2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <linux/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
94 #include <net/dst_metadata.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/lwtunnel.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
111 #include <linux/kmemleak.h>
113 #include <net/secure_seq.h>
114 #include <net/ip_tunnels.h>
115 #include <net/l3mdev.h>
117 #include "fib_lookup.h"
119 #define RT_FL_TOS(oldflp4) \
120 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
122 #define RT_GC_TIMEOUT (300*HZ)
124 static int ip_rt_max_size;
125 static int ip_rt_redirect_number __read_mostly = 9;
126 static int ip_rt_redirect_load __read_mostly = HZ / 50;
127 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
128 static int ip_rt_error_cost __read_mostly = HZ;
129 static int ip_rt_error_burst __read_mostly = 5 * HZ;
130 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
131 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
132 static int ip_rt_min_advmss __read_mostly = 256;
134 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
137 * Interface to generic destination cache.
140 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
141 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
142 static unsigned int ipv4_mtu(const struct dst_entry *dst);
143 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
144 static void ipv4_link_failure(struct sk_buff *skb);
145 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
146 struct sk_buff *skb, u32 mtu);
147 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
148 struct sk_buff *skb);
149 static void ipv4_dst_destroy(struct dst_entry *dst);
151 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
157 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
160 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
162 static struct dst_ops ipv4_dst_ops = {
164 .check = ipv4_dst_check,
165 .default_advmss = ipv4_default_advmss,
167 .cow_metrics = ipv4_cow_metrics,
168 .destroy = ipv4_dst_destroy,
169 .negative_advice = ipv4_negative_advice,
170 .link_failure = ipv4_link_failure,
171 .update_pmtu = ip_rt_update_pmtu,
172 .redirect = ip_do_redirect,
173 .local_out = __ip_local_out,
174 .neigh_lookup = ipv4_neigh_lookup,
175 .confirm_neigh = ipv4_confirm_neigh,
178 #define ECN_OR_COST(class) TC_PRIO_##class
180 const __u8 ip_tos2prio[16] = {
182 ECN_OR_COST(BESTEFFORT),
184 ECN_OR_COST(BESTEFFORT),
190 ECN_OR_COST(INTERACTIVE),
192 ECN_OR_COST(INTERACTIVE),
193 TC_PRIO_INTERACTIVE_BULK,
194 ECN_OR_COST(INTERACTIVE_BULK),
195 TC_PRIO_INTERACTIVE_BULK,
196 ECN_OR_COST(INTERACTIVE_BULK)
198 EXPORT_SYMBOL(ip_tos2prio);
200 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
201 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
203 #ifdef CONFIG_PROC_FS
204 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
208 return SEQ_START_TOKEN;
211 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
217 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
221 static int rt_cache_seq_show(struct seq_file *seq, void *v)
223 if (v == SEQ_START_TOKEN)
224 seq_printf(seq, "%-127s\n",
225 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
226 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
231 static const struct seq_operations rt_cache_seq_ops = {
232 .start = rt_cache_seq_start,
233 .next = rt_cache_seq_next,
234 .stop = rt_cache_seq_stop,
235 .show = rt_cache_seq_show,
238 static int rt_cache_seq_open(struct inode *inode, struct file *file)
240 return seq_open(file, &rt_cache_seq_ops);
243 static const struct file_operations rt_cache_seq_fops = {
244 .open = rt_cache_seq_open,
247 .release = seq_release,
251 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
256 return SEQ_START_TOKEN;
258 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
259 if (!cpu_possible(cpu))
262 return &per_cpu(rt_cache_stat, cpu);
267 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
271 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
272 if (!cpu_possible(cpu))
275 return &per_cpu(rt_cache_stat, cpu);
281 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
286 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
288 struct rt_cache_stat *st = v;
290 if (v == SEQ_START_TOKEN) {
291 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
295 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
296 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
297 dst_entries_get_slow(&ipv4_dst_ops),
310 0, /* st->gc_total */
311 0, /* st->gc_ignored */
312 0, /* st->gc_goal_miss */
313 0, /* st->gc_dst_overflow */
314 0, /* st->in_hlist_search */
315 0 /* st->out_hlist_search */
320 static const struct seq_operations rt_cpu_seq_ops = {
321 .start = rt_cpu_seq_start,
322 .next = rt_cpu_seq_next,
323 .stop = rt_cpu_seq_stop,
324 .show = rt_cpu_seq_show,
328 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
330 return seq_open(file, &rt_cpu_seq_ops);
333 static const struct file_operations rt_cpu_seq_fops = {
334 .open = rt_cpu_seq_open,
337 .release = seq_release,
340 #ifdef CONFIG_IP_ROUTE_CLASSID
341 static int rt_acct_proc_show(struct seq_file *m, void *v)
343 struct ip_rt_acct *dst, *src;
346 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
350 for_each_possible_cpu(i) {
351 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
352 for (j = 0; j < 256; j++) {
353 dst[j].o_bytes += src[j].o_bytes;
354 dst[j].o_packets += src[j].o_packets;
355 dst[j].i_bytes += src[j].i_bytes;
356 dst[j].i_packets += src[j].i_packets;
360 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
365 static int rt_acct_proc_open(struct inode *inode, struct file *file)
367 return single_open(file, rt_acct_proc_show, NULL);
370 static const struct file_operations rt_acct_proc_fops = {
371 .open = rt_acct_proc_open,
374 .release = single_release,
378 static int __net_init ip_rt_do_proc_init(struct net *net)
380 struct proc_dir_entry *pde;
382 pde = proc_create("rt_cache", 0444, net->proc_net,
387 pde = proc_create("rt_cache", 0444,
388 net->proc_net_stat, &rt_cpu_seq_fops);
392 #ifdef CONFIG_IP_ROUTE_CLASSID
393 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
399 #ifdef CONFIG_IP_ROUTE_CLASSID
401 remove_proc_entry("rt_cache", net->proc_net_stat);
404 remove_proc_entry("rt_cache", net->proc_net);
409 static void __net_exit ip_rt_do_proc_exit(struct net *net)
411 remove_proc_entry("rt_cache", net->proc_net_stat);
412 remove_proc_entry("rt_cache", net->proc_net);
413 #ifdef CONFIG_IP_ROUTE_CLASSID
414 remove_proc_entry("rt_acct", net->proc_net);
418 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
419 .init = ip_rt_do_proc_init,
420 .exit = ip_rt_do_proc_exit,
423 static int __init ip_rt_proc_init(void)
425 return register_pernet_subsys(&ip_rt_proc_ops);
429 static inline int ip_rt_proc_init(void)
433 #endif /* CONFIG_PROC_FS */
435 static inline bool rt_is_expired(const struct rtable *rth)
437 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
440 void rt_cache_flush(struct net *net)
442 rt_genid_bump_ipv4(net);
445 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
449 struct net_device *dev = dst->dev;
450 const __be32 *pkey = daddr;
451 const struct rtable *rt;
454 rt = (const struct rtable *) dst;
456 pkey = (const __be32 *) &rt->rt_gateway;
458 pkey = &ip_hdr(skb)->daddr;
460 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
463 return neigh_create(&arp_tbl, pkey, dev);
466 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
468 struct net_device *dev = dst->dev;
469 const __be32 *pkey = daddr;
470 const struct rtable *rt;
472 rt = (const struct rtable *)dst;
474 pkey = (const __be32 *)&rt->rt_gateway;
477 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
480 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
483 #define IP_IDENTS_SZ 2048u
485 static atomic_t *ip_idents __read_mostly;
486 static u32 *ip_tstamps __read_mostly;
488 /* In order to protect privacy, we add a perturbation to identifiers
489 * if one generator is seldom used. This makes hard for an attacker
490 * to infer how many packets were sent between two points in time.
492 u32 ip_idents_reserve(u32 hash, int segs)
494 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
495 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
496 u32 old = READ_ONCE(*p_tstamp);
497 u32 now = (u32)jiffies;
500 if (old != now && cmpxchg(p_tstamp, old, now) == old)
501 delta = prandom_u32_max(now - old);
503 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
505 old = (u32)atomic_read(p_id);
506 new = old + delta + segs;
507 } while (atomic_cmpxchg(p_id, old, new) != old);
511 EXPORT_SYMBOL(ip_idents_reserve);
513 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
515 static u32 ip_idents_hashrnd __read_mostly;
518 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
520 hash = jhash_3words((__force u32)iph->daddr,
521 (__force u32)iph->saddr,
522 iph->protocol ^ net_hash_mix(net),
524 id = ip_idents_reserve(hash, segs);
527 EXPORT_SYMBOL(__ip_select_ident);
529 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
530 const struct sock *sk,
531 const struct iphdr *iph,
533 u8 prot, u32 mark, int flow_flags)
536 const struct inet_sock *inet = inet_sk(sk);
538 oif = sk->sk_bound_dev_if;
540 tos = RT_CONN_FLAGS(sk);
541 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
543 flowi4_init_output(fl4, oif, mark, tos,
544 RT_SCOPE_UNIVERSE, prot,
546 iph->daddr, iph->saddr, 0, 0,
547 sock_net_uid(net, sk));
550 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
551 const struct sock *sk)
553 const struct net *net = dev_net(skb->dev);
554 const struct iphdr *iph = ip_hdr(skb);
555 int oif = skb->dev->ifindex;
556 u8 tos = RT_TOS(iph->tos);
557 u8 prot = iph->protocol;
558 u32 mark = skb->mark;
560 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
563 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
565 const struct inet_sock *inet = inet_sk(sk);
566 const struct ip_options_rcu *inet_opt;
567 __be32 daddr = inet->inet_daddr;
570 inet_opt = rcu_dereference(inet->inet_opt);
571 if (inet_opt && inet_opt->opt.srr)
572 daddr = inet_opt->opt.faddr;
573 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
574 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
575 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
576 inet_sk_flowi_flags(sk),
577 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
581 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
582 const struct sk_buff *skb)
585 build_skb_flow_key(fl4, skb, sk);
587 build_sk_flow_key(fl4, sk);
590 static DEFINE_SPINLOCK(fnhe_lock);
592 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
596 rt = rcu_dereference(fnhe->fnhe_rth_input);
598 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
599 dst_dev_put(&rt->dst);
600 dst_release(&rt->dst);
602 rt = rcu_dereference(fnhe->fnhe_rth_output);
604 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
605 dst_dev_put(&rt->dst);
606 dst_release(&rt->dst);
610 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
612 struct fib_nh_exception *fnhe, *oldest;
614 oldest = rcu_dereference(hash->chain);
615 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
616 fnhe = rcu_dereference(fnhe->fnhe_next)) {
617 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
620 fnhe_flush_routes(oldest);
624 static inline u32 fnhe_hashfun(__be32 daddr)
626 static u32 fnhe_hashrnd __read_mostly;
629 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
630 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
631 return hash_32(hval, FNHE_HASH_SHIFT);
634 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
636 rt->rt_pmtu = fnhe->fnhe_pmtu;
637 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
638 rt->dst.expires = fnhe->fnhe_expires;
641 rt->rt_flags |= RTCF_REDIRECTED;
642 rt->rt_gateway = fnhe->fnhe_gw;
643 rt->rt_uses_gateway = 1;
647 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
648 u32 pmtu, bool lock, unsigned long expires)
650 struct fnhe_hash_bucket *hash;
651 struct fib_nh_exception *fnhe;
657 genid = fnhe_genid(dev_net(nh->nh_dev));
658 hval = fnhe_hashfun(daddr);
660 spin_lock_bh(&fnhe_lock);
662 hash = rcu_dereference(nh->nh_exceptions);
664 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
667 rcu_assign_pointer(nh->nh_exceptions, hash);
673 for (fnhe = rcu_dereference(hash->chain); fnhe;
674 fnhe = rcu_dereference(fnhe->fnhe_next)) {
675 if (fnhe->fnhe_daddr == daddr)
681 if (fnhe->fnhe_genid != genid)
682 fnhe->fnhe_genid = genid;
686 fnhe->fnhe_pmtu = pmtu;
687 fnhe->fnhe_mtu_locked = lock;
689 fnhe->fnhe_expires = max(1UL, expires);
690 /* Update all cached dsts too */
691 rt = rcu_dereference(fnhe->fnhe_rth_input);
693 fill_route_from_fnhe(rt, fnhe);
694 rt = rcu_dereference(fnhe->fnhe_rth_output);
696 fill_route_from_fnhe(rt, fnhe);
698 if (depth > FNHE_RECLAIM_DEPTH)
699 fnhe = fnhe_oldest(hash);
701 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
705 fnhe->fnhe_next = hash->chain;
706 rcu_assign_pointer(hash->chain, fnhe);
708 fnhe->fnhe_genid = genid;
709 fnhe->fnhe_daddr = daddr;
711 fnhe->fnhe_pmtu = pmtu;
712 fnhe->fnhe_mtu_locked = lock;
713 fnhe->fnhe_expires = expires;
715 /* Exception created; mark the cached routes for the nexthop
716 * stale, so anyone caching it rechecks if this exception
719 rt = rcu_dereference(nh->nh_rth_input);
721 rt->dst.obsolete = DST_OBSOLETE_KILL;
723 for_each_possible_cpu(i) {
724 struct rtable __rcu **prt;
725 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
726 rt = rcu_dereference(*prt);
728 rt->dst.obsolete = DST_OBSOLETE_KILL;
732 fnhe->fnhe_stamp = jiffies;
735 spin_unlock_bh(&fnhe_lock);
738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
741 __be32 new_gw = icmp_hdr(skb)->un.gateway;
742 __be32 old_gw = ip_hdr(skb)->saddr;
743 struct net_device *dev = skb->dev;
744 struct in_device *in_dev;
745 struct fib_result res;
749 switch (icmp_hdr(skb)->code & 7) {
751 case ICMP_REDIR_NETTOS:
752 case ICMP_REDIR_HOST:
753 case ICMP_REDIR_HOSTTOS:
760 if (rt->rt_gateway != old_gw)
763 in_dev = __in_dev_get_rcu(dev);
768 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
769 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
770 ipv4_is_zeronet(new_gw))
771 goto reject_redirect;
773 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
774 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
775 goto reject_redirect;
776 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
777 goto reject_redirect;
779 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
780 goto reject_redirect;
783 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
785 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
787 if (!(n->nud_state & NUD_VALID)) {
788 neigh_event_send(n, NULL);
790 if (fib_lookup(net, fl4, &res, 0) == 0) {
791 struct fib_nh *nh = &FIB_RES_NH(res);
793 update_or_create_fnhe(nh, fl4->daddr, new_gw,
795 jiffies + ip_rt_gc_timeout);
798 rt->dst.obsolete = DST_OBSOLETE_KILL;
799 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
806 #ifdef CONFIG_IP_ROUTE_VERBOSE
807 if (IN_DEV_LOG_MARTIANS(in_dev)) {
808 const struct iphdr *iph = (const struct iphdr *) skb->data;
809 __be32 daddr = iph->daddr;
810 __be32 saddr = iph->saddr;
812 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
813 " Advised path = %pI4 -> %pI4\n",
814 &old_gw, dev->name, &new_gw,
821 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
825 const struct iphdr *iph = (const struct iphdr *) skb->data;
826 struct net *net = dev_net(skb->dev);
827 int oif = skb->dev->ifindex;
828 u8 tos = RT_TOS(iph->tos);
829 u8 prot = iph->protocol;
830 u32 mark = skb->mark;
832 rt = (struct rtable *) dst;
834 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
835 __ip_do_redirect(rt, skb, &fl4, true);
838 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
840 struct rtable *rt = (struct rtable *)dst;
841 struct dst_entry *ret = dst;
844 if (dst->obsolete > 0) {
847 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
858 * 1. The first ip_rt_redirect_number redirects are sent
859 * with exponential backoff, then we stop sending them at all,
860 * assuming that the host ignores our redirects.
861 * 2. If we did not see packets requiring redirects
862 * during ip_rt_redirect_silence, we assume that the host
863 * forgot redirected route and start to send redirects again.
865 * This algorithm is much cheaper and more intelligent than dumb load limiting
868 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
869 * and "frag. need" (breaks PMTU discovery) in icmp.c.
872 void ip_rt_send_redirect(struct sk_buff *skb)
874 struct rtable *rt = skb_rtable(skb);
875 struct in_device *in_dev;
876 struct inet_peer *peer;
882 in_dev = __in_dev_get_rcu(rt->dst.dev);
883 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
887 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
888 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
891 net = dev_net(rt->dst.dev);
892 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
894 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
895 rt_nexthop(rt, ip_hdr(skb)->daddr));
899 /* No redirected packets during ip_rt_redirect_silence;
900 * reset the algorithm.
902 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
903 peer->rate_tokens = 0;
905 /* Too many ignored redirects; do not send anything
906 * set dst.rate_last to the last seen redirected packet.
908 if (peer->rate_tokens >= ip_rt_redirect_number) {
909 peer->rate_last = jiffies;
913 /* Check for load limit; set rate_last to the latest sent
916 if (peer->rate_tokens == 0 ||
919 (ip_rt_redirect_load << peer->rate_tokens)))) {
920 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
922 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
923 peer->rate_last = jiffies;
925 #ifdef CONFIG_IP_ROUTE_VERBOSE
927 peer->rate_tokens == ip_rt_redirect_number)
928 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
929 &ip_hdr(skb)->saddr, inet_iif(skb),
930 &ip_hdr(skb)->daddr, &gw);
937 static int ip_error(struct sk_buff *skb)
939 struct rtable *rt = skb_rtable(skb);
940 struct net_device *dev = skb->dev;
941 struct in_device *in_dev;
942 struct inet_peer *peer;
948 if (netif_is_l3_master(skb->dev)) {
949 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
954 in_dev = __in_dev_get_rcu(dev);
956 /* IP on this device is disabled. */
960 net = dev_net(rt->dst.dev);
961 if (!IN_DEV_FORWARD(in_dev)) {
962 switch (rt->dst.error) {
964 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
974 switch (rt->dst.error) {
979 code = ICMP_HOST_UNREACH;
982 code = ICMP_NET_UNREACH;
983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
986 code = ICMP_PKT_FILTERED;
990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 l3mdev_master_ifindex(skb->dev), 1);
996 peer->rate_tokens += now - peer->rate_last;
997 if (peer->rate_tokens > ip_rt_error_burst)
998 peer->rate_tokens = ip_rt_error_burst;
999 peer->rate_last = now;
1000 if (peer->rate_tokens >= ip_rt_error_cost)
1001 peer->rate_tokens -= ip_rt_error_cost;
1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1009 out: kfree_skb(skb);
1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1015 struct dst_entry *dst = &rt->dst;
1016 struct fib_result res;
1019 if (ip_mtu_locked(dst))
1022 if (ipv4_mtu(dst) < mtu)
1025 if (mtu < ip_rt_min_pmtu) {
1027 mtu = ip_rt_min_pmtu;
1030 if (rt->rt_pmtu == mtu &&
1031 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1035 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1036 struct fib_nh *nh = &FIB_RES_NH(res);
1038 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1039 jiffies + ip_rt_mtu_expires);
1044 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1045 struct sk_buff *skb, u32 mtu)
1047 struct rtable *rt = (struct rtable *) dst;
1050 ip_rt_build_flow_key(&fl4, sk, skb);
1051 __ip_rt_update_pmtu(rt, &fl4, mtu);
1054 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1055 int oif, u32 mark, u8 protocol, int flow_flags)
1057 const struct iphdr *iph = (const struct iphdr *) skb->data;
1062 mark = IP4_REPLY_MARK(net, skb->mark);
1064 __build_flow_key(net, &fl4, NULL, iph, oif,
1065 RT_TOS(iph->tos), protocol, mark, flow_flags);
1066 rt = __ip_route_output_key(net, &fl4);
1068 __ip_rt_update_pmtu(rt, &fl4, mtu);
1072 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1074 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1076 const struct iphdr *iph = (const struct iphdr *) skb->data;
1080 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1082 if (!fl4.flowi4_mark)
1083 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1085 rt = __ip_route_output_key(sock_net(sk), &fl4);
1087 __ip_rt_update_pmtu(rt, &fl4, mtu);
1092 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1094 const struct iphdr *iph = (const struct iphdr *) skb->data;
1097 struct dst_entry *odst = NULL;
1099 struct net *net = sock_net(sk);
1103 if (!ip_sk_accept_pmtu(sk))
1106 odst = sk_dst_get(sk);
1108 if (sock_owned_by_user(sk) || !odst) {
1109 __ipv4_sk_update_pmtu(skb, sk, mtu);
1113 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1115 rt = (struct rtable *)odst;
1116 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1117 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1124 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1126 if (!dst_check(&rt->dst, 0)) {
1128 dst_release(&rt->dst);
1130 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1138 sk_dst_set(sk, &rt->dst);
1144 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1146 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1147 int oif, u32 mark, u8 protocol, int flow_flags)
1149 const struct iphdr *iph = (const struct iphdr *) skb->data;
1153 __build_flow_key(net, &fl4, NULL, iph, oif,
1154 RT_TOS(iph->tos), protocol, mark, flow_flags);
1155 rt = __ip_route_output_key(net, &fl4);
1157 __ip_do_redirect(rt, skb, &fl4, false);
1161 EXPORT_SYMBOL_GPL(ipv4_redirect);
1163 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1165 const struct iphdr *iph = (const struct iphdr *) skb->data;
1168 struct net *net = sock_net(sk);
1170 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1171 rt = __ip_route_output_key(net, &fl4);
1173 __ip_do_redirect(rt, skb, &fl4, false);
1177 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1179 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1181 struct rtable *rt = (struct rtable *) dst;
1183 /* All IPV4 dsts are created with ->obsolete set to the value
1184 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1185 * into this function always.
1187 * When a PMTU/redirect information update invalidates a route,
1188 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1189 * DST_OBSOLETE_DEAD by dst_free().
1191 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1196 static void ipv4_link_failure(struct sk_buff *skb)
1200 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1202 rt = skb_rtable(skb);
1204 dst_set_expires(&rt->dst, 0);
1207 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1209 pr_debug("%s: %pI4 -> %pI4, %s\n",
1210 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1211 skb->dev ? skb->dev->name : "?");
1218 We do not cache source address of outgoing interface,
1219 because it is used only by IP RR, TS and SRR options,
1220 so that it out of fast path.
1222 BTW remember: "addr" is allowed to be not aligned
1226 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1230 if (rt_is_output_route(rt))
1231 src = ip_hdr(skb)->saddr;
1233 struct fib_result res;
1239 memset(&fl4, 0, sizeof(fl4));
1240 fl4.daddr = iph->daddr;
1241 fl4.saddr = iph->saddr;
1242 fl4.flowi4_tos = RT_TOS(iph->tos);
1243 fl4.flowi4_oif = rt->dst.dev->ifindex;
1244 fl4.flowi4_iif = skb->dev->ifindex;
1245 fl4.flowi4_mark = skb->mark;
1248 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1249 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1251 src = inet_select_addr(rt->dst.dev,
1252 rt_nexthop(rt, iph->daddr),
1256 memcpy(addr, &src, 4);
1259 #ifdef CONFIG_IP_ROUTE_CLASSID
1260 static void set_class_tag(struct rtable *rt, u32 tag)
1262 if (!(rt->dst.tclassid & 0xFFFF))
1263 rt->dst.tclassid |= tag & 0xFFFF;
1264 if (!(rt->dst.tclassid & 0xFFFF0000))
1265 rt->dst.tclassid |= tag & 0xFFFF0000;
1269 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1271 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1272 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1275 return min(advmss, IPV4_MAX_PMTU - header_size);
1278 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1280 const struct rtable *rt = (const struct rtable *) dst;
1281 unsigned int mtu = rt->rt_pmtu;
1283 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1284 mtu = dst_metric_raw(dst, RTAX_MTU);
1289 mtu = READ_ONCE(dst->dev->mtu);
1291 if (unlikely(ip_mtu_locked(dst))) {
1292 if (rt->rt_uses_gateway && mtu > 576)
1296 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1298 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1301 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1303 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1304 struct fib_nh_exception *fnhe;
1310 hval = fnhe_hashfun(daddr);
1312 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1313 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1314 if (fnhe->fnhe_daddr == daddr)
1320 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1321 __be32 daddr, const bool do_cache)
1325 spin_lock_bh(&fnhe_lock);
1327 if (daddr == fnhe->fnhe_daddr) {
1328 struct rtable __rcu **porig;
1329 struct rtable *orig;
1330 int genid = fnhe_genid(dev_net(rt->dst.dev));
1332 if (rt_is_input_route(rt))
1333 porig = &fnhe->fnhe_rth_input;
1335 porig = &fnhe->fnhe_rth_output;
1336 orig = rcu_dereference(*porig);
1338 if (fnhe->fnhe_genid != genid) {
1339 fnhe->fnhe_genid = genid;
1341 fnhe->fnhe_pmtu = 0;
1342 fnhe->fnhe_expires = 0;
1343 fnhe_flush_routes(fnhe);
1346 fill_route_from_fnhe(rt, fnhe);
1347 if (!rt->rt_gateway)
1348 rt->rt_gateway = daddr;
1352 rcu_assign_pointer(*porig, rt);
1354 dst_dev_put(&orig->dst);
1355 dst_release(&orig->dst);
1360 fnhe->fnhe_stamp = jiffies;
1362 spin_unlock_bh(&fnhe_lock);
1367 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1369 struct rtable *orig, *prev, **p;
1372 if (rt_is_input_route(rt)) {
1373 p = (struct rtable **)&nh->nh_rth_input;
1375 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1379 /* hold dst before doing cmpxchg() to avoid race condition
1383 prev = cmpxchg(p, orig, rt);
1386 dst_dev_put(&orig->dst);
1387 dst_release(&orig->dst);
1390 dst_release(&rt->dst);
1397 struct uncached_list {
1399 struct list_head head;
1402 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1404 void rt_add_uncached_list(struct rtable *rt)
1406 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1408 rt->rt_uncached_list = ul;
1410 spin_lock_bh(&ul->lock);
1411 list_add_tail(&rt->rt_uncached, &ul->head);
1412 spin_unlock_bh(&ul->lock);
1415 void rt_del_uncached_list(struct rtable *rt)
1417 if (!list_empty(&rt->rt_uncached)) {
1418 struct uncached_list *ul = rt->rt_uncached_list;
1420 spin_lock_bh(&ul->lock);
1421 list_del(&rt->rt_uncached);
1422 spin_unlock_bh(&ul->lock);
1426 static void ipv4_dst_destroy(struct dst_entry *dst)
1428 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1429 struct rtable *rt = (struct rtable *)dst;
1431 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1434 rt_del_uncached_list(rt);
1437 void rt_flush_dev(struct net_device *dev)
1439 struct net *net = dev_net(dev);
1443 for_each_possible_cpu(cpu) {
1444 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1446 spin_lock_bh(&ul->lock);
1447 list_for_each_entry(rt, &ul->head, rt_uncached) {
1448 if (rt->dst.dev != dev)
1450 rt->dst.dev = net->loopback_dev;
1451 dev_hold(rt->dst.dev);
1454 spin_unlock_bh(&ul->lock);
1458 static bool rt_cache_valid(const struct rtable *rt)
1461 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1465 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1466 const struct fib_result *res,
1467 struct fib_nh_exception *fnhe,
1468 struct fib_info *fi, u16 type, u32 itag,
1469 const bool do_cache)
1471 bool cached = false;
1474 struct fib_nh *nh = &FIB_RES_NH(*res);
1476 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1477 rt->rt_gateway = nh->nh_gw;
1478 rt->rt_uses_gateway = 1;
1480 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1481 if (fi->fib_metrics != &dst_default_metrics) {
1482 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1483 refcount_inc(&fi->fib_metrics->refcnt);
1485 #ifdef CONFIG_IP_ROUTE_CLASSID
1486 rt->dst.tclassid = nh->nh_tclassid;
1488 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1490 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1492 cached = rt_cache_route(nh, rt);
1493 if (unlikely(!cached)) {
1494 /* Routes we intend to cache in nexthop exception or
1495 * FIB nexthop have the DST_NOCACHE bit clear.
1496 * However, if we are unsuccessful at storing this
1497 * route into the cache we really need to set it.
1499 if (!rt->rt_gateway)
1500 rt->rt_gateway = daddr;
1501 rt_add_uncached_list(rt);
1504 rt_add_uncached_list(rt);
1506 #ifdef CONFIG_IP_ROUTE_CLASSID
1507 #ifdef CONFIG_IP_MULTIPLE_TABLES
1508 set_class_tag(rt, res->tclassid);
1510 set_class_tag(rt, itag);
1514 struct rtable *rt_dst_alloc(struct net_device *dev,
1515 unsigned int flags, u16 type,
1516 bool nopolicy, bool noxfrm, bool will_cache)
1520 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1521 (will_cache ? 0 : DST_HOST) |
1522 (nopolicy ? DST_NOPOLICY : 0) |
1523 (noxfrm ? DST_NOXFRM : 0));
1526 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1527 rt->rt_flags = flags;
1529 rt->rt_is_input = 0;
1532 rt->rt_mtu_locked = 0;
1534 rt->rt_uses_gateway = 0;
1535 INIT_LIST_HEAD(&rt->rt_uncached);
1537 rt->dst.output = ip_output;
1538 if (flags & RTCF_LOCAL)
1539 rt->dst.input = ip_local_deliver;
1544 EXPORT_SYMBOL(rt_dst_alloc);
1546 /* called in rcu_read_lock() section */
1547 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1548 u8 tos, struct net_device *dev,
1549 struct in_device *in_dev, u32 *itag)
1553 /* Primary sanity checks. */
1557 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1558 skb->protocol != htons(ETH_P_IP))
1561 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1564 if (ipv4_is_zeronet(saddr)) {
1565 if (!ipv4_is_local_multicast(daddr))
1568 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1576 /* called in rcu_read_lock() section */
1577 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1578 u8 tos, struct net_device *dev, int our)
1580 struct in_device *in_dev = __in_dev_get_rcu(dev);
1581 unsigned int flags = RTCF_MULTICAST;
1586 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1591 flags |= RTCF_LOCAL;
1593 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1594 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1598 #ifdef CONFIG_IP_ROUTE_CLASSID
1599 rth->dst.tclassid = itag;
1601 rth->dst.output = ip_rt_bug;
1602 rth->rt_is_input= 1;
1604 #ifdef CONFIG_IP_MROUTE
1605 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1606 rth->dst.input = ip_mr_input;
1608 RT_CACHE_STAT_INC(in_slow_mc);
1610 skb_dst_set(skb, &rth->dst);
1615 static void ip_handle_martian_source(struct net_device *dev,
1616 struct in_device *in_dev,
1617 struct sk_buff *skb,
1621 RT_CACHE_STAT_INC(in_martian_src);
1622 #ifdef CONFIG_IP_ROUTE_VERBOSE
1623 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1625 * RFC1812 recommendation, if source is martian,
1626 * the only hint is MAC header.
1628 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1629 &daddr, &saddr, dev->name);
1630 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1631 print_hex_dump(KERN_WARNING, "ll header: ",
1632 DUMP_PREFIX_OFFSET, 16, 1,
1633 skb_mac_header(skb),
1634 dev->hard_header_len, true);
1640 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1642 struct fnhe_hash_bucket *hash;
1643 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1644 u32 hval = fnhe_hashfun(daddr);
1646 spin_lock_bh(&fnhe_lock);
1648 hash = rcu_dereference_protected(nh->nh_exceptions,
1649 lockdep_is_held(&fnhe_lock));
1652 fnhe_p = &hash->chain;
1653 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1655 if (fnhe->fnhe_daddr == daddr) {
1656 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1657 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1658 fnhe_flush_routes(fnhe);
1659 kfree_rcu(fnhe, rcu);
1662 fnhe_p = &fnhe->fnhe_next;
1663 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1664 lockdep_is_held(&fnhe_lock));
1667 spin_unlock_bh(&fnhe_lock);
1670 /* called in rcu_read_lock() section */
1671 static int __mkroute_input(struct sk_buff *skb,
1672 const struct fib_result *res,
1673 struct in_device *in_dev,
1674 __be32 daddr, __be32 saddr, u32 tos)
1676 struct fib_nh_exception *fnhe;
1679 struct in_device *out_dev;
1683 /* get a working reference to the output device */
1684 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1686 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1690 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1691 in_dev->dev, in_dev, &itag);
1693 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1699 do_cache = res->fi && !itag;
1700 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1701 skb->protocol == htons(ETH_P_IP) &&
1702 (IN_DEV_SHARED_MEDIA(out_dev) ||
1703 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1704 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1706 if (skb->protocol != htons(ETH_P_IP)) {
1707 /* Not IP (i.e. ARP). Do not create route, if it is
1708 * invalid for proxy arp. DNAT routes are always valid.
1710 * Proxy arp feature have been extended to allow, ARP
1711 * replies back to the same interface, to support
1712 * Private VLAN switch technologies. See arp.c.
1714 if (out_dev == in_dev &&
1715 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1721 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1724 rth = rcu_dereference(fnhe->fnhe_rth_input);
1725 if (rth && rth->dst.expires &&
1726 time_after(jiffies, rth->dst.expires)) {
1727 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1734 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1737 if (rt_cache_valid(rth)) {
1738 skb_dst_set_noref(skb, &rth->dst);
1743 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1744 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1745 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1751 rth->rt_is_input = 1;
1752 RT_CACHE_STAT_INC(in_slow_tot);
1754 rth->dst.input = ip_forward;
1756 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1758 lwtunnel_set_redirect(&rth->dst);
1759 skb_dst_set(skb, &rth->dst);
1766 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1767 /* To make ICMP packets follow the right flow, the multipath hash is
1768 * calculated from the inner IP addresses.
1770 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1771 struct flow_keys *hash_keys)
1773 const struct iphdr *outer_iph = ip_hdr(skb);
1774 const struct iphdr *key_iph = outer_iph;
1775 const struct iphdr *inner_iph;
1776 const struct icmphdr *icmph;
1777 struct iphdr _inner_iph;
1778 struct icmphdr _icmph;
1780 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1783 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1786 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1791 if (icmph->type != ICMP_DEST_UNREACH &&
1792 icmph->type != ICMP_REDIRECT &&
1793 icmph->type != ICMP_TIME_EXCEEDED &&
1794 icmph->type != ICMP_PARAMETERPROB)
1797 inner_iph = skb_header_pointer(skb,
1798 outer_iph->ihl * 4 + sizeof(_icmph),
1799 sizeof(_inner_iph), &_inner_iph);
1803 key_iph = inner_iph;
1805 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1806 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1809 /* if skb is set it will be used and fl4 can be NULL */
1810 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1811 const struct sk_buff *skb, struct flow_keys *flkeys)
1813 struct flow_keys hash_keys;
1816 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1818 memset(&hash_keys, 0, sizeof(hash_keys));
1819 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1821 ip_multipath_l3_keys(skb, &hash_keys);
1823 hash_keys.addrs.v4addrs.src = fl4->saddr;
1824 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1828 /* skb is currently provided only when forwarding */
1830 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1831 struct flow_keys keys;
1833 /* short-circuit if we already have L4 hash present */
1835 return skb_get_hash_raw(skb) >> 1;
1837 memset(&hash_keys, 0, sizeof(hash_keys));
1840 skb_flow_dissect_flow_keys(skb, &keys, flag);
1844 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1845 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1846 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1847 hash_keys.ports.src = flkeys->ports.src;
1848 hash_keys.ports.dst = flkeys->ports.dst;
1849 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1851 memset(&hash_keys, 0, sizeof(hash_keys));
1852 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1853 hash_keys.addrs.v4addrs.src = fl4->saddr;
1854 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1855 hash_keys.ports.src = fl4->fl4_sport;
1856 hash_keys.ports.dst = fl4->fl4_dport;
1857 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1861 mhash = flow_hash_from_keys(&hash_keys);
1865 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1867 static int ip_mkroute_input(struct sk_buff *skb,
1868 struct fib_result *res,
1869 struct in_device *in_dev,
1870 __be32 daddr, __be32 saddr, u32 tos,
1871 struct flow_keys *hkeys)
1873 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1874 if (res->fi && res->fi->fib_nhs > 1) {
1875 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1877 fib_select_multipath(res, h);
1881 /* create a routing cache entry */
1882 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1886 * NOTE. We drop all the packets that has local source
1887 * addresses, because every properly looped back packet
1888 * must have correct destination already attached by output routine.
1890 * Such approach solves two big problems:
1891 * 1. Not simplex devices are handled properly.
1892 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1893 * called with rcu_read_lock()
1896 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1897 u8 tos, struct net_device *dev,
1898 struct fib_result *res)
1900 struct in_device *in_dev = __in_dev_get_rcu(dev);
1901 struct flow_keys *flkeys = NULL, _flkeys;
1902 struct net *net = dev_net(dev);
1903 struct ip_tunnel_info *tun_info;
1905 unsigned int flags = 0;
1911 /* IP on this device is disabled. */
1916 /* Check for the most weird martians, which can be not detected
1920 tun_info = skb_tunnel_info(skb);
1921 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1922 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1924 fl4.flowi4_tun_key.tun_id = 0;
1927 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1928 goto martian_source;
1932 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1935 /* Accept zero addresses only to limited broadcast;
1936 * I even do not know to fix it or not. Waiting for complains :-)
1938 if (ipv4_is_zeronet(saddr))
1939 goto martian_source;
1941 if (ipv4_is_zeronet(daddr))
1942 goto martian_destination;
1944 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1945 * and call it once if daddr or/and saddr are loopback addresses
1947 if (ipv4_is_loopback(daddr)) {
1948 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1949 goto martian_destination;
1950 } else if (ipv4_is_loopback(saddr)) {
1951 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1952 goto martian_source;
1956 * Now we are ready to route packet.
1959 fl4.flowi4_iif = dev->ifindex;
1960 fl4.flowi4_mark = skb->mark;
1961 fl4.flowi4_tos = tos;
1962 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1963 fl4.flowi4_flags = 0;
1966 fl4.flowi4_uid = sock_net_uid(net, NULL);
1968 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys))
1971 err = fib_lookup(net, &fl4, res, 0);
1973 if (!IN_DEV_FORWARD(in_dev))
1974 err = -EHOSTUNREACH;
1978 if (res->type == RTN_BROADCAST)
1981 if (res->type == RTN_LOCAL) {
1982 err = fib_validate_source(skb, saddr, daddr, tos,
1983 0, dev, in_dev, &itag);
1985 goto martian_source;
1989 if (!IN_DEV_FORWARD(in_dev)) {
1990 err = -EHOSTUNREACH;
1993 if (res->type != RTN_UNICAST)
1994 goto martian_destination;
1996 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2000 if (skb->protocol != htons(ETH_P_IP))
2003 if (!ipv4_is_zeronet(saddr)) {
2004 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2007 goto martian_source;
2009 flags |= RTCF_BROADCAST;
2010 res->type = RTN_BROADCAST;
2011 RT_CACHE_STAT_INC(in_brd);
2017 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2018 if (rt_cache_valid(rth)) {
2019 skb_dst_set_noref(skb, &rth->dst);
2027 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2028 flags | RTCF_LOCAL, res->type,
2029 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2033 rth->dst.output= ip_rt_bug;
2034 #ifdef CONFIG_IP_ROUTE_CLASSID
2035 rth->dst.tclassid = itag;
2037 rth->rt_is_input = 1;
2039 RT_CACHE_STAT_INC(in_slow_tot);
2040 if (res->type == RTN_UNREACHABLE) {
2041 rth->dst.input= ip_error;
2042 rth->dst.error= -err;
2043 rth->rt_flags &= ~RTCF_LOCAL;
2047 struct fib_nh *nh = &FIB_RES_NH(*res);
2049 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2050 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2051 WARN_ON(rth->dst.input == lwtunnel_input);
2052 rth->dst.lwtstate->orig_input = rth->dst.input;
2053 rth->dst.input = lwtunnel_input;
2056 if (unlikely(!rt_cache_route(nh, rth)))
2057 rt_add_uncached_list(rth);
2059 skb_dst_set(skb, &rth->dst);
2064 RT_CACHE_STAT_INC(in_no_route);
2065 res->type = RTN_UNREACHABLE;
2071 * Do not cache martian addresses: they should be logged (RFC1812)
2073 martian_destination:
2074 RT_CACHE_STAT_INC(in_martian_dst);
2075 #ifdef CONFIG_IP_ROUTE_VERBOSE
2076 if (IN_DEV_LOG_MARTIANS(in_dev))
2077 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2078 &daddr, &saddr, dev->name);
2090 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2094 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2095 u8 tos, struct net_device *dev)
2097 struct fib_result res;
2100 tos &= IPTOS_RT_MASK;
2102 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2107 EXPORT_SYMBOL(ip_route_input_noref);
2109 /* called with rcu_read_lock held */
2110 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2111 u8 tos, struct net_device *dev, struct fib_result *res)
2113 /* Multicast recognition logic is moved from route cache to here.
2114 The problem was that too many Ethernet cards have broken/missing
2115 hardware multicast filters :-( As result the host on multicasting
2116 network acquires a lot of useless route cache entries, sort of
2117 SDR messages from all the world. Now we try to get rid of them.
2118 Really, provided software IP multicast filter is organized
2119 reasonably (at least, hashed), it does not result in a slowdown
2120 comparing with route cache reject entries.
2121 Note, that multicast routers are not affected, because
2122 route cache entry is created eventually.
2124 if (ipv4_is_multicast(daddr)) {
2125 struct in_device *in_dev = __in_dev_get_rcu(dev);
2130 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2131 ip_hdr(skb)->protocol);
2133 /* check l3 master if no match yet */
2134 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2135 struct in_device *l3_in_dev;
2137 l3_in_dev = __in_dev_get_rcu(skb->dev);
2139 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2140 ip_hdr(skb)->protocol);
2144 #ifdef CONFIG_IP_MROUTE
2146 (!ipv4_is_local_multicast(daddr) &&
2147 IN_DEV_MFORWARD(in_dev))
2150 err = ip_route_input_mc(skb, daddr, saddr,
2156 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2159 /* called with rcu_read_lock() */
2160 static struct rtable *__mkroute_output(const struct fib_result *res,
2161 const struct flowi4 *fl4, int orig_oif,
2162 struct net_device *dev_out,
2165 struct fib_info *fi = res->fi;
2166 struct fib_nh_exception *fnhe;
2167 struct in_device *in_dev;
2168 u16 type = res->type;
2172 in_dev = __in_dev_get_rcu(dev_out);
2174 return ERR_PTR(-EINVAL);
2176 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2177 if (ipv4_is_loopback(fl4->saddr) &&
2178 !(dev_out->flags & IFF_LOOPBACK) &&
2179 !netif_is_l3_master(dev_out))
2180 return ERR_PTR(-EINVAL);
2182 if (ipv4_is_lbcast(fl4->daddr))
2183 type = RTN_BROADCAST;
2184 else if (ipv4_is_multicast(fl4->daddr))
2185 type = RTN_MULTICAST;
2186 else if (ipv4_is_zeronet(fl4->daddr))
2187 return ERR_PTR(-EINVAL);
2189 if (dev_out->flags & IFF_LOOPBACK)
2190 flags |= RTCF_LOCAL;
2193 if (type == RTN_BROADCAST) {
2194 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2196 } else if (type == RTN_MULTICAST) {
2197 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2198 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2200 flags &= ~RTCF_LOCAL;
2203 /* If multicast route do not exist use
2204 * default one, but do not gateway in this case.
2207 if (fi && res->prefixlen < 4)
2209 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2210 (orig_oif != dev_out->ifindex)) {
2211 /* For local routes that require a particular output interface
2212 * we do not want to cache the result. Caching the result
2213 * causes incorrect behaviour when there are multiple source
2214 * addresses on the interface, the end result being that if the
2215 * intended recipient is waiting on that interface for the
2216 * packet he won't receive it because it will be delivered on
2217 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2218 * be set to the loopback interface as well.
2224 do_cache &= fi != NULL;
2226 struct rtable __rcu **prth;
2227 struct fib_nh *nh = &FIB_RES_NH(*res);
2229 fnhe = find_exception(nh, fl4->daddr);
2231 prth = &fnhe->fnhe_rth_output;
2232 rth = rcu_dereference(*prth);
2233 if (rth && rth->dst.expires &&
2234 time_after(jiffies, rth->dst.expires)) {
2235 ip_del_fnhe(nh, fl4->daddr);
2242 if (unlikely(fl4->flowi4_flags &
2243 FLOWI_FLAG_KNOWN_NH &&
2245 nh->nh_scope == RT_SCOPE_LINK))) {
2249 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2250 rth = rcu_dereference(*prth);
2253 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2258 rth = rt_dst_alloc(dev_out, flags, type,
2259 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2260 IN_DEV_CONF_GET(in_dev, NOXFRM),
2263 return ERR_PTR(-ENOBUFS);
2265 rth->rt_iif = orig_oif;
2267 RT_CACHE_STAT_INC(out_slow_tot);
2269 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2270 if (flags & RTCF_LOCAL &&
2271 !(dev_out->flags & IFF_LOOPBACK)) {
2272 rth->dst.output = ip_mc_output;
2273 RT_CACHE_STAT_INC(out_slow_mc);
2275 #ifdef CONFIG_IP_MROUTE
2276 if (type == RTN_MULTICAST) {
2277 if (IN_DEV_MFORWARD(in_dev) &&
2278 !ipv4_is_local_multicast(fl4->daddr)) {
2279 rth->dst.input = ip_mr_input;
2280 rth->dst.output = ip_mc_output;
2286 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2287 lwtunnel_set_redirect(&rth->dst);
2293 * Major route resolver routine.
2296 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2297 const struct sk_buff *skb)
2299 __u8 tos = RT_FL_TOS(fl4);
2300 struct fib_result res = {
2308 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2309 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2310 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2311 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2314 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2319 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2321 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2322 struct fib_result *res,
2323 const struct sk_buff *skb)
2325 struct net_device *dev_out = NULL;
2326 int orig_oif = fl4->flowi4_oif;
2327 unsigned int flags = 0;
2329 int err = -ENETUNREACH;
2332 rth = ERR_PTR(-EINVAL);
2333 if (ipv4_is_multicast(fl4->saddr) ||
2334 ipv4_is_lbcast(fl4->saddr) ||
2335 ipv4_is_zeronet(fl4->saddr))
2338 /* I removed check for oif == dev_out->oif here.
2339 It was wrong for two reasons:
2340 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2341 is assigned to multiple interfaces.
2342 2. Moreover, we are allowed to send packets with saddr
2343 of another iface. --ANK
2346 if (fl4->flowi4_oif == 0 &&
2347 (ipv4_is_multicast(fl4->daddr) ||
2348 ipv4_is_lbcast(fl4->daddr))) {
2349 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2350 dev_out = __ip_dev_find(net, fl4->saddr, false);
2354 /* Special hack: user can direct multicasts
2355 and limited broadcast via necessary interface
2356 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2357 This hack is not just for fun, it allows
2358 vic,vat and friends to work.
2359 They bind socket to loopback, set ttl to zero
2360 and expect that it will work.
2361 From the viewpoint of routing cache they are broken,
2362 because we are not allowed to build multicast path
2363 with loopback source addr (look, routing cache
2364 cannot know, that ttl is zero, so that packet
2365 will not leave this host and route is valid).
2366 Luckily, this hack is good workaround.
2369 fl4->flowi4_oif = dev_out->ifindex;
2373 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2374 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2375 if (!__ip_dev_find(net, fl4->saddr, false))
2381 if (fl4->flowi4_oif) {
2382 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2383 rth = ERR_PTR(-ENODEV);
2387 /* RACE: Check return value of inet_select_addr instead. */
2388 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2389 rth = ERR_PTR(-ENETUNREACH);
2392 if (ipv4_is_local_multicast(fl4->daddr) ||
2393 ipv4_is_lbcast(fl4->daddr) ||
2394 fl4->flowi4_proto == IPPROTO_IGMP) {
2396 fl4->saddr = inet_select_addr(dev_out, 0,
2401 if (ipv4_is_multicast(fl4->daddr))
2402 fl4->saddr = inet_select_addr(dev_out, 0,
2404 else if (!fl4->daddr)
2405 fl4->saddr = inet_select_addr(dev_out, 0,
2411 fl4->daddr = fl4->saddr;
2413 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2414 dev_out = net->loopback_dev;
2415 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2416 res->type = RTN_LOCAL;
2417 flags |= RTCF_LOCAL;
2421 err = fib_lookup(net, fl4, res, 0);
2425 if (fl4->flowi4_oif &&
2426 (ipv4_is_multicast(fl4->daddr) ||
2427 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2428 /* Apparently, routing tables are wrong. Assume,
2429 that the destination is on link.
2432 Because we are allowed to send to iface
2433 even if it has NO routes and NO assigned
2434 addresses. When oif is specified, routing
2435 tables are looked up with only one purpose:
2436 to catch if destination is gatewayed, rather than
2437 direct. Moreover, if MSG_DONTROUTE is set,
2438 we send packet, ignoring both routing tables
2439 and ifaddr state. --ANK
2442 We could make it even if oif is unknown,
2443 likely IPv6, but we do not.
2446 if (fl4->saddr == 0)
2447 fl4->saddr = inet_select_addr(dev_out, 0,
2449 res->type = RTN_UNICAST;
2456 if (res->type == RTN_LOCAL) {
2458 if (res->fi->fib_prefsrc)
2459 fl4->saddr = res->fi->fib_prefsrc;
2461 fl4->saddr = fl4->daddr;
2464 /* L3 master device is the loopback for that domain */
2465 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2468 /* make sure orig_oif points to fib result device even
2469 * though packet rx/tx happens over loopback or l3mdev
2471 orig_oif = FIB_RES_OIF(*res);
2473 fl4->flowi4_oif = dev_out->ifindex;
2474 flags |= RTCF_LOCAL;
2478 fib_select_path(net, res, fl4, skb);
2480 dev_out = FIB_RES_DEV(*res);
2481 fl4->flowi4_oif = dev_out->ifindex;
2485 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2491 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2496 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2498 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2500 return mtu ? : dst->dev->mtu;
2503 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2504 struct sk_buff *skb, u32 mtu)
2508 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2509 struct sk_buff *skb)
2513 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2519 static struct dst_ops ipv4_dst_blackhole_ops = {
2521 .check = ipv4_blackhole_dst_check,
2522 .mtu = ipv4_blackhole_mtu,
2523 .default_advmss = ipv4_default_advmss,
2524 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2525 .redirect = ipv4_rt_blackhole_redirect,
2526 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2527 .neigh_lookup = ipv4_neigh_lookup,
2530 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2532 struct rtable *ort = (struct rtable *) dst_orig;
2535 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2537 struct dst_entry *new = &rt->dst;
2540 new->input = dst_discard;
2541 new->output = dst_discard_out;
2543 new->dev = net->loopback_dev;
2547 rt->rt_is_input = ort->rt_is_input;
2548 rt->rt_iif = ort->rt_iif;
2549 rt->rt_pmtu = ort->rt_pmtu;
2550 rt->rt_mtu_locked = ort->rt_mtu_locked;
2552 rt->rt_genid = rt_genid_ipv4(net);
2553 rt->rt_flags = ort->rt_flags;
2554 rt->rt_type = ort->rt_type;
2555 rt->rt_gateway = ort->rt_gateway;
2556 rt->rt_uses_gateway = ort->rt_uses_gateway;
2558 INIT_LIST_HEAD(&rt->rt_uncached);
2561 dst_release(dst_orig);
2563 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2566 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2567 const struct sock *sk)
2569 struct rtable *rt = __ip_route_output_key(net, flp4);
2574 if (flp4->flowi4_proto)
2575 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2576 flowi4_to_flowi(flp4),
2581 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2583 /* called with rcu_read_lock held */
2584 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2585 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2588 struct rtable *rt = skb_rtable(skb);
2590 struct nlmsghdr *nlh;
2591 unsigned long expires = 0;
2593 u32 metrics[RTAX_MAX];
2595 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2599 r = nlmsg_data(nlh);
2600 r->rtm_family = AF_INET;
2601 r->rtm_dst_len = 32;
2603 r->rtm_tos = fl4->flowi4_tos;
2604 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2605 if (nla_put_u32(skb, RTA_TABLE, table_id))
2606 goto nla_put_failure;
2607 r->rtm_type = rt->rt_type;
2608 r->rtm_scope = RT_SCOPE_UNIVERSE;
2609 r->rtm_protocol = RTPROT_UNSPEC;
2610 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2611 if (rt->rt_flags & RTCF_NOTIFY)
2612 r->rtm_flags |= RTM_F_NOTIFY;
2613 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2614 r->rtm_flags |= RTCF_DOREDIRECT;
2616 if (nla_put_in_addr(skb, RTA_DST, dst))
2617 goto nla_put_failure;
2619 r->rtm_src_len = 32;
2620 if (nla_put_in_addr(skb, RTA_SRC, src))
2621 goto nla_put_failure;
2624 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2625 goto nla_put_failure;
2626 #ifdef CONFIG_IP_ROUTE_CLASSID
2627 if (rt->dst.tclassid &&
2628 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2629 goto nla_put_failure;
2631 if (!rt_is_input_route(rt) &&
2632 fl4->saddr != src) {
2633 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2634 goto nla_put_failure;
2636 if (rt->rt_uses_gateway &&
2637 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2638 goto nla_put_failure;
2640 expires = rt->dst.expires;
2642 unsigned long now = jiffies;
2644 if (time_before(now, expires))
2650 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2651 if (rt->rt_pmtu && expires)
2652 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2653 if (rt->rt_mtu_locked && expires)
2654 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2655 if (rtnetlink_put_metrics(skb, metrics) < 0)
2656 goto nla_put_failure;
2658 if (fl4->flowi4_mark &&
2659 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2660 goto nla_put_failure;
2662 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2663 nla_put_u32(skb, RTA_UID,
2664 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2665 goto nla_put_failure;
2667 error = rt->dst.error;
2669 if (rt_is_input_route(rt)) {
2670 #ifdef CONFIG_IP_MROUTE
2671 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2672 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2673 int err = ipmr_get_route(net, skb,
2674 fl4->saddr, fl4->daddr,
2680 goto nla_put_failure;
2684 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2685 goto nla_put_failure;
2688 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2689 goto nla_put_failure;
2691 nlmsg_end(skb, nlh);
2695 nlmsg_cancel(skb, nlh);
2699 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2700 struct netlink_ext_ack *extack)
2702 struct net *net = sock_net(in_skb->sk);
2704 struct nlattr *tb[RTA_MAX+1];
2705 struct fib_result res = {};
2706 struct rtable *rt = NULL;
2713 struct sk_buff *skb;
2714 u32 table_id = RT_TABLE_MAIN;
2717 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2722 rtm = nlmsg_data(nlh);
2724 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2730 /* Reserve room for dummy headers, this skb can pass
2731 through good chunk of routing engine.
2733 skb_reset_mac_header(skb);
2734 skb_reset_network_header(skb);
2736 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2737 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2738 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2739 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2741 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2743 uid = (iif ? INVALID_UID : current_uid());
2745 /* Bugfix: need to give ip_route_input enough of an IP header to
2748 ip_hdr(skb)->protocol = IPPROTO_UDP;
2749 ip_hdr(skb)->saddr = src;
2750 ip_hdr(skb)->daddr = dst;
2752 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2754 memset(&fl4, 0, sizeof(fl4));
2757 fl4.flowi4_tos = rtm->rtm_tos;
2758 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2759 fl4.flowi4_mark = mark;
2760 fl4.flowi4_uid = uid;
2765 struct net_device *dev;
2767 dev = dev_get_by_index_rcu(net, iif);
2773 skb->protocol = htons(ETH_P_IP);
2776 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2779 rt = skb_rtable(skb);
2780 if (err == 0 && rt->dst.error)
2781 err = -rt->dst.error;
2783 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2784 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2789 skb_dst_set(skb, &rt->dst);
2795 if (rtm->rtm_flags & RTM_F_NOTIFY)
2796 rt->rt_flags |= RTCF_NOTIFY;
2798 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2799 table_id = res.table ? res.table->tb_id : 0;
2801 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2803 err = fib_props[res.type].error;
2805 err = -EHOSTUNREACH;
2808 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2809 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2810 rt->rt_type, res.prefix, res.prefixlen,
2811 fl4.flowi4_tos, res.fi, 0);
2813 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2814 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2821 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2831 void ip_rt_multicast_event(struct in_device *in_dev)
2833 rt_cache_flush(dev_net(in_dev->dev));
2836 #ifdef CONFIG_SYSCTL
2837 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2838 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2839 static int ip_rt_gc_elasticity __read_mostly = 8;
2840 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
2842 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2843 void __user *buffer,
2844 size_t *lenp, loff_t *ppos)
2846 struct net *net = (struct net *)__ctl->extra1;
2849 rt_cache_flush(net);
2850 fnhe_genid_bump(net);
2857 static struct ctl_table ipv4_route_table[] = {
2859 .procname = "gc_thresh",
2860 .data = &ipv4_dst_ops.gc_thresh,
2861 .maxlen = sizeof(int),
2863 .proc_handler = proc_dointvec,
2866 .procname = "max_size",
2867 .data = &ip_rt_max_size,
2868 .maxlen = sizeof(int),
2870 .proc_handler = proc_dointvec,
2873 /* Deprecated. Use gc_min_interval_ms */
2875 .procname = "gc_min_interval",
2876 .data = &ip_rt_gc_min_interval,
2877 .maxlen = sizeof(int),
2879 .proc_handler = proc_dointvec_jiffies,
2882 .procname = "gc_min_interval_ms",
2883 .data = &ip_rt_gc_min_interval,
2884 .maxlen = sizeof(int),
2886 .proc_handler = proc_dointvec_ms_jiffies,
2889 .procname = "gc_timeout",
2890 .data = &ip_rt_gc_timeout,
2891 .maxlen = sizeof(int),
2893 .proc_handler = proc_dointvec_jiffies,
2896 .procname = "gc_interval",
2897 .data = &ip_rt_gc_interval,
2898 .maxlen = sizeof(int),
2900 .proc_handler = proc_dointvec_jiffies,
2903 .procname = "redirect_load",
2904 .data = &ip_rt_redirect_load,
2905 .maxlen = sizeof(int),
2907 .proc_handler = proc_dointvec,
2910 .procname = "redirect_number",
2911 .data = &ip_rt_redirect_number,
2912 .maxlen = sizeof(int),
2914 .proc_handler = proc_dointvec,
2917 .procname = "redirect_silence",
2918 .data = &ip_rt_redirect_silence,
2919 .maxlen = sizeof(int),
2921 .proc_handler = proc_dointvec,
2924 .procname = "error_cost",
2925 .data = &ip_rt_error_cost,
2926 .maxlen = sizeof(int),
2928 .proc_handler = proc_dointvec,
2931 .procname = "error_burst",
2932 .data = &ip_rt_error_burst,
2933 .maxlen = sizeof(int),
2935 .proc_handler = proc_dointvec,
2938 .procname = "gc_elasticity",
2939 .data = &ip_rt_gc_elasticity,
2940 .maxlen = sizeof(int),
2942 .proc_handler = proc_dointvec,
2945 .procname = "mtu_expires",
2946 .data = &ip_rt_mtu_expires,
2947 .maxlen = sizeof(int),
2949 .proc_handler = proc_dointvec_jiffies,
2952 .procname = "min_pmtu",
2953 .data = &ip_rt_min_pmtu,
2954 .maxlen = sizeof(int),
2956 .proc_handler = proc_dointvec_minmax,
2957 .extra1 = &ip_min_valid_pmtu,
2960 .procname = "min_adv_mss",
2961 .data = &ip_rt_min_advmss,
2962 .maxlen = sizeof(int),
2964 .proc_handler = proc_dointvec,
2969 static struct ctl_table ipv4_route_flush_table[] = {
2971 .procname = "flush",
2972 .maxlen = sizeof(int),
2974 .proc_handler = ipv4_sysctl_rtcache_flush,
2979 static __net_init int sysctl_route_net_init(struct net *net)
2981 struct ctl_table *tbl;
2983 tbl = ipv4_route_flush_table;
2984 if (!net_eq(net, &init_net)) {
2985 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2989 /* Don't export sysctls to unprivileged users */
2990 if (net->user_ns != &init_user_ns)
2991 tbl[0].procname = NULL;
2993 tbl[0].extra1 = net;
2995 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2996 if (!net->ipv4.route_hdr)
3001 if (tbl != ipv4_route_flush_table)
3007 static __net_exit void sysctl_route_net_exit(struct net *net)
3009 struct ctl_table *tbl;
3011 tbl = net->ipv4.route_hdr->ctl_table_arg;
3012 unregister_net_sysctl_table(net->ipv4.route_hdr);
3013 BUG_ON(tbl == ipv4_route_flush_table);
3017 static __net_initdata struct pernet_operations sysctl_route_ops = {
3018 .init = sysctl_route_net_init,
3019 .exit = sysctl_route_net_exit,
3023 static __net_init int rt_genid_init(struct net *net)
3025 atomic_set(&net->ipv4.rt_genid, 0);
3026 atomic_set(&net->fnhe_genid, 0);
3027 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3031 static __net_initdata struct pernet_operations rt_genid_ops = {
3032 .init = rt_genid_init,
3035 static int __net_init ipv4_inetpeer_init(struct net *net)
3037 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3041 inet_peer_base_init(bp);
3042 net->ipv4.peers = bp;
3046 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3048 struct inet_peer_base *bp = net->ipv4.peers;
3050 net->ipv4.peers = NULL;
3051 inetpeer_invalidate_tree(bp);
3055 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3056 .init = ipv4_inetpeer_init,
3057 .exit = ipv4_inetpeer_exit,
3060 #ifdef CONFIG_IP_ROUTE_CLASSID
3061 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3062 #endif /* CONFIG_IP_ROUTE_CLASSID */
3064 int __init ip_rt_init(void)
3068 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3070 panic("IP: failed to allocate ip_idents\n");
3072 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3074 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3076 panic("IP: failed to allocate ip_tstamps\n");
3078 for_each_possible_cpu(cpu) {
3079 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3081 INIT_LIST_HEAD(&ul->head);
3082 spin_lock_init(&ul->lock);
3084 #ifdef CONFIG_IP_ROUTE_CLASSID
3085 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3087 panic("IP: failed to allocate ip_rt_acct\n");
3090 ipv4_dst_ops.kmem_cachep =
3091 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3092 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3094 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3096 if (dst_entries_init(&ipv4_dst_ops) < 0)
3097 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3099 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3100 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3102 ipv4_dst_ops.gc_thresh = ~0;
3103 ip_rt_max_size = INT_MAX;
3108 if (ip_rt_proc_init())
3109 pr_err("Unable to create route proc files\n");
3114 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3115 RTNL_FLAG_DOIT_UNLOCKED);
3117 #ifdef CONFIG_SYSCTL
3118 register_pernet_subsys(&sysctl_route_ops);
3120 register_pernet_subsys(&rt_genid_ops);
3121 register_pernet_subsys(&ipv4_inetpeer_ops);
3125 #ifdef CONFIG_SYSCTL
3127 * We really need to sanitize the damn ipv4 init order, then all
3128 * this nonsense will go away.
3130 void __init ip_static_sysctl_init(void)
3132 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);