1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
4 *		operating system.  INET is implemented using the  BSD Socket
5 *		interface as the means of communication with the user level.
6 *
7 *		IPv4 Forwarding Information Base: semantics.
8 *
9 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 */
11
12#include <linux/uaccess.h>
13#include <linux/bitops.h>
14#include <linux/types.h>
15#include <linux/kernel.h>
16#include <linux/jiffies.h>
17#include <linux/mm.h>
18#include <linux/string.h>
19#include <linux/socket.h>
20#include <linux/sockios.h>
21#include <linux/errno.h>
22#include <linux/in.h>
23#include <linux/inet.h>
24#include <linux/inetdevice.h>
25#include <linux/netdevice.h>
26#include <linux/if_arp.h>
27#include <linux/proc_fs.h>
28#include <linux/skbuff.h>
29#include <linux/init.h>
30#include <linux/slab.h>
31#include <linux/netlink.h>
32#include <linux/hash.h>
33#include <linux/nospec.h>
34
35#include <net/arp.h>
36#include <net/ip.h>
37#include <net/protocol.h>
38#include <net/route.h>
39#include <net/tcp.h>
40#include <net/sock.h>
41#include <net/ip_fib.h>
42#include <net/ip6_fib.h>
43#include <net/nexthop.h>
44#include <net/netlink.h>
45#include <net/rtnh.h>
46#include <net/lwtunnel.h>
47#include <net/fib_notifier.h>
48#include <net/addrconf.h>
49
50#include "fib_lookup.h"
51
52static DEFINE_SPINLOCK(fib_info_lock);
53static struct hlist_head *fib_info_hash;
54static struct hlist_head *fib_info_laddrhash;
55static unsigned int fib_info_hash_size;
56static unsigned int fib_info_cnt;
57
58#define DEVINDEX_HASHBITS 8
59#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
60static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
61
62/* for_nexthops and change_nexthops only used when nexthop object
63 * is not set in a fib_info. The logic within can reference fib_nh.
64 */
65#ifdef CONFIG_IP_ROUTE_MULTIPATH
66
67#define for_nexthops(fi) {						\
68	int nhsel; const struct fib_nh *nh;				\
69	for (nhsel = 0, nh = (fi)->fib_nh;				\
70	     nhsel < fib_info_num_path((fi));				\
71	     nh++, nhsel++)
72
73#define change_nexthops(fi) {						\
74	int nhsel; struct fib_nh *nexthop_nh;				\
75	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
76	     nhsel < fib_info_num_path((fi));				\
77	     nexthop_nh++, nhsel++)
78
79#else /* CONFIG_IP_ROUTE_MULTIPATH */
80
81/* Hope, that gcc will optimize it to get rid of dummy loop */
82
83#define for_nexthops(fi) {						\
84	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
85	for (nhsel = 0; nhsel < 1; nhsel++)
86
87#define change_nexthops(fi) {						\
88	int nhsel;							\
89	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
90	for (nhsel = 0; nhsel < 1; nhsel++)
91
92#endif /* CONFIG_IP_ROUTE_MULTIPATH */
93
94#define endfor_nexthops(fi) }
95
96
97const struct fib_prop fib_props[RTN_MAX + 1] = {
98	[RTN_UNSPEC] = {
99		.error	= 0,
100		.scope	= RT_SCOPE_NOWHERE,
101	},
102	[RTN_UNICAST] = {
103		.error	= 0,
104		.scope	= RT_SCOPE_UNIVERSE,
105	},
106	[RTN_LOCAL] = {
107		.error	= 0,
108		.scope	= RT_SCOPE_HOST,
109	},
110	[RTN_BROADCAST] = {
111		.error	= 0,
112		.scope	= RT_SCOPE_LINK,
113	},
114	[RTN_ANYCAST] = {
115		.error	= 0,
116		.scope	= RT_SCOPE_LINK,
117	},
118	[RTN_MULTICAST] = {
119		.error	= 0,
120		.scope	= RT_SCOPE_UNIVERSE,
121	},
122	[RTN_BLACKHOLE] = {
123		.error	= -EINVAL,
124		.scope	= RT_SCOPE_UNIVERSE,
125	},
126	[RTN_UNREACHABLE] = {
127		.error	= -EHOSTUNREACH,
128		.scope	= RT_SCOPE_UNIVERSE,
129	},
130	[RTN_PROHIBIT] = {
131		.error	= -EACCES,
132		.scope	= RT_SCOPE_UNIVERSE,
133	},
134	[RTN_THROW] = {
135		.error	= -EAGAIN,
136		.scope	= RT_SCOPE_UNIVERSE,
137	},
138	[RTN_NAT] = {
139		.error	= -EINVAL,
140		.scope	= RT_SCOPE_NOWHERE,
141	},
142	[RTN_XRESOLVE] = {
143		.error	= -EINVAL,
144		.scope	= RT_SCOPE_NOWHERE,
145	},
146};
147
148static void rt_fibinfo_free(struct rtable __rcu **rtp)
149{
150	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
151
152	if (!rt)
153		return;
154
155	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
156	 * because we waited an RCU grace period before calling
157	 * free_fib_info_rcu()
158	 */
159
160	dst_dev_put(&rt->dst);
161	dst_release_immediate(&rt->dst);
162}
163
164static void free_nh_exceptions(struct fib_nh_common *nhc)
165{
166	struct fnhe_hash_bucket *hash;
167	int i;
168
169	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
170	if (!hash)
171		return;
172	for (i = 0; i < FNHE_HASH_SIZE; i++) {
173		struct fib_nh_exception *fnhe;
174
175		fnhe = rcu_dereference_protected(hash[i].chain, 1);
176		while (fnhe) {
177			struct fib_nh_exception *next;
178
179			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
180
181			rt_fibinfo_free(&fnhe->fnhe_rth_input);
182			rt_fibinfo_free(&fnhe->fnhe_rth_output);
183
184			kfree(fnhe);
185
186			fnhe = next;
187		}
188	}
189	kfree(hash);
190}
191
192static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
193{
194	int cpu;
195
196	if (!rtp)
197		return;
198
199	for_each_possible_cpu(cpu) {
200		struct rtable *rt;
201
202		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
203		if (rt) {
204			dst_dev_put(&rt->dst);
205			dst_release_immediate(&rt->dst);
206		}
207	}
208	free_percpu(rtp);
209}
210
211void fib_nh_common_release(struct fib_nh_common *nhc)
212{
213	if (nhc->nhc_dev)
214		dev_put(nhc->nhc_dev);
215
216	lwtstate_put(nhc->nhc_lwtstate);
217	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
218	rt_fibinfo_free(&nhc->nhc_rth_input);
219	free_nh_exceptions(nhc);
220}
221EXPORT_SYMBOL_GPL(fib_nh_common_release);
222
223void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
224{
225#ifdef CONFIG_IP_ROUTE_CLASSID
226	if (fib_nh->nh_tclassid)
227		atomic_dec(&net->ipv4.fib_num_tclassid_users);
228#endif
229	fib_nh_common_release(&fib_nh->nh_common);
230}
231
232/* Release a nexthop info record */
233static void free_fib_info_rcu(struct rcu_head *head)
234{
235	struct fib_info *fi = container_of(head, struct fib_info, rcu);
236
237	if (fi->nh) {
238		nexthop_put(fi->nh);
239	} else {
240		change_nexthops(fi) {
241			fib_nh_release(fi->fib_net, nexthop_nh);
242		} endfor_nexthops(fi);
243	}
244
245	ip_fib_metrics_put(fi->fib_metrics);
246
247	kfree(fi);
248}
249
250void free_fib_info(struct fib_info *fi)
251{
252	if (fi->fib_dead == 0) {
253		pr_warn("Freeing alive fib_info %p\n", fi);
254		return;
255	}
256
257	call_rcu(&fi->rcu, free_fib_info_rcu);
258}
259EXPORT_SYMBOL_GPL(free_fib_info);
260
261void fib_release_info(struct fib_info *fi)
262{
263	spin_lock_bh(&fib_info_lock);
264	if (fi && --fi->fib_treeref == 0) {
265		hlist_del(&fi->fib_hash);
266
267		/* Paired with READ_ONCE() in fib_create_info(). */
268		WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
269
270		if (fi->fib_prefsrc)
271			hlist_del(&fi->fib_lhash);
272		if (fi->nh) {
273			list_del(&fi->nh_list);
274		} else {
275			change_nexthops(fi) {
276				if (!nexthop_nh->fib_nh_dev)
277					continue;
278				hlist_del(&nexthop_nh->nh_hash);
279			} endfor_nexthops(fi)
280		}
281		/* Paired with READ_ONCE() from fib_table_lookup() */
282		WRITE_ONCE(fi->fib_dead, 1);
283		fib_info_put(fi);
284	}
285	spin_unlock_bh(&fib_info_lock);
286}
287
288static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
289{
290	const struct fib_nh *onh;
291
292	if (fi->nh || ofi->nh)
293		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
294
295	if (ofi->fib_nhs == 0)
296		return 0;
297
298	for_nexthops(fi) {
299		onh = fib_info_nh(ofi, nhsel);
300
301		if (nh->fib_nh_oif != onh->fib_nh_oif ||
302		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
303		    nh->fib_nh_scope != onh->fib_nh_scope ||
304#ifdef CONFIG_IP_ROUTE_MULTIPATH
305		    nh->fib_nh_weight != onh->fib_nh_weight ||
306#endif
307#ifdef CONFIG_IP_ROUTE_CLASSID
308		    nh->nh_tclassid != onh->nh_tclassid ||
309#endif
310		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
311		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
312			return -1;
313
314		if (nh->fib_nh_gw_family == AF_INET &&
315		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
316			return -1;
317
318		if (nh->fib_nh_gw_family == AF_INET6 &&
319		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
320			return -1;
321	} endfor_nexthops(fi);
322	return 0;
323}
324
325static inline unsigned int fib_devindex_hashfn(unsigned int val)
326{
327	return hash_32(val, DEVINDEX_HASHBITS);
328}
329
330static struct hlist_head *
331fib_info_devhash_bucket(const struct net_device *dev)
332{
333	u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
334
335	return &fib_info_devhash[fib_devindex_hashfn(val)];
336}
337
338static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
339				      u32 prefsrc, u32 priority)
340{
341	unsigned int val = init_val;
342
343	val ^= (protocol << 8) | scope;
344	val ^= prefsrc;
345	val ^= priority;
346
347	return val;
348}
349
350static unsigned int fib_info_hashfn_result(unsigned int val)
351{
352	unsigned int mask = (fib_info_hash_size - 1);
353
354	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
355}
356
357static inline unsigned int fib_info_hashfn(struct fib_info *fi)
358{
359	unsigned int val;
360
361	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
362				fi->fib_scope, (__force u32)fi->fib_prefsrc,
363				fi->fib_priority);
364
365	if (fi->nh) {
366		val ^= fib_devindex_hashfn(fi->nh->id);
367	} else {
368		for_nexthops(fi) {
369			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
370		} endfor_nexthops(fi)
371	}
372
373	return fib_info_hashfn_result(val);
374}
375
376/* no metrics, only nexthop id */
377static struct fib_info *fib_find_info_nh(struct net *net,
378					 const struct fib_config *cfg)
379{
380	struct hlist_head *head;
381	struct fib_info *fi;
382	unsigned int hash;
383
384	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
385				 cfg->fc_protocol, cfg->fc_scope,
386				 (__force u32)cfg->fc_prefsrc,
387				 cfg->fc_priority);
388	hash = fib_info_hashfn_result(hash);
389	head = &fib_info_hash[hash];
390
391	hlist_for_each_entry(fi, head, fib_hash) {
392		if (!net_eq(fi->fib_net, net))
393			continue;
394		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
395			continue;
396		if (cfg->fc_protocol == fi->fib_protocol &&
397		    cfg->fc_scope == fi->fib_scope &&
398		    cfg->fc_prefsrc == fi->fib_prefsrc &&
399		    cfg->fc_priority == fi->fib_priority &&
400		    cfg->fc_type == fi->fib_type &&
401		    cfg->fc_table == fi->fib_tb_id &&
402		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
403			return fi;
404	}
405
406	return NULL;
407}
408
409static struct fib_info *fib_find_info(struct fib_info *nfi)
410{
411	struct hlist_head *head;
412	struct fib_info *fi;
413	unsigned int hash;
414
415	hash = fib_info_hashfn(nfi);
416	head = &fib_info_hash[hash];
417
418	hlist_for_each_entry(fi, head, fib_hash) {
419		if (!net_eq(fi->fib_net, nfi->fib_net))
420			continue;
421		if (fi->fib_nhs != nfi->fib_nhs)
422			continue;
423		if (nfi->fib_protocol == fi->fib_protocol &&
424		    nfi->fib_scope == fi->fib_scope &&
425		    nfi->fib_prefsrc == fi->fib_prefsrc &&
426		    nfi->fib_priority == fi->fib_priority &&
427		    nfi->fib_type == fi->fib_type &&
428		    nfi->fib_tb_id == fi->fib_tb_id &&
429		    memcmp(nfi->fib_metrics, fi->fib_metrics,
430			   sizeof(u32) * RTAX_MAX) == 0 &&
431		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
432		    nh_comp(fi, nfi) == 0)
433			return fi;
434	}
435
436	return NULL;
437}
438
439/* Check, that the gateway is already configured.
440 * Used only by redirect accept routine.
441 */
442int ip_fib_check_default(__be32 gw, struct net_device *dev)
443{
444	struct hlist_head *head;
445	struct fib_nh *nh;
446
447	spin_lock(&fib_info_lock);
448
449	head = fib_info_devhash_bucket(dev);
450
451	hlist_for_each_entry(nh, head, nh_hash) {
452		if (nh->fib_nh_dev == dev &&
453		    nh->fib_nh_gw4 == gw &&
454		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
455			spin_unlock(&fib_info_lock);
456			return 0;
457		}
458	}
459
460	spin_unlock(&fib_info_lock);
461
462	return -1;
463}
464
465static inline size_t fib_nlmsg_size(struct fib_info *fi)
466{
467	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
468			 + nla_total_size(4) /* RTA_TABLE */
469			 + nla_total_size(4) /* RTA_DST */
470			 + nla_total_size(4) /* RTA_PRIORITY */
471			 + nla_total_size(4) /* RTA_PREFSRC */
472			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
473	unsigned int nhs = fib_info_num_path(fi);
474
475	/* space for nested metrics */
476	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
477
478	if (fi->nh)
479		payload += nla_total_size(4); /* RTA_NH_ID */
480
481	if (nhs) {
482		size_t nh_encapsize = 0;
483		/* Also handles the special case nhs == 1 */
484
485		/* each nexthop is packed in an attribute */
486		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
487		unsigned int i;
488
489		/* may contain flow and gateway attribute */
490		nhsize += 2 * nla_total_size(4);
491
492		/* grab encap info */
493		for (i = 0; i < fib_info_num_path(fi); i++) {
494			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
495
496			if (nhc->nhc_lwtstate) {
497				/* RTA_ENCAP_TYPE */
498				nh_encapsize += lwtunnel_get_encap_size(
499						nhc->nhc_lwtstate);
500				/* RTA_ENCAP */
501				nh_encapsize +=  nla_total_size(2);
502			}
503		}
504
505		/* all nexthops are packed in a nested attribute */
506		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
507
508	}
509
510	return payload;
511}
512
513void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
514	       int dst_len, u32 tb_id, const struct nl_info *info,
515	       unsigned int nlm_flags)
516{
517	struct fib_rt_info fri;
518	struct sk_buff *skb;
519	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
520	int err = -ENOBUFS;
521
522	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
523	if (!skb)
524		goto errout;
525
526	fri.fi = fa->fa_info;
527	fri.tb_id = tb_id;
528	fri.dst = key;
529	fri.dst_len = dst_len;
530	fri.tos = fa->fa_tos;
531	fri.type = fa->fa_type;
532	fri.offload = fa->offload;
533	fri.trap = fa->trap;
534	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
535	if (err < 0) {
536		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
537		WARN_ON(err == -EMSGSIZE);
538		kfree_skb(skb);
539		goto errout;
540	}
541	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
542		    info->nlh, GFP_KERNEL);
543	return;
544errout:
545	if (err < 0)
546		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
547}
548
549static int fib_detect_death(struct fib_info *fi, int order,
550			    struct fib_info **last_resort, int *last_idx,
551			    int dflt)
552{
553	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
554	struct neighbour *n;
555	int state = NUD_NONE;
556
557	if (likely(nhc->nhc_gw_family == AF_INET))
558		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
559	else if (nhc->nhc_gw_family == AF_INET6)
560		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
561				 nhc->nhc_dev);
562	else
563		n = NULL;
564
565	if (n) {
566		state = n->nud_state;
567		neigh_release(n);
568	} else {
569		return 0;
570	}
571	if (state == NUD_REACHABLE)
572		return 0;
573	if ((state & NUD_VALID) && order != dflt)
574		return 0;
575	if ((state & NUD_VALID) ||
576	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
577		*last_resort = fi;
578		*last_idx = order;
579	}
580	return 1;
581}
582
583int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
584		       struct nlattr *encap, u16 encap_type,
585		       void *cfg, gfp_t gfp_flags,
586		       struct netlink_ext_ack *extack)
587{
588	int err;
589
590	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
591						    gfp_flags);
592	if (!nhc->nhc_pcpu_rth_output)
593		return -ENOMEM;
594
595	if (encap) {
596		struct lwtunnel_state *lwtstate;
597
598		if (encap_type == LWTUNNEL_ENCAP_NONE) {
599			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
600			err = -EINVAL;
601			goto lwt_failure;
602		}
603		err = lwtunnel_build_state(net, encap_type, encap,
604					   nhc->nhc_family, cfg, &lwtstate,
605					   extack);
606		if (err)
607			goto lwt_failure;
608
609		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
610	}
611
612	return 0;
613
614lwt_failure:
615	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
616	nhc->nhc_pcpu_rth_output = NULL;
617	return err;
618}
619EXPORT_SYMBOL_GPL(fib_nh_common_init);
620
621int fib_nh_init(struct net *net, struct fib_nh *nh,
622		struct fib_config *cfg, int nh_weight,
623		struct netlink_ext_ack *extack)
624{
625	int err;
626
627	nh->fib_nh_family = AF_INET;
628
629	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
630				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
631	if (err)
632		return err;
633
634	nh->fib_nh_oif = cfg->fc_oif;
635	nh->fib_nh_gw_family = cfg->fc_gw_family;
636	if (cfg->fc_gw_family == AF_INET)
637		nh->fib_nh_gw4 = cfg->fc_gw4;
638	else if (cfg->fc_gw_family == AF_INET6)
639		nh->fib_nh_gw6 = cfg->fc_gw6;
640
641	nh->fib_nh_flags = cfg->fc_flags;
642
643#ifdef CONFIG_IP_ROUTE_CLASSID
644	nh->nh_tclassid = cfg->fc_flow;
645	if (nh->nh_tclassid)
646		atomic_inc(&net->ipv4.fib_num_tclassid_users);
647#endif
648#ifdef CONFIG_IP_ROUTE_MULTIPATH
649	nh->fib_nh_weight = nh_weight;
650#endif
651	return 0;
652}
653
654#ifdef CONFIG_IP_ROUTE_MULTIPATH
655
656static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
657			      struct netlink_ext_ack *extack)
658{
659	int nhs = 0;
660
661	while (rtnh_ok(rtnh, remaining)) {
662		nhs++;
663		rtnh = rtnh_next(rtnh, &remaining);
664	}
665
666	/* leftover implies invalid nexthop configuration, discard it */
667	if (remaining > 0) {
668		NL_SET_ERR_MSG(extack,
669			       "Invalid nexthop configuration - extra data after nexthops");
670		nhs = 0;
671	}
672
673	return nhs;
674}
675
676static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
677			    struct netlink_ext_ack *extack)
678{
679	if (nla_len(nla) < sizeof(*gw)) {
680		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
681		return -EINVAL;
682	}
683
684	*gw = nla_get_in_addr(nla);
685
686	return 0;
687}
688
689/* only called when fib_nh is integrated into fib_info */
690static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
691		       int remaining, struct fib_config *cfg,
692		       struct netlink_ext_ack *extack)
693{
694	struct net *net = fi->fib_net;
695	struct fib_config fib_cfg;
696	struct fib_nh *nh;
697	int ret;
698
699	change_nexthops(fi) {
700		int attrlen;
701
702		memset(&fib_cfg, 0, sizeof(fib_cfg));
703
704		if (!rtnh_ok(rtnh, remaining)) {
705			NL_SET_ERR_MSG(extack,
706				       "Invalid nexthop configuration - extra data after nexthop");
707			return -EINVAL;
708		}
709
710		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
711			NL_SET_ERR_MSG(extack,
712				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
713			return -EINVAL;
714		}
715
716		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
717		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
718
719		attrlen = rtnh_attrlen(rtnh);
720		if (attrlen > 0) {
721			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
722
723			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
724			nlav = nla_find(attrs, attrlen, RTA_VIA);
725			if (nla && nlav) {
726				NL_SET_ERR_MSG(extack,
727					       "Nexthop configuration can not contain both GATEWAY and VIA");
728				return -EINVAL;
729			}
730			if (nla) {
731				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
732						       extack);
733				if (ret)
734					goto errout;
735
736				if (fib_cfg.fc_gw4)
737					fib_cfg.fc_gw_family = AF_INET;
738			} else if (nlav) {
739				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
740				if (ret)
741					goto errout;
742			}
743
744			nla = nla_find(attrs, attrlen, RTA_FLOW);
745			if (nla) {
746				if (nla_len(nla) < sizeof(u32)) {
747					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
748					return -EINVAL;
749				}
750				fib_cfg.fc_flow = nla_get_u32(nla);
751			}
752
753			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
754			/* RTA_ENCAP_TYPE length checked in
755			 * lwtunnel_valid_encap_type_attr
756			 */
757			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
758			if (nla)
759				fib_cfg.fc_encap_type = nla_get_u16(nla);
760		}
761
762		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
763				  rtnh->rtnh_hops + 1, extack);
764		if (ret)
765			goto errout;
766
767		rtnh = rtnh_next(rtnh, &remaining);
768	} endfor_nexthops(fi);
769
770	ret = -EINVAL;
771	nh = fib_info_nh(fi, 0);
772	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
773		NL_SET_ERR_MSG(extack,
774			       "Nexthop device index does not match RTA_OIF");
775		goto errout;
776	}
777	if (cfg->fc_gw_family) {
778		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
779		    (cfg->fc_gw_family == AF_INET &&
780		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
781		    (cfg->fc_gw_family == AF_INET6 &&
782		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
783			NL_SET_ERR_MSG(extack,
784				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
785			goto errout;
786		}
787	}
788#ifdef CONFIG_IP_ROUTE_CLASSID
789	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
790		NL_SET_ERR_MSG(extack,
791			       "Nexthop class id does not match RTA_FLOW");
792		goto errout;
793	}
794#endif
795	ret = 0;
796errout:
797	return ret;
798}
799
800/* only called when fib_nh is integrated into fib_info */
801static void fib_rebalance(struct fib_info *fi)
802{
803	int total;
804	int w;
805
806	if (fib_info_num_path(fi) < 2)
807		return;
808
809	total = 0;
810	for_nexthops(fi) {
811		if (nh->fib_nh_flags & RTNH_F_DEAD)
812			continue;
813
814		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
815		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
816			continue;
817
818		total += nh->fib_nh_weight;
819	} endfor_nexthops(fi);
820
821	w = 0;
822	change_nexthops(fi) {
823		int upper_bound;
824
825		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
826			upper_bound = -1;
827		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
828			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
829			upper_bound = -1;
830		} else {
831			w += nexthop_nh->fib_nh_weight;
832			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
833							    total) - 1;
834		}
835
836		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
837	} endfor_nexthops(fi);
838}
839#else /* CONFIG_IP_ROUTE_MULTIPATH */
840
841static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
842		       int remaining, struct fib_config *cfg,
843		       struct netlink_ext_ack *extack)
844{
845	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
846
847	return -EINVAL;
848}
849
850#define fib_rebalance(fi) do { } while (0)
851
852#endif /* CONFIG_IP_ROUTE_MULTIPATH */
853
854static int fib_encap_match(struct net *net, u16 encap_type,
855			   struct nlattr *encap,
856			   const struct fib_nh *nh,
857			   const struct fib_config *cfg,
858			   struct netlink_ext_ack *extack)
859{
860	struct lwtunnel_state *lwtstate;
861	int ret, result = 0;
862
863	if (encap_type == LWTUNNEL_ENCAP_NONE)
864		return 0;
865
866	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
867				   cfg, &lwtstate, extack);
868	if (!ret) {
869		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
870		lwtstate_free(lwtstate);
871	}
872
873	return result;
874}
875
876int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
877		 struct netlink_ext_ack *extack)
878{
879#ifdef CONFIG_IP_ROUTE_MULTIPATH
880	struct rtnexthop *rtnh;
881	int remaining;
882#endif
883
884	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
885		return 1;
886
887	if (cfg->fc_nh_id) {
888		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
889			return 0;
890		return 1;
891	}
892
893	if (fi->nh) {
894		if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
895			return 1;
896		return 0;
897	}
898
899	if (cfg->fc_oif || cfg->fc_gw_family) {
900		struct fib_nh *nh;
901
902		nh = fib_info_nh(fi, 0);
903		if (cfg->fc_encap) {
904			if (fib_encap_match(net, cfg->fc_encap_type,
905					    cfg->fc_encap, nh, cfg, extack))
906				return 1;
907		}
908#ifdef CONFIG_IP_ROUTE_CLASSID
909		if (cfg->fc_flow &&
910		    cfg->fc_flow != nh->nh_tclassid)
911			return 1;
912#endif
913		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
914		    (cfg->fc_gw_family &&
915		     cfg->fc_gw_family != nh->fib_nh_gw_family))
916			return 1;
917
918		if (cfg->fc_gw_family == AF_INET &&
919		    cfg->fc_gw4 != nh->fib_nh_gw4)
920			return 1;
921
922		if (cfg->fc_gw_family == AF_INET6 &&
923		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
924			return 1;
925
926		return 0;
927	}
928
929#ifdef CONFIG_IP_ROUTE_MULTIPATH
930	if (!cfg->fc_mp)
931		return 0;
932
933	rtnh = cfg->fc_mp;
934	remaining = cfg->fc_mp_len;
935
936	for_nexthops(fi) {
937		int attrlen;
938
939		if (!rtnh_ok(rtnh, remaining))
940			return -EINVAL;
941
942		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
943			return 1;
944
945		attrlen = rtnh_attrlen(rtnh);
946		if (attrlen > 0) {
947			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
948			int err;
949
950			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
951			nlav = nla_find(attrs, attrlen, RTA_VIA);
952			if (nla && nlav) {
953				NL_SET_ERR_MSG(extack,
954					       "Nexthop configuration can not contain both GATEWAY and VIA");
955				return -EINVAL;
956			}
957
958			if (nla) {
959				__be32 gw;
960
961				err = fib_gw_from_attr(&gw, nla, extack);
962				if (err)
963					return err;
964
965				if (nh->fib_nh_gw_family != AF_INET ||
966				    gw != nh->fib_nh_gw4)
967					return 1;
968			} else if (nlav) {
969				struct fib_config cfg2;
970
971				err = fib_gw_from_via(&cfg2, nlav, extack);
972				if (err)
973					return err;
974
975				switch (nh->fib_nh_gw_family) {
976				case AF_INET:
977					if (cfg2.fc_gw_family != AF_INET ||
978					    cfg2.fc_gw4 != nh->fib_nh_gw4)
979						return 1;
980					break;
981				case AF_INET6:
982					if (cfg2.fc_gw_family != AF_INET6 ||
983					    ipv6_addr_cmp(&cfg2.fc_gw6,
984							  &nh->fib_nh_gw6))
985						return 1;
986					break;
987				}
988			}
989
990#ifdef CONFIG_IP_ROUTE_CLASSID
991			nla = nla_find(attrs, attrlen, RTA_FLOW);
992			if (nla) {
993				if (nla_len(nla) < sizeof(u32)) {
994					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
995					return -EINVAL;
996				}
997				if (nla_get_u32(nla) != nh->nh_tclassid)
998					return 1;
999			}
1000#endif
1001		}
1002
1003		rtnh = rtnh_next(rtnh, &remaining);
1004	} endfor_nexthops(fi);
1005#endif
1006	return 0;
1007}
1008
1009bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1010{
1011	struct nlattr *nla;
1012	int remaining;
1013
1014	if (!cfg->fc_mx)
1015		return true;
1016
1017	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1018		int type = nla_type(nla);
1019		u32 fi_val, val;
1020
1021		if (!type)
1022			continue;
1023		if (type > RTAX_MAX)
1024			return false;
1025
1026		type = array_index_nospec(type, RTAX_MAX + 1);
1027		if (type == RTAX_CC_ALGO) {
1028			char tmp[TCP_CA_NAME_MAX];
1029			bool ecn_ca = false;
1030
1031			nla_strlcpy(tmp, nla, sizeof(tmp));
1032			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1033		} else {
1034			if (nla_len(nla) != sizeof(u32))
1035				return false;
1036			val = nla_get_u32(nla);
1037		}
1038
1039		fi_val = fi->fib_metrics->metrics[type - 1];
1040		if (type == RTAX_FEATURES)
1041			fi_val &= ~DST_FEATURE_ECN_CA;
1042
1043		if (fi_val != val)
1044			return false;
1045	}
1046
1047	return true;
1048}
1049
1050static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1051			      u32 table, struct netlink_ext_ack *extack)
1052{
1053	struct fib6_config cfg = {
1054		.fc_table = table,
1055		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1056		.fc_ifindex = nh->fib_nh_oif,
1057		.fc_gateway = nh->fib_nh_gw6,
1058	};
1059	struct fib6_nh fib6_nh = {};
1060	int err;
1061
1062	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1063	if (!err) {
1064		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1065		dev_hold(nh->fib_nh_dev);
1066		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1067		nh->fib_nh_scope = RT_SCOPE_LINK;
1068
1069		ipv6_stub->fib6_nh_release(&fib6_nh);
1070	}
1071
1072	return err;
1073}
1074
1075/*
1076 * Picture
1077 * -------
1078 *
1079 * Semantics of nexthop is very messy by historical reasons.
1080 * We have to take into account, that:
1081 * a) gateway can be actually local interface address,
1082 *    so that gatewayed route is direct.
1083 * b) gateway must be on-link address, possibly
1084 *    described not by an ifaddr, but also by a direct route.
1085 * c) If both gateway and interface are specified, they should not
1086 *    contradict.
1087 * d) If we use tunnel routes, gateway could be not on-link.
1088 *
1089 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1090 * results in pretty ugly and hairy code with obscure logic.
1091 *
1092 * I chose to generalized it instead, so that the size
1093 * of code does not increase practically, but it becomes
1094 * much more general.
1095 * Every prefix is assigned a "scope" value: "host" is local address,
1096 * "link" is direct route,
1097 * [ ... "site" ... "interior" ... ]
1098 * and "universe" is true gateway route with global meaning.
1099 *
1100 * Every prefix refers to a set of "nexthop"s (gw, oif),
1101 * where gw must have narrower scope. This recursion stops
1102 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1103 * which means that gw is forced to be on link.
1104 *
1105 * Code is still hairy, but now it is apparently logically
1106 * consistent and very flexible. F.e. as by-product it allows
1107 * to co-exists in peace independent exterior and interior
1108 * routing processes.
1109 *
1110 * Normally it looks as following.
1111 *
1112 * {universe prefix}  -> (gw, oif) [scope link]
1113 *		  |
1114 *		  |-> {link prefix} -> (gw, oif) [scope local]
1115 *					|
1116 *					|-> {local prefix} (terminal node)
1117 */
1118static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1119			      u8 scope, struct netlink_ext_ack *extack)
1120{
1121	struct net_device *dev;
1122	struct fib_result res;
1123	int err = 0;
1124
1125	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1126		unsigned int addr_type;
1127
1128		if (scope >= RT_SCOPE_LINK) {
1129			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1130			return -EINVAL;
1131		}
1132		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1133		if (!dev) {
1134			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1135			return -ENODEV;
1136		}
1137		if (!(dev->flags & IFF_UP)) {
1138			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1139			return -ENETDOWN;
1140		}
1141		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1142		if (addr_type != RTN_UNICAST) {
1143			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1144			return -EINVAL;
1145		}
1146		if (!netif_carrier_ok(dev))
1147			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1148		nh->fib_nh_dev = dev;
1149		dev_hold(dev);
1150		nh->fib_nh_scope = RT_SCOPE_LINK;
1151		return 0;
1152	}
1153	rcu_read_lock();
1154	{
1155		struct fib_table *tbl = NULL;
1156		struct flowi4 fl4 = {
1157			.daddr = nh->fib_nh_gw4,
1158			.flowi4_scope = scope + 1,
1159			.flowi4_oif = nh->fib_nh_oif,
1160			.flowi4_iif = LOOPBACK_IFINDEX,
1161		};
1162
1163		/* It is not necessary, but requires a bit of thinking */
1164		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1165			fl4.flowi4_scope = RT_SCOPE_LINK;
1166
1167		if (table && table != RT_TABLE_MAIN)
1168			tbl = fib_get_table(net, table);
1169
1170		if (tbl)
1171			err = fib_table_lookup(tbl, &fl4, &res,
1172					       FIB_LOOKUP_IGNORE_LINKSTATE |
1173					       FIB_LOOKUP_NOREF);
1174
1175		/* on error or if no table given do full lookup. This
1176		 * is needed for example when nexthops are in the local
1177		 * table rather than the given table
1178		 */
1179		if (!tbl || err) {
1180			err = fib_lookup(net, &fl4, &res,
1181					 FIB_LOOKUP_IGNORE_LINKSTATE);
1182		}
1183
1184		if (err) {
1185			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1186			goto out;
1187		}
1188	}
1189
1190	err = -EINVAL;
1191	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1192		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1193		goto out;
1194	}
1195	nh->fib_nh_scope = res.scope;
1196	nh->fib_nh_oif = FIB_RES_OIF(res);
1197	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1198	if (!dev) {
1199		NL_SET_ERR_MSG(extack,
1200			       "No egress device for nexthop gateway");
1201		goto out;
1202	}
1203	dev_hold(dev);
1204	if (!netif_carrier_ok(dev))
1205		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1206	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1207out:
1208	rcu_read_unlock();
1209	return err;
1210}
1211
1212static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1213			      struct netlink_ext_ack *extack)
1214{
1215	struct in_device *in_dev;
1216	int err;
1217
1218	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1219		NL_SET_ERR_MSG(extack,
1220			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1221		return -EINVAL;
1222	}
1223
1224	rcu_read_lock();
1225
1226	err = -ENODEV;
1227	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1228	if (!in_dev)
1229		goto out;
1230	err = -ENETDOWN;
1231	if (!(in_dev->dev->flags & IFF_UP)) {
1232		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1233		goto out;
1234	}
1235
1236	nh->fib_nh_dev = in_dev->dev;
1237	dev_hold(nh->fib_nh_dev);
1238	nh->fib_nh_scope = RT_SCOPE_LINK;
1239	if (!netif_carrier_ok(nh->fib_nh_dev))
1240		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1241	err = 0;
1242out:
1243	rcu_read_unlock();
1244	return err;
1245}
1246
1247int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1248		 struct netlink_ext_ack *extack)
1249{
1250	int err;
1251
1252	if (nh->fib_nh_gw_family == AF_INET)
1253		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1254	else if (nh->fib_nh_gw_family == AF_INET6)
1255		err = fib_check_nh_v6_gw(net, nh, table, extack);
1256	else
1257		err = fib_check_nh_nongw(net, nh, extack);
1258
1259	return err;
1260}
1261
1262static inline unsigned int fib_laddr_hashfn(__be32 val)
1263{
1264	unsigned int mask = (fib_info_hash_size - 1);
1265
1266	return ((__force u32)val ^
1267		((__force u32)val >> 7) ^
1268		((__force u32)val >> 14)) & mask;
1269}
1270
1271static struct hlist_head *fib_info_hash_alloc(int bytes)
1272{
1273	if (bytes <= PAGE_SIZE)
1274		return kzalloc(bytes, GFP_KERNEL);
1275	else
1276		return (struct hlist_head *)
1277			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1278					 get_order(bytes));
1279}
1280
1281static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1282{
1283	if (!hash)
1284		return;
1285
1286	if (bytes <= PAGE_SIZE)
1287		kfree(hash);
1288	else
1289		free_pages((unsigned long) hash, get_order(bytes));
1290}
1291
1292static void fib_info_hash_move(struct hlist_head *new_info_hash,
1293			       struct hlist_head *new_laddrhash,
1294			       unsigned int new_size)
1295{
1296	struct hlist_head *old_info_hash, *old_laddrhash;
1297	unsigned int old_size = fib_info_hash_size;
1298	unsigned int i, bytes;
1299
1300	spin_lock_bh(&fib_info_lock);
1301	old_info_hash = fib_info_hash;
1302	old_laddrhash = fib_info_laddrhash;
1303	fib_info_hash_size = new_size;
1304
1305	for (i = 0; i < old_size; i++) {
1306		struct hlist_head *head = &fib_info_hash[i];
1307		struct hlist_node *n;
1308		struct fib_info *fi;
1309
1310		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1311			struct hlist_head *dest;
1312			unsigned int new_hash;
1313
1314			new_hash = fib_info_hashfn(fi);
1315			dest = &new_info_hash[new_hash];
1316			hlist_add_head(&fi->fib_hash, dest);
1317		}
1318	}
1319	fib_info_hash = new_info_hash;
1320
1321	for (i = 0; i < old_size; i++) {
1322		struct hlist_head *lhead = &fib_info_laddrhash[i];
1323		struct hlist_node *n;
1324		struct fib_info *fi;
1325
1326		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1327			struct hlist_head *ldest;
1328			unsigned int new_hash;
1329
1330			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1331			ldest = &new_laddrhash[new_hash];
1332			hlist_add_head(&fi->fib_lhash, ldest);
1333		}
1334	}
1335	fib_info_laddrhash = new_laddrhash;
1336
1337	spin_unlock_bh(&fib_info_lock);
1338
1339	bytes = old_size * sizeof(struct hlist_head *);
1340	fib_info_hash_free(old_info_hash, bytes);
1341	fib_info_hash_free(old_laddrhash, bytes);
1342}
1343
1344__be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1345				 unsigned char scope)
1346{
1347	struct fib_nh *nh;
1348	__be32 saddr;
1349
1350	if (nhc->nhc_family != AF_INET)
1351		return inet_select_addr(nhc->nhc_dev, 0, scope);
1352
1353	nh = container_of(nhc, struct fib_nh, nh_common);
1354	saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1355
1356	WRITE_ONCE(nh->nh_saddr, saddr);
1357	WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1358
1359	return saddr;
1360}
1361
1362__be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1363{
1364	struct fib_nh_common *nhc = res->nhc;
1365
1366	if (res->fi->fib_prefsrc)
1367		return res->fi->fib_prefsrc;
1368
1369	if (nhc->nhc_family == AF_INET) {
1370		struct fib_nh *nh;
1371
1372		nh = container_of(nhc, struct fib_nh, nh_common);
1373		if (READ_ONCE(nh->nh_saddr_genid) ==
1374		    atomic_read(&net->ipv4.dev_addr_genid))
1375			return READ_ONCE(nh->nh_saddr);
1376	}
1377
1378	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1379}
1380
1381static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1382{
1383	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1384	    fib_prefsrc != cfg->fc_dst) {
1385		u32 tb_id = cfg->fc_table;
1386		int rc;
1387
1388		if (tb_id == RT_TABLE_MAIN)
1389			tb_id = RT_TABLE_LOCAL;
1390
1391		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1392					  fib_prefsrc, tb_id);
1393
1394		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1395			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1396						  fib_prefsrc, RT_TABLE_LOCAL);
1397		}
1398
1399		if (rc != RTN_LOCAL)
1400			return false;
1401	}
1402	return true;
1403}
1404
1405struct fib_info *fib_create_info(struct fib_config *cfg,
1406				 struct netlink_ext_ack *extack)
1407{
1408	int err;
1409	struct fib_info *fi = NULL;
1410	struct nexthop *nh = NULL;
1411	struct fib_info *ofi;
1412	int nhs = 1;
1413	struct net *net = cfg->fc_nlinfo.nl_net;
1414
1415	if (cfg->fc_type > RTN_MAX)
1416		goto err_inval;
1417
1418	/* Fast check to catch the most weird cases */
1419	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1420		NL_SET_ERR_MSG(extack, "Invalid scope");
1421		goto err_inval;
1422	}
1423
1424	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1425		NL_SET_ERR_MSG(extack,
1426			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1427		goto err_inval;
1428	}
1429
1430	if (cfg->fc_nh_id) {
1431		if (!cfg->fc_mx) {
1432			fi = fib_find_info_nh(net, cfg);
1433			if (fi) {
1434				fi->fib_treeref++;
1435				return fi;
1436			}
1437		}
1438
1439		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1440		if (!nh) {
1441			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1442			goto err_inval;
1443		}
1444		nhs = 0;
1445	}
1446
1447#ifdef CONFIG_IP_ROUTE_MULTIPATH
1448	if (cfg->fc_mp) {
1449		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1450		if (nhs == 0)
1451			goto err_inval;
1452	}
1453#endif
1454
1455	err = -ENOBUFS;
1456
1457	/* Paired with WRITE_ONCE() in fib_release_info() */
1458	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1459		unsigned int new_size = fib_info_hash_size << 1;
1460		struct hlist_head *new_info_hash;
1461		struct hlist_head *new_laddrhash;
1462		unsigned int bytes;
1463
1464		if (!new_size)
1465			new_size = 16;
1466		bytes = new_size * sizeof(struct hlist_head *);
1467		new_info_hash = fib_info_hash_alloc(bytes);
1468		new_laddrhash = fib_info_hash_alloc(bytes);
1469		if (!new_info_hash || !new_laddrhash) {
1470			fib_info_hash_free(new_info_hash, bytes);
1471			fib_info_hash_free(new_laddrhash, bytes);
1472		} else
1473			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1474
1475		if (!fib_info_hash_size)
1476			goto failure;
1477	}
1478
1479	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1480	if (!fi)
1481		goto failure;
1482	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1483					      cfg->fc_mx_len, extack);
1484	if (IS_ERR(fi->fib_metrics)) {
1485		err = PTR_ERR(fi->fib_metrics);
1486		kfree(fi);
1487		return ERR_PTR(err);
1488	}
1489
1490	fi->fib_net = net;
1491	fi->fib_protocol = cfg->fc_protocol;
1492	fi->fib_scope = cfg->fc_scope;
1493	fi->fib_flags = cfg->fc_flags;
1494	fi->fib_priority = cfg->fc_priority;
1495	fi->fib_prefsrc = cfg->fc_prefsrc;
1496	fi->fib_type = cfg->fc_type;
1497	fi->fib_tb_id = cfg->fc_table;
1498
1499	fi->fib_nhs = nhs;
1500	if (nh) {
1501		if (!nexthop_get(nh)) {
1502			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1503			err = -EINVAL;
1504		} else {
1505			err = 0;
1506			fi->nh = nh;
1507		}
1508	} else {
1509		change_nexthops(fi) {
1510			nexthop_nh->nh_parent = fi;
1511		} endfor_nexthops(fi)
1512
1513		if (cfg->fc_mp)
1514			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1515					  extack);
1516		else
1517			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1518	}
1519
1520	if (err != 0)
1521		goto failure;
1522
1523	if (fib_props[cfg->fc_type].error) {
1524		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1525			NL_SET_ERR_MSG(extack,
1526				       "Gateway, device and multipath can not be specified for this route type");
1527			goto err_inval;
1528		}
1529		goto link_it;
1530	} else {
1531		switch (cfg->fc_type) {
1532		case RTN_UNICAST:
1533		case RTN_LOCAL:
1534		case RTN_BROADCAST:
1535		case RTN_ANYCAST:
1536		case RTN_MULTICAST:
1537			break;
1538		default:
1539			NL_SET_ERR_MSG(extack, "Invalid route type");
1540			goto err_inval;
1541		}
1542	}
1543
1544	if (cfg->fc_scope > RT_SCOPE_HOST) {
1545		NL_SET_ERR_MSG(extack, "Invalid scope");
1546		goto err_inval;
1547	}
1548
1549	if (fi->nh) {
1550		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1551		if (err)
1552			goto failure;
1553	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1554		struct fib_nh *nh = fi->fib_nh;
1555
1556		/* Local address is added. */
1557		if (nhs != 1) {
1558			NL_SET_ERR_MSG(extack,
1559				       "Route with host scope can not have multiple nexthops");
1560			goto err_inval;
1561		}
1562		if (nh->fib_nh_gw_family) {
1563			NL_SET_ERR_MSG(extack,
1564				       "Route with host scope can not have a gateway");
1565			goto err_inval;
1566		}
1567		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1568		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1569		err = -ENODEV;
1570		if (!nh->fib_nh_dev)
1571			goto failure;
1572	} else {
1573		int linkdown = 0;
1574
1575		change_nexthops(fi) {
1576			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1577					   cfg->fc_table, cfg->fc_scope,
1578					   extack);
1579			if (err != 0)
1580				goto failure;
1581			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1582				linkdown++;
1583		} endfor_nexthops(fi)
1584		if (linkdown == fi->fib_nhs)
1585			fi->fib_flags |= RTNH_F_LINKDOWN;
1586	}
1587
1588	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1589		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1590		goto err_inval;
1591	}
1592
1593	if (!fi->nh) {
1594		change_nexthops(fi) {
1595			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1596						  fi->fib_scope);
1597			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1598				fi->fib_nh_is_v6 = true;
1599		} endfor_nexthops(fi)
1600
1601		fib_rebalance(fi);
1602	}
1603
1604link_it:
1605	ofi = fib_find_info(fi);
1606	if (ofi) {
1607		/* fib_table_lookup() should not see @fi yet. */
1608		fi->fib_dead = 1;
1609		free_fib_info(fi);
1610		ofi->fib_treeref++;
1611		return ofi;
1612	}
1613
1614	fi->fib_treeref++;
1615	refcount_set(&fi->fib_clntref, 1);
1616	spin_lock_bh(&fib_info_lock);
1617	fib_info_cnt++;
1618	hlist_add_head(&fi->fib_hash,
1619		       &fib_info_hash[fib_info_hashfn(fi)]);
1620	if (fi->fib_prefsrc) {
1621		struct hlist_head *head;
1622
1623		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1624		hlist_add_head(&fi->fib_lhash, head);
1625	}
1626	if (fi->nh) {
1627		list_add(&fi->nh_list, &nh->fi_list);
1628	} else {
1629		change_nexthops(fi) {
1630			struct hlist_head *head;
1631
1632			if (!nexthop_nh->fib_nh_dev)
1633				continue;
1634			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1635			hlist_add_head(&nexthop_nh->nh_hash, head);
1636		} endfor_nexthops(fi)
1637	}
1638	spin_unlock_bh(&fib_info_lock);
1639	return fi;
1640
1641err_inval:
1642	err = -EINVAL;
1643
1644failure:
1645	if (fi) {
1646		/* fib_table_lookup() should not see @fi yet. */
1647		fi->fib_dead = 1;
1648		free_fib_info(fi);
1649	}
1650
1651	return ERR_PTR(err);
1652}
1653
1654int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1655		     u8 rt_family, unsigned char *flags, bool skip_oif)
1656{
1657	if (nhc->nhc_flags & RTNH_F_DEAD)
1658		*flags |= RTNH_F_DEAD;
1659
1660	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1661		*flags |= RTNH_F_LINKDOWN;
1662
1663		rcu_read_lock();
1664		switch (nhc->nhc_family) {
1665		case AF_INET:
1666			if (ip_ignore_linkdown(nhc->nhc_dev))
1667				*flags |= RTNH_F_DEAD;
1668			break;
1669		case AF_INET6:
1670			if (ip6_ignore_linkdown(nhc->nhc_dev))
1671				*flags |= RTNH_F_DEAD;
1672			break;
1673		}
1674		rcu_read_unlock();
1675	}
1676
1677	switch (nhc->nhc_gw_family) {
1678	case AF_INET:
1679		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1680			goto nla_put_failure;
1681		break;
1682	case AF_INET6:
1683		/* if gateway family does not match nexthop family
1684		 * gateway is encoded as RTA_VIA
1685		 */
1686		if (rt_family != nhc->nhc_gw_family) {
1687			int alen = sizeof(struct in6_addr);
1688			struct nlattr *nla;
1689			struct rtvia *via;
1690
1691			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1692			if (!nla)
1693				goto nla_put_failure;
1694
1695			via = nla_data(nla);
1696			via->rtvia_family = AF_INET6;
1697			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1698		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1699					    &nhc->nhc_gw.ipv6) < 0) {
1700			goto nla_put_failure;
1701		}
1702		break;
1703	}
1704
1705	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1706	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1707		*flags |= RTNH_F_OFFLOAD;
1708
1709	if (!skip_oif && nhc->nhc_dev &&
1710	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1711		goto nla_put_failure;
1712
1713	if (nhc->nhc_lwtstate &&
1714	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1715				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1716		goto nla_put_failure;
1717
1718	return 0;
1719
1720nla_put_failure:
1721	return -EMSGSIZE;
1722}
1723EXPORT_SYMBOL_GPL(fib_nexthop_info);
1724
1725#if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1726int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1727		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1728{
1729	const struct net_device *dev = nhc->nhc_dev;
1730	struct rtnexthop *rtnh;
1731	unsigned char flags = 0;
1732
1733	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1734	if (!rtnh)
1735		goto nla_put_failure;
1736
1737	rtnh->rtnh_hops = nh_weight - 1;
1738	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1739
1740	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1741		goto nla_put_failure;
1742
1743	rtnh->rtnh_flags = flags;
1744
1745	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1746		goto nla_put_failure;
1747
1748	/* length of rtnetlink header + attributes */
1749	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1750
1751	return 0;
1752
1753nla_put_failure:
1754	return -EMSGSIZE;
1755}
1756EXPORT_SYMBOL_GPL(fib_add_nexthop);
1757#endif
1758
1759#ifdef CONFIG_IP_ROUTE_MULTIPATH
1760static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1761{
1762	struct nlattr *mp;
1763
1764	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1765	if (!mp)
1766		goto nla_put_failure;
1767
1768	if (unlikely(fi->nh)) {
1769		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1770			goto nla_put_failure;
1771		goto mp_end;
1772	}
1773
1774	for_nexthops(fi) {
1775		u32 nh_tclassid = 0;
1776#ifdef CONFIG_IP_ROUTE_CLASSID
1777		nh_tclassid = nh->nh_tclassid;
1778#endif
1779		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1780				    AF_INET, nh_tclassid) < 0)
1781			goto nla_put_failure;
1782	} endfor_nexthops(fi);
1783
1784mp_end:
1785	nla_nest_end(skb, mp);
1786
1787	return 0;
1788
1789nla_put_failure:
1790	return -EMSGSIZE;
1791}
1792#else
1793static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1794{
1795	return 0;
1796}
1797#endif
1798
1799int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1800		  struct fib_rt_info *fri, unsigned int flags)
1801{
1802	unsigned int nhs = fib_info_num_path(fri->fi);
1803	struct fib_info *fi = fri->fi;
1804	u32 tb_id = fri->tb_id;
1805	struct nlmsghdr *nlh;
1806	struct rtmsg *rtm;
1807
1808	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1809	if (!nlh)
1810		return -EMSGSIZE;
1811
1812	rtm = nlmsg_data(nlh);
1813	rtm->rtm_family = AF_INET;
1814	rtm->rtm_dst_len = fri->dst_len;
1815	rtm->rtm_src_len = 0;
1816	rtm->rtm_tos = fri->tos;
1817	if (tb_id < 256)
1818		rtm->rtm_table = tb_id;
1819	else
1820		rtm->rtm_table = RT_TABLE_COMPAT;
1821	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1822		goto nla_put_failure;
1823	rtm->rtm_type = fri->type;
1824	rtm->rtm_flags = fi->fib_flags;
1825	rtm->rtm_scope = fi->fib_scope;
1826	rtm->rtm_protocol = fi->fib_protocol;
1827
1828	if (rtm->rtm_dst_len &&
1829	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1830		goto nla_put_failure;
1831	if (fi->fib_priority &&
1832	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1833		goto nla_put_failure;
1834	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1835		goto nla_put_failure;
1836
1837	if (fi->fib_prefsrc &&
1838	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1839		goto nla_put_failure;
1840
1841	if (fi->nh) {
1842		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1843			goto nla_put_failure;
1844		if (nexthop_is_blackhole(fi->nh))
1845			rtm->rtm_type = RTN_BLACKHOLE;
1846		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1847			goto offload;
1848	}
1849
1850	if (nhs == 1) {
1851		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1852		unsigned char flags = 0;
1853
1854		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1855			goto nla_put_failure;
1856
1857		rtm->rtm_flags = flags;
1858#ifdef CONFIG_IP_ROUTE_CLASSID
1859		if (nhc->nhc_family == AF_INET) {
1860			struct fib_nh *nh;
1861
1862			nh = container_of(nhc, struct fib_nh, nh_common);
1863			if (nh->nh_tclassid &&
1864			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1865				goto nla_put_failure;
1866		}
1867#endif
1868	} else {
1869		if (fib_add_multipath(skb, fi) < 0)
1870			goto nla_put_failure;
1871	}
1872
1873offload:
1874	if (fri->offload)
1875		rtm->rtm_flags |= RTM_F_OFFLOAD;
1876	if (fri->trap)
1877		rtm->rtm_flags |= RTM_F_TRAP;
1878
1879	nlmsg_end(skb, nlh);
1880	return 0;
1881
1882nla_put_failure:
1883	nlmsg_cancel(skb, nlh);
1884	return -EMSGSIZE;
1885}
1886
1887/*
1888 * Update FIB if:
1889 * - local address disappeared -> we must delete all the entries
1890 *   referring to it.
1891 * - device went down -> we must shutdown all nexthops going via it.
1892 */
1893int fib_sync_down_addr(struct net_device *dev, __be32 local)
1894{
1895	int ret = 0;
1896	unsigned int hash = fib_laddr_hashfn(local);
1897	struct hlist_head *head = &fib_info_laddrhash[hash];
1898	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1899	struct net *net = dev_net(dev);
1900	struct fib_info *fi;
1901
1902	if (!fib_info_laddrhash || local == 0)
1903		return 0;
1904
1905	hlist_for_each_entry(fi, head, fib_lhash) {
1906		if (!net_eq(fi->fib_net, net) ||
1907		    fi->fib_tb_id != tb_id)
1908			continue;
1909		if (fi->fib_prefsrc == local) {
1910			fi->fib_flags |= RTNH_F_DEAD;
1911			fi->pfsrc_removed = true;
1912			ret++;
1913		}
1914	}
1915	return ret;
1916}
1917
1918static int call_fib_nh_notifiers(struct fib_nh *nh,
1919				 enum fib_event_type event_type)
1920{
1921	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1922	struct fib_nh_notifier_info info = {
1923		.fib_nh = nh,
1924	};
1925
1926	switch (event_type) {
1927	case FIB_EVENT_NH_ADD:
1928		if (nh->fib_nh_flags & RTNH_F_DEAD)
1929			break;
1930		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1931			break;
1932		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1933					   &info.info);
1934	case FIB_EVENT_NH_DEL:
1935		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1936		    (nh->fib_nh_flags & RTNH_F_DEAD))
1937			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1938						   event_type, &info.info);
1939	default:
1940		break;
1941	}
1942
1943	return NOTIFY_DONE;
1944}
1945
1946/* Update the PMTU of exceptions when:
1947 * - the new MTU of the first hop becomes smaller than the PMTU
1948 * - the old MTU was the same as the PMTU, and it limited discovery of
1949 *   larger MTUs on the path. With that limit raised, we can now
1950 *   discover larger MTUs
1951 * A special case is locked exceptions, for which the PMTU is smaller
1952 * than the minimal accepted PMTU:
1953 * - if the new MTU is greater than the PMTU, don't make any change
1954 * - otherwise, unlock and set PMTU
1955 */
1956void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1957{
1958	struct fnhe_hash_bucket *bucket;
1959	int i;
1960
1961	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1962	if (!bucket)
1963		return;
1964
1965	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1966		struct fib_nh_exception *fnhe;
1967
1968		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1969		     fnhe;
1970		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1971			if (fnhe->fnhe_mtu_locked) {
1972				if (new <= fnhe->fnhe_pmtu) {
1973					fnhe->fnhe_pmtu = new;
1974					fnhe->fnhe_mtu_locked = false;
1975				}
1976			} else if (new < fnhe->fnhe_pmtu ||
1977				   orig == fnhe->fnhe_pmtu) {
1978				fnhe->fnhe_pmtu = new;
1979			}
1980		}
1981	}
1982}
1983
1984void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1985{
1986	struct hlist_head *head = fib_info_devhash_bucket(dev);
1987	struct fib_nh *nh;
1988
1989	hlist_for_each_entry(nh, head, nh_hash) {
1990		if (nh->fib_nh_dev == dev)
1991			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1992	}
1993}
1994
1995/* Event              force Flags           Description
1996 * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1997 * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1998 * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1999 * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
2000 *
2001 * only used when fib_nh is built into fib_info
2002 */
2003int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
2004{
2005	struct hlist_head *head = fib_info_devhash_bucket(dev);
2006	struct fib_info *prev_fi = NULL;
2007	int scope = RT_SCOPE_NOWHERE;
2008	struct fib_nh *nh;
2009	int ret = 0;
2010
2011	if (force)
2012		scope = -1;
2013
2014	hlist_for_each_entry(nh, head, nh_hash) {
2015		struct fib_info *fi = nh->nh_parent;
2016		int dead;
2017
2018		BUG_ON(!fi->fib_nhs);
2019		if (nh->fib_nh_dev != dev || fi == prev_fi)
2020			continue;
2021		prev_fi = fi;
2022		dead = 0;
2023		change_nexthops(fi) {
2024			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2025				dead++;
2026			else if (nexthop_nh->fib_nh_dev == dev &&
2027				 nexthop_nh->fib_nh_scope != scope) {
2028				switch (event) {
2029				case NETDEV_DOWN:
2030				case NETDEV_UNREGISTER:
2031					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2032					fallthrough;
2033				case NETDEV_CHANGE:
2034					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2035					break;
2036				}
2037				call_fib_nh_notifiers(nexthop_nh,
2038						      FIB_EVENT_NH_DEL);
2039				dead++;
2040			}
2041#ifdef CONFIG_IP_ROUTE_MULTIPATH
2042			if (event == NETDEV_UNREGISTER &&
2043			    nexthop_nh->fib_nh_dev == dev) {
2044				dead = fi->fib_nhs;
2045				break;
2046			}
2047#endif
2048		} endfor_nexthops(fi)
2049		if (dead == fi->fib_nhs) {
2050			switch (event) {
2051			case NETDEV_DOWN:
2052			case NETDEV_UNREGISTER:
2053				fi->fib_flags |= RTNH_F_DEAD;
2054				fallthrough;
2055			case NETDEV_CHANGE:
2056				fi->fib_flags |= RTNH_F_LINKDOWN;
2057				break;
2058			}
2059			ret++;
2060		}
2061
2062		fib_rebalance(fi);
2063	}
2064
2065	return ret;
2066}
2067
2068/* Must be invoked inside of an RCU protected region.  */
2069static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2070{
2071	struct fib_info *fi = NULL, *last_resort = NULL;
2072	struct hlist_head *fa_head = res->fa_head;
2073	struct fib_table *tb = res->table;
2074	u8 slen = 32 - res->prefixlen;
2075	int order = -1, last_idx = -1;
2076	struct fib_alias *fa, *fa1 = NULL;
2077	u32 last_prio = res->fi->fib_priority;
2078	u8 last_tos = 0;
2079
2080	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2081		struct fib_info *next_fi = fa->fa_info;
2082		struct fib_nh_common *nhc;
2083
2084		if (fa->fa_slen != slen)
2085			continue;
2086		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2087			continue;
2088		if (fa->tb_id != tb->tb_id)
2089			continue;
2090		if (next_fi->fib_priority > last_prio &&
2091		    fa->fa_tos == last_tos) {
2092			if (last_tos)
2093				continue;
2094			break;
2095		}
2096		if (next_fi->fib_flags & RTNH_F_DEAD)
2097			continue;
2098		last_tos = fa->fa_tos;
2099		last_prio = next_fi->fib_priority;
2100
2101		if (next_fi->fib_scope != res->scope ||
2102		    fa->fa_type != RTN_UNICAST)
2103			continue;
2104
2105		nhc = fib_info_nhc(next_fi, 0);
2106		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2107			continue;
2108
2109		fib_alias_accessed(fa);
2110
2111		if (!fi) {
2112			if (next_fi != res->fi)
2113				break;
2114			fa1 = fa;
2115		} else if (!fib_detect_death(fi, order, &last_resort,
2116					     &last_idx, fa1->fa_default)) {
2117			fib_result_assign(res, fi);
2118			fa1->fa_default = order;
2119			goto out;
2120		}
2121		fi = next_fi;
2122		order++;
2123	}
2124
2125	if (order <= 0 || !fi) {
2126		if (fa1)
2127			fa1->fa_default = -1;
2128		goto out;
2129	}
2130
2131	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2132			      fa1->fa_default)) {
2133		fib_result_assign(res, fi);
2134		fa1->fa_default = order;
2135		goto out;
2136	}
2137
2138	if (last_idx >= 0)
2139		fib_result_assign(res, last_resort);
2140	fa1->fa_default = last_idx;
2141out:
2142	return;
2143}
2144
2145/*
2146 * Dead device goes up. We wake up dead nexthops.
2147 * It takes sense only on multipath routes.
2148 *
2149 * only used when fib_nh is built into fib_info
2150 */
2151int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2152{
2153	struct fib_info *prev_fi;
2154	struct hlist_head *head;
2155	struct fib_nh *nh;
2156	int ret;
2157
2158	if (!(dev->flags & IFF_UP))
2159		return 0;
2160
2161	if (nh_flags & RTNH_F_DEAD) {
2162		unsigned int flags = dev_get_flags(dev);
2163
2164		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2165			nh_flags |= RTNH_F_LINKDOWN;
2166	}
2167
2168	prev_fi = NULL;
2169	head = fib_info_devhash_bucket(dev);
2170	ret = 0;
2171
2172	hlist_for_each_entry(nh, head, nh_hash) {
2173		struct fib_info *fi = nh->nh_parent;
2174		int alive;
2175
2176		BUG_ON(!fi->fib_nhs);
2177		if (nh->fib_nh_dev != dev || fi == prev_fi)
2178			continue;
2179
2180		prev_fi = fi;
2181		alive = 0;
2182		change_nexthops(fi) {
2183			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2184				alive++;
2185				continue;
2186			}
2187			if (!nexthop_nh->fib_nh_dev ||
2188			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2189				continue;
2190			if (nexthop_nh->fib_nh_dev != dev ||
2191			    !__in_dev_get_rtnl(dev))
2192				continue;
2193			alive++;
2194			nexthop_nh->fib_nh_flags &= ~nh_flags;
2195			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2196		} endfor_nexthops(fi)
2197
2198		if (alive > 0) {
2199			fi->fib_flags &= ~nh_flags;
2200			ret++;
2201		}
2202
2203		fib_rebalance(fi);
2204	}
2205
2206	return ret;
2207}
2208
2209#ifdef CONFIG_IP_ROUTE_MULTIPATH
2210static bool fib_good_nh(const struct fib_nh *nh)
2211{
2212	int state = NUD_REACHABLE;
2213
2214	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2215		struct neighbour *n;
2216
2217		rcu_read_lock_bh();
2218
2219		if (likely(nh->fib_nh_gw_family == AF_INET))
2220			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2221						   (__force u32)nh->fib_nh_gw4);
2222		else if (nh->fib_nh_gw_family == AF_INET6)
2223			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2224							   &nh->fib_nh_gw6);
2225		else
2226			n = NULL;
2227		if (n)
2228			state = n->nud_state;
2229
2230		rcu_read_unlock_bh();
2231	}
2232
2233	return !!(state & NUD_VALID);
2234}
2235
2236void fib_select_multipath(struct fib_result *res, int hash)
2237{
2238	struct fib_info *fi = res->fi;
2239	struct net *net = fi->fib_net;
2240	bool first = false;
2241
2242	if (unlikely(res->fi->nh)) {
2243		nexthop_path_fib_result(res, hash);
2244		return;
2245	}
2246
2247	change_nexthops(fi) {
2248		if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2249			if (!fib_good_nh(nexthop_nh))
2250				continue;
2251			if (!first) {
2252				res->nh_sel = nhsel;
2253				res->nhc = &nexthop_nh->nh_common;
2254				first = true;
2255			}
2256		}
2257
2258		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2259			continue;
2260
2261		res->nh_sel = nhsel;
2262		res->nhc = &nexthop_nh->nh_common;
2263		return;
2264	} endfor_nexthops(fi);
2265}
2266#endif
2267
2268void fib_select_path(struct net *net, struct fib_result *res,
2269		     struct flowi4 *fl4, const struct sk_buff *skb)
2270{
2271	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2272		goto check_saddr;
2273
2274#ifdef CONFIG_IP_ROUTE_MULTIPATH
2275	if (fib_info_num_path(res->fi) > 1) {
2276		int h = fib_multipath_hash(net, fl4, skb, NULL);
2277
2278		fib_select_multipath(res, h);
2279	}
2280	else
2281#endif
2282	if (!res->prefixlen &&
2283	    res->table->tb_num_default > 1 &&
2284	    res->type == RTN_UNICAST)
2285		fib_select_default(fl4, res);
2286
2287check_saddr:
2288	if (!fl4->saddr)
2289		fl4->saddr = fib_result_prefsrc(net, res);
2290}
2291