2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #include <linux/module.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <linux/bitops.h>
69 #include <linux/types.h>
70 #include <linux/kernel.h>
72 #include <linux/bootmem.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/workqueue.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/jhash.h>
91 #include <linux/rcupdate.h>
92 #include <linux/times.h>
93 #include <linux/slab.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/netevent.h>
107 #include <net/rtnetlink.h>
109 #include <linux/sysctl.h>
112 #define RT_FL_TOS(oldflp) \
113 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
115 #define IP_MAX_MTU 0xFFF0
117 #define RT_GC_TIMEOUT (300*HZ)
119 static int ip_rt_max_size;
120 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
121 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
122 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_gc_elasticity __read_mostly = 8;
129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly = 256;
132 static int rt_chain_length_max __read_mostly = 20;
134 static struct delayed_work expires_work;
135 static unsigned long expires_ljiffies;
138 * Interface to generic destination cache.
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static void ipv4_dst_destroy(struct dst_entry *dst);
143 static void ipv4_dst_ifdown(struct dst_entry *dst,
144 struct net_device *dev, int how);
145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146 static void ipv4_link_failure(struct sk_buff *skb);
147 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
148 static int rt_garbage_collect(struct dst_ops *ops);
151 static struct dst_ops ipv4_dst_ops = {
153 .protocol = cpu_to_be16(ETH_P_IP),
154 .gc = rt_garbage_collect,
155 .check = ipv4_dst_check,
156 .destroy = ipv4_dst_destroy,
157 .ifdown = ipv4_dst_ifdown,
158 .negative_advice = ipv4_negative_advice,
159 .link_failure = ipv4_link_failure,
160 .update_pmtu = ip_rt_update_pmtu,
161 .local_out = __ip_local_out,
162 .entries = ATOMIC_INIT(0),
165 #define ECN_OR_COST(class) TC_PRIO_##class
167 const __u8 ip_tos2prio[16] = {
171 ECN_OR_COST(BESTEFFORT),
177 ECN_OR_COST(INTERACTIVE),
179 ECN_OR_COST(INTERACTIVE),
180 TC_PRIO_INTERACTIVE_BULK,
181 ECN_OR_COST(INTERACTIVE_BULK),
182 TC_PRIO_INTERACTIVE_BULK,
183 ECN_OR_COST(INTERACTIVE_BULK)
191 /* The locking scheme is rather straight forward:
193 * 1) Read-Copy Update protects the buckets of the central route hash.
194 * 2) Only writers remove entries, and they hold the lock
195 * as they look at rtable reference counts.
196 * 3) Only readers acquire references to rtable entries,
197 * they do so with atomic increments and with the
201 struct rt_hash_bucket {
202 struct rtable *chain;
205 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
206 defined(CONFIG_PROVE_LOCKING)
208 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
209 * The size of this table is a power of two and depends on the number of CPUS.
210 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
212 #ifdef CONFIG_LOCKDEP
213 # define RT_HASH_LOCK_SZ 256
216 # define RT_HASH_LOCK_SZ 4096
218 # define RT_HASH_LOCK_SZ 2048
220 # define RT_HASH_LOCK_SZ 1024
222 # define RT_HASH_LOCK_SZ 512
224 # define RT_HASH_LOCK_SZ 256
228 static spinlock_t *rt_hash_locks;
229 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
231 static __init void rt_hash_lock_init(void)
235 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
238 panic("IP: failed to allocate rt_hash_locks\n");
240 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
241 spin_lock_init(&rt_hash_locks[i]);
244 # define rt_hash_lock_addr(slot) NULL
246 static inline void rt_hash_lock_init(void)
251 static struct rt_hash_bucket *rt_hash_table __read_mostly;
252 static unsigned rt_hash_mask __read_mostly;
253 static unsigned int rt_hash_log __read_mostly;
255 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
256 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
258 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
261 return jhash_3words((__force u32)daddr, (__force u32)saddr,
266 static inline int rt_genid(struct net *net)
268 return atomic_read(&net->ipv4.rt_genid);
271 #ifdef CONFIG_PROC_FS
272 struct rt_cache_iter_state {
273 struct seq_net_private p;
278 static struct rtable *rt_cache_get_first(struct seq_file *seq)
280 struct rt_cache_iter_state *st = seq->private;
281 struct rtable *r = NULL;
283 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
284 if (!rt_hash_table[st->bucket].chain)
287 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
289 if (dev_net(r->u.dst.dev) == seq_file_net(seq) &&
290 r->rt_genid == st->genid)
292 r = rcu_dereference_bh(r->u.dst.rt_next);
294 rcu_read_unlock_bh();
299 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
302 struct rt_cache_iter_state *st = seq->private;
304 r = r->u.dst.rt_next;
306 rcu_read_unlock_bh();
308 if (--st->bucket < 0)
310 } while (!rt_hash_table[st->bucket].chain);
312 r = rt_hash_table[st->bucket].chain;
314 return rcu_dereference_bh(r);
317 static struct rtable *rt_cache_get_next(struct seq_file *seq,
320 struct rt_cache_iter_state *st = seq->private;
321 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
322 if (dev_net(r->u.dst.dev) != seq_file_net(seq))
324 if (r->rt_genid == st->genid)
330 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
332 struct rtable *r = rt_cache_get_first(seq);
335 while (pos && (r = rt_cache_get_next(seq, r)))
337 return pos ? NULL : r;
340 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
342 struct rt_cache_iter_state *st = seq->private;
344 return rt_cache_get_idx(seq, *pos - 1);
345 st->genid = rt_genid(seq_file_net(seq));
346 return SEQ_START_TOKEN;
349 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
353 if (v == SEQ_START_TOKEN)
354 r = rt_cache_get_first(seq);
356 r = rt_cache_get_next(seq, v);
361 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
363 if (v && v != SEQ_START_TOKEN)
364 rcu_read_unlock_bh();
367 static int rt_cache_seq_show(struct seq_file *seq, void *v)
369 if (v == SEQ_START_TOKEN)
370 seq_printf(seq, "%-127s\n",
371 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
372 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
375 struct rtable *r = v;
378 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
379 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
380 r->u.dst.dev ? r->u.dst.dev->name : "*",
381 (__force u32)r->rt_dst,
382 (__force u32)r->rt_gateway,
383 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
384 r->u.dst.__use, 0, (__force u32)r->rt_src,
385 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
386 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
387 dst_metric(&r->u.dst, RTAX_WINDOW),
388 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
389 dst_metric(&r->u.dst, RTAX_RTTVAR)),
391 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
392 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
394 r->rt_spec_dst, &len);
396 seq_printf(seq, "%*s\n", 127 - len, "");
401 static const struct seq_operations rt_cache_seq_ops = {
402 .start = rt_cache_seq_start,
403 .next = rt_cache_seq_next,
404 .stop = rt_cache_seq_stop,
405 .show = rt_cache_seq_show,
408 static int rt_cache_seq_open(struct inode *inode, struct file *file)
410 return seq_open_net(inode, file, &rt_cache_seq_ops,
411 sizeof(struct rt_cache_iter_state));
414 static const struct file_operations rt_cache_seq_fops = {
415 .owner = THIS_MODULE,
416 .open = rt_cache_seq_open,
419 .release = seq_release_net,
423 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
428 return SEQ_START_TOKEN;
430 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
431 if (!cpu_possible(cpu))
434 return &per_cpu(rt_cache_stat, cpu);
439 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
443 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
444 if (!cpu_possible(cpu))
447 return &per_cpu(rt_cache_stat, cpu);
453 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
458 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
460 struct rt_cache_stat *st = v;
462 if (v == SEQ_START_TOKEN) {
463 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");
467 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
468 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
469 atomic_read(&ipv4_dst_ops.entries),
492 static const struct seq_operations rt_cpu_seq_ops = {
493 .start = rt_cpu_seq_start,
494 .next = rt_cpu_seq_next,
495 .stop = rt_cpu_seq_stop,
496 .show = rt_cpu_seq_show,
500 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
502 return seq_open(file, &rt_cpu_seq_ops);
505 static const struct file_operations rt_cpu_seq_fops = {
506 .owner = THIS_MODULE,
507 .open = rt_cpu_seq_open,
510 .release = seq_release,
513 #ifdef CONFIG_NET_CLS_ROUTE
514 static int rt_acct_proc_show(struct seq_file *m, void *v)
516 struct ip_rt_acct *dst, *src;
519 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
523 for_each_possible_cpu(i) {
524 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
525 for (j = 0; j < 256; j++) {
526 dst[j].o_bytes += src[j].o_bytes;
527 dst[j].o_packets += src[j].o_packets;
528 dst[j].i_bytes += src[j].i_bytes;
529 dst[j].i_packets += src[j].i_packets;
533 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
538 static int rt_acct_proc_open(struct inode *inode, struct file *file)
540 return single_open(file, rt_acct_proc_show, NULL);
543 static const struct file_operations rt_acct_proc_fops = {
544 .owner = THIS_MODULE,
545 .open = rt_acct_proc_open,
548 .release = single_release,
552 static int __net_init ip_rt_do_proc_init(struct net *net)
554 struct proc_dir_entry *pde;
556 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
561 pde = proc_create("rt_cache", S_IRUGO,
562 net->proc_net_stat, &rt_cpu_seq_fops);
566 #ifdef CONFIG_NET_CLS_ROUTE
567 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
573 #ifdef CONFIG_NET_CLS_ROUTE
575 remove_proc_entry("rt_cache", net->proc_net_stat);
578 remove_proc_entry("rt_cache", net->proc_net);
583 static void __net_exit ip_rt_do_proc_exit(struct net *net)
585 remove_proc_entry("rt_cache", net->proc_net_stat);
586 remove_proc_entry("rt_cache", net->proc_net);
587 #ifdef CONFIG_NET_CLS_ROUTE
588 remove_proc_entry("rt_acct", net->proc_net);
592 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
593 .init = ip_rt_do_proc_init,
594 .exit = ip_rt_do_proc_exit,
597 static int __init ip_rt_proc_init(void)
599 return register_pernet_subsys(&ip_rt_proc_ops);
603 static inline int ip_rt_proc_init(void)
607 #endif /* CONFIG_PROC_FS */
609 static inline void rt_free(struct rtable *rt)
611 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
614 static inline void rt_drop(struct rtable *rt)
617 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
620 static inline int rt_fast_clean(struct rtable *rth)
622 /* Kill broadcast/multicast entries very aggresively, if they
623 collide in hash table with more useful entries */
624 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
625 rth->fl.iif && rth->u.dst.rt_next;
628 static inline int rt_valuable(struct rtable *rth)
630 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
634 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
639 if (atomic_read(&rth->u.dst.__refcnt))
643 if (rth->u.dst.expires &&
644 time_after_eq(jiffies, rth->u.dst.expires))
647 age = jiffies - rth->u.dst.lastuse;
649 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
650 (age <= tmo2 && rt_valuable(rth)))
656 /* Bits of score are:
658 * 30: not quite useless
659 * 29..0: usage counter
661 static inline u32 rt_score(struct rtable *rt)
663 u32 score = jiffies - rt->u.dst.lastuse;
665 score = ~score & ~(3<<30);
671 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
677 static inline bool rt_caching(const struct net *net)
679 return net->ipv4.current_rt_cache_rebuild_count <=
680 net->ipv4.sysctl_rt_cache_rebuild_count;
683 static inline bool compare_hash_inputs(const struct flowi *fl1,
684 const struct flowi *fl2)
686 return ((((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
687 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
688 (fl1->iif ^ fl2->iif)) == 0);
691 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
693 return (((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
694 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
695 (fl1->mark ^ fl2->mark) |
696 (*(u16 *)&fl1->nl_u.ip4_u.tos ^ *(u16 *)&fl2->nl_u.ip4_u.tos) |
697 (fl1->oif ^ fl2->oif) |
698 (fl1->iif ^ fl2->iif)) == 0;
701 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
703 return net_eq(dev_net(rt1->u.dst.dev), dev_net(rt2->u.dst.dev));
706 static inline int rt_is_expired(struct rtable *rth)
708 return rth->rt_genid != rt_genid(dev_net(rth->u.dst.dev));
712 * Perform a full scan of hash table and free all entries.
713 * Can be called by a softirq or a process.
714 * In the later case, we want to be reschedule if necessary
716 static void rt_do_flush(int process_context)
719 struct rtable *rth, *next;
720 struct rtable * tail;
722 for (i = 0; i <= rt_hash_mask; i++) {
723 if (process_context && need_resched())
725 rth = rt_hash_table[i].chain;
729 spin_lock_bh(rt_hash_lock_addr(i));
732 struct rtable ** prev, * p;
734 rth = rt_hash_table[i].chain;
736 /* defer releasing the head of the list after spin_unlock */
737 for (tail = rth; tail; tail = tail->u.dst.rt_next)
738 if (!rt_is_expired(tail))
741 rt_hash_table[i].chain = tail;
743 /* call rt_free on entries after the tail requiring flush */
744 prev = &rt_hash_table[i].chain;
745 for (p = *prev; p; p = next) {
746 next = p->u.dst.rt_next;
747 if (!rt_is_expired(p)) {
748 prev = &p->u.dst.rt_next;
756 rth = rt_hash_table[i].chain;
757 rt_hash_table[i].chain = NULL;
760 spin_unlock_bh(rt_hash_lock_addr(i));
762 for (; rth != tail; rth = next) {
763 next = rth->u.dst.rt_next;
770 * While freeing expired entries, we compute average chain length
771 * and standard deviation, using fixed-point arithmetic.
772 * This to have an estimation of rt_chain_length_max
773 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
774 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
778 #define ONE (1UL << FRACT_BITS)
781 * Given a hash chain and an item in this hash chain,
782 * find if a previous entry has the same hash_inputs
783 * (but differs on tos, mark or oif)
784 * Returns 0 if an alias is found.
785 * Returns ONE if rth has no alias before itself.
787 static int has_noalias(const struct rtable *head, const struct rtable *rth)
789 const struct rtable *aux = head;
792 if (compare_hash_inputs(&aux->fl, &rth->fl))
794 aux = aux->u.dst.rt_next;
799 static void rt_check_expire(void)
801 static unsigned int rover;
802 unsigned int i = rover, goal;
803 struct rtable *rth, **rthp;
804 unsigned long samples = 0;
805 unsigned long sum = 0, sum2 = 0;
809 delta = jiffies - expires_ljiffies;
810 expires_ljiffies = jiffies;
811 mult = ((u64)delta) << rt_hash_log;
812 if (ip_rt_gc_timeout > 1)
813 do_div(mult, ip_rt_gc_timeout);
814 goal = (unsigned int)mult;
815 if (goal > rt_hash_mask)
816 goal = rt_hash_mask + 1;
817 for (; goal > 0; goal--) {
818 unsigned long tmo = ip_rt_gc_timeout;
819 unsigned long length;
821 i = (i + 1) & rt_hash_mask;
822 rthp = &rt_hash_table[i].chain;
832 spin_lock_bh(rt_hash_lock_addr(i));
833 while ((rth = *rthp) != NULL) {
834 prefetch(rth->u.dst.rt_next);
835 if (rt_is_expired(rth)) {
836 *rthp = rth->u.dst.rt_next;
840 if (rth->u.dst.expires) {
841 /* Entry is expired even if it is in use */
842 if (time_before_eq(jiffies, rth->u.dst.expires)) {
845 rthp = &rth->u.dst.rt_next;
847 * We only count entries on
848 * a chain with equal hash inputs once
849 * so that entries for different QOS
850 * levels, and other non-hash input
851 * attributes don't unfairly skew
852 * the length computation
854 length += has_noalias(rt_hash_table[i].chain, rth);
857 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
860 /* Cleanup aged off entries. */
861 *rthp = rth->u.dst.rt_next;
864 spin_unlock_bh(rt_hash_lock_addr(i));
866 sum2 += length*length;
869 unsigned long avg = sum / samples;
870 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
871 rt_chain_length_max = max_t(unsigned long,
873 (avg + 4*sd) >> FRACT_BITS);
879 * rt_worker_func() is run in process context.
880 * we call rt_check_expire() to scan part of the hash table
882 static void rt_worker_func(struct work_struct *work)
885 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
889 * Pertubation of rt_genid by a small quantity [1..256]
890 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
891 * many times (2^24) without giving recent rt_genid.
892 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
894 static void rt_cache_invalidate(struct net *net)
896 unsigned char shuffle;
898 get_random_bytes(&shuffle, sizeof(shuffle));
899 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
903 * delay < 0 : invalidate cache (fast : entries will be deleted later)
904 * delay >= 0 : invalidate & flush cache (can be long)
906 void rt_cache_flush(struct net *net, int delay)
908 rt_cache_invalidate(net);
910 rt_do_flush(!in_softirq());
913 /* Flush previous cache invalidated entries from the cache */
914 void rt_cache_flush_batch(void)
916 rt_do_flush(!in_softirq());
919 static void rt_emergency_hash_rebuild(struct net *net)
922 printk(KERN_WARNING "Route hash chain too long!\n");
923 rt_cache_invalidate(net);
927 Short description of GC goals.
929 We want to build algorithm, which will keep routing cache
930 at some equilibrium point, when number of aged off entries
931 is kept approximately equal to newly generated ones.
933 Current expiration strength is variable "expire".
934 We try to adjust it dynamically, so that if networking
935 is idle expires is large enough to keep enough of warm entries,
936 and when load increases it reduces to limit cache size.
939 static int rt_garbage_collect(struct dst_ops *ops)
941 static unsigned long expire = RT_GC_TIMEOUT;
942 static unsigned long last_gc;
944 static int equilibrium;
945 struct rtable *rth, **rthp;
946 unsigned long now = jiffies;
950 * Garbage collection is pretty expensive,
951 * do not make it too frequently.
954 RT_CACHE_STAT_INC(gc_total);
956 if (now - last_gc < ip_rt_gc_min_interval &&
957 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
958 RT_CACHE_STAT_INC(gc_ignored);
962 /* Calculate number of entries, which we want to expire now. */
963 goal = atomic_read(&ipv4_dst_ops.entries) -
964 (ip_rt_gc_elasticity << rt_hash_log);
966 if (equilibrium < ipv4_dst_ops.gc_thresh)
967 equilibrium = ipv4_dst_ops.gc_thresh;
968 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
970 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
971 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
974 /* We are in dangerous area. Try to reduce cache really
977 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
978 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
981 if (now - last_gc >= ip_rt_gc_min_interval)
992 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
993 unsigned long tmo = expire;
995 k = (k + 1) & rt_hash_mask;
996 rthp = &rt_hash_table[k].chain;
997 spin_lock_bh(rt_hash_lock_addr(k));
998 while ((rth = *rthp) != NULL) {
999 if (!rt_is_expired(rth) &&
1000 !rt_may_expire(rth, tmo, expire)) {
1002 rthp = &rth->u.dst.rt_next;
1005 *rthp = rth->u.dst.rt_next;
1009 spin_unlock_bh(rt_hash_lock_addr(k));
1018 /* Goal is not achieved. We stop process if:
1020 - if expire reduced to zero. Otherwise, expire is halfed.
1021 - if table is not full.
1022 - if we are called from interrupt.
1023 - jiffies check is just fallback/debug loop breaker.
1024 We will not spin here for long time in any case.
1027 RT_CACHE_STAT_INC(gc_goal_miss);
1033 #if RT_CACHE_DEBUG >= 2
1034 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1035 atomic_read(&ipv4_dst_ops.entries), goal, i);
1038 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1040 } while (!in_softirq() && time_before_eq(jiffies, now));
1042 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1044 if (net_ratelimit())
1045 printk(KERN_WARNING "dst cache overflow\n");
1046 RT_CACHE_STAT_INC(gc_dst_overflow);
1050 expire += ip_rt_gc_min_interval;
1051 if (expire > ip_rt_gc_timeout ||
1052 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
1053 expire = ip_rt_gc_timeout;
1054 #if RT_CACHE_DEBUG >= 2
1055 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1056 atomic_read(&ipv4_dst_ops.entries), goal, rover);
1062 * Returns number of entries in a hash chain that have different hash_inputs
1064 static int slow_chain_length(const struct rtable *head)
1067 const struct rtable *rth = head;
1070 length += has_noalias(head, rth);
1071 rth = rth->u.dst.rt_next;
1073 return length >> FRACT_BITS;
1076 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1077 struct rtable **rp, struct sk_buff *skb, int ifindex)
1079 struct rtable *rth, **rthp;
1081 struct rtable *cand, **candp;
1084 int attempts = !in_softirq();
1088 min_score = ~(u32)0;
1093 if (!rt_caching(dev_net(rt->u.dst.dev))) {
1095 * If we're not caching, just tell the caller we
1096 * were successful and don't touch the route. The
1097 * caller hold the sole reference to the cache entry, and
1098 * it will be released when the caller is done with it.
1099 * If we drop it here, the callers have no way to resolve routes
1100 * when we're not caching. Instead, just point *rp at rt, so
1101 * the caller gets a single use out of the route
1102 * Note that we do rt_free on this new route entry, so that
1103 * once its refcount hits zero, we are still able to reap it
1105 * Note also the rt_free uses call_rcu. We don't actually
1106 * need rcu protection here, this is just our path to get
1107 * on the route gc list.
1110 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1111 int err = arp_bind_neighbour(&rt->u.dst);
1113 if (net_ratelimit())
1115 "Neighbour table failure & not caching routes.\n");
1125 rthp = &rt_hash_table[hash].chain;
1127 spin_lock_bh(rt_hash_lock_addr(hash));
1128 while ((rth = *rthp) != NULL) {
1129 if (rt_is_expired(rth)) {
1130 *rthp = rth->u.dst.rt_next;
1134 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1136 *rthp = rth->u.dst.rt_next;
1138 * Since lookup is lockfree, the deletion
1139 * must be visible to another weakly ordered CPU before
1140 * the insertion at the start of the hash chain.
1142 rcu_assign_pointer(rth->u.dst.rt_next,
1143 rt_hash_table[hash].chain);
1145 * Since lookup is lockfree, the update writes
1146 * must be ordered for consistency on SMP.
1148 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1150 dst_use(&rth->u.dst, now);
1151 spin_unlock_bh(rt_hash_lock_addr(hash));
1157 skb_dst_set(skb, &rth->u.dst);
1161 if (!atomic_read(&rth->u.dst.__refcnt)) {
1162 u32 score = rt_score(rth);
1164 if (score <= min_score) {
1173 rthp = &rth->u.dst.rt_next;
1177 /* ip_rt_gc_elasticity used to be average length of chain
1178 * length, when exceeded gc becomes really aggressive.
1180 * The second limit is less certain. At the moment it allows
1181 * only 2 entries per bucket. We will see.
1183 if (chain_length > ip_rt_gc_elasticity) {
1184 *candp = cand->u.dst.rt_next;
1188 if (chain_length > rt_chain_length_max &&
1189 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1190 struct net *net = dev_net(rt->u.dst.dev);
1191 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1192 if (!rt_caching(net)) {
1193 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1194 rt->u.dst.dev->name, num);
1196 rt_emergency_hash_rebuild(net);
1197 spin_unlock_bh(rt_hash_lock_addr(hash));
1199 hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1200 ifindex, rt_genid(net));
1205 /* Try to bind route to arp only if it is output
1206 route or unicast forwarding path.
1208 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1209 int err = arp_bind_neighbour(&rt->u.dst);
1211 spin_unlock_bh(rt_hash_lock_addr(hash));
1213 if (err != -ENOBUFS) {
1218 /* Neighbour tables are full and nothing
1219 can be released. Try to shrink route cache,
1220 it is most likely it holds some neighbour records.
1222 if (attempts-- > 0) {
1223 int saved_elasticity = ip_rt_gc_elasticity;
1224 int saved_int = ip_rt_gc_min_interval;
1225 ip_rt_gc_elasticity = 1;
1226 ip_rt_gc_min_interval = 0;
1227 rt_garbage_collect(&ipv4_dst_ops);
1228 ip_rt_gc_min_interval = saved_int;
1229 ip_rt_gc_elasticity = saved_elasticity;
1233 if (net_ratelimit())
1234 printk(KERN_WARNING "Neighbour table overflow.\n");
1240 rt->u.dst.rt_next = rt_hash_table[hash].chain;
1242 #if RT_CACHE_DEBUG >= 2
1243 if (rt->u.dst.rt_next) {
1245 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1247 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1248 printk(" . %pI4", &trt->rt_dst);
1253 * Since lookup is lockfree, we must make sure
1254 * previous writes to rt are comitted to memory
1255 * before making rt visible to other CPUS.
1257 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1259 spin_unlock_bh(rt_hash_lock_addr(hash));
1265 skb_dst_set(skb, &rt->u.dst);
1269 void rt_bind_peer(struct rtable *rt, int create)
1271 static DEFINE_SPINLOCK(rt_peer_lock);
1272 struct inet_peer *peer;
1274 peer = inet_getpeer(rt->rt_dst, create);
1276 spin_lock_bh(&rt_peer_lock);
1277 if (rt->peer == NULL) {
1281 spin_unlock_bh(&rt_peer_lock);
1287 * Peer allocation may fail only in serious out-of-memory conditions. However
1288 * we still can generate some output.
1289 * Random ID selection looks a bit dangerous because we have no chances to
1290 * select ID being unique in a reasonable period of time.
1291 * But broken packet identifier may be better than no packet at all.
1293 static void ip_select_fb_ident(struct iphdr *iph)
1295 static DEFINE_SPINLOCK(ip_fb_id_lock);
1296 static u32 ip_fallback_id;
1299 spin_lock_bh(&ip_fb_id_lock);
1300 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1301 iph->id = htons(salt & 0xFFFF);
1302 ip_fallback_id = salt;
1303 spin_unlock_bh(&ip_fb_id_lock);
1306 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1308 struct rtable *rt = (struct rtable *) dst;
1311 if (rt->peer == NULL)
1312 rt_bind_peer(rt, 1);
1314 /* If peer is attached to destination, it is never detached,
1315 so that we need not to grab a lock to dereference it.
1318 iph->id = htons(inet_getid(rt->peer, more));
1322 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1323 __builtin_return_address(0));
1325 ip_select_fb_ident(iph);
1328 static void rt_del(unsigned hash, struct rtable *rt)
1330 struct rtable **rthp, *aux;
1332 rthp = &rt_hash_table[hash].chain;
1333 spin_lock_bh(rt_hash_lock_addr(hash));
1335 while ((aux = *rthp) != NULL) {
1336 if (aux == rt || rt_is_expired(aux)) {
1337 *rthp = aux->u.dst.rt_next;
1341 rthp = &aux->u.dst.rt_next;
1343 spin_unlock_bh(rt_hash_lock_addr(hash));
1346 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1347 __be32 saddr, struct net_device *dev)
1350 struct in_device *in_dev = in_dev_get(dev);
1351 struct rtable *rth, **rthp;
1352 __be32 skeys[2] = { saddr, 0 };
1353 int ikeys[2] = { dev->ifindex, 0 };
1354 struct netevent_redirect netevent;
1361 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1362 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1363 ipv4_is_zeronet(new_gw))
1364 goto reject_redirect;
1366 if (!rt_caching(net))
1367 goto reject_redirect;
1369 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1370 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1371 goto reject_redirect;
1372 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1373 goto reject_redirect;
1375 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1376 goto reject_redirect;
1379 for (i = 0; i < 2; i++) {
1380 for (k = 0; k < 2; k++) {
1381 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1384 rthp=&rt_hash_table[hash].chain;
1387 while ((rth = rcu_dereference(*rthp)) != NULL) {
1390 if (rth->fl.fl4_dst != daddr ||
1391 rth->fl.fl4_src != skeys[i] ||
1392 rth->fl.oif != ikeys[k] ||
1394 rt_is_expired(rth) ||
1395 !net_eq(dev_net(rth->u.dst.dev), net)) {
1396 rthp = &rth->u.dst.rt_next;
1400 if (rth->rt_dst != daddr ||
1401 rth->rt_src != saddr ||
1403 rth->rt_gateway != old_gw ||
1404 rth->u.dst.dev != dev)
1407 dst_hold(&rth->u.dst);
1410 rt = dst_alloc(&ipv4_dst_ops);
1417 /* Copy all the information. */
1419 rt->u.dst.__use = 1;
1420 atomic_set(&rt->u.dst.__refcnt, 1);
1421 rt->u.dst.child = NULL;
1423 dev_hold(rt->u.dst.dev);
1425 in_dev_hold(rt->idev);
1426 rt->u.dst.obsolete = -1;
1427 rt->u.dst.lastuse = jiffies;
1428 rt->u.dst.path = &rt->u.dst;
1429 rt->u.dst.neighbour = NULL;
1430 rt->u.dst.hh = NULL;
1432 rt->u.dst.xfrm = NULL;
1434 rt->rt_genid = rt_genid(net);
1435 rt->rt_flags |= RTCF_REDIRECTED;
1437 /* Gateway is different ... */
1438 rt->rt_gateway = new_gw;
1440 /* Redirect received -> path was valid */
1441 dst_confirm(&rth->u.dst);
1444 atomic_inc(&rt->peer->refcnt);
1446 if (arp_bind_neighbour(&rt->u.dst) ||
1447 !(rt->u.dst.neighbour->nud_state &
1449 if (rt->u.dst.neighbour)
1450 neigh_event_send(rt->u.dst.neighbour, NULL);
1456 netevent.old = &rth->u.dst;
1457 netevent.new = &rt->u.dst;
1458 call_netevent_notifiers(NETEVENT_REDIRECT,
1462 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1475 #ifdef CONFIG_IP_ROUTE_VERBOSE
1476 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1477 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1478 " Advised path = %pI4 -> %pI4\n",
1479 &old_gw, dev->name, &new_gw,
1485 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1487 struct rtable *rt = (struct rtable *)dst;
1488 struct dst_entry *ret = dst;
1491 if (dst->obsolete > 0) {
1494 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1495 (rt->u.dst.expires &&
1496 time_after_eq(jiffies, rt->u.dst.expires))) {
1497 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1499 rt_genid(dev_net(dst->dev)));
1500 #if RT_CACHE_DEBUG >= 1
1501 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1502 &rt->rt_dst, rt->fl.fl4_tos);
1513 * 1. The first ip_rt_redirect_number redirects are sent
1514 * with exponential backoff, then we stop sending them at all,
1515 * assuming that the host ignores our redirects.
1516 * 2. If we did not see packets requiring redirects
1517 * during ip_rt_redirect_silence, we assume that the host
1518 * forgot redirected route and start to send redirects again.
1520 * This algorithm is much cheaper and more intelligent than dumb load limiting
1523 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1524 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1527 void ip_rt_send_redirect(struct sk_buff *skb)
1529 struct rtable *rt = skb_rtable(skb);
1530 struct in_device *in_dev;
1534 in_dev = __in_dev_get_rcu(rt->u.dst.dev);
1535 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1539 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1542 /* No redirected packets during ip_rt_redirect_silence;
1543 * reset the algorithm.
1545 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1546 rt->u.dst.rate_tokens = 0;
1548 /* Too many ignored redirects; do not send anything
1549 * set u.dst.rate_last to the last seen redirected packet.
1551 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1552 rt->u.dst.rate_last = jiffies;
1556 /* Check for load limit; set rate_last to the latest sent
1559 if (rt->u.dst.rate_tokens == 0 ||
1561 (rt->u.dst.rate_last +
1562 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1563 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1564 rt->u.dst.rate_last = jiffies;
1565 ++rt->u.dst.rate_tokens;
1566 #ifdef CONFIG_IP_ROUTE_VERBOSE
1568 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1570 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1571 &rt->rt_src, rt->rt_iif,
1572 &rt->rt_dst, &rt->rt_gateway);
1577 static int ip_error(struct sk_buff *skb)
1579 struct rtable *rt = skb_rtable(skb);
1583 switch (rt->u.dst.error) {
1588 code = ICMP_HOST_UNREACH;
1591 code = ICMP_NET_UNREACH;
1592 IP_INC_STATS_BH(dev_net(rt->u.dst.dev),
1593 IPSTATS_MIB_INNOROUTES);
1596 code = ICMP_PKT_FILTERED;
1601 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1602 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1603 rt->u.dst.rate_tokens = ip_rt_error_burst;
1604 rt->u.dst.rate_last = now;
1605 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1606 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1607 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1610 out: kfree_skb(skb);
1615 * The last two values are not from the RFC but
1616 * are needed for AMPRnet AX.25 paths.
1619 static const unsigned short mtu_plateau[] =
1620 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1622 static inline unsigned short guess_mtu(unsigned short old_mtu)
1626 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1627 if (old_mtu > mtu_plateau[i])
1628 return mtu_plateau[i];
1632 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1633 unsigned short new_mtu,
1634 struct net_device *dev)
1637 unsigned short old_mtu = ntohs(iph->tot_len);
1639 int ikeys[2] = { dev->ifindex, 0 };
1640 __be32 skeys[2] = { iph->saddr, 0, };
1641 __be32 daddr = iph->daddr;
1642 unsigned short est_mtu = 0;
1644 for (k = 0; k < 2; k++) {
1645 for (i = 0; i < 2; i++) {
1646 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1650 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1651 rth = rcu_dereference(rth->u.dst.rt_next)) {
1652 unsigned short mtu = new_mtu;
1654 if (rth->fl.fl4_dst != daddr ||
1655 rth->fl.fl4_src != skeys[i] ||
1656 rth->rt_dst != daddr ||
1657 rth->rt_src != iph->saddr ||
1658 rth->fl.oif != ikeys[k] ||
1660 dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1661 !net_eq(dev_net(rth->u.dst.dev), net) ||
1665 if (new_mtu < 68 || new_mtu >= old_mtu) {
1667 /* BSD 4.2 compatibility hack :-( */
1669 old_mtu >= dst_mtu(&rth->u.dst) &&
1670 old_mtu >= 68 + (iph->ihl << 2))
1671 old_mtu -= iph->ihl << 2;
1673 mtu = guess_mtu(old_mtu);
1675 if (mtu <= dst_mtu(&rth->u.dst)) {
1676 if (mtu < dst_mtu(&rth->u.dst)) {
1677 dst_confirm(&rth->u.dst);
1678 if (mtu < ip_rt_min_pmtu) {
1679 mtu = ip_rt_min_pmtu;
1680 rth->u.dst.metrics[RTAX_LOCK-1] |=
1683 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1684 dst_set_expires(&rth->u.dst,
1693 return est_mtu ? : new_mtu;
1696 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1698 if (dst_mtu(dst) > mtu && mtu >= 68 &&
1699 !(dst_metric_locked(dst, RTAX_MTU))) {
1700 if (mtu < ip_rt_min_pmtu) {
1701 mtu = ip_rt_min_pmtu;
1702 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1704 dst->metrics[RTAX_MTU-1] = mtu;
1705 dst_set_expires(dst, ip_rt_mtu_expires);
1706 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1710 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1712 if (rt_is_expired((struct rtable *)dst))
1717 static void ipv4_dst_destroy(struct dst_entry *dst)
1719 struct rtable *rt = (struct rtable *) dst;
1720 struct inet_peer *peer = rt->peer;
1721 struct in_device *idev = rt->idev;
1734 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1737 struct rtable *rt = (struct rtable *) dst;
1738 struct in_device *idev = rt->idev;
1739 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1740 struct in_device *loopback_idev =
1741 in_dev_get(dev_net(dev)->loopback_dev);
1742 if (loopback_idev) {
1743 rt->idev = loopback_idev;
1749 static void ipv4_link_failure(struct sk_buff *skb)
1753 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1755 rt = skb_rtable(skb);
1757 dst_set_expires(&rt->u.dst, 0);
1760 static int ip_rt_bug(struct sk_buff *skb)
1762 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1763 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1764 skb->dev ? skb->dev->name : "?");
1770 We do not cache source address of outgoing interface,
1771 because it is used only by IP RR, TS and SRR options,
1772 so that it out of fast path.
1774 BTW remember: "addr" is allowed to be not aligned
1778 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1781 struct fib_result res;
1783 if (rt->fl.iif == 0)
1785 else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1786 src = FIB_RES_PREFSRC(res);
1789 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1791 memcpy(addr, &src, 4);
1794 #ifdef CONFIG_NET_CLS_ROUTE
1795 static void set_class_tag(struct rtable *rt, u32 tag)
1797 if (!(rt->u.dst.tclassid & 0xFFFF))
1798 rt->u.dst.tclassid |= tag & 0xFFFF;
1799 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1800 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1804 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1806 struct fib_info *fi = res->fi;
1809 if (FIB_RES_GW(*res) &&
1810 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1811 rt->rt_gateway = FIB_RES_GW(*res);
1812 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1813 sizeof(rt->u.dst.metrics));
1814 if (fi->fib_mtu == 0) {
1815 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1816 if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1817 rt->rt_gateway != rt->rt_dst &&
1818 rt->u.dst.dev->mtu > 576)
1819 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1821 #ifdef CONFIG_NET_CLS_ROUTE
1822 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1825 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1827 if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1828 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1829 if (dst_mtu(&rt->u.dst) > IP_MAX_MTU)
1830 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1831 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1832 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1834 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1835 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1837 #ifdef CONFIG_NET_CLS_ROUTE
1838 #ifdef CONFIG_IP_MULTIPLE_TABLES
1839 set_class_tag(rt, fib_rules_tclass(res));
1841 set_class_tag(rt, itag);
1843 rt->rt_type = res->type;
1846 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1847 u8 tos, struct net_device *dev, int our)
1852 struct in_device *in_dev = in_dev_get(dev);
1856 /* Primary sanity checks. */
1861 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1862 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1865 if (ipv4_is_zeronet(saddr)) {
1866 if (!ipv4_is_local_multicast(daddr))
1868 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1870 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1875 rth = dst_alloc(&ipv4_dst_ops);
1879 rth->u.dst.output = ip_rt_bug;
1880 rth->u.dst.obsolete = -1;
1882 atomic_set(&rth->u.dst.__refcnt, 1);
1883 rth->u.dst.flags= DST_HOST;
1884 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1885 rth->u.dst.flags |= DST_NOPOLICY;
1886 rth->fl.fl4_dst = daddr;
1887 rth->rt_dst = daddr;
1888 rth->fl.fl4_tos = tos;
1889 rth->fl.mark = skb->mark;
1890 rth->fl.fl4_src = saddr;
1891 rth->rt_src = saddr;
1892 #ifdef CONFIG_NET_CLS_ROUTE
1893 rth->u.dst.tclassid = itag;
1896 rth->fl.iif = dev->ifindex;
1897 rth->u.dst.dev = init_net.loopback_dev;
1898 dev_hold(rth->u.dst.dev);
1899 rth->idev = in_dev_get(rth->u.dst.dev);
1901 rth->rt_gateway = daddr;
1902 rth->rt_spec_dst= spec_dst;
1903 rth->rt_genid = rt_genid(dev_net(dev));
1904 rth->rt_flags = RTCF_MULTICAST;
1905 rth->rt_type = RTN_MULTICAST;
1907 rth->u.dst.input= ip_local_deliver;
1908 rth->rt_flags |= RTCF_LOCAL;
1911 #ifdef CONFIG_IP_MROUTE
1912 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1913 rth->u.dst.input = ip_mr_input;
1915 RT_CACHE_STAT_INC(in_slow_mc);
1918 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1919 return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1933 static void ip_handle_martian_source(struct net_device *dev,
1934 struct in_device *in_dev,
1935 struct sk_buff *skb,
1939 RT_CACHE_STAT_INC(in_martian_src);
1940 #ifdef CONFIG_IP_ROUTE_VERBOSE
1941 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1943 * RFC1812 recommendation, if source is martian,
1944 * the only hint is MAC header.
1946 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1947 &daddr, &saddr, dev->name);
1948 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1950 const unsigned char *p = skb_mac_header(skb);
1951 printk(KERN_WARNING "ll header: ");
1952 for (i = 0; i < dev->hard_header_len; i++, p++) {
1954 if (i < (dev->hard_header_len - 1))
1963 static int __mkroute_input(struct sk_buff *skb,
1964 struct fib_result *res,
1965 struct in_device *in_dev,
1966 __be32 daddr, __be32 saddr, u32 tos,
1967 struct rtable **result)
1972 struct in_device *out_dev;
1977 /* get a working reference to the output device */
1978 out_dev = in_dev_get(FIB_RES_DEV(*res));
1979 if (out_dev == NULL) {
1980 if (net_ratelimit())
1981 printk(KERN_CRIT "Bug in ip_route_input" \
1982 "_slow(). Please, report\n");
1987 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1988 in_dev->dev, &spec_dst, &itag, skb->mark);
1990 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1997 flags |= RTCF_DIRECTSRC;
1999 if (out_dev == in_dev && err &&
2000 (IN_DEV_SHARED_MEDIA(out_dev) ||
2001 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2002 flags |= RTCF_DOREDIRECT;
2004 if (skb->protocol != htons(ETH_P_IP)) {
2005 /* Not IP (i.e. ARP). Do not create route, if it is
2006 * invalid for proxy arp. DNAT routes are always valid.
2008 * Proxy arp feature have been extended to allow, ARP
2009 * replies back to the same interface, to support
2010 * Private VLAN switch technologies. See arp.c.
2012 if (out_dev == in_dev &&
2013 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2020 rth = dst_alloc(&ipv4_dst_ops);
2026 atomic_set(&rth->u.dst.__refcnt, 1);
2027 rth->u.dst.flags= DST_HOST;
2028 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2029 rth->u.dst.flags |= DST_NOPOLICY;
2030 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2031 rth->u.dst.flags |= DST_NOXFRM;
2032 rth->fl.fl4_dst = daddr;
2033 rth->rt_dst = daddr;
2034 rth->fl.fl4_tos = tos;
2035 rth->fl.mark = skb->mark;
2036 rth->fl.fl4_src = saddr;
2037 rth->rt_src = saddr;
2038 rth->rt_gateway = daddr;
2040 rth->fl.iif = in_dev->dev->ifindex;
2041 rth->u.dst.dev = (out_dev)->dev;
2042 dev_hold(rth->u.dst.dev);
2043 rth->idev = in_dev_get(rth->u.dst.dev);
2045 rth->rt_spec_dst= spec_dst;
2047 rth->u.dst.obsolete = -1;
2048 rth->u.dst.input = ip_forward;
2049 rth->u.dst.output = ip_output;
2050 rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
2052 rt_set_nexthop(rth, res, itag);
2054 rth->rt_flags = flags;
2059 /* release the working reference to the output device */
2060 in_dev_put(out_dev);
2064 static int ip_mkroute_input(struct sk_buff *skb,
2065 struct fib_result *res,
2066 const struct flowi *fl,
2067 struct in_device *in_dev,
2068 __be32 daddr, __be32 saddr, u32 tos)
2070 struct rtable* rth = NULL;
2074 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2075 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2076 fib_select_multipath(fl, res);
2079 /* create a routing cache entry */
2080 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2084 /* put it into the cache */
2085 hash = rt_hash(daddr, saddr, fl->iif,
2086 rt_genid(dev_net(rth->u.dst.dev)));
2087 return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2091 * NOTE. We drop all the packets that has local source
2092 * addresses, because every properly looped back packet
2093 * must have correct destination already attached by output routine.
2095 * Such approach solves two big problems:
2096 * 1. Not simplex devices are handled properly.
2097 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2100 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2101 u8 tos, struct net_device *dev)
2103 struct fib_result res;
2104 struct in_device *in_dev = in_dev_get(dev);
2105 struct flowi fl = { .nl_u = { .ip4_u =
2109 .scope = RT_SCOPE_UNIVERSE,
2112 .iif = dev->ifindex };
2115 struct rtable * rth;
2120 struct net * net = dev_net(dev);
2122 /* IP on this device is disabled. */
2127 /* Check for the most weird martians, which can be not detected
2131 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2132 ipv4_is_loopback(saddr))
2133 goto martian_source;
2135 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2138 /* Accept zero addresses only to limited broadcast;
2139 * I even do not know to fix it or not. Waiting for complains :-)
2141 if (ipv4_is_zeronet(saddr))
2142 goto martian_source;
2144 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2145 ipv4_is_loopback(daddr))
2146 goto martian_destination;
2149 * Now we are ready to route packet.
2151 if ((err = fib_lookup(net, &fl, &res)) != 0) {
2152 if (!IN_DEV_FORWARD(in_dev))
2158 RT_CACHE_STAT_INC(in_slow_tot);
2160 if (res.type == RTN_BROADCAST)
2163 if (res.type == RTN_LOCAL) {
2164 err = fib_validate_source(saddr, daddr, tos,
2165 net->loopback_dev->ifindex,
2166 dev, &spec_dst, &itag, skb->mark);
2168 goto martian_source_keep_err;
2170 flags |= RTCF_DIRECTSRC;
2175 if (!IN_DEV_FORWARD(in_dev))
2177 if (res.type != RTN_UNICAST)
2178 goto martian_destination;
2180 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2188 if (skb->protocol != htons(ETH_P_IP))
2191 if (ipv4_is_zeronet(saddr))
2192 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2194 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2197 goto martian_source_keep_err;
2199 flags |= RTCF_DIRECTSRC;
2201 flags |= RTCF_BROADCAST;
2202 res.type = RTN_BROADCAST;
2203 RT_CACHE_STAT_INC(in_brd);
2206 rth = dst_alloc(&ipv4_dst_ops);
2210 rth->u.dst.output= ip_rt_bug;
2211 rth->u.dst.obsolete = -1;
2212 rth->rt_genid = rt_genid(net);
2214 atomic_set(&rth->u.dst.__refcnt, 1);
2215 rth->u.dst.flags= DST_HOST;
2216 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2217 rth->u.dst.flags |= DST_NOPOLICY;
2218 rth->fl.fl4_dst = daddr;
2219 rth->rt_dst = daddr;
2220 rth->fl.fl4_tos = tos;
2221 rth->fl.mark = skb->mark;
2222 rth->fl.fl4_src = saddr;
2223 rth->rt_src = saddr;
2224 #ifdef CONFIG_NET_CLS_ROUTE
2225 rth->u.dst.tclassid = itag;
2228 rth->fl.iif = dev->ifindex;
2229 rth->u.dst.dev = net->loopback_dev;
2230 dev_hold(rth->u.dst.dev);
2231 rth->idev = in_dev_get(rth->u.dst.dev);
2232 rth->rt_gateway = daddr;
2233 rth->rt_spec_dst= spec_dst;
2234 rth->u.dst.input= ip_local_deliver;
2235 rth->rt_flags = flags|RTCF_LOCAL;
2236 if (res.type == RTN_UNREACHABLE) {
2237 rth->u.dst.input= ip_error;
2238 rth->u.dst.error= -err;
2239 rth->rt_flags &= ~RTCF_LOCAL;
2241 rth->rt_type = res.type;
2242 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2243 err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2247 RT_CACHE_STAT_INC(in_no_route);
2248 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2249 res.type = RTN_UNREACHABLE;
2255 * Do not cache martian addresses: they should be logged (RFC1812)
2257 martian_destination:
2258 RT_CACHE_STAT_INC(in_martian_dst);
2259 #ifdef CONFIG_IP_ROUTE_VERBOSE
2260 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2261 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2262 &daddr, &saddr, dev->name);
2266 err = -EHOSTUNREACH;
2279 martian_source_keep_err:
2280 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2284 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2285 u8 tos, struct net_device *dev, bool noref)
2287 struct rtable * rth;
2289 int iif = dev->ifindex;
2294 if (!rt_caching(net))
2297 tos &= IPTOS_RT_MASK;
2298 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2301 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2302 rth = rcu_dereference(rth->u.dst.rt_next)) {
2303 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2304 ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2305 (rth->fl.iif ^ iif) |
2307 (rth->fl.fl4_tos ^ tos)) == 0 &&
2308 rth->fl.mark == skb->mark &&
2309 net_eq(dev_net(rth->u.dst.dev), net) &&
2310 !rt_is_expired(rth)) {
2312 dst_use_noref(&rth->u.dst, jiffies);
2313 skb_dst_set_noref(skb, &rth->u.dst);
2315 dst_use(&rth->u.dst, jiffies);
2316 skb_dst_set(skb, &rth->u.dst);
2318 RT_CACHE_STAT_INC(in_hit);
2322 RT_CACHE_STAT_INC(in_hlist_search);
2327 /* Multicast recognition logic is moved from route cache to here.
2328 The problem was that too many Ethernet cards have broken/missing
2329 hardware multicast filters :-( As result the host on multicasting
2330 network acquires a lot of useless route cache entries, sort of
2331 SDR messages from all the world. Now we try to get rid of them.
2332 Really, provided software IP multicast filter is organized
2333 reasonably (at least, hashed), it does not result in a slowdown
2334 comparing with route cache reject entries.
2335 Note, that multicast routers are not affected, because
2336 route cache entry is created eventually.
2338 if (ipv4_is_multicast(daddr)) {
2339 struct in_device *in_dev;
2342 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2343 int our = ip_check_mc(in_dev, daddr, saddr,
2344 ip_hdr(skb)->protocol);
2346 #ifdef CONFIG_IP_MROUTE
2348 (!ipv4_is_local_multicast(daddr) &&
2349 IN_DEV_MFORWARD(in_dev))
2353 return ip_route_input_mc(skb, daddr, saddr,
2360 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2362 EXPORT_SYMBOL(ip_route_input_common);
2364 static int __mkroute_output(struct rtable **result,
2365 struct fib_result *res,
2366 const struct flowi *fl,
2367 const struct flowi *oldflp,
2368 struct net_device *dev_out,
2372 struct in_device *in_dev;
2373 u32 tos = RT_FL_TOS(oldflp);
2376 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2379 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2380 res->type = RTN_BROADCAST;
2381 else if (ipv4_is_multicast(fl->fl4_dst))
2382 res->type = RTN_MULTICAST;
2383 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2386 if (dev_out->flags & IFF_LOOPBACK)
2387 flags |= RTCF_LOCAL;
2389 /* get work reference to inet device */
2390 in_dev = in_dev_get(dev_out);
2394 if (res->type == RTN_BROADCAST) {
2395 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2397 fib_info_put(res->fi);
2400 } else if (res->type == RTN_MULTICAST) {
2401 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2402 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2404 flags &= ~RTCF_LOCAL;
2405 /* If multicast route do not exist use
2406 default one, but do not gateway in this case.
2409 if (res->fi && res->prefixlen < 4) {
2410 fib_info_put(res->fi);
2416 rth = dst_alloc(&ipv4_dst_ops);
2422 atomic_set(&rth->u.dst.__refcnt, 1);
2423 rth->u.dst.flags= DST_HOST;
2424 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2425 rth->u.dst.flags |= DST_NOXFRM;
2426 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2427 rth->u.dst.flags |= DST_NOPOLICY;
2429 rth->fl.fl4_dst = oldflp->fl4_dst;
2430 rth->fl.fl4_tos = tos;
2431 rth->fl.fl4_src = oldflp->fl4_src;
2432 rth->fl.oif = oldflp->oif;
2433 rth->fl.mark = oldflp->mark;
2434 rth->rt_dst = fl->fl4_dst;
2435 rth->rt_src = fl->fl4_src;
2436 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2437 /* get references to the devices that are to be hold by the routing
2439 rth->u.dst.dev = dev_out;
2441 rth->idev = in_dev_get(dev_out);
2442 rth->rt_gateway = fl->fl4_dst;
2443 rth->rt_spec_dst= fl->fl4_src;
2445 rth->u.dst.output=ip_output;
2446 rth->u.dst.obsolete = -1;
2447 rth->rt_genid = rt_genid(dev_net(dev_out));
2449 RT_CACHE_STAT_INC(out_slow_tot);
2451 if (flags & RTCF_LOCAL) {
2452 rth->u.dst.input = ip_local_deliver;
2453 rth->rt_spec_dst = fl->fl4_dst;
2455 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2456 rth->rt_spec_dst = fl->fl4_src;
2457 if (flags & RTCF_LOCAL &&
2458 !(dev_out->flags & IFF_LOOPBACK)) {
2459 rth->u.dst.output = ip_mc_output;
2460 RT_CACHE_STAT_INC(out_slow_mc);
2462 #ifdef CONFIG_IP_MROUTE
2463 if (res->type == RTN_MULTICAST) {
2464 if (IN_DEV_MFORWARD(in_dev) &&
2465 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2466 rth->u.dst.input = ip_mr_input;
2467 rth->u.dst.output = ip_mc_output;
2473 rt_set_nexthop(rth, res, 0);
2475 rth->rt_flags = flags;
2479 /* release work reference to inet device */
2485 static int ip_mkroute_output(struct rtable **rp,
2486 struct fib_result *res,
2487 const struct flowi *fl,
2488 const struct flowi *oldflp,
2489 struct net_device *dev_out,
2492 struct rtable *rth = NULL;
2493 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2496 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2497 rt_genid(dev_net(dev_out)));
2498 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2505 * Major route resolver routine.
2508 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2509 const struct flowi *oldflp)
2511 u32 tos = RT_FL_TOS(oldflp);
2512 struct flowi fl = { .nl_u = { .ip4_u =
2513 { .daddr = oldflp->fl4_dst,
2514 .saddr = oldflp->fl4_src,
2515 .tos = tos & IPTOS_RT_MASK,
2516 .scope = ((tos & RTO_ONLINK) ?
2520 .mark = oldflp->mark,
2521 .iif = net->loopback_dev->ifindex,
2522 .oif = oldflp->oif };
2523 struct fib_result res;
2525 struct net_device *dev_out = NULL;
2531 #ifdef CONFIG_IP_MULTIPLE_TABLES
2535 if (oldflp->fl4_src) {
2537 if (ipv4_is_multicast(oldflp->fl4_src) ||
2538 ipv4_is_lbcast(oldflp->fl4_src) ||
2539 ipv4_is_zeronet(oldflp->fl4_src))
2542 /* I removed check for oif == dev_out->oif here.
2543 It was wrong for two reasons:
2544 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2545 is assigned to multiple interfaces.
2546 2. Moreover, we are allowed to send packets with saddr
2547 of another iface. --ANK
2550 if (oldflp->oif == 0 &&
2551 (ipv4_is_multicast(oldflp->fl4_dst) ||
2552 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2553 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2554 dev_out = ip_dev_find(net, oldflp->fl4_src);
2555 if (dev_out == NULL)
2558 /* Special hack: user can direct multicasts
2559 and limited broadcast via necessary interface
2560 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2561 This hack is not just for fun, it allows
2562 vic,vat and friends to work.
2563 They bind socket to loopback, set ttl to zero
2564 and expect that it will work.
2565 From the viewpoint of routing cache they are broken,
2566 because we are not allowed to build multicast path
2567 with loopback source addr (look, routing cache
2568 cannot know, that ttl is zero, so that packet
2569 will not leave this host and route is valid).
2570 Luckily, this hack is good workaround.
2573 fl.oif = dev_out->ifindex;
2577 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2578 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2579 dev_out = ip_dev_find(net, oldflp->fl4_src);
2580 if (dev_out == NULL)
2589 dev_out = dev_get_by_index(net, oldflp->oif);
2591 if (dev_out == NULL)
2594 /* RACE: Check return value of inet_select_addr instead. */
2595 if (__in_dev_get_rtnl(dev_out) == NULL) {
2597 goto out; /* Wrong error code */
2600 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2601 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2603 fl.fl4_src = inet_select_addr(dev_out, 0,
2608 if (ipv4_is_multicast(oldflp->fl4_dst))
2609 fl.fl4_src = inet_select_addr(dev_out, 0,
2611 else if (!oldflp->fl4_dst)
2612 fl.fl4_src = inet_select_addr(dev_out, 0,
2618 fl.fl4_dst = fl.fl4_src;
2620 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2623 dev_out = net->loopback_dev;
2625 fl.oif = net->loopback_dev->ifindex;
2626 res.type = RTN_LOCAL;
2627 flags |= RTCF_LOCAL;
2631 if (fib_lookup(net, &fl, &res)) {
2634 /* Apparently, routing tables are wrong. Assume,
2635 that the destination is on link.
2638 Because we are allowed to send to iface
2639 even if it has NO routes and NO assigned
2640 addresses. When oif is specified, routing
2641 tables are looked up with only one purpose:
2642 to catch if destination is gatewayed, rather than
2643 direct. Moreover, if MSG_DONTROUTE is set,
2644 we send packet, ignoring both routing tables
2645 and ifaddr state. --ANK
2648 We could make it even if oif is unknown,
2649 likely IPv6, but we do not.
2652 if (fl.fl4_src == 0)
2653 fl.fl4_src = inet_select_addr(dev_out, 0,
2655 res.type = RTN_UNICAST;
2665 if (res.type == RTN_LOCAL) {
2667 fl.fl4_src = fl.fl4_dst;
2670 dev_out = net->loopback_dev;
2672 fl.oif = dev_out->ifindex;
2674 fib_info_put(res.fi);
2676 flags |= RTCF_LOCAL;
2680 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2681 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2682 fib_select_multipath(&fl, &res);
2685 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2686 fib_select_default(net, &fl, &res);
2689 fl.fl4_src = FIB_RES_PREFSRC(res);
2693 dev_out = FIB_RES_DEV(res);
2695 fl.oif = dev_out->ifindex;
2699 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2709 int __ip_route_output_key(struct net *net, struct rtable **rp,
2710 const struct flowi *flp)
2715 if (!rt_caching(net))
2718 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2721 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2722 rth = rcu_dereference_bh(rth->u.dst.rt_next)) {
2723 if (rth->fl.fl4_dst == flp->fl4_dst &&
2724 rth->fl.fl4_src == flp->fl4_src &&
2726 rth->fl.oif == flp->oif &&
2727 rth->fl.mark == flp->mark &&
2728 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2729 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2730 net_eq(dev_net(rth->u.dst.dev), net) &&
2731 !rt_is_expired(rth)) {
2732 dst_use(&rth->u.dst, jiffies);
2733 RT_CACHE_STAT_INC(out_hit);
2734 rcu_read_unlock_bh();
2738 RT_CACHE_STAT_INC(out_hlist_search);
2740 rcu_read_unlock_bh();
2743 return ip_route_output_slow(net, rp, flp);
2746 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2748 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2752 static struct dst_ops ipv4_dst_blackhole_ops = {
2754 .protocol = cpu_to_be16(ETH_P_IP),
2755 .destroy = ipv4_dst_destroy,
2756 .check = ipv4_dst_check,
2757 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2758 .entries = ATOMIC_INIT(0),
2762 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2764 struct rtable *ort = *rp;
2765 struct rtable *rt = (struct rtable *)
2766 dst_alloc(&ipv4_dst_blackhole_ops);
2769 struct dst_entry *new = &rt->u.dst;
2771 atomic_set(&new->__refcnt, 1);
2773 new->input = dst_discard;
2774 new->output = dst_discard;
2775 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2777 new->dev = ort->u.dst.dev;
2783 rt->idev = ort->idev;
2785 in_dev_hold(rt->idev);
2786 rt->rt_genid = rt_genid(net);
2787 rt->rt_flags = ort->rt_flags;
2788 rt->rt_type = ort->rt_type;
2789 rt->rt_dst = ort->rt_dst;
2790 rt->rt_src = ort->rt_src;
2791 rt->rt_iif = ort->rt_iif;
2792 rt->rt_gateway = ort->rt_gateway;
2793 rt->rt_spec_dst = ort->rt_spec_dst;
2794 rt->peer = ort->peer;
2796 atomic_inc(&rt->peer->refcnt);
2801 dst_release(&(*rp)->u.dst);
2803 return (rt ? 0 : -ENOMEM);
2806 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2807 struct sock *sk, int flags)
2811 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2816 flp->fl4_src = (*rp)->rt_src;
2818 flp->fl4_dst = (*rp)->rt_dst;
2819 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2820 flags ? XFRM_LOOKUP_WAIT : 0);
2821 if (err == -EREMOTE)
2822 err = ipv4_dst_blackhole(net, rp, flp);
2830 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2832 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2834 return ip_route_output_flow(net, rp, flp, NULL, 0);
2837 static int rt_fill_info(struct net *net,
2838 struct sk_buff *skb, u32 pid, u32 seq, int event,
2839 int nowait, unsigned int flags)
2841 struct rtable *rt = skb_rtable(skb);
2843 struct nlmsghdr *nlh;
2845 u32 id = 0, ts = 0, tsage = 0, error;
2847 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2851 r = nlmsg_data(nlh);
2852 r->rtm_family = AF_INET;
2853 r->rtm_dst_len = 32;
2855 r->rtm_tos = rt->fl.fl4_tos;
2856 r->rtm_table = RT_TABLE_MAIN;
2857 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2858 r->rtm_type = rt->rt_type;
2859 r->rtm_scope = RT_SCOPE_UNIVERSE;
2860 r->rtm_protocol = RTPROT_UNSPEC;
2861 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2862 if (rt->rt_flags & RTCF_NOTIFY)
2863 r->rtm_flags |= RTM_F_NOTIFY;
2865 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2867 if (rt->fl.fl4_src) {
2868 r->rtm_src_len = 32;
2869 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2872 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2873 #ifdef CONFIG_NET_CLS_ROUTE
2874 if (rt->u.dst.tclassid)
2875 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2878 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2879 else if (rt->rt_src != rt->fl.fl4_src)
2880 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2882 if (rt->rt_dst != rt->rt_gateway)
2883 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2885 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2886 goto nla_put_failure;
2888 error = rt->u.dst.error;
2889 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2891 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2892 if (rt->peer->tcp_ts_stamp) {
2893 ts = rt->peer->tcp_ts;
2894 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2899 #ifdef CONFIG_IP_MROUTE
2900 __be32 dst = rt->rt_dst;
2902 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2903 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2904 int err = ipmr_get_route(net, skb, r, nowait);
2909 goto nla_put_failure;
2911 if (err == -EMSGSIZE)
2912 goto nla_put_failure;
2918 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2921 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2922 expires, error) < 0)
2923 goto nla_put_failure;
2925 return nlmsg_end(skb, nlh);
2928 nlmsg_cancel(skb, nlh);
2932 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2934 struct net *net = sock_net(in_skb->sk);
2936 struct nlattr *tb[RTA_MAX+1];
2937 struct rtable *rt = NULL;
2942 struct sk_buff *skb;
2944 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2948 rtm = nlmsg_data(nlh);
2950 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2956 /* Reserve room for dummy headers, this skb can pass
2957 through good chunk of routing engine.
2959 skb_reset_mac_header(skb);
2960 skb_reset_network_header(skb);
2962 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2963 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2964 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2966 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2967 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2968 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2971 struct net_device *dev;
2973 dev = __dev_get_by_index(net, iif);
2979 skb->protocol = htons(ETH_P_IP);
2982 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2985 rt = skb_rtable(skb);
2986 if (err == 0 && rt->u.dst.error)
2987 err = -rt->u.dst.error;
2994 .tos = rtm->rtm_tos,
2997 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2999 err = ip_route_output_key(net, &rt, &fl);
3005 skb_dst_set(skb, &rt->u.dst);
3006 if (rtm->rtm_flags & RTM_F_NOTIFY)
3007 rt->rt_flags |= RTCF_NOTIFY;
3009 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3010 RTM_NEWROUTE, 0, 0);
3014 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3023 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3030 net = sock_net(skb->sk);
3035 s_idx = idx = cb->args[1];
3036 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3037 if (!rt_hash_table[h].chain)
3040 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3041 rt = rcu_dereference_bh(rt->u.dst.rt_next), idx++) {
3042 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
3044 if (rt_is_expired(rt))
3046 skb_dst_set_noref(skb, &rt->u.dst);
3047 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3048 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3049 1, NLM_F_MULTI) <= 0) {
3051 rcu_read_unlock_bh();
3056 rcu_read_unlock_bh();
3065 void ip_rt_multicast_event(struct in_device *in_dev)
3067 rt_cache_flush(dev_net(in_dev->dev), 0);
3070 #ifdef CONFIG_SYSCTL
3071 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3072 void __user *buffer,
3073 size_t *lenp, loff_t *ppos)
3080 memcpy(&ctl, __ctl, sizeof(ctl));
3081 ctl.data = &flush_delay;
3082 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3084 net = (struct net *)__ctl->extra1;
3085 rt_cache_flush(net, flush_delay);
3092 static ctl_table ipv4_route_table[] = {
3094 .procname = "gc_thresh",
3095 .data = &ipv4_dst_ops.gc_thresh,
3096 .maxlen = sizeof(int),
3098 .proc_handler = proc_dointvec,
3101 .procname = "max_size",
3102 .data = &ip_rt_max_size,
3103 .maxlen = sizeof(int),
3105 .proc_handler = proc_dointvec,
3108 /* Deprecated. Use gc_min_interval_ms */
3110 .procname = "gc_min_interval",
3111 .data = &ip_rt_gc_min_interval,
3112 .maxlen = sizeof(int),
3114 .proc_handler = proc_dointvec_jiffies,
3117 .procname = "gc_min_interval_ms",
3118 .data = &ip_rt_gc_min_interval,
3119 .maxlen = sizeof(int),
3121 .proc_handler = proc_dointvec_ms_jiffies,
3124 .procname = "gc_timeout",
3125 .data = &ip_rt_gc_timeout,
3126 .maxlen = sizeof(int),
3128 .proc_handler = proc_dointvec_jiffies,
3131 .procname = "gc_interval",
3132 .data = &ip_rt_gc_interval,
3133 .maxlen = sizeof(int),
3135 .proc_handler = proc_dointvec_jiffies,
3138 .procname = "redirect_load",
3139 .data = &ip_rt_redirect_load,
3140 .maxlen = sizeof(int),
3142 .proc_handler = proc_dointvec,
3145 .procname = "redirect_number",
3146 .data = &ip_rt_redirect_number,
3147 .maxlen = sizeof(int),
3149 .proc_handler = proc_dointvec,
3152 .procname = "redirect_silence",
3153 .data = &ip_rt_redirect_silence,
3154 .maxlen = sizeof(int),
3156 .proc_handler = proc_dointvec,
3159 .procname = "error_cost",
3160 .data = &ip_rt_error_cost,
3161 .maxlen = sizeof(int),
3163 .proc_handler = proc_dointvec,
3166 .procname = "error_burst",
3167 .data = &ip_rt_error_burst,
3168 .maxlen = sizeof(int),
3170 .proc_handler = proc_dointvec,
3173 .procname = "gc_elasticity",
3174 .data = &ip_rt_gc_elasticity,
3175 .maxlen = sizeof(int),
3177 .proc_handler = proc_dointvec,
3180 .procname = "mtu_expires",
3181 .data = &ip_rt_mtu_expires,
3182 .maxlen = sizeof(int),
3184 .proc_handler = proc_dointvec_jiffies,
3187 .procname = "min_pmtu",
3188 .data = &ip_rt_min_pmtu,
3189 .maxlen = sizeof(int),
3191 .proc_handler = proc_dointvec,
3194 .procname = "min_adv_mss",
3195 .data = &ip_rt_min_advmss,
3196 .maxlen = sizeof(int),
3198 .proc_handler = proc_dointvec,
3203 static struct ctl_table empty[1];
3205 static struct ctl_table ipv4_skeleton[] =
3207 { .procname = "route",
3208 .mode = 0555, .child = ipv4_route_table},
3209 { .procname = "neigh",
3210 .mode = 0555, .child = empty},
3214 static __net_initdata struct ctl_path ipv4_path[] = {
3215 { .procname = "net", },
3216 { .procname = "ipv4", },
3220 static struct ctl_table ipv4_route_flush_table[] = {
3222 .procname = "flush",
3223 .maxlen = sizeof(int),
3225 .proc_handler = ipv4_sysctl_rtcache_flush,
3230 static __net_initdata struct ctl_path ipv4_route_path[] = {
3231 { .procname = "net", },
3232 { .procname = "ipv4", },
3233 { .procname = "route", },
3237 static __net_init int sysctl_route_net_init(struct net *net)
3239 struct ctl_table *tbl;
3241 tbl = ipv4_route_flush_table;
3242 if (!net_eq(net, &init_net)) {
3243 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3247 tbl[0].extra1 = net;
3249 net->ipv4.route_hdr =
3250 register_net_sysctl_table(net, ipv4_route_path, tbl);
3251 if (net->ipv4.route_hdr == NULL)
3256 if (tbl != ipv4_route_flush_table)
3262 static __net_exit void sysctl_route_net_exit(struct net *net)
3264 struct ctl_table *tbl;
3266 tbl = net->ipv4.route_hdr->ctl_table_arg;
3267 unregister_net_sysctl_table(net->ipv4.route_hdr);
3268 BUG_ON(tbl == ipv4_route_flush_table);
3272 static __net_initdata struct pernet_operations sysctl_route_ops = {
3273 .init = sysctl_route_net_init,
3274 .exit = sysctl_route_net_exit,
3278 static __net_init int rt_genid_init(struct net *net)
3280 get_random_bytes(&net->ipv4.rt_genid,
3281 sizeof(net->ipv4.rt_genid));
3285 static __net_initdata struct pernet_operations rt_genid_ops = {
3286 .init = rt_genid_init,
3290 #ifdef CONFIG_NET_CLS_ROUTE
3291 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3292 #endif /* CONFIG_NET_CLS_ROUTE */
3294 static __initdata unsigned long rhash_entries;
3295 static int __init set_rhash_entries(char *str)
3299 rhash_entries = simple_strtoul(str, &str, 0);
3302 __setup("rhash_entries=", set_rhash_entries);
3304 int __init ip_rt_init(void)
3308 #ifdef CONFIG_NET_CLS_ROUTE
3309 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3311 panic("IP: failed to allocate ip_rt_acct\n");
3314 ipv4_dst_ops.kmem_cachep =
3315 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3316 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3318 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3320 rt_hash_table = (struct rt_hash_bucket *)
3321 alloc_large_system_hash("IP route cache",
3322 sizeof(struct rt_hash_bucket),
3324 (totalram_pages >= 128 * 1024) ?
3329 rhash_entries ? 0 : 512 * 1024);
3330 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3331 rt_hash_lock_init();
3333 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3334 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3339 /* All the timers, started at system startup tend
3340 to synchronize. Perturb it a bit.
3342 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3343 expires_ljiffies = jiffies;
3344 schedule_delayed_work(&expires_work,
3345 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3347 if (ip_rt_proc_init())
3348 printk(KERN_ERR "Unable to create route proc files\n");
3351 xfrm4_init(ip_rt_max_size);
3353 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3355 #ifdef CONFIG_SYSCTL
3356 register_pernet_subsys(&sysctl_route_ops);
3358 register_pernet_subsys(&rt_genid_ops);
3362 #ifdef CONFIG_SYSCTL
3364 * We really need to sanitize the damn ipv4 init order, then all
3365 * this nonsense will go away.
3367 void __init ip_static_sysctl_init(void)
3369 register_sysctl_paths(ipv4_path, ipv4_skeleton);
3373 EXPORT_SYMBOL(__ip_select_ident);
3374 EXPORT_SYMBOL(ip_route_output_key);