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/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/rcupdate.h>
91 #include <linux/times.h>
92 #include <linux/slab.h>
93 #include <linux/jhash.h>
95 #include <net/dst_metadata.h>
96 #include <net/net_namespace.h>
97 #include <net/protocol.h>
99 #include <net/route.h>
100 #include <net/inetpeer.h>
101 #include <net/sock.h>
102 #include <net/ip_fib.h>
105 #include <net/icmp.h>
106 #include <net/xfrm.h>
107 #include <net/lwtunnel.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
114 #include <net/secure_seq.h>
115 #include <net/ip_tunnels.h>
116 #include <net/l3mdev.h>
118 #define RT_FL_TOS(oldflp4) \
119 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size;
124 static int ip_rt_redirect_number __read_mostly = 9;
125 static int ip_rt_redirect_load __read_mostly = HZ / 50;
126 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly = HZ;
128 static int ip_rt_error_burst __read_mostly = 5 * HZ;
129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly = 256;
133 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
136 * Interface to generic destination cache.
139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
140 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
141 static unsigned int ipv4_mtu(const struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void ipv4_link_failure(struct sk_buff *skb);
144 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb, u32 mtu);
146 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
147 struct sk_buff *skb);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
160 static struct dst_ops ipv4_dst_ops = {
162 .check = ipv4_dst_check,
163 .default_advmss = ipv4_default_advmss,
165 .cow_metrics = ipv4_cow_metrics,
166 .destroy = ipv4_dst_destroy,
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) raw_cpu_inc(rt_cache_stat.field)
200 #ifdef CONFIG_PROC_FS
201 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
205 return SEQ_START_TOKEN;
208 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
214 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
218 static int rt_cache_seq_show(struct seq_file *seq, void *v)
220 if (v == SEQ_START_TOKEN)
221 seq_printf(seq, "%-127s\n",
222 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
223 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
228 static const struct seq_operations rt_cache_seq_ops = {
229 .start = rt_cache_seq_start,
230 .next = rt_cache_seq_next,
231 .stop = rt_cache_seq_stop,
232 .show = rt_cache_seq_show,
235 static int rt_cache_seq_open(struct inode *inode, struct file *file)
237 return seq_open(file, &rt_cache_seq_ops);
240 static const struct file_operations rt_cache_seq_fops = {
241 .owner = THIS_MODULE,
242 .open = rt_cache_seq_open,
245 .release = seq_release,
249 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
254 return SEQ_START_TOKEN;
256 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
257 if (!cpu_possible(cpu))
260 return &per_cpu(rt_cache_stat, cpu);
265 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
269 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
270 if (!cpu_possible(cpu))
273 return &per_cpu(rt_cache_stat, cpu);
280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
285 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
287 struct rt_cache_stat *st = v;
289 if (v == SEQ_START_TOKEN) {
290 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
294 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
295 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
296 dst_entries_get_slow(&ipv4_dst_ops),
309 0, /* st->gc_total */
310 0, /* st->gc_ignored */
311 0, /* st->gc_goal_miss */
312 0, /* st->gc_dst_overflow */
313 0, /* st->in_hlist_search */
314 0 /* st->out_hlist_search */
319 static const struct seq_operations rt_cpu_seq_ops = {
320 .start = rt_cpu_seq_start,
321 .next = rt_cpu_seq_next,
322 .stop = rt_cpu_seq_stop,
323 .show = rt_cpu_seq_show,
327 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
329 return seq_open(file, &rt_cpu_seq_ops);
332 static const struct file_operations rt_cpu_seq_fops = {
333 .owner = THIS_MODULE,
334 .open = rt_cpu_seq_open,
337 .release = seq_release,
340 #ifdef CONFIG_IP_ROUTE_CLASSID
341 static int rt_acct_proc_show(struct seq_file *m, void *v)
343 struct ip_rt_acct *dst, *src;
346 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
350 for_each_possible_cpu(i) {
351 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
352 for (j = 0; j < 256; j++) {
353 dst[j].o_bytes += src[j].o_bytes;
354 dst[j].o_packets += src[j].o_packets;
355 dst[j].i_bytes += src[j].i_bytes;
356 dst[j].i_packets += src[j].i_packets;
360 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
365 static int rt_acct_proc_open(struct inode *inode, struct file *file)
367 return single_open(file, rt_acct_proc_show, NULL);
370 static const struct file_operations rt_acct_proc_fops = {
371 .owner = THIS_MODULE,
372 .open = rt_acct_proc_open,
375 .release = single_release,
379 static int __net_init ip_rt_do_proc_init(struct net *net)
381 struct proc_dir_entry *pde;
383 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
388 pde = proc_create("rt_cache", S_IRUGO,
389 net->proc_net_stat, &rt_cpu_seq_fops);
393 #ifdef CONFIG_IP_ROUTE_CLASSID
394 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
400 #ifdef CONFIG_IP_ROUTE_CLASSID
402 remove_proc_entry("rt_cache", net->proc_net_stat);
405 remove_proc_entry("rt_cache", net->proc_net);
410 static void __net_exit ip_rt_do_proc_exit(struct net *net)
412 remove_proc_entry("rt_cache", net->proc_net_stat);
413 remove_proc_entry("rt_cache", net->proc_net);
414 #ifdef CONFIG_IP_ROUTE_CLASSID
415 remove_proc_entry("rt_acct", net->proc_net);
419 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
420 .init = ip_rt_do_proc_init,
421 .exit = ip_rt_do_proc_exit,
424 static int __init ip_rt_proc_init(void)
426 return register_pernet_subsys(&ip_rt_proc_ops);
430 static inline int ip_rt_proc_init(void)
434 #endif /* CONFIG_PROC_FS */
436 static inline bool rt_is_expired(const struct rtable *rth)
438 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
441 void rt_cache_flush(struct net *net)
443 rt_genid_bump_ipv4(net);
446 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
450 struct net_device *dev = dst->dev;
451 const __be32 *pkey = daddr;
452 const struct rtable *rt;
455 rt = (const struct rtable *) dst;
457 pkey = (const __be32 *) &rt->rt_gateway;
459 pkey = &ip_hdr(skb)->daddr;
461 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
464 return neigh_create(&arp_tbl, pkey, dev);
467 /* Hash tables of size 2048..262144 depending on RAM size.
468 * Each bucket uses 8 bytes.
470 static u32 ip_idents_mask __read_mostly;
471 static atomic_t *ip_idents __read_mostly;
472 static u32 *ip_tstamps __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 u32 bucket, old, now = (u32)jiffies;
485 bucket = hash & ip_idents_mask;
486 p_tstamp = ip_tstamps + bucket;
487 p_id = ip_idents + bucket;
488 old = ACCESS_ONCE(*p_tstamp);
490 if (old != now && cmpxchg(p_tstamp, old, now) == old)
491 delta = prandom_u32_max(now - old);
493 /* If UBSAN reports an error there, please make sure your compiler
494 * supports -fno-strict-overflow before reporting it that was a bug
495 * in UBSAN, and it has been fixed in GCC-8.
497 return atomic_add_return(segs + delta, p_id) - segs;
499 EXPORT_SYMBOL(ip_idents_reserve);
501 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
505 /* Note the following code is not safe, but this is okay. */
506 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
507 get_random_bytes(&net->ipv4.ip_id_key,
508 sizeof(net->ipv4.ip_id_key));
510 hash = siphash_3u32((__force u32)iph->daddr,
511 (__force u32)iph->saddr,
513 &net->ipv4.ip_id_key);
514 id = ip_idents_reserve(hash, segs);
517 EXPORT_SYMBOL(__ip_select_ident);
519 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
520 const struct iphdr *iph,
522 u8 prot, u32 mark, int flow_flags)
525 const struct inet_sock *inet = inet_sk(sk);
527 oif = sk->sk_bound_dev_if;
529 tos = RT_CONN_FLAGS(sk);
530 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
532 flowi4_init_output(fl4, oif, mark, tos,
533 RT_SCOPE_UNIVERSE, prot,
535 iph->daddr, iph->saddr, 0, 0);
538 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
539 const struct sock *sk)
541 const struct iphdr *iph = ip_hdr(skb);
542 int oif = skb->dev->ifindex;
543 u8 tos = RT_TOS(iph->tos);
544 u8 prot = iph->protocol;
545 u32 mark = skb->mark;
547 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
550 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
552 const struct inet_sock *inet = inet_sk(sk);
553 const struct ip_options_rcu *inet_opt;
554 __be32 daddr = inet->inet_daddr;
557 inet_opt = rcu_dereference(inet->inet_opt);
558 if (inet_opt && inet_opt->opt.srr)
559 daddr = inet_opt->opt.faddr;
560 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
561 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
562 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
563 inet_sk_flowi_flags(sk),
564 daddr, inet->inet_saddr, 0, 0);
568 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
569 const struct sk_buff *skb)
572 build_skb_flow_key(fl4, skb, sk);
574 build_sk_flow_key(fl4, sk);
577 static inline void rt_free(struct rtable *rt)
579 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
582 static DEFINE_SPINLOCK(fnhe_lock);
584 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
588 rt = rcu_dereference(fnhe->fnhe_rth_input);
590 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
593 rt = rcu_dereference(fnhe->fnhe_rth_output);
595 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
600 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
602 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
603 struct fib_nh_exception *fnhe, *oldest = NULL;
605 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
606 fnhe = rcu_dereference_protected(*fnhe_p,
607 lockdep_is_held(&fnhe_lock));
611 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
616 fnhe_flush_routes(oldest);
617 *oldest_p = oldest->fnhe_next;
618 kfree_rcu(oldest, rcu);
621 static inline u32 fnhe_hashfun(__be32 daddr)
623 static u32 fnhe_hashrnd __read_mostly;
626 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
627 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
628 return hash_32(hval, FNHE_HASH_SHIFT);
631 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
633 rt->rt_pmtu = fnhe->fnhe_pmtu;
634 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
635 rt->dst.expires = fnhe->fnhe_expires;
638 rt->rt_flags |= RTCF_REDIRECTED;
639 rt->rt_gateway = fnhe->fnhe_gw;
640 rt->rt_uses_gateway = 1;
644 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
645 u32 pmtu, bool lock, unsigned long expires)
647 struct fnhe_hash_bucket *hash;
648 struct fib_nh_exception *fnhe;
654 genid = fnhe_genid(dev_net(nh->nh_dev));
655 hval = fnhe_hashfun(daddr);
657 spin_lock_bh(&fnhe_lock);
659 hash = rcu_dereference(nh->nh_exceptions);
661 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
664 rcu_assign_pointer(nh->nh_exceptions, hash);
670 for (fnhe = rcu_dereference(hash->chain); fnhe;
671 fnhe = rcu_dereference(fnhe->fnhe_next)) {
672 if (fnhe->fnhe_daddr == daddr)
678 if (fnhe->fnhe_genid != genid)
679 fnhe->fnhe_genid = genid;
683 fnhe->fnhe_pmtu = pmtu;
684 fnhe->fnhe_mtu_locked = lock;
686 fnhe->fnhe_expires = max(1UL, expires);
687 /* Update all cached dsts too */
688 rt = rcu_dereference(fnhe->fnhe_rth_input);
690 fill_route_from_fnhe(rt, fnhe);
691 rt = rcu_dereference(fnhe->fnhe_rth_output);
693 fill_route_from_fnhe(rt, fnhe);
695 /* Randomize max depth to avoid some side channels attacks. */
696 int max_depth = FNHE_RECLAIM_DEPTH +
697 prandom_u32_max(FNHE_RECLAIM_DEPTH);
699 while (depth > max_depth) {
700 fnhe_remove_oldest(hash);
704 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
708 fnhe->fnhe_next = hash->chain;
710 fnhe->fnhe_genid = genid;
711 fnhe->fnhe_daddr = daddr;
713 fnhe->fnhe_pmtu = pmtu;
714 fnhe->fnhe_mtu_locked = lock;
715 fnhe->fnhe_expires = expires;
717 rcu_assign_pointer(hash->chain, fnhe);
719 /* Exception created; mark the cached routes for the nexthop
720 * stale, so anyone caching it rechecks if this exception
723 rt = rcu_dereference(nh->nh_rth_input);
725 rt->dst.obsolete = DST_OBSOLETE_KILL;
727 for_each_possible_cpu(i) {
728 struct rtable __rcu **prt;
729 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
730 rt = rcu_dereference(*prt);
732 rt->dst.obsolete = DST_OBSOLETE_KILL;
736 fnhe->fnhe_stamp = jiffies;
739 spin_unlock_bh(&fnhe_lock);
742 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
745 __be32 new_gw = icmp_hdr(skb)->un.gateway;
746 __be32 old_gw = ip_hdr(skb)->saddr;
747 struct net_device *dev = skb->dev;
748 struct in_device *in_dev;
749 struct fib_result res;
753 switch (icmp_hdr(skb)->code & 7) {
755 case ICMP_REDIR_NETTOS:
756 case ICMP_REDIR_HOST:
757 case ICMP_REDIR_HOSTTOS:
764 if (rt->rt_gateway != old_gw)
767 in_dev = __in_dev_get_rcu(dev);
772 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
773 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
774 ipv4_is_zeronet(new_gw))
775 goto reject_redirect;
777 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
778 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
779 goto reject_redirect;
780 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
781 goto reject_redirect;
783 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
784 goto reject_redirect;
787 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
789 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
791 if (!(n->nud_state & NUD_VALID)) {
792 neigh_event_send(n, NULL);
794 if (fib_lookup(net, fl4, &res, 0) == 0) {
795 struct fib_nh *nh = &FIB_RES_NH(res);
797 update_or_create_fnhe(nh, fl4->daddr, new_gw,
799 jiffies + ip_rt_gc_timeout);
802 rt->dst.obsolete = DST_OBSOLETE_KILL;
803 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
810 #ifdef CONFIG_IP_ROUTE_VERBOSE
811 if (IN_DEV_LOG_MARTIANS(in_dev)) {
812 const struct iphdr *iph = (const struct iphdr *) skb->data;
813 __be32 daddr = iph->daddr;
814 __be32 saddr = iph->saddr;
816 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
817 " Advised path = %pI4 -> %pI4\n",
818 &old_gw, dev->name, &new_gw,
825 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
829 const struct iphdr *iph = (const struct iphdr *) skb->data;
830 int oif = skb->dev->ifindex;
831 u8 tos = RT_TOS(iph->tos);
832 u8 prot = iph->protocol;
833 u32 mark = skb->mark;
835 rt = (struct rtable *) dst;
837 __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
838 __ip_do_redirect(rt, skb, &fl4, true);
841 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
843 struct rtable *rt = (struct rtable *)dst;
844 struct dst_entry *ret = dst;
847 if (dst->obsolete > 0) {
850 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
861 * 1. The first ip_rt_redirect_number redirects are sent
862 * with exponential backoff, then we stop sending them at all,
863 * assuming that the host ignores our redirects.
864 * 2. If we did not see packets requiring redirects
865 * during ip_rt_redirect_silence, we assume that the host
866 * forgot redirected route and start to send redirects again.
868 * This algorithm is much cheaper and more intelligent than dumb load limiting
871 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
872 * and "frag. need" (breaks PMTU discovery) in icmp.c.
875 void ip_rt_send_redirect(struct sk_buff *skb)
877 struct rtable *rt = skb_rtable(skb);
878 struct in_device *in_dev;
879 struct inet_peer *peer;
885 in_dev = __in_dev_get_rcu(rt->dst.dev);
886 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
890 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
891 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
894 net = dev_net(rt->dst.dev);
895 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
897 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
898 rt_nexthop(rt, ip_hdr(skb)->daddr));
902 /* No redirected packets during ip_rt_redirect_silence;
903 * reset the algorithm.
905 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
906 peer->rate_tokens = 0;
907 peer->n_redirects = 0;
910 /* Too many ignored redirects; do not send anything
911 * set dst.rate_last to the last seen redirected packet.
913 if (peer->n_redirects >= ip_rt_redirect_number) {
914 peer->rate_last = jiffies;
918 /* Check for load limit; set rate_last to the latest sent
921 if (peer->n_redirects == 0 ||
924 (ip_rt_redirect_load << peer->n_redirects)))) {
925 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
927 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
928 peer->rate_last = jiffies;
930 #ifdef CONFIG_IP_ROUTE_VERBOSE
932 peer->n_redirects == ip_rt_redirect_number)
933 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
934 &ip_hdr(skb)->saddr, inet_iif(skb),
935 &ip_hdr(skb)->daddr, &gw);
942 static int ip_error(struct sk_buff *skb)
944 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
945 struct rtable *rt = skb_rtable(skb);
946 struct inet_peer *peer;
952 /* IP on this device is disabled. */
956 net = dev_net(rt->dst.dev);
957 if (!IN_DEV_FORWARD(in_dev)) {
958 switch (rt->dst.error) {
960 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
964 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
970 switch (rt->dst.error) {
975 code = ICMP_HOST_UNREACH;
978 code = ICMP_NET_UNREACH;
979 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
982 code = ICMP_PKT_FILTERED;
986 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
987 l3mdev_master_ifindex(skb->dev), 1);
992 peer->rate_tokens += now - peer->rate_last;
993 if (peer->rate_tokens > ip_rt_error_burst)
994 peer->rate_tokens = ip_rt_error_burst;
995 peer->rate_last = now;
996 if (peer->rate_tokens >= ip_rt_error_cost)
997 peer->rate_tokens -= ip_rt_error_cost;
1003 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1005 out: kfree_skb(skb);
1009 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1011 struct dst_entry *dst = &rt->dst;
1012 u32 old_mtu = ipv4_mtu(dst);
1013 struct fib_result res;
1016 if (ip_mtu_locked(dst))
1022 if (mtu < ip_rt_min_pmtu) {
1024 mtu = min(old_mtu, ip_rt_min_pmtu);
1027 if (rt->rt_pmtu == mtu && !lock &&
1028 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1032 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1033 struct fib_nh *nh = &FIB_RES_NH(res);
1035 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1036 jiffies + ip_rt_mtu_expires);
1041 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1042 struct sk_buff *skb, u32 mtu)
1044 struct rtable *rt = (struct rtable *) dst;
1047 ip_rt_build_flow_key(&fl4, sk, skb);
1048 __ip_rt_update_pmtu(rt, &fl4, mtu);
1051 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1052 int oif, u32 mark, u8 protocol, int flow_flags)
1054 const struct iphdr *iph = (const struct iphdr *) skb->data;
1059 mark = IP4_REPLY_MARK(net, skb->mark);
1061 __build_flow_key(&fl4, NULL, iph, oif,
1062 RT_TOS(iph->tos), protocol, mark, flow_flags);
1063 rt = __ip_route_output_key(net, &fl4);
1065 __ip_rt_update_pmtu(rt, &fl4, mtu);
1069 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1071 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1073 const struct iphdr *iph = (const struct iphdr *) skb->data;
1077 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1079 if (!fl4.flowi4_mark)
1080 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1082 rt = __ip_route_output_key(sock_net(sk), &fl4);
1084 __ip_rt_update_pmtu(rt, &fl4, mtu);
1089 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1091 const struct iphdr *iph = (const struct iphdr *) skb->data;
1094 struct dst_entry *odst = NULL;
1099 if (!ip_sk_accept_pmtu(sk))
1102 odst = sk_dst_get(sk);
1104 if (sock_owned_by_user(sk) || !odst) {
1105 __ipv4_sk_update_pmtu(skb, sk, mtu);
1109 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1111 rt = (struct rtable *)odst;
1112 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1113 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1120 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1122 if (!dst_check(&rt->dst, 0)) {
1124 dst_release(&rt->dst);
1126 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1134 sk_dst_set(sk, &rt->dst);
1140 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1142 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1143 int oif, u32 mark, u8 protocol, int flow_flags)
1145 const struct iphdr *iph = (const struct iphdr *) skb->data;
1149 __build_flow_key(&fl4, NULL, iph, oif,
1150 RT_TOS(iph->tos), protocol, mark, flow_flags);
1151 rt = __ip_route_output_key(net, &fl4);
1153 __ip_do_redirect(rt, skb, &fl4, false);
1157 EXPORT_SYMBOL_GPL(ipv4_redirect);
1159 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1161 const struct iphdr *iph = (const struct iphdr *) skb->data;
1165 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1166 rt = __ip_route_output_key(sock_net(sk), &fl4);
1168 __ip_do_redirect(rt, skb, &fl4, false);
1172 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1174 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1176 struct rtable *rt = (struct rtable *) dst;
1178 /* All IPV4 dsts are created with ->obsolete set to the value
1179 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1180 * into this function always.
1182 * When a PMTU/redirect information update invalidates a route,
1183 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1184 * DST_OBSOLETE_DEAD by dst_free().
1186 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1191 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1193 struct ip_options opt;
1196 /* Recompile ip options since IPCB may not be valid anymore.
1197 * Also check we have a reasonable ipv4 header.
1199 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1200 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1203 memset(&opt, 0, sizeof(opt));
1204 if (ip_hdr(skb)->ihl > 5) {
1205 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1207 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1210 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1216 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1219 static void ipv4_link_failure(struct sk_buff *skb)
1223 ipv4_send_dest_unreach(skb);
1225 rt = skb_rtable(skb);
1227 dst_set_expires(&rt->dst, 0);
1230 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1232 pr_debug("%s: %pI4 -> %pI4, %s\n",
1233 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1234 skb->dev ? skb->dev->name : "?");
1241 We do not cache source address of outgoing interface,
1242 because it is used only by IP RR, TS and SRR options,
1243 so that it out of fast path.
1245 BTW remember: "addr" is allowed to be not aligned
1249 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1253 if (rt_is_output_route(rt))
1254 src = ip_hdr(skb)->saddr;
1256 struct fib_result res;
1262 memset(&fl4, 0, sizeof(fl4));
1263 fl4.daddr = iph->daddr;
1264 fl4.saddr = iph->saddr;
1265 fl4.flowi4_tos = RT_TOS(iph->tos);
1266 fl4.flowi4_oif = rt->dst.dev->ifindex;
1267 fl4.flowi4_iif = skb->dev->ifindex;
1268 fl4.flowi4_mark = skb->mark;
1271 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1272 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1274 src = inet_select_addr(rt->dst.dev,
1275 rt_nexthop(rt, iph->daddr),
1279 memcpy(addr, &src, 4);
1282 #ifdef CONFIG_IP_ROUTE_CLASSID
1283 static void set_class_tag(struct rtable *rt, u32 tag)
1285 if (!(rt->dst.tclassid & 0xFFFF))
1286 rt->dst.tclassid |= tag & 0xFFFF;
1287 if (!(rt->dst.tclassid & 0xFFFF0000))
1288 rt->dst.tclassid |= tag & 0xFFFF0000;
1292 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1294 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1297 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1299 if (advmss > 65535 - 40)
1300 advmss = 65535 - 40;
1305 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1307 const struct rtable *rt = (const struct rtable *) dst;
1308 unsigned int mtu = rt->rt_pmtu;
1310 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1311 mtu = dst_metric_raw(dst, RTAX_MTU);
1316 mtu = READ_ONCE(dst->dev->mtu);
1318 if (unlikely(ip_mtu_locked(dst))) {
1319 if (rt->rt_uses_gateway && mtu > 576)
1323 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1325 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1328 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1330 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1331 struct fib_nh_exception *fnhe;
1337 hval = fnhe_hashfun(daddr);
1339 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1340 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1341 if (fnhe->fnhe_daddr == daddr)
1347 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1352 spin_lock_bh(&fnhe_lock);
1354 if (daddr == fnhe->fnhe_daddr) {
1355 struct rtable __rcu **porig;
1356 struct rtable *orig;
1357 int genid = fnhe_genid(dev_net(rt->dst.dev));
1359 if (rt_is_input_route(rt))
1360 porig = &fnhe->fnhe_rth_input;
1362 porig = &fnhe->fnhe_rth_output;
1363 orig = rcu_dereference(*porig);
1365 if (fnhe->fnhe_genid != genid) {
1366 fnhe->fnhe_genid = genid;
1368 fnhe->fnhe_pmtu = 0;
1369 fnhe->fnhe_expires = 0;
1370 fnhe_flush_routes(fnhe);
1373 fill_route_from_fnhe(rt, fnhe);
1374 if (!rt->rt_gateway)
1375 rt->rt_gateway = daddr;
1377 if (!(rt->dst.flags & DST_NOCACHE)) {
1378 rcu_assign_pointer(*porig, rt);
1384 fnhe->fnhe_stamp = jiffies;
1386 spin_unlock_bh(&fnhe_lock);
1391 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1393 struct rtable *orig, *prev, **p;
1396 if (rt_is_input_route(rt)) {
1397 p = (struct rtable **)&nh->nh_rth_input;
1399 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1403 prev = cmpxchg(p, orig, rt);
1413 struct uncached_list {
1415 struct list_head head;
1418 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1420 static void rt_add_uncached_list(struct rtable *rt)
1422 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1424 rt->rt_uncached_list = ul;
1426 spin_lock_bh(&ul->lock);
1427 list_add_tail(&rt->rt_uncached, &ul->head);
1428 spin_unlock_bh(&ul->lock);
1431 static void ipv4_dst_destroy(struct dst_entry *dst)
1433 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1434 struct rtable *rt = (struct rtable *) dst;
1436 if (p != &dst_default_metrics && atomic_dec_and_test(&p->refcnt))
1439 if (!list_empty(&rt->rt_uncached)) {
1440 struct uncached_list *ul = rt->rt_uncached_list;
1442 spin_lock_bh(&ul->lock);
1443 list_del(&rt->rt_uncached);
1444 spin_unlock_bh(&ul->lock);
1448 void rt_flush_dev(struct net_device *dev)
1450 struct net *net = dev_net(dev);
1454 for_each_possible_cpu(cpu) {
1455 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1457 spin_lock_bh(&ul->lock);
1458 list_for_each_entry(rt, &ul->head, rt_uncached) {
1459 if (rt->dst.dev != dev)
1461 rt->dst.dev = net->loopback_dev;
1462 dev_hold(rt->dst.dev);
1465 spin_unlock_bh(&ul->lock);
1469 static bool rt_cache_valid(const struct rtable *rt)
1472 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1476 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1477 const struct fib_result *res,
1478 struct fib_nh_exception *fnhe,
1479 struct fib_info *fi, u16 type, u32 itag)
1481 bool cached = false;
1484 struct fib_nh *nh = &FIB_RES_NH(*res);
1486 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1487 rt->rt_gateway = nh->nh_gw;
1488 rt->rt_uses_gateway = 1;
1490 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1491 if (fi->fib_metrics != &dst_default_metrics) {
1492 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1493 atomic_inc(&fi->fib_metrics->refcnt);
1495 #ifdef CONFIG_IP_ROUTE_CLASSID
1496 rt->dst.tclassid = nh->nh_tclassid;
1498 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1500 cached = rt_bind_exception(rt, fnhe, daddr);
1501 else if (!(rt->dst.flags & DST_NOCACHE))
1502 cached = rt_cache_route(nh, rt);
1503 if (unlikely(!cached)) {
1504 /* Routes we intend to cache in nexthop exception or
1505 * FIB nexthop have the DST_NOCACHE bit clear.
1506 * However, if we are unsuccessful at storing this
1507 * route into the cache we really need to set it.
1509 rt->dst.flags |= DST_NOCACHE;
1510 if (!rt->rt_gateway)
1511 rt->rt_gateway = daddr;
1512 rt_add_uncached_list(rt);
1515 rt_add_uncached_list(rt);
1517 #ifdef CONFIG_IP_ROUTE_CLASSID
1518 #ifdef CONFIG_IP_MULTIPLE_TABLES
1519 set_class_tag(rt, res->tclassid);
1521 set_class_tag(rt, itag);
1525 struct rtable *rt_dst_alloc(struct net_device *dev,
1526 unsigned int flags, u16 type,
1527 bool nopolicy, bool noxfrm, bool will_cache)
1531 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1532 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1533 (nopolicy ? DST_NOPOLICY : 0) |
1534 (noxfrm ? DST_NOXFRM : 0));
1537 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1538 rt->rt_flags = flags;
1540 rt->rt_is_input = 0;
1543 rt->rt_mtu_locked = 0;
1545 rt->rt_uses_gateway = 0;
1546 rt->rt_table_id = 0;
1547 INIT_LIST_HEAD(&rt->rt_uncached);
1549 rt->dst.output = ip_output;
1550 if (flags & RTCF_LOCAL)
1551 rt->dst.input = ip_local_deliver;
1556 EXPORT_SYMBOL(rt_dst_alloc);
1558 /* called in rcu_read_lock() section */
1559 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1560 u8 tos, struct net_device *dev, int our)
1563 struct in_device *in_dev = __in_dev_get_rcu(dev);
1564 unsigned int flags = RTCF_MULTICAST;
1568 /* Primary sanity checks. */
1573 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1574 skb->protocol != htons(ETH_P_IP))
1577 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1580 if (ipv4_is_zeronet(saddr)) {
1581 if (!ipv4_is_local_multicast(daddr))
1584 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1590 flags |= RTCF_LOCAL;
1592 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1593 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1597 #ifdef CONFIG_IP_ROUTE_CLASSID
1598 rth->dst.tclassid = itag;
1600 rth->dst.output = ip_rt_bug;
1601 rth->rt_is_input= 1;
1603 #ifdef CONFIG_IP_MROUTE
1604 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1605 rth->dst.input = ip_mr_input;
1607 RT_CACHE_STAT_INC(in_slow_mc);
1609 skb_dst_set(skb, &rth->dst);
1621 static void ip_handle_martian_source(struct net_device *dev,
1622 struct in_device *in_dev,
1623 struct sk_buff *skb,
1627 RT_CACHE_STAT_INC(in_martian_src);
1628 #ifdef CONFIG_IP_ROUTE_VERBOSE
1629 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1631 * RFC1812 recommendation, if source is martian,
1632 * the only hint is MAC header.
1634 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1635 &daddr, &saddr, dev->name);
1636 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1637 print_hex_dump(KERN_WARNING, "ll header: ",
1638 DUMP_PREFIX_OFFSET, 16, 1,
1639 skb_mac_header(skb),
1640 dev->hard_header_len, true);
1646 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1648 struct fnhe_hash_bucket *hash;
1649 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1650 u32 hval = fnhe_hashfun(daddr);
1652 spin_lock_bh(&fnhe_lock);
1654 hash = rcu_dereference_protected(nh->nh_exceptions,
1655 lockdep_is_held(&fnhe_lock));
1658 fnhe_p = &hash->chain;
1659 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1661 if (fnhe->fnhe_daddr == daddr) {
1662 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1663 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1664 /* set fnhe_daddr to 0 to ensure it won't bind with
1665 * new dsts in rt_bind_exception().
1667 fnhe->fnhe_daddr = 0;
1668 fnhe_flush_routes(fnhe);
1669 kfree_rcu(fnhe, rcu);
1672 fnhe_p = &fnhe->fnhe_next;
1673 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1674 lockdep_is_held(&fnhe_lock));
1677 spin_unlock_bh(&fnhe_lock);
1680 /* called in rcu_read_lock() section */
1681 static int __mkroute_input(struct sk_buff *skb,
1682 const struct fib_result *res,
1683 struct in_device *in_dev,
1684 __be32 daddr, __be32 saddr, u32 tos)
1686 struct fib_nh_exception *fnhe;
1689 struct in_device *out_dev;
1693 /* get a working reference to the output device */
1694 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1696 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1700 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1701 in_dev->dev, in_dev, &itag);
1703 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1709 do_cache = res->fi && !itag;
1710 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1711 skb->protocol == htons(ETH_P_IP) &&
1712 (IN_DEV_SHARED_MEDIA(out_dev) ||
1713 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1714 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1716 if (skb->protocol != htons(ETH_P_IP)) {
1717 /* Not IP (i.e. ARP). Do not create route, if it is
1718 * invalid for proxy arp. DNAT routes are always valid.
1720 * Proxy arp feature have been extended to allow, ARP
1721 * replies back to the same interface, to support
1722 * Private VLAN switch technologies. See arp.c.
1724 if (out_dev == in_dev &&
1725 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1731 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1734 rth = rcu_dereference(fnhe->fnhe_rth_input);
1735 if (rth && rth->dst.expires &&
1736 time_after(jiffies, rth->dst.expires)) {
1737 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1744 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1747 if (rt_cache_valid(rth)) {
1748 skb_dst_set_noref(skb, &rth->dst);
1753 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1754 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1755 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1761 rth->rt_is_input = 1;
1763 rth->rt_table_id = res->table->tb_id;
1764 RT_CACHE_STAT_INC(in_slow_tot);
1766 rth->dst.input = ip_forward;
1768 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1769 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1770 rth->dst.lwtstate->orig_output = rth->dst.output;
1771 rth->dst.output = lwtunnel_output;
1773 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1774 rth->dst.lwtstate->orig_input = rth->dst.input;
1775 rth->dst.input = lwtunnel_input;
1777 skb_dst_set(skb, &rth->dst);
1784 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1786 /* To make ICMP packets follow the right flow, the multipath hash is
1787 * calculated from the inner IP addresses in reverse order.
1789 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1791 const struct iphdr *outer_iph = ip_hdr(skb);
1792 struct icmphdr _icmph;
1793 const struct icmphdr *icmph;
1794 struct iphdr _inner_iph;
1795 const struct iphdr *inner_iph;
1797 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1800 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1805 if (icmph->type != ICMP_DEST_UNREACH &&
1806 icmph->type != ICMP_REDIRECT &&
1807 icmph->type != ICMP_TIME_EXCEEDED &&
1808 icmph->type != ICMP_PARAMETERPROB) {
1812 inner_iph = skb_header_pointer(skb,
1813 outer_iph->ihl * 4 + sizeof(_icmph),
1814 sizeof(_inner_iph), &_inner_iph);
1818 return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1821 return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1824 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1826 static int ip_mkroute_input(struct sk_buff *skb,
1827 struct fib_result *res,
1828 const struct flowi4 *fl4,
1829 struct in_device *in_dev,
1830 __be32 daddr, __be32 saddr, u32 tos)
1832 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1833 if (res->fi && res->fi->fib_nhs > 1) {
1836 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1837 h = ip_multipath_icmp_hash(skb);
1839 h = fib_multipath_hash(saddr, daddr);
1840 fib_select_multipath(res, h);
1844 /* create a routing cache entry */
1845 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1849 * NOTE. We drop all the packets that has local source
1850 * addresses, because every properly looped back packet
1851 * must have correct destination already attached by output routine.
1853 * Such approach solves two big problems:
1854 * 1. Not simplex devices are handled properly.
1855 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1856 * called with rcu_read_lock()
1859 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1860 u8 tos, struct net_device *dev)
1862 struct fib_result res;
1863 struct in_device *in_dev = __in_dev_get_rcu(dev);
1864 struct ip_tunnel_info *tun_info;
1866 unsigned int flags = 0;
1870 struct net *net = dev_net(dev);
1873 /* IP on this device is disabled. */
1878 /* Check for the most weird martians, which can be not detected
1882 tun_info = skb_tunnel_info(skb);
1883 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1884 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1886 fl4.flowi4_tun_key.tun_id = 0;
1889 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1890 goto martian_source;
1894 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1897 /* Accept zero addresses only to limited broadcast;
1898 * I even do not know to fix it or not. Waiting for complains :-)
1900 if (ipv4_is_zeronet(saddr))
1901 goto martian_source;
1903 if (ipv4_is_zeronet(daddr))
1904 goto martian_destination;
1906 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1907 * and call it once if daddr or/and saddr are loopback addresses
1909 if (ipv4_is_loopback(daddr)) {
1910 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1911 goto martian_destination;
1912 } else if (ipv4_is_loopback(saddr)) {
1913 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1914 goto martian_source;
1918 * Now we are ready to route packet.
1921 fl4.flowi4_iif = dev->ifindex;
1922 fl4.flowi4_mark = skb->mark;
1923 fl4.flowi4_tos = tos;
1924 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1925 fl4.flowi4_flags = 0;
1928 err = fib_lookup(net, &fl4, &res, 0);
1930 if (!IN_DEV_FORWARD(in_dev))
1931 err = -EHOSTUNREACH;
1935 if (res.type == RTN_BROADCAST)
1938 if (res.type == RTN_LOCAL) {
1939 err = fib_validate_source(skb, saddr, daddr, tos,
1940 0, dev, in_dev, &itag);
1942 goto martian_source;
1946 if (!IN_DEV_FORWARD(in_dev)) {
1947 err = -EHOSTUNREACH;
1950 if (res.type != RTN_UNICAST)
1951 goto martian_destination;
1953 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1957 if (skb->protocol != htons(ETH_P_IP))
1960 if (!ipv4_is_zeronet(saddr)) {
1961 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1964 goto martian_source;
1966 flags |= RTCF_BROADCAST;
1967 res.type = RTN_BROADCAST;
1968 RT_CACHE_STAT_INC(in_brd);
1974 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1975 if (rt_cache_valid(rth)) {
1976 skb_dst_set_noref(skb, &rth->dst);
1984 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
1985 flags | RTCF_LOCAL, res.type,
1986 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1990 rth->dst.output= ip_rt_bug;
1991 #ifdef CONFIG_IP_ROUTE_CLASSID
1992 rth->dst.tclassid = itag;
1994 rth->rt_is_input = 1;
1996 rth->rt_table_id = res.table->tb_id;
1998 RT_CACHE_STAT_INC(in_slow_tot);
1999 if (res.type == RTN_UNREACHABLE) {
2000 rth->dst.input= ip_error;
2001 rth->dst.error= -err;
2002 rth->rt_flags &= ~RTCF_LOCAL;
2005 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
2006 rth->dst.flags |= DST_NOCACHE;
2007 rt_add_uncached_list(rth);
2010 skb_dst_set(skb, &rth->dst);
2015 RT_CACHE_STAT_INC(in_no_route);
2016 res.type = RTN_UNREACHABLE;
2022 * Do not cache martian addresses: they should be logged (RFC1812)
2024 martian_destination:
2025 RT_CACHE_STAT_INC(in_martian_dst);
2026 #ifdef CONFIG_IP_ROUTE_VERBOSE
2027 if (IN_DEV_LOG_MARTIANS(in_dev))
2028 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2029 &daddr, &saddr, dev->name);
2041 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2045 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2046 u8 tos, struct net_device *dev)
2050 tos &= IPTOS_RT_MASK;
2053 /* Multicast recognition logic is moved from route cache to here.
2054 The problem was that too many Ethernet cards have broken/missing
2055 hardware multicast filters :-( As result the host on multicasting
2056 network acquires a lot of useless route cache entries, sort of
2057 SDR messages from all the world. Now we try to get rid of them.
2058 Really, provided software IP multicast filter is organized
2059 reasonably (at least, hashed), it does not result in a slowdown
2060 comparing with route cache reject entries.
2061 Note, that multicast routers are not affected, because
2062 route cache entry is created eventually.
2064 if (ipv4_is_multicast(daddr)) {
2065 struct in_device *in_dev = __in_dev_get_rcu(dev);
2068 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2069 ip_hdr(skb)->protocol);
2071 #ifdef CONFIG_IP_MROUTE
2073 (!ipv4_is_local_multicast(daddr) &&
2074 IN_DEV_MFORWARD(in_dev))
2077 int res = ip_route_input_mc(skb, daddr, saddr,
2086 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2090 EXPORT_SYMBOL(ip_route_input_noref);
2092 /* called with rcu_read_lock() */
2093 static struct rtable *__mkroute_output(const struct fib_result *res,
2094 const struct flowi4 *fl4, int orig_oif,
2095 struct net_device *dev_out,
2098 struct fib_info *fi = res->fi;
2099 struct fib_nh_exception *fnhe;
2100 struct in_device *in_dev;
2101 u16 type = res->type;
2105 in_dev = __in_dev_get_rcu(dev_out);
2107 return ERR_PTR(-EINVAL);
2109 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2110 if (ipv4_is_loopback(fl4->saddr) &&
2111 !(dev_out->flags & IFF_LOOPBACK) &&
2112 !netif_is_l3_master(dev_out))
2113 return ERR_PTR(-EINVAL);
2115 if (ipv4_is_lbcast(fl4->daddr))
2116 type = RTN_BROADCAST;
2117 else if (ipv4_is_multicast(fl4->daddr))
2118 type = RTN_MULTICAST;
2119 else if (ipv4_is_zeronet(fl4->daddr))
2120 return ERR_PTR(-EINVAL);
2122 if (dev_out->flags & IFF_LOOPBACK)
2123 flags |= RTCF_LOCAL;
2126 if (type == RTN_BROADCAST) {
2127 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2129 } else if (type == RTN_MULTICAST) {
2130 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2131 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2133 flags &= ~RTCF_LOCAL;
2136 /* If multicast route do not exist use
2137 * default one, but do not gateway in this case.
2140 if (fi && res->prefixlen < 4)
2142 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2143 (orig_oif != dev_out->ifindex)) {
2144 /* For local routes that require a particular output interface
2145 * we do not want to cache the result. Caching the result
2146 * causes incorrect behaviour when there are multiple source
2147 * addresses on the interface, the end result being that if the
2148 * intended recipient is waiting on that interface for the
2149 * packet he won't receive it because it will be delivered on
2150 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2151 * be set to the loopback interface as well.
2157 do_cache &= fi != NULL;
2159 struct rtable __rcu **prth;
2160 struct fib_nh *nh = &FIB_RES_NH(*res);
2162 fnhe = find_exception(nh, fl4->daddr);
2164 prth = &fnhe->fnhe_rth_output;
2165 rth = rcu_dereference(*prth);
2166 if (rth && rth->dst.expires &&
2167 time_after(jiffies, rth->dst.expires)) {
2168 ip_del_fnhe(nh, fl4->daddr);
2175 if (unlikely(fl4->flowi4_flags &
2176 FLOWI_FLAG_KNOWN_NH &&
2178 nh->nh_scope == RT_SCOPE_LINK))) {
2182 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2183 rth = rcu_dereference(*prth);
2186 if (rt_cache_valid(rth)) {
2187 dst_hold(&rth->dst);
2193 rth = rt_dst_alloc(dev_out, flags, type,
2194 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2195 IN_DEV_CONF_GET(in_dev, NOXFRM),
2198 return ERR_PTR(-ENOBUFS);
2200 rth->rt_iif = orig_oif ? : 0;
2202 rth->rt_table_id = res->table->tb_id;
2204 RT_CACHE_STAT_INC(out_slow_tot);
2206 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2207 if (flags & RTCF_LOCAL &&
2208 !(dev_out->flags & IFF_LOOPBACK)) {
2209 rth->dst.output = ip_mc_output;
2210 RT_CACHE_STAT_INC(out_slow_mc);
2212 #ifdef CONFIG_IP_MROUTE
2213 if (type == RTN_MULTICAST) {
2214 if (IN_DEV_MFORWARD(in_dev) &&
2215 !ipv4_is_local_multicast(fl4->daddr)) {
2216 rth->dst.input = ip_mr_input;
2217 rth->dst.output = ip_mc_output;
2223 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2224 if (lwtunnel_output_redirect(rth->dst.lwtstate))
2225 rth->dst.output = lwtunnel_output;
2231 * Major route resolver routine.
2234 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2237 struct net_device *dev_out = NULL;
2238 __u8 tos = RT_FL_TOS(fl4);
2239 unsigned int flags = 0;
2240 struct fib_result res;
2249 orig_oif = fl4->flowi4_oif;
2251 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2252 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2253 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2254 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2258 if (ipv4_is_multicast(fl4->saddr) ||
2259 ipv4_is_lbcast(fl4->saddr) ||
2260 ipv4_is_zeronet(fl4->saddr)) {
2261 rth = ERR_PTR(-EINVAL);
2265 rth = ERR_PTR(-ENETUNREACH);
2267 /* I removed check for oif == dev_out->oif here.
2268 It was wrong for two reasons:
2269 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2270 is assigned to multiple interfaces.
2271 2. Moreover, we are allowed to send packets with saddr
2272 of another iface. --ANK
2275 if (fl4->flowi4_oif == 0 &&
2276 (ipv4_is_multicast(fl4->daddr) ||
2277 ipv4_is_lbcast(fl4->daddr))) {
2278 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2279 dev_out = __ip_dev_find(net, fl4->saddr, false);
2283 /* Special hack: user can direct multicasts
2284 and limited broadcast via necessary interface
2285 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2286 This hack is not just for fun, it allows
2287 vic,vat and friends to work.
2288 They bind socket to loopback, set ttl to zero
2289 and expect that it will work.
2290 From the viewpoint of routing cache they are broken,
2291 because we are not allowed to build multicast path
2292 with loopback source addr (look, routing cache
2293 cannot know, that ttl is zero, so that packet
2294 will not leave this host and route is valid).
2295 Luckily, this hack is good workaround.
2298 fl4->flowi4_oif = dev_out->ifindex;
2302 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2303 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2304 if (!__ip_dev_find(net, fl4->saddr, false))
2310 if (fl4->flowi4_oif) {
2311 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2312 rth = ERR_PTR(-ENODEV);
2316 /* RACE: Check return value of inet_select_addr instead. */
2317 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2318 rth = ERR_PTR(-ENETUNREACH);
2321 if (ipv4_is_local_multicast(fl4->daddr) ||
2322 ipv4_is_lbcast(fl4->daddr) ||
2323 fl4->flowi4_proto == IPPROTO_IGMP) {
2325 fl4->saddr = inet_select_addr(dev_out, 0,
2330 if (ipv4_is_multicast(fl4->daddr))
2331 fl4->saddr = inet_select_addr(dev_out, 0,
2333 else if (!fl4->daddr)
2334 fl4->saddr = inet_select_addr(dev_out, 0,
2340 fl4->daddr = fl4->saddr;
2342 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2343 dev_out = net->loopback_dev;
2344 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2345 res.type = RTN_LOCAL;
2346 flags |= RTCF_LOCAL;
2350 err = fib_lookup(net, fl4, &res, 0);
2354 if (fl4->flowi4_oif &&
2355 !netif_index_is_l3_master(net, fl4->flowi4_oif)) {
2356 /* Apparently, routing tables are wrong. Assume,
2357 that the destination is on link.
2360 Because we are allowed to send to iface
2361 even if it has NO routes and NO assigned
2362 addresses. When oif is specified, routing
2363 tables are looked up with only one purpose:
2364 to catch if destination is gatewayed, rather than
2365 direct. Moreover, if MSG_DONTROUTE is set,
2366 we send packet, ignoring both routing tables
2367 and ifaddr state. --ANK
2370 We could make it even if oif is unknown,
2371 likely IPv6, but we do not.
2374 if (fl4->saddr == 0)
2375 fl4->saddr = inet_select_addr(dev_out, 0,
2377 res.type = RTN_UNICAST;
2384 if (res.type == RTN_LOCAL) {
2386 if (res.fi->fib_prefsrc)
2387 fl4->saddr = res.fi->fib_prefsrc;
2389 fl4->saddr = fl4->daddr;
2392 /* L3 master device is the loopback for that domain */
2393 dev_out = l3mdev_master_dev_rcu(dev_out) ? : net->loopback_dev;
2394 fl4->flowi4_oif = dev_out->ifindex;
2395 flags |= RTCF_LOCAL;
2399 fib_select_path(net, &res, fl4, mp_hash);
2401 dev_out = FIB_RES_DEV(res);
2402 fl4->flowi4_oif = dev_out->ifindex;
2406 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2412 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2414 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2419 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2421 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2423 return mtu ? : dst->dev->mtu;
2426 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2427 struct sk_buff *skb, u32 mtu)
2431 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2432 struct sk_buff *skb)
2436 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2442 static struct dst_ops ipv4_dst_blackhole_ops = {
2444 .check = ipv4_blackhole_dst_check,
2445 .mtu = ipv4_blackhole_mtu,
2446 .default_advmss = ipv4_default_advmss,
2447 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2448 .redirect = ipv4_rt_blackhole_redirect,
2449 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2450 .neigh_lookup = ipv4_neigh_lookup,
2453 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2455 struct rtable *ort = (struct rtable *) dst_orig;
2458 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2460 struct dst_entry *new = &rt->dst;
2463 new->input = dst_discard;
2464 new->output = dst_discard_out;
2466 new->dev = ort->dst.dev;
2470 rt->rt_is_input = ort->rt_is_input;
2471 rt->rt_iif = ort->rt_iif;
2472 rt->rt_pmtu = ort->rt_pmtu;
2473 rt->rt_mtu_locked = ort->rt_mtu_locked;
2475 rt->rt_genid = rt_genid_ipv4(net);
2476 rt->rt_flags = ort->rt_flags;
2477 rt->rt_type = ort->rt_type;
2478 rt->rt_gateway = ort->rt_gateway;
2479 rt->rt_uses_gateway = ort->rt_uses_gateway;
2481 INIT_LIST_HEAD(&rt->rt_uncached);
2485 dst_release(dst_orig);
2487 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2490 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2491 const struct sock *sk)
2493 struct rtable *rt = __ip_route_output_key(net, flp4);
2498 if (flp4->flowi4_proto)
2499 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2500 flowi4_to_flowi(flp4),
2505 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2507 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2508 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2509 u32 seq, int event, int nowait, unsigned int flags)
2511 struct rtable *rt = skb_rtable(skb);
2513 struct nlmsghdr *nlh;
2514 unsigned long expires = 0;
2516 u32 metrics[RTAX_MAX];
2518 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2522 r = nlmsg_data(nlh);
2523 r->rtm_family = AF_INET;
2524 r->rtm_dst_len = 32;
2526 r->rtm_tos = fl4->flowi4_tos;
2527 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2528 if (nla_put_u32(skb, RTA_TABLE, table_id))
2529 goto nla_put_failure;
2530 r->rtm_type = rt->rt_type;
2531 r->rtm_scope = RT_SCOPE_UNIVERSE;
2532 r->rtm_protocol = RTPROT_UNSPEC;
2533 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2534 if (rt->rt_flags & RTCF_NOTIFY)
2535 r->rtm_flags |= RTM_F_NOTIFY;
2536 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2537 r->rtm_flags |= RTCF_DOREDIRECT;
2539 if (nla_put_in_addr(skb, RTA_DST, dst))
2540 goto nla_put_failure;
2542 r->rtm_src_len = 32;
2543 if (nla_put_in_addr(skb, RTA_SRC, src))
2544 goto nla_put_failure;
2547 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2548 goto nla_put_failure;
2549 #ifdef CONFIG_IP_ROUTE_CLASSID
2550 if (rt->dst.tclassid &&
2551 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2552 goto nla_put_failure;
2554 if (!rt_is_input_route(rt) &&
2555 fl4->saddr != src) {
2556 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2557 goto nla_put_failure;
2559 if (rt->rt_uses_gateway &&
2560 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2561 goto nla_put_failure;
2563 expires = rt->dst.expires;
2565 unsigned long now = jiffies;
2567 if (time_before(now, expires))
2573 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2574 if (rt->rt_pmtu && expires)
2575 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2576 if (rt->rt_mtu_locked && expires)
2577 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2578 if (rtnetlink_put_metrics(skb, metrics) < 0)
2579 goto nla_put_failure;
2581 if (fl4->flowi4_mark &&
2582 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2583 goto nla_put_failure;
2585 error = rt->dst.error;
2587 if (rt_is_input_route(rt)) {
2588 #ifdef CONFIG_IP_MROUTE
2589 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2590 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2591 int err = ipmr_get_route(net, skb,
2592 fl4->saddr, fl4->daddr,
2599 goto nla_put_failure;
2601 if (err == -EMSGSIZE)
2602 goto nla_put_failure;
2608 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2609 goto nla_put_failure;
2612 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2613 goto nla_put_failure;
2615 nlmsg_end(skb, nlh);
2619 nlmsg_cancel(skb, nlh);
2623 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2625 struct net *net = sock_net(in_skb->sk);
2627 struct nlattr *tb[RTA_MAX+1];
2628 struct rtable *rt = NULL;
2635 struct sk_buff *skb;
2636 u32 table_id = RT_TABLE_MAIN;
2638 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2642 rtm = nlmsg_data(nlh);
2644 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2650 /* Reserve room for dummy headers, this skb can pass
2651 through good chunk of routing engine.
2653 skb_reset_mac_header(skb);
2654 skb_reset_network_header(skb);
2656 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2657 ip_hdr(skb)->protocol = IPPROTO_UDP;
2658 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2660 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2661 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2662 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2663 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2665 memset(&fl4, 0, sizeof(fl4));
2668 fl4.flowi4_tos = rtm->rtm_tos;
2669 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2670 fl4.flowi4_mark = mark;
2673 struct net_device *dev;
2675 dev = __dev_get_by_index(net, iif);
2681 skb->protocol = htons(ETH_P_IP);
2685 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2688 rt = skb_rtable(skb);
2689 if (err == 0 && rt->dst.error)
2690 err = -rt->dst.error;
2692 rt = ip_route_output_key(net, &fl4);
2702 skb_dst_set(skb, &rt->dst);
2703 if (rtm->rtm_flags & RTM_F_NOTIFY)
2704 rt->rt_flags |= RTCF_NOTIFY;
2706 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2707 table_id = rt->rt_table_id;
2709 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2710 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2711 RTM_NEWROUTE, 0, 0);
2715 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2724 void ip_rt_multicast_event(struct in_device *in_dev)
2726 rt_cache_flush(dev_net(in_dev->dev));
2729 #ifdef CONFIG_SYSCTL
2730 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2731 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2732 static int ip_rt_gc_elasticity __read_mostly = 8;
2733 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
2735 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2736 void __user *buffer,
2737 size_t *lenp, loff_t *ppos)
2739 struct net *net = (struct net *)__ctl->extra1;
2742 rt_cache_flush(net);
2743 fnhe_genid_bump(net);
2750 static struct ctl_table ipv4_route_table[] = {
2752 .procname = "gc_thresh",
2753 .data = &ipv4_dst_ops.gc_thresh,
2754 .maxlen = sizeof(int),
2756 .proc_handler = proc_dointvec,
2759 .procname = "max_size",
2760 .data = &ip_rt_max_size,
2761 .maxlen = sizeof(int),
2763 .proc_handler = proc_dointvec,
2766 /* Deprecated. Use gc_min_interval_ms */
2768 .procname = "gc_min_interval",
2769 .data = &ip_rt_gc_min_interval,
2770 .maxlen = sizeof(int),
2772 .proc_handler = proc_dointvec_jiffies,
2775 .procname = "gc_min_interval_ms",
2776 .data = &ip_rt_gc_min_interval,
2777 .maxlen = sizeof(int),
2779 .proc_handler = proc_dointvec_ms_jiffies,
2782 .procname = "gc_timeout",
2783 .data = &ip_rt_gc_timeout,
2784 .maxlen = sizeof(int),
2786 .proc_handler = proc_dointvec_jiffies,
2789 .procname = "gc_interval",
2790 .data = &ip_rt_gc_interval,
2791 .maxlen = sizeof(int),
2793 .proc_handler = proc_dointvec_jiffies,
2796 .procname = "redirect_load",
2797 .data = &ip_rt_redirect_load,
2798 .maxlen = sizeof(int),
2800 .proc_handler = proc_dointvec,
2803 .procname = "redirect_number",
2804 .data = &ip_rt_redirect_number,
2805 .maxlen = sizeof(int),
2807 .proc_handler = proc_dointvec,
2810 .procname = "redirect_silence",
2811 .data = &ip_rt_redirect_silence,
2812 .maxlen = sizeof(int),
2814 .proc_handler = proc_dointvec,
2817 .procname = "error_cost",
2818 .data = &ip_rt_error_cost,
2819 .maxlen = sizeof(int),
2821 .proc_handler = proc_dointvec,
2824 .procname = "error_burst",
2825 .data = &ip_rt_error_burst,
2826 .maxlen = sizeof(int),
2828 .proc_handler = proc_dointvec,
2831 .procname = "gc_elasticity",
2832 .data = &ip_rt_gc_elasticity,
2833 .maxlen = sizeof(int),
2835 .proc_handler = proc_dointvec,
2838 .procname = "mtu_expires",
2839 .data = &ip_rt_mtu_expires,
2840 .maxlen = sizeof(int),
2842 .proc_handler = proc_dointvec_jiffies,
2845 .procname = "min_pmtu",
2846 .data = &ip_rt_min_pmtu,
2847 .maxlen = sizeof(int),
2849 .proc_handler = proc_dointvec_minmax,
2850 .extra1 = &ip_min_valid_pmtu,
2853 .procname = "min_adv_mss",
2854 .data = &ip_rt_min_advmss,
2855 .maxlen = sizeof(int),
2857 .proc_handler = proc_dointvec,
2862 static struct ctl_table ipv4_route_flush_table[] = {
2864 .procname = "flush",
2865 .maxlen = sizeof(int),
2867 .proc_handler = ipv4_sysctl_rtcache_flush,
2872 static __net_init int sysctl_route_net_init(struct net *net)
2874 struct ctl_table *tbl;
2876 tbl = ipv4_route_flush_table;
2877 if (!net_eq(net, &init_net)) {
2878 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2882 /* Don't export sysctls to unprivileged users */
2883 if (net->user_ns != &init_user_ns)
2884 tbl[0].procname = NULL;
2886 tbl[0].extra1 = net;
2888 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2889 if (!net->ipv4.route_hdr)
2894 if (tbl != ipv4_route_flush_table)
2900 static __net_exit void sysctl_route_net_exit(struct net *net)
2902 struct ctl_table *tbl;
2904 tbl = net->ipv4.route_hdr->ctl_table_arg;
2905 unregister_net_sysctl_table(net->ipv4.route_hdr);
2906 BUG_ON(tbl == ipv4_route_flush_table);
2910 static __net_initdata struct pernet_operations sysctl_route_ops = {
2911 .init = sysctl_route_net_init,
2912 .exit = sysctl_route_net_exit,
2916 static __net_init int rt_genid_init(struct net *net)
2918 atomic_set(&net->ipv4.rt_genid, 0);
2919 atomic_set(&net->fnhe_genid, 0);
2920 get_random_bytes(&net->ipv4.dev_addr_genid,
2921 sizeof(net->ipv4.dev_addr_genid));
2925 static __net_initdata struct pernet_operations rt_genid_ops = {
2926 .init = rt_genid_init,
2929 static int __net_init ipv4_inetpeer_init(struct net *net)
2931 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2935 inet_peer_base_init(bp);
2936 net->ipv4.peers = bp;
2940 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2942 struct inet_peer_base *bp = net->ipv4.peers;
2944 net->ipv4.peers = NULL;
2945 inetpeer_invalidate_tree(bp);
2949 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2950 .init = ipv4_inetpeer_init,
2951 .exit = ipv4_inetpeer_exit,
2954 #ifdef CONFIG_IP_ROUTE_CLASSID
2955 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2956 #endif /* CONFIG_IP_ROUTE_CLASSID */
2958 int __init ip_rt_init(void)
2964 /* For modern hosts, this will use 2 MB of memory */
2965 idents_hash = alloc_large_system_hash("IP idents",
2966 sizeof(*ip_idents) + sizeof(*ip_tstamps),
2968 16, /* one bucket per 64 KB */
2975 ip_idents = idents_hash;
2977 prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
2979 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
2980 memset(ip_tstamps, 0, (ip_idents_mask + 1) * sizeof(*ip_tstamps));
2982 for_each_possible_cpu(cpu) {
2983 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2985 INIT_LIST_HEAD(&ul->head);
2986 spin_lock_init(&ul->lock);
2988 #ifdef CONFIG_IP_ROUTE_CLASSID
2989 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2991 panic("IP: failed to allocate ip_rt_acct\n");
2994 ipv4_dst_ops.kmem_cachep =
2995 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2996 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2998 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3000 if (dst_entries_init(&ipv4_dst_ops) < 0)
3001 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3003 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3004 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3006 ipv4_dst_ops.gc_thresh = ~0;
3007 ip_rt_max_size = INT_MAX;
3012 if (ip_rt_proc_init())
3013 pr_err("Unable to create route proc files\n");
3018 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3020 #ifdef CONFIG_SYSCTL
3021 register_pernet_subsys(&sysctl_route_ops);
3023 register_pernet_subsys(&rt_genid_ops);
3024 register_pernet_subsys(&ipv4_inetpeer_ops);
3028 #ifdef CONFIG_SYSCTL
3030 * We really need to sanitize the damn ipv4 init order, then all
3031 * this nonsense will go away.
3033 void __init ip_static_sysctl_init(void)
3035 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);