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 dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1389 struct rtable *rt = (struct rtable *) dst;
1391 if (p != &dst_default_metrics && atomic_dec_and_test(&p->refcnt))
1394 if (!list_empty(&rt->rt_uncached)) {
1395 struct uncached_list *ul = rt->rt_uncached_list;
1397 spin_lock_bh(&ul->lock);
1398 list_del(&rt->rt_uncached);
1399 spin_unlock_bh(&ul->lock);
1403 void rt_flush_dev(struct net_device *dev)
1405 struct net *net = dev_net(dev);
1409 for_each_possible_cpu(cpu) {
1410 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1412 spin_lock_bh(&ul->lock);
1413 list_for_each_entry(rt, &ul->head, rt_uncached) {
1414 if (rt->dst.dev != dev)
1416 rt->dst.dev = net->loopback_dev;
1417 dev_hold(rt->dst.dev);
1420 spin_unlock_bh(&ul->lock);
1424 static bool rt_cache_valid(const struct rtable *rt)
1427 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1431 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1432 const struct fib_result *res,
1433 struct fib_nh_exception *fnhe,
1434 struct fib_info *fi, u16 type, u32 itag)
1436 bool cached = false;
1439 struct fib_nh *nh = &FIB_RES_NH(*res);
1441 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1442 rt->rt_gateway = nh->nh_gw;
1443 rt->rt_uses_gateway = 1;
1445 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1446 if (fi->fib_metrics != &dst_default_metrics) {
1447 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1448 atomic_inc(&fi->fib_metrics->refcnt);
1450 #ifdef CONFIG_IP_ROUTE_CLASSID
1451 rt->dst.tclassid = nh->nh_tclassid;
1453 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1455 cached = rt_bind_exception(rt, fnhe, daddr);
1456 else if (!(rt->dst.flags & DST_NOCACHE))
1457 cached = rt_cache_route(nh, rt);
1458 if (unlikely(!cached)) {
1459 /* Routes we intend to cache in nexthop exception or
1460 * FIB nexthop have the DST_NOCACHE bit clear.
1461 * However, if we are unsuccessful at storing this
1462 * route into the cache we really need to set it.
1464 rt->dst.flags |= DST_NOCACHE;
1465 if (!rt->rt_gateway)
1466 rt->rt_gateway = daddr;
1467 rt_add_uncached_list(rt);
1470 rt_add_uncached_list(rt);
1472 #ifdef CONFIG_IP_ROUTE_CLASSID
1473 #ifdef CONFIG_IP_MULTIPLE_TABLES
1474 set_class_tag(rt, res->tclassid);
1476 set_class_tag(rt, itag);
1480 struct rtable *rt_dst_alloc(struct net_device *dev,
1481 unsigned int flags, u16 type,
1482 bool nopolicy, bool noxfrm, bool will_cache)
1486 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1487 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1488 (nopolicy ? DST_NOPOLICY : 0) |
1489 (noxfrm ? DST_NOXFRM : 0));
1492 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1493 rt->rt_flags = flags;
1495 rt->rt_is_input = 0;
1499 rt->rt_uses_gateway = 0;
1500 rt->rt_table_id = 0;
1501 INIT_LIST_HEAD(&rt->rt_uncached);
1503 rt->dst.output = ip_output;
1504 if (flags & RTCF_LOCAL)
1505 rt->dst.input = ip_local_deliver;
1510 EXPORT_SYMBOL(rt_dst_alloc);
1512 /* called in rcu_read_lock() section */
1513 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1514 u8 tos, struct net_device *dev, int our)
1517 struct in_device *in_dev = __in_dev_get_rcu(dev);
1518 unsigned int flags = RTCF_MULTICAST;
1522 /* Primary sanity checks. */
1527 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1528 skb->protocol != htons(ETH_P_IP))
1531 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1534 if (ipv4_is_zeronet(saddr)) {
1535 if (!ipv4_is_local_multicast(daddr))
1538 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1544 flags |= RTCF_LOCAL;
1546 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1547 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1551 #ifdef CONFIG_IP_ROUTE_CLASSID
1552 rth->dst.tclassid = itag;
1554 rth->dst.output = ip_rt_bug;
1555 rth->rt_is_input= 1;
1557 #ifdef CONFIG_IP_MROUTE
1558 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1559 rth->dst.input = ip_mr_input;
1561 RT_CACHE_STAT_INC(in_slow_mc);
1563 skb_dst_set(skb, &rth->dst);
1575 static void ip_handle_martian_source(struct net_device *dev,
1576 struct in_device *in_dev,
1577 struct sk_buff *skb,
1581 RT_CACHE_STAT_INC(in_martian_src);
1582 #ifdef CONFIG_IP_ROUTE_VERBOSE
1583 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1585 * RFC1812 recommendation, if source is martian,
1586 * the only hint is MAC header.
1588 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1589 &daddr, &saddr, dev->name);
1590 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1591 print_hex_dump(KERN_WARNING, "ll header: ",
1592 DUMP_PREFIX_OFFSET, 16, 1,
1593 skb_mac_header(skb),
1594 dev->hard_header_len, true);
1600 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1602 struct fnhe_hash_bucket *hash;
1603 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1604 u32 hval = fnhe_hashfun(daddr);
1606 spin_lock_bh(&fnhe_lock);
1608 hash = rcu_dereference_protected(nh->nh_exceptions,
1609 lockdep_is_held(&fnhe_lock));
1612 fnhe_p = &hash->chain;
1613 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1615 if (fnhe->fnhe_daddr == daddr) {
1616 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1617 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1618 fnhe_flush_routes(fnhe);
1619 kfree_rcu(fnhe, rcu);
1622 fnhe_p = &fnhe->fnhe_next;
1623 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1624 lockdep_is_held(&fnhe_lock));
1627 spin_unlock_bh(&fnhe_lock);
1630 static void set_lwt_redirect(struct rtable *rth)
1632 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1633 rth->dst.lwtstate->orig_output = rth->dst.output;
1634 rth->dst.output = lwtunnel_output;
1637 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1638 rth->dst.lwtstate->orig_input = rth->dst.input;
1639 rth->dst.input = lwtunnel_input;
1643 /* called in rcu_read_lock() section */
1644 static int __mkroute_input(struct sk_buff *skb,
1645 const struct fib_result *res,
1646 struct in_device *in_dev,
1647 __be32 daddr, __be32 saddr, u32 tos)
1649 struct fib_nh_exception *fnhe;
1652 struct in_device *out_dev;
1656 /* get a working reference to the output device */
1657 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1659 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1663 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1664 in_dev->dev, in_dev, &itag);
1666 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1672 do_cache = res->fi && !itag;
1673 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1674 skb->protocol == htons(ETH_P_IP) &&
1675 (IN_DEV_SHARED_MEDIA(out_dev) ||
1676 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1677 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1679 if (skb->protocol != htons(ETH_P_IP)) {
1680 /* Not IP (i.e. ARP). Do not create route, if it is
1681 * invalid for proxy arp. DNAT routes are always valid.
1683 * Proxy arp feature have been extended to allow, ARP
1684 * replies back to the same interface, to support
1685 * Private VLAN switch technologies. See arp.c.
1687 if (out_dev == in_dev &&
1688 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1694 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1697 rth = rcu_dereference(fnhe->fnhe_rth_input);
1698 if (rth && rth->dst.expires &&
1699 time_after(jiffies, rth->dst.expires)) {
1700 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1707 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1710 if (rt_cache_valid(rth)) {
1711 skb_dst_set_noref(skb, &rth->dst);
1716 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1717 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1718 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1724 rth->rt_is_input = 1;
1726 rth->rt_table_id = res->table->tb_id;
1727 RT_CACHE_STAT_INC(in_slow_tot);
1729 rth->dst.input = ip_forward;
1731 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1732 set_lwt_redirect(rth);
1733 skb_dst_set(skb, &rth->dst);
1740 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1741 /* To make ICMP packets follow the right flow, the multipath hash is
1742 * calculated from the inner IP addresses.
1744 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1745 struct flow_keys *hash_keys)
1747 const struct iphdr *outer_iph = ip_hdr(skb);
1748 const struct iphdr *inner_iph;
1749 const struct icmphdr *icmph;
1750 struct iphdr _inner_iph;
1751 struct icmphdr _icmph;
1753 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1754 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1755 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1758 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1761 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1766 if (icmph->type != ICMP_DEST_UNREACH &&
1767 icmph->type != ICMP_REDIRECT &&
1768 icmph->type != ICMP_TIME_EXCEEDED &&
1769 icmph->type != ICMP_PARAMETERPROB)
1772 inner_iph = skb_header_pointer(skb,
1773 outer_iph->ihl * 4 + sizeof(_icmph),
1774 sizeof(_inner_iph), &_inner_iph);
1777 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1778 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1781 /* if skb is set it will be used and fl4 can be NULL */
1782 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1783 const struct sk_buff *skb)
1785 struct net *net = fi->fib_net;
1786 struct flow_keys hash_keys;
1789 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1791 memset(&hash_keys, 0, sizeof(hash_keys));
1792 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1794 ip_multipath_l3_keys(skb, &hash_keys);
1796 hash_keys.addrs.v4addrs.src = fl4->saddr;
1797 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1801 /* skb is currently provided only when forwarding */
1803 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1804 struct flow_keys keys;
1806 /* short-circuit if we already have L4 hash present */
1808 return skb_get_hash_raw(skb) >> 1;
1809 memset(&hash_keys, 0, sizeof(hash_keys));
1810 skb_flow_dissect_flow_keys(skb, &keys, flag);
1811 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1812 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1813 hash_keys.ports.src = keys.ports.src;
1814 hash_keys.ports.dst = keys.ports.dst;
1815 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1817 memset(&hash_keys, 0, sizeof(hash_keys));
1818 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1819 hash_keys.addrs.v4addrs.src = fl4->saddr;
1820 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1821 hash_keys.ports.src = fl4->fl4_sport;
1822 hash_keys.ports.dst = fl4->fl4_dport;
1823 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1827 mhash = flow_hash_from_keys(&hash_keys);
1831 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1832 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1834 static int ip_mkroute_input(struct sk_buff *skb,
1835 struct fib_result *res,
1836 struct in_device *in_dev,
1837 __be32 daddr, __be32 saddr, u32 tos)
1839 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1840 if (res->fi && res->fi->fib_nhs > 1) {
1841 int h = fib_multipath_hash(res->fi, NULL, skb);
1843 fib_select_multipath(res, h);
1847 /* create a routing cache entry */
1848 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1852 * NOTE. We drop all the packets that has local source
1853 * addresses, because every properly looped back packet
1854 * must have correct destination already attached by output routine.
1856 * Such approach solves two big problems:
1857 * 1. Not simplex devices are handled properly.
1858 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1859 * called with rcu_read_lock()
1862 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1863 u8 tos, struct net_device *dev)
1865 struct fib_result res;
1866 struct in_device *in_dev = __in_dev_get_rcu(dev);
1867 struct ip_tunnel_info *tun_info;
1869 unsigned int flags = 0;
1873 struct net *net = dev_net(dev);
1876 /* IP on this device is disabled. */
1881 /* Check for the most weird martians, which can be not detected
1885 tun_info = skb_tunnel_info(skb);
1886 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1887 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1889 fl4.flowi4_tun_key.tun_id = 0;
1892 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1893 goto martian_source;
1897 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1900 /* Accept zero addresses only to limited broadcast;
1901 * I even do not know to fix it or not. Waiting for complains :-)
1903 if (ipv4_is_zeronet(saddr))
1904 goto martian_source;
1906 if (ipv4_is_zeronet(daddr))
1907 goto martian_destination;
1909 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1910 * and call it once if daddr or/and saddr are loopback addresses
1912 if (ipv4_is_loopback(daddr)) {
1913 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1914 goto martian_destination;
1915 } else if (ipv4_is_loopback(saddr)) {
1916 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1917 goto martian_source;
1921 * Now we are ready to route packet.
1924 fl4.flowi4_iif = dev->ifindex;
1925 fl4.flowi4_mark = skb->mark;
1926 fl4.flowi4_tos = tos;
1927 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1928 fl4.flowi4_flags = 0;
1931 fl4.flowi4_uid = sock_net_uid(net, NULL);
1932 err = fib_lookup(net, &fl4, &res, 0);
1934 if (!IN_DEV_FORWARD(in_dev))
1935 err = -EHOSTUNREACH;
1939 if (res.type == RTN_BROADCAST)
1942 if (res.type == RTN_LOCAL) {
1943 err = fib_validate_source(skb, saddr, daddr, tos,
1944 0, dev, in_dev, &itag);
1946 goto martian_source;
1950 if (!IN_DEV_FORWARD(in_dev)) {
1951 err = -EHOSTUNREACH;
1954 if (res.type != RTN_UNICAST)
1955 goto martian_destination;
1957 err = ip_mkroute_input(skb, &res, in_dev, daddr, saddr, tos);
1961 if (skb->protocol != htons(ETH_P_IP))
1964 if (!ipv4_is_zeronet(saddr)) {
1965 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1968 goto martian_source;
1970 flags |= RTCF_BROADCAST;
1971 res.type = RTN_BROADCAST;
1972 RT_CACHE_STAT_INC(in_brd);
1978 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1979 if (rt_cache_valid(rth)) {
1980 skb_dst_set_noref(skb, &rth->dst);
1988 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
1989 flags | RTCF_LOCAL, res.type,
1990 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1994 rth->dst.output= ip_rt_bug;
1995 #ifdef CONFIG_IP_ROUTE_CLASSID
1996 rth->dst.tclassid = itag;
1998 rth->rt_is_input = 1;
2000 rth->rt_table_id = res.table->tb_id;
2002 RT_CACHE_STAT_INC(in_slow_tot);
2003 if (res.type == RTN_UNREACHABLE) {
2004 rth->dst.input= ip_error;
2005 rth->dst.error= -err;
2006 rth->rt_flags &= ~RTCF_LOCAL;
2010 struct fib_nh *nh = &FIB_RES_NH(res);
2012 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2013 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2014 WARN_ON(rth->dst.input == lwtunnel_input);
2015 rth->dst.lwtstate->orig_input = rth->dst.input;
2016 rth->dst.input = lwtunnel_input;
2019 if (unlikely(!rt_cache_route(nh, rth))) {
2020 rth->dst.flags |= DST_NOCACHE;
2021 rt_add_uncached_list(rth);
2024 skb_dst_set(skb, &rth->dst);
2029 RT_CACHE_STAT_INC(in_no_route);
2030 res.type = RTN_UNREACHABLE;
2036 * Do not cache martian addresses: they should be logged (RFC1812)
2038 martian_destination:
2039 RT_CACHE_STAT_INC(in_martian_dst);
2040 #ifdef CONFIG_IP_ROUTE_VERBOSE
2041 if (IN_DEV_LOG_MARTIANS(in_dev))
2042 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2043 &daddr, &saddr, dev->name);
2055 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2059 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2060 u8 tos, struct net_device *dev)
2064 tos &= IPTOS_RT_MASK;
2067 /* Multicast recognition logic is moved from route cache to here.
2068 The problem was that too many Ethernet cards have broken/missing
2069 hardware multicast filters :-( As result the host on multicasting
2070 network acquires a lot of useless route cache entries, sort of
2071 SDR messages from all the world. Now we try to get rid of them.
2072 Really, provided software IP multicast filter is organized
2073 reasonably (at least, hashed), it does not result in a slowdown
2074 comparing with route cache reject entries.
2075 Note, that multicast routers are not affected, because
2076 route cache entry is created eventually.
2078 if (ipv4_is_multicast(daddr)) {
2079 struct in_device *in_dev = __in_dev_get_rcu(dev);
2083 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2084 ip_hdr(skb)->protocol);
2086 /* check l3 master if no match yet */
2087 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2088 struct in_device *l3_in_dev;
2090 l3_in_dev = __in_dev_get_rcu(skb->dev);
2092 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2093 ip_hdr(skb)->protocol);
2098 #ifdef CONFIG_IP_MROUTE
2100 (!ipv4_is_local_multicast(daddr) &&
2101 IN_DEV_MFORWARD(in_dev))
2104 res = ip_route_input_mc(skb, daddr, saddr,
2110 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2114 EXPORT_SYMBOL(ip_route_input_noref);
2116 /* called with rcu_read_lock() */
2117 static struct rtable *__mkroute_output(const struct fib_result *res,
2118 const struct flowi4 *fl4, int orig_oif,
2119 struct net_device *dev_out,
2122 struct fib_info *fi = res->fi;
2123 struct fib_nh_exception *fnhe;
2124 struct in_device *in_dev;
2125 u16 type = res->type;
2129 in_dev = __in_dev_get_rcu(dev_out);
2131 return ERR_PTR(-EINVAL);
2133 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2134 if (ipv4_is_loopback(fl4->saddr) &&
2135 !(dev_out->flags & IFF_LOOPBACK) &&
2136 !netif_is_l3_master(dev_out))
2137 return ERR_PTR(-EINVAL);
2139 if (ipv4_is_lbcast(fl4->daddr))
2140 type = RTN_BROADCAST;
2141 else if (ipv4_is_multicast(fl4->daddr))
2142 type = RTN_MULTICAST;
2143 else if (ipv4_is_zeronet(fl4->daddr))
2144 return ERR_PTR(-EINVAL);
2146 if (dev_out->flags & IFF_LOOPBACK)
2147 flags |= RTCF_LOCAL;
2150 if (type == RTN_BROADCAST) {
2151 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2153 } else if (type == RTN_MULTICAST) {
2154 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2155 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2157 flags &= ~RTCF_LOCAL;
2160 /* If multicast route do not exist use
2161 * default one, but do not gateway in this case.
2164 if (fi && res->prefixlen < 4)
2166 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2167 (orig_oif != dev_out->ifindex)) {
2168 /* For local routes that require a particular output interface
2169 * we do not want to cache the result. Caching the result
2170 * causes incorrect behaviour when there are multiple source
2171 * addresses on the interface, the end result being that if the
2172 * intended recipient is waiting on that interface for the
2173 * packet he won't receive it because it will be delivered on
2174 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2175 * be set to the loopback interface as well.
2181 do_cache &= fi != NULL;
2183 struct rtable __rcu **prth;
2184 struct fib_nh *nh = &FIB_RES_NH(*res);
2186 fnhe = find_exception(nh, fl4->daddr);
2188 prth = &fnhe->fnhe_rth_output;
2189 rth = rcu_dereference(*prth);
2190 if (rth && rth->dst.expires &&
2191 time_after(jiffies, rth->dst.expires)) {
2192 ip_del_fnhe(nh, fl4->daddr);
2199 if (unlikely(fl4->flowi4_flags &
2200 FLOWI_FLAG_KNOWN_NH &&
2202 nh->nh_scope == RT_SCOPE_LINK))) {
2206 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2207 rth = rcu_dereference(*prth);
2210 if (rt_cache_valid(rth)) {
2211 dst_hold(&rth->dst);
2217 rth = rt_dst_alloc(dev_out, flags, type,
2218 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2219 IN_DEV_CONF_GET(in_dev, NOXFRM),
2222 return ERR_PTR(-ENOBUFS);
2224 rth->rt_iif = orig_oif ? : 0;
2226 rth->rt_table_id = res->table->tb_id;
2228 RT_CACHE_STAT_INC(out_slow_tot);
2230 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2231 if (flags & RTCF_LOCAL &&
2232 !(dev_out->flags & IFF_LOOPBACK)) {
2233 rth->dst.output = ip_mc_output;
2234 RT_CACHE_STAT_INC(out_slow_mc);
2236 #ifdef CONFIG_IP_MROUTE
2237 if (type == RTN_MULTICAST) {
2238 if (IN_DEV_MFORWARD(in_dev) &&
2239 !ipv4_is_local_multicast(fl4->daddr)) {
2240 rth->dst.input = ip_mr_input;
2241 rth->dst.output = ip_mc_output;
2247 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2248 set_lwt_redirect(rth);
2254 * Major route resolver routine.
2257 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2258 const struct sk_buff *skb)
2260 struct net_device *dev_out = NULL;
2261 __u8 tos = RT_FL_TOS(fl4);
2262 unsigned int flags = 0;
2263 struct fib_result res;
2266 int err = -ENETUNREACH;
2272 orig_oif = fl4->flowi4_oif;
2274 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2275 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2276 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2277 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2281 rth = ERR_PTR(-EINVAL);
2282 if (ipv4_is_multicast(fl4->saddr) ||
2283 ipv4_is_lbcast(fl4->saddr) ||
2284 ipv4_is_zeronet(fl4->saddr))
2287 /* I removed check for oif == dev_out->oif here.
2288 It was wrong for two reasons:
2289 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2290 is assigned to multiple interfaces.
2291 2. Moreover, we are allowed to send packets with saddr
2292 of another iface. --ANK
2295 if (fl4->flowi4_oif == 0 &&
2296 (ipv4_is_multicast(fl4->daddr) ||
2297 ipv4_is_lbcast(fl4->daddr))) {
2298 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2299 dev_out = __ip_dev_find(net, fl4->saddr, false);
2303 /* Special hack: user can direct multicasts
2304 and limited broadcast via necessary interface
2305 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2306 This hack is not just for fun, it allows
2307 vic,vat and friends to work.
2308 They bind socket to loopback, set ttl to zero
2309 and expect that it will work.
2310 From the viewpoint of routing cache they are broken,
2311 because we are not allowed to build multicast path
2312 with loopback source addr (look, routing cache
2313 cannot know, that ttl is zero, so that packet
2314 will not leave this host and route is valid).
2315 Luckily, this hack is good workaround.
2318 fl4->flowi4_oif = dev_out->ifindex;
2322 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2323 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2324 if (!__ip_dev_find(net, fl4->saddr, false))
2330 if (fl4->flowi4_oif) {
2331 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2332 rth = ERR_PTR(-ENODEV);
2336 /* RACE: Check return value of inet_select_addr instead. */
2337 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2338 rth = ERR_PTR(-ENETUNREACH);
2341 if (ipv4_is_local_multicast(fl4->daddr) ||
2342 ipv4_is_lbcast(fl4->daddr) ||
2343 fl4->flowi4_proto == IPPROTO_IGMP) {
2345 fl4->saddr = inet_select_addr(dev_out, 0,
2350 if (ipv4_is_multicast(fl4->daddr))
2351 fl4->saddr = inet_select_addr(dev_out, 0,
2353 else if (!fl4->daddr)
2354 fl4->saddr = inet_select_addr(dev_out, 0,
2360 fl4->daddr = fl4->saddr;
2362 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2363 dev_out = net->loopback_dev;
2364 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2365 res.type = RTN_LOCAL;
2366 flags |= RTCF_LOCAL;
2370 err = fib_lookup(net, fl4, &res, 0);
2374 if (fl4->flowi4_oif &&
2375 (ipv4_is_multicast(fl4->daddr) ||
2376 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2377 /* Apparently, routing tables are wrong. Assume,
2378 that the destination is on link.
2381 Because we are allowed to send to iface
2382 even if it has NO routes and NO assigned
2383 addresses. When oif is specified, routing
2384 tables are looked up with only one purpose:
2385 to catch if destination is gatewayed, rather than
2386 direct. Moreover, if MSG_DONTROUTE is set,
2387 we send packet, ignoring both routing tables
2388 and ifaddr state. --ANK
2391 We could make it even if oif is unknown,
2392 likely IPv6, but we do not.
2395 if (fl4->saddr == 0)
2396 fl4->saddr = inet_select_addr(dev_out, 0,
2398 res.type = RTN_UNICAST;
2405 if (res.type == RTN_LOCAL) {
2407 if (res.fi->fib_prefsrc)
2408 fl4->saddr = res.fi->fib_prefsrc;
2410 fl4->saddr = fl4->daddr;
2413 /* L3 master device is the loopback for that domain */
2414 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(res)) ? :
2416 fl4->flowi4_oif = dev_out->ifindex;
2417 flags |= RTCF_LOCAL;
2421 fib_select_path(net, &res, fl4, skb);
2423 dev_out = FIB_RES_DEV(res);
2424 fl4->flowi4_oif = dev_out->ifindex;
2428 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2434 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2436 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2441 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2443 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2445 return mtu ? : dst->dev->mtu;
2448 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2449 struct sk_buff *skb, u32 mtu)
2453 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2454 struct sk_buff *skb)
2458 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2464 static struct dst_ops ipv4_dst_blackhole_ops = {
2466 .check = ipv4_blackhole_dst_check,
2467 .mtu = ipv4_blackhole_mtu,
2468 .default_advmss = ipv4_default_advmss,
2469 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2470 .redirect = ipv4_rt_blackhole_redirect,
2471 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2472 .neigh_lookup = ipv4_neigh_lookup,
2475 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2477 struct rtable *ort = (struct rtable *) dst_orig;
2480 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2482 struct dst_entry *new = &rt->dst;
2485 new->input = dst_discard;
2486 new->output = dst_discard_out;
2488 new->dev = ort->dst.dev;
2492 rt->rt_is_input = ort->rt_is_input;
2493 rt->rt_iif = ort->rt_iif;
2494 rt->rt_pmtu = ort->rt_pmtu;
2496 rt->rt_genid = rt_genid_ipv4(net);
2497 rt->rt_flags = ort->rt_flags;
2498 rt->rt_type = ort->rt_type;
2499 rt->rt_gateway = ort->rt_gateway;
2500 rt->rt_uses_gateway = ort->rt_uses_gateway;
2502 INIT_LIST_HEAD(&rt->rt_uncached);
2506 dst_release(dst_orig);
2508 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2511 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2512 const struct sock *sk)
2514 struct rtable *rt = __ip_route_output_key(net, flp4);
2519 if (flp4->flowi4_proto)
2520 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2521 flowi4_to_flowi(flp4),
2526 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2528 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2529 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2532 struct rtable *rt = skb_rtable(skb);
2534 struct nlmsghdr *nlh;
2535 unsigned long expires = 0;
2537 u32 metrics[RTAX_MAX];
2539 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), 0);
2543 r = nlmsg_data(nlh);
2544 r->rtm_family = AF_INET;
2545 r->rtm_dst_len = 32;
2547 r->rtm_tos = fl4->flowi4_tos;
2548 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2549 if (nla_put_u32(skb, RTA_TABLE, table_id))
2550 goto nla_put_failure;
2551 r->rtm_type = rt->rt_type;
2552 r->rtm_scope = RT_SCOPE_UNIVERSE;
2553 r->rtm_protocol = RTPROT_UNSPEC;
2554 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2555 if (rt->rt_flags & RTCF_NOTIFY)
2556 r->rtm_flags |= RTM_F_NOTIFY;
2557 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2558 r->rtm_flags |= RTCF_DOREDIRECT;
2560 if (nla_put_in_addr(skb, RTA_DST, dst))
2561 goto nla_put_failure;
2563 r->rtm_src_len = 32;
2564 if (nla_put_in_addr(skb, RTA_SRC, src))
2565 goto nla_put_failure;
2568 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2569 goto nla_put_failure;
2570 #ifdef CONFIG_IP_ROUTE_CLASSID
2571 if (rt->dst.tclassid &&
2572 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2573 goto nla_put_failure;
2575 if (!rt_is_input_route(rt) &&
2576 fl4->saddr != src) {
2577 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2578 goto nla_put_failure;
2580 if (rt->rt_uses_gateway &&
2581 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2582 goto nla_put_failure;
2584 expires = rt->dst.expires;
2586 unsigned long now = jiffies;
2588 if (time_before(now, expires))
2594 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2595 if (rt->rt_pmtu && expires)
2596 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2597 if (rtnetlink_put_metrics(skb, metrics) < 0)
2598 goto nla_put_failure;
2600 if (fl4->flowi4_mark &&
2601 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2602 goto nla_put_failure;
2604 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2605 nla_put_u32(skb, RTA_UID,
2606 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2607 goto nla_put_failure;
2609 error = rt->dst.error;
2611 if (rt_is_input_route(rt)) {
2612 #ifdef CONFIG_IP_MROUTE
2613 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2614 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2615 int err = ipmr_get_route(net, skb,
2616 fl4->saddr, fl4->daddr,
2622 goto nla_put_failure;
2626 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2627 goto nla_put_failure;
2630 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2631 goto nla_put_failure;
2633 nlmsg_end(skb, nlh);
2637 nlmsg_cancel(skb, nlh);
2641 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2642 struct netlink_ext_ack *extack)
2644 struct net *net = sock_net(in_skb->sk);
2646 struct nlattr *tb[RTA_MAX+1];
2647 struct rtable *rt = NULL;
2654 struct sk_buff *skb;
2655 u32 table_id = RT_TABLE_MAIN;
2658 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2663 rtm = nlmsg_data(nlh);
2665 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2671 /* Reserve room for dummy headers, this skb can pass
2672 through good chunk of routing engine.
2674 skb_reset_mac_header(skb);
2675 skb_reset_network_header(skb);
2677 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2678 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2679 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2680 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2682 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2684 uid = (iif ? INVALID_UID : current_uid());
2686 /* Bugfix: need to give ip_route_input enough of an IP header to
2689 ip_hdr(skb)->protocol = IPPROTO_UDP;
2690 ip_hdr(skb)->saddr = src;
2691 ip_hdr(skb)->daddr = dst;
2693 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2695 memset(&fl4, 0, sizeof(fl4));
2698 fl4.flowi4_tos = rtm->rtm_tos;
2699 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2700 fl4.flowi4_mark = mark;
2701 fl4.flowi4_uid = uid;
2704 struct net_device *dev;
2706 dev = __dev_get_by_index(net, iif);
2712 skb->protocol = htons(ETH_P_IP);
2715 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2717 rt = skb_rtable(skb);
2718 if (err == 0 && rt->dst.error)
2719 err = -rt->dst.error;
2721 rt = ip_route_output_key(net, &fl4);
2731 skb_dst_set(skb, &rt->dst);
2732 if (rtm->rtm_flags & RTM_F_NOTIFY)
2733 rt->rt_flags |= RTCF_NOTIFY;
2735 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2736 table_id = rt->rt_table_id;
2738 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2739 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2744 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2753 void ip_rt_multicast_event(struct in_device *in_dev)
2755 rt_cache_flush(dev_net(in_dev->dev));
2758 #ifdef CONFIG_SYSCTL
2759 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2760 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2761 static int ip_rt_gc_elasticity __read_mostly = 8;
2763 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2764 void __user *buffer,
2765 size_t *lenp, loff_t *ppos)
2767 struct net *net = (struct net *)__ctl->extra1;
2770 rt_cache_flush(net);
2771 fnhe_genid_bump(net);
2778 static struct ctl_table ipv4_route_table[] = {
2780 .procname = "gc_thresh",
2781 .data = &ipv4_dst_ops.gc_thresh,
2782 .maxlen = sizeof(int),
2784 .proc_handler = proc_dointvec,
2787 .procname = "max_size",
2788 .data = &ip_rt_max_size,
2789 .maxlen = sizeof(int),
2791 .proc_handler = proc_dointvec,
2794 /* Deprecated. Use gc_min_interval_ms */
2796 .procname = "gc_min_interval",
2797 .data = &ip_rt_gc_min_interval,
2798 .maxlen = sizeof(int),
2800 .proc_handler = proc_dointvec_jiffies,
2803 .procname = "gc_min_interval_ms",
2804 .data = &ip_rt_gc_min_interval,
2805 .maxlen = sizeof(int),
2807 .proc_handler = proc_dointvec_ms_jiffies,
2810 .procname = "gc_timeout",
2811 .data = &ip_rt_gc_timeout,
2812 .maxlen = sizeof(int),
2814 .proc_handler = proc_dointvec_jiffies,
2817 .procname = "gc_interval",
2818 .data = &ip_rt_gc_interval,
2819 .maxlen = sizeof(int),
2821 .proc_handler = proc_dointvec_jiffies,
2824 .procname = "redirect_load",
2825 .data = &ip_rt_redirect_load,
2826 .maxlen = sizeof(int),
2828 .proc_handler = proc_dointvec,
2831 .procname = "redirect_number",
2832 .data = &ip_rt_redirect_number,
2833 .maxlen = sizeof(int),
2835 .proc_handler = proc_dointvec,
2838 .procname = "redirect_silence",
2839 .data = &ip_rt_redirect_silence,
2840 .maxlen = sizeof(int),
2842 .proc_handler = proc_dointvec,
2845 .procname = "error_cost",
2846 .data = &ip_rt_error_cost,
2847 .maxlen = sizeof(int),
2849 .proc_handler = proc_dointvec,
2852 .procname = "error_burst",
2853 .data = &ip_rt_error_burst,
2854 .maxlen = sizeof(int),
2856 .proc_handler = proc_dointvec,
2859 .procname = "gc_elasticity",
2860 .data = &ip_rt_gc_elasticity,
2861 .maxlen = sizeof(int),
2863 .proc_handler = proc_dointvec,
2866 .procname = "mtu_expires",
2867 .data = &ip_rt_mtu_expires,
2868 .maxlen = sizeof(int),
2870 .proc_handler = proc_dointvec_jiffies,
2873 .procname = "min_pmtu",
2874 .data = &ip_rt_min_pmtu,
2875 .maxlen = sizeof(int),
2877 .proc_handler = proc_dointvec,
2880 .procname = "min_adv_mss",
2881 .data = &ip_rt_min_advmss,
2882 .maxlen = sizeof(int),
2884 .proc_handler = proc_dointvec,
2889 static struct ctl_table ipv4_route_flush_table[] = {
2891 .procname = "flush",
2892 .maxlen = sizeof(int),
2894 .proc_handler = ipv4_sysctl_rtcache_flush,
2899 static __net_init int sysctl_route_net_init(struct net *net)
2901 struct ctl_table *tbl;
2903 tbl = ipv4_route_flush_table;
2904 if (!net_eq(net, &init_net)) {
2905 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2909 /* Don't export sysctls to unprivileged users */
2910 if (net->user_ns != &init_user_ns)
2911 tbl[0].procname = NULL;
2913 tbl[0].extra1 = net;
2915 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2916 if (!net->ipv4.route_hdr)
2921 if (tbl != ipv4_route_flush_table)
2927 static __net_exit void sysctl_route_net_exit(struct net *net)
2929 struct ctl_table *tbl;
2931 tbl = net->ipv4.route_hdr->ctl_table_arg;
2932 unregister_net_sysctl_table(net->ipv4.route_hdr);
2933 BUG_ON(tbl == ipv4_route_flush_table);
2937 static __net_initdata struct pernet_operations sysctl_route_ops = {
2938 .init = sysctl_route_net_init,
2939 .exit = sysctl_route_net_exit,
2943 static __net_init int rt_genid_init(struct net *net)
2945 atomic_set(&net->ipv4.rt_genid, 0);
2946 atomic_set(&net->fnhe_genid, 0);
2947 get_random_bytes(&net->ipv4.dev_addr_genid,
2948 sizeof(net->ipv4.dev_addr_genid));
2952 static __net_initdata struct pernet_operations rt_genid_ops = {
2953 .init = rt_genid_init,
2956 static int __net_init ipv4_inetpeer_init(struct net *net)
2958 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2962 inet_peer_base_init(bp);
2963 net->ipv4.peers = bp;
2967 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2969 struct inet_peer_base *bp = net->ipv4.peers;
2971 net->ipv4.peers = NULL;
2972 inetpeer_invalidate_tree(bp);
2976 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2977 .init = ipv4_inetpeer_init,
2978 .exit = ipv4_inetpeer_exit,
2981 #ifdef CONFIG_IP_ROUTE_CLASSID
2982 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2983 #endif /* CONFIG_IP_ROUTE_CLASSID */
2985 int __init ip_rt_init(void)
2990 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2992 panic("IP: failed to allocate ip_idents\n");
2994 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2996 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2998 panic("IP: failed to allocate ip_tstamps\n");
3000 for_each_possible_cpu(cpu) {
3001 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3003 INIT_LIST_HEAD(&ul->head);
3004 spin_lock_init(&ul->lock);
3006 #ifdef CONFIG_IP_ROUTE_CLASSID
3007 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3009 panic("IP: failed to allocate ip_rt_acct\n");
3012 ipv4_dst_ops.kmem_cachep =
3013 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3014 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3016 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3018 if (dst_entries_init(&ipv4_dst_ops) < 0)
3019 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3021 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3022 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3024 ipv4_dst_ops.gc_thresh = ~0;
3025 ip_rt_max_size = INT_MAX;
3030 if (ip_rt_proc_init())
3031 pr_err("Unable to create route proc files\n");
3036 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3038 #ifdef CONFIG_SYSCTL
3039 register_pernet_subsys(&sysctl_route_ops);
3041 register_pernet_subsys(&rt_genid_ops);
3042 register_pernet_subsys(&ipv4_inetpeer_ops);
3046 #ifdef CONFIG_SYSCTL
3048 * We really need to sanitize the damn ipv4 init order, then all
3049 * this nonsense will go away.
3051 void __init ip_static_sysctl_init(void)
3053 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);