xref: /linux/net/core/fib_rules.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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 static const struct fib_kuid_range fib_kuid_range_unset = {
22 	KUIDT_INIT(0),
23 	KUIDT_INIT(~0),
24 };
25 
26 bool fib_rule_matchall(const struct fib_rule *rule)
27 {
28 	if (rule->iifindex || rule->oifindex || rule->mark || rule->tun_id ||
29 	    rule->flags)
30 		return false;
31 	if (rule->suppress_ifgroup != -1 || rule->suppress_prefixlen != -1)
32 		return false;
33 	if (!uid_eq(rule->uid_range.start, fib_kuid_range_unset.start) ||
34 	    !uid_eq(rule->uid_range.end, fib_kuid_range_unset.end))
35 		return false;
36 	return true;
37 }
38 EXPORT_SYMBOL_GPL(fib_rule_matchall);
39 
40 int fib_default_rule_add(struct fib_rules_ops *ops,
41 			 u32 pref, u32 table, u32 flags)
42 {
43 	struct fib_rule *r;
44 
45 	r = kzalloc(ops->rule_size, GFP_KERNEL);
46 	if (r == NULL)
47 		return -ENOMEM;
48 
49 	refcount_set(&r->refcnt, 1);
50 	r->action = FR_ACT_TO_TBL;
51 	r->pref = pref;
52 	r->table = table;
53 	r->flags = flags;
54 	r->fr_net = ops->fro_net;
55 	r->uid_range = fib_kuid_range_unset;
56 
57 	r->suppress_prefixlen = -1;
58 	r->suppress_ifgroup = -1;
59 
60 	/* The lock is not required here, the list in unreacheable
61 	 * at the moment this function is called */
62 	list_add_tail(&r->list, &ops->rules_list);
63 	return 0;
64 }
65 EXPORT_SYMBOL(fib_default_rule_add);
66 
67 static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
68 {
69 	struct list_head *pos;
70 	struct fib_rule *rule;
71 
72 	if (!list_empty(&ops->rules_list)) {
73 		pos = ops->rules_list.next;
74 		if (pos->next != &ops->rules_list) {
75 			rule = list_entry(pos->next, struct fib_rule, list);
76 			if (rule->pref)
77 				return rule->pref - 1;
78 		}
79 	}
80 
81 	return 0;
82 }
83 
84 static void notify_rule_change(int event, struct fib_rule *rule,
85 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
86 			       u32 pid);
87 
88 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
89 {
90 	struct fib_rules_ops *ops;
91 
92 	rcu_read_lock();
93 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
94 		if (ops->family == family) {
95 			if (!try_module_get(ops->owner))
96 				ops = NULL;
97 			rcu_read_unlock();
98 			return ops;
99 		}
100 	}
101 	rcu_read_unlock();
102 
103 	return NULL;
104 }
105 
106 static void rules_ops_put(struct fib_rules_ops *ops)
107 {
108 	if (ops)
109 		module_put(ops->owner);
110 }
111 
112 static void flush_route_cache(struct fib_rules_ops *ops)
113 {
114 	if (ops->flush_cache)
115 		ops->flush_cache(ops);
116 }
117 
118 static int __fib_rules_register(struct fib_rules_ops *ops)
119 {
120 	int err = -EEXIST;
121 	struct fib_rules_ops *o;
122 	struct net *net;
123 
124 	net = ops->fro_net;
125 
126 	if (ops->rule_size < sizeof(struct fib_rule))
127 		return -EINVAL;
128 
129 	if (ops->match == NULL || ops->configure == NULL ||
130 	    ops->compare == NULL || ops->fill == NULL ||
131 	    ops->action == NULL)
132 		return -EINVAL;
133 
134 	spin_lock(&net->rules_mod_lock);
135 	list_for_each_entry(o, &net->rules_ops, list)
136 		if (ops->family == o->family)
137 			goto errout;
138 
139 	list_add_tail_rcu(&ops->list, &net->rules_ops);
140 	err = 0;
141 errout:
142 	spin_unlock(&net->rules_mod_lock);
143 
144 	return err;
145 }
146 
147 struct fib_rules_ops *
148 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
149 {
150 	struct fib_rules_ops *ops;
151 	int err;
152 
153 	ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
154 	if (ops == NULL)
155 		return ERR_PTR(-ENOMEM);
156 
157 	INIT_LIST_HEAD(&ops->rules_list);
158 	ops->fro_net = net;
159 
160 	err = __fib_rules_register(ops);
161 	if (err) {
162 		kfree(ops);
163 		ops = ERR_PTR(err);
164 	}
165 
166 	return ops;
167 }
168 EXPORT_SYMBOL_GPL(fib_rules_register);
169 
170 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
171 {
172 	struct fib_rule *rule, *tmp;
173 
174 	list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
175 		list_del_rcu(&rule->list);
176 		if (ops->delete)
177 			ops->delete(rule);
178 		fib_rule_put(rule);
179 	}
180 }
181 
182 void fib_rules_unregister(struct fib_rules_ops *ops)
183 {
184 	struct net *net = ops->fro_net;
185 
186 	spin_lock(&net->rules_mod_lock);
187 	list_del_rcu(&ops->list);
188 	spin_unlock(&net->rules_mod_lock);
189 
190 	fib_rules_cleanup_ops(ops);
191 	kfree_rcu(ops, rcu);
192 }
193 EXPORT_SYMBOL_GPL(fib_rules_unregister);
194 
195 static int uid_range_set(struct fib_kuid_range *range)
196 {
197 	return uid_valid(range->start) && uid_valid(range->end);
198 }
199 
200 static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
201 {
202 	struct fib_rule_uid_range *in;
203 	struct fib_kuid_range out;
204 
205 	in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
206 
207 	out.start = make_kuid(current_user_ns(), in->start);
208 	out.end = make_kuid(current_user_ns(), in->end);
209 
210 	return out;
211 }
212 
213 static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
214 {
215 	struct fib_rule_uid_range out = {
216 		from_kuid_munged(current_user_ns(), range->start),
217 		from_kuid_munged(current_user_ns(), range->end)
218 	};
219 
220 	return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
221 }
222 
223 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
224 			  struct flowi *fl, int flags,
225 			  struct fib_lookup_arg *arg)
226 {
227 	int ret = 0;
228 
229 	if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
230 		goto out;
231 
232 	if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
233 		goto out;
234 
235 	if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
236 		goto out;
237 
238 	if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
239 		goto out;
240 
241 	if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
242 		goto out;
243 
244 	if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
245 	    uid_gt(fl->flowi_uid, rule->uid_range.end))
246 		goto out;
247 
248 	ret = ops->match(rule, fl, flags);
249 out:
250 	return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
251 }
252 
253 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
254 		     int flags, struct fib_lookup_arg *arg)
255 {
256 	struct fib_rule *rule;
257 	int err;
258 
259 	rcu_read_lock();
260 
261 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
262 jumped:
263 		if (!fib_rule_match(rule, ops, fl, flags, arg))
264 			continue;
265 
266 		if (rule->action == FR_ACT_GOTO) {
267 			struct fib_rule *target;
268 
269 			target = rcu_dereference(rule->ctarget);
270 			if (target == NULL) {
271 				continue;
272 			} else {
273 				rule = target;
274 				goto jumped;
275 			}
276 		} else if (rule->action == FR_ACT_NOP)
277 			continue;
278 		else
279 			err = ops->action(rule, fl, flags, arg);
280 
281 		if (!err && ops->suppress && ops->suppress(rule, arg))
282 			continue;
283 
284 		if (err != -EAGAIN) {
285 			if ((arg->flags & FIB_LOOKUP_NOREF) ||
286 			    likely(refcount_inc_not_zero(&rule->refcnt))) {
287 				arg->rule = rule;
288 				goto out;
289 			}
290 			break;
291 		}
292 	}
293 
294 	err = -ESRCH;
295 out:
296 	rcu_read_unlock();
297 
298 	return err;
299 }
300 EXPORT_SYMBOL_GPL(fib_rules_lookup);
301 
302 static int call_fib_rule_notifier(struct notifier_block *nb, struct net *net,
303 				  enum fib_event_type event_type,
304 				  struct fib_rule *rule, int family)
305 {
306 	struct fib_rule_notifier_info info = {
307 		.info.family = family,
308 		.rule = rule,
309 	};
310 
311 	return call_fib_notifier(nb, net, event_type, &info.info);
312 }
313 
314 static int call_fib_rule_notifiers(struct net *net,
315 				   enum fib_event_type event_type,
316 				   struct fib_rule *rule,
317 				   struct fib_rules_ops *ops,
318 				   struct netlink_ext_ack *extack)
319 {
320 	struct fib_rule_notifier_info info = {
321 		.info.family = ops->family,
322 		.info.extack = extack,
323 		.rule = rule,
324 	};
325 
326 	ops->fib_rules_seq++;
327 	return call_fib_notifiers(net, event_type, &info.info);
328 }
329 
330 /* Called with rcu_read_lock() */
331 int fib_rules_dump(struct net *net, struct notifier_block *nb, int family)
332 {
333 	struct fib_rules_ops *ops;
334 	struct fib_rule *rule;
335 
336 	ops = lookup_rules_ops(net, family);
337 	if (!ops)
338 		return -EAFNOSUPPORT;
339 	list_for_each_entry_rcu(rule, &ops->rules_list, list)
340 		call_fib_rule_notifier(nb, net, FIB_EVENT_RULE_ADD, rule,
341 				       family);
342 	rules_ops_put(ops);
343 
344 	return 0;
345 }
346 EXPORT_SYMBOL_GPL(fib_rules_dump);
347 
348 unsigned int fib_rules_seq_read(struct net *net, int family)
349 {
350 	unsigned int fib_rules_seq;
351 	struct fib_rules_ops *ops;
352 
353 	ASSERT_RTNL();
354 
355 	ops = lookup_rules_ops(net, family);
356 	if (!ops)
357 		return 0;
358 	fib_rules_seq = ops->fib_rules_seq;
359 	rules_ops_put(ops);
360 
361 	return fib_rules_seq;
362 }
363 EXPORT_SYMBOL_GPL(fib_rules_seq_read);
364 
365 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
366 			    struct fib_rules_ops *ops)
367 {
368 	int err = -EINVAL;
369 
370 	if (frh->src_len)
371 		if (tb[FRA_SRC] == NULL ||
372 		    frh->src_len > (ops->addr_size * 8) ||
373 		    nla_len(tb[FRA_SRC]) != ops->addr_size)
374 			goto errout;
375 
376 	if (frh->dst_len)
377 		if (tb[FRA_DST] == NULL ||
378 		    frh->dst_len > (ops->addr_size * 8) ||
379 		    nla_len(tb[FRA_DST]) != ops->addr_size)
380 			goto errout;
381 
382 	err = 0;
383 errout:
384 	return err;
385 }
386 
387 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
388 		       struct nlattr **tb, struct fib_rule *rule)
389 {
390 	struct fib_rule *r;
391 
392 	list_for_each_entry(r, &ops->rules_list, list) {
393 		if (r->action != rule->action)
394 			continue;
395 
396 		if (r->table != rule->table)
397 			continue;
398 
399 		if (r->pref != rule->pref)
400 			continue;
401 
402 		if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
403 			continue;
404 
405 		if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
406 			continue;
407 
408 		if (r->mark != rule->mark)
409 			continue;
410 
411 		if (r->mark_mask != rule->mark_mask)
412 			continue;
413 
414 		if (r->tun_id != rule->tun_id)
415 			continue;
416 
417 		if (r->fr_net != rule->fr_net)
418 			continue;
419 
420 		if (r->l3mdev != rule->l3mdev)
421 			continue;
422 
423 		if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
424 		    !uid_eq(r->uid_range.end, rule->uid_range.end))
425 			continue;
426 
427 		if (!ops->compare(r, frh, tb))
428 			continue;
429 		return 1;
430 	}
431 	return 0;
432 }
433 
434 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh,
435 		   struct netlink_ext_ack *extack)
436 {
437 	struct net *net = sock_net(skb->sk);
438 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
439 	struct fib_rules_ops *ops = NULL;
440 	struct fib_rule *rule, *r, *last = NULL;
441 	struct nlattr *tb[FRA_MAX+1];
442 	int err = -EINVAL, unresolved = 0;
443 
444 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
445 		goto errout;
446 
447 	ops = lookup_rules_ops(net, frh->family);
448 	if (ops == NULL) {
449 		err = -EAFNOSUPPORT;
450 		goto errout;
451 	}
452 
453 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy, extack);
454 	if (err < 0)
455 		goto errout;
456 
457 	err = validate_rulemsg(frh, tb, ops);
458 	if (err < 0)
459 		goto errout;
460 
461 	rule = kzalloc(ops->rule_size, GFP_KERNEL);
462 	if (rule == NULL) {
463 		err = -ENOMEM;
464 		goto errout;
465 	}
466 	refcount_set(&rule->refcnt, 1);
467 	rule->fr_net = net;
468 
469 	rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
470 	                              : fib_default_rule_pref(ops);
471 
472 	if (tb[FRA_IIFNAME]) {
473 		struct net_device *dev;
474 
475 		rule->iifindex = -1;
476 		nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
477 		dev = __dev_get_by_name(net, rule->iifname);
478 		if (dev)
479 			rule->iifindex = dev->ifindex;
480 	}
481 
482 	if (tb[FRA_OIFNAME]) {
483 		struct net_device *dev;
484 
485 		rule->oifindex = -1;
486 		nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
487 		dev = __dev_get_by_name(net, rule->oifname);
488 		if (dev)
489 			rule->oifindex = dev->ifindex;
490 	}
491 
492 	if (tb[FRA_FWMARK]) {
493 		rule->mark = nla_get_u32(tb[FRA_FWMARK]);
494 		if (rule->mark)
495 			/* compatibility: if the mark value is non-zero all bits
496 			 * are compared unless a mask is explicitly specified.
497 			 */
498 			rule->mark_mask = 0xFFFFFFFF;
499 	}
500 
501 	if (tb[FRA_FWMASK])
502 		rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
503 
504 	if (tb[FRA_TUN_ID])
505 		rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
506 
507 	err = -EINVAL;
508 	if (tb[FRA_L3MDEV]) {
509 #ifdef CONFIG_NET_L3_MASTER_DEV
510 		rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
511 		if (rule->l3mdev != 1)
512 #endif
513 			goto errout_free;
514 	}
515 
516 	rule->action = frh->action;
517 	rule->flags = frh->flags;
518 	rule->table = frh_get_table(frh, tb);
519 	if (tb[FRA_SUPPRESS_PREFIXLEN])
520 		rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
521 	else
522 		rule->suppress_prefixlen = -1;
523 
524 	if (tb[FRA_SUPPRESS_IFGROUP])
525 		rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
526 	else
527 		rule->suppress_ifgroup = -1;
528 
529 	if (tb[FRA_GOTO]) {
530 		if (rule->action != FR_ACT_GOTO)
531 			goto errout_free;
532 
533 		rule->target = nla_get_u32(tb[FRA_GOTO]);
534 		/* Backward jumps are prohibited to avoid endless loops */
535 		if (rule->target <= rule->pref)
536 			goto errout_free;
537 
538 		list_for_each_entry(r, &ops->rules_list, list) {
539 			if (r->pref == rule->target) {
540 				RCU_INIT_POINTER(rule->ctarget, r);
541 				break;
542 			}
543 		}
544 
545 		if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
546 			unresolved = 1;
547 	} else if (rule->action == FR_ACT_GOTO)
548 		goto errout_free;
549 
550 	if (rule->l3mdev && rule->table)
551 		goto errout_free;
552 
553 	if (tb[FRA_UID_RANGE]) {
554 		if (current_user_ns() != net->user_ns) {
555 			err = -EPERM;
556 			goto errout_free;
557 		}
558 
559 		rule->uid_range = nla_get_kuid_range(tb);
560 
561 		if (!uid_range_set(&rule->uid_range) ||
562 		    !uid_lte(rule->uid_range.start, rule->uid_range.end))
563 			goto errout_free;
564 	} else {
565 		rule->uid_range = fib_kuid_range_unset;
566 	}
567 
568 	if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
569 	    rule_exists(ops, frh, tb, rule)) {
570 		err = -EEXIST;
571 		goto errout_free;
572 	}
573 
574 	err = ops->configure(rule, skb, frh, tb);
575 	if (err < 0)
576 		goto errout_free;
577 
578 	list_for_each_entry(r, &ops->rules_list, list) {
579 		if (r->pref > rule->pref)
580 			break;
581 		last = r;
582 	}
583 
584 	if (last)
585 		list_add_rcu(&rule->list, &last->list);
586 	else
587 		list_add_rcu(&rule->list, &ops->rules_list);
588 
589 	if (ops->unresolved_rules) {
590 		/*
591 		 * There are unresolved goto rules in the list, check if
592 		 * any of them are pointing to this new rule.
593 		 */
594 		list_for_each_entry(r, &ops->rules_list, list) {
595 			if (r->action == FR_ACT_GOTO &&
596 			    r->target == rule->pref &&
597 			    rtnl_dereference(r->ctarget) == NULL) {
598 				rcu_assign_pointer(r->ctarget, rule);
599 				if (--ops->unresolved_rules == 0)
600 					break;
601 			}
602 		}
603 	}
604 
605 	if (rule->action == FR_ACT_GOTO)
606 		ops->nr_goto_rules++;
607 
608 	if (unresolved)
609 		ops->unresolved_rules++;
610 
611 	if (rule->tun_id)
612 		ip_tunnel_need_metadata();
613 
614 	call_fib_rule_notifiers(net, FIB_EVENT_RULE_ADD, rule, ops, extack);
615 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
616 	flush_route_cache(ops);
617 	rules_ops_put(ops);
618 	return 0;
619 
620 errout_free:
621 	kfree(rule);
622 errout:
623 	rules_ops_put(ops);
624 	return err;
625 }
626 EXPORT_SYMBOL_GPL(fib_nl_newrule);
627 
628 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh,
629 		   struct netlink_ext_ack *extack)
630 {
631 	struct net *net = sock_net(skb->sk);
632 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
633 	struct fib_rules_ops *ops = NULL;
634 	struct fib_rule *rule, *r;
635 	struct nlattr *tb[FRA_MAX+1];
636 	struct fib_kuid_range range;
637 	int err = -EINVAL;
638 
639 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
640 		goto errout;
641 
642 	ops = lookup_rules_ops(net, frh->family);
643 	if (ops == NULL) {
644 		err = -EAFNOSUPPORT;
645 		goto errout;
646 	}
647 
648 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy, extack);
649 	if (err < 0)
650 		goto errout;
651 
652 	err = validate_rulemsg(frh, tb, ops);
653 	if (err < 0)
654 		goto errout;
655 
656 	if (tb[FRA_UID_RANGE]) {
657 		range = nla_get_kuid_range(tb);
658 		if (!uid_range_set(&range)) {
659 			err = -EINVAL;
660 			goto errout;
661 		}
662 	} else {
663 		range = fib_kuid_range_unset;
664 	}
665 
666 	list_for_each_entry(rule, &ops->rules_list, list) {
667 		if (frh->action && (frh->action != rule->action))
668 			continue;
669 
670 		if (frh_get_table(frh, tb) &&
671 		    (frh_get_table(frh, tb) != rule->table))
672 			continue;
673 
674 		if (tb[FRA_PRIORITY] &&
675 		    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
676 			continue;
677 
678 		if (tb[FRA_IIFNAME] &&
679 		    nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
680 			continue;
681 
682 		if (tb[FRA_OIFNAME] &&
683 		    nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
684 			continue;
685 
686 		if (tb[FRA_FWMARK] &&
687 		    (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
688 			continue;
689 
690 		if (tb[FRA_FWMASK] &&
691 		    (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
692 			continue;
693 
694 		if (tb[FRA_TUN_ID] &&
695 		    (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
696 			continue;
697 
698 		if (tb[FRA_L3MDEV] &&
699 		    (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
700 			continue;
701 
702 		if (uid_range_set(&range) &&
703 		    (!uid_eq(rule->uid_range.start, range.start) ||
704 		     !uid_eq(rule->uid_range.end, range.end)))
705 			continue;
706 
707 		if (!ops->compare(rule, frh, tb))
708 			continue;
709 
710 		if (rule->flags & FIB_RULE_PERMANENT) {
711 			err = -EPERM;
712 			goto errout;
713 		}
714 
715 		if (ops->delete) {
716 			err = ops->delete(rule);
717 			if (err)
718 				goto errout;
719 		}
720 
721 		if (rule->tun_id)
722 			ip_tunnel_unneed_metadata();
723 
724 		list_del_rcu(&rule->list);
725 
726 		if (rule->action == FR_ACT_GOTO) {
727 			ops->nr_goto_rules--;
728 			if (rtnl_dereference(rule->ctarget) == NULL)
729 				ops->unresolved_rules--;
730 		}
731 
732 		/*
733 		 * Check if this rule is a target to any of them. If so,
734 		 * adjust to the next one with the same preference or
735 		 * disable them. As this operation is eventually very
736 		 * expensive, it is only performed if goto rules, except
737 		 * current if it is goto rule, have actually been added.
738 		 */
739 		if (ops->nr_goto_rules > 0) {
740 			struct fib_rule *n;
741 
742 			n = list_next_entry(rule, list);
743 			if (&n->list == &ops->rules_list || n->pref != rule->pref)
744 				n = NULL;
745 			list_for_each_entry(r, &ops->rules_list, list) {
746 				if (rtnl_dereference(r->ctarget) != rule)
747 					continue;
748 				rcu_assign_pointer(r->ctarget, n);
749 				if (!n)
750 					ops->unresolved_rules++;
751 			}
752 		}
753 
754 		call_fib_rule_notifiers(net, FIB_EVENT_RULE_DEL, rule, ops,
755 					NULL);
756 		notify_rule_change(RTM_DELRULE, rule, ops, nlh,
757 				   NETLINK_CB(skb).portid);
758 		fib_rule_put(rule);
759 		flush_route_cache(ops);
760 		rules_ops_put(ops);
761 		return 0;
762 	}
763 
764 	err = -ENOENT;
765 errout:
766 	rules_ops_put(ops);
767 	return err;
768 }
769 EXPORT_SYMBOL_GPL(fib_nl_delrule);
770 
771 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
772 					 struct fib_rule *rule)
773 {
774 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
775 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
776 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
777 			 + nla_total_size(4) /* FRA_PRIORITY */
778 			 + nla_total_size(4) /* FRA_TABLE */
779 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
780 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
781 			 + nla_total_size(4) /* FRA_FWMARK */
782 			 + nla_total_size(4) /* FRA_FWMASK */
783 			 + nla_total_size_64bit(8) /* FRA_TUN_ID */
784 			 + nla_total_size(sizeof(struct fib_kuid_range));
785 
786 	if (ops->nlmsg_payload)
787 		payload += ops->nlmsg_payload(rule);
788 
789 	return payload;
790 }
791 
792 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
793 			    u32 pid, u32 seq, int type, int flags,
794 			    struct fib_rules_ops *ops)
795 {
796 	struct nlmsghdr *nlh;
797 	struct fib_rule_hdr *frh;
798 
799 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
800 	if (nlh == NULL)
801 		return -EMSGSIZE;
802 
803 	frh = nlmsg_data(nlh);
804 	frh->family = ops->family;
805 	frh->table = rule->table;
806 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
807 		goto nla_put_failure;
808 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
809 		goto nla_put_failure;
810 	frh->res1 = 0;
811 	frh->res2 = 0;
812 	frh->action = rule->action;
813 	frh->flags = rule->flags;
814 
815 	if (rule->action == FR_ACT_GOTO &&
816 	    rcu_access_pointer(rule->ctarget) == NULL)
817 		frh->flags |= FIB_RULE_UNRESOLVED;
818 
819 	if (rule->iifname[0]) {
820 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
821 			goto nla_put_failure;
822 		if (rule->iifindex == -1)
823 			frh->flags |= FIB_RULE_IIF_DETACHED;
824 	}
825 
826 	if (rule->oifname[0]) {
827 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
828 			goto nla_put_failure;
829 		if (rule->oifindex == -1)
830 			frh->flags |= FIB_RULE_OIF_DETACHED;
831 	}
832 
833 	if ((rule->pref &&
834 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
835 	    (rule->mark &&
836 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
837 	    ((rule->mark_mask || rule->mark) &&
838 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
839 	    (rule->target &&
840 	     nla_put_u32(skb, FRA_GOTO, rule->target)) ||
841 	    (rule->tun_id &&
842 	     nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
843 	    (rule->l3mdev &&
844 	     nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
845 	    (uid_range_set(&rule->uid_range) &&
846 	     nla_put_uid_range(skb, &rule->uid_range)))
847 		goto nla_put_failure;
848 
849 	if (rule->suppress_ifgroup != -1) {
850 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
851 			goto nla_put_failure;
852 	}
853 
854 	if (ops->fill(rule, skb, frh) < 0)
855 		goto nla_put_failure;
856 
857 	nlmsg_end(skb, nlh);
858 	return 0;
859 
860 nla_put_failure:
861 	nlmsg_cancel(skb, nlh);
862 	return -EMSGSIZE;
863 }
864 
865 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
866 		      struct fib_rules_ops *ops)
867 {
868 	int idx = 0;
869 	struct fib_rule *rule;
870 	int err = 0;
871 
872 	rcu_read_lock();
873 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
874 		if (idx < cb->args[1])
875 			goto skip;
876 
877 		err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
878 				       cb->nlh->nlmsg_seq, RTM_NEWRULE,
879 				       NLM_F_MULTI, ops);
880 		if (err)
881 			break;
882 skip:
883 		idx++;
884 	}
885 	rcu_read_unlock();
886 	cb->args[1] = idx;
887 	rules_ops_put(ops);
888 
889 	return err;
890 }
891 
892 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
893 {
894 	struct net *net = sock_net(skb->sk);
895 	struct fib_rules_ops *ops;
896 	int idx = 0, family;
897 
898 	family = rtnl_msg_family(cb->nlh);
899 	if (family != AF_UNSPEC) {
900 		/* Protocol specific dump request */
901 		ops = lookup_rules_ops(net, family);
902 		if (ops == NULL)
903 			return -EAFNOSUPPORT;
904 
905 		dump_rules(skb, cb, ops);
906 
907 		return skb->len;
908 	}
909 
910 	rcu_read_lock();
911 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
912 		if (idx < cb->args[0] || !try_module_get(ops->owner))
913 			goto skip;
914 
915 		if (dump_rules(skb, cb, ops) < 0)
916 			break;
917 
918 		cb->args[1] = 0;
919 skip:
920 		idx++;
921 	}
922 	rcu_read_unlock();
923 	cb->args[0] = idx;
924 
925 	return skb->len;
926 }
927 
928 static void notify_rule_change(int event, struct fib_rule *rule,
929 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
930 			       u32 pid)
931 {
932 	struct net *net;
933 	struct sk_buff *skb;
934 	int err = -ENOBUFS;
935 
936 	net = ops->fro_net;
937 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
938 	if (skb == NULL)
939 		goto errout;
940 
941 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
942 	if (err < 0) {
943 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
944 		WARN_ON(err == -EMSGSIZE);
945 		kfree_skb(skb);
946 		goto errout;
947 	}
948 
949 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
950 	return;
951 errout:
952 	if (err < 0)
953 		rtnl_set_sk_err(net, ops->nlgroup, err);
954 }
955 
956 static void attach_rules(struct list_head *rules, struct net_device *dev)
957 {
958 	struct fib_rule *rule;
959 
960 	list_for_each_entry(rule, rules, list) {
961 		if (rule->iifindex == -1 &&
962 		    strcmp(dev->name, rule->iifname) == 0)
963 			rule->iifindex = dev->ifindex;
964 		if (rule->oifindex == -1 &&
965 		    strcmp(dev->name, rule->oifname) == 0)
966 			rule->oifindex = dev->ifindex;
967 	}
968 }
969 
970 static void detach_rules(struct list_head *rules, struct net_device *dev)
971 {
972 	struct fib_rule *rule;
973 
974 	list_for_each_entry(rule, rules, list) {
975 		if (rule->iifindex == dev->ifindex)
976 			rule->iifindex = -1;
977 		if (rule->oifindex == dev->ifindex)
978 			rule->oifindex = -1;
979 	}
980 }
981 
982 
983 static int fib_rules_event(struct notifier_block *this, unsigned long event,
984 			   void *ptr)
985 {
986 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
987 	struct net *net = dev_net(dev);
988 	struct fib_rules_ops *ops;
989 
990 	ASSERT_RTNL();
991 
992 	switch (event) {
993 	case NETDEV_REGISTER:
994 		list_for_each_entry(ops, &net->rules_ops, list)
995 			attach_rules(&ops->rules_list, dev);
996 		break;
997 
998 	case NETDEV_CHANGENAME:
999 		list_for_each_entry(ops, &net->rules_ops, list) {
1000 			detach_rules(&ops->rules_list, dev);
1001 			attach_rules(&ops->rules_list, dev);
1002 		}
1003 		break;
1004 
1005 	case NETDEV_UNREGISTER:
1006 		list_for_each_entry(ops, &net->rules_ops, list)
1007 			detach_rules(&ops->rules_list, dev);
1008 		break;
1009 	}
1010 
1011 	return NOTIFY_DONE;
1012 }
1013 
1014 static struct notifier_block fib_rules_notifier = {
1015 	.notifier_call = fib_rules_event,
1016 };
1017 
1018 static int __net_init fib_rules_net_init(struct net *net)
1019 {
1020 	INIT_LIST_HEAD(&net->rules_ops);
1021 	spin_lock_init(&net->rules_mod_lock);
1022 	return 0;
1023 }
1024 
1025 static void __net_exit fib_rules_net_exit(struct net *net)
1026 {
1027 	WARN_ON_ONCE(!list_empty(&net->rules_ops));
1028 }
1029 
1030 static struct pernet_operations fib_rules_net_ops = {
1031 	.init = fib_rules_net_init,
1032 	.exit = fib_rules_net_exit,
1033 };
1034 
1035 static int __init fib_rules_init(void)
1036 {
1037 	int err;
1038 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, 0);
1039 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, 0);
1040 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, 0);
1041 
1042 	err = register_pernet_subsys(&fib_rules_net_ops);
1043 	if (err < 0)
1044 		goto fail;
1045 
1046 	err = register_netdevice_notifier(&fib_rules_notifier);
1047 	if (err < 0)
1048 		goto fail_unregister;
1049 
1050 	return 0;
1051 
1052 fail_unregister:
1053 	unregister_pernet_subsys(&fib_rules_net_ops);
1054 fail:
1055 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
1056 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
1057 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
1058 	return err;
1059 }
1060 
1061 subsys_initcall(fib_rules_init);
1062