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 <[email protected]>
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/kernel.h>
  13 #include <linux/slab.h>
  14 #include <linux/list.h>
  15 #include <linux/module.h>
  16 #include <net/net_namespace.h>
  17 #include <net/sock.h>
  18 #include <net/fib_rules.h>
  19 #include <net/ip_tunnels.h>
  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         r->fr_net = ops->fro_net;
  36
  37         r->suppress_prefixlen = -1;
  38         r->suppress_ifgroup = -1;
  39
  40         /* The lock is not required here, the list in unreacheable
  41          * at the moment this function is called */
  42         list_add_tail(&r->list, &ops->rules_list);
  43         return 0;
  44 }
  45 EXPORT_SYMBOL(fib_default_rule_add);
  46
  47 u32 fib_default_rule_pref(struct fib_rules_ops *ops)
  48 {
  49         struct list_head *pos;
  50         struct fib_rule *rule;
  51
  52         if (!list_empty(&ops->rules_list)) {
  53                 pos = ops->rules_list.next;
  54                 if (pos->next != &ops->rules_list) {
  55                         rule = list_entry(pos->next, struct fib_rule, list);
  56                         if (rule->pref)
  57                                 return rule->pref - 1;
  58                 }
  59         }
  60
  61         return 0;
  62 }
  63 EXPORT_SYMBOL(fib_default_rule_pref);
  64
  65 static void notify_rule_change(int event, struct fib_rule *rule,
  66                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
  67                                u32 pid);
  68
  69 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
  70 {
  71         struct fib_rules_ops *ops;
  72
  73         rcu_read_lock();
  74         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
  75                 if (ops->family == family) {
  76                         if (!try_module_get(ops->owner))
  77                                 ops = NULL;
  78                         rcu_read_unlock();
  79                         return ops;
  80                 }
  81         }
  82         rcu_read_unlock();
  83
  84         return NULL;
  85 }
  86
  87 static void rules_ops_put(struct fib_rules_ops *ops)
  88 {
  89         if (ops)
  90                 module_put(ops->owner);
  91 }
  92
  93 static void flush_route_cache(struct fib_rules_ops *ops)
  94 {
  95         if (ops->flush_cache)
  96                 ops->flush_cache(ops);
  97 }
  98
  99 static int __fib_rules_register(struct fib_rules_ops *ops)
 100 {
 101         int err = -EEXIST;
 102         struct fib_rules_ops *o;
 103         struct net *net;
 104
 105         net = ops->fro_net;
 106
 107         if (ops->rule_size < sizeof(struct fib_rule))
 108                 return -EINVAL;
 109
 110         if (ops->match == NULL || ops->configure == NULL ||
 111             ops->compare == NULL || ops->fill == NULL ||
 112             ops->action == NULL)
 113                 return -EINVAL;
 114
 115         spin_lock(&net->rules_mod_lock);
 116         list_for_each_entry(o, &net->rules_ops, list)
 117                 if (ops->family == o->family)
 118                         goto errout;
 119
 120         list_add_tail_rcu(&ops->list, &net->rules_ops);
 121         err = 0;
 122 errout:
 123         spin_unlock(&net->rules_mod_lock);
 124
 125         return err;
 126 }
 127
 128 struct fib_rules_ops *
 129 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
 130 {
 131         struct fib_rules_ops *ops;
 132         int err;
 133
 134         ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
 135         if (ops == NULL)
 136                 return ERR_PTR(-ENOMEM);
 137
 138         INIT_LIST_HEAD(&ops->rules_list);
 139         ops->fro_net = net;
 140
 141         err = __fib_rules_register(ops);
 142         if (err) {
 143                 kfree(ops);
 144                 ops = ERR_PTR(err);
 145         }
 146
 147         return ops;
 148 }
 149 EXPORT_SYMBOL_GPL(fib_rules_register);
 150
 151 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
 152 {
 153         struct fib_rule *rule, *tmp;
 154
 155         list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
 156                 list_del_rcu(&rule->list);
 157                 if (ops->delete)
 158                         ops->delete(rule);
 159                 fib_rule_put(rule);
 160         }
 161 }
 162
 163 void fib_rules_unregister(struct fib_rules_ops *ops)
 164 {
 165         struct net *net = ops->fro_net;
 166
 167         spin_lock(&net->rules_mod_lock);
 168         list_del_rcu(&ops->list);
 169         spin_unlock(&net->rules_mod_lock);
 170
 171         fib_rules_cleanup_ops(ops);
 172         kfree_rcu(ops, rcu);
 173 }
 174 EXPORT_SYMBOL_GPL(fib_rules_unregister);
 175
 176 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
 177                           struct flowi *fl, int flags)
 178 {
 179         int ret = 0;
 180
 181         if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
 182                 goto out;
 183
 184         if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
 185                 goto out;
 186
 187         if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
 188                 goto out;
 189
 190         if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
 191                 goto out;
 192
 193         ret = ops->match(rule, fl, flags);
 194 out:
 195         return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
 196 }
 197
 198 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
 199                      int flags, struct fib_lookup_arg *arg)
 200 {
 201         struct fib_rule *rule;
 202         int err;
 203
 204         rcu_read_lock();
 205
 206         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 207 jumped:
 208                 if (!fib_rule_match(rule, ops, fl, flags))
 209                         continue;
 210
 211                 if (rule->action == FR_ACT_GOTO) {
 212                         struct fib_rule *target;
 213
 214                         target = rcu_dereference(rule->ctarget);
 215                         if (target == NULL) {
 216                                 continue;
 217                         } else {
 218                                 rule = target;
 219                                 goto jumped;
 220                         }
 221                 } else if (rule->action == FR_ACT_NOP)
 222                         continue;
 223                 else
 224                         err = ops->action(rule, fl, flags, arg);
 225
 226                 if (!err && ops->suppress && ops->suppress(rule, arg))
 227                         continue;
 228
 229                 if (err != -EAGAIN) {
 230                         if ((arg->flags & FIB_LOOKUP_NOREF) ||
 231                             likely(atomic_inc_not_zero(&rule->refcnt))) {
 232                                 arg->rule = rule;
 233                                 goto out;
 234                         }
 235                         break;
 236                 }
 237         }
 238
 239         err = -ESRCH;
 240 out:
 241         rcu_read_unlock();
 242
 243         return err;
 244 }
 245 EXPORT_SYMBOL_GPL(fib_rules_lookup);
 246
 247 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
 248                             struct fib_rules_ops *ops)
 249 {
 250         int err = -EINVAL;
 251
 252         if (frh->src_len)
 253                 if (tb[FRA_SRC] == NULL ||
 254                     frh->src_len > (ops->addr_size * 8) ||
 255                     nla_len(tb[FRA_SRC]) != ops->addr_size)
 256                         goto errout;
 257
 258         if (frh->dst_len)
 259                 if (tb[FRA_DST] == NULL ||
 260                     frh->dst_len > (ops->addr_size * 8) ||
 261                     nla_len(tb[FRA_DST]) != ops->addr_size)
 262                         goto errout;
 263
 264         err = 0;
 265 errout:
 266         return err;
 267 }
 268
 269 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh)
 270 {
 271         struct net *net = sock_net(skb->sk);
 272         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 273         struct fib_rules_ops *ops = NULL;
 274         struct fib_rule *rule, *r, *last = NULL;
 275         struct nlattr *tb[FRA_MAX+1];
 276         int err = -EINVAL, unresolved = 0;
 277
 278         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 279                 goto errout;
 280
 281         ops = lookup_rules_ops(net, frh->family);
 282         if (ops == NULL) {
 283                 err = -EAFNOSUPPORT;
 284                 goto errout;
 285         }
 286
 287         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 288         if (err < 0)
 289                 goto errout;
 290
 291         err = validate_rulemsg(frh, tb, ops);
 292         if (err < 0)
 293                 goto errout;
 294
 295         rule = kzalloc(ops->rule_size, GFP_KERNEL);
 296         if (rule == NULL) {
 297                 err = -ENOMEM;
 298                 goto errout;
 299         }
 300         rule->fr_net = net;
 301
 302         if (tb[FRA_PRIORITY])
 303                 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
 304
 305         if (tb[FRA_IIFNAME]) {
 306                 struct net_device *dev;
 307
 308                 rule->iifindex = -1;
 309                 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
 310                 dev = __dev_get_by_name(net, rule->iifname);
 311                 if (dev)
 312                         rule->iifindex = dev->ifindex;
 313         }
 314
 315         if (tb[FRA_OIFNAME]) {
 316                 struct net_device *dev;
 317
 318                 rule->oifindex = -1;
 319                 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
 320                 dev = __dev_get_by_name(net, rule->oifname);
 321                 if (dev)
 322                         rule->oifindex = dev->ifindex;
 323         }
 324
 325         if (tb[FRA_FWMARK]) {
 326                 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
 327                 if (rule->mark)
 328                         /* compatibility: if the mark value is non-zero all bits
 329                          * are compared unless a mask is explicitly specified.
 330                          */
 331                         rule->mark_mask = 0xFFFFFFFF;
 332         }
 333
 334         if (tb[FRA_FWMASK])
 335                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
 336
 337         if (tb[FRA_TUN_ID])
 338                 rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
 339
 340         rule->action = frh->action;
 341         rule->flags = frh->flags;
 342         rule->table = frh_get_table(frh, tb);
 343         if (tb[FRA_SUPPRESS_PREFIXLEN])
 344                 rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
 345         else
 346                 rule->suppress_prefixlen = -1;
 347
 348         if (tb[FRA_SUPPRESS_IFGROUP])
 349                 rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
 350         else
 351                 rule->suppress_ifgroup = -1;
 352
 353         if (!tb[FRA_PRIORITY] && ops->default_pref)
 354                 rule->pref = ops->default_pref(ops);
 355
 356         err = -EINVAL;
 357         if (tb[FRA_GOTO]) {
 358                 if (rule->action != FR_ACT_GOTO)
 359                         goto errout_free;
 360
 361                 rule->target = nla_get_u32(tb[FRA_GOTO]);
 362                 /* Backward jumps are prohibited to avoid endless loops */
 363                 if (rule->target <= rule->pref)
 364                         goto errout_free;
 365
 366                 list_for_each_entry(r, &ops->rules_list, list) {
 367                         if (r->pref == rule->target) {
 368                                 RCU_INIT_POINTER(rule->ctarget, r);
 369                                 break;
 370                         }
 371                 }
 372
 373                 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
 374                         unresolved = 1;
 375         } else if (rule->action == FR_ACT_GOTO)
 376                 goto errout_free;
 377
 378         err = ops->configure(rule, skb, frh, tb);
 379         if (err < 0)
 380                 goto errout_free;
 381
 382         list_for_each_entry(r, &ops->rules_list, list) {
 383                 if (r->pref > rule->pref)
 384                         break;
 385                 last = r;
 386         }
 387
 388         fib_rule_get(rule);
 389
 390         if (last)
 391                 list_add_rcu(&rule->list, &last->list);
 392         else
 393                 list_add_rcu(&rule->list, &ops->rules_list);
 394
 395         if (ops->unresolved_rules) {
 396                 /*
 397                  * There are unresolved goto rules in the list, check if
 398                  * any of them are pointing to this new rule.
 399                  */
 400                 list_for_each_entry(r, &ops->rules_list, list) {
 401                         if (r->action == FR_ACT_GOTO &&
 402                             r->target == rule->pref &&
 403                             rtnl_dereference(r->ctarget) == NULL) {
 404                                 rcu_assign_pointer(r->ctarget, rule);
 405                                 if (--ops->unresolved_rules == 0)
 406                                         break;
 407                         }
 408                 }
 409         }
 410
 411         if (rule->action == FR_ACT_GOTO)
 412                 ops->nr_goto_rules++;
 413
 414         if (unresolved)
 415                 ops->unresolved_rules++;
 416
 417         if (rule->tun_id)
 418                 ip_tunnel_need_metadata();
 419
 420         notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
 421         flush_route_cache(ops);
 422         rules_ops_put(ops);
 423         return 0;
 424
 425 errout_free:
 426         kfree(rule);
 427 errout:
 428         rules_ops_put(ops);
 429         return err;
 430 }
 431
 432 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh)
 433 {
 434         struct net *net = sock_net(skb->sk);
 435         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 436         struct fib_rules_ops *ops = NULL;
 437         struct fib_rule *rule, *tmp;
 438         struct nlattr *tb[FRA_MAX+1];
 439         int err = -EINVAL;
 440
 441         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 442                 goto errout;
 443
 444         ops = lookup_rules_ops(net, frh->family);
 445         if (ops == NULL) {
 446                 err = -EAFNOSUPPORT;
 447                 goto errout;
 448         }
 449
 450         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 451         if (err < 0)
 452                 goto errout;
 453
 454         err = validate_rulemsg(frh, tb, ops);
 455         if (err < 0)
 456                 goto errout;
 457
 458         list_for_each_entry(rule, &ops->rules_list, list) {
 459                 if (frh->action && (frh->action != rule->action))
 460                         continue;
 461
 462                 if (frh_get_table(frh, tb) &&
 463                     (frh_get_table(frh, tb) != rule->table))
 464                         continue;
 465
 466                 if (tb[FRA_PRIORITY] &&
 467                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
 468                         continue;
 469
 470                 if (tb[FRA_IIFNAME] &&
 471                     nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
 472                         continue;
 473
 474                 if (tb[FRA_OIFNAME] &&
 475                     nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
 476                         continue;
 477
 478                 if (tb[FRA_FWMARK] &&
 479                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
 480                         continue;
 481
 482                 if (tb[FRA_FWMASK] &&
 483                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
 484                         continue;
 485
 486                 if (tb[FRA_TUN_ID] &&
 487                     (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
 488                         continue;
 489
 490                 if (!ops->compare(rule, frh, tb))
 491                         continue;
 492
 493                 if (rule->flags & FIB_RULE_PERMANENT) {
 494                         err = -EPERM;
 495                         goto errout;
 496                 }
 497
 498                 if (ops->delete) {
 499                         err = ops->delete(rule);
 500                         if (err)
 501                                 goto errout;
 502                 }
 503
 504                 if (rule->tun_id)
 505                         ip_tunnel_unneed_metadata();
 506
 507                 list_del_rcu(&rule->list);
 508
 509                 if (rule->action == FR_ACT_GOTO) {
 510                         ops->nr_goto_rules--;
 511                         if (rtnl_dereference(rule->ctarget) == NULL)
 512                                 ops->unresolved_rules--;
 513                 }
 514
 515                 /*
 516                  * Check if this rule is a target to any of them. If so,
 517                  * disable them. As this operation is eventually very
 518                  * expensive, it is only performed if goto rules have
 519                  * actually been added.
 520                  */
 521                 if (ops->nr_goto_rules > 0) {
 522                         list_for_each_entry(tmp, &ops->rules_list, list) {
 523                                 if (rtnl_dereference(tmp->ctarget) == rule) {
 524                                         RCU_INIT_POINTER(tmp->ctarget, NULL);
 525                                         ops->unresolved_rules++;
 526                                 }
 527                         }
 528                 }
 529
 530                 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
 531                                    NETLINK_CB(skb).portid);
 532                 fib_rule_put(rule);
 533                 flush_route_cache(ops);
 534                 rules_ops_put(ops);
 535                 return 0;
 536         }
 537
 538         err = -ENOENT;
 539 errout:
 540         rules_ops_put(ops);
 541         return err;
 542 }
 543
 544 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
 545                                          struct fib_rule *rule)
 546 {
 547         size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
 548                          + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
 549                          + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
 550                          + nla_total_size(4) /* FRA_PRIORITY */
 551                          + nla_total_size(4) /* FRA_TABLE */
 552                          + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
 553                          + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
 554                          + nla_total_size(4) /* FRA_FWMARK */
 555                          + nla_total_size(4) /* FRA_FWMASK */
 556                          + nla_total_size(8); /* FRA_TUN_ID */
 557
 558         if (ops->nlmsg_payload)
 559                 payload += ops->nlmsg_payload(rule);
 560
 561         return payload;
 562 }
 563
 564 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
 565                             u32 pid, u32 seq, int type, int flags,
 566                             struct fib_rules_ops *ops)
 567 {
 568         struct nlmsghdr *nlh;
 569         struct fib_rule_hdr *frh;
 570
 571         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
 572         if (nlh == NULL)
 573                 return -EMSGSIZE;
 574
 575         frh = nlmsg_data(nlh);
 576         frh->family = ops->family;
 577         frh->table = rule->table;
 578         if (nla_put_u32(skb, FRA_TABLE, rule->table))
 579                 goto nla_put_failure;
 580         if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
 581                 goto nla_put_failure;
 582         frh->res1 = 0;
 583         frh->res2 = 0;
 584         frh->action = rule->action;
 585         frh->flags = rule->flags;
 586
 587         if (rule->action == FR_ACT_GOTO &&
 588             rcu_access_pointer(rule->ctarget) == NULL)
 589                 frh->flags |= FIB_RULE_UNRESOLVED;
 590
 591         if (rule->iifname[0]) {
 592                 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
 593                         goto nla_put_failure;
 594                 if (rule->iifindex == -1)
 595                         frh->flags |= FIB_RULE_IIF_DETACHED;
 596         }
 597
 598         if (rule->oifname[0]) {
 599                 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
 600                         goto nla_put_failure;
 601                 if (rule->oifindex == -1)
 602                         frh->flags |= FIB_RULE_OIF_DETACHED;
 603         }
 604
 605         if ((rule->pref &&
 606              nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
 607             (rule->mark &&
 608              nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
 609             ((rule->mark_mask || rule->mark) &&
 610              nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
 611             (rule->target &&
 612              nla_put_u32(skb, FRA_GOTO, rule->target)) ||
 613             (rule->tun_id &&
 614              nla_put_be64(skb, FRA_TUN_ID, rule->tun_id)))
 615                 goto nla_put_failure;
 616
 617         if (rule->suppress_ifgroup != -1) {
 618                 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
 619                         goto nla_put_failure;
 620         }
 621
 622         if (ops->fill(rule, skb, frh) < 0)
 623                 goto nla_put_failure;
 624
 625         nlmsg_end(skb, nlh);
 626         return 0;
 627
 628 nla_put_failure:
 629         nlmsg_cancel(skb, nlh);
 630         return -EMSGSIZE;
 631 }
 632
 633 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
 634                       struct fib_rules_ops *ops)
 635 {
 636         int idx = 0;
 637         struct fib_rule *rule;
 638
 639         rcu_read_lock();
 640         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 641                 if (idx < cb->args[1])
 642                         goto skip;
 643
 644                 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
 645                                      cb->nlh->nlmsg_seq, RTM_NEWRULE,
 646                                      NLM_F_MULTI, ops) < 0)
 647                         break;
 648 skip:
 649                 idx++;
 650         }
 651         rcu_read_unlock();
 652         cb->args[1] = idx;
 653         rules_ops_put(ops);
 654
 655         return skb->len;
 656 }
 657
 658 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
 659 {
 660         struct net *net = sock_net(skb->sk);
 661         struct fib_rules_ops *ops;
 662         int idx = 0, family;
 663
 664         family = rtnl_msg_family(cb->nlh);
 665         if (family != AF_UNSPEC) {
 666                 /* Protocol specific dump request */
 667                 ops = lookup_rules_ops(net, family);
 668                 if (ops == NULL)
 669                         return -EAFNOSUPPORT;
 670
 671                 return dump_rules(skb, cb, ops);
 672         }
 673
 674         rcu_read_lock();
 675         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
 676                 if (idx < cb->args[0] || !try_module_get(ops->owner))
 677                         goto skip;
 678
 679                 if (dump_rules(skb, cb, ops) < 0)
 680                         break;
 681
 682                 cb->args[1] = 0;
 683 skip:
 684                 idx++;
 685         }
 686         rcu_read_unlock();
 687         cb->args[0] = idx;
 688
 689         return skb->len;
 690 }
 691
 692 static void notify_rule_change(int event, struct fib_rule *rule,
 693                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
 694                                u32 pid)
 695 {
 696         struct net *net;
 697         struct sk_buff *skb;
 698         int err = -ENOBUFS;
 699
 700         net = ops->fro_net;
 701         skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
 702         if (skb == NULL)
 703                 goto errout;
 704
 705         err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
 706         if (err < 0) {
 707                 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
 708                 WARN_ON(err == -EMSGSIZE);
 709                 kfree_skb(skb);
 710                 goto errout;
 711         }
 712
 713         rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
 714         return;
 715 errout:
 716         if (err < 0)
 717                 rtnl_set_sk_err(net, ops->nlgroup, err);
 718 }
 719
 720 static void attach_rules(struct list_head *rules, struct net_device *dev)
 721 {
 722         struct fib_rule *rule;
 723
 724         list_for_each_entry(rule, rules, list) {
 725                 if (rule->iifindex == -1 &&
 726                     strcmp(dev->name, rule->iifname) == 0)
 727                         rule->iifindex = dev->ifindex;
 728                 if (rule->oifindex == -1 &&
 729                     strcmp(dev->name, rule->oifname) == 0)
 730                         rule->oifindex = dev->ifindex;
 731         }
 732 }
 733
 734 static void detach_rules(struct list_head *rules, struct net_device *dev)
 735 {
 736         struct fib_rule *rule;
 737
 738         list_for_each_entry(rule, rules, list) {
 739                 if (rule->iifindex == dev->ifindex)
 740                         rule->iifindex = -1;
 741                 if (rule->oifindex == dev->ifindex)
 742                         rule->oifindex = -1;
 743         }
 744 }
 745
 746
 747 static int fib_rules_event(struct notifier_block *this, unsigned long event,
 748                            void *ptr)
 749 {
 750         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 751         struct net *net = dev_net(dev);
 752         struct fib_rules_ops *ops;
 753
 754         ASSERT_RTNL();
 755
 756         switch (event) {
 757         case NETDEV_REGISTER:
 758                 list_for_each_entry(ops, &net->rules_ops, list)
 759                         attach_rules(&ops->rules_list, dev);
 760                 break;
 761
 762         case NETDEV_CHANGENAME:
 763                 list_for_each_entry(ops, &net->rules_ops, list) {
 764                         detach_rules(&ops->rules_list, dev);
 765                         attach_rules(&ops->rules_list, dev);
 766                 }
 767                 break;
 768
 769         case NETDEV_UNREGISTER:
 770                 list_for_each_entry(ops, &net->rules_ops, list)
 771                         detach_rules(&ops->rules_list, dev);
 772                 break;
 773         }
 774
 775         return NOTIFY_DONE;
 776 }
 777
 778 static struct notifier_block fib_rules_notifier = {
 779         .notifier_call = fib_rules_event,
 780 };
 781
 782 static int __net_init fib_rules_net_init(struct net *net)
 783 {
 784         INIT_LIST_HEAD(&net->rules_ops);
 785         spin_lock_init(&net->rules_mod_lock);
 786         return 0;
 787 }
 788
 789 static struct pernet_operations fib_rules_net_ops = {
 790         .init = fib_rules_net_init,
 791 };
 792
 793 static int __init fib_rules_init(void)
 794 {
 795         int err;
 796         rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
 797         rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
 798         rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
 799
 800         err = register_pernet_subsys(&fib_rules_net_ops);
 801         if (err < 0)
 802                 goto fail;
 803
 804         err = register_netdevice_notifier(&fib_rules_notifier);
 805         if (err < 0)
 806                 goto fail_unregister;
 807
 808         return 0;
 809
 810 fail_unregister:
 811         unregister_pernet_subsys(&fib_rules_net_ops);
 812 fail:
 813         rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
 814         rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
 815         rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
 816         return err;
 817 }
 818
 819 subsys_initcall(fib_rules_init);