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 = kzalloc(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, u32 mark, u8 protocol, int flow_flags)
1045 const struct iphdr *iph = (const struct iphdr *) skb->data;
1050 mark = IP4_REPLY_MARK(net, skb->mark);
1052 __build_flow_key(net, &fl4, NULL, iph, oif,
1053 RT_TOS(iph->tos), protocol, mark, flow_flags);
1054 rt = __ip_route_output_key(net, &fl4);
1056 __ip_rt_update_pmtu(rt, &fl4, mtu);
1060 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1062 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1064 const struct iphdr *iph = (const struct iphdr *) skb->data;
1068 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1070 if (!fl4.flowi4_mark)
1071 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1073 rt = __ip_route_output_key(sock_net(sk), &fl4);
1075 __ip_rt_update_pmtu(rt, &fl4, mtu);
1080 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1082 const struct iphdr *iph = (const struct iphdr *) skb->data;
1085 struct dst_entry *odst = NULL;
1087 struct net *net = sock_net(sk);
1091 if (!ip_sk_accept_pmtu(sk))
1094 odst = sk_dst_get(sk);
1096 if (sock_owned_by_user(sk) || !odst) {
1097 __ipv4_sk_update_pmtu(skb, sk, mtu);
1101 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1103 rt = (struct rtable *)odst;
1104 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1105 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1112 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1114 if (!dst_check(&rt->dst, 0)) {
1116 dst_release(&rt->dst);
1118 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1126 sk_dst_set(sk, &rt->dst);
1132 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1134 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1135 int oif, u32 mark, u8 protocol, int flow_flags)
1137 const struct iphdr *iph = (const struct iphdr *) skb->data;
1141 __build_flow_key(net, &fl4, NULL, iph, oif,
1142 RT_TOS(iph->tos), protocol, mark, flow_flags);
1143 rt = __ip_route_output_key(net, &fl4);
1145 __ip_do_redirect(rt, skb, &fl4, false);
1149 EXPORT_SYMBOL_GPL(ipv4_redirect);
1151 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1153 const struct iphdr *iph = (const struct iphdr *) skb->data;
1156 struct net *net = sock_net(sk);
1158 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1159 rt = __ip_route_output_key(net, &fl4);
1161 __ip_do_redirect(rt, skb, &fl4, false);
1165 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1167 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1169 struct rtable *rt = (struct rtable *) dst;
1171 /* All IPV4 dsts are created with ->obsolete set to the value
1172 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1173 * into this function always.
1175 * When a PMTU/redirect information update invalidates a route,
1176 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1177 * DST_OBSOLETE_DEAD by dst_free().
1179 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1184 static void ipv4_link_failure(struct sk_buff *skb)
1188 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1190 rt = skb_rtable(skb);
1192 dst_set_expires(&rt->dst, 0);
1195 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1197 pr_debug("%s: %pI4 -> %pI4, %s\n",
1198 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1199 skb->dev ? skb->dev->name : "?");
1206 We do not cache source address of outgoing interface,
1207 because it is used only by IP RR, TS and SRR options,
1208 so that it out of fast path.
1210 BTW remember: "addr" is allowed to be not aligned
1214 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1218 if (rt_is_output_route(rt))
1219 src = ip_hdr(skb)->saddr;
1221 struct fib_result res;
1227 memset(&fl4, 0, sizeof(fl4));
1228 fl4.daddr = iph->daddr;
1229 fl4.saddr = iph->saddr;
1230 fl4.flowi4_tos = RT_TOS(iph->tos);
1231 fl4.flowi4_oif = rt->dst.dev->ifindex;
1232 fl4.flowi4_iif = skb->dev->ifindex;
1233 fl4.flowi4_mark = skb->mark;
1236 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1237 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1239 src = inet_select_addr(rt->dst.dev,
1240 rt_nexthop(rt, iph->daddr),
1244 memcpy(addr, &src, 4);
1247 #ifdef CONFIG_IP_ROUTE_CLASSID
1248 static void set_class_tag(struct rtable *rt, u32 tag)
1250 if (!(rt->dst.tclassid & 0xFFFF))
1251 rt->dst.tclassid |= tag & 0xFFFF;
1252 if (!(rt->dst.tclassid & 0xFFFF0000))
1253 rt->dst.tclassid |= tag & 0xFFFF0000;
1257 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1259 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1260 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1263 return min(advmss, IPV4_MAX_PMTU - header_size);
1266 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1268 const struct rtable *rt = (const struct rtable *) dst;
1269 unsigned int mtu = rt->rt_pmtu;
1271 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1272 mtu = dst_metric_raw(dst, RTAX_MTU);
1277 mtu = READ_ONCE(dst->dev->mtu);
1279 if (unlikely(ip_mtu_locked(dst))) {
1280 if (rt->rt_uses_gateway && mtu > 576)
1284 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1286 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1289 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1291 struct fnhe_hash_bucket *hash;
1292 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1293 u32 hval = fnhe_hashfun(daddr);
1295 spin_lock_bh(&fnhe_lock);
1297 hash = rcu_dereference_protected(nh->nh_exceptions,
1298 lockdep_is_held(&fnhe_lock));
1301 fnhe_p = &hash->chain;
1302 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1304 if (fnhe->fnhe_daddr == daddr) {
1305 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1306 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1307 fnhe_flush_routes(fnhe);
1308 kfree_rcu(fnhe, rcu);
1311 fnhe_p = &fnhe->fnhe_next;
1312 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1313 lockdep_is_held(&fnhe_lock));
1316 spin_unlock_bh(&fnhe_lock);
1319 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1321 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1322 struct fib_nh_exception *fnhe;
1328 hval = fnhe_hashfun(daddr);
1330 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1331 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1332 if (fnhe->fnhe_daddr == daddr) {
1333 if (fnhe->fnhe_expires &&
1334 time_after(jiffies, fnhe->fnhe_expires)) {
1335 ip_del_fnhe(nh, daddr);
1344 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1345 __be32 daddr, const bool do_cache)
1349 spin_lock_bh(&fnhe_lock);
1351 if (daddr == fnhe->fnhe_daddr) {
1352 struct rtable __rcu **porig;
1353 struct rtable *orig;
1354 int genid = fnhe_genid(dev_net(rt->dst.dev));
1356 if (rt_is_input_route(rt))
1357 porig = &fnhe->fnhe_rth_input;
1359 porig = &fnhe->fnhe_rth_output;
1360 orig = rcu_dereference(*porig);
1362 if (fnhe->fnhe_genid != genid) {
1363 fnhe->fnhe_genid = genid;
1365 fnhe->fnhe_pmtu = 0;
1366 fnhe->fnhe_expires = 0;
1367 fnhe->fnhe_mtu_locked = false;
1368 fnhe_flush_routes(fnhe);
1371 fill_route_from_fnhe(rt, fnhe);
1372 if (!rt->rt_gateway)
1373 rt->rt_gateway = daddr;
1377 rcu_assign_pointer(*porig, rt);
1379 dst_dev_put(&orig->dst);
1380 dst_release(&orig->dst);
1385 fnhe->fnhe_stamp = jiffies;
1387 spin_unlock_bh(&fnhe_lock);
1392 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1394 struct rtable *orig, *prev, **p;
1397 if (rt_is_input_route(rt)) {
1398 p = (struct rtable **)&nh->nh_rth_input;
1400 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1404 /* hold dst before doing cmpxchg() to avoid race condition
1408 prev = cmpxchg(p, orig, rt);
1411 dst_dev_put(&orig->dst);
1412 dst_release(&orig->dst);
1415 dst_release(&rt->dst);
1422 struct uncached_list {
1424 struct list_head head;
1427 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1429 void rt_add_uncached_list(struct rtable *rt)
1431 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1433 rt->rt_uncached_list = ul;
1435 spin_lock_bh(&ul->lock);
1436 list_add_tail(&rt->rt_uncached, &ul->head);
1437 spin_unlock_bh(&ul->lock);
1440 void rt_del_uncached_list(struct rtable *rt)
1442 if (!list_empty(&rt->rt_uncached)) {
1443 struct uncached_list *ul = rt->rt_uncached_list;
1445 spin_lock_bh(&ul->lock);
1446 list_del(&rt->rt_uncached);
1447 spin_unlock_bh(&ul->lock);
1451 static void ipv4_dst_destroy(struct dst_entry *dst)
1453 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1454 struct rtable *rt = (struct rtable *)dst;
1456 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1459 rt_del_uncached_list(rt);
1462 void rt_flush_dev(struct net_device *dev)
1464 struct net *net = dev_net(dev);
1468 for_each_possible_cpu(cpu) {
1469 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1471 spin_lock_bh(&ul->lock);
1472 list_for_each_entry(rt, &ul->head, rt_uncached) {
1473 if (rt->dst.dev != dev)
1475 rt->dst.dev = net->loopback_dev;
1476 dev_hold(rt->dst.dev);
1479 spin_unlock_bh(&ul->lock);
1483 static bool rt_cache_valid(const struct rtable *rt)
1486 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1490 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1491 const struct fib_result *res,
1492 struct fib_nh_exception *fnhe,
1493 struct fib_info *fi, u16 type, u32 itag,
1494 const bool do_cache)
1496 bool cached = false;
1499 struct fib_nh *nh = &FIB_RES_NH(*res);
1501 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1502 rt->rt_gateway = nh->nh_gw;
1503 rt->rt_uses_gateway = 1;
1505 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1506 if (fi->fib_metrics != &dst_default_metrics) {
1507 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1508 refcount_inc(&fi->fib_metrics->refcnt);
1510 #ifdef CONFIG_IP_ROUTE_CLASSID
1511 rt->dst.tclassid = nh->nh_tclassid;
1513 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1515 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1517 cached = rt_cache_route(nh, rt);
1518 if (unlikely(!cached)) {
1519 /* Routes we intend to cache in nexthop exception or
1520 * FIB nexthop have the DST_NOCACHE bit clear.
1521 * However, if we are unsuccessful at storing this
1522 * route into the cache we really need to set it.
1524 if (!rt->rt_gateway)
1525 rt->rt_gateway = daddr;
1526 rt_add_uncached_list(rt);
1529 rt_add_uncached_list(rt);
1531 #ifdef CONFIG_IP_ROUTE_CLASSID
1532 #ifdef CONFIG_IP_MULTIPLE_TABLES
1533 set_class_tag(rt, res->tclassid);
1535 set_class_tag(rt, itag);
1539 struct rtable *rt_dst_alloc(struct net_device *dev,
1540 unsigned int flags, u16 type,
1541 bool nopolicy, bool noxfrm, bool will_cache)
1545 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1546 (will_cache ? 0 : DST_HOST) |
1547 (nopolicy ? DST_NOPOLICY : 0) |
1548 (noxfrm ? DST_NOXFRM : 0));
1551 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1552 rt->rt_flags = flags;
1554 rt->rt_is_input = 0;
1557 rt->rt_mtu_locked = 0;
1559 rt->rt_uses_gateway = 0;
1560 INIT_LIST_HEAD(&rt->rt_uncached);
1562 rt->dst.output = ip_output;
1563 if (flags & RTCF_LOCAL)
1564 rt->dst.input = ip_local_deliver;
1569 EXPORT_SYMBOL(rt_dst_alloc);
1571 /* called in rcu_read_lock() section */
1572 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1573 u8 tos, struct net_device *dev,
1574 struct in_device *in_dev, u32 *itag)
1578 /* Primary sanity checks. */
1582 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1583 skb->protocol != htons(ETH_P_IP))
1586 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1589 if (ipv4_is_zeronet(saddr)) {
1590 if (!ipv4_is_local_multicast(daddr))
1593 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1601 /* called in rcu_read_lock() section */
1602 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1603 u8 tos, struct net_device *dev, int our)
1605 struct in_device *in_dev = __in_dev_get_rcu(dev);
1606 unsigned int flags = RTCF_MULTICAST;
1611 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1616 flags |= RTCF_LOCAL;
1618 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1619 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1623 #ifdef CONFIG_IP_ROUTE_CLASSID
1624 rth->dst.tclassid = itag;
1626 rth->dst.output = ip_rt_bug;
1627 rth->rt_is_input= 1;
1629 #ifdef CONFIG_IP_MROUTE
1630 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1631 rth->dst.input = ip_mr_input;
1633 RT_CACHE_STAT_INC(in_slow_mc);
1635 skb_dst_set(skb, &rth->dst);
1640 static void ip_handle_martian_source(struct net_device *dev,
1641 struct in_device *in_dev,
1642 struct sk_buff *skb,
1646 RT_CACHE_STAT_INC(in_martian_src);
1647 #ifdef CONFIG_IP_ROUTE_VERBOSE
1648 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1650 * RFC1812 recommendation, if source is martian,
1651 * the only hint is MAC header.
1653 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1654 &daddr, &saddr, dev->name);
1655 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1656 print_hex_dump(KERN_WARNING, "ll header: ",
1657 DUMP_PREFIX_OFFSET, 16, 1,
1658 skb_mac_header(skb),
1659 dev->hard_header_len, true);
1665 /* called in rcu_read_lock() section */
1666 static int __mkroute_input(struct sk_buff *skb,
1667 const struct fib_result *res,
1668 struct in_device *in_dev,
1669 __be32 daddr, __be32 saddr, u32 tos)
1671 struct fib_nh_exception *fnhe;
1674 struct in_device *out_dev;
1678 /* get a working reference to the output device */
1679 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1681 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1685 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1686 in_dev->dev, in_dev, &itag);
1688 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1694 do_cache = res->fi && !itag;
1695 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1696 skb->protocol == htons(ETH_P_IP) &&
1697 (IN_DEV_SHARED_MEDIA(out_dev) ||
1698 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1699 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1701 if (skb->protocol != htons(ETH_P_IP)) {
1702 /* Not IP (i.e. ARP). Do not create route, if it is
1703 * invalid for proxy arp. DNAT routes are always valid.
1705 * Proxy arp feature have been extended to allow, ARP
1706 * replies back to the same interface, to support
1707 * Private VLAN switch technologies. See arp.c.
1709 if (out_dev == in_dev &&
1710 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1716 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1719 rth = rcu_dereference(fnhe->fnhe_rth_input);
1721 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1722 if (rt_cache_valid(rth)) {
1723 skb_dst_set_noref(skb, &rth->dst);
1728 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1729 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1730 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1736 rth->rt_is_input = 1;
1737 RT_CACHE_STAT_INC(in_slow_tot);
1739 rth->dst.input = ip_forward;
1741 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1743 lwtunnel_set_redirect(&rth->dst);
1744 skb_dst_set(skb, &rth->dst);
1751 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1752 /* To make ICMP packets follow the right flow, the multipath hash is
1753 * calculated from the inner IP addresses.
1755 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1756 struct flow_keys *hash_keys)
1758 const struct iphdr *outer_iph = ip_hdr(skb);
1759 const struct iphdr *key_iph = outer_iph;
1760 const struct iphdr *inner_iph;
1761 const struct icmphdr *icmph;
1762 struct iphdr _inner_iph;
1763 struct icmphdr _icmph;
1765 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1768 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1771 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1776 if (icmph->type != ICMP_DEST_UNREACH &&
1777 icmph->type != ICMP_REDIRECT &&
1778 icmph->type != ICMP_TIME_EXCEEDED &&
1779 icmph->type != ICMP_PARAMETERPROB)
1782 inner_iph = skb_header_pointer(skb,
1783 outer_iph->ihl * 4 + sizeof(_icmph),
1784 sizeof(_inner_iph), &_inner_iph);
1788 key_iph = inner_iph;
1790 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1791 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1794 /* if skb is set it will be used and fl4 can be NULL */
1795 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1796 const struct sk_buff *skb, struct flow_keys *flkeys)
1798 struct flow_keys hash_keys;
1801 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1803 memset(&hash_keys, 0, sizeof(hash_keys));
1804 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1806 ip_multipath_l3_keys(skb, &hash_keys);
1808 hash_keys.addrs.v4addrs.src = fl4->saddr;
1809 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1813 /* skb is currently provided only when forwarding */
1815 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1816 struct flow_keys keys;
1818 /* short-circuit if we already have L4 hash present */
1820 return skb_get_hash_raw(skb) >> 1;
1822 memset(&hash_keys, 0, sizeof(hash_keys));
1825 skb_flow_dissect_flow_keys(skb, &keys, flag);
1829 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1830 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1831 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1832 hash_keys.ports.src = flkeys->ports.src;
1833 hash_keys.ports.dst = flkeys->ports.dst;
1834 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1836 memset(&hash_keys, 0, sizeof(hash_keys));
1837 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1838 hash_keys.addrs.v4addrs.src = fl4->saddr;
1839 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1840 hash_keys.ports.src = fl4->fl4_sport;
1841 hash_keys.ports.dst = fl4->fl4_dport;
1842 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1846 mhash = flow_hash_from_keys(&hash_keys);
1850 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1852 static int ip_mkroute_input(struct sk_buff *skb,
1853 struct fib_result *res,
1854 struct in_device *in_dev,
1855 __be32 daddr, __be32 saddr, u32 tos,
1856 struct flow_keys *hkeys)
1858 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1859 if (res->fi && res->fi->fib_nhs > 1) {
1860 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1862 fib_select_multipath(res, h);
1866 /* create a routing cache entry */
1867 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1871 * NOTE. We drop all the packets that has local source
1872 * addresses, because every properly looped back packet
1873 * must have correct destination already attached by output routine.
1875 * Such approach solves two big problems:
1876 * 1. Not simplex devices are handled properly.
1877 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1878 * called with rcu_read_lock()
1881 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1882 u8 tos, struct net_device *dev,
1883 struct fib_result *res)
1885 struct in_device *in_dev = __in_dev_get_rcu(dev);
1886 struct flow_keys *flkeys = NULL, _flkeys;
1887 struct net *net = dev_net(dev);
1888 struct ip_tunnel_info *tun_info;
1890 unsigned int flags = 0;
1896 /* IP on this device is disabled. */
1901 /* Check for the most weird martians, which can be not detected
1905 tun_info = skb_tunnel_info(skb);
1906 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1907 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1909 fl4.flowi4_tun_key.tun_id = 0;
1912 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1913 goto martian_source;
1917 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1920 /* Accept zero addresses only to limited broadcast;
1921 * I even do not know to fix it or not. Waiting for complains :-)
1923 if (ipv4_is_zeronet(saddr))
1924 goto martian_source;
1926 if (ipv4_is_zeronet(daddr))
1927 goto martian_destination;
1929 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1930 * and call it once if daddr or/and saddr are loopback addresses
1932 if (ipv4_is_loopback(daddr)) {
1933 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1934 goto martian_destination;
1935 } else if (ipv4_is_loopback(saddr)) {
1936 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1937 goto martian_source;
1941 * Now we are ready to route packet.
1944 fl4.flowi4_iif = dev->ifindex;
1945 fl4.flowi4_mark = skb->mark;
1946 fl4.flowi4_tos = tos;
1947 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1948 fl4.flowi4_flags = 0;
1951 fl4.flowi4_uid = sock_net_uid(net, NULL);
1953 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys))
1956 err = fib_lookup(net, &fl4, res, 0);
1958 if (!IN_DEV_FORWARD(in_dev))
1959 err = -EHOSTUNREACH;
1963 if (res->type == RTN_BROADCAST)
1966 if (res->type == RTN_LOCAL) {
1967 err = fib_validate_source(skb, saddr, daddr, tos,
1968 0, dev, in_dev, &itag);
1970 goto martian_source;
1974 if (!IN_DEV_FORWARD(in_dev)) {
1975 err = -EHOSTUNREACH;
1978 if (res->type != RTN_UNICAST)
1979 goto martian_destination;
1981 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
1985 if (skb->protocol != htons(ETH_P_IP))
1988 if (!ipv4_is_zeronet(saddr)) {
1989 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1992 goto martian_source;
1994 flags |= RTCF_BROADCAST;
1995 res->type = RTN_BROADCAST;
1996 RT_CACHE_STAT_INC(in_brd);
2002 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2003 if (rt_cache_valid(rth)) {
2004 skb_dst_set_noref(skb, &rth->dst);
2012 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2013 flags | RTCF_LOCAL, res->type,
2014 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2018 rth->dst.output= ip_rt_bug;
2019 #ifdef CONFIG_IP_ROUTE_CLASSID
2020 rth->dst.tclassid = itag;
2022 rth->rt_is_input = 1;
2024 RT_CACHE_STAT_INC(in_slow_tot);
2025 if (res->type == RTN_UNREACHABLE) {
2026 rth->dst.input= ip_error;
2027 rth->dst.error= -err;
2028 rth->rt_flags &= ~RTCF_LOCAL;
2032 struct fib_nh *nh = &FIB_RES_NH(*res);
2034 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2035 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2036 WARN_ON(rth->dst.input == lwtunnel_input);
2037 rth->dst.lwtstate->orig_input = rth->dst.input;
2038 rth->dst.input = lwtunnel_input;
2041 if (unlikely(!rt_cache_route(nh, rth)))
2042 rt_add_uncached_list(rth);
2044 skb_dst_set(skb, &rth->dst);
2049 RT_CACHE_STAT_INC(in_no_route);
2050 res->type = RTN_UNREACHABLE;
2056 * Do not cache martian addresses: they should be logged (RFC1812)
2058 martian_destination:
2059 RT_CACHE_STAT_INC(in_martian_dst);
2060 #ifdef CONFIG_IP_ROUTE_VERBOSE
2061 if (IN_DEV_LOG_MARTIANS(in_dev))
2062 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2063 &daddr, &saddr, dev->name);
2075 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2079 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2080 u8 tos, struct net_device *dev)
2082 struct fib_result res;
2085 tos &= IPTOS_RT_MASK;
2087 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2092 EXPORT_SYMBOL(ip_route_input_noref);
2094 /* called with rcu_read_lock held */
2095 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2096 u8 tos, struct net_device *dev, struct fib_result *res)
2098 /* Multicast recognition logic is moved from route cache to here.
2099 The problem was that too many Ethernet cards have broken/missing
2100 hardware multicast filters :-( As result the host on multicasting
2101 network acquires a lot of useless route cache entries, sort of
2102 SDR messages from all the world. Now we try to get rid of them.
2103 Really, provided software IP multicast filter is organized
2104 reasonably (at least, hashed), it does not result in a slowdown
2105 comparing with route cache reject entries.
2106 Note, that multicast routers are not affected, because
2107 route cache entry is created eventually.
2109 if (ipv4_is_multicast(daddr)) {
2110 struct in_device *in_dev = __in_dev_get_rcu(dev);
2115 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2116 ip_hdr(skb)->protocol);
2118 /* check l3 master if no match yet */
2119 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2120 struct in_device *l3_in_dev;
2122 l3_in_dev = __in_dev_get_rcu(skb->dev);
2124 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2125 ip_hdr(skb)->protocol);
2129 #ifdef CONFIG_IP_MROUTE
2131 (!ipv4_is_local_multicast(daddr) &&
2132 IN_DEV_MFORWARD(in_dev))
2135 err = ip_route_input_mc(skb, daddr, saddr,
2141 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2144 /* called with rcu_read_lock() */
2145 static struct rtable *__mkroute_output(const struct fib_result *res,
2146 const struct flowi4 *fl4, int orig_oif,
2147 struct net_device *dev_out,
2150 struct fib_info *fi = res->fi;
2151 struct fib_nh_exception *fnhe;
2152 struct in_device *in_dev;
2153 u16 type = res->type;
2157 in_dev = __in_dev_get_rcu(dev_out);
2159 return ERR_PTR(-EINVAL);
2161 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2162 if (ipv4_is_loopback(fl4->saddr) &&
2163 !(dev_out->flags & IFF_LOOPBACK) &&
2164 !netif_is_l3_master(dev_out))
2165 return ERR_PTR(-EINVAL);
2167 if (ipv4_is_lbcast(fl4->daddr))
2168 type = RTN_BROADCAST;
2169 else if (ipv4_is_multicast(fl4->daddr))
2170 type = RTN_MULTICAST;
2171 else if (ipv4_is_zeronet(fl4->daddr))
2172 return ERR_PTR(-EINVAL);
2174 if (dev_out->flags & IFF_LOOPBACK)
2175 flags |= RTCF_LOCAL;
2178 if (type == RTN_BROADCAST) {
2179 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2181 } else if (type == RTN_MULTICAST) {
2182 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2183 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2185 flags &= ~RTCF_LOCAL;
2188 /* If multicast route do not exist use
2189 * default one, but do not gateway in this case.
2192 if (fi && res->prefixlen < 4)
2194 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2195 (orig_oif != dev_out->ifindex)) {
2196 /* For local routes that require a particular output interface
2197 * we do not want to cache the result. Caching the result
2198 * causes incorrect behaviour when there are multiple source
2199 * addresses on the interface, the end result being that if the
2200 * intended recipient is waiting on that interface for the
2201 * packet he won't receive it because it will be delivered on
2202 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2203 * be set to the loopback interface as well.
2209 do_cache &= fi != NULL;
2211 struct rtable __rcu **prth;
2212 struct fib_nh *nh = &FIB_RES_NH(*res);
2214 fnhe = find_exception(nh, fl4->daddr);
2218 prth = &fnhe->fnhe_rth_output;
2220 if (unlikely(fl4->flowi4_flags &
2221 FLOWI_FLAG_KNOWN_NH &&
2223 nh->nh_scope == RT_SCOPE_LINK))) {
2227 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2229 rth = rcu_dereference(*prth);
2230 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2235 rth = rt_dst_alloc(dev_out, flags, type,
2236 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2237 IN_DEV_CONF_GET(in_dev, NOXFRM),
2240 return ERR_PTR(-ENOBUFS);
2242 rth->rt_iif = orig_oif;
2244 RT_CACHE_STAT_INC(out_slow_tot);
2246 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2247 if (flags & RTCF_LOCAL &&
2248 !(dev_out->flags & IFF_LOOPBACK)) {
2249 rth->dst.output = ip_mc_output;
2250 RT_CACHE_STAT_INC(out_slow_mc);
2252 #ifdef CONFIG_IP_MROUTE
2253 if (type == RTN_MULTICAST) {
2254 if (IN_DEV_MFORWARD(in_dev) &&
2255 !ipv4_is_local_multicast(fl4->daddr)) {
2256 rth->dst.input = ip_mr_input;
2257 rth->dst.output = ip_mc_output;
2263 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2264 lwtunnel_set_redirect(&rth->dst);
2270 * Major route resolver routine.
2273 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2274 const struct sk_buff *skb)
2276 __u8 tos = RT_FL_TOS(fl4);
2277 struct fib_result res = {
2285 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2286 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2287 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2288 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2291 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2296 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2298 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2299 struct fib_result *res,
2300 const struct sk_buff *skb)
2302 struct net_device *dev_out = NULL;
2303 int orig_oif = fl4->flowi4_oif;
2304 unsigned int flags = 0;
2306 int err = -ENETUNREACH;
2309 rth = ERR_PTR(-EINVAL);
2310 if (ipv4_is_multicast(fl4->saddr) ||
2311 ipv4_is_lbcast(fl4->saddr) ||
2312 ipv4_is_zeronet(fl4->saddr))
2315 /* I removed check for oif == dev_out->oif here.
2316 It was wrong for two reasons:
2317 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2318 is assigned to multiple interfaces.
2319 2. Moreover, we are allowed to send packets with saddr
2320 of another iface. --ANK
2323 if (fl4->flowi4_oif == 0 &&
2324 (ipv4_is_multicast(fl4->daddr) ||
2325 ipv4_is_lbcast(fl4->daddr))) {
2326 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2327 dev_out = __ip_dev_find(net, fl4->saddr, false);
2331 /* Special hack: user can direct multicasts
2332 and limited broadcast via necessary interface
2333 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2334 This hack is not just for fun, it allows
2335 vic,vat and friends to work.
2336 They bind socket to loopback, set ttl to zero
2337 and expect that it will work.
2338 From the viewpoint of routing cache they are broken,
2339 because we are not allowed to build multicast path
2340 with loopback source addr (look, routing cache
2341 cannot know, that ttl is zero, so that packet
2342 will not leave this host and route is valid).
2343 Luckily, this hack is good workaround.
2346 fl4->flowi4_oif = dev_out->ifindex;
2350 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2351 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2352 if (!__ip_dev_find(net, fl4->saddr, false))
2358 if (fl4->flowi4_oif) {
2359 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2360 rth = ERR_PTR(-ENODEV);
2364 /* RACE: Check return value of inet_select_addr instead. */
2365 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2366 rth = ERR_PTR(-ENETUNREACH);
2369 if (ipv4_is_local_multicast(fl4->daddr) ||
2370 ipv4_is_lbcast(fl4->daddr) ||
2371 fl4->flowi4_proto == IPPROTO_IGMP) {
2373 fl4->saddr = inet_select_addr(dev_out, 0,
2378 if (ipv4_is_multicast(fl4->daddr))
2379 fl4->saddr = inet_select_addr(dev_out, 0,
2381 else if (!fl4->daddr)
2382 fl4->saddr = inet_select_addr(dev_out, 0,
2388 fl4->daddr = fl4->saddr;
2390 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2391 dev_out = net->loopback_dev;
2392 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2393 res->type = RTN_LOCAL;
2394 flags |= RTCF_LOCAL;
2398 err = fib_lookup(net, fl4, res, 0);
2402 if (fl4->flowi4_oif &&
2403 (ipv4_is_multicast(fl4->daddr) ||
2404 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2405 /* Apparently, routing tables are wrong. Assume,
2406 that the destination is on link.
2409 Because we are allowed to send to iface
2410 even if it has NO routes and NO assigned
2411 addresses. When oif is specified, routing
2412 tables are looked up with only one purpose:
2413 to catch if destination is gatewayed, rather than
2414 direct. Moreover, if MSG_DONTROUTE is set,
2415 we send packet, ignoring both routing tables
2416 and ifaddr state. --ANK
2419 We could make it even if oif is unknown,
2420 likely IPv6, but we do not.
2423 if (fl4->saddr == 0)
2424 fl4->saddr = inet_select_addr(dev_out, 0,
2426 res->type = RTN_UNICAST;
2433 if (res->type == RTN_LOCAL) {
2435 if (res->fi->fib_prefsrc)
2436 fl4->saddr = res->fi->fib_prefsrc;
2438 fl4->saddr = fl4->daddr;
2441 /* L3 master device is the loopback for that domain */
2442 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2445 /* make sure orig_oif points to fib result device even
2446 * though packet rx/tx happens over loopback or l3mdev
2448 orig_oif = FIB_RES_OIF(*res);
2450 fl4->flowi4_oif = dev_out->ifindex;
2451 flags |= RTCF_LOCAL;
2455 fib_select_path(net, res, fl4, skb);
2457 dev_out = FIB_RES_DEV(*res);
2458 fl4->flowi4_oif = dev_out->ifindex;
2462 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2468 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2473 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2475 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2477 return mtu ? : dst->dev->mtu;
2480 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2481 struct sk_buff *skb, u32 mtu)
2485 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2486 struct sk_buff *skb)
2490 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2496 static struct dst_ops ipv4_dst_blackhole_ops = {
2498 .check = ipv4_blackhole_dst_check,
2499 .mtu = ipv4_blackhole_mtu,
2500 .default_advmss = ipv4_default_advmss,
2501 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2502 .redirect = ipv4_rt_blackhole_redirect,
2503 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2504 .neigh_lookup = ipv4_neigh_lookup,
2507 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2509 struct rtable *ort = (struct rtable *) dst_orig;
2512 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2514 struct dst_entry *new = &rt->dst;
2517 new->input = dst_discard;
2518 new->output = dst_discard_out;
2520 new->dev = net->loopback_dev;
2524 rt->rt_is_input = ort->rt_is_input;
2525 rt->rt_iif = ort->rt_iif;
2526 rt->rt_pmtu = ort->rt_pmtu;
2527 rt->rt_mtu_locked = ort->rt_mtu_locked;
2529 rt->rt_genid = rt_genid_ipv4(net);
2530 rt->rt_flags = ort->rt_flags;
2531 rt->rt_type = ort->rt_type;
2532 rt->rt_gateway = ort->rt_gateway;
2533 rt->rt_uses_gateway = ort->rt_uses_gateway;
2535 INIT_LIST_HEAD(&rt->rt_uncached);
2538 dst_release(dst_orig);
2540 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2543 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2544 const struct sock *sk)
2546 struct rtable *rt = __ip_route_output_key(net, flp4);
2551 if (flp4->flowi4_proto)
2552 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2553 flowi4_to_flowi(flp4),
2558 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2560 /* called with rcu_read_lock held */
2561 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2562 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2565 struct rtable *rt = skb_rtable(skb);
2567 struct nlmsghdr *nlh;
2568 unsigned long expires = 0;
2570 u32 metrics[RTAX_MAX];
2572 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2576 r = nlmsg_data(nlh);
2577 r->rtm_family = AF_INET;
2578 r->rtm_dst_len = 32;
2580 r->rtm_tos = fl4->flowi4_tos;
2581 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2582 if (nla_put_u32(skb, RTA_TABLE, table_id))
2583 goto nla_put_failure;
2584 r->rtm_type = rt->rt_type;
2585 r->rtm_scope = RT_SCOPE_UNIVERSE;
2586 r->rtm_protocol = RTPROT_UNSPEC;
2587 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2588 if (rt->rt_flags & RTCF_NOTIFY)
2589 r->rtm_flags |= RTM_F_NOTIFY;
2590 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2591 r->rtm_flags |= RTCF_DOREDIRECT;
2593 if (nla_put_in_addr(skb, RTA_DST, dst))
2594 goto nla_put_failure;
2596 r->rtm_src_len = 32;
2597 if (nla_put_in_addr(skb, RTA_SRC, src))
2598 goto nla_put_failure;
2601 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2602 goto nla_put_failure;
2603 #ifdef CONFIG_IP_ROUTE_CLASSID
2604 if (rt->dst.tclassid &&
2605 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2606 goto nla_put_failure;
2608 if (!rt_is_input_route(rt) &&
2609 fl4->saddr != src) {
2610 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2611 goto nla_put_failure;
2613 if (rt->rt_uses_gateway &&
2614 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2615 goto nla_put_failure;
2617 expires = rt->dst.expires;
2619 unsigned long now = jiffies;
2621 if (time_before(now, expires))
2627 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2628 if (rt->rt_pmtu && expires)
2629 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2630 if (rt->rt_mtu_locked && expires)
2631 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2632 if (rtnetlink_put_metrics(skb, metrics) < 0)
2633 goto nla_put_failure;
2635 if (fl4->flowi4_mark &&
2636 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2637 goto nla_put_failure;
2639 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2640 nla_put_u32(skb, RTA_UID,
2641 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2642 goto nla_put_failure;
2644 error = rt->dst.error;
2646 if (rt_is_input_route(rt)) {
2647 #ifdef CONFIG_IP_MROUTE
2648 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2649 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2650 int err = ipmr_get_route(net, skb,
2651 fl4->saddr, fl4->daddr,
2657 goto nla_put_failure;
2661 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2662 goto nla_put_failure;
2665 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2666 goto nla_put_failure;
2668 nlmsg_end(skb, nlh);
2672 nlmsg_cancel(skb, nlh);
2676 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2677 struct netlink_ext_ack *extack)
2679 struct net *net = sock_net(in_skb->sk);
2681 struct nlattr *tb[RTA_MAX+1];
2682 struct fib_result res = {};
2683 struct rtable *rt = NULL;
2690 struct sk_buff *skb;
2691 u32 table_id = RT_TABLE_MAIN;
2694 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2699 rtm = nlmsg_data(nlh);
2701 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2707 /* Reserve room for dummy headers, this skb can pass
2708 through good chunk of routing engine.
2710 skb_reset_mac_header(skb);
2711 skb_reset_network_header(skb);
2713 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2714 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2715 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2716 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2718 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2720 uid = (iif ? INVALID_UID : current_uid());
2722 /* Bugfix: need to give ip_route_input enough of an IP header to
2725 ip_hdr(skb)->protocol = IPPROTO_UDP;
2726 ip_hdr(skb)->saddr = src;
2727 ip_hdr(skb)->daddr = dst;
2729 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2731 memset(&fl4, 0, sizeof(fl4));
2734 fl4.flowi4_tos = rtm->rtm_tos;
2735 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2736 fl4.flowi4_mark = mark;
2737 fl4.flowi4_uid = uid;
2742 struct net_device *dev;
2744 dev = dev_get_by_index_rcu(net, iif);
2750 skb->protocol = htons(ETH_P_IP);
2753 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2756 rt = skb_rtable(skb);
2757 if (err == 0 && rt->dst.error)
2758 err = -rt->dst.error;
2760 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2761 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2766 skb_dst_set(skb, &rt->dst);
2772 if (rtm->rtm_flags & RTM_F_NOTIFY)
2773 rt->rt_flags |= RTCF_NOTIFY;
2775 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2776 table_id = res.table ? res.table->tb_id : 0;
2778 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2780 err = fib_props[res.type].error;
2782 err = -EHOSTUNREACH;
2785 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2786 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2787 rt->rt_type, res.prefix, res.prefixlen,
2788 fl4.flowi4_tos, res.fi, 0);
2790 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2791 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2798 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2808 void ip_rt_multicast_event(struct in_device *in_dev)
2810 rt_cache_flush(dev_net(in_dev->dev));
2813 #ifdef CONFIG_SYSCTL
2814 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2815 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2816 static int ip_rt_gc_elasticity __read_mostly = 8;
2817 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
2819 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2820 void __user *buffer,
2821 size_t *lenp, loff_t *ppos)
2823 struct net *net = (struct net *)__ctl->extra1;
2826 rt_cache_flush(net);
2827 fnhe_genid_bump(net);
2834 static struct ctl_table ipv4_route_table[] = {
2836 .procname = "gc_thresh",
2837 .data = &ipv4_dst_ops.gc_thresh,
2838 .maxlen = sizeof(int),
2840 .proc_handler = proc_dointvec,
2843 .procname = "max_size",
2844 .data = &ip_rt_max_size,
2845 .maxlen = sizeof(int),
2847 .proc_handler = proc_dointvec,
2850 /* Deprecated. Use gc_min_interval_ms */
2852 .procname = "gc_min_interval",
2853 .data = &ip_rt_gc_min_interval,
2854 .maxlen = sizeof(int),
2856 .proc_handler = proc_dointvec_jiffies,
2859 .procname = "gc_min_interval_ms",
2860 .data = &ip_rt_gc_min_interval,
2861 .maxlen = sizeof(int),
2863 .proc_handler = proc_dointvec_ms_jiffies,
2866 .procname = "gc_timeout",
2867 .data = &ip_rt_gc_timeout,
2868 .maxlen = sizeof(int),
2870 .proc_handler = proc_dointvec_jiffies,
2873 .procname = "gc_interval",
2874 .data = &ip_rt_gc_interval,
2875 .maxlen = sizeof(int),
2877 .proc_handler = proc_dointvec_jiffies,
2880 .procname = "redirect_load",
2881 .data = &ip_rt_redirect_load,
2882 .maxlen = sizeof(int),
2884 .proc_handler = proc_dointvec,
2887 .procname = "redirect_number",
2888 .data = &ip_rt_redirect_number,
2889 .maxlen = sizeof(int),
2891 .proc_handler = proc_dointvec,
2894 .procname = "redirect_silence",
2895 .data = &ip_rt_redirect_silence,
2896 .maxlen = sizeof(int),
2898 .proc_handler = proc_dointvec,
2901 .procname = "error_cost",
2902 .data = &ip_rt_error_cost,
2903 .maxlen = sizeof(int),
2905 .proc_handler = proc_dointvec,
2908 .procname = "error_burst",
2909 .data = &ip_rt_error_burst,
2910 .maxlen = sizeof(int),
2912 .proc_handler = proc_dointvec,
2915 .procname = "gc_elasticity",
2916 .data = &ip_rt_gc_elasticity,
2917 .maxlen = sizeof(int),
2919 .proc_handler = proc_dointvec,
2922 .procname = "mtu_expires",
2923 .data = &ip_rt_mtu_expires,
2924 .maxlen = sizeof(int),
2926 .proc_handler = proc_dointvec_jiffies,
2929 .procname = "min_pmtu",
2930 .data = &ip_rt_min_pmtu,
2931 .maxlen = sizeof(int),
2933 .proc_handler = proc_dointvec_minmax,
2934 .extra1 = &ip_min_valid_pmtu,
2937 .procname = "min_adv_mss",
2938 .data = &ip_rt_min_advmss,
2939 .maxlen = sizeof(int),
2941 .proc_handler = proc_dointvec,
2946 static struct ctl_table ipv4_route_flush_table[] = {
2948 .procname = "flush",
2949 .maxlen = sizeof(int),
2951 .proc_handler = ipv4_sysctl_rtcache_flush,
2956 static __net_init int sysctl_route_net_init(struct net *net)
2958 struct ctl_table *tbl;
2960 tbl = ipv4_route_flush_table;
2961 if (!net_eq(net, &init_net)) {
2962 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2966 /* Don't export sysctls to unprivileged users */
2967 if (net->user_ns != &init_user_ns)
2968 tbl[0].procname = NULL;
2970 tbl[0].extra1 = net;
2972 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2973 if (!net->ipv4.route_hdr)
2978 if (tbl != ipv4_route_flush_table)
2984 static __net_exit void sysctl_route_net_exit(struct net *net)
2986 struct ctl_table *tbl;
2988 tbl = net->ipv4.route_hdr->ctl_table_arg;
2989 unregister_net_sysctl_table(net->ipv4.route_hdr);
2990 BUG_ON(tbl == ipv4_route_flush_table);
2994 static __net_initdata struct pernet_operations sysctl_route_ops = {
2995 .init = sysctl_route_net_init,
2996 .exit = sysctl_route_net_exit,
3000 static __net_init int rt_genid_init(struct net *net)
3002 atomic_set(&net->ipv4.rt_genid, 0);
3003 atomic_set(&net->fnhe_genid, 0);
3004 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3008 static __net_initdata struct pernet_operations rt_genid_ops = {
3009 .init = rt_genid_init,
3012 static int __net_init ipv4_inetpeer_init(struct net *net)
3014 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3018 inet_peer_base_init(bp);
3019 net->ipv4.peers = bp;
3023 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3025 struct inet_peer_base *bp = net->ipv4.peers;
3027 net->ipv4.peers = NULL;
3028 inetpeer_invalidate_tree(bp);
3032 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3033 .init = ipv4_inetpeer_init,
3034 .exit = ipv4_inetpeer_exit,
3037 #ifdef CONFIG_IP_ROUTE_CLASSID
3038 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3039 #endif /* CONFIG_IP_ROUTE_CLASSID */
3041 int __init ip_rt_init(void)
3045 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3047 panic("IP: failed to allocate ip_idents\n");
3049 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3051 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3053 panic("IP: failed to allocate ip_tstamps\n");
3055 for_each_possible_cpu(cpu) {
3056 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3058 INIT_LIST_HEAD(&ul->head);
3059 spin_lock_init(&ul->lock);
3061 #ifdef CONFIG_IP_ROUTE_CLASSID
3062 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3064 panic("IP: failed to allocate ip_rt_acct\n");
3067 ipv4_dst_ops.kmem_cachep =
3068 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3069 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3071 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3073 if (dst_entries_init(&ipv4_dst_ops) < 0)
3074 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3076 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3077 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3079 ipv4_dst_ops.gc_thresh = ~0;
3080 ip_rt_max_size = INT_MAX;
3085 if (ip_rt_proc_init())
3086 pr_err("Unable to create route proc files\n");
3091 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3092 RTNL_FLAG_DOIT_UNLOCKED);
3094 #ifdef CONFIG_SYSCTL
3095 register_pernet_subsys(&sysctl_route_ops);
3097 register_pernet_subsys(&rt_genid_ops);
3098 register_pernet_subsys(&ipv4_inetpeer_ops);
3102 #ifdef CONFIG_SYSCTL
3104 * We really need to sanitize the damn ipv4 init order, then all
3105 * this nonsense will go away.
3107 void __init ip_static_sysctl_init(void)
3109 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);