2 * net/core/fib_rules.c Generic Routing Rules
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.
8 * Authors: Thomas Graf <tgraf@suug.ch>
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>
18 #include <net/fib_rules.h>
19 #include <net/ip_tunnels.h>
21 static const struct fib_kuid_range fib_kuid_range_unset = {
26 int fib_default_rule_add(struct fib_rules_ops *ops,
27 u32 pref, u32 table, u32 flags)
31 r = kzalloc(ops->rule_size, GFP_KERNEL);
35 atomic_set(&r->refcnt, 1);
36 r->action = FR_ACT_TO_TBL;
40 r->fr_net = ops->fro_net;
41 r->uid_range = fib_kuid_range_unset;
43 r->suppress_prefixlen = -1;
44 r->suppress_ifgroup = -1;
46 /* The lock is not required here, the list in unreacheable
47 * at the moment this function is called */
48 list_add_tail(&r->list, &ops->rules_list);
51 EXPORT_SYMBOL(fib_default_rule_add);
53 static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
55 struct list_head *pos;
56 struct fib_rule *rule;
58 if (!list_empty(&ops->rules_list)) {
59 pos = ops->rules_list.next;
60 if (pos->next != &ops->rules_list) {
61 rule = list_entry(pos->next, struct fib_rule, list);
63 return rule->pref - 1;
70 static void notify_rule_change(int event, struct fib_rule *rule,
71 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
74 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
76 struct fib_rules_ops *ops;
79 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
80 if (ops->family == family) {
81 if (!try_module_get(ops->owner))
92 static void rules_ops_put(struct fib_rules_ops *ops)
95 module_put(ops->owner);
98 static void flush_route_cache(struct fib_rules_ops *ops)
100 if (ops->flush_cache)
101 ops->flush_cache(ops);
104 static int __fib_rules_register(struct fib_rules_ops *ops)
107 struct fib_rules_ops *o;
112 if (ops->rule_size < sizeof(struct fib_rule))
115 if (ops->match == NULL || ops->configure == NULL ||
116 ops->compare == NULL || ops->fill == NULL ||
120 spin_lock(&net->rules_mod_lock);
121 list_for_each_entry(o, &net->rules_ops, list)
122 if (ops->family == o->family)
125 list_add_tail_rcu(&ops->list, &net->rules_ops);
128 spin_unlock(&net->rules_mod_lock);
133 struct fib_rules_ops *
134 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
136 struct fib_rules_ops *ops;
139 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
141 return ERR_PTR(-ENOMEM);
143 INIT_LIST_HEAD(&ops->rules_list);
146 err = __fib_rules_register(ops);
154 EXPORT_SYMBOL_GPL(fib_rules_register);
156 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
158 struct fib_rule *rule, *tmp;
160 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
161 list_del_rcu(&rule->list);
168 void fib_rules_unregister(struct fib_rules_ops *ops)
170 struct net *net = ops->fro_net;
172 spin_lock(&net->rules_mod_lock);
173 list_del_rcu(&ops->list);
174 spin_unlock(&net->rules_mod_lock);
176 fib_rules_cleanup_ops(ops);
179 EXPORT_SYMBOL_GPL(fib_rules_unregister);
181 static int uid_range_set(struct fib_kuid_range *range)
183 return uid_valid(range->start) && uid_valid(range->end);
186 static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
188 struct fib_rule_uid_range *in;
189 struct fib_kuid_range out;
191 in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
193 out.start = make_kuid(current_user_ns(), in->start);
194 out.end = make_kuid(current_user_ns(), in->end);
199 static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
201 struct fib_rule_uid_range out = {
202 from_kuid_munged(current_user_ns(), range->start),
203 from_kuid_munged(current_user_ns(), range->end)
206 return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
209 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
210 struct flowi *fl, int flags,
211 struct fib_lookup_arg *arg)
215 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
218 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
221 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
224 if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
227 if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
230 if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
231 uid_gt(fl->flowi_uid, rule->uid_range.end))
234 ret = ops->match(rule, fl, flags);
236 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
239 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
240 int flags, struct fib_lookup_arg *arg)
242 struct fib_rule *rule;
247 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
249 if (!fib_rule_match(rule, ops, fl, flags, arg))
252 if (rule->action == FR_ACT_GOTO) {
253 struct fib_rule *target;
255 target = rcu_dereference(rule->ctarget);
256 if (target == NULL) {
262 } else if (rule->action == FR_ACT_NOP)
265 err = ops->action(rule, fl, flags, arg);
267 if (!err && ops->suppress && ops->suppress(rule, arg))
270 if (err != -EAGAIN) {
271 if ((arg->flags & FIB_LOOKUP_NOREF) ||
272 likely(atomic_inc_not_zero(&rule->refcnt))) {
286 EXPORT_SYMBOL_GPL(fib_rules_lookup);
288 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
289 struct fib_rules_ops *ops)
294 if (tb[FRA_SRC] == NULL ||
295 frh->src_len > (ops->addr_size * 8) ||
296 nla_len(tb[FRA_SRC]) != ops->addr_size)
300 if (tb[FRA_DST] == NULL ||
301 frh->dst_len > (ops->addr_size * 8) ||
302 nla_len(tb[FRA_DST]) != ops->addr_size)
310 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
311 struct nlattr **tb, struct fib_rule *rule)
315 list_for_each_entry(r, &ops->rules_list, list) {
316 if (r->action != rule->action)
319 if (r->table != rule->table)
322 if (r->pref != rule->pref)
325 if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
328 if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
331 if (r->mark != rule->mark)
334 if (r->mark_mask != rule->mark_mask)
337 if (r->tun_id != rule->tun_id)
340 if (r->fr_net != rule->fr_net)
343 if (r->l3mdev != rule->l3mdev)
346 if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
347 !uid_eq(r->uid_range.end, rule->uid_range.end))
350 if (!ops->compare(r, frh, tb))
357 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh)
359 struct net *net = sock_net(skb->sk);
360 struct fib_rule_hdr *frh = nlmsg_data(nlh);
361 struct fib_rules_ops *ops = NULL;
362 struct fib_rule *rule, *r, *last = NULL;
363 struct nlattr *tb[FRA_MAX+1];
364 int err = -EINVAL, unresolved = 0;
366 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
369 ops = lookup_rules_ops(net, frh->family);
375 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
379 err = validate_rulemsg(frh, tb, ops);
383 rule = kzalloc(ops->rule_size, GFP_KERNEL);
390 rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
391 : fib_default_rule_pref(ops);
393 if (tb[FRA_IIFNAME]) {
394 struct net_device *dev;
397 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
398 dev = __dev_get_by_name(net, rule->iifname);
400 rule->iifindex = dev->ifindex;
403 if (tb[FRA_OIFNAME]) {
404 struct net_device *dev;
407 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
408 dev = __dev_get_by_name(net, rule->oifname);
410 rule->oifindex = dev->ifindex;
413 if (tb[FRA_FWMARK]) {
414 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
416 /* compatibility: if the mark value is non-zero all bits
417 * are compared unless a mask is explicitly specified.
419 rule->mark_mask = 0xFFFFFFFF;
423 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
426 rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
428 if (tb[FRA_L3MDEV]) {
429 #ifdef CONFIG_NET_L3_MASTER_DEV
430 rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
431 if (rule->l3mdev != 1)
436 rule->action = frh->action;
437 rule->flags = frh->flags;
438 rule->table = frh_get_table(frh, tb);
439 if (tb[FRA_SUPPRESS_PREFIXLEN])
440 rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
442 rule->suppress_prefixlen = -1;
444 if (tb[FRA_SUPPRESS_IFGROUP])
445 rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
447 rule->suppress_ifgroup = -1;
451 if (rule->action != FR_ACT_GOTO)
454 rule->target = nla_get_u32(tb[FRA_GOTO]);
455 /* Backward jumps are prohibited to avoid endless loops */
456 if (rule->target <= rule->pref)
459 list_for_each_entry(r, &ops->rules_list, list) {
460 if (r->pref == rule->target) {
461 RCU_INIT_POINTER(rule->ctarget, r);
466 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
468 } else if (rule->action == FR_ACT_GOTO)
471 if (rule->l3mdev && rule->table)
474 if (tb[FRA_UID_RANGE]) {
475 if (current_user_ns() != net->user_ns) {
480 rule->uid_range = nla_get_kuid_range(tb);
482 if (!uid_range_set(&rule->uid_range) ||
483 !uid_lte(rule->uid_range.start, rule->uid_range.end))
486 rule->uid_range = fib_kuid_range_unset;
489 if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
490 rule_exists(ops, frh, tb, rule)) {
495 err = ops->configure(rule, skb, frh, tb);
499 list_for_each_entry(r, &ops->rules_list, list) {
500 if (r->pref > rule->pref)
508 list_add_rcu(&rule->list, &last->list);
510 list_add_rcu(&rule->list, &ops->rules_list);
512 if (ops->unresolved_rules) {
514 * There are unresolved goto rules in the list, check if
515 * any of them are pointing to this new rule.
517 list_for_each_entry(r, &ops->rules_list, list) {
518 if (r->action == FR_ACT_GOTO &&
519 r->target == rule->pref &&
520 rtnl_dereference(r->ctarget) == NULL) {
521 rcu_assign_pointer(r->ctarget, rule);
522 if (--ops->unresolved_rules == 0)
528 if (rule->action == FR_ACT_GOTO)
529 ops->nr_goto_rules++;
532 ops->unresolved_rules++;
535 ip_tunnel_need_metadata();
537 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
538 flush_route_cache(ops);
548 EXPORT_SYMBOL_GPL(fib_nl_newrule);
550 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh)
552 struct net *net = sock_net(skb->sk);
553 struct fib_rule_hdr *frh = nlmsg_data(nlh);
554 struct fib_rules_ops *ops = NULL;
555 struct fib_rule *rule, *tmp;
556 struct nlattr *tb[FRA_MAX+1];
557 struct fib_kuid_range range;
560 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
563 ops = lookup_rules_ops(net, frh->family);
569 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
573 err = validate_rulemsg(frh, tb, ops);
577 if (tb[FRA_UID_RANGE]) {
578 range = nla_get_kuid_range(tb);
579 if (!uid_range_set(&range))
582 range = fib_kuid_range_unset;
585 list_for_each_entry(rule, &ops->rules_list, list) {
586 if (frh->action && (frh->action != rule->action))
589 if (frh_get_table(frh, tb) &&
590 (frh_get_table(frh, tb) != rule->table))
593 if (tb[FRA_PRIORITY] &&
594 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
597 if (tb[FRA_IIFNAME] &&
598 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
601 if (tb[FRA_OIFNAME] &&
602 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
605 if (tb[FRA_FWMARK] &&
606 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
609 if (tb[FRA_FWMASK] &&
610 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
613 if (tb[FRA_TUN_ID] &&
614 (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
617 if (tb[FRA_L3MDEV] &&
618 (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
621 if (uid_range_set(&range) &&
622 (!uid_eq(rule->uid_range.start, range.start) ||
623 !uid_eq(rule->uid_range.end, range.end)))
626 if (!ops->compare(rule, frh, tb))
629 if (rule->flags & FIB_RULE_PERMANENT) {
635 err = ops->delete(rule);
641 ip_tunnel_unneed_metadata();
643 list_del_rcu(&rule->list);
645 if (rule->action == FR_ACT_GOTO) {
646 ops->nr_goto_rules--;
647 if (rtnl_dereference(rule->ctarget) == NULL)
648 ops->unresolved_rules--;
652 * Check if this rule is a target to any of them. If so,
653 * disable them. As this operation is eventually very
654 * expensive, it is only performed if goto rules have
655 * actually been added.
657 if (ops->nr_goto_rules > 0) {
658 list_for_each_entry(tmp, &ops->rules_list, list) {
659 if (rtnl_dereference(tmp->ctarget) == rule) {
660 RCU_INIT_POINTER(tmp->ctarget, NULL);
661 ops->unresolved_rules++;
666 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
667 NETLINK_CB(skb).portid);
669 flush_route_cache(ops);
679 EXPORT_SYMBOL_GPL(fib_nl_delrule);
681 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
682 struct fib_rule *rule)
684 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
685 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
686 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
687 + nla_total_size(4) /* FRA_PRIORITY */
688 + nla_total_size(4) /* FRA_TABLE */
689 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
690 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
691 + nla_total_size(4) /* FRA_FWMARK */
692 + nla_total_size(4) /* FRA_FWMASK */
693 + nla_total_size_64bit(8) /* FRA_TUN_ID */
694 + nla_total_size(sizeof(struct fib_kuid_range));
696 if (ops->nlmsg_payload)
697 payload += ops->nlmsg_payload(rule);
702 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
703 u32 pid, u32 seq, int type, int flags,
704 struct fib_rules_ops *ops)
706 struct nlmsghdr *nlh;
707 struct fib_rule_hdr *frh;
709 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
713 frh = nlmsg_data(nlh);
714 frh->family = ops->family;
715 frh->table = rule->table;
716 if (nla_put_u32(skb, FRA_TABLE, rule->table))
717 goto nla_put_failure;
718 if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
719 goto nla_put_failure;
722 frh->action = rule->action;
723 frh->flags = rule->flags;
725 if (rule->action == FR_ACT_GOTO &&
726 rcu_access_pointer(rule->ctarget) == NULL)
727 frh->flags |= FIB_RULE_UNRESOLVED;
729 if (rule->iifname[0]) {
730 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
731 goto nla_put_failure;
732 if (rule->iifindex == -1)
733 frh->flags |= FIB_RULE_IIF_DETACHED;
736 if (rule->oifname[0]) {
737 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
738 goto nla_put_failure;
739 if (rule->oifindex == -1)
740 frh->flags |= FIB_RULE_OIF_DETACHED;
744 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
746 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
747 ((rule->mark_mask || rule->mark) &&
748 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
750 nla_put_u32(skb, FRA_GOTO, rule->target)) ||
752 nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
754 nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
755 (uid_range_set(&rule->uid_range) &&
756 nla_put_uid_range(skb, &rule->uid_range)))
757 goto nla_put_failure;
759 if (rule->suppress_ifgroup != -1) {
760 if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
761 goto nla_put_failure;
764 if (ops->fill(rule, skb, frh) < 0)
765 goto nla_put_failure;
771 nlmsg_cancel(skb, nlh);
775 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
776 struct fib_rules_ops *ops)
779 struct fib_rule *rule;
783 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
784 if (idx < cb->args[1])
787 err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
788 cb->nlh->nlmsg_seq, RTM_NEWRULE,
802 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
804 struct net *net = sock_net(skb->sk);
805 struct fib_rules_ops *ops;
808 family = rtnl_msg_family(cb->nlh);
809 if (family != AF_UNSPEC) {
810 /* Protocol specific dump request */
811 ops = lookup_rules_ops(net, family);
813 return -EAFNOSUPPORT;
815 dump_rules(skb, cb, ops);
821 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
822 if (idx < cb->args[0] || !try_module_get(ops->owner))
825 if (dump_rules(skb, cb, ops) < 0)
838 static void notify_rule_change(int event, struct fib_rule *rule,
839 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
847 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
851 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
853 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
854 WARN_ON(err == -EMSGSIZE);
859 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
863 rtnl_set_sk_err(net, ops->nlgroup, err);
866 static void attach_rules(struct list_head *rules, struct net_device *dev)
868 struct fib_rule *rule;
870 list_for_each_entry(rule, rules, list) {
871 if (rule->iifindex == -1 &&
872 strcmp(dev->name, rule->iifname) == 0)
873 rule->iifindex = dev->ifindex;
874 if (rule->oifindex == -1 &&
875 strcmp(dev->name, rule->oifname) == 0)
876 rule->oifindex = dev->ifindex;
880 static void detach_rules(struct list_head *rules, struct net_device *dev)
882 struct fib_rule *rule;
884 list_for_each_entry(rule, rules, list) {
885 if (rule->iifindex == dev->ifindex)
887 if (rule->oifindex == dev->ifindex)
893 static int fib_rules_event(struct notifier_block *this, unsigned long event,
896 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
897 struct net *net = dev_net(dev);
898 struct fib_rules_ops *ops;
903 case NETDEV_REGISTER:
904 list_for_each_entry(ops, &net->rules_ops, list)
905 attach_rules(&ops->rules_list, dev);
908 case NETDEV_CHANGENAME:
909 list_for_each_entry(ops, &net->rules_ops, list) {
910 detach_rules(&ops->rules_list, dev);
911 attach_rules(&ops->rules_list, dev);
915 case NETDEV_UNREGISTER:
916 list_for_each_entry(ops, &net->rules_ops, list)
917 detach_rules(&ops->rules_list, dev);
924 static struct notifier_block fib_rules_notifier = {
925 .notifier_call = fib_rules_event,
928 static int __net_init fib_rules_net_init(struct net *net)
930 INIT_LIST_HEAD(&net->rules_ops);
931 spin_lock_init(&net->rules_mod_lock);
935 static struct pernet_operations fib_rules_net_ops = {
936 .init = fib_rules_net_init,
939 static int __init fib_rules_init(void)
942 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
943 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
944 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
946 err = register_pernet_subsys(&fib_rules_net_ops);
950 err = register_netdevice_notifier(&fib_rules_notifier);
952 goto fail_unregister;
957 unregister_pernet_subsys(&fib_rules_net_ops);
959 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
960 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
961 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
965 subsys_initcall(fib_rules_init);