xref: /openbmc/linux/net/core/fib_rules.c (revision 12eb4683)
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/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>
19 
20 int fib_default_rule_add(struct fib_rules_ops *ops,
21 			 u32 pref, u32 table, u32 flags)
22 {
23 	struct fib_rule *r;
24 
25 	r = kzalloc(ops->rule_size, GFP_KERNEL);
26 	if (r == NULL)
27 		return -ENOMEM;
28 
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);
35 
36 	r->suppress_prefixlen = -1;
37 	r->suppress_ifgroup = -1;
38 
39 	/* The lock is not required here, the list in unreacheable
40 	 * at the moment this function is called */
41 	list_add_tail(&r->list, &ops->rules_list);
42 	return 0;
43 }
44 EXPORT_SYMBOL(fib_default_rule_add);
45 
46 u32 fib_default_rule_pref(struct fib_rules_ops *ops)
47 {
48 	struct list_head *pos;
49 	struct fib_rule *rule;
50 
51 	if (!list_empty(&ops->rules_list)) {
52 		pos = ops->rules_list.next;
53 		if (pos->next != &ops->rules_list) {
54 			rule = list_entry(pos->next, struct fib_rule, list);
55 			if (rule->pref)
56 				return rule->pref - 1;
57 		}
58 	}
59 
60 	return 0;
61 }
62 EXPORT_SYMBOL(fib_default_rule_pref);
63 
64 static void notify_rule_change(int event, struct fib_rule *rule,
65 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
66 			       u32 pid);
67 
68 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
69 {
70 	struct fib_rules_ops *ops;
71 
72 	rcu_read_lock();
73 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
74 		if (ops->family == family) {
75 			if (!try_module_get(ops->owner))
76 				ops = NULL;
77 			rcu_read_unlock();
78 			return ops;
79 		}
80 	}
81 	rcu_read_unlock();
82 
83 	return NULL;
84 }
85 
86 static void rules_ops_put(struct fib_rules_ops *ops)
87 {
88 	if (ops)
89 		module_put(ops->owner);
90 }
91 
92 static void flush_route_cache(struct fib_rules_ops *ops)
93 {
94 	if (ops->flush_cache)
95 		ops->flush_cache(ops);
96 }
97 
98 static int __fib_rules_register(struct fib_rules_ops *ops)
99 {
100 	int err = -EEXIST;
101 	struct fib_rules_ops *o;
102 	struct net *net;
103 
104 	net = ops->fro_net;
105 
106 	if (ops->rule_size < sizeof(struct fib_rule))
107 		return -EINVAL;
108 
109 	if (ops->match == NULL || ops->configure == NULL ||
110 	    ops->compare == NULL || ops->fill == NULL ||
111 	    ops->action == NULL)
112 		return -EINVAL;
113 
114 	spin_lock(&net->rules_mod_lock);
115 	list_for_each_entry(o, &net->rules_ops, list)
116 		if (ops->family == o->family)
117 			goto errout;
118 
119 	hold_net(net);
120 	list_add_tail_rcu(&ops->list, &net->rules_ops);
121 	err = 0;
122 errout:
123 	spin_unlock(&net->rules_mod_lock);
124 
125 	return err;
126 }
127 
128 struct fib_rules_ops *
129 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
130 {
131 	struct fib_rules_ops *ops;
132 	int err;
133 
134 	ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
135 	if (ops == NULL)
136 		return ERR_PTR(-ENOMEM);
137 
138 	INIT_LIST_HEAD(&ops->rules_list);
139 	ops->fro_net = net;
140 
141 	err = __fib_rules_register(ops);
142 	if (err) {
143 		kfree(ops);
144 		ops = ERR_PTR(err);
145 	}
146 
147 	return ops;
148 }
149 EXPORT_SYMBOL_GPL(fib_rules_register);
150 
151 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
152 {
153 	struct fib_rule *rule, *tmp;
154 
155 	list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
156 		list_del_rcu(&rule->list);
157 		if (ops->delete)
158 			ops->delete(rule);
159 		fib_rule_put(rule);
160 	}
161 }
162 
163 static void fib_rules_put_rcu(struct rcu_head *head)
164 {
165 	struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu);
166 	struct net *net = ops->fro_net;
167 
168 	release_net(net);
169 	kfree(ops);
170 }
171 
172 void fib_rules_unregister(struct fib_rules_ops *ops)
173 {
174 	struct net *net = ops->fro_net;
175 
176 	spin_lock(&net->rules_mod_lock);
177 	list_del_rcu(&ops->list);
178 	fib_rules_cleanup_ops(ops);
179 	spin_unlock(&net->rules_mod_lock);
180 
181 	call_rcu(&ops->rcu, fib_rules_put_rcu);
182 }
183 EXPORT_SYMBOL_GPL(fib_rules_unregister);
184 
185 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
186 			  struct flowi *fl, int flags)
187 {
188 	int ret = 0;
189 
190 	if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
191 		goto out;
192 
193 	if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
194 		goto out;
195 
196 	if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
197 		goto out;
198 
199 	ret = ops->match(rule, fl, flags);
200 out:
201 	return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
202 }
203 
204 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
205 		     int flags, struct fib_lookup_arg *arg)
206 {
207 	struct fib_rule *rule;
208 	int err;
209 
210 	rcu_read_lock();
211 
212 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
213 jumped:
214 		if (!fib_rule_match(rule, ops, fl, flags))
215 			continue;
216 
217 		if (rule->action == FR_ACT_GOTO) {
218 			struct fib_rule *target;
219 
220 			target = rcu_dereference(rule->ctarget);
221 			if (target == NULL) {
222 				continue;
223 			} else {
224 				rule = target;
225 				goto jumped;
226 			}
227 		} else if (rule->action == FR_ACT_NOP)
228 			continue;
229 		else
230 			err = ops->action(rule, fl, flags, arg);
231 
232 		if (!err && ops->suppress && ops->suppress(rule, arg))
233 			continue;
234 
235 		if (err != -EAGAIN) {
236 			if ((arg->flags & FIB_LOOKUP_NOREF) ||
237 			    likely(atomic_inc_not_zero(&rule->refcnt))) {
238 				arg->rule = rule;
239 				goto out;
240 			}
241 			break;
242 		}
243 	}
244 
245 	err = -ESRCH;
246 out:
247 	rcu_read_unlock();
248 
249 	return err;
250 }
251 EXPORT_SYMBOL_GPL(fib_rules_lookup);
252 
253 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
254 			    struct fib_rules_ops *ops)
255 {
256 	int err = -EINVAL;
257 
258 	if (frh->src_len)
259 		if (tb[FRA_SRC] == NULL ||
260 		    frh->src_len > (ops->addr_size * 8) ||
261 		    nla_len(tb[FRA_SRC]) != ops->addr_size)
262 			goto errout;
263 
264 	if (frh->dst_len)
265 		if (tb[FRA_DST] == NULL ||
266 		    frh->dst_len > (ops->addr_size * 8) ||
267 		    nla_len(tb[FRA_DST]) != ops->addr_size)
268 			goto errout;
269 
270 	err = 0;
271 errout:
272 	return err;
273 }
274 
275 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh)
276 {
277 	struct net *net = sock_net(skb->sk);
278 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
279 	struct fib_rules_ops *ops = NULL;
280 	struct fib_rule *rule, *r, *last = NULL;
281 	struct nlattr *tb[FRA_MAX+1];
282 	int err = -EINVAL, unresolved = 0;
283 
284 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
285 		goto errout;
286 
287 	ops = lookup_rules_ops(net, frh->family);
288 	if (ops == NULL) {
289 		err = -EAFNOSUPPORT;
290 		goto errout;
291 	}
292 
293 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
294 	if (err < 0)
295 		goto errout;
296 
297 	err = validate_rulemsg(frh, tb, ops);
298 	if (err < 0)
299 		goto errout;
300 
301 	rule = kzalloc(ops->rule_size, GFP_KERNEL);
302 	if (rule == NULL) {
303 		err = -ENOMEM;
304 		goto errout;
305 	}
306 	rule->fr_net = hold_net(net);
307 
308 	if (tb[FRA_PRIORITY])
309 		rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
310 
311 	if (tb[FRA_IIFNAME]) {
312 		struct net_device *dev;
313 
314 		rule->iifindex = -1;
315 		nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
316 		dev = __dev_get_by_name(net, rule->iifname);
317 		if (dev)
318 			rule->iifindex = dev->ifindex;
319 	}
320 
321 	if (tb[FRA_OIFNAME]) {
322 		struct net_device *dev;
323 
324 		rule->oifindex = -1;
325 		nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
326 		dev = __dev_get_by_name(net, rule->oifname);
327 		if (dev)
328 			rule->oifindex = dev->ifindex;
329 	}
330 
331 	if (tb[FRA_FWMARK]) {
332 		rule->mark = nla_get_u32(tb[FRA_FWMARK]);
333 		if (rule->mark)
334 			/* compatibility: if the mark value is non-zero all bits
335 			 * are compared unless a mask is explicitly specified.
336 			 */
337 			rule->mark_mask = 0xFFFFFFFF;
338 	}
339 
340 	if (tb[FRA_FWMASK])
341 		rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
342 
343 	rule->action = frh->action;
344 	rule->flags = frh->flags;
345 	rule->table = frh_get_table(frh, tb);
346 	if (tb[FRA_SUPPRESS_PREFIXLEN])
347 		rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
348 	else
349 		rule->suppress_prefixlen = -1;
350 
351 	if (tb[FRA_SUPPRESS_IFGROUP])
352 		rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
353 	else
354 		rule->suppress_ifgroup = -1;
355 
356 	if (!tb[FRA_PRIORITY] && ops->default_pref)
357 		rule->pref = ops->default_pref(ops);
358 
359 	err = -EINVAL;
360 	if (tb[FRA_GOTO]) {
361 		if (rule->action != FR_ACT_GOTO)
362 			goto errout_free;
363 
364 		rule->target = nla_get_u32(tb[FRA_GOTO]);
365 		/* Backward jumps are prohibited to avoid endless loops */
366 		if (rule->target <= rule->pref)
367 			goto errout_free;
368 
369 		list_for_each_entry(r, &ops->rules_list, list) {
370 			if (r->pref == rule->target) {
371 				RCU_INIT_POINTER(rule->ctarget, r);
372 				break;
373 			}
374 		}
375 
376 		if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
377 			unresolved = 1;
378 	} else if (rule->action == FR_ACT_GOTO)
379 		goto errout_free;
380 
381 	err = ops->configure(rule, skb, frh, tb);
382 	if (err < 0)
383 		goto errout_free;
384 
385 	list_for_each_entry(r, &ops->rules_list, list) {
386 		if (r->pref > rule->pref)
387 			break;
388 		last = r;
389 	}
390 
391 	fib_rule_get(rule);
392 
393 	if (last)
394 		list_add_rcu(&rule->list, &last->list);
395 	else
396 		list_add_rcu(&rule->list, &ops->rules_list);
397 
398 	if (ops->unresolved_rules) {
399 		/*
400 		 * There are unresolved goto rules in the list, check if
401 		 * any of them are pointing to this new rule.
402 		 */
403 		list_for_each_entry(r, &ops->rules_list, list) {
404 			if (r->action == FR_ACT_GOTO &&
405 			    r->target == rule->pref &&
406 			    rtnl_dereference(r->ctarget) == NULL) {
407 				rcu_assign_pointer(r->ctarget, rule);
408 				if (--ops->unresolved_rules == 0)
409 					break;
410 			}
411 		}
412 	}
413 
414 	if (rule->action == FR_ACT_GOTO)
415 		ops->nr_goto_rules++;
416 
417 	if (unresolved)
418 		ops->unresolved_rules++;
419 
420 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
421 	flush_route_cache(ops);
422 	rules_ops_put(ops);
423 	return 0;
424 
425 errout_free:
426 	release_net(rule->fr_net);
427 	kfree(rule);
428 errout:
429 	rules_ops_put(ops);
430 	return err;
431 }
432 
433 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh)
434 {
435 	struct net *net = sock_net(skb->sk);
436 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
437 	struct fib_rules_ops *ops = NULL;
438 	struct fib_rule *rule, *tmp;
439 	struct nlattr *tb[FRA_MAX+1];
440 	int err = -EINVAL;
441 
442 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
443 		goto errout;
444 
445 	ops = lookup_rules_ops(net, frh->family);
446 	if (ops == NULL) {
447 		err = -EAFNOSUPPORT;
448 		goto errout;
449 	}
450 
451 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
452 	if (err < 0)
453 		goto errout;
454 
455 	err = validate_rulemsg(frh, tb, ops);
456 	if (err < 0)
457 		goto errout;
458 
459 	list_for_each_entry(rule, &ops->rules_list, list) {
460 		if (frh->action && (frh->action != rule->action))
461 			continue;
462 
463 		if (frh_get_table(frh, tb) &&
464 		    (frh_get_table(frh, tb) != rule->table))
465 			continue;
466 
467 		if (tb[FRA_PRIORITY] &&
468 		    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
469 			continue;
470 
471 		if (tb[FRA_IIFNAME] &&
472 		    nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
473 			continue;
474 
475 		if (tb[FRA_OIFNAME] &&
476 		    nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
477 			continue;
478 
479 		if (tb[FRA_FWMARK] &&
480 		    (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
481 			continue;
482 
483 		if (tb[FRA_FWMASK] &&
484 		    (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
485 			continue;
486 
487 		if (!ops->compare(rule, frh, tb))
488 			continue;
489 
490 		if (rule->flags & FIB_RULE_PERMANENT) {
491 			err = -EPERM;
492 			goto errout;
493 		}
494 
495 		list_del_rcu(&rule->list);
496 
497 		if (rule->action == FR_ACT_GOTO) {
498 			ops->nr_goto_rules--;
499 			if (rtnl_dereference(rule->ctarget) == NULL)
500 				ops->unresolved_rules--;
501 		}
502 
503 		/*
504 		 * Check if this rule is a target to any of them. If so,
505 		 * disable them. As this operation is eventually very
506 		 * expensive, it is only performed if goto rules have
507 		 * actually been added.
508 		 */
509 		if (ops->nr_goto_rules > 0) {
510 			list_for_each_entry(tmp, &ops->rules_list, list) {
511 				if (rtnl_dereference(tmp->ctarget) == rule) {
512 					RCU_INIT_POINTER(tmp->ctarget, NULL);
513 					ops->unresolved_rules++;
514 				}
515 			}
516 		}
517 
518 		notify_rule_change(RTM_DELRULE, rule, ops, nlh,
519 				   NETLINK_CB(skb).portid);
520 		if (ops->delete)
521 			ops->delete(rule);
522 		fib_rule_put(rule);
523 		flush_route_cache(ops);
524 		rules_ops_put(ops);
525 		return 0;
526 	}
527 
528 	err = -ENOENT;
529 errout:
530 	rules_ops_put(ops);
531 	return err;
532 }
533 
534 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
535 					 struct fib_rule *rule)
536 {
537 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
538 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
539 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
540 			 + nla_total_size(4) /* FRA_PRIORITY */
541 			 + nla_total_size(4) /* FRA_TABLE */
542 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
543 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
544 			 + nla_total_size(4) /* FRA_FWMARK */
545 			 + nla_total_size(4); /* FRA_FWMASK */
546 
547 	if (ops->nlmsg_payload)
548 		payload += ops->nlmsg_payload(rule);
549 
550 	return payload;
551 }
552 
553 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
554 			    u32 pid, u32 seq, int type, int flags,
555 			    struct fib_rules_ops *ops)
556 {
557 	struct nlmsghdr *nlh;
558 	struct fib_rule_hdr *frh;
559 
560 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
561 	if (nlh == NULL)
562 		return -EMSGSIZE;
563 
564 	frh = nlmsg_data(nlh);
565 	frh->family = ops->family;
566 	frh->table = rule->table;
567 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
568 		goto nla_put_failure;
569 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
570 		goto nla_put_failure;
571 	frh->res1 = 0;
572 	frh->res2 = 0;
573 	frh->action = rule->action;
574 	frh->flags = rule->flags;
575 
576 	if (rule->action == FR_ACT_GOTO &&
577 	    rcu_access_pointer(rule->ctarget) == NULL)
578 		frh->flags |= FIB_RULE_UNRESOLVED;
579 
580 	if (rule->iifname[0]) {
581 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
582 			goto nla_put_failure;
583 		if (rule->iifindex == -1)
584 			frh->flags |= FIB_RULE_IIF_DETACHED;
585 	}
586 
587 	if (rule->oifname[0]) {
588 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
589 			goto nla_put_failure;
590 		if (rule->oifindex == -1)
591 			frh->flags |= FIB_RULE_OIF_DETACHED;
592 	}
593 
594 	if ((rule->pref &&
595 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
596 	    (rule->mark &&
597 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
598 	    ((rule->mark_mask || rule->mark) &&
599 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
600 	    (rule->target &&
601 	     nla_put_u32(skb, FRA_GOTO, rule->target)))
602 		goto nla_put_failure;
603 
604 	if (rule->suppress_ifgroup != -1) {
605 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
606 			goto nla_put_failure;
607 	}
608 
609 	if (ops->fill(rule, skb, frh) < 0)
610 		goto nla_put_failure;
611 
612 	return nlmsg_end(skb, nlh);
613 
614 nla_put_failure:
615 	nlmsg_cancel(skb, nlh);
616 	return -EMSGSIZE;
617 }
618 
619 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
620 		      struct fib_rules_ops *ops)
621 {
622 	int idx = 0;
623 	struct fib_rule *rule;
624 
625 	rcu_read_lock();
626 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
627 		if (idx < cb->args[1])
628 			goto skip;
629 
630 		if (fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
631 				     cb->nlh->nlmsg_seq, RTM_NEWRULE,
632 				     NLM_F_MULTI, ops) < 0)
633 			break;
634 skip:
635 		idx++;
636 	}
637 	rcu_read_unlock();
638 	cb->args[1] = idx;
639 	rules_ops_put(ops);
640 
641 	return skb->len;
642 }
643 
644 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
645 {
646 	struct net *net = sock_net(skb->sk);
647 	struct fib_rules_ops *ops;
648 	int idx = 0, family;
649 
650 	family = rtnl_msg_family(cb->nlh);
651 	if (family != AF_UNSPEC) {
652 		/* Protocol specific dump request */
653 		ops = lookup_rules_ops(net, family);
654 		if (ops == NULL)
655 			return -EAFNOSUPPORT;
656 
657 		return dump_rules(skb, cb, ops);
658 	}
659 
660 	rcu_read_lock();
661 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
662 		if (idx < cb->args[0] || !try_module_get(ops->owner))
663 			goto skip;
664 
665 		if (dump_rules(skb, cb, ops) < 0)
666 			break;
667 
668 		cb->args[1] = 0;
669 skip:
670 		idx++;
671 	}
672 	rcu_read_unlock();
673 	cb->args[0] = idx;
674 
675 	return skb->len;
676 }
677 
678 static void notify_rule_change(int event, struct fib_rule *rule,
679 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
680 			       u32 pid)
681 {
682 	struct net *net;
683 	struct sk_buff *skb;
684 	int err = -ENOBUFS;
685 
686 	net = ops->fro_net;
687 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
688 	if (skb == NULL)
689 		goto errout;
690 
691 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
692 	if (err < 0) {
693 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
694 		WARN_ON(err == -EMSGSIZE);
695 		kfree_skb(skb);
696 		goto errout;
697 	}
698 
699 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
700 	return;
701 errout:
702 	if (err < 0)
703 		rtnl_set_sk_err(net, ops->nlgroup, err);
704 }
705 
706 static void attach_rules(struct list_head *rules, struct net_device *dev)
707 {
708 	struct fib_rule *rule;
709 
710 	list_for_each_entry(rule, rules, list) {
711 		if (rule->iifindex == -1 &&
712 		    strcmp(dev->name, rule->iifname) == 0)
713 			rule->iifindex = dev->ifindex;
714 		if (rule->oifindex == -1 &&
715 		    strcmp(dev->name, rule->oifname) == 0)
716 			rule->oifindex = dev->ifindex;
717 	}
718 }
719 
720 static void detach_rules(struct list_head *rules, struct net_device *dev)
721 {
722 	struct fib_rule *rule;
723 
724 	list_for_each_entry(rule, rules, list) {
725 		if (rule->iifindex == dev->ifindex)
726 			rule->iifindex = -1;
727 		if (rule->oifindex == dev->ifindex)
728 			rule->oifindex = -1;
729 	}
730 }
731 
732 
733 static int fib_rules_event(struct notifier_block *this, unsigned long event,
734 			   void *ptr)
735 {
736 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
737 	struct net *net = dev_net(dev);
738 	struct fib_rules_ops *ops;
739 
740 	ASSERT_RTNL();
741 
742 	switch (event) {
743 	case NETDEV_REGISTER:
744 		list_for_each_entry(ops, &net->rules_ops, list)
745 			attach_rules(&ops->rules_list, dev);
746 		break;
747 
748 	case NETDEV_UNREGISTER:
749 		list_for_each_entry(ops, &net->rules_ops, list)
750 			detach_rules(&ops->rules_list, dev);
751 		break;
752 	}
753 
754 	return NOTIFY_DONE;
755 }
756 
757 static struct notifier_block fib_rules_notifier = {
758 	.notifier_call = fib_rules_event,
759 };
760 
761 static int __net_init fib_rules_net_init(struct net *net)
762 {
763 	INIT_LIST_HEAD(&net->rules_ops);
764 	spin_lock_init(&net->rules_mod_lock);
765 	return 0;
766 }
767 
768 static struct pernet_operations fib_rules_net_ops = {
769 	.init = fib_rules_net_init,
770 };
771 
772 static int __init fib_rules_init(void)
773 {
774 	int err;
775 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
776 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
777 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
778 
779 	err = register_pernet_subsys(&fib_rules_net_ops);
780 	if (err < 0)
781 		goto fail;
782 
783 	err = register_netdevice_notifier(&fib_rules_notifier);
784 	if (err < 0)
785 		goto fail_unregister;
786 
787 	return 0;
788 
789 fail_unregister:
790 	unregister_pernet_subsys(&fib_rules_net_ops);
791 fail:
792 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
793 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
794 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
795 	return err;
796 }
797 
798 subsys_initcall(fib_rules_init);
799