Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / 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/net_namespace.h>
15 #include <net/sock.h>
16 #include <net/fib_rules.h>
17
18 int fib_default_rule_add(struct fib_rules_ops *ops,
19                          u32 pref, u32 table, u32 flags)
20 {
21         struct fib_rule *r;
22
23         r = kzalloc(ops->rule_size, GFP_KERNEL);
24         if (r == NULL)
25                 return -ENOMEM;
26
27         atomic_set(&r->refcnt, 1);
28         r->action = FR_ACT_TO_TBL;
29         r->pref = pref;
30         r->table = table;
31         r->flags = flags;
32         r->fr_net = hold_net(ops->fro_net);
33
34         /* The lock is not required here, the list in unreacheable
35          * at the moment this function is called */
36         list_add_tail(&r->list, &ops->rules_list);
37         return 0;
38 }
39 EXPORT_SYMBOL(fib_default_rule_add);
40
41 static void notify_rule_change(int event, struct fib_rule *rule,
42                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
43                                u32 pid);
44
45 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
46 {
47         struct fib_rules_ops *ops;
48
49         rcu_read_lock();
50         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
51                 if (ops->family == family) {
52                         if (!try_module_get(ops->owner))
53                                 ops = NULL;
54                         rcu_read_unlock();
55                         return ops;
56                 }
57         }
58         rcu_read_unlock();
59
60         return NULL;
61 }
62
63 static void rules_ops_put(struct fib_rules_ops *ops)
64 {
65         if (ops)
66                 module_put(ops->owner);
67 }
68
69 static void flush_route_cache(struct fib_rules_ops *ops)
70 {
71         if (ops->flush_cache)
72                 ops->flush_cache(ops);
73 }
74
75 int fib_rules_register(struct fib_rules_ops *ops)
76 {
77         int err = -EEXIST;
78         struct fib_rules_ops *o;
79         struct net *net;
80
81         net = ops->fro_net;
82
83         if (ops->rule_size < sizeof(struct fib_rule))
84                 return -EINVAL;
85
86         if (ops->match == NULL || ops->configure == NULL ||
87             ops->compare == NULL || ops->fill == NULL ||
88             ops->action == NULL)
89                 return -EINVAL;
90
91         spin_lock(&net->rules_mod_lock);
92         list_for_each_entry(o, &net->rules_ops, list)
93                 if (ops->family == o->family)
94                         goto errout;
95
96         hold_net(net);
97         list_add_tail_rcu(&ops->list, &net->rules_ops);
98         err = 0;
99 errout:
100         spin_unlock(&net->rules_mod_lock);
101
102         return err;
103 }
104
105 EXPORT_SYMBOL_GPL(fib_rules_register);
106
107 void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
108 {
109         struct fib_rule *rule, *tmp;
110
111         list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
112                 list_del_rcu(&rule->list);
113                 fib_rule_put(rule);
114         }
115 }
116 EXPORT_SYMBOL_GPL(fib_rules_cleanup_ops);
117
118 void fib_rules_unregister(struct fib_rules_ops *ops)
119 {
120         struct net *net = ops->fro_net;
121
122         spin_lock(&net->rules_mod_lock);
123         list_del_rcu(&ops->list);
124         fib_rules_cleanup_ops(ops);
125         spin_unlock(&net->rules_mod_lock);
126
127         synchronize_rcu();
128         release_net(net);
129 }
130
131 EXPORT_SYMBOL_GPL(fib_rules_unregister);
132
133 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
134                           struct flowi *fl, int flags)
135 {
136         int ret = 0;
137
138         if (rule->ifindex && (rule->ifindex != fl->iif))
139                 goto out;
140
141         if ((rule->mark ^ fl->mark) & rule->mark_mask)
142                 goto out;
143
144         ret = ops->match(rule, fl, flags);
145 out:
146         return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
147 }
148
149 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
150                      int flags, struct fib_lookup_arg *arg)
151 {
152         struct fib_rule *rule;
153         int err;
154
155         rcu_read_lock();
156
157         list_for_each_entry_rcu(rule, &ops->rules_list, list) {
158 jumped:
159                 if (!fib_rule_match(rule, ops, fl, flags))
160                         continue;
161
162                 if (rule->action == FR_ACT_GOTO) {
163                         struct fib_rule *target;
164
165                         target = rcu_dereference(rule->ctarget);
166                         if (target == NULL) {
167                                 continue;
168                         } else {
169                                 rule = target;
170                                 goto jumped;
171                         }
172                 } else if (rule->action == FR_ACT_NOP)
173                         continue;
174                 else
175                         err = ops->action(rule, fl, flags, arg);
176
177                 if (err != -EAGAIN) {
178                         fib_rule_get(rule);
179                         arg->rule = rule;
180                         goto out;
181                 }
182         }
183
184         err = -ESRCH;
185 out:
186         rcu_read_unlock();
187
188         return err;
189 }
190
191 EXPORT_SYMBOL_GPL(fib_rules_lookup);
192
193 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
194                             struct fib_rules_ops *ops)
195 {
196         int err = -EINVAL;
197
198         if (frh->src_len)
199                 if (tb[FRA_SRC] == NULL ||
200                     frh->src_len > (ops->addr_size * 8) ||
201                     nla_len(tb[FRA_SRC]) != ops->addr_size)
202                         goto errout;
203
204         if (frh->dst_len)
205                 if (tb[FRA_DST] == NULL ||
206                     frh->dst_len > (ops->addr_size * 8) ||
207                     nla_len(tb[FRA_DST]) != ops->addr_size)
208                         goto errout;
209
210         err = 0;
211 errout:
212         return err;
213 }
214
215 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
216 {
217         struct net *net = sock_net(skb->sk);
218         struct fib_rule_hdr *frh = nlmsg_data(nlh);
219         struct fib_rules_ops *ops = NULL;
220         struct fib_rule *rule, *r, *last = NULL;
221         struct nlattr *tb[FRA_MAX+1];
222         int err = -EINVAL, unresolved = 0;
223
224         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
225                 goto errout;
226
227         ops = lookup_rules_ops(net, frh->family);
228         if (ops == NULL) {
229                 err = -EAFNOSUPPORT;
230                 goto errout;
231         }
232
233         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
234         if (err < 0)
235                 goto errout;
236
237         err = validate_rulemsg(frh, tb, ops);
238         if (err < 0)
239                 goto errout;
240
241         rule = kzalloc(ops->rule_size, GFP_KERNEL);
242         if (rule == NULL) {
243                 err = -ENOMEM;
244                 goto errout;
245         }
246         rule->fr_net = hold_net(net);
247
248         if (tb[FRA_PRIORITY])
249                 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
250
251         if (tb[FRA_IFNAME]) {
252                 struct net_device *dev;
253
254                 rule->ifindex = -1;
255                 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
256                 dev = __dev_get_by_name(net, rule->ifname);
257                 if (dev)
258                         rule->ifindex = dev->ifindex;
259         }
260
261         if (tb[FRA_FWMARK]) {
262                 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
263                 if (rule->mark)
264                         /* compatibility: if the mark value is non-zero all bits
265                          * are compared unless a mask is explicitly specified.
266                          */
267                         rule->mark_mask = 0xFFFFFFFF;
268         }
269
270         if (tb[FRA_FWMASK])
271                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
272
273         rule->action = frh->action;
274         rule->flags = frh->flags;
275         rule->table = frh_get_table(frh, tb);
276
277         if (!rule->pref && ops->default_pref)
278                 rule->pref = ops->default_pref(ops);
279
280         err = -EINVAL;
281         if (tb[FRA_GOTO]) {
282                 if (rule->action != FR_ACT_GOTO)
283                         goto errout_free;
284
285                 rule->target = nla_get_u32(tb[FRA_GOTO]);
286                 /* Backward jumps are prohibited to avoid endless loops */
287                 if (rule->target <= rule->pref)
288                         goto errout_free;
289
290                 list_for_each_entry(r, &ops->rules_list, list) {
291                         if (r->pref == rule->target) {
292                                 rule->ctarget = r;
293                                 break;
294                         }
295                 }
296
297                 if (rule->ctarget == NULL)
298                         unresolved = 1;
299         } else if (rule->action == FR_ACT_GOTO)
300                 goto errout_free;
301
302         err = ops->configure(rule, skb, nlh, frh, tb);
303         if (err < 0)
304                 goto errout_free;
305
306         list_for_each_entry(r, &ops->rules_list, list) {
307                 if (r->pref > rule->pref)
308                         break;
309                 last = r;
310         }
311
312         fib_rule_get(rule);
313
314         if (ops->unresolved_rules) {
315                 /*
316                  * There are unresolved goto rules in the list, check if
317                  * any of them are pointing to this new rule.
318                  */
319                 list_for_each_entry(r, &ops->rules_list, list) {
320                         if (r->action == FR_ACT_GOTO &&
321                             r->target == rule->pref) {
322                                 BUG_ON(r->ctarget != NULL);
323                                 rcu_assign_pointer(r->ctarget, rule);
324                                 if (--ops->unresolved_rules == 0)
325                                         break;
326                         }
327                 }
328         }
329
330         if (rule->action == FR_ACT_GOTO)
331                 ops->nr_goto_rules++;
332
333         if (unresolved)
334                 ops->unresolved_rules++;
335
336         if (last)
337                 list_add_rcu(&rule->list, &last->list);
338         else
339                 list_add_rcu(&rule->list, &ops->rules_list);
340
341         notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
342         flush_route_cache(ops);
343         rules_ops_put(ops);
344         return 0;
345
346 errout_free:
347         release_net(rule->fr_net);
348         kfree(rule);
349 errout:
350         rules_ops_put(ops);
351         return err;
352 }
353
354 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
355 {
356         struct net *net = sock_net(skb->sk);
357         struct fib_rule_hdr *frh = nlmsg_data(nlh);
358         struct fib_rules_ops *ops = NULL;
359         struct fib_rule *rule, *tmp;
360         struct nlattr *tb[FRA_MAX+1];
361         int err = -EINVAL;
362
363         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
364                 goto errout;
365
366         ops = lookup_rules_ops(net, frh->family);
367         if (ops == NULL) {
368                 err = -EAFNOSUPPORT;
369                 goto errout;
370         }
371
372         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
373         if (err < 0)
374                 goto errout;
375
376         err = validate_rulemsg(frh, tb, ops);
377         if (err < 0)
378                 goto errout;
379
380         list_for_each_entry(rule, &ops->rules_list, list) {
381                 if (frh->action && (frh->action != rule->action))
382                         continue;
383
384                 if (frh->table && (frh_get_table(frh, tb) != rule->table))
385                         continue;
386
387                 if (tb[FRA_PRIORITY] &&
388                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
389                         continue;
390
391                 if (tb[FRA_IFNAME] &&
392                     nla_strcmp(tb[FRA_IFNAME], rule->ifname))
393                         continue;
394
395                 if (tb[FRA_FWMARK] &&
396                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
397                         continue;
398
399                 if (tb[FRA_FWMASK] &&
400                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
401                         continue;
402
403                 if (!ops->compare(rule, frh, tb))
404                         continue;
405
406                 if (rule->flags & FIB_RULE_PERMANENT) {
407                         err = -EPERM;
408                         goto errout;
409                 }
410
411                 list_del_rcu(&rule->list);
412
413                 if (rule->action == FR_ACT_GOTO)
414                         ops->nr_goto_rules--;
415
416                 /*
417                  * Check if this rule is a target to any of them. If so,
418                  * disable them. As this operation is eventually very
419                  * expensive, it is only performed if goto rules have
420                  * actually been added.
421                  */
422                 if (ops->nr_goto_rules > 0) {
423                         list_for_each_entry(tmp, &ops->rules_list, list) {
424                                 if (tmp->ctarget == rule) {
425                                         rcu_assign_pointer(tmp->ctarget, NULL);
426                                         ops->unresolved_rules++;
427                                 }
428                         }
429                 }
430
431                 synchronize_rcu();
432                 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
433                                    NETLINK_CB(skb).pid);
434                 fib_rule_put(rule);
435                 flush_route_cache(ops);
436                 rules_ops_put(ops);
437                 return 0;
438         }
439
440         err = -ENOENT;
441 errout:
442         rules_ops_put(ops);
443         return err;
444 }
445
446 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
447                                          struct fib_rule *rule)
448 {
449         size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
450                          + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
451                          + nla_total_size(4) /* FRA_PRIORITY */
452                          + nla_total_size(4) /* FRA_TABLE */
453                          + nla_total_size(4) /* FRA_FWMARK */
454                          + nla_total_size(4); /* FRA_FWMASK */
455
456         if (ops->nlmsg_payload)
457                 payload += ops->nlmsg_payload(rule);
458
459         return payload;
460 }
461
462 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
463                             u32 pid, u32 seq, int type, int flags,
464                             struct fib_rules_ops *ops)
465 {
466         struct nlmsghdr *nlh;
467         struct fib_rule_hdr *frh;
468
469         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
470         if (nlh == NULL)
471                 return -EMSGSIZE;
472
473         frh = nlmsg_data(nlh);
474         frh->table = rule->table;
475         NLA_PUT_U32(skb, FRA_TABLE, rule->table);
476         frh->res1 = 0;
477         frh->res2 = 0;
478         frh->action = rule->action;
479         frh->flags = rule->flags;
480
481         if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
482                 frh->flags |= FIB_RULE_UNRESOLVED;
483
484         if (rule->ifname[0]) {
485                 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
486
487                 if (rule->ifindex == -1)
488                         frh->flags |= FIB_RULE_DEV_DETACHED;
489         }
490
491         if (rule->pref)
492                 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
493
494         if (rule->mark)
495                 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
496
497         if (rule->mark_mask || rule->mark)
498                 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
499
500         if (rule->target)
501                 NLA_PUT_U32(skb, FRA_GOTO, rule->target);
502
503         if (ops->fill(rule, skb, nlh, frh) < 0)
504                 goto nla_put_failure;
505
506         return nlmsg_end(skb, nlh);
507
508 nla_put_failure:
509         nlmsg_cancel(skb, nlh);
510         return -EMSGSIZE;
511 }
512
513 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
514                       struct fib_rules_ops *ops)
515 {
516         int idx = 0;
517         struct fib_rule *rule;
518
519         list_for_each_entry(rule, &ops->rules_list, list) {
520                 if (idx < cb->args[1])
521                         goto skip;
522
523                 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
524                                      cb->nlh->nlmsg_seq, RTM_NEWRULE,
525                                      NLM_F_MULTI, ops) < 0)
526                         break;
527 skip:
528                 idx++;
529         }
530         cb->args[1] = idx;
531         rules_ops_put(ops);
532
533         return skb->len;
534 }
535
536 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
537 {
538         struct net *net = sock_net(skb->sk);
539         struct fib_rules_ops *ops;
540         int idx = 0, family;
541
542         family = rtnl_msg_family(cb->nlh);
543         if (family != AF_UNSPEC) {
544                 /* Protocol specific dump request */
545                 ops = lookup_rules_ops(net, family);
546                 if (ops == NULL)
547                         return -EAFNOSUPPORT;
548
549                 return dump_rules(skb, cb, ops);
550         }
551
552         rcu_read_lock();
553         list_for_each_entry_rcu(ops, &net->rules_ops, list) {
554                 if (idx < cb->args[0] || !try_module_get(ops->owner))
555                         goto skip;
556
557                 if (dump_rules(skb, cb, ops) < 0)
558                         break;
559
560                 cb->args[1] = 0;
561         skip:
562                 idx++;
563         }
564         rcu_read_unlock();
565         cb->args[0] = idx;
566
567         return skb->len;
568 }
569
570 static void notify_rule_change(int event, struct fib_rule *rule,
571                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
572                                u32 pid)
573 {
574         struct net *net;
575         struct sk_buff *skb;
576         int err = -ENOBUFS;
577
578         net = ops->fro_net;
579         skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
580         if (skb == NULL)
581                 goto errout;
582
583         err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
584         if (err < 0) {
585                 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
586                 WARN_ON(err == -EMSGSIZE);
587                 kfree_skb(skb);
588                 goto errout;
589         }
590
591         err = rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
592 errout:
593         if (err < 0)
594                 rtnl_set_sk_err(net, ops->nlgroup, err);
595 }
596
597 static void attach_rules(struct list_head *rules, struct net_device *dev)
598 {
599         struct fib_rule *rule;
600
601         list_for_each_entry(rule, rules, list) {
602                 if (rule->ifindex == -1 &&
603                     strcmp(dev->name, rule->ifname) == 0)
604                         rule->ifindex = dev->ifindex;
605         }
606 }
607
608 static void detach_rules(struct list_head *rules, struct net_device *dev)
609 {
610         struct fib_rule *rule;
611
612         list_for_each_entry(rule, rules, list)
613                 if (rule->ifindex == dev->ifindex)
614                         rule->ifindex = -1;
615 }
616
617
618 static int fib_rules_event(struct notifier_block *this, unsigned long event,
619                             void *ptr)
620 {
621         struct net_device *dev = ptr;
622         struct net *net = dev_net(dev);
623         struct fib_rules_ops *ops;
624
625         ASSERT_RTNL();
626         rcu_read_lock();
627
628         switch (event) {
629         case NETDEV_REGISTER:
630                 list_for_each_entry(ops, &net->rules_ops, list)
631                         attach_rules(&ops->rules_list, dev);
632                 break;
633
634         case NETDEV_UNREGISTER:
635                 list_for_each_entry(ops, &net->rules_ops, list)
636                         detach_rules(&ops->rules_list, dev);
637                 break;
638         }
639
640         rcu_read_unlock();
641
642         return NOTIFY_DONE;
643 }
644
645 static struct notifier_block fib_rules_notifier = {
646         .notifier_call = fib_rules_event,
647 };
648
649 static int fib_rules_net_init(struct net *net)
650 {
651         INIT_LIST_HEAD(&net->rules_ops);
652         spin_lock_init(&net->rules_mod_lock);
653         return 0;
654 }
655
656 static struct pernet_operations fib_rules_net_ops = {
657         .init = fib_rules_net_init,
658 };
659
660 static int __init fib_rules_init(void)
661 {
662         int err;
663         rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
664         rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
665         rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
666
667         err = register_netdevice_notifier(&fib_rules_notifier);
668         if (err < 0)
669                 goto fail;
670
671         err = register_pernet_subsys(&fib_rules_net_ops);
672         if (err < 0)
673                 goto fail_unregister;
674         return 0;
675
676 fail_unregister:
677         unregister_netdevice_notifier(&fib_rules_notifier);
678 fail:
679         rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
680         rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
681         rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
682         return err;
683 }
684
685 subsys_initcall(fib_rules_init);