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>
93 #include <net/net_namespace.h>
94 #include <net/protocol.h>
96 #include <net/route.h>
97 #include <net/inetpeer.h>
99 #include <net/ip_fib.h>
102 #include <net/icmp.h>
103 #include <net/xfrm.h>
104 #include <net/netevent.h>
105 #include <net/rtnetlink.h>
107 #include <linux/sysctl.h>
108 #include <linux/kmemleak.h>
110 #include <net/secure_seq.h>
112 #define RT_FL_TOS(oldflp4) \
113 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
115 /* IPv4 datagram length is stored into 16bit field (tot_len) */
116 #define IP_MAX_MTU 0xFFFF
118 #define RT_GC_TIMEOUT (300*HZ)
120 static int ip_rt_max_size;
121 static int ip_rt_redirect_number __read_mostly = 9;
122 static int ip_rt_redirect_load __read_mostly = HZ / 50;
123 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
124 static int ip_rt_error_cost __read_mostly = HZ;
125 static int ip_rt_error_burst __read_mostly = 5 * HZ;
126 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
127 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
128 static int ip_rt_min_advmss __read_mostly = 256;
131 * Interface to generic destination cache.
134 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
135 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
136 static unsigned int ipv4_mtu(const struct dst_entry *dst);
137 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
138 static void ipv4_link_failure(struct sk_buff *skb);
139 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
140 struct sk_buff *skb, u32 mtu);
141 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb);
143 static void ipv4_dst_destroy(struct dst_entry *dst);
145 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
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,
160 static struct dst_ops ipv4_dst_ops = {
162 .protocol = cpu_to_be16(ETH_P_IP),
163 .check = ipv4_dst_check,
164 .default_advmss = ipv4_default_advmss,
166 .cow_metrics = ipv4_cow_metrics,
167 .destroy = ipv4_dst_destroy,
168 .ifdown = ipv4_dst_ifdown,
169 .negative_advice = ipv4_negative_advice,
170 .link_failure = ipv4_link_failure,
171 .update_pmtu = ip_rt_update_pmtu,
172 .redirect = ip_do_redirect,
173 .local_out = __ip_local_out,
174 .neigh_lookup = ipv4_neigh_lookup,
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) __this_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 .owner = THIS_MODULE,
244 .open = rt_cache_seq_open,
247 .release = seq_release,
251 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
256 return SEQ_START_TOKEN;
258 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
259 if (!cpu_possible(cpu))
262 return &per_cpu(rt_cache_stat, cpu);
267 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
271 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
272 if (!cpu_possible(cpu))
275 return &per_cpu(rt_cache_stat, cpu);
281 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
286 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
288 struct rt_cache_stat *st = v;
290 if (v == SEQ_START_TOKEN) {
291 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");
295 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
296 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
297 dst_entries_get_slow(&ipv4_dst_ops),
310 0, /* st->gc_total */
311 0, /* st->gc_ignored */
312 0, /* st->gc_goal_miss */
313 0, /* st->gc_dst_overflow */
314 0, /* st->in_hlist_search */
315 0 /* st->out_hlist_search */
320 static const struct seq_operations rt_cpu_seq_ops = {
321 .start = rt_cpu_seq_start,
322 .next = rt_cpu_seq_next,
323 .stop = rt_cpu_seq_stop,
324 .show = rt_cpu_seq_show,
328 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
330 return seq_open(file, &rt_cpu_seq_ops);
333 static const struct file_operations rt_cpu_seq_fops = {
334 .owner = THIS_MODULE,
335 .open = rt_cpu_seq_open,
338 .release = seq_release,
341 #ifdef CONFIG_IP_ROUTE_CLASSID
342 static int rt_acct_proc_show(struct seq_file *m, void *v)
344 struct ip_rt_acct *dst, *src;
347 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
351 for_each_possible_cpu(i) {
352 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
353 for (j = 0; j < 256; j++) {
354 dst[j].o_bytes += src[j].o_bytes;
355 dst[j].o_packets += src[j].o_packets;
356 dst[j].i_bytes += src[j].i_bytes;
357 dst[j].i_packets += src[j].i_packets;
361 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
366 static int rt_acct_proc_open(struct inode *inode, struct file *file)
368 return single_open(file, rt_acct_proc_show, NULL);
371 static const struct file_operations rt_acct_proc_fops = {
372 .owner = THIS_MODULE,
373 .open = rt_acct_proc_open,
376 .release = single_release,
380 static int __net_init ip_rt_do_proc_init(struct net *net)
382 struct proc_dir_entry *pde;
384 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
389 pde = proc_create("rt_cache", S_IRUGO,
390 net->proc_net_stat, &rt_cpu_seq_fops);
394 #ifdef CONFIG_IP_ROUTE_CLASSID
395 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
401 #ifdef CONFIG_IP_ROUTE_CLASSID
403 remove_proc_entry("rt_cache", net->proc_net_stat);
406 remove_proc_entry("rt_cache", net->proc_net);
411 static void __net_exit ip_rt_do_proc_exit(struct net *net)
413 remove_proc_entry("rt_cache", net->proc_net_stat);
414 remove_proc_entry("rt_cache", net->proc_net);
415 #ifdef CONFIG_IP_ROUTE_CLASSID
416 remove_proc_entry("rt_acct", net->proc_net);
420 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
421 .init = ip_rt_do_proc_init,
422 .exit = ip_rt_do_proc_exit,
425 static int __init ip_rt_proc_init(void)
427 return register_pernet_subsys(&ip_rt_proc_ops);
431 static inline int ip_rt_proc_init(void)
435 #endif /* CONFIG_PROC_FS */
437 static inline bool rt_is_expired(const struct rtable *rth)
439 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
442 void rt_cache_flush(struct net *net)
444 rt_genid_bump_ipv4(net);
447 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
451 struct net_device *dev = dst->dev;
452 const __be32 *pkey = daddr;
453 const struct rtable *rt;
456 rt = (const struct rtable *) dst;
458 pkey = (const __be32 *) &rt->rt_gateway;
460 pkey = &ip_hdr(skb)->daddr;
462 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
465 return neigh_create(&arp_tbl, pkey, dev);
469 * Peer allocation may fail only in serious out-of-memory conditions. However
470 * we still can generate some output.
471 * Random ID selection looks a bit dangerous because we have no chances to
472 * select ID being unique in a reasonable period of time.
473 * But broken packet identifier may be better than no packet at all.
475 static void ip_select_fb_ident(struct iphdr *iph)
477 static DEFINE_SPINLOCK(ip_fb_id_lock);
478 static u32 ip_fallback_id;
481 spin_lock_bh(&ip_fb_id_lock);
482 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
483 iph->id = htons(salt & 0xFFFF);
484 ip_fallback_id = salt;
485 spin_unlock_bh(&ip_fb_id_lock);
488 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
490 struct net *net = dev_net(dst->dev);
491 struct inet_peer *peer;
493 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
495 iph->id = htons(inet_getid(peer, more));
500 ip_select_fb_ident(iph);
502 EXPORT_SYMBOL(__ip_select_ident);
504 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
505 const struct iphdr *iph,
507 u8 prot, u32 mark, int flow_flags)
510 const struct inet_sock *inet = inet_sk(sk);
512 oif = sk->sk_bound_dev_if;
514 tos = RT_CONN_FLAGS(sk);
515 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
517 flowi4_init_output(fl4, oif, mark, tos,
518 RT_SCOPE_UNIVERSE, prot,
520 iph->daddr, iph->saddr, 0, 0);
523 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
524 const struct sock *sk)
526 const struct iphdr *iph = ip_hdr(skb);
527 int oif = skb->dev->ifindex;
528 u8 tos = RT_TOS(iph->tos);
529 u8 prot = iph->protocol;
530 u32 mark = skb->mark;
532 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
535 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
537 const struct inet_sock *inet = inet_sk(sk);
538 const struct ip_options_rcu *inet_opt;
539 __be32 daddr = inet->inet_daddr;
542 inet_opt = rcu_dereference(inet->inet_opt);
543 if (inet_opt && inet_opt->opt.srr)
544 daddr = inet_opt->opt.faddr;
545 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
546 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
547 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
548 inet_sk_flowi_flags(sk),
549 daddr, inet->inet_saddr, 0, 0);
553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
554 const struct sk_buff *skb)
557 build_skb_flow_key(fl4, skb, sk);
559 build_sk_flow_key(fl4, sk);
562 static inline void rt_free(struct rtable *rt)
564 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
567 static DEFINE_SPINLOCK(fnhe_lock);
569 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
573 rt = rcu_dereference(fnhe->fnhe_rth_input);
575 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
578 rt = rcu_dereference(fnhe->fnhe_rth_output);
580 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
585 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
587 struct fib_nh_exception *fnhe, *oldest;
589 oldest = rcu_dereference(hash->chain);
590 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
591 fnhe = rcu_dereference(fnhe->fnhe_next)) {
592 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
595 fnhe_flush_routes(oldest);
599 static inline u32 fnhe_hashfun(__be32 daddr)
603 hval = (__force u32) daddr;
604 hval ^= (hval >> 11) ^ (hval >> 22);
606 return hval & (FNHE_HASH_SIZE - 1);
609 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
611 rt->rt_pmtu = fnhe->fnhe_pmtu;
612 rt->dst.expires = fnhe->fnhe_expires;
615 rt->rt_flags |= RTCF_REDIRECTED;
616 rt->rt_gateway = fnhe->fnhe_gw;
617 rt->rt_uses_gateway = 1;
621 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
622 u32 pmtu, unsigned long expires)
624 struct fnhe_hash_bucket *hash;
625 struct fib_nh_exception *fnhe;
629 u32 hval = fnhe_hashfun(daddr);
631 spin_lock_bh(&fnhe_lock);
633 hash = nh->nh_exceptions;
635 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
638 nh->nh_exceptions = hash;
644 for (fnhe = rcu_dereference(hash->chain); fnhe;
645 fnhe = rcu_dereference(fnhe->fnhe_next)) {
646 if (fnhe->fnhe_daddr == daddr)
655 fnhe->fnhe_pmtu = pmtu;
656 fnhe->fnhe_expires = max(1UL, expires);
658 /* Update all cached dsts too */
659 rt = rcu_dereference(fnhe->fnhe_rth_input);
661 fill_route_from_fnhe(rt, fnhe);
662 rt = rcu_dereference(fnhe->fnhe_rth_output);
664 fill_route_from_fnhe(rt, fnhe);
666 if (depth > FNHE_RECLAIM_DEPTH)
667 fnhe = fnhe_oldest(hash);
669 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
673 fnhe->fnhe_next = hash->chain;
674 rcu_assign_pointer(hash->chain, fnhe);
676 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
677 fnhe->fnhe_daddr = daddr;
679 fnhe->fnhe_pmtu = pmtu;
680 fnhe->fnhe_expires = expires;
682 /* Exception created; mark the cached routes for the nexthop
683 * stale, so anyone caching it rechecks if this exception
686 rt = rcu_dereference(nh->nh_rth_input);
688 rt->dst.obsolete = DST_OBSOLETE_KILL;
690 for_each_possible_cpu(i) {
691 struct rtable __rcu **prt;
692 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
693 rt = rcu_dereference(*prt);
695 rt->dst.obsolete = DST_OBSOLETE_KILL;
699 fnhe->fnhe_stamp = jiffies;
702 spin_unlock_bh(&fnhe_lock);
706 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
709 __be32 new_gw = icmp_hdr(skb)->un.gateway;
710 __be32 old_gw = ip_hdr(skb)->saddr;
711 struct net_device *dev = skb->dev;
712 struct in_device *in_dev;
713 struct fib_result res;
717 switch (icmp_hdr(skb)->code & 7) {
719 case ICMP_REDIR_NETTOS:
720 case ICMP_REDIR_HOST:
721 case ICMP_REDIR_HOSTTOS:
728 if (rt->rt_gateway != old_gw)
731 in_dev = __in_dev_get_rcu(dev);
736 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
737 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
738 ipv4_is_zeronet(new_gw))
739 goto reject_redirect;
741 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
742 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
743 goto reject_redirect;
744 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
745 goto reject_redirect;
747 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
748 goto reject_redirect;
751 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
753 if (!(n->nud_state & NUD_VALID)) {
754 neigh_event_send(n, NULL);
756 if (fib_lookup(net, fl4, &res) == 0) {
757 struct fib_nh *nh = &FIB_RES_NH(res);
759 update_or_create_fnhe(nh, fl4->daddr, new_gw,
763 rt->dst.obsolete = DST_OBSOLETE_KILL;
764 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
771 #ifdef CONFIG_IP_ROUTE_VERBOSE
772 if (IN_DEV_LOG_MARTIANS(in_dev)) {
773 const struct iphdr *iph = (const struct iphdr *) skb->data;
774 __be32 daddr = iph->daddr;
775 __be32 saddr = iph->saddr;
777 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
778 " Advised path = %pI4 -> %pI4\n",
779 &old_gw, dev->name, &new_gw,
786 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
790 const struct iphdr *iph = (const struct iphdr *) skb->data;
791 int oif = skb->dev->ifindex;
792 u8 tos = RT_TOS(iph->tos);
793 u8 prot = iph->protocol;
794 u32 mark = skb->mark;
796 rt = (struct rtable *) dst;
798 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
799 __ip_do_redirect(rt, skb, &fl4, true);
802 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
804 struct rtable *rt = (struct rtable *)dst;
805 struct dst_entry *ret = dst;
808 if (dst->obsolete > 0) {
811 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
822 * 1. The first ip_rt_redirect_number redirects are sent
823 * with exponential backoff, then we stop sending them at all,
824 * assuming that the host ignores our redirects.
825 * 2. If we did not see packets requiring redirects
826 * during ip_rt_redirect_silence, we assume that the host
827 * forgot redirected route and start to send redirects again.
829 * This algorithm is much cheaper and more intelligent than dumb load limiting
832 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
833 * and "frag. need" (breaks PMTU discovery) in icmp.c.
836 void ip_rt_send_redirect(struct sk_buff *skb)
838 struct rtable *rt = skb_rtable(skb);
839 struct in_device *in_dev;
840 struct inet_peer *peer;
845 in_dev = __in_dev_get_rcu(rt->dst.dev);
846 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
850 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
853 net = dev_net(rt->dst.dev);
854 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
856 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
857 rt_nexthop(rt, ip_hdr(skb)->daddr));
861 /* No redirected packets during ip_rt_redirect_silence;
862 * reset the algorithm.
864 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
865 peer->rate_tokens = 0;
867 /* Too many ignored redirects; do not send anything
868 * set dst.rate_last to the last seen redirected packet.
870 if (peer->rate_tokens >= ip_rt_redirect_number) {
871 peer->rate_last = jiffies;
875 /* Check for load limit; set rate_last to the latest sent
878 if (peer->rate_tokens == 0 ||
881 (ip_rt_redirect_load << peer->rate_tokens)))) {
882 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
884 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
885 peer->rate_last = jiffies;
887 #ifdef CONFIG_IP_ROUTE_VERBOSE
889 peer->rate_tokens == ip_rt_redirect_number)
890 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
891 &ip_hdr(skb)->saddr, inet_iif(skb),
892 &ip_hdr(skb)->daddr, &gw);
899 static int ip_error(struct sk_buff *skb)
901 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
902 struct rtable *rt = skb_rtable(skb);
903 struct inet_peer *peer;
909 net = dev_net(rt->dst.dev);
910 if (!IN_DEV_FORWARD(in_dev)) {
911 switch (rt->dst.error) {
913 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
917 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
923 switch (rt->dst.error) {
928 code = ICMP_HOST_UNREACH;
931 code = ICMP_NET_UNREACH;
932 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
935 code = ICMP_PKT_FILTERED;
939 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
944 peer->rate_tokens += now - peer->rate_last;
945 if (peer->rate_tokens > ip_rt_error_burst)
946 peer->rate_tokens = ip_rt_error_burst;
947 peer->rate_last = now;
948 if (peer->rate_tokens >= ip_rt_error_cost)
949 peer->rate_tokens -= ip_rt_error_cost;
955 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
961 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
963 struct dst_entry *dst = &rt->dst;
964 struct fib_result res;
966 if (dst_metric_locked(dst, RTAX_MTU))
969 if (dst->dev->mtu < mtu)
972 if (mtu < ip_rt_min_pmtu)
973 mtu = ip_rt_min_pmtu;
975 if (rt->rt_pmtu == mtu &&
976 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
980 if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
981 struct fib_nh *nh = &FIB_RES_NH(res);
983 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
984 jiffies + ip_rt_mtu_expires);
989 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
990 struct sk_buff *skb, u32 mtu)
992 struct rtable *rt = (struct rtable *) dst;
995 ip_rt_build_flow_key(&fl4, sk, skb);
996 __ip_rt_update_pmtu(rt, &fl4, mtu);
999 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1000 int oif, u32 mark, u8 protocol, int flow_flags)
1002 const struct iphdr *iph = (const struct iphdr *) skb->data;
1006 __build_flow_key(&fl4, NULL, iph, oif,
1007 RT_TOS(iph->tos), protocol, mark, flow_flags);
1008 rt = __ip_route_output_key(net, &fl4);
1010 __ip_rt_update_pmtu(rt, &fl4, mtu);
1014 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1016 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1018 const struct iphdr *iph = (const struct iphdr *) skb->data;
1022 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1023 rt = __ip_route_output_key(sock_net(sk), &fl4);
1025 __ip_rt_update_pmtu(rt, &fl4, mtu);
1030 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1032 const struct iphdr *iph = (const struct iphdr *) skb->data;
1035 struct dst_entry *dst;
1040 if (!ip_sk_accept_pmtu(sk))
1043 rt = (struct rtable *) __sk_dst_get(sk);
1045 if (sock_owned_by_user(sk) || !rt) {
1046 __ipv4_sk_update_pmtu(skb, sk, mtu);
1050 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1052 if (!__sk_dst_check(sk, 0)) {
1053 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1060 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1062 dst = dst_check(&rt->dst, 0);
1065 dst_release(&rt->dst);
1067 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1075 __sk_dst_set(sk, &rt->dst);
1080 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1082 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1083 int oif, u32 mark, u8 protocol, int flow_flags)
1085 const struct iphdr *iph = (const struct iphdr *) skb->data;
1089 __build_flow_key(&fl4, NULL, iph, oif,
1090 RT_TOS(iph->tos), protocol, mark, flow_flags);
1091 rt = __ip_route_output_key(net, &fl4);
1093 __ip_do_redirect(rt, skb, &fl4, false);
1097 EXPORT_SYMBOL_GPL(ipv4_redirect);
1099 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1101 const struct iphdr *iph = (const struct iphdr *) skb->data;
1105 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1106 rt = __ip_route_output_key(sock_net(sk), &fl4);
1108 __ip_do_redirect(rt, skb, &fl4, false);
1112 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1114 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1116 struct rtable *rt = (struct rtable *) dst;
1118 /* All IPV4 dsts are created with ->obsolete set to the value
1119 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1120 * into this function always.
1122 * When a PMTU/redirect information update invalidates a route,
1123 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1124 * DST_OBSOLETE_DEAD by dst_free().
1126 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1131 static void ipv4_link_failure(struct sk_buff *skb)
1135 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1137 rt = skb_rtable(skb);
1139 dst_set_expires(&rt->dst, 0);
1142 static int ip_rt_bug(struct sk_buff *skb)
1144 pr_debug("%s: %pI4 -> %pI4, %s\n",
1145 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1146 skb->dev ? skb->dev->name : "?");
1153 We do not cache source address of outgoing interface,
1154 because it is used only by IP RR, TS and SRR options,
1155 so that it out of fast path.
1157 BTW remember: "addr" is allowed to be not aligned
1161 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1165 if (rt_is_output_route(rt))
1166 src = ip_hdr(skb)->saddr;
1168 struct fib_result res;
1174 memset(&fl4, 0, sizeof(fl4));
1175 fl4.daddr = iph->daddr;
1176 fl4.saddr = iph->saddr;
1177 fl4.flowi4_tos = RT_TOS(iph->tos);
1178 fl4.flowi4_oif = rt->dst.dev->ifindex;
1179 fl4.flowi4_iif = skb->dev->ifindex;
1180 fl4.flowi4_mark = skb->mark;
1183 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1184 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1186 src = inet_select_addr(rt->dst.dev,
1187 rt_nexthop(rt, iph->daddr),
1191 memcpy(addr, &src, 4);
1194 #ifdef CONFIG_IP_ROUTE_CLASSID
1195 static void set_class_tag(struct rtable *rt, u32 tag)
1197 if (!(rt->dst.tclassid & 0xFFFF))
1198 rt->dst.tclassid |= tag & 0xFFFF;
1199 if (!(rt->dst.tclassid & 0xFFFF0000))
1200 rt->dst.tclassid |= tag & 0xFFFF0000;
1204 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1206 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1209 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1211 if (advmss > 65535 - 40)
1212 advmss = 65535 - 40;
1217 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1219 const struct rtable *rt = (const struct rtable *) dst;
1220 unsigned int mtu = rt->rt_pmtu;
1222 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1223 mtu = dst_metric_raw(dst, RTAX_MTU);
1228 mtu = dst->dev->mtu;
1230 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1231 if (rt->rt_uses_gateway && mtu > 576)
1235 return min_t(unsigned int, mtu, IP_MAX_MTU);
1238 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1240 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1241 struct fib_nh_exception *fnhe;
1247 hval = fnhe_hashfun(daddr);
1249 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1250 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1251 if (fnhe->fnhe_daddr == daddr)
1257 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1262 spin_lock_bh(&fnhe_lock);
1264 if (daddr == fnhe->fnhe_daddr) {
1265 struct rtable __rcu **porig;
1266 struct rtable *orig;
1267 int genid = fnhe_genid(dev_net(rt->dst.dev));
1269 if (rt_is_input_route(rt))
1270 porig = &fnhe->fnhe_rth_input;
1272 porig = &fnhe->fnhe_rth_output;
1273 orig = rcu_dereference(*porig);
1275 if (fnhe->fnhe_genid != genid) {
1276 fnhe->fnhe_genid = genid;
1278 fnhe->fnhe_pmtu = 0;
1279 fnhe->fnhe_expires = 0;
1280 fnhe_flush_routes(fnhe);
1283 fill_route_from_fnhe(rt, fnhe);
1284 if (!rt->rt_gateway)
1285 rt->rt_gateway = daddr;
1287 if (!(rt->dst.flags & DST_NOCACHE)) {
1288 rcu_assign_pointer(*porig, rt);
1294 fnhe->fnhe_stamp = jiffies;
1296 spin_unlock_bh(&fnhe_lock);
1301 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1303 struct rtable *orig, *prev, **p;
1306 if (rt_is_input_route(rt)) {
1307 p = (struct rtable **)&nh->nh_rth_input;
1309 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1313 prev = cmpxchg(p, orig, rt);
1323 static DEFINE_SPINLOCK(rt_uncached_lock);
1324 static LIST_HEAD(rt_uncached_list);
1326 static void rt_add_uncached_list(struct rtable *rt)
1328 spin_lock_bh(&rt_uncached_lock);
1329 list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1330 spin_unlock_bh(&rt_uncached_lock);
1333 static void ipv4_dst_destroy(struct dst_entry *dst)
1335 struct rtable *rt = (struct rtable *) dst;
1337 if (!list_empty(&rt->rt_uncached)) {
1338 spin_lock_bh(&rt_uncached_lock);
1339 list_del(&rt->rt_uncached);
1340 spin_unlock_bh(&rt_uncached_lock);
1344 void rt_flush_dev(struct net_device *dev)
1346 if (!list_empty(&rt_uncached_list)) {
1347 struct net *net = dev_net(dev);
1350 spin_lock_bh(&rt_uncached_lock);
1351 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1352 if (rt->dst.dev != dev)
1354 rt->dst.dev = net->loopback_dev;
1355 dev_hold(rt->dst.dev);
1358 spin_unlock_bh(&rt_uncached_lock);
1362 static bool rt_cache_valid(const struct rtable *rt)
1365 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1369 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1370 const struct fib_result *res,
1371 struct fib_nh_exception *fnhe,
1372 struct fib_info *fi, u16 type, u32 itag)
1374 bool cached = false;
1377 struct fib_nh *nh = &FIB_RES_NH(*res);
1379 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1380 rt->rt_gateway = nh->nh_gw;
1381 rt->rt_uses_gateway = 1;
1383 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1384 #ifdef CONFIG_IP_ROUTE_CLASSID
1385 rt->dst.tclassid = nh->nh_tclassid;
1388 cached = rt_bind_exception(rt, fnhe, daddr);
1389 else if (!(rt->dst.flags & DST_NOCACHE))
1390 cached = rt_cache_route(nh, rt);
1391 if (unlikely(!cached)) {
1392 /* Routes we intend to cache in nexthop exception or
1393 * FIB nexthop have the DST_NOCACHE bit clear.
1394 * However, if we are unsuccessful at storing this
1395 * route into the cache we really need to set it.
1397 rt->dst.flags |= DST_NOCACHE;
1398 if (!rt->rt_gateway)
1399 rt->rt_gateway = daddr;
1400 rt_add_uncached_list(rt);
1403 rt_add_uncached_list(rt);
1405 #ifdef CONFIG_IP_ROUTE_CLASSID
1406 #ifdef CONFIG_IP_MULTIPLE_TABLES
1407 set_class_tag(rt, res->tclassid);
1409 set_class_tag(rt, itag);
1413 static struct rtable *rt_dst_alloc(struct net_device *dev,
1414 bool nopolicy, bool noxfrm, bool will_cache)
1416 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1417 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1418 (nopolicy ? DST_NOPOLICY : 0) |
1419 (noxfrm ? DST_NOXFRM : 0));
1422 /* called in rcu_read_lock() section */
1423 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1424 u8 tos, struct net_device *dev, int our)
1427 struct in_device *in_dev = __in_dev_get_rcu(dev);
1431 /* Primary sanity checks. */
1436 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1437 skb->protocol != htons(ETH_P_IP))
1440 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1441 if (ipv4_is_loopback(saddr))
1444 if (ipv4_is_zeronet(saddr)) {
1445 if (!ipv4_is_local_multicast(daddr))
1448 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1453 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1454 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1458 #ifdef CONFIG_IP_ROUTE_CLASSID
1459 rth->dst.tclassid = itag;
1461 rth->dst.output = ip_rt_bug;
1463 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
1464 rth->rt_flags = RTCF_MULTICAST;
1465 rth->rt_type = RTN_MULTICAST;
1466 rth->rt_is_input= 1;
1469 rth->rt_gateway = 0;
1470 rth->rt_uses_gateway = 0;
1471 INIT_LIST_HEAD(&rth->rt_uncached);
1473 rth->dst.input= ip_local_deliver;
1474 rth->rt_flags |= RTCF_LOCAL;
1477 #ifdef CONFIG_IP_MROUTE
1478 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1479 rth->dst.input = ip_mr_input;
1481 RT_CACHE_STAT_INC(in_slow_mc);
1483 skb_dst_set(skb, &rth->dst);
1495 static void ip_handle_martian_source(struct net_device *dev,
1496 struct in_device *in_dev,
1497 struct sk_buff *skb,
1501 RT_CACHE_STAT_INC(in_martian_src);
1502 #ifdef CONFIG_IP_ROUTE_VERBOSE
1503 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1505 * RFC1812 recommendation, if source is martian,
1506 * the only hint is MAC header.
1508 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1509 &daddr, &saddr, dev->name);
1510 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1511 print_hex_dump(KERN_WARNING, "ll header: ",
1512 DUMP_PREFIX_OFFSET, 16, 1,
1513 skb_mac_header(skb),
1514 dev->hard_header_len, true);
1520 /* called in rcu_read_lock() section */
1521 static int __mkroute_input(struct sk_buff *skb,
1522 const struct fib_result *res,
1523 struct in_device *in_dev,
1524 __be32 daddr, __be32 saddr, u32 tos)
1526 struct fib_nh_exception *fnhe;
1529 struct in_device *out_dev;
1530 unsigned int flags = 0;
1534 /* get a working reference to the output device */
1535 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1536 if (out_dev == NULL) {
1537 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1541 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1542 in_dev->dev, in_dev, &itag);
1544 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1550 do_cache = res->fi && !itag;
1551 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1552 (IN_DEV_SHARED_MEDIA(out_dev) ||
1553 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1554 flags |= RTCF_DOREDIRECT;
1558 if (skb->protocol != htons(ETH_P_IP)) {
1559 /* Not IP (i.e. ARP). Do not create route, if it is
1560 * invalid for proxy arp. DNAT routes are always valid.
1562 * Proxy arp feature have been extended to allow, ARP
1563 * replies back to the same interface, to support
1564 * Private VLAN switch technologies. See arp.c.
1566 if (out_dev == in_dev &&
1567 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1573 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1576 rth = rcu_dereference(fnhe->fnhe_rth_input);
1578 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1580 if (rt_cache_valid(rth)) {
1581 skb_dst_set_noref(skb, &rth->dst);
1586 rth = rt_dst_alloc(out_dev->dev,
1587 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1588 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1594 rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1595 rth->rt_flags = flags;
1596 rth->rt_type = res->type;
1597 rth->rt_is_input = 1;
1600 rth->rt_gateway = 0;
1601 rth->rt_uses_gateway = 0;
1602 INIT_LIST_HEAD(&rth->rt_uncached);
1604 rth->dst.input = ip_forward;
1605 rth->dst.output = ip_output;
1607 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1608 skb_dst_set(skb, &rth->dst);
1615 static int ip_mkroute_input(struct sk_buff *skb,
1616 struct fib_result *res,
1617 const struct flowi4 *fl4,
1618 struct in_device *in_dev,
1619 __be32 daddr, __be32 saddr, u32 tos)
1621 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1622 if (res->fi && res->fi->fib_nhs > 1)
1623 fib_select_multipath(res);
1626 /* create a routing cache entry */
1627 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1631 * NOTE. We drop all the packets that has local source
1632 * addresses, because every properly looped back packet
1633 * must have correct destination already attached by output routine.
1635 * Such approach solves two big problems:
1636 * 1. Not simplex devices are handled properly.
1637 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1638 * called with rcu_read_lock()
1641 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1642 u8 tos, struct net_device *dev)
1644 struct fib_result res;
1645 struct in_device *in_dev = __in_dev_get_rcu(dev);
1647 unsigned int flags = 0;
1651 struct net *net = dev_net(dev);
1654 /* IP on this device is disabled. */
1659 /* Check for the most weird martians, which can be not detected
1663 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1664 goto martian_source;
1667 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1670 /* Accept zero addresses only to limited broadcast;
1671 * I even do not know to fix it or not. Waiting for complains :-)
1673 if (ipv4_is_zeronet(saddr))
1674 goto martian_source;
1676 if (ipv4_is_zeronet(daddr))
1677 goto martian_destination;
1679 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1680 * and call it once if daddr or/and saddr are loopback addresses
1682 if (ipv4_is_loopback(daddr)) {
1683 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1684 goto martian_destination;
1685 } else if (ipv4_is_loopback(saddr)) {
1686 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1687 goto martian_source;
1691 * Now we are ready to route packet.
1694 fl4.flowi4_iif = dev->ifindex;
1695 fl4.flowi4_mark = skb->mark;
1696 fl4.flowi4_tos = tos;
1697 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1700 err = fib_lookup(net, &fl4, &res);
1704 RT_CACHE_STAT_INC(in_slow_tot);
1706 if (res.type == RTN_BROADCAST)
1709 if (res.type == RTN_LOCAL) {
1710 err = fib_validate_source(skb, saddr, daddr, tos,
1712 dev, in_dev, &itag);
1714 goto martian_source_keep_err;
1718 if (!IN_DEV_FORWARD(in_dev))
1720 if (res.type != RTN_UNICAST)
1721 goto martian_destination;
1723 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1727 if (skb->protocol != htons(ETH_P_IP))
1730 if (!ipv4_is_zeronet(saddr)) {
1731 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1734 goto martian_source_keep_err;
1736 flags |= RTCF_BROADCAST;
1737 res.type = RTN_BROADCAST;
1738 RT_CACHE_STAT_INC(in_brd);
1744 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1745 if (rt_cache_valid(rth)) {
1746 skb_dst_set_noref(skb, &rth->dst);
1754 rth = rt_dst_alloc(net->loopback_dev,
1755 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1759 rth->dst.input= ip_local_deliver;
1760 rth->dst.output= ip_rt_bug;
1761 #ifdef CONFIG_IP_ROUTE_CLASSID
1762 rth->dst.tclassid = itag;
1765 rth->rt_genid = rt_genid_ipv4(net);
1766 rth->rt_flags = flags|RTCF_LOCAL;
1767 rth->rt_type = res.type;
1768 rth->rt_is_input = 1;
1771 rth->rt_gateway = 0;
1772 rth->rt_uses_gateway = 0;
1773 INIT_LIST_HEAD(&rth->rt_uncached);
1774 if (res.type == RTN_UNREACHABLE) {
1775 rth->dst.input= ip_error;
1776 rth->dst.error= -err;
1777 rth->rt_flags &= ~RTCF_LOCAL;
1780 rt_cache_route(&FIB_RES_NH(res), rth);
1781 skb_dst_set(skb, &rth->dst);
1786 RT_CACHE_STAT_INC(in_no_route);
1787 res.type = RTN_UNREACHABLE;
1793 * Do not cache martian addresses: they should be logged (RFC1812)
1795 martian_destination:
1796 RT_CACHE_STAT_INC(in_martian_dst);
1797 #ifdef CONFIG_IP_ROUTE_VERBOSE
1798 if (IN_DEV_LOG_MARTIANS(in_dev))
1799 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1800 &daddr, &saddr, dev->name);
1813 martian_source_keep_err:
1814 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1818 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1819 u8 tos, struct net_device *dev)
1825 /* Multicast recognition logic is moved from route cache to here.
1826 The problem was that too many Ethernet cards have broken/missing
1827 hardware multicast filters :-( As result the host on multicasting
1828 network acquires a lot of useless route cache entries, sort of
1829 SDR messages from all the world. Now we try to get rid of them.
1830 Really, provided software IP multicast filter is organized
1831 reasonably (at least, hashed), it does not result in a slowdown
1832 comparing with route cache reject entries.
1833 Note, that multicast routers are not affected, because
1834 route cache entry is created eventually.
1836 if (ipv4_is_multicast(daddr)) {
1837 struct in_device *in_dev = __in_dev_get_rcu(dev);
1840 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1841 ip_hdr(skb)->protocol);
1843 #ifdef CONFIG_IP_MROUTE
1845 (!ipv4_is_local_multicast(daddr) &&
1846 IN_DEV_MFORWARD(in_dev))
1849 int res = ip_route_input_mc(skb, daddr, saddr,
1858 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1862 EXPORT_SYMBOL(ip_route_input_noref);
1864 /* called with rcu_read_lock() */
1865 static struct rtable *__mkroute_output(const struct fib_result *res,
1866 const struct flowi4 *fl4, int orig_oif,
1867 struct net_device *dev_out,
1870 struct fib_info *fi = res->fi;
1871 struct fib_nh_exception *fnhe;
1872 struct in_device *in_dev;
1873 u16 type = res->type;
1877 in_dev = __in_dev_get_rcu(dev_out);
1879 return ERR_PTR(-EINVAL);
1881 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1882 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1883 return ERR_PTR(-EINVAL);
1885 if (ipv4_is_lbcast(fl4->daddr))
1886 type = RTN_BROADCAST;
1887 else if (ipv4_is_multicast(fl4->daddr))
1888 type = RTN_MULTICAST;
1889 else if (ipv4_is_zeronet(fl4->daddr))
1890 return ERR_PTR(-EINVAL);
1892 if (dev_out->flags & IFF_LOOPBACK)
1893 flags |= RTCF_LOCAL;
1896 if (type == RTN_BROADCAST) {
1897 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1899 } else if (type == RTN_MULTICAST) {
1900 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1901 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1903 flags &= ~RTCF_LOCAL;
1906 /* If multicast route do not exist use
1907 * default one, but do not gateway in this case.
1910 if (fi && res->prefixlen < 4)
1915 do_cache &= fi != NULL;
1917 struct rtable __rcu **prth;
1918 struct fib_nh *nh = &FIB_RES_NH(*res);
1920 fnhe = find_exception(nh, fl4->daddr);
1922 prth = &fnhe->fnhe_rth_output;
1924 if (unlikely(fl4->flowi4_flags &
1925 FLOWI_FLAG_KNOWN_NH &&
1927 nh->nh_scope == RT_SCOPE_LINK))) {
1931 prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1933 rth = rcu_dereference(*prth);
1934 if (rt_cache_valid(rth)) {
1935 dst_hold(&rth->dst);
1941 rth = rt_dst_alloc(dev_out,
1942 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1943 IN_DEV_CONF_GET(in_dev, NOXFRM),
1946 return ERR_PTR(-ENOBUFS);
1948 rth->dst.output = ip_output;
1950 rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1951 rth->rt_flags = flags;
1952 rth->rt_type = type;
1953 rth->rt_is_input = 0;
1954 rth->rt_iif = orig_oif ? : 0;
1956 rth->rt_gateway = 0;
1957 rth->rt_uses_gateway = 0;
1958 INIT_LIST_HEAD(&rth->rt_uncached);
1960 RT_CACHE_STAT_INC(out_slow_tot);
1962 if (flags & RTCF_LOCAL)
1963 rth->dst.input = ip_local_deliver;
1964 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1965 if (flags & RTCF_LOCAL &&
1966 !(dev_out->flags & IFF_LOOPBACK)) {
1967 rth->dst.output = ip_mc_output;
1968 RT_CACHE_STAT_INC(out_slow_mc);
1970 #ifdef CONFIG_IP_MROUTE
1971 if (type == RTN_MULTICAST) {
1972 if (IN_DEV_MFORWARD(in_dev) &&
1973 !ipv4_is_local_multicast(fl4->daddr)) {
1974 rth->dst.input = ip_mr_input;
1975 rth->dst.output = ip_mc_output;
1981 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1987 * Major route resolver routine.
1990 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1992 struct net_device *dev_out = NULL;
1993 __u8 tos = RT_FL_TOS(fl4);
1994 unsigned int flags = 0;
1995 struct fib_result res;
2003 orig_oif = fl4->flowi4_oif;
2005 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2006 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2007 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2008 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2012 rth = ERR_PTR(-EINVAL);
2013 if (ipv4_is_multicast(fl4->saddr) ||
2014 ipv4_is_lbcast(fl4->saddr) ||
2015 ipv4_is_zeronet(fl4->saddr))
2018 /* I removed check for oif == dev_out->oif here.
2019 It was wrong for two reasons:
2020 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2021 is assigned to multiple interfaces.
2022 2. Moreover, we are allowed to send packets with saddr
2023 of another iface. --ANK
2026 if (fl4->flowi4_oif == 0 &&
2027 (ipv4_is_multicast(fl4->daddr) ||
2028 ipv4_is_lbcast(fl4->daddr))) {
2029 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2030 dev_out = __ip_dev_find(net, fl4->saddr, false);
2031 if (dev_out == NULL)
2034 /* Special hack: user can direct multicasts
2035 and limited broadcast via necessary interface
2036 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2037 This hack is not just for fun, it allows
2038 vic,vat and friends to work.
2039 They bind socket to loopback, set ttl to zero
2040 and expect that it will work.
2041 From the viewpoint of routing cache they are broken,
2042 because we are not allowed to build multicast path
2043 with loopback source addr (look, routing cache
2044 cannot know, that ttl is zero, so that packet
2045 will not leave this host and route is valid).
2046 Luckily, this hack is good workaround.
2049 fl4->flowi4_oif = dev_out->ifindex;
2053 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2054 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2055 if (!__ip_dev_find(net, fl4->saddr, false))
2061 if (fl4->flowi4_oif) {
2062 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2063 rth = ERR_PTR(-ENODEV);
2064 if (dev_out == NULL)
2067 /* RACE: Check return value of inet_select_addr instead. */
2068 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2069 rth = ERR_PTR(-ENETUNREACH);
2072 if (ipv4_is_local_multicast(fl4->daddr) ||
2073 ipv4_is_lbcast(fl4->daddr)) {
2075 fl4->saddr = inet_select_addr(dev_out, 0,
2080 if (ipv4_is_multicast(fl4->daddr))
2081 fl4->saddr = inet_select_addr(dev_out, 0,
2083 else if (!fl4->daddr)
2084 fl4->saddr = inet_select_addr(dev_out, 0,
2090 fl4->daddr = fl4->saddr;
2092 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2093 dev_out = net->loopback_dev;
2094 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2095 res.type = RTN_LOCAL;
2096 flags |= RTCF_LOCAL;
2100 if (fib_lookup(net, fl4, &res)) {
2103 if (fl4->flowi4_oif) {
2104 /* Apparently, routing tables are wrong. Assume,
2105 that the destination is on link.
2108 Because we are allowed to send to iface
2109 even if it has NO routes and NO assigned
2110 addresses. When oif is specified, routing
2111 tables are looked up with only one purpose:
2112 to catch if destination is gatewayed, rather than
2113 direct. Moreover, if MSG_DONTROUTE is set,
2114 we send packet, ignoring both routing tables
2115 and ifaddr state. --ANK
2118 We could make it even if oif is unknown,
2119 likely IPv6, but we do not.
2122 if (fl4->saddr == 0)
2123 fl4->saddr = inet_select_addr(dev_out, 0,
2125 res.type = RTN_UNICAST;
2128 rth = ERR_PTR(-ENETUNREACH);
2132 if (res.type == RTN_LOCAL) {
2134 if (res.fi->fib_prefsrc)
2135 fl4->saddr = res.fi->fib_prefsrc;
2137 fl4->saddr = fl4->daddr;
2139 dev_out = net->loopback_dev;
2140 fl4->flowi4_oif = dev_out->ifindex;
2141 flags |= RTCF_LOCAL;
2145 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2146 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2147 fib_select_multipath(&res);
2150 if (!res.prefixlen &&
2151 res.table->tb_num_default > 1 &&
2152 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2153 fib_select_default(&res);
2156 fl4->saddr = FIB_RES_PREFSRC(net, res);
2158 dev_out = FIB_RES_DEV(res);
2159 fl4->flowi4_oif = dev_out->ifindex;
2163 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2169 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2171 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2176 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2178 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2180 return mtu ? : dst->dev->mtu;
2183 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2184 struct sk_buff *skb, u32 mtu)
2188 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2189 struct sk_buff *skb)
2193 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2199 static struct dst_ops ipv4_dst_blackhole_ops = {
2201 .protocol = cpu_to_be16(ETH_P_IP),
2202 .check = ipv4_blackhole_dst_check,
2203 .mtu = ipv4_blackhole_mtu,
2204 .default_advmss = ipv4_default_advmss,
2205 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2206 .redirect = ipv4_rt_blackhole_redirect,
2207 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2208 .neigh_lookup = ipv4_neigh_lookup,
2211 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2213 struct rtable *ort = (struct rtable *) dst_orig;
2216 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2218 struct dst_entry *new = &rt->dst;
2221 new->input = dst_discard;
2222 new->output = dst_discard;
2224 new->dev = ort->dst.dev;
2228 rt->rt_is_input = ort->rt_is_input;
2229 rt->rt_iif = ort->rt_iif;
2230 rt->rt_pmtu = ort->rt_pmtu;
2232 rt->rt_genid = rt_genid_ipv4(net);
2233 rt->rt_flags = ort->rt_flags;
2234 rt->rt_type = ort->rt_type;
2235 rt->rt_gateway = ort->rt_gateway;
2236 rt->rt_uses_gateway = ort->rt_uses_gateway;
2238 INIT_LIST_HEAD(&rt->rt_uncached);
2243 dst_release(dst_orig);
2245 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2248 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2251 struct rtable *rt = __ip_route_output_key(net, flp4);
2256 if (flp4->flowi4_proto)
2257 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2258 flowi4_to_flowi(flp4),
2263 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2265 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2266 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2267 u32 seq, int event, int nowait, unsigned int flags)
2269 struct rtable *rt = skb_rtable(skb);
2271 struct nlmsghdr *nlh;
2272 unsigned long expires = 0;
2274 u32 metrics[RTAX_MAX];
2276 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2280 r = nlmsg_data(nlh);
2281 r->rtm_family = AF_INET;
2282 r->rtm_dst_len = 32;
2284 r->rtm_tos = fl4->flowi4_tos;
2285 r->rtm_table = RT_TABLE_MAIN;
2286 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2287 goto nla_put_failure;
2288 r->rtm_type = rt->rt_type;
2289 r->rtm_scope = RT_SCOPE_UNIVERSE;
2290 r->rtm_protocol = RTPROT_UNSPEC;
2291 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2292 if (rt->rt_flags & RTCF_NOTIFY)
2293 r->rtm_flags |= RTM_F_NOTIFY;
2295 if (nla_put_be32(skb, RTA_DST, dst))
2296 goto nla_put_failure;
2298 r->rtm_src_len = 32;
2299 if (nla_put_be32(skb, RTA_SRC, src))
2300 goto nla_put_failure;
2303 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2304 goto nla_put_failure;
2305 #ifdef CONFIG_IP_ROUTE_CLASSID
2306 if (rt->dst.tclassid &&
2307 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2308 goto nla_put_failure;
2310 if (!rt_is_input_route(rt) &&
2311 fl4->saddr != src) {
2312 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2313 goto nla_put_failure;
2315 if (rt->rt_uses_gateway &&
2316 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2317 goto nla_put_failure;
2319 expires = rt->dst.expires;
2321 unsigned long now = jiffies;
2323 if (time_before(now, expires))
2329 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2330 if (rt->rt_pmtu && expires)
2331 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2332 if (rtnetlink_put_metrics(skb, metrics) < 0)
2333 goto nla_put_failure;
2335 if (fl4->flowi4_mark &&
2336 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2337 goto nla_put_failure;
2339 error = rt->dst.error;
2341 if (rt_is_input_route(rt)) {
2342 #ifdef CONFIG_IP_MROUTE
2343 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2344 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2345 int err = ipmr_get_route(net, skb,
2346 fl4->saddr, fl4->daddr,
2352 goto nla_put_failure;
2354 if (err == -EMSGSIZE)
2355 goto nla_put_failure;
2361 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2362 goto nla_put_failure;
2365 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2366 goto nla_put_failure;
2368 return nlmsg_end(skb, nlh);
2371 nlmsg_cancel(skb, nlh);
2375 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2377 struct net *net = sock_net(in_skb->sk);
2379 struct nlattr *tb[RTA_MAX+1];
2380 struct rtable *rt = NULL;
2387 struct sk_buff *skb;
2389 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2393 rtm = nlmsg_data(nlh);
2395 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2401 /* Reserve room for dummy headers, this skb can pass
2402 through good chunk of routing engine.
2404 skb_reset_mac_header(skb);
2405 skb_reset_network_header(skb);
2407 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2408 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2409 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2411 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2412 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2413 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2414 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2416 memset(&fl4, 0, sizeof(fl4));
2419 fl4.flowi4_tos = rtm->rtm_tos;
2420 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2421 fl4.flowi4_mark = mark;
2424 struct net_device *dev;
2426 dev = __dev_get_by_index(net, iif);
2432 skb->protocol = htons(ETH_P_IP);
2436 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2439 rt = skb_rtable(skb);
2440 if (err == 0 && rt->dst.error)
2441 err = -rt->dst.error;
2443 rt = ip_route_output_key(net, &fl4);
2453 skb_dst_set(skb, &rt->dst);
2454 if (rtm->rtm_flags & RTM_F_NOTIFY)
2455 rt->rt_flags |= RTCF_NOTIFY;
2457 err = rt_fill_info(net, dst, src, &fl4, skb,
2458 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2459 RTM_NEWROUTE, 0, 0);
2463 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2472 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2477 void ip_rt_multicast_event(struct in_device *in_dev)
2479 rt_cache_flush(dev_net(in_dev->dev));
2482 #ifdef CONFIG_SYSCTL
2483 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
2484 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2485 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2486 static int ip_rt_gc_elasticity __read_mostly = 8;
2488 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2489 void __user *buffer,
2490 size_t *lenp, loff_t *ppos)
2492 struct net *net = (struct net *)__ctl->extra1;
2495 rt_cache_flush(net);
2496 fnhe_genid_bump(net);
2503 static struct ctl_table ipv4_route_table[] = {
2505 .procname = "gc_thresh",
2506 .data = &ipv4_dst_ops.gc_thresh,
2507 .maxlen = sizeof(int),
2509 .proc_handler = proc_dointvec,
2512 .procname = "max_size",
2513 .data = &ip_rt_max_size,
2514 .maxlen = sizeof(int),
2516 .proc_handler = proc_dointvec,
2519 /* Deprecated. Use gc_min_interval_ms */
2521 .procname = "gc_min_interval",
2522 .data = &ip_rt_gc_min_interval,
2523 .maxlen = sizeof(int),
2525 .proc_handler = proc_dointvec_jiffies,
2528 .procname = "gc_min_interval_ms",
2529 .data = &ip_rt_gc_min_interval,
2530 .maxlen = sizeof(int),
2532 .proc_handler = proc_dointvec_ms_jiffies,
2535 .procname = "gc_timeout",
2536 .data = &ip_rt_gc_timeout,
2537 .maxlen = sizeof(int),
2539 .proc_handler = proc_dointvec_jiffies,
2542 .procname = "gc_interval",
2543 .data = &ip_rt_gc_interval,
2544 .maxlen = sizeof(int),
2546 .proc_handler = proc_dointvec_jiffies,
2549 .procname = "redirect_load",
2550 .data = &ip_rt_redirect_load,
2551 .maxlen = sizeof(int),
2553 .proc_handler = proc_dointvec,
2556 .procname = "redirect_number",
2557 .data = &ip_rt_redirect_number,
2558 .maxlen = sizeof(int),
2560 .proc_handler = proc_dointvec,
2563 .procname = "redirect_silence",
2564 .data = &ip_rt_redirect_silence,
2565 .maxlen = sizeof(int),
2567 .proc_handler = proc_dointvec,
2570 .procname = "error_cost",
2571 .data = &ip_rt_error_cost,
2572 .maxlen = sizeof(int),
2574 .proc_handler = proc_dointvec,
2577 .procname = "error_burst",
2578 .data = &ip_rt_error_burst,
2579 .maxlen = sizeof(int),
2581 .proc_handler = proc_dointvec,
2584 .procname = "gc_elasticity",
2585 .data = &ip_rt_gc_elasticity,
2586 .maxlen = sizeof(int),
2588 .proc_handler = proc_dointvec,
2591 .procname = "mtu_expires",
2592 .data = &ip_rt_mtu_expires,
2593 .maxlen = sizeof(int),
2595 .proc_handler = proc_dointvec_jiffies,
2598 .procname = "min_pmtu",
2599 .data = &ip_rt_min_pmtu,
2600 .maxlen = sizeof(int),
2602 .proc_handler = proc_dointvec,
2605 .procname = "min_adv_mss",
2606 .data = &ip_rt_min_advmss,
2607 .maxlen = sizeof(int),
2609 .proc_handler = proc_dointvec,
2614 static struct ctl_table ipv4_route_flush_table[] = {
2616 .procname = "flush",
2617 .maxlen = sizeof(int),
2619 .proc_handler = ipv4_sysctl_rtcache_flush,
2624 static __net_init int sysctl_route_net_init(struct net *net)
2626 struct ctl_table *tbl;
2628 tbl = ipv4_route_flush_table;
2629 if (!net_eq(net, &init_net)) {
2630 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2634 /* Don't export sysctls to unprivileged users */
2635 if (net->user_ns != &init_user_ns)
2636 tbl[0].procname = NULL;
2638 tbl[0].extra1 = net;
2640 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2641 if (net->ipv4.route_hdr == NULL)
2646 if (tbl != ipv4_route_flush_table)
2652 static __net_exit void sysctl_route_net_exit(struct net *net)
2654 struct ctl_table *tbl;
2656 tbl = net->ipv4.route_hdr->ctl_table_arg;
2657 unregister_net_sysctl_table(net->ipv4.route_hdr);
2658 BUG_ON(tbl == ipv4_route_flush_table);
2662 static __net_initdata struct pernet_operations sysctl_route_ops = {
2663 .init = sysctl_route_net_init,
2664 .exit = sysctl_route_net_exit,
2668 static __net_init int rt_genid_init(struct net *net)
2670 atomic_set(&net->ipv4.rt_genid, 0);
2671 atomic_set(&net->fnhe_genid, 0);
2672 get_random_bytes(&net->ipv4.dev_addr_genid,
2673 sizeof(net->ipv4.dev_addr_genid));
2677 static __net_initdata struct pernet_operations rt_genid_ops = {
2678 .init = rt_genid_init,
2681 static int __net_init ipv4_inetpeer_init(struct net *net)
2683 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2687 inet_peer_base_init(bp);
2688 net->ipv4.peers = bp;
2692 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2694 struct inet_peer_base *bp = net->ipv4.peers;
2696 net->ipv4.peers = NULL;
2697 inetpeer_invalidate_tree(bp);
2701 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2702 .init = ipv4_inetpeer_init,
2703 .exit = ipv4_inetpeer_exit,
2706 #ifdef CONFIG_IP_ROUTE_CLASSID
2707 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2708 #endif /* CONFIG_IP_ROUTE_CLASSID */
2710 int __init ip_rt_init(void)
2714 #ifdef CONFIG_IP_ROUTE_CLASSID
2715 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2717 panic("IP: failed to allocate ip_rt_acct\n");
2720 ipv4_dst_ops.kmem_cachep =
2721 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2722 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2724 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2726 if (dst_entries_init(&ipv4_dst_ops) < 0)
2727 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2729 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2730 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2732 ipv4_dst_ops.gc_thresh = ~0;
2733 ip_rt_max_size = INT_MAX;
2738 if (ip_rt_proc_init())
2739 pr_err("Unable to create route proc files\n");
2744 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2746 #ifdef CONFIG_SYSCTL
2747 register_pernet_subsys(&sysctl_route_ops);
2749 register_pernet_subsys(&rt_genid_ops);
2750 register_pernet_subsys(&ipv4_inetpeer_ops);
2754 #ifdef CONFIG_SYSCTL
2756 * We really need to sanitize the damn ipv4 init order, then all
2757 * this nonsense will go away.
2759 void __init ip_static_sysctl_init(void)
2761 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);