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 <asm/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 #define RT_FL_TOS(oldflp4) \
118 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
120 #define RT_GC_TIMEOUT (300*HZ)
122 static int ip_rt_max_size;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
129 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly = 256;
132 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
134 * Interface to generic destination cache.
137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
138 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
139 static unsigned int ipv4_mtu(const struct dst_entry *dst);
140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
141 static void ipv4_link_failure(struct sk_buff *skb);
142 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
143 struct sk_buff *skb, u32 mtu);
144 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb);
146 static void ipv4_dst_destroy(struct dst_entry *dst);
148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
158 static struct dst_ops ipv4_dst_ops = {
160 .check = ipv4_dst_check,
161 .default_advmss = ipv4_default_advmss,
163 .cow_metrics = ipv4_cow_metrics,
164 .destroy = ipv4_dst_destroy,
165 .negative_advice = ipv4_negative_advice,
166 .link_failure = ipv4_link_failure,
167 .update_pmtu = ip_rt_update_pmtu,
168 .redirect = ip_do_redirect,
169 .local_out = __ip_local_out,
170 .neigh_lookup = ipv4_neigh_lookup,
173 #define ECN_OR_COST(class) TC_PRIO_##class
175 const __u8 ip_tos2prio[16] = {
177 ECN_OR_COST(BESTEFFORT),
179 ECN_OR_COST(BESTEFFORT),
185 ECN_OR_COST(INTERACTIVE),
187 ECN_OR_COST(INTERACTIVE),
188 TC_PRIO_INTERACTIVE_BULK,
189 ECN_OR_COST(INTERACTIVE_BULK),
190 TC_PRIO_INTERACTIVE_BULK,
191 ECN_OR_COST(INTERACTIVE_BULK)
193 EXPORT_SYMBOL(ip_tos2prio);
195 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
196 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
198 #ifdef CONFIG_PROC_FS
199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
203 return SEQ_START_TOKEN;
206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 if (v == SEQ_START_TOKEN)
219 seq_printf(seq, "%-127s\n",
220 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
221 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
226 static const struct seq_operations rt_cache_seq_ops = {
227 .start = rt_cache_seq_start,
228 .next = rt_cache_seq_next,
229 .stop = rt_cache_seq_stop,
230 .show = rt_cache_seq_show,
233 static int rt_cache_seq_open(struct inode *inode, struct file *file)
235 return seq_open(file, &rt_cache_seq_ops);
238 static const struct file_operations rt_cache_seq_fops = {
239 .owner = THIS_MODULE,
240 .open = rt_cache_seq_open,
243 .release = seq_release,
247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
252 return SEQ_START_TOKEN;
254 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
255 if (!cpu_possible(cpu))
258 return &per_cpu(rt_cache_stat, cpu);
263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
267 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
268 if (!cpu_possible(cpu))
271 return &per_cpu(rt_cache_stat, cpu);
277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
284 struct rt_cache_stat *st = v;
286 if (v == SEQ_START_TOKEN) {
287 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");
291 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
292 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
293 dst_entries_get_slow(&ipv4_dst_ops),
306 0, /* st->gc_total */
307 0, /* st->gc_ignored */
308 0, /* st->gc_goal_miss */
309 0, /* st->gc_dst_overflow */
310 0, /* st->in_hlist_search */
311 0 /* st->out_hlist_search */
316 static const struct seq_operations rt_cpu_seq_ops = {
317 .start = rt_cpu_seq_start,
318 .next = rt_cpu_seq_next,
319 .stop = rt_cpu_seq_stop,
320 .show = rt_cpu_seq_show,
324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
326 return seq_open(file, &rt_cpu_seq_ops);
329 static const struct file_operations rt_cpu_seq_fops = {
330 .owner = THIS_MODULE,
331 .open = rt_cpu_seq_open,
334 .release = seq_release,
337 #ifdef CONFIG_IP_ROUTE_CLASSID
338 static int rt_acct_proc_show(struct seq_file *m, void *v)
340 struct ip_rt_acct *dst, *src;
343 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
347 for_each_possible_cpu(i) {
348 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
349 for (j = 0; j < 256; j++) {
350 dst[j].o_bytes += src[j].o_bytes;
351 dst[j].o_packets += src[j].o_packets;
352 dst[j].i_bytes += src[j].i_bytes;
353 dst[j].i_packets += src[j].i_packets;
357 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
362 static int rt_acct_proc_open(struct inode *inode, struct file *file)
364 return single_open(file, rt_acct_proc_show, NULL);
367 static const struct file_operations rt_acct_proc_fops = {
368 .owner = THIS_MODULE,
369 .open = rt_acct_proc_open,
372 .release = single_release,
376 static int __net_init ip_rt_do_proc_init(struct net *net)
378 struct proc_dir_entry *pde;
380 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
385 pde = proc_create("rt_cache", S_IRUGO,
386 net->proc_net_stat, &rt_cpu_seq_fops);
390 #ifdef CONFIG_IP_ROUTE_CLASSID
391 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
397 #ifdef CONFIG_IP_ROUTE_CLASSID
399 remove_proc_entry("rt_cache", net->proc_net_stat);
402 remove_proc_entry("rt_cache", net->proc_net);
407 static void __net_exit ip_rt_do_proc_exit(struct net *net)
409 remove_proc_entry("rt_cache", net->proc_net_stat);
410 remove_proc_entry("rt_cache", net->proc_net);
411 #ifdef CONFIG_IP_ROUTE_CLASSID
412 remove_proc_entry("rt_acct", net->proc_net);
416 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
417 .init = ip_rt_do_proc_init,
418 .exit = ip_rt_do_proc_exit,
421 static int __init ip_rt_proc_init(void)
423 return register_pernet_subsys(&ip_rt_proc_ops);
427 static inline int ip_rt_proc_init(void)
431 #endif /* CONFIG_PROC_FS */
433 static inline bool rt_is_expired(const struct rtable *rth)
435 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
438 void rt_cache_flush(struct net *net)
440 rt_genid_bump_ipv4(net);
443 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
447 struct net_device *dev = dst->dev;
448 const __be32 *pkey = daddr;
449 const struct rtable *rt;
452 rt = (const struct rtable *) dst;
454 pkey = (const __be32 *) &rt->rt_gateway;
456 pkey = &ip_hdr(skb)->daddr;
458 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
461 return neigh_create(&arp_tbl, pkey, dev);
464 #define IP_IDENTS_SZ 2048u
466 static atomic_t *ip_idents __read_mostly;
467 static u32 *ip_tstamps __read_mostly;
469 /* In order to protect privacy, we add a perturbation to identifiers
470 * if one generator is seldom used. This makes hard for an attacker
471 * to infer how many packets were sent between two points in time.
473 u32 ip_idents_reserve(u32 hash, int segs)
475 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
476 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
477 u32 old = ACCESS_ONCE(*p_tstamp);
478 u32 now = (u32)jiffies;
481 if (old != now && cmpxchg(p_tstamp, old, now) == old)
482 delta = prandom_u32_max(now - old);
484 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
486 old = (u32)atomic_read(p_id);
487 new = old + delta + segs;
488 } while (atomic_cmpxchg(p_id, old, new) != old);
492 EXPORT_SYMBOL(ip_idents_reserve);
494 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
496 static u32 ip_idents_hashrnd __read_mostly;
499 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
501 hash = jhash_3words((__force u32)iph->daddr,
502 (__force u32)iph->saddr,
503 iph->protocol ^ net_hash_mix(net),
505 id = ip_idents_reserve(hash, segs);
508 EXPORT_SYMBOL(__ip_select_ident);
510 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
511 const struct iphdr *iph,
513 u8 prot, u32 mark, int flow_flags)
516 const struct inet_sock *inet = inet_sk(sk);
518 oif = sk->sk_bound_dev_if;
520 tos = RT_CONN_FLAGS(sk);
521 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
523 flowi4_init_output(fl4, oif, mark, tos,
524 RT_SCOPE_UNIVERSE, prot,
526 iph->daddr, iph->saddr, 0, 0);
529 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
530 const struct sock *sk)
532 const struct iphdr *iph = ip_hdr(skb);
533 int oif = skb->dev->ifindex;
534 u8 tos = RT_TOS(iph->tos);
535 u8 prot = iph->protocol;
536 u32 mark = skb->mark;
538 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
541 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
543 const struct inet_sock *inet = inet_sk(sk);
544 const struct ip_options_rcu *inet_opt;
545 __be32 daddr = inet->inet_daddr;
548 inet_opt = rcu_dereference(inet->inet_opt);
549 if (inet_opt && inet_opt->opt.srr)
550 daddr = inet_opt->opt.faddr;
551 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
552 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
553 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
554 inet_sk_flowi_flags(sk),
555 daddr, inet->inet_saddr, 0, 0);
559 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
560 const struct sk_buff *skb)
563 build_skb_flow_key(fl4, skb, sk);
565 build_sk_flow_key(fl4, sk);
568 static inline void rt_free(struct rtable *rt)
570 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
573 static DEFINE_SPINLOCK(fnhe_lock);
575 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
579 rt = rcu_dereference(fnhe->fnhe_rth_input);
581 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
584 rt = rcu_dereference(fnhe->fnhe_rth_output);
586 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
591 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
593 struct fib_nh_exception *fnhe, *oldest;
595 oldest = rcu_dereference(hash->chain);
596 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
597 fnhe = rcu_dereference(fnhe->fnhe_next)) {
598 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
601 fnhe_flush_routes(oldest);
605 static inline u32 fnhe_hashfun(__be32 daddr)
607 static u32 fnhe_hashrnd __read_mostly;
610 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
611 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
612 return hash_32(hval, FNHE_HASH_SHIFT);
615 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
617 rt->rt_pmtu = fnhe->fnhe_pmtu;
618 rt->dst.expires = fnhe->fnhe_expires;
621 rt->rt_flags |= RTCF_REDIRECTED;
622 rt->rt_gateway = fnhe->fnhe_gw;
623 rt->rt_uses_gateway = 1;
627 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
628 u32 pmtu, unsigned long expires)
630 struct fnhe_hash_bucket *hash;
631 struct fib_nh_exception *fnhe;
635 u32 hval = fnhe_hashfun(daddr);
637 spin_lock_bh(&fnhe_lock);
639 hash = rcu_dereference(nh->nh_exceptions);
641 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
644 rcu_assign_pointer(nh->nh_exceptions, hash);
650 for (fnhe = rcu_dereference(hash->chain); fnhe;
651 fnhe = rcu_dereference(fnhe->fnhe_next)) {
652 if (fnhe->fnhe_daddr == daddr)
661 fnhe->fnhe_pmtu = pmtu;
662 fnhe->fnhe_expires = max(1UL, expires);
664 /* Update all cached dsts too */
665 rt = rcu_dereference(fnhe->fnhe_rth_input);
667 fill_route_from_fnhe(rt, fnhe);
668 rt = rcu_dereference(fnhe->fnhe_rth_output);
670 fill_route_from_fnhe(rt, fnhe);
672 if (depth > FNHE_RECLAIM_DEPTH)
673 fnhe = fnhe_oldest(hash);
675 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
679 fnhe->fnhe_next = hash->chain;
680 rcu_assign_pointer(hash->chain, fnhe);
682 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
683 fnhe->fnhe_daddr = daddr;
685 fnhe->fnhe_pmtu = pmtu;
686 fnhe->fnhe_expires = expires;
688 /* Exception created; mark the cached routes for the nexthop
689 * stale, so anyone caching it rechecks if this exception
692 rt = rcu_dereference(nh->nh_rth_input);
694 rt->dst.obsolete = DST_OBSOLETE_KILL;
696 for_each_possible_cpu(i) {
697 struct rtable __rcu **prt;
698 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
699 rt = rcu_dereference(*prt);
701 rt->dst.obsolete = DST_OBSOLETE_KILL;
705 fnhe->fnhe_stamp = jiffies;
708 spin_unlock_bh(&fnhe_lock);
711 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
714 __be32 new_gw = icmp_hdr(skb)->un.gateway;
715 __be32 old_gw = ip_hdr(skb)->saddr;
716 struct net_device *dev = skb->dev;
717 struct in_device *in_dev;
718 struct fib_result res;
722 switch (icmp_hdr(skb)->code & 7) {
724 case ICMP_REDIR_NETTOS:
725 case ICMP_REDIR_HOST:
726 case ICMP_REDIR_HOSTTOS:
733 if (rt->rt_gateway != old_gw)
736 in_dev = __in_dev_get_rcu(dev);
741 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
742 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
743 ipv4_is_zeronet(new_gw))
744 goto reject_redirect;
746 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
747 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
748 goto reject_redirect;
749 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
750 goto reject_redirect;
752 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
753 goto reject_redirect;
756 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
758 if (!(n->nud_state & NUD_VALID)) {
759 neigh_event_send(n, NULL);
761 if (fib_lookup(net, fl4, &res, 0) == 0) {
762 struct fib_nh *nh = &FIB_RES_NH(res);
764 update_or_create_fnhe(nh, fl4->daddr, new_gw,
765 0, jiffies + ip_rt_gc_timeout);
768 rt->dst.obsolete = DST_OBSOLETE_KILL;
769 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
776 #ifdef CONFIG_IP_ROUTE_VERBOSE
777 if (IN_DEV_LOG_MARTIANS(in_dev)) {
778 const struct iphdr *iph = (const struct iphdr *) skb->data;
779 __be32 daddr = iph->daddr;
780 __be32 saddr = iph->saddr;
782 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
783 " Advised path = %pI4 -> %pI4\n",
784 &old_gw, dev->name, &new_gw,
791 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
795 const struct iphdr *iph = (const struct iphdr *) skb->data;
796 int oif = skb->dev->ifindex;
797 u8 tos = RT_TOS(iph->tos);
798 u8 prot = iph->protocol;
799 u32 mark = skb->mark;
801 rt = (struct rtable *) dst;
803 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
804 __ip_do_redirect(rt, skb, &fl4, true);
807 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
809 struct rtable *rt = (struct rtable *)dst;
810 struct dst_entry *ret = dst;
813 if (dst->obsolete > 0) {
816 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
827 * 1. The first ip_rt_redirect_number redirects are sent
828 * with exponential backoff, then we stop sending them at all,
829 * assuming that the host ignores our redirects.
830 * 2. If we did not see packets requiring redirects
831 * during ip_rt_redirect_silence, we assume that the host
832 * forgot redirected route and start to send redirects again.
834 * This algorithm is much cheaper and more intelligent than dumb load limiting
837 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
838 * and "frag. need" (breaks PMTU discovery) in icmp.c.
841 void ip_rt_send_redirect(struct sk_buff *skb)
843 struct rtable *rt = skb_rtable(skb);
844 struct in_device *in_dev;
845 struct inet_peer *peer;
851 in_dev = __in_dev_get_rcu(rt->dst.dev);
852 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
856 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
857 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
860 net = dev_net(rt->dst.dev);
861 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
863 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
864 rt_nexthop(rt, ip_hdr(skb)->daddr));
868 /* No redirected packets during ip_rt_redirect_silence;
869 * reset the algorithm.
871 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
872 peer->rate_tokens = 0;
874 /* Too many ignored redirects; do not send anything
875 * set dst.rate_last to the last seen redirected packet.
877 if (peer->rate_tokens >= ip_rt_redirect_number) {
878 peer->rate_last = jiffies;
882 /* Check for load limit; set rate_last to the latest sent
885 if (peer->rate_tokens == 0 ||
888 (ip_rt_redirect_load << peer->rate_tokens)))) {
889 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
891 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
892 peer->rate_last = jiffies;
894 #ifdef CONFIG_IP_ROUTE_VERBOSE
896 peer->rate_tokens == ip_rt_redirect_number)
897 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
898 &ip_hdr(skb)->saddr, inet_iif(skb),
899 &ip_hdr(skb)->daddr, &gw);
906 static int ip_error(struct sk_buff *skb)
908 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
909 struct rtable *rt = skb_rtable(skb);
910 struct inet_peer *peer;
916 /* IP on this device is disabled. */
920 net = dev_net(rt->dst.dev);
921 if (!IN_DEV_FORWARD(in_dev)) {
922 switch (rt->dst.error) {
924 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
928 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
934 switch (rt->dst.error) {
939 code = ICMP_HOST_UNREACH;
942 code = ICMP_NET_UNREACH;
943 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
946 code = ICMP_PKT_FILTERED;
950 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
951 l3mdev_master_ifindex(skb->dev), 1);
956 peer->rate_tokens += now - peer->rate_last;
957 if (peer->rate_tokens > ip_rt_error_burst)
958 peer->rate_tokens = ip_rt_error_burst;
959 peer->rate_last = now;
960 if (peer->rate_tokens >= ip_rt_error_cost)
961 peer->rate_tokens -= ip_rt_error_cost;
967 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
973 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
975 struct dst_entry *dst = &rt->dst;
976 struct fib_result res;
978 if (dst_metric_locked(dst, RTAX_MTU))
981 if (ipv4_mtu(dst) < mtu)
984 if (mtu < ip_rt_min_pmtu)
985 mtu = ip_rt_min_pmtu;
987 if (rt->rt_pmtu == mtu &&
988 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
992 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
993 struct fib_nh *nh = &FIB_RES_NH(res);
995 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
996 jiffies + ip_rt_mtu_expires);
1001 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1002 struct sk_buff *skb, u32 mtu)
1004 struct rtable *rt = (struct rtable *) dst;
1007 ip_rt_build_flow_key(&fl4, sk, skb);
1008 __ip_rt_update_pmtu(rt, &fl4, mtu);
1011 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1012 int oif, u32 mark, u8 protocol, int flow_flags)
1014 const struct iphdr *iph = (const struct iphdr *) skb->data;
1019 mark = IP4_REPLY_MARK(net, skb->mark);
1021 __build_flow_key(&fl4, NULL, iph, oif,
1022 RT_TOS(iph->tos), protocol, mark, flow_flags);
1023 rt = __ip_route_output_key(net, &fl4);
1025 __ip_rt_update_pmtu(rt, &fl4, mtu);
1029 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1031 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1033 const struct iphdr *iph = (const struct iphdr *) skb->data;
1037 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1039 if (!fl4.flowi4_mark)
1040 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1042 rt = __ip_route_output_key(sock_net(sk), &fl4);
1044 __ip_rt_update_pmtu(rt, &fl4, mtu);
1049 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1051 const struct iphdr *iph = (const struct iphdr *) skb->data;
1054 struct dst_entry *odst = NULL;
1059 if (!ip_sk_accept_pmtu(sk))
1062 odst = sk_dst_get(sk);
1064 if (sock_owned_by_user(sk) || !odst) {
1065 __ipv4_sk_update_pmtu(skb, sk, mtu);
1069 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1071 rt = (struct rtable *)odst;
1072 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1073 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1080 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1082 if (!dst_check(&rt->dst, 0)) {
1084 dst_release(&rt->dst);
1086 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1094 sk_dst_set(sk, &rt->dst);
1100 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1102 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1103 int oif, u32 mark, u8 protocol, int flow_flags)
1105 const struct iphdr *iph = (const struct iphdr *) skb->data;
1109 __build_flow_key(&fl4, NULL, iph, oif,
1110 RT_TOS(iph->tos), protocol, mark, flow_flags);
1111 rt = __ip_route_output_key(net, &fl4);
1113 __ip_do_redirect(rt, skb, &fl4, false);
1117 EXPORT_SYMBOL_GPL(ipv4_redirect);
1119 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1121 const struct iphdr *iph = (const struct iphdr *) skb->data;
1125 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1126 rt = __ip_route_output_key(sock_net(sk), &fl4);
1128 __ip_do_redirect(rt, skb, &fl4, false);
1132 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1134 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1136 struct rtable *rt = (struct rtable *) dst;
1138 /* All IPV4 dsts are created with ->obsolete set to the value
1139 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1140 * into this function always.
1142 * When a PMTU/redirect information update invalidates a route,
1143 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1144 * DST_OBSOLETE_DEAD by dst_free().
1146 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1151 static void ipv4_link_failure(struct sk_buff *skb)
1155 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1157 rt = skb_rtable(skb);
1159 dst_set_expires(&rt->dst, 0);
1162 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1164 pr_debug("%s: %pI4 -> %pI4, %s\n",
1165 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1166 skb->dev ? skb->dev->name : "?");
1173 We do not cache source address of outgoing interface,
1174 because it is used only by IP RR, TS and SRR options,
1175 so that it out of fast path.
1177 BTW remember: "addr" is allowed to be not aligned
1181 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1185 if (rt_is_output_route(rt))
1186 src = ip_hdr(skb)->saddr;
1188 struct fib_result res;
1194 memset(&fl4, 0, sizeof(fl4));
1195 fl4.daddr = iph->daddr;
1196 fl4.saddr = iph->saddr;
1197 fl4.flowi4_tos = RT_TOS(iph->tos);
1198 fl4.flowi4_oif = rt->dst.dev->ifindex;
1199 fl4.flowi4_iif = skb->dev->ifindex;
1200 fl4.flowi4_mark = skb->mark;
1203 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1204 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1206 src = inet_select_addr(rt->dst.dev,
1207 rt_nexthop(rt, iph->daddr),
1211 memcpy(addr, &src, 4);
1214 #ifdef CONFIG_IP_ROUTE_CLASSID
1215 static void set_class_tag(struct rtable *rt, u32 tag)
1217 if (!(rt->dst.tclassid & 0xFFFF))
1218 rt->dst.tclassid |= tag & 0xFFFF;
1219 if (!(rt->dst.tclassid & 0xFFFF0000))
1220 rt->dst.tclassid |= tag & 0xFFFF0000;
1224 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1226 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1229 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1231 if (advmss > 65535 - 40)
1232 advmss = 65535 - 40;
1237 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1239 const struct rtable *rt = (const struct rtable *) dst;
1240 unsigned int mtu = rt->rt_pmtu;
1242 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1243 mtu = dst_metric_raw(dst, RTAX_MTU);
1248 mtu = dst->dev->mtu;
1250 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1251 if (rt->rt_uses_gateway && mtu > 576)
1255 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1257 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1260 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1262 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1263 struct fib_nh_exception *fnhe;
1269 hval = fnhe_hashfun(daddr);
1271 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1272 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1273 if (fnhe->fnhe_daddr == daddr)
1279 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1284 spin_lock_bh(&fnhe_lock);
1286 if (daddr == fnhe->fnhe_daddr) {
1287 struct rtable __rcu **porig;
1288 struct rtable *orig;
1289 int genid = fnhe_genid(dev_net(rt->dst.dev));
1291 if (rt_is_input_route(rt))
1292 porig = &fnhe->fnhe_rth_input;
1294 porig = &fnhe->fnhe_rth_output;
1295 orig = rcu_dereference(*porig);
1297 if (fnhe->fnhe_genid != genid) {
1298 fnhe->fnhe_genid = genid;
1300 fnhe->fnhe_pmtu = 0;
1301 fnhe->fnhe_expires = 0;
1302 fnhe_flush_routes(fnhe);
1305 fill_route_from_fnhe(rt, fnhe);
1306 if (!rt->rt_gateway)
1307 rt->rt_gateway = daddr;
1309 if (!(rt->dst.flags & DST_NOCACHE)) {
1310 rcu_assign_pointer(*porig, rt);
1316 fnhe->fnhe_stamp = jiffies;
1318 spin_unlock_bh(&fnhe_lock);
1323 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1325 struct rtable *orig, *prev, **p;
1328 if (rt_is_input_route(rt)) {
1329 p = (struct rtable **)&nh->nh_rth_input;
1331 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1335 prev = cmpxchg(p, orig, rt);
1345 struct uncached_list {
1347 struct list_head head;
1350 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1352 static void rt_add_uncached_list(struct rtable *rt)
1354 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1356 rt->rt_uncached_list = ul;
1358 spin_lock_bh(&ul->lock);
1359 list_add_tail(&rt->rt_uncached, &ul->head);
1360 spin_unlock_bh(&ul->lock);
1363 static void ipv4_dst_destroy(struct dst_entry *dst)
1365 struct rtable *rt = (struct rtable *) dst;
1367 if (!list_empty(&rt->rt_uncached)) {
1368 struct uncached_list *ul = rt->rt_uncached_list;
1370 spin_lock_bh(&ul->lock);
1371 list_del(&rt->rt_uncached);
1372 spin_unlock_bh(&ul->lock);
1376 void rt_flush_dev(struct net_device *dev)
1378 struct net *net = dev_net(dev);
1382 for_each_possible_cpu(cpu) {
1383 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1385 spin_lock_bh(&ul->lock);
1386 list_for_each_entry(rt, &ul->head, rt_uncached) {
1387 if (rt->dst.dev != dev)
1389 rt->dst.dev = net->loopback_dev;
1390 dev_hold(rt->dst.dev);
1393 spin_unlock_bh(&ul->lock);
1397 static bool rt_cache_valid(const struct rtable *rt)
1400 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1404 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1405 const struct fib_result *res,
1406 struct fib_nh_exception *fnhe,
1407 struct fib_info *fi, u16 type, u32 itag)
1409 bool cached = false;
1412 struct fib_nh *nh = &FIB_RES_NH(*res);
1414 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1415 rt->rt_gateway = nh->nh_gw;
1416 rt->rt_uses_gateway = 1;
1418 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1419 #ifdef CONFIG_IP_ROUTE_CLASSID
1420 rt->dst.tclassid = nh->nh_tclassid;
1422 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1424 cached = rt_bind_exception(rt, fnhe, daddr);
1425 else if (!(rt->dst.flags & DST_NOCACHE))
1426 cached = rt_cache_route(nh, rt);
1427 if (unlikely(!cached)) {
1428 /* Routes we intend to cache in nexthop exception or
1429 * FIB nexthop have the DST_NOCACHE bit clear.
1430 * However, if we are unsuccessful at storing this
1431 * route into the cache we really need to set it.
1433 rt->dst.flags |= DST_NOCACHE;
1434 if (!rt->rt_gateway)
1435 rt->rt_gateway = daddr;
1436 rt_add_uncached_list(rt);
1439 rt_add_uncached_list(rt);
1441 #ifdef CONFIG_IP_ROUTE_CLASSID
1442 #ifdef CONFIG_IP_MULTIPLE_TABLES
1443 set_class_tag(rt, res->tclassid);
1445 set_class_tag(rt, itag);
1449 struct rtable *rt_dst_alloc(struct net_device *dev,
1450 unsigned int flags, u16 type,
1451 bool nopolicy, bool noxfrm, bool will_cache)
1455 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1456 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1457 (nopolicy ? DST_NOPOLICY : 0) |
1458 (noxfrm ? DST_NOXFRM : 0));
1461 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1462 rt->rt_flags = flags;
1464 rt->rt_is_input = 0;
1468 rt->rt_uses_gateway = 0;
1469 rt->rt_table_id = 0;
1470 INIT_LIST_HEAD(&rt->rt_uncached);
1472 rt->dst.output = ip_output;
1473 if (flags & RTCF_LOCAL)
1474 rt->dst.input = ip_local_deliver;
1479 EXPORT_SYMBOL(rt_dst_alloc);
1481 /* called in rcu_read_lock() section */
1482 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1483 u8 tos, struct net_device *dev, int our)
1486 struct in_device *in_dev = __in_dev_get_rcu(dev);
1487 unsigned int flags = RTCF_MULTICAST;
1491 /* Primary sanity checks. */
1496 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1497 skb->protocol != htons(ETH_P_IP))
1500 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1503 if (ipv4_is_zeronet(saddr)) {
1504 if (!ipv4_is_local_multicast(daddr))
1507 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1513 flags |= RTCF_LOCAL;
1515 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1516 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1520 #ifdef CONFIG_IP_ROUTE_CLASSID
1521 rth->dst.tclassid = itag;
1523 rth->dst.output = ip_rt_bug;
1524 rth->rt_is_input= 1;
1526 #ifdef CONFIG_IP_MROUTE
1527 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1528 rth->dst.input = ip_mr_input;
1530 RT_CACHE_STAT_INC(in_slow_mc);
1532 skb_dst_set(skb, &rth->dst);
1544 static void ip_handle_martian_source(struct net_device *dev,
1545 struct in_device *in_dev,
1546 struct sk_buff *skb,
1550 RT_CACHE_STAT_INC(in_martian_src);
1551 #ifdef CONFIG_IP_ROUTE_VERBOSE
1552 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1554 * RFC1812 recommendation, if source is martian,
1555 * the only hint is MAC header.
1557 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1558 &daddr, &saddr, dev->name);
1559 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1560 print_hex_dump(KERN_WARNING, "ll header: ",
1561 DUMP_PREFIX_OFFSET, 16, 1,
1562 skb_mac_header(skb),
1563 dev->hard_header_len, true);
1569 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1571 struct fnhe_hash_bucket *hash;
1572 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1573 u32 hval = fnhe_hashfun(daddr);
1575 spin_lock_bh(&fnhe_lock);
1577 hash = rcu_dereference_protected(nh->nh_exceptions,
1578 lockdep_is_held(&fnhe_lock));
1581 fnhe_p = &hash->chain;
1582 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1584 if (fnhe->fnhe_daddr == daddr) {
1585 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1586 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1587 fnhe_flush_routes(fnhe);
1588 kfree_rcu(fnhe, rcu);
1591 fnhe_p = &fnhe->fnhe_next;
1592 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1593 lockdep_is_held(&fnhe_lock));
1596 spin_unlock_bh(&fnhe_lock);
1599 /* called in rcu_read_lock() section */
1600 static int __mkroute_input(struct sk_buff *skb,
1601 const struct fib_result *res,
1602 struct in_device *in_dev,
1603 __be32 daddr, __be32 saddr, u32 tos)
1605 struct fib_nh_exception *fnhe;
1608 struct in_device *out_dev;
1612 /* get a working reference to the output device */
1613 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1615 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1619 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1620 in_dev->dev, in_dev, &itag);
1622 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1628 do_cache = res->fi && !itag;
1629 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1630 skb->protocol == htons(ETH_P_IP) &&
1631 (IN_DEV_SHARED_MEDIA(out_dev) ||
1632 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1633 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1635 if (skb->protocol != htons(ETH_P_IP)) {
1636 /* Not IP (i.e. ARP). Do not create route, if it is
1637 * invalid for proxy arp. DNAT routes are always valid.
1639 * Proxy arp feature have been extended to allow, ARP
1640 * replies back to the same interface, to support
1641 * Private VLAN switch technologies. See arp.c.
1643 if (out_dev == in_dev &&
1644 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1650 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1653 rth = rcu_dereference(fnhe->fnhe_rth_input);
1654 if (rth && rth->dst.expires &&
1655 time_after(jiffies, rth->dst.expires)) {
1656 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1663 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1666 if (rt_cache_valid(rth)) {
1667 skb_dst_set_noref(skb, &rth->dst);
1672 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1673 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1674 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1680 rth->rt_is_input = 1;
1682 rth->rt_table_id = res->table->tb_id;
1683 RT_CACHE_STAT_INC(in_slow_tot);
1685 rth->dst.input = ip_forward;
1687 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1688 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1689 rth->dst.lwtstate->orig_output = rth->dst.output;
1690 rth->dst.output = lwtunnel_output;
1692 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1693 rth->dst.lwtstate->orig_input = rth->dst.input;
1694 rth->dst.input = lwtunnel_input;
1696 skb_dst_set(skb, &rth->dst);
1703 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1705 /* To make ICMP packets follow the right flow, the multipath hash is
1706 * calculated from the inner IP addresses in reverse order.
1708 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1710 const struct iphdr *outer_iph = ip_hdr(skb);
1711 struct icmphdr _icmph;
1712 const struct icmphdr *icmph;
1713 struct iphdr _inner_iph;
1714 const struct iphdr *inner_iph;
1716 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1719 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1724 if (icmph->type != ICMP_DEST_UNREACH &&
1725 icmph->type != ICMP_REDIRECT &&
1726 icmph->type != ICMP_TIME_EXCEEDED &&
1727 icmph->type != ICMP_PARAMETERPROB) {
1731 inner_iph = skb_header_pointer(skb,
1732 outer_iph->ihl * 4 + sizeof(_icmph),
1733 sizeof(_inner_iph), &_inner_iph);
1737 return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1740 return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1743 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1745 static int ip_mkroute_input(struct sk_buff *skb,
1746 struct fib_result *res,
1747 const struct flowi4 *fl4,
1748 struct in_device *in_dev,
1749 __be32 daddr, __be32 saddr, u32 tos)
1751 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1752 if (res->fi && res->fi->fib_nhs > 1) {
1755 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1756 h = ip_multipath_icmp_hash(skb);
1758 h = fib_multipath_hash(saddr, daddr);
1759 fib_select_multipath(res, h);
1763 /* create a routing cache entry */
1764 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1768 * NOTE. We drop all the packets that has local source
1769 * addresses, because every properly looped back packet
1770 * must have correct destination already attached by output routine.
1772 * Such approach solves two big problems:
1773 * 1. Not simplex devices are handled properly.
1774 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1775 * called with rcu_read_lock()
1778 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1779 u8 tos, struct net_device *dev)
1781 struct fib_result res;
1782 struct in_device *in_dev = __in_dev_get_rcu(dev);
1783 struct ip_tunnel_info *tun_info;
1785 unsigned int flags = 0;
1789 struct net *net = dev_net(dev);
1792 /* IP on this device is disabled. */
1797 /* Check for the most weird martians, which can be not detected
1801 tun_info = skb_tunnel_info(skb);
1802 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1803 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1805 fl4.flowi4_tun_key.tun_id = 0;
1808 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1809 goto martian_source;
1813 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1816 /* Accept zero addresses only to limited broadcast;
1817 * I even do not know to fix it or not. Waiting for complains :-)
1819 if (ipv4_is_zeronet(saddr))
1820 goto martian_source;
1822 if (ipv4_is_zeronet(daddr))
1823 goto martian_destination;
1825 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1826 * and call it once if daddr or/and saddr are loopback addresses
1828 if (ipv4_is_loopback(daddr)) {
1829 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1830 goto martian_destination;
1831 } else if (ipv4_is_loopback(saddr)) {
1832 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1833 goto martian_source;
1837 * Now we are ready to route packet.
1840 fl4.flowi4_iif = dev->ifindex;
1841 fl4.flowi4_mark = skb->mark;
1842 fl4.flowi4_tos = tos;
1843 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1844 fl4.flowi4_flags = 0;
1847 err = fib_lookup(net, &fl4, &res, 0);
1849 if (!IN_DEV_FORWARD(in_dev))
1850 err = -EHOSTUNREACH;
1854 if (res.type == RTN_BROADCAST)
1857 if (res.type == RTN_LOCAL) {
1858 err = fib_validate_source(skb, saddr, daddr, tos,
1859 0, dev, in_dev, &itag);
1861 goto martian_source;
1865 if (!IN_DEV_FORWARD(in_dev)) {
1866 err = -EHOSTUNREACH;
1869 if (res.type != RTN_UNICAST)
1870 goto martian_destination;
1872 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1876 if (skb->protocol != htons(ETH_P_IP))
1879 if (!ipv4_is_zeronet(saddr)) {
1880 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1883 goto martian_source;
1885 flags |= RTCF_BROADCAST;
1886 res.type = RTN_BROADCAST;
1887 RT_CACHE_STAT_INC(in_brd);
1893 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1894 if (rt_cache_valid(rth)) {
1895 skb_dst_set_noref(skb, &rth->dst);
1903 rth = rt_dst_alloc(net->loopback_dev, flags | RTCF_LOCAL, res.type,
1904 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1908 rth->dst.output= ip_rt_bug;
1909 #ifdef CONFIG_IP_ROUTE_CLASSID
1910 rth->dst.tclassid = itag;
1912 rth->rt_is_input = 1;
1914 rth->rt_table_id = res.table->tb_id;
1916 RT_CACHE_STAT_INC(in_slow_tot);
1917 if (res.type == RTN_UNREACHABLE) {
1918 rth->dst.input= ip_error;
1919 rth->dst.error= -err;
1920 rth->rt_flags &= ~RTCF_LOCAL;
1923 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1924 rth->dst.flags |= DST_NOCACHE;
1925 rt_add_uncached_list(rth);
1928 skb_dst_set(skb, &rth->dst);
1933 RT_CACHE_STAT_INC(in_no_route);
1934 res.type = RTN_UNREACHABLE;
1940 * Do not cache martian addresses: they should be logged (RFC1812)
1942 martian_destination:
1943 RT_CACHE_STAT_INC(in_martian_dst);
1944 #ifdef CONFIG_IP_ROUTE_VERBOSE
1945 if (IN_DEV_LOG_MARTIANS(in_dev))
1946 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1947 &daddr, &saddr, dev->name);
1959 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1963 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1964 u8 tos, struct net_device *dev)
1970 /* Multicast recognition logic is moved from route cache to here.
1971 The problem was that too many Ethernet cards have broken/missing
1972 hardware multicast filters :-( As result the host on multicasting
1973 network acquires a lot of useless route cache entries, sort of
1974 SDR messages from all the world. Now we try to get rid of them.
1975 Really, provided software IP multicast filter is organized
1976 reasonably (at least, hashed), it does not result in a slowdown
1977 comparing with route cache reject entries.
1978 Note, that multicast routers are not affected, because
1979 route cache entry is created eventually.
1981 if (ipv4_is_multicast(daddr)) {
1982 struct in_device *in_dev = __in_dev_get_rcu(dev);
1985 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1986 ip_hdr(skb)->protocol);
1988 #ifdef CONFIG_IP_MROUTE
1990 (!ipv4_is_local_multicast(daddr) &&
1991 IN_DEV_MFORWARD(in_dev))
1994 int res = ip_route_input_mc(skb, daddr, saddr,
2003 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2007 EXPORT_SYMBOL(ip_route_input_noref);
2009 /* called with rcu_read_lock() */
2010 static struct rtable *__mkroute_output(const struct fib_result *res,
2011 const struct flowi4 *fl4, int orig_oif,
2012 struct net_device *dev_out,
2015 struct fib_info *fi = res->fi;
2016 struct fib_nh_exception *fnhe;
2017 struct in_device *in_dev;
2018 u16 type = res->type;
2022 in_dev = __in_dev_get_rcu(dev_out);
2024 return ERR_PTR(-EINVAL);
2026 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2027 if (ipv4_is_loopback(fl4->saddr) &&
2028 !(dev_out->flags & IFF_LOOPBACK) &&
2029 !netif_is_l3_master(dev_out))
2030 return ERR_PTR(-EINVAL);
2032 if (ipv4_is_lbcast(fl4->daddr))
2033 type = RTN_BROADCAST;
2034 else if (ipv4_is_multicast(fl4->daddr))
2035 type = RTN_MULTICAST;
2036 else if (ipv4_is_zeronet(fl4->daddr))
2037 return ERR_PTR(-EINVAL);
2039 if (dev_out->flags & IFF_LOOPBACK)
2040 flags |= RTCF_LOCAL;
2043 if (type == RTN_BROADCAST) {
2044 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2046 } else if (type == RTN_MULTICAST) {
2047 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2048 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2050 flags &= ~RTCF_LOCAL;
2053 /* If multicast route do not exist use
2054 * default one, but do not gateway in this case.
2057 if (fi && res->prefixlen < 4)
2059 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2060 (orig_oif != dev_out->ifindex)) {
2061 /* For local routes that require a particular output interface
2062 * we do not want to cache the result. Caching the result
2063 * causes incorrect behaviour when there are multiple source
2064 * addresses on the interface, the end result being that if the
2065 * intended recipient is waiting on that interface for the
2066 * packet he won't receive it because it will be delivered on
2067 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2068 * be set to the loopback interface as well.
2074 do_cache &= fi != NULL;
2076 struct rtable __rcu **prth;
2077 struct fib_nh *nh = &FIB_RES_NH(*res);
2079 fnhe = find_exception(nh, fl4->daddr);
2081 prth = &fnhe->fnhe_rth_output;
2082 rth = rcu_dereference(*prth);
2083 if (rth && rth->dst.expires &&
2084 time_after(jiffies, rth->dst.expires)) {
2085 ip_del_fnhe(nh, fl4->daddr);
2092 if (unlikely(fl4->flowi4_flags &
2093 FLOWI_FLAG_KNOWN_NH &&
2095 nh->nh_scope == RT_SCOPE_LINK))) {
2099 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2100 rth = rcu_dereference(*prth);
2103 if (rt_cache_valid(rth)) {
2104 dst_hold(&rth->dst);
2110 rth = rt_dst_alloc(dev_out, flags, type,
2111 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2112 IN_DEV_CONF_GET(in_dev, NOXFRM),
2115 return ERR_PTR(-ENOBUFS);
2117 rth->rt_iif = orig_oif ? : 0;
2119 rth->rt_table_id = res->table->tb_id;
2121 RT_CACHE_STAT_INC(out_slow_tot);
2123 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2124 if (flags & RTCF_LOCAL &&
2125 !(dev_out->flags & IFF_LOOPBACK)) {
2126 rth->dst.output = ip_mc_output;
2127 RT_CACHE_STAT_INC(out_slow_mc);
2129 #ifdef CONFIG_IP_MROUTE
2130 if (type == RTN_MULTICAST) {
2131 if (IN_DEV_MFORWARD(in_dev) &&
2132 !ipv4_is_local_multicast(fl4->daddr)) {
2133 rth->dst.input = ip_mr_input;
2134 rth->dst.output = ip_mc_output;
2140 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2141 if (lwtunnel_output_redirect(rth->dst.lwtstate))
2142 rth->dst.output = lwtunnel_output;
2148 * Major route resolver routine.
2151 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2154 struct net_device *dev_out = NULL;
2155 __u8 tos = RT_FL_TOS(fl4);
2156 unsigned int flags = 0;
2157 struct fib_result res;
2160 int err = -ENETUNREACH;
2166 orig_oif = fl4->flowi4_oif;
2168 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2169 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2170 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2171 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2175 rth = ERR_PTR(-EINVAL);
2176 if (ipv4_is_multicast(fl4->saddr) ||
2177 ipv4_is_lbcast(fl4->saddr) ||
2178 ipv4_is_zeronet(fl4->saddr))
2181 /* I removed check for oif == dev_out->oif here.
2182 It was wrong for two reasons:
2183 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2184 is assigned to multiple interfaces.
2185 2. Moreover, we are allowed to send packets with saddr
2186 of another iface. --ANK
2189 if (fl4->flowi4_oif == 0 &&
2190 (ipv4_is_multicast(fl4->daddr) ||
2191 ipv4_is_lbcast(fl4->daddr))) {
2192 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2193 dev_out = __ip_dev_find(net, fl4->saddr, false);
2197 /* Special hack: user can direct multicasts
2198 and limited broadcast via necessary interface
2199 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2200 This hack is not just for fun, it allows
2201 vic,vat and friends to work.
2202 They bind socket to loopback, set ttl to zero
2203 and expect that it will work.
2204 From the viewpoint of routing cache they are broken,
2205 because we are not allowed to build multicast path
2206 with loopback source addr (look, routing cache
2207 cannot know, that ttl is zero, so that packet
2208 will not leave this host and route is valid).
2209 Luckily, this hack is good workaround.
2212 fl4->flowi4_oif = dev_out->ifindex;
2216 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2217 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2218 if (!__ip_dev_find(net, fl4->saddr, false))
2224 if (fl4->flowi4_oif) {
2225 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2226 rth = ERR_PTR(-ENODEV);
2230 /* RACE: Check return value of inet_select_addr instead. */
2231 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2232 rth = ERR_PTR(-ENETUNREACH);
2235 if (ipv4_is_local_multicast(fl4->daddr) ||
2236 ipv4_is_lbcast(fl4->daddr) ||
2237 fl4->flowi4_proto == IPPROTO_IGMP) {
2239 fl4->saddr = inet_select_addr(dev_out, 0,
2244 if (ipv4_is_multicast(fl4->daddr))
2245 fl4->saddr = inet_select_addr(dev_out, 0,
2247 else if (!fl4->daddr)
2248 fl4->saddr = inet_select_addr(dev_out, 0,
2254 fl4->daddr = fl4->saddr;
2256 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2257 dev_out = net->loopback_dev;
2258 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2259 res.type = RTN_LOCAL;
2260 flags |= RTCF_LOCAL;
2264 err = fib_lookup(net, fl4, &res, 0);
2268 if (fl4->flowi4_oif) {
2269 /* Apparently, routing tables are wrong. Assume,
2270 that the destination is on link.
2273 Because we are allowed to send to iface
2274 even if it has NO routes and NO assigned
2275 addresses. When oif is specified, routing
2276 tables are looked up with only one purpose:
2277 to catch if destination is gatewayed, rather than
2278 direct. Moreover, if MSG_DONTROUTE is set,
2279 we send packet, ignoring both routing tables
2280 and ifaddr state. --ANK
2283 We could make it even if oif is unknown,
2284 likely IPv6, but we do not.
2287 if (fl4->saddr == 0)
2288 fl4->saddr = inet_select_addr(dev_out, 0,
2290 res.type = RTN_UNICAST;
2297 if (res.type == RTN_LOCAL) {
2299 if (res.fi->fib_prefsrc)
2300 fl4->saddr = res.fi->fib_prefsrc;
2302 fl4->saddr = fl4->daddr;
2305 /* L3 master device is the loopback for that domain */
2306 dev_out = l3mdev_master_dev_rcu(dev_out) ? : net->loopback_dev;
2307 fl4->flowi4_oif = dev_out->ifindex;
2308 flags |= RTCF_LOCAL;
2312 fib_select_path(net, &res, fl4, mp_hash);
2314 dev_out = FIB_RES_DEV(res);
2315 fl4->flowi4_oif = dev_out->ifindex;
2319 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2325 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2327 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2332 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2334 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2336 return mtu ? : dst->dev->mtu;
2339 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2340 struct sk_buff *skb, u32 mtu)
2344 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2345 struct sk_buff *skb)
2349 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2355 static struct dst_ops ipv4_dst_blackhole_ops = {
2357 .check = ipv4_blackhole_dst_check,
2358 .mtu = ipv4_blackhole_mtu,
2359 .default_advmss = ipv4_default_advmss,
2360 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2361 .redirect = ipv4_rt_blackhole_redirect,
2362 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2363 .neigh_lookup = ipv4_neigh_lookup,
2366 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2368 struct rtable *ort = (struct rtable *) dst_orig;
2371 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2373 struct dst_entry *new = &rt->dst;
2376 new->input = dst_discard;
2377 new->output = dst_discard_out;
2379 new->dev = ort->dst.dev;
2383 rt->rt_is_input = ort->rt_is_input;
2384 rt->rt_iif = ort->rt_iif;
2385 rt->rt_pmtu = ort->rt_pmtu;
2387 rt->rt_genid = rt_genid_ipv4(net);
2388 rt->rt_flags = ort->rt_flags;
2389 rt->rt_type = ort->rt_type;
2390 rt->rt_gateway = ort->rt_gateway;
2391 rt->rt_uses_gateway = ort->rt_uses_gateway;
2393 INIT_LIST_HEAD(&rt->rt_uncached);
2397 dst_release(dst_orig);
2399 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2402 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2403 const struct sock *sk)
2405 struct rtable *rt = __ip_route_output_key(net, flp4);
2410 if (flp4->flowi4_proto)
2411 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2412 flowi4_to_flowi(flp4),
2417 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2419 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2420 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2421 u32 seq, int event, int nowait, unsigned int flags)
2423 struct rtable *rt = skb_rtable(skb);
2425 struct nlmsghdr *nlh;
2426 unsigned long expires = 0;
2428 u32 metrics[RTAX_MAX];
2430 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2434 r = nlmsg_data(nlh);
2435 r->rtm_family = AF_INET;
2436 r->rtm_dst_len = 32;
2438 r->rtm_tos = fl4->flowi4_tos;
2439 r->rtm_table = table_id;
2440 if (nla_put_u32(skb, RTA_TABLE, table_id))
2441 goto nla_put_failure;
2442 r->rtm_type = rt->rt_type;
2443 r->rtm_scope = RT_SCOPE_UNIVERSE;
2444 r->rtm_protocol = RTPROT_UNSPEC;
2445 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2446 if (rt->rt_flags & RTCF_NOTIFY)
2447 r->rtm_flags |= RTM_F_NOTIFY;
2448 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2449 r->rtm_flags |= RTCF_DOREDIRECT;
2451 if (nla_put_in_addr(skb, RTA_DST, dst))
2452 goto nla_put_failure;
2454 r->rtm_src_len = 32;
2455 if (nla_put_in_addr(skb, RTA_SRC, src))
2456 goto nla_put_failure;
2459 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2460 goto nla_put_failure;
2461 #ifdef CONFIG_IP_ROUTE_CLASSID
2462 if (rt->dst.tclassid &&
2463 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2464 goto nla_put_failure;
2466 if (!rt_is_input_route(rt) &&
2467 fl4->saddr != src) {
2468 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2469 goto nla_put_failure;
2471 if (rt->rt_uses_gateway &&
2472 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2473 goto nla_put_failure;
2475 expires = rt->dst.expires;
2477 unsigned long now = jiffies;
2479 if (time_before(now, expires))
2485 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2486 if (rt->rt_pmtu && expires)
2487 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2488 if (rtnetlink_put_metrics(skb, metrics) < 0)
2489 goto nla_put_failure;
2491 if (fl4->flowi4_mark &&
2492 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2493 goto nla_put_failure;
2495 error = rt->dst.error;
2497 if (rt_is_input_route(rt)) {
2498 #ifdef CONFIG_IP_MROUTE
2499 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2500 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2501 int err = ipmr_get_route(net, skb,
2502 fl4->saddr, fl4->daddr,
2509 goto nla_put_failure;
2511 if (err == -EMSGSIZE)
2512 goto nla_put_failure;
2518 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2519 goto nla_put_failure;
2522 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2523 goto nla_put_failure;
2525 nlmsg_end(skb, nlh);
2529 nlmsg_cancel(skb, nlh);
2533 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2535 struct net *net = sock_net(in_skb->sk);
2537 struct nlattr *tb[RTA_MAX+1];
2538 struct rtable *rt = NULL;
2545 struct sk_buff *skb;
2546 u32 table_id = RT_TABLE_MAIN;
2548 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2552 rtm = nlmsg_data(nlh);
2554 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2560 /* Reserve room for dummy headers, this skb can pass
2561 through good chunk of routing engine.
2563 skb_reset_mac_header(skb);
2564 skb_reset_network_header(skb);
2566 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2567 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2568 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2570 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2571 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2572 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2573 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2575 memset(&fl4, 0, sizeof(fl4));
2578 fl4.flowi4_tos = rtm->rtm_tos;
2579 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2580 fl4.flowi4_mark = mark;
2583 struct net_device *dev;
2585 dev = __dev_get_by_index(net, iif);
2591 skb->protocol = htons(ETH_P_IP);
2595 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2598 rt = skb_rtable(skb);
2599 if (err == 0 && rt->dst.error)
2600 err = -rt->dst.error;
2602 rt = ip_route_output_key(net, &fl4);
2612 skb_dst_set(skb, &rt->dst);
2613 if (rtm->rtm_flags & RTM_F_NOTIFY)
2614 rt->rt_flags |= RTCF_NOTIFY;
2616 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2617 table_id = rt->rt_table_id;
2619 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2620 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2621 RTM_NEWROUTE, 0, 0);
2625 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2634 void ip_rt_multicast_event(struct in_device *in_dev)
2636 rt_cache_flush(dev_net(in_dev->dev));
2639 #ifdef CONFIG_SYSCTL
2640 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2641 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2642 static int ip_rt_gc_elasticity __read_mostly = 8;
2644 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2645 void __user *buffer,
2646 size_t *lenp, loff_t *ppos)
2648 struct net *net = (struct net *)__ctl->extra1;
2651 rt_cache_flush(net);
2652 fnhe_genid_bump(net);
2659 static struct ctl_table ipv4_route_table[] = {
2661 .procname = "gc_thresh",
2662 .data = &ipv4_dst_ops.gc_thresh,
2663 .maxlen = sizeof(int),
2665 .proc_handler = proc_dointvec,
2668 .procname = "max_size",
2669 .data = &ip_rt_max_size,
2670 .maxlen = sizeof(int),
2672 .proc_handler = proc_dointvec,
2675 /* Deprecated. Use gc_min_interval_ms */
2677 .procname = "gc_min_interval",
2678 .data = &ip_rt_gc_min_interval,
2679 .maxlen = sizeof(int),
2681 .proc_handler = proc_dointvec_jiffies,
2684 .procname = "gc_min_interval_ms",
2685 .data = &ip_rt_gc_min_interval,
2686 .maxlen = sizeof(int),
2688 .proc_handler = proc_dointvec_ms_jiffies,
2691 .procname = "gc_timeout",
2692 .data = &ip_rt_gc_timeout,
2693 .maxlen = sizeof(int),
2695 .proc_handler = proc_dointvec_jiffies,
2698 .procname = "gc_interval",
2699 .data = &ip_rt_gc_interval,
2700 .maxlen = sizeof(int),
2702 .proc_handler = proc_dointvec_jiffies,
2705 .procname = "redirect_load",
2706 .data = &ip_rt_redirect_load,
2707 .maxlen = sizeof(int),
2709 .proc_handler = proc_dointvec,
2712 .procname = "redirect_number",
2713 .data = &ip_rt_redirect_number,
2714 .maxlen = sizeof(int),
2716 .proc_handler = proc_dointvec,
2719 .procname = "redirect_silence",
2720 .data = &ip_rt_redirect_silence,
2721 .maxlen = sizeof(int),
2723 .proc_handler = proc_dointvec,
2726 .procname = "error_cost",
2727 .data = &ip_rt_error_cost,
2728 .maxlen = sizeof(int),
2730 .proc_handler = proc_dointvec,
2733 .procname = "error_burst",
2734 .data = &ip_rt_error_burst,
2735 .maxlen = sizeof(int),
2737 .proc_handler = proc_dointvec,
2740 .procname = "gc_elasticity",
2741 .data = &ip_rt_gc_elasticity,
2742 .maxlen = sizeof(int),
2744 .proc_handler = proc_dointvec,
2747 .procname = "mtu_expires",
2748 .data = &ip_rt_mtu_expires,
2749 .maxlen = sizeof(int),
2751 .proc_handler = proc_dointvec_jiffies,
2754 .procname = "min_pmtu",
2755 .data = &ip_rt_min_pmtu,
2756 .maxlen = sizeof(int),
2758 .proc_handler = proc_dointvec,
2761 .procname = "min_adv_mss",
2762 .data = &ip_rt_min_advmss,
2763 .maxlen = sizeof(int),
2765 .proc_handler = proc_dointvec,
2770 static struct ctl_table ipv4_route_flush_table[] = {
2772 .procname = "flush",
2773 .maxlen = sizeof(int),
2775 .proc_handler = ipv4_sysctl_rtcache_flush,
2780 static __net_init int sysctl_route_net_init(struct net *net)
2782 struct ctl_table *tbl;
2784 tbl = ipv4_route_flush_table;
2785 if (!net_eq(net, &init_net)) {
2786 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2790 /* Don't export sysctls to unprivileged users */
2791 if (net->user_ns != &init_user_ns)
2792 tbl[0].procname = NULL;
2794 tbl[0].extra1 = net;
2796 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2797 if (!net->ipv4.route_hdr)
2802 if (tbl != ipv4_route_flush_table)
2808 static __net_exit void sysctl_route_net_exit(struct net *net)
2810 struct ctl_table *tbl;
2812 tbl = net->ipv4.route_hdr->ctl_table_arg;
2813 unregister_net_sysctl_table(net->ipv4.route_hdr);
2814 BUG_ON(tbl == ipv4_route_flush_table);
2818 static __net_initdata struct pernet_operations sysctl_route_ops = {
2819 .init = sysctl_route_net_init,
2820 .exit = sysctl_route_net_exit,
2824 static __net_init int rt_genid_init(struct net *net)
2826 atomic_set(&net->ipv4.rt_genid, 0);
2827 atomic_set(&net->fnhe_genid, 0);
2828 get_random_bytes(&net->ipv4.dev_addr_genid,
2829 sizeof(net->ipv4.dev_addr_genid));
2833 static __net_initdata struct pernet_operations rt_genid_ops = {
2834 .init = rt_genid_init,
2837 static int __net_init ipv4_inetpeer_init(struct net *net)
2839 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2843 inet_peer_base_init(bp);
2844 net->ipv4.peers = bp;
2848 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2850 struct inet_peer_base *bp = net->ipv4.peers;
2852 net->ipv4.peers = NULL;
2853 inetpeer_invalidate_tree(bp);
2857 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2858 .init = ipv4_inetpeer_init,
2859 .exit = ipv4_inetpeer_exit,
2862 #ifdef CONFIG_IP_ROUTE_CLASSID
2863 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2864 #endif /* CONFIG_IP_ROUTE_CLASSID */
2866 int __init ip_rt_init(void)
2871 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2873 panic("IP: failed to allocate ip_idents\n");
2875 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2877 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2879 panic("IP: failed to allocate ip_tstamps\n");
2881 for_each_possible_cpu(cpu) {
2882 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2884 INIT_LIST_HEAD(&ul->head);
2885 spin_lock_init(&ul->lock);
2887 #ifdef CONFIG_IP_ROUTE_CLASSID
2888 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2890 panic("IP: failed to allocate ip_rt_acct\n");
2893 ipv4_dst_ops.kmem_cachep =
2894 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2895 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2897 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2899 if (dst_entries_init(&ipv4_dst_ops) < 0)
2900 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2902 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2903 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2905 ipv4_dst_ops.gc_thresh = ~0;
2906 ip_rt_max_size = INT_MAX;
2911 if (ip_rt_proc_init())
2912 pr_err("Unable to create route proc files\n");
2917 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2919 #ifdef CONFIG_SYSCTL
2920 register_pernet_subsys(&sysctl_route_ops);
2922 register_pernet_subsys(&rt_genid_ops);
2923 register_pernet_subsys(&ipv4_inetpeer_ops);
2927 #ifdef CONFIG_SYSCTL
2929 * We really need to sanitize the damn ipv4 init order, then all
2930 * this nonsense will go away.
2932 void __init ip_static_sysctl_init(void)
2934 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);