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>
112 #include <net/secure_seq.h>
113 #include <net/ip_tunnels.h>
114 #include <net/l3mdev.h>
116 #include "fib_lookup.h"
118 #define RT_FL_TOS(oldflp4) \
119 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size;
124 static int ip_rt_redirect_number __read_mostly = 9;
125 static int ip_rt_redirect_load __read_mostly = HZ / 50;
126 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly = HZ;
128 static int ip_rt_error_burst __read_mostly = 5 * HZ;
129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly = 256;
133 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
136 * Interface to generic destination cache.
139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
140 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
141 static unsigned int ipv4_mtu(const struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void ipv4_link_failure(struct sk_buff *skb);
144 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb, u32 mtu);
146 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
147 struct sk_buff *skb);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
159 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
161 static struct dst_ops ipv4_dst_ops = {
163 .check = ipv4_dst_check,
164 .default_advmss = ipv4_default_advmss,
166 .cow_metrics = ipv4_cow_metrics,
167 .destroy = ipv4_dst_destroy,
168 .negative_advice = ipv4_negative_advice,
169 .link_failure = ipv4_link_failure,
170 .update_pmtu = ip_rt_update_pmtu,
171 .redirect = ip_do_redirect,
172 .local_out = __ip_local_out,
173 .neigh_lookup = ipv4_neigh_lookup,
174 .confirm_neigh = ipv4_confirm_neigh,
177 #define ECN_OR_COST(class) TC_PRIO_##class
179 const __u8 ip_tos2prio[16] = {
181 ECN_OR_COST(BESTEFFORT),
183 ECN_OR_COST(BESTEFFORT),
189 ECN_OR_COST(INTERACTIVE),
191 ECN_OR_COST(INTERACTIVE),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK),
194 TC_PRIO_INTERACTIVE_BULK,
195 ECN_OR_COST(INTERACTIVE_BULK)
197 EXPORT_SYMBOL(ip_tos2prio);
199 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
200 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
202 #ifdef CONFIG_PROC_FS
203 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
207 return SEQ_START_TOKEN;
210 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
216 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
220 static int rt_cache_seq_show(struct seq_file *seq, void *v)
222 if (v == SEQ_START_TOKEN)
223 seq_printf(seq, "%-127s\n",
224 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
225 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
230 static const struct seq_operations rt_cache_seq_ops = {
231 .start = rt_cache_seq_start,
232 .next = rt_cache_seq_next,
233 .stop = rt_cache_seq_stop,
234 .show = rt_cache_seq_show,
237 static int rt_cache_seq_open(struct inode *inode, struct file *file)
239 return seq_open(file, &rt_cache_seq_ops);
242 static const struct file_operations rt_cache_seq_fops = {
243 .open = rt_cache_seq_open,
246 .release = seq_release,
250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
255 return SEQ_START_TOKEN;
257 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
258 if (!cpu_possible(cpu))
261 return &per_cpu(rt_cache_stat, cpu);
266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
270 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
271 if (!cpu_possible(cpu))
274 return &per_cpu(rt_cache_stat, cpu);
280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
285 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
287 struct rt_cache_stat *st = v;
289 if (v == SEQ_START_TOKEN) {
290 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");
294 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
295 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
296 dst_entries_get_slow(&ipv4_dst_ops),
309 0, /* st->gc_total */
310 0, /* st->gc_ignored */
311 0, /* st->gc_goal_miss */
312 0, /* st->gc_dst_overflow */
313 0, /* st->in_hlist_search */
314 0 /* st->out_hlist_search */
319 static const struct seq_operations rt_cpu_seq_ops = {
320 .start = rt_cpu_seq_start,
321 .next = rt_cpu_seq_next,
322 .stop = rt_cpu_seq_stop,
323 .show = rt_cpu_seq_show,
327 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
329 return seq_open(file, &rt_cpu_seq_ops);
332 static const struct file_operations rt_cpu_seq_fops = {
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));
365 static int __net_init ip_rt_do_proc_init(struct net *net)
367 struct proc_dir_entry *pde;
369 pde = proc_create("rt_cache", 0444, net->proc_net,
374 pde = proc_create("rt_cache", 0444,
375 net->proc_net_stat, &rt_cpu_seq_fops);
379 #ifdef CONFIG_IP_ROUTE_CLASSID
380 pde = proc_create_single("rt_acct", 0, net->proc_net,
387 #ifdef CONFIG_IP_ROUTE_CLASSID
389 remove_proc_entry("rt_cache", net->proc_net_stat);
392 remove_proc_entry("rt_cache", net->proc_net);
397 static void __net_exit ip_rt_do_proc_exit(struct net *net)
399 remove_proc_entry("rt_cache", net->proc_net_stat);
400 remove_proc_entry("rt_cache", net->proc_net);
401 #ifdef CONFIG_IP_ROUTE_CLASSID
402 remove_proc_entry("rt_acct", net->proc_net);
406 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
407 .init = ip_rt_do_proc_init,
408 .exit = ip_rt_do_proc_exit,
411 static int __init ip_rt_proc_init(void)
413 return register_pernet_subsys(&ip_rt_proc_ops);
417 static inline int ip_rt_proc_init(void)
421 #endif /* CONFIG_PROC_FS */
423 static inline bool rt_is_expired(const struct rtable *rth)
425 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
428 void rt_cache_flush(struct net *net)
430 rt_genid_bump_ipv4(net);
433 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
437 struct net_device *dev = dst->dev;
438 const __be32 *pkey = daddr;
439 const struct rtable *rt;
442 rt = (const struct rtable *) dst;
444 pkey = (const __be32 *) &rt->rt_gateway;
446 pkey = &ip_hdr(skb)->daddr;
448 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
451 return neigh_create(&arp_tbl, pkey, dev);
454 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
456 struct net_device *dev = dst->dev;
457 const __be32 *pkey = daddr;
458 const struct rtable *rt;
460 rt = (const struct rtable *)dst;
462 pkey = (const __be32 *)&rt->rt_gateway;
465 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
468 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
471 #define IP_IDENTS_SZ 2048u
473 static atomic_t *ip_idents __read_mostly;
474 static u32 *ip_tstamps __read_mostly;
476 /* In order to protect privacy, we add a perturbation to identifiers
477 * if one generator is seldom used. This makes hard for an attacker
478 * to infer how many packets were sent between two points in time.
480 u32 ip_idents_reserve(u32 hash, int segs)
482 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
483 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
484 u32 old = READ_ONCE(*p_tstamp);
485 u32 now = (u32)jiffies;
488 if (old != now && cmpxchg(p_tstamp, old, now) == old)
489 delta = prandom_u32_max(now - old);
491 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
493 old = (u32)atomic_read(p_id);
494 new = old + delta + segs;
495 } while (atomic_cmpxchg(p_id, old, new) != old);
499 EXPORT_SYMBOL(ip_idents_reserve);
501 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
503 static u32 ip_idents_hashrnd __read_mostly;
506 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
508 hash = jhash_3words((__force u32)iph->daddr,
509 (__force u32)iph->saddr,
510 iph->protocol ^ net_hash_mix(net),
512 id = ip_idents_reserve(hash, segs);
515 EXPORT_SYMBOL(__ip_select_ident);
517 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
518 const struct sock *sk,
519 const struct iphdr *iph,
521 u8 prot, u32 mark, int flow_flags)
524 const struct inet_sock *inet = inet_sk(sk);
526 oif = sk->sk_bound_dev_if;
528 tos = RT_CONN_FLAGS(sk);
529 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
531 flowi4_init_output(fl4, oif, mark, tos,
532 RT_SCOPE_UNIVERSE, prot,
534 iph->daddr, iph->saddr, 0, 0,
535 sock_net_uid(net, sk));
538 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
539 const struct sock *sk)
541 const struct net *net = dev_net(skb->dev);
542 const struct iphdr *iph = ip_hdr(skb);
543 int oif = skb->dev->ifindex;
544 u8 tos = RT_TOS(iph->tos);
545 u8 prot = iph->protocol;
546 u32 mark = skb->mark;
548 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
551 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
553 const struct inet_sock *inet = inet_sk(sk);
554 const struct ip_options_rcu *inet_opt;
555 __be32 daddr = inet->inet_daddr;
558 inet_opt = rcu_dereference(inet->inet_opt);
559 if (inet_opt && inet_opt->opt.srr)
560 daddr = inet_opt->opt.faddr;
561 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
562 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
563 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
564 inet_sk_flowi_flags(sk),
565 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
569 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
570 const struct sk_buff *skb)
573 build_skb_flow_key(fl4, skb, sk);
575 build_sk_flow_key(fl4, sk);
578 static DEFINE_SPINLOCK(fnhe_lock);
580 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
584 rt = rcu_dereference(fnhe->fnhe_rth_input);
586 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
587 dst_dev_put(&rt->dst);
588 dst_release(&rt->dst);
590 rt = rcu_dereference(fnhe->fnhe_rth_output);
592 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
593 dst_dev_put(&rt->dst);
594 dst_release(&rt->dst);
598 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
600 struct fib_nh_exception *fnhe, *oldest;
602 oldest = rcu_dereference(hash->chain);
603 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
604 fnhe = rcu_dereference(fnhe->fnhe_next)) {
605 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
608 fnhe_flush_routes(oldest);
612 static inline u32 fnhe_hashfun(__be32 daddr)
614 static u32 fnhe_hashrnd __read_mostly;
617 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
618 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
619 return hash_32(hval, FNHE_HASH_SHIFT);
622 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
624 rt->rt_pmtu = fnhe->fnhe_pmtu;
625 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
626 rt->dst.expires = fnhe->fnhe_expires;
629 rt->rt_flags |= RTCF_REDIRECTED;
630 rt->rt_gateway = fnhe->fnhe_gw;
631 rt->rt_uses_gateway = 1;
635 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
636 u32 pmtu, bool lock, unsigned long expires)
638 struct fnhe_hash_bucket *hash;
639 struct fib_nh_exception *fnhe;
645 genid = fnhe_genid(dev_net(nh->nh_dev));
646 hval = fnhe_hashfun(daddr);
648 spin_lock_bh(&fnhe_lock);
650 hash = rcu_dereference(nh->nh_exceptions);
652 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
655 rcu_assign_pointer(nh->nh_exceptions, hash);
661 for (fnhe = rcu_dereference(hash->chain); fnhe;
662 fnhe = rcu_dereference(fnhe->fnhe_next)) {
663 if (fnhe->fnhe_daddr == daddr)
669 if (fnhe->fnhe_genid != genid)
670 fnhe->fnhe_genid = genid;
674 fnhe->fnhe_pmtu = pmtu;
675 fnhe->fnhe_mtu_locked = lock;
677 fnhe->fnhe_expires = max(1UL, expires);
678 /* Update all cached dsts too */
679 rt = rcu_dereference(fnhe->fnhe_rth_input);
681 fill_route_from_fnhe(rt, fnhe);
682 rt = rcu_dereference(fnhe->fnhe_rth_output);
684 fill_route_from_fnhe(rt, fnhe);
686 if (depth > FNHE_RECLAIM_DEPTH)
687 fnhe = fnhe_oldest(hash);
689 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
693 fnhe->fnhe_next = hash->chain;
694 rcu_assign_pointer(hash->chain, fnhe);
696 fnhe->fnhe_genid = genid;
697 fnhe->fnhe_daddr = daddr;
699 fnhe->fnhe_pmtu = pmtu;
700 fnhe->fnhe_mtu_locked = lock;
701 fnhe->fnhe_expires = max(1UL, expires);
703 /* Exception created; mark the cached routes for the nexthop
704 * stale, so anyone caching it rechecks if this exception
707 rt = rcu_dereference(nh->nh_rth_input);
709 rt->dst.obsolete = DST_OBSOLETE_KILL;
711 for_each_possible_cpu(i) {
712 struct rtable __rcu **prt;
713 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
714 rt = rcu_dereference(*prt);
716 rt->dst.obsolete = DST_OBSOLETE_KILL;
720 fnhe->fnhe_stamp = jiffies;
723 spin_unlock_bh(&fnhe_lock);
726 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
729 __be32 new_gw = icmp_hdr(skb)->un.gateway;
730 __be32 old_gw = ip_hdr(skb)->saddr;
731 struct net_device *dev = skb->dev;
732 struct in_device *in_dev;
733 struct fib_result res;
737 switch (icmp_hdr(skb)->code & 7) {
739 case ICMP_REDIR_NETTOS:
740 case ICMP_REDIR_HOST:
741 case ICMP_REDIR_HOSTTOS:
748 if (rt->rt_gateway != old_gw)
751 in_dev = __in_dev_get_rcu(dev);
756 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
757 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
758 ipv4_is_zeronet(new_gw))
759 goto reject_redirect;
761 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
762 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
763 goto reject_redirect;
764 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
765 goto reject_redirect;
767 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
768 goto reject_redirect;
771 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
773 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
775 if (!(n->nud_state & NUD_VALID)) {
776 neigh_event_send(n, NULL);
778 if (fib_lookup(net, fl4, &res, 0) == 0) {
779 struct fib_nh *nh = &FIB_RES_NH(res);
781 update_or_create_fnhe(nh, fl4->daddr, new_gw,
783 jiffies + ip_rt_gc_timeout);
786 rt->dst.obsolete = DST_OBSOLETE_KILL;
787 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
794 #ifdef CONFIG_IP_ROUTE_VERBOSE
795 if (IN_DEV_LOG_MARTIANS(in_dev)) {
796 const struct iphdr *iph = (const struct iphdr *) skb->data;
797 __be32 daddr = iph->daddr;
798 __be32 saddr = iph->saddr;
800 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
801 " Advised path = %pI4 -> %pI4\n",
802 &old_gw, dev->name, &new_gw,
809 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
813 const struct iphdr *iph = (const struct iphdr *) skb->data;
814 struct net *net = dev_net(skb->dev);
815 int oif = skb->dev->ifindex;
816 u8 tos = RT_TOS(iph->tos);
817 u8 prot = iph->protocol;
818 u32 mark = skb->mark;
820 rt = (struct rtable *) dst;
822 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
823 __ip_do_redirect(rt, skb, &fl4, true);
826 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
828 struct rtable *rt = (struct rtable *)dst;
829 struct dst_entry *ret = dst;
832 if (dst->obsolete > 0) {
835 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
846 * 1. The first ip_rt_redirect_number redirects are sent
847 * with exponential backoff, then we stop sending them at all,
848 * assuming that the host ignores our redirects.
849 * 2. If we did not see packets requiring redirects
850 * during ip_rt_redirect_silence, we assume that the host
851 * forgot redirected route and start to send redirects again.
853 * This algorithm is much cheaper and more intelligent than dumb load limiting
856 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
857 * and "frag. need" (breaks PMTU discovery) in icmp.c.
860 void ip_rt_send_redirect(struct sk_buff *skb)
862 struct rtable *rt = skb_rtable(skb);
863 struct in_device *in_dev;
864 struct inet_peer *peer;
870 in_dev = __in_dev_get_rcu(rt->dst.dev);
871 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
875 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
876 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
879 net = dev_net(rt->dst.dev);
880 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
882 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
883 rt_nexthop(rt, ip_hdr(skb)->daddr));
887 /* No redirected packets during ip_rt_redirect_silence;
888 * reset the algorithm.
890 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
891 peer->rate_tokens = 0;
893 /* Too many ignored redirects; do not send anything
894 * set dst.rate_last to the last seen redirected packet.
896 if (peer->rate_tokens >= ip_rt_redirect_number) {
897 peer->rate_last = jiffies;
901 /* Check for load limit; set rate_last to the latest sent
904 if (peer->rate_tokens == 0 ||
907 (ip_rt_redirect_load << peer->rate_tokens)))) {
908 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
910 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
911 peer->rate_last = jiffies;
913 #ifdef CONFIG_IP_ROUTE_VERBOSE
915 peer->rate_tokens == ip_rt_redirect_number)
916 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
917 &ip_hdr(skb)->saddr, inet_iif(skb),
918 &ip_hdr(skb)->daddr, &gw);
925 static int ip_error(struct sk_buff *skb)
927 struct rtable *rt = skb_rtable(skb);
928 struct net_device *dev = skb->dev;
929 struct in_device *in_dev;
930 struct inet_peer *peer;
936 if (netif_is_l3_master(skb->dev)) {
937 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
942 in_dev = __in_dev_get_rcu(dev);
944 /* IP on this device is disabled. */
948 net = dev_net(rt->dst.dev);
949 if (!IN_DEV_FORWARD(in_dev)) {
950 switch (rt->dst.error) {
952 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
956 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
962 switch (rt->dst.error) {
967 code = ICMP_HOST_UNREACH;
970 code = ICMP_NET_UNREACH;
971 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
974 code = ICMP_PKT_FILTERED;
978 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
979 l3mdev_master_ifindex(skb->dev), 1);
984 peer->rate_tokens += now - peer->rate_last;
985 if (peer->rate_tokens > ip_rt_error_burst)
986 peer->rate_tokens = ip_rt_error_burst;
987 peer->rate_last = now;
988 if (peer->rate_tokens >= ip_rt_error_cost)
989 peer->rate_tokens -= ip_rt_error_cost;
995 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1001 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1003 struct dst_entry *dst = &rt->dst;
1004 struct fib_result res;
1007 if (ip_mtu_locked(dst))
1010 if (ipv4_mtu(dst) < mtu)
1013 if (mtu < ip_rt_min_pmtu) {
1015 mtu = ip_rt_min_pmtu;
1018 if (rt->rt_pmtu == mtu &&
1019 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1023 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1024 struct fib_nh *nh = &FIB_RES_NH(res);
1026 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1027 jiffies + ip_rt_mtu_expires);
1032 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1033 struct sk_buff *skb, u32 mtu)
1035 struct rtable *rt = (struct rtable *) dst;
1038 ip_rt_build_flow_key(&fl4, sk, skb);
1039 __ip_rt_update_pmtu(rt, &fl4, mtu);
1042 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1043 int oif, u8 protocol)
1045 const struct iphdr *iph = (const struct iphdr *) skb->data;
1048 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1050 __build_flow_key(net, &fl4, NULL, iph, oif,
1051 RT_TOS(iph->tos), protocol, mark, 0);
1052 rt = __ip_route_output_key(net, &fl4);
1054 __ip_rt_update_pmtu(rt, &fl4, mtu);
1058 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1060 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1062 const struct iphdr *iph = (const struct iphdr *) skb->data;
1066 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1068 if (!fl4.flowi4_mark)
1069 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1071 rt = __ip_route_output_key(sock_net(sk), &fl4);
1073 __ip_rt_update_pmtu(rt, &fl4, mtu);
1078 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1080 const struct iphdr *iph = (const struct iphdr *) skb->data;
1083 struct dst_entry *odst = NULL;
1085 struct net *net = sock_net(sk);
1089 if (!ip_sk_accept_pmtu(sk))
1092 odst = sk_dst_get(sk);
1094 if (sock_owned_by_user(sk) || !odst) {
1095 __ipv4_sk_update_pmtu(skb, sk, mtu);
1099 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1101 rt = (struct rtable *)odst;
1102 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1103 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1110 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1112 if (!dst_check(&rt->dst, 0)) {
1114 dst_release(&rt->dst);
1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1124 sk_dst_set(sk, &rt->dst);
1130 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1132 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1133 int oif, u8 protocol)
1135 const struct iphdr *iph = (const struct iphdr *) skb->data;
1139 __build_flow_key(net, &fl4, NULL, iph, oif,
1140 RT_TOS(iph->tos), protocol, 0, 0);
1141 rt = __ip_route_output_key(net, &fl4);
1143 __ip_do_redirect(rt, skb, &fl4, false);
1147 EXPORT_SYMBOL_GPL(ipv4_redirect);
1149 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1151 const struct iphdr *iph = (const struct iphdr *) skb->data;
1154 struct net *net = sock_net(sk);
1156 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1157 rt = __ip_route_output_key(net, &fl4);
1159 __ip_do_redirect(rt, skb, &fl4, false);
1163 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1165 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1167 struct rtable *rt = (struct rtable *) dst;
1169 /* All IPV4 dsts are created with ->obsolete set to the value
1170 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1171 * into this function always.
1173 * When a PMTU/redirect information update invalidates a route,
1174 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1175 * DST_OBSOLETE_DEAD by dst_free().
1177 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1182 static void ipv4_link_failure(struct sk_buff *skb)
1186 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1188 rt = skb_rtable(skb);
1190 dst_set_expires(&rt->dst, 0);
1193 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1195 pr_debug("%s: %pI4 -> %pI4, %s\n",
1196 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1197 skb->dev ? skb->dev->name : "?");
1204 We do not cache source address of outgoing interface,
1205 because it is used only by IP RR, TS and SRR options,
1206 so that it out of fast path.
1208 BTW remember: "addr" is allowed to be not aligned
1212 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1216 if (rt_is_output_route(rt))
1217 src = ip_hdr(skb)->saddr;
1219 struct fib_result res;
1220 struct iphdr *iph = ip_hdr(skb);
1221 struct flowi4 fl4 = {
1222 .daddr = iph->daddr,
1223 .saddr = iph->saddr,
1224 .flowi4_tos = RT_TOS(iph->tos),
1225 .flowi4_oif = rt->dst.dev->ifindex,
1226 .flowi4_iif = skb->dev->ifindex,
1227 .flowi4_mark = skb->mark,
1231 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1232 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1234 src = inet_select_addr(rt->dst.dev,
1235 rt_nexthop(rt, iph->daddr),
1239 memcpy(addr, &src, 4);
1242 #ifdef CONFIG_IP_ROUTE_CLASSID
1243 static void set_class_tag(struct rtable *rt, u32 tag)
1245 if (!(rt->dst.tclassid & 0xFFFF))
1246 rt->dst.tclassid |= tag & 0xFFFF;
1247 if (!(rt->dst.tclassid & 0xFFFF0000))
1248 rt->dst.tclassid |= tag & 0xFFFF0000;
1252 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1254 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1255 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1258 return min(advmss, IPV4_MAX_PMTU - header_size);
1261 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1263 const struct rtable *rt = (const struct rtable *) dst;
1264 unsigned int mtu = rt->rt_pmtu;
1266 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1267 mtu = dst_metric_raw(dst, RTAX_MTU);
1272 mtu = READ_ONCE(dst->dev->mtu);
1274 if (unlikely(ip_mtu_locked(dst))) {
1275 if (rt->rt_uses_gateway && mtu > 576)
1279 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1281 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1284 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1286 struct fnhe_hash_bucket *hash;
1287 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1288 u32 hval = fnhe_hashfun(daddr);
1290 spin_lock_bh(&fnhe_lock);
1292 hash = rcu_dereference_protected(nh->nh_exceptions,
1293 lockdep_is_held(&fnhe_lock));
1296 fnhe_p = &hash->chain;
1297 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1299 if (fnhe->fnhe_daddr == daddr) {
1300 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1301 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1302 fnhe_flush_routes(fnhe);
1303 kfree_rcu(fnhe, rcu);
1306 fnhe_p = &fnhe->fnhe_next;
1307 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1308 lockdep_is_held(&fnhe_lock));
1311 spin_unlock_bh(&fnhe_lock);
1314 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1316 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1317 struct fib_nh_exception *fnhe;
1323 hval = fnhe_hashfun(daddr);
1325 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1326 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1327 if (fnhe->fnhe_daddr == daddr) {
1328 if (fnhe->fnhe_expires &&
1329 time_after(jiffies, fnhe->fnhe_expires)) {
1330 ip_del_fnhe(nh, daddr);
1340 * 1. mtu on route is locked - use it
1341 * 2. mtu from nexthop exception
1342 * 3. mtu from egress device
1345 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1347 struct fib_info *fi = res->fi;
1348 struct fib_nh *nh = &fi->fib_nh[res->nh_sel];
1349 struct net_device *dev = nh->nh_dev;
1352 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1353 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1357 struct fib_nh_exception *fnhe;
1359 fnhe = find_exception(nh, daddr);
1360 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1361 mtu = fnhe->fnhe_pmtu;
1365 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1367 return mtu - lwtunnel_headroom(nh->nh_lwtstate, mtu);
1370 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1371 __be32 daddr, const bool do_cache)
1375 spin_lock_bh(&fnhe_lock);
1377 if (daddr == fnhe->fnhe_daddr) {
1378 struct rtable __rcu **porig;
1379 struct rtable *orig;
1380 int genid = fnhe_genid(dev_net(rt->dst.dev));
1382 if (rt_is_input_route(rt))
1383 porig = &fnhe->fnhe_rth_input;
1385 porig = &fnhe->fnhe_rth_output;
1386 orig = rcu_dereference(*porig);
1388 if (fnhe->fnhe_genid != genid) {
1389 fnhe->fnhe_genid = genid;
1391 fnhe->fnhe_pmtu = 0;
1392 fnhe->fnhe_expires = 0;
1393 fnhe->fnhe_mtu_locked = false;
1394 fnhe_flush_routes(fnhe);
1397 fill_route_from_fnhe(rt, fnhe);
1398 if (!rt->rt_gateway)
1399 rt->rt_gateway = daddr;
1403 rcu_assign_pointer(*porig, rt);
1405 dst_dev_put(&orig->dst);
1406 dst_release(&orig->dst);
1411 fnhe->fnhe_stamp = jiffies;
1413 spin_unlock_bh(&fnhe_lock);
1418 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1420 struct rtable *orig, *prev, **p;
1423 if (rt_is_input_route(rt)) {
1424 p = (struct rtable **)&nh->nh_rth_input;
1426 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1430 /* hold dst before doing cmpxchg() to avoid race condition
1434 prev = cmpxchg(p, orig, rt);
1437 dst_dev_put(&orig->dst);
1438 dst_release(&orig->dst);
1441 dst_release(&rt->dst);
1448 struct uncached_list {
1450 struct list_head head;
1453 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1455 void rt_add_uncached_list(struct rtable *rt)
1457 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1459 rt->rt_uncached_list = ul;
1461 spin_lock_bh(&ul->lock);
1462 list_add_tail(&rt->rt_uncached, &ul->head);
1463 spin_unlock_bh(&ul->lock);
1466 void rt_del_uncached_list(struct rtable *rt)
1468 if (!list_empty(&rt->rt_uncached)) {
1469 struct uncached_list *ul = rt->rt_uncached_list;
1471 spin_lock_bh(&ul->lock);
1472 list_del(&rt->rt_uncached);
1473 spin_unlock_bh(&ul->lock);
1477 static void ipv4_dst_destroy(struct dst_entry *dst)
1479 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1480 struct rtable *rt = (struct rtable *)dst;
1482 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1485 rt_del_uncached_list(rt);
1488 void rt_flush_dev(struct net_device *dev)
1490 struct net *net = dev_net(dev);
1494 for_each_possible_cpu(cpu) {
1495 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1497 spin_lock_bh(&ul->lock);
1498 list_for_each_entry(rt, &ul->head, rt_uncached) {
1499 if (rt->dst.dev != dev)
1501 rt->dst.dev = net->loopback_dev;
1502 dev_hold(rt->dst.dev);
1505 spin_unlock_bh(&ul->lock);
1509 static bool rt_cache_valid(const struct rtable *rt)
1512 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1516 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1517 const struct fib_result *res,
1518 struct fib_nh_exception *fnhe,
1519 struct fib_info *fi, u16 type, u32 itag,
1520 const bool do_cache)
1522 bool cached = false;
1525 struct fib_nh *nh = &FIB_RES_NH(*res);
1527 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1528 rt->rt_gateway = nh->nh_gw;
1529 rt->rt_uses_gateway = 1;
1531 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1533 #ifdef CONFIG_IP_ROUTE_CLASSID
1534 rt->dst.tclassid = nh->nh_tclassid;
1536 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1538 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1540 cached = rt_cache_route(nh, rt);
1541 if (unlikely(!cached)) {
1542 /* Routes we intend to cache in nexthop exception or
1543 * FIB nexthop have the DST_NOCACHE bit clear.
1544 * However, if we are unsuccessful at storing this
1545 * route into the cache we really need to set it.
1547 if (!rt->rt_gateway)
1548 rt->rt_gateway = daddr;
1549 rt_add_uncached_list(rt);
1552 rt_add_uncached_list(rt);
1554 #ifdef CONFIG_IP_ROUTE_CLASSID
1555 #ifdef CONFIG_IP_MULTIPLE_TABLES
1556 set_class_tag(rt, res->tclassid);
1558 set_class_tag(rt, itag);
1562 struct rtable *rt_dst_alloc(struct net_device *dev,
1563 unsigned int flags, u16 type,
1564 bool nopolicy, bool noxfrm, bool will_cache)
1568 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1569 (will_cache ? 0 : DST_HOST) |
1570 (nopolicy ? DST_NOPOLICY : 0) |
1571 (noxfrm ? DST_NOXFRM : 0));
1574 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1575 rt->rt_flags = flags;
1577 rt->rt_is_input = 0;
1580 rt->rt_mtu_locked = 0;
1582 rt->rt_uses_gateway = 0;
1583 INIT_LIST_HEAD(&rt->rt_uncached);
1585 rt->dst.output = ip_output;
1586 if (flags & RTCF_LOCAL)
1587 rt->dst.input = ip_local_deliver;
1592 EXPORT_SYMBOL(rt_dst_alloc);
1594 /* called in rcu_read_lock() section */
1595 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1596 u8 tos, struct net_device *dev,
1597 struct in_device *in_dev, u32 *itag)
1601 /* Primary sanity checks. */
1605 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1606 skb->protocol != htons(ETH_P_IP))
1609 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1612 if (ipv4_is_zeronet(saddr)) {
1613 if (!ipv4_is_local_multicast(daddr))
1616 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1624 /* called in rcu_read_lock() section */
1625 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1626 u8 tos, struct net_device *dev, int our)
1628 struct in_device *in_dev = __in_dev_get_rcu(dev);
1629 unsigned int flags = RTCF_MULTICAST;
1634 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1639 flags |= RTCF_LOCAL;
1641 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1642 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1646 #ifdef CONFIG_IP_ROUTE_CLASSID
1647 rth->dst.tclassid = itag;
1649 rth->dst.output = ip_rt_bug;
1650 rth->rt_is_input= 1;
1652 #ifdef CONFIG_IP_MROUTE
1653 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1654 rth->dst.input = ip_mr_input;
1656 RT_CACHE_STAT_INC(in_slow_mc);
1658 skb_dst_set(skb, &rth->dst);
1663 static void ip_handle_martian_source(struct net_device *dev,
1664 struct in_device *in_dev,
1665 struct sk_buff *skb,
1669 RT_CACHE_STAT_INC(in_martian_src);
1670 #ifdef CONFIG_IP_ROUTE_VERBOSE
1671 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1673 * RFC1812 recommendation, if source is martian,
1674 * the only hint is MAC header.
1676 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1677 &daddr, &saddr, dev->name);
1678 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1679 print_hex_dump(KERN_WARNING, "ll header: ",
1680 DUMP_PREFIX_OFFSET, 16, 1,
1681 skb_mac_header(skb),
1682 dev->hard_header_len, true);
1688 /* called in rcu_read_lock() section */
1689 static int __mkroute_input(struct sk_buff *skb,
1690 const struct fib_result *res,
1691 struct in_device *in_dev,
1692 __be32 daddr, __be32 saddr, u32 tos)
1694 struct fib_nh_exception *fnhe;
1697 struct in_device *out_dev;
1701 /* get a working reference to the output device */
1702 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1704 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1708 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1709 in_dev->dev, in_dev, &itag);
1711 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1717 do_cache = res->fi && !itag;
1718 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1719 skb->protocol == htons(ETH_P_IP) &&
1720 (IN_DEV_SHARED_MEDIA(out_dev) ||
1721 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1722 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1724 if (skb->protocol != htons(ETH_P_IP)) {
1725 /* Not IP (i.e. ARP). Do not create route, if it is
1726 * invalid for proxy arp. DNAT routes are always valid.
1728 * Proxy arp feature have been extended to allow, ARP
1729 * replies back to the same interface, to support
1730 * Private VLAN switch technologies. See arp.c.
1732 if (out_dev == in_dev &&
1733 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1739 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1742 rth = rcu_dereference(fnhe->fnhe_rth_input);
1744 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1745 if (rt_cache_valid(rth)) {
1746 skb_dst_set_noref(skb, &rth->dst);
1751 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1752 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1753 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1759 rth->rt_is_input = 1;
1760 RT_CACHE_STAT_INC(in_slow_tot);
1762 rth->dst.input = ip_forward;
1764 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1766 lwtunnel_set_redirect(&rth->dst);
1767 skb_dst_set(skb, &rth->dst);
1774 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1775 /* To make ICMP packets follow the right flow, the multipath hash is
1776 * calculated from the inner IP addresses.
1778 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1779 struct flow_keys *hash_keys)
1781 const struct iphdr *outer_iph = ip_hdr(skb);
1782 const struct iphdr *key_iph = outer_iph;
1783 const struct iphdr *inner_iph;
1784 const struct icmphdr *icmph;
1785 struct iphdr _inner_iph;
1786 struct icmphdr _icmph;
1788 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1791 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1794 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1799 if (icmph->type != ICMP_DEST_UNREACH &&
1800 icmph->type != ICMP_REDIRECT &&
1801 icmph->type != ICMP_TIME_EXCEEDED &&
1802 icmph->type != ICMP_PARAMETERPROB)
1805 inner_iph = skb_header_pointer(skb,
1806 outer_iph->ihl * 4 + sizeof(_icmph),
1807 sizeof(_inner_iph), &_inner_iph);
1811 key_iph = inner_iph;
1813 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1814 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1817 /* if skb is set it will be used and fl4 can be NULL */
1818 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1819 const struct sk_buff *skb, struct flow_keys *flkeys)
1821 struct flow_keys hash_keys;
1824 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1826 memset(&hash_keys, 0, sizeof(hash_keys));
1827 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1829 ip_multipath_l3_keys(skb, &hash_keys);
1831 hash_keys.addrs.v4addrs.src = fl4->saddr;
1832 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1836 /* skb is currently provided only when forwarding */
1838 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1839 struct flow_keys keys;
1841 /* short-circuit if we already have L4 hash present */
1843 return skb_get_hash_raw(skb) >> 1;
1845 memset(&hash_keys, 0, sizeof(hash_keys));
1848 skb_flow_dissect_flow_keys(skb, &keys, flag);
1852 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1853 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1854 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1855 hash_keys.ports.src = flkeys->ports.src;
1856 hash_keys.ports.dst = flkeys->ports.dst;
1857 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1859 memset(&hash_keys, 0, sizeof(hash_keys));
1860 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1861 hash_keys.addrs.v4addrs.src = fl4->saddr;
1862 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1863 hash_keys.ports.src = fl4->fl4_sport;
1864 hash_keys.ports.dst = fl4->fl4_dport;
1865 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1869 mhash = flow_hash_from_keys(&hash_keys);
1873 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1875 static int ip_mkroute_input(struct sk_buff *skb,
1876 struct fib_result *res,
1877 struct in_device *in_dev,
1878 __be32 daddr, __be32 saddr, u32 tos,
1879 struct flow_keys *hkeys)
1881 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1882 if (res->fi && res->fi->fib_nhs > 1) {
1883 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1885 fib_select_multipath(res, h);
1889 /* create a routing cache entry */
1890 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1894 * NOTE. We drop all the packets that has local source
1895 * addresses, because every properly looped back packet
1896 * must have correct destination already attached by output routine.
1898 * Such approach solves two big problems:
1899 * 1. Not simplex devices are handled properly.
1900 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1901 * called with rcu_read_lock()
1904 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1905 u8 tos, struct net_device *dev,
1906 struct fib_result *res)
1908 struct in_device *in_dev = __in_dev_get_rcu(dev);
1909 struct flow_keys *flkeys = NULL, _flkeys;
1910 struct net *net = dev_net(dev);
1911 struct ip_tunnel_info *tun_info;
1913 unsigned int flags = 0;
1919 /* IP on this device is disabled. */
1924 /* Check for the most weird martians, which can be not detected
1928 tun_info = skb_tunnel_info(skb);
1929 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1930 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1932 fl4.flowi4_tun_key.tun_id = 0;
1935 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1936 goto martian_source;
1940 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1943 /* Accept zero addresses only to limited broadcast;
1944 * I even do not know to fix it or not. Waiting for complains :-)
1946 if (ipv4_is_zeronet(saddr))
1947 goto martian_source;
1949 if (ipv4_is_zeronet(daddr))
1950 goto martian_destination;
1952 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1953 * and call it once if daddr or/and saddr are loopback addresses
1955 if (ipv4_is_loopback(daddr)) {
1956 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1957 goto martian_destination;
1958 } else if (ipv4_is_loopback(saddr)) {
1959 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1960 goto martian_source;
1964 * Now we are ready to route packet.
1967 fl4.flowi4_iif = dev->ifindex;
1968 fl4.flowi4_mark = skb->mark;
1969 fl4.flowi4_tos = tos;
1970 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1971 fl4.flowi4_flags = 0;
1974 fl4.flowi4_uid = sock_net_uid(net, NULL);
1976 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
1979 fl4.flowi4_proto = 0;
1984 err = fib_lookup(net, &fl4, res, 0);
1986 if (!IN_DEV_FORWARD(in_dev))
1987 err = -EHOSTUNREACH;
1991 if (res->type == RTN_BROADCAST) {
1992 if (IN_DEV_BFORWARD(in_dev))
1997 if (res->type == RTN_LOCAL) {
1998 err = fib_validate_source(skb, saddr, daddr, tos,
1999 0, dev, in_dev, &itag);
2001 goto martian_source;
2005 if (!IN_DEV_FORWARD(in_dev)) {
2006 err = -EHOSTUNREACH;
2009 if (res->type != RTN_UNICAST)
2010 goto martian_destination;
2013 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2017 if (skb->protocol != htons(ETH_P_IP))
2020 if (!ipv4_is_zeronet(saddr)) {
2021 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2024 goto martian_source;
2026 flags |= RTCF_BROADCAST;
2027 res->type = RTN_BROADCAST;
2028 RT_CACHE_STAT_INC(in_brd);
2034 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2035 if (rt_cache_valid(rth)) {
2036 skb_dst_set_noref(skb, &rth->dst);
2044 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2045 flags | RTCF_LOCAL, res->type,
2046 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2050 rth->dst.output= ip_rt_bug;
2051 #ifdef CONFIG_IP_ROUTE_CLASSID
2052 rth->dst.tclassid = itag;
2054 rth->rt_is_input = 1;
2056 RT_CACHE_STAT_INC(in_slow_tot);
2057 if (res->type == RTN_UNREACHABLE) {
2058 rth->dst.input= ip_error;
2059 rth->dst.error= -err;
2060 rth->rt_flags &= ~RTCF_LOCAL;
2064 struct fib_nh *nh = &FIB_RES_NH(*res);
2066 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2067 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2068 WARN_ON(rth->dst.input == lwtunnel_input);
2069 rth->dst.lwtstate->orig_input = rth->dst.input;
2070 rth->dst.input = lwtunnel_input;
2073 if (unlikely(!rt_cache_route(nh, rth)))
2074 rt_add_uncached_list(rth);
2076 skb_dst_set(skb, &rth->dst);
2081 RT_CACHE_STAT_INC(in_no_route);
2082 res->type = RTN_UNREACHABLE;
2088 * Do not cache martian addresses: they should be logged (RFC1812)
2090 martian_destination:
2091 RT_CACHE_STAT_INC(in_martian_dst);
2092 #ifdef CONFIG_IP_ROUTE_VERBOSE
2093 if (IN_DEV_LOG_MARTIANS(in_dev))
2094 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2095 &daddr, &saddr, dev->name);
2107 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2111 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2112 u8 tos, struct net_device *dev)
2114 struct fib_result res;
2117 tos &= IPTOS_RT_MASK;
2119 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2124 EXPORT_SYMBOL(ip_route_input_noref);
2126 /* called with rcu_read_lock held */
2127 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2128 u8 tos, struct net_device *dev, struct fib_result *res)
2130 /* Multicast recognition logic is moved from route cache to here.
2131 The problem was that too many Ethernet cards have broken/missing
2132 hardware multicast filters :-( As result the host on multicasting
2133 network acquires a lot of useless route cache entries, sort of
2134 SDR messages from all the world. Now we try to get rid of them.
2135 Really, provided software IP multicast filter is organized
2136 reasonably (at least, hashed), it does not result in a slowdown
2137 comparing with route cache reject entries.
2138 Note, that multicast routers are not affected, because
2139 route cache entry is created eventually.
2141 if (ipv4_is_multicast(daddr)) {
2142 struct in_device *in_dev = __in_dev_get_rcu(dev);
2147 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2148 ip_hdr(skb)->protocol);
2150 /* check l3 master if no match yet */
2151 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2152 struct in_device *l3_in_dev;
2154 l3_in_dev = __in_dev_get_rcu(skb->dev);
2156 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2157 ip_hdr(skb)->protocol);
2161 #ifdef CONFIG_IP_MROUTE
2163 (!ipv4_is_local_multicast(daddr) &&
2164 IN_DEV_MFORWARD(in_dev))
2167 err = ip_route_input_mc(skb, daddr, saddr,
2173 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2176 /* called with rcu_read_lock() */
2177 static struct rtable *__mkroute_output(const struct fib_result *res,
2178 const struct flowi4 *fl4, int orig_oif,
2179 struct net_device *dev_out,
2182 struct fib_info *fi = res->fi;
2183 struct fib_nh_exception *fnhe;
2184 struct in_device *in_dev;
2185 u16 type = res->type;
2189 in_dev = __in_dev_get_rcu(dev_out);
2191 return ERR_PTR(-EINVAL);
2193 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2194 if (ipv4_is_loopback(fl4->saddr) &&
2195 !(dev_out->flags & IFF_LOOPBACK) &&
2196 !netif_is_l3_master(dev_out))
2197 return ERR_PTR(-EINVAL);
2199 if (ipv4_is_lbcast(fl4->daddr))
2200 type = RTN_BROADCAST;
2201 else if (ipv4_is_multicast(fl4->daddr))
2202 type = RTN_MULTICAST;
2203 else if (ipv4_is_zeronet(fl4->daddr))
2204 return ERR_PTR(-EINVAL);
2206 if (dev_out->flags & IFF_LOOPBACK)
2207 flags |= RTCF_LOCAL;
2210 if (type == RTN_BROADCAST) {
2211 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2213 } else if (type == RTN_MULTICAST) {
2214 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2215 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2217 flags &= ~RTCF_LOCAL;
2220 /* If multicast route do not exist use
2221 * default one, but do not gateway in this case.
2224 if (fi && res->prefixlen < 4)
2226 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2227 (orig_oif != dev_out->ifindex)) {
2228 /* For local routes that require a particular output interface
2229 * we do not want to cache the result. Caching the result
2230 * causes incorrect behaviour when there are multiple source
2231 * addresses on the interface, the end result being that if the
2232 * intended recipient is waiting on that interface for the
2233 * packet he won't receive it because it will be delivered on
2234 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2235 * be set to the loopback interface as well.
2241 do_cache &= fi != NULL;
2243 struct rtable __rcu **prth;
2244 struct fib_nh *nh = &FIB_RES_NH(*res);
2246 fnhe = find_exception(nh, fl4->daddr);
2250 prth = &fnhe->fnhe_rth_output;
2252 if (unlikely(fl4->flowi4_flags &
2253 FLOWI_FLAG_KNOWN_NH &&
2255 nh->nh_scope == RT_SCOPE_LINK))) {
2259 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2261 rth = rcu_dereference(*prth);
2262 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2267 rth = rt_dst_alloc(dev_out, flags, type,
2268 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2269 IN_DEV_CONF_GET(in_dev, NOXFRM),
2272 return ERR_PTR(-ENOBUFS);
2274 rth->rt_iif = orig_oif;
2276 RT_CACHE_STAT_INC(out_slow_tot);
2278 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2279 if (flags & RTCF_LOCAL &&
2280 !(dev_out->flags & IFF_LOOPBACK)) {
2281 rth->dst.output = ip_mc_output;
2282 RT_CACHE_STAT_INC(out_slow_mc);
2284 #ifdef CONFIG_IP_MROUTE
2285 if (type == RTN_MULTICAST) {
2286 if (IN_DEV_MFORWARD(in_dev) &&
2287 !ipv4_is_local_multicast(fl4->daddr)) {
2288 rth->dst.input = ip_mr_input;
2289 rth->dst.output = ip_mc_output;
2295 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2296 lwtunnel_set_redirect(&rth->dst);
2302 * Major route resolver routine.
2305 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2306 const struct sk_buff *skb)
2308 __u8 tos = RT_FL_TOS(fl4);
2309 struct fib_result res = {
2317 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2318 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2319 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2320 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2323 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2328 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2330 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2331 struct fib_result *res,
2332 const struct sk_buff *skb)
2334 struct net_device *dev_out = NULL;
2335 int orig_oif = fl4->flowi4_oif;
2336 unsigned int flags = 0;
2338 int err = -ENETUNREACH;
2341 rth = ERR_PTR(-EINVAL);
2342 if (ipv4_is_multicast(fl4->saddr) ||
2343 ipv4_is_lbcast(fl4->saddr) ||
2344 ipv4_is_zeronet(fl4->saddr))
2347 /* I removed check for oif == dev_out->oif here.
2348 It was wrong for two reasons:
2349 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2350 is assigned to multiple interfaces.
2351 2. Moreover, we are allowed to send packets with saddr
2352 of another iface. --ANK
2355 if (fl4->flowi4_oif == 0 &&
2356 (ipv4_is_multicast(fl4->daddr) ||
2357 ipv4_is_lbcast(fl4->daddr))) {
2358 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2359 dev_out = __ip_dev_find(net, fl4->saddr, false);
2363 /* Special hack: user can direct multicasts
2364 and limited broadcast via necessary interface
2365 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2366 This hack is not just for fun, it allows
2367 vic,vat and friends to work.
2368 They bind socket to loopback, set ttl to zero
2369 and expect that it will work.
2370 From the viewpoint of routing cache they are broken,
2371 because we are not allowed to build multicast path
2372 with loopback source addr (look, routing cache
2373 cannot know, that ttl is zero, so that packet
2374 will not leave this host and route is valid).
2375 Luckily, this hack is good workaround.
2378 fl4->flowi4_oif = dev_out->ifindex;
2382 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2383 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2384 if (!__ip_dev_find(net, fl4->saddr, false))
2390 if (fl4->flowi4_oif) {
2391 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2392 rth = ERR_PTR(-ENODEV);
2396 /* RACE: Check return value of inet_select_addr instead. */
2397 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2398 rth = ERR_PTR(-ENETUNREACH);
2401 if (ipv4_is_local_multicast(fl4->daddr) ||
2402 ipv4_is_lbcast(fl4->daddr) ||
2403 fl4->flowi4_proto == IPPROTO_IGMP) {
2405 fl4->saddr = inet_select_addr(dev_out, 0,
2410 if (ipv4_is_multicast(fl4->daddr))
2411 fl4->saddr = inet_select_addr(dev_out, 0,
2413 else if (!fl4->daddr)
2414 fl4->saddr = inet_select_addr(dev_out, 0,
2420 fl4->daddr = fl4->saddr;
2422 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2423 dev_out = net->loopback_dev;
2424 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2425 res->type = RTN_LOCAL;
2426 flags |= RTCF_LOCAL;
2430 err = fib_lookup(net, fl4, res, 0);
2434 if (fl4->flowi4_oif &&
2435 (ipv4_is_multicast(fl4->daddr) ||
2436 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2437 /* Apparently, routing tables are wrong. Assume,
2438 that the destination is on link.
2441 Because we are allowed to send to iface
2442 even if it has NO routes and NO assigned
2443 addresses. When oif is specified, routing
2444 tables are looked up with only one purpose:
2445 to catch if destination is gatewayed, rather than
2446 direct. Moreover, if MSG_DONTROUTE is set,
2447 we send packet, ignoring both routing tables
2448 and ifaddr state. --ANK
2451 We could make it even if oif is unknown,
2452 likely IPv6, but we do not.
2455 if (fl4->saddr == 0)
2456 fl4->saddr = inet_select_addr(dev_out, 0,
2458 res->type = RTN_UNICAST;
2465 if (res->type == RTN_LOCAL) {
2467 if (res->fi->fib_prefsrc)
2468 fl4->saddr = res->fi->fib_prefsrc;
2470 fl4->saddr = fl4->daddr;
2473 /* L3 master device is the loopback for that domain */
2474 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2477 /* make sure orig_oif points to fib result device even
2478 * though packet rx/tx happens over loopback or l3mdev
2480 orig_oif = FIB_RES_OIF(*res);
2482 fl4->flowi4_oif = dev_out->ifindex;
2483 flags |= RTCF_LOCAL;
2487 fib_select_path(net, res, fl4, skb);
2489 dev_out = FIB_RES_DEV(*res);
2490 fl4->flowi4_oif = dev_out->ifindex;
2494 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2500 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2505 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2507 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2509 return mtu ? : dst->dev->mtu;
2512 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2513 struct sk_buff *skb, u32 mtu)
2517 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2518 struct sk_buff *skb)
2522 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2528 static struct dst_ops ipv4_dst_blackhole_ops = {
2530 .check = ipv4_blackhole_dst_check,
2531 .mtu = ipv4_blackhole_mtu,
2532 .default_advmss = ipv4_default_advmss,
2533 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2534 .redirect = ipv4_rt_blackhole_redirect,
2535 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2536 .neigh_lookup = ipv4_neigh_lookup,
2539 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2541 struct rtable *ort = (struct rtable *) dst_orig;
2544 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2546 struct dst_entry *new = &rt->dst;
2549 new->input = dst_discard;
2550 new->output = dst_discard_out;
2552 new->dev = net->loopback_dev;
2556 rt->rt_is_input = ort->rt_is_input;
2557 rt->rt_iif = ort->rt_iif;
2558 rt->rt_pmtu = ort->rt_pmtu;
2559 rt->rt_mtu_locked = ort->rt_mtu_locked;
2561 rt->rt_genid = rt_genid_ipv4(net);
2562 rt->rt_flags = ort->rt_flags;
2563 rt->rt_type = ort->rt_type;
2564 rt->rt_gateway = ort->rt_gateway;
2565 rt->rt_uses_gateway = ort->rt_uses_gateway;
2567 INIT_LIST_HEAD(&rt->rt_uncached);
2570 dst_release(dst_orig);
2572 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2575 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2576 const struct sock *sk)
2578 struct rtable *rt = __ip_route_output_key(net, flp4);
2583 if (flp4->flowi4_proto)
2584 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2585 flowi4_to_flowi(flp4),
2590 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2592 /* called with rcu_read_lock held */
2593 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2594 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2595 struct sk_buff *skb, u32 portid, u32 seq)
2598 struct nlmsghdr *nlh;
2599 unsigned long expires = 0;
2601 u32 metrics[RTAX_MAX];
2603 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2607 r = nlmsg_data(nlh);
2608 r->rtm_family = AF_INET;
2609 r->rtm_dst_len = 32;
2611 r->rtm_tos = fl4->flowi4_tos;
2612 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2613 if (nla_put_u32(skb, RTA_TABLE, table_id))
2614 goto nla_put_failure;
2615 r->rtm_type = rt->rt_type;
2616 r->rtm_scope = RT_SCOPE_UNIVERSE;
2617 r->rtm_protocol = RTPROT_UNSPEC;
2618 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2619 if (rt->rt_flags & RTCF_NOTIFY)
2620 r->rtm_flags |= RTM_F_NOTIFY;
2621 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2622 r->rtm_flags |= RTCF_DOREDIRECT;
2624 if (nla_put_in_addr(skb, RTA_DST, dst))
2625 goto nla_put_failure;
2627 r->rtm_src_len = 32;
2628 if (nla_put_in_addr(skb, RTA_SRC, src))
2629 goto nla_put_failure;
2632 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2633 goto nla_put_failure;
2634 #ifdef CONFIG_IP_ROUTE_CLASSID
2635 if (rt->dst.tclassid &&
2636 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2637 goto nla_put_failure;
2639 if (!rt_is_input_route(rt) &&
2640 fl4->saddr != src) {
2641 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2642 goto nla_put_failure;
2644 if (rt->rt_uses_gateway &&
2645 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2646 goto nla_put_failure;
2648 expires = rt->dst.expires;
2650 unsigned long now = jiffies;
2652 if (time_before(now, expires))
2658 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2659 if (rt->rt_pmtu && expires)
2660 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2661 if (rt->rt_mtu_locked && expires)
2662 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2663 if (rtnetlink_put_metrics(skb, metrics) < 0)
2664 goto nla_put_failure;
2666 if (fl4->flowi4_mark &&
2667 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2668 goto nla_put_failure;
2670 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2671 nla_put_u32(skb, RTA_UID,
2672 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2673 goto nla_put_failure;
2675 error = rt->dst.error;
2677 if (rt_is_input_route(rt)) {
2678 #ifdef CONFIG_IP_MROUTE
2679 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2680 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2681 int err = ipmr_get_route(net, skb,
2682 fl4->saddr, fl4->daddr,
2688 goto nla_put_failure;
2692 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2693 goto nla_put_failure;
2696 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2697 goto nla_put_failure;
2699 nlmsg_end(skb, nlh);
2703 nlmsg_cancel(skb, nlh);
2707 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2708 u8 ip_proto, __be16 sport,
2711 struct sk_buff *skb;
2714 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2718 /* Reserve room for dummy headers, this skb can pass
2719 * through good chunk of routing engine.
2721 skb_reset_mac_header(skb);
2722 skb_reset_network_header(skb);
2723 skb->protocol = htons(ETH_P_IP);
2724 iph = skb_put(skb, sizeof(struct iphdr));
2725 iph->protocol = ip_proto;
2731 skb_set_transport_header(skb, skb->len);
2733 switch (iph->protocol) {
2735 struct udphdr *udph;
2737 udph = skb_put_zero(skb, sizeof(struct udphdr));
2738 udph->source = sport;
2740 udph->len = sizeof(struct udphdr);
2745 struct tcphdr *tcph;
2747 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2748 tcph->source = sport;
2750 tcph->doff = sizeof(struct tcphdr) / 4;
2752 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2756 case IPPROTO_ICMP: {
2757 struct icmphdr *icmph;
2759 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2760 icmph->type = ICMP_ECHO;
2768 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2769 struct netlink_ext_ack *extack)
2771 struct net *net = sock_net(in_skb->sk);
2772 struct nlattr *tb[RTA_MAX+1];
2773 u32 table_id = RT_TABLE_MAIN;
2774 __be16 sport = 0, dport = 0;
2775 struct fib_result res = {};
2776 u8 ip_proto = IPPROTO_UDP;
2777 struct rtable *rt = NULL;
2778 struct sk_buff *skb;
2780 struct flowi4 fl4 = {};
2788 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2793 rtm = nlmsg_data(nlh);
2794 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2795 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2796 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2797 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2799 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2801 uid = (iif ? INVALID_UID : current_uid());
2803 if (tb[RTA_IP_PROTO]) {
2804 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
2811 sport = nla_get_be16(tb[RTA_SPORT]);
2814 dport = nla_get_be16(tb[RTA_DPORT]);
2816 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
2822 fl4.flowi4_tos = rtm->rtm_tos;
2823 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2824 fl4.flowi4_mark = mark;
2825 fl4.flowi4_uid = uid;
2827 fl4.fl4_sport = sport;
2829 fl4.fl4_dport = dport;
2830 fl4.flowi4_proto = ip_proto;
2835 struct net_device *dev;
2837 dev = dev_get_by_index_rcu(net, iif);
2843 fl4.flowi4_iif = iif; /* for rt_fill_info */
2846 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2849 rt = skb_rtable(skb);
2850 if (err == 0 && rt->dst.error)
2851 err = -rt->dst.error;
2853 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2854 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2859 skb_dst_set(skb, &rt->dst);
2865 if (rtm->rtm_flags & RTM_F_NOTIFY)
2866 rt->rt_flags |= RTCF_NOTIFY;
2868 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2869 table_id = res.table ? res.table->tb_id : 0;
2871 /* reset skb for netlink reply msg */
2873 skb_reset_network_header(skb);
2874 skb_reset_transport_header(skb);
2875 skb_reset_mac_header(skb);
2877 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2879 err = fib_props[res.type].error;
2881 err = -EHOSTUNREACH;
2884 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2885 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2886 rt->rt_type, res.prefix, res.prefixlen,
2887 fl4.flowi4_tos, res.fi, 0);
2889 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
2890 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2897 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2907 void ip_rt_multicast_event(struct in_device *in_dev)
2909 rt_cache_flush(dev_net(in_dev->dev));
2912 #ifdef CONFIG_SYSCTL
2913 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2914 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2915 static int ip_rt_gc_elasticity __read_mostly = 8;
2916 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
2918 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2919 void __user *buffer,
2920 size_t *lenp, loff_t *ppos)
2922 struct net *net = (struct net *)__ctl->extra1;
2925 rt_cache_flush(net);
2926 fnhe_genid_bump(net);
2933 static struct ctl_table ipv4_route_table[] = {
2935 .procname = "gc_thresh",
2936 .data = &ipv4_dst_ops.gc_thresh,
2937 .maxlen = sizeof(int),
2939 .proc_handler = proc_dointvec,
2942 .procname = "max_size",
2943 .data = &ip_rt_max_size,
2944 .maxlen = sizeof(int),
2946 .proc_handler = proc_dointvec,
2949 /* Deprecated. Use gc_min_interval_ms */
2951 .procname = "gc_min_interval",
2952 .data = &ip_rt_gc_min_interval,
2953 .maxlen = sizeof(int),
2955 .proc_handler = proc_dointvec_jiffies,
2958 .procname = "gc_min_interval_ms",
2959 .data = &ip_rt_gc_min_interval,
2960 .maxlen = sizeof(int),
2962 .proc_handler = proc_dointvec_ms_jiffies,
2965 .procname = "gc_timeout",
2966 .data = &ip_rt_gc_timeout,
2967 .maxlen = sizeof(int),
2969 .proc_handler = proc_dointvec_jiffies,
2972 .procname = "gc_interval",
2973 .data = &ip_rt_gc_interval,
2974 .maxlen = sizeof(int),
2976 .proc_handler = proc_dointvec_jiffies,
2979 .procname = "redirect_load",
2980 .data = &ip_rt_redirect_load,
2981 .maxlen = sizeof(int),
2983 .proc_handler = proc_dointvec,
2986 .procname = "redirect_number",
2987 .data = &ip_rt_redirect_number,
2988 .maxlen = sizeof(int),
2990 .proc_handler = proc_dointvec,
2993 .procname = "redirect_silence",
2994 .data = &ip_rt_redirect_silence,
2995 .maxlen = sizeof(int),
2997 .proc_handler = proc_dointvec,
3000 .procname = "error_cost",
3001 .data = &ip_rt_error_cost,
3002 .maxlen = sizeof(int),
3004 .proc_handler = proc_dointvec,
3007 .procname = "error_burst",
3008 .data = &ip_rt_error_burst,
3009 .maxlen = sizeof(int),
3011 .proc_handler = proc_dointvec,
3014 .procname = "gc_elasticity",
3015 .data = &ip_rt_gc_elasticity,
3016 .maxlen = sizeof(int),
3018 .proc_handler = proc_dointvec,
3021 .procname = "mtu_expires",
3022 .data = &ip_rt_mtu_expires,
3023 .maxlen = sizeof(int),
3025 .proc_handler = proc_dointvec_jiffies,
3028 .procname = "min_pmtu",
3029 .data = &ip_rt_min_pmtu,
3030 .maxlen = sizeof(int),
3032 .proc_handler = proc_dointvec_minmax,
3033 .extra1 = &ip_min_valid_pmtu,
3036 .procname = "min_adv_mss",
3037 .data = &ip_rt_min_advmss,
3038 .maxlen = sizeof(int),
3040 .proc_handler = proc_dointvec,
3045 static struct ctl_table ipv4_route_flush_table[] = {
3047 .procname = "flush",
3048 .maxlen = sizeof(int),
3050 .proc_handler = ipv4_sysctl_rtcache_flush,
3055 static __net_init int sysctl_route_net_init(struct net *net)
3057 struct ctl_table *tbl;
3059 tbl = ipv4_route_flush_table;
3060 if (!net_eq(net, &init_net)) {
3061 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3065 /* Don't export sysctls to unprivileged users */
3066 if (net->user_ns != &init_user_ns)
3067 tbl[0].procname = NULL;
3069 tbl[0].extra1 = net;
3071 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3072 if (!net->ipv4.route_hdr)
3077 if (tbl != ipv4_route_flush_table)
3083 static __net_exit void sysctl_route_net_exit(struct net *net)
3085 struct ctl_table *tbl;
3087 tbl = net->ipv4.route_hdr->ctl_table_arg;
3088 unregister_net_sysctl_table(net->ipv4.route_hdr);
3089 BUG_ON(tbl == ipv4_route_flush_table);
3093 static __net_initdata struct pernet_operations sysctl_route_ops = {
3094 .init = sysctl_route_net_init,
3095 .exit = sysctl_route_net_exit,
3099 static __net_init int rt_genid_init(struct net *net)
3101 atomic_set(&net->ipv4.rt_genid, 0);
3102 atomic_set(&net->fnhe_genid, 0);
3103 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3107 static __net_initdata struct pernet_operations rt_genid_ops = {
3108 .init = rt_genid_init,
3111 static int __net_init ipv4_inetpeer_init(struct net *net)
3113 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3117 inet_peer_base_init(bp);
3118 net->ipv4.peers = bp;
3122 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3124 struct inet_peer_base *bp = net->ipv4.peers;
3126 net->ipv4.peers = NULL;
3127 inetpeer_invalidate_tree(bp);
3131 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3132 .init = ipv4_inetpeer_init,
3133 .exit = ipv4_inetpeer_exit,
3136 #ifdef CONFIG_IP_ROUTE_CLASSID
3137 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3138 #endif /* CONFIG_IP_ROUTE_CLASSID */
3140 int __init ip_rt_init(void)
3144 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3147 panic("IP: failed to allocate ip_idents\n");
3149 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3151 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3153 panic("IP: failed to allocate ip_tstamps\n");
3155 for_each_possible_cpu(cpu) {
3156 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3158 INIT_LIST_HEAD(&ul->head);
3159 spin_lock_init(&ul->lock);
3161 #ifdef CONFIG_IP_ROUTE_CLASSID
3162 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3164 panic("IP: failed to allocate ip_rt_acct\n");
3167 ipv4_dst_ops.kmem_cachep =
3168 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3169 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3171 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3173 if (dst_entries_init(&ipv4_dst_ops) < 0)
3174 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3176 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3177 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3179 ipv4_dst_ops.gc_thresh = ~0;
3180 ip_rt_max_size = INT_MAX;
3185 if (ip_rt_proc_init())
3186 pr_err("Unable to create route proc files\n");
3191 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3192 RTNL_FLAG_DOIT_UNLOCKED);
3194 #ifdef CONFIG_SYSCTL
3195 register_pernet_subsys(&sysctl_route_ops);
3197 register_pernet_subsys(&rt_genid_ops);
3198 register_pernet_subsys(&ipv4_inetpeer_ops);
3202 #ifdef CONFIG_SYSCTL
3204 * We really need to sanitize the damn ipv4 init order, then all
3205 * this nonsense will go away.
3207 void __init ip_static_sysctl_init(void)
3209 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);