2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <linux/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/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 #include "fib_lookup.h"
120 #define RT_FL_TOS(oldflp4) \
121 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
123 #define RT_GC_TIMEOUT (300*HZ)
125 static int ip_rt_max_size;
126 static int ip_rt_redirect_number __read_mostly = 9;
127 static int ip_rt_redirect_load __read_mostly = HZ / 50;
128 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
129 static int ip_rt_error_cost __read_mostly = HZ;
130 static int ip_rt_error_burst __read_mostly = 5 * HZ;
131 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
132 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
133 static int ip_rt_min_advmss __read_mostly = 256;
135 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
138 * Interface to generic destination cache.
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
143 static unsigned int ipv4_mtu(const struct dst_entry *dst);
144 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
145 static void ipv4_link_failure(struct sk_buff *skb);
146 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
147 struct sk_buff *skb, u32 mtu,
149 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
150 struct sk_buff *skb);
151 static void ipv4_dst_destroy(struct dst_entry *dst);
153 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
159 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
162 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
164 static struct dst_ops ipv4_dst_ops = {
166 .check = ipv4_dst_check,
167 .default_advmss = ipv4_default_advmss,
169 .cow_metrics = ipv4_cow_metrics,
170 .destroy = ipv4_dst_destroy,
171 .negative_advice = ipv4_negative_advice,
172 .link_failure = ipv4_link_failure,
173 .update_pmtu = ip_rt_update_pmtu,
174 .redirect = ip_do_redirect,
175 .local_out = __ip_local_out,
176 .neigh_lookup = ipv4_neigh_lookup,
177 .confirm_neigh = ipv4_confirm_neigh,
180 #define ECN_OR_COST(class) TC_PRIO_##class
182 const __u8 ip_tos2prio[16] = {
184 ECN_OR_COST(BESTEFFORT),
186 ECN_OR_COST(BESTEFFORT),
192 ECN_OR_COST(INTERACTIVE),
194 ECN_OR_COST(INTERACTIVE),
195 TC_PRIO_INTERACTIVE_BULK,
196 ECN_OR_COST(INTERACTIVE_BULK),
197 TC_PRIO_INTERACTIVE_BULK,
198 ECN_OR_COST(INTERACTIVE_BULK)
200 EXPORT_SYMBOL(ip_tos2prio);
202 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
203 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
205 #ifdef CONFIG_PROC_FS
206 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
210 return SEQ_START_TOKEN;
213 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
219 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
223 static int rt_cache_seq_show(struct seq_file *seq, void *v)
225 if (v == SEQ_START_TOKEN)
226 seq_printf(seq, "%-127s\n",
227 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
228 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
233 static const struct seq_operations rt_cache_seq_ops = {
234 .start = rt_cache_seq_start,
235 .next = rt_cache_seq_next,
236 .stop = rt_cache_seq_stop,
237 .show = rt_cache_seq_show,
240 static int rt_cache_seq_open(struct inode *inode, struct file *file)
242 return seq_open(file, &rt_cache_seq_ops);
245 static const struct file_operations rt_cache_seq_fops = {
246 .owner = THIS_MODULE,
247 .open = rt_cache_seq_open,
250 .release = seq_release,
254 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
259 return SEQ_START_TOKEN;
261 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
262 if (!cpu_possible(cpu))
265 return &per_cpu(rt_cache_stat, cpu);
270 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
274 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
275 if (!cpu_possible(cpu))
278 return &per_cpu(rt_cache_stat, cpu);
285 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
290 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
292 struct rt_cache_stat *st = v;
294 if (v == SEQ_START_TOKEN) {
295 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");
299 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
300 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
301 dst_entries_get_slow(&ipv4_dst_ops),
314 0, /* st->gc_total */
315 0, /* st->gc_ignored */
316 0, /* st->gc_goal_miss */
317 0, /* st->gc_dst_overflow */
318 0, /* st->in_hlist_search */
319 0 /* st->out_hlist_search */
324 static const struct seq_operations rt_cpu_seq_ops = {
325 .start = rt_cpu_seq_start,
326 .next = rt_cpu_seq_next,
327 .stop = rt_cpu_seq_stop,
328 .show = rt_cpu_seq_show,
332 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
334 return seq_open(file, &rt_cpu_seq_ops);
337 static const struct file_operations rt_cpu_seq_fops = {
338 .owner = THIS_MODULE,
339 .open = rt_cpu_seq_open,
342 .release = seq_release,
345 #ifdef CONFIG_IP_ROUTE_CLASSID
346 static int rt_acct_proc_show(struct seq_file *m, void *v)
348 struct ip_rt_acct *dst, *src;
351 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
355 for_each_possible_cpu(i) {
356 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
357 for (j = 0; j < 256; j++) {
358 dst[j].o_bytes += src[j].o_bytes;
359 dst[j].o_packets += src[j].o_packets;
360 dst[j].i_bytes += src[j].i_bytes;
361 dst[j].i_packets += src[j].i_packets;
365 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
370 static int rt_acct_proc_open(struct inode *inode, struct file *file)
372 return single_open(file, rt_acct_proc_show, NULL);
375 static const struct file_operations rt_acct_proc_fops = {
376 .owner = THIS_MODULE,
377 .open = rt_acct_proc_open,
380 .release = single_release,
384 static int __net_init ip_rt_do_proc_init(struct net *net)
386 struct proc_dir_entry *pde;
388 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
393 pde = proc_create("rt_cache", S_IRUGO,
394 net->proc_net_stat, &rt_cpu_seq_fops);
398 #ifdef CONFIG_IP_ROUTE_CLASSID
399 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
405 #ifdef CONFIG_IP_ROUTE_CLASSID
407 remove_proc_entry("rt_cache", net->proc_net_stat);
410 remove_proc_entry("rt_cache", net->proc_net);
415 static void __net_exit ip_rt_do_proc_exit(struct net *net)
417 remove_proc_entry("rt_cache", net->proc_net_stat);
418 remove_proc_entry("rt_cache", net->proc_net);
419 #ifdef CONFIG_IP_ROUTE_CLASSID
420 remove_proc_entry("rt_acct", net->proc_net);
424 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
425 .init = ip_rt_do_proc_init,
426 .exit = ip_rt_do_proc_exit,
429 static int __init ip_rt_proc_init(void)
431 return register_pernet_subsys(&ip_rt_proc_ops);
435 static inline int ip_rt_proc_init(void)
439 #endif /* CONFIG_PROC_FS */
441 static inline bool rt_is_expired(const struct rtable *rth)
443 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
446 void rt_cache_flush(struct net *net)
448 rt_genid_bump_ipv4(net);
451 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
455 struct net_device *dev = dst->dev;
456 const __be32 *pkey = daddr;
457 const struct rtable *rt;
460 rt = (const struct rtable *) dst;
462 pkey = (const __be32 *) &rt->rt_gateway;
464 pkey = &ip_hdr(skb)->daddr;
466 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
469 return neigh_create(&arp_tbl, pkey, dev);
472 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
474 struct net_device *dev = dst->dev;
475 const __be32 *pkey = daddr;
476 const struct rtable *rt;
478 rt = (const struct rtable *)dst;
480 pkey = (const __be32 *)&rt->rt_gateway;
483 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
486 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
489 /* Hash tables of size 2048..262144 depending on RAM size.
490 * Each bucket uses 8 bytes.
492 static u32 ip_idents_mask __read_mostly;
493 static atomic_t *ip_idents __read_mostly;
494 static u32 *ip_tstamps __read_mostly;
496 /* In order to protect privacy, we add a perturbation to identifiers
497 * if one generator is seldom used. This makes hard for an attacker
498 * to infer how many packets were sent between two points in time.
500 u32 ip_idents_reserve(u32 hash, int segs)
502 u32 bucket, old, now = (u32)jiffies;
507 bucket = hash & ip_idents_mask;
508 p_tstamp = ip_tstamps + bucket;
509 p_id = ip_idents + bucket;
510 old = ACCESS_ONCE(*p_tstamp);
512 if (old != now && cmpxchg(p_tstamp, old, now) == old)
513 delta = prandom_u32_max(now - old);
515 /* If UBSAN reports an error there, please make sure your compiler
516 * supports -fno-strict-overflow before reporting it that was a bug
517 * in UBSAN, and it has been fixed in GCC-8.
519 return atomic_add_return(segs + delta, p_id) - segs;
521 EXPORT_SYMBOL(ip_idents_reserve);
523 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
527 /* Note the following code is not safe, but this is okay. */
528 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
529 get_random_bytes(&net->ipv4.ip_id_key,
530 sizeof(net->ipv4.ip_id_key));
532 hash = siphash_3u32((__force u32)iph->daddr,
533 (__force u32)iph->saddr,
535 &net->ipv4.ip_id_key);
536 id = ip_idents_reserve(hash, segs);
539 EXPORT_SYMBOL(__ip_select_ident);
541 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
542 const struct sock *sk,
543 const struct iphdr *iph,
545 u8 prot, u32 mark, int flow_flags)
548 const struct inet_sock *inet = inet_sk(sk);
550 oif = sk->sk_bound_dev_if;
552 tos = RT_CONN_FLAGS(sk);
553 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
555 flowi4_init_output(fl4, oif, mark, tos,
556 RT_SCOPE_UNIVERSE, prot,
558 iph->daddr, iph->saddr, 0, 0,
559 sock_net_uid(net, sk));
562 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
563 const struct sock *sk)
565 const struct net *net = dev_net(skb->dev);
566 const struct iphdr *iph = ip_hdr(skb);
567 int oif = skb->dev->ifindex;
568 u8 tos = RT_TOS(iph->tos);
569 u8 prot = iph->protocol;
570 u32 mark = skb->mark;
572 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
575 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
577 const struct inet_sock *inet = inet_sk(sk);
578 const struct ip_options_rcu *inet_opt;
579 __be32 daddr = inet->inet_daddr;
582 inet_opt = rcu_dereference(inet->inet_opt);
583 if (inet_opt && inet_opt->opt.srr)
584 daddr = inet_opt->opt.faddr;
585 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
586 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
587 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
588 inet_sk_flowi_flags(sk),
589 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
593 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
594 const struct sk_buff *skb)
597 build_skb_flow_key(fl4, skb, sk);
599 build_sk_flow_key(fl4, sk);
602 static DEFINE_SPINLOCK(fnhe_lock);
604 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
608 rt = rcu_dereference(fnhe->fnhe_rth_input);
610 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
611 dst_dev_put(&rt->dst);
612 dst_release(&rt->dst);
614 rt = rcu_dereference(fnhe->fnhe_rth_output);
616 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
617 dst_dev_put(&rt->dst);
618 dst_release(&rt->dst);
622 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
624 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
625 struct fib_nh_exception *fnhe, *oldest = NULL;
627 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
628 fnhe = rcu_dereference_protected(*fnhe_p,
629 lockdep_is_held(&fnhe_lock));
633 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
638 fnhe_flush_routes(oldest);
639 *oldest_p = oldest->fnhe_next;
640 kfree_rcu(oldest, rcu);
643 static u32 fnhe_hashfun(__be32 daddr)
645 static siphash_key_t fnhe_hash_key __read_mostly;
648 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
649 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
650 return hash_64(hval, FNHE_HASH_SHIFT);
653 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
655 rt->rt_pmtu = fnhe->fnhe_pmtu;
656 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
657 rt->dst.expires = fnhe->fnhe_expires;
660 rt->rt_flags |= RTCF_REDIRECTED;
661 rt->rt_gateway = fnhe->fnhe_gw;
662 rt->rt_uses_gateway = 1;
666 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
667 u32 pmtu, bool lock, unsigned long expires)
669 struct fnhe_hash_bucket *hash;
670 struct fib_nh_exception *fnhe;
676 genid = fnhe_genid(dev_net(nh->nh_dev));
677 hval = fnhe_hashfun(daddr);
679 spin_lock_bh(&fnhe_lock);
681 hash = rcu_dereference(nh->nh_exceptions);
683 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
686 rcu_assign_pointer(nh->nh_exceptions, hash);
692 for (fnhe = rcu_dereference(hash->chain); fnhe;
693 fnhe = rcu_dereference(fnhe->fnhe_next)) {
694 if (fnhe->fnhe_daddr == daddr)
700 if (fnhe->fnhe_genid != genid)
701 fnhe->fnhe_genid = genid;
705 fnhe->fnhe_pmtu = pmtu;
706 fnhe->fnhe_mtu_locked = lock;
708 fnhe->fnhe_expires = max(1UL, expires);
709 /* Update all cached dsts too */
710 rt = rcu_dereference(fnhe->fnhe_rth_input);
712 fill_route_from_fnhe(rt, fnhe);
713 rt = rcu_dereference(fnhe->fnhe_rth_output);
715 fill_route_from_fnhe(rt, fnhe);
717 /* Randomize max depth to avoid some side channels attacks. */
718 int max_depth = FNHE_RECLAIM_DEPTH +
719 prandom_u32_max(FNHE_RECLAIM_DEPTH);
721 while (depth > max_depth) {
722 fnhe_remove_oldest(hash);
726 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
730 fnhe->fnhe_next = hash->chain;
732 fnhe->fnhe_genid = genid;
733 fnhe->fnhe_daddr = daddr;
735 fnhe->fnhe_pmtu = pmtu;
736 fnhe->fnhe_mtu_locked = lock;
737 fnhe->fnhe_expires = max(1UL, expires);
739 rcu_assign_pointer(hash->chain, fnhe);
741 /* Exception created; mark the cached routes for the nexthop
742 * stale, so anyone caching it rechecks if this exception
745 rt = rcu_dereference(nh->nh_rth_input);
747 rt->dst.obsolete = DST_OBSOLETE_KILL;
749 for_each_possible_cpu(i) {
750 struct rtable __rcu **prt;
751 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
752 rt = rcu_dereference(*prt);
754 rt->dst.obsolete = DST_OBSOLETE_KILL;
758 fnhe->fnhe_stamp = jiffies;
761 spin_unlock_bh(&fnhe_lock);
764 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
767 __be32 new_gw = icmp_hdr(skb)->un.gateway;
768 __be32 old_gw = ip_hdr(skb)->saddr;
769 struct net_device *dev = skb->dev;
770 struct in_device *in_dev;
771 struct fib_result res;
775 switch (icmp_hdr(skb)->code & 7) {
777 case ICMP_REDIR_NETTOS:
778 case ICMP_REDIR_HOST:
779 case ICMP_REDIR_HOSTTOS:
786 if (rt->rt_gateway != old_gw)
789 in_dev = __in_dev_get_rcu(dev);
794 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
795 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
796 ipv4_is_zeronet(new_gw))
797 goto reject_redirect;
799 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
800 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
801 goto reject_redirect;
802 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
803 goto reject_redirect;
805 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
806 goto reject_redirect;
809 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
811 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
813 if (!(n->nud_state & NUD_VALID)) {
814 neigh_event_send(n, NULL);
816 if (fib_lookup(net, fl4, &res, 0) == 0) {
817 struct fib_nh *nh = &FIB_RES_NH(res);
819 fib_select_path(net, &res, fl4, skb);
820 nh = &FIB_RES_NH(res);
821 update_or_create_fnhe(nh, fl4->daddr, new_gw,
823 jiffies + ip_rt_gc_timeout);
826 rt->dst.obsolete = DST_OBSOLETE_KILL;
827 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
834 #ifdef CONFIG_IP_ROUTE_VERBOSE
835 if (IN_DEV_LOG_MARTIANS(in_dev)) {
836 const struct iphdr *iph = (const struct iphdr *) skb->data;
837 __be32 daddr = iph->daddr;
838 __be32 saddr = iph->saddr;
840 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
841 " Advised path = %pI4 -> %pI4\n",
842 &old_gw, dev->name, &new_gw,
849 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
853 const struct iphdr *iph = (const struct iphdr *) skb->data;
854 struct net *net = dev_net(skb->dev);
855 int oif = skb->dev->ifindex;
856 u8 tos = RT_TOS(iph->tos);
857 u8 prot = iph->protocol;
858 u32 mark = skb->mark;
860 rt = (struct rtable *) dst;
862 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
863 __ip_do_redirect(rt, skb, &fl4, true);
866 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
868 struct rtable *rt = (struct rtable *)dst;
869 struct dst_entry *ret = dst;
872 if (dst->obsolete > 0) {
875 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
886 * 1. The first ip_rt_redirect_number redirects are sent
887 * with exponential backoff, then we stop sending them at all,
888 * assuming that the host ignores our redirects.
889 * 2. If we did not see packets requiring redirects
890 * during ip_rt_redirect_silence, we assume that the host
891 * forgot redirected route and start to send redirects again.
893 * This algorithm is much cheaper and more intelligent than dumb load limiting
896 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
897 * and "frag. need" (breaks PMTU discovery) in icmp.c.
900 void ip_rt_send_redirect(struct sk_buff *skb)
902 struct rtable *rt = skb_rtable(skb);
903 struct in_device *in_dev;
904 struct inet_peer *peer;
910 in_dev = __in_dev_get_rcu(rt->dst.dev);
911 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
915 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
916 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
919 net = dev_net(rt->dst.dev);
920 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
922 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
923 rt_nexthop(rt, ip_hdr(skb)->daddr));
927 /* No redirected packets during ip_rt_redirect_silence;
928 * reset the algorithm.
930 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
931 peer->rate_tokens = 0;
932 peer->n_redirects = 0;
935 /* Too many ignored redirects; do not send anything
936 * set dst.rate_last to the last seen redirected packet.
938 if (peer->n_redirects >= ip_rt_redirect_number) {
939 peer->rate_last = jiffies;
943 /* Check for load limit; set rate_last to the latest sent
946 if (peer->n_redirects == 0 ||
949 (ip_rt_redirect_load << peer->n_redirects)))) {
950 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
952 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
953 peer->rate_last = jiffies;
955 #ifdef CONFIG_IP_ROUTE_VERBOSE
957 peer->n_redirects == ip_rt_redirect_number)
958 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
959 &ip_hdr(skb)->saddr, inet_iif(skb),
960 &ip_hdr(skb)->daddr, &gw);
967 static int ip_error(struct sk_buff *skb)
969 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
970 struct rtable *rt = skb_rtable(skb);
971 struct inet_peer *peer;
977 /* IP on this device is disabled. */
981 net = dev_net(rt->dst.dev);
982 if (!IN_DEV_FORWARD(in_dev)) {
983 switch (rt->dst.error) {
985 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
989 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
995 switch (rt->dst.error) {
1000 code = ICMP_HOST_UNREACH;
1003 code = ICMP_NET_UNREACH;
1004 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
1007 code = ICMP_PKT_FILTERED;
1011 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
1012 l3mdev_master_ifindex(skb->dev), 1);
1017 peer->rate_tokens += now - peer->rate_last;
1018 if (peer->rate_tokens > ip_rt_error_burst)
1019 peer->rate_tokens = ip_rt_error_burst;
1020 peer->rate_last = now;
1021 if (peer->rate_tokens >= ip_rt_error_cost)
1022 peer->rate_tokens -= ip_rt_error_cost;
1028 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1030 out: kfree_skb(skb);
1034 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1036 struct dst_entry *dst = &rt->dst;
1037 struct net *net = dev_net(dst->dev);
1038 u32 old_mtu = ipv4_mtu(dst);
1039 struct fib_result res;
1042 if (ip_mtu_locked(dst))
1048 if (mtu < ip_rt_min_pmtu) {
1050 mtu = min(old_mtu, ip_rt_min_pmtu);
1053 if (rt->rt_pmtu == mtu && !lock &&
1054 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1058 if (fib_lookup(net, fl4, &res, 0) == 0) {
1061 fib_select_path(net, &res, fl4, NULL);
1062 nh = &FIB_RES_NH(res);
1063 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1064 jiffies + ip_rt_mtu_expires);
1069 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1070 struct sk_buff *skb, u32 mtu,
1073 struct rtable *rt = (struct rtable *) dst;
1076 ip_rt_build_flow_key(&fl4, sk, skb);
1077 __ip_rt_update_pmtu(rt, &fl4, mtu);
1080 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1081 int oif, u32 mark, u8 protocol, int flow_flags)
1083 const struct iphdr *iph = (const struct iphdr *) skb->data;
1088 mark = IP4_REPLY_MARK(net, skb->mark);
1090 __build_flow_key(net, &fl4, NULL, iph, oif,
1091 RT_TOS(iph->tos), protocol, mark, flow_flags);
1092 rt = __ip_route_output_key(net, &fl4);
1094 __ip_rt_update_pmtu(rt, &fl4, mtu);
1098 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1100 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1102 const struct iphdr *iph = (const struct iphdr *) skb->data;
1106 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1108 if (!fl4.flowi4_mark)
1109 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1111 rt = __ip_route_output_key(sock_net(sk), &fl4);
1113 __ip_rt_update_pmtu(rt, &fl4, mtu);
1118 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1120 const struct iphdr *iph = (const struct iphdr *) skb->data;
1123 struct dst_entry *odst = NULL;
1125 struct net *net = sock_net(sk);
1129 if (!ip_sk_accept_pmtu(sk))
1132 odst = sk_dst_get(sk);
1134 if (sock_owned_by_user(sk) || !odst) {
1135 __ipv4_sk_update_pmtu(skb, sk, mtu);
1139 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1141 rt = (struct rtable *)odst;
1142 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1143 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1150 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1152 if (!dst_check(&rt->dst, 0)) {
1154 dst_release(&rt->dst);
1156 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1164 sk_dst_set(sk, &rt->dst);
1170 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1172 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1173 int oif, u32 mark, u8 protocol, int flow_flags)
1175 const struct iphdr *iph = (const struct iphdr *) skb->data;
1179 __build_flow_key(net, &fl4, NULL, iph, oif,
1180 RT_TOS(iph->tos), protocol, mark, flow_flags);
1181 rt = __ip_route_output_key(net, &fl4);
1183 __ip_do_redirect(rt, skb, &fl4, false);
1187 EXPORT_SYMBOL_GPL(ipv4_redirect);
1189 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1191 const struct iphdr *iph = (const struct iphdr *) skb->data;
1194 struct net *net = sock_net(sk);
1196 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1197 rt = __ip_route_output_key(net, &fl4);
1199 __ip_do_redirect(rt, skb, &fl4, false);
1203 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1205 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1207 struct rtable *rt = (struct rtable *) dst;
1209 /* All IPV4 dsts are created with ->obsolete set to the value
1210 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1211 * into this function always.
1213 * When a PMTU/redirect information update invalidates a route,
1214 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1215 * DST_OBSOLETE_DEAD by dst_free().
1217 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1222 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1224 struct ip_options opt;
1227 /* Recompile ip options since IPCB may not be valid anymore.
1228 * Also check we have a reasonable ipv4 header.
1230 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1231 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1234 memset(&opt, 0, sizeof(opt));
1235 if (ip_hdr(skb)->ihl > 5) {
1236 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1238 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1241 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1247 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1250 static void ipv4_link_failure(struct sk_buff *skb)
1254 ipv4_send_dest_unreach(skb);
1256 rt = skb_rtable(skb);
1258 dst_set_expires(&rt->dst, 0);
1261 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1263 pr_debug("%s: %pI4 -> %pI4, %s\n",
1264 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1265 skb->dev ? skb->dev->name : "?");
1272 We do not cache source address of outgoing interface,
1273 because it is used only by IP RR, TS and SRR options,
1274 so that it out of fast path.
1276 BTW remember: "addr" is allowed to be not aligned
1280 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1284 if (rt_is_output_route(rt))
1285 src = ip_hdr(skb)->saddr;
1287 struct fib_result res;
1293 memset(&fl4, 0, sizeof(fl4));
1294 fl4.daddr = iph->daddr;
1295 fl4.saddr = iph->saddr;
1296 fl4.flowi4_tos = RT_TOS(iph->tos);
1297 fl4.flowi4_oif = rt->dst.dev->ifindex;
1298 fl4.flowi4_iif = skb->dev->ifindex;
1299 fl4.flowi4_mark = skb->mark;
1302 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1303 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1305 src = inet_select_addr(rt->dst.dev,
1306 rt_nexthop(rt, iph->daddr),
1310 memcpy(addr, &src, 4);
1313 #ifdef CONFIG_IP_ROUTE_CLASSID
1314 static void set_class_tag(struct rtable *rt, u32 tag)
1316 if (!(rt->dst.tclassid & 0xFFFF))
1317 rt->dst.tclassid |= tag & 0xFFFF;
1318 if (!(rt->dst.tclassid & 0xFFFF0000))
1319 rt->dst.tclassid |= tag & 0xFFFF0000;
1323 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1325 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1326 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1329 return min(advmss, IPV4_MAX_PMTU - header_size);
1332 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1334 const struct rtable *rt = (const struct rtable *) dst;
1335 unsigned int mtu = rt->rt_pmtu;
1337 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1338 mtu = dst_metric_raw(dst, RTAX_MTU);
1343 mtu = READ_ONCE(dst->dev->mtu);
1345 if (unlikely(ip_mtu_locked(dst))) {
1346 if (rt->rt_uses_gateway && mtu > 576)
1350 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1352 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1355 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1357 struct fnhe_hash_bucket *hash;
1358 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1359 u32 hval = fnhe_hashfun(daddr);
1361 spin_lock_bh(&fnhe_lock);
1363 hash = rcu_dereference_protected(nh->nh_exceptions,
1364 lockdep_is_held(&fnhe_lock));
1367 fnhe_p = &hash->chain;
1368 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1370 if (fnhe->fnhe_daddr == daddr) {
1371 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1372 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1373 /* set fnhe_daddr to 0 to ensure it won't bind with
1374 * new dsts in rt_bind_exception().
1376 fnhe->fnhe_daddr = 0;
1377 fnhe_flush_routes(fnhe);
1378 kfree_rcu(fnhe, rcu);
1381 fnhe_p = &fnhe->fnhe_next;
1382 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1383 lockdep_is_held(&fnhe_lock));
1386 spin_unlock_bh(&fnhe_lock);
1389 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1391 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1392 struct fib_nh_exception *fnhe;
1398 hval = fnhe_hashfun(daddr);
1400 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1401 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1402 if (fnhe->fnhe_daddr == daddr) {
1403 if (fnhe->fnhe_expires &&
1404 time_after(jiffies, fnhe->fnhe_expires)) {
1405 ip_del_fnhe(nh, daddr);
1414 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1415 __be32 daddr, const bool do_cache)
1419 spin_lock_bh(&fnhe_lock);
1421 if (daddr == fnhe->fnhe_daddr) {
1422 struct rtable __rcu **porig;
1423 struct rtable *orig;
1424 int genid = fnhe_genid(dev_net(rt->dst.dev));
1426 if (rt_is_input_route(rt))
1427 porig = &fnhe->fnhe_rth_input;
1429 porig = &fnhe->fnhe_rth_output;
1430 orig = rcu_dereference(*porig);
1432 if (fnhe->fnhe_genid != genid) {
1433 fnhe->fnhe_genid = genid;
1435 fnhe->fnhe_pmtu = 0;
1436 fnhe->fnhe_expires = 0;
1437 fnhe_flush_routes(fnhe);
1440 fill_route_from_fnhe(rt, fnhe);
1441 if (!rt->rt_gateway)
1442 rt->rt_gateway = daddr;
1446 rcu_assign_pointer(*porig, rt);
1448 dst_dev_put(&orig->dst);
1449 dst_release(&orig->dst);
1454 fnhe->fnhe_stamp = jiffies;
1456 spin_unlock_bh(&fnhe_lock);
1461 struct uncached_list {
1463 struct list_head head;
1466 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1468 static void rt_add_uncached_list(struct rtable *rt)
1470 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1472 rt->rt_uncached_list = ul;
1474 spin_lock_bh(&ul->lock);
1475 list_add_tail(&rt->rt_uncached, &ul->head);
1476 spin_unlock_bh(&ul->lock);
1479 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1481 struct rtable *orig, *prev, **p;
1484 if (rt_is_input_route(rt)) {
1485 p = (struct rtable **)&nh->nh_rth_input;
1487 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1491 /* hold dst before doing cmpxchg() to avoid race condition
1495 prev = cmpxchg(p, orig, rt);
1498 rt_add_uncached_list(orig);
1499 dst_release(&orig->dst);
1502 dst_release(&rt->dst);
1509 static void ipv4_dst_destroy(struct dst_entry *dst)
1511 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1512 struct rtable *rt = (struct rtable *) dst;
1514 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1517 if (!list_empty(&rt->rt_uncached)) {
1518 struct uncached_list *ul = rt->rt_uncached_list;
1520 spin_lock_bh(&ul->lock);
1521 list_del(&rt->rt_uncached);
1522 spin_unlock_bh(&ul->lock);
1526 void rt_flush_dev(struct net_device *dev)
1528 struct net *net = dev_net(dev);
1532 for_each_possible_cpu(cpu) {
1533 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1535 spin_lock_bh(&ul->lock);
1536 list_for_each_entry(rt, &ul->head, rt_uncached) {
1537 if (rt->dst.dev != dev)
1539 rt->dst.dev = net->loopback_dev;
1540 dev_hold(rt->dst.dev);
1543 spin_unlock_bh(&ul->lock);
1547 static bool rt_cache_valid(const struct rtable *rt)
1550 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1554 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1555 const struct fib_result *res,
1556 struct fib_nh_exception *fnhe,
1557 struct fib_info *fi, u16 type, u32 itag,
1558 const bool do_cache)
1560 bool cached = false;
1563 struct fib_nh *nh = &FIB_RES_NH(*res);
1565 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1566 rt->rt_gateway = nh->nh_gw;
1567 rt->rt_uses_gateway = 1;
1569 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1570 if (fi->fib_metrics != &dst_default_metrics) {
1571 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1572 refcount_inc(&fi->fib_metrics->refcnt);
1574 #ifdef CONFIG_IP_ROUTE_CLASSID
1575 rt->dst.tclassid = nh->nh_tclassid;
1577 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1579 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1581 cached = rt_cache_route(nh, rt);
1582 if (unlikely(!cached)) {
1583 /* Routes we intend to cache in nexthop exception or
1584 * FIB nexthop have the DST_NOCACHE bit clear.
1585 * However, if we are unsuccessful at storing this
1586 * route into the cache we really need to set it.
1588 if (!rt->rt_gateway)
1589 rt->rt_gateway = daddr;
1590 rt_add_uncached_list(rt);
1593 rt_add_uncached_list(rt);
1595 #ifdef CONFIG_IP_ROUTE_CLASSID
1596 #ifdef CONFIG_IP_MULTIPLE_TABLES
1597 set_class_tag(rt, res->tclassid);
1599 set_class_tag(rt, itag);
1603 struct rtable *rt_dst_alloc(struct net_device *dev,
1604 unsigned int flags, u16 type,
1605 bool nopolicy, bool noxfrm, bool will_cache)
1609 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1610 (will_cache ? 0 : DST_HOST) |
1611 (nopolicy ? DST_NOPOLICY : 0) |
1612 (noxfrm ? DST_NOXFRM : 0));
1615 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1616 rt->rt_flags = flags;
1618 rt->rt_is_input = 0;
1621 rt->rt_mtu_locked = 0;
1623 rt->rt_uses_gateway = 0;
1624 rt->rt_table_id = 0;
1625 INIT_LIST_HEAD(&rt->rt_uncached);
1627 rt->dst.output = ip_output;
1628 if (flags & RTCF_LOCAL)
1629 rt->dst.input = ip_local_deliver;
1634 EXPORT_SYMBOL(rt_dst_alloc);
1636 /* called in rcu_read_lock() section */
1637 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1638 u8 tos, struct net_device *dev,
1639 struct in_device *in_dev, u32 *itag)
1643 /* Primary sanity checks. */
1647 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1648 skb->protocol != htons(ETH_P_IP))
1651 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1654 if (ipv4_is_zeronet(saddr)) {
1655 if (!ipv4_is_local_multicast(daddr))
1658 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1666 /* called in rcu_read_lock() section */
1667 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1668 u8 tos, struct net_device *dev, int our)
1670 struct in_device *in_dev = __in_dev_get_rcu(dev);
1671 unsigned int flags = RTCF_MULTICAST;
1676 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1681 flags |= RTCF_LOCAL;
1683 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1684 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1688 #ifdef CONFIG_IP_ROUTE_CLASSID
1689 rth->dst.tclassid = itag;
1691 rth->dst.output = ip_rt_bug;
1692 rth->rt_is_input= 1;
1694 #ifdef CONFIG_IP_MROUTE
1695 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1696 rth->dst.input = ip_mr_input;
1698 RT_CACHE_STAT_INC(in_slow_mc);
1700 skb_dst_set(skb, &rth->dst);
1705 static void ip_handle_martian_source(struct net_device *dev,
1706 struct in_device *in_dev,
1707 struct sk_buff *skb,
1711 RT_CACHE_STAT_INC(in_martian_src);
1712 #ifdef CONFIG_IP_ROUTE_VERBOSE
1713 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1715 * RFC1812 recommendation, if source is martian,
1716 * the only hint is MAC header.
1718 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1719 &daddr, &saddr, dev->name);
1720 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1721 print_hex_dump(KERN_WARNING, "ll header: ",
1722 DUMP_PREFIX_OFFSET, 16, 1,
1723 skb_mac_header(skb),
1724 dev->hard_header_len, true);
1730 static void set_lwt_redirect(struct rtable *rth)
1732 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1733 rth->dst.lwtstate->orig_output = rth->dst.output;
1734 rth->dst.output = lwtunnel_output;
1737 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1738 rth->dst.lwtstate->orig_input = rth->dst.input;
1739 rth->dst.input = lwtunnel_input;
1743 /* called in rcu_read_lock() section */
1744 static int __mkroute_input(struct sk_buff *skb,
1745 const struct fib_result *res,
1746 struct in_device *in_dev,
1747 __be32 daddr, __be32 saddr, u32 tos)
1749 struct fib_nh_exception *fnhe;
1752 struct in_device *out_dev;
1756 /* get a working reference to the output device */
1757 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1759 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1763 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1764 in_dev->dev, in_dev, &itag);
1766 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1772 do_cache = res->fi && !itag;
1773 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1774 skb->protocol == htons(ETH_P_IP) &&
1775 (IN_DEV_SHARED_MEDIA(out_dev) ||
1776 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1777 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1779 if (skb->protocol != htons(ETH_P_IP)) {
1780 /* Not IP (i.e. ARP). Do not create route, if it is
1781 * invalid for proxy arp. DNAT routes are always valid.
1783 * Proxy arp feature have been extended to allow, ARP
1784 * replies back to the same interface, to support
1785 * Private VLAN switch technologies. See arp.c.
1787 if (out_dev == in_dev &&
1788 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1794 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1797 rth = rcu_dereference(fnhe->fnhe_rth_input);
1799 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1800 if (rt_cache_valid(rth)) {
1801 skb_dst_set_noref(skb, &rth->dst);
1806 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1807 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1808 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1814 rth->rt_is_input = 1;
1816 rth->rt_table_id = res->table->tb_id;
1817 RT_CACHE_STAT_INC(in_slow_tot);
1819 rth->dst.input = ip_forward;
1821 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1823 set_lwt_redirect(rth);
1824 skb_dst_set(skb, &rth->dst);
1831 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1832 /* To make ICMP packets follow the right flow, the multipath hash is
1833 * calculated from the inner IP addresses.
1835 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1836 struct flow_keys *hash_keys)
1838 const struct iphdr *outer_iph = ip_hdr(skb);
1839 const struct iphdr *inner_iph;
1840 const struct icmphdr *icmph;
1841 struct iphdr _inner_iph;
1842 struct icmphdr _icmph;
1844 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1845 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1846 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1849 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1852 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1857 if (icmph->type != ICMP_DEST_UNREACH &&
1858 icmph->type != ICMP_REDIRECT &&
1859 icmph->type != ICMP_TIME_EXCEEDED &&
1860 icmph->type != ICMP_PARAMETERPROB)
1863 inner_iph = skb_header_pointer(skb,
1864 outer_iph->ihl * 4 + sizeof(_icmph),
1865 sizeof(_inner_iph), &_inner_iph);
1868 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1869 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1872 /* if skb is set it will be used and fl4 can be NULL */
1873 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1874 const struct sk_buff *skb)
1876 struct net *net = fi->fib_net;
1877 struct flow_keys hash_keys;
1880 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1882 memset(&hash_keys, 0, sizeof(hash_keys));
1883 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1885 ip_multipath_l3_keys(skb, &hash_keys);
1887 hash_keys.addrs.v4addrs.src = fl4->saddr;
1888 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1892 /* skb is currently provided only when forwarding */
1894 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1895 struct flow_keys keys;
1897 /* short-circuit if we already have L4 hash present */
1899 return skb_get_hash_raw(skb) >> 1;
1900 memset(&hash_keys, 0, sizeof(hash_keys));
1901 skb_flow_dissect_flow_keys(skb, &keys, flag);
1903 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1904 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1905 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1906 hash_keys.ports.src = keys.ports.src;
1907 hash_keys.ports.dst = keys.ports.dst;
1908 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1910 memset(&hash_keys, 0, sizeof(hash_keys));
1911 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1912 hash_keys.addrs.v4addrs.src = fl4->saddr;
1913 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1914 hash_keys.ports.src = fl4->fl4_sport;
1915 hash_keys.ports.dst = fl4->fl4_dport;
1916 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1920 mhash = flow_hash_from_keys(&hash_keys);
1924 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1925 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1927 static int ip_mkroute_input(struct sk_buff *skb,
1928 struct fib_result *res,
1929 struct in_device *in_dev,
1930 __be32 daddr, __be32 saddr, u32 tos)
1932 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1933 if (res->fi && res->fi->fib_nhs > 1) {
1934 int h = fib_multipath_hash(res->fi, NULL, skb);
1936 fib_select_multipath(res, h);
1940 /* create a routing cache entry */
1941 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1945 * NOTE. We drop all the packets that has local source
1946 * addresses, because every properly looped back packet
1947 * must have correct destination already attached by output routine.
1949 * Such approach solves two big problems:
1950 * 1. Not simplex devices are handled properly.
1951 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1952 * called with rcu_read_lock()
1955 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1956 u8 tos, struct net_device *dev,
1957 struct fib_result *res)
1959 struct in_device *in_dev = __in_dev_get_rcu(dev);
1960 struct ip_tunnel_info *tun_info;
1962 unsigned int flags = 0;
1966 struct net *net = dev_net(dev);
1969 /* IP on this device is disabled. */
1974 /* Check for the most weird martians, which can be not detected
1978 tun_info = skb_tunnel_info(skb);
1979 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1980 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1982 fl4.flowi4_tun_key.tun_id = 0;
1985 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1986 goto martian_source;
1990 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1993 /* Accept zero addresses only to limited broadcast;
1994 * I even do not know to fix it or not. Waiting for complains :-)
1996 if (ipv4_is_zeronet(saddr))
1997 goto martian_source;
1999 if (ipv4_is_zeronet(daddr))
2000 goto martian_destination;
2002 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2003 * and call it once if daddr or/and saddr are loopback addresses
2005 if (ipv4_is_loopback(daddr)) {
2006 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2007 goto martian_destination;
2008 } else if (ipv4_is_loopback(saddr)) {
2009 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2010 goto martian_source;
2014 * Now we are ready to route packet.
2017 fl4.flowi4_iif = dev->ifindex;
2018 fl4.flowi4_mark = skb->mark;
2019 fl4.flowi4_tos = tos;
2020 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2021 fl4.flowi4_flags = 0;
2024 fl4.flowi4_uid = sock_net_uid(net, NULL);
2025 err = fib_lookup(net, &fl4, res, 0);
2027 if (!IN_DEV_FORWARD(in_dev))
2028 err = -EHOSTUNREACH;
2032 if (res->type == RTN_BROADCAST)
2035 if (res->type == RTN_LOCAL) {
2036 err = fib_validate_source(skb, saddr, daddr, tos,
2037 0, dev, in_dev, &itag);
2039 goto martian_source;
2043 if (!IN_DEV_FORWARD(in_dev)) {
2044 err = -EHOSTUNREACH;
2047 if (res->type != RTN_UNICAST)
2048 goto martian_destination;
2050 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2054 if (skb->protocol != htons(ETH_P_IP))
2057 if (!ipv4_is_zeronet(saddr)) {
2058 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2061 goto martian_source;
2063 flags |= RTCF_BROADCAST;
2064 res->type = RTN_BROADCAST;
2065 RT_CACHE_STAT_INC(in_brd);
2071 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2072 if (rt_cache_valid(rth)) {
2073 skb_dst_set_noref(skb, &rth->dst);
2081 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2082 flags | RTCF_LOCAL, res->type,
2083 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2087 rth->dst.output= ip_rt_bug;
2088 #ifdef CONFIG_IP_ROUTE_CLASSID
2089 rth->dst.tclassid = itag;
2091 rth->rt_is_input = 1;
2093 rth->rt_table_id = res->table->tb_id;
2095 RT_CACHE_STAT_INC(in_slow_tot);
2096 if (res->type == RTN_UNREACHABLE) {
2097 rth->dst.input= ip_error;
2098 rth->dst.error= -err;
2099 rth->rt_flags &= ~RTCF_LOCAL;
2103 struct fib_nh *nh = &FIB_RES_NH(*res);
2105 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2106 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2107 WARN_ON(rth->dst.input == lwtunnel_input);
2108 rth->dst.lwtstate->orig_input = rth->dst.input;
2109 rth->dst.input = lwtunnel_input;
2112 if (unlikely(!rt_cache_route(nh, rth)))
2113 rt_add_uncached_list(rth);
2115 skb_dst_set(skb, &rth->dst);
2120 RT_CACHE_STAT_INC(in_no_route);
2121 res->type = RTN_UNREACHABLE;
2127 * Do not cache martian addresses: they should be logged (RFC1812)
2129 martian_destination:
2130 RT_CACHE_STAT_INC(in_martian_dst);
2131 #ifdef CONFIG_IP_ROUTE_VERBOSE
2132 if (IN_DEV_LOG_MARTIANS(in_dev))
2133 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2134 &daddr, &saddr, dev->name);
2146 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2150 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2151 u8 tos, struct net_device *dev)
2153 struct fib_result res;
2156 tos &= IPTOS_RT_MASK;
2158 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2163 EXPORT_SYMBOL(ip_route_input_noref);
2165 /* called with rcu_read_lock held */
2166 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2167 u8 tos, struct net_device *dev, struct fib_result *res)
2169 /* Multicast recognition logic is moved from route cache to here.
2170 The problem was that too many Ethernet cards have broken/missing
2171 hardware multicast filters :-( As result the host on multicasting
2172 network acquires a lot of useless route cache entries, sort of
2173 SDR messages from all the world. Now we try to get rid of them.
2174 Really, provided software IP multicast filter is organized
2175 reasonably (at least, hashed), it does not result in a slowdown
2176 comparing with route cache reject entries.
2177 Note, that multicast routers are not affected, because
2178 route cache entry is created eventually.
2180 if (ipv4_is_multicast(daddr)) {
2181 struct in_device *in_dev = __in_dev_get_rcu(dev);
2187 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2188 ip_hdr(skb)->protocol);
2190 /* check l3 master if no match yet */
2191 if (!our && netif_is_l3_slave(dev)) {
2192 struct in_device *l3_in_dev;
2194 l3_in_dev = __in_dev_get_rcu(skb->dev);
2196 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2197 ip_hdr(skb)->protocol);
2201 #ifdef CONFIG_IP_MROUTE
2203 (!ipv4_is_local_multicast(daddr) &&
2204 IN_DEV_MFORWARD(in_dev))
2207 err = ip_route_input_mc(skb, daddr, saddr,
2213 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2216 /* called with rcu_read_lock() */
2217 static struct rtable *__mkroute_output(const struct fib_result *res,
2218 const struct flowi4 *fl4, int orig_oif,
2219 struct net_device *dev_out,
2222 struct fib_info *fi = res->fi;
2223 struct fib_nh_exception *fnhe;
2224 struct in_device *in_dev;
2225 u16 type = res->type;
2229 in_dev = __in_dev_get_rcu(dev_out);
2231 return ERR_PTR(-EINVAL);
2233 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2234 if (ipv4_is_loopback(fl4->saddr) &&
2235 !(dev_out->flags & IFF_LOOPBACK) &&
2236 !netif_is_l3_master(dev_out))
2237 return ERR_PTR(-EINVAL);
2239 if (ipv4_is_lbcast(fl4->daddr))
2240 type = RTN_BROADCAST;
2241 else if (ipv4_is_multicast(fl4->daddr))
2242 type = RTN_MULTICAST;
2243 else if (ipv4_is_zeronet(fl4->daddr))
2244 return ERR_PTR(-EINVAL);
2246 if (dev_out->flags & IFF_LOOPBACK)
2247 flags |= RTCF_LOCAL;
2250 if (type == RTN_BROADCAST) {
2251 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2253 } else if (type == RTN_MULTICAST) {
2254 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2255 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2257 flags &= ~RTCF_LOCAL;
2260 /* If multicast route do not exist use
2261 * default one, but do not gateway in this case.
2264 if (fi && res->prefixlen < 4)
2266 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2267 (orig_oif != dev_out->ifindex)) {
2268 /* For local routes that require a particular output interface
2269 * we do not want to cache the result. Caching the result
2270 * causes incorrect behaviour when there are multiple source
2271 * addresses on the interface, the end result being that if the
2272 * intended recipient is waiting on that interface for the
2273 * packet he won't receive it because it will be delivered on
2274 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2275 * be set to the loopback interface as well.
2281 do_cache &= fi != NULL;
2283 struct rtable __rcu **prth;
2284 struct fib_nh *nh = &FIB_RES_NH(*res);
2286 fnhe = find_exception(nh, fl4->daddr);
2290 prth = &fnhe->fnhe_rth_output;
2292 if (unlikely(fl4->flowi4_flags &
2293 FLOWI_FLAG_KNOWN_NH &&
2295 nh->nh_scope == RT_SCOPE_LINK))) {
2299 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2301 rth = rcu_dereference(*prth);
2302 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2307 rth = rt_dst_alloc(dev_out, flags, type,
2308 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2309 IN_DEV_CONF_GET(in_dev, NOXFRM),
2312 return ERR_PTR(-ENOBUFS);
2314 rth->rt_iif = orig_oif;
2316 rth->rt_table_id = res->table->tb_id;
2318 RT_CACHE_STAT_INC(out_slow_tot);
2320 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2321 if (flags & RTCF_LOCAL &&
2322 !(dev_out->flags & IFF_LOOPBACK)) {
2323 rth->dst.output = ip_mc_output;
2324 RT_CACHE_STAT_INC(out_slow_mc);
2326 #ifdef CONFIG_IP_MROUTE
2327 if (type == RTN_MULTICAST) {
2328 if (IN_DEV_MFORWARD(in_dev) &&
2329 !ipv4_is_local_multicast(fl4->daddr)) {
2330 rth->dst.input = ip_mr_input;
2331 rth->dst.output = ip_mc_output;
2337 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2338 set_lwt_redirect(rth);
2344 * Major route resolver routine.
2347 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2348 const struct sk_buff *skb)
2350 __u8 tos = RT_FL_TOS(fl4);
2351 struct fib_result res = {
2359 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2360 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2361 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2362 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2365 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2370 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2372 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2373 struct fib_result *res,
2374 const struct sk_buff *skb)
2376 struct net_device *dev_out = NULL;
2377 int orig_oif = fl4->flowi4_oif;
2378 unsigned int flags = 0;
2383 if (ipv4_is_multicast(fl4->saddr) ||
2384 ipv4_is_lbcast(fl4->saddr) ||
2385 ipv4_is_zeronet(fl4->saddr)) {
2386 rth = ERR_PTR(-EINVAL);
2390 rth = ERR_PTR(-ENETUNREACH);
2392 /* I removed check for oif == dev_out->oif here.
2393 It was wrong for two reasons:
2394 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2395 is assigned to multiple interfaces.
2396 2. Moreover, we are allowed to send packets with saddr
2397 of another iface. --ANK
2400 if (fl4->flowi4_oif == 0 &&
2401 (ipv4_is_multicast(fl4->daddr) ||
2402 ipv4_is_lbcast(fl4->daddr))) {
2403 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2404 dev_out = __ip_dev_find(net, fl4->saddr, false);
2408 /* Special hack: user can direct multicasts
2409 and limited broadcast via necessary interface
2410 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2411 This hack is not just for fun, it allows
2412 vic,vat and friends to work.
2413 They bind socket to loopback, set ttl to zero
2414 and expect that it will work.
2415 From the viewpoint of routing cache they are broken,
2416 because we are not allowed to build multicast path
2417 with loopback source addr (look, routing cache
2418 cannot know, that ttl is zero, so that packet
2419 will not leave this host and route is valid).
2420 Luckily, this hack is good workaround.
2423 fl4->flowi4_oif = dev_out->ifindex;
2427 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2428 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2429 if (!__ip_dev_find(net, fl4->saddr, false))
2435 if (fl4->flowi4_oif) {
2436 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2437 rth = ERR_PTR(-ENODEV);
2441 /* RACE: Check return value of inet_select_addr instead. */
2442 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2443 rth = ERR_PTR(-ENETUNREACH);
2446 if (ipv4_is_local_multicast(fl4->daddr) ||
2447 ipv4_is_lbcast(fl4->daddr) ||
2448 fl4->flowi4_proto == IPPROTO_IGMP) {
2450 fl4->saddr = inet_select_addr(dev_out, 0,
2455 if (ipv4_is_multicast(fl4->daddr))
2456 fl4->saddr = inet_select_addr(dev_out, 0,
2458 else if (!fl4->daddr)
2459 fl4->saddr = inet_select_addr(dev_out, 0,
2465 fl4->daddr = fl4->saddr;
2467 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2468 dev_out = net->loopback_dev;
2469 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2470 res->type = RTN_LOCAL;
2471 flags |= RTCF_LOCAL;
2475 err = fib_lookup(net, fl4, res, 0);
2479 if (fl4->flowi4_oif &&
2480 (ipv4_is_multicast(fl4->daddr) ||
2481 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2482 /* Apparently, routing tables are wrong. Assume,
2483 that the destination is on link.
2486 Because we are allowed to send to iface
2487 even if it has NO routes and NO assigned
2488 addresses. When oif is specified, routing
2489 tables are looked up with only one purpose:
2490 to catch if destination is gatewayed, rather than
2491 direct. Moreover, if MSG_DONTROUTE is set,
2492 we send packet, ignoring both routing tables
2493 and ifaddr state. --ANK
2496 We could make it even if oif is unknown,
2497 likely IPv6, but we do not.
2500 if (fl4->saddr == 0)
2501 fl4->saddr = inet_select_addr(dev_out, 0,
2503 res->type = RTN_UNICAST;
2510 if (res->type == RTN_LOCAL) {
2512 if (res->fi->fib_prefsrc)
2513 fl4->saddr = res->fi->fib_prefsrc;
2515 fl4->saddr = fl4->daddr;
2518 /* L3 master device is the loopback for that domain */
2519 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2522 /* make sure orig_oif points to fib result device even
2523 * though packet rx/tx happens over loopback or l3mdev
2525 orig_oif = FIB_RES_OIF(*res);
2527 fl4->flowi4_oif = dev_out->ifindex;
2528 flags |= RTCF_LOCAL;
2532 fib_select_path(net, res, fl4, skb);
2534 dev_out = FIB_RES_DEV(*res);
2537 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2543 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2548 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2550 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2552 return mtu ? : dst->dev->mtu;
2555 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2556 struct sk_buff *skb, u32 mtu,
2561 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2562 struct sk_buff *skb)
2566 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2572 static struct dst_ops ipv4_dst_blackhole_ops = {
2574 .check = ipv4_blackhole_dst_check,
2575 .mtu = ipv4_blackhole_mtu,
2576 .default_advmss = ipv4_default_advmss,
2577 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2578 .redirect = ipv4_rt_blackhole_redirect,
2579 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2580 .neigh_lookup = ipv4_neigh_lookup,
2583 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2585 struct rtable *ort = (struct rtable *) dst_orig;
2588 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2590 struct dst_entry *new = &rt->dst;
2593 new->input = dst_discard;
2594 new->output = dst_discard_out;
2596 new->dev = net->loopback_dev;
2600 rt->rt_is_input = ort->rt_is_input;
2601 rt->rt_iif = ort->rt_iif;
2602 rt->rt_pmtu = ort->rt_pmtu;
2603 rt->rt_mtu_locked = ort->rt_mtu_locked;
2605 rt->rt_genid = rt_genid_ipv4(net);
2606 rt->rt_flags = ort->rt_flags;
2607 rt->rt_type = ort->rt_type;
2608 rt->rt_gateway = ort->rt_gateway;
2609 rt->rt_uses_gateway = ort->rt_uses_gateway;
2611 INIT_LIST_HEAD(&rt->rt_uncached);
2614 dst_release(dst_orig);
2616 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2619 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2620 const struct sock *sk)
2622 struct rtable *rt = __ip_route_output_key(net, flp4);
2627 if (flp4->flowi4_proto) {
2628 flp4->flowi4_oif = rt->dst.dev->ifindex;
2629 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2630 flowi4_to_flowi(flp4),
2636 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2638 /* called with rcu_read_lock held */
2639 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2640 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2643 struct rtable *rt = skb_rtable(skb);
2645 struct nlmsghdr *nlh;
2646 unsigned long expires = 0;
2648 u32 metrics[RTAX_MAX];
2650 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2654 r = nlmsg_data(nlh);
2655 r->rtm_family = AF_INET;
2656 r->rtm_dst_len = 32;
2658 r->rtm_tos = fl4->flowi4_tos;
2659 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2660 if (nla_put_u32(skb, RTA_TABLE, table_id))
2661 goto nla_put_failure;
2662 r->rtm_type = rt->rt_type;
2663 r->rtm_scope = RT_SCOPE_UNIVERSE;
2664 r->rtm_protocol = RTPROT_UNSPEC;
2665 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2666 if (rt->rt_flags & RTCF_NOTIFY)
2667 r->rtm_flags |= RTM_F_NOTIFY;
2668 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2669 r->rtm_flags |= RTCF_DOREDIRECT;
2671 if (nla_put_in_addr(skb, RTA_DST, dst))
2672 goto nla_put_failure;
2674 r->rtm_src_len = 32;
2675 if (nla_put_in_addr(skb, RTA_SRC, src))
2676 goto nla_put_failure;
2679 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2680 goto nla_put_failure;
2681 #ifdef CONFIG_IP_ROUTE_CLASSID
2682 if (rt->dst.tclassid &&
2683 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2684 goto nla_put_failure;
2686 if (!rt_is_input_route(rt) &&
2687 fl4->saddr != src) {
2688 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2689 goto nla_put_failure;
2691 if (rt->rt_uses_gateway &&
2692 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2693 goto nla_put_failure;
2695 expires = rt->dst.expires;
2697 unsigned long now = jiffies;
2699 if (time_before(now, expires))
2705 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2706 if (rt->rt_pmtu && expires)
2707 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2708 if (rt->rt_mtu_locked && expires)
2709 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2710 if (rtnetlink_put_metrics(skb, metrics) < 0)
2711 goto nla_put_failure;
2713 if (fl4->flowi4_mark &&
2714 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2715 goto nla_put_failure;
2717 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2718 nla_put_u32(skb, RTA_UID,
2719 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2720 goto nla_put_failure;
2722 error = rt->dst.error;
2724 if (rt_is_input_route(rt)) {
2725 #ifdef CONFIG_IP_MROUTE
2726 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2727 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2728 int err = ipmr_get_route(net, skb,
2729 fl4->saddr, fl4->daddr,
2735 goto nla_put_failure;
2739 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2740 goto nla_put_failure;
2743 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2744 goto nla_put_failure;
2746 nlmsg_end(skb, nlh);
2750 nlmsg_cancel(skb, nlh);
2754 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2755 struct netlink_ext_ack *extack)
2757 struct net *net = sock_net(in_skb->sk);
2759 struct nlattr *tb[RTA_MAX+1];
2760 struct fib_result res = {};
2761 struct rtable *rt = NULL;
2768 struct sk_buff *skb;
2769 u32 table_id = RT_TABLE_MAIN;
2772 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2777 rtm = nlmsg_data(nlh);
2779 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2785 /* Reserve room for dummy headers, this skb can pass
2786 through good chunk of routing engine.
2788 skb_reset_mac_header(skb);
2789 skb_reset_network_header(skb);
2791 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2792 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2793 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2794 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2796 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2798 uid = (iif ? INVALID_UID : current_uid());
2800 /* Bugfix: need to give ip_route_input enough of an IP header to
2803 ip_hdr(skb)->protocol = IPPROTO_UDP;
2804 ip_hdr(skb)->saddr = src;
2805 ip_hdr(skb)->daddr = dst;
2807 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2809 memset(&fl4, 0, sizeof(fl4));
2812 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
2813 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2814 fl4.flowi4_mark = mark;
2815 fl4.flowi4_uid = uid;
2820 struct net_device *dev;
2822 dev = dev_get_by_index_rcu(net, iif);
2828 skb->protocol = htons(ETH_P_IP);
2831 err = ip_route_input_rcu(skb, dst, src,
2832 rtm->rtm_tos & IPTOS_RT_MASK, dev,
2835 rt = skb_rtable(skb);
2836 if (err == 0 && rt->dst.error)
2837 err = -rt->dst.error;
2839 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2840 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2845 skb_dst_set(skb, &rt->dst);
2851 if (rtm->rtm_flags & RTM_F_NOTIFY)
2852 rt->rt_flags |= RTCF_NOTIFY;
2854 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2855 table_id = rt->rt_table_id;
2857 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2859 err = fib_props[res.type].error;
2861 err = -EHOSTUNREACH;
2864 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2865 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2866 rt->rt_type, res.prefix, res.prefixlen,
2867 fl4.flowi4_tos, res.fi, 0);
2869 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2870 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2877 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2887 void ip_rt_multicast_event(struct in_device *in_dev)
2889 rt_cache_flush(dev_net(in_dev->dev));
2892 #ifdef CONFIG_SYSCTL
2893 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2894 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2895 static int ip_rt_gc_elasticity __read_mostly = 8;
2896 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
2898 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2899 void __user *buffer,
2900 size_t *lenp, loff_t *ppos)
2902 struct net *net = (struct net *)__ctl->extra1;
2905 rt_cache_flush(net);
2906 fnhe_genid_bump(net);
2913 static struct ctl_table ipv4_route_table[] = {
2915 .procname = "gc_thresh",
2916 .data = &ipv4_dst_ops.gc_thresh,
2917 .maxlen = sizeof(int),
2919 .proc_handler = proc_dointvec,
2922 .procname = "max_size",
2923 .data = &ip_rt_max_size,
2924 .maxlen = sizeof(int),
2926 .proc_handler = proc_dointvec,
2929 /* Deprecated. Use gc_min_interval_ms */
2931 .procname = "gc_min_interval",
2932 .data = &ip_rt_gc_min_interval,
2933 .maxlen = sizeof(int),
2935 .proc_handler = proc_dointvec_jiffies,
2938 .procname = "gc_min_interval_ms",
2939 .data = &ip_rt_gc_min_interval,
2940 .maxlen = sizeof(int),
2942 .proc_handler = proc_dointvec_ms_jiffies,
2945 .procname = "gc_timeout",
2946 .data = &ip_rt_gc_timeout,
2947 .maxlen = sizeof(int),
2949 .proc_handler = proc_dointvec_jiffies,
2952 .procname = "gc_interval",
2953 .data = &ip_rt_gc_interval,
2954 .maxlen = sizeof(int),
2956 .proc_handler = proc_dointvec_jiffies,
2959 .procname = "redirect_load",
2960 .data = &ip_rt_redirect_load,
2961 .maxlen = sizeof(int),
2963 .proc_handler = proc_dointvec,
2966 .procname = "redirect_number",
2967 .data = &ip_rt_redirect_number,
2968 .maxlen = sizeof(int),
2970 .proc_handler = proc_dointvec,
2973 .procname = "redirect_silence",
2974 .data = &ip_rt_redirect_silence,
2975 .maxlen = sizeof(int),
2977 .proc_handler = proc_dointvec,
2980 .procname = "error_cost",
2981 .data = &ip_rt_error_cost,
2982 .maxlen = sizeof(int),
2984 .proc_handler = proc_dointvec,
2987 .procname = "error_burst",
2988 .data = &ip_rt_error_burst,
2989 .maxlen = sizeof(int),
2991 .proc_handler = proc_dointvec,
2994 .procname = "gc_elasticity",
2995 .data = &ip_rt_gc_elasticity,
2996 .maxlen = sizeof(int),
2998 .proc_handler = proc_dointvec,
3001 .procname = "mtu_expires",
3002 .data = &ip_rt_mtu_expires,
3003 .maxlen = sizeof(int),
3005 .proc_handler = proc_dointvec_jiffies,
3008 .procname = "min_pmtu",
3009 .data = &ip_rt_min_pmtu,
3010 .maxlen = sizeof(int),
3012 .proc_handler = proc_dointvec_minmax,
3013 .extra1 = &ip_min_valid_pmtu,
3016 .procname = "min_adv_mss",
3017 .data = &ip_rt_min_advmss,
3018 .maxlen = sizeof(int),
3020 .proc_handler = proc_dointvec,
3025 static struct ctl_table ipv4_route_flush_table[] = {
3027 .procname = "flush",
3028 .maxlen = sizeof(int),
3030 .proc_handler = ipv4_sysctl_rtcache_flush,
3035 static __net_init int sysctl_route_net_init(struct net *net)
3037 struct ctl_table *tbl;
3039 tbl = ipv4_route_flush_table;
3040 if (!net_eq(net, &init_net)) {
3041 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3045 /* Don't export sysctls to unprivileged users */
3046 if (net->user_ns != &init_user_ns)
3047 tbl[0].procname = NULL;
3049 tbl[0].extra1 = net;
3051 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3052 if (!net->ipv4.route_hdr)
3057 if (tbl != ipv4_route_flush_table)
3063 static __net_exit void sysctl_route_net_exit(struct net *net)
3065 struct ctl_table *tbl;
3067 tbl = net->ipv4.route_hdr->ctl_table_arg;
3068 unregister_net_sysctl_table(net->ipv4.route_hdr);
3069 BUG_ON(tbl == ipv4_route_flush_table);
3073 static __net_initdata struct pernet_operations sysctl_route_ops = {
3074 .init = sysctl_route_net_init,
3075 .exit = sysctl_route_net_exit,
3079 static __net_init int rt_genid_init(struct net *net)
3081 atomic_set(&net->ipv4.rt_genid, 0);
3082 atomic_set(&net->fnhe_genid, 0);
3083 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3087 static __net_initdata struct pernet_operations rt_genid_ops = {
3088 .init = rt_genid_init,
3091 static int __net_init ipv4_inetpeer_init(struct net *net)
3093 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3097 inet_peer_base_init(bp);
3098 net->ipv4.peers = bp;
3102 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3104 struct inet_peer_base *bp = net->ipv4.peers;
3106 net->ipv4.peers = NULL;
3107 inetpeer_invalidate_tree(bp);
3111 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3112 .init = ipv4_inetpeer_init,
3113 .exit = ipv4_inetpeer_exit,
3116 #ifdef CONFIG_IP_ROUTE_CLASSID
3117 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3118 #endif /* CONFIG_IP_ROUTE_CLASSID */
3120 int __init ip_rt_init(void)
3126 /* For modern hosts, this will use 2 MB of memory */
3127 idents_hash = alloc_large_system_hash("IP idents",
3128 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3130 16, /* one bucket per 64 KB */
3137 ip_idents = idents_hash;
3139 prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3141 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3143 for_each_possible_cpu(cpu) {
3144 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3146 INIT_LIST_HEAD(&ul->head);
3147 spin_lock_init(&ul->lock);
3149 #ifdef CONFIG_IP_ROUTE_CLASSID
3150 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3152 panic("IP: failed to allocate ip_rt_acct\n");
3155 ipv4_dst_ops.kmem_cachep =
3156 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3157 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3159 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3161 if (dst_entries_init(&ipv4_dst_ops) < 0)
3162 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3164 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3165 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3167 ipv4_dst_ops.gc_thresh = ~0;
3168 ip_rt_max_size = INT_MAX;
3173 if (ip_rt_proc_init())
3174 pr_err("Unable to create route proc files\n");
3179 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3180 RTNL_FLAG_DOIT_UNLOCKED);
3182 #ifdef CONFIG_SYSCTL
3183 register_pernet_subsys(&sysctl_route_ops);
3185 register_pernet_subsys(&rt_genid_ops);
3186 register_pernet_subsys(&ipv4_inetpeer_ops);
3190 #ifdef CONFIG_SYSCTL
3192 * We really need to sanitize the damn ipv4 init order, then all
3193 * this nonsense will go away.
3195 void __init ip_static_sysctl_init(void)
3197 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);