1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2020 Alexander V. Chernikov 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 #include "opt_inet.h" 31 #include "opt_inet6.h" 32 #include "opt_route.h" 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/lock.h> 37 #include <sys/rwlock.h> 38 #include <sys/malloc.h> 39 #include <sys/socket.h> 40 #include <sys/sysctl.h> 41 #include <sys/kernel.h> 42 #include <sys/epoch.h> 43 44 #include <net/if.h> 45 #include <net/if_var.h> 46 #include <net/if_dl.h> 47 #include <net/route.h> 48 #include <net/route/route_ctl.h> 49 #include <net/route/route_var.h> 50 #include <net/route/nhop_utils.h> 51 #include <net/route/nhop.h> 52 #include <net/route/nhop_var.h> 53 #include <net/vnet.h> 54 55 #define DEBUG_MOD_NAME nhop_ctl 56 #define DEBUG_MAX_LEVEL LOG_DEBUG 57 #include <net/route/route_debug.h> 58 _DECLARE_DEBUG(LOG_INFO); 59 60 /* 61 * This file contains core functionality for the nexthop ("nhop") route subsystem. 62 * The business logic needed to create nexhop objects is implemented here. 63 * 64 * Nexthops in the original sense are the objects containing all the necessary 65 * information to forward the packet to the selected destination. 66 * In particular, nexthop is defined by a combination of 67 * ifp, ifa, aifp, mtu, gw addr(if set), nh_type, nh_upper_family, mask of rt_flags and 68 * NHF_DEFAULT 69 * 70 * Additionally, each nexthop gets assigned its unique index (nexthop index). 71 * It serves two purposes: first one is to ease the ability of userland programs to 72 * reference nexthops by their index. The second one allows lookup algorithms to 73 * to store index instead of pointer (2 bytes vs 8) as a lookup result. 74 * All nexthops are stored in the resizable hash table. 75 * 76 * Basically, this file revolves around supporting 3 functions: 77 * 1) nhop_create_from_info / nhop_create_from_nhop, which contains all 78 * business logic on filling the nexthop fields based on the provided request. 79 * 2) nhop_get(), which gets a usable referenced nexthops. 80 * 81 * Conventions: 82 * 1) non-exported functions start with verb 83 * 2) exported function starts with the subsystem prefix: "nhop" 84 */ 85 86 static int dump_nhop_entry(struct rib_head *rh, struct nhop_object *nh, struct sysctl_req *w); 87 88 static struct nhop_priv *alloc_nhop_structure(void); 89 static int get_nhop(struct rib_head *rnh, struct rt_addrinfo *info, 90 struct nhop_priv **pnh_priv); 91 static int finalize_nhop(struct nh_control *ctl, struct rt_addrinfo *info, 92 struct nhop_priv *nh_priv); 93 static struct ifnet *get_aifp(const struct nhop_object *nh); 94 static void fill_sdl_from_ifp(struct sockaddr_dl_short *sdl, const struct ifnet *ifp); 95 96 static void destroy_nhop_epoch(epoch_context_t ctx); 97 static void destroy_nhop(struct nhop_priv *nh_priv); 98 99 _Static_assert(__offsetof(struct nhop_object, nh_ifp) == 32, 100 "nhop_object: wrong nh_ifp offset"); 101 _Static_assert(sizeof(struct nhop_object) <= 128, 102 "nhop_object: size exceeds 128 bytes"); 103 104 static uma_zone_t nhops_zone; /* Global zone for each and every nexthop */ 105 106 #define NHOP_OBJECT_ALIGNED_SIZE roundup2(sizeof(struct nhop_object), \ 107 2 * CACHE_LINE_SIZE) 108 #define NHOP_PRIV_ALIGNED_SIZE roundup2(sizeof(struct nhop_priv), \ 109 2 * CACHE_LINE_SIZE) 110 void 111 nhops_init(void) 112 { 113 114 nhops_zone = uma_zcreate("routing nhops", 115 NHOP_OBJECT_ALIGNED_SIZE + NHOP_PRIV_ALIGNED_SIZE, 116 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 117 } 118 119 /* 120 * Fetches the interface of source address used by the route. 121 * In all cases except interface-address-route it would be the 122 * same as the transmit interfaces. 123 * However, for the interface address this function will return 124 * this interface ifp instead of loopback. This is needed to support 125 * link-local IPv6 loopback communications. 126 * 127 * Returns found ifp. 128 */ 129 static struct ifnet * 130 get_aifp(const struct nhop_object *nh) 131 { 132 struct ifnet *aifp = NULL; 133 134 /* 135 * Adjust the "outgoing" interface. If we're going to loop 136 * the packet back to ourselves, the ifp would be the loopback 137 * interface. However, we'd rather know the interface associated 138 * to the destination address (which should probably be one of 139 * our own addresses). 140 */ 141 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) && 142 nh->gw_sa.sa_family == AF_LINK) { 143 aifp = ifnet_byindex(nh->gwl_sa.sdl_index); 144 if (aifp == NULL) { 145 FIB_NH_LOG(LOG_WARNING, nh, "unable to get aifp for %s index %d", 146 if_name(nh->nh_ifp), nh->gwl_sa.sdl_index); 147 } 148 } 149 150 if (aifp == NULL) 151 aifp = nh->nh_ifp; 152 153 return (aifp); 154 } 155 156 int 157 cmp_priv(const struct nhop_priv *_one, const struct nhop_priv *_two) 158 { 159 160 if (memcmp(_one->nh, _two->nh, NHOP_END_CMP) != 0) 161 return (0); 162 163 if (memcmp(_one, _two, NH_PRIV_END_CMP) != 0) 164 return (0); 165 166 return (1); 167 } 168 169 /* 170 * Conditionally sets @nh mtu data based on the @info data. 171 */ 172 static void 173 set_nhop_mtu_from_info(struct nhop_object *nh, const struct rt_addrinfo *info) 174 { 175 176 if (info->rti_mflags & RTV_MTU) { 177 if (info->rti_rmx->rmx_mtu != 0) { 178 /* 179 * MTU was explicitly provided by user. 180 * Keep it. 181 */ 182 183 nh->nh_priv->rt_flags |= RTF_FIXEDMTU; 184 } else { 185 /* 186 * User explicitly sets MTU to 0. 187 * Assume rollback to default. 188 */ 189 nh->nh_priv->rt_flags &= ~RTF_FIXEDMTU; 190 } 191 nh->nh_mtu = info->rti_rmx->rmx_mtu; 192 } 193 } 194 195 /* 196 * Fills in shorted link-level sockadd version suitable to be stored inside the 197 * nexthop gateway buffer. 198 */ 199 static void 200 fill_sdl_from_ifp(struct sockaddr_dl_short *sdl, const struct ifnet *ifp) 201 { 202 203 bzero(sdl, sizeof(struct sockaddr_dl_short)); 204 sdl->sdl_family = AF_LINK; 205 sdl->sdl_len = sizeof(struct sockaddr_dl_short); 206 sdl->sdl_index = ifp->if_index; 207 sdl->sdl_type = ifp->if_type; 208 } 209 210 static int 211 set_nhop_gw_from_info(struct nhop_object *nh, struct rt_addrinfo *info) 212 { 213 struct sockaddr *gw; 214 215 gw = info->rti_info[RTAX_GATEWAY]; 216 KASSERT(gw != NULL, ("gw is NULL")); 217 218 if ((gw->sa_family == AF_LINK) && !(info->rti_flags & RTF_GATEWAY)) { 219 220 /* 221 * Interface route with interface specified by the interface 222 * index in sockadd_dl structure. It is used in the IPv6 loopback 223 * output code, where we need to preserve the original interface 224 * to maintain proper scoping. 225 * Despite the fact that nexthop code stores original interface 226 * in the separate field (nh_aifp, see below), write AF_LINK 227 * compatible sa with shorter total length. 228 */ 229 struct sockaddr_dl *sdl = (struct sockaddr_dl *)gw; 230 struct ifnet *ifp = ifnet_byindex(sdl->sdl_index); 231 if (ifp == NULL) { 232 FIB_NH_LOG(LOG_WARNING, nh, "invalid ifindex %d", sdl->sdl_index); 233 return (EINVAL); 234 } 235 fill_sdl_from_ifp(&nh->gwl_sa, ifp); 236 } else { 237 238 /* 239 * Multiple options here: 240 * 241 * 1) RTF_GATEWAY with IPv4/IPv6 gateway data 242 * 2) Interface route with IPv4/IPv6 address of the 243 * matching interface. Some routing daemons do that 244 * instead of specifying ifindex in AF_LINK. 245 * 246 * In both cases, save the original nexthop to make the callers 247 * happy. 248 */ 249 if (gw->sa_len > sizeof(struct sockaddr_in6)) { 250 FIB_NH_LOG(LOG_WARNING, nh, "nhop SA size too big: AF %d len %u", 251 gw->sa_family, gw->sa_len); 252 return (ENOMEM); 253 } 254 memcpy(&nh->gw_sa, gw, gw->sa_len); 255 } 256 return (0); 257 } 258 259 static uint16_t 260 convert_rt_to_nh_flags(int rt_flags) 261 { 262 uint16_t res; 263 264 res = (rt_flags & RTF_REJECT) ? NHF_REJECT : 0; 265 res |= (rt_flags & RTF_HOST) ? NHF_HOST : 0; 266 res |= (rt_flags & RTF_BLACKHOLE) ? NHF_BLACKHOLE : 0; 267 res |= (rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) ? NHF_REDIRECT : 0; 268 res |= (rt_flags & RTF_BROADCAST) ? NHF_BROADCAST : 0; 269 res |= (rt_flags & RTF_GATEWAY) ? NHF_GATEWAY : 0; 270 271 return (res); 272 } 273 274 static int 275 fill_nhop_from_info(struct nhop_priv *nh_priv, struct rt_addrinfo *info) 276 { 277 int error, rt_flags; 278 struct nhop_object *nh; 279 280 nh = nh_priv->nh; 281 282 rt_flags = info->rti_flags & NHOP_RT_FLAG_MASK; 283 284 nh->nh_priv->rt_flags = rt_flags; 285 nh_priv->nh_upper_family = info->rti_info[RTAX_DST]->sa_family; 286 nh_priv->nh_type = 0; // hook responsibility to set nhop type 287 nh->nh_flags = convert_rt_to_nh_flags(rt_flags); 288 289 set_nhop_mtu_from_info(nh, info); 290 if ((error = set_nhop_gw_from_info(nh, info)) != 0) 291 return (error); 292 if (nh->gw_sa.sa_family == AF_LINK) 293 nh_priv->nh_neigh_family = nh_priv->nh_upper_family; 294 else 295 nh_priv->nh_neigh_family = nh->gw_sa.sa_family; 296 297 nh->nh_ifp = (info->rti_ifp != NULL) ? info->rti_ifp : info->rti_ifa->ifa_ifp; 298 nh->nh_ifa = info->rti_ifa; 299 /* depends on the gateway */ 300 nh->nh_aifp = get_aifp(nh); 301 302 /* 303 * Note some of the remaining data is set by the 304 * per-address-family pre-add hook. 305 */ 306 307 return (0); 308 } 309 310 /* 311 * Creates a new nexthop based on the information in @info. 312 * 313 * Returns: 314 * 0 on success, filling @nh_ret with the desired nexthop object ptr 315 * errno otherwise 316 */ 317 int 318 nhop_create_from_info(struct rib_head *rnh, struct rt_addrinfo *info, 319 struct nhop_object **nh_ret) 320 { 321 struct nhop_priv *nh_priv; 322 int error; 323 324 NET_EPOCH_ASSERT(); 325 326 if (info->rti_info[RTAX_GATEWAY] == NULL) 327 return (EINVAL); 328 329 nh_priv = alloc_nhop_structure(); 330 331 error = fill_nhop_from_info(nh_priv, info); 332 if (error != 0) { 333 uma_zfree(nhops_zone, nh_priv->nh); 334 return (error); 335 } 336 337 error = get_nhop(rnh, info, &nh_priv); 338 if (error == 0) 339 *nh_ret = nh_priv->nh; 340 341 return (error); 342 } 343 344 /* 345 * Gets linked nhop using the provided @pnh_priv nexhop data. 346 * If linked nhop is found, returns it, freeing the provided one. 347 * If there is no such nexthop, attaches the remaining data to the 348 * provided nexthop and links it. 349 * 350 * Returns 0 on success, storing referenced nexthop in @pnh_priv. 351 * Otherwise, errno is returned. 352 */ 353 static int 354 get_nhop(struct rib_head *rnh, struct rt_addrinfo *info, 355 struct nhop_priv **pnh_priv) 356 { 357 const struct sockaddr *dst, *netmask; 358 struct nhop_priv *nh_priv, *tmp_priv; 359 int error; 360 361 nh_priv = *pnh_priv; 362 363 /* Give the protocols chance to augment the request data */ 364 dst = info->rti_info[RTAX_DST]; 365 netmask = info->rti_info[RTAX_NETMASK]; 366 367 error = rnh->rnh_preadd(rnh->rib_fibnum, dst, netmask, nh_priv->nh); 368 if (error != 0) { 369 uma_zfree(nhops_zone, nh_priv->nh); 370 return (error); 371 } 372 373 tmp_priv = find_nhop(rnh->nh_control, nh_priv); 374 if (tmp_priv != NULL) { 375 uma_zfree(nhops_zone, nh_priv->nh); 376 *pnh_priv = tmp_priv; 377 return (0); 378 } 379 380 /* 381 * Existing nexthop not found, need to create new one. 382 * Note: multiple simultaneous get_nhop() requests 383 * can result in multiple equal nexhops existing in the 384 * nexthop table. This is not a not a problem until the 385 * relative number of such nexthops is significant, which 386 * is extremely unlikely. 387 */ 388 389 error = finalize_nhop(rnh->nh_control, info, nh_priv); 390 if (error != 0) 391 return (error); 392 393 return (0); 394 } 395 396 /* 397 * Update @nh with data supplied in @info. 398 * This is a helper function to support route changes. 399 * 400 * It limits the changes that can be done to the route to the following: 401 * 1) all combination of gateway changes (gw, interface, blackhole/reject) 402 * 2) route flags (FLAG[123],STATIC,BLACKHOLE,REJECT) 403 * 3) route MTU 404 * 405 * Returns: 406 * 0 on success 407 */ 408 static int 409 alter_nhop_from_info(struct nhop_object *nh, struct rt_addrinfo *info) 410 { 411 struct nhop_priv *nh_priv = nh->nh_priv; 412 struct sockaddr *info_gw; 413 int error; 414 415 /* Update MTU if set in the request*/ 416 set_nhop_mtu_from_info(nh, info); 417 418 /* XXX: allow only one of BLACKHOLE,REJECT,GATEWAY */ 419 420 /* Allow some flags (FLAG1,STATIC,BLACKHOLE,REJECT) to be toggled on change. */ 421 nh_priv->rt_flags &= ~RTF_FMASK; 422 nh_priv->rt_flags |= info->rti_flags & RTF_FMASK; 423 424 /* Consider gateway change */ 425 info_gw = info->rti_info[RTAX_GATEWAY]; 426 if (info_gw != NULL) { 427 error = set_nhop_gw_from_info(nh, info); 428 if (error != 0) 429 return (error); 430 if (nh->gw_sa.sa_family == AF_LINK) 431 nh_priv->nh_neigh_family = nh_priv->nh_upper_family; 432 else 433 nh_priv->nh_neigh_family = nh->gw_sa.sa_family; 434 /* Update RTF_GATEWAY flag status */ 435 nh_priv->rt_flags &= ~RTF_GATEWAY; 436 nh_priv->rt_flags |= (RTF_GATEWAY & info->rti_flags); 437 } 438 /* Update datapath flags */ 439 nh->nh_flags = convert_rt_to_nh_flags(nh_priv->rt_flags); 440 441 if (info->rti_ifa != NULL) 442 nh->nh_ifa = info->rti_ifa; 443 if (info->rti_ifp != NULL) 444 nh->nh_ifp = info->rti_ifp; 445 nh->nh_aifp = get_aifp(nh); 446 447 return (0); 448 } 449 450 /* 451 * Creates new nexthop based on @nh_orig and augmentation data from @info. 452 * Helper function used in the route changes, please see 453 * alter_nhop_from_info() comments for more details. 454 * 455 * Returns: 456 * 0 on success, filling @nh_ret with the desired nexthop object 457 * errno otherwise 458 */ 459 int 460 nhop_create_from_nhop(struct rib_head *rnh, const struct nhop_object *nh_orig, 461 struct rt_addrinfo *info, struct nhop_object **pnh) 462 { 463 struct nhop_priv *nh_priv; 464 struct nhop_object *nh; 465 int error; 466 467 NET_EPOCH_ASSERT(); 468 469 nh_priv = alloc_nhop_structure(); 470 nh = nh_priv->nh; 471 472 /* Start with copying data from original nexthop */ 473 nh_priv->nh_upper_family = nh_orig->nh_priv->nh_upper_family; 474 nh_priv->nh_neigh_family = nh_orig->nh_priv->nh_neigh_family; 475 nh_priv->rt_flags = nh_orig->nh_priv->rt_flags; 476 nh_priv->nh_type = nh_orig->nh_priv->nh_type; 477 nh_priv->nh_fibnum = nh_orig->nh_priv->nh_fibnum; 478 479 nh->nh_ifp = nh_orig->nh_ifp; 480 nh->nh_ifa = nh_orig->nh_ifa; 481 nh->nh_aifp = nh_orig->nh_aifp; 482 nh->nh_mtu = nh_orig->nh_mtu; 483 nh->nh_flags = nh_orig->nh_flags; 484 memcpy(&nh->gw_sa, &nh_orig->gw_sa, nh_orig->gw_sa.sa_len); 485 486 error = alter_nhop_from_info(nh, info); 487 if (error != 0) { 488 uma_zfree(nhops_zone, nh_priv->nh); 489 return (error); 490 } 491 492 error = get_nhop(rnh, info, &nh_priv); 493 if (error == 0) 494 *pnh = nh_priv->nh; 495 496 return (error); 497 } 498 499 /* 500 * Allocates memory for public/private nexthop structures. 501 * 502 * Returns pointer to nhop_priv or NULL. 503 */ 504 static struct nhop_priv * 505 alloc_nhop_structure() 506 { 507 struct nhop_object *nh; 508 struct nhop_priv *nh_priv; 509 510 nh = (struct nhop_object *)uma_zalloc(nhops_zone, M_NOWAIT | M_ZERO); 511 if (nh == NULL) 512 return (NULL); 513 nh_priv = (struct nhop_priv *)((char *)nh + NHOP_OBJECT_ALIGNED_SIZE); 514 515 nh->nh_priv = nh_priv; 516 nh_priv->nh = nh; 517 518 return (nh_priv); 519 } 520 521 static bool 522 reference_nhop_deps(struct nhop_object *nh) 523 { 524 if (!ifa_try_ref(nh->nh_ifa)) 525 return (false); 526 nh->nh_aifp = get_aifp(nh); 527 if (!if_try_ref(nh->nh_aifp)) { 528 ifa_free(nh->nh_ifa); 529 return (false); 530 } 531 FIB_NH_LOG(LOG_DEBUG, nh, "AIFP: %p nh_ifp %p", nh->nh_aifp, nh->nh_ifp); 532 if (!if_try_ref(nh->nh_ifp)) { 533 ifa_free(nh->nh_ifa); 534 if_rele(nh->nh_aifp); 535 return (false); 536 } 537 538 return (true); 539 } 540 541 /* 542 * Alocates/references the remaining bits of nexthop data and links 543 * it to the hash table. 544 * Returns 0 if successful, 545 * errno otherwise. @nh_priv is freed in case of error. 546 */ 547 static int 548 finalize_nhop(struct nh_control *ctl, struct rt_addrinfo *info, 549 struct nhop_priv *nh_priv) 550 { 551 struct nhop_object *nh = nh_priv->nh; 552 553 /* Allocate per-cpu packet counter */ 554 nh->nh_pksent = counter_u64_alloc(M_NOWAIT); 555 if (nh->nh_pksent == NULL) { 556 uma_zfree(nhops_zone, nh); 557 RTSTAT_INC(rts_nh_alloc_failure); 558 FIB_NH_LOG(LOG_WARNING, nh, "counter_u64_alloc() failed"); 559 return (ENOMEM); 560 } 561 562 if (!reference_nhop_deps(nh)) { 563 counter_u64_free(nh->nh_pksent); 564 uma_zfree(nhops_zone, nh); 565 RTSTAT_INC(rts_nh_alloc_failure); 566 FIB_NH_LOG(LOG_WARNING, nh, "interface reference failed"); 567 return (EAGAIN); 568 } 569 570 /* Save vnet to ease destruction */ 571 nh_priv->nh_vnet = curvnet; 572 573 refcount_init(&nh_priv->nh_refcnt, 1); 574 575 /* Please see nhop_free() comments on the initial value */ 576 refcount_init(&nh_priv->nh_linked, 2); 577 578 nh_priv->nh_fibnum = ctl->ctl_rh->rib_fibnum; 579 580 if (link_nhop(ctl, nh_priv) == 0) { 581 /* 582 * Adding nexthop to the datastructures 583 * failed. Call destructor w/o waiting for 584 * the epoch end, as nexthop is not used 585 * and return. 586 */ 587 char nhbuf[NHOP_PRINT_BUFSIZE]; 588 FIB_NH_LOG(LOG_WARNING, nh, "failed to link %s", 589 nhop_print_buf(nh, nhbuf, sizeof(nhbuf))); 590 destroy_nhop(nh_priv); 591 592 return (ENOBUFS); 593 } 594 595 #if DEBUG_MAX_LEVEL >= LOG_DEBUG 596 char nhbuf[NHOP_PRINT_BUFSIZE]; 597 FIB_NH_LOG(LOG_DEBUG, nh, "finalized: %s", nhop_print_buf(nh, nhbuf, sizeof(nhbuf))); 598 #endif 599 600 return (0); 601 } 602 603 static void 604 destroy_nhop(struct nhop_priv *nh_priv) 605 { 606 struct nhop_object *nh; 607 608 nh = nh_priv->nh; 609 610 if_rele(nh->nh_ifp); 611 if_rele(nh->nh_aifp); 612 ifa_free(nh->nh_ifa); 613 counter_u64_free(nh->nh_pksent); 614 615 uma_zfree(nhops_zone, nh); 616 } 617 618 /* 619 * Epoch callback indicating nhop is safe to destroy 620 */ 621 static void 622 destroy_nhop_epoch(epoch_context_t ctx) 623 { 624 struct nhop_priv *nh_priv; 625 626 nh_priv = __containerof(ctx, struct nhop_priv, nh_epoch_ctx); 627 628 destroy_nhop(nh_priv); 629 } 630 631 void 632 nhop_ref_object(struct nhop_object *nh) 633 { 634 u_int old __diagused; 635 636 old = refcount_acquire(&nh->nh_priv->nh_refcnt); 637 KASSERT(old > 0, ("%s: nhop object %p has 0 refs", __func__, nh)); 638 } 639 640 int 641 nhop_try_ref_object(struct nhop_object *nh) 642 { 643 644 return (refcount_acquire_if_not_zero(&nh->nh_priv->nh_refcnt)); 645 } 646 647 void 648 nhop_free(struct nhop_object *nh) 649 { 650 struct nh_control *ctl; 651 struct nhop_priv *nh_priv = nh->nh_priv; 652 struct epoch_tracker et; 653 654 if (!refcount_release(&nh_priv->nh_refcnt)) 655 return; 656 657 #if DEBUG_MAX_LEVEL >= LOG_DEBUG 658 char nhbuf[NHOP_PRINT_BUFSIZE]; 659 FIB_NH_LOG(LOG_DEBUG, nh, "deleting %s", nhop_print_buf(nh, nhbuf, sizeof(nhbuf))); 660 #endif 661 662 /* 663 * There are only 2 places, where nh_linked can be decreased: 664 * rib destroy (nhops_destroy_rib) and this function. 665 * nh_link can never be increased. 666 * 667 * Hence, use initial value of 2 to make use of 668 * refcount_release_if_not_last(). 669 * 670 * There can be two scenarious when calling this function: 671 * 672 * 1) nh_linked value is 2. This means that either 673 * nhops_destroy_rib() has not been called OR it is running, 674 * but we are guaranteed that nh_control won't be freed in 675 * this epoch. Hence, nexthop can be safely unlinked. 676 * 677 * 2) nh_linked value is 1. In that case, nhops_destroy_rib() 678 * has been called and nhop unlink can be skipped. 679 */ 680 681 NET_EPOCH_ENTER(et); 682 if (refcount_release_if_not_last(&nh_priv->nh_linked)) { 683 ctl = nh_priv->nh_control; 684 if (unlink_nhop(ctl, nh_priv) == NULL) { 685 /* Do not try to reclaim */ 686 char nhbuf[NHOP_PRINT_BUFSIZE]; 687 FIB_NH_LOG(LOG_WARNING, nh, "failed to unlink %s", 688 nhop_print_buf(nh, nhbuf, sizeof(nhbuf))); 689 NET_EPOCH_EXIT(et); 690 return; 691 } 692 } 693 NET_EPOCH_EXIT(et); 694 695 epoch_call(net_epoch_preempt, destroy_nhop_epoch, 696 &nh_priv->nh_epoch_ctx); 697 } 698 699 void 700 nhop_ref_any(struct nhop_object *nh) 701 { 702 #ifdef ROUTE_MPATH 703 if (!NH_IS_NHGRP(nh)) 704 nhop_ref_object(nh); 705 else 706 nhgrp_ref_object((struct nhgrp_object *)nh); 707 #else 708 nhop_ref_object(nh); 709 #endif 710 } 711 712 void 713 nhop_free_any(struct nhop_object *nh) 714 { 715 716 #ifdef ROUTE_MPATH 717 if (!NH_IS_NHGRP(nh)) 718 nhop_free(nh); 719 else 720 nhgrp_free((struct nhgrp_object *)nh); 721 #else 722 nhop_free(nh); 723 #endif 724 } 725 726 /* Helper functions */ 727 728 uint32_t 729 nhop_get_idx(const struct nhop_object *nh) 730 { 731 732 return (nh->nh_priv->nh_idx); 733 } 734 735 enum nhop_type 736 nhop_get_type(const struct nhop_object *nh) 737 { 738 739 return (nh->nh_priv->nh_type); 740 } 741 742 void 743 nhop_set_type(struct nhop_object *nh, enum nhop_type nh_type) 744 { 745 746 nh->nh_priv->nh_type = nh_type; 747 } 748 749 int 750 nhop_get_rtflags(const struct nhop_object *nh) 751 { 752 753 return (nh->nh_priv->rt_flags); 754 } 755 756 void 757 nhop_set_rtflags(struct nhop_object *nh, int rt_flags) 758 { 759 760 nh->nh_priv->rt_flags = rt_flags; 761 } 762 763 struct vnet * 764 nhop_get_vnet(const struct nhop_object *nh) 765 { 766 767 return (nh->nh_priv->nh_vnet); 768 } 769 770 struct nhop_object * 771 nhop_select_func(struct nhop_object *nh, uint32_t flowid) 772 { 773 774 return (nhop_select(nh, flowid)); 775 } 776 777 /* 778 * Returns address family of the traffic uses the nexthop. 779 */ 780 int 781 nhop_get_upper_family(const struct nhop_object *nh) 782 { 783 return (nh->nh_priv->nh_upper_family); 784 } 785 786 /* 787 * Returns address family of the LLE or gateway that is used 788 * to forward the traffic to. 789 */ 790 int 791 nhop_get_neigh_family(const struct nhop_object *nh) 792 { 793 return (nh->nh_priv->nh_neigh_family); 794 } 795 796 uint32_t 797 nhop_get_fibnum(const struct nhop_object *nh) 798 { 799 return (nh->nh_priv->nh_fibnum); 800 } 801 802 void 803 nhops_update_ifmtu(struct rib_head *rh, struct ifnet *ifp, uint32_t mtu) 804 { 805 struct nh_control *ctl; 806 struct nhop_priv *nh_priv; 807 struct nhop_object *nh; 808 809 ctl = rh->nh_control; 810 811 NHOPS_WLOCK(ctl); 812 CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) { 813 nh = nh_priv->nh; 814 if (nh->nh_ifp == ifp) { 815 if ((nh_priv->rt_flags & RTF_FIXEDMTU) == 0 || 816 nh->nh_mtu > mtu) { 817 /* Update MTU directly */ 818 nh->nh_mtu = mtu; 819 } 820 } 821 } CHT_SLIST_FOREACH_END; 822 NHOPS_WUNLOCK(ctl); 823 824 } 825 826 /* 827 * Prints nexthop @nh data in the provided @buf. 828 * Example: nh#33/inet/em0/192.168.0.1 829 */ 830 char * 831 nhop_print_buf(const struct nhop_object *nh, char *buf, size_t bufsize) 832 { 833 #if defined(INET) || defined(INET6) 834 char abuf[INET6_ADDRSTRLEN]; 835 #endif 836 struct nhop_priv *nh_priv = nh->nh_priv; 837 const char *upper_str = rib_print_family(nh->nh_priv->nh_upper_family); 838 839 switch (nh->gw_sa.sa_family) { 840 #ifdef INET 841 case AF_INET: 842 inet_ntop(AF_INET, &nh->gw4_sa.sin_addr, abuf, sizeof(abuf)); 843 snprintf(buf, bufsize, "nh#%d/%s/%s/%s", nh_priv->nh_idx, upper_str, 844 if_name(nh->nh_ifp), abuf); 845 break; 846 #endif 847 #ifdef INET6 848 case AF_INET6: 849 inet_ntop(AF_INET6, &nh->gw6_sa.sin6_addr, abuf, sizeof(abuf)); 850 snprintf(buf, bufsize, "nh#%d/%s/%s/%s", nh_priv->nh_idx, upper_str, 851 if_name(nh->nh_ifp), abuf); 852 break; 853 #endif 854 case AF_LINK: 855 snprintf(buf, bufsize, "nh#%d/%s/%s/resolve", nh_priv->nh_idx, upper_str, 856 if_name(nh->nh_ifp)); 857 break; 858 default: 859 snprintf(buf, bufsize, "nh#%d/%s/%s/????", nh_priv->nh_idx, upper_str, 860 if_name(nh->nh_ifp)); 861 break; 862 } 863 864 return (buf); 865 } 866 867 char * 868 nhop_print_buf_any(const struct nhop_object *nh, char *buf, size_t bufsize) 869 { 870 #ifdef ROUTE_MPATH 871 if (NH_IS_NHGRP(nh)) 872 return (nhgrp_print_buf((const struct nhgrp_object *)nh, buf, bufsize)); 873 else 874 #endif 875 return (nhop_print_buf(nh, buf, bufsize)); 876 } 877 878 /* 879 * Dumps a single entry to sysctl buffer. 880 * 881 * Layout: 882 * rt_msghdr - generic RTM header to allow users to skip non-understood messages 883 * nhop_external - nexhop description structure (with length) 884 * nhop_addrs - structure encapsulating GW/SRC sockaddrs 885 */ 886 static int 887 dump_nhop_entry(struct rib_head *rh, struct nhop_object *nh, struct sysctl_req *w) 888 { 889 struct { 890 struct rt_msghdr rtm; 891 struct nhop_external nhe; 892 struct nhop_addrs na; 893 } arpc; 894 struct nhop_external *pnhe; 895 struct sockaddr *gw_sa, *src_sa; 896 struct sockaddr_storage ss; 897 size_t addrs_len; 898 int error; 899 900 memset(&arpc, 0, sizeof(arpc)); 901 902 arpc.rtm.rtm_msglen = sizeof(arpc); 903 arpc.rtm.rtm_version = RTM_VERSION; 904 arpc.rtm.rtm_type = RTM_GET; 905 //arpc.rtm.rtm_flags = RTF_UP; 906 arpc.rtm.rtm_flags = nh->nh_priv->rt_flags; 907 908 /* nhop_external */ 909 pnhe = &arpc.nhe; 910 pnhe->nh_len = sizeof(struct nhop_external); 911 pnhe->nh_idx = nh->nh_priv->nh_idx; 912 pnhe->nh_fib = rh->rib_fibnum; 913 pnhe->ifindex = nh->nh_ifp->if_index; 914 pnhe->aifindex = nh->nh_aifp->if_index; 915 pnhe->nh_family = nh->nh_priv->nh_upper_family; 916 pnhe->nh_type = nh->nh_priv->nh_type; 917 pnhe->nh_mtu = nh->nh_mtu; 918 pnhe->nh_flags = nh->nh_flags; 919 920 memcpy(pnhe->nh_prepend, nh->nh_prepend, sizeof(nh->nh_prepend)); 921 pnhe->prepend_len = nh->nh_prepend_len; 922 pnhe->nh_refcount = nh->nh_priv->nh_refcnt; 923 pnhe->nh_pksent = counter_u64_fetch(nh->nh_pksent); 924 925 /* sockaddr container */ 926 addrs_len = sizeof(struct nhop_addrs); 927 arpc.na.gw_sa_off = addrs_len; 928 gw_sa = (struct sockaddr *)&nh->gw4_sa; 929 addrs_len += gw_sa->sa_len; 930 931 src_sa = nh->nh_ifa->ifa_addr; 932 if (src_sa->sa_family == AF_LINK) { 933 /* Shorten structure */ 934 memset(&ss, 0, sizeof(struct sockaddr_storage)); 935 fill_sdl_from_ifp((struct sockaddr_dl_short *)&ss, 936 nh->nh_ifa->ifa_ifp); 937 src_sa = (struct sockaddr *)&ss; 938 } 939 arpc.na.src_sa_off = addrs_len; 940 addrs_len += src_sa->sa_len; 941 942 /* Write total container length */ 943 arpc.na.na_len = addrs_len; 944 945 arpc.rtm.rtm_msglen += arpc.na.na_len - sizeof(struct nhop_addrs); 946 947 error = SYSCTL_OUT(w, &arpc, sizeof(arpc)); 948 if (error == 0) 949 error = SYSCTL_OUT(w, gw_sa, gw_sa->sa_len); 950 if (error == 0) 951 error = SYSCTL_OUT(w, src_sa, src_sa->sa_len); 952 953 return (error); 954 } 955 956 uint32_t 957 nhops_get_count(struct rib_head *rh) 958 { 959 struct nh_control *ctl; 960 uint32_t count; 961 962 ctl = rh->nh_control; 963 964 NHOPS_RLOCK(ctl); 965 count = ctl->nh_head.items_count; 966 NHOPS_RUNLOCK(ctl); 967 968 return (count); 969 } 970 971 int 972 nhops_dump_sysctl(struct rib_head *rh, struct sysctl_req *w) 973 { 974 struct nh_control *ctl; 975 struct nhop_priv *nh_priv; 976 int error; 977 978 ctl = rh->nh_control; 979 980 NHOPS_RLOCK(ctl); 981 #if DEBUG_MAX_LEVEL >= LOG_DEBUG 982 FIB_LOG(LOG_DEBUG, rh->rib_fibnum, rh->rib_family, "dump %u items", 983 ctl->nh_head.items_count); 984 #endif 985 CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) { 986 error = dump_nhop_entry(rh, nh_priv->nh, w); 987 if (error != 0) { 988 NHOPS_RUNLOCK(ctl); 989 return (error); 990 } 991 } CHT_SLIST_FOREACH_END; 992 NHOPS_RUNLOCK(ctl); 993 994 return (0); 995 } 996