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/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/workqueue.h>
84 #include <linux/skbuff.h>
85 #include <linux/inetdevice.h>
86 #include <linux/igmp.h>
87 #include <linux/pkt_sched.h>
88 #include <linux/mroute.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/jhash.h>
92 #include <linux/rcupdate.h>
93 #include <linux/times.h>
94 #include <linux/slab.h>
95 #include <linux/prefetch.h>
97 #include <net/net_namespace.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/inetpeer.h>
102 #include <net/sock.h>
103 #include <net/ip_fib.h>
106 #include <net/icmp.h>
107 #include <net/xfrm.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
114 #include <net/secure_seq.h>
116 #define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119 #define IP_MAX_MTU 0xFFF0
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size;
124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
126 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
127 static int ip_rt_redirect_number __read_mostly = 9;
128 static int ip_rt_redirect_load __read_mostly = HZ / 50;
129 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130 static int ip_rt_error_cost __read_mostly = HZ;
131 static int ip_rt_error_burst __read_mostly = 5 * HZ;
132 static int ip_rt_gc_elasticity __read_mostly = 8;
133 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
134 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
135 static int ip_rt_min_advmss __read_mostly = 256;
138 * Interface to generic destination cache.
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
143 static unsigned int ipv4_mtu(const struct dst_entry *dst);
144 static void ipv4_dst_destroy(struct dst_entry *dst);
145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146 static void ipv4_link_failure(struct sk_buff *skb);
147 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
148 struct sk_buff *skb, u32 mtu);
149 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
150 struct sk_buff *skb);
152 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
157 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
163 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
167 static struct dst_ops ipv4_dst_ops = {
169 .protocol = cpu_to_be16(ETH_P_IP),
170 .check = ipv4_dst_check,
171 .default_advmss = ipv4_default_advmss,
173 .cow_metrics = ipv4_cow_metrics,
174 .destroy = ipv4_dst_destroy,
175 .ifdown = ipv4_dst_ifdown,
176 .negative_advice = ipv4_negative_advice,
177 .link_failure = ipv4_link_failure,
178 .update_pmtu = ip_rt_update_pmtu,
179 .redirect = ip_do_redirect,
180 .local_out = __ip_local_out,
181 .neigh_lookup = ipv4_neigh_lookup,
184 #define ECN_OR_COST(class) TC_PRIO_##class
186 const __u8 ip_tos2prio[16] = {
188 ECN_OR_COST(BESTEFFORT),
190 ECN_OR_COST(BESTEFFORT),
196 ECN_OR_COST(INTERACTIVE),
198 ECN_OR_COST(INTERACTIVE),
199 TC_PRIO_INTERACTIVE_BULK,
200 ECN_OR_COST(INTERACTIVE_BULK),
201 TC_PRIO_INTERACTIVE_BULK,
202 ECN_OR_COST(INTERACTIVE_BULK)
204 EXPORT_SYMBOL(ip_tos2prio);
206 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
207 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
209 static inline int rt_genid(struct net *net)
211 return atomic_read(&net->ipv4.rt_genid);
214 #ifdef CONFIG_PROC_FS
215 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
219 return SEQ_START_TOKEN;
222 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
228 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
232 static int rt_cache_seq_show(struct seq_file *seq, void *v)
234 if (v == SEQ_START_TOKEN)
235 seq_printf(seq, "%-127s\n",
236 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
237 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
242 static const struct seq_operations rt_cache_seq_ops = {
243 .start = rt_cache_seq_start,
244 .next = rt_cache_seq_next,
245 .stop = rt_cache_seq_stop,
246 .show = rt_cache_seq_show,
249 static int rt_cache_seq_open(struct inode *inode, struct file *file)
251 return seq_open(file, &rt_cache_seq_ops);
254 static const struct file_operations rt_cache_seq_fops = {
255 .owner = THIS_MODULE,
256 .open = rt_cache_seq_open,
259 .release = seq_release,
263 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
268 return SEQ_START_TOKEN;
270 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
271 if (!cpu_possible(cpu))
274 return &per_cpu(rt_cache_stat, cpu);
279 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
283 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
284 if (!cpu_possible(cpu))
287 return &per_cpu(rt_cache_stat, cpu);
293 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
298 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
300 struct rt_cache_stat *st = v;
302 if (v == SEQ_START_TOKEN) {
303 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");
307 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
308 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
309 dst_entries_get_slow(&ipv4_dst_ops),
332 static const struct seq_operations rt_cpu_seq_ops = {
333 .start = rt_cpu_seq_start,
334 .next = rt_cpu_seq_next,
335 .stop = rt_cpu_seq_stop,
336 .show = rt_cpu_seq_show,
340 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
342 return seq_open(file, &rt_cpu_seq_ops);
345 static const struct file_operations rt_cpu_seq_fops = {
346 .owner = THIS_MODULE,
347 .open = rt_cpu_seq_open,
350 .release = seq_release,
353 #ifdef CONFIG_IP_ROUTE_CLASSID
354 static int rt_acct_proc_show(struct seq_file *m, void *v)
356 struct ip_rt_acct *dst, *src;
359 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
363 for_each_possible_cpu(i) {
364 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
365 for (j = 0; j < 256; j++) {
366 dst[j].o_bytes += src[j].o_bytes;
367 dst[j].o_packets += src[j].o_packets;
368 dst[j].i_bytes += src[j].i_bytes;
369 dst[j].i_packets += src[j].i_packets;
373 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
378 static int rt_acct_proc_open(struct inode *inode, struct file *file)
380 return single_open(file, rt_acct_proc_show, NULL);
383 static const struct file_operations rt_acct_proc_fops = {
384 .owner = THIS_MODULE,
385 .open = rt_acct_proc_open,
388 .release = single_release,
392 static int __net_init ip_rt_do_proc_init(struct net *net)
394 struct proc_dir_entry *pde;
396 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
401 pde = proc_create("rt_cache", S_IRUGO,
402 net->proc_net_stat, &rt_cpu_seq_fops);
406 #ifdef CONFIG_IP_ROUTE_CLASSID
407 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
413 #ifdef CONFIG_IP_ROUTE_CLASSID
415 remove_proc_entry("rt_cache", net->proc_net_stat);
418 remove_proc_entry("rt_cache", net->proc_net);
423 static void __net_exit ip_rt_do_proc_exit(struct net *net)
425 remove_proc_entry("rt_cache", net->proc_net_stat);
426 remove_proc_entry("rt_cache", net->proc_net);
427 #ifdef CONFIG_IP_ROUTE_CLASSID
428 remove_proc_entry("rt_acct", net->proc_net);
432 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
433 .init = ip_rt_do_proc_init,
434 .exit = ip_rt_do_proc_exit,
437 static int __init ip_rt_proc_init(void)
439 return register_pernet_subsys(&ip_rt_proc_ops);
443 static inline int ip_rt_proc_init(void)
447 #endif /* CONFIG_PROC_FS */
449 static inline int rt_is_expired(struct rtable *rth)
451 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
455 * Perturbation of rt_genid by a small quantity [1..256]
456 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
457 * many times (2^24) without giving recent rt_genid.
458 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
460 static void rt_cache_invalidate(struct net *net)
462 unsigned char shuffle;
464 get_random_bytes(&shuffle, sizeof(shuffle));
465 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
469 * delay < 0 : invalidate cache (fast : entries will be deleted later)
470 * delay >= 0 : invalidate & flush cache (can be long)
472 void rt_cache_flush(struct net *net, int delay)
474 rt_cache_invalidate(net);
477 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
481 struct net_device *dev = dst->dev;
482 const __be32 *pkey = daddr;
483 const struct rtable *rt;
486 rt = (const struct rtable *) dst;
488 pkey = (const __be32 *) &rt->rt_gateway;
490 pkey = &ip_hdr(skb)->daddr;
492 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
495 return neigh_create(&arp_tbl, pkey, dev);
499 * Peer allocation may fail only in serious out-of-memory conditions. However
500 * we still can generate some output.
501 * Random ID selection looks a bit dangerous because we have no chances to
502 * select ID being unique in a reasonable period of time.
503 * But broken packet identifier may be better than no packet at all.
505 static void ip_select_fb_ident(struct iphdr *iph)
507 static DEFINE_SPINLOCK(ip_fb_id_lock);
508 static u32 ip_fallback_id;
511 spin_lock_bh(&ip_fb_id_lock);
512 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
513 iph->id = htons(salt & 0xFFFF);
514 ip_fallback_id = salt;
515 spin_unlock_bh(&ip_fb_id_lock);
518 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
520 struct net *net = dev_net(dst->dev);
521 struct inet_peer *peer;
523 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
525 iph->id = htons(inet_getid(peer, more));
530 ip_select_fb_ident(iph);
532 EXPORT_SYMBOL(__ip_select_ident);
534 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
535 const struct iphdr *iph,
537 u8 prot, u32 mark, int flow_flags)
540 const struct inet_sock *inet = inet_sk(sk);
542 oif = sk->sk_bound_dev_if;
544 tos = RT_CONN_FLAGS(sk);
545 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
547 flowi4_init_output(fl4, oif, mark, tos,
548 RT_SCOPE_UNIVERSE, prot,
550 iph->daddr, iph->saddr, 0, 0);
553 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
554 const struct sock *sk)
556 const struct iphdr *iph = ip_hdr(skb);
557 int oif = skb->dev->ifindex;
558 u8 tos = RT_TOS(iph->tos);
559 u8 prot = iph->protocol;
560 u32 mark = skb->mark;
562 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
565 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
567 const struct inet_sock *inet = inet_sk(sk);
568 const struct ip_options_rcu *inet_opt;
569 __be32 daddr = inet->inet_daddr;
572 inet_opt = rcu_dereference(inet->inet_opt);
573 if (inet_opt && inet_opt->opt.srr)
574 daddr = inet_opt->opt.faddr;
575 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
576 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
577 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
578 inet_sk_flowi_flags(sk),
579 daddr, inet->inet_saddr, 0, 0);
583 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
584 const struct sk_buff *skb)
587 build_skb_flow_key(fl4, skb, sk);
589 build_sk_flow_key(fl4, sk);
592 static DEFINE_SEQLOCK(fnhe_seqlock);
594 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
596 struct fib_nh_exception *fnhe, *oldest;
598 oldest = rcu_dereference(hash->chain);
599 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
600 fnhe = rcu_dereference(fnhe->fnhe_next)) {
601 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
607 static inline u32 fnhe_hashfun(__be32 daddr)
611 hval = (__force u32) daddr;
612 hval ^= (hval >> 11) ^ (hval >> 22);
614 return hval & (FNHE_HASH_SIZE - 1);
617 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
618 u32 pmtu, unsigned long expires)
620 struct fnhe_hash_bucket *hash;
621 struct fib_nh_exception *fnhe;
623 u32 hval = fnhe_hashfun(daddr);
625 write_seqlock_bh(&fnhe_seqlock);
627 hash = nh->nh_exceptions;
629 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
632 nh->nh_exceptions = hash;
638 for (fnhe = rcu_dereference(hash->chain); fnhe;
639 fnhe = rcu_dereference(fnhe->fnhe_next)) {
640 if (fnhe->fnhe_daddr == daddr)
649 fnhe->fnhe_pmtu = pmtu;
650 fnhe->fnhe_expires = expires;
653 if (depth > FNHE_RECLAIM_DEPTH)
654 fnhe = fnhe_oldest(hash);
656 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
660 fnhe->fnhe_next = hash->chain;
661 rcu_assign_pointer(hash->chain, fnhe);
663 fnhe->fnhe_daddr = daddr;
665 fnhe->fnhe_pmtu = pmtu;
666 fnhe->fnhe_expires = expires;
669 fnhe->fnhe_stamp = jiffies;
672 write_sequnlock_bh(&fnhe_seqlock);
676 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4)
678 __be32 new_gw = icmp_hdr(skb)->un.gateway;
679 __be32 old_gw = ip_hdr(skb)->saddr;
680 struct net_device *dev = skb->dev;
681 struct in_device *in_dev;
682 struct fib_result res;
686 switch (icmp_hdr(skb)->code & 7) {
688 case ICMP_REDIR_NETTOS:
689 case ICMP_REDIR_HOST:
690 case ICMP_REDIR_HOSTTOS:
697 if (rt->rt_gateway != old_gw)
700 in_dev = __in_dev_get_rcu(dev);
705 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
706 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
707 ipv4_is_zeronet(new_gw))
708 goto reject_redirect;
710 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
711 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
712 goto reject_redirect;
713 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
714 goto reject_redirect;
716 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
717 goto reject_redirect;
720 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
722 if (!(n->nud_state & NUD_VALID)) {
723 neigh_event_send(n, NULL);
725 if (fib_lookup(net, fl4, &res) == 0) {
726 struct fib_nh *nh = &FIB_RES_NH(res);
728 update_or_create_fnhe(nh, fl4->daddr, new_gw,
731 rt->rt_gateway = new_gw;
732 rt->rt_flags |= RTCF_REDIRECTED;
733 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
740 #ifdef CONFIG_IP_ROUTE_VERBOSE
741 if (IN_DEV_LOG_MARTIANS(in_dev)) {
742 const struct iphdr *iph = (const struct iphdr *) skb->data;
743 __be32 daddr = iph->daddr;
744 __be32 saddr = iph->saddr;
746 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
747 " Advised path = %pI4 -> %pI4\n",
748 &old_gw, dev->name, &new_gw,
755 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
760 rt = (struct rtable *) dst;
762 ip_rt_build_flow_key(&fl4, sk, skb);
763 __ip_do_redirect(rt, skb, &fl4);
766 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
768 struct rtable *rt = (struct rtable *)dst;
769 struct dst_entry *ret = dst;
772 if (dst->obsolete > 0) {
775 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
786 * 1. The first ip_rt_redirect_number redirects are sent
787 * with exponential backoff, then we stop sending them at all,
788 * assuming that the host ignores our redirects.
789 * 2. If we did not see packets requiring redirects
790 * during ip_rt_redirect_silence, we assume that the host
791 * forgot redirected route and start to send redirects again.
793 * This algorithm is much cheaper and more intelligent than dumb load limiting
796 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
797 * and "frag. need" (breaks PMTU discovery) in icmp.c.
800 void ip_rt_send_redirect(struct sk_buff *skb)
802 struct rtable *rt = skb_rtable(skb);
803 struct in_device *in_dev;
804 struct inet_peer *peer;
809 in_dev = __in_dev_get_rcu(rt->dst.dev);
810 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
814 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
817 net = dev_net(rt->dst.dev);
818 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
820 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
824 /* No redirected packets during ip_rt_redirect_silence;
825 * reset the algorithm.
827 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
828 peer->rate_tokens = 0;
830 /* Too many ignored redirects; do not send anything
831 * set dst.rate_last to the last seen redirected packet.
833 if (peer->rate_tokens >= ip_rt_redirect_number) {
834 peer->rate_last = jiffies;
838 /* Check for load limit; set rate_last to the latest sent
841 if (peer->rate_tokens == 0 ||
844 (ip_rt_redirect_load << peer->rate_tokens)))) {
845 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
846 peer->rate_last = jiffies;
848 #ifdef CONFIG_IP_ROUTE_VERBOSE
850 peer->rate_tokens == ip_rt_redirect_number)
851 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
852 &ip_hdr(skb)->saddr, rt->rt_iif,
853 &ip_hdr(skb)->daddr, &rt->rt_gateway);
860 static int ip_error(struct sk_buff *skb)
862 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
863 struct rtable *rt = skb_rtable(skb);
864 struct inet_peer *peer;
870 net = dev_net(rt->dst.dev);
871 if (!IN_DEV_FORWARD(in_dev)) {
872 switch (rt->dst.error) {
874 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
878 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
884 switch (rt->dst.error) {
889 code = ICMP_HOST_UNREACH;
892 code = ICMP_NET_UNREACH;
893 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
896 code = ICMP_PKT_FILTERED;
900 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
905 peer->rate_tokens += now - peer->rate_last;
906 if (peer->rate_tokens > ip_rt_error_burst)
907 peer->rate_tokens = ip_rt_error_burst;
908 peer->rate_last = now;
909 if (peer->rate_tokens >= ip_rt_error_cost)
910 peer->rate_tokens -= ip_rt_error_cost;
916 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
922 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
924 struct fib_result res;
926 if (mtu < ip_rt_min_pmtu)
927 mtu = ip_rt_min_pmtu;
929 if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) {
930 struct fib_nh *nh = &FIB_RES_NH(res);
932 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
933 jiffies + ip_rt_mtu_expires);
936 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
939 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
940 struct sk_buff *skb, u32 mtu)
942 struct rtable *rt = (struct rtable *) dst;
945 ip_rt_build_flow_key(&fl4, sk, skb);
946 __ip_rt_update_pmtu(rt, &fl4, mtu);
949 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
950 int oif, u32 mark, u8 protocol, int flow_flags)
952 const struct iphdr *iph = (const struct iphdr *) skb->data;
956 __build_flow_key(&fl4, NULL, iph, oif,
957 RT_TOS(iph->tos), protocol, mark, flow_flags);
958 rt = __ip_route_output_key(net, &fl4);
960 __ip_rt_update_pmtu(rt, &fl4, mtu);
964 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
966 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
968 const struct iphdr *iph = (const struct iphdr *) skb->data;
972 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
973 rt = __ip_route_output_key(sock_net(sk), &fl4);
975 __ip_rt_update_pmtu(rt, &fl4, mtu);
979 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
981 void ipv4_redirect(struct sk_buff *skb, struct net *net,
982 int oif, u32 mark, u8 protocol, int flow_flags)
984 const struct iphdr *iph = (const struct iphdr *) skb->data;
988 __build_flow_key(&fl4, NULL, iph, oif,
989 RT_TOS(iph->tos), protocol, mark, flow_flags);
990 rt = __ip_route_output_key(net, &fl4);
992 __ip_do_redirect(rt, skb, &fl4);
996 EXPORT_SYMBOL_GPL(ipv4_redirect);
998 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1000 const struct iphdr *iph = (const struct iphdr *) skb->data;
1004 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1005 rt = __ip_route_output_key(sock_net(sk), &fl4);
1007 __ip_do_redirect(rt, skb, &fl4);
1011 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1013 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1015 struct rtable *rt = (struct rtable *) dst;
1017 if (rt_is_expired(rt))
1022 static void ipv4_dst_destroy(struct dst_entry *dst)
1024 struct rtable *rt = (struct rtable *) dst;
1027 fib_info_put(rt->fi);
1033 static void ipv4_link_failure(struct sk_buff *skb)
1037 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1039 rt = skb_rtable(skb);
1041 dst_set_expires(&rt->dst, 0);
1044 static int ip_rt_bug(struct sk_buff *skb)
1046 pr_debug("%s: %pI4 -> %pI4, %s\n",
1047 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1048 skb->dev ? skb->dev->name : "?");
1055 We do not cache source address of outgoing interface,
1056 because it is used only by IP RR, TS and SRR options,
1057 so that it out of fast path.
1059 BTW remember: "addr" is allowed to be not aligned
1063 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1067 if (rt_is_output_route(rt))
1068 src = ip_hdr(skb)->saddr;
1070 struct fib_result res;
1076 memset(&fl4, 0, sizeof(fl4));
1077 fl4.daddr = iph->daddr;
1078 fl4.saddr = iph->saddr;
1079 fl4.flowi4_tos = RT_TOS(iph->tos);
1080 fl4.flowi4_oif = rt->dst.dev->ifindex;
1081 fl4.flowi4_iif = skb->dev->ifindex;
1082 fl4.flowi4_mark = skb->mark;
1085 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1086 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1088 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1092 memcpy(addr, &src, 4);
1095 #ifdef CONFIG_IP_ROUTE_CLASSID
1096 static void set_class_tag(struct rtable *rt, u32 tag)
1098 if (!(rt->dst.tclassid & 0xFFFF))
1099 rt->dst.tclassid |= tag & 0xFFFF;
1100 if (!(rt->dst.tclassid & 0xFFFF0000))
1101 rt->dst.tclassid |= tag & 0xFFFF0000;
1105 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1107 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1110 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1112 if (advmss > 65535 - 40)
1113 advmss = 65535 - 40;
1118 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1120 const struct rtable *rt = (const struct rtable *) dst;
1121 unsigned int mtu = rt->rt_pmtu;
1123 if (mtu && time_after_eq(jiffies, rt->dst.expires))
1127 mtu = dst_metric_raw(dst, RTAX_MTU);
1129 if (mtu && rt_is_output_route(rt))
1132 mtu = dst->dev->mtu;
1134 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1135 if (rt->rt_gateway != 0 && mtu > 576)
1139 if (mtu > IP_MAX_MTU)
1145 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1146 struct fib_info *fi)
1148 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1150 atomic_inc(&fi->fib_clntref);
1152 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1155 static void rt_bind_exception(struct rtable *rt, struct fib_nh *nh, __be32 daddr)
1157 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1158 struct fib_nh_exception *fnhe;
1161 hval = fnhe_hashfun(daddr);
1164 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1165 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1166 __be32 fnhe_daddr, gw;
1167 unsigned long expires;
1171 seq = read_seqbegin(&fnhe_seqlock);
1172 fnhe_daddr = fnhe->fnhe_daddr;
1174 pmtu = fnhe->fnhe_pmtu;
1175 expires = fnhe->fnhe_expires;
1176 if (read_seqretry(&fnhe_seqlock, seq))
1178 if (daddr != fnhe_daddr)
1181 unsigned long diff = expires - jiffies;
1183 if (time_before(jiffies, expires)) {
1185 dst_set_expires(&rt->dst, diff);
1189 rt->rt_gateway = gw;
1190 fnhe->fnhe_stamp = jiffies;
1195 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1196 const struct fib_result *res,
1197 struct fib_info *fi, u16 type, u32 itag)
1200 struct fib_nh *nh = &FIB_RES_NH(*res);
1202 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1203 rt->rt_gateway = nh->nh_gw;
1204 if (unlikely(nh->nh_exceptions))
1205 rt_bind_exception(rt, nh, fl4->daddr);
1206 rt_init_metrics(rt, fl4, fi);
1207 #ifdef CONFIG_IP_ROUTE_CLASSID
1208 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1212 #ifdef CONFIG_IP_ROUTE_CLASSID
1213 #ifdef CONFIG_IP_MULTIPLE_TABLES
1214 set_class_tag(rt, res->tclassid);
1216 set_class_tag(rt, itag);
1220 static struct rtable *rt_dst_alloc(struct net_device *dev,
1221 bool nopolicy, bool noxfrm)
1223 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1224 DST_HOST | DST_NOCACHE |
1225 (nopolicy ? DST_NOPOLICY : 0) |
1226 (noxfrm ? DST_NOXFRM : 0));
1229 /* called in rcu_read_lock() section */
1230 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1231 u8 tos, struct net_device *dev, int our)
1234 struct in_device *in_dev = __in_dev_get_rcu(dev);
1238 /* Primary sanity checks. */
1243 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1244 skb->protocol != htons(ETH_P_IP))
1247 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1248 if (ipv4_is_loopback(saddr))
1251 if (ipv4_is_zeronet(saddr)) {
1252 if (!ipv4_is_local_multicast(daddr))
1255 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1260 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1261 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1265 #ifdef CONFIG_IP_ROUTE_CLASSID
1266 rth->dst.tclassid = itag;
1268 rth->dst.output = ip_rt_bug;
1270 rth->rt_genid = rt_genid(dev_net(dev));
1271 rth->rt_flags = RTCF_MULTICAST;
1272 rth->rt_type = RTN_MULTICAST;
1273 rth->rt_route_iif = dev->ifindex;
1274 rth->rt_iif = dev->ifindex;
1277 rth->rt_gateway = daddr;
1280 rth->dst.input= ip_local_deliver;
1281 rth->rt_flags |= RTCF_LOCAL;
1284 #ifdef CONFIG_IP_MROUTE
1285 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1286 rth->dst.input = ip_mr_input;
1288 RT_CACHE_STAT_INC(in_slow_mc);
1290 skb_dst_set(skb, &rth->dst);
1302 static void ip_handle_martian_source(struct net_device *dev,
1303 struct in_device *in_dev,
1304 struct sk_buff *skb,
1308 RT_CACHE_STAT_INC(in_martian_src);
1309 #ifdef CONFIG_IP_ROUTE_VERBOSE
1310 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1312 * RFC1812 recommendation, if source is martian,
1313 * the only hint is MAC header.
1315 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1316 &daddr, &saddr, dev->name);
1317 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1318 print_hex_dump(KERN_WARNING, "ll header: ",
1319 DUMP_PREFIX_OFFSET, 16, 1,
1320 skb_mac_header(skb),
1321 dev->hard_header_len, true);
1327 /* called in rcu_read_lock() section */
1328 static int __mkroute_input(struct sk_buff *skb,
1329 const struct fib_result *res,
1330 struct in_device *in_dev,
1331 __be32 daddr, __be32 saddr, u32 tos,
1332 struct rtable **result)
1336 struct in_device *out_dev;
1337 unsigned int flags = 0;
1340 /* get a working reference to the output device */
1341 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1342 if (out_dev == NULL) {
1343 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1348 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1349 in_dev->dev, in_dev, &itag);
1351 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1358 flags |= RTCF_DIRECTSRC;
1360 if (out_dev == in_dev && err &&
1361 (IN_DEV_SHARED_MEDIA(out_dev) ||
1362 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1363 flags |= RTCF_DOREDIRECT;
1365 if (skb->protocol != htons(ETH_P_IP)) {
1366 /* Not IP (i.e. ARP). Do not create route, if it is
1367 * invalid for proxy arp. DNAT routes are always valid.
1369 * Proxy arp feature have been extended to allow, ARP
1370 * replies back to the same interface, to support
1371 * Private VLAN switch technologies. See arp.c.
1373 if (out_dev == in_dev &&
1374 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1380 rth = rt_dst_alloc(out_dev->dev,
1381 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1382 IN_DEV_CONF_GET(out_dev, NOXFRM));
1388 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1389 rth->rt_flags = flags;
1390 rth->rt_type = res->type;
1391 rth->rt_route_iif = in_dev->dev->ifindex;
1392 rth->rt_iif = in_dev->dev->ifindex;
1395 rth->rt_gateway = daddr;
1398 rth->dst.input = ip_forward;
1399 rth->dst.output = ip_output;
1401 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
1409 static int ip_mkroute_input(struct sk_buff *skb,
1410 struct fib_result *res,
1411 const struct flowi4 *fl4,
1412 struct in_device *in_dev,
1413 __be32 daddr, __be32 saddr, u32 tos)
1415 struct rtable *rth = NULL;
1418 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1419 if (res->fi && res->fi->fib_nhs > 1)
1420 fib_select_multipath(res);
1423 /* create a routing cache entry */
1424 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1428 skb_dst_set(skb, &rth->dst);
1433 * NOTE. We drop all the packets that has local source
1434 * addresses, because every properly looped back packet
1435 * must have correct destination already attached by output routine.
1437 * Such approach solves two big problems:
1438 * 1. Not simplex devices are handled properly.
1439 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1440 * called with rcu_read_lock()
1443 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1444 u8 tos, struct net_device *dev)
1446 struct fib_result res;
1447 struct in_device *in_dev = __in_dev_get_rcu(dev);
1449 unsigned int flags = 0;
1453 struct net *net = dev_net(dev);
1455 /* IP on this device is disabled. */
1460 /* Check for the most weird martians, which can be not detected
1464 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1465 goto martian_source;
1467 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1470 /* Accept zero addresses only to limited broadcast;
1471 * I even do not know to fix it or not. Waiting for complains :-)
1473 if (ipv4_is_zeronet(saddr))
1474 goto martian_source;
1476 if (ipv4_is_zeronet(daddr))
1477 goto martian_destination;
1479 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
1480 if (ipv4_is_loopback(daddr))
1481 goto martian_destination;
1483 if (ipv4_is_loopback(saddr))
1484 goto martian_source;
1488 * Now we are ready to route packet.
1491 fl4.flowi4_iif = dev->ifindex;
1492 fl4.flowi4_mark = skb->mark;
1493 fl4.flowi4_tos = tos;
1494 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1497 err = fib_lookup(net, &fl4, &res);
1501 RT_CACHE_STAT_INC(in_slow_tot);
1503 if (res.type == RTN_BROADCAST)
1506 if (res.type == RTN_LOCAL) {
1507 err = fib_validate_source(skb, saddr, daddr, tos,
1508 net->loopback_dev->ifindex,
1509 dev, in_dev, &itag);
1511 goto martian_source_keep_err;
1513 flags |= RTCF_DIRECTSRC;
1517 if (!IN_DEV_FORWARD(in_dev))
1519 if (res.type != RTN_UNICAST)
1520 goto martian_destination;
1522 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1526 if (skb->protocol != htons(ETH_P_IP))
1529 if (!ipv4_is_zeronet(saddr)) {
1530 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1533 goto martian_source_keep_err;
1535 flags |= RTCF_DIRECTSRC;
1537 flags |= RTCF_BROADCAST;
1538 res.type = RTN_BROADCAST;
1539 RT_CACHE_STAT_INC(in_brd);
1542 rth = rt_dst_alloc(net->loopback_dev,
1543 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1547 rth->dst.input= ip_local_deliver;
1548 rth->dst.output= ip_rt_bug;
1549 #ifdef CONFIG_IP_ROUTE_CLASSID
1550 rth->dst.tclassid = itag;
1553 rth->rt_genid = rt_genid(net);
1554 rth->rt_flags = flags|RTCF_LOCAL;
1555 rth->rt_type = res.type;
1556 rth->rt_route_iif = dev->ifindex;
1557 rth->rt_iif = dev->ifindex;
1560 rth->rt_gateway = daddr;
1562 if (res.type == RTN_UNREACHABLE) {
1563 rth->dst.input= ip_error;
1564 rth->dst.error= -err;
1565 rth->rt_flags &= ~RTCF_LOCAL;
1567 skb_dst_set(skb, &rth->dst);
1572 RT_CACHE_STAT_INC(in_no_route);
1573 res.type = RTN_UNREACHABLE;
1579 * Do not cache martian addresses: they should be logged (RFC1812)
1581 martian_destination:
1582 RT_CACHE_STAT_INC(in_martian_dst);
1583 #ifdef CONFIG_IP_ROUTE_VERBOSE
1584 if (IN_DEV_LOG_MARTIANS(in_dev))
1585 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1586 &daddr, &saddr, dev->name);
1599 martian_source_keep_err:
1600 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1604 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1605 u8 tos, struct net_device *dev)
1611 /* Multicast recognition logic is moved from route cache to here.
1612 The problem was that too many Ethernet cards have broken/missing
1613 hardware multicast filters :-( As result the host on multicasting
1614 network acquires a lot of useless route cache entries, sort of
1615 SDR messages from all the world. Now we try to get rid of them.
1616 Really, provided software IP multicast filter is organized
1617 reasonably (at least, hashed), it does not result in a slowdown
1618 comparing with route cache reject entries.
1619 Note, that multicast routers are not affected, because
1620 route cache entry is created eventually.
1622 if (ipv4_is_multicast(daddr)) {
1623 struct in_device *in_dev = __in_dev_get_rcu(dev);
1626 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1627 ip_hdr(skb)->protocol);
1629 #ifdef CONFIG_IP_MROUTE
1631 (!ipv4_is_local_multicast(daddr) &&
1632 IN_DEV_MFORWARD(in_dev))
1635 int res = ip_route_input_mc(skb, daddr, saddr,
1644 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1648 EXPORT_SYMBOL(ip_route_input);
1650 /* called with rcu_read_lock() */
1651 static struct rtable *__mkroute_output(const struct fib_result *res,
1652 const struct flowi4 *fl4, int orig_oif,
1653 struct net_device *dev_out,
1656 struct fib_info *fi = res->fi;
1657 struct in_device *in_dev;
1658 u16 type = res->type;
1661 in_dev = __in_dev_get_rcu(dev_out);
1663 return ERR_PTR(-EINVAL);
1665 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1666 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1667 return ERR_PTR(-EINVAL);
1669 if (ipv4_is_lbcast(fl4->daddr))
1670 type = RTN_BROADCAST;
1671 else if (ipv4_is_multicast(fl4->daddr))
1672 type = RTN_MULTICAST;
1673 else if (ipv4_is_zeronet(fl4->daddr))
1674 return ERR_PTR(-EINVAL);
1676 if (dev_out->flags & IFF_LOOPBACK)
1677 flags |= RTCF_LOCAL;
1679 if (type == RTN_BROADCAST) {
1680 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1682 } else if (type == RTN_MULTICAST) {
1683 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1684 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1686 flags &= ~RTCF_LOCAL;
1687 /* If multicast route do not exist use
1688 * default one, but do not gateway in this case.
1691 if (fi && res->prefixlen < 4)
1695 rth = rt_dst_alloc(dev_out,
1696 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1697 IN_DEV_CONF_GET(in_dev, NOXFRM));
1699 return ERR_PTR(-ENOBUFS);
1701 rth->dst.output = ip_output;
1703 rth->rt_genid = rt_genid(dev_net(dev_out));
1704 rth->rt_flags = flags;
1705 rth->rt_type = type;
1706 rth->rt_route_iif = 0;
1707 rth->rt_iif = orig_oif ? : dev_out->ifindex;
1708 rth->rt_oif = orig_oif;
1710 rth->rt_gateway = fl4->daddr;
1713 RT_CACHE_STAT_INC(out_slow_tot);
1715 if (flags & RTCF_LOCAL)
1716 rth->dst.input = ip_local_deliver;
1717 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1718 if (flags & RTCF_LOCAL &&
1719 !(dev_out->flags & IFF_LOOPBACK)) {
1720 rth->dst.output = ip_mc_output;
1721 RT_CACHE_STAT_INC(out_slow_mc);
1723 #ifdef CONFIG_IP_MROUTE
1724 if (type == RTN_MULTICAST) {
1725 if (IN_DEV_MFORWARD(in_dev) &&
1726 !ipv4_is_local_multicast(fl4->daddr)) {
1727 rth->dst.input = ip_mr_input;
1728 rth->dst.output = ip_mc_output;
1734 rt_set_nexthop(rth, fl4, res, fi, type, 0);
1736 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
1737 rth->dst.flags |= DST_NOCACHE;
1743 * Major route resolver routine.
1746 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1748 struct net_device *dev_out = NULL;
1749 __u8 tos = RT_FL_TOS(fl4);
1750 unsigned int flags = 0;
1751 struct fib_result res;
1759 orig_oif = fl4->flowi4_oif;
1761 fl4->flowi4_iif = net->loopback_dev->ifindex;
1762 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1763 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1764 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1768 rth = ERR_PTR(-EINVAL);
1769 if (ipv4_is_multicast(fl4->saddr) ||
1770 ipv4_is_lbcast(fl4->saddr) ||
1771 ipv4_is_zeronet(fl4->saddr))
1774 /* I removed check for oif == dev_out->oif here.
1775 It was wrong for two reasons:
1776 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1777 is assigned to multiple interfaces.
1778 2. Moreover, we are allowed to send packets with saddr
1779 of another iface. --ANK
1782 if (fl4->flowi4_oif == 0 &&
1783 (ipv4_is_multicast(fl4->daddr) ||
1784 ipv4_is_lbcast(fl4->daddr))) {
1785 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1786 dev_out = __ip_dev_find(net, fl4->saddr, false);
1787 if (dev_out == NULL)
1790 /* Special hack: user can direct multicasts
1791 and limited broadcast via necessary interface
1792 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1793 This hack is not just for fun, it allows
1794 vic,vat and friends to work.
1795 They bind socket to loopback, set ttl to zero
1796 and expect that it will work.
1797 From the viewpoint of routing cache they are broken,
1798 because we are not allowed to build multicast path
1799 with loopback source addr (look, routing cache
1800 cannot know, that ttl is zero, so that packet
1801 will not leave this host and route is valid).
1802 Luckily, this hack is good workaround.
1805 fl4->flowi4_oif = dev_out->ifindex;
1809 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1810 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1811 if (!__ip_dev_find(net, fl4->saddr, false))
1817 if (fl4->flowi4_oif) {
1818 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1819 rth = ERR_PTR(-ENODEV);
1820 if (dev_out == NULL)
1823 /* RACE: Check return value of inet_select_addr instead. */
1824 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1825 rth = ERR_PTR(-ENETUNREACH);
1828 if (ipv4_is_local_multicast(fl4->daddr) ||
1829 ipv4_is_lbcast(fl4->daddr)) {
1831 fl4->saddr = inet_select_addr(dev_out, 0,
1836 if (ipv4_is_multicast(fl4->daddr))
1837 fl4->saddr = inet_select_addr(dev_out, 0,
1839 else if (!fl4->daddr)
1840 fl4->saddr = inet_select_addr(dev_out, 0,
1846 fl4->daddr = fl4->saddr;
1848 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1849 dev_out = net->loopback_dev;
1850 fl4->flowi4_oif = net->loopback_dev->ifindex;
1851 res.type = RTN_LOCAL;
1852 flags |= RTCF_LOCAL;
1856 if (fib_lookup(net, fl4, &res)) {
1859 if (fl4->flowi4_oif) {
1860 /* Apparently, routing tables are wrong. Assume,
1861 that the destination is on link.
1864 Because we are allowed to send to iface
1865 even if it has NO routes and NO assigned
1866 addresses. When oif is specified, routing
1867 tables are looked up with only one purpose:
1868 to catch if destination is gatewayed, rather than
1869 direct. Moreover, if MSG_DONTROUTE is set,
1870 we send packet, ignoring both routing tables
1871 and ifaddr state. --ANK
1874 We could make it even if oif is unknown,
1875 likely IPv6, but we do not.
1878 if (fl4->saddr == 0)
1879 fl4->saddr = inet_select_addr(dev_out, 0,
1881 res.type = RTN_UNICAST;
1884 rth = ERR_PTR(-ENETUNREACH);
1888 if (res.type == RTN_LOCAL) {
1890 if (res.fi->fib_prefsrc)
1891 fl4->saddr = res.fi->fib_prefsrc;
1893 fl4->saddr = fl4->daddr;
1895 dev_out = net->loopback_dev;
1896 fl4->flowi4_oif = dev_out->ifindex;
1898 flags |= RTCF_LOCAL;
1902 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1903 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
1904 fib_select_multipath(&res);
1907 if (!res.prefixlen &&
1908 res.table->tb_num_default > 1 &&
1909 res.type == RTN_UNICAST && !fl4->flowi4_oif)
1910 fib_select_default(&res);
1913 fl4->saddr = FIB_RES_PREFSRC(net, res);
1915 dev_out = FIB_RES_DEV(res);
1916 fl4->flowi4_oif = dev_out->ifindex;
1920 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
1926 EXPORT_SYMBOL_GPL(__ip_route_output_key);
1928 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
1933 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
1935 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1937 return mtu ? : dst->dev->mtu;
1940 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
1941 struct sk_buff *skb, u32 mtu)
1945 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
1946 struct sk_buff *skb)
1950 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
1956 static struct dst_ops ipv4_dst_blackhole_ops = {
1958 .protocol = cpu_to_be16(ETH_P_IP),
1959 .destroy = ipv4_dst_destroy,
1960 .check = ipv4_blackhole_dst_check,
1961 .mtu = ipv4_blackhole_mtu,
1962 .default_advmss = ipv4_default_advmss,
1963 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
1964 .redirect = ipv4_rt_blackhole_redirect,
1965 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
1966 .neigh_lookup = ipv4_neigh_lookup,
1969 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
1971 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
1972 struct rtable *ort = (struct rtable *) dst_orig;
1975 struct dst_entry *new = &rt->dst;
1978 new->input = dst_discard;
1979 new->output = dst_discard;
1981 new->dev = ort->dst.dev;
1985 rt->rt_route_iif = ort->rt_route_iif;
1986 rt->rt_iif = ort->rt_iif;
1987 rt->rt_oif = ort->rt_oif;
1988 rt->rt_pmtu = ort->rt_pmtu;
1990 rt->rt_genid = rt_genid(net);
1991 rt->rt_flags = ort->rt_flags;
1992 rt->rt_type = ort->rt_type;
1993 rt->rt_gateway = ort->rt_gateway;
1996 atomic_inc(&rt->fi->fib_clntref);
2001 dst_release(dst_orig);
2003 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2006 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2009 struct rtable *rt = __ip_route_output_key(net, flp4);
2014 if (flp4->flowi4_proto)
2015 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2016 flowi4_to_flowi(flp4),
2021 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2023 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2024 struct flowi4 *fl4, struct sk_buff *skb, u32 pid,
2025 u32 seq, int event, int nowait, unsigned int flags)
2027 struct rtable *rt = skb_rtable(skb);
2029 struct nlmsghdr *nlh;
2030 unsigned long expires = 0;
2032 u32 metrics[RTAX_MAX];
2034 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2038 r = nlmsg_data(nlh);
2039 r->rtm_family = AF_INET;
2040 r->rtm_dst_len = 32;
2042 r->rtm_tos = fl4->flowi4_tos;
2043 r->rtm_table = RT_TABLE_MAIN;
2044 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2045 goto nla_put_failure;
2046 r->rtm_type = rt->rt_type;
2047 r->rtm_scope = RT_SCOPE_UNIVERSE;
2048 r->rtm_protocol = RTPROT_UNSPEC;
2049 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2050 if (rt->rt_flags & RTCF_NOTIFY)
2051 r->rtm_flags |= RTM_F_NOTIFY;
2053 if (nla_put_be32(skb, RTA_DST, dst))
2054 goto nla_put_failure;
2056 r->rtm_src_len = 32;
2057 if (nla_put_be32(skb, RTA_SRC, src))
2058 goto nla_put_failure;
2061 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2062 goto nla_put_failure;
2063 #ifdef CONFIG_IP_ROUTE_CLASSID
2064 if (rt->dst.tclassid &&
2065 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2066 goto nla_put_failure;
2068 if (!rt_is_input_route(rt) &&
2069 fl4->saddr != src) {
2070 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2071 goto nla_put_failure;
2073 if (fl4->daddr != rt->rt_gateway &&
2074 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2075 goto nla_put_failure;
2077 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2079 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2080 if (rtnetlink_put_metrics(skb, metrics) < 0)
2081 goto nla_put_failure;
2083 if (fl4->flowi4_mark &&
2084 nla_put_be32(skb, RTA_MARK, fl4->flowi4_mark))
2085 goto nla_put_failure;
2087 error = rt->dst.error;
2088 expires = rt->dst.expires;
2090 if (time_before(jiffies, expires))
2096 if (rt_is_input_route(rt)) {
2097 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2098 goto nla_put_failure;
2101 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2102 goto nla_put_failure;
2104 return nlmsg_end(skb, nlh);
2107 nlmsg_cancel(skb, nlh);
2111 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2113 struct net *net = sock_net(in_skb->sk);
2115 struct nlattr *tb[RTA_MAX+1];
2116 struct rtable *rt = NULL;
2123 struct sk_buff *skb;
2125 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2129 rtm = nlmsg_data(nlh);
2131 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2137 /* Reserve room for dummy headers, this skb can pass
2138 through good chunk of routing engine.
2140 skb_reset_mac_header(skb);
2141 skb_reset_network_header(skb);
2143 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2144 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2145 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2147 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2148 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2149 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2150 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2152 memset(&fl4, 0, sizeof(fl4));
2155 fl4.flowi4_tos = rtm->rtm_tos;
2156 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2157 fl4.flowi4_mark = mark;
2160 struct net_device *dev;
2162 dev = __dev_get_by_index(net, iif);
2168 skb->protocol = htons(ETH_P_IP);
2172 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2175 rt = skb_rtable(skb);
2176 if (err == 0 && rt->dst.error)
2177 err = -rt->dst.error;
2179 rt = ip_route_output_key(net, &fl4);
2189 skb_dst_set(skb, &rt->dst);
2190 if (rtm->rtm_flags & RTM_F_NOTIFY)
2191 rt->rt_flags |= RTCF_NOTIFY;
2193 err = rt_fill_info(net, dst, src, &fl4, skb,
2194 NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2195 RTM_NEWROUTE, 0, 0);
2199 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2208 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2213 void ip_rt_multicast_event(struct in_device *in_dev)
2215 rt_cache_flush(dev_net(in_dev->dev), 0);
2218 #ifdef CONFIG_SYSCTL
2219 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2220 void __user *buffer,
2221 size_t *lenp, loff_t *ppos)
2228 memcpy(&ctl, __ctl, sizeof(ctl));
2229 ctl.data = &flush_delay;
2230 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2232 net = (struct net *)__ctl->extra1;
2233 rt_cache_flush(net, flush_delay);
2240 static ctl_table ipv4_route_table[] = {
2242 .procname = "gc_thresh",
2243 .data = &ipv4_dst_ops.gc_thresh,
2244 .maxlen = sizeof(int),
2246 .proc_handler = proc_dointvec,
2249 .procname = "max_size",
2250 .data = &ip_rt_max_size,
2251 .maxlen = sizeof(int),
2253 .proc_handler = proc_dointvec,
2256 /* Deprecated. Use gc_min_interval_ms */
2258 .procname = "gc_min_interval",
2259 .data = &ip_rt_gc_min_interval,
2260 .maxlen = sizeof(int),
2262 .proc_handler = proc_dointvec_jiffies,
2265 .procname = "gc_min_interval_ms",
2266 .data = &ip_rt_gc_min_interval,
2267 .maxlen = sizeof(int),
2269 .proc_handler = proc_dointvec_ms_jiffies,
2272 .procname = "gc_timeout",
2273 .data = &ip_rt_gc_timeout,
2274 .maxlen = sizeof(int),
2276 .proc_handler = proc_dointvec_jiffies,
2279 .procname = "gc_interval",
2280 .data = &ip_rt_gc_interval,
2281 .maxlen = sizeof(int),
2283 .proc_handler = proc_dointvec_jiffies,
2286 .procname = "redirect_load",
2287 .data = &ip_rt_redirect_load,
2288 .maxlen = sizeof(int),
2290 .proc_handler = proc_dointvec,
2293 .procname = "redirect_number",
2294 .data = &ip_rt_redirect_number,
2295 .maxlen = sizeof(int),
2297 .proc_handler = proc_dointvec,
2300 .procname = "redirect_silence",
2301 .data = &ip_rt_redirect_silence,
2302 .maxlen = sizeof(int),
2304 .proc_handler = proc_dointvec,
2307 .procname = "error_cost",
2308 .data = &ip_rt_error_cost,
2309 .maxlen = sizeof(int),
2311 .proc_handler = proc_dointvec,
2314 .procname = "error_burst",
2315 .data = &ip_rt_error_burst,
2316 .maxlen = sizeof(int),
2318 .proc_handler = proc_dointvec,
2321 .procname = "gc_elasticity",
2322 .data = &ip_rt_gc_elasticity,
2323 .maxlen = sizeof(int),
2325 .proc_handler = proc_dointvec,
2328 .procname = "mtu_expires",
2329 .data = &ip_rt_mtu_expires,
2330 .maxlen = sizeof(int),
2332 .proc_handler = proc_dointvec_jiffies,
2335 .procname = "min_pmtu",
2336 .data = &ip_rt_min_pmtu,
2337 .maxlen = sizeof(int),
2339 .proc_handler = proc_dointvec,
2342 .procname = "min_adv_mss",
2343 .data = &ip_rt_min_advmss,
2344 .maxlen = sizeof(int),
2346 .proc_handler = proc_dointvec,
2351 static struct ctl_table ipv4_route_flush_table[] = {
2353 .procname = "flush",
2354 .maxlen = sizeof(int),
2356 .proc_handler = ipv4_sysctl_rtcache_flush,
2361 static __net_init int sysctl_route_net_init(struct net *net)
2363 struct ctl_table *tbl;
2365 tbl = ipv4_route_flush_table;
2366 if (!net_eq(net, &init_net)) {
2367 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2371 tbl[0].extra1 = net;
2373 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2374 if (net->ipv4.route_hdr == NULL)
2379 if (tbl != ipv4_route_flush_table)
2385 static __net_exit void sysctl_route_net_exit(struct net *net)
2387 struct ctl_table *tbl;
2389 tbl = net->ipv4.route_hdr->ctl_table_arg;
2390 unregister_net_sysctl_table(net->ipv4.route_hdr);
2391 BUG_ON(tbl == ipv4_route_flush_table);
2395 static __net_initdata struct pernet_operations sysctl_route_ops = {
2396 .init = sysctl_route_net_init,
2397 .exit = sysctl_route_net_exit,
2401 static __net_init int rt_genid_init(struct net *net)
2403 get_random_bytes(&net->ipv4.rt_genid,
2404 sizeof(net->ipv4.rt_genid));
2405 get_random_bytes(&net->ipv4.dev_addr_genid,
2406 sizeof(net->ipv4.dev_addr_genid));
2410 static __net_initdata struct pernet_operations rt_genid_ops = {
2411 .init = rt_genid_init,
2414 static int __net_init ipv4_inetpeer_init(struct net *net)
2416 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2420 inet_peer_base_init(bp);
2421 net->ipv4.peers = bp;
2425 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2427 struct inet_peer_base *bp = net->ipv4.peers;
2429 net->ipv4.peers = NULL;
2430 inetpeer_invalidate_tree(bp);
2434 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2435 .init = ipv4_inetpeer_init,
2436 .exit = ipv4_inetpeer_exit,
2439 #ifdef CONFIG_IP_ROUTE_CLASSID
2440 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2441 #endif /* CONFIG_IP_ROUTE_CLASSID */
2443 int __init ip_rt_init(void)
2447 #ifdef CONFIG_IP_ROUTE_CLASSID
2448 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2450 panic("IP: failed to allocate ip_rt_acct\n");
2453 ipv4_dst_ops.kmem_cachep =
2454 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2455 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2457 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2459 if (dst_entries_init(&ipv4_dst_ops) < 0)
2460 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2462 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2463 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2465 ipv4_dst_ops.gc_thresh = ~0;
2466 ip_rt_max_size = INT_MAX;
2471 if (ip_rt_proc_init())
2472 pr_err("Unable to create route proc files\n");
2475 xfrm4_init(ip_rt_max_size);
2477 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2479 #ifdef CONFIG_SYSCTL
2480 register_pernet_subsys(&sysctl_route_ops);
2482 register_pernet_subsys(&rt_genid_ops);
2483 register_pernet_subsys(&ipv4_inetpeer_ops);
2487 #ifdef CONFIG_SYSCTL
2489 * We really need to sanitize the damn ipv4 init order, then all
2490 * this nonsense will go away.
2492 void __init ip_static_sysctl_init(void)
2494 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);