1 /* 2 * net/core/fib_rules.c Generic Routing Rules 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License as 6 * published by the Free Software Foundation, version 2. 7 * 8 * Authors: Thomas Graf <tgraf@suug.ch> 9 */ 10 11 #include <linux/types.h> 12 #include <linux/kernel.h> 13 #include <linux/slab.h> 14 #include <linux/list.h> 15 #include <net/net_namespace.h> 16 #include <net/sock.h> 17 #include <net/fib_rules.h> 18 19 int fib_default_rule_add(struct fib_rules_ops *ops, 20 u32 pref, u32 table, u32 flags) 21 { 22 struct fib_rule *r; 23 24 r = kzalloc(ops->rule_size, GFP_KERNEL); 25 if (r == NULL) 26 return -ENOMEM; 27 28 atomic_set(&r->refcnt, 1); 29 r->action = FR_ACT_TO_TBL; 30 r->pref = pref; 31 r->table = table; 32 r->flags = flags; 33 r->fr_net = hold_net(ops->fro_net); 34 35 /* The lock is not required here, the list in unreacheable 36 * at the moment this function is called */ 37 list_add_tail(&r->list, &ops->rules_list); 38 return 0; 39 } 40 EXPORT_SYMBOL(fib_default_rule_add); 41 42 static void notify_rule_change(int event, struct fib_rule *rule, 43 struct fib_rules_ops *ops, struct nlmsghdr *nlh, 44 u32 pid); 45 46 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family) 47 { 48 struct fib_rules_ops *ops; 49 50 rcu_read_lock(); 51 list_for_each_entry_rcu(ops, &net->rules_ops, list) { 52 if (ops->family == family) { 53 if (!try_module_get(ops->owner)) 54 ops = NULL; 55 rcu_read_unlock(); 56 return ops; 57 } 58 } 59 rcu_read_unlock(); 60 61 return NULL; 62 } 63 64 static void rules_ops_put(struct fib_rules_ops *ops) 65 { 66 if (ops) 67 module_put(ops->owner); 68 } 69 70 static void flush_route_cache(struct fib_rules_ops *ops) 71 { 72 if (ops->flush_cache) 73 ops->flush_cache(ops); 74 } 75 76 static int __fib_rules_register(struct fib_rules_ops *ops) 77 { 78 int err = -EEXIST; 79 struct fib_rules_ops *o; 80 struct net *net; 81 82 net = ops->fro_net; 83 84 if (ops->rule_size < sizeof(struct fib_rule)) 85 return -EINVAL; 86 87 if (ops->match == NULL || ops->configure == NULL || 88 ops->compare == NULL || ops->fill == NULL || 89 ops->action == NULL) 90 return -EINVAL; 91 92 spin_lock(&net->rules_mod_lock); 93 list_for_each_entry(o, &net->rules_ops, list) 94 if (ops->family == o->family) 95 goto errout; 96 97 hold_net(net); 98 list_add_tail_rcu(&ops->list, &net->rules_ops); 99 err = 0; 100 errout: 101 spin_unlock(&net->rules_mod_lock); 102 103 return err; 104 } 105 106 struct fib_rules_ops * 107 fib_rules_register(struct fib_rules_ops *tmpl, struct net *net) 108 { 109 struct fib_rules_ops *ops; 110 int err; 111 112 ops = kmemdup(tmpl, sizeof (*ops), GFP_KERNEL); 113 if (ops == NULL) 114 return ERR_PTR(-ENOMEM); 115 116 INIT_LIST_HEAD(&ops->rules_list); 117 ops->fro_net = net; 118 119 err = __fib_rules_register(ops); 120 if (err) { 121 kfree(ops); 122 ops = ERR_PTR(err); 123 } 124 125 return ops; 126 } 127 128 EXPORT_SYMBOL_GPL(fib_rules_register); 129 130 void fib_rules_cleanup_ops(struct fib_rules_ops *ops) 131 { 132 struct fib_rule *rule, *tmp; 133 134 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) { 135 list_del_rcu(&rule->list); 136 fib_rule_put(rule); 137 } 138 } 139 EXPORT_SYMBOL_GPL(fib_rules_cleanup_ops); 140 141 static void fib_rules_put_rcu(struct rcu_head *head) 142 { 143 struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu); 144 struct net *net = ops->fro_net; 145 146 release_net(net); 147 kfree(ops); 148 } 149 150 void fib_rules_unregister(struct fib_rules_ops *ops) 151 { 152 struct net *net = ops->fro_net; 153 154 spin_lock(&net->rules_mod_lock); 155 list_del_rcu(&ops->list); 156 fib_rules_cleanup_ops(ops); 157 spin_unlock(&net->rules_mod_lock); 158 159 call_rcu(&ops->rcu, fib_rules_put_rcu); 160 } 161 162 EXPORT_SYMBOL_GPL(fib_rules_unregister); 163 164 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops, 165 struct flowi *fl, int flags) 166 { 167 int ret = 0; 168 169 if (rule->iifindex && (rule->iifindex != fl->iif)) 170 goto out; 171 172 if (rule->oifindex && (rule->oifindex != fl->oif)) 173 goto out; 174 175 if ((rule->mark ^ fl->mark) & rule->mark_mask) 176 goto out; 177 178 ret = ops->match(rule, fl, flags); 179 out: 180 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret; 181 } 182 183 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl, 184 int flags, struct fib_lookup_arg *arg) 185 { 186 struct fib_rule *rule; 187 int err; 188 189 rcu_read_lock(); 190 191 list_for_each_entry_rcu(rule, &ops->rules_list, list) { 192 jumped: 193 if (!fib_rule_match(rule, ops, fl, flags)) 194 continue; 195 196 if (rule->action == FR_ACT_GOTO) { 197 struct fib_rule *target; 198 199 target = rcu_dereference(rule->ctarget); 200 if (target == NULL) { 201 continue; 202 } else { 203 rule = target; 204 goto jumped; 205 } 206 } else if (rule->action == FR_ACT_NOP) 207 continue; 208 else 209 err = ops->action(rule, fl, flags, arg); 210 211 if (err != -EAGAIN) { 212 fib_rule_get(rule); 213 arg->rule = rule; 214 goto out; 215 } 216 } 217 218 err = -ESRCH; 219 out: 220 rcu_read_unlock(); 221 222 return err; 223 } 224 225 EXPORT_SYMBOL_GPL(fib_rules_lookup); 226 227 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb, 228 struct fib_rules_ops *ops) 229 { 230 int err = -EINVAL; 231 232 if (frh->src_len) 233 if (tb[FRA_SRC] == NULL || 234 frh->src_len > (ops->addr_size * 8) || 235 nla_len(tb[FRA_SRC]) != ops->addr_size) 236 goto errout; 237 238 if (frh->dst_len) 239 if (tb[FRA_DST] == NULL || 240 frh->dst_len > (ops->addr_size * 8) || 241 nla_len(tb[FRA_DST]) != ops->addr_size) 242 goto errout; 243 244 err = 0; 245 errout: 246 return err; 247 } 248 249 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg) 250 { 251 struct net *net = sock_net(skb->sk); 252 struct fib_rule_hdr *frh = nlmsg_data(nlh); 253 struct fib_rules_ops *ops = NULL; 254 struct fib_rule *rule, *r, *last = NULL; 255 struct nlattr *tb[FRA_MAX+1]; 256 int err = -EINVAL, unresolved = 0; 257 258 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) 259 goto errout; 260 261 ops = lookup_rules_ops(net, frh->family); 262 if (ops == NULL) { 263 err = -EAFNOSUPPORT; 264 goto errout; 265 } 266 267 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy); 268 if (err < 0) 269 goto errout; 270 271 err = validate_rulemsg(frh, tb, ops); 272 if (err < 0) 273 goto errout; 274 275 rule = kzalloc(ops->rule_size, GFP_KERNEL); 276 if (rule == NULL) { 277 err = -ENOMEM; 278 goto errout; 279 } 280 rule->fr_net = hold_net(net); 281 282 if (tb[FRA_PRIORITY]) 283 rule->pref = nla_get_u32(tb[FRA_PRIORITY]); 284 285 if (tb[FRA_IIFNAME]) { 286 struct net_device *dev; 287 288 rule->iifindex = -1; 289 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ); 290 dev = __dev_get_by_name(net, rule->iifname); 291 if (dev) 292 rule->iifindex = dev->ifindex; 293 } 294 295 if (tb[FRA_OIFNAME]) { 296 struct net_device *dev; 297 298 rule->oifindex = -1; 299 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ); 300 dev = __dev_get_by_name(net, rule->oifname); 301 if (dev) 302 rule->oifindex = dev->ifindex; 303 } 304 305 if (tb[FRA_FWMARK]) { 306 rule->mark = nla_get_u32(tb[FRA_FWMARK]); 307 if (rule->mark) 308 /* compatibility: if the mark value is non-zero all bits 309 * are compared unless a mask is explicitly specified. 310 */ 311 rule->mark_mask = 0xFFFFFFFF; 312 } 313 314 if (tb[FRA_FWMASK]) 315 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]); 316 317 rule->action = frh->action; 318 rule->flags = frh->flags; 319 rule->table = frh_get_table(frh, tb); 320 321 if (!tb[FRA_PRIORITY] && ops->default_pref) 322 rule->pref = ops->default_pref(ops); 323 324 err = -EINVAL; 325 if (tb[FRA_GOTO]) { 326 if (rule->action != FR_ACT_GOTO) 327 goto errout_free; 328 329 rule->target = nla_get_u32(tb[FRA_GOTO]); 330 /* Backward jumps are prohibited to avoid endless loops */ 331 if (rule->target <= rule->pref) 332 goto errout_free; 333 334 list_for_each_entry(r, &ops->rules_list, list) { 335 if (r->pref == rule->target) { 336 rule->ctarget = r; 337 break; 338 } 339 } 340 341 if (rule->ctarget == NULL) 342 unresolved = 1; 343 } else if (rule->action == FR_ACT_GOTO) 344 goto errout_free; 345 346 err = ops->configure(rule, skb, frh, tb); 347 if (err < 0) 348 goto errout_free; 349 350 list_for_each_entry(r, &ops->rules_list, list) { 351 if (r->pref > rule->pref) 352 break; 353 last = r; 354 } 355 356 fib_rule_get(rule); 357 358 if (ops->unresolved_rules) { 359 /* 360 * There are unresolved goto rules in the list, check if 361 * any of them are pointing to this new rule. 362 */ 363 list_for_each_entry(r, &ops->rules_list, list) { 364 if (r->action == FR_ACT_GOTO && 365 r->target == rule->pref) { 366 BUG_ON(r->ctarget != NULL); 367 rcu_assign_pointer(r->ctarget, rule); 368 if (--ops->unresolved_rules == 0) 369 break; 370 } 371 } 372 } 373 374 if (rule->action == FR_ACT_GOTO) 375 ops->nr_goto_rules++; 376 377 if (unresolved) 378 ops->unresolved_rules++; 379 380 if (last) 381 list_add_rcu(&rule->list, &last->list); 382 else 383 list_add_rcu(&rule->list, &ops->rules_list); 384 385 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid); 386 flush_route_cache(ops); 387 rules_ops_put(ops); 388 return 0; 389 390 errout_free: 391 release_net(rule->fr_net); 392 kfree(rule); 393 errout: 394 rules_ops_put(ops); 395 return err; 396 } 397 398 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg) 399 { 400 struct net *net = sock_net(skb->sk); 401 struct fib_rule_hdr *frh = nlmsg_data(nlh); 402 struct fib_rules_ops *ops = NULL; 403 struct fib_rule *rule, *tmp; 404 struct nlattr *tb[FRA_MAX+1]; 405 int err = -EINVAL; 406 407 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) 408 goto errout; 409 410 ops = lookup_rules_ops(net, frh->family); 411 if (ops == NULL) { 412 err = -EAFNOSUPPORT; 413 goto errout; 414 } 415 416 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy); 417 if (err < 0) 418 goto errout; 419 420 err = validate_rulemsg(frh, tb, ops); 421 if (err < 0) 422 goto errout; 423 424 list_for_each_entry(rule, &ops->rules_list, list) { 425 if (frh->action && (frh->action != rule->action)) 426 continue; 427 428 if (frh->table && (frh_get_table(frh, tb) != rule->table)) 429 continue; 430 431 if (tb[FRA_PRIORITY] && 432 (rule->pref != nla_get_u32(tb[FRA_PRIORITY]))) 433 continue; 434 435 if (tb[FRA_IIFNAME] && 436 nla_strcmp(tb[FRA_IIFNAME], rule->iifname)) 437 continue; 438 439 if (tb[FRA_OIFNAME] && 440 nla_strcmp(tb[FRA_OIFNAME], rule->oifname)) 441 continue; 442 443 if (tb[FRA_FWMARK] && 444 (rule->mark != nla_get_u32(tb[FRA_FWMARK]))) 445 continue; 446 447 if (tb[FRA_FWMASK] && 448 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK]))) 449 continue; 450 451 if (!ops->compare(rule, frh, tb)) 452 continue; 453 454 if (rule->flags & FIB_RULE_PERMANENT) { 455 err = -EPERM; 456 goto errout; 457 } 458 459 list_del_rcu(&rule->list); 460 461 if (rule->action == FR_ACT_GOTO) 462 ops->nr_goto_rules--; 463 464 /* 465 * Check if this rule is a target to any of them. If so, 466 * disable them. As this operation is eventually very 467 * expensive, it is only performed if goto rules have 468 * actually been added. 469 */ 470 if (ops->nr_goto_rules > 0) { 471 list_for_each_entry(tmp, &ops->rules_list, list) { 472 if (tmp->ctarget == rule) { 473 rcu_assign_pointer(tmp->ctarget, NULL); 474 ops->unresolved_rules++; 475 } 476 } 477 } 478 479 synchronize_rcu(); 480 notify_rule_change(RTM_DELRULE, rule, ops, nlh, 481 NETLINK_CB(skb).pid); 482 fib_rule_put(rule); 483 flush_route_cache(ops); 484 rules_ops_put(ops); 485 return 0; 486 } 487 488 err = -ENOENT; 489 errout: 490 rules_ops_put(ops); 491 return err; 492 } 493 494 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops, 495 struct fib_rule *rule) 496 { 497 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)) 498 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */ 499 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */ 500 + nla_total_size(4) /* FRA_PRIORITY */ 501 + nla_total_size(4) /* FRA_TABLE */ 502 + nla_total_size(4) /* FRA_FWMARK */ 503 + nla_total_size(4); /* FRA_FWMASK */ 504 505 if (ops->nlmsg_payload) 506 payload += ops->nlmsg_payload(rule); 507 508 return payload; 509 } 510 511 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule, 512 u32 pid, u32 seq, int type, int flags, 513 struct fib_rules_ops *ops) 514 { 515 struct nlmsghdr *nlh; 516 struct fib_rule_hdr *frh; 517 518 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags); 519 if (nlh == NULL) 520 return -EMSGSIZE; 521 522 frh = nlmsg_data(nlh); 523 frh->table = rule->table; 524 NLA_PUT_U32(skb, FRA_TABLE, rule->table); 525 frh->res1 = 0; 526 frh->res2 = 0; 527 frh->action = rule->action; 528 frh->flags = rule->flags; 529 530 if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL) 531 frh->flags |= FIB_RULE_UNRESOLVED; 532 533 if (rule->iifname[0]) { 534 NLA_PUT_STRING(skb, FRA_IIFNAME, rule->iifname); 535 536 if (rule->iifindex == -1) 537 frh->flags |= FIB_RULE_IIF_DETACHED; 538 } 539 540 if (rule->oifname[0]) { 541 NLA_PUT_STRING(skb, FRA_OIFNAME, rule->oifname); 542 543 if (rule->oifindex == -1) 544 frh->flags |= FIB_RULE_OIF_DETACHED; 545 } 546 547 if (rule->pref) 548 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref); 549 550 if (rule->mark) 551 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark); 552 553 if (rule->mark_mask || rule->mark) 554 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask); 555 556 if (rule->target) 557 NLA_PUT_U32(skb, FRA_GOTO, rule->target); 558 559 if (ops->fill(rule, skb, frh) < 0) 560 goto nla_put_failure; 561 562 return nlmsg_end(skb, nlh); 563 564 nla_put_failure: 565 nlmsg_cancel(skb, nlh); 566 return -EMSGSIZE; 567 } 568 569 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb, 570 struct fib_rules_ops *ops) 571 { 572 int idx = 0; 573 struct fib_rule *rule; 574 575 list_for_each_entry(rule, &ops->rules_list, list) { 576 if (idx < cb->args[1]) 577 goto skip; 578 579 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid, 580 cb->nlh->nlmsg_seq, RTM_NEWRULE, 581 NLM_F_MULTI, ops) < 0) 582 break; 583 skip: 584 idx++; 585 } 586 cb->args[1] = idx; 587 rules_ops_put(ops); 588 589 return skb->len; 590 } 591 592 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb) 593 { 594 struct net *net = sock_net(skb->sk); 595 struct fib_rules_ops *ops; 596 int idx = 0, family; 597 598 family = rtnl_msg_family(cb->nlh); 599 if (family != AF_UNSPEC) { 600 /* Protocol specific dump request */ 601 ops = lookup_rules_ops(net, family); 602 if (ops == NULL) 603 return -EAFNOSUPPORT; 604 605 return dump_rules(skb, cb, ops); 606 } 607 608 rcu_read_lock(); 609 list_for_each_entry_rcu(ops, &net->rules_ops, list) { 610 if (idx < cb->args[0] || !try_module_get(ops->owner)) 611 goto skip; 612 613 if (dump_rules(skb, cb, ops) < 0) 614 break; 615 616 cb->args[1] = 0; 617 skip: 618 idx++; 619 } 620 rcu_read_unlock(); 621 cb->args[0] = idx; 622 623 return skb->len; 624 } 625 626 static void notify_rule_change(int event, struct fib_rule *rule, 627 struct fib_rules_ops *ops, struct nlmsghdr *nlh, 628 u32 pid) 629 { 630 struct net *net; 631 struct sk_buff *skb; 632 int err = -ENOBUFS; 633 634 net = ops->fro_net; 635 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL); 636 if (skb == NULL) 637 goto errout; 638 639 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops); 640 if (err < 0) { 641 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */ 642 WARN_ON(err == -EMSGSIZE); 643 kfree_skb(skb); 644 goto errout; 645 } 646 647 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL); 648 return; 649 errout: 650 if (err < 0) 651 rtnl_set_sk_err(net, ops->nlgroup, err); 652 } 653 654 static void attach_rules(struct list_head *rules, struct net_device *dev) 655 { 656 struct fib_rule *rule; 657 658 list_for_each_entry(rule, rules, list) { 659 if (rule->iifindex == -1 && 660 strcmp(dev->name, rule->iifname) == 0) 661 rule->iifindex = dev->ifindex; 662 if (rule->oifindex == -1 && 663 strcmp(dev->name, rule->oifname) == 0) 664 rule->oifindex = dev->ifindex; 665 } 666 } 667 668 static void detach_rules(struct list_head *rules, struct net_device *dev) 669 { 670 struct fib_rule *rule; 671 672 list_for_each_entry(rule, rules, list) { 673 if (rule->iifindex == dev->ifindex) 674 rule->iifindex = -1; 675 if (rule->oifindex == dev->ifindex) 676 rule->oifindex = -1; 677 } 678 } 679 680 681 static int fib_rules_event(struct notifier_block *this, unsigned long event, 682 void *ptr) 683 { 684 struct net_device *dev = ptr; 685 struct net *net = dev_net(dev); 686 struct fib_rules_ops *ops; 687 688 ASSERT_RTNL(); 689 rcu_read_lock(); 690 691 switch (event) { 692 case NETDEV_REGISTER: 693 list_for_each_entry(ops, &net->rules_ops, list) 694 attach_rules(&ops->rules_list, dev); 695 break; 696 697 case NETDEV_UNREGISTER: 698 list_for_each_entry(ops, &net->rules_ops, list) 699 detach_rules(&ops->rules_list, dev); 700 break; 701 } 702 703 rcu_read_unlock(); 704 705 return NOTIFY_DONE; 706 } 707 708 static struct notifier_block fib_rules_notifier = { 709 .notifier_call = fib_rules_event, 710 }; 711 712 static int __net_init fib_rules_net_init(struct net *net) 713 { 714 INIT_LIST_HEAD(&net->rules_ops); 715 spin_lock_init(&net->rules_mod_lock); 716 return 0; 717 } 718 719 static struct pernet_operations fib_rules_net_ops = { 720 .init = fib_rules_net_init, 721 }; 722 723 static int __init fib_rules_init(void) 724 { 725 int err; 726 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL); 727 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL); 728 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule); 729 730 err = register_pernet_subsys(&fib_rules_net_ops); 731 if (err < 0) 732 goto fail; 733 734 err = register_netdevice_notifier(&fib_rules_notifier); 735 if (err < 0) 736 goto fail_unregister; 737 738 return 0; 739 740 fail_unregister: 741 unregister_pernet_subsys(&fib_rules_net_ops); 742 fail: 743 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE); 744 rtnl_unregister(PF_UNSPEC, RTM_DELRULE); 745 rtnl_unregister(PF_UNSPEC, RTM_GETRULE); 746 return err; 747 } 748 749 subsys_initcall(fib_rules_init); 750