1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * ROUTE - implementation of the IP router.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * (rco@di.uminho.pt) Routing table insertion and update
20 * Linus Torvalds : Rewrote bits to be sensible
21 * Alan Cox : Added BSD route gw semantics
22 * Alan Cox : Super /proc >4K
23 * Alan Cox : MTU in route table
24 * Alan Cox : MSS actually. Also added the window
26 * Sam Lantinga : Fixed route matching in rt_del()
27 * Alan Cox : Routing cache support.
28 * Alan Cox : Removed compatibility cruft.
29 * Alan Cox : RTF_REJECT support.
30 * Alan Cox : TCP irtt support.
31 * Jonathan Naylor : Added Metric support.
32 * Miquel van Smoorenburg : BSD API fixes.
33 * Miquel van Smoorenburg : Metrics.
34 * Alan Cox : Use __u32 properly
35 * Alan Cox : Aligned routing errors more closely with BSD
36 * our system is still very different.
37 * Alan Cox : Faster /proc handling
38 * Alexey Kuznetsov : Massive rework to support tree based routing,
39 * routing caches and better behaviour.
41 * Olaf Erb : irtt wasn't being copied right.
42 * Bjorn Ekwall : Kerneld route support.
43 * Alan Cox : Multicast fixed (I hope)
44 * Pavel Krauz : Limited broadcast fixed
45 * Mike McLagan : Routing by source
46 * Alexey Kuznetsov : End of old history. Split to fib.c and
47 * route.c and rewritten from scratch.
48 * Andi Kleen : Load-limit warning messages.
49 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
53 * Marc Boucher : routing by fwmark
54 * Robert Olsson : Added rt_cache statistics
55 * Arnaldo C. Melo : Convert proc stuff to seq_file
56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
58 * Ilia Sotnikov : Removed TOS from hash calculations
61 #define pr_fmt(fmt) "IPv4: " fmt
63 #include <linux/module.h>
64 #include <linux/uaccess.h>
65 #include <linux/bitops.h>
66 #include <linux/types.h>
67 #include <linux/kernel.h>
69 #include <linux/string.h>
70 #include <linux/socket.h>
71 #include <linux/sockios.h>
72 #include <linux/errno.h>
74 #include <linux/inet.h>
75 #include <linux/netdevice.h>
76 #include <linux/proc_fs.h>
77 #include <linux/init.h>
78 #include <linux/skbuff.h>
79 #include <linux/inetdevice.h>
80 #include <linux/igmp.h>
81 #include <linux/pkt_sched.h>
82 #include <linux/mroute.h>
83 #include <linux/netfilter_ipv4.h>
84 #include <linux/random.h>
85 #include <linux/rcupdate.h>
86 #include <linux/times.h>
87 #include <linux/slab.h>
88 #include <linux/jhash.h>
90 #include <net/dst_metadata.h>
91 #include <net/net_namespace.h>
92 #include <net/protocol.h>
94 #include <net/route.h>
95 #include <net/inetpeer.h>
97 #include <net/ip_fib.h>
98 #include <net/nexthop.h>
101 #include <net/icmp.h>
102 #include <net/xfrm.h>
103 #include <net/lwtunnel.h>
104 #include <net/netevent.h>
105 #include <net/rtnetlink.h>
107 #include <linux/sysctl.h>
109 #include <net/secure_seq.h>
110 #include <net/ip_tunnels.h>
111 #include <net/l3mdev.h>
113 #include "fib_lookup.h"
115 #define RT_FL_TOS(oldflp4) \
116 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118 #define RT_GC_TIMEOUT (300*HZ)
120 static int ip_rt_max_size;
121 static int ip_rt_redirect_number __read_mostly = 9;
122 static int ip_rt_redirect_load __read_mostly = HZ / 50;
123 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
124 static int ip_rt_error_cost __read_mostly = HZ;
125 static int ip_rt_error_burst __read_mostly = 5 * HZ;
126 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
127 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
128 static int ip_rt_min_advmss __read_mostly = 256;
130 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
133 * Interface to generic destination cache.
136 INDIRECT_CALLABLE_SCOPE
137 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
138 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
139 INDIRECT_CALLABLE_SCOPE
140 unsigned int ipv4_mtu(const struct dst_entry *dst);
141 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
142 static void ipv4_link_failure(struct sk_buff *skb);
143 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
144 struct sk_buff *skb, u32 mtu,
146 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
147 struct sk_buff *skb);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
159 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
161 static struct dst_ops ipv4_dst_ops = {
163 .check = ipv4_dst_check,
164 .default_advmss = ipv4_default_advmss,
166 .cow_metrics = ipv4_cow_metrics,
167 .destroy = ipv4_dst_destroy,
168 .negative_advice = ipv4_negative_advice,
169 .link_failure = ipv4_link_failure,
170 .update_pmtu = ip_rt_update_pmtu,
171 .redirect = ip_do_redirect,
172 .local_out = __ip_local_out,
173 .neigh_lookup = ipv4_neigh_lookup,
174 .confirm_neigh = ipv4_confirm_neigh,
177 #define ECN_OR_COST(class) TC_PRIO_##class
179 const __u8 ip_tos2prio[16] = {
181 ECN_OR_COST(BESTEFFORT),
183 ECN_OR_COST(BESTEFFORT),
189 ECN_OR_COST(INTERACTIVE),
191 ECN_OR_COST(INTERACTIVE),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK),
194 TC_PRIO_INTERACTIVE_BULK,
195 ECN_OR_COST(INTERACTIVE_BULK)
197 EXPORT_SYMBOL(ip_tos2prio);
199 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
200 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
202 #ifdef CONFIG_PROC_FS
203 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
207 return SEQ_START_TOKEN;
210 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
216 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
220 static int rt_cache_seq_show(struct seq_file *seq, void *v)
222 if (v == SEQ_START_TOKEN)
223 seq_printf(seq, "%-127s\n",
224 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
225 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
230 static const struct seq_operations rt_cache_seq_ops = {
231 .start = rt_cache_seq_start,
232 .next = rt_cache_seq_next,
233 .stop = rt_cache_seq_stop,
234 .show = rt_cache_seq_show,
237 static int rt_cache_seq_open(struct inode *inode, struct file *file)
239 return seq_open(file, &rt_cache_seq_ops);
242 static const struct proc_ops rt_cache_proc_ops = {
243 .proc_open = rt_cache_seq_open,
244 .proc_read = seq_read,
245 .proc_lseek = seq_lseek,
246 .proc_release = seq_release,
250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
255 return SEQ_START_TOKEN;
257 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
258 if (!cpu_possible(cpu))
261 return &per_cpu(rt_cache_stat, cpu);
266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
270 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
271 if (!cpu_possible(cpu))
274 return &per_cpu(rt_cache_stat, cpu);
281 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
286 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
288 struct rt_cache_stat *st = v;
290 if (v == SEQ_START_TOKEN) {
291 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
295 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
296 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
297 dst_entries_get_slow(&ipv4_dst_ops),
310 0, /* st->gc_total */
311 0, /* st->gc_ignored */
312 0, /* st->gc_goal_miss */
313 0, /* st->gc_dst_overflow */
314 0, /* st->in_hlist_search */
315 0 /* st->out_hlist_search */
320 static const struct seq_operations rt_cpu_seq_ops = {
321 .start = rt_cpu_seq_start,
322 .next = rt_cpu_seq_next,
323 .stop = rt_cpu_seq_stop,
324 .show = rt_cpu_seq_show,
328 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
330 return seq_open(file, &rt_cpu_seq_ops);
333 static const struct proc_ops rt_cpu_proc_ops = {
334 .proc_open = rt_cpu_seq_open,
335 .proc_read = seq_read,
336 .proc_lseek = seq_lseek,
337 .proc_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));
366 static int __net_init ip_rt_do_proc_init(struct net *net)
368 struct proc_dir_entry *pde;
370 pde = proc_create("rt_cache", 0444, net->proc_net,
375 pde = proc_create("rt_cache", 0444,
376 net->proc_net_stat, &rt_cpu_proc_ops);
380 #ifdef CONFIG_IP_ROUTE_CLASSID
381 pde = proc_create_single("rt_acct", 0, net->proc_net,
388 #ifdef CONFIG_IP_ROUTE_CLASSID
390 remove_proc_entry("rt_cache", net->proc_net_stat);
393 remove_proc_entry("rt_cache", net->proc_net);
398 static void __net_exit ip_rt_do_proc_exit(struct net *net)
400 remove_proc_entry("rt_cache", net->proc_net_stat);
401 remove_proc_entry("rt_cache", net->proc_net);
402 #ifdef CONFIG_IP_ROUTE_CLASSID
403 remove_proc_entry("rt_acct", net->proc_net);
407 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
408 .init = ip_rt_do_proc_init,
409 .exit = ip_rt_do_proc_exit,
412 static int __init ip_rt_proc_init(void)
414 return register_pernet_subsys(&ip_rt_proc_ops);
418 static inline int ip_rt_proc_init(void)
422 #endif /* CONFIG_PROC_FS */
424 static inline bool rt_is_expired(const struct rtable *rth)
426 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
429 void rt_cache_flush(struct net *net)
431 rt_genid_bump_ipv4(net);
434 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
438 const struct rtable *rt = container_of(dst, struct rtable, dst);
439 struct net_device *dev = dst->dev;
444 if (likely(rt->rt_gw_family == AF_INET)) {
445 n = ip_neigh_gw4(dev, rt->rt_gw4);
446 } else if (rt->rt_gw_family == AF_INET6) {
447 n = ip_neigh_gw6(dev, &rt->rt_gw6);
451 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
452 n = ip_neigh_gw4(dev, pkey);
455 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
458 rcu_read_unlock_bh();
463 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
465 const struct rtable *rt = container_of(dst, struct rtable, dst);
466 struct net_device *dev = dst->dev;
467 const __be32 *pkey = daddr;
469 if (rt->rt_gw_family == AF_INET) {
470 pkey = (const __be32 *)&rt->rt_gw4;
471 } else if (rt->rt_gw_family == AF_INET6) {
472 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
475 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
478 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
481 #define IP_IDENTS_SZ 2048u
483 static atomic_t *ip_idents __read_mostly;
484 static u32 *ip_tstamps __read_mostly;
486 /* In order to protect privacy, we add a perturbation to identifiers
487 * if one generator is seldom used. This makes hard for an attacker
488 * to infer how many packets were sent between two points in time.
490 u32 ip_idents_reserve(u32 hash, int segs)
492 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
493 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
494 u32 old = READ_ONCE(*p_tstamp);
495 u32 now = (u32)jiffies;
498 if (old != now && cmpxchg(p_tstamp, old, now) == old)
499 delta = prandom_u32_max(now - old);
501 /* If UBSAN reports an error there, please make sure your compiler
502 * supports -fno-strict-overflow before reporting it that was a bug
503 * in UBSAN, and it has been fixed in GCC-8.
505 return atomic_add_return(segs + delta, p_id) - segs;
507 EXPORT_SYMBOL(ip_idents_reserve);
509 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
513 /* Note the following code is not safe, but this is okay. */
514 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
515 get_random_bytes(&net->ipv4.ip_id_key,
516 sizeof(net->ipv4.ip_id_key));
518 hash = siphash_3u32((__force u32)iph->daddr,
519 (__force u32)iph->saddr,
521 &net->ipv4.ip_id_key);
522 id = ip_idents_reserve(hash, segs);
525 EXPORT_SYMBOL(__ip_select_ident);
527 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
528 const struct sock *sk,
529 const struct iphdr *iph,
531 u8 prot, u32 mark, int flow_flags)
534 const struct inet_sock *inet = inet_sk(sk);
536 oif = sk->sk_bound_dev_if;
538 tos = RT_CONN_FLAGS(sk);
539 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
541 flowi4_init_output(fl4, oif, mark, tos,
542 RT_SCOPE_UNIVERSE, prot,
544 iph->daddr, iph->saddr, 0, 0,
545 sock_net_uid(net, sk));
548 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
549 const struct sock *sk)
551 const struct net *net = dev_net(skb->dev);
552 const struct iphdr *iph = ip_hdr(skb);
553 int oif = skb->dev->ifindex;
554 u8 tos = RT_TOS(iph->tos);
555 u8 prot = iph->protocol;
556 u32 mark = skb->mark;
558 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
561 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
563 const struct inet_sock *inet = inet_sk(sk);
564 const struct ip_options_rcu *inet_opt;
565 __be32 daddr = inet->inet_daddr;
568 inet_opt = rcu_dereference(inet->inet_opt);
569 if (inet_opt && inet_opt->opt.srr)
570 daddr = inet_opt->opt.faddr;
571 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
572 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
573 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
574 inet_sk_flowi_flags(sk),
575 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
579 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
580 const struct sk_buff *skb)
583 build_skb_flow_key(fl4, skb, sk);
585 build_sk_flow_key(fl4, sk);
588 static DEFINE_SPINLOCK(fnhe_lock);
590 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
594 rt = rcu_dereference(fnhe->fnhe_rth_input);
596 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
597 dst_dev_put(&rt->dst);
598 dst_release(&rt->dst);
600 rt = rcu_dereference(fnhe->fnhe_rth_output);
602 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
603 dst_dev_put(&rt->dst);
604 dst_release(&rt->dst);
608 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
610 struct fib_nh_exception *fnhe, *oldest;
612 oldest = rcu_dereference(hash->chain);
613 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
614 fnhe = rcu_dereference(fnhe->fnhe_next)) {
615 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
618 fnhe_flush_routes(oldest);
622 static inline u32 fnhe_hashfun(__be32 daddr)
624 static u32 fnhe_hashrnd __read_mostly;
627 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
628 hval = jhash_1word((__force u32)daddr, fnhe_hashrnd);
629 return hash_32(hval, FNHE_HASH_SHIFT);
632 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
634 rt->rt_pmtu = fnhe->fnhe_pmtu;
635 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
636 rt->dst.expires = fnhe->fnhe_expires;
639 rt->rt_flags |= RTCF_REDIRECTED;
640 rt->rt_uses_gateway = 1;
641 rt->rt_gw_family = AF_INET;
642 rt->rt_gw4 = fnhe->fnhe_gw;
646 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
647 __be32 gw, u32 pmtu, bool lock,
648 unsigned long expires)
650 struct fnhe_hash_bucket *hash;
651 struct fib_nh_exception *fnhe;
657 genid = fnhe_genid(dev_net(nhc->nhc_dev));
658 hval = fnhe_hashfun(daddr);
660 spin_lock_bh(&fnhe_lock);
662 hash = rcu_dereference(nhc->nhc_exceptions);
664 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
667 rcu_assign_pointer(nhc->nhc_exceptions, hash);
673 for (fnhe = rcu_dereference(hash->chain); fnhe;
674 fnhe = rcu_dereference(fnhe->fnhe_next)) {
675 if (fnhe->fnhe_daddr == daddr)
681 if (fnhe->fnhe_genid != genid)
682 fnhe->fnhe_genid = genid;
686 fnhe->fnhe_pmtu = pmtu;
687 fnhe->fnhe_mtu_locked = lock;
689 fnhe->fnhe_expires = max(1UL, expires);
690 /* Update all cached dsts too */
691 rt = rcu_dereference(fnhe->fnhe_rth_input);
693 fill_route_from_fnhe(rt, fnhe);
694 rt = rcu_dereference(fnhe->fnhe_rth_output);
696 fill_route_from_fnhe(rt, fnhe);
698 if (depth > FNHE_RECLAIM_DEPTH)
699 fnhe = fnhe_oldest(hash);
701 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
705 fnhe->fnhe_next = hash->chain;
706 rcu_assign_pointer(hash->chain, fnhe);
708 fnhe->fnhe_genid = genid;
709 fnhe->fnhe_daddr = daddr;
711 fnhe->fnhe_pmtu = pmtu;
712 fnhe->fnhe_mtu_locked = lock;
713 fnhe->fnhe_expires = max(1UL, expires);
715 /* Exception created; mark the cached routes for the nexthop
716 * stale, so anyone caching it rechecks if this exception
719 rt = rcu_dereference(nhc->nhc_rth_input);
721 rt->dst.obsolete = DST_OBSOLETE_KILL;
723 for_each_possible_cpu(i) {
724 struct rtable __rcu **prt;
725 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
726 rt = rcu_dereference(*prt);
728 rt->dst.obsolete = DST_OBSOLETE_KILL;
732 fnhe->fnhe_stamp = jiffies;
735 spin_unlock_bh(&fnhe_lock);
738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
741 __be32 new_gw = icmp_hdr(skb)->un.gateway;
742 __be32 old_gw = ip_hdr(skb)->saddr;
743 struct net_device *dev = skb->dev;
744 struct in_device *in_dev;
745 struct fib_result res;
749 switch (icmp_hdr(skb)->code & 7) {
751 case ICMP_REDIR_NETTOS:
752 case ICMP_REDIR_HOST:
753 case ICMP_REDIR_HOSTTOS:
760 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
763 in_dev = __in_dev_get_rcu(dev);
768 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
769 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
770 ipv4_is_zeronet(new_gw))
771 goto reject_redirect;
773 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
774 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
775 goto reject_redirect;
776 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
777 goto reject_redirect;
779 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
780 goto reject_redirect;
783 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
785 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
787 if (!(n->nud_state & NUD_VALID)) {
788 neigh_event_send(n, NULL);
790 if (fib_lookup(net, fl4, &res, 0) == 0) {
791 struct fib_nh_common *nhc;
793 fib_select_path(net, &res, fl4, skb);
794 nhc = FIB_RES_NHC(res);
795 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
797 jiffies + ip_rt_gc_timeout);
800 rt->dst.obsolete = DST_OBSOLETE_KILL;
801 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
808 #ifdef CONFIG_IP_ROUTE_VERBOSE
809 if (IN_DEV_LOG_MARTIANS(in_dev)) {
810 const struct iphdr *iph = (const struct iphdr *) skb->data;
811 __be32 daddr = iph->daddr;
812 __be32 saddr = iph->saddr;
814 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
815 " Advised path = %pI4 -> %pI4\n",
816 &old_gw, dev->name, &new_gw,
823 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
827 const struct iphdr *iph = (const struct iphdr *) skb->data;
828 struct net *net = dev_net(skb->dev);
829 int oif = skb->dev->ifindex;
830 u8 tos = RT_TOS(iph->tos);
831 u8 prot = iph->protocol;
832 u32 mark = skb->mark;
834 rt = (struct rtable *) dst;
836 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
837 __ip_do_redirect(rt, skb, &fl4, true);
840 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
842 struct rtable *rt = (struct rtable *)dst;
843 struct dst_entry *ret = dst;
846 if (dst->obsolete > 0) {
849 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
860 * 1. The first ip_rt_redirect_number redirects are sent
861 * with exponential backoff, then we stop sending them at all,
862 * assuming that the host ignores our redirects.
863 * 2. If we did not see packets requiring redirects
864 * during ip_rt_redirect_silence, we assume that the host
865 * forgot redirected route and start to send redirects again.
867 * This algorithm is much cheaper and more intelligent than dumb load limiting
870 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
871 * and "frag. need" (breaks PMTU discovery) in icmp.c.
874 void ip_rt_send_redirect(struct sk_buff *skb)
876 struct rtable *rt = skb_rtable(skb);
877 struct in_device *in_dev;
878 struct inet_peer *peer;
884 in_dev = __in_dev_get_rcu(rt->dst.dev);
885 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
889 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
890 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
893 net = dev_net(rt->dst.dev);
894 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
896 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
897 rt_nexthop(rt, ip_hdr(skb)->daddr));
901 /* No redirected packets during ip_rt_redirect_silence;
902 * reset the algorithm.
904 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
905 peer->rate_tokens = 0;
906 peer->n_redirects = 0;
909 /* Too many ignored redirects; do not send anything
910 * set dst.rate_last to the last seen redirected packet.
912 if (peer->n_redirects >= ip_rt_redirect_number) {
913 peer->rate_last = jiffies;
917 /* Check for load limit; set rate_last to the latest sent
920 if (peer->n_redirects == 0 ||
923 (ip_rt_redirect_load << peer->n_redirects)))) {
924 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
926 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
927 peer->rate_last = jiffies;
929 #ifdef CONFIG_IP_ROUTE_VERBOSE
931 peer->n_redirects == ip_rt_redirect_number)
932 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
933 &ip_hdr(skb)->saddr, inet_iif(skb),
934 &ip_hdr(skb)->daddr, &gw);
941 static int ip_error(struct sk_buff *skb)
943 struct rtable *rt = skb_rtable(skb);
944 struct net_device *dev = skb->dev;
945 struct in_device *in_dev;
946 struct inet_peer *peer;
952 if (netif_is_l3_master(skb->dev)) {
953 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
958 in_dev = __in_dev_get_rcu(dev);
960 /* IP on this device is disabled. */
964 net = dev_net(rt->dst.dev);
965 if (!IN_DEV_FORWARD(in_dev)) {
966 switch (rt->dst.error) {
968 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
972 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
978 switch (rt->dst.error) {
983 code = ICMP_HOST_UNREACH;
986 code = ICMP_NET_UNREACH;
987 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
990 code = ICMP_PKT_FILTERED;
994 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
995 l3mdev_master_ifindex(skb->dev), 1);
1000 peer->rate_tokens += now - peer->rate_last;
1001 if (peer->rate_tokens > ip_rt_error_burst)
1002 peer->rate_tokens = ip_rt_error_burst;
1003 peer->rate_last = now;
1004 if (peer->rate_tokens >= ip_rt_error_cost)
1005 peer->rate_tokens -= ip_rt_error_cost;
1011 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1013 out: kfree_skb(skb);
1017 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1019 struct dst_entry *dst = &rt->dst;
1020 struct net *net = dev_net(dst->dev);
1021 struct fib_result res;
1025 if (ip_mtu_locked(dst))
1028 old_mtu = ipv4_mtu(dst);
1032 if (mtu < ip_rt_min_pmtu) {
1034 mtu = min(old_mtu, ip_rt_min_pmtu);
1037 if (rt->rt_pmtu == mtu && !lock &&
1038 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1042 if (fib_lookup(net, fl4, &res, 0) == 0) {
1043 struct fib_nh_common *nhc;
1045 fib_select_path(net, &res, fl4, NULL);
1046 nhc = FIB_RES_NHC(res);
1047 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1048 jiffies + ip_rt_mtu_expires);
1053 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1054 struct sk_buff *skb, u32 mtu,
1057 struct rtable *rt = (struct rtable *) dst;
1060 ip_rt_build_flow_key(&fl4, sk, skb);
1062 /* Don't make lookup fail for bridged encapsulations */
1063 if (skb && netif_is_any_bridge_port(skb->dev))
1066 __ip_rt_update_pmtu(rt, &fl4, mtu);
1069 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1070 int oif, u8 protocol)
1072 const struct iphdr *iph = (const struct iphdr *)skb->data;
1075 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1077 __build_flow_key(net, &fl4, NULL, iph, oif,
1078 RT_TOS(iph->tos), protocol, mark, 0);
1079 rt = __ip_route_output_key(net, &fl4);
1081 __ip_rt_update_pmtu(rt, &fl4, mtu);
1085 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1087 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1089 const struct iphdr *iph = (const struct iphdr *)skb->data;
1093 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1095 if (!fl4.flowi4_mark)
1096 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1098 rt = __ip_route_output_key(sock_net(sk), &fl4);
1100 __ip_rt_update_pmtu(rt, &fl4, mtu);
1105 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1107 const struct iphdr *iph = (const struct iphdr *)skb->data;
1110 struct dst_entry *odst = NULL;
1112 struct net *net = sock_net(sk);
1116 if (!ip_sk_accept_pmtu(sk))
1119 odst = sk_dst_get(sk);
1121 if (sock_owned_by_user(sk) || !odst) {
1122 __ipv4_sk_update_pmtu(skb, sk, mtu);
1126 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1128 rt = (struct rtable *)odst;
1129 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1130 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1137 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1139 if (!dst_check(&rt->dst, 0)) {
1141 dst_release(&rt->dst);
1143 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1151 sk_dst_set(sk, &rt->dst);
1157 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1159 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1160 int oif, u8 protocol)
1162 const struct iphdr *iph = (const struct iphdr *)skb->data;
1166 __build_flow_key(net, &fl4, NULL, iph, oif,
1167 RT_TOS(iph->tos), protocol, 0, 0);
1168 rt = __ip_route_output_key(net, &fl4);
1170 __ip_do_redirect(rt, skb, &fl4, false);
1174 EXPORT_SYMBOL_GPL(ipv4_redirect);
1176 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1178 const struct iphdr *iph = (const struct iphdr *)skb->data;
1181 struct net *net = sock_net(sk);
1183 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1184 rt = __ip_route_output_key(net, &fl4);
1186 __ip_do_redirect(rt, skb, &fl4, false);
1190 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1192 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1195 struct rtable *rt = (struct rtable *) dst;
1197 /* All IPV4 dsts are created with ->obsolete set to the value
1198 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1199 * into this function always.
1201 * When a PMTU/redirect information update invalidates a route,
1202 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1203 * DST_OBSOLETE_DEAD.
1205 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1209 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1211 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1213 struct ip_options opt;
1216 /* Recompile ip options since IPCB may not be valid anymore.
1217 * Also check we have a reasonable ipv4 header.
1219 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1220 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1223 memset(&opt, 0, sizeof(opt));
1224 if (ip_hdr(skb)->ihl > 5) {
1225 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1227 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1230 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1236 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1239 static void ipv4_link_failure(struct sk_buff *skb)
1243 ipv4_send_dest_unreach(skb);
1245 rt = skb_rtable(skb);
1247 dst_set_expires(&rt->dst, 0);
1250 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1252 pr_debug("%s: %pI4 -> %pI4, %s\n",
1253 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1254 skb->dev ? skb->dev->name : "?");
1261 We do not cache source address of outgoing interface,
1262 because it is used only by IP RR, TS and SRR options,
1263 so that it out of fast path.
1265 BTW remember: "addr" is allowed to be not aligned
1269 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1273 if (rt_is_output_route(rt))
1274 src = ip_hdr(skb)->saddr;
1276 struct fib_result res;
1277 struct iphdr *iph = ip_hdr(skb);
1278 struct flowi4 fl4 = {
1279 .daddr = iph->daddr,
1280 .saddr = iph->saddr,
1281 .flowi4_tos = RT_TOS(iph->tos),
1282 .flowi4_oif = rt->dst.dev->ifindex,
1283 .flowi4_iif = skb->dev->ifindex,
1284 .flowi4_mark = skb->mark,
1288 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1289 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1291 src = inet_select_addr(rt->dst.dev,
1292 rt_nexthop(rt, iph->daddr),
1296 memcpy(addr, &src, 4);
1299 #ifdef CONFIG_IP_ROUTE_CLASSID
1300 static void set_class_tag(struct rtable *rt, u32 tag)
1302 if (!(rt->dst.tclassid & 0xFFFF))
1303 rt->dst.tclassid |= tag & 0xFFFF;
1304 if (!(rt->dst.tclassid & 0xFFFF0000))
1305 rt->dst.tclassid |= tag & 0xFFFF0000;
1309 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1311 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1312 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1315 return min(advmss, IPV4_MAX_PMTU - header_size);
1318 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1320 const struct rtable *rt = (const struct rtable *)dst;
1321 unsigned int mtu = rt->rt_pmtu;
1323 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1324 mtu = dst_metric_raw(dst, RTAX_MTU);
1329 mtu = READ_ONCE(dst->dev->mtu);
1331 if (unlikely(ip_mtu_locked(dst))) {
1332 if (rt->rt_uses_gateway && mtu > 576)
1336 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1338 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1340 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1342 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1344 struct fnhe_hash_bucket *hash;
1345 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1346 u32 hval = fnhe_hashfun(daddr);
1348 spin_lock_bh(&fnhe_lock);
1350 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1351 lockdep_is_held(&fnhe_lock));
1354 fnhe_p = &hash->chain;
1355 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1357 if (fnhe->fnhe_daddr == daddr) {
1358 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1359 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1360 /* set fnhe_daddr to 0 to ensure it won't bind with
1361 * new dsts in rt_bind_exception().
1363 fnhe->fnhe_daddr = 0;
1364 fnhe_flush_routes(fnhe);
1365 kfree_rcu(fnhe, rcu);
1368 fnhe_p = &fnhe->fnhe_next;
1369 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1370 lockdep_is_held(&fnhe_lock));
1373 spin_unlock_bh(&fnhe_lock);
1376 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1379 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1380 struct fib_nh_exception *fnhe;
1386 hval = fnhe_hashfun(daddr);
1388 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1389 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1390 if (fnhe->fnhe_daddr == daddr) {
1391 if (fnhe->fnhe_expires &&
1392 time_after(jiffies, fnhe->fnhe_expires)) {
1393 ip_del_fnhe(nhc, daddr);
1403 * 1. mtu on route is locked - use it
1404 * 2. mtu from nexthop exception
1405 * 3. mtu from egress device
1408 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1410 struct fib_nh_common *nhc = res->nhc;
1411 struct net_device *dev = nhc->nhc_dev;
1412 struct fib_info *fi = res->fi;
1415 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1416 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1420 struct fib_nh_exception *fnhe;
1422 fnhe = find_exception(nhc, daddr);
1423 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1424 mtu = fnhe->fnhe_pmtu;
1428 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1430 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1433 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1434 __be32 daddr, const bool do_cache)
1438 spin_lock_bh(&fnhe_lock);
1440 if (daddr == fnhe->fnhe_daddr) {
1441 struct rtable __rcu **porig;
1442 struct rtable *orig;
1443 int genid = fnhe_genid(dev_net(rt->dst.dev));
1445 if (rt_is_input_route(rt))
1446 porig = &fnhe->fnhe_rth_input;
1448 porig = &fnhe->fnhe_rth_output;
1449 orig = rcu_dereference(*porig);
1451 if (fnhe->fnhe_genid != genid) {
1452 fnhe->fnhe_genid = genid;
1454 fnhe->fnhe_pmtu = 0;
1455 fnhe->fnhe_expires = 0;
1456 fnhe->fnhe_mtu_locked = false;
1457 fnhe_flush_routes(fnhe);
1460 fill_route_from_fnhe(rt, fnhe);
1463 rt->rt_gw_family = AF_INET;
1468 rcu_assign_pointer(*porig, rt);
1470 dst_dev_put(&orig->dst);
1471 dst_release(&orig->dst);
1476 fnhe->fnhe_stamp = jiffies;
1478 spin_unlock_bh(&fnhe_lock);
1483 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1485 struct rtable *orig, *prev, **p;
1488 if (rt_is_input_route(rt)) {
1489 p = (struct rtable **)&nhc->nhc_rth_input;
1491 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1495 /* hold dst before doing cmpxchg() to avoid race condition
1499 prev = cmpxchg(p, orig, rt);
1502 rt_add_uncached_list(orig);
1503 dst_release(&orig->dst);
1506 dst_release(&rt->dst);
1513 struct uncached_list {
1515 struct list_head head;
1518 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1520 void rt_add_uncached_list(struct rtable *rt)
1522 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1524 rt->rt_uncached_list = ul;
1526 spin_lock_bh(&ul->lock);
1527 list_add_tail(&rt->rt_uncached, &ul->head);
1528 spin_unlock_bh(&ul->lock);
1531 void rt_del_uncached_list(struct rtable *rt)
1533 if (!list_empty(&rt->rt_uncached)) {
1534 struct uncached_list *ul = rt->rt_uncached_list;
1536 spin_lock_bh(&ul->lock);
1537 list_del(&rt->rt_uncached);
1538 spin_unlock_bh(&ul->lock);
1542 static void ipv4_dst_destroy(struct dst_entry *dst)
1544 struct rtable *rt = (struct rtable *)dst;
1546 ip_dst_metrics_put(dst);
1547 rt_del_uncached_list(rt);
1550 void rt_flush_dev(struct net_device *dev)
1555 for_each_possible_cpu(cpu) {
1556 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1558 spin_lock_bh(&ul->lock);
1559 list_for_each_entry(rt, &ul->head, rt_uncached) {
1560 if (rt->dst.dev != dev)
1562 rt->dst.dev = blackhole_netdev;
1563 dev_hold(rt->dst.dev);
1566 spin_unlock_bh(&ul->lock);
1570 static bool rt_cache_valid(const struct rtable *rt)
1573 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1577 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1578 const struct fib_result *res,
1579 struct fib_nh_exception *fnhe,
1580 struct fib_info *fi, u16 type, u32 itag,
1581 const bool do_cache)
1583 bool cached = false;
1586 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1588 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1589 rt->rt_uses_gateway = 1;
1590 rt->rt_gw_family = nhc->nhc_gw_family;
1591 /* only INET and INET6 are supported */
1592 if (likely(nhc->nhc_gw_family == AF_INET))
1593 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1595 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1598 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1600 #ifdef CONFIG_IP_ROUTE_CLASSID
1601 if (nhc->nhc_family == AF_INET) {
1604 nh = container_of(nhc, struct fib_nh, nh_common);
1605 rt->dst.tclassid = nh->nh_tclassid;
1608 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1610 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1612 cached = rt_cache_route(nhc, rt);
1613 if (unlikely(!cached)) {
1614 /* Routes we intend to cache in nexthop exception or
1615 * FIB nexthop have the DST_NOCACHE bit clear.
1616 * However, if we are unsuccessful at storing this
1617 * route into the cache we really need to set it.
1620 rt->rt_gw_family = AF_INET;
1623 rt_add_uncached_list(rt);
1626 rt_add_uncached_list(rt);
1628 #ifdef CONFIG_IP_ROUTE_CLASSID
1629 #ifdef CONFIG_IP_MULTIPLE_TABLES
1630 set_class_tag(rt, res->tclassid);
1632 set_class_tag(rt, itag);
1636 struct rtable *rt_dst_alloc(struct net_device *dev,
1637 unsigned int flags, u16 type,
1638 bool nopolicy, bool noxfrm)
1642 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1643 (nopolicy ? DST_NOPOLICY : 0) |
1644 (noxfrm ? DST_NOXFRM : 0));
1647 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1648 rt->rt_flags = flags;
1650 rt->rt_is_input = 0;
1653 rt->rt_mtu_locked = 0;
1654 rt->rt_uses_gateway = 0;
1655 rt->rt_gw_family = 0;
1657 INIT_LIST_HEAD(&rt->rt_uncached);
1659 rt->dst.output = ip_output;
1660 if (flags & RTCF_LOCAL)
1661 rt->dst.input = ip_local_deliver;
1666 EXPORT_SYMBOL(rt_dst_alloc);
1668 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1670 struct rtable *new_rt;
1672 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1676 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1677 new_rt->rt_flags = rt->rt_flags;
1678 new_rt->rt_type = rt->rt_type;
1679 new_rt->rt_is_input = rt->rt_is_input;
1680 new_rt->rt_iif = rt->rt_iif;
1681 new_rt->rt_pmtu = rt->rt_pmtu;
1682 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1683 new_rt->rt_gw_family = rt->rt_gw_family;
1684 if (rt->rt_gw_family == AF_INET)
1685 new_rt->rt_gw4 = rt->rt_gw4;
1686 else if (rt->rt_gw_family == AF_INET6)
1687 new_rt->rt_gw6 = rt->rt_gw6;
1688 INIT_LIST_HEAD(&new_rt->rt_uncached);
1690 new_rt->dst.input = rt->dst.input;
1691 new_rt->dst.output = rt->dst.output;
1692 new_rt->dst.error = rt->dst.error;
1693 new_rt->dst.lastuse = jiffies;
1694 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1698 EXPORT_SYMBOL(rt_dst_clone);
1700 /* called in rcu_read_lock() section */
1701 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1702 u8 tos, struct net_device *dev,
1703 struct in_device *in_dev, u32 *itag)
1707 /* Primary sanity checks. */
1711 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1712 skb->protocol != htons(ETH_P_IP))
1715 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1718 if (ipv4_is_zeronet(saddr)) {
1719 if (!ipv4_is_local_multicast(daddr) &&
1720 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1723 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1731 /* called in rcu_read_lock() section */
1732 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1733 u8 tos, struct net_device *dev, int our)
1735 struct in_device *in_dev = __in_dev_get_rcu(dev);
1736 unsigned int flags = RTCF_MULTICAST;
1741 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1746 flags |= RTCF_LOCAL;
1748 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1749 IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1753 #ifdef CONFIG_IP_ROUTE_CLASSID
1754 rth->dst.tclassid = itag;
1756 rth->dst.output = ip_rt_bug;
1757 rth->rt_is_input= 1;
1759 #ifdef CONFIG_IP_MROUTE
1760 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1761 rth->dst.input = ip_mr_input;
1763 RT_CACHE_STAT_INC(in_slow_mc);
1765 skb_dst_set(skb, &rth->dst);
1770 static void ip_handle_martian_source(struct net_device *dev,
1771 struct in_device *in_dev,
1772 struct sk_buff *skb,
1776 RT_CACHE_STAT_INC(in_martian_src);
1777 #ifdef CONFIG_IP_ROUTE_VERBOSE
1778 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1780 * RFC1812 recommendation, if source is martian,
1781 * the only hint is MAC header.
1783 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1784 &daddr, &saddr, dev->name);
1785 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1786 print_hex_dump(KERN_WARNING, "ll header: ",
1787 DUMP_PREFIX_OFFSET, 16, 1,
1788 skb_mac_header(skb),
1789 dev->hard_header_len, false);
1795 /* called in rcu_read_lock() section */
1796 static int __mkroute_input(struct sk_buff *skb,
1797 const struct fib_result *res,
1798 struct in_device *in_dev,
1799 __be32 daddr, __be32 saddr, u32 tos)
1801 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1802 struct net_device *dev = nhc->nhc_dev;
1803 struct fib_nh_exception *fnhe;
1806 struct in_device *out_dev;
1810 /* get a working reference to the output device */
1811 out_dev = __in_dev_get_rcu(dev);
1813 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1817 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1818 in_dev->dev, in_dev, &itag);
1820 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1826 do_cache = res->fi && !itag;
1827 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1828 skb->protocol == htons(ETH_P_IP)) {
1831 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1832 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1833 inet_addr_onlink(out_dev, saddr, gw))
1834 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1837 if (skb->protocol != htons(ETH_P_IP)) {
1838 /* Not IP (i.e. ARP). Do not create route, if it is
1839 * invalid for proxy arp. DNAT routes are always valid.
1841 * Proxy arp feature have been extended to allow, ARP
1842 * replies back to the same interface, to support
1843 * Private VLAN switch technologies. See arp.c.
1845 if (out_dev == in_dev &&
1846 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1852 fnhe = find_exception(nhc, daddr);
1855 rth = rcu_dereference(fnhe->fnhe_rth_input);
1857 rth = rcu_dereference(nhc->nhc_rth_input);
1858 if (rt_cache_valid(rth)) {
1859 skb_dst_set_noref(skb, &rth->dst);
1864 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1865 IN_DEV_ORCONF(in_dev, NOPOLICY),
1866 IN_DEV_ORCONF(out_dev, NOXFRM));
1872 rth->rt_is_input = 1;
1873 RT_CACHE_STAT_INC(in_slow_tot);
1875 rth->dst.input = ip_forward;
1877 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1879 lwtunnel_set_redirect(&rth->dst);
1880 skb_dst_set(skb, &rth->dst);
1887 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1888 /* To make ICMP packets follow the right flow, the multipath hash is
1889 * calculated from the inner IP addresses.
1891 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1892 struct flow_keys *hash_keys)
1894 const struct iphdr *outer_iph = ip_hdr(skb);
1895 const struct iphdr *key_iph = outer_iph;
1896 const struct iphdr *inner_iph;
1897 const struct icmphdr *icmph;
1898 struct iphdr _inner_iph;
1899 struct icmphdr _icmph;
1901 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1904 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1907 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1912 if (!icmp_is_err(icmph->type))
1915 inner_iph = skb_header_pointer(skb,
1916 outer_iph->ihl * 4 + sizeof(_icmph),
1917 sizeof(_inner_iph), &_inner_iph);
1921 key_iph = inner_iph;
1923 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1924 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1927 /* if skb is set it will be used and fl4 can be NULL */
1928 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1929 const struct sk_buff *skb, struct flow_keys *flkeys)
1931 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1932 struct flow_keys hash_keys;
1935 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1937 memset(&hash_keys, 0, sizeof(hash_keys));
1938 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1940 ip_multipath_l3_keys(skb, &hash_keys);
1942 hash_keys.addrs.v4addrs.src = fl4->saddr;
1943 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1947 /* skb is currently provided only when forwarding */
1949 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1950 struct flow_keys keys;
1952 /* short-circuit if we already have L4 hash present */
1954 return skb_get_hash_raw(skb) >> 1;
1956 memset(&hash_keys, 0, sizeof(hash_keys));
1959 skb_flow_dissect_flow_keys(skb, &keys, flag);
1963 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1964 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1965 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1966 hash_keys.ports.src = flkeys->ports.src;
1967 hash_keys.ports.dst = flkeys->ports.dst;
1968 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1970 memset(&hash_keys, 0, sizeof(hash_keys));
1971 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1972 hash_keys.addrs.v4addrs.src = fl4->saddr;
1973 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1974 hash_keys.ports.src = fl4->fl4_sport;
1975 hash_keys.ports.dst = fl4->fl4_dport;
1976 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1980 memset(&hash_keys, 0, sizeof(hash_keys));
1981 /* skb is currently provided only when forwarding */
1983 struct flow_keys keys;
1985 skb_flow_dissect_flow_keys(skb, &keys, 0);
1986 /* Inner can be v4 or v6 */
1987 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1988 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1989 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1990 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1991 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1992 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1993 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1994 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1995 hash_keys.tags.flow_label = keys.tags.flow_label;
1996 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1998 /* Same as case 0 */
1999 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2000 ip_multipath_l3_keys(skb, &hash_keys);
2003 /* Same as case 0 */
2004 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2005 hash_keys.addrs.v4addrs.src = fl4->saddr;
2006 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2010 mhash = flow_hash_from_keys(&hash_keys);
2013 mhash = jhash_2words(mhash, multipath_hash, 0);
2017 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2019 static int ip_mkroute_input(struct sk_buff *skb,
2020 struct fib_result *res,
2021 struct in_device *in_dev,
2022 __be32 daddr, __be32 saddr, u32 tos,
2023 struct flow_keys *hkeys)
2025 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2026 if (res->fi && fib_info_num_path(res->fi) > 1) {
2027 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2029 fib_select_multipath(res, h);
2033 /* create a routing cache entry */
2034 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2037 /* Implements all the saddr-related checks as ip_route_input_slow(),
2038 * assuming daddr is valid and the destination is not a local broadcast one.
2039 * Uses the provided hint instead of performing a route lookup.
2041 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2042 u8 tos, struct net_device *dev,
2043 const struct sk_buff *hint)
2045 struct in_device *in_dev = __in_dev_get_rcu(dev);
2046 struct rtable *rt = skb_rtable(hint);
2047 struct net *net = dev_net(dev);
2051 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2052 goto martian_source;
2054 if (ipv4_is_zeronet(saddr))
2055 goto martian_source;
2057 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2058 goto martian_source;
2060 if (rt->rt_type != RTN_LOCAL)
2061 goto skip_validate_source;
2063 tos &= IPTOS_RT_MASK;
2064 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2066 goto martian_source;
2068 skip_validate_source:
2069 skb_dst_copy(skb, hint);
2073 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2078 * NOTE. We drop all the packets that has local source
2079 * addresses, because every properly looped back packet
2080 * must have correct destination already attached by output routine.
2081 * Changes in the enforced policies must be applied also to
2082 * ip_route_use_hint().
2084 * Such approach solves two big problems:
2085 * 1. Not simplex devices are handled properly.
2086 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2087 * called with rcu_read_lock()
2090 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2091 u8 tos, struct net_device *dev,
2092 struct fib_result *res)
2094 struct in_device *in_dev = __in_dev_get_rcu(dev);
2095 struct flow_keys *flkeys = NULL, _flkeys;
2096 struct net *net = dev_net(dev);
2097 struct ip_tunnel_info *tun_info;
2099 unsigned int flags = 0;
2103 bool do_cache = true;
2105 /* IP on this device is disabled. */
2110 /* Check for the most weird martians, which can be not detected
2114 tun_info = skb_tunnel_info(skb);
2115 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2116 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2118 fl4.flowi4_tun_key.tun_id = 0;
2121 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2122 goto martian_source;
2126 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2129 /* Accept zero addresses only to limited broadcast;
2130 * I even do not know to fix it or not. Waiting for complains :-)
2132 if (ipv4_is_zeronet(saddr))
2133 goto martian_source;
2135 if (ipv4_is_zeronet(daddr))
2136 goto martian_destination;
2138 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2139 * and call it once if daddr or/and saddr are loopback addresses
2141 if (ipv4_is_loopback(daddr)) {
2142 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2143 goto martian_destination;
2144 } else if (ipv4_is_loopback(saddr)) {
2145 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2146 goto martian_source;
2150 * Now we are ready to route packet.
2153 fl4.flowi4_iif = dev->ifindex;
2154 fl4.flowi4_mark = skb->mark;
2155 fl4.flowi4_tos = tos;
2156 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2157 fl4.flowi4_flags = 0;
2160 fl4.flowi4_uid = sock_net_uid(net, NULL);
2161 fl4.flowi4_multipath_hash = 0;
2163 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2166 fl4.flowi4_proto = 0;
2171 err = fib_lookup(net, &fl4, res, 0);
2173 if (!IN_DEV_FORWARD(in_dev))
2174 err = -EHOSTUNREACH;
2178 if (res->type == RTN_BROADCAST) {
2179 if (IN_DEV_BFORWARD(in_dev))
2181 /* not do cache if bc_forwarding is enabled */
2182 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2187 if (res->type == RTN_LOCAL) {
2188 err = fib_validate_source(skb, saddr, daddr, tos,
2189 0, dev, in_dev, &itag);
2191 goto martian_source;
2195 if (!IN_DEV_FORWARD(in_dev)) {
2196 err = -EHOSTUNREACH;
2199 if (res->type != RTN_UNICAST)
2200 goto martian_destination;
2203 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2207 if (skb->protocol != htons(ETH_P_IP))
2210 if (!ipv4_is_zeronet(saddr)) {
2211 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2214 goto martian_source;
2216 flags |= RTCF_BROADCAST;
2217 res->type = RTN_BROADCAST;
2218 RT_CACHE_STAT_INC(in_brd);
2221 do_cache &= res->fi && !itag;
2223 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2225 rth = rcu_dereference(nhc->nhc_rth_input);
2226 if (rt_cache_valid(rth)) {
2227 skb_dst_set_noref(skb, &rth->dst);
2233 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2234 flags | RTCF_LOCAL, res->type,
2235 IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2239 rth->dst.output= ip_rt_bug;
2240 #ifdef CONFIG_IP_ROUTE_CLASSID
2241 rth->dst.tclassid = itag;
2243 rth->rt_is_input = 1;
2245 RT_CACHE_STAT_INC(in_slow_tot);
2246 if (res->type == RTN_UNREACHABLE) {
2247 rth->dst.input= ip_error;
2248 rth->dst.error= -err;
2249 rth->rt_flags &= ~RTCF_LOCAL;
2253 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2255 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2256 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2257 WARN_ON(rth->dst.input == lwtunnel_input);
2258 rth->dst.lwtstate->orig_input = rth->dst.input;
2259 rth->dst.input = lwtunnel_input;
2262 if (unlikely(!rt_cache_route(nhc, rth)))
2263 rt_add_uncached_list(rth);
2265 skb_dst_set(skb, &rth->dst);
2270 RT_CACHE_STAT_INC(in_no_route);
2271 res->type = RTN_UNREACHABLE;
2277 * Do not cache martian addresses: they should be logged (RFC1812)
2279 martian_destination:
2280 RT_CACHE_STAT_INC(in_martian_dst);
2281 #ifdef CONFIG_IP_ROUTE_VERBOSE
2282 if (IN_DEV_LOG_MARTIANS(in_dev))
2283 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2284 &daddr, &saddr, dev->name);
2296 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2300 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2301 u8 tos, struct net_device *dev)
2303 struct fib_result res;
2306 tos &= IPTOS_RT_MASK;
2308 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2313 EXPORT_SYMBOL(ip_route_input_noref);
2315 /* called with rcu_read_lock held */
2316 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2317 u8 tos, struct net_device *dev, struct fib_result *res)
2319 /* Multicast recognition logic is moved from route cache to here.
2320 The problem was that too many Ethernet cards have broken/missing
2321 hardware multicast filters :-( As result the host on multicasting
2322 network acquires a lot of useless route cache entries, sort of
2323 SDR messages from all the world. Now we try to get rid of them.
2324 Really, provided software IP multicast filter is organized
2325 reasonably (at least, hashed), it does not result in a slowdown
2326 comparing with route cache reject entries.
2327 Note, that multicast routers are not affected, because
2328 route cache entry is created eventually.
2330 if (ipv4_is_multicast(daddr)) {
2331 struct in_device *in_dev = __in_dev_get_rcu(dev);
2337 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2338 ip_hdr(skb)->protocol);
2340 /* check l3 master if no match yet */
2341 if (!our && netif_is_l3_slave(dev)) {
2342 struct in_device *l3_in_dev;
2344 l3_in_dev = __in_dev_get_rcu(skb->dev);
2346 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2347 ip_hdr(skb)->protocol);
2351 #ifdef CONFIG_IP_MROUTE
2353 (!ipv4_is_local_multicast(daddr) &&
2354 IN_DEV_MFORWARD(in_dev))
2357 err = ip_route_input_mc(skb, daddr, saddr,
2363 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2366 /* called with rcu_read_lock() */
2367 static struct rtable *__mkroute_output(const struct fib_result *res,
2368 const struct flowi4 *fl4, int orig_oif,
2369 struct net_device *dev_out,
2372 struct fib_info *fi = res->fi;
2373 struct fib_nh_exception *fnhe;
2374 struct in_device *in_dev;
2375 u16 type = res->type;
2379 in_dev = __in_dev_get_rcu(dev_out);
2381 return ERR_PTR(-EINVAL);
2383 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2384 if (ipv4_is_loopback(fl4->saddr) &&
2385 !(dev_out->flags & IFF_LOOPBACK) &&
2386 !netif_is_l3_master(dev_out))
2387 return ERR_PTR(-EINVAL);
2389 if (ipv4_is_lbcast(fl4->daddr))
2390 type = RTN_BROADCAST;
2391 else if (ipv4_is_multicast(fl4->daddr))
2392 type = RTN_MULTICAST;
2393 else if (ipv4_is_zeronet(fl4->daddr))
2394 return ERR_PTR(-EINVAL);
2396 if (dev_out->flags & IFF_LOOPBACK)
2397 flags |= RTCF_LOCAL;
2400 if (type == RTN_BROADCAST) {
2401 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2403 } else if (type == RTN_MULTICAST) {
2404 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2405 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2407 flags &= ~RTCF_LOCAL;
2410 /* If multicast route do not exist use
2411 * default one, but do not gateway in this case.
2414 if (fi && res->prefixlen < 4)
2416 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2417 (orig_oif != dev_out->ifindex)) {
2418 /* For local routes that require a particular output interface
2419 * we do not want to cache the result. Caching the result
2420 * causes incorrect behaviour when there are multiple source
2421 * addresses on the interface, the end result being that if the
2422 * intended recipient is waiting on that interface for the
2423 * packet he won't receive it because it will be delivered on
2424 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2425 * be set to the loopback interface as well.
2431 do_cache &= fi != NULL;
2433 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2434 struct rtable __rcu **prth;
2436 fnhe = find_exception(nhc, fl4->daddr);
2440 prth = &fnhe->fnhe_rth_output;
2442 if (unlikely(fl4->flowi4_flags &
2443 FLOWI_FLAG_KNOWN_NH &&
2444 !(nhc->nhc_gw_family &&
2445 nhc->nhc_scope == RT_SCOPE_LINK))) {
2449 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2451 rth = rcu_dereference(*prth);
2452 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2457 rth = rt_dst_alloc(dev_out, flags, type,
2458 IN_DEV_ORCONF(in_dev, NOPOLICY),
2459 IN_DEV_ORCONF(in_dev, NOXFRM));
2461 return ERR_PTR(-ENOBUFS);
2463 rth->rt_iif = orig_oif;
2465 RT_CACHE_STAT_INC(out_slow_tot);
2467 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2468 if (flags & RTCF_LOCAL &&
2469 !(dev_out->flags & IFF_LOOPBACK)) {
2470 rth->dst.output = ip_mc_output;
2471 RT_CACHE_STAT_INC(out_slow_mc);
2473 #ifdef CONFIG_IP_MROUTE
2474 if (type == RTN_MULTICAST) {
2475 if (IN_DEV_MFORWARD(in_dev) &&
2476 !ipv4_is_local_multicast(fl4->daddr)) {
2477 rth->dst.input = ip_mr_input;
2478 rth->dst.output = ip_mc_output;
2484 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2485 lwtunnel_set_redirect(&rth->dst);
2491 * Major route resolver routine.
2494 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2495 const struct sk_buff *skb)
2497 __u8 tos = RT_FL_TOS(fl4);
2498 struct fib_result res = {
2506 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2507 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2508 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2509 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2512 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2517 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2519 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2520 struct fib_result *res,
2521 const struct sk_buff *skb)
2523 struct net_device *dev_out = NULL;
2524 int orig_oif = fl4->flowi4_oif;
2525 unsigned int flags = 0;
2530 if (ipv4_is_multicast(fl4->saddr) ||
2531 ipv4_is_lbcast(fl4->saddr) ||
2532 ipv4_is_zeronet(fl4->saddr)) {
2533 rth = ERR_PTR(-EINVAL);
2537 rth = ERR_PTR(-ENETUNREACH);
2539 /* I removed check for oif == dev_out->oif here.
2540 It was wrong for two reasons:
2541 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2542 is assigned to multiple interfaces.
2543 2. Moreover, we are allowed to send packets with saddr
2544 of another iface. --ANK
2547 if (fl4->flowi4_oif == 0 &&
2548 (ipv4_is_multicast(fl4->daddr) ||
2549 ipv4_is_lbcast(fl4->daddr))) {
2550 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2551 dev_out = __ip_dev_find(net, fl4->saddr, false);
2555 /* Special hack: user can direct multicasts
2556 and limited broadcast via necessary interface
2557 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2558 This hack is not just for fun, it allows
2559 vic,vat and friends to work.
2560 They bind socket to loopback, set ttl to zero
2561 and expect that it will work.
2562 From the viewpoint of routing cache they are broken,
2563 because we are not allowed to build multicast path
2564 with loopback source addr (look, routing cache
2565 cannot know, that ttl is zero, so that packet
2566 will not leave this host and route is valid).
2567 Luckily, this hack is good workaround.
2570 fl4->flowi4_oif = dev_out->ifindex;
2574 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2575 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2576 if (!__ip_dev_find(net, fl4->saddr, false))
2582 if (fl4->flowi4_oif) {
2583 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2584 rth = ERR_PTR(-ENODEV);
2588 /* RACE: Check return value of inet_select_addr instead. */
2589 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2590 rth = ERR_PTR(-ENETUNREACH);
2593 if (ipv4_is_local_multicast(fl4->daddr) ||
2594 ipv4_is_lbcast(fl4->daddr) ||
2595 fl4->flowi4_proto == IPPROTO_IGMP) {
2597 fl4->saddr = inet_select_addr(dev_out, 0,
2602 if (ipv4_is_multicast(fl4->daddr))
2603 fl4->saddr = inet_select_addr(dev_out, 0,
2605 else if (!fl4->daddr)
2606 fl4->saddr = inet_select_addr(dev_out, 0,
2612 fl4->daddr = fl4->saddr;
2614 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2615 dev_out = net->loopback_dev;
2616 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2617 res->type = RTN_LOCAL;
2618 flags |= RTCF_LOCAL;
2622 err = fib_lookup(net, fl4, res, 0);
2626 if (fl4->flowi4_oif &&
2627 (ipv4_is_multicast(fl4->daddr) ||
2628 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2629 /* Apparently, routing tables are wrong. Assume,
2630 that the destination is on link.
2633 Because we are allowed to send to iface
2634 even if it has NO routes and NO assigned
2635 addresses. When oif is specified, routing
2636 tables are looked up with only one purpose:
2637 to catch if destination is gatewayed, rather than
2638 direct. Moreover, if MSG_DONTROUTE is set,
2639 we send packet, ignoring both routing tables
2640 and ifaddr state. --ANK
2643 We could make it even if oif is unknown,
2644 likely IPv6, but we do not.
2647 if (fl4->saddr == 0)
2648 fl4->saddr = inet_select_addr(dev_out, 0,
2650 res->type = RTN_UNICAST;
2657 if (res->type == RTN_LOCAL) {
2659 if (res->fi->fib_prefsrc)
2660 fl4->saddr = res->fi->fib_prefsrc;
2662 fl4->saddr = fl4->daddr;
2665 /* L3 master device is the loopback for that domain */
2666 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2669 /* make sure orig_oif points to fib result device even
2670 * though packet rx/tx happens over loopback or l3mdev
2672 orig_oif = FIB_RES_OIF(*res);
2674 fl4->flowi4_oif = dev_out->ifindex;
2675 flags |= RTCF_LOCAL;
2679 fib_select_path(net, res, fl4, skb);
2681 dev_out = FIB_RES_DEV(*res);
2684 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2690 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2695 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2697 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2699 return mtu ? : dst->dev->mtu;
2702 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2703 struct sk_buff *skb, u32 mtu,
2708 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2709 struct sk_buff *skb)
2713 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2719 static struct dst_ops ipv4_dst_blackhole_ops = {
2721 .check = ipv4_blackhole_dst_check,
2722 .mtu = ipv4_blackhole_mtu,
2723 .default_advmss = ipv4_default_advmss,
2724 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2725 .redirect = ipv4_rt_blackhole_redirect,
2726 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2727 .neigh_lookup = ipv4_neigh_lookup,
2730 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2732 struct rtable *ort = (struct rtable *) dst_orig;
2735 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2737 struct dst_entry *new = &rt->dst;
2740 new->input = dst_discard;
2741 new->output = dst_discard_out;
2743 new->dev = net->loopback_dev;
2747 rt->rt_is_input = ort->rt_is_input;
2748 rt->rt_iif = ort->rt_iif;
2749 rt->rt_pmtu = ort->rt_pmtu;
2750 rt->rt_mtu_locked = ort->rt_mtu_locked;
2752 rt->rt_genid = rt_genid_ipv4(net);
2753 rt->rt_flags = ort->rt_flags;
2754 rt->rt_type = ort->rt_type;
2755 rt->rt_uses_gateway = ort->rt_uses_gateway;
2756 rt->rt_gw_family = ort->rt_gw_family;
2757 if (rt->rt_gw_family == AF_INET)
2758 rt->rt_gw4 = ort->rt_gw4;
2759 else if (rt->rt_gw_family == AF_INET6)
2760 rt->rt_gw6 = ort->rt_gw6;
2762 INIT_LIST_HEAD(&rt->rt_uncached);
2765 dst_release(dst_orig);
2767 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2770 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2771 const struct sock *sk)
2773 struct rtable *rt = __ip_route_output_key(net, flp4);
2778 if (flp4->flowi4_proto) {
2779 flp4->flowi4_oif = rt->dst.dev->ifindex;
2780 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2781 flowi4_to_flowi(flp4),
2787 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2789 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2790 struct net_device *dev,
2791 struct net *net, __be32 *saddr,
2792 const struct ip_tunnel_info *info,
2793 u8 protocol, bool use_cache)
2795 #ifdef CONFIG_DST_CACHE
2796 struct dst_cache *dst_cache;
2798 struct rtable *rt = NULL;
2802 #ifdef CONFIG_DST_CACHE
2803 dst_cache = (struct dst_cache *)&info->dst_cache;
2805 rt = dst_cache_get_ip4(dst_cache, saddr);
2810 memset(&fl4, 0, sizeof(fl4));
2811 fl4.flowi4_mark = skb->mark;
2812 fl4.flowi4_proto = protocol;
2813 fl4.daddr = info->key.u.ipv4.dst;
2814 fl4.saddr = info->key.u.ipv4.src;
2815 tos = info->key.tos;
2816 fl4.flowi4_tos = RT_TOS(tos);
2818 rt = ip_route_output_key(net, &fl4);
2820 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2821 return ERR_PTR(-ENETUNREACH);
2823 if (rt->dst.dev == dev) { /* is this necessary? */
2824 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2826 return ERR_PTR(-ELOOP);
2828 #ifdef CONFIG_DST_CACHE
2830 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2835 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2837 /* called with rcu_read_lock held */
2838 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2839 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2840 struct sk_buff *skb, u32 portid, u32 seq,
2844 struct nlmsghdr *nlh;
2845 unsigned long expires = 0;
2847 u32 metrics[RTAX_MAX];
2849 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2853 r = nlmsg_data(nlh);
2854 r->rtm_family = AF_INET;
2855 r->rtm_dst_len = 32;
2857 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2858 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2859 if (nla_put_u32(skb, RTA_TABLE, table_id))
2860 goto nla_put_failure;
2861 r->rtm_type = rt->rt_type;
2862 r->rtm_scope = RT_SCOPE_UNIVERSE;
2863 r->rtm_protocol = RTPROT_UNSPEC;
2864 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2865 if (rt->rt_flags & RTCF_NOTIFY)
2866 r->rtm_flags |= RTM_F_NOTIFY;
2867 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2868 r->rtm_flags |= RTCF_DOREDIRECT;
2870 if (nla_put_in_addr(skb, RTA_DST, dst))
2871 goto nla_put_failure;
2873 r->rtm_src_len = 32;
2874 if (nla_put_in_addr(skb, RTA_SRC, src))
2875 goto nla_put_failure;
2878 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2879 goto nla_put_failure;
2880 if (rt->dst.lwtstate &&
2881 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2882 goto nla_put_failure;
2883 #ifdef CONFIG_IP_ROUTE_CLASSID
2884 if (rt->dst.tclassid &&
2885 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2886 goto nla_put_failure;
2888 if (fl4 && !rt_is_input_route(rt) &&
2889 fl4->saddr != src) {
2890 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2891 goto nla_put_failure;
2893 if (rt->rt_uses_gateway) {
2894 if (rt->rt_gw_family == AF_INET &&
2895 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2896 goto nla_put_failure;
2897 } else if (rt->rt_gw_family == AF_INET6) {
2898 int alen = sizeof(struct in6_addr);
2902 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2904 goto nla_put_failure;
2906 via = nla_data(nla);
2907 via->rtvia_family = AF_INET6;
2908 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2912 expires = rt->dst.expires;
2914 unsigned long now = jiffies;
2916 if (time_before(now, expires))
2922 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2923 if (rt->rt_pmtu && expires)
2924 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2925 if (rt->rt_mtu_locked && expires)
2926 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2927 if (rtnetlink_put_metrics(skb, metrics) < 0)
2928 goto nla_put_failure;
2931 if (fl4->flowi4_mark &&
2932 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2933 goto nla_put_failure;
2935 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2936 nla_put_u32(skb, RTA_UID,
2937 from_kuid_munged(current_user_ns(),
2939 goto nla_put_failure;
2941 if (rt_is_input_route(rt)) {
2942 #ifdef CONFIG_IP_MROUTE
2943 if (ipv4_is_multicast(dst) &&
2944 !ipv4_is_local_multicast(dst) &&
2945 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2946 int err = ipmr_get_route(net, skb,
2947 fl4->saddr, fl4->daddr,
2953 goto nla_put_failure;
2957 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2958 goto nla_put_failure;
2962 error = rt->dst.error;
2964 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2965 goto nla_put_failure;
2967 nlmsg_end(skb, nlh);
2971 nlmsg_cancel(skb, nlh);
2975 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
2976 struct netlink_callback *cb, u32 table_id,
2977 struct fnhe_hash_bucket *bucket, int genid,
2978 int *fa_index, int fa_start, unsigned int flags)
2982 for (i = 0; i < FNHE_HASH_SIZE; i++) {
2983 struct fib_nh_exception *fnhe;
2985 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
2986 fnhe = rcu_dereference(fnhe->fnhe_next)) {
2990 if (*fa_index < fa_start)
2993 if (fnhe->fnhe_genid != genid)
2996 if (fnhe->fnhe_expires &&
2997 time_after(jiffies, fnhe->fnhe_expires))
3000 rt = rcu_dereference(fnhe->fnhe_rth_input);
3002 rt = rcu_dereference(fnhe->fnhe_rth_output);
3006 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3007 table_id, NULL, skb,
3008 NETLINK_CB(cb->skb).portid,
3009 cb->nlh->nlmsg_seq, flags);
3020 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3021 u32 table_id, struct fib_info *fi,
3022 int *fa_index, int fa_start, unsigned int flags)
3024 struct net *net = sock_net(cb->skb->sk);
3025 int nhsel, genid = fnhe_genid(net);
3027 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3028 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3029 struct fnhe_hash_bucket *bucket;
3032 if (nhc->nhc_flags & RTNH_F_DEAD)
3036 bucket = rcu_dereference(nhc->nhc_exceptions);
3039 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3040 genid, fa_index, fa_start,
3050 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3051 u8 ip_proto, __be16 sport,
3054 struct sk_buff *skb;
3057 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3061 /* Reserve room for dummy headers, this skb can pass
3062 * through good chunk of routing engine.
3064 skb_reset_mac_header(skb);
3065 skb_reset_network_header(skb);
3066 skb->protocol = htons(ETH_P_IP);
3067 iph = skb_put(skb, sizeof(struct iphdr));
3068 iph->protocol = ip_proto;
3074 skb_set_transport_header(skb, skb->len);
3076 switch (iph->protocol) {
3078 struct udphdr *udph;
3080 udph = skb_put_zero(skb, sizeof(struct udphdr));
3081 udph->source = sport;
3083 udph->len = sizeof(struct udphdr);
3088 struct tcphdr *tcph;
3090 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3091 tcph->source = sport;
3093 tcph->doff = sizeof(struct tcphdr) / 4;
3095 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3099 case IPPROTO_ICMP: {
3100 struct icmphdr *icmph;
3102 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3103 icmph->type = ICMP_ECHO;
3111 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3112 const struct nlmsghdr *nlh,
3114 struct netlink_ext_ack *extack)
3119 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3120 NL_SET_ERR_MSG(extack,
3121 "ipv4: Invalid header for route get request");
3125 if (!netlink_strict_get_check(skb))
3126 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3127 rtm_ipv4_policy, extack);
3129 rtm = nlmsg_data(nlh);
3130 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3131 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3132 rtm->rtm_table || rtm->rtm_protocol ||
3133 rtm->rtm_scope || rtm->rtm_type) {
3134 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3138 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3139 RTM_F_LOOKUP_TABLE |
3141 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3145 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3146 rtm_ipv4_policy, extack);
3150 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3151 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3152 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3156 for (i = 0; i <= RTA_MAX; i++) {
3172 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3180 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3181 struct netlink_ext_ack *extack)
3183 struct net *net = sock_net(in_skb->sk);
3184 struct nlattr *tb[RTA_MAX+1];
3185 u32 table_id = RT_TABLE_MAIN;
3186 __be16 sport = 0, dport = 0;
3187 struct fib_result res = {};
3188 u8 ip_proto = IPPROTO_UDP;
3189 struct rtable *rt = NULL;
3190 struct sk_buff *skb;
3192 struct flowi4 fl4 = {};
3200 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3204 rtm = nlmsg_data(nlh);
3205 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3206 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3207 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3208 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3210 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3212 uid = (iif ? INVALID_UID : current_uid());
3214 if (tb[RTA_IP_PROTO]) {
3215 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3216 &ip_proto, AF_INET, extack);
3222 sport = nla_get_be16(tb[RTA_SPORT]);
3225 dport = nla_get_be16(tb[RTA_DPORT]);
3227 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3233 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3234 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3235 fl4.flowi4_mark = mark;
3236 fl4.flowi4_uid = uid;
3238 fl4.fl4_sport = sport;
3240 fl4.fl4_dport = dport;
3241 fl4.flowi4_proto = ip_proto;
3246 struct net_device *dev;
3248 dev = dev_get_by_index_rcu(net, iif);
3254 fl4.flowi4_iif = iif; /* for rt_fill_info */
3257 err = ip_route_input_rcu(skb, dst, src,
3258 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3261 rt = skb_rtable(skb);
3262 if (err == 0 && rt->dst.error)
3263 err = -rt->dst.error;
3265 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3266 skb->dev = net->loopback_dev;
3267 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3272 skb_dst_set(skb, &rt->dst);
3278 if (rtm->rtm_flags & RTM_F_NOTIFY)
3279 rt->rt_flags |= RTCF_NOTIFY;
3281 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3282 table_id = res.table ? res.table->tb_id : 0;
3284 /* reset skb for netlink reply msg */
3286 skb_reset_network_header(skb);
3287 skb_reset_transport_header(skb);
3288 skb_reset_mac_header(skb);
3290 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3291 struct fib_rt_info fri;
3294 err = fib_props[res.type].error;
3296 err = -EHOSTUNREACH;
3300 fri.tb_id = table_id;
3301 fri.dst = res.prefix;
3302 fri.dst_len = res.prefixlen;
3303 fri.tos = fl4.flowi4_tos;
3304 fri.type = rt->rt_type;
3307 fri.offload_failed = 0;
3309 struct fib_alias *fa;
3311 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3312 u8 slen = 32 - fri.dst_len;
3314 if (fa->fa_slen == slen &&
3315 fa->tb_id == fri.tb_id &&
3316 fa->fa_tos == fri.tos &&
3317 fa->fa_info == res.fi &&
3318 fa->fa_type == fri.type) {
3319 fri.offload = fa->offload;
3320 fri.trap = fa->trap;
3325 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3326 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3328 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3329 NETLINK_CB(in_skb).portid,
3337 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3347 void ip_rt_multicast_event(struct in_device *in_dev)
3349 rt_cache_flush(dev_net(in_dev->dev));
3352 #ifdef CONFIG_SYSCTL
3353 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3354 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3355 static int ip_rt_gc_elasticity __read_mostly = 8;
3356 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3358 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3359 void *buffer, size_t *lenp, loff_t *ppos)
3361 struct net *net = (struct net *)__ctl->extra1;
3364 rt_cache_flush(net);
3365 fnhe_genid_bump(net);
3372 static struct ctl_table ipv4_route_table[] = {
3374 .procname = "gc_thresh",
3375 .data = &ipv4_dst_ops.gc_thresh,
3376 .maxlen = sizeof(int),
3378 .proc_handler = proc_dointvec,
3381 .procname = "max_size",
3382 .data = &ip_rt_max_size,
3383 .maxlen = sizeof(int),
3385 .proc_handler = proc_dointvec,
3388 /* Deprecated. Use gc_min_interval_ms */
3390 .procname = "gc_min_interval",
3391 .data = &ip_rt_gc_min_interval,
3392 .maxlen = sizeof(int),
3394 .proc_handler = proc_dointvec_jiffies,
3397 .procname = "gc_min_interval_ms",
3398 .data = &ip_rt_gc_min_interval,
3399 .maxlen = sizeof(int),
3401 .proc_handler = proc_dointvec_ms_jiffies,
3404 .procname = "gc_timeout",
3405 .data = &ip_rt_gc_timeout,
3406 .maxlen = sizeof(int),
3408 .proc_handler = proc_dointvec_jiffies,
3411 .procname = "gc_interval",
3412 .data = &ip_rt_gc_interval,
3413 .maxlen = sizeof(int),
3415 .proc_handler = proc_dointvec_jiffies,
3418 .procname = "redirect_load",
3419 .data = &ip_rt_redirect_load,
3420 .maxlen = sizeof(int),
3422 .proc_handler = proc_dointvec,
3425 .procname = "redirect_number",
3426 .data = &ip_rt_redirect_number,
3427 .maxlen = sizeof(int),
3429 .proc_handler = proc_dointvec,
3432 .procname = "redirect_silence",
3433 .data = &ip_rt_redirect_silence,
3434 .maxlen = sizeof(int),
3436 .proc_handler = proc_dointvec,
3439 .procname = "error_cost",
3440 .data = &ip_rt_error_cost,
3441 .maxlen = sizeof(int),
3443 .proc_handler = proc_dointvec,
3446 .procname = "error_burst",
3447 .data = &ip_rt_error_burst,
3448 .maxlen = sizeof(int),
3450 .proc_handler = proc_dointvec,
3453 .procname = "gc_elasticity",
3454 .data = &ip_rt_gc_elasticity,
3455 .maxlen = sizeof(int),
3457 .proc_handler = proc_dointvec,
3460 .procname = "mtu_expires",
3461 .data = &ip_rt_mtu_expires,
3462 .maxlen = sizeof(int),
3464 .proc_handler = proc_dointvec_jiffies,
3467 .procname = "min_pmtu",
3468 .data = &ip_rt_min_pmtu,
3469 .maxlen = sizeof(int),
3471 .proc_handler = proc_dointvec_minmax,
3472 .extra1 = &ip_min_valid_pmtu,
3475 .procname = "min_adv_mss",
3476 .data = &ip_rt_min_advmss,
3477 .maxlen = sizeof(int),
3479 .proc_handler = proc_dointvec,
3484 static const char ipv4_route_flush_procname[] = "flush";
3486 static struct ctl_table ipv4_route_flush_table[] = {
3488 .procname = ipv4_route_flush_procname,
3489 .maxlen = sizeof(int),
3491 .proc_handler = ipv4_sysctl_rtcache_flush,
3496 static __net_init int sysctl_route_net_init(struct net *net)
3498 struct ctl_table *tbl;
3500 tbl = ipv4_route_flush_table;
3501 if (!net_eq(net, &init_net)) {
3502 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3506 /* Don't export non-whitelisted sysctls to unprivileged users */
3507 if (net->user_ns != &init_user_ns) {
3508 if (tbl[0].procname != ipv4_route_flush_procname)
3509 tbl[0].procname = NULL;
3512 tbl[0].extra1 = net;
3514 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3515 if (!net->ipv4.route_hdr)
3520 if (tbl != ipv4_route_flush_table)
3526 static __net_exit void sysctl_route_net_exit(struct net *net)
3528 struct ctl_table *tbl;
3530 tbl = net->ipv4.route_hdr->ctl_table_arg;
3531 unregister_net_sysctl_table(net->ipv4.route_hdr);
3532 BUG_ON(tbl == ipv4_route_flush_table);
3536 static __net_initdata struct pernet_operations sysctl_route_ops = {
3537 .init = sysctl_route_net_init,
3538 .exit = sysctl_route_net_exit,
3542 static __net_init int rt_genid_init(struct net *net)
3544 atomic_set(&net->ipv4.rt_genid, 0);
3545 atomic_set(&net->fnhe_genid, 0);
3546 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3550 static __net_initdata struct pernet_operations rt_genid_ops = {
3551 .init = rt_genid_init,
3554 static int __net_init ipv4_inetpeer_init(struct net *net)
3556 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3560 inet_peer_base_init(bp);
3561 net->ipv4.peers = bp;
3565 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3567 struct inet_peer_base *bp = net->ipv4.peers;
3569 net->ipv4.peers = NULL;
3570 inetpeer_invalidate_tree(bp);
3574 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3575 .init = ipv4_inetpeer_init,
3576 .exit = ipv4_inetpeer_exit,
3579 #ifdef CONFIG_IP_ROUTE_CLASSID
3580 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3581 #endif /* CONFIG_IP_ROUTE_CLASSID */
3583 int __init ip_rt_init(void)
3587 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3590 panic("IP: failed to allocate ip_idents\n");
3592 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3594 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3596 panic("IP: failed to allocate ip_tstamps\n");
3598 for_each_possible_cpu(cpu) {
3599 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3601 INIT_LIST_HEAD(&ul->head);
3602 spin_lock_init(&ul->lock);
3604 #ifdef CONFIG_IP_ROUTE_CLASSID
3605 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3607 panic("IP: failed to allocate ip_rt_acct\n");
3610 ipv4_dst_ops.kmem_cachep =
3611 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3612 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3614 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3616 if (dst_entries_init(&ipv4_dst_ops) < 0)
3617 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3619 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3620 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3622 ipv4_dst_ops.gc_thresh = ~0;
3623 ip_rt_max_size = INT_MAX;
3628 if (ip_rt_proc_init())
3629 pr_err("Unable to create route proc files\n");
3634 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3635 RTNL_FLAG_DOIT_UNLOCKED);
3637 #ifdef CONFIG_SYSCTL
3638 register_pernet_subsys(&sysctl_route_ops);
3640 register_pernet_subsys(&rt_genid_ops);
3641 register_pernet_subsys(&ipv4_inetpeer_ops);
3645 #ifdef CONFIG_SYSCTL
3647 * We really need to sanitize the damn ipv4 init order, then all
3648 * this nonsense will go away.
3650 void __init ip_static_sysctl_init(void)
3652 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);