[NET] rules: Protocol independant mark selector
[safe/jmp/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/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 int fib_rules_register(struct fib_rules_ops *ops)
48 {
49         int err = -EEXIST;
50         struct fib_rules_ops *o;
51
52         if (ops->rule_size < sizeof(struct fib_rule))
53                 return -EINVAL;
54
55         if (ops->match == NULL || ops->configure == NULL ||
56             ops->compare == NULL || ops->fill == NULL ||
57             ops->action == NULL)
58                 return -EINVAL;
59
60         spin_lock(&rules_mod_lock);
61         list_for_each_entry(o, &rules_ops, list)
62                 if (ops->family == o->family)
63                         goto errout;
64
65         list_add_tail_rcu(&ops->list, &rules_ops);
66         err = 0;
67 errout:
68         spin_unlock(&rules_mod_lock);
69
70         return err;
71 }
72
73 EXPORT_SYMBOL_GPL(fib_rules_register);
74
75 static void cleanup_ops(struct fib_rules_ops *ops)
76 {
77         struct fib_rule *rule, *tmp;
78
79         list_for_each_entry_safe(rule, tmp, ops->rules_list, list) {
80                 list_del_rcu(&rule->list);
81                 fib_rule_put(rule);
82         }
83 }
84
85 int fib_rules_unregister(struct fib_rules_ops *ops)
86 {
87         int err = 0;
88         struct fib_rules_ops *o;
89
90         spin_lock(&rules_mod_lock);
91         list_for_each_entry(o, &rules_ops, list) {
92                 if (o == ops) {
93                         list_del_rcu(&o->list);
94                         cleanup_ops(ops);
95                         goto out;
96                 }
97         }
98
99         err = -ENOENT;
100 out:
101         spin_unlock(&rules_mod_lock);
102
103         synchronize_rcu();
104
105         return err;
106 }
107
108 EXPORT_SYMBOL_GPL(fib_rules_unregister);
109
110 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
111                      int flags, struct fib_lookup_arg *arg)
112 {
113         struct fib_rule *rule;
114         int err;
115
116         rcu_read_lock();
117
118         list_for_each_entry_rcu(rule, ops->rules_list, list) {
119                 if (rule->ifindex && (rule->ifindex != fl->iif))
120                         continue;
121
122                 if ((rule->mark ^ fl->mark) & rule->mark_mask)
123                         continue;
124
125                 if (!ops->match(rule, fl, flags))
126                         continue;
127
128                 err = ops->action(rule, fl, flags, arg);
129                 if (err != -EAGAIN) {
130                         fib_rule_get(rule);
131                         arg->rule = rule;
132                         goto out;
133                 }
134         }
135
136         err = -ENETUNREACH;
137 out:
138         rcu_read_unlock();
139
140         return err;
141 }
142
143 EXPORT_SYMBOL_GPL(fib_rules_lookup);
144
145 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
146 {
147         struct fib_rule_hdr *frh = nlmsg_data(nlh);
148         struct fib_rules_ops *ops = NULL;
149         struct fib_rule *rule, *r, *last = NULL;
150         struct nlattr *tb[FRA_MAX+1];
151         int err = -EINVAL;
152
153         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
154                 goto errout;
155
156         ops = lookup_rules_ops(frh->family);
157         if (ops == NULL) {
158                 err = EAFNOSUPPORT;
159                 goto errout;
160         }
161
162         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
163         if (err < 0)
164                 goto errout;
165
166         rule = kzalloc(ops->rule_size, GFP_KERNEL);
167         if (rule == NULL) {
168                 err = -ENOMEM;
169                 goto errout;
170         }
171
172         if (tb[FRA_PRIORITY])
173                 rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
174
175         if (tb[FRA_IFNAME]) {
176                 struct net_device *dev;
177
178                 rule->ifindex = -1;
179                 nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
180                 dev = __dev_get_by_name(rule->ifname);
181                 if (dev)
182                         rule->ifindex = dev->ifindex;
183         }
184
185         if (tb[FRA_FWMARK]) {
186                 rule->mark = nla_get_u32(tb[FRA_FWMARK]);
187                 if (rule->mark)
188                         /* compatibility: if the mark value is non-zero all bits
189                          * are compared unless a mask is explicitly specified.
190                          */
191                         rule->mark_mask = 0xFFFFFFFF;
192         }
193
194         if (tb[FRA_FWMASK])
195                 rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
196
197         rule->action = frh->action;
198         rule->flags = frh->flags;
199         rule->table = frh_get_table(frh, tb);
200
201         if (!rule->pref && ops->default_pref)
202                 rule->pref = ops->default_pref();
203
204         err = ops->configure(rule, skb, nlh, frh, tb);
205         if (err < 0)
206                 goto errout_free;
207
208         list_for_each_entry(r, ops->rules_list, list) {
209                 if (r->pref > rule->pref)
210                         break;
211                 last = r;
212         }
213
214         fib_rule_get(rule);
215
216         if (last)
217                 list_add_rcu(&rule->list, &last->list);
218         else
219                 list_add_rcu(&rule->list, ops->rules_list);
220
221         notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
222         rules_ops_put(ops);
223         return 0;
224
225 errout_free:
226         kfree(rule);
227 errout:
228         rules_ops_put(ops);
229         return err;
230 }
231
232 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
233 {
234         struct fib_rule_hdr *frh = nlmsg_data(nlh);
235         struct fib_rules_ops *ops = NULL;
236         struct fib_rule *rule;
237         struct nlattr *tb[FRA_MAX+1];
238         int err = -EINVAL;
239
240         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
241                 goto errout;
242
243         ops = lookup_rules_ops(frh->family);
244         if (ops == NULL) {
245                 err = EAFNOSUPPORT;
246                 goto errout;
247         }
248
249         err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
250         if (err < 0)
251                 goto errout;
252
253         list_for_each_entry(rule, ops->rules_list, list) {
254                 if (frh->action && (frh->action != rule->action))
255                         continue;
256
257                 if (frh->table && (frh_get_table(frh, tb) != rule->table))
258                         continue;
259
260                 if (tb[FRA_PRIORITY] &&
261                     (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
262                         continue;
263
264                 if (tb[FRA_IFNAME] &&
265                     nla_strcmp(tb[FRA_IFNAME], rule->ifname))
266                         continue;
267
268                 if (tb[FRA_FWMARK] &&
269                     (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
270                         continue;
271
272                 if (tb[FRA_FWMASK] &&
273                     (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
274                         continue;
275
276                 if (!ops->compare(rule, frh, tb))
277                         continue;
278
279                 if (rule->flags & FIB_RULE_PERMANENT) {
280                         err = -EPERM;
281                         goto errout;
282                 }
283
284                 list_del_rcu(&rule->list);
285                 synchronize_rcu();
286                 notify_rule_change(RTM_DELRULE, rule, ops, nlh,
287                                    NETLINK_CB(skb).pid);
288                 fib_rule_put(rule);
289                 rules_ops_put(ops);
290                 return 0;
291         }
292
293         err = -ENOENT;
294 errout:
295         rules_ops_put(ops);
296         return err;
297 }
298
299 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
300                             u32 pid, u32 seq, int type, int flags,
301                             struct fib_rules_ops *ops)
302 {
303         struct nlmsghdr *nlh;
304         struct fib_rule_hdr *frh;
305
306         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
307         if (nlh == NULL)
308                 return -1;
309
310         frh = nlmsg_data(nlh);
311         frh->table = rule->table;
312         NLA_PUT_U32(skb, FRA_TABLE, rule->table);
313         frh->res1 = 0;
314         frh->res2 = 0;
315         frh->action = rule->action;
316         frh->flags = rule->flags;
317
318         if (rule->ifname[0])
319                 NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
320
321         if (rule->pref)
322                 NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
323
324         if (rule->mark)
325                 NLA_PUT_U32(skb, FRA_FWMARK, rule->mark);
326
327         if (rule->mark_mask || rule->mark)
328                 NLA_PUT_U32(skb, FRA_FWMASK, rule->mark_mask);
329
330         if (ops->fill(rule, skb, nlh, frh) < 0)
331                 goto nla_put_failure;
332
333         return nlmsg_end(skb, nlh);
334
335 nla_put_failure:
336         return nlmsg_cancel(skb, nlh);
337 }
338
339 int fib_rules_dump(struct sk_buff *skb, struct netlink_callback *cb, int family)
340 {
341         int idx = 0;
342         struct fib_rule *rule;
343         struct fib_rules_ops *ops;
344
345         ops = lookup_rules_ops(family);
346         if (ops == NULL)
347                 return -EAFNOSUPPORT;
348
349         rcu_read_lock();
350         list_for_each_entry(rule, ops->rules_list, list) {
351                 if (idx < cb->args[0])
352                         goto skip;
353
354                 if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).pid,
355                                      cb->nlh->nlmsg_seq, RTM_NEWRULE,
356                                      NLM_F_MULTI, ops) < 0)
357                         break;
358 skip:
359                 idx++;
360         }
361         rcu_read_unlock();
362         cb->args[0] = idx;
363         rules_ops_put(ops);
364
365         return skb->len;
366 }
367
368 EXPORT_SYMBOL_GPL(fib_rules_dump);
369
370 static void notify_rule_change(int event, struct fib_rule *rule,
371                                struct fib_rules_ops *ops, struct nlmsghdr *nlh,
372                                u32 pid)
373 {
374         struct sk_buff *skb;
375         int err = -ENOBUFS;
376
377         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
378         if (skb == NULL)
379                 goto errout;
380
381         err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
382         if (err < 0) {
383                 kfree_skb(skb);
384                 goto errout;
385         }
386
387         err = rtnl_notify(skb, pid, ops->nlgroup, nlh, GFP_KERNEL);
388 errout:
389         if (err < 0)
390                 rtnl_set_sk_err(ops->nlgroup, err);
391 }
392
393 static void attach_rules(struct list_head *rules, struct net_device *dev)
394 {
395         struct fib_rule *rule;
396
397         list_for_each_entry(rule, rules, list) {
398                 if (rule->ifindex == -1 &&
399                     strcmp(dev->name, rule->ifname) == 0)
400                         rule->ifindex = dev->ifindex;
401         }
402 }
403
404 static void detach_rules(struct list_head *rules, struct net_device *dev)
405 {
406         struct fib_rule *rule;
407
408         list_for_each_entry(rule, rules, list)
409                 if (rule->ifindex == dev->ifindex)
410                         rule->ifindex = -1;
411 }
412
413
414 static int fib_rules_event(struct notifier_block *this, unsigned long event,
415                             void *ptr)
416 {
417         struct net_device *dev = ptr;
418         struct fib_rules_ops *ops;
419
420         ASSERT_RTNL();
421         rcu_read_lock();
422
423         switch (event) {
424         case NETDEV_REGISTER:
425                 list_for_each_entry(ops, &rules_ops, list)
426                         attach_rules(ops->rules_list, dev);
427                 break;
428
429         case NETDEV_UNREGISTER:
430                 list_for_each_entry(ops, &rules_ops, list)
431                         detach_rules(ops->rules_list, dev);
432                 break;
433         }
434
435         rcu_read_unlock();
436
437         return NOTIFY_DONE;
438 }
439
440 static struct notifier_block fib_rules_notifier = {
441         .notifier_call = fib_rules_event,
442 };
443
444 static int __init fib_rules_init(void)
445 {
446         return register_netdevice_notifier(&fib_rules_notifier);
447 }
448
449 subsys_initcall(fib_rules_init);