xref: /linux/net/core/fib_rules.c (revision 96c63fa7393d0a346acfe5a91e0c7d4c7782641b)
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 #include <net/ip_tunnels.h>
20 
21 int fib_default_rule_add(struct fib_rules_ops *ops,
22 			 u32 pref, u32 table, u32 flags)
23 {
24 	struct fib_rule *r;
25 
26 	r = kzalloc(ops->rule_size, GFP_KERNEL);
27 	if (r == NULL)
28 		return -ENOMEM;
29 
30 	atomic_set(&r->refcnt, 1);
31 	r->action = FR_ACT_TO_TBL;
32 	r->pref = pref;
33 	r->table = table;
34 	r->flags = flags;
35 	r->fr_net = ops->fro_net;
36 
37 	r->suppress_prefixlen = -1;
38 	r->suppress_ifgroup = -1;
39 
40 	/* The lock is not required here, the list in unreacheable
41 	 * at the moment this function is called */
42 	list_add_tail(&r->list, &ops->rules_list);
43 	return 0;
44 }
45 EXPORT_SYMBOL(fib_default_rule_add);
46 
47 static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
48 {
49 	struct list_head *pos;
50 	struct fib_rule *rule;
51 
52 	if (!list_empty(&ops->rules_list)) {
53 		pos = ops->rules_list.next;
54 		if (pos->next != &ops->rules_list) {
55 			rule = list_entry(pos->next, struct fib_rule, list);
56 			if (rule->pref)
57 				return rule->pref - 1;
58 		}
59 	}
60 
61 	return 0;
62 }
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 	list_add_tail_rcu(&ops->list, &net->rules_ops);
120 	err = 0;
121 errout:
122 	spin_unlock(&net->rules_mod_lock);
123 
124 	return err;
125 }
126 
127 struct fib_rules_ops *
128 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
129 {
130 	struct fib_rules_ops *ops;
131 	int err;
132 
133 	ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
134 	if (ops == NULL)
135 		return ERR_PTR(-ENOMEM);
136 
137 	INIT_LIST_HEAD(&ops->rules_list);
138 	ops->fro_net = net;
139 
140 	err = __fib_rules_register(ops);
141 	if (err) {
142 		kfree(ops);
143 		ops = ERR_PTR(err);
144 	}
145 
146 	return ops;
147 }
148 EXPORT_SYMBOL_GPL(fib_rules_register);
149 
150 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
151 {
152 	struct fib_rule *rule, *tmp;
153 
154 	list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
155 		list_del_rcu(&rule->list);
156 		if (ops->delete)
157 			ops->delete(rule);
158 		fib_rule_put(rule);
159 	}
160 }
161 
162 void fib_rules_unregister(struct fib_rules_ops *ops)
163 {
164 	struct net *net = ops->fro_net;
165 
166 	spin_lock(&net->rules_mod_lock);
167 	list_del_rcu(&ops->list);
168 	spin_unlock(&net->rules_mod_lock);
169 
170 	fib_rules_cleanup_ops(ops);
171 	kfree_rcu(ops, rcu);
172 }
173 EXPORT_SYMBOL_GPL(fib_rules_unregister);
174 
175 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
176 			  struct flowi *fl, int flags,
177 			  struct fib_lookup_arg *arg)
178 {
179 	int ret = 0;
180 
181 	if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
182 		goto out;
183 
184 	if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
185 		goto out;
186 
187 	if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
188 		goto out;
189 
190 	if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
191 		goto out;
192 
193 	if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
194 		goto out;
195 
196 	ret = ops->match(rule, fl, flags);
197 out:
198 	return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
199 }
200 
201 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
202 		     int flags, struct fib_lookup_arg *arg)
203 {
204 	struct fib_rule *rule;
205 	int err;
206 
207 	rcu_read_lock();
208 
209 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
210 jumped:
211 		if (!fib_rule_match(rule, ops, fl, flags, arg))
212 			continue;
213 
214 		if (rule->action == FR_ACT_GOTO) {
215 			struct fib_rule *target;
216 
217 			target = rcu_dereference(rule->ctarget);
218 			if (target == NULL) {
219 				continue;
220 			} else {
221 				rule = target;
222 				goto jumped;
223 			}
224 		} else if (rule->action == FR_ACT_NOP)
225 			continue;
226 		else
227 			err = ops->action(rule, fl, flags, arg);
228 
229 		if (!err && ops->suppress && ops->suppress(rule, arg))
230 			continue;
231 
232 		if (err != -EAGAIN) {
233 			if ((arg->flags & FIB_LOOKUP_NOREF) ||
234 			    likely(atomic_inc_not_zero(&rule->refcnt))) {
235 				arg->rule = rule;
236 				goto out;
237 			}
238 			break;
239 		}
240 	}
241 
242 	err = -ESRCH;
243 out:
244 	rcu_read_unlock();
245 
246 	return err;
247 }
248 EXPORT_SYMBOL_GPL(fib_rules_lookup);
249 
250 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
251 			    struct fib_rules_ops *ops)
252 {
253 	int err = -EINVAL;
254 
255 	if (frh->src_len)
256 		if (tb[FRA_SRC] == NULL ||
257 		    frh->src_len > (ops->addr_size * 8) ||
258 		    nla_len(tb[FRA_SRC]) != ops->addr_size)
259 			goto errout;
260 
261 	if (frh->dst_len)
262 		if (tb[FRA_DST] == NULL ||
263 		    frh->dst_len > (ops->addr_size * 8) ||
264 		    nla_len(tb[FRA_DST]) != ops->addr_size)
265 			goto errout;
266 
267 	err = 0;
268 errout:
269 	return err;
270 }
271 
272 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh)
273 {
274 	struct net *net = sock_net(skb->sk);
275 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
276 	struct fib_rules_ops *ops = NULL;
277 	struct fib_rule *rule, *r, *last = NULL;
278 	struct nlattr *tb[FRA_MAX+1];
279 	int err = -EINVAL, unresolved = 0;
280 
281 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
282 		goto errout;
283 
284 	ops = lookup_rules_ops(net, frh->family);
285 	if (ops == NULL) {
286 		err = -EAFNOSUPPORT;
287 		goto errout;
288 	}
289 
290 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
291 	if (err < 0)
292 		goto errout;
293 
294 	err = validate_rulemsg(frh, tb, ops);
295 	if (err < 0)
296 		goto errout;
297 
298 	rule = kzalloc(ops->rule_size, GFP_KERNEL);
299 	if (rule == NULL) {
300 		err = -ENOMEM;
301 		goto errout;
302 	}
303 	rule->fr_net = net;
304 
305 	rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
306 	                              : fib_default_rule_pref(ops);
307 
308 	if (tb[FRA_IIFNAME]) {
309 		struct net_device *dev;
310 
311 		rule->iifindex = -1;
312 		nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
313 		dev = __dev_get_by_name(net, rule->iifname);
314 		if (dev)
315 			rule->iifindex = dev->ifindex;
316 	}
317 
318 	if (tb[FRA_OIFNAME]) {
319 		struct net_device *dev;
320 
321 		rule->oifindex = -1;
322 		nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
323 		dev = __dev_get_by_name(net, rule->oifname);
324 		if (dev)
325 			rule->oifindex = dev->ifindex;
326 	}
327 
328 	if (tb[FRA_FWMARK]) {
329 		rule->mark = nla_get_u32(tb[FRA_FWMARK]);
330 		if (rule->mark)
331 			/* compatibility: if the mark value is non-zero all bits
332 			 * are compared unless a mask is explicitly specified.
333 			 */
334 			rule->mark_mask = 0xFFFFFFFF;
335 	}
336 
337 	if (tb[FRA_FWMASK])
338 		rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
339 
340 	if (tb[FRA_TUN_ID])
341 		rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
342 
343 	if (tb[FRA_L3MDEV]) {
344 #ifdef CONFIG_NET_L3_MASTER_DEV
345 		rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
346 		if (rule->l3mdev != 1)
347 #endif
348 			goto errout_free;
349 	}
350 
351 	rule->action = frh->action;
352 	rule->flags = frh->flags;
353 	rule->table = frh_get_table(frh, tb);
354 	if (tb[FRA_SUPPRESS_PREFIXLEN])
355 		rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
356 	else
357 		rule->suppress_prefixlen = -1;
358 
359 	if (tb[FRA_SUPPRESS_IFGROUP])
360 		rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
361 	else
362 		rule->suppress_ifgroup = -1;
363 
364 	err = -EINVAL;
365 	if (tb[FRA_GOTO]) {
366 		if (rule->action != FR_ACT_GOTO)
367 			goto errout_free;
368 
369 		rule->target = nla_get_u32(tb[FRA_GOTO]);
370 		/* Backward jumps are prohibited to avoid endless loops */
371 		if (rule->target <= rule->pref)
372 			goto errout_free;
373 
374 		list_for_each_entry(r, &ops->rules_list, list) {
375 			if (r->pref == rule->target) {
376 				RCU_INIT_POINTER(rule->ctarget, r);
377 				break;
378 			}
379 		}
380 
381 		if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
382 			unresolved = 1;
383 	} else if (rule->action == FR_ACT_GOTO)
384 		goto errout_free;
385 
386 	if (rule->l3mdev && rule->table)
387 		goto errout_free;
388 
389 	err = ops->configure(rule, skb, frh, tb);
390 	if (err < 0)
391 		goto errout_free;
392 
393 	list_for_each_entry(r, &ops->rules_list, list) {
394 		if (r->pref > rule->pref)
395 			break;
396 		last = r;
397 	}
398 
399 	fib_rule_get(rule);
400 
401 	if (last)
402 		list_add_rcu(&rule->list, &last->list);
403 	else
404 		list_add_rcu(&rule->list, &ops->rules_list);
405 
406 	if (ops->unresolved_rules) {
407 		/*
408 		 * There are unresolved goto rules in the list, check if
409 		 * any of them are pointing to this new rule.
410 		 */
411 		list_for_each_entry(r, &ops->rules_list, list) {
412 			if (r->action == FR_ACT_GOTO &&
413 			    r->target == rule->pref &&
414 			    rtnl_dereference(r->ctarget) == NULL) {
415 				rcu_assign_pointer(r->ctarget, rule);
416 				if (--ops->unresolved_rules == 0)
417 					break;
418 			}
419 		}
420 	}
421 
422 	if (rule->action == FR_ACT_GOTO)
423 		ops->nr_goto_rules++;
424 
425 	if (unresolved)
426 		ops->unresolved_rules++;
427 
428 	if (rule->tun_id)
429 		ip_tunnel_need_metadata();
430 
431 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
432 	flush_route_cache(ops);
433 	rules_ops_put(ops);
434 	return 0;
435 
436 errout_free:
437 	kfree(rule);
438 errout:
439 	rules_ops_put(ops);
440 	return err;
441 }
442 EXPORT_SYMBOL_GPL(fib_nl_newrule);
443 
444 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh)
445 {
446 	struct net *net = sock_net(skb->sk);
447 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
448 	struct fib_rules_ops *ops = NULL;
449 	struct fib_rule *rule, *tmp;
450 	struct nlattr *tb[FRA_MAX+1];
451 	int err = -EINVAL;
452 
453 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
454 		goto errout;
455 
456 	ops = lookup_rules_ops(net, frh->family);
457 	if (ops == NULL) {
458 		err = -EAFNOSUPPORT;
459 		goto errout;
460 	}
461 
462 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
463 	if (err < 0)
464 		goto errout;
465 
466 	err = validate_rulemsg(frh, tb, ops);
467 	if (err < 0)
468 		goto errout;
469 
470 	list_for_each_entry(rule, &ops->rules_list, list) {
471 		if (frh->action && (frh->action != rule->action))
472 			continue;
473 
474 		if (frh_get_table(frh, tb) &&
475 		    (frh_get_table(frh, tb) != rule->table))
476 			continue;
477 
478 		if (tb[FRA_PRIORITY] &&
479 		    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
480 			continue;
481 
482 		if (tb[FRA_IIFNAME] &&
483 		    nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
484 			continue;
485 
486 		if (tb[FRA_OIFNAME] &&
487 		    nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
488 			continue;
489 
490 		if (tb[FRA_FWMARK] &&
491 		    (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
492 			continue;
493 
494 		if (tb[FRA_FWMASK] &&
495 		    (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
496 			continue;
497 
498 		if (tb[FRA_TUN_ID] &&
499 		    (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
500 			continue;
501 
502 		if (tb[FRA_L3MDEV] &&
503 		    (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
504 			continue;
505 
506 		if (!ops->compare(rule, frh, tb))
507 			continue;
508 
509 		if (rule->flags & FIB_RULE_PERMANENT) {
510 			err = -EPERM;
511 			goto errout;
512 		}
513 
514 		if (ops->delete) {
515 			err = ops->delete(rule);
516 			if (err)
517 				goto errout;
518 		}
519 
520 		if (rule->tun_id)
521 			ip_tunnel_unneed_metadata();
522 
523 		list_del_rcu(&rule->list);
524 
525 		if (rule->action == FR_ACT_GOTO) {
526 			ops->nr_goto_rules--;
527 			if (rtnl_dereference(rule->ctarget) == NULL)
528 				ops->unresolved_rules--;
529 		}
530 
531 		/*
532 		 * Check if this rule is a target to any of them. If so,
533 		 * disable them. As this operation is eventually very
534 		 * expensive, it is only performed if goto rules have
535 		 * actually been added.
536 		 */
537 		if (ops->nr_goto_rules > 0) {
538 			list_for_each_entry(tmp, &ops->rules_list, list) {
539 				if (rtnl_dereference(tmp->ctarget) == rule) {
540 					RCU_INIT_POINTER(tmp->ctarget, NULL);
541 					ops->unresolved_rules++;
542 				}
543 			}
544 		}
545 
546 		notify_rule_change(RTM_DELRULE, rule, ops, nlh,
547 				   NETLINK_CB(skb).portid);
548 		fib_rule_put(rule);
549 		flush_route_cache(ops);
550 		rules_ops_put(ops);
551 		return 0;
552 	}
553 
554 	err = -ENOENT;
555 errout:
556 	rules_ops_put(ops);
557 	return err;
558 }
559 EXPORT_SYMBOL_GPL(fib_nl_delrule);
560 
561 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
562 					 struct fib_rule *rule)
563 {
564 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
565 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
566 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
567 			 + nla_total_size(4) /* FRA_PRIORITY */
568 			 + nla_total_size(4) /* FRA_TABLE */
569 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
570 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
571 			 + nla_total_size(4) /* FRA_FWMARK */
572 			 + nla_total_size(4) /* FRA_FWMASK */
573 			 + nla_total_size_64bit(8); /* FRA_TUN_ID */
574 
575 	if (ops->nlmsg_payload)
576 		payload += ops->nlmsg_payload(rule);
577 
578 	return payload;
579 }
580 
581 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
582 			    u32 pid, u32 seq, int type, int flags,
583 			    struct fib_rules_ops *ops)
584 {
585 	struct nlmsghdr *nlh;
586 	struct fib_rule_hdr *frh;
587 
588 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
589 	if (nlh == NULL)
590 		return -EMSGSIZE;
591 
592 	frh = nlmsg_data(nlh);
593 	frh->family = ops->family;
594 	frh->table = rule->table;
595 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
596 		goto nla_put_failure;
597 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
598 		goto nla_put_failure;
599 	frh->res1 = 0;
600 	frh->res2 = 0;
601 	frh->action = rule->action;
602 	frh->flags = rule->flags;
603 
604 	if (rule->action == FR_ACT_GOTO &&
605 	    rcu_access_pointer(rule->ctarget) == NULL)
606 		frh->flags |= FIB_RULE_UNRESOLVED;
607 
608 	if (rule->iifname[0]) {
609 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
610 			goto nla_put_failure;
611 		if (rule->iifindex == -1)
612 			frh->flags |= FIB_RULE_IIF_DETACHED;
613 	}
614 
615 	if (rule->oifname[0]) {
616 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
617 			goto nla_put_failure;
618 		if (rule->oifindex == -1)
619 			frh->flags |= FIB_RULE_OIF_DETACHED;
620 	}
621 
622 	if ((rule->pref &&
623 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
624 	    (rule->mark &&
625 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
626 	    ((rule->mark_mask || rule->mark) &&
627 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
628 	    (rule->target &&
629 	     nla_put_u32(skb, FRA_GOTO, rule->target)) ||
630 	    (rule->tun_id &&
631 	     nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
632 	    (rule->l3mdev &&
633 	     nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)))
634 		goto nla_put_failure;
635 
636 	if (rule->suppress_ifgroup != -1) {
637 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
638 			goto nla_put_failure;
639 	}
640 
641 	if (ops->fill(rule, skb, frh) < 0)
642 		goto nla_put_failure;
643 
644 	nlmsg_end(skb, nlh);
645 	return 0;
646 
647 nla_put_failure:
648 	nlmsg_cancel(skb, nlh);
649 	return -EMSGSIZE;
650 }
651 
652 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
653 		      struct fib_rules_ops *ops)
654 {
655 	int idx = 0;
656 	struct fib_rule *rule;
657 	int err = 0;
658 
659 	rcu_read_lock();
660 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
661 		if (idx < cb->args[1])
662 			goto skip;
663 
664 		err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
665 				       cb->nlh->nlmsg_seq, RTM_NEWRULE,
666 				       NLM_F_MULTI, ops);
667 		if (err)
668 			break;
669 skip:
670 		idx++;
671 	}
672 	rcu_read_unlock();
673 	cb->args[1] = idx;
674 	rules_ops_put(ops);
675 
676 	return err;
677 }
678 
679 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
680 {
681 	struct net *net = sock_net(skb->sk);
682 	struct fib_rules_ops *ops;
683 	int idx = 0, family;
684 
685 	family = rtnl_msg_family(cb->nlh);
686 	if (family != AF_UNSPEC) {
687 		/* Protocol specific dump request */
688 		ops = lookup_rules_ops(net, family);
689 		if (ops == NULL)
690 			return -EAFNOSUPPORT;
691 
692 		dump_rules(skb, cb, ops);
693 
694 		return skb->len;
695 	}
696 
697 	rcu_read_lock();
698 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
699 		if (idx < cb->args[0] || !try_module_get(ops->owner))
700 			goto skip;
701 
702 		if (dump_rules(skb, cb, ops) < 0)
703 			break;
704 
705 		cb->args[1] = 0;
706 skip:
707 		idx++;
708 	}
709 	rcu_read_unlock();
710 	cb->args[0] = idx;
711 
712 	return skb->len;
713 }
714 
715 static void notify_rule_change(int event, struct fib_rule *rule,
716 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
717 			       u32 pid)
718 {
719 	struct net *net;
720 	struct sk_buff *skb;
721 	int err = -ENOBUFS;
722 
723 	net = ops->fro_net;
724 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
725 	if (skb == NULL)
726 		goto errout;
727 
728 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
729 	if (err < 0) {
730 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
731 		WARN_ON(err == -EMSGSIZE);
732 		kfree_skb(skb);
733 		goto errout;
734 	}
735 
736 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
737 	return;
738 errout:
739 	if (err < 0)
740 		rtnl_set_sk_err(net, ops->nlgroup, err);
741 }
742 
743 static void attach_rules(struct list_head *rules, struct net_device *dev)
744 {
745 	struct fib_rule *rule;
746 
747 	list_for_each_entry(rule, rules, list) {
748 		if (rule->iifindex == -1 &&
749 		    strcmp(dev->name, rule->iifname) == 0)
750 			rule->iifindex = dev->ifindex;
751 		if (rule->oifindex == -1 &&
752 		    strcmp(dev->name, rule->oifname) == 0)
753 			rule->oifindex = dev->ifindex;
754 	}
755 }
756 
757 static void detach_rules(struct list_head *rules, struct net_device *dev)
758 {
759 	struct fib_rule *rule;
760 
761 	list_for_each_entry(rule, rules, list) {
762 		if (rule->iifindex == dev->ifindex)
763 			rule->iifindex = -1;
764 		if (rule->oifindex == dev->ifindex)
765 			rule->oifindex = -1;
766 	}
767 }
768 
769 
770 static int fib_rules_event(struct notifier_block *this, unsigned long event,
771 			   void *ptr)
772 {
773 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
774 	struct net *net = dev_net(dev);
775 	struct fib_rules_ops *ops;
776 
777 	ASSERT_RTNL();
778 
779 	switch (event) {
780 	case NETDEV_REGISTER:
781 		list_for_each_entry(ops, &net->rules_ops, list)
782 			attach_rules(&ops->rules_list, dev);
783 		break;
784 
785 	case NETDEV_CHANGENAME:
786 		list_for_each_entry(ops, &net->rules_ops, list) {
787 			detach_rules(&ops->rules_list, dev);
788 			attach_rules(&ops->rules_list, dev);
789 		}
790 		break;
791 
792 	case NETDEV_UNREGISTER:
793 		list_for_each_entry(ops, &net->rules_ops, list)
794 			detach_rules(&ops->rules_list, dev);
795 		break;
796 	}
797 
798 	return NOTIFY_DONE;
799 }
800 
801 static struct notifier_block fib_rules_notifier = {
802 	.notifier_call = fib_rules_event,
803 };
804 
805 static int __net_init fib_rules_net_init(struct net *net)
806 {
807 	INIT_LIST_HEAD(&net->rules_ops);
808 	spin_lock_init(&net->rules_mod_lock);
809 	return 0;
810 }
811 
812 static struct pernet_operations fib_rules_net_ops = {
813 	.init = fib_rules_net_init,
814 };
815 
816 static int __init fib_rules_init(void)
817 {
818 	int err;
819 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
820 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
821 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
822 
823 	err = register_pernet_subsys(&fib_rules_net_ops);
824 	if (err < 0)
825 		goto fail;
826 
827 	err = register_netdevice_notifier(&fib_rules_notifier);
828 	if (err < 0)
829 		goto fail_unregister;
830 
831 	return 0;
832 
833 fail_unregister:
834 	unregister_pernet_subsys(&fib_rules_net_ops);
835 fail:
836 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
837 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
838 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
839 	return err;
840 }
841 
842 subsys_initcall(fib_rules_init);
843