xref: /openbmc/linux/net/core/fib_rules.c (revision 4f139972b489f8bc2c821aa25ac65018d92af3f7)
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 	atomic_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(atomic_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 validate_rulemsg(struct fib_rule_hdr *frh, struct nlattr **tb,
303 			    struct fib_rules_ops *ops)
304 {
305 	int err = -EINVAL;
306 
307 	if (frh->src_len)
308 		if (tb[FRA_SRC] == NULL ||
309 		    frh->src_len > (ops->addr_size * 8) ||
310 		    nla_len(tb[FRA_SRC]) != ops->addr_size)
311 			goto errout;
312 
313 	if (frh->dst_len)
314 		if (tb[FRA_DST] == NULL ||
315 		    frh->dst_len > (ops->addr_size * 8) ||
316 		    nla_len(tb[FRA_DST]) != ops->addr_size)
317 			goto errout;
318 
319 	err = 0;
320 errout:
321 	return err;
322 }
323 
324 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
325 		       struct nlattr **tb, struct fib_rule *rule)
326 {
327 	struct fib_rule *r;
328 
329 	list_for_each_entry(r, &ops->rules_list, list) {
330 		if (r->action != rule->action)
331 			continue;
332 
333 		if (r->table != rule->table)
334 			continue;
335 
336 		if (r->pref != rule->pref)
337 			continue;
338 
339 		if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
340 			continue;
341 
342 		if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
343 			continue;
344 
345 		if (r->mark != rule->mark)
346 			continue;
347 
348 		if (r->mark_mask != rule->mark_mask)
349 			continue;
350 
351 		if (r->tun_id != rule->tun_id)
352 			continue;
353 
354 		if (r->fr_net != rule->fr_net)
355 			continue;
356 
357 		if (r->l3mdev != rule->l3mdev)
358 			continue;
359 
360 		if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
361 		    !uid_eq(r->uid_range.end, rule->uid_range.end))
362 			continue;
363 
364 		if (!ops->compare(r, frh, tb))
365 			continue;
366 		return 1;
367 	}
368 	return 0;
369 }
370 
371 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh)
372 {
373 	struct net *net = sock_net(skb->sk);
374 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
375 	struct fib_rules_ops *ops = NULL;
376 	struct fib_rule *rule, *r, *last = NULL;
377 	struct nlattr *tb[FRA_MAX+1];
378 	int err = -EINVAL, unresolved = 0;
379 
380 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
381 		goto errout;
382 
383 	ops = lookup_rules_ops(net, frh->family);
384 	if (ops == NULL) {
385 		err = -EAFNOSUPPORT;
386 		goto errout;
387 	}
388 
389 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
390 	if (err < 0)
391 		goto errout;
392 
393 	err = validate_rulemsg(frh, tb, ops);
394 	if (err < 0)
395 		goto errout;
396 
397 	rule = kzalloc(ops->rule_size, GFP_KERNEL);
398 	if (rule == NULL) {
399 		err = -ENOMEM;
400 		goto errout;
401 	}
402 	rule->fr_net = net;
403 
404 	rule->pref = tb[FRA_PRIORITY] ? nla_get_u32(tb[FRA_PRIORITY])
405 	                              : fib_default_rule_pref(ops);
406 
407 	if (tb[FRA_IIFNAME]) {
408 		struct net_device *dev;
409 
410 		rule->iifindex = -1;
411 		nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
412 		dev = __dev_get_by_name(net, rule->iifname);
413 		if (dev)
414 			rule->iifindex = dev->ifindex;
415 	}
416 
417 	if (tb[FRA_OIFNAME]) {
418 		struct net_device *dev;
419 
420 		rule->oifindex = -1;
421 		nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
422 		dev = __dev_get_by_name(net, rule->oifname);
423 		if (dev)
424 			rule->oifindex = dev->ifindex;
425 	}
426 
427 	if (tb[FRA_FWMARK]) {
428 		rule->mark = nla_get_u32(tb[FRA_FWMARK]);
429 		if (rule->mark)
430 			/* compatibility: if the mark value is non-zero all bits
431 			 * are compared unless a mask is explicitly specified.
432 			 */
433 			rule->mark_mask = 0xFFFFFFFF;
434 	}
435 
436 	if (tb[FRA_FWMASK])
437 		rule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
438 
439 	if (tb[FRA_TUN_ID])
440 		rule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
441 
442 	if (tb[FRA_L3MDEV]) {
443 #ifdef CONFIG_NET_L3_MASTER_DEV
444 		rule->l3mdev = nla_get_u8(tb[FRA_L3MDEV]);
445 		if (rule->l3mdev != 1)
446 #endif
447 			goto errout_free;
448 	}
449 
450 	rule->action = frh->action;
451 	rule->flags = frh->flags;
452 	rule->table = frh_get_table(frh, tb);
453 	if (tb[FRA_SUPPRESS_PREFIXLEN])
454 		rule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
455 	else
456 		rule->suppress_prefixlen = -1;
457 
458 	if (tb[FRA_SUPPRESS_IFGROUP])
459 		rule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
460 	else
461 		rule->suppress_ifgroup = -1;
462 
463 	err = -EINVAL;
464 	if (tb[FRA_GOTO]) {
465 		if (rule->action != FR_ACT_GOTO)
466 			goto errout_free;
467 
468 		rule->target = nla_get_u32(tb[FRA_GOTO]);
469 		/* Backward jumps are prohibited to avoid endless loops */
470 		if (rule->target <= rule->pref)
471 			goto errout_free;
472 
473 		list_for_each_entry(r, &ops->rules_list, list) {
474 			if (r->pref == rule->target) {
475 				RCU_INIT_POINTER(rule->ctarget, r);
476 				break;
477 			}
478 		}
479 
480 		if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
481 			unresolved = 1;
482 	} else if (rule->action == FR_ACT_GOTO)
483 		goto errout_free;
484 
485 	if (rule->l3mdev && rule->table)
486 		goto errout_free;
487 
488 	if (tb[FRA_UID_RANGE]) {
489 		if (current_user_ns() != net->user_ns) {
490 			err = -EPERM;
491 			goto errout_free;
492 		}
493 
494 		rule->uid_range = nla_get_kuid_range(tb);
495 
496 		if (!uid_range_set(&rule->uid_range) ||
497 		    !uid_lte(rule->uid_range.start, rule->uid_range.end))
498 			goto errout_free;
499 	} else {
500 		rule->uid_range = fib_kuid_range_unset;
501 	}
502 
503 	if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
504 	    rule_exists(ops, frh, tb, rule)) {
505 		err = -EEXIST;
506 		goto errout_free;
507 	}
508 
509 	err = ops->configure(rule, skb, frh, tb);
510 	if (err < 0)
511 		goto errout_free;
512 
513 	list_for_each_entry(r, &ops->rules_list, list) {
514 		if (r->pref > rule->pref)
515 			break;
516 		last = r;
517 	}
518 
519 	fib_rule_get(rule);
520 
521 	if (last)
522 		list_add_rcu(&rule->list, &last->list);
523 	else
524 		list_add_rcu(&rule->list, &ops->rules_list);
525 
526 	if (ops->unresolved_rules) {
527 		/*
528 		 * There are unresolved goto rules in the list, check if
529 		 * any of them are pointing to this new rule.
530 		 */
531 		list_for_each_entry(r, &ops->rules_list, list) {
532 			if (r->action == FR_ACT_GOTO &&
533 			    r->target == rule->pref &&
534 			    rtnl_dereference(r->ctarget) == NULL) {
535 				rcu_assign_pointer(r->ctarget, rule);
536 				if (--ops->unresolved_rules == 0)
537 					break;
538 			}
539 		}
540 	}
541 
542 	if (rule->action == FR_ACT_GOTO)
543 		ops->nr_goto_rules++;
544 
545 	if (unresolved)
546 		ops->unresolved_rules++;
547 
548 	if (rule->tun_id)
549 		ip_tunnel_need_metadata();
550 
551 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
552 	flush_route_cache(ops);
553 	rules_ops_put(ops);
554 	return 0;
555 
556 errout_free:
557 	kfree(rule);
558 errout:
559 	rules_ops_put(ops);
560 	return err;
561 }
562 EXPORT_SYMBOL_GPL(fib_nl_newrule);
563 
564 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh)
565 {
566 	struct net *net = sock_net(skb->sk);
567 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
568 	struct fib_rules_ops *ops = NULL;
569 	struct fib_rule *rule, *tmp;
570 	struct nlattr *tb[FRA_MAX+1];
571 	struct fib_kuid_range range;
572 	int err = -EINVAL;
573 
574 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
575 		goto errout;
576 
577 	ops = lookup_rules_ops(net, frh->family);
578 	if (ops == NULL) {
579 		err = -EAFNOSUPPORT;
580 		goto errout;
581 	}
582 
583 	err = nlmsg_parse(nlh, sizeof(*frh), tb, FRA_MAX, ops->policy);
584 	if (err < 0)
585 		goto errout;
586 
587 	err = validate_rulemsg(frh, tb, ops);
588 	if (err < 0)
589 		goto errout;
590 
591 	if (tb[FRA_UID_RANGE]) {
592 		range = nla_get_kuid_range(tb);
593 		if (!uid_range_set(&range))
594 			goto errout;
595 	} else {
596 		range = fib_kuid_range_unset;
597 	}
598 
599 	list_for_each_entry(rule, &ops->rules_list, list) {
600 		if (frh->action && (frh->action != rule->action))
601 			continue;
602 
603 		if (frh_get_table(frh, tb) &&
604 		    (frh_get_table(frh, tb) != rule->table))
605 			continue;
606 
607 		if (tb[FRA_PRIORITY] &&
608 		    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
609 			continue;
610 
611 		if (tb[FRA_IIFNAME] &&
612 		    nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
613 			continue;
614 
615 		if (tb[FRA_OIFNAME] &&
616 		    nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
617 			continue;
618 
619 		if (tb[FRA_FWMARK] &&
620 		    (rule->mark != nla_get_u32(tb[FRA_FWMARK])))
621 			continue;
622 
623 		if (tb[FRA_FWMASK] &&
624 		    (rule->mark_mask != nla_get_u32(tb[FRA_FWMASK])))
625 			continue;
626 
627 		if (tb[FRA_TUN_ID] &&
628 		    (rule->tun_id != nla_get_be64(tb[FRA_TUN_ID])))
629 			continue;
630 
631 		if (tb[FRA_L3MDEV] &&
632 		    (rule->l3mdev != nla_get_u8(tb[FRA_L3MDEV])))
633 			continue;
634 
635 		if (uid_range_set(&range) &&
636 		    (!uid_eq(rule->uid_range.start, range.start) ||
637 		     !uid_eq(rule->uid_range.end, range.end)))
638 			continue;
639 
640 		if (!ops->compare(rule, frh, tb))
641 			continue;
642 
643 		if (rule->flags & FIB_RULE_PERMANENT) {
644 			err = -EPERM;
645 			goto errout;
646 		}
647 
648 		if (ops->delete) {
649 			err = ops->delete(rule);
650 			if (err)
651 				goto errout;
652 		}
653 
654 		if (rule->tun_id)
655 			ip_tunnel_unneed_metadata();
656 
657 		list_del_rcu(&rule->list);
658 
659 		if (rule->action == FR_ACT_GOTO) {
660 			ops->nr_goto_rules--;
661 			if (rtnl_dereference(rule->ctarget) == NULL)
662 				ops->unresolved_rules--;
663 		}
664 
665 		/*
666 		 * Check if this rule is a target to any of them. If so,
667 		 * disable them. As this operation is eventually very
668 		 * expensive, it is only performed if goto rules have
669 		 * actually been added.
670 		 */
671 		if (ops->nr_goto_rules > 0) {
672 			list_for_each_entry(tmp, &ops->rules_list, list) {
673 				if (rtnl_dereference(tmp->ctarget) == rule) {
674 					RCU_INIT_POINTER(tmp->ctarget, NULL);
675 					ops->unresolved_rules++;
676 				}
677 			}
678 		}
679 
680 		notify_rule_change(RTM_DELRULE, rule, ops, nlh,
681 				   NETLINK_CB(skb).portid);
682 		fib_rule_put(rule);
683 		flush_route_cache(ops);
684 		rules_ops_put(ops);
685 		return 0;
686 	}
687 
688 	err = -ENOENT;
689 errout:
690 	rules_ops_put(ops);
691 	return err;
692 }
693 EXPORT_SYMBOL_GPL(fib_nl_delrule);
694 
695 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
696 					 struct fib_rule *rule)
697 {
698 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
699 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
700 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
701 			 + nla_total_size(4) /* FRA_PRIORITY */
702 			 + nla_total_size(4) /* FRA_TABLE */
703 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
704 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
705 			 + nla_total_size(4) /* FRA_FWMARK */
706 			 + nla_total_size(4) /* FRA_FWMASK */
707 			 + nla_total_size_64bit(8) /* FRA_TUN_ID */
708 			 + nla_total_size(sizeof(struct fib_kuid_range));
709 
710 	if (ops->nlmsg_payload)
711 		payload += ops->nlmsg_payload(rule);
712 
713 	return payload;
714 }
715 
716 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
717 			    u32 pid, u32 seq, int type, int flags,
718 			    struct fib_rules_ops *ops)
719 {
720 	struct nlmsghdr *nlh;
721 	struct fib_rule_hdr *frh;
722 
723 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
724 	if (nlh == NULL)
725 		return -EMSGSIZE;
726 
727 	frh = nlmsg_data(nlh);
728 	frh->family = ops->family;
729 	frh->table = rule->table;
730 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
731 		goto nla_put_failure;
732 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
733 		goto nla_put_failure;
734 	frh->res1 = 0;
735 	frh->res2 = 0;
736 	frh->action = rule->action;
737 	frh->flags = rule->flags;
738 
739 	if (rule->action == FR_ACT_GOTO &&
740 	    rcu_access_pointer(rule->ctarget) == NULL)
741 		frh->flags |= FIB_RULE_UNRESOLVED;
742 
743 	if (rule->iifname[0]) {
744 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
745 			goto nla_put_failure;
746 		if (rule->iifindex == -1)
747 			frh->flags |= FIB_RULE_IIF_DETACHED;
748 	}
749 
750 	if (rule->oifname[0]) {
751 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
752 			goto nla_put_failure;
753 		if (rule->oifindex == -1)
754 			frh->flags |= FIB_RULE_OIF_DETACHED;
755 	}
756 
757 	if ((rule->pref &&
758 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
759 	    (rule->mark &&
760 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
761 	    ((rule->mark_mask || rule->mark) &&
762 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
763 	    (rule->target &&
764 	     nla_put_u32(skb, FRA_GOTO, rule->target)) ||
765 	    (rule->tun_id &&
766 	     nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
767 	    (rule->l3mdev &&
768 	     nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
769 	    (uid_range_set(&rule->uid_range) &&
770 	     nla_put_uid_range(skb, &rule->uid_range)))
771 		goto nla_put_failure;
772 
773 	if (rule->suppress_ifgroup != -1) {
774 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
775 			goto nla_put_failure;
776 	}
777 
778 	if (ops->fill(rule, skb, frh) < 0)
779 		goto nla_put_failure;
780 
781 	nlmsg_end(skb, nlh);
782 	return 0;
783 
784 nla_put_failure:
785 	nlmsg_cancel(skb, nlh);
786 	return -EMSGSIZE;
787 }
788 
789 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
790 		      struct fib_rules_ops *ops)
791 {
792 	int idx = 0;
793 	struct fib_rule *rule;
794 	int err = 0;
795 
796 	rcu_read_lock();
797 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
798 		if (idx < cb->args[1])
799 			goto skip;
800 
801 		err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
802 				       cb->nlh->nlmsg_seq, RTM_NEWRULE,
803 				       NLM_F_MULTI, ops);
804 		if (err)
805 			break;
806 skip:
807 		idx++;
808 	}
809 	rcu_read_unlock();
810 	cb->args[1] = idx;
811 	rules_ops_put(ops);
812 
813 	return err;
814 }
815 
816 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
817 {
818 	struct net *net = sock_net(skb->sk);
819 	struct fib_rules_ops *ops;
820 	int idx = 0, family;
821 
822 	family = rtnl_msg_family(cb->nlh);
823 	if (family != AF_UNSPEC) {
824 		/* Protocol specific dump request */
825 		ops = lookup_rules_ops(net, family);
826 		if (ops == NULL)
827 			return -EAFNOSUPPORT;
828 
829 		dump_rules(skb, cb, ops);
830 
831 		return skb->len;
832 	}
833 
834 	rcu_read_lock();
835 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
836 		if (idx < cb->args[0] || !try_module_get(ops->owner))
837 			goto skip;
838 
839 		if (dump_rules(skb, cb, ops) < 0)
840 			break;
841 
842 		cb->args[1] = 0;
843 skip:
844 		idx++;
845 	}
846 	rcu_read_unlock();
847 	cb->args[0] = idx;
848 
849 	return skb->len;
850 }
851 
852 static void notify_rule_change(int event, struct fib_rule *rule,
853 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
854 			       u32 pid)
855 {
856 	struct net *net;
857 	struct sk_buff *skb;
858 	int err = -ENOBUFS;
859 
860 	net = ops->fro_net;
861 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
862 	if (skb == NULL)
863 		goto errout;
864 
865 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
866 	if (err < 0) {
867 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
868 		WARN_ON(err == -EMSGSIZE);
869 		kfree_skb(skb);
870 		goto errout;
871 	}
872 
873 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
874 	return;
875 errout:
876 	if (err < 0)
877 		rtnl_set_sk_err(net, ops->nlgroup, err);
878 }
879 
880 static void attach_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 == -1 &&
886 		    strcmp(dev->name, rule->iifname) == 0)
887 			rule->iifindex = dev->ifindex;
888 		if (rule->oifindex == -1 &&
889 		    strcmp(dev->name, rule->oifname) == 0)
890 			rule->oifindex = dev->ifindex;
891 	}
892 }
893 
894 static void detach_rules(struct list_head *rules, struct net_device *dev)
895 {
896 	struct fib_rule *rule;
897 
898 	list_for_each_entry(rule, rules, list) {
899 		if (rule->iifindex == dev->ifindex)
900 			rule->iifindex = -1;
901 		if (rule->oifindex == dev->ifindex)
902 			rule->oifindex = -1;
903 	}
904 }
905 
906 
907 static int fib_rules_event(struct notifier_block *this, unsigned long event,
908 			   void *ptr)
909 {
910 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
911 	struct net *net = dev_net(dev);
912 	struct fib_rules_ops *ops;
913 
914 	ASSERT_RTNL();
915 
916 	switch (event) {
917 	case NETDEV_REGISTER:
918 		list_for_each_entry(ops, &net->rules_ops, list)
919 			attach_rules(&ops->rules_list, dev);
920 		break;
921 
922 	case NETDEV_CHANGENAME:
923 		list_for_each_entry(ops, &net->rules_ops, list) {
924 			detach_rules(&ops->rules_list, dev);
925 			attach_rules(&ops->rules_list, dev);
926 		}
927 		break;
928 
929 	case NETDEV_UNREGISTER:
930 		list_for_each_entry(ops, &net->rules_ops, list)
931 			detach_rules(&ops->rules_list, dev);
932 		break;
933 	}
934 
935 	return NOTIFY_DONE;
936 }
937 
938 static struct notifier_block fib_rules_notifier = {
939 	.notifier_call = fib_rules_event,
940 };
941 
942 static int __net_init fib_rules_net_init(struct net *net)
943 {
944 	INIT_LIST_HEAD(&net->rules_ops);
945 	spin_lock_init(&net->rules_mod_lock);
946 	return 0;
947 }
948 
949 static struct pernet_operations fib_rules_net_ops = {
950 	.init = fib_rules_net_init,
951 };
952 
953 static int __init fib_rules_init(void)
954 {
955 	int err;
956 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, NULL);
957 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, NULL);
958 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, NULL);
959 
960 	err = register_pernet_subsys(&fib_rules_net_ops);
961 	if (err < 0)
962 		goto fail;
963 
964 	err = register_netdevice_notifier(&fib_rules_notifier);
965 	if (err < 0)
966 		goto fail_unregister;
967 
968 	return 0;
969 
970 fail_unregister:
971 	unregister_pernet_subsys(&fib_rules_net_ops);
972 fail:
973 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
974 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
975 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
976 	return err;
977 }
978 
979 subsys_initcall(fib_rules_init);
980