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// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/core/fib_rules.c Generic Routing Rules
4 *
5 * Authors: Thomas Graf <tgraf@suug.ch>
6 */
7
8#include <linux/types.h>
9#include <linux/kernel.h>
10#include <linux/slab.h>
11#include <linux/list.h>
12#include <linux/module.h>
13#include <net/net_namespace.h>
14#include <net/sock.h>
15#include <net/fib_rules.h>
16#include <net/ip_tunnels.h>
17
18static const struct fib_kuid_range fib_kuid_range_unset = {
19 KUIDT_INIT(0),
20 KUIDT_INIT(~0),
21};
22
23bool fib_rule_matchall(const struct fib_rule *rule)
24{
25 if (rule->iifindex || rule->oifindex || rule->mark || rule->tun_id ||
26 rule->flags)
27 return false;
28 if (rule->suppress_ifgroup != -1 || rule->suppress_prefixlen != -1)
29 return false;
30 if (!uid_eq(rule->uid_range.start, fib_kuid_range_unset.start) ||
31 !uid_eq(rule->uid_range.end, fib_kuid_range_unset.end))
32 return false;
33 if (fib_rule_port_range_set(&rule->sport_range))
34 return false;
35 if (fib_rule_port_range_set(&rule->dport_range))
36 return false;
37 return true;
38}
39EXPORT_SYMBOL_GPL(fib_rule_matchall);
40
41int fib_default_rule_add(struct fib_rules_ops *ops,
42 u32 pref, u32 table, u32 flags)
43{
44 struct fib_rule *r;
45
46 r = kzalloc(ops->rule_size, GFP_KERNEL);
47 if (r == NULL)
48 return -ENOMEM;
49
50 refcount_set(&r->refcnt, 1);
51 r->action = FR_ACT_TO_TBL;
52 r->pref = pref;
53 r->table = table;
54 r->flags = flags;
55 r->proto = RTPROT_KERNEL;
56 r->fr_net = ops->fro_net;
57 r->uid_range = fib_kuid_range_unset;
58
59 r->suppress_prefixlen = -1;
60 r->suppress_ifgroup = -1;
61
62 /* The lock is not required here, the list in unreacheable
63 * at the moment this function is called */
64 list_add_tail(&r->list, &ops->rules_list);
65 return 0;
66}
67EXPORT_SYMBOL(fib_default_rule_add);
68
69static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
70{
71 struct list_head *pos;
72 struct fib_rule *rule;
73
74 if (!list_empty(&ops->rules_list)) {
75 pos = ops->rules_list.next;
76 if (pos->next != &ops->rules_list) {
77 rule = list_entry(pos->next, struct fib_rule, list);
78 if (rule->pref)
79 return rule->pref - 1;
80 }
81 }
82
83 return 0;
84}
85
86static void notify_rule_change(int event, struct fib_rule *rule,
87 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
88 u32 pid);
89
90static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
91{
92 struct fib_rules_ops *ops;
93
94 rcu_read_lock();
95 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
96 if (ops->family == family) {
97 if (!try_module_get(ops->owner))
98 ops = NULL;
99 rcu_read_unlock();
100 return ops;
101 }
102 }
103 rcu_read_unlock();
104
105 return NULL;
106}
107
108static void rules_ops_put(struct fib_rules_ops *ops)
109{
110 if (ops)
111 module_put(ops->owner);
112}
113
114static void flush_route_cache(struct fib_rules_ops *ops)
115{
116 if (ops->flush_cache)
117 ops->flush_cache(ops);
118}
119
120static int __fib_rules_register(struct fib_rules_ops *ops)
121{
122 int err = -EEXIST;
123 struct fib_rules_ops *o;
124 struct net *net;
125
126 net = ops->fro_net;
127
128 if (ops->rule_size < sizeof(struct fib_rule))
129 return -EINVAL;
130
131 if (ops->match == NULL || ops->configure == NULL ||
132 ops->compare == NULL || ops->fill == NULL ||
133 ops->action == NULL)
134 return -EINVAL;
135
136 spin_lock(&net->rules_mod_lock);
137 list_for_each_entry(o, &net->rules_ops, list)
138 if (ops->family == o->family)
139 goto errout;
140
141 list_add_tail_rcu(&ops->list, &net->rules_ops);
142 err = 0;
143errout:
144 spin_unlock(&net->rules_mod_lock);
145
146 return err;
147}
148
149struct fib_rules_ops *
150fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
151{
152 struct fib_rules_ops *ops;
153 int err;
154
155 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
156 if (ops == NULL)
157 return ERR_PTR(-ENOMEM);
158
159 INIT_LIST_HEAD(&ops->rules_list);
160 ops->fro_net = net;
161
162 err = __fib_rules_register(ops);
163 if (err) {
164 kfree(ops);
165 ops = ERR_PTR(err);
166 }
167
168 return ops;
169}
170EXPORT_SYMBOL_GPL(fib_rules_register);
171
172static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
173{
174 struct fib_rule *rule, *tmp;
175
176 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
177 list_del_rcu(&rule->list);
178 if (ops->delete)
179 ops->delete(rule);
180 fib_rule_put(rule);
181 }
182}
183
184void fib_rules_unregister(struct fib_rules_ops *ops)
185{
186 struct net *net = ops->fro_net;
187
188 spin_lock(&net->rules_mod_lock);
189 list_del_rcu(&ops->list);
190 spin_unlock(&net->rules_mod_lock);
191
192 fib_rules_cleanup_ops(ops);
193 kfree_rcu(ops, rcu);
194}
195EXPORT_SYMBOL_GPL(fib_rules_unregister);
196
197static int uid_range_set(struct fib_kuid_range *range)
198{
199 return uid_valid(range->start) && uid_valid(range->end);
200}
201
202static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
203{
204 struct fib_rule_uid_range *in;
205 struct fib_kuid_range out;
206
207 in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
208
209 out.start = make_kuid(current_user_ns(), in->start);
210 out.end = make_kuid(current_user_ns(), in->end);
211
212 return out;
213}
214
215static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
216{
217 struct fib_rule_uid_range out = {
218 from_kuid_munged(current_user_ns(), range->start),
219 from_kuid_munged(current_user_ns(), range->end)
220 };
221
222 return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
223}
224
225static int nla_get_port_range(struct nlattr *pattr,
226 struct fib_rule_port_range *port_range)
227{
228 const struct fib_rule_port_range *pr = nla_data(pattr);
229
230 if (!fib_rule_port_range_valid(pr))
231 return -EINVAL;
232
233 port_range->start = pr->start;
234 port_range->end = pr->end;
235
236 return 0;
237}
238
239static int nla_put_port_range(struct sk_buff *skb, int attrtype,
240 struct fib_rule_port_range *range)
241{
242 return nla_put(skb, attrtype, sizeof(*range), range);
243}
244
245static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
246 struct flowi *fl, int flags,
247 struct fib_lookup_arg *arg)
248{
249 int ret = 0;
250
251 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
252 goto out;
253
254 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
255 goto out;
256
257 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
258 goto out;
259
260 if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
261 goto out;
262
263 if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
264 goto out;
265
266 if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
267 uid_gt(fl->flowi_uid, rule->uid_range.end))
268 goto out;
269
270 ret = ops->match(rule, fl, flags);
271out:
272 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
273}
274
275int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
276 int flags, struct fib_lookup_arg *arg)
277{
278 struct fib_rule *rule;
279 int err;
280
281 rcu_read_lock();
282
283 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
284jumped:
285 if (!fib_rule_match(rule, ops, fl, flags, arg))
286 continue;
287
288 if (rule->action == FR_ACT_GOTO) {
289 struct fib_rule *target;
290
291 target = rcu_dereference(rule->ctarget);
292 if (target == NULL) {
293 continue;
294 } else {
295 rule = target;
296 goto jumped;
297 }
298 } else if (rule->action == FR_ACT_NOP)
299 continue;
300 else
301 err = ops->action(rule, fl, flags, arg);
302
303 if (!err && ops->suppress && ops->suppress(rule, arg))
304 continue;
305
306 if (err != -EAGAIN) {
307 if ((arg->flags & FIB_LOOKUP_NOREF) ||
308 likely(refcount_inc_not_zero(&rule->refcnt))) {
309 arg->rule = rule;
310 goto out;
311 }
312 break;
313 }
314 }
315
316 err = -ESRCH;
317out:
318 rcu_read_unlock();
319
320 return err;
321}
322EXPORT_SYMBOL_GPL(fib_rules_lookup);
323
324static int call_fib_rule_notifier(struct notifier_block *nb, struct net *net,
325 enum fib_event_type event_type,
326 struct fib_rule *rule, int family)
327{
328 struct fib_rule_notifier_info info = {
329 .info.family = family,
330 .rule = rule,
331 };
332
333 return call_fib_notifier(nb, net, event_type, &info.info);
334}
335
336static int call_fib_rule_notifiers(struct net *net,
337 enum fib_event_type event_type,
338 struct fib_rule *rule,
339 struct fib_rules_ops *ops,
340 struct netlink_ext_ack *extack)
341{
342 struct fib_rule_notifier_info info = {
343 .info.family = ops->family,
344 .info.extack = extack,
345 .rule = rule,
346 };
347
348 ops->fib_rules_seq++;
349 return call_fib_notifiers(net, event_type, &info.info);
350}
351
352/* Called with rcu_read_lock() */
353int fib_rules_dump(struct net *net, struct notifier_block *nb, int family)
354{
355 struct fib_rules_ops *ops;
356 struct fib_rule *rule;
357
358 ops = lookup_rules_ops(net, family);
359 if (!ops)
360 return -EAFNOSUPPORT;
361 list_for_each_entry_rcu(rule, &ops->rules_list, list)
362 call_fib_rule_notifier(nb, net, FIB_EVENT_RULE_ADD, rule,
363 family);
364 rules_ops_put(ops);
365
366 return 0;
367}
368EXPORT_SYMBOL_GPL(fib_rules_dump);
369
370unsigned int fib_rules_seq_read(struct net *net, int family)
371{
372 unsigned int fib_rules_seq;
373 struct fib_rules_ops *ops;
374
375 ASSERT_RTNL();
376
377 ops = lookup_rules_ops(net, family);
378 if (!ops)
379 return 0;
380 fib_rules_seq = ops->fib_rules_seq;
381 rules_ops_put(ops);
382
383 return fib_rules_seq;
384}
385EXPORT_SYMBOL_GPL(fib_rules_seq_read);
386
387static struct fib_rule *rule_find(struct fib_rules_ops *ops,
388 struct fib_rule_hdr *frh,
389 struct nlattr **tb,
390 struct fib_rule *rule,
391 bool user_priority)
392{
393 struct fib_rule *r;
394
395 list_for_each_entry(r, &ops->rules_list, list) {
396 if (rule->action && r->action != rule->action)
397 continue;
398
399 if (rule->table && r->table != rule->table)
400 continue;
401
402 if (user_priority && r->pref != rule->pref)
403 continue;
404
405 if (rule->iifname[0] &&
406 memcmp(r->iifname, rule->iifname, IFNAMSIZ))
407 continue;
408
409 if (rule->oifname[0] &&
410 memcmp(r->oifname, rule->oifname, IFNAMSIZ))
411 continue;
412
413 if (rule->mark && r->mark != rule->mark)
414 continue;
415
416 if (rule->suppress_ifgroup != -1 &&
417 r->suppress_ifgroup != rule->suppress_ifgroup)
418 continue;
419
420 if (rule->suppress_prefixlen != -1 &&
421 r->suppress_prefixlen != rule->suppress_prefixlen)
422 continue;
423
424 if (rule->mark_mask && r->mark_mask != rule->mark_mask)
425 continue;
426
427 if (rule->tun_id && r->tun_id != rule->tun_id)
428 continue;
429
430 if (r->fr_net != rule->fr_net)
431 continue;
432
433 if (rule->l3mdev && r->l3mdev != rule->l3mdev)
434 continue;
435
436 if (uid_range_set(&rule->uid_range) &&
437 (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
438 !uid_eq(r->uid_range.end, rule->uid_range.end)))
439 continue;
440
441 if (rule->ip_proto && r->ip_proto != rule->ip_proto)
442 continue;
443
444 if (rule->proto && r->proto != rule->proto)
445 continue;
446
447 if (fib_rule_port_range_set(&rule->sport_range) &&
448 !fib_rule_port_range_compare(&r->sport_range,
449 &rule->sport_range))
450 continue;
451
452 if (fib_rule_port_range_set(&rule->dport_range) &&
453 !fib_rule_port_range_compare(&r->dport_range,
454 &rule->dport_range))
455 continue;
456
457 if (!ops->compare(r, frh, tb))
458 continue;
459 return r;
460 }
461
462 return NULL;
463}
464
465#ifdef CONFIG_NET_L3_MASTER_DEV
466static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
467 struct netlink_ext_ack *extack)
468{
469 nlrule->l3mdev = nla_get_u8(nla);
470 if (nlrule->l3mdev != 1) {
471 NL_SET_ERR_MSG(extack, "Invalid l3mdev attribute");
472 return -1;
473 }
474
475 return 0;
476}
477#else
478static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
479 struct netlink_ext_ack *extack)
480{
481 NL_SET_ERR_MSG(extack, "l3mdev support is not enabled in kernel");
482 return -1;
483}
484#endif
485
486static int fib_nl2rule(struct sk_buff *skb, struct nlmsghdr *nlh,
487 struct netlink_ext_ack *extack,
488 struct fib_rules_ops *ops,
489 struct nlattr *tb[],
490 struct fib_rule **rule,
491 bool *user_priority)
492{
493 struct net *net = sock_net(skb->sk);
494 struct fib_rule_hdr *frh = nlmsg_data(nlh);
495 struct fib_rule *nlrule = NULL;
496 int err = -EINVAL;
497
498 if (frh->src_len)
499 if (!tb[FRA_SRC] ||
500 frh->src_len > (ops->addr_size * 8) ||
501 nla_len(tb[FRA_SRC]) != ops->addr_size) {
502 NL_SET_ERR_MSG(extack, "Invalid source address");
503 goto errout;
504 }
505
506 if (frh->dst_len)
507 if (!tb[FRA_DST] ||
508 frh->dst_len > (ops->addr_size * 8) ||
509 nla_len(tb[FRA_DST]) != ops->addr_size) {
510 NL_SET_ERR_MSG(extack, "Invalid dst address");
511 goto errout;
512 }
513
514 nlrule = kzalloc(ops->rule_size, GFP_KERNEL);
515 if (!nlrule) {
516 err = -ENOMEM;
517 goto errout;
518 }
519 refcount_set(&nlrule->refcnt, 1);
520 nlrule->fr_net = net;
521
522 if (tb[FRA_PRIORITY]) {
523 nlrule->pref = nla_get_u32(tb[FRA_PRIORITY]);
524 *user_priority = true;
525 } else {
526 nlrule->pref = fib_default_rule_pref(ops);
527 }
528
529 nlrule->proto = tb[FRA_PROTOCOL] ?
530 nla_get_u8(tb[FRA_PROTOCOL]) : RTPROT_UNSPEC;
531
532 if (tb[FRA_IIFNAME]) {
533 struct net_device *dev;
534
535 nlrule->iifindex = -1;
536 nla_strlcpy(nlrule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
537 dev = __dev_get_by_name(net, nlrule->iifname);
538 if (dev)
539 nlrule->iifindex = dev->ifindex;
540 }
541
542 if (tb[FRA_OIFNAME]) {
543 struct net_device *dev;
544
545 nlrule->oifindex = -1;
546 nla_strlcpy(nlrule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
547 dev = __dev_get_by_name(net, nlrule->oifname);
548 if (dev)
549 nlrule->oifindex = dev->ifindex;
550 }
551
552 if (tb[FRA_FWMARK]) {
553 nlrule->mark = nla_get_u32(tb[FRA_FWMARK]);
554 if (nlrule->mark)
555 /* compatibility: if the mark value is non-zero all bits
556 * are compared unless a mask is explicitly specified.
557 */
558 nlrule->mark_mask = 0xFFFFFFFF;
559 }
560
561 if (tb[FRA_FWMASK])
562 nlrule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
563
564 if (tb[FRA_TUN_ID])
565 nlrule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
566
567 err = -EINVAL;
568 if (tb[FRA_L3MDEV] &&
569 fib_nl2rule_l3mdev(tb[FRA_L3MDEV], nlrule, extack) < 0)
570 goto errout_free;
571
572 nlrule->action = frh->action;
573 nlrule->flags = frh->flags;
574 nlrule->table = frh_get_table(frh, tb);
575 if (tb[FRA_SUPPRESS_PREFIXLEN])
576 nlrule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
577 else
578 nlrule->suppress_prefixlen = -1;
579
580 if (tb[FRA_SUPPRESS_IFGROUP])
581 nlrule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
582 else
583 nlrule->suppress_ifgroup = -1;
584
585 if (tb[FRA_GOTO]) {
586 if (nlrule->action != FR_ACT_GOTO) {
587 NL_SET_ERR_MSG(extack, "Unexpected goto");
588 goto errout_free;
589 }
590
591 nlrule->target = nla_get_u32(tb[FRA_GOTO]);
592 /* Backward jumps are prohibited to avoid endless loops */
593 if (nlrule->target <= nlrule->pref) {
594 NL_SET_ERR_MSG(extack, "Backward goto not supported");
595 goto errout_free;
596 }
597 } else if (nlrule->action == FR_ACT_GOTO) {
598 NL_SET_ERR_MSG(extack, "Missing goto target for action goto");
599 goto errout_free;
600 }
601
602 if (nlrule->l3mdev && nlrule->table) {
603 NL_SET_ERR_MSG(extack, "l3mdev and table are mutually exclusive");
604 goto errout_free;
605 }
606
607 if (tb[FRA_UID_RANGE]) {
608 if (current_user_ns() != net->user_ns) {
609 err = -EPERM;
610 NL_SET_ERR_MSG(extack, "No permission to set uid");
611 goto errout_free;
612 }
613
614 nlrule->uid_range = nla_get_kuid_range(tb);
615
616 if (!uid_range_set(&nlrule->uid_range) ||
617 !uid_lte(nlrule->uid_range.start, nlrule->uid_range.end)) {
618 NL_SET_ERR_MSG(extack, "Invalid uid range");
619 goto errout_free;
620 }
621 } else {
622 nlrule->uid_range = fib_kuid_range_unset;
623 }
624
625 if (tb[FRA_IP_PROTO])
626 nlrule->ip_proto = nla_get_u8(tb[FRA_IP_PROTO]);
627
628 if (tb[FRA_SPORT_RANGE]) {
629 err = nla_get_port_range(tb[FRA_SPORT_RANGE],
630 &nlrule->sport_range);
631 if (err) {
632 NL_SET_ERR_MSG(extack, "Invalid sport range");
633 goto errout_free;
634 }
635 }
636
637 if (tb[FRA_DPORT_RANGE]) {
638 err = nla_get_port_range(tb[FRA_DPORT_RANGE],
639 &nlrule->dport_range);
640 if (err) {
641 NL_SET_ERR_MSG(extack, "Invalid dport range");
642 goto errout_free;
643 }
644 }
645
646 *rule = nlrule;
647
648 return 0;
649
650errout_free:
651 kfree(nlrule);
652errout:
653 return err;
654}
655
656static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
657 struct nlattr **tb, struct fib_rule *rule)
658{
659 struct fib_rule *r;
660
661 list_for_each_entry(r, &ops->rules_list, list) {
662 if (r->action != rule->action)
663 continue;
664
665 if (r->table != rule->table)
666 continue;
667
668 if (r->pref != rule->pref)
669 continue;
670
671 if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
672 continue;
673
674 if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
675 continue;
676
677 if (r->mark != rule->mark)
678 continue;
679
680 if (r->suppress_ifgroup != rule->suppress_ifgroup)
681 continue;
682
683 if (r->suppress_prefixlen != rule->suppress_prefixlen)
684 continue;
685
686 if (r->mark_mask != rule->mark_mask)
687 continue;
688
689 if (r->tun_id != rule->tun_id)
690 continue;
691
692 if (r->fr_net != rule->fr_net)
693 continue;
694
695 if (r->l3mdev != rule->l3mdev)
696 continue;
697
698 if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
699 !uid_eq(r->uid_range.end, rule->uid_range.end))
700 continue;
701
702 if (r->ip_proto != rule->ip_proto)
703 continue;
704
705 if (r->proto != rule->proto)
706 continue;
707
708 if (!fib_rule_port_range_compare(&r->sport_range,
709 &rule->sport_range))
710 continue;
711
712 if (!fib_rule_port_range_compare(&r->dport_range,
713 &rule->dport_range))
714 continue;
715
716 if (!ops->compare(r, frh, tb))
717 continue;
718 return 1;
719 }
720 return 0;
721}
722
723int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh,
724 struct netlink_ext_ack *extack)
725{
726 struct net *net = sock_net(skb->sk);
727 struct fib_rule_hdr *frh = nlmsg_data(nlh);
728 struct fib_rules_ops *ops = NULL;
729 struct fib_rule *rule = NULL, *r, *last = NULL;
730 struct nlattr *tb[FRA_MAX + 1];
731 int err = -EINVAL, unresolved = 0;
732 bool user_priority = false;
733
734 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
735 NL_SET_ERR_MSG(extack, "Invalid msg length");
736 goto errout;
737 }
738
739 ops = lookup_rules_ops(net, frh->family);
740 if (!ops) {
741 err = -EAFNOSUPPORT;
742 NL_SET_ERR_MSG(extack, "Rule family not supported");
743 goto errout;
744 }
745
746 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
747 ops->policy, extack);
748 if (err < 0) {
749 NL_SET_ERR_MSG(extack, "Error parsing msg");
750 goto errout;
751 }
752
753 err = fib_nl2rule(skb, nlh, extack, ops, tb, &rule, &user_priority);
754 if (err)
755 goto errout;
756
757 if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
758 rule_exists(ops, frh, tb, rule)) {
759 err = -EEXIST;
760 goto errout_free;
761 }
762
763 err = ops->configure(rule, skb, frh, tb, extack);
764 if (err < 0)
765 goto errout_free;
766
767 err = call_fib_rule_notifiers(net, FIB_EVENT_RULE_ADD, rule, ops,
768 extack);
769 if (err < 0)
770 goto errout_free;
771
772 list_for_each_entry(r, &ops->rules_list, list) {
773 if (r->pref == rule->target) {
774 RCU_INIT_POINTER(rule->ctarget, r);
775 break;
776 }
777 }
778
779 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
780 unresolved = 1;
781
782 list_for_each_entry(r, &ops->rules_list, list) {
783 if (r->pref > rule->pref)
784 break;
785 last = r;
786 }
787
788 if (last)
789 list_add_rcu(&rule->list, &last->list);
790 else
791 list_add_rcu(&rule->list, &ops->rules_list);
792
793 if (ops->unresolved_rules) {
794 /*
795 * There are unresolved goto rules in the list, check if
796 * any of them are pointing to this new rule.
797 */
798 list_for_each_entry(r, &ops->rules_list, list) {
799 if (r->action == FR_ACT_GOTO &&
800 r->target == rule->pref &&
801 rtnl_dereference(r->ctarget) == NULL) {
802 rcu_assign_pointer(r->ctarget, rule);
803 if (--ops->unresolved_rules == 0)
804 break;
805 }
806 }
807 }
808
809 if (rule->action == FR_ACT_GOTO)
810 ops->nr_goto_rules++;
811
812 if (unresolved)
813 ops->unresolved_rules++;
814
815 if (rule->tun_id)
816 ip_tunnel_need_metadata();
817
818 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
819 flush_route_cache(ops);
820 rules_ops_put(ops);
821 return 0;
822
823errout_free:
824 kfree(rule);
825errout:
826 rules_ops_put(ops);
827 return err;
828}
829EXPORT_SYMBOL_GPL(fib_nl_newrule);
830
831int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh,
832 struct netlink_ext_ack *extack)
833{
834 struct net *net = sock_net(skb->sk);
835 struct fib_rule_hdr *frh = nlmsg_data(nlh);
836 struct fib_rules_ops *ops = NULL;
837 struct fib_rule *rule = NULL, *r, *nlrule = NULL;
838 struct nlattr *tb[FRA_MAX+1];
839 int err = -EINVAL;
840 bool user_priority = false;
841
842 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
843 NL_SET_ERR_MSG(extack, "Invalid msg length");
844 goto errout;
845 }
846
847 ops = lookup_rules_ops(net, frh->family);
848 if (ops == NULL) {
849 err = -EAFNOSUPPORT;
850 NL_SET_ERR_MSG(extack, "Rule family not supported");
851 goto errout;
852 }
853
854 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
855 ops->policy, extack);
856 if (err < 0) {
857 NL_SET_ERR_MSG(extack, "Error parsing msg");
858 goto errout;
859 }
860
861 err = fib_nl2rule(skb, nlh, extack, ops, tb, &nlrule, &user_priority);
862 if (err)
863 goto errout;
864
865 rule = rule_find(ops, frh, tb, nlrule, user_priority);
866 if (!rule) {
867 err = -ENOENT;
868 goto errout;
869 }
870
871 if (rule->flags & FIB_RULE_PERMANENT) {
872 err = -EPERM;
873 goto errout;
874 }
875
876 if (ops->delete) {
877 err = ops->delete(rule);
878 if (err)
879 goto errout;
880 }
881
882 if (rule->tun_id)
883 ip_tunnel_unneed_metadata();
884
885 list_del_rcu(&rule->list);
886
887 if (rule->action == FR_ACT_GOTO) {
888 ops->nr_goto_rules--;
889 if (rtnl_dereference(rule->ctarget) == NULL)
890 ops->unresolved_rules--;
891 }
892
893 /*
894 * Check if this rule is a target to any of them. If so,
895 * adjust to the next one with the same preference or
896 * disable them. As this operation is eventually very
897 * expensive, it is only performed if goto rules, except
898 * current if it is goto rule, have actually been added.
899 */
900 if (ops->nr_goto_rules > 0) {
901 struct fib_rule *n;
902
903 n = list_next_entry(rule, list);
904 if (&n->list == &ops->rules_list || n->pref != rule->pref)
905 n = NULL;
906 list_for_each_entry(r, &ops->rules_list, list) {
907 if (rtnl_dereference(r->ctarget) != rule)
908 continue;
909 rcu_assign_pointer(r->ctarget, n);
910 if (!n)
911 ops->unresolved_rules++;
912 }
913 }
914
915 call_fib_rule_notifiers(net, FIB_EVENT_RULE_DEL, rule, ops,
916 NULL);
917 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
918 NETLINK_CB(skb).portid);
919 fib_rule_put(rule);
920 flush_route_cache(ops);
921 rules_ops_put(ops);
922 kfree(nlrule);
923 return 0;
924
925errout:
926 kfree(nlrule);
927 rules_ops_put(ops);
928 return err;
929}
930EXPORT_SYMBOL_GPL(fib_nl_delrule);
931
932static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
933 struct fib_rule *rule)
934{
935 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
936 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
937 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
938 + nla_total_size(4) /* FRA_PRIORITY */
939 + nla_total_size(4) /* FRA_TABLE */
940 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
941 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
942 + nla_total_size(4) /* FRA_FWMARK */
943 + nla_total_size(4) /* FRA_FWMASK */
944 + nla_total_size_64bit(8) /* FRA_TUN_ID */
945 + nla_total_size(sizeof(struct fib_kuid_range))
946 + nla_total_size(1) /* FRA_PROTOCOL */
947 + nla_total_size(1) /* FRA_IP_PROTO */
948 + nla_total_size(sizeof(struct fib_rule_port_range)) /* FRA_SPORT_RANGE */
949 + nla_total_size(sizeof(struct fib_rule_port_range)); /* FRA_DPORT_RANGE */
950
951 if (ops->nlmsg_payload)
952 payload += ops->nlmsg_payload(rule);
953
954 return payload;
955}
956
957static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
958 u32 pid, u32 seq, int type, int flags,
959 struct fib_rules_ops *ops)
960{
961 struct nlmsghdr *nlh;
962 struct fib_rule_hdr *frh;
963
964 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
965 if (nlh == NULL)
966 return -EMSGSIZE;
967
968 frh = nlmsg_data(nlh);
969 frh->family = ops->family;
970 frh->table = rule->table;
971 if (nla_put_u32(skb, FRA_TABLE, rule->table))
972 goto nla_put_failure;
973 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
974 goto nla_put_failure;
975 frh->res1 = 0;
976 frh->res2 = 0;
977 frh->action = rule->action;
978 frh->flags = rule->flags;
979
980 if (nla_put_u8(skb, FRA_PROTOCOL, rule->proto))
981 goto nla_put_failure;
982
983 if (rule->action == FR_ACT_GOTO &&
984 rcu_access_pointer(rule->ctarget) == NULL)
985 frh->flags |= FIB_RULE_UNRESOLVED;
986
987 if (rule->iifname[0]) {
988 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
989 goto nla_put_failure;
990 if (rule->iifindex == -1)
991 frh->flags |= FIB_RULE_IIF_DETACHED;
992 }
993
994 if (rule->oifname[0]) {
995 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
996 goto nla_put_failure;
997 if (rule->oifindex == -1)
998 frh->flags |= FIB_RULE_OIF_DETACHED;
999 }
1000
1001 if ((rule->pref &&
1002 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
1003 (rule->mark &&
1004 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
1005 ((rule->mark_mask || rule->mark) &&
1006 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
1007 (rule->target &&
1008 nla_put_u32(skb, FRA_GOTO, rule->target)) ||
1009 (rule->tun_id &&
1010 nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
1011 (rule->l3mdev &&
1012 nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
1013 (uid_range_set(&rule->uid_range) &&
1014 nla_put_uid_range(skb, &rule->uid_range)) ||
1015 (fib_rule_port_range_set(&rule->sport_range) &&
1016 nla_put_port_range(skb, FRA_SPORT_RANGE, &rule->sport_range)) ||
1017 (fib_rule_port_range_set(&rule->dport_range) &&
1018 nla_put_port_range(skb, FRA_DPORT_RANGE, &rule->dport_range)) ||
1019 (rule->ip_proto && nla_put_u8(skb, FRA_IP_PROTO, rule->ip_proto)))
1020 goto nla_put_failure;
1021
1022 if (rule->suppress_ifgroup != -1) {
1023 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
1024 goto nla_put_failure;
1025 }
1026
1027 if (ops->fill(rule, skb, frh) < 0)
1028 goto nla_put_failure;
1029
1030 nlmsg_end(skb, nlh);
1031 return 0;
1032
1033nla_put_failure:
1034 nlmsg_cancel(skb, nlh);
1035 return -EMSGSIZE;
1036}
1037
1038static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
1039 struct fib_rules_ops *ops)
1040{
1041 int idx = 0;
1042 struct fib_rule *rule;
1043 int err = 0;
1044
1045 rcu_read_lock();
1046 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
1047 if (idx < cb->args[1])
1048 goto skip;
1049
1050 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
1051 cb->nlh->nlmsg_seq, RTM_NEWRULE,
1052 NLM_F_MULTI, ops);
1053 if (err)
1054 break;
1055skip:
1056 idx++;
1057 }
1058 rcu_read_unlock();
1059 cb->args[1] = idx;
1060 rules_ops_put(ops);
1061
1062 return err;
1063}
1064
1065static int fib_valid_dumprule_req(const struct nlmsghdr *nlh,
1066 struct netlink_ext_ack *extack)
1067{
1068 struct fib_rule_hdr *frh;
1069
1070 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
1071 NL_SET_ERR_MSG(extack, "Invalid header for fib rule dump request");
1072 return -EINVAL;
1073 }
1074
1075 frh = nlmsg_data(nlh);
1076 if (frh->dst_len || frh->src_len || frh->tos || frh->table ||
1077 frh->res1 || frh->res2 || frh->action || frh->flags) {
1078 NL_SET_ERR_MSG(extack,
1079 "Invalid values in header for fib rule dump request");
1080 return -EINVAL;
1081 }
1082
1083 if (nlmsg_attrlen(nlh, sizeof(*frh))) {
1084 NL_SET_ERR_MSG(extack, "Invalid data after header in fib rule dump request");
1085 return -EINVAL;
1086 }
1087
1088 return 0;
1089}
1090
1091static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
1092{
1093 const struct nlmsghdr *nlh = cb->nlh;
1094 struct net *net = sock_net(skb->sk);
1095 struct fib_rules_ops *ops;
1096 int idx = 0, family;
1097
1098 if (cb->strict_check) {
1099 int err = fib_valid_dumprule_req(nlh, cb->extack);
1100
1101 if (err < 0)
1102 return err;
1103 }
1104
1105 family = rtnl_msg_family(nlh);
1106 if (family != AF_UNSPEC) {
1107 /* Protocol specific dump request */
1108 ops = lookup_rules_ops(net, family);
1109 if (ops == NULL)
1110 return -EAFNOSUPPORT;
1111
1112 dump_rules(skb, cb, ops);
1113
1114 return skb->len;
1115 }
1116
1117 rcu_read_lock();
1118 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
1119 if (idx < cb->args[0] || !try_module_get(ops->owner))
1120 goto skip;
1121
1122 if (dump_rules(skb, cb, ops) < 0)
1123 break;
1124
1125 cb->args[1] = 0;
1126skip:
1127 idx++;
1128 }
1129 rcu_read_unlock();
1130 cb->args[0] = idx;
1131
1132 return skb->len;
1133}
1134
1135static void notify_rule_change(int event, struct fib_rule *rule,
1136 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
1137 u32 pid)
1138{
1139 struct net *net;
1140 struct sk_buff *skb;
1141 int err = -ENOBUFS;
1142
1143 net = ops->fro_net;
1144 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
1145 if (skb == NULL)
1146 goto errout;
1147
1148 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
1149 if (err < 0) {
1150 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
1151 WARN_ON(err == -EMSGSIZE);
1152 kfree_skb(skb);
1153 goto errout;
1154 }
1155
1156 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
1157 return;
1158errout:
1159 if (err < 0)
1160 rtnl_set_sk_err(net, ops->nlgroup, err);
1161}
1162
1163static void attach_rules(struct list_head *rules, struct net_device *dev)
1164{
1165 struct fib_rule *rule;
1166
1167 list_for_each_entry(rule, rules, list) {
1168 if (rule->iifindex == -1 &&
1169 strcmp(dev->name, rule->iifname) == 0)
1170 rule->iifindex = dev->ifindex;
1171 if (rule->oifindex == -1 &&
1172 strcmp(dev->name, rule->oifname) == 0)
1173 rule->oifindex = dev->ifindex;
1174 }
1175}
1176
1177static void detach_rules(struct list_head *rules, struct net_device *dev)
1178{
1179 struct fib_rule *rule;
1180
1181 list_for_each_entry(rule, rules, list) {
1182 if (rule->iifindex == dev->ifindex)
1183 rule->iifindex = -1;
1184 if (rule->oifindex == dev->ifindex)
1185 rule->oifindex = -1;
1186 }
1187}
1188
1189
1190static int fib_rules_event(struct notifier_block *this, unsigned long event,
1191 void *ptr)
1192{
1193 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1194 struct net *net = dev_net(dev);
1195 struct fib_rules_ops *ops;
1196
1197 ASSERT_RTNL();
1198
1199 switch (event) {
1200 case NETDEV_REGISTER:
1201 list_for_each_entry(ops, &net->rules_ops, list)
1202 attach_rules(&ops->rules_list, dev);
1203 break;
1204
1205 case NETDEV_CHANGENAME:
1206 list_for_each_entry(ops, &net->rules_ops, list) {
1207 detach_rules(&ops->rules_list, dev);
1208 attach_rules(&ops->rules_list, dev);
1209 }
1210 break;
1211
1212 case NETDEV_UNREGISTER:
1213 list_for_each_entry(ops, &net->rules_ops, list)
1214 detach_rules(&ops->rules_list, dev);
1215 break;
1216 }
1217
1218 return NOTIFY_DONE;
1219}
1220
1221static struct notifier_block fib_rules_notifier = {
1222 .notifier_call = fib_rules_event,
1223};
1224
1225static int __net_init fib_rules_net_init(struct net *net)
1226{
1227 INIT_LIST_HEAD(&net->rules_ops);
1228 spin_lock_init(&net->rules_mod_lock);
1229 return 0;
1230}
1231
1232static void __net_exit fib_rules_net_exit(struct net *net)
1233{
1234 WARN_ON_ONCE(!list_empty(&net->rules_ops));
1235}
1236
1237static struct pernet_operations fib_rules_net_ops = {
1238 .init = fib_rules_net_init,
1239 .exit = fib_rules_net_exit,
1240};
1241
1242static int __init fib_rules_init(void)
1243{
1244 int err;
1245 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, 0);
1246 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, 0);
1247 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, 0);
1248
1249 err = register_pernet_subsys(&fib_rules_net_ops);
1250 if (err < 0)
1251 goto fail;
1252
1253 err = register_netdevice_notifier(&fib_rules_notifier);
1254 if (err < 0)
1255 goto fail_unregister;
1256
1257 return 0;
1258
1259fail_unregister:
1260 unregister_pernet_subsys(&fib_rules_net_ops);
1261fail:
1262 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
1263 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
1264 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
1265 return err;
1266}
1267
1268subsys_initcall(fib_rules_init);