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/list.h>
14 #include <net/net_namespace.h>
16 #include <net/fib_rules.h>
18 int fib_default_rule_add(struct fib_rules_ops *ops,
19 u32 pref, u32 table, u32 flags)
23 r = kzalloc(ops->rule_size, GFP_KERNEL);
27 atomic_set(&r->refcnt, 1);
28 r->action = FR_ACT_TO_TBL;
32 r->fr_net = hold_net(ops->fro_net);
34 /* The lock is not required here, the list in unreacheable
35 * at the moment this function is called */
36 list_add_tail(&r->list, &ops->rules_list);
39 EXPORT_SYMBOL(fib_default_rule_add);
41 static void notify_rule_change(int event, struct fib_rule *rule,
42 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
45 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
47 struct fib_rules_ops *ops;
50 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
51 if (ops->family == family) {
52 if (!try_module_get(ops->owner))
63 static void rules_ops_put(struct fib_rules_ops *ops)
66 module_put(ops->owner);
69 static void flush_route_cache(struct fib_rules_ops *ops)
72 ops->flush_cache(ops);
75 int fib_rules_register(struct fib_rules_ops *ops)
78 struct fib_rules_ops *o;
83 if (ops->rule_size < sizeof(struct fib_rule))
86 if (ops->match == NULL || ops->configure == NULL ||
87 ops->compare == NULL || ops->fill == NULL ||
91 spin_lock(&net->rules_mod_lock);
92 list_for_each_entry(o, &net->rules_ops, list)
93 if (ops->family == o->family)
97 list_add_tail_rcu(&ops->list, &net->rules_ops);
100 spin_unlock(&net->rules_mod_lock);
105 EXPORT_SYMBOL_GPL(fib_rules_register);
107 void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
109 struct fib_rule *rule, *tmp;
111 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
112 list_del_rcu(&rule->list);
116 EXPORT_SYMBOL_GPL(fib_rules_cleanup_ops);
118 void fib_rules_unregister(struct fib_rules_ops *ops)
120 struct net *net = ops->fro_net;
122 spin_lock(&net->rules_mod_lock);
123 list_del_rcu(&ops->list);
124 fib_rules_cleanup_ops(ops);
125 spin_unlock(&net->rules_mod_lock);
131 EXPORT_SYMBOL_GPL(fib_rules_unregister);
133 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
134 struct flowi *fl, int flags)
138 if (rule->ifindex && (rule->ifindex != fl->iif))
141 if ((rule->mark ^ fl->mark) & rule->mark_mask)
144 ret = ops->match(rule, fl, flags);
146 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
149 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
150 int flags, struct fib_lookup_arg *arg)
152 struct fib_rule *rule;
157 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
159 if (!fib_rule_match(rule, ops, fl, flags))
162 if (rule->action == FR_ACT_GOTO) {
163 struct fib_rule *target;
165 target = rcu_dereference(rule->ctarget);
166 if (target == NULL) {
172 } else if (rule->action == FR_ACT_NOP)
175 err = ops->action(rule, fl, flags, arg);
177 if (err != -EAGAIN) {
191 EXPORT_SYMBOL_GPL(fib_rules_lookup);
193 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
194 struct fib_rules_ops *ops)
199 if (tb[FRA_SRC] == NULL ||
200 frh->src_len > (ops->addr_size * 8) ||
201 nla_len(tb[FRA_SRC]) != ops->addr_size)
205 if (tb[FRA_DST] == NULL ||
206 frh->dst_len > (ops->addr_size * 8) ||
207 nla_len(tb[FRA_DST]) != ops->addr_size)
215 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
217 struct net *net = sock_net(skb->sk);
218 struct fib_rule_hdr *frh = nlmsg_data(nlh);
219 struct fib_rules_ops *ops = NULL;
220 struct fib_rule *rule, *r, *last = NULL;
221 struct nlattr *tb[FRA_MAX+1];
222 int err = -EINVAL, unresolved = 0;
224 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
227 ops = lookup_rules_ops(net, frh->family);
233 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
237 err = validate_rulemsg(frh, tb, ops);
241 rule = kzalloc(ops->rule_size, GFP_KERNEL);
246 rule->fr_net = hold_net(net);
248 if (tb[FRA_PRIORITY])
249 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
251 if (tb[FRA_IFNAME]) {
252 struct net_device *dev;
255 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
256 dev = __dev_get_by_name(net, rule->ifname);
258 rule->ifindex = dev->ifindex;
261 if (tb[FRA_FWMARK]) {
262 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
264 /* compatibility: if the mark value is non-zero all bits
265 * are compared unless a mask is explicitly specified.
267 rule->mark_mask = 0xFFFFFFFF;
271 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
273 rule->action = frh->action;
274 rule->flags = frh->flags;
275 rule->table = frh_get_table(frh, tb);
277 if (!rule->pref && ops->default_pref)
278 rule->pref = ops->default_pref(ops);
282 if (rule->action != FR_ACT_GOTO)
285 rule->target = nla_get_u32(tb[FRA_GOTO]);
286 /* Backward jumps are prohibited to avoid endless loops */
287 if (rule->target <= rule->pref)
290 list_for_each_entry(r, &ops->rules_list, list) {
291 if (r->pref == rule->target) {
297 if (rule->ctarget == NULL)
299 } else if (rule->action == FR_ACT_GOTO)
302 err = ops->configure(rule, skb, frh, tb);
306 list_for_each_entry(r, &ops->rules_list, list) {
307 if (r->pref > rule->pref)
314 if (ops->unresolved_rules) {
316 * There are unresolved goto rules in the list, check if
317 * any of them are pointing to this new rule.
319 list_for_each_entry(r, &ops->rules_list, list) {
320 if (r->action == FR_ACT_GOTO &&
321 r->target == rule->pref) {
322 BUG_ON(r->ctarget != NULL);
323 rcu_assign_pointer(r->ctarget, rule);
324 if (--ops->unresolved_rules == 0)
330 if (rule->action == FR_ACT_GOTO)
331 ops->nr_goto_rules++;
334 ops->unresolved_rules++;
337 list_add_rcu(&rule->list, &last->list);
339 list_add_rcu(&rule->list, &ops->rules_list);
341 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
342 flush_route_cache(ops);
347 release_net(rule->fr_net);
354 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
356 struct net *net = sock_net(skb->sk);
357 struct fib_rule_hdr *frh = nlmsg_data(nlh);
358 struct fib_rules_ops *ops = NULL;
359 struct fib_rule *rule, *tmp;
360 struct nlattr *tb[FRA_MAX+1];
363 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
366 ops = lookup_rules_ops(net, frh->family);
372 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
376 err = validate_rulemsg(frh, tb, ops);
380 list_for_each_entry(rule, &ops->rules_list, list) {
381 if (frh->action && (frh->action != rule->action))
384 if (frh->table && (frh_get_table(frh, tb) != rule->table))
387 if (tb[FRA_PRIORITY] &&
388 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
391 if (tb[FRA_IFNAME] &&
392 nla_strcmp(tb[FRA_IFNAME], rule->ifname))
395 if (tb[FRA_FWMARK] &&
396 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
399 if (tb[FRA_FWMASK] &&
400 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
403 if (!ops->compare(rule, frh, tb))
406 if (rule->flags & FIB_RULE_PERMANENT) {
411 list_del_rcu(&rule->list);
413 if (rule->action == FR_ACT_GOTO)
414 ops->nr_goto_rules--;
417 * Check if this rule is a target to any of them. If so,
418 * disable them. As this operation is eventually very
419 * expensive, it is only performed if goto rules have
420 * actually been added.
422 if (ops->nr_goto_rules > 0) {
423 list_for_each_entry(tmp, &ops->rules_list, list) {
424 if (tmp->ctarget == rule) {
425 rcu_assign_pointer(tmp->ctarget, NULL);
426 ops->unresolved_rules++;
432 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
433 NETLINK_CB(skb).pid);
435 flush_route_cache(ops);
446 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
447 struct fib_rule *rule)
449 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
450 + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
451 + nla_total_size(4) /* FRA_PRIORITY */
452 + nla_total_size(4) /* FRA_TABLE */
453 + nla_total_size(4) /* FRA_FWMARK */
454 + nla_total_size(4); /* FRA_FWMASK */
456 if (ops->nlmsg_payload)
457 payload += ops->nlmsg_payload(rule);
462 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
463 u32 pid, u32 seq, int type, int flags,
464 struct fib_rules_ops *ops)
466 struct nlmsghdr *nlh;
467 struct fib_rule_hdr *frh;
469 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
473 frh = nlmsg_data(nlh);
474 frh->table = rule->table;
475 NLA_PUT_U32(skb, FRA_TABLE, rule->table);
478 frh->action = rule->action;
479 frh->flags = rule->flags;
481 if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
482 frh->flags |= FIB_RULE_UNRESOLVED;
484 if (rule->ifname[0]) {
485 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
487 if (rule->ifindex == -1)
488 frh->flags |= FIB_RULE_DEV_DETACHED;
492 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
495 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
497 if (rule->mark_mask || rule->mark)
498 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
501 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
503 if (ops->fill(rule, skb, frh) < 0)
504 goto nla_put_failure;
506 return nlmsg_end(skb, nlh);
509 nlmsg_cancel(skb, nlh);
513 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
514 struct fib_rules_ops *ops)
517 struct fib_rule *rule;
519 list_for_each_entry(rule, &ops->rules_list, list) {
520 if (idx < cb->args[1])
523 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
524 cb->nlh->nlmsg_seq, RTM_NEWRULE,
525 NLM_F_MULTI, ops) < 0)
536 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
538 struct net *net = sock_net(skb->sk);
539 struct fib_rules_ops *ops;
542 family = rtnl_msg_family(cb->nlh);
543 if (family != AF_UNSPEC) {
544 /* Protocol specific dump request */
545 ops = lookup_rules_ops(net, family);
547 return -EAFNOSUPPORT;
549 return dump_rules(skb, cb, ops);
553 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
554 if (idx < cb->args[0] || !try_module_get(ops->owner))
557 if (dump_rules(skb, cb, ops) < 0)
570 static void notify_rule_change(int event, struct fib_rule *rule,
571 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
579 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
583 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
585 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
586 WARN_ON(err == -EMSGSIZE);
591 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
595 rtnl_set_sk_err(net, ops->nlgroup, err);
598 static void attach_rules(struct list_head *rules, struct net_device *dev)
600 struct fib_rule *rule;
602 list_for_each_entry(rule, rules, list) {
603 if (rule->ifindex == -1 &&
604 strcmp(dev->name, rule->ifname) == 0)
605 rule->ifindex = dev->ifindex;
609 static void detach_rules(struct list_head *rules, struct net_device *dev)
611 struct fib_rule *rule;
613 list_for_each_entry(rule, rules, list)
614 if (rule->ifindex == dev->ifindex)
619 static int fib_rules_event(struct notifier_block *this, unsigned long event,
622 struct net_device *dev = ptr;
623 struct net *net = dev_net(dev);
624 struct fib_rules_ops *ops;
630 case NETDEV_REGISTER:
631 list_for_each_entry(ops, &net->rules_ops, list)
632 attach_rules(&ops->rules_list, dev);
635 case NETDEV_UNREGISTER:
636 list_for_each_entry(ops, &net->rules_ops, list)
637 detach_rules(&ops->rules_list, dev);
646 static struct notifier_block fib_rules_notifier = {
647 .notifier_call = fib_rules_event,
650 static int fib_rules_net_init(struct net *net)
652 INIT_LIST_HEAD(&net->rules_ops);
653 spin_lock_init(&net->rules_mod_lock);
657 static struct pernet_operations fib_rules_net_ops = {
658 .init = fib_rules_net_init,
661 static int __init fib_rules_init(void)
664 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
665 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
666 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
668 err = register_pernet_subsys(&fib_rules_net_ops);
672 err = register_netdevice_notifier(&fib_rules_notifier);
674 goto fail_unregister;
679 unregister_pernet_subsys(&fib_rules_net_ops);
681 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
682 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
683 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
687 subsys_initcall(fib_rules_init);