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;
135 * Interface to generic destination cache.
138 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
139 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
140 static unsigned int ipv4_default_mtu(const struct dst_entry *dst);
141 static void ipv4_dst_destroy(struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void ipv4_link_failure(struct sk_buff *skb);
144 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
145 static int rt_garbage_collect(struct dst_ops *ops);
147 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
152 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 struct rtable *rt = (struct rtable *) dst;
155 struct inet_peer *peer;
163 u32 *old_p = __DST_METRICS_PTR(old);
164 unsigned long prev, new;
167 if (inet_metrics_new(peer))
168 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
170 new = (unsigned long) p;
171 prev = cmpxchg(&dst->_metrics, old, new);
174 p = __DST_METRICS_PTR(prev);
175 if (prev & DST_METRICS_READ_ONLY)
179 fib_info_put(rt->fi);
187 static struct dst_ops ipv4_dst_ops = {
189 .protocol = cpu_to_be16(ETH_P_IP),
190 .gc = rt_garbage_collect,
191 .check = ipv4_dst_check,
192 .default_advmss = ipv4_default_advmss,
193 .default_mtu = ipv4_default_mtu,
194 .cow_metrics = ipv4_cow_metrics,
195 .destroy = ipv4_dst_destroy,
196 .ifdown = ipv4_dst_ifdown,
197 .negative_advice = ipv4_negative_advice,
198 .link_failure = ipv4_link_failure,
199 .update_pmtu = ip_rt_update_pmtu,
200 .local_out = __ip_local_out,
203 #define ECN_OR_COST(class) TC_PRIO_##class
205 const __u8 ip_tos2prio[16] = {
209 ECN_OR_COST(BESTEFFORT),
215 ECN_OR_COST(INTERACTIVE),
217 ECN_OR_COST(INTERACTIVE),
218 TC_PRIO_INTERACTIVE_BULK,
219 ECN_OR_COST(INTERACTIVE_BULK),
220 TC_PRIO_INTERACTIVE_BULK,
221 ECN_OR_COST(INTERACTIVE_BULK)
229 /* The locking scheme is rather straight forward:
231 * 1) Read-Copy Update protects the buckets of the central route hash.
232 * 2) Only writers remove entries, and they hold the lock
233 * as they look at rtable reference counts.
234 * 3) Only readers acquire references to rtable entries,
235 * they do so with atomic increments and with the
239 struct rt_hash_bucket {
240 struct rtable __rcu *chain;
243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
244 defined(CONFIG_PROVE_LOCKING)
246 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
247 * The size of this table is a power of two and depends on the number of CPUS.
248 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
250 #ifdef CONFIG_LOCKDEP
251 # define RT_HASH_LOCK_SZ 256
254 # define RT_HASH_LOCK_SZ 4096
256 # define RT_HASH_LOCK_SZ 2048
258 # define RT_HASH_LOCK_SZ 1024
260 # define RT_HASH_LOCK_SZ 512
262 # define RT_HASH_LOCK_SZ 256
266 static spinlock_t *rt_hash_locks;
267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
269 static __init void rt_hash_lock_init(void)
273 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
276 panic("IP: failed to allocate rt_hash_locks\n");
278 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
279 spin_lock_init(&rt_hash_locks[i]);
282 # define rt_hash_lock_addr(slot) NULL
284 static inline void rt_hash_lock_init(void)
289 static struct rt_hash_bucket *rt_hash_table __read_mostly;
290 static unsigned rt_hash_mask __read_mostly;
291 static unsigned int rt_hash_log __read_mostly;
293 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
296 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
299 return jhash_3words((__force u32)daddr, (__force u32)saddr,
304 static inline int rt_genid(struct net *net)
306 return atomic_read(&net->ipv4.rt_genid);
309 #ifdef CONFIG_PROC_FS
310 struct rt_cache_iter_state {
311 struct seq_net_private p;
316 static struct rtable *rt_cache_get_first(struct seq_file *seq)
318 struct rt_cache_iter_state *st = seq->private;
319 struct rtable *r = NULL;
321 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
322 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
325 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
327 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
328 r->rt_genid == st->genid)
330 r = rcu_dereference_bh(r->dst.rt_next);
332 rcu_read_unlock_bh();
337 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
340 struct rt_cache_iter_state *st = seq->private;
342 r = rcu_dereference_bh(r->dst.rt_next);
344 rcu_read_unlock_bh();
346 if (--st->bucket < 0)
348 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
350 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
355 static struct rtable *rt_cache_get_next(struct seq_file *seq,
358 struct rt_cache_iter_state *st = seq->private;
359 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
360 if (dev_net(r->dst.dev) != seq_file_net(seq))
362 if (r->rt_genid == st->genid)
368 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
370 struct rtable *r = rt_cache_get_first(seq);
373 while (pos && (r = rt_cache_get_next(seq, r)))
375 return pos ? NULL : r;
378 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
380 struct rt_cache_iter_state *st = seq->private;
382 return rt_cache_get_idx(seq, *pos - 1);
383 st->genid = rt_genid(seq_file_net(seq));
384 return SEQ_START_TOKEN;
387 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
391 if (v == SEQ_START_TOKEN)
392 r = rt_cache_get_first(seq);
394 r = rt_cache_get_next(seq, v);
399 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
401 if (v && v != SEQ_START_TOKEN)
402 rcu_read_unlock_bh();
405 static int rt_cache_seq_show(struct seq_file *seq, void *v)
407 if (v == SEQ_START_TOKEN)
408 seq_printf(seq, "%-127s\n",
409 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
410 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
413 struct rtable *r = v;
416 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
417 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
418 r->dst.dev ? r->dst.dev->name : "*",
419 (__force u32)r->rt_dst,
420 (__force u32)r->rt_gateway,
421 r->rt_flags, atomic_read(&r->dst.__refcnt),
422 r->dst.__use, 0, (__force u32)r->rt_src,
423 dst_metric_advmss(&r->dst) + 40,
424 dst_metric(&r->dst, RTAX_WINDOW),
425 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
426 dst_metric(&r->dst, RTAX_RTTVAR)),
428 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
429 r->dst.hh ? (r->dst.hh->hh_output ==
431 r->rt_spec_dst, &len);
433 seq_printf(seq, "%*s\n", 127 - len, "");
438 static const struct seq_operations rt_cache_seq_ops = {
439 .start = rt_cache_seq_start,
440 .next = rt_cache_seq_next,
441 .stop = rt_cache_seq_stop,
442 .show = rt_cache_seq_show,
445 static int rt_cache_seq_open(struct inode *inode, struct file *file)
447 return seq_open_net(inode, file, &rt_cache_seq_ops,
448 sizeof(struct rt_cache_iter_state));
451 static const struct file_operations rt_cache_seq_fops = {
452 .owner = THIS_MODULE,
453 .open = rt_cache_seq_open,
456 .release = seq_release_net,
460 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
465 return SEQ_START_TOKEN;
467 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
468 if (!cpu_possible(cpu))
471 return &per_cpu(rt_cache_stat, cpu);
476 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
480 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
481 if (!cpu_possible(cpu))
484 return &per_cpu(rt_cache_stat, cpu);
490 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
495 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
497 struct rt_cache_stat *st = v;
499 if (v == SEQ_START_TOKEN) {
500 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");
504 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
505 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
506 dst_entries_get_slow(&ipv4_dst_ops),
529 static const struct seq_operations rt_cpu_seq_ops = {
530 .start = rt_cpu_seq_start,
531 .next = rt_cpu_seq_next,
532 .stop = rt_cpu_seq_stop,
533 .show = rt_cpu_seq_show,
537 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
539 return seq_open(file, &rt_cpu_seq_ops);
542 static const struct file_operations rt_cpu_seq_fops = {
543 .owner = THIS_MODULE,
544 .open = rt_cpu_seq_open,
547 .release = seq_release,
550 #ifdef CONFIG_IP_ROUTE_CLASSID
551 static int rt_acct_proc_show(struct seq_file *m, void *v)
553 struct ip_rt_acct *dst, *src;
556 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
560 for_each_possible_cpu(i) {
561 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
562 for (j = 0; j < 256; j++) {
563 dst[j].o_bytes += src[j].o_bytes;
564 dst[j].o_packets += src[j].o_packets;
565 dst[j].i_bytes += src[j].i_bytes;
566 dst[j].i_packets += src[j].i_packets;
570 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
575 static int rt_acct_proc_open(struct inode *inode, struct file *file)
577 return single_open(file, rt_acct_proc_show, NULL);
580 static const struct file_operations rt_acct_proc_fops = {
581 .owner = THIS_MODULE,
582 .open = rt_acct_proc_open,
585 .release = single_release,
589 static int __net_init ip_rt_do_proc_init(struct net *net)
591 struct proc_dir_entry *pde;
593 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
598 pde = proc_create("rt_cache", S_IRUGO,
599 net->proc_net_stat, &rt_cpu_seq_fops);
603 #ifdef CONFIG_IP_ROUTE_CLASSID
604 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
610 #ifdef CONFIG_IP_ROUTE_CLASSID
612 remove_proc_entry("rt_cache", net->proc_net_stat);
615 remove_proc_entry("rt_cache", net->proc_net);
620 static void __net_exit ip_rt_do_proc_exit(struct net *net)
622 remove_proc_entry("rt_cache", net->proc_net_stat);
623 remove_proc_entry("rt_cache", net->proc_net);
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625 remove_proc_entry("rt_acct", net->proc_net);
629 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
630 .init = ip_rt_do_proc_init,
631 .exit = ip_rt_do_proc_exit,
634 static int __init ip_rt_proc_init(void)
636 return register_pernet_subsys(&ip_rt_proc_ops);
640 static inline int ip_rt_proc_init(void)
644 #endif /* CONFIG_PROC_FS */
646 static inline void rt_free(struct rtable *rt)
648 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
651 static inline void rt_drop(struct rtable *rt)
654 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
657 static inline int rt_fast_clean(struct rtable *rth)
659 /* Kill broadcast/multicast entries very aggresively, if they
660 collide in hash table with more useful entries */
661 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
662 rt_is_input_route(rth) && rth->dst.rt_next;
665 static inline int rt_valuable(struct rtable *rth)
667 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
668 (rth->peer && rth->peer->pmtu_expires);
671 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
676 if (atomic_read(&rth->dst.__refcnt))
679 age = jiffies - rth->dst.lastuse;
680 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
681 (age <= tmo2 && rt_valuable(rth)))
687 /* Bits of score are:
689 * 30: not quite useless
690 * 29..0: usage counter
692 static inline u32 rt_score(struct rtable *rt)
694 u32 score = jiffies - rt->dst.lastuse;
696 score = ~score & ~(3<<30);
701 if (rt_is_output_route(rt) ||
702 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
708 static inline bool rt_caching(const struct net *net)
710 return net->ipv4.current_rt_cache_rebuild_count <=
711 net->ipv4.sysctl_rt_cache_rebuild_count;
714 static inline bool compare_hash_inputs(const struct flowi *fl1,
715 const struct flowi *fl2)
717 return ((((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
718 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
719 (fl1->iif ^ fl2->iif)) == 0);
722 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
724 return (((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
725 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
726 (fl1->mark ^ fl2->mark) |
727 (*(u16 *)&fl1->fl4_tos ^ *(u16 *)&fl2->fl4_tos) |
728 (fl1->oif ^ fl2->oif) |
729 (fl1->iif ^ fl2->iif)) == 0;
732 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
734 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
737 static inline int rt_is_expired(struct rtable *rth)
739 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
743 * Perform a full scan of hash table and free all entries.
744 * Can be called by a softirq or a process.
745 * In the later case, we want to be reschedule if necessary
747 static void rt_do_flush(struct net *net, int process_context)
750 struct rtable *rth, *next;
752 for (i = 0; i <= rt_hash_mask; i++) {
753 struct rtable __rcu **pprev;
756 if (process_context && need_resched())
758 rth = rcu_dereference_raw(rt_hash_table[i].chain);
762 spin_lock_bh(rt_hash_lock_addr(i));
765 pprev = &rt_hash_table[i].chain;
766 rth = rcu_dereference_protected(*pprev,
767 lockdep_is_held(rt_hash_lock_addr(i)));
770 next = rcu_dereference_protected(rth->dst.rt_next,
771 lockdep_is_held(rt_hash_lock_addr(i)));
774 net_eq(dev_net(rth->dst.dev), net)) {
775 rcu_assign_pointer(*pprev, next);
776 rcu_assign_pointer(rth->dst.rt_next, list);
779 pprev = &rth->dst.rt_next;
784 spin_unlock_bh(rt_hash_lock_addr(i));
786 for (; list; list = next) {
787 next = rcu_dereference_protected(list->dst.rt_next, 1);
794 * While freeing expired entries, we compute average chain length
795 * and standard deviation, using fixed-point arithmetic.
796 * This to have an estimation of rt_chain_length_max
797 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
798 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
802 #define ONE (1UL << FRACT_BITS)
805 * Given a hash chain and an item in this hash chain,
806 * find if a previous entry has the same hash_inputs
807 * (but differs on tos, mark or oif)
808 * Returns 0 if an alias is found.
809 * Returns ONE if rth has no alias before itself.
811 static int has_noalias(const struct rtable *head, const struct rtable *rth)
813 const struct rtable *aux = head;
816 if (compare_hash_inputs(&aux->fl, &rth->fl))
818 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
824 * Pertubation of rt_genid by a small quantity [1..256]
825 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
826 * many times (2^24) without giving recent rt_genid.
827 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
829 static void rt_cache_invalidate(struct net *net)
831 unsigned char shuffle;
833 get_random_bytes(&shuffle, sizeof(shuffle));
834 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
838 * delay < 0 : invalidate cache (fast : entries will be deleted later)
839 * delay >= 0 : invalidate & flush cache (can be long)
841 void rt_cache_flush(struct net *net, int delay)
843 rt_cache_invalidate(net);
845 rt_do_flush(net, !in_softirq());
848 /* Flush previous cache invalidated entries from the cache */
849 void rt_cache_flush_batch(struct net *net)
851 rt_do_flush(net, !in_softirq());
854 static void rt_emergency_hash_rebuild(struct net *net)
857 printk(KERN_WARNING "Route hash chain too long!\n");
858 rt_cache_invalidate(net);
862 Short description of GC goals.
864 We want to build algorithm, which will keep routing cache
865 at some equilibrium point, when number of aged off entries
866 is kept approximately equal to newly generated ones.
868 Current expiration strength is variable "expire".
869 We try to adjust it dynamically, so that if networking
870 is idle expires is large enough to keep enough of warm entries,
871 and when load increases it reduces to limit cache size.
874 static int rt_garbage_collect(struct dst_ops *ops)
876 static unsigned long expire = RT_GC_TIMEOUT;
877 static unsigned long last_gc;
879 static int equilibrium;
881 struct rtable __rcu **rthp;
882 unsigned long now = jiffies;
884 int entries = dst_entries_get_fast(&ipv4_dst_ops);
887 * Garbage collection is pretty expensive,
888 * do not make it too frequently.
891 RT_CACHE_STAT_INC(gc_total);
893 if (now - last_gc < ip_rt_gc_min_interval &&
894 entries < ip_rt_max_size) {
895 RT_CACHE_STAT_INC(gc_ignored);
899 entries = dst_entries_get_slow(&ipv4_dst_ops);
900 /* Calculate number of entries, which we want to expire now. */
901 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
903 if (equilibrium < ipv4_dst_ops.gc_thresh)
904 equilibrium = ipv4_dst_ops.gc_thresh;
905 goal = entries - equilibrium;
907 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
908 goal = entries - equilibrium;
911 /* We are in dangerous area. Try to reduce cache really
914 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
915 equilibrium = entries - goal;
918 if (now - last_gc >= ip_rt_gc_min_interval)
929 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
930 unsigned long tmo = expire;
932 k = (k + 1) & rt_hash_mask;
933 rthp = &rt_hash_table[k].chain;
934 spin_lock_bh(rt_hash_lock_addr(k));
935 while ((rth = rcu_dereference_protected(*rthp,
936 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
937 if (!rt_is_expired(rth) &&
938 !rt_may_expire(rth, tmo, expire)) {
940 rthp = &rth->dst.rt_next;
943 *rthp = rth->dst.rt_next;
947 spin_unlock_bh(rt_hash_lock_addr(k));
956 /* Goal is not achieved. We stop process if:
958 - if expire reduced to zero. Otherwise, expire is halfed.
959 - if table is not full.
960 - if we are called from interrupt.
961 - jiffies check is just fallback/debug loop breaker.
962 We will not spin here for long time in any case.
965 RT_CACHE_STAT_INC(gc_goal_miss);
971 #if RT_CACHE_DEBUG >= 2
972 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
973 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
976 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
978 } while (!in_softirq() && time_before_eq(jiffies, now));
980 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
982 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
985 printk(KERN_WARNING "dst cache overflow\n");
986 RT_CACHE_STAT_INC(gc_dst_overflow);
990 expire += ip_rt_gc_min_interval;
991 if (expire > ip_rt_gc_timeout ||
992 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
993 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
994 expire = ip_rt_gc_timeout;
995 #if RT_CACHE_DEBUG >= 2
996 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
997 dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
1003 * Returns number of entries in a hash chain that have different hash_inputs
1005 static int slow_chain_length(const struct rtable *head)
1008 const struct rtable *rth = head;
1011 length += has_noalias(head, rth);
1012 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1014 return length >> FRACT_BITS;
1017 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1018 struct rtable **rp, struct sk_buff *skb, int ifindex)
1020 struct rtable *rth, *cand;
1021 struct rtable __rcu **rthp, **candp;
1025 int attempts = !in_softirq();
1029 min_score = ~(u32)0;
1034 if (!rt_caching(dev_net(rt->dst.dev))) {
1036 * If we're not caching, just tell the caller we
1037 * were successful and don't touch the route. The
1038 * caller hold the sole reference to the cache entry, and
1039 * it will be released when the caller is done with it.
1040 * If we drop it here, the callers have no way to resolve routes
1041 * when we're not caching. Instead, just point *rp at rt, so
1042 * the caller gets a single use out of the route
1043 * Note that we do rt_free on this new route entry, so that
1044 * once its refcount hits zero, we are still able to reap it
1046 * Note: To avoid expensive rcu stuff for this uncached dst,
1047 * we set DST_NOCACHE so that dst_release() can free dst without
1048 * waiting a grace period.
1051 rt->dst.flags |= DST_NOCACHE;
1052 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1053 int err = arp_bind_neighbour(&rt->dst);
1055 if (net_ratelimit())
1057 "Neighbour table failure & not caching routes.\n");
1066 rthp = &rt_hash_table[hash].chain;
1068 spin_lock_bh(rt_hash_lock_addr(hash));
1069 while ((rth = rcu_dereference_protected(*rthp,
1070 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1071 if (rt_is_expired(rth)) {
1072 *rthp = rth->dst.rt_next;
1076 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1078 *rthp = rth->dst.rt_next;
1080 * Since lookup is lockfree, the deletion
1081 * must be visible to another weakly ordered CPU before
1082 * the insertion at the start of the hash chain.
1084 rcu_assign_pointer(rth->dst.rt_next,
1085 rt_hash_table[hash].chain);
1087 * Since lookup is lockfree, the update writes
1088 * must be ordered for consistency on SMP.
1090 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1092 dst_use(&rth->dst, now);
1093 spin_unlock_bh(rt_hash_lock_addr(hash));
1099 skb_dst_set(skb, &rth->dst);
1103 if (!atomic_read(&rth->dst.__refcnt)) {
1104 u32 score = rt_score(rth);
1106 if (score <= min_score) {
1115 rthp = &rth->dst.rt_next;
1119 /* ip_rt_gc_elasticity used to be average length of chain
1120 * length, when exceeded gc becomes really aggressive.
1122 * The second limit is less certain. At the moment it allows
1123 * only 2 entries per bucket. We will see.
1125 if (chain_length > ip_rt_gc_elasticity) {
1126 *candp = cand->dst.rt_next;
1130 if (chain_length > rt_chain_length_max &&
1131 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1132 struct net *net = dev_net(rt->dst.dev);
1133 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1134 if (!rt_caching(net)) {
1135 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1136 rt->dst.dev->name, num);
1138 rt_emergency_hash_rebuild(net);
1139 spin_unlock_bh(rt_hash_lock_addr(hash));
1141 hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1142 ifindex, rt_genid(net));
1147 /* Try to bind route to arp only if it is output
1148 route or unicast forwarding path.
1150 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1151 int err = arp_bind_neighbour(&rt->dst);
1153 spin_unlock_bh(rt_hash_lock_addr(hash));
1155 if (err != -ENOBUFS) {
1160 /* Neighbour tables are full and nothing
1161 can be released. Try to shrink route cache,
1162 it is most likely it holds some neighbour records.
1164 if (attempts-- > 0) {
1165 int saved_elasticity = ip_rt_gc_elasticity;
1166 int saved_int = ip_rt_gc_min_interval;
1167 ip_rt_gc_elasticity = 1;
1168 ip_rt_gc_min_interval = 0;
1169 rt_garbage_collect(&ipv4_dst_ops);
1170 ip_rt_gc_min_interval = saved_int;
1171 ip_rt_gc_elasticity = saved_elasticity;
1175 if (net_ratelimit())
1176 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1182 rt->dst.rt_next = rt_hash_table[hash].chain;
1184 #if RT_CACHE_DEBUG >= 2
1185 if (rt->dst.rt_next) {
1187 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1189 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1190 printk(" . %pI4", &trt->rt_dst);
1195 * Since lookup is lockfree, we must make sure
1196 * previous writes to rt are comitted to memory
1197 * before making rt visible to other CPUS.
1199 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1201 spin_unlock_bh(rt_hash_lock_addr(hash));
1207 skb_dst_set(skb, &rt->dst);
1211 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1213 static u32 rt_peer_genid(void)
1215 return atomic_read(&__rt_peer_genid);
1218 void rt_bind_peer(struct rtable *rt, int create)
1220 struct inet_peer *peer;
1222 peer = inet_getpeer_v4(rt->rt_dst, create);
1224 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1227 rt->rt_peer_genid = rt_peer_genid();
1231 * Peer allocation may fail only in serious out-of-memory conditions. However
1232 * we still can generate some output.
1233 * Random ID selection looks a bit dangerous because we have no chances to
1234 * select ID being unique in a reasonable period of time.
1235 * But broken packet identifier may be better than no packet at all.
1237 static void ip_select_fb_ident(struct iphdr *iph)
1239 static DEFINE_SPINLOCK(ip_fb_id_lock);
1240 static u32 ip_fallback_id;
1243 spin_lock_bh(&ip_fb_id_lock);
1244 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1245 iph->id = htons(salt & 0xFFFF);
1246 ip_fallback_id = salt;
1247 spin_unlock_bh(&ip_fb_id_lock);
1250 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1252 struct rtable *rt = (struct rtable *) dst;
1255 if (rt->peer == NULL)
1256 rt_bind_peer(rt, 1);
1258 /* If peer is attached to destination, it is never detached,
1259 so that we need not to grab a lock to dereference it.
1262 iph->id = htons(inet_getid(rt->peer, more));
1266 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1267 __builtin_return_address(0));
1269 ip_select_fb_ident(iph);
1271 EXPORT_SYMBOL(__ip_select_ident);
1273 static void rt_del(unsigned hash, struct rtable *rt)
1275 struct rtable __rcu **rthp;
1278 rthp = &rt_hash_table[hash].chain;
1279 spin_lock_bh(rt_hash_lock_addr(hash));
1281 while ((aux = rcu_dereference_protected(*rthp,
1282 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1283 if (aux == rt || rt_is_expired(aux)) {
1284 *rthp = aux->dst.rt_next;
1288 rthp = &aux->dst.rt_next;
1290 spin_unlock_bh(rt_hash_lock_addr(hash));
1293 /* called in rcu_read_lock() section */
1294 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1295 __be32 saddr, struct net_device *dev)
1298 struct in_device *in_dev = __in_dev_get_rcu(dev);
1300 struct rtable __rcu **rthp;
1301 __be32 skeys[2] = { saddr, 0 };
1302 int ikeys[2] = { dev->ifindex, 0 };
1303 struct netevent_redirect netevent;
1310 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1311 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1312 ipv4_is_zeronet(new_gw))
1313 goto reject_redirect;
1315 if (!rt_caching(net))
1316 goto reject_redirect;
1318 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1319 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1320 goto reject_redirect;
1321 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1322 goto reject_redirect;
1324 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1325 goto reject_redirect;
1328 for (i = 0; i < 2; i++) {
1329 for (k = 0; k < 2; k++) {
1330 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1333 rthp = &rt_hash_table[hash].chain;
1335 while ((rth = rcu_dereference(*rthp)) != NULL) {
1338 if (rth->fl.fl4_dst != daddr ||
1339 rth->fl.fl4_src != skeys[i] ||
1340 rth->fl.oif != ikeys[k] ||
1341 rt_is_input_route(rth) ||
1342 rt_is_expired(rth) ||
1343 !net_eq(dev_net(rth->dst.dev), net)) {
1344 rthp = &rth->dst.rt_next;
1348 if (rth->rt_dst != daddr ||
1349 rth->rt_src != saddr ||
1351 rth->rt_gateway != old_gw ||
1352 rth->dst.dev != dev)
1355 dst_hold(&rth->dst);
1357 rt = dst_alloc(&ipv4_dst_ops);
1363 /* Copy all the information. */
1366 atomic_set(&rt->dst.__refcnt, 1);
1367 rt->dst.child = NULL;
1369 dev_hold(rt->dst.dev);
1370 rt->dst.obsolete = -1;
1371 rt->dst.lastuse = jiffies;
1372 rt->dst.path = &rt->dst;
1373 rt->dst.neighbour = NULL;
1376 rt->dst.xfrm = NULL;
1378 rt->rt_genid = rt_genid(net);
1379 rt->rt_flags |= RTCF_REDIRECTED;
1381 /* Gateway is different ... */
1382 rt->rt_gateway = new_gw;
1384 /* Redirect received -> path was valid */
1385 dst_confirm(&rth->dst);
1388 atomic_inc(&rt->peer->refcnt);
1390 atomic_inc(&rt->fi->fib_clntref);
1392 if (arp_bind_neighbour(&rt->dst) ||
1393 !(rt->dst.neighbour->nud_state &
1395 if (rt->dst.neighbour)
1396 neigh_event_send(rt->dst.neighbour, NULL);
1402 netevent.old = &rth->dst;
1403 netevent.new = &rt->dst;
1404 call_netevent_notifiers(NETEVENT_REDIRECT,
1408 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1419 #ifdef CONFIG_IP_ROUTE_VERBOSE
1420 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1421 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1422 " Advised path = %pI4 -> %pI4\n",
1423 &old_gw, dev->name, &new_gw,
1429 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1431 struct rtable *rt = (struct rtable *)dst;
1432 struct dst_entry *ret = dst;
1435 if (dst->obsolete > 0) {
1438 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1439 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1441 rt_genid(dev_net(dst->dev)));
1442 #if RT_CACHE_DEBUG >= 1
1443 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1444 &rt->rt_dst, rt->fl.fl4_tos);
1448 } else if (rt->peer &&
1449 rt->peer->pmtu_expires &&
1450 time_after_eq(jiffies, rt->peer->pmtu_expires)) {
1451 unsigned long orig = rt->peer->pmtu_expires;
1453 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1454 dst_metric_set(dst, RTAX_MTU,
1455 rt->peer->pmtu_orig);
1463 * 1. The first ip_rt_redirect_number redirects are sent
1464 * with exponential backoff, then we stop sending them at all,
1465 * assuming that the host ignores our redirects.
1466 * 2. If we did not see packets requiring redirects
1467 * during ip_rt_redirect_silence, we assume that the host
1468 * forgot redirected route and start to send redirects again.
1470 * This algorithm is much cheaper and more intelligent than dumb load limiting
1473 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1474 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1477 void ip_rt_send_redirect(struct sk_buff *skb)
1479 struct rtable *rt = skb_rtable(skb);
1480 struct in_device *in_dev;
1481 struct inet_peer *peer;
1485 in_dev = __in_dev_get_rcu(rt->dst.dev);
1486 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1490 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1494 rt_bind_peer(rt, 1);
1497 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1501 /* No redirected packets during ip_rt_redirect_silence;
1502 * reset the algorithm.
1504 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1505 peer->rate_tokens = 0;
1507 /* Too many ignored redirects; do not send anything
1508 * set dst.rate_last to the last seen redirected packet.
1510 if (peer->rate_tokens >= ip_rt_redirect_number) {
1511 peer->rate_last = jiffies;
1515 /* Check for load limit; set rate_last to the latest sent
1518 if (peer->rate_tokens == 0 ||
1521 (ip_rt_redirect_load << peer->rate_tokens)))) {
1522 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1523 peer->rate_last = jiffies;
1524 ++peer->rate_tokens;
1525 #ifdef CONFIG_IP_ROUTE_VERBOSE
1527 peer->rate_tokens == ip_rt_redirect_number &&
1529 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1530 &rt->rt_src, rt->rt_iif,
1531 &rt->rt_dst, &rt->rt_gateway);
1536 static int ip_error(struct sk_buff *skb)
1538 struct rtable *rt = skb_rtable(skb);
1539 struct inet_peer *peer;
1544 switch (rt->dst.error) {
1549 code = ICMP_HOST_UNREACH;
1552 code = ICMP_NET_UNREACH;
1553 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1554 IPSTATS_MIB_INNOROUTES);
1557 code = ICMP_PKT_FILTERED;
1562 rt_bind_peer(rt, 1);
1568 peer->rate_tokens += now - peer->rate_last;
1569 if (peer->rate_tokens > ip_rt_error_burst)
1570 peer->rate_tokens = ip_rt_error_burst;
1571 peer->rate_last = now;
1572 if (peer->rate_tokens >= ip_rt_error_cost)
1573 peer->rate_tokens -= ip_rt_error_cost;
1578 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1580 out: kfree_skb(skb);
1585 * The last two values are not from the RFC but
1586 * are needed for AMPRnet AX.25 paths.
1589 static const unsigned short mtu_plateau[] =
1590 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1592 static inline unsigned short guess_mtu(unsigned short old_mtu)
1596 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1597 if (old_mtu > mtu_plateau[i])
1598 return mtu_plateau[i];
1602 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1603 unsigned short new_mtu,
1604 struct net_device *dev)
1606 unsigned short old_mtu = ntohs(iph->tot_len);
1607 unsigned short est_mtu = 0;
1608 struct inet_peer *peer;
1610 peer = inet_getpeer_v4(iph->daddr, 1);
1612 unsigned short mtu = new_mtu;
1614 if (new_mtu < 68 || new_mtu >= old_mtu) {
1615 /* BSD 4.2 derived systems incorrectly adjust
1616 * tot_len by the IP header length, and report
1617 * a zero MTU in the ICMP message.
1620 old_mtu >= 68 + (iph->ihl << 2))
1621 old_mtu -= iph->ihl << 2;
1622 mtu = guess_mtu(old_mtu);
1625 if (mtu < ip_rt_min_pmtu)
1626 mtu = ip_rt_min_pmtu;
1627 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1629 peer->pmtu_learned = mtu;
1630 peer->pmtu_expires = jiffies + ip_rt_mtu_expires;
1635 atomic_inc(&__rt_peer_genid);
1637 return est_mtu ? : new_mtu;
1640 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1642 unsigned long expires = peer->pmtu_expires;
1644 if (time_before(expires, jiffies)) {
1645 u32 orig_dst_mtu = dst_mtu(dst);
1646 if (peer->pmtu_learned < orig_dst_mtu) {
1647 if (!peer->pmtu_orig)
1648 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1649 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1651 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1652 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1655 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1657 struct rtable *rt = (struct rtable *) dst;
1658 struct inet_peer *peer;
1663 rt_bind_peer(rt, 1);
1666 if (mtu < ip_rt_min_pmtu)
1667 mtu = ip_rt_min_pmtu;
1668 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1669 peer->pmtu_learned = mtu;
1670 peer->pmtu_expires = jiffies + ip_rt_mtu_expires;
1672 atomic_inc(&__rt_peer_genid);
1673 rt->rt_peer_genid = rt_peer_genid();
1675 check_peer_pmtu(dst, peer);
1681 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1683 struct rtable *rt = (struct rtable *) dst;
1685 if (rt_is_expired(rt))
1687 if (rt->rt_peer_genid != rt_peer_genid()) {
1688 struct inet_peer *peer;
1691 rt_bind_peer(rt, 0);
1694 if (peer && peer->pmtu_expires)
1695 check_peer_pmtu(dst, peer);
1697 rt->rt_peer_genid = rt_peer_genid();
1702 static void ipv4_dst_destroy(struct dst_entry *dst)
1704 struct rtable *rt = (struct rtable *) dst;
1705 struct inet_peer *peer = rt->peer;
1708 fib_info_put(rt->fi);
1718 static void ipv4_link_failure(struct sk_buff *skb)
1722 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1724 rt = skb_rtable(skb);
1727 rt->peer->pmtu_expires) {
1728 unsigned long orig = rt->peer->pmtu_expires;
1730 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1731 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1735 static int ip_rt_bug(struct sk_buff *skb)
1737 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1738 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1739 skb->dev ? skb->dev->name : "?");
1745 We do not cache source address of outgoing interface,
1746 because it is used only by IP RR, TS and SRR options,
1747 so that it out of fast path.
1749 BTW remember: "addr" is allowed to be not aligned
1753 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1756 struct fib_result res;
1758 if (rt_is_output_route(rt))
1762 if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0)
1763 src = FIB_RES_PREFSRC(res);
1765 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1769 memcpy(addr, &src, 4);
1772 #ifdef CONFIG_IP_ROUTE_CLASSID
1773 static void set_class_tag(struct rtable *rt, u32 tag)
1775 if (!(rt->dst.tclassid & 0xFFFF))
1776 rt->dst.tclassid |= tag & 0xFFFF;
1777 if (!(rt->dst.tclassid & 0xFFFF0000))
1778 rt->dst.tclassid |= tag & 0xFFFF0000;
1782 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1784 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1787 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1789 if (advmss > 65535 - 40)
1790 advmss = 65535 - 40;
1795 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1797 unsigned int mtu = dst->dev->mtu;
1799 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1800 const struct rtable *rt = (const struct rtable *) dst;
1802 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1806 if (mtu > IP_MAX_MTU)
1812 static void rt_init_metrics(struct rtable *rt, struct fib_info *fi)
1814 struct inet_peer *peer;
1817 /* If a peer entry exists for this destination, we must hook
1818 * it up in order to get at cached metrics.
1820 if (rt->fl.flags & FLOWI_FLAG_PRECOW_METRICS)
1823 rt_bind_peer(rt, create);
1826 if (inet_metrics_new(peer))
1827 memcpy(peer->metrics, fi->fib_metrics,
1828 sizeof(u32) * RTAX_MAX);
1829 dst_init_metrics(&rt->dst, peer->metrics, false);
1831 if (peer->pmtu_expires)
1832 check_peer_pmtu(&rt->dst, peer);
1834 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1836 atomic_inc(&fi->fib_clntref);
1838 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1842 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1844 struct dst_entry *dst = &rt->dst;
1845 struct fib_info *fi = res->fi;
1848 if (FIB_RES_GW(*res) &&
1849 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1850 rt->rt_gateway = FIB_RES_GW(*res);
1851 rt_init_metrics(rt, fi);
1852 #ifdef CONFIG_IP_ROUTE_CLASSID
1853 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1857 if (dst_mtu(dst) > IP_MAX_MTU)
1858 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1859 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1860 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1862 #ifdef CONFIG_IP_ROUTE_CLASSID
1863 #ifdef CONFIG_IP_MULTIPLE_TABLES
1864 set_class_tag(rt, fib_rules_tclass(res));
1866 set_class_tag(rt, itag);
1868 rt->rt_type = res->type;
1871 /* called in rcu_read_lock() section */
1872 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1873 u8 tos, struct net_device *dev, int our)
1878 struct in_device *in_dev = __in_dev_get_rcu(dev);
1882 /* Primary sanity checks. */
1887 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1888 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1891 if (ipv4_is_zeronet(saddr)) {
1892 if (!ipv4_is_local_multicast(daddr))
1894 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1896 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1901 rth = dst_alloc(&ipv4_dst_ops);
1905 rth->dst.output = ip_rt_bug;
1906 rth->dst.obsolete = -1;
1908 atomic_set(&rth->dst.__refcnt, 1);
1909 rth->dst.flags= DST_HOST;
1910 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1911 rth->dst.flags |= DST_NOPOLICY;
1912 rth->fl.fl4_dst = daddr;
1913 rth->rt_dst = daddr;
1914 rth->fl.fl4_tos = tos;
1915 rth->fl.mark = skb->mark;
1916 rth->fl.fl4_src = saddr;
1917 rth->rt_src = saddr;
1918 #ifdef CONFIG_IP_ROUTE_CLASSID
1919 rth->dst.tclassid = itag;
1922 rth->fl.iif = dev->ifindex;
1923 rth->dst.dev = init_net.loopback_dev;
1924 dev_hold(rth->dst.dev);
1926 rth->rt_gateway = daddr;
1927 rth->rt_spec_dst= spec_dst;
1928 rth->rt_genid = rt_genid(dev_net(dev));
1929 rth->rt_flags = RTCF_MULTICAST;
1930 rth->rt_type = RTN_MULTICAST;
1932 rth->dst.input= ip_local_deliver;
1933 rth->rt_flags |= RTCF_LOCAL;
1936 #ifdef CONFIG_IP_MROUTE
1937 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1938 rth->dst.input = ip_mr_input;
1940 RT_CACHE_STAT_INC(in_slow_mc);
1942 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1943 return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1954 static void ip_handle_martian_source(struct net_device *dev,
1955 struct in_device *in_dev,
1956 struct sk_buff *skb,
1960 RT_CACHE_STAT_INC(in_martian_src);
1961 #ifdef CONFIG_IP_ROUTE_VERBOSE
1962 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1964 * RFC1812 recommendation, if source is martian,
1965 * the only hint is MAC header.
1967 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1968 &daddr, &saddr, dev->name);
1969 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1971 const unsigned char *p = skb_mac_header(skb);
1972 printk(KERN_WARNING "ll header: ");
1973 for (i = 0; i < dev->hard_header_len; i++, p++) {
1975 if (i < (dev->hard_header_len - 1))
1984 /* called in rcu_read_lock() section */
1985 static int __mkroute_input(struct sk_buff *skb,
1986 struct fib_result *res,
1987 struct in_device *in_dev,
1988 __be32 daddr, __be32 saddr, u32 tos,
1989 struct rtable **result)
1993 struct in_device *out_dev;
1994 unsigned int flags = 0;
1998 /* get a working reference to the output device */
1999 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2000 if (out_dev == NULL) {
2001 if (net_ratelimit())
2002 printk(KERN_CRIT "Bug in ip_route_input" \
2003 "_slow(). Please, report\n");
2008 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
2009 in_dev->dev, &spec_dst, &itag, skb->mark);
2011 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2018 flags |= RTCF_DIRECTSRC;
2020 if (out_dev == in_dev && err &&
2021 (IN_DEV_SHARED_MEDIA(out_dev) ||
2022 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2023 flags |= RTCF_DOREDIRECT;
2025 if (skb->protocol != htons(ETH_P_IP)) {
2026 /* Not IP (i.e. ARP). Do not create route, if it is
2027 * invalid for proxy arp. DNAT routes are always valid.
2029 * Proxy arp feature have been extended to allow, ARP
2030 * replies back to the same interface, to support
2031 * Private VLAN switch technologies. See arp.c.
2033 if (out_dev == in_dev &&
2034 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2041 rth = dst_alloc(&ipv4_dst_ops);
2047 atomic_set(&rth->dst.__refcnt, 1);
2048 rth->dst.flags= DST_HOST;
2049 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2050 rth->dst.flags |= DST_NOPOLICY;
2051 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2052 rth->dst.flags |= DST_NOXFRM;
2053 rth->fl.fl4_dst = daddr;
2054 rth->rt_dst = daddr;
2055 rth->fl.fl4_tos = tos;
2056 rth->fl.mark = skb->mark;
2057 rth->fl.fl4_src = saddr;
2058 rth->rt_src = saddr;
2059 rth->rt_gateway = daddr;
2061 rth->fl.iif = in_dev->dev->ifindex;
2062 rth->dst.dev = (out_dev)->dev;
2063 dev_hold(rth->dst.dev);
2065 rth->rt_spec_dst= spec_dst;
2067 rth->dst.obsolete = -1;
2068 rth->dst.input = ip_forward;
2069 rth->dst.output = ip_output;
2070 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2072 rt_set_nexthop(rth, res, itag);
2074 rth->rt_flags = flags;
2082 static int ip_mkroute_input(struct sk_buff *skb,
2083 struct fib_result *res,
2084 const struct flowi *fl,
2085 struct in_device *in_dev,
2086 __be32 daddr, __be32 saddr, u32 tos)
2088 struct rtable* rth = NULL;
2092 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2093 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2094 fib_select_multipath(fl, res);
2097 /* create a routing cache entry */
2098 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2102 /* put it into the cache */
2103 hash = rt_hash(daddr, saddr, fl->iif,
2104 rt_genid(dev_net(rth->dst.dev)));
2105 return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2109 * NOTE. We drop all the packets that has local source
2110 * addresses, because every properly looped back packet
2111 * must have correct destination already attached by output routine.
2113 * Such approach solves two big problems:
2114 * 1. Not simplex devices are handled properly.
2115 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2116 * called with rcu_read_lock()
2119 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2120 u8 tos, struct net_device *dev)
2122 struct fib_result res;
2123 struct in_device *in_dev = __in_dev_get_rcu(dev);
2124 struct flowi fl = { .fl4_dst = daddr,
2127 .fl4_scope = RT_SCOPE_UNIVERSE,
2129 .iif = dev->ifindex };
2132 struct rtable * rth;
2136 struct net * net = dev_net(dev);
2138 /* IP on this device is disabled. */
2143 /* Check for the most weird martians, which can be not detected
2147 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2148 ipv4_is_loopback(saddr))
2149 goto martian_source;
2151 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2154 /* Accept zero addresses only to limited broadcast;
2155 * I even do not know to fix it or not. Waiting for complains :-)
2157 if (ipv4_is_zeronet(saddr))
2158 goto martian_source;
2160 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2161 goto martian_destination;
2164 * Now we are ready to route packet.
2166 err = fib_lookup(net, &fl, &res);
2168 if (!IN_DEV_FORWARD(in_dev))
2173 RT_CACHE_STAT_INC(in_slow_tot);
2175 if (res.type == RTN_BROADCAST)
2178 if (res.type == RTN_LOCAL) {
2179 err = fib_validate_source(saddr, daddr, tos,
2180 net->loopback_dev->ifindex,
2181 dev, &spec_dst, &itag, skb->mark);
2183 goto martian_source_keep_err;
2185 flags |= RTCF_DIRECTSRC;
2190 if (!IN_DEV_FORWARD(in_dev))
2192 if (res.type != RTN_UNICAST)
2193 goto martian_destination;
2195 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2199 if (skb->protocol != htons(ETH_P_IP))
2202 if (ipv4_is_zeronet(saddr))
2203 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2205 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2208 goto martian_source_keep_err;
2210 flags |= RTCF_DIRECTSRC;
2212 flags |= RTCF_BROADCAST;
2213 res.type = RTN_BROADCAST;
2214 RT_CACHE_STAT_INC(in_brd);
2217 rth = dst_alloc(&ipv4_dst_ops);
2221 rth->dst.output= ip_rt_bug;
2222 rth->dst.obsolete = -1;
2223 rth->rt_genid = rt_genid(net);
2225 atomic_set(&rth->dst.__refcnt, 1);
2226 rth->dst.flags= DST_HOST;
2227 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2228 rth->dst.flags |= DST_NOPOLICY;
2229 rth->fl.fl4_dst = daddr;
2230 rth->rt_dst = daddr;
2231 rth->fl.fl4_tos = tos;
2232 rth->fl.mark = skb->mark;
2233 rth->fl.fl4_src = saddr;
2234 rth->rt_src = saddr;
2235 #ifdef CONFIG_IP_ROUTE_CLASSID
2236 rth->dst.tclassid = itag;
2239 rth->fl.iif = dev->ifindex;
2240 rth->dst.dev = net->loopback_dev;
2241 dev_hold(rth->dst.dev);
2242 rth->rt_gateway = daddr;
2243 rth->rt_spec_dst= spec_dst;
2244 rth->dst.input= ip_local_deliver;
2245 rth->rt_flags = flags|RTCF_LOCAL;
2246 if (res.type == RTN_UNREACHABLE) {
2247 rth->dst.input= ip_error;
2248 rth->dst.error= -err;
2249 rth->rt_flags &= ~RTCF_LOCAL;
2251 rth->rt_type = res.type;
2252 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2253 err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2257 RT_CACHE_STAT_INC(in_no_route);
2258 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2259 res.type = RTN_UNREACHABLE;
2265 * Do not cache martian addresses: they should be logged (RFC1812)
2267 martian_destination:
2268 RT_CACHE_STAT_INC(in_martian_dst);
2269 #ifdef CONFIG_IP_ROUTE_VERBOSE
2270 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2271 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2272 &daddr, &saddr, dev->name);
2276 err = -EHOSTUNREACH;
2289 martian_source_keep_err:
2290 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2294 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2295 u8 tos, struct net_device *dev, bool noref)
2297 struct rtable * rth;
2299 int iif = dev->ifindex;
2307 if (!rt_caching(net))
2310 tos &= IPTOS_RT_MASK;
2311 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2313 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2314 rth = rcu_dereference(rth->dst.rt_next)) {
2315 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2316 ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2317 (rth->fl.iif ^ iif) |
2319 (rth->fl.fl4_tos ^ tos)) == 0 &&
2320 rth->fl.mark == skb->mark &&
2321 net_eq(dev_net(rth->dst.dev), net) &&
2322 !rt_is_expired(rth)) {
2324 dst_use_noref(&rth->dst, jiffies);
2325 skb_dst_set_noref(skb, &rth->dst);
2327 dst_use(&rth->dst, jiffies);
2328 skb_dst_set(skb, &rth->dst);
2330 RT_CACHE_STAT_INC(in_hit);
2334 RT_CACHE_STAT_INC(in_hlist_search);
2338 /* Multicast recognition logic is moved from route cache to here.
2339 The problem was that too many Ethernet cards have broken/missing
2340 hardware multicast filters :-( As result the host on multicasting
2341 network acquires a lot of useless route cache entries, sort of
2342 SDR messages from all the world. Now we try to get rid of them.
2343 Really, provided software IP multicast filter is organized
2344 reasonably (at least, hashed), it does not result in a slowdown
2345 comparing with route cache reject entries.
2346 Note, that multicast routers are not affected, because
2347 route cache entry is created eventually.
2349 if (ipv4_is_multicast(daddr)) {
2350 struct in_device *in_dev = __in_dev_get_rcu(dev);
2353 int our = ip_check_mc(in_dev, daddr, saddr,
2354 ip_hdr(skb)->protocol);
2356 #ifdef CONFIG_IP_MROUTE
2358 (!ipv4_is_local_multicast(daddr) &&
2359 IN_DEV_MFORWARD(in_dev))
2362 int res = ip_route_input_mc(skb, daddr, saddr,
2371 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2375 EXPORT_SYMBOL(ip_route_input_common);
2377 /* called with rcu_read_lock() */
2378 static int __mkroute_output(struct rtable **result,
2379 struct fib_result *res,
2380 const struct flowi *fl,
2381 const struct flowi *oldflp,
2382 struct net_device *dev_out,
2386 struct in_device *in_dev;
2387 u32 tos = RT_FL_TOS(oldflp);
2389 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2392 if (ipv4_is_lbcast(fl->fl4_dst))
2393 res->type = RTN_BROADCAST;
2394 else if (ipv4_is_multicast(fl->fl4_dst))
2395 res->type = RTN_MULTICAST;
2396 else if (ipv4_is_zeronet(fl->fl4_dst))
2399 if (dev_out->flags & IFF_LOOPBACK)
2400 flags |= RTCF_LOCAL;
2402 in_dev = __in_dev_get_rcu(dev_out);
2406 if (res->type == RTN_BROADCAST) {
2407 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2409 } else if (res->type == RTN_MULTICAST) {
2410 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2411 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2413 flags &= ~RTCF_LOCAL;
2414 /* If multicast route do not exist use
2415 * default one, but do not gateway in this case.
2418 if (res->fi && res->prefixlen < 4)
2423 rth = dst_alloc(&ipv4_dst_ops);
2427 atomic_set(&rth->dst.__refcnt, 1);
2428 rth->dst.flags= DST_HOST;
2429 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2430 rth->dst.flags |= DST_NOXFRM;
2431 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2432 rth->dst.flags |= DST_NOPOLICY;
2434 rth->fl.fl4_dst = oldflp->fl4_dst;
2435 rth->fl.fl4_tos = tos;
2436 rth->fl.fl4_src = oldflp->fl4_src;
2437 rth->fl.oif = oldflp->oif;
2438 rth->fl.mark = oldflp->mark;
2439 rth->rt_dst = fl->fl4_dst;
2440 rth->rt_src = fl->fl4_src;
2441 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2442 /* get references to the devices that are to be hold by the routing
2444 rth->dst.dev = dev_out;
2446 rth->rt_gateway = fl->fl4_dst;
2447 rth->rt_spec_dst= fl->fl4_src;
2449 rth->dst.output=ip_output;
2450 rth->dst.obsolete = -1;
2451 rth->rt_genid = rt_genid(dev_net(dev_out));
2453 RT_CACHE_STAT_INC(out_slow_tot);
2455 if (flags & RTCF_LOCAL) {
2456 rth->dst.input = ip_local_deliver;
2457 rth->rt_spec_dst = fl->fl4_dst;
2459 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2460 rth->rt_spec_dst = fl->fl4_src;
2461 if (flags & RTCF_LOCAL &&
2462 !(dev_out->flags & IFF_LOOPBACK)) {
2463 rth->dst.output = ip_mc_output;
2464 RT_CACHE_STAT_INC(out_slow_mc);
2466 #ifdef CONFIG_IP_MROUTE
2467 if (res->type == RTN_MULTICAST) {
2468 if (IN_DEV_MFORWARD(in_dev) &&
2469 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2470 rth->dst.input = ip_mr_input;
2471 rth->dst.output = ip_mc_output;
2477 rt_set_nexthop(rth, res, 0);
2479 rth->rt_flags = flags;
2484 /* called with rcu_read_lock() */
2485 static int ip_mkroute_output(struct rtable **rp,
2486 struct fib_result *res,
2487 const struct flowi *fl,
2488 const struct flowi *oldflp,
2489 struct net_device *dev_out,
2492 struct rtable *rth = NULL;
2493 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2496 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2497 rt_genid(dev_net(dev_out)));
2498 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2505 * Major route resolver routine.
2506 * called with rcu_read_lock();
2509 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2510 const struct flowi *oldflp)
2512 u32 tos = RT_FL_TOS(oldflp);
2513 struct flowi fl = { .fl4_dst = oldflp->fl4_dst,
2514 .fl4_src = oldflp->fl4_src,
2515 .fl4_tos = tos & IPTOS_RT_MASK,
2516 .fl4_scope = ((tos & RTO_ONLINK) ?
2517 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE),
2518 .mark = oldflp->mark,
2519 .iif = net->loopback_dev->ifindex,
2520 .oif = oldflp->oif };
2521 struct fib_result res;
2522 unsigned int flags = 0;
2523 struct net_device *dev_out = NULL;
2528 #ifdef CONFIG_IP_MULTIPLE_TABLES
2532 if (oldflp->fl4_src) {
2534 if (ipv4_is_multicast(oldflp->fl4_src) ||
2535 ipv4_is_lbcast(oldflp->fl4_src) ||
2536 ipv4_is_zeronet(oldflp->fl4_src))
2539 /* I removed check for oif == dev_out->oif here.
2540 It was wrong for two reasons:
2541 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2542 is assigned to multiple interfaces.
2543 2. Moreover, we are allowed to send packets with saddr
2544 of another iface. --ANK
2547 if (oldflp->oif == 0 &&
2548 (ipv4_is_multicast(oldflp->fl4_dst) ||
2549 ipv4_is_lbcast(oldflp->fl4_dst))) {
2550 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2551 dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2552 if (dev_out == NULL)
2555 /* Special hack: user can direct multicasts
2556 and limited broadcast via necessary interface
2557 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2558 This hack is not just for fun, it allows
2559 vic,vat and friends to work.
2560 They bind socket to loopback, set ttl to zero
2561 and expect that it will work.
2562 From the viewpoint of routing cache they are broken,
2563 because we are not allowed to build multicast path
2564 with loopback source addr (look, routing cache
2565 cannot know, that ttl is zero, so that packet
2566 will not leave this host and route is valid).
2567 Luckily, this hack is good workaround.
2570 fl.oif = dev_out->ifindex;
2574 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2575 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2576 if (!__ip_dev_find(net, oldflp->fl4_src, false))
2583 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2585 if (dev_out == NULL)
2588 /* RACE: Check return value of inet_select_addr instead. */
2589 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2593 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2594 ipv4_is_lbcast(oldflp->fl4_dst)) {
2596 fl.fl4_src = inet_select_addr(dev_out, 0,
2601 if (ipv4_is_multicast(oldflp->fl4_dst))
2602 fl.fl4_src = inet_select_addr(dev_out, 0,
2604 else if (!oldflp->fl4_dst)
2605 fl.fl4_src = inet_select_addr(dev_out, 0,
2611 fl.fl4_dst = fl.fl4_src;
2613 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2614 dev_out = net->loopback_dev;
2615 fl.oif = net->loopback_dev->ifindex;
2616 res.type = RTN_LOCAL;
2617 flags |= RTCF_LOCAL;
2621 if (fib_lookup(net, &fl, &res)) {
2624 /* Apparently, routing tables are wrong. Assume,
2625 that the destination is on link.
2628 Because we are allowed to send to iface
2629 even if it has NO routes and NO assigned
2630 addresses. When oif is specified, routing
2631 tables are looked up with only one purpose:
2632 to catch if destination is gatewayed, rather than
2633 direct. Moreover, if MSG_DONTROUTE is set,
2634 we send packet, ignoring both routing tables
2635 and ifaddr state. --ANK
2638 We could make it even if oif is unknown,
2639 likely IPv6, but we do not.
2642 if (fl.fl4_src == 0)
2643 fl.fl4_src = inet_select_addr(dev_out, 0,
2645 res.type = RTN_UNICAST;
2652 if (res.type == RTN_LOCAL) {
2654 if (res.fi->fib_prefsrc)
2655 fl.fl4_src = res.fi->fib_prefsrc;
2657 fl.fl4_src = fl.fl4_dst;
2659 dev_out = net->loopback_dev;
2660 fl.oif = dev_out->ifindex;
2662 flags |= RTCF_LOCAL;
2666 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2667 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2668 fib_select_multipath(&fl, &res);
2671 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2672 fib_select_default(&res);
2675 fl.fl4_src = FIB_RES_PREFSRC(res);
2677 dev_out = FIB_RES_DEV(res);
2678 fl.oif = dev_out->ifindex;
2682 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2687 int __ip_route_output_key(struct net *net, struct rtable **rp,
2688 const struct flowi *flp)
2694 if (!rt_caching(net))
2697 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2700 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2701 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2702 if (rth->fl.fl4_dst == flp->fl4_dst &&
2703 rth->fl.fl4_src == flp->fl4_src &&
2704 rt_is_output_route(rth) &&
2705 rth->fl.oif == flp->oif &&
2706 rth->fl.mark == flp->mark &&
2707 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2708 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2709 net_eq(dev_net(rth->dst.dev), net) &&
2710 !rt_is_expired(rth)) {
2711 dst_use(&rth->dst, jiffies);
2712 RT_CACHE_STAT_INC(out_hit);
2713 rcu_read_unlock_bh();
2717 RT_CACHE_STAT_INC(out_hlist_search);
2719 rcu_read_unlock_bh();
2723 res = ip_route_output_slow(net, rp, flp);
2727 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2729 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2734 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2739 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2743 static struct dst_ops ipv4_dst_blackhole_ops = {
2745 .protocol = cpu_to_be16(ETH_P_IP),
2746 .destroy = ipv4_dst_destroy,
2747 .check = ipv4_blackhole_dst_check,
2748 .default_mtu = ipv4_blackhole_default_mtu,
2749 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2753 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2755 struct rtable *ort = *rp;
2756 struct rtable *rt = (struct rtable *)
2757 dst_alloc(&ipv4_dst_blackhole_ops);
2760 struct dst_entry *new = &rt->dst;
2762 atomic_set(&new->__refcnt, 1);
2764 new->input = dst_discard;
2765 new->output = dst_discard;
2766 dst_copy_metrics(new, &ort->dst);
2768 new->dev = ort->dst.dev;
2774 rt->rt_genid = rt_genid(net);
2775 rt->rt_flags = ort->rt_flags;
2776 rt->rt_type = ort->rt_type;
2777 rt->rt_dst = ort->rt_dst;
2778 rt->rt_src = ort->rt_src;
2779 rt->rt_iif = ort->rt_iif;
2780 rt->rt_gateway = ort->rt_gateway;
2781 rt->rt_spec_dst = ort->rt_spec_dst;
2782 rt->peer = ort->peer;
2784 atomic_inc(&rt->peer->refcnt);
2787 atomic_inc(&rt->fi->fib_clntref);
2792 dst_release(&(*rp)->dst);
2794 return rt ? 0 : -ENOMEM;
2797 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2798 struct sock *sk, int flags)
2802 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2807 flp->fl4_src = (*rp)->rt_src;
2809 flp->fl4_dst = (*rp)->rt_dst;
2810 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2811 flags ? XFRM_LOOKUP_WAIT : 0);
2812 if (err == -EREMOTE)
2813 err = ipv4_dst_blackhole(net, rp, flp);
2820 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2822 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2824 return ip_route_output_flow(net, rp, flp, NULL, 0);
2826 EXPORT_SYMBOL(ip_route_output_key);
2828 static int rt_fill_info(struct net *net,
2829 struct sk_buff *skb, u32 pid, u32 seq, int event,
2830 int nowait, unsigned int flags)
2832 struct rtable *rt = skb_rtable(skb);
2834 struct nlmsghdr *nlh;
2836 u32 id = 0, ts = 0, tsage = 0, error;
2838 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2842 r = nlmsg_data(nlh);
2843 r->rtm_family = AF_INET;
2844 r->rtm_dst_len = 32;
2846 r->rtm_tos = rt->fl.fl4_tos;
2847 r->rtm_table = RT_TABLE_MAIN;
2848 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2849 r->rtm_type = rt->rt_type;
2850 r->rtm_scope = RT_SCOPE_UNIVERSE;
2851 r->rtm_protocol = RTPROT_UNSPEC;
2852 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2853 if (rt->rt_flags & RTCF_NOTIFY)
2854 r->rtm_flags |= RTM_F_NOTIFY;
2856 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2858 if (rt->fl.fl4_src) {
2859 r->rtm_src_len = 32;
2860 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2863 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2864 #ifdef CONFIG_IP_ROUTE_CLASSID
2865 if (rt->dst.tclassid)
2866 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2868 if (rt_is_input_route(rt))
2869 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2870 else if (rt->rt_src != rt->fl.fl4_src)
2871 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2873 if (rt->rt_dst != rt->rt_gateway)
2874 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2876 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2877 goto nla_put_failure;
2880 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2882 error = rt->dst.error;
2883 expires = (rt->peer && rt->peer->pmtu_expires) ?
2884 rt->peer->pmtu_expires - jiffies : 0;
2886 inet_peer_refcheck(rt->peer);
2887 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2888 if (rt->peer->tcp_ts_stamp) {
2889 ts = rt->peer->tcp_ts;
2890 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2894 if (rt_is_input_route(rt)) {
2895 #ifdef CONFIG_IP_MROUTE
2896 __be32 dst = rt->rt_dst;
2898 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2899 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2900 int err = ipmr_get_route(net, skb, r, nowait);
2905 goto nla_put_failure;
2907 if (err == -EMSGSIZE)
2908 goto nla_put_failure;
2914 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2917 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2918 expires, error) < 0)
2919 goto nla_put_failure;
2921 return nlmsg_end(skb, nlh);
2924 nlmsg_cancel(skb, nlh);
2928 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2930 struct net *net = sock_net(in_skb->sk);
2932 struct nlattr *tb[RTA_MAX+1];
2933 struct rtable *rt = NULL;
2939 struct sk_buff *skb;
2941 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2945 rtm = nlmsg_data(nlh);
2947 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2953 /* Reserve room for dummy headers, this skb can pass
2954 through good chunk of routing engine.
2956 skb_reset_mac_header(skb);
2957 skb_reset_network_header(skb);
2959 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2960 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2961 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2963 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2964 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2965 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2966 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2969 struct net_device *dev;
2971 dev = __dev_get_by_index(net, iif);
2977 skb->protocol = htons(ETH_P_IP);
2981 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2984 rt = skb_rtable(skb);
2985 if (err == 0 && rt->dst.error)
2986 err = -rt->dst.error;
2991 .fl4_tos = rtm->rtm_tos,
2992 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2995 err = ip_route_output_key(net, &rt, &fl);
3001 skb_dst_set(skb, &rt->dst);
3002 if (rtm->rtm_flags & RTM_F_NOTIFY)
3003 rt->rt_flags |= RTCF_NOTIFY;
3005 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3006 RTM_NEWROUTE, 0, 0);
3010 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3019 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3026 net = sock_net(skb->sk);
3031 s_idx = idx = cb->args[1];
3032 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3033 if (!rt_hash_table[h].chain)
3036 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3037 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3038 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3040 if (rt_is_expired(rt))
3042 skb_dst_set_noref(skb, &rt->dst);
3043 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3044 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3045 1, NLM_F_MULTI) <= 0) {
3047 rcu_read_unlock_bh();
3052 rcu_read_unlock_bh();
3061 void ip_rt_multicast_event(struct in_device *in_dev)
3063 rt_cache_flush(dev_net(in_dev->dev), 0);
3066 #ifdef CONFIG_SYSCTL
3067 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3068 void __user *buffer,
3069 size_t *lenp, loff_t *ppos)
3076 memcpy(&ctl, __ctl, sizeof(ctl));
3077 ctl.data = &flush_delay;
3078 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3080 net = (struct net *)__ctl->extra1;
3081 rt_cache_flush(net, flush_delay);
3088 static ctl_table ipv4_route_table[] = {
3090 .procname = "gc_thresh",
3091 .data = &ipv4_dst_ops.gc_thresh,
3092 .maxlen = sizeof(int),
3094 .proc_handler = proc_dointvec,
3097 .procname = "max_size",
3098 .data = &ip_rt_max_size,
3099 .maxlen = sizeof(int),
3101 .proc_handler = proc_dointvec,
3104 /* Deprecated. Use gc_min_interval_ms */
3106 .procname = "gc_min_interval",
3107 .data = &ip_rt_gc_min_interval,
3108 .maxlen = sizeof(int),
3110 .proc_handler = proc_dointvec_jiffies,
3113 .procname = "gc_min_interval_ms",
3114 .data = &ip_rt_gc_min_interval,
3115 .maxlen = sizeof(int),
3117 .proc_handler = proc_dointvec_ms_jiffies,
3120 .procname = "gc_timeout",
3121 .data = &ip_rt_gc_timeout,
3122 .maxlen = sizeof(int),
3124 .proc_handler = proc_dointvec_jiffies,
3127 .procname = "gc_interval",
3128 .data = &ip_rt_gc_interval,
3129 .maxlen = sizeof(int),
3131 .proc_handler = proc_dointvec_jiffies,
3134 .procname = "redirect_load",
3135 .data = &ip_rt_redirect_load,
3136 .maxlen = sizeof(int),
3138 .proc_handler = proc_dointvec,
3141 .procname = "redirect_number",
3142 .data = &ip_rt_redirect_number,
3143 .maxlen = sizeof(int),
3145 .proc_handler = proc_dointvec,
3148 .procname = "redirect_silence",
3149 .data = &ip_rt_redirect_silence,
3150 .maxlen = sizeof(int),
3152 .proc_handler = proc_dointvec,
3155 .procname = "error_cost",
3156 .data = &ip_rt_error_cost,
3157 .maxlen = sizeof(int),
3159 .proc_handler = proc_dointvec,
3162 .procname = "error_burst",
3163 .data = &ip_rt_error_burst,
3164 .maxlen = sizeof(int),
3166 .proc_handler = proc_dointvec,
3169 .procname = "gc_elasticity",
3170 .data = &ip_rt_gc_elasticity,
3171 .maxlen = sizeof(int),
3173 .proc_handler = proc_dointvec,
3176 .procname = "mtu_expires",
3177 .data = &ip_rt_mtu_expires,
3178 .maxlen = sizeof(int),
3180 .proc_handler = proc_dointvec_jiffies,
3183 .procname = "min_pmtu",
3184 .data = &ip_rt_min_pmtu,
3185 .maxlen = sizeof(int),
3187 .proc_handler = proc_dointvec,
3190 .procname = "min_adv_mss",
3191 .data = &ip_rt_min_advmss,
3192 .maxlen = sizeof(int),
3194 .proc_handler = proc_dointvec,
3199 static struct ctl_table empty[1];
3201 static struct ctl_table ipv4_skeleton[] =
3203 { .procname = "route",
3204 .mode = 0555, .child = ipv4_route_table},
3205 { .procname = "neigh",
3206 .mode = 0555, .child = empty},
3210 static __net_initdata struct ctl_path ipv4_path[] = {
3211 { .procname = "net", },
3212 { .procname = "ipv4", },
3216 static struct ctl_table ipv4_route_flush_table[] = {
3218 .procname = "flush",
3219 .maxlen = sizeof(int),
3221 .proc_handler = ipv4_sysctl_rtcache_flush,
3226 static __net_initdata struct ctl_path ipv4_route_path[] = {
3227 { .procname = "net", },
3228 { .procname = "ipv4", },
3229 { .procname = "route", },
3233 static __net_init int sysctl_route_net_init(struct net *net)
3235 struct ctl_table *tbl;
3237 tbl = ipv4_route_flush_table;
3238 if (!net_eq(net, &init_net)) {
3239 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3243 tbl[0].extra1 = net;
3245 net->ipv4.route_hdr =
3246 register_net_sysctl_table(net, ipv4_route_path, tbl);
3247 if (net->ipv4.route_hdr == NULL)
3252 if (tbl != ipv4_route_flush_table)
3258 static __net_exit void sysctl_route_net_exit(struct net *net)
3260 struct ctl_table *tbl;
3262 tbl = net->ipv4.route_hdr->ctl_table_arg;
3263 unregister_net_sysctl_table(net->ipv4.route_hdr);
3264 BUG_ON(tbl == ipv4_route_flush_table);
3268 static __net_initdata struct pernet_operations sysctl_route_ops = {
3269 .init = sysctl_route_net_init,
3270 .exit = sysctl_route_net_exit,
3274 static __net_init int rt_genid_init(struct net *net)
3276 get_random_bytes(&net->ipv4.rt_genid,
3277 sizeof(net->ipv4.rt_genid));
3281 static __net_initdata struct pernet_operations rt_genid_ops = {
3282 .init = rt_genid_init,
3286 #ifdef CONFIG_IP_ROUTE_CLASSID
3287 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3288 #endif /* CONFIG_IP_ROUTE_CLASSID */
3290 static __initdata unsigned long rhash_entries;
3291 static int __init set_rhash_entries(char *str)
3295 rhash_entries = simple_strtoul(str, &str, 0);
3298 __setup("rhash_entries=", set_rhash_entries);
3300 int __init ip_rt_init(void)
3304 #ifdef CONFIG_IP_ROUTE_CLASSID
3305 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3307 panic("IP: failed to allocate ip_rt_acct\n");
3310 ipv4_dst_ops.kmem_cachep =
3311 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3312 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3314 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3316 if (dst_entries_init(&ipv4_dst_ops) < 0)
3317 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3319 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3320 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3322 rt_hash_table = (struct rt_hash_bucket *)
3323 alloc_large_system_hash("IP route cache",
3324 sizeof(struct rt_hash_bucket),
3326 (totalram_pages >= 128 * 1024) ?
3331 rhash_entries ? 0 : 512 * 1024);
3332 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3333 rt_hash_lock_init();
3335 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3336 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3341 if (ip_rt_proc_init())
3342 printk(KERN_ERR "Unable to create route proc files\n");
3345 xfrm4_init(ip_rt_max_size);
3347 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3349 #ifdef CONFIG_SYSCTL
3350 register_pernet_subsys(&sysctl_route_ops);
3352 register_pernet_subsys(&rt_genid_ops);
3356 #ifdef CONFIG_SYSCTL
3358 * We really need to sanitize the damn ipv4 init order, then all
3359 * this nonsense will go away.
3361 void __init ip_static_sysctl_init(void)
3363 register_sysctl_paths(ipv4_path, ipv4_skeleton);