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>
93 #include <net/net_namespace.h>
94 #include <net/protocol.h>
96 #include <net/route.h>
97 #include <net/inetpeer.h>
99 #include <net/ip_fib.h>
102 #include <net/icmp.h>
103 #include <net/xfrm.h>
104 #include <net/netevent.h>
105 #include <net/rtnetlink.h>
107 #include <linux/sysctl.h>
108 #include <linux/kmemleak.h>
110 #include <net/secure_seq.h>
112 #define RT_FL_TOS(oldflp4) \
113 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
115 #define IP_MAX_MTU 0xFFF0
117 #define RT_GC_TIMEOUT (300*HZ)
119 static int ip_rt_max_size;
120 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
121 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
122 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
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_gc_elasticity __read_mostly = 8;
129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly = 256;
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 void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
153 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
159 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
163 static struct dst_ops ipv4_dst_ops = {
165 .protocol = cpu_to_be16(ETH_P_IP),
166 .check = ipv4_dst_check,
167 .default_advmss = ipv4_default_advmss,
169 .cow_metrics = ipv4_cow_metrics,
170 .destroy = ipv4_dst_destroy,
171 .ifdown = ipv4_dst_ifdown,
172 .negative_advice = ipv4_negative_advice,
173 .link_failure = ipv4_link_failure,
174 .update_pmtu = ip_rt_update_pmtu,
175 .redirect = ip_do_redirect,
176 .local_out = __ip_local_out,
177 .neigh_lookup = ipv4_neigh_lookup,
180 #define ECN_OR_COST(class) TC_PRIO_##class
182 const __u8 ip_tos2prio[16] = {
184 ECN_OR_COST(BESTEFFORT),
186 ECN_OR_COST(BESTEFFORT),
192 ECN_OR_COST(INTERACTIVE),
194 ECN_OR_COST(INTERACTIVE),
195 TC_PRIO_INTERACTIVE_BULK,
196 ECN_OR_COST(INTERACTIVE_BULK),
197 TC_PRIO_INTERACTIVE_BULK,
198 ECN_OR_COST(INTERACTIVE_BULK)
200 EXPORT_SYMBOL(ip_tos2prio);
202 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
203 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
205 #ifdef CONFIG_PROC_FS
206 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
210 return SEQ_START_TOKEN;
213 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
219 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
223 static int rt_cache_seq_show(struct seq_file *seq, void *v)
225 if (v == SEQ_START_TOKEN)
226 seq_printf(seq, "%-127s\n",
227 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
228 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
233 static const struct seq_operations rt_cache_seq_ops = {
234 .start = rt_cache_seq_start,
235 .next = rt_cache_seq_next,
236 .stop = rt_cache_seq_stop,
237 .show = rt_cache_seq_show,
240 static int rt_cache_seq_open(struct inode *inode, struct file *file)
242 return seq_open(file, &rt_cache_seq_ops);
245 static const struct file_operations rt_cache_seq_fops = {
246 .owner = THIS_MODULE,
247 .open = rt_cache_seq_open,
250 .release = seq_release,
254 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
259 return SEQ_START_TOKEN;
261 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
262 if (!cpu_possible(cpu))
265 return &per_cpu(rt_cache_stat, cpu);
270 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
274 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
275 if (!cpu_possible(cpu))
278 return &per_cpu(rt_cache_stat, cpu);
284 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
289 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
291 struct rt_cache_stat *st = v;
293 if (v == SEQ_START_TOKEN) {
294 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");
298 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
299 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
300 dst_entries_get_slow(&ipv4_dst_ops),
323 static const struct seq_operations rt_cpu_seq_ops = {
324 .start = rt_cpu_seq_start,
325 .next = rt_cpu_seq_next,
326 .stop = rt_cpu_seq_stop,
327 .show = rt_cpu_seq_show,
331 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
333 return seq_open(file, &rt_cpu_seq_ops);
336 static const struct file_operations rt_cpu_seq_fops = {
337 .owner = THIS_MODULE,
338 .open = rt_cpu_seq_open,
341 .release = seq_release,
344 #ifdef CONFIG_IP_ROUTE_CLASSID
345 static int rt_acct_proc_show(struct seq_file *m, void *v)
347 struct ip_rt_acct *dst, *src;
350 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
354 for_each_possible_cpu(i) {
355 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
356 for (j = 0; j < 256; j++) {
357 dst[j].o_bytes += src[j].o_bytes;
358 dst[j].o_packets += src[j].o_packets;
359 dst[j].i_bytes += src[j].i_bytes;
360 dst[j].i_packets += src[j].i_packets;
364 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
369 static int rt_acct_proc_open(struct inode *inode, struct file *file)
371 return single_open(file, rt_acct_proc_show, NULL);
374 static const struct file_operations rt_acct_proc_fops = {
375 .owner = THIS_MODULE,
376 .open = rt_acct_proc_open,
379 .release = single_release,
383 static int __net_init ip_rt_do_proc_init(struct net *net)
385 struct proc_dir_entry *pde;
387 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
392 pde = proc_create("rt_cache", S_IRUGO,
393 net->proc_net_stat, &rt_cpu_seq_fops);
397 #ifdef CONFIG_IP_ROUTE_CLASSID
398 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
404 #ifdef CONFIG_IP_ROUTE_CLASSID
406 remove_proc_entry("rt_cache", net->proc_net_stat);
409 remove_proc_entry("rt_cache", net->proc_net);
414 static void __net_exit ip_rt_do_proc_exit(struct net *net)
416 remove_proc_entry("rt_cache", net->proc_net_stat);
417 remove_proc_entry("rt_cache", net->proc_net);
418 #ifdef CONFIG_IP_ROUTE_CLASSID
419 remove_proc_entry("rt_acct", net->proc_net);
423 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
424 .init = ip_rt_do_proc_init,
425 .exit = ip_rt_do_proc_exit,
428 static int __init ip_rt_proc_init(void)
430 return register_pernet_subsys(&ip_rt_proc_ops);
434 static inline int ip_rt_proc_init(void)
438 #endif /* CONFIG_PROC_FS */
440 static inline bool rt_is_expired(const struct rtable *rth)
442 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
445 void rt_cache_flush(struct net *net)
450 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
454 struct net_device *dev = dst->dev;
455 const __be32 *pkey = daddr;
456 const struct rtable *rt;
459 rt = (const struct rtable *) dst;
461 pkey = (const __be32 *) &rt->rt_gateway;
463 pkey = &ip_hdr(skb)->daddr;
465 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
468 return neigh_create(&arp_tbl, pkey, dev);
472 * Peer allocation may fail only in serious out-of-memory conditions. However
473 * we still can generate some output.
474 * Random ID selection looks a bit dangerous because we have no chances to
475 * select ID being unique in a reasonable period of time.
476 * But broken packet identifier may be better than no packet at all.
478 static void ip_select_fb_ident(struct iphdr *iph)
480 static DEFINE_SPINLOCK(ip_fb_id_lock);
481 static u32 ip_fallback_id;
484 spin_lock_bh(&ip_fb_id_lock);
485 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
486 iph->id = htons(salt & 0xFFFF);
487 ip_fallback_id = salt;
488 spin_unlock_bh(&ip_fb_id_lock);
491 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
493 struct net *net = dev_net(dst->dev);
494 struct inet_peer *peer;
496 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
498 iph->id = htons(inet_getid(peer, more));
503 ip_select_fb_ident(iph);
505 EXPORT_SYMBOL(__ip_select_ident);
507 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
508 const struct iphdr *iph,
510 u8 prot, u32 mark, int flow_flags)
513 const struct inet_sock *inet = inet_sk(sk);
515 oif = sk->sk_bound_dev_if;
517 tos = RT_CONN_FLAGS(sk);
518 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
520 flowi4_init_output(fl4, oif, mark, tos,
521 RT_SCOPE_UNIVERSE, prot,
523 iph->daddr, iph->saddr, 0, 0);
526 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
527 const struct sock *sk)
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(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);
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 struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
574 struct fib_nh_exception *fnhe, *oldest;
577 oldest = rcu_dereference(hash->chain);
578 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
579 fnhe = rcu_dereference(fnhe->fnhe_next)) {
580 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
583 orig = rcu_dereference(oldest->fnhe_rth);
585 RCU_INIT_POINTER(oldest->fnhe_rth, NULL);
591 static inline u32 fnhe_hashfun(__be32 daddr)
595 hval = (__force u32) daddr;
596 hval ^= (hval >> 11) ^ (hval >> 22);
598 return hval & (FNHE_HASH_SIZE - 1);
601 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
602 u32 pmtu, unsigned long expires)
604 struct fnhe_hash_bucket *hash;
605 struct fib_nh_exception *fnhe;
607 u32 hval = fnhe_hashfun(daddr);
609 spin_lock_bh(&fnhe_lock);
611 hash = nh->nh_exceptions;
613 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
616 nh->nh_exceptions = hash;
622 for (fnhe = rcu_dereference(hash->chain); fnhe;
623 fnhe = rcu_dereference(fnhe->fnhe_next)) {
624 if (fnhe->fnhe_daddr == daddr)
633 fnhe->fnhe_pmtu = pmtu;
634 fnhe->fnhe_expires = expires;
637 if (depth > FNHE_RECLAIM_DEPTH)
638 fnhe = fnhe_oldest(hash);
640 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
644 fnhe->fnhe_next = hash->chain;
645 rcu_assign_pointer(hash->chain, fnhe);
647 fnhe->fnhe_daddr = daddr;
649 fnhe->fnhe_pmtu = pmtu;
650 fnhe->fnhe_expires = expires;
653 fnhe->fnhe_stamp = jiffies;
656 spin_unlock_bh(&fnhe_lock);
660 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
663 __be32 new_gw = icmp_hdr(skb)->un.gateway;
664 __be32 old_gw = ip_hdr(skb)->saddr;
665 struct net_device *dev = skb->dev;
666 struct in_device *in_dev;
667 struct fib_result res;
671 switch (icmp_hdr(skb)->code & 7) {
673 case ICMP_REDIR_NETTOS:
674 case ICMP_REDIR_HOST:
675 case ICMP_REDIR_HOSTTOS:
682 if (rt->rt_gateway != old_gw)
685 in_dev = __in_dev_get_rcu(dev);
690 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
691 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
692 ipv4_is_zeronet(new_gw))
693 goto reject_redirect;
695 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
696 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
697 goto reject_redirect;
698 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
699 goto reject_redirect;
701 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
702 goto reject_redirect;
705 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
707 if (!(n->nud_state & NUD_VALID)) {
708 neigh_event_send(n, NULL);
710 if (fib_lookup(net, fl4, &res) == 0) {
711 struct fib_nh *nh = &FIB_RES_NH(res);
713 update_or_create_fnhe(nh, fl4->daddr, new_gw,
717 rt->dst.obsolete = DST_OBSOLETE_KILL;
718 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
725 #ifdef CONFIG_IP_ROUTE_VERBOSE
726 if (IN_DEV_LOG_MARTIANS(in_dev)) {
727 const struct iphdr *iph = (const struct iphdr *) skb->data;
728 __be32 daddr = iph->daddr;
729 __be32 saddr = iph->saddr;
731 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
732 " Advised path = %pI4 -> %pI4\n",
733 &old_gw, dev->name, &new_gw,
740 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
745 rt = (struct rtable *) dst;
747 ip_rt_build_flow_key(&fl4, sk, skb);
748 __ip_do_redirect(rt, skb, &fl4, true);
751 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
753 struct rtable *rt = (struct rtable *)dst;
754 struct dst_entry *ret = dst;
757 if (dst->obsolete > 0) {
760 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
771 * 1. The first ip_rt_redirect_number redirects are sent
772 * with exponential backoff, then we stop sending them at all,
773 * assuming that the host ignores our redirects.
774 * 2. If we did not see packets requiring redirects
775 * during ip_rt_redirect_silence, we assume that the host
776 * forgot redirected route and start to send redirects again.
778 * This algorithm is much cheaper and more intelligent than dumb load limiting
781 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
782 * and "frag. need" (breaks PMTU discovery) in icmp.c.
785 void ip_rt_send_redirect(struct sk_buff *skb)
787 struct rtable *rt = skb_rtable(skb);
788 struct in_device *in_dev;
789 struct inet_peer *peer;
794 in_dev = __in_dev_get_rcu(rt->dst.dev);
795 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
799 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
802 net = dev_net(rt->dst.dev);
803 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
805 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
806 rt_nexthop(rt, ip_hdr(skb)->daddr));
810 /* No redirected packets during ip_rt_redirect_silence;
811 * reset the algorithm.
813 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
814 peer->rate_tokens = 0;
816 /* Too many ignored redirects; do not send anything
817 * set dst.rate_last to the last seen redirected packet.
819 if (peer->rate_tokens >= ip_rt_redirect_number) {
820 peer->rate_last = jiffies;
824 /* Check for load limit; set rate_last to the latest sent
827 if (peer->rate_tokens == 0 ||
830 (ip_rt_redirect_load << peer->rate_tokens)))) {
831 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
833 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
834 peer->rate_last = jiffies;
836 #ifdef CONFIG_IP_ROUTE_VERBOSE
838 peer->rate_tokens == ip_rt_redirect_number)
839 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
840 &ip_hdr(skb)->saddr, inet_iif(skb),
841 &ip_hdr(skb)->daddr, &gw);
848 static int ip_error(struct sk_buff *skb)
850 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
851 struct rtable *rt = skb_rtable(skb);
852 struct inet_peer *peer;
858 net = dev_net(rt->dst.dev);
859 if (!IN_DEV_FORWARD(in_dev)) {
860 switch (rt->dst.error) {
862 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
866 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
872 switch (rt->dst.error) {
877 code = ICMP_HOST_UNREACH;
880 code = ICMP_NET_UNREACH;
881 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
884 code = ICMP_PKT_FILTERED;
888 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
893 peer->rate_tokens += now - peer->rate_last;
894 if (peer->rate_tokens > ip_rt_error_burst)
895 peer->rate_tokens = ip_rt_error_burst;
896 peer->rate_last = now;
897 if (peer->rate_tokens >= ip_rt_error_cost)
898 peer->rate_tokens -= ip_rt_error_cost;
904 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
910 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
912 struct dst_entry *dst = &rt->dst;
913 struct fib_result res;
915 if (dst_metric_locked(dst, RTAX_MTU))
918 if (dst->dev->mtu < mtu)
921 if (mtu < ip_rt_min_pmtu)
922 mtu = ip_rt_min_pmtu;
925 dst->obsolete = DST_OBSOLETE_KILL;
928 dst->expires = max(1UL, jiffies + ip_rt_mtu_expires);
932 if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
933 struct fib_nh *nh = &FIB_RES_NH(res);
935 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
936 jiffies + ip_rt_mtu_expires);
941 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
942 struct sk_buff *skb, u32 mtu)
944 struct rtable *rt = (struct rtable *) dst;
947 ip_rt_build_flow_key(&fl4, sk, skb);
948 __ip_rt_update_pmtu(rt, &fl4, mtu);
951 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
952 int oif, u32 mark, u8 protocol, int flow_flags)
954 const struct iphdr *iph = (const struct iphdr *) skb->data;
958 __build_flow_key(&fl4, NULL, iph, oif,
959 RT_TOS(iph->tos), protocol, mark, flow_flags);
960 rt = __ip_route_output_key(net, &fl4);
962 __ip_rt_update_pmtu(rt, &fl4, mtu);
966 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
968 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
970 const struct iphdr *iph = (const struct iphdr *) skb->data;
974 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
975 rt = __ip_route_output_key(sock_net(sk), &fl4);
977 __ip_rt_update_pmtu(rt, &fl4, mtu);
981 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
983 void ipv4_redirect(struct sk_buff *skb, struct net *net,
984 int oif, u32 mark, u8 protocol, int flow_flags)
986 const struct iphdr *iph = (const struct iphdr *) skb->data;
990 __build_flow_key(&fl4, NULL, iph, oif,
991 RT_TOS(iph->tos), protocol, mark, flow_flags);
992 rt = __ip_route_output_key(net, &fl4);
994 __ip_do_redirect(rt, skb, &fl4, false);
998 EXPORT_SYMBOL_GPL(ipv4_redirect);
1000 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1002 const struct iphdr *iph = (const struct iphdr *) skb->data;
1006 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1007 rt = __ip_route_output_key(sock_net(sk), &fl4);
1009 __ip_do_redirect(rt, skb, &fl4, false);
1013 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1015 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1017 struct rtable *rt = (struct rtable *) dst;
1019 /* All IPV4 dsts are created with ->obsolete set to the value
1020 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1021 * into this function always.
1023 * When a PMTU/redirect information update invalidates a
1024 * route, this is indicated by setting obsolete to
1025 * DST_OBSOLETE_KILL.
1027 if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1032 static void ipv4_link_failure(struct sk_buff *skb)
1036 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1038 rt = skb_rtable(skb);
1040 dst_set_expires(&rt->dst, 0);
1043 static int ip_rt_bug(struct sk_buff *skb)
1045 pr_debug("%s: %pI4 -> %pI4, %s\n",
1046 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1047 skb->dev ? skb->dev->name : "?");
1054 We do not cache source address of outgoing interface,
1055 because it is used only by IP RR, TS and SRR options,
1056 so that it out of fast path.
1058 BTW remember: "addr" is allowed to be not aligned
1062 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1066 if (rt_is_output_route(rt))
1067 src = ip_hdr(skb)->saddr;
1069 struct fib_result res;
1075 memset(&fl4, 0, sizeof(fl4));
1076 fl4.daddr = iph->daddr;
1077 fl4.saddr = iph->saddr;
1078 fl4.flowi4_tos = RT_TOS(iph->tos);
1079 fl4.flowi4_oif = rt->dst.dev->ifindex;
1080 fl4.flowi4_iif = skb->dev->ifindex;
1081 fl4.flowi4_mark = skb->mark;
1084 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1085 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1087 src = inet_select_addr(rt->dst.dev,
1088 rt_nexthop(rt, iph->daddr),
1092 memcpy(addr, &src, 4);
1095 #ifdef CONFIG_IP_ROUTE_CLASSID
1096 static void set_class_tag(struct rtable *rt, u32 tag)
1098 if (!(rt->dst.tclassid & 0xFFFF))
1099 rt->dst.tclassid |= tag & 0xFFFF;
1100 if (!(rt->dst.tclassid & 0xFFFF0000))
1101 rt->dst.tclassid |= tag & 0xFFFF0000;
1105 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1107 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1110 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1112 if (advmss > 65535 - 40)
1113 advmss = 65535 - 40;
1118 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1120 const struct rtable *rt = (const struct rtable *) dst;
1121 unsigned int mtu = rt->rt_pmtu;
1123 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1124 mtu = dst_metric_raw(dst, RTAX_MTU);
1129 mtu = dst->dev->mtu;
1131 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1132 if (rt->rt_uses_gateway && mtu > 576)
1136 if (mtu > IP_MAX_MTU)
1142 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1144 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1145 struct fib_nh_exception *fnhe;
1151 hval = fnhe_hashfun(daddr);
1153 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1154 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1155 if (fnhe->fnhe_daddr == daddr)
1161 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1166 spin_lock_bh(&fnhe_lock);
1168 if (daddr == fnhe->fnhe_daddr) {
1169 struct rtable *orig = rcu_dereference(fnhe->fnhe_rth);
1170 if (orig && rt_is_expired(orig)) {
1172 fnhe->fnhe_pmtu = 0;
1173 fnhe->fnhe_expires = 0;
1175 if (fnhe->fnhe_pmtu) {
1176 unsigned long expires = fnhe->fnhe_expires;
1177 unsigned long diff = expires - jiffies;
1179 if (time_before(jiffies, expires)) {
1180 rt->rt_pmtu = fnhe->fnhe_pmtu;
1181 dst_set_expires(&rt->dst, diff);
1184 if (fnhe->fnhe_gw) {
1185 rt->rt_flags |= RTCF_REDIRECTED;
1186 rt->rt_gateway = fnhe->fnhe_gw;
1187 rt->rt_uses_gateway = 1;
1188 } else if (!rt->rt_gateway)
1189 rt->rt_gateway = daddr;
1191 rcu_assign_pointer(fnhe->fnhe_rth, rt);
1195 fnhe->fnhe_stamp = jiffies;
1198 spin_unlock_bh(&fnhe_lock);
1203 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1205 struct rtable *orig, *prev, **p;
1208 if (rt_is_input_route(rt)) {
1209 p = (struct rtable **)&nh->nh_rth_input;
1211 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1215 prev = cmpxchg(p, orig, rt);
1225 static DEFINE_SPINLOCK(rt_uncached_lock);
1226 static LIST_HEAD(rt_uncached_list);
1228 static void rt_add_uncached_list(struct rtable *rt)
1230 spin_lock_bh(&rt_uncached_lock);
1231 list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1232 spin_unlock_bh(&rt_uncached_lock);
1235 static void ipv4_dst_destroy(struct dst_entry *dst)
1237 struct rtable *rt = (struct rtable *) dst;
1239 if (!list_empty(&rt->rt_uncached)) {
1240 spin_lock_bh(&rt_uncached_lock);
1241 list_del(&rt->rt_uncached);
1242 spin_unlock_bh(&rt_uncached_lock);
1246 void rt_flush_dev(struct net_device *dev)
1248 if (!list_empty(&rt_uncached_list)) {
1249 struct net *net = dev_net(dev);
1252 spin_lock_bh(&rt_uncached_lock);
1253 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1254 if (rt->dst.dev != dev)
1256 rt->dst.dev = net->loopback_dev;
1257 dev_hold(rt->dst.dev);
1260 spin_unlock_bh(&rt_uncached_lock);
1264 static bool rt_cache_valid(const struct rtable *rt)
1267 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1271 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1272 const struct fib_result *res,
1273 struct fib_nh_exception *fnhe,
1274 struct fib_info *fi, u16 type, u32 itag)
1276 bool cached = false;
1279 struct fib_nh *nh = &FIB_RES_NH(*res);
1281 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1282 rt->rt_gateway = nh->nh_gw;
1283 rt->rt_uses_gateway = 1;
1285 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1286 #ifdef CONFIG_IP_ROUTE_CLASSID
1287 rt->dst.tclassid = nh->nh_tclassid;
1290 cached = rt_bind_exception(rt, fnhe, daddr);
1291 else if (!(rt->dst.flags & DST_NOCACHE))
1292 cached = rt_cache_route(nh, rt);
1293 if (unlikely(!cached)) {
1294 /* Routes we intend to cache in nexthop exception or
1295 * FIB nexthop have the DST_NOCACHE bit clear.
1296 * However, if we are unsuccessful at storing this
1297 * route into the cache we really need to set it.
1299 rt->dst.flags |= DST_NOCACHE;
1300 if (!rt->rt_gateway)
1301 rt->rt_gateway = daddr;
1302 rt_add_uncached_list(rt);
1305 rt_add_uncached_list(rt);
1307 #ifdef CONFIG_IP_ROUTE_CLASSID
1308 #ifdef CONFIG_IP_MULTIPLE_TABLES
1309 set_class_tag(rt, res->tclassid);
1311 set_class_tag(rt, itag);
1315 static struct rtable *rt_dst_alloc(struct net_device *dev,
1316 bool nopolicy, bool noxfrm, bool will_cache)
1318 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1319 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1320 (nopolicy ? DST_NOPOLICY : 0) |
1321 (noxfrm ? DST_NOXFRM : 0));
1324 /* called in rcu_read_lock() section */
1325 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1326 u8 tos, struct net_device *dev, int our)
1329 struct in_device *in_dev = __in_dev_get_rcu(dev);
1333 /* Primary sanity checks. */
1338 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1339 skb->protocol != htons(ETH_P_IP))
1342 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1343 if (ipv4_is_loopback(saddr))
1346 if (ipv4_is_zeronet(saddr)) {
1347 if (!ipv4_is_local_multicast(daddr))
1350 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1355 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1356 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1360 #ifdef CONFIG_IP_ROUTE_CLASSID
1361 rth->dst.tclassid = itag;
1363 rth->dst.output = ip_rt_bug;
1365 rth->rt_genid = rt_genid(dev_net(dev));
1366 rth->rt_flags = RTCF_MULTICAST;
1367 rth->rt_type = RTN_MULTICAST;
1368 rth->rt_is_input= 1;
1371 rth->rt_gateway = 0;
1372 rth->rt_uses_gateway = 0;
1373 INIT_LIST_HEAD(&rth->rt_uncached);
1375 rth->dst.input= ip_local_deliver;
1376 rth->rt_flags |= RTCF_LOCAL;
1379 #ifdef CONFIG_IP_MROUTE
1380 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1381 rth->dst.input = ip_mr_input;
1383 RT_CACHE_STAT_INC(in_slow_mc);
1385 skb_dst_set(skb, &rth->dst);
1397 static void ip_handle_martian_source(struct net_device *dev,
1398 struct in_device *in_dev,
1399 struct sk_buff *skb,
1403 RT_CACHE_STAT_INC(in_martian_src);
1404 #ifdef CONFIG_IP_ROUTE_VERBOSE
1405 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1407 * RFC1812 recommendation, if source is martian,
1408 * the only hint is MAC header.
1410 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1411 &daddr, &saddr, dev->name);
1412 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1413 print_hex_dump(KERN_WARNING, "ll header: ",
1414 DUMP_PREFIX_OFFSET, 16, 1,
1415 skb_mac_header(skb),
1416 dev->hard_header_len, true);
1422 /* called in rcu_read_lock() section */
1423 static int __mkroute_input(struct sk_buff *skb,
1424 const struct fib_result *res,
1425 struct in_device *in_dev,
1426 __be32 daddr, __be32 saddr, u32 tos)
1430 struct in_device *out_dev;
1431 unsigned int flags = 0;
1435 /* get a working reference to the output device */
1436 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1437 if (out_dev == NULL) {
1438 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1442 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1443 in_dev->dev, in_dev, &itag);
1445 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1451 do_cache = res->fi && !itag;
1452 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1453 (IN_DEV_SHARED_MEDIA(out_dev) ||
1454 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1455 flags |= RTCF_DOREDIRECT;
1459 if (skb->protocol != htons(ETH_P_IP)) {
1460 /* Not IP (i.e. ARP). Do not create route, if it is
1461 * invalid for proxy arp. DNAT routes are always valid.
1463 * Proxy arp feature have been extended to allow, ARP
1464 * replies back to the same interface, to support
1465 * Private VLAN switch technologies. See arp.c.
1467 if (out_dev == in_dev &&
1468 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1475 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1476 if (rt_cache_valid(rth)) {
1477 skb_dst_set_noref(skb, &rth->dst);
1482 rth = rt_dst_alloc(out_dev->dev,
1483 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1484 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1490 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1491 rth->rt_flags = flags;
1492 rth->rt_type = res->type;
1493 rth->rt_is_input = 1;
1496 rth->rt_gateway = 0;
1497 rth->rt_uses_gateway = 0;
1498 INIT_LIST_HEAD(&rth->rt_uncached);
1500 rth->dst.input = ip_forward;
1501 rth->dst.output = ip_output;
1503 rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1504 skb_dst_set(skb, &rth->dst);
1511 static int ip_mkroute_input(struct sk_buff *skb,
1512 struct fib_result *res,
1513 const struct flowi4 *fl4,
1514 struct in_device *in_dev,
1515 __be32 daddr, __be32 saddr, u32 tos)
1517 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1518 if (res->fi && res->fi->fib_nhs > 1)
1519 fib_select_multipath(res);
1522 /* create a routing cache entry */
1523 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1527 * NOTE. We drop all the packets that has local source
1528 * addresses, because every properly looped back packet
1529 * must have correct destination already attached by output routine.
1531 * Such approach solves two big problems:
1532 * 1. Not simplex devices are handled properly.
1533 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1534 * called with rcu_read_lock()
1537 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1538 u8 tos, struct net_device *dev)
1540 struct fib_result res;
1541 struct in_device *in_dev = __in_dev_get_rcu(dev);
1543 unsigned int flags = 0;
1547 struct net *net = dev_net(dev);
1550 /* IP on this device is disabled. */
1555 /* Check for the most weird martians, which can be not detected
1559 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1560 goto martian_source;
1563 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1566 /* Accept zero addresses only to limited broadcast;
1567 * I even do not know to fix it or not. Waiting for complains :-)
1569 if (ipv4_is_zeronet(saddr))
1570 goto martian_source;
1572 if (ipv4_is_zeronet(daddr))
1573 goto martian_destination;
1575 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1576 * and call it once if daddr or/and saddr are loopback addresses
1578 if (ipv4_is_loopback(daddr)) {
1579 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1580 goto martian_destination;
1581 } else if (ipv4_is_loopback(saddr)) {
1582 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1583 goto martian_source;
1587 * Now we are ready to route packet.
1590 fl4.flowi4_iif = dev->ifindex;
1591 fl4.flowi4_mark = skb->mark;
1592 fl4.flowi4_tos = tos;
1593 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1596 err = fib_lookup(net, &fl4, &res);
1600 RT_CACHE_STAT_INC(in_slow_tot);
1602 if (res.type == RTN_BROADCAST)
1605 if (res.type == RTN_LOCAL) {
1606 err = fib_validate_source(skb, saddr, daddr, tos,
1608 dev, in_dev, &itag);
1610 goto martian_source_keep_err;
1614 if (!IN_DEV_FORWARD(in_dev))
1616 if (res.type != RTN_UNICAST)
1617 goto martian_destination;
1619 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1623 if (skb->protocol != htons(ETH_P_IP))
1626 if (!ipv4_is_zeronet(saddr)) {
1627 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1630 goto martian_source_keep_err;
1632 flags |= RTCF_BROADCAST;
1633 res.type = RTN_BROADCAST;
1634 RT_CACHE_STAT_INC(in_brd);
1640 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1641 if (rt_cache_valid(rth)) {
1642 skb_dst_set_noref(skb, &rth->dst);
1650 rth = rt_dst_alloc(net->loopback_dev,
1651 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1655 rth->dst.input= ip_local_deliver;
1656 rth->dst.output= ip_rt_bug;
1657 #ifdef CONFIG_IP_ROUTE_CLASSID
1658 rth->dst.tclassid = itag;
1661 rth->rt_genid = rt_genid(net);
1662 rth->rt_flags = flags|RTCF_LOCAL;
1663 rth->rt_type = res.type;
1664 rth->rt_is_input = 1;
1667 rth->rt_gateway = 0;
1668 rth->rt_uses_gateway = 0;
1669 INIT_LIST_HEAD(&rth->rt_uncached);
1670 if (res.type == RTN_UNREACHABLE) {
1671 rth->dst.input= ip_error;
1672 rth->dst.error= -err;
1673 rth->rt_flags &= ~RTCF_LOCAL;
1676 rt_cache_route(&FIB_RES_NH(res), rth);
1677 skb_dst_set(skb, &rth->dst);
1682 RT_CACHE_STAT_INC(in_no_route);
1683 res.type = RTN_UNREACHABLE;
1689 * Do not cache martian addresses: they should be logged (RFC1812)
1691 martian_destination:
1692 RT_CACHE_STAT_INC(in_martian_dst);
1693 #ifdef CONFIG_IP_ROUTE_VERBOSE
1694 if (IN_DEV_LOG_MARTIANS(in_dev))
1695 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1696 &daddr, &saddr, dev->name);
1709 martian_source_keep_err:
1710 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1714 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1715 u8 tos, struct net_device *dev)
1721 /* Multicast recognition logic is moved from route cache to here.
1722 The problem was that too many Ethernet cards have broken/missing
1723 hardware multicast filters :-( As result the host on multicasting
1724 network acquires a lot of useless route cache entries, sort of
1725 SDR messages from all the world. Now we try to get rid of them.
1726 Really, provided software IP multicast filter is organized
1727 reasonably (at least, hashed), it does not result in a slowdown
1728 comparing with route cache reject entries.
1729 Note, that multicast routers are not affected, because
1730 route cache entry is created eventually.
1732 if (ipv4_is_multicast(daddr)) {
1733 struct in_device *in_dev = __in_dev_get_rcu(dev);
1736 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1737 ip_hdr(skb)->protocol);
1739 #ifdef CONFIG_IP_MROUTE
1741 (!ipv4_is_local_multicast(daddr) &&
1742 IN_DEV_MFORWARD(in_dev))
1745 int res = ip_route_input_mc(skb, daddr, saddr,
1754 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1758 EXPORT_SYMBOL(ip_route_input_noref);
1760 /* called with rcu_read_lock() */
1761 static struct rtable *__mkroute_output(const struct fib_result *res,
1762 const struct flowi4 *fl4, int orig_oif,
1763 struct net_device *dev_out,
1766 struct fib_info *fi = res->fi;
1767 struct fib_nh_exception *fnhe;
1768 struct in_device *in_dev;
1769 u16 type = res->type;
1773 in_dev = __in_dev_get_rcu(dev_out);
1775 return ERR_PTR(-EINVAL);
1777 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1778 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1779 return ERR_PTR(-EINVAL);
1781 if (ipv4_is_lbcast(fl4->daddr))
1782 type = RTN_BROADCAST;
1783 else if (ipv4_is_multicast(fl4->daddr))
1784 type = RTN_MULTICAST;
1785 else if (ipv4_is_zeronet(fl4->daddr))
1786 return ERR_PTR(-EINVAL);
1788 if (dev_out->flags & IFF_LOOPBACK)
1789 flags |= RTCF_LOCAL;
1792 if (type == RTN_BROADCAST) {
1793 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1795 } else if (type == RTN_MULTICAST) {
1796 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1797 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1799 flags &= ~RTCF_LOCAL;
1802 /* If multicast route do not exist use
1803 * default one, but do not gateway in this case.
1806 if (fi && res->prefixlen < 4)
1811 do_cache &= fi != NULL;
1813 struct rtable __rcu **prth;
1814 struct fib_nh *nh = &FIB_RES_NH(*res);
1816 fnhe = find_exception(nh, fl4->daddr);
1818 prth = &fnhe->fnhe_rth;
1820 if (unlikely(fl4->flowi4_flags &
1821 FLOWI_FLAG_KNOWN_NH &&
1823 nh->nh_scope == RT_SCOPE_LINK))) {
1827 prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1829 rth = rcu_dereference(*prth);
1830 if (rt_cache_valid(rth)) {
1831 dst_hold(&rth->dst);
1837 rth = rt_dst_alloc(dev_out,
1838 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1839 IN_DEV_CONF_GET(in_dev, NOXFRM),
1842 return ERR_PTR(-ENOBUFS);
1844 rth->dst.output = ip_output;
1846 rth->rt_genid = rt_genid(dev_net(dev_out));
1847 rth->rt_flags = flags;
1848 rth->rt_type = type;
1849 rth->rt_is_input = 0;
1850 rth->rt_iif = orig_oif ? : 0;
1852 rth->rt_gateway = 0;
1853 rth->rt_uses_gateway = 0;
1854 INIT_LIST_HEAD(&rth->rt_uncached);
1856 RT_CACHE_STAT_INC(out_slow_tot);
1858 if (flags & RTCF_LOCAL)
1859 rth->dst.input = ip_local_deliver;
1860 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1861 if (flags & RTCF_LOCAL &&
1862 !(dev_out->flags & IFF_LOOPBACK)) {
1863 rth->dst.output = ip_mc_output;
1864 RT_CACHE_STAT_INC(out_slow_mc);
1866 #ifdef CONFIG_IP_MROUTE
1867 if (type == RTN_MULTICAST) {
1868 if (IN_DEV_MFORWARD(in_dev) &&
1869 !ipv4_is_local_multicast(fl4->daddr)) {
1870 rth->dst.input = ip_mr_input;
1871 rth->dst.output = ip_mc_output;
1877 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1883 * Major route resolver routine.
1886 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1888 struct net_device *dev_out = NULL;
1889 __u8 tos = RT_FL_TOS(fl4);
1890 unsigned int flags = 0;
1891 struct fib_result res;
1899 orig_oif = fl4->flowi4_oif;
1901 fl4->flowi4_iif = LOOPBACK_IFINDEX;
1902 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1903 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1904 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1908 rth = ERR_PTR(-EINVAL);
1909 if (ipv4_is_multicast(fl4->saddr) ||
1910 ipv4_is_lbcast(fl4->saddr) ||
1911 ipv4_is_zeronet(fl4->saddr))
1914 /* I removed check for oif == dev_out->oif here.
1915 It was wrong for two reasons:
1916 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1917 is assigned to multiple interfaces.
1918 2. Moreover, we are allowed to send packets with saddr
1919 of another iface. --ANK
1922 if (fl4->flowi4_oif == 0 &&
1923 (ipv4_is_multicast(fl4->daddr) ||
1924 ipv4_is_lbcast(fl4->daddr))) {
1925 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1926 dev_out = __ip_dev_find(net, fl4->saddr, false);
1927 if (dev_out == NULL)
1930 /* Special hack: user can direct multicasts
1931 and limited broadcast via necessary interface
1932 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1933 This hack is not just for fun, it allows
1934 vic,vat and friends to work.
1935 They bind socket to loopback, set ttl to zero
1936 and expect that it will work.
1937 From the viewpoint of routing cache they are broken,
1938 because we are not allowed to build multicast path
1939 with loopback source addr (look, routing cache
1940 cannot know, that ttl is zero, so that packet
1941 will not leave this host and route is valid).
1942 Luckily, this hack is good workaround.
1945 fl4->flowi4_oif = dev_out->ifindex;
1949 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1950 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1951 if (!__ip_dev_find(net, fl4->saddr, false))
1957 if (fl4->flowi4_oif) {
1958 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1959 rth = ERR_PTR(-ENODEV);
1960 if (dev_out == NULL)
1963 /* RACE: Check return value of inet_select_addr instead. */
1964 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1965 rth = ERR_PTR(-ENETUNREACH);
1968 if (ipv4_is_local_multicast(fl4->daddr) ||
1969 ipv4_is_lbcast(fl4->daddr)) {
1971 fl4->saddr = inet_select_addr(dev_out, 0,
1976 if (ipv4_is_multicast(fl4->daddr))
1977 fl4->saddr = inet_select_addr(dev_out, 0,
1979 else if (!fl4->daddr)
1980 fl4->saddr = inet_select_addr(dev_out, 0,
1986 fl4->daddr = fl4->saddr;
1988 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1989 dev_out = net->loopback_dev;
1990 fl4->flowi4_oif = LOOPBACK_IFINDEX;
1991 res.type = RTN_LOCAL;
1992 flags |= RTCF_LOCAL;
1996 if (fib_lookup(net, fl4, &res)) {
1999 if (fl4->flowi4_oif) {
2000 /* Apparently, routing tables are wrong. Assume,
2001 that the destination is on link.
2004 Because we are allowed to send to iface
2005 even if it has NO routes and NO assigned
2006 addresses. When oif is specified, routing
2007 tables are looked up with only one purpose:
2008 to catch if destination is gatewayed, rather than
2009 direct. Moreover, if MSG_DONTROUTE is set,
2010 we send packet, ignoring both routing tables
2011 and ifaddr state. --ANK
2014 We could make it even if oif is unknown,
2015 likely IPv6, but we do not.
2018 if (fl4->saddr == 0)
2019 fl4->saddr = inet_select_addr(dev_out, 0,
2021 res.type = RTN_UNICAST;
2024 rth = ERR_PTR(-ENETUNREACH);
2028 if (res.type == RTN_LOCAL) {
2030 if (res.fi->fib_prefsrc)
2031 fl4->saddr = res.fi->fib_prefsrc;
2033 fl4->saddr = fl4->daddr;
2035 dev_out = net->loopback_dev;
2036 fl4->flowi4_oif = dev_out->ifindex;
2037 flags |= RTCF_LOCAL;
2041 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2042 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2043 fib_select_multipath(&res);
2046 if (!res.prefixlen &&
2047 res.table->tb_num_default > 1 &&
2048 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2049 fib_select_default(&res);
2052 fl4->saddr = FIB_RES_PREFSRC(net, res);
2054 dev_out = FIB_RES_DEV(res);
2055 fl4->flowi4_oif = dev_out->ifindex;
2059 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2065 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2067 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2072 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2074 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2076 return mtu ? : dst->dev->mtu;
2079 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2080 struct sk_buff *skb, u32 mtu)
2084 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2085 struct sk_buff *skb)
2089 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2095 static struct dst_ops ipv4_dst_blackhole_ops = {
2097 .protocol = cpu_to_be16(ETH_P_IP),
2098 .check = ipv4_blackhole_dst_check,
2099 .mtu = ipv4_blackhole_mtu,
2100 .default_advmss = ipv4_default_advmss,
2101 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2102 .redirect = ipv4_rt_blackhole_redirect,
2103 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2104 .neigh_lookup = ipv4_neigh_lookup,
2107 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2109 struct rtable *ort = (struct rtable *) dst_orig;
2112 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2114 struct dst_entry *new = &rt->dst;
2117 new->input = dst_discard;
2118 new->output = dst_discard;
2120 new->dev = ort->dst.dev;
2124 rt->rt_is_input = ort->rt_is_input;
2125 rt->rt_iif = ort->rt_iif;
2126 rt->rt_pmtu = ort->rt_pmtu;
2128 rt->rt_genid = rt_genid(net);
2129 rt->rt_flags = ort->rt_flags;
2130 rt->rt_type = ort->rt_type;
2131 rt->rt_gateway = ort->rt_gateway;
2132 rt->rt_uses_gateway = ort->rt_uses_gateway;
2134 INIT_LIST_HEAD(&rt->rt_uncached);
2139 dst_release(dst_orig);
2141 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2144 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2147 struct rtable *rt = __ip_route_output_key(net, flp4);
2152 if (flp4->flowi4_proto)
2153 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2154 flowi4_to_flowi(flp4),
2159 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2161 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2162 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2163 u32 seq, int event, int nowait, unsigned int flags)
2165 struct rtable *rt = skb_rtable(skb);
2167 struct nlmsghdr *nlh;
2168 unsigned long expires = 0;
2170 u32 metrics[RTAX_MAX];
2172 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2176 r = nlmsg_data(nlh);
2177 r->rtm_family = AF_INET;
2178 r->rtm_dst_len = 32;
2180 r->rtm_tos = fl4->flowi4_tos;
2181 r->rtm_table = RT_TABLE_MAIN;
2182 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2183 goto nla_put_failure;
2184 r->rtm_type = rt->rt_type;
2185 r->rtm_scope = RT_SCOPE_UNIVERSE;
2186 r->rtm_protocol = RTPROT_UNSPEC;
2187 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2188 if (rt->rt_flags & RTCF_NOTIFY)
2189 r->rtm_flags |= RTM_F_NOTIFY;
2191 if (nla_put_be32(skb, RTA_DST, dst))
2192 goto nla_put_failure;
2194 r->rtm_src_len = 32;
2195 if (nla_put_be32(skb, RTA_SRC, src))
2196 goto nla_put_failure;
2199 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2200 goto nla_put_failure;
2201 #ifdef CONFIG_IP_ROUTE_CLASSID
2202 if (rt->dst.tclassid &&
2203 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2204 goto nla_put_failure;
2206 if (!rt_is_input_route(rt) &&
2207 fl4->saddr != src) {
2208 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2209 goto nla_put_failure;
2211 if (rt->rt_uses_gateway &&
2212 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2213 goto nla_put_failure;
2215 expires = rt->dst.expires;
2217 unsigned long now = jiffies;
2219 if (time_before(now, expires))
2225 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2226 if (rt->rt_pmtu && expires)
2227 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2228 if (rtnetlink_put_metrics(skb, metrics) < 0)
2229 goto nla_put_failure;
2231 if (fl4->flowi4_mark &&
2232 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2233 goto nla_put_failure;
2235 error = rt->dst.error;
2237 if (rt_is_input_route(rt)) {
2238 #ifdef CONFIG_IP_MROUTE
2239 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2240 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2241 int err = ipmr_get_route(net, skb,
2242 fl4->saddr, fl4->daddr,
2248 goto nla_put_failure;
2250 if (err == -EMSGSIZE)
2251 goto nla_put_failure;
2257 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2258 goto nla_put_failure;
2261 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2262 goto nla_put_failure;
2264 return nlmsg_end(skb, nlh);
2267 nlmsg_cancel(skb, nlh);
2271 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2273 struct net *net = sock_net(in_skb->sk);
2275 struct nlattr *tb[RTA_MAX+1];
2276 struct rtable *rt = NULL;
2283 struct sk_buff *skb;
2285 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2289 rtm = nlmsg_data(nlh);
2291 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2297 /* Reserve room for dummy headers, this skb can pass
2298 through good chunk of routing engine.
2300 skb_reset_mac_header(skb);
2301 skb_reset_network_header(skb);
2303 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2304 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2305 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2307 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2308 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2309 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2310 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2312 memset(&fl4, 0, sizeof(fl4));
2315 fl4.flowi4_tos = rtm->rtm_tos;
2316 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2317 fl4.flowi4_mark = mark;
2320 struct net_device *dev;
2322 dev = __dev_get_by_index(net, iif);
2328 skb->protocol = htons(ETH_P_IP);
2332 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2335 rt = skb_rtable(skb);
2336 if (err == 0 && rt->dst.error)
2337 err = -rt->dst.error;
2339 rt = ip_route_output_key(net, &fl4);
2349 skb_dst_set(skb, &rt->dst);
2350 if (rtm->rtm_flags & RTM_F_NOTIFY)
2351 rt->rt_flags |= RTCF_NOTIFY;
2353 err = rt_fill_info(net, dst, src, &fl4, skb,
2354 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2355 RTM_NEWROUTE, 0, 0);
2359 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2368 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2373 void ip_rt_multicast_event(struct in_device *in_dev)
2375 rt_cache_flush(dev_net(in_dev->dev));
2378 #ifdef CONFIG_SYSCTL
2379 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2380 void __user *buffer,
2381 size_t *lenp, loff_t *ppos)
2384 rt_cache_flush((struct net *)__ctl->extra1);
2391 static ctl_table ipv4_route_table[] = {
2393 .procname = "gc_thresh",
2394 .data = &ipv4_dst_ops.gc_thresh,
2395 .maxlen = sizeof(int),
2397 .proc_handler = proc_dointvec,
2400 .procname = "max_size",
2401 .data = &ip_rt_max_size,
2402 .maxlen = sizeof(int),
2404 .proc_handler = proc_dointvec,
2407 /* Deprecated. Use gc_min_interval_ms */
2409 .procname = "gc_min_interval",
2410 .data = &ip_rt_gc_min_interval,
2411 .maxlen = sizeof(int),
2413 .proc_handler = proc_dointvec_jiffies,
2416 .procname = "gc_min_interval_ms",
2417 .data = &ip_rt_gc_min_interval,
2418 .maxlen = sizeof(int),
2420 .proc_handler = proc_dointvec_ms_jiffies,
2423 .procname = "gc_timeout",
2424 .data = &ip_rt_gc_timeout,
2425 .maxlen = sizeof(int),
2427 .proc_handler = proc_dointvec_jiffies,
2430 .procname = "gc_interval",
2431 .data = &ip_rt_gc_interval,
2432 .maxlen = sizeof(int),
2434 .proc_handler = proc_dointvec_jiffies,
2437 .procname = "redirect_load",
2438 .data = &ip_rt_redirect_load,
2439 .maxlen = sizeof(int),
2441 .proc_handler = proc_dointvec,
2444 .procname = "redirect_number",
2445 .data = &ip_rt_redirect_number,
2446 .maxlen = sizeof(int),
2448 .proc_handler = proc_dointvec,
2451 .procname = "redirect_silence",
2452 .data = &ip_rt_redirect_silence,
2453 .maxlen = sizeof(int),
2455 .proc_handler = proc_dointvec,
2458 .procname = "error_cost",
2459 .data = &ip_rt_error_cost,
2460 .maxlen = sizeof(int),
2462 .proc_handler = proc_dointvec,
2465 .procname = "error_burst",
2466 .data = &ip_rt_error_burst,
2467 .maxlen = sizeof(int),
2469 .proc_handler = proc_dointvec,
2472 .procname = "gc_elasticity",
2473 .data = &ip_rt_gc_elasticity,
2474 .maxlen = sizeof(int),
2476 .proc_handler = proc_dointvec,
2479 .procname = "mtu_expires",
2480 .data = &ip_rt_mtu_expires,
2481 .maxlen = sizeof(int),
2483 .proc_handler = proc_dointvec_jiffies,
2486 .procname = "min_pmtu",
2487 .data = &ip_rt_min_pmtu,
2488 .maxlen = sizeof(int),
2490 .proc_handler = proc_dointvec,
2493 .procname = "min_adv_mss",
2494 .data = &ip_rt_min_advmss,
2495 .maxlen = sizeof(int),
2497 .proc_handler = proc_dointvec,
2502 static struct ctl_table ipv4_route_flush_table[] = {
2504 .procname = "flush",
2505 .maxlen = sizeof(int),
2507 .proc_handler = ipv4_sysctl_rtcache_flush,
2512 static __net_init int sysctl_route_net_init(struct net *net)
2514 struct ctl_table *tbl;
2516 tbl = ipv4_route_flush_table;
2517 if (!net_eq(net, &init_net)) {
2518 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2522 /* Don't export sysctls to unprivileged users */
2523 if (net->user_ns != &init_user_ns)
2524 tbl[0].procname = NULL;
2526 tbl[0].extra1 = net;
2528 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2529 if (net->ipv4.route_hdr == NULL)
2534 if (tbl != ipv4_route_flush_table)
2540 static __net_exit void sysctl_route_net_exit(struct net *net)
2542 struct ctl_table *tbl;
2544 tbl = net->ipv4.route_hdr->ctl_table_arg;
2545 unregister_net_sysctl_table(net->ipv4.route_hdr);
2546 BUG_ON(tbl == ipv4_route_flush_table);
2550 static __net_initdata struct pernet_operations sysctl_route_ops = {
2551 .init = sysctl_route_net_init,
2552 .exit = sysctl_route_net_exit,
2556 static __net_init int rt_genid_init(struct net *net)
2558 atomic_set(&net->rt_genid, 0);
2559 get_random_bytes(&net->ipv4.dev_addr_genid,
2560 sizeof(net->ipv4.dev_addr_genid));
2564 static __net_initdata struct pernet_operations rt_genid_ops = {
2565 .init = rt_genid_init,
2568 static int __net_init ipv4_inetpeer_init(struct net *net)
2570 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2574 inet_peer_base_init(bp);
2575 net->ipv4.peers = bp;
2579 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2581 struct inet_peer_base *bp = net->ipv4.peers;
2583 net->ipv4.peers = NULL;
2584 inetpeer_invalidate_tree(bp);
2588 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2589 .init = ipv4_inetpeer_init,
2590 .exit = ipv4_inetpeer_exit,
2593 #ifdef CONFIG_IP_ROUTE_CLASSID
2594 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2595 #endif /* CONFIG_IP_ROUTE_CLASSID */
2597 int __init ip_rt_init(void)
2601 #ifdef CONFIG_IP_ROUTE_CLASSID
2602 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2604 panic("IP: failed to allocate ip_rt_acct\n");
2607 ipv4_dst_ops.kmem_cachep =
2608 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2609 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2611 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2613 if (dst_entries_init(&ipv4_dst_ops) < 0)
2614 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2616 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2617 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2619 ipv4_dst_ops.gc_thresh = ~0;
2620 ip_rt_max_size = INT_MAX;
2625 if (ip_rt_proc_init())
2626 pr_err("Unable to create route proc files\n");
2631 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2633 #ifdef CONFIG_SYSCTL
2634 register_pernet_subsys(&sysctl_route_ops);
2636 register_pernet_subsys(&rt_genid_ops);
2637 register_pernet_subsys(&ipv4_inetpeer_ops);
2641 #ifdef CONFIG_SYSCTL
2643 * We really need to sanitize the damn ipv4 init order, then all
2644 * this nonsense will go away.
2646 void __init ip_static_sysctl_init(void)
2648 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);