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 RT_GC_TIMEOUT (300*HZ)
117 static int ip_rt_max_size;
118 static int ip_rt_redirect_number __read_mostly = 9;
119 static int ip_rt_redirect_load __read_mostly = HZ / 50;
120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
121 static int ip_rt_error_cost __read_mostly = HZ;
122 static int ip_rt_error_burst __read_mostly = 5 * HZ;
123 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
124 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
125 static int ip_rt_min_advmss __read_mostly = 256;
128 * Interface to generic destination cache.
131 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
133 static unsigned int ipv4_mtu(const struct dst_entry *dst);
134 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
135 static void ipv4_link_failure(struct sk_buff *skb);
136 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
137 struct sk_buff *skb, u32 mtu);
138 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb);
140 static void ipv4_dst_destroy(struct dst_entry *dst);
142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
152 static struct dst_ops ipv4_dst_ops = {
154 .protocol = cpu_to_be16(ETH_P_IP),
155 .check = ipv4_dst_check,
156 .default_advmss = ipv4_default_advmss,
158 .cow_metrics = ipv4_cow_metrics,
159 .destroy = ipv4_dst_destroy,
160 .negative_advice = ipv4_negative_advice,
161 .link_failure = ipv4_link_failure,
162 .update_pmtu = ip_rt_update_pmtu,
163 .redirect = ip_do_redirect,
164 .local_out = __ip_local_out,
165 .neigh_lookup = ipv4_neigh_lookup,
168 #define ECN_OR_COST(class) TC_PRIO_##class
170 const __u8 ip_tos2prio[16] = {
172 ECN_OR_COST(BESTEFFORT),
174 ECN_OR_COST(BESTEFFORT),
180 ECN_OR_COST(INTERACTIVE),
182 ECN_OR_COST(INTERACTIVE),
183 TC_PRIO_INTERACTIVE_BULK,
184 ECN_OR_COST(INTERACTIVE_BULK),
185 TC_PRIO_INTERACTIVE_BULK,
186 ECN_OR_COST(INTERACTIVE_BULK)
188 EXPORT_SYMBOL(ip_tos2prio);
190 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
191 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
193 #ifdef CONFIG_PROC_FS
194 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
198 return SEQ_START_TOKEN;
201 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
207 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
211 static int rt_cache_seq_show(struct seq_file *seq, void *v)
213 if (v == SEQ_START_TOKEN)
214 seq_printf(seq, "%-127s\n",
215 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
216 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
221 static const struct seq_operations rt_cache_seq_ops = {
222 .start = rt_cache_seq_start,
223 .next = rt_cache_seq_next,
224 .stop = rt_cache_seq_stop,
225 .show = rt_cache_seq_show,
228 static int rt_cache_seq_open(struct inode *inode, struct file *file)
230 return seq_open(file, &rt_cache_seq_ops);
233 static const struct file_operations rt_cache_seq_fops = {
234 .owner = THIS_MODULE,
235 .open = rt_cache_seq_open,
238 .release = seq_release,
242 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
247 return SEQ_START_TOKEN;
249 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
250 if (!cpu_possible(cpu))
253 return &per_cpu(rt_cache_stat, cpu);
258 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
262 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
263 if (!cpu_possible(cpu))
266 return &per_cpu(rt_cache_stat, cpu);
272 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
277 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
279 struct rt_cache_stat *st = v;
281 if (v == SEQ_START_TOKEN) {
282 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");
286 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
287 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
288 dst_entries_get_slow(&ipv4_dst_ops),
301 0, /* st->gc_total */
302 0, /* st->gc_ignored */
303 0, /* st->gc_goal_miss */
304 0, /* st->gc_dst_overflow */
305 0, /* st->in_hlist_search */
306 0 /* st->out_hlist_search */
311 static const struct seq_operations rt_cpu_seq_ops = {
312 .start = rt_cpu_seq_start,
313 .next = rt_cpu_seq_next,
314 .stop = rt_cpu_seq_stop,
315 .show = rt_cpu_seq_show,
319 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
321 return seq_open(file, &rt_cpu_seq_ops);
324 static const struct file_operations rt_cpu_seq_fops = {
325 .owner = THIS_MODULE,
326 .open = rt_cpu_seq_open,
329 .release = seq_release,
332 #ifdef CONFIG_IP_ROUTE_CLASSID
333 static int rt_acct_proc_show(struct seq_file *m, void *v)
335 struct ip_rt_acct *dst, *src;
338 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
342 for_each_possible_cpu(i) {
343 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
344 for (j = 0; j < 256; j++) {
345 dst[j].o_bytes += src[j].o_bytes;
346 dst[j].o_packets += src[j].o_packets;
347 dst[j].i_bytes += src[j].i_bytes;
348 dst[j].i_packets += src[j].i_packets;
352 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
357 static int rt_acct_proc_open(struct inode *inode, struct file *file)
359 return single_open(file, rt_acct_proc_show, NULL);
362 static const struct file_operations rt_acct_proc_fops = {
363 .owner = THIS_MODULE,
364 .open = rt_acct_proc_open,
367 .release = single_release,
371 static int __net_init ip_rt_do_proc_init(struct net *net)
373 struct proc_dir_entry *pde;
375 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
380 pde = proc_create("rt_cache", S_IRUGO,
381 net->proc_net_stat, &rt_cpu_seq_fops);
385 #ifdef CONFIG_IP_ROUTE_CLASSID
386 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
392 #ifdef CONFIG_IP_ROUTE_CLASSID
394 remove_proc_entry("rt_cache", net->proc_net_stat);
397 remove_proc_entry("rt_cache", net->proc_net);
402 static void __net_exit ip_rt_do_proc_exit(struct net *net)
404 remove_proc_entry("rt_cache", net->proc_net_stat);
405 remove_proc_entry("rt_cache", net->proc_net);
406 #ifdef CONFIG_IP_ROUTE_CLASSID
407 remove_proc_entry("rt_acct", net->proc_net);
411 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
412 .init = ip_rt_do_proc_init,
413 .exit = ip_rt_do_proc_exit,
416 static int __init ip_rt_proc_init(void)
418 return register_pernet_subsys(&ip_rt_proc_ops);
422 static inline int ip_rt_proc_init(void)
426 #endif /* CONFIG_PROC_FS */
428 static inline bool rt_is_expired(const struct rtable *rth)
430 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
433 void rt_cache_flush(struct net *net)
435 rt_genid_bump_ipv4(net);
438 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
442 struct net_device *dev = dst->dev;
443 const __be32 *pkey = daddr;
444 const struct rtable *rt;
447 rt = (const struct rtable *) dst;
449 pkey = (const __be32 *) &rt->rt_gateway;
451 pkey = &ip_hdr(skb)->daddr;
453 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
456 return neigh_create(&arp_tbl, pkey, dev);
460 * Peer allocation may fail only in serious out-of-memory conditions. However
461 * we still can generate some output.
462 * Random ID selection looks a bit dangerous because we have no chances to
463 * select ID being unique in a reasonable period of time.
464 * But broken packet identifier may be better than no packet at all.
466 static void ip_select_fb_ident(struct iphdr *iph)
468 static DEFINE_SPINLOCK(ip_fb_id_lock);
469 static u32 ip_fallback_id;
472 spin_lock_bh(&ip_fb_id_lock);
473 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
474 iph->id = htons(salt & 0xFFFF);
475 ip_fallback_id = salt;
476 spin_unlock_bh(&ip_fb_id_lock);
479 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
481 struct net *net = dev_net(dst->dev);
482 struct inet_peer *peer;
484 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
486 iph->id = htons(inet_getid(peer, more));
491 ip_select_fb_ident(iph);
493 EXPORT_SYMBOL(__ip_select_ident);
495 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
496 const struct iphdr *iph,
498 u8 prot, u32 mark, int flow_flags)
501 const struct inet_sock *inet = inet_sk(sk);
503 oif = sk->sk_bound_dev_if;
505 tos = RT_CONN_FLAGS(sk);
506 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
508 flowi4_init_output(fl4, oif, mark, tos,
509 RT_SCOPE_UNIVERSE, prot,
511 iph->daddr, iph->saddr, 0, 0);
514 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
515 const struct sock *sk)
517 const struct iphdr *iph = ip_hdr(skb);
518 int oif = skb->dev->ifindex;
519 u8 tos = RT_TOS(iph->tos);
520 u8 prot = iph->protocol;
521 u32 mark = skb->mark;
523 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
526 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
528 const struct inet_sock *inet = inet_sk(sk);
529 const struct ip_options_rcu *inet_opt;
530 __be32 daddr = inet->inet_daddr;
533 inet_opt = rcu_dereference(inet->inet_opt);
534 if (inet_opt && inet_opt->opt.srr)
535 daddr = inet_opt->opt.faddr;
536 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
537 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
538 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
539 inet_sk_flowi_flags(sk),
540 daddr, inet->inet_saddr, 0, 0);
544 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
545 const struct sk_buff *skb)
548 build_skb_flow_key(fl4, skb, sk);
550 build_sk_flow_key(fl4, sk);
553 static inline void rt_free(struct rtable *rt)
555 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
558 static DEFINE_SPINLOCK(fnhe_lock);
560 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
564 rt = rcu_dereference(fnhe->fnhe_rth_input);
566 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
569 rt = rcu_dereference(fnhe->fnhe_rth_output);
571 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
576 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
578 struct fib_nh_exception *fnhe, *oldest;
580 oldest = rcu_dereference(hash->chain);
581 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
582 fnhe = rcu_dereference(fnhe->fnhe_next)) {
583 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
586 fnhe_flush_routes(oldest);
590 static inline u32 fnhe_hashfun(__be32 daddr)
594 hval = (__force u32) daddr;
595 hval ^= (hval >> 11) ^ (hval >> 22);
597 return hval & (FNHE_HASH_SIZE - 1);
600 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
602 rt->rt_pmtu = fnhe->fnhe_pmtu;
603 rt->dst.expires = fnhe->fnhe_expires;
606 rt->rt_flags |= RTCF_REDIRECTED;
607 rt->rt_gateway = fnhe->fnhe_gw;
608 rt->rt_uses_gateway = 1;
612 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
613 u32 pmtu, unsigned long expires)
615 struct fnhe_hash_bucket *hash;
616 struct fib_nh_exception *fnhe;
620 u32 hval = fnhe_hashfun(daddr);
622 spin_lock_bh(&fnhe_lock);
624 hash = nh->nh_exceptions;
626 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
629 nh->nh_exceptions = hash;
635 for (fnhe = rcu_dereference(hash->chain); fnhe;
636 fnhe = rcu_dereference(fnhe->fnhe_next)) {
637 if (fnhe->fnhe_daddr == daddr)
646 fnhe->fnhe_pmtu = pmtu;
647 fnhe->fnhe_expires = max(1UL, expires);
649 /* Update all cached dsts too */
650 rt = rcu_dereference(fnhe->fnhe_rth_input);
652 fill_route_from_fnhe(rt, fnhe);
653 rt = rcu_dereference(fnhe->fnhe_rth_output);
655 fill_route_from_fnhe(rt, fnhe);
657 if (depth > FNHE_RECLAIM_DEPTH)
658 fnhe = fnhe_oldest(hash);
660 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
664 fnhe->fnhe_next = hash->chain;
665 rcu_assign_pointer(hash->chain, fnhe);
667 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
668 fnhe->fnhe_daddr = daddr;
670 fnhe->fnhe_pmtu = pmtu;
671 fnhe->fnhe_expires = expires;
673 /* Exception created; mark the cached routes for the nexthop
674 * stale, so anyone caching it rechecks if this exception
677 rt = rcu_dereference(nh->nh_rth_input);
679 rt->dst.obsolete = DST_OBSOLETE_KILL;
681 for_each_possible_cpu(i) {
682 struct rtable __rcu **prt;
683 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
684 rt = rcu_dereference(*prt);
686 rt->dst.obsolete = DST_OBSOLETE_KILL;
690 fnhe->fnhe_stamp = jiffies;
693 spin_unlock_bh(&fnhe_lock);
696 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
699 __be32 new_gw = icmp_hdr(skb)->un.gateway;
700 __be32 old_gw = ip_hdr(skb)->saddr;
701 struct net_device *dev = skb->dev;
702 struct in_device *in_dev;
703 struct fib_result res;
707 switch (icmp_hdr(skb)->code & 7) {
709 case ICMP_REDIR_NETTOS:
710 case ICMP_REDIR_HOST:
711 case ICMP_REDIR_HOSTTOS:
718 if (rt->rt_gateway != old_gw)
721 in_dev = __in_dev_get_rcu(dev);
726 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
727 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
728 ipv4_is_zeronet(new_gw))
729 goto reject_redirect;
731 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
732 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
733 goto reject_redirect;
734 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
735 goto reject_redirect;
737 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
738 goto reject_redirect;
741 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
743 if (!(n->nud_state & NUD_VALID)) {
744 neigh_event_send(n, NULL);
746 if (fib_lookup(net, fl4, &res) == 0) {
747 struct fib_nh *nh = &FIB_RES_NH(res);
749 update_or_create_fnhe(nh, fl4->daddr, new_gw,
753 rt->dst.obsolete = DST_OBSOLETE_KILL;
754 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
761 #ifdef CONFIG_IP_ROUTE_VERBOSE
762 if (IN_DEV_LOG_MARTIANS(in_dev)) {
763 const struct iphdr *iph = (const struct iphdr *) skb->data;
764 __be32 daddr = iph->daddr;
765 __be32 saddr = iph->saddr;
767 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
768 " Advised path = %pI4 -> %pI4\n",
769 &old_gw, dev->name, &new_gw,
776 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
780 const struct iphdr *iph = (const struct iphdr *) skb->data;
781 int oif = skb->dev->ifindex;
782 u8 tos = RT_TOS(iph->tos);
783 u8 prot = iph->protocol;
784 u32 mark = skb->mark;
786 rt = (struct rtable *) dst;
788 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
789 __ip_do_redirect(rt, skb, &fl4, true);
792 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
794 struct rtable *rt = (struct rtable *)dst;
795 struct dst_entry *ret = dst;
798 if (dst->obsolete > 0) {
801 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
812 * 1. The first ip_rt_redirect_number redirects are sent
813 * with exponential backoff, then we stop sending them at all,
814 * assuming that the host ignores our redirects.
815 * 2. If we did not see packets requiring redirects
816 * during ip_rt_redirect_silence, we assume that the host
817 * forgot redirected route and start to send redirects again.
819 * This algorithm is much cheaper and more intelligent than dumb load limiting
822 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
823 * and "frag. need" (breaks PMTU discovery) in icmp.c.
826 void ip_rt_send_redirect(struct sk_buff *skb)
828 struct rtable *rt = skb_rtable(skb);
829 struct in_device *in_dev;
830 struct inet_peer *peer;
835 in_dev = __in_dev_get_rcu(rt->dst.dev);
836 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
840 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
843 net = dev_net(rt->dst.dev);
844 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
846 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
847 rt_nexthop(rt, ip_hdr(skb)->daddr));
851 /* No redirected packets during ip_rt_redirect_silence;
852 * reset the algorithm.
854 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
855 peer->rate_tokens = 0;
857 /* Too many ignored redirects; do not send anything
858 * set dst.rate_last to the last seen redirected packet.
860 if (peer->rate_tokens >= ip_rt_redirect_number) {
861 peer->rate_last = jiffies;
865 /* Check for load limit; set rate_last to the latest sent
868 if (peer->rate_tokens == 0 ||
871 (ip_rt_redirect_load << peer->rate_tokens)))) {
872 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
874 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
875 peer->rate_last = jiffies;
877 #ifdef CONFIG_IP_ROUTE_VERBOSE
879 peer->rate_tokens == ip_rt_redirect_number)
880 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
881 &ip_hdr(skb)->saddr, inet_iif(skb),
882 &ip_hdr(skb)->daddr, &gw);
889 static int ip_error(struct sk_buff *skb)
891 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
892 struct rtable *rt = skb_rtable(skb);
893 struct inet_peer *peer;
899 net = dev_net(rt->dst.dev);
900 if (!IN_DEV_FORWARD(in_dev)) {
901 switch (rt->dst.error) {
903 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
907 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
913 switch (rt->dst.error) {
918 code = ICMP_HOST_UNREACH;
921 code = ICMP_NET_UNREACH;
922 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
925 code = ICMP_PKT_FILTERED;
929 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
934 peer->rate_tokens += now - peer->rate_last;
935 if (peer->rate_tokens > ip_rt_error_burst)
936 peer->rate_tokens = ip_rt_error_burst;
937 peer->rate_last = now;
938 if (peer->rate_tokens >= ip_rt_error_cost)
939 peer->rate_tokens -= ip_rt_error_cost;
945 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
951 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
953 struct dst_entry *dst = &rt->dst;
954 struct fib_result res;
956 if (dst_metric_locked(dst, RTAX_MTU))
959 if (dst->dev->mtu < mtu)
962 if (mtu < ip_rt_min_pmtu)
963 mtu = ip_rt_min_pmtu;
965 if (rt->rt_pmtu == mtu &&
966 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
970 if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
971 struct fib_nh *nh = &FIB_RES_NH(res);
973 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
974 jiffies + ip_rt_mtu_expires);
979 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
980 struct sk_buff *skb, u32 mtu)
982 struct rtable *rt = (struct rtable *) dst;
985 ip_rt_build_flow_key(&fl4, sk, skb);
986 __ip_rt_update_pmtu(rt, &fl4, mtu);
989 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
990 int oif, u32 mark, u8 protocol, int flow_flags)
992 const struct iphdr *iph = (const struct iphdr *) skb->data;
996 __build_flow_key(&fl4, NULL, iph, oif,
997 RT_TOS(iph->tos), protocol, mark, flow_flags);
998 rt = __ip_route_output_key(net, &fl4);
1000 __ip_rt_update_pmtu(rt, &fl4, mtu);
1004 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1006 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1008 const struct iphdr *iph = (const struct iphdr *) skb->data;
1012 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1013 rt = __ip_route_output_key(sock_net(sk), &fl4);
1015 __ip_rt_update_pmtu(rt, &fl4, mtu);
1020 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1022 const struct iphdr *iph = (const struct iphdr *) skb->data;
1025 struct dst_entry *dst;
1030 if (!ip_sk_accept_pmtu(sk))
1033 rt = (struct rtable *) __sk_dst_get(sk);
1035 if (sock_owned_by_user(sk) || !rt) {
1036 __ipv4_sk_update_pmtu(skb, sk, mtu);
1040 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1042 if (!__sk_dst_check(sk, 0)) {
1043 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1050 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1052 dst = dst_check(&rt->dst, 0);
1055 dst_release(&rt->dst);
1057 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1065 __sk_dst_set(sk, &rt->dst);
1070 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1072 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1073 int oif, u32 mark, u8 protocol, int flow_flags)
1075 const struct iphdr *iph = (const struct iphdr *) skb->data;
1079 __build_flow_key(&fl4, NULL, iph, oif,
1080 RT_TOS(iph->tos), protocol, mark, flow_flags);
1081 rt = __ip_route_output_key(net, &fl4);
1083 __ip_do_redirect(rt, skb, &fl4, false);
1087 EXPORT_SYMBOL_GPL(ipv4_redirect);
1089 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1091 const struct iphdr *iph = (const struct iphdr *) skb->data;
1095 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1096 rt = __ip_route_output_key(sock_net(sk), &fl4);
1098 __ip_do_redirect(rt, skb, &fl4, false);
1102 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1104 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1106 struct rtable *rt = (struct rtable *) dst;
1108 /* All IPV4 dsts are created with ->obsolete set to the value
1109 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1110 * into this function always.
1112 * When a PMTU/redirect information update invalidates a route,
1113 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1114 * DST_OBSOLETE_DEAD by dst_free().
1116 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1121 static void ipv4_link_failure(struct sk_buff *skb)
1125 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1127 rt = skb_rtable(skb);
1129 dst_set_expires(&rt->dst, 0);
1132 static int ip_rt_bug(struct sock *sk, struct sk_buff *skb)
1134 pr_debug("%s: %pI4 -> %pI4, %s\n",
1135 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1136 skb->dev ? skb->dev->name : "?");
1143 We do not cache source address of outgoing interface,
1144 because it is used only by IP RR, TS and SRR options,
1145 so that it out of fast path.
1147 BTW remember: "addr" is allowed to be not aligned
1151 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1155 if (rt_is_output_route(rt))
1156 src = ip_hdr(skb)->saddr;
1158 struct fib_result res;
1164 memset(&fl4, 0, sizeof(fl4));
1165 fl4.daddr = iph->daddr;
1166 fl4.saddr = iph->saddr;
1167 fl4.flowi4_tos = RT_TOS(iph->tos);
1168 fl4.flowi4_oif = rt->dst.dev->ifindex;
1169 fl4.flowi4_iif = skb->dev->ifindex;
1170 fl4.flowi4_mark = skb->mark;
1173 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1174 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1176 src = inet_select_addr(rt->dst.dev,
1177 rt_nexthop(rt, iph->daddr),
1181 memcpy(addr, &src, 4);
1184 #ifdef CONFIG_IP_ROUTE_CLASSID
1185 static void set_class_tag(struct rtable *rt, u32 tag)
1187 if (!(rt->dst.tclassid & 0xFFFF))
1188 rt->dst.tclassid |= tag & 0xFFFF;
1189 if (!(rt->dst.tclassid & 0xFFFF0000))
1190 rt->dst.tclassid |= tag & 0xFFFF0000;
1194 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1196 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1199 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1201 if (advmss > 65535 - 40)
1202 advmss = 65535 - 40;
1207 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1209 const struct rtable *rt = (const struct rtable *) dst;
1210 unsigned int mtu = rt->rt_pmtu;
1212 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1213 mtu = dst_metric_raw(dst, RTAX_MTU);
1218 mtu = dst->dev->mtu;
1220 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1221 if (rt->rt_uses_gateway && mtu > 576)
1225 return min_t(unsigned int, mtu, IP_MAX_MTU);
1228 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1230 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1231 struct fib_nh_exception *fnhe;
1237 hval = fnhe_hashfun(daddr);
1239 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1240 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1241 if (fnhe->fnhe_daddr == daddr)
1247 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1252 spin_lock_bh(&fnhe_lock);
1254 if (daddr == fnhe->fnhe_daddr) {
1255 struct rtable __rcu **porig;
1256 struct rtable *orig;
1257 int genid = fnhe_genid(dev_net(rt->dst.dev));
1259 if (rt_is_input_route(rt))
1260 porig = &fnhe->fnhe_rth_input;
1262 porig = &fnhe->fnhe_rth_output;
1263 orig = rcu_dereference(*porig);
1265 if (fnhe->fnhe_genid != genid) {
1266 fnhe->fnhe_genid = genid;
1268 fnhe->fnhe_pmtu = 0;
1269 fnhe->fnhe_expires = 0;
1270 fnhe_flush_routes(fnhe);
1273 fill_route_from_fnhe(rt, fnhe);
1274 if (!rt->rt_gateway)
1275 rt->rt_gateway = daddr;
1277 if (!(rt->dst.flags & DST_NOCACHE)) {
1278 rcu_assign_pointer(*porig, rt);
1284 fnhe->fnhe_stamp = jiffies;
1286 spin_unlock_bh(&fnhe_lock);
1291 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1293 struct rtable *orig, *prev, **p;
1296 if (rt_is_input_route(rt)) {
1297 p = (struct rtable **)&nh->nh_rth_input;
1299 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1303 prev = cmpxchg(p, orig, rt);
1313 static DEFINE_SPINLOCK(rt_uncached_lock);
1314 static LIST_HEAD(rt_uncached_list);
1316 static void rt_add_uncached_list(struct rtable *rt)
1318 spin_lock_bh(&rt_uncached_lock);
1319 list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1320 spin_unlock_bh(&rt_uncached_lock);
1323 static void ipv4_dst_destroy(struct dst_entry *dst)
1325 struct rtable *rt = (struct rtable *) dst;
1327 if (!list_empty(&rt->rt_uncached)) {
1328 spin_lock_bh(&rt_uncached_lock);
1329 list_del(&rt->rt_uncached);
1330 spin_unlock_bh(&rt_uncached_lock);
1334 void rt_flush_dev(struct net_device *dev)
1336 if (!list_empty(&rt_uncached_list)) {
1337 struct net *net = dev_net(dev);
1340 spin_lock_bh(&rt_uncached_lock);
1341 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1342 if (rt->dst.dev != dev)
1344 rt->dst.dev = net->loopback_dev;
1345 dev_hold(rt->dst.dev);
1348 spin_unlock_bh(&rt_uncached_lock);
1352 static bool rt_cache_valid(const struct rtable *rt)
1355 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1359 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1360 const struct fib_result *res,
1361 struct fib_nh_exception *fnhe,
1362 struct fib_info *fi, u16 type, u32 itag)
1364 bool cached = false;
1367 struct fib_nh *nh = &FIB_RES_NH(*res);
1369 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1370 rt->rt_gateway = nh->nh_gw;
1371 rt->rt_uses_gateway = 1;
1373 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1374 #ifdef CONFIG_IP_ROUTE_CLASSID
1375 rt->dst.tclassid = nh->nh_tclassid;
1378 cached = rt_bind_exception(rt, fnhe, daddr);
1379 else if (!(rt->dst.flags & DST_NOCACHE))
1380 cached = rt_cache_route(nh, rt);
1381 if (unlikely(!cached)) {
1382 /* Routes we intend to cache in nexthop exception or
1383 * FIB nexthop have the DST_NOCACHE bit clear.
1384 * However, if we are unsuccessful at storing this
1385 * route into the cache we really need to set it.
1387 rt->dst.flags |= DST_NOCACHE;
1388 if (!rt->rt_gateway)
1389 rt->rt_gateway = daddr;
1390 rt_add_uncached_list(rt);
1393 rt_add_uncached_list(rt);
1395 #ifdef CONFIG_IP_ROUTE_CLASSID
1396 #ifdef CONFIG_IP_MULTIPLE_TABLES
1397 set_class_tag(rt, res->tclassid);
1399 set_class_tag(rt, itag);
1403 static struct rtable *rt_dst_alloc(struct net_device *dev,
1404 bool nopolicy, bool noxfrm, bool will_cache)
1406 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1407 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1408 (nopolicy ? DST_NOPOLICY : 0) |
1409 (noxfrm ? DST_NOXFRM : 0));
1412 /* called in rcu_read_lock() section */
1413 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1414 u8 tos, struct net_device *dev, int our)
1417 struct in_device *in_dev = __in_dev_get_rcu(dev);
1421 /* Primary sanity checks. */
1426 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1427 skb->protocol != htons(ETH_P_IP))
1430 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1431 if (ipv4_is_loopback(saddr))
1434 if (ipv4_is_zeronet(saddr)) {
1435 if (!ipv4_is_local_multicast(daddr))
1438 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1443 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1444 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1448 #ifdef CONFIG_IP_ROUTE_CLASSID
1449 rth->dst.tclassid = itag;
1451 rth->dst.output = ip_rt_bug;
1453 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
1454 rth->rt_flags = RTCF_MULTICAST;
1455 rth->rt_type = RTN_MULTICAST;
1456 rth->rt_is_input= 1;
1459 rth->rt_gateway = 0;
1460 rth->rt_uses_gateway = 0;
1461 INIT_LIST_HEAD(&rth->rt_uncached);
1463 rth->dst.input= ip_local_deliver;
1464 rth->rt_flags |= RTCF_LOCAL;
1467 #ifdef CONFIG_IP_MROUTE
1468 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1469 rth->dst.input = ip_mr_input;
1471 RT_CACHE_STAT_INC(in_slow_mc);
1473 skb_dst_set(skb, &rth->dst);
1485 static void ip_handle_martian_source(struct net_device *dev,
1486 struct in_device *in_dev,
1487 struct sk_buff *skb,
1491 RT_CACHE_STAT_INC(in_martian_src);
1492 #ifdef CONFIG_IP_ROUTE_VERBOSE
1493 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1495 * RFC1812 recommendation, if source is martian,
1496 * the only hint is MAC header.
1498 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1499 &daddr, &saddr, dev->name);
1500 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1501 print_hex_dump(KERN_WARNING, "ll header: ",
1502 DUMP_PREFIX_OFFSET, 16, 1,
1503 skb_mac_header(skb),
1504 dev->hard_header_len, true);
1510 /* called in rcu_read_lock() section */
1511 static int __mkroute_input(struct sk_buff *skb,
1512 const struct fib_result *res,
1513 struct in_device *in_dev,
1514 __be32 daddr, __be32 saddr, u32 tos)
1516 struct fib_nh_exception *fnhe;
1519 struct in_device *out_dev;
1520 unsigned int flags = 0;
1524 /* get a working reference to the output device */
1525 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1526 if (out_dev == NULL) {
1527 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1531 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1532 in_dev->dev, in_dev, &itag);
1534 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1540 do_cache = res->fi && !itag;
1541 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1542 (IN_DEV_SHARED_MEDIA(out_dev) ||
1543 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1544 flags |= RTCF_DOREDIRECT;
1548 if (skb->protocol != htons(ETH_P_IP)) {
1549 /* Not IP (i.e. ARP). Do not create route, if it is
1550 * invalid for proxy arp. DNAT routes are always valid.
1552 * Proxy arp feature have been extended to allow, ARP
1553 * replies back to the same interface, to support
1554 * Private VLAN switch technologies. See arp.c.
1556 if (out_dev == in_dev &&
1557 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1563 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1566 rth = rcu_dereference(fnhe->fnhe_rth_input);
1568 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1570 if (rt_cache_valid(rth)) {
1571 skb_dst_set_noref(skb, &rth->dst);
1576 rth = rt_dst_alloc(out_dev->dev,
1577 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1578 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1584 rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1585 rth->rt_flags = flags;
1586 rth->rt_type = res->type;
1587 rth->rt_is_input = 1;
1590 rth->rt_gateway = 0;
1591 rth->rt_uses_gateway = 0;
1592 INIT_LIST_HEAD(&rth->rt_uncached);
1593 RT_CACHE_STAT_INC(in_slow_tot);
1595 rth->dst.input = ip_forward;
1596 rth->dst.output = ip_output;
1598 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1599 skb_dst_set(skb, &rth->dst);
1606 static int ip_mkroute_input(struct sk_buff *skb,
1607 struct fib_result *res,
1608 const struct flowi4 *fl4,
1609 struct in_device *in_dev,
1610 __be32 daddr, __be32 saddr, u32 tos)
1612 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1613 if (res->fi && res->fi->fib_nhs > 1)
1614 fib_select_multipath(res);
1617 /* create a routing cache entry */
1618 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1622 * NOTE. We drop all the packets that has local source
1623 * addresses, because every properly looped back packet
1624 * must have correct destination already attached by output routine.
1626 * Such approach solves two big problems:
1627 * 1. Not simplex devices are handled properly.
1628 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1629 * called with rcu_read_lock()
1632 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1633 u8 tos, struct net_device *dev)
1635 struct fib_result res;
1636 struct in_device *in_dev = __in_dev_get_rcu(dev);
1638 unsigned int flags = 0;
1642 struct net *net = dev_net(dev);
1645 /* IP on this device is disabled. */
1650 /* Check for the most weird martians, which can be not detected
1654 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1655 goto martian_source;
1658 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1661 /* Accept zero addresses only to limited broadcast;
1662 * I even do not know to fix it or not. Waiting for complains :-)
1664 if (ipv4_is_zeronet(saddr))
1665 goto martian_source;
1667 if (ipv4_is_zeronet(daddr))
1668 goto martian_destination;
1670 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1671 * and call it once if daddr or/and saddr are loopback addresses
1673 if (ipv4_is_loopback(daddr)) {
1674 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1675 goto martian_destination;
1676 } else if (ipv4_is_loopback(saddr)) {
1677 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1678 goto martian_source;
1682 * Now we are ready to route packet.
1685 fl4.flowi4_iif = dev->ifindex;
1686 fl4.flowi4_mark = skb->mark;
1687 fl4.flowi4_tos = tos;
1688 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1691 err = fib_lookup(net, &fl4, &res);
1693 if (!IN_DEV_FORWARD(in_dev))
1694 err = -EHOSTUNREACH;
1698 if (res.type == RTN_BROADCAST)
1701 if (res.type == RTN_LOCAL) {
1702 err = fib_validate_source(skb, saddr, daddr, tos,
1703 0, dev, in_dev, &itag);
1705 goto martian_source_keep_err;
1709 if (!IN_DEV_FORWARD(in_dev)) {
1710 err = -EHOSTUNREACH;
1713 if (res.type != RTN_UNICAST)
1714 goto martian_destination;
1716 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1720 if (skb->protocol != htons(ETH_P_IP))
1723 if (!ipv4_is_zeronet(saddr)) {
1724 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1727 goto martian_source_keep_err;
1729 flags |= RTCF_BROADCAST;
1730 res.type = RTN_BROADCAST;
1731 RT_CACHE_STAT_INC(in_brd);
1737 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1738 if (rt_cache_valid(rth)) {
1739 skb_dst_set_noref(skb, &rth->dst);
1747 rth = rt_dst_alloc(net->loopback_dev,
1748 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1752 rth->dst.input= ip_local_deliver;
1753 rth->dst.output= ip_rt_bug;
1754 #ifdef CONFIG_IP_ROUTE_CLASSID
1755 rth->dst.tclassid = itag;
1758 rth->rt_genid = rt_genid_ipv4(net);
1759 rth->rt_flags = flags|RTCF_LOCAL;
1760 rth->rt_type = res.type;
1761 rth->rt_is_input = 1;
1764 rth->rt_gateway = 0;
1765 rth->rt_uses_gateway = 0;
1766 INIT_LIST_HEAD(&rth->rt_uncached);
1767 RT_CACHE_STAT_INC(in_slow_tot);
1768 if (res.type == RTN_UNREACHABLE) {
1769 rth->dst.input= ip_error;
1770 rth->dst.error= -err;
1771 rth->rt_flags &= ~RTCF_LOCAL;
1774 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1775 rth->dst.flags |= DST_NOCACHE;
1776 rt_add_uncached_list(rth);
1779 skb_dst_set(skb, &rth->dst);
1784 RT_CACHE_STAT_INC(in_no_route);
1785 res.type = RTN_UNREACHABLE;
1791 * Do not cache martian addresses: they should be logged (RFC1812)
1793 martian_destination:
1794 RT_CACHE_STAT_INC(in_martian_dst);
1795 #ifdef CONFIG_IP_ROUTE_VERBOSE
1796 if (IN_DEV_LOG_MARTIANS(in_dev))
1797 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1798 &daddr, &saddr, dev->name);
1811 martian_source_keep_err:
1812 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1816 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1817 u8 tos, struct net_device *dev)
1823 /* Multicast recognition logic is moved from route cache to here.
1824 The problem was that too many Ethernet cards have broken/missing
1825 hardware multicast filters :-( As result the host on multicasting
1826 network acquires a lot of useless route cache entries, sort of
1827 SDR messages from all the world. Now we try to get rid of them.
1828 Really, provided software IP multicast filter is organized
1829 reasonably (at least, hashed), it does not result in a slowdown
1830 comparing with route cache reject entries.
1831 Note, that multicast routers are not affected, because
1832 route cache entry is created eventually.
1834 if (ipv4_is_multicast(daddr)) {
1835 struct in_device *in_dev = __in_dev_get_rcu(dev);
1838 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1839 ip_hdr(skb)->protocol);
1841 #ifdef CONFIG_IP_MROUTE
1843 (!ipv4_is_local_multicast(daddr) &&
1844 IN_DEV_MFORWARD(in_dev))
1847 int res = ip_route_input_mc(skb, daddr, saddr,
1856 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1860 EXPORT_SYMBOL(ip_route_input_noref);
1862 /* called with rcu_read_lock() */
1863 static struct rtable *__mkroute_output(const struct fib_result *res,
1864 const struct flowi4 *fl4, int orig_oif,
1865 struct net_device *dev_out,
1868 struct fib_info *fi = res->fi;
1869 struct fib_nh_exception *fnhe;
1870 struct in_device *in_dev;
1871 u16 type = res->type;
1875 in_dev = __in_dev_get_rcu(dev_out);
1877 return ERR_PTR(-EINVAL);
1879 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1880 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1881 return ERR_PTR(-EINVAL);
1883 if (ipv4_is_lbcast(fl4->daddr))
1884 type = RTN_BROADCAST;
1885 else if (ipv4_is_multicast(fl4->daddr))
1886 type = RTN_MULTICAST;
1887 else if (ipv4_is_zeronet(fl4->daddr))
1888 return ERR_PTR(-EINVAL);
1890 if (dev_out->flags & IFF_LOOPBACK)
1891 flags |= RTCF_LOCAL;
1894 if (type == RTN_BROADCAST) {
1895 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1897 } else if (type == RTN_MULTICAST) {
1898 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1899 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1901 flags &= ~RTCF_LOCAL;
1904 /* If multicast route do not exist use
1905 * default one, but do not gateway in this case.
1908 if (fi && res->prefixlen < 4)
1913 do_cache &= fi != NULL;
1915 struct rtable __rcu **prth;
1916 struct fib_nh *nh = &FIB_RES_NH(*res);
1918 fnhe = find_exception(nh, fl4->daddr);
1920 prth = &fnhe->fnhe_rth_output;
1922 if (unlikely(fl4->flowi4_flags &
1923 FLOWI_FLAG_KNOWN_NH &&
1925 nh->nh_scope == RT_SCOPE_LINK))) {
1929 prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1931 rth = rcu_dereference(*prth);
1932 if (rt_cache_valid(rth)) {
1933 dst_hold(&rth->dst);
1939 rth = rt_dst_alloc(dev_out,
1940 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1941 IN_DEV_CONF_GET(in_dev, NOXFRM),
1944 return ERR_PTR(-ENOBUFS);
1946 rth->dst.output = ip_output;
1948 rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1949 rth->rt_flags = flags;
1950 rth->rt_type = type;
1951 rth->rt_is_input = 0;
1952 rth->rt_iif = orig_oif ? : 0;
1954 rth->rt_gateway = 0;
1955 rth->rt_uses_gateway = 0;
1956 INIT_LIST_HEAD(&rth->rt_uncached);
1958 RT_CACHE_STAT_INC(out_slow_tot);
1960 if (flags & RTCF_LOCAL)
1961 rth->dst.input = ip_local_deliver;
1962 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1963 if (flags & RTCF_LOCAL &&
1964 !(dev_out->flags & IFF_LOOPBACK)) {
1965 rth->dst.output = ip_mc_output;
1966 RT_CACHE_STAT_INC(out_slow_mc);
1968 #ifdef CONFIG_IP_MROUTE
1969 if (type == RTN_MULTICAST) {
1970 if (IN_DEV_MFORWARD(in_dev) &&
1971 !ipv4_is_local_multicast(fl4->daddr)) {
1972 rth->dst.input = ip_mr_input;
1973 rth->dst.output = ip_mc_output;
1979 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1985 * Major route resolver routine.
1988 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1990 struct net_device *dev_out = NULL;
1991 __u8 tos = RT_FL_TOS(fl4);
1992 unsigned int flags = 0;
1993 struct fib_result res;
2001 orig_oif = fl4->flowi4_oif;
2003 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2004 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2005 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2006 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2010 rth = ERR_PTR(-EINVAL);
2011 if (ipv4_is_multicast(fl4->saddr) ||
2012 ipv4_is_lbcast(fl4->saddr) ||
2013 ipv4_is_zeronet(fl4->saddr))
2016 /* I removed check for oif == dev_out->oif here.
2017 It was wrong for two reasons:
2018 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2019 is assigned to multiple interfaces.
2020 2. Moreover, we are allowed to send packets with saddr
2021 of another iface. --ANK
2024 if (fl4->flowi4_oif == 0 &&
2025 (ipv4_is_multicast(fl4->daddr) ||
2026 ipv4_is_lbcast(fl4->daddr))) {
2027 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2028 dev_out = __ip_dev_find(net, fl4->saddr, false);
2029 if (dev_out == NULL)
2032 /* Special hack: user can direct multicasts
2033 and limited broadcast via necessary interface
2034 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2035 This hack is not just for fun, it allows
2036 vic,vat and friends to work.
2037 They bind socket to loopback, set ttl to zero
2038 and expect that it will work.
2039 From the viewpoint of routing cache they are broken,
2040 because we are not allowed to build multicast path
2041 with loopback source addr (look, routing cache
2042 cannot know, that ttl is zero, so that packet
2043 will not leave this host and route is valid).
2044 Luckily, this hack is good workaround.
2047 fl4->flowi4_oif = dev_out->ifindex;
2051 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2052 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2053 if (!__ip_dev_find(net, fl4->saddr, false))
2059 if (fl4->flowi4_oif) {
2060 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2061 rth = ERR_PTR(-ENODEV);
2062 if (dev_out == NULL)
2065 /* RACE: Check return value of inet_select_addr instead. */
2066 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2067 rth = ERR_PTR(-ENETUNREACH);
2070 if (ipv4_is_local_multicast(fl4->daddr) ||
2071 ipv4_is_lbcast(fl4->daddr)) {
2073 fl4->saddr = inet_select_addr(dev_out, 0,
2078 if (ipv4_is_multicast(fl4->daddr))
2079 fl4->saddr = inet_select_addr(dev_out, 0,
2081 else if (!fl4->daddr)
2082 fl4->saddr = inet_select_addr(dev_out, 0,
2088 fl4->daddr = fl4->saddr;
2090 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2091 dev_out = net->loopback_dev;
2092 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2093 res.type = RTN_LOCAL;
2094 flags |= RTCF_LOCAL;
2098 if (fib_lookup(net, fl4, &res)) {
2101 if (fl4->flowi4_oif) {
2102 /* Apparently, routing tables are wrong. Assume,
2103 that the destination is on link.
2106 Because we are allowed to send to iface
2107 even if it has NO routes and NO assigned
2108 addresses. When oif is specified, routing
2109 tables are looked up with only one purpose:
2110 to catch if destination is gatewayed, rather than
2111 direct. Moreover, if MSG_DONTROUTE is set,
2112 we send packet, ignoring both routing tables
2113 and ifaddr state. --ANK
2116 We could make it even if oif is unknown,
2117 likely IPv6, but we do not.
2120 if (fl4->saddr == 0)
2121 fl4->saddr = inet_select_addr(dev_out, 0,
2123 res.type = RTN_UNICAST;
2126 rth = ERR_PTR(-ENETUNREACH);
2130 if (res.type == RTN_LOCAL) {
2132 if (res.fi->fib_prefsrc)
2133 fl4->saddr = res.fi->fib_prefsrc;
2135 fl4->saddr = fl4->daddr;
2137 dev_out = net->loopback_dev;
2138 fl4->flowi4_oif = dev_out->ifindex;
2139 flags |= RTCF_LOCAL;
2143 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2144 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2145 fib_select_multipath(&res);
2148 if (!res.prefixlen &&
2149 res.table->tb_num_default > 1 &&
2150 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2151 fib_select_default(&res);
2154 fl4->saddr = FIB_RES_PREFSRC(net, res);
2156 dev_out = FIB_RES_DEV(res);
2157 fl4->flowi4_oif = dev_out->ifindex;
2161 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2167 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2169 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2174 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2176 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2178 return mtu ? : dst->dev->mtu;
2181 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2182 struct sk_buff *skb, u32 mtu)
2186 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2187 struct sk_buff *skb)
2191 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2197 static struct dst_ops ipv4_dst_blackhole_ops = {
2199 .protocol = cpu_to_be16(ETH_P_IP),
2200 .check = ipv4_blackhole_dst_check,
2201 .mtu = ipv4_blackhole_mtu,
2202 .default_advmss = ipv4_default_advmss,
2203 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2204 .redirect = ipv4_rt_blackhole_redirect,
2205 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2206 .neigh_lookup = ipv4_neigh_lookup,
2209 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2211 struct rtable *ort = (struct rtable *) dst_orig;
2214 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2216 struct dst_entry *new = &rt->dst;
2219 new->input = dst_discard;
2220 new->output = dst_discard_sk;
2222 new->dev = ort->dst.dev;
2226 rt->rt_is_input = ort->rt_is_input;
2227 rt->rt_iif = ort->rt_iif;
2228 rt->rt_pmtu = ort->rt_pmtu;
2230 rt->rt_genid = rt_genid_ipv4(net);
2231 rt->rt_flags = ort->rt_flags;
2232 rt->rt_type = ort->rt_type;
2233 rt->rt_gateway = ort->rt_gateway;
2234 rt->rt_uses_gateway = ort->rt_uses_gateway;
2236 INIT_LIST_HEAD(&rt->rt_uncached);
2241 dst_release(dst_orig);
2243 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2246 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2249 struct rtable *rt = __ip_route_output_key(net, flp4);
2254 if (flp4->flowi4_proto)
2255 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2256 flowi4_to_flowi(flp4),
2261 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2263 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2264 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2265 u32 seq, int event, int nowait, unsigned int flags)
2267 struct rtable *rt = skb_rtable(skb);
2269 struct nlmsghdr *nlh;
2270 unsigned long expires = 0;
2272 u32 metrics[RTAX_MAX];
2274 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2278 r = nlmsg_data(nlh);
2279 r->rtm_family = AF_INET;
2280 r->rtm_dst_len = 32;
2282 r->rtm_tos = fl4->flowi4_tos;
2283 r->rtm_table = RT_TABLE_MAIN;
2284 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2285 goto nla_put_failure;
2286 r->rtm_type = rt->rt_type;
2287 r->rtm_scope = RT_SCOPE_UNIVERSE;
2288 r->rtm_protocol = RTPROT_UNSPEC;
2289 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2290 if (rt->rt_flags & RTCF_NOTIFY)
2291 r->rtm_flags |= RTM_F_NOTIFY;
2293 if (nla_put_be32(skb, RTA_DST, dst))
2294 goto nla_put_failure;
2296 r->rtm_src_len = 32;
2297 if (nla_put_be32(skb, RTA_SRC, src))
2298 goto nla_put_failure;
2301 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2302 goto nla_put_failure;
2303 #ifdef CONFIG_IP_ROUTE_CLASSID
2304 if (rt->dst.tclassid &&
2305 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2306 goto nla_put_failure;
2308 if (!rt_is_input_route(rt) &&
2309 fl4->saddr != src) {
2310 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2311 goto nla_put_failure;
2313 if (rt->rt_uses_gateway &&
2314 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2315 goto nla_put_failure;
2317 expires = rt->dst.expires;
2319 unsigned long now = jiffies;
2321 if (time_before(now, expires))
2327 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2328 if (rt->rt_pmtu && expires)
2329 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2330 if (rtnetlink_put_metrics(skb, metrics) < 0)
2331 goto nla_put_failure;
2333 if (fl4->flowi4_mark &&
2334 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2335 goto nla_put_failure;
2337 error = rt->dst.error;
2339 if (rt_is_input_route(rt)) {
2340 #ifdef CONFIG_IP_MROUTE
2341 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2342 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2343 int err = ipmr_get_route(net, skb,
2344 fl4->saddr, fl4->daddr,
2350 goto nla_put_failure;
2352 if (err == -EMSGSIZE)
2353 goto nla_put_failure;
2359 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2360 goto nla_put_failure;
2363 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2364 goto nla_put_failure;
2366 return nlmsg_end(skb, nlh);
2369 nlmsg_cancel(skb, nlh);
2373 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2375 struct net *net = sock_net(in_skb->sk);
2377 struct nlattr *tb[RTA_MAX+1];
2378 struct rtable *rt = NULL;
2385 struct sk_buff *skb;
2387 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2391 rtm = nlmsg_data(nlh);
2393 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2399 /* Reserve room for dummy headers, this skb can pass
2400 through good chunk of routing engine.
2402 skb_reset_mac_header(skb);
2403 skb_reset_network_header(skb);
2405 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2406 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2407 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2409 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2410 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2411 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2412 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2414 memset(&fl4, 0, sizeof(fl4));
2417 fl4.flowi4_tos = rtm->rtm_tos;
2418 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2419 fl4.flowi4_mark = mark;
2422 struct net_device *dev;
2424 dev = __dev_get_by_index(net, iif);
2430 skb->protocol = htons(ETH_P_IP);
2434 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2437 rt = skb_rtable(skb);
2438 if (err == 0 && rt->dst.error)
2439 err = -rt->dst.error;
2441 rt = ip_route_output_key(net, &fl4);
2451 skb_dst_set(skb, &rt->dst);
2452 if (rtm->rtm_flags & RTM_F_NOTIFY)
2453 rt->rt_flags |= RTCF_NOTIFY;
2455 err = rt_fill_info(net, dst, src, &fl4, skb,
2456 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2457 RTM_NEWROUTE, 0, 0);
2461 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2470 void ip_rt_multicast_event(struct in_device *in_dev)
2472 rt_cache_flush(dev_net(in_dev->dev));
2475 #ifdef CONFIG_SYSCTL
2476 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2477 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2478 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2479 static int ip_rt_gc_elasticity __read_mostly = 8;
2481 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2482 void __user *buffer,
2483 size_t *lenp, loff_t *ppos)
2485 struct net *net = (struct net *)__ctl->extra1;
2488 rt_cache_flush(net);
2489 fnhe_genid_bump(net);
2496 static struct ctl_table ipv4_route_table[] = {
2498 .procname = "gc_thresh",
2499 .data = &ipv4_dst_ops.gc_thresh,
2500 .maxlen = sizeof(int),
2502 .proc_handler = proc_dointvec,
2505 .procname = "max_size",
2506 .data = &ip_rt_max_size,
2507 .maxlen = sizeof(int),
2509 .proc_handler = proc_dointvec,
2512 /* Deprecated. Use gc_min_interval_ms */
2514 .procname = "gc_min_interval",
2515 .data = &ip_rt_gc_min_interval,
2516 .maxlen = sizeof(int),
2518 .proc_handler = proc_dointvec_jiffies,
2521 .procname = "gc_min_interval_ms",
2522 .data = &ip_rt_gc_min_interval,
2523 .maxlen = sizeof(int),
2525 .proc_handler = proc_dointvec_ms_jiffies,
2528 .procname = "gc_timeout",
2529 .data = &ip_rt_gc_timeout,
2530 .maxlen = sizeof(int),
2532 .proc_handler = proc_dointvec_jiffies,
2535 .procname = "gc_interval",
2536 .data = &ip_rt_gc_interval,
2537 .maxlen = sizeof(int),
2539 .proc_handler = proc_dointvec_jiffies,
2542 .procname = "redirect_load",
2543 .data = &ip_rt_redirect_load,
2544 .maxlen = sizeof(int),
2546 .proc_handler = proc_dointvec,
2549 .procname = "redirect_number",
2550 .data = &ip_rt_redirect_number,
2551 .maxlen = sizeof(int),
2553 .proc_handler = proc_dointvec,
2556 .procname = "redirect_silence",
2557 .data = &ip_rt_redirect_silence,
2558 .maxlen = sizeof(int),
2560 .proc_handler = proc_dointvec,
2563 .procname = "error_cost",
2564 .data = &ip_rt_error_cost,
2565 .maxlen = sizeof(int),
2567 .proc_handler = proc_dointvec,
2570 .procname = "error_burst",
2571 .data = &ip_rt_error_burst,
2572 .maxlen = sizeof(int),
2574 .proc_handler = proc_dointvec,
2577 .procname = "gc_elasticity",
2578 .data = &ip_rt_gc_elasticity,
2579 .maxlen = sizeof(int),
2581 .proc_handler = proc_dointvec,
2584 .procname = "mtu_expires",
2585 .data = &ip_rt_mtu_expires,
2586 .maxlen = sizeof(int),
2588 .proc_handler = proc_dointvec_jiffies,
2591 .procname = "min_pmtu",
2592 .data = &ip_rt_min_pmtu,
2593 .maxlen = sizeof(int),
2595 .proc_handler = proc_dointvec,
2598 .procname = "min_adv_mss",
2599 .data = &ip_rt_min_advmss,
2600 .maxlen = sizeof(int),
2602 .proc_handler = proc_dointvec,
2607 static struct ctl_table ipv4_route_flush_table[] = {
2609 .procname = "flush",
2610 .maxlen = sizeof(int),
2612 .proc_handler = ipv4_sysctl_rtcache_flush,
2617 static __net_init int sysctl_route_net_init(struct net *net)
2619 struct ctl_table *tbl;
2621 tbl = ipv4_route_flush_table;
2622 if (!net_eq(net, &init_net)) {
2623 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2627 /* Don't export sysctls to unprivileged users */
2628 if (net->user_ns != &init_user_ns)
2629 tbl[0].procname = NULL;
2631 tbl[0].extra1 = net;
2633 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2634 if (net->ipv4.route_hdr == NULL)
2639 if (tbl != ipv4_route_flush_table)
2645 static __net_exit void sysctl_route_net_exit(struct net *net)
2647 struct ctl_table *tbl;
2649 tbl = net->ipv4.route_hdr->ctl_table_arg;
2650 unregister_net_sysctl_table(net->ipv4.route_hdr);
2651 BUG_ON(tbl == ipv4_route_flush_table);
2655 static __net_initdata struct pernet_operations sysctl_route_ops = {
2656 .init = sysctl_route_net_init,
2657 .exit = sysctl_route_net_exit,
2661 static __net_init int rt_genid_init(struct net *net)
2663 atomic_set(&net->ipv4.rt_genid, 0);
2664 atomic_set(&net->fnhe_genid, 0);
2665 get_random_bytes(&net->ipv4.dev_addr_genid,
2666 sizeof(net->ipv4.dev_addr_genid));
2670 static __net_initdata struct pernet_operations rt_genid_ops = {
2671 .init = rt_genid_init,
2674 static int __net_init ipv4_inetpeer_init(struct net *net)
2676 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2680 inet_peer_base_init(bp);
2681 net->ipv4.peers = bp;
2685 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2687 struct inet_peer_base *bp = net->ipv4.peers;
2689 net->ipv4.peers = NULL;
2690 inetpeer_invalidate_tree(bp);
2694 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2695 .init = ipv4_inetpeer_init,
2696 .exit = ipv4_inetpeer_exit,
2699 #ifdef CONFIG_IP_ROUTE_CLASSID
2700 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2701 #endif /* CONFIG_IP_ROUTE_CLASSID */
2703 int __init ip_rt_init(void)
2707 #ifdef CONFIG_IP_ROUTE_CLASSID
2708 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2710 panic("IP: failed to allocate ip_rt_acct\n");
2713 ipv4_dst_ops.kmem_cachep =
2714 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2715 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2717 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2719 if (dst_entries_init(&ipv4_dst_ops) < 0)
2720 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2722 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2723 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2725 ipv4_dst_ops.gc_thresh = ~0;
2726 ip_rt_max_size = INT_MAX;
2731 if (ip_rt_proc_init())
2732 pr_err("Unable to create route proc files\n");
2737 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2739 #ifdef CONFIG_SYSCTL
2740 register_pernet_subsys(&sysctl_route_ops);
2742 register_pernet_subsys(&rt_genid_ops);
2743 register_pernet_subsys(&ipv4_inetpeer_ops);
2747 #ifdef CONFIG_SYSCTL
2749 * We really need to sanitize the damn ipv4 init order, then all
2750 * this nonsense will go away.
2752 void __init ip_static_sysctl_init(void)
2754 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);