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 (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_gw_family = AF_INET;
639 rt->rt_gw4 = fnhe->fnhe_gw;
643 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
644 __be32 gw, u32 pmtu, bool lock,
645 unsigned long expires)
647 struct fnhe_hash_bucket *hash;
648 struct fib_nh_exception *fnhe;
654 genid = fnhe_genid(dev_net(nhc->nhc_dev));
655 hval = fnhe_hashfun(daddr);
657 spin_lock_bh(&fnhe_lock);
659 hash = rcu_dereference(nhc->nhc_exceptions);
661 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
664 rcu_assign_pointer(nhc->nhc_exceptions, hash);
670 for (fnhe = rcu_dereference(hash->chain); fnhe;
671 fnhe = rcu_dereference(fnhe->fnhe_next)) {
672 if (fnhe->fnhe_daddr == daddr)
678 if (fnhe->fnhe_genid != genid)
679 fnhe->fnhe_genid = genid;
683 fnhe->fnhe_pmtu = pmtu;
684 fnhe->fnhe_mtu_locked = lock;
686 fnhe->fnhe_expires = max(1UL, expires);
687 /* Update all cached dsts too */
688 rt = rcu_dereference(fnhe->fnhe_rth_input);
690 fill_route_from_fnhe(rt, fnhe);
691 rt = rcu_dereference(fnhe->fnhe_rth_output);
693 fill_route_from_fnhe(rt, fnhe);
695 if (depth > FNHE_RECLAIM_DEPTH)
696 fnhe = fnhe_oldest(hash);
698 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
702 fnhe->fnhe_next = hash->chain;
703 rcu_assign_pointer(hash->chain, fnhe);
705 fnhe->fnhe_genid = genid;
706 fnhe->fnhe_daddr = daddr;
708 fnhe->fnhe_pmtu = pmtu;
709 fnhe->fnhe_mtu_locked = lock;
710 fnhe->fnhe_expires = max(1UL, expires);
712 /* Exception created; mark the cached routes for the nexthop
713 * stale, so anyone caching it rechecks if this exception
716 rt = rcu_dereference(nhc->nhc_rth_input);
718 rt->dst.obsolete = DST_OBSOLETE_KILL;
720 for_each_possible_cpu(i) {
721 struct rtable __rcu **prt;
722 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
723 rt = rcu_dereference(*prt);
725 rt->dst.obsolete = DST_OBSOLETE_KILL;
729 fnhe->fnhe_stamp = jiffies;
732 spin_unlock_bh(&fnhe_lock);
735 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
738 __be32 new_gw = icmp_hdr(skb)->un.gateway;
739 __be32 old_gw = ip_hdr(skb)->saddr;
740 struct net_device *dev = skb->dev;
741 struct in_device *in_dev;
742 struct fib_result res;
746 switch (icmp_hdr(skb)->code & 7) {
748 case ICMP_REDIR_NETTOS:
749 case ICMP_REDIR_HOST:
750 case ICMP_REDIR_HOSTTOS:
757 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
760 in_dev = __in_dev_get_rcu(dev);
765 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
766 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
767 ipv4_is_zeronet(new_gw))
768 goto reject_redirect;
770 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
771 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
772 goto reject_redirect;
773 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
774 goto reject_redirect;
776 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
777 goto reject_redirect;
780 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
782 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
784 if (!(n->nud_state & NUD_VALID)) {
785 neigh_event_send(n, NULL);
787 if (fib_lookup(net, fl4, &res, 0) == 0) {
788 struct fib_nh_common *nhc = FIB_RES_NHC(res);
790 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
792 jiffies + ip_rt_gc_timeout);
795 rt->dst.obsolete = DST_OBSOLETE_KILL;
796 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
803 #ifdef CONFIG_IP_ROUTE_VERBOSE
804 if (IN_DEV_LOG_MARTIANS(in_dev)) {
805 const struct iphdr *iph = (const struct iphdr *) skb->data;
806 __be32 daddr = iph->daddr;
807 __be32 saddr = iph->saddr;
809 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
810 " Advised path = %pI4 -> %pI4\n",
811 &old_gw, dev->name, &new_gw,
818 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
822 const struct iphdr *iph = (const struct iphdr *) skb->data;
823 struct net *net = dev_net(skb->dev);
824 int oif = skb->dev->ifindex;
825 u8 tos = RT_TOS(iph->tos);
826 u8 prot = iph->protocol;
827 u32 mark = skb->mark;
829 rt = (struct rtable *) dst;
831 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
832 __ip_do_redirect(rt, skb, &fl4, true);
835 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
837 struct rtable *rt = (struct rtable *)dst;
838 struct dst_entry *ret = dst;
841 if (dst->obsolete > 0) {
844 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
855 * 1. The first ip_rt_redirect_number redirects are sent
856 * with exponential backoff, then we stop sending them at all,
857 * assuming that the host ignores our redirects.
858 * 2. If we did not see packets requiring redirects
859 * during ip_rt_redirect_silence, we assume that the host
860 * forgot redirected route and start to send redirects again.
862 * This algorithm is much cheaper and more intelligent than dumb load limiting
865 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
866 * and "frag. need" (breaks PMTU discovery) in icmp.c.
869 void ip_rt_send_redirect(struct sk_buff *skb)
871 struct rtable *rt = skb_rtable(skb);
872 struct in_device *in_dev;
873 struct inet_peer *peer;
879 in_dev = __in_dev_get_rcu(rt->dst.dev);
880 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
884 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
885 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
888 net = dev_net(rt->dst.dev);
889 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
891 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
892 rt_nexthop(rt, ip_hdr(skb)->daddr));
896 /* No redirected packets during ip_rt_redirect_silence;
897 * reset the algorithm.
899 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
900 peer->rate_tokens = 0;
901 peer->n_redirects = 0;
904 /* Too many ignored redirects; do not send anything
905 * set dst.rate_last to the last seen redirected packet.
907 if (peer->n_redirects >= ip_rt_redirect_number) {
908 peer->rate_last = jiffies;
912 /* Check for load limit; set rate_last to the latest sent
915 if (peer->rate_tokens == 0 ||
918 (ip_rt_redirect_load << peer->rate_tokens)))) {
919 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
921 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
922 peer->rate_last = jiffies;
925 #ifdef CONFIG_IP_ROUTE_VERBOSE
927 peer->rate_tokens == ip_rt_redirect_number)
928 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
929 &ip_hdr(skb)->saddr, inet_iif(skb),
930 &ip_hdr(skb)->daddr, &gw);
937 static int ip_error(struct sk_buff *skb)
939 struct rtable *rt = skb_rtable(skb);
940 struct net_device *dev = skb->dev;
941 struct in_device *in_dev;
942 struct inet_peer *peer;
948 if (netif_is_l3_master(skb->dev)) {
949 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
954 in_dev = __in_dev_get_rcu(dev);
956 /* IP on this device is disabled. */
960 net = dev_net(rt->dst.dev);
961 if (!IN_DEV_FORWARD(in_dev)) {
962 switch (rt->dst.error) {
964 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
974 switch (rt->dst.error) {
979 code = ICMP_HOST_UNREACH;
982 code = ICMP_NET_UNREACH;
983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
986 code = ICMP_PKT_FILTERED;
990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 l3mdev_master_ifindex(skb->dev), 1);
996 peer->rate_tokens += now - peer->rate_last;
997 if (peer->rate_tokens > ip_rt_error_burst)
998 peer->rate_tokens = ip_rt_error_burst;
999 peer->rate_last = now;
1000 if (peer->rate_tokens >= ip_rt_error_cost)
1001 peer->rate_tokens -= ip_rt_error_cost;
1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1009 out: kfree_skb(skb);
1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1015 struct dst_entry *dst = &rt->dst;
1016 u32 old_mtu = ipv4_mtu(dst);
1017 struct fib_result res;
1020 if (ip_mtu_locked(dst))
1026 if (mtu < ip_rt_min_pmtu) {
1028 mtu = min(old_mtu, ip_rt_min_pmtu);
1031 if (rt->rt_pmtu == mtu && !lock &&
1032 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1036 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1037 struct fib_nh_common *nhc = FIB_RES_NHC(res);
1039 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1040 jiffies + ip_rt_mtu_expires);
1045 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1046 struct sk_buff *skb, u32 mtu)
1048 struct rtable *rt = (struct rtable *) dst;
1051 ip_rt_build_flow_key(&fl4, sk, skb);
1052 __ip_rt_update_pmtu(rt, &fl4, mtu);
1055 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1056 int oif, u8 protocol)
1058 const struct iphdr *iph = (const struct iphdr *) skb->data;
1061 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1063 __build_flow_key(net, &fl4, NULL, iph, oif,
1064 RT_TOS(iph->tos), protocol, mark, 0);
1065 rt = __ip_route_output_key(net, &fl4);
1067 __ip_rt_update_pmtu(rt, &fl4, mtu);
1071 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1073 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1075 const struct iphdr *iph = (const struct iphdr *) skb->data;
1079 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1081 if (!fl4.flowi4_mark)
1082 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1084 rt = __ip_route_output_key(sock_net(sk), &fl4);
1086 __ip_rt_update_pmtu(rt, &fl4, mtu);
1091 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1093 const struct iphdr *iph = (const struct iphdr *) skb->data;
1096 struct dst_entry *odst = NULL;
1098 struct net *net = sock_net(sk);
1102 if (!ip_sk_accept_pmtu(sk))
1105 odst = sk_dst_get(sk);
1107 if (sock_owned_by_user(sk) || !odst) {
1108 __ipv4_sk_update_pmtu(skb, sk, mtu);
1112 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1114 rt = (struct rtable *)odst;
1115 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1123 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1125 if (!dst_check(&rt->dst, 0)) {
1127 dst_release(&rt->dst);
1129 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1137 sk_dst_set(sk, &rt->dst);
1143 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1145 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1146 int oif, u8 protocol)
1148 const struct iphdr *iph = (const struct iphdr *) skb->data;
1152 __build_flow_key(net, &fl4, NULL, iph, oif,
1153 RT_TOS(iph->tos), protocol, 0, 0);
1154 rt = __ip_route_output_key(net, &fl4);
1156 __ip_do_redirect(rt, skb, &fl4, false);
1160 EXPORT_SYMBOL_GPL(ipv4_redirect);
1162 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1164 const struct iphdr *iph = (const struct iphdr *) skb->data;
1167 struct net *net = sock_net(sk);
1169 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1170 rt = __ip_route_output_key(net, &fl4);
1172 __ip_do_redirect(rt, skb, &fl4, false);
1176 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1178 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1180 struct rtable *rt = (struct rtable *) dst;
1182 /* All IPV4 dsts are created with ->obsolete set to the value
1183 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1184 * into this function always.
1186 * When a PMTU/redirect information update invalidates a route,
1187 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1188 * DST_OBSOLETE_DEAD.
1190 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1195 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1197 struct ip_options opt;
1200 /* Recompile ip options since IPCB may not be valid anymore.
1201 * Also check we have a reasonable ipv4 header.
1203 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1207 memset(&opt, 0, sizeof(opt));
1208 if (ip_hdr(skb)->ihl > 5) {
1209 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1211 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1214 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1220 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1223 static void ipv4_link_failure(struct sk_buff *skb)
1227 ipv4_send_dest_unreach(skb);
1229 rt = skb_rtable(skb);
1231 dst_set_expires(&rt->dst, 0);
1234 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1236 pr_debug("%s: %pI4 -> %pI4, %s\n",
1237 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1238 skb->dev ? skb->dev->name : "?");
1245 We do not cache source address of outgoing interface,
1246 because it is used only by IP RR, TS and SRR options,
1247 so that it out of fast path.
1249 BTW remember: "addr" is allowed to be not aligned
1253 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1257 if (rt_is_output_route(rt))
1258 src = ip_hdr(skb)->saddr;
1260 struct fib_result res;
1261 struct iphdr *iph = ip_hdr(skb);
1262 struct flowi4 fl4 = {
1263 .daddr = iph->daddr,
1264 .saddr = iph->saddr,
1265 .flowi4_tos = RT_TOS(iph->tos),
1266 .flowi4_oif = rt->dst.dev->ifindex,
1267 .flowi4_iif = skb->dev->ifindex,
1268 .flowi4_mark = skb->mark,
1272 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1273 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1275 src = inet_select_addr(rt->dst.dev,
1276 rt_nexthop(rt, iph->daddr),
1280 memcpy(addr, &src, 4);
1283 #ifdef CONFIG_IP_ROUTE_CLASSID
1284 static void set_class_tag(struct rtable *rt, u32 tag)
1286 if (!(rt->dst.tclassid & 0xFFFF))
1287 rt->dst.tclassid |= tag & 0xFFFF;
1288 if (!(rt->dst.tclassid & 0xFFFF0000))
1289 rt->dst.tclassid |= tag & 0xFFFF0000;
1293 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1295 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1296 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1299 return min(advmss, IPV4_MAX_PMTU - header_size);
1302 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1304 const struct rtable *rt = (const struct rtable *) dst;
1305 unsigned int mtu = rt->rt_pmtu;
1307 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1308 mtu = dst_metric_raw(dst, RTAX_MTU);
1313 mtu = READ_ONCE(dst->dev->mtu);
1315 if (unlikely(ip_mtu_locked(dst))) {
1316 if (rt->rt_gw_family && mtu > 576)
1320 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1322 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1325 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1327 struct fnhe_hash_bucket *hash;
1328 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1329 u32 hval = fnhe_hashfun(daddr);
1331 spin_lock_bh(&fnhe_lock);
1333 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1334 lockdep_is_held(&fnhe_lock));
1337 fnhe_p = &hash->chain;
1338 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1340 if (fnhe->fnhe_daddr == daddr) {
1341 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1342 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1343 /* set fnhe_daddr to 0 to ensure it won't bind with
1344 * new dsts in rt_bind_exception().
1346 fnhe->fnhe_daddr = 0;
1347 fnhe_flush_routes(fnhe);
1348 kfree_rcu(fnhe, rcu);
1351 fnhe_p = &fnhe->fnhe_next;
1352 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1353 lockdep_is_held(&fnhe_lock));
1356 spin_unlock_bh(&fnhe_lock);
1359 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1362 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1363 struct fib_nh_exception *fnhe;
1369 hval = fnhe_hashfun(daddr);
1371 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1372 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1373 if (fnhe->fnhe_daddr == daddr) {
1374 if (fnhe->fnhe_expires &&
1375 time_after(jiffies, fnhe->fnhe_expires)) {
1376 ip_del_fnhe(nhc, daddr);
1386 * 1. mtu on route is locked - use it
1387 * 2. mtu from nexthop exception
1388 * 3. mtu from egress device
1391 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1393 struct fib_nh_common *nhc = res->nhc;
1394 struct net_device *dev = nhc->nhc_dev;
1395 struct fib_info *fi = res->fi;
1398 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1399 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1403 struct fib_nh_exception *fnhe;
1405 fnhe = find_exception(nhc, daddr);
1406 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1407 mtu = fnhe->fnhe_pmtu;
1411 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1413 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1416 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1417 __be32 daddr, const bool do_cache)
1421 spin_lock_bh(&fnhe_lock);
1423 if (daddr == fnhe->fnhe_daddr) {
1424 struct rtable __rcu **porig;
1425 struct rtable *orig;
1426 int genid = fnhe_genid(dev_net(rt->dst.dev));
1428 if (rt_is_input_route(rt))
1429 porig = &fnhe->fnhe_rth_input;
1431 porig = &fnhe->fnhe_rth_output;
1432 orig = rcu_dereference(*porig);
1434 if (fnhe->fnhe_genid != genid) {
1435 fnhe->fnhe_genid = genid;
1437 fnhe->fnhe_pmtu = 0;
1438 fnhe->fnhe_expires = 0;
1439 fnhe->fnhe_mtu_locked = false;
1440 fnhe_flush_routes(fnhe);
1443 fill_route_from_fnhe(rt, fnhe);
1446 rt->rt_gw_family = AF_INET;
1451 rcu_assign_pointer(*porig, rt);
1453 dst_dev_put(&orig->dst);
1454 dst_release(&orig->dst);
1459 fnhe->fnhe_stamp = jiffies;
1461 spin_unlock_bh(&fnhe_lock);
1466 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1468 struct rtable *orig, *prev, **p;
1471 if (rt_is_input_route(rt)) {
1472 p = (struct rtable **)&nhc->nhc_rth_input;
1474 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1478 /* hold dst before doing cmpxchg() to avoid race condition
1482 prev = cmpxchg(p, orig, rt);
1485 dst_dev_put(&orig->dst);
1486 dst_release(&orig->dst);
1489 dst_release(&rt->dst);
1496 struct uncached_list {
1498 struct list_head head;
1501 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1503 void rt_add_uncached_list(struct rtable *rt)
1505 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1507 rt->rt_uncached_list = ul;
1509 spin_lock_bh(&ul->lock);
1510 list_add_tail(&rt->rt_uncached, &ul->head);
1511 spin_unlock_bh(&ul->lock);
1514 void rt_del_uncached_list(struct rtable *rt)
1516 if (!list_empty(&rt->rt_uncached)) {
1517 struct uncached_list *ul = rt->rt_uncached_list;
1519 spin_lock_bh(&ul->lock);
1520 list_del(&rt->rt_uncached);
1521 spin_unlock_bh(&ul->lock);
1525 static void ipv4_dst_destroy(struct dst_entry *dst)
1527 struct rtable *rt = (struct rtable *)dst;
1529 ip_dst_metrics_put(dst);
1530 rt_del_uncached_list(rt);
1533 void rt_flush_dev(struct net_device *dev)
1535 struct net *net = dev_net(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 = net->loopback_dev;
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_gw_family = nhc->nhc_gw_family;
1574 /* only INET and INET6 are supported */
1575 if (likely(nhc->nhc_gw_family == AF_INET))
1576 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1578 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1581 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1583 #ifdef CONFIG_IP_ROUTE_CLASSID
1584 if (nhc->nhc_family == AF_INET) {
1587 nh = container_of(nhc, struct fib_nh, nh_common);
1588 rt->dst.tclassid = nh->nh_tclassid;
1591 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1593 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1595 cached = rt_cache_route(nhc, rt);
1596 if (unlikely(!cached)) {
1597 /* Routes we intend to cache in nexthop exception or
1598 * FIB nexthop have the DST_NOCACHE bit clear.
1599 * However, if we are unsuccessful at storing this
1600 * route into the cache we really need to set it.
1603 rt->rt_gw_family = AF_INET;
1606 rt_add_uncached_list(rt);
1609 rt_add_uncached_list(rt);
1611 #ifdef CONFIG_IP_ROUTE_CLASSID
1612 #ifdef CONFIG_IP_MULTIPLE_TABLES
1613 set_class_tag(rt, res->tclassid);
1615 set_class_tag(rt, itag);
1619 struct rtable *rt_dst_alloc(struct net_device *dev,
1620 unsigned int flags, u16 type,
1621 bool nopolicy, bool noxfrm, bool will_cache)
1625 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1626 (will_cache ? 0 : DST_HOST) |
1627 (nopolicy ? DST_NOPOLICY : 0) |
1628 (noxfrm ? DST_NOXFRM : 0));
1631 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1632 rt->rt_flags = flags;
1634 rt->rt_is_input = 0;
1637 rt->rt_mtu_locked = 0;
1638 rt->rt_gw_family = 0;
1640 INIT_LIST_HEAD(&rt->rt_uncached);
1642 rt->dst.output = ip_output;
1643 if (flags & RTCF_LOCAL)
1644 rt->dst.input = ip_local_deliver;
1649 EXPORT_SYMBOL(rt_dst_alloc);
1651 /* called in rcu_read_lock() section */
1652 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1653 u8 tos, struct net_device *dev,
1654 struct in_device *in_dev, u32 *itag)
1658 /* Primary sanity checks. */
1662 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1663 skb->protocol != htons(ETH_P_IP))
1666 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1669 if (ipv4_is_zeronet(saddr)) {
1670 if (!ipv4_is_local_multicast(daddr) &&
1671 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1674 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1682 /* called in rcu_read_lock() section */
1683 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1684 u8 tos, struct net_device *dev, int our)
1686 struct in_device *in_dev = __in_dev_get_rcu(dev);
1687 unsigned int flags = RTCF_MULTICAST;
1692 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1697 flags |= RTCF_LOCAL;
1699 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1700 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1704 #ifdef CONFIG_IP_ROUTE_CLASSID
1705 rth->dst.tclassid = itag;
1707 rth->dst.output = ip_rt_bug;
1708 rth->rt_is_input= 1;
1710 #ifdef CONFIG_IP_MROUTE
1711 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1712 rth->dst.input = ip_mr_input;
1714 RT_CACHE_STAT_INC(in_slow_mc);
1716 skb_dst_set(skb, &rth->dst);
1721 static void ip_handle_martian_source(struct net_device *dev,
1722 struct in_device *in_dev,
1723 struct sk_buff *skb,
1727 RT_CACHE_STAT_INC(in_martian_src);
1728 #ifdef CONFIG_IP_ROUTE_VERBOSE
1729 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1731 * RFC1812 recommendation, if source is martian,
1732 * the only hint is MAC header.
1734 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1735 &daddr, &saddr, dev->name);
1736 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1737 print_hex_dump(KERN_WARNING, "ll header: ",
1738 DUMP_PREFIX_OFFSET, 16, 1,
1739 skb_mac_header(skb),
1740 dev->hard_header_len, false);
1746 /* called in rcu_read_lock() section */
1747 static int __mkroute_input(struct sk_buff *skb,
1748 const struct fib_result *res,
1749 struct in_device *in_dev,
1750 __be32 daddr, __be32 saddr, u32 tos)
1752 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1753 struct net_device *dev = nhc->nhc_dev;
1754 struct fib_nh_exception *fnhe;
1757 struct in_device *out_dev;
1761 /* get a working reference to the output device */
1762 out_dev = __in_dev_get_rcu(dev);
1764 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1768 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1769 in_dev->dev, in_dev, &itag);
1771 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1777 do_cache = res->fi && !itag;
1778 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1779 skb->protocol == htons(ETH_P_IP)) {
1782 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1783 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1784 inet_addr_onlink(out_dev, saddr, gw))
1785 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1788 if (skb->protocol != htons(ETH_P_IP)) {
1789 /* Not IP (i.e. ARP). Do not create route, if it is
1790 * invalid for proxy arp. DNAT routes are always valid.
1792 * Proxy arp feature have been extended to allow, ARP
1793 * replies back to the same interface, to support
1794 * Private VLAN switch technologies. See arp.c.
1796 if (out_dev == in_dev &&
1797 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1803 fnhe = find_exception(nhc, daddr);
1806 rth = rcu_dereference(fnhe->fnhe_rth_input);
1808 rth = rcu_dereference(nhc->nhc_rth_input);
1809 if (rt_cache_valid(rth)) {
1810 skb_dst_set_noref(skb, &rth->dst);
1815 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1816 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1817 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1823 rth->rt_is_input = 1;
1824 RT_CACHE_STAT_INC(in_slow_tot);
1826 rth->dst.input = ip_forward;
1828 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1830 lwtunnel_set_redirect(&rth->dst);
1831 skb_dst_set(skb, &rth->dst);
1838 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1839 /* To make ICMP packets follow the right flow, the multipath hash is
1840 * calculated from the inner IP addresses.
1842 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1843 struct flow_keys *hash_keys)
1845 const struct iphdr *outer_iph = ip_hdr(skb);
1846 const struct iphdr *key_iph = outer_iph;
1847 const struct iphdr *inner_iph;
1848 const struct icmphdr *icmph;
1849 struct iphdr _inner_iph;
1850 struct icmphdr _icmph;
1852 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1855 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1858 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1863 if (icmph->type != ICMP_DEST_UNREACH &&
1864 icmph->type != ICMP_REDIRECT &&
1865 icmph->type != ICMP_TIME_EXCEEDED &&
1866 icmph->type != ICMP_PARAMETERPROB)
1869 inner_iph = skb_header_pointer(skb,
1870 outer_iph->ihl * 4 + sizeof(_icmph),
1871 sizeof(_inner_iph), &_inner_iph);
1875 key_iph = inner_iph;
1877 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1878 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1881 /* if skb is set it will be used and fl4 can be NULL */
1882 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1883 const struct sk_buff *skb, struct flow_keys *flkeys)
1885 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1886 struct flow_keys hash_keys;
1889 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1891 memset(&hash_keys, 0, sizeof(hash_keys));
1892 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1894 ip_multipath_l3_keys(skb, &hash_keys);
1896 hash_keys.addrs.v4addrs.src = fl4->saddr;
1897 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1901 /* skb is currently provided only when forwarding */
1903 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1904 struct flow_keys keys;
1906 /* short-circuit if we already have L4 hash present */
1908 return skb_get_hash_raw(skb) >> 1;
1910 memset(&hash_keys, 0, sizeof(hash_keys));
1913 skb_flow_dissect_flow_keys(skb, &keys, flag);
1917 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1918 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1919 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1920 hash_keys.ports.src = flkeys->ports.src;
1921 hash_keys.ports.dst = flkeys->ports.dst;
1922 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1924 memset(&hash_keys, 0, sizeof(hash_keys));
1925 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1926 hash_keys.addrs.v4addrs.src = fl4->saddr;
1927 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1928 hash_keys.ports.src = fl4->fl4_sport;
1929 hash_keys.ports.dst = fl4->fl4_dport;
1930 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1934 memset(&hash_keys, 0, sizeof(hash_keys));
1935 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1936 /* skb is currently provided only when forwarding */
1938 struct flow_keys keys;
1940 skb_flow_dissect_flow_keys(skb, &keys, 0);
1942 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1943 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1945 /* Same as case 0 */
1946 hash_keys.addrs.v4addrs.src = fl4->saddr;
1947 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1951 mhash = flow_hash_from_keys(&hash_keys);
1954 mhash = jhash_2words(mhash, multipath_hash, 0);
1958 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1960 static int ip_mkroute_input(struct sk_buff *skb,
1961 struct fib_result *res,
1962 struct in_device *in_dev,
1963 __be32 daddr, __be32 saddr, u32 tos,
1964 struct flow_keys *hkeys)
1966 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1967 if (res->fi && fib_info_num_path(res->fi) > 1) {
1968 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1970 fib_select_multipath(res, h);
1974 /* create a routing cache entry */
1975 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1979 * NOTE. We drop all the packets that has local source
1980 * addresses, because every properly looped back packet
1981 * must have correct destination already attached by output routine.
1983 * Such approach solves two big problems:
1984 * 1. Not simplex devices are handled properly.
1985 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1986 * called with rcu_read_lock()
1989 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1990 u8 tos, struct net_device *dev,
1991 struct fib_result *res)
1993 struct in_device *in_dev = __in_dev_get_rcu(dev);
1994 struct flow_keys *flkeys = NULL, _flkeys;
1995 struct net *net = dev_net(dev);
1996 struct ip_tunnel_info *tun_info;
1998 unsigned int flags = 0;
2002 bool do_cache = true;
2004 /* IP on this device is disabled. */
2009 /* Check for the most weird martians, which can be not detected
2013 tun_info = skb_tunnel_info(skb);
2014 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2015 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2017 fl4.flowi4_tun_key.tun_id = 0;
2020 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2021 goto martian_source;
2025 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2028 /* Accept zero addresses only to limited broadcast;
2029 * I even do not know to fix it or not. Waiting for complains :-)
2031 if (ipv4_is_zeronet(saddr))
2032 goto martian_source;
2034 if (ipv4_is_zeronet(daddr))
2035 goto martian_destination;
2037 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2038 * and call it once if daddr or/and saddr are loopback addresses
2040 if (ipv4_is_loopback(daddr)) {
2041 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2042 goto martian_destination;
2043 } else if (ipv4_is_loopback(saddr)) {
2044 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2045 goto martian_source;
2049 * Now we are ready to route packet.
2052 fl4.flowi4_iif = dev->ifindex;
2053 fl4.flowi4_mark = skb->mark;
2054 fl4.flowi4_tos = tos;
2055 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2056 fl4.flowi4_flags = 0;
2059 fl4.flowi4_uid = sock_net_uid(net, NULL);
2061 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2064 fl4.flowi4_proto = 0;
2069 err = fib_lookup(net, &fl4, res, 0);
2071 if (!IN_DEV_FORWARD(in_dev))
2072 err = -EHOSTUNREACH;
2076 if (res->type == RTN_BROADCAST) {
2077 if (IN_DEV_BFORWARD(in_dev))
2079 /* not do cache if bc_forwarding is enabled */
2080 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2085 if (res->type == RTN_LOCAL) {
2086 err = fib_validate_source(skb, saddr, daddr, tos,
2087 0, dev, in_dev, &itag);
2089 goto martian_source;
2093 if (!IN_DEV_FORWARD(in_dev)) {
2094 err = -EHOSTUNREACH;
2097 if (res->type != RTN_UNICAST)
2098 goto martian_destination;
2101 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2105 if (skb->protocol != htons(ETH_P_IP))
2108 if (!ipv4_is_zeronet(saddr)) {
2109 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2112 goto martian_source;
2114 flags |= RTCF_BROADCAST;
2115 res->type = RTN_BROADCAST;
2116 RT_CACHE_STAT_INC(in_brd);
2119 do_cache &= res->fi && !itag;
2121 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2123 rth = rcu_dereference(nhc->nhc_rth_input);
2124 if (rt_cache_valid(rth)) {
2125 skb_dst_set_noref(skb, &rth->dst);
2131 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2132 flags | RTCF_LOCAL, res->type,
2133 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2137 rth->dst.output= ip_rt_bug;
2138 #ifdef CONFIG_IP_ROUTE_CLASSID
2139 rth->dst.tclassid = itag;
2141 rth->rt_is_input = 1;
2143 RT_CACHE_STAT_INC(in_slow_tot);
2144 if (res->type == RTN_UNREACHABLE) {
2145 rth->dst.input= ip_error;
2146 rth->dst.error= -err;
2147 rth->rt_flags &= ~RTCF_LOCAL;
2151 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2153 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2154 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2155 WARN_ON(rth->dst.input == lwtunnel_input);
2156 rth->dst.lwtstate->orig_input = rth->dst.input;
2157 rth->dst.input = lwtunnel_input;
2160 if (unlikely(!rt_cache_route(nhc, rth)))
2161 rt_add_uncached_list(rth);
2163 skb_dst_set(skb, &rth->dst);
2168 RT_CACHE_STAT_INC(in_no_route);
2169 res->type = RTN_UNREACHABLE;
2175 * Do not cache martian addresses: they should be logged (RFC1812)
2177 martian_destination:
2178 RT_CACHE_STAT_INC(in_martian_dst);
2179 #ifdef CONFIG_IP_ROUTE_VERBOSE
2180 if (IN_DEV_LOG_MARTIANS(in_dev))
2181 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2182 &daddr, &saddr, dev->name);
2194 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2198 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2199 u8 tos, struct net_device *dev)
2201 struct fib_result res;
2204 tos &= IPTOS_RT_MASK;
2206 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2211 EXPORT_SYMBOL(ip_route_input_noref);
2213 /* called with rcu_read_lock held */
2214 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2215 u8 tos, struct net_device *dev, struct fib_result *res)
2217 /* Multicast recognition logic is moved from route cache to here.
2218 The problem was that too many Ethernet cards have broken/missing
2219 hardware multicast filters :-( As result the host on multicasting
2220 network acquires a lot of useless route cache entries, sort of
2221 SDR messages from all the world. Now we try to get rid of them.
2222 Really, provided software IP multicast filter is organized
2223 reasonably (at least, hashed), it does not result in a slowdown
2224 comparing with route cache reject entries.
2225 Note, that multicast routers are not affected, because
2226 route cache entry is created eventually.
2228 if (ipv4_is_multicast(daddr)) {
2229 struct in_device *in_dev = __in_dev_get_rcu(dev);
2235 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2236 ip_hdr(skb)->protocol);
2238 /* check l3 master if no match yet */
2239 if (!our && netif_is_l3_slave(dev)) {
2240 struct in_device *l3_in_dev;
2242 l3_in_dev = __in_dev_get_rcu(skb->dev);
2244 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2245 ip_hdr(skb)->protocol);
2249 #ifdef CONFIG_IP_MROUTE
2251 (!ipv4_is_local_multicast(daddr) &&
2252 IN_DEV_MFORWARD(in_dev))
2255 err = ip_route_input_mc(skb, daddr, saddr,
2261 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2264 /* called with rcu_read_lock() */
2265 static struct rtable *__mkroute_output(const struct fib_result *res,
2266 const struct flowi4 *fl4, int orig_oif,
2267 struct net_device *dev_out,
2270 struct fib_info *fi = res->fi;
2271 struct fib_nh_exception *fnhe;
2272 struct in_device *in_dev;
2273 u16 type = res->type;
2277 in_dev = __in_dev_get_rcu(dev_out);
2279 return ERR_PTR(-EINVAL);
2281 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2282 if (ipv4_is_loopback(fl4->saddr) &&
2283 !(dev_out->flags & IFF_LOOPBACK) &&
2284 !netif_is_l3_master(dev_out))
2285 return ERR_PTR(-EINVAL);
2287 if (ipv4_is_lbcast(fl4->daddr))
2288 type = RTN_BROADCAST;
2289 else if (ipv4_is_multicast(fl4->daddr))
2290 type = RTN_MULTICAST;
2291 else if (ipv4_is_zeronet(fl4->daddr))
2292 return ERR_PTR(-EINVAL);
2294 if (dev_out->flags & IFF_LOOPBACK)
2295 flags |= RTCF_LOCAL;
2298 if (type == RTN_BROADCAST) {
2299 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2301 } else if (type == RTN_MULTICAST) {
2302 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2303 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2305 flags &= ~RTCF_LOCAL;
2308 /* If multicast route do not exist use
2309 * default one, but do not gateway in this case.
2312 if (fi && res->prefixlen < 4)
2314 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2315 (orig_oif != dev_out->ifindex)) {
2316 /* For local routes that require a particular output interface
2317 * we do not want to cache the result. Caching the result
2318 * causes incorrect behaviour when there are multiple source
2319 * addresses on the interface, the end result being that if the
2320 * intended recipient is waiting on that interface for the
2321 * packet he won't receive it because it will be delivered on
2322 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2323 * be set to the loopback interface as well.
2329 do_cache &= fi != NULL;
2331 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2332 struct rtable __rcu **prth;
2334 fnhe = find_exception(nhc, fl4->daddr);
2338 prth = &fnhe->fnhe_rth_output;
2340 if (unlikely(fl4->flowi4_flags &
2341 FLOWI_FLAG_KNOWN_NH &&
2342 !(nhc->nhc_gw_family &&
2343 nhc->nhc_scope == RT_SCOPE_LINK))) {
2347 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2349 rth = rcu_dereference(*prth);
2350 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2355 rth = rt_dst_alloc(dev_out, flags, type,
2356 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2357 IN_DEV_CONF_GET(in_dev, NOXFRM),
2360 return ERR_PTR(-ENOBUFS);
2362 rth->rt_iif = orig_oif;
2364 RT_CACHE_STAT_INC(out_slow_tot);
2366 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2367 if (flags & RTCF_LOCAL &&
2368 !(dev_out->flags & IFF_LOOPBACK)) {
2369 rth->dst.output = ip_mc_output;
2370 RT_CACHE_STAT_INC(out_slow_mc);
2372 #ifdef CONFIG_IP_MROUTE
2373 if (type == RTN_MULTICAST) {
2374 if (IN_DEV_MFORWARD(in_dev) &&
2375 !ipv4_is_local_multicast(fl4->daddr)) {
2376 rth->dst.input = ip_mr_input;
2377 rth->dst.output = ip_mc_output;
2383 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2384 lwtunnel_set_redirect(&rth->dst);
2390 * Major route resolver routine.
2393 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2394 const struct sk_buff *skb)
2396 __u8 tos = RT_FL_TOS(fl4);
2397 struct fib_result res = {
2405 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2406 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2407 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2408 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2411 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2416 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2418 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2419 struct fib_result *res,
2420 const struct sk_buff *skb)
2422 struct net_device *dev_out = NULL;
2423 int orig_oif = fl4->flowi4_oif;
2424 unsigned int flags = 0;
2426 int err = -ENETUNREACH;
2429 rth = ERR_PTR(-EINVAL);
2430 if (ipv4_is_multicast(fl4->saddr) ||
2431 ipv4_is_lbcast(fl4->saddr) ||
2432 ipv4_is_zeronet(fl4->saddr))
2435 /* I removed check for oif == dev_out->oif here.
2436 It was wrong for two reasons:
2437 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2438 is assigned to multiple interfaces.
2439 2. Moreover, we are allowed to send packets with saddr
2440 of another iface. --ANK
2443 if (fl4->flowi4_oif == 0 &&
2444 (ipv4_is_multicast(fl4->daddr) ||
2445 ipv4_is_lbcast(fl4->daddr))) {
2446 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2447 dev_out = __ip_dev_find(net, fl4->saddr, false);
2451 /* Special hack: user can direct multicasts
2452 and limited broadcast via necessary interface
2453 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2454 This hack is not just for fun, it allows
2455 vic,vat and friends to work.
2456 They bind socket to loopback, set ttl to zero
2457 and expect that it will work.
2458 From the viewpoint of routing cache they are broken,
2459 because we are not allowed to build multicast path
2460 with loopback source addr (look, routing cache
2461 cannot know, that ttl is zero, so that packet
2462 will not leave this host and route is valid).
2463 Luckily, this hack is good workaround.
2466 fl4->flowi4_oif = dev_out->ifindex;
2470 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2471 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2472 if (!__ip_dev_find(net, fl4->saddr, false))
2478 if (fl4->flowi4_oif) {
2479 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2480 rth = ERR_PTR(-ENODEV);
2484 /* RACE: Check return value of inet_select_addr instead. */
2485 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2486 rth = ERR_PTR(-ENETUNREACH);
2489 if (ipv4_is_local_multicast(fl4->daddr) ||
2490 ipv4_is_lbcast(fl4->daddr) ||
2491 fl4->flowi4_proto == IPPROTO_IGMP) {
2493 fl4->saddr = inet_select_addr(dev_out, 0,
2498 if (ipv4_is_multicast(fl4->daddr))
2499 fl4->saddr = inet_select_addr(dev_out, 0,
2501 else if (!fl4->daddr)
2502 fl4->saddr = inet_select_addr(dev_out, 0,
2508 fl4->daddr = fl4->saddr;
2510 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2511 dev_out = net->loopback_dev;
2512 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2513 res->type = RTN_LOCAL;
2514 flags |= RTCF_LOCAL;
2518 err = fib_lookup(net, fl4, res, 0);
2522 if (fl4->flowi4_oif &&
2523 (ipv4_is_multicast(fl4->daddr) ||
2524 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2525 /* Apparently, routing tables are wrong. Assume,
2526 that the destination is on link.
2529 Because we are allowed to send to iface
2530 even if it has NO routes and NO assigned
2531 addresses. When oif is specified, routing
2532 tables are looked up with only one purpose:
2533 to catch if destination is gatewayed, rather than
2534 direct. Moreover, if MSG_DONTROUTE is set,
2535 we send packet, ignoring both routing tables
2536 and ifaddr state. --ANK
2539 We could make it even if oif is unknown,
2540 likely IPv6, but we do not.
2543 if (fl4->saddr == 0)
2544 fl4->saddr = inet_select_addr(dev_out, 0,
2546 res->type = RTN_UNICAST;
2553 if (res->type == RTN_LOCAL) {
2555 if (res->fi->fib_prefsrc)
2556 fl4->saddr = res->fi->fib_prefsrc;
2558 fl4->saddr = fl4->daddr;
2561 /* L3 master device is the loopback for that domain */
2562 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2565 /* make sure orig_oif points to fib result device even
2566 * though packet rx/tx happens over loopback or l3mdev
2568 orig_oif = FIB_RES_OIF(*res);
2570 fl4->flowi4_oif = dev_out->ifindex;
2571 flags |= RTCF_LOCAL;
2575 fib_select_path(net, res, fl4, skb);
2577 dev_out = FIB_RES_DEV(*res);
2578 fl4->flowi4_oif = dev_out->ifindex;
2582 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2588 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2593 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2595 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2597 return mtu ? : dst->dev->mtu;
2600 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2601 struct sk_buff *skb, u32 mtu)
2605 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2606 struct sk_buff *skb)
2610 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2616 static struct dst_ops ipv4_dst_blackhole_ops = {
2618 .check = ipv4_blackhole_dst_check,
2619 .mtu = ipv4_blackhole_mtu,
2620 .default_advmss = ipv4_default_advmss,
2621 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2622 .redirect = ipv4_rt_blackhole_redirect,
2623 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2624 .neigh_lookup = ipv4_neigh_lookup,
2627 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2629 struct rtable *ort = (struct rtable *) dst_orig;
2632 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2634 struct dst_entry *new = &rt->dst;
2637 new->input = dst_discard;
2638 new->output = dst_discard_out;
2640 new->dev = net->loopback_dev;
2644 rt->rt_is_input = ort->rt_is_input;
2645 rt->rt_iif = ort->rt_iif;
2646 rt->rt_pmtu = ort->rt_pmtu;
2647 rt->rt_mtu_locked = ort->rt_mtu_locked;
2649 rt->rt_genid = rt_genid_ipv4(net);
2650 rt->rt_flags = ort->rt_flags;
2651 rt->rt_type = ort->rt_type;
2652 rt->rt_gw_family = ort->rt_gw_family;
2653 if (rt->rt_gw_family == AF_INET)
2654 rt->rt_gw4 = ort->rt_gw4;
2655 else if (rt->rt_gw_family == AF_INET6)
2656 rt->rt_gw6 = ort->rt_gw6;
2658 INIT_LIST_HEAD(&rt->rt_uncached);
2661 dst_release(dst_orig);
2663 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2666 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2667 const struct sock *sk)
2669 struct rtable *rt = __ip_route_output_key(net, flp4);
2674 if (flp4->flowi4_proto)
2675 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2676 flowi4_to_flowi(flp4),
2681 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2683 /* called with rcu_read_lock held */
2684 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2685 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2686 struct sk_buff *skb, u32 portid, u32 seq)
2689 struct nlmsghdr *nlh;
2690 unsigned long expires = 0;
2692 u32 metrics[RTAX_MAX];
2694 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2698 r = nlmsg_data(nlh);
2699 r->rtm_family = AF_INET;
2700 r->rtm_dst_len = 32;
2702 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2703 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2704 if (nla_put_u32(skb, RTA_TABLE, table_id))
2705 goto nla_put_failure;
2706 r->rtm_type = rt->rt_type;
2707 r->rtm_scope = RT_SCOPE_UNIVERSE;
2708 r->rtm_protocol = RTPROT_UNSPEC;
2709 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2710 if (rt->rt_flags & RTCF_NOTIFY)
2711 r->rtm_flags |= RTM_F_NOTIFY;
2712 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2713 r->rtm_flags |= RTCF_DOREDIRECT;
2715 if (nla_put_in_addr(skb, RTA_DST, dst))
2716 goto nla_put_failure;
2718 r->rtm_src_len = 32;
2719 if (nla_put_in_addr(skb, RTA_SRC, src))
2720 goto nla_put_failure;
2723 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2724 goto nla_put_failure;
2725 #ifdef CONFIG_IP_ROUTE_CLASSID
2726 if (rt->dst.tclassid &&
2727 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2728 goto nla_put_failure;
2730 if (fl4 && !rt_is_input_route(rt) &&
2731 fl4->saddr != src) {
2732 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2733 goto nla_put_failure;
2735 if (rt->rt_gw_family == AF_INET &&
2736 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2737 goto nla_put_failure;
2738 } else if (rt->rt_gw_family == AF_INET6) {
2739 int alen = sizeof(struct in6_addr);
2743 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2745 goto nla_put_failure;
2747 via = nla_data(nla);
2748 via->rtvia_family = AF_INET6;
2749 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2752 expires = rt->dst.expires;
2754 unsigned long now = jiffies;
2756 if (time_before(now, expires))
2762 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2763 if (rt->rt_pmtu && expires)
2764 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2765 if (rt->rt_mtu_locked && expires)
2766 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2767 if (rtnetlink_put_metrics(skb, metrics) < 0)
2768 goto nla_put_failure;
2771 if (fl4->flowi4_mark &&
2772 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2773 goto nla_put_failure;
2775 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2776 nla_put_u32(skb, RTA_UID,
2777 from_kuid_munged(current_user_ns(),
2779 goto nla_put_failure;
2781 if (rt_is_input_route(rt)) {
2782 #ifdef CONFIG_IP_MROUTE
2783 if (ipv4_is_multicast(dst) &&
2784 !ipv4_is_local_multicast(dst) &&
2785 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2786 int err = ipmr_get_route(net, skb,
2787 fl4->saddr, fl4->daddr,
2793 goto nla_put_failure;
2797 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2798 goto nla_put_failure;
2802 error = rt->dst.error;
2804 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2805 goto nla_put_failure;
2807 nlmsg_end(skb, nlh);
2811 nlmsg_cancel(skb, nlh);
2815 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
2816 struct netlink_callback *cb, u32 table_id,
2817 struct fnhe_hash_bucket *bucket, int genid,
2818 int *fa_index, int fa_start)
2822 for (i = 0; i < FNHE_HASH_SIZE; i++) {
2823 struct fib_nh_exception *fnhe;
2825 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
2826 fnhe = rcu_dereference(fnhe->fnhe_next)) {
2830 if (*fa_index < fa_start)
2833 if (fnhe->fnhe_genid != genid)
2836 if (fnhe->fnhe_expires &&
2837 time_after(jiffies, fnhe->fnhe_expires))
2840 rt = rcu_dereference(fnhe->fnhe_rth_input);
2842 rt = rcu_dereference(fnhe->fnhe_rth_output);
2846 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
2847 table_id, NULL, skb,
2848 NETLINK_CB(cb->skb).portid,
2849 cb->nlh->nlmsg_seq);
2860 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
2861 u32 table_id, struct fib_info *fi,
2862 int *fa_index, int fa_start)
2864 struct net *net = sock_net(cb->skb->sk);
2865 int nhsel, genid = fnhe_genid(net);
2867 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
2868 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
2869 struct fnhe_hash_bucket *bucket;
2872 if (nhc->nhc_flags & RTNH_F_DEAD)
2875 bucket = rcu_dereference(nhc->nhc_exceptions);
2879 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, genid,
2880 fa_index, fa_start);
2888 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2889 u8 ip_proto, __be16 sport,
2892 struct sk_buff *skb;
2895 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2899 /* Reserve room for dummy headers, this skb can pass
2900 * through good chunk of routing engine.
2902 skb_reset_mac_header(skb);
2903 skb_reset_network_header(skb);
2904 skb->protocol = htons(ETH_P_IP);
2905 iph = skb_put(skb, sizeof(struct iphdr));
2906 iph->protocol = ip_proto;
2912 skb_set_transport_header(skb, skb->len);
2914 switch (iph->protocol) {
2916 struct udphdr *udph;
2918 udph = skb_put_zero(skb, sizeof(struct udphdr));
2919 udph->source = sport;
2921 udph->len = sizeof(struct udphdr);
2926 struct tcphdr *tcph;
2928 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2929 tcph->source = sport;
2931 tcph->doff = sizeof(struct tcphdr) / 4;
2933 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2937 case IPPROTO_ICMP: {
2938 struct icmphdr *icmph;
2940 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2941 icmph->type = ICMP_ECHO;
2949 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
2950 const struct nlmsghdr *nlh,
2952 struct netlink_ext_ack *extack)
2957 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2958 NL_SET_ERR_MSG(extack,
2959 "ipv4: Invalid header for route get request");
2963 if (!netlink_strict_get_check(skb))
2964 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2965 rtm_ipv4_policy, extack);
2967 rtm = nlmsg_data(nlh);
2968 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2969 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2970 rtm->rtm_table || rtm->rtm_protocol ||
2971 rtm->rtm_scope || rtm->rtm_type) {
2972 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
2976 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
2977 RTM_F_LOOKUP_TABLE |
2979 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
2983 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2984 rtm_ipv4_policy, extack);
2988 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2989 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2990 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2994 for (i = 0; i <= RTA_MAX; i++) {
3010 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3018 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3019 struct netlink_ext_ack *extack)
3021 struct net *net = sock_net(in_skb->sk);
3022 struct nlattr *tb[RTA_MAX+1];
3023 u32 table_id = RT_TABLE_MAIN;
3024 __be16 sport = 0, dport = 0;
3025 struct fib_result res = {};
3026 u8 ip_proto = IPPROTO_UDP;
3027 struct rtable *rt = NULL;
3028 struct sk_buff *skb;
3030 struct flowi4 fl4 = {};
3038 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3042 rtm = nlmsg_data(nlh);
3043 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3044 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3045 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3046 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3048 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3050 uid = (iif ? INVALID_UID : current_uid());
3052 if (tb[RTA_IP_PROTO]) {
3053 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3054 &ip_proto, AF_INET, extack);
3060 sport = nla_get_be16(tb[RTA_SPORT]);
3063 dport = nla_get_be16(tb[RTA_DPORT]);
3065 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3071 fl4.flowi4_tos = rtm->rtm_tos;
3072 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3073 fl4.flowi4_mark = mark;
3074 fl4.flowi4_uid = uid;
3076 fl4.fl4_sport = sport;
3078 fl4.fl4_dport = dport;
3079 fl4.flowi4_proto = ip_proto;
3084 struct net_device *dev;
3086 dev = dev_get_by_index_rcu(net, iif);
3092 fl4.flowi4_iif = iif; /* for rt_fill_info */
3095 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
3098 rt = skb_rtable(skb);
3099 if (err == 0 && rt->dst.error)
3100 err = -rt->dst.error;
3102 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3103 skb->dev = net->loopback_dev;
3104 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3109 skb_dst_set(skb, &rt->dst);
3115 if (rtm->rtm_flags & RTM_F_NOTIFY)
3116 rt->rt_flags |= RTCF_NOTIFY;
3118 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3119 table_id = res.table ? res.table->tb_id : 0;
3121 /* reset skb for netlink reply msg */
3123 skb_reset_network_header(skb);
3124 skb_reset_transport_header(skb);
3125 skb_reset_mac_header(skb);
3127 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3129 err = fib_props[res.type].error;
3131 err = -EHOSTUNREACH;
3134 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3135 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3136 rt->rt_type, res.prefix, res.prefixlen,
3137 fl4.flowi4_tos, res.fi, 0);
3139 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3140 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
3147 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3157 void ip_rt_multicast_event(struct in_device *in_dev)
3159 rt_cache_flush(dev_net(in_dev->dev));
3162 #ifdef CONFIG_SYSCTL
3163 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3164 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3165 static int ip_rt_gc_elasticity __read_mostly = 8;
3166 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3168 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3169 void __user *buffer,
3170 size_t *lenp, loff_t *ppos)
3172 struct net *net = (struct net *)__ctl->extra1;
3175 rt_cache_flush(net);
3176 fnhe_genid_bump(net);
3183 static struct ctl_table ipv4_route_table[] = {
3185 .procname = "gc_thresh",
3186 .data = &ipv4_dst_ops.gc_thresh,
3187 .maxlen = sizeof(int),
3189 .proc_handler = proc_dointvec,
3192 .procname = "max_size",
3193 .data = &ip_rt_max_size,
3194 .maxlen = sizeof(int),
3196 .proc_handler = proc_dointvec,
3199 /* Deprecated. Use gc_min_interval_ms */
3201 .procname = "gc_min_interval",
3202 .data = &ip_rt_gc_min_interval,
3203 .maxlen = sizeof(int),
3205 .proc_handler = proc_dointvec_jiffies,
3208 .procname = "gc_min_interval_ms",
3209 .data = &ip_rt_gc_min_interval,
3210 .maxlen = sizeof(int),
3212 .proc_handler = proc_dointvec_ms_jiffies,
3215 .procname = "gc_timeout",
3216 .data = &ip_rt_gc_timeout,
3217 .maxlen = sizeof(int),
3219 .proc_handler = proc_dointvec_jiffies,
3222 .procname = "gc_interval",
3223 .data = &ip_rt_gc_interval,
3224 .maxlen = sizeof(int),
3226 .proc_handler = proc_dointvec_jiffies,
3229 .procname = "redirect_load",
3230 .data = &ip_rt_redirect_load,
3231 .maxlen = sizeof(int),
3233 .proc_handler = proc_dointvec,
3236 .procname = "redirect_number",
3237 .data = &ip_rt_redirect_number,
3238 .maxlen = sizeof(int),
3240 .proc_handler = proc_dointvec,
3243 .procname = "redirect_silence",
3244 .data = &ip_rt_redirect_silence,
3245 .maxlen = sizeof(int),
3247 .proc_handler = proc_dointvec,
3250 .procname = "error_cost",
3251 .data = &ip_rt_error_cost,
3252 .maxlen = sizeof(int),
3254 .proc_handler = proc_dointvec,
3257 .procname = "error_burst",
3258 .data = &ip_rt_error_burst,
3259 .maxlen = sizeof(int),
3261 .proc_handler = proc_dointvec,
3264 .procname = "gc_elasticity",
3265 .data = &ip_rt_gc_elasticity,
3266 .maxlen = sizeof(int),
3268 .proc_handler = proc_dointvec,
3271 .procname = "mtu_expires",
3272 .data = &ip_rt_mtu_expires,
3273 .maxlen = sizeof(int),
3275 .proc_handler = proc_dointvec_jiffies,
3278 .procname = "min_pmtu",
3279 .data = &ip_rt_min_pmtu,
3280 .maxlen = sizeof(int),
3282 .proc_handler = proc_dointvec_minmax,
3283 .extra1 = &ip_min_valid_pmtu,
3286 .procname = "min_adv_mss",
3287 .data = &ip_rt_min_advmss,
3288 .maxlen = sizeof(int),
3290 .proc_handler = proc_dointvec,
3295 static struct ctl_table ipv4_route_flush_table[] = {
3297 .procname = "flush",
3298 .maxlen = sizeof(int),
3300 .proc_handler = ipv4_sysctl_rtcache_flush,
3305 static __net_init int sysctl_route_net_init(struct net *net)
3307 struct ctl_table *tbl;
3309 tbl = ipv4_route_flush_table;
3310 if (!net_eq(net, &init_net)) {
3311 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3315 /* Don't export sysctls to unprivileged users */
3316 if (net->user_ns != &init_user_ns)
3317 tbl[0].procname = NULL;
3319 tbl[0].extra1 = net;
3321 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3322 if (!net->ipv4.route_hdr)
3327 if (tbl != ipv4_route_flush_table)
3333 static __net_exit void sysctl_route_net_exit(struct net *net)
3335 struct ctl_table *tbl;
3337 tbl = net->ipv4.route_hdr->ctl_table_arg;
3338 unregister_net_sysctl_table(net->ipv4.route_hdr);
3339 BUG_ON(tbl == ipv4_route_flush_table);
3343 static __net_initdata struct pernet_operations sysctl_route_ops = {
3344 .init = sysctl_route_net_init,
3345 .exit = sysctl_route_net_exit,
3349 static __net_init int rt_genid_init(struct net *net)
3351 atomic_set(&net->ipv4.rt_genid, 0);
3352 atomic_set(&net->fnhe_genid, 0);
3353 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3357 static __net_initdata struct pernet_operations rt_genid_ops = {
3358 .init = rt_genid_init,
3361 static int __net_init ipv4_inetpeer_init(struct net *net)
3363 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3367 inet_peer_base_init(bp);
3368 net->ipv4.peers = bp;
3372 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3374 struct inet_peer_base *bp = net->ipv4.peers;
3376 net->ipv4.peers = NULL;
3377 inetpeer_invalidate_tree(bp);
3381 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3382 .init = ipv4_inetpeer_init,
3383 .exit = ipv4_inetpeer_exit,
3386 #ifdef CONFIG_IP_ROUTE_CLASSID
3387 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3388 #endif /* CONFIG_IP_ROUTE_CLASSID */
3390 int __init ip_rt_init(void)
3394 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3397 panic("IP: failed to allocate ip_idents\n");
3399 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3401 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3403 panic("IP: failed to allocate ip_tstamps\n");
3405 for_each_possible_cpu(cpu) {
3406 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3408 INIT_LIST_HEAD(&ul->head);
3409 spin_lock_init(&ul->lock);
3411 #ifdef CONFIG_IP_ROUTE_CLASSID
3412 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3414 panic("IP: failed to allocate ip_rt_acct\n");
3417 ipv4_dst_ops.kmem_cachep =
3418 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3419 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3421 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3423 if (dst_entries_init(&ipv4_dst_ops) < 0)
3424 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3426 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3427 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3429 ipv4_dst_ops.gc_thresh = ~0;
3430 ip_rt_max_size = INT_MAX;
3435 if (ip_rt_proc_init())
3436 pr_err("Unable to create route proc files\n");
3441 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3442 RTNL_FLAG_DOIT_UNLOCKED);
3444 #ifdef CONFIG_SYSCTL
3445 register_pernet_subsys(&sysctl_route_ops);
3447 register_pernet_subsys(&rt_genid_ops);
3448 register_pernet_subsys(&ipv4_inetpeer_ops);
3452 #ifdef CONFIG_SYSCTL
3454 * We really need to sanitize the damn ipv4 init order, then all
3455 * this nonsense will go away.
3457 void __init ip_static_sysctl_init(void)
3459 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);