; 397 } else { 398 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0; 399 memset(edst, 0, MAX_ADDR_LEN); 400 error = arpresolve(ifp, is_gw, NULL, is_gw ? 401 &nh->gw_sa : (const struct sockaddr *)&dst_tmp, 402 edst, NULL, NULL); 403 if (error != 0) 404 goto error_put_ifp; 405 else if (is_gw) 406 addr->network = RDMA_NETWORK_IPV4; 407 } 408 409 /* 410 * Step 4 - update source address, if any 411 */ 412 if (saddr != NULL) { 413 src_port = src_in->sin_port; 414 memcpy(src_in, saddr, rdma_addr_size(saddr)); 415 src_in->sin_port = src_port; /* preserve port number */ 416 } 417 418 *ifpp = ifp; 419 420 goto done; 421 422 error_put_ifp: 423 dev_put(ifp); 424 done: 425 CURVNET_RESTORE(); 426 427 if (error == EWOULDBLOCK || error == EAGAIN) 428 error = ENODATA; 429 return (-error); 430 } 431 #else 432 static int addr4_resolve(struct sockaddr_in *src_in, 433 const struct sockaddr_in *dst_in, 434 struct rdma_dev_addr *addr, 435 u8 *edst, 436 struct ifnet **ifpp) 437 { 438 return -EADDRNOTAVAIL; 439 } 440 #endif 441 442 #ifdef INET6 443 static int addr6_resolve(struct sockaddr_in6 *src_in, 444 const struct sockaddr_in6 *dst_in, 445 struct rdma_dev_addr *addr, 446 u8 *edst, 447 struct ifnet **ifpp) 448 { 449 enum { 450 ADDR_VALID = 0, 451 ADDR_SRC_ANY = 1, 452 ADDR_DST_ANY = 2, 453 }; 454 struct sockaddr_in6 dst_tmp = *dst_in; 455 in_port_t src_port; 456 struct sockaddr *saddr = NULL; 457 struct nhop_object *nh; 458 struct ifnet *ifp; 459 int error; 460 int type; 461 462 NET_EPOCH_ASSERT(); 463 464 /* set VNET, if any */ 465 CURVNET_SET(addr->net); 466 467 /* set default TTL limit */ 468 addr->hoplimit = V_ip_defttl; 469 470 type = ADDR_VALID; 471 if (ipv6_addr_any(&src_in->sin6_addr)) 472 type |= ADDR_SRC_ANY; 473 if (ipv6_addr_any(&dst_tmp.sin6_addr)) 474 type |= ADDR_DST_ANY; 475 476 /* 477 * Make sure the socket address length field is set. 478 */ 479 dst_tmp.sin6_len = sizeof(dst_tmp); 480 481 /* 482 * Make sure the scope ID gets embedded, else nd6_resolve() will 483 * not find the record. 484 */ 485 dst_tmp.sin6_scope_id = addr->bound_dev_if; 486 sa6_embedscope(&dst_tmp, 0); 487 488 /* Step 1 - lookup destination route if any */ 489 switch (type) { 490 case ADDR_VALID: 491 /* sanity check for IPv4 addresses */ 492 if (ipv6_addr_v4mapped(&src_in->sin6_addr) != 493 ipv6_addr_v4mapped(&dst_tmp.sin6_addr)) { 494 error = EAFNOSUPPORT; 495 goto done; 496 } 497 /* FALLTHROUGH */ 498 case ADDR_SRC_ANY: 499 /* regular destination route lookup */ 500 nh = fib6_lookup(RT_DEFAULT_FIB, &dst_in->sin6_addr, 501 addr->bound_dev_if, NHR_NONE, 0); 502 if (nh == NULL) { 503 error = EHOSTUNREACH; 504 goto done; 505 } 506 break; 507 default: 508 error = ENETUNREACH; 509 goto done; 510 } 511 512 /* Step 2 - find outgoing network interface */ 513 switch (type) { 514 case ADDR_VALID: 515 /* get source interface */ 516 if (addr->bound_dev_if != 0) { 517 ifp = dev_get_by_index(addr->net, addr->bound_dev_if); 518 } else { 519 ifp = ip6_dev_find(addr->net, src_in->sin6_addr, 0); 520 } 521 522 /* check source interface */ 523 if (ifp == NULL) { 524 error = ENETUNREACH; 525 goto done; 526 } else if (ifp->if_flags & IFF_LOOPBACK) { 527 /* 528 * Source address cannot be a loopback device. 529 */ 530 error = EHOSTUNREACH; 531 goto error_put_ifp; 532 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) { 533 if (memcmp(&src_in->sin6_addr, &dst_in->sin6_addr, 534 sizeof(src_in->sin6_addr))) { 535 /* 536 * Destination is loopback, but source 537 * and destination address is not the 538 * same. 539 */ 540 error = EHOSTUNREACH; 541 goto error_put_ifp; 542 } 543 /* get destination network interface from route */ 544 dev_put(ifp); 545 ifp = nh->nh_ifp; 546 dev_hold(ifp); 547 } else if (ifp != nh->nh_ifp) { 548 /* 549 * Source and destination interfaces are 550 * different. 551 */ 552 error = ENETUNREACH; 553 goto error_put_ifp; 554 } 555 break; 556 case ADDR_SRC_ANY: 557 /* check for loopback device */ 558 if (nh->nh_ifp->if_flags & IFF_LOOPBACK) 559 saddr = (struct sockaddr *)&dst_tmp; 560 else 561 saddr = nh->nh_ifa->ifa_addr; 562 563 /* get destination network interface from route */ 564 ifp = nh->nh_ifp; 565 dev_hold(ifp); 566 break; 567 default: 568 break; 569 } 570 571 /* 572 * Step 3 - resolve destination MAC address 573 */ 574 if (IN6_IS_ADDR_MULTICAST(&dst_tmp.sin6_addr)) { 575 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0; 576 error = addr_resolve_multi(edst, ifp, 577 (struct sockaddr *)&dst_tmp); 578 if (error != 0) 579 goto error_put_ifp; 580 else if (is_gw) 581 addr->network = RDMA_NETWORK_IPV6; 582 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) { 583 memset(edst, 0, MAX_ADDR_LEN); 584 error = 0; 585 } else { 586 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0; 587 memset(edst, 0, MAX_ADDR_LEN); 588 error = nd6_resolve(ifp, is_gw, NULL, is_gw ? 589 &nh->gw_sa : (const struct sockaddr *)&dst_tmp, 590 edst, NULL, NULL); 591 if (error != 0) 592 goto error_put_ifp; 593 else if (is_gw) 594 addr->network = RDMA_NETWORK_IPV6; 595 } 596 597 /* 598 * Step 4 - update source address, if any 599 */ 600 if (saddr != NULL) { 601 src_port = src_in->sin6_port; 602 memcpy(src_in, saddr, rdma_addr_size(saddr)); 603 src_in->sin6_port = src_port; /* preserve port number */ 604 } 605 606 *ifpp = ifp; 607 608 goto done; 609 610 error_put_ifp: 611 dev_put(ifp); 612 done: 613 CURVNET_RESTORE(); 614 615 if (error == EWOULDBLOCK || error == EAGAIN) 616 error = ENODATA; 617 return (-error); 618 } 619 #else 620 static int addr6_resolve(struct sockaddr_in6 *src_in, 621 const struct sockaddr_in6 *dst_in, 622 struct rdma_dev_addr *addr, 623 u8 *edst, 624 struct ifnet **ifpp) 625 { 626 return -EADDRNOTAVAIL; 627 } 628 #endif 629 630 static int addr_resolve_neigh(struct ifnet *dev, 631 const struct sockaddr *dst_in, 632 u8 *edst, 633 struct rdma_dev_addr *addr) 634 { 635 if (dev->if_flags & IFF_LOOPBACK) { 636 int ret; 637 638 /* 639 * Binding to a loopback device is not allowed. Make 640 * sure the destination device address is global by 641 * clearing the bound device interface: 642 */ 643 if (addr->bound_dev_if == dev->if_index) 644 addr->bound_dev_if = 0; 645 646 ret = rdma_translate_ip(dst_in, addr); 647 if (ret == 0) { 648 memcpy(addr->dst_dev_addr, addr->src_dev_addr, 649 MAX_ADDR_LEN); 650 } 651 return ret; 652 } 653 654 /* If the device doesn't do ARP internally */ 655 if (!(dev->if_flags & IFF_NOARP)) 656 return rdma_copy_addr(addr, dev, edst); 657 658 return rdma_copy_addr(addr, dev, NULL); 659 } 660 661 static int addr_resolve(struct sockaddr *src_in, 662 const struct sockaddr *dst_in, 663 struct rdma_dev_addr *addr) 664 { 665 struct epoch_tracker et; 666 struct net_device *ndev = NULL; 667 u8 edst[MAX_ADDR_LEN]; 668 int ret; 669 670 if (dst_in->sa_family != src_in->sa_family) 671 return -EINVAL; 672 673 NET_EPOCH_ENTER(et); 674 switch (src_in->sa_family) { 675 case AF_INET: 676 ret = addr4_resolve((struct sockaddr_in *)src_in, 677 (const struct sockaddr_in *)dst_in, 678 addr, edst, &ndev); 679 break; 680 case AF_INET6: 681 ret = addr6_resolve((struct sockaddr_in6 *)src_in, 682 (const struct sockaddr_in6 *)dst_in, addr, 683 edst, &ndev); 684 break; 685 default: 686 ret = -EADDRNOTAVAIL; 687 break; 688 } 689 NET_EPOCH_EXIT(et); 690 691 /* check for error */ 692 if (ret != 0) 693 return ret; 694 695 /* store MAC addresses and check for loopback */ 696 ret = addr_resolve_neigh(ndev, dst_in, edst, addr); 697 698 /* set belonging VNET, if any */ 699 addr->net = dev_net(ndev); 700 dev_put(ndev); 701 702 return ret; 703 } 704 705 static void process_req(struct work_struct *work) 706 { 707 struct addr_req *req, *temp_req; 708 struct sockaddr *src_in, *dst_in; 709 struct list_head done_list; 710 711 INIT_LIST_HEAD(&done_list); 712 713 mutex_lock(&lock); 714 list_for_each_entry_safe(req, temp_req, &req_list, list) { 715 if (req->status == -ENODATA) { 716 src_in = (struct sockaddr *) &req->src_addr; 717 dst_in = (struct sockaddr *) &req->dst_addr; 718 req->status = addr_resolve(src_in, dst_in, req->addr); 719 if (req->status && time_after_eq(jiffies, req->timeout)) 720 req->status = -ETIMEDOUT; 721 else if (req->status == -ENODATA) 722 continue; 723 } 724 list_move_tail(&req->list, &done_list); 725 } 726 727 if (!list_empty(&req_list)) { 728 req = list_entry(req_list.next, struct addr_req, list); 729 set_timeout(req->timeout); 730 } 731 mutex_unlock(&lock); 732 733 list_for_each_entry_safe(req, temp_req, &done_list, list) { 734 list_del(&req->list); 735 req->callback(req->status, (struct sockaddr *) &req->src_addr, 736 req->addr, req->context); 737 put_client(req->client); 738 kfree(req); 739 } 740 } 741 742 int rdma_resolve_ip(struct rdma_addr_client *client, 743 struct sockaddr *src_addr, struct sockaddr *dst_addr, 744 struct rdma_dev_addr *addr, int timeout_ms, 745 void (*callback)(int status, struct sockaddr *src_addr, 746 struct rdma_dev_addr *addr, void *context), 747 void *context) 748 { 749 struct sockaddr *src_in, *dst_in; 750 struct addr_req *req; 751 int ret = 0; 752 753 req = kzalloc(sizeof *req, GFP_KERNEL); 754 if (!req) 755 return -ENOMEM; 756 757 src_in = (struct sockaddr *) &req->src_addr; 758 dst_in = (struct sockaddr *) &req->dst_addr; 759 760 if (src_addr) { 761 if (src_addr->sa_family != dst_addr->sa_family) { 762 ret = -EINVAL; 763 goto err; 764 } 765 766 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 767 } else { 768 src_in->sa_family = dst_addr->sa_family; 769 } 770 771 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr)); 772 req->addr = addr; 773 req->callback = callback; 774 req->context = context; 775 req->client = client; 776 atomic_inc(&client->refcount); 777 778 req->status = addr_resolve(src_in, dst_in, addr); 779 switch (req->status) { 780 case 0: 781 req->timeout = jiffies; 782 queue_req(req); 783 break; 784 case -ENODATA: 785 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies; 786 queue_req(req); 787 break; 788 default: 789 ret = req->status; 790 atomic_dec(&client->refcount); 791 goto err; 792 } 793 return ret; 794 err: 795 kfree(req); 796 return ret; 797 } 798 EXPORT_SYMBOL(rdma_resolve_ip); 799 800 int rdma_resolve_ip_route(struct sockaddr *src_addr, 801 const struct sockaddr *dst_addr, 802 struct rdma_dev_addr *addr) 803 { 804 struct sockaddr_storage ssrc_addr = {}; 805 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr; 806 807 if (src_addr) { 808 if (src_addr->sa_family != dst_addr->sa_family) 809 return -EINVAL; 810 811 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 812 } else { 813 src_in->sa_family = dst_addr->sa_family; 814 } 815 816 return addr_resolve(src_in, dst_addr, addr); 817 } 818 EXPORT_SYMBOL(rdma_resolve_ip_route); 819 820 void rdma_addr_cancel(struct rdma_dev_addr *addr) 821 { 822 struct addr_req *req, *temp_req; 823 824 mutex_lock(&lock); 825 list_for_each_entry_safe(req, temp_req, &req_list, list) { 826 if (req->addr == addr) { 827 req->status = -ECANCELED; 828 req->timeout = jiffies; 829 list_move(&req->list, &req_list); 830 set_timeout(req->timeout); 831 break; 832 } 833 } 834 mutex_unlock(&lock); 835 } 836 EXPORT_SYMBOL(rdma_addr_cancel); 837 838 struct resolve_cb_context { 839 struct rdma_dev_addr *addr; 840 struct completion comp; 841 int status; 842 }; 843 844 static void resolve_cb(int status, struct sockaddr *src_addr, 845 struct rdma_dev_addr *addr, void *context) 846 { 847 if (!status) 848 memcpy(((struct resolve_cb_context *)context)->addr, 849 addr, sizeof(struct rdma_dev_addr)); 850 ((struct resolve_cb_context *)context)->status = status; 851 complete(&((struct resolve_cb_context *)context)->comp); 852 } 853 854 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid, 855 const union ib_gid *dgid, 856 u8 *dmac, struct net_device *dev, 857 int *hoplimit) 858 { 859 int ret = 0; 860 struct rdma_dev_addr dev_addr; 861 struct resolve_cb_context ctx; 862 863 union { 864 struct sockaddr _sockaddr; 865 struct sockaddr_in _sockaddr_in; 866 struct sockaddr_in6 _sockaddr_in6; 867 } sgid_addr, dgid_addr; 868 869 rdma_gid2ip(&sgid_addr._sockaddr, sgid); 870 rdma_gid2ip(&dgid_addr._sockaddr, dgid); 871 872 memset(&dev_addr, 0, sizeof(dev_addr)); 873 874 dev_addr.bound_dev_if = dev->if_index; 875 dev_addr.net = dev_net(dev); 876 877 ctx.addr = &dev_addr; 878 init_completion(&ctx.comp); 879 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr, 880 &dev_addr, 1000, resolve_cb, &ctx); 881 if (ret) 882 return ret; 883 884 wait_for_completion(&ctx.comp); 885 886 ret = ctx.status; 887 if (ret) 888 return ret; 889 890 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN); 891 if (hoplimit) 892 *hoplimit = dev_addr.hoplimit; 893 return ret; 894 } 895 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh); 896 897 int addr_init(void) 898 { 899 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0); 900 if (!addr_wq) 901 return -ENOMEM; 902 903 rdma_addr_register_client(&self); 904 905 return 0; 906 } 907 908 void addr_cleanup(void) 909 { 910 rdma_addr_unregister_client(&self); 911 destroy_workqueue(addr_wq); 912 } 913