Add linux-next specific files for 20110426
[linux-2.6/next.git] / net / core / fib_rules.c
blob8248ebb5891d0764eebb9d4cbc2256c091ff357c
1 /*
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>
9 */
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>
16 #include <net/sock.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)
22 struct fib_rule *r;
24 r = kzalloc(ops->rule_size, GFP_KERNEL);
25 if (r == NULL)
26 return -ENOMEM;
28 atomic_set(&r->refcnt, 1);
29 r->action = FR_ACT_TO_TBL;
30 r->pref = pref;
31 r->table = table;
32 r->flags = flags;
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);
38 return 0;
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);
51 if (rule->pref)
52 return rule->pref - 1;
56 return 0;
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,
62 u32 pid);
64 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
66 struct fib_rules_ops *ops;
68 rcu_read_lock();
69 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
70 if (ops->family == family) {
71 if (!try_module_get(ops->owner))
72 ops = NULL;
73 rcu_read_unlock();
74 return ops;
77 rcu_read_unlock();
79 return NULL;
82 static void rules_ops_put(struct fib_rules_ops *ops)
84 if (ops)
85 module_put(ops->owner);
88 static void flush_route_cache(struct fib_rules_ops *ops)
90 if (ops->flush_cache)
91 ops->flush_cache(ops);
94 static int __fib_rules_register(struct fib_rules_ops *ops)
96 int err = -EEXIST;
97 struct fib_rules_ops *o;
98 struct net *net;
100 net = ops->fro_net;
102 if (ops->rule_size < sizeof(struct fib_rule))
103 return -EINVAL;
105 if (ops->match == NULL || ops->configure == NULL ||
106 ops->compare == NULL || ops->fill == NULL ||
107 ops->action == NULL)
108 return -EINVAL;
110 spin_lock(&net->rules_mod_lock);
111 list_for_each_entry(o, &net->rules_ops, list)
112 if (ops->family == o->family)
113 goto errout;
115 hold_net(net);
116 list_add_tail_rcu(&ops->list, &net->rules_ops);
117 err = 0;
118 errout:
119 spin_unlock(&net->rules_mod_lock);
121 return err;
124 struct fib_rules_ops *
125 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
127 struct fib_rules_ops *ops;
128 int err;
130 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
131 if (ops == NULL)
132 return ERR_PTR(-ENOMEM);
134 INIT_LIST_HEAD(&ops->rules_list);
135 ops->fro_net = net;
137 err = __fib_rules_register(ops);
138 if (err) {
139 kfree(ops);
140 ops = ERR_PTR(err);
143 return 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);
153 fib_rule_put(rule);
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;
162 release_net(net);
163 kfree(ops);
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)
182 int ret = 0;
184 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
185 goto out;
187 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
188 goto out;
190 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
191 goto out;
193 ret = ops->match(rule, fl, flags);
194 out:
195 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
198 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
199 int flags, struct fib_lookup_arg *arg)
201 struct fib_rule *rule;
202 int err;
204 rcu_read_lock();
206 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
207 jumped:
208 if (!fib_rule_match(rule, ops, fl, flags))
209 continue;
211 if (rule->action == FR_ACT_GOTO) {
212 struct fib_rule *target;
214 target = rcu_dereference(rule->ctarget);
215 if (target == NULL) {
216 continue;
217 } else {
218 rule = target;
219 goto jumped;
221 } else if (rule->action == FR_ACT_NOP)
222 continue;
223 else
224 err = ops->action(rule, fl, flags, arg);
226 if (err != -EAGAIN) {
227 if ((arg->flags & FIB_LOOKUP_NOREF) ||
228 likely(atomic_inc_not_zero(&rule->refcnt))) {
229 arg->rule = rule;
230 goto out;
232 break;
236 err = -ESRCH;
237 out:
238 rcu_read_unlock();
240 return err;
242 EXPORT_SYMBOL_GPL(fib_rules_lookup);
244 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
245 struct fib_rules_ops *ops)
247 int err = -EINVAL;
249 if (frh->src_len)
250 if (tb[FRA_SRC] == NULL ||
251 frh->src_len > (ops->addr_size * 8) ||
252 nla_len(tb[FRA_SRC]) != ops->addr_size)
253 goto errout;
255 if (frh->dst_len)
256 if (tb[FRA_DST] == NULL ||
257 frh->dst_len > (ops->addr_size * 8) ||
258 nla_len(tb[FRA_DST]) != ops->addr_size)
259 goto errout;
261 err = 0;
262 errout:
263 return err;
266 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
268 struct net *net = sock_net(skb->sk);
269 struct fib_rule_hdr *frh = nlmsg_data(nlh);
270 struct fib_rules_ops *ops = NULL;
271 struct fib_rule *rule, *r, *last = NULL;
272 struct nlattr *tb[FRA_MAX+1];
273 int err = -EINVAL, unresolved = 0;
275 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
276 goto errout;
278 ops = lookup_rules_ops(net, frh->family);
279 if (ops == NULL) {
280 err = -EAFNOSUPPORT;
281 goto errout;
284 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
285 if (err < 0)
286 goto errout;
288 err = validate_rulemsg(frh, tb, ops);
289 if (err < 0)
290 goto errout;
292 rule = kzalloc(ops->rule_size, GFP_KERNEL);
293 if (rule == NULL) {
294 err = -ENOMEM;
295 goto errout;
297 rule->fr_net = hold_net(net);
299 if (tb[FRA_PRIORITY])
300 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
302 if (tb[FRA_IIFNAME]) {
303 struct net_device *dev;
305 rule->iifindex = -1;
306 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
307 dev = __dev_get_by_name(net, rule->iifname);
308 if (dev)
309 rule->iifindex = dev->ifindex;
312 if (tb[FRA_OIFNAME]) {
313 struct net_device *dev;
315 rule->oifindex = -1;
316 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
317 dev = __dev_get_by_name(net, rule->oifname);
318 if (dev)
319 rule->oifindex = dev->ifindex;
322 if (tb[FRA_FWMARK]) {
323 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
324 if (rule->mark)
325 /* compatibility: if the mark value is non-zero all bits
326 * are compared unless a mask is explicitly specified.
328 rule->mark_mask = 0xFFFFFFFF;
331 if (tb[FRA_FWMASK])
332 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
334 rule->action = frh->action;
335 rule->flags = frh->flags;
336 rule->table = frh_get_table(frh, tb);
338 if (!tb[FRA_PRIORITY] && ops->default_pref)
339 rule->pref = ops->default_pref(ops);
341 err = -EINVAL;
342 if (tb[FRA_GOTO]) {
343 if (rule->action != FR_ACT_GOTO)
344 goto errout_free;
346 rule->target = nla_get_u32(tb[FRA_GOTO]);
347 /* Backward jumps are prohibited to avoid endless loops */
348 if (rule->target <= rule->pref)
349 goto errout_free;
351 list_for_each_entry(r, &ops->rules_list, list) {
352 if (r->pref == rule->target) {
353 RCU_INIT_POINTER(rule->ctarget, r);
354 break;
358 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
359 unresolved = 1;
360 } else if (rule->action == FR_ACT_GOTO)
361 goto errout_free;
363 err = ops->configure(rule, skb, frh, tb);
364 if (err < 0)
365 goto errout_free;
367 list_for_each_entry(r, &ops->rules_list, list) {
368 if (r->pref > rule->pref)
369 break;
370 last = r;
373 fib_rule_get(rule);
375 if (last)
376 list_add_rcu(&rule->list, &last->list);
377 else
378 list_add_rcu(&rule->list, &ops->rules_list);
380 if (ops->unresolved_rules) {
382 * There are unresolved goto rules in the list, check if
383 * any of them are pointing to this new rule.
385 list_for_each_entry(r, &ops->rules_list, list) {
386 if (r->action == FR_ACT_GOTO &&
387 r->target == rule->pref) {
388 BUG_ON(rtnl_dereference(r->ctarget) != NULL);
389 rcu_assign_pointer(r->ctarget, rule);
390 if (--ops->unresolved_rules == 0)
391 break;
396 if (rule->action == FR_ACT_GOTO)
397 ops->nr_goto_rules++;
399 if (unresolved)
400 ops->unresolved_rules++;
402 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
403 flush_route_cache(ops);
404 rules_ops_put(ops);
405 return 0;
407 errout_free:
408 release_net(rule->fr_net);
409 kfree(rule);
410 errout:
411 rules_ops_put(ops);
412 return err;
415 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
417 struct net *net = sock_net(skb->sk);
418 struct fib_rule_hdr *frh = nlmsg_data(nlh);
419 struct fib_rules_ops *ops = NULL;
420 struct fib_rule *rule, *tmp;
421 struct nlattr *tb[FRA_MAX+1];
422 int err = -EINVAL;
424 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
425 goto errout;
427 ops = lookup_rules_ops(net, frh->family);
428 if (ops == NULL) {
429 err = -EAFNOSUPPORT;
430 goto errout;
433 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
434 if (err < 0)
435 goto errout;
437 err = validate_rulemsg(frh, tb, ops);
438 if (err < 0)
439 goto errout;
441 list_for_each_entry(rule, &ops->rules_list, list) {
442 if (frh->action && (frh->action != rule->action))
443 continue;
445 if (frh->table && (frh_get_table(frh, tb) != rule->table))
446 continue;
448 if (tb[FRA_PRIORITY] &&
449 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
450 continue;
452 if (tb[FRA_IIFNAME] &&
453 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
454 continue;
456 if (tb[FRA_OIFNAME] &&
457 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
458 continue;
460 if (tb[FRA_FWMARK] &&
461 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
462 continue;
464 if (tb[FRA_FWMASK] &&
465 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
466 continue;
468 if (!ops->compare(rule, frh, tb))
469 continue;
471 if (rule->flags & FIB_RULE_PERMANENT) {
472 err = -EPERM;
473 goto errout;
476 list_del_rcu(&rule->list);
478 if (rule->action == FR_ACT_GOTO)
479 ops->nr_goto_rules--;
482 * Check if this rule is a target to any of them. If so,
483 * disable them. As this operation is eventually very
484 * expensive, it is only performed if goto rules have
485 * actually been added.
487 if (ops->nr_goto_rules > 0) {
488 list_for_each_entry(tmp, &ops->rules_list, list) {
489 if (rtnl_dereference(tmp->ctarget) == rule) {
490 rcu_assign_pointer(tmp->ctarget, NULL);
491 ops->unresolved_rules++;
496 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
497 NETLINK_CB(skb).pid);
498 fib_rule_put(rule);
499 flush_route_cache(ops);
500 rules_ops_put(ops);
501 return 0;
504 err = -ENOENT;
505 errout:
506 rules_ops_put(ops);
507 return err;
510 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
511 struct fib_rule *rule)
513 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
514 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
515 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
516 + nla_total_size(4) /* FRA_PRIORITY */
517 + nla_total_size(4) /* FRA_TABLE */
518 + nla_total_size(4) /* FRA_FWMARK */
519 + nla_total_size(4); /* FRA_FWMASK */
521 if (ops->nlmsg_payload)
522 payload += ops->nlmsg_payload(rule);
524 return payload;
527 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
528 u32 pid, u32 seq, int type, int flags,
529 struct fib_rules_ops *ops)
531 struct nlmsghdr *nlh;
532 struct fib_rule_hdr *frh;
534 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
535 if (nlh == NULL)
536 return -EMSGSIZE;
538 frh = nlmsg_data(nlh);
539 frh->family = ops->family;
540 frh->table = rule->table;
541 NLA_PUT_U32(skb, FRA_TABLE, rule->table);
542 frh->res1 = 0;
543 frh->res2 = 0;
544 frh->action = rule->action;
545 frh->flags = rule->flags;
547 if (rule->action == FR_ACT_GOTO &&
548 rcu_dereference_raw(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;
565 if (rule->pref)
566 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
568 if (rule->mark)
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);
574 if (rule->target)
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);
582 nla_put_failure:
583 nlmsg_cancel(skb, nlh);
584 return -EMSGSIZE;
587 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
588 struct fib_rules_ops *ops)
590 int idx = 0;
591 struct fib_rule *rule;
593 list_for_each_entry(rule, &ops->rules_list, list) {
594 if (idx < cb->args[1])
595 goto skip;
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)
600 break;
601 skip:
602 idx++;
604 cb->args[1] = idx;
605 rules_ops_put(ops);
607 return skb->len;
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;
614 int idx = 0, family;
616 family = rtnl_msg_family(cb->nlh);
617 if (family != AF_UNSPEC) {
618 /* Protocol specific dump request */
619 ops = lookup_rules_ops(net, family);
620 if (ops == NULL)
621 return -EAFNOSUPPORT;
623 return dump_rules(skb, cb, ops);
626 rcu_read_lock();
627 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
628 if (idx < cb->args[0] || !try_module_get(ops->owner))
629 goto skip;
631 if (dump_rules(skb, cb, ops) < 0)
632 break;
634 cb->args[1] = 0;
635 skip:
636 idx++;
638 rcu_read_unlock();
639 cb->args[0] = idx;
641 return skb->len;
644 static void notify_rule_change(int event, struct fib_rule *rule,
645 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
646 u32 pid)
648 struct net *net;
649 struct sk_buff *skb;
650 int err = -ENOBUFS;
652 net = ops->fro_net;
653 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
654 if (skb == NULL)
655 goto errout;
657 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
658 if (err < 0) {
659 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
660 WARN_ON(err == -EMSGSIZE);
661 kfree_skb(skb);
662 goto errout;
665 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
666 return;
667 errout:
668 if (err < 0)
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)
692 rule->iifindex = -1;
693 if (rule->oifindex == dev->ifindex)
694 rule->oifindex = -1;
699 static int fib_rules_event(struct notifier_block *this, unsigned long event,
700 void *ptr)
702 struct net_device *dev = ptr;
703 struct net *net = dev_net(dev);
704 struct fib_rules_ops *ops;
706 ASSERT_RTNL();
708 switch (event) {
709 case NETDEV_REGISTER:
710 list_for_each_entry(ops, &net->rules_ops, list)
711 attach_rules(&ops->rules_list, dev);
712 break;
714 case NETDEV_UNREGISTER:
715 list_for_each_entry(ops, &net->rules_ops, list)
716 detach_rules(&ops->rules_list, dev);
717 break;
720 return NOTIFY_DONE;
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);
731 return 0;
734 static struct pernet_operations fib_rules_net_ops = {
735 .init = fib_rules_net_init,
738 static int __init fib_rules_init(void)
740 int err;
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);
746 if (err < 0)
747 goto fail;
749 err = register_netdevice_notifier(&fib_rules_notifier);
750 if (err < 0)
751 goto fail_unregister;
753 return 0;
755 fail_unregister:
756 unregister_pernet_subsys(&fib_rules_net_ops);
757 fail:
758 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
759 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
760 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
761 return err;
764 subsys_initcall(fib_rules_init);