net/core/fib_rules.c

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