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/inet_dscp.h>
15#include <net/sock.h>
16#include <net/fib_rules.h>
17#include <net/ip_tunnels.h>
18#include <linux/indirect_call_wrapper.h>
19
20#if defined(CONFIG_IPV6) && defined(CONFIG_IPV6_MULTIPLE_TABLES)
21#ifdef CONFIG_IP_MULTIPLE_TABLES
22#define INDIRECT_CALL_MT(f, f2, f1, ...) \
23 INDIRECT_CALL_INET(f, f2, f1, __VA_ARGS__)
24#else
25#define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
26#endif
27#elif defined(CONFIG_IP_MULTIPLE_TABLES)
28#define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f1, __VA_ARGS__)
29#else
30#define INDIRECT_CALL_MT(f, f2, f1, ...) f(__VA_ARGS__)
31#endif
32
33static const struct fib_kuid_range fib_kuid_range_unset = {
34 KUIDT_INIT(0),
35 KUIDT_INIT(~0),
36};
37
38bool fib_rule_matchall(const struct fib_rule *rule)
39{
40 if (READ_ONCE(rule->iifindex) || READ_ONCE(rule->oifindex) ||
41 rule->mark || rule->tun_id || rule->flags)
42 return false;
43 if (rule->suppress_ifgroup != -1 || rule->suppress_prefixlen != -1)
44 return false;
45 if (!uid_eq(rule->uid_range.start, fib_kuid_range_unset.start) ||
46 !uid_eq(rule->uid_range.end, fib_kuid_range_unset.end))
47 return false;
48 if (fib_rule_port_range_set(&rule->sport_range))
49 return false;
50 if (fib_rule_port_range_set(&rule->dport_range))
51 return false;
52 return true;
53}
54EXPORT_SYMBOL_GPL(fib_rule_matchall);
55
56int fib_default_rule_add(struct fib_rules_ops *ops,
57 u32 pref, u32 table)
58{
59 struct fib_rule *r;
60
61 r = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT);
62 if (r == NULL)
63 return -ENOMEM;
64
65 refcount_set(&r->refcnt, 1);
66 r->action = FR_ACT_TO_TBL;
67 r->pref = pref;
68 r->table = table;
69 r->proto = RTPROT_KERNEL;
70 r->fr_net = ops->fro_net;
71 r->uid_range = fib_kuid_range_unset;
72
73 r->suppress_prefixlen = -1;
74 r->suppress_ifgroup = -1;
75
76 /* The lock is not required here, the list in unreachable
77 * at the moment this function is called */
78 list_add_tail(&r->list, &ops->rules_list);
79 return 0;
80}
81EXPORT_SYMBOL(fib_default_rule_add);
82
83static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
84{
85 struct list_head *pos;
86 struct fib_rule *rule;
87
88 if (!list_empty(&ops->rules_list)) {
89 pos = ops->rules_list.next;
90 if (pos->next != &ops->rules_list) {
91 rule = list_entry(pos->next, struct fib_rule, list);
92 if (rule->pref)
93 return rule->pref - 1;
94 }
95 }
96
97 return 0;
98}
99
100static void notify_rule_change(int event, struct fib_rule *rule,
101 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
102 u32 pid);
103
104static struct fib_rules_ops *lookup_rules_ops(const struct net *net,
105 int family)
106{
107 struct fib_rules_ops *ops;
108
109 rcu_read_lock();
110 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
111 if (ops->family == family) {
112 if (!try_module_get(ops->owner))
113 ops = NULL;
114 rcu_read_unlock();
115 return ops;
116 }
117 }
118 rcu_read_unlock();
119
120 return NULL;
121}
122
123static void rules_ops_put(struct fib_rules_ops *ops)
124{
125 if (ops)
126 module_put(ops->owner);
127}
128
129static void flush_route_cache(struct fib_rules_ops *ops)
130{
131 if (ops->flush_cache)
132 ops->flush_cache(ops);
133}
134
135static int __fib_rules_register(struct fib_rules_ops *ops)
136{
137 int err = -EEXIST;
138 struct fib_rules_ops *o;
139 struct net *net;
140
141 net = ops->fro_net;
142
143 if (ops->rule_size < sizeof(struct fib_rule))
144 return -EINVAL;
145
146 if (ops->match == NULL || ops->configure == NULL ||
147 ops->compare == NULL || ops->fill == NULL ||
148 ops->action == NULL)
149 return -EINVAL;
150
151 spin_lock(&net->rules_mod_lock);
152 list_for_each_entry(o, &net->rules_ops, list)
153 if (ops->family == o->family)
154 goto errout;
155
156 list_add_tail_rcu(&ops->list, &net->rules_ops);
157 err = 0;
158errout:
159 spin_unlock(&net->rules_mod_lock);
160
161 return err;
162}
163
164struct fib_rules_ops *
165fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
166{
167 struct fib_rules_ops *ops;
168 int err;
169
170 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
171 if (ops == NULL)
172 return ERR_PTR(-ENOMEM);
173
174 INIT_LIST_HEAD(&ops->rules_list);
175 ops->fro_net = net;
176
177 err = __fib_rules_register(ops);
178 if (err) {
179 kfree(ops);
180 ops = ERR_PTR(err);
181 }
182
183 return ops;
184}
185EXPORT_SYMBOL_GPL(fib_rules_register);
186
187static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
188{
189 struct fib_rule *rule, *tmp;
190
191 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
192 list_del_rcu(&rule->list);
193 if (ops->delete)
194 ops->delete(rule);
195 fib_rule_put(rule);
196 }
197}
198
199void fib_rules_unregister(struct fib_rules_ops *ops)
200{
201 struct net *net = ops->fro_net;
202
203 spin_lock(&net->rules_mod_lock);
204 list_del_rcu(&ops->list);
205 spin_unlock(&net->rules_mod_lock);
206
207 fib_rules_cleanup_ops(ops);
208 kfree_rcu(ops, rcu);
209}
210EXPORT_SYMBOL_GPL(fib_rules_unregister);
211
212static int uid_range_set(struct fib_kuid_range *range)
213{
214 return uid_valid(range->start) && uid_valid(range->end);
215}
216
217static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
218{
219 struct fib_rule_uid_range *in;
220 struct fib_kuid_range out;
221
222 in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
223
224 out.start = make_kuid(current_user_ns(), in->start);
225 out.end = make_kuid(current_user_ns(), in->end);
226
227 return out;
228}
229
230static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
231{
232 struct fib_rule_uid_range out = {
233 from_kuid_munged(current_user_ns(), range->start),
234 from_kuid_munged(current_user_ns(), range->end)
235 };
236
237 return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
238}
239
240static int nla_get_port_range(struct nlattr *pattr,
241 struct fib_rule_port_range *port_range)
242{
243 const struct fib_rule_port_range *pr = nla_data(pattr);
244
245 if (!fib_rule_port_range_valid(pr))
246 return -EINVAL;
247
248 port_range->start = pr->start;
249 port_range->end = pr->end;
250
251 return 0;
252}
253
254static int nla_put_port_range(struct sk_buff *skb, int attrtype,
255 struct fib_rule_port_range *range)
256{
257 return nla_put(skb, attrtype, sizeof(*range), range);
258}
259
260static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
261 struct flowi *fl, int flags,
262 struct fib_lookup_arg *arg)
263{
264 int iifindex, oifindex, ret = 0;
265
266 iifindex = READ_ONCE(rule->iifindex);
267 if (iifindex && (iifindex != fl->flowi_iif))
268 goto out;
269
270 oifindex = READ_ONCE(rule->oifindex);
271 if (oifindex && (oifindex != fl->flowi_oif))
272 goto out;
273
274 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
275 goto out;
276
277 if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
278 goto out;
279
280 if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
281 goto out;
282
283 if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
284 uid_gt(fl->flowi_uid, rule->uid_range.end))
285 goto out;
286
287 ret = INDIRECT_CALL_MT(ops->match,
288 fib6_rule_match,
289 fib4_rule_match,
290 rule, fl, flags);
291out:
292 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
293}
294
295int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
296 int flags, struct fib_lookup_arg *arg)
297{
298 struct fib_rule *rule;
299 int err;
300
301 rcu_read_lock();
302
303 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
304jumped:
305 if (!fib_rule_match(rule, ops, fl, flags, arg))
306 continue;
307
308 if (rule->action == FR_ACT_GOTO) {
309 struct fib_rule *target;
310
311 target = rcu_dereference(rule->ctarget);
312 if (target == NULL) {
313 continue;
314 } else {
315 rule = target;
316 goto jumped;
317 }
318 } else if (rule->action == FR_ACT_NOP)
319 continue;
320 else
321 err = INDIRECT_CALL_MT(ops->action,
322 fib6_rule_action,
323 fib4_rule_action,
324 rule, fl, flags, arg);
325
326 if (!err && ops->suppress && INDIRECT_CALL_MT(ops->suppress,
327 fib6_rule_suppress,
328 fib4_rule_suppress,
329 rule, flags, arg))
330 continue;
331
332 if (err != -EAGAIN) {
333 if ((arg->flags & FIB_LOOKUP_NOREF) ||
334 likely(refcount_inc_not_zero(&rule->refcnt))) {
335 arg->rule = rule;
336 goto out;
337 }
338 break;
339 }
340 }
341
342 err = -ESRCH;
343out:
344 rcu_read_unlock();
345
346 return err;
347}
348EXPORT_SYMBOL_GPL(fib_rules_lookup);
349
350static int call_fib_rule_notifier(struct notifier_block *nb,
351 enum fib_event_type event_type,
352 struct fib_rule *rule, int family,
353 struct netlink_ext_ack *extack)
354{
355 struct fib_rule_notifier_info info = {
356 .info.family = family,
357 .info.extack = extack,
358 .rule = rule,
359 };
360
361 return call_fib_notifier(nb, event_type, &info.info);
362}
363
364static int call_fib_rule_notifiers(struct net *net,
365 enum fib_event_type event_type,
366 struct fib_rule *rule,
367 struct fib_rules_ops *ops,
368 struct netlink_ext_ack *extack)
369{
370 struct fib_rule_notifier_info info = {
371 .info.family = ops->family,
372 .info.extack = extack,
373 .rule = rule,
374 };
375
376 ASSERT_RTNL();
377 /* Paired with READ_ONCE() in fib_rules_seq() */
378 WRITE_ONCE(ops->fib_rules_seq, ops->fib_rules_seq + 1);
379 return call_fib_notifiers(net, event_type, &info.info);
380}
381
382/* Called with rcu_read_lock() */
383int fib_rules_dump(struct net *net, struct notifier_block *nb, int family,
384 struct netlink_ext_ack *extack)
385{
386 struct fib_rules_ops *ops;
387 struct fib_rule *rule;
388 int err = 0;
389
390 ops = lookup_rules_ops(net, family);
391 if (!ops)
392 return -EAFNOSUPPORT;
393 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
394 err = call_fib_rule_notifier(nb, FIB_EVENT_RULE_ADD,
395 rule, family, extack);
396 if (err)
397 break;
398 }
399 rules_ops_put(ops);
400
401 return err;
402}
403EXPORT_SYMBOL_GPL(fib_rules_dump);
404
405unsigned int fib_rules_seq_read(const struct net *net, int family)
406{
407 unsigned int fib_rules_seq;
408 struct fib_rules_ops *ops;
409
410 ops = lookup_rules_ops(net, family);
411 if (!ops)
412 return 0;
413 /* Paired with WRITE_ONCE() in call_fib_rule_notifiers() */
414 fib_rules_seq = READ_ONCE(ops->fib_rules_seq);
415 rules_ops_put(ops);
416
417 return fib_rules_seq;
418}
419EXPORT_SYMBOL_GPL(fib_rules_seq_read);
420
421static struct fib_rule *rule_find(struct fib_rules_ops *ops,
422 struct fib_rule_hdr *frh,
423 struct nlattr **tb,
424 struct fib_rule *rule,
425 bool user_priority)
426{
427 struct fib_rule *r;
428
429 list_for_each_entry(r, &ops->rules_list, list) {
430 if (rule->action && r->action != rule->action)
431 continue;
432
433 if (rule->table && r->table != rule->table)
434 continue;
435
436 if (user_priority && r->pref != rule->pref)
437 continue;
438
439 if (rule->iifname[0] &&
440 memcmp(r->iifname, rule->iifname, IFNAMSIZ))
441 continue;
442
443 if (rule->oifname[0] &&
444 memcmp(r->oifname, rule->oifname, IFNAMSIZ))
445 continue;
446
447 if (rule->mark && r->mark != rule->mark)
448 continue;
449
450 if (rule->suppress_ifgroup != -1 &&
451 r->suppress_ifgroup != rule->suppress_ifgroup)
452 continue;
453
454 if (rule->suppress_prefixlen != -1 &&
455 r->suppress_prefixlen != rule->suppress_prefixlen)
456 continue;
457
458 if (rule->mark_mask && r->mark_mask != rule->mark_mask)
459 continue;
460
461 if (rule->tun_id && r->tun_id != rule->tun_id)
462 continue;
463
464 if (r->fr_net != rule->fr_net)
465 continue;
466
467 if (rule->l3mdev && r->l3mdev != rule->l3mdev)
468 continue;
469
470 if (uid_range_set(&rule->uid_range) &&
471 (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
472 !uid_eq(r->uid_range.end, rule->uid_range.end)))
473 continue;
474
475 if (rule->ip_proto && r->ip_proto != rule->ip_proto)
476 continue;
477
478 if (rule->proto && r->proto != rule->proto)
479 continue;
480
481 if (fib_rule_port_range_set(&rule->sport_range) &&
482 !fib_rule_port_range_compare(&r->sport_range,
483 &rule->sport_range))
484 continue;
485
486 if (fib_rule_port_range_set(&rule->dport_range) &&
487 !fib_rule_port_range_compare(&r->dport_range,
488 &rule->dport_range))
489 continue;
490
491 if (!ops->compare(r, frh, tb))
492 continue;
493 return r;
494 }
495
496 return NULL;
497}
498
499#ifdef CONFIG_NET_L3_MASTER_DEV
500static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
501 struct netlink_ext_ack *extack)
502{
503 nlrule->l3mdev = nla_get_u8(nla);
504 if (nlrule->l3mdev != 1) {
505 NL_SET_ERR_MSG(extack, "Invalid l3mdev attribute");
506 return -1;
507 }
508
509 return 0;
510}
511#else
512static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
513 struct netlink_ext_ack *extack)
514{
515 NL_SET_ERR_MSG(extack, "l3mdev support is not enabled in kernel");
516 return -1;
517}
518#endif
519
520static int fib_nl2rule(struct sk_buff *skb, struct nlmsghdr *nlh,
521 struct netlink_ext_ack *extack,
522 struct fib_rules_ops *ops,
523 struct nlattr *tb[],
524 struct fib_rule **rule,
525 bool *user_priority)
526{
527 struct net *net = sock_net(skb->sk);
528 struct fib_rule_hdr *frh = nlmsg_data(nlh);
529 struct fib_rule *nlrule = NULL;
530 int err = -EINVAL;
531
532 if (frh->src_len)
533 if (!tb[FRA_SRC] ||
534 frh->src_len > (ops->addr_size * 8) ||
535 nla_len(tb[FRA_SRC]) != ops->addr_size) {
536 NL_SET_ERR_MSG(extack, "Invalid source address");
537 goto errout;
538 }
539
540 if (frh->dst_len)
541 if (!tb[FRA_DST] ||
542 frh->dst_len > (ops->addr_size * 8) ||
543 nla_len(tb[FRA_DST]) != ops->addr_size) {
544 NL_SET_ERR_MSG(extack, "Invalid dst address");
545 goto errout;
546 }
547
548 nlrule = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT);
549 if (!nlrule) {
550 err = -ENOMEM;
551 goto errout;
552 }
553 refcount_set(&nlrule->refcnt, 1);
554 nlrule->fr_net = net;
555
556 if (tb[FRA_PRIORITY]) {
557 nlrule->pref = nla_get_u32(tb[FRA_PRIORITY]);
558 *user_priority = true;
559 } else {
560 nlrule->pref = fib_default_rule_pref(ops);
561 }
562
563 nlrule->proto = nla_get_u8_default(tb[FRA_PROTOCOL], RTPROT_UNSPEC);
564
565 if (tb[FRA_IIFNAME]) {
566 struct net_device *dev;
567
568 nlrule->iifindex = -1;
569 nla_strscpy(nlrule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
570 dev = __dev_get_by_name(net, nlrule->iifname);
571 if (dev)
572 nlrule->iifindex = dev->ifindex;
573 }
574
575 if (tb[FRA_OIFNAME]) {
576 struct net_device *dev;
577
578 nlrule->oifindex = -1;
579 nla_strscpy(nlrule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
580 dev = __dev_get_by_name(net, nlrule->oifname);
581 if (dev)
582 nlrule->oifindex = dev->ifindex;
583 }
584
585 if (tb[FRA_FWMARK]) {
586 nlrule->mark = nla_get_u32(tb[FRA_FWMARK]);
587 if (nlrule->mark)
588 /* compatibility: if the mark value is non-zero all bits
589 * are compared unless a mask is explicitly specified.
590 */
591 nlrule->mark_mask = 0xFFFFFFFF;
592 }
593
594 if (tb[FRA_FWMASK])
595 nlrule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
596
597 if (tb[FRA_TUN_ID])
598 nlrule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
599
600 if (tb[FRA_L3MDEV] &&
601 fib_nl2rule_l3mdev(tb[FRA_L3MDEV], nlrule, extack) < 0)
602 goto errout_free;
603
604 nlrule->action = frh->action;
605 nlrule->flags = frh->flags;
606 nlrule->table = frh_get_table(frh, tb);
607 if (tb[FRA_SUPPRESS_PREFIXLEN])
608 nlrule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
609 else
610 nlrule->suppress_prefixlen = -1;
611
612 if (tb[FRA_SUPPRESS_IFGROUP])
613 nlrule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
614 else
615 nlrule->suppress_ifgroup = -1;
616
617 if (tb[FRA_GOTO]) {
618 if (nlrule->action != FR_ACT_GOTO) {
619 NL_SET_ERR_MSG(extack, "Unexpected goto");
620 goto errout_free;
621 }
622
623 nlrule->target = nla_get_u32(tb[FRA_GOTO]);
624 /* Backward jumps are prohibited to avoid endless loops */
625 if (nlrule->target <= nlrule->pref) {
626 NL_SET_ERR_MSG(extack, "Backward goto not supported");
627 goto errout_free;
628 }
629 } else if (nlrule->action == FR_ACT_GOTO) {
630 NL_SET_ERR_MSG(extack, "Missing goto target for action goto");
631 goto errout_free;
632 }
633
634 if (nlrule->l3mdev && nlrule->table) {
635 NL_SET_ERR_MSG(extack, "l3mdev and table are mutually exclusive");
636 goto errout_free;
637 }
638
639 if (tb[FRA_UID_RANGE]) {
640 if (current_user_ns() != net->user_ns) {
641 err = -EPERM;
642 NL_SET_ERR_MSG(extack, "No permission to set uid");
643 goto errout_free;
644 }
645
646 nlrule->uid_range = nla_get_kuid_range(tb);
647
648 if (!uid_range_set(&nlrule->uid_range) ||
649 !uid_lte(nlrule->uid_range.start, nlrule->uid_range.end)) {
650 NL_SET_ERR_MSG(extack, "Invalid uid range");
651 goto errout_free;
652 }
653 } else {
654 nlrule->uid_range = fib_kuid_range_unset;
655 }
656
657 if (tb[FRA_IP_PROTO])
658 nlrule->ip_proto = nla_get_u8(tb[FRA_IP_PROTO]);
659
660 if (tb[FRA_SPORT_RANGE]) {
661 err = nla_get_port_range(tb[FRA_SPORT_RANGE],
662 &nlrule->sport_range);
663 if (err) {
664 NL_SET_ERR_MSG(extack, "Invalid sport range");
665 goto errout_free;
666 }
667 }
668
669 if (tb[FRA_DPORT_RANGE]) {
670 err = nla_get_port_range(tb[FRA_DPORT_RANGE],
671 &nlrule->dport_range);
672 if (err) {
673 NL_SET_ERR_MSG(extack, "Invalid dport range");
674 goto errout_free;
675 }
676 }
677
678 *rule = nlrule;
679
680 return 0;
681
682errout_free:
683 kfree(nlrule);
684errout:
685 return err;
686}
687
688static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
689 struct nlattr **tb, struct fib_rule *rule)
690{
691 struct fib_rule *r;
692
693 list_for_each_entry(r, &ops->rules_list, list) {
694 if (r->action != rule->action)
695 continue;
696
697 if (r->table != rule->table)
698 continue;
699
700 if (r->pref != rule->pref)
701 continue;
702
703 if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
704 continue;
705
706 if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
707 continue;
708
709 if (r->mark != rule->mark)
710 continue;
711
712 if (r->suppress_ifgroup != rule->suppress_ifgroup)
713 continue;
714
715 if (r->suppress_prefixlen != rule->suppress_prefixlen)
716 continue;
717
718 if (r->mark_mask != rule->mark_mask)
719 continue;
720
721 if (r->tun_id != rule->tun_id)
722 continue;
723
724 if (r->fr_net != rule->fr_net)
725 continue;
726
727 if (r->l3mdev != rule->l3mdev)
728 continue;
729
730 if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
731 !uid_eq(r->uid_range.end, rule->uid_range.end))
732 continue;
733
734 if (r->ip_proto != rule->ip_proto)
735 continue;
736
737 if (r->proto != rule->proto)
738 continue;
739
740 if (!fib_rule_port_range_compare(&r->sport_range,
741 &rule->sport_range))
742 continue;
743
744 if (!fib_rule_port_range_compare(&r->dport_range,
745 &rule->dport_range))
746 continue;
747
748 if (!ops->compare(r, frh, tb))
749 continue;
750 return 1;
751 }
752 return 0;
753}
754
755static const struct nla_policy fib_rule_policy[FRA_MAX + 1] = {
756 [FRA_UNSPEC] = { .strict_start_type = FRA_DPORT_RANGE + 1 },
757 [FRA_IIFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ - 1 },
758 [FRA_OIFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ - 1 },
759 [FRA_PRIORITY] = { .type = NLA_U32 },
760 [FRA_FWMARK] = { .type = NLA_U32 },
761 [FRA_FLOW] = { .type = NLA_U32 },
762 [FRA_TUN_ID] = { .type = NLA_U64 },
763 [FRA_FWMASK] = { .type = NLA_U32 },
764 [FRA_TABLE] = { .type = NLA_U32 },
765 [FRA_SUPPRESS_PREFIXLEN] = { .type = NLA_U32 },
766 [FRA_SUPPRESS_IFGROUP] = { .type = NLA_U32 },
767 [FRA_GOTO] = { .type = NLA_U32 },
768 [FRA_L3MDEV] = { .type = NLA_U8 },
769 [FRA_UID_RANGE] = { .len = sizeof(struct fib_rule_uid_range) },
770 [FRA_PROTOCOL] = { .type = NLA_U8 },
771 [FRA_IP_PROTO] = { .type = NLA_U8 },
772 [FRA_SPORT_RANGE] = { .len = sizeof(struct fib_rule_port_range) },
773 [FRA_DPORT_RANGE] = { .len = sizeof(struct fib_rule_port_range) },
774 [FRA_DSCP] = NLA_POLICY_MAX(NLA_U8, INET_DSCP_MASK >> 2),
775};
776
777int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh,
778 struct netlink_ext_ack *extack)
779{
780 struct net *net = sock_net(skb->sk);
781 struct fib_rule_hdr *frh = nlmsg_data(nlh);
782 struct fib_rules_ops *ops = NULL;
783 struct fib_rule *rule = NULL, *r, *last = NULL;
784 struct nlattr *tb[FRA_MAX + 1];
785 int err = -EINVAL, unresolved = 0;
786 bool user_priority = false;
787
788 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
789 NL_SET_ERR_MSG(extack, "Invalid msg length");
790 goto errout;
791 }
792
793 ops = lookup_rules_ops(net, frh->family);
794 if (!ops) {
795 err = -EAFNOSUPPORT;
796 NL_SET_ERR_MSG(extack, "Rule family not supported");
797 goto errout;
798 }
799
800 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
801 fib_rule_policy, extack);
802 if (err < 0) {
803 NL_SET_ERR_MSG(extack, "Error parsing msg");
804 goto errout;
805 }
806
807 err = fib_nl2rule(skb, nlh, extack, ops, tb, &rule, &user_priority);
808 if (err)
809 goto errout;
810
811 if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
812 rule_exists(ops, frh, tb, rule)) {
813 err = -EEXIST;
814 goto errout_free;
815 }
816
817 err = ops->configure(rule, skb, frh, tb, extack);
818 if (err < 0)
819 goto errout_free;
820
821 err = call_fib_rule_notifiers(net, FIB_EVENT_RULE_ADD, rule, ops,
822 extack);
823 if (err < 0)
824 goto errout_free;
825
826 list_for_each_entry(r, &ops->rules_list, list) {
827 if (r->pref == rule->target) {
828 RCU_INIT_POINTER(rule->ctarget, r);
829 break;
830 }
831 }
832
833 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
834 unresolved = 1;
835
836 list_for_each_entry(r, &ops->rules_list, list) {
837 if (r->pref > rule->pref)
838 break;
839 last = r;
840 }
841
842 if (last)
843 list_add_rcu(&rule->list, &last->list);
844 else
845 list_add_rcu(&rule->list, &ops->rules_list);
846
847 if (ops->unresolved_rules) {
848 /*
849 * There are unresolved goto rules in the list, check if
850 * any of them are pointing to this new rule.
851 */
852 list_for_each_entry(r, &ops->rules_list, list) {
853 if (r->action == FR_ACT_GOTO &&
854 r->target == rule->pref &&
855 rtnl_dereference(r->ctarget) == NULL) {
856 rcu_assign_pointer(r->ctarget, rule);
857 if (--ops->unresolved_rules == 0)
858 break;
859 }
860 }
861 }
862
863 if (rule->action == FR_ACT_GOTO)
864 ops->nr_goto_rules++;
865
866 if (unresolved)
867 ops->unresolved_rules++;
868
869 if (rule->tun_id)
870 ip_tunnel_need_metadata();
871
872 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
873 flush_route_cache(ops);
874 rules_ops_put(ops);
875 return 0;
876
877errout_free:
878 kfree(rule);
879errout:
880 rules_ops_put(ops);
881 return err;
882}
883EXPORT_SYMBOL_GPL(fib_nl_newrule);
884
885int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh,
886 struct netlink_ext_ack *extack)
887{
888 struct net *net = sock_net(skb->sk);
889 struct fib_rule_hdr *frh = nlmsg_data(nlh);
890 struct fib_rules_ops *ops = NULL;
891 struct fib_rule *rule = NULL, *r, *nlrule = NULL;
892 struct nlattr *tb[FRA_MAX+1];
893 int err = -EINVAL;
894 bool user_priority = false;
895
896 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
897 NL_SET_ERR_MSG(extack, "Invalid msg length");
898 goto errout;
899 }
900
901 ops = lookup_rules_ops(net, frh->family);
902 if (ops == NULL) {
903 err = -EAFNOSUPPORT;
904 NL_SET_ERR_MSG(extack, "Rule family not supported");
905 goto errout;
906 }
907
908 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
909 fib_rule_policy, extack);
910 if (err < 0) {
911 NL_SET_ERR_MSG(extack, "Error parsing msg");
912 goto errout;
913 }
914
915 err = fib_nl2rule(skb, nlh, extack, ops, tb, &nlrule, &user_priority);
916 if (err)
917 goto errout;
918
919 rule = rule_find(ops, frh, tb, nlrule, user_priority);
920 if (!rule) {
921 err = -ENOENT;
922 goto errout;
923 }
924
925 if (rule->flags & FIB_RULE_PERMANENT) {
926 err = -EPERM;
927 goto errout;
928 }
929
930 if (ops->delete) {
931 err = ops->delete(rule);
932 if (err)
933 goto errout;
934 }
935
936 if (rule->tun_id)
937 ip_tunnel_unneed_metadata();
938
939 list_del_rcu(&rule->list);
940
941 if (rule->action == FR_ACT_GOTO) {
942 ops->nr_goto_rules--;
943 if (rtnl_dereference(rule->ctarget) == NULL)
944 ops->unresolved_rules--;
945 }
946
947 /*
948 * Check if this rule is a target to any of them. If so,
949 * adjust to the next one with the same preference or
950 * disable them. As this operation is eventually very
951 * expensive, it is only performed if goto rules, except
952 * current if it is goto rule, have actually been added.
953 */
954 if (ops->nr_goto_rules > 0) {
955 struct fib_rule *n;
956
957 n = list_next_entry(rule, list);
958 if (&n->list == &ops->rules_list || n->pref != rule->pref)
959 n = NULL;
960 list_for_each_entry(r, &ops->rules_list, list) {
961 if (rtnl_dereference(r->ctarget) != rule)
962 continue;
963 rcu_assign_pointer(r->ctarget, n);
964 if (!n)
965 ops->unresolved_rules++;
966 }
967 }
968
969 call_fib_rule_notifiers(net, FIB_EVENT_RULE_DEL, rule, ops,
970 NULL);
971 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
972 NETLINK_CB(skb).portid);
973 fib_rule_put(rule);
974 flush_route_cache(ops);
975 rules_ops_put(ops);
976 kfree(nlrule);
977 return 0;
978
979errout:
980 kfree(nlrule);
981 rules_ops_put(ops);
982 return err;
983}
984EXPORT_SYMBOL_GPL(fib_nl_delrule);
985
986static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
987 struct fib_rule *rule)
988{
989 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
990 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
991 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
992 + nla_total_size(4) /* FRA_PRIORITY */
993 + nla_total_size(4) /* FRA_TABLE */
994 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
995 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
996 + nla_total_size(4) /* FRA_FWMARK */
997 + nla_total_size(4) /* FRA_FWMASK */
998 + nla_total_size_64bit(8) /* FRA_TUN_ID */
999 + nla_total_size(sizeof(struct fib_kuid_range))
1000 + nla_total_size(1) /* FRA_PROTOCOL */
1001 + nla_total_size(1) /* FRA_IP_PROTO */
1002 + nla_total_size(sizeof(struct fib_rule_port_range)) /* FRA_SPORT_RANGE */
1003 + nla_total_size(sizeof(struct fib_rule_port_range)); /* FRA_DPORT_RANGE */
1004
1005 if (ops->nlmsg_payload)
1006 payload += ops->nlmsg_payload(rule);
1007
1008 return payload;
1009}
1010
1011static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
1012 u32 pid, u32 seq, int type, int flags,
1013 struct fib_rules_ops *ops)
1014{
1015 struct nlmsghdr *nlh;
1016 struct fib_rule_hdr *frh;
1017
1018 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
1019 if (nlh == NULL)
1020 return -EMSGSIZE;
1021
1022 frh = nlmsg_data(nlh);
1023 frh->family = ops->family;
1024 frh->table = rule->table < 256 ? rule->table : RT_TABLE_COMPAT;
1025 if (nla_put_u32(skb, FRA_TABLE, rule->table))
1026 goto nla_put_failure;
1027 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
1028 goto nla_put_failure;
1029 frh->res1 = 0;
1030 frh->res2 = 0;
1031 frh->action = rule->action;
1032 frh->flags = rule->flags;
1033
1034 if (nla_put_u8(skb, FRA_PROTOCOL, rule->proto))
1035 goto nla_put_failure;
1036
1037 if (rule->action == FR_ACT_GOTO &&
1038 rcu_access_pointer(rule->ctarget) == NULL)
1039 frh->flags |= FIB_RULE_UNRESOLVED;
1040
1041 if (rule->iifname[0]) {
1042 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
1043 goto nla_put_failure;
1044 if (READ_ONCE(rule->iifindex) == -1)
1045 frh->flags |= FIB_RULE_IIF_DETACHED;
1046 }
1047
1048 if (rule->oifname[0]) {
1049 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
1050 goto nla_put_failure;
1051 if (READ_ONCE(rule->oifindex) == -1)
1052 frh->flags |= FIB_RULE_OIF_DETACHED;
1053 }
1054
1055 if ((rule->pref &&
1056 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
1057 (rule->mark &&
1058 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
1059 ((rule->mark_mask || rule->mark) &&
1060 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
1061 (rule->target &&
1062 nla_put_u32(skb, FRA_GOTO, rule->target)) ||
1063 (rule->tun_id &&
1064 nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
1065 (rule->l3mdev &&
1066 nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
1067 (uid_range_set(&rule->uid_range) &&
1068 nla_put_uid_range(skb, &rule->uid_range)) ||
1069 (fib_rule_port_range_set(&rule->sport_range) &&
1070 nla_put_port_range(skb, FRA_SPORT_RANGE, &rule->sport_range)) ||
1071 (fib_rule_port_range_set(&rule->dport_range) &&
1072 nla_put_port_range(skb, FRA_DPORT_RANGE, &rule->dport_range)) ||
1073 (rule->ip_proto && nla_put_u8(skb, FRA_IP_PROTO, rule->ip_proto)))
1074 goto nla_put_failure;
1075
1076 if (rule->suppress_ifgroup != -1) {
1077 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
1078 goto nla_put_failure;
1079 }
1080
1081 if (ops->fill(rule, skb, frh) < 0)
1082 goto nla_put_failure;
1083
1084 nlmsg_end(skb, nlh);
1085 return 0;
1086
1087nla_put_failure:
1088 nlmsg_cancel(skb, nlh);
1089 return -EMSGSIZE;
1090}
1091
1092static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
1093 struct fib_rules_ops *ops)
1094{
1095 int idx = 0;
1096 struct fib_rule *rule;
1097 int err = 0;
1098
1099 rcu_read_lock();
1100 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
1101 if (idx < cb->args[1])
1102 goto skip;
1103
1104 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
1105 cb->nlh->nlmsg_seq, RTM_NEWRULE,
1106 NLM_F_MULTI, ops);
1107 if (err)
1108 break;
1109skip:
1110 idx++;
1111 }
1112 rcu_read_unlock();
1113 cb->args[1] = idx;
1114 rules_ops_put(ops);
1115
1116 return err;
1117}
1118
1119static int fib_valid_dumprule_req(const struct nlmsghdr *nlh,
1120 struct netlink_ext_ack *extack)
1121{
1122 struct fib_rule_hdr *frh;
1123
1124 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
1125 NL_SET_ERR_MSG(extack, "Invalid header for fib rule dump request");
1126 return -EINVAL;
1127 }
1128
1129 frh = nlmsg_data(nlh);
1130 if (frh->dst_len || frh->src_len || frh->tos || frh->table ||
1131 frh->res1 || frh->res2 || frh->action || frh->flags) {
1132 NL_SET_ERR_MSG(extack,
1133 "Invalid values in header for fib rule dump request");
1134 return -EINVAL;
1135 }
1136
1137 if (nlmsg_attrlen(nlh, sizeof(*frh))) {
1138 NL_SET_ERR_MSG(extack, "Invalid data after header in fib rule dump request");
1139 return -EINVAL;
1140 }
1141
1142 return 0;
1143}
1144
1145static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
1146{
1147 const struct nlmsghdr *nlh = cb->nlh;
1148 struct net *net = sock_net(skb->sk);
1149 struct fib_rules_ops *ops;
1150 int err, idx = 0, family;
1151
1152 if (cb->strict_check) {
1153 err = fib_valid_dumprule_req(nlh, cb->extack);
1154
1155 if (err < 0)
1156 return err;
1157 }
1158
1159 family = rtnl_msg_family(nlh);
1160 if (family != AF_UNSPEC) {
1161 /* Protocol specific dump request */
1162 ops = lookup_rules_ops(net, family);
1163 if (ops == NULL)
1164 return -EAFNOSUPPORT;
1165
1166 return dump_rules(skb, cb, ops);
1167 }
1168
1169 err = 0;
1170 rcu_read_lock();
1171 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
1172 if (idx < cb->args[0] || !try_module_get(ops->owner))
1173 goto skip;
1174
1175 err = dump_rules(skb, cb, ops);
1176 if (err < 0)
1177 break;
1178
1179 cb->args[1] = 0;
1180skip:
1181 idx++;
1182 }
1183 rcu_read_unlock();
1184 cb->args[0] = idx;
1185
1186 return err;
1187}
1188
1189static void notify_rule_change(int event, struct fib_rule *rule,
1190 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
1191 u32 pid)
1192{
1193 struct net *net;
1194 struct sk_buff *skb;
1195 int err = -ENOMEM;
1196
1197 net = ops->fro_net;
1198 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
1199 if (skb == NULL)
1200 goto errout;
1201
1202 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
1203 if (err < 0) {
1204 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
1205 WARN_ON(err == -EMSGSIZE);
1206 kfree_skb(skb);
1207 goto errout;
1208 }
1209
1210 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
1211 return;
1212errout:
1213 rtnl_set_sk_err(net, ops->nlgroup, err);
1214}
1215
1216static void attach_rules(struct list_head *rules, struct net_device *dev)
1217{
1218 struct fib_rule *rule;
1219
1220 list_for_each_entry(rule, rules, list) {
1221 if (rule->iifindex == -1 &&
1222 strcmp(dev->name, rule->iifname) == 0)
1223 WRITE_ONCE(rule->iifindex, dev->ifindex);
1224 if (rule->oifindex == -1 &&
1225 strcmp(dev->name, rule->oifname) == 0)
1226 WRITE_ONCE(rule->oifindex, dev->ifindex);
1227 }
1228}
1229
1230static void detach_rules(struct list_head *rules, struct net_device *dev)
1231{
1232 struct fib_rule *rule;
1233
1234 list_for_each_entry(rule, rules, list) {
1235 if (rule->iifindex == dev->ifindex)
1236 WRITE_ONCE(rule->iifindex, -1);
1237 if (rule->oifindex == dev->ifindex)
1238 WRITE_ONCE(rule->oifindex, -1);
1239 }
1240}
1241
1242
1243static int fib_rules_event(struct notifier_block *this, unsigned long event,
1244 void *ptr)
1245{
1246 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1247 struct net *net = dev_net(dev);
1248 struct fib_rules_ops *ops;
1249
1250 ASSERT_RTNL();
1251
1252 switch (event) {
1253 case NETDEV_REGISTER:
1254 list_for_each_entry(ops, &net->rules_ops, list)
1255 attach_rules(&ops->rules_list, dev);
1256 break;
1257
1258 case NETDEV_CHANGENAME:
1259 list_for_each_entry(ops, &net->rules_ops, list) {
1260 detach_rules(&ops->rules_list, dev);
1261 attach_rules(&ops->rules_list, dev);
1262 }
1263 break;
1264
1265 case NETDEV_UNREGISTER:
1266 list_for_each_entry(ops, &net->rules_ops, list)
1267 detach_rules(&ops->rules_list, dev);
1268 break;
1269 }
1270
1271 return NOTIFY_DONE;
1272}
1273
1274static struct notifier_block fib_rules_notifier = {
1275 .notifier_call = fib_rules_event,
1276};
1277
1278static int __net_init fib_rules_net_init(struct net *net)
1279{
1280 INIT_LIST_HEAD(&net->rules_ops);
1281 spin_lock_init(&net->rules_mod_lock);
1282 return 0;
1283}
1284
1285static void __net_exit fib_rules_net_exit(struct net *net)
1286{
1287 WARN_ON_ONCE(!list_empty(&net->rules_ops));
1288}
1289
1290static struct pernet_operations fib_rules_net_ops = {
1291 .init = fib_rules_net_init,
1292 .exit = fib_rules_net_exit,
1293};
1294
1295static const struct rtnl_msg_handler fib_rules_rtnl_msg_handlers[] __initconst = {
1296 {.msgtype = RTM_NEWRULE, .doit = fib_nl_newrule},
1297 {.msgtype = RTM_DELRULE, .doit = fib_nl_delrule},
1298 {.msgtype = RTM_GETRULE, .dumpit = fib_nl_dumprule,
1299 .flags = RTNL_FLAG_DUMP_UNLOCKED},
1300};
1301
1302static int __init fib_rules_init(void)
1303{
1304 int err;
1305
1306 rtnl_register_many(fib_rules_rtnl_msg_handlers);
1307
1308 err = register_pernet_subsys(&fib_rules_net_ops);
1309 if (err < 0)
1310 goto fail;
1311
1312 err = register_netdevice_notifier(&fib_rules_notifier);
1313 if (err < 0)
1314 goto fail_unregister;
1315
1316 return 0;
1317
1318fail_unregister:
1319 unregister_pernet_subsys(&fib_rules_net_ops);
1320fail:
1321 rtnl_unregister_many(fib_rules_rtnl_msg_handlers);
1322 return err;
1323}
1324
1325subsys_initcall(fib_rules_init);