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