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 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
137 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
138 static unsigned int ipv4_mtu(const struct dst_entry *dst);
139 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
140 static void ipv4_link_failure(struct sk_buff *skb);
141 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb, u32 mtu);
143 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
144 struct sk_buff *skb);
145 static void ipv4_dst_destroy(struct dst_entry *dst);
147 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
153 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
156 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
158 static struct dst_ops ipv4_dst_ops = {
160 .check = ipv4_dst_check,
161 .default_advmss = ipv4_default_advmss,
163 .cow_metrics = ipv4_cow_metrics,
164 .destroy = ipv4_dst_destroy,
165 .negative_advice = ipv4_negative_advice,
166 .link_failure = ipv4_link_failure,
167 .update_pmtu = ip_rt_update_pmtu,
168 .redirect = ip_do_redirect,
169 .local_out = __ip_local_out,
170 .neigh_lookup = ipv4_neigh_lookup,
171 .confirm_neigh = ipv4_confirm_neigh,
174 #define ECN_OR_COST(class) TC_PRIO_##class
176 const __u8 ip_tos2prio[16] = {
178 ECN_OR_COST(BESTEFFORT),
180 ECN_OR_COST(BESTEFFORT),
186 ECN_OR_COST(INTERACTIVE),
188 ECN_OR_COST(INTERACTIVE),
189 TC_PRIO_INTERACTIVE_BULK,
190 ECN_OR_COST(INTERACTIVE_BULK),
191 TC_PRIO_INTERACTIVE_BULK,
192 ECN_OR_COST(INTERACTIVE_BULK)
194 EXPORT_SYMBOL(ip_tos2prio);
196 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
197 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
199 #ifdef CONFIG_PROC_FS
200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
204 return SEQ_START_TOKEN;
207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
219 if (v == SEQ_START_TOKEN)
220 seq_printf(seq, "%-127s\n",
221 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
227 static const struct seq_operations rt_cache_seq_ops = {
228 .start = rt_cache_seq_start,
229 .next = rt_cache_seq_next,
230 .stop = rt_cache_seq_stop,
231 .show = rt_cache_seq_show,
234 static int rt_cache_seq_open(struct inode *inode, struct file *file)
236 return seq_open(file, &rt_cache_seq_ops);
239 static const struct file_operations rt_cache_seq_fops = {
240 .open = rt_cache_seq_open,
243 .release = seq_release,
247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
252 return SEQ_START_TOKEN;
254 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
255 if (!cpu_possible(cpu))
258 return &per_cpu(rt_cache_stat, cpu);
263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
267 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
268 if (!cpu_possible(cpu))
271 return &per_cpu(rt_cache_stat, cpu);
277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
284 struct rt_cache_stat *st = v;
286 if (v == SEQ_START_TOKEN) {
287 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");
291 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
292 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
293 dst_entries_get_slow(&ipv4_dst_ops),
306 0, /* st->gc_total */
307 0, /* st->gc_ignored */
308 0, /* st->gc_goal_miss */
309 0, /* st->gc_dst_overflow */
310 0, /* st->in_hlist_search */
311 0 /* st->out_hlist_search */
316 static const struct seq_operations rt_cpu_seq_ops = {
317 .start = rt_cpu_seq_start,
318 .next = rt_cpu_seq_next,
319 .stop = rt_cpu_seq_stop,
320 .show = rt_cpu_seq_show,
324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
326 return seq_open(file, &rt_cpu_seq_ops);
329 static const struct file_operations rt_cpu_seq_fops = {
330 .open = rt_cpu_seq_open,
333 .release = seq_release,
336 #ifdef CONFIG_IP_ROUTE_CLASSID
337 static int rt_acct_proc_show(struct seq_file *m, void *v)
339 struct ip_rt_acct *dst, *src;
342 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
346 for_each_possible_cpu(i) {
347 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
348 for (j = 0; j < 256; j++) {
349 dst[j].o_bytes += src[j].o_bytes;
350 dst[j].o_packets += src[j].o_packets;
351 dst[j].i_bytes += src[j].i_bytes;
352 dst[j].i_packets += src[j].i_packets;
356 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
362 static int __net_init ip_rt_do_proc_init(struct net *net)
364 struct proc_dir_entry *pde;
366 pde = proc_create("rt_cache", 0444, net->proc_net,
371 pde = proc_create("rt_cache", 0444,
372 net->proc_net_stat, &rt_cpu_seq_fops);
376 #ifdef CONFIG_IP_ROUTE_CLASSID
377 pde = proc_create_single("rt_acct", 0, net->proc_net,
384 #ifdef CONFIG_IP_ROUTE_CLASSID
386 remove_proc_entry("rt_cache", net->proc_net_stat);
389 remove_proc_entry("rt_cache", net->proc_net);
394 static void __net_exit ip_rt_do_proc_exit(struct net *net)
396 remove_proc_entry("rt_cache", net->proc_net_stat);
397 remove_proc_entry("rt_cache", net->proc_net);
398 #ifdef CONFIG_IP_ROUTE_CLASSID
399 remove_proc_entry("rt_acct", net->proc_net);
403 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
404 .init = ip_rt_do_proc_init,
405 .exit = ip_rt_do_proc_exit,
408 static int __init ip_rt_proc_init(void)
410 return register_pernet_subsys(&ip_rt_proc_ops);
414 static inline int ip_rt_proc_init(void)
418 #endif /* CONFIG_PROC_FS */
420 static inline bool rt_is_expired(const struct rtable *rth)
422 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
425 void rt_cache_flush(struct net *net)
427 rt_genid_bump_ipv4(net);
430 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
434 const struct rtable *rt = container_of(dst, struct rtable, dst);
435 struct net_device *dev = dst->dev;
440 if (likely(rt->rt_gw_family == AF_INET)) {
441 n = ip_neigh_gw4(dev, rt->rt_gw4);
442 } else if (rt->rt_gw_family == AF_INET6) {
443 n = ip_neigh_gw6(dev, &rt->rt_gw6);
447 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
448 n = ip_neigh_gw4(dev, pkey);
451 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
454 rcu_read_unlock_bh();
459 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
461 const struct rtable *rt = container_of(dst, struct rtable, dst);
462 struct net_device *dev = dst->dev;
463 const __be32 *pkey = daddr;
465 if (rt->rt_gw_family == AF_INET) {
466 pkey = (const __be32 *)&rt->rt_gw4;
467 } else if (rt->rt_gw_family == AF_INET6) {
468 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
471 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
474 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
477 #define IP_IDENTS_SZ 2048u
479 static atomic_t *ip_idents __read_mostly;
480 static u32 *ip_tstamps __read_mostly;
482 /* In order to protect privacy, we add a perturbation to identifiers
483 * if one generator is seldom used. This makes hard for an attacker
484 * to infer how many packets were sent between two points in time.
486 u32 ip_idents_reserve(u32 hash, int segs)
488 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
489 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
490 u32 old = READ_ONCE(*p_tstamp);
491 u32 now = (u32)jiffies;
494 if (old != now && cmpxchg(p_tstamp, old, now) == old)
495 delta = prandom_u32_max(now - old);
497 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
499 old = (u32)atomic_read(p_id);
500 new = old + delta + segs;
501 } while (atomic_cmpxchg(p_id, old, new) != old);
505 EXPORT_SYMBOL(ip_idents_reserve);
507 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
511 /* Note the following code is not safe, but this is okay. */
512 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
513 get_random_bytes(&net->ipv4.ip_id_key,
514 sizeof(net->ipv4.ip_id_key));
516 hash = siphash_3u32((__force u32)iph->daddr,
517 (__force u32)iph->saddr,
519 &net->ipv4.ip_id_key);
520 id = ip_idents_reserve(hash, segs);
523 EXPORT_SYMBOL(__ip_select_ident);
525 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
526 const struct sock *sk,
527 const struct iphdr *iph,
529 u8 prot, u32 mark, int flow_flags)
532 const struct inet_sock *inet = inet_sk(sk);
534 oif = sk->sk_bound_dev_if;
536 tos = RT_CONN_FLAGS(sk);
537 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
539 flowi4_init_output(fl4, oif, mark, tos,
540 RT_SCOPE_UNIVERSE, prot,
542 iph->daddr, iph->saddr, 0, 0,
543 sock_net_uid(net, sk));
546 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
547 const struct sock *sk)
549 const struct net *net = dev_net(skb->dev);
550 const struct iphdr *iph = ip_hdr(skb);
551 int oif = skb->dev->ifindex;
552 u8 tos = RT_TOS(iph->tos);
553 u8 prot = iph->protocol;
554 u32 mark = skb->mark;
556 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
559 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
561 const struct inet_sock *inet = inet_sk(sk);
562 const struct ip_options_rcu *inet_opt;
563 __be32 daddr = inet->inet_daddr;
566 inet_opt = rcu_dereference(inet->inet_opt);
567 if (inet_opt && inet_opt->opt.srr)
568 daddr = inet_opt->opt.faddr;
569 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
570 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
571 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
572 inet_sk_flowi_flags(sk),
573 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
577 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
578 const struct sk_buff *skb)
581 build_skb_flow_key(fl4, skb, sk);
583 build_sk_flow_key(fl4, sk);
586 static DEFINE_SPINLOCK(fnhe_lock);
588 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
592 rt = rcu_dereference(fnhe->fnhe_rth_input);
594 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
595 dst_dev_put(&rt->dst);
596 dst_release(&rt->dst);
598 rt = rcu_dereference(fnhe->fnhe_rth_output);
600 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
601 dst_dev_put(&rt->dst);
602 dst_release(&rt->dst);
606 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
608 struct fib_nh_exception *fnhe, *oldest;
610 oldest = rcu_dereference(hash->chain);
611 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
612 fnhe = rcu_dereference(fnhe->fnhe_next)) {
613 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
616 fnhe_flush_routes(oldest);
620 static inline u32 fnhe_hashfun(__be32 daddr)
622 static u32 fnhe_hashrnd __read_mostly;
625 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
626 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
627 return hash_32(hval, FNHE_HASH_SHIFT);
630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
632 rt->rt_pmtu = fnhe->fnhe_pmtu;
633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 rt->dst.expires = fnhe->fnhe_expires;
637 rt->rt_flags |= RTCF_REDIRECTED;
638 rt->rt_uses_gateway = 1;
639 rt->rt_gw_family = AF_INET;
640 rt->rt_gw4 = fnhe->fnhe_gw;
644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 __be32 gw, u32 pmtu, bool lock,
646 unsigned long expires)
648 struct fnhe_hash_bucket *hash;
649 struct fib_nh_exception *fnhe;
655 genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 hval = fnhe_hashfun(daddr);
658 spin_lock_bh(&fnhe_lock);
660 hash = rcu_dereference(nhc->nhc_exceptions);
662 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
665 rcu_assign_pointer(nhc->nhc_exceptions, hash);
671 for (fnhe = rcu_dereference(hash->chain); fnhe;
672 fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 if (fnhe->fnhe_daddr == daddr)
679 if (fnhe->fnhe_genid != genid)
680 fnhe->fnhe_genid = genid;
684 fnhe->fnhe_pmtu = pmtu;
685 fnhe->fnhe_mtu_locked = lock;
687 fnhe->fnhe_expires = max(1UL, expires);
688 /* Update all cached dsts too */
689 rt = rcu_dereference(fnhe->fnhe_rth_input);
691 fill_route_from_fnhe(rt, fnhe);
692 rt = rcu_dereference(fnhe->fnhe_rth_output);
694 fill_route_from_fnhe(rt, fnhe);
696 if (depth > FNHE_RECLAIM_DEPTH)
697 fnhe = fnhe_oldest(hash);
699 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
703 fnhe->fnhe_next = hash->chain;
704 rcu_assign_pointer(hash->chain, fnhe);
706 fnhe->fnhe_genid = genid;
707 fnhe->fnhe_daddr = daddr;
709 fnhe->fnhe_pmtu = pmtu;
710 fnhe->fnhe_mtu_locked = lock;
711 fnhe->fnhe_expires = max(1UL, expires);
713 /* Exception created; mark the cached routes for the nexthop
714 * stale, so anyone caching it rechecks if this exception
717 rt = rcu_dereference(nhc->nhc_rth_input);
719 rt->dst.obsolete = DST_OBSOLETE_KILL;
721 for_each_possible_cpu(i) {
722 struct rtable __rcu **prt;
723 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
724 rt = rcu_dereference(*prt);
726 rt->dst.obsolete = DST_OBSOLETE_KILL;
730 fnhe->fnhe_stamp = jiffies;
733 spin_unlock_bh(&fnhe_lock);
736 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
739 __be32 new_gw = icmp_hdr(skb)->un.gateway;
740 __be32 old_gw = ip_hdr(skb)->saddr;
741 struct net_device *dev = skb->dev;
742 struct in_device *in_dev;
743 struct fib_result res;
747 switch (icmp_hdr(skb)->code & 7) {
749 case ICMP_REDIR_NETTOS:
750 case ICMP_REDIR_HOST:
751 case ICMP_REDIR_HOSTTOS:
758 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
761 in_dev = __in_dev_get_rcu(dev);
766 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
767 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
768 ipv4_is_zeronet(new_gw))
769 goto reject_redirect;
771 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
772 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
773 goto reject_redirect;
774 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
775 goto reject_redirect;
777 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
778 goto reject_redirect;
781 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
783 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
785 if (!(n->nud_state & NUD_VALID)) {
786 neigh_event_send(n, NULL);
788 if (fib_lookup(net, fl4, &res, 0) == 0) {
789 struct fib_nh_common *nhc = FIB_RES_NHC(res);
791 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
793 jiffies + ip_rt_gc_timeout);
796 rt->dst.obsolete = DST_OBSOLETE_KILL;
797 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
804 #ifdef CONFIG_IP_ROUTE_VERBOSE
805 if (IN_DEV_LOG_MARTIANS(in_dev)) {
806 const struct iphdr *iph = (const struct iphdr *) skb->data;
807 __be32 daddr = iph->daddr;
808 __be32 saddr = iph->saddr;
810 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
811 " Advised path = %pI4 -> %pI4\n",
812 &old_gw, dev->name, &new_gw,
819 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
823 const struct iphdr *iph = (const struct iphdr *) skb->data;
824 struct net *net = dev_net(skb->dev);
825 int oif = skb->dev->ifindex;
826 u8 tos = RT_TOS(iph->tos);
827 u8 prot = iph->protocol;
828 u32 mark = skb->mark;
830 rt = (struct rtable *) dst;
832 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
833 __ip_do_redirect(rt, skb, &fl4, true);
836 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
838 struct rtable *rt = (struct rtable *)dst;
839 struct dst_entry *ret = dst;
842 if (dst->obsolete > 0) {
845 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
856 * 1. The first ip_rt_redirect_number redirects are sent
857 * with exponential backoff, then we stop sending them at all,
858 * assuming that the host ignores our redirects.
859 * 2. If we did not see packets requiring redirects
860 * during ip_rt_redirect_silence, we assume that the host
861 * forgot redirected route and start to send redirects again.
863 * This algorithm is much cheaper and more intelligent than dumb load limiting
866 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
867 * and "frag. need" (breaks PMTU discovery) in icmp.c.
870 void ip_rt_send_redirect(struct sk_buff *skb)
872 struct rtable *rt = skb_rtable(skb);
873 struct in_device *in_dev;
874 struct inet_peer *peer;
880 in_dev = __in_dev_get_rcu(rt->dst.dev);
881 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
885 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
886 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
889 net = dev_net(rt->dst.dev);
890 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
892 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
893 rt_nexthop(rt, ip_hdr(skb)->daddr));
897 /* No redirected packets during ip_rt_redirect_silence;
898 * reset the algorithm.
900 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
901 peer->rate_tokens = 0;
902 peer->n_redirects = 0;
905 /* Too many ignored redirects; do not send anything
906 * set dst.rate_last to the last seen redirected packet.
908 if (peer->n_redirects >= ip_rt_redirect_number) {
909 peer->rate_last = jiffies;
913 /* Check for load limit; set rate_last to the latest sent
916 if (peer->rate_tokens == 0 ||
919 (ip_rt_redirect_load << peer->rate_tokens)))) {
920 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
922 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
923 peer->rate_last = jiffies;
926 #ifdef CONFIG_IP_ROUTE_VERBOSE
928 peer->rate_tokens == ip_rt_redirect_number)
929 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
930 &ip_hdr(skb)->saddr, inet_iif(skb),
931 &ip_hdr(skb)->daddr, &gw);
938 static int ip_error(struct sk_buff *skb)
940 struct rtable *rt = skb_rtable(skb);
941 struct net_device *dev = skb->dev;
942 struct in_device *in_dev;
943 struct inet_peer *peer;
949 if (netif_is_l3_master(skb->dev)) {
950 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
955 in_dev = __in_dev_get_rcu(dev);
957 /* IP on this device is disabled. */
961 net = dev_net(rt->dst.dev);
962 if (!IN_DEV_FORWARD(in_dev)) {
963 switch (rt->dst.error) {
965 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
969 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
975 switch (rt->dst.error) {
980 code = ICMP_HOST_UNREACH;
983 code = ICMP_NET_UNREACH;
984 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
987 code = ICMP_PKT_FILTERED;
991 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
992 l3mdev_master_ifindex(skb->dev), 1);
997 peer->rate_tokens += now - peer->rate_last;
998 if (peer->rate_tokens > ip_rt_error_burst)
999 peer->rate_tokens = ip_rt_error_burst;
1000 peer->rate_last = now;
1001 if (peer->rate_tokens >= ip_rt_error_cost)
1002 peer->rate_tokens -= ip_rt_error_cost;
1008 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1010 out: kfree_skb(skb);
1014 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1016 struct dst_entry *dst = &rt->dst;
1017 u32 old_mtu = ipv4_mtu(dst);
1018 struct fib_result res;
1021 if (ip_mtu_locked(dst))
1027 if (mtu < ip_rt_min_pmtu) {
1029 mtu = min(old_mtu, ip_rt_min_pmtu);
1032 if (rt->rt_pmtu == mtu && !lock &&
1033 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1037 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1038 struct fib_nh_common *nhc = FIB_RES_NHC(res);
1040 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1041 jiffies + ip_rt_mtu_expires);
1046 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1047 struct sk_buff *skb, u32 mtu)
1049 struct rtable *rt = (struct rtable *) dst;
1052 ip_rt_build_flow_key(&fl4, sk, skb);
1053 __ip_rt_update_pmtu(rt, &fl4, mtu);
1056 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1057 int oif, u8 protocol)
1059 const struct iphdr *iph = (const struct iphdr *) skb->data;
1062 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1064 __build_flow_key(net, &fl4, NULL, iph, oif,
1065 RT_TOS(iph->tos), protocol, mark, 0);
1066 rt = __ip_route_output_key(net, &fl4);
1068 __ip_rt_update_pmtu(rt, &fl4, mtu);
1072 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1074 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1076 const struct iphdr *iph = (const struct iphdr *) skb->data;
1080 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1082 if (!fl4.flowi4_mark)
1083 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1085 rt = __ip_route_output_key(sock_net(sk), &fl4);
1087 __ip_rt_update_pmtu(rt, &fl4, mtu);
1092 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1094 const struct iphdr *iph = (const struct iphdr *) skb->data;
1097 struct dst_entry *odst = NULL;
1099 struct net *net = sock_net(sk);
1103 if (!ip_sk_accept_pmtu(sk))
1106 odst = sk_dst_get(sk);
1108 if (sock_owned_by_user(sk) || !odst) {
1109 __ipv4_sk_update_pmtu(skb, sk, mtu);
1113 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1115 rt = (struct rtable *)odst;
1116 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1117 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1124 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1126 if (!dst_check(&rt->dst, 0)) {
1128 dst_release(&rt->dst);
1130 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1138 sk_dst_set(sk, &rt->dst);
1144 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1146 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1147 int oif, u8 protocol)
1149 const struct iphdr *iph = (const struct iphdr *) skb->data;
1153 __build_flow_key(net, &fl4, NULL, iph, oif,
1154 RT_TOS(iph->tos), protocol, 0, 0);
1155 rt = __ip_route_output_key(net, &fl4);
1157 __ip_do_redirect(rt, skb, &fl4, false);
1161 EXPORT_SYMBOL_GPL(ipv4_redirect);
1163 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1165 const struct iphdr *iph = (const struct iphdr *) skb->data;
1168 struct net *net = sock_net(sk);
1170 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1171 rt = __ip_route_output_key(net, &fl4);
1173 __ip_do_redirect(rt, skb, &fl4, false);
1177 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1179 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1181 struct rtable *rt = (struct rtable *) dst;
1183 /* All IPV4 dsts are created with ->obsolete set to the value
1184 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1185 * into this function always.
1187 * When a PMTU/redirect information update invalidates a route,
1188 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1189 * DST_OBSOLETE_DEAD.
1191 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1196 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1198 struct ip_options opt;
1201 /* Recompile ip options since IPCB may not be valid anymore.
1202 * Also check we have a reasonable ipv4 header.
1204 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1205 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1208 memset(&opt, 0, sizeof(opt));
1209 if (ip_hdr(skb)->ihl > 5) {
1210 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1212 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1215 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1221 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1224 static void ipv4_link_failure(struct sk_buff *skb)
1228 ipv4_send_dest_unreach(skb);
1230 rt = skb_rtable(skb);
1232 dst_set_expires(&rt->dst, 0);
1235 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1237 pr_debug("%s: %pI4 -> %pI4, %s\n",
1238 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1239 skb->dev ? skb->dev->name : "?");
1246 We do not cache source address of outgoing interface,
1247 because it is used only by IP RR, TS and SRR options,
1248 so that it out of fast path.
1250 BTW remember: "addr" is allowed to be not aligned
1254 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1258 if (rt_is_output_route(rt))
1259 src = ip_hdr(skb)->saddr;
1261 struct fib_result res;
1262 struct iphdr *iph = ip_hdr(skb);
1263 struct flowi4 fl4 = {
1264 .daddr = iph->daddr,
1265 .saddr = iph->saddr,
1266 .flowi4_tos = RT_TOS(iph->tos),
1267 .flowi4_oif = rt->dst.dev->ifindex,
1268 .flowi4_iif = skb->dev->ifindex,
1269 .flowi4_mark = skb->mark,
1273 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1274 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1276 src = inet_select_addr(rt->dst.dev,
1277 rt_nexthop(rt, iph->daddr),
1281 memcpy(addr, &src, 4);
1284 #ifdef CONFIG_IP_ROUTE_CLASSID
1285 static void set_class_tag(struct rtable *rt, u32 tag)
1287 if (!(rt->dst.tclassid & 0xFFFF))
1288 rt->dst.tclassid |= tag & 0xFFFF;
1289 if (!(rt->dst.tclassid & 0xFFFF0000))
1290 rt->dst.tclassid |= tag & 0xFFFF0000;
1294 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1296 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1297 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1300 return min(advmss, IPV4_MAX_PMTU - header_size);
1303 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1305 const struct rtable *rt = (const struct rtable *) dst;
1306 unsigned int mtu = rt->rt_pmtu;
1308 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1309 mtu = dst_metric_raw(dst, RTAX_MTU);
1314 mtu = READ_ONCE(dst->dev->mtu);
1316 if (unlikely(ip_mtu_locked(dst))) {
1317 if (rt->rt_uses_gateway && mtu > 576)
1321 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1323 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1326 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1328 struct fnhe_hash_bucket *hash;
1329 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1330 u32 hval = fnhe_hashfun(daddr);
1332 spin_lock_bh(&fnhe_lock);
1334 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1335 lockdep_is_held(&fnhe_lock));
1338 fnhe_p = &hash->chain;
1339 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1341 if (fnhe->fnhe_daddr == daddr) {
1342 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1343 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1344 /* set fnhe_daddr to 0 to ensure it won't bind with
1345 * new dsts in rt_bind_exception().
1347 fnhe->fnhe_daddr = 0;
1348 fnhe_flush_routes(fnhe);
1349 kfree_rcu(fnhe, rcu);
1352 fnhe_p = &fnhe->fnhe_next;
1353 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1354 lockdep_is_held(&fnhe_lock));
1357 spin_unlock_bh(&fnhe_lock);
1360 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1363 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1364 struct fib_nh_exception *fnhe;
1370 hval = fnhe_hashfun(daddr);
1372 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1373 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1374 if (fnhe->fnhe_daddr == daddr) {
1375 if (fnhe->fnhe_expires &&
1376 time_after(jiffies, fnhe->fnhe_expires)) {
1377 ip_del_fnhe(nhc, daddr);
1387 * 1. mtu on route is locked - use it
1388 * 2. mtu from nexthop exception
1389 * 3. mtu from egress device
1392 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1394 struct fib_nh_common *nhc = res->nhc;
1395 struct net_device *dev = nhc->nhc_dev;
1396 struct fib_info *fi = res->fi;
1399 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1400 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1404 struct fib_nh_exception *fnhe;
1406 fnhe = find_exception(nhc, daddr);
1407 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1408 mtu = fnhe->fnhe_pmtu;
1412 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1414 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1417 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1418 __be32 daddr, const bool do_cache)
1422 spin_lock_bh(&fnhe_lock);
1424 if (daddr == fnhe->fnhe_daddr) {
1425 struct rtable __rcu **porig;
1426 struct rtable *orig;
1427 int genid = fnhe_genid(dev_net(rt->dst.dev));
1429 if (rt_is_input_route(rt))
1430 porig = &fnhe->fnhe_rth_input;
1432 porig = &fnhe->fnhe_rth_output;
1433 orig = rcu_dereference(*porig);
1435 if (fnhe->fnhe_genid != genid) {
1436 fnhe->fnhe_genid = genid;
1438 fnhe->fnhe_pmtu = 0;
1439 fnhe->fnhe_expires = 0;
1440 fnhe->fnhe_mtu_locked = false;
1441 fnhe_flush_routes(fnhe);
1444 fill_route_from_fnhe(rt, fnhe);
1447 rt->rt_gw_family = AF_INET;
1452 rcu_assign_pointer(*porig, rt);
1454 dst_dev_put(&orig->dst);
1455 dst_release(&orig->dst);
1460 fnhe->fnhe_stamp = jiffies;
1462 spin_unlock_bh(&fnhe_lock);
1467 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1469 struct rtable *orig, *prev, **p;
1472 if (rt_is_input_route(rt)) {
1473 p = (struct rtable **)&nhc->nhc_rth_input;
1475 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1479 /* hold dst before doing cmpxchg() to avoid race condition
1483 prev = cmpxchg(p, orig, rt);
1486 dst_dev_put(&orig->dst);
1487 dst_release(&orig->dst);
1490 dst_release(&rt->dst);
1497 struct uncached_list {
1499 struct list_head head;
1502 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1504 void rt_add_uncached_list(struct rtable *rt)
1506 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1508 rt->rt_uncached_list = ul;
1510 spin_lock_bh(&ul->lock);
1511 list_add_tail(&rt->rt_uncached, &ul->head);
1512 spin_unlock_bh(&ul->lock);
1515 void rt_del_uncached_list(struct rtable *rt)
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 static void ipv4_dst_destroy(struct dst_entry *dst)
1528 struct rtable *rt = (struct rtable *)dst;
1530 ip_dst_metrics_put(dst);
1531 rt_del_uncached_list(rt);
1534 void rt_flush_dev(struct net_device *dev)
1539 for_each_possible_cpu(cpu) {
1540 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1542 spin_lock_bh(&ul->lock);
1543 list_for_each_entry(rt, &ul->head, rt_uncached) {
1544 if (rt->dst.dev != dev)
1546 rt->dst.dev = blackhole_netdev;
1547 dev_hold(rt->dst.dev);
1550 spin_unlock_bh(&ul->lock);
1554 static bool rt_cache_valid(const struct rtable *rt)
1557 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1561 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1562 const struct fib_result *res,
1563 struct fib_nh_exception *fnhe,
1564 struct fib_info *fi, u16 type, u32 itag,
1565 const bool do_cache)
1567 bool cached = false;
1570 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1572 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1573 rt->rt_uses_gateway = 1;
1574 rt->rt_gw_family = nhc->nhc_gw_family;
1575 /* only INET and INET6 are supported */
1576 if (likely(nhc->nhc_gw_family == AF_INET))
1577 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1579 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1582 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1584 #ifdef CONFIG_IP_ROUTE_CLASSID
1585 if (nhc->nhc_family == AF_INET) {
1588 nh = container_of(nhc, struct fib_nh, nh_common);
1589 rt->dst.tclassid = nh->nh_tclassid;
1592 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1594 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1596 cached = rt_cache_route(nhc, rt);
1597 if (unlikely(!cached)) {
1598 /* Routes we intend to cache in nexthop exception or
1599 * FIB nexthop have the DST_NOCACHE bit clear.
1600 * However, if we are unsuccessful at storing this
1601 * route into the cache we really need to set it.
1604 rt->rt_gw_family = AF_INET;
1607 rt_add_uncached_list(rt);
1610 rt_add_uncached_list(rt);
1612 #ifdef CONFIG_IP_ROUTE_CLASSID
1613 #ifdef CONFIG_IP_MULTIPLE_TABLES
1614 set_class_tag(rt, res->tclassid);
1616 set_class_tag(rt, itag);
1620 struct rtable *rt_dst_alloc(struct net_device *dev,
1621 unsigned int flags, u16 type,
1622 bool nopolicy, bool noxfrm, bool will_cache)
1626 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1627 (will_cache ? 0 : DST_HOST) |
1628 (nopolicy ? DST_NOPOLICY : 0) |
1629 (noxfrm ? DST_NOXFRM : 0));
1632 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1633 rt->rt_flags = flags;
1635 rt->rt_is_input = 0;
1638 rt->rt_mtu_locked = 0;
1639 rt->rt_uses_gateway = 0;
1640 rt->rt_gw_family = 0;
1642 INIT_LIST_HEAD(&rt->rt_uncached);
1644 rt->dst.output = ip_output;
1645 if (flags & RTCF_LOCAL)
1646 rt->dst.input = ip_local_deliver;
1651 EXPORT_SYMBOL(rt_dst_alloc);
1653 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1655 struct rtable *new_rt;
1657 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1661 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1662 new_rt->rt_flags = rt->rt_flags;
1663 new_rt->rt_type = rt->rt_type;
1664 new_rt->rt_is_input = rt->rt_is_input;
1665 new_rt->rt_iif = rt->rt_iif;
1666 new_rt->rt_pmtu = rt->rt_pmtu;
1667 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1668 new_rt->rt_gw_family = rt->rt_gw_family;
1669 if (rt->rt_gw_family == AF_INET)
1670 new_rt->rt_gw4 = rt->rt_gw4;
1671 else if (rt->rt_gw_family == AF_INET6)
1672 new_rt->rt_gw6 = rt->rt_gw6;
1673 INIT_LIST_HEAD(&new_rt->rt_uncached);
1675 new_rt->dst.flags |= DST_HOST;
1676 new_rt->dst.input = rt->dst.input;
1677 new_rt->dst.output = rt->dst.output;
1678 new_rt->dst.error = rt->dst.error;
1679 new_rt->dst.lastuse = jiffies;
1680 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1684 EXPORT_SYMBOL(rt_dst_clone);
1686 /* called in rcu_read_lock() section */
1687 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1688 u8 tos, struct net_device *dev,
1689 struct in_device *in_dev, u32 *itag)
1693 /* Primary sanity checks. */
1697 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1698 skb->protocol != htons(ETH_P_IP))
1701 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1704 if (ipv4_is_zeronet(saddr)) {
1705 if (!ipv4_is_local_multicast(daddr) &&
1706 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1709 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1717 /* called in rcu_read_lock() section */
1718 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1719 u8 tos, struct net_device *dev, int our)
1721 struct in_device *in_dev = __in_dev_get_rcu(dev);
1722 unsigned int flags = RTCF_MULTICAST;
1727 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1732 flags |= RTCF_LOCAL;
1734 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1735 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1739 #ifdef CONFIG_IP_ROUTE_CLASSID
1740 rth->dst.tclassid = itag;
1742 rth->dst.output = ip_rt_bug;
1743 rth->rt_is_input= 1;
1745 #ifdef CONFIG_IP_MROUTE
1746 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1747 rth->dst.input = ip_mr_input;
1749 RT_CACHE_STAT_INC(in_slow_mc);
1751 skb_dst_set(skb, &rth->dst);
1756 static void ip_handle_martian_source(struct net_device *dev,
1757 struct in_device *in_dev,
1758 struct sk_buff *skb,
1762 RT_CACHE_STAT_INC(in_martian_src);
1763 #ifdef CONFIG_IP_ROUTE_VERBOSE
1764 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1766 * RFC1812 recommendation, if source is martian,
1767 * the only hint is MAC header.
1769 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1770 &daddr, &saddr, dev->name);
1771 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1772 print_hex_dump(KERN_WARNING, "ll header: ",
1773 DUMP_PREFIX_OFFSET, 16, 1,
1774 skb_mac_header(skb),
1775 dev->hard_header_len, false);
1781 /* called in rcu_read_lock() section */
1782 static int __mkroute_input(struct sk_buff *skb,
1783 const struct fib_result *res,
1784 struct in_device *in_dev,
1785 __be32 daddr, __be32 saddr, u32 tos)
1787 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1788 struct net_device *dev = nhc->nhc_dev;
1789 struct fib_nh_exception *fnhe;
1792 struct in_device *out_dev;
1796 /* get a working reference to the output device */
1797 out_dev = __in_dev_get_rcu(dev);
1799 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1803 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1804 in_dev->dev, in_dev, &itag);
1806 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1812 do_cache = res->fi && !itag;
1813 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1814 skb->protocol == htons(ETH_P_IP)) {
1817 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1818 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1819 inet_addr_onlink(out_dev, saddr, gw))
1820 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1823 if (skb->protocol != htons(ETH_P_IP)) {
1824 /* Not IP (i.e. ARP). Do not create route, if it is
1825 * invalid for proxy arp. DNAT routes are always valid.
1827 * Proxy arp feature have been extended to allow, ARP
1828 * replies back to the same interface, to support
1829 * Private VLAN switch technologies. See arp.c.
1831 if (out_dev == in_dev &&
1832 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1838 fnhe = find_exception(nhc, daddr);
1841 rth = rcu_dereference(fnhe->fnhe_rth_input);
1843 rth = rcu_dereference(nhc->nhc_rth_input);
1844 if (rt_cache_valid(rth)) {
1845 skb_dst_set_noref(skb, &rth->dst);
1850 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1851 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1852 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1858 rth->rt_is_input = 1;
1859 RT_CACHE_STAT_INC(in_slow_tot);
1861 rth->dst.input = ip_forward;
1863 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1865 lwtunnel_set_redirect(&rth->dst);
1866 skb_dst_set(skb, &rth->dst);
1873 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1874 /* To make ICMP packets follow the right flow, the multipath hash is
1875 * calculated from the inner IP addresses.
1877 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1878 struct flow_keys *hash_keys)
1880 const struct iphdr *outer_iph = ip_hdr(skb);
1881 const struct iphdr *key_iph = outer_iph;
1882 const struct iphdr *inner_iph;
1883 const struct icmphdr *icmph;
1884 struct iphdr _inner_iph;
1885 struct icmphdr _icmph;
1887 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1890 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1893 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1898 if (icmph->type != ICMP_DEST_UNREACH &&
1899 icmph->type != ICMP_REDIRECT &&
1900 icmph->type != ICMP_TIME_EXCEEDED &&
1901 icmph->type != ICMP_PARAMETERPROB)
1904 inner_iph = skb_header_pointer(skb,
1905 outer_iph->ihl * 4 + sizeof(_icmph),
1906 sizeof(_inner_iph), &_inner_iph);
1910 key_iph = inner_iph;
1912 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1913 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1916 /* if skb is set it will be used and fl4 can be NULL */
1917 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1918 const struct sk_buff *skb, struct flow_keys *flkeys)
1920 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1921 struct flow_keys hash_keys;
1924 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1926 memset(&hash_keys, 0, sizeof(hash_keys));
1927 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1929 ip_multipath_l3_keys(skb, &hash_keys);
1931 hash_keys.addrs.v4addrs.src = fl4->saddr;
1932 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1936 /* skb is currently provided only when forwarding */
1938 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1939 struct flow_keys keys;
1941 /* short-circuit if we already have L4 hash present */
1943 return skb_get_hash_raw(skb) >> 1;
1945 memset(&hash_keys, 0, sizeof(hash_keys));
1948 skb_flow_dissect_flow_keys(skb, &keys, flag);
1952 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1953 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1954 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1955 hash_keys.ports.src = flkeys->ports.src;
1956 hash_keys.ports.dst = flkeys->ports.dst;
1957 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1959 memset(&hash_keys, 0, sizeof(hash_keys));
1960 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1961 hash_keys.addrs.v4addrs.src = fl4->saddr;
1962 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1963 hash_keys.ports.src = fl4->fl4_sport;
1964 hash_keys.ports.dst = fl4->fl4_dport;
1965 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1969 memset(&hash_keys, 0, sizeof(hash_keys));
1970 /* skb is currently provided only when forwarding */
1972 struct flow_keys keys;
1974 skb_flow_dissect_flow_keys(skb, &keys, 0);
1975 /* Inner can be v4 or v6 */
1976 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1977 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1978 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1979 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1980 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1981 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1982 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1983 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1984 hash_keys.tags.flow_label = keys.tags.flow_label;
1985 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1987 /* Same as case 0 */
1988 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1989 ip_multipath_l3_keys(skb, &hash_keys);
1992 /* Same as case 0 */
1993 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1994 hash_keys.addrs.v4addrs.src = fl4->saddr;
1995 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1999 mhash = flow_hash_from_keys(&hash_keys);
2002 mhash = jhash_2words(mhash, multipath_hash, 0);
2006 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2008 static int ip_mkroute_input(struct sk_buff *skb,
2009 struct fib_result *res,
2010 struct in_device *in_dev,
2011 __be32 daddr, __be32 saddr, u32 tos,
2012 struct flow_keys *hkeys)
2014 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2015 if (res->fi && fib_info_num_path(res->fi) > 1) {
2016 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2018 fib_select_multipath(res, h);
2022 /* create a routing cache entry */
2023 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2027 * NOTE. We drop all the packets that has local source
2028 * addresses, because every properly looped back packet
2029 * must have correct destination already attached by output routine.
2031 * Such approach solves two big problems:
2032 * 1. Not simplex devices are handled properly.
2033 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2034 * called with rcu_read_lock()
2037 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2038 u8 tos, struct net_device *dev,
2039 struct fib_result *res)
2041 struct in_device *in_dev = __in_dev_get_rcu(dev);
2042 struct flow_keys *flkeys = NULL, _flkeys;
2043 struct net *net = dev_net(dev);
2044 struct ip_tunnel_info *tun_info;
2046 unsigned int flags = 0;
2050 bool do_cache = true;
2052 /* IP on this device is disabled. */
2057 /* Check for the most weird martians, which can be not detected
2061 tun_info = skb_tunnel_info(skb);
2062 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2063 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2065 fl4.flowi4_tun_key.tun_id = 0;
2068 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2069 goto martian_source;
2073 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2076 /* Accept zero addresses only to limited broadcast;
2077 * I even do not know to fix it or not. Waiting for complains :-)
2079 if (ipv4_is_zeronet(saddr))
2080 goto martian_source;
2082 if (ipv4_is_zeronet(daddr))
2083 goto martian_destination;
2085 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2086 * and call it once if daddr or/and saddr are loopback addresses
2088 if (ipv4_is_loopback(daddr)) {
2089 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2090 goto martian_destination;
2091 } else if (ipv4_is_loopback(saddr)) {
2092 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2093 goto martian_source;
2097 * Now we are ready to route packet.
2100 fl4.flowi4_iif = dev->ifindex;
2101 fl4.flowi4_mark = skb->mark;
2102 fl4.flowi4_tos = tos;
2103 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2104 fl4.flowi4_flags = 0;
2107 fl4.flowi4_uid = sock_net_uid(net, NULL);
2109 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2112 fl4.flowi4_proto = 0;
2117 err = fib_lookup(net, &fl4, res, 0);
2119 if (!IN_DEV_FORWARD(in_dev))
2120 err = -EHOSTUNREACH;
2124 if (res->type == RTN_BROADCAST) {
2125 if (IN_DEV_BFORWARD(in_dev))
2127 /* not do cache if bc_forwarding is enabled */
2128 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2133 if (res->type == RTN_LOCAL) {
2134 err = fib_validate_source(skb, saddr, daddr, tos,
2135 0, dev, in_dev, &itag);
2137 goto martian_source;
2141 if (!IN_DEV_FORWARD(in_dev)) {
2142 err = -EHOSTUNREACH;
2145 if (res->type != RTN_UNICAST)
2146 goto martian_destination;
2149 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2153 if (skb->protocol != htons(ETH_P_IP))
2156 if (!ipv4_is_zeronet(saddr)) {
2157 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2160 goto martian_source;
2162 flags |= RTCF_BROADCAST;
2163 res->type = RTN_BROADCAST;
2164 RT_CACHE_STAT_INC(in_brd);
2167 do_cache &= res->fi && !itag;
2169 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2171 rth = rcu_dereference(nhc->nhc_rth_input);
2172 if (rt_cache_valid(rth)) {
2173 skb_dst_set_noref(skb, &rth->dst);
2179 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2180 flags | RTCF_LOCAL, res->type,
2181 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2185 rth->dst.output= ip_rt_bug;
2186 #ifdef CONFIG_IP_ROUTE_CLASSID
2187 rth->dst.tclassid = itag;
2189 rth->rt_is_input = 1;
2191 RT_CACHE_STAT_INC(in_slow_tot);
2192 if (res->type == RTN_UNREACHABLE) {
2193 rth->dst.input= ip_error;
2194 rth->dst.error= -err;
2195 rth->rt_flags &= ~RTCF_LOCAL;
2199 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2201 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2202 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2203 WARN_ON(rth->dst.input == lwtunnel_input);
2204 rth->dst.lwtstate->orig_input = rth->dst.input;
2205 rth->dst.input = lwtunnel_input;
2208 if (unlikely(!rt_cache_route(nhc, rth)))
2209 rt_add_uncached_list(rth);
2211 skb_dst_set(skb, &rth->dst);
2216 RT_CACHE_STAT_INC(in_no_route);
2217 res->type = RTN_UNREACHABLE;
2223 * Do not cache martian addresses: they should be logged (RFC1812)
2225 martian_destination:
2226 RT_CACHE_STAT_INC(in_martian_dst);
2227 #ifdef CONFIG_IP_ROUTE_VERBOSE
2228 if (IN_DEV_LOG_MARTIANS(in_dev))
2229 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2230 &daddr, &saddr, dev->name);
2242 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2246 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2247 u8 tos, struct net_device *dev)
2249 struct fib_result res;
2252 tos &= IPTOS_RT_MASK;
2254 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2259 EXPORT_SYMBOL(ip_route_input_noref);
2261 /* called with rcu_read_lock held */
2262 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2263 u8 tos, struct net_device *dev, struct fib_result *res)
2265 /* Multicast recognition logic is moved from route cache to here.
2266 The problem was that too many Ethernet cards have broken/missing
2267 hardware multicast filters :-( As result the host on multicasting
2268 network acquires a lot of useless route cache entries, sort of
2269 SDR messages from all the world. Now we try to get rid of them.
2270 Really, provided software IP multicast filter is organized
2271 reasonably (at least, hashed), it does not result in a slowdown
2272 comparing with route cache reject entries.
2273 Note, that multicast routers are not affected, because
2274 route cache entry is created eventually.
2276 if (ipv4_is_multicast(daddr)) {
2277 struct in_device *in_dev = __in_dev_get_rcu(dev);
2283 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2284 ip_hdr(skb)->protocol);
2286 /* check l3 master if no match yet */
2287 if (!our && netif_is_l3_slave(dev)) {
2288 struct in_device *l3_in_dev;
2290 l3_in_dev = __in_dev_get_rcu(skb->dev);
2292 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2293 ip_hdr(skb)->protocol);
2297 #ifdef CONFIG_IP_MROUTE
2299 (!ipv4_is_local_multicast(daddr) &&
2300 IN_DEV_MFORWARD(in_dev))
2303 err = ip_route_input_mc(skb, daddr, saddr,
2309 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2312 /* called with rcu_read_lock() */
2313 static struct rtable *__mkroute_output(const struct fib_result *res,
2314 const struct flowi4 *fl4, int orig_oif,
2315 struct net_device *dev_out,
2318 struct fib_info *fi = res->fi;
2319 struct fib_nh_exception *fnhe;
2320 struct in_device *in_dev;
2321 u16 type = res->type;
2325 in_dev = __in_dev_get_rcu(dev_out);
2327 return ERR_PTR(-EINVAL);
2329 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2330 if (ipv4_is_loopback(fl4->saddr) &&
2331 !(dev_out->flags & IFF_LOOPBACK) &&
2332 !netif_is_l3_master(dev_out))
2333 return ERR_PTR(-EINVAL);
2335 if (ipv4_is_lbcast(fl4->daddr))
2336 type = RTN_BROADCAST;
2337 else if (ipv4_is_multicast(fl4->daddr))
2338 type = RTN_MULTICAST;
2339 else if (ipv4_is_zeronet(fl4->daddr))
2340 return ERR_PTR(-EINVAL);
2342 if (dev_out->flags & IFF_LOOPBACK)
2343 flags |= RTCF_LOCAL;
2346 if (type == RTN_BROADCAST) {
2347 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2349 } else if (type == RTN_MULTICAST) {
2350 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2351 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2353 flags &= ~RTCF_LOCAL;
2356 /* If multicast route do not exist use
2357 * default one, but do not gateway in this case.
2360 if (fi && res->prefixlen < 4)
2362 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2363 (orig_oif != dev_out->ifindex)) {
2364 /* For local routes that require a particular output interface
2365 * we do not want to cache the result. Caching the result
2366 * causes incorrect behaviour when there are multiple source
2367 * addresses on the interface, the end result being that if the
2368 * intended recipient is waiting on that interface for the
2369 * packet he won't receive it because it will be delivered on
2370 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2371 * be set to the loopback interface as well.
2377 do_cache &= fi != NULL;
2379 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2380 struct rtable __rcu **prth;
2382 fnhe = find_exception(nhc, fl4->daddr);
2386 prth = &fnhe->fnhe_rth_output;
2388 if (unlikely(fl4->flowi4_flags &
2389 FLOWI_FLAG_KNOWN_NH &&
2390 !(nhc->nhc_gw_family &&
2391 nhc->nhc_scope == RT_SCOPE_LINK))) {
2395 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2397 rth = rcu_dereference(*prth);
2398 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2403 rth = rt_dst_alloc(dev_out, flags, type,
2404 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2405 IN_DEV_CONF_GET(in_dev, NOXFRM),
2408 return ERR_PTR(-ENOBUFS);
2410 rth->rt_iif = orig_oif;
2412 RT_CACHE_STAT_INC(out_slow_tot);
2414 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2415 if (flags & RTCF_LOCAL &&
2416 !(dev_out->flags & IFF_LOOPBACK)) {
2417 rth->dst.output = ip_mc_output;
2418 RT_CACHE_STAT_INC(out_slow_mc);
2420 #ifdef CONFIG_IP_MROUTE
2421 if (type == RTN_MULTICAST) {
2422 if (IN_DEV_MFORWARD(in_dev) &&
2423 !ipv4_is_local_multicast(fl4->daddr)) {
2424 rth->dst.input = ip_mr_input;
2425 rth->dst.output = ip_mc_output;
2431 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2432 lwtunnel_set_redirect(&rth->dst);
2438 * Major route resolver routine.
2441 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2442 const struct sk_buff *skb)
2444 __u8 tos = RT_FL_TOS(fl4);
2445 struct fib_result res = {
2453 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2454 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2455 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2456 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2459 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2464 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2466 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2467 struct fib_result *res,
2468 const struct sk_buff *skb)
2470 struct net_device *dev_out = NULL;
2471 int orig_oif = fl4->flowi4_oif;
2472 unsigned int flags = 0;
2474 int err = -ENETUNREACH;
2477 rth = ERR_PTR(-EINVAL);
2478 if (ipv4_is_multicast(fl4->saddr) ||
2479 ipv4_is_lbcast(fl4->saddr) ||
2480 ipv4_is_zeronet(fl4->saddr))
2483 /* I removed check for oif == dev_out->oif here.
2484 It was wrong for two reasons:
2485 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2486 is assigned to multiple interfaces.
2487 2. Moreover, we are allowed to send packets with saddr
2488 of another iface. --ANK
2491 if (fl4->flowi4_oif == 0 &&
2492 (ipv4_is_multicast(fl4->daddr) ||
2493 ipv4_is_lbcast(fl4->daddr))) {
2494 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2495 dev_out = __ip_dev_find(net, fl4->saddr, false);
2499 /* Special hack: user can direct multicasts
2500 and limited broadcast via necessary interface
2501 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2502 This hack is not just for fun, it allows
2503 vic,vat and friends to work.
2504 They bind socket to loopback, set ttl to zero
2505 and expect that it will work.
2506 From the viewpoint of routing cache they are broken,
2507 because we are not allowed to build multicast path
2508 with loopback source addr (look, routing cache
2509 cannot know, that ttl is zero, so that packet
2510 will not leave this host and route is valid).
2511 Luckily, this hack is good workaround.
2514 fl4->flowi4_oif = dev_out->ifindex;
2518 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2519 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2520 if (!__ip_dev_find(net, fl4->saddr, false))
2526 if (fl4->flowi4_oif) {
2527 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2528 rth = ERR_PTR(-ENODEV);
2532 /* RACE: Check return value of inet_select_addr instead. */
2533 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2534 rth = ERR_PTR(-ENETUNREACH);
2537 if (ipv4_is_local_multicast(fl4->daddr) ||
2538 ipv4_is_lbcast(fl4->daddr) ||
2539 fl4->flowi4_proto == IPPROTO_IGMP) {
2541 fl4->saddr = inet_select_addr(dev_out, 0,
2546 if (ipv4_is_multicast(fl4->daddr))
2547 fl4->saddr = inet_select_addr(dev_out, 0,
2549 else if (!fl4->daddr)
2550 fl4->saddr = inet_select_addr(dev_out, 0,
2556 fl4->daddr = fl4->saddr;
2558 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2559 dev_out = net->loopback_dev;
2560 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2561 res->type = RTN_LOCAL;
2562 flags |= RTCF_LOCAL;
2566 err = fib_lookup(net, fl4, res, 0);
2570 if (fl4->flowi4_oif &&
2571 (ipv4_is_multicast(fl4->daddr) ||
2572 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2573 /* Apparently, routing tables are wrong. Assume,
2574 that the destination is on link.
2577 Because we are allowed to send to iface
2578 even if it has NO routes and NO assigned
2579 addresses. When oif is specified, routing
2580 tables are looked up with only one purpose:
2581 to catch if destination is gatewayed, rather than
2582 direct. Moreover, if MSG_DONTROUTE is set,
2583 we send packet, ignoring both routing tables
2584 and ifaddr state. --ANK
2587 We could make it even if oif is unknown,
2588 likely IPv6, but we do not.
2591 if (fl4->saddr == 0)
2592 fl4->saddr = inet_select_addr(dev_out, 0,
2594 res->type = RTN_UNICAST;
2601 if (res->type == RTN_LOCAL) {
2603 if (res->fi->fib_prefsrc)
2604 fl4->saddr = res->fi->fib_prefsrc;
2606 fl4->saddr = fl4->daddr;
2609 /* L3 master device is the loopback for that domain */
2610 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2613 /* make sure orig_oif points to fib result device even
2614 * though packet rx/tx happens over loopback or l3mdev
2616 orig_oif = FIB_RES_OIF(*res);
2618 fl4->flowi4_oif = dev_out->ifindex;
2619 flags |= RTCF_LOCAL;
2623 fib_select_path(net, res, fl4, skb);
2625 dev_out = FIB_RES_DEV(*res);
2626 fl4->flowi4_oif = dev_out->ifindex;
2630 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2636 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2641 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2643 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2645 return mtu ? : dst->dev->mtu;
2648 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2649 struct sk_buff *skb, u32 mtu)
2653 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2654 struct sk_buff *skb)
2658 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2664 static struct dst_ops ipv4_dst_blackhole_ops = {
2666 .check = ipv4_blackhole_dst_check,
2667 .mtu = ipv4_blackhole_mtu,
2668 .default_advmss = ipv4_default_advmss,
2669 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2670 .redirect = ipv4_rt_blackhole_redirect,
2671 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2672 .neigh_lookup = ipv4_neigh_lookup,
2675 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2677 struct rtable *ort = (struct rtable *) dst_orig;
2680 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2682 struct dst_entry *new = &rt->dst;
2685 new->input = dst_discard;
2686 new->output = dst_discard_out;
2688 new->dev = net->loopback_dev;
2692 rt->rt_is_input = ort->rt_is_input;
2693 rt->rt_iif = ort->rt_iif;
2694 rt->rt_pmtu = ort->rt_pmtu;
2695 rt->rt_mtu_locked = ort->rt_mtu_locked;
2697 rt->rt_genid = rt_genid_ipv4(net);
2698 rt->rt_flags = ort->rt_flags;
2699 rt->rt_type = ort->rt_type;
2700 rt->rt_uses_gateway = ort->rt_uses_gateway;
2701 rt->rt_gw_family = ort->rt_gw_family;
2702 if (rt->rt_gw_family == AF_INET)
2703 rt->rt_gw4 = ort->rt_gw4;
2704 else if (rt->rt_gw_family == AF_INET6)
2705 rt->rt_gw6 = ort->rt_gw6;
2707 INIT_LIST_HEAD(&rt->rt_uncached);
2710 dst_release(dst_orig);
2712 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2715 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2716 const struct sock *sk)
2718 struct rtable *rt = __ip_route_output_key(net, flp4);
2723 if (flp4->flowi4_proto)
2724 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2725 flowi4_to_flowi(flp4),
2730 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2732 /* called with rcu_read_lock held */
2733 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2734 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2735 struct sk_buff *skb, u32 portid, u32 seq,
2739 struct nlmsghdr *nlh;
2740 unsigned long expires = 0;
2742 u32 metrics[RTAX_MAX];
2744 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2748 r = nlmsg_data(nlh);
2749 r->rtm_family = AF_INET;
2750 r->rtm_dst_len = 32;
2752 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2753 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2754 if (nla_put_u32(skb, RTA_TABLE, table_id))
2755 goto nla_put_failure;
2756 r->rtm_type = rt->rt_type;
2757 r->rtm_scope = RT_SCOPE_UNIVERSE;
2758 r->rtm_protocol = RTPROT_UNSPEC;
2759 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2760 if (rt->rt_flags & RTCF_NOTIFY)
2761 r->rtm_flags |= RTM_F_NOTIFY;
2762 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2763 r->rtm_flags |= RTCF_DOREDIRECT;
2765 if (nla_put_in_addr(skb, RTA_DST, dst))
2766 goto nla_put_failure;
2768 r->rtm_src_len = 32;
2769 if (nla_put_in_addr(skb, RTA_SRC, src))
2770 goto nla_put_failure;
2773 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2774 goto nla_put_failure;
2775 #ifdef CONFIG_IP_ROUTE_CLASSID
2776 if (rt->dst.tclassid &&
2777 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2778 goto nla_put_failure;
2780 if (fl4 && !rt_is_input_route(rt) &&
2781 fl4->saddr != src) {
2782 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2783 goto nla_put_failure;
2785 if (rt->rt_uses_gateway) {
2786 if (rt->rt_gw_family == AF_INET &&
2787 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2788 goto nla_put_failure;
2789 } else if (rt->rt_gw_family == AF_INET6) {
2790 int alen = sizeof(struct in6_addr);
2794 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2796 goto nla_put_failure;
2798 via = nla_data(nla);
2799 via->rtvia_family = AF_INET6;
2800 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2804 expires = rt->dst.expires;
2806 unsigned long now = jiffies;
2808 if (time_before(now, expires))
2814 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2815 if (rt->rt_pmtu && expires)
2816 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2817 if (rt->rt_mtu_locked && expires)
2818 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2819 if (rtnetlink_put_metrics(skb, metrics) < 0)
2820 goto nla_put_failure;
2823 if (fl4->flowi4_mark &&
2824 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2825 goto nla_put_failure;
2827 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2828 nla_put_u32(skb, RTA_UID,
2829 from_kuid_munged(current_user_ns(),
2831 goto nla_put_failure;
2833 if (rt_is_input_route(rt)) {
2834 #ifdef CONFIG_IP_MROUTE
2835 if (ipv4_is_multicast(dst) &&
2836 !ipv4_is_local_multicast(dst) &&
2837 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2838 int err = ipmr_get_route(net, skb,
2839 fl4->saddr, fl4->daddr,
2845 goto nla_put_failure;
2849 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2850 goto nla_put_failure;
2854 error = rt->dst.error;
2856 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2857 goto nla_put_failure;
2859 nlmsg_end(skb, nlh);
2863 nlmsg_cancel(skb, nlh);
2867 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
2868 struct netlink_callback *cb, u32 table_id,
2869 struct fnhe_hash_bucket *bucket, int genid,
2870 int *fa_index, int fa_start, unsigned int flags)
2874 for (i = 0; i < FNHE_HASH_SIZE; i++) {
2875 struct fib_nh_exception *fnhe;
2877 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
2878 fnhe = rcu_dereference(fnhe->fnhe_next)) {
2882 if (*fa_index < fa_start)
2885 if (fnhe->fnhe_genid != genid)
2888 if (fnhe->fnhe_expires &&
2889 time_after(jiffies, fnhe->fnhe_expires))
2892 rt = rcu_dereference(fnhe->fnhe_rth_input);
2894 rt = rcu_dereference(fnhe->fnhe_rth_output);
2898 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
2899 table_id, NULL, skb,
2900 NETLINK_CB(cb->skb).portid,
2901 cb->nlh->nlmsg_seq, flags);
2912 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
2913 u32 table_id, struct fib_info *fi,
2914 int *fa_index, int fa_start, unsigned int flags)
2916 struct net *net = sock_net(cb->skb->sk);
2917 int nhsel, genid = fnhe_genid(net);
2919 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
2920 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
2921 struct fnhe_hash_bucket *bucket;
2924 if (nhc->nhc_flags & RTNH_F_DEAD)
2928 bucket = rcu_dereference(nhc->nhc_exceptions);
2931 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
2932 genid, fa_index, fa_start,
2942 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2943 u8 ip_proto, __be16 sport,
2946 struct sk_buff *skb;
2949 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2953 /* Reserve room for dummy headers, this skb can pass
2954 * through good chunk of routing engine.
2956 skb_reset_mac_header(skb);
2957 skb_reset_network_header(skb);
2958 skb->protocol = htons(ETH_P_IP);
2959 iph = skb_put(skb, sizeof(struct iphdr));
2960 iph->protocol = ip_proto;
2966 skb_set_transport_header(skb, skb->len);
2968 switch (iph->protocol) {
2970 struct udphdr *udph;
2972 udph = skb_put_zero(skb, sizeof(struct udphdr));
2973 udph->source = sport;
2975 udph->len = sizeof(struct udphdr);
2980 struct tcphdr *tcph;
2982 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2983 tcph->source = sport;
2985 tcph->doff = sizeof(struct tcphdr) / 4;
2987 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2991 case IPPROTO_ICMP: {
2992 struct icmphdr *icmph;
2994 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2995 icmph->type = ICMP_ECHO;
3003 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3004 const struct nlmsghdr *nlh,
3006 struct netlink_ext_ack *extack)
3011 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3012 NL_SET_ERR_MSG(extack,
3013 "ipv4: Invalid header for route get request");
3017 if (!netlink_strict_get_check(skb))
3018 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3019 rtm_ipv4_policy, extack);
3021 rtm = nlmsg_data(nlh);
3022 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3023 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3024 rtm->rtm_table || rtm->rtm_protocol ||
3025 rtm->rtm_scope || rtm->rtm_type) {
3026 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3030 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3031 RTM_F_LOOKUP_TABLE |
3033 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3037 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3038 rtm_ipv4_policy, extack);
3042 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3043 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3044 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3048 for (i = 0; i <= RTA_MAX; i++) {
3064 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3072 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3073 struct netlink_ext_ack *extack)
3075 struct net *net = sock_net(in_skb->sk);
3076 struct nlattr *tb[RTA_MAX+1];
3077 u32 table_id = RT_TABLE_MAIN;
3078 __be16 sport = 0, dport = 0;
3079 struct fib_result res = {};
3080 u8 ip_proto = IPPROTO_UDP;
3081 struct rtable *rt = NULL;
3082 struct sk_buff *skb;
3084 struct flowi4 fl4 = {};
3092 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3096 rtm = nlmsg_data(nlh);
3097 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3098 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3099 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3100 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3102 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3104 uid = (iif ? INVALID_UID : current_uid());
3106 if (tb[RTA_IP_PROTO]) {
3107 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3108 &ip_proto, AF_INET, extack);
3114 sport = nla_get_be16(tb[RTA_SPORT]);
3117 dport = nla_get_be16(tb[RTA_DPORT]);
3119 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3125 fl4.flowi4_tos = rtm->rtm_tos;
3126 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3127 fl4.flowi4_mark = mark;
3128 fl4.flowi4_uid = uid;
3130 fl4.fl4_sport = sport;
3132 fl4.fl4_dport = dport;
3133 fl4.flowi4_proto = ip_proto;
3138 struct net_device *dev;
3140 dev = dev_get_by_index_rcu(net, iif);
3146 fl4.flowi4_iif = iif; /* for rt_fill_info */
3149 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
3152 rt = skb_rtable(skb);
3153 if (err == 0 && rt->dst.error)
3154 err = -rt->dst.error;
3156 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3157 skb->dev = net->loopback_dev;
3158 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3163 skb_dst_set(skb, &rt->dst);
3169 if (rtm->rtm_flags & RTM_F_NOTIFY)
3170 rt->rt_flags |= RTCF_NOTIFY;
3172 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3173 table_id = res.table ? res.table->tb_id : 0;
3175 /* reset skb for netlink reply msg */
3177 skb_reset_network_header(skb);
3178 skb_reset_transport_header(skb);
3179 skb_reset_mac_header(skb);
3181 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3183 err = fib_props[res.type].error;
3185 err = -EHOSTUNREACH;
3188 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3189 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3190 rt->rt_type, res.prefix, res.prefixlen,
3191 fl4.flowi4_tos, res.fi, 0);
3193 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3194 NETLINK_CB(in_skb).portid,
3202 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3212 void ip_rt_multicast_event(struct in_device *in_dev)
3214 rt_cache_flush(dev_net(in_dev->dev));
3217 #ifdef CONFIG_SYSCTL
3218 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3219 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3220 static int ip_rt_gc_elasticity __read_mostly = 8;
3221 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3223 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3224 void __user *buffer,
3225 size_t *lenp, loff_t *ppos)
3227 struct net *net = (struct net *)__ctl->extra1;
3230 rt_cache_flush(net);
3231 fnhe_genid_bump(net);
3238 static struct ctl_table ipv4_route_table[] = {
3240 .procname = "gc_thresh",
3241 .data = &ipv4_dst_ops.gc_thresh,
3242 .maxlen = sizeof(int),
3244 .proc_handler = proc_dointvec,
3247 .procname = "max_size",
3248 .data = &ip_rt_max_size,
3249 .maxlen = sizeof(int),
3251 .proc_handler = proc_dointvec,
3254 /* Deprecated. Use gc_min_interval_ms */
3256 .procname = "gc_min_interval",
3257 .data = &ip_rt_gc_min_interval,
3258 .maxlen = sizeof(int),
3260 .proc_handler = proc_dointvec_jiffies,
3263 .procname = "gc_min_interval_ms",
3264 .data = &ip_rt_gc_min_interval,
3265 .maxlen = sizeof(int),
3267 .proc_handler = proc_dointvec_ms_jiffies,
3270 .procname = "gc_timeout",
3271 .data = &ip_rt_gc_timeout,
3272 .maxlen = sizeof(int),
3274 .proc_handler = proc_dointvec_jiffies,
3277 .procname = "gc_interval",
3278 .data = &ip_rt_gc_interval,
3279 .maxlen = sizeof(int),
3281 .proc_handler = proc_dointvec_jiffies,
3284 .procname = "redirect_load",
3285 .data = &ip_rt_redirect_load,
3286 .maxlen = sizeof(int),
3288 .proc_handler = proc_dointvec,
3291 .procname = "redirect_number",
3292 .data = &ip_rt_redirect_number,
3293 .maxlen = sizeof(int),
3295 .proc_handler = proc_dointvec,
3298 .procname = "redirect_silence",
3299 .data = &ip_rt_redirect_silence,
3300 .maxlen = sizeof(int),
3302 .proc_handler = proc_dointvec,
3305 .procname = "error_cost",
3306 .data = &ip_rt_error_cost,
3307 .maxlen = sizeof(int),
3309 .proc_handler = proc_dointvec,
3312 .procname = "error_burst",
3313 .data = &ip_rt_error_burst,
3314 .maxlen = sizeof(int),
3316 .proc_handler = proc_dointvec,
3319 .procname = "gc_elasticity",
3320 .data = &ip_rt_gc_elasticity,
3321 .maxlen = sizeof(int),
3323 .proc_handler = proc_dointvec,
3326 .procname = "mtu_expires",
3327 .data = &ip_rt_mtu_expires,
3328 .maxlen = sizeof(int),
3330 .proc_handler = proc_dointvec_jiffies,
3333 .procname = "min_pmtu",
3334 .data = &ip_rt_min_pmtu,
3335 .maxlen = sizeof(int),
3337 .proc_handler = proc_dointvec_minmax,
3338 .extra1 = &ip_min_valid_pmtu,
3341 .procname = "min_adv_mss",
3342 .data = &ip_rt_min_advmss,
3343 .maxlen = sizeof(int),
3345 .proc_handler = proc_dointvec,
3350 static const char ipv4_route_flush_procname[] = "flush";
3352 static struct ctl_table ipv4_route_flush_table[] = {
3354 .procname = ipv4_route_flush_procname,
3355 .maxlen = sizeof(int),
3357 .proc_handler = ipv4_sysctl_rtcache_flush,
3362 static __net_init int sysctl_route_net_init(struct net *net)
3364 struct ctl_table *tbl;
3366 tbl = ipv4_route_flush_table;
3367 if (!net_eq(net, &init_net)) {
3368 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3372 /* Don't export non-whitelisted sysctls to unprivileged users */
3373 if (net->user_ns != &init_user_ns) {
3374 if (tbl[0].procname != ipv4_route_flush_procname)
3375 tbl[0].procname = NULL;
3378 tbl[0].extra1 = net;
3380 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3381 if (!net->ipv4.route_hdr)
3386 if (tbl != ipv4_route_flush_table)
3392 static __net_exit void sysctl_route_net_exit(struct net *net)
3394 struct ctl_table *tbl;
3396 tbl = net->ipv4.route_hdr->ctl_table_arg;
3397 unregister_net_sysctl_table(net->ipv4.route_hdr);
3398 BUG_ON(tbl == ipv4_route_flush_table);
3402 static __net_initdata struct pernet_operations sysctl_route_ops = {
3403 .init = sysctl_route_net_init,
3404 .exit = sysctl_route_net_exit,
3408 static __net_init int rt_genid_init(struct net *net)
3410 atomic_set(&net->ipv4.rt_genid, 0);
3411 atomic_set(&net->fnhe_genid, 0);
3412 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3416 static __net_initdata struct pernet_operations rt_genid_ops = {
3417 .init = rt_genid_init,
3420 static int __net_init ipv4_inetpeer_init(struct net *net)
3422 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3426 inet_peer_base_init(bp);
3427 net->ipv4.peers = bp;
3431 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3433 struct inet_peer_base *bp = net->ipv4.peers;
3435 net->ipv4.peers = NULL;
3436 inetpeer_invalidate_tree(bp);
3440 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3441 .init = ipv4_inetpeer_init,
3442 .exit = ipv4_inetpeer_exit,
3445 #ifdef CONFIG_IP_ROUTE_CLASSID
3446 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3447 #endif /* CONFIG_IP_ROUTE_CLASSID */
3449 int __init ip_rt_init(void)
3453 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3456 panic("IP: failed to allocate ip_idents\n");
3458 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3460 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3462 panic("IP: failed to allocate ip_tstamps\n");
3464 for_each_possible_cpu(cpu) {
3465 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3467 INIT_LIST_HEAD(&ul->head);
3468 spin_lock_init(&ul->lock);
3470 #ifdef CONFIG_IP_ROUTE_CLASSID
3471 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3473 panic("IP: failed to allocate ip_rt_acct\n");
3476 ipv4_dst_ops.kmem_cachep =
3477 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3478 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3480 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3482 if (dst_entries_init(&ipv4_dst_ops) < 0)
3483 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3485 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3486 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3488 ipv4_dst_ops.gc_thresh = ~0;
3489 ip_rt_max_size = INT_MAX;
3494 if (ip_rt_proc_init())
3495 pr_err("Unable to create route proc files\n");
3500 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3501 RTNL_FLAG_DOIT_UNLOCKED);
3503 #ifdef CONFIG_SYSCTL
3504 register_pernet_subsys(&sysctl_route_ops);
3506 register_pernet_subsys(&rt_genid_ops);
3507 register_pernet_subsys(&ipv4_inetpeer_ops);
3511 #ifdef CONFIG_SYSCTL
3513 * We really need to sanitize the damn ipv4 init order, then all
3514 * this nonsense will go away.
3516 void __init ip_static_sysctl_init(void)
3518 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);