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>
116 #define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119 #define RT_GC_TIMEOUT (300*HZ)
121 static int ip_rt_max_size;
122 static int ip_rt_redirect_number __read_mostly = 9;
123 static int ip_rt_redirect_load __read_mostly = HZ / 50;
124 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
125 static int ip_rt_error_cost __read_mostly = HZ;
126 static int ip_rt_error_burst __read_mostly = 5 * HZ;
127 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
128 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
129 static int ip_rt_min_advmss __read_mostly = 256;
132 * Interface to generic destination cache.
135 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
136 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
137 static unsigned int ipv4_mtu(const struct dst_entry *dst);
138 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
139 static void ipv4_link_failure(struct sk_buff *skb);
140 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
141 struct sk_buff *skb, u32 mtu);
142 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
143 struct sk_buff *skb);
144 static void ipv4_dst_destroy(struct dst_entry *dst);
146 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
152 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
156 static struct dst_ops ipv4_dst_ops = {
158 .check = ipv4_dst_check,
159 .default_advmss = ipv4_default_advmss,
161 .cow_metrics = ipv4_cow_metrics,
162 .destroy = ipv4_dst_destroy,
163 .negative_advice = ipv4_negative_advice,
164 .link_failure = ipv4_link_failure,
165 .update_pmtu = ip_rt_update_pmtu,
166 .redirect = ip_do_redirect,
167 .local_out = __ip_local_out,
168 .neigh_lookup = ipv4_neigh_lookup,
171 #define ECN_OR_COST(class) TC_PRIO_##class
173 const __u8 ip_tos2prio[16] = {
175 ECN_OR_COST(BESTEFFORT),
177 ECN_OR_COST(BESTEFFORT),
183 ECN_OR_COST(INTERACTIVE),
185 ECN_OR_COST(INTERACTIVE),
186 TC_PRIO_INTERACTIVE_BULK,
187 ECN_OR_COST(INTERACTIVE_BULK),
188 TC_PRIO_INTERACTIVE_BULK,
189 ECN_OR_COST(INTERACTIVE_BULK)
191 EXPORT_SYMBOL(ip_tos2prio);
193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
196 #ifdef CONFIG_PROC_FS
197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201 return SEQ_START_TOKEN;
204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
216 if (v == SEQ_START_TOKEN)
217 seq_printf(seq, "%-127s\n",
218 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
219 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
224 static const struct seq_operations rt_cache_seq_ops = {
225 .start = rt_cache_seq_start,
226 .next = rt_cache_seq_next,
227 .stop = rt_cache_seq_stop,
228 .show = rt_cache_seq_show,
231 static int rt_cache_seq_open(struct inode *inode, struct file *file)
233 return seq_open(file, &rt_cache_seq_ops);
236 static const struct file_operations rt_cache_seq_fops = {
237 .owner = THIS_MODULE,
238 .open = rt_cache_seq_open,
241 .release = seq_release,
245 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
250 return SEQ_START_TOKEN;
252 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
253 if (!cpu_possible(cpu))
256 return &per_cpu(rt_cache_stat, cpu);
261 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
265 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
266 if (!cpu_possible(cpu))
269 return &per_cpu(rt_cache_stat, cpu);
275 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
280 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
282 struct rt_cache_stat *st = v;
284 if (v == SEQ_START_TOKEN) {
285 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");
289 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
290 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
291 dst_entries_get_slow(&ipv4_dst_ops),
304 0, /* st->gc_total */
305 0, /* st->gc_ignored */
306 0, /* st->gc_goal_miss */
307 0, /* st->gc_dst_overflow */
308 0, /* st->in_hlist_search */
309 0 /* st->out_hlist_search */
314 static const struct seq_operations rt_cpu_seq_ops = {
315 .start = rt_cpu_seq_start,
316 .next = rt_cpu_seq_next,
317 .stop = rt_cpu_seq_stop,
318 .show = rt_cpu_seq_show,
322 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
324 return seq_open(file, &rt_cpu_seq_ops);
327 static const struct file_operations rt_cpu_seq_fops = {
328 .owner = THIS_MODULE,
329 .open = rt_cpu_seq_open,
332 .release = seq_release,
335 #ifdef CONFIG_IP_ROUTE_CLASSID
336 static int rt_acct_proc_show(struct seq_file *m, void *v)
338 struct ip_rt_acct *dst, *src;
341 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
345 for_each_possible_cpu(i) {
346 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
347 for (j = 0; j < 256; j++) {
348 dst[j].o_bytes += src[j].o_bytes;
349 dst[j].o_packets += src[j].o_packets;
350 dst[j].i_bytes += src[j].i_bytes;
351 dst[j].i_packets += src[j].i_packets;
355 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
360 static int rt_acct_proc_open(struct inode *inode, struct file *file)
362 return single_open(file, rt_acct_proc_show, NULL);
365 static const struct file_operations rt_acct_proc_fops = {
366 .owner = THIS_MODULE,
367 .open = rt_acct_proc_open,
370 .release = single_release,
374 static int __net_init ip_rt_do_proc_init(struct net *net)
376 struct proc_dir_entry *pde;
378 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
383 pde = proc_create("rt_cache", S_IRUGO,
384 net->proc_net_stat, &rt_cpu_seq_fops);
388 #ifdef CONFIG_IP_ROUTE_CLASSID
389 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
395 #ifdef CONFIG_IP_ROUTE_CLASSID
397 remove_proc_entry("rt_cache", net->proc_net_stat);
400 remove_proc_entry("rt_cache", net->proc_net);
405 static void __net_exit ip_rt_do_proc_exit(struct net *net)
407 remove_proc_entry("rt_cache", net->proc_net_stat);
408 remove_proc_entry("rt_cache", net->proc_net);
409 #ifdef CONFIG_IP_ROUTE_CLASSID
410 remove_proc_entry("rt_acct", net->proc_net);
414 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
415 .init = ip_rt_do_proc_init,
416 .exit = ip_rt_do_proc_exit,
419 static int __init ip_rt_proc_init(void)
421 return register_pernet_subsys(&ip_rt_proc_ops);
425 static inline int ip_rt_proc_init(void)
429 #endif /* CONFIG_PROC_FS */
431 static inline bool rt_is_expired(const struct rtable *rth)
433 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
436 void rt_cache_flush(struct net *net)
438 rt_genid_bump_ipv4(net);
441 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
445 struct net_device *dev = dst->dev;
446 const __be32 *pkey = daddr;
447 const struct rtable *rt;
450 rt = (const struct rtable *) dst;
452 pkey = (const __be32 *) &rt->rt_gateway;
454 pkey = &ip_hdr(skb)->daddr;
456 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
459 return neigh_create(&arp_tbl, pkey, dev);
462 #define IP_IDENTS_SZ 2048u
464 static atomic_t *ip_idents __read_mostly;
465 static u32 *ip_tstamps __read_mostly;
467 /* In order to protect privacy, we add a perturbation to identifiers
468 * if one generator is seldom used. This makes hard for an attacker
469 * to infer how many packets were sent between two points in time.
471 u32 ip_idents_reserve(u32 hash, int segs)
473 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
474 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
475 u32 old = ACCESS_ONCE(*p_tstamp);
476 u32 now = (u32)jiffies;
479 if (old != now && cmpxchg(p_tstamp, old, now) == old)
480 delta = prandom_u32_max(now - old);
482 return atomic_add_return(segs + delta, p_id) - segs;
484 EXPORT_SYMBOL(ip_idents_reserve);
486 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
488 static u32 ip_idents_hashrnd __read_mostly;
491 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
493 hash = jhash_3words((__force u32)iph->daddr,
494 (__force u32)iph->saddr,
495 iph->protocol ^ net_hash_mix(net),
497 id = ip_idents_reserve(hash, segs);
500 EXPORT_SYMBOL(__ip_select_ident);
502 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
503 const struct iphdr *iph,
505 u8 prot, u32 mark, int flow_flags)
508 const struct inet_sock *inet = inet_sk(sk);
510 oif = sk->sk_bound_dev_if;
512 tos = RT_CONN_FLAGS(sk);
513 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
515 flowi4_init_output(fl4, oif, mark, tos,
516 RT_SCOPE_UNIVERSE, prot,
518 iph->daddr, iph->saddr, 0, 0);
521 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
522 const struct sock *sk)
524 const struct iphdr *iph = ip_hdr(skb);
525 int oif = skb->dev->ifindex;
526 u8 tos = RT_TOS(iph->tos);
527 u8 prot = iph->protocol;
528 u32 mark = skb->mark;
530 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
533 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
535 const struct inet_sock *inet = inet_sk(sk);
536 const struct ip_options_rcu *inet_opt;
537 __be32 daddr = inet->inet_daddr;
540 inet_opt = rcu_dereference(inet->inet_opt);
541 if (inet_opt && inet_opt->opt.srr)
542 daddr = inet_opt->opt.faddr;
543 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
544 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
545 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
546 inet_sk_flowi_flags(sk),
547 daddr, inet->inet_saddr, 0, 0);
551 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
552 const struct sk_buff *skb)
555 build_skb_flow_key(fl4, skb, sk);
557 build_sk_flow_key(fl4, sk);
560 static inline void rt_free(struct rtable *rt)
562 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
565 static DEFINE_SPINLOCK(fnhe_lock);
567 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
571 rt = rcu_dereference(fnhe->fnhe_rth_input);
573 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
576 rt = rcu_dereference(fnhe->fnhe_rth_output);
578 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
583 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
585 struct fib_nh_exception *fnhe, *oldest;
587 oldest = rcu_dereference(hash->chain);
588 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
589 fnhe = rcu_dereference(fnhe->fnhe_next)) {
590 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
593 fnhe_flush_routes(oldest);
597 static inline u32 fnhe_hashfun(__be32 daddr)
599 static u32 fnhe_hashrnd __read_mostly;
602 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
603 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
604 return hash_32(hval, FNHE_HASH_SHIFT);
607 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
609 rt->rt_pmtu = fnhe->fnhe_pmtu;
610 rt->dst.expires = fnhe->fnhe_expires;
613 rt->rt_flags |= RTCF_REDIRECTED;
614 rt->rt_gateway = fnhe->fnhe_gw;
615 rt->rt_uses_gateway = 1;
619 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
620 u32 pmtu, unsigned long expires)
622 struct fnhe_hash_bucket *hash;
623 struct fib_nh_exception *fnhe;
627 u32 hval = fnhe_hashfun(daddr);
629 spin_lock_bh(&fnhe_lock);
631 hash = rcu_dereference(nh->nh_exceptions);
633 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
636 rcu_assign_pointer(nh->nh_exceptions, hash);
642 for (fnhe = rcu_dereference(hash->chain); fnhe;
643 fnhe = rcu_dereference(fnhe->fnhe_next)) {
644 if (fnhe->fnhe_daddr == daddr)
653 fnhe->fnhe_pmtu = pmtu;
654 fnhe->fnhe_expires = max(1UL, expires);
656 /* Update all cached dsts too */
657 rt = rcu_dereference(fnhe->fnhe_rth_input);
659 fill_route_from_fnhe(rt, fnhe);
660 rt = rcu_dereference(fnhe->fnhe_rth_output);
662 fill_route_from_fnhe(rt, fnhe);
664 if (depth > FNHE_RECLAIM_DEPTH)
665 fnhe = fnhe_oldest(hash);
667 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
671 fnhe->fnhe_next = hash->chain;
672 rcu_assign_pointer(hash->chain, fnhe);
674 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
675 fnhe->fnhe_daddr = daddr;
677 fnhe->fnhe_pmtu = pmtu;
678 fnhe->fnhe_expires = expires;
680 /* Exception created; mark the cached routes for the nexthop
681 * stale, so anyone caching it rechecks if this exception
684 rt = rcu_dereference(nh->nh_rth_input);
686 rt->dst.obsolete = DST_OBSOLETE_KILL;
688 for_each_possible_cpu(i) {
689 struct rtable __rcu **prt;
690 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
691 rt = rcu_dereference(*prt);
693 rt->dst.obsolete = DST_OBSOLETE_KILL;
697 fnhe->fnhe_stamp = jiffies;
700 spin_unlock_bh(&fnhe_lock);
703 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
706 __be32 new_gw = icmp_hdr(skb)->un.gateway;
707 __be32 old_gw = ip_hdr(skb)->saddr;
708 struct net_device *dev = skb->dev;
709 struct in_device *in_dev;
710 struct fib_result res;
714 switch (icmp_hdr(skb)->code & 7) {
716 case ICMP_REDIR_NETTOS:
717 case ICMP_REDIR_HOST:
718 case ICMP_REDIR_HOSTTOS:
725 if (rt->rt_gateway != old_gw)
728 in_dev = __in_dev_get_rcu(dev);
733 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
734 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
735 ipv4_is_zeronet(new_gw))
736 goto reject_redirect;
738 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
739 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
740 goto reject_redirect;
741 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
742 goto reject_redirect;
744 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
745 goto reject_redirect;
748 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
750 if (!(n->nud_state & NUD_VALID)) {
751 neigh_event_send(n, NULL);
753 if (fib_lookup(net, fl4, &res, 0) == 0) {
754 struct fib_nh *nh = &FIB_RES_NH(res);
756 update_or_create_fnhe(nh, fl4->daddr, new_gw,
760 rt->dst.obsolete = DST_OBSOLETE_KILL;
761 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
768 #ifdef CONFIG_IP_ROUTE_VERBOSE
769 if (IN_DEV_LOG_MARTIANS(in_dev)) {
770 const struct iphdr *iph = (const struct iphdr *) skb->data;
771 __be32 daddr = iph->daddr;
772 __be32 saddr = iph->saddr;
774 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
775 " Advised path = %pI4 -> %pI4\n",
776 &old_gw, dev->name, &new_gw,
783 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
787 const struct iphdr *iph = (const struct iphdr *) skb->data;
788 int oif = skb->dev->ifindex;
789 u8 tos = RT_TOS(iph->tos);
790 u8 prot = iph->protocol;
791 u32 mark = skb->mark;
793 rt = (struct rtable *) dst;
795 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
796 __ip_do_redirect(rt, skb, &fl4, true);
799 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
801 struct rtable *rt = (struct rtable *)dst;
802 struct dst_entry *ret = dst;
805 if (dst->obsolete > 0) {
808 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
819 * 1. The first ip_rt_redirect_number redirects are sent
820 * with exponential backoff, then we stop sending them at all,
821 * assuming that the host ignores our redirects.
822 * 2. If we did not see packets requiring redirects
823 * during ip_rt_redirect_silence, we assume that the host
824 * forgot redirected route and start to send redirects again.
826 * This algorithm is much cheaper and more intelligent than dumb load limiting
829 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
830 * and "frag. need" (breaks PMTU discovery) in icmp.c.
833 void ip_rt_send_redirect(struct sk_buff *skb)
835 struct rtable *rt = skb_rtable(skb);
836 struct in_device *in_dev;
837 struct inet_peer *peer;
842 in_dev = __in_dev_get_rcu(rt->dst.dev);
843 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
847 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
850 net = dev_net(rt->dst.dev);
851 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
853 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
854 rt_nexthop(rt, ip_hdr(skb)->daddr));
858 /* No redirected packets during ip_rt_redirect_silence;
859 * reset the algorithm.
861 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
862 peer->rate_tokens = 0;
864 /* Too many ignored redirects; do not send anything
865 * set dst.rate_last to the last seen redirected packet.
867 if (peer->rate_tokens >= ip_rt_redirect_number) {
868 peer->rate_last = jiffies;
872 /* Check for load limit; set rate_last to the latest sent
875 if (peer->rate_tokens == 0 ||
878 (ip_rt_redirect_load << peer->rate_tokens)))) {
879 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
881 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
882 peer->rate_last = jiffies;
884 #ifdef CONFIG_IP_ROUTE_VERBOSE
886 peer->rate_tokens == ip_rt_redirect_number)
887 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
888 &ip_hdr(skb)->saddr, inet_iif(skb),
889 &ip_hdr(skb)->daddr, &gw);
896 static int ip_error(struct sk_buff *skb)
898 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
899 struct rtable *rt = skb_rtable(skb);
900 struct inet_peer *peer;
906 /* IP on this device is disabled. */
910 net = dev_net(rt->dst.dev);
911 if (!IN_DEV_FORWARD(in_dev)) {
912 switch (rt->dst.error) {
914 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
918 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
924 switch (rt->dst.error) {
929 code = ICMP_HOST_UNREACH;
932 code = ICMP_NET_UNREACH;
933 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
936 code = ICMP_PKT_FILTERED;
940 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
945 peer->rate_tokens += now - peer->rate_last;
946 if (peer->rate_tokens > ip_rt_error_burst)
947 peer->rate_tokens = ip_rt_error_burst;
948 peer->rate_last = now;
949 if (peer->rate_tokens >= ip_rt_error_cost)
950 peer->rate_tokens -= ip_rt_error_cost;
956 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
962 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
964 struct dst_entry *dst = &rt->dst;
965 struct fib_result res;
967 if (dst_metric_locked(dst, RTAX_MTU))
970 if (ipv4_mtu(dst) < mtu)
973 if (mtu < ip_rt_min_pmtu)
974 mtu = ip_rt_min_pmtu;
976 if (rt->rt_pmtu == mtu &&
977 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
981 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
982 struct fib_nh *nh = &FIB_RES_NH(res);
984 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
985 jiffies + ip_rt_mtu_expires);
990 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
991 struct sk_buff *skb, u32 mtu)
993 struct rtable *rt = (struct rtable *) dst;
996 ip_rt_build_flow_key(&fl4, sk, skb);
997 __ip_rt_update_pmtu(rt, &fl4, mtu);
1000 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1001 int oif, u32 mark, u8 protocol, int flow_flags)
1003 const struct iphdr *iph = (const struct iphdr *) skb->data;
1008 mark = IP4_REPLY_MARK(net, skb->mark);
1010 __build_flow_key(&fl4, NULL, iph, oif,
1011 RT_TOS(iph->tos), protocol, mark, flow_flags);
1012 rt = __ip_route_output_key(net, &fl4);
1014 __ip_rt_update_pmtu(rt, &fl4, mtu);
1018 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1020 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1022 const struct iphdr *iph = (const struct iphdr *) skb->data;
1026 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1028 if (!fl4.flowi4_mark)
1029 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1031 rt = __ip_route_output_key(sock_net(sk), &fl4);
1033 __ip_rt_update_pmtu(rt, &fl4, mtu);
1038 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1040 const struct iphdr *iph = (const struct iphdr *) skb->data;
1043 struct dst_entry *odst = NULL;
1048 if (!ip_sk_accept_pmtu(sk))
1051 odst = sk_dst_get(sk);
1053 if (sock_owned_by_user(sk) || !odst) {
1054 __ipv4_sk_update_pmtu(skb, sk, mtu);
1058 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1060 rt = (struct rtable *)odst;
1061 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1062 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1069 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1071 if (!dst_check(&rt->dst, 0)) {
1073 dst_release(&rt->dst);
1075 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1083 sk_dst_set(sk, &rt->dst);
1089 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1091 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1092 int oif, u32 mark, u8 protocol, int flow_flags)
1094 const struct iphdr *iph = (const struct iphdr *) skb->data;
1098 __build_flow_key(&fl4, NULL, iph, oif,
1099 RT_TOS(iph->tos), protocol, mark, flow_flags);
1100 rt = __ip_route_output_key(net, &fl4);
1102 __ip_do_redirect(rt, skb, &fl4, false);
1106 EXPORT_SYMBOL_GPL(ipv4_redirect);
1108 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1110 const struct iphdr *iph = (const struct iphdr *) skb->data;
1114 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1115 rt = __ip_route_output_key(sock_net(sk), &fl4);
1117 __ip_do_redirect(rt, skb, &fl4, false);
1121 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1123 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1125 struct rtable *rt = (struct rtable *) dst;
1127 /* All IPV4 dsts are created with ->obsolete set to the value
1128 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1129 * into this function always.
1131 * When a PMTU/redirect information update invalidates a route,
1132 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1133 * DST_OBSOLETE_DEAD by dst_free().
1135 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1140 static void ipv4_link_failure(struct sk_buff *skb)
1144 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1146 rt = skb_rtable(skb);
1148 dst_set_expires(&rt->dst, 0);
1151 static int ip_rt_bug(struct sock *sk, struct sk_buff *skb)
1153 pr_debug("%s: %pI4 -> %pI4, %s\n",
1154 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1155 skb->dev ? skb->dev->name : "?");
1162 We do not cache source address of outgoing interface,
1163 because it is used only by IP RR, TS and SRR options,
1164 so that it out of fast path.
1166 BTW remember: "addr" is allowed to be not aligned
1170 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1174 if (rt_is_output_route(rt))
1175 src = ip_hdr(skb)->saddr;
1177 struct fib_result res;
1183 memset(&fl4, 0, sizeof(fl4));
1184 fl4.daddr = iph->daddr;
1185 fl4.saddr = iph->saddr;
1186 fl4.flowi4_tos = RT_TOS(iph->tos);
1187 fl4.flowi4_oif = rt->dst.dev->ifindex;
1188 fl4.flowi4_iif = skb->dev->ifindex;
1189 fl4.flowi4_mark = skb->mark;
1192 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1193 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1195 src = inet_select_addr(rt->dst.dev,
1196 rt_nexthop(rt, iph->daddr),
1200 memcpy(addr, &src, 4);
1203 #ifdef CONFIG_IP_ROUTE_CLASSID
1204 static void set_class_tag(struct rtable *rt, u32 tag)
1206 if (!(rt->dst.tclassid & 0xFFFF))
1207 rt->dst.tclassid |= tag & 0xFFFF;
1208 if (!(rt->dst.tclassid & 0xFFFF0000))
1209 rt->dst.tclassid |= tag & 0xFFFF0000;
1213 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1215 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1218 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1220 if (advmss > 65535 - 40)
1221 advmss = 65535 - 40;
1226 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1228 const struct rtable *rt = (const struct rtable *) dst;
1229 unsigned int mtu = rt->rt_pmtu;
1231 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1232 mtu = dst_metric_raw(dst, RTAX_MTU);
1237 mtu = dst->dev->mtu;
1239 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1240 if (rt->rt_uses_gateway && mtu > 576)
1244 return min_t(unsigned int, mtu, IP_MAX_MTU);
1247 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1249 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1250 struct fib_nh_exception *fnhe;
1256 hval = fnhe_hashfun(daddr);
1258 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1259 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1260 if (fnhe->fnhe_daddr == daddr)
1266 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1271 spin_lock_bh(&fnhe_lock);
1273 if (daddr == fnhe->fnhe_daddr) {
1274 struct rtable __rcu **porig;
1275 struct rtable *orig;
1276 int genid = fnhe_genid(dev_net(rt->dst.dev));
1278 if (rt_is_input_route(rt))
1279 porig = &fnhe->fnhe_rth_input;
1281 porig = &fnhe->fnhe_rth_output;
1282 orig = rcu_dereference(*porig);
1284 if (fnhe->fnhe_genid != genid) {
1285 fnhe->fnhe_genid = genid;
1287 fnhe->fnhe_pmtu = 0;
1288 fnhe->fnhe_expires = 0;
1289 fnhe_flush_routes(fnhe);
1292 fill_route_from_fnhe(rt, fnhe);
1293 if (!rt->rt_gateway)
1294 rt->rt_gateway = daddr;
1296 if (!(rt->dst.flags & DST_NOCACHE)) {
1297 rcu_assign_pointer(*porig, rt);
1303 fnhe->fnhe_stamp = jiffies;
1305 spin_unlock_bh(&fnhe_lock);
1310 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1312 struct rtable *orig, *prev, **p;
1315 if (rt_is_input_route(rt)) {
1316 p = (struct rtable **)&nh->nh_rth_input;
1318 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1322 prev = cmpxchg(p, orig, rt);
1332 struct uncached_list {
1334 struct list_head head;
1337 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1339 static void rt_add_uncached_list(struct rtable *rt)
1341 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1343 rt->rt_uncached_list = ul;
1345 spin_lock_bh(&ul->lock);
1346 list_add_tail(&rt->rt_uncached, &ul->head);
1347 spin_unlock_bh(&ul->lock);
1350 static void ipv4_dst_destroy(struct dst_entry *dst)
1352 struct rtable *rt = (struct rtable *) dst;
1354 if (!list_empty(&rt->rt_uncached)) {
1355 struct uncached_list *ul = rt->rt_uncached_list;
1357 spin_lock_bh(&ul->lock);
1358 list_del(&rt->rt_uncached);
1359 spin_unlock_bh(&ul->lock);
1361 lwtstate_put(rt->rt_lwtstate);
1364 void rt_flush_dev(struct net_device *dev)
1366 struct net *net = dev_net(dev);
1370 for_each_possible_cpu(cpu) {
1371 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1373 spin_lock_bh(&ul->lock);
1374 list_for_each_entry(rt, &ul->head, rt_uncached) {
1375 if (rt->dst.dev != dev)
1377 rt->dst.dev = net->loopback_dev;
1378 dev_hold(rt->dst.dev);
1381 spin_unlock_bh(&ul->lock);
1385 static bool rt_cache_valid(const struct rtable *rt)
1388 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1392 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1393 const struct fib_result *res,
1394 struct fib_nh_exception *fnhe,
1395 struct fib_info *fi, u16 type, u32 itag)
1397 bool cached = false;
1400 struct fib_nh *nh = &FIB_RES_NH(*res);
1402 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1403 rt->rt_gateway = nh->nh_gw;
1404 rt->rt_uses_gateway = 1;
1406 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1407 #ifdef CONFIG_IP_ROUTE_CLASSID
1408 rt->dst.tclassid = nh->nh_tclassid;
1410 rt->rt_lwtstate = lwtstate_get(nh->nh_lwtstate);
1412 cached = rt_bind_exception(rt, fnhe, daddr);
1413 else if (!(rt->dst.flags & DST_NOCACHE))
1414 cached = rt_cache_route(nh, rt);
1415 if (unlikely(!cached)) {
1416 /* Routes we intend to cache in nexthop exception or
1417 * FIB nexthop have the DST_NOCACHE bit clear.
1418 * However, if we are unsuccessful at storing this
1419 * route into the cache we really need to set it.
1421 rt->dst.flags |= DST_NOCACHE;
1422 if (!rt->rt_gateway)
1423 rt->rt_gateway = daddr;
1424 rt_add_uncached_list(rt);
1427 rt_add_uncached_list(rt);
1429 #ifdef CONFIG_IP_ROUTE_CLASSID
1430 #ifdef CONFIG_IP_MULTIPLE_TABLES
1431 set_class_tag(rt, res->tclassid);
1433 set_class_tag(rt, itag);
1437 static struct rtable *rt_dst_alloc(struct net_device *dev,
1438 bool nopolicy, bool noxfrm, bool will_cache)
1440 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1441 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1442 (nopolicy ? DST_NOPOLICY : 0) |
1443 (noxfrm ? DST_NOXFRM : 0));
1446 /* called in rcu_read_lock() section */
1447 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1448 u8 tos, struct net_device *dev, int our)
1451 struct in_device *in_dev = __in_dev_get_rcu(dev);
1455 /* Primary sanity checks. */
1460 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1461 skb->protocol != htons(ETH_P_IP))
1464 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1465 if (ipv4_is_loopback(saddr))
1468 if (ipv4_is_zeronet(saddr)) {
1469 if (!ipv4_is_local_multicast(daddr))
1472 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1477 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1478 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1482 #ifdef CONFIG_IP_ROUTE_CLASSID
1483 rth->dst.tclassid = itag;
1485 rth->dst.output = ip_rt_bug;
1487 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
1488 rth->rt_flags = RTCF_MULTICAST;
1489 rth->rt_type = RTN_MULTICAST;
1490 rth->rt_is_input= 1;
1493 rth->rt_gateway = 0;
1494 rth->rt_uses_gateway = 0;
1495 INIT_LIST_HEAD(&rth->rt_uncached);
1496 rth->rt_lwtstate = NULL;
1498 rth->dst.input= ip_local_deliver;
1499 rth->rt_flags |= RTCF_LOCAL;
1502 #ifdef CONFIG_IP_MROUTE
1503 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1504 rth->dst.input = ip_mr_input;
1506 RT_CACHE_STAT_INC(in_slow_mc);
1508 skb_dst_set(skb, &rth->dst);
1520 static void ip_handle_martian_source(struct net_device *dev,
1521 struct in_device *in_dev,
1522 struct sk_buff *skb,
1526 RT_CACHE_STAT_INC(in_martian_src);
1527 #ifdef CONFIG_IP_ROUTE_VERBOSE
1528 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1530 * RFC1812 recommendation, if source is martian,
1531 * the only hint is MAC header.
1533 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1534 &daddr, &saddr, dev->name);
1535 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1536 print_hex_dump(KERN_WARNING, "ll header: ",
1537 DUMP_PREFIX_OFFSET, 16, 1,
1538 skb_mac_header(skb),
1539 dev->hard_header_len, true);
1545 /* called in rcu_read_lock() section */
1546 static int __mkroute_input(struct sk_buff *skb,
1547 const struct fib_result *res,
1548 struct in_device *in_dev,
1549 __be32 daddr, __be32 saddr, u32 tos)
1551 struct fib_nh_exception *fnhe;
1554 struct in_device *out_dev;
1558 /* get a working reference to the output device */
1559 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1561 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1565 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1566 in_dev->dev, in_dev, &itag);
1568 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1574 do_cache = res->fi && !itag;
1575 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1576 skb->protocol == htons(ETH_P_IP) &&
1577 (IN_DEV_SHARED_MEDIA(out_dev) ||
1578 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1579 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1581 if (skb->protocol != htons(ETH_P_IP)) {
1582 /* Not IP (i.e. ARP). Do not create route, if it is
1583 * invalid for proxy arp. DNAT routes are always valid.
1585 * Proxy arp feature have been extended to allow, ARP
1586 * replies back to the same interface, to support
1587 * Private VLAN switch technologies. See arp.c.
1589 if (out_dev == in_dev &&
1590 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1596 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1599 rth = rcu_dereference(fnhe->fnhe_rth_input);
1601 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1603 if (rt_cache_valid(rth)) {
1604 skb_dst_set_noref(skb, &rth->dst);
1609 rth = rt_dst_alloc(out_dev->dev,
1610 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1611 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1617 rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1619 rth->rt_type = res->type;
1620 rth->rt_is_input = 1;
1623 rth->rt_gateway = 0;
1624 rth->rt_uses_gateway = 0;
1625 INIT_LIST_HEAD(&rth->rt_uncached);
1626 rth->rt_lwtstate = NULL;
1627 RT_CACHE_STAT_INC(in_slow_tot);
1629 rth->dst.input = ip_forward;
1630 rth->dst.output = ip_output;
1632 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1633 if (lwtunnel_output_redirect(rth->rt_lwtstate))
1634 rth->dst.output = lwtunnel_output;
1635 skb_dst_set(skb, &rth->dst);
1642 static int ip_mkroute_input(struct sk_buff *skb,
1643 struct fib_result *res,
1644 const struct flowi4 *fl4,
1645 struct in_device *in_dev,
1646 __be32 daddr, __be32 saddr, u32 tos)
1648 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1649 if (res->fi && res->fi->fib_nhs > 1)
1650 fib_select_multipath(res);
1653 /* create a routing cache entry */
1654 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1658 * NOTE. We drop all the packets that has local source
1659 * addresses, because every properly looped back packet
1660 * must have correct destination already attached by output routine.
1662 * Such approach solves two big problems:
1663 * 1. Not simplex devices are handled properly.
1664 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1665 * called with rcu_read_lock()
1668 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1669 u8 tos, struct net_device *dev)
1671 struct fib_result res;
1672 struct in_device *in_dev = __in_dev_get_rcu(dev);
1673 struct ip_tunnel_info *tun_info;
1675 unsigned int flags = 0;
1679 struct net *net = dev_net(dev);
1682 /* IP on this device is disabled. */
1687 /* Check for the most weird martians, which can be not detected
1691 tun_info = skb_tunnel_info(skb, AF_INET);
1692 if (tun_info && tun_info->mode == IP_TUNNEL_INFO_RX)
1693 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1695 fl4.flowi4_tun_key.tun_id = 0;
1698 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1699 goto martian_source;
1702 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1705 /* Accept zero addresses only to limited broadcast;
1706 * I even do not know to fix it or not. Waiting for complains :-)
1708 if (ipv4_is_zeronet(saddr))
1709 goto martian_source;
1711 if (ipv4_is_zeronet(daddr))
1712 goto martian_destination;
1714 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1715 * and call it once if daddr or/and saddr are loopback addresses
1717 if (ipv4_is_loopback(daddr)) {
1718 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1719 goto martian_destination;
1720 } else if (ipv4_is_loopback(saddr)) {
1721 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1722 goto martian_source;
1726 * Now we are ready to route packet.
1729 fl4.flowi4_iif = dev->ifindex;
1730 fl4.flowi4_mark = skb->mark;
1731 fl4.flowi4_tos = tos;
1732 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1735 err = fib_lookup(net, &fl4, &res, 0);
1737 if (!IN_DEV_FORWARD(in_dev))
1738 err = -EHOSTUNREACH;
1742 if (res.type == RTN_BROADCAST)
1745 if (res.type == RTN_LOCAL) {
1746 err = fib_validate_source(skb, saddr, daddr, tos,
1747 0, dev, in_dev, &itag);
1749 goto martian_source_keep_err;
1753 if (!IN_DEV_FORWARD(in_dev)) {
1754 err = -EHOSTUNREACH;
1757 if (res.type != RTN_UNICAST)
1758 goto martian_destination;
1760 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1764 if (skb->protocol != htons(ETH_P_IP))
1767 if (!ipv4_is_zeronet(saddr)) {
1768 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1771 goto martian_source_keep_err;
1773 flags |= RTCF_BROADCAST;
1774 res.type = RTN_BROADCAST;
1775 RT_CACHE_STAT_INC(in_brd);
1781 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1782 if (rt_cache_valid(rth)) {
1783 skb_dst_set_noref(skb, &rth->dst);
1791 rth = rt_dst_alloc(net->loopback_dev,
1792 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1796 rth->dst.input= ip_local_deliver;
1797 rth->dst.output= ip_rt_bug;
1798 #ifdef CONFIG_IP_ROUTE_CLASSID
1799 rth->dst.tclassid = itag;
1802 rth->rt_genid = rt_genid_ipv4(net);
1803 rth->rt_flags = flags|RTCF_LOCAL;
1804 rth->rt_type = res.type;
1805 rth->rt_is_input = 1;
1808 rth->rt_gateway = 0;
1809 rth->rt_uses_gateway = 0;
1810 INIT_LIST_HEAD(&rth->rt_uncached);
1811 rth->rt_lwtstate = NULL;
1813 RT_CACHE_STAT_INC(in_slow_tot);
1814 if (res.type == RTN_UNREACHABLE) {
1815 rth->dst.input= ip_error;
1816 rth->dst.error= -err;
1817 rth->rt_flags &= ~RTCF_LOCAL;
1820 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1821 rth->dst.flags |= DST_NOCACHE;
1822 rt_add_uncached_list(rth);
1825 skb_dst_set(skb, &rth->dst);
1830 RT_CACHE_STAT_INC(in_no_route);
1831 res.type = RTN_UNREACHABLE;
1836 * Do not cache martian addresses: they should be logged (RFC1812)
1838 martian_destination:
1839 RT_CACHE_STAT_INC(in_martian_dst);
1840 #ifdef CONFIG_IP_ROUTE_VERBOSE
1841 if (IN_DEV_LOG_MARTIANS(in_dev))
1842 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1843 &daddr, &saddr, dev->name);
1856 martian_source_keep_err:
1857 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1861 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1862 u8 tos, struct net_device *dev)
1868 /* Multicast recognition logic is moved from route cache to here.
1869 The problem was that too many Ethernet cards have broken/missing
1870 hardware multicast filters :-( As result the host on multicasting
1871 network acquires a lot of useless route cache entries, sort of
1872 SDR messages from all the world. Now we try to get rid of them.
1873 Really, provided software IP multicast filter is organized
1874 reasonably (at least, hashed), it does not result in a slowdown
1875 comparing with route cache reject entries.
1876 Note, that multicast routers are not affected, because
1877 route cache entry is created eventually.
1879 if (ipv4_is_multicast(daddr)) {
1880 struct in_device *in_dev = __in_dev_get_rcu(dev);
1883 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1884 ip_hdr(skb)->protocol);
1886 #ifdef CONFIG_IP_MROUTE
1888 (!ipv4_is_local_multicast(daddr) &&
1889 IN_DEV_MFORWARD(in_dev))
1892 int res = ip_route_input_mc(skb, daddr, saddr,
1901 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1905 EXPORT_SYMBOL(ip_route_input_noref);
1907 /* called with rcu_read_lock() */
1908 static struct rtable *__mkroute_output(const struct fib_result *res,
1909 const struct flowi4 *fl4, int orig_oif,
1910 struct net_device *dev_out,
1913 struct fib_info *fi = res->fi;
1914 struct fib_nh_exception *fnhe;
1915 struct in_device *in_dev;
1916 u16 type = res->type;
1920 in_dev = __in_dev_get_rcu(dev_out);
1922 return ERR_PTR(-EINVAL);
1924 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1925 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1926 return ERR_PTR(-EINVAL);
1928 if (ipv4_is_lbcast(fl4->daddr))
1929 type = RTN_BROADCAST;
1930 else if (ipv4_is_multicast(fl4->daddr))
1931 type = RTN_MULTICAST;
1932 else if (ipv4_is_zeronet(fl4->daddr))
1933 return ERR_PTR(-EINVAL);
1935 if (dev_out->flags & IFF_LOOPBACK)
1936 flags |= RTCF_LOCAL;
1939 if (type == RTN_BROADCAST) {
1940 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1942 } else if (type == RTN_MULTICAST) {
1943 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1944 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1946 flags &= ~RTCF_LOCAL;
1949 /* If multicast route do not exist use
1950 * default one, but do not gateway in this case.
1953 if (fi && res->prefixlen < 4)
1958 do_cache &= fi != NULL;
1960 struct rtable __rcu **prth;
1961 struct fib_nh *nh = &FIB_RES_NH(*res);
1963 fnhe = find_exception(nh, fl4->daddr);
1965 prth = &fnhe->fnhe_rth_output;
1967 if (unlikely(fl4->flowi4_flags &
1968 FLOWI_FLAG_KNOWN_NH &&
1970 nh->nh_scope == RT_SCOPE_LINK))) {
1974 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
1976 rth = rcu_dereference(*prth);
1977 if (rt_cache_valid(rth)) {
1978 dst_hold(&rth->dst);
1984 rth = rt_dst_alloc(dev_out,
1985 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1986 IN_DEV_CONF_GET(in_dev, NOXFRM),
1989 return ERR_PTR(-ENOBUFS);
1991 rth->dst.output = ip_output;
1993 rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1994 rth->rt_flags = flags;
1995 rth->rt_type = type;
1996 rth->rt_is_input = 0;
1997 rth->rt_iif = orig_oif ? : 0;
1999 rth->rt_gateway = 0;
2000 rth->rt_uses_gateway = 0;
2001 INIT_LIST_HEAD(&rth->rt_uncached);
2002 rth->rt_lwtstate = NULL;
2003 RT_CACHE_STAT_INC(out_slow_tot);
2005 if (flags & RTCF_LOCAL)
2006 rth->dst.input = ip_local_deliver;
2007 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2008 if (flags & RTCF_LOCAL &&
2009 !(dev_out->flags & IFF_LOOPBACK)) {
2010 rth->dst.output = ip_mc_output;
2011 RT_CACHE_STAT_INC(out_slow_mc);
2013 #ifdef CONFIG_IP_MROUTE
2014 if (type == RTN_MULTICAST) {
2015 if (IN_DEV_MFORWARD(in_dev) &&
2016 !ipv4_is_local_multicast(fl4->daddr)) {
2017 rth->dst.input = ip_mr_input;
2018 rth->dst.output = ip_mc_output;
2024 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2030 * Major route resolver routine.
2033 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2035 struct net_device *dev_out = NULL;
2036 __u8 tos = RT_FL_TOS(fl4);
2037 unsigned int flags = 0;
2038 struct fib_result res;
2046 orig_oif = fl4->flowi4_oif;
2048 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2049 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2050 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2051 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2055 rth = ERR_PTR(-EINVAL);
2056 if (ipv4_is_multicast(fl4->saddr) ||
2057 ipv4_is_lbcast(fl4->saddr) ||
2058 ipv4_is_zeronet(fl4->saddr))
2061 /* I removed check for oif == dev_out->oif here.
2062 It was wrong for two reasons:
2063 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2064 is assigned to multiple interfaces.
2065 2. Moreover, we are allowed to send packets with saddr
2066 of another iface. --ANK
2069 if (fl4->flowi4_oif == 0 &&
2070 (ipv4_is_multicast(fl4->daddr) ||
2071 ipv4_is_lbcast(fl4->daddr))) {
2072 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2073 dev_out = __ip_dev_find(net, fl4->saddr, false);
2077 /* Special hack: user can direct multicasts
2078 and limited broadcast via necessary interface
2079 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2080 This hack is not just for fun, it allows
2081 vic,vat and friends to work.
2082 They bind socket to loopback, set ttl to zero
2083 and expect that it will work.
2084 From the viewpoint of routing cache they are broken,
2085 because we are not allowed to build multicast path
2086 with loopback source addr (look, routing cache
2087 cannot know, that ttl is zero, so that packet
2088 will not leave this host and route is valid).
2089 Luckily, this hack is good workaround.
2092 fl4->flowi4_oif = dev_out->ifindex;
2096 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2097 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2098 if (!__ip_dev_find(net, fl4->saddr, false))
2104 if (fl4->flowi4_oif) {
2105 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2106 rth = ERR_PTR(-ENODEV);
2110 /* RACE: Check return value of inet_select_addr instead. */
2111 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2112 rth = ERR_PTR(-ENETUNREACH);
2115 if (ipv4_is_local_multicast(fl4->daddr) ||
2116 ipv4_is_lbcast(fl4->daddr) ||
2117 fl4->flowi4_proto == IPPROTO_IGMP) {
2119 fl4->saddr = inet_select_addr(dev_out, 0,
2124 if (ipv4_is_multicast(fl4->daddr))
2125 fl4->saddr = inet_select_addr(dev_out, 0,
2127 else if (!fl4->daddr)
2128 fl4->saddr = inet_select_addr(dev_out, 0,
2134 fl4->daddr = fl4->saddr;
2136 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2137 dev_out = net->loopback_dev;
2138 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2139 res.type = RTN_LOCAL;
2140 flags |= RTCF_LOCAL;
2144 if (fib_lookup(net, fl4, &res, 0)) {
2147 if (fl4->flowi4_oif) {
2148 /* Apparently, routing tables are wrong. Assume,
2149 that the destination is on link.
2152 Because we are allowed to send to iface
2153 even if it has NO routes and NO assigned
2154 addresses. When oif is specified, routing
2155 tables are looked up with only one purpose:
2156 to catch if destination is gatewayed, rather than
2157 direct. Moreover, if MSG_DONTROUTE is set,
2158 we send packet, ignoring both routing tables
2159 and ifaddr state. --ANK
2162 We could make it even if oif is unknown,
2163 likely IPv6, but we do not.
2166 if (fl4->saddr == 0)
2167 fl4->saddr = inet_select_addr(dev_out, 0,
2169 res.type = RTN_UNICAST;
2172 rth = ERR_PTR(-ENETUNREACH);
2176 if (res.type == RTN_LOCAL) {
2178 if (res.fi->fib_prefsrc)
2179 fl4->saddr = res.fi->fib_prefsrc;
2181 fl4->saddr = fl4->daddr;
2183 dev_out = net->loopback_dev;
2184 fl4->flowi4_oif = dev_out->ifindex;
2185 flags |= RTCF_LOCAL;
2189 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2190 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2191 fib_select_multipath(&res);
2194 if (!res.prefixlen &&
2195 res.table->tb_num_default > 1 &&
2196 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2197 fib_select_default(fl4, &res);
2200 fl4->saddr = FIB_RES_PREFSRC(net, res);
2202 dev_out = FIB_RES_DEV(res);
2203 fl4->flowi4_oif = dev_out->ifindex;
2207 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2213 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2215 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2220 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2222 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2224 return mtu ? : dst->dev->mtu;
2227 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2228 struct sk_buff *skb, u32 mtu)
2232 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2233 struct sk_buff *skb)
2237 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2243 static struct dst_ops ipv4_dst_blackhole_ops = {
2245 .check = ipv4_blackhole_dst_check,
2246 .mtu = ipv4_blackhole_mtu,
2247 .default_advmss = ipv4_default_advmss,
2248 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2249 .redirect = ipv4_rt_blackhole_redirect,
2250 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2251 .neigh_lookup = ipv4_neigh_lookup,
2254 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2256 struct rtable *ort = (struct rtable *) dst_orig;
2259 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2261 struct dst_entry *new = &rt->dst;
2264 new->input = dst_discard;
2265 new->output = dst_discard_sk;
2267 new->dev = ort->dst.dev;
2271 rt->rt_is_input = ort->rt_is_input;
2272 rt->rt_iif = ort->rt_iif;
2273 rt->rt_pmtu = ort->rt_pmtu;
2275 rt->rt_genid = rt_genid_ipv4(net);
2276 rt->rt_flags = ort->rt_flags;
2277 rt->rt_type = ort->rt_type;
2278 rt->rt_gateway = ort->rt_gateway;
2279 rt->rt_uses_gateway = ort->rt_uses_gateway;
2281 INIT_LIST_HEAD(&rt->rt_uncached);
2282 rt->rt_lwtstate = NULL;
2286 dst_release(dst_orig);
2288 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2291 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2294 struct rtable *rt = __ip_route_output_key(net, flp4);
2299 if (flp4->flowi4_proto)
2300 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2301 flowi4_to_flowi(flp4),
2306 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2308 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2309 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2310 u32 seq, int event, int nowait, unsigned int flags)
2312 struct rtable *rt = skb_rtable(skb);
2314 struct nlmsghdr *nlh;
2315 unsigned long expires = 0;
2317 u32 metrics[RTAX_MAX];
2319 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2323 r = nlmsg_data(nlh);
2324 r->rtm_family = AF_INET;
2325 r->rtm_dst_len = 32;
2327 r->rtm_tos = fl4->flowi4_tos;
2328 r->rtm_table = RT_TABLE_MAIN;
2329 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2330 goto nla_put_failure;
2331 r->rtm_type = rt->rt_type;
2332 r->rtm_scope = RT_SCOPE_UNIVERSE;
2333 r->rtm_protocol = RTPROT_UNSPEC;
2334 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2335 if (rt->rt_flags & RTCF_NOTIFY)
2336 r->rtm_flags |= RTM_F_NOTIFY;
2337 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2338 r->rtm_flags |= RTCF_DOREDIRECT;
2340 if (nla_put_in_addr(skb, RTA_DST, dst))
2341 goto nla_put_failure;
2343 r->rtm_src_len = 32;
2344 if (nla_put_in_addr(skb, RTA_SRC, src))
2345 goto nla_put_failure;
2348 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2349 goto nla_put_failure;
2350 #ifdef CONFIG_IP_ROUTE_CLASSID
2351 if (rt->dst.tclassid &&
2352 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2353 goto nla_put_failure;
2355 if (!rt_is_input_route(rt) &&
2356 fl4->saddr != src) {
2357 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2358 goto nla_put_failure;
2360 if (rt->rt_uses_gateway &&
2361 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2362 goto nla_put_failure;
2364 expires = rt->dst.expires;
2366 unsigned long now = jiffies;
2368 if (time_before(now, expires))
2374 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2375 if (rt->rt_pmtu && expires)
2376 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2377 if (rtnetlink_put_metrics(skb, metrics) < 0)
2378 goto nla_put_failure;
2380 if (fl4->flowi4_mark &&
2381 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2382 goto nla_put_failure;
2384 error = rt->dst.error;
2386 if (rt_is_input_route(rt)) {
2387 #ifdef CONFIG_IP_MROUTE
2388 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2389 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2390 int err = ipmr_get_route(net, skb,
2391 fl4->saddr, fl4->daddr,
2397 goto nla_put_failure;
2399 if (err == -EMSGSIZE)
2400 goto nla_put_failure;
2406 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2407 goto nla_put_failure;
2410 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2411 goto nla_put_failure;
2413 nlmsg_end(skb, nlh);
2417 nlmsg_cancel(skb, nlh);
2421 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2423 struct net *net = sock_net(in_skb->sk);
2425 struct nlattr *tb[RTA_MAX+1];
2426 struct rtable *rt = NULL;
2433 struct sk_buff *skb;
2435 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2439 rtm = nlmsg_data(nlh);
2441 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2447 /* Reserve room for dummy headers, this skb can pass
2448 through good chunk of routing engine.
2450 skb_reset_mac_header(skb);
2451 skb_reset_network_header(skb);
2453 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2454 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2455 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2457 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2458 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2459 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2460 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2462 memset(&fl4, 0, sizeof(fl4));
2465 fl4.flowi4_tos = rtm->rtm_tos;
2466 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2467 fl4.flowi4_mark = mark;
2470 struct net_device *dev;
2472 dev = __dev_get_by_index(net, iif);
2478 skb->protocol = htons(ETH_P_IP);
2482 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2485 rt = skb_rtable(skb);
2486 if (err == 0 && rt->dst.error)
2487 err = -rt->dst.error;
2489 rt = ip_route_output_key(net, &fl4);
2499 skb_dst_set(skb, &rt->dst);
2500 if (rtm->rtm_flags & RTM_F_NOTIFY)
2501 rt->rt_flags |= RTCF_NOTIFY;
2503 err = rt_fill_info(net, dst, src, &fl4, skb,
2504 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2505 RTM_NEWROUTE, 0, 0);
2509 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2518 void ip_rt_multicast_event(struct in_device *in_dev)
2520 rt_cache_flush(dev_net(in_dev->dev));
2523 #ifdef CONFIG_SYSCTL
2524 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2525 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2526 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2527 static int ip_rt_gc_elasticity __read_mostly = 8;
2529 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2530 void __user *buffer,
2531 size_t *lenp, loff_t *ppos)
2533 struct net *net = (struct net *)__ctl->extra1;
2536 rt_cache_flush(net);
2537 fnhe_genid_bump(net);
2544 static struct ctl_table ipv4_route_table[] = {
2546 .procname = "gc_thresh",
2547 .data = &ipv4_dst_ops.gc_thresh,
2548 .maxlen = sizeof(int),
2550 .proc_handler = proc_dointvec,
2553 .procname = "max_size",
2554 .data = &ip_rt_max_size,
2555 .maxlen = sizeof(int),
2557 .proc_handler = proc_dointvec,
2560 /* Deprecated. Use gc_min_interval_ms */
2562 .procname = "gc_min_interval",
2563 .data = &ip_rt_gc_min_interval,
2564 .maxlen = sizeof(int),
2566 .proc_handler = proc_dointvec_jiffies,
2569 .procname = "gc_min_interval_ms",
2570 .data = &ip_rt_gc_min_interval,
2571 .maxlen = sizeof(int),
2573 .proc_handler = proc_dointvec_ms_jiffies,
2576 .procname = "gc_timeout",
2577 .data = &ip_rt_gc_timeout,
2578 .maxlen = sizeof(int),
2580 .proc_handler = proc_dointvec_jiffies,
2583 .procname = "gc_interval",
2584 .data = &ip_rt_gc_interval,
2585 .maxlen = sizeof(int),
2587 .proc_handler = proc_dointvec_jiffies,
2590 .procname = "redirect_load",
2591 .data = &ip_rt_redirect_load,
2592 .maxlen = sizeof(int),
2594 .proc_handler = proc_dointvec,
2597 .procname = "redirect_number",
2598 .data = &ip_rt_redirect_number,
2599 .maxlen = sizeof(int),
2601 .proc_handler = proc_dointvec,
2604 .procname = "redirect_silence",
2605 .data = &ip_rt_redirect_silence,
2606 .maxlen = sizeof(int),
2608 .proc_handler = proc_dointvec,
2611 .procname = "error_cost",
2612 .data = &ip_rt_error_cost,
2613 .maxlen = sizeof(int),
2615 .proc_handler = proc_dointvec,
2618 .procname = "error_burst",
2619 .data = &ip_rt_error_burst,
2620 .maxlen = sizeof(int),
2622 .proc_handler = proc_dointvec,
2625 .procname = "gc_elasticity",
2626 .data = &ip_rt_gc_elasticity,
2627 .maxlen = sizeof(int),
2629 .proc_handler = proc_dointvec,
2632 .procname = "mtu_expires",
2633 .data = &ip_rt_mtu_expires,
2634 .maxlen = sizeof(int),
2636 .proc_handler = proc_dointvec_jiffies,
2639 .procname = "min_pmtu",
2640 .data = &ip_rt_min_pmtu,
2641 .maxlen = sizeof(int),
2643 .proc_handler = proc_dointvec,
2646 .procname = "min_adv_mss",
2647 .data = &ip_rt_min_advmss,
2648 .maxlen = sizeof(int),
2650 .proc_handler = proc_dointvec,
2655 static struct ctl_table ipv4_route_flush_table[] = {
2657 .procname = "flush",
2658 .maxlen = sizeof(int),
2660 .proc_handler = ipv4_sysctl_rtcache_flush,
2665 static __net_init int sysctl_route_net_init(struct net *net)
2667 struct ctl_table *tbl;
2669 tbl = ipv4_route_flush_table;
2670 if (!net_eq(net, &init_net)) {
2671 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2675 /* Don't export sysctls to unprivileged users */
2676 if (net->user_ns != &init_user_ns)
2677 tbl[0].procname = NULL;
2679 tbl[0].extra1 = net;
2681 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2682 if (!net->ipv4.route_hdr)
2687 if (tbl != ipv4_route_flush_table)
2693 static __net_exit void sysctl_route_net_exit(struct net *net)
2695 struct ctl_table *tbl;
2697 tbl = net->ipv4.route_hdr->ctl_table_arg;
2698 unregister_net_sysctl_table(net->ipv4.route_hdr);
2699 BUG_ON(tbl == ipv4_route_flush_table);
2703 static __net_initdata struct pernet_operations sysctl_route_ops = {
2704 .init = sysctl_route_net_init,
2705 .exit = sysctl_route_net_exit,
2709 static __net_init int rt_genid_init(struct net *net)
2711 atomic_set(&net->ipv4.rt_genid, 0);
2712 atomic_set(&net->fnhe_genid, 0);
2713 get_random_bytes(&net->ipv4.dev_addr_genid,
2714 sizeof(net->ipv4.dev_addr_genid));
2718 static __net_initdata struct pernet_operations rt_genid_ops = {
2719 .init = rt_genid_init,
2722 static int __net_init ipv4_inetpeer_init(struct net *net)
2724 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2728 inet_peer_base_init(bp);
2729 net->ipv4.peers = bp;
2733 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2735 struct inet_peer_base *bp = net->ipv4.peers;
2737 net->ipv4.peers = NULL;
2738 inetpeer_invalidate_tree(bp);
2742 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2743 .init = ipv4_inetpeer_init,
2744 .exit = ipv4_inetpeer_exit,
2747 #ifdef CONFIG_IP_ROUTE_CLASSID
2748 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2749 #endif /* CONFIG_IP_ROUTE_CLASSID */
2751 int __init ip_rt_init(void)
2756 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2758 panic("IP: failed to allocate ip_idents\n");
2760 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2762 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2764 panic("IP: failed to allocate ip_tstamps\n");
2766 for_each_possible_cpu(cpu) {
2767 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2769 INIT_LIST_HEAD(&ul->head);
2770 spin_lock_init(&ul->lock);
2772 #ifdef CONFIG_IP_ROUTE_CLASSID
2773 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2775 panic("IP: failed to allocate ip_rt_acct\n");
2778 ipv4_dst_ops.kmem_cachep =
2779 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2780 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2782 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2784 if (dst_entries_init(&ipv4_dst_ops) < 0)
2785 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2787 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2788 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2790 ipv4_dst_ops.gc_thresh = ~0;
2791 ip_rt_max_size = INT_MAX;
2796 if (ip_rt_proc_init())
2797 pr_err("Unable to create route proc files\n");
2802 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2804 #ifdef CONFIG_SYSCTL
2805 register_pernet_subsys(&sysctl_route_ops);
2807 register_pernet_subsys(&rt_genid_ops);
2808 register_pernet_subsys(&ipv4_inetpeer_ops);
2812 #ifdef CONFIG_SYSCTL
2814 * We really need to sanitize the damn ipv4 init order, then all
2815 * this nonsense will go away.
2817 void __init ip_static_sysctl_init(void)
2819 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);