xref: /openbmc/linux/net/core/fib_rules.c (revision d9679d00)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * net/core/fib_rules.c		Generic Routing Rules
4  *
5  * Authors:	Thomas Graf <tgraf@suug.ch>
6  */
7 
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/list.h>
12 #include <linux/module.h>
13 #include <net/net_namespace.h>
14 #include <net/sock.h>
15 #include <net/fib_rules.h>
16 #include <net/ip_tunnels.h>
17 #include <linux/indirect_call_wrapper.h>
18 
19 #if defined(CONFIG_IPV6) && defined(CONFIG_IPV6_MULTIPLE_TABLES)
20 #ifdef CONFIG_IP_MULTIPLE_TABLES
21 #define INDIRECT_CALL_MT(f, f2, f1, ...) \
22 	INDIRECT_CALL_INET(f, f2, f1, __VA_ARGS__)
23 #else
24 #define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
25 #endif
26 #elif defined(CONFIG_IP_MULTIPLE_TABLES)
27 #define INDIRECT_CALL_MT(f, f2, f1, ...) INDIRECT_CALL_1(f, f1, __VA_ARGS__)
28 #else
29 #define INDIRECT_CALL_MT(f, f2, f1, ...) f(__VA_ARGS__)
30 #endif
31 
32 static const struct fib_kuid_range fib_kuid_range_unset = {
33 	KUIDT_INIT(0),
34 	KUIDT_INIT(~0),
35 };
36 
37 bool fib_rule_matchall(const struct fib_rule *rule)
38 {
39 	if (rule->iifindex || rule->oifindex || rule->mark || rule->tun_id ||
40 	    rule->flags)
41 		return false;
42 	if (rule->suppress_ifgroup != -1 || rule->suppress_prefixlen != -1)
43 		return false;
44 	if (!uid_eq(rule->uid_range.start, fib_kuid_range_unset.start) ||
45 	    !uid_eq(rule->uid_range.end, fib_kuid_range_unset.end))
46 		return false;
47 	if (fib_rule_port_range_set(&rule->sport_range))
48 		return false;
49 	if (fib_rule_port_range_set(&rule->dport_range))
50 		return false;
51 	return true;
52 }
53 EXPORT_SYMBOL_GPL(fib_rule_matchall);
54 
55 int fib_default_rule_add(struct fib_rules_ops *ops,
56 			 u32 pref, u32 table, u32 flags)
57 {
58 	struct fib_rule *r;
59 
60 	r = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT);
61 	if (r == NULL)
62 		return -ENOMEM;
63 
64 	refcount_set(&r->refcnt, 1);
65 	r->action = FR_ACT_TO_TBL;
66 	r->pref = pref;
67 	r->table = table;
68 	r->flags = flags;
69 	r->proto = RTPROT_KERNEL;
70 	r->fr_net = ops->fro_net;
71 	r->uid_range = fib_kuid_range_unset;
72 
73 	r->suppress_prefixlen = -1;
74 	r->suppress_ifgroup = -1;
75 
76 	/* The lock is not required here, the list in unreacheable
77 	 * at the moment this function is called */
78 	list_add_tail(&r->list, &ops->rules_list);
79 	return 0;
80 }
81 EXPORT_SYMBOL(fib_default_rule_add);
82 
83 static u32 fib_default_rule_pref(struct fib_rules_ops *ops)
84 {
85 	struct list_head *pos;
86 	struct fib_rule *rule;
87 
88 	if (!list_empty(&ops->rules_list)) {
89 		pos = ops->rules_list.next;
90 		if (pos->next != &ops->rules_list) {
91 			rule = list_entry(pos->next, struct fib_rule, list);
92 			if (rule->pref)
93 				return rule->pref - 1;
94 		}
95 	}
96 
97 	return 0;
98 }
99 
100 static void notify_rule_change(int event, struct fib_rule *rule,
101 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
102 			       u32 pid);
103 
104 static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
105 {
106 	struct fib_rules_ops *ops;
107 
108 	rcu_read_lock();
109 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
110 		if (ops->family == family) {
111 			if (!try_module_get(ops->owner))
112 				ops = NULL;
113 			rcu_read_unlock();
114 			return ops;
115 		}
116 	}
117 	rcu_read_unlock();
118 
119 	return NULL;
120 }
121 
122 static void rules_ops_put(struct fib_rules_ops *ops)
123 {
124 	if (ops)
125 		module_put(ops->owner);
126 }
127 
128 static void flush_route_cache(struct fib_rules_ops *ops)
129 {
130 	if (ops->flush_cache)
131 		ops->flush_cache(ops);
132 }
133 
134 static int __fib_rules_register(struct fib_rules_ops *ops)
135 {
136 	int err = -EEXIST;
137 	struct fib_rules_ops *o;
138 	struct net *net;
139 
140 	net = ops->fro_net;
141 
142 	if (ops->rule_size < sizeof(struct fib_rule))
143 		return -EINVAL;
144 
145 	if (ops->match == NULL || ops->configure == NULL ||
146 	    ops->compare == NULL || ops->fill == NULL ||
147 	    ops->action == NULL)
148 		return -EINVAL;
149 
150 	spin_lock(&net->rules_mod_lock);
151 	list_for_each_entry(o, &net->rules_ops, list)
152 		if (ops->family == o->family)
153 			goto errout;
154 
155 	list_add_tail_rcu(&ops->list, &net->rules_ops);
156 	err = 0;
157 errout:
158 	spin_unlock(&net->rules_mod_lock);
159 
160 	return err;
161 }
162 
163 struct fib_rules_ops *
164 fib_rules_register(const struct fib_rules_ops *tmpl, struct net *net)
165 {
166 	struct fib_rules_ops *ops;
167 	int err;
168 
169 	ops = kmemdup(tmpl, sizeof(*ops), GFP_KERNEL);
170 	if (ops == NULL)
171 		return ERR_PTR(-ENOMEM);
172 
173 	INIT_LIST_HEAD(&ops->rules_list);
174 	ops->fro_net = net;
175 
176 	err = __fib_rules_register(ops);
177 	if (err) {
178 		kfree(ops);
179 		ops = ERR_PTR(err);
180 	}
181 
182 	return ops;
183 }
184 EXPORT_SYMBOL_GPL(fib_rules_register);
185 
186 static void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
187 {
188 	struct fib_rule *rule, *tmp;
189 
190 	list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
191 		list_del_rcu(&rule->list);
192 		if (ops->delete)
193 			ops->delete(rule);
194 		fib_rule_put(rule);
195 	}
196 }
197 
198 void fib_rules_unregister(struct fib_rules_ops *ops)
199 {
200 	struct net *net = ops->fro_net;
201 
202 	spin_lock(&net->rules_mod_lock);
203 	list_del_rcu(&ops->list);
204 	spin_unlock(&net->rules_mod_lock);
205 
206 	fib_rules_cleanup_ops(ops);
207 	kfree_rcu(ops, rcu);
208 }
209 EXPORT_SYMBOL_GPL(fib_rules_unregister);
210 
211 static int uid_range_set(struct fib_kuid_range *range)
212 {
213 	return uid_valid(range->start) && uid_valid(range->end);
214 }
215 
216 static struct fib_kuid_range nla_get_kuid_range(struct nlattr **tb)
217 {
218 	struct fib_rule_uid_range *in;
219 	struct fib_kuid_range out;
220 
221 	in = (struct fib_rule_uid_range *)nla_data(tb[FRA_UID_RANGE]);
222 
223 	out.start = make_kuid(current_user_ns(), in->start);
224 	out.end = make_kuid(current_user_ns(), in->end);
225 
226 	return out;
227 }
228 
229 static int nla_put_uid_range(struct sk_buff *skb, struct fib_kuid_range *range)
230 {
231 	struct fib_rule_uid_range out = {
232 		from_kuid_munged(current_user_ns(), range->start),
233 		from_kuid_munged(current_user_ns(), range->end)
234 	};
235 
236 	return nla_put(skb, FRA_UID_RANGE, sizeof(out), &out);
237 }
238 
239 static int nla_get_port_range(struct nlattr *pattr,
240 			      struct fib_rule_port_range *port_range)
241 {
242 	const struct fib_rule_port_range *pr = nla_data(pattr);
243 
244 	if (!fib_rule_port_range_valid(pr))
245 		return -EINVAL;
246 
247 	port_range->start = pr->start;
248 	port_range->end = pr->end;
249 
250 	return 0;
251 }
252 
253 static int nla_put_port_range(struct sk_buff *skb, int attrtype,
254 			      struct fib_rule_port_range *range)
255 {
256 	return nla_put(skb, attrtype, sizeof(*range), range);
257 }
258 
259 static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
260 			  struct flowi *fl, int flags,
261 			  struct fib_lookup_arg *arg)
262 {
263 	int ret = 0;
264 
265 	if (rule->iifindex && (rule->iifindex != fl->flowi_iif))
266 		goto out;
267 
268 	if (rule->oifindex && (rule->oifindex != fl->flowi_oif))
269 		goto out;
270 
271 	if ((rule->mark ^ fl->flowi_mark) & rule->mark_mask)
272 		goto out;
273 
274 	if (rule->tun_id && (rule->tun_id != fl->flowi_tun_key.tun_id))
275 		goto out;
276 
277 	if (rule->l3mdev && !l3mdev_fib_rule_match(rule->fr_net, fl, arg))
278 		goto out;
279 
280 	if (uid_lt(fl->flowi_uid, rule->uid_range.start) ||
281 	    uid_gt(fl->flowi_uid, rule->uid_range.end))
282 		goto out;
283 
284 	ret = INDIRECT_CALL_MT(ops->match,
285 			       fib6_rule_match,
286 			       fib4_rule_match,
287 			       rule, fl, flags);
288 out:
289 	return (rule->flags & FIB_RULE_INVERT) ? !ret : ret;
290 }
291 
292 int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
293 		     int flags, struct fib_lookup_arg *arg)
294 {
295 	struct fib_rule *rule;
296 	int err;
297 
298 	rcu_read_lock();
299 
300 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
301 jumped:
302 		if (!fib_rule_match(rule, ops, fl, flags, arg))
303 			continue;
304 
305 		if (rule->action == FR_ACT_GOTO) {
306 			struct fib_rule *target;
307 
308 			target = rcu_dereference(rule->ctarget);
309 			if (target == NULL) {
310 				continue;
311 			} else {
312 				rule = target;
313 				goto jumped;
314 			}
315 		} else if (rule->action == FR_ACT_NOP)
316 			continue;
317 		else
318 			err = INDIRECT_CALL_MT(ops->action,
319 					       fib6_rule_action,
320 					       fib4_rule_action,
321 					       rule, fl, flags, arg);
322 
323 		if (!err && ops->suppress && INDIRECT_CALL_MT(ops->suppress,
324 							      fib6_rule_suppress,
325 							      fib4_rule_suppress,
326 							      rule, flags, arg))
327 			continue;
328 
329 		if (err != -EAGAIN) {
330 			if ((arg->flags & FIB_LOOKUP_NOREF) ||
331 			    likely(refcount_inc_not_zero(&rule->refcnt))) {
332 				arg->rule = rule;
333 				goto out;
334 			}
335 			break;
336 		}
337 	}
338 
339 	err = -ESRCH;
340 out:
341 	rcu_read_unlock();
342 
343 	return err;
344 }
345 EXPORT_SYMBOL_GPL(fib_rules_lookup);
346 
347 static int call_fib_rule_notifier(struct notifier_block *nb,
348 				  enum fib_event_type event_type,
349 				  struct fib_rule *rule, int family,
350 				  struct netlink_ext_ack *extack)
351 {
352 	struct fib_rule_notifier_info info = {
353 		.info.family = family,
354 		.info.extack = extack,
355 		.rule = rule,
356 	};
357 
358 	return call_fib_notifier(nb, event_type, &info.info);
359 }
360 
361 static int call_fib_rule_notifiers(struct net *net,
362 				   enum fib_event_type event_type,
363 				   struct fib_rule *rule,
364 				   struct fib_rules_ops *ops,
365 				   struct netlink_ext_ack *extack)
366 {
367 	struct fib_rule_notifier_info info = {
368 		.info.family = ops->family,
369 		.info.extack = extack,
370 		.rule = rule,
371 	};
372 
373 	ops->fib_rules_seq++;
374 	return call_fib_notifiers(net, event_type, &info.info);
375 }
376 
377 /* Called with rcu_read_lock() */
378 int fib_rules_dump(struct net *net, struct notifier_block *nb, int family,
379 		   struct netlink_ext_ack *extack)
380 {
381 	struct fib_rules_ops *ops;
382 	struct fib_rule *rule;
383 	int err = 0;
384 
385 	ops = lookup_rules_ops(net, family);
386 	if (!ops)
387 		return -EAFNOSUPPORT;
388 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
389 		err = call_fib_rule_notifier(nb, FIB_EVENT_RULE_ADD,
390 					     rule, family, extack);
391 		if (err)
392 			break;
393 	}
394 	rules_ops_put(ops);
395 
396 	return err;
397 }
398 EXPORT_SYMBOL_GPL(fib_rules_dump);
399 
400 unsigned int fib_rules_seq_read(struct net *net, int family)
401 {
402 	unsigned int fib_rules_seq;
403 	struct fib_rules_ops *ops;
404 
405 	ASSERT_RTNL();
406 
407 	ops = lookup_rules_ops(net, family);
408 	if (!ops)
409 		return 0;
410 	fib_rules_seq = ops->fib_rules_seq;
411 	rules_ops_put(ops);
412 
413 	return fib_rules_seq;
414 }
415 EXPORT_SYMBOL_GPL(fib_rules_seq_read);
416 
417 static struct fib_rule *rule_find(struct fib_rules_ops *ops,
418 				  struct fib_rule_hdr *frh,
419 				  struct nlattr **tb,
420 				  struct fib_rule *rule,
421 				  bool user_priority)
422 {
423 	struct fib_rule *r;
424 
425 	list_for_each_entry(r, &ops->rules_list, list) {
426 		if (rule->action && r->action != rule->action)
427 			continue;
428 
429 		if (rule->table && r->table != rule->table)
430 			continue;
431 
432 		if (user_priority && r->pref != rule->pref)
433 			continue;
434 
435 		if (rule->iifname[0] &&
436 		    memcmp(r->iifname, rule->iifname, IFNAMSIZ))
437 			continue;
438 
439 		if (rule->oifname[0] &&
440 		    memcmp(r->oifname, rule->oifname, IFNAMSIZ))
441 			continue;
442 
443 		if (rule->mark && r->mark != rule->mark)
444 			continue;
445 
446 		if (rule->suppress_ifgroup != -1 &&
447 		    r->suppress_ifgroup != rule->suppress_ifgroup)
448 			continue;
449 
450 		if (rule->suppress_prefixlen != -1 &&
451 		    r->suppress_prefixlen != rule->suppress_prefixlen)
452 			continue;
453 
454 		if (rule->mark_mask && r->mark_mask != rule->mark_mask)
455 			continue;
456 
457 		if (rule->tun_id && r->tun_id != rule->tun_id)
458 			continue;
459 
460 		if (r->fr_net != rule->fr_net)
461 			continue;
462 
463 		if (rule->l3mdev && r->l3mdev != rule->l3mdev)
464 			continue;
465 
466 		if (uid_range_set(&rule->uid_range) &&
467 		    (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
468 		    !uid_eq(r->uid_range.end, rule->uid_range.end)))
469 			continue;
470 
471 		if (rule->ip_proto && r->ip_proto != rule->ip_proto)
472 			continue;
473 
474 		if (rule->proto && r->proto != rule->proto)
475 			continue;
476 
477 		if (fib_rule_port_range_set(&rule->sport_range) &&
478 		    !fib_rule_port_range_compare(&r->sport_range,
479 						 &rule->sport_range))
480 			continue;
481 
482 		if (fib_rule_port_range_set(&rule->dport_range) &&
483 		    !fib_rule_port_range_compare(&r->dport_range,
484 						 &rule->dport_range))
485 			continue;
486 
487 		if (!ops->compare(r, frh, tb))
488 			continue;
489 		return r;
490 	}
491 
492 	return NULL;
493 }
494 
495 #ifdef CONFIG_NET_L3_MASTER_DEV
496 static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
497 			      struct netlink_ext_ack *extack)
498 {
499 	nlrule->l3mdev = nla_get_u8(nla);
500 	if (nlrule->l3mdev != 1) {
501 		NL_SET_ERR_MSG(extack, "Invalid l3mdev attribute");
502 		return -1;
503 	}
504 
505 	return 0;
506 }
507 #else
508 static int fib_nl2rule_l3mdev(struct nlattr *nla, struct fib_rule *nlrule,
509 			      struct netlink_ext_ack *extack)
510 {
511 	NL_SET_ERR_MSG(extack, "l3mdev support is not enabled in kernel");
512 	return -1;
513 }
514 #endif
515 
516 static int fib_nl2rule(struct sk_buff *skb, struct nlmsghdr *nlh,
517 		       struct netlink_ext_ack *extack,
518 		       struct fib_rules_ops *ops,
519 		       struct nlattr *tb[],
520 		       struct fib_rule **rule,
521 		       bool *user_priority)
522 {
523 	struct net *net = sock_net(skb->sk);
524 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
525 	struct fib_rule *nlrule = NULL;
526 	int err = -EINVAL;
527 
528 	if (frh->src_len)
529 		if (!tb[FRA_SRC] ||
530 		    frh->src_len > (ops->addr_size * 8) ||
531 		    nla_len(tb[FRA_SRC]) != ops->addr_size) {
532 			NL_SET_ERR_MSG(extack, "Invalid source address");
533 			goto errout;
534 	}
535 
536 	if (frh->dst_len)
537 		if (!tb[FRA_DST] ||
538 		    frh->dst_len > (ops->addr_size * 8) ||
539 		    nla_len(tb[FRA_DST]) != ops->addr_size) {
540 			NL_SET_ERR_MSG(extack, "Invalid dst address");
541 			goto errout;
542 	}
543 
544 	nlrule = kzalloc(ops->rule_size, GFP_KERNEL_ACCOUNT);
545 	if (!nlrule) {
546 		err = -ENOMEM;
547 		goto errout;
548 	}
549 	refcount_set(&nlrule->refcnt, 1);
550 	nlrule->fr_net = net;
551 
552 	if (tb[FRA_PRIORITY]) {
553 		nlrule->pref = nla_get_u32(tb[FRA_PRIORITY]);
554 		*user_priority = true;
555 	} else {
556 		nlrule->pref = fib_default_rule_pref(ops);
557 	}
558 
559 	nlrule->proto = tb[FRA_PROTOCOL] ?
560 		nla_get_u8(tb[FRA_PROTOCOL]) : RTPROT_UNSPEC;
561 
562 	if (tb[FRA_IIFNAME]) {
563 		struct net_device *dev;
564 
565 		nlrule->iifindex = -1;
566 		nla_strscpy(nlrule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
567 		dev = __dev_get_by_name(net, nlrule->iifname);
568 		if (dev)
569 			nlrule->iifindex = dev->ifindex;
570 	}
571 
572 	if (tb[FRA_OIFNAME]) {
573 		struct net_device *dev;
574 
575 		nlrule->oifindex = -1;
576 		nla_strscpy(nlrule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
577 		dev = __dev_get_by_name(net, nlrule->oifname);
578 		if (dev)
579 			nlrule->oifindex = dev->ifindex;
580 	}
581 
582 	if (tb[FRA_FWMARK]) {
583 		nlrule->mark = nla_get_u32(tb[FRA_FWMARK]);
584 		if (nlrule->mark)
585 			/* compatibility: if the mark value is non-zero all bits
586 			 * are compared unless a mask is explicitly specified.
587 			 */
588 			nlrule->mark_mask = 0xFFFFFFFF;
589 	}
590 
591 	if (tb[FRA_FWMASK])
592 		nlrule->mark_mask = nla_get_u32(tb[FRA_FWMASK]);
593 
594 	if (tb[FRA_TUN_ID])
595 		nlrule->tun_id = nla_get_be64(tb[FRA_TUN_ID]);
596 
597 	err = -EINVAL;
598 	if (tb[FRA_L3MDEV] &&
599 	    fib_nl2rule_l3mdev(tb[FRA_L3MDEV], nlrule, extack) < 0)
600 		goto errout_free;
601 
602 	nlrule->action = frh->action;
603 	nlrule->flags = frh->flags;
604 	nlrule->table = frh_get_table(frh, tb);
605 	if (tb[FRA_SUPPRESS_PREFIXLEN])
606 		nlrule->suppress_prefixlen = nla_get_u32(tb[FRA_SUPPRESS_PREFIXLEN]);
607 	else
608 		nlrule->suppress_prefixlen = -1;
609 
610 	if (tb[FRA_SUPPRESS_IFGROUP])
611 		nlrule->suppress_ifgroup = nla_get_u32(tb[FRA_SUPPRESS_IFGROUP]);
612 	else
613 		nlrule->suppress_ifgroup = -1;
614 
615 	if (tb[FRA_GOTO]) {
616 		if (nlrule->action != FR_ACT_GOTO) {
617 			NL_SET_ERR_MSG(extack, "Unexpected goto");
618 			goto errout_free;
619 		}
620 
621 		nlrule->target = nla_get_u32(tb[FRA_GOTO]);
622 		/* Backward jumps are prohibited to avoid endless loops */
623 		if (nlrule->target <= nlrule->pref) {
624 			NL_SET_ERR_MSG(extack, "Backward goto not supported");
625 			goto errout_free;
626 		}
627 	} else if (nlrule->action == FR_ACT_GOTO) {
628 		NL_SET_ERR_MSG(extack, "Missing goto target for action goto");
629 		goto errout_free;
630 	}
631 
632 	if (nlrule->l3mdev && nlrule->table) {
633 		NL_SET_ERR_MSG(extack, "l3mdev and table are mutually exclusive");
634 		goto errout_free;
635 	}
636 
637 	if (tb[FRA_UID_RANGE]) {
638 		if (current_user_ns() != net->user_ns) {
639 			err = -EPERM;
640 			NL_SET_ERR_MSG(extack, "No permission to set uid");
641 			goto errout_free;
642 		}
643 
644 		nlrule->uid_range = nla_get_kuid_range(tb);
645 
646 		if (!uid_range_set(&nlrule->uid_range) ||
647 		    !uid_lte(nlrule->uid_range.start, nlrule->uid_range.end)) {
648 			NL_SET_ERR_MSG(extack, "Invalid uid range");
649 			goto errout_free;
650 		}
651 	} else {
652 		nlrule->uid_range = fib_kuid_range_unset;
653 	}
654 
655 	if (tb[FRA_IP_PROTO])
656 		nlrule->ip_proto = nla_get_u8(tb[FRA_IP_PROTO]);
657 
658 	if (tb[FRA_SPORT_RANGE]) {
659 		err = nla_get_port_range(tb[FRA_SPORT_RANGE],
660 					 &nlrule->sport_range);
661 		if (err) {
662 			NL_SET_ERR_MSG(extack, "Invalid sport range");
663 			goto errout_free;
664 		}
665 	}
666 
667 	if (tb[FRA_DPORT_RANGE]) {
668 		err = nla_get_port_range(tb[FRA_DPORT_RANGE],
669 					 &nlrule->dport_range);
670 		if (err) {
671 			NL_SET_ERR_MSG(extack, "Invalid dport range");
672 			goto errout_free;
673 		}
674 	}
675 
676 	*rule = nlrule;
677 
678 	return 0;
679 
680 errout_free:
681 	kfree(nlrule);
682 errout:
683 	return err;
684 }
685 
686 static int rule_exists(struct fib_rules_ops *ops, struct fib_rule_hdr *frh,
687 		       struct nlattr **tb, struct fib_rule *rule)
688 {
689 	struct fib_rule *r;
690 
691 	list_for_each_entry(r, &ops->rules_list, list) {
692 		if (r->action != rule->action)
693 			continue;
694 
695 		if (r->table != rule->table)
696 			continue;
697 
698 		if (r->pref != rule->pref)
699 			continue;
700 
701 		if (memcmp(r->iifname, rule->iifname, IFNAMSIZ))
702 			continue;
703 
704 		if (memcmp(r->oifname, rule->oifname, IFNAMSIZ))
705 			continue;
706 
707 		if (r->mark != rule->mark)
708 			continue;
709 
710 		if (r->suppress_ifgroup != rule->suppress_ifgroup)
711 			continue;
712 
713 		if (r->suppress_prefixlen != rule->suppress_prefixlen)
714 			continue;
715 
716 		if (r->mark_mask != rule->mark_mask)
717 			continue;
718 
719 		if (r->tun_id != rule->tun_id)
720 			continue;
721 
722 		if (r->fr_net != rule->fr_net)
723 			continue;
724 
725 		if (r->l3mdev != rule->l3mdev)
726 			continue;
727 
728 		if (!uid_eq(r->uid_range.start, rule->uid_range.start) ||
729 		    !uid_eq(r->uid_range.end, rule->uid_range.end))
730 			continue;
731 
732 		if (r->ip_proto != rule->ip_proto)
733 			continue;
734 
735 		if (r->proto != rule->proto)
736 			continue;
737 
738 		if (!fib_rule_port_range_compare(&r->sport_range,
739 						 &rule->sport_range))
740 			continue;
741 
742 		if (!fib_rule_port_range_compare(&r->dport_range,
743 						 &rule->dport_range))
744 			continue;
745 
746 		if (!ops->compare(r, frh, tb))
747 			continue;
748 		return 1;
749 	}
750 	return 0;
751 }
752 
753 int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr *nlh,
754 		   struct netlink_ext_ack *extack)
755 {
756 	struct net *net = sock_net(skb->sk);
757 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
758 	struct fib_rules_ops *ops = NULL;
759 	struct fib_rule *rule = NULL, *r, *last = NULL;
760 	struct nlattr *tb[FRA_MAX + 1];
761 	int err = -EINVAL, unresolved = 0;
762 	bool user_priority = false;
763 
764 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
765 		NL_SET_ERR_MSG(extack, "Invalid msg length");
766 		goto errout;
767 	}
768 
769 	ops = lookup_rules_ops(net, frh->family);
770 	if (!ops) {
771 		err = -EAFNOSUPPORT;
772 		NL_SET_ERR_MSG(extack, "Rule family not supported");
773 		goto errout;
774 	}
775 
776 	err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
777 				     ops->policy, extack);
778 	if (err < 0) {
779 		NL_SET_ERR_MSG(extack, "Error parsing msg");
780 		goto errout;
781 	}
782 
783 	err = fib_nl2rule(skb, nlh, extack, ops, tb, &rule, &user_priority);
784 	if (err)
785 		goto errout;
786 
787 	if ((nlh->nlmsg_flags & NLM_F_EXCL) &&
788 	    rule_exists(ops, frh, tb, rule)) {
789 		err = -EEXIST;
790 		goto errout_free;
791 	}
792 
793 	err = ops->configure(rule, skb, frh, tb, extack);
794 	if (err < 0)
795 		goto errout_free;
796 
797 	err = call_fib_rule_notifiers(net, FIB_EVENT_RULE_ADD, rule, ops,
798 				      extack);
799 	if (err < 0)
800 		goto errout_free;
801 
802 	list_for_each_entry(r, &ops->rules_list, list) {
803 		if (r->pref == rule->target) {
804 			RCU_INIT_POINTER(rule->ctarget, r);
805 			break;
806 		}
807 	}
808 
809 	if (rcu_dereference_protected(rule->ctarget, 1) == NULL)
810 		unresolved = 1;
811 
812 	list_for_each_entry(r, &ops->rules_list, list) {
813 		if (r->pref > rule->pref)
814 			break;
815 		last = r;
816 	}
817 
818 	if (last)
819 		list_add_rcu(&rule->list, &last->list);
820 	else
821 		list_add_rcu(&rule->list, &ops->rules_list);
822 
823 	if (ops->unresolved_rules) {
824 		/*
825 		 * There are unresolved goto rules in the list, check if
826 		 * any of them are pointing to this new rule.
827 		 */
828 		list_for_each_entry(r, &ops->rules_list, list) {
829 			if (r->action == FR_ACT_GOTO &&
830 			    r->target == rule->pref &&
831 			    rtnl_dereference(r->ctarget) == NULL) {
832 				rcu_assign_pointer(r->ctarget, rule);
833 				if (--ops->unresolved_rules == 0)
834 					break;
835 			}
836 		}
837 	}
838 
839 	if (rule->action == FR_ACT_GOTO)
840 		ops->nr_goto_rules++;
841 
842 	if (unresolved)
843 		ops->unresolved_rules++;
844 
845 	if (rule->tun_id)
846 		ip_tunnel_need_metadata();
847 
848 	notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).portid);
849 	flush_route_cache(ops);
850 	rules_ops_put(ops);
851 	return 0;
852 
853 errout_free:
854 	kfree(rule);
855 errout:
856 	rules_ops_put(ops);
857 	return err;
858 }
859 EXPORT_SYMBOL_GPL(fib_nl_newrule);
860 
861 int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr *nlh,
862 		   struct netlink_ext_ack *extack)
863 {
864 	struct net *net = sock_net(skb->sk);
865 	struct fib_rule_hdr *frh = nlmsg_data(nlh);
866 	struct fib_rules_ops *ops = NULL;
867 	struct fib_rule *rule = NULL, *r, *nlrule = NULL;
868 	struct nlattr *tb[FRA_MAX+1];
869 	int err = -EINVAL;
870 	bool user_priority = false;
871 
872 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
873 		NL_SET_ERR_MSG(extack, "Invalid msg length");
874 		goto errout;
875 	}
876 
877 	ops = lookup_rules_ops(net, frh->family);
878 	if (ops == NULL) {
879 		err = -EAFNOSUPPORT;
880 		NL_SET_ERR_MSG(extack, "Rule family not supported");
881 		goto errout;
882 	}
883 
884 	err = nlmsg_parse_deprecated(nlh, sizeof(*frh), tb, FRA_MAX,
885 				     ops->policy, extack);
886 	if (err < 0) {
887 		NL_SET_ERR_MSG(extack, "Error parsing msg");
888 		goto errout;
889 	}
890 
891 	err = fib_nl2rule(skb, nlh, extack, ops, tb, &nlrule, &user_priority);
892 	if (err)
893 		goto errout;
894 
895 	rule = rule_find(ops, frh, tb, nlrule, user_priority);
896 	if (!rule) {
897 		err = -ENOENT;
898 		goto errout;
899 	}
900 
901 	if (rule->flags & FIB_RULE_PERMANENT) {
902 		err = -EPERM;
903 		goto errout;
904 	}
905 
906 	if (ops->delete) {
907 		err = ops->delete(rule);
908 		if (err)
909 			goto errout;
910 	}
911 
912 	if (rule->tun_id)
913 		ip_tunnel_unneed_metadata();
914 
915 	list_del_rcu(&rule->list);
916 
917 	if (rule->action == FR_ACT_GOTO) {
918 		ops->nr_goto_rules--;
919 		if (rtnl_dereference(rule->ctarget) == NULL)
920 			ops->unresolved_rules--;
921 	}
922 
923 	/*
924 	 * Check if this rule is a target to any of them. If so,
925 	 * adjust to the next one with the same preference or
926 	 * disable them. As this operation is eventually very
927 	 * expensive, it is only performed if goto rules, except
928 	 * current if it is goto rule, have actually been added.
929 	 */
930 	if (ops->nr_goto_rules > 0) {
931 		struct fib_rule *n;
932 
933 		n = list_next_entry(rule, list);
934 		if (&n->list == &ops->rules_list || n->pref != rule->pref)
935 			n = NULL;
936 		list_for_each_entry(r, &ops->rules_list, list) {
937 			if (rtnl_dereference(r->ctarget) != rule)
938 				continue;
939 			rcu_assign_pointer(r->ctarget, n);
940 			if (!n)
941 				ops->unresolved_rules++;
942 		}
943 	}
944 
945 	call_fib_rule_notifiers(net, FIB_EVENT_RULE_DEL, rule, ops,
946 				NULL);
947 	notify_rule_change(RTM_DELRULE, rule, ops, nlh,
948 			   NETLINK_CB(skb).portid);
949 	fib_rule_put(rule);
950 	flush_route_cache(ops);
951 	rules_ops_put(ops);
952 	kfree(nlrule);
953 	return 0;
954 
955 errout:
956 	kfree(nlrule);
957 	rules_ops_put(ops);
958 	return err;
959 }
960 EXPORT_SYMBOL_GPL(fib_nl_delrule);
961 
962 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
963 					 struct fib_rule *rule)
964 {
965 	size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
966 			 + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
967 			 + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
968 			 + nla_total_size(4) /* FRA_PRIORITY */
969 			 + nla_total_size(4) /* FRA_TABLE */
970 			 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
971 			 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
972 			 + nla_total_size(4) /* FRA_FWMARK */
973 			 + nla_total_size(4) /* FRA_FWMASK */
974 			 + nla_total_size_64bit(8) /* FRA_TUN_ID */
975 			 + nla_total_size(sizeof(struct fib_kuid_range))
976 			 + nla_total_size(1) /* FRA_PROTOCOL */
977 			 + nla_total_size(1) /* FRA_IP_PROTO */
978 			 + nla_total_size(sizeof(struct fib_rule_port_range)) /* FRA_SPORT_RANGE */
979 			 + nla_total_size(sizeof(struct fib_rule_port_range)); /* FRA_DPORT_RANGE */
980 
981 	if (ops->nlmsg_payload)
982 		payload += ops->nlmsg_payload(rule);
983 
984 	return payload;
985 }
986 
987 static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
988 			    u32 pid, u32 seq, int type, int flags,
989 			    struct fib_rules_ops *ops)
990 {
991 	struct nlmsghdr *nlh;
992 	struct fib_rule_hdr *frh;
993 
994 	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*frh), flags);
995 	if (nlh == NULL)
996 		return -EMSGSIZE;
997 
998 	frh = nlmsg_data(nlh);
999 	frh->family = ops->family;
1000 	frh->table = rule->table < 256 ? rule->table : RT_TABLE_COMPAT;
1001 	if (nla_put_u32(skb, FRA_TABLE, rule->table))
1002 		goto nla_put_failure;
1003 	if (nla_put_u32(skb, FRA_SUPPRESS_PREFIXLEN, rule->suppress_prefixlen))
1004 		goto nla_put_failure;
1005 	frh->res1 = 0;
1006 	frh->res2 = 0;
1007 	frh->action = rule->action;
1008 	frh->flags = rule->flags;
1009 
1010 	if (nla_put_u8(skb, FRA_PROTOCOL, rule->proto))
1011 		goto nla_put_failure;
1012 
1013 	if (rule->action == FR_ACT_GOTO &&
1014 	    rcu_access_pointer(rule->ctarget) == NULL)
1015 		frh->flags |= FIB_RULE_UNRESOLVED;
1016 
1017 	if (rule->iifname[0]) {
1018 		if (nla_put_string(skb, FRA_IIFNAME, rule->iifname))
1019 			goto nla_put_failure;
1020 		if (rule->iifindex == -1)
1021 			frh->flags |= FIB_RULE_IIF_DETACHED;
1022 	}
1023 
1024 	if (rule->oifname[0]) {
1025 		if (nla_put_string(skb, FRA_OIFNAME, rule->oifname))
1026 			goto nla_put_failure;
1027 		if (rule->oifindex == -1)
1028 			frh->flags |= FIB_RULE_OIF_DETACHED;
1029 	}
1030 
1031 	if ((rule->pref &&
1032 	     nla_put_u32(skb, FRA_PRIORITY, rule->pref)) ||
1033 	    (rule->mark &&
1034 	     nla_put_u32(skb, FRA_FWMARK, rule->mark)) ||
1035 	    ((rule->mark_mask || rule->mark) &&
1036 	     nla_put_u32(skb, FRA_FWMASK, rule->mark_mask)) ||
1037 	    (rule->target &&
1038 	     nla_put_u32(skb, FRA_GOTO, rule->target)) ||
1039 	    (rule->tun_id &&
1040 	     nla_put_be64(skb, FRA_TUN_ID, rule->tun_id, FRA_PAD)) ||
1041 	    (rule->l3mdev &&
1042 	     nla_put_u8(skb, FRA_L3MDEV, rule->l3mdev)) ||
1043 	    (uid_range_set(&rule->uid_range) &&
1044 	     nla_put_uid_range(skb, &rule->uid_range)) ||
1045 	    (fib_rule_port_range_set(&rule->sport_range) &&
1046 	     nla_put_port_range(skb, FRA_SPORT_RANGE, &rule->sport_range)) ||
1047 	    (fib_rule_port_range_set(&rule->dport_range) &&
1048 	     nla_put_port_range(skb, FRA_DPORT_RANGE, &rule->dport_range)) ||
1049 	    (rule->ip_proto && nla_put_u8(skb, FRA_IP_PROTO, rule->ip_proto)))
1050 		goto nla_put_failure;
1051 
1052 	if (rule->suppress_ifgroup != -1) {
1053 		if (nla_put_u32(skb, FRA_SUPPRESS_IFGROUP, rule->suppress_ifgroup))
1054 			goto nla_put_failure;
1055 	}
1056 
1057 	if (ops->fill(rule, skb, frh) < 0)
1058 		goto nla_put_failure;
1059 
1060 	nlmsg_end(skb, nlh);
1061 	return 0;
1062 
1063 nla_put_failure:
1064 	nlmsg_cancel(skb, nlh);
1065 	return -EMSGSIZE;
1066 }
1067 
1068 static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
1069 		      struct fib_rules_ops *ops)
1070 {
1071 	int idx = 0;
1072 	struct fib_rule *rule;
1073 	int err = 0;
1074 
1075 	rcu_read_lock();
1076 	list_for_each_entry_rcu(rule, &ops->rules_list, list) {
1077 		if (idx < cb->args[1])
1078 			goto skip;
1079 
1080 		err = fib_nl_fill_rule(skb, rule, NETLINK_CB(cb->skb).portid,
1081 				       cb->nlh->nlmsg_seq, RTM_NEWRULE,
1082 				       NLM_F_MULTI, ops);
1083 		if (err)
1084 			break;
1085 skip:
1086 		idx++;
1087 	}
1088 	rcu_read_unlock();
1089 	cb->args[1] = idx;
1090 	rules_ops_put(ops);
1091 
1092 	return err;
1093 }
1094 
1095 static int fib_valid_dumprule_req(const struct nlmsghdr *nlh,
1096 				   struct netlink_ext_ack *extack)
1097 {
1098 	struct fib_rule_hdr *frh;
1099 
1100 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh))) {
1101 		NL_SET_ERR_MSG(extack, "Invalid header for fib rule dump request");
1102 		return -EINVAL;
1103 	}
1104 
1105 	frh = nlmsg_data(nlh);
1106 	if (frh->dst_len || frh->src_len || frh->tos || frh->table ||
1107 	    frh->res1 || frh->res2 || frh->action || frh->flags) {
1108 		NL_SET_ERR_MSG(extack,
1109 			       "Invalid values in header for fib rule dump request");
1110 		return -EINVAL;
1111 	}
1112 
1113 	if (nlmsg_attrlen(nlh, sizeof(*frh))) {
1114 		NL_SET_ERR_MSG(extack, "Invalid data after header in fib rule dump request");
1115 		return -EINVAL;
1116 	}
1117 
1118 	return 0;
1119 }
1120 
1121 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
1122 {
1123 	const struct nlmsghdr *nlh = cb->nlh;
1124 	struct net *net = sock_net(skb->sk);
1125 	struct fib_rules_ops *ops;
1126 	int idx = 0, family;
1127 
1128 	if (cb->strict_check) {
1129 		int err = fib_valid_dumprule_req(nlh, cb->extack);
1130 
1131 		if (err < 0)
1132 			return err;
1133 	}
1134 
1135 	family = rtnl_msg_family(nlh);
1136 	if (family != AF_UNSPEC) {
1137 		/* Protocol specific dump request */
1138 		ops = lookup_rules_ops(net, family);
1139 		if (ops == NULL)
1140 			return -EAFNOSUPPORT;
1141 
1142 		dump_rules(skb, cb, ops);
1143 
1144 		return skb->len;
1145 	}
1146 
1147 	rcu_read_lock();
1148 	list_for_each_entry_rcu(ops, &net->rules_ops, list) {
1149 		if (idx < cb->args[0] || !try_module_get(ops->owner))
1150 			goto skip;
1151 
1152 		if (dump_rules(skb, cb, ops) < 0)
1153 			break;
1154 
1155 		cb->args[1] = 0;
1156 skip:
1157 		idx++;
1158 	}
1159 	rcu_read_unlock();
1160 	cb->args[0] = idx;
1161 
1162 	return skb->len;
1163 }
1164 
1165 static void notify_rule_change(int event, struct fib_rule *rule,
1166 			       struct fib_rules_ops *ops, struct nlmsghdr *nlh,
1167 			       u32 pid)
1168 {
1169 	struct net *net;
1170 	struct sk_buff *skb;
1171 	int err = -ENOMEM;
1172 
1173 	net = ops->fro_net;
1174 	skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
1175 	if (skb == NULL)
1176 		goto errout;
1177 
1178 	err = fib_nl_fill_rule(skb, rule, pid, nlh->nlmsg_seq, event, 0, ops);
1179 	if (err < 0) {
1180 		/* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
1181 		WARN_ON(err == -EMSGSIZE);
1182 		kfree_skb(skb);
1183 		goto errout;
1184 	}
1185 
1186 	rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
1187 	return;
1188 errout:
1189 	if (err < 0)
1190 		rtnl_set_sk_err(net, ops->nlgroup, err);
1191 }
1192 
1193 static void attach_rules(struct list_head *rules, struct net_device *dev)
1194 {
1195 	struct fib_rule *rule;
1196 
1197 	list_for_each_entry(rule, rules, list) {
1198 		if (rule->iifindex == -1 &&
1199 		    strcmp(dev->name, rule->iifname) == 0)
1200 			rule->iifindex = dev->ifindex;
1201 		if (rule->oifindex == -1 &&
1202 		    strcmp(dev->name, rule->oifname) == 0)
1203 			rule->oifindex = dev->ifindex;
1204 	}
1205 }
1206 
1207 static void detach_rules(struct list_head *rules, struct net_device *dev)
1208 {
1209 	struct fib_rule *rule;
1210 
1211 	list_for_each_entry(rule, rules, list) {
1212 		if (rule->iifindex == dev->ifindex)
1213 			rule->iifindex = -1;
1214 		if (rule->oifindex == dev->ifindex)
1215 			rule->oifindex = -1;
1216 	}
1217 }
1218 
1219 
1220 static int fib_rules_event(struct notifier_block *this, unsigned long event,
1221 			   void *ptr)
1222 {
1223 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1224 	struct net *net = dev_net(dev);
1225 	struct fib_rules_ops *ops;
1226 
1227 	ASSERT_RTNL();
1228 
1229 	switch (event) {
1230 	case NETDEV_REGISTER:
1231 		list_for_each_entry(ops, &net->rules_ops, list)
1232 			attach_rules(&ops->rules_list, dev);
1233 		break;
1234 
1235 	case NETDEV_CHANGENAME:
1236 		list_for_each_entry(ops, &net->rules_ops, list) {
1237 			detach_rules(&ops->rules_list, dev);
1238 			attach_rules(&ops->rules_list, dev);
1239 		}
1240 		break;
1241 
1242 	case NETDEV_UNREGISTER:
1243 		list_for_each_entry(ops, &net->rules_ops, list)
1244 			detach_rules(&ops->rules_list, dev);
1245 		break;
1246 	}
1247 
1248 	return NOTIFY_DONE;
1249 }
1250 
1251 static struct notifier_block fib_rules_notifier = {
1252 	.notifier_call = fib_rules_event,
1253 };
1254 
1255 static int __net_init fib_rules_net_init(struct net *net)
1256 {
1257 	INIT_LIST_HEAD(&net->rules_ops);
1258 	spin_lock_init(&net->rules_mod_lock);
1259 	return 0;
1260 }
1261 
1262 static void __net_exit fib_rules_net_exit(struct net *net)
1263 {
1264 	WARN_ON_ONCE(!list_empty(&net->rules_ops));
1265 }
1266 
1267 static struct pernet_operations fib_rules_net_ops = {
1268 	.init = fib_rules_net_init,
1269 	.exit = fib_rules_net_exit,
1270 };
1271 
1272 static int __init fib_rules_init(void)
1273 {
1274 	int err;
1275 	rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL, 0);
1276 	rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL, 0);
1277 	rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule, 0);
1278 
1279 	err = register_pernet_subsys(&fib_rules_net_ops);
1280 	if (err < 0)
1281 		goto fail;
1282 
1283 	err = register_netdevice_notifier(&fib_rules_notifier);
1284 	if (err < 0)
1285 		goto fail_unregister;
1286 
1287 	return 0;
1288 
1289 fail_unregister:
1290 	unregister_pernet_subsys(&fib_rules_net_ops);
1291 fail:
1292 	rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
1293 	rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
1294 	rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
1295 	return err;
1296 }
1297 
1298 subsys_initcall(fib_rules_init);
1299