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 return atomic_add_return(segs + delta, p_id) - segs;
486 EXPORT_SYMBOL(ip_idents_reserve);
488 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
490 static u32 ip_idents_hashrnd __read_mostly;
493 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
495 hash = jhash_3words((__force u32)iph->daddr,
496 (__force u32)iph->saddr,
497 iph->protocol ^ net_hash_mix(net),
499 id = ip_idents_reserve(hash, segs);
502 EXPORT_SYMBOL(__ip_select_ident);
504 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
505 const struct sock *sk,
506 const struct iphdr *iph,
508 u8 prot, u32 mark, int flow_flags)
511 const struct inet_sock *inet = inet_sk(sk);
513 oif = sk->sk_bound_dev_if;
515 tos = RT_CONN_FLAGS(sk);
516 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
518 flowi4_init_output(fl4, oif, mark, tos,
519 RT_SCOPE_UNIVERSE, prot,
521 iph->daddr, iph->saddr, 0, 0,
522 sock_net_uid(net, sk));
525 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
526 const struct sock *sk)
528 const struct net *net = dev_net(skb->dev);
529 const struct iphdr *iph = ip_hdr(skb);
530 int oif = skb->dev->ifindex;
531 u8 tos = RT_TOS(iph->tos);
532 u8 prot = iph->protocol;
533 u32 mark = skb->mark;
535 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
538 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
540 const struct inet_sock *inet = inet_sk(sk);
541 const struct ip_options_rcu *inet_opt;
542 __be32 daddr = inet->inet_daddr;
545 inet_opt = rcu_dereference(inet->inet_opt);
546 if (inet_opt && inet_opt->opt.srr)
547 daddr = inet_opt->opt.faddr;
548 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
549 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
550 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
551 inet_sk_flowi_flags(sk),
552 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
556 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
557 const struct sk_buff *skb)
560 build_skb_flow_key(fl4, skb, sk);
562 build_sk_flow_key(fl4, sk);
565 static inline void rt_free(struct rtable *rt)
567 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
570 static DEFINE_SPINLOCK(fnhe_lock);
572 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
576 rt = rcu_dereference(fnhe->fnhe_rth_input);
578 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
581 rt = rcu_dereference(fnhe->fnhe_rth_output);
583 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
588 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
590 struct fib_nh_exception *fnhe, *oldest;
592 oldest = rcu_dereference(hash->chain);
593 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
594 fnhe = rcu_dereference(fnhe->fnhe_next)) {
595 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
598 fnhe_flush_routes(oldest);
602 static inline u32 fnhe_hashfun(__be32 daddr)
604 static u32 fnhe_hashrnd __read_mostly;
607 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
608 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
609 return hash_32(hval, FNHE_HASH_SHIFT);
612 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
614 rt->rt_pmtu = fnhe->fnhe_pmtu;
615 rt->dst.expires = fnhe->fnhe_expires;
618 rt->rt_flags |= RTCF_REDIRECTED;
619 rt->rt_gateway = fnhe->fnhe_gw;
620 rt->rt_uses_gateway = 1;
624 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
625 u32 pmtu, unsigned long expires)
627 struct fnhe_hash_bucket *hash;
628 struct fib_nh_exception *fnhe;
632 u32 hval = fnhe_hashfun(daddr);
634 spin_lock_bh(&fnhe_lock);
636 hash = rcu_dereference(nh->nh_exceptions);
638 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
641 rcu_assign_pointer(nh->nh_exceptions, hash);
647 for (fnhe = rcu_dereference(hash->chain); fnhe;
648 fnhe = rcu_dereference(fnhe->fnhe_next)) {
649 if (fnhe->fnhe_daddr == daddr)
658 fnhe->fnhe_pmtu = pmtu;
659 fnhe->fnhe_expires = max(1UL, expires);
661 /* Update all cached dsts too */
662 rt = rcu_dereference(fnhe->fnhe_rth_input);
664 fill_route_from_fnhe(rt, fnhe);
665 rt = rcu_dereference(fnhe->fnhe_rth_output);
667 fill_route_from_fnhe(rt, fnhe);
669 if (depth > FNHE_RECLAIM_DEPTH)
670 fnhe = fnhe_oldest(hash);
672 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
676 fnhe->fnhe_next = hash->chain;
677 rcu_assign_pointer(hash->chain, fnhe);
679 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
680 fnhe->fnhe_daddr = daddr;
682 fnhe->fnhe_pmtu = pmtu;
683 fnhe->fnhe_expires = expires;
685 /* Exception created; mark the cached routes for the nexthop
686 * stale, so anyone caching it rechecks if this exception
689 rt = rcu_dereference(nh->nh_rth_input);
691 rt->dst.obsolete = DST_OBSOLETE_KILL;
693 for_each_possible_cpu(i) {
694 struct rtable __rcu **prt;
695 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
696 rt = rcu_dereference(*prt);
698 rt->dst.obsolete = DST_OBSOLETE_KILL;
702 fnhe->fnhe_stamp = jiffies;
705 spin_unlock_bh(&fnhe_lock);
708 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
711 __be32 new_gw = icmp_hdr(skb)->un.gateway;
712 __be32 old_gw = ip_hdr(skb)->saddr;
713 struct net_device *dev = skb->dev;
714 struct in_device *in_dev;
715 struct fib_result res;
719 switch (icmp_hdr(skb)->code & 7) {
721 case ICMP_REDIR_NETTOS:
722 case ICMP_REDIR_HOST:
723 case ICMP_REDIR_HOSTTOS:
730 if (rt->rt_gateway != old_gw)
733 in_dev = __in_dev_get_rcu(dev);
738 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
739 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
740 ipv4_is_zeronet(new_gw))
741 goto reject_redirect;
743 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
744 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
745 goto reject_redirect;
746 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
747 goto reject_redirect;
749 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
750 goto reject_redirect;
753 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
755 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
757 if (!(n->nud_state & NUD_VALID)) {
758 neigh_event_send(n, NULL);
760 if (fib_lookup(net, fl4, &res, 0) == 0) {
761 struct fib_nh *nh = &FIB_RES_NH(res);
763 update_or_create_fnhe(nh, fl4->daddr, new_gw,
764 0, jiffies + ip_rt_gc_timeout);
767 rt->dst.obsolete = DST_OBSOLETE_KILL;
768 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
775 #ifdef CONFIG_IP_ROUTE_VERBOSE
776 if (IN_DEV_LOG_MARTIANS(in_dev)) {
777 const struct iphdr *iph = (const struct iphdr *) skb->data;
778 __be32 daddr = iph->daddr;
779 __be32 saddr = iph->saddr;
781 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
782 " Advised path = %pI4 -> %pI4\n",
783 &old_gw, dev->name, &new_gw,
790 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
794 const struct iphdr *iph = (const struct iphdr *) skb->data;
795 int oif = skb->dev->ifindex;
796 u8 tos = RT_TOS(iph->tos);
797 u8 prot = iph->protocol;
798 u32 mark = skb->mark;
800 rt = (struct rtable *) dst;
802 __build_flow_key(sock_net(sk), &fl4, sk, iph, oif, tos, prot, mark, 0);
803 __ip_do_redirect(rt, skb, &fl4, true);
806 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
808 struct rtable *rt = (struct rtable *)dst;
809 struct dst_entry *ret = dst;
812 if (dst->obsolete > 0) {
815 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
826 * 1. The first ip_rt_redirect_number redirects are sent
827 * with exponential backoff, then we stop sending them at all,
828 * assuming that the host ignores our redirects.
829 * 2. If we did not see packets requiring redirects
830 * during ip_rt_redirect_silence, we assume that the host
831 * forgot redirected route and start to send redirects again.
833 * This algorithm is much cheaper and more intelligent than dumb load limiting
836 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
837 * and "frag. need" (breaks PMTU discovery) in icmp.c.
840 void ip_rt_send_redirect(struct sk_buff *skb)
842 struct rtable *rt = skb_rtable(skb);
843 struct in_device *in_dev;
844 struct inet_peer *peer;
850 in_dev = __in_dev_get_rcu(rt->dst.dev);
851 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
855 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
856 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
859 net = dev_net(rt->dst.dev);
860 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
862 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
863 rt_nexthop(rt, ip_hdr(skb)->daddr));
867 /* No redirected packets during ip_rt_redirect_silence;
868 * reset the algorithm.
870 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
871 peer->rate_tokens = 0;
873 /* Too many ignored redirects; do not send anything
874 * set dst.rate_last to the last seen redirected packet.
876 if (peer->rate_tokens >= ip_rt_redirect_number) {
877 peer->rate_last = jiffies;
881 /* Check for load limit; set rate_last to the latest sent
884 if (peer->rate_tokens == 0 ||
887 (ip_rt_redirect_load << peer->rate_tokens)))) {
888 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
890 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
891 peer->rate_last = jiffies;
893 #ifdef CONFIG_IP_ROUTE_VERBOSE
895 peer->rate_tokens == ip_rt_redirect_number)
896 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
897 &ip_hdr(skb)->saddr, inet_iif(skb),
898 &ip_hdr(skb)->daddr, &gw);
905 static int ip_error(struct sk_buff *skb)
907 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
908 struct rtable *rt = skb_rtable(skb);
909 struct inet_peer *peer;
915 /* IP on this device is disabled. */
919 net = dev_net(rt->dst.dev);
920 if (!IN_DEV_FORWARD(in_dev)) {
921 switch (rt->dst.error) {
923 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
927 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
933 switch (rt->dst.error) {
938 code = ICMP_HOST_UNREACH;
941 code = ICMP_NET_UNREACH;
942 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
945 code = ICMP_PKT_FILTERED;
949 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
950 l3mdev_master_ifindex(skb->dev), 1);
955 peer->rate_tokens += now - peer->rate_last;
956 if (peer->rate_tokens > ip_rt_error_burst)
957 peer->rate_tokens = ip_rt_error_burst;
958 peer->rate_last = now;
959 if (peer->rate_tokens >= ip_rt_error_cost)
960 peer->rate_tokens -= ip_rt_error_cost;
966 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
972 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
974 struct dst_entry *dst = &rt->dst;
975 struct fib_result res;
977 if (dst_metric_locked(dst, RTAX_MTU))
980 if (ipv4_mtu(dst) < mtu)
983 if (mtu < ip_rt_min_pmtu)
984 mtu = ip_rt_min_pmtu;
986 if (rt->rt_pmtu == mtu &&
987 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
991 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
992 struct fib_nh *nh = &FIB_RES_NH(res);
994 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
995 jiffies + ip_rt_mtu_expires);
1000 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1001 struct sk_buff *skb, u32 mtu)
1003 struct rtable *rt = (struct rtable *) dst;
1006 ip_rt_build_flow_key(&fl4, sk, skb);
1007 __ip_rt_update_pmtu(rt, &fl4, mtu);
1010 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1011 int oif, u32 mark, u8 protocol, int flow_flags)
1013 const struct iphdr *iph = (const struct iphdr *) skb->data;
1018 mark = IP4_REPLY_MARK(net, skb->mark);
1020 __build_flow_key(net, &fl4, NULL, iph, oif,
1021 RT_TOS(iph->tos), protocol, mark, flow_flags);
1022 rt = __ip_route_output_key(net, &fl4);
1024 __ip_rt_update_pmtu(rt, &fl4, mtu);
1028 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1030 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1032 const struct iphdr *iph = (const struct iphdr *) skb->data;
1036 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1038 if (!fl4.flowi4_mark)
1039 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1041 rt = __ip_route_output_key(sock_net(sk), &fl4);
1043 __ip_rt_update_pmtu(rt, &fl4, mtu);
1048 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1050 const struct iphdr *iph = (const struct iphdr *) skb->data;
1053 struct dst_entry *odst = NULL;
1055 struct net *net = sock_net(sk);
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(net, &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(net, &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;
1124 struct net *net = sock_net(sk);
1126 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1127 rt = __ip_route_output_key(net, &fl4);
1129 __ip_do_redirect(rt, skb, &fl4, false);
1133 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1135 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1137 struct rtable *rt = (struct rtable *) dst;
1139 /* All IPV4 dsts are created with ->obsolete set to the value
1140 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1141 * into this function always.
1143 * When a PMTU/redirect information update invalidates a route,
1144 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1145 * DST_OBSOLETE_DEAD by dst_free().
1147 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1152 static void ipv4_link_failure(struct sk_buff *skb)
1156 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1158 rt = skb_rtable(skb);
1160 dst_set_expires(&rt->dst, 0);
1163 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1165 pr_debug("%s: %pI4 -> %pI4, %s\n",
1166 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1167 skb->dev ? skb->dev->name : "?");
1174 We do not cache source address of outgoing interface,
1175 because it is used only by IP RR, TS and SRR options,
1176 so that it out of fast path.
1178 BTW remember: "addr" is allowed to be not aligned
1182 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1186 if (rt_is_output_route(rt))
1187 src = ip_hdr(skb)->saddr;
1189 struct fib_result res;
1195 memset(&fl4, 0, sizeof(fl4));
1196 fl4.daddr = iph->daddr;
1197 fl4.saddr = iph->saddr;
1198 fl4.flowi4_tos = RT_TOS(iph->tos);
1199 fl4.flowi4_oif = rt->dst.dev->ifindex;
1200 fl4.flowi4_iif = skb->dev->ifindex;
1201 fl4.flowi4_mark = skb->mark;
1204 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1205 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1207 src = inet_select_addr(rt->dst.dev,
1208 rt_nexthop(rt, iph->daddr),
1212 memcpy(addr, &src, 4);
1215 #ifdef CONFIG_IP_ROUTE_CLASSID
1216 static void set_class_tag(struct rtable *rt, u32 tag)
1218 if (!(rt->dst.tclassid & 0xFFFF))
1219 rt->dst.tclassid |= tag & 0xFFFF;
1220 if (!(rt->dst.tclassid & 0xFFFF0000))
1221 rt->dst.tclassid |= tag & 0xFFFF0000;
1225 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1227 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1230 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1232 if (advmss > 65535 - 40)
1233 advmss = 65535 - 40;
1238 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1240 const struct rtable *rt = (const struct rtable *) dst;
1241 unsigned int mtu = rt->rt_pmtu;
1243 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1244 mtu = dst_metric_raw(dst, RTAX_MTU);
1249 mtu = dst->dev->mtu;
1251 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1252 if (rt->rt_uses_gateway && mtu > 576)
1256 return min_t(unsigned int, mtu, IP_MAX_MTU);
1259 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1261 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1262 struct fib_nh_exception *fnhe;
1268 hval = fnhe_hashfun(daddr);
1270 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1271 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1272 if (fnhe->fnhe_daddr == daddr)
1278 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1283 spin_lock_bh(&fnhe_lock);
1285 if (daddr == fnhe->fnhe_daddr) {
1286 struct rtable __rcu **porig;
1287 struct rtable *orig;
1288 int genid = fnhe_genid(dev_net(rt->dst.dev));
1290 if (rt_is_input_route(rt))
1291 porig = &fnhe->fnhe_rth_input;
1293 porig = &fnhe->fnhe_rth_output;
1294 orig = rcu_dereference(*porig);
1296 if (fnhe->fnhe_genid != genid) {
1297 fnhe->fnhe_genid = genid;
1299 fnhe->fnhe_pmtu = 0;
1300 fnhe->fnhe_expires = 0;
1301 fnhe_flush_routes(fnhe);
1304 fill_route_from_fnhe(rt, fnhe);
1305 if (!rt->rt_gateway)
1306 rt->rt_gateway = daddr;
1308 if (!(rt->dst.flags & DST_NOCACHE)) {
1309 rcu_assign_pointer(*porig, rt);
1315 fnhe->fnhe_stamp = jiffies;
1317 spin_unlock_bh(&fnhe_lock);
1322 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1324 struct rtable *orig, *prev, **p;
1327 if (rt_is_input_route(rt)) {
1328 p = (struct rtable **)&nh->nh_rth_input;
1330 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1334 prev = cmpxchg(p, orig, rt);
1344 struct uncached_list {
1346 struct list_head head;
1349 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1351 static void rt_add_uncached_list(struct rtable *rt)
1353 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1355 rt->rt_uncached_list = ul;
1357 spin_lock_bh(&ul->lock);
1358 list_add_tail(&rt->rt_uncached, &ul->head);
1359 spin_unlock_bh(&ul->lock);
1362 static void ipv4_dst_destroy(struct dst_entry *dst)
1364 struct rtable *rt = (struct rtable *) dst;
1366 if (!list_empty(&rt->rt_uncached)) {
1367 struct uncached_list *ul = rt->rt_uncached_list;
1369 spin_lock_bh(&ul->lock);
1370 list_del(&rt->rt_uncached);
1371 spin_unlock_bh(&ul->lock);
1375 void rt_flush_dev(struct net_device *dev)
1377 struct net *net = dev_net(dev);
1381 for_each_possible_cpu(cpu) {
1382 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1384 spin_lock_bh(&ul->lock);
1385 list_for_each_entry(rt, &ul->head, rt_uncached) {
1386 if (rt->dst.dev != dev)
1388 rt->dst.dev = net->loopback_dev;
1389 dev_hold(rt->dst.dev);
1392 spin_unlock_bh(&ul->lock);
1396 static bool rt_cache_valid(const struct rtable *rt)
1399 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1403 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1404 const struct fib_result *res,
1405 struct fib_nh_exception *fnhe,
1406 struct fib_info *fi, u16 type, u32 itag)
1408 bool cached = false;
1411 struct fib_nh *nh = &FIB_RES_NH(*res);
1413 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1414 rt->rt_gateway = nh->nh_gw;
1415 rt->rt_uses_gateway = 1;
1417 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1418 #ifdef CONFIG_IP_ROUTE_CLASSID
1419 rt->dst.tclassid = nh->nh_tclassid;
1421 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1423 cached = rt_bind_exception(rt, fnhe, daddr);
1424 else if (!(rt->dst.flags & DST_NOCACHE))
1425 cached = rt_cache_route(nh, rt);
1426 if (unlikely(!cached)) {
1427 /* Routes we intend to cache in nexthop exception or
1428 * FIB nexthop have the DST_NOCACHE bit clear.
1429 * However, if we are unsuccessful at storing this
1430 * route into the cache we really need to set it.
1432 rt->dst.flags |= DST_NOCACHE;
1433 if (!rt->rt_gateway)
1434 rt->rt_gateway = daddr;
1435 rt_add_uncached_list(rt);
1438 rt_add_uncached_list(rt);
1440 #ifdef CONFIG_IP_ROUTE_CLASSID
1441 #ifdef CONFIG_IP_MULTIPLE_TABLES
1442 set_class_tag(rt, res->tclassid);
1444 set_class_tag(rt, itag);
1448 static struct rtable *rt_dst_alloc(struct net_device *dev,
1449 unsigned int flags, u16 type,
1450 bool nopolicy, bool noxfrm, bool will_cache)
1454 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1455 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1456 (nopolicy ? DST_NOPOLICY : 0) |
1457 (noxfrm ? DST_NOXFRM : 0));
1460 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1461 rt->rt_flags = flags;
1463 rt->rt_is_input = 0;
1467 rt->rt_uses_gateway = 0;
1468 rt->rt_table_id = 0;
1469 INIT_LIST_HEAD(&rt->rt_uncached);
1471 rt->dst.output = ip_output;
1472 if (flags & RTCF_LOCAL)
1473 rt->dst.input = ip_local_deliver;
1479 /* called in rcu_read_lock() section */
1480 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1481 u8 tos, struct net_device *dev, int our)
1484 struct in_device *in_dev = __in_dev_get_rcu(dev);
1485 unsigned int flags = RTCF_MULTICAST;
1489 /* Primary sanity checks. */
1494 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1495 skb->protocol != htons(ETH_P_IP))
1498 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1501 if (ipv4_is_zeronet(saddr)) {
1502 if (!ipv4_is_local_multicast(daddr))
1505 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1511 flags |= RTCF_LOCAL;
1513 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1514 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1518 #ifdef CONFIG_IP_ROUTE_CLASSID
1519 rth->dst.tclassid = itag;
1521 rth->dst.output = ip_rt_bug;
1522 rth->rt_is_input= 1;
1524 #ifdef CONFIG_IP_MROUTE
1525 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1526 rth->dst.input = ip_mr_input;
1528 RT_CACHE_STAT_INC(in_slow_mc);
1530 skb_dst_set(skb, &rth->dst);
1542 static void ip_handle_martian_source(struct net_device *dev,
1543 struct in_device *in_dev,
1544 struct sk_buff *skb,
1548 RT_CACHE_STAT_INC(in_martian_src);
1549 #ifdef CONFIG_IP_ROUTE_VERBOSE
1550 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1552 * RFC1812 recommendation, if source is martian,
1553 * the only hint is MAC header.
1555 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1556 &daddr, &saddr, dev->name);
1557 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1558 print_hex_dump(KERN_WARNING, "ll header: ",
1559 DUMP_PREFIX_OFFSET, 16, 1,
1560 skb_mac_header(skb),
1561 dev->hard_header_len, true);
1567 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1569 struct fnhe_hash_bucket *hash;
1570 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1571 u32 hval = fnhe_hashfun(daddr);
1573 spin_lock_bh(&fnhe_lock);
1575 hash = rcu_dereference_protected(nh->nh_exceptions,
1576 lockdep_is_held(&fnhe_lock));
1579 fnhe_p = &hash->chain;
1580 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1582 if (fnhe->fnhe_daddr == daddr) {
1583 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1584 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1585 fnhe_flush_routes(fnhe);
1586 kfree_rcu(fnhe, rcu);
1589 fnhe_p = &fnhe->fnhe_next;
1590 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1591 lockdep_is_held(&fnhe_lock));
1594 spin_unlock_bh(&fnhe_lock);
1597 /* called in rcu_read_lock() section */
1598 static int __mkroute_input(struct sk_buff *skb,
1599 const struct fib_result *res,
1600 struct in_device *in_dev,
1601 __be32 daddr, __be32 saddr, u32 tos)
1603 struct fib_nh_exception *fnhe;
1606 struct in_device *out_dev;
1610 /* get a working reference to the output device */
1611 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1613 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1617 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1618 in_dev->dev, in_dev, &itag);
1620 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1626 do_cache = res->fi && !itag;
1627 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1628 skb->protocol == htons(ETH_P_IP) &&
1629 (IN_DEV_SHARED_MEDIA(out_dev) ||
1630 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1631 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1633 if (skb->protocol != htons(ETH_P_IP)) {
1634 /* Not IP (i.e. ARP). Do not create route, if it is
1635 * invalid for proxy arp. DNAT routes are always valid.
1637 * Proxy arp feature have been extended to allow, ARP
1638 * replies back to the same interface, to support
1639 * Private VLAN switch technologies. See arp.c.
1641 if (out_dev == in_dev &&
1642 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1648 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1651 rth = rcu_dereference(fnhe->fnhe_rth_input);
1652 if (rth && rth->dst.expires &&
1653 time_after(jiffies, rth->dst.expires)) {
1654 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1661 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1664 if (rt_cache_valid(rth)) {
1665 skb_dst_set_noref(skb, &rth->dst);
1670 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1671 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1672 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1678 rth->rt_is_input = 1;
1680 rth->rt_table_id = res->table->tb_id;
1681 RT_CACHE_STAT_INC(in_slow_tot);
1683 rth->dst.input = ip_forward;
1685 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1686 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1687 rth->dst.lwtstate->orig_output = rth->dst.output;
1688 rth->dst.output = lwtunnel_output;
1690 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1691 rth->dst.lwtstate->orig_input = rth->dst.input;
1692 rth->dst.input = lwtunnel_input;
1694 skb_dst_set(skb, &rth->dst);
1701 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1703 /* To make ICMP packets follow the right flow, the multipath hash is
1704 * calculated from the inner IP addresses in reverse order.
1706 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1708 const struct iphdr *outer_iph = ip_hdr(skb);
1709 struct icmphdr _icmph;
1710 const struct icmphdr *icmph;
1711 struct iphdr _inner_iph;
1712 const struct iphdr *inner_iph;
1714 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1717 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1722 if (icmph->type != ICMP_DEST_UNREACH &&
1723 icmph->type != ICMP_REDIRECT &&
1724 icmph->type != ICMP_TIME_EXCEEDED &&
1725 icmph->type != ICMP_PARAMETERPROB) {
1729 inner_iph = skb_header_pointer(skb,
1730 outer_iph->ihl * 4 + sizeof(_icmph),
1731 sizeof(_inner_iph), &_inner_iph);
1735 return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1738 return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1741 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1743 static int ip_mkroute_input(struct sk_buff *skb,
1744 struct fib_result *res,
1745 const struct flowi4 *fl4,
1746 struct in_device *in_dev,
1747 __be32 daddr, __be32 saddr, u32 tos)
1749 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1750 if (res->fi && res->fi->fib_nhs > 1) {
1753 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1754 h = ip_multipath_icmp_hash(skb);
1756 h = fib_multipath_hash(saddr, daddr);
1757 fib_select_multipath(res, h);
1761 /* create a routing cache entry */
1762 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1766 * NOTE. We drop all the packets that has local source
1767 * addresses, because every properly looped back packet
1768 * must have correct destination already attached by output routine.
1770 * Such approach solves two big problems:
1771 * 1. Not simplex devices are handled properly.
1772 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1773 * called with rcu_read_lock()
1776 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1777 u8 tos, struct net_device *dev)
1779 struct fib_result res;
1780 struct in_device *in_dev = __in_dev_get_rcu(dev);
1781 struct ip_tunnel_info *tun_info;
1783 unsigned int flags = 0;
1787 struct net *net = dev_net(dev);
1790 /* IP on this device is disabled. */
1795 /* Check for the most weird martians, which can be not detected
1799 tun_info = skb_tunnel_info(skb);
1800 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1801 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1803 fl4.flowi4_tun_key.tun_id = 0;
1806 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1807 goto martian_source;
1811 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1814 /* Accept zero addresses only to limited broadcast;
1815 * I even do not know to fix it or not. Waiting for complains :-)
1817 if (ipv4_is_zeronet(saddr))
1818 goto martian_source;
1820 if (ipv4_is_zeronet(daddr))
1821 goto martian_destination;
1823 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1824 * and call it once if daddr or/and saddr are loopback addresses
1826 if (ipv4_is_loopback(daddr)) {
1827 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1828 goto martian_destination;
1829 } else if (ipv4_is_loopback(saddr)) {
1830 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1831 goto martian_source;
1835 * Now we are ready to route packet.
1838 fl4.flowi4_iif = l3mdev_fib_oif_rcu(dev);
1839 fl4.flowi4_mark = skb->mark;
1840 fl4.flowi4_tos = tos;
1841 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1842 fl4.flowi4_flags = 0;
1845 err = fib_lookup(net, &fl4, &res, 0);
1847 if (!IN_DEV_FORWARD(in_dev))
1848 err = -EHOSTUNREACH;
1852 if (res.type == RTN_BROADCAST)
1855 if (res.type == RTN_LOCAL) {
1856 err = fib_validate_source(skb, saddr, daddr, tos,
1857 0, dev, in_dev, &itag);
1859 goto martian_source;
1863 if (!IN_DEV_FORWARD(in_dev)) {
1864 err = -EHOSTUNREACH;
1867 if (res.type != RTN_UNICAST)
1868 goto martian_destination;
1870 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1874 if (skb->protocol != htons(ETH_P_IP))
1877 if (!ipv4_is_zeronet(saddr)) {
1878 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1881 goto martian_source;
1883 flags |= RTCF_BROADCAST;
1884 res.type = RTN_BROADCAST;
1885 RT_CACHE_STAT_INC(in_brd);
1891 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1892 if (rt_cache_valid(rth)) {
1893 skb_dst_set_noref(skb, &rth->dst);
1901 rth = rt_dst_alloc(net->loopback_dev, flags | RTCF_LOCAL, res.type,
1902 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1906 rth->dst.output= ip_rt_bug;
1907 #ifdef CONFIG_IP_ROUTE_CLASSID
1908 rth->dst.tclassid = itag;
1910 rth->rt_is_input = 1;
1912 rth->rt_table_id = res.table->tb_id;
1914 RT_CACHE_STAT_INC(in_slow_tot);
1915 if (res.type == RTN_UNREACHABLE) {
1916 rth->dst.input= ip_error;
1917 rth->dst.error= -err;
1918 rth->rt_flags &= ~RTCF_LOCAL;
1921 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1922 rth->dst.flags |= DST_NOCACHE;
1923 rt_add_uncached_list(rth);
1926 skb_dst_set(skb, &rth->dst);
1931 RT_CACHE_STAT_INC(in_no_route);
1932 res.type = RTN_UNREACHABLE;
1938 * Do not cache martian addresses: they should be logged (RFC1812)
1940 martian_destination:
1941 RT_CACHE_STAT_INC(in_martian_dst);
1942 #ifdef CONFIG_IP_ROUTE_VERBOSE
1943 if (IN_DEV_LOG_MARTIANS(in_dev))
1944 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1945 &daddr, &saddr, dev->name);
1957 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1961 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1962 u8 tos, struct net_device *dev)
1968 /* Multicast recognition logic is moved from route cache to here.
1969 The problem was that too many Ethernet cards have broken/missing
1970 hardware multicast filters :-( As result the host on multicasting
1971 network acquires a lot of useless route cache entries, sort of
1972 SDR messages from all the world. Now we try to get rid of them.
1973 Really, provided software IP multicast filter is organized
1974 reasonably (at least, hashed), it does not result in a slowdown
1975 comparing with route cache reject entries.
1976 Note, that multicast routers are not affected, because
1977 route cache entry is created eventually.
1979 if (ipv4_is_multicast(daddr)) {
1980 struct in_device *in_dev = __in_dev_get_rcu(dev);
1983 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1984 ip_hdr(skb)->protocol);
1986 #ifdef CONFIG_IP_MROUTE
1988 (!ipv4_is_local_multicast(daddr) &&
1989 IN_DEV_MFORWARD(in_dev))
1992 int res = ip_route_input_mc(skb, daddr, saddr,
2001 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2005 EXPORT_SYMBOL(ip_route_input_noref);
2007 /* called with rcu_read_lock() */
2008 static struct rtable *__mkroute_output(const struct fib_result *res,
2009 const struct flowi4 *fl4, int orig_oif,
2010 struct net_device *dev_out,
2013 struct fib_info *fi = res->fi;
2014 struct fib_nh_exception *fnhe;
2015 struct in_device *in_dev;
2016 u16 type = res->type;
2020 in_dev = __in_dev_get_rcu(dev_out);
2022 return ERR_PTR(-EINVAL);
2024 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2025 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2026 return ERR_PTR(-EINVAL);
2028 if (ipv4_is_lbcast(fl4->daddr))
2029 type = RTN_BROADCAST;
2030 else if (ipv4_is_multicast(fl4->daddr))
2031 type = RTN_MULTICAST;
2032 else if (ipv4_is_zeronet(fl4->daddr))
2033 return ERR_PTR(-EINVAL);
2035 if (dev_out->flags & IFF_LOOPBACK)
2036 flags |= RTCF_LOCAL;
2039 if (type == RTN_BROADCAST) {
2040 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2042 } else if (type == RTN_MULTICAST) {
2043 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2044 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2046 flags &= ~RTCF_LOCAL;
2049 /* If multicast route do not exist use
2050 * default one, but do not gateway in this case.
2053 if (fi && res->prefixlen < 4)
2055 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2056 (orig_oif != dev_out->ifindex)) {
2057 /* For local routes that require a particular output interface
2058 * we do not want to cache the result. Caching the result
2059 * causes incorrect behaviour when there are multiple source
2060 * addresses on the interface, the end result being that if the
2061 * intended recipient is waiting on that interface for the
2062 * packet he won't receive it because it will be delivered on
2063 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2064 * be set to the loopback interface as well.
2070 do_cache &= fi != NULL;
2072 struct rtable __rcu **prth;
2073 struct fib_nh *nh = &FIB_RES_NH(*res);
2075 fnhe = find_exception(nh, fl4->daddr);
2077 prth = &fnhe->fnhe_rth_output;
2078 rth = rcu_dereference(*prth);
2079 if (rth && rth->dst.expires &&
2080 time_after(jiffies, rth->dst.expires)) {
2081 ip_del_fnhe(nh, fl4->daddr);
2088 if (unlikely(fl4->flowi4_flags &
2089 FLOWI_FLAG_KNOWN_NH &&
2091 nh->nh_scope == RT_SCOPE_LINK))) {
2095 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2096 rth = rcu_dereference(*prth);
2099 if (rt_cache_valid(rth)) {
2100 dst_hold(&rth->dst);
2106 rth = rt_dst_alloc(dev_out, flags, type,
2107 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2108 IN_DEV_CONF_GET(in_dev, NOXFRM),
2111 return ERR_PTR(-ENOBUFS);
2113 rth->rt_iif = orig_oif ? : 0;
2115 rth->rt_table_id = res->table->tb_id;
2117 RT_CACHE_STAT_INC(out_slow_tot);
2119 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2120 if (flags & RTCF_LOCAL &&
2121 !(dev_out->flags & IFF_LOOPBACK)) {
2122 rth->dst.output = ip_mc_output;
2123 RT_CACHE_STAT_INC(out_slow_mc);
2125 #ifdef CONFIG_IP_MROUTE
2126 if (type == RTN_MULTICAST) {
2127 if (IN_DEV_MFORWARD(in_dev) &&
2128 !ipv4_is_local_multicast(fl4->daddr)) {
2129 rth->dst.input = ip_mr_input;
2130 rth->dst.output = ip_mc_output;
2136 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2137 if (lwtunnel_output_redirect(rth->dst.lwtstate))
2138 rth->dst.output = lwtunnel_output;
2144 * Major route resolver routine.
2147 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2150 struct net_device *dev_out = NULL;
2151 __u8 tos = RT_FL_TOS(fl4);
2152 unsigned int flags = 0;
2153 struct fib_result res;
2156 int err = -ENETUNREACH;
2162 orig_oif = fl4->flowi4_oif;
2164 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2165 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2166 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2167 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2171 rth = ERR_PTR(-EINVAL);
2172 if (ipv4_is_multicast(fl4->saddr) ||
2173 ipv4_is_lbcast(fl4->saddr) ||
2174 ipv4_is_zeronet(fl4->saddr))
2177 /* I removed check for oif == dev_out->oif here.
2178 It was wrong for two reasons:
2179 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2180 is assigned to multiple interfaces.
2181 2. Moreover, we are allowed to send packets with saddr
2182 of another iface. --ANK
2185 if (fl4->flowi4_oif == 0 &&
2186 (ipv4_is_multicast(fl4->daddr) ||
2187 ipv4_is_lbcast(fl4->daddr))) {
2188 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2189 dev_out = __ip_dev_find(net, fl4->saddr, false);
2193 /* Special hack: user can direct multicasts
2194 and limited broadcast via necessary interface
2195 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2196 This hack is not just for fun, it allows
2197 vic,vat and friends to work.
2198 They bind socket to loopback, set ttl to zero
2199 and expect that it will work.
2200 From the viewpoint of routing cache they are broken,
2201 because we are not allowed to build multicast path
2202 with loopback source addr (look, routing cache
2203 cannot know, that ttl is zero, so that packet
2204 will not leave this host and route is valid).
2205 Luckily, this hack is good workaround.
2208 fl4->flowi4_oif = dev_out->ifindex;
2212 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2213 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2214 if (!__ip_dev_find(net, fl4->saddr, false))
2220 if (fl4->flowi4_oif) {
2221 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2222 rth = ERR_PTR(-ENODEV);
2226 /* RACE: Check return value of inet_select_addr instead. */
2227 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2228 rth = ERR_PTR(-ENETUNREACH);
2231 if (ipv4_is_local_multicast(fl4->daddr) ||
2232 ipv4_is_lbcast(fl4->daddr) ||
2233 fl4->flowi4_proto == IPPROTO_IGMP) {
2235 fl4->saddr = inet_select_addr(dev_out, 0,
2240 if (ipv4_is_multicast(fl4->daddr))
2241 fl4->saddr = inet_select_addr(dev_out, 0,
2243 else if (!fl4->daddr)
2244 fl4->saddr = inet_select_addr(dev_out, 0,
2248 rth = l3mdev_get_rtable(dev_out, fl4);
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 !netif_index_is_l3_master(net, fl4->flowi4_oif)) {
2270 /* Apparently, routing tables are wrong. Assume,
2271 that the destination is on link.
2274 Because we are allowed to send to iface
2275 even if it has NO routes and NO assigned
2276 addresses. When oif is specified, routing
2277 tables are looked up with only one purpose:
2278 to catch if destination is gatewayed, rather than
2279 direct. Moreover, if MSG_DONTROUTE is set,
2280 we send packet, ignoring both routing tables
2281 and ifaddr state. --ANK
2284 We could make it even if oif is unknown,
2285 likely IPv6, but we do not.
2288 if (fl4->saddr == 0)
2289 fl4->saddr = inet_select_addr(dev_out, 0,
2291 res.type = RTN_UNICAST;
2298 if (res.type == RTN_LOCAL) {
2300 if (res.fi->fib_prefsrc)
2301 fl4->saddr = res.fi->fib_prefsrc;
2303 fl4->saddr = fl4->daddr;
2305 dev_out = net->loopback_dev;
2306 fl4->flowi4_oif = dev_out->ifindex;
2307 flags |= RTCF_LOCAL;
2311 fib_select_path(net, &res, fl4, mp_hash);
2313 dev_out = FIB_RES_DEV(res);
2314 fl4->flowi4_oif = dev_out->ifindex;
2318 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2324 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2326 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2331 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2333 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2335 return mtu ? : dst->dev->mtu;
2338 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2339 struct sk_buff *skb, u32 mtu)
2343 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2344 struct sk_buff *skb)
2348 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2354 static struct dst_ops ipv4_dst_blackhole_ops = {
2356 .check = ipv4_blackhole_dst_check,
2357 .mtu = ipv4_blackhole_mtu,
2358 .default_advmss = ipv4_default_advmss,
2359 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2360 .redirect = ipv4_rt_blackhole_redirect,
2361 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2362 .neigh_lookup = ipv4_neigh_lookup,
2365 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2367 struct rtable *ort = (struct rtable *) dst_orig;
2370 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2372 struct dst_entry *new = &rt->dst;
2375 new->input = dst_discard;
2376 new->output = dst_discard_out;
2378 new->dev = ort->dst.dev;
2382 rt->rt_is_input = ort->rt_is_input;
2383 rt->rt_iif = ort->rt_iif;
2384 rt->rt_pmtu = ort->rt_pmtu;
2386 rt->rt_genid = rt_genid_ipv4(net);
2387 rt->rt_flags = ort->rt_flags;
2388 rt->rt_type = ort->rt_type;
2389 rt->rt_gateway = ort->rt_gateway;
2390 rt->rt_uses_gateway = ort->rt_uses_gateway;
2392 INIT_LIST_HEAD(&rt->rt_uncached);
2396 dst_release(dst_orig);
2398 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2401 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2402 const struct sock *sk)
2404 struct rtable *rt = __ip_route_output_key(net, flp4);
2409 if (flp4->flowi4_proto)
2410 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2411 flowi4_to_flowi(flp4),
2416 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2418 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2419 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2420 u32 seq, int event, int nowait, unsigned int flags)
2422 struct rtable *rt = skb_rtable(skb);
2424 struct nlmsghdr *nlh;
2425 unsigned long expires = 0;
2427 u32 metrics[RTAX_MAX];
2429 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2433 r = nlmsg_data(nlh);
2434 r->rtm_family = AF_INET;
2435 r->rtm_dst_len = 32;
2437 r->rtm_tos = fl4->flowi4_tos;
2438 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2439 if (nla_put_u32(skb, RTA_TABLE, table_id))
2440 goto nla_put_failure;
2441 r->rtm_type = rt->rt_type;
2442 r->rtm_scope = RT_SCOPE_UNIVERSE;
2443 r->rtm_protocol = RTPROT_UNSPEC;
2444 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2445 if (rt->rt_flags & RTCF_NOTIFY)
2446 r->rtm_flags |= RTM_F_NOTIFY;
2447 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2448 r->rtm_flags |= RTCF_DOREDIRECT;
2450 if (nla_put_in_addr(skb, RTA_DST, dst))
2451 goto nla_put_failure;
2453 r->rtm_src_len = 32;
2454 if (nla_put_in_addr(skb, RTA_SRC, src))
2455 goto nla_put_failure;
2458 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2459 goto nla_put_failure;
2460 #ifdef CONFIG_IP_ROUTE_CLASSID
2461 if (rt->dst.tclassid &&
2462 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2463 goto nla_put_failure;
2465 if (!rt_is_input_route(rt) &&
2466 fl4->saddr != src) {
2467 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2468 goto nla_put_failure;
2470 if (rt->rt_uses_gateway &&
2471 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2472 goto nla_put_failure;
2474 expires = rt->dst.expires;
2476 unsigned long now = jiffies;
2478 if (time_before(now, expires))
2484 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2485 if (rt->rt_pmtu && expires)
2486 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2487 if (rtnetlink_put_metrics(skb, metrics) < 0)
2488 goto nla_put_failure;
2490 if (fl4->flowi4_mark &&
2491 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2492 goto nla_put_failure;
2494 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2495 nla_put_u32(skb, RTA_UID,
2496 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2497 goto nla_put_failure;
2499 error = rt->dst.error;
2501 if (rt_is_input_route(rt)) {
2502 #ifdef CONFIG_IP_MROUTE
2503 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2504 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2505 int err = ipmr_get_route(net, skb,
2506 fl4->saddr, fl4->daddr,
2513 goto nla_put_failure;
2515 if (err == -EMSGSIZE)
2516 goto nla_put_failure;
2522 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2523 goto nla_put_failure;
2526 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2527 goto nla_put_failure;
2529 nlmsg_end(skb, nlh);
2533 nlmsg_cancel(skb, nlh);
2537 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2539 struct net *net = sock_net(in_skb->sk);
2541 struct nlattr *tb[RTA_MAX+1];
2542 struct rtable *rt = NULL;
2549 struct sk_buff *skb;
2550 u32 table_id = RT_TABLE_MAIN;
2553 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2557 rtm = nlmsg_data(nlh);
2559 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2565 /* Reserve room for dummy headers, this skb can pass
2566 through good chunk of routing engine.
2568 skb_reset_mac_header(skb);
2569 skb_reset_network_header(skb);
2571 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2572 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2573 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2575 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2576 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2577 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2578 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2580 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2582 uid = (iif ? INVALID_UID : current_uid());
2584 memset(&fl4, 0, sizeof(fl4));
2587 fl4.flowi4_tos = rtm->rtm_tos;
2588 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2589 fl4.flowi4_mark = mark;
2590 fl4.flowi4_uid = uid;
2592 if (netif_index_is_l3_master(net, fl4.flowi4_oif))
2593 fl4.flowi4_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF;
2596 struct net_device *dev;
2598 dev = __dev_get_by_index(net, iif);
2604 skb->protocol = htons(ETH_P_IP);
2608 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2611 rt = skb_rtable(skb);
2612 if (err == 0 && rt->dst.error)
2613 err = -rt->dst.error;
2615 rt = ip_route_output_key(net, &fl4);
2625 skb_dst_set(skb, &rt->dst);
2626 if (rtm->rtm_flags & RTM_F_NOTIFY)
2627 rt->rt_flags |= RTCF_NOTIFY;
2629 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2630 table_id = rt->rt_table_id;
2632 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2633 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2634 RTM_NEWROUTE, 0, 0);
2638 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2647 void ip_rt_multicast_event(struct in_device *in_dev)
2649 rt_cache_flush(dev_net(in_dev->dev));
2652 #ifdef CONFIG_SYSCTL
2653 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2654 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2655 static int ip_rt_gc_elasticity __read_mostly = 8;
2657 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2658 void __user *buffer,
2659 size_t *lenp, loff_t *ppos)
2661 struct net *net = (struct net *)__ctl->extra1;
2664 rt_cache_flush(net);
2665 fnhe_genid_bump(net);
2672 static struct ctl_table ipv4_route_table[] = {
2674 .procname = "gc_thresh",
2675 .data = &ipv4_dst_ops.gc_thresh,
2676 .maxlen = sizeof(int),
2678 .proc_handler = proc_dointvec,
2681 .procname = "max_size",
2682 .data = &ip_rt_max_size,
2683 .maxlen = sizeof(int),
2685 .proc_handler = proc_dointvec,
2688 /* Deprecated. Use gc_min_interval_ms */
2690 .procname = "gc_min_interval",
2691 .data = &ip_rt_gc_min_interval,
2692 .maxlen = sizeof(int),
2694 .proc_handler = proc_dointvec_jiffies,
2697 .procname = "gc_min_interval_ms",
2698 .data = &ip_rt_gc_min_interval,
2699 .maxlen = sizeof(int),
2701 .proc_handler = proc_dointvec_ms_jiffies,
2704 .procname = "gc_timeout",
2705 .data = &ip_rt_gc_timeout,
2706 .maxlen = sizeof(int),
2708 .proc_handler = proc_dointvec_jiffies,
2711 .procname = "gc_interval",
2712 .data = &ip_rt_gc_interval,
2713 .maxlen = sizeof(int),
2715 .proc_handler = proc_dointvec_jiffies,
2718 .procname = "redirect_load",
2719 .data = &ip_rt_redirect_load,
2720 .maxlen = sizeof(int),
2722 .proc_handler = proc_dointvec,
2725 .procname = "redirect_number",
2726 .data = &ip_rt_redirect_number,
2727 .maxlen = sizeof(int),
2729 .proc_handler = proc_dointvec,
2732 .procname = "redirect_silence",
2733 .data = &ip_rt_redirect_silence,
2734 .maxlen = sizeof(int),
2736 .proc_handler = proc_dointvec,
2739 .procname = "error_cost",
2740 .data = &ip_rt_error_cost,
2741 .maxlen = sizeof(int),
2743 .proc_handler = proc_dointvec,
2746 .procname = "error_burst",
2747 .data = &ip_rt_error_burst,
2748 .maxlen = sizeof(int),
2750 .proc_handler = proc_dointvec,
2753 .procname = "gc_elasticity",
2754 .data = &ip_rt_gc_elasticity,
2755 .maxlen = sizeof(int),
2757 .proc_handler = proc_dointvec,
2760 .procname = "mtu_expires",
2761 .data = &ip_rt_mtu_expires,
2762 .maxlen = sizeof(int),
2764 .proc_handler = proc_dointvec_jiffies,
2767 .procname = "min_pmtu",
2768 .data = &ip_rt_min_pmtu,
2769 .maxlen = sizeof(int),
2771 .proc_handler = proc_dointvec,
2774 .procname = "min_adv_mss",
2775 .data = &ip_rt_min_advmss,
2776 .maxlen = sizeof(int),
2778 .proc_handler = proc_dointvec,
2783 static struct ctl_table ipv4_route_flush_table[] = {
2785 .procname = "flush",
2786 .maxlen = sizeof(int),
2788 .proc_handler = ipv4_sysctl_rtcache_flush,
2793 static __net_init int sysctl_route_net_init(struct net *net)
2795 struct ctl_table *tbl;
2797 tbl = ipv4_route_flush_table;
2798 if (!net_eq(net, &init_net)) {
2799 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2803 /* Don't export sysctls to unprivileged users */
2804 if (net->user_ns != &init_user_ns)
2805 tbl[0].procname = NULL;
2807 tbl[0].extra1 = net;
2809 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2810 if (!net->ipv4.route_hdr)
2815 if (tbl != ipv4_route_flush_table)
2821 static __net_exit void sysctl_route_net_exit(struct net *net)
2823 struct ctl_table *tbl;
2825 tbl = net->ipv4.route_hdr->ctl_table_arg;
2826 unregister_net_sysctl_table(net->ipv4.route_hdr);
2827 BUG_ON(tbl == ipv4_route_flush_table);
2831 static __net_initdata struct pernet_operations sysctl_route_ops = {
2832 .init = sysctl_route_net_init,
2833 .exit = sysctl_route_net_exit,
2837 static __net_init int rt_genid_init(struct net *net)
2839 atomic_set(&net->ipv4.rt_genid, 0);
2840 atomic_set(&net->fnhe_genid, 0);
2841 get_random_bytes(&net->ipv4.dev_addr_genid,
2842 sizeof(net->ipv4.dev_addr_genid));
2846 static __net_initdata struct pernet_operations rt_genid_ops = {
2847 .init = rt_genid_init,
2850 static int __net_init ipv4_inetpeer_init(struct net *net)
2852 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2856 inet_peer_base_init(bp);
2857 net->ipv4.peers = bp;
2861 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2863 struct inet_peer_base *bp = net->ipv4.peers;
2865 net->ipv4.peers = NULL;
2866 inetpeer_invalidate_tree(bp);
2870 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2871 .init = ipv4_inetpeer_init,
2872 .exit = ipv4_inetpeer_exit,
2875 #ifdef CONFIG_IP_ROUTE_CLASSID
2876 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2877 #endif /* CONFIG_IP_ROUTE_CLASSID */
2879 int __init ip_rt_init(void)
2884 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2886 panic("IP: failed to allocate ip_idents\n");
2888 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2890 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2892 panic("IP: failed to allocate ip_tstamps\n");
2894 for_each_possible_cpu(cpu) {
2895 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2897 INIT_LIST_HEAD(&ul->head);
2898 spin_lock_init(&ul->lock);
2900 #ifdef CONFIG_IP_ROUTE_CLASSID
2901 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2903 panic("IP: failed to allocate ip_rt_acct\n");
2906 ipv4_dst_ops.kmem_cachep =
2907 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2908 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2910 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2912 if (dst_entries_init(&ipv4_dst_ops) < 0)
2913 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2915 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2916 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2918 ipv4_dst_ops.gc_thresh = ~0;
2919 ip_rt_max_size = INT_MAX;
2924 if (ip_rt_proc_init())
2925 pr_err("Unable to create route proc files\n");
2930 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2932 #ifdef CONFIG_SYSCTL
2933 register_pernet_subsys(&sysctl_route_ops);
2935 register_pernet_subsys(&rt_genid_ops);
2936 register_pernet_subsys(&ipv4_inetpeer_ops);
2940 #ifdef CONFIG_SYSCTL
2942 * We really need to sanitize the damn ipv4 init order, then all
2943 * this nonsense will go away.
2945 void __init ip_static_sysctl_init(void)
2947 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);