Loading...
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
21int 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}
45EXPORT_SYMBOL(fib_default_rule_add);
46
47static 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
64static void notify_rule_change(int event, struct fib_rule *rule,
65 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
66 u32 pid);
67
68static 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
86static void rules_ops_put(struct fib_rules_ops *ops)
87{
88 if (ops)
89 module_put(ops->owner);
90}
91
92static void flush_route_cache(struct fib_rules_ops *ops)
93{
94 if (ops->flush_cache)
95 ops->flush_cache(ops);
96}
97
98static 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;
121errout:
122 spin_unlock(&net->rules_mod_lock);
123
124 return err;
125}
126
127struct fib_rules_ops *
128fib_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}
148EXPORT_SYMBOL_GPL(fib_rules_register);
149
150static 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
162void 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}
173EXPORT_SYMBOL_GPL(fib_rules_unregister);
174
175static 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);
193out:
194 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
195}
196
197int 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) {
206jumped:
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;
239out:
240 rcu_read_unlock();
241
242 return err;
243}
244EXPORT_SYMBOL_GPL(fib_rules_lookup);
245
246static 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;
264errout:
265 return err;
266}
267
268static 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
421errout_free:
422 kfree(rule);
423errout:
424 rules_ops_put(ops);
425 return err;
426}
427
428static 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;
535errout:
536 rules_ops_put(ops);
537 return err;
538}
539
540static 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
560static 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
624nla_put_failure:
625 nlmsg_cancel(skb, nlh);
626 return -EMSGSIZE;
627}
628
629static 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;
646skip:
647 idx++;
648 }
649 rcu_read_unlock();
650 cb->args[1] = idx;
651 rules_ops_put(ops);
652
653 return err;
654}
655
656static 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;
683skip:
684 idx++;
685 }
686 rcu_read_unlock();
687 cb->args[0] = idx;
688
689 return skb->len;
690}
691
692static 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;
715errout:
716 if (err < 0)
717 rtnl_set_sk_err(net, ops->nlgroup, err);
718}
719
720static 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
734static 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
747static 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
778static struct notifier_block fib_rules_notifier = {
779 .notifier_call = fib_rules_event,
780};
781
782static 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
789static struct pernet_operations fib_rules_net_ops = {
790 .init = fib_rules_net_init,
791};
792
793static 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
810fail_unregister:
811 unregister_pernet_subsys(&fib_rules_net_ops);
812fail:
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
819subsys_initcall(fib_rules_init);
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
21static const struct fib_kuid_range fib_kuid_range_unset = {
22 KUIDT_INIT(0),
23 KUIDT_INIT(~0),
24};
25
26int fib_default_rule_add(struct fib_rules_ops *ops,
27 u32 pref, u32 table, u32 flags)
28{
29 struct fib_rule *r;
30
31 r = kzalloc(ops->rule_size, GFP_KERNEL);
32 if (r == NULL)
33 return -ENOMEM;
34
35 atomic_set(&r->refcnt, 1);
36 r->action = FR_ACT_TO_TBL;
37 r->pref = pref;
38 r->table = table;
39 r->flags = flags;
40 r->fr_net = ops->fro_net;
41 r->uid_range = fib_kuid_range_unset;
42
43 r->suppress_prefixlen = -1;
44 r->suppress_ifgroup = -1;
45
46 /* The lock is not required here, the list in unreacheable
47 * at the moment this function is called */
48 list_add_tail(&r->list, &ops->rules_list);
49 return 0;
50}
51EXPORT_SYMBOL(fib_default_rule_add);
52
53static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
54{
55 struct list_head *pos;
56 struct fib_rule *rule;
57
58 if (!list_empty(&ops->rules_list)) {
59 pos = ops->rules_list.next;
60 if (pos->next != &ops->rules_list) {
61 rule = list_entry(pos->next, struct fib_rule, list);
62 if (rule->pref)
63 return rule->pref - 1;
64 }
65 }
66
67 return 0;
68}
69
70static void notify_rule_change(int event, struct fib_rule *rule,
71 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
72 u32 pid);
73
74static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
75{
76 struct fib_rules_ops *ops;
77
78 rcu_read_lock();
79 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
80 if (ops->family == family) {
81 if (!try_module_get(ops->owner))
82 ops = NULL;
83 rcu_read_unlock();
84 return ops;
85 }
86 }
87 rcu_read_unlock();
88
89 return NULL;
90}
91
92static void rules_ops_put(struct fib_rules_ops *ops)
93{
94 if (ops)
95 module_put(ops->owner);
96}
97
98static void flush_route_cache(struct fib_rules_ops *ops)
99{
100 if (ops->flush_cache)
101 ops->flush_cache(ops);
102}
103
104static int __fib_rules_register(struct fib_rules_ops *ops)
105{
106 int err = -EEXIST;
107 struct fib_rules_ops *o;
108 struct net *net;
109
110 net = ops->fro_net;
111
112 if (ops->rule_size < sizeof(struct fib_rule))
113 return -EINVAL;
114
115 if (ops->match == NULL || ops->configure == NULL ||
116 ops->compare == NULL || ops->fill == NULL ||
117 ops->action == NULL)
118 return -EINVAL;
119
120 spin_lock(&net->rules_mod_lock);
121 list_for_each_entry(o, &net->rules_ops, list)
122 if (ops->family == o->family)
123 goto errout;
124
125 list_add_tail_rcu(&ops->list, &net->rules_ops);
126 err = 0;
127errout:
128 spin_unlock(&net->rules_mod_lock);
129
130 return err;
131}
132
133struct fib_rules_ops *
134fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
135{
136 struct fib_rules_ops *ops;
137 int err;
138
139 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
140 if (ops == NULL)
141 return ERR_PTR(-ENOMEM);
142
143 INIT_LIST_HEAD(&ops->rules_list);
144 ops->fro_net = net;
145
146 err = __fib_rules_register(ops);
147 if (err) {
148 kfree(ops);
149 ops = ERR_PTR(err);
150 }
151
152 return ops;
153}
154EXPORT_SYMBOL_GPL(fib_rules_register);
155
156static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
157{
158 struct fib_rule *rule, *tmp;
159
160 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
161 list_del_rcu(&rule->list);
162 if (ops->delete)
163 ops->delete(rule);
164 fib_rule_put(rule);
165 }
166}
167
168void fib_rules_unregister(struct fib_rules_ops *ops)
169{
170 struct net *net = ops->fro_net;
171
172 spin_lock(&net->rules_mod_lock);
173 list_del_rcu(&ops->list);
174 spin_unlock(&net->rules_mod_lock);
175
176 fib_rules_cleanup_ops(ops);
177 kfree_rcu(ops, rcu);
178}
179EXPORT_SYMBOL_GPL(fib_rules_unregister);
180
181static int uid_range_set(struct fib_kuid_range *range)
182{
183 return uid_valid(range->start) && uid_valid(range->end);
184}
185
186static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
187{
188 struct fib_rule_uid_range *in;
189 struct fib_kuid_range out;
190
191 in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
192
193 out.start = make_kuid(current_user_ns(), in->start);
194 out.end = make_kuid(current_user_ns(), in->end);
195
196 return out;
197}
198
199static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
200{
201 struct fib_rule_uid_range out = {
202 from_kuid_munged(current_user_ns(), range->start),
203 from_kuid_munged(current_user_ns(), range->end)
204 };
205
206 return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
207}
208
209static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
210 struct flowi *fl, int flags,
211 struct fib_lookup_arg *arg)
212{
213 int ret = 0;
214
215 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
216 goto out;
217
218 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
219 goto out;
220
221 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
222 goto out;
223
224 if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
225 goto out;
226
227 if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
228 goto out;
229
230 if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
231 uid_gt(fl->flowi_uid, rule->uid_range.end))
232 goto out;
233
234 ret = ops->match(rule, fl, flags);
235out:
236 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
237}
238
239int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
240 int flags, struct fib_lookup_arg *arg)
241{
242 struct fib_rule *rule;
243 int err;
244
245 rcu_read_lock();
246
247 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
248jumped:
249 if (!fib_rule_match(rule, ops, fl, flags, arg))
250 continue;
251
252 if (rule->action == FR_ACT_GOTO) {
253 struct fib_rule *target;
254
255 target = rcu_dereference(rule->ctarget);
256 if (target == NULL) {
257 continue;
258 } else {
259 rule = target;
260 goto jumped;
261 }
262 } else if (rule->action == FR_ACT_NOP)
263 continue;
264 else
265 err = ops->action(rule, fl, flags, arg);
266
267 if (!err && ops->suppress && ops->suppress(rule, arg))
268 continue;
269
270 if (err != -EAGAIN) {
271 if ((arg->flags & FIB_LOOKUP_NOREF) ||
272 likely(atomic_inc_not_zero(&rule->refcnt))) {
273 arg->rule = rule;
274 goto out;
275 }
276 break;
277 }
278 }
279
280 err = -ESRCH;
281out:
282 rcu_read_unlock();
283
284 return err;
285}
286EXPORT_SYMBOL_GPL(fib_rules_lookup);
287
288static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
289 struct fib_rules_ops *ops)
290{
291 int err = -EINVAL;
292
293 if (frh->src_len)
294 if (tb[FRA_SRC] == NULL ||
295 frh->src_len > (ops->addr_size * 8) ||
296 nla_len(tb[FRA_SRC]) != ops->addr_size)
297 goto errout;
298
299 if (frh->dst_len)
300 if (tb[FRA_DST] == NULL ||
301 frh->dst_len > (ops->addr_size * 8) ||
302 nla_len(tb[FRA_DST]) != ops->addr_size)
303 goto errout;
304
305 err = 0;
306errout:
307 return err;
308}
309
310static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
311 struct nlattr **tb, struct fib_rule *rule)
312{
313 struct fib_rule *r;
314
315 list_for_each_entry(r, &ops->rules_list, list) {
316 if (r->action != rule->action)
317 continue;
318
319 if (r->table != rule->table)
320 continue;
321
322 if (r->pref != rule->pref)
323 continue;
324
325 if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
326 continue;
327
328 if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
329 continue;
330
331 if (r->mark != rule->mark)
332 continue;
333
334 if (r->mark_mask != rule->mark_mask)
335 continue;
336
337 if (r->tun_id != rule->tun_id)
338 continue;
339
340 if (r->fr_net != rule->fr_net)
341 continue;
342
343 if (r->l3mdev != rule->l3mdev)
344 continue;
345
346 if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
347 !uid_eq(r->uid_range.end, rule->uid_range.end))
348 continue;
349
350 if (!ops->compare(r, frh, tb))
351 continue;
352 return 1;
353 }
354 return 0;
355}
356
357int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh)
358{
359 struct net *net = sock_net(skb->sk);
360 struct fib_rule_hdr *frh = nlmsg_data(nlh);
361 struct fib_rules_ops *ops = NULL;
362 struct fib_rule *rule, *r, *last = NULL;
363 struct nlattr *tb[FRA_MAX+1];
364 int err = -EINVAL, unresolved = 0;
365
366 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
367 goto errout;
368
369 ops = lookup_rules_ops(net, frh->family);
370 if (ops == NULL) {
371 err = -EAFNOSUPPORT;
372 goto errout;
373 }
374
375 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
376 if (err < 0)
377 goto errout;
378
379 err = validate_rulemsg(frh, tb, ops);
380 if (err < 0)
381 goto errout;
382
383 rule = kzalloc(ops->rule_size, GFP_KERNEL);
384 if (rule == NULL) {
385 err = -ENOMEM;
386 goto errout;
387 }
388 rule->fr_net = net;
389
390 rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
391 : fib_default_rule_pref(ops);
392
393 if (tb[FRA_IIFNAME]) {
394 struct net_device *dev;
395
396 rule->iifindex = -1;
397 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
398 dev = __dev_get_by_name(net, rule->iifname);
399 if (dev)
400 rule->iifindex = dev->ifindex;
401 }
402
403 if (tb[FRA_OIFNAME]) {
404 struct net_device *dev;
405
406 rule->oifindex = -1;
407 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
408 dev = __dev_get_by_name(net, rule->oifname);
409 if (dev)
410 rule->oifindex = dev->ifindex;
411 }
412
413 if (tb[FRA_FWMARK]) {
414 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
415 if (rule->mark)
416 /* compatibility: if the mark value is non-zero all bits
417 * are compared unless a mask is explicitly specified.
418 */
419 rule->mark_mask = 0xFFFFFFFF;
420 }
421
422 if (tb[FRA_FWMASK])
423 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
424
425 if (tb[FRA_TUN_ID])
426 rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
427
428 if (tb[FRA_L3MDEV]) {
429#ifdef CONFIG_NET_L3_MASTER_DEV
430 rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
431 if (rule->l3mdev != 1)
432#endif
433 goto errout_free;
434 }
435
436 rule->action = frh->action;
437 rule->flags = frh->flags;
438 rule->table = frh_get_table(frh, tb);
439 if (tb[FRA_SUPPRESS_PREFIXLEN])
440 rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
441 else
442 rule->suppress_prefixlen = -1;
443
444 if (tb[FRA_SUPPRESS_IFGROUP])
445 rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
446 else
447 rule->suppress_ifgroup = -1;
448
449 err = -EINVAL;
450 if (tb[FRA_GOTO]) {
451 if (rule->action != FR_ACT_GOTO)
452 goto errout_free;
453
454 rule->target = nla_get_u32(tb[FRA_GOTO]);
455 /* Backward jumps are prohibited to avoid endless loops */
456 if (rule->target <= rule->pref)
457 goto errout_free;
458
459 list_for_each_entry(r, &ops->rules_list, list) {
460 if (r->pref == rule->target) {
461 RCU_INIT_POINTER(rule->ctarget, r);
462 break;
463 }
464 }
465
466 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
467 unresolved = 1;
468 } else if (rule->action == FR_ACT_GOTO)
469 goto errout_free;
470
471 if (rule->l3mdev && rule->table)
472 goto errout_free;
473
474 if (tb[FRA_UID_RANGE]) {
475 if (current_user_ns() != net->user_ns) {
476 err = -EPERM;
477 goto errout_free;
478 }
479
480 rule->uid_range = nla_get_kuid_range(tb);
481
482 if (!uid_range_set(&rule->uid_range) ||
483 !uid_lte(rule->uid_range.start, rule->uid_range.end))
484 goto errout_free;
485 } else {
486 rule->uid_range = fib_kuid_range_unset;
487 }
488
489 if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
490 rule_exists(ops, frh, tb, rule)) {
491 err = -EEXIST;
492 goto errout_free;
493 }
494
495 err = ops->configure(rule, skb, frh, tb);
496 if (err < 0)
497 goto errout_free;
498
499 list_for_each_entry(r, &ops->rules_list, list) {
500 if (r->pref > rule->pref)
501 break;
502 last = r;
503 }
504
505 fib_rule_get(rule);
506
507 if (last)
508 list_add_rcu(&rule->list, &last->list);
509 else
510 list_add_rcu(&rule->list, &ops->rules_list);
511
512 if (ops->unresolved_rules) {
513 /*
514 * There are unresolved goto rules in the list, check if
515 * any of them are pointing to this new rule.
516 */
517 list_for_each_entry(r, &ops->rules_list, list) {
518 if (r->action == FR_ACT_GOTO &&
519 r->target == rule->pref &&
520 rtnl_dereference(r->ctarget) == NULL) {
521 rcu_assign_pointer(r->ctarget, rule);
522 if (--ops->unresolved_rules == 0)
523 break;
524 }
525 }
526 }
527
528 if (rule->action == FR_ACT_GOTO)
529 ops->nr_goto_rules++;
530
531 if (unresolved)
532 ops->unresolved_rules++;
533
534 if (rule->tun_id)
535 ip_tunnel_need_metadata();
536
537 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
538 flush_route_cache(ops);
539 rules_ops_put(ops);
540 return 0;
541
542errout_free:
543 kfree(rule);
544errout:
545 rules_ops_put(ops);
546 return err;
547}
548EXPORT_SYMBOL_GPL(fib_nl_newrule);
549
550int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh)
551{
552 struct net *net = sock_net(skb->sk);
553 struct fib_rule_hdr *frh = nlmsg_data(nlh);
554 struct fib_rules_ops *ops = NULL;
555 struct fib_rule *rule, *tmp;
556 struct nlattr *tb[FRA_MAX+1];
557 struct fib_kuid_range range;
558 int err = -EINVAL;
559
560 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
561 goto errout;
562
563 ops = lookup_rules_ops(net, frh->family);
564 if (ops == NULL) {
565 err = -EAFNOSUPPORT;
566 goto errout;
567 }
568
569 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
570 if (err < 0)
571 goto errout;
572
573 err = validate_rulemsg(frh, tb, ops);
574 if (err < 0)
575 goto errout;
576
577 if (tb[FRA_UID_RANGE]) {
578 range = nla_get_kuid_range(tb);
579 if (!uid_range_set(&range))
580 goto errout;
581 } else {
582 range = fib_kuid_range_unset;
583 }
584
585 list_for_each_entry(rule, &ops->rules_list, list) {
586 if (frh->action && (frh->action != rule->action))
587 continue;
588
589 if (frh_get_table(frh, tb) &&
590 (frh_get_table(frh, tb) != rule->table))
591 continue;
592
593 if (tb[FRA_PRIORITY] &&
594 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
595 continue;
596
597 if (tb[FRA_IIFNAME] &&
598 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
599 continue;
600
601 if (tb[FRA_OIFNAME] &&
602 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
603 continue;
604
605 if (tb[FRA_FWMARK] &&
606 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
607 continue;
608
609 if (tb[FRA_FWMASK] &&
610 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
611 continue;
612
613 if (tb[FRA_TUN_ID] &&
614 (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
615 continue;
616
617 if (tb[FRA_L3MDEV] &&
618 (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
619 continue;
620
621 if (uid_range_set(&range) &&
622 (!uid_eq(rule->uid_range.start, range.start) ||
623 !uid_eq(rule->uid_range.end, range.end)))
624 continue;
625
626 if (!ops->compare(rule, frh, tb))
627 continue;
628
629 if (rule->flags & FIB_RULE_PERMANENT) {
630 err = -EPERM;
631 goto errout;
632 }
633
634 if (ops->delete) {
635 err = ops->delete(rule);
636 if (err)
637 goto errout;
638 }
639
640 if (rule->tun_id)
641 ip_tunnel_unneed_metadata();
642
643 list_del_rcu(&rule->list);
644
645 if (rule->action == FR_ACT_GOTO) {
646 ops->nr_goto_rules--;
647 if (rtnl_dereference(rule->ctarget) == NULL)
648 ops->unresolved_rules--;
649 }
650
651 /*
652 * Check if this rule is a target to any of them. If so,
653 * disable them. As this operation is eventually very
654 * expensive, it is only performed if goto rules have
655 * actually been added.
656 */
657 if (ops->nr_goto_rules > 0) {
658 list_for_each_entry(tmp, &ops->rules_list, list) {
659 if (rtnl_dereference(tmp->ctarget) == rule) {
660 RCU_INIT_POINTER(tmp->ctarget, NULL);
661 ops->unresolved_rules++;
662 }
663 }
664 }
665
666 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
667 NETLINK_CB(skb).portid);
668 fib_rule_put(rule);
669 flush_route_cache(ops);
670 rules_ops_put(ops);
671 return 0;
672 }
673
674 err = -ENOENT;
675errout:
676 rules_ops_put(ops);
677 return err;
678}
679EXPORT_SYMBOL_GPL(fib_nl_delrule);
680
681static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
682 struct fib_rule *rule)
683{
684 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
685 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
686 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
687 + nla_total_size(4) /* FRA_PRIORITY */
688 + nla_total_size(4) /* FRA_TABLE */
689 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
690 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
691 + nla_total_size(4) /* FRA_FWMARK */
692 + nla_total_size(4) /* FRA_FWMASK */
693 + nla_total_size_64bit(8) /* FRA_TUN_ID */
694 + nla_total_size(sizeof(struct fib_kuid_range));
695
696 if (ops->nlmsg_payload)
697 payload += ops->nlmsg_payload(rule);
698
699 return payload;
700}
701
702static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
703 u32 pid, u32 seq, int type, int flags,
704 struct fib_rules_ops *ops)
705{
706 struct nlmsghdr *nlh;
707 struct fib_rule_hdr *frh;
708
709 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
710 if (nlh == NULL)
711 return -EMSGSIZE;
712
713 frh = nlmsg_data(nlh);
714 frh->family = ops->family;
715 frh->table = rule->table;
716 if (nla_put_u32(skb, FRA_TABLE, rule->table))
717 goto nla_put_failure;
718 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
719 goto nla_put_failure;
720 frh->res1 = 0;
721 frh->res2 = 0;
722 frh->action = rule->action;
723 frh->flags = rule->flags;
724
725 if (rule->action == FR_ACT_GOTO &&
726 rcu_access_pointer(rule->ctarget) == NULL)
727 frh->flags |= FIB_RULE_UNRESOLVED;
728
729 if (rule->iifname[0]) {
730 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
731 goto nla_put_failure;
732 if (rule->iifindex == -1)
733 frh->flags |= FIB_RULE_IIF_DETACHED;
734 }
735
736 if (rule->oifname[0]) {
737 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
738 goto nla_put_failure;
739 if (rule->oifindex == -1)
740 frh->flags |= FIB_RULE_OIF_DETACHED;
741 }
742
743 if ((rule->pref &&
744 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
745 (rule->mark &&
746 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
747 ((rule->mark_mask || rule->mark) &&
748 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
749 (rule->target &&
750 nla_put_u32(skb, FRA_GOTO, rule->target)) ||
751 (rule->tun_id &&
752 nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
753 (rule->l3mdev &&
754 nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
755 (uid_range_set(&rule->uid_range) &&
756 nla_put_uid_range(skb, &rule->uid_range)))
757 goto nla_put_failure;
758
759 if (rule->suppress_ifgroup != -1) {
760 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
761 goto nla_put_failure;
762 }
763
764 if (ops->fill(rule, skb, frh) < 0)
765 goto nla_put_failure;
766
767 nlmsg_end(skb, nlh);
768 return 0;
769
770nla_put_failure:
771 nlmsg_cancel(skb, nlh);
772 return -EMSGSIZE;
773}
774
775static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
776 struct fib_rules_ops *ops)
777{
778 int idx = 0;
779 struct fib_rule *rule;
780 int err = 0;
781
782 rcu_read_lock();
783 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
784 if (idx < cb->args[1])
785 goto skip;
786
787 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
788 cb->nlh->nlmsg_seq, RTM_NEWRULE,
789 NLM_F_MULTI, ops);
790 if (err)
791 break;
792skip:
793 idx++;
794 }
795 rcu_read_unlock();
796 cb->args[1] = idx;
797 rules_ops_put(ops);
798
799 return err;
800}
801
802static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
803{
804 struct net *net = sock_net(skb->sk);
805 struct fib_rules_ops *ops;
806 int idx = 0, family;
807
808 family = rtnl_msg_family(cb->nlh);
809 if (family != AF_UNSPEC) {
810 /* Protocol specific dump request */
811 ops = lookup_rules_ops(net, family);
812 if (ops == NULL)
813 return -EAFNOSUPPORT;
814
815 dump_rules(skb, cb, ops);
816
817 return skb->len;
818 }
819
820 rcu_read_lock();
821 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
822 if (idx < cb->args[0] || !try_module_get(ops->owner))
823 goto skip;
824
825 if (dump_rules(skb, cb, ops) < 0)
826 break;
827
828 cb->args[1] = 0;
829skip:
830 idx++;
831 }
832 rcu_read_unlock();
833 cb->args[0] = idx;
834
835 return skb->len;
836}
837
838static void notify_rule_change(int event, struct fib_rule *rule,
839 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
840 u32 pid)
841{
842 struct net *net;
843 struct sk_buff *skb;
844 int err = -ENOBUFS;
845
846 net = ops->fro_net;
847 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
848 if (skb == NULL)
849 goto errout;
850
851 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
852 if (err < 0) {
853 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
854 WARN_ON(err == -EMSGSIZE);
855 kfree_skb(skb);
856 goto errout;
857 }
858
859 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
860 return;
861errout:
862 if (err < 0)
863 rtnl_set_sk_err(net, ops->nlgroup, err);
864}
865
866static void attach_rules(struct list_head *rules, struct net_device *dev)
867{
868 struct fib_rule *rule;
869
870 list_for_each_entry(rule, rules, list) {
871 if (rule->iifindex == -1 &&
872 strcmp(dev->name, rule->iifname) == 0)
873 rule->iifindex = dev->ifindex;
874 if (rule->oifindex == -1 &&
875 strcmp(dev->name, rule->oifname) == 0)
876 rule->oifindex = dev->ifindex;
877 }
878}
879
880static void detach_rules(struct list_head *rules, struct net_device *dev)
881{
882 struct fib_rule *rule;
883
884 list_for_each_entry(rule, rules, list) {
885 if (rule->iifindex == dev->ifindex)
886 rule->iifindex = -1;
887 if (rule->oifindex == dev->ifindex)
888 rule->oifindex = -1;
889 }
890}
891
892
893static int fib_rules_event(struct notifier_block *this, unsigned long event,
894 void *ptr)
895{
896 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
897 struct net *net = dev_net(dev);
898 struct fib_rules_ops *ops;
899
900 ASSERT_RTNL();
901
902 switch (event) {
903 case NETDEV_REGISTER:
904 list_for_each_entry(ops, &net->rules_ops, list)
905 attach_rules(&ops->rules_list, dev);
906 break;
907
908 case NETDEV_CHANGENAME:
909 list_for_each_entry(ops, &net->rules_ops, list) {
910 detach_rules(&ops->rules_list, dev);
911 attach_rules(&ops->rules_list, dev);
912 }
913 break;
914
915 case NETDEV_UNREGISTER:
916 list_for_each_entry(ops, &net->rules_ops, list)
917 detach_rules(&ops->rules_list, dev);
918 break;
919 }
920
921 return NOTIFY_DONE;
922}
923
924static struct notifier_block fib_rules_notifier = {
925 .notifier_call = fib_rules_event,
926};
927
928static int __net_init fib_rules_net_init(struct net *net)
929{
930 INIT_LIST_HEAD(&net->rules_ops);
931 spin_lock_init(&net->rules_mod_lock);
932 return 0;
933}
934
935static struct pernet_operations fib_rules_net_ops = {
936 .init = fib_rules_net_init,
937};
938
939static int __init fib_rules_init(void)
940{
941 int err;
942 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
943 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
944 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
945
946 err = register_pernet_subsys(&fib_rules_net_ops);
947 if (err < 0)
948 goto fail;
949
950 err = register_netdevice_notifier(&fib_rules_notifier);
951 if (err < 0)
952 goto fail_unregister;
953
954 return 0;
955
956fail_unregister:
957 unregister_pernet_subsys(&fib_rules_net_ops);
958fail:
959 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
960 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
961 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
962 return err;
963}
964
965subsys_initcall(fib_rules_init);