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/dst_metadata.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/lwtunnel.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
111 #include <linux/kmemleak.h>
113 #include <net/secure_seq.h>
114 #include <net/ip_tunnels.h>
115 #include <net/l3mdev.h>
117 #define RT_FL_TOS(oldflp4) \
118 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
120 #define RT_GC_TIMEOUT (300*HZ)
122 static int ip_rt_max_size;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
129 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly = 256;
132 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
134 * Interface to generic destination cache.
137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
138 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
139 static unsigned int ipv4_mtu(const struct dst_entry *dst);
140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
141 static void ipv4_link_failure(struct sk_buff *skb);
142 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
143 struct sk_buff *skb, u32 mtu);
144 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb);
146 static void ipv4_dst_destroy(struct dst_entry *dst);
148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
158 static struct dst_ops ipv4_dst_ops = {
160 .check = ipv4_dst_check,
161 .default_advmss = ipv4_default_advmss,
163 .cow_metrics = ipv4_cow_metrics,
164 .destroy = ipv4_dst_destroy,
165 .negative_advice = ipv4_negative_advice,
166 .link_failure = ipv4_link_failure,
167 .update_pmtu = ip_rt_update_pmtu,
168 .redirect = ip_do_redirect,
169 .local_out = __ip_local_out,
170 .neigh_lookup = ipv4_neigh_lookup,
173 #define ECN_OR_COST(class) TC_PRIO_##class
175 const __u8 ip_tos2prio[16] = {
177 ECN_OR_COST(BESTEFFORT),
179 ECN_OR_COST(BESTEFFORT),
185 ECN_OR_COST(INTERACTIVE),
187 ECN_OR_COST(INTERACTIVE),
188 TC_PRIO_INTERACTIVE_BULK,
189 ECN_OR_COST(INTERACTIVE_BULK),
190 TC_PRIO_INTERACTIVE_BULK,
191 ECN_OR_COST(INTERACTIVE_BULK)
193 EXPORT_SYMBOL(ip_tos2prio);
195 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
196 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
198 #ifdef CONFIG_PROC_FS
199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
203 return SEQ_START_TOKEN;
206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 if (v == SEQ_START_TOKEN)
219 seq_printf(seq, "%-127s\n",
220 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
221 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
226 static const struct seq_operations rt_cache_seq_ops = {
227 .start = rt_cache_seq_start,
228 .next = rt_cache_seq_next,
229 .stop = rt_cache_seq_stop,
230 .show = rt_cache_seq_show,
233 static int rt_cache_seq_open(struct inode *inode, struct file *file)
235 return seq_open(file, &rt_cache_seq_ops);
238 static const struct file_operations rt_cache_seq_fops = {
239 .owner = THIS_MODULE,
240 .open = rt_cache_seq_open,
243 .release = seq_release,
247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
252 return SEQ_START_TOKEN;
254 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
255 if (!cpu_possible(cpu))
258 return &per_cpu(rt_cache_stat, cpu);
263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
267 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
268 if (!cpu_possible(cpu))
271 return &per_cpu(rt_cache_stat, cpu);
277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
284 struct rt_cache_stat *st = v;
286 if (v == SEQ_START_TOKEN) {
287 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");
291 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
292 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
293 dst_entries_get_slow(&ipv4_dst_ops),
306 0, /* st->gc_total */
307 0, /* st->gc_ignored */
308 0, /* st->gc_goal_miss */
309 0, /* st->gc_dst_overflow */
310 0, /* st->in_hlist_search */
311 0 /* st->out_hlist_search */
316 static const struct seq_operations rt_cpu_seq_ops = {
317 .start = rt_cpu_seq_start,
318 .next = rt_cpu_seq_next,
319 .stop = rt_cpu_seq_stop,
320 .show = rt_cpu_seq_show,
324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
326 return seq_open(file, &rt_cpu_seq_ops);
329 static const struct file_operations rt_cpu_seq_fops = {
330 .owner = THIS_MODULE,
331 .open = rt_cpu_seq_open,
334 .release = seq_release,
337 #ifdef CONFIG_IP_ROUTE_CLASSID
338 static int rt_acct_proc_show(struct seq_file *m, void *v)
340 struct ip_rt_acct *dst, *src;
343 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
347 for_each_possible_cpu(i) {
348 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
349 for (j = 0; j < 256; j++) {
350 dst[j].o_bytes += src[j].o_bytes;
351 dst[j].o_packets += src[j].o_packets;
352 dst[j].i_bytes += src[j].i_bytes;
353 dst[j].i_packets += src[j].i_packets;
357 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
362 static int rt_acct_proc_open(struct inode *inode, struct file *file)
364 return single_open(file, rt_acct_proc_show, NULL);
367 static const struct file_operations rt_acct_proc_fops = {
368 .owner = THIS_MODULE,
369 .open = rt_acct_proc_open,
372 .release = single_release,
376 static int __net_init ip_rt_do_proc_init(struct net *net)
378 struct proc_dir_entry *pde;
380 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
385 pde = proc_create("rt_cache", S_IRUGO,
386 net->proc_net_stat, &rt_cpu_seq_fops);
390 #ifdef CONFIG_IP_ROUTE_CLASSID
391 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
397 #ifdef CONFIG_IP_ROUTE_CLASSID
399 remove_proc_entry("rt_cache", net->proc_net_stat);
402 remove_proc_entry("rt_cache", net->proc_net);
407 static void __net_exit ip_rt_do_proc_exit(struct net *net)
409 remove_proc_entry("rt_cache", net->proc_net_stat);
410 remove_proc_entry("rt_cache", net->proc_net);
411 #ifdef CONFIG_IP_ROUTE_CLASSID
412 remove_proc_entry("rt_acct", net->proc_net);
416 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
417 .init = ip_rt_do_proc_init,
418 .exit = ip_rt_do_proc_exit,
421 static int __init ip_rt_proc_init(void)
423 return register_pernet_subsys(&ip_rt_proc_ops);
427 static inline int ip_rt_proc_init(void)
431 #endif /* CONFIG_PROC_FS */
433 static inline bool rt_is_expired(const struct rtable *rth)
435 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
438 void rt_cache_flush(struct net *net)
440 rt_genid_bump_ipv4(net);
443 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
447 struct net_device *dev = dst->dev;
448 const __be32 *pkey = daddr;
449 const struct rtable *rt;
452 rt = (const struct rtable *) dst;
454 pkey = (const __be32 *) &rt->rt_gateway;
456 pkey = &ip_hdr(skb)->daddr;
458 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
461 return neigh_create(&arp_tbl, pkey, dev);
464 #define IP_IDENTS_SZ 2048u
466 static atomic_t *ip_idents __read_mostly;
467 static u32 *ip_tstamps __read_mostly;
469 /* In order to protect privacy, we add a perturbation to identifiers
470 * if one generator is seldom used. This makes hard for an attacker
471 * to infer how many packets were sent between two points in time.
473 u32 ip_idents_reserve(u32 hash, int segs)
475 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
476 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
477 u32 old = ACCESS_ONCE(*p_tstamp);
478 u32 now = (u32)jiffies;
481 if (old != now && cmpxchg(p_tstamp, old, now) == old)
482 delta = prandom_u32_max(now - old);
484 return atomic_add_return(segs + delta, p_id) - segs;
486 EXPORT_SYMBOL(ip_idents_reserve);
488 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
490 static u32 ip_idents_hashrnd __read_mostly;
493 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
495 hash = jhash_3words((__force u32)iph->daddr,
496 (__force u32)iph->saddr,
497 iph->protocol ^ net_hash_mix(net),
499 id = ip_idents_reserve(hash, segs);
502 EXPORT_SYMBOL(__ip_select_ident);
504 static void __build_flow_key(struct flowi4 *fl4, struct sock *sk,
505 const struct iphdr *iph,
507 u8 prot, u32 mark, int flow_flags)
510 const struct inet_sock *inet = inet_sk(sk);
512 oif = sk->sk_bound_dev_if;
514 tos = RT_CONN_FLAGS(sk);
515 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
517 flowi4_init_output(fl4, oif, mark, tos,
518 RT_SCOPE_UNIVERSE, prot,
520 iph->daddr, iph->saddr, 0, 0,
521 sk ? sock_i_uid(sk) : GLOBAL_ROOT_UID);
524 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
527 const struct iphdr *iph = ip_hdr(skb);
528 int oif = skb->dev->ifindex;
529 u8 tos = RT_TOS(iph->tos);
530 u8 prot = iph->protocol;
531 u32 mark = skb->mark;
533 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
536 static void build_sk_flow_key(struct flowi4 *fl4, struct sock *sk)
538 const struct inet_sock *inet = inet_sk(sk);
539 const struct ip_options_rcu *inet_opt;
540 __be32 daddr = inet->inet_daddr;
543 inet_opt = rcu_dereference(inet->inet_opt);
544 if (inet_opt && inet_opt->opt.srr)
545 daddr = inet_opt->opt.faddr;
546 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
547 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
548 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
549 inet_sk_flowi_flags(sk),
550 daddr, inet->inet_saddr, 0, 0,
555 static void ip_rt_build_flow_key(struct flowi4 *fl4, struct sock *sk,
556 const struct sk_buff *skb)
559 build_skb_flow_key(fl4, skb, sk);
561 build_sk_flow_key(fl4, sk);
564 static inline void rt_free(struct rtable *rt)
566 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
569 static DEFINE_SPINLOCK(fnhe_lock);
571 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
575 rt = rcu_dereference(fnhe->fnhe_rth_input);
577 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
580 rt = rcu_dereference(fnhe->fnhe_rth_output);
582 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
587 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
589 struct fib_nh_exception *fnhe, *oldest;
591 oldest = rcu_dereference(hash->chain);
592 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
593 fnhe = rcu_dereference(fnhe->fnhe_next)) {
594 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
597 fnhe_flush_routes(oldest);
601 static inline u32 fnhe_hashfun(__be32 daddr)
603 static u32 fnhe_hashrnd __read_mostly;
606 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
607 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
608 return hash_32(hval, FNHE_HASH_SHIFT);
611 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
613 rt->rt_pmtu = fnhe->fnhe_pmtu;
614 rt->dst.expires = fnhe->fnhe_expires;
617 rt->rt_flags |= RTCF_REDIRECTED;
618 rt->rt_gateway = fnhe->fnhe_gw;
619 rt->rt_uses_gateway = 1;
623 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
624 u32 pmtu, unsigned long expires)
626 struct fnhe_hash_bucket *hash;
627 struct fib_nh_exception *fnhe;
631 u32 hval = fnhe_hashfun(daddr);
633 spin_lock_bh(&fnhe_lock);
635 hash = rcu_dereference(nh->nh_exceptions);
637 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
640 rcu_assign_pointer(nh->nh_exceptions, hash);
646 for (fnhe = rcu_dereference(hash->chain); fnhe;
647 fnhe = rcu_dereference(fnhe->fnhe_next)) {
648 if (fnhe->fnhe_daddr == daddr)
657 fnhe->fnhe_pmtu = pmtu;
658 fnhe->fnhe_expires = max(1UL, expires);
660 /* Update all cached dsts too */
661 rt = rcu_dereference(fnhe->fnhe_rth_input);
663 fill_route_from_fnhe(rt, fnhe);
664 rt = rcu_dereference(fnhe->fnhe_rth_output);
666 fill_route_from_fnhe(rt, fnhe);
668 if (depth > FNHE_RECLAIM_DEPTH)
669 fnhe = fnhe_oldest(hash);
671 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
675 fnhe->fnhe_next = hash->chain;
676 rcu_assign_pointer(hash->chain, fnhe);
678 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
679 fnhe->fnhe_daddr = daddr;
681 fnhe->fnhe_pmtu = pmtu;
682 fnhe->fnhe_expires = expires;
684 /* Exception created; mark the cached routes for the nexthop
685 * stale, so anyone caching it rechecks if this exception
688 rt = rcu_dereference(nh->nh_rth_input);
690 rt->dst.obsolete = DST_OBSOLETE_KILL;
692 for_each_possible_cpu(i) {
693 struct rtable __rcu **prt;
694 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
695 rt = rcu_dereference(*prt);
697 rt->dst.obsolete = DST_OBSOLETE_KILL;
701 fnhe->fnhe_stamp = jiffies;
704 spin_unlock_bh(&fnhe_lock);
707 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
710 __be32 new_gw = icmp_hdr(skb)->un.gateway;
711 __be32 old_gw = ip_hdr(skb)->saddr;
712 struct net_device *dev = skb->dev;
713 struct in_device *in_dev;
714 struct fib_result res;
718 switch (icmp_hdr(skb)->code & 7) {
720 case ICMP_REDIR_NETTOS:
721 case ICMP_REDIR_HOST:
722 case ICMP_REDIR_HOSTTOS:
729 if (rt->rt_gateway != old_gw)
732 in_dev = __in_dev_get_rcu(dev);
737 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
738 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
739 ipv4_is_zeronet(new_gw))
740 goto reject_redirect;
742 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
743 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
744 goto reject_redirect;
745 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
746 goto reject_redirect;
748 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
749 goto reject_redirect;
752 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
754 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
756 if (!(n->nud_state & NUD_VALID)) {
757 neigh_event_send(n, NULL);
759 if (fib_lookup(net, fl4, &res, 0) == 0) {
760 struct fib_nh *nh = &FIB_RES_NH(res);
762 update_or_create_fnhe(nh, fl4->daddr, new_gw,
763 0, jiffies + ip_rt_gc_timeout);
766 rt->dst.obsolete = DST_OBSOLETE_KILL;
767 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
774 #ifdef CONFIG_IP_ROUTE_VERBOSE
775 if (IN_DEV_LOG_MARTIANS(in_dev)) {
776 const struct iphdr *iph = (const struct iphdr *) skb->data;
777 __be32 daddr = iph->daddr;
778 __be32 saddr = iph->saddr;
780 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
781 " Advised path = %pI4 -> %pI4\n",
782 &old_gw, dev->name, &new_gw,
789 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
793 const struct iphdr *iph = (const struct iphdr *) skb->data;
794 int oif = skb->dev->ifindex;
795 u8 tos = RT_TOS(iph->tos);
796 u8 prot = iph->protocol;
797 u32 mark = skb->mark;
799 rt = (struct rtable *) dst;
801 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
802 __ip_do_redirect(rt, skb, &fl4, true);
805 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
807 struct rtable *rt = (struct rtable *)dst;
808 struct dst_entry *ret = dst;
811 if (dst->obsolete > 0) {
814 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
825 * 1. The first ip_rt_redirect_number redirects are sent
826 * with exponential backoff, then we stop sending them at all,
827 * assuming that the host ignores our redirects.
828 * 2. If we did not see packets requiring redirects
829 * during ip_rt_redirect_silence, we assume that the host
830 * forgot redirected route and start to send redirects again.
832 * This algorithm is much cheaper and more intelligent than dumb load limiting
835 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
836 * and "frag. need" (breaks PMTU discovery) in icmp.c.
839 void ip_rt_send_redirect(struct sk_buff *skb)
841 struct rtable *rt = skb_rtable(skb);
842 struct in_device *in_dev;
843 struct inet_peer *peer;
849 in_dev = __in_dev_get_rcu(rt->dst.dev);
850 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
854 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
855 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
858 net = dev_net(rt->dst.dev);
859 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
861 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
862 rt_nexthop(rt, ip_hdr(skb)->daddr));
866 /* No redirected packets during ip_rt_redirect_silence;
867 * reset the algorithm.
869 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
870 peer->rate_tokens = 0;
872 /* Too many ignored redirects; do not send anything
873 * set dst.rate_last to the last seen redirected packet.
875 if (peer->rate_tokens >= ip_rt_redirect_number) {
876 peer->rate_last = jiffies;
880 /* Check for load limit; set rate_last to the latest sent
883 if (peer->rate_tokens == 0 ||
886 (ip_rt_redirect_load << peer->rate_tokens)))) {
887 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
889 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
890 peer->rate_last = jiffies;
892 #ifdef CONFIG_IP_ROUTE_VERBOSE
894 peer->rate_tokens == ip_rt_redirect_number)
895 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
896 &ip_hdr(skb)->saddr, inet_iif(skb),
897 &ip_hdr(skb)->daddr, &gw);
904 static int ip_error(struct sk_buff *skb)
906 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
907 struct rtable *rt = skb_rtable(skb);
908 struct inet_peer *peer;
914 /* IP on this device is disabled. */
918 net = dev_net(rt->dst.dev);
919 if (!IN_DEV_FORWARD(in_dev)) {
920 switch (rt->dst.error) {
922 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
926 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
932 switch (rt->dst.error) {
937 code = ICMP_HOST_UNREACH;
940 code = ICMP_NET_UNREACH;
941 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
944 code = ICMP_PKT_FILTERED;
948 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
949 l3mdev_master_ifindex(skb->dev), 1);
954 peer->rate_tokens += now - peer->rate_last;
955 if (peer->rate_tokens > ip_rt_error_burst)
956 peer->rate_tokens = ip_rt_error_burst;
957 peer->rate_last = now;
958 if (peer->rate_tokens >= ip_rt_error_cost)
959 peer->rate_tokens -= ip_rt_error_cost;
965 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
971 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
973 struct dst_entry *dst = &rt->dst;
974 struct fib_result res;
976 if (dst_metric_locked(dst, RTAX_MTU))
979 if (ipv4_mtu(dst) < mtu)
982 if (mtu < ip_rt_min_pmtu)
983 mtu = ip_rt_min_pmtu;
985 if (rt->rt_pmtu == mtu &&
986 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
990 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
991 struct fib_nh *nh = &FIB_RES_NH(res);
993 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
994 jiffies + ip_rt_mtu_expires);
999 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1000 struct sk_buff *skb, u32 mtu)
1002 struct rtable *rt = (struct rtable *) dst;
1005 ip_rt_build_flow_key(&fl4, sk, skb);
1006 __ip_rt_update_pmtu(rt, &fl4, mtu);
1009 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1010 int oif, u32 mark, u8 protocol, int flow_flags)
1012 const struct iphdr *iph = (const struct iphdr *) skb->data;
1017 mark = IP4_REPLY_MARK(net, skb->mark);
1019 __build_flow_key(&fl4, NULL, iph, oif,
1020 RT_TOS(iph->tos), protocol, mark, flow_flags);
1021 rt = __ip_route_output_key(net, &fl4);
1023 __ip_rt_update_pmtu(rt, &fl4, mtu);
1027 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1029 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1031 const struct iphdr *iph = (const struct iphdr *) skb->data;
1035 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1037 if (!fl4.flowi4_mark)
1038 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1040 rt = __ip_route_output_key(sock_net(sk), &fl4);
1042 __ip_rt_update_pmtu(rt, &fl4, mtu);
1047 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1049 const struct iphdr *iph = (const struct iphdr *) skb->data;
1052 struct dst_entry *odst = NULL;
1057 if (!ip_sk_accept_pmtu(sk))
1060 odst = sk_dst_get(sk);
1062 if (sock_owned_by_user(sk) || !odst) {
1063 __ipv4_sk_update_pmtu(skb, sk, mtu);
1067 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1069 rt = (struct rtable *)odst;
1070 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1071 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1078 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1080 if (!dst_check(&rt->dst, 0)) {
1082 dst_release(&rt->dst);
1084 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1092 sk_dst_set(sk, &rt->dst);
1098 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1100 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1101 int oif, u32 mark, u8 protocol, int flow_flags)
1103 const struct iphdr *iph = (const struct iphdr *) skb->data;
1107 __build_flow_key(&fl4, NULL, iph, oif,
1108 RT_TOS(iph->tos), protocol, mark, flow_flags);
1109 rt = __ip_route_output_key(net, &fl4);
1111 __ip_do_redirect(rt, skb, &fl4, false);
1115 EXPORT_SYMBOL_GPL(ipv4_redirect);
1117 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1119 const struct iphdr *iph = (const struct iphdr *) skb->data;
1123 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1124 rt = __ip_route_output_key(sock_net(sk), &fl4);
1126 __ip_do_redirect(rt, skb, &fl4, false);
1130 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1132 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1134 struct rtable *rt = (struct rtable *) dst;
1136 /* All IPV4 dsts are created with ->obsolete set to the value
1137 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1138 * into this function always.
1140 * When a PMTU/redirect information update invalidates a route,
1141 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1142 * DST_OBSOLETE_DEAD by dst_free().
1144 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1149 static void ipv4_link_failure(struct sk_buff *skb)
1153 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1155 rt = skb_rtable(skb);
1157 dst_set_expires(&rt->dst, 0);
1160 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1162 pr_debug("%s: %pI4 -> %pI4, %s\n",
1163 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1164 skb->dev ? skb->dev->name : "?");
1171 We do not cache source address of outgoing interface,
1172 because it is used only by IP RR, TS and SRR options,
1173 so that it out of fast path.
1175 BTW remember: "addr" is allowed to be not aligned
1179 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1183 if (rt_is_output_route(rt))
1184 src = ip_hdr(skb)->saddr;
1186 struct fib_result res;
1192 memset(&fl4, 0, sizeof(fl4));
1193 fl4.daddr = iph->daddr;
1194 fl4.saddr = iph->saddr;
1195 fl4.flowi4_tos = RT_TOS(iph->tos);
1196 fl4.flowi4_oif = rt->dst.dev->ifindex;
1197 fl4.flowi4_iif = skb->dev->ifindex;
1198 fl4.flowi4_mark = skb->mark;
1201 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1202 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1204 src = inet_select_addr(rt->dst.dev,
1205 rt_nexthop(rt, iph->daddr),
1209 memcpy(addr, &src, 4);
1212 #ifdef CONFIG_IP_ROUTE_CLASSID
1213 static void set_class_tag(struct rtable *rt, u32 tag)
1215 if (!(rt->dst.tclassid & 0xFFFF))
1216 rt->dst.tclassid |= tag & 0xFFFF;
1217 if (!(rt->dst.tclassid & 0xFFFF0000))
1218 rt->dst.tclassid |= tag & 0xFFFF0000;
1222 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1224 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1227 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1229 if (advmss > 65535 - 40)
1230 advmss = 65535 - 40;
1235 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1237 const struct rtable *rt = (const struct rtable *) dst;
1238 unsigned int mtu = rt->rt_pmtu;
1240 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1241 mtu = dst_metric_raw(dst, RTAX_MTU);
1246 mtu = dst->dev->mtu;
1248 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1249 if (rt->rt_uses_gateway && mtu > 576)
1253 return min_t(unsigned int, mtu, IP_MAX_MTU);
1256 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1258 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1259 struct fib_nh_exception *fnhe;
1265 hval = fnhe_hashfun(daddr);
1267 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1268 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1269 if (fnhe->fnhe_daddr == daddr)
1275 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1280 spin_lock_bh(&fnhe_lock);
1282 if (daddr == fnhe->fnhe_daddr) {
1283 struct rtable __rcu **porig;
1284 struct rtable *orig;
1285 int genid = fnhe_genid(dev_net(rt->dst.dev));
1287 if (rt_is_input_route(rt))
1288 porig = &fnhe->fnhe_rth_input;
1290 porig = &fnhe->fnhe_rth_output;
1291 orig = rcu_dereference(*porig);
1293 if (fnhe->fnhe_genid != genid) {
1294 fnhe->fnhe_genid = genid;
1296 fnhe->fnhe_pmtu = 0;
1297 fnhe->fnhe_expires = 0;
1298 fnhe_flush_routes(fnhe);
1301 fill_route_from_fnhe(rt, fnhe);
1302 if (!rt->rt_gateway)
1303 rt->rt_gateway = daddr;
1305 if (!(rt->dst.flags & DST_NOCACHE)) {
1306 rcu_assign_pointer(*porig, rt);
1312 fnhe->fnhe_stamp = jiffies;
1314 spin_unlock_bh(&fnhe_lock);
1319 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1321 struct rtable *orig, *prev, **p;
1324 if (rt_is_input_route(rt)) {
1325 p = (struct rtable **)&nh->nh_rth_input;
1327 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1331 prev = cmpxchg(p, orig, rt);
1341 struct uncached_list {
1343 struct list_head head;
1346 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1348 static void rt_add_uncached_list(struct rtable *rt)
1350 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1352 rt->rt_uncached_list = ul;
1354 spin_lock_bh(&ul->lock);
1355 list_add_tail(&rt->rt_uncached, &ul->head);
1356 spin_unlock_bh(&ul->lock);
1359 static void ipv4_dst_destroy(struct dst_entry *dst)
1361 struct rtable *rt = (struct rtable *) dst;
1363 if (!list_empty(&rt->rt_uncached)) {
1364 struct uncached_list *ul = rt->rt_uncached_list;
1366 spin_lock_bh(&ul->lock);
1367 list_del(&rt->rt_uncached);
1368 spin_unlock_bh(&ul->lock);
1372 void rt_flush_dev(struct net_device *dev)
1374 struct net *net = dev_net(dev);
1378 for_each_possible_cpu(cpu) {
1379 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1381 spin_lock_bh(&ul->lock);
1382 list_for_each_entry(rt, &ul->head, rt_uncached) {
1383 if (rt->dst.dev != dev)
1385 rt->dst.dev = net->loopback_dev;
1386 dev_hold(rt->dst.dev);
1389 spin_unlock_bh(&ul->lock);
1393 static bool rt_cache_valid(const struct rtable *rt)
1396 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1400 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1401 const struct fib_result *res,
1402 struct fib_nh_exception *fnhe,
1403 struct fib_info *fi, u16 type, u32 itag)
1405 bool cached = false;
1408 struct fib_nh *nh = &FIB_RES_NH(*res);
1410 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1411 rt->rt_gateway = nh->nh_gw;
1412 rt->rt_uses_gateway = 1;
1414 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1415 #ifdef CONFIG_IP_ROUTE_CLASSID
1416 rt->dst.tclassid = nh->nh_tclassid;
1418 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1420 cached = rt_bind_exception(rt, fnhe, daddr);
1421 else if (!(rt->dst.flags & DST_NOCACHE))
1422 cached = rt_cache_route(nh, rt);
1423 if (unlikely(!cached)) {
1424 /* Routes we intend to cache in nexthop exception or
1425 * FIB nexthop have the DST_NOCACHE bit clear.
1426 * However, if we are unsuccessful at storing this
1427 * route into the cache we really need to set it.
1429 rt->dst.flags |= DST_NOCACHE;
1430 if (!rt->rt_gateway)
1431 rt->rt_gateway = daddr;
1432 rt_add_uncached_list(rt);
1435 rt_add_uncached_list(rt);
1437 #ifdef CONFIG_IP_ROUTE_CLASSID
1438 #ifdef CONFIG_IP_MULTIPLE_TABLES
1439 set_class_tag(rt, res->tclassid);
1441 set_class_tag(rt, itag);
1445 static struct rtable *rt_dst_alloc(struct net_device *dev,
1446 unsigned int flags, u16 type,
1447 bool nopolicy, bool noxfrm, bool will_cache)
1451 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1452 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1453 (nopolicy ? DST_NOPOLICY : 0) |
1454 (noxfrm ? DST_NOXFRM : 0));
1457 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1458 rt->rt_flags = flags;
1460 rt->rt_is_input = 0;
1464 rt->rt_uses_gateway = 0;
1465 rt->rt_table_id = 0;
1466 INIT_LIST_HEAD(&rt->rt_uncached);
1468 rt->dst.output = ip_output;
1469 if (flags & RTCF_LOCAL)
1470 rt->dst.input = ip_local_deliver;
1476 /* called in rcu_read_lock() section */
1477 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1478 u8 tos, struct net_device *dev, int our)
1481 struct in_device *in_dev = __in_dev_get_rcu(dev);
1482 unsigned int flags = RTCF_MULTICAST;
1486 /* Primary sanity checks. */
1491 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1492 skb->protocol != htons(ETH_P_IP))
1495 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1498 if (ipv4_is_zeronet(saddr)) {
1499 if (!ipv4_is_local_multicast(daddr))
1502 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1508 flags |= RTCF_LOCAL;
1510 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1511 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1515 #ifdef CONFIG_IP_ROUTE_CLASSID
1516 rth->dst.tclassid = itag;
1518 rth->dst.output = ip_rt_bug;
1519 rth->rt_is_input= 1;
1521 #ifdef CONFIG_IP_MROUTE
1522 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1523 rth->dst.input = ip_mr_input;
1525 RT_CACHE_STAT_INC(in_slow_mc);
1527 skb_dst_set(skb, &rth->dst);
1539 static void ip_handle_martian_source(struct net_device *dev,
1540 struct in_device *in_dev,
1541 struct sk_buff *skb,
1545 RT_CACHE_STAT_INC(in_martian_src);
1546 #ifdef CONFIG_IP_ROUTE_VERBOSE
1547 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1549 * RFC1812 recommendation, if source is martian,
1550 * the only hint is MAC header.
1552 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1553 &daddr, &saddr, dev->name);
1554 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1555 print_hex_dump(KERN_WARNING, "ll header: ",
1556 DUMP_PREFIX_OFFSET, 16, 1,
1557 skb_mac_header(skb),
1558 dev->hard_header_len, true);
1564 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1566 struct fnhe_hash_bucket *hash;
1567 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1568 u32 hval = fnhe_hashfun(daddr);
1570 spin_lock_bh(&fnhe_lock);
1572 hash = rcu_dereference_protected(nh->nh_exceptions,
1573 lockdep_is_held(&fnhe_lock));
1576 fnhe_p = &hash->chain;
1577 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1579 if (fnhe->fnhe_daddr == daddr) {
1580 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1581 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1582 fnhe_flush_routes(fnhe);
1583 kfree_rcu(fnhe, rcu);
1586 fnhe_p = &fnhe->fnhe_next;
1587 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1588 lockdep_is_held(&fnhe_lock));
1591 spin_unlock_bh(&fnhe_lock);
1594 /* called in rcu_read_lock() section */
1595 static int __mkroute_input(struct sk_buff *skb,
1596 const struct fib_result *res,
1597 struct in_device *in_dev,
1598 __be32 daddr, __be32 saddr, u32 tos)
1600 struct fib_nh_exception *fnhe;
1603 struct in_device *out_dev;
1607 /* get a working reference to the output device */
1608 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1610 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1614 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1615 in_dev->dev, in_dev, &itag);
1617 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1623 do_cache = res->fi && !itag;
1624 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1625 skb->protocol == htons(ETH_P_IP) &&
1626 (IN_DEV_SHARED_MEDIA(out_dev) ||
1627 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1628 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1630 if (skb->protocol != htons(ETH_P_IP)) {
1631 /* Not IP (i.e. ARP). Do not create route, if it is
1632 * invalid for proxy arp. DNAT routes are always valid.
1634 * Proxy arp feature have been extended to allow, ARP
1635 * replies back to the same interface, to support
1636 * Private VLAN switch technologies. See arp.c.
1638 if (out_dev == in_dev &&
1639 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1645 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1648 rth = rcu_dereference(fnhe->fnhe_rth_input);
1649 if (rth && rth->dst.expires &&
1650 time_after(jiffies, rth->dst.expires)) {
1651 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1658 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1661 if (rt_cache_valid(rth)) {
1662 skb_dst_set_noref(skb, &rth->dst);
1667 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1668 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1669 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1675 rth->rt_is_input = 1;
1677 rth->rt_table_id = res->table->tb_id;
1678 RT_CACHE_STAT_INC(in_slow_tot);
1680 rth->dst.input = ip_forward;
1682 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1683 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1684 rth->dst.lwtstate->orig_output = rth->dst.output;
1685 rth->dst.output = lwtunnel_output;
1687 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1688 rth->dst.lwtstate->orig_input = rth->dst.input;
1689 rth->dst.input = lwtunnel_input;
1691 skb_dst_set(skb, &rth->dst);
1698 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1700 /* To make ICMP packets follow the right flow, the multipath hash is
1701 * calculated from the inner IP addresses in reverse order.
1703 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1705 const struct iphdr *outer_iph = ip_hdr(skb);
1706 struct icmphdr _icmph;
1707 const struct icmphdr *icmph;
1708 struct iphdr _inner_iph;
1709 const struct iphdr *inner_iph;
1711 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1714 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1719 if (icmph->type != ICMP_DEST_UNREACH &&
1720 icmph->type != ICMP_REDIRECT &&
1721 icmph->type != ICMP_TIME_EXCEEDED &&
1722 icmph->type != ICMP_PARAMETERPROB) {
1726 inner_iph = skb_header_pointer(skb,
1727 outer_iph->ihl * 4 + sizeof(_icmph),
1728 sizeof(_inner_iph), &_inner_iph);
1732 return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1735 return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1738 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1740 static int ip_mkroute_input(struct sk_buff *skb,
1741 struct fib_result *res,
1742 const struct flowi4 *fl4,
1743 struct in_device *in_dev,
1744 __be32 daddr, __be32 saddr, u32 tos)
1746 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1747 if (res->fi && res->fi->fib_nhs > 1) {
1750 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1751 h = ip_multipath_icmp_hash(skb);
1753 h = fib_multipath_hash(saddr, daddr);
1754 fib_select_multipath(res, h);
1758 /* create a routing cache entry */
1759 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1763 * NOTE. We drop all the packets that has local source
1764 * addresses, because every properly looped back packet
1765 * must have correct destination already attached by output routine.
1767 * Such approach solves two big problems:
1768 * 1. Not simplex devices are handled properly.
1769 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1770 * called with rcu_read_lock()
1773 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1774 u8 tos, struct net_device *dev)
1776 struct fib_result res;
1777 struct in_device *in_dev = __in_dev_get_rcu(dev);
1778 struct ip_tunnel_info *tun_info;
1780 unsigned int flags = 0;
1784 struct net *net = dev_net(dev);
1787 /* IP on this device is disabled. */
1792 /* Check for the most weird martians, which can be not detected
1796 tun_info = skb_tunnel_info(skb);
1797 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1798 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1800 fl4.flowi4_tun_key.tun_id = 0;
1803 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1804 goto martian_source;
1808 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1811 /* Accept zero addresses only to limited broadcast;
1812 * I even do not know to fix it or not. Waiting for complains :-)
1814 if (ipv4_is_zeronet(saddr))
1815 goto martian_source;
1817 if (ipv4_is_zeronet(daddr))
1818 goto martian_destination;
1820 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1821 * and call it once if daddr or/and saddr are loopback addresses
1823 if (ipv4_is_loopback(daddr)) {
1824 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1825 goto martian_destination;
1826 } else if (ipv4_is_loopback(saddr)) {
1827 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1828 goto martian_source;
1832 * Now we are ready to route packet.
1835 fl4.flowi4_iif = l3mdev_fib_oif_rcu(dev);
1836 fl4.flowi4_mark = skb->mark;
1837 fl4.flowi4_tos = tos;
1838 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1839 fl4.flowi4_flags = 0;
1842 err = fib_lookup(net, &fl4, &res, 0);
1844 if (!IN_DEV_FORWARD(in_dev))
1845 err = -EHOSTUNREACH;
1849 if (res.type == RTN_BROADCAST)
1852 if (res.type == RTN_LOCAL) {
1853 err = fib_validate_source(skb, saddr, daddr, tos,
1854 0, dev, in_dev, &itag);
1856 goto martian_source;
1860 if (!IN_DEV_FORWARD(in_dev)) {
1861 err = -EHOSTUNREACH;
1864 if (res.type != RTN_UNICAST)
1865 goto martian_destination;
1867 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1871 if (skb->protocol != htons(ETH_P_IP))
1874 if (!ipv4_is_zeronet(saddr)) {
1875 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1878 goto martian_source;
1880 flags |= RTCF_BROADCAST;
1881 res.type = RTN_BROADCAST;
1882 RT_CACHE_STAT_INC(in_brd);
1888 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1889 if (rt_cache_valid(rth)) {
1890 skb_dst_set_noref(skb, &rth->dst);
1898 rth = rt_dst_alloc(net->loopback_dev, flags | RTCF_LOCAL, res.type,
1899 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1903 rth->dst.output= ip_rt_bug;
1904 #ifdef CONFIG_IP_ROUTE_CLASSID
1905 rth->dst.tclassid = itag;
1907 rth->rt_is_input = 1;
1909 rth->rt_table_id = res.table->tb_id;
1911 RT_CACHE_STAT_INC(in_slow_tot);
1912 if (res.type == RTN_UNREACHABLE) {
1913 rth->dst.input= ip_error;
1914 rth->dst.error= -err;
1915 rth->rt_flags &= ~RTCF_LOCAL;
1918 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1919 rth->dst.flags |= DST_NOCACHE;
1920 rt_add_uncached_list(rth);
1923 skb_dst_set(skb, &rth->dst);
1928 RT_CACHE_STAT_INC(in_no_route);
1929 res.type = RTN_UNREACHABLE;
1935 * Do not cache martian addresses: they should be logged (RFC1812)
1937 martian_destination:
1938 RT_CACHE_STAT_INC(in_martian_dst);
1939 #ifdef CONFIG_IP_ROUTE_VERBOSE
1940 if (IN_DEV_LOG_MARTIANS(in_dev))
1941 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1942 &daddr, &saddr, dev->name);
1954 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1958 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1959 u8 tos, struct net_device *dev)
1965 /* Multicast recognition logic is moved from route cache to here.
1966 The problem was that too many Ethernet cards have broken/missing
1967 hardware multicast filters :-( As result the host on multicasting
1968 network acquires a lot of useless route cache entries, sort of
1969 SDR messages from all the world. Now we try to get rid of them.
1970 Really, provided software IP multicast filter is organized
1971 reasonably (at least, hashed), it does not result in a slowdown
1972 comparing with route cache reject entries.
1973 Note, that multicast routers are not affected, because
1974 route cache entry is created eventually.
1976 if (ipv4_is_multicast(daddr)) {
1977 struct in_device *in_dev = __in_dev_get_rcu(dev);
1980 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1981 ip_hdr(skb)->protocol);
1983 #ifdef CONFIG_IP_MROUTE
1985 (!ipv4_is_local_multicast(daddr) &&
1986 IN_DEV_MFORWARD(in_dev))
1989 int res = ip_route_input_mc(skb, daddr, saddr,
1998 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2002 EXPORT_SYMBOL(ip_route_input_noref);
2004 /* called with rcu_read_lock() */
2005 static struct rtable *__mkroute_output(const struct fib_result *res,
2006 const struct flowi4 *fl4, int orig_oif,
2007 struct net_device *dev_out,
2010 struct fib_info *fi = res->fi;
2011 struct fib_nh_exception *fnhe;
2012 struct in_device *in_dev;
2013 u16 type = res->type;
2017 in_dev = __in_dev_get_rcu(dev_out);
2019 return ERR_PTR(-EINVAL);
2021 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2022 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2023 return ERR_PTR(-EINVAL);
2025 if (ipv4_is_lbcast(fl4->daddr))
2026 type = RTN_BROADCAST;
2027 else if (ipv4_is_multicast(fl4->daddr))
2028 type = RTN_MULTICAST;
2029 else if (ipv4_is_zeronet(fl4->daddr))
2030 return ERR_PTR(-EINVAL);
2032 if (dev_out->flags & IFF_LOOPBACK)
2033 flags |= RTCF_LOCAL;
2036 if (type == RTN_BROADCAST) {
2037 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2039 } else if (type == RTN_MULTICAST) {
2040 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2041 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2043 flags &= ~RTCF_LOCAL;
2046 /* If multicast route do not exist use
2047 * default one, but do not gateway in this case.
2050 if (fi && res->prefixlen < 4)
2052 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2053 (orig_oif != dev_out->ifindex)) {
2054 /* For local routes that require a particular output interface
2055 * we do not want to cache the result. Caching the result
2056 * causes incorrect behaviour when there are multiple source
2057 * addresses on the interface, the end result being that if the
2058 * intended recipient is waiting on that interface for the
2059 * packet he won't receive it because it will be delivered on
2060 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2061 * be set to the loopback interface as well.
2067 do_cache &= fi != NULL;
2069 struct rtable __rcu **prth;
2070 struct fib_nh *nh = &FIB_RES_NH(*res);
2072 fnhe = find_exception(nh, fl4->daddr);
2074 prth = &fnhe->fnhe_rth_output;
2075 rth = rcu_dereference(*prth);
2076 if (rth && rth->dst.expires &&
2077 time_after(jiffies, rth->dst.expires)) {
2078 ip_del_fnhe(nh, fl4->daddr);
2085 if (unlikely(fl4->flowi4_flags &
2086 FLOWI_FLAG_KNOWN_NH &&
2088 nh->nh_scope == RT_SCOPE_LINK))) {
2092 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2093 rth = rcu_dereference(*prth);
2096 if (rt_cache_valid(rth)) {
2097 dst_hold(&rth->dst);
2103 rth = rt_dst_alloc(dev_out, flags, type,
2104 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2105 IN_DEV_CONF_GET(in_dev, NOXFRM),
2108 return ERR_PTR(-ENOBUFS);
2110 rth->rt_iif = orig_oif ? : 0;
2112 rth->rt_table_id = res->table->tb_id;
2114 RT_CACHE_STAT_INC(out_slow_tot);
2116 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2117 if (flags & RTCF_LOCAL &&
2118 !(dev_out->flags & IFF_LOOPBACK)) {
2119 rth->dst.output = ip_mc_output;
2120 RT_CACHE_STAT_INC(out_slow_mc);
2122 #ifdef CONFIG_IP_MROUTE
2123 if (type == RTN_MULTICAST) {
2124 if (IN_DEV_MFORWARD(in_dev) &&
2125 !ipv4_is_local_multicast(fl4->daddr)) {
2126 rth->dst.input = ip_mr_input;
2127 rth->dst.output = ip_mc_output;
2133 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2134 if (lwtunnel_output_redirect(rth->dst.lwtstate))
2135 rth->dst.output = lwtunnel_output;
2141 * Major route resolver routine.
2144 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2147 struct net_device *dev_out = NULL;
2148 __u8 tos = RT_FL_TOS(fl4);
2149 unsigned int flags = 0;
2150 struct fib_result res;
2153 int err = -ENETUNREACH;
2159 orig_oif = fl4->flowi4_oif;
2161 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2162 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2163 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2164 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2168 rth = ERR_PTR(-EINVAL);
2169 if (ipv4_is_multicast(fl4->saddr) ||
2170 ipv4_is_lbcast(fl4->saddr) ||
2171 ipv4_is_zeronet(fl4->saddr))
2174 /* I removed check for oif == dev_out->oif here.
2175 It was wrong for two reasons:
2176 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2177 is assigned to multiple interfaces.
2178 2. Moreover, we are allowed to send packets with saddr
2179 of another iface. --ANK
2182 if (fl4->flowi4_oif == 0 &&
2183 (ipv4_is_multicast(fl4->daddr) ||
2184 ipv4_is_lbcast(fl4->daddr))) {
2185 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2186 dev_out = __ip_dev_find(net, fl4->saddr, false);
2190 /* Special hack: user can direct multicasts
2191 and limited broadcast via necessary interface
2192 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2193 This hack is not just for fun, it allows
2194 vic,vat and friends to work.
2195 They bind socket to loopback, set ttl to zero
2196 and expect that it will work.
2197 From the viewpoint of routing cache they are broken,
2198 because we are not allowed to build multicast path
2199 with loopback source addr (look, routing cache
2200 cannot know, that ttl is zero, so that packet
2201 will not leave this host and route is valid).
2202 Luckily, this hack is good workaround.
2205 fl4->flowi4_oif = dev_out->ifindex;
2209 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2210 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2211 if (!__ip_dev_find(net, fl4->saddr, false))
2217 if (fl4->flowi4_oif) {
2218 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2219 rth = ERR_PTR(-ENODEV);
2223 /* RACE: Check return value of inet_select_addr instead. */
2224 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2225 rth = ERR_PTR(-ENETUNREACH);
2228 if (ipv4_is_local_multicast(fl4->daddr) ||
2229 ipv4_is_lbcast(fl4->daddr) ||
2230 fl4->flowi4_proto == IPPROTO_IGMP) {
2232 fl4->saddr = inet_select_addr(dev_out, 0,
2237 if (ipv4_is_multicast(fl4->daddr))
2238 fl4->saddr = inet_select_addr(dev_out, 0,
2240 else if (!fl4->daddr)
2241 fl4->saddr = inet_select_addr(dev_out, 0,
2245 rth = l3mdev_get_rtable(dev_out, fl4);
2251 fl4->daddr = fl4->saddr;
2253 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2254 dev_out = net->loopback_dev;
2255 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2256 res.type = RTN_LOCAL;
2257 flags |= RTCF_LOCAL;
2261 err = fib_lookup(net, fl4, &res, 0);
2265 if (fl4->flowi4_oif &&
2266 !netif_index_is_l3_master(net, fl4->flowi4_oif)) {
2267 /* Apparently, routing tables are wrong. Assume,
2268 that the destination is on link.
2271 Because we are allowed to send to iface
2272 even if it has NO routes and NO assigned
2273 addresses. When oif is specified, routing
2274 tables are looked up with only one purpose:
2275 to catch if destination is gatewayed, rather than
2276 direct. Moreover, if MSG_DONTROUTE is set,
2277 we send packet, ignoring both routing tables
2278 and ifaddr state. --ANK
2281 We could make it even if oif is unknown,
2282 likely IPv6, but we do not.
2285 if (fl4->saddr == 0)
2286 fl4->saddr = inet_select_addr(dev_out, 0,
2288 res.type = RTN_UNICAST;
2295 if (res.type == RTN_LOCAL) {
2297 if (res.fi->fib_prefsrc)
2298 fl4->saddr = res.fi->fib_prefsrc;
2300 fl4->saddr = fl4->daddr;
2302 dev_out = net->loopback_dev;
2303 fl4->flowi4_oif = dev_out->ifindex;
2304 flags |= RTCF_LOCAL;
2308 fib_select_path(net, &res, fl4, mp_hash);
2310 dev_out = FIB_RES_DEV(res);
2311 fl4->flowi4_oif = dev_out->ifindex;
2315 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2321 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2323 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2328 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2330 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2332 return mtu ? : dst->dev->mtu;
2335 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2336 struct sk_buff *skb, u32 mtu)
2340 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2341 struct sk_buff *skb)
2345 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2351 static struct dst_ops ipv4_dst_blackhole_ops = {
2353 .check = ipv4_blackhole_dst_check,
2354 .mtu = ipv4_blackhole_mtu,
2355 .default_advmss = ipv4_default_advmss,
2356 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2357 .redirect = ipv4_rt_blackhole_redirect,
2358 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2359 .neigh_lookup = ipv4_neigh_lookup,
2362 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2364 struct rtable *ort = (struct rtable *) dst_orig;
2367 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2369 struct dst_entry *new = &rt->dst;
2372 new->input = dst_discard;
2373 new->output = dst_discard_out;
2375 new->dev = ort->dst.dev;
2379 rt->rt_is_input = ort->rt_is_input;
2380 rt->rt_iif = ort->rt_iif;
2381 rt->rt_pmtu = ort->rt_pmtu;
2383 rt->rt_genid = rt_genid_ipv4(net);
2384 rt->rt_flags = ort->rt_flags;
2385 rt->rt_type = ort->rt_type;
2386 rt->rt_gateway = ort->rt_gateway;
2387 rt->rt_uses_gateway = ort->rt_uses_gateway;
2389 INIT_LIST_HEAD(&rt->rt_uncached);
2393 dst_release(dst_orig);
2395 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2398 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2399 const struct sock *sk)
2401 struct rtable *rt = __ip_route_output_key(net, flp4);
2406 if (flp4->flowi4_proto)
2407 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2408 flowi4_to_flowi(flp4),
2413 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2415 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2416 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2417 u32 seq, int event, int nowait, unsigned int flags)
2419 struct rtable *rt = skb_rtable(skb);
2421 struct nlmsghdr *nlh;
2422 unsigned long expires = 0;
2424 u32 metrics[RTAX_MAX];
2426 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2430 r = nlmsg_data(nlh);
2431 r->rtm_family = AF_INET;
2432 r->rtm_dst_len = 32;
2434 r->rtm_tos = fl4->flowi4_tos;
2435 r->rtm_table = table_id;
2436 if (nla_put_u32(skb, RTA_TABLE, table_id))
2437 goto nla_put_failure;
2438 r->rtm_type = rt->rt_type;
2439 r->rtm_scope = RT_SCOPE_UNIVERSE;
2440 r->rtm_protocol = RTPROT_UNSPEC;
2441 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2442 if (rt->rt_flags & RTCF_NOTIFY)
2443 r->rtm_flags |= RTM_F_NOTIFY;
2444 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2445 r->rtm_flags |= RTCF_DOREDIRECT;
2447 if (nla_put_in_addr(skb, RTA_DST, dst))
2448 goto nla_put_failure;
2450 r->rtm_src_len = 32;
2451 if (nla_put_in_addr(skb, RTA_SRC, src))
2452 goto nla_put_failure;
2455 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2456 goto nla_put_failure;
2457 #ifdef CONFIG_IP_ROUTE_CLASSID
2458 if (rt->dst.tclassid &&
2459 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2460 goto nla_put_failure;
2462 if (!rt_is_input_route(rt) &&
2463 fl4->saddr != src) {
2464 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2465 goto nla_put_failure;
2467 if (rt->rt_uses_gateway &&
2468 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2469 goto nla_put_failure;
2471 expires = rt->dst.expires;
2473 unsigned long now = jiffies;
2475 if (time_before(now, expires))
2481 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2482 if (rt->rt_pmtu && expires)
2483 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2484 if (rtnetlink_put_metrics(skb, metrics) < 0)
2485 goto nla_put_failure;
2487 if (fl4->flowi4_mark &&
2488 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2489 goto nla_put_failure;
2491 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2492 nla_put_u32(skb, RTA_UID,
2493 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2494 goto nla_put_failure;
2496 error = rt->dst.error;
2498 if (rt_is_input_route(rt)) {
2499 #ifdef CONFIG_IP_MROUTE
2500 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2501 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2502 int err = ipmr_get_route(net, skb,
2503 fl4->saddr, fl4->daddr,
2510 goto nla_put_failure;
2512 if (err == -EMSGSIZE)
2513 goto nla_put_failure;
2519 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2520 goto nla_put_failure;
2523 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2524 goto nla_put_failure;
2526 nlmsg_end(skb, nlh);
2530 nlmsg_cancel(skb, nlh);
2534 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2536 struct net *net = sock_net(in_skb->sk);
2538 struct nlattr *tb[RTA_MAX+1];
2539 struct rtable *rt = NULL;
2546 struct sk_buff *skb;
2547 u32 table_id = RT_TABLE_MAIN;
2550 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2554 rtm = nlmsg_data(nlh);
2556 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2562 /* Reserve room for dummy headers, this skb can pass
2563 through good chunk of routing engine.
2565 skb_reset_mac_header(skb);
2566 skb_reset_network_header(skb);
2568 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2569 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2570 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2572 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2573 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2574 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2575 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2577 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2579 uid = (iif ? INVALID_UID : current_uid());
2581 memset(&fl4, 0, sizeof(fl4));
2584 fl4.flowi4_tos = rtm->rtm_tos;
2585 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2586 fl4.flowi4_mark = mark;
2587 fl4.flowi4_uid = uid;
2589 if (netif_index_is_l3_master(net, fl4.flowi4_oif))
2590 fl4.flowi4_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF;
2593 struct net_device *dev;
2595 dev = __dev_get_by_index(net, iif);
2601 skb->protocol = htons(ETH_P_IP);
2605 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2608 rt = skb_rtable(skb);
2609 if (err == 0 && rt->dst.error)
2610 err = -rt->dst.error;
2612 rt = ip_route_output_key(net, &fl4);
2622 skb_dst_set(skb, &rt->dst);
2623 if (rtm->rtm_flags & RTM_F_NOTIFY)
2624 rt->rt_flags |= RTCF_NOTIFY;
2626 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2627 table_id = rt->rt_table_id;
2629 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2630 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2631 RTM_NEWROUTE, 0, 0);
2635 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2644 void ip_rt_multicast_event(struct in_device *in_dev)
2646 rt_cache_flush(dev_net(in_dev->dev));
2649 #ifdef CONFIG_SYSCTL
2650 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2651 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2652 static int ip_rt_gc_elasticity __read_mostly = 8;
2654 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2655 void __user *buffer,
2656 size_t *lenp, loff_t *ppos)
2658 struct net *net = (struct net *)__ctl->extra1;
2661 rt_cache_flush(net);
2662 fnhe_genid_bump(net);
2669 static struct ctl_table ipv4_route_table[] = {
2671 .procname = "gc_thresh",
2672 .data = &ipv4_dst_ops.gc_thresh,
2673 .maxlen = sizeof(int),
2675 .proc_handler = proc_dointvec,
2678 .procname = "max_size",
2679 .data = &ip_rt_max_size,
2680 .maxlen = sizeof(int),
2682 .proc_handler = proc_dointvec,
2685 /* Deprecated. Use gc_min_interval_ms */
2687 .procname = "gc_min_interval",
2688 .data = &ip_rt_gc_min_interval,
2689 .maxlen = sizeof(int),
2691 .proc_handler = proc_dointvec_jiffies,
2694 .procname = "gc_min_interval_ms",
2695 .data = &ip_rt_gc_min_interval,
2696 .maxlen = sizeof(int),
2698 .proc_handler = proc_dointvec_ms_jiffies,
2701 .procname = "gc_timeout",
2702 .data = &ip_rt_gc_timeout,
2703 .maxlen = sizeof(int),
2705 .proc_handler = proc_dointvec_jiffies,
2708 .procname = "gc_interval",
2709 .data = &ip_rt_gc_interval,
2710 .maxlen = sizeof(int),
2712 .proc_handler = proc_dointvec_jiffies,
2715 .procname = "redirect_load",
2716 .data = &ip_rt_redirect_load,
2717 .maxlen = sizeof(int),
2719 .proc_handler = proc_dointvec,
2722 .procname = "redirect_number",
2723 .data = &ip_rt_redirect_number,
2724 .maxlen = sizeof(int),
2726 .proc_handler = proc_dointvec,
2729 .procname = "redirect_silence",
2730 .data = &ip_rt_redirect_silence,
2731 .maxlen = sizeof(int),
2733 .proc_handler = proc_dointvec,
2736 .procname = "error_cost",
2737 .data = &ip_rt_error_cost,
2738 .maxlen = sizeof(int),
2740 .proc_handler = proc_dointvec,
2743 .procname = "error_burst",
2744 .data = &ip_rt_error_burst,
2745 .maxlen = sizeof(int),
2747 .proc_handler = proc_dointvec,
2750 .procname = "gc_elasticity",
2751 .data = &ip_rt_gc_elasticity,
2752 .maxlen = sizeof(int),
2754 .proc_handler = proc_dointvec,
2757 .procname = "mtu_expires",
2758 .data = &ip_rt_mtu_expires,
2759 .maxlen = sizeof(int),
2761 .proc_handler = proc_dointvec_jiffies,
2764 .procname = "min_pmtu",
2765 .data = &ip_rt_min_pmtu,
2766 .maxlen = sizeof(int),
2768 .proc_handler = proc_dointvec,
2771 .procname = "min_adv_mss",
2772 .data = &ip_rt_min_advmss,
2773 .maxlen = sizeof(int),
2775 .proc_handler = proc_dointvec,
2780 static struct ctl_table ipv4_route_flush_table[] = {
2782 .procname = "flush",
2783 .maxlen = sizeof(int),
2785 .proc_handler = ipv4_sysctl_rtcache_flush,
2790 static __net_init int sysctl_route_net_init(struct net *net)
2792 struct ctl_table *tbl;
2794 tbl = ipv4_route_flush_table;
2795 if (!net_eq(net, &init_net)) {
2796 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2800 /* Don't export sysctls to unprivileged users */
2801 if (net->user_ns != &init_user_ns)
2802 tbl[0].procname = NULL;
2804 tbl[0].extra1 = net;
2806 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2807 if (!net->ipv4.route_hdr)
2812 if (tbl != ipv4_route_flush_table)
2818 static __net_exit void sysctl_route_net_exit(struct net *net)
2820 struct ctl_table *tbl;
2822 tbl = net->ipv4.route_hdr->ctl_table_arg;
2823 unregister_net_sysctl_table(net->ipv4.route_hdr);
2824 BUG_ON(tbl == ipv4_route_flush_table);
2828 static __net_initdata struct pernet_operations sysctl_route_ops = {
2829 .init = sysctl_route_net_init,
2830 .exit = sysctl_route_net_exit,
2834 static __net_init int rt_genid_init(struct net *net)
2836 atomic_set(&net->ipv4.rt_genid, 0);
2837 atomic_set(&net->fnhe_genid, 0);
2838 get_random_bytes(&net->ipv4.dev_addr_genid,
2839 sizeof(net->ipv4.dev_addr_genid));
2843 static __net_initdata struct pernet_operations rt_genid_ops = {
2844 .init = rt_genid_init,
2847 static int __net_init ipv4_inetpeer_init(struct net *net)
2849 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2853 inet_peer_base_init(bp);
2854 net->ipv4.peers = bp;
2858 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2860 struct inet_peer_base *bp = net->ipv4.peers;
2862 net->ipv4.peers = NULL;
2863 inetpeer_invalidate_tree(bp);
2867 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2868 .init = ipv4_inetpeer_init,
2869 .exit = ipv4_inetpeer_exit,
2872 #ifdef CONFIG_IP_ROUTE_CLASSID
2873 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2874 #endif /* CONFIG_IP_ROUTE_CLASSID */
2876 int __init ip_rt_init(void)
2881 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2883 panic("IP: failed to allocate ip_idents\n");
2885 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2887 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2889 panic("IP: failed to allocate ip_tstamps\n");
2891 for_each_possible_cpu(cpu) {
2892 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2894 INIT_LIST_HEAD(&ul->head);
2895 spin_lock_init(&ul->lock);
2897 #ifdef CONFIG_IP_ROUTE_CLASSID
2898 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2900 panic("IP: failed to allocate ip_rt_acct\n");
2903 ipv4_dst_ops.kmem_cachep =
2904 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2905 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2907 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2909 if (dst_entries_init(&ipv4_dst_ops) < 0)
2910 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2912 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2913 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2915 ipv4_dst_ops.gc_thresh = ~0;
2916 ip_rt_max_size = INT_MAX;
2921 if (ip_rt_proc_init())
2922 pr_err("Unable to create route proc files\n");
2927 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2929 #ifdef CONFIG_SYSCTL
2930 register_pernet_subsys(&sysctl_route_ops);
2932 register_pernet_subsys(&rt_genid_ops);
2933 register_pernet_subsys(&ipv4_inetpeer_ops);
2937 #ifdef CONFIG_SYSCTL
2939 * We really need to sanitize the damn ipv4 init order, then all
2940 * this nonsense will go away.
2942 void __init ip_static_sysctl_init(void)
2944 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);