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/slab.h> 43 #include <linux/workqueue.h> 44 #include <linux/module.h> 45 #include <net/if_llatbl.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 ifnet *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 || dev->if_type == IFT_L2VLAN) 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 ifnet *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_ifp_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_ifp_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_ifp_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 #ifdef INET6 401 if (is_gw && nh->gw_sa.sa_family == AF_INET6) 402 error = nd6_resolve(ifp, LLE_SF(AF_INET, is_gw), NULL, 403 &nh->gw_sa, edst, NULL, NULL); 404 else 405 #endif 406 error = arpresolve(ifp, is_gw, NULL, is_gw ? 407 &nh->gw_sa : (const struct sockaddr *)&dst_tmp, 408 edst, NULL, NULL); 409 410 if (error != 0) 411 goto error_put_ifp; 412 else if (is_gw) 413 addr->network = RDMA_NETWORK_IPV4; 414 } 415 416 /* 417 * Step 4 - update source address, if any 418 */ 419 if (saddr != NULL) { 420 src_port = src_in->sin_port; 421 memcpy(src_in, saddr, rdma_addr_size(saddr)); 422 src_in->sin_port = src_port; /* preserve port number */ 423 } 424 425 *ifpp = ifp; 426 427 goto done; 428 429 error_put_ifp: 430 dev_put(ifp); 431 done: 432 CURVNET_RESTORE(); 433 434 if (error == EWOULDBLOCK || error == EAGAIN) 435 error = ENODATA; 436 return (-error); 437 } 438 #else 439 static int addr4_resolve(struct sockaddr_in *src_in, 440 const struct sockaddr_in *dst_in, 441 struct rdma_dev_addr *addr, 442 u8 *edst, 443 struct ifnet **ifpp) 444 { 445 return -EADDRNOTAVAIL; 446 } 447 #endif 448 449 #ifdef INET6 450 static int addr6_resolve(struct sockaddr_in6 *src_in, 451 const struct sockaddr_in6 *dst_in, 452 struct rdma_dev_addr *addr, 453 u8 *edst, 454 struct ifnet **ifpp) 455 { 456 enum { 457 ADDR_VALID = 0, 458 ADDR_SRC_ANY = 1, 459 ADDR_DST_ANY = 2, 460 }; 461 struct sockaddr_in6 dst_tmp = *dst_in; 462 in_port_t src_port; 463 struct sockaddr *saddr = NULL; 464 struct nhop_object *nh; 465 struct ifnet *ifp; 466 int error; 467 int type; 468 469 NET_EPOCH_ASSERT(); 470 471 /* set VNET, if any */ 472 CURVNET_SET(addr->net); 473 474 /* set default TTL limit */ 475 addr->hoplimit = V_ip_defttl; 476 477 type = ADDR_VALID; 478 if (ipv6_addr_any(&src_in->sin6_addr)) 479 type |= ADDR_SRC_ANY; 480 if (ipv6_addr_any(&dst_tmp.sin6_addr)) 481 type |= ADDR_DST_ANY; 482 483 /* 484 * Make sure the socket address length field is set. 485 */ 486 dst_tmp.sin6_len = sizeof(dst_tmp); 487 488 /* 489 * Make sure the scope ID gets embedded, else nd6_resolve() will 490 * not find the record. 491 */ 492 dst_tmp.sin6_scope_id = addr->bound_dev_if; 493 sa6_embedscope(&dst_tmp, 0); 494 495 /* Step 1 - lookup destination route if any */ 496 switch (type) { 497 case ADDR_VALID: 498 /* sanity check for IPv4 addresses */ 499 if (ipv6_addr_v4mapped(&src_in->sin6_addr) != 500 ipv6_addr_v4mapped(&dst_tmp.sin6_addr)) { 501 error = EAFNOSUPPORT; 502 goto done; 503 } 504 /* FALLTHROUGH */ 505 case ADDR_SRC_ANY: 506 /* regular destination route lookup */ 507 nh = fib6_lookup(RT_DEFAULT_FIB, &dst_in->sin6_addr, 508 addr->bound_dev_if, NHR_NONE, 0); 509 if (nh == NULL) { 510 error = EHOSTUNREACH; 511 goto done; 512 } 513 break; 514 default: 515 error = ENETUNREACH; 516 goto done; 517 } 518 519 /* Step 2 - find outgoing network interface */ 520 switch (type) { 521 case ADDR_VALID: 522 /* get source interface */ 523 if (addr->bound_dev_if != 0) { 524 ifp = dev_get_by_index(addr->net, addr->bound_dev_if); 525 } else { 526 ifp = ip6_ifp_find(addr->net, src_in->sin6_addr, 0); 527 } 528 529 /* check source interface */ 530 if (ifp == NULL) { 531 error = ENETUNREACH; 532 goto done; 533 } else if (ifp->if_flags & IFF_LOOPBACK) { 534 /* 535 * Source address cannot be a loopback device. 536 */ 537 error = EHOSTUNREACH; 538 goto error_put_ifp; 539 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) { 540 if (memcmp(&src_in->sin6_addr, &dst_in->sin6_addr, 541 sizeof(src_in->sin6_addr))) { 542 /* 543 * Destination is loopback, but source 544 * and destination address is not the 545 * same. 546 */ 547 error = EHOSTUNREACH; 548 goto error_put_ifp; 549 } 550 /* get destination network interface from route */ 551 dev_put(ifp); 552 ifp = nh->nh_ifp; 553 dev_hold(ifp); 554 } else if (ifp != nh->nh_ifp) { 555 /* 556 * Source and destination interfaces are 557 * different. 558 */ 559 error = ENETUNREACH; 560 goto error_put_ifp; 561 } 562 break; 563 case ADDR_SRC_ANY: 564 /* check for loopback device */ 565 if (nh->nh_ifp->if_flags & IFF_LOOPBACK) 566 saddr = (struct sockaddr *)&dst_tmp; 567 else 568 saddr = nh->nh_ifa->ifa_addr; 569 570 /* get destination network interface from route */ 571 ifp = nh->nh_ifp; 572 dev_hold(ifp); 573 break; 574 default: 575 break; 576 } 577 578 /* 579 * Step 3 - resolve destination MAC address 580 */ 581 if (IN6_IS_ADDR_MULTICAST(&dst_tmp.sin6_addr)) { 582 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0; 583 error = addr_resolve_multi(edst, ifp, 584 (struct sockaddr *)&dst_tmp); 585 if (error != 0) 586 goto error_put_ifp; 587 else if (is_gw) 588 addr->network = RDMA_NETWORK_IPV6; 589 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) { 590 memset(edst, 0, MAX_ADDR_LEN); 591 error = 0; 592 } else { 593 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0; 594 memset(edst, 0, MAX_ADDR_LEN); 595 error = nd6_resolve(ifp, LLE_SF(AF_INET6, is_gw), NULL, 596 is_gw ? &nh->gw_sa : (const struct sockaddr *)&dst_tmp, 597 edst, NULL, NULL); 598 if (error != 0) 599 goto error_put_ifp; 600 else if (is_gw) 601 addr->network = RDMA_NETWORK_IPV6; 602 } 603 604 /* 605 * Step 4 - update source address, if any 606 */ 607 if (saddr != NULL) { 608 src_port = src_in->sin6_port; 609 memcpy(src_in, saddr, rdma_addr_size(saddr)); 610 src_in->sin6_port = src_port; /* preserve port number */ 611 } 612 613 *ifpp = ifp; 614 615 goto done; 616 617 error_put_ifp: 618 dev_put(ifp); 619 done: 620 CURVNET_RESTORE(); 621 622 if (error == EWOULDBLOCK || error == EAGAIN) 623 error = ENODATA; 624 return (-error); 625 } 626 #else 627 static int addr6_resolve(struct sockaddr_in6 *src_in, 628 const struct sockaddr_in6 *dst_in, 629 struct rdma_dev_addr *addr, 630 u8 *edst, 631 struct ifnet **ifpp) 632 { 633 return -EADDRNOTAVAIL; 634 } 635 #endif 636 637 static int addr_resolve_neigh(struct ifnet *dev, 638 const struct sockaddr *dst_in, 639 u8 *edst, 640 struct rdma_dev_addr *addr) 641 { 642 if (dev->if_flags & IFF_LOOPBACK) { 643 int ret; 644 645 /* 646 * Binding to a loopback device is not allowed. Make 647 * sure the destination device address is global by 648 * clearing the bound device interface: 649 */ 650 if (addr->bound_dev_if == dev->if_index) 651 addr->bound_dev_if = 0; 652 653 ret = rdma_translate_ip(dst_in, addr); 654 if (ret == 0) { 655 memcpy(addr->dst_dev_addr, addr->src_dev_addr, 656 MAX_ADDR_LEN); 657 } 658 return ret; 659 } 660 661 /* If the device doesn't do ARP internally */ 662 if (!(dev->if_flags & IFF_NOARP)) 663 return rdma_copy_addr(addr, dev, edst); 664 665 return rdma_copy_addr(addr, dev, NULL); 666 } 667 668 static int addr_resolve(struct sockaddr *src_in, 669 const struct sockaddr *dst_in, 670 struct rdma_dev_addr *addr) 671 { 672 struct epoch_tracker et; 673 struct ifnet *ndev = NULL; 674 u8 edst[MAX_ADDR_LEN]; 675 int ret; 676 677 if (dst_in->sa_family != src_in->sa_family) 678 return -EINVAL; 679 680 NET_EPOCH_ENTER(et); 681 switch (src_in->sa_family) { 682 case AF_INET: 683 ret = addr4_resolve((struct sockaddr_in *)src_in, 684 (const struct sockaddr_in *)dst_in, 685 addr, edst, &ndev); 686 break; 687 case AF_INET6: 688 ret = addr6_resolve((struct sockaddr_in6 *)src_in, 689 (const struct sockaddr_in6 *)dst_in, addr, 690 edst, &ndev); 691 break; 692 default: 693 ret = -EADDRNOTAVAIL; 694 break; 695 } 696 NET_EPOCH_EXIT(et); 697 698 /* check for error */ 699 if (ret != 0) 700 return ret; 701 702 /* store MAC addresses and check for loopback */ 703 ret = addr_resolve_neigh(ndev, dst_in, edst, addr); 704 705 /* set belonging VNET, if any */ 706 addr->net = dev_net(ndev); 707 dev_put(ndev); 708 709 return ret; 710 } 711 712 static void process_req(struct work_struct *work) 713 { 714 struct addr_req *req, *temp_req; 715 struct sockaddr *src_in, *dst_in; 716 struct list_head done_list; 717 718 INIT_LIST_HEAD(&done_list); 719 720 mutex_lock(&lock); 721 list_for_each_entry_safe(req, temp_req, &req_list, list) { 722 if (req->status == -ENODATA) { 723 src_in = (struct sockaddr *) &req->src_addr; 724 dst_in = (struct sockaddr *) &req->dst_addr; 725 req->status = addr_resolve(src_in, dst_in, req->addr); 726 if (req->status && time_after_eq(jiffies, req->timeout)) 727 req->status = -ETIMEDOUT; 728 else if (req->status == -ENODATA) 729 continue; 730 } 731 list_move_tail(&req->list, &done_list); 732 } 733 734 if (!list_empty(&req_list)) { 735 req = list_entry(req_list.next, struct addr_req, list); 736 set_timeout(req->timeout); 737 } 738 mutex_unlock(&lock); 739 740 list_for_each_entry_safe(req, temp_req, &done_list, list) { 741 list_del(&req->list); 742 req->callback(req->status, (struct sockaddr *) &req->src_addr, 743 req->addr, req->context); 744 put_client(req->client); 745 kfree(req); 746 } 747 } 748 749 int rdma_resolve_ip(struct rdma_addr_client *client, 750 struct sockaddr *src_addr, struct sockaddr *dst_addr, 751 struct rdma_dev_addr *addr, int timeout_ms, 752 void (*callback)(int status, struct sockaddr *src_addr, 753 struct rdma_dev_addr *addr, void *context), 754 void *context) 755 { 756 struct sockaddr *src_in, *dst_in; 757 struct addr_req *req; 758 int ret = 0; 759 760 req = kzalloc(sizeof *req, GFP_KERNEL); 761 if (!req) 762 return -ENOMEM; 763 764 src_in = (struct sockaddr *) &req->src_addr; 765 dst_in = (struct sockaddr *) &req->dst_addr; 766 767 if (src_addr) { 768 if (src_addr->sa_family != dst_addr->sa_family) { 769 ret = -EINVAL; 770 goto err; 771 } 772 773 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 774 } else { 775 src_in->sa_family = dst_addr->sa_family; 776 } 777 778 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr)); 779 req->addr = addr; 780 req->callback = callback; 781 req->context = context; 782 req->client = client; 783 atomic_inc(&client->refcount); 784 785 req->status = addr_resolve(src_in, dst_in, addr); 786 switch (req->status) { 787 case 0: 788 req->timeout = jiffies; 789 queue_req(req); 790 break; 791 case -ENODATA: 792 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies; 793 queue_req(req); 794 break; 795 default: 796 ret = req->status; 797 atomic_dec(&client->refcount); 798 goto err; 799 } 800 return ret; 801 err: 802 kfree(req); 803 return ret; 804 } 805 EXPORT_SYMBOL(rdma_resolve_ip); 806 807 int rdma_resolve_ip_route(struct sockaddr *src_addr, 808 const struct sockaddr *dst_addr, 809 struct rdma_dev_addr *addr) 810 { 811 struct sockaddr_storage ssrc_addr = {}; 812 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr; 813 814 if (src_addr) { 815 if (src_addr->sa_family != dst_addr->sa_family) 816 return -EINVAL; 817 818 memcpy(src_in, src_addr, rdma_addr_size(src_addr)); 819 } else { 820 src_in->sa_family = dst_addr->sa_family; 821 } 822 823 return addr_resolve(src_in, dst_addr, addr); 824 } 825 EXPORT_SYMBOL(rdma_resolve_ip_route); 826 827 void rdma_addr_cancel(struct rdma_dev_addr *addr) 828 { 829 struct addr_req *req, *temp_req; 830 831 mutex_lock(&lock); 832 list_for_each_entry_safe(req, temp_req, &req_list, list) { 833 if (req->addr == addr) { 834 req->status = -ECANCELED; 835 req->timeout = jiffies; 836 list_move(&req->list, &req_list); 837 set_timeout(req->timeout); 838 break; 839 } 840 } 841 mutex_unlock(&lock); 842 } 843 EXPORT_SYMBOL(rdma_addr_cancel); 844 845 struct resolve_cb_context { 846 struct rdma_dev_addr *addr; 847 struct completion comp; 848 int status; 849 }; 850 851 static void resolve_cb(int status, struct sockaddr *src_addr, 852 struct rdma_dev_addr *addr, void *context) 853 { 854 if (!status) 855 memcpy(((struct resolve_cb_context *)context)->addr, 856 addr, sizeof(struct rdma_dev_addr)); 857 ((struct resolve_cb_context *)context)->status = status; 858 complete(&((struct resolve_cb_context *)context)->comp); 859 } 860 861 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid, 862 const union ib_gid *dgid, 863 u8 *dmac, struct ifnet *dev, 864 int *hoplimit) 865 { 866 int ret = 0; 867 struct rdma_dev_addr dev_addr; 868 struct resolve_cb_context ctx; 869 870 union rdma_sockaddr sgid_addr, dgid_addr; 871 872 rdma_gid2ip(&sgid_addr._sockaddr, sgid); 873 rdma_gid2ip(&dgid_addr._sockaddr, dgid); 874 875 memset(&dev_addr, 0, sizeof(dev_addr)); 876 877 dev_addr.bound_dev_if = dev->if_index; 878 dev_addr.net = dev_net(dev); 879 880 ctx.addr = &dev_addr; 881 init_completion(&ctx.comp); 882 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr, 883 &dev_addr, 1000, resolve_cb, &ctx); 884 if (ret) 885 return ret; 886 887 wait_for_completion(&ctx.comp); 888 889 ret = ctx.status; 890 if (ret) 891 return ret; 892 893 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN); 894 if (hoplimit) 895 *hoplimit = dev_addr.hoplimit; 896 return ret; 897 } 898 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh); 899 900 int addr_init(void) 901 { 902 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0); 903 if (!addr_wq) 904 return -ENOMEM; 905 906 rdma_addr_register_client(&self); 907 908 return 0; 909 } 910 911 void addr_cleanup(void) 912 { 913 rdma_addr_unregister_client(&self); 914 destroy_workqueue(addr_wq); 915 } 916