xref: /openbmc/linux/net/core/fib_rules.c (revision aac5987a)
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 int fib_default_rule_add(struct fib_rules_ops *ops,
27 			 u32 pref, u32 table, u32 flags)
28 {
29 	struct fib_rule *r;
30 
31 	r = kzalloc(ops->rule_size, GFP_KERNEL);
32 	if (r == NULL)
33 		return -ENOMEM;
34 
35 	atomic_set(&r->refcnt, 1);
36 	r->action = FR_ACT_TO_TBL;
37 	r->pref = pref;
38 	r->table = table;
39 	r->flags = flags;
40 	r->fr_net = ops->fro_net;
41 	r->uid_range = fib_kuid_range_unset;
42 
43 	r->suppress_prefixlen = -1;
44 	r->suppress_ifgroup = -1;
45 
46 	/* The lock is not required here, the list in unreacheable
47 	 * at the moment this function is called */
48 	list_add_tail(&r->list, &ops->rules_list);
49 	return 0;
50 }
51 EXPORT_SYMBOL(fib_default_rule_add);
52 
53 static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
54 {
55 	struct list_head *pos;
56 	struct fib_rule *rule;
57 
58 	if (!list_empty(&ops->rules_list)) {
59 		pos = ops->rules_list.next;
60 		if (pos->next != &ops->rules_list) {
61 			rule = list_entry(pos->next, struct fib_rule, list);
62 			if (rule->pref)
63 				return rule->pref - 1;
64 		}
65 	}
66 
67 	return 0;
68 }
69 
70 static void notify_rule_change(int event, struct fib_rule *rule,
71 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
72 			       u32 pid);
73 
74 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
75 {
76 	struct fib_rules_ops *ops;
77 
78 	rcu_read_lock();
79 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
80 		if (ops->family == family) {
81 			if (!try_module_get(ops->owner))
82 				ops = NULL;
83 			rcu_read_unlock();
84 			return ops;
85 		}
86 	}
87 	rcu_read_unlock();
88 
89 	return NULL;
90 }
91 
92 static void rules_ops_put(struct fib_rules_ops *ops)
93 {
94 	if (ops)
95 		module_put(ops->owner);
96 }
97 
98 static void flush_route_cache(struct fib_rules_ops *ops)
99 {
100 	if (ops->flush_cache)
101 		ops->flush_cache(ops);
102 }
103 
104 static int __fib_rules_register(struct fib_rules_ops *ops)
105 {
106 	int err = -EEXIST;
107 	struct fib_rules_ops *o;
108 	struct net *net;
109 
110 	net = ops->fro_net;
111 
112 	if (ops->rule_size < sizeof(struct fib_rule))
113 		return -EINVAL;
114 
115 	if (ops->match == NULL || ops->configure == NULL ||
116 	    ops->compare == NULL || ops->fill == NULL ||
117 	    ops->action == NULL)
118 		return -EINVAL;
119 
120 	spin_lock(&net->rules_mod_lock);
121 	list_for_each_entry(o, &net->rules_ops, list)
122 		if (ops->family == o->family)
123 			goto errout;
124 
125 	list_add_tail_rcu(&ops->list, &net->rules_ops);
126 	err = 0;
127 errout:
128 	spin_unlock(&net->rules_mod_lock);
129 
130 	return err;
131 }
132 
133 struct fib_rules_ops *
134 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
135 {
136 	struct fib_rules_ops *ops;
137 	int err;
138 
139 	ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
140 	if (ops == NULL)
141 		return ERR_PTR(-ENOMEM);
142 
143 	INIT_LIST_HEAD(&ops->rules_list);
144 	ops->fro_net = net;
145 
146 	err = __fib_rules_register(ops);
147 	if (err) {
148 		kfree(ops);
149 		ops = ERR_PTR(err);
150 	}
151 
152 	return ops;
153 }
154 EXPORT_SYMBOL_GPL(fib_rules_register);
155 
156 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
157 {
158 	struct fib_rule *rule, *tmp;
159 
160 	list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
161 		list_del_rcu(&rule->list);
162 		if (ops->delete)
163 			ops->delete(rule);
164 		fib_rule_put(rule);
165 	}
166 }
167 
168 void fib_rules_unregister(struct fib_rules_ops *ops)
169 {
170 	struct net *net = ops->fro_net;
171 
172 	spin_lock(&net->rules_mod_lock);
173 	list_del_rcu(&ops->list);
174 	spin_unlock(&net->rules_mod_lock);
175 
176 	fib_rules_cleanup_ops(ops);
177 	kfree_rcu(ops, rcu);
178 }
179 EXPORT_SYMBOL_GPL(fib_rules_unregister);
180 
181 static int uid_range_set(struct fib_kuid_range *range)
182 {
183 	return uid_valid(range->start) && uid_valid(range->end);
184 }
185 
186 static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
187 {
188 	struct fib_rule_uid_range *in;
189 	struct fib_kuid_range out;
190 
191 	in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
192 
193 	out.start = make_kuid(current_user_ns(), in->start);
194 	out.end = make_kuid(current_user_ns(), in->end);
195 
196 	return out;
197 }
198 
199 static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
200 {
201 	struct fib_rule_uid_range out = {
202 		from_kuid_munged(current_user_ns(), range->start),
203 		from_kuid_munged(current_user_ns(), range->end)
204 	};
205 
206 	return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
207 }
208 
209 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
210 			  struct flowi *fl, int flags,
211 			  struct fib_lookup_arg *arg)
212 {
213 	int ret = 0;
214 
215 	if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
216 		goto out;
217 
218 	if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
219 		goto out;
220 
221 	if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
222 		goto out;
223 
224 	if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
225 		goto out;
226 
227 	if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
228 		goto out;
229 
230 	if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
231 	    uid_gt(fl->flowi_uid, rule->uid_range.end))
232 		goto out;
233 
234 	ret = ops->match(rule, fl, flags);
235 out:
236 	return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
237 }
238 
239 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
240 		     int flags, struct fib_lookup_arg *arg)
241 {
242 	struct fib_rule *rule;
243 	int err;
244 
245 	rcu_read_lock();
246 
247 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
248 jumped:
249 		if (!fib_rule_match(rule, ops, fl, flags, arg))
250 			continue;
251 
252 		if (rule->action == FR_ACT_GOTO) {
253 			struct fib_rule *target;
254 
255 			target = rcu_dereference(rule->ctarget);
256 			if (target == NULL) {
257 				continue;
258 			} else {
259 				rule = target;
260 				goto jumped;
261 			}
262 		} else if (rule->action == FR_ACT_NOP)
263 			continue;
264 		else
265 			err = ops->action(rule, fl, flags, arg);
266 
267 		if (!err && ops->suppress && ops->suppress(rule, arg))
268 			continue;
269 
270 		if (err != -EAGAIN) {
271 			if ((arg->flags & FIB_LOOKUP_NOREF) ||
272 			    likely(atomic_inc_not_zero(&rule->refcnt))) {
273 				arg->rule = rule;
274 				goto out;
275 			}
276 			break;
277 		}
278 	}
279 
280 	err = -ESRCH;
281 out:
282 	rcu_read_unlock();
283 
284 	return err;
285 }
286 EXPORT_SYMBOL_GPL(fib_rules_lookup);
287 
288 static int validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
289 			    struct fib_rules_ops *ops)
290 {
291 	int err = -EINVAL;
292 
293 	if (frh->src_len)
294 		if (tb[FRA_SRC] == NULL ||
295 		    frh->src_len > (ops->addr_size * 8) ||
296 		    nla_len(tb[FRA_SRC]) != ops->addr_size)
297 			goto errout;
298 
299 	if (frh->dst_len)
300 		if (tb[FRA_DST] == NULL ||
301 		    frh->dst_len > (ops->addr_size * 8) ||
302 		    nla_len(tb[FRA_DST]) != ops->addr_size)
303 			goto errout;
304 
305 	err = 0;
306 errout:
307 	return err;
308 }
309 
310 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
311 		       struct nlattr **tb, struct fib_rule *rule)
312 {
313 	struct fib_rule *r;
314 
315 	list_for_each_entry(r, &ops->rules_list, list) {
316 		if (r->action != rule->action)
317 			continue;
318 
319 		if (r->table != rule->table)
320 			continue;
321 
322 		if (r->pref != rule->pref)
323 			continue;
324 
325 		if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
326 			continue;
327 
328 		if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
329 			continue;
330 
331 		if (r->mark != rule->mark)
332 			continue;
333 
334 		if (r->mark_mask != rule->mark_mask)
335 			continue;
336 
337 		if (r->tun_id != rule->tun_id)
338 			continue;
339 
340 		if (r->fr_net != rule->fr_net)
341 			continue;
342 
343 		if (r->l3mdev != rule->l3mdev)
344 			continue;
345 
346 		if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
347 		    !uid_eq(r->uid_range.end, rule->uid_range.end))
348 			continue;
349 
350 		if (!ops->compare(r, frh, tb))
351 			continue;
352 		return 1;
353 	}
354 	return 0;
355 }
356 
357 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh)
358 {
359 	struct net *net = sock_net(skb->sk);
360 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
361 	struct fib_rules_ops *ops = NULL;
362 	struct fib_rule *rule, *r, *last = NULL;
363 	struct nlattr *tb[FRA_MAX+1];
364 	int err = -EINVAL, unresolved = 0;
365 
366 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
367 		goto errout;
368 
369 	ops = lookup_rules_ops(net, frh->family);
370 	if (ops == NULL) {
371 		err = -EAFNOSUPPORT;
372 		goto errout;
373 	}
374 
375 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
376 	if (err < 0)
377 		goto errout;
378 
379 	err = validate_rulemsg(frh, tb, ops);
380 	if (err < 0)
381 		goto errout;
382 
383 	rule = kzalloc(ops->rule_size, GFP_KERNEL);
384 	if (rule == NULL) {
385 		err = -ENOMEM;
386 		goto errout;
387 	}
388 	rule->fr_net = net;
389 
390 	rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
391 	                              : fib_default_rule_pref(ops);
392 
393 	if (tb[FRA_IIFNAME]) {
394 		struct net_device *dev;
395 
396 		rule->iifindex = -1;
397 		nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
398 		dev = __dev_get_by_name(net, rule->iifname);
399 		if (dev)
400 			rule->iifindex = dev->ifindex;
401 	}
402 
403 	if (tb[FRA_OIFNAME]) {
404 		struct net_device *dev;
405 
406 		rule->oifindex = -1;
407 		nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
408 		dev = __dev_get_by_name(net, rule->oifname);
409 		if (dev)
410 			rule->oifindex = dev->ifindex;
411 	}
412 
413 	if (tb[FRA_FWMARK]) {
414 		rule->mark = nla_get_u32(tb[FRA_FWMARK]);
415 		if (rule->mark)
416 			/* compatibility: if the mark value is non-zero all bits
417 			 * are compared unless a mask is explicitly specified.
418 			 */
419 			rule->mark_mask = 0xFFFFFFFF;
420 	}
421 
422 	if (tb[FRA_FWMASK])
423 		rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
424 
425 	if (tb[FRA_TUN_ID])
426 		rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
427 
428 	if (tb[FRA_L3MDEV]) {
429 #ifdef CONFIG_NET_L3_MASTER_DEV
430 		rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
431 		if (rule->l3mdev != 1)
432 #endif
433 			goto errout_free;
434 	}
435 
436 	rule->action = frh->action;
437 	rule->flags = frh->flags;
438 	rule->table = frh_get_table(frh, tb);
439 	if (tb[FRA_SUPPRESS_PREFIXLEN])
440 		rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
441 	else
442 		rule->suppress_prefixlen = -1;
443 
444 	if (tb[FRA_SUPPRESS_IFGROUP])
445 		rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
446 	else
447 		rule->suppress_ifgroup = -1;
448 
449 	err = -EINVAL;
450 	if (tb[FRA_GOTO]) {
451 		if (rule->action != FR_ACT_GOTO)
452 			goto errout_free;
453 
454 		rule->target = nla_get_u32(tb[FRA_GOTO]);
455 		/* Backward jumps are prohibited to avoid endless loops */
456 		if (rule->target <= rule->pref)
457 			goto errout_free;
458 
459 		list_for_each_entry(r, &ops->rules_list, list) {
460 			if (r->pref == rule->target) {
461 				RCU_INIT_POINTER(rule->ctarget, r);
462 				break;
463 			}
464 		}
465 
466 		if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
467 			unresolved = 1;
468 	} else if (rule->action == FR_ACT_GOTO)
469 		goto errout_free;
470 
471 	if (rule->l3mdev && rule->table)
472 		goto errout_free;
473 
474 	if (tb[FRA_UID_RANGE]) {
475 		if (current_user_ns() != net->user_ns) {
476 			err = -EPERM;
477 			goto errout_free;
478 		}
479 
480 		rule->uid_range = nla_get_kuid_range(tb);
481 
482 		if (!uid_range_set(&rule->uid_range) ||
483 		    !uid_lte(rule->uid_range.start, rule->uid_range.end))
484 			goto errout_free;
485 	} else {
486 		rule->uid_range = fib_kuid_range_unset;
487 	}
488 
489 	if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
490 	    rule_exists(ops, frh, tb, rule)) {
491 		err = -EEXIST;
492 		goto errout_free;
493 	}
494 
495 	err = ops->configure(rule, skb, frh, tb);
496 	if (err < 0)
497 		goto errout_free;
498 
499 	list_for_each_entry(r, &ops->rules_list, list) {
500 		if (r->pref > rule->pref)
501 			break;
502 		last = r;
503 	}
504 
505 	fib_rule_get(rule);
506 
507 	if (last)
508 		list_add_rcu(&rule->list, &last->list);
509 	else
510 		list_add_rcu(&rule->list, &ops->rules_list);
511 
512 	if (ops->unresolved_rules) {
513 		/*
514 		 * There are unresolved goto rules in the list, check if
515 		 * any of them are pointing to this new rule.
516 		 */
517 		list_for_each_entry(r, &ops->rules_list, list) {
518 			if (r->action == FR_ACT_GOTO &&
519 			    r->target == rule->pref &&
520 			    rtnl_dereference(r->ctarget) == NULL) {
521 				rcu_assign_pointer(r->ctarget, rule);
522 				if (--ops->unresolved_rules == 0)
523 					break;
524 			}
525 		}
526 	}
527 
528 	if (rule->action == FR_ACT_GOTO)
529 		ops->nr_goto_rules++;
530 
531 	if (unresolved)
532 		ops->unresolved_rules++;
533 
534 	if (rule->tun_id)
535 		ip_tunnel_need_metadata();
536 
537 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
538 	flush_route_cache(ops);
539 	rules_ops_put(ops);
540 	return 0;
541 
542 errout_free:
543 	kfree(rule);
544 errout:
545 	rules_ops_put(ops);
546 	return err;
547 }
548 EXPORT_SYMBOL_GPL(fib_nl_newrule);
549 
550 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh)
551 {
552 	struct net *net = sock_net(skb->sk);
553 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
554 	struct fib_rules_ops *ops = NULL;
555 	struct fib_rule *rule, *tmp;
556 	struct nlattr *tb[FRA_MAX+1];
557 	struct fib_kuid_range range;
558 	int err = -EINVAL;
559 
560 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
561 		goto errout;
562 
563 	ops = lookup_rules_ops(net, frh->family);
564 	if (ops == NULL) {
565 		err = -EAFNOSUPPORT;
566 		goto errout;
567 	}
568 
569 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
570 	if (err < 0)
571 		goto errout;
572 
573 	err = validate_rulemsg(frh, tb, ops);
574 	if (err < 0)
575 		goto errout;
576 
577 	if (tb[FRA_UID_RANGE]) {
578 		range = nla_get_kuid_range(tb);
579 		if (!uid_range_set(&range))
580 			goto errout;
581 	} else {
582 		range = fib_kuid_range_unset;
583 	}
584 
585 	list_for_each_entry(rule, &ops->rules_list, list) {
586 		if (frh->action && (frh->action != rule->action))
587 			continue;
588 
589 		if (frh_get_table(frh, tb) &&
590 		    (frh_get_table(frh, tb) != rule->table))
591 			continue;
592 
593 		if (tb[FRA_PRIORITY] &&
594 		    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
595 			continue;
596 
597 		if (tb[FRA_IIFNAME] &&
598 		    nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
599 			continue;
600 
601 		if (tb[FRA_OIFNAME] &&
602 		    nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
603 			continue;
604 
605 		if (tb[FRA_FWMARK] &&
606 		    (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
607 			continue;
608 
609 		if (tb[FRA_FWMASK] &&
610 		    (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
611 			continue;
612 
613 		if (tb[FRA_TUN_ID] &&
614 		    (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
615 			continue;
616 
617 		if (tb[FRA_L3MDEV] &&
618 		    (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
619 			continue;
620 
621 		if (uid_range_set(&range) &&
622 		    (!uid_eq(rule->uid_range.start, range.start) ||
623 		     !uid_eq(rule->uid_range.end, range.end)))
624 			continue;
625 
626 		if (!ops->compare(rule, frh, tb))
627 			continue;
628 
629 		if (rule->flags & FIB_RULE_PERMANENT) {
630 			err = -EPERM;
631 			goto errout;
632 		}
633 
634 		if (ops->delete) {
635 			err = ops->delete(rule);
636 			if (err)
637 				goto errout;
638 		}
639 
640 		if (rule->tun_id)
641 			ip_tunnel_unneed_metadata();
642 
643 		list_del_rcu(&rule->list);
644 
645 		if (rule->action == FR_ACT_GOTO) {
646 			ops->nr_goto_rules--;
647 			if (rtnl_dereference(rule->ctarget) == NULL)
648 				ops->unresolved_rules--;
649 		}
650 
651 		/*
652 		 * Check if this rule is a target to any of them. If so,
653 		 * disable them. As this operation is eventually very
654 		 * expensive, it is only performed if goto rules have
655 		 * actually been added.
656 		 */
657 		if (ops->nr_goto_rules > 0) {
658 			list_for_each_entry(tmp, &ops->rules_list, list) {
659 				if (rtnl_dereference(tmp->ctarget) == rule) {
660 					RCU_INIT_POINTER(tmp->ctarget, NULL);
661 					ops->unresolved_rules++;
662 				}
663 			}
664 		}
665 
666 		notify_rule_change(RTM_DELRULE, rule, ops, nlh,
667 				   NETLINK_CB(skb).portid);
668 		fib_rule_put(rule);
669 		flush_route_cache(ops);
670 		rules_ops_put(ops);
671 		return 0;
672 	}
673 
674 	err = -ENOENT;
675 errout:
676 	rules_ops_put(ops);
677 	return err;
678 }
679 EXPORT_SYMBOL_GPL(fib_nl_delrule);
680 
681 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
682 					 struct fib_rule *rule)
683 {
684 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
685 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
686 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
687 			 + nla_total_size(4) /* FRA_PRIORITY */
688 			 + nla_total_size(4) /* FRA_TABLE */
689 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
690 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
691 			 + nla_total_size(4) /* FRA_FWMARK */
692 			 + nla_total_size(4) /* FRA_FWMASK */
693 			 + nla_total_size_64bit(8) /* FRA_TUN_ID */
694 			 + nla_total_size(sizeof(struct fib_kuid_range));
695 
696 	if (ops->nlmsg_payload)
697 		payload += ops->nlmsg_payload(rule);
698 
699 	return payload;
700 }
701 
702 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
703 			    u32 pid, u32 seq, int type, int flags,
704 			    struct fib_rules_ops *ops)
705 {
706 	struct nlmsghdr *nlh;
707 	struct fib_rule_hdr *frh;
708 
709 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
710 	if (nlh == NULL)
711 		return -EMSGSIZE;
712 
713 	frh = nlmsg_data(nlh);
714 	frh->family = ops->family;
715 	frh->table = rule->table;
716 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
717 		goto nla_put_failure;
718 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
719 		goto nla_put_failure;
720 	frh->res1 = 0;
721 	frh->res2 = 0;
722 	frh->action = rule->action;
723 	frh->flags = rule->flags;
724 
725 	if (rule->action == FR_ACT_GOTO &&
726 	    rcu_access_pointer(rule->ctarget) == NULL)
727 		frh->flags |= FIB_RULE_UNRESOLVED;
728 
729 	if (rule->iifname[0]) {
730 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
731 			goto nla_put_failure;
732 		if (rule->iifindex == -1)
733 			frh->flags |= FIB_RULE_IIF_DETACHED;
734 	}
735 
736 	if (rule->oifname[0]) {
737 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
738 			goto nla_put_failure;
739 		if (rule->oifindex == -1)
740 			frh->flags |= FIB_RULE_OIF_DETACHED;
741 	}
742 
743 	if ((rule->pref &&
744 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
745 	    (rule->mark &&
746 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
747 	    ((rule->mark_mask || rule->mark) &&
748 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
749 	    (rule->target &&
750 	     nla_put_u32(skb, FRA_GOTO, rule->target)) ||
751 	    (rule->tun_id &&
752 	     nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
753 	    (rule->l3mdev &&
754 	     nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
755 	    (uid_range_set(&rule->uid_range) &&
756 	     nla_put_uid_range(skb, &rule->uid_range)))
757 		goto nla_put_failure;
758 
759 	if (rule->suppress_ifgroup != -1) {
760 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
761 			goto nla_put_failure;
762 	}
763 
764 	if (ops->fill(rule, skb, frh) < 0)
765 		goto nla_put_failure;
766 
767 	nlmsg_end(skb, nlh);
768 	return 0;
769 
770 nla_put_failure:
771 	nlmsg_cancel(skb, nlh);
772 	return -EMSGSIZE;
773 }
774 
775 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
776 		      struct fib_rules_ops *ops)
777 {
778 	int idx = 0;
779 	struct fib_rule *rule;
780 	int err = 0;
781 
782 	rcu_read_lock();
783 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
784 		if (idx < cb->args[1])
785 			goto skip;
786 
787 		err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
788 				       cb->nlh->nlmsg_seq, RTM_NEWRULE,
789 				       NLM_F_MULTI, ops);
790 		if (err)
791 			break;
792 skip:
793 		idx++;
794 	}
795 	rcu_read_unlock();
796 	cb->args[1] = idx;
797 	rules_ops_put(ops);
798 
799 	return err;
800 }
801 
802 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
803 {
804 	struct net *net = sock_net(skb->sk);
805 	struct fib_rules_ops *ops;
806 	int idx = 0, family;
807 
808 	family = rtnl_msg_family(cb->nlh);
809 	if (family != AF_UNSPEC) {
810 		/* Protocol specific dump request */
811 		ops = lookup_rules_ops(net, family);
812 		if (ops == NULL)
813 			return -EAFNOSUPPORT;
814 
815 		dump_rules(skb, cb, ops);
816 
817 		return skb->len;
818 	}
819 
820 	rcu_read_lock();
821 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
822 		if (idx < cb->args[0] || !try_module_get(ops->owner))
823 			goto skip;
824 
825 		if (dump_rules(skb, cb, ops) < 0)
826 			break;
827 
828 		cb->args[1] = 0;
829 skip:
830 		idx++;
831 	}
832 	rcu_read_unlock();
833 	cb->args[0] = idx;
834 
835 	return skb->len;
836 }
837 
838 static void notify_rule_change(int event, struct fib_rule *rule,
839 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
840 			       u32 pid)
841 {
842 	struct net *net;
843 	struct sk_buff *skb;
844 	int err = -ENOBUFS;
845 
846 	net = ops->fro_net;
847 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
848 	if (skb == NULL)
849 		goto errout;
850 
851 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
852 	if (err < 0) {
853 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
854 		WARN_ON(err == -EMSGSIZE);
855 		kfree_skb(skb);
856 		goto errout;
857 	}
858 
859 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
860 	return;
861 errout:
862 	if (err < 0)
863 		rtnl_set_sk_err(net, ops->nlgroup, err);
864 }
865 
866 static void attach_rules(struct list_head *rules, struct net_device *dev)
867 {
868 	struct fib_rule *rule;
869 
870 	list_for_each_entry(rule, rules, list) {
871 		if (rule->iifindex == -1 &&
872 		    strcmp(dev->name, rule->iifname) == 0)
873 			rule->iifindex = dev->ifindex;
874 		if (rule->oifindex == -1 &&
875 		    strcmp(dev->name, rule->oifname) == 0)
876 			rule->oifindex = dev->ifindex;
877 	}
878 }
879 
880 static void detach_rules(struct list_head *rules, struct net_device *dev)
881 {
882 	struct fib_rule *rule;
883 
884 	list_for_each_entry(rule, rules, list) {
885 		if (rule->iifindex == dev->ifindex)
886 			rule->iifindex = -1;
887 		if (rule->oifindex == dev->ifindex)
888 			rule->oifindex = -1;
889 	}
890 }
891 
892 
893 static int fib_rules_event(struct notifier_block *this, unsigned long event,
894 			   void *ptr)
895 {
896 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
897 	struct net *net = dev_net(dev);
898 	struct fib_rules_ops *ops;
899 
900 	ASSERT_RTNL();
901 
902 	switch (event) {
903 	case NETDEV_REGISTER:
904 		list_for_each_entry(ops, &net->rules_ops, list)
905 			attach_rules(&ops->rules_list, dev);
906 		break;
907 
908 	case NETDEV_CHANGENAME:
909 		list_for_each_entry(ops, &net->rules_ops, list) {
910 			detach_rules(&ops->rules_list, dev);
911 			attach_rules(&ops->rules_list, dev);
912 		}
913 		break;
914 
915 	case NETDEV_UNREGISTER:
916 		list_for_each_entry(ops, &net->rules_ops, list)
917 			detach_rules(&ops->rules_list, dev);
918 		break;
919 	}
920 
921 	return NOTIFY_DONE;
922 }
923 
924 static struct notifier_block fib_rules_notifier = {
925 	.notifier_call = fib_rules_event,
926 };
927 
928 static int __net_init fib_rules_net_init(struct net *net)
929 {
930 	INIT_LIST_HEAD(&net->rules_ops);
931 	spin_lock_init(&net->rules_mod_lock);
932 	return 0;
933 }
934 
935 static struct pernet_operations fib_rules_net_ops = {
936 	.init = fib_rules_net_init,
937 };
938 
939 static int __init fib_rules_init(void)
940 {
941 	int err;
942 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
943 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
944 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
945 
946 	err = register_pernet_subsys(&fib_rules_net_ops);
947 	if (err < 0)
948 		goto fail;
949 
950 	err = register_netdevice_notifier(&fib_rules_notifier);
951 	if (err < 0)
952 		goto fail_unregister;
953 
954 	return 0;
955 
956 fail_unregister:
957 	unregister_pernet_subsys(&fib_rules_net_ops);
958 fail:
959 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
960 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
961 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
962 	return err;
963 }
964 
965 subsys_initcall(fib_rules_init);
966