xref: /openbmc/linux/net/core/fib_rules.c (revision bc5aa3a0)
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 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
273 		       struct nlattr **tb, struct fib_rule *rule)
274 {
275 	struct fib_rule *r;
276 
277 	list_for_each_entry(r, &ops->rules_list, list) {
278 		if (r->action != rule->action)
279 			continue;
280 
281 		if (r->table != rule->table)
282 			continue;
283 
284 		if (r->pref != rule->pref)
285 			continue;
286 
287 		if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
288 			continue;
289 
290 		if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
291 			continue;
292 
293 		if (r->mark != rule->mark)
294 			continue;
295 
296 		if (r->mark_mask != rule->mark_mask)
297 			continue;
298 
299 		if (r->tun_id != rule->tun_id)
300 			continue;
301 
302 		if (r->fr_net != rule->fr_net)
303 			continue;
304 
305 		if (r->l3mdev != rule->l3mdev)
306 			continue;
307 
308 		if (!ops->compare(r, frh, tb))
309 			continue;
310 		return 1;
311 	}
312 	return 0;
313 }
314 
315 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh)
316 {
317 	struct net *net = sock_net(skb->sk);
318 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
319 	struct fib_rules_ops *ops = NULL;
320 	struct fib_rule *rule, *r, *last = NULL;
321 	struct nlattr *tb[FRA_MAX+1];
322 	int err = -EINVAL, unresolved = 0;
323 
324 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
325 		goto errout;
326 
327 	ops = lookup_rules_ops(net, frh->family);
328 	if (ops == NULL) {
329 		err = -EAFNOSUPPORT;
330 		goto errout;
331 	}
332 
333 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
334 	if (err < 0)
335 		goto errout;
336 
337 	err = validate_rulemsg(frh, tb, ops);
338 	if (err < 0)
339 		goto errout;
340 
341 	rule = kzalloc(ops->rule_size, GFP_KERNEL);
342 	if (rule == NULL) {
343 		err = -ENOMEM;
344 		goto errout;
345 	}
346 	rule->fr_net = net;
347 
348 	rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
349 	                              : fib_default_rule_pref(ops);
350 
351 	if (tb[FRA_IIFNAME]) {
352 		struct net_device *dev;
353 
354 		rule->iifindex = -1;
355 		nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
356 		dev = __dev_get_by_name(net, rule->iifname);
357 		if (dev)
358 			rule->iifindex = dev->ifindex;
359 	}
360 
361 	if (tb[FRA_OIFNAME]) {
362 		struct net_device *dev;
363 
364 		rule->oifindex = -1;
365 		nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
366 		dev = __dev_get_by_name(net, rule->oifname);
367 		if (dev)
368 			rule->oifindex = dev->ifindex;
369 	}
370 
371 	if (tb[FRA_FWMARK]) {
372 		rule->mark = nla_get_u32(tb[FRA_FWMARK]);
373 		if (rule->mark)
374 			/* compatibility: if the mark value is non-zero all bits
375 			 * are compared unless a mask is explicitly specified.
376 			 */
377 			rule->mark_mask = 0xFFFFFFFF;
378 	}
379 
380 	if (tb[FRA_FWMASK])
381 		rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
382 
383 	if (tb[FRA_TUN_ID])
384 		rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
385 
386 	if (tb[FRA_L3MDEV]) {
387 #ifdef CONFIG_NET_L3_MASTER_DEV
388 		rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
389 		if (rule->l3mdev != 1)
390 #endif
391 			goto errout_free;
392 	}
393 
394 	rule->action = frh->action;
395 	rule->flags = frh->flags;
396 	rule->table = frh_get_table(frh, tb);
397 	if (tb[FRA_SUPPRESS_PREFIXLEN])
398 		rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
399 	else
400 		rule->suppress_prefixlen = -1;
401 
402 	if (tb[FRA_SUPPRESS_IFGROUP])
403 		rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
404 	else
405 		rule->suppress_ifgroup = -1;
406 
407 	err = -EINVAL;
408 	if (tb[FRA_GOTO]) {
409 		if (rule->action != FR_ACT_GOTO)
410 			goto errout_free;
411 
412 		rule->target = nla_get_u32(tb[FRA_GOTO]);
413 		/* Backward jumps are prohibited to avoid endless loops */
414 		if (rule->target <= rule->pref)
415 			goto errout_free;
416 
417 		list_for_each_entry(r, &ops->rules_list, list) {
418 			if (r->pref == rule->target) {
419 				RCU_INIT_POINTER(rule->ctarget, r);
420 				break;
421 			}
422 		}
423 
424 		if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
425 			unresolved = 1;
426 	} else if (rule->action == FR_ACT_GOTO)
427 		goto errout_free;
428 
429 	if (rule->l3mdev && rule->table)
430 		goto errout_free;
431 
432 	if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
433 	    rule_exists(ops, frh, tb, rule)) {
434 		err = -EEXIST;
435 		goto errout_free;
436 	}
437 
438 	err = ops->configure(rule, skb, frh, tb);
439 	if (err < 0)
440 		goto errout_free;
441 
442 	list_for_each_entry(r, &ops->rules_list, list) {
443 		if (r->pref > rule->pref)
444 			break;
445 		last = r;
446 	}
447 
448 	fib_rule_get(rule);
449 
450 	if (last)
451 		list_add_rcu(&rule->list, &last->list);
452 	else
453 		list_add_rcu(&rule->list, &ops->rules_list);
454 
455 	if (ops->unresolved_rules) {
456 		/*
457 		 * There are unresolved goto rules in the list, check if
458 		 * any of them are pointing to this new rule.
459 		 */
460 		list_for_each_entry(r, &ops->rules_list, list) {
461 			if (r->action == FR_ACT_GOTO &&
462 			    r->target == rule->pref &&
463 			    rtnl_dereference(r->ctarget) == NULL) {
464 				rcu_assign_pointer(r->ctarget, rule);
465 				if (--ops->unresolved_rules == 0)
466 					break;
467 			}
468 		}
469 	}
470 
471 	if (rule->action == FR_ACT_GOTO)
472 		ops->nr_goto_rules++;
473 
474 	if (unresolved)
475 		ops->unresolved_rules++;
476 
477 	if (rule->tun_id)
478 		ip_tunnel_need_metadata();
479 
480 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
481 	flush_route_cache(ops);
482 	rules_ops_put(ops);
483 	return 0;
484 
485 errout_free:
486 	kfree(rule);
487 errout:
488 	rules_ops_put(ops);
489 	return err;
490 }
491 EXPORT_SYMBOL_GPL(fib_nl_newrule);
492 
493 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh)
494 {
495 	struct net *net = sock_net(skb->sk);
496 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
497 	struct fib_rules_ops *ops = NULL;
498 	struct fib_rule *rule, *tmp;
499 	struct nlattr *tb[FRA_MAX+1];
500 	int err = -EINVAL;
501 
502 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
503 		goto errout;
504 
505 	ops = lookup_rules_ops(net, frh->family);
506 	if (ops == NULL) {
507 		err = -EAFNOSUPPORT;
508 		goto errout;
509 	}
510 
511 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
512 	if (err < 0)
513 		goto errout;
514 
515 	err = validate_rulemsg(frh, tb, ops);
516 	if (err < 0)
517 		goto errout;
518 
519 	list_for_each_entry(rule, &ops->rules_list, list) {
520 		if (frh->action && (frh->action != rule->action))
521 			continue;
522 
523 		if (frh_get_table(frh, tb) &&
524 		    (frh_get_table(frh, tb) != rule->table))
525 			continue;
526 
527 		if (tb[FRA_PRIORITY] &&
528 		    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
529 			continue;
530 
531 		if (tb[FRA_IIFNAME] &&
532 		    nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
533 			continue;
534 
535 		if (tb[FRA_OIFNAME] &&
536 		    nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
537 			continue;
538 
539 		if (tb[FRA_FWMARK] &&
540 		    (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
541 			continue;
542 
543 		if (tb[FRA_FWMASK] &&
544 		    (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
545 			continue;
546 
547 		if (tb[FRA_TUN_ID] &&
548 		    (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
549 			continue;
550 
551 		if (tb[FRA_L3MDEV] &&
552 		    (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
553 			continue;
554 
555 		if (!ops->compare(rule, frh, tb))
556 			continue;
557 
558 		if (rule->flags & FIB_RULE_PERMANENT) {
559 			err = -EPERM;
560 			goto errout;
561 		}
562 
563 		if (ops->delete) {
564 			err = ops->delete(rule);
565 			if (err)
566 				goto errout;
567 		}
568 
569 		if (rule->tun_id)
570 			ip_tunnel_unneed_metadata();
571 
572 		list_del_rcu(&rule->list);
573 
574 		if (rule->action == FR_ACT_GOTO) {
575 			ops->nr_goto_rules--;
576 			if (rtnl_dereference(rule->ctarget) == NULL)
577 				ops->unresolved_rules--;
578 		}
579 
580 		/*
581 		 * Check if this rule is a target to any of them. If so,
582 		 * disable them. As this operation is eventually very
583 		 * expensive, it is only performed if goto rules have
584 		 * actually been added.
585 		 */
586 		if (ops->nr_goto_rules > 0) {
587 			list_for_each_entry(tmp, &ops->rules_list, list) {
588 				if (rtnl_dereference(tmp->ctarget) == rule) {
589 					RCU_INIT_POINTER(tmp->ctarget, NULL);
590 					ops->unresolved_rules++;
591 				}
592 			}
593 		}
594 
595 		notify_rule_change(RTM_DELRULE, rule, ops, nlh,
596 				   NETLINK_CB(skb).portid);
597 		fib_rule_put(rule);
598 		flush_route_cache(ops);
599 		rules_ops_put(ops);
600 		return 0;
601 	}
602 
603 	err = -ENOENT;
604 errout:
605 	rules_ops_put(ops);
606 	return err;
607 }
608 EXPORT_SYMBOL_GPL(fib_nl_delrule);
609 
610 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
611 					 struct fib_rule *rule)
612 {
613 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
614 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
615 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
616 			 + nla_total_size(4) /* FRA_PRIORITY */
617 			 + nla_total_size(4) /* FRA_TABLE */
618 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
619 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
620 			 + nla_total_size(4) /* FRA_FWMARK */
621 			 + nla_total_size(4) /* FRA_FWMASK */
622 			 + nla_total_size_64bit(8); /* FRA_TUN_ID */
623 
624 	if (ops->nlmsg_payload)
625 		payload += ops->nlmsg_payload(rule);
626 
627 	return payload;
628 }
629 
630 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
631 			    u32 pid, u32 seq, int type, int flags,
632 			    struct fib_rules_ops *ops)
633 {
634 	struct nlmsghdr *nlh;
635 	struct fib_rule_hdr *frh;
636 
637 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
638 	if (nlh == NULL)
639 		return -EMSGSIZE;
640 
641 	frh = nlmsg_data(nlh);
642 	frh->family = ops->family;
643 	frh->table = rule->table;
644 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
645 		goto nla_put_failure;
646 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
647 		goto nla_put_failure;
648 	frh->res1 = 0;
649 	frh->res2 = 0;
650 	frh->action = rule->action;
651 	frh->flags = rule->flags;
652 
653 	if (rule->action == FR_ACT_GOTO &&
654 	    rcu_access_pointer(rule->ctarget) == NULL)
655 		frh->flags |= FIB_RULE_UNRESOLVED;
656 
657 	if (rule->iifname[0]) {
658 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
659 			goto nla_put_failure;
660 		if (rule->iifindex == -1)
661 			frh->flags |= FIB_RULE_IIF_DETACHED;
662 	}
663 
664 	if (rule->oifname[0]) {
665 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
666 			goto nla_put_failure;
667 		if (rule->oifindex == -1)
668 			frh->flags |= FIB_RULE_OIF_DETACHED;
669 	}
670 
671 	if ((rule->pref &&
672 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
673 	    (rule->mark &&
674 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
675 	    ((rule->mark_mask || rule->mark) &&
676 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
677 	    (rule->target &&
678 	     nla_put_u32(skb, FRA_GOTO, rule->target)) ||
679 	    (rule->tun_id &&
680 	     nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
681 	    (rule->l3mdev &&
682 	     nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)))
683 		goto nla_put_failure;
684 
685 	if (rule->suppress_ifgroup != -1) {
686 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
687 			goto nla_put_failure;
688 	}
689 
690 	if (ops->fill(rule, skb, frh) < 0)
691 		goto nla_put_failure;
692 
693 	nlmsg_end(skb, nlh);
694 	return 0;
695 
696 nla_put_failure:
697 	nlmsg_cancel(skb, nlh);
698 	return -EMSGSIZE;
699 }
700 
701 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
702 		      struct fib_rules_ops *ops)
703 {
704 	int idx = 0;
705 	struct fib_rule *rule;
706 	int err = 0;
707 
708 	rcu_read_lock();
709 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
710 		if (idx < cb->args[1])
711 			goto skip;
712 
713 		err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
714 				       cb->nlh->nlmsg_seq, RTM_NEWRULE,
715 				       NLM_F_MULTI, ops);
716 		if (err)
717 			break;
718 skip:
719 		idx++;
720 	}
721 	rcu_read_unlock();
722 	cb->args[1] = idx;
723 	rules_ops_put(ops);
724 
725 	return err;
726 }
727 
728 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
729 {
730 	struct net *net = sock_net(skb->sk);
731 	struct fib_rules_ops *ops;
732 	int idx = 0, family;
733 
734 	family = rtnl_msg_family(cb->nlh);
735 	if (family != AF_UNSPEC) {
736 		/* Protocol specific dump request */
737 		ops = lookup_rules_ops(net, family);
738 		if (ops == NULL)
739 			return -EAFNOSUPPORT;
740 
741 		dump_rules(skb, cb, ops);
742 
743 		return skb->len;
744 	}
745 
746 	rcu_read_lock();
747 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
748 		if (idx < cb->args[0] || !try_module_get(ops->owner))
749 			goto skip;
750 
751 		if (dump_rules(skb, cb, ops) < 0)
752 			break;
753 
754 		cb->args[1] = 0;
755 skip:
756 		idx++;
757 	}
758 	rcu_read_unlock();
759 	cb->args[0] = idx;
760 
761 	return skb->len;
762 }
763 
764 static void notify_rule_change(int event, struct fib_rule *rule,
765 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
766 			       u32 pid)
767 {
768 	struct net *net;
769 	struct sk_buff *skb;
770 	int err = -ENOBUFS;
771 
772 	net = ops->fro_net;
773 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
774 	if (skb == NULL)
775 		goto errout;
776 
777 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
778 	if (err < 0) {
779 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
780 		WARN_ON(err == -EMSGSIZE);
781 		kfree_skb(skb);
782 		goto errout;
783 	}
784 
785 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
786 	return;
787 errout:
788 	if (err < 0)
789 		rtnl_set_sk_err(net, ops->nlgroup, err);
790 }
791 
792 static void attach_rules(struct list_head *rules, struct net_device *dev)
793 {
794 	struct fib_rule *rule;
795 
796 	list_for_each_entry(rule, rules, list) {
797 		if (rule->iifindex == -1 &&
798 		    strcmp(dev->name, rule->iifname) == 0)
799 			rule->iifindex = dev->ifindex;
800 		if (rule->oifindex == -1 &&
801 		    strcmp(dev->name, rule->oifname) == 0)
802 			rule->oifindex = dev->ifindex;
803 	}
804 }
805 
806 static void detach_rules(struct list_head *rules, struct net_device *dev)
807 {
808 	struct fib_rule *rule;
809 
810 	list_for_each_entry(rule, rules, list) {
811 		if (rule->iifindex == dev->ifindex)
812 			rule->iifindex = -1;
813 		if (rule->oifindex == dev->ifindex)
814 			rule->oifindex = -1;
815 	}
816 }
817 
818 
819 static int fib_rules_event(struct notifier_block *this, unsigned long event,
820 			   void *ptr)
821 {
822 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
823 	struct net *net = dev_net(dev);
824 	struct fib_rules_ops *ops;
825 
826 	ASSERT_RTNL();
827 
828 	switch (event) {
829 	case NETDEV_REGISTER:
830 		list_for_each_entry(ops, &net->rules_ops, list)
831 			attach_rules(&ops->rules_list, dev);
832 		break;
833 
834 	case NETDEV_CHANGENAME:
835 		list_for_each_entry(ops, &net->rules_ops, list) {
836 			detach_rules(&ops->rules_list, dev);
837 			attach_rules(&ops->rules_list, dev);
838 		}
839 		break;
840 
841 	case NETDEV_UNREGISTER:
842 		list_for_each_entry(ops, &net->rules_ops, list)
843 			detach_rules(&ops->rules_list, dev);
844 		break;
845 	}
846 
847 	return NOTIFY_DONE;
848 }
849 
850 static struct notifier_block fib_rules_notifier = {
851 	.notifier_call = fib_rules_event,
852 };
853 
854 static int __net_init fib_rules_net_init(struct net *net)
855 {
856 	INIT_LIST_HEAD(&net->rules_ops);
857 	spin_lock_init(&net->rules_mod_lock);
858 	return 0;
859 }
860 
861 static struct pernet_operations fib_rules_net_ops = {
862 	.init = fib_rules_net_init,
863 };
864 
865 static int __init fib_rules_init(void)
866 {
867 	int err;
868 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
869 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
870 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
871 
872 	err = register_pernet_subsys(&fib_rules_net_ops);
873 	if (err < 0)
874 		goto fail;
875 
876 	err = register_netdevice_notifier(&fib_rules_notifier);
877 	if (err < 0)
878 		goto fail_unregister;
879 
880 	return 0;
881 
882 fail_unregister:
883 	unregister_pernet_subsys(&fib_rules_net_ops);
884 fail:
885 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
886 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
887 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
888 	return err;
889 }
890 
891 subsys_initcall(fib_rules_init);
892