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 <net/net_namespace.h>
17 #include <net/fib_rules.h>
19 int fib_default_rule_add(struct fib_rules_ops *ops,
20 u32 pref, u32 table, u32 flags)
24 r = kzalloc(ops->rule_size, GFP_KERNEL);
28 atomic_set(&r->refcnt, 1);
29 r->action = FR_ACT_TO_TBL;
33 r->fr_net = hold_net(ops->fro_net);
35 /* The lock is not required here, the list in unreacheable
36 * at the moment this function is called */
37 list_add_tail(&r->list, &ops->rules_list);
40 EXPORT_SYMBOL(fib_default_rule_add);
42 u32 fib_default_rule_pref(struct fib_rules_ops *ops)
44 struct list_head *pos;
45 struct fib_rule *rule;
47 if (!list_empty(&ops->rules_list)) {
48 pos = ops->rules_list.next;
49 if (pos->next != &ops->rules_list) {
50 rule = list_entry(pos->next, struct fib_rule, list);
52 return rule->pref - 1;
58 EXPORT_SYMBOL(fib_default_rule_pref);
60 static void notify_rule_change(int event, struct fib_rule *rule,
61 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
64 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
66 struct fib_rules_ops *ops;
69 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
70 if (ops->family == family) {
71 if (!try_module_get(ops->owner))
82 static void rules_ops_put(struct fib_rules_ops *ops)
85 module_put(ops->owner);
88 static void flush_route_cache(struct fib_rules_ops *ops)
91 ops->flush_cache(ops);
94 static int __fib_rules_register(struct fib_rules_ops *ops)
97 struct fib_rules_ops *o;
102 if (ops->rule_size < sizeof(struct fib_rule))
105 if (ops->match == NULL || ops->configure == NULL ||
106 ops->compare == NULL || ops->fill == NULL ||
110 spin_lock(&net->rules_mod_lock);
111 list_for_each_entry(o, &net->rules_ops, list)
112 if (ops->family == o->family)
116 list_add_tail_rcu(&ops->list, &net->rules_ops);
119 spin_unlock(&net->rules_mod_lock);
124 struct fib_rules_ops *
125 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
127 struct fib_rules_ops *ops;
130 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
132 return ERR_PTR(-ENOMEM);
134 INIT_LIST_HEAD(&ops->rules_list);
137 err = __fib_rules_register(ops);
145 EXPORT_SYMBOL_GPL(fib_rules_register);
147 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
149 struct fib_rule *rule, *tmp;
151 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
152 list_del_rcu(&rule->list);
157 static void fib_rules_put_rcu(struct rcu_head *head)
159 struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu);
160 struct net *net = ops->fro_net;
166 void fib_rules_unregister(struct fib_rules_ops *ops)
168 struct net *net = ops->fro_net;
170 spin_lock(&net->rules_mod_lock);
171 list_del_rcu(&ops->list);
172 fib_rules_cleanup_ops(ops);
173 spin_unlock(&net->rules_mod_lock);
175 call_rcu(&ops->rcu, fib_rules_put_rcu);
177 EXPORT_SYMBOL_GPL(fib_rules_unregister);
179 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
180 struct flowi *fl, int flags)
184 if (rule->iifindex && (rule->iifindex != fl->iif) &&
185 !(fl->flags & FLOWI_FLAG_MATCH_ANY_IIF))
188 if (rule->oifindex && (rule->oifindex != fl->oif))
191 if ((rule->mark ^ fl->mark) & rule->mark_mask)
194 ret = ops->match(rule, fl, flags);
196 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
199 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
200 int flags, struct fib_lookup_arg *arg)
202 struct fib_rule *rule;
207 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
209 if (!fib_rule_match(rule, ops, fl, flags))
212 if (rule->action == FR_ACT_GOTO) {
213 struct fib_rule *target;
215 target = rcu_dereference(rule->ctarget);
216 if (target == NULL) {
222 } else if (rule->action == FR_ACT_NOP)
225 err = ops->action(rule, fl, flags, arg);
227 if (err != -EAGAIN) {
228 if ((arg->flags & FIB_LOOKUP_NOREF) ||
229 likely(atomic_inc_not_zero(&rule->refcnt))) {
243 EXPORT_SYMBOL_GPL(fib_rules_lookup);
245 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
246 struct fib_rules_ops *ops)
251 if (tb[FRA_SRC] == NULL ||
252 frh->src_len > (ops->addr_size * 8) ||
253 nla_len(tb[FRA_SRC]) != ops->addr_size)
257 if (tb[FRA_DST] == NULL ||
258 frh->dst_len > (ops->addr_size * 8) ||
259 nla_len(tb[FRA_DST]) != ops->addr_size)
267 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
269 struct net *net = sock_net(skb->sk);
270 struct fib_rule_hdr *frh = nlmsg_data(nlh);
271 struct fib_rules_ops *ops = NULL;
272 struct fib_rule *rule, *r, *last = NULL;
273 struct nlattr *tb[FRA_MAX+1];
274 int err = -EINVAL, unresolved = 0;
276 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
279 ops = lookup_rules_ops(net, frh->family);
285 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
289 err = validate_rulemsg(frh, tb, ops);
293 rule = kzalloc(ops->rule_size, GFP_KERNEL);
298 rule->fr_net = hold_net(net);
300 if (tb[FRA_PRIORITY])
301 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
303 if (tb[FRA_IIFNAME]) {
304 struct net_device *dev;
307 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
308 dev = __dev_get_by_name(net, rule->iifname);
310 rule->iifindex = dev->ifindex;
313 if (tb[FRA_OIFNAME]) {
314 struct net_device *dev;
317 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
318 dev = __dev_get_by_name(net, rule->oifname);
320 rule->oifindex = dev->ifindex;
323 if (tb[FRA_FWMARK]) {
324 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
326 /* compatibility: if the mark value is non-zero all bits
327 * are compared unless a mask is explicitly specified.
329 rule->mark_mask = 0xFFFFFFFF;
333 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
335 rule->action = frh->action;
336 rule->flags = frh->flags;
337 rule->table = frh_get_table(frh, tb);
339 if (!tb[FRA_PRIORITY] && ops->default_pref)
340 rule->pref = ops->default_pref(ops);
344 if (rule->action != FR_ACT_GOTO)
347 rule->target = nla_get_u32(tb[FRA_GOTO]);
348 /* Backward jumps are prohibited to avoid endless loops */
349 if (rule->target <= rule->pref)
352 list_for_each_entry(r, &ops->rules_list, list) {
353 if (r->pref == rule->target) {
359 if (rule->ctarget == NULL)
361 } else if (rule->action == FR_ACT_GOTO)
364 err = ops->configure(rule, skb, frh, tb);
368 list_for_each_entry(r, &ops->rules_list, list) {
369 if (r->pref > rule->pref)
376 if (ops->unresolved_rules) {
378 * There are unresolved goto rules in the list, check if
379 * any of them are pointing to this new rule.
381 list_for_each_entry(r, &ops->rules_list, list) {
382 if (r->action == FR_ACT_GOTO &&
383 r->target == rule->pref) {
384 BUG_ON(r->ctarget != NULL);
385 rcu_assign_pointer(r->ctarget, rule);
386 if (--ops->unresolved_rules == 0)
392 if (rule->action == FR_ACT_GOTO)
393 ops->nr_goto_rules++;
396 ops->unresolved_rules++;
399 list_add_rcu(&rule->list, &last->list);
401 list_add_rcu(&rule->list, &ops->rules_list);
403 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
404 flush_route_cache(ops);
409 release_net(rule->fr_net);
416 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
418 struct net *net = sock_net(skb->sk);
419 struct fib_rule_hdr *frh = nlmsg_data(nlh);
420 struct fib_rules_ops *ops = NULL;
421 struct fib_rule *rule, *tmp;
422 struct nlattr *tb[FRA_MAX+1];
425 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
428 ops = lookup_rules_ops(net, frh->family);
434 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
438 err = validate_rulemsg(frh, tb, ops);
442 list_for_each_entry(rule, &ops->rules_list, list) {
443 if (frh->action && (frh->action != rule->action))
446 if (frh->table && (frh_get_table(frh, tb) != rule->table))
449 if (tb[FRA_PRIORITY] &&
450 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
453 if (tb[FRA_IIFNAME] &&
454 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
457 if (tb[FRA_OIFNAME] &&
458 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
461 if (tb[FRA_FWMARK] &&
462 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
465 if (tb[FRA_FWMASK] &&
466 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
469 if (!ops->compare(rule, frh, tb))
472 if (rule->flags & FIB_RULE_PERMANENT) {
477 list_del_rcu(&rule->list);
479 if (rule->action == FR_ACT_GOTO)
480 ops->nr_goto_rules--;
483 * Check if this rule is a target to any of them. If so,
484 * disable them. As this operation is eventually very
485 * expensive, it is only performed if goto rules have
486 * actually been added.
488 if (ops->nr_goto_rules > 0) {
489 list_for_each_entry(tmp, &ops->rules_list, list) {
490 if (tmp->ctarget == rule) {
491 rcu_assign_pointer(tmp->ctarget, NULL);
492 ops->unresolved_rules++;
497 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
498 NETLINK_CB(skb).pid);
500 flush_route_cache(ops);
511 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
512 struct fib_rule *rule)
514 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
515 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
516 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
517 + nla_total_size(4) /* FRA_PRIORITY */
518 + nla_total_size(4) /* FRA_TABLE */
519 + nla_total_size(4) /* FRA_FWMARK */
520 + nla_total_size(4); /* FRA_FWMASK */
522 if (ops->nlmsg_payload)
523 payload += ops->nlmsg_payload(rule);
528 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
529 u32 pid, u32 seq, int type, int flags,
530 struct fib_rules_ops *ops)
532 struct nlmsghdr *nlh;
533 struct fib_rule_hdr *frh;
535 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
539 frh = nlmsg_data(nlh);
540 frh->family = ops->family;
541 frh->table = rule->table;
542 NLA_PUT_U32(skb, FRA_TABLE, rule->table);
545 frh->action = rule->action;
546 frh->flags = rule->flags;
548 if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
549 frh->flags |= FIB_RULE_UNRESOLVED;
551 if (rule->iifname[0]) {
552 NLA_PUT_STRING(skb, FRA_IIFNAME, rule->iifname);
554 if (rule->iifindex == -1)
555 frh->flags |= FIB_RULE_IIF_DETACHED;
558 if (rule->oifname[0]) {
559 NLA_PUT_STRING(skb, FRA_OIFNAME, rule->oifname);
561 if (rule->oifindex == -1)
562 frh->flags |= FIB_RULE_OIF_DETACHED;
566 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
569 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
571 if (rule->mark_mask || rule->mark)
572 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
575 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
577 if (ops->fill(rule, skb, frh) < 0)
578 goto nla_put_failure;
580 return nlmsg_end(skb, nlh);
583 nlmsg_cancel(skb, nlh);
587 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
588 struct fib_rules_ops *ops)
591 struct fib_rule *rule;
593 list_for_each_entry(rule, &ops->rules_list, list) {
594 if (idx < cb->args[1])
597 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
598 cb->nlh->nlmsg_seq, RTM_NEWRULE,
599 NLM_F_MULTI, ops) < 0)
610 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
612 struct net *net = sock_net(skb->sk);
613 struct fib_rules_ops *ops;
616 family = rtnl_msg_family(cb->nlh);
617 if (family != AF_UNSPEC) {
618 /* Protocol specific dump request */
619 ops = lookup_rules_ops(net, family);
621 return -EAFNOSUPPORT;
623 return dump_rules(skb, cb, ops);
627 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
628 if (idx < cb->args[0] || !try_module_get(ops->owner))
631 if (dump_rules(skb, cb, ops) < 0)
644 static void notify_rule_change(int event, struct fib_rule *rule,
645 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
653 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
657 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
659 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
660 WARN_ON(err == -EMSGSIZE);
665 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
669 rtnl_set_sk_err(net, ops->nlgroup, err);
672 static void attach_rules(struct list_head *rules, struct net_device *dev)
674 struct fib_rule *rule;
676 list_for_each_entry(rule, rules, list) {
677 if (rule->iifindex == -1 &&
678 strcmp(dev->name, rule->iifname) == 0)
679 rule->iifindex = dev->ifindex;
680 if (rule->oifindex == -1 &&
681 strcmp(dev->name, rule->oifname) == 0)
682 rule->oifindex = dev->ifindex;
686 static void detach_rules(struct list_head *rules, struct net_device *dev)
688 struct fib_rule *rule;
690 list_for_each_entry(rule, rules, list) {
691 if (rule->iifindex == dev->ifindex)
693 if (rule->oifindex == dev->ifindex)
699 static int fib_rules_event(struct notifier_block *this, unsigned long event,
702 struct net_device *dev = ptr;
703 struct net *net = dev_net(dev);
704 struct fib_rules_ops *ops;
709 case NETDEV_REGISTER:
710 list_for_each_entry(ops, &net->rules_ops, list)
711 attach_rules(&ops->rules_list, dev);
714 case NETDEV_UNREGISTER:
715 list_for_each_entry(ops, &net->rules_ops, list)
716 detach_rules(&ops->rules_list, dev);
723 static struct notifier_block fib_rules_notifier = {
724 .notifier_call = fib_rules_event,
727 static int __net_init fib_rules_net_init(struct net *net)
729 INIT_LIST_HEAD(&net->rules_ops);
730 spin_lock_init(&net->rules_mod_lock);
734 static struct pernet_operations fib_rules_net_ops = {
735 .init = fib_rules_net_init,
738 static int __init fib_rules_init(void)
741 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
742 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
743 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
745 err = register_pernet_subsys(&fib_rules_net_ops);
749 err = register_netdevice_notifier(&fib_rules_notifier);
751 goto fail_unregister;
756 unregister_pernet_subsys(&fib_rules_net_ops);
758 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
759 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
760 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
764 subsys_initcall(fib_rules_init);