block: remove WARN_ON_ONCE() in exit_io_context()
[linux-2.6.git] / net / core / fib_rules.c
blobc02e63c908da6972228735ba247dfcda6798b233
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 <linux/module.h>
16 #include <net/net_namespace.h>
17 #include <net/sock.h>
18 #include <net/fib_rules.h>
20 int fib_default_rule_add(struct fib_rules_ops *ops,
21 u32 pref, u32 table, u32 flags)
23 struct fib_rule *r;
25 r = kzalloc(ops->rule_size, GFP_KERNEL);
26 if (r == NULL)
27 return -ENOMEM;
29 atomic_set(&r->refcnt, 1);
30 r->action = FR_ACT_TO_TBL;
31 r->pref = pref;
32 r->table = table;
33 r->flags = flags;
34 r->fr_net = hold_net(ops->fro_net);
36 /* The lock is not required here, the list in unreacheable
37 * at the moment this function is called */
38 list_add_tail(&r->list, &ops->rules_list);
39 return 0;
41 EXPORT_SYMBOL(fib_default_rule_add);
43 u32 fib_default_rule_pref(struct fib_rules_ops *ops)
45 struct list_head *pos;
46 struct fib_rule *rule;
48 if (!list_empty(&ops->rules_list)) {
49 pos = ops->rules_list.next;
50 if (pos->next != &ops->rules_list) {
51 rule = list_entry(pos->next, struct fib_rule, list);
52 if (rule->pref)
53 return rule->pref - 1;
57 return 0;
59 EXPORT_SYMBOL(fib_default_rule_pref);
61 static void notify_rule_change(int event, struct fib_rule *rule,
62 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
63 u32 pid);
65 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
67 struct fib_rules_ops *ops;
69 rcu_read_lock();
70 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
71 if (ops->family == family) {
72 if (!try_module_get(ops->owner))
73 ops = NULL;
74 rcu_read_unlock();
75 return ops;
78 rcu_read_unlock();
80 return NULL;
83 static void rules_ops_put(struct fib_rules_ops *ops)
85 if (ops)
86 module_put(ops->owner);
89 static void flush_route_cache(struct fib_rules_ops *ops)
91 if (ops->flush_cache)
92 ops->flush_cache(ops);
95 static int __fib_rules_register(struct fib_rules_ops *ops)
97 int err = -EEXIST;
98 struct fib_rules_ops *o;
99 struct net *net;
101 net = ops->fro_net;
103 if (ops->rule_size < sizeof(struct fib_rule))
104 return -EINVAL;
106 if (ops->match == NULL || ops->configure == NULL ||
107 ops->compare == NULL || ops->fill == NULL ||
108 ops->action == NULL)
109 return -EINVAL;
111 spin_lock(&net->rules_mod_lock);
112 list_for_each_entry(o, &net->rules_ops, list)
113 if (ops->family == o->family)
114 goto errout;
116 hold_net(net);
117 list_add_tail_rcu(&ops->list, &net->rules_ops);
118 err = 0;
119 errout:
120 spin_unlock(&net->rules_mod_lock);
122 return err;
125 struct fib_rules_ops *
126 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
128 struct fib_rules_ops *ops;
129 int err;
131 ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
132 if (ops == NULL)
133 return ERR_PTR(-ENOMEM);
135 INIT_LIST_HEAD(&ops->rules_list);
136 ops->fro_net = net;
138 err = __fib_rules_register(ops);
139 if (err) {
140 kfree(ops);
141 ops = ERR_PTR(err);
144 return ops;
146 EXPORT_SYMBOL_GPL(fib_rules_register);
148 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
150 struct fib_rule *rule, *tmp;
152 list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
153 list_del_rcu(&rule->list);
154 fib_rule_put(rule);
158 static void fib_rules_put_rcu(struct rcu_head *head)
160 struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu);
161 struct net *net = ops->fro_net;
163 release_net(net);
164 kfree(ops);
167 void fib_rules_unregister(struct fib_rules_ops *ops)
169 struct net *net = ops->fro_net;
171 spin_lock(&net->rules_mod_lock);
172 list_del_rcu(&ops->list);
173 fib_rules_cleanup_ops(ops);
174 spin_unlock(&net->rules_mod_lock);
176 call_rcu(&ops->rcu, fib_rules_put_rcu);
178 EXPORT_SYMBOL_GPL(fib_rules_unregister);
180 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
181 struct flowi *fl, int flags)
183 int ret = 0;
185 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
186 goto out;
188 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
189 goto out;
191 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
192 goto out;
194 ret = ops->match(rule, fl, flags);
195 out:
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;
203 int err;
205 rcu_read_lock();
207 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
208 jumped:
209 if (!fib_rule_match(rule, ops, fl, flags))
210 continue;
212 if (rule->action == FR_ACT_GOTO) {
213 struct fib_rule *target;
215 target = rcu_dereference(rule->ctarget);
216 if (target == NULL) {
217 continue;
218 } else {
219 rule = target;
220 goto jumped;
222 } else if (rule->action == FR_ACT_NOP)
223 continue;
224 else
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))) {
230 arg->rule = rule;
231 goto out;
233 break;
237 err = -ESRCH;
238 out:
239 rcu_read_unlock();
241 return err;
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)
248 int err = -EINVAL;
250 if (frh->src_len)
251 if (tb[FRA_SRC] == NULL ||
252 frh->src_len > (ops->addr_size * 8) ||
253 nla_len(tb[FRA_SRC]) != ops->addr_size)
254 goto errout;
256 if (frh->dst_len)
257 if (tb[FRA_DST] == NULL ||
258 frh->dst_len > (ops->addr_size * 8) ||
259 nla_len(tb[FRA_DST]) != ops->addr_size)
260 goto errout;
262 err = 0;
263 errout:
264 return err;
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)))
277 goto errout;
279 ops = lookup_rules_ops(net, frh->family);
280 if (ops == NULL) {
281 err = -EAFNOSUPPORT;
282 goto errout;
285 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
286 if (err < 0)
287 goto errout;
289 err = validate_rulemsg(frh, tb, ops);
290 if (err < 0)
291 goto errout;
293 rule = kzalloc(ops->rule_size, GFP_KERNEL);
294 if (rule == NULL) {
295 err = -ENOMEM;
296 goto errout;
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;
306 rule->iifindex = -1;
307 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
308 dev = __dev_get_by_name(net, rule->iifname);
309 if (dev)
310 rule->iifindex = dev->ifindex;
313 if (tb[FRA_OIFNAME]) {
314 struct net_device *dev;
316 rule->oifindex = -1;
317 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
318 dev = __dev_get_by_name(net, rule->oifname);
319 if (dev)
320 rule->oifindex = dev->ifindex;
323 if (tb[FRA_FWMARK]) {
324 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
325 if (rule->mark)
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;
332 if (tb[FRA_FWMASK])
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);
342 err = -EINVAL;
343 if (tb[FRA_GOTO]) {
344 if (rule->action != FR_ACT_GOTO)
345 goto errout_free;
347 rule->target = nla_get_u32(tb[FRA_GOTO]);
348 /* Backward jumps are prohibited to avoid endless loops */
349 if (rule->target <= rule->pref)
350 goto errout_free;
352 list_for_each_entry(r, &ops->rules_list, list) {
353 if (r->pref == rule->target) {
354 RCU_INIT_POINTER(rule->ctarget, r);
355 break;
359 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
360 unresolved = 1;
361 } else if (rule->action == FR_ACT_GOTO)
362 goto errout_free;
364 err = ops->configure(rule, skb, frh, tb);
365 if (err < 0)
366 goto errout_free;
368 list_for_each_entry(r, &ops->rules_list, list) {
369 if (r->pref > rule->pref)
370 break;
371 last = r;
374 fib_rule_get(rule);
376 if (last)
377 list_add_rcu(&rule->list, &last->list);
378 else
379 list_add_rcu(&rule->list, &ops->rules_list);
381 if (ops->unresolved_rules) {
383 * There are unresolved goto rules in the list, check if
384 * any of them are pointing to this new rule.
386 list_for_each_entry(r, &ops->rules_list, list) {
387 if (r->action == FR_ACT_GOTO &&
388 r->target == rule->pref &&
389 rtnl_dereference(r->ctarget) == NULL) {
390 rcu_assign_pointer(r->ctarget, rule);
391 if (--ops->unresolved_rules == 0)
392 break;
397 if (rule->action == FR_ACT_GOTO)
398 ops->nr_goto_rules++;
400 if (unresolved)
401 ops->unresolved_rules++;
403 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
404 flush_route_cache(ops);
405 rules_ops_put(ops);
406 return 0;
408 errout_free:
409 release_net(rule->fr_net);
410 kfree(rule);
411 errout:
412 rules_ops_put(ops);
413 return err;
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];
423 int err = -EINVAL;
425 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
426 goto errout;
428 ops = lookup_rules_ops(net, frh->family);
429 if (ops == NULL) {
430 err = -EAFNOSUPPORT;
431 goto errout;
434 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
435 if (err < 0)
436 goto errout;
438 err = validate_rulemsg(frh, tb, ops);
439 if (err < 0)
440 goto errout;
442 list_for_each_entry(rule, &ops->rules_list, list) {
443 if (frh->action && (frh->action != rule->action))
444 continue;
446 if (frh->table && (frh_get_table(frh, tb) != rule->table))
447 continue;
449 if (tb[FRA_PRIORITY] &&
450 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
451 continue;
453 if (tb[FRA_IIFNAME] &&
454 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
455 continue;
457 if (tb[FRA_OIFNAME] &&
458 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
459 continue;
461 if (tb[FRA_FWMARK] &&
462 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
463 continue;
465 if (tb[FRA_FWMASK] &&
466 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
467 continue;
469 if (!ops->compare(rule, frh, tb))
470 continue;
472 if (rule->flags & FIB_RULE_PERMANENT) {
473 err = -EPERM;
474 goto errout;
477 list_del_rcu(&rule->list);
479 if (rule->action == FR_ACT_GOTO) {
480 ops->nr_goto_rules--;
481 if (rtnl_dereference(rule->ctarget) == NULL)
482 ops->unresolved_rules--;
486 * Check if this rule is a target to any of them. If so,
487 * disable them. As this operation is eventually very
488 * expensive, it is only performed if goto rules have
489 * actually been added.
491 if (ops->nr_goto_rules > 0) {
492 list_for_each_entry(tmp, &ops->rules_list, list) {
493 if (rtnl_dereference(tmp->ctarget) == rule) {
494 RCU_INIT_POINTER(tmp->ctarget, NULL);
495 ops->unresolved_rules++;
500 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
501 NETLINK_CB(skb).pid);
502 fib_rule_put(rule);
503 flush_route_cache(ops);
504 rules_ops_put(ops);
505 return 0;
508 err = -ENOENT;
509 errout:
510 rules_ops_put(ops);
511 return err;
514 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
515 struct fib_rule *rule)
517 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
518 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
519 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
520 + nla_total_size(4) /* FRA_PRIORITY */
521 + nla_total_size(4) /* FRA_TABLE */
522 + nla_total_size(4) /* FRA_FWMARK */
523 + nla_total_size(4); /* FRA_FWMASK */
525 if (ops->nlmsg_payload)
526 payload += ops->nlmsg_payload(rule);
528 return payload;
531 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
532 u32 pid, u32 seq, int type, int flags,
533 struct fib_rules_ops *ops)
535 struct nlmsghdr *nlh;
536 struct fib_rule_hdr *frh;
538 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
539 if (nlh == NULL)
540 return -EMSGSIZE;
542 frh = nlmsg_data(nlh);
543 frh->family = ops->family;
544 frh->table = rule->table;
545 NLA_PUT_U32(skb, FRA_TABLE, rule->table);
546 frh->res1 = 0;
547 frh->res2 = 0;
548 frh->action = rule->action;
549 frh->flags = rule->flags;
551 if (rule->action == FR_ACT_GOTO &&
552 rcu_access_pointer(rule->ctarget) == NULL)
553 frh->flags |= FIB_RULE_UNRESOLVED;
555 if (rule->iifname[0]) {
556 NLA_PUT_STRING(skb, FRA_IIFNAME, rule->iifname);
558 if (rule->iifindex == -1)
559 frh->flags |= FIB_RULE_IIF_DETACHED;
562 if (rule->oifname[0]) {
563 NLA_PUT_STRING(skb, FRA_OIFNAME, rule->oifname);
565 if (rule->oifindex == -1)
566 frh->flags |= FIB_RULE_OIF_DETACHED;
569 if (rule->pref)
570 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
572 if (rule->mark)
573 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
575 if (rule->mark_mask || rule->mark)
576 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
578 if (rule->target)
579 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
581 if (ops->fill(rule, skb, frh) < 0)
582 goto nla_put_failure;
584 return nlmsg_end(skb, nlh);
586 nla_put_failure:
587 nlmsg_cancel(skb, nlh);
588 return -EMSGSIZE;
591 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
592 struct fib_rules_ops *ops)
594 int idx = 0;
595 struct fib_rule *rule;
597 rcu_read_lock();
598 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
599 if (idx < cb->args[1])
600 goto skip;
602 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
603 cb->nlh->nlmsg_seq, RTM_NEWRULE,
604 NLM_F_MULTI, ops) < 0)
605 break;
606 skip:
607 idx++;
609 rcu_read_unlock();
610 cb->args[1] = idx;
611 rules_ops_put(ops);
613 return skb->len;
616 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
618 struct net *net = sock_net(skb->sk);
619 struct fib_rules_ops *ops;
620 int idx = 0, family;
622 family = rtnl_msg_family(cb->nlh);
623 if (family != AF_UNSPEC) {
624 /* Protocol specific dump request */
625 ops = lookup_rules_ops(net, family);
626 if (ops == NULL)
627 return -EAFNOSUPPORT;
629 return dump_rules(skb, cb, ops);
632 rcu_read_lock();
633 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
634 if (idx < cb->args[0] || !try_module_get(ops->owner))
635 goto skip;
637 if (dump_rules(skb, cb, ops) < 0)
638 break;
640 cb->args[1] = 0;
641 skip:
642 idx++;
644 rcu_read_unlock();
645 cb->args[0] = idx;
647 return skb->len;
650 static void notify_rule_change(int event, struct fib_rule *rule,
651 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
652 u32 pid)
654 struct net *net;
655 struct sk_buff *skb;
656 int err = -ENOBUFS;
658 net = ops->fro_net;
659 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
660 if (skb == NULL)
661 goto errout;
663 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
664 if (err < 0) {
665 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
666 WARN_ON(err == -EMSGSIZE);
667 kfree_skb(skb);
668 goto errout;
671 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
672 return;
673 errout:
674 if (err < 0)
675 rtnl_set_sk_err(net, ops->nlgroup, err);
678 static void attach_rules(struct list_head *rules, struct net_device *dev)
680 struct fib_rule *rule;
682 list_for_each_entry(rule, rules, list) {
683 if (rule->iifindex == -1 &&
684 strcmp(dev->name, rule->iifname) == 0)
685 rule->iifindex = dev->ifindex;
686 if (rule->oifindex == -1 &&
687 strcmp(dev->name, rule->oifname) == 0)
688 rule->oifindex = dev->ifindex;
692 static void detach_rules(struct list_head *rules, struct net_device *dev)
694 struct fib_rule *rule;
696 list_for_each_entry(rule, rules, list) {
697 if (rule->iifindex == dev->ifindex)
698 rule->iifindex = -1;
699 if (rule->oifindex == dev->ifindex)
700 rule->oifindex = -1;
705 static int fib_rules_event(struct notifier_block *this, unsigned long event,
706 void *ptr)
708 struct net_device *dev = ptr;
709 struct net *net = dev_net(dev);
710 struct fib_rules_ops *ops;
712 ASSERT_RTNL();
714 switch (event) {
715 case NETDEV_REGISTER:
716 list_for_each_entry(ops, &net->rules_ops, list)
717 attach_rules(&ops->rules_list, dev);
718 break;
720 case NETDEV_UNREGISTER:
721 list_for_each_entry(ops, &net->rules_ops, list)
722 detach_rules(&ops->rules_list, dev);
723 break;
726 return NOTIFY_DONE;
729 static struct notifier_block fib_rules_notifier = {
730 .notifier_call = fib_rules_event,
733 static int __net_init fib_rules_net_init(struct net *net)
735 INIT_LIST_HEAD(&net->rules_ops);
736 spin_lock_init(&net->rules_mod_lock);
737 return 0;
740 static struct pernet_operations fib_rules_net_ops = {
741 .init = fib_rules_net_init,
744 static int __init fib_rules_init(void)
746 int err;
747 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
748 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
749 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
751 err = register_pernet_subsys(&fib_rules_net_ops);
752 if (err < 0)
753 goto fail;
755 err = register_netdevice_notifier(&fib_rules_notifier);
756 if (err < 0)
757 goto fail_unregister;
759 return 0;
761 fail_unregister:
762 unregister_pernet_subsys(&fib_rules_net_ops);
763 fail:
764 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
765 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
766 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
767 return err;
770 subsys_initcall(fib_rules_init);