SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  33         /* The lock is not required here, the list in unreacheable
  34          * at the moment this function is called */
  35         list_add_tail(&r->list, &ops->rules_list);
  36         return 0;
  37 }
  38 EXPORT_SYMBOL(fib_default_rule_add);
  39
  40 static void notify_rule_change(struct net *net, int event,
  41                                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 net *net, struct fib_rules_ops *ops)
  76 {
  77         int err = -EEXIST;
  78         struct fib_rules_ops *o;
  79
  80         if (ops->rule_size < sizeof(struct fib_rule))
  81                 return -EINVAL;
  82
  83         if (ops->match == NULL || ops->configure == NULL ||
  84             ops->compare == NULL || ops->fill == NULL ||
  85             ops->action == NULL)
  86                 return -EINVAL;
  87
  88         spin_lock(&net->rules_mod_lock);
  89         list_for_each_entry(o, &net->rules_ops, list)
  90                 if (ops->family == o->family)
  91                         goto errout;
  92
  93         hold_net(net);
  94         list_add_tail_rcu(&ops->list, &net->rules_ops);
  95         err = 0;
  96 errout:
  97         spin_unlock(&net->rules_mod_lock);
  98
  99         return err;
 100 }
 101
 102 EXPORT_SYMBOL_GPL(fib_rules_register);
 103
 104 void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
 105 {
 106         struct fib_rule *rule, *tmp;
 107
 108         list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
 109                 list_del_rcu(&rule->list);
 110                 fib_rule_put(rule);
 111         }
 112 }
 113 EXPORT_SYMBOL_GPL(fib_rules_cleanup_ops);
 114
 115 void fib_rules_unregister(struct net *net, struct fib_rules_ops *ops)
 116 {
 117
 118         spin_lock(&net->rules_mod_lock);
 119         list_del_rcu(&ops->list);
 120         fib_rules_cleanup_ops(ops);
 121         spin_unlock(&net->rules_mod_lock);
 122
 123         synchronize_rcu();
 124         release_net(net);
 125 }
 126
 127 EXPORT_SYMBOL_GPL(fib_rules_unregister);
 128
 129 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
 130                           struct flowi *fl, int flags)
 131 {
 132         int ret = 0;
 133
 134         if (rule->ifindex && (rule->ifindex != fl->iif))
 135                 goto out;
 136
 137         if ((rule->mark ^ fl->mark) & rule->mark_mask)
 138                 goto out;
 139
 140         ret = ops->match(rule, fl, flags);
 141 out:
 142         return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
 143 }
 144
 145 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
 146                      int flags, struct fib_lookup_arg *arg)
 147 {
 148         struct fib_rule *rule;
 149         int err;
 150
 151         rcu_read_lock();
 152
 153         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 154 jumped:
 155                 if (!fib_rule_match(rule, ops, fl, flags))
 156                         continue;
 157
 158                 if (rule->action == FR_ACT_GOTO) {
 159                         struct fib_rule *target;
 160
 161                         target = rcu_dereference(rule->ctarget);
 162                         if (target == NULL) {
 163                                 continue;
 164                         } else {
 165                                 rule = target;
 166                                 goto jumped;
 167                         }
 168                 } else if (rule->action == FR_ACT_NOP)
 169                         continue;
 170                 else
 171                         err = ops->action(rule, fl, flags, arg);
 172
 173                 if (err != -EAGAIN) {
 174                         fib_rule_get(rule);
 175                         arg->rule = rule;
 176                         goto out;
 177                 }
 178         }
 179
 180         err = -ESRCH;
 181 out:
 182         rcu_read_unlock();
 183
 184         return err;
 185 }
 186
 187 EXPORT_SYMBOL_GPL(fib_rules_lookup);
 188
 189 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
 190                             struct fib_rules_ops *ops)
 191 {
 192         int err = -EINVAL;
 193
 194         if (frh->src_len)
 195                 if (tb[FRA_SRC] == NULL ||
 196                     frh->src_len > (ops->addr_size * 8) ||
 197                     nla_len(tb[FRA_SRC]) != ops->addr_size)
 198                         goto errout;
 199
 200         if (frh->dst_len)
 201                 if (tb[FRA_DST] == NULL ||
 202                     frh->dst_len > (ops->addr_size * 8) ||
 203                     nla_len(tb[FRA_DST]) != ops->addr_size)
 204                         goto errout;
 205
 206         err = 0;
 207 errout:
 208         return err;
 209 }
 210
 211 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 212 {
 213         struct net *net = skb->sk->sk_net;
 214         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 215         struct fib_rules_ops *ops = NULL;
 216         struct fib_rule *rule, *r, *last = NULL;
 217         struct nlattr *tb[FRA_MAX+1];
 218         int err = -EINVAL, unresolved = 0;
 219
 220         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 221                 goto errout;
 222
 223         ops = lookup_rules_ops(net, frh->family);
 224         if (ops == NULL) {
 225                 err = EAFNOSUPPORT;
 226                 goto errout;
 227         }
 228
 229         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 230         if (err < 0)
 231                 goto errout;
 232
 233         err = validate_rulemsg(frh, tb, ops);
 234         if (err < 0)
 235                 goto errout;
 236
 237         rule = kzalloc(ops->rule_size, GFP_KERNEL);
 238         if (rule == NULL) {
 239                 err = -ENOMEM;
 240                 goto errout;
 241         }
 242
 243         if (tb[FRA_PRIORITY])
 244                 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
 245
 246         if (tb[FRA_IFNAME]) {
 247                 struct net_device *dev;
 248
 249                 rule->ifindex = -1;
 250                 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
 251                 dev = __dev_get_by_name(net, rule->ifname);
 252                 if (dev)
 253                         rule->ifindex = dev->ifindex;
 254         }
 255
 256         if (tb[FRA_FWMARK]) {
 257                 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
 258                 if (rule->mark)
 259                         /* compatibility: if the mark value is non-zero all bits
 260                          * are compared unless a mask is explicitly specified.
 261                          */
 262                         rule->mark_mask = 0xFFFFFFFF;
 263         }
 264
 265         if (tb[FRA_FWMASK])
 266                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
 267
 268         rule->action = frh->action;
 269         rule->flags = frh->flags;
 270         rule->table = frh_get_table(frh, tb);
 271
 272         if (!rule->pref && ops->default_pref)
 273                 rule->pref = ops->default_pref(ops);
 274
 275         err = -EINVAL;
 276         if (tb[FRA_GOTO]) {
 277                 if (rule->action != FR_ACT_GOTO)
 278                         goto errout_free;
 279
 280                 rule->target = nla_get_u32(tb[FRA_GOTO]);
 281                 /* Backward jumps are prohibited to avoid endless loops */
 282                 if (rule->target <= rule->pref)
 283                         goto errout_free;
 284
 285                 list_for_each_entry(r, &ops->rules_list, list) {
 286                         if (r->pref == rule->target) {
 287                                 rule->ctarget = r;
 288                                 break;
 289                         }
 290                 }
 291
 292                 if (rule->ctarget == NULL)
 293                         unresolved = 1;
 294         } else if (rule->action == FR_ACT_GOTO)
 295                 goto errout_free;
 296
 297         err = ops->configure(rule, skb, nlh, frh, tb);
 298         if (err < 0)
 299                 goto errout_free;
 300
 301         list_for_each_entry(r, &ops->rules_list, list) {
 302                 if (r->pref > rule->pref)
 303                         break;
 304                 last = r;
 305         }
 306
 307         fib_rule_get(rule);
 308
 309         if (ops->unresolved_rules) {
 310                 /*
 311                  * There are unresolved goto rules in the list, check if
 312                  * any of them are pointing to this new rule.
 313                  */
 314                 list_for_each_entry(r, &ops->rules_list, list) {
 315                         if (r->action == FR_ACT_GOTO &&
 316                             r->target == rule->pref) {
 317                                 BUG_ON(r->ctarget != NULL);
 318                                 rcu_assign_pointer(r->ctarget, rule);
 319                                 if (--ops->unresolved_rules == 0)
 320                                         break;
 321                         }
 322                 }
 323         }
 324
 325         if (rule->action == FR_ACT_GOTO)
 326                 ops->nr_goto_rules++;
 327
 328         if (unresolved)
 329                 ops->unresolved_rules++;
 330
 331         if (last)
 332                 list_add_rcu(&rule->list, &last->list);
 333         else
 334                 list_add_rcu(&rule->list, &ops->rules_list);
 335
 336         notify_rule_change(net, RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
 337         flush_route_cache(ops);
 338         rules_ops_put(ops);
 339         return 0;
 340
 341 errout_free:
 342         kfree(rule);
 343 errout:
 344         rules_ops_put(ops);
 345         return err;
 346 }
 347
 348 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 349 {
 350         struct net *net = skb->sk->sk_net;
 351         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 352         struct fib_rules_ops *ops = NULL;
 353         struct fib_rule *rule, *tmp;
 354         struct nlattr *tb[FRA_MAX+1];
 355         int err = -EINVAL;
 356
 357         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 358                 goto errout;
 359
 360         ops = lookup_rules_ops(net, frh->family);
 361         if (ops == NULL) {
 362                 err = EAFNOSUPPORT;
 363                 goto errout;
 364         }
 365
 366         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 367         if (err < 0)
 368                 goto errout;
 369
 370         err = validate_rulemsg(frh, tb, ops);
 371         if (err < 0)
 372                 goto errout;
 373
 374         list_for_each_entry(rule, &ops->rules_list, list) {
 375                 if (frh->action && (frh->action != rule->action))
 376                         continue;
 377
 378                 if (frh->table && (frh_get_table(frh, tb) != rule->table))
 379                         continue;
 380
 381                 if (tb[FRA_PRIORITY] &&
 382                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
 383                         continue;
 384
 385                 if (tb[FRA_IFNAME] &&
 386                     nla_strcmp(tb[FRA_IFNAME], rule->ifname))
 387                         continue;
 388
 389                 if (tb[FRA_FWMARK] &&
 390                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
 391                         continue;
 392
 393                 if (tb[FRA_FWMASK] &&
 394                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
 395                         continue;
 396
 397                 if (!ops->compare(rule, frh, tb))
 398                         continue;
 399
 400                 if (rule->flags & FIB_RULE_PERMANENT) {
 401                         err = -EPERM;
 402                         goto errout;
 403                 }
 404
 405                 list_del_rcu(&rule->list);
 406
 407                 if (rule->action == FR_ACT_GOTO)
 408                         ops->nr_goto_rules--;
 409
 410                 /*
 411                  * Check if this rule is a target to any of them. If so,
 412                  * disable them. As this operation is eventually very
 413                  * expensive, it is only performed if goto rules have
 414                  * actually been added.
 415                  */
 416                 if (ops->nr_goto_rules > 0) {
 417                         list_for_each_entry(tmp, &ops->rules_list, list) {
 418                                 if (tmp->ctarget == rule) {
 419                                         rcu_assign_pointer(tmp->ctarget, NULL);
 420                                         ops->unresolved_rules++;
 421                                 }
 422                         }
 423                 }
 424
 425                 synchronize_rcu();
 426                 notify_rule_change(net, RTM_DELRULE, rule, ops, nlh,
 427                                    NETLINK_CB(skb).pid);
 428                 fib_rule_put(rule);
 429                 flush_route_cache(ops);
 430                 rules_ops_put(ops);
 431                 return 0;
 432         }
 433
 434         err = -ENOENT;
 435 errout:
 436         rules_ops_put(ops);
 437         return err;
 438 }
 439
 440 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
 441                                          struct fib_rule *rule)
 442 {
 443         size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
 444                          + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
 445                          + nla_total_size(4) /* FRA_PRIORITY */
 446                          + nla_total_size(4) /* FRA_TABLE */
 447                          + nla_total_size(4) /* FRA_FWMARK */
 448                          + nla_total_size(4); /* FRA_FWMASK */
 449
 450         if (ops->nlmsg_payload)
 451                 payload += ops->nlmsg_payload(rule);
 452
 453         return payload;
 454 }
 455
 456 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
 457                             u32 pid, u32 seq, int type, int flags,
 458                             struct fib_rules_ops *ops)
 459 {
 460         struct nlmsghdr *nlh;
 461         struct fib_rule_hdr *frh;
 462
 463         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
 464         if (nlh == NULL)
 465                 return -EMSGSIZE;
 466
 467         frh = nlmsg_data(nlh);
 468         frh->table = rule->table;
 469         NLA_PUT_U32(skb, FRA_TABLE, rule->table);
 470         frh->res1 = 0;
 471         frh->res2 = 0;
 472         frh->action = rule->action;
 473         frh->flags = rule->flags;
 474
 475         if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
 476                 frh->flags |= FIB_RULE_UNRESOLVED;
 477
 478         if (rule->ifname[0]) {
 479                 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
 480
 481                 if (rule->ifindex == -1)
 482                         frh->flags |= FIB_RULE_DEV_DETACHED;
 483         }
 484
 485         if (rule->pref)
 486                 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
 487
 488         if (rule->mark)
 489                 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
 490
 491         if (rule->mark_mask || rule->mark)
 492                 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
 493
 494         if (rule->target)
 495                 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
 496
 497         if (ops->fill(rule, skb, nlh, frh) < 0)
 498                 goto nla_put_failure;
 499
 500         return nlmsg_end(skb, nlh);
 501
 502 nla_put_failure:
 503         nlmsg_cancel(skb, nlh);
 504         return -EMSGSIZE;
 505 }
 506
 507 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
 508                       struct fib_rules_ops *ops)
 509 {
 510         int idx = 0;
 511         struct fib_rule *rule;
 512
 513         list_for_each_entry(rule, &ops->rules_list, list) {
 514                 if (idx < cb->args[1])
 515                         goto skip;
 516
 517                 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
 518                                      cb->nlh->nlmsg_seq, RTM_NEWRULE,
 519                                      NLM_F_MULTI, ops) < 0)
 520                         break;
 521 skip:
 522                 idx++;
 523         }
 524         cb->args[1] = idx;
 525         rules_ops_put(ops);
 526
 527         return skb->len;
 528 }
 529
 530 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
 531 {
 532         struct net *net = skb->sk->sk_net;
 533         struct fib_rules_ops *ops;
 534         int idx = 0, family;
 535
 536         family = rtnl_msg_family(cb->nlh);
 537         if (family != AF_UNSPEC) {
 538                 /* Protocol specific dump request */
 539                 ops = lookup_rules_ops(net, family);
 540                 if (ops == NULL)
 541                         return -EAFNOSUPPORT;
 542
 543                 return dump_rules(skb, cb, ops);
 544         }
 545
 546         rcu_read_lock();
 547         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
 548                 if (idx < cb->args[0] || !try_module_get(ops->owner))
 549                         goto skip;
 550
 551                 if (dump_rules(skb, cb, ops) < 0)
 552                         break;
 553
 554                 cb->args[1] = 0;
 555         skip:
 556                 idx++;
 557         }
 558         rcu_read_unlock();
 559         cb->args[0] = idx;
 560
 561         return skb->len;
 562 }
 563
 564 static void notify_rule_change(struct net *net, int event, struct fib_rule *rule,
 565                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
 566                                u32 pid)
 567 {
 568         struct sk_buff *skb;
 569         int err = -ENOBUFS;
 570
 571         skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
 572         if (skb == NULL)
 573                 goto errout;
 574
 575         err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
 576         if (err < 0) {
 577                 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
 578                 WARN_ON(err == -EMSGSIZE);
 579                 kfree_skb(skb);
 580                 goto errout;
 581         }
 582         err = rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
 583 errout:
 584         if (err < 0)
 585                 rtnl_set_sk_err(net, ops->nlgroup, err);
 586 }
 587
 588 static void attach_rules(struct list_head *rules, struct net_device *dev)
 589 {
 590         struct fib_rule *rule;
 591
 592         list_for_each_entry(rule, rules, list) {
 593                 if (rule->ifindex == -1 &&
 594                     strcmp(dev->name, rule->ifname) == 0)
 595                         rule->ifindex = dev->ifindex;
 596         }
 597 }
 598
 599 static void detach_rules(struct list_head *rules, struct net_device *dev)
 600 {
 601         struct fib_rule *rule;
 602
 603         list_for_each_entry(rule, rules, list)
 604                 if (rule->ifindex == dev->ifindex)
 605                         rule->ifindex = -1;
 606 }
 607
 608
 609 static int fib_rules_event(struct notifier_block *this, unsigned long event,
 610                             void *ptr)
 611 {
 612         struct net_device *dev = ptr;
 613         struct net *net = dev->nd_net;
 614         struct fib_rules_ops *ops;
 615
 616         ASSERT_RTNL();
 617         rcu_read_lock();
 618
 619         switch (event) {
 620         case NETDEV_REGISTER:
 621                 list_for_each_entry(ops, &net->rules_ops, list)
 622                         attach_rules(&ops->rules_list, dev);
 623                 break;
 624
 625         case NETDEV_UNREGISTER:
 626                 list_for_each_entry(ops, &net->rules_ops, list)
 627                         detach_rules(&ops->rules_list, dev);
 628                 break;
 629         }
 630
 631         rcu_read_unlock();
 632
 633         return NOTIFY_DONE;
 634 }
 635
 636 static struct notifier_block fib_rules_notifier = {
 637         .notifier_call = fib_rules_event,
 638 };
 639
 640 static int fib_rules_net_init(struct net *net)
 641 {
 642         INIT_LIST_HEAD(&net->rules_ops);
 643         spin_lock_init(&net->rules_mod_lock);
 644         return 0;
 645 }
 646
 647 static struct pernet_operations fib_rules_net_ops = {
 648         .init = fib_rules_net_init,
 649 };
 650
 651 static int __init fib_rules_init(void)
 652 {
 653         int err;
 654         rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
 655         rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
 656         rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
 657
 658         err = register_netdevice_notifier(&fib_rules_notifier);
 659         if (err < 0)
 660                 goto fail;
 661
 662         err = register_pernet_subsys(&fib_rules_net_ops);
 663         if (err < 0)
 664                 goto fail_unregister;
 665         return 0;
 666
 667 fail_unregister:
 668         unregister_netdevice_notifier(&fib_rules_notifier);
 669 fail:
 670         rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
 671         rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
 672         rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
 673         return err;
 674 }
 675
 676 subsys_initcall(fib_rules_init);