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 int fib_rules_unregister(struct net *net, struct fib_rules_ops *ops)
 116 {
 117         int err = 0;
 118         struct fib_rules_ops *o;
 119
 120         spin_lock(&net->rules_mod_lock);
 121         list_for_each_entry(o, &net->rules_ops, list) {
 122                 if (o == ops) {
 123                         list_del_rcu(&o->list);
 124                         fib_rules_cleanup_ops(ops);
 125                         goto out;
 126                 }
 127         }
 128
 129         err = -ENOENT;
 130 out:
 131         spin_unlock(&net->rules_mod_lock);
 132
 133         synchronize_rcu();
 134         if (!err)
 135                 release_net(net);
 136
 137         return err;
 138 }
 139
 140 EXPORT_SYMBOL_GPL(fib_rules_unregister);
 141
 142 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
 143                           struct flowi *fl, int flags)
 144 {
 145         int ret = 0;
 146
 147         if (rule->ifindex && (rule->ifindex != fl->iif))
 148                 goto out;
 149
 150         if ((rule->mark ^ fl->mark) & rule->mark_mask)
 151                 goto out;
 152
 153         ret = ops->match(rule, fl, flags);
 154 out:
 155         return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
 156 }
 157
 158 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
 159                      int flags, struct fib_lookup_arg *arg)
 160 {
 161         struct fib_rule *rule;
 162         int err;
 163
 164         rcu_read_lock();
 165
 166         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 167 jumped:
 168                 if (!fib_rule_match(rule, ops, fl, flags))
 169                         continue;
 170
 171                 if (rule->action == FR_ACT_GOTO) {
 172                         struct fib_rule *target;
 173
 174                         target = rcu_dereference(rule->ctarget);
 175                         if (target == NULL) {
 176                                 continue;
 177                         } else {
 178                                 rule = target;
 179                                 goto jumped;
 180                         }
 181                 } else if (rule->action == FR_ACT_NOP)
 182                         continue;
 183                 else
 184                         err = ops->action(rule, fl, flags, arg);
 185
 186                 if (err != -EAGAIN) {
 187                         fib_rule_get(rule);
 188                         arg->rule = rule;
 189                         goto out;
 190                 }
 191         }
 192
 193         err = -ESRCH;
 194 out:
 195         rcu_read_unlock();
 196
 197         return err;
 198 }
 199
 200 EXPORT_SYMBOL_GPL(fib_rules_lookup);
 201
 202 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
 203                             struct fib_rules_ops *ops)
 204 {
 205         int err = -EINVAL;
 206
 207         if (frh->src_len)
 208                 if (tb[FRA_SRC] == NULL ||
 209                     frh->src_len > (ops->addr_size * 8) ||
 210                     nla_len(tb[FRA_SRC]) != ops->addr_size)
 211                         goto errout;
 212
 213         if (frh->dst_len)
 214                 if (tb[FRA_DST] == NULL ||
 215                     frh->dst_len > (ops->addr_size * 8) ||
 216                     nla_len(tb[FRA_DST]) != ops->addr_size)
 217                         goto errout;
 218
 219         err = 0;
 220 errout:
 221         return err;
 222 }
 223
 224 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 225 {
 226         struct net *net = skb->sk->sk_net;
 227         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 228         struct fib_rules_ops *ops = NULL;
 229         struct fib_rule *rule, *r, *last = NULL;
 230         struct nlattr *tb[FRA_MAX+1];
 231         int err = -EINVAL, unresolved = 0;
 232
 233         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 234                 goto errout;
 235
 236         ops = lookup_rules_ops(net, frh->family);
 237         if (ops == NULL) {
 238                 err = EAFNOSUPPORT;
 239                 goto errout;
 240         }
 241
 242         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 243         if (err < 0)
 244                 goto errout;
 245
 246         err = validate_rulemsg(frh, tb, ops);
 247         if (err < 0)
 248                 goto errout;
 249
 250         rule = kzalloc(ops->rule_size, GFP_KERNEL);
 251         if (rule == NULL) {
 252                 err = -ENOMEM;
 253                 goto errout;
 254         }
 255
 256         if (tb[FRA_PRIORITY])
 257                 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
 258
 259         if (tb[FRA_IFNAME]) {
 260                 struct net_device *dev;
 261
 262                 rule->ifindex = -1;
 263                 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
 264                 dev = __dev_get_by_name(net, rule->ifname);
 265                 if (dev)
 266                         rule->ifindex = dev->ifindex;
 267         }
 268
 269         if (tb[FRA_FWMARK]) {
 270                 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
 271                 if (rule->mark)
 272                         /* compatibility: if the mark value is non-zero all bits
 273                          * are compared unless a mask is explicitly specified.
 274                          */
 275                         rule->mark_mask = 0xFFFFFFFF;
 276         }
 277
 278         if (tb[FRA_FWMASK])
 279                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
 280
 281         rule->action = frh->action;
 282         rule->flags = frh->flags;
 283         rule->table = frh_get_table(frh, tb);
 284
 285         if (!rule->pref && ops->default_pref)
 286                 rule->pref = ops->default_pref(ops);
 287
 288         err = -EINVAL;
 289         if (tb[FRA_GOTO]) {
 290                 if (rule->action != FR_ACT_GOTO)
 291                         goto errout_free;
 292
 293                 rule->target = nla_get_u32(tb[FRA_GOTO]);
 294                 /* Backward jumps are prohibited to avoid endless loops */
 295                 if (rule->target <= rule->pref)
 296                         goto errout_free;
 297
 298                 list_for_each_entry(r, &ops->rules_list, list) {
 299                         if (r->pref == rule->target) {
 300                                 rule->ctarget = r;
 301                                 break;
 302                         }
 303                 }
 304
 305                 if (rule->ctarget == NULL)
 306                         unresolved = 1;
 307         } else if (rule->action == FR_ACT_GOTO)
 308                 goto errout_free;
 309
 310         err = ops->configure(rule, skb, nlh, frh, tb);
 311         if (err < 0)
 312                 goto errout_free;
 313
 314         list_for_each_entry(r, &ops->rules_list, list) {
 315                 if (r->pref > rule->pref)
 316                         break;
 317                 last = r;
 318         }
 319
 320         fib_rule_get(rule);
 321
 322         if (ops->unresolved_rules) {
 323                 /*
 324                  * There are unresolved goto rules in the list, check if
 325                  * any of them are pointing to this new rule.
 326                  */
 327                 list_for_each_entry(r, &ops->rules_list, list) {
 328                         if (r->action == FR_ACT_GOTO &&
 329                             r->target == rule->pref) {
 330                                 BUG_ON(r->ctarget != NULL);
 331                                 rcu_assign_pointer(r->ctarget, rule);
 332                                 if (--ops->unresolved_rules == 0)
 333                                         break;
 334                         }
 335                 }
 336         }
 337
 338         if (rule->action == FR_ACT_GOTO)
 339                 ops->nr_goto_rules++;
 340
 341         if (unresolved)
 342                 ops->unresolved_rules++;
 343
 344         if (last)
 345                 list_add_rcu(&rule->list, &last->list);
 346         else
 347                 list_add_rcu(&rule->list, &ops->rules_list);
 348
 349         notify_rule_change(net, RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
 350         flush_route_cache(ops);
 351         rules_ops_put(ops);
 352         return 0;
 353
 354 errout_free:
 355         kfree(rule);
 356 errout:
 357         rules_ops_put(ops);
 358         return err;
 359 }
 360
 361 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 362 {
 363         struct net *net = skb->sk->sk_net;
 364         struct fib_rule_hdr *frh = nlmsg_data(nlh);
 365         struct fib_rules_ops *ops = NULL;
 366         struct fib_rule *rule, *tmp;
 367         struct nlattr *tb[FRA_MAX+1];
 368         int err = -EINVAL;
 369
 370         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
 371                 goto errout;
 372
 373         ops = lookup_rules_ops(net, frh->family);
 374         if (ops == NULL) {
 375                 err = EAFNOSUPPORT;
 376                 goto errout;
 377         }
 378
 379         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
 380         if (err < 0)
 381                 goto errout;
 382
 383         err = validate_rulemsg(frh, tb, ops);
 384         if (err < 0)
 385                 goto errout;
 386
 387         list_for_each_entry(rule, &ops->rules_list, list) {
 388                 if (frh->action && (frh->action != rule->action))
 389                         continue;
 390
 391                 if (frh->table && (frh_get_table(frh, tb) != rule->table))
 392                         continue;
 393
 394                 if (tb[FRA_PRIORITY] &&
 395                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
 396                         continue;
 397
 398                 if (tb[FRA_IFNAME] &&
 399                     nla_strcmp(tb[FRA_IFNAME], rule->ifname))
 400                         continue;
 401
 402                 if (tb[FRA_FWMARK] &&
 403                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
 404                         continue;
 405
 406                 if (tb[FRA_FWMASK] &&
 407                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
 408                         continue;
 409
 410                 if (!ops->compare(rule, frh, tb))
 411                         continue;
 412
 413                 if (rule->flags & FIB_RULE_PERMANENT) {
 414                         err = -EPERM;
 415                         goto errout;
 416                 }
 417
 418                 list_del_rcu(&rule->list);
 419
 420                 if (rule->action == FR_ACT_GOTO)
 421                         ops->nr_goto_rules--;
 422
 423                 /*
 424                  * Check if this rule is a target to any of them. If so,
 425                  * disable them. As this operation is eventually very
 426                  * expensive, it is only performed if goto rules have
 427                  * actually been added.
 428                  */
 429                 if (ops->nr_goto_rules > 0) {
 430                         list_for_each_entry(tmp, &ops->rules_list, list) {
 431                                 if (tmp->ctarget == rule) {
 432                                         rcu_assign_pointer(tmp->ctarget, NULL);
 433                                         ops->unresolved_rules++;
 434                                 }
 435                         }
 436                 }
 437
 438                 synchronize_rcu();
 439                 notify_rule_change(net, RTM_DELRULE, rule, ops, nlh,
 440                                    NETLINK_CB(skb).pid);
 441                 fib_rule_put(rule);
 442                 flush_route_cache(ops);
 443                 rules_ops_put(ops);
 444                 return 0;
 445         }
 446
 447         err = -ENOENT;
 448 errout:
 449         rules_ops_put(ops);
 450         return err;
 451 }
 452
 453 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
 454                                          struct fib_rule *rule)
 455 {
 456         size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
 457                          + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
 458                          + nla_total_size(4) /* FRA_PRIORITY */
 459                          + nla_total_size(4) /* FRA_TABLE */
 460                          + nla_total_size(4) /* FRA_FWMARK */
 461                          + nla_total_size(4); /* FRA_FWMASK */
 462
 463         if (ops->nlmsg_payload)
 464                 payload += ops->nlmsg_payload(rule);
 465
 466         return payload;
 467 }
 468
 469 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
 470                             u32 pid, u32 seq, int type, int flags,
 471                             struct fib_rules_ops *ops)
 472 {
 473         struct nlmsghdr *nlh;
 474         struct fib_rule_hdr *frh;
 475
 476         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
 477         if (nlh == NULL)
 478                 return -EMSGSIZE;
 479
 480         frh = nlmsg_data(nlh);
 481         frh->table = rule->table;
 482         NLA_PUT_U32(skb, FRA_TABLE, rule->table);
 483         frh->res1 = 0;
 484         frh->res2 = 0;
 485         frh->action = rule->action;
 486         frh->flags = rule->flags;
 487
 488         if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
 489                 frh->flags |= FIB_RULE_UNRESOLVED;
 490
 491         if (rule->ifname[0]) {
 492                 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
 493
 494                 if (rule->ifindex == -1)
 495                         frh->flags |= FIB_RULE_DEV_DETACHED;
 496         }
 497
 498         if (rule->pref)
 499                 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
 500
 501         if (rule->mark)
 502                 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
 503
 504         if (rule->mark_mask || rule->mark)
 505                 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
 506
 507         if (rule->target)
 508                 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
 509
 510         if (ops->fill(rule, skb, nlh, frh) < 0)
 511                 goto nla_put_failure;
 512
 513         return nlmsg_end(skb, nlh);
 514
 515 nla_put_failure:
 516         nlmsg_cancel(skb, nlh);
 517         return -EMSGSIZE;
 518 }
 519
 520 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
 521                       struct fib_rules_ops *ops)
 522 {
 523         int idx = 0;
 524         struct fib_rule *rule;
 525
 526         list_for_each_entry(rule, &ops->rules_list, list) {
 527                 if (idx < cb->args[1])
 528                         goto skip;
 529
 530                 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
 531                                      cb->nlh->nlmsg_seq, RTM_NEWRULE,
 532                                      NLM_F_MULTI, ops) < 0)
 533                         break;
 534 skip:
 535                 idx++;
 536         }
 537         cb->args[1] = idx;
 538         rules_ops_put(ops);
 539
 540         return skb->len;
 541 }
 542
 543 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
 544 {
 545         struct net *net = skb->sk->sk_net;
 546         struct fib_rules_ops *ops;
 547         int idx = 0, family;
 548
 549         family = rtnl_msg_family(cb->nlh);
 550         if (family != AF_UNSPEC) {
 551                 /* Protocol specific dump request */
 552                 ops = lookup_rules_ops(net, family);
 553                 if (ops == NULL)
 554                         return -EAFNOSUPPORT;
 555
 556                 return dump_rules(skb, cb, ops);
 557         }
 558
 559         rcu_read_lock();
 560         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
 561                 if (idx < cb->args[0] || !try_module_get(ops->owner))
 562                         goto skip;
 563
 564                 if (dump_rules(skb, cb, ops) < 0)
 565                         break;
 566
 567                 cb->args[1] = 0;
 568         skip:
 569                 idx++;
 570         }
 571         rcu_read_unlock();
 572         cb->args[0] = idx;
 573
 574         return skb->len;
 575 }
 576
 577 static void notify_rule_change(struct net *net, int event, struct fib_rule *rule,
 578                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
 579                                u32 pid)
 580 {
 581         struct sk_buff *skb;
 582         int err = -ENOBUFS;
 583
 584         skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
 585         if (skb == NULL)
 586                 goto errout;
 587
 588         err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
 589         if (err < 0) {
 590                 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
 591                 WARN_ON(err == -EMSGSIZE);
 592                 kfree_skb(skb);
 593                 goto errout;
 594         }
 595         err = rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
 596 errout:
 597         if (err < 0)
 598                 rtnl_set_sk_err(net, ops->nlgroup, err);
 599 }
 600
 601 static void attach_rules(struct list_head *rules, struct net_device *dev)
 602 {
 603         struct fib_rule *rule;
 604
 605         list_for_each_entry(rule, rules, list) {
 606                 if (rule->ifindex == -1 &&
 607                     strcmp(dev->name, rule->ifname) == 0)
 608                         rule->ifindex = dev->ifindex;
 609         }
 610 }
 611
 612 static void detach_rules(struct list_head *rules, struct net_device *dev)
 613 {
 614         struct fib_rule *rule;
 615
 616         list_for_each_entry(rule, rules, list)
 617                 if (rule->ifindex == dev->ifindex)
 618                         rule->ifindex = -1;
 619 }
 620
 621
 622 static int fib_rules_event(struct notifier_block *this, unsigned long event,
 623                             void *ptr)
 624 {
 625         struct net_device *dev = ptr;
 626         struct net *net = dev->nd_net;
 627         struct fib_rules_ops *ops;
 628
 629         ASSERT_RTNL();
 630         rcu_read_lock();
 631
 632         switch (event) {
 633         case NETDEV_REGISTER:
 634                 list_for_each_entry(ops, &net->rules_ops, list)
 635                         attach_rules(&ops->rules_list, dev);
 636                 break;
 637
 638         case NETDEV_UNREGISTER:
 639                 list_for_each_entry(ops, &net->rules_ops, list)
 640                         detach_rules(&ops->rules_list, dev);
 641                 break;
 642         }
 643
 644         rcu_read_unlock();
 645
 646         return NOTIFY_DONE;
 647 }
 648
 649 static struct notifier_block fib_rules_notifier = {
 650         .notifier_call = fib_rules_event,
 651 };
 652
 653 static int fib_rules_net_init(struct net *net)
 654 {
 655         INIT_LIST_HEAD(&net->rules_ops);
 656         spin_lock_init(&net->rules_mod_lock);
 657         return 0;
 658 }
 659
 660 static struct pernet_operations fib_rules_net_ops = {
 661         .init = fib_rules_net_init,
 662 };
 663
 664 static int __init fib_rules_init(void)
 665 {
 666         int err;
 667         rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
 668         rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
 669         rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
 670
 671         err = register_netdevice_notifier(&fib_rules_notifier);
 672         if (err < 0)
 673                 goto fail;
 674
 675         err = register_pernet_subsys(&fib_rules_net_ops);
 676         if (err < 0)
 677                 goto fail_unregister;
 678         return 0;
 679
 680 fail_unregister:
 681         unregister_netdevice_notifier(&fib_rules_notifier);
 682 fail:
 683         rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
 684         rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
 685         rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
 686         return err;
 687 }
 688
 689 subsys_initcall(fib_rules_init);