2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
94 #include <net/net_namespace.h>
95 #include <net/protocol.h>
97 #include <net/route.h>
98 #include <net/inetpeer.h>
100 #include <net/ip_fib.h>
103 #include <net/icmp.h>
104 #include <net/xfrm.h>
105 #include <net/netevent.h>
106 #include <net/rtnetlink.h>
108 #include <linux/sysctl.h>
109 #include <linux/kmemleak.h>
111 #include <net/secure_seq.h>
113 #define RT_FL_TOS(oldflp4) \
114 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
116 #define RT_GC_TIMEOUT (300*HZ)
118 static int ip_rt_max_size;
119 static int ip_rt_redirect_number __read_mostly = 9;
120 static int ip_rt_redirect_load __read_mostly = HZ / 50;
121 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
122 static int ip_rt_error_cost __read_mostly = HZ;
123 static int ip_rt_error_burst __read_mostly = 5 * HZ;
124 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
125 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
126 static int ip_rt_min_advmss __read_mostly = 256;
129 * Interface to generic destination cache.
132 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
133 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
134 static unsigned int ipv4_mtu(const struct dst_entry *dst);
135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
136 static void ipv4_link_failure(struct sk_buff *skb);
137 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
138 struct sk_buff *skb, u32 mtu);
139 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
140 struct sk_buff *skb);
141 static void ipv4_dst_destroy(struct dst_entry *dst);
143 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
158 static struct dst_ops ipv4_dst_ops = {
160 .protocol = cpu_to_be16(ETH_P_IP),
161 .check = ipv4_dst_check,
162 .default_advmss = ipv4_default_advmss,
164 .cow_metrics = ipv4_cow_metrics,
165 .destroy = ipv4_dst_destroy,
166 .ifdown = ipv4_dst_ifdown,
167 .negative_advice = ipv4_negative_advice,
168 .link_failure = ipv4_link_failure,
169 .update_pmtu = ip_rt_update_pmtu,
170 .redirect = ip_do_redirect,
171 .local_out = __ip_local_out,
172 .neigh_lookup = ipv4_neigh_lookup,
175 #define ECN_OR_COST(class) TC_PRIO_##class
177 const __u8 ip_tos2prio[16] = {
179 ECN_OR_COST(BESTEFFORT),
181 ECN_OR_COST(BESTEFFORT),
187 ECN_OR_COST(INTERACTIVE),
189 ECN_OR_COST(INTERACTIVE),
190 TC_PRIO_INTERACTIVE_BULK,
191 ECN_OR_COST(INTERACTIVE_BULK),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK)
195 EXPORT_SYMBOL(ip_tos2prio);
197 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
198 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
200 #ifdef CONFIG_PROC_FS
201 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
205 return SEQ_START_TOKEN;
208 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
214 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
218 static int rt_cache_seq_show(struct seq_file *seq, void *v)
220 if (v == SEQ_START_TOKEN)
221 seq_printf(seq, "%-127s\n",
222 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
223 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
228 static const struct seq_operations rt_cache_seq_ops = {
229 .start = rt_cache_seq_start,
230 .next = rt_cache_seq_next,
231 .stop = rt_cache_seq_stop,
232 .show = rt_cache_seq_show,
235 static int rt_cache_seq_open(struct inode *inode, struct file *file)
237 return seq_open(file, &rt_cache_seq_ops);
240 static const struct file_operations rt_cache_seq_fops = {
241 .owner = THIS_MODULE,
242 .open = rt_cache_seq_open,
245 .release = seq_release,
249 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
254 return SEQ_START_TOKEN;
256 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
257 if (!cpu_possible(cpu))
260 return &per_cpu(rt_cache_stat, cpu);
265 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
269 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
270 if (!cpu_possible(cpu))
273 return &per_cpu(rt_cache_stat, cpu);
279 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
284 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
286 struct rt_cache_stat *st = v;
288 if (v == SEQ_START_TOKEN) {
289 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
293 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
294 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
295 dst_entries_get_slow(&ipv4_dst_ops),
308 0, /* st->gc_total */
309 0, /* st->gc_ignored */
310 0, /* st->gc_goal_miss */
311 0, /* st->gc_dst_overflow */
312 0, /* st->in_hlist_search */
313 0 /* st->out_hlist_search */
318 static const struct seq_operations rt_cpu_seq_ops = {
319 .start = rt_cpu_seq_start,
320 .next = rt_cpu_seq_next,
321 .stop = rt_cpu_seq_stop,
322 .show = rt_cpu_seq_show,
326 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
328 return seq_open(file, &rt_cpu_seq_ops);
331 static const struct file_operations rt_cpu_seq_fops = {
332 .owner = THIS_MODULE,
333 .open = rt_cpu_seq_open,
336 .release = seq_release,
339 #ifdef CONFIG_IP_ROUTE_CLASSID
340 static int rt_acct_proc_show(struct seq_file *m, void *v)
342 struct ip_rt_acct *dst, *src;
345 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
349 for_each_possible_cpu(i) {
350 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
351 for (j = 0; j < 256; j++) {
352 dst[j].o_bytes += src[j].o_bytes;
353 dst[j].o_packets += src[j].o_packets;
354 dst[j].i_bytes += src[j].i_bytes;
355 dst[j].i_packets += src[j].i_packets;
359 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
364 static int rt_acct_proc_open(struct inode *inode, struct file *file)
366 return single_open(file, rt_acct_proc_show, NULL);
369 static const struct file_operations rt_acct_proc_fops = {
370 .owner = THIS_MODULE,
371 .open = rt_acct_proc_open,
374 .release = single_release,
378 static int __net_init ip_rt_do_proc_init(struct net *net)
380 struct proc_dir_entry *pde;
382 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
387 pde = proc_create("rt_cache", S_IRUGO,
388 net->proc_net_stat, &rt_cpu_seq_fops);
392 #ifdef CONFIG_IP_ROUTE_CLASSID
393 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
399 #ifdef CONFIG_IP_ROUTE_CLASSID
401 remove_proc_entry("rt_cache", net->proc_net_stat);
404 remove_proc_entry("rt_cache", net->proc_net);
409 static void __net_exit ip_rt_do_proc_exit(struct net *net)
411 remove_proc_entry("rt_cache", net->proc_net_stat);
412 remove_proc_entry("rt_cache", net->proc_net);
413 #ifdef CONFIG_IP_ROUTE_CLASSID
414 remove_proc_entry("rt_acct", net->proc_net);
418 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
419 .init = ip_rt_do_proc_init,
420 .exit = ip_rt_do_proc_exit,
423 static int __init ip_rt_proc_init(void)
425 return register_pernet_subsys(&ip_rt_proc_ops);
429 static inline int ip_rt_proc_init(void)
433 #endif /* CONFIG_PROC_FS */
435 static inline bool rt_is_expired(const struct rtable *rth)
437 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
440 void rt_cache_flush(struct net *net)
442 rt_genid_bump_ipv4(net);
445 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
449 struct net_device *dev = dst->dev;
450 const __be32 *pkey = daddr;
451 const struct rtable *rt;
454 rt = (const struct rtable *) dst;
456 pkey = (const __be32 *) &rt->rt_gateway;
458 pkey = &ip_hdr(skb)->daddr;
460 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
463 return neigh_create(&arp_tbl, pkey, dev);
466 #define IP_IDENTS_SZ 2048u
467 struct ip_ident_bucket {
472 static struct ip_ident_bucket *ip_idents __read_mostly;
474 /* In order to protect privacy, we add a perturbation to identifiers
475 * if one generator is seldom used. This makes hard for an attacker
476 * to infer how many packets were sent between two points in time.
478 u32 ip_idents_reserve(u32 hash, int segs)
480 struct ip_ident_bucket *bucket = ip_idents + hash % IP_IDENTS_SZ;
481 u32 old = ACCESS_ONCE(bucket->stamp32);
482 u32 now = (u32)jiffies;
485 if (old != now && cmpxchg(&bucket->stamp32, old, now) == old)
486 delta = prandom_u32_max(now - old);
488 return atomic_add_return(segs + delta, &bucket->id) - segs;
490 EXPORT_SYMBOL(ip_idents_reserve);
492 void __ip_select_ident(struct iphdr *iph, int segs)
494 static u32 ip_idents_hashrnd __read_mostly;
497 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
499 hash = jhash_3words((__force u32)iph->daddr,
500 (__force u32)iph->saddr,
503 id = ip_idents_reserve(hash, segs);
506 EXPORT_SYMBOL(__ip_select_ident);
508 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
509 const struct iphdr *iph,
511 u8 prot, u32 mark, int flow_flags)
514 const struct inet_sock *inet = inet_sk(sk);
516 oif = sk->sk_bound_dev_if;
518 tos = RT_CONN_FLAGS(sk);
519 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
521 flowi4_init_output(fl4, oif, mark, tos,
522 RT_SCOPE_UNIVERSE, prot,
524 iph->daddr, iph->saddr, 0, 0);
527 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
528 const struct sock *sk)
530 const struct iphdr *iph = ip_hdr(skb);
531 int oif = skb->dev->ifindex;
532 u8 tos = RT_TOS(iph->tos);
533 u8 prot = iph->protocol;
534 u32 mark = skb->mark;
536 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
539 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
541 const struct inet_sock *inet = inet_sk(sk);
542 const struct ip_options_rcu *inet_opt;
543 __be32 daddr = inet->inet_daddr;
546 inet_opt = rcu_dereference(inet->inet_opt);
547 if (inet_opt && inet_opt->opt.srr)
548 daddr = inet_opt->opt.faddr;
549 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
550 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
551 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
552 inet_sk_flowi_flags(sk),
553 daddr, inet->inet_saddr, 0, 0);
557 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
558 const struct sk_buff *skb)
561 build_skb_flow_key(fl4, skb, sk);
563 build_sk_flow_key(fl4, sk);
566 static inline void rt_free(struct rtable *rt)
568 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
571 static DEFINE_SPINLOCK(fnhe_lock);
573 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
577 rt = rcu_dereference(fnhe->fnhe_rth_input);
579 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
582 rt = rcu_dereference(fnhe->fnhe_rth_output);
584 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
589 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
591 struct fib_nh_exception *fnhe, *oldest;
593 oldest = rcu_dereference(hash->chain);
594 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
595 fnhe = rcu_dereference(fnhe->fnhe_next)) {
596 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
599 fnhe_flush_routes(oldest);
603 static inline u32 fnhe_hashfun(__be32 daddr)
607 hval = (__force u32) daddr;
608 hval ^= (hval >> 11) ^ (hval >> 22);
610 return hval & (FNHE_HASH_SIZE - 1);
613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
615 rt->rt_pmtu = fnhe->fnhe_pmtu;
616 rt->dst.expires = fnhe->fnhe_expires;
619 rt->rt_flags |= RTCF_REDIRECTED;
620 rt->rt_gateway = fnhe->fnhe_gw;
621 rt->rt_uses_gateway = 1;
625 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
626 u32 pmtu, unsigned long expires)
628 struct fnhe_hash_bucket *hash;
629 struct fib_nh_exception *fnhe;
633 u32 hval = fnhe_hashfun(daddr);
635 spin_lock_bh(&fnhe_lock);
637 hash = nh->nh_exceptions;
639 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
642 nh->nh_exceptions = hash;
648 for (fnhe = rcu_dereference(hash->chain); fnhe;
649 fnhe = rcu_dereference(fnhe->fnhe_next)) {
650 if (fnhe->fnhe_daddr == daddr)
659 fnhe->fnhe_pmtu = pmtu;
660 fnhe->fnhe_expires = max(1UL, expires);
662 /* Update all cached dsts too */
663 rt = rcu_dereference(fnhe->fnhe_rth_input);
665 fill_route_from_fnhe(rt, fnhe);
666 rt = rcu_dereference(fnhe->fnhe_rth_output);
668 fill_route_from_fnhe(rt, fnhe);
670 if (depth > FNHE_RECLAIM_DEPTH)
671 fnhe = fnhe_oldest(hash);
673 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
677 fnhe->fnhe_next = hash->chain;
678 rcu_assign_pointer(hash->chain, fnhe);
680 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
681 fnhe->fnhe_daddr = daddr;
683 fnhe->fnhe_pmtu = pmtu;
684 fnhe->fnhe_expires = expires;
686 /* Exception created; mark the cached routes for the nexthop
687 * stale, so anyone caching it rechecks if this exception
690 rt = rcu_dereference(nh->nh_rth_input);
692 rt->dst.obsolete = DST_OBSOLETE_KILL;
694 for_each_possible_cpu(i) {
695 struct rtable __rcu **prt;
696 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
697 rt = rcu_dereference(*prt);
699 rt->dst.obsolete = DST_OBSOLETE_KILL;
703 fnhe->fnhe_stamp = jiffies;
706 spin_unlock_bh(&fnhe_lock);
710 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
713 __be32 new_gw = icmp_hdr(skb)->un.gateway;
714 __be32 old_gw = ip_hdr(skb)->saddr;
715 struct net_device *dev = skb->dev;
716 struct in_device *in_dev;
717 struct fib_result res;
721 switch (icmp_hdr(skb)->code & 7) {
723 case ICMP_REDIR_NETTOS:
724 case ICMP_REDIR_HOST:
725 case ICMP_REDIR_HOSTTOS:
732 if (rt->rt_gateway != old_gw)
735 in_dev = __in_dev_get_rcu(dev);
740 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
741 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
742 ipv4_is_zeronet(new_gw))
743 goto reject_redirect;
745 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
746 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
747 goto reject_redirect;
748 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
749 goto reject_redirect;
751 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
752 goto reject_redirect;
755 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
757 if (!(n->nud_state & NUD_VALID)) {
758 neigh_event_send(n, NULL);
760 if (fib_lookup(net, fl4, &res) == 0) {
761 struct fib_nh *nh = &FIB_RES_NH(res);
763 update_or_create_fnhe(nh, fl4->daddr, new_gw,
767 rt->dst.obsolete = DST_OBSOLETE_KILL;
768 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
775 #ifdef CONFIG_IP_ROUTE_VERBOSE
776 if (IN_DEV_LOG_MARTIANS(in_dev)) {
777 const struct iphdr *iph = (const struct iphdr *) skb->data;
778 __be32 daddr = iph->daddr;
779 __be32 saddr = iph->saddr;
781 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
782 " Advised path = %pI4 -> %pI4\n",
783 &old_gw, dev->name, &new_gw,
790 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
794 const struct iphdr *iph = (const struct iphdr *) skb->data;
795 int oif = skb->dev->ifindex;
796 u8 tos = RT_TOS(iph->tos);
797 u8 prot = iph->protocol;
798 u32 mark = skb->mark;
800 rt = (struct rtable *) dst;
802 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
803 __ip_do_redirect(rt, skb, &fl4, true);
806 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
808 struct rtable *rt = (struct rtable *)dst;
809 struct dst_entry *ret = dst;
812 if (dst->obsolete > 0) {
815 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
826 * 1. The first ip_rt_redirect_number redirects are sent
827 * with exponential backoff, then we stop sending them at all,
828 * assuming that the host ignores our redirects.
829 * 2. If we did not see packets requiring redirects
830 * during ip_rt_redirect_silence, we assume that the host
831 * forgot redirected route and start to send redirects again.
833 * This algorithm is much cheaper and more intelligent than dumb load limiting
836 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
837 * and "frag. need" (breaks PMTU discovery) in icmp.c.
840 void ip_rt_send_redirect(struct sk_buff *skb)
842 struct rtable *rt = skb_rtable(skb);
843 struct in_device *in_dev;
844 struct inet_peer *peer;
849 in_dev = __in_dev_get_rcu(rt->dst.dev);
850 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
854 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
857 net = dev_net(rt->dst.dev);
858 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
860 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
861 rt_nexthop(rt, ip_hdr(skb)->daddr));
865 /* No redirected packets during ip_rt_redirect_silence;
866 * reset the algorithm.
868 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
869 peer->rate_tokens = 0;
871 /* Too many ignored redirects; do not send anything
872 * set dst.rate_last to the last seen redirected packet.
874 if (peer->rate_tokens >= ip_rt_redirect_number) {
875 peer->rate_last = jiffies;
879 /* Check for load limit; set rate_last to the latest sent
882 if (peer->rate_tokens == 0 ||
885 (ip_rt_redirect_load << peer->rate_tokens)))) {
886 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
888 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
889 peer->rate_last = jiffies;
891 #ifdef CONFIG_IP_ROUTE_VERBOSE
893 peer->rate_tokens == ip_rt_redirect_number)
894 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
895 &ip_hdr(skb)->saddr, inet_iif(skb),
896 &ip_hdr(skb)->daddr, &gw);
903 static int ip_error(struct sk_buff *skb)
905 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
906 struct rtable *rt = skb_rtable(skb);
907 struct inet_peer *peer;
913 net = dev_net(rt->dst.dev);
914 if (!IN_DEV_FORWARD(in_dev)) {
915 switch (rt->dst.error) {
917 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
921 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
927 switch (rt->dst.error) {
932 code = ICMP_HOST_UNREACH;
935 code = ICMP_NET_UNREACH;
936 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
939 code = ICMP_PKT_FILTERED;
943 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
948 peer->rate_tokens += now - peer->rate_last;
949 if (peer->rate_tokens > ip_rt_error_burst)
950 peer->rate_tokens = ip_rt_error_burst;
951 peer->rate_last = now;
952 if (peer->rate_tokens >= ip_rt_error_cost)
953 peer->rate_tokens -= ip_rt_error_cost;
959 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
965 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
967 struct dst_entry *dst = &rt->dst;
968 struct fib_result res;
970 if (dst_metric_locked(dst, RTAX_MTU))
973 if (dst->dev->mtu < mtu)
976 if (mtu < ip_rt_min_pmtu)
977 mtu = ip_rt_min_pmtu;
979 if (rt->rt_pmtu == mtu &&
980 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
984 if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
985 struct fib_nh *nh = &FIB_RES_NH(res);
987 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
988 jiffies + ip_rt_mtu_expires);
993 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
994 struct sk_buff *skb, u32 mtu)
996 struct rtable *rt = (struct rtable *) dst;
999 ip_rt_build_flow_key(&fl4, sk, skb);
1000 __ip_rt_update_pmtu(rt, &fl4, mtu);
1003 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1004 int oif, u32 mark, u8 protocol, int flow_flags)
1006 const struct iphdr *iph = (const struct iphdr *) skb->data;
1010 __build_flow_key(&fl4, NULL, iph, oif,
1011 RT_TOS(iph->tos), protocol, mark, flow_flags);
1012 rt = __ip_route_output_key(net, &fl4);
1014 __ip_rt_update_pmtu(rt, &fl4, mtu);
1018 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1020 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1022 const struct iphdr *iph = (const struct iphdr *) skb->data;
1026 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1027 rt = __ip_route_output_key(sock_net(sk), &fl4);
1029 __ip_rt_update_pmtu(rt, &fl4, mtu);
1034 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1036 const struct iphdr *iph = (const struct iphdr *) skb->data;
1039 struct dst_entry *odst = NULL;
1044 if (!ip_sk_accept_pmtu(sk))
1047 odst = sk_dst_get(sk);
1049 if (sock_owned_by_user(sk) || !odst) {
1050 __ipv4_sk_update_pmtu(skb, sk, mtu);
1054 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1056 rt = (struct rtable *)odst;
1057 if (odst->obsolete && odst->ops->check(odst, 0) == NULL) {
1058 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1065 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1067 if (!dst_check(&rt->dst, 0)) {
1069 dst_release(&rt->dst);
1071 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1079 sk_dst_set(sk, &rt->dst);
1085 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1087 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1088 int oif, u32 mark, u8 protocol, int flow_flags)
1090 const struct iphdr *iph = (const struct iphdr *) skb->data;
1094 __build_flow_key(&fl4, NULL, iph, oif,
1095 RT_TOS(iph->tos), protocol, mark, flow_flags);
1096 rt = __ip_route_output_key(net, &fl4);
1098 __ip_do_redirect(rt, skb, &fl4, false);
1102 EXPORT_SYMBOL_GPL(ipv4_redirect);
1104 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1106 const struct iphdr *iph = (const struct iphdr *) skb->data;
1110 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1111 rt = __ip_route_output_key(sock_net(sk), &fl4);
1113 __ip_do_redirect(rt, skb, &fl4, false);
1117 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1119 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1121 struct rtable *rt = (struct rtable *) dst;
1123 /* All IPV4 dsts are created with ->obsolete set to the value
1124 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1125 * into this function always.
1127 * When a PMTU/redirect information update invalidates a route,
1128 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1129 * DST_OBSOLETE_DEAD by dst_free().
1131 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1136 static void ipv4_link_failure(struct sk_buff *skb)
1140 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1142 rt = skb_rtable(skb);
1144 dst_set_expires(&rt->dst, 0);
1147 static int ip_rt_bug(struct sk_buff *skb)
1149 pr_debug("%s: %pI4 -> %pI4, %s\n",
1150 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1151 skb->dev ? skb->dev->name : "?");
1158 We do not cache source address of outgoing interface,
1159 because it is used only by IP RR, TS and SRR options,
1160 so that it out of fast path.
1162 BTW remember: "addr" is allowed to be not aligned
1166 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1170 if (rt_is_output_route(rt))
1171 src = ip_hdr(skb)->saddr;
1173 struct fib_result res;
1179 memset(&fl4, 0, sizeof(fl4));
1180 fl4.daddr = iph->daddr;
1181 fl4.saddr = iph->saddr;
1182 fl4.flowi4_tos = RT_TOS(iph->tos);
1183 fl4.flowi4_oif = rt->dst.dev->ifindex;
1184 fl4.flowi4_iif = skb->dev->ifindex;
1185 fl4.flowi4_mark = skb->mark;
1188 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1189 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1191 src = inet_select_addr(rt->dst.dev,
1192 rt_nexthop(rt, iph->daddr),
1196 memcpy(addr, &src, 4);
1199 #ifdef CONFIG_IP_ROUTE_CLASSID
1200 static void set_class_tag(struct rtable *rt, u32 tag)
1202 if (!(rt->dst.tclassid & 0xFFFF))
1203 rt->dst.tclassid |= tag & 0xFFFF;
1204 if (!(rt->dst.tclassid & 0xFFFF0000))
1205 rt->dst.tclassid |= tag & 0xFFFF0000;
1209 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1211 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1214 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1216 if (advmss > 65535 - 40)
1217 advmss = 65535 - 40;
1222 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1224 const struct rtable *rt = (const struct rtable *) dst;
1225 unsigned int mtu = rt->rt_pmtu;
1227 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1228 mtu = dst_metric_raw(dst, RTAX_MTU);
1233 mtu = dst->dev->mtu;
1235 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1236 if (rt->rt_uses_gateway && mtu > 576)
1240 return min_t(unsigned int, mtu, IP_MAX_MTU);
1243 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1245 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1246 struct fib_nh_exception *fnhe;
1252 hval = fnhe_hashfun(daddr);
1254 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1255 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1256 if (fnhe->fnhe_daddr == daddr)
1262 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1267 spin_lock_bh(&fnhe_lock);
1269 if (daddr == fnhe->fnhe_daddr) {
1270 struct rtable __rcu **porig;
1271 struct rtable *orig;
1272 int genid = fnhe_genid(dev_net(rt->dst.dev));
1274 if (rt_is_input_route(rt))
1275 porig = &fnhe->fnhe_rth_input;
1277 porig = &fnhe->fnhe_rth_output;
1278 orig = rcu_dereference(*porig);
1280 if (fnhe->fnhe_genid != genid) {
1281 fnhe->fnhe_genid = genid;
1283 fnhe->fnhe_pmtu = 0;
1284 fnhe->fnhe_expires = 0;
1285 fnhe_flush_routes(fnhe);
1288 fill_route_from_fnhe(rt, fnhe);
1289 if (!rt->rt_gateway)
1290 rt->rt_gateway = daddr;
1292 if (!(rt->dst.flags & DST_NOCACHE)) {
1293 rcu_assign_pointer(*porig, rt);
1299 fnhe->fnhe_stamp = jiffies;
1301 spin_unlock_bh(&fnhe_lock);
1306 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1308 struct rtable *orig, *prev, **p;
1311 if (rt_is_input_route(rt)) {
1312 p = (struct rtable **)&nh->nh_rth_input;
1314 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1318 prev = cmpxchg(p, orig, rt);
1328 static DEFINE_SPINLOCK(rt_uncached_lock);
1329 static LIST_HEAD(rt_uncached_list);
1331 static void rt_add_uncached_list(struct rtable *rt)
1333 spin_lock_bh(&rt_uncached_lock);
1334 list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1335 spin_unlock_bh(&rt_uncached_lock);
1338 static void ipv4_dst_destroy(struct dst_entry *dst)
1340 struct rtable *rt = (struct rtable *) dst;
1342 if (!list_empty(&rt->rt_uncached)) {
1343 spin_lock_bh(&rt_uncached_lock);
1344 list_del(&rt->rt_uncached);
1345 spin_unlock_bh(&rt_uncached_lock);
1349 void rt_flush_dev(struct net_device *dev)
1351 if (!list_empty(&rt_uncached_list)) {
1352 struct net *net = dev_net(dev);
1355 spin_lock_bh(&rt_uncached_lock);
1356 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1357 if (rt->dst.dev != dev)
1359 rt->dst.dev = net->loopback_dev;
1360 dev_hold(rt->dst.dev);
1363 spin_unlock_bh(&rt_uncached_lock);
1367 static bool rt_cache_valid(const struct rtable *rt)
1370 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1374 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1375 const struct fib_result *res,
1376 struct fib_nh_exception *fnhe,
1377 struct fib_info *fi, u16 type, u32 itag)
1379 bool cached = false;
1382 struct fib_nh *nh = &FIB_RES_NH(*res);
1384 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1385 rt->rt_gateway = nh->nh_gw;
1386 rt->rt_uses_gateway = 1;
1388 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1389 #ifdef CONFIG_IP_ROUTE_CLASSID
1390 rt->dst.tclassid = nh->nh_tclassid;
1393 cached = rt_bind_exception(rt, fnhe, daddr);
1394 else if (!(rt->dst.flags & DST_NOCACHE))
1395 cached = rt_cache_route(nh, rt);
1396 if (unlikely(!cached)) {
1397 /* Routes we intend to cache in nexthop exception or
1398 * FIB nexthop have the DST_NOCACHE bit clear.
1399 * However, if we are unsuccessful at storing this
1400 * route into the cache we really need to set it.
1402 rt->dst.flags |= DST_NOCACHE;
1403 if (!rt->rt_gateway)
1404 rt->rt_gateway = daddr;
1405 rt_add_uncached_list(rt);
1408 rt_add_uncached_list(rt);
1410 #ifdef CONFIG_IP_ROUTE_CLASSID
1411 #ifdef CONFIG_IP_MULTIPLE_TABLES
1412 set_class_tag(rt, res->tclassid);
1414 set_class_tag(rt, itag);
1418 static struct rtable *rt_dst_alloc(struct net_device *dev,
1419 bool nopolicy, bool noxfrm, bool will_cache)
1421 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1422 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1423 (nopolicy ? DST_NOPOLICY : 0) |
1424 (noxfrm ? DST_NOXFRM : 0));
1427 /* called in rcu_read_lock() section */
1428 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1429 u8 tos, struct net_device *dev, int our)
1432 struct in_device *in_dev = __in_dev_get_rcu(dev);
1436 /* Primary sanity checks. */
1441 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1442 skb->protocol != htons(ETH_P_IP))
1445 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1446 if (ipv4_is_loopback(saddr))
1449 if (ipv4_is_zeronet(saddr)) {
1450 if (!ipv4_is_local_multicast(daddr))
1453 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1458 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1459 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1463 #ifdef CONFIG_IP_ROUTE_CLASSID
1464 rth->dst.tclassid = itag;
1466 rth->dst.output = ip_rt_bug;
1468 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
1469 rth->rt_flags = RTCF_MULTICAST;
1470 rth->rt_type = RTN_MULTICAST;
1471 rth->rt_is_input= 1;
1474 rth->rt_gateway = 0;
1475 rth->rt_uses_gateway = 0;
1476 INIT_LIST_HEAD(&rth->rt_uncached);
1478 rth->dst.input= ip_local_deliver;
1479 rth->rt_flags |= RTCF_LOCAL;
1482 #ifdef CONFIG_IP_MROUTE
1483 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1484 rth->dst.input = ip_mr_input;
1486 RT_CACHE_STAT_INC(in_slow_mc);
1488 skb_dst_set(skb, &rth->dst);
1500 static void ip_handle_martian_source(struct net_device *dev,
1501 struct in_device *in_dev,
1502 struct sk_buff *skb,
1506 RT_CACHE_STAT_INC(in_martian_src);
1507 #ifdef CONFIG_IP_ROUTE_VERBOSE
1508 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1510 * RFC1812 recommendation, if source is martian,
1511 * the only hint is MAC header.
1513 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1514 &daddr, &saddr, dev->name);
1515 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1516 print_hex_dump(KERN_WARNING, "ll header: ",
1517 DUMP_PREFIX_OFFSET, 16, 1,
1518 skb_mac_header(skb),
1519 dev->hard_header_len, true);
1525 /* called in rcu_read_lock() section */
1526 static int __mkroute_input(struct sk_buff *skb,
1527 const struct fib_result *res,
1528 struct in_device *in_dev,
1529 __be32 daddr, __be32 saddr, u32 tos)
1531 struct fib_nh_exception *fnhe;
1534 struct in_device *out_dev;
1535 unsigned int flags = 0;
1539 /* get a working reference to the output device */
1540 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1541 if (out_dev == NULL) {
1542 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1546 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1547 in_dev->dev, in_dev, &itag);
1549 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1555 do_cache = res->fi && !itag;
1556 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1557 (IN_DEV_SHARED_MEDIA(out_dev) ||
1558 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1559 flags |= RTCF_DOREDIRECT;
1563 if (skb->protocol != htons(ETH_P_IP)) {
1564 /* Not IP (i.e. ARP). Do not create route, if it is
1565 * invalid for proxy arp. DNAT routes are always valid.
1567 * Proxy arp feature have been extended to allow, ARP
1568 * replies back to the same interface, to support
1569 * Private VLAN switch technologies. See arp.c.
1571 if (out_dev == in_dev &&
1572 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1578 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1581 rth = rcu_dereference(fnhe->fnhe_rth_input);
1583 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1585 if (rt_cache_valid(rth)) {
1586 skb_dst_set_noref(skb, &rth->dst);
1591 rth = rt_dst_alloc(out_dev->dev,
1592 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1593 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1599 rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1600 rth->rt_flags = flags;
1601 rth->rt_type = res->type;
1602 rth->rt_is_input = 1;
1605 rth->rt_gateway = 0;
1606 rth->rt_uses_gateway = 0;
1607 INIT_LIST_HEAD(&rth->rt_uncached);
1608 RT_CACHE_STAT_INC(in_slow_tot);
1610 rth->dst.input = ip_forward;
1611 rth->dst.output = ip_output;
1613 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1614 skb_dst_set(skb, &rth->dst);
1621 static int ip_mkroute_input(struct sk_buff *skb,
1622 struct fib_result *res,
1623 const struct flowi4 *fl4,
1624 struct in_device *in_dev,
1625 __be32 daddr, __be32 saddr, u32 tos)
1627 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1628 if (res->fi && res->fi->fib_nhs > 1)
1629 fib_select_multipath(res);
1632 /* create a routing cache entry */
1633 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1637 * NOTE. We drop all the packets that has local source
1638 * addresses, because every properly looped back packet
1639 * must have correct destination already attached by output routine.
1641 * Such approach solves two big problems:
1642 * 1. Not simplex devices are handled properly.
1643 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1644 * called with rcu_read_lock()
1647 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1648 u8 tos, struct net_device *dev)
1650 struct fib_result res;
1651 struct in_device *in_dev = __in_dev_get_rcu(dev);
1653 unsigned int flags = 0;
1657 struct net *net = dev_net(dev);
1660 /* IP on this device is disabled. */
1665 /* Check for the most weird martians, which can be not detected
1669 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1670 goto martian_source;
1673 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1676 /* Accept zero addresses only to limited broadcast;
1677 * I even do not know to fix it or not. Waiting for complains :-)
1679 if (ipv4_is_zeronet(saddr))
1680 goto martian_source;
1682 if (ipv4_is_zeronet(daddr))
1683 goto martian_destination;
1685 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1686 * and call it once if daddr or/and saddr are loopback addresses
1688 if (ipv4_is_loopback(daddr)) {
1689 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1690 goto martian_destination;
1691 } else if (ipv4_is_loopback(saddr)) {
1692 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1693 goto martian_source;
1697 * Now we are ready to route packet.
1700 fl4.flowi4_iif = dev->ifindex;
1701 fl4.flowi4_mark = skb->mark;
1702 fl4.flowi4_tos = tos;
1703 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1706 err = fib_lookup(net, &fl4, &res);
1708 if (!IN_DEV_FORWARD(in_dev))
1709 err = -EHOSTUNREACH;
1713 if (res.type == RTN_BROADCAST)
1716 if (res.type == RTN_LOCAL) {
1717 err = fib_validate_source(skb, saddr, daddr, tos,
1719 dev, in_dev, &itag);
1721 goto martian_source_keep_err;
1725 if (!IN_DEV_FORWARD(in_dev)) {
1726 err = -EHOSTUNREACH;
1729 if (res.type != RTN_UNICAST)
1730 goto martian_destination;
1732 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1736 if (skb->protocol != htons(ETH_P_IP))
1739 if (!ipv4_is_zeronet(saddr)) {
1740 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1743 goto martian_source_keep_err;
1745 flags |= RTCF_BROADCAST;
1746 res.type = RTN_BROADCAST;
1747 RT_CACHE_STAT_INC(in_brd);
1753 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1754 if (rt_cache_valid(rth)) {
1755 skb_dst_set_noref(skb, &rth->dst);
1763 rth = rt_dst_alloc(net->loopback_dev,
1764 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1768 rth->dst.input= ip_local_deliver;
1769 rth->dst.output= ip_rt_bug;
1770 #ifdef CONFIG_IP_ROUTE_CLASSID
1771 rth->dst.tclassid = itag;
1774 rth->rt_genid = rt_genid_ipv4(net);
1775 rth->rt_flags = flags|RTCF_LOCAL;
1776 rth->rt_type = res.type;
1777 rth->rt_is_input = 1;
1780 rth->rt_gateway = 0;
1781 rth->rt_uses_gateway = 0;
1782 INIT_LIST_HEAD(&rth->rt_uncached);
1783 RT_CACHE_STAT_INC(in_slow_tot);
1784 if (res.type == RTN_UNREACHABLE) {
1785 rth->dst.input= ip_error;
1786 rth->dst.error= -err;
1787 rth->rt_flags &= ~RTCF_LOCAL;
1790 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1791 rth->dst.flags |= DST_NOCACHE;
1792 rt_add_uncached_list(rth);
1795 skb_dst_set(skb, &rth->dst);
1800 RT_CACHE_STAT_INC(in_no_route);
1801 res.type = RTN_UNREACHABLE;
1807 * Do not cache martian addresses: they should be logged (RFC1812)
1809 martian_destination:
1810 RT_CACHE_STAT_INC(in_martian_dst);
1811 #ifdef CONFIG_IP_ROUTE_VERBOSE
1812 if (IN_DEV_LOG_MARTIANS(in_dev))
1813 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1814 &daddr, &saddr, dev->name);
1827 martian_source_keep_err:
1828 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1832 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1833 u8 tos, struct net_device *dev)
1839 /* Multicast recognition logic is moved from route cache to here.
1840 The problem was that too many Ethernet cards have broken/missing
1841 hardware multicast filters :-( As result the host on multicasting
1842 network acquires a lot of useless route cache entries, sort of
1843 SDR messages from all the world. Now we try to get rid of them.
1844 Really, provided software IP multicast filter is organized
1845 reasonably (at least, hashed), it does not result in a slowdown
1846 comparing with route cache reject entries.
1847 Note, that multicast routers are not affected, because
1848 route cache entry is created eventually.
1850 if (ipv4_is_multicast(daddr)) {
1851 struct in_device *in_dev = __in_dev_get_rcu(dev);
1854 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1855 ip_hdr(skb)->protocol);
1857 #ifdef CONFIG_IP_MROUTE
1859 (!ipv4_is_local_multicast(daddr) &&
1860 IN_DEV_MFORWARD(in_dev))
1863 int res = ip_route_input_mc(skb, daddr, saddr,
1872 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1876 EXPORT_SYMBOL(ip_route_input_noref);
1878 /* called with rcu_read_lock() */
1879 static struct rtable *__mkroute_output(const struct fib_result *res,
1880 const struct flowi4 *fl4, int orig_oif,
1881 struct net_device *dev_out,
1884 struct fib_info *fi = res->fi;
1885 struct fib_nh_exception *fnhe;
1886 struct in_device *in_dev;
1887 u16 type = res->type;
1891 in_dev = __in_dev_get_rcu(dev_out);
1893 return ERR_PTR(-EINVAL);
1895 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1896 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1897 return ERR_PTR(-EINVAL);
1899 if (ipv4_is_lbcast(fl4->daddr))
1900 type = RTN_BROADCAST;
1901 else if (ipv4_is_multicast(fl4->daddr))
1902 type = RTN_MULTICAST;
1903 else if (ipv4_is_zeronet(fl4->daddr))
1904 return ERR_PTR(-EINVAL);
1906 if (dev_out->flags & IFF_LOOPBACK)
1907 flags |= RTCF_LOCAL;
1910 if (type == RTN_BROADCAST) {
1911 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1913 } else if (type == RTN_MULTICAST) {
1914 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1915 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1917 flags &= ~RTCF_LOCAL;
1920 /* If multicast route do not exist use
1921 * default one, but do not gateway in this case.
1924 if (fi && res->prefixlen < 4)
1929 do_cache &= fi != NULL;
1931 struct rtable __rcu **prth;
1932 struct fib_nh *nh = &FIB_RES_NH(*res);
1934 fnhe = find_exception(nh, fl4->daddr);
1936 prth = &fnhe->fnhe_rth_output;
1938 if (unlikely(fl4->flowi4_flags &
1939 FLOWI_FLAG_KNOWN_NH &&
1941 nh->nh_scope == RT_SCOPE_LINK))) {
1945 prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1947 rth = rcu_dereference(*prth);
1948 if (rt_cache_valid(rth)) {
1949 dst_hold(&rth->dst);
1955 rth = rt_dst_alloc(dev_out,
1956 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1957 IN_DEV_CONF_GET(in_dev, NOXFRM),
1960 return ERR_PTR(-ENOBUFS);
1962 rth->dst.output = ip_output;
1964 rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1965 rth->rt_flags = flags;
1966 rth->rt_type = type;
1967 rth->rt_is_input = 0;
1968 rth->rt_iif = orig_oif ? : 0;
1970 rth->rt_gateway = 0;
1971 rth->rt_uses_gateway = 0;
1972 INIT_LIST_HEAD(&rth->rt_uncached);
1974 RT_CACHE_STAT_INC(out_slow_tot);
1976 if (flags & RTCF_LOCAL)
1977 rth->dst.input = ip_local_deliver;
1978 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1979 if (flags & RTCF_LOCAL &&
1980 !(dev_out->flags & IFF_LOOPBACK)) {
1981 rth->dst.output = ip_mc_output;
1982 RT_CACHE_STAT_INC(out_slow_mc);
1984 #ifdef CONFIG_IP_MROUTE
1985 if (type == RTN_MULTICAST) {
1986 if (IN_DEV_MFORWARD(in_dev) &&
1987 !ipv4_is_local_multicast(fl4->daddr)) {
1988 rth->dst.input = ip_mr_input;
1989 rth->dst.output = ip_mc_output;
1995 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2001 * Major route resolver routine.
2004 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2006 struct net_device *dev_out = NULL;
2007 __u8 tos = RT_FL_TOS(fl4);
2008 unsigned int flags = 0;
2009 struct fib_result res;
2017 orig_oif = fl4->flowi4_oif;
2019 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2020 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2021 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2022 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2026 rth = ERR_PTR(-EINVAL);
2027 if (ipv4_is_multicast(fl4->saddr) ||
2028 ipv4_is_lbcast(fl4->saddr) ||
2029 ipv4_is_zeronet(fl4->saddr))
2032 /* I removed check for oif == dev_out->oif here.
2033 It was wrong for two reasons:
2034 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2035 is assigned to multiple interfaces.
2036 2. Moreover, we are allowed to send packets with saddr
2037 of another iface. --ANK
2040 if (fl4->flowi4_oif == 0 &&
2041 (ipv4_is_multicast(fl4->daddr) ||
2042 ipv4_is_lbcast(fl4->daddr))) {
2043 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2044 dev_out = __ip_dev_find(net, fl4->saddr, false);
2045 if (dev_out == NULL)
2048 /* Special hack: user can direct multicasts
2049 and limited broadcast via necessary interface
2050 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2051 This hack is not just for fun, it allows
2052 vic,vat and friends to work.
2053 They bind socket to loopback, set ttl to zero
2054 and expect that it will work.
2055 From the viewpoint of routing cache they are broken,
2056 because we are not allowed to build multicast path
2057 with loopback source addr (look, routing cache
2058 cannot know, that ttl is zero, so that packet
2059 will not leave this host and route is valid).
2060 Luckily, this hack is good workaround.
2063 fl4->flowi4_oif = dev_out->ifindex;
2067 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2068 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2069 if (!__ip_dev_find(net, fl4->saddr, false))
2075 if (fl4->flowi4_oif) {
2076 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2077 rth = ERR_PTR(-ENODEV);
2078 if (dev_out == NULL)
2081 /* RACE: Check return value of inet_select_addr instead. */
2082 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2083 rth = ERR_PTR(-ENETUNREACH);
2086 if (ipv4_is_local_multicast(fl4->daddr) ||
2087 ipv4_is_lbcast(fl4->daddr)) {
2089 fl4->saddr = inet_select_addr(dev_out, 0,
2094 if (ipv4_is_multicast(fl4->daddr))
2095 fl4->saddr = inet_select_addr(dev_out, 0,
2097 else if (!fl4->daddr)
2098 fl4->saddr = inet_select_addr(dev_out, 0,
2104 fl4->daddr = fl4->saddr;
2106 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2107 dev_out = net->loopback_dev;
2108 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2109 res.type = RTN_LOCAL;
2110 flags |= RTCF_LOCAL;
2114 if (fib_lookup(net, fl4, &res)) {
2117 if (fl4->flowi4_oif) {
2118 /* Apparently, routing tables are wrong. Assume,
2119 that the destination is on link.
2122 Because we are allowed to send to iface
2123 even if it has NO routes and NO assigned
2124 addresses. When oif is specified, routing
2125 tables are looked up with only one purpose:
2126 to catch if destination is gatewayed, rather than
2127 direct. Moreover, if MSG_DONTROUTE is set,
2128 we send packet, ignoring both routing tables
2129 and ifaddr state. --ANK
2132 We could make it even if oif is unknown,
2133 likely IPv6, but we do not.
2136 if (fl4->saddr == 0)
2137 fl4->saddr = inet_select_addr(dev_out, 0,
2139 res.type = RTN_UNICAST;
2142 rth = ERR_PTR(-ENETUNREACH);
2146 if (res.type == RTN_LOCAL) {
2148 if (res.fi->fib_prefsrc)
2149 fl4->saddr = res.fi->fib_prefsrc;
2151 fl4->saddr = fl4->daddr;
2153 dev_out = net->loopback_dev;
2154 fl4->flowi4_oif = dev_out->ifindex;
2155 flags |= RTCF_LOCAL;
2159 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2160 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2161 fib_select_multipath(&res);
2164 if (!res.prefixlen &&
2165 res.table->tb_num_default > 1 &&
2166 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2167 fib_select_default(&res);
2170 fl4->saddr = FIB_RES_PREFSRC(net, res);
2172 dev_out = FIB_RES_DEV(res);
2173 fl4->flowi4_oif = dev_out->ifindex;
2177 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2183 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2185 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2190 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2192 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2194 return mtu ? : dst->dev->mtu;
2197 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2198 struct sk_buff *skb, u32 mtu)
2202 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2203 struct sk_buff *skb)
2207 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2213 static struct dst_ops ipv4_dst_blackhole_ops = {
2215 .protocol = cpu_to_be16(ETH_P_IP),
2216 .check = ipv4_blackhole_dst_check,
2217 .mtu = ipv4_blackhole_mtu,
2218 .default_advmss = ipv4_default_advmss,
2219 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2220 .redirect = ipv4_rt_blackhole_redirect,
2221 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2222 .neigh_lookup = ipv4_neigh_lookup,
2225 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2227 struct rtable *ort = (struct rtable *) dst_orig;
2230 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2232 struct dst_entry *new = &rt->dst;
2235 new->input = dst_discard;
2236 new->output = dst_discard;
2238 new->dev = ort->dst.dev;
2242 rt->rt_is_input = ort->rt_is_input;
2243 rt->rt_iif = ort->rt_iif;
2244 rt->rt_pmtu = ort->rt_pmtu;
2246 rt->rt_genid = rt_genid_ipv4(net);
2247 rt->rt_flags = ort->rt_flags;
2248 rt->rt_type = ort->rt_type;
2249 rt->rt_gateway = ort->rt_gateway;
2250 rt->rt_uses_gateway = ort->rt_uses_gateway;
2252 INIT_LIST_HEAD(&rt->rt_uncached);
2257 dst_release(dst_orig);
2259 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2262 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2265 struct rtable *rt = __ip_route_output_key(net, flp4);
2270 if (flp4->flowi4_proto)
2271 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2272 flowi4_to_flowi(flp4),
2277 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2279 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2280 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2281 u32 seq, int event, int nowait, unsigned int flags)
2283 struct rtable *rt = skb_rtable(skb);
2285 struct nlmsghdr *nlh;
2286 unsigned long expires = 0;
2288 u32 metrics[RTAX_MAX];
2290 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2294 r = nlmsg_data(nlh);
2295 r->rtm_family = AF_INET;
2296 r->rtm_dst_len = 32;
2298 r->rtm_tos = fl4->flowi4_tos;
2299 r->rtm_table = RT_TABLE_MAIN;
2300 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2301 goto nla_put_failure;
2302 r->rtm_type = rt->rt_type;
2303 r->rtm_scope = RT_SCOPE_UNIVERSE;
2304 r->rtm_protocol = RTPROT_UNSPEC;
2305 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2306 if (rt->rt_flags & RTCF_NOTIFY)
2307 r->rtm_flags |= RTM_F_NOTIFY;
2309 if (nla_put_be32(skb, RTA_DST, dst))
2310 goto nla_put_failure;
2312 r->rtm_src_len = 32;
2313 if (nla_put_be32(skb, RTA_SRC, src))
2314 goto nla_put_failure;
2317 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2318 goto nla_put_failure;
2319 #ifdef CONFIG_IP_ROUTE_CLASSID
2320 if (rt->dst.tclassid &&
2321 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2322 goto nla_put_failure;
2324 if (!rt_is_input_route(rt) &&
2325 fl4->saddr != src) {
2326 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2327 goto nla_put_failure;
2329 if (rt->rt_uses_gateway &&
2330 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2331 goto nla_put_failure;
2333 expires = rt->dst.expires;
2335 unsigned long now = jiffies;
2337 if (time_before(now, expires))
2343 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2344 if (rt->rt_pmtu && expires)
2345 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2346 if (rtnetlink_put_metrics(skb, metrics) < 0)
2347 goto nla_put_failure;
2349 if (fl4->flowi4_mark &&
2350 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2351 goto nla_put_failure;
2353 error = rt->dst.error;
2355 if (rt_is_input_route(rt)) {
2356 #ifdef CONFIG_IP_MROUTE
2357 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2358 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2359 int err = ipmr_get_route(net, skb,
2360 fl4->saddr, fl4->daddr,
2366 goto nla_put_failure;
2368 if (err == -EMSGSIZE)
2369 goto nla_put_failure;
2375 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2376 goto nla_put_failure;
2379 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2380 goto nla_put_failure;
2382 return nlmsg_end(skb, nlh);
2385 nlmsg_cancel(skb, nlh);
2389 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2391 struct net *net = sock_net(in_skb->sk);
2393 struct nlattr *tb[RTA_MAX+1];
2394 struct rtable *rt = NULL;
2401 struct sk_buff *skb;
2403 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2407 rtm = nlmsg_data(nlh);
2409 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2415 /* Reserve room for dummy headers, this skb can pass
2416 through good chunk of routing engine.
2418 skb_reset_mac_header(skb);
2419 skb_reset_network_header(skb);
2421 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2422 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2423 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2425 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2426 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2427 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2428 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2430 memset(&fl4, 0, sizeof(fl4));
2433 fl4.flowi4_tos = rtm->rtm_tos;
2434 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2435 fl4.flowi4_mark = mark;
2438 struct net_device *dev;
2440 dev = __dev_get_by_index(net, iif);
2446 skb->protocol = htons(ETH_P_IP);
2450 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2453 rt = skb_rtable(skb);
2454 if (err == 0 && rt->dst.error)
2455 err = -rt->dst.error;
2457 rt = ip_route_output_key(net, &fl4);
2467 skb_dst_set(skb, &rt->dst);
2468 if (rtm->rtm_flags & RTM_F_NOTIFY)
2469 rt->rt_flags |= RTCF_NOTIFY;
2471 err = rt_fill_info(net, dst, src, &fl4, skb,
2472 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2473 RTM_NEWROUTE, 0, 0);
2477 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2486 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2491 void ip_rt_multicast_event(struct in_device *in_dev)
2493 rt_cache_flush(dev_net(in_dev->dev));
2496 #ifdef CONFIG_SYSCTL
2497 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2498 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2499 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2500 static int ip_rt_gc_elasticity __read_mostly = 8;
2502 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2503 void __user *buffer,
2504 size_t *lenp, loff_t *ppos)
2506 struct net *net = (struct net *)__ctl->extra1;
2509 rt_cache_flush(net);
2510 fnhe_genid_bump(net);
2517 static struct ctl_table ipv4_route_table[] = {
2519 .procname = "gc_thresh",
2520 .data = &ipv4_dst_ops.gc_thresh,
2521 .maxlen = sizeof(int),
2523 .proc_handler = proc_dointvec,
2526 .procname = "max_size",
2527 .data = &ip_rt_max_size,
2528 .maxlen = sizeof(int),
2530 .proc_handler = proc_dointvec,
2533 /* Deprecated. Use gc_min_interval_ms */
2535 .procname = "gc_min_interval",
2536 .data = &ip_rt_gc_min_interval,
2537 .maxlen = sizeof(int),
2539 .proc_handler = proc_dointvec_jiffies,
2542 .procname = "gc_min_interval_ms",
2543 .data = &ip_rt_gc_min_interval,
2544 .maxlen = sizeof(int),
2546 .proc_handler = proc_dointvec_ms_jiffies,
2549 .procname = "gc_timeout",
2550 .data = &ip_rt_gc_timeout,
2551 .maxlen = sizeof(int),
2553 .proc_handler = proc_dointvec_jiffies,
2556 .procname = "gc_interval",
2557 .data = &ip_rt_gc_interval,
2558 .maxlen = sizeof(int),
2560 .proc_handler = proc_dointvec_jiffies,
2563 .procname = "redirect_load",
2564 .data = &ip_rt_redirect_load,
2565 .maxlen = sizeof(int),
2567 .proc_handler = proc_dointvec,
2570 .procname = "redirect_number",
2571 .data = &ip_rt_redirect_number,
2572 .maxlen = sizeof(int),
2574 .proc_handler = proc_dointvec,
2577 .procname = "redirect_silence",
2578 .data = &ip_rt_redirect_silence,
2579 .maxlen = sizeof(int),
2581 .proc_handler = proc_dointvec,
2584 .procname = "error_cost",
2585 .data = &ip_rt_error_cost,
2586 .maxlen = sizeof(int),
2588 .proc_handler = proc_dointvec,
2591 .procname = "error_burst",
2592 .data = &ip_rt_error_burst,
2593 .maxlen = sizeof(int),
2595 .proc_handler = proc_dointvec,
2598 .procname = "gc_elasticity",
2599 .data = &ip_rt_gc_elasticity,
2600 .maxlen = sizeof(int),
2602 .proc_handler = proc_dointvec,
2605 .procname = "mtu_expires",
2606 .data = &ip_rt_mtu_expires,
2607 .maxlen = sizeof(int),
2609 .proc_handler = proc_dointvec_jiffies,
2612 .procname = "min_pmtu",
2613 .data = &ip_rt_min_pmtu,
2614 .maxlen = sizeof(int),
2616 .proc_handler = proc_dointvec,
2619 .procname = "min_adv_mss",
2620 .data = &ip_rt_min_advmss,
2621 .maxlen = sizeof(int),
2623 .proc_handler = proc_dointvec,
2628 static struct ctl_table ipv4_route_flush_table[] = {
2630 .procname = "flush",
2631 .maxlen = sizeof(int),
2633 .proc_handler = ipv4_sysctl_rtcache_flush,
2638 static __net_init int sysctl_route_net_init(struct net *net)
2640 struct ctl_table *tbl;
2642 tbl = ipv4_route_flush_table;
2643 if (!net_eq(net, &init_net)) {
2644 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2648 /* Don't export sysctls to unprivileged users */
2649 if (net->user_ns != &init_user_ns)
2650 tbl[0].procname = NULL;
2652 tbl[0].extra1 = net;
2654 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2655 if (net->ipv4.route_hdr == NULL)
2660 if (tbl != ipv4_route_flush_table)
2666 static __net_exit void sysctl_route_net_exit(struct net *net)
2668 struct ctl_table *tbl;
2670 tbl = net->ipv4.route_hdr->ctl_table_arg;
2671 unregister_net_sysctl_table(net->ipv4.route_hdr);
2672 BUG_ON(tbl == ipv4_route_flush_table);
2676 static __net_initdata struct pernet_operations sysctl_route_ops = {
2677 .init = sysctl_route_net_init,
2678 .exit = sysctl_route_net_exit,
2682 static __net_init int rt_genid_init(struct net *net)
2684 atomic_set(&net->ipv4.rt_genid, 0);
2685 atomic_set(&net->fnhe_genid, 0);
2686 get_random_bytes(&net->ipv4.dev_addr_genid,
2687 sizeof(net->ipv4.dev_addr_genid));
2691 static __net_initdata struct pernet_operations rt_genid_ops = {
2692 .init = rt_genid_init,
2695 static int __net_init ipv4_inetpeer_init(struct net *net)
2697 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2701 inet_peer_base_init(bp);
2702 net->ipv4.peers = bp;
2706 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2708 struct inet_peer_base *bp = net->ipv4.peers;
2710 net->ipv4.peers = NULL;
2711 inetpeer_invalidate_tree(bp);
2715 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2716 .init = ipv4_inetpeer_init,
2717 .exit = ipv4_inetpeer_exit,
2720 #ifdef CONFIG_IP_ROUTE_CLASSID
2721 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2722 #endif /* CONFIG_IP_ROUTE_CLASSID */
2724 int __init ip_rt_init(void)
2728 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2730 panic("IP: failed to allocate ip_idents\n");
2732 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2734 #ifdef CONFIG_IP_ROUTE_CLASSID
2735 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2737 panic("IP: failed to allocate ip_rt_acct\n");
2740 ipv4_dst_ops.kmem_cachep =
2741 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2742 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2744 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2746 if (dst_entries_init(&ipv4_dst_ops) < 0)
2747 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2749 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2750 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2752 ipv4_dst_ops.gc_thresh = ~0;
2753 ip_rt_max_size = INT_MAX;
2758 if (ip_rt_proc_init())
2759 pr_err("Unable to create route proc files\n");
2764 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2766 #ifdef CONFIG_SYSCTL
2767 register_pernet_subsys(&sysctl_route_ops);
2769 register_pernet_subsys(&rt_genid_ops);
2770 register_pernet_subsys(&ipv4_inetpeer_ops);
2774 #ifdef CONFIG_SYSCTL
2776 * We really need to sanitize the damn ipv4 init order, then all
2777 * this nonsense will go away.
2779 void __init ip_static_sysctl_init(void)
2781 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);