1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * IPv4 Forwarding Information Base: semantics. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 */ 11 12 #include <linux/uaccess.h> 13 #include <linux/bitops.h> 14 #include <linux/types.h> 15 #include <linux/kernel.h> 16 #include <linux/jiffies.h> 17 #include <linux/mm.h> 18 #include <linux/string.h> 19 #include <linux/socket.h> 20 #include <linux/sockios.h> 21 #include <linux/errno.h> 22 #include <linux/in.h> 23 #include <linux/inet.h> 24 #include <linux/inetdevice.h> 25 #include <linux/netdevice.h> 26 #include <linux/if_arp.h> 27 #include <linux/proc_fs.h> 28 #include <linux/skbuff.h> 29 #include <linux/init.h> 30 #include <linux/slab.h> 31 #include <linux/netlink.h> 32 33 #include <net/arp.h> 34 #include <net/ip.h> 35 #include <net/protocol.h> 36 #include <net/route.h> 37 #include <net/tcp.h> 38 #include <net/sock.h> 39 #include <net/ip_fib.h> 40 #include <net/ip6_fib.h> 41 #include <net/nexthop.h> 42 #include <net/netlink.h> 43 #include <net/rtnh.h> 44 #include <net/lwtunnel.h> 45 #include <net/fib_notifier.h> 46 #include <net/addrconf.h> 47 48 #include "fib_lookup.h" 49 50 static DEFINE_SPINLOCK(fib_info_lock); 51 static struct hlist_head *fib_info_hash; 52 static struct hlist_head *fib_info_laddrhash; 53 static unsigned int fib_info_hash_size; 54 static unsigned int fib_info_cnt; 55 56 #define DEVINDEX_HASHBITS 8 57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS) 58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE]; 59 60 /* for_nexthops and change_nexthops only used when nexthop object 61 * is not set in a fib_info. The logic within can reference fib_nh. 62 */ 63 #ifdef CONFIG_IP_ROUTE_MULTIPATH 64 65 #define for_nexthops(fi) { \ 66 int nhsel; const struct fib_nh *nh; \ 67 for (nhsel = 0, nh = (fi)->fib_nh; \ 68 nhsel < fib_info_num_path((fi)); \ 69 nh++, nhsel++) 70 71 #define change_nexthops(fi) { \ 72 int nhsel; struct fib_nh *nexthop_nh; \ 73 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 74 nhsel < fib_info_num_path((fi)); \ 75 nexthop_nh++, nhsel++) 76 77 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 78 79 /* Hope, that gcc will optimize it to get rid of dummy loop */ 80 81 #define for_nexthops(fi) { \ 82 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \ 83 for (nhsel = 0; nhsel < 1; nhsel++) 84 85 #define change_nexthops(fi) { \ 86 int nhsel; \ 87 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 88 for (nhsel = 0; nhsel < 1; nhsel++) 89 90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 91 92 #define endfor_nexthops(fi) } 93 94 95 const struct fib_prop fib_props[RTN_MAX + 1] = { 96 [RTN_UNSPEC] = { 97 .error = 0, 98 .scope = RT_SCOPE_NOWHERE, 99 }, 100 [RTN_UNICAST] = { 101 .error = 0, 102 .scope = RT_SCOPE_UNIVERSE, 103 }, 104 [RTN_LOCAL] = { 105 .error = 0, 106 .scope = RT_SCOPE_HOST, 107 }, 108 [RTN_BROADCAST] = { 109 .error = 0, 110 .scope = RT_SCOPE_LINK, 111 }, 112 [RTN_ANYCAST] = { 113 .error = 0, 114 .scope = RT_SCOPE_LINK, 115 }, 116 [RTN_MULTICAST] = { 117 .error = 0, 118 .scope = RT_SCOPE_UNIVERSE, 119 }, 120 [RTN_BLACKHOLE] = { 121 .error = -EINVAL, 122 .scope = RT_SCOPE_UNIVERSE, 123 }, 124 [RTN_UNREACHABLE] = { 125 .error = -EHOSTUNREACH, 126 .scope = RT_SCOPE_UNIVERSE, 127 }, 128 [RTN_PROHIBIT] = { 129 .error = -EACCES, 130 .scope = RT_SCOPE_UNIVERSE, 131 }, 132 [RTN_THROW] = { 133 .error = -EAGAIN, 134 .scope = RT_SCOPE_UNIVERSE, 135 }, 136 [RTN_NAT] = { 137 .error = -EINVAL, 138 .scope = RT_SCOPE_NOWHERE, 139 }, 140 [RTN_XRESOLVE] = { 141 .error = -EINVAL, 142 .scope = RT_SCOPE_NOWHERE, 143 }, 144 }; 145 146 static void rt_fibinfo_free(struct rtable __rcu **rtp) 147 { 148 struct rtable *rt = rcu_dereference_protected(*rtp, 1); 149 150 if (!rt) 151 return; 152 153 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL); 154 * because we waited an RCU grace period before calling 155 * free_fib_info_rcu() 156 */ 157 158 dst_dev_put(&rt->dst); 159 dst_release_immediate(&rt->dst); 160 } 161 162 static void free_nh_exceptions(struct fib_nh_common *nhc) 163 { 164 struct fnhe_hash_bucket *hash; 165 int i; 166 167 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1); 168 if (!hash) 169 return; 170 for (i = 0; i < FNHE_HASH_SIZE; i++) { 171 struct fib_nh_exception *fnhe; 172 173 fnhe = rcu_dereference_protected(hash[i].chain, 1); 174 while (fnhe) { 175 struct fib_nh_exception *next; 176 177 next = rcu_dereference_protected(fnhe->fnhe_next, 1); 178 179 rt_fibinfo_free(&fnhe->fnhe_rth_input); 180 rt_fibinfo_free(&fnhe->fnhe_rth_output); 181 182 kfree(fnhe); 183 184 fnhe = next; 185 } 186 } 187 kfree(hash); 188 } 189 190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) 191 { 192 int cpu; 193 194 if (!rtp) 195 return; 196 197 for_each_possible_cpu(cpu) { 198 struct rtable *rt; 199 200 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); 201 if (rt) { 202 dst_dev_put(&rt->dst); 203 dst_release_immediate(&rt->dst); 204 } 205 } 206 free_percpu(rtp); 207 } 208 209 void fib_nh_common_release(struct fib_nh_common *nhc) 210 { 211 dev_put_track(nhc->nhc_dev, &nhc->nhc_dev_tracker); 212 lwtstate_put(nhc->nhc_lwtstate); 213 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output); 214 rt_fibinfo_free(&nhc->nhc_rth_input); 215 free_nh_exceptions(nhc); 216 } 217 EXPORT_SYMBOL_GPL(fib_nh_common_release); 218 219 void fib_nh_release(struct net *net, struct fib_nh *fib_nh) 220 { 221 #ifdef CONFIG_IP_ROUTE_CLASSID 222 if (fib_nh->nh_tclassid) 223 atomic_dec(&net->ipv4.fib_num_tclassid_users); 224 #endif 225 fib_nh_common_release(&fib_nh->nh_common); 226 } 227 228 /* Release a nexthop info record */ 229 static void free_fib_info_rcu(struct rcu_head *head) 230 { 231 struct fib_info *fi = container_of(head, struct fib_info, rcu); 232 233 if (fi->nh) { 234 nexthop_put(fi->nh); 235 } else { 236 change_nexthops(fi) { 237 fib_nh_release(fi->fib_net, nexthop_nh); 238 } endfor_nexthops(fi); 239 } 240 241 ip_fib_metrics_put(fi->fib_metrics); 242 243 kfree(fi); 244 } 245 246 void free_fib_info(struct fib_info *fi) 247 { 248 if (fi->fib_dead == 0) { 249 pr_warn("Freeing alive fib_info %p\n", fi); 250 return; 251 } 252 fib_info_cnt--; 253 254 call_rcu(&fi->rcu, free_fib_info_rcu); 255 } 256 EXPORT_SYMBOL_GPL(free_fib_info); 257 258 void fib_release_info(struct fib_info *fi) 259 { 260 spin_lock_bh(&fib_info_lock); 261 if (fi && refcount_dec_and_test(&fi->fib_treeref)) { 262 hlist_del(&fi->fib_hash); 263 if (fi->fib_prefsrc) 264 hlist_del(&fi->fib_lhash); 265 if (fi->nh) { 266 list_del(&fi->nh_list); 267 } else { 268 change_nexthops(fi) { 269 if (!nexthop_nh->fib_nh_dev) 270 continue; 271 hlist_del(&nexthop_nh->nh_hash); 272 } endfor_nexthops(fi) 273 } 274 fi->fib_dead = 1; 275 fib_info_put(fi); 276 } 277 spin_unlock_bh(&fib_info_lock); 278 } 279 280 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi) 281 { 282 const struct fib_nh *onh; 283 284 if (fi->nh || ofi->nh) 285 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1; 286 287 if (ofi->fib_nhs == 0) 288 return 0; 289 290 for_nexthops(fi) { 291 onh = fib_info_nh(ofi, nhsel); 292 293 if (nh->fib_nh_oif != onh->fib_nh_oif || 294 nh->fib_nh_gw_family != onh->fib_nh_gw_family || 295 nh->fib_nh_scope != onh->fib_nh_scope || 296 #ifdef CONFIG_IP_ROUTE_MULTIPATH 297 nh->fib_nh_weight != onh->fib_nh_weight || 298 #endif 299 #ifdef CONFIG_IP_ROUTE_CLASSID 300 nh->nh_tclassid != onh->nh_tclassid || 301 #endif 302 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) || 303 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK)) 304 return -1; 305 306 if (nh->fib_nh_gw_family == AF_INET && 307 nh->fib_nh_gw4 != onh->fib_nh_gw4) 308 return -1; 309 310 if (nh->fib_nh_gw_family == AF_INET6 && 311 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6)) 312 return -1; 313 } endfor_nexthops(fi); 314 return 0; 315 } 316 317 static inline unsigned int fib_devindex_hashfn(unsigned int val) 318 { 319 unsigned int mask = DEVINDEX_HASHSIZE - 1; 320 321 return (val ^ 322 (val >> DEVINDEX_HASHBITS) ^ 323 (val >> (DEVINDEX_HASHBITS * 2))) & mask; 324 } 325 326 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope, 327 u32 prefsrc, u32 priority) 328 { 329 unsigned int val = init_val; 330 331 val ^= (protocol << 8) | scope; 332 val ^= prefsrc; 333 val ^= priority; 334 335 return val; 336 } 337 338 static unsigned int fib_info_hashfn_result(unsigned int val) 339 { 340 unsigned int mask = (fib_info_hash_size - 1); 341 342 return (val ^ (val >> 7) ^ (val >> 12)) & mask; 343 } 344 345 static inline unsigned int fib_info_hashfn(struct fib_info *fi) 346 { 347 unsigned int val; 348 349 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol, 350 fi->fib_scope, (__force u32)fi->fib_prefsrc, 351 fi->fib_priority); 352 353 if (fi->nh) { 354 val ^= fib_devindex_hashfn(fi->nh->id); 355 } else { 356 for_nexthops(fi) { 357 val ^= fib_devindex_hashfn(nh->fib_nh_oif); 358 } endfor_nexthops(fi) 359 } 360 361 return fib_info_hashfn_result(val); 362 } 363 364 /* no metrics, only nexthop id */ 365 static struct fib_info *fib_find_info_nh(struct net *net, 366 const struct fib_config *cfg) 367 { 368 struct hlist_head *head; 369 struct fib_info *fi; 370 unsigned int hash; 371 372 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id), 373 cfg->fc_protocol, cfg->fc_scope, 374 (__force u32)cfg->fc_prefsrc, 375 cfg->fc_priority); 376 hash = fib_info_hashfn_result(hash); 377 head = &fib_info_hash[hash]; 378 379 hlist_for_each_entry(fi, head, fib_hash) { 380 if (!net_eq(fi->fib_net, net)) 381 continue; 382 if (!fi->nh || fi->nh->id != cfg->fc_nh_id) 383 continue; 384 if (cfg->fc_protocol == fi->fib_protocol && 385 cfg->fc_scope == fi->fib_scope && 386 cfg->fc_prefsrc == fi->fib_prefsrc && 387 cfg->fc_priority == fi->fib_priority && 388 cfg->fc_type == fi->fib_type && 389 cfg->fc_table == fi->fib_tb_id && 390 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK)) 391 return fi; 392 } 393 394 return NULL; 395 } 396 397 static struct fib_info *fib_find_info(struct fib_info *nfi) 398 { 399 struct hlist_head *head; 400 struct fib_info *fi; 401 unsigned int hash; 402 403 hash = fib_info_hashfn(nfi); 404 head = &fib_info_hash[hash]; 405 406 hlist_for_each_entry(fi, head, fib_hash) { 407 if (!net_eq(fi->fib_net, nfi->fib_net)) 408 continue; 409 if (fi->fib_nhs != nfi->fib_nhs) 410 continue; 411 if (nfi->fib_protocol == fi->fib_protocol && 412 nfi->fib_scope == fi->fib_scope && 413 nfi->fib_prefsrc == fi->fib_prefsrc && 414 nfi->fib_priority == fi->fib_priority && 415 nfi->fib_type == fi->fib_type && 416 memcmp(nfi->fib_metrics, fi->fib_metrics, 417 sizeof(u32) * RTAX_MAX) == 0 && 418 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) && 419 nh_comp(fi, nfi) == 0) 420 return fi; 421 } 422 423 return NULL; 424 } 425 426 /* Check, that the gateway is already configured. 427 * Used only by redirect accept routine. 428 */ 429 int ip_fib_check_default(__be32 gw, struct net_device *dev) 430 { 431 struct hlist_head *head; 432 struct fib_nh *nh; 433 unsigned int hash; 434 435 spin_lock(&fib_info_lock); 436 437 hash = fib_devindex_hashfn(dev->ifindex); 438 head = &fib_info_devhash[hash]; 439 hlist_for_each_entry(nh, head, nh_hash) { 440 if (nh->fib_nh_dev == dev && 441 nh->fib_nh_gw4 == gw && 442 !(nh->fib_nh_flags & RTNH_F_DEAD)) { 443 spin_unlock(&fib_info_lock); 444 return 0; 445 } 446 } 447 448 spin_unlock(&fib_info_lock); 449 450 return -1; 451 } 452 453 size_t fib_nlmsg_size(struct fib_info *fi) 454 { 455 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) 456 + nla_total_size(4) /* RTA_TABLE */ 457 + nla_total_size(4) /* RTA_DST */ 458 + nla_total_size(4) /* RTA_PRIORITY */ 459 + nla_total_size(4) /* RTA_PREFSRC */ 460 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ 461 unsigned int nhs = fib_info_num_path(fi); 462 463 /* space for nested metrics */ 464 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 465 466 if (fi->nh) 467 payload += nla_total_size(4); /* RTA_NH_ID */ 468 469 if (nhs) { 470 size_t nh_encapsize = 0; 471 /* Also handles the special case nhs == 1 */ 472 473 /* each nexthop is packed in an attribute */ 474 size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); 475 unsigned int i; 476 477 /* may contain flow and gateway attribute */ 478 nhsize += 2 * nla_total_size(4); 479 480 /* grab encap info */ 481 for (i = 0; i < fib_info_num_path(fi); i++) { 482 struct fib_nh_common *nhc = fib_info_nhc(fi, i); 483 484 if (nhc->nhc_lwtstate) { 485 /* RTA_ENCAP_TYPE */ 486 nh_encapsize += lwtunnel_get_encap_size( 487 nhc->nhc_lwtstate); 488 /* RTA_ENCAP */ 489 nh_encapsize += nla_total_size(2); 490 } 491 } 492 493 /* all nexthops are packed in a nested attribute */ 494 payload += nla_total_size((nhs * nhsize) + nh_encapsize); 495 496 } 497 498 return payload; 499 } 500 501 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 502 int dst_len, u32 tb_id, const struct nl_info *info, 503 unsigned int nlm_flags) 504 { 505 struct fib_rt_info fri; 506 struct sk_buff *skb; 507 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 508 int err = -ENOBUFS; 509 510 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 511 if (!skb) 512 goto errout; 513 514 fri.fi = fa->fa_info; 515 fri.tb_id = tb_id; 516 fri.dst = key; 517 fri.dst_len = dst_len; 518 fri.tos = fa->fa_tos; 519 fri.type = fa->fa_type; 520 fri.offload = fa->offload; 521 fri.trap = fa->trap; 522 fri.offload_failed = fa->offload_failed; 523 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags); 524 if (err < 0) { 525 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 526 WARN_ON(err == -EMSGSIZE); 527 kfree_skb(skb); 528 goto errout; 529 } 530 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 531 info->nlh, GFP_KERNEL); 532 return; 533 errout: 534 if (err < 0) 535 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 536 } 537 538 static int fib_detect_death(struct fib_info *fi, int order, 539 struct fib_info **last_resort, int *last_idx, 540 int dflt) 541 { 542 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0); 543 struct neighbour *n; 544 int state = NUD_NONE; 545 546 if (likely(nhc->nhc_gw_family == AF_INET)) 547 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev); 548 else if (nhc->nhc_gw_family == AF_INET6) 549 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6, 550 nhc->nhc_dev); 551 else 552 n = NULL; 553 554 if (n) { 555 state = n->nud_state; 556 neigh_release(n); 557 } else { 558 return 0; 559 } 560 if (state == NUD_REACHABLE) 561 return 0; 562 if ((state & NUD_VALID) && order != dflt) 563 return 0; 564 if ((state & NUD_VALID) || 565 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) { 566 *last_resort = fi; 567 *last_idx = order; 568 } 569 return 1; 570 } 571 572 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc, 573 struct nlattr *encap, u16 encap_type, 574 void *cfg, gfp_t gfp_flags, 575 struct netlink_ext_ack *extack) 576 { 577 int err; 578 579 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *, 580 gfp_flags); 581 if (!nhc->nhc_pcpu_rth_output) 582 return -ENOMEM; 583 584 if (encap) { 585 struct lwtunnel_state *lwtstate; 586 587 if (encap_type == LWTUNNEL_ENCAP_NONE) { 588 NL_SET_ERR_MSG(extack, "LWT encap type not specified"); 589 err = -EINVAL; 590 goto lwt_failure; 591 } 592 err = lwtunnel_build_state(net, encap_type, encap, 593 nhc->nhc_family, cfg, &lwtstate, 594 extack); 595 if (err) 596 goto lwt_failure; 597 598 nhc->nhc_lwtstate = lwtstate_get(lwtstate); 599 } 600 601 return 0; 602 603 lwt_failure: 604 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output); 605 nhc->nhc_pcpu_rth_output = NULL; 606 return err; 607 } 608 EXPORT_SYMBOL_GPL(fib_nh_common_init); 609 610 int fib_nh_init(struct net *net, struct fib_nh *nh, 611 struct fib_config *cfg, int nh_weight, 612 struct netlink_ext_ack *extack) 613 { 614 int err; 615 616 nh->fib_nh_family = AF_INET; 617 618 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap, 619 cfg->fc_encap_type, cfg, GFP_KERNEL, extack); 620 if (err) 621 return err; 622 623 nh->fib_nh_oif = cfg->fc_oif; 624 nh->fib_nh_gw_family = cfg->fc_gw_family; 625 if (cfg->fc_gw_family == AF_INET) 626 nh->fib_nh_gw4 = cfg->fc_gw4; 627 else if (cfg->fc_gw_family == AF_INET6) 628 nh->fib_nh_gw6 = cfg->fc_gw6; 629 630 nh->fib_nh_flags = cfg->fc_flags; 631 632 #ifdef CONFIG_IP_ROUTE_CLASSID 633 nh->nh_tclassid = cfg->fc_flow; 634 if (nh->nh_tclassid) 635 atomic_inc(&net->ipv4.fib_num_tclassid_users); 636 #endif 637 #ifdef CONFIG_IP_ROUTE_MULTIPATH 638 nh->fib_nh_weight = nh_weight; 639 #endif 640 return 0; 641 } 642 643 #ifdef CONFIG_IP_ROUTE_MULTIPATH 644 645 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining, 646 struct netlink_ext_ack *extack) 647 { 648 int nhs = 0; 649 650 while (rtnh_ok(rtnh, remaining)) { 651 nhs++; 652 rtnh = rtnh_next(rtnh, &remaining); 653 } 654 655 /* leftover implies invalid nexthop configuration, discard it */ 656 if (remaining > 0) { 657 NL_SET_ERR_MSG(extack, 658 "Invalid nexthop configuration - extra data after nexthops"); 659 nhs = 0; 660 } 661 662 return nhs; 663 } 664 665 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla, 666 struct netlink_ext_ack *extack) 667 { 668 if (nla_len(nla) < sizeof(*gw)) { 669 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY"); 670 return -EINVAL; 671 } 672 673 *gw = nla_get_in_addr(nla); 674 675 return 0; 676 } 677 678 /* only called when fib_nh is integrated into fib_info */ 679 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 680 int remaining, struct fib_config *cfg, 681 struct netlink_ext_ack *extack) 682 { 683 struct net *net = fi->fib_net; 684 struct fib_config fib_cfg; 685 struct fib_nh *nh; 686 int ret; 687 688 change_nexthops(fi) { 689 int attrlen; 690 691 memset(&fib_cfg, 0, sizeof(fib_cfg)); 692 693 if (!rtnh_ok(rtnh, remaining)) { 694 NL_SET_ERR_MSG(extack, 695 "Invalid nexthop configuration - extra data after nexthop"); 696 return -EINVAL; 697 } 698 699 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 700 NL_SET_ERR_MSG(extack, 701 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN"); 702 return -EINVAL; 703 } 704 705 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 706 fib_cfg.fc_oif = rtnh->rtnh_ifindex; 707 708 attrlen = rtnh_attrlen(rtnh); 709 if (attrlen > 0) { 710 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh); 711 712 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 713 nlav = nla_find(attrs, attrlen, RTA_VIA); 714 if (nla && nlav) { 715 NL_SET_ERR_MSG(extack, 716 "Nexthop configuration can not contain both GATEWAY and VIA"); 717 return -EINVAL; 718 } 719 if (nla) { 720 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla, 721 extack); 722 if (ret) 723 goto errout; 724 725 if (fib_cfg.fc_gw4) 726 fib_cfg.fc_gw_family = AF_INET; 727 } else if (nlav) { 728 ret = fib_gw_from_via(&fib_cfg, nlav, extack); 729 if (ret) 730 goto errout; 731 } 732 733 nla = nla_find(attrs, attrlen, RTA_FLOW); 734 if (nla) { 735 if (nla_len(nla) < sizeof(u32)) { 736 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW"); 737 return -EINVAL; 738 } 739 fib_cfg.fc_flow = nla_get_u32(nla); 740 } 741 742 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP); 743 /* RTA_ENCAP_TYPE length checked in 744 * lwtunnel_valid_encap_type_attr 745 */ 746 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE); 747 if (nla) 748 fib_cfg.fc_encap_type = nla_get_u16(nla); 749 } 750 751 ret = fib_nh_init(net, nexthop_nh, &fib_cfg, 752 rtnh->rtnh_hops + 1, extack); 753 if (ret) 754 goto errout; 755 756 rtnh = rtnh_next(rtnh, &remaining); 757 } endfor_nexthops(fi); 758 759 ret = -EINVAL; 760 nh = fib_info_nh(fi, 0); 761 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) { 762 NL_SET_ERR_MSG(extack, 763 "Nexthop device index does not match RTA_OIF"); 764 goto errout; 765 } 766 if (cfg->fc_gw_family) { 767 if (cfg->fc_gw_family != nh->fib_nh_gw_family || 768 (cfg->fc_gw_family == AF_INET && 769 nh->fib_nh_gw4 != cfg->fc_gw4) || 770 (cfg->fc_gw_family == AF_INET6 && 771 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) { 772 NL_SET_ERR_MSG(extack, 773 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA"); 774 goto errout; 775 } 776 } 777 #ifdef CONFIG_IP_ROUTE_CLASSID 778 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) { 779 NL_SET_ERR_MSG(extack, 780 "Nexthop class id does not match RTA_FLOW"); 781 goto errout; 782 } 783 #endif 784 ret = 0; 785 errout: 786 return ret; 787 } 788 789 /* only called when fib_nh is integrated into fib_info */ 790 static void fib_rebalance(struct fib_info *fi) 791 { 792 int total; 793 int w; 794 795 if (fib_info_num_path(fi) < 2) 796 return; 797 798 total = 0; 799 for_nexthops(fi) { 800 if (nh->fib_nh_flags & RTNH_F_DEAD) 801 continue; 802 803 if (ip_ignore_linkdown(nh->fib_nh_dev) && 804 nh->fib_nh_flags & RTNH_F_LINKDOWN) 805 continue; 806 807 total += nh->fib_nh_weight; 808 } endfor_nexthops(fi); 809 810 w = 0; 811 change_nexthops(fi) { 812 int upper_bound; 813 814 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) { 815 upper_bound = -1; 816 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) && 817 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) { 818 upper_bound = -1; 819 } else { 820 w += nexthop_nh->fib_nh_weight; 821 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, 822 total) - 1; 823 } 824 825 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound); 826 } endfor_nexthops(fi); 827 } 828 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 829 830 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 831 int remaining, struct fib_config *cfg, 832 struct netlink_ext_ack *extack) 833 { 834 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel"); 835 836 return -EINVAL; 837 } 838 839 #define fib_rebalance(fi) do { } while (0) 840 841 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 842 843 static int fib_encap_match(struct net *net, u16 encap_type, 844 struct nlattr *encap, 845 const struct fib_nh *nh, 846 const struct fib_config *cfg, 847 struct netlink_ext_ack *extack) 848 { 849 struct lwtunnel_state *lwtstate; 850 int ret, result = 0; 851 852 if (encap_type == LWTUNNEL_ENCAP_NONE) 853 return 0; 854 855 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET, 856 cfg, &lwtstate, extack); 857 if (!ret) { 858 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws); 859 lwtstate_free(lwtstate); 860 } 861 862 return result; 863 } 864 865 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi, 866 struct netlink_ext_ack *extack) 867 { 868 #ifdef CONFIG_IP_ROUTE_MULTIPATH 869 struct rtnexthop *rtnh; 870 int remaining; 871 #endif 872 873 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 874 return 1; 875 876 if (cfg->fc_nh_id) { 877 if (fi->nh && cfg->fc_nh_id == fi->nh->id) 878 return 0; 879 return 1; 880 } 881 882 if (cfg->fc_oif || cfg->fc_gw_family) { 883 struct fib_nh *nh = fib_info_nh(fi, 0); 884 885 if (cfg->fc_encap) { 886 if (fib_encap_match(net, cfg->fc_encap_type, 887 cfg->fc_encap, nh, cfg, extack)) 888 return 1; 889 } 890 #ifdef CONFIG_IP_ROUTE_CLASSID 891 if (cfg->fc_flow && 892 cfg->fc_flow != nh->nh_tclassid) 893 return 1; 894 #endif 895 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) || 896 (cfg->fc_gw_family && 897 cfg->fc_gw_family != nh->fib_nh_gw_family)) 898 return 1; 899 900 if (cfg->fc_gw_family == AF_INET && 901 cfg->fc_gw4 != nh->fib_nh_gw4) 902 return 1; 903 904 if (cfg->fc_gw_family == AF_INET6 && 905 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6)) 906 return 1; 907 908 return 0; 909 } 910 911 #ifdef CONFIG_IP_ROUTE_MULTIPATH 912 if (!cfg->fc_mp) 913 return 0; 914 915 rtnh = cfg->fc_mp; 916 remaining = cfg->fc_mp_len; 917 918 for_nexthops(fi) { 919 int attrlen; 920 921 if (!rtnh_ok(rtnh, remaining)) 922 return -EINVAL; 923 924 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif) 925 return 1; 926 927 attrlen = rtnh_attrlen(rtnh); 928 if (attrlen > 0) { 929 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh); 930 int err; 931 932 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 933 nlav = nla_find(attrs, attrlen, RTA_VIA); 934 if (nla && nlav) { 935 NL_SET_ERR_MSG(extack, 936 "Nexthop configuration can not contain both GATEWAY and VIA"); 937 return -EINVAL; 938 } 939 940 if (nla) { 941 __be32 gw; 942 943 err = fib_gw_from_attr(&gw, nla, extack); 944 if (err) 945 return err; 946 947 if (nh->fib_nh_gw_family != AF_INET || 948 gw != nh->fib_nh_gw4) 949 return 1; 950 } else if (nlav) { 951 struct fib_config cfg2; 952 953 err = fib_gw_from_via(&cfg2, nlav, extack); 954 if (err) 955 return err; 956 957 switch (nh->fib_nh_gw_family) { 958 case AF_INET: 959 if (cfg2.fc_gw_family != AF_INET || 960 cfg2.fc_gw4 != nh->fib_nh_gw4) 961 return 1; 962 break; 963 case AF_INET6: 964 if (cfg2.fc_gw_family != AF_INET6 || 965 ipv6_addr_cmp(&cfg2.fc_gw6, 966 &nh->fib_nh_gw6)) 967 return 1; 968 break; 969 } 970 } 971 972 #ifdef CONFIG_IP_ROUTE_CLASSID 973 nla = nla_find(attrs, attrlen, RTA_FLOW); 974 if (nla) { 975 if (nla_len(nla) < sizeof(u32)) { 976 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW"); 977 return -EINVAL; 978 } 979 if (nla_get_u32(nla) != nh->nh_tclassid) 980 return 1; 981 } 982 #endif 983 } 984 985 rtnh = rtnh_next(rtnh, &remaining); 986 } endfor_nexthops(fi); 987 #endif 988 return 0; 989 } 990 991 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi) 992 { 993 struct nlattr *nla; 994 int remaining; 995 996 if (!cfg->fc_mx) 997 return true; 998 999 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 1000 int type = nla_type(nla); 1001 u32 fi_val, val; 1002 1003 if (!type) 1004 continue; 1005 if (type > RTAX_MAX) 1006 return false; 1007 1008 if (type == RTAX_CC_ALGO) { 1009 char tmp[TCP_CA_NAME_MAX]; 1010 bool ecn_ca = false; 1011 1012 nla_strscpy(tmp, nla, sizeof(tmp)); 1013 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca); 1014 } else { 1015 if (nla_len(nla) != sizeof(u32)) 1016 return false; 1017 val = nla_get_u32(nla); 1018 } 1019 1020 fi_val = fi->fib_metrics->metrics[type - 1]; 1021 if (type == RTAX_FEATURES) 1022 fi_val &= ~DST_FEATURE_ECN_CA; 1023 1024 if (fi_val != val) 1025 return false; 1026 } 1027 1028 return true; 1029 } 1030 1031 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh, 1032 u32 table, struct netlink_ext_ack *extack) 1033 { 1034 struct fib6_config cfg = { 1035 .fc_table = table, 1036 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY, 1037 .fc_ifindex = nh->fib_nh_oif, 1038 .fc_gateway = nh->fib_nh_gw6, 1039 }; 1040 struct fib6_nh fib6_nh = {}; 1041 int err; 1042 1043 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack); 1044 if (!err) { 1045 nh->fib_nh_dev = fib6_nh.fib_nh_dev; 1046 dev_hold_track(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_KERNEL); 1047 nh->fib_nh_oif = nh->fib_nh_dev->ifindex; 1048 nh->fib_nh_scope = RT_SCOPE_LINK; 1049 1050 ipv6_stub->fib6_nh_release(&fib6_nh); 1051 } 1052 1053 return err; 1054 } 1055 1056 /* 1057 * Picture 1058 * ------- 1059 * 1060 * Semantics of nexthop is very messy by historical reasons. 1061 * We have to take into account, that: 1062 * a) gateway can be actually local interface address, 1063 * so that gatewayed route is direct. 1064 * b) gateway must be on-link address, possibly 1065 * described not by an ifaddr, but also by a direct route. 1066 * c) If both gateway and interface are specified, they should not 1067 * contradict. 1068 * d) If we use tunnel routes, gateway could be not on-link. 1069 * 1070 * Attempt to reconcile all of these (alas, self-contradictory) conditions 1071 * results in pretty ugly and hairy code with obscure logic. 1072 * 1073 * I chose to generalized it instead, so that the size 1074 * of code does not increase practically, but it becomes 1075 * much more general. 1076 * Every prefix is assigned a "scope" value: "host" is local address, 1077 * "link" is direct route, 1078 * [ ... "site" ... "interior" ... ] 1079 * and "universe" is true gateway route with global meaning. 1080 * 1081 * Every prefix refers to a set of "nexthop"s (gw, oif), 1082 * where gw must have narrower scope. This recursion stops 1083 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 1084 * which means that gw is forced to be on link. 1085 * 1086 * Code is still hairy, but now it is apparently logically 1087 * consistent and very flexible. F.e. as by-product it allows 1088 * to co-exists in peace independent exterior and interior 1089 * routing processes. 1090 * 1091 * Normally it looks as following. 1092 * 1093 * {universe prefix} -> (gw, oif) [scope link] 1094 * | 1095 * |-> {link prefix} -> (gw, oif) [scope local] 1096 * | 1097 * |-> {local prefix} (terminal node) 1098 */ 1099 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table, 1100 u8 scope, struct netlink_ext_ack *extack) 1101 { 1102 struct net_device *dev; 1103 struct fib_result res; 1104 int err = 0; 1105 1106 if (nh->fib_nh_flags & RTNH_F_ONLINK) { 1107 unsigned int addr_type; 1108 1109 if (scope >= RT_SCOPE_LINK) { 1110 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope"); 1111 return -EINVAL; 1112 } 1113 dev = __dev_get_by_index(net, nh->fib_nh_oif); 1114 if (!dev) { 1115 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink"); 1116 return -ENODEV; 1117 } 1118 if (!(dev->flags & IFF_UP)) { 1119 NL_SET_ERR_MSG(extack, "Nexthop device is not up"); 1120 return -ENETDOWN; 1121 } 1122 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4); 1123 if (addr_type != RTN_UNICAST) { 1124 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1125 return -EINVAL; 1126 } 1127 if (!netif_carrier_ok(dev)) 1128 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1129 nh->fib_nh_dev = dev; 1130 dev_hold_track(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1131 nh->fib_nh_scope = RT_SCOPE_LINK; 1132 return 0; 1133 } 1134 rcu_read_lock(); 1135 { 1136 struct fib_table *tbl = NULL; 1137 struct flowi4 fl4 = { 1138 .daddr = nh->fib_nh_gw4, 1139 .flowi4_scope = scope + 1, 1140 .flowi4_oif = nh->fib_nh_oif, 1141 .flowi4_iif = LOOPBACK_IFINDEX, 1142 }; 1143 1144 /* It is not necessary, but requires a bit of thinking */ 1145 if (fl4.flowi4_scope < RT_SCOPE_LINK) 1146 fl4.flowi4_scope = RT_SCOPE_LINK; 1147 1148 if (table && table != RT_TABLE_MAIN) 1149 tbl = fib_get_table(net, table); 1150 1151 if (tbl) 1152 err = fib_table_lookup(tbl, &fl4, &res, 1153 FIB_LOOKUP_IGNORE_LINKSTATE | 1154 FIB_LOOKUP_NOREF); 1155 1156 /* on error or if no table given do full lookup. This 1157 * is needed for example when nexthops are in the local 1158 * table rather than the given table 1159 */ 1160 if (!tbl || err) { 1161 err = fib_lookup(net, &fl4, &res, 1162 FIB_LOOKUP_IGNORE_LINKSTATE); 1163 } 1164 1165 if (err) { 1166 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1167 goto out; 1168 } 1169 } 1170 1171 err = -EINVAL; 1172 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) { 1173 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1174 goto out; 1175 } 1176 nh->fib_nh_scope = res.scope; 1177 nh->fib_nh_oif = FIB_RES_OIF(res); 1178 nh->fib_nh_dev = dev = FIB_RES_DEV(res); 1179 if (!dev) { 1180 NL_SET_ERR_MSG(extack, 1181 "No egress device for nexthop gateway"); 1182 goto out; 1183 } 1184 dev_hold_track(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1185 if (!netif_carrier_ok(dev)) 1186 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1187 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 1188 out: 1189 rcu_read_unlock(); 1190 return err; 1191 } 1192 1193 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh, 1194 struct netlink_ext_ack *extack) 1195 { 1196 struct in_device *in_dev; 1197 int err; 1198 1199 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) { 1200 NL_SET_ERR_MSG(extack, 1201 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set"); 1202 return -EINVAL; 1203 } 1204 1205 rcu_read_lock(); 1206 1207 err = -ENODEV; 1208 in_dev = inetdev_by_index(net, nh->fib_nh_oif); 1209 if (!in_dev) 1210 goto out; 1211 err = -ENETDOWN; 1212 if (!(in_dev->dev->flags & IFF_UP)) { 1213 NL_SET_ERR_MSG(extack, "Device for nexthop is not up"); 1214 goto out; 1215 } 1216 1217 nh->fib_nh_dev = in_dev->dev; 1218 dev_hold_track(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1219 nh->fib_nh_scope = RT_SCOPE_HOST; 1220 if (!netif_carrier_ok(nh->fib_nh_dev)) 1221 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1222 err = 0; 1223 out: 1224 rcu_read_unlock(); 1225 return err; 1226 } 1227 1228 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope, 1229 struct netlink_ext_ack *extack) 1230 { 1231 int err; 1232 1233 if (nh->fib_nh_gw_family == AF_INET) 1234 err = fib_check_nh_v4_gw(net, nh, table, scope, extack); 1235 else if (nh->fib_nh_gw_family == AF_INET6) 1236 err = fib_check_nh_v6_gw(net, nh, table, extack); 1237 else 1238 err = fib_check_nh_nongw(net, nh, extack); 1239 1240 return err; 1241 } 1242 1243 static inline unsigned int fib_laddr_hashfn(__be32 val) 1244 { 1245 unsigned int mask = (fib_info_hash_size - 1); 1246 1247 return ((__force u32)val ^ 1248 ((__force u32)val >> 7) ^ 1249 ((__force u32)val >> 14)) & mask; 1250 } 1251 1252 static struct hlist_head *fib_info_hash_alloc(int bytes) 1253 { 1254 if (bytes <= PAGE_SIZE) 1255 return kzalloc(bytes, GFP_KERNEL); 1256 else 1257 return (struct hlist_head *) 1258 __get_free_pages(GFP_KERNEL | __GFP_ZERO, 1259 get_order(bytes)); 1260 } 1261 1262 static void fib_info_hash_free(struct hlist_head *hash, int bytes) 1263 { 1264 if (!hash) 1265 return; 1266 1267 if (bytes <= PAGE_SIZE) 1268 kfree(hash); 1269 else 1270 free_pages((unsigned long) hash, get_order(bytes)); 1271 } 1272 1273 static void fib_info_hash_move(struct hlist_head *new_info_hash, 1274 struct hlist_head *new_laddrhash, 1275 unsigned int new_size) 1276 { 1277 struct hlist_head *old_info_hash, *old_laddrhash; 1278 unsigned int old_size = fib_info_hash_size; 1279 unsigned int i, bytes; 1280 1281 spin_lock_bh(&fib_info_lock); 1282 old_info_hash = fib_info_hash; 1283 old_laddrhash = fib_info_laddrhash; 1284 fib_info_hash_size = new_size; 1285 1286 for (i = 0; i < old_size; i++) { 1287 struct hlist_head *head = &fib_info_hash[i]; 1288 struct hlist_node *n; 1289 struct fib_info *fi; 1290 1291 hlist_for_each_entry_safe(fi, n, head, fib_hash) { 1292 struct hlist_head *dest; 1293 unsigned int new_hash; 1294 1295 new_hash = fib_info_hashfn(fi); 1296 dest = &new_info_hash[new_hash]; 1297 hlist_add_head(&fi->fib_hash, dest); 1298 } 1299 } 1300 fib_info_hash = new_info_hash; 1301 1302 for (i = 0; i < old_size; i++) { 1303 struct hlist_head *lhead = &fib_info_laddrhash[i]; 1304 struct hlist_node *n; 1305 struct fib_info *fi; 1306 1307 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) { 1308 struct hlist_head *ldest; 1309 unsigned int new_hash; 1310 1311 new_hash = fib_laddr_hashfn(fi->fib_prefsrc); 1312 ldest = &new_laddrhash[new_hash]; 1313 hlist_add_head(&fi->fib_lhash, ldest); 1314 } 1315 } 1316 fib_info_laddrhash = new_laddrhash; 1317 1318 spin_unlock_bh(&fib_info_lock); 1319 1320 bytes = old_size * sizeof(struct hlist_head *); 1321 fib_info_hash_free(old_info_hash, bytes); 1322 fib_info_hash_free(old_laddrhash, bytes); 1323 } 1324 1325 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc, 1326 unsigned char scope) 1327 { 1328 struct fib_nh *nh; 1329 1330 if (nhc->nhc_family != AF_INET) 1331 return inet_select_addr(nhc->nhc_dev, 0, scope); 1332 1333 nh = container_of(nhc, struct fib_nh, nh_common); 1334 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope); 1335 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid); 1336 1337 return nh->nh_saddr; 1338 } 1339 1340 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res) 1341 { 1342 struct fib_nh_common *nhc = res->nhc; 1343 1344 if (res->fi->fib_prefsrc) 1345 return res->fi->fib_prefsrc; 1346 1347 if (nhc->nhc_family == AF_INET) { 1348 struct fib_nh *nh; 1349 1350 nh = container_of(nhc, struct fib_nh, nh_common); 1351 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid)) 1352 return nh->nh_saddr; 1353 } 1354 1355 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope); 1356 } 1357 1358 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc) 1359 { 1360 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 1361 fib_prefsrc != cfg->fc_dst) { 1362 u32 tb_id = cfg->fc_table; 1363 int rc; 1364 1365 if (tb_id == RT_TABLE_MAIN) 1366 tb_id = RT_TABLE_LOCAL; 1367 1368 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1369 fib_prefsrc, tb_id); 1370 1371 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) { 1372 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1373 fib_prefsrc, RT_TABLE_LOCAL); 1374 } 1375 1376 if (rc != RTN_LOCAL) 1377 return false; 1378 } 1379 return true; 1380 } 1381 1382 struct fib_info *fib_create_info(struct fib_config *cfg, 1383 struct netlink_ext_ack *extack) 1384 { 1385 int err; 1386 struct fib_info *fi = NULL; 1387 struct nexthop *nh = NULL; 1388 struct fib_info *ofi; 1389 int nhs = 1; 1390 struct net *net = cfg->fc_nlinfo.nl_net; 1391 1392 if (cfg->fc_type > RTN_MAX) 1393 goto err_inval; 1394 1395 /* Fast check to catch the most weird cases */ 1396 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) { 1397 NL_SET_ERR_MSG(extack, "Invalid scope"); 1398 goto err_inval; 1399 } 1400 1401 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 1402 NL_SET_ERR_MSG(extack, 1403 "Invalid rtm_flags - can not contain DEAD or LINKDOWN"); 1404 goto err_inval; 1405 } 1406 1407 if (cfg->fc_nh_id) { 1408 if (!cfg->fc_mx) { 1409 fi = fib_find_info_nh(net, cfg); 1410 if (fi) { 1411 refcount_inc(&fi->fib_treeref); 1412 return fi; 1413 } 1414 } 1415 1416 nh = nexthop_find_by_id(net, cfg->fc_nh_id); 1417 if (!nh) { 1418 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 1419 goto err_inval; 1420 } 1421 nhs = 0; 1422 } 1423 1424 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1425 if (cfg->fc_mp) { 1426 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack); 1427 if (nhs == 0) 1428 goto err_inval; 1429 } 1430 #endif 1431 1432 err = -ENOBUFS; 1433 if (fib_info_cnt >= fib_info_hash_size) { 1434 unsigned int new_size = fib_info_hash_size << 1; 1435 struct hlist_head *new_info_hash; 1436 struct hlist_head *new_laddrhash; 1437 unsigned int bytes; 1438 1439 if (!new_size) 1440 new_size = 16; 1441 bytes = new_size * sizeof(struct hlist_head *); 1442 new_info_hash = fib_info_hash_alloc(bytes); 1443 new_laddrhash = fib_info_hash_alloc(bytes); 1444 if (!new_info_hash || !new_laddrhash) { 1445 fib_info_hash_free(new_info_hash, bytes); 1446 fib_info_hash_free(new_laddrhash, bytes); 1447 } else 1448 fib_info_hash_move(new_info_hash, new_laddrhash, new_size); 1449 1450 if (!fib_info_hash_size) 1451 goto failure; 1452 } 1453 1454 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL); 1455 if (!fi) 1456 goto failure; 1457 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx, 1458 cfg->fc_mx_len, extack); 1459 if (IS_ERR(fi->fib_metrics)) { 1460 err = PTR_ERR(fi->fib_metrics); 1461 kfree(fi); 1462 return ERR_PTR(err); 1463 } 1464 1465 fib_info_cnt++; 1466 fi->fib_net = net; 1467 fi->fib_protocol = cfg->fc_protocol; 1468 fi->fib_scope = cfg->fc_scope; 1469 fi->fib_flags = cfg->fc_flags; 1470 fi->fib_priority = cfg->fc_priority; 1471 fi->fib_prefsrc = cfg->fc_prefsrc; 1472 fi->fib_type = cfg->fc_type; 1473 fi->fib_tb_id = cfg->fc_table; 1474 1475 fi->fib_nhs = nhs; 1476 if (nh) { 1477 if (!nexthop_get(nh)) { 1478 NL_SET_ERR_MSG(extack, "Nexthop has been deleted"); 1479 err = -EINVAL; 1480 } else { 1481 err = 0; 1482 fi->nh = nh; 1483 } 1484 } else { 1485 change_nexthops(fi) { 1486 nexthop_nh->nh_parent = fi; 1487 } endfor_nexthops(fi) 1488 1489 if (cfg->fc_mp) 1490 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, 1491 extack); 1492 else 1493 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack); 1494 } 1495 1496 if (err != 0) 1497 goto failure; 1498 1499 if (fib_props[cfg->fc_type].error) { 1500 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) { 1501 NL_SET_ERR_MSG(extack, 1502 "Gateway, device and multipath can not be specified for this route type"); 1503 goto err_inval; 1504 } 1505 goto link_it; 1506 } else { 1507 switch (cfg->fc_type) { 1508 case RTN_UNICAST: 1509 case RTN_LOCAL: 1510 case RTN_BROADCAST: 1511 case RTN_ANYCAST: 1512 case RTN_MULTICAST: 1513 break; 1514 default: 1515 NL_SET_ERR_MSG(extack, "Invalid route type"); 1516 goto err_inval; 1517 } 1518 } 1519 1520 if (cfg->fc_scope > RT_SCOPE_HOST) { 1521 NL_SET_ERR_MSG(extack, "Invalid scope"); 1522 goto err_inval; 1523 } 1524 1525 if (fi->nh) { 1526 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack); 1527 if (err) 1528 goto failure; 1529 } else if (cfg->fc_scope == RT_SCOPE_HOST) { 1530 struct fib_nh *nh = fi->fib_nh; 1531 1532 /* Local address is added. */ 1533 if (nhs != 1) { 1534 NL_SET_ERR_MSG(extack, 1535 "Route with host scope can not have multiple nexthops"); 1536 goto err_inval; 1537 } 1538 if (nh->fib_nh_gw_family) { 1539 NL_SET_ERR_MSG(extack, 1540 "Route with host scope can not have a gateway"); 1541 goto err_inval; 1542 } 1543 nh->fib_nh_scope = RT_SCOPE_NOWHERE; 1544 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif); 1545 err = -ENODEV; 1546 if (!nh->fib_nh_dev) 1547 goto failure; 1548 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, 1549 GFP_KERNEL); 1550 } else { 1551 int linkdown = 0; 1552 1553 change_nexthops(fi) { 1554 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh, 1555 cfg->fc_table, cfg->fc_scope, 1556 extack); 1557 if (err != 0) 1558 goto failure; 1559 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) 1560 linkdown++; 1561 } endfor_nexthops(fi) 1562 if (linkdown == fi->fib_nhs) 1563 fi->fib_flags |= RTNH_F_LINKDOWN; 1564 } 1565 1566 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) { 1567 NL_SET_ERR_MSG(extack, "Invalid prefsrc address"); 1568 goto err_inval; 1569 } 1570 1571 if (!fi->nh) { 1572 change_nexthops(fi) { 1573 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common, 1574 fi->fib_scope); 1575 if (nexthop_nh->fib_nh_gw_family == AF_INET6) 1576 fi->fib_nh_is_v6 = true; 1577 } endfor_nexthops(fi) 1578 1579 fib_rebalance(fi); 1580 } 1581 1582 link_it: 1583 ofi = fib_find_info(fi); 1584 if (ofi) { 1585 fi->fib_dead = 1; 1586 free_fib_info(fi); 1587 refcount_inc(&ofi->fib_treeref); 1588 return ofi; 1589 } 1590 1591 refcount_set(&fi->fib_treeref, 1); 1592 refcount_set(&fi->fib_clntref, 1); 1593 spin_lock_bh(&fib_info_lock); 1594 hlist_add_head(&fi->fib_hash, 1595 &fib_info_hash[fib_info_hashfn(fi)]); 1596 if (fi->fib_prefsrc) { 1597 struct hlist_head *head; 1598 1599 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)]; 1600 hlist_add_head(&fi->fib_lhash, head); 1601 } 1602 if (fi->nh) { 1603 list_add(&fi->nh_list, &nh->fi_list); 1604 } else { 1605 change_nexthops(fi) { 1606 struct hlist_head *head; 1607 unsigned int hash; 1608 1609 if (!nexthop_nh->fib_nh_dev) 1610 continue; 1611 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex); 1612 head = &fib_info_devhash[hash]; 1613 hlist_add_head(&nexthop_nh->nh_hash, head); 1614 } endfor_nexthops(fi) 1615 } 1616 spin_unlock_bh(&fib_info_lock); 1617 return fi; 1618 1619 err_inval: 1620 err = -EINVAL; 1621 1622 failure: 1623 if (fi) { 1624 fi->fib_dead = 1; 1625 free_fib_info(fi); 1626 } 1627 1628 return ERR_PTR(err); 1629 } 1630 1631 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc, 1632 u8 rt_family, unsigned char *flags, bool skip_oif) 1633 { 1634 if (nhc->nhc_flags & RTNH_F_DEAD) 1635 *flags |= RTNH_F_DEAD; 1636 1637 if (nhc->nhc_flags & RTNH_F_LINKDOWN) { 1638 *flags |= RTNH_F_LINKDOWN; 1639 1640 rcu_read_lock(); 1641 switch (nhc->nhc_family) { 1642 case AF_INET: 1643 if (ip_ignore_linkdown(nhc->nhc_dev)) 1644 *flags |= RTNH_F_DEAD; 1645 break; 1646 case AF_INET6: 1647 if (ip6_ignore_linkdown(nhc->nhc_dev)) 1648 *flags |= RTNH_F_DEAD; 1649 break; 1650 } 1651 rcu_read_unlock(); 1652 } 1653 1654 switch (nhc->nhc_gw_family) { 1655 case AF_INET: 1656 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4)) 1657 goto nla_put_failure; 1658 break; 1659 case AF_INET6: 1660 /* if gateway family does not match nexthop family 1661 * gateway is encoded as RTA_VIA 1662 */ 1663 if (rt_family != nhc->nhc_gw_family) { 1664 int alen = sizeof(struct in6_addr); 1665 struct nlattr *nla; 1666 struct rtvia *via; 1667 1668 nla = nla_reserve(skb, RTA_VIA, alen + 2); 1669 if (!nla) 1670 goto nla_put_failure; 1671 1672 via = nla_data(nla); 1673 via->rtvia_family = AF_INET6; 1674 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen); 1675 } else if (nla_put_in6_addr(skb, RTA_GATEWAY, 1676 &nhc->nhc_gw.ipv6) < 0) { 1677 goto nla_put_failure; 1678 } 1679 break; 1680 } 1681 1682 *flags |= (nhc->nhc_flags & 1683 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP)); 1684 1685 if (!skip_oif && nhc->nhc_dev && 1686 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex)) 1687 goto nla_put_failure; 1688 1689 if (nhc->nhc_lwtstate && 1690 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate, 1691 RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 1692 goto nla_put_failure; 1693 1694 return 0; 1695 1696 nla_put_failure: 1697 return -EMSGSIZE; 1698 } 1699 EXPORT_SYMBOL_GPL(fib_nexthop_info); 1700 1701 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6) 1702 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc, 1703 int nh_weight, u8 rt_family, u32 nh_tclassid) 1704 { 1705 const struct net_device *dev = nhc->nhc_dev; 1706 struct rtnexthop *rtnh; 1707 unsigned char flags = 0; 1708 1709 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1710 if (!rtnh) 1711 goto nla_put_failure; 1712 1713 rtnh->rtnh_hops = nh_weight - 1; 1714 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0; 1715 1716 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0) 1717 goto nla_put_failure; 1718 1719 rtnh->rtnh_flags = flags; 1720 1721 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid)) 1722 goto nla_put_failure; 1723 1724 /* length of rtnetlink header + attributes */ 1725 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh; 1726 1727 return 0; 1728 1729 nla_put_failure: 1730 return -EMSGSIZE; 1731 } 1732 EXPORT_SYMBOL_GPL(fib_add_nexthop); 1733 #endif 1734 1735 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1736 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1737 { 1738 struct nlattr *mp; 1739 1740 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH); 1741 if (!mp) 1742 goto nla_put_failure; 1743 1744 if (unlikely(fi->nh)) { 1745 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0) 1746 goto nla_put_failure; 1747 goto mp_end; 1748 } 1749 1750 for_nexthops(fi) { 1751 u32 nh_tclassid = 0; 1752 #ifdef CONFIG_IP_ROUTE_CLASSID 1753 nh_tclassid = nh->nh_tclassid; 1754 #endif 1755 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight, 1756 AF_INET, nh_tclassid) < 0) 1757 goto nla_put_failure; 1758 } endfor_nexthops(fi); 1759 1760 mp_end: 1761 nla_nest_end(skb, mp); 1762 1763 return 0; 1764 1765 nla_put_failure: 1766 return -EMSGSIZE; 1767 } 1768 #else 1769 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1770 { 1771 return 0; 1772 } 1773 #endif 1774 1775 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 1776 const struct fib_rt_info *fri, unsigned int flags) 1777 { 1778 unsigned int nhs = fib_info_num_path(fri->fi); 1779 struct fib_info *fi = fri->fi; 1780 u32 tb_id = fri->tb_id; 1781 struct nlmsghdr *nlh; 1782 struct rtmsg *rtm; 1783 1784 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 1785 if (!nlh) 1786 return -EMSGSIZE; 1787 1788 rtm = nlmsg_data(nlh); 1789 rtm->rtm_family = AF_INET; 1790 rtm->rtm_dst_len = fri->dst_len; 1791 rtm->rtm_src_len = 0; 1792 rtm->rtm_tos = fri->tos; 1793 if (tb_id < 256) 1794 rtm->rtm_table = tb_id; 1795 else 1796 rtm->rtm_table = RT_TABLE_COMPAT; 1797 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1798 goto nla_put_failure; 1799 rtm->rtm_type = fri->type; 1800 rtm->rtm_flags = fi->fib_flags; 1801 rtm->rtm_scope = fi->fib_scope; 1802 rtm->rtm_protocol = fi->fib_protocol; 1803 1804 if (rtm->rtm_dst_len && 1805 nla_put_in_addr(skb, RTA_DST, fri->dst)) 1806 goto nla_put_failure; 1807 if (fi->fib_priority && 1808 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1809 goto nla_put_failure; 1810 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0) 1811 goto nla_put_failure; 1812 1813 if (fi->fib_prefsrc && 1814 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1815 goto nla_put_failure; 1816 1817 if (fi->nh) { 1818 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id)) 1819 goto nla_put_failure; 1820 if (nexthop_is_blackhole(fi->nh)) 1821 rtm->rtm_type = RTN_BLACKHOLE; 1822 if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode) 1823 goto offload; 1824 } 1825 1826 if (nhs == 1) { 1827 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0); 1828 unsigned char flags = 0; 1829 1830 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0) 1831 goto nla_put_failure; 1832 1833 rtm->rtm_flags = flags; 1834 #ifdef CONFIG_IP_ROUTE_CLASSID 1835 if (nhc->nhc_family == AF_INET) { 1836 struct fib_nh *nh; 1837 1838 nh = container_of(nhc, struct fib_nh, nh_common); 1839 if (nh->nh_tclassid && 1840 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1841 goto nla_put_failure; 1842 } 1843 #endif 1844 } else { 1845 if (fib_add_multipath(skb, fi) < 0) 1846 goto nla_put_failure; 1847 } 1848 1849 offload: 1850 if (fri->offload) 1851 rtm->rtm_flags |= RTM_F_OFFLOAD; 1852 if (fri->trap) 1853 rtm->rtm_flags |= RTM_F_TRAP; 1854 if (fri->offload_failed) 1855 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED; 1856 1857 nlmsg_end(skb, nlh); 1858 return 0; 1859 1860 nla_put_failure: 1861 nlmsg_cancel(skb, nlh); 1862 return -EMSGSIZE; 1863 } 1864 1865 /* 1866 * Update FIB if: 1867 * - local address disappeared -> we must delete all the entries 1868 * referring to it. 1869 * - device went down -> we must shutdown all nexthops going via it. 1870 */ 1871 int fib_sync_down_addr(struct net_device *dev, __be32 local) 1872 { 1873 int ret = 0; 1874 unsigned int hash = fib_laddr_hashfn(local); 1875 struct hlist_head *head = &fib_info_laddrhash[hash]; 1876 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN; 1877 struct net *net = dev_net(dev); 1878 struct fib_info *fi; 1879 1880 if (!fib_info_laddrhash || local == 0) 1881 return 0; 1882 1883 hlist_for_each_entry(fi, head, fib_lhash) { 1884 if (!net_eq(fi->fib_net, net) || 1885 fi->fib_tb_id != tb_id) 1886 continue; 1887 if (fi->fib_prefsrc == local) { 1888 fi->fib_flags |= RTNH_F_DEAD; 1889 ret++; 1890 } 1891 } 1892 return ret; 1893 } 1894 1895 static int call_fib_nh_notifiers(struct fib_nh *nh, 1896 enum fib_event_type event_type) 1897 { 1898 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev); 1899 struct fib_nh_notifier_info info = { 1900 .fib_nh = nh, 1901 }; 1902 1903 switch (event_type) { 1904 case FIB_EVENT_NH_ADD: 1905 if (nh->fib_nh_flags & RTNH_F_DEAD) 1906 break; 1907 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) 1908 break; 1909 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type, 1910 &info.info); 1911 case FIB_EVENT_NH_DEL: 1912 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) || 1913 (nh->fib_nh_flags & RTNH_F_DEAD)) 1914 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), 1915 event_type, &info.info); 1916 break; 1917 default: 1918 break; 1919 } 1920 1921 return NOTIFY_DONE; 1922 } 1923 1924 /* Update the PMTU of exceptions when: 1925 * - the new MTU of the first hop becomes smaller than the PMTU 1926 * - the old MTU was the same as the PMTU, and it limited discovery of 1927 * larger MTUs on the path. With that limit raised, we can now 1928 * discover larger MTUs 1929 * A special case is locked exceptions, for which the PMTU is smaller 1930 * than the minimal accepted PMTU: 1931 * - if the new MTU is greater than the PMTU, don't make any change 1932 * - otherwise, unlock and set PMTU 1933 */ 1934 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig) 1935 { 1936 struct fnhe_hash_bucket *bucket; 1937 int i; 1938 1939 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1); 1940 if (!bucket) 1941 return; 1942 1943 for (i = 0; i < FNHE_HASH_SIZE; i++) { 1944 struct fib_nh_exception *fnhe; 1945 1946 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1); 1947 fnhe; 1948 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) { 1949 if (fnhe->fnhe_mtu_locked) { 1950 if (new <= fnhe->fnhe_pmtu) { 1951 fnhe->fnhe_pmtu = new; 1952 fnhe->fnhe_mtu_locked = false; 1953 } 1954 } else if (new < fnhe->fnhe_pmtu || 1955 orig == fnhe->fnhe_pmtu) { 1956 fnhe->fnhe_pmtu = new; 1957 } 1958 } 1959 } 1960 } 1961 1962 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu) 1963 { 1964 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1965 struct hlist_head *head = &fib_info_devhash[hash]; 1966 struct fib_nh *nh; 1967 1968 hlist_for_each_entry(nh, head, nh_hash) { 1969 if (nh->fib_nh_dev == dev) 1970 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu); 1971 } 1972 } 1973 1974 /* Event force Flags Description 1975 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host 1976 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host 1977 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed 1978 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed 1979 * 1980 * only used when fib_nh is built into fib_info 1981 */ 1982 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force) 1983 { 1984 int ret = 0; 1985 int scope = RT_SCOPE_NOWHERE; 1986 struct fib_info *prev_fi = NULL; 1987 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1988 struct hlist_head *head = &fib_info_devhash[hash]; 1989 struct fib_nh *nh; 1990 1991 if (force) 1992 scope = -1; 1993 1994 hlist_for_each_entry(nh, head, nh_hash) { 1995 struct fib_info *fi = nh->nh_parent; 1996 int dead; 1997 1998 BUG_ON(!fi->fib_nhs); 1999 if (nh->fib_nh_dev != dev || fi == prev_fi) 2000 continue; 2001 prev_fi = fi; 2002 dead = 0; 2003 change_nexthops(fi) { 2004 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) 2005 dead++; 2006 else if (nexthop_nh->fib_nh_dev == dev && 2007 nexthop_nh->fib_nh_scope != scope) { 2008 switch (event) { 2009 case NETDEV_DOWN: 2010 case NETDEV_UNREGISTER: 2011 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD; 2012 fallthrough; 2013 case NETDEV_CHANGE: 2014 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN; 2015 break; 2016 } 2017 call_fib_nh_notifiers(nexthop_nh, 2018 FIB_EVENT_NH_DEL); 2019 dead++; 2020 } 2021 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2022 if (event == NETDEV_UNREGISTER && 2023 nexthop_nh->fib_nh_dev == dev) { 2024 dead = fi->fib_nhs; 2025 break; 2026 } 2027 #endif 2028 } endfor_nexthops(fi) 2029 if (dead == fi->fib_nhs) { 2030 switch (event) { 2031 case NETDEV_DOWN: 2032 case NETDEV_UNREGISTER: 2033 fi->fib_flags |= RTNH_F_DEAD; 2034 fallthrough; 2035 case NETDEV_CHANGE: 2036 fi->fib_flags |= RTNH_F_LINKDOWN; 2037 break; 2038 } 2039 ret++; 2040 } 2041 2042 fib_rebalance(fi); 2043 } 2044 2045 return ret; 2046 } 2047 2048 /* Must be invoked inside of an RCU protected region. */ 2049 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res) 2050 { 2051 struct fib_info *fi = NULL, *last_resort = NULL; 2052 struct hlist_head *fa_head = res->fa_head; 2053 struct fib_table *tb = res->table; 2054 u8 slen = 32 - res->prefixlen; 2055 int order = -1, last_idx = -1; 2056 struct fib_alias *fa, *fa1 = NULL; 2057 u32 last_prio = res->fi->fib_priority; 2058 u8 last_tos = 0; 2059 2060 hlist_for_each_entry_rcu(fa, fa_head, fa_list) { 2061 struct fib_info *next_fi = fa->fa_info; 2062 struct fib_nh_common *nhc; 2063 2064 if (fa->fa_slen != slen) 2065 continue; 2066 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos) 2067 continue; 2068 if (fa->tb_id != tb->tb_id) 2069 continue; 2070 if (next_fi->fib_priority > last_prio && 2071 fa->fa_tos == last_tos) { 2072 if (last_tos) 2073 continue; 2074 break; 2075 } 2076 if (next_fi->fib_flags & RTNH_F_DEAD) 2077 continue; 2078 last_tos = fa->fa_tos; 2079 last_prio = next_fi->fib_priority; 2080 2081 if (next_fi->fib_scope != res->scope || 2082 fa->fa_type != RTN_UNICAST) 2083 continue; 2084 2085 nhc = fib_info_nhc(next_fi, 0); 2086 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK) 2087 continue; 2088 2089 fib_alias_accessed(fa); 2090 2091 if (!fi) { 2092 if (next_fi != res->fi) 2093 break; 2094 fa1 = fa; 2095 } else if (!fib_detect_death(fi, order, &last_resort, 2096 &last_idx, fa1->fa_default)) { 2097 fib_result_assign(res, fi); 2098 fa1->fa_default = order; 2099 goto out; 2100 } 2101 fi = next_fi; 2102 order++; 2103 } 2104 2105 if (order <= 0 || !fi) { 2106 if (fa1) 2107 fa1->fa_default = -1; 2108 goto out; 2109 } 2110 2111 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 2112 fa1->fa_default)) { 2113 fib_result_assign(res, fi); 2114 fa1->fa_default = order; 2115 goto out; 2116 } 2117 2118 if (last_idx >= 0) 2119 fib_result_assign(res, last_resort); 2120 fa1->fa_default = last_idx; 2121 out: 2122 return; 2123 } 2124 2125 /* 2126 * Dead device goes up. We wake up dead nexthops. 2127 * It takes sense only on multipath routes. 2128 * 2129 * only used when fib_nh is built into fib_info 2130 */ 2131 int fib_sync_up(struct net_device *dev, unsigned char nh_flags) 2132 { 2133 struct fib_info *prev_fi; 2134 unsigned int hash; 2135 struct hlist_head *head; 2136 struct fib_nh *nh; 2137 int ret; 2138 2139 if (!(dev->flags & IFF_UP)) 2140 return 0; 2141 2142 if (nh_flags & RTNH_F_DEAD) { 2143 unsigned int flags = dev_get_flags(dev); 2144 2145 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 2146 nh_flags |= RTNH_F_LINKDOWN; 2147 } 2148 2149 prev_fi = NULL; 2150 hash = fib_devindex_hashfn(dev->ifindex); 2151 head = &fib_info_devhash[hash]; 2152 ret = 0; 2153 2154 hlist_for_each_entry(nh, head, nh_hash) { 2155 struct fib_info *fi = nh->nh_parent; 2156 int alive; 2157 2158 BUG_ON(!fi->fib_nhs); 2159 if (nh->fib_nh_dev != dev || fi == prev_fi) 2160 continue; 2161 2162 prev_fi = fi; 2163 alive = 0; 2164 change_nexthops(fi) { 2165 if (!(nexthop_nh->fib_nh_flags & nh_flags)) { 2166 alive++; 2167 continue; 2168 } 2169 if (!nexthop_nh->fib_nh_dev || 2170 !(nexthop_nh->fib_nh_dev->flags & IFF_UP)) 2171 continue; 2172 if (nexthop_nh->fib_nh_dev != dev || 2173 !__in_dev_get_rtnl(dev)) 2174 continue; 2175 alive++; 2176 nexthop_nh->fib_nh_flags &= ~nh_flags; 2177 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD); 2178 } endfor_nexthops(fi) 2179 2180 if (alive > 0) { 2181 fi->fib_flags &= ~nh_flags; 2182 ret++; 2183 } 2184 2185 fib_rebalance(fi); 2186 } 2187 2188 return ret; 2189 } 2190 2191 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2192 static bool fib_good_nh(const struct fib_nh *nh) 2193 { 2194 int state = NUD_REACHABLE; 2195 2196 if (nh->fib_nh_scope == RT_SCOPE_LINK) { 2197 struct neighbour *n; 2198 2199 rcu_read_lock_bh(); 2200 2201 if (likely(nh->fib_nh_gw_family == AF_INET)) 2202 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev, 2203 (__force u32)nh->fib_nh_gw4); 2204 else if (nh->fib_nh_gw_family == AF_INET6) 2205 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, 2206 &nh->fib_nh_gw6); 2207 else 2208 n = NULL; 2209 if (n) 2210 state = n->nud_state; 2211 2212 rcu_read_unlock_bh(); 2213 } 2214 2215 return !!(state & NUD_VALID); 2216 } 2217 2218 void fib_select_multipath(struct fib_result *res, int hash) 2219 { 2220 struct fib_info *fi = res->fi; 2221 struct net *net = fi->fib_net; 2222 bool first = false; 2223 2224 if (unlikely(res->fi->nh)) { 2225 nexthop_path_fib_result(res, hash); 2226 return; 2227 } 2228 2229 change_nexthops(fi) { 2230 if (net->ipv4.sysctl_fib_multipath_use_neigh) { 2231 if (!fib_good_nh(nexthop_nh)) 2232 continue; 2233 if (!first) { 2234 res->nh_sel = nhsel; 2235 res->nhc = &nexthop_nh->nh_common; 2236 first = true; 2237 } 2238 } 2239 2240 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound)) 2241 continue; 2242 2243 res->nh_sel = nhsel; 2244 res->nhc = &nexthop_nh->nh_common; 2245 return; 2246 } endfor_nexthops(fi); 2247 } 2248 #endif 2249 2250 void fib_select_path(struct net *net, struct fib_result *res, 2251 struct flowi4 *fl4, const struct sk_buff *skb) 2252 { 2253 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) 2254 goto check_saddr; 2255 2256 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2257 if (fib_info_num_path(res->fi) > 1) { 2258 int h = fib_multipath_hash(net, fl4, skb, NULL); 2259 2260 fib_select_multipath(res, h); 2261 } 2262 else 2263 #endif 2264 if (!res->prefixlen && 2265 res->table->tb_num_default > 1 && 2266 res->type == RTN_UNICAST) 2267 fib_select_default(fl4, res); 2268 2269 check_saddr: 2270 if (!fl4->saddr) 2271 fl4->saddr = fib_result_prefsrc(net, res); 2272 } 2273