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, NULL, &new_gw);
754 if (!(n->nud_state & NUD_VALID)) {
755 neigh_event_send(n, NULL);
757 if (fib_lookup(net, fl4, &res, 0) == 0) {
758 struct fib_nh *nh = &FIB_RES_NH(res);
760 update_or_create_fnhe(nh, fl4->daddr, new_gw,
761 0, jiffies + ip_rt_gc_timeout);
764 rt->dst.obsolete = DST_OBSOLETE_KILL;
765 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
772 #ifdef CONFIG_IP_ROUTE_VERBOSE
773 if (IN_DEV_LOG_MARTIANS(in_dev)) {
774 const struct iphdr *iph = (const struct iphdr *) skb->data;
775 __be32 daddr = iph->daddr;
776 __be32 saddr = iph->saddr;
778 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
779 " Advised path = %pI4 -> %pI4\n",
780 &old_gw, dev->name, &new_gw,
787 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
791 const struct iphdr *iph = (const struct iphdr *) skb->data;
792 int oif = skb->dev->ifindex;
793 u8 tos = RT_TOS(iph->tos);
794 u8 prot = iph->protocol;
795 u32 mark = skb->mark;
797 rt = (struct rtable *) dst;
799 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
800 __ip_do_redirect(rt, skb, &fl4, true);
803 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
805 struct rtable *rt = (struct rtable *)dst;
806 struct dst_entry *ret = dst;
809 if (dst->obsolete > 0) {
812 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
823 * 1. The first ip_rt_redirect_number redirects are sent
824 * with exponential backoff, then we stop sending them at all,
825 * assuming that the host ignores our redirects.
826 * 2. If we did not see packets requiring redirects
827 * during ip_rt_redirect_silence, we assume that the host
828 * forgot redirected route and start to send redirects again.
830 * This algorithm is much cheaper and more intelligent than dumb load limiting
833 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
834 * and "frag. need" (breaks PMTU discovery) in icmp.c.
837 void ip_rt_send_redirect(struct sk_buff *skb)
839 struct rtable *rt = skb_rtable(skb);
840 struct in_device *in_dev;
841 struct inet_peer *peer;
847 in_dev = __in_dev_get_rcu(rt->dst.dev);
848 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
852 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
853 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
856 net = dev_net(rt->dst.dev);
857 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
859 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
860 rt_nexthop(rt, ip_hdr(skb)->daddr));
864 /* No redirected packets during ip_rt_redirect_silence;
865 * reset the algorithm.
867 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
868 peer->rate_tokens = 0;
870 /* Too many ignored redirects; do not send anything
871 * set dst.rate_last to the last seen redirected packet.
873 if (peer->rate_tokens >= ip_rt_redirect_number) {
874 peer->rate_last = jiffies;
878 /* Check for load limit; set rate_last to the latest sent
881 if (peer->rate_tokens == 0 ||
884 (ip_rt_redirect_load << peer->rate_tokens)))) {
885 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
887 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
888 peer->rate_last = jiffies;
890 #ifdef CONFIG_IP_ROUTE_VERBOSE
892 peer->rate_tokens == ip_rt_redirect_number)
893 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
894 &ip_hdr(skb)->saddr, inet_iif(skb),
895 &ip_hdr(skb)->daddr, &gw);
902 static int ip_error(struct sk_buff *skb)
904 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
905 struct rtable *rt = skb_rtable(skb);
906 struct inet_peer *peer;
912 /* IP on this device is disabled. */
916 net = dev_net(rt->dst.dev);
917 if (!IN_DEV_FORWARD(in_dev)) {
918 switch (rt->dst.error) {
920 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
924 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
930 switch (rt->dst.error) {
935 code = ICMP_HOST_UNREACH;
938 code = ICMP_NET_UNREACH;
939 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
942 code = ICMP_PKT_FILTERED;
946 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
947 l3mdev_master_ifindex(skb->dev), 1);
952 peer->rate_tokens += now - peer->rate_last;
953 if (peer->rate_tokens > ip_rt_error_burst)
954 peer->rate_tokens = ip_rt_error_burst;
955 peer->rate_last = now;
956 if (peer->rate_tokens >= ip_rt_error_cost)
957 peer->rate_tokens -= ip_rt_error_cost;
963 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
969 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
971 struct dst_entry *dst = &rt->dst;
972 struct fib_result res;
974 if (dst_metric_locked(dst, RTAX_MTU))
977 if (ipv4_mtu(dst) < mtu)
980 if (mtu < ip_rt_min_pmtu)
981 mtu = ip_rt_min_pmtu;
983 if (rt->rt_pmtu == mtu &&
984 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
988 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
989 struct fib_nh *nh = &FIB_RES_NH(res);
991 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
992 jiffies + ip_rt_mtu_expires);
997 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
998 struct sk_buff *skb, u32 mtu)
1000 struct rtable *rt = (struct rtable *) dst;
1003 ip_rt_build_flow_key(&fl4, sk, skb);
1004 __ip_rt_update_pmtu(rt, &fl4, mtu);
1007 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1008 int oif, u32 mark, u8 protocol, int flow_flags)
1010 const struct iphdr *iph = (const struct iphdr *) skb->data;
1015 mark = IP4_REPLY_MARK(net, skb->mark);
1017 __build_flow_key(&fl4, NULL, iph, oif,
1018 RT_TOS(iph->tos), protocol, mark, flow_flags);
1019 rt = __ip_route_output_key(net, &fl4);
1021 __ip_rt_update_pmtu(rt, &fl4, mtu);
1025 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1027 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1029 const struct iphdr *iph = (const struct iphdr *) skb->data;
1033 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1035 if (!fl4.flowi4_mark)
1036 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1038 rt = __ip_route_output_key(sock_net(sk), &fl4);
1040 __ip_rt_update_pmtu(rt, &fl4, mtu);
1045 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1047 const struct iphdr *iph = (const struct iphdr *) skb->data;
1050 struct dst_entry *odst = NULL;
1055 if (!ip_sk_accept_pmtu(sk))
1058 odst = sk_dst_get(sk);
1060 if (sock_owned_by_user(sk) || !odst) {
1061 __ipv4_sk_update_pmtu(skb, sk, mtu);
1065 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1067 rt = (struct rtable *)odst;
1068 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1069 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1076 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1078 if (!dst_check(&rt->dst, 0)) {
1080 dst_release(&rt->dst);
1082 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1090 sk_dst_set(sk, &rt->dst);
1096 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1098 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1099 int oif, u32 mark, u8 protocol, int flow_flags)
1101 const struct iphdr *iph = (const struct iphdr *) skb->data;
1105 __build_flow_key(&fl4, NULL, iph, oif,
1106 RT_TOS(iph->tos), protocol, mark, flow_flags);
1107 rt = __ip_route_output_key(net, &fl4);
1109 __ip_do_redirect(rt, skb, &fl4, false);
1113 EXPORT_SYMBOL_GPL(ipv4_redirect);
1115 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1117 const struct iphdr *iph = (const struct iphdr *) skb->data;
1121 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1122 rt = __ip_route_output_key(sock_net(sk), &fl4);
1124 __ip_do_redirect(rt, skb, &fl4, false);
1128 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1130 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1132 struct rtable *rt = (struct rtable *) dst;
1134 /* All IPV4 dsts are created with ->obsolete set to the value
1135 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1136 * into this function always.
1138 * When a PMTU/redirect information update invalidates a route,
1139 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1140 * DST_OBSOLETE_DEAD by dst_free().
1142 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1147 static void ipv4_link_failure(struct sk_buff *skb)
1151 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1153 rt = skb_rtable(skb);
1155 dst_set_expires(&rt->dst, 0);
1158 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1160 pr_debug("%s: %pI4 -> %pI4, %s\n",
1161 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1162 skb->dev ? skb->dev->name : "?");
1169 We do not cache source address of outgoing interface,
1170 because it is used only by IP RR, TS and SRR options,
1171 so that it out of fast path.
1173 BTW remember: "addr" is allowed to be not aligned
1177 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1181 if (rt_is_output_route(rt))
1182 src = ip_hdr(skb)->saddr;
1184 struct fib_result res;
1190 memset(&fl4, 0, sizeof(fl4));
1191 fl4.daddr = iph->daddr;
1192 fl4.saddr = iph->saddr;
1193 fl4.flowi4_tos = RT_TOS(iph->tos);
1194 fl4.flowi4_oif = rt->dst.dev->ifindex;
1195 fl4.flowi4_iif = skb->dev->ifindex;
1196 fl4.flowi4_mark = skb->mark;
1199 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1200 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1202 src = inet_select_addr(rt->dst.dev,
1203 rt_nexthop(rt, iph->daddr),
1207 memcpy(addr, &src, 4);
1210 #ifdef CONFIG_IP_ROUTE_CLASSID
1211 static void set_class_tag(struct rtable *rt, u32 tag)
1213 if (!(rt->dst.tclassid & 0xFFFF))
1214 rt->dst.tclassid |= tag & 0xFFFF;
1215 if (!(rt->dst.tclassid & 0xFFFF0000))
1216 rt->dst.tclassid |= tag & 0xFFFF0000;
1220 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1222 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1225 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1227 if (advmss > 65535 - 40)
1228 advmss = 65535 - 40;
1233 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1235 const struct rtable *rt = (const struct rtable *) dst;
1236 unsigned int mtu = rt->rt_pmtu;
1238 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1239 mtu = dst_metric_raw(dst, RTAX_MTU);
1244 mtu = dst->dev->mtu;
1246 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1247 if (rt->rt_uses_gateway && mtu > 576)
1251 return min_t(unsigned int, mtu, IP_MAX_MTU);
1254 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1256 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1257 struct fib_nh_exception *fnhe;
1263 hval = fnhe_hashfun(daddr);
1265 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1266 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1267 if (fnhe->fnhe_daddr == daddr)
1273 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1278 spin_lock_bh(&fnhe_lock);
1280 if (daddr == fnhe->fnhe_daddr) {
1281 struct rtable __rcu **porig;
1282 struct rtable *orig;
1283 int genid = fnhe_genid(dev_net(rt->dst.dev));
1285 if (rt_is_input_route(rt))
1286 porig = &fnhe->fnhe_rth_input;
1288 porig = &fnhe->fnhe_rth_output;
1289 orig = rcu_dereference(*porig);
1291 if (fnhe->fnhe_genid != genid) {
1292 fnhe->fnhe_genid = genid;
1294 fnhe->fnhe_pmtu = 0;
1295 fnhe->fnhe_expires = 0;
1296 fnhe_flush_routes(fnhe);
1299 fill_route_from_fnhe(rt, fnhe);
1300 if (!rt->rt_gateway)
1301 rt->rt_gateway = daddr;
1303 if (!(rt->dst.flags & DST_NOCACHE)) {
1304 rcu_assign_pointer(*porig, rt);
1310 fnhe->fnhe_stamp = jiffies;
1312 spin_unlock_bh(&fnhe_lock);
1317 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1319 struct rtable *orig, *prev, **p;
1322 if (rt_is_input_route(rt)) {
1323 p = (struct rtable **)&nh->nh_rth_input;
1325 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1329 prev = cmpxchg(p, orig, rt);
1339 struct uncached_list {
1341 struct list_head head;
1344 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1346 static void rt_add_uncached_list(struct rtable *rt)
1348 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1350 rt->rt_uncached_list = ul;
1352 spin_lock_bh(&ul->lock);
1353 list_add_tail(&rt->rt_uncached, &ul->head);
1354 spin_unlock_bh(&ul->lock);
1357 static void ipv4_dst_destroy(struct dst_entry *dst)
1359 struct rtable *rt = (struct rtable *) dst;
1361 if (!list_empty(&rt->rt_uncached)) {
1362 struct uncached_list *ul = rt->rt_uncached_list;
1364 spin_lock_bh(&ul->lock);
1365 list_del(&rt->rt_uncached);
1366 spin_unlock_bh(&ul->lock);
1370 void rt_flush_dev(struct net_device *dev)
1372 struct net *net = dev_net(dev);
1376 for_each_possible_cpu(cpu) {
1377 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1379 spin_lock_bh(&ul->lock);
1380 list_for_each_entry(rt, &ul->head, rt_uncached) {
1381 if (rt->dst.dev != dev)
1383 rt->dst.dev = net->loopback_dev;
1384 dev_hold(rt->dst.dev);
1387 spin_unlock_bh(&ul->lock);
1391 static bool rt_cache_valid(const struct rtable *rt)
1394 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1398 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1399 const struct fib_result *res,
1400 struct fib_nh_exception *fnhe,
1401 struct fib_info *fi, u16 type, u32 itag)
1403 bool cached = false;
1406 struct fib_nh *nh = &FIB_RES_NH(*res);
1408 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1409 rt->rt_gateway = nh->nh_gw;
1410 rt->rt_uses_gateway = 1;
1412 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1413 #ifdef CONFIG_IP_ROUTE_CLASSID
1414 rt->dst.tclassid = nh->nh_tclassid;
1416 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1418 cached = rt_bind_exception(rt, fnhe, daddr);
1419 else if (!(rt->dst.flags & DST_NOCACHE))
1420 cached = rt_cache_route(nh, rt);
1421 if (unlikely(!cached)) {
1422 /* Routes we intend to cache in nexthop exception or
1423 * FIB nexthop have the DST_NOCACHE bit clear.
1424 * However, if we are unsuccessful at storing this
1425 * route into the cache we really need to set it.
1427 rt->dst.flags |= DST_NOCACHE;
1428 if (!rt->rt_gateway)
1429 rt->rt_gateway = daddr;
1430 rt_add_uncached_list(rt);
1433 rt_add_uncached_list(rt);
1435 #ifdef CONFIG_IP_ROUTE_CLASSID
1436 #ifdef CONFIG_IP_MULTIPLE_TABLES
1437 set_class_tag(rt, res->tclassid);
1439 set_class_tag(rt, itag);
1443 static struct rtable *rt_dst_alloc(struct net_device *dev,
1444 unsigned int flags, u16 type,
1445 bool nopolicy, bool noxfrm, bool will_cache)
1449 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1450 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1451 (nopolicy ? DST_NOPOLICY : 0) |
1452 (noxfrm ? DST_NOXFRM : 0));
1455 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1456 rt->rt_flags = flags;
1458 rt->rt_is_input = 0;
1462 rt->rt_uses_gateway = 0;
1463 rt->rt_table_id = 0;
1464 INIT_LIST_HEAD(&rt->rt_uncached);
1466 rt->dst.output = ip_output;
1467 if (flags & RTCF_LOCAL)
1468 rt->dst.input = ip_local_deliver;
1474 /* called in rcu_read_lock() section */
1475 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1476 u8 tos, struct net_device *dev, int our)
1479 struct in_device *in_dev = __in_dev_get_rcu(dev);
1480 unsigned int flags = RTCF_MULTICAST;
1484 /* Primary sanity checks. */
1489 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1490 skb->protocol != htons(ETH_P_IP))
1493 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1496 if (ipv4_is_zeronet(saddr)) {
1497 if (!ipv4_is_local_multicast(daddr))
1500 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1506 flags |= RTCF_LOCAL;
1508 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1509 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1513 #ifdef CONFIG_IP_ROUTE_CLASSID
1514 rth->dst.tclassid = itag;
1516 rth->dst.output = ip_rt_bug;
1517 rth->rt_is_input= 1;
1519 #ifdef CONFIG_IP_MROUTE
1520 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1521 rth->dst.input = ip_mr_input;
1523 RT_CACHE_STAT_INC(in_slow_mc);
1525 skb_dst_set(skb, &rth->dst);
1537 static void ip_handle_martian_source(struct net_device *dev,
1538 struct in_device *in_dev,
1539 struct sk_buff *skb,
1543 RT_CACHE_STAT_INC(in_martian_src);
1544 #ifdef CONFIG_IP_ROUTE_VERBOSE
1545 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1547 * RFC1812 recommendation, if source is martian,
1548 * the only hint is MAC header.
1550 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1551 &daddr, &saddr, dev->name);
1552 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1553 print_hex_dump(KERN_WARNING, "ll header: ",
1554 DUMP_PREFIX_OFFSET, 16, 1,
1555 skb_mac_header(skb),
1556 dev->hard_header_len, true);
1562 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1564 struct fnhe_hash_bucket *hash;
1565 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1566 u32 hval = fnhe_hashfun(daddr);
1568 spin_lock_bh(&fnhe_lock);
1570 hash = rcu_dereference_protected(nh->nh_exceptions,
1571 lockdep_is_held(&fnhe_lock));
1574 fnhe_p = &hash->chain;
1575 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1577 if (fnhe->fnhe_daddr == daddr) {
1578 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1579 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1580 fnhe_flush_routes(fnhe);
1581 kfree_rcu(fnhe, rcu);
1584 fnhe_p = &fnhe->fnhe_next;
1585 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1586 lockdep_is_held(&fnhe_lock));
1589 spin_unlock_bh(&fnhe_lock);
1592 /* called in rcu_read_lock() section */
1593 static int __mkroute_input(struct sk_buff *skb,
1594 const struct fib_result *res,
1595 struct in_device *in_dev,
1596 __be32 daddr, __be32 saddr, u32 tos)
1598 struct fib_nh_exception *fnhe;
1601 struct in_device *out_dev;
1605 /* get a working reference to the output device */
1606 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1608 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1612 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1613 in_dev->dev, in_dev, &itag);
1615 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1621 do_cache = res->fi && !itag;
1622 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1623 skb->protocol == htons(ETH_P_IP) &&
1624 (IN_DEV_SHARED_MEDIA(out_dev) ||
1625 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1626 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1628 if (skb->protocol != htons(ETH_P_IP)) {
1629 /* Not IP (i.e. ARP). Do not create route, if it is
1630 * invalid for proxy arp. DNAT routes are always valid.
1632 * Proxy arp feature have been extended to allow, ARP
1633 * replies back to the same interface, to support
1634 * Private VLAN switch technologies. See arp.c.
1636 if (out_dev == in_dev &&
1637 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1643 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1646 rth = rcu_dereference(fnhe->fnhe_rth_input);
1647 if (rth && rth->dst.expires &&
1648 time_after(jiffies, rth->dst.expires)) {
1649 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1656 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1659 if (rt_cache_valid(rth)) {
1660 skb_dst_set_noref(skb, &rth->dst);
1665 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1666 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1667 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1673 rth->rt_is_input = 1;
1675 rth->rt_table_id = res->table->tb_id;
1676 RT_CACHE_STAT_INC(in_slow_tot);
1678 rth->dst.input = ip_forward;
1680 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1681 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1682 rth->dst.lwtstate->orig_output = rth->dst.output;
1683 rth->dst.output = lwtunnel_output;
1685 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1686 rth->dst.lwtstate->orig_input = rth->dst.input;
1687 rth->dst.input = lwtunnel_input;
1689 skb_dst_set(skb, &rth->dst);
1696 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1698 /* To make ICMP packets follow the right flow, the multipath hash is
1699 * calculated from the inner IP addresses in reverse order.
1701 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1703 const struct iphdr *outer_iph = ip_hdr(skb);
1704 struct icmphdr _icmph;
1705 const struct icmphdr *icmph;
1706 struct iphdr _inner_iph;
1707 const struct iphdr *inner_iph;
1709 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1712 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1717 if (icmph->type != ICMP_DEST_UNREACH &&
1718 icmph->type != ICMP_REDIRECT &&
1719 icmph->type != ICMP_TIME_EXCEEDED &&
1720 icmph->type != ICMP_PARAMETERPROB) {
1724 inner_iph = skb_header_pointer(skb,
1725 outer_iph->ihl * 4 + sizeof(_icmph),
1726 sizeof(_inner_iph), &_inner_iph);
1730 return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1733 return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1736 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1738 static int ip_mkroute_input(struct sk_buff *skb,
1739 struct fib_result *res,
1740 const struct flowi4 *fl4,
1741 struct in_device *in_dev,
1742 __be32 daddr, __be32 saddr, u32 tos)
1744 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1745 if (res->fi && res->fi->fib_nhs > 1) {
1748 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1749 h = ip_multipath_icmp_hash(skb);
1751 h = fib_multipath_hash(saddr, daddr);
1752 fib_select_multipath(res, h);
1756 /* create a routing cache entry */
1757 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1761 * NOTE. We drop all the packets that has local source
1762 * addresses, because every properly looped back packet
1763 * must have correct destination already attached by output routine.
1765 * Such approach solves two big problems:
1766 * 1. Not simplex devices are handled properly.
1767 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1768 * called with rcu_read_lock()
1771 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1772 u8 tos, struct net_device *dev)
1774 struct fib_result res;
1775 struct in_device *in_dev = __in_dev_get_rcu(dev);
1776 struct ip_tunnel_info *tun_info;
1778 unsigned int flags = 0;
1782 struct net *net = dev_net(dev);
1785 /* IP on this device is disabled. */
1790 /* Check for the most weird martians, which can be not detected
1794 tun_info = skb_tunnel_info(skb);
1795 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1796 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1798 fl4.flowi4_tun_key.tun_id = 0;
1801 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1802 goto martian_source;
1806 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1809 /* Accept zero addresses only to limited broadcast;
1810 * I even do not know to fix it or not. Waiting for complains :-)
1812 if (ipv4_is_zeronet(saddr))
1813 goto martian_source;
1815 if (ipv4_is_zeronet(daddr))
1816 goto martian_destination;
1818 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1819 * and call it once if daddr or/and saddr are loopback addresses
1821 if (ipv4_is_loopback(daddr)) {
1822 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1823 goto martian_destination;
1824 } else if (ipv4_is_loopback(saddr)) {
1825 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1826 goto martian_source;
1830 * Now we are ready to route packet.
1833 fl4.flowi4_iif = l3mdev_fib_oif_rcu(dev);
1834 fl4.flowi4_mark = skb->mark;
1835 fl4.flowi4_tos = tos;
1836 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1837 fl4.flowi4_flags = 0;
1840 err = fib_lookup(net, &fl4, &res, 0);
1842 if (!IN_DEV_FORWARD(in_dev))
1843 err = -EHOSTUNREACH;
1847 if (res.type == RTN_BROADCAST)
1850 if (res.type == RTN_LOCAL) {
1851 err = fib_validate_source(skb, saddr, daddr, tos,
1852 0, dev, in_dev, &itag);
1854 goto martian_source;
1858 if (!IN_DEV_FORWARD(in_dev)) {
1859 err = -EHOSTUNREACH;
1862 if (res.type != RTN_UNICAST)
1863 goto martian_destination;
1865 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1869 if (skb->protocol != htons(ETH_P_IP))
1872 if (!ipv4_is_zeronet(saddr)) {
1873 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1876 goto martian_source;
1878 flags |= RTCF_BROADCAST;
1879 res.type = RTN_BROADCAST;
1880 RT_CACHE_STAT_INC(in_brd);
1886 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1887 if (rt_cache_valid(rth)) {
1888 skb_dst_set_noref(skb, &rth->dst);
1896 rth = rt_dst_alloc(net->loopback_dev, flags | RTCF_LOCAL, res.type,
1897 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1901 rth->dst.output= ip_rt_bug;
1902 #ifdef CONFIG_IP_ROUTE_CLASSID
1903 rth->dst.tclassid = itag;
1905 rth->rt_is_input = 1;
1907 rth->rt_table_id = res.table->tb_id;
1909 RT_CACHE_STAT_INC(in_slow_tot);
1910 if (res.type == RTN_UNREACHABLE) {
1911 rth->dst.input= ip_error;
1912 rth->dst.error= -err;
1913 rth->rt_flags &= ~RTCF_LOCAL;
1916 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1917 rth->dst.flags |= DST_NOCACHE;
1918 rt_add_uncached_list(rth);
1921 skb_dst_set(skb, &rth->dst);
1926 RT_CACHE_STAT_INC(in_no_route);
1927 res.type = RTN_UNREACHABLE;
1933 * Do not cache martian addresses: they should be logged (RFC1812)
1935 martian_destination:
1936 RT_CACHE_STAT_INC(in_martian_dst);
1937 #ifdef CONFIG_IP_ROUTE_VERBOSE
1938 if (IN_DEV_LOG_MARTIANS(in_dev))
1939 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1940 &daddr, &saddr, dev->name);
1952 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1956 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1957 u8 tos, struct net_device *dev)
1963 /* Multicast recognition logic is moved from route cache to here.
1964 The problem was that too many Ethernet cards have broken/missing
1965 hardware multicast filters :-( As result the host on multicasting
1966 network acquires a lot of useless route cache entries, sort of
1967 SDR messages from all the world. Now we try to get rid of them.
1968 Really, provided software IP multicast filter is organized
1969 reasonably (at least, hashed), it does not result in a slowdown
1970 comparing with route cache reject entries.
1971 Note, that multicast routers are not affected, because
1972 route cache entry is created eventually.
1974 if (ipv4_is_multicast(daddr)) {
1975 struct in_device *in_dev = __in_dev_get_rcu(dev);
1978 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1979 ip_hdr(skb)->protocol);
1981 #ifdef CONFIG_IP_MROUTE
1983 (!ipv4_is_local_multicast(daddr) &&
1984 IN_DEV_MFORWARD(in_dev))
1987 int res = ip_route_input_mc(skb, daddr, saddr,
1996 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2000 EXPORT_SYMBOL(ip_route_input_noref);
2002 /* called with rcu_read_lock() */
2003 static struct rtable *__mkroute_output(const struct fib_result *res,
2004 const struct flowi4 *fl4, int orig_oif,
2005 struct net_device *dev_out,
2008 struct fib_info *fi = res->fi;
2009 struct fib_nh_exception *fnhe;
2010 struct in_device *in_dev;
2011 u16 type = res->type;
2015 in_dev = __in_dev_get_rcu(dev_out);
2017 return ERR_PTR(-EINVAL);
2019 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2020 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2021 return ERR_PTR(-EINVAL);
2023 if (ipv4_is_lbcast(fl4->daddr))
2024 type = RTN_BROADCAST;
2025 else if (ipv4_is_multicast(fl4->daddr))
2026 type = RTN_MULTICAST;
2027 else if (ipv4_is_zeronet(fl4->daddr))
2028 return ERR_PTR(-EINVAL);
2030 if (dev_out->flags & IFF_LOOPBACK)
2031 flags |= RTCF_LOCAL;
2034 if (type == RTN_BROADCAST) {
2035 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2037 } else if (type == RTN_MULTICAST) {
2038 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2039 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2041 flags &= ~RTCF_LOCAL;
2044 /* If multicast route do not exist use
2045 * default one, but do not gateway in this case.
2048 if (fi && res->prefixlen < 4)
2053 do_cache &= fi != NULL;
2055 struct rtable __rcu **prth;
2056 struct fib_nh *nh = &FIB_RES_NH(*res);
2058 fnhe = find_exception(nh, fl4->daddr);
2060 prth = &fnhe->fnhe_rth_output;
2061 rth = rcu_dereference(*prth);
2062 if (rth && rth->dst.expires &&
2063 time_after(jiffies, rth->dst.expires)) {
2064 ip_del_fnhe(nh, fl4->daddr);
2071 if (unlikely(fl4->flowi4_flags &
2072 FLOWI_FLAG_KNOWN_NH &&
2074 nh->nh_scope == RT_SCOPE_LINK))) {
2078 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2079 rth = rcu_dereference(*prth);
2082 if (rt_cache_valid(rth)) {
2083 dst_hold(&rth->dst);
2089 rth = rt_dst_alloc(dev_out, flags, type,
2090 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2091 IN_DEV_CONF_GET(in_dev, NOXFRM),
2094 return ERR_PTR(-ENOBUFS);
2096 rth->rt_iif = orig_oif ? : 0;
2098 rth->rt_table_id = res->table->tb_id;
2100 RT_CACHE_STAT_INC(out_slow_tot);
2102 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2103 if (flags & RTCF_LOCAL &&
2104 !(dev_out->flags & IFF_LOOPBACK)) {
2105 rth->dst.output = ip_mc_output;
2106 RT_CACHE_STAT_INC(out_slow_mc);
2108 #ifdef CONFIG_IP_MROUTE
2109 if (type == RTN_MULTICAST) {
2110 if (IN_DEV_MFORWARD(in_dev) &&
2111 !ipv4_is_local_multicast(fl4->daddr)) {
2112 rth->dst.input = ip_mr_input;
2113 rth->dst.output = ip_mc_output;
2119 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2120 if (lwtunnel_output_redirect(rth->dst.lwtstate))
2121 rth->dst.output = lwtunnel_output;
2127 * Major route resolver routine.
2130 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2133 struct net_device *dev_out = NULL;
2134 __u8 tos = RT_FL_TOS(fl4);
2135 unsigned int flags = 0;
2136 struct fib_result res;
2139 int err = -ENETUNREACH;
2145 orig_oif = fl4->flowi4_oif;
2147 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2148 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2149 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2150 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2154 rth = ERR_PTR(-EINVAL);
2155 if (ipv4_is_multicast(fl4->saddr) ||
2156 ipv4_is_lbcast(fl4->saddr) ||
2157 ipv4_is_zeronet(fl4->saddr))
2160 /* I removed check for oif == dev_out->oif here.
2161 It was wrong for two reasons:
2162 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2163 is assigned to multiple interfaces.
2164 2. Moreover, we are allowed to send packets with saddr
2165 of another iface. --ANK
2168 if (fl4->flowi4_oif == 0 &&
2169 (ipv4_is_multicast(fl4->daddr) ||
2170 ipv4_is_lbcast(fl4->daddr))) {
2171 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2172 dev_out = __ip_dev_find(net, fl4->saddr, false);
2176 /* Special hack: user can direct multicasts
2177 and limited broadcast via necessary interface
2178 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2179 This hack is not just for fun, it allows
2180 vic,vat and friends to work.
2181 They bind socket to loopback, set ttl to zero
2182 and expect that it will work.
2183 From the viewpoint of routing cache they are broken,
2184 because we are not allowed to build multicast path
2185 with loopback source addr (look, routing cache
2186 cannot know, that ttl is zero, so that packet
2187 will not leave this host and route is valid).
2188 Luckily, this hack is good workaround.
2191 fl4->flowi4_oif = dev_out->ifindex;
2195 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2196 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2197 if (!__ip_dev_find(net, fl4->saddr, false))
2203 if (fl4->flowi4_oif) {
2204 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2205 rth = ERR_PTR(-ENODEV);
2209 /* RACE: Check return value of inet_select_addr instead. */
2210 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2211 rth = ERR_PTR(-ENETUNREACH);
2214 if (ipv4_is_local_multicast(fl4->daddr) ||
2215 ipv4_is_lbcast(fl4->daddr) ||
2216 fl4->flowi4_proto == IPPROTO_IGMP) {
2218 fl4->saddr = inet_select_addr(dev_out, 0,
2223 if (ipv4_is_multicast(fl4->daddr))
2224 fl4->saddr = inet_select_addr(dev_out, 0,
2226 else if (!fl4->daddr)
2227 fl4->saddr = inet_select_addr(dev_out, 0,
2231 rth = l3mdev_get_rtable(dev_out, fl4);
2237 fl4->daddr = fl4->saddr;
2239 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2240 dev_out = net->loopback_dev;
2241 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2242 res.type = RTN_LOCAL;
2243 flags |= RTCF_LOCAL;
2247 err = fib_lookup(net, fl4, &res, 0);
2251 if (fl4->flowi4_oif &&
2252 !netif_index_is_l3_master(net, fl4->flowi4_oif)) {
2253 /* Apparently, routing tables are wrong. Assume,
2254 that the destination is on link.
2257 Because we are allowed to send to iface
2258 even if it has NO routes and NO assigned
2259 addresses. When oif is specified, routing
2260 tables are looked up with only one purpose:
2261 to catch if destination is gatewayed, rather than
2262 direct. Moreover, if MSG_DONTROUTE is set,
2263 we send packet, ignoring both routing tables
2264 and ifaddr state. --ANK
2267 We could make it even if oif is unknown,
2268 likely IPv6, but we do not.
2271 if (fl4->saddr == 0)
2272 fl4->saddr = inet_select_addr(dev_out, 0,
2274 res.type = RTN_UNICAST;
2281 if (res.type == RTN_LOCAL) {
2283 if (res.fi->fib_prefsrc)
2284 fl4->saddr = res.fi->fib_prefsrc;
2286 fl4->saddr = fl4->daddr;
2288 dev_out = net->loopback_dev;
2289 fl4->flowi4_oif = dev_out->ifindex;
2290 flags |= RTCF_LOCAL;
2294 fib_select_path(net, &res, fl4, mp_hash);
2296 dev_out = FIB_RES_DEV(res);
2297 fl4->flowi4_oif = dev_out->ifindex;
2301 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2307 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2309 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2314 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2316 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2318 return mtu ? : dst->dev->mtu;
2321 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2322 struct sk_buff *skb, u32 mtu)
2326 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2327 struct sk_buff *skb)
2331 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2337 static struct dst_ops ipv4_dst_blackhole_ops = {
2339 .check = ipv4_blackhole_dst_check,
2340 .mtu = ipv4_blackhole_mtu,
2341 .default_advmss = ipv4_default_advmss,
2342 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2343 .redirect = ipv4_rt_blackhole_redirect,
2344 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2345 .neigh_lookup = ipv4_neigh_lookup,
2348 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2350 struct rtable *ort = (struct rtable *) dst_orig;
2353 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2355 struct dst_entry *new = &rt->dst;
2358 new->input = dst_discard;
2359 new->output = dst_discard_out;
2361 new->dev = ort->dst.dev;
2365 rt->rt_is_input = ort->rt_is_input;
2366 rt->rt_iif = ort->rt_iif;
2367 rt->rt_pmtu = ort->rt_pmtu;
2369 rt->rt_genid = rt_genid_ipv4(net);
2370 rt->rt_flags = ort->rt_flags;
2371 rt->rt_type = ort->rt_type;
2372 rt->rt_gateway = ort->rt_gateway;
2373 rt->rt_uses_gateway = ort->rt_uses_gateway;
2375 INIT_LIST_HEAD(&rt->rt_uncached);
2379 dst_release(dst_orig);
2381 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2384 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2385 const struct sock *sk)
2387 struct rtable *rt = __ip_route_output_key(net, flp4);
2392 if (flp4->flowi4_proto)
2393 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2394 flowi4_to_flowi(flp4),
2399 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2401 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2402 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2403 u32 seq, int event, int nowait, unsigned int flags)
2405 struct rtable *rt = skb_rtable(skb);
2407 struct nlmsghdr *nlh;
2408 unsigned long expires = 0;
2410 u32 metrics[RTAX_MAX];
2412 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2416 r = nlmsg_data(nlh);
2417 r->rtm_family = AF_INET;
2418 r->rtm_dst_len = 32;
2420 r->rtm_tos = fl4->flowi4_tos;
2421 r->rtm_table = table_id;
2422 if (nla_put_u32(skb, RTA_TABLE, table_id))
2423 goto nla_put_failure;
2424 r->rtm_type = rt->rt_type;
2425 r->rtm_scope = RT_SCOPE_UNIVERSE;
2426 r->rtm_protocol = RTPROT_UNSPEC;
2427 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2428 if (rt->rt_flags & RTCF_NOTIFY)
2429 r->rtm_flags |= RTM_F_NOTIFY;
2430 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2431 r->rtm_flags |= RTCF_DOREDIRECT;
2433 if (nla_put_in_addr(skb, RTA_DST, dst))
2434 goto nla_put_failure;
2436 r->rtm_src_len = 32;
2437 if (nla_put_in_addr(skb, RTA_SRC, src))
2438 goto nla_put_failure;
2441 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2442 goto nla_put_failure;
2443 #ifdef CONFIG_IP_ROUTE_CLASSID
2444 if (rt->dst.tclassid &&
2445 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2446 goto nla_put_failure;
2448 if (!rt_is_input_route(rt) &&
2449 fl4->saddr != src) {
2450 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2451 goto nla_put_failure;
2453 if (rt->rt_uses_gateway &&
2454 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2455 goto nla_put_failure;
2457 expires = rt->dst.expires;
2459 unsigned long now = jiffies;
2461 if (time_before(now, expires))
2467 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2468 if (rt->rt_pmtu && expires)
2469 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2470 if (rtnetlink_put_metrics(skb, metrics) < 0)
2471 goto nla_put_failure;
2473 if (fl4->flowi4_mark &&
2474 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2475 goto nla_put_failure;
2477 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2478 nla_put_u32(skb, RTA_UID,
2479 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2480 goto nla_put_failure;
2482 error = rt->dst.error;
2484 if (rt_is_input_route(rt)) {
2485 #ifdef CONFIG_IP_MROUTE
2486 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2487 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2488 int err = ipmr_get_route(net, skb,
2489 fl4->saddr, fl4->daddr,
2495 goto nla_put_failure;
2497 if (err == -EMSGSIZE)
2498 goto nla_put_failure;
2504 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2505 goto nla_put_failure;
2508 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2509 goto nla_put_failure;
2511 nlmsg_end(skb, nlh);
2515 nlmsg_cancel(skb, nlh);
2519 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2521 struct net *net = sock_net(in_skb->sk);
2523 struct nlattr *tb[RTA_MAX+1];
2524 struct rtable *rt = NULL;
2531 struct sk_buff *skb;
2532 u32 table_id = RT_TABLE_MAIN;
2535 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2539 rtm = nlmsg_data(nlh);
2541 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2547 /* Reserve room for dummy headers, this skb can pass
2548 through good chunk of routing engine.
2550 skb_reset_mac_header(skb);
2551 skb_reset_network_header(skb);
2553 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2554 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2555 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2557 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2558 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2559 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2560 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2562 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2564 uid = (iif ? INVALID_UID : current_uid());
2566 memset(&fl4, 0, sizeof(fl4));
2569 fl4.flowi4_tos = rtm->rtm_tos;
2570 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2571 fl4.flowi4_mark = mark;
2572 fl4.flowi4_uid = uid;
2574 if (netif_index_is_l3_master(net, fl4.flowi4_oif))
2575 fl4.flowi4_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF;
2578 struct net_device *dev;
2580 dev = __dev_get_by_index(net, iif);
2586 skb->protocol = htons(ETH_P_IP);
2590 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2593 rt = skb_rtable(skb);
2594 if (err == 0 && rt->dst.error)
2595 err = -rt->dst.error;
2597 rt = ip_route_output_key(net, &fl4);
2607 skb_dst_set(skb, &rt->dst);
2608 if (rtm->rtm_flags & RTM_F_NOTIFY)
2609 rt->rt_flags |= RTCF_NOTIFY;
2611 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2612 table_id = rt->rt_table_id;
2614 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2615 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2616 RTM_NEWROUTE, 0, 0);
2620 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2629 void ip_rt_multicast_event(struct in_device *in_dev)
2631 rt_cache_flush(dev_net(in_dev->dev));
2634 #ifdef CONFIG_SYSCTL
2635 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2636 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2637 static int ip_rt_gc_elasticity __read_mostly = 8;
2639 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2640 void __user *buffer,
2641 size_t *lenp, loff_t *ppos)
2643 struct net *net = (struct net *)__ctl->extra1;
2646 rt_cache_flush(net);
2647 fnhe_genid_bump(net);
2654 static struct ctl_table ipv4_route_table[] = {
2656 .procname = "gc_thresh",
2657 .data = &ipv4_dst_ops.gc_thresh,
2658 .maxlen = sizeof(int),
2660 .proc_handler = proc_dointvec,
2663 .procname = "max_size",
2664 .data = &ip_rt_max_size,
2665 .maxlen = sizeof(int),
2667 .proc_handler = proc_dointvec,
2670 /* Deprecated. Use gc_min_interval_ms */
2672 .procname = "gc_min_interval",
2673 .data = &ip_rt_gc_min_interval,
2674 .maxlen = sizeof(int),
2676 .proc_handler = proc_dointvec_jiffies,
2679 .procname = "gc_min_interval_ms",
2680 .data = &ip_rt_gc_min_interval,
2681 .maxlen = sizeof(int),
2683 .proc_handler = proc_dointvec_ms_jiffies,
2686 .procname = "gc_timeout",
2687 .data = &ip_rt_gc_timeout,
2688 .maxlen = sizeof(int),
2690 .proc_handler = proc_dointvec_jiffies,
2693 .procname = "gc_interval",
2694 .data = &ip_rt_gc_interval,
2695 .maxlen = sizeof(int),
2697 .proc_handler = proc_dointvec_jiffies,
2700 .procname = "redirect_load",
2701 .data = &ip_rt_redirect_load,
2702 .maxlen = sizeof(int),
2704 .proc_handler = proc_dointvec,
2707 .procname = "redirect_number",
2708 .data = &ip_rt_redirect_number,
2709 .maxlen = sizeof(int),
2711 .proc_handler = proc_dointvec,
2714 .procname = "redirect_silence",
2715 .data = &ip_rt_redirect_silence,
2716 .maxlen = sizeof(int),
2718 .proc_handler = proc_dointvec,
2721 .procname = "error_cost",
2722 .data = &ip_rt_error_cost,
2723 .maxlen = sizeof(int),
2725 .proc_handler = proc_dointvec,
2728 .procname = "error_burst",
2729 .data = &ip_rt_error_burst,
2730 .maxlen = sizeof(int),
2732 .proc_handler = proc_dointvec,
2735 .procname = "gc_elasticity",
2736 .data = &ip_rt_gc_elasticity,
2737 .maxlen = sizeof(int),
2739 .proc_handler = proc_dointvec,
2742 .procname = "mtu_expires",
2743 .data = &ip_rt_mtu_expires,
2744 .maxlen = sizeof(int),
2746 .proc_handler = proc_dointvec_jiffies,
2749 .procname = "min_pmtu",
2750 .data = &ip_rt_min_pmtu,
2751 .maxlen = sizeof(int),
2753 .proc_handler = proc_dointvec,
2756 .procname = "min_adv_mss",
2757 .data = &ip_rt_min_advmss,
2758 .maxlen = sizeof(int),
2760 .proc_handler = proc_dointvec,
2765 static struct ctl_table ipv4_route_flush_table[] = {
2767 .procname = "flush",
2768 .maxlen = sizeof(int),
2770 .proc_handler = ipv4_sysctl_rtcache_flush,
2775 static __net_init int sysctl_route_net_init(struct net *net)
2777 struct ctl_table *tbl;
2779 tbl = ipv4_route_flush_table;
2780 if (!net_eq(net, &init_net)) {
2781 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2785 /* Don't export sysctls to unprivileged users */
2786 if (net->user_ns != &init_user_ns)
2787 tbl[0].procname = NULL;
2789 tbl[0].extra1 = net;
2791 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2792 if (!net->ipv4.route_hdr)
2797 if (tbl != ipv4_route_flush_table)
2803 static __net_exit void sysctl_route_net_exit(struct net *net)
2805 struct ctl_table *tbl;
2807 tbl = net->ipv4.route_hdr->ctl_table_arg;
2808 unregister_net_sysctl_table(net->ipv4.route_hdr);
2809 BUG_ON(tbl == ipv4_route_flush_table);
2813 static __net_initdata struct pernet_operations sysctl_route_ops = {
2814 .init = sysctl_route_net_init,
2815 .exit = sysctl_route_net_exit,
2819 static __net_init int rt_genid_init(struct net *net)
2821 atomic_set(&net->ipv4.rt_genid, 0);
2822 atomic_set(&net->fnhe_genid, 0);
2823 get_random_bytes(&net->ipv4.dev_addr_genid,
2824 sizeof(net->ipv4.dev_addr_genid));
2828 static __net_initdata struct pernet_operations rt_genid_ops = {
2829 .init = rt_genid_init,
2832 static int __net_init ipv4_inetpeer_init(struct net *net)
2834 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2838 inet_peer_base_init(bp);
2839 net->ipv4.peers = bp;
2843 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2845 struct inet_peer_base *bp = net->ipv4.peers;
2847 net->ipv4.peers = NULL;
2848 inetpeer_invalidate_tree(bp);
2852 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2853 .init = ipv4_inetpeer_init,
2854 .exit = ipv4_inetpeer_exit,
2857 #ifdef CONFIG_IP_ROUTE_CLASSID
2858 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2859 #endif /* CONFIG_IP_ROUTE_CLASSID */
2861 int __init ip_rt_init(void)
2866 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2868 panic("IP: failed to allocate ip_idents\n");
2870 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2872 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2874 panic("IP: failed to allocate ip_tstamps\n");
2876 for_each_possible_cpu(cpu) {
2877 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2879 INIT_LIST_HEAD(&ul->head);
2880 spin_lock_init(&ul->lock);
2882 #ifdef CONFIG_IP_ROUTE_CLASSID
2883 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2885 panic("IP: failed to allocate ip_rt_acct\n");
2888 ipv4_dst_ops.kmem_cachep =
2889 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2890 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2892 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2894 if (dst_entries_init(&ipv4_dst_ops) < 0)
2895 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2897 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2898 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2900 ipv4_dst_ops.gc_thresh = ~0;
2901 ip_rt_max_size = INT_MAX;
2906 if (ip_rt_proc_init())
2907 pr_err("Unable to create route proc files\n");
2912 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2914 #ifdef CONFIG_SYSCTL
2915 register_pernet_subsys(&sysctl_route_ops);
2917 register_pernet_subsys(&rt_genid_ops);
2918 register_pernet_subsys(&ipv4_inetpeer_ops);
2922 #ifdef CONFIG_SYSCTL
2924 * We really need to sanitize the damn ipv4 init order, then all
2925 * this nonsense will go away.
2927 void __init ip_static_sysctl_init(void)
2929 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);