1 // SPDX-License-Identifier: GPL-2.0 2 /* Generic nexthop implementation 3 * 4 * Copyright (c) 2017-19 Cumulus Networks 5 * Copyright (c) 2017-19 David Ahern <dsa@cumulusnetworks.com> 6 */ 7 8 #include <linux/nexthop.h> 9 #include <linux/rtnetlink.h> 10 #include <linux/slab.h> 11 #include <net/arp.h> 12 #include <net/ipv6_stubs.h> 13 #include <net/lwtunnel.h> 14 #include <net/ndisc.h> 15 #include <net/nexthop.h> 16 #include <net/route.h> 17 #include <net/sock.h> 18 19 static void remove_nexthop(struct net *net, struct nexthop *nh, 20 struct nl_info *nlinfo); 21 22 #define NH_DEV_HASHBITS 8 23 #define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS) 24 25 static const struct nla_policy rtm_nh_policy[NHA_MAX + 1] = { 26 [NHA_ID] = { .type = NLA_U32 }, 27 [NHA_GROUP] = { .type = NLA_BINARY }, 28 [NHA_GROUP_TYPE] = { .type = NLA_U16 }, 29 [NHA_BLACKHOLE] = { .type = NLA_FLAG }, 30 [NHA_OIF] = { .type = NLA_U32 }, 31 [NHA_GATEWAY] = { .type = NLA_BINARY }, 32 [NHA_ENCAP_TYPE] = { .type = NLA_U16 }, 33 [NHA_ENCAP] = { .type = NLA_NESTED }, 34 [NHA_GROUPS] = { .type = NLA_FLAG }, 35 [NHA_MASTER] = { .type = NLA_U32 }, 36 [NHA_FDB] = { .type = NLA_FLAG }, 37 }; 38 39 static int call_nexthop_notifiers(struct net *net, 40 enum nexthop_event_type event_type, 41 struct nexthop *nh) 42 { 43 int err; 44 45 err = atomic_notifier_call_chain(&net->nexthop.notifier_chain, 46 event_type, nh); 47 return notifier_to_errno(err); 48 } 49 50 static unsigned int nh_dev_hashfn(unsigned int val) 51 { 52 unsigned int mask = NH_DEV_HASHSIZE - 1; 53 54 return (val ^ 55 (val >> NH_DEV_HASHBITS) ^ 56 (val >> (NH_DEV_HASHBITS * 2))) & mask; 57 } 58 59 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi) 60 { 61 struct net_device *dev = nhi->fib_nhc.nhc_dev; 62 struct hlist_head *head; 63 unsigned int hash; 64 65 WARN_ON(!dev); 66 67 hash = nh_dev_hashfn(dev->ifindex); 68 head = &net->nexthop.devhash[hash]; 69 hlist_add_head(&nhi->dev_hash, head); 70 } 71 72 static void nexthop_free_mpath(struct nexthop *nh) 73 { 74 struct nh_group *nhg; 75 int i; 76 77 nhg = rcu_dereference_raw(nh->nh_grp); 78 for (i = 0; i < nhg->num_nh; ++i) { 79 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 80 81 WARN_ON(!list_empty(&nhge->nh_list)); 82 nexthop_put(nhge->nh); 83 } 84 85 WARN_ON(nhg->spare == nhg); 86 87 kfree(nhg->spare); 88 kfree(nhg); 89 } 90 91 static void nexthop_free_single(struct nexthop *nh) 92 { 93 struct nh_info *nhi; 94 95 nhi = rcu_dereference_raw(nh->nh_info); 96 switch (nhi->family) { 97 case AF_INET: 98 fib_nh_release(nh->net, &nhi->fib_nh); 99 break; 100 case AF_INET6: 101 ipv6_stub->fib6_nh_release(&nhi->fib6_nh); 102 break; 103 } 104 kfree(nhi); 105 } 106 107 void nexthop_free_rcu(struct rcu_head *head) 108 { 109 struct nexthop *nh = container_of(head, struct nexthop, rcu); 110 111 if (nh->is_group) 112 nexthop_free_mpath(nh); 113 else 114 nexthop_free_single(nh); 115 116 kfree(nh); 117 } 118 EXPORT_SYMBOL_GPL(nexthop_free_rcu); 119 120 static struct nexthop *nexthop_alloc(void) 121 { 122 struct nexthop *nh; 123 124 nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL); 125 if (nh) { 126 INIT_LIST_HEAD(&nh->fi_list); 127 INIT_LIST_HEAD(&nh->f6i_list); 128 INIT_LIST_HEAD(&nh->grp_list); 129 INIT_LIST_HEAD(&nh->fdb_list); 130 } 131 return nh; 132 } 133 134 static struct nh_group *nexthop_grp_alloc(u16 num_nh) 135 { 136 struct nh_group *nhg; 137 138 nhg = kzalloc(struct_size(nhg, nh_entries, num_nh), GFP_KERNEL); 139 if (nhg) 140 nhg->num_nh = num_nh; 141 142 return nhg; 143 } 144 145 static void nh_base_seq_inc(struct net *net) 146 { 147 while (++net->nexthop.seq == 0) 148 ; 149 } 150 151 /* no reference taken; rcu lock or rtnl must be held */ 152 struct nexthop *nexthop_find_by_id(struct net *net, u32 id) 153 { 154 struct rb_node **pp, *parent = NULL, *next; 155 156 pp = &net->nexthop.rb_root.rb_node; 157 while (1) { 158 struct nexthop *nh; 159 160 next = rcu_dereference_raw(*pp); 161 if (!next) 162 break; 163 parent = next; 164 165 nh = rb_entry(parent, struct nexthop, rb_node); 166 if (id < nh->id) 167 pp = &next->rb_left; 168 else if (id > nh->id) 169 pp = &next->rb_right; 170 else 171 return nh; 172 } 173 return NULL; 174 } 175 EXPORT_SYMBOL_GPL(nexthop_find_by_id); 176 177 /* used for auto id allocation; called with rtnl held */ 178 static u32 nh_find_unused_id(struct net *net) 179 { 180 u32 id_start = net->nexthop.last_id_allocated; 181 182 while (1) { 183 net->nexthop.last_id_allocated++; 184 if (net->nexthop.last_id_allocated == id_start) 185 break; 186 187 if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated)) 188 return net->nexthop.last_id_allocated; 189 } 190 return 0; 191 } 192 193 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg) 194 { 195 struct nexthop_grp *p; 196 size_t len = nhg->num_nh * sizeof(*p); 197 struct nlattr *nla; 198 u16 group_type = 0; 199 int i; 200 201 if (nhg->mpath) 202 group_type = NEXTHOP_GRP_TYPE_MPATH; 203 204 if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type)) 205 goto nla_put_failure; 206 207 nla = nla_reserve(skb, NHA_GROUP, len); 208 if (!nla) 209 goto nla_put_failure; 210 211 p = nla_data(nla); 212 for (i = 0; i < nhg->num_nh; ++i) { 213 p->id = nhg->nh_entries[i].nh->id; 214 p->weight = nhg->nh_entries[i].weight - 1; 215 p += 1; 216 } 217 218 return 0; 219 220 nla_put_failure: 221 return -EMSGSIZE; 222 } 223 224 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh, 225 int event, u32 portid, u32 seq, unsigned int nlflags) 226 { 227 struct fib6_nh *fib6_nh; 228 struct fib_nh *fib_nh; 229 struct nlmsghdr *nlh; 230 struct nh_info *nhi; 231 struct nhmsg *nhm; 232 233 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags); 234 if (!nlh) 235 return -EMSGSIZE; 236 237 nhm = nlmsg_data(nlh); 238 nhm->nh_family = AF_UNSPEC; 239 nhm->nh_flags = nh->nh_flags; 240 nhm->nh_protocol = nh->protocol; 241 nhm->nh_scope = 0; 242 nhm->resvd = 0; 243 244 if (nla_put_u32(skb, NHA_ID, nh->id)) 245 goto nla_put_failure; 246 247 if (nh->is_group) { 248 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 249 250 if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB)) 251 goto nla_put_failure; 252 if (nla_put_nh_group(skb, nhg)) 253 goto nla_put_failure; 254 goto out; 255 } 256 257 nhi = rtnl_dereference(nh->nh_info); 258 nhm->nh_family = nhi->family; 259 if (nhi->reject_nh) { 260 if (nla_put_flag(skb, NHA_BLACKHOLE)) 261 goto nla_put_failure; 262 goto out; 263 } else if (nhi->fdb_nh) { 264 if (nla_put_flag(skb, NHA_FDB)) 265 goto nla_put_failure; 266 } else { 267 const struct net_device *dev; 268 269 dev = nhi->fib_nhc.nhc_dev; 270 if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex)) 271 goto nla_put_failure; 272 } 273 274 nhm->nh_scope = nhi->fib_nhc.nhc_scope; 275 switch (nhi->family) { 276 case AF_INET: 277 fib_nh = &nhi->fib_nh; 278 if (fib_nh->fib_nh_gw_family && 279 nla_put_be32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4)) 280 goto nla_put_failure; 281 break; 282 283 case AF_INET6: 284 fib6_nh = &nhi->fib6_nh; 285 if (fib6_nh->fib_nh_gw_family && 286 nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6)) 287 goto nla_put_failure; 288 break; 289 } 290 291 if (nhi->fib_nhc.nhc_lwtstate && 292 lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate, 293 NHA_ENCAP, NHA_ENCAP_TYPE) < 0) 294 goto nla_put_failure; 295 296 out: 297 nlmsg_end(skb, nlh); 298 return 0; 299 300 nla_put_failure: 301 nlmsg_cancel(skb, nlh); 302 return -EMSGSIZE; 303 } 304 305 static size_t nh_nlmsg_size_grp(struct nexthop *nh) 306 { 307 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 308 size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh; 309 310 return nla_total_size(sz) + 311 nla_total_size(2); /* NHA_GROUP_TYPE */ 312 } 313 314 static size_t nh_nlmsg_size_single(struct nexthop *nh) 315 { 316 struct nh_info *nhi = rtnl_dereference(nh->nh_info); 317 size_t sz; 318 319 /* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE 320 * are mutually exclusive 321 */ 322 sz = nla_total_size(4); /* NHA_OIF */ 323 324 switch (nhi->family) { 325 case AF_INET: 326 if (nhi->fib_nh.fib_nh_gw_family) 327 sz += nla_total_size(4); /* NHA_GATEWAY */ 328 break; 329 330 case AF_INET6: 331 /* NHA_GATEWAY */ 332 if (nhi->fib6_nh.fib_nh_gw_family) 333 sz += nla_total_size(sizeof(const struct in6_addr)); 334 break; 335 } 336 337 if (nhi->fib_nhc.nhc_lwtstate) { 338 sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate); 339 sz += nla_total_size(2); /* NHA_ENCAP_TYPE */ 340 } 341 342 return sz; 343 } 344 345 static size_t nh_nlmsg_size(struct nexthop *nh) 346 { 347 size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg)); 348 349 sz += nla_total_size(4); /* NHA_ID */ 350 351 if (nh->is_group) 352 sz += nh_nlmsg_size_grp(nh); 353 else 354 sz += nh_nlmsg_size_single(nh); 355 356 return sz; 357 } 358 359 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info) 360 { 361 unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0; 362 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 363 struct sk_buff *skb; 364 int err = -ENOBUFS; 365 366 skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any()); 367 if (!skb) 368 goto errout; 369 370 err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags); 371 if (err < 0) { 372 /* -EMSGSIZE implies BUG in nh_nlmsg_size() */ 373 WARN_ON(err == -EMSGSIZE); 374 kfree_skb(skb); 375 goto errout; 376 } 377 378 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP, 379 info->nlh, gfp_any()); 380 return; 381 errout: 382 if (err < 0) 383 rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err); 384 } 385 386 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths, 387 bool *is_fdb, struct netlink_ext_ack *extack) 388 { 389 if (nh->is_group) { 390 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 391 392 /* nested multipath (group within a group) is not 393 * supported 394 */ 395 if (nhg->mpath) { 396 NL_SET_ERR_MSG(extack, 397 "Multipath group can not be a nexthop within a group"); 398 return false; 399 } 400 *is_fdb = nhg->fdb_nh; 401 } else { 402 struct nh_info *nhi = rtnl_dereference(nh->nh_info); 403 404 if (nhi->reject_nh && npaths > 1) { 405 NL_SET_ERR_MSG(extack, 406 "Blackhole nexthop can not be used in a group with more than 1 path"); 407 return false; 408 } 409 *is_fdb = nhi->fdb_nh; 410 } 411 412 return true; 413 } 414 415 static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family, 416 struct netlink_ext_ack *extack) 417 { 418 struct nh_info *nhi; 419 420 nhi = rtnl_dereference(nh->nh_info); 421 422 if (!nhi->fdb_nh) { 423 NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops"); 424 return -EINVAL; 425 } 426 427 if (*nh_family == AF_UNSPEC) { 428 *nh_family = nhi->family; 429 } else if (*nh_family != nhi->family) { 430 NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops"); 431 return -EINVAL; 432 } 433 434 return 0; 435 } 436 437 static int nh_check_attr_group(struct net *net, struct nlattr *tb[], 438 struct netlink_ext_ack *extack) 439 { 440 unsigned int len = nla_len(tb[NHA_GROUP]); 441 u8 nh_family = AF_UNSPEC; 442 struct nexthop_grp *nhg; 443 unsigned int i, j; 444 u8 nhg_fdb = 0; 445 446 if (!len || len & (sizeof(struct nexthop_grp) - 1)) { 447 NL_SET_ERR_MSG(extack, 448 "Invalid length for nexthop group attribute"); 449 return -EINVAL; 450 } 451 452 /* convert len to number of nexthop ids */ 453 len /= sizeof(*nhg); 454 455 nhg = nla_data(tb[NHA_GROUP]); 456 for (i = 0; i < len; ++i) { 457 if (nhg[i].resvd1 || nhg[i].resvd2) { 458 NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0"); 459 return -EINVAL; 460 } 461 if (nhg[i].weight > 254) { 462 NL_SET_ERR_MSG(extack, "Invalid value for weight"); 463 return -EINVAL; 464 } 465 for (j = i + 1; j < len; ++j) { 466 if (nhg[i].id == nhg[j].id) { 467 NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group"); 468 return -EINVAL; 469 } 470 } 471 } 472 473 if (tb[NHA_FDB]) 474 nhg_fdb = 1; 475 nhg = nla_data(tb[NHA_GROUP]); 476 for (i = 0; i < len; ++i) { 477 struct nexthop *nh; 478 bool is_fdb_nh; 479 480 nh = nexthop_find_by_id(net, nhg[i].id); 481 if (!nh) { 482 NL_SET_ERR_MSG(extack, "Invalid nexthop id"); 483 return -EINVAL; 484 } 485 if (!valid_group_nh(nh, len, &is_fdb_nh, extack)) 486 return -EINVAL; 487 488 if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack)) 489 return -EINVAL; 490 491 if (!nhg_fdb && is_fdb_nh) { 492 NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops"); 493 return -EINVAL; 494 } 495 } 496 for (i = NHA_GROUP_TYPE + 1; i < __NHA_MAX; ++i) { 497 if (!tb[i]) 498 continue; 499 if (tb[NHA_FDB]) 500 continue; 501 NL_SET_ERR_MSG(extack, 502 "No other attributes can be set in nexthop groups"); 503 return -EINVAL; 504 } 505 506 return 0; 507 } 508 509 static bool ipv6_good_nh(const struct fib6_nh *nh) 510 { 511 int state = NUD_REACHABLE; 512 struct neighbour *n; 513 514 rcu_read_lock_bh(); 515 516 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6); 517 if (n) 518 state = n->nud_state; 519 520 rcu_read_unlock_bh(); 521 522 return !!(state & NUD_VALID); 523 } 524 525 static bool ipv4_good_nh(const struct fib_nh *nh) 526 { 527 int state = NUD_REACHABLE; 528 struct neighbour *n; 529 530 rcu_read_lock_bh(); 531 532 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev, 533 (__force u32)nh->fib_nh_gw4); 534 if (n) 535 state = n->nud_state; 536 537 rcu_read_unlock_bh(); 538 539 return !!(state & NUD_VALID); 540 } 541 542 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash) 543 { 544 struct nexthop *rc = NULL; 545 struct nh_group *nhg; 546 int i; 547 548 if (!nh->is_group) 549 return nh; 550 551 nhg = rcu_dereference(nh->nh_grp); 552 for (i = 0; i < nhg->num_nh; ++i) { 553 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 554 struct nh_info *nhi; 555 556 if (hash > atomic_read(&nhge->upper_bound)) 557 continue; 558 559 nhi = rcu_dereference(nhge->nh->nh_info); 560 if (nhi->fdb_nh) 561 return nhge->nh; 562 563 /* nexthops always check if it is good and does 564 * not rely on a sysctl for this behavior 565 */ 566 switch (nhi->family) { 567 case AF_INET: 568 if (ipv4_good_nh(&nhi->fib_nh)) 569 return nhge->nh; 570 break; 571 case AF_INET6: 572 if (ipv6_good_nh(&nhi->fib6_nh)) 573 return nhge->nh; 574 break; 575 } 576 577 if (!rc) 578 rc = nhge->nh; 579 } 580 581 return rc; 582 } 583 EXPORT_SYMBOL_GPL(nexthop_select_path); 584 585 int nexthop_for_each_fib6_nh(struct nexthop *nh, 586 int (*cb)(struct fib6_nh *nh, void *arg), 587 void *arg) 588 { 589 struct nh_info *nhi; 590 int err; 591 592 if (nh->is_group) { 593 struct nh_group *nhg; 594 int i; 595 596 nhg = rcu_dereference_rtnl(nh->nh_grp); 597 for (i = 0; i < nhg->num_nh; i++) { 598 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 599 600 nhi = rcu_dereference_rtnl(nhge->nh->nh_info); 601 err = cb(&nhi->fib6_nh, arg); 602 if (err) 603 return err; 604 } 605 } else { 606 nhi = rcu_dereference_rtnl(nh->nh_info); 607 err = cb(&nhi->fib6_nh, arg); 608 if (err) 609 return err; 610 } 611 612 return 0; 613 } 614 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh); 615 616 static int check_src_addr(const struct in6_addr *saddr, 617 struct netlink_ext_ack *extack) 618 { 619 if (!ipv6_addr_any(saddr)) { 620 NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects"); 621 return -EINVAL; 622 } 623 return 0; 624 } 625 626 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg, 627 struct netlink_ext_ack *extack) 628 { 629 struct nh_info *nhi; 630 bool is_fdb_nh; 631 632 /* fib6_src is unique to a fib6_info and limits the ability to cache 633 * routes in fib6_nh within a nexthop that is potentially shared 634 * across multiple fib entries. If the config wants to use source 635 * routing it can not use nexthop objects. mlxsw also does not allow 636 * fib6_src on routes. 637 */ 638 if (cfg && check_src_addr(&cfg->fc_src, extack) < 0) 639 return -EINVAL; 640 641 if (nh->is_group) { 642 struct nh_group *nhg; 643 644 nhg = rtnl_dereference(nh->nh_grp); 645 if (nhg->has_v4) 646 goto no_v4_nh; 647 is_fdb_nh = nhg->fdb_nh; 648 } else { 649 nhi = rtnl_dereference(nh->nh_info); 650 if (nhi->family == AF_INET) 651 goto no_v4_nh; 652 is_fdb_nh = nhi->fdb_nh; 653 } 654 655 if (is_fdb_nh) { 656 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop"); 657 return -EINVAL; 658 } 659 660 return 0; 661 no_v4_nh: 662 NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop"); 663 return -EINVAL; 664 } 665 EXPORT_SYMBOL_GPL(fib6_check_nexthop); 666 667 /* if existing nexthop has ipv6 routes linked to it, need 668 * to verify this new spec works with ipv6 669 */ 670 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new, 671 struct netlink_ext_ack *extack) 672 { 673 struct fib6_info *f6i; 674 675 if (list_empty(&old->f6i_list)) 676 return 0; 677 678 list_for_each_entry(f6i, &old->f6i_list, nh_list) { 679 if (check_src_addr(&f6i->fib6_src.addr, extack) < 0) 680 return -EINVAL; 681 } 682 683 return fib6_check_nexthop(new, NULL, extack); 684 } 685 686 static int nexthop_check_scope(struct nh_info *nhi, u8 scope, 687 struct netlink_ext_ack *extack) 688 { 689 if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) { 690 NL_SET_ERR_MSG(extack, 691 "Route with host scope can not have a gateway"); 692 return -EINVAL; 693 } 694 695 if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) { 696 NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop"); 697 return -EINVAL; 698 } 699 700 return 0; 701 } 702 703 /* Invoked by fib add code to verify nexthop by id is ok with 704 * config for prefix; parts of fib_check_nh not done when nexthop 705 * object is used. 706 */ 707 int fib_check_nexthop(struct nexthop *nh, u8 scope, 708 struct netlink_ext_ack *extack) 709 { 710 struct nh_info *nhi; 711 int err = 0; 712 713 if (nh->is_group) { 714 struct nh_group *nhg; 715 716 nhg = rtnl_dereference(nh->nh_grp); 717 if (nhg->fdb_nh) { 718 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop"); 719 err = -EINVAL; 720 goto out; 721 } 722 723 if (scope == RT_SCOPE_HOST) { 724 NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops"); 725 err = -EINVAL; 726 goto out; 727 } 728 729 /* all nexthops in a group have the same scope */ 730 nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info); 731 err = nexthop_check_scope(nhi, scope, extack); 732 } else { 733 nhi = rtnl_dereference(nh->nh_info); 734 if (nhi->fdb_nh) { 735 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop"); 736 err = -EINVAL; 737 goto out; 738 } 739 err = nexthop_check_scope(nhi, scope, extack); 740 } 741 742 out: 743 return err; 744 } 745 746 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new, 747 struct netlink_ext_ack *extack) 748 { 749 struct fib_info *fi; 750 751 list_for_each_entry(fi, &old->fi_list, nh_list) { 752 int err; 753 754 err = fib_check_nexthop(new, fi->fib_scope, extack); 755 if (err) 756 return err; 757 } 758 return 0; 759 } 760 761 static void nh_group_rebalance(struct nh_group *nhg) 762 { 763 int total = 0; 764 int w = 0; 765 int i; 766 767 for (i = 0; i < nhg->num_nh; ++i) 768 total += nhg->nh_entries[i].weight; 769 770 for (i = 0; i < nhg->num_nh; ++i) { 771 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 772 int upper_bound; 773 774 w += nhge->weight; 775 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1; 776 atomic_set(&nhge->upper_bound, upper_bound); 777 } 778 } 779 780 static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge, 781 struct nl_info *nlinfo) 782 { 783 struct nh_grp_entry *nhges, *new_nhges; 784 struct nexthop *nhp = nhge->nh_parent; 785 struct nexthop *nh = nhge->nh; 786 struct nh_group *nhg, *newg; 787 int i, j; 788 789 WARN_ON(!nh); 790 791 nhg = rtnl_dereference(nhp->nh_grp); 792 newg = nhg->spare; 793 794 /* last entry, keep it visible and remove the parent */ 795 if (nhg->num_nh == 1) { 796 remove_nexthop(net, nhp, nlinfo); 797 return; 798 } 799 800 newg->has_v4 = false; 801 newg->mpath = nhg->mpath; 802 newg->fdb_nh = nhg->fdb_nh; 803 newg->num_nh = nhg->num_nh; 804 805 /* copy old entries to new except the one getting removed */ 806 nhges = nhg->nh_entries; 807 new_nhges = newg->nh_entries; 808 for (i = 0, j = 0; i < nhg->num_nh; ++i) { 809 struct nh_info *nhi; 810 811 /* current nexthop getting removed */ 812 if (nhg->nh_entries[i].nh == nh) { 813 newg->num_nh--; 814 continue; 815 } 816 817 nhi = rtnl_dereference(nhges[i].nh->nh_info); 818 if (nhi->family == AF_INET) 819 newg->has_v4 = true; 820 821 list_del(&nhges[i].nh_list); 822 new_nhges[j].nh_parent = nhges[i].nh_parent; 823 new_nhges[j].nh = nhges[i].nh; 824 new_nhges[j].weight = nhges[i].weight; 825 list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list); 826 j++; 827 } 828 829 nh_group_rebalance(newg); 830 rcu_assign_pointer(nhp->nh_grp, newg); 831 832 list_del(&nhge->nh_list); 833 nexthop_put(nhge->nh); 834 835 if (nlinfo) 836 nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo); 837 } 838 839 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh, 840 struct nl_info *nlinfo) 841 { 842 struct nh_grp_entry *nhge, *tmp; 843 844 list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list) 845 remove_nh_grp_entry(net, nhge, nlinfo); 846 847 /* make sure all see the newly published array before releasing rtnl */ 848 synchronize_rcu(); 849 } 850 851 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo) 852 { 853 struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp); 854 int i, num_nh = nhg->num_nh; 855 856 for (i = 0; i < num_nh; ++i) { 857 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 858 859 if (WARN_ON(!nhge->nh)) 860 continue; 861 862 list_del_init(&nhge->nh_list); 863 } 864 } 865 866 /* not called for nexthop replace */ 867 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh) 868 { 869 struct fib6_info *f6i, *tmp; 870 bool do_flush = false; 871 struct fib_info *fi; 872 873 call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh); 874 875 list_for_each_entry(fi, &nh->fi_list, nh_list) { 876 fi->fib_flags |= RTNH_F_DEAD; 877 do_flush = true; 878 } 879 if (do_flush) 880 fib_flush(net); 881 882 /* ip6_del_rt removes the entry from this list hence the _safe */ 883 list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) { 884 /* __ip6_del_rt does a release, so do a hold here */ 885 fib6_info_hold(f6i); 886 ipv6_stub->ip6_del_rt(net, f6i, 887 !net->ipv4.sysctl_nexthop_compat_mode); 888 } 889 } 890 891 static void __remove_nexthop(struct net *net, struct nexthop *nh, 892 struct nl_info *nlinfo) 893 { 894 __remove_nexthop_fib(net, nh); 895 896 if (nh->is_group) { 897 remove_nexthop_group(nh, nlinfo); 898 } else { 899 struct nh_info *nhi; 900 901 nhi = rtnl_dereference(nh->nh_info); 902 if (nhi->fib_nhc.nhc_dev) 903 hlist_del(&nhi->dev_hash); 904 905 remove_nexthop_from_groups(net, nh, nlinfo); 906 } 907 } 908 909 static void remove_nexthop(struct net *net, struct nexthop *nh, 910 struct nl_info *nlinfo) 911 { 912 /* remove from the tree */ 913 rb_erase(&nh->rb_node, &net->nexthop.rb_root); 914 915 if (nlinfo) 916 nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo); 917 918 __remove_nexthop(net, nh, nlinfo); 919 nh_base_seq_inc(net); 920 921 nexthop_put(nh); 922 } 923 924 /* if any FIB entries reference this nexthop, any dst entries 925 * need to be regenerated 926 */ 927 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh) 928 { 929 struct fib6_info *f6i; 930 931 if (!list_empty(&nh->fi_list)) 932 rt_cache_flush(net); 933 934 list_for_each_entry(f6i, &nh->f6i_list, nh_list) 935 ipv6_stub->fib6_update_sernum(net, f6i); 936 } 937 938 static int replace_nexthop_grp(struct net *net, struct nexthop *old, 939 struct nexthop *new, 940 struct netlink_ext_ack *extack) 941 { 942 struct nh_group *oldg, *newg; 943 int i; 944 945 if (!new->is_group) { 946 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop."); 947 return -EINVAL; 948 } 949 950 oldg = rtnl_dereference(old->nh_grp); 951 newg = rtnl_dereference(new->nh_grp); 952 953 /* update parents - used by nexthop code for cleanup */ 954 for (i = 0; i < newg->num_nh; i++) 955 newg->nh_entries[i].nh_parent = old; 956 957 rcu_assign_pointer(old->nh_grp, newg); 958 959 for (i = 0; i < oldg->num_nh; i++) 960 oldg->nh_entries[i].nh_parent = new; 961 962 rcu_assign_pointer(new->nh_grp, oldg); 963 964 return 0; 965 } 966 967 static void nh_group_v4_update(struct nh_group *nhg) 968 { 969 struct nh_grp_entry *nhges; 970 bool has_v4 = false; 971 int i; 972 973 nhges = nhg->nh_entries; 974 for (i = 0; i < nhg->num_nh; i++) { 975 struct nh_info *nhi; 976 977 nhi = rtnl_dereference(nhges[i].nh->nh_info); 978 if (nhi->family == AF_INET) 979 has_v4 = true; 980 } 981 nhg->has_v4 = has_v4; 982 } 983 984 static int replace_nexthop_single(struct net *net, struct nexthop *old, 985 struct nexthop *new, 986 struct netlink_ext_ack *extack) 987 { 988 struct nh_info *oldi, *newi; 989 990 if (new->is_group) { 991 NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group."); 992 return -EINVAL; 993 } 994 995 oldi = rtnl_dereference(old->nh_info); 996 newi = rtnl_dereference(new->nh_info); 997 998 newi->nh_parent = old; 999 oldi->nh_parent = new; 1000 1001 old->protocol = new->protocol; 1002 old->nh_flags = new->nh_flags; 1003 1004 rcu_assign_pointer(old->nh_info, newi); 1005 rcu_assign_pointer(new->nh_info, oldi); 1006 1007 /* When replacing an IPv4 nexthop with an IPv6 nexthop, potentially 1008 * update IPv4 indication in all the groups using the nexthop. 1009 */ 1010 if (oldi->family == AF_INET && newi->family == AF_INET6) { 1011 struct nh_grp_entry *nhge; 1012 1013 list_for_each_entry(nhge, &old->grp_list, nh_list) { 1014 struct nexthop *nhp = nhge->nh_parent; 1015 struct nh_group *nhg; 1016 1017 nhg = rtnl_dereference(nhp->nh_grp); 1018 nh_group_v4_update(nhg); 1019 } 1020 } 1021 1022 return 0; 1023 } 1024 1025 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh, 1026 struct nl_info *info) 1027 { 1028 struct fib6_info *f6i; 1029 1030 if (!list_empty(&nh->fi_list)) { 1031 struct fib_info *fi; 1032 1033 /* expectation is a few fib_info per nexthop and then 1034 * a lot of routes per fib_info. So mark the fib_info 1035 * and then walk the fib tables once 1036 */ 1037 list_for_each_entry(fi, &nh->fi_list, nh_list) 1038 fi->nh_updated = true; 1039 1040 fib_info_notify_update(net, info); 1041 1042 list_for_each_entry(fi, &nh->fi_list, nh_list) 1043 fi->nh_updated = false; 1044 } 1045 1046 list_for_each_entry(f6i, &nh->f6i_list, nh_list) 1047 ipv6_stub->fib6_rt_update(net, f6i, info); 1048 } 1049 1050 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries 1051 * linked to this nexthop and for all groups that the nexthop 1052 * is a member of 1053 */ 1054 static void nexthop_replace_notify(struct net *net, struct nexthop *nh, 1055 struct nl_info *info) 1056 { 1057 struct nh_grp_entry *nhge; 1058 1059 __nexthop_replace_notify(net, nh, info); 1060 1061 list_for_each_entry(nhge, &nh->grp_list, nh_list) 1062 __nexthop_replace_notify(net, nhge->nh_parent, info); 1063 } 1064 1065 static int replace_nexthop(struct net *net, struct nexthop *old, 1066 struct nexthop *new, struct netlink_ext_ack *extack) 1067 { 1068 bool new_is_reject = false; 1069 struct nh_grp_entry *nhge; 1070 int err; 1071 1072 /* check that existing FIB entries are ok with the 1073 * new nexthop definition 1074 */ 1075 err = fib_check_nh_list(old, new, extack); 1076 if (err) 1077 return err; 1078 1079 err = fib6_check_nh_list(old, new, extack); 1080 if (err) 1081 return err; 1082 1083 if (!new->is_group) { 1084 struct nh_info *nhi = rtnl_dereference(new->nh_info); 1085 1086 new_is_reject = nhi->reject_nh; 1087 } 1088 1089 list_for_each_entry(nhge, &old->grp_list, nh_list) { 1090 /* if new nexthop is a blackhole, any groups using this 1091 * nexthop cannot have more than 1 path 1092 */ 1093 if (new_is_reject && 1094 nexthop_num_path(nhge->nh_parent) > 1) { 1095 NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path"); 1096 return -EINVAL; 1097 } 1098 1099 err = fib_check_nh_list(nhge->nh_parent, new, extack); 1100 if (err) 1101 return err; 1102 1103 err = fib6_check_nh_list(nhge->nh_parent, new, extack); 1104 if (err) 1105 return err; 1106 } 1107 1108 if (old->is_group) 1109 err = replace_nexthop_grp(net, old, new, extack); 1110 else 1111 err = replace_nexthop_single(net, old, new, extack); 1112 1113 if (!err) { 1114 nh_rt_cache_flush(net, old); 1115 1116 __remove_nexthop(net, new, NULL); 1117 nexthop_put(new); 1118 } 1119 1120 return err; 1121 } 1122 1123 /* called with rtnl_lock held */ 1124 static int insert_nexthop(struct net *net, struct nexthop *new_nh, 1125 struct nh_config *cfg, struct netlink_ext_ack *extack) 1126 { 1127 struct rb_node **pp, *parent = NULL, *next; 1128 struct rb_root *root = &net->nexthop.rb_root; 1129 bool replace = !!(cfg->nlflags & NLM_F_REPLACE); 1130 bool create = !!(cfg->nlflags & NLM_F_CREATE); 1131 u32 new_id = new_nh->id; 1132 int replace_notify = 0; 1133 int rc = -EEXIST; 1134 1135 pp = &root->rb_node; 1136 while (1) { 1137 struct nexthop *nh; 1138 1139 next = *pp; 1140 if (!next) 1141 break; 1142 1143 parent = next; 1144 1145 nh = rb_entry(parent, struct nexthop, rb_node); 1146 if (new_id < nh->id) { 1147 pp = &next->rb_left; 1148 } else if (new_id > nh->id) { 1149 pp = &next->rb_right; 1150 } else if (replace) { 1151 rc = replace_nexthop(net, nh, new_nh, extack); 1152 if (!rc) { 1153 new_nh = nh; /* send notification with old nh */ 1154 replace_notify = 1; 1155 } 1156 goto out; 1157 } else { 1158 /* id already exists and not a replace */ 1159 goto out; 1160 } 1161 } 1162 1163 if (replace && !create) { 1164 NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists"); 1165 rc = -ENOENT; 1166 goto out; 1167 } 1168 1169 rb_link_node_rcu(&new_nh->rb_node, parent, pp); 1170 rb_insert_color(&new_nh->rb_node, root); 1171 rc = 0; 1172 out: 1173 if (!rc) { 1174 nh_base_seq_inc(net); 1175 nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo); 1176 if (replace_notify && net->ipv4.sysctl_nexthop_compat_mode) 1177 nexthop_replace_notify(net, new_nh, &cfg->nlinfo); 1178 } 1179 1180 return rc; 1181 } 1182 1183 /* rtnl */ 1184 /* remove all nexthops tied to a device being deleted */ 1185 static void nexthop_flush_dev(struct net_device *dev) 1186 { 1187 unsigned int hash = nh_dev_hashfn(dev->ifindex); 1188 struct net *net = dev_net(dev); 1189 struct hlist_head *head = &net->nexthop.devhash[hash]; 1190 struct hlist_node *n; 1191 struct nh_info *nhi; 1192 1193 hlist_for_each_entry_safe(nhi, n, head, dev_hash) { 1194 if (nhi->fib_nhc.nhc_dev != dev) 1195 continue; 1196 1197 remove_nexthop(net, nhi->nh_parent, NULL); 1198 } 1199 } 1200 1201 /* rtnl; called when net namespace is deleted */ 1202 static void flush_all_nexthops(struct net *net) 1203 { 1204 struct rb_root *root = &net->nexthop.rb_root; 1205 struct rb_node *node; 1206 struct nexthop *nh; 1207 1208 while ((node = rb_first(root))) { 1209 nh = rb_entry(node, struct nexthop, rb_node); 1210 remove_nexthop(net, nh, NULL); 1211 cond_resched(); 1212 } 1213 } 1214 1215 static struct nexthop *nexthop_create_group(struct net *net, 1216 struct nh_config *cfg) 1217 { 1218 struct nlattr *grps_attr = cfg->nh_grp; 1219 struct nexthop_grp *entry = nla_data(grps_attr); 1220 u16 num_nh = nla_len(grps_attr) / sizeof(*entry); 1221 struct nh_group *nhg; 1222 struct nexthop *nh; 1223 int i; 1224 1225 if (WARN_ON(!num_nh)) 1226 return ERR_PTR(-EINVAL); 1227 1228 nh = nexthop_alloc(); 1229 if (!nh) 1230 return ERR_PTR(-ENOMEM); 1231 1232 nh->is_group = 1; 1233 1234 nhg = nexthop_grp_alloc(num_nh); 1235 if (!nhg) { 1236 kfree(nh); 1237 return ERR_PTR(-ENOMEM); 1238 } 1239 1240 /* spare group used for removals */ 1241 nhg->spare = nexthop_grp_alloc(num_nh); 1242 if (!nhg->spare) { 1243 kfree(nhg); 1244 kfree(nh); 1245 return ERR_PTR(-ENOMEM); 1246 } 1247 nhg->spare->spare = nhg; 1248 1249 for (i = 0; i < nhg->num_nh; ++i) { 1250 struct nexthop *nhe; 1251 struct nh_info *nhi; 1252 1253 nhe = nexthop_find_by_id(net, entry[i].id); 1254 if (!nexthop_get(nhe)) 1255 goto out_no_nh; 1256 1257 nhi = rtnl_dereference(nhe->nh_info); 1258 if (nhi->family == AF_INET) 1259 nhg->has_v4 = true; 1260 1261 nhg->nh_entries[i].nh = nhe; 1262 nhg->nh_entries[i].weight = entry[i].weight + 1; 1263 list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list); 1264 nhg->nh_entries[i].nh_parent = nh; 1265 } 1266 1267 if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) { 1268 nhg->mpath = 1; 1269 nh_group_rebalance(nhg); 1270 } 1271 1272 if (cfg->nh_fdb) 1273 nhg->fdb_nh = 1; 1274 1275 rcu_assign_pointer(nh->nh_grp, nhg); 1276 1277 return nh; 1278 1279 out_no_nh: 1280 for (; i >= 0; --i) 1281 nexthop_put(nhg->nh_entries[i].nh); 1282 1283 kfree(nhg->spare); 1284 kfree(nhg); 1285 kfree(nh); 1286 1287 return ERR_PTR(-ENOENT); 1288 } 1289 1290 static int nh_create_ipv4(struct net *net, struct nexthop *nh, 1291 struct nh_info *nhi, struct nh_config *cfg, 1292 struct netlink_ext_ack *extack) 1293 { 1294 struct fib_nh *fib_nh = &nhi->fib_nh; 1295 struct fib_config fib_cfg = { 1296 .fc_oif = cfg->nh_ifindex, 1297 .fc_gw4 = cfg->gw.ipv4, 1298 .fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0, 1299 .fc_flags = cfg->nh_flags, 1300 .fc_encap = cfg->nh_encap, 1301 .fc_encap_type = cfg->nh_encap_type, 1302 }; 1303 u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN); 1304 int err; 1305 1306 err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack); 1307 if (err) { 1308 fib_nh_release(net, fib_nh); 1309 goto out; 1310 } 1311 1312 if (nhi->fdb_nh) 1313 goto out; 1314 1315 /* sets nh_dev if successful */ 1316 err = fib_check_nh(net, fib_nh, tb_id, 0, extack); 1317 if (!err) { 1318 nh->nh_flags = fib_nh->fib_nh_flags; 1319 fib_info_update_nhc_saddr(net, &fib_nh->nh_common, 1320 fib_nh->fib_nh_scope); 1321 } else { 1322 fib_nh_release(net, fib_nh); 1323 } 1324 out: 1325 return err; 1326 } 1327 1328 static int nh_create_ipv6(struct net *net, struct nexthop *nh, 1329 struct nh_info *nhi, struct nh_config *cfg, 1330 struct netlink_ext_ack *extack) 1331 { 1332 struct fib6_nh *fib6_nh = &nhi->fib6_nh; 1333 struct fib6_config fib6_cfg = { 1334 .fc_table = l3mdev_fib_table(cfg->dev), 1335 .fc_ifindex = cfg->nh_ifindex, 1336 .fc_gateway = cfg->gw.ipv6, 1337 .fc_flags = cfg->nh_flags, 1338 .fc_encap = cfg->nh_encap, 1339 .fc_encap_type = cfg->nh_encap_type, 1340 .fc_is_fdb = cfg->nh_fdb, 1341 }; 1342 int err; 1343 1344 if (!ipv6_addr_any(&cfg->gw.ipv6)) 1345 fib6_cfg.fc_flags |= RTF_GATEWAY; 1346 1347 /* sets nh_dev if successful */ 1348 err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL, 1349 extack); 1350 if (err) 1351 ipv6_stub->fib6_nh_release(fib6_nh); 1352 else 1353 nh->nh_flags = fib6_nh->fib_nh_flags; 1354 1355 return err; 1356 } 1357 1358 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg, 1359 struct netlink_ext_ack *extack) 1360 { 1361 struct nh_info *nhi; 1362 struct nexthop *nh; 1363 int err = 0; 1364 1365 nh = nexthop_alloc(); 1366 if (!nh) 1367 return ERR_PTR(-ENOMEM); 1368 1369 nhi = kzalloc(sizeof(*nhi), GFP_KERNEL); 1370 if (!nhi) { 1371 kfree(nh); 1372 return ERR_PTR(-ENOMEM); 1373 } 1374 1375 nh->nh_flags = cfg->nh_flags; 1376 nh->net = net; 1377 1378 nhi->nh_parent = nh; 1379 nhi->family = cfg->nh_family; 1380 nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK; 1381 1382 if (cfg->nh_fdb) 1383 nhi->fdb_nh = 1; 1384 1385 if (cfg->nh_blackhole) { 1386 nhi->reject_nh = 1; 1387 cfg->nh_ifindex = net->loopback_dev->ifindex; 1388 } 1389 1390 switch (cfg->nh_family) { 1391 case AF_INET: 1392 err = nh_create_ipv4(net, nh, nhi, cfg, extack); 1393 break; 1394 case AF_INET6: 1395 err = nh_create_ipv6(net, nh, nhi, cfg, extack); 1396 break; 1397 } 1398 1399 if (err) { 1400 kfree(nhi); 1401 kfree(nh); 1402 return ERR_PTR(err); 1403 } 1404 1405 /* add the entry to the device based hash */ 1406 if (!nhi->fdb_nh) 1407 nexthop_devhash_add(net, nhi); 1408 1409 rcu_assign_pointer(nh->nh_info, nhi); 1410 1411 return nh; 1412 } 1413 1414 /* called with rtnl lock held */ 1415 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg, 1416 struct netlink_ext_ack *extack) 1417 { 1418 struct nexthop *nh; 1419 int err; 1420 1421 if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) { 1422 NL_SET_ERR_MSG(extack, "Replace requires nexthop id"); 1423 return ERR_PTR(-EINVAL); 1424 } 1425 1426 if (!cfg->nh_id) { 1427 cfg->nh_id = nh_find_unused_id(net); 1428 if (!cfg->nh_id) { 1429 NL_SET_ERR_MSG(extack, "No unused id"); 1430 return ERR_PTR(-EINVAL); 1431 } 1432 } 1433 1434 if (cfg->nh_grp) 1435 nh = nexthop_create_group(net, cfg); 1436 else 1437 nh = nexthop_create(net, cfg, extack); 1438 1439 if (IS_ERR(nh)) 1440 return nh; 1441 1442 refcount_set(&nh->refcnt, 1); 1443 nh->id = cfg->nh_id; 1444 nh->protocol = cfg->nh_protocol; 1445 nh->net = net; 1446 1447 err = insert_nexthop(net, nh, cfg, extack); 1448 if (err) { 1449 __remove_nexthop(net, nh, NULL); 1450 nexthop_put(nh); 1451 nh = ERR_PTR(err); 1452 } 1453 1454 return nh; 1455 } 1456 1457 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb, 1458 struct nlmsghdr *nlh, struct nh_config *cfg, 1459 struct netlink_ext_ack *extack) 1460 { 1461 struct nhmsg *nhm = nlmsg_data(nlh); 1462 struct nlattr *tb[NHA_MAX + 1]; 1463 int err; 1464 1465 err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy, 1466 extack); 1467 if (err < 0) 1468 return err; 1469 1470 err = -EINVAL; 1471 if (nhm->resvd || nhm->nh_scope) { 1472 NL_SET_ERR_MSG(extack, "Invalid values in ancillary header"); 1473 goto out; 1474 } 1475 if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) { 1476 NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header"); 1477 goto out; 1478 } 1479 1480 switch (nhm->nh_family) { 1481 case AF_INET: 1482 case AF_INET6: 1483 break; 1484 case AF_UNSPEC: 1485 if (tb[NHA_GROUP]) 1486 break; 1487 fallthrough; 1488 default: 1489 NL_SET_ERR_MSG(extack, "Invalid address family"); 1490 goto out; 1491 } 1492 1493 if (tb[NHA_GROUPS] || tb[NHA_MASTER]) { 1494 NL_SET_ERR_MSG(extack, "Invalid attributes in request"); 1495 goto out; 1496 } 1497 1498 memset(cfg, 0, sizeof(*cfg)); 1499 cfg->nlflags = nlh->nlmsg_flags; 1500 cfg->nlinfo.portid = NETLINK_CB(skb).portid; 1501 cfg->nlinfo.nlh = nlh; 1502 cfg->nlinfo.nl_net = net; 1503 1504 cfg->nh_family = nhm->nh_family; 1505 cfg->nh_protocol = nhm->nh_protocol; 1506 cfg->nh_flags = nhm->nh_flags; 1507 1508 if (tb[NHA_ID]) 1509 cfg->nh_id = nla_get_u32(tb[NHA_ID]); 1510 1511 if (tb[NHA_FDB]) { 1512 if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] || 1513 tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE]) { 1514 NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole"); 1515 goto out; 1516 } 1517 if (nhm->nh_flags) { 1518 NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header"); 1519 goto out; 1520 } 1521 cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]); 1522 } 1523 1524 if (tb[NHA_GROUP]) { 1525 if (nhm->nh_family != AF_UNSPEC) { 1526 NL_SET_ERR_MSG(extack, "Invalid family for group"); 1527 goto out; 1528 } 1529 cfg->nh_grp = tb[NHA_GROUP]; 1530 1531 cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH; 1532 if (tb[NHA_GROUP_TYPE]) 1533 cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]); 1534 1535 if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) { 1536 NL_SET_ERR_MSG(extack, "Invalid group type"); 1537 goto out; 1538 } 1539 err = nh_check_attr_group(net, tb, extack); 1540 1541 /* no other attributes should be set */ 1542 goto out; 1543 } 1544 1545 if (tb[NHA_BLACKHOLE]) { 1546 if (tb[NHA_GATEWAY] || tb[NHA_OIF] || 1547 tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) { 1548 NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb"); 1549 goto out; 1550 } 1551 1552 cfg->nh_blackhole = 1; 1553 err = 0; 1554 goto out; 1555 } 1556 1557 if (!cfg->nh_fdb && !tb[NHA_OIF]) { 1558 NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops"); 1559 goto out; 1560 } 1561 1562 if (!cfg->nh_fdb && tb[NHA_OIF]) { 1563 cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]); 1564 if (cfg->nh_ifindex) 1565 cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex); 1566 1567 if (!cfg->dev) { 1568 NL_SET_ERR_MSG(extack, "Invalid device index"); 1569 goto out; 1570 } else if (!(cfg->dev->flags & IFF_UP)) { 1571 NL_SET_ERR_MSG(extack, "Nexthop device is not up"); 1572 err = -ENETDOWN; 1573 goto out; 1574 } else if (!netif_carrier_ok(cfg->dev)) { 1575 NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down"); 1576 err = -ENETDOWN; 1577 goto out; 1578 } 1579 } 1580 1581 err = -EINVAL; 1582 if (tb[NHA_GATEWAY]) { 1583 struct nlattr *gwa = tb[NHA_GATEWAY]; 1584 1585 switch (cfg->nh_family) { 1586 case AF_INET: 1587 if (nla_len(gwa) != sizeof(u32)) { 1588 NL_SET_ERR_MSG(extack, "Invalid gateway"); 1589 goto out; 1590 } 1591 cfg->gw.ipv4 = nla_get_be32(gwa); 1592 break; 1593 case AF_INET6: 1594 if (nla_len(gwa) != sizeof(struct in6_addr)) { 1595 NL_SET_ERR_MSG(extack, "Invalid gateway"); 1596 goto out; 1597 } 1598 cfg->gw.ipv6 = nla_get_in6_addr(gwa); 1599 break; 1600 default: 1601 NL_SET_ERR_MSG(extack, 1602 "Unknown address family for gateway"); 1603 goto out; 1604 } 1605 } else { 1606 /* device only nexthop (no gateway) */ 1607 if (cfg->nh_flags & RTNH_F_ONLINK) { 1608 NL_SET_ERR_MSG(extack, 1609 "ONLINK flag can not be set for nexthop without a gateway"); 1610 goto out; 1611 } 1612 } 1613 1614 if (tb[NHA_ENCAP]) { 1615 cfg->nh_encap = tb[NHA_ENCAP]; 1616 1617 if (!tb[NHA_ENCAP_TYPE]) { 1618 NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing"); 1619 goto out; 1620 } 1621 1622 cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]); 1623 err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack); 1624 if (err < 0) 1625 goto out; 1626 1627 } else if (tb[NHA_ENCAP_TYPE]) { 1628 NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing"); 1629 goto out; 1630 } 1631 1632 1633 err = 0; 1634 out: 1635 return err; 1636 } 1637 1638 /* rtnl */ 1639 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh, 1640 struct netlink_ext_ack *extack) 1641 { 1642 struct net *net = sock_net(skb->sk); 1643 struct nh_config cfg; 1644 struct nexthop *nh; 1645 int err; 1646 1647 err = rtm_to_nh_config(net, skb, nlh, &cfg, extack); 1648 if (!err) { 1649 nh = nexthop_add(net, &cfg, extack); 1650 if (IS_ERR(nh)) 1651 err = PTR_ERR(nh); 1652 } 1653 1654 return err; 1655 } 1656 1657 static int nh_valid_get_del_req(struct nlmsghdr *nlh, u32 *id, 1658 struct netlink_ext_ack *extack) 1659 { 1660 struct nhmsg *nhm = nlmsg_data(nlh); 1661 struct nlattr *tb[NHA_MAX + 1]; 1662 int err, i; 1663 1664 err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy, 1665 extack); 1666 if (err < 0) 1667 return err; 1668 1669 err = -EINVAL; 1670 for (i = 0; i < __NHA_MAX; ++i) { 1671 if (!tb[i]) 1672 continue; 1673 1674 switch (i) { 1675 case NHA_ID: 1676 break; 1677 default: 1678 NL_SET_ERR_MSG_ATTR(extack, tb[i], 1679 "Unexpected attribute in request"); 1680 goto out; 1681 } 1682 } 1683 if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) { 1684 NL_SET_ERR_MSG(extack, "Invalid values in header"); 1685 goto out; 1686 } 1687 1688 if (!tb[NHA_ID]) { 1689 NL_SET_ERR_MSG(extack, "Nexthop id is missing"); 1690 goto out; 1691 } 1692 1693 *id = nla_get_u32(tb[NHA_ID]); 1694 if (!(*id)) 1695 NL_SET_ERR_MSG(extack, "Invalid nexthop id"); 1696 else 1697 err = 0; 1698 out: 1699 return err; 1700 } 1701 1702 /* rtnl */ 1703 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh, 1704 struct netlink_ext_ack *extack) 1705 { 1706 struct net *net = sock_net(skb->sk); 1707 struct nl_info nlinfo = { 1708 .nlh = nlh, 1709 .nl_net = net, 1710 .portid = NETLINK_CB(skb).portid, 1711 }; 1712 struct nexthop *nh; 1713 int err; 1714 u32 id; 1715 1716 err = nh_valid_get_del_req(nlh, &id, extack); 1717 if (err) 1718 return err; 1719 1720 nh = nexthop_find_by_id(net, id); 1721 if (!nh) 1722 return -ENOENT; 1723 1724 remove_nexthop(net, nh, &nlinfo); 1725 1726 return 0; 1727 } 1728 1729 /* rtnl */ 1730 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh, 1731 struct netlink_ext_ack *extack) 1732 { 1733 struct net *net = sock_net(in_skb->sk); 1734 struct sk_buff *skb = NULL; 1735 struct nexthop *nh; 1736 int err; 1737 u32 id; 1738 1739 err = nh_valid_get_del_req(nlh, &id, extack); 1740 if (err) 1741 return err; 1742 1743 err = -ENOBUFS; 1744 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1745 if (!skb) 1746 goto out; 1747 1748 err = -ENOENT; 1749 nh = nexthop_find_by_id(net, id); 1750 if (!nh) 1751 goto errout_free; 1752 1753 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid, 1754 nlh->nlmsg_seq, 0); 1755 if (err < 0) { 1756 WARN_ON(err == -EMSGSIZE); 1757 goto errout_free; 1758 } 1759 1760 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 1761 out: 1762 return err; 1763 errout_free: 1764 kfree_skb(skb); 1765 goto out; 1766 } 1767 1768 static bool nh_dump_filtered(struct nexthop *nh, int dev_idx, int master_idx, 1769 bool group_filter, u8 family) 1770 { 1771 const struct net_device *dev; 1772 const struct nh_info *nhi; 1773 1774 if (group_filter && !nh->is_group) 1775 return true; 1776 1777 if (!dev_idx && !master_idx && !family) 1778 return false; 1779 1780 if (nh->is_group) 1781 return true; 1782 1783 nhi = rtnl_dereference(nh->nh_info); 1784 if (family && nhi->family != family) 1785 return true; 1786 1787 dev = nhi->fib_nhc.nhc_dev; 1788 if (dev_idx && (!dev || dev->ifindex != dev_idx)) 1789 return true; 1790 1791 if (master_idx) { 1792 struct net_device *master; 1793 1794 if (!dev) 1795 return true; 1796 1797 master = netdev_master_upper_dev_get((struct net_device *)dev); 1798 if (!master || master->ifindex != master_idx) 1799 return true; 1800 } 1801 1802 return false; 1803 } 1804 1805 static int nh_valid_dump_req(const struct nlmsghdr *nlh, int *dev_idx, 1806 int *master_idx, bool *group_filter, 1807 bool *fdb_filter, struct netlink_callback *cb) 1808 { 1809 struct netlink_ext_ack *extack = cb->extack; 1810 struct nlattr *tb[NHA_MAX + 1]; 1811 struct nhmsg *nhm; 1812 int err, i; 1813 u32 idx; 1814 1815 err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy, 1816 NULL); 1817 if (err < 0) 1818 return err; 1819 1820 for (i = 0; i <= NHA_MAX; ++i) { 1821 if (!tb[i]) 1822 continue; 1823 1824 switch (i) { 1825 case NHA_OIF: 1826 idx = nla_get_u32(tb[i]); 1827 if (idx > INT_MAX) { 1828 NL_SET_ERR_MSG(extack, "Invalid device index"); 1829 return -EINVAL; 1830 } 1831 *dev_idx = idx; 1832 break; 1833 case NHA_MASTER: 1834 idx = nla_get_u32(tb[i]); 1835 if (idx > INT_MAX) { 1836 NL_SET_ERR_MSG(extack, "Invalid master device index"); 1837 return -EINVAL; 1838 } 1839 *master_idx = idx; 1840 break; 1841 case NHA_GROUPS: 1842 *group_filter = true; 1843 break; 1844 case NHA_FDB: 1845 *fdb_filter = true; 1846 break; 1847 default: 1848 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request"); 1849 return -EINVAL; 1850 } 1851 } 1852 1853 nhm = nlmsg_data(nlh); 1854 if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) { 1855 NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request"); 1856 return -EINVAL; 1857 } 1858 1859 return 0; 1860 } 1861 1862 /* rtnl */ 1863 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb) 1864 { 1865 bool group_filter = false, fdb_filter = false; 1866 struct nhmsg *nhm = nlmsg_data(cb->nlh); 1867 int dev_filter_idx = 0, master_idx = 0; 1868 struct net *net = sock_net(skb->sk); 1869 struct rb_root *root = &net->nexthop.rb_root; 1870 struct rb_node *node; 1871 int idx = 0, s_idx; 1872 int err; 1873 1874 err = nh_valid_dump_req(cb->nlh, &dev_filter_idx, &master_idx, 1875 &group_filter, &fdb_filter, cb); 1876 if (err < 0) 1877 return err; 1878 1879 s_idx = cb->args[0]; 1880 for (node = rb_first(root); node; node = rb_next(node)) { 1881 struct nexthop *nh; 1882 1883 if (idx < s_idx) 1884 goto cont; 1885 1886 nh = rb_entry(node, struct nexthop, rb_node); 1887 if (nh_dump_filtered(nh, dev_filter_idx, master_idx, 1888 group_filter, nhm->nh_family)) 1889 goto cont; 1890 1891 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, 1892 NETLINK_CB(cb->skb).portid, 1893 cb->nlh->nlmsg_seq, NLM_F_MULTI); 1894 if (err < 0) { 1895 if (likely(skb->len)) 1896 goto out; 1897 1898 goto out_err; 1899 } 1900 cont: 1901 idx++; 1902 } 1903 1904 out: 1905 err = skb->len; 1906 out_err: 1907 cb->args[0] = idx; 1908 cb->seq = net->nexthop.seq; 1909 nl_dump_check_consistent(cb, nlmsg_hdr(skb)); 1910 1911 return err; 1912 } 1913 1914 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu) 1915 { 1916 unsigned int hash = nh_dev_hashfn(dev->ifindex); 1917 struct net *net = dev_net(dev); 1918 struct hlist_head *head = &net->nexthop.devhash[hash]; 1919 struct hlist_node *n; 1920 struct nh_info *nhi; 1921 1922 hlist_for_each_entry_safe(nhi, n, head, dev_hash) { 1923 if (nhi->fib_nhc.nhc_dev == dev) { 1924 if (nhi->family == AF_INET) 1925 fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu, 1926 orig_mtu); 1927 } 1928 } 1929 } 1930 1931 /* rtnl */ 1932 static int nh_netdev_event(struct notifier_block *this, 1933 unsigned long event, void *ptr) 1934 { 1935 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1936 struct netdev_notifier_info_ext *info_ext; 1937 1938 switch (event) { 1939 case NETDEV_DOWN: 1940 case NETDEV_UNREGISTER: 1941 nexthop_flush_dev(dev); 1942 break; 1943 case NETDEV_CHANGE: 1944 if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP))) 1945 nexthop_flush_dev(dev); 1946 break; 1947 case NETDEV_CHANGEMTU: 1948 info_ext = ptr; 1949 nexthop_sync_mtu(dev, info_ext->ext.mtu); 1950 rt_cache_flush(dev_net(dev)); 1951 break; 1952 } 1953 return NOTIFY_DONE; 1954 } 1955 1956 static struct notifier_block nh_netdev_notifier = { 1957 .notifier_call = nh_netdev_event, 1958 }; 1959 1960 int register_nexthop_notifier(struct net *net, struct notifier_block *nb) 1961 { 1962 return atomic_notifier_chain_register(&net->nexthop.notifier_chain, nb); 1963 } 1964 EXPORT_SYMBOL(register_nexthop_notifier); 1965 1966 int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb) 1967 { 1968 return atomic_notifier_chain_unregister(&net->nexthop.notifier_chain, 1969 nb); 1970 } 1971 EXPORT_SYMBOL(unregister_nexthop_notifier); 1972 1973 static void __net_exit nexthop_net_exit(struct net *net) 1974 { 1975 rtnl_lock(); 1976 flush_all_nexthops(net); 1977 rtnl_unlock(); 1978 kfree(net->nexthop.devhash); 1979 } 1980 1981 static int __net_init nexthop_net_init(struct net *net) 1982 { 1983 size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE; 1984 1985 net->nexthop.rb_root = RB_ROOT; 1986 net->nexthop.devhash = kzalloc(sz, GFP_KERNEL); 1987 if (!net->nexthop.devhash) 1988 return -ENOMEM; 1989 ATOMIC_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain); 1990 1991 return 0; 1992 } 1993 1994 static struct pernet_operations nexthop_net_ops = { 1995 .init = nexthop_net_init, 1996 .exit = nexthop_net_exit, 1997 }; 1998 1999 static int __init nexthop_init(void) 2000 { 2001 register_pernet_subsys(&nexthop_net_ops); 2002 2003 register_netdevice_notifier(&nh_netdev_notifier); 2004 2005 rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0); 2006 rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0); 2007 rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop, 2008 rtm_dump_nexthop, 0); 2009 2010 rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0); 2011 rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0); 2012 2013 rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0); 2014 rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0); 2015 2016 return 0; 2017 } 2018 subsys_initcall(nexthop_init); 2019