2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
94 #include <net/net_namespace.h>
95 #include <net/protocol.h>
97 #include <net/route.h>
98 #include <net/inetpeer.h>
100 #include <net/ip_fib.h>
103 #include <net/icmp.h>
104 #include <net/xfrm.h>
105 #include <net/netevent.h>
106 #include <net/rtnetlink.h>
108 #include <linux/sysctl.h>
109 #include <linux/kmemleak.h>
111 #include <net/secure_seq.h>
113 #define RT_FL_TOS(oldflp4) \
114 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
116 #define IP_MAX_MTU 0xFFF0
118 #define RT_GC_TIMEOUT (300*HZ)
120 static int ip_rt_max_size;
121 static int ip_rt_redirect_number __read_mostly = 9;
122 static int ip_rt_redirect_load __read_mostly = HZ / 50;
123 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
124 static int ip_rt_error_cost __read_mostly = HZ;
125 static int ip_rt_error_burst __read_mostly = 5 * HZ;
126 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
127 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
128 static int ip_rt_min_advmss __read_mostly = 256;
131 * Interface to generic destination cache.
134 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
135 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
136 static unsigned int ipv4_mtu(const struct dst_entry *dst);
137 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
138 static void ipv4_link_failure(struct sk_buff *skb);
139 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
140 struct sk_buff *skb, u32 mtu);
141 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb);
143 static void ipv4_dst_destroy(struct dst_entry *dst);
145 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
160 static struct dst_ops ipv4_dst_ops = {
162 .protocol = cpu_to_be16(ETH_P_IP),
163 .check = ipv4_dst_check,
164 .default_advmss = ipv4_default_advmss,
166 .cow_metrics = ipv4_cow_metrics,
167 .destroy = ipv4_dst_destroy,
168 .ifdown = ipv4_dst_ifdown,
169 .negative_advice = ipv4_negative_advice,
170 .link_failure = ipv4_link_failure,
171 .update_pmtu = ip_rt_update_pmtu,
172 .redirect = ip_do_redirect,
173 .local_out = __ip_local_out,
174 .neigh_lookup = ipv4_neigh_lookup,
177 #define ECN_OR_COST(class) TC_PRIO_##class
179 const __u8 ip_tos2prio[16] = {
181 ECN_OR_COST(BESTEFFORT),
183 ECN_OR_COST(BESTEFFORT),
189 ECN_OR_COST(INTERACTIVE),
191 ECN_OR_COST(INTERACTIVE),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK),
194 TC_PRIO_INTERACTIVE_BULK,
195 ECN_OR_COST(INTERACTIVE_BULK)
197 EXPORT_SYMBOL(ip_tos2prio);
199 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
200 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
202 #ifdef CONFIG_PROC_FS
203 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
207 return SEQ_START_TOKEN;
210 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
216 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
220 static int rt_cache_seq_show(struct seq_file *seq, void *v)
222 if (v == SEQ_START_TOKEN)
223 seq_printf(seq, "%-127s\n",
224 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
225 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
230 static const struct seq_operations rt_cache_seq_ops = {
231 .start = rt_cache_seq_start,
232 .next = rt_cache_seq_next,
233 .stop = rt_cache_seq_stop,
234 .show = rt_cache_seq_show,
237 static int rt_cache_seq_open(struct inode *inode, struct file *file)
239 return seq_open(file, &rt_cache_seq_ops);
242 static const struct file_operations rt_cache_seq_fops = {
243 .owner = THIS_MODULE,
244 .open = rt_cache_seq_open,
247 .release = seq_release,
251 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
256 return SEQ_START_TOKEN;
258 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
259 if (!cpu_possible(cpu))
262 return &per_cpu(rt_cache_stat, cpu);
267 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
271 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
272 if (!cpu_possible(cpu))
275 return &per_cpu(rt_cache_stat, cpu);
281 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
286 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
288 struct rt_cache_stat *st = v;
290 if (v == SEQ_START_TOKEN) {
291 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");
295 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
296 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
297 dst_entries_get_slow(&ipv4_dst_ops),
320 static const struct seq_operations rt_cpu_seq_ops = {
321 .start = rt_cpu_seq_start,
322 .next = rt_cpu_seq_next,
323 .stop = rt_cpu_seq_stop,
324 .show = rt_cpu_seq_show,
328 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
330 return seq_open(file, &rt_cpu_seq_ops);
333 static const struct file_operations rt_cpu_seq_fops = {
334 .owner = THIS_MODULE,
335 .open = rt_cpu_seq_open,
338 .release = seq_release,
341 #ifdef CONFIG_IP_ROUTE_CLASSID
342 static int rt_acct_proc_show(struct seq_file *m, void *v)
344 struct ip_rt_acct *dst, *src;
347 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
351 for_each_possible_cpu(i) {
352 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
353 for (j = 0; j < 256; j++) {
354 dst[j].o_bytes += src[j].o_bytes;
355 dst[j].o_packets += src[j].o_packets;
356 dst[j].i_bytes += src[j].i_bytes;
357 dst[j].i_packets += src[j].i_packets;
361 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
366 static int rt_acct_proc_open(struct inode *inode, struct file *file)
368 return single_open(file, rt_acct_proc_show, NULL);
371 static const struct file_operations rt_acct_proc_fops = {
372 .owner = THIS_MODULE,
373 .open = rt_acct_proc_open,
376 .release = single_release,
380 static int __net_init ip_rt_do_proc_init(struct net *net)
382 struct proc_dir_entry *pde;
384 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
389 pde = proc_create("rt_cache", S_IRUGO,
390 net->proc_net_stat, &rt_cpu_seq_fops);
394 #ifdef CONFIG_IP_ROUTE_CLASSID
395 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
401 #ifdef CONFIG_IP_ROUTE_CLASSID
403 remove_proc_entry("rt_cache", net->proc_net_stat);
406 remove_proc_entry("rt_cache", net->proc_net);
411 static void __net_exit ip_rt_do_proc_exit(struct net *net)
413 remove_proc_entry("rt_cache", net->proc_net_stat);
414 remove_proc_entry("rt_cache", net->proc_net);
415 #ifdef CONFIG_IP_ROUTE_CLASSID
416 remove_proc_entry("rt_acct", net->proc_net);
420 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
421 .init = ip_rt_do_proc_init,
422 .exit = ip_rt_do_proc_exit,
425 static int __init ip_rt_proc_init(void)
427 return register_pernet_subsys(&ip_rt_proc_ops);
431 static inline int ip_rt_proc_init(void)
435 #endif /* CONFIG_PROC_FS */
437 static inline bool rt_is_expired(const struct rtable *rth)
439 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
442 void rt_cache_flush(struct net *net)
447 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
451 struct net_device *dev = dst->dev;
452 const __be32 *pkey = daddr;
453 const struct rtable *rt;
456 rt = (const struct rtable *) dst;
458 pkey = (const __be32 *) &rt->rt_gateway;
460 pkey = &ip_hdr(skb)->daddr;
462 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
465 return neigh_create(&arp_tbl, pkey, dev);
468 #define IP_IDENTS_SZ 2048u
469 struct ip_ident_bucket {
474 static struct ip_ident_bucket *ip_idents __read_mostly;
476 /* In order to protect privacy, we add a perturbation to identifiers
477 * if one generator is seldom used. This makes hard for an attacker
478 * to infer how many packets were sent between two points in time.
480 u32 ip_idents_reserve(u32 hash, int segs)
482 struct ip_ident_bucket *bucket = ip_idents + hash % IP_IDENTS_SZ;
483 u32 old = ACCESS_ONCE(bucket->stamp32);
484 u32 now = (u32)jiffies;
487 if (old != now && cmpxchg(&bucket->stamp32, old, now) == old) {
488 u64 x = prandom_u32();
491 delta = (u32)(x >> 32);
494 return atomic_add_return(segs + delta, &bucket->id) - segs;
496 EXPORT_SYMBOL(ip_idents_reserve);
498 void __ip_select_ident(struct iphdr *iph, int segs)
500 static u32 ip_idents_hashrnd __read_mostly;
501 static bool hashrnd_initialized = false;
504 if (unlikely(!hashrnd_initialized)) {
505 hashrnd_initialized = true;
506 get_random_bytes(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
509 hash = jhash_3words((__force u32)iph->daddr,
510 (__force u32)iph->saddr,
513 id = ip_idents_reserve(hash, segs);
516 EXPORT_SYMBOL(__ip_select_ident);
518 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
519 const struct iphdr *iph,
521 u8 prot, u32 mark, int flow_flags)
524 const struct inet_sock *inet = inet_sk(sk);
526 oif = sk->sk_bound_dev_if;
528 tos = RT_CONN_FLAGS(sk);
529 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
531 flowi4_init_output(fl4, oif, mark, tos,
532 RT_SCOPE_UNIVERSE, prot,
534 iph->daddr, iph->saddr, 0, 0);
537 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
538 const struct sock *sk)
540 const struct iphdr *iph = ip_hdr(skb);
541 int oif = skb->dev->ifindex;
542 u8 tos = RT_TOS(iph->tos);
543 u8 prot = iph->protocol;
544 u32 mark = skb->mark;
546 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
549 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
551 const struct inet_sock *inet = inet_sk(sk);
552 const struct ip_options_rcu *inet_opt;
553 __be32 daddr = inet->inet_daddr;
556 inet_opt = rcu_dereference(inet->inet_opt);
557 if (inet_opt && inet_opt->opt.srr)
558 daddr = inet_opt->opt.faddr;
559 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
560 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
561 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
562 inet_sk_flowi_flags(sk),
563 daddr, inet->inet_saddr, 0, 0);
567 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
568 const struct sk_buff *skb)
571 build_skb_flow_key(fl4, skb, sk);
573 build_sk_flow_key(fl4, sk);
576 static inline void rt_free(struct rtable *rt)
578 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
581 static DEFINE_SPINLOCK(fnhe_lock);
583 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
585 struct fib_nh_exception *fnhe, *oldest;
588 oldest = rcu_dereference(hash->chain);
589 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
590 fnhe = rcu_dereference(fnhe->fnhe_next)) {
591 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
594 orig = rcu_dereference(oldest->fnhe_rth);
596 RCU_INIT_POINTER(oldest->fnhe_rth, NULL);
602 static inline u32 fnhe_hashfun(__be32 daddr)
606 hval = (__force u32) daddr;
607 hval ^= (hval >> 11) ^ (hval >> 22);
609 return hval & (FNHE_HASH_SIZE - 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;
618 u32 hval = fnhe_hashfun(daddr);
620 spin_lock_bh(&fnhe_lock);
622 hash = nh->nh_exceptions;
624 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
627 nh->nh_exceptions = hash;
633 for (fnhe = rcu_dereference(hash->chain); fnhe;
634 fnhe = rcu_dereference(fnhe->fnhe_next)) {
635 if (fnhe->fnhe_daddr == daddr)
644 fnhe->fnhe_pmtu = pmtu;
645 fnhe->fnhe_expires = expires;
648 if (depth > FNHE_RECLAIM_DEPTH)
649 fnhe = fnhe_oldest(hash);
651 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
655 fnhe->fnhe_next = hash->chain;
656 rcu_assign_pointer(hash->chain, fnhe);
658 fnhe->fnhe_daddr = daddr;
660 fnhe->fnhe_pmtu = pmtu;
661 fnhe->fnhe_expires = expires;
664 fnhe->fnhe_stamp = jiffies;
667 spin_unlock_bh(&fnhe_lock);
671 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
674 __be32 new_gw = icmp_hdr(skb)->un.gateway;
675 __be32 old_gw = ip_hdr(skb)->saddr;
676 struct net_device *dev = skb->dev;
677 struct in_device *in_dev;
678 struct fib_result res;
682 switch (icmp_hdr(skb)->code & 7) {
684 case ICMP_REDIR_NETTOS:
685 case ICMP_REDIR_HOST:
686 case ICMP_REDIR_HOSTTOS:
693 if (rt->rt_gateway != old_gw)
696 in_dev = __in_dev_get_rcu(dev);
701 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
702 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
703 ipv4_is_zeronet(new_gw))
704 goto reject_redirect;
706 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
707 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
708 goto reject_redirect;
709 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
710 goto reject_redirect;
712 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
713 goto reject_redirect;
716 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
718 if (!(n->nud_state & NUD_VALID)) {
719 neigh_event_send(n, NULL);
721 if (fib_lookup(net, fl4, &res) == 0) {
722 struct fib_nh *nh = &FIB_RES_NH(res);
724 update_or_create_fnhe(nh, fl4->daddr, new_gw,
728 rt->dst.obsolete = DST_OBSOLETE_KILL;
729 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
736 #ifdef CONFIG_IP_ROUTE_VERBOSE
737 if (IN_DEV_LOG_MARTIANS(in_dev)) {
738 const struct iphdr *iph = (const struct iphdr *) skb->data;
739 __be32 daddr = iph->daddr;
740 __be32 saddr = iph->saddr;
742 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
743 " Advised path = %pI4 -> %pI4\n",
744 &old_gw, dev->name, &new_gw,
751 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
755 const struct iphdr *iph = (const struct iphdr *) skb->data;
756 int oif = skb->dev->ifindex;
757 u8 tos = RT_TOS(iph->tos);
758 u8 prot = iph->protocol;
759 u32 mark = skb->mark;
761 rt = (struct rtable *) dst;
763 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
764 __ip_do_redirect(rt, skb, &fl4, true);
767 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
769 struct rtable *rt = (struct rtable *)dst;
770 struct dst_entry *ret = dst;
773 if (dst->obsolete > 0) {
776 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
787 * 1. The first ip_rt_redirect_number redirects are sent
788 * with exponential backoff, then we stop sending them at all,
789 * assuming that the host ignores our redirects.
790 * 2. If we did not see packets requiring redirects
791 * during ip_rt_redirect_silence, we assume that the host
792 * forgot redirected route and start to send redirects again.
794 * This algorithm is much cheaper and more intelligent than dumb load limiting
797 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
798 * and "frag. need" (breaks PMTU discovery) in icmp.c.
801 void ip_rt_send_redirect(struct sk_buff *skb)
803 struct rtable *rt = skb_rtable(skb);
804 struct in_device *in_dev;
805 struct inet_peer *peer;
810 in_dev = __in_dev_get_rcu(rt->dst.dev);
811 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
815 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
818 net = dev_net(rt->dst.dev);
819 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
821 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
822 rt_nexthop(rt, ip_hdr(skb)->daddr));
826 /* No redirected packets during ip_rt_redirect_silence;
827 * reset the algorithm.
829 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
830 peer->rate_tokens = 0;
832 /* Too many ignored redirects; do not send anything
833 * set dst.rate_last to the last seen redirected packet.
835 if (peer->rate_tokens >= ip_rt_redirect_number) {
836 peer->rate_last = jiffies;
840 /* Check for load limit; set rate_last to the latest sent
843 if (peer->rate_tokens == 0 ||
846 (ip_rt_redirect_load << peer->rate_tokens)))) {
847 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
849 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
850 peer->rate_last = jiffies;
852 #ifdef CONFIG_IP_ROUTE_VERBOSE
854 peer->rate_tokens == ip_rt_redirect_number)
855 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
856 &ip_hdr(skb)->saddr, inet_iif(skb),
857 &ip_hdr(skb)->daddr, &gw);
864 static int ip_error(struct sk_buff *skb)
866 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
867 struct rtable *rt = skb_rtable(skb);
868 struct inet_peer *peer;
874 net = dev_net(rt->dst.dev);
875 if (!IN_DEV_FORWARD(in_dev)) {
876 switch (rt->dst.error) {
878 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
882 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
888 switch (rt->dst.error) {
893 code = ICMP_HOST_UNREACH;
896 code = ICMP_NET_UNREACH;
897 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
900 code = ICMP_PKT_FILTERED;
904 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
909 peer->rate_tokens += now - peer->rate_last;
910 if (peer->rate_tokens > ip_rt_error_burst)
911 peer->rate_tokens = ip_rt_error_burst;
912 peer->rate_last = now;
913 if (peer->rate_tokens >= ip_rt_error_cost)
914 peer->rate_tokens -= ip_rt_error_cost;
920 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
926 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
928 struct dst_entry *dst = &rt->dst;
929 struct fib_result res;
931 if (dst_metric_locked(dst, RTAX_MTU))
934 if (dst->dev->mtu < mtu)
937 if (mtu < ip_rt_min_pmtu)
938 mtu = ip_rt_min_pmtu;
941 dst->obsolete = DST_OBSOLETE_KILL;
944 dst->expires = max(1UL, jiffies + ip_rt_mtu_expires);
948 if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
949 struct fib_nh *nh = &FIB_RES_NH(res);
951 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
952 jiffies + ip_rt_mtu_expires);
957 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
958 struct sk_buff *skb, u32 mtu)
960 struct rtable *rt = (struct rtable *) dst;
963 ip_rt_build_flow_key(&fl4, sk, skb);
964 __ip_rt_update_pmtu(rt, &fl4, mtu);
967 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
968 int oif, u32 mark, u8 protocol, int flow_flags)
970 const struct iphdr *iph = (const struct iphdr *) skb->data;
974 __build_flow_key(&fl4, NULL, iph, oif,
975 RT_TOS(iph->tos), protocol, mark, flow_flags);
976 rt = __ip_route_output_key(net, &fl4);
978 __ip_rt_update_pmtu(rt, &fl4, mtu);
982 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
984 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
986 const struct iphdr *iph = (const struct iphdr *) skb->data;
990 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
991 rt = __ip_route_output_key(sock_net(sk), &fl4);
993 __ip_rt_update_pmtu(rt, &fl4, mtu);
998 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1000 const struct iphdr *iph = (const struct iphdr *) skb->data;
1003 struct dst_entry *odst = NULL;
1007 odst = sk_dst_get(sk);
1009 if (sock_owned_by_user(sk) || !odst) {
1010 __ipv4_sk_update_pmtu(skb, sk, mtu);
1014 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1016 rt = (struct rtable *)odst;
1017 if (odst->obsolete && odst->ops->check(odst, 0) == NULL) {
1018 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1025 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1027 if (!dst_check(&rt->dst, 0)) {
1029 dst_release(&rt->dst);
1031 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1039 sk_dst_set(sk, &rt->dst);
1045 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1047 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1048 int oif, u32 mark, u8 protocol, int flow_flags)
1050 const struct iphdr *iph = (const struct iphdr *) skb->data;
1054 __build_flow_key(&fl4, NULL, iph, oif,
1055 RT_TOS(iph->tos), protocol, mark, flow_flags);
1056 rt = __ip_route_output_key(net, &fl4);
1058 __ip_do_redirect(rt, skb, &fl4, false);
1062 EXPORT_SYMBOL_GPL(ipv4_redirect);
1064 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1066 const struct iphdr *iph = (const struct iphdr *) skb->data;
1070 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1071 rt = __ip_route_output_key(sock_net(sk), &fl4);
1073 __ip_do_redirect(rt, skb, &fl4, false);
1077 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1079 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1081 struct rtable *rt = (struct rtable *) dst;
1083 /* All IPV4 dsts are created with ->obsolete set to the value
1084 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1085 * into this function always.
1087 * When a PMTU/redirect information update invalidates a
1088 * route, this is indicated by setting obsolete to
1089 * DST_OBSOLETE_KILL.
1091 if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1096 static void ipv4_link_failure(struct sk_buff *skb)
1100 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1102 rt = skb_rtable(skb);
1104 dst_set_expires(&rt->dst, 0);
1107 static int ip_rt_bug(struct sk_buff *skb)
1109 pr_debug("%s: %pI4 -> %pI4, %s\n",
1110 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1111 skb->dev ? skb->dev->name : "?");
1118 We do not cache source address of outgoing interface,
1119 because it is used only by IP RR, TS and SRR options,
1120 so that it out of fast path.
1122 BTW remember: "addr" is allowed to be not aligned
1126 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1130 if (rt_is_output_route(rt))
1131 src = ip_hdr(skb)->saddr;
1133 struct fib_result res;
1139 memset(&fl4, 0, sizeof(fl4));
1140 fl4.daddr = iph->daddr;
1141 fl4.saddr = iph->saddr;
1142 fl4.flowi4_tos = RT_TOS(iph->tos);
1143 fl4.flowi4_oif = rt->dst.dev->ifindex;
1144 fl4.flowi4_iif = skb->dev->ifindex;
1145 fl4.flowi4_mark = skb->mark;
1148 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1149 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1151 src = inet_select_addr(rt->dst.dev,
1152 rt_nexthop(rt, iph->daddr),
1156 memcpy(addr, &src, 4);
1159 #ifdef CONFIG_IP_ROUTE_CLASSID
1160 static void set_class_tag(struct rtable *rt, u32 tag)
1162 if (!(rt->dst.tclassid & 0xFFFF))
1163 rt->dst.tclassid |= tag & 0xFFFF;
1164 if (!(rt->dst.tclassid & 0xFFFF0000))
1165 rt->dst.tclassid |= tag & 0xFFFF0000;
1169 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1171 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1174 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1176 if (advmss > 65535 - 40)
1177 advmss = 65535 - 40;
1182 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1184 const struct rtable *rt = (const struct rtable *) dst;
1185 unsigned int mtu = rt->rt_pmtu;
1187 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1188 mtu = dst_metric_raw(dst, RTAX_MTU);
1193 mtu = dst->dev->mtu;
1195 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1196 if (rt->rt_uses_gateway && mtu > 576)
1200 if (mtu > IP_MAX_MTU)
1206 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1208 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1209 struct fib_nh_exception *fnhe;
1215 hval = fnhe_hashfun(daddr);
1217 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1218 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1219 if (fnhe->fnhe_daddr == daddr)
1225 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1230 spin_lock_bh(&fnhe_lock);
1232 if (daddr == fnhe->fnhe_daddr) {
1233 struct rtable *orig = rcu_dereference(fnhe->fnhe_rth);
1234 if (orig && rt_is_expired(orig)) {
1236 fnhe->fnhe_pmtu = 0;
1237 fnhe->fnhe_expires = 0;
1239 if (fnhe->fnhe_pmtu) {
1240 unsigned long expires = fnhe->fnhe_expires;
1241 unsigned long diff = expires - jiffies;
1243 if (time_before(jiffies, expires)) {
1244 rt->rt_pmtu = fnhe->fnhe_pmtu;
1245 dst_set_expires(&rt->dst, diff);
1248 if (fnhe->fnhe_gw) {
1249 rt->rt_flags |= RTCF_REDIRECTED;
1250 rt->rt_gateway = fnhe->fnhe_gw;
1251 rt->rt_uses_gateway = 1;
1252 } else if (!rt->rt_gateway)
1253 rt->rt_gateway = daddr;
1255 rcu_assign_pointer(fnhe->fnhe_rth, rt);
1259 fnhe->fnhe_stamp = jiffies;
1262 spin_unlock_bh(&fnhe_lock);
1267 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1269 struct rtable *orig, *prev, **p;
1272 if (rt_is_input_route(rt)) {
1273 p = (struct rtable **)&nh->nh_rth_input;
1275 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1279 prev = cmpxchg(p, orig, rt);
1289 static DEFINE_SPINLOCK(rt_uncached_lock);
1290 static LIST_HEAD(rt_uncached_list);
1292 static void rt_add_uncached_list(struct rtable *rt)
1294 spin_lock_bh(&rt_uncached_lock);
1295 list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1296 spin_unlock_bh(&rt_uncached_lock);
1299 static void ipv4_dst_destroy(struct dst_entry *dst)
1301 struct rtable *rt = (struct rtable *) dst;
1303 if (!list_empty(&rt->rt_uncached)) {
1304 spin_lock_bh(&rt_uncached_lock);
1305 list_del(&rt->rt_uncached);
1306 spin_unlock_bh(&rt_uncached_lock);
1310 void rt_flush_dev(struct net_device *dev)
1312 if (!list_empty(&rt_uncached_list)) {
1313 struct net *net = dev_net(dev);
1316 spin_lock_bh(&rt_uncached_lock);
1317 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1318 if (rt->dst.dev != dev)
1320 rt->dst.dev = net->loopback_dev;
1321 dev_hold(rt->dst.dev);
1324 spin_unlock_bh(&rt_uncached_lock);
1328 static bool rt_cache_valid(const struct rtable *rt)
1331 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1335 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1336 const struct fib_result *res,
1337 struct fib_nh_exception *fnhe,
1338 struct fib_info *fi, u16 type, u32 itag)
1340 bool cached = false;
1343 struct fib_nh *nh = &FIB_RES_NH(*res);
1345 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1346 rt->rt_gateway = nh->nh_gw;
1347 rt->rt_uses_gateway = 1;
1349 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1350 #ifdef CONFIG_IP_ROUTE_CLASSID
1351 rt->dst.tclassid = nh->nh_tclassid;
1354 cached = rt_bind_exception(rt, fnhe, daddr);
1355 else if (!(rt->dst.flags & DST_NOCACHE))
1356 cached = rt_cache_route(nh, rt);
1357 if (unlikely(!cached)) {
1358 /* Routes we intend to cache in nexthop exception or
1359 * FIB nexthop have the DST_NOCACHE bit clear.
1360 * However, if we are unsuccessful at storing this
1361 * route into the cache we really need to set it.
1363 rt->dst.flags |= DST_NOCACHE;
1364 if (!rt->rt_gateway)
1365 rt->rt_gateway = daddr;
1366 rt_add_uncached_list(rt);
1369 rt_add_uncached_list(rt);
1371 #ifdef CONFIG_IP_ROUTE_CLASSID
1372 #ifdef CONFIG_IP_MULTIPLE_TABLES
1373 set_class_tag(rt, res->tclassid);
1375 set_class_tag(rt, itag);
1379 static struct rtable *rt_dst_alloc(struct net_device *dev,
1380 bool nopolicy, bool noxfrm, bool will_cache)
1382 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1383 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1384 (nopolicy ? DST_NOPOLICY : 0) |
1385 (noxfrm ? DST_NOXFRM : 0));
1388 /* called in rcu_read_lock() section */
1389 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1390 u8 tos, struct net_device *dev, int our)
1393 struct in_device *in_dev = __in_dev_get_rcu(dev);
1397 /* Primary sanity checks. */
1402 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1403 skb->protocol != htons(ETH_P_IP))
1406 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1407 if (ipv4_is_loopback(saddr))
1410 if (ipv4_is_zeronet(saddr)) {
1411 if (!ipv4_is_local_multicast(daddr))
1414 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1419 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1420 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1424 #ifdef CONFIG_IP_ROUTE_CLASSID
1425 rth->dst.tclassid = itag;
1427 rth->dst.output = ip_rt_bug;
1429 rth->rt_genid = rt_genid(dev_net(dev));
1430 rth->rt_flags = RTCF_MULTICAST;
1431 rth->rt_type = RTN_MULTICAST;
1432 rth->rt_is_input= 1;
1435 rth->rt_gateway = 0;
1436 rth->rt_uses_gateway = 0;
1437 INIT_LIST_HEAD(&rth->rt_uncached);
1439 rth->dst.input= ip_local_deliver;
1440 rth->rt_flags |= RTCF_LOCAL;
1443 #ifdef CONFIG_IP_MROUTE
1444 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1445 rth->dst.input = ip_mr_input;
1447 RT_CACHE_STAT_INC(in_slow_mc);
1449 skb_dst_set(skb, &rth->dst);
1461 static void ip_handle_martian_source(struct net_device *dev,
1462 struct in_device *in_dev,
1463 struct sk_buff *skb,
1467 RT_CACHE_STAT_INC(in_martian_src);
1468 #ifdef CONFIG_IP_ROUTE_VERBOSE
1469 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1471 * RFC1812 recommendation, if source is martian,
1472 * the only hint is MAC header.
1474 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1475 &daddr, &saddr, dev->name);
1476 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1477 print_hex_dump(KERN_WARNING, "ll header: ",
1478 DUMP_PREFIX_OFFSET, 16, 1,
1479 skb_mac_header(skb),
1480 dev->hard_header_len, true);
1486 /* called in rcu_read_lock() section */
1487 static int __mkroute_input(struct sk_buff *skb,
1488 const struct fib_result *res,
1489 struct in_device *in_dev,
1490 __be32 daddr, __be32 saddr, u32 tos)
1494 struct in_device *out_dev;
1495 unsigned int flags = 0;
1499 /* get a working reference to the output device */
1500 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1501 if (out_dev == NULL) {
1502 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1506 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1507 in_dev->dev, in_dev, &itag);
1509 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1515 do_cache = res->fi && !itag;
1516 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1517 (IN_DEV_SHARED_MEDIA(out_dev) ||
1518 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1519 flags |= RTCF_DOREDIRECT;
1523 if (skb->protocol != htons(ETH_P_IP)) {
1524 /* Not IP (i.e. ARP). Do not create route, if it is
1525 * invalid for proxy arp. DNAT routes are always valid.
1527 * Proxy arp feature have been extended to allow, ARP
1528 * replies back to the same interface, to support
1529 * Private VLAN switch technologies. See arp.c.
1531 if (out_dev == in_dev &&
1532 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1539 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1540 if (rt_cache_valid(rth)) {
1541 skb_dst_set_noref(skb, &rth->dst);
1546 rth = rt_dst_alloc(out_dev->dev,
1547 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1548 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1554 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1555 rth->rt_flags = flags;
1556 rth->rt_type = res->type;
1557 rth->rt_is_input = 1;
1560 rth->rt_gateway = 0;
1561 rth->rt_uses_gateway = 0;
1562 INIT_LIST_HEAD(&rth->rt_uncached);
1563 RT_CACHE_STAT_INC(in_slow_tot);
1565 rth->dst.input = ip_forward;
1566 rth->dst.output = ip_output;
1568 rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1569 skb_dst_set(skb, &rth->dst);
1576 static int ip_mkroute_input(struct sk_buff *skb,
1577 struct fib_result *res,
1578 const struct flowi4 *fl4,
1579 struct in_device *in_dev,
1580 __be32 daddr, __be32 saddr, u32 tos)
1582 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1583 if (res->fi && res->fi->fib_nhs > 1)
1584 fib_select_multipath(res);
1587 /* create a routing cache entry */
1588 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1592 * NOTE. We drop all the packets that has local source
1593 * addresses, because every properly looped back packet
1594 * must have correct destination already attached by output routine.
1596 * Such approach solves two big problems:
1597 * 1. Not simplex devices are handled properly.
1598 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1599 * called with rcu_read_lock()
1602 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1603 u8 tos, struct net_device *dev)
1605 struct fib_result res;
1606 struct in_device *in_dev = __in_dev_get_rcu(dev);
1608 unsigned int flags = 0;
1612 struct net *net = dev_net(dev);
1615 /* IP on this device is disabled. */
1620 /* Check for the most weird martians, which can be not detected
1624 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1625 goto martian_source;
1628 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1631 /* Accept zero addresses only to limited broadcast;
1632 * I even do not know to fix it or not. Waiting for complains :-)
1634 if (ipv4_is_zeronet(saddr))
1635 goto martian_source;
1637 if (ipv4_is_zeronet(daddr))
1638 goto martian_destination;
1640 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1641 * and call it once if daddr or/and saddr are loopback addresses
1643 if (ipv4_is_loopback(daddr)) {
1644 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1645 goto martian_destination;
1646 } else if (ipv4_is_loopback(saddr)) {
1647 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1648 goto martian_source;
1652 * Now we are ready to route packet.
1655 fl4.flowi4_iif = dev->ifindex;
1656 fl4.flowi4_mark = skb->mark;
1657 fl4.flowi4_tos = tos;
1658 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1661 err = fib_lookup(net, &fl4, &res);
1665 if (res.type == RTN_BROADCAST)
1668 if (res.type == RTN_LOCAL) {
1669 err = fib_validate_source(skb, saddr, daddr, tos,
1671 dev, in_dev, &itag);
1673 goto martian_source_keep_err;
1677 if (!IN_DEV_FORWARD(in_dev))
1679 if (res.type != RTN_UNICAST)
1680 goto martian_destination;
1682 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1686 if (skb->protocol != htons(ETH_P_IP))
1689 if (!ipv4_is_zeronet(saddr)) {
1690 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1693 goto martian_source_keep_err;
1695 flags |= RTCF_BROADCAST;
1696 res.type = RTN_BROADCAST;
1697 RT_CACHE_STAT_INC(in_brd);
1703 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1704 if (rt_cache_valid(rth)) {
1705 skb_dst_set_noref(skb, &rth->dst);
1713 rth = rt_dst_alloc(net->loopback_dev,
1714 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1718 rth->dst.input= ip_local_deliver;
1719 rth->dst.output= ip_rt_bug;
1720 #ifdef CONFIG_IP_ROUTE_CLASSID
1721 rth->dst.tclassid = itag;
1724 rth->rt_genid = rt_genid(net);
1725 rth->rt_flags = flags|RTCF_LOCAL;
1726 rth->rt_type = res.type;
1727 rth->rt_is_input = 1;
1730 rth->rt_gateway = 0;
1731 rth->rt_uses_gateway = 0;
1732 INIT_LIST_HEAD(&rth->rt_uncached);
1733 RT_CACHE_STAT_INC(in_slow_tot);
1734 if (res.type == RTN_UNREACHABLE) {
1735 rth->dst.input= ip_error;
1736 rth->dst.error= -err;
1737 rth->rt_flags &= ~RTCF_LOCAL;
1740 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1741 rth->dst.flags |= DST_NOCACHE;
1742 rt_add_uncached_list(rth);
1745 skb_dst_set(skb, &rth->dst);
1750 RT_CACHE_STAT_INC(in_no_route);
1751 res.type = RTN_UNREACHABLE;
1757 * Do not cache martian addresses: they should be logged (RFC1812)
1759 martian_destination:
1760 RT_CACHE_STAT_INC(in_martian_dst);
1761 #ifdef CONFIG_IP_ROUTE_VERBOSE
1762 if (IN_DEV_LOG_MARTIANS(in_dev))
1763 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1764 &daddr, &saddr, dev->name);
1777 martian_source_keep_err:
1778 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1782 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1783 u8 tos, struct net_device *dev)
1789 /* Multicast recognition logic is moved from route cache to here.
1790 The problem was that too many Ethernet cards have broken/missing
1791 hardware multicast filters :-( As result the host on multicasting
1792 network acquires a lot of useless route cache entries, sort of
1793 SDR messages from all the world. Now we try to get rid of them.
1794 Really, provided software IP multicast filter is organized
1795 reasonably (at least, hashed), it does not result in a slowdown
1796 comparing with route cache reject entries.
1797 Note, that multicast routers are not affected, because
1798 route cache entry is created eventually.
1800 if (ipv4_is_multicast(daddr)) {
1801 struct in_device *in_dev = __in_dev_get_rcu(dev);
1804 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1805 ip_hdr(skb)->protocol);
1807 #ifdef CONFIG_IP_MROUTE
1809 (!ipv4_is_local_multicast(daddr) &&
1810 IN_DEV_MFORWARD(in_dev))
1813 int res = ip_route_input_mc(skb, daddr, saddr,
1822 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1826 EXPORT_SYMBOL(ip_route_input_noref);
1828 /* called with rcu_read_lock() */
1829 static struct rtable *__mkroute_output(const struct fib_result *res,
1830 const struct flowi4 *fl4, int orig_oif,
1831 struct net_device *dev_out,
1834 struct fib_info *fi = res->fi;
1835 struct fib_nh_exception *fnhe;
1836 struct in_device *in_dev;
1837 u16 type = res->type;
1841 in_dev = __in_dev_get_rcu(dev_out);
1843 return ERR_PTR(-EINVAL);
1845 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1846 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1847 return ERR_PTR(-EINVAL);
1849 if (ipv4_is_lbcast(fl4->daddr))
1850 type = RTN_BROADCAST;
1851 else if (ipv4_is_multicast(fl4->daddr))
1852 type = RTN_MULTICAST;
1853 else if (ipv4_is_zeronet(fl4->daddr))
1854 return ERR_PTR(-EINVAL);
1856 if (dev_out->flags & IFF_LOOPBACK)
1857 flags |= RTCF_LOCAL;
1860 if (type == RTN_BROADCAST) {
1861 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1863 } else if (type == RTN_MULTICAST) {
1864 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1865 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1867 flags &= ~RTCF_LOCAL;
1870 /* If multicast route do not exist use
1871 * default one, but do not gateway in this case.
1874 if (fi && res->prefixlen < 4)
1879 do_cache &= fi != NULL;
1881 struct rtable __rcu **prth;
1882 struct fib_nh *nh = &FIB_RES_NH(*res);
1884 fnhe = find_exception(nh, fl4->daddr);
1886 prth = &fnhe->fnhe_rth;
1888 if (unlikely(fl4->flowi4_flags &
1889 FLOWI_FLAG_KNOWN_NH &&
1891 nh->nh_scope == RT_SCOPE_LINK))) {
1895 prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1897 rth = rcu_dereference(*prth);
1898 if (rt_cache_valid(rth)) {
1899 dst_hold(&rth->dst);
1905 rth = rt_dst_alloc(dev_out,
1906 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1907 IN_DEV_CONF_GET(in_dev, NOXFRM),
1910 return ERR_PTR(-ENOBUFS);
1912 rth->dst.output = ip_output;
1914 rth->rt_genid = rt_genid(dev_net(dev_out));
1915 rth->rt_flags = flags;
1916 rth->rt_type = type;
1917 rth->rt_is_input = 0;
1918 rth->rt_iif = orig_oif ? : 0;
1920 rth->rt_gateway = 0;
1921 rth->rt_uses_gateway = 0;
1922 INIT_LIST_HEAD(&rth->rt_uncached);
1924 RT_CACHE_STAT_INC(out_slow_tot);
1926 if (flags & RTCF_LOCAL)
1927 rth->dst.input = ip_local_deliver;
1928 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1929 if (flags & RTCF_LOCAL &&
1930 !(dev_out->flags & IFF_LOOPBACK)) {
1931 rth->dst.output = ip_mc_output;
1932 RT_CACHE_STAT_INC(out_slow_mc);
1934 #ifdef CONFIG_IP_MROUTE
1935 if (type == RTN_MULTICAST) {
1936 if (IN_DEV_MFORWARD(in_dev) &&
1937 !ipv4_is_local_multicast(fl4->daddr)) {
1938 rth->dst.input = ip_mr_input;
1939 rth->dst.output = ip_mc_output;
1945 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1951 * Major route resolver routine.
1954 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1956 struct net_device *dev_out = NULL;
1957 __u8 tos = RT_FL_TOS(fl4);
1958 unsigned int flags = 0;
1959 struct fib_result res;
1967 orig_oif = fl4->flowi4_oif;
1969 fl4->flowi4_iif = LOOPBACK_IFINDEX;
1970 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1971 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1972 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1976 rth = ERR_PTR(-EINVAL);
1977 if (ipv4_is_multicast(fl4->saddr) ||
1978 ipv4_is_lbcast(fl4->saddr) ||
1979 ipv4_is_zeronet(fl4->saddr))
1982 /* I removed check for oif == dev_out->oif here.
1983 It was wrong for two reasons:
1984 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1985 is assigned to multiple interfaces.
1986 2. Moreover, we are allowed to send packets with saddr
1987 of another iface. --ANK
1990 if (fl4->flowi4_oif == 0 &&
1991 (ipv4_is_multicast(fl4->daddr) ||
1992 ipv4_is_lbcast(fl4->daddr))) {
1993 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1994 dev_out = __ip_dev_find(net, fl4->saddr, false);
1995 if (dev_out == NULL)
1998 /* Special hack: user can direct multicasts
1999 and limited broadcast via necessary interface
2000 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2001 This hack is not just for fun, it allows
2002 vic,vat and friends to work.
2003 They bind socket to loopback, set ttl to zero
2004 and expect that it will work.
2005 From the viewpoint of routing cache they are broken,
2006 because we are not allowed to build multicast path
2007 with loopback source addr (look, routing cache
2008 cannot know, that ttl is zero, so that packet
2009 will not leave this host and route is valid).
2010 Luckily, this hack is good workaround.
2013 fl4->flowi4_oif = dev_out->ifindex;
2017 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2018 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2019 if (!__ip_dev_find(net, fl4->saddr, false))
2025 if (fl4->flowi4_oif) {
2026 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2027 rth = ERR_PTR(-ENODEV);
2028 if (dev_out == NULL)
2031 /* RACE: Check return value of inet_select_addr instead. */
2032 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2033 rth = ERR_PTR(-ENETUNREACH);
2036 if (ipv4_is_local_multicast(fl4->daddr) ||
2037 ipv4_is_lbcast(fl4->daddr)) {
2039 fl4->saddr = inet_select_addr(dev_out, 0,
2044 if (ipv4_is_multicast(fl4->daddr))
2045 fl4->saddr = inet_select_addr(dev_out, 0,
2047 else if (!fl4->daddr)
2048 fl4->saddr = inet_select_addr(dev_out, 0,
2054 fl4->daddr = fl4->saddr;
2056 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2057 dev_out = net->loopback_dev;
2058 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2059 res.type = RTN_LOCAL;
2060 flags |= RTCF_LOCAL;
2064 if (fib_lookup(net, fl4, &res)) {
2067 if (fl4->flowi4_oif) {
2068 /* Apparently, routing tables are wrong. Assume,
2069 that the destination is on link.
2072 Because we are allowed to send to iface
2073 even if it has NO routes and NO assigned
2074 addresses. When oif is specified, routing
2075 tables are looked up with only one purpose:
2076 to catch if destination is gatewayed, rather than
2077 direct. Moreover, if MSG_DONTROUTE is set,
2078 we send packet, ignoring both routing tables
2079 and ifaddr state. --ANK
2082 We could make it even if oif is unknown,
2083 likely IPv6, but we do not.
2086 if (fl4->saddr == 0)
2087 fl4->saddr = inet_select_addr(dev_out, 0,
2089 res.type = RTN_UNICAST;
2092 rth = ERR_PTR(-ENETUNREACH);
2096 if (res.type == RTN_LOCAL) {
2098 if (res.fi->fib_prefsrc)
2099 fl4->saddr = res.fi->fib_prefsrc;
2101 fl4->saddr = fl4->daddr;
2103 dev_out = net->loopback_dev;
2104 fl4->flowi4_oif = dev_out->ifindex;
2105 flags |= RTCF_LOCAL;
2109 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2110 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2111 fib_select_multipath(&res);
2114 if (!res.prefixlen &&
2115 res.table->tb_num_default > 1 &&
2116 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2117 fib_select_default(&res);
2120 fl4->saddr = FIB_RES_PREFSRC(net, res);
2122 dev_out = FIB_RES_DEV(res);
2123 fl4->flowi4_oif = dev_out->ifindex;
2127 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2133 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2135 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2140 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2142 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2144 return mtu ? : dst->dev->mtu;
2147 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2148 struct sk_buff *skb, u32 mtu)
2152 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2153 struct sk_buff *skb)
2157 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2163 static struct dst_ops ipv4_dst_blackhole_ops = {
2165 .protocol = cpu_to_be16(ETH_P_IP),
2166 .check = ipv4_blackhole_dst_check,
2167 .mtu = ipv4_blackhole_mtu,
2168 .default_advmss = ipv4_default_advmss,
2169 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2170 .redirect = ipv4_rt_blackhole_redirect,
2171 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2172 .neigh_lookup = ipv4_neigh_lookup,
2175 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2177 struct rtable *ort = (struct rtable *) dst_orig;
2180 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2182 struct dst_entry *new = &rt->dst;
2185 new->input = dst_discard;
2186 new->output = dst_discard;
2188 new->dev = ort->dst.dev;
2192 rt->rt_is_input = ort->rt_is_input;
2193 rt->rt_iif = ort->rt_iif;
2194 rt->rt_pmtu = ort->rt_pmtu;
2196 rt->rt_genid = rt_genid(net);
2197 rt->rt_flags = ort->rt_flags;
2198 rt->rt_type = ort->rt_type;
2199 rt->rt_gateway = ort->rt_gateway;
2200 rt->rt_uses_gateway = ort->rt_uses_gateway;
2202 INIT_LIST_HEAD(&rt->rt_uncached);
2207 dst_release(dst_orig);
2209 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2212 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2215 struct rtable *rt = __ip_route_output_key(net, flp4);
2220 if (flp4->flowi4_proto)
2221 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2222 flowi4_to_flowi(flp4),
2227 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2229 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2230 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2231 u32 seq, int event, int nowait, unsigned int flags)
2233 struct rtable *rt = skb_rtable(skb);
2235 struct nlmsghdr *nlh;
2236 unsigned long expires = 0;
2238 u32 metrics[RTAX_MAX];
2240 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2244 r = nlmsg_data(nlh);
2245 r->rtm_family = AF_INET;
2246 r->rtm_dst_len = 32;
2248 r->rtm_tos = fl4->flowi4_tos;
2249 r->rtm_table = RT_TABLE_MAIN;
2250 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2251 goto nla_put_failure;
2252 r->rtm_type = rt->rt_type;
2253 r->rtm_scope = RT_SCOPE_UNIVERSE;
2254 r->rtm_protocol = RTPROT_UNSPEC;
2255 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2256 if (rt->rt_flags & RTCF_NOTIFY)
2257 r->rtm_flags |= RTM_F_NOTIFY;
2259 if (nla_put_be32(skb, RTA_DST, dst))
2260 goto nla_put_failure;
2262 r->rtm_src_len = 32;
2263 if (nla_put_be32(skb, RTA_SRC, src))
2264 goto nla_put_failure;
2267 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2268 goto nla_put_failure;
2269 #ifdef CONFIG_IP_ROUTE_CLASSID
2270 if (rt->dst.tclassid &&
2271 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2272 goto nla_put_failure;
2274 if (!rt_is_input_route(rt) &&
2275 fl4->saddr != src) {
2276 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2277 goto nla_put_failure;
2279 if (rt->rt_uses_gateway &&
2280 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2281 goto nla_put_failure;
2283 expires = rt->dst.expires;
2285 unsigned long now = jiffies;
2287 if (time_before(now, expires))
2293 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2294 if (rt->rt_pmtu && expires)
2295 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2296 if (rtnetlink_put_metrics(skb, metrics) < 0)
2297 goto nla_put_failure;
2299 if (fl4->flowi4_mark &&
2300 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2301 goto nla_put_failure;
2303 error = rt->dst.error;
2305 if (rt_is_input_route(rt)) {
2306 #ifdef CONFIG_IP_MROUTE
2307 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2308 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2309 int err = ipmr_get_route(net, skb,
2310 fl4->saddr, fl4->daddr,
2316 goto nla_put_failure;
2318 if (err == -EMSGSIZE)
2319 goto nla_put_failure;
2325 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2326 goto nla_put_failure;
2329 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2330 goto nla_put_failure;
2332 return nlmsg_end(skb, nlh);
2335 nlmsg_cancel(skb, nlh);
2339 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2341 struct net *net = sock_net(in_skb->sk);
2343 struct nlattr *tb[RTA_MAX+1];
2344 struct rtable *rt = NULL;
2351 struct sk_buff *skb;
2353 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2357 rtm = nlmsg_data(nlh);
2359 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2365 /* Reserve room for dummy headers, this skb can pass
2366 through good chunk of routing engine.
2368 skb_reset_mac_header(skb);
2369 skb_reset_network_header(skb);
2371 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2372 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2373 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2375 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2376 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2377 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2378 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2380 memset(&fl4, 0, sizeof(fl4));
2383 fl4.flowi4_tos = rtm->rtm_tos;
2384 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2385 fl4.flowi4_mark = mark;
2388 struct net_device *dev;
2390 dev = __dev_get_by_index(net, iif);
2396 skb->protocol = htons(ETH_P_IP);
2400 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2403 rt = skb_rtable(skb);
2404 if (err == 0 && rt->dst.error)
2405 err = -rt->dst.error;
2407 rt = ip_route_output_key(net, &fl4);
2417 skb_dst_set(skb, &rt->dst);
2418 if (rtm->rtm_flags & RTM_F_NOTIFY)
2419 rt->rt_flags |= RTCF_NOTIFY;
2421 err = rt_fill_info(net, dst, src, &fl4, skb,
2422 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2423 RTM_NEWROUTE, 0, 0);
2427 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2436 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2441 void ip_rt_multicast_event(struct in_device *in_dev)
2443 rt_cache_flush(dev_net(in_dev->dev));
2446 #ifdef CONFIG_SYSCTL
2447 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2448 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2449 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2450 static int ip_rt_gc_elasticity __read_mostly = 8;
2452 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2453 void __user *buffer,
2454 size_t *lenp, loff_t *ppos)
2457 rt_cache_flush((struct net *)__ctl->extra1);
2464 static ctl_table ipv4_route_table[] = {
2466 .procname = "gc_thresh",
2467 .data = &ipv4_dst_ops.gc_thresh,
2468 .maxlen = sizeof(int),
2470 .proc_handler = proc_dointvec,
2473 .procname = "max_size",
2474 .data = &ip_rt_max_size,
2475 .maxlen = sizeof(int),
2477 .proc_handler = proc_dointvec,
2480 /* Deprecated. Use gc_min_interval_ms */
2482 .procname = "gc_min_interval",
2483 .data = &ip_rt_gc_min_interval,
2484 .maxlen = sizeof(int),
2486 .proc_handler = proc_dointvec_jiffies,
2489 .procname = "gc_min_interval_ms",
2490 .data = &ip_rt_gc_min_interval,
2491 .maxlen = sizeof(int),
2493 .proc_handler = proc_dointvec_ms_jiffies,
2496 .procname = "gc_timeout",
2497 .data = &ip_rt_gc_timeout,
2498 .maxlen = sizeof(int),
2500 .proc_handler = proc_dointvec_jiffies,
2503 .procname = "gc_interval",
2504 .data = &ip_rt_gc_interval,
2505 .maxlen = sizeof(int),
2507 .proc_handler = proc_dointvec_jiffies,
2510 .procname = "redirect_load",
2511 .data = &ip_rt_redirect_load,
2512 .maxlen = sizeof(int),
2514 .proc_handler = proc_dointvec,
2517 .procname = "redirect_number",
2518 .data = &ip_rt_redirect_number,
2519 .maxlen = sizeof(int),
2521 .proc_handler = proc_dointvec,
2524 .procname = "redirect_silence",
2525 .data = &ip_rt_redirect_silence,
2526 .maxlen = sizeof(int),
2528 .proc_handler = proc_dointvec,
2531 .procname = "error_cost",
2532 .data = &ip_rt_error_cost,
2533 .maxlen = sizeof(int),
2535 .proc_handler = proc_dointvec,
2538 .procname = "error_burst",
2539 .data = &ip_rt_error_burst,
2540 .maxlen = sizeof(int),
2542 .proc_handler = proc_dointvec,
2545 .procname = "gc_elasticity",
2546 .data = &ip_rt_gc_elasticity,
2547 .maxlen = sizeof(int),
2549 .proc_handler = proc_dointvec,
2552 .procname = "mtu_expires",
2553 .data = &ip_rt_mtu_expires,
2554 .maxlen = sizeof(int),
2556 .proc_handler = proc_dointvec_jiffies,
2559 .procname = "min_pmtu",
2560 .data = &ip_rt_min_pmtu,
2561 .maxlen = sizeof(int),
2563 .proc_handler = proc_dointvec,
2566 .procname = "min_adv_mss",
2567 .data = &ip_rt_min_advmss,
2568 .maxlen = sizeof(int),
2570 .proc_handler = proc_dointvec,
2575 static struct ctl_table ipv4_route_flush_table[] = {
2577 .procname = "flush",
2578 .maxlen = sizeof(int),
2580 .proc_handler = ipv4_sysctl_rtcache_flush,
2585 static __net_init int sysctl_route_net_init(struct net *net)
2587 struct ctl_table *tbl;
2589 tbl = ipv4_route_flush_table;
2590 if (!net_eq(net, &init_net)) {
2591 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2595 /* Don't export sysctls to unprivileged users */
2596 if (net->user_ns != &init_user_ns)
2597 tbl[0].procname = NULL;
2599 tbl[0].extra1 = net;
2601 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2602 if (net->ipv4.route_hdr == NULL)
2607 if (tbl != ipv4_route_flush_table)
2613 static __net_exit void sysctl_route_net_exit(struct net *net)
2615 struct ctl_table *tbl;
2617 tbl = net->ipv4.route_hdr->ctl_table_arg;
2618 unregister_net_sysctl_table(net->ipv4.route_hdr);
2619 BUG_ON(tbl == ipv4_route_flush_table);
2623 static __net_initdata struct pernet_operations sysctl_route_ops = {
2624 .init = sysctl_route_net_init,
2625 .exit = sysctl_route_net_exit,
2629 static __net_init int rt_genid_init(struct net *net)
2631 atomic_set(&net->rt_genid, 0);
2632 get_random_bytes(&net->ipv4.dev_addr_genid,
2633 sizeof(net->ipv4.dev_addr_genid));
2637 static __net_initdata struct pernet_operations rt_genid_ops = {
2638 .init = rt_genid_init,
2641 static int __net_init ipv4_inetpeer_init(struct net *net)
2643 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2647 inet_peer_base_init(bp);
2648 net->ipv4.peers = bp;
2652 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2654 struct inet_peer_base *bp = net->ipv4.peers;
2656 net->ipv4.peers = NULL;
2657 inetpeer_invalidate_tree(bp);
2661 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2662 .init = ipv4_inetpeer_init,
2663 .exit = ipv4_inetpeer_exit,
2666 #ifdef CONFIG_IP_ROUTE_CLASSID
2667 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2668 #endif /* CONFIG_IP_ROUTE_CLASSID */
2670 int __init ip_rt_init(void)
2674 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2676 panic("IP: failed to allocate ip_idents\n");
2678 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2680 #ifdef CONFIG_IP_ROUTE_CLASSID
2681 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2683 panic("IP: failed to allocate ip_rt_acct\n");
2686 ipv4_dst_ops.kmem_cachep =
2687 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2688 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2690 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2692 if (dst_entries_init(&ipv4_dst_ops) < 0)
2693 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2695 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2696 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2698 ipv4_dst_ops.gc_thresh = ~0;
2699 ip_rt_max_size = INT_MAX;
2704 if (ip_rt_proc_init())
2705 pr_err("Unable to create route proc files\n");
2710 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2712 #ifdef CONFIG_SYSCTL
2713 register_pernet_subsys(&sysctl_route_ops);
2715 register_pernet_subsys(&rt_genid_ops);
2716 register_pernet_subsys(&ipv4_inetpeer_ops);
2720 #ifdef CONFIG_SYSCTL
2722 * We really need to sanitize the damn ipv4 init order, then all
2723 * this nonsense will go away.
2725 void __init ip_static_sysctl_init(void)
2727 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);