1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * net/core/fib_rules.c Generic Routing Rules 4 * 5 * Authors: Thomas Graf <tgraf@suug.ch> 6 */ 7 8 #include <linux/types.h> 9 #include <linux/kernel.h> 10 #include <linux/slab.h> 11 #include <linux/list.h> 12 #include <linux/module.h> 13 #include <net/net_namespace.h> 14 #include <net/inet_dscp.h> 15 #include <net/sock.h> 16 #include <net/fib_rules.h> 17 #include <net/ip_tunnels.h> 18 #include <linux/indirect_call_wrapper.h> 19 20 #if defined(CONFIG_IPV6) && defined(CONFIG_IPV6_MULTIPLE_TABLES) 21 #ifdef CONFIG_IP_MULTIPLE_TABLES 22 #define INDIRECT_CALL_MT(f, f2, f1, ...) \ 23 INDIRECT_CALL_INET(f, f2, f1, __VA_ARGS__) 24 #else 25 #define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__) 26 #endif 27 #elif defined(CONFIG_IP_MULTIPLE_TABLES) 28 #define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f1, __VA_ARGS__) 29 #else 30 #define INDIRECT_CALL_MT(f, f2, f1, ...) f(__VA_ARGS__) 31 #endif 32 33 static const struct fib_kuid_range fib_kuid_range_unset = { 34 KUIDT_INIT(0), 35 KUIDT_INIT(~0), 36 }; 37 38 bool fib_rule_matchall(const struct fib_rule *rule) 39 { 40 if (rule->iifindex || rule->oifindex || rule->mark || rule->tun_id || 41 rule->flags) 42 return false; 43 if (rule->suppress_ifgroup != -1 || rule->suppress_prefixlen != -1) 44 return false; 45 if (!uid_eq(rule->uid_range.start, fib_kuid_range_unset.start) || 46 !uid_eq(rule->uid_range.end, fib_kuid_range_unset.end)) 47 return false; 48 if (fib_rule_port_range_set(&rule->sport_range)) 49 return false; 50 if (fib_rule_port_range_set(&rule->dport_range)) 51 return false; 52 return true; 53 } 54 EXPORT_SYMBOL_GPL(fib_rule_matchall); 55 56 int fib_default_rule_add(struct fib_rules_ops *ops, 57 u32 pref, u32 table) 58 { 59 struct fib_rule *r; 60 61 r = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT); 62 if (r == NULL) 63 return -ENOMEM; 64 65 refcount_set(&r->refcnt, 1); 66 r->action = FR_ACT_TO_TBL; 67 r->pref = pref; 68 r->table = table; 69 r->proto = RTPROT_KERNEL; 70 r->fr_net = ops->fro_net; 71 r->uid_range = fib_kuid_range_unset; 72 73 r->suppress_prefixlen = -1; 74 r->suppress_ifgroup = -1; 75 76 /* The lock is not required here, the list in unreachable 77 * at the moment this function is called */ 78 list_add_tail(&r->list, &ops->rules_list); 79 return 0; 80 } 81 EXPORT_SYMBOL(fib_default_rule_add); 82 83 static u32 fib_default_rule_pref(struct fib_rules_ops *ops) 84 { 85 struct list_head *pos; 86 struct fib_rule *rule; 87 88 if (!list_empty(&ops->rules_list)) { 89 pos = ops->rules_list.next; 90 if (pos->next != &ops->rules_list) { 91 rule = list_entry(pos->next, struct fib_rule, list); 92 if (rule->pref) 93 return rule->pref - 1; 94 } 95 } 96 97 return 0; 98 } 99 100 static void notify_rule_change(int event, struct fib_rule *rule, 101 struct fib_rules_ops *ops, struct nlmsghdr *nlh, 102 u32 pid); 103 104 static struct fib_rules_ops *lookup_rules_ops(const struct net *net, 105 int family) 106 { 107 struct fib_rules_ops *ops; 108 109 rcu_read_lock(); 110 list_for_each_entry_rcu(ops, &net->rules_ops, list) { 111 if (ops->family == family) { 112 if (!try_module_get(ops->owner)) 113 ops = NULL; 114 rcu_read_unlock(); 115 return ops; 116 } 117 } 118 rcu_read_unlock(); 119 120 return NULL; 121 } 122 123 static void rules_ops_put(struct fib_rules_ops *ops) 124 { 125 if (ops) 126 module_put(ops->owner); 127 } 128 129 static void flush_route_cache(struct fib_rules_ops *ops) 130 { 131 if (ops->flush_cache) 132 ops->flush_cache(ops); 133 } 134 135 static int __fib_rules_register(struct fib_rules_ops *ops) 136 { 137 int err = -EEXIST; 138 struct fib_rules_ops *o; 139 struct net *net; 140 141 net = ops->fro_net; 142 143 if (ops->rule_size < sizeof(struct fib_rule)) 144 return -EINVAL; 145 146 if (ops->match == NULL || ops->configure == NULL || 147 ops->compare == NULL || ops->fill == NULL || 148 ops->action == NULL) 149 return -EINVAL; 150 151 spin_lock(&net->rules_mod_lock); 152 list_for_each_entry(o, &net->rules_ops, list) 153 if (ops->family == o->family) 154 goto errout; 155 156 list_add_tail_rcu(&ops->list, &net->rules_ops); 157 err = 0; 158 errout: 159 spin_unlock(&net->rules_mod_lock); 160 161 return err; 162 } 163 164 struct fib_rules_ops * 165 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net) 166 { 167 struct fib_rules_ops *ops; 168 int err; 169 170 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL); 171 if (ops == NULL) 172 return ERR_PTR(-ENOMEM); 173 174 INIT_LIST_HEAD(&ops->rules_list); 175 ops->fro_net = net; 176 177 err = __fib_rules_register(ops); 178 if (err) { 179 kfree(ops); 180 ops = ERR_PTR(err); 181 } 182 183 return ops; 184 } 185 EXPORT_SYMBOL_GPL(fib_rules_register); 186 187 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops) 188 { 189 struct fib_rule *rule, *tmp; 190 191 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) { 192 list_del_rcu(&rule->list); 193 if (ops->delete) 194 ops->delete(rule); 195 fib_rule_put(rule); 196 } 197 } 198 199 void fib_rules_unregister(struct fib_rules_ops *ops) 200 { 201 struct net *net = ops->fro_net; 202 203 spin_lock(&net->rules_mod_lock); 204 list_del_rcu(&ops->list); 205 spin_unlock(&net->rules_mod_lock); 206 207 fib_rules_cleanup_ops(ops); 208 kfree_rcu(ops, rcu); 209 } 210 EXPORT_SYMBOL_GPL(fib_rules_unregister); 211 212 static int uid_range_set(struct fib_kuid_range *range) 213 { 214 return uid_valid(range->start) && uid_valid(range->end); 215 } 216 217 static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb) 218 { 219 struct fib_rule_uid_range *in; 220 struct fib_kuid_range out; 221 222 in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]); 223 224 out.start = make_kuid(current_user_ns(), in->start); 225 out.end = make_kuid(current_user_ns(), in->end); 226 227 return out; 228 } 229 230 static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range) 231 { 232 struct fib_rule_uid_range out = { 233 from_kuid_munged(current_user_ns(), range->start), 234 from_kuid_munged(current_user_ns(), range->end) 235 }; 236 237 return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out); 238 } 239 240 static int nla_get_port_range(struct nlattr *pattr, 241 struct fib_rule_port_range *port_range) 242 { 243 const struct fib_rule_port_range *pr = nla_data(pattr); 244 245 if (!fib_rule_port_range_valid(pr)) 246 return -EINVAL; 247 248 port_range->start = pr->start; 249 port_range->end = pr->end; 250 251 return 0; 252 } 253 254 static int nla_put_port_range(struct sk_buff *skb, int attrtype, 255 struct fib_rule_port_range *range) 256 { 257 return nla_put(skb, attrtype, sizeof(*range), range); 258 } 259 260 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops, 261 struct flowi *fl, int flags, 262 struct fib_lookup_arg *arg) 263 { 264 int ret = 0; 265 266 if (rule->iifindex && (rule->iifindex != fl->flowi_iif)) 267 goto out; 268 269 if (rule->oifindex && (rule->oifindex != fl->flowi_oif)) 270 goto out; 271 272 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask) 273 goto out; 274 275 if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id)) 276 goto out; 277 278 if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg)) 279 goto out; 280 281 if (uid_lt(fl->flowi_uid, rule->uid_range.start) || 282 uid_gt(fl->flowi_uid, rule->uid_range.end)) 283 goto out; 284 285 ret = INDIRECT_CALL_MT(ops->match, 286 fib6_rule_match, 287 fib4_rule_match, 288 rule, fl, flags); 289 out: 290 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret; 291 } 292 293 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl, 294 int flags, struct fib_lookup_arg *arg) 295 { 296 struct fib_rule *rule; 297 int err; 298 299 rcu_read_lock(); 300 301 list_for_each_entry_rcu(rule, &ops->rules_list, list) { 302 jumped: 303 if (!fib_rule_match(rule, ops, fl, flags, arg)) 304 continue; 305 306 if (rule->action == FR_ACT_GOTO) { 307 struct fib_rule *target; 308 309 target = rcu_dereference(rule->ctarget); 310 if (target == NULL) { 311 continue; 312 } else { 313 rule = target; 314 goto jumped; 315 } 316 } else if (rule->action == FR_ACT_NOP) 317 continue; 318 else 319 err = INDIRECT_CALL_MT(ops->action, 320 fib6_rule_action, 321 fib4_rule_action, 322 rule, fl, flags, arg); 323 324 if (!err && ops->suppress && INDIRECT_CALL_MT(ops->suppress, 325 fib6_rule_suppress, 326 fib4_rule_suppress, 327 rule, flags, arg)) 328 continue; 329 330 if (err != -EAGAIN) { 331 if ((arg->flags & FIB_LOOKUP_NOREF) || 332 likely(refcount_inc_not_zero(&rule->refcnt))) { 333 arg->rule = rule; 334 goto out; 335 } 336 break; 337 } 338 } 339 340 err = -ESRCH; 341 out: 342 rcu_read_unlock(); 343 344 return err; 345 } 346 EXPORT_SYMBOL_GPL(fib_rules_lookup); 347 348 static int call_fib_rule_notifier(struct notifier_block *nb, 349 enum fib_event_type event_type, 350 struct fib_rule *rule, int family, 351 struct netlink_ext_ack *extack) 352 { 353 struct fib_rule_notifier_info info = { 354 .info.family = family, 355 .info.extack = extack, 356 .rule = rule, 357 }; 358 359 return call_fib_notifier(nb, event_type, &info.info); 360 } 361 362 static int call_fib_rule_notifiers(struct net *net, 363 enum fib_event_type event_type, 364 struct fib_rule *rule, 365 struct fib_rules_ops *ops, 366 struct netlink_ext_ack *extack) 367 { 368 struct fib_rule_notifier_info info = { 369 .info.family = ops->family, 370 .info.extack = extack, 371 .rule = rule, 372 }; 373 374 ASSERT_RTNL_NET(net); 375 376 /* Paired with READ_ONCE() in fib_rules_seq() */ 377 WRITE_ONCE(ops->fib_rules_seq, ops->fib_rules_seq + 1); 378 return call_fib_notifiers(net, event_type, &info.info); 379 } 380 381 /* Called with rcu_read_lock() */ 382 int fib_rules_dump(struct net *net, struct notifier_block *nb, int family, 383 struct netlink_ext_ack *extack) 384 { 385 struct fib_rules_ops *ops; 386 struct fib_rule *rule; 387 int err = 0; 388 389 ops = lookup_rules_ops(net, family); 390 if (!ops) 391 return -EAFNOSUPPORT; 392 list_for_each_entry_rcu(rule, &ops->rules_list, list) { 393 err = call_fib_rule_notifier(nb, FIB_EVENT_RULE_ADD, 394 rule, family, extack); 395 if (err) 396 break; 397 } 398 rules_ops_put(ops); 399 400 return err; 401 } 402 EXPORT_SYMBOL_GPL(fib_rules_dump); 403 404 unsigned int fib_rules_seq_read(const struct net *net, int family) 405 { 406 unsigned int fib_rules_seq; 407 struct fib_rules_ops *ops; 408 409 ops = lookup_rules_ops(net, family); 410 if (!ops) 411 return 0; 412 /* Paired with WRITE_ONCE() in call_fib_rule_notifiers() */ 413 fib_rules_seq = READ_ONCE(ops->fib_rules_seq); 414 rules_ops_put(ops); 415 416 return fib_rules_seq; 417 } 418 EXPORT_SYMBOL_GPL(fib_rules_seq_read); 419 420 static struct fib_rule *rule_find(struct fib_rules_ops *ops, 421 struct fib_rule_hdr *frh, 422 struct nlattr **tb, 423 struct fib_rule *rule, 424 bool user_priority) 425 { 426 struct fib_rule *r; 427 428 list_for_each_entry(r, &ops->rules_list, list) { 429 if (rule->action && r->action != rule->action) 430 continue; 431 432 if (rule->table && r->table != rule->table) 433 continue; 434 435 if (user_priority && r->pref != rule->pref) 436 continue; 437 438 if (rule->iifname[0] && 439 memcmp(r->iifname, rule->iifname, IFNAMSIZ)) 440 continue; 441 442 if (rule->oifname[0] && 443 memcmp(r->oifname, rule->oifname, IFNAMSIZ)) 444 continue; 445 446 if (rule->mark && r->mark != rule->mark) 447 continue; 448 449 if (rule->suppress_ifgroup != -1 && 450 r->suppress_ifgroup != rule->suppress_ifgroup) 451 continue; 452 453 if (rule->suppress_prefixlen != -1 && 454 r->suppress_prefixlen != rule->suppress_prefixlen) 455 continue; 456 457 if (rule->mark_mask && r->mark_mask != rule->mark_mask) 458 continue; 459 460 if (rule->tun_id && r->tun_id != rule->tun_id) 461 continue; 462 463 if (rule->l3mdev && r->l3mdev != rule->l3mdev) 464 continue; 465 466 if (uid_range_set(&rule->uid_range) && 467 (!uid_eq(r->uid_range.start, rule->uid_range.start) || 468 !uid_eq(r->uid_range.end, rule->uid_range.end))) 469 continue; 470 471 if (rule->ip_proto && r->ip_proto != rule->ip_proto) 472 continue; 473 474 if (rule->proto && r->proto != rule->proto) 475 continue; 476 477 if (fib_rule_port_range_set(&rule->sport_range) && 478 !fib_rule_port_range_compare(&r->sport_range, 479 &rule->sport_range)) 480 continue; 481 482 if (fib_rule_port_range_set(&rule->dport_range) && 483 !fib_rule_port_range_compare(&r->dport_range, 484 &rule->dport_range)) 485 continue; 486 487 if (!ops->compare(r, frh, tb)) 488 continue; 489 return r; 490 } 491 492 return NULL; 493 } 494 495 #ifdef CONFIG_NET_L3_MASTER_DEV 496 static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule, 497 struct netlink_ext_ack *extack) 498 { 499 nlrule->l3mdev = nla_get_u8(nla); 500 if (nlrule->l3mdev != 1) { 501 NL_SET_ERR_MSG(extack, "Invalid l3mdev attribute"); 502 return -1; 503 } 504 505 return 0; 506 } 507 #else 508 static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule, 509 struct netlink_ext_ack *extack) 510 { 511 NL_SET_ERR_MSG(extack, "l3mdev support is not enabled in kernel"); 512 return -1; 513 } 514 #endif 515 516 static int fib_nl2rule(struct net *net, struct nlmsghdr *nlh, 517 struct netlink_ext_ack *extack, 518 struct fib_rules_ops *ops, 519 struct nlattr *tb[], 520 struct fib_rule **rule, 521 bool *user_priority) 522 { 523 struct fib_rule_hdr *frh = nlmsg_data(nlh); 524 struct fib_rule *nlrule = NULL; 525 int err = -EINVAL; 526 527 if (frh->src_len) 528 if (!tb[FRA_SRC] || 529 frh->src_len > (ops->addr_size * 8) || 530 nla_len(tb[FRA_SRC]) != ops->addr_size) { 531 NL_SET_ERR_MSG(extack, "Invalid source address"); 532 goto errout; 533 } 534 535 if (frh->dst_len) 536 if (!tb[FRA_DST] || 537 frh->dst_len > (ops->addr_size * 8) || 538 nla_len(tb[FRA_DST]) != ops->addr_size) { 539 NL_SET_ERR_MSG(extack, "Invalid dst address"); 540 goto errout; 541 } 542 543 nlrule = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT); 544 if (!nlrule) { 545 err = -ENOMEM; 546 goto errout; 547 } 548 refcount_set(&nlrule->refcnt, 1); 549 nlrule->fr_net = net; 550 551 if (tb[FRA_PRIORITY]) { 552 nlrule->pref = nla_get_u32(tb[FRA_PRIORITY]); 553 *user_priority = true; 554 } 555 556 nlrule->proto = nla_get_u8_default(tb[FRA_PROTOCOL], RTPROT_UNSPEC); 557 558 if (tb[FRA_IIFNAME]) { 559 nlrule->iifindex = -1; 560 nla_strscpy(nlrule->iifname, tb[FRA_IIFNAME], IFNAMSIZ); 561 } 562 563 if (tb[FRA_OIFNAME]) { 564 nlrule->oifindex = -1; 565 nla_strscpy(nlrule->oifname, tb[FRA_OIFNAME], IFNAMSIZ); 566 } 567 568 if (tb[FRA_FWMARK]) { 569 nlrule->mark = nla_get_u32(tb[FRA_FWMARK]); 570 if (nlrule->mark) 571 /* compatibility: if the mark value is non-zero all bits 572 * are compared unless a mask is explicitly specified. 573 */ 574 nlrule->mark_mask = 0xFFFFFFFF; 575 } 576 577 if (tb[FRA_FWMASK]) 578 nlrule->mark_mask = nla_get_u32(tb[FRA_FWMASK]); 579 580 if (tb[FRA_TUN_ID]) 581 nlrule->tun_id = nla_get_be64(tb[FRA_TUN_ID]); 582 583 if (tb[FRA_L3MDEV] && 584 fib_nl2rule_l3mdev(tb[FRA_L3MDEV], nlrule, extack) < 0) 585 goto errout_free; 586 587 nlrule->action = frh->action; 588 nlrule->flags = frh->flags; 589 nlrule->table = frh_get_table(frh, tb); 590 if (tb[FRA_SUPPRESS_PREFIXLEN]) 591 nlrule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]); 592 else 593 nlrule->suppress_prefixlen = -1; 594 595 if (tb[FRA_SUPPRESS_IFGROUP]) 596 nlrule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]); 597 else 598 nlrule->suppress_ifgroup = -1; 599 600 if (tb[FRA_GOTO]) { 601 if (nlrule->action != FR_ACT_GOTO) { 602 NL_SET_ERR_MSG(extack, "Unexpected goto"); 603 goto errout_free; 604 } 605 606 nlrule->target = nla_get_u32(tb[FRA_GOTO]); 607 } else if (nlrule->action == FR_ACT_GOTO) { 608 NL_SET_ERR_MSG(extack, "Missing goto target for action goto"); 609 goto errout_free; 610 } 611 612 if (nlrule->l3mdev && nlrule->table) { 613 NL_SET_ERR_MSG(extack, "l3mdev and table are mutually exclusive"); 614 goto errout_free; 615 } 616 617 if (tb[FRA_UID_RANGE]) { 618 if (current_user_ns() != net->user_ns) { 619 err = -EPERM; 620 NL_SET_ERR_MSG(extack, "No permission to set uid"); 621 goto errout_free; 622 } 623 624 nlrule->uid_range = nla_get_kuid_range(tb); 625 626 if (!uid_range_set(&nlrule->uid_range) || 627 !uid_lte(nlrule->uid_range.start, nlrule->uid_range.end)) { 628 NL_SET_ERR_MSG(extack, "Invalid uid range"); 629 goto errout_free; 630 } 631 } else { 632 nlrule->uid_range = fib_kuid_range_unset; 633 } 634 635 if (tb[FRA_IP_PROTO]) 636 nlrule->ip_proto = nla_get_u8(tb[FRA_IP_PROTO]); 637 638 if (tb[FRA_SPORT_RANGE]) { 639 err = nla_get_port_range(tb[FRA_SPORT_RANGE], 640 &nlrule->sport_range); 641 if (err) { 642 NL_SET_ERR_MSG(extack, "Invalid sport range"); 643 goto errout_free; 644 } 645 } 646 647 if (tb[FRA_DPORT_RANGE]) { 648 err = nla_get_port_range(tb[FRA_DPORT_RANGE], 649 &nlrule->dport_range); 650 if (err) { 651 NL_SET_ERR_MSG(extack, "Invalid dport range"); 652 goto errout_free; 653 } 654 } 655 656 *rule = nlrule; 657 658 return 0; 659 660 errout_free: 661 kfree(nlrule); 662 errout: 663 return err; 664 } 665 666 static int fib_nl2rule_rtnl(struct fib_rule *nlrule, 667 struct fib_rules_ops *ops, 668 struct nlattr *tb[], 669 struct netlink_ext_ack *extack) 670 { 671 if (!tb[FRA_PRIORITY]) 672 nlrule->pref = fib_default_rule_pref(ops); 673 674 /* Backward jumps are prohibited to avoid endless loops */ 675 if (tb[FRA_GOTO] && nlrule->target <= nlrule->pref) { 676 NL_SET_ERR_MSG(extack, "Backward goto not supported"); 677 return -EINVAL; 678 } 679 680 if (tb[FRA_IIFNAME]) { 681 struct net_device *dev; 682 683 dev = __dev_get_by_name(nlrule->fr_net, nlrule->iifname); 684 if (dev) 685 nlrule->iifindex = dev->ifindex; 686 } 687 688 if (tb[FRA_OIFNAME]) { 689 struct net_device *dev; 690 691 dev = __dev_get_by_name(nlrule->fr_net, nlrule->oifname); 692 if (dev) 693 nlrule->oifindex = dev->ifindex; 694 } 695 696 return 0; 697 } 698 699 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh, 700 struct nlattr **tb, struct fib_rule *rule) 701 { 702 struct fib_rule *r; 703 704 list_for_each_entry(r, &ops->rules_list, list) { 705 if (r->action != rule->action) 706 continue; 707 708 if (r->table != rule->table) 709 continue; 710 711 if (r->pref != rule->pref) 712 continue; 713 714 if (memcmp(r->iifname, rule->iifname, IFNAMSIZ)) 715 continue; 716 717 if (memcmp(r->oifname, rule->oifname, IFNAMSIZ)) 718 continue; 719 720 if (r->mark != rule->mark) 721 continue; 722 723 if (r->suppress_ifgroup != rule->suppress_ifgroup) 724 continue; 725 726 if (r->suppress_prefixlen != rule->suppress_prefixlen) 727 continue; 728 729 if (r->mark_mask != rule->mark_mask) 730 continue; 731 732 if (r->tun_id != rule->tun_id) 733 continue; 734 735 if (r->l3mdev != rule->l3mdev) 736 continue; 737 738 if (!uid_eq(r->uid_range.start, rule->uid_range.start) || 739 !uid_eq(r->uid_range.end, rule->uid_range.end)) 740 continue; 741 742 if (r->ip_proto != rule->ip_proto) 743 continue; 744 745 if (r->proto != rule->proto) 746 continue; 747 748 if (!fib_rule_port_range_compare(&r->sport_range, 749 &rule->sport_range)) 750 continue; 751 752 if (!fib_rule_port_range_compare(&r->dport_range, 753 &rule->dport_range)) 754 continue; 755 756 if (!ops->compare(r, frh, tb)) 757 continue; 758 return 1; 759 } 760 return 0; 761 } 762 763 static const struct nla_policy fib_rule_policy[FRA_MAX + 1] = { 764 [FRA_UNSPEC] = { .strict_start_type = FRA_DPORT_RANGE + 1 }, 765 [FRA_IIFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ - 1 }, 766 [FRA_OIFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ - 1 }, 767 [FRA_PRIORITY] = { .type = NLA_U32 }, 768 [FRA_FWMARK] = { .type = NLA_U32 }, 769 [FRA_FLOW] = { .type = NLA_U32 }, 770 [FRA_TUN_ID] = { .type = NLA_U64 }, 771 [FRA_FWMASK] = { .type = NLA_U32 }, 772 [FRA_TABLE] = { .type = NLA_U32 }, 773 [FRA_SUPPRESS_PREFIXLEN] = { .type = NLA_U32 }, 774 [FRA_SUPPRESS_IFGROUP] = { .type = NLA_U32 }, 775 [FRA_GOTO] = { .type = NLA_U32 }, 776 [FRA_L3MDEV] = { .type = NLA_U8 }, 777 [FRA_UID_RANGE] = { .len = sizeof(struct fib_rule_uid_range) }, 778 [FRA_PROTOCOL] = { .type = NLA_U8 }, 779 [FRA_IP_PROTO] = { .type = NLA_U8 }, 780 [FRA_SPORT_RANGE] = { .len = sizeof(struct fib_rule_port_range) }, 781 [FRA_DPORT_RANGE] = { .len = sizeof(struct fib_rule_port_range) }, 782 [FRA_DSCP] = NLA_POLICY_MAX(NLA_U8, INET_DSCP_MASK >> 2), 783 [FRA_FLOWLABEL] = { .type = NLA_BE32 }, 784 [FRA_FLOWLABEL_MASK] = { .type = NLA_BE32 }, 785 }; 786 787 int fib_newrule(struct net *net, struct sk_buff *skb, struct nlmsghdr *nlh, 788 struct netlink_ext_ack *extack, bool rtnl_held) 789 { 790 struct fib_rule *rule = NULL, *r, *last = NULL; 791 struct fib_rule_hdr *frh = nlmsg_data(nlh); 792 int err = -EINVAL, unresolved = 0; 793 struct fib_rules_ops *ops = NULL; 794 struct nlattr *tb[FRA_MAX + 1]; 795 bool user_priority = false; 796 797 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) { 798 NL_SET_ERR_MSG(extack, "Invalid msg length"); 799 goto errout; 800 } 801 802 ops = lookup_rules_ops(net, frh->family); 803 if (!ops) { 804 err = -EAFNOSUPPORT; 805 NL_SET_ERR_MSG(extack, "Rule family not supported"); 806 goto errout; 807 } 808 809 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX, 810 fib_rule_policy, extack); 811 if (err < 0) { 812 NL_SET_ERR_MSG(extack, "Error parsing msg"); 813 goto errout; 814 } 815 816 err = fib_nl2rule(net, nlh, extack, ops, tb, &rule, &user_priority); 817 if (err) 818 goto errout; 819 820 if (!rtnl_held) 821 rtnl_net_lock(net); 822 823 err = fib_nl2rule_rtnl(rule, ops, tb, extack); 824 if (err) 825 goto errout_free; 826 827 if ((nlh->nlmsg_flags & NLM_F_EXCL) && 828 rule_exists(ops, frh, tb, rule)) { 829 err = -EEXIST; 830 goto errout_free; 831 } 832 833 err = ops->configure(rule, skb, frh, tb, extack); 834 if (err < 0) 835 goto errout_free; 836 837 err = call_fib_rule_notifiers(net, FIB_EVENT_RULE_ADD, rule, ops, 838 extack); 839 if (err < 0) 840 goto errout_free; 841 842 list_for_each_entry(r, &ops->rules_list, list) { 843 if (r->pref == rule->target) { 844 RCU_INIT_POINTER(rule->ctarget, r); 845 break; 846 } 847 } 848 849 if (rcu_dereference_protected(rule->ctarget, 1) == NULL) 850 unresolved = 1; 851 852 list_for_each_entry(r, &ops->rules_list, list) { 853 if (r->pref > rule->pref) 854 break; 855 last = r; 856 } 857 858 if (last) 859 list_add_rcu(&rule->list, &last->list); 860 else 861 list_add_rcu(&rule->list, &ops->rules_list); 862 863 if (ops->unresolved_rules) { 864 /* 865 * There are unresolved goto rules in the list, check if 866 * any of them are pointing to this new rule. 867 */ 868 list_for_each_entry(r, &ops->rules_list, list) { 869 if (r->action == FR_ACT_GOTO && 870 r->target == rule->pref && 871 rtnl_dereference(r->ctarget) == NULL) { 872 rcu_assign_pointer(r->ctarget, rule); 873 if (--ops->unresolved_rules == 0) 874 break; 875 } 876 } 877 } 878 879 if (rule->action == FR_ACT_GOTO) 880 ops->nr_goto_rules++; 881 882 if (unresolved) 883 ops->unresolved_rules++; 884 885 if (rule->tun_id) 886 ip_tunnel_need_metadata(); 887 888 fib_rule_get(rule); 889 890 if (!rtnl_held) 891 rtnl_net_unlock(net); 892 893 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid); 894 fib_rule_put(rule); 895 flush_route_cache(ops); 896 rules_ops_put(ops); 897 return 0; 898 899 errout_free: 900 if (!rtnl_held) 901 rtnl_net_unlock(net); 902 kfree(rule); 903 errout: 904 rules_ops_put(ops); 905 return err; 906 } 907 EXPORT_SYMBOL_GPL(fib_newrule); 908 909 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh, 910 struct netlink_ext_ack *extack) 911 { 912 return fib_newrule(sock_net(skb->sk), skb, nlh, extack, false); 913 } 914 915 int fib_delrule(struct net *net, struct sk_buff *skb, struct nlmsghdr *nlh, 916 struct netlink_ext_ack *extack, bool rtnl_held) 917 { 918 struct fib_rule *rule = NULL, *nlrule = NULL; 919 struct fib_rule_hdr *frh = nlmsg_data(nlh); 920 struct fib_rules_ops *ops = NULL; 921 struct nlattr *tb[FRA_MAX+1]; 922 bool user_priority = false; 923 int err = -EINVAL; 924 925 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) { 926 NL_SET_ERR_MSG(extack, "Invalid msg length"); 927 goto errout; 928 } 929 930 ops = lookup_rules_ops(net, frh->family); 931 if (ops == NULL) { 932 err = -EAFNOSUPPORT; 933 NL_SET_ERR_MSG(extack, "Rule family not supported"); 934 goto errout; 935 } 936 937 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX, 938 fib_rule_policy, extack); 939 if (err < 0) { 940 NL_SET_ERR_MSG(extack, "Error parsing msg"); 941 goto errout; 942 } 943 944 err = fib_nl2rule(net, nlh, extack, ops, tb, &nlrule, &user_priority); 945 if (err) 946 goto errout; 947 948 if (!rtnl_held) 949 rtnl_net_lock(net); 950 951 err = fib_nl2rule_rtnl(nlrule, ops, tb, extack); 952 if (err) 953 goto errout_free; 954 955 rule = rule_find(ops, frh, tb, nlrule, user_priority); 956 if (!rule) { 957 err = -ENOENT; 958 goto errout_free; 959 } 960 961 if (rule->flags & FIB_RULE_PERMANENT) { 962 err = -EPERM; 963 goto errout_free; 964 } 965 966 if (ops->delete) { 967 err = ops->delete(rule); 968 if (err) 969 goto errout_free; 970 } 971 972 if (rule->tun_id) 973 ip_tunnel_unneed_metadata(); 974 975 list_del_rcu(&rule->list); 976 977 if (rule->action == FR_ACT_GOTO) { 978 ops->nr_goto_rules--; 979 if (rtnl_dereference(rule->ctarget) == NULL) 980 ops->unresolved_rules--; 981 } 982 983 /* 984 * Check if this rule is a target to any of them. If so, 985 * adjust to the next one with the same preference or 986 * disable them. As this operation is eventually very 987 * expensive, it is only performed if goto rules, except 988 * current if it is goto rule, have actually been added. 989 */ 990 if (ops->nr_goto_rules > 0) { 991 struct fib_rule *n, *r; 992 993 n = list_next_entry(rule, list); 994 if (&n->list == &ops->rules_list || n->pref != rule->pref) 995 n = NULL; 996 list_for_each_entry(r, &ops->rules_list, list) { 997 if (rtnl_dereference(r->ctarget) != rule) 998 continue; 999 rcu_assign_pointer(r->ctarget, n); 1000 if (!n) 1001 ops->unresolved_rules++; 1002 } 1003 } 1004 1005 call_fib_rule_notifiers(net, FIB_EVENT_RULE_DEL, rule, ops, NULL); 1006 1007 if (!rtnl_held) 1008 rtnl_net_unlock(net); 1009 1010 notify_rule_change(RTM_DELRULE, rule, ops, nlh, NETLINK_CB(skb).portid); 1011 fib_rule_put(rule); 1012 flush_route_cache(ops); 1013 rules_ops_put(ops); 1014 kfree(nlrule); 1015 return 0; 1016 1017 errout_free: 1018 if (!rtnl_held) 1019 rtnl_net_unlock(net); 1020 kfree(nlrule); 1021 errout: 1022 rules_ops_put(ops); 1023 return err; 1024 } 1025 EXPORT_SYMBOL_GPL(fib_delrule); 1026 1027 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh, 1028 struct netlink_ext_ack *extack) 1029 { 1030 return fib_delrule(sock_net(skb->sk), skb, nlh, extack, false); 1031 } 1032 1033 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops, 1034 struct fib_rule *rule) 1035 { 1036 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)) 1037 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */ 1038 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */ 1039 + nla_total_size(4) /* FRA_PRIORITY */ 1040 + nla_total_size(4) /* FRA_TABLE */ 1041 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */ 1042 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */ 1043 + nla_total_size(4) /* FRA_FWMARK */ 1044 + nla_total_size(4) /* FRA_FWMASK */ 1045 + nla_total_size_64bit(8) /* FRA_TUN_ID */ 1046 + nla_total_size(sizeof(struct fib_kuid_range)) 1047 + nla_total_size(1) /* FRA_PROTOCOL */ 1048 + nla_total_size(1) /* FRA_IP_PROTO */ 1049 + nla_total_size(sizeof(struct fib_rule_port_range)) /* FRA_SPORT_RANGE */ 1050 + nla_total_size(sizeof(struct fib_rule_port_range)); /* FRA_DPORT_RANGE */ 1051 1052 if (ops->nlmsg_payload) 1053 payload += ops->nlmsg_payload(rule); 1054 1055 return payload; 1056 } 1057 1058 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule, 1059 u32 pid, u32 seq, int type, int flags, 1060 struct fib_rules_ops *ops) 1061 { 1062 struct nlmsghdr *nlh; 1063 struct fib_rule_hdr *frh; 1064 1065 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags); 1066 if (nlh == NULL) 1067 return -EMSGSIZE; 1068 1069 frh = nlmsg_data(nlh); 1070 frh->family = ops->family; 1071 frh->table = rule->table < 256 ? rule->table : RT_TABLE_COMPAT; 1072 if (nla_put_u32(skb, FRA_TABLE, rule->table)) 1073 goto nla_put_failure; 1074 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen)) 1075 goto nla_put_failure; 1076 frh->res1 = 0; 1077 frh->res2 = 0; 1078 frh->action = rule->action; 1079 frh->flags = rule->flags; 1080 1081 if (nla_put_u8(skb, FRA_PROTOCOL, rule->proto)) 1082 goto nla_put_failure; 1083 1084 if (rule->action == FR_ACT_GOTO && 1085 rcu_access_pointer(rule->ctarget) == NULL) 1086 frh->flags |= FIB_RULE_UNRESOLVED; 1087 1088 if (rule->iifname[0]) { 1089 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname)) 1090 goto nla_put_failure; 1091 if (rule->iifindex == -1) 1092 frh->flags |= FIB_RULE_IIF_DETACHED; 1093 } 1094 1095 if (rule->oifname[0]) { 1096 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname)) 1097 goto nla_put_failure; 1098 if (rule->oifindex == -1) 1099 frh->flags |= FIB_RULE_OIF_DETACHED; 1100 } 1101 1102 if ((rule->pref && 1103 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) || 1104 (rule->mark && 1105 nla_put_u32(skb, FRA_FWMARK, rule->mark)) || 1106 ((rule->mark_mask || rule->mark) && 1107 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) || 1108 (rule->target && 1109 nla_put_u32(skb, FRA_GOTO, rule->target)) || 1110 (rule->tun_id && 1111 nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) || 1112 (rule->l3mdev && 1113 nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) || 1114 (uid_range_set(&rule->uid_range) && 1115 nla_put_uid_range(skb, &rule->uid_range)) || 1116 (fib_rule_port_range_set(&rule->sport_range) && 1117 nla_put_port_range(skb, FRA_SPORT_RANGE, &rule->sport_range)) || 1118 (fib_rule_port_range_set(&rule->dport_range) && 1119 nla_put_port_range(skb, FRA_DPORT_RANGE, &rule->dport_range)) || 1120 (rule->ip_proto && nla_put_u8(skb, FRA_IP_PROTO, rule->ip_proto))) 1121 goto nla_put_failure; 1122 1123 if (rule->suppress_ifgroup != -1) { 1124 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup)) 1125 goto nla_put_failure; 1126 } 1127 1128 if (ops->fill(rule, skb, frh) < 0) 1129 goto nla_put_failure; 1130 1131 nlmsg_end(skb, nlh); 1132 return 0; 1133 1134 nla_put_failure: 1135 nlmsg_cancel(skb, nlh); 1136 return -EMSGSIZE; 1137 } 1138 1139 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb, 1140 struct fib_rules_ops *ops) 1141 { 1142 int idx = 0; 1143 struct fib_rule *rule; 1144 int err = 0; 1145 1146 rcu_read_lock(); 1147 list_for_each_entry_rcu(rule, &ops->rules_list, list) { 1148 if (idx < cb->args[1]) 1149 goto skip; 1150 1151 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid, 1152 cb->nlh->nlmsg_seq, RTM_NEWRULE, 1153 NLM_F_MULTI, ops); 1154 if (err) 1155 break; 1156 skip: 1157 idx++; 1158 } 1159 rcu_read_unlock(); 1160 cb->args[1] = idx; 1161 rules_ops_put(ops); 1162 1163 return err; 1164 } 1165 1166 static int fib_valid_dumprule_req(const struct nlmsghdr *nlh, 1167 struct netlink_ext_ack *extack) 1168 { 1169 struct fib_rule_hdr *frh; 1170 1171 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) { 1172 NL_SET_ERR_MSG(extack, "Invalid header for fib rule dump request"); 1173 return -EINVAL; 1174 } 1175 1176 frh = nlmsg_data(nlh); 1177 if (frh->dst_len || frh->src_len || frh->tos || frh->table || 1178 frh->res1 || frh->res2 || frh->action || frh->flags) { 1179 NL_SET_ERR_MSG(extack, 1180 "Invalid values in header for fib rule dump request"); 1181 return -EINVAL; 1182 } 1183 1184 if (nlmsg_attrlen(nlh, sizeof(*frh))) { 1185 NL_SET_ERR_MSG(extack, "Invalid data after header in fib rule dump request"); 1186 return -EINVAL; 1187 } 1188 1189 return 0; 1190 } 1191 1192 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb) 1193 { 1194 const struct nlmsghdr *nlh = cb->nlh; 1195 struct net *net = sock_net(skb->sk); 1196 struct fib_rules_ops *ops; 1197 int err, idx = 0, family; 1198 1199 if (cb->strict_check) { 1200 err = fib_valid_dumprule_req(nlh, cb->extack); 1201 1202 if (err < 0) 1203 return err; 1204 } 1205 1206 family = rtnl_msg_family(nlh); 1207 if (family != AF_UNSPEC) { 1208 /* Protocol specific dump request */ 1209 ops = lookup_rules_ops(net, family); 1210 if (ops == NULL) 1211 return -EAFNOSUPPORT; 1212 1213 return dump_rules(skb, cb, ops); 1214 } 1215 1216 err = 0; 1217 rcu_read_lock(); 1218 list_for_each_entry_rcu(ops, &net->rules_ops, list) { 1219 if (idx < cb->args[0] || !try_module_get(ops->owner)) 1220 goto skip; 1221 1222 err = dump_rules(skb, cb, ops); 1223 if (err < 0) 1224 break; 1225 1226 cb->args[1] = 0; 1227 skip: 1228 idx++; 1229 } 1230 rcu_read_unlock(); 1231 cb->args[0] = idx; 1232 1233 return err; 1234 } 1235 1236 static void notify_rule_change(int event, struct fib_rule *rule, 1237 struct fib_rules_ops *ops, struct nlmsghdr *nlh, 1238 u32 pid) 1239 { 1240 struct net *net; 1241 struct sk_buff *skb; 1242 int err = -ENOMEM; 1243 1244 net = ops->fro_net; 1245 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL); 1246 if (skb == NULL) 1247 goto errout; 1248 1249 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops); 1250 if (err < 0) { 1251 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */ 1252 WARN_ON(err == -EMSGSIZE); 1253 kfree_skb(skb); 1254 goto errout; 1255 } 1256 1257 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL); 1258 return; 1259 errout: 1260 rtnl_set_sk_err(net, ops->nlgroup, err); 1261 } 1262 1263 static void attach_rules(struct list_head *rules, struct net_device *dev) 1264 { 1265 struct fib_rule *rule; 1266 1267 list_for_each_entry(rule, rules, list) { 1268 if (rule->iifindex == -1 && 1269 strcmp(dev->name, rule->iifname) == 0) 1270 rule->iifindex = dev->ifindex; 1271 if (rule->oifindex == -1 && 1272 strcmp(dev->name, rule->oifname) == 0) 1273 rule->oifindex = dev->ifindex; 1274 } 1275 } 1276 1277 static void detach_rules(struct list_head *rules, struct net_device *dev) 1278 { 1279 struct fib_rule *rule; 1280 1281 list_for_each_entry(rule, rules, list) { 1282 if (rule->iifindex == dev->ifindex) 1283 rule->iifindex = -1; 1284 if (rule->oifindex == dev->ifindex) 1285 rule->oifindex = -1; 1286 } 1287 } 1288 1289 1290 static int fib_rules_event(struct notifier_block *this, unsigned long event, 1291 void *ptr) 1292 { 1293 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1294 struct net *net = dev_net(dev); 1295 struct fib_rules_ops *ops; 1296 1297 ASSERT_RTNL(); 1298 1299 switch (event) { 1300 case NETDEV_REGISTER: 1301 list_for_each_entry(ops, &net->rules_ops, list) 1302 attach_rules(&ops->rules_list, dev); 1303 break; 1304 1305 case NETDEV_CHANGENAME: 1306 list_for_each_entry(ops, &net->rules_ops, list) { 1307 detach_rules(&ops->rules_list, dev); 1308 attach_rules(&ops->rules_list, dev); 1309 } 1310 break; 1311 1312 case NETDEV_UNREGISTER: 1313 list_for_each_entry(ops, &net->rules_ops, list) 1314 detach_rules(&ops->rules_list, dev); 1315 break; 1316 } 1317 1318 return NOTIFY_DONE; 1319 } 1320 1321 static struct notifier_block fib_rules_notifier = { 1322 .notifier_call = fib_rules_event, 1323 }; 1324 1325 static int __net_init fib_rules_net_init(struct net *net) 1326 { 1327 INIT_LIST_HEAD(&net->rules_ops); 1328 spin_lock_init(&net->rules_mod_lock); 1329 return 0; 1330 } 1331 1332 static void __net_exit fib_rules_net_exit(struct net *net) 1333 { 1334 WARN_ON_ONCE(!list_empty(&net->rules_ops)); 1335 } 1336 1337 static struct pernet_operations fib_rules_net_ops = { 1338 .init = fib_rules_net_init, 1339 .exit = fib_rules_net_exit, 1340 }; 1341 1342 static const struct rtnl_msg_handler fib_rules_rtnl_msg_handlers[] __initconst = { 1343 {.msgtype = RTM_NEWRULE, .doit = fib_nl_newrule, 1344 .flags = RTNL_FLAG_DOIT_PERNET}, 1345 {.msgtype = RTM_DELRULE, .doit = fib_nl_delrule, 1346 .flags = RTNL_FLAG_DOIT_PERNET}, 1347 {.msgtype = RTM_GETRULE, .dumpit = fib_nl_dumprule, 1348 .flags = RTNL_FLAG_DUMP_UNLOCKED}, 1349 }; 1350 1351 static int __init fib_rules_init(void) 1352 { 1353 int err; 1354 1355 rtnl_register_many(fib_rules_rtnl_msg_handlers); 1356 1357 err = register_pernet_subsys(&fib_rules_net_ops); 1358 if (err < 0) 1359 goto fail; 1360 1361 err = register_netdevice_notifier(&fib_rules_notifier); 1362 if (err < 0) 1363 goto fail_unregister; 1364 1365 return 0; 1366 1367 fail_unregister: 1368 unregister_pernet_subsys(&fib_rules_net_ops); 1369 fail: 1370 rtnl_unregister_many(fib_rules_rtnl_msg_handlers); 1371 return err; 1372 } 1373 1374 subsys_initcall(fib_rules_init); 1375