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 <linux/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
94 #include <net/dst_metadata.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/lwtunnel.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
111 #include <linux/kmemleak.h>
113 #include <net/secure_seq.h>
114 #include <net/ip_tunnels.h>
115 #include <net/l3mdev.h>
117 #define RT_FL_TOS(oldflp4) \
118 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
120 #define RT_GC_TIMEOUT (300*HZ)
122 static int ip_rt_max_size;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
129 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly = 256;
132 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
134 * Interface to generic destination cache.
137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
138 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
139 static unsigned int ipv4_mtu(const struct dst_entry *dst);
140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
141 static void ipv4_link_failure(struct sk_buff *skb);
142 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
143 struct sk_buff *skb, u32 mtu);
144 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb);
146 static void ipv4_dst_destroy(struct dst_entry *dst);
148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
157 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
159 static struct dst_ops ipv4_dst_ops = {
161 .check = ipv4_dst_check,
162 .default_advmss = ipv4_default_advmss,
164 .cow_metrics = ipv4_cow_metrics,
165 .destroy = ipv4_dst_destroy,
166 .negative_advice = ipv4_negative_advice,
167 .link_failure = ipv4_link_failure,
168 .update_pmtu = ip_rt_update_pmtu,
169 .redirect = ip_do_redirect,
170 .local_out = __ip_local_out,
171 .neigh_lookup = ipv4_neigh_lookup,
172 .confirm_neigh = ipv4_confirm_neigh,
175 #define ECN_OR_COST(class) TC_PRIO_##class
177 const __u8 ip_tos2prio[16] = {
179 ECN_OR_COST(BESTEFFORT),
181 ECN_OR_COST(BESTEFFORT),
187 ECN_OR_COST(INTERACTIVE),
189 ECN_OR_COST(INTERACTIVE),
190 TC_PRIO_INTERACTIVE_BULK,
191 ECN_OR_COST(INTERACTIVE_BULK),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK)
195 EXPORT_SYMBOL(ip_tos2prio);
197 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
198 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
200 #ifdef CONFIG_PROC_FS
201 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
205 return SEQ_START_TOKEN;
208 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
214 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
218 static int rt_cache_seq_show(struct seq_file *seq, void *v)
220 if (v == SEQ_START_TOKEN)
221 seq_printf(seq, "%-127s\n",
222 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
223 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
228 static const struct seq_operations rt_cache_seq_ops = {
229 .start = rt_cache_seq_start,
230 .next = rt_cache_seq_next,
231 .stop = rt_cache_seq_stop,
232 .show = rt_cache_seq_show,
235 static int rt_cache_seq_open(struct inode *inode, struct file *file)
237 return seq_open(file, &rt_cache_seq_ops);
240 static const struct file_operations rt_cache_seq_fops = {
241 .owner = THIS_MODULE,
242 .open = rt_cache_seq_open,
245 .release = seq_release,
249 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
254 return SEQ_START_TOKEN;
256 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
257 if (!cpu_possible(cpu))
260 return &per_cpu(rt_cache_stat, cpu);
265 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
269 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
270 if (!cpu_possible(cpu))
273 return &per_cpu(rt_cache_stat, cpu);
279 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
284 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
286 struct rt_cache_stat *st = v;
288 if (v == SEQ_START_TOKEN) {
289 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");
293 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
294 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
295 dst_entries_get_slow(&ipv4_dst_ops),
308 0, /* st->gc_total */
309 0, /* st->gc_ignored */
310 0, /* st->gc_goal_miss */
311 0, /* st->gc_dst_overflow */
312 0, /* st->in_hlist_search */
313 0 /* st->out_hlist_search */
318 static const struct seq_operations rt_cpu_seq_ops = {
319 .start = rt_cpu_seq_start,
320 .next = rt_cpu_seq_next,
321 .stop = rt_cpu_seq_stop,
322 .show = rt_cpu_seq_show,
326 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
328 return seq_open(file, &rt_cpu_seq_ops);
331 static const struct file_operations rt_cpu_seq_fops = {
332 .owner = THIS_MODULE,
333 .open = rt_cpu_seq_open,
336 .release = seq_release,
339 #ifdef CONFIG_IP_ROUTE_CLASSID
340 static int rt_acct_proc_show(struct seq_file *m, void *v)
342 struct ip_rt_acct *dst, *src;
345 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
349 for_each_possible_cpu(i) {
350 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
351 for (j = 0; j < 256; j++) {
352 dst[j].o_bytes += src[j].o_bytes;
353 dst[j].o_packets += src[j].o_packets;
354 dst[j].i_bytes += src[j].i_bytes;
355 dst[j].i_packets += src[j].i_packets;
359 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
364 static int rt_acct_proc_open(struct inode *inode, struct file *file)
366 return single_open(file, rt_acct_proc_show, NULL);
369 static const struct file_operations rt_acct_proc_fops = {
370 .owner = THIS_MODULE,
371 .open = rt_acct_proc_open,
374 .release = single_release,
378 static int __net_init ip_rt_do_proc_init(struct net *net)
380 struct proc_dir_entry *pde;
382 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
387 pde = proc_create("rt_cache", S_IRUGO,
388 net->proc_net_stat, &rt_cpu_seq_fops);
392 #ifdef CONFIG_IP_ROUTE_CLASSID
393 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
399 #ifdef CONFIG_IP_ROUTE_CLASSID
401 remove_proc_entry("rt_cache", net->proc_net_stat);
404 remove_proc_entry("rt_cache", net->proc_net);
409 static void __net_exit ip_rt_do_proc_exit(struct net *net)
411 remove_proc_entry("rt_cache", net->proc_net_stat);
412 remove_proc_entry("rt_cache", net->proc_net);
413 #ifdef CONFIG_IP_ROUTE_CLASSID
414 remove_proc_entry("rt_acct", net->proc_net);
418 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
419 .init = ip_rt_do_proc_init,
420 .exit = ip_rt_do_proc_exit,
423 static int __init ip_rt_proc_init(void)
425 return register_pernet_subsys(&ip_rt_proc_ops);
429 static inline int ip_rt_proc_init(void)
433 #endif /* CONFIG_PROC_FS */
435 static inline bool rt_is_expired(const struct rtable *rth)
437 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
440 void rt_cache_flush(struct net *net)
442 rt_genid_bump_ipv4(net);
445 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
449 struct net_device *dev = dst->dev;
450 const __be32 *pkey = daddr;
451 const struct rtable *rt;
454 rt = (const struct rtable *) dst;
456 pkey = (const __be32 *) &rt->rt_gateway;
458 pkey = &ip_hdr(skb)->daddr;
460 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
463 return neigh_create(&arp_tbl, pkey, dev);
466 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
468 struct net_device *dev = dst->dev;
469 const __be32 *pkey = daddr;
470 const struct rtable *rt;
472 rt = (const struct rtable *)dst;
474 pkey = (const __be32 *)&rt->rt_gateway;
477 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
480 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
483 #define IP_IDENTS_SZ 2048u
485 static atomic_t *ip_idents __read_mostly;
486 static u32 *ip_tstamps __read_mostly;
488 /* In order to protect privacy, we add a perturbation to identifiers
489 * if one generator is seldom used. This makes hard for an attacker
490 * to infer how many packets were sent between two points in time.
492 u32 ip_idents_reserve(u32 hash, int segs)
494 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
495 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
496 u32 old = ACCESS_ONCE(*p_tstamp);
497 u32 now = (u32)jiffies;
500 if (old != now && cmpxchg(p_tstamp, old, now) == old)
501 delta = prandom_u32_max(now - old);
503 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
505 old = (u32)atomic_read(p_id);
506 new = old + delta + segs;
507 } while (atomic_cmpxchg(p_id, old, new) != old);
511 EXPORT_SYMBOL(ip_idents_reserve);
513 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
515 static u32 ip_idents_hashrnd __read_mostly;
518 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
520 hash = jhash_3words((__force u32)iph->daddr,
521 (__force u32)iph->saddr,
522 iph->protocol ^ net_hash_mix(net),
524 id = ip_idents_reserve(hash, segs);
527 EXPORT_SYMBOL(__ip_select_ident);
529 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
530 const struct sock *sk,
531 const struct iphdr *iph,
533 u8 prot, u32 mark, int flow_flags)
536 const struct inet_sock *inet = inet_sk(sk);
538 oif = sk->sk_bound_dev_if;
540 tos = RT_CONN_FLAGS(sk);
541 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
543 flowi4_init_output(fl4, oif, mark, tos,
544 RT_SCOPE_UNIVERSE, prot,
546 iph->daddr, iph->saddr, 0, 0,
547 sock_net_uid(net, sk));
550 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
551 const struct sock *sk)
553 const struct net *net = dev_net(skb->dev);
554 const struct iphdr *iph = ip_hdr(skb);
555 int oif = skb->dev->ifindex;
556 u8 tos = RT_TOS(iph->tos);
557 u8 prot = iph->protocol;
558 u32 mark = skb->mark;
560 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
563 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
565 const struct inet_sock *inet = inet_sk(sk);
566 const struct ip_options_rcu *inet_opt;
567 __be32 daddr = inet->inet_daddr;
570 inet_opt = rcu_dereference(inet->inet_opt);
571 if (inet_opt && inet_opt->opt.srr)
572 daddr = inet_opt->opt.faddr;
573 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
574 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
575 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
576 inet_sk_flowi_flags(sk),
577 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
581 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
582 const struct sk_buff *skb)
585 build_skb_flow_key(fl4, skb, sk);
587 build_sk_flow_key(fl4, sk);
590 static inline void rt_free(struct rtable *rt)
592 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
595 static DEFINE_SPINLOCK(fnhe_lock);
597 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
601 rt = rcu_dereference(fnhe->fnhe_rth_input);
603 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
606 rt = rcu_dereference(fnhe->fnhe_rth_output);
608 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
613 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
615 struct fib_nh_exception *fnhe, *oldest;
617 oldest = rcu_dereference(hash->chain);
618 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
619 fnhe = rcu_dereference(fnhe->fnhe_next)) {
620 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
623 fnhe_flush_routes(oldest);
627 static inline u32 fnhe_hashfun(__be32 daddr)
629 static u32 fnhe_hashrnd __read_mostly;
632 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
633 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
634 return hash_32(hval, FNHE_HASH_SHIFT);
637 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
639 rt->rt_pmtu = fnhe->fnhe_pmtu;
640 rt->dst.expires = fnhe->fnhe_expires;
643 rt->rt_flags |= RTCF_REDIRECTED;
644 rt->rt_gateway = fnhe->fnhe_gw;
645 rt->rt_uses_gateway = 1;
649 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
650 u32 pmtu, unsigned long expires)
652 struct fnhe_hash_bucket *hash;
653 struct fib_nh_exception *fnhe;
657 u32 hval = fnhe_hashfun(daddr);
659 spin_lock_bh(&fnhe_lock);
661 hash = rcu_dereference(nh->nh_exceptions);
663 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
666 rcu_assign_pointer(nh->nh_exceptions, hash);
672 for (fnhe = rcu_dereference(hash->chain); fnhe;
673 fnhe = rcu_dereference(fnhe->fnhe_next)) {
674 if (fnhe->fnhe_daddr == daddr)
683 fnhe->fnhe_pmtu = pmtu;
684 fnhe->fnhe_expires = max(1UL, expires);
686 /* Update all cached dsts too */
687 rt = rcu_dereference(fnhe->fnhe_rth_input);
689 fill_route_from_fnhe(rt, fnhe);
690 rt = rcu_dereference(fnhe->fnhe_rth_output);
692 fill_route_from_fnhe(rt, fnhe);
694 if (depth > FNHE_RECLAIM_DEPTH)
695 fnhe = fnhe_oldest(hash);
697 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
701 fnhe->fnhe_next = hash->chain;
702 rcu_assign_pointer(hash->chain, fnhe);
704 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
705 fnhe->fnhe_daddr = daddr;
707 fnhe->fnhe_pmtu = pmtu;
708 fnhe->fnhe_expires = expires;
710 /* Exception created; mark the cached routes for the nexthop
711 * stale, so anyone caching it rechecks if this exception
714 rt = rcu_dereference(nh->nh_rth_input);
716 rt->dst.obsolete = DST_OBSOLETE_KILL;
718 for_each_possible_cpu(i) {
719 struct rtable __rcu **prt;
720 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
721 rt = rcu_dereference(*prt);
723 rt->dst.obsolete = DST_OBSOLETE_KILL;
727 fnhe->fnhe_stamp = jiffies;
730 spin_unlock_bh(&fnhe_lock);
733 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
736 __be32 new_gw = icmp_hdr(skb)->un.gateway;
737 __be32 old_gw = ip_hdr(skb)->saddr;
738 struct net_device *dev = skb->dev;
739 struct in_device *in_dev;
740 struct fib_result res;
744 switch (icmp_hdr(skb)->code & 7) {
746 case ICMP_REDIR_NETTOS:
747 case ICMP_REDIR_HOST:
748 case ICMP_REDIR_HOSTTOS:
755 if (rt->rt_gateway != old_gw)
758 in_dev = __in_dev_get_rcu(dev);
763 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
764 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
765 ipv4_is_zeronet(new_gw))
766 goto reject_redirect;
768 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
769 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
770 goto reject_redirect;
771 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
772 goto reject_redirect;
774 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
775 goto reject_redirect;
778 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
780 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
782 if (!(n->nud_state & NUD_VALID)) {
783 neigh_event_send(n, NULL);
785 if (fib_lookup(net, fl4, &res, 0) == 0) {
786 struct fib_nh *nh = &FIB_RES_NH(res);
788 update_or_create_fnhe(nh, fl4->daddr, new_gw,
789 0, jiffies + ip_rt_gc_timeout);
792 rt->dst.obsolete = DST_OBSOLETE_KILL;
793 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
800 #ifdef CONFIG_IP_ROUTE_VERBOSE
801 if (IN_DEV_LOG_MARTIANS(in_dev)) {
802 const struct iphdr *iph = (const struct iphdr *) skb->data;
803 __be32 daddr = iph->daddr;
804 __be32 saddr = iph->saddr;
806 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
807 " Advised path = %pI4 -> %pI4\n",
808 &old_gw, dev->name, &new_gw,
815 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
819 const struct iphdr *iph = (const struct iphdr *) skb->data;
820 struct net *net = dev_net(skb->dev);
821 int oif = skb->dev->ifindex;
822 u8 tos = RT_TOS(iph->tos);
823 u8 prot = iph->protocol;
824 u32 mark = skb->mark;
826 rt = (struct rtable *) dst;
828 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
829 __ip_do_redirect(rt, skb, &fl4, true);
832 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
834 struct rtable *rt = (struct rtable *)dst;
835 struct dst_entry *ret = dst;
838 if (dst->obsolete > 0) {
841 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
852 * 1. The first ip_rt_redirect_number redirects are sent
853 * with exponential backoff, then we stop sending them at all,
854 * assuming that the host ignores our redirects.
855 * 2. If we did not see packets requiring redirects
856 * during ip_rt_redirect_silence, we assume that the host
857 * forgot redirected route and start to send redirects again.
859 * This algorithm is much cheaper and more intelligent than dumb load limiting
862 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
863 * and "frag. need" (breaks PMTU discovery) in icmp.c.
866 void ip_rt_send_redirect(struct sk_buff *skb)
868 struct rtable *rt = skb_rtable(skb);
869 struct in_device *in_dev;
870 struct inet_peer *peer;
876 in_dev = __in_dev_get_rcu(rt->dst.dev);
877 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
881 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
882 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
885 net = dev_net(rt->dst.dev);
886 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
888 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
889 rt_nexthop(rt, ip_hdr(skb)->daddr));
893 /* No redirected packets during ip_rt_redirect_silence;
894 * reset the algorithm.
896 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
897 peer->rate_tokens = 0;
899 /* Too many ignored redirects; do not send anything
900 * set dst.rate_last to the last seen redirected packet.
902 if (peer->rate_tokens >= ip_rt_redirect_number) {
903 peer->rate_last = jiffies;
907 /* Check for load limit; set rate_last to the latest sent
910 if (peer->rate_tokens == 0 ||
913 (ip_rt_redirect_load << peer->rate_tokens)))) {
914 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
916 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
917 peer->rate_last = jiffies;
919 #ifdef CONFIG_IP_ROUTE_VERBOSE
921 peer->rate_tokens == ip_rt_redirect_number)
922 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
923 &ip_hdr(skb)->saddr, inet_iif(skb),
924 &ip_hdr(skb)->daddr, &gw);
931 static int ip_error(struct sk_buff *skb)
933 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
934 struct rtable *rt = skb_rtable(skb);
935 struct inet_peer *peer;
941 /* IP on this device is disabled. */
945 net = dev_net(rt->dst.dev);
946 if (!IN_DEV_FORWARD(in_dev)) {
947 switch (rt->dst.error) {
949 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
953 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
959 switch (rt->dst.error) {
964 code = ICMP_HOST_UNREACH;
967 code = ICMP_NET_UNREACH;
968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
971 code = ICMP_PKT_FILTERED;
975 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
976 l3mdev_master_ifindex(skb->dev), 1);
981 peer->rate_tokens += now - peer->rate_last;
982 if (peer->rate_tokens > ip_rt_error_burst)
983 peer->rate_tokens = ip_rt_error_burst;
984 peer->rate_last = now;
985 if (peer->rate_tokens >= ip_rt_error_cost)
986 peer->rate_tokens -= ip_rt_error_cost;
992 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
998 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1000 struct dst_entry *dst = &rt->dst;
1001 struct fib_result res;
1003 if (dst_metric_locked(dst, RTAX_MTU))
1006 if (ipv4_mtu(dst) < mtu)
1009 if (mtu < ip_rt_min_pmtu)
1010 mtu = ip_rt_min_pmtu;
1012 if (rt->rt_pmtu == mtu &&
1013 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1017 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1018 struct fib_nh *nh = &FIB_RES_NH(res);
1020 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
1021 jiffies + ip_rt_mtu_expires);
1026 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1027 struct sk_buff *skb, u32 mtu)
1029 struct rtable *rt = (struct rtable *) dst;
1032 ip_rt_build_flow_key(&fl4, sk, skb);
1033 __ip_rt_update_pmtu(rt, &fl4, mtu);
1036 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1037 int oif, u32 mark, u8 protocol, int flow_flags)
1039 const struct iphdr *iph = (const struct iphdr *) skb->data;
1044 mark = IP4_REPLY_MARK(net, skb->mark);
1046 __build_flow_key(net, &fl4, NULL, iph, oif,
1047 RT_TOS(iph->tos), protocol, mark, flow_flags);
1048 rt = __ip_route_output_key(net, &fl4);
1050 __ip_rt_update_pmtu(rt, &fl4, mtu);
1054 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1056 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1058 const struct iphdr *iph = (const struct iphdr *) skb->data;
1062 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1064 if (!fl4.flowi4_mark)
1065 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1067 rt = __ip_route_output_key(sock_net(sk), &fl4);
1069 __ip_rt_update_pmtu(rt, &fl4, mtu);
1074 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1076 const struct iphdr *iph = (const struct iphdr *) skb->data;
1079 struct dst_entry *odst = NULL;
1081 struct net *net = sock_net(sk);
1085 if (!ip_sk_accept_pmtu(sk))
1088 odst = sk_dst_get(sk);
1090 if (sock_owned_by_user(sk) || !odst) {
1091 __ipv4_sk_update_pmtu(skb, sk, mtu);
1095 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1097 rt = (struct rtable *)odst;
1098 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1099 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1106 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1108 if (!dst_check(&rt->dst, 0)) {
1110 dst_release(&rt->dst);
1112 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1120 sk_dst_set(sk, &rt->dst);
1126 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1128 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1129 int oif, u32 mark, u8 protocol, int flow_flags)
1131 const struct iphdr *iph = (const struct iphdr *) skb->data;
1135 __build_flow_key(net, &fl4, NULL, iph, oif,
1136 RT_TOS(iph->tos), protocol, mark, flow_flags);
1137 rt = __ip_route_output_key(net, &fl4);
1139 __ip_do_redirect(rt, skb, &fl4, false);
1143 EXPORT_SYMBOL_GPL(ipv4_redirect);
1145 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1147 const struct iphdr *iph = (const struct iphdr *) skb->data;
1150 struct net *net = sock_net(sk);
1152 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1153 rt = __ip_route_output_key(net, &fl4);
1155 __ip_do_redirect(rt, skb, &fl4, false);
1159 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1161 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1163 struct rtable *rt = (struct rtable *) dst;
1165 /* All IPV4 dsts are created with ->obsolete set to the value
1166 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1167 * into this function always.
1169 * When a PMTU/redirect information update invalidates a route,
1170 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1171 * DST_OBSOLETE_DEAD by dst_free().
1173 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1178 static void ipv4_link_failure(struct sk_buff *skb)
1182 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1184 rt = skb_rtable(skb);
1186 dst_set_expires(&rt->dst, 0);
1189 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1191 pr_debug("%s: %pI4 -> %pI4, %s\n",
1192 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1193 skb->dev ? skb->dev->name : "?");
1200 We do not cache source address of outgoing interface,
1201 because it is used only by IP RR, TS and SRR options,
1202 so that it out of fast path.
1204 BTW remember: "addr" is allowed to be not aligned
1208 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1212 if (rt_is_output_route(rt))
1213 src = ip_hdr(skb)->saddr;
1215 struct fib_result res;
1221 memset(&fl4, 0, sizeof(fl4));
1222 fl4.daddr = iph->daddr;
1223 fl4.saddr = iph->saddr;
1224 fl4.flowi4_tos = RT_TOS(iph->tos);
1225 fl4.flowi4_oif = rt->dst.dev->ifindex;
1226 fl4.flowi4_iif = skb->dev->ifindex;
1227 fl4.flowi4_mark = skb->mark;
1230 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1231 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1233 src = inet_select_addr(rt->dst.dev,
1234 rt_nexthop(rt, iph->daddr),
1238 memcpy(addr, &src, 4);
1241 #ifdef CONFIG_IP_ROUTE_CLASSID
1242 static void set_class_tag(struct rtable *rt, u32 tag)
1244 if (!(rt->dst.tclassid & 0xFFFF))
1245 rt->dst.tclassid |= tag & 0xFFFF;
1246 if (!(rt->dst.tclassid & 0xFFFF0000))
1247 rt->dst.tclassid |= tag & 0xFFFF0000;
1251 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1253 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1254 unsigned int advmss = max_t(unsigned int, dst->dev->mtu - header_size,
1257 return min(advmss, IPV4_MAX_PMTU - header_size);
1260 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1262 const struct rtable *rt = (const struct rtable *) dst;
1263 unsigned int mtu = rt->rt_pmtu;
1265 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1266 mtu = dst_metric_raw(dst, RTAX_MTU);
1271 mtu = dst->dev->mtu;
1273 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1274 if (rt->rt_uses_gateway && mtu > 576)
1278 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1280 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1283 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1285 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1286 struct fib_nh_exception *fnhe;
1292 hval = fnhe_hashfun(daddr);
1294 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1295 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1296 if (fnhe->fnhe_daddr == daddr)
1302 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1307 spin_lock_bh(&fnhe_lock);
1309 if (daddr == fnhe->fnhe_daddr) {
1310 struct rtable __rcu **porig;
1311 struct rtable *orig;
1312 int genid = fnhe_genid(dev_net(rt->dst.dev));
1314 if (rt_is_input_route(rt))
1315 porig = &fnhe->fnhe_rth_input;
1317 porig = &fnhe->fnhe_rth_output;
1318 orig = rcu_dereference(*porig);
1320 if (fnhe->fnhe_genid != genid) {
1321 fnhe->fnhe_genid = genid;
1323 fnhe->fnhe_pmtu = 0;
1324 fnhe->fnhe_expires = 0;
1325 fnhe_flush_routes(fnhe);
1328 fill_route_from_fnhe(rt, fnhe);
1329 if (!rt->rt_gateway)
1330 rt->rt_gateway = daddr;
1332 if (!(rt->dst.flags & DST_NOCACHE)) {
1333 rcu_assign_pointer(*porig, rt);
1339 fnhe->fnhe_stamp = jiffies;
1341 spin_unlock_bh(&fnhe_lock);
1346 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1348 struct rtable *orig, *prev, **p;
1351 if (rt_is_input_route(rt)) {
1352 p = (struct rtable **)&nh->nh_rth_input;
1354 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1358 prev = cmpxchg(p, orig, rt);
1368 struct uncached_list {
1370 struct list_head head;
1373 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1375 static void rt_add_uncached_list(struct rtable *rt)
1377 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1379 rt->rt_uncached_list = ul;
1381 spin_lock_bh(&ul->lock);
1382 list_add_tail(&rt->rt_uncached, &ul->head);
1383 spin_unlock_bh(&ul->lock);
1386 static void ipv4_dst_destroy(struct dst_entry *dst)
1388 struct rtable *rt = (struct rtable *) dst;
1391 fib_info_put(rt->fi);
1395 if (!list_empty(&rt->rt_uncached)) {
1396 struct uncached_list *ul = rt->rt_uncached_list;
1398 spin_lock_bh(&ul->lock);
1399 list_del(&rt->rt_uncached);
1400 spin_unlock_bh(&ul->lock);
1404 void rt_flush_dev(struct net_device *dev)
1406 struct net *net = dev_net(dev);
1410 for_each_possible_cpu(cpu) {
1411 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1413 spin_lock_bh(&ul->lock);
1414 list_for_each_entry(rt, &ul->head, rt_uncached) {
1415 if (rt->dst.dev != dev)
1417 rt->dst.dev = net->loopback_dev;
1418 dev_hold(rt->dst.dev);
1421 spin_unlock_bh(&ul->lock);
1425 static bool rt_cache_valid(const struct rtable *rt)
1428 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1432 static void rt_init_metrics(struct rtable *rt, struct fib_info *fi)
1434 if (fi->fib_metrics != (u32 *)dst_default_metrics) {
1439 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1442 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1443 const struct fib_result *res,
1444 struct fib_nh_exception *fnhe,
1445 struct fib_info *fi, u16 type, u32 itag)
1447 bool cached = false;
1450 struct fib_nh *nh = &FIB_RES_NH(*res);
1452 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1453 rt->rt_gateway = nh->nh_gw;
1454 rt->rt_uses_gateway = 1;
1456 rt_init_metrics(rt, fi);
1457 #ifdef CONFIG_IP_ROUTE_CLASSID
1458 rt->dst.tclassid = nh->nh_tclassid;
1460 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1462 cached = rt_bind_exception(rt, fnhe, daddr);
1463 else if (!(rt->dst.flags & DST_NOCACHE))
1464 cached = rt_cache_route(nh, rt);
1465 if (unlikely(!cached)) {
1466 /* Routes we intend to cache in nexthop exception or
1467 * FIB nexthop have the DST_NOCACHE bit clear.
1468 * However, if we are unsuccessful at storing this
1469 * route into the cache we really need to set it.
1471 rt->dst.flags |= DST_NOCACHE;
1472 if (!rt->rt_gateway)
1473 rt->rt_gateway = daddr;
1474 rt_add_uncached_list(rt);
1477 rt_add_uncached_list(rt);
1479 #ifdef CONFIG_IP_ROUTE_CLASSID
1480 #ifdef CONFIG_IP_MULTIPLE_TABLES
1481 set_class_tag(rt, res->tclassid);
1483 set_class_tag(rt, itag);
1487 struct rtable *rt_dst_alloc(struct net_device *dev,
1488 unsigned int flags, u16 type,
1489 bool nopolicy, bool noxfrm, bool will_cache)
1493 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1494 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1495 (nopolicy ? DST_NOPOLICY : 0) |
1496 (noxfrm ? DST_NOXFRM : 0));
1499 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1500 rt->rt_flags = flags;
1502 rt->rt_is_input = 0;
1506 rt->rt_uses_gateway = 0;
1507 rt->rt_table_id = 0;
1509 INIT_LIST_HEAD(&rt->rt_uncached);
1511 rt->dst.output = ip_output;
1512 if (flags & RTCF_LOCAL)
1513 rt->dst.input = ip_local_deliver;
1518 EXPORT_SYMBOL(rt_dst_alloc);
1520 /* called in rcu_read_lock() section */
1521 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1522 u8 tos, struct net_device *dev, int our)
1525 struct in_device *in_dev = __in_dev_get_rcu(dev);
1526 unsigned int flags = RTCF_MULTICAST;
1530 /* Primary sanity checks. */
1535 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1536 skb->protocol != htons(ETH_P_IP))
1539 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1542 if (ipv4_is_zeronet(saddr)) {
1543 if (!ipv4_is_local_multicast(daddr))
1546 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1552 flags |= RTCF_LOCAL;
1554 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1555 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1559 #ifdef CONFIG_IP_ROUTE_CLASSID
1560 rth->dst.tclassid = itag;
1562 rth->dst.output = ip_rt_bug;
1563 rth->rt_is_input= 1;
1565 #ifdef CONFIG_IP_MROUTE
1566 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1567 rth->dst.input = ip_mr_input;
1569 RT_CACHE_STAT_INC(in_slow_mc);
1571 skb_dst_set(skb, &rth->dst);
1583 static void ip_handle_martian_source(struct net_device *dev,
1584 struct in_device *in_dev,
1585 struct sk_buff *skb,
1589 RT_CACHE_STAT_INC(in_martian_src);
1590 #ifdef CONFIG_IP_ROUTE_VERBOSE
1591 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1593 * RFC1812 recommendation, if source is martian,
1594 * the only hint is MAC header.
1596 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1597 &daddr, &saddr, dev->name);
1598 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1599 print_hex_dump(KERN_WARNING, "ll header: ",
1600 DUMP_PREFIX_OFFSET, 16, 1,
1601 skb_mac_header(skb),
1602 dev->hard_header_len, true);
1608 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1610 struct fnhe_hash_bucket *hash;
1611 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1612 u32 hval = fnhe_hashfun(daddr);
1614 spin_lock_bh(&fnhe_lock);
1616 hash = rcu_dereference_protected(nh->nh_exceptions,
1617 lockdep_is_held(&fnhe_lock));
1620 fnhe_p = &hash->chain;
1621 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1623 if (fnhe->fnhe_daddr == daddr) {
1624 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1625 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1626 fnhe_flush_routes(fnhe);
1627 kfree_rcu(fnhe, rcu);
1630 fnhe_p = &fnhe->fnhe_next;
1631 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1632 lockdep_is_held(&fnhe_lock));
1635 spin_unlock_bh(&fnhe_lock);
1638 static void set_lwt_redirect(struct rtable *rth)
1640 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1641 rth->dst.lwtstate->orig_output = rth->dst.output;
1642 rth->dst.output = lwtunnel_output;
1645 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1646 rth->dst.lwtstate->orig_input = rth->dst.input;
1647 rth->dst.input = lwtunnel_input;
1651 /* called in rcu_read_lock() section */
1652 static int __mkroute_input(struct sk_buff *skb,
1653 const struct fib_result *res,
1654 struct in_device *in_dev,
1655 __be32 daddr, __be32 saddr, u32 tos)
1657 struct fib_nh_exception *fnhe;
1660 struct in_device *out_dev;
1664 /* get a working reference to the output device */
1665 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1667 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1671 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1672 in_dev->dev, in_dev, &itag);
1674 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1680 do_cache = res->fi && !itag;
1681 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1682 skb->protocol == htons(ETH_P_IP) &&
1683 (IN_DEV_SHARED_MEDIA(out_dev) ||
1684 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1685 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1687 if (skb->protocol != htons(ETH_P_IP)) {
1688 /* Not IP (i.e. ARP). Do not create route, if it is
1689 * invalid for proxy arp. DNAT routes are always valid.
1691 * Proxy arp feature have been extended to allow, ARP
1692 * replies back to the same interface, to support
1693 * Private VLAN switch technologies. See arp.c.
1695 if (out_dev == in_dev &&
1696 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1702 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1705 rth = rcu_dereference(fnhe->fnhe_rth_input);
1706 if (rth && rth->dst.expires &&
1707 time_after(jiffies, rth->dst.expires)) {
1708 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1715 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1718 if (rt_cache_valid(rth)) {
1719 skb_dst_set_noref(skb, &rth->dst);
1724 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1725 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1726 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1732 rth->rt_is_input = 1;
1734 rth->rt_table_id = res->table->tb_id;
1735 RT_CACHE_STAT_INC(in_slow_tot);
1737 rth->dst.input = ip_forward;
1739 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1740 set_lwt_redirect(rth);
1741 skb_dst_set(skb, &rth->dst);
1748 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1749 /* To make ICMP packets follow the right flow, the multipath hash is
1750 * calculated from the inner IP addresses.
1752 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1753 struct flow_keys *hash_keys)
1755 const struct iphdr *outer_iph = ip_hdr(skb);
1756 const struct iphdr *inner_iph;
1757 const struct icmphdr *icmph;
1758 struct iphdr _inner_iph;
1759 struct icmphdr _icmph;
1761 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1762 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1763 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1766 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1769 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1774 if (icmph->type != ICMP_DEST_UNREACH &&
1775 icmph->type != ICMP_REDIRECT &&
1776 icmph->type != ICMP_TIME_EXCEEDED &&
1777 icmph->type != ICMP_PARAMETERPROB)
1780 inner_iph = skb_header_pointer(skb,
1781 outer_iph->ihl * 4 + sizeof(_icmph),
1782 sizeof(_inner_iph), &_inner_iph);
1785 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1786 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1789 /* if skb is set it will be used and fl4 can be NULL */
1790 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1791 const struct sk_buff *skb)
1793 struct net *net = fi->fib_net;
1794 struct flow_keys hash_keys;
1797 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1799 memset(&hash_keys, 0, sizeof(hash_keys));
1800 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1802 ip_multipath_l3_keys(skb, &hash_keys);
1804 hash_keys.addrs.v4addrs.src = fl4->saddr;
1805 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1809 /* skb is currently provided only when forwarding */
1811 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1812 struct flow_keys keys;
1814 /* short-circuit if we already have L4 hash present */
1816 return skb_get_hash_raw(skb) >> 1;
1817 memset(&hash_keys, 0, sizeof(hash_keys));
1818 skb_flow_dissect_flow_keys(skb, &keys, flag);
1819 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1820 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1821 hash_keys.ports.src = keys.ports.src;
1822 hash_keys.ports.dst = keys.ports.dst;
1823 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1825 memset(&hash_keys, 0, sizeof(hash_keys));
1826 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1827 hash_keys.addrs.v4addrs.src = fl4->saddr;
1828 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1829 hash_keys.ports.src = fl4->fl4_sport;
1830 hash_keys.ports.dst = fl4->fl4_dport;
1831 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1835 mhash = flow_hash_from_keys(&hash_keys);
1839 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1840 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1842 static int ip_mkroute_input(struct sk_buff *skb,
1843 struct fib_result *res,
1844 struct in_device *in_dev,
1845 __be32 daddr, __be32 saddr, u32 tos)
1847 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1848 if (res->fi && res->fi->fib_nhs > 1) {
1849 int h = fib_multipath_hash(res->fi, NULL, skb);
1851 fib_select_multipath(res, h);
1855 /* create a routing cache entry */
1856 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1860 * NOTE. We drop all the packets that has local source
1861 * addresses, because every properly looped back packet
1862 * must have correct destination already attached by output routine.
1864 * Such approach solves two big problems:
1865 * 1. Not simplex devices are handled properly.
1866 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1867 * called with rcu_read_lock()
1870 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1871 u8 tos, struct net_device *dev)
1873 struct fib_result res;
1874 struct in_device *in_dev = __in_dev_get_rcu(dev);
1875 struct ip_tunnel_info *tun_info;
1877 unsigned int flags = 0;
1881 struct net *net = dev_net(dev);
1884 /* IP on this device is disabled. */
1889 /* Check for the most weird martians, which can be not detected
1893 tun_info = skb_tunnel_info(skb);
1894 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1895 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1897 fl4.flowi4_tun_key.tun_id = 0;
1900 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1901 goto martian_source;
1905 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1908 /* Accept zero addresses only to limited broadcast;
1909 * I even do not know to fix it or not. Waiting for complains :-)
1911 if (ipv4_is_zeronet(saddr))
1912 goto martian_source;
1914 if (ipv4_is_zeronet(daddr))
1915 goto martian_destination;
1917 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1918 * and call it once if daddr or/and saddr are loopback addresses
1920 if (ipv4_is_loopback(daddr)) {
1921 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1922 goto martian_destination;
1923 } else if (ipv4_is_loopback(saddr)) {
1924 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1925 goto martian_source;
1929 * Now we are ready to route packet.
1932 fl4.flowi4_iif = dev->ifindex;
1933 fl4.flowi4_mark = skb->mark;
1934 fl4.flowi4_tos = tos;
1935 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1936 fl4.flowi4_flags = 0;
1939 fl4.flowi4_uid = sock_net_uid(net, NULL);
1940 err = fib_lookup(net, &fl4, &res, 0);
1942 if (!IN_DEV_FORWARD(in_dev))
1943 err = -EHOSTUNREACH;
1947 if (res.type == RTN_BROADCAST)
1950 if (res.type == RTN_LOCAL) {
1951 err = fib_validate_source(skb, saddr, daddr, tos,
1952 0, dev, in_dev, &itag);
1954 goto martian_source;
1958 if (!IN_DEV_FORWARD(in_dev)) {
1959 err = -EHOSTUNREACH;
1962 if (res.type != RTN_UNICAST)
1963 goto martian_destination;
1965 err = ip_mkroute_input(skb, &res, in_dev, daddr, saddr, tos);
1969 if (skb->protocol != htons(ETH_P_IP))
1972 if (!ipv4_is_zeronet(saddr)) {
1973 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1976 goto martian_source;
1978 flags |= RTCF_BROADCAST;
1979 res.type = RTN_BROADCAST;
1980 RT_CACHE_STAT_INC(in_brd);
1986 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1987 if (rt_cache_valid(rth)) {
1988 skb_dst_set_noref(skb, &rth->dst);
1996 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
1997 flags | RTCF_LOCAL, res.type,
1998 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2002 rth->dst.output= ip_rt_bug;
2003 #ifdef CONFIG_IP_ROUTE_CLASSID
2004 rth->dst.tclassid = itag;
2006 rth->rt_is_input = 1;
2008 rth->rt_table_id = res.table->tb_id;
2010 RT_CACHE_STAT_INC(in_slow_tot);
2011 if (res.type == RTN_UNREACHABLE) {
2012 rth->dst.input= ip_error;
2013 rth->dst.error= -err;
2014 rth->rt_flags &= ~RTCF_LOCAL;
2018 struct fib_nh *nh = &FIB_RES_NH(res);
2020 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2021 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2022 WARN_ON(rth->dst.input == lwtunnel_input);
2023 rth->dst.lwtstate->orig_input = rth->dst.input;
2024 rth->dst.input = lwtunnel_input;
2027 if (unlikely(!rt_cache_route(nh, rth))) {
2028 rth->dst.flags |= DST_NOCACHE;
2029 rt_add_uncached_list(rth);
2032 skb_dst_set(skb, &rth->dst);
2037 RT_CACHE_STAT_INC(in_no_route);
2038 res.type = RTN_UNREACHABLE;
2044 * Do not cache martian addresses: they should be logged (RFC1812)
2046 martian_destination:
2047 RT_CACHE_STAT_INC(in_martian_dst);
2048 #ifdef CONFIG_IP_ROUTE_VERBOSE
2049 if (IN_DEV_LOG_MARTIANS(in_dev))
2050 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2051 &daddr, &saddr, dev->name);
2063 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2067 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2068 u8 tos, struct net_device *dev)
2072 tos &= IPTOS_RT_MASK;
2075 /* Multicast recognition logic is moved from route cache to here.
2076 The problem was that too many Ethernet cards have broken/missing
2077 hardware multicast filters :-( As result the host on multicasting
2078 network acquires a lot of useless route cache entries, sort of
2079 SDR messages from all the world. Now we try to get rid of them.
2080 Really, provided software IP multicast filter is organized
2081 reasonably (at least, hashed), it does not result in a slowdown
2082 comparing with route cache reject entries.
2083 Note, that multicast routers are not affected, because
2084 route cache entry is created eventually.
2086 if (ipv4_is_multicast(daddr)) {
2087 struct in_device *in_dev = __in_dev_get_rcu(dev);
2091 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2092 ip_hdr(skb)->protocol);
2094 /* check l3 master if no match yet */
2095 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2096 struct in_device *l3_in_dev;
2098 l3_in_dev = __in_dev_get_rcu(skb->dev);
2100 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2101 ip_hdr(skb)->protocol);
2106 #ifdef CONFIG_IP_MROUTE
2108 (!ipv4_is_local_multicast(daddr) &&
2109 IN_DEV_MFORWARD(in_dev))
2112 res = ip_route_input_mc(skb, daddr, saddr,
2118 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2122 EXPORT_SYMBOL(ip_route_input_noref);
2124 /* called with rcu_read_lock() */
2125 static struct rtable *__mkroute_output(const struct fib_result *res,
2126 const struct flowi4 *fl4, int orig_oif,
2127 struct net_device *dev_out,
2130 struct fib_info *fi = res->fi;
2131 struct fib_nh_exception *fnhe;
2132 struct in_device *in_dev;
2133 u16 type = res->type;
2137 in_dev = __in_dev_get_rcu(dev_out);
2139 return ERR_PTR(-EINVAL);
2141 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2142 if (ipv4_is_loopback(fl4->saddr) &&
2143 !(dev_out->flags & IFF_LOOPBACK) &&
2144 !netif_is_l3_master(dev_out))
2145 return ERR_PTR(-EINVAL);
2147 if (ipv4_is_lbcast(fl4->daddr))
2148 type = RTN_BROADCAST;
2149 else if (ipv4_is_multicast(fl4->daddr))
2150 type = RTN_MULTICAST;
2151 else if (ipv4_is_zeronet(fl4->daddr))
2152 return ERR_PTR(-EINVAL);
2154 if (dev_out->flags & IFF_LOOPBACK)
2155 flags |= RTCF_LOCAL;
2158 if (type == RTN_BROADCAST) {
2159 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2161 } else if (type == RTN_MULTICAST) {
2162 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2163 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2165 flags &= ~RTCF_LOCAL;
2168 /* If multicast route do not exist use
2169 * default one, but do not gateway in this case.
2172 if (fi && res->prefixlen < 4)
2174 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2175 (orig_oif != dev_out->ifindex)) {
2176 /* For local routes that require a particular output interface
2177 * we do not want to cache the result. Caching the result
2178 * causes incorrect behaviour when there are multiple source
2179 * addresses on the interface, the end result being that if the
2180 * intended recipient is waiting on that interface for the
2181 * packet he won't receive it because it will be delivered on
2182 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2183 * be set to the loopback interface as well.
2189 do_cache &= fi != NULL;
2191 struct rtable __rcu **prth;
2192 struct fib_nh *nh = &FIB_RES_NH(*res);
2194 fnhe = find_exception(nh, fl4->daddr);
2196 prth = &fnhe->fnhe_rth_output;
2197 rth = rcu_dereference(*prth);
2198 if (rth && rth->dst.expires &&
2199 time_after(jiffies, rth->dst.expires)) {
2200 ip_del_fnhe(nh, fl4->daddr);
2207 if (unlikely(fl4->flowi4_flags &
2208 FLOWI_FLAG_KNOWN_NH &&
2210 nh->nh_scope == RT_SCOPE_LINK))) {
2214 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2215 rth = rcu_dereference(*prth);
2218 if (rt_cache_valid(rth)) {
2219 dst_hold(&rth->dst);
2225 rth = rt_dst_alloc(dev_out, flags, type,
2226 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2227 IN_DEV_CONF_GET(in_dev, NOXFRM),
2230 return ERR_PTR(-ENOBUFS);
2232 rth->rt_iif = orig_oif ? : 0;
2234 rth->rt_table_id = res->table->tb_id;
2236 RT_CACHE_STAT_INC(out_slow_tot);
2238 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2239 if (flags & RTCF_LOCAL &&
2240 !(dev_out->flags & IFF_LOOPBACK)) {
2241 rth->dst.output = ip_mc_output;
2242 RT_CACHE_STAT_INC(out_slow_mc);
2244 #ifdef CONFIG_IP_MROUTE
2245 if (type == RTN_MULTICAST) {
2246 if (IN_DEV_MFORWARD(in_dev) &&
2247 !ipv4_is_local_multicast(fl4->daddr)) {
2248 rth->dst.input = ip_mr_input;
2249 rth->dst.output = ip_mc_output;
2255 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2256 set_lwt_redirect(rth);
2262 * Major route resolver routine.
2265 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2266 const struct sk_buff *skb)
2268 struct net_device *dev_out = NULL;
2269 __u8 tos = RT_FL_TOS(fl4);
2270 unsigned int flags = 0;
2271 struct fib_result res;
2274 int err = -ENETUNREACH;
2280 orig_oif = fl4->flowi4_oif;
2282 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2283 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2284 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2285 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2289 rth = ERR_PTR(-EINVAL);
2290 if (ipv4_is_multicast(fl4->saddr) ||
2291 ipv4_is_lbcast(fl4->saddr) ||
2292 ipv4_is_zeronet(fl4->saddr))
2295 /* I removed check for oif == dev_out->oif here.
2296 It was wrong for two reasons:
2297 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2298 is assigned to multiple interfaces.
2299 2. Moreover, we are allowed to send packets with saddr
2300 of another iface. --ANK
2303 if (fl4->flowi4_oif == 0 &&
2304 (ipv4_is_multicast(fl4->daddr) ||
2305 ipv4_is_lbcast(fl4->daddr))) {
2306 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2307 dev_out = __ip_dev_find(net, fl4->saddr, false);
2311 /* Special hack: user can direct multicasts
2312 and limited broadcast via necessary interface
2313 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2314 This hack is not just for fun, it allows
2315 vic,vat and friends to work.
2316 They bind socket to loopback, set ttl to zero
2317 and expect that it will work.
2318 From the viewpoint of routing cache they are broken,
2319 because we are not allowed to build multicast path
2320 with loopback source addr (look, routing cache
2321 cannot know, that ttl is zero, so that packet
2322 will not leave this host and route is valid).
2323 Luckily, this hack is good workaround.
2326 fl4->flowi4_oif = dev_out->ifindex;
2330 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2331 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2332 if (!__ip_dev_find(net, fl4->saddr, false))
2338 if (fl4->flowi4_oif) {
2339 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2340 rth = ERR_PTR(-ENODEV);
2344 /* RACE: Check return value of inet_select_addr instead. */
2345 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2346 rth = ERR_PTR(-ENETUNREACH);
2349 if (ipv4_is_local_multicast(fl4->daddr) ||
2350 ipv4_is_lbcast(fl4->daddr) ||
2351 fl4->flowi4_proto == IPPROTO_IGMP) {
2353 fl4->saddr = inet_select_addr(dev_out, 0,
2358 if (ipv4_is_multicast(fl4->daddr))
2359 fl4->saddr = inet_select_addr(dev_out, 0,
2361 else if (!fl4->daddr)
2362 fl4->saddr = inet_select_addr(dev_out, 0,
2368 fl4->daddr = fl4->saddr;
2370 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2371 dev_out = net->loopback_dev;
2372 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2373 res.type = RTN_LOCAL;
2374 flags |= RTCF_LOCAL;
2378 err = fib_lookup(net, fl4, &res, 0);
2382 if (fl4->flowi4_oif &&
2383 (ipv4_is_multicast(fl4->daddr) ||
2384 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2385 /* Apparently, routing tables are wrong. Assume,
2386 that the destination is on link.
2389 Because we are allowed to send to iface
2390 even if it has NO routes and NO assigned
2391 addresses. When oif is specified, routing
2392 tables are looked up with only one purpose:
2393 to catch if destination is gatewayed, rather than
2394 direct. Moreover, if MSG_DONTROUTE is set,
2395 we send packet, ignoring both routing tables
2396 and ifaddr state. --ANK
2399 We could make it even if oif is unknown,
2400 likely IPv6, but we do not.
2403 if (fl4->saddr == 0)
2404 fl4->saddr = inet_select_addr(dev_out, 0,
2406 res.type = RTN_UNICAST;
2413 if (res.type == RTN_LOCAL) {
2415 if (res.fi->fib_prefsrc)
2416 fl4->saddr = res.fi->fib_prefsrc;
2418 fl4->saddr = fl4->daddr;
2421 /* L3 master device is the loopback for that domain */
2422 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(res)) ? :
2424 fl4->flowi4_oif = dev_out->ifindex;
2425 flags |= RTCF_LOCAL;
2429 fib_select_path(net, &res, fl4, skb);
2431 dev_out = FIB_RES_DEV(res);
2432 fl4->flowi4_oif = dev_out->ifindex;
2436 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2442 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2444 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2449 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2451 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2453 return mtu ? : dst->dev->mtu;
2456 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2457 struct sk_buff *skb, u32 mtu)
2461 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2462 struct sk_buff *skb)
2466 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2472 static struct dst_ops ipv4_dst_blackhole_ops = {
2474 .check = ipv4_blackhole_dst_check,
2475 .mtu = ipv4_blackhole_mtu,
2476 .default_advmss = ipv4_default_advmss,
2477 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2478 .redirect = ipv4_rt_blackhole_redirect,
2479 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2480 .neigh_lookup = ipv4_neigh_lookup,
2483 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2485 struct rtable *ort = (struct rtable *) dst_orig;
2488 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2490 struct dst_entry *new = &rt->dst;
2493 new->input = dst_discard;
2494 new->output = dst_discard_out;
2496 new->dev = ort->dst.dev;
2500 rt->rt_is_input = ort->rt_is_input;
2501 rt->rt_iif = ort->rt_iif;
2502 rt->rt_pmtu = ort->rt_pmtu;
2504 rt->rt_genid = rt_genid_ipv4(net);
2505 rt->rt_flags = ort->rt_flags;
2506 rt->rt_type = ort->rt_type;
2507 rt->rt_gateway = ort->rt_gateway;
2508 rt->rt_uses_gateway = ort->rt_uses_gateway;
2510 INIT_LIST_HEAD(&rt->rt_uncached);
2514 dst_release(dst_orig);
2516 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2519 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2520 const struct sock *sk)
2522 struct rtable *rt = __ip_route_output_key(net, flp4);
2527 if (flp4->flowi4_proto)
2528 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2529 flowi4_to_flowi(flp4),
2534 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2536 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2537 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2540 struct rtable *rt = skb_rtable(skb);
2542 struct nlmsghdr *nlh;
2543 unsigned long expires = 0;
2545 u32 metrics[RTAX_MAX];
2547 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), 0);
2551 r = nlmsg_data(nlh);
2552 r->rtm_family = AF_INET;
2553 r->rtm_dst_len = 32;
2555 r->rtm_tos = fl4->flowi4_tos;
2556 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2557 if (nla_put_u32(skb, RTA_TABLE, table_id))
2558 goto nla_put_failure;
2559 r->rtm_type = rt->rt_type;
2560 r->rtm_scope = RT_SCOPE_UNIVERSE;
2561 r->rtm_protocol = RTPROT_UNSPEC;
2562 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2563 if (rt->rt_flags & RTCF_NOTIFY)
2564 r->rtm_flags |= RTM_F_NOTIFY;
2565 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2566 r->rtm_flags |= RTCF_DOREDIRECT;
2568 if (nla_put_in_addr(skb, RTA_DST, dst))
2569 goto nla_put_failure;
2571 r->rtm_src_len = 32;
2572 if (nla_put_in_addr(skb, RTA_SRC, src))
2573 goto nla_put_failure;
2576 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2577 goto nla_put_failure;
2578 #ifdef CONFIG_IP_ROUTE_CLASSID
2579 if (rt->dst.tclassid &&
2580 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2581 goto nla_put_failure;
2583 if (!rt_is_input_route(rt) &&
2584 fl4->saddr != src) {
2585 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2586 goto nla_put_failure;
2588 if (rt->rt_uses_gateway &&
2589 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2590 goto nla_put_failure;
2592 expires = rt->dst.expires;
2594 unsigned long now = jiffies;
2596 if (time_before(now, expires))
2602 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2603 if (rt->rt_pmtu && expires)
2604 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2605 if (rtnetlink_put_metrics(skb, metrics) < 0)
2606 goto nla_put_failure;
2608 if (fl4->flowi4_mark &&
2609 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2610 goto nla_put_failure;
2612 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2613 nla_put_u32(skb, RTA_UID,
2614 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2615 goto nla_put_failure;
2617 error = rt->dst.error;
2619 if (rt_is_input_route(rt)) {
2620 #ifdef CONFIG_IP_MROUTE
2621 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2622 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2623 int err = ipmr_get_route(net, skb,
2624 fl4->saddr, fl4->daddr,
2630 goto nla_put_failure;
2634 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2635 goto nla_put_failure;
2638 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2639 goto nla_put_failure;
2641 nlmsg_end(skb, nlh);
2645 nlmsg_cancel(skb, nlh);
2649 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2650 struct netlink_ext_ack *extack)
2652 struct net *net = sock_net(in_skb->sk);
2654 struct nlattr *tb[RTA_MAX+1];
2655 struct rtable *rt = NULL;
2662 struct sk_buff *skb;
2663 u32 table_id = RT_TABLE_MAIN;
2666 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2671 rtm = nlmsg_data(nlh);
2673 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2679 /* Reserve room for dummy headers, this skb can pass
2680 through good chunk of routing engine.
2682 skb_reset_mac_header(skb);
2683 skb_reset_network_header(skb);
2685 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2686 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2687 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2688 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2690 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2692 uid = (iif ? INVALID_UID : current_uid());
2694 /* Bugfix: need to give ip_route_input enough of an IP header to
2697 ip_hdr(skb)->protocol = IPPROTO_UDP;
2698 ip_hdr(skb)->saddr = src;
2699 ip_hdr(skb)->daddr = dst;
2701 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2703 memset(&fl4, 0, sizeof(fl4));
2706 fl4.flowi4_tos = rtm->rtm_tos;
2707 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2708 fl4.flowi4_mark = mark;
2709 fl4.flowi4_uid = uid;
2712 struct net_device *dev;
2714 dev = __dev_get_by_index(net, iif);
2720 skb->protocol = htons(ETH_P_IP);
2723 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2725 rt = skb_rtable(skb);
2726 if (err == 0 && rt->dst.error)
2727 err = -rt->dst.error;
2729 rt = ip_route_output_key(net, &fl4);
2739 skb_dst_set(skb, &rt->dst);
2740 if (rtm->rtm_flags & RTM_F_NOTIFY)
2741 rt->rt_flags |= RTCF_NOTIFY;
2743 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2744 table_id = rt->rt_table_id;
2746 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2747 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2752 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2761 void ip_rt_multicast_event(struct in_device *in_dev)
2763 rt_cache_flush(dev_net(in_dev->dev));
2766 #ifdef CONFIG_SYSCTL
2767 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2768 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2769 static int ip_rt_gc_elasticity __read_mostly = 8;
2771 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2772 void __user *buffer,
2773 size_t *lenp, loff_t *ppos)
2775 struct net *net = (struct net *)__ctl->extra1;
2778 rt_cache_flush(net);
2779 fnhe_genid_bump(net);
2786 static struct ctl_table ipv4_route_table[] = {
2788 .procname = "gc_thresh",
2789 .data = &ipv4_dst_ops.gc_thresh,
2790 .maxlen = sizeof(int),
2792 .proc_handler = proc_dointvec,
2795 .procname = "max_size",
2796 .data = &ip_rt_max_size,
2797 .maxlen = sizeof(int),
2799 .proc_handler = proc_dointvec,
2802 /* Deprecated. Use gc_min_interval_ms */
2804 .procname = "gc_min_interval",
2805 .data = &ip_rt_gc_min_interval,
2806 .maxlen = sizeof(int),
2808 .proc_handler = proc_dointvec_jiffies,
2811 .procname = "gc_min_interval_ms",
2812 .data = &ip_rt_gc_min_interval,
2813 .maxlen = sizeof(int),
2815 .proc_handler = proc_dointvec_ms_jiffies,
2818 .procname = "gc_timeout",
2819 .data = &ip_rt_gc_timeout,
2820 .maxlen = sizeof(int),
2822 .proc_handler = proc_dointvec_jiffies,
2825 .procname = "gc_interval",
2826 .data = &ip_rt_gc_interval,
2827 .maxlen = sizeof(int),
2829 .proc_handler = proc_dointvec_jiffies,
2832 .procname = "redirect_load",
2833 .data = &ip_rt_redirect_load,
2834 .maxlen = sizeof(int),
2836 .proc_handler = proc_dointvec,
2839 .procname = "redirect_number",
2840 .data = &ip_rt_redirect_number,
2841 .maxlen = sizeof(int),
2843 .proc_handler = proc_dointvec,
2846 .procname = "redirect_silence",
2847 .data = &ip_rt_redirect_silence,
2848 .maxlen = sizeof(int),
2850 .proc_handler = proc_dointvec,
2853 .procname = "error_cost",
2854 .data = &ip_rt_error_cost,
2855 .maxlen = sizeof(int),
2857 .proc_handler = proc_dointvec,
2860 .procname = "error_burst",
2861 .data = &ip_rt_error_burst,
2862 .maxlen = sizeof(int),
2864 .proc_handler = proc_dointvec,
2867 .procname = "gc_elasticity",
2868 .data = &ip_rt_gc_elasticity,
2869 .maxlen = sizeof(int),
2871 .proc_handler = proc_dointvec,
2874 .procname = "mtu_expires",
2875 .data = &ip_rt_mtu_expires,
2876 .maxlen = sizeof(int),
2878 .proc_handler = proc_dointvec_jiffies,
2881 .procname = "min_pmtu",
2882 .data = &ip_rt_min_pmtu,
2883 .maxlen = sizeof(int),
2885 .proc_handler = proc_dointvec,
2888 .procname = "min_adv_mss",
2889 .data = &ip_rt_min_advmss,
2890 .maxlen = sizeof(int),
2892 .proc_handler = proc_dointvec,
2897 static struct ctl_table ipv4_route_flush_table[] = {
2899 .procname = "flush",
2900 .maxlen = sizeof(int),
2902 .proc_handler = ipv4_sysctl_rtcache_flush,
2907 static __net_init int sysctl_route_net_init(struct net *net)
2909 struct ctl_table *tbl;
2911 tbl = ipv4_route_flush_table;
2912 if (!net_eq(net, &init_net)) {
2913 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2917 /* Don't export sysctls to unprivileged users */
2918 if (net->user_ns != &init_user_ns)
2919 tbl[0].procname = NULL;
2921 tbl[0].extra1 = net;
2923 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2924 if (!net->ipv4.route_hdr)
2929 if (tbl != ipv4_route_flush_table)
2935 static __net_exit void sysctl_route_net_exit(struct net *net)
2937 struct ctl_table *tbl;
2939 tbl = net->ipv4.route_hdr->ctl_table_arg;
2940 unregister_net_sysctl_table(net->ipv4.route_hdr);
2941 BUG_ON(tbl == ipv4_route_flush_table);
2945 static __net_initdata struct pernet_operations sysctl_route_ops = {
2946 .init = sysctl_route_net_init,
2947 .exit = sysctl_route_net_exit,
2951 static __net_init int rt_genid_init(struct net *net)
2953 atomic_set(&net->ipv4.rt_genid, 0);
2954 atomic_set(&net->fnhe_genid, 0);
2955 get_random_bytes(&net->ipv4.dev_addr_genid,
2956 sizeof(net->ipv4.dev_addr_genid));
2960 static __net_initdata struct pernet_operations rt_genid_ops = {
2961 .init = rt_genid_init,
2964 static int __net_init ipv4_inetpeer_init(struct net *net)
2966 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2970 inet_peer_base_init(bp);
2971 net->ipv4.peers = bp;
2975 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2977 struct inet_peer_base *bp = net->ipv4.peers;
2979 net->ipv4.peers = NULL;
2980 inetpeer_invalidate_tree(bp);
2984 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2985 .init = ipv4_inetpeer_init,
2986 .exit = ipv4_inetpeer_exit,
2989 #ifdef CONFIG_IP_ROUTE_CLASSID
2990 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2991 #endif /* CONFIG_IP_ROUTE_CLASSID */
2993 int __init ip_rt_init(void)
2998 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3000 panic("IP: failed to allocate ip_idents\n");
3002 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3004 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3006 panic("IP: failed to allocate ip_tstamps\n");
3008 for_each_possible_cpu(cpu) {
3009 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3011 INIT_LIST_HEAD(&ul->head);
3012 spin_lock_init(&ul->lock);
3014 #ifdef CONFIG_IP_ROUTE_CLASSID
3015 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3017 panic("IP: failed to allocate ip_rt_acct\n");
3020 ipv4_dst_ops.kmem_cachep =
3021 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3022 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3024 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3026 if (dst_entries_init(&ipv4_dst_ops) < 0)
3027 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3029 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3030 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3032 ipv4_dst_ops.gc_thresh = ~0;
3033 ip_rt_max_size = INT_MAX;
3038 if (ip_rt_proc_init())
3039 pr_err("Unable to create route proc files\n");
3044 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3046 #ifdef CONFIG_SYSCTL
3047 register_pernet_subsys(&sysctl_route_ops);
3049 register_pernet_subsys(&rt_genid_ops);
3050 register_pernet_subsys(&ipv4_inetpeer_ops);
3054 #ifdef CONFIG_SYSCTL
3056 * We really need to sanitize the damn ipv4 init order, then all
3057 * this nonsense will go away.
3059 void __init ip_static_sysctl_init(void)
3061 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);