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>
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;
133 * Interface to generic destination cache.
136 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
137 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
138 static unsigned int ipv4_mtu(const struct dst_entry *dst);
139 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
140 static void ipv4_link_failure(struct sk_buff *skb);
141 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb, u32 mtu);
143 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
144 struct sk_buff *skb);
145 static void ipv4_dst_destroy(struct dst_entry *dst);
147 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
153 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
157 static struct dst_ops ipv4_dst_ops = {
159 .check = ipv4_dst_check,
160 .default_advmss = ipv4_default_advmss,
162 .cow_metrics = ipv4_cow_metrics,
163 .destroy = ipv4_dst_destroy,
164 .negative_advice = ipv4_negative_advice,
165 .link_failure = ipv4_link_failure,
166 .update_pmtu = ip_rt_update_pmtu,
167 .redirect = ip_do_redirect,
168 .local_out = __ip_local_out,
169 .neigh_lookup = ipv4_neigh_lookup,
172 #define ECN_OR_COST(class) TC_PRIO_##class
174 const __u8 ip_tos2prio[16] = {
176 ECN_OR_COST(BESTEFFORT),
178 ECN_OR_COST(BESTEFFORT),
184 ECN_OR_COST(INTERACTIVE),
186 ECN_OR_COST(INTERACTIVE),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK),
189 TC_PRIO_INTERACTIVE_BULK,
190 ECN_OR_COST(INTERACTIVE_BULK)
192 EXPORT_SYMBOL(ip_tos2prio);
194 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
195 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
197 #ifdef CONFIG_PROC_FS
198 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
202 return SEQ_START_TOKEN;
205 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
211 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
215 static int rt_cache_seq_show(struct seq_file *seq, void *v)
217 if (v == SEQ_START_TOKEN)
218 seq_printf(seq, "%-127s\n",
219 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
220 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
225 static const struct seq_operations rt_cache_seq_ops = {
226 .start = rt_cache_seq_start,
227 .next = rt_cache_seq_next,
228 .stop = rt_cache_seq_stop,
229 .show = rt_cache_seq_show,
232 static int rt_cache_seq_open(struct inode *inode, struct file *file)
234 return seq_open(file, &rt_cache_seq_ops);
237 static const struct file_operations rt_cache_seq_fops = {
238 .owner = THIS_MODULE,
239 .open = rt_cache_seq_open,
242 .release = seq_release,
246 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
251 return SEQ_START_TOKEN;
253 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
254 if (!cpu_possible(cpu))
257 return &per_cpu(rt_cache_stat, cpu);
262 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
266 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
267 if (!cpu_possible(cpu))
270 return &per_cpu(rt_cache_stat, cpu);
276 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
281 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
283 struct rt_cache_stat *st = v;
285 if (v == SEQ_START_TOKEN) {
286 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");
290 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
291 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
292 dst_entries_get_slow(&ipv4_dst_ops),
305 0, /* st->gc_total */
306 0, /* st->gc_ignored */
307 0, /* st->gc_goal_miss */
308 0, /* st->gc_dst_overflow */
309 0, /* st->in_hlist_search */
310 0 /* st->out_hlist_search */
315 static const struct seq_operations rt_cpu_seq_ops = {
316 .start = rt_cpu_seq_start,
317 .next = rt_cpu_seq_next,
318 .stop = rt_cpu_seq_stop,
319 .show = rt_cpu_seq_show,
323 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
325 return seq_open(file, &rt_cpu_seq_ops);
328 static const struct file_operations rt_cpu_seq_fops = {
329 .owner = THIS_MODULE,
330 .open = rt_cpu_seq_open,
333 .release = seq_release,
336 #ifdef CONFIG_IP_ROUTE_CLASSID
337 static int rt_acct_proc_show(struct seq_file *m, void *v)
339 struct ip_rt_acct *dst, *src;
342 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
346 for_each_possible_cpu(i) {
347 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
348 for (j = 0; j < 256; j++) {
349 dst[j].o_bytes += src[j].o_bytes;
350 dst[j].o_packets += src[j].o_packets;
351 dst[j].i_bytes += src[j].i_bytes;
352 dst[j].i_packets += src[j].i_packets;
356 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
361 static int rt_acct_proc_open(struct inode *inode, struct file *file)
363 return single_open(file, rt_acct_proc_show, NULL);
366 static const struct file_operations rt_acct_proc_fops = {
367 .owner = THIS_MODULE,
368 .open = rt_acct_proc_open,
371 .release = single_release,
375 static int __net_init ip_rt_do_proc_init(struct net *net)
377 struct proc_dir_entry *pde;
379 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
384 pde = proc_create("rt_cache", S_IRUGO,
385 net->proc_net_stat, &rt_cpu_seq_fops);
389 #ifdef CONFIG_IP_ROUTE_CLASSID
390 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
396 #ifdef CONFIG_IP_ROUTE_CLASSID
398 remove_proc_entry("rt_cache", net->proc_net_stat);
401 remove_proc_entry("rt_cache", net->proc_net);
406 static void __net_exit ip_rt_do_proc_exit(struct net *net)
408 remove_proc_entry("rt_cache", net->proc_net_stat);
409 remove_proc_entry("rt_cache", net->proc_net);
410 #ifdef CONFIG_IP_ROUTE_CLASSID
411 remove_proc_entry("rt_acct", net->proc_net);
415 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
416 .init = ip_rt_do_proc_init,
417 .exit = ip_rt_do_proc_exit,
420 static int __init ip_rt_proc_init(void)
422 return register_pernet_subsys(&ip_rt_proc_ops);
426 static inline int ip_rt_proc_init(void)
430 #endif /* CONFIG_PROC_FS */
432 static inline bool rt_is_expired(const struct rtable *rth)
434 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
437 void rt_cache_flush(struct net *net)
439 rt_genid_bump_ipv4(net);
442 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
446 struct net_device *dev = dst->dev;
447 const __be32 *pkey = daddr;
448 const struct rtable *rt;
451 rt = (const struct rtable *) dst;
453 pkey = (const __be32 *) &rt->rt_gateway;
455 pkey = &ip_hdr(skb)->daddr;
457 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
460 return neigh_create(&arp_tbl, pkey, dev);
463 #define IP_IDENTS_SZ 2048u
465 static atomic_t *ip_idents __read_mostly;
466 static u32 *ip_tstamps __read_mostly;
468 /* In order to protect privacy, we add a perturbation to identifiers
469 * if one generator is seldom used. This makes hard for an attacker
470 * to infer how many packets were sent between two points in time.
472 u32 ip_idents_reserve(u32 hash, int segs)
474 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
475 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
476 u32 old = ACCESS_ONCE(*p_tstamp);
477 u32 now = (u32)jiffies;
480 if (old != now && cmpxchg(p_tstamp, old, now) == old)
481 delta = prandom_u32_max(now - old);
483 return atomic_add_return(segs + delta, p_id) - segs;
485 EXPORT_SYMBOL(ip_idents_reserve);
487 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
489 static u32 ip_idents_hashrnd __read_mostly;
492 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
494 hash = jhash_3words((__force u32)iph->daddr,
495 (__force u32)iph->saddr,
496 iph->protocol ^ net_hash_mix(net),
498 id = ip_idents_reserve(hash, segs);
501 EXPORT_SYMBOL(__ip_select_ident);
503 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
504 const struct iphdr *iph,
506 u8 prot, u32 mark, int flow_flags)
509 const struct inet_sock *inet = inet_sk(sk);
511 oif = sk->sk_bound_dev_if;
513 tos = RT_CONN_FLAGS(sk);
514 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
516 flowi4_init_output(fl4, oif, mark, tos,
517 RT_SCOPE_UNIVERSE, prot,
519 iph->daddr, iph->saddr, 0, 0);
522 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
523 const struct sock *sk)
525 const struct iphdr *iph = ip_hdr(skb);
526 int oif = skb->dev->ifindex;
527 u8 tos = RT_TOS(iph->tos);
528 u8 prot = iph->protocol;
529 u32 mark = skb->mark;
531 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
534 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
536 const struct inet_sock *inet = inet_sk(sk);
537 const struct ip_options_rcu *inet_opt;
538 __be32 daddr = inet->inet_daddr;
541 inet_opt = rcu_dereference(inet->inet_opt);
542 if (inet_opt && inet_opt->opt.srr)
543 daddr = inet_opt->opt.faddr;
544 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
545 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
546 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
547 inet_sk_flowi_flags(sk),
548 daddr, inet->inet_saddr, 0, 0);
552 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
553 const struct sk_buff *skb)
556 build_skb_flow_key(fl4, skb, sk);
558 build_sk_flow_key(fl4, sk);
561 static inline void rt_free(struct rtable *rt)
563 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
566 static DEFINE_SPINLOCK(fnhe_lock);
568 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
572 rt = rcu_dereference(fnhe->fnhe_rth_input);
574 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
577 rt = rcu_dereference(fnhe->fnhe_rth_output);
579 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
584 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
586 struct fib_nh_exception *fnhe, *oldest;
588 oldest = rcu_dereference(hash->chain);
589 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
590 fnhe = rcu_dereference(fnhe->fnhe_next)) {
591 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
594 fnhe_flush_routes(oldest);
598 static inline u32 fnhe_hashfun(__be32 daddr)
600 static u32 fnhe_hashrnd __read_mostly;
603 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
604 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
605 return hash_32(hval, FNHE_HASH_SHIFT);
608 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
610 rt->rt_pmtu = fnhe->fnhe_pmtu;
611 rt->dst.expires = fnhe->fnhe_expires;
614 rt->rt_flags |= RTCF_REDIRECTED;
615 rt->rt_gateway = fnhe->fnhe_gw;
616 rt->rt_uses_gateway = 1;
620 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
621 u32 pmtu, unsigned long expires)
623 struct fnhe_hash_bucket *hash;
624 struct fib_nh_exception *fnhe;
628 u32 hval = fnhe_hashfun(daddr);
630 spin_lock_bh(&fnhe_lock);
632 hash = rcu_dereference(nh->nh_exceptions);
634 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
637 rcu_assign_pointer(nh->nh_exceptions, hash);
643 for (fnhe = rcu_dereference(hash->chain); fnhe;
644 fnhe = rcu_dereference(fnhe->fnhe_next)) {
645 if (fnhe->fnhe_daddr == daddr)
654 fnhe->fnhe_pmtu = pmtu;
655 fnhe->fnhe_expires = max(1UL, expires);
657 /* Update all cached dsts too */
658 rt = rcu_dereference(fnhe->fnhe_rth_input);
660 fill_route_from_fnhe(rt, fnhe);
661 rt = rcu_dereference(fnhe->fnhe_rth_output);
663 fill_route_from_fnhe(rt, fnhe);
665 if (depth > FNHE_RECLAIM_DEPTH)
666 fnhe = fnhe_oldest(hash);
668 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
672 fnhe->fnhe_next = hash->chain;
673 rcu_assign_pointer(hash->chain, fnhe);
675 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
676 fnhe->fnhe_daddr = daddr;
678 fnhe->fnhe_pmtu = pmtu;
679 fnhe->fnhe_expires = expires;
681 /* Exception created; mark the cached routes for the nexthop
682 * stale, so anyone caching it rechecks if this exception
685 rt = rcu_dereference(nh->nh_rth_input);
687 rt->dst.obsolete = DST_OBSOLETE_KILL;
689 for_each_possible_cpu(i) {
690 struct rtable __rcu **prt;
691 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
692 rt = rcu_dereference(*prt);
694 rt->dst.obsolete = DST_OBSOLETE_KILL;
698 fnhe->fnhe_stamp = jiffies;
701 spin_unlock_bh(&fnhe_lock);
704 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
707 __be32 new_gw = icmp_hdr(skb)->un.gateway;
708 __be32 old_gw = ip_hdr(skb)->saddr;
709 struct net_device *dev = skb->dev;
710 struct in_device *in_dev;
711 struct fib_result res;
715 switch (icmp_hdr(skb)->code & 7) {
717 case ICMP_REDIR_NETTOS:
718 case ICMP_REDIR_HOST:
719 case ICMP_REDIR_HOSTTOS:
726 if (rt->rt_gateway != old_gw)
729 in_dev = __in_dev_get_rcu(dev);
734 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
735 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
736 ipv4_is_zeronet(new_gw))
737 goto reject_redirect;
739 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
740 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
741 goto reject_redirect;
742 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
743 goto reject_redirect;
745 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
746 goto reject_redirect;
749 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
751 if (!(n->nud_state & NUD_VALID)) {
752 neigh_event_send(n, NULL);
754 if (fib_lookup(net, fl4, &res, 0) == 0) {
755 struct fib_nh *nh = &FIB_RES_NH(res);
757 update_or_create_fnhe(nh, fl4->daddr, new_gw,
761 rt->dst.obsolete = DST_OBSOLETE_KILL;
762 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
769 #ifdef CONFIG_IP_ROUTE_VERBOSE
770 if (IN_DEV_LOG_MARTIANS(in_dev)) {
771 const struct iphdr *iph = (const struct iphdr *) skb->data;
772 __be32 daddr = iph->daddr;
773 __be32 saddr = iph->saddr;
775 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
776 " Advised path = %pI4 -> %pI4\n",
777 &old_gw, dev->name, &new_gw,
784 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
788 const struct iphdr *iph = (const struct iphdr *) skb->data;
789 int oif = skb->dev->ifindex;
790 u8 tos = RT_TOS(iph->tos);
791 u8 prot = iph->protocol;
792 u32 mark = skb->mark;
794 rt = (struct rtable *) dst;
796 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
797 __ip_do_redirect(rt, skb, &fl4, true);
800 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
802 struct rtable *rt = (struct rtable *)dst;
803 struct dst_entry *ret = dst;
806 if (dst->obsolete > 0) {
809 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
820 * 1. The first ip_rt_redirect_number redirects are sent
821 * with exponential backoff, then we stop sending them at all,
822 * assuming that the host ignores our redirects.
823 * 2. If we did not see packets requiring redirects
824 * during ip_rt_redirect_silence, we assume that the host
825 * forgot redirected route and start to send redirects again.
827 * This algorithm is much cheaper and more intelligent than dumb load limiting
830 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
831 * and "frag. need" (breaks PMTU discovery) in icmp.c.
834 void ip_rt_send_redirect(struct sk_buff *skb)
836 struct rtable *rt = skb_rtable(skb);
837 struct in_device *in_dev;
838 struct inet_peer *peer;
844 in_dev = __in_dev_get_rcu(rt->dst.dev);
845 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
849 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
850 vif = vrf_master_ifindex_rcu(rt->dst.dev);
853 net = dev_net(rt->dst.dev);
854 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
856 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
857 rt_nexthop(rt, ip_hdr(skb)->daddr));
861 /* No redirected packets during ip_rt_redirect_silence;
862 * reset the algorithm.
864 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
865 peer->rate_tokens = 0;
867 /* Too many ignored redirects; do not send anything
868 * set dst.rate_last to the last seen redirected packet.
870 if (peer->rate_tokens >= ip_rt_redirect_number) {
871 peer->rate_last = jiffies;
875 /* Check for load limit; set rate_last to the latest sent
878 if (peer->rate_tokens == 0 ||
881 (ip_rt_redirect_load << peer->rate_tokens)))) {
882 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
884 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
885 peer->rate_last = jiffies;
887 #ifdef CONFIG_IP_ROUTE_VERBOSE
889 peer->rate_tokens == ip_rt_redirect_number)
890 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
891 &ip_hdr(skb)->saddr, inet_iif(skb),
892 &ip_hdr(skb)->daddr, &gw);
899 static int ip_error(struct sk_buff *skb)
901 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
902 struct rtable *rt = skb_rtable(skb);
903 struct inet_peer *peer;
909 /* IP on this device is disabled. */
913 net = dev_net(rt->dst.dev);
914 if (!IN_DEV_FORWARD(in_dev)) {
915 switch (rt->dst.error) {
917 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
921 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
927 switch (rt->dst.error) {
932 code = ICMP_HOST_UNREACH;
935 code = ICMP_NET_UNREACH;
936 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
939 code = ICMP_PKT_FILTERED;
943 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
944 vrf_master_ifindex(skb->dev), 1);
949 peer->rate_tokens += now - peer->rate_last;
950 if (peer->rate_tokens > ip_rt_error_burst)
951 peer->rate_tokens = ip_rt_error_burst;
952 peer->rate_last = now;
953 if (peer->rate_tokens >= ip_rt_error_cost)
954 peer->rate_tokens -= ip_rt_error_cost;
960 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
966 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
968 struct dst_entry *dst = &rt->dst;
969 struct fib_result res;
971 if (dst_metric_locked(dst, RTAX_MTU))
974 if (ipv4_mtu(dst) < mtu)
977 if (mtu < ip_rt_min_pmtu)
978 mtu = ip_rt_min_pmtu;
980 if (rt->rt_pmtu == mtu &&
981 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
985 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
986 struct fib_nh *nh = &FIB_RES_NH(res);
988 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
989 jiffies + ip_rt_mtu_expires);
994 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
995 struct sk_buff *skb, u32 mtu)
997 struct rtable *rt = (struct rtable *) dst;
1000 ip_rt_build_flow_key(&fl4, sk, skb);
1001 __ip_rt_update_pmtu(rt, &fl4, mtu);
1004 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1005 int oif, u32 mark, u8 protocol, int flow_flags)
1007 const struct iphdr *iph = (const struct iphdr *) skb->data;
1012 mark = IP4_REPLY_MARK(net, skb->mark);
1014 __build_flow_key(&fl4, NULL, iph, oif,
1015 RT_TOS(iph->tos), protocol, mark, flow_flags);
1016 rt = __ip_route_output_key(net, &fl4);
1018 __ip_rt_update_pmtu(rt, &fl4, mtu);
1022 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1024 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1026 const struct iphdr *iph = (const struct iphdr *) skb->data;
1030 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1032 if (!fl4.flowi4_mark)
1033 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1035 rt = __ip_route_output_key(sock_net(sk), &fl4);
1037 __ip_rt_update_pmtu(rt, &fl4, mtu);
1042 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1044 const struct iphdr *iph = (const struct iphdr *) skb->data;
1047 struct dst_entry *odst = NULL;
1052 if (!ip_sk_accept_pmtu(sk))
1055 odst = sk_dst_get(sk);
1057 if (sock_owned_by_user(sk) || !odst) {
1058 __ipv4_sk_update_pmtu(skb, sk, mtu);
1062 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1064 rt = (struct rtable *)odst;
1065 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1066 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1073 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1075 if (!dst_check(&rt->dst, 0)) {
1077 dst_release(&rt->dst);
1079 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1087 sk_dst_set(sk, &rt->dst);
1093 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1095 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1096 int oif, u32 mark, u8 protocol, int flow_flags)
1098 const struct iphdr *iph = (const struct iphdr *) skb->data;
1102 __build_flow_key(&fl4, NULL, iph, oif,
1103 RT_TOS(iph->tos), protocol, mark, flow_flags);
1104 rt = __ip_route_output_key(net, &fl4);
1106 __ip_do_redirect(rt, skb, &fl4, false);
1110 EXPORT_SYMBOL_GPL(ipv4_redirect);
1112 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1114 const struct iphdr *iph = (const struct iphdr *) skb->data;
1118 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1119 rt = __ip_route_output_key(sock_net(sk), &fl4);
1121 __ip_do_redirect(rt, skb, &fl4, false);
1125 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1127 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1129 struct rtable *rt = (struct rtable *) dst;
1131 /* All IPV4 dsts are created with ->obsolete set to the value
1132 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1133 * into this function always.
1135 * When a PMTU/redirect information update invalidates a route,
1136 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1137 * DST_OBSOLETE_DEAD by dst_free().
1139 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1144 static void ipv4_link_failure(struct sk_buff *skb)
1148 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1150 rt = skb_rtable(skb);
1152 dst_set_expires(&rt->dst, 0);
1155 static int ip_rt_bug(struct sock *sk, struct sk_buff *skb)
1157 pr_debug("%s: %pI4 -> %pI4, %s\n",
1158 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1159 skb->dev ? skb->dev->name : "?");
1166 We do not cache source address of outgoing interface,
1167 because it is used only by IP RR, TS and SRR options,
1168 so that it out of fast path.
1170 BTW remember: "addr" is allowed to be not aligned
1174 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1178 if (rt_is_output_route(rt))
1179 src = ip_hdr(skb)->saddr;
1181 struct fib_result res;
1187 memset(&fl4, 0, sizeof(fl4));
1188 fl4.daddr = iph->daddr;
1189 fl4.saddr = iph->saddr;
1190 fl4.flowi4_tos = RT_TOS(iph->tos);
1191 fl4.flowi4_oif = rt->dst.dev->ifindex;
1192 fl4.flowi4_iif = skb->dev->ifindex;
1193 fl4.flowi4_mark = skb->mark;
1196 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1197 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1199 src = inet_select_addr(rt->dst.dev,
1200 rt_nexthop(rt, iph->daddr),
1204 memcpy(addr, &src, 4);
1207 #ifdef CONFIG_IP_ROUTE_CLASSID
1208 static void set_class_tag(struct rtable *rt, u32 tag)
1210 if (!(rt->dst.tclassid & 0xFFFF))
1211 rt->dst.tclassid |= tag & 0xFFFF;
1212 if (!(rt->dst.tclassid & 0xFFFF0000))
1213 rt->dst.tclassid |= tag & 0xFFFF0000;
1217 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1219 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1222 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1224 if (advmss > 65535 - 40)
1225 advmss = 65535 - 40;
1230 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1232 const struct rtable *rt = (const struct rtable *) dst;
1233 unsigned int mtu = rt->rt_pmtu;
1235 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1236 mtu = dst_metric_raw(dst, RTAX_MTU);
1241 mtu = dst->dev->mtu;
1243 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1244 if (rt->rt_uses_gateway && mtu > 576)
1248 return min_t(unsigned int, mtu, IP_MAX_MTU);
1251 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1253 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1254 struct fib_nh_exception *fnhe;
1260 hval = fnhe_hashfun(daddr);
1262 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1263 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1264 if (fnhe->fnhe_daddr == daddr)
1270 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1275 spin_lock_bh(&fnhe_lock);
1277 if (daddr == fnhe->fnhe_daddr) {
1278 struct rtable __rcu **porig;
1279 struct rtable *orig;
1280 int genid = fnhe_genid(dev_net(rt->dst.dev));
1282 if (rt_is_input_route(rt))
1283 porig = &fnhe->fnhe_rth_input;
1285 porig = &fnhe->fnhe_rth_output;
1286 orig = rcu_dereference(*porig);
1288 if (fnhe->fnhe_genid != genid) {
1289 fnhe->fnhe_genid = genid;
1291 fnhe->fnhe_pmtu = 0;
1292 fnhe->fnhe_expires = 0;
1293 fnhe_flush_routes(fnhe);
1296 fill_route_from_fnhe(rt, fnhe);
1297 if (!rt->rt_gateway)
1298 rt->rt_gateway = daddr;
1300 if (!(rt->dst.flags & DST_NOCACHE)) {
1301 rcu_assign_pointer(*porig, rt);
1307 fnhe->fnhe_stamp = jiffies;
1309 spin_unlock_bh(&fnhe_lock);
1314 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1316 struct rtable *orig, *prev, **p;
1319 if (rt_is_input_route(rt)) {
1320 p = (struct rtable **)&nh->nh_rth_input;
1322 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1326 prev = cmpxchg(p, orig, rt);
1336 struct uncached_list {
1338 struct list_head head;
1341 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1343 static void rt_add_uncached_list(struct rtable *rt)
1345 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1347 rt->rt_uncached_list = ul;
1349 spin_lock_bh(&ul->lock);
1350 list_add_tail(&rt->rt_uncached, &ul->head);
1351 spin_unlock_bh(&ul->lock);
1354 static void ipv4_dst_destroy(struct dst_entry *dst)
1356 struct rtable *rt = (struct rtable *) dst;
1358 if (!list_empty(&rt->rt_uncached)) {
1359 struct uncached_list *ul = rt->rt_uncached_list;
1361 spin_lock_bh(&ul->lock);
1362 list_del(&rt->rt_uncached);
1363 spin_unlock_bh(&ul->lock);
1367 void rt_flush_dev(struct net_device *dev)
1369 struct net *net = dev_net(dev);
1373 for_each_possible_cpu(cpu) {
1374 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1376 spin_lock_bh(&ul->lock);
1377 list_for_each_entry(rt, &ul->head, rt_uncached) {
1378 if (rt->dst.dev != dev)
1380 rt->dst.dev = net->loopback_dev;
1381 dev_hold(rt->dst.dev);
1384 spin_unlock_bh(&ul->lock);
1388 static bool rt_cache_valid(const struct rtable *rt)
1391 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1395 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1396 const struct fib_result *res,
1397 struct fib_nh_exception *fnhe,
1398 struct fib_info *fi, u16 type, u32 itag)
1400 bool cached = false;
1403 struct fib_nh *nh = &FIB_RES_NH(*res);
1405 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1406 rt->rt_gateway = nh->nh_gw;
1407 rt->rt_uses_gateway = 1;
1409 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1410 #ifdef CONFIG_IP_ROUTE_CLASSID
1411 rt->dst.tclassid = nh->nh_tclassid;
1413 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1415 cached = rt_bind_exception(rt, fnhe, daddr);
1416 else if (!(rt->dst.flags & DST_NOCACHE))
1417 cached = rt_cache_route(nh, rt);
1418 if (unlikely(!cached)) {
1419 /* Routes we intend to cache in nexthop exception or
1420 * FIB nexthop have the DST_NOCACHE bit clear.
1421 * However, if we are unsuccessful at storing this
1422 * route into the cache we really need to set it.
1424 rt->dst.flags |= DST_NOCACHE;
1425 if (!rt->rt_gateway)
1426 rt->rt_gateway = daddr;
1427 rt_add_uncached_list(rt);
1430 rt_add_uncached_list(rt);
1432 #ifdef CONFIG_IP_ROUTE_CLASSID
1433 #ifdef CONFIG_IP_MULTIPLE_TABLES
1434 set_class_tag(rt, res->tclassid);
1436 set_class_tag(rt, itag);
1440 static struct rtable *rt_dst_alloc(struct net_device *dev,
1441 bool nopolicy, bool noxfrm, bool will_cache)
1443 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1444 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1445 (nopolicy ? DST_NOPOLICY : 0) |
1446 (noxfrm ? DST_NOXFRM : 0));
1449 /* called in rcu_read_lock() section */
1450 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1451 u8 tos, struct net_device *dev, int our)
1454 struct in_device *in_dev = __in_dev_get_rcu(dev);
1458 /* Primary sanity checks. */
1463 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1464 skb->protocol != htons(ETH_P_IP))
1467 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1468 if (ipv4_is_loopback(saddr))
1471 if (ipv4_is_zeronet(saddr)) {
1472 if (!ipv4_is_local_multicast(daddr))
1475 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1480 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1481 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1485 #ifdef CONFIG_IP_ROUTE_CLASSID
1486 rth->dst.tclassid = itag;
1488 rth->dst.output = ip_rt_bug;
1490 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
1491 rth->rt_flags = RTCF_MULTICAST;
1492 rth->rt_type = RTN_MULTICAST;
1493 rth->rt_is_input= 1;
1496 rth->rt_gateway = 0;
1497 rth->rt_uses_gateway = 0;
1498 INIT_LIST_HEAD(&rth->rt_uncached);
1500 rth->dst.input= ip_local_deliver;
1501 rth->rt_flags |= RTCF_LOCAL;
1504 #ifdef CONFIG_IP_MROUTE
1505 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1506 rth->dst.input = ip_mr_input;
1508 RT_CACHE_STAT_INC(in_slow_mc);
1510 skb_dst_set(skb, &rth->dst);
1522 static void ip_handle_martian_source(struct net_device *dev,
1523 struct in_device *in_dev,
1524 struct sk_buff *skb,
1528 RT_CACHE_STAT_INC(in_martian_src);
1529 #ifdef CONFIG_IP_ROUTE_VERBOSE
1530 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1532 * RFC1812 recommendation, if source is martian,
1533 * the only hint is MAC header.
1535 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1536 &daddr, &saddr, dev->name);
1537 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1538 print_hex_dump(KERN_WARNING, "ll header: ",
1539 DUMP_PREFIX_OFFSET, 16, 1,
1540 skb_mac_header(skb),
1541 dev->hard_header_len, true);
1547 /* called in rcu_read_lock() section */
1548 static int __mkroute_input(struct sk_buff *skb,
1549 const struct fib_result *res,
1550 struct in_device *in_dev,
1551 __be32 daddr, __be32 saddr, u32 tos)
1553 struct fib_nh_exception *fnhe;
1556 struct in_device *out_dev;
1560 /* get a working reference to the output device */
1561 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1563 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1567 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1568 in_dev->dev, in_dev, &itag);
1570 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1576 do_cache = res->fi && !itag;
1577 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1578 skb->protocol == htons(ETH_P_IP) &&
1579 (IN_DEV_SHARED_MEDIA(out_dev) ||
1580 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1581 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1583 if (skb->protocol != htons(ETH_P_IP)) {
1584 /* Not IP (i.e. ARP). Do not create route, if it is
1585 * invalid for proxy arp. DNAT routes are always valid.
1587 * Proxy arp feature have been extended to allow, ARP
1588 * replies back to the same interface, to support
1589 * Private VLAN switch technologies. See arp.c.
1591 if (out_dev == in_dev &&
1592 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1598 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1601 rth = rcu_dereference(fnhe->fnhe_rth_input);
1603 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1605 if (rt_cache_valid(rth)) {
1606 skb_dst_set_noref(skb, &rth->dst);
1611 rth = rt_dst_alloc(out_dev->dev,
1612 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1613 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1619 rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1621 rth->rt_type = res->type;
1622 rth->rt_is_input = 1;
1625 rth->rt_gateway = 0;
1626 rth->rt_uses_gateway = 0;
1627 INIT_LIST_HEAD(&rth->rt_uncached);
1628 RT_CACHE_STAT_INC(in_slow_tot);
1630 rth->dst.input = ip_forward;
1631 rth->dst.output = ip_output;
1633 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1634 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1635 rth->dst.lwtstate->orig_output = rth->dst.output;
1636 rth->dst.output = lwtunnel_output;
1638 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1639 rth->dst.lwtstate->orig_input = rth->dst.input;
1640 rth->dst.input = lwtunnel_input;
1642 skb_dst_set(skb, &rth->dst);
1649 static int ip_mkroute_input(struct sk_buff *skb,
1650 struct fib_result *res,
1651 const struct flowi4 *fl4,
1652 struct in_device *in_dev,
1653 __be32 daddr, __be32 saddr, u32 tos)
1655 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1656 if (res->fi && res->fi->fib_nhs > 1)
1657 fib_select_multipath(res);
1660 /* create a routing cache entry */
1661 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1665 * NOTE. We drop all the packets that has local source
1666 * addresses, because every properly looped back packet
1667 * must have correct destination already attached by output routine.
1669 * Such approach solves two big problems:
1670 * 1. Not simplex devices are handled properly.
1671 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1672 * called with rcu_read_lock()
1675 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1676 u8 tos, struct net_device *dev)
1678 struct fib_result res;
1679 struct in_device *in_dev = __in_dev_get_rcu(dev);
1680 struct ip_tunnel_info *tun_info;
1682 unsigned int flags = 0;
1686 struct net *net = dev_net(dev);
1689 /* IP on this device is disabled. */
1694 /* Check for the most weird martians, which can be not detected
1698 tun_info = skb_tunnel_info(skb);
1699 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1700 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1702 fl4.flowi4_tun_key.tun_id = 0;
1705 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1706 goto martian_source;
1709 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1712 /* Accept zero addresses only to limited broadcast;
1713 * I even do not know to fix it or not. Waiting for complains :-)
1715 if (ipv4_is_zeronet(saddr))
1716 goto martian_source;
1718 if (ipv4_is_zeronet(daddr))
1719 goto martian_destination;
1721 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1722 * and call it once if daddr or/and saddr are loopback addresses
1724 if (ipv4_is_loopback(daddr)) {
1725 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1726 goto martian_destination;
1727 } else if (ipv4_is_loopback(saddr)) {
1728 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1729 goto martian_source;
1733 * Now we are ready to route packet.
1736 fl4.flowi4_iif = vrf_master_ifindex_rcu(dev) ? : dev->ifindex;
1737 fl4.flowi4_mark = skb->mark;
1738 fl4.flowi4_tos = tos;
1739 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1742 err = fib_lookup(net, &fl4, &res, 0);
1744 if (!IN_DEV_FORWARD(in_dev))
1745 err = -EHOSTUNREACH;
1749 if (res.type == RTN_BROADCAST)
1752 if (res.type == RTN_LOCAL) {
1753 err = fib_validate_source(skb, saddr, daddr, tos,
1754 0, dev, in_dev, &itag);
1756 goto martian_source_keep_err;
1760 if (!IN_DEV_FORWARD(in_dev)) {
1761 err = -EHOSTUNREACH;
1764 if (res.type != RTN_UNICAST)
1765 goto martian_destination;
1767 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1771 if (skb->protocol != htons(ETH_P_IP))
1774 if (!ipv4_is_zeronet(saddr)) {
1775 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1778 goto martian_source_keep_err;
1780 flags |= RTCF_BROADCAST;
1781 res.type = RTN_BROADCAST;
1782 RT_CACHE_STAT_INC(in_brd);
1788 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1789 if (rt_cache_valid(rth)) {
1790 skb_dst_set_noref(skb, &rth->dst);
1798 rth = rt_dst_alloc(net->loopback_dev,
1799 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1803 rth->dst.input= ip_local_deliver;
1804 rth->dst.output= ip_rt_bug;
1805 #ifdef CONFIG_IP_ROUTE_CLASSID
1806 rth->dst.tclassid = itag;
1809 rth->rt_genid = rt_genid_ipv4(net);
1810 rth->rt_flags = flags|RTCF_LOCAL;
1811 rth->rt_type = res.type;
1812 rth->rt_is_input = 1;
1815 rth->rt_gateway = 0;
1816 rth->rt_uses_gateway = 0;
1817 INIT_LIST_HEAD(&rth->rt_uncached);
1819 RT_CACHE_STAT_INC(in_slow_tot);
1820 if (res.type == RTN_UNREACHABLE) {
1821 rth->dst.input= ip_error;
1822 rth->dst.error= -err;
1823 rth->rt_flags &= ~RTCF_LOCAL;
1826 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1827 rth->dst.flags |= DST_NOCACHE;
1828 rt_add_uncached_list(rth);
1831 skb_dst_set(skb, &rth->dst);
1836 RT_CACHE_STAT_INC(in_no_route);
1837 res.type = RTN_UNREACHABLE;
1842 * Do not cache martian addresses: they should be logged (RFC1812)
1844 martian_destination:
1845 RT_CACHE_STAT_INC(in_martian_dst);
1846 #ifdef CONFIG_IP_ROUTE_VERBOSE
1847 if (IN_DEV_LOG_MARTIANS(in_dev))
1848 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1849 &daddr, &saddr, dev->name);
1862 martian_source_keep_err:
1863 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1867 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1868 u8 tos, struct net_device *dev)
1874 /* Multicast recognition logic is moved from route cache to here.
1875 The problem was that too many Ethernet cards have broken/missing
1876 hardware multicast filters :-( As result the host on multicasting
1877 network acquires a lot of useless route cache entries, sort of
1878 SDR messages from all the world. Now we try to get rid of them.
1879 Really, provided software IP multicast filter is organized
1880 reasonably (at least, hashed), it does not result in a slowdown
1881 comparing with route cache reject entries.
1882 Note, that multicast routers are not affected, because
1883 route cache entry is created eventually.
1885 if (ipv4_is_multicast(daddr)) {
1886 struct in_device *in_dev = __in_dev_get_rcu(dev);
1889 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1890 ip_hdr(skb)->protocol);
1892 #ifdef CONFIG_IP_MROUTE
1894 (!ipv4_is_local_multicast(daddr) &&
1895 IN_DEV_MFORWARD(in_dev))
1898 int res = ip_route_input_mc(skb, daddr, saddr,
1907 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1911 EXPORT_SYMBOL(ip_route_input_noref);
1913 /* called with rcu_read_lock() */
1914 static struct rtable *__mkroute_output(const struct fib_result *res,
1915 const struct flowi4 *fl4, int orig_oif,
1916 struct net_device *dev_out,
1919 struct fib_info *fi = res->fi;
1920 struct fib_nh_exception *fnhe;
1921 struct in_device *in_dev;
1922 u16 type = res->type;
1926 in_dev = __in_dev_get_rcu(dev_out);
1928 return ERR_PTR(-EINVAL);
1930 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1931 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1932 return ERR_PTR(-EINVAL);
1934 if (ipv4_is_lbcast(fl4->daddr))
1935 type = RTN_BROADCAST;
1936 else if (ipv4_is_multicast(fl4->daddr))
1937 type = RTN_MULTICAST;
1938 else if (ipv4_is_zeronet(fl4->daddr))
1939 return ERR_PTR(-EINVAL);
1941 if (dev_out->flags & IFF_LOOPBACK)
1942 flags |= RTCF_LOCAL;
1945 if (type == RTN_BROADCAST) {
1946 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1948 } else if (type == RTN_MULTICAST) {
1949 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1950 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1952 flags &= ~RTCF_LOCAL;
1955 /* If multicast route do not exist use
1956 * default one, but do not gateway in this case.
1959 if (fi && res->prefixlen < 4)
1964 do_cache &= fi != NULL;
1966 struct rtable __rcu **prth;
1967 struct fib_nh *nh = &FIB_RES_NH(*res);
1969 fnhe = find_exception(nh, fl4->daddr);
1971 prth = &fnhe->fnhe_rth_output;
1973 if (unlikely(fl4->flowi4_flags &
1974 FLOWI_FLAG_KNOWN_NH &&
1976 nh->nh_scope == RT_SCOPE_LINK))) {
1980 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
1982 rth = rcu_dereference(*prth);
1983 if (rt_cache_valid(rth)) {
1984 dst_hold(&rth->dst);
1990 rth = rt_dst_alloc(dev_out,
1991 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1992 IN_DEV_CONF_GET(in_dev, NOXFRM),
1995 return ERR_PTR(-ENOBUFS);
1997 rth->dst.output = ip_output;
1999 rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
2000 rth->rt_flags = flags;
2001 rth->rt_type = type;
2002 rth->rt_is_input = 0;
2003 rth->rt_iif = orig_oif ? : 0;
2005 rth->rt_gateway = 0;
2006 rth->rt_uses_gateway = 0;
2007 INIT_LIST_HEAD(&rth->rt_uncached);
2008 RT_CACHE_STAT_INC(out_slow_tot);
2010 if (flags & RTCF_LOCAL)
2011 rth->dst.input = ip_local_deliver;
2012 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2013 if (flags & RTCF_LOCAL &&
2014 !(dev_out->flags & IFF_LOOPBACK)) {
2015 rth->dst.output = ip_mc_output;
2016 RT_CACHE_STAT_INC(out_slow_mc);
2018 #ifdef CONFIG_IP_MROUTE
2019 if (type == RTN_MULTICAST) {
2020 if (IN_DEV_MFORWARD(in_dev) &&
2021 !ipv4_is_local_multicast(fl4->daddr)) {
2022 rth->dst.input = ip_mr_input;
2023 rth->dst.output = ip_mc_output;
2029 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2030 if (lwtunnel_output_redirect(rth->dst.lwtstate))
2031 rth->dst.output = lwtunnel_output;
2037 * Major route resolver routine.
2040 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2042 struct net_device *dev_out = NULL;
2043 __u8 tos = RT_FL_TOS(fl4);
2044 unsigned int flags = 0;
2045 struct fib_result res;
2053 orig_oif = fl4->flowi4_oif;
2055 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2056 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2057 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2058 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2062 rth = ERR_PTR(-EINVAL);
2063 if (ipv4_is_multicast(fl4->saddr) ||
2064 ipv4_is_lbcast(fl4->saddr) ||
2065 ipv4_is_zeronet(fl4->saddr))
2068 /* I removed check for oif == dev_out->oif here.
2069 It was wrong for two reasons:
2070 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2071 is assigned to multiple interfaces.
2072 2. Moreover, we are allowed to send packets with saddr
2073 of another iface. --ANK
2076 if (fl4->flowi4_oif == 0 &&
2077 (ipv4_is_multicast(fl4->daddr) ||
2078 ipv4_is_lbcast(fl4->daddr))) {
2079 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2080 dev_out = __ip_dev_find(net, fl4->saddr, false);
2084 /* Special hack: user can direct multicasts
2085 and limited broadcast via necessary interface
2086 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2087 This hack is not just for fun, it allows
2088 vic,vat and friends to work.
2089 They bind socket to loopback, set ttl to zero
2090 and expect that it will work.
2091 From the viewpoint of routing cache they are broken,
2092 because we are not allowed to build multicast path
2093 with loopback source addr (look, routing cache
2094 cannot know, that ttl is zero, so that packet
2095 will not leave this host and route is valid).
2096 Luckily, this hack is good workaround.
2099 fl4->flowi4_oif = dev_out->ifindex;
2103 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2104 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2105 if (!__ip_dev_find(net, fl4->saddr, false))
2111 if (fl4->flowi4_oif) {
2112 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2113 rth = ERR_PTR(-ENODEV);
2117 /* RACE: Check return value of inet_select_addr instead. */
2118 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2119 rth = ERR_PTR(-ENETUNREACH);
2122 if (ipv4_is_local_multicast(fl4->daddr) ||
2123 ipv4_is_lbcast(fl4->daddr) ||
2124 fl4->flowi4_proto == IPPROTO_IGMP) {
2126 fl4->saddr = inet_select_addr(dev_out, 0,
2131 if (ipv4_is_multicast(fl4->daddr))
2132 fl4->saddr = inet_select_addr(dev_out, 0,
2134 else if (!fl4->daddr)
2135 fl4->saddr = inet_select_addr(dev_out, 0,
2138 if (netif_is_vrf(dev_out) &&
2139 !(fl4->flowi4_flags & FLOWI_FLAG_VRFSRC)) {
2140 rth = vrf_dev_get_rth(dev_out);
2146 fl4->daddr = fl4->saddr;
2148 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2149 dev_out = net->loopback_dev;
2150 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2151 res.type = RTN_LOCAL;
2152 flags |= RTCF_LOCAL;
2156 if (fib_lookup(net, fl4, &res, 0)) {
2159 if (fl4->flowi4_oif) {
2160 /* Apparently, routing tables are wrong. Assume,
2161 that the destination is on link.
2164 Because we are allowed to send to iface
2165 even if it has NO routes and NO assigned
2166 addresses. When oif is specified, routing
2167 tables are looked up with only one purpose:
2168 to catch if destination is gatewayed, rather than
2169 direct. Moreover, if MSG_DONTROUTE is set,
2170 we send packet, ignoring both routing tables
2171 and ifaddr state. --ANK
2174 We could make it even if oif is unknown,
2175 likely IPv6, but we do not.
2178 if (fl4->saddr == 0)
2179 fl4->saddr = inet_select_addr(dev_out, 0,
2181 res.type = RTN_UNICAST;
2184 rth = ERR_PTR(-ENETUNREACH);
2188 if (res.type == RTN_LOCAL) {
2190 if (res.fi->fib_prefsrc)
2191 fl4->saddr = res.fi->fib_prefsrc;
2193 fl4->saddr = fl4->daddr;
2195 dev_out = net->loopback_dev;
2196 fl4->flowi4_oif = dev_out->ifindex;
2197 flags |= RTCF_LOCAL;
2201 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2202 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2203 fib_select_multipath(&res);
2206 if (!res.prefixlen &&
2207 res.table->tb_num_default > 1 &&
2208 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2209 fib_select_default(fl4, &res);
2212 fl4->saddr = FIB_RES_PREFSRC(net, res);
2214 dev_out = FIB_RES_DEV(res);
2215 fl4->flowi4_oif = dev_out->ifindex;
2219 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2225 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2227 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2232 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2234 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2236 return mtu ? : dst->dev->mtu;
2239 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2240 struct sk_buff *skb, u32 mtu)
2244 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2245 struct sk_buff *skb)
2249 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2255 static struct dst_ops ipv4_dst_blackhole_ops = {
2257 .check = ipv4_blackhole_dst_check,
2258 .mtu = ipv4_blackhole_mtu,
2259 .default_advmss = ipv4_default_advmss,
2260 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2261 .redirect = ipv4_rt_blackhole_redirect,
2262 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2263 .neigh_lookup = ipv4_neigh_lookup,
2266 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2268 struct rtable *ort = (struct rtable *) dst_orig;
2271 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2273 struct dst_entry *new = &rt->dst;
2276 new->input = dst_discard;
2277 new->output = dst_discard_sk;
2279 new->dev = ort->dst.dev;
2283 rt->rt_is_input = ort->rt_is_input;
2284 rt->rt_iif = ort->rt_iif;
2285 rt->rt_pmtu = ort->rt_pmtu;
2287 rt->rt_genid = rt_genid_ipv4(net);
2288 rt->rt_flags = ort->rt_flags;
2289 rt->rt_type = ort->rt_type;
2290 rt->rt_gateway = ort->rt_gateway;
2291 rt->rt_uses_gateway = ort->rt_uses_gateway;
2293 INIT_LIST_HEAD(&rt->rt_uncached);
2297 dst_release(dst_orig);
2299 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2302 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2305 struct rtable *rt = __ip_route_output_key(net, flp4);
2310 if (flp4->flowi4_proto)
2311 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2312 flowi4_to_flowi(flp4),
2317 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2319 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2320 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2321 u32 seq, int event, int nowait, unsigned int flags)
2323 struct rtable *rt = skb_rtable(skb);
2325 struct nlmsghdr *nlh;
2326 unsigned long expires = 0;
2328 u32 metrics[RTAX_MAX];
2330 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2334 r = nlmsg_data(nlh);
2335 r->rtm_family = AF_INET;
2336 r->rtm_dst_len = 32;
2338 r->rtm_tos = fl4->flowi4_tos;
2339 r->rtm_table = RT_TABLE_MAIN;
2340 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2341 goto nla_put_failure;
2342 r->rtm_type = rt->rt_type;
2343 r->rtm_scope = RT_SCOPE_UNIVERSE;
2344 r->rtm_protocol = RTPROT_UNSPEC;
2345 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2346 if (rt->rt_flags & RTCF_NOTIFY)
2347 r->rtm_flags |= RTM_F_NOTIFY;
2348 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2349 r->rtm_flags |= RTCF_DOREDIRECT;
2351 if (nla_put_in_addr(skb, RTA_DST, dst))
2352 goto nla_put_failure;
2354 r->rtm_src_len = 32;
2355 if (nla_put_in_addr(skb, RTA_SRC, src))
2356 goto nla_put_failure;
2359 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2360 goto nla_put_failure;
2361 #ifdef CONFIG_IP_ROUTE_CLASSID
2362 if (rt->dst.tclassid &&
2363 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2364 goto nla_put_failure;
2366 if (!rt_is_input_route(rt) &&
2367 fl4->saddr != src) {
2368 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2369 goto nla_put_failure;
2371 if (rt->rt_uses_gateway &&
2372 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2373 goto nla_put_failure;
2375 expires = rt->dst.expires;
2377 unsigned long now = jiffies;
2379 if (time_before(now, expires))
2385 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2386 if (rt->rt_pmtu && expires)
2387 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2388 if (rtnetlink_put_metrics(skb, metrics) < 0)
2389 goto nla_put_failure;
2391 if (fl4->flowi4_mark &&
2392 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2393 goto nla_put_failure;
2395 error = rt->dst.error;
2397 if (rt_is_input_route(rt)) {
2398 #ifdef CONFIG_IP_MROUTE
2399 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2400 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2401 int err = ipmr_get_route(net, skb,
2402 fl4->saddr, fl4->daddr,
2408 goto nla_put_failure;
2410 if (err == -EMSGSIZE)
2411 goto nla_put_failure;
2417 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2418 goto nla_put_failure;
2421 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2422 goto nla_put_failure;
2424 nlmsg_end(skb, nlh);
2428 nlmsg_cancel(skb, nlh);
2432 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2434 struct net *net = sock_net(in_skb->sk);
2436 struct nlattr *tb[RTA_MAX+1];
2437 struct rtable *rt = NULL;
2444 struct sk_buff *skb;
2446 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2450 rtm = nlmsg_data(nlh);
2452 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2458 /* Reserve room for dummy headers, this skb can pass
2459 through good chunk of routing engine.
2461 skb_reset_mac_header(skb);
2462 skb_reset_network_header(skb);
2464 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2465 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2466 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2468 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2469 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2470 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2471 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2473 memset(&fl4, 0, sizeof(fl4));
2476 fl4.flowi4_tos = rtm->rtm_tos;
2477 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2478 fl4.flowi4_mark = mark;
2481 struct net_device *dev;
2483 dev = __dev_get_by_index(net, iif);
2489 skb->protocol = htons(ETH_P_IP);
2493 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2496 rt = skb_rtable(skb);
2497 if (err == 0 && rt->dst.error)
2498 err = -rt->dst.error;
2500 rt = ip_route_output_key(net, &fl4);
2510 skb_dst_set(skb, &rt->dst);
2511 if (rtm->rtm_flags & RTM_F_NOTIFY)
2512 rt->rt_flags |= RTCF_NOTIFY;
2514 err = rt_fill_info(net, dst, src, &fl4, skb,
2515 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2516 RTM_NEWROUTE, 0, 0);
2520 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2529 void ip_rt_multicast_event(struct in_device *in_dev)
2531 rt_cache_flush(dev_net(in_dev->dev));
2534 #ifdef CONFIG_SYSCTL
2535 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2536 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2537 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2538 static int ip_rt_gc_elasticity __read_mostly = 8;
2540 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2541 void __user *buffer,
2542 size_t *lenp, loff_t *ppos)
2544 struct net *net = (struct net *)__ctl->extra1;
2547 rt_cache_flush(net);
2548 fnhe_genid_bump(net);
2555 static struct ctl_table ipv4_route_table[] = {
2557 .procname = "gc_thresh",
2558 .data = &ipv4_dst_ops.gc_thresh,
2559 .maxlen = sizeof(int),
2561 .proc_handler = proc_dointvec,
2564 .procname = "max_size",
2565 .data = &ip_rt_max_size,
2566 .maxlen = sizeof(int),
2568 .proc_handler = proc_dointvec,
2571 /* Deprecated. Use gc_min_interval_ms */
2573 .procname = "gc_min_interval",
2574 .data = &ip_rt_gc_min_interval,
2575 .maxlen = sizeof(int),
2577 .proc_handler = proc_dointvec_jiffies,
2580 .procname = "gc_min_interval_ms",
2581 .data = &ip_rt_gc_min_interval,
2582 .maxlen = sizeof(int),
2584 .proc_handler = proc_dointvec_ms_jiffies,
2587 .procname = "gc_timeout",
2588 .data = &ip_rt_gc_timeout,
2589 .maxlen = sizeof(int),
2591 .proc_handler = proc_dointvec_jiffies,
2594 .procname = "gc_interval",
2595 .data = &ip_rt_gc_interval,
2596 .maxlen = sizeof(int),
2598 .proc_handler = proc_dointvec_jiffies,
2601 .procname = "redirect_load",
2602 .data = &ip_rt_redirect_load,
2603 .maxlen = sizeof(int),
2605 .proc_handler = proc_dointvec,
2608 .procname = "redirect_number",
2609 .data = &ip_rt_redirect_number,
2610 .maxlen = sizeof(int),
2612 .proc_handler = proc_dointvec,
2615 .procname = "redirect_silence",
2616 .data = &ip_rt_redirect_silence,
2617 .maxlen = sizeof(int),
2619 .proc_handler = proc_dointvec,
2622 .procname = "error_cost",
2623 .data = &ip_rt_error_cost,
2624 .maxlen = sizeof(int),
2626 .proc_handler = proc_dointvec,
2629 .procname = "error_burst",
2630 .data = &ip_rt_error_burst,
2631 .maxlen = sizeof(int),
2633 .proc_handler = proc_dointvec,
2636 .procname = "gc_elasticity",
2637 .data = &ip_rt_gc_elasticity,
2638 .maxlen = sizeof(int),
2640 .proc_handler = proc_dointvec,
2643 .procname = "mtu_expires",
2644 .data = &ip_rt_mtu_expires,
2645 .maxlen = sizeof(int),
2647 .proc_handler = proc_dointvec_jiffies,
2650 .procname = "min_pmtu",
2651 .data = &ip_rt_min_pmtu,
2652 .maxlen = sizeof(int),
2654 .proc_handler = proc_dointvec,
2657 .procname = "min_adv_mss",
2658 .data = &ip_rt_min_advmss,
2659 .maxlen = sizeof(int),
2661 .proc_handler = proc_dointvec,
2666 static struct ctl_table ipv4_route_flush_table[] = {
2668 .procname = "flush",
2669 .maxlen = sizeof(int),
2671 .proc_handler = ipv4_sysctl_rtcache_flush,
2676 static __net_init int sysctl_route_net_init(struct net *net)
2678 struct ctl_table *tbl;
2680 tbl = ipv4_route_flush_table;
2681 if (!net_eq(net, &init_net)) {
2682 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2686 /* Don't export sysctls to unprivileged users */
2687 if (net->user_ns != &init_user_ns)
2688 tbl[0].procname = NULL;
2690 tbl[0].extra1 = net;
2692 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2693 if (!net->ipv4.route_hdr)
2698 if (tbl != ipv4_route_flush_table)
2704 static __net_exit void sysctl_route_net_exit(struct net *net)
2706 struct ctl_table *tbl;
2708 tbl = net->ipv4.route_hdr->ctl_table_arg;
2709 unregister_net_sysctl_table(net->ipv4.route_hdr);
2710 BUG_ON(tbl == ipv4_route_flush_table);
2714 static __net_initdata struct pernet_operations sysctl_route_ops = {
2715 .init = sysctl_route_net_init,
2716 .exit = sysctl_route_net_exit,
2720 static __net_init int rt_genid_init(struct net *net)
2722 atomic_set(&net->ipv4.rt_genid, 0);
2723 atomic_set(&net->fnhe_genid, 0);
2724 get_random_bytes(&net->ipv4.dev_addr_genid,
2725 sizeof(net->ipv4.dev_addr_genid));
2729 static __net_initdata struct pernet_operations rt_genid_ops = {
2730 .init = rt_genid_init,
2733 static int __net_init ipv4_inetpeer_init(struct net *net)
2735 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2739 inet_peer_base_init(bp);
2740 net->ipv4.peers = bp;
2744 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2746 struct inet_peer_base *bp = net->ipv4.peers;
2748 net->ipv4.peers = NULL;
2749 inetpeer_invalidate_tree(bp);
2753 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2754 .init = ipv4_inetpeer_init,
2755 .exit = ipv4_inetpeer_exit,
2758 #ifdef CONFIG_IP_ROUTE_CLASSID
2759 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2760 #endif /* CONFIG_IP_ROUTE_CLASSID */
2762 int __init ip_rt_init(void)
2767 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2769 panic("IP: failed to allocate ip_idents\n");
2771 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2773 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2775 panic("IP: failed to allocate ip_tstamps\n");
2777 for_each_possible_cpu(cpu) {
2778 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2780 INIT_LIST_HEAD(&ul->head);
2781 spin_lock_init(&ul->lock);
2783 #ifdef CONFIG_IP_ROUTE_CLASSID
2784 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2786 panic("IP: failed to allocate ip_rt_acct\n");
2789 ipv4_dst_ops.kmem_cachep =
2790 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2791 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2793 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2795 if (dst_entries_init(&ipv4_dst_ops) < 0)
2796 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2798 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2799 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2801 ipv4_dst_ops.gc_thresh = ~0;
2802 ip_rt_max_size = INT_MAX;
2807 if (ip_rt_proc_init())
2808 pr_err("Unable to create route proc files\n");
2813 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2815 #ifdef CONFIG_SYSCTL
2816 register_pernet_subsys(&sysctl_route_ops);
2818 register_pernet_subsys(&rt_genid_ops);
2819 register_pernet_subsys(&ipv4_inetpeer_ops);
2823 #ifdef CONFIG_SYSCTL
2825 * We really need to sanitize the damn ipv4 init order, then all
2826 * this nonsense will go away.
2828 void __init ip_static_sysctl_init(void)
2830 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);