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/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 static 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
  64 static void notify_rule_change(int event, struct fib_rule *rule,
  65                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
  66                                u32 pid);
  67
  68 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
  69 {
  70         struct fib_rules_ops *ops;
  71
  72         rcu_read_lock();
  73         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
  74                 if (ops->family == family) {
  75                         if (!try_module_get(ops->owner))
  76                                 ops = NULL;
  77                         rcu_read_unlock();
  78                         return ops;
  79                 }
  80         }
  81         rcu_read_unlock();
  82
  83         return NULL;
  84 }
  85
  86 static void rules_ops_put(struct fib_rules_ops *ops)
  87 {
  88         if (ops)
  89                 module_put(ops->owner);
  90 }
  91
  92 static void flush_route_cache(struct fib_rules_ops *ops)
  93 {
  94         if (ops->flush_cache)
  95                 ops->flush_cache(ops);
  96 }
  97
  98 static int __fib_rules_register(struct fib_rules_ops *ops)
  99 {
 100         int err = -EEXIST;
 101         struct fib_rules_ops *o;
 102         struct net *net;
 103
 104         net = ops->fro_net;
 105
 106         if (ops->rule_size < sizeof(struct fib_rule))
 107                 return -EINVAL;
 108
 109         if (ops->match == NULL || ops->configure == NULL ||
 110             ops->compare == NULL || ops->fill == NULL ||
 111             ops->action == NULL)
 112                 return -EINVAL;
 113
 114         spin_lock(&net->rules_mod_lock);
 115         list_for_each_entry(o, &net->rules_ops, list)
 116                 if (ops->family == o->family)
 117                         goto errout;
 118
 119         list_add_tail_rcu(&ops->list, &net->rules_ops);
 120         err = 0;
 121 errout:
 122         spin_unlock(&net->rules_mod_lock);
 123
 124         return err;
 125 }
 126
 127 struct fib_rules_ops *
 128 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
 129 {
 130         struct fib_rules_ops *ops;
 131         int err;
 132
 133         ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
 134         if (ops == NULL)
 135                 return ERR_PTR(-ENOMEM);
 136
 137         INIT_LIST_HEAD(&ops->rules_list);
 138         ops->fro_net = net;
 139
 140         err = __fib_rules_register(ops);
 141         if (err) {
 142                 kfree(ops);
 143                 ops = ERR_PTR(err);
 144         }
 145
 146         return ops;
 147 }
 148 EXPORT_SYMBOL_GPL(fib_rules_register);
 149
 150 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
 151 {
 152         struct fib_rule *rule, *tmp;
 153
 154         list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
 155                 list_del_rcu(&rule->list);
 156                 if (ops->delete)
 157                         ops->delete(rule);
 158                 fib_rule_put(rule);
 159         }
 160 }
 161
 162 void fib_rules_unregister(struct fib_rules_ops *ops)
 163 {
 164         struct net *net = ops->fro_net;
 165
 166         spin_lock(&net->rules_mod_lock);
 167         list_del_rcu(&ops->list);
 168         spin_unlock(&net->rules_mod_lock);
 169
 170         fib_rules_cleanup_ops(ops);
 171         kfree_rcu(ops, rcu);
 172 }
 173 EXPORT_SYMBOL_GPL(fib_rules_unregister);
 174
 175 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
 176                           struct flowi *fl, int flags)
 177 {
 178         int ret = 0;
 179
 180         if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
 181                 goto out;
 182
 183         if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
 184                 goto out;
 185
 186         if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
 187                 goto out;
 188
 189         if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
 190                 goto out;
 191
 192         ret = ops->match(rule, fl, flags);
 193 out:
 194         return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
 195 }
 196
 197 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
 198                      int flags, struct fib_lookup_arg *arg)
 199 {
 200         struct fib_rule *rule;
 201         int err;
 202
 203         rcu_read_lock();
 204
 205         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 206 jumped:
 207                 if (!fib_rule_match(rule, ops, fl, flags))
 208                         continue;
 209
 210                 if (rule->action == FR_ACT_GOTO) {
 211                         struct fib_rule *target;
 212
 213                         target = rcu_dereference(rule->ctarget);
 214                         if (target == NULL) {
 215                                 continue;
 216                         } else {
 217                                 rule = target;
 218                                 goto jumped;
 219                         }
 220                 } else if (rule->action == FR_ACT_NOP)
 221                         continue;
 222                 else
 223                         err = ops->action(rule, fl, flags, arg);
 224
 225                 if (!err && ops->suppress && ops->suppress(rule, arg))
 226                         continue;
 227
 228                 if (err != -EAGAIN) {
 229                         if ((arg->flags & FIB_LOOKUP_NOREF) ||
 230                             likely(atomic_inc_not_zero(&rule->refcnt))) {
 231                                 arg->rule = rule;
 232                                 goto out;
 233                         }
 234                         break;
 235                 }
 236         }
 237
 238         err = -ESRCH;
 239 out:
 240         rcu_read_unlock();
 241
 242         return err;
 243 }
 244 EXPORT_SYMBOL_GPL(fib_rules_lookup);
 245
 246 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
 247                             struct fib_rules_ops *ops)
 248 {
 249         int err = -EINVAL;
 250
 251         if (frh->src_len)
 252                 if (tb[FRA_SRC] == NULL ||
 253                     frh->src_len > (ops->addr_size * 8) ||
 254                     nla_len(tb[FRA_SRC]) != ops->addr_size)
 255                         goto errout;
 256
 257         if (frh->dst_len)
 258                 if (tb[FRA_DST] == NULL ||
 259                     frh->dst_len > (ops->addr_size * 8) ||
 260                     nla_len(tb[FRA_DST]) != ops->addr_size)
 261                         goto errout;
 262
 263         err = 0;
 264 errout:
 265         return err;
 266 }
 267
 268 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh)
 269 {
 270         struct net *net = sock_net(skb->sk);
 271         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 272         struct fib_rules_ops *ops = NULL;
 273         struct fib_rule *rule, *r, *last = NULL;
 274         struct nlattr *tb[FRA_MAX+1];
 275         int err = -EINVAL, unresolved = 0;
 276
 277         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 278                 goto errout;
 279
 280         ops = lookup_rules_ops(net, frh->family);
 281         if (ops == NULL) {
 282                 err = -EAFNOSUPPORT;
 283                 goto errout;
 284         }
 285
 286         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 287         if (err < 0)
 288                 goto errout;
 289
 290         err = validate_rulemsg(frh, tb, ops);
 291         if (err < 0)
 292                 goto errout;
 293
 294         rule = kzalloc(ops->rule_size, GFP_KERNEL);
 295         if (rule == NULL) {
 296                 err = -ENOMEM;
 297                 goto errout;
 298         }
 299         rule->fr_net = net;
 300
 301         rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
 302                                       : fib_default_rule_pref(ops);
 303
 304         if (tb[FRA_IIFNAME]) {
 305                 struct net_device *dev;
 306
 307                 rule->iifindex = -1;
 308                 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
 309                 dev = __dev_get_by_name(net, rule->iifname);
 310                 if (dev)
 311                         rule->iifindex = dev->ifindex;
 312         }
 313
 314         if (tb[FRA_OIFNAME]) {
 315                 struct net_device *dev;
 316
 317                 rule->oifindex = -1;
 318                 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
 319                 dev = __dev_get_by_name(net, rule->oifname);
 320                 if (dev)
 321                         rule->oifindex = dev->ifindex;
 322         }
 323
 324         if (tb[FRA_FWMARK]) {
 325                 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
 326                 if (rule->mark)
 327                         /* compatibility: if the mark value is non-zero all bits
 328                          * are compared unless a mask is explicitly specified.
 329                          */
 330                         rule->mark_mask = 0xFFFFFFFF;
 331         }
 332
 333         if (tb[FRA_FWMASK])
 334                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
 335
 336         if (tb[FRA_TUN_ID])
 337                 rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
 338
 339         rule->action = frh->action;
 340         rule->flags = frh->flags;
 341         rule->table = frh_get_table(frh, tb);
 342         if (tb[FRA_SUPPRESS_PREFIXLEN])
 343                 rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
 344         else
 345                 rule->suppress_prefixlen = -1;
 346
 347         if (tb[FRA_SUPPRESS_IFGROUP])
 348                 rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
 349         else
 350                 rule->suppress_ifgroup = -1;
 351
 352         err = -EINVAL;
 353         if (tb[FRA_GOTO]) {
 354                 if (rule->action != FR_ACT_GOTO)
 355                         goto errout_free;
 356
 357                 rule->target = nla_get_u32(tb[FRA_GOTO]);
 358                 /* Backward jumps are prohibited to avoid endless loops */
 359                 if (rule->target <= rule->pref)
 360                         goto errout_free;
 361
 362                 list_for_each_entry(r, &ops->rules_list, list) {
 363                         if (r->pref == rule->target) {
 364                                 RCU_INIT_POINTER(rule->ctarget, r);
 365                                 break;
 366                         }
 367                 }
 368
 369                 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
 370                         unresolved = 1;
 371         } else if (rule->action == FR_ACT_GOTO)
 372                 goto errout_free;
 373
 374         err = ops->configure(rule, skb, frh, tb);
 375         if (err < 0)
 376                 goto errout_free;
 377
 378         list_for_each_entry(r, &ops->rules_list, list) {
 379                 if (r->pref > rule->pref)
 380                         break;
 381                 last = r;
 382         }
 383
 384         fib_rule_get(rule);
 385
 386         if (last)
 387                 list_add_rcu(&rule->list, &last->list);
 388         else
 389                 list_add_rcu(&rule->list, &ops->rules_list);
 390
 391         if (ops->unresolved_rules) {
 392                 /*
 393                  * There are unresolved goto rules in the list, check if
 394                  * any of them are pointing to this new rule.
 395                  */
 396                 list_for_each_entry(r, &ops->rules_list, list) {
 397                         if (r->action == FR_ACT_GOTO &&
 398                             r->target == rule->pref &&
 399                             rtnl_dereference(r->ctarget) == NULL) {
 400                                 rcu_assign_pointer(r->ctarget, rule);
 401                                 if (--ops->unresolved_rules == 0)
 402                                         break;
 403                         }
 404                 }
 405         }
 406
 407         if (rule->action == FR_ACT_GOTO)
 408                 ops->nr_goto_rules++;
 409
 410         if (unresolved)
 411                 ops->unresolved_rules++;
 412
 413         if (rule->tun_id)
 414                 ip_tunnel_need_metadata();
 415
 416         notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
 417         flush_route_cache(ops);
 418         rules_ops_put(ops);
 419         return 0;
 420
 421 errout_free:
 422         kfree(rule);
 423 errout:
 424         rules_ops_put(ops);
 425         return err;
 426 }
 427
 428 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh)
 429 {
 430         struct net *net = sock_net(skb->sk);
 431         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 432         struct fib_rules_ops *ops = NULL;
 433         struct fib_rule *rule, *tmp;
 434         struct nlattr *tb[FRA_MAX+1];
 435         int err = -EINVAL;
 436
 437         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 438                 goto errout;
 439
 440         ops = lookup_rules_ops(net, frh->family);
 441         if (ops == NULL) {
 442                 err = -EAFNOSUPPORT;
 443                 goto errout;
 444         }
 445
 446         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 447         if (err < 0)
 448                 goto errout;
 449
 450         err = validate_rulemsg(frh, tb, ops);
 451         if (err < 0)
 452                 goto errout;
 453
 454         list_for_each_entry(rule, &ops->rules_list, list) {
 455                 if (frh->action && (frh->action != rule->action))
 456                         continue;
 457
 458                 if (frh_get_table(frh, tb) &&
 459                     (frh_get_table(frh, tb) != rule->table))
 460                         continue;
 461
 462                 if (tb[FRA_PRIORITY] &&
 463                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
 464                         continue;
 465
 466                 if (tb[FRA_IIFNAME] &&
 467                     nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
 468                         continue;
 469
 470                 if (tb[FRA_OIFNAME] &&
 471                     nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
 472                         continue;
 473
 474                 if (tb[FRA_FWMARK] &&
 475                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
 476                         continue;
 477
 478                 if (tb[FRA_FWMASK] &&
 479                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
 480                         continue;
 481
 482                 if (tb[FRA_TUN_ID] &&
 483                     (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
 484                         continue;
 485
 486                 if (!ops->compare(rule, frh, tb))
 487                         continue;
 488
 489                 if (rule->flags & FIB_RULE_PERMANENT) {
 490                         err = -EPERM;
 491                         goto errout;
 492                 }
 493
 494                 if (ops->delete) {
 495                         err = ops->delete(rule);
 496                         if (err)
 497                                 goto errout;
 498                 }
 499
 500                 if (rule->tun_id)
 501                         ip_tunnel_unneed_metadata();
 502
 503                 list_del_rcu(&rule->list);
 504
 505                 if (rule->action == FR_ACT_GOTO) {
 506                         ops->nr_goto_rules--;
 507                         if (rtnl_dereference(rule->ctarget) == NULL)
 508                                 ops->unresolved_rules--;
 509                 }
 510
 511                 /*
 512                  * Check if this rule is a target to any of them. If so,
 513                  * disable them. As this operation is eventually very
 514                  * expensive, it is only performed if goto rules have
 515                  * actually been added.
 516                  */
 517                 if (ops->nr_goto_rules > 0) {
 518                         list_for_each_entry(tmp, &ops->rules_list, list) {
 519                                 if (rtnl_dereference(tmp->ctarget) == rule) {
 520                                         RCU_INIT_POINTER(tmp->ctarget, NULL);
 521                                         ops->unresolved_rules++;
 522                                 }
 523                         }
 524                 }
 525
 526                 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
 527                                    NETLINK_CB(skb).portid);
 528                 fib_rule_put(rule);
 529                 flush_route_cache(ops);
 530                 rules_ops_put(ops);
 531                 return 0;
 532         }
 533
 534         err = -ENOENT;
 535 errout:
 536         rules_ops_put(ops);
 537         return err;
 538 }
 539
 540 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
 541                                          struct fib_rule *rule)
 542 {
 543         size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
 544                          + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
 545                          + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
 546                          + nla_total_size(4) /* FRA_PRIORITY */
 547                          + nla_total_size(4) /* FRA_TABLE */
 548                          + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
 549                          + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
 550                          + nla_total_size(4) /* FRA_FWMARK */
 551                          + nla_total_size(4) /* FRA_FWMASK */
 552                          + nla_total_size(8); /* FRA_TUN_ID */
 553
 554         if (ops->nlmsg_payload)
 555                 payload += ops->nlmsg_payload(rule);
 556
 557         return payload;
 558 }
 559
 560 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
 561                             u32 pid, u32 seq, int type, int flags,
 562                             struct fib_rules_ops *ops)
 563 {
 564         struct nlmsghdr *nlh;
 565         struct fib_rule_hdr *frh;
 566
 567         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
 568         if (nlh == NULL)
 569                 return -EMSGSIZE;
 570
 571         frh = nlmsg_data(nlh);
 572         frh->family = ops->family;
 573         frh->table = rule->table;
 574         if (nla_put_u32(skb, FRA_TABLE, rule->table))
 575                 goto nla_put_failure;
 576         if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
 577                 goto nla_put_failure;
 578         frh->res1 = 0;
 579         frh->res2 = 0;
 580         frh->action = rule->action;
 581         frh->flags = rule->flags;
 582
 583         if (rule->action == FR_ACT_GOTO &&
 584             rcu_access_pointer(rule->ctarget) == NULL)
 585                 frh->flags |= FIB_RULE_UNRESOLVED;
 586
 587         if (rule->iifname[0]) {
 588                 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
 589                         goto nla_put_failure;
 590                 if (rule->iifindex == -1)
 591                         frh->flags |= FIB_RULE_IIF_DETACHED;
 592         }
 593
 594         if (rule->oifname[0]) {
 595                 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
 596                         goto nla_put_failure;
 597                 if (rule->oifindex == -1)
 598                         frh->flags |= FIB_RULE_OIF_DETACHED;
 599         }
 600
 601         if ((rule->pref &&
 602              nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
 603             (rule->mark &&
 604              nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
 605             ((rule->mark_mask || rule->mark) &&
 606              nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
 607             (rule->target &&
 608              nla_put_u32(skb, FRA_GOTO, rule->target)) ||
 609             (rule->tun_id &&
 610              nla_put_be64(skb, FRA_TUN_ID, rule->tun_id)))
 611                 goto nla_put_failure;
 612
 613         if (rule->suppress_ifgroup != -1) {
 614                 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
 615                         goto nla_put_failure;
 616         }
 617
 618         if (ops->fill(rule, skb, frh) < 0)
 619                 goto nla_put_failure;
 620
 621         nlmsg_end(skb, nlh);
 622         return 0;
 623
 624 nla_put_failure:
 625         nlmsg_cancel(skb, nlh);
 626         return -EMSGSIZE;
 627 }
 628
 629 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
 630                       struct fib_rules_ops *ops)
 631 {
 632         int idx = 0;
 633         struct fib_rule *rule;
 634         int err = 0;
 635
 636         rcu_read_lock();
 637         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 638                 if (idx < cb->args[1])
 639                         goto skip;
 640
 641                 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
 642                                        cb->nlh->nlmsg_seq, RTM_NEWRULE,
 643                                        NLM_F_MULTI, ops);
 644                 if (err)
 645                         break;
 646 skip:
 647                 idx++;
 648         }
 649         rcu_read_unlock();
 650         cb->args[1] = idx;
 651         rules_ops_put(ops);
 652
 653         return err;
 654 }
 655
 656 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
 657 {
 658         struct net *net = sock_net(skb->sk);
 659         struct fib_rules_ops *ops;
 660         int idx = 0, family;
 661
 662         family = rtnl_msg_family(cb->nlh);
 663         if (family != AF_UNSPEC) {
 664                 /* Protocol specific dump request */
 665                 ops = lookup_rules_ops(net, family);
 666                 if (ops == NULL)
 667                         return -EAFNOSUPPORT;
 668
 669                 dump_rules(skb, cb, ops);
 670
 671                 return skb->len;
 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);