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->dst.dev) == seq_file_net(seq) &&
290 r->rt_genid == st->genid)
292 r = rcu_dereference_bh(r->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;
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->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->dst.dev ? r->dst.dev->name : "*",
381 (__force u32)r->rt_dst,
382 (__force u32)r->rt_gateway,
383 r->rt_flags, atomic_read(&r->dst.__refcnt),
384 r->dst.__use, 0, (__force u32)r->rt_src,
385 (dst_metric(&r->dst, RTAX_ADVMSS) ?
386 (int)dst_metric(&r->dst, RTAX_ADVMSS) + 40 : 0),
387 dst_metric(&r->dst, RTAX_WINDOW),
388 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
389 dst_metric(&r->dst, RTAX_RTTVAR)),
391 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
392 r->dst.hh ? (r->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->dst.rcu_head, dst_rcu_free);
614 static inline void rt_drop(struct rtable *rt)
617 call_rcu_bh(&rt->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->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->dst.__refcnt))
643 if (rth->dst.expires &&
644 time_after_eq(jiffies, rth->dst.expires))
647 age = jiffies - rth->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->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->dst.dev), dev_net(rt2->dst.dev));
706 static inline int rt_is_expired(struct rtable *rth)
708 return rth->rt_genid != rt_genid(dev_net(rth->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->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->dst.rt_next;
747 if (!rt_is_expired(p)) {
748 prev = &p->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->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->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->dst.rt_next);
835 if (rt_is_expired(rth)) {
836 *rthp = rth->dst.rt_next;
840 if (rth->dst.expires) {
841 /* Entry is expired even if it is in use */
842 if (time_before_eq(jiffies, rth->dst.expires)) {
845 rthp = &rth->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->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->dst.rt_next;
1005 *rthp = rth->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->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->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 rt->dst.flags |= DST_NOCACHE;
1111 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1112 int err = arp_bind_neighbour(&rt->dst);
1114 if (net_ratelimit())
1116 "Neighbour table failure & not caching routes.\n");
1126 rthp = &rt_hash_table[hash].chain;
1128 spin_lock_bh(rt_hash_lock_addr(hash));
1129 while ((rth = *rthp) != NULL) {
1130 if (rt_is_expired(rth)) {
1131 *rthp = rth->dst.rt_next;
1135 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1137 *rthp = rth->dst.rt_next;
1139 * Since lookup is lockfree, the deletion
1140 * must be visible to another weakly ordered CPU before
1141 * the insertion at the start of the hash chain.
1143 rcu_assign_pointer(rth->dst.rt_next,
1144 rt_hash_table[hash].chain);
1146 * Since lookup is lockfree, the update writes
1147 * must be ordered for consistency on SMP.
1149 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1151 dst_use(&rth->dst, now);
1152 spin_unlock_bh(rt_hash_lock_addr(hash));
1158 skb_dst_set(skb, &rth->dst);
1162 if (!atomic_read(&rth->dst.__refcnt)) {
1163 u32 score = rt_score(rth);
1165 if (score <= min_score) {
1174 rthp = &rth->dst.rt_next;
1178 /* ip_rt_gc_elasticity used to be average length of chain
1179 * length, when exceeded gc becomes really aggressive.
1181 * The second limit is less certain. At the moment it allows
1182 * only 2 entries per bucket. We will see.
1184 if (chain_length > ip_rt_gc_elasticity) {
1185 *candp = cand->dst.rt_next;
1189 if (chain_length > rt_chain_length_max &&
1190 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1191 struct net *net = dev_net(rt->dst.dev);
1192 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1193 if (!rt_caching(net)) {
1194 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1195 rt->dst.dev->name, num);
1197 rt_emergency_hash_rebuild(net);
1198 spin_unlock_bh(rt_hash_lock_addr(hash));
1200 hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1201 ifindex, rt_genid(net));
1206 /* Try to bind route to arp only if it is output
1207 route or unicast forwarding path.
1209 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1210 int err = arp_bind_neighbour(&rt->dst);
1212 spin_unlock_bh(rt_hash_lock_addr(hash));
1214 if (err != -ENOBUFS) {
1219 /* Neighbour tables are full and nothing
1220 can be released. Try to shrink route cache,
1221 it is most likely it holds some neighbour records.
1223 if (attempts-- > 0) {
1224 int saved_elasticity = ip_rt_gc_elasticity;
1225 int saved_int = ip_rt_gc_min_interval;
1226 ip_rt_gc_elasticity = 1;
1227 ip_rt_gc_min_interval = 0;
1228 rt_garbage_collect(&ipv4_dst_ops);
1229 ip_rt_gc_min_interval = saved_int;
1230 ip_rt_gc_elasticity = saved_elasticity;
1234 if (net_ratelimit())
1235 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1241 rt->dst.rt_next = rt_hash_table[hash].chain;
1243 #if RT_CACHE_DEBUG >= 2
1244 if (rt->dst.rt_next) {
1246 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1248 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1249 printk(" . %pI4", &trt->rt_dst);
1254 * Since lookup is lockfree, we must make sure
1255 * previous writes to rt are comitted to memory
1256 * before making rt visible to other CPUS.
1258 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1260 spin_unlock_bh(rt_hash_lock_addr(hash));
1266 skb_dst_set(skb, &rt->dst);
1270 void rt_bind_peer(struct rtable *rt, int create)
1272 struct inet_peer *peer;
1274 peer = inet_getpeer(rt->rt_dst, create);
1276 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1281 * Peer allocation may fail only in serious out-of-memory conditions. However
1282 * we still can generate some output.
1283 * Random ID selection looks a bit dangerous because we have no chances to
1284 * select ID being unique in a reasonable period of time.
1285 * But broken packet identifier may be better than no packet at all.
1287 static void ip_select_fb_ident(struct iphdr *iph)
1289 static DEFINE_SPINLOCK(ip_fb_id_lock);
1290 static u32 ip_fallback_id;
1293 spin_lock_bh(&ip_fb_id_lock);
1294 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1295 iph->id = htons(salt & 0xFFFF);
1296 ip_fallback_id = salt;
1297 spin_unlock_bh(&ip_fb_id_lock);
1300 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1302 struct rtable *rt = (struct rtable *) dst;
1305 if (rt->peer == NULL)
1306 rt_bind_peer(rt, 1);
1308 /* If peer is attached to destination, it is never detached,
1309 so that we need not to grab a lock to dereference it.
1312 iph->id = htons(inet_getid(rt->peer, more));
1316 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1317 __builtin_return_address(0));
1319 ip_select_fb_ident(iph);
1321 EXPORT_SYMBOL(__ip_select_ident);
1323 static void rt_del(unsigned hash, struct rtable *rt)
1325 struct rtable **rthp, *aux;
1327 rthp = &rt_hash_table[hash].chain;
1328 spin_lock_bh(rt_hash_lock_addr(hash));
1330 while ((aux = *rthp) != NULL) {
1331 if (aux == rt || rt_is_expired(aux)) {
1332 *rthp = aux->dst.rt_next;
1336 rthp = &aux->dst.rt_next;
1338 spin_unlock_bh(rt_hash_lock_addr(hash));
1341 /* called in rcu_read_lock() section */
1342 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1343 __be32 saddr, struct net_device *dev)
1346 struct in_device *in_dev = __in_dev_get_rcu(dev);
1347 struct rtable *rth, **rthp;
1348 __be32 skeys[2] = { saddr, 0 };
1349 int ikeys[2] = { dev->ifindex, 0 };
1350 struct netevent_redirect netevent;
1357 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1358 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1359 ipv4_is_zeronet(new_gw))
1360 goto reject_redirect;
1362 if (!rt_caching(net))
1363 goto reject_redirect;
1365 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1366 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1367 goto reject_redirect;
1368 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1369 goto reject_redirect;
1371 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1372 goto reject_redirect;
1375 for (i = 0; i < 2; i++) {
1376 for (k = 0; k < 2; k++) {
1377 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1380 rthp=&rt_hash_table[hash].chain;
1382 while ((rth = rcu_dereference(*rthp)) != NULL) {
1385 if (rth->fl.fl4_dst != daddr ||
1386 rth->fl.fl4_src != skeys[i] ||
1387 rth->fl.oif != ikeys[k] ||
1389 rt_is_expired(rth) ||
1390 !net_eq(dev_net(rth->dst.dev), net)) {
1391 rthp = &rth->dst.rt_next;
1395 if (rth->rt_dst != daddr ||
1396 rth->rt_src != saddr ||
1398 rth->rt_gateway != old_gw ||
1399 rth->dst.dev != dev)
1402 dst_hold(&rth->dst);
1404 rt = dst_alloc(&ipv4_dst_ops);
1410 /* Copy all the information. */
1413 atomic_set(&rt->dst.__refcnt, 1);
1414 rt->dst.child = NULL;
1416 dev_hold(rt->dst.dev);
1418 in_dev_hold(rt->idev);
1419 rt->dst.obsolete = -1;
1420 rt->dst.lastuse = jiffies;
1421 rt->dst.path = &rt->dst;
1422 rt->dst.neighbour = NULL;
1425 rt->dst.xfrm = NULL;
1427 rt->rt_genid = rt_genid(net);
1428 rt->rt_flags |= RTCF_REDIRECTED;
1430 /* Gateway is different ... */
1431 rt->rt_gateway = new_gw;
1433 /* Redirect received -> path was valid */
1434 dst_confirm(&rth->dst);
1437 atomic_inc(&rt->peer->refcnt);
1439 if (arp_bind_neighbour(&rt->dst) ||
1440 !(rt->dst.neighbour->nud_state &
1442 if (rt->dst.neighbour)
1443 neigh_event_send(rt->dst.neighbour, NULL);
1449 netevent.old = &rth->dst;
1450 netevent.new = &rt->dst;
1451 call_netevent_notifiers(NETEVENT_REDIRECT,
1455 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1466 #ifdef CONFIG_IP_ROUTE_VERBOSE
1467 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1468 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1469 " Advised path = %pI4 -> %pI4\n",
1470 &old_gw, dev->name, &new_gw,
1476 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1478 struct rtable *rt = (struct rtable *)dst;
1479 struct dst_entry *ret = dst;
1482 if (dst->obsolete > 0) {
1485 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1487 time_after_eq(jiffies, rt->dst.expires))) {
1488 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1490 rt_genid(dev_net(dst->dev)));
1491 #if RT_CACHE_DEBUG >= 1
1492 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1493 &rt->rt_dst, rt->fl.fl4_tos);
1504 * 1. The first ip_rt_redirect_number redirects are sent
1505 * with exponential backoff, then we stop sending them at all,
1506 * assuming that the host ignores our redirects.
1507 * 2. If we did not see packets requiring redirects
1508 * during ip_rt_redirect_silence, we assume that the host
1509 * forgot redirected route and start to send redirects again.
1511 * This algorithm is much cheaper and more intelligent than dumb load limiting
1514 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1515 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1518 void ip_rt_send_redirect(struct sk_buff *skb)
1520 struct rtable *rt = skb_rtable(skb);
1521 struct in_device *in_dev;
1525 in_dev = __in_dev_get_rcu(rt->dst.dev);
1526 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1530 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1533 /* No redirected packets during ip_rt_redirect_silence;
1534 * reset the algorithm.
1536 if (time_after(jiffies, rt->dst.rate_last + ip_rt_redirect_silence))
1537 rt->dst.rate_tokens = 0;
1539 /* Too many ignored redirects; do not send anything
1540 * set dst.rate_last to the last seen redirected packet.
1542 if (rt->dst.rate_tokens >= ip_rt_redirect_number) {
1543 rt->dst.rate_last = jiffies;
1547 /* Check for load limit; set rate_last to the latest sent
1550 if (rt->dst.rate_tokens == 0 ||
1552 (rt->dst.rate_last +
1553 (ip_rt_redirect_load << rt->dst.rate_tokens)))) {
1554 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1555 rt->dst.rate_last = jiffies;
1556 ++rt->dst.rate_tokens;
1557 #ifdef CONFIG_IP_ROUTE_VERBOSE
1559 rt->dst.rate_tokens == ip_rt_redirect_number &&
1561 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1562 &rt->rt_src, rt->rt_iif,
1563 &rt->rt_dst, &rt->rt_gateway);
1568 static int ip_error(struct sk_buff *skb)
1570 struct rtable *rt = skb_rtable(skb);
1574 switch (rt->dst.error) {
1579 code = ICMP_HOST_UNREACH;
1582 code = ICMP_NET_UNREACH;
1583 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1584 IPSTATS_MIB_INNOROUTES);
1587 code = ICMP_PKT_FILTERED;
1592 rt->dst.rate_tokens += now - rt->dst.rate_last;
1593 if (rt->dst.rate_tokens > ip_rt_error_burst)
1594 rt->dst.rate_tokens = ip_rt_error_burst;
1595 rt->dst.rate_last = now;
1596 if (rt->dst.rate_tokens >= ip_rt_error_cost) {
1597 rt->dst.rate_tokens -= ip_rt_error_cost;
1598 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1601 out: kfree_skb(skb);
1606 * The last two values are not from the RFC but
1607 * are needed for AMPRnet AX.25 paths.
1610 static const unsigned short mtu_plateau[] =
1611 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1613 static inline unsigned short guess_mtu(unsigned short old_mtu)
1617 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1618 if (old_mtu > mtu_plateau[i])
1619 return mtu_plateau[i];
1623 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1624 unsigned short new_mtu,
1625 struct net_device *dev)
1628 unsigned short old_mtu = ntohs(iph->tot_len);
1630 int ikeys[2] = { dev->ifindex, 0 };
1631 __be32 skeys[2] = { iph->saddr, 0, };
1632 __be32 daddr = iph->daddr;
1633 unsigned short est_mtu = 0;
1635 for (k = 0; k < 2; k++) {
1636 for (i = 0; i < 2; i++) {
1637 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1641 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1642 rth = rcu_dereference(rth->dst.rt_next)) {
1643 unsigned short mtu = new_mtu;
1645 if (rth->fl.fl4_dst != daddr ||
1646 rth->fl.fl4_src != skeys[i] ||
1647 rth->rt_dst != daddr ||
1648 rth->rt_src != iph->saddr ||
1649 rth->fl.oif != ikeys[k] ||
1651 dst_metric_locked(&rth->dst, RTAX_MTU) ||
1652 !net_eq(dev_net(rth->dst.dev), net) ||
1656 if (new_mtu < 68 || new_mtu >= old_mtu) {
1658 /* BSD 4.2 compatibility hack :-( */
1660 old_mtu >= dst_mtu(&rth->dst) &&
1661 old_mtu >= 68 + (iph->ihl << 2))
1662 old_mtu -= iph->ihl << 2;
1664 mtu = guess_mtu(old_mtu);
1666 if (mtu <= dst_mtu(&rth->dst)) {
1667 if (mtu < dst_mtu(&rth->dst)) {
1668 dst_confirm(&rth->dst);
1669 if (mtu < ip_rt_min_pmtu) {
1670 mtu = ip_rt_min_pmtu;
1671 rth->dst.metrics[RTAX_LOCK-1] |=
1674 rth->dst.metrics[RTAX_MTU-1] = mtu;
1675 dst_set_expires(&rth->dst,
1684 return est_mtu ? : new_mtu;
1687 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1689 if (dst_mtu(dst) > mtu && mtu >= 68 &&
1690 !(dst_metric_locked(dst, RTAX_MTU))) {
1691 if (mtu < ip_rt_min_pmtu) {
1692 mtu = ip_rt_min_pmtu;
1693 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1695 dst->metrics[RTAX_MTU-1] = mtu;
1696 dst_set_expires(dst, ip_rt_mtu_expires);
1697 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1701 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1703 if (rt_is_expired((struct rtable *)dst))
1708 static void ipv4_dst_destroy(struct dst_entry *dst)
1710 struct rtable *rt = (struct rtable *) dst;
1711 struct inet_peer *peer = rt->peer;
1712 struct in_device *idev = rt->idev;
1725 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1728 struct rtable *rt = (struct rtable *) dst;
1729 struct in_device *idev = rt->idev;
1730 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1731 struct in_device *loopback_idev =
1732 in_dev_get(dev_net(dev)->loopback_dev);
1733 if (loopback_idev) {
1734 rt->idev = loopback_idev;
1740 static void ipv4_link_failure(struct sk_buff *skb)
1744 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1746 rt = skb_rtable(skb);
1748 dst_set_expires(&rt->dst, 0);
1751 static int ip_rt_bug(struct sk_buff *skb)
1753 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1754 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1755 skb->dev ? skb->dev->name : "?");
1761 We do not cache source address of outgoing interface,
1762 because it is used only by IP RR, TS and SRR options,
1763 so that it out of fast path.
1765 BTW remember: "addr" is allowed to be not aligned
1769 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1772 struct fib_result res;
1774 if (rt->fl.iif == 0)
1776 else if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0) {
1777 src = FIB_RES_PREFSRC(res);
1780 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1782 memcpy(addr, &src, 4);
1785 #ifdef CONFIG_NET_CLS_ROUTE
1786 static void set_class_tag(struct rtable *rt, u32 tag)
1788 if (!(rt->dst.tclassid & 0xFFFF))
1789 rt->dst.tclassid |= tag & 0xFFFF;
1790 if (!(rt->dst.tclassid & 0xFFFF0000))
1791 rt->dst.tclassid |= tag & 0xFFFF0000;
1795 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1797 struct fib_info *fi = res->fi;
1800 if (FIB_RES_GW(*res) &&
1801 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1802 rt->rt_gateway = FIB_RES_GW(*res);
1803 memcpy(rt->dst.metrics, fi->fib_metrics,
1804 sizeof(rt->dst.metrics));
1805 if (fi->fib_mtu == 0) {
1806 rt->dst.metrics[RTAX_MTU-1] = rt->dst.dev->mtu;
1807 if (dst_metric_locked(&rt->dst, RTAX_MTU) &&
1808 rt->rt_gateway != rt->rt_dst &&
1809 rt->dst.dev->mtu > 576)
1810 rt->dst.metrics[RTAX_MTU-1] = 576;
1812 #ifdef CONFIG_NET_CLS_ROUTE
1813 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1816 rt->dst.metrics[RTAX_MTU-1]= rt->dst.dev->mtu;
1818 if (dst_metric(&rt->dst, RTAX_HOPLIMIT) == 0)
1819 rt->dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1820 if (dst_mtu(&rt->dst) > IP_MAX_MTU)
1821 rt->dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1822 if (dst_metric(&rt->dst, RTAX_ADVMSS) == 0)
1823 rt->dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->dst.dev->mtu - 40,
1825 if (dst_metric(&rt->dst, RTAX_ADVMSS) > 65535 - 40)
1826 rt->dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1828 #ifdef CONFIG_NET_CLS_ROUTE
1829 #ifdef CONFIG_IP_MULTIPLE_TABLES
1830 set_class_tag(rt, fib_rules_tclass(res));
1832 set_class_tag(rt, itag);
1834 rt->rt_type = res->type;
1837 /* called in rcu_read_lock() section */
1838 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1839 u8 tos, struct net_device *dev, int our)
1844 struct in_device *in_dev = __in_dev_get_rcu(dev);
1848 /* Primary sanity checks. */
1853 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1854 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1857 if (ipv4_is_zeronet(saddr)) {
1858 if (!ipv4_is_local_multicast(daddr))
1860 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1862 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1867 rth = dst_alloc(&ipv4_dst_ops);
1871 rth->dst.output = ip_rt_bug;
1872 rth->dst.obsolete = -1;
1874 atomic_set(&rth->dst.__refcnt, 1);
1875 rth->dst.flags= DST_HOST;
1876 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1877 rth->dst.flags |= DST_NOPOLICY;
1878 rth->fl.fl4_dst = daddr;
1879 rth->rt_dst = daddr;
1880 rth->fl.fl4_tos = tos;
1881 rth->fl.mark = skb->mark;
1882 rth->fl.fl4_src = saddr;
1883 rth->rt_src = saddr;
1884 #ifdef CONFIG_NET_CLS_ROUTE
1885 rth->dst.tclassid = itag;
1888 rth->fl.iif = dev->ifindex;
1889 rth->dst.dev = init_net.loopback_dev;
1890 dev_hold(rth->dst.dev);
1891 rth->idev = in_dev_get(rth->dst.dev);
1893 rth->rt_gateway = daddr;
1894 rth->rt_spec_dst= spec_dst;
1895 rth->rt_genid = rt_genid(dev_net(dev));
1896 rth->rt_flags = RTCF_MULTICAST;
1897 rth->rt_type = RTN_MULTICAST;
1899 rth->dst.input= ip_local_deliver;
1900 rth->rt_flags |= RTCF_LOCAL;
1903 #ifdef CONFIG_IP_MROUTE
1904 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1905 rth->dst.input = ip_mr_input;
1907 RT_CACHE_STAT_INC(in_slow_mc);
1909 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1910 return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1921 static void ip_handle_martian_source(struct net_device *dev,
1922 struct in_device *in_dev,
1923 struct sk_buff *skb,
1927 RT_CACHE_STAT_INC(in_martian_src);
1928 #ifdef CONFIG_IP_ROUTE_VERBOSE
1929 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1931 * RFC1812 recommendation, if source is martian,
1932 * the only hint is MAC header.
1934 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1935 &daddr, &saddr, dev->name);
1936 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1938 const unsigned char *p = skb_mac_header(skb);
1939 printk(KERN_WARNING "ll header: ");
1940 for (i = 0; i < dev->hard_header_len; i++, p++) {
1942 if (i < (dev->hard_header_len - 1))
1951 /* called in rcu_read_lock() section */
1952 static int __mkroute_input(struct sk_buff *skb,
1953 struct fib_result *res,
1954 struct in_device *in_dev,
1955 __be32 daddr, __be32 saddr, u32 tos,
1956 struct rtable **result)
1960 struct in_device *out_dev;
1961 unsigned int flags = 0;
1965 /* get a working reference to the output device */
1966 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1967 if (out_dev == NULL) {
1968 if (net_ratelimit())
1969 printk(KERN_CRIT "Bug in ip_route_input" \
1970 "_slow(). Please, report\n");
1975 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1976 in_dev->dev, &spec_dst, &itag, skb->mark);
1978 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1985 flags |= RTCF_DIRECTSRC;
1987 if (out_dev == in_dev && err &&
1988 (IN_DEV_SHARED_MEDIA(out_dev) ||
1989 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1990 flags |= RTCF_DOREDIRECT;
1992 if (skb->protocol != htons(ETH_P_IP)) {
1993 /* Not IP (i.e. ARP). Do not create route, if it is
1994 * invalid for proxy arp. DNAT routes are always valid.
1996 * Proxy arp feature have been extended to allow, ARP
1997 * replies back to the same interface, to support
1998 * Private VLAN switch technologies. See arp.c.
2000 if (out_dev == in_dev &&
2001 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2008 rth = dst_alloc(&ipv4_dst_ops);
2014 atomic_set(&rth->dst.__refcnt, 1);
2015 rth->dst.flags= DST_HOST;
2016 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2017 rth->dst.flags |= DST_NOPOLICY;
2018 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2019 rth->dst.flags |= DST_NOXFRM;
2020 rth->fl.fl4_dst = daddr;
2021 rth->rt_dst = daddr;
2022 rth->fl.fl4_tos = tos;
2023 rth->fl.mark = skb->mark;
2024 rth->fl.fl4_src = saddr;
2025 rth->rt_src = saddr;
2026 rth->rt_gateway = daddr;
2028 rth->fl.iif = in_dev->dev->ifindex;
2029 rth->dst.dev = (out_dev)->dev;
2030 dev_hold(rth->dst.dev);
2031 rth->idev = in_dev_get(rth->dst.dev);
2033 rth->rt_spec_dst= spec_dst;
2035 rth->dst.obsolete = -1;
2036 rth->dst.input = ip_forward;
2037 rth->dst.output = ip_output;
2038 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2040 rt_set_nexthop(rth, res, itag);
2042 rth->rt_flags = flags;
2050 static int ip_mkroute_input(struct sk_buff *skb,
2051 struct fib_result *res,
2052 const struct flowi *fl,
2053 struct in_device *in_dev,
2054 __be32 daddr, __be32 saddr, u32 tos)
2056 struct rtable* rth = NULL;
2060 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2061 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2062 fib_select_multipath(fl, res);
2065 /* create a routing cache entry */
2066 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2070 /* put it into the cache */
2071 hash = rt_hash(daddr, saddr, fl->iif,
2072 rt_genid(dev_net(rth->dst.dev)));
2073 return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2077 * NOTE. We drop all the packets that has local source
2078 * addresses, because every properly looped back packet
2079 * must have correct destination already attached by output routine.
2081 * Such approach solves two big problems:
2082 * 1. Not simplex devices are handled properly.
2083 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2086 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2087 u8 tos, struct net_device *dev)
2089 struct fib_result res;
2090 struct in_device *in_dev = __in_dev_get_rcu(dev);
2091 struct flowi fl = { .nl_u = { .ip4_u =
2095 .scope = RT_SCOPE_UNIVERSE,
2098 .iif = dev->ifindex };
2101 struct rtable * rth;
2106 struct net * net = dev_net(dev);
2108 /* IP on this device is disabled. */
2113 /* Check for the most weird martians, which can be not detected
2117 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2118 ipv4_is_loopback(saddr))
2119 goto martian_source;
2121 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2124 /* Accept zero addresses only to limited broadcast;
2125 * I even do not know to fix it or not. Waiting for complains :-)
2127 if (ipv4_is_zeronet(saddr))
2128 goto martian_source;
2130 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2131 ipv4_is_loopback(daddr))
2132 goto martian_destination;
2135 * Now we are ready to route packet.
2137 if ((err = fib_lookup(net, &fl, &res)) != 0) {
2138 if (!IN_DEV_FORWARD(in_dev))
2144 RT_CACHE_STAT_INC(in_slow_tot);
2146 if (res.type == RTN_BROADCAST)
2149 if (res.type == RTN_LOCAL) {
2150 err = fib_validate_source(saddr, daddr, tos,
2151 net->loopback_dev->ifindex,
2152 dev, &spec_dst, &itag, skb->mark);
2154 goto martian_source_keep_err;
2156 flags |= RTCF_DIRECTSRC;
2161 if (!IN_DEV_FORWARD(in_dev))
2163 if (res.type != RTN_UNICAST)
2164 goto martian_destination;
2166 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2173 if (skb->protocol != htons(ETH_P_IP))
2176 if (ipv4_is_zeronet(saddr))
2177 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2179 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2182 goto martian_source_keep_err;
2184 flags |= RTCF_DIRECTSRC;
2186 flags |= RTCF_BROADCAST;
2187 res.type = RTN_BROADCAST;
2188 RT_CACHE_STAT_INC(in_brd);
2191 rth = dst_alloc(&ipv4_dst_ops);
2195 rth->dst.output= ip_rt_bug;
2196 rth->dst.obsolete = -1;
2197 rth->rt_genid = rt_genid(net);
2199 atomic_set(&rth->dst.__refcnt, 1);
2200 rth->dst.flags= DST_HOST;
2201 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2202 rth->dst.flags |= DST_NOPOLICY;
2203 rth->fl.fl4_dst = daddr;
2204 rth->rt_dst = daddr;
2205 rth->fl.fl4_tos = tos;
2206 rth->fl.mark = skb->mark;
2207 rth->fl.fl4_src = saddr;
2208 rth->rt_src = saddr;
2209 #ifdef CONFIG_NET_CLS_ROUTE
2210 rth->dst.tclassid = itag;
2213 rth->fl.iif = dev->ifindex;
2214 rth->dst.dev = net->loopback_dev;
2215 dev_hold(rth->dst.dev);
2216 rth->idev = in_dev_get(rth->dst.dev);
2217 rth->rt_gateway = daddr;
2218 rth->rt_spec_dst= spec_dst;
2219 rth->dst.input= ip_local_deliver;
2220 rth->rt_flags = flags|RTCF_LOCAL;
2221 if (res.type == RTN_UNREACHABLE) {
2222 rth->dst.input= ip_error;
2223 rth->dst.error= -err;
2224 rth->rt_flags &= ~RTCF_LOCAL;
2226 rth->rt_type = res.type;
2227 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2228 err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2232 RT_CACHE_STAT_INC(in_no_route);
2233 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2234 res.type = RTN_UNREACHABLE;
2240 * Do not cache martian addresses: they should be logged (RFC1812)
2242 martian_destination:
2243 RT_CACHE_STAT_INC(in_martian_dst);
2244 #ifdef CONFIG_IP_ROUTE_VERBOSE
2245 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2246 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2247 &daddr, &saddr, dev->name);
2251 err = -EHOSTUNREACH;
2264 martian_source_keep_err:
2265 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2269 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2270 u8 tos, struct net_device *dev, bool noref)
2272 struct rtable * rth;
2274 int iif = dev->ifindex;
2282 if (!rt_caching(net))
2285 tos &= IPTOS_RT_MASK;
2286 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2288 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2289 rth = rcu_dereference(rth->dst.rt_next)) {
2290 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2291 ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2292 (rth->fl.iif ^ iif) |
2294 (rth->fl.fl4_tos ^ tos)) == 0 &&
2295 rth->fl.mark == skb->mark &&
2296 net_eq(dev_net(rth->dst.dev), net) &&
2297 !rt_is_expired(rth)) {
2299 dst_use_noref(&rth->dst, jiffies);
2300 skb_dst_set_noref(skb, &rth->dst);
2302 dst_use(&rth->dst, jiffies);
2303 skb_dst_set(skb, &rth->dst);
2305 RT_CACHE_STAT_INC(in_hit);
2309 RT_CACHE_STAT_INC(in_hlist_search);
2313 /* Multicast recognition logic is moved from route cache to here.
2314 The problem was that too many Ethernet cards have broken/missing
2315 hardware multicast filters :-( As result the host on multicasting
2316 network acquires a lot of useless route cache entries, sort of
2317 SDR messages from all the world. Now we try to get rid of them.
2318 Really, provided software IP multicast filter is organized
2319 reasonably (at least, hashed), it does not result in a slowdown
2320 comparing with route cache reject entries.
2321 Note, that multicast routers are not affected, because
2322 route cache entry is created eventually.
2324 if (ipv4_is_multicast(daddr)) {
2325 struct in_device *in_dev = __in_dev_get_rcu(dev);
2328 int our = ip_check_mc(in_dev, daddr, saddr,
2329 ip_hdr(skb)->protocol);
2331 #ifdef CONFIG_IP_MROUTE
2333 (!ipv4_is_local_multicast(daddr) &&
2334 IN_DEV_MFORWARD(in_dev))
2337 int res = ip_route_input_mc(skb, daddr, saddr,
2346 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2350 EXPORT_SYMBOL(ip_route_input_common);
2352 static int __mkroute_output(struct rtable **result,
2353 struct fib_result *res,
2354 const struct flowi *fl,
2355 const struct flowi *oldflp,
2356 struct net_device *dev_out,
2360 struct in_device *in_dev;
2361 u32 tos = RT_FL_TOS(oldflp);
2363 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2366 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2367 res->type = RTN_BROADCAST;
2368 else if (ipv4_is_multicast(fl->fl4_dst))
2369 res->type = RTN_MULTICAST;
2370 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2373 if (dev_out->flags & IFF_LOOPBACK)
2374 flags |= RTCF_LOCAL;
2377 in_dev = __in_dev_get_rcu(dev_out);
2382 if (res->type == RTN_BROADCAST) {
2383 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2385 fib_info_put(res->fi);
2388 } else if (res->type == RTN_MULTICAST) {
2389 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2390 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2392 flags &= ~RTCF_LOCAL;
2393 /* If multicast route do not exist use
2394 * default one, but do not gateway in this case.
2397 if (res->fi && res->prefixlen < 4) {
2398 fib_info_put(res->fi);
2404 rth = dst_alloc(&ipv4_dst_ops);
2409 in_dev_hold(in_dev);
2413 atomic_set(&rth->dst.__refcnt, 1);
2414 rth->dst.flags= DST_HOST;
2415 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2416 rth->dst.flags |= DST_NOXFRM;
2417 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2418 rth->dst.flags |= DST_NOPOLICY;
2420 rth->fl.fl4_dst = oldflp->fl4_dst;
2421 rth->fl.fl4_tos = tos;
2422 rth->fl.fl4_src = oldflp->fl4_src;
2423 rth->fl.oif = oldflp->oif;
2424 rth->fl.mark = oldflp->mark;
2425 rth->rt_dst = fl->fl4_dst;
2426 rth->rt_src = fl->fl4_src;
2427 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2428 /* get references to the devices that are to be hold by the routing
2430 rth->dst.dev = dev_out;
2432 rth->rt_gateway = fl->fl4_dst;
2433 rth->rt_spec_dst= fl->fl4_src;
2435 rth->dst.output=ip_output;
2436 rth->dst.obsolete = -1;
2437 rth->rt_genid = rt_genid(dev_net(dev_out));
2439 RT_CACHE_STAT_INC(out_slow_tot);
2441 if (flags & RTCF_LOCAL) {
2442 rth->dst.input = ip_local_deliver;
2443 rth->rt_spec_dst = fl->fl4_dst;
2445 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2446 rth->rt_spec_dst = fl->fl4_src;
2447 if (flags & RTCF_LOCAL &&
2448 !(dev_out->flags & IFF_LOOPBACK)) {
2449 rth->dst.output = ip_mc_output;
2450 RT_CACHE_STAT_INC(out_slow_mc);
2452 #ifdef CONFIG_IP_MROUTE
2453 if (res->type == RTN_MULTICAST) {
2454 if (IN_DEV_MFORWARD(in_dev) &&
2455 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2456 rth->dst.input = ip_mr_input;
2457 rth->dst.output = ip_mc_output;
2463 rt_set_nexthop(rth, res, 0);
2465 rth->rt_flags = flags;
2470 static int ip_mkroute_output(struct rtable **rp,
2471 struct fib_result *res,
2472 const struct flowi *fl,
2473 const struct flowi *oldflp,
2474 struct net_device *dev_out,
2477 struct rtable *rth = NULL;
2478 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2481 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2482 rt_genid(dev_net(dev_out)));
2483 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2490 * Major route resolver routine.
2491 * called with rcu_read_lock();
2494 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2495 const struct flowi *oldflp)
2497 u32 tos = RT_FL_TOS(oldflp);
2498 struct flowi fl = { .nl_u = { .ip4_u =
2499 { .daddr = oldflp->fl4_dst,
2500 .saddr = oldflp->fl4_src,
2501 .tos = tos & IPTOS_RT_MASK,
2502 .scope = ((tos & RTO_ONLINK) ?
2506 .mark = oldflp->mark,
2507 .iif = net->loopback_dev->ifindex,
2508 .oif = oldflp->oif };
2509 struct fib_result res;
2510 unsigned int flags = 0;
2511 struct net_device *dev_out = NULL;
2517 #ifdef CONFIG_IP_MULTIPLE_TABLES
2521 if (oldflp->fl4_src) {
2523 if (ipv4_is_multicast(oldflp->fl4_src) ||
2524 ipv4_is_lbcast(oldflp->fl4_src) ||
2525 ipv4_is_zeronet(oldflp->fl4_src))
2528 /* I removed check for oif == dev_out->oif here.
2529 It was wrong for two reasons:
2530 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2531 is assigned to multiple interfaces.
2532 2. Moreover, we are allowed to send packets with saddr
2533 of another iface. --ANK
2536 if (oldflp->oif == 0 &&
2537 (ipv4_is_multicast(oldflp->fl4_dst) ||
2538 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2539 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2540 dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2541 if (dev_out == NULL)
2544 /* Special hack: user can direct multicasts
2545 and limited broadcast via necessary interface
2546 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2547 This hack is not just for fun, it allows
2548 vic,vat and friends to work.
2549 They bind socket to loopback, set ttl to zero
2550 and expect that it will work.
2551 From the viewpoint of routing cache they are broken,
2552 because we are not allowed to build multicast path
2553 with loopback source addr (look, routing cache
2554 cannot know, that ttl is zero, so that packet
2555 will not leave this host and route is valid).
2556 Luckily, this hack is good workaround.
2559 fl.oif = dev_out->ifindex;
2563 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2564 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2565 if (!__ip_dev_find(net, oldflp->fl4_src, false))
2572 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2574 if (dev_out == NULL)
2577 /* RACE: Check return value of inet_select_addr instead. */
2578 if (rcu_dereference(dev_out->ip_ptr) == NULL)
2579 goto out; /* Wrong error code */
2581 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2582 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2584 fl.fl4_src = inet_select_addr(dev_out, 0,
2589 if (ipv4_is_multicast(oldflp->fl4_dst))
2590 fl.fl4_src = inet_select_addr(dev_out, 0,
2592 else if (!oldflp->fl4_dst)
2593 fl.fl4_src = inet_select_addr(dev_out, 0,
2599 fl.fl4_dst = fl.fl4_src;
2601 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2602 dev_out = net->loopback_dev;
2603 fl.oif = net->loopback_dev->ifindex;
2604 res.type = RTN_LOCAL;
2605 flags |= RTCF_LOCAL;
2609 if (fib_lookup(net, &fl, &res)) {
2612 /* Apparently, routing tables are wrong. Assume,
2613 that the destination is on link.
2616 Because we are allowed to send to iface
2617 even if it has NO routes and NO assigned
2618 addresses. When oif is specified, routing
2619 tables are looked up with only one purpose:
2620 to catch if destination is gatewayed, rather than
2621 direct. Moreover, if MSG_DONTROUTE is set,
2622 we send packet, ignoring both routing tables
2623 and ifaddr state. --ANK
2626 We could make it even if oif is unknown,
2627 likely IPv6, but we do not.
2630 if (fl.fl4_src == 0)
2631 fl.fl4_src = inet_select_addr(dev_out, 0,
2633 res.type = RTN_UNICAST;
2641 if (res.type == RTN_LOCAL) {
2643 fl.fl4_src = fl.fl4_dst;
2644 dev_out = net->loopback_dev;
2645 fl.oif = dev_out->ifindex;
2647 fib_info_put(res.fi);
2649 flags |= RTCF_LOCAL;
2653 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2654 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2655 fib_select_multipath(&fl, &res);
2658 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2659 fib_select_default(net, &fl, &res);
2662 fl.fl4_src = FIB_RES_PREFSRC(res);
2664 dev_out = FIB_RES_DEV(res);
2665 fl.oif = dev_out->ifindex;
2669 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2676 int __ip_route_output_key(struct net *net, struct rtable **rp,
2677 const struct flowi *flp)
2683 if (!rt_caching(net))
2686 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2689 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2690 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2691 if (rth->fl.fl4_dst == flp->fl4_dst &&
2692 rth->fl.fl4_src == flp->fl4_src &&
2694 rth->fl.oif == flp->oif &&
2695 rth->fl.mark == flp->mark &&
2696 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2697 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2698 net_eq(dev_net(rth->dst.dev), net) &&
2699 !rt_is_expired(rth)) {
2700 dst_use(&rth->dst, jiffies);
2701 RT_CACHE_STAT_INC(out_hit);
2702 rcu_read_unlock_bh();
2706 RT_CACHE_STAT_INC(out_hlist_search);
2708 rcu_read_unlock_bh();
2712 res = ip_route_output_slow(net, rp, flp);
2716 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2718 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2723 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2727 static struct dst_ops ipv4_dst_blackhole_ops = {
2729 .protocol = cpu_to_be16(ETH_P_IP),
2730 .destroy = ipv4_dst_destroy,
2731 .check = ipv4_blackhole_dst_check,
2732 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2733 .entries = ATOMIC_INIT(0),
2737 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2739 struct rtable *ort = *rp;
2740 struct rtable *rt = (struct rtable *)
2741 dst_alloc(&ipv4_dst_blackhole_ops);
2744 struct dst_entry *new = &rt->dst;
2746 atomic_set(&new->__refcnt, 1);
2748 new->input = dst_discard;
2749 new->output = dst_discard;
2750 memcpy(new->metrics, ort->dst.metrics, RTAX_MAX*sizeof(u32));
2752 new->dev = ort->dst.dev;
2758 rt->idev = ort->idev;
2760 in_dev_hold(rt->idev);
2761 rt->rt_genid = rt_genid(net);
2762 rt->rt_flags = ort->rt_flags;
2763 rt->rt_type = ort->rt_type;
2764 rt->rt_dst = ort->rt_dst;
2765 rt->rt_src = ort->rt_src;
2766 rt->rt_iif = ort->rt_iif;
2767 rt->rt_gateway = ort->rt_gateway;
2768 rt->rt_spec_dst = ort->rt_spec_dst;
2769 rt->peer = ort->peer;
2771 atomic_inc(&rt->peer->refcnt);
2776 dst_release(&(*rp)->dst);
2778 return rt ? 0 : -ENOMEM;
2781 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2782 struct sock *sk, int flags)
2786 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2791 flp->fl4_src = (*rp)->rt_src;
2793 flp->fl4_dst = (*rp)->rt_dst;
2794 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2795 flags ? XFRM_LOOKUP_WAIT : 0);
2796 if (err == -EREMOTE)
2797 err = ipv4_dst_blackhole(net, rp, flp);
2804 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2806 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2808 return ip_route_output_flow(net, rp, flp, NULL, 0);
2810 EXPORT_SYMBOL(ip_route_output_key);
2812 static int rt_fill_info(struct net *net,
2813 struct sk_buff *skb, u32 pid, u32 seq, int event,
2814 int nowait, unsigned int flags)
2816 struct rtable *rt = skb_rtable(skb);
2818 struct nlmsghdr *nlh;
2820 u32 id = 0, ts = 0, tsage = 0, error;
2822 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2826 r = nlmsg_data(nlh);
2827 r->rtm_family = AF_INET;
2828 r->rtm_dst_len = 32;
2830 r->rtm_tos = rt->fl.fl4_tos;
2831 r->rtm_table = RT_TABLE_MAIN;
2832 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2833 r->rtm_type = rt->rt_type;
2834 r->rtm_scope = RT_SCOPE_UNIVERSE;
2835 r->rtm_protocol = RTPROT_UNSPEC;
2836 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2837 if (rt->rt_flags & RTCF_NOTIFY)
2838 r->rtm_flags |= RTM_F_NOTIFY;
2840 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2842 if (rt->fl.fl4_src) {
2843 r->rtm_src_len = 32;
2844 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2847 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2848 #ifdef CONFIG_NET_CLS_ROUTE
2849 if (rt->dst.tclassid)
2850 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2853 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2854 else if (rt->rt_src != rt->fl.fl4_src)
2855 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2857 if (rt->rt_dst != rt->rt_gateway)
2858 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2860 if (rtnetlink_put_metrics(skb, rt->dst.metrics) < 0)
2861 goto nla_put_failure;
2864 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2866 error = rt->dst.error;
2867 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
2869 inet_peer_refcheck(rt->peer);
2870 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2871 if (rt->peer->tcp_ts_stamp) {
2872 ts = rt->peer->tcp_ts;
2873 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2878 #ifdef CONFIG_IP_MROUTE
2879 __be32 dst = rt->rt_dst;
2881 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2882 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2883 int err = ipmr_get_route(net, skb, r, nowait);
2888 goto nla_put_failure;
2890 if (err == -EMSGSIZE)
2891 goto nla_put_failure;
2897 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2900 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2901 expires, error) < 0)
2902 goto nla_put_failure;
2904 return nlmsg_end(skb, nlh);
2907 nlmsg_cancel(skb, nlh);
2911 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2913 struct net *net = sock_net(in_skb->sk);
2915 struct nlattr *tb[RTA_MAX+1];
2916 struct rtable *rt = NULL;
2922 struct sk_buff *skb;
2924 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2928 rtm = nlmsg_data(nlh);
2930 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2936 /* Reserve room for dummy headers, this skb can pass
2937 through good chunk of routing engine.
2939 skb_reset_mac_header(skb);
2940 skb_reset_network_header(skb);
2942 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2943 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2944 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2946 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2947 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2948 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2949 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2952 struct net_device *dev;
2954 dev = __dev_get_by_index(net, iif);
2960 skb->protocol = htons(ETH_P_IP);
2964 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2967 rt = skb_rtable(skb);
2968 if (err == 0 && rt->dst.error)
2969 err = -rt->dst.error;
2976 .tos = rtm->rtm_tos,
2979 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2982 err = ip_route_output_key(net, &rt, &fl);
2988 skb_dst_set(skb, &rt->dst);
2989 if (rtm->rtm_flags & RTM_F_NOTIFY)
2990 rt->rt_flags |= RTCF_NOTIFY;
2992 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2993 RTM_NEWROUTE, 0, 0);
2997 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3006 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3013 net = sock_net(skb->sk);
3018 s_idx = idx = cb->args[1];
3019 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3020 if (!rt_hash_table[h].chain)
3023 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3024 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3025 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3027 if (rt_is_expired(rt))
3029 skb_dst_set_noref(skb, &rt->dst);
3030 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3031 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3032 1, NLM_F_MULTI) <= 0) {
3034 rcu_read_unlock_bh();
3039 rcu_read_unlock_bh();
3048 void ip_rt_multicast_event(struct in_device *in_dev)
3050 rt_cache_flush(dev_net(in_dev->dev), 0);
3053 #ifdef CONFIG_SYSCTL
3054 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3055 void __user *buffer,
3056 size_t *lenp, loff_t *ppos)
3063 memcpy(&ctl, __ctl, sizeof(ctl));
3064 ctl.data = &flush_delay;
3065 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3067 net = (struct net *)__ctl->extra1;
3068 rt_cache_flush(net, flush_delay);
3075 static ctl_table ipv4_route_table[] = {
3077 .procname = "gc_thresh",
3078 .data = &ipv4_dst_ops.gc_thresh,
3079 .maxlen = sizeof(int),
3081 .proc_handler = proc_dointvec,
3084 .procname = "max_size",
3085 .data = &ip_rt_max_size,
3086 .maxlen = sizeof(int),
3088 .proc_handler = proc_dointvec,
3091 /* Deprecated. Use gc_min_interval_ms */
3093 .procname = "gc_min_interval",
3094 .data = &ip_rt_gc_min_interval,
3095 .maxlen = sizeof(int),
3097 .proc_handler = proc_dointvec_jiffies,
3100 .procname = "gc_min_interval_ms",
3101 .data = &ip_rt_gc_min_interval,
3102 .maxlen = sizeof(int),
3104 .proc_handler = proc_dointvec_ms_jiffies,
3107 .procname = "gc_timeout",
3108 .data = &ip_rt_gc_timeout,
3109 .maxlen = sizeof(int),
3111 .proc_handler = proc_dointvec_jiffies,
3114 .procname = "gc_interval",
3115 .data = &ip_rt_gc_interval,
3116 .maxlen = sizeof(int),
3118 .proc_handler = proc_dointvec_jiffies,
3121 .procname = "redirect_load",
3122 .data = &ip_rt_redirect_load,
3123 .maxlen = sizeof(int),
3125 .proc_handler = proc_dointvec,
3128 .procname = "redirect_number",
3129 .data = &ip_rt_redirect_number,
3130 .maxlen = sizeof(int),
3132 .proc_handler = proc_dointvec,
3135 .procname = "redirect_silence",
3136 .data = &ip_rt_redirect_silence,
3137 .maxlen = sizeof(int),
3139 .proc_handler = proc_dointvec,
3142 .procname = "error_cost",
3143 .data = &ip_rt_error_cost,
3144 .maxlen = sizeof(int),
3146 .proc_handler = proc_dointvec,
3149 .procname = "error_burst",
3150 .data = &ip_rt_error_burst,
3151 .maxlen = sizeof(int),
3153 .proc_handler = proc_dointvec,
3156 .procname = "gc_elasticity",
3157 .data = &ip_rt_gc_elasticity,
3158 .maxlen = sizeof(int),
3160 .proc_handler = proc_dointvec,
3163 .procname = "mtu_expires",
3164 .data = &ip_rt_mtu_expires,
3165 .maxlen = sizeof(int),
3167 .proc_handler = proc_dointvec_jiffies,
3170 .procname = "min_pmtu",
3171 .data = &ip_rt_min_pmtu,
3172 .maxlen = sizeof(int),
3174 .proc_handler = proc_dointvec,
3177 .procname = "min_adv_mss",
3178 .data = &ip_rt_min_advmss,
3179 .maxlen = sizeof(int),
3181 .proc_handler = proc_dointvec,
3186 static struct ctl_table empty[1];
3188 static struct ctl_table ipv4_skeleton[] =
3190 { .procname = "route",
3191 .mode = 0555, .child = ipv4_route_table},
3192 { .procname = "neigh",
3193 .mode = 0555, .child = empty},
3197 static __net_initdata struct ctl_path ipv4_path[] = {
3198 { .procname = "net", },
3199 { .procname = "ipv4", },
3203 static struct ctl_table ipv4_route_flush_table[] = {
3205 .procname = "flush",
3206 .maxlen = sizeof(int),
3208 .proc_handler = ipv4_sysctl_rtcache_flush,
3213 static __net_initdata struct ctl_path ipv4_route_path[] = {
3214 { .procname = "net", },
3215 { .procname = "ipv4", },
3216 { .procname = "route", },
3220 static __net_init int sysctl_route_net_init(struct net *net)
3222 struct ctl_table *tbl;
3224 tbl = ipv4_route_flush_table;
3225 if (!net_eq(net, &init_net)) {
3226 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3230 tbl[0].extra1 = net;
3232 net->ipv4.route_hdr =
3233 register_net_sysctl_table(net, ipv4_route_path, tbl);
3234 if (net->ipv4.route_hdr == NULL)
3239 if (tbl != ipv4_route_flush_table)
3245 static __net_exit void sysctl_route_net_exit(struct net *net)
3247 struct ctl_table *tbl;
3249 tbl = net->ipv4.route_hdr->ctl_table_arg;
3250 unregister_net_sysctl_table(net->ipv4.route_hdr);
3251 BUG_ON(tbl == ipv4_route_flush_table);
3255 static __net_initdata struct pernet_operations sysctl_route_ops = {
3256 .init = sysctl_route_net_init,
3257 .exit = sysctl_route_net_exit,
3261 static __net_init int rt_genid_init(struct net *net)
3263 get_random_bytes(&net->ipv4.rt_genid,
3264 sizeof(net->ipv4.rt_genid));
3268 static __net_initdata struct pernet_operations rt_genid_ops = {
3269 .init = rt_genid_init,
3273 #ifdef CONFIG_NET_CLS_ROUTE
3274 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3275 #endif /* CONFIG_NET_CLS_ROUTE */
3277 static __initdata unsigned long rhash_entries;
3278 static int __init set_rhash_entries(char *str)
3282 rhash_entries = simple_strtoul(str, &str, 0);
3285 __setup("rhash_entries=", set_rhash_entries);
3287 int __init ip_rt_init(void)
3291 #ifdef CONFIG_NET_CLS_ROUTE
3292 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3294 panic("IP: failed to allocate ip_rt_acct\n");
3297 ipv4_dst_ops.kmem_cachep =
3298 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3299 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3301 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3303 rt_hash_table = (struct rt_hash_bucket *)
3304 alloc_large_system_hash("IP route cache",
3305 sizeof(struct rt_hash_bucket),
3307 (totalram_pages >= 128 * 1024) ?
3312 rhash_entries ? 0 : 512 * 1024);
3313 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3314 rt_hash_lock_init();
3316 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3317 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3322 /* All the timers, started at system startup tend
3323 to synchronize. Perturb it a bit.
3325 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3326 expires_ljiffies = jiffies;
3327 schedule_delayed_work(&expires_work,
3328 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3330 if (ip_rt_proc_init())
3331 printk(KERN_ERR "Unable to create route proc files\n");
3334 xfrm4_init(ip_rt_max_size);
3336 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3338 #ifdef CONFIG_SYSCTL
3339 register_pernet_subsys(&sysctl_route_ops);
3341 register_pernet_subsys(&rt_genid_ops);
3345 #ifdef CONFIG_SYSCTL
3347 * We really need to sanitize the damn ipv4 init order, then all
3348 * this nonsense will go away.
3350 void __init ip_static_sysctl_init(void)
3352 register_sysctl_paths(ipv4_path, ipv4_skeleton);