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,
164 #define ECN_OR_COST(class) TC_PRIO_##class
166 const __u8 ip_tos2prio[16] = {
170 ECN_OR_COST(BESTEFFORT),
176 ECN_OR_COST(INTERACTIVE),
178 ECN_OR_COST(INTERACTIVE),
179 TC_PRIO_INTERACTIVE_BULK,
180 ECN_OR_COST(INTERACTIVE_BULK),
181 TC_PRIO_INTERACTIVE_BULK,
182 ECN_OR_COST(INTERACTIVE_BULK)
190 /* The locking scheme is rather straight forward:
192 * 1) Read-Copy Update protects the buckets of the central route hash.
193 * 2) Only writers remove entries, and they hold the lock
194 * as they look at rtable reference counts.
195 * 3) Only readers acquire references to rtable entries,
196 * they do so with atomic increments and with the
200 struct rt_hash_bucket {
201 struct rtable *chain;
204 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
205 defined(CONFIG_PROVE_LOCKING)
207 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
208 * The size of this table is a power of two and depends on the number of CPUS.
209 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
211 #ifdef CONFIG_LOCKDEP
212 # define RT_HASH_LOCK_SZ 256
215 # define RT_HASH_LOCK_SZ 4096
217 # define RT_HASH_LOCK_SZ 2048
219 # define RT_HASH_LOCK_SZ 1024
221 # define RT_HASH_LOCK_SZ 512
223 # define RT_HASH_LOCK_SZ 256
227 static spinlock_t *rt_hash_locks;
228 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
230 static __init void rt_hash_lock_init(void)
234 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
237 panic("IP: failed to allocate rt_hash_locks\n");
239 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
240 spin_lock_init(&rt_hash_locks[i]);
243 # define rt_hash_lock_addr(slot) NULL
245 static inline void rt_hash_lock_init(void)
250 static struct rt_hash_bucket *rt_hash_table __read_mostly;
251 static unsigned rt_hash_mask __read_mostly;
252 static unsigned int rt_hash_log __read_mostly;
254 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
255 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
257 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
260 return jhash_3words((__force u32)daddr, (__force u32)saddr,
265 static inline int rt_genid(struct net *net)
267 return atomic_read(&net->ipv4.rt_genid);
270 #ifdef CONFIG_PROC_FS
271 struct rt_cache_iter_state {
272 struct seq_net_private p;
277 static struct rtable *rt_cache_get_first(struct seq_file *seq)
279 struct rt_cache_iter_state *st = seq->private;
280 struct rtable *r = NULL;
282 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
283 if (!rt_hash_table[st->bucket].chain)
286 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
288 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
289 r->rt_genid == st->genid)
291 r = rcu_dereference_bh(r->dst.rt_next);
293 rcu_read_unlock_bh();
298 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
301 struct rt_cache_iter_state *st = seq->private;
305 rcu_read_unlock_bh();
307 if (--st->bucket < 0)
309 } while (!rt_hash_table[st->bucket].chain);
311 r = rt_hash_table[st->bucket].chain;
313 return rcu_dereference_bh(r);
316 static struct rtable *rt_cache_get_next(struct seq_file *seq,
319 struct rt_cache_iter_state *st = seq->private;
320 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
321 if (dev_net(r->dst.dev) != seq_file_net(seq))
323 if (r->rt_genid == st->genid)
329 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
331 struct rtable *r = rt_cache_get_first(seq);
334 while (pos && (r = rt_cache_get_next(seq, r)))
336 return pos ? NULL : r;
339 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
341 struct rt_cache_iter_state *st = seq->private;
343 return rt_cache_get_idx(seq, *pos - 1);
344 st->genid = rt_genid(seq_file_net(seq));
345 return SEQ_START_TOKEN;
348 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
352 if (v == SEQ_START_TOKEN)
353 r = rt_cache_get_first(seq);
355 r = rt_cache_get_next(seq, v);
360 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
362 if (v && v != SEQ_START_TOKEN)
363 rcu_read_unlock_bh();
366 static int rt_cache_seq_show(struct seq_file *seq, void *v)
368 if (v == SEQ_START_TOKEN)
369 seq_printf(seq, "%-127s\n",
370 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
371 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
374 struct rtable *r = v;
377 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
378 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
379 r->dst.dev ? r->dst.dev->name : "*",
380 (__force u32)r->rt_dst,
381 (__force u32)r->rt_gateway,
382 r->rt_flags, atomic_read(&r->dst.__refcnt),
383 r->dst.__use, 0, (__force u32)r->rt_src,
384 (dst_metric(&r->dst, RTAX_ADVMSS) ?
385 (int)dst_metric(&r->dst, RTAX_ADVMSS) + 40 : 0),
386 dst_metric(&r->dst, RTAX_WINDOW),
387 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
388 dst_metric(&r->dst, RTAX_RTTVAR)),
390 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
391 r->dst.hh ? (r->dst.hh->hh_output ==
393 r->rt_spec_dst, &len);
395 seq_printf(seq, "%*s\n", 127 - len, "");
400 static const struct seq_operations rt_cache_seq_ops = {
401 .start = rt_cache_seq_start,
402 .next = rt_cache_seq_next,
403 .stop = rt_cache_seq_stop,
404 .show = rt_cache_seq_show,
407 static int rt_cache_seq_open(struct inode *inode, struct file *file)
409 return seq_open_net(inode, file, &rt_cache_seq_ops,
410 sizeof(struct rt_cache_iter_state));
413 static const struct file_operations rt_cache_seq_fops = {
414 .owner = THIS_MODULE,
415 .open = rt_cache_seq_open,
418 .release = seq_release_net,
422 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
427 return SEQ_START_TOKEN;
429 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
430 if (!cpu_possible(cpu))
433 return &per_cpu(rt_cache_stat, cpu);
438 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
442 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
443 if (!cpu_possible(cpu))
446 return &per_cpu(rt_cache_stat, cpu);
452 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
457 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
459 struct rt_cache_stat *st = v;
461 if (v == SEQ_START_TOKEN) {
462 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");
466 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
467 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
468 dst_entries_get_slow(&ipv4_dst_ops),
491 static const struct seq_operations rt_cpu_seq_ops = {
492 .start = rt_cpu_seq_start,
493 .next = rt_cpu_seq_next,
494 .stop = rt_cpu_seq_stop,
495 .show = rt_cpu_seq_show,
499 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
501 return seq_open(file, &rt_cpu_seq_ops);
504 static const struct file_operations rt_cpu_seq_fops = {
505 .owner = THIS_MODULE,
506 .open = rt_cpu_seq_open,
509 .release = seq_release,
512 #ifdef CONFIG_NET_CLS_ROUTE
513 static int rt_acct_proc_show(struct seq_file *m, void *v)
515 struct ip_rt_acct *dst, *src;
518 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
522 for_each_possible_cpu(i) {
523 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
524 for (j = 0; j < 256; j++) {
525 dst[j].o_bytes += src[j].o_bytes;
526 dst[j].o_packets += src[j].o_packets;
527 dst[j].i_bytes += src[j].i_bytes;
528 dst[j].i_packets += src[j].i_packets;
532 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
537 static int rt_acct_proc_open(struct inode *inode, struct file *file)
539 return single_open(file, rt_acct_proc_show, NULL);
542 static const struct file_operations rt_acct_proc_fops = {
543 .owner = THIS_MODULE,
544 .open = rt_acct_proc_open,
547 .release = single_release,
551 static int __net_init ip_rt_do_proc_init(struct net *net)
553 struct proc_dir_entry *pde;
555 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
560 pde = proc_create("rt_cache", S_IRUGO,
561 net->proc_net_stat, &rt_cpu_seq_fops);
565 #ifdef CONFIG_NET_CLS_ROUTE
566 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
572 #ifdef CONFIG_NET_CLS_ROUTE
574 remove_proc_entry("rt_cache", net->proc_net_stat);
577 remove_proc_entry("rt_cache", net->proc_net);
582 static void __net_exit ip_rt_do_proc_exit(struct net *net)
584 remove_proc_entry("rt_cache", net->proc_net_stat);
585 remove_proc_entry("rt_cache", net->proc_net);
586 #ifdef CONFIG_NET_CLS_ROUTE
587 remove_proc_entry("rt_acct", net->proc_net);
591 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
592 .init = ip_rt_do_proc_init,
593 .exit = ip_rt_do_proc_exit,
596 static int __init ip_rt_proc_init(void)
598 return register_pernet_subsys(&ip_rt_proc_ops);
602 static inline int ip_rt_proc_init(void)
606 #endif /* CONFIG_PROC_FS */
608 static inline void rt_free(struct rtable *rt)
610 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
613 static inline void rt_drop(struct rtable *rt)
616 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
619 static inline int rt_fast_clean(struct rtable *rth)
621 /* Kill broadcast/multicast entries very aggresively, if they
622 collide in hash table with more useful entries */
623 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
624 rth->fl.iif && rth->dst.rt_next;
627 static inline int rt_valuable(struct rtable *rth)
629 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
633 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
638 if (atomic_read(&rth->dst.__refcnt))
642 if (rth->dst.expires &&
643 time_after_eq(jiffies, rth->dst.expires))
646 age = jiffies - rth->dst.lastuse;
648 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
649 (age <= tmo2 && rt_valuable(rth)))
655 /* Bits of score are:
657 * 30: not quite useless
658 * 29..0: usage counter
660 static inline u32 rt_score(struct rtable *rt)
662 u32 score = jiffies - rt->dst.lastuse;
664 score = ~score & ~(3<<30);
670 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
676 static inline bool rt_caching(const struct net *net)
678 return net->ipv4.current_rt_cache_rebuild_count <=
679 net->ipv4.sysctl_rt_cache_rebuild_count;
682 static inline bool compare_hash_inputs(const struct flowi *fl1,
683 const struct flowi *fl2)
685 return ((((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
686 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
687 (fl1->iif ^ fl2->iif)) == 0);
690 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
692 return (((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
693 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
694 (fl1->mark ^ fl2->mark) |
695 (*(u16 *)&fl1->nl_u.ip4_u.tos ^ *(u16 *)&fl2->nl_u.ip4_u.tos) |
696 (fl1->oif ^ fl2->oif) |
697 (fl1->iif ^ fl2->iif)) == 0;
700 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
702 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
705 static inline int rt_is_expired(struct rtable *rth)
707 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
711 * Perform a full scan of hash table and free all entries.
712 * Can be called by a softirq or a process.
713 * In the later case, we want to be reschedule if necessary
715 static void rt_do_flush(int process_context)
718 struct rtable *rth, *next;
719 struct rtable * tail;
721 for (i = 0; i <= rt_hash_mask; i++) {
722 if (process_context && need_resched())
724 rth = rt_hash_table[i].chain;
728 spin_lock_bh(rt_hash_lock_addr(i));
731 struct rtable ** prev, * p;
733 rth = rt_hash_table[i].chain;
735 /* defer releasing the head of the list after spin_unlock */
736 for (tail = rth; tail; tail = tail->dst.rt_next)
737 if (!rt_is_expired(tail))
740 rt_hash_table[i].chain = tail;
742 /* call rt_free on entries after the tail requiring flush */
743 prev = &rt_hash_table[i].chain;
744 for (p = *prev; p; p = next) {
745 next = p->dst.rt_next;
746 if (!rt_is_expired(p)) {
747 prev = &p->dst.rt_next;
755 rth = rt_hash_table[i].chain;
756 rt_hash_table[i].chain = NULL;
759 spin_unlock_bh(rt_hash_lock_addr(i));
761 for (; rth != tail; rth = next) {
762 next = rth->dst.rt_next;
769 * While freeing expired entries, we compute average chain length
770 * and standard deviation, using fixed-point arithmetic.
771 * This to have an estimation of rt_chain_length_max
772 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
773 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
777 #define ONE (1UL << FRACT_BITS)
780 * Given a hash chain and an item in this hash chain,
781 * find if a previous entry has the same hash_inputs
782 * (but differs on tos, mark or oif)
783 * Returns 0 if an alias is found.
784 * Returns ONE if rth has no alias before itself.
786 static int has_noalias(const struct rtable *head, const struct rtable *rth)
788 const struct rtable *aux = head;
791 if (compare_hash_inputs(&aux->fl, &rth->fl))
793 aux = aux->dst.rt_next;
798 static void rt_check_expire(void)
800 static unsigned int rover;
801 unsigned int i = rover, goal;
802 struct rtable *rth, **rthp;
803 unsigned long samples = 0;
804 unsigned long sum = 0, sum2 = 0;
808 delta = jiffies - expires_ljiffies;
809 expires_ljiffies = jiffies;
810 mult = ((u64)delta) << rt_hash_log;
811 if (ip_rt_gc_timeout > 1)
812 do_div(mult, ip_rt_gc_timeout);
813 goal = (unsigned int)mult;
814 if (goal > rt_hash_mask)
815 goal = rt_hash_mask + 1;
816 for (; goal > 0; goal--) {
817 unsigned long tmo = ip_rt_gc_timeout;
818 unsigned long length;
820 i = (i + 1) & rt_hash_mask;
821 rthp = &rt_hash_table[i].chain;
831 spin_lock_bh(rt_hash_lock_addr(i));
832 while ((rth = *rthp) != NULL) {
833 prefetch(rth->dst.rt_next);
834 if (rt_is_expired(rth)) {
835 *rthp = rth->dst.rt_next;
839 if (rth->dst.expires) {
840 /* Entry is expired even if it is in use */
841 if (time_before_eq(jiffies, rth->dst.expires)) {
844 rthp = &rth->dst.rt_next;
846 * We only count entries on
847 * a chain with equal hash inputs once
848 * so that entries for different QOS
849 * levels, and other non-hash input
850 * attributes don't unfairly skew
851 * the length computation
853 length += has_noalias(rt_hash_table[i].chain, rth);
856 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
859 /* Cleanup aged off entries. */
860 *rthp = rth->dst.rt_next;
863 spin_unlock_bh(rt_hash_lock_addr(i));
865 sum2 += length*length;
868 unsigned long avg = sum / samples;
869 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
870 rt_chain_length_max = max_t(unsigned long,
872 (avg + 4*sd) >> FRACT_BITS);
878 * rt_worker_func() is run in process context.
879 * we call rt_check_expire() to scan part of the hash table
881 static void rt_worker_func(struct work_struct *work)
884 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
888 * Pertubation of rt_genid by a small quantity [1..256]
889 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
890 * many times (2^24) without giving recent rt_genid.
891 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
893 static void rt_cache_invalidate(struct net *net)
895 unsigned char shuffle;
897 get_random_bytes(&shuffle, sizeof(shuffle));
898 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
902 * delay < 0 : invalidate cache (fast : entries will be deleted later)
903 * delay >= 0 : invalidate & flush cache (can be long)
905 void rt_cache_flush(struct net *net, int delay)
907 rt_cache_invalidate(net);
909 rt_do_flush(!in_softirq());
912 /* Flush previous cache invalidated entries from the cache */
913 void rt_cache_flush_batch(void)
915 rt_do_flush(!in_softirq());
918 static void rt_emergency_hash_rebuild(struct net *net)
921 printk(KERN_WARNING "Route hash chain too long!\n");
922 rt_cache_invalidate(net);
926 Short description of GC goals.
928 We want to build algorithm, which will keep routing cache
929 at some equilibrium point, when number of aged off entries
930 is kept approximately equal to newly generated ones.
932 Current expiration strength is variable "expire".
933 We try to adjust it dynamically, so that if networking
934 is idle expires is large enough to keep enough of warm entries,
935 and when load increases it reduces to limit cache size.
938 static int rt_garbage_collect(struct dst_ops *ops)
940 static unsigned long expire = RT_GC_TIMEOUT;
941 static unsigned long last_gc;
943 static int equilibrium;
944 struct rtable *rth, **rthp;
945 unsigned long now = jiffies;
947 int entries = dst_entries_get_fast(&ipv4_dst_ops);
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 entries < ip_rt_max_size) {
958 RT_CACHE_STAT_INC(gc_ignored);
962 entries = dst_entries_get_slow(&ipv4_dst_ops);
963 /* Calculate number of entries, which we want to expire now. */
964 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
966 if (equilibrium < ipv4_dst_ops.gc_thresh)
967 equilibrium = ipv4_dst_ops.gc_thresh;
968 goal = entries - equilibrium;
970 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
971 goal = 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 = 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 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
1038 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1040 } while (!in_softirq() && time_before_eq(jiffies, now));
1042 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1044 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1046 if (net_ratelimit())
1047 printk(KERN_WARNING "dst cache overflow\n");
1048 RT_CACHE_STAT_INC(gc_dst_overflow);
1052 expire += ip_rt_gc_min_interval;
1053 if (expire > ip_rt_gc_timeout ||
1054 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1055 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1056 expire = ip_rt_gc_timeout;
1057 #if RT_CACHE_DEBUG >= 2
1058 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1059 dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
1065 * Returns number of entries in a hash chain that have different hash_inputs
1067 static int slow_chain_length(const struct rtable *head)
1070 const struct rtable *rth = head;
1073 length += has_noalias(head, rth);
1074 rth = rth->dst.rt_next;
1076 return length >> FRACT_BITS;
1079 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1080 struct rtable **rp, struct sk_buff *skb, int ifindex)
1082 struct rtable *rth, **rthp;
1084 struct rtable *cand, **candp;
1087 int attempts = !in_softirq();
1091 min_score = ~(u32)0;
1096 if (!rt_caching(dev_net(rt->dst.dev))) {
1098 * If we're not caching, just tell the caller we
1099 * were successful and don't touch the route. The
1100 * caller hold the sole reference to the cache entry, and
1101 * it will be released when the caller is done with it.
1102 * If we drop it here, the callers have no way to resolve routes
1103 * when we're not caching. Instead, just point *rp at rt, so
1104 * the caller gets a single use out of the route
1105 * Note that we do rt_free on this new route entry, so that
1106 * once its refcount hits zero, we are still able to reap it
1108 * Note: To avoid expensive rcu stuff for this uncached dst,
1109 * we set DST_NOCACHE so that dst_release() can free dst without
1110 * waiting a grace period.
1113 rt->dst.flags |= DST_NOCACHE;
1114 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1115 int err = arp_bind_neighbour(&rt->dst);
1117 if (net_ratelimit())
1119 "Neighbour table failure & not caching routes.\n");
1128 rthp = &rt_hash_table[hash].chain;
1130 spin_lock_bh(rt_hash_lock_addr(hash));
1131 while ((rth = *rthp) != NULL) {
1132 if (rt_is_expired(rth)) {
1133 *rthp = rth->dst.rt_next;
1137 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1139 *rthp = rth->dst.rt_next;
1141 * Since lookup is lockfree, the deletion
1142 * must be visible to another weakly ordered CPU before
1143 * the insertion at the start of the hash chain.
1145 rcu_assign_pointer(rth->dst.rt_next,
1146 rt_hash_table[hash].chain);
1148 * Since lookup is lockfree, the update writes
1149 * must be ordered for consistency on SMP.
1151 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1153 dst_use(&rth->dst, now);
1154 spin_unlock_bh(rt_hash_lock_addr(hash));
1160 skb_dst_set(skb, &rth->dst);
1164 if (!atomic_read(&rth->dst.__refcnt)) {
1165 u32 score = rt_score(rth);
1167 if (score <= min_score) {
1176 rthp = &rth->dst.rt_next;
1180 /* ip_rt_gc_elasticity used to be average length of chain
1181 * length, when exceeded gc becomes really aggressive.
1183 * The second limit is less certain. At the moment it allows
1184 * only 2 entries per bucket. We will see.
1186 if (chain_length > ip_rt_gc_elasticity) {
1187 *candp = cand->dst.rt_next;
1191 if (chain_length > rt_chain_length_max &&
1192 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1193 struct net *net = dev_net(rt->dst.dev);
1194 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1195 if (!rt_caching(net)) {
1196 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1197 rt->dst.dev->name, num);
1199 rt_emergency_hash_rebuild(net);
1200 spin_unlock_bh(rt_hash_lock_addr(hash));
1202 hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1203 ifindex, rt_genid(net));
1208 /* Try to bind route to arp only if it is output
1209 route or unicast forwarding path.
1211 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1212 int err = arp_bind_neighbour(&rt->dst);
1214 spin_unlock_bh(rt_hash_lock_addr(hash));
1216 if (err != -ENOBUFS) {
1221 /* Neighbour tables are full and nothing
1222 can be released. Try to shrink route cache,
1223 it is most likely it holds some neighbour records.
1225 if (attempts-- > 0) {
1226 int saved_elasticity = ip_rt_gc_elasticity;
1227 int saved_int = ip_rt_gc_min_interval;
1228 ip_rt_gc_elasticity = 1;
1229 ip_rt_gc_min_interval = 0;
1230 rt_garbage_collect(&ipv4_dst_ops);
1231 ip_rt_gc_min_interval = saved_int;
1232 ip_rt_gc_elasticity = saved_elasticity;
1236 if (net_ratelimit())
1237 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1243 rt->dst.rt_next = rt_hash_table[hash].chain;
1245 #if RT_CACHE_DEBUG >= 2
1246 if (rt->dst.rt_next) {
1248 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1250 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1251 printk(" . %pI4", &trt->rt_dst);
1256 * Since lookup is lockfree, we must make sure
1257 * previous writes to rt are comitted to memory
1258 * before making rt visible to other CPUS.
1260 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1262 spin_unlock_bh(rt_hash_lock_addr(hash));
1268 skb_dst_set(skb, &rt->dst);
1272 void rt_bind_peer(struct rtable *rt, int create)
1274 struct inet_peer *peer;
1276 peer = inet_getpeer(rt->rt_dst, create);
1278 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1283 * Peer allocation may fail only in serious out-of-memory conditions. However
1284 * we still can generate some output.
1285 * Random ID selection looks a bit dangerous because we have no chances to
1286 * select ID being unique in a reasonable period of time.
1287 * But broken packet identifier may be better than no packet at all.
1289 static void ip_select_fb_ident(struct iphdr *iph)
1291 static DEFINE_SPINLOCK(ip_fb_id_lock);
1292 static u32 ip_fallback_id;
1295 spin_lock_bh(&ip_fb_id_lock);
1296 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1297 iph->id = htons(salt & 0xFFFF);
1298 ip_fallback_id = salt;
1299 spin_unlock_bh(&ip_fb_id_lock);
1302 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1304 struct rtable *rt = (struct rtable *) dst;
1307 if (rt->peer == NULL)
1308 rt_bind_peer(rt, 1);
1310 /* If peer is attached to destination, it is never detached,
1311 so that we need not to grab a lock to dereference it.
1314 iph->id = htons(inet_getid(rt->peer, more));
1318 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1319 __builtin_return_address(0));
1321 ip_select_fb_ident(iph);
1323 EXPORT_SYMBOL(__ip_select_ident);
1325 static void rt_del(unsigned hash, struct rtable *rt)
1327 struct rtable **rthp, *aux;
1329 rthp = &rt_hash_table[hash].chain;
1330 spin_lock_bh(rt_hash_lock_addr(hash));
1332 while ((aux = *rthp) != NULL) {
1333 if (aux == rt || rt_is_expired(aux)) {
1334 *rthp = aux->dst.rt_next;
1338 rthp = &aux->dst.rt_next;
1340 spin_unlock_bh(rt_hash_lock_addr(hash));
1343 /* called in rcu_read_lock() section */
1344 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1345 __be32 saddr, struct net_device *dev)
1348 struct in_device *in_dev = __in_dev_get_rcu(dev);
1349 struct rtable *rth, **rthp;
1350 __be32 skeys[2] = { saddr, 0 };
1351 int ikeys[2] = { dev->ifindex, 0 };
1352 struct netevent_redirect netevent;
1359 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1360 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1361 ipv4_is_zeronet(new_gw))
1362 goto reject_redirect;
1364 if (!rt_caching(net))
1365 goto reject_redirect;
1367 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1368 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1369 goto reject_redirect;
1370 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1371 goto reject_redirect;
1373 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1374 goto reject_redirect;
1377 for (i = 0; i < 2; i++) {
1378 for (k = 0; k < 2; k++) {
1379 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1382 rthp=&rt_hash_table[hash].chain;
1384 while ((rth = rcu_dereference(*rthp)) != NULL) {
1387 if (rth->fl.fl4_dst != daddr ||
1388 rth->fl.fl4_src != skeys[i] ||
1389 rth->fl.oif != ikeys[k] ||
1391 rt_is_expired(rth) ||
1392 !net_eq(dev_net(rth->dst.dev), net)) {
1393 rthp = &rth->dst.rt_next;
1397 if (rth->rt_dst != daddr ||
1398 rth->rt_src != saddr ||
1400 rth->rt_gateway != old_gw ||
1401 rth->dst.dev != dev)
1404 dst_hold(&rth->dst);
1406 rt = dst_alloc(&ipv4_dst_ops);
1412 /* Copy all the information. */
1415 atomic_set(&rt->dst.__refcnt, 1);
1416 rt->dst.child = NULL;
1418 dev_hold(rt->dst.dev);
1420 in_dev_hold(rt->idev);
1421 rt->dst.obsolete = -1;
1422 rt->dst.lastuse = jiffies;
1423 rt->dst.path = &rt->dst;
1424 rt->dst.neighbour = NULL;
1427 rt->dst.xfrm = NULL;
1429 rt->rt_genid = rt_genid(net);
1430 rt->rt_flags |= RTCF_REDIRECTED;
1432 /* Gateway is different ... */
1433 rt->rt_gateway = new_gw;
1435 /* Redirect received -> path was valid */
1436 dst_confirm(&rth->dst);
1439 atomic_inc(&rt->peer->refcnt);
1441 if (arp_bind_neighbour(&rt->dst) ||
1442 !(rt->dst.neighbour->nud_state &
1444 if (rt->dst.neighbour)
1445 neigh_event_send(rt->dst.neighbour, NULL);
1451 netevent.old = &rth->dst;
1452 netevent.new = &rt->dst;
1453 call_netevent_notifiers(NETEVENT_REDIRECT,
1457 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1468 #ifdef CONFIG_IP_ROUTE_VERBOSE
1469 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1470 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1471 " Advised path = %pI4 -> %pI4\n",
1472 &old_gw, dev->name, &new_gw,
1478 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1480 struct rtable *rt = (struct rtable *)dst;
1481 struct dst_entry *ret = dst;
1484 if (dst->obsolete > 0) {
1487 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1489 time_after_eq(jiffies, rt->dst.expires))) {
1490 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1492 rt_genid(dev_net(dst->dev)));
1493 #if RT_CACHE_DEBUG >= 1
1494 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1495 &rt->rt_dst, rt->fl.fl4_tos);
1506 * 1. The first ip_rt_redirect_number redirects are sent
1507 * with exponential backoff, then we stop sending them at all,
1508 * assuming that the host ignores our redirects.
1509 * 2. If we did not see packets requiring redirects
1510 * during ip_rt_redirect_silence, we assume that the host
1511 * forgot redirected route and start to send redirects again.
1513 * This algorithm is much cheaper and more intelligent than dumb load limiting
1516 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1517 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1520 void ip_rt_send_redirect(struct sk_buff *skb)
1522 struct rtable *rt = skb_rtable(skb);
1523 struct in_device *in_dev;
1527 in_dev = __in_dev_get_rcu(rt->dst.dev);
1528 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1532 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1535 /* No redirected packets during ip_rt_redirect_silence;
1536 * reset the algorithm.
1538 if (time_after(jiffies, rt->dst.rate_last + ip_rt_redirect_silence))
1539 rt->dst.rate_tokens = 0;
1541 /* Too many ignored redirects; do not send anything
1542 * set dst.rate_last to the last seen redirected packet.
1544 if (rt->dst.rate_tokens >= ip_rt_redirect_number) {
1545 rt->dst.rate_last = jiffies;
1549 /* Check for load limit; set rate_last to the latest sent
1552 if (rt->dst.rate_tokens == 0 ||
1554 (rt->dst.rate_last +
1555 (ip_rt_redirect_load << rt->dst.rate_tokens)))) {
1556 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1557 rt->dst.rate_last = jiffies;
1558 ++rt->dst.rate_tokens;
1559 #ifdef CONFIG_IP_ROUTE_VERBOSE
1561 rt->dst.rate_tokens == ip_rt_redirect_number &&
1563 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1564 &rt->rt_src, rt->rt_iif,
1565 &rt->rt_dst, &rt->rt_gateway);
1570 static int ip_error(struct sk_buff *skb)
1572 struct rtable *rt = skb_rtable(skb);
1576 switch (rt->dst.error) {
1581 code = ICMP_HOST_UNREACH;
1584 code = ICMP_NET_UNREACH;
1585 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1586 IPSTATS_MIB_INNOROUTES);
1589 code = ICMP_PKT_FILTERED;
1594 rt->dst.rate_tokens += now - rt->dst.rate_last;
1595 if (rt->dst.rate_tokens > ip_rt_error_burst)
1596 rt->dst.rate_tokens = ip_rt_error_burst;
1597 rt->dst.rate_last = now;
1598 if (rt->dst.rate_tokens >= ip_rt_error_cost) {
1599 rt->dst.rate_tokens -= ip_rt_error_cost;
1600 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1603 out: kfree_skb(skb);
1608 * The last two values are not from the RFC but
1609 * are needed for AMPRnet AX.25 paths.
1612 static const unsigned short mtu_plateau[] =
1613 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1615 static inline unsigned short guess_mtu(unsigned short old_mtu)
1619 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1620 if (old_mtu > mtu_plateau[i])
1621 return mtu_plateau[i];
1625 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1626 unsigned short new_mtu,
1627 struct net_device *dev)
1630 unsigned short old_mtu = ntohs(iph->tot_len);
1632 int ikeys[2] = { dev->ifindex, 0 };
1633 __be32 skeys[2] = { iph->saddr, 0, };
1634 __be32 daddr = iph->daddr;
1635 unsigned short est_mtu = 0;
1637 for (k = 0; k < 2; k++) {
1638 for (i = 0; i < 2; i++) {
1639 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1643 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1644 rth = rcu_dereference(rth->dst.rt_next)) {
1645 unsigned short mtu = new_mtu;
1647 if (rth->fl.fl4_dst != daddr ||
1648 rth->fl.fl4_src != skeys[i] ||
1649 rth->rt_dst != daddr ||
1650 rth->rt_src != iph->saddr ||
1651 rth->fl.oif != ikeys[k] ||
1653 dst_metric_locked(&rth->dst, RTAX_MTU) ||
1654 !net_eq(dev_net(rth->dst.dev), net) ||
1658 if (new_mtu < 68 || new_mtu >= old_mtu) {
1660 /* BSD 4.2 compatibility hack :-( */
1662 old_mtu >= dst_mtu(&rth->dst) &&
1663 old_mtu >= 68 + (iph->ihl << 2))
1664 old_mtu -= iph->ihl << 2;
1666 mtu = guess_mtu(old_mtu);
1668 if (mtu <= dst_mtu(&rth->dst)) {
1669 if (mtu < dst_mtu(&rth->dst)) {
1670 dst_confirm(&rth->dst);
1671 if (mtu < ip_rt_min_pmtu) {
1672 mtu = ip_rt_min_pmtu;
1673 rth->dst.metrics[RTAX_LOCK-1] |=
1676 rth->dst.metrics[RTAX_MTU-1] = mtu;
1677 dst_set_expires(&rth->dst,
1686 return est_mtu ? : new_mtu;
1689 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1691 if (dst_mtu(dst) > mtu && mtu >= 68 &&
1692 !(dst_metric_locked(dst, RTAX_MTU))) {
1693 if (mtu < ip_rt_min_pmtu) {
1694 mtu = ip_rt_min_pmtu;
1695 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1697 dst->metrics[RTAX_MTU-1] = mtu;
1698 dst_set_expires(dst, ip_rt_mtu_expires);
1699 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1703 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1705 if (rt_is_expired((struct rtable *)dst))
1710 static void ipv4_dst_destroy(struct dst_entry *dst)
1712 struct rtable *rt = (struct rtable *) dst;
1713 struct inet_peer *peer = rt->peer;
1714 struct in_device *idev = rt->idev;
1727 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1730 struct rtable *rt = (struct rtable *) dst;
1731 struct in_device *idev = rt->idev;
1732 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1733 struct in_device *loopback_idev =
1734 in_dev_get(dev_net(dev)->loopback_dev);
1735 if (loopback_idev) {
1736 rt->idev = loopback_idev;
1742 static void ipv4_link_failure(struct sk_buff *skb)
1746 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1748 rt = skb_rtable(skb);
1750 dst_set_expires(&rt->dst, 0);
1753 static int ip_rt_bug(struct sk_buff *skb)
1755 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1756 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1757 skb->dev ? skb->dev->name : "?");
1763 We do not cache source address of outgoing interface,
1764 because it is used only by IP RR, TS and SRR options,
1765 so that it out of fast path.
1767 BTW remember: "addr" is allowed to be not aligned
1771 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1774 struct fib_result res;
1776 if (rt->fl.iif == 0)
1780 if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0)
1781 src = FIB_RES_PREFSRC(res);
1783 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1787 memcpy(addr, &src, 4);
1790 #ifdef CONFIG_NET_CLS_ROUTE
1791 static void set_class_tag(struct rtable *rt, u32 tag)
1793 if (!(rt->dst.tclassid & 0xFFFF))
1794 rt->dst.tclassid |= tag & 0xFFFF;
1795 if (!(rt->dst.tclassid & 0xFFFF0000))
1796 rt->dst.tclassid |= tag & 0xFFFF0000;
1800 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1802 struct fib_info *fi = res->fi;
1805 if (FIB_RES_GW(*res) &&
1806 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1807 rt->rt_gateway = FIB_RES_GW(*res);
1808 memcpy(rt->dst.metrics, fi->fib_metrics,
1809 sizeof(rt->dst.metrics));
1810 if (fi->fib_mtu == 0) {
1811 rt->dst.metrics[RTAX_MTU-1] = rt->dst.dev->mtu;
1812 if (dst_metric_locked(&rt->dst, RTAX_MTU) &&
1813 rt->rt_gateway != rt->rt_dst &&
1814 rt->dst.dev->mtu > 576)
1815 rt->dst.metrics[RTAX_MTU-1] = 576;
1817 #ifdef CONFIG_NET_CLS_ROUTE
1818 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1821 rt->dst.metrics[RTAX_MTU-1]= rt->dst.dev->mtu;
1823 if (dst_metric(&rt->dst, RTAX_HOPLIMIT) == 0)
1824 rt->dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1825 if (dst_mtu(&rt->dst) > IP_MAX_MTU)
1826 rt->dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1827 if (dst_metric(&rt->dst, RTAX_ADVMSS) == 0)
1828 rt->dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->dst.dev->mtu - 40,
1830 if (dst_metric(&rt->dst, RTAX_ADVMSS) > 65535 - 40)
1831 rt->dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1833 #ifdef CONFIG_NET_CLS_ROUTE
1834 #ifdef CONFIG_IP_MULTIPLE_TABLES
1835 set_class_tag(rt, fib_rules_tclass(res));
1837 set_class_tag(rt, itag);
1839 rt->rt_type = res->type;
1842 /* called in rcu_read_lock() section */
1843 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1844 u8 tos, struct net_device *dev, int our)
1849 struct in_device *in_dev = __in_dev_get_rcu(dev);
1853 /* Primary sanity checks. */
1858 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1859 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1862 if (ipv4_is_zeronet(saddr)) {
1863 if (!ipv4_is_local_multicast(daddr))
1865 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1867 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1872 rth = dst_alloc(&ipv4_dst_ops);
1876 rth->dst.output = ip_rt_bug;
1877 rth->dst.obsolete = -1;
1879 atomic_set(&rth->dst.__refcnt, 1);
1880 rth->dst.flags= DST_HOST;
1881 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1882 rth->dst.flags |= DST_NOPOLICY;
1883 rth->fl.fl4_dst = daddr;
1884 rth->rt_dst = daddr;
1885 rth->fl.fl4_tos = tos;
1886 rth->fl.mark = skb->mark;
1887 rth->fl.fl4_src = saddr;
1888 rth->rt_src = saddr;
1889 #ifdef CONFIG_NET_CLS_ROUTE
1890 rth->dst.tclassid = itag;
1893 rth->fl.iif = dev->ifindex;
1894 rth->dst.dev = init_net.loopback_dev;
1895 dev_hold(rth->dst.dev);
1896 rth->idev = in_dev_get(rth->dst.dev);
1898 rth->rt_gateway = daddr;
1899 rth->rt_spec_dst= spec_dst;
1900 rth->rt_genid = rt_genid(dev_net(dev));
1901 rth->rt_flags = RTCF_MULTICAST;
1902 rth->rt_type = RTN_MULTICAST;
1904 rth->dst.input= ip_local_deliver;
1905 rth->rt_flags |= RTCF_LOCAL;
1908 #ifdef CONFIG_IP_MROUTE
1909 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1910 rth->dst.input = ip_mr_input;
1912 RT_CACHE_STAT_INC(in_slow_mc);
1914 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1915 return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1926 static void ip_handle_martian_source(struct net_device *dev,
1927 struct in_device *in_dev,
1928 struct sk_buff *skb,
1932 RT_CACHE_STAT_INC(in_martian_src);
1933 #ifdef CONFIG_IP_ROUTE_VERBOSE
1934 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1936 * RFC1812 recommendation, if source is martian,
1937 * the only hint is MAC header.
1939 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1940 &daddr, &saddr, dev->name);
1941 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1943 const unsigned char *p = skb_mac_header(skb);
1944 printk(KERN_WARNING "ll header: ");
1945 for (i = 0; i < dev->hard_header_len; i++, p++) {
1947 if (i < (dev->hard_header_len - 1))
1956 /* called in rcu_read_lock() section */
1957 static int __mkroute_input(struct sk_buff *skb,
1958 struct fib_result *res,
1959 struct in_device *in_dev,
1960 __be32 daddr, __be32 saddr, u32 tos,
1961 struct rtable **result)
1965 struct in_device *out_dev;
1966 unsigned int flags = 0;
1970 /* get a working reference to the output device */
1971 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1972 if (out_dev == NULL) {
1973 if (net_ratelimit())
1974 printk(KERN_CRIT "Bug in ip_route_input" \
1975 "_slow(). Please, report\n");
1980 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1981 in_dev->dev, &spec_dst, &itag, skb->mark);
1983 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1990 flags |= RTCF_DIRECTSRC;
1992 if (out_dev == in_dev && err &&
1993 (IN_DEV_SHARED_MEDIA(out_dev) ||
1994 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1995 flags |= RTCF_DOREDIRECT;
1997 if (skb->protocol != htons(ETH_P_IP)) {
1998 /* Not IP (i.e. ARP). Do not create route, if it is
1999 * invalid for proxy arp. DNAT routes are always valid.
2001 * Proxy arp feature have been extended to allow, ARP
2002 * replies back to the same interface, to support
2003 * Private VLAN switch technologies. See arp.c.
2005 if (out_dev == in_dev &&
2006 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2013 rth = dst_alloc(&ipv4_dst_ops);
2019 atomic_set(&rth->dst.__refcnt, 1);
2020 rth->dst.flags= DST_HOST;
2021 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2022 rth->dst.flags |= DST_NOPOLICY;
2023 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2024 rth->dst.flags |= DST_NOXFRM;
2025 rth->fl.fl4_dst = daddr;
2026 rth->rt_dst = daddr;
2027 rth->fl.fl4_tos = tos;
2028 rth->fl.mark = skb->mark;
2029 rth->fl.fl4_src = saddr;
2030 rth->rt_src = saddr;
2031 rth->rt_gateway = daddr;
2033 rth->fl.iif = in_dev->dev->ifindex;
2034 rth->dst.dev = (out_dev)->dev;
2035 dev_hold(rth->dst.dev);
2036 rth->idev = in_dev_get(rth->dst.dev);
2038 rth->rt_spec_dst= spec_dst;
2040 rth->dst.obsolete = -1;
2041 rth->dst.input = ip_forward;
2042 rth->dst.output = ip_output;
2043 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2045 rt_set_nexthop(rth, res, itag);
2047 rth->rt_flags = flags;
2055 static int ip_mkroute_input(struct sk_buff *skb,
2056 struct fib_result *res,
2057 const struct flowi *fl,
2058 struct in_device *in_dev,
2059 __be32 daddr, __be32 saddr, u32 tos)
2061 struct rtable* rth = NULL;
2065 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2066 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2067 fib_select_multipath(fl, res);
2070 /* create a routing cache entry */
2071 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2075 /* put it into the cache */
2076 hash = rt_hash(daddr, saddr, fl->iif,
2077 rt_genid(dev_net(rth->dst.dev)));
2078 return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2082 * NOTE. We drop all the packets that has local source
2083 * addresses, because every properly looped back packet
2084 * must have correct destination already attached by output routine.
2086 * Such approach solves two big problems:
2087 * 1. Not simplex devices are handled properly.
2088 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2089 * called with rcu_read_lock()
2092 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2093 u8 tos, struct net_device *dev)
2095 struct fib_result res;
2096 struct in_device *in_dev = __in_dev_get_rcu(dev);
2097 struct flowi fl = { .nl_u = { .ip4_u =
2101 .scope = RT_SCOPE_UNIVERSE,
2104 .iif = dev->ifindex };
2107 struct rtable * rth;
2111 struct net * net = dev_net(dev);
2113 /* IP on this device is disabled. */
2118 /* Check for the most weird martians, which can be not detected
2122 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2123 ipv4_is_loopback(saddr))
2124 goto martian_source;
2126 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2129 /* Accept zero addresses only to limited broadcast;
2130 * I even do not know to fix it or not. Waiting for complains :-)
2132 if (ipv4_is_zeronet(saddr))
2133 goto martian_source;
2135 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2136 goto martian_destination;
2139 * Now we are ready to route packet.
2141 err = fib_lookup(net, &fl, &res);
2143 if (!IN_DEV_FORWARD(in_dev))
2148 RT_CACHE_STAT_INC(in_slow_tot);
2150 if (res.type == RTN_BROADCAST)
2153 if (res.type == RTN_LOCAL) {
2154 err = fib_validate_source(saddr, daddr, tos,
2155 net->loopback_dev->ifindex,
2156 dev, &spec_dst, &itag, skb->mark);
2158 goto martian_source_keep_err;
2160 flags |= RTCF_DIRECTSRC;
2165 if (!IN_DEV_FORWARD(in_dev))
2167 if (res.type != RTN_UNICAST)
2168 goto martian_destination;
2170 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2174 if (skb->protocol != htons(ETH_P_IP))
2177 if (ipv4_is_zeronet(saddr))
2178 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2180 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2183 goto martian_source_keep_err;
2185 flags |= RTCF_DIRECTSRC;
2187 flags |= RTCF_BROADCAST;
2188 res.type = RTN_BROADCAST;
2189 RT_CACHE_STAT_INC(in_brd);
2192 rth = dst_alloc(&ipv4_dst_ops);
2196 rth->dst.output= ip_rt_bug;
2197 rth->dst.obsolete = -1;
2198 rth->rt_genid = rt_genid(net);
2200 atomic_set(&rth->dst.__refcnt, 1);
2201 rth->dst.flags= DST_HOST;
2202 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2203 rth->dst.flags |= DST_NOPOLICY;
2204 rth->fl.fl4_dst = daddr;
2205 rth->rt_dst = daddr;
2206 rth->fl.fl4_tos = tos;
2207 rth->fl.mark = skb->mark;
2208 rth->fl.fl4_src = saddr;
2209 rth->rt_src = saddr;
2210 #ifdef CONFIG_NET_CLS_ROUTE
2211 rth->dst.tclassid = itag;
2214 rth->fl.iif = dev->ifindex;
2215 rth->dst.dev = net->loopback_dev;
2216 dev_hold(rth->dst.dev);
2217 rth->idev = in_dev_get(rth->dst.dev);
2218 rth->rt_gateway = daddr;
2219 rth->rt_spec_dst= spec_dst;
2220 rth->dst.input= ip_local_deliver;
2221 rth->rt_flags = flags|RTCF_LOCAL;
2222 if (res.type == RTN_UNREACHABLE) {
2223 rth->dst.input= ip_error;
2224 rth->dst.error= -err;
2225 rth->rt_flags &= ~RTCF_LOCAL;
2227 rth->rt_type = res.type;
2228 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2229 err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2233 RT_CACHE_STAT_INC(in_no_route);
2234 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2235 res.type = RTN_UNREACHABLE;
2241 * Do not cache martian addresses: they should be logged (RFC1812)
2243 martian_destination:
2244 RT_CACHE_STAT_INC(in_martian_dst);
2245 #ifdef CONFIG_IP_ROUTE_VERBOSE
2246 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2247 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2248 &daddr, &saddr, dev->name);
2252 err = -EHOSTUNREACH;
2265 martian_source_keep_err:
2266 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2270 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2271 u8 tos, struct net_device *dev, bool noref)
2273 struct rtable * rth;
2275 int iif = dev->ifindex;
2283 if (!rt_caching(net))
2286 tos &= IPTOS_RT_MASK;
2287 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2289 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2290 rth = rcu_dereference(rth->dst.rt_next)) {
2291 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2292 ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2293 (rth->fl.iif ^ iif) |
2295 (rth->fl.fl4_tos ^ tos)) == 0 &&
2296 rth->fl.mark == skb->mark &&
2297 net_eq(dev_net(rth->dst.dev), net) &&
2298 !rt_is_expired(rth)) {
2300 dst_use_noref(&rth->dst, jiffies);
2301 skb_dst_set_noref(skb, &rth->dst);
2303 dst_use(&rth->dst, jiffies);
2304 skb_dst_set(skb, &rth->dst);
2306 RT_CACHE_STAT_INC(in_hit);
2310 RT_CACHE_STAT_INC(in_hlist_search);
2314 /* Multicast recognition logic is moved from route cache to here.
2315 The problem was that too many Ethernet cards have broken/missing
2316 hardware multicast filters :-( As result the host on multicasting
2317 network acquires a lot of useless route cache entries, sort of
2318 SDR messages from all the world. Now we try to get rid of them.
2319 Really, provided software IP multicast filter is organized
2320 reasonably (at least, hashed), it does not result in a slowdown
2321 comparing with route cache reject entries.
2322 Note, that multicast routers are not affected, because
2323 route cache entry is created eventually.
2325 if (ipv4_is_multicast(daddr)) {
2326 struct in_device *in_dev = __in_dev_get_rcu(dev);
2329 int our = ip_check_mc(in_dev, daddr, saddr,
2330 ip_hdr(skb)->protocol);
2332 #ifdef CONFIG_IP_MROUTE
2334 (!ipv4_is_local_multicast(daddr) &&
2335 IN_DEV_MFORWARD(in_dev))
2338 int res = ip_route_input_mc(skb, daddr, saddr,
2347 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2351 EXPORT_SYMBOL(ip_route_input_common);
2353 /* called with rcu_read_lock() */
2354 static int __mkroute_output(struct rtable **result,
2355 struct fib_result *res,
2356 const struct flowi *fl,
2357 const struct flowi *oldflp,
2358 struct net_device *dev_out,
2362 struct in_device *in_dev;
2363 u32 tos = RT_FL_TOS(oldflp);
2365 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2368 if (ipv4_is_lbcast(fl->fl4_dst))
2369 res->type = RTN_BROADCAST;
2370 else if (ipv4_is_multicast(fl->fl4_dst))
2371 res->type = RTN_MULTICAST;
2372 else if (ipv4_is_zeronet(fl->fl4_dst))
2375 if (dev_out->flags & IFF_LOOPBACK)
2376 flags |= RTCF_LOCAL;
2378 in_dev = __in_dev_get_rcu(dev_out);
2382 if (res->type == RTN_BROADCAST) {
2383 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2385 } else if (res->type == RTN_MULTICAST) {
2386 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2387 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2389 flags &= ~RTCF_LOCAL;
2390 /* If multicast route do not exist use
2391 * default one, but do not gateway in this case.
2394 if (res->fi && res->prefixlen < 4)
2399 rth = dst_alloc(&ipv4_dst_ops);
2403 in_dev_hold(in_dev);
2406 atomic_set(&rth->dst.__refcnt, 1);
2407 rth->dst.flags= DST_HOST;
2408 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2409 rth->dst.flags |= DST_NOXFRM;
2410 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2411 rth->dst.flags |= DST_NOPOLICY;
2413 rth->fl.fl4_dst = oldflp->fl4_dst;
2414 rth->fl.fl4_tos = tos;
2415 rth->fl.fl4_src = oldflp->fl4_src;
2416 rth->fl.oif = oldflp->oif;
2417 rth->fl.mark = oldflp->mark;
2418 rth->rt_dst = fl->fl4_dst;
2419 rth->rt_src = fl->fl4_src;
2420 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2421 /* get references to the devices that are to be hold by the routing
2423 rth->dst.dev = dev_out;
2425 rth->rt_gateway = fl->fl4_dst;
2426 rth->rt_spec_dst= fl->fl4_src;
2428 rth->dst.output=ip_output;
2429 rth->dst.obsolete = -1;
2430 rth->rt_genid = rt_genid(dev_net(dev_out));
2432 RT_CACHE_STAT_INC(out_slow_tot);
2434 if (flags & RTCF_LOCAL) {
2435 rth->dst.input = ip_local_deliver;
2436 rth->rt_spec_dst = fl->fl4_dst;
2438 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2439 rth->rt_spec_dst = fl->fl4_src;
2440 if (flags & RTCF_LOCAL &&
2441 !(dev_out->flags & IFF_LOOPBACK)) {
2442 rth->dst.output = ip_mc_output;
2443 RT_CACHE_STAT_INC(out_slow_mc);
2445 #ifdef CONFIG_IP_MROUTE
2446 if (res->type == RTN_MULTICAST) {
2447 if (IN_DEV_MFORWARD(in_dev) &&
2448 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2449 rth->dst.input = ip_mr_input;
2450 rth->dst.output = ip_mc_output;
2456 rt_set_nexthop(rth, res, 0);
2458 rth->rt_flags = flags;
2463 /* called with rcu_read_lock() */
2464 static int ip_mkroute_output(struct rtable **rp,
2465 struct fib_result *res,
2466 const struct flowi *fl,
2467 const struct flowi *oldflp,
2468 struct net_device *dev_out,
2471 struct rtable *rth = NULL;
2472 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2475 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2476 rt_genid(dev_net(dev_out)));
2477 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2484 * Major route resolver routine.
2485 * called with rcu_read_lock();
2488 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2489 const struct flowi *oldflp)
2491 u32 tos = RT_FL_TOS(oldflp);
2492 struct flowi fl = { .nl_u = { .ip4_u =
2493 { .daddr = oldflp->fl4_dst,
2494 .saddr = oldflp->fl4_src,
2495 .tos = tos & IPTOS_RT_MASK,
2496 .scope = ((tos & RTO_ONLINK) ?
2500 .mark = oldflp->mark,
2501 .iif = net->loopback_dev->ifindex,
2502 .oif = oldflp->oif };
2503 struct fib_result res;
2504 unsigned int flags = 0;
2505 struct net_device *dev_out = NULL;
2510 #ifdef CONFIG_IP_MULTIPLE_TABLES
2514 if (oldflp->fl4_src) {
2516 if (ipv4_is_multicast(oldflp->fl4_src) ||
2517 ipv4_is_lbcast(oldflp->fl4_src) ||
2518 ipv4_is_zeronet(oldflp->fl4_src))
2521 /* I removed check for oif == dev_out->oif here.
2522 It was wrong for two reasons:
2523 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2524 is assigned to multiple interfaces.
2525 2. Moreover, we are allowed to send packets with saddr
2526 of another iface. --ANK
2529 if (oldflp->oif == 0 &&
2530 (ipv4_is_multicast(oldflp->fl4_dst) ||
2531 ipv4_is_lbcast(oldflp->fl4_dst))) {
2532 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2533 dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2534 if (dev_out == NULL)
2537 /* Special hack: user can direct multicasts
2538 and limited broadcast via necessary interface
2539 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2540 This hack is not just for fun, it allows
2541 vic,vat and friends to work.
2542 They bind socket to loopback, set ttl to zero
2543 and expect that it will work.
2544 From the viewpoint of routing cache they are broken,
2545 because we are not allowed to build multicast path
2546 with loopback source addr (look, routing cache
2547 cannot know, that ttl is zero, so that packet
2548 will not leave this host and route is valid).
2549 Luckily, this hack is good workaround.
2552 fl.oif = dev_out->ifindex;
2556 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2557 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2558 if (!__ip_dev_find(net, oldflp->fl4_src, false))
2565 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2567 if (dev_out == NULL)
2570 /* RACE: Check return value of inet_select_addr instead. */
2571 if (rcu_dereference(dev_out->ip_ptr) == NULL)
2572 goto out; /* Wrong error code */
2574 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2575 ipv4_is_lbcast(oldflp->fl4_dst)) {
2577 fl.fl4_src = inet_select_addr(dev_out, 0,
2582 if (ipv4_is_multicast(oldflp->fl4_dst))
2583 fl.fl4_src = inet_select_addr(dev_out, 0,
2585 else if (!oldflp->fl4_dst)
2586 fl.fl4_src = inet_select_addr(dev_out, 0,
2592 fl.fl4_dst = fl.fl4_src;
2594 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2595 dev_out = net->loopback_dev;
2596 fl.oif = net->loopback_dev->ifindex;
2597 res.type = RTN_LOCAL;
2598 flags |= RTCF_LOCAL;
2602 if (fib_lookup(net, &fl, &res)) {
2605 /* Apparently, routing tables are wrong. Assume,
2606 that the destination is on link.
2609 Because we are allowed to send to iface
2610 even if it has NO routes and NO assigned
2611 addresses. When oif is specified, routing
2612 tables are looked up with only one purpose:
2613 to catch if destination is gatewayed, rather than
2614 direct. Moreover, if MSG_DONTROUTE is set,
2615 we send packet, ignoring both routing tables
2616 and ifaddr state. --ANK
2619 We could make it even if oif is unknown,
2620 likely IPv6, but we do not.
2623 if (fl.fl4_src == 0)
2624 fl.fl4_src = inet_select_addr(dev_out, 0,
2626 res.type = RTN_UNICAST;
2633 if (res.type == RTN_LOCAL) {
2635 fl.fl4_src = fl.fl4_dst;
2636 dev_out = net->loopback_dev;
2637 fl.oif = dev_out->ifindex;
2639 flags |= RTCF_LOCAL;
2643 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2644 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2645 fib_select_multipath(&fl, &res);
2648 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2649 fib_select_default(net, &fl, &res);
2652 fl.fl4_src = FIB_RES_PREFSRC(res);
2654 dev_out = FIB_RES_DEV(res);
2655 fl.oif = dev_out->ifindex;
2659 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2664 int __ip_route_output_key(struct net *net, struct rtable **rp,
2665 const struct flowi *flp)
2671 if (!rt_caching(net))
2674 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2677 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2678 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2679 if (rth->fl.fl4_dst == flp->fl4_dst &&
2680 rth->fl.fl4_src == flp->fl4_src &&
2682 rth->fl.oif == flp->oif &&
2683 rth->fl.mark == flp->mark &&
2684 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2685 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2686 net_eq(dev_net(rth->dst.dev), net) &&
2687 !rt_is_expired(rth)) {
2688 dst_use(&rth->dst, jiffies);
2689 RT_CACHE_STAT_INC(out_hit);
2690 rcu_read_unlock_bh();
2694 RT_CACHE_STAT_INC(out_hlist_search);
2696 rcu_read_unlock_bh();
2700 res = ip_route_output_slow(net, rp, flp);
2704 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2706 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2711 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2715 static struct dst_ops ipv4_dst_blackhole_ops = {
2717 .protocol = cpu_to_be16(ETH_P_IP),
2718 .destroy = ipv4_dst_destroy,
2719 .check = ipv4_blackhole_dst_check,
2720 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2724 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2726 struct rtable *ort = *rp;
2727 struct rtable *rt = (struct rtable *)
2728 dst_alloc(&ipv4_dst_blackhole_ops);
2731 struct dst_entry *new = &rt->dst;
2733 atomic_set(&new->__refcnt, 1);
2735 new->input = dst_discard;
2736 new->output = dst_discard;
2737 memcpy(new->metrics, ort->dst.metrics, RTAX_MAX*sizeof(u32));
2739 new->dev = ort->dst.dev;
2745 rt->idev = ort->idev;
2747 in_dev_hold(rt->idev);
2748 rt->rt_genid = rt_genid(net);
2749 rt->rt_flags = ort->rt_flags;
2750 rt->rt_type = ort->rt_type;
2751 rt->rt_dst = ort->rt_dst;
2752 rt->rt_src = ort->rt_src;
2753 rt->rt_iif = ort->rt_iif;
2754 rt->rt_gateway = ort->rt_gateway;
2755 rt->rt_spec_dst = ort->rt_spec_dst;
2756 rt->peer = ort->peer;
2758 atomic_inc(&rt->peer->refcnt);
2763 dst_release(&(*rp)->dst);
2765 return rt ? 0 : -ENOMEM;
2768 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2769 struct sock *sk, int flags)
2773 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2778 flp->fl4_src = (*rp)->rt_src;
2780 flp->fl4_dst = (*rp)->rt_dst;
2781 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2782 flags ? XFRM_LOOKUP_WAIT : 0);
2783 if (err == -EREMOTE)
2784 err = ipv4_dst_blackhole(net, rp, flp);
2791 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2793 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2795 return ip_route_output_flow(net, rp, flp, NULL, 0);
2797 EXPORT_SYMBOL(ip_route_output_key);
2799 static int rt_fill_info(struct net *net,
2800 struct sk_buff *skb, u32 pid, u32 seq, int event,
2801 int nowait, unsigned int flags)
2803 struct rtable *rt = skb_rtable(skb);
2805 struct nlmsghdr *nlh;
2807 u32 id = 0, ts = 0, tsage = 0, error;
2809 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2813 r = nlmsg_data(nlh);
2814 r->rtm_family = AF_INET;
2815 r->rtm_dst_len = 32;
2817 r->rtm_tos = rt->fl.fl4_tos;
2818 r->rtm_table = RT_TABLE_MAIN;
2819 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2820 r->rtm_type = rt->rt_type;
2821 r->rtm_scope = RT_SCOPE_UNIVERSE;
2822 r->rtm_protocol = RTPROT_UNSPEC;
2823 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2824 if (rt->rt_flags & RTCF_NOTIFY)
2825 r->rtm_flags |= RTM_F_NOTIFY;
2827 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2829 if (rt->fl.fl4_src) {
2830 r->rtm_src_len = 32;
2831 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2834 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2835 #ifdef CONFIG_NET_CLS_ROUTE
2836 if (rt->dst.tclassid)
2837 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2840 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2841 else if (rt->rt_src != rt->fl.fl4_src)
2842 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2844 if (rt->rt_dst != rt->rt_gateway)
2845 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2847 if (rtnetlink_put_metrics(skb, rt->dst.metrics) < 0)
2848 goto nla_put_failure;
2851 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2853 error = rt->dst.error;
2854 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
2856 inet_peer_refcheck(rt->peer);
2857 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2858 if (rt->peer->tcp_ts_stamp) {
2859 ts = rt->peer->tcp_ts;
2860 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2865 #ifdef CONFIG_IP_MROUTE
2866 __be32 dst = rt->rt_dst;
2868 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2869 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2870 int err = ipmr_get_route(net, skb, r, nowait);
2875 goto nla_put_failure;
2877 if (err == -EMSGSIZE)
2878 goto nla_put_failure;
2884 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2887 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2888 expires, error) < 0)
2889 goto nla_put_failure;
2891 return nlmsg_end(skb, nlh);
2894 nlmsg_cancel(skb, nlh);
2898 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2900 struct net *net = sock_net(in_skb->sk);
2902 struct nlattr *tb[RTA_MAX+1];
2903 struct rtable *rt = NULL;
2909 struct sk_buff *skb;
2911 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2915 rtm = nlmsg_data(nlh);
2917 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2923 /* Reserve room for dummy headers, this skb can pass
2924 through good chunk of routing engine.
2926 skb_reset_mac_header(skb);
2927 skb_reset_network_header(skb);
2929 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2930 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2931 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2933 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2934 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2935 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2936 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2939 struct net_device *dev;
2941 dev = __dev_get_by_index(net, iif);
2947 skb->protocol = htons(ETH_P_IP);
2951 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2954 rt = skb_rtable(skb);
2955 if (err == 0 && rt->dst.error)
2956 err = -rt->dst.error;
2963 .tos = rtm->rtm_tos,
2966 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2969 err = ip_route_output_key(net, &rt, &fl);
2975 skb_dst_set(skb, &rt->dst);
2976 if (rtm->rtm_flags & RTM_F_NOTIFY)
2977 rt->rt_flags |= RTCF_NOTIFY;
2979 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2980 RTM_NEWROUTE, 0, 0);
2984 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2993 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3000 net = sock_net(skb->sk);
3005 s_idx = idx = cb->args[1];
3006 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3007 if (!rt_hash_table[h].chain)
3010 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3011 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3012 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3014 if (rt_is_expired(rt))
3016 skb_dst_set_noref(skb, &rt->dst);
3017 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3018 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3019 1, NLM_F_MULTI) <= 0) {
3021 rcu_read_unlock_bh();
3026 rcu_read_unlock_bh();
3035 void ip_rt_multicast_event(struct in_device *in_dev)
3037 rt_cache_flush(dev_net(in_dev->dev), 0);
3040 #ifdef CONFIG_SYSCTL
3041 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3042 void __user *buffer,
3043 size_t *lenp, loff_t *ppos)
3050 memcpy(&ctl, __ctl, sizeof(ctl));
3051 ctl.data = &flush_delay;
3052 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3054 net = (struct net *)__ctl->extra1;
3055 rt_cache_flush(net, flush_delay);
3062 static ctl_table ipv4_route_table[] = {
3064 .procname = "gc_thresh",
3065 .data = &ipv4_dst_ops.gc_thresh,
3066 .maxlen = sizeof(int),
3068 .proc_handler = proc_dointvec,
3071 .procname = "max_size",
3072 .data = &ip_rt_max_size,
3073 .maxlen = sizeof(int),
3075 .proc_handler = proc_dointvec,
3078 /* Deprecated. Use gc_min_interval_ms */
3080 .procname = "gc_min_interval",
3081 .data = &ip_rt_gc_min_interval,
3082 .maxlen = sizeof(int),
3084 .proc_handler = proc_dointvec_jiffies,
3087 .procname = "gc_min_interval_ms",
3088 .data = &ip_rt_gc_min_interval,
3089 .maxlen = sizeof(int),
3091 .proc_handler = proc_dointvec_ms_jiffies,
3094 .procname = "gc_timeout",
3095 .data = &ip_rt_gc_timeout,
3096 .maxlen = sizeof(int),
3098 .proc_handler = proc_dointvec_jiffies,
3101 .procname = "gc_interval",
3102 .data = &ip_rt_gc_interval,
3103 .maxlen = sizeof(int),
3105 .proc_handler = proc_dointvec_jiffies,
3108 .procname = "redirect_load",
3109 .data = &ip_rt_redirect_load,
3110 .maxlen = sizeof(int),
3112 .proc_handler = proc_dointvec,
3115 .procname = "redirect_number",
3116 .data = &ip_rt_redirect_number,
3117 .maxlen = sizeof(int),
3119 .proc_handler = proc_dointvec,
3122 .procname = "redirect_silence",
3123 .data = &ip_rt_redirect_silence,
3124 .maxlen = sizeof(int),
3126 .proc_handler = proc_dointvec,
3129 .procname = "error_cost",
3130 .data = &ip_rt_error_cost,
3131 .maxlen = sizeof(int),
3133 .proc_handler = proc_dointvec,
3136 .procname = "error_burst",
3137 .data = &ip_rt_error_burst,
3138 .maxlen = sizeof(int),
3140 .proc_handler = proc_dointvec,
3143 .procname = "gc_elasticity",
3144 .data = &ip_rt_gc_elasticity,
3145 .maxlen = sizeof(int),
3147 .proc_handler = proc_dointvec,
3150 .procname = "mtu_expires",
3151 .data = &ip_rt_mtu_expires,
3152 .maxlen = sizeof(int),
3154 .proc_handler = proc_dointvec_jiffies,
3157 .procname = "min_pmtu",
3158 .data = &ip_rt_min_pmtu,
3159 .maxlen = sizeof(int),
3161 .proc_handler = proc_dointvec,
3164 .procname = "min_adv_mss",
3165 .data = &ip_rt_min_advmss,
3166 .maxlen = sizeof(int),
3168 .proc_handler = proc_dointvec,
3173 static struct ctl_table empty[1];
3175 static struct ctl_table ipv4_skeleton[] =
3177 { .procname = "route",
3178 .mode = 0555, .child = ipv4_route_table},
3179 { .procname = "neigh",
3180 .mode = 0555, .child = empty},
3184 static __net_initdata struct ctl_path ipv4_path[] = {
3185 { .procname = "net", },
3186 { .procname = "ipv4", },
3190 static struct ctl_table ipv4_route_flush_table[] = {
3192 .procname = "flush",
3193 .maxlen = sizeof(int),
3195 .proc_handler = ipv4_sysctl_rtcache_flush,
3200 static __net_initdata struct ctl_path ipv4_route_path[] = {
3201 { .procname = "net", },
3202 { .procname = "ipv4", },
3203 { .procname = "route", },
3207 static __net_init int sysctl_route_net_init(struct net *net)
3209 struct ctl_table *tbl;
3211 tbl = ipv4_route_flush_table;
3212 if (!net_eq(net, &init_net)) {
3213 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3217 tbl[0].extra1 = net;
3219 net->ipv4.route_hdr =
3220 register_net_sysctl_table(net, ipv4_route_path, tbl);
3221 if (net->ipv4.route_hdr == NULL)
3226 if (tbl != ipv4_route_flush_table)
3232 static __net_exit void sysctl_route_net_exit(struct net *net)
3234 struct ctl_table *tbl;
3236 tbl = net->ipv4.route_hdr->ctl_table_arg;
3237 unregister_net_sysctl_table(net->ipv4.route_hdr);
3238 BUG_ON(tbl == ipv4_route_flush_table);
3242 static __net_initdata struct pernet_operations sysctl_route_ops = {
3243 .init = sysctl_route_net_init,
3244 .exit = sysctl_route_net_exit,
3248 static __net_init int rt_genid_init(struct net *net)
3250 get_random_bytes(&net->ipv4.rt_genid,
3251 sizeof(net->ipv4.rt_genid));
3255 static __net_initdata struct pernet_operations rt_genid_ops = {
3256 .init = rt_genid_init,
3260 #ifdef CONFIG_NET_CLS_ROUTE
3261 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3262 #endif /* CONFIG_NET_CLS_ROUTE */
3264 static __initdata unsigned long rhash_entries;
3265 static int __init set_rhash_entries(char *str)
3269 rhash_entries = simple_strtoul(str, &str, 0);
3272 __setup("rhash_entries=", set_rhash_entries);
3274 int __init ip_rt_init(void)
3278 #ifdef CONFIG_NET_CLS_ROUTE
3279 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3281 panic("IP: failed to allocate ip_rt_acct\n");
3284 ipv4_dst_ops.kmem_cachep =
3285 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3286 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3288 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3290 if (dst_entries_init(&ipv4_dst_ops) < 0)
3291 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3293 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3294 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3296 rt_hash_table = (struct rt_hash_bucket *)
3297 alloc_large_system_hash("IP route cache",
3298 sizeof(struct rt_hash_bucket),
3300 (totalram_pages >= 128 * 1024) ?
3305 rhash_entries ? 0 : 512 * 1024);
3306 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3307 rt_hash_lock_init();
3309 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3310 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3315 /* All the timers, started at system startup tend
3316 to synchronize. Perturb it a bit.
3318 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3319 expires_ljiffies = jiffies;
3320 schedule_delayed_work(&expires_work,
3321 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3323 if (ip_rt_proc_init())
3324 printk(KERN_ERR "Unable to create route proc files\n");
3327 xfrm4_init(ip_rt_max_size);
3329 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3331 #ifdef CONFIG_SYSCTL
3332 register_pernet_subsys(&sysctl_route_ops);
3334 register_pernet_subsys(&rt_genid_ops);
3338 #ifdef CONFIG_SYSCTL
3340 * We really need to sanitize the damn ipv4 init order, then all
3341 * this nonsense will go away.
3343 void __init ip_static_sysctl_init(void)
3345 register_sysctl_paths(ipv4_path, ipv4_skeleton);