Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
[pv_ops_mirror.git] / net / core / fib_rules.c
blob8c5474e16683aa22971b1bed0390354b728a0f64
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/list.h>
14 #include <net/fib_rules.h>
16 static LIST_HEAD(rules_ops);
17 static DEFINE_SPINLOCK(rules_mod_lock);
19 static void notify_rule_change(int event, struct fib_rule *rule,
20 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
21 u32 pid);
23 static struct fib_rules_ops *lookup_rules_ops(int family)
25 struct fib_rules_ops *ops;
27 rcu_read_lock();
28 list_for_each_entry_rcu(ops, &rules_ops, list) {
29 if (ops->family == family) {
30 if (!try_module_get(ops->owner))
31 ops = NULL;
32 rcu_read_unlock();
33 return ops;
36 rcu_read_unlock();
38 return NULL;
41 static void rules_ops_put(struct fib_rules_ops *ops)
43 if (ops)
44 module_put(ops->owner);
47 static void flush_route_cache(struct fib_rules_ops *ops)
49 if (ops->flush_cache)
50 ops->flush_cache();
53 int fib_rules_register(struct fib_rules_ops *ops)
55 int err = -EEXIST;
56 struct fib_rules_ops *o;
58 if (ops->rule_size < sizeof(struct fib_rule))
59 return -EINVAL;
61 if (ops->match == NULL || ops->configure == NULL ||
62 ops->compare == NULL || ops->fill == NULL ||
63 ops->action == NULL)
64 return -EINVAL;
66 spin_lock(&rules_mod_lock);
67 list_for_each_entry(o, &rules_ops, list)
68 if (ops->family == o->family)
69 goto errout;
71 list_add_tail_rcu(&ops->list, &rules_ops);
72 err = 0;
73 errout:
74 spin_unlock(&rules_mod_lock);
76 return err;
79 EXPORT_SYMBOL_GPL(fib_rules_register);
81 static void cleanup_ops(struct fib_rules_ops *ops)
83 struct fib_rule *rule, *tmp;
85 list_for_each_entry_safe(rule, tmp, ops->rules_list, list) {
86 list_del_rcu(&rule->list);
87 fib_rule_put(rule);
91 int fib_rules_unregister(struct fib_rules_ops *ops)
93 int err = 0;
94 struct fib_rules_ops *o;
96 spin_lock(&rules_mod_lock);
97 list_for_each_entry(o, &rules_ops, list) {
98 if (o == ops) {
99 list_del_rcu(&o->list);
100 cleanup_ops(ops);
101 goto out;
105 err = -ENOENT;
106 out:
107 spin_unlock(&rules_mod_lock);
109 synchronize_rcu();
111 return err;
114 EXPORT_SYMBOL_GPL(fib_rules_unregister);
116 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
117 struct flowi *fl, int flags)
119 int ret = 0;
121 if (rule->ifindex && (rule->ifindex != fl->iif))
122 goto out;
124 if ((rule->mark ^ fl->mark) & rule->mark_mask)
125 goto out;
127 ret = ops->match(rule, fl, flags);
128 out:
129 return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
132 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
133 int flags, struct fib_lookup_arg *arg)
135 struct fib_rule *rule;
136 int err;
138 rcu_read_lock();
140 list_for_each_entry_rcu(rule, ops->rules_list, list) {
141 jumped:
142 if (!fib_rule_match(rule, ops, fl, flags))
143 continue;
145 if (rule->action == FR_ACT_GOTO) {
146 struct fib_rule *target;
148 target = rcu_dereference(rule->ctarget);
149 if (target == NULL) {
150 continue;
151 } else {
152 rule = target;
153 goto jumped;
155 } else if (rule->action == FR_ACT_NOP)
156 continue;
157 else
158 err = ops->action(rule, fl, flags, arg);
160 if (err != -EAGAIN) {
161 fib_rule_get(rule);
162 arg->rule = rule;
163 goto out;
167 err = -ESRCH;
168 out:
169 rcu_read_unlock();
171 return err;
174 EXPORT_SYMBOL_GPL(fib_rules_lookup);
176 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
177 struct fib_rules_ops *ops)
179 int err = -EINVAL;
181 if (frh->src_len)
182 if (tb[FRA_SRC] == NULL ||
183 frh->src_len > (ops->addr_size * 8) ||
184 nla_len(tb[FRA_SRC]) != ops->addr_size)
185 goto errout;
187 if (frh->dst_len)
188 if (tb[FRA_DST] == NULL ||
189 frh->dst_len > (ops->addr_size * 8) ||
190 nla_len(tb[FRA_DST]) != ops->addr_size)
191 goto errout;
193 err = 0;
194 errout:
195 return err;
198 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
200 struct fib_rule_hdr *frh = nlmsg_data(nlh);
201 struct fib_rules_ops *ops = NULL;
202 struct fib_rule *rule, *r, *last = NULL;
203 struct nlattr *tb[FRA_MAX+1];
204 int err = -EINVAL, unresolved = 0;
206 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
207 goto errout;
209 ops = lookup_rules_ops(frh->family);
210 if (ops == NULL) {
211 err = EAFNOSUPPORT;
212 goto errout;
215 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
216 if (err < 0)
217 goto errout;
219 err = validate_rulemsg(frh, tb, ops);
220 if (err < 0)
221 goto errout;
223 rule = kzalloc(ops->rule_size, GFP_KERNEL);
224 if (rule == NULL) {
225 err = -ENOMEM;
226 goto errout;
229 if (tb[FRA_PRIORITY])
230 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
232 if (tb[FRA_IFNAME]) {
233 struct net_device *dev;
235 rule->ifindex = -1;
236 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
237 dev = __dev_get_by_name(rule->ifname);
238 if (dev)
239 rule->ifindex = dev->ifindex;
242 if (tb[FRA_FWMARK]) {
243 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
244 if (rule->mark)
245 /* compatibility: if the mark value is non-zero all bits
246 * are compared unless a mask is explicitly specified.
248 rule->mark_mask = 0xFFFFFFFF;
251 if (tb[FRA_FWMASK])
252 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
254 rule->action = frh->action;
255 rule->flags = frh->flags;
256 rule->table = frh_get_table(frh, tb);
258 if (!rule->pref && ops->default_pref)
259 rule->pref = ops->default_pref();
261 err = -EINVAL;
262 if (tb[FRA_GOTO]) {
263 if (rule->action != FR_ACT_GOTO)
264 goto errout_free;
266 rule->target = nla_get_u32(tb[FRA_GOTO]);
267 /* Backward jumps are prohibited to avoid endless loops */
268 if (rule->target <= rule->pref)
269 goto errout_free;
271 list_for_each_entry(r, ops->rules_list, list) {
272 if (r->pref == rule->target) {
273 rule->ctarget = r;
274 break;
278 if (rule->ctarget == NULL)
279 unresolved = 1;
280 } else if (rule->action == FR_ACT_GOTO)
281 goto errout_free;
283 err = ops->configure(rule, skb, nlh, frh, tb);
284 if (err < 0)
285 goto errout_free;
287 list_for_each_entry(r, ops->rules_list, list) {
288 if (r->pref > rule->pref)
289 break;
290 last = r;
293 fib_rule_get(rule);
295 if (ops->unresolved_rules) {
297 * There are unresolved goto rules in the list, check if
298 * any of them are pointing to this new rule.
300 list_for_each_entry(r, ops->rules_list, list) {
301 if (r->action == FR_ACT_GOTO &&
302 r->target == rule->pref) {
303 BUG_ON(r->ctarget != NULL);
304 rcu_assign_pointer(r->ctarget, rule);
305 if (--ops->unresolved_rules == 0)
306 break;
311 if (rule->action == FR_ACT_GOTO)
312 ops->nr_goto_rules++;
314 if (unresolved)
315 ops->unresolved_rules++;
317 if (last)
318 list_add_rcu(&rule->list, &last->list);
319 else
320 list_add_rcu(&rule->list, ops->rules_list);
322 notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
323 flush_route_cache(ops);
324 rules_ops_put(ops);
325 return 0;
327 errout_free:
328 kfree(rule);
329 errout:
330 rules_ops_put(ops);
331 return err;
334 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
336 struct fib_rule_hdr *frh = nlmsg_data(nlh);
337 struct fib_rules_ops *ops = NULL;
338 struct fib_rule *rule, *tmp;
339 struct nlattr *tb[FRA_MAX+1];
340 int err = -EINVAL;
342 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
343 goto errout;
345 ops = lookup_rules_ops(frh->family);
346 if (ops == NULL) {
347 err = EAFNOSUPPORT;
348 goto errout;
351 err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
352 if (err < 0)
353 goto errout;
355 err = validate_rulemsg(frh, tb, ops);
356 if (err < 0)
357 goto errout;
359 list_for_each_entry(rule, ops->rules_list, list) {
360 if (frh->action && (frh->action != rule->action))
361 continue;
363 if (frh->table && (frh_get_table(frh, tb) != rule->table))
364 continue;
366 if (tb[FRA_PRIORITY] &&
367 (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
368 continue;
370 if (tb[FRA_IFNAME] &&
371 nla_strcmp(tb[FRA_IFNAME], rule->ifname))
372 continue;
374 if (tb[FRA_FWMARK] &&
375 (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
376 continue;
378 if (tb[FRA_FWMASK] &&
379 (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
380 continue;
382 if (!ops->compare(rule, frh, tb))
383 continue;
385 if (rule->flags & FIB_RULE_PERMANENT) {
386 err = -EPERM;
387 goto errout;
390 list_del_rcu(&rule->list);
392 if (rule->action == FR_ACT_GOTO)
393 ops->nr_goto_rules--;
396 * Check if this rule is a target to any of them. If so,
397 * disable them. As this operation is eventually very
398 * expensive, it is only performed if goto rules have
399 * actually been added.
401 if (ops->nr_goto_rules > 0) {
402 list_for_each_entry(tmp, ops->rules_list, list) {
403 if (tmp->ctarget == rule) {
404 rcu_assign_pointer(tmp->ctarget, NULL);
405 ops->unresolved_rules++;
410 synchronize_rcu();
411 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
412 NETLINK_CB(skb).pid);
413 fib_rule_put(rule);
414 flush_route_cache(ops);
415 rules_ops_put(ops);
416 return 0;
419 err = -ENOENT;
420 errout:
421 rules_ops_put(ops);
422 return err;
425 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
426 struct fib_rule *rule)
428 size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
429 + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
430 + nla_total_size(4) /* FRA_PRIORITY */
431 + nla_total_size(4) /* FRA_TABLE */
432 + nla_total_size(4) /* FRA_FWMARK */
433 + nla_total_size(4); /* FRA_FWMASK */
435 if (ops->nlmsg_payload)
436 payload += ops->nlmsg_payload(rule);
438 return payload;
441 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
442 u32 pid, u32 seq, int type, int flags,
443 struct fib_rules_ops *ops)
445 struct nlmsghdr *nlh;
446 struct fib_rule_hdr *frh;
448 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
449 if (nlh == NULL)
450 return -EMSGSIZE;
452 frh = nlmsg_data(nlh);
453 frh->table = rule->table;
454 NLA_PUT_U32(skb, FRA_TABLE, rule->table);
455 frh->res1 = 0;
456 frh->res2 = 0;
457 frh->action = rule->action;
458 frh->flags = rule->flags;
460 if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
461 frh->flags |= FIB_RULE_UNRESOLVED;
463 if (rule->ifname[0]) {
464 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
466 if (rule->ifindex == -1)
467 frh->flags |= FIB_RULE_DEV_DETACHED;
470 if (rule->pref)
471 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
473 if (rule->mark)
474 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
476 if (rule->mark_mask || rule->mark)
477 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
479 if (rule->target)
480 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
482 if (ops->fill(rule, skb, nlh, frh) < 0)
483 goto nla_put_failure;
485 return nlmsg_end(skb, nlh);
487 nla_put_failure:
488 nlmsg_cancel(skb, nlh);
489 return -EMSGSIZE;
492 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
493 struct fib_rules_ops *ops)
495 int idx = 0;
496 struct fib_rule *rule;
498 list_for_each_entry(rule, ops->rules_list, list) {
499 if (idx < cb->args[1])
500 goto skip;
502 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
503 cb->nlh->nlmsg_seq, RTM_NEWRULE,
504 NLM_F_MULTI, ops) < 0)
505 break;
506 skip:
507 idx++;
509 cb->args[1] = idx;
510 rules_ops_put(ops);
512 return skb->len;
515 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
517 struct fib_rules_ops *ops;
518 int idx = 0, family;
520 family = rtnl_msg_family(cb->nlh);
521 if (family != AF_UNSPEC) {
522 /* Protocol specific dump request */
523 ops = lookup_rules_ops(family);
524 if (ops == NULL)
525 return -EAFNOSUPPORT;
527 return dump_rules(skb, cb, ops);
530 rcu_read_lock();
531 list_for_each_entry_rcu(ops, &rules_ops, list) {
532 if (idx < cb->args[0] || !try_module_get(ops->owner))
533 goto skip;
535 if (dump_rules(skb, cb, ops) < 0)
536 break;
538 cb->args[1] = 0;
539 skip:
540 idx++;
542 rcu_read_unlock();
543 cb->args[0] = idx;
545 return skb->len;
548 static void notify_rule_change(int event, struct fib_rule *rule,
549 struct fib_rules_ops *ops, struct nlmsghdr *nlh,
550 u32 pid)
552 struct sk_buff *skb;
553 int err = -ENOBUFS;
555 skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
556 if (skb == NULL)
557 goto errout;
559 err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
560 if (err < 0) {
561 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
562 WARN_ON(err == -EMSGSIZE);
563 kfree_skb(skb);
564 goto errout;
566 err = rtnl_notify(skb, pid, ops->nlgroup, nlh, GFP_KERNEL);
567 errout:
568 if (err < 0)
569 rtnl_set_sk_err(ops->nlgroup, err);
572 static void attach_rules(struct list_head *rules, struct net_device *dev)
574 struct fib_rule *rule;
576 list_for_each_entry(rule, rules, list) {
577 if (rule->ifindex == -1 &&
578 strcmp(dev->name, rule->ifname) == 0)
579 rule->ifindex = dev->ifindex;
583 static void detach_rules(struct list_head *rules, struct net_device *dev)
585 struct fib_rule *rule;
587 list_for_each_entry(rule, rules, list)
588 if (rule->ifindex == dev->ifindex)
589 rule->ifindex = -1;
593 static int fib_rules_event(struct notifier_block *this, unsigned long event,
594 void *ptr)
596 struct net_device *dev = ptr;
597 struct fib_rules_ops *ops;
599 ASSERT_RTNL();
600 rcu_read_lock();
602 switch (event) {
603 case NETDEV_REGISTER:
604 list_for_each_entry(ops, &rules_ops, list)
605 attach_rules(ops->rules_list, dev);
606 break;
608 case NETDEV_UNREGISTER:
609 list_for_each_entry(ops, &rules_ops, list)
610 detach_rules(ops->rules_list, dev);
611 break;
614 rcu_read_unlock();
616 return NOTIFY_DONE;
619 static struct notifier_block fib_rules_notifier = {
620 .notifier_call = fib_rules_event,
623 static int __init fib_rules_init(void)
625 rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
626 rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
627 rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
629 return register_netdevice_notifier(&fib_rules_notifier);
632 subsys_initcall(fib_rules_init);