dm thin: wake worker when discard is prepared
[linux-2.6.git] / net / core / fib_rules.c
blob58a4ba27dfe3117d439ada122000e1339a23f48f
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 if (ops->delete)
155 ops->delete(rule);
156 fib_rule_put(rule);
160 static void fib_rules_put_rcu(struct rcu_head *head)
162 struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu);
163 struct net *net = ops->fro_net;
165 release_net(net);
166 kfree(ops);
169 void fib_rules_unregister(struct fib_rules_ops *ops)
171 struct net *net = ops->fro_net;
173 spin_lock(&net->rules_mod_lock);
174 list_del_rcu(&ops->list);
175 fib_rules_cleanup_ops(ops);
176 spin_unlock(&net->rules_mod_lock);
178 call_rcu(&ops->rcu, fib_rules_put_rcu);
180 EXPORT_SYMBOL_GPL(fib_rules_unregister);
182 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
183 struct flowi *fl, int flags)
185 int ret = 0;
187 if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
188 goto out;
190 if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
191 goto out;
193 if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
194 goto out;
196 ret = ops->match(rule, fl, flags);
197 out:
198 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
201 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
202 int flags, struct fib_lookup_arg *arg)
204 struct fib_rule *rule;
205 int err;
207 rcu_read_lock();
209 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
210 jumped:
211 if (!fib_rule_match(rule, ops, fl, flags))
212 continue;
214 if (rule->action == FR_ACT_GOTO) {
215 struct fib_rule *target;
217 target = rcu_dereference(rule->ctarget);
218 if (target == NULL) {
219 continue;
220 } else {
221 rule = target;
222 goto jumped;
224 } else if (rule->action == FR_ACT_NOP)
225 continue;
226 else
227 err = ops->action(rule, fl, flags, arg);
229 if (err != -EAGAIN) {
230 if ((arg->flags & FIB_LOOKUP_NOREF) ||
231 likely(atomic_inc_not_zero(&rule->refcnt))) {
232 arg->rule = rule;
233 goto out;
235 break;
239 err = -ESRCH;
240 out:
241 rcu_read_unlock();
243 return err;
245 EXPORT_SYMBOL_GPL(fib_rules_lookup);
247 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
248 struct fib_rules_ops *ops)
250 int err = -EINVAL;
252 if (frh->src_len)
253 if (tb[FRA_SRC] == NULL ||
254 frh->src_len > (ops->addr_size * 8) ||
255 nla_len(tb[FRA_SRC]) != ops->addr_size)
256 goto errout;
258 if (frh->dst_len)
259 if (tb[FRA_DST] == NULL ||
260 frh->dst_len > (ops->addr_size * 8) ||
261 nla_len(tb[FRA_DST]) != ops->addr_size)
262 goto errout;
264 err = 0;
265 errout:
266 return err;
269 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
271 struct net *net = sock_net(skb->sk);
272 struct fib_rule_hdr *frh = nlmsg_data(nlh);
273 struct fib_rules_ops *ops = NULL;
274 struct fib_rule *rule, *r, *last = NULL;
275 struct nlattr *tb[FRA_MAX+1];
276 int err = -EINVAL, unresolved = 0;
278 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
279 goto errout;
281 ops = lookup_rules_ops(net, frh->family);
282 if (ops == NULL) {
283 err = -EAFNOSUPPORT;
284 goto errout;
287 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
288 if (err < 0)
289 goto errout;
291 err = validate_rulemsg(frh, tb, ops);
292 if (err < 0)
293 goto errout;
295 rule = kzalloc(ops->rule_size, GFP_KERNEL);
296 if (rule == NULL) {
297 err = -ENOMEM;
298 goto errout;
300 rule->fr_net = hold_net(net);
302 if (tb[FRA_PRIORITY])
303 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
305 if (tb[FRA_IIFNAME]) {
306 struct net_device *dev;
308 rule->iifindex = -1;
309 nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
310 dev = __dev_get_by_name(net, rule->iifname);
311 if (dev)
312 rule->iifindex = dev->ifindex;
315 if (tb[FRA_OIFNAME]) {
316 struct net_device *dev;
318 rule->oifindex = -1;
319 nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
320 dev = __dev_get_by_name(net, rule->oifname);
321 if (dev)
322 rule->oifindex = dev->ifindex;
325 if (tb[FRA_FWMARK]) {
326 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
327 if (rule->mark)
328 /* compatibility: if the mark value is non-zero all bits
329 * are compared unless a mask is explicitly specified.
331 rule->mark_mask = 0xFFFFFFFF;
334 if (tb[FRA_FWMASK])
335 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
337 rule->action = frh->action;
338 rule->flags = frh->flags;
339 rule->table = frh_get_table(frh, tb);
341 if (!tb[FRA_PRIORITY] && ops->default_pref)
342 rule->pref = ops->default_pref(ops);
344 err = -EINVAL;
345 if (tb[FRA_GOTO]) {
346 if (rule->action != FR_ACT_GOTO)
347 goto errout_free;
349 rule->target = nla_get_u32(tb[FRA_GOTO]);
350 /* Backward jumps are prohibited to avoid endless loops */
351 if (rule->target <= rule->pref)
352 goto errout_free;
354 list_for_each_entry(r, &ops->rules_list, list) {
355 if (r->pref == rule->target) {
356 RCU_INIT_POINTER(rule->ctarget, r);
357 break;
361 if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
362 unresolved = 1;
363 } else if (rule->action == FR_ACT_GOTO)
364 goto errout_free;
366 err = ops->configure(rule, skb, frh, tb);
367 if (err < 0)
368 goto errout_free;
370 list_for_each_entry(r, &ops->rules_list, list) {
371 if (r->pref > rule->pref)
372 break;
373 last = r;
376 fib_rule_get(rule);
378 if (last)
379 list_add_rcu(&rule->list, &last->list);
380 else
381 list_add_rcu(&rule->list, &ops->rules_list);
383 if (ops->unresolved_rules) {
385 * There are unresolved goto rules in the list, check if
386 * any of them are pointing to this new rule.
388 list_for_each_entry(r, &ops->rules_list, list) {
389 if (r->action == FR_ACT_GOTO &&
390 r->target == rule->pref &&
391 rtnl_dereference(r->ctarget) == NULL) {
392 rcu_assign_pointer(r->ctarget, rule);
393 if (--ops->unresolved_rules == 0)
394 break;
399 if (rule->action == FR_ACT_GOTO)
400 ops->nr_goto_rules++;
402 if (unresolved)
403 ops->unresolved_rules++;
405 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
406 flush_route_cache(ops);
407 rules_ops_put(ops);
408 return 0;
410 errout_free:
411 release_net(rule->fr_net);
412 kfree(rule);
413 errout:
414 rules_ops_put(ops);
415 return err;
418 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
420 struct net *net = sock_net(skb->sk);
421 struct fib_rule_hdr *frh = nlmsg_data(nlh);
422 struct fib_rules_ops *ops = NULL;
423 struct fib_rule *rule, *tmp;
424 struct nlattr *tb[FRA_MAX+1];
425 int err = -EINVAL;
427 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
428 goto errout;
430 ops = lookup_rules_ops(net, frh->family);
431 if (ops == NULL) {
432 err = -EAFNOSUPPORT;
433 goto errout;
436 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
437 if (err < 0)
438 goto errout;
440 err = validate_rulemsg(frh, tb, ops);
441 if (err < 0)
442 goto errout;
444 list_for_each_entry(rule, &ops->rules_list, list) {
445 if (frh->action && (frh->action != rule->action))
446 continue;
448 if (frh->table && (frh_get_table(frh, tb) != rule->table))
449 continue;
451 if (tb[FRA_PRIORITY] &&
452 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
453 continue;
455 if (tb[FRA_IIFNAME] &&
456 nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
457 continue;
459 if (tb[FRA_OIFNAME] &&
460 nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
461 continue;
463 if (tb[FRA_FWMARK] &&
464 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
465 continue;
467 if (tb[FRA_FWMASK] &&
468 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
469 continue;
471 if (!ops->compare(rule, frh, tb))
472 continue;
474 if (rule->flags & FIB_RULE_PERMANENT) {
475 err = -EPERM;
476 goto errout;
479 list_del_rcu(&rule->list);
481 if (rule->action == FR_ACT_GOTO) {
482 ops->nr_goto_rules--;
483 if (rtnl_dereference(rule->ctarget) == NULL)
484 ops->unresolved_rules--;
488 * Check if this rule is a target to any of them. If so,
489 * disable them. As this operation is eventually very
490 * expensive, it is only performed if goto rules have
491 * actually been added.
493 if (ops->nr_goto_rules > 0) {
494 list_for_each_entry(tmp, &ops->rules_list, list) {
495 if (rtnl_dereference(tmp->ctarget) == rule) {
496 RCU_INIT_POINTER(tmp->ctarget, NULL);
497 ops->unresolved_rules++;
502 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
503 NETLINK_CB(skb).portid);
504 if (ops->delete)
505 ops->delete(rule);
506 fib_rule_put(rule);
507 flush_route_cache(ops);
508 rules_ops_put(ops);
509 return 0;
512 err = -ENOENT;
513 errout:
514 rules_ops_put(ops);
515 return err;
518 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
519 struct fib_rule *rule)
521 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
522 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
523 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
524 + nla_total_size(4) /* FRA_PRIORITY */
525 + nla_total_size(4) /* FRA_TABLE */
526 + nla_total_size(4) /* FRA_FWMARK */
527 + nla_total_size(4); /* FRA_FWMASK */
529 if (ops->nlmsg_payload)
530 payload += ops->nlmsg_payload(rule);
532 return payload;
535 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
536 u32 pid, u32 seq, int type, int flags,
537 struct fib_rules_ops *ops)
539 struct nlmsghdr *nlh;
540 struct fib_rule_hdr *frh;
542 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
543 if (nlh == NULL)
544 return -EMSGSIZE;
546 frh = nlmsg_data(nlh);
547 frh->family = ops->family;
548 frh->table = rule->table;
549 if (nla_put_u32(skb, FRA_TABLE, rule->table))
550 goto nla_put_failure;
551 frh->res1 = 0;
552 frh->res2 = 0;
553 frh->action = rule->action;
554 frh->flags = rule->flags;
556 if (rule->action == FR_ACT_GOTO &&
557 rcu_access_pointer(rule->ctarget) == NULL)
558 frh->flags |= FIB_RULE_UNRESOLVED;
560 if (rule->iifname[0]) {
561 if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
562 goto nla_put_failure;
563 if (rule->iifindex == -1)
564 frh->flags |= FIB_RULE_IIF_DETACHED;
567 if (rule->oifname[0]) {
568 if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
569 goto nla_put_failure;
570 if (rule->oifindex == -1)
571 frh->flags |= FIB_RULE_OIF_DETACHED;
574 if ((rule->pref &&
575 nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
576 (rule->mark &&
577 nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
578 ((rule->mark_mask || rule->mark) &&
579 nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
580 (rule->target &&
581 nla_put_u32(skb, FRA_GOTO, rule->target)))
582 goto nla_put_failure;
583 if (ops->fill(rule, skb, frh) < 0)
584 goto nla_put_failure;
586 return nlmsg_end(skb, nlh);
588 nla_put_failure:
589 nlmsg_cancel(skb, nlh);
590 return -EMSGSIZE;
593 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
594 struct fib_rules_ops *ops)
596 int idx = 0;
597 struct fib_rule *rule;
599 rcu_read_lock();
600 list_for_each_entry_rcu(rule, &ops->rules_list, list) {
601 if (idx < cb->args[1])
602 goto skip;
604 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
605 cb->nlh->nlmsg_seq, RTM_NEWRULE,
606 NLM_F_MULTI, ops) < 0)
607 break;
608 skip:
609 idx++;
611 rcu_read_unlock();
612 cb->args[1] = idx;
613 rules_ops_put(ops);
615 return skb->len;
618 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
620 struct net *net = sock_net(skb->sk);
621 struct fib_rules_ops *ops;
622 int idx = 0, family;
624 family = rtnl_msg_family(cb->nlh);
625 if (family != AF_UNSPEC) {
626 /* Protocol specific dump request */
627 ops = lookup_rules_ops(net, family);
628 if (ops == NULL)
629 return -EAFNOSUPPORT;
631 return dump_rules(skb, cb, ops);
634 rcu_read_lock();
635 list_for_each_entry_rcu(ops, &net->rules_ops, list) {
636 if (idx < cb->args[0] || !try_module_get(ops->owner))
637 goto skip;
639 if (dump_rules(skb, cb, ops) < 0)
640 break;
642 cb->args[1] = 0;
643 skip:
644 idx++;
646 rcu_read_unlock();
647 cb->args[0] = idx;
649 return skb->len;
652 static void notify_rule_change(int event, struct fib_rule *rule,
653 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
654 u32 pid)
656 struct net *net;
657 struct sk_buff *skb;
658 int err = -ENOBUFS;
660 net = ops->fro_net;
661 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
662 if (skb == NULL)
663 goto errout;
665 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
666 if (err < 0) {
667 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
668 WARN_ON(err == -EMSGSIZE);
669 kfree_skb(skb);
670 goto errout;
673 rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
674 return;
675 errout:
676 if (err < 0)
677 rtnl_set_sk_err(net, ops->nlgroup, err);
680 static void attach_rules(struct list_head *rules, struct net_device *dev)
682 struct fib_rule *rule;
684 list_for_each_entry(rule, rules, list) {
685 if (rule->iifindex == -1 &&
686 strcmp(dev->name, rule->iifname) == 0)
687 rule->iifindex = dev->ifindex;
688 if (rule->oifindex == -1 &&
689 strcmp(dev->name, rule->oifname) == 0)
690 rule->oifindex = dev->ifindex;
694 static void detach_rules(struct list_head *rules, struct net_device *dev)
696 struct fib_rule *rule;
698 list_for_each_entry(rule, rules, list) {
699 if (rule->iifindex == dev->ifindex)
700 rule->iifindex = -1;
701 if (rule->oifindex == dev->ifindex)
702 rule->oifindex = -1;
707 static int fib_rules_event(struct notifier_block *this, unsigned long event,
708 void *ptr)
710 struct net_device *dev = ptr;
711 struct net *net = dev_net(dev);
712 struct fib_rules_ops *ops;
714 ASSERT_RTNL();
716 switch (event) {
717 case NETDEV_REGISTER:
718 list_for_each_entry(ops, &net->rules_ops, list)
719 attach_rules(&ops->rules_list, dev);
720 break;
722 case NETDEV_UNREGISTER:
723 list_for_each_entry(ops, &net->rules_ops, list)
724 detach_rules(&ops->rules_list, dev);
725 break;
728 return NOTIFY_DONE;
731 static struct notifier_block fib_rules_notifier = {
732 .notifier_call = fib_rules_event,
735 static int __net_init fib_rules_net_init(struct net *net)
737 INIT_LIST_HEAD(&net->rules_ops);
738 spin_lock_init(&net->rules_mod_lock);
739 return 0;
742 static struct pernet_operations fib_rules_net_ops = {
743 .init = fib_rules_net_init,
746 static int __init fib_rules_init(void)
748 int err;
749 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
750 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
751 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
753 err = register_pernet_subsys(&fib_rules_net_ops);
754 if (err < 0)
755 goto fail;
757 err = register_netdevice_notifier(&fib_rules_notifier);
758 if (err < 0)
759 goto fail_unregister;
761 return 0;
763 fail_unregister:
764 unregister_pernet_subsys(&fib_rules_net_ops);
765 fail:
766 rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
767 rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
768 rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
769 return err;
772 subsys_initcall(fib_rules_init);