Loading...
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#include <linux/indirect_call_wrapper.h>
18
19#if defined(CONFIG_IPV6) && defined(CONFIG_IPV6_MULTIPLE_TABLES)
20#ifdef CONFIG_IP_MULTIPLE_TABLES
21#define INDIRECT_CALL_MT(f, f2, f1, ...) \
22 INDIRECT_CALL_INET(f, f2, f1, __VA_ARGS__)
23#else
24#define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
25#endif
26#elif defined(CONFIG_IP_MULTIPLE_TABLES)
27#define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f1, __VA_ARGS__)
28#else
29#define INDIRECT_CALL_MT(f, f2, f1, ...) f(__VA_ARGS__)
30#endif
31
32static const struct fib_kuid_range fib_kuid_range_unset = {
33 KUIDT_INIT(0),
34 KUIDT_INIT(~0),
35};
36
37bool fib_rule_matchall(const struct fib_rule *rule)
38{
39 if (rule->iifindex || rule->oifindex || rule->mark || rule->tun_id ||
40 rule->flags)
41 return false;
42 if (rule->suppress_ifgroup != -1 || rule->suppress_prefixlen != -1)
43 return false;
44 if (!uid_eq(rule->uid_range.start, fib_kuid_range_unset.start) ||
45 !uid_eq(rule->uid_range.end, fib_kuid_range_unset.end))
46 return false;
47 if (fib_rule_port_range_set(&rule->sport_range))
48 return false;
49 if (fib_rule_port_range_set(&rule->dport_range))
50 return false;
51 return true;
52}
53EXPORT_SYMBOL_GPL(fib_rule_matchall);
54
55int fib_default_rule_add(struct fib_rules_ops *ops,
56 u32 pref, u32 table, u32 flags)
57{
58 struct fib_rule *r;
59
60 r = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT);
61 if (r == NULL)
62 return -ENOMEM;
63
64 refcount_set(&r->refcnt, 1);
65 r->action = FR_ACT_TO_TBL;
66 r->pref = pref;
67 r->table = table;
68 r->flags = flags;
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 unreacheable
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(struct net *net, int family)
105{
106 struct fib_rules_ops *ops;
107
108 rcu_read_lock();
109 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
110 if (ops->family == family) {
111 if (!try_module_get(ops->owner))
112 ops = NULL;
113 rcu_read_unlock();
114 return ops;
115 }
116 }
117 rcu_read_unlock();
118
119 return NULL;
120}
121
122static void rules_ops_put(struct fib_rules_ops *ops)
123{
124 if (ops)
125 module_put(ops->owner);
126}
127
128static void flush_route_cache(struct fib_rules_ops *ops)
129{
130 if (ops->flush_cache)
131 ops->flush_cache(ops);
132}
133
134static int __fib_rules_register(struct fib_rules_ops *ops)
135{
136 int err = -EEXIST;
137 struct fib_rules_ops *o;
138 struct net *net;
139
140 net = ops->fro_net;
141
142 if (ops->rule_size < sizeof(struct fib_rule))
143 return -EINVAL;
144
145 if (ops->match == NULL || ops->configure == NULL ||
146 ops->compare == NULL || ops->fill == NULL ||
147 ops->action == NULL)
148 return -EINVAL;
149
150 spin_lock(&net->rules_mod_lock);
151 list_for_each_entry(o, &net->rules_ops, list)
152 if (ops->family == o->family)
153 goto errout;
154
155 list_add_tail_rcu(&ops->list, &net->rules_ops);
156 err = 0;
157errout:
158 spin_unlock(&net->rules_mod_lock);
159
160 return err;
161}
162
163struct fib_rules_ops *
164fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
165{
166 struct fib_rules_ops *ops;
167 int err;
168
169 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
170 if (ops == NULL)
171 return ERR_PTR(-ENOMEM);
172
173 INIT_LIST_HEAD(&ops->rules_list);
174 ops->fro_net = net;
175
176 err = __fib_rules_register(ops);
177 if (err) {
178 kfree(ops);
179 ops = ERR_PTR(err);
180 }
181
182 return ops;
183}
184EXPORT_SYMBOL_GPL(fib_rules_register);
185
186static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
187{
188 struct fib_rule *rule, *tmp;
189
190 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
191 list_del_rcu(&rule->list);
192 if (ops->delete)
193 ops->delete(rule);
194 fib_rule_put(rule);
195 }
196}
197
198void fib_rules_unregister(struct fib_rules_ops *ops)
199{
200 struct net *net = ops->fro_net;
201
202 spin_lock(&net->rules_mod_lock);
203 list_del_rcu(&ops->list);
204 spin_unlock(&net->rules_mod_lock);
205
206 fib_rules_cleanup_ops(ops);
207 kfree_rcu(ops, rcu);
208}
209EXPORT_SYMBOL_GPL(fib_rules_unregister);
210
211static int uid_range_set(struct fib_kuid_range *range)
212{
213 return uid_valid(range->start) && uid_valid(range->end);
214}
215
216static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
217{
218 struct fib_rule_uid_range *in;
219 struct fib_kuid_range out;
220
221 in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
222
223 out.start = make_kuid(current_user_ns(), in->start);
224 out.end = make_kuid(current_user_ns(), in->end);
225
226 return out;
227}
228
229static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
230{
231 struct fib_rule_uid_range out = {
232 from_kuid_munged(current_user_ns(), range->start),
233 from_kuid_munged(current_user_ns(), range->end)
234 };
235
236 return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
237}
238
239static int nla_get_port_range(struct nlattr *pattr,
240 struct fib_rule_port_range *port_range)
241{
242 const struct fib_rule_port_range *pr = nla_data(pattr);
243
244 if (!fib_rule_port_range_valid(pr))
245 return -EINVAL;
246
247 port_range->start = pr->start;
248 port_range->end = pr->end;
249
250 return 0;
251}
252
253static int nla_put_port_range(struct sk_buff *skb, int attrtype,
254 struct fib_rule_port_range *range)
255{
256 return nla_put(skb, attrtype, sizeof(*range), range);
257}
258
259static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
260 struct flowi *fl, int flags,
261 struct fib_lookup_arg *arg)
262{
263 int ret = 0;
264
265 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
266 goto out;
267
268 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
269 goto out;
270
271 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
272 goto out;
273
274 if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
275 goto out;
276
277 if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
278 goto out;
279
280 if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
281 uid_gt(fl->flowi_uid, rule->uid_range.end))
282 goto out;
283
284 ret = INDIRECT_CALL_MT(ops->match,
285 fib6_rule_match,
286 fib4_rule_match,
287 rule, fl, flags);
288out:
289 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
290}
291
292int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
293 int flags, struct fib_lookup_arg *arg)
294{
295 struct fib_rule *rule;
296 int err;
297
298 rcu_read_lock();
299
300 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
301jumped:
302 if (!fib_rule_match(rule, ops, fl, flags, arg))
303 continue;
304
305 if (rule->action == FR_ACT_GOTO) {
306 struct fib_rule *target;
307
308 target = rcu_dereference(rule->ctarget);
309 if (target == NULL) {
310 continue;
311 } else {
312 rule = target;
313 goto jumped;
314 }
315 } else if (rule->action == FR_ACT_NOP)
316 continue;
317 else
318 err = INDIRECT_CALL_MT(ops->action,
319 fib6_rule_action,
320 fib4_rule_action,
321 rule, fl, flags, arg);
322
323 if (!err && ops->suppress && INDIRECT_CALL_MT(ops->suppress,
324 fib6_rule_suppress,
325 fib4_rule_suppress,
326 rule, flags, arg))
327 continue;
328
329 if (err != -EAGAIN) {
330 if ((arg->flags & FIB_LOOKUP_NOREF) ||
331 likely(refcount_inc_not_zero(&rule->refcnt))) {
332 arg->rule = rule;
333 goto out;
334 }
335 break;
336 }
337 }
338
339 err = -ESRCH;
340out:
341 rcu_read_unlock();
342
343 return err;
344}
345EXPORT_SYMBOL_GPL(fib_rules_lookup);
346
347static int call_fib_rule_notifier(struct notifier_block *nb,
348 enum fib_event_type event_type,
349 struct fib_rule *rule, int family,
350 struct netlink_ext_ack *extack)
351{
352 struct fib_rule_notifier_info info = {
353 .info.family = family,
354 .info.extack = extack,
355 .rule = rule,
356 };
357
358 return call_fib_notifier(nb, event_type, &info.info);
359}
360
361static int call_fib_rule_notifiers(struct net *net,
362 enum fib_event_type event_type,
363 struct fib_rule *rule,
364 struct fib_rules_ops *ops,
365 struct netlink_ext_ack *extack)
366{
367 struct fib_rule_notifier_info info = {
368 .info.family = ops->family,
369 .info.extack = extack,
370 .rule = rule,
371 };
372
373 ops->fib_rules_seq++;
374 return call_fib_notifiers(net, event_type, &info.info);
375}
376
377/* Called with rcu_read_lock() */
378int fib_rules_dump(struct net *net, struct notifier_block *nb, int family,
379 struct netlink_ext_ack *extack)
380{
381 struct fib_rules_ops *ops;
382 struct fib_rule *rule;
383 int err = 0;
384
385 ops = lookup_rules_ops(net, family);
386 if (!ops)
387 return -EAFNOSUPPORT;
388 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
389 err = call_fib_rule_notifier(nb, FIB_EVENT_RULE_ADD,
390 rule, family, extack);
391 if (err)
392 break;
393 }
394 rules_ops_put(ops);
395
396 return err;
397}
398EXPORT_SYMBOL_GPL(fib_rules_dump);
399
400unsigned int fib_rules_seq_read(struct net *net, int family)
401{
402 unsigned int fib_rules_seq;
403 struct fib_rules_ops *ops;
404
405 ASSERT_RTNL();
406
407 ops = lookup_rules_ops(net, family);
408 if (!ops)
409 return 0;
410 fib_rules_seq = ops->fib_rules_seq;
411 rules_ops_put(ops);
412
413 return fib_rules_seq;
414}
415EXPORT_SYMBOL_GPL(fib_rules_seq_read);
416
417static struct fib_rule *rule_find(struct fib_rules_ops *ops,
418 struct fib_rule_hdr *frh,
419 struct nlattr **tb,
420 struct fib_rule *rule,
421 bool user_priority)
422{
423 struct fib_rule *r;
424
425 list_for_each_entry(r, &ops->rules_list, list) {
426 if (rule->action && r->action != rule->action)
427 continue;
428
429 if (rule->table && r->table != rule->table)
430 continue;
431
432 if (user_priority && r->pref != rule->pref)
433 continue;
434
435 if (rule->iifname[0] &&
436 memcmp(r->iifname, rule->iifname, IFNAMSIZ))
437 continue;
438
439 if (rule->oifname[0] &&
440 memcmp(r->oifname, rule->oifname, IFNAMSIZ))
441 continue;
442
443 if (rule->mark && r->mark != rule->mark)
444 continue;
445
446 if (rule->suppress_ifgroup != -1 &&
447 r->suppress_ifgroup != rule->suppress_ifgroup)
448 continue;
449
450 if (rule->suppress_prefixlen != -1 &&
451 r->suppress_prefixlen != rule->suppress_prefixlen)
452 continue;
453
454 if (rule->mark_mask && r->mark_mask != rule->mark_mask)
455 continue;
456
457 if (rule->tun_id && r->tun_id != rule->tun_id)
458 continue;
459
460 if (r->fr_net != rule->fr_net)
461 continue;
462
463 if (rule->l3mdev && r->l3mdev != rule->l3mdev)
464 continue;
465
466 if (uid_range_set(&rule->uid_range) &&
467 (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
468 !uid_eq(r->uid_range.end, rule->uid_range.end)))
469 continue;
470
471 if (rule->ip_proto && r->ip_proto != rule->ip_proto)
472 continue;
473
474 if (rule->proto && r->proto != rule->proto)
475 continue;
476
477 if (fib_rule_port_range_set(&rule->sport_range) &&
478 !fib_rule_port_range_compare(&r->sport_range,
479 &rule->sport_range))
480 continue;
481
482 if (fib_rule_port_range_set(&rule->dport_range) &&
483 !fib_rule_port_range_compare(&r->dport_range,
484 &rule->dport_range))
485 continue;
486
487 if (!ops->compare(r, frh, tb))
488 continue;
489 return r;
490 }
491
492 return NULL;
493}
494
495#ifdef CONFIG_NET_L3_MASTER_DEV
496static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
497 struct netlink_ext_ack *extack)
498{
499 nlrule->l3mdev = nla_get_u8(nla);
500 if (nlrule->l3mdev != 1) {
501 NL_SET_ERR_MSG(extack, "Invalid l3mdev attribute");
502 return -1;
503 }
504
505 return 0;
506}
507#else
508static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
509 struct netlink_ext_ack *extack)
510{
511 NL_SET_ERR_MSG(extack, "l3mdev support is not enabled in kernel");
512 return -1;
513}
514#endif
515
516static int fib_nl2rule(struct sk_buff *skb, struct nlmsghdr *nlh,
517 struct netlink_ext_ack *extack,
518 struct fib_rules_ops *ops,
519 struct nlattr *tb[],
520 struct fib_rule **rule,
521 bool *user_priority)
522{
523 struct net *net = sock_net(skb->sk);
524 struct fib_rule_hdr *frh = nlmsg_data(nlh);
525 struct fib_rule *nlrule = NULL;
526 int err = -EINVAL;
527
528 if (frh->src_len)
529 if (!tb[FRA_SRC] ||
530 frh->src_len > (ops->addr_size * 8) ||
531 nla_len(tb[FRA_SRC]) != ops->addr_size) {
532 NL_SET_ERR_MSG(extack, "Invalid source address");
533 goto errout;
534 }
535
536 if (frh->dst_len)
537 if (!tb[FRA_DST] ||
538 frh->dst_len > (ops->addr_size * 8) ||
539 nla_len(tb[FRA_DST]) != ops->addr_size) {
540 NL_SET_ERR_MSG(extack, "Invalid dst address");
541 goto errout;
542 }
543
544 nlrule = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT);
545 if (!nlrule) {
546 err = -ENOMEM;
547 goto errout;
548 }
549 refcount_set(&nlrule->refcnt, 1);
550 nlrule->fr_net = net;
551
552 if (tb[FRA_PRIORITY]) {
553 nlrule->pref = nla_get_u32(tb[FRA_PRIORITY]);
554 *user_priority = true;
555 } else {
556 nlrule->pref = fib_default_rule_pref(ops);
557 }
558
559 nlrule->proto = tb[FRA_PROTOCOL] ?
560 nla_get_u8(tb[FRA_PROTOCOL]) : RTPROT_UNSPEC;
561
562 if (tb[FRA_IIFNAME]) {
563 struct net_device *dev;
564
565 nlrule->iifindex = -1;
566 nla_strscpy(nlrule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
567 dev = __dev_get_by_name(net, nlrule->iifname);
568 if (dev)
569 nlrule->iifindex = dev->ifindex;
570 }
571
572 if (tb[FRA_OIFNAME]) {
573 struct net_device *dev;
574
575 nlrule->oifindex = -1;
576 nla_strscpy(nlrule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
577 dev = __dev_get_by_name(net, nlrule->oifname);
578 if (dev)
579 nlrule->oifindex = dev->ifindex;
580 }
581
582 if (tb[FRA_FWMARK]) {
583 nlrule->mark = nla_get_u32(tb[FRA_FWMARK]);
584 if (nlrule->mark)
585 /* compatibility: if the mark value is non-zero all bits
586 * are compared unless a mask is explicitly specified.
587 */
588 nlrule->mark_mask = 0xFFFFFFFF;
589 }
590
591 if (tb[FRA_FWMASK])
592 nlrule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
593
594 if (tb[FRA_TUN_ID])
595 nlrule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
596
597 err = -EINVAL;
598 if (tb[FRA_L3MDEV] &&
599 fib_nl2rule_l3mdev(tb[FRA_L3MDEV], nlrule, extack) < 0)
600 goto errout_free;
601
602 nlrule->action = frh->action;
603 nlrule->flags = frh->flags;
604 nlrule->table = frh_get_table(frh, tb);
605 if (tb[FRA_SUPPRESS_PREFIXLEN])
606 nlrule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
607 else
608 nlrule->suppress_prefixlen = -1;
609
610 if (tb[FRA_SUPPRESS_IFGROUP])
611 nlrule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
612 else
613 nlrule->suppress_ifgroup = -1;
614
615 if (tb[FRA_GOTO]) {
616 if (nlrule->action != FR_ACT_GOTO) {
617 NL_SET_ERR_MSG(extack, "Unexpected goto");
618 goto errout_free;
619 }
620
621 nlrule->target = nla_get_u32(tb[FRA_GOTO]);
622 /* Backward jumps are prohibited to avoid endless loops */
623 if (nlrule->target <= nlrule->pref) {
624 NL_SET_ERR_MSG(extack, "Backward goto not supported");
625 goto errout_free;
626 }
627 } else if (nlrule->action == FR_ACT_GOTO) {
628 NL_SET_ERR_MSG(extack, "Missing goto target for action goto");
629 goto errout_free;
630 }
631
632 if (nlrule->l3mdev && nlrule->table) {
633 NL_SET_ERR_MSG(extack, "l3mdev and table are mutually exclusive");
634 goto errout_free;
635 }
636
637 if (tb[FRA_UID_RANGE]) {
638 if (current_user_ns() != net->user_ns) {
639 err = -EPERM;
640 NL_SET_ERR_MSG(extack, "No permission to set uid");
641 goto errout_free;
642 }
643
644 nlrule->uid_range = nla_get_kuid_range(tb);
645
646 if (!uid_range_set(&nlrule->uid_range) ||
647 !uid_lte(nlrule->uid_range.start, nlrule->uid_range.end)) {
648 NL_SET_ERR_MSG(extack, "Invalid uid range");
649 goto errout_free;
650 }
651 } else {
652 nlrule->uid_range = fib_kuid_range_unset;
653 }
654
655 if (tb[FRA_IP_PROTO])
656 nlrule->ip_proto = nla_get_u8(tb[FRA_IP_PROTO]);
657
658 if (tb[FRA_SPORT_RANGE]) {
659 err = nla_get_port_range(tb[FRA_SPORT_RANGE],
660 &nlrule->sport_range);
661 if (err) {
662 NL_SET_ERR_MSG(extack, "Invalid sport range");
663 goto errout_free;
664 }
665 }
666
667 if (tb[FRA_DPORT_RANGE]) {
668 err = nla_get_port_range(tb[FRA_DPORT_RANGE],
669 &nlrule->dport_range);
670 if (err) {
671 NL_SET_ERR_MSG(extack, "Invalid dport range");
672 goto errout_free;
673 }
674 }
675
676 *rule = nlrule;
677
678 return 0;
679
680errout_free:
681 kfree(nlrule);
682errout:
683 return err;
684}
685
686static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
687 struct nlattr **tb, struct fib_rule *rule)
688{
689 struct fib_rule *r;
690
691 list_for_each_entry(r, &ops->rules_list, list) {
692 if (r->action != rule->action)
693 continue;
694
695 if (r->table != rule->table)
696 continue;
697
698 if (r->pref != rule->pref)
699 continue;
700
701 if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
702 continue;
703
704 if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
705 continue;
706
707 if (r->mark != rule->mark)
708 continue;
709
710 if (r->suppress_ifgroup != rule->suppress_ifgroup)
711 continue;
712
713 if (r->suppress_prefixlen != rule->suppress_prefixlen)
714 continue;
715
716 if (r->mark_mask != rule->mark_mask)
717 continue;
718
719 if (r->tun_id != rule->tun_id)
720 continue;
721
722 if (r->fr_net != rule->fr_net)
723 continue;
724
725 if (r->l3mdev != rule->l3mdev)
726 continue;
727
728 if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
729 !uid_eq(r->uid_range.end, rule->uid_range.end))
730 continue;
731
732 if (r->ip_proto != rule->ip_proto)
733 continue;
734
735 if (r->proto != rule->proto)
736 continue;
737
738 if (!fib_rule_port_range_compare(&r->sport_range,
739 &rule->sport_range))
740 continue;
741
742 if (!fib_rule_port_range_compare(&r->dport_range,
743 &rule->dport_range))
744 continue;
745
746 if (!ops->compare(r, frh, tb))
747 continue;
748 return 1;
749 }
750 return 0;
751}
752
753static const struct nla_policy fib_rule_policy[FRA_MAX + 1] = {
754 [FRA_UNSPEC] = { .strict_start_type = FRA_DPORT_RANGE + 1 },
755 [FRA_IIFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ - 1 },
756 [FRA_OIFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ - 1 },
757 [FRA_PRIORITY] = { .type = NLA_U32 },
758 [FRA_FWMARK] = { .type = NLA_U32 },
759 [FRA_FLOW] = { .type = NLA_U32 },
760 [FRA_TUN_ID] = { .type = NLA_U64 },
761 [FRA_FWMASK] = { .type = NLA_U32 },
762 [FRA_TABLE] = { .type = NLA_U32 },
763 [FRA_SUPPRESS_PREFIXLEN] = { .type = NLA_U32 },
764 [FRA_SUPPRESS_IFGROUP] = { .type = NLA_U32 },
765 [FRA_GOTO] = { .type = NLA_U32 },
766 [FRA_L3MDEV] = { .type = NLA_U8 },
767 [FRA_UID_RANGE] = { .len = sizeof(struct fib_rule_uid_range) },
768 [FRA_PROTOCOL] = { .type = NLA_U8 },
769 [FRA_IP_PROTO] = { .type = NLA_U8 },
770 [FRA_SPORT_RANGE] = { .len = sizeof(struct fib_rule_port_range) },
771 [FRA_DPORT_RANGE] = { .len = sizeof(struct fib_rule_port_range) }
772};
773
774int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh,
775 struct netlink_ext_ack *extack)
776{
777 struct net *net = sock_net(skb->sk);
778 struct fib_rule_hdr *frh = nlmsg_data(nlh);
779 struct fib_rules_ops *ops = NULL;
780 struct fib_rule *rule = NULL, *r, *last = NULL;
781 struct nlattr *tb[FRA_MAX + 1];
782 int err = -EINVAL, unresolved = 0;
783 bool user_priority = false;
784
785 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
786 NL_SET_ERR_MSG(extack, "Invalid msg length");
787 goto errout;
788 }
789
790 ops = lookup_rules_ops(net, frh->family);
791 if (!ops) {
792 err = -EAFNOSUPPORT;
793 NL_SET_ERR_MSG(extack, "Rule family not supported");
794 goto errout;
795 }
796
797 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
798 fib_rule_policy, extack);
799 if (err < 0) {
800 NL_SET_ERR_MSG(extack, "Error parsing msg");
801 goto errout;
802 }
803
804 err = fib_nl2rule(skb, nlh, extack, ops, tb, &rule, &user_priority);
805 if (err)
806 goto errout;
807
808 if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
809 rule_exists(ops, frh, tb, rule)) {
810 err = -EEXIST;
811 goto errout_free;
812 }
813
814 err = ops->configure(rule, skb, frh, tb, extack);
815 if (err < 0)
816 goto errout_free;
817
818 err = call_fib_rule_notifiers(net, FIB_EVENT_RULE_ADD, rule, ops,
819 extack);
820 if (err < 0)
821 goto errout_free;
822
823 list_for_each_entry(r, &ops->rules_list, list) {
824 if (r->pref == rule->target) {
825 RCU_INIT_POINTER(rule->ctarget, r);
826 break;
827 }
828 }
829
830 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
831 unresolved = 1;
832
833 list_for_each_entry(r, &ops->rules_list, list) {
834 if (r->pref > rule->pref)
835 break;
836 last = r;
837 }
838
839 if (last)
840 list_add_rcu(&rule->list, &last->list);
841 else
842 list_add_rcu(&rule->list, &ops->rules_list);
843
844 if (ops->unresolved_rules) {
845 /*
846 * There are unresolved goto rules in the list, check if
847 * any of them are pointing to this new rule.
848 */
849 list_for_each_entry(r, &ops->rules_list, list) {
850 if (r->action == FR_ACT_GOTO &&
851 r->target == rule->pref &&
852 rtnl_dereference(r->ctarget) == NULL) {
853 rcu_assign_pointer(r->ctarget, rule);
854 if (--ops->unresolved_rules == 0)
855 break;
856 }
857 }
858 }
859
860 if (rule->action == FR_ACT_GOTO)
861 ops->nr_goto_rules++;
862
863 if (unresolved)
864 ops->unresolved_rules++;
865
866 if (rule->tun_id)
867 ip_tunnel_need_metadata();
868
869 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
870 flush_route_cache(ops);
871 rules_ops_put(ops);
872 return 0;
873
874errout_free:
875 kfree(rule);
876errout:
877 rules_ops_put(ops);
878 return err;
879}
880EXPORT_SYMBOL_GPL(fib_nl_newrule);
881
882int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh,
883 struct netlink_ext_ack *extack)
884{
885 struct net *net = sock_net(skb->sk);
886 struct fib_rule_hdr *frh = nlmsg_data(nlh);
887 struct fib_rules_ops *ops = NULL;
888 struct fib_rule *rule = NULL, *r, *nlrule = NULL;
889 struct nlattr *tb[FRA_MAX+1];
890 int err = -EINVAL;
891 bool user_priority = false;
892
893 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
894 NL_SET_ERR_MSG(extack, "Invalid msg length");
895 goto errout;
896 }
897
898 ops = lookup_rules_ops(net, frh->family);
899 if (ops == NULL) {
900 err = -EAFNOSUPPORT;
901 NL_SET_ERR_MSG(extack, "Rule family not supported");
902 goto errout;
903 }
904
905 err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
906 fib_rule_policy, extack);
907 if (err < 0) {
908 NL_SET_ERR_MSG(extack, "Error parsing msg");
909 goto errout;
910 }
911
912 err = fib_nl2rule(skb, nlh, extack, ops, tb, &nlrule, &user_priority);
913 if (err)
914 goto errout;
915
916 rule = rule_find(ops, frh, tb, nlrule, user_priority);
917 if (!rule) {
918 err = -ENOENT;
919 goto errout;
920 }
921
922 if (rule->flags & FIB_RULE_PERMANENT) {
923 err = -EPERM;
924 goto errout;
925 }
926
927 if (ops->delete) {
928 err = ops->delete(rule);
929 if (err)
930 goto errout;
931 }
932
933 if (rule->tun_id)
934 ip_tunnel_unneed_metadata();
935
936 list_del_rcu(&rule->list);
937
938 if (rule->action == FR_ACT_GOTO) {
939 ops->nr_goto_rules--;
940 if (rtnl_dereference(rule->ctarget) == NULL)
941 ops->unresolved_rules--;
942 }
943
944 /*
945 * Check if this rule is a target to any of them. If so,
946 * adjust to the next one with the same preference or
947 * disable them. As this operation is eventually very
948 * expensive, it is only performed if goto rules, except
949 * current if it is goto rule, have actually been added.
950 */
951 if (ops->nr_goto_rules > 0) {
952 struct fib_rule *n;
953
954 n = list_next_entry(rule, list);
955 if (&n->list == &ops->rules_list || n->pref != rule->pref)
956 n = NULL;
957 list_for_each_entry(r, &ops->rules_list, list) {
958 if (rtnl_dereference(r->ctarget) != rule)
959 continue;
960 rcu_assign_pointer(r->ctarget, n);
961 if (!n)
962 ops->unresolved_rules++;
963 }
964 }
965
966 call_fib_rule_notifiers(net, FIB_EVENT_RULE_DEL, rule, ops,
967 NULL);
968 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
969 NETLINK_CB(skb).portid);
970 fib_rule_put(rule);
971 flush_route_cache(ops);
972 rules_ops_put(ops);
973 kfree(nlrule);
974 return 0;
975
976errout:
977 kfree(nlrule);
978 rules_ops_put(ops);
979 return err;
980}
981EXPORT_SYMBOL_GPL(fib_nl_delrule);
982
983static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
984 struct fib_rule *rule)
985{
986 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
987 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
988 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
989 + nla_total_size(4) /* FRA_PRIORITY */
990 + nla_total_size(4) /* FRA_TABLE */
991 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
992 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
993 + nla_total_size(4) /* FRA_FWMARK */
994 + nla_total_size(4) /* FRA_FWMASK */
995 + nla_total_size_64bit(8) /* FRA_TUN_ID */
996 + nla_total_size(sizeof(struct fib_kuid_range))
997 + nla_total_size(1) /* FRA_PROTOCOL */
998 + nla_total_size(1) /* FRA_IP_PROTO */
999 + nla_total_size(sizeof(struct fib_rule_port_range)) /* FRA_SPORT_RANGE */
1000 + nla_total_size(sizeof(struct fib_rule_port_range)); /* FRA_DPORT_RANGE */
1001
1002 if (ops->nlmsg_payload)
1003 payload += ops->nlmsg_payload(rule);
1004
1005 return payload;
1006}
1007
1008static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
1009 u32 pid, u32 seq, int type, int flags,
1010 struct fib_rules_ops *ops)
1011{
1012 struct nlmsghdr *nlh;
1013 struct fib_rule_hdr *frh;
1014
1015 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
1016 if (nlh == NULL)
1017 return -EMSGSIZE;
1018
1019 frh = nlmsg_data(nlh);
1020 frh->family = ops->family;
1021 frh->table = rule->table < 256 ? rule->table : RT_TABLE_COMPAT;
1022 if (nla_put_u32(skb, FRA_TABLE, rule->table))
1023 goto nla_put_failure;
1024 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
1025 goto nla_put_failure;
1026 frh->res1 = 0;
1027 frh->res2 = 0;
1028 frh->action = rule->action;
1029 frh->flags = rule->flags;
1030
1031 if (nla_put_u8(skb, FRA_PROTOCOL, rule->proto))
1032 goto nla_put_failure;
1033
1034 if (rule->action == FR_ACT_GOTO &&
1035 rcu_access_pointer(rule->ctarget) == NULL)
1036 frh->flags |= FIB_RULE_UNRESOLVED;
1037
1038 if (rule->iifname[0]) {
1039 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
1040 goto nla_put_failure;
1041 if (rule->iifindex == -1)
1042 frh->flags |= FIB_RULE_IIF_DETACHED;
1043 }
1044
1045 if (rule->oifname[0]) {
1046 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
1047 goto nla_put_failure;
1048 if (rule->oifindex == -1)
1049 frh->flags |= FIB_RULE_OIF_DETACHED;
1050 }
1051
1052 if ((rule->pref &&
1053 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
1054 (rule->mark &&
1055 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
1056 ((rule->mark_mask || rule->mark) &&
1057 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
1058 (rule->target &&
1059 nla_put_u32(skb, FRA_GOTO, rule->target)) ||
1060 (rule->tun_id &&
1061 nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
1062 (rule->l3mdev &&
1063 nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
1064 (uid_range_set(&rule->uid_range) &&
1065 nla_put_uid_range(skb, &rule->uid_range)) ||
1066 (fib_rule_port_range_set(&rule->sport_range) &&
1067 nla_put_port_range(skb, FRA_SPORT_RANGE, &rule->sport_range)) ||
1068 (fib_rule_port_range_set(&rule->dport_range) &&
1069 nla_put_port_range(skb, FRA_DPORT_RANGE, &rule->dport_range)) ||
1070 (rule->ip_proto && nla_put_u8(skb, FRA_IP_PROTO, rule->ip_proto)))
1071 goto nla_put_failure;
1072
1073 if (rule->suppress_ifgroup != -1) {
1074 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
1075 goto nla_put_failure;
1076 }
1077
1078 if (ops->fill(rule, skb, frh) < 0)
1079 goto nla_put_failure;
1080
1081 nlmsg_end(skb, nlh);
1082 return 0;
1083
1084nla_put_failure:
1085 nlmsg_cancel(skb, nlh);
1086 return -EMSGSIZE;
1087}
1088
1089static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
1090 struct fib_rules_ops *ops)
1091{
1092 int idx = 0;
1093 struct fib_rule *rule;
1094 int err = 0;
1095
1096 rcu_read_lock();
1097 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
1098 if (idx < cb->args[1])
1099 goto skip;
1100
1101 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
1102 cb->nlh->nlmsg_seq, RTM_NEWRULE,
1103 NLM_F_MULTI, ops);
1104 if (err)
1105 break;
1106skip:
1107 idx++;
1108 }
1109 rcu_read_unlock();
1110 cb->args[1] = idx;
1111 rules_ops_put(ops);
1112
1113 return err;
1114}
1115
1116static int fib_valid_dumprule_req(const struct nlmsghdr *nlh,
1117 struct netlink_ext_ack *extack)
1118{
1119 struct fib_rule_hdr *frh;
1120
1121 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
1122 NL_SET_ERR_MSG(extack, "Invalid header for fib rule dump request");
1123 return -EINVAL;
1124 }
1125
1126 frh = nlmsg_data(nlh);
1127 if (frh->dst_len || frh->src_len || frh->tos || frh->table ||
1128 frh->res1 || frh->res2 || frh->action || frh->flags) {
1129 NL_SET_ERR_MSG(extack,
1130 "Invalid values in header for fib rule dump request");
1131 return -EINVAL;
1132 }
1133
1134 if (nlmsg_attrlen(nlh, sizeof(*frh))) {
1135 NL_SET_ERR_MSG(extack, "Invalid data after header in fib rule dump request");
1136 return -EINVAL;
1137 }
1138
1139 return 0;
1140}
1141
1142static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
1143{
1144 const struct nlmsghdr *nlh = cb->nlh;
1145 struct net *net = sock_net(skb->sk);
1146 struct fib_rules_ops *ops;
1147 int idx = 0, family;
1148
1149 if (cb->strict_check) {
1150 int err = fib_valid_dumprule_req(nlh, cb->extack);
1151
1152 if (err < 0)
1153 return err;
1154 }
1155
1156 family = rtnl_msg_family(nlh);
1157 if (family != AF_UNSPEC) {
1158 /* Protocol specific dump request */
1159 ops = lookup_rules_ops(net, family);
1160 if (ops == NULL)
1161 return -EAFNOSUPPORT;
1162
1163 dump_rules(skb, cb, ops);
1164
1165 return skb->len;
1166 }
1167
1168 rcu_read_lock();
1169 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
1170 if (idx < cb->args[0] || !try_module_get(ops->owner))
1171 goto skip;
1172
1173 if (dump_rules(skb, cb, ops) < 0)
1174 break;
1175
1176 cb->args[1] = 0;
1177skip:
1178 idx++;
1179 }
1180 rcu_read_unlock();
1181 cb->args[0] = idx;
1182
1183 return skb->len;
1184}
1185
1186static void notify_rule_change(int event, struct fib_rule *rule,
1187 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
1188 u32 pid)
1189{
1190 struct net *net;
1191 struct sk_buff *skb;
1192 int err = -ENOMEM;
1193
1194 net = ops->fro_net;
1195 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
1196 if (skb == NULL)
1197 goto errout;
1198
1199 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
1200 if (err < 0) {
1201 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
1202 WARN_ON(err == -EMSGSIZE);
1203 kfree_skb(skb);
1204 goto errout;
1205 }
1206
1207 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
1208 return;
1209errout:
1210 if (err < 0)
1211 rtnl_set_sk_err(net, ops->nlgroup, err);
1212}
1213
1214static void attach_rules(struct list_head *rules, struct net_device *dev)
1215{
1216 struct fib_rule *rule;
1217
1218 list_for_each_entry(rule, rules, list) {
1219 if (rule->iifindex == -1 &&
1220 strcmp(dev->name, rule->iifname) == 0)
1221 rule->iifindex = dev->ifindex;
1222 if (rule->oifindex == -1 &&
1223 strcmp(dev->name, rule->oifname) == 0)
1224 rule->oifindex = dev->ifindex;
1225 }
1226}
1227
1228static void detach_rules(struct list_head *rules, struct net_device *dev)
1229{
1230 struct fib_rule *rule;
1231
1232 list_for_each_entry(rule, rules, list) {
1233 if (rule->iifindex == dev->ifindex)
1234 rule->iifindex = -1;
1235 if (rule->oifindex == dev->ifindex)
1236 rule->oifindex = -1;
1237 }
1238}
1239
1240
1241static int fib_rules_event(struct notifier_block *this, unsigned long event,
1242 void *ptr)
1243{
1244 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1245 struct net *net = dev_net(dev);
1246 struct fib_rules_ops *ops;
1247
1248 ASSERT_RTNL();
1249
1250 switch (event) {
1251 case NETDEV_REGISTER:
1252 list_for_each_entry(ops, &net->rules_ops, list)
1253 attach_rules(&ops->rules_list, dev);
1254 break;
1255
1256 case NETDEV_CHANGENAME:
1257 list_for_each_entry(ops, &net->rules_ops, list) {
1258 detach_rules(&ops->rules_list, dev);
1259 attach_rules(&ops->rules_list, dev);
1260 }
1261 break;
1262
1263 case NETDEV_UNREGISTER:
1264 list_for_each_entry(ops, &net->rules_ops, list)
1265 detach_rules(&ops->rules_list, dev);
1266 break;
1267 }
1268
1269 return NOTIFY_DONE;
1270}
1271
1272static struct notifier_block fib_rules_notifier = {
1273 .notifier_call = fib_rules_event,
1274};
1275
1276static int __net_init fib_rules_net_init(struct net *net)
1277{
1278 INIT_LIST_HEAD(&net->rules_ops);
1279 spin_lock_init(&net->rules_mod_lock);
1280 return 0;
1281}
1282
1283static void __net_exit fib_rules_net_exit(struct net *net)
1284{
1285 WARN_ON_ONCE(!list_empty(&net->rules_ops));
1286}
1287
1288static struct pernet_operations fib_rules_net_ops = {
1289 .init = fib_rules_net_init,
1290 .exit = fib_rules_net_exit,
1291};
1292
1293static int __init fib_rules_init(void)
1294{
1295 int err;
1296 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, 0);
1297 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, 0);
1298 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, 0);
1299
1300 err = register_pernet_subsys(&fib_rules_net_ops);
1301 if (err < 0)
1302 goto fail;
1303
1304 err = register_netdevice_notifier(&fib_rules_notifier);
1305 if (err < 0)
1306 goto fail_unregister;
1307
1308 return 0;
1309
1310fail_unregister:
1311 unregister_pernet_subsys(&fib_rules_net_ops);
1312fail:
1313 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
1314 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
1315 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
1316 return err;
1317}
1318
1319subsys_initcall(fib_rules_init);
1/*
2 * net/core/fib_rules.c Generic Routing Rules
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License as
6 * published by the Free Software Foundation, version 2.
7 *
8 * Authors: Thomas Graf <tgraf@suug.ch>
9 */
10
11#include <linux/types.h>
12#include <linux/kernel.h>
13#include <linux/slab.h>
14#include <linux/list.h>
15#include <linux/module.h>
16#include <net/net_namespace.h>
17#include <net/sock.h>
18#include <net/fib_rules.h>
19
20int fib_default_rule_add(struct fib_rules_ops *ops,
21 u32 pref, u32 table, u32 flags)
22{
23 struct fib_rule *r;
24
25 r = kzalloc(ops->rule_size, GFP_KERNEL);
26 if (r == NULL)
27 return -ENOMEM;
28
29 atomic_set(&r->refcnt, 1);
30 r->action = FR_ACT_TO_TBL;
31 r->pref = pref;
32 r->table = table;
33 r->flags = flags;
34 r->fr_net = hold_net(ops->fro_net);
35
36 r->suppress_prefixlen = -1;
37 r->suppress_ifgroup = -1;
38
39 /* The lock is not required here, the list in unreacheable
40 * at the moment this function is called */
41 list_add_tail(&r->list, &ops->rules_list);
42 return 0;
43}
44EXPORT_SYMBOL(fib_default_rule_add);
45
46u32 fib_default_rule_pref(struct fib_rules_ops *ops)
47{
48 struct list_head *pos;
49 struct fib_rule *rule;
50
51 if (!list_empty(&ops->rules_list)) {
52 pos = ops->rules_list.next;
53 if (pos->next != &ops->rules_list) {
54 rule = list_entry(pos->next, struct fib_rule, list);
55 if (rule->pref)
56 return rule->pref - 1;
57 }
58 }
59
60 return 0;
61}
62EXPORT_SYMBOL(fib_default_rule_pref);
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 hold_net(net);
120 list_add_tail_rcu(&ops->list, &net->rules_ops);
121 err = 0;
122errout:
123 spin_unlock(&net->rules_mod_lock);
124
125 return err;
126}
127
128struct fib_rules_ops *
129fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
130{
131 struct fib_rules_ops *ops;
132 int err;
133
134 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
135 if (ops == NULL)
136 return ERR_PTR(-ENOMEM);
137
138 INIT_LIST_HEAD(&ops->rules_list);
139 ops->fro_net = net;
140
141 err = __fib_rules_register(ops);
142 if (err) {
143 kfree(ops);
144 ops = ERR_PTR(err);
145 }
146
147 return ops;
148}
149EXPORT_SYMBOL_GPL(fib_rules_register);
150
151static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
152{
153 struct fib_rule *rule, *tmp;
154
155 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
156 list_del_rcu(&rule->list);
157 if (ops->delete)
158 ops->delete(rule);
159 fib_rule_put(rule);
160 }
161}
162
163static void fib_rules_put_rcu(struct rcu_head *head)
164{
165 struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu);
166 struct net *net = ops->fro_net;
167
168 release_net(net);
169 kfree(ops);
170}
171
172void fib_rules_unregister(struct fib_rules_ops *ops)
173{
174 struct net *net = ops->fro_net;
175
176 spin_lock(&net->rules_mod_lock);
177 list_del_rcu(&ops->list);
178 fib_rules_cleanup_ops(ops);
179 spin_unlock(&net->rules_mod_lock);
180
181 call_rcu(&ops->rcu, fib_rules_put_rcu);
182}
183EXPORT_SYMBOL_GPL(fib_rules_unregister);
184
185static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
186 struct flowi *fl, int flags)
187{
188 int ret = 0;
189
190 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
191 goto out;
192
193 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
194 goto out;
195
196 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
197 goto out;
198
199 ret = ops->match(rule, fl, flags);
200out:
201 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
202}
203
204int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
205 int flags, struct fib_lookup_arg *arg)
206{
207 struct fib_rule *rule;
208 int err;
209
210 rcu_read_lock();
211
212 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
213jumped:
214 if (!fib_rule_match(rule, ops, fl, flags))
215 continue;
216
217 if (rule->action == FR_ACT_GOTO) {
218 struct fib_rule *target;
219
220 target = rcu_dereference(rule->ctarget);
221 if (target == NULL) {
222 continue;
223 } else {
224 rule = target;
225 goto jumped;
226 }
227 } else if (rule->action == FR_ACT_NOP)
228 continue;
229 else
230 err = ops->action(rule, fl, flags, arg);
231
232 if (!err && ops->suppress && ops->suppress(rule, arg))
233 continue;
234
235 if (err != -EAGAIN) {
236 if ((arg->flags & FIB_LOOKUP_NOREF) ||
237 likely(atomic_inc_not_zero(&rule->refcnt))) {
238 arg->rule = rule;
239 goto out;
240 }
241 break;
242 }
243 }
244
245 err = -ESRCH;
246out:
247 rcu_read_unlock();
248
249 return err;
250}
251EXPORT_SYMBOL_GPL(fib_rules_lookup);
252
253static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
254 struct fib_rules_ops *ops)
255{
256 int err = -EINVAL;
257
258 if (frh->src_len)
259 if (tb[FRA_SRC] == NULL ||
260 frh->src_len > (ops->addr_size * 8) ||
261 nla_len(tb[FRA_SRC]) != ops->addr_size)
262 goto errout;
263
264 if (frh->dst_len)
265 if (tb[FRA_DST] == NULL ||
266 frh->dst_len > (ops->addr_size * 8) ||
267 nla_len(tb[FRA_DST]) != ops->addr_size)
268 goto errout;
269
270 err = 0;
271errout:
272 return err;
273}
274
275static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh)
276{
277 struct net *net = sock_net(skb->sk);
278 struct fib_rule_hdr *frh = nlmsg_data(nlh);
279 struct fib_rules_ops *ops = NULL;
280 struct fib_rule *rule, *r, *last = NULL;
281 struct nlattr *tb[FRA_MAX+1];
282 int err = -EINVAL, unresolved = 0;
283
284 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
285 goto errout;
286
287 ops = lookup_rules_ops(net, frh->family);
288 if (ops == NULL) {
289 err = -EAFNOSUPPORT;
290 goto errout;
291 }
292
293 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
294 if (err < 0)
295 goto errout;
296
297 err = validate_rulemsg(frh, tb, ops);
298 if (err < 0)
299 goto errout;
300
301 rule = kzalloc(ops->rule_size, GFP_KERNEL);
302 if (rule == NULL) {
303 err = -ENOMEM;
304 goto errout;
305 }
306 rule->fr_net = hold_net(net);
307
308 if (tb[FRA_PRIORITY])
309 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
310
311 if (tb[FRA_IIFNAME]) {
312 struct net_device *dev;
313
314 rule->iifindex = -1;
315 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
316 dev = __dev_get_by_name(net, rule->iifname);
317 if (dev)
318 rule->iifindex = dev->ifindex;
319 }
320
321 if (tb[FRA_OIFNAME]) {
322 struct net_device *dev;
323
324 rule->oifindex = -1;
325 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
326 dev = __dev_get_by_name(net, rule->oifname);
327 if (dev)
328 rule->oifindex = dev->ifindex;
329 }
330
331 if (tb[FRA_FWMARK]) {
332 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
333 if (rule->mark)
334 /* compatibility: if the mark value is non-zero all bits
335 * are compared unless a mask is explicitly specified.
336 */
337 rule->mark_mask = 0xFFFFFFFF;
338 }
339
340 if (tb[FRA_FWMASK])
341 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
342
343 rule->action = frh->action;
344 rule->flags = frh->flags;
345 rule->table = frh_get_table(frh, tb);
346 if (tb[FRA_SUPPRESS_PREFIXLEN])
347 rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
348 else
349 rule->suppress_prefixlen = -1;
350
351 if (tb[FRA_SUPPRESS_IFGROUP])
352 rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
353 else
354 rule->suppress_ifgroup = -1;
355
356 if (!tb[FRA_PRIORITY] && ops->default_pref)
357 rule->pref = ops->default_pref(ops);
358
359 err = -EINVAL;
360 if (tb[FRA_GOTO]) {
361 if (rule->action != FR_ACT_GOTO)
362 goto errout_free;
363
364 rule->target = nla_get_u32(tb[FRA_GOTO]);
365 /* Backward jumps are prohibited to avoid endless loops */
366 if (rule->target <= rule->pref)
367 goto errout_free;
368
369 list_for_each_entry(r, &ops->rules_list, list) {
370 if (r->pref == rule->target) {
371 RCU_INIT_POINTER(rule->ctarget, r);
372 break;
373 }
374 }
375
376 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
377 unresolved = 1;
378 } else if (rule->action == FR_ACT_GOTO)
379 goto errout_free;
380
381 err = ops->configure(rule, skb, frh, tb);
382 if (err < 0)
383 goto errout_free;
384
385 list_for_each_entry(r, &ops->rules_list, list) {
386 if (r->pref > rule->pref)
387 break;
388 last = r;
389 }
390
391 fib_rule_get(rule);
392
393 if (last)
394 list_add_rcu(&rule->list, &last->list);
395 else
396 list_add_rcu(&rule->list, &ops->rules_list);
397
398 if (ops->unresolved_rules) {
399 /*
400 * There are unresolved goto rules in the list, check if
401 * any of them are pointing to this new rule.
402 */
403 list_for_each_entry(r, &ops->rules_list, list) {
404 if (r->action == FR_ACT_GOTO &&
405 r->target == rule->pref &&
406 rtnl_dereference(r->ctarget) == NULL) {
407 rcu_assign_pointer(r->ctarget, rule);
408 if (--ops->unresolved_rules == 0)
409 break;
410 }
411 }
412 }
413
414 if (rule->action == FR_ACT_GOTO)
415 ops->nr_goto_rules++;
416
417 if (unresolved)
418 ops->unresolved_rules++;
419
420 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
421 flush_route_cache(ops);
422 rules_ops_put(ops);
423 return 0;
424
425errout_free:
426 release_net(rule->fr_net);
427 kfree(rule);
428errout:
429 rules_ops_put(ops);
430 return err;
431}
432
433static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh)
434{
435 struct net *net = sock_net(skb->sk);
436 struct fib_rule_hdr *frh = nlmsg_data(nlh);
437 struct fib_rules_ops *ops = NULL;
438 struct fib_rule *rule, *tmp;
439 struct nlattr *tb[FRA_MAX+1];
440 int err = -EINVAL;
441
442 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
443 goto errout;
444
445 ops = lookup_rules_ops(net, frh->family);
446 if (ops == NULL) {
447 err = -EAFNOSUPPORT;
448 goto errout;
449 }
450
451 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
452 if (err < 0)
453 goto errout;
454
455 err = validate_rulemsg(frh, tb, ops);
456 if (err < 0)
457 goto errout;
458
459 list_for_each_entry(rule, &ops->rules_list, list) {
460 if (frh->action && (frh->action != rule->action))
461 continue;
462
463 if (frh_get_table(frh, tb) &&
464 (frh_get_table(frh, tb) != rule->table))
465 continue;
466
467 if (tb[FRA_PRIORITY] &&
468 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
469 continue;
470
471 if (tb[FRA_IIFNAME] &&
472 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
473 continue;
474
475 if (tb[FRA_OIFNAME] &&
476 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
477 continue;
478
479 if (tb[FRA_FWMARK] &&
480 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
481 continue;
482
483 if (tb[FRA_FWMASK] &&
484 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
485 continue;
486
487 if (!ops->compare(rule, frh, tb))
488 continue;
489
490 if (rule->flags & FIB_RULE_PERMANENT) {
491 err = -EPERM;
492 goto errout;
493 }
494
495 list_del_rcu(&rule->list);
496
497 if (rule->action == FR_ACT_GOTO) {
498 ops->nr_goto_rules--;
499 if (rtnl_dereference(rule->ctarget) == NULL)
500 ops->unresolved_rules--;
501 }
502
503 /*
504 * Check if this rule is a target to any of them. If so,
505 * disable them. As this operation is eventually very
506 * expensive, it is only performed if goto rules have
507 * actually been added.
508 */
509 if (ops->nr_goto_rules > 0) {
510 list_for_each_entry(tmp, &ops->rules_list, list) {
511 if (rtnl_dereference(tmp->ctarget) == rule) {
512 RCU_INIT_POINTER(tmp->ctarget, NULL);
513 ops->unresolved_rules++;
514 }
515 }
516 }
517
518 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
519 NETLINK_CB(skb).portid);
520 if (ops->delete)
521 ops->delete(rule);
522 fib_rule_put(rule);
523 flush_route_cache(ops);
524 rules_ops_put(ops);
525 return 0;
526 }
527
528 err = -ENOENT;
529errout:
530 rules_ops_put(ops);
531 return err;
532}
533
534static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
535 struct fib_rule *rule)
536{
537 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
538 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
539 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
540 + nla_total_size(4) /* FRA_PRIORITY */
541 + nla_total_size(4) /* FRA_TABLE */
542 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
543 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
544 + nla_total_size(4) /* FRA_FWMARK */
545 + nla_total_size(4); /* FRA_FWMASK */
546
547 if (ops->nlmsg_payload)
548 payload += ops->nlmsg_payload(rule);
549
550 return payload;
551}
552
553static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
554 u32 pid, u32 seq, int type, int flags,
555 struct fib_rules_ops *ops)
556{
557 struct nlmsghdr *nlh;
558 struct fib_rule_hdr *frh;
559
560 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
561 if (nlh == NULL)
562 return -EMSGSIZE;
563
564 frh = nlmsg_data(nlh);
565 frh->family = ops->family;
566 frh->table = rule->table;
567 if (nla_put_u32(skb, FRA_TABLE, rule->table))
568 goto nla_put_failure;
569 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
570 goto nla_put_failure;
571 frh->res1 = 0;
572 frh->res2 = 0;
573 frh->action = rule->action;
574 frh->flags = rule->flags;
575
576 if (rule->action == FR_ACT_GOTO &&
577 rcu_access_pointer(rule->ctarget) == NULL)
578 frh->flags |= FIB_RULE_UNRESOLVED;
579
580 if (rule->iifname[0]) {
581 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
582 goto nla_put_failure;
583 if (rule->iifindex == -1)
584 frh->flags |= FIB_RULE_IIF_DETACHED;
585 }
586
587 if (rule->oifname[0]) {
588 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
589 goto nla_put_failure;
590 if (rule->oifindex == -1)
591 frh->flags |= FIB_RULE_OIF_DETACHED;
592 }
593
594 if ((rule->pref &&
595 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
596 (rule->mark &&
597 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
598 ((rule->mark_mask || rule->mark) &&
599 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
600 (rule->target &&
601 nla_put_u32(skb, FRA_GOTO, rule->target)))
602 goto nla_put_failure;
603
604 if (rule->suppress_ifgroup != -1) {
605 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
606 goto nla_put_failure;
607 }
608
609 if (ops->fill(rule, skb, frh) < 0)
610 goto nla_put_failure;
611
612 return nlmsg_end(skb, nlh);
613
614nla_put_failure:
615 nlmsg_cancel(skb, nlh);
616 return -EMSGSIZE;
617}
618
619static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
620 struct fib_rules_ops *ops)
621{
622 int idx = 0;
623 struct fib_rule *rule;
624
625 rcu_read_lock();
626 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
627 if (idx < cb->args[1])
628 goto skip;
629
630 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
631 cb->nlh->nlmsg_seq, RTM_NEWRULE,
632 NLM_F_MULTI, ops) < 0)
633 break;
634skip:
635 idx++;
636 }
637 rcu_read_unlock();
638 cb->args[1] = idx;
639 rules_ops_put(ops);
640
641 return skb->len;
642}
643
644static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
645{
646 struct net *net = sock_net(skb->sk);
647 struct fib_rules_ops *ops;
648 int idx = 0, family;
649
650 family = rtnl_msg_family(cb->nlh);
651 if (family != AF_UNSPEC) {
652 /* Protocol specific dump request */
653 ops = lookup_rules_ops(net, family);
654 if (ops == NULL)
655 return -EAFNOSUPPORT;
656
657 return dump_rules(skb, cb, ops);
658 }
659
660 rcu_read_lock();
661 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
662 if (idx < cb->args[0] || !try_module_get(ops->owner))
663 goto skip;
664
665 if (dump_rules(skb, cb, ops) < 0)
666 break;
667
668 cb->args[1] = 0;
669skip:
670 idx++;
671 }
672 rcu_read_unlock();
673 cb->args[0] = idx;
674
675 return skb->len;
676}
677
678static void notify_rule_change(int event, struct fib_rule *rule,
679 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
680 u32 pid)
681{
682 struct net *net;
683 struct sk_buff *skb;
684 int err = -ENOBUFS;
685
686 net = ops->fro_net;
687 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
688 if (skb == NULL)
689 goto errout;
690
691 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
692 if (err < 0) {
693 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
694 WARN_ON(err == -EMSGSIZE);
695 kfree_skb(skb);
696 goto errout;
697 }
698
699 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
700 return;
701errout:
702 if (err < 0)
703 rtnl_set_sk_err(net, ops->nlgroup, err);
704}
705
706static void attach_rules(struct list_head *rules, struct net_device *dev)
707{
708 struct fib_rule *rule;
709
710 list_for_each_entry(rule, rules, list) {
711 if (rule->iifindex == -1 &&
712 strcmp(dev->name, rule->iifname) == 0)
713 rule->iifindex = dev->ifindex;
714 if (rule->oifindex == -1 &&
715 strcmp(dev->name, rule->oifname) == 0)
716 rule->oifindex = dev->ifindex;
717 }
718}
719
720static void detach_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 == dev->ifindex)
726 rule->iifindex = -1;
727 if (rule->oifindex == dev->ifindex)
728 rule->oifindex = -1;
729 }
730}
731
732
733static int fib_rules_event(struct notifier_block *this, unsigned long event,
734 void *ptr)
735{
736 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
737 struct net *net = dev_net(dev);
738 struct fib_rules_ops *ops;
739
740 ASSERT_RTNL();
741
742 switch (event) {
743 case NETDEV_REGISTER:
744 list_for_each_entry(ops, &net->rules_ops, list)
745 attach_rules(&ops->rules_list, dev);
746 break;
747
748 case NETDEV_CHANGENAME:
749 list_for_each_entry(ops, &net->rules_ops, list) {
750 detach_rules(&ops->rules_list, dev);
751 attach_rules(&ops->rules_list, dev);
752 }
753 break;
754
755 case NETDEV_UNREGISTER:
756 list_for_each_entry(ops, &net->rules_ops, list)
757 detach_rules(&ops->rules_list, dev);
758 break;
759 }
760
761 return NOTIFY_DONE;
762}
763
764static struct notifier_block fib_rules_notifier = {
765 .notifier_call = fib_rules_event,
766};
767
768static int __net_init fib_rules_net_init(struct net *net)
769{
770 INIT_LIST_HEAD(&net->rules_ops);
771 spin_lock_init(&net->rules_mod_lock);
772 return 0;
773}
774
775static struct pernet_operations fib_rules_net_ops = {
776 .init = fib_rules_net_init,
777};
778
779static int __init fib_rules_init(void)
780{
781 int err;
782 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
783 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
784 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
785
786 err = register_pernet_subsys(&fib_rules_net_ops);
787 if (err < 0)
788 goto fail;
789
790 err = register_netdevice_notifier(&fib_rules_notifier);
791 if (err < 0)
792 goto fail_unregister;
793
794 return 0;
795
796fail_unregister:
797 unregister_pernet_subsys(&fib_rules_net_ops);
798fail:
799 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
800 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
801 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
802 return err;
803}
804
805subsys_initcall(fib_rules_init);