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>
15 #include <net/fib_rules.h>
17 static LIST_HEAD(rules_ops);
18 static DEFINE_SPINLOCK(rules_mod_lock);
20 static void notify_rule_change(int event, struct fib_rule *rule,
21 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
24 static struct fib_rules_ops *lookup_rules_ops(int family)
26 struct fib_rules_ops *ops;
29 list_for_each_entry_rcu(ops, &rules_ops, list) {
30 if (ops->family == family) {
31 if (!try_module_get(ops->owner))
42 static void rules_ops_put(struct fib_rules_ops *ops)
45 module_put(ops->owner);
48 static void flush_route_cache(struct fib_rules_ops *ops)
54 int fib_rules_register(struct fib_rules_ops *ops)
57 struct fib_rules_ops *o;
59 if (ops->rule_size < sizeof(struct fib_rule))
62 if (ops->match == NULL || ops->configure == NULL ||
63 ops->compare == NULL || ops->fill == NULL ||
67 spin_lock(&rules_mod_lock);
68 list_for_each_entry(o, &rules_ops, list)
69 if (ops->family == o->family)
72 list_add_tail_rcu(&ops->list, &rules_ops);
75 spin_unlock(&rules_mod_lock);
80 EXPORT_SYMBOL_GPL(fib_rules_register);
82 static void cleanup_ops(struct fib_rules_ops *ops)
84 struct fib_rule *rule, *tmp;
86 list_for_each_entry_safe(rule, tmp, ops->rules_list, list) {
87 list_del_rcu(&rule->list);
92 int fib_rules_unregister(struct fib_rules_ops *ops)
95 struct fib_rules_ops *o;
97 spin_lock(&rules_mod_lock);
98 list_for_each_entry(o, &rules_ops, list) {
100 list_del_rcu(&o->list);
108 spin_unlock(&rules_mod_lock);
115 EXPORT_SYMBOL_GPL(fib_rules_unregister);
117 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
118 struct flowi *fl, int flags)
122 if (rule->ifindex && (rule->ifindex != fl->iif))
125 if ((rule->mark ^ fl->mark) & rule->mark_mask)
128 ret = ops->match(rule, fl, flags);
130 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
133 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
134 int flags, struct fib_lookup_arg *arg)
136 struct fib_rule *rule;
141 list_for_each_entry_rcu(rule, ops->rules_list, list) {
143 if (!fib_rule_match(rule, ops, fl, flags))
146 if (rule->action == FR_ACT_GOTO) {
147 struct fib_rule *target;
149 target = rcu_dereference(rule->ctarget);
150 if (target == NULL) {
156 } else if (rule->action == FR_ACT_NOP)
159 err = ops->action(rule, fl, flags, arg);
161 if (err != -EAGAIN) {
175 EXPORT_SYMBOL_GPL(fib_rules_lookup);
177 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
178 struct fib_rules_ops *ops)
183 if (tb[FRA_SRC] == NULL ||
184 frh->src_len > (ops->addr_size * 8) ||
185 nla_len(tb[FRA_SRC]) != ops->addr_size)
189 if (tb[FRA_DST] == NULL ||
190 frh->dst_len > (ops->addr_size * 8) ||
191 nla_len(tb[FRA_DST]) != ops->addr_size)
199 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
201 struct fib_rule_hdr *frh = nlmsg_data(nlh);
202 struct fib_rules_ops *ops = NULL;
203 struct fib_rule *rule, *r, *last = NULL;
204 struct nlattr *tb[FRA_MAX+1];
205 int err = -EINVAL, unresolved = 0;
207 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
210 ops = lookup_rules_ops(frh->family);
216 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
220 err = validate_rulemsg(frh, tb, ops);
224 rule = kzalloc(ops->rule_size, GFP_KERNEL);
230 if (tb[FRA_PRIORITY])
231 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
233 if (tb[FRA_IFNAME]) {
234 struct net_device *dev;
237 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
238 dev = __dev_get_by_name(rule->ifname);
240 rule->ifindex = dev->ifindex;
243 if (tb[FRA_FWMARK]) {
244 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
246 /* compatibility: if the mark value is non-zero all bits
247 * are compared unless a mask is explicitly specified.
249 rule->mark_mask = 0xFFFFFFFF;
253 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
255 rule->action = frh->action;
256 rule->flags = frh->flags;
257 rule->table = frh_get_table(frh, tb);
259 if (!rule->pref && ops->default_pref)
260 rule->pref = ops->default_pref();
264 if (rule->action != FR_ACT_GOTO)
267 rule->target = nla_get_u32(tb[FRA_GOTO]);
268 /* Backward jumps are prohibited to avoid endless loops */
269 if (rule->target <= rule->pref)
272 list_for_each_entry(r, ops->rules_list, list) {
273 if (r->pref == rule->target) {
279 if (rule->ctarget == NULL)
281 } else if (rule->action == FR_ACT_GOTO)
284 err = ops->configure(rule, skb, nlh, frh, tb);
288 list_for_each_entry(r, ops->rules_list, list) {
289 if (r->pref > rule->pref)
296 if (ops->unresolved_rules) {
298 * There are unresolved goto rules in the list, check if
299 * any of them are pointing to this new rule.
301 list_for_each_entry(r, ops->rules_list, list) {
302 if (r->action == FR_ACT_GOTO &&
303 r->target == rule->pref) {
304 BUG_ON(r->ctarget != NULL);
305 rcu_assign_pointer(r->ctarget, rule);
306 if (--ops->unresolved_rules == 0)
312 if (rule->action == FR_ACT_GOTO)
313 ops->nr_goto_rules++;
316 ops->unresolved_rules++;
319 list_add_rcu(&rule->list, &last->list);
321 list_add_rcu(&rule->list, ops->rules_list);
323 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
324 flush_route_cache(ops);
335 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
337 struct fib_rule_hdr *frh = nlmsg_data(nlh);
338 struct fib_rules_ops *ops = NULL;
339 struct fib_rule *rule, *tmp;
340 struct nlattr *tb[FRA_MAX+1];
343 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
346 ops = lookup_rules_ops(frh->family);
352 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
356 err = validate_rulemsg(frh, tb, ops);
360 list_for_each_entry(rule, ops->rules_list, list) {
361 if (frh->action && (frh->action != rule->action))
364 if (frh->table && (frh_get_table(frh, tb) != rule->table))
367 if (tb[FRA_PRIORITY] &&
368 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
371 if (tb[FRA_IFNAME] &&
372 nla_strcmp(tb[FRA_IFNAME], rule->ifname))
375 if (tb[FRA_FWMARK] &&
376 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
379 if (tb[FRA_FWMASK] &&
380 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
383 if (!ops->compare(rule, frh, tb))
386 if (rule->flags & FIB_RULE_PERMANENT) {
391 list_del_rcu(&rule->list);
393 if (rule->action == FR_ACT_GOTO)
394 ops->nr_goto_rules--;
397 * Check if this rule is a target to any of them. If so,
398 * disable them. As this operation is eventually very
399 * expensive, it is only performed if goto rules have
400 * actually been added.
402 if (ops->nr_goto_rules > 0) {
403 list_for_each_entry(tmp, ops->rules_list, list) {
404 if (tmp->ctarget == rule) {
405 rcu_assign_pointer(tmp->ctarget, NULL);
406 ops->unresolved_rules++;
412 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
413 NETLINK_CB(skb).pid);
415 flush_route_cache(ops);
426 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
427 struct fib_rule *rule)
429 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
430 + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
431 + nla_total_size(4) /* FRA_PRIORITY */
432 + nla_total_size(4) /* FRA_TABLE */
433 + nla_total_size(4) /* FRA_FWMARK */
434 + nla_total_size(4); /* FRA_FWMASK */
436 if (ops->nlmsg_payload)
437 payload += ops->nlmsg_payload(rule);
442 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
443 u32 pid, u32 seq, int type, int flags,
444 struct fib_rules_ops *ops)
446 struct nlmsghdr *nlh;
447 struct fib_rule_hdr *frh;
449 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
453 frh = nlmsg_data(nlh);
454 frh->table = rule->table;
455 NLA_PUT_U32(skb, FRA_TABLE, rule->table);
458 frh->action = rule->action;
459 frh->flags = rule->flags;
461 if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
462 frh->flags |= FIB_RULE_UNRESOLVED;
464 if (rule->ifname[0]) {
465 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
467 if (rule->ifindex == -1)
468 frh->flags |= FIB_RULE_DEV_DETACHED;
472 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
475 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
477 if (rule->mark_mask || rule->mark)
478 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
481 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
483 if (ops->fill(rule, skb, nlh, frh) < 0)
484 goto nla_put_failure;
486 return nlmsg_end(skb, nlh);
489 nlmsg_cancel(skb, nlh);
493 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
494 struct fib_rules_ops *ops)
497 struct fib_rule *rule;
499 list_for_each_entry(rule, ops->rules_list, list) {
500 if (idx < cb->args[1])
503 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
504 cb->nlh->nlmsg_seq, RTM_NEWRULE,
505 NLM_F_MULTI, ops) < 0)
516 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
518 struct fib_rules_ops *ops;
521 family = rtnl_msg_family(cb->nlh);
522 if (family != AF_UNSPEC) {
523 /* Protocol specific dump request */
524 ops = lookup_rules_ops(family);
526 return -EAFNOSUPPORT;
528 return dump_rules(skb, cb, ops);
532 list_for_each_entry_rcu(ops, &rules_ops, list) {
533 if (idx < cb->args[0] || !try_module_get(ops->owner))
536 if (dump_rules(skb, cb, ops) < 0)
549 static void notify_rule_change(int event, struct fib_rule *rule,
550 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
556 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
560 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
562 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
563 WARN_ON(err == -EMSGSIZE);
567 err = rtnl_notify(skb, pid, ops->nlgroup, nlh, GFP_KERNEL);
570 rtnl_set_sk_err(ops->nlgroup, err);
573 static void attach_rules(struct list_head *rules, struct net_device *dev)
575 struct fib_rule *rule;
577 list_for_each_entry(rule, rules, list) {
578 if (rule->ifindex == -1 &&
579 strcmp(dev->name, rule->ifname) == 0)
580 rule->ifindex = dev->ifindex;
584 static void detach_rules(struct list_head *rules, struct net_device *dev)
586 struct fib_rule *rule;
588 list_for_each_entry(rule, rules, list)
589 if (rule->ifindex == dev->ifindex)
594 static int fib_rules_event(struct notifier_block *this, unsigned long event,
597 struct net_device *dev = ptr;
598 struct fib_rules_ops *ops;
600 if (dev->nd_net != &init_net)
607 case NETDEV_REGISTER:
608 list_for_each_entry(ops, &rules_ops, list)
609 attach_rules(ops->rules_list, dev);
612 case NETDEV_UNREGISTER:
613 list_for_each_entry(ops, &rules_ops, list)
614 detach_rules(ops->rules_list, dev);
623 static struct notifier_block fib_rules_notifier = {
624 .notifier_call = fib_rules_event,
627 static int __init fib_rules_init(void)
629 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
630 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
631 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
633 return register_netdevice_notifier(&fib_rules_notifier);
636 subsys_initcall(fib_rules_init);