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_redirect_number __read_mostly = 9;
121 static int ip_rt_redirect_load __read_mostly = HZ / 50;
122 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
123 static int ip_rt_error_cost __read_mostly = HZ;
124 static int ip_rt_error_burst __read_mostly = 5 * HZ;
125 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
126 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
127 static int ip_rt_min_advmss __read_mostly = 256;
130 * Interface to generic destination cache.
133 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
134 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
135 static unsigned int ipv4_mtu(const struct dst_entry *dst);
136 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
137 static void ipv4_link_failure(struct sk_buff *skb);
138 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb, u32 mtu);
140 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
141 struct sk_buff *skb);
142 static void ipv4_dst_destroy(struct dst_entry *dst);
144 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
149 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
155 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
159 static struct dst_ops ipv4_dst_ops = {
161 .protocol = cpu_to_be16(ETH_P_IP),
162 .check = ipv4_dst_check,
163 .default_advmss = ipv4_default_advmss,
165 .cow_metrics = ipv4_cow_metrics,
166 .destroy = ipv4_dst_destroy,
167 .ifdown = ipv4_dst_ifdown,
168 .negative_advice = ipv4_negative_advice,
169 .link_failure = ipv4_link_failure,
170 .update_pmtu = ip_rt_update_pmtu,
171 .redirect = ip_do_redirect,
172 .local_out = __ip_local_out,
173 .neigh_lookup = ipv4_neigh_lookup,
176 #define ECN_OR_COST(class) TC_PRIO_##class
178 const __u8 ip_tos2prio[16] = {
180 ECN_OR_COST(BESTEFFORT),
182 ECN_OR_COST(BESTEFFORT),
188 ECN_OR_COST(INTERACTIVE),
190 ECN_OR_COST(INTERACTIVE),
191 TC_PRIO_INTERACTIVE_BULK,
192 ECN_OR_COST(INTERACTIVE_BULK),
193 TC_PRIO_INTERACTIVE_BULK,
194 ECN_OR_COST(INTERACTIVE_BULK)
196 EXPORT_SYMBOL(ip_tos2prio);
198 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
199 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
201 #ifdef CONFIG_PROC_FS
202 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
206 return SEQ_START_TOKEN;
209 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
215 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
219 static int rt_cache_seq_show(struct seq_file *seq, void *v)
221 if (v == SEQ_START_TOKEN)
222 seq_printf(seq, "%-127s\n",
223 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
224 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
229 static const struct seq_operations rt_cache_seq_ops = {
230 .start = rt_cache_seq_start,
231 .next = rt_cache_seq_next,
232 .stop = rt_cache_seq_stop,
233 .show = rt_cache_seq_show,
236 static int rt_cache_seq_open(struct inode *inode, struct file *file)
238 return seq_open(file, &rt_cache_seq_ops);
241 static const struct file_operations rt_cache_seq_fops = {
242 .owner = THIS_MODULE,
243 .open = rt_cache_seq_open,
246 .release = seq_release,
250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
255 return SEQ_START_TOKEN;
257 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
258 if (!cpu_possible(cpu))
261 return &per_cpu(rt_cache_stat, cpu);
266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
270 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
271 if (!cpu_possible(cpu))
274 return &per_cpu(rt_cache_stat, cpu);
280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
285 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
287 struct rt_cache_stat *st = v;
289 if (v == SEQ_START_TOKEN) {
290 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");
294 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
295 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
296 dst_entries_get_slow(&ipv4_dst_ops),
319 static const struct seq_operations rt_cpu_seq_ops = {
320 .start = rt_cpu_seq_start,
321 .next = rt_cpu_seq_next,
322 .stop = rt_cpu_seq_stop,
323 .show = rt_cpu_seq_show,
327 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
329 return seq_open(file, &rt_cpu_seq_ops);
332 static const struct file_operations rt_cpu_seq_fops = {
333 .owner = THIS_MODULE,
334 .open = rt_cpu_seq_open,
337 .release = seq_release,
340 #ifdef CONFIG_IP_ROUTE_CLASSID
341 static int rt_acct_proc_show(struct seq_file *m, void *v)
343 struct ip_rt_acct *dst, *src;
346 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
350 for_each_possible_cpu(i) {
351 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
352 for (j = 0; j < 256; j++) {
353 dst[j].o_bytes += src[j].o_bytes;
354 dst[j].o_packets += src[j].o_packets;
355 dst[j].i_bytes += src[j].i_bytes;
356 dst[j].i_packets += src[j].i_packets;
360 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
365 static int rt_acct_proc_open(struct inode *inode, struct file *file)
367 return single_open(file, rt_acct_proc_show, NULL);
370 static const struct file_operations rt_acct_proc_fops = {
371 .owner = THIS_MODULE,
372 .open = rt_acct_proc_open,
375 .release = single_release,
379 static int __net_init ip_rt_do_proc_init(struct net *net)
381 struct proc_dir_entry *pde;
383 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
388 pde = proc_create("rt_cache", S_IRUGO,
389 net->proc_net_stat, &rt_cpu_seq_fops);
393 #ifdef CONFIG_IP_ROUTE_CLASSID
394 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
400 #ifdef CONFIG_IP_ROUTE_CLASSID
402 remove_proc_entry("rt_cache", net->proc_net_stat);
405 remove_proc_entry("rt_cache", net->proc_net);
410 static void __net_exit ip_rt_do_proc_exit(struct net *net)
412 remove_proc_entry("rt_cache", net->proc_net_stat);
413 remove_proc_entry("rt_cache", net->proc_net);
414 #ifdef CONFIG_IP_ROUTE_CLASSID
415 remove_proc_entry("rt_acct", net->proc_net);
419 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
420 .init = ip_rt_do_proc_init,
421 .exit = ip_rt_do_proc_exit,
424 static int __init ip_rt_proc_init(void)
426 return register_pernet_subsys(&ip_rt_proc_ops);
430 static inline int ip_rt_proc_init(void)
434 #endif /* CONFIG_PROC_FS */
436 static inline bool rt_is_expired(const struct rtable *rth)
438 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
441 void rt_cache_flush(struct net *net)
446 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
450 struct net_device *dev = dst->dev;
451 const __be32 *pkey = daddr;
452 const struct rtable *rt;
455 rt = (const struct rtable *) dst;
457 pkey = (const __be32 *) &rt->rt_gateway;
459 pkey = &ip_hdr(skb)->daddr;
461 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
464 return neigh_create(&arp_tbl, pkey, dev);
468 * Peer allocation may fail only in serious out-of-memory conditions. However
469 * we still can generate some output.
470 * Random ID selection looks a bit dangerous because we have no chances to
471 * select ID being unique in a reasonable period of time.
472 * But broken packet identifier may be better than no packet at all.
474 static void ip_select_fb_ident(struct iphdr *iph)
476 static DEFINE_SPINLOCK(ip_fb_id_lock);
477 static u32 ip_fallback_id;
480 spin_lock_bh(&ip_fb_id_lock);
481 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
482 iph->id = htons(salt & 0xFFFF);
483 ip_fallback_id = salt;
484 spin_unlock_bh(&ip_fb_id_lock);
487 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
489 struct net *net = dev_net(dst->dev);
490 struct inet_peer *peer;
492 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
494 iph->id = htons(inet_getid(peer, more));
499 ip_select_fb_ident(iph);
501 EXPORT_SYMBOL(__ip_select_ident);
503 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
504 const struct iphdr *iph,
506 u8 prot, u32 mark, int flow_flags)
509 const struct inet_sock *inet = inet_sk(sk);
511 oif = sk->sk_bound_dev_if;
513 tos = RT_CONN_FLAGS(sk);
514 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
516 flowi4_init_output(fl4, oif, mark, tos,
517 RT_SCOPE_UNIVERSE, prot,
519 iph->daddr, iph->saddr, 0, 0);
522 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
523 const struct sock *sk)
525 const struct iphdr *iph = ip_hdr(skb);
526 int oif = skb->dev->ifindex;
527 u8 tos = RT_TOS(iph->tos);
528 u8 prot = iph->protocol;
529 u32 mark = skb->mark;
531 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
534 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
536 const struct inet_sock *inet = inet_sk(sk);
537 const struct ip_options_rcu *inet_opt;
538 __be32 daddr = inet->inet_daddr;
541 inet_opt = rcu_dereference(inet->inet_opt);
542 if (inet_opt && inet_opt->opt.srr)
543 daddr = inet_opt->opt.faddr;
544 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
545 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
546 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
547 inet_sk_flowi_flags(sk),
548 daddr, inet->inet_saddr, 0, 0);
552 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
553 const struct sk_buff *skb)
556 build_skb_flow_key(fl4, skb, sk);
558 build_sk_flow_key(fl4, sk);
561 static inline void rt_free(struct rtable *rt)
563 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
566 static DEFINE_SPINLOCK(fnhe_lock);
568 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
570 struct fib_nh_exception *fnhe, *oldest;
573 oldest = rcu_dereference(hash->chain);
574 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
575 fnhe = rcu_dereference(fnhe->fnhe_next)) {
576 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
579 orig = rcu_dereference(oldest->fnhe_rth);
581 RCU_INIT_POINTER(oldest->fnhe_rth, NULL);
587 static inline u32 fnhe_hashfun(__be32 daddr)
591 hval = (__force u32) daddr;
592 hval ^= (hval >> 11) ^ (hval >> 22);
594 return hval & (FNHE_HASH_SIZE - 1);
597 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
598 u32 pmtu, unsigned long expires)
600 struct fnhe_hash_bucket *hash;
601 struct fib_nh_exception *fnhe;
603 u32 hval = fnhe_hashfun(daddr);
605 spin_lock_bh(&fnhe_lock);
607 hash = nh->nh_exceptions;
609 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
612 nh->nh_exceptions = hash;
618 for (fnhe = rcu_dereference(hash->chain); fnhe;
619 fnhe = rcu_dereference(fnhe->fnhe_next)) {
620 if (fnhe->fnhe_daddr == daddr)
629 fnhe->fnhe_pmtu = pmtu;
630 fnhe->fnhe_expires = expires;
633 if (depth > FNHE_RECLAIM_DEPTH)
634 fnhe = fnhe_oldest(hash);
636 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
640 fnhe->fnhe_next = hash->chain;
641 rcu_assign_pointer(hash->chain, fnhe);
643 fnhe->fnhe_daddr = daddr;
645 fnhe->fnhe_pmtu = pmtu;
646 fnhe->fnhe_expires = expires;
649 fnhe->fnhe_stamp = jiffies;
652 spin_unlock_bh(&fnhe_lock);
656 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
659 __be32 new_gw = icmp_hdr(skb)->un.gateway;
660 __be32 old_gw = ip_hdr(skb)->saddr;
661 struct net_device *dev = skb->dev;
662 struct in_device *in_dev;
663 struct fib_result res;
667 switch (icmp_hdr(skb)->code & 7) {
669 case ICMP_REDIR_NETTOS:
670 case ICMP_REDIR_HOST:
671 case ICMP_REDIR_HOSTTOS:
678 if (rt->rt_gateway != old_gw)
681 in_dev = __in_dev_get_rcu(dev);
686 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
687 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
688 ipv4_is_zeronet(new_gw))
689 goto reject_redirect;
691 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
692 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
693 goto reject_redirect;
694 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
695 goto reject_redirect;
697 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
698 goto reject_redirect;
701 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
703 if (!(n->nud_state & NUD_VALID)) {
704 neigh_event_send(n, NULL);
706 if (fib_lookup(net, fl4, &res) == 0) {
707 struct fib_nh *nh = &FIB_RES_NH(res);
709 update_or_create_fnhe(nh, fl4->daddr, new_gw,
713 rt->dst.obsolete = DST_OBSOLETE_KILL;
714 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
721 #ifdef CONFIG_IP_ROUTE_VERBOSE
722 if (IN_DEV_LOG_MARTIANS(in_dev)) {
723 const struct iphdr *iph = (const struct iphdr *) skb->data;
724 __be32 daddr = iph->daddr;
725 __be32 saddr = iph->saddr;
727 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
728 " Advised path = %pI4 -> %pI4\n",
729 &old_gw, dev->name, &new_gw,
736 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
740 const struct iphdr *iph = (const struct iphdr *) skb->data;
741 int oif = skb->dev->ifindex;
742 u8 tos = RT_TOS(iph->tos);
743 u8 prot = iph->protocol;
744 u32 mark = skb->mark;
746 rt = (struct rtable *) dst;
748 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
749 __ip_do_redirect(rt, skb, &fl4, true);
752 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
754 struct rtable *rt = (struct rtable *)dst;
755 struct dst_entry *ret = dst;
758 if (dst->obsolete > 0) {
761 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
772 * 1. The first ip_rt_redirect_number redirects are sent
773 * with exponential backoff, then we stop sending them at all,
774 * assuming that the host ignores our redirects.
775 * 2. If we did not see packets requiring redirects
776 * during ip_rt_redirect_silence, we assume that the host
777 * forgot redirected route and start to send redirects again.
779 * This algorithm is much cheaper and more intelligent than dumb load limiting
782 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
783 * and "frag. need" (breaks PMTU discovery) in icmp.c.
786 void ip_rt_send_redirect(struct sk_buff *skb)
788 struct rtable *rt = skb_rtable(skb);
789 struct in_device *in_dev;
790 struct inet_peer *peer;
795 in_dev = __in_dev_get_rcu(rt->dst.dev);
796 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
800 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
803 net = dev_net(rt->dst.dev);
804 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
806 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
807 rt_nexthop(rt, ip_hdr(skb)->daddr));
811 /* No redirected packets during ip_rt_redirect_silence;
812 * reset the algorithm.
814 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
815 peer->rate_tokens = 0;
817 /* Too many ignored redirects; do not send anything
818 * set dst.rate_last to the last seen redirected packet.
820 if (peer->rate_tokens >= ip_rt_redirect_number) {
821 peer->rate_last = jiffies;
825 /* Check for load limit; set rate_last to the latest sent
828 if (peer->rate_tokens == 0 ||
831 (ip_rt_redirect_load << peer->rate_tokens)))) {
832 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
834 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
835 peer->rate_last = jiffies;
837 #ifdef CONFIG_IP_ROUTE_VERBOSE
839 peer->rate_tokens == ip_rt_redirect_number)
840 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
841 &ip_hdr(skb)->saddr, inet_iif(skb),
842 &ip_hdr(skb)->daddr, &gw);
849 static int ip_error(struct sk_buff *skb)
851 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
852 struct rtable *rt = skb_rtable(skb);
853 struct inet_peer *peer;
859 net = dev_net(rt->dst.dev);
860 if (!IN_DEV_FORWARD(in_dev)) {
861 switch (rt->dst.error) {
863 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
867 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
873 switch (rt->dst.error) {
878 code = ICMP_HOST_UNREACH;
881 code = ICMP_NET_UNREACH;
882 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
885 code = ICMP_PKT_FILTERED;
889 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
894 peer->rate_tokens += now - peer->rate_last;
895 if (peer->rate_tokens > ip_rt_error_burst)
896 peer->rate_tokens = ip_rt_error_burst;
897 peer->rate_last = now;
898 if (peer->rate_tokens >= ip_rt_error_cost)
899 peer->rate_tokens -= ip_rt_error_cost;
905 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
911 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
913 struct dst_entry *dst = &rt->dst;
914 struct fib_result res;
916 if (dst_metric_locked(dst, RTAX_MTU))
919 if (dst->dev->mtu < mtu)
922 if (mtu < ip_rt_min_pmtu)
923 mtu = ip_rt_min_pmtu;
926 dst->obsolete = DST_OBSOLETE_KILL;
929 dst->expires = max(1UL, jiffies + ip_rt_mtu_expires);
933 if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
934 struct fib_nh *nh = &FIB_RES_NH(res);
936 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
937 jiffies + ip_rt_mtu_expires);
942 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
943 struct sk_buff *skb, u32 mtu)
945 struct rtable *rt = (struct rtable *) dst;
948 ip_rt_build_flow_key(&fl4, sk, skb);
949 __ip_rt_update_pmtu(rt, &fl4, mtu);
952 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
953 int oif, u32 mark, u8 protocol, int flow_flags)
955 const struct iphdr *iph = (const struct iphdr *) skb->data;
960 mark = IP4_REPLY_MARK(net, skb->mark);
962 __build_flow_key(&fl4, NULL, iph, oif,
963 RT_TOS(iph->tos), protocol, mark, flow_flags);
964 rt = __ip_route_output_key(net, &fl4);
966 __ip_rt_update_pmtu(rt, &fl4, mtu);
970 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
972 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
974 const struct iphdr *iph = (const struct iphdr *) skb->data;
978 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
980 if (!fl4.flowi4_mark)
981 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
983 rt = __ip_route_output_key(sock_net(sk), &fl4);
985 __ip_rt_update_pmtu(rt, &fl4, mtu);
990 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
992 const struct iphdr *iph = (const struct iphdr *) skb->data;
995 struct dst_entry *dst;
999 rt = (struct rtable *) __sk_dst_get(sk);
1001 if (sock_owned_by_user(sk) || !rt) {
1002 __ipv4_sk_update_pmtu(skb, sk, mtu);
1006 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1008 if (!__sk_dst_check(sk, 0)) {
1009 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1016 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1018 dst = dst_check(&rt->dst, 0);
1021 dst_release(&rt->dst);
1023 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1031 __sk_dst_set(sk, &rt->dst);
1036 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1038 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1039 int oif, u32 mark, u8 protocol, int flow_flags)
1041 const struct iphdr *iph = (const struct iphdr *) skb->data;
1045 __build_flow_key(&fl4, NULL, iph, oif,
1046 RT_TOS(iph->tos), protocol, mark, flow_flags);
1047 rt = __ip_route_output_key(net, &fl4);
1049 __ip_do_redirect(rt, skb, &fl4, false);
1053 EXPORT_SYMBOL_GPL(ipv4_redirect);
1055 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1057 const struct iphdr *iph = (const struct iphdr *) skb->data;
1061 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1062 rt = __ip_route_output_key(sock_net(sk), &fl4);
1064 __ip_do_redirect(rt, skb, &fl4, false);
1068 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1070 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1072 struct rtable *rt = (struct rtable *) dst;
1074 /* All IPV4 dsts are created with ->obsolete set to the value
1075 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1076 * into this function always.
1078 * When a PMTU/redirect information update invalidates a
1079 * route, this is indicated by setting obsolete to
1080 * DST_OBSOLETE_KILL.
1082 if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1087 static void ipv4_link_failure(struct sk_buff *skb)
1091 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1093 rt = skb_rtable(skb);
1095 dst_set_expires(&rt->dst, 0);
1098 static int ip_rt_bug(struct sk_buff *skb)
1100 pr_debug("%s: %pI4 -> %pI4, %s\n",
1101 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1102 skb->dev ? skb->dev->name : "?");
1109 We do not cache source address of outgoing interface,
1110 because it is used only by IP RR, TS and SRR options,
1111 so that it out of fast path.
1113 BTW remember: "addr" is allowed to be not aligned
1117 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1121 if (rt_is_output_route(rt))
1122 src = ip_hdr(skb)->saddr;
1124 struct fib_result res;
1130 memset(&fl4, 0, sizeof(fl4));
1131 fl4.daddr = iph->daddr;
1132 fl4.saddr = iph->saddr;
1133 fl4.flowi4_tos = RT_TOS(iph->tos);
1134 fl4.flowi4_oif = rt->dst.dev->ifindex;
1135 fl4.flowi4_iif = skb->dev->ifindex;
1136 fl4.flowi4_mark = skb->mark;
1139 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1140 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1142 src = inet_select_addr(rt->dst.dev,
1143 rt_nexthop(rt, iph->daddr),
1147 memcpy(addr, &src, 4);
1150 #ifdef CONFIG_IP_ROUTE_CLASSID
1151 static void set_class_tag(struct rtable *rt, u32 tag)
1153 if (!(rt->dst.tclassid & 0xFFFF))
1154 rt->dst.tclassid |= tag & 0xFFFF;
1155 if (!(rt->dst.tclassid & 0xFFFF0000))
1156 rt->dst.tclassid |= tag & 0xFFFF0000;
1160 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1162 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1165 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1167 if (advmss > 65535 - 40)
1168 advmss = 65535 - 40;
1173 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1175 const struct rtable *rt = (const struct rtable *) dst;
1176 unsigned int mtu = rt->rt_pmtu;
1178 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1179 mtu = dst_metric_raw(dst, RTAX_MTU);
1184 mtu = dst->dev->mtu;
1186 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1187 if (rt->rt_uses_gateway && mtu > 576)
1191 if (mtu > IP_MAX_MTU)
1197 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1199 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1200 struct fib_nh_exception *fnhe;
1206 hval = fnhe_hashfun(daddr);
1208 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1209 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1210 if (fnhe->fnhe_daddr == daddr)
1216 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1221 spin_lock_bh(&fnhe_lock);
1223 if (daddr == fnhe->fnhe_daddr) {
1224 struct rtable *orig = rcu_dereference(fnhe->fnhe_rth);
1225 if (orig && rt_is_expired(orig)) {
1227 fnhe->fnhe_pmtu = 0;
1228 fnhe->fnhe_expires = 0;
1230 if (fnhe->fnhe_pmtu) {
1231 unsigned long expires = fnhe->fnhe_expires;
1232 unsigned long diff = expires - jiffies;
1234 if (time_before(jiffies, expires)) {
1235 rt->rt_pmtu = fnhe->fnhe_pmtu;
1236 dst_set_expires(&rt->dst, diff);
1239 if (fnhe->fnhe_gw) {
1240 rt->rt_flags |= RTCF_REDIRECTED;
1241 rt->rt_gateway = fnhe->fnhe_gw;
1242 rt->rt_uses_gateway = 1;
1243 } else if (!rt->rt_gateway)
1244 rt->rt_gateway = daddr;
1246 rcu_assign_pointer(fnhe->fnhe_rth, rt);
1250 fnhe->fnhe_stamp = jiffies;
1253 spin_unlock_bh(&fnhe_lock);
1258 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1260 struct rtable *orig, *prev, **p;
1263 if (rt_is_input_route(rt)) {
1264 p = (struct rtable **)&nh->nh_rth_input;
1266 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1270 prev = cmpxchg(p, orig, rt);
1280 static DEFINE_SPINLOCK(rt_uncached_lock);
1281 static LIST_HEAD(rt_uncached_list);
1283 static void rt_add_uncached_list(struct rtable *rt)
1285 spin_lock_bh(&rt_uncached_lock);
1286 list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1287 spin_unlock_bh(&rt_uncached_lock);
1290 static void ipv4_dst_destroy(struct dst_entry *dst)
1292 struct rtable *rt = (struct rtable *) dst;
1294 if (!list_empty(&rt->rt_uncached)) {
1295 spin_lock_bh(&rt_uncached_lock);
1296 list_del(&rt->rt_uncached);
1297 spin_unlock_bh(&rt_uncached_lock);
1301 void rt_flush_dev(struct net_device *dev)
1303 if (!list_empty(&rt_uncached_list)) {
1304 struct net *net = dev_net(dev);
1307 spin_lock_bh(&rt_uncached_lock);
1308 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1309 if (rt->dst.dev != dev)
1311 rt->dst.dev = net->loopback_dev;
1312 dev_hold(rt->dst.dev);
1315 spin_unlock_bh(&rt_uncached_lock);
1319 static bool rt_cache_valid(const struct rtable *rt)
1322 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1326 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1327 const struct fib_result *res,
1328 struct fib_nh_exception *fnhe,
1329 struct fib_info *fi, u16 type, u32 itag)
1331 bool cached = false;
1334 struct fib_nh *nh = &FIB_RES_NH(*res);
1336 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1337 rt->rt_gateway = nh->nh_gw;
1338 rt->rt_uses_gateway = 1;
1340 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1341 #ifdef CONFIG_IP_ROUTE_CLASSID
1342 rt->dst.tclassid = nh->nh_tclassid;
1345 cached = rt_bind_exception(rt, fnhe, daddr);
1346 else if (!(rt->dst.flags & DST_NOCACHE))
1347 cached = rt_cache_route(nh, rt);
1348 if (unlikely(!cached)) {
1349 /* Routes we intend to cache in nexthop exception or
1350 * FIB nexthop have the DST_NOCACHE bit clear.
1351 * However, if we are unsuccessful at storing this
1352 * route into the cache we really need to set it.
1354 rt->dst.flags |= DST_NOCACHE;
1355 if (!rt->rt_gateway)
1356 rt->rt_gateway = daddr;
1357 rt_add_uncached_list(rt);
1360 rt_add_uncached_list(rt);
1362 #ifdef CONFIG_IP_ROUTE_CLASSID
1363 #ifdef CONFIG_IP_MULTIPLE_TABLES
1364 set_class_tag(rt, res->tclassid);
1366 set_class_tag(rt, itag);
1370 static struct rtable *rt_dst_alloc(struct net_device *dev,
1371 bool nopolicy, bool noxfrm, bool will_cache)
1373 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1374 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1375 (nopolicy ? DST_NOPOLICY : 0) |
1376 (noxfrm ? DST_NOXFRM : 0));
1379 /* called in rcu_read_lock() section */
1380 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1381 u8 tos, struct net_device *dev, int our)
1384 struct in_device *in_dev = __in_dev_get_rcu(dev);
1388 /* Primary sanity checks. */
1393 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1394 skb->protocol != htons(ETH_P_IP))
1397 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1398 if (ipv4_is_loopback(saddr))
1401 if (ipv4_is_zeronet(saddr)) {
1402 if (!ipv4_is_local_multicast(daddr))
1405 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1410 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1411 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1415 #ifdef CONFIG_IP_ROUTE_CLASSID
1416 rth->dst.tclassid = itag;
1418 rth->dst.output = ip_rt_bug;
1420 rth->rt_genid = rt_genid(dev_net(dev));
1421 rth->rt_flags = RTCF_MULTICAST;
1422 rth->rt_type = RTN_MULTICAST;
1423 rth->rt_is_input= 1;
1426 rth->rt_gateway = 0;
1427 rth->rt_uses_gateway = 0;
1428 INIT_LIST_HEAD(&rth->rt_uncached);
1430 rth->dst.input= ip_local_deliver;
1431 rth->rt_flags |= RTCF_LOCAL;
1434 #ifdef CONFIG_IP_MROUTE
1435 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1436 rth->dst.input = ip_mr_input;
1438 RT_CACHE_STAT_INC(in_slow_mc);
1440 skb_dst_set(skb, &rth->dst);
1452 static void ip_handle_martian_source(struct net_device *dev,
1453 struct in_device *in_dev,
1454 struct sk_buff *skb,
1458 RT_CACHE_STAT_INC(in_martian_src);
1459 #ifdef CONFIG_IP_ROUTE_VERBOSE
1460 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1462 * RFC1812 recommendation, if source is martian,
1463 * the only hint is MAC header.
1465 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1466 &daddr, &saddr, dev->name);
1467 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1468 print_hex_dump(KERN_WARNING, "ll header: ",
1469 DUMP_PREFIX_OFFSET, 16, 1,
1470 skb_mac_header(skb),
1471 dev->hard_header_len, true);
1477 /* called in rcu_read_lock() section */
1478 static int __mkroute_input(struct sk_buff *skb,
1479 const struct fib_result *res,
1480 struct in_device *in_dev,
1481 __be32 daddr, __be32 saddr, u32 tos)
1485 struct in_device *out_dev;
1486 unsigned int flags = 0;
1490 /* get a working reference to the output device */
1491 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1492 if (out_dev == NULL) {
1493 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1497 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1498 in_dev->dev, in_dev, &itag);
1500 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1506 do_cache = res->fi && !itag;
1507 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1508 (IN_DEV_SHARED_MEDIA(out_dev) ||
1509 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1510 flags |= RTCF_DOREDIRECT;
1514 if (skb->protocol != htons(ETH_P_IP)) {
1515 /* Not IP (i.e. ARP). Do not create route, if it is
1516 * invalid for proxy arp. DNAT routes are always valid.
1518 * Proxy arp feature have been extended to allow, ARP
1519 * replies back to the same interface, to support
1520 * Private VLAN switch technologies. See arp.c.
1522 if (out_dev == in_dev &&
1523 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1530 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1531 if (rt_cache_valid(rth)) {
1532 skb_dst_set_noref(skb, &rth->dst);
1537 rth = rt_dst_alloc(out_dev->dev,
1538 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1539 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1545 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1546 rth->rt_flags = flags;
1547 rth->rt_type = res->type;
1548 rth->rt_is_input = 1;
1551 rth->rt_gateway = 0;
1552 rth->rt_uses_gateway = 0;
1553 INIT_LIST_HEAD(&rth->rt_uncached);
1554 RT_CACHE_STAT_INC(in_slow_tot);
1556 rth->dst.input = ip_forward;
1557 rth->dst.output = ip_output;
1559 rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1560 skb_dst_set(skb, &rth->dst);
1567 static int ip_mkroute_input(struct sk_buff *skb,
1568 struct fib_result *res,
1569 const struct flowi4 *fl4,
1570 struct in_device *in_dev,
1571 __be32 daddr, __be32 saddr, u32 tos)
1573 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1574 if (res->fi && res->fi->fib_nhs > 1)
1575 fib_select_multipath(res);
1578 /* create a routing cache entry */
1579 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1583 * NOTE. We drop all the packets that has local source
1584 * addresses, because every properly looped back packet
1585 * must have correct destination already attached by output routine.
1587 * Such approach solves two big problems:
1588 * 1. Not simplex devices are handled properly.
1589 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1590 * called with rcu_read_lock()
1593 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1594 u8 tos, struct net_device *dev)
1596 struct fib_result res;
1597 struct in_device *in_dev = __in_dev_get_rcu(dev);
1599 unsigned int flags = 0;
1603 struct net *net = dev_net(dev);
1606 /* IP on this device is disabled. */
1611 /* Check for the most weird martians, which can be not detected
1615 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1616 goto martian_source;
1619 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1622 /* Accept zero addresses only to limited broadcast;
1623 * I even do not know to fix it or not. Waiting for complains :-)
1625 if (ipv4_is_zeronet(saddr))
1626 goto martian_source;
1628 if (ipv4_is_zeronet(daddr))
1629 goto martian_destination;
1631 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1632 * and call it once if daddr or/and saddr are loopback addresses
1634 if (ipv4_is_loopback(daddr)) {
1635 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1636 goto martian_destination;
1637 } else if (ipv4_is_loopback(saddr)) {
1638 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1639 goto martian_source;
1643 * Now we are ready to route packet.
1646 fl4.flowi4_iif = dev->ifindex;
1647 fl4.flowi4_mark = skb->mark;
1648 fl4.flowi4_tos = tos;
1649 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1652 err = fib_lookup(net, &fl4, &res);
1656 if (res.type == RTN_BROADCAST)
1659 if (res.type == RTN_LOCAL) {
1660 err = fib_validate_source(skb, saddr, daddr, tos,
1662 dev, in_dev, &itag);
1664 goto martian_source_keep_err;
1668 if (!IN_DEV_FORWARD(in_dev))
1670 if (res.type != RTN_UNICAST)
1671 goto martian_destination;
1673 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1677 if (skb->protocol != htons(ETH_P_IP))
1680 if (!ipv4_is_zeronet(saddr)) {
1681 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1684 goto martian_source_keep_err;
1686 flags |= RTCF_BROADCAST;
1687 res.type = RTN_BROADCAST;
1688 RT_CACHE_STAT_INC(in_brd);
1694 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1695 if (rt_cache_valid(rth)) {
1696 skb_dst_set_noref(skb, &rth->dst);
1704 rth = rt_dst_alloc(net->loopback_dev,
1705 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1709 rth->dst.input= ip_local_deliver;
1710 rth->dst.output= ip_rt_bug;
1711 #ifdef CONFIG_IP_ROUTE_CLASSID
1712 rth->dst.tclassid = itag;
1715 rth->rt_genid = rt_genid(net);
1716 rth->rt_flags = flags|RTCF_LOCAL;
1717 rth->rt_type = res.type;
1718 rth->rt_is_input = 1;
1721 rth->rt_gateway = 0;
1722 rth->rt_uses_gateway = 0;
1723 INIT_LIST_HEAD(&rth->rt_uncached);
1724 RT_CACHE_STAT_INC(in_slow_tot);
1725 if (res.type == RTN_UNREACHABLE) {
1726 rth->dst.input= ip_error;
1727 rth->dst.error= -err;
1728 rth->rt_flags &= ~RTCF_LOCAL;
1731 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1732 rth->dst.flags |= DST_NOCACHE;
1733 rt_add_uncached_list(rth);
1736 skb_dst_set(skb, &rth->dst);
1741 RT_CACHE_STAT_INC(in_no_route);
1742 res.type = RTN_UNREACHABLE;
1748 * Do not cache martian addresses: they should be logged (RFC1812)
1750 martian_destination:
1751 RT_CACHE_STAT_INC(in_martian_dst);
1752 #ifdef CONFIG_IP_ROUTE_VERBOSE
1753 if (IN_DEV_LOG_MARTIANS(in_dev))
1754 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1755 &daddr, &saddr, dev->name);
1768 martian_source_keep_err:
1769 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1773 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1774 u8 tos, struct net_device *dev)
1780 /* Multicast recognition logic is moved from route cache to here.
1781 The problem was that too many Ethernet cards have broken/missing
1782 hardware multicast filters :-( As result the host on multicasting
1783 network acquires a lot of useless route cache entries, sort of
1784 SDR messages from all the world. Now we try to get rid of them.
1785 Really, provided software IP multicast filter is organized
1786 reasonably (at least, hashed), it does not result in a slowdown
1787 comparing with route cache reject entries.
1788 Note, that multicast routers are not affected, because
1789 route cache entry is created eventually.
1791 if (ipv4_is_multicast(daddr)) {
1792 struct in_device *in_dev = __in_dev_get_rcu(dev);
1795 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1796 ip_hdr(skb)->protocol);
1798 #ifdef CONFIG_IP_MROUTE
1800 (!ipv4_is_local_multicast(daddr) &&
1801 IN_DEV_MFORWARD(in_dev))
1804 int res = ip_route_input_mc(skb, daddr, saddr,
1813 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1817 EXPORT_SYMBOL(ip_route_input_noref);
1819 /* called with rcu_read_lock() */
1820 static struct rtable *__mkroute_output(const struct fib_result *res,
1821 const struct flowi4 *fl4, int orig_oif,
1822 struct net_device *dev_out,
1825 struct fib_info *fi = res->fi;
1826 struct fib_nh_exception *fnhe;
1827 struct in_device *in_dev;
1828 u16 type = res->type;
1832 in_dev = __in_dev_get_rcu(dev_out);
1834 return ERR_PTR(-EINVAL);
1836 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1837 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1838 return ERR_PTR(-EINVAL);
1840 if (ipv4_is_lbcast(fl4->daddr))
1841 type = RTN_BROADCAST;
1842 else if (ipv4_is_multicast(fl4->daddr))
1843 type = RTN_MULTICAST;
1844 else if (ipv4_is_zeronet(fl4->daddr))
1845 return ERR_PTR(-EINVAL);
1847 if (dev_out->flags & IFF_LOOPBACK)
1848 flags |= RTCF_LOCAL;
1851 if (type == RTN_BROADCAST) {
1852 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1854 } else if (type == RTN_MULTICAST) {
1855 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1856 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1858 flags &= ~RTCF_LOCAL;
1861 /* If multicast route do not exist use
1862 * default one, but do not gateway in this case.
1865 if (fi && res->prefixlen < 4)
1870 do_cache &= fi != NULL;
1872 struct rtable __rcu **prth;
1873 struct fib_nh *nh = &FIB_RES_NH(*res);
1875 fnhe = find_exception(nh, fl4->daddr);
1877 prth = &fnhe->fnhe_rth;
1879 if (unlikely(fl4->flowi4_flags &
1880 FLOWI_FLAG_KNOWN_NH &&
1882 nh->nh_scope == RT_SCOPE_LINK))) {
1886 prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1888 rth = rcu_dereference(*prth);
1889 if (rt_cache_valid(rth)) {
1890 dst_hold(&rth->dst);
1896 rth = rt_dst_alloc(dev_out,
1897 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1898 IN_DEV_CONF_GET(in_dev, NOXFRM),
1901 return ERR_PTR(-ENOBUFS);
1903 rth->dst.output = ip_output;
1905 rth->rt_genid = rt_genid(dev_net(dev_out));
1906 rth->rt_flags = flags;
1907 rth->rt_type = type;
1908 rth->rt_is_input = 0;
1909 rth->rt_iif = orig_oif ? : 0;
1911 rth->rt_gateway = 0;
1912 rth->rt_uses_gateway = 0;
1913 INIT_LIST_HEAD(&rth->rt_uncached);
1915 RT_CACHE_STAT_INC(out_slow_tot);
1917 if (flags & RTCF_LOCAL)
1918 rth->dst.input = ip_local_deliver;
1919 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1920 if (flags & RTCF_LOCAL &&
1921 !(dev_out->flags & IFF_LOOPBACK)) {
1922 rth->dst.output = ip_mc_output;
1923 RT_CACHE_STAT_INC(out_slow_mc);
1925 #ifdef CONFIG_IP_MROUTE
1926 if (type == RTN_MULTICAST) {
1927 if (IN_DEV_MFORWARD(in_dev) &&
1928 !ipv4_is_local_multicast(fl4->daddr)) {
1929 rth->dst.input = ip_mr_input;
1930 rth->dst.output = ip_mc_output;
1936 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1942 * Major route resolver routine.
1945 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1947 struct net_device *dev_out = NULL;
1948 __u8 tos = RT_FL_TOS(fl4);
1949 unsigned int flags = 0;
1950 struct fib_result res;
1958 orig_oif = fl4->flowi4_oif;
1960 fl4->flowi4_iif = LOOPBACK_IFINDEX;
1961 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1962 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1963 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1967 rth = ERR_PTR(-EINVAL);
1968 if (ipv4_is_multicast(fl4->saddr) ||
1969 ipv4_is_lbcast(fl4->saddr) ||
1970 ipv4_is_zeronet(fl4->saddr))
1973 /* I removed check for oif == dev_out->oif here.
1974 It was wrong for two reasons:
1975 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1976 is assigned to multiple interfaces.
1977 2. Moreover, we are allowed to send packets with saddr
1978 of another iface. --ANK
1981 if (fl4->flowi4_oif == 0 &&
1982 (ipv4_is_multicast(fl4->daddr) ||
1983 ipv4_is_lbcast(fl4->daddr))) {
1984 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1985 dev_out = __ip_dev_find(net, fl4->saddr, false);
1986 if (dev_out == NULL)
1989 /* Special hack: user can direct multicasts
1990 and limited broadcast via necessary interface
1991 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1992 This hack is not just for fun, it allows
1993 vic,vat and friends to work.
1994 They bind socket to loopback, set ttl to zero
1995 and expect that it will work.
1996 From the viewpoint of routing cache they are broken,
1997 because we are not allowed to build multicast path
1998 with loopback source addr (look, routing cache
1999 cannot know, that ttl is zero, so that packet
2000 will not leave this host and route is valid).
2001 Luckily, this hack is good workaround.
2004 fl4->flowi4_oif = dev_out->ifindex;
2008 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2009 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2010 if (!__ip_dev_find(net, fl4->saddr, false))
2016 if (fl4->flowi4_oif) {
2017 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2018 rth = ERR_PTR(-ENODEV);
2019 if (dev_out == NULL)
2022 /* RACE: Check return value of inet_select_addr instead. */
2023 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2024 rth = ERR_PTR(-ENETUNREACH);
2027 if (ipv4_is_local_multicast(fl4->daddr) ||
2028 ipv4_is_lbcast(fl4->daddr)) {
2030 fl4->saddr = inet_select_addr(dev_out, 0,
2035 if (ipv4_is_multicast(fl4->daddr))
2036 fl4->saddr = inet_select_addr(dev_out, 0,
2038 else if (!fl4->daddr)
2039 fl4->saddr = inet_select_addr(dev_out, 0,
2045 fl4->daddr = fl4->saddr;
2047 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2048 dev_out = net->loopback_dev;
2049 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2050 res.type = RTN_LOCAL;
2051 flags |= RTCF_LOCAL;
2055 if (fib_lookup(net, fl4, &res)) {
2058 if (fl4->flowi4_oif) {
2059 /* Apparently, routing tables are wrong. Assume,
2060 that the destination is on link.
2063 Because we are allowed to send to iface
2064 even if it has NO routes and NO assigned
2065 addresses. When oif is specified, routing
2066 tables are looked up with only one purpose:
2067 to catch if destination is gatewayed, rather than
2068 direct. Moreover, if MSG_DONTROUTE is set,
2069 we send packet, ignoring both routing tables
2070 and ifaddr state. --ANK
2073 We could make it even if oif is unknown,
2074 likely IPv6, but we do not.
2077 if (fl4->saddr == 0)
2078 fl4->saddr = inet_select_addr(dev_out, 0,
2080 res.type = RTN_UNICAST;
2083 rth = ERR_PTR(-ENETUNREACH);
2087 if (res.type == RTN_LOCAL) {
2089 if (res.fi->fib_prefsrc)
2090 fl4->saddr = res.fi->fib_prefsrc;
2092 fl4->saddr = fl4->daddr;
2094 dev_out = net->loopback_dev;
2095 fl4->flowi4_oif = dev_out->ifindex;
2096 flags |= RTCF_LOCAL;
2100 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2101 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2102 fib_select_multipath(&res);
2105 if (!res.prefixlen &&
2106 res.table->tb_num_default > 1 &&
2107 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2108 fib_select_default(&res);
2111 fl4->saddr = FIB_RES_PREFSRC(net, res);
2113 dev_out = FIB_RES_DEV(res);
2114 fl4->flowi4_oif = dev_out->ifindex;
2118 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2124 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2126 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2131 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2133 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2135 return mtu ? : dst->dev->mtu;
2138 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2139 struct sk_buff *skb, u32 mtu)
2143 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2144 struct sk_buff *skb)
2148 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2154 static struct dst_ops ipv4_dst_blackhole_ops = {
2156 .protocol = cpu_to_be16(ETH_P_IP),
2157 .check = ipv4_blackhole_dst_check,
2158 .mtu = ipv4_blackhole_mtu,
2159 .default_advmss = ipv4_default_advmss,
2160 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2161 .redirect = ipv4_rt_blackhole_redirect,
2162 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2163 .neigh_lookup = ipv4_neigh_lookup,
2166 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2168 struct rtable *ort = (struct rtable *) dst_orig;
2171 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2173 struct dst_entry *new = &rt->dst;
2176 new->input = dst_discard;
2177 new->output = dst_discard;
2179 new->dev = ort->dst.dev;
2183 rt->rt_is_input = ort->rt_is_input;
2184 rt->rt_iif = ort->rt_iif;
2185 rt->rt_pmtu = ort->rt_pmtu;
2187 rt->rt_genid = rt_genid(net);
2188 rt->rt_flags = ort->rt_flags;
2189 rt->rt_type = ort->rt_type;
2190 rt->rt_gateway = ort->rt_gateway;
2191 rt->rt_uses_gateway = ort->rt_uses_gateway;
2193 INIT_LIST_HEAD(&rt->rt_uncached);
2198 dst_release(dst_orig);
2200 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2203 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2206 struct rtable *rt = __ip_route_output_key(net, flp4);
2211 if (flp4->flowi4_proto)
2212 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2213 flowi4_to_flowi(flp4),
2218 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2220 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2221 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2222 u32 seq, int event, int nowait, unsigned int flags)
2224 struct rtable *rt = skb_rtable(skb);
2226 struct nlmsghdr *nlh;
2227 unsigned long expires = 0;
2229 u32 metrics[RTAX_MAX];
2231 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2235 r = nlmsg_data(nlh);
2236 r->rtm_family = AF_INET;
2237 r->rtm_dst_len = 32;
2239 r->rtm_tos = fl4->flowi4_tos;
2240 r->rtm_table = RT_TABLE_MAIN;
2241 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2242 goto nla_put_failure;
2243 r->rtm_type = rt->rt_type;
2244 r->rtm_scope = RT_SCOPE_UNIVERSE;
2245 r->rtm_protocol = RTPROT_UNSPEC;
2246 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2247 if (rt->rt_flags & RTCF_NOTIFY)
2248 r->rtm_flags |= RTM_F_NOTIFY;
2250 if (nla_put_be32(skb, RTA_DST, dst))
2251 goto nla_put_failure;
2253 r->rtm_src_len = 32;
2254 if (nla_put_be32(skb, RTA_SRC, src))
2255 goto nla_put_failure;
2258 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2259 goto nla_put_failure;
2260 #ifdef CONFIG_IP_ROUTE_CLASSID
2261 if (rt->dst.tclassid &&
2262 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2263 goto nla_put_failure;
2265 if (!rt_is_input_route(rt) &&
2266 fl4->saddr != src) {
2267 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2268 goto nla_put_failure;
2270 if (rt->rt_uses_gateway &&
2271 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2272 goto nla_put_failure;
2274 expires = rt->dst.expires;
2276 unsigned long now = jiffies;
2278 if (time_before(now, expires))
2284 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2285 if (rt->rt_pmtu && expires)
2286 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2287 if (rtnetlink_put_metrics(skb, metrics) < 0)
2288 goto nla_put_failure;
2290 if (fl4->flowi4_mark &&
2291 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2292 goto nla_put_failure;
2294 error = rt->dst.error;
2296 if (rt_is_input_route(rt)) {
2297 #ifdef CONFIG_IP_MROUTE
2298 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2299 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2300 int err = ipmr_get_route(net, skb,
2301 fl4->saddr, fl4->daddr,
2307 goto nla_put_failure;
2309 if (err == -EMSGSIZE)
2310 goto nla_put_failure;
2316 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2317 goto nla_put_failure;
2320 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2321 goto nla_put_failure;
2323 return nlmsg_end(skb, nlh);
2326 nlmsg_cancel(skb, nlh);
2330 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2332 struct net *net = sock_net(in_skb->sk);
2334 struct nlattr *tb[RTA_MAX+1];
2335 struct rtable *rt = NULL;
2342 struct sk_buff *skb;
2344 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2348 rtm = nlmsg_data(nlh);
2350 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2356 /* Reserve room for dummy headers, this skb can pass
2357 through good chunk of routing engine.
2359 skb_reset_mac_header(skb);
2360 skb_reset_network_header(skb);
2362 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2363 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2364 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2366 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2367 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2368 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2369 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2371 memset(&fl4, 0, sizeof(fl4));
2374 fl4.flowi4_tos = rtm->rtm_tos;
2375 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2376 fl4.flowi4_mark = mark;
2379 struct net_device *dev;
2381 dev = __dev_get_by_index(net, iif);
2387 skb->protocol = htons(ETH_P_IP);
2391 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2394 rt = skb_rtable(skb);
2395 if (err == 0 && rt->dst.error)
2396 err = -rt->dst.error;
2398 rt = ip_route_output_key(net, &fl4);
2408 skb_dst_set(skb, &rt->dst);
2409 if (rtm->rtm_flags & RTM_F_NOTIFY)
2410 rt->rt_flags |= RTCF_NOTIFY;
2412 err = rt_fill_info(net, dst, src, &fl4, skb,
2413 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2414 RTM_NEWROUTE, 0, 0);
2418 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2427 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2432 void ip_rt_multicast_event(struct in_device *in_dev)
2434 rt_cache_flush(dev_net(in_dev->dev));
2437 #ifdef CONFIG_SYSCTL
2438 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2439 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2440 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2441 static int ip_rt_gc_elasticity __read_mostly = 8;
2443 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2444 void __user *buffer,
2445 size_t *lenp, loff_t *ppos)
2448 rt_cache_flush((struct net *)__ctl->extra1);
2455 static ctl_table ipv4_route_table[] = {
2457 .procname = "gc_thresh",
2458 .data = &ipv4_dst_ops.gc_thresh,
2459 .maxlen = sizeof(int),
2461 .proc_handler = proc_dointvec,
2464 .procname = "max_size",
2465 .data = &ip_rt_max_size,
2466 .maxlen = sizeof(int),
2468 .proc_handler = proc_dointvec,
2471 /* Deprecated. Use gc_min_interval_ms */
2473 .procname = "gc_min_interval",
2474 .data = &ip_rt_gc_min_interval,
2475 .maxlen = sizeof(int),
2477 .proc_handler = proc_dointvec_jiffies,
2480 .procname = "gc_min_interval_ms",
2481 .data = &ip_rt_gc_min_interval,
2482 .maxlen = sizeof(int),
2484 .proc_handler = proc_dointvec_ms_jiffies,
2487 .procname = "gc_timeout",
2488 .data = &ip_rt_gc_timeout,
2489 .maxlen = sizeof(int),
2491 .proc_handler = proc_dointvec_jiffies,
2494 .procname = "gc_interval",
2495 .data = &ip_rt_gc_interval,
2496 .maxlen = sizeof(int),
2498 .proc_handler = proc_dointvec_jiffies,
2501 .procname = "redirect_load",
2502 .data = &ip_rt_redirect_load,
2503 .maxlen = sizeof(int),
2505 .proc_handler = proc_dointvec,
2508 .procname = "redirect_number",
2509 .data = &ip_rt_redirect_number,
2510 .maxlen = sizeof(int),
2512 .proc_handler = proc_dointvec,
2515 .procname = "redirect_silence",
2516 .data = &ip_rt_redirect_silence,
2517 .maxlen = sizeof(int),
2519 .proc_handler = proc_dointvec,
2522 .procname = "error_cost",
2523 .data = &ip_rt_error_cost,
2524 .maxlen = sizeof(int),
2526 .proc_handler = proc_dointvec,
2529 .procname = "error_burst",
2530 .data = &ip_rt_error_burst,
2531 .maxlen = sizeof(int),
2533 .proc_handler = proc_dointvec,
2536 .procname = "gc_elasticity",
2537 .data = &ip_rt_gc_elasticity,
2538 .maxlen = sizeof(int),
2540 .proc_handler = proc_dointvec,
2543 .procname = "mtu_expires",
2544 .data = &ip_rt_mtu_expires,
2545 .maxlen = sizeof(int),
2547 .proc_handler = proc_dointvec_jiffies,
2550 .procname = "min_pmtu",
2551 .data = &ip_rt_min_pmtu,
2552 .maxlen = sizeof(int),
2554 .proc_handler = proc_dointvec,
2557 .procname = "min_adv_mss",
2558 .data = &ip_rt_min_advmss,
2559 .maxlen = sizeof(int),
2561 .proc_handler = proc_dointvec,
2566 static struct ctl_table ipv4_route_flush_table[] = {
2568 .procname = "flush",
2569 .maxlen = sizeof(int),
2571 .proc_handler = ipv4_sysctl_rtcache_flush,
2576 static __net_init int sysctl_route_net_init(struct net *net)
2578 struct ctl_table *tbl;
2580 tbl = ipv4_route_flush_table;
2581 if (!net_eq(net, &init_net)) {
2582 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2586 /* Don't export sysctls to unprivileged users */
2587 if (net->user_ns != &init_user_ns)
2588 tbl[0].procname = NULL;
2590 tbl[0].extra1 = net;
2592 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2593 if (net->ipv4.route_hdr == NULL)
2598 if (tbl != ipv4_route_flush_table)
2604 static __net_exit void sysctl_route_net_exit(struct net *net)
2606 struct ctl_table *tbl;
2608 tbl = net->ipv4.route_hdr->ctl_table_arg;
2609 unregister_net_sysctl_table(net->ipv4.route_hdr);
2610 BUG_ON(tbl == ipv4_route_flush_table);
2614 static __net_initdata struct pernet_operations sysctl_route_ops = {
2615 .init = sysctl_route_net_init,
2616 .exit = sysctl_route_net_exit,
2620 static __net_init int rt_genid_init(struct net *net)
2622 atomic_set(&net->rt_genid, 0);
2623 get_random_bytes(&net->ipv4.dev_addr_genid,
2624 sizeof(net->ipv4.dev_addr_genid));
2628 static __net_initdata struct pernet_operations rt_genid_ops = {
2629 .init = rt_genid_init,
2632 static int __net_init ipv4_inetpeer_init(struct net *net)
2634 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2638 inet_peer_base_init(bp);
2639 net->ipv4.peers = bp;
2643 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2645 struct inet_peer_base *bp = net->ipv4.peers;
2647 net->ipv4.peers = NULL;
2648 inetpeer_invalidate_tree(bp);
2652 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2653 .init = ipv4_inetpeer_init,
2654 .exit = ipv4_inetpeer_exit,
2657 #ifdef CONFIG_IP_ROUTE_CLASSID
2658 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2659 #endif /* CONFIG_IP_ROUTE_CLASSID */
2661 int __init ip_rt_init(void)
2665 #ifdef CONFIG_IP_ROUTE_CLASSID
2666 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2668 panic("IP: failed to allocate ip_rt_acct\n");
2671 ipv4_dst_ops.kmem_cachep =
2672 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2673 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2675 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2677 if (dst_entries_init(&ipv4_dst_ops) < 0)
2678 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2680 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2681 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2683 ipv4_dst_ops.gc_thresh = ~0;
2684 ip_rt_max_size = INT_MAX;
2689 if (ip_rt_proc_init())
2690 pr_err("Unable to create route proc files\n");
2695 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2697 #ifdef CONFIG_SYSCTL
2698 register_pernet_subsys(&sysctl_route_ops);
2700 register_pernet_subsys(&rt_genid_ops);
2701 register_pernet_subsys(&ipv4_inetpeer_ops);
2705 #ifdef CONFIG_SYSCTL
2707 * We really need to sanitize the damn ipv4 init order, then all
2708 * this nonsense will go away.
2710 void __init ip_static_sysctl_init(void)
2712 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);