[NET] fib_rules: Flush route cache after rule modifications
[linux-2.6.git] / net / core / fib_rules.c
1 /*
2  * net/core/fib_rules.c         Generic Routing Rules
3  *
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.
7  *
8  * Authors:     Thomas Graf <tgraf@suug.ch>
9  */
10
11 #include <linux/types.h>
12 #include <linux/kernel.h>
13 #include <linux/list.h>
14 #include <net/fib_rules.h>
15
16 static LIST_HEAD(rules_ops);
17 static DEFINE_SPINLOCK(rules_mod_lock);
18
19 static void notify_rule_change(int event, struct fib_rule *rule,
20                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
21                                u32 pid);
22
23 static struct fib_rules_ops *lookup_rules_ops(int family)
24 {
25         struct fib_rules_ops *ops;
26
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;
34                 }
35         }
36         rcu_read_unlock();
37
38         return NULL;
39 }
40
41 static void rules_ops_put(struct fib_rules_ops *ops)
42 {
43         if (ops)
44                 module_put(ops->owner);
45 }
46
47 static void flush_route_cache(struct fib_rules_ops *ops)
48 {
49         if (ops->flush_cache)
50                 ops->flush_cache();
51 }
52
53 int fib_rules_register(struct fib_rules_ops *ops)
54 {
55         int err = -EEXIST;
56         struct fib_rules_ops *o;
57
58         if (ops->rule_size < sizeof(struct fib_rule))
59                 return -EINVAL;
60
61         if (ops->match == NULL || ops->configure == NULL ||
62             ops->compare == NULL || ops->fill == NULL ||
63             ops->action == NULL)
64                 return -EINVAL;
65
66         spin_lock(&rules_mod_lock);
67         list_for_each_entry(o, &rules_ops, list)
68                 if (ops->family == o->family)
69                         goto errout;
70
71         list_add_tail_rcu(&ops->list, &rules_ops);
72         err = 0;
73 errout:
74         spin_unlock(&rules_mod_lock);
75
76         return err;
77 }
78
79 EXPORT_SYMBOL_GPL(fib_rules_register);
80
81 static void cleanup_ops(struct fib_rules_ops *ops)
82 {
83         struct fib_rule *rule, *tmp;
84
85         list_for_each_entry_safe(rule, tmp, ops->rules_list, list) {
86                 list_del_rcu(&rule->list);
87                 fib_rule_put(rule);
88         }
89 }
90
91 int fib_rules_unregister(struct fib_rules_ops *ops)
92 {
93         int err = 0;
94         struct fib_rules_ops *o;
95
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;
102                 }
103         }
104
105         err = -ENOENT;
106 out:
107         spin_unlock(&rules_mod_lock);
108
109         synchronize_rcu();
110
111         return err;
112 }
113
114 EXPORT_SYMBOL_GPL(fib_rules_unregister);
115
116 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
117                           struct flowi *fl, int flags)
118 {
119         int ret = 0;
120
121         if (rule->ifindex && (rule->ifindex != fl->iif))
122                 goto out;
123
124         if ((rule->mark ^ fl->mark) & rule->mark_mask)
125                 goto out;
126
127         ret = ops->match(rule, fl, flags);
128 out:
129         return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
130 }
131
132 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
133                      int flags, struct fib_lookup_arg *arg)
134 {
135         struct fib_rule *rule;
136         int err;
137
138         rcu_read_lock();
139
140         list_for_each_entry_rcu(rule, ops->rules_list, list) {
141 jumped:
142                 if (!fib_rule_match(rule, ops, fl, flags))
143                         continue;
144
145                 if (rule->action == FR_ACT_GOTO) {
146                         struct fib_rule *target;
147
148                         target = rcu_dereference(rule->ctarget);
149                         if (target == NULL) {
150                                 continue;
151                         } else {
152                                 rule = target;
153                                 goto jumped;
154                         }
155                 } else if (rule->action == FR_ACT_NOP)
156                         continue;
157                 else
158                         err = ops->action(rule, fl, flags, arg);
159
160                 if (err != -EAGAIN) {
161                         fib_rule_get(rule);
162                         arg->rule = rule;
163                         goto out;
164                 }
165         }
166
167         err = -ESRCH;
168 out:
169         rcu_read_unlock();
170
171         return err;
172 }
173
174 EXPORT_SYMBOL_GPL(fib_rules_lookup);
175
176 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
177                             struct fib_rules_ops *ops)
178 {
179         int err = -EINVAL;
180
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;
186
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;
192
193         err = 0;
194 errout:
195         return err;
196 }
197
198 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
199 {
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;
205
206         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
207                 goto errout;
208
209         ops = lookup_rules_ops(frh->family);
210         if (ops == NULL) {
211                 err = EAFNOSUPPORT;
212                 goto errout;
213         }
214
215         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
216         if (err < 0)
217                 goto errout;
218
219         err = validate_rulemsg(frh, tb, ops);
220         if (err < 0)
221                 goto errout;
222
223         rule = kzalloc(ops->rule_size, GFP_KERNEL);
224         if (rule == NULL) {
225                 err = -ENOMEM;
226                 goto errout;
227         }
228
229         if (tb[FRA_PRIORITY])
230                 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
231
232         if (tb[FRA_IFNAME]) {
233                 struct net_device *dev;
234
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;
240         }
241
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.
247                          */
248                         rule->mark_mask = 0xFFFFFFFF;
249         }
250
251         if (tb[FRA_FWMASK])
252                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
253
254         rule->action = frh->action;
255         rule->flags = frh->flags;
256         rule->table = frh_get_table(frh, tb);
257
258         if (!rule->pref && ops->default_pref)
259                 rule->pref = ops->default_pref();
260
261         err = -EINVAL;
262         if (tb[FRA_GOTO]) {
263                 if (rule->action != FR_ACT_GOTO)
264                         goto errout_free;
265
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;
270
271                 list_for_each_entry(r, ops->rules_list, list) {
272                         if (r->pref == rule->target) {
273                                 rule->ctarget = r;
274                                 break;
275                         }
276                 }
277
278                 if (rule->ctarget == NULL)
279                         unresolved = 1;
280         } else if (rule->action == FR_ACT_GOTO)
281                 goto errout_free;
282
283         err = ops->configure(rule, skb, nlh, frh, tb);
284         if (err < 0)
285                 goto errout_free;
286
287         list_for_each_entry(r, ops->rules_list, list) {
288                 if (r->pref > rule->pref)
289                         break;
290                 last = r;
291         }
292
293         fib_rule_get(rule);
294
295         if (ops->unresolved_rules) {
296                 /*
297                  * There are unresolved goto rules in the list, check if
298                  * any of them are pointing to this new rule.
299                  */
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;
307                         }
308                 }
309         }
310
311         if (rule->action == FR_ACT_GOTO)
312                 ops->nr_goto_rules++;
313
314         if (unresolved)
315                 ops->unresolved_rules++;
316
317         if (last)
318                 list_add_rcu(&rule->list, &last->list);
319         else
320                 list_add_rcu(&rule->list, ops->rules_list);
321
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;
326
327 errout_free:
328         kfree(rule);
329 errout:
330         rules_ops_put(ops);
331         return err;
332 }
333
334 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
335 {
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;
341
342         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
343                 goto errout;
344
345         ops = lookup_rules_ops(frh->family);
346         if (ops == NULL) {
347                 err = EAFNOSUPPORT;
348                 goto errout;
349         }
350
351         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
352         if (err < 0)
353                 goto errout;
354
355         err = validate_rulemsg(frh, tb, ops);
356         if (err < 0)
357                 goto errout;
358
359         list_for_each_entry(rule, ops->rules_list, list) {
360                 if (frh->action && (frh->action != rule->action))
361                         continue;
362
363                 if (frh->table && (frh_get_table(frh, tb) != rule->table))
364                         continue;
365
366                 if (tb[FRA_PRIORITY] &&
367                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
368                         continue;
369
370                 if (tb[FRA_IFNAME] &&
371                     nla_strcmp(tb[FRA_IFNAME], rule->ifname))
372                         continue;
373
374                 if (tb[FRA_FWMARK] &&
375                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
376                         continue;
377
378                 if (tb[FRA_FWMASK] &&
379                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
380                         continue;
381
382                 if (!ops->compare(rule, frh, tb))
383                         continue;
384
385                 if (rule->flags & FIB_RULE_PERMANENT) {
386                         err = -EPERM;
387                         goto errout;
388                 }
389
390                 list_del_rcu(&rule->list);
391
392                 if (rule->action == FR_ACT_GOTO)
393                         ops->nr_goto_rules--;
394
395                 /*
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.
400                  */
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++;
406                                 }
407                         }
408                 }
409
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;
417         }
418
419         err = -ENOENT;
420 errout:
421         rules_ops_put(ops);
422         return err;
423 }
424
425 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
426                                          struct fib_rule *rule)
427 {
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 */
434
435         if (ops->nlmsg_payload)
436                 payload += ops->nlmsg_payload(rule);
437
438         return payload;
439 }
440
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)
444 {
445         struct nlmsghdr *nlh;
446         struct fib_rule_hdr *frh;
447
448         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
449         if (nlh == NULL)
450                 return -EMSGSIZE;
451
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;
459
460         if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
461                 frh->flags |= FIB_RULE_UNRESOLVED;
462
463         if (rule->ifname[0]) {
464                 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
465
466                 if (rule->ifindex == -1)
467                         frh->flags |= FIB_RULE_DEV_DETACHED;
468         }
469
470         if (rule->pref)
471                 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
472
473         if (rule->mark)
474                 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
475
476         if (rule->mark_mask || rule->mark)
477                 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
478
479         if (rule->target)
480                 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
481
482         if (ops->fill(rule, skb, nlh, frh) < 0)
483                 goto nla_put_failure;
484
485         return nlmsg_end(skb, nlh);
486
487 nla_put_failure:
488         nlmsg_cancel(skb, nlh);
489         return -EMSGSIZE;
490 }
491
492 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
493                       struct fib_rules_ops *ops)
494 {
495         int idx = 0;
496         struct fib_rule *rule;
497
498         rcu_read_lock();
499         list_for_each_entry_rcu(rule, ops->rules_list, list) {
500                 if (idx < cb->args[1])
501                         goto skip;
502
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)
506                         break;
507 skip:
508                 idx++;
509         }
510         rcu_read_unlock();
511         cb->args[1] = idx;
512         rules_ops_put(ops);
513
514         return skb->len;
515 }
516
517 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
518 {
519         struct fib_rules_ops *ops;
520         int idx = 0, family;
521
522         family = rtnl_msg_family(cb->nlh);
523         if (family != AF_UNSPEC) {
524                 /* Protocol specific dump request */
525                 ops = lookup_rules_ops(family);
526                 if (ops == NULL)
527                         return -EAFNOSUPPORT;
528
529                 return dump_rules(skb, cb, ops);
530         }
531
532         rcu_read_lock();
533         list_for_each_entry_rcu(ops, &rules_ops, list) {
534                 if (idx < cb->args[0] || !try_module_get(ops->owner))
535                         goto skip;
536
537                 if (dump_rules(skb, cb, ops) < 0)
538                         break;
539
540                 cb->args[1] = 0;
541         skip:
542                 idx++;
543         }
544         rcu_read_unlock();
545         cb->args[0] = idx;
546
547         return skb->len;
548 }
549
550 static void notify_rule_change(int event, struct fib_rule *rule,
551                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
552                                u32 pid)
553 {
554         struct sk_buff *skb;
555         int err = -ENOBUFS;
556
557         skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
558         if (skb == NULL)
559                 goto errout;
560
561         err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
562         if (err < 0) {
563                 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
564                 WARN_ON(err == -EMSGSIZE);
565                 kfree_skb(skb);
566                 goto errout;
567         }
568         err = rtnl_notify(skb, pid, ops->nlgroup, nlh, GFP_KERNEL);
569 errout:
570         if (err < 0)
571                 rtnl_set_sk_err(ops->nlgroup, err);
572 }
573
574 static void attach_rules(struct list_head *rules, struct net_device *dev)
575 {
576         struct fib_rule *rule;
577
578         list_for_each_entry(rule, rules, list) {
579                 if (rule->ifindex == -1 &&
580                     strcmp(dev->name, rule->ifname) == 0)
581                         rule->ifindex = dev->ifindex;
582         }
583 }
584
585 static void detach_rules(struct list_head *rules, struct net_device *dev)
586 {
587         struct fib_rule *rule;
588
589         list_for_each_entry(rule, rules, list)
590                 if (rule->ifindex == dev->ifindex)
591                         rule->ifindex = -1;
592 }
593
594
595 static int fib_rules_event(struct notifier_block *this, unsigned long event,
596                             void *ptr)
597 {
598         struct net_device *dev = ptr;
599         struct fib_rules_ops *ops;
600
601         ASSERT_RTNL();
602         rcu_read_lock();
603
604         switch (event) {
605         case NETDEV_REGISTER:
606                 list_for_each_entry(ops, &rules_ops, list)
607                         attach_rules(ops->rules_list, dev);
608                 break;
609
610         case NETDEV_UNREGISTER:
611                 list_for_each_entry(ops, &rules_ops, list)
612                         detach_rules(ops->rules_list, dev);
613                 break;
614         }
615
616         rcu_read_unlock();
617
618         return NOTIFY_DONE;
619 }
620
621 static struct notifier_block fib_rules_notifier = {
622         .notifier_call = fib_rules_event,
623 };
624
625 static int __init fib_rules_init(void)
626 {
627         rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
628         rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
629         rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
630
631         return register_netdevice_notifier(&fib_rules_notifier);
632 }
633
634 subsys_initcall(fib_rules_init);