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>
111 #include <net/atmclip.h>
113 #define RT_FL_TOS(oldflp4) \
114 ((u32)(oldflp4->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
116 #define IP_MAX_MTU 0xFFF0
118 #define RT_GC_TIMEOUT (300*HZ)
120 static int ip_rt_max_size;
121 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
122 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
123 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
124 static int ip_rt_redirect_number __read_mostly = 9;
125 static int ip_rt_redirect_load __read_mostly = HZ / 50;
126 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly = HZ;
128 static int ip_rt_error_burst __read_mostly = 5 * HZ;
129 static int ip_rt_gc_elasticity __read_mostly = 8;
130 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
131 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
132 static int ip_rt_min_advmss __read_mostly = 256;
133 static int rt_chain_length_max __read_mostly = 20;
136 * Interface to generic destination cache.
139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
140 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
141 static unsigned int ipv4_default_mtu(const struct dst_entry *dst);
142 static void ipv4_dst_destroy(struct dst_entry *dst);
143 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
144 static void ipv4_link_failure(struct sk_buff *skb);
145 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
146 static int rt_garbage_collect(struct dst_ops *ops);
148 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
153 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
155 struct rtable *rt = (struct rtable *) dst;
156 struct inet_peer *peer;
160 rt_bind_peer(rt, rt->rt_dst, 1);
164 u32 *old_p = __DST_METRICS_PTR(old);
165 unsigned long prev, new;
168 if (inet_metrics_new(peer))
169 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
171 new = (unsigned long) p;
172 prev = cmpxchg(&dst->_metrics, old, new);
175 p = __DST_METRICS_PTR(prev);
176 if (prev & DST_METRICS_READ_ONLY)
180 fib_info_put(rt->fi);
188 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr);
190 static struct dst_ops ipv4_dst_ops = {
192 .protocol = cpu_to_be16(ETH_P_IP),
193 .gc = rt_garbage_collect,
194 .check = ipv4_dst_check,
195 .default_advmss = ipv4_default_advmss,
196 .default_mtu = ipv4_default_mtu,
197 .cow_metrics = ipv4_cow_metrics,
198 .destroy = ipv4_dst_destroy,
199 .ifdown = ipv4_dst_ifdown,
200 .negative_advice = ipv4_negative_advice,
201 .link_failure = ipv4_link_failure,
202 .update_pmtu = ip_rt_update_pmtu,
203 .local_out = __ip_local_out,
204 .neigh_lookup = ipv4_neigh_lookup,
207 #define ECN_OR_COST(class) TC_PRIO_##class
209 const __u8 ip_tos2prio[16] = {
211 ECN_OR_COST(BESTEFFORT),
213 ECN_OR_COST(BESTEFFORT),
219 ECN_OR_COST(INTERACTIVE),
221 ECN_OR_COST(INTERACTIVE),
222 TC_PRIO_INTERACTIVE_BULK,
223 ECN_OR_COST(INTERACTIVE_BULK),
224 TC_PRIO_INTERACTIVE_BULK,
225 ECN_OR_COST(INTERACTIVE_BULK)
233 /* The locking scheme is rather straight forward:
235 * 1) Read-Copy Update protects the buckets of the central route hash.
236 * 2) Only writers remove entries, and they hold the lock
237 * as they look at rtable reference counts.
238 * 3) Only readers acquire references to rtable entries,
239 * they do so with atomic increments and with the
243 struct rt_hash_bucket {
244 struct rtable __rcu *chain;
247 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
248 defined(CONFIG_PROVE_LOCKING)
250 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
251 * The size of this table is a power of two and depends on the number of CPUS.
252 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
254 #ifdef CONFIG_LOCKDEP
255 # define RT_HASH_LOCK_SZ 256
258 # define RT_HASH_LOCK_SZ 4096
260 # define RT_HASH_LOCK_SZ 2048
262 # define RT_HASH_LOCK_SZ 1024
264 # define RT_HASH_LOCK_SZ 512
266 # define RT_HASH_LOCK_SZ 256
270 static spinlock_t *rt_hash_locks;
271 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
273 static __init void rt_hash_lock_init(void)
277 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
280 panic("IP: failed to allocate rt_hash_locks\n");
282 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
283 spin_lock_init(&rt_hash_locks[i]);
286 # define rt_hash_lock_addr(slot) NULL
288 static inline void rt_hash_lock_init(void)
293 static struct rt_hash_bucket *rt_hash_table __read_mostly;
294 static unsigned rt_hash_mask __read_mostly;
295 static unsigned int rt_hash_log __read_mostly;
297 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
298 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
300 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
303 return jhash_3words((__force u32)daddr, (__force u32)saddr,
308 static inline int rt_genid(struct net *net)
310 return atomic_read(&net->ipv4.rt_genid);
313 #ifdef CONFIG_PROC_FS
314 struct rt_cache_iter_state {
315 struct seq_net_private p;
320 static struct rtable *rt_cache_get_first(struct seq_file *seq)
322 struct rt_cache_iter_state *st = seq->private;
323 struct rtable *r = NULL;
325 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
326 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
329 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
331 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
332 r->rt_genid == st->genid)
334 r = rcu_dereference_bh(r->dst.rt_next);
336 rcu_read_unlock_bh();
341 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
344 struct rt_cache_iter_state *st = seq->private;
346 r = rcu_dereference_bh(r->dst.rt_next);
348 rcu_read_unlock_bh();
350 if (--st->bucket < 0)
352 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
354 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
359 static struct rtable *rt_cache_get_next(struct seq_file *seq,
362 struct rt_cache_iter_state *st = seq->private;
363 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
364 if (dev_net(r->dst.dev) != seq_file_net(seq))
366 if (r->rt_genid == st->genid)
372 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
374 struct rtable *r = rt_cache_get_first(seq);
377 while (pos && (r = rt_cache_get_next(seq, r)))
379 return pos ? NULL : r;
382 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
384 struct rt_cache_iter_state *st = seq->private;
386 return rt_cache_get_idx(seq, *pos - 1);
387 st->genid = rt_genid(seq_file_net(seq));
388 return SEQ_START_TOKEN;
391 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
395 if (v == SEQ_START_TOKEN)
396 r = rt_cache_get_first(seq);
398 r = rt_cache_get_next(seq, v);
403 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
405 if (v && v != SEQ_START_TOKEN)
406 rcu_read_unlock_bh();
409 static int rt_cache_seq_show(struct seq_file *seq, void *v)
411 if (v == SEQ_START_TOKEN)
412 seq_printf(seq, "%-127s\n",
413 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
414 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
417 struct rtable *r = v;
421 n = dst_get_neighbour(&r->dst);
422 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
423 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
424 r->dst.dev ? r->dst.dev->name : "*",
425 (__force u32)r->rt_dst,
426 (__force u32)r->rt_gateway,
427 r->rt_flags, atomic_read(&r->dst.__refcnt),
428 r->dst.__use, 0, (__force u32)r->rt_src,
429 dst_metric_advmss(&r->dst) + 40,
430 dst_metric(&r->dst, RTAX_WINDOW),
431 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
432 dst_metric(&r->dst, RTAX_RTTVAR)),
435 (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0,
436 r->rt_spec_dst, &len);
438 seq_printf(seq, "%*s\n", 127 - len, "");
443 static const struct seq_operations rt_cache_seq_ops = {
444 .start = rt_cache_seq_start,
445 .next = rt_cache_seq_next,
446 .stop = rt_cache_seq_stop,
447 .show = rt_cache_seq_show,
450 static int rt_cache_seq_open(struct inode *inode, struct file *file)
452 return seq_open_net(inode, file, &rt_cache_seq_ops,
453 sizeof(struct rt_cache_iter_state));
456 static const struct file_operations rt_cache_seq_fops = {
457 .owner = THIS_MODULE,
458 .open = rt_cache_seq_open,
461 .release = seq_release_net,
465 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
470 return SEQ_START_TOKEN;
472 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
473 if (!cpu_possible(cpu))
476 return &per_cpu(rt_cache_stat, cpu);
481 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
485 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
486 if (!cpu_possible(cpu))
489 return &per_cpu(rt_cache_stat, cpu);
495 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
500 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
502 struct rt_cache_stat *st = v;
504 if (v == SEQ_START_TOKEN) {
505 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");
509 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
510 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
511 dst_entries_get_slow(&ipv4_dst_ops),
534 static const struct seq_operations rt_cpu_seq_ops = {
535 .start = rt_cpu_seq_start,
536 .next = rt_cpu_seq_next,
537 .stop = rt_cpu_seq_stop,
538 .show = rt_cpu_seq_show,
542 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
544 return seq_open(file, &rt_cpu_seq_ops);
547 static const struct file_operations rt_cpu_seq_fops = {
548 .owner = THIS_MODULE,
549 .open = rt_cpu_seq_open,
552 .release = seq_release,
555 #ifdef CONFIG_IP_ROUTE_CLASSID
556 static int rt_acct_proc_show(struct seq_file *m, void *v)
558 struct ip_rt_acct *dst, *src;
561 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
565 for_each_possible_cpu(i) {
566 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
567 for (j = 0; j < 256; j++) {
568 dst[j].o_bytes += src[j].o_bytes;
569 dst[j].o_packets += src[j].o_packets;
570 dst[j].i_bytes += src[j].i_bytes;
571 dst[j].i_packets += src[j].i_packets;
575 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
580 static int rt_acct_proc_open(struct inode *inode, struct file *file)
582 return single_open(file, rt_acct_proc_show, NULL);
585 static const struct file_operations rt_acct_proc_fops = {
586 .owner = THIS_MODULE,
587 .open = rt_acct_proc_open,
590 .release = single_release,
594 static int __net_init ip_rt_do_proc_init(struct net *net)
596 struct proc_dir_entry *pde;
598 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
603 pde = proc_create("rt_cache", S_IRUGO,
604 net->proc_net_stat, &rt_cpu_seq_fops);
608 #ifdef CONFIG_IP_ROUTE_CLASSID
609 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
615 #ifdef CONFIG_IP_ROUTE_CLASSID
617 remove_proc_entry("rt_cache", net->proc_net_stat);
620 remove_proc_entry("rt_cache", net->proc_net);
625 static void __net_exit ip_rt_do_proc_exit(struct net *net)
627 remove_proc_entry("rt_cache", net->proc_net_stat);
628 remove_proc_entry("rt_cache", net->proc_net);
629 #ifdef CONFIG_IP_ROUTE_CLASSID
630 remove_proc_entry("rt_acct", net->proc_net);
634 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
635 .init = ip_rt_do_proc_init,
636 .exit = ip_rt_do_proc_exit,
639 static int __init ip_rt_proc_init(void)
641 return register_pernet_subsys(&ip_rt_proc_ops);
645 static inline int ip_rt_proc_init(void)
649 #endif /* CONFIG_PROC_FS */
651 static inline void rt_free(struct rtable *rt)
653 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
656 static inline void rt_drop(struct rtable *rt)
659 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
662 static inline int rt_fast_clean(struct rtable *rth)
664 /* Kill broadcast/multicast entries very aggresively, if they
665 collide in hash table with more useful entries */
666 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
667 rt_is_input_route(rth) && rth->dst.rt_next;
670 static inline int rt_valuable(struct rtable *rth)
672 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
673 (rth->peer && rth->peer->pmtu_expires);
676 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
681 if (atomic_read(&rth->dst.__refcnt))
684 age = jiffies - rth->dst.lastuse;
685 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
686 (age <= tmo2 && rt_valuable(rth)))
692 /* Bits of score are:
694 * 30: not quite useless
695 * 29..0: usage counter
697 static inline u32 rt_score(struct rtable *rt)
699 u32 score = jiffies - rt->dst.lastuse;
701 score = ~score & ~(3<<30);
706 if (rt_is_output_route(rt) ||
707 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
713 static inline bool rt_caching(const struct net *net)
715 return net->ipv4.current_rt_cache_rebuild_count <=
716 net->ipv4.sysctl_rt_cache_rebuild_count;
719 static inline bool compare_hash_inputs(const struct rtable *rt1,
720 const struct rtable *rt2)
722 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
723 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
724 (rt1->rt_iif ^ rt2->rt_iif)) == 0);
727 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
729 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
730 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
731 (rt1->rt_mark ^ rt2->rt_mark) |
732 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
733 (rt1->rt_oif ^ rt2->rt_oif) |
734 (rt1->rt_iif ^ rt2->rt_iif)) == 0;
737 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
739 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
742 static inline int rt_is_expired(struct rtable *rth)
744 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
748 * Perform a full scan of hash table and free all entries.
749 * Can be called by a softirq or a process.
750 * In the later case, we want to be reschedule if necessary
752 static void rt_do_flush(struct net *net, int process_context)
755 struct rtable *rth, *next;
757 for (i = 0; i <= rt_hash_mask; i++) {
758 struct rtable __rcu **pprev;
761 if (process_context && need_resched())
763 rth = rcu_dereference_raw(rt_hash_table[i].chain);
767 spin_lock_bh(rt_hash_lock_addr(i));
770 pprev = &rt_hash_table[i].chain;
771 rth = rcu_dereference_protected(*pprev,
772 lockdep_is_held(rt_hash_lock_addr(i)));
775 next = rcu_dereference_protected(rth->dst.rt_next,
776 lockdep_is_held(rt_hash_lock_addr(i)));
779 net_eq(dev_net(rth->dst.dev), net)) {
780 rcu_assign_pointer(*pprev, next);
781 rcu_assign_pointer(rth->dst.rt_next, list);
784 pprev = &rth->dst.rt_next;
789 spin_unlock_bh(rt_hash_lock_addr(i));
791 for (; list; list = next) {
792 next = rcu_dereference_protected(list->dst.rt_next, 1);
799 * While freeing expired entries, we compute average chain length
800 * and standard deviation, using fixed-point arithmetic.
801 * This to have an estimation of rt_chain_length_max
802 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
803 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
807 #define ONE (1UL << FRACT_BITS)
810 * Given a hash chain and an item in this hash chain,
811 * find if a previous entry has the same hash_inputs
812 * (but differs on tos, mark or oif)
813 * Returns 0 if an alias is found.
814 * Returns ONE if rth has no alias before itself.
816 static int has_noalias(const struct rtable *head, const struct rtable *rth)
818 const struct rtable *aux = head;
821 if (compare_hash_inputs(aux, rth))
823 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
829 * Perturbation of rt_genid by a small quantity [1..256]
830 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
831 * many times (2^24) without giving recent rt_genid.
832 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
834 static void rt_cache_invalidate(struct net *net)
836 unsigned char shuffle;
838 get_random_bytes(&shuffle, sizeof(shuffle));
839 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
843 * delay < 0 : invalidate cache (fast : entries will be deleted later)
844 * delay >= 0 : invalidate & flush cache (can be long)
846 void rt_cache_flush(struct net *net, int delay)
848 rt_cache_invalidate(net);
850 rt_do_flush(net, !in_softirq());
853 /* Flush previous cache invalidated entries from the cache */
854 void rt_cache_flush_batch(struct net *net)
856 rt_do_flush(net, !in_softirq());
859 static void rt_emergency_hash_rebuild(struct net *net)
862 printk(KERN_WARNING "Route hash chain too long!\n");
863 rt_cache_invalidate(net);
867 Short description of GC goals.
869 We want to build algorithm, which will keep routing cache
870 at some equilibrium point, when number of aged off entries
871 is kept approximately equal to newly generated ones.
873 Current expiration strength is variable "expire".
874 We try to adjust it dynamically, so that if networking
875 is idle expires is large enough to keep enough of warm entries,
876 and when load increases it reduces to limit cache size.
879 static int rt_garbage_collect(struct dst_ops *ops)
881 static unsigned long expire = RT_GC_TIMEOUT;
882 static unsigned long last_gc;
884 static int equilibrium;
886 struct rtable __rcu **rthp;
887 unsigned long now = jiffies;
889 int entries = dst_entries_get_fast(&ipv4_dst_ops);
892 * Garbage collection is pretty expensive,
893 * do not make it too frequently.
896 RT_CACHE_STAT_INC(gc_total);
898 if (now - last_gc < ip_rt_gc_min_interval &&
899 entries < ip_rt_max_size) {
900 RT_CACHE_STAT_INC(gc_ignored);
904 entries = dst_entries_get_slow(&ipv4_dst_ops);
905 /* Calculate number of entries, which we want to expire now. */
906 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
908 if (equilibrium < ipv4_dst_ops.gc_thresh)
909 equilibrium = ipv4_dst_ops.gc_thresh;
910 goal = entries - equilibrium;
912 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
913 goal = entries - equilibrium;
916 /* We are in dangerous area. Try to reduce cache really
919 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
920 equilibrium = entries - goal;
923 if (now - last_gc >= ip_rt_gc_min_interval)
934 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
935 unsigned long tmo = expire;
937 k = (k + 1) & rt_hash_mask;
938 rthp = &rt_hash_table[k].chain;
939 spin_lock_bh(rt_hash_lock_addr(k));
940 while ((rth = rcu_dereference_protected(*rthp,
941 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
942 if (!rt_is_expired(rth) &&
943 !rt_may_expire(rth, tmo, expire)) {
945 rthp = &rth->dst.rt_next;
948 *rthp = rth->dst.rt_next;
952 spin_unlock_bh(rt_hash_lock_addr(k));
961 /* Goal is not achieved. We stop process if:
963 - if expire reduced to zero. Otherwise, expire is halfed.
964 - if table is not full.
965 - if we are called from interrupt.
966 - jiffies check is just fallback/debug loop breaker.
967 We will not spin here for long time in any case.
970 RT_CACHE_STAT_INC(gc_goal_miss);
977 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
979 } while (!in_softirq() && time_before_eq(jiffies, now));
981 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
983 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
986 printk(KERN_WARNING "dst cache overflow\n");
987 RT_CACHE_STAT_INC(gc_dst_overflow);
991 expire += ip_rt_gc_min_interval;
992 if (expire > ip_rt_gc_timeout ||
993 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
994 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
995 expire = ip_rt_gc_timeout;
1000 * Returns number of entries in a hash chain that have different hash_inputs
1002 static int slow_chain_length(const struct rtable *head)
1005 const struct rtable *rth = head;
1008 length += has_noalias(head, rth);
1009 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1011 return length >> FRACT_BITS;
1014 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1016 struct neigh_table *tbl = &arp_tbl;
1017 static const __be32 inaddr_any = 0;
1018 struct net_device *dev = dst->dev;
1019 const __be32 *pkey = daddr;
1020 struct neighbour *n;
1022 #if defined(CONFIG_ATM_CLIP) || defined(CONFIG_ATM_CLIP_MODULE)
1023 if (dev->type == ARPHRD_ATM)
1024 tbl = clip_tbl_hook;
1026 if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1029 n = __ipv4_neigh_lookup(tbl, dev, *(__force u32 *)pkey);
1032 return neigh_create(tbl, pkey, dev);
1035 static int rt_bind_neighbour(struct rtable *rt)
1037 struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1040 dst_set_neighbour(&rt->dst, n);
1045 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1046 struct sk_buff *skb, int ifindex)
1048 struct rtable *rth, *cand;
1049 struct rtable __rcu **rthp, **candp;
1053 int attempts = !in_softirq();
1057 min_score = ~(u32)0;
1062 if (!rt_caching(dev_net(rt->dst.dev))) {
1064 * If we're not caching, just tell the caller we
1065 * were successful and don't touch the route. The
1066 * caller hold the sole reference to the cache entry, and
1067 * it will be released when the caller is done with it.
1068 * If we drop it here, the callers have no way to resolve routes
1069 * when we're not caching. Instead, just point *rp at rt, so
1070 * the caller gets a single use out of the route
1071 * Note that we do rt_free on this new route entry, so that
1072 * once its refcount hits zero, we are still able to reap it
1074 * Note: To avoid expensive rcu stuff for this uncached dst,
1075 * we set DST_NOCACHE so that dst_release() can free dst without
1076 * waiting a grace period.
1079 rt->dst.flags |= DST_NOCACHE;
1080 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1081 int err = rt_bind_neighbour(rt);
1083 if (net_ratelimit())
1085 "Neighbour table failure & not caching routes.\n");
1087 return ERR_PTR(err);
1094 rthp = &rt_hash_table[hash].chain;
1096 spin_lock_bh(rt_hash_lock_addr(hash));
1097 while ((rth = rcu_dereference_protected(*rthp,
1098 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1099 if (rt_is_expired(rth)) {
1100 *rthp = rth->dst.rt_next;
1104 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1106 *rthp = rth->dst.rt_next;
1108 * Since lookup is lockfree, the deletion
1109 * must be visible to another weakly ordered CPU before
1110 * the insertion at the start of the hash chain.
1112 rcu_assign_pointer(rth->dst.rt_next,
1113 rt_hash_table[hash].chain);
1115 * Since lookup is lockfree, the update writes
1116 * must be ordered for consistency on SMP.
1118 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1120 dst_use(&rth->dst, now);
1121 spin_unlock_bh(rt_hash_lock_addr(hash));
1125 skb_dst_set(skb, &rth->dst);
1129 if (!atomic_read(&rth->dst.__refcnt)) {
1130 u32 score = rt_score(rth);
1132 if (score <= min_score) {
1141 rthp = &rth->dst.rt_next;
1145 /* ip_rt_gc_elasticity used to be average length of chain
1146 * length, when exceeded gc becomes really aggressive.
1148 * The second limit is less certain. At the moment it allows
1149 * only 2 entries per bucket. We will see.
1151 if (chain_length > ip_rt_gc_elasticity) {
1152 *candp = cand->dst.rt_next;
1156 if (chain_length > rt_chain_length_max &&
1157 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1158 struct net *net = dev_net(rt->dst.dev);
1159 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1160 if (!rt_caching(net)) {
1161 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1162 rt->dst.dev->name, num);
1164 rt_emergency_hash_rebuild(net);
1165 spin_unlock_bh(rt_hash_lock_addr(hash));
1167 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1168 ifindex, rt_genid(net));
1173 /* Try to bind route to arp only if it is output
1174 route or unicast forwarding path.
1176 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1177 int err = rt_bind_neighbour(rt);
1179 spin_unlock_bh(rt_hash_lock_addr(hash));
1181 if (err != -ENOBUFS) {
1183 return ERR_PTR(err);
1186 /* Neighbour tables are full and nothing
1187 can be released. Try to shrink route cache,
1188 it is most likely it holds some neighbour records.
1190 if (attempts-- > 0) {
1191 int saved_elasticity = ip_rt_gc_elasticity;
1192 int saved_int = ip_rt_gc_min_interval;
1193 ip_rt_gc_elasticity = 1;
1194 ip_rt_gc_min_interval = 0;
1195 rt_garbage_collect(&ipv4_dst_ops);
1196 ip_rt_gc_min_interval = saved_int;
1197 ip_rt_gc_elasticity = saved_elasticity;
1201 if (net_ratelimit())
1202 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1204 return ERR_PTR(-ENOBUFS);
1208 rt->dst.rt_next = rt_hash_table[hash].chain;
1211 * Since lookup is lockfree, we must make sure
1212 * previous writes to rt are committed to memory
1213 * before making rt visible to other CPUS.
1215 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1217 spin_unlock_bh(rt_hash_lock_addr(hash));
1221 skb_dst_set(skb, &rt->dst);
1225 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1227 static u32 rt_peer_genid(void)
1229 return atomic_read(&__rt_peer_genid);
1232 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1234 struct inet_peer *peer;
1236 peer = inet_getpeer_v4(daddr, create);
1238 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1241 rt->rt_peer_genid = rt_peer_genid();
1245 * Peer allocation may fail only in serious out-of-memory conditions. However
1246 * we still can generate some output.
1247 * Random ID selection looks a bit dangerous because we have no chances to
1248 * select ID being unique in a reasonable period of time.
1249 * But broken packet identifier may be better than no packet at all.
1251 static void ip_select_fb_ident(struct iphdr *iph)
1253 static DEFINE_SPINLOCK(ip_fb_id_lock);
1254 static u32 ip_fallback_id;
1257 spin_lock_bh(&ip_fb_id_lock);
1258 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1259 iph->id = htons(salt & 0xFFFF);
1260 ip_fallback_id = salt;
1261 spin_unlock_bh(&ip_fb_id_lock);
1264 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1266 struct rtable *rt = (struct rtable *) dst;
1269 if (rt->peer == NULL)
1270 rt_bind_peer(rt, rt->rt_dst, 1);
1272 /* If peer is attached to destination, it is never detached,
1273 so that we need not to grab a lock to dereference it.
1276 iph->id = htons(inet_getid(rt->peer, more));
1280 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1281 __builtin_return_address(0));
1283 ip_select_fb_ident(iph);
1285 EXPORT_SYMBOL(__ip_select_ident);
1287 static void rt_del(unsigned hash, struct rtable *rt)
1289 struct rtable __rcu **rthp;
1292 rthp = &rt_hash_table[hash].chain;
1293 spin_lock_bh(rt_hash_lock_addr(hash));
1295 while ((aux = rcu_dereference_protected(*rthp,
1296 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1297 if (aux == rt || rt_is_expired(aux)) {
1298 *rthp = aux->dst.rt_next;
1302 rthp = &aux->dst.rt_next;
1304 spin_unlock_bh(rt_hash_lock_addr(hash));
1307 /* called in rcu_read_lock() section */
1308 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1309 __be32 saddr, struct net_device *dev)
1311 struct in_device *in_dev = __in_dev_get_rcu(dev);
1312 struct inet_peer *peer;
1319 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1320 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1321 ipv4_is_zeronet(new_gw))
1322 goto reject_redirect;
1324 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1325 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1326 goto reject_redirect;
1327 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1328 goto reject_redirect;
1330 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1331 goto reject_redirect;
1334 peer = inet_getpeer_v4(daddr, 1);
1336 peer->redirect_learned.a4 = new_gw;
1340 atomic_inc(&__rt_peer_genid);
1345 #ifdef CONFIG_IP_ROUTE_VERBOSE
1346 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1347 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1348 " Advised path = %pI4 -> %pI4\n",
1349 &old_gw, dev->name, &new_gw,
1355 static bool peer_pmtu_expired(struct inet_peer *peer)
1357 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1360 time_after_eq(jiffies, orig) &&
1361 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1364 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1366 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1369 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1372 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1374 struct rtable *rt = (struct rtable *)dst;
1375 struct dst_entry *ret = dst;
1378 if (dst->obsolete > 0) {
1381 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1382 unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1384 rt_genid(dev_net(dst->dev)));
1387 } else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1388 dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1396 * 1. The first ip_rt_redirect_number redirects are sent
1397 * with exponential backoff, then we stop sending them at all,
1398 * assuming that the host ignores our redirects.
1399 * 2. If we did not see packets requiring redirects
1400 * during ip_rt_redirect_silence, we assume that the host
1401 * forgot redirected route and start to send redirects again.
1403 * This algorithm is much cheaper and more intelligent than dumb load limiting
1406 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1407 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1410 void ip_rt_send_redirect(struct sk_buff *skb)
1412 struct rtable *rt = skb_rtable(skb);
1413 struct in_device *in_dev;
1414 struct inet_peer *peer;
1418 in_dev = __in_dev_get_rcu(rt->dst.dev);
1419 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1423 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1427 rt_bind_peer(rt, rt->rt_dst, 1);
1430 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1434 /* No redirected packets during ip_rt_redirect_silence;
1435 * reset the algorithm.
1437 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1438 peer->rate_tokens = 0;
1440 /* Too many ignored redirects; do not send anything
1441 * set dst.rate_last to the last seen redirected packet.
1443 if (peer->rate_tokens >= ip_rt_redirect_number) {
1444 peer->rate_last = jiffies;
1448 /* Check for load limit; set rate_last to the latest sent
1451 if (peer->rate_tokens == 0 ||
1454 (ip_rt_redirect_load << peer->rate_tokens)))) {
1455 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1456 peer->rate_last = jiffies;
1457 ++peer->rate_tokens;
1458 #ifdef CONFIG_IP_ROUTE_VERBOSE
1460 peer->rate_tokens == ip_rt_redirect_number &&
1462 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1463 &ip_hdr(skb)->saddr, rt->rt_iif,
1464 &rt->rt_dst, &rt->rt_gateway);
1469 static int ip_error(struct sk_buff *skb)
1471 struct rtable *rt = skb_rtable(skb);
1472 struct inet_peer *peer;
1477 switch (rt->dst.error) {
1482 code = ICMP_HOST_UNREACH;
1485 code = ICMP_NET_UNREACH;
1486 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1487 IPSTATS_MIB_INNOROUTES);
1490 code = ICMP_PKT_FILTERED;
1495 rt_bind_peer(rt, rt->rt_dst, 1);
1501 peer->rate_tokens += now - peer->rate_last;
1502 if (peer->rate_tokens > ip_rt_error_burst)
1503 peer->rate_tokens = ip_rt_error_burst;
1504 peer->rate_last = now;
1505 if (peer->rate_tokens >= ip_rt_error_cost)
1506 peer->rate_tokens -= ip_rt_error_cost;
1511 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1513 out: kfree_skb(skb);
1518 * The last two values are not from the RFC but
1519 * are needed for AMPRnet AX.25 paths.
1522 static const unsigned short mtu_plateau[] =
1523 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1525 static inline unsigned short guess_mtu(unsigned short old_mtu)
1529 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1530 if (old_mtu > mtu_plateau[i])
1531 return mtu_plateau[i];
1535 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1536 unsigned short new_mtu,
1537 struct net_device *dev)
1539 unsigned short old_mtu = ntohs(iph->tot_len);
1540 unsigned short est_mtu = 0;
1541 struct inet_peer *peer;
1543 peer = inet_getpeer_v4(iph->daddr, 1);
1545 unsigned short mtu = new_mtu;
1547 if (new_mtu < 68 || new_mtu >= old_mtu) {
1548 /* BSD 4.2 derived systems incorrectly adjust
1549 * tot_len by the IP header length, and report
1550 * a zero MTU in the ICMP message.
1553 old_mtu >= 68 + (iph->ihl << 2))
1554 old_mtu -= iph->ihl << 2;
1555 mtu = guess_mtu(old_mtu);
1558 if (mtu < ip_rt_min_pmtu)
1559 mtu = ip_rt_min_pmtu;
1560 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1561 unsigned long pmtu_expires;
1563 pmtu_expires = jiffies + ip_rt_mtu_expires;
1568 peer->pmtu_learned = mtu;
1569 peer->pmtu_expires = pmtu_expires;
1574 atomic_inc(&__rt_peer_genid);
1576 return est_mtu ? : new_mtu;
1579 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1581 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1585 if (time_before(jiffies, expires)) {
1586 u32 orig_dst_mtu = dst_mtu(dst);
1587 if (peer->pmtu_learned < orig_dst_mtu) {
1588 if (!peer->pmtu_orig)
1589 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1590 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1592 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1593 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1596 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1598 struct rtable *rt = (struct rtable *) dst;
1599 struct inet_peer *peer;
1604 rt_bind_peer(rt, rt->rt_dst, 1);
1607 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1609 if (mtu < ip_rt_min_pmtu)
1610 mtu = ip_rt_min_pmtu;
1611 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1613 pmtu_expires = jiffies + ip_rt_mtu_expires;
1617 peer->pmtu_learned = mtu;
1618 peer->pmtu_expires = pmtu_expires;
1620 atomic_inc(&__rt_peer_genid);
1621 rt->rt_peer_genid = rt_peer_genid();
1623 check_peer_pmtu(dst, peer);
1627 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1629 struct rtable *rt = (struct rtable *) dst;
1630 __be32 orig_gw = rt->rt_gateway;
1631 struct neighbour *n, *old_n;
1633 dst_confirm(&rt->dst);
1635 rt->rt_gateway = peer->redirect_learned.a4;
1637 n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1640 old_n = xchg(&rt->dst._neighbour, n);
1642 neigh_release(old_n);
1643 if (!n || !(n->nud_state & NUD_VALID)) {
1645 neigh_event_send(n, NULL);
1646 rt->rt_gateway = orig_gw;
1649 rt->rt_flags |= RTCF_REDIRECTED;
1650 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1655 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1657 struct rtable *rt = (struct rtable *) dst;
1659 if (rt_is_expired(rt))
1661 if (rt->rt_peer_genid != rt_peer_genid()) {
1662 struct inet_peer *peer;
1665 rt_bind_peer(rt, rt->rt_dst, 0);
1669 check_peer_pmtu(dst, peer);
1671 if (peer->redirect_learned.a4 &&
1672 peer->redirect_learned.a4 != rt->rt_gateway) {
1673 if (check_peer_redir(dst, peer))
1678 rt->rt_peer_genid = rt_peer_genid();
1683 static void ipv4_dst_destroy(struct dst_entry *dst)
1685 struct rtable *rt = (struct rtable *) dst;
1686 struct inet_peer *peer = rt->peer;
1689 fib_info_put(rt->fi);
1699 static void ipv4_link_failure(struct sk_buff *skb)
1703 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1705 rt = skb_rtable(skb);
1706 if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1707 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1710 static int ip_rt_bug(struct sk_buff *skb)
1712 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1713 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1714 skb->dev ? skb->dev->name : "?");
1721 We do not cache source address of outgoing interface,
1722 because it is used only by IP RR, TS and SRR options,
1723 so that it out of fast path.
1725 BTW remember: "addr" is allowed to be not aligned
1729 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1733 if (rt_is_output_route(rt))
1734 src = ip_hdr(skb)->saddr;
1736 struct fib_result res;
1742 memset(&fl4, 0, sizeof(fl4));
1743 fl4.daddr = iph->daddr;
1744 fl4.saddr = iph->saddr;
1745 fl4.flowi4_tos = RT_TOS(iph->tos);
1746 fl4.flowi4_oif = rt->dst.dev->ifindex;
1747 fl4.flowi4_iif = skb->dev->ifindex;
1748 fl4.flowi4_mark = skb->mark;
1751 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1752 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1754 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1758 memcpy(addr, &src, 4);
1761 #ifdef CONFIG_IP_ROUTE_CLASSID
1762 static void set_class_tag(struct rtable *rt, u32 tag)
1764 if (!(rt->dst.tclassid & 0xFFFF))
1765 rt->dst.tclassid |= tag & 0xFFFF;
1766 if (!(rt->dst.tclassid & 0xFFFF0000))
1767 rt->dst.tclassid |= tag & 0xFFFF0000;
1771 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1773 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1776 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1778 if (advmss > 65535 - 40)
1779 advmss = 65535 - 40;
1784 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1786 unsigned int mtu = dst->dev->mtu;
1788 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1789 const struct rtable *rt = (const struct rtable *) dst;
1791 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1795 if (mtu > IP_MAX_MTU)
1801 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1802 struct fib_info *fi)
1804 struct inet_peer *peer;
1807 /* If a peer entry exists for this destination, we must hook
1808 * it up in order to get at cached metrics.
1810 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1813 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1815 rt->rt_peer_genid = rt_peer_genid();
1816 if (inet_metrics_new(peer))
1817 memcpy(peer->metrics, fi->fib_metrics,
1818 sizeof(u32) * RTAX_MAX);
1819 dst_init_metrics(&rt->dst, peer->metrics, false);
1821 check_peer_pmtu(&rt->dst, peer);
1822 if (peer->redirect_learned.a4 &&
1823 peer->redirect_learned.a4 != rt->rt_gateway) {
1824 rt->rt_gateway = peer->redirect_learned.a4;
1825 rt->rt_flags |= RTCF_REDIRECTED;
1828 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1830 atomic_inc(&fi->fib_clntref);
1832 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1836 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1837 const struct fib_result *res,
1838 struct fib_info *fi, u16 type, u32 itag)
1840 struct dst_entry *dst = &rt->dst;
1843 if (FIB_RES_GW(*res) &&
1844 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1845 rt->rt_gateway = FIB_RES_GW(*res);
1846 rt_init_metrics(rt, fl4, fi);
1847 #ifdef CONFIG_IP_ROUTE_CLASSID
1848 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1852 if (dst_mtu(dst) > IP_MAX_MTU)
1853 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1854 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1855 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1857 #ifdef CONFIG_IP_ROUTE_CLASSID
1858 #ifdef CONFIG_IP_MULTIPLE_TABLES
1859 set_class_tag(rt, fib_rules_tclass(res));
1861 set_class_tag(rt, itag);
1865 static struct rtable *rt_dst_alloc(struct net_device *dev,
1866 bool nopolicy, bool noxfrm)
1868 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1870 (nopolicy ? DST_NOPOLICY : 0) |
1871 (noxfrm ? DST_NOXFRM : 0));
1874 /* called in rcu_read_lock() section */
1875 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1876 u8 tos, struct net_device *dev, int our)
1881 struct in_device *in_dev = __in_dev_get_rcu(dev);
1885 /* Primary sanity checks. */
1890 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1891 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1894 if (ipv4_is_zeronet(saddr)) {
1895 if (!ipv4_is_local_multicast(daddr))
1897 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1899 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1904 rth = rt_dst_alloc(init_net.loopback_dev,
1905 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1909 #ifdef CONFIG_IP_ROUTE_CLASSID
1910 rth->dst.tclassid = itag;
1912 rth->dst.output = ip_rt_bug;
1914 rth->rt_key_dst = daddr;
1915 rth->rt_key_src = saddr;
1916 rth->rt_genid = rt_genid(dev_net(dev));
1917 rth->rt_flags = RTCF_MULTICAST;
1918 rth->rt_type = RTN_MULTICAST;
1919 rth->rt_key_tos = tos;
1920 rth->rt_dst = daddr;
1921 rth->rt_src = saddr;
1922 rth->rt_route_iif = dev->ifindex;
1923 rth->rt_iif = dev->ifindex;
1925 rth->rt_mark = skb->mark;
1926 rth->rt_gateway = daddr;
1927 rth->rt_spec_dst= spec_dst;
1928 rth->rt_peer_genid = 0;
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 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1944 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1955 static void ip_handle_martian_source(struct net_device *dev,
1956 struct in_device *in_dev,
1957 struct sk_buff *skb,
1961 RT_CACHE_STAT_INC(in_martian_src);
1962 #ifdef CONFIG_IP_ROUTE_VERBOSE
1963 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1965 * RFC1812 recommendation, if source is martian,
1966 * the only hint is MAC header.
1968 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1969 &daddr, &saddr, dev->name);
1970 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1972 const unsigned char *p = skb_mac_header(skb);
1973 printk(KERN_WARNING "ll header: ");
1974 for (i = 0; i < dev->hard_header_len; i++, p++) {
1976 if (i < (dev->hard_header_len - 1))
1985 /* called in rcu_read_lock() section */
1986 static int __mkroute_input(struct sk_buff *skb,
1987 const struct fib_result *res,
1988 struct in_device *in_dev,
1989 __be32 daddr, __be32 saddr, u32 tos,
1990 struct rtable **result)
1994 struct in_device *out_dev;
1995 unsigned int flags = 0;
1999 /* get a working reference to the output device */
2000 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2001 if (out_dev == NULL) {
2002 if (net_ratelimit())
2003 printk(KERN_CRIT "Bug in ip_route_input" \
2004 "_slow(). Please, report\n");
2009 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2010 in_dev->dev, &spec_dst, &itag);
2012 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2019 flags |= RTCF_DIRECTSRC;
2021 if (out_dev == in_dev && err &&
2022 (IN_DEV_SHARED_MEDIA(out_dev) ||
2023 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2024 flags |= RTCF_DOREDIRECT;
2026 if (skb->protocol != htons(ETH_P_IP)) {
2027 /* Not IP (i.e. ARP). Do not create route, if it is
2028 * invalid for proxy arp. DNAT routes are always valid.
2030 * Proxy arp feature have been extended to allow, ARP
2031 * replies back to the same interface, to support
2032 * Private VLAN switch technologies. See arp.c.
2034 if (out_dev == in_dev &&
2035 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2041 rth = rt_dst_alloc(out_dev->dev,
2042 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2043 IN_DEV_CONF_GET(out_dev, NOXFRM));
2049 rth->rt_key_dst = daddr;
2050 rth->rt_key_src = saddr;
2051 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2052 rth->rt_flags = flags;
2053 rth->rt_type = res->type;
2054 rth->rt_key_tos = tos;
2055 rth->rt_dst = daddr;
2056 rth->rt_src = saddr;
2057 rth->rt_route_iif = in_dev->dev->ifindex;
2058 rth->rt_iif = in_dev->dev->ifindex;
2060 rth->rt_mark = skb->mark;
2061 rth->rt_gateway = daddr;
2062 rth->rt_spec_dst= spec_dst;
2063 rth->rt_peer_genid = 0;
2067 rth->dst.input = ip_forward;
2068 rth->dst.output = ip_output;
2070 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2078 static int ip_mkroute_input(struct sk_buff *skb,
2079 struct fib_result *res,
2080 const struct flowi4 *fl4,
2081 struct in_device *in_dev,
2082 __be32 daddr, __be32 saddr, u32 tos)
2084 struct rtable* rth = NULL;
2088 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2089 if (res->fi && res->fi->fib_nhs > 1)
2090 fib_select_multipath(res);
2093 /* create a routing cache entry */
2094 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2098 /* put it into the cache */
2099 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2100 rt_genid(dev_net(rth->dst.dev)));
2101 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2103 return PTR_ERR(rth);
2108 * NOTE. We drop all the packets that has local source
2109 * addresses, because every properly looped back packet
2110 * must have correct destination already attached by output routine.
2112 * Such approach solves two big problems:
2113 * 1. Not simplex devices are handled properly.
2114 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2115 * called with rcu_read_lock()
2118 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2119 u8 tos, struct net_device *dev)
2121 struct fib_result res;
2122 struct in_device *in_dev = __in_dev_get_rcu(dev);
2126 struct rtable * rth;
2130 struct net * net = dev_net(dev);
2132 /* IP on this device is disabled. */
2137 /* Check for the most weird martians, which can be not detected
2141 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2142 ipv4_is_loopback(saddr))
2143 goto martian_source;
2145 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2148 /* Accept zero addresses only to limited broadcast;
2149 * I even do not know to fix it or not. Waiting for complains :-)
2151 if (ipv4_is_zeronet(saddr))
2152 goto martian_source;
2154 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2155 goto martian_destination;
2158 * Now we are ready to route packet.
2161 fl4.flowi4_iif = dev->ifindex;
2162 fl4.flowi4_mark = skb->mark;
2163 fl4.flowi4_tos = tos;
2164 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2167 err = fib_lookup(net, &fl4, &res);
2169 if (!IN_DEV_FORWARD(in_dev))
2174 RT_CACHE_STAT_INC(in_slow_tot);
2176 if (res.type == RTN_BROADCAST)
2179 if (res.type == RTN_LOCAL) {
2180 err = fib_validate_source(skb, saddr, daddr, tos,
2181 net->loopback_dev->ifindex,
2182 dev, &spec_dst, &itag);
2184 goto martian_source_keep_err;
2186 flags |= RTCF_DIRECTSRC;
2191 if (!IN_DEV_FORWARD(in_dev))
2193 if (res.type != RTN_UNICAST)
2194 goto martian_destination;
2196 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2200 if (skb->protocol != htons(ETH_P_IP))
2203 if (ipv4_is_zeronet(saddr))
2204 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2206 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2209 goto martian_source_keep_err;
2211 flags |= RTCF_DIRECTSRC;
2213 flags |= RTCF_BROADCAST;
2214 res.type = RTN_BROADCAST;
2215 RT_CACHE_STAT_INC(in_brd);
2218 rth = rt_dst_alloc(net->loopback_dev,
2219 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2223 rth->dst.input= ip_local_deliver;
2224 rth->dst.output= ip_rt_bug;
2225 #ifdef CONFIG_IP_ROUTE_CLASSID
2226 rth->dst.tclassid = itag;
2229 rth->rt_key_dst = daddr;
2230 rth->rt_key_src = saddr;
2231 rth->rt_genid = rt_genid(net);
2232 rth->rt_flags = flags|RTCF_LOCAL;
2233 rth->rt_type = res.type;
2234 rth->rt_key_tos = tos;
2235 rth->rt_dst = daddr;
2236 rth->rt_src = saddr;
2237 #ifdef CONFIG_IP_ROUTE_CLASSID
2238 rth->dst.tclassid = itag;
2240 rth->rt_route_iif = dev->ifindex;
2241 rth->rt_iif = dev->ifindex;
2243 rth->rt_mark = skb->mark;
2244 rth->rt_gateway = daddr;
2245 rth->rt_spec_dst= spec_dst;
2246 rth->rt_peer_genid = 0;
2249 if (res.type == RTN_UNREACHABLE) {
2250 rth->dst.input= ip_error;
2251 rth->dst.error= -err;
2252 rth->rt_flags &= ~RTCF_LOCAL;
2254 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2255 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2262 RT_CACHE_STAT_INC(in_no_route);
2263 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2264 res.type = RTN_UNREACHABLE;
2270 * Do not cache martian addresses: they should be logged (RFC1812)
2272 martian_destination:
2273 RT_CACHE_STAT_INC(in_martian_dst);
2274 #ifdef CONFIG_IP_ROUTE_VERBOSE
2275 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2276 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2277 &daddr, &saddr, dev->name);
2281 err = -EHOSTUNREACH;
2294 martian_source_keep_err:
2295 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2299 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2300 u8 tos, struct net_device *dev, bool noref)
2302 struct rtable * rth;
2304 int iif = dev->ifindex;
2312 if (!rt_caching(net))
2315 tos &= IPTOS_RT_MASK;
2316 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2318 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2319 rth = rcu_dereference(rth->dst.rt_next)) {
2320 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2321 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2322 (rth->rt_iif ^ iif) |
2324 (rth->rt_key_tos ^ tos)) == 0 &&
2325 rth->rt_mark == skb->mark &&
2326 net_eq(dev_net(rth->dst.dev), net) &&
2327 !rt_is_expired(rth)) {
2329 dst_use_noref(&rth->dst, jiffies);
2330 skb_dst_set_noref(skb, &rth->dst);
2332 dst_use(&rth->dst, jiffies);
2333 skb_dst_set(skb, &rth->dst);
2335 RT_CACHE_STAT_INC(in_hit);
2339 RT_CACHE_STAT_INC(in_hlist_search);
2343 /* Multicast recognition logic is moved from route cache to here.
2344 The problem was that too many Ethernet cards have broken/missing
2345 hardware multicast filters :-( As result the host on multicasting
2346 network acquires a lot of useless route cache entries, sort of
2347 SDR messages from all the world. Now we try to get rid of them.
2348 Really, provided software IP multicast filter is organized
2349 reasonably (at least, hashed), it does not result in a slowdown
2350 comparing with route cache reject entries.
2351 Note, that multicast routers are not affected, because
2352 route cache entry is created eventually.
2354 if (ipv4_is_multicast(daddr)) {
2355 struct in_device *in_dev = __in_dev_get_rcu(dev);
2358 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2359 ip_hdr(skb)->protocol);
2361 #ifdef CONFIG_IP_MROUTE
2363 (!ipv4_is_local_multicast(daddr) &&
2364 IN_DEV_MFORWARD(in_dev))
2367 int res = ip_route_input_mc(skb, daddr, saddr,
2376 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2380 EXPORT_SYMBOL(ip_route_input_common);
2382 /* called with rcu_read_lock() */
2383 static struct rtable *__mkroute_output(const struct fib_result *res,
2384 const struct flowi4 *fl4,
2385 __be32 orig_daddr, __be32 orig_saddr,
2386 int orig_oif, struct net_device *dev_out,
2389 struct fib_info *fi = res->fi;
2390 u32 tos = RT_FL_TOS(fl4);
2391 struct in_device *in_dev;
2392 u16 type = res->type;
2395 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2396 return ERR_PTR(-EINVAL);
2398 if (ipv4_is_lbcast(fl4->daddr))
2399 type = RTN_BROADCAST;
2400 else if (ipv4_is_multicast(fl4->daddr))
2401 type = RTN_MULTICAST;
2402 else if (ipv4_is_zeronet(fl4->daddr))
2403 return ERR_PTR(-EINVAL);
2405 if (dev_out->flags & IFF_LOOPBACK)
2406 flags |= RTCF_LOCAL;
2408 in_dev = __in_dev_get_rcu(dev_out);
2410 return ERR_PTR(-EINVAL);
2412 if (type == RTN_BROADCAST) {
2413 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2415 } else if (type == RTN_MULTICAST) {
2416 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2417 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2419 flags &= ~RTCF_LOCAL;
2420 /* If multicast route do not exist use
2421 * default one, but do not gateway in this case.
2424 if (fi && res->prefixlen < 4)
2428 rth = rt_dst_alloc(dev_out,
2429 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2430 IN_DEV_CONF_GET(in_dev, NOXFRM));
2432 return ERR_PTR(-ENOBUFS);
2434 rth->dst.output = ip_output;
2436 rth->rt_key_dst = orig_daddr;
2437 rth->rt_key_src = orig_saddr;
2438 rth->rt_genid = rt_genid(dev_net(dev_out));
2439 rth->rt_flags = flags;
2440 rth->rt_type = type;
2441 rth->rt_key_tos = tos;
2442 rth->rt_dst = fl4->daddr;
2443 rth->rt_src = fl4->saddr;
2444 rth->rt_route_iif = 0;
2445 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2446 rth->rt_oif = orig_oif;
2447 rth->rt_mark = fl4->flowi4_mark;
2448 rth->rt_gateway = fl4->daddr;
2449 rth->rt_spec_dst= fl4->saddr;
2450 rth->rt_peer_genid = 0;
2454 RT_CACHE_STAT_INC(out_slow_tot);
2456 if (flags & RTCF_LOCAL) {
2457 rth->dst.input = ip_local_deliver;
2458 rth->rt_spec_dst = fl4->daddr;
2460 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2461 rth->rt_spec_dst = fl4->saddr;
2462 if (flags & RTCF_LOCAL &&
2463 !(dev_out->flags & IFF_LOOPBACK)) {
2464 rth->dst.output = ip_mc_output;
2465 RT_CACHE_STAT_INC(out_slow_mc);
2467 #ifdef CONFIG_IP_MROUTE
2468 if (type == RTN_MULTICAST) {
2469 if (IN_DEV_MFORWARD(in_dev) &&
2470 !ipv4_is_local_multicast(fl4->daddr)) {
2471 rth->dst.input = ip_mr_input;
2472 rth->dst.output = ip_mc_output;
2478 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2484 * Major route resolver routine.
2485 * called with rcu_read_lock();
2488 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2490 struct net_device *dev_out = NULL;
2491 u32 tos = RT_FL_TOS(fl4);
2492 unsigned int flags = 0;
2493 struct fib_result res;
2500 #ifdef CONFIG_IP_MULTIPLE_TABLES
2504 orig_daddr = fl4->daddr;
2505 orig_saddr = fl4->saddr;
2506 orig_oif = fl4->flowi4_oif;
2508 fl4->flowi4_iif = net->loopback_dev->ifindex;
2509 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2510 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2511 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2515 rth = ERR_PTR(-EINVAL);
2516 if (ipv4_is_multicast(fl4->saddr) ||
2517 ipv4_is_lbcast(fl4->saddr) ||
2518 ipv4_is_zeronet(fl4->saddr))
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 (fl4->flowi4_oif == 0 &&
2530 (ipv4_is_multicast(fl4->daddr) ||
2531 ipv4_is_lbcast(fl4->daddr))) {
2532 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2533 dev_out = __ip_dev_find(net, fl4->saddr, 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 fl4->flowi4_oif = dev_out->ifindex;
2556 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2557 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2558 if (!__ip_dev_find(net, fl4->saddr, false))
2564 if (fl4->flowi4_oif) {
2565 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2566 rth = ERR_PTR(-ENODEV);
2567 if (dev_out == NULL)
2570 /* RACE: Check return value of inet_select_addr instead. */
2571 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2572 rth = ERR_PTR(-ENETUNREACH);
2575 if (ipv4_is_local_multicast(fl4->daddr) ||
2576 ipv4_is_lbcast(fl4->daddr)) {
2578 fl4->saddr = inet_select_addr(dev_out, 0,
2583 if (ipv4_is_multicast(fl4->daddr))
2584 fl4->saddr = inet_select_addr(dev_out, 0,
2586 else if (!fl4->daddr)
2587 fl4->saddr = inet_select_addr(dev_out, 0,
2593 fl4->daddr = fl4->saddr;
2595 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2596 dev_out = net->loopback_dev;
2597 fl4->flowi4_oif = net->loopback_dev->ifindex;
2598 res.type = RTN_LOCAL;
2599 flags |= RTCF_LOCAL;
2603 if (fib_lookup(net, fl4, &res)) {
2605 if (fl4->flowi4_oif) {
2606 /* Apparently, routing tables are wrong. Assume,
2607 that the destination is on link.
2610 Because we are allowed to send to iface
2611 even if it has NO routes and NO assigned
2612 addresses. When oif is specified, routing
2613 tables are looked up with only one purpose:
2614 to catch if destination is gatewayed, rather than
2615 direct. Moreover, if MSG_DONTROUTE is set,
2616 we send packet, ignoring both routing tables
2617 and ifaddr state. --ANK
2620 We could make it even if oif is unknown,
2621 likely IPv6, but we do not.
2624 if (fl4->saddr == 0)
2625 fl4->saddr = inet_select_addr(dev_out, 0,
2627 res.type = RTN_UNICAST;
2630 rth = ERR_PTR(-ENETUNREACH);
2634 if (res.type == RTN_LOCAL) {
2636 if (res.fi->fib_prefsrc)
2637 fl4->saddr = res.fi->fib_prefsrc;
2639 fl4->saddr = fl4->daddr;
2641 dev_out = net->loopback_dev;
2642 fl4->flowi4_oif = dev_out->ifindex;
2644 flags |= RTCF_LOCAL;
2648 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2649 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2650 fib_select_multipath(&res);
2653 if (!res.prefixlen &&
2654 res.table->tb_num_default > 1 &&
2655 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2656 fib_select_default(&res);
2659 fl4->saddr = FIB_RES_PREFSRC(net, res);
2661 dev_out = FIB_RES_DEV(res);
2662 fl4->flowi4_oif = dev_out->ifindex;
2666 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2671 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2672 rt_genid(dev_net(dev_out)));
2673 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2681 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2686 if (!rt_caching(net))
2689 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2692 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2693 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2694 if (rth->rt_key_dst == flp4->daddr &&
2695 rth->rt_key_src == flp4->saddr &&
2696 rt_is_output_route(rth) &&
2697 rth->rt_oif == flp4->flowi4_oif &&
2698 rth->rt_mark == flp4->flowi4_mark &&
2699 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2700 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2701 net_eq(dev_net(rth->dst.dev), net) &&
2702 !rt_is_expired(rth)) {
2703 dst_use(&rth->dst, jiffies);
2704 RT_CACHE_STAT_INC(out_hit);
2705 rcu_read_unlock_bh();
2707 flp4->saddr = rth->rt_src;
2709 flp4->daddr = rth->rt_dst;
2712 RT_CACHE_STAT_INC(out_hlist_search);
2714 rcu_read_unlock_bh();
2717 return ip_route_output_slow(net, flp4);
2719 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2721 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2726 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2731 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2735 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2741 static struct dst_ops ipv4_dst_blackhole_ops = {
2743 .protocol = cpu_to_be16(ETH_P_IP),
2744 .destroy = ipv4_dst_destroy,
2745 .check = ipv4_blackhole_dst_check,
2746 .default_mtu = ipv4_blackhole_default_mtu,
2747 .default_advmss = ipv4_default_advmss,
2748 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2749 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2750 .neigh_lookup = ipv4_neigh_lookup,
2753 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2755 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2756 struct rtable *ort = (struct rtable *) dst_orig;
2759 struct dst_entry *new = &rt->dst;
2762 new->input = dst_discard;
2763 new->output = dst_discard;
2764 dst_copy_metrics(new, &ort->dst);
2766 new->dev = ort->dst.dev;
2770 rt->rt_key_dst = ort->rt_key_dst;
2771 rt->rt_key_src = ort->rt_key_src;
2772 rt->rt_key_tos = ort->rt_key_tos;
2773 rt->rt_route_iif = ort->rt_route_iif;
2774 rt->rt_iif = ort->rt_iif;
2775 rt->rt_oif = ort->rt_oif;
2776 rt->rt_mark = ort->rt_mark;
2778 rt->rt_genid = rt_genid(net);
2779 rt->rt_flags = ort->rt_flags;
2780 rt->rt_type = ort->rt_type;
2781 rt->rt_dst = ort->rt_dst;
2782 rt->rt_src = ort->rt_src;
2783 rt->rt_gateway = ort->rt_gateway;
2784 rt->rt_spec_dst = ort->rt_spec_dst;
2785 rt->peer = ort->peer;
2787 atomic_inc(&rt->peer->refcnt);
2790 atomic_inc(&rt->fi->fib_clntref);
2795 dst_release(dst_orig);
2797 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2800 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2803 struct rtable *rt = __ip_route_output_key(net, flp4);
2808 if (flp4->flowi4_proto)
2809 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2810 flowi4_to_flowi(flp4),
2815 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2817 static int rt_fill_info(struct net *net,
2818 struct sk_buff *skb, u32 pid, u32 seq, int event,
2819 int nowait, unsigned int flags)
2821 struct rtable *rt = skb_rtable(skb);
2823 struct nlmsghdr *nlh;
2825 const struct inet_peer *peer = rt->peer;
2826 u32 id = 0, ts = 0, tsage = 0, error;
2828 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2832 r = nlmsg_data(nlh);
2833 r->rtm_family = AF_INET;
2834 r->rtm_dst_len = 32;
2836 r->rtm_tos = rt->rt_key_tos;
2837 r->rtm_table = RT_TABLE_MAIN;
2838 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2839 r->rtm_type = rt->rt_type;
2840 r->rtm_scope = RT_SCOPE_UNIVERSE;
2841 r->rtm_protocol = RTPROT_UNSPEC;
2842 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2843 if (rt->rt_flags & RTCF_NOTIFY)
2844 r->rtm_flags |= RTM_F_NOTIFY;
2846 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2848 if (rt->rt_key_src) {
2849 r->rtm_src_len = 32;
2850 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2853 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2854 #ifdef CONFIG_IP_ROUTE_CLASSID
2855 if (rt->dst.tclassid)
2856 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2858 if (rt_is_input_route(rt))
2859 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2860 else if (rt->rt_src != rt->rt_key_src)
2861 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2863 if (rt->rt_dst != rt->rt_gateway)
2864 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2866 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2867 goto nla_put_failure;
2870 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2872 error = rt->dst.error;
2874 inet_peer_refcheck(rt->peer);
2875 id = atomic_read(&peer->ip_id_count) & 0xffff;
2876 if (peer->tcp_ts_stamp) {
2878 tsage = get_seconds() - peer->tcp_ts_stamp;
2880 expires = ACCESS_ONCE(peer->pmtu_expires);
2885 if (rt_is_input_route(rt)) {
2886 #ifdef CONFIG_IP_MROUTE
2887 __be32 dst = rt->rt_dst;
2889 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2890 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2891 int err = ipmr_get_route(net, skb,
2892 rt->rt_src, rt->rt_dst,
2898 goto nla_put_failure;
2900 if (err == -EMSGSIZE)
2901 goto nla_put_failure;
2907 NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2910 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2911 expires, error) < 0)
2912 goto nla_put_failure;
2914 return nlmsg_end(skb, nlh);
2917 nlmsg_cancel(skb, nlh);
2921 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2923 struct net *net = sock_net(in_skb->sk);
2925 struct nlattr *tb[RTA_MAX+1];
2926 struct rtable *rt = NULL;
2932 struct sk_buff *skb;
2934 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2938 rtm = nlmsg_data(nlh);
2940 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2946 /* Reserve room for dummy headers, this skb can pass
2947 through good chunk of routing engine.
2949 skb_reset_mac_header(skb);
2950 skb_reset_network_header(skb);
2952 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2953 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2954 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2956 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2957 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2958 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2959 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2962 struct net_device *dev;
2964 dev = __dev_get_by_index(net, iif);
2970 skb->protocol = htons(ETH_P_IP);
2974 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2977 rt = skb_rtable(skb);
2978 if (err == 0 && rt->dst.error)
2979 err = -rt->dst.error;
2981 struct flowi4 fl4 = {
2984 .flowi4_tos = rtm->rtm_tos,
2985 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2986 .flowi4_mark = mark,
2988 rt = ip_route_output_key(net, &fl4);
2998 skb_dst_set(skb, &rt->dst);
2999 if (rtm->rtm_flags & RTM_F_NOTIFY)
3000 rt->rt_flags |= RTCF_NOTIFY;
3002 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3003 RTM_NEWROUTE, 0, 0);
3007 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3016 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3023 net = sock_net(skb->sk);
3028 s_idx = idx = cb->args[1];
3029 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3030 if (!rt_hash_table[h].chain)
3033 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3034 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3035 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3037 if (rt_is_expired(rt))
3039 skb_dst_set_noref(skb, &rt->dst);
3040 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3041 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3042 1, NLM_F_MULTI) <= 0) {
3044 rcu_read_unlock_bh();
3049 rcu_read_unlock_bh();
3058 void ip_rt_multicast_event(struct in_device *in_dev)
3060 rt_cache_flush(dev_net(in_dev->dev), 0);
3063 #ifdef CONFIG_SYSCTL
3064 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3065 void __user *buffer,
3066 size_t *lenp, loff_t *ppos)
3073 memcpy(&ctl, __ctl, sizeof(ctl));
3074 ctl.data = &flush_delay;
3075 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3077 net = (struct net *)__ctl->extra1;
3078 rt_cache_flush(net, flush_delay);
3085 static ctl_table ipv4_route_table[] = {
3087 .procname = "gc_thresh",
3088 .data = &ipv4_dst_ops.gc_thresh,
3089 .maxlen = sizeof(int),
3091 .proc_handler = proc_dointvec,
3094 .procname = "max_size",
3095 .data = &ip_rt_max_size,
3096 .maxlen = sizeof(int),
3098 .proc_handler = proc_dointvec,
3101 /* Deprecated. Use gc_min_interval_ms */
3103 .procname = "gc_min_interval",
3104 .data = &ip_rt_gc_min_interval,
3105 .maxlen = sizeof(int),
3107 .proc_handler = proc_dointvec_jiffies,
3110 .procname = "gc_min_interval_ms",
3111 .data = &ip_rt_gc_min_interval,
3112 .maxlen = sizeof(int),
3114 .proc_handler = proc_dointvec_ms_jiffies,
3117 .procname = "gc_timeout",
3118 .data = &ip_rt_gc_timeout,
3119 .maxlen = sizeof(int),
3121 .proc_handler = proc_dointvec_jiffies,
3124 .procname = "gc_interval",
3125 .data = &ip_rt_gc_interval,
3126 .maxlen = sizeof(int),
3128 .proc_handler = proc_dointvec_jiffies,
3131 .procname = "redirect_load",
3132 .data = &ip_rt_redirect_load,
3133 .maxlen = sizeof(int),
3135 .proc_handler = proc_dointvec,
3138 .procname = "redirect_number",
3139 .data = &ip_rt_redirect_number,
3140 .maxlen = sizeof(int),
3142 .proc_handler = proc_dointvec,
3145 .procname = "redirect_silence",
3146 .data = &ip_rt_redirect_silence,
3147 .maxlen = sizeof(int),
3149 .proc_handler = proc_dointvec,
3152 .procname = "error_cost",
3153 .data = &ip_rt_error_cost,
3154 .maxlen = sizeof(int),
3156 .proc_handler = proc_dointvec,
3159 .procname = "error_burst",
3160 .data = &ip_rt_error_burst,
3161 .maxlen = sizeof(int),
3163 .proc_handler = proc_dointvec,
3166 .procname = "gc_elasticity",
3167 .data = &ip_rt_gc_elasticity,
3168 .maxlen = sizeof(int),
3170 .proc_handler = proc_dointvec,
3173 .procname = "mtu_expires",
3174 .data = &ip_rt_mtu_expires,
3175 .maxlen = sizeof(int),
3177 .proc_handler = proc_dointvec_jiffies,
3180 .procname = "min_pmtu",
3181 .data = &ip_rt_min_pmtu,
3182 .maxlen = sizeof(int),
3184 .proc_handler = proc_dointvec,
3187 .procname = "min_adv_mss",
3188 .data = &ip_rt_min_advmss,
3189 .maxlen = sizeof(int),
3191 .proc_handler = proc_dointvec,
3196 static struct ctl_table empty[1];
3198 static struct ctl_table ipv4_skeleton[] =
3200 { .procname = "route",
3201 .mode = 0555, .child = ipv4_route_table},
3202 { .procname = "neigh",
3203 .mode = 0555, .child = empty},
3207 static __net_initdata struct ctl_path ipv4_path[] = {
3208 { .procname = "net", },
3209 { .procname = "ipv4", },
3213 static struct ctl_table ipv4_route_flush_table[] = {
3215 .procname = "flush",
3216 .maxlen = sizeof(int),
3218 .proc_handler = ipv4_sysctl_rtcache_flush,
3223 static __net_initdata struct ctl_path ipv4_route_path[] = {
3224 { .procname = "net", },
3225 { .procname = "ipv4", },
3226 { .procname = "route", },
3230 static __net_init int sysctl_route_net_init(struct net *net)
3232 struct ctl_table *tbl;
3234 tbl = ipv4_route_flush_table;
3235 if (!net_eq(net, &init_net)) {
3236 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3240 tbl[0].extra1 = net;
3242 net->ipv4.route_hdr =
3243 register_net_sysctl_table(net, ipv4_route_path, tbl);
3244 if (net->ipv4.route_hdr == NULL)
3249 if (tbl != ipv4_route_flush_table)
3255 static __net_exit void sysctl_route_net_exit(struct net *net)
3257 struct ctl_table *tbl;
3259 tbl = net->ipv4.route_hdr->ctl_table_arg;
3260 unregister_net_sysctl_table(net->ipv4.route_hdr);
3261 BUG_ON(tbl == ipv4_route_flush_table);
3265 static __net_initdata struct pernet_operations sysctl_route_ops = {
3266 .init = sysctl_route_net_init,
3267 .exit = sysctl_route_net_exit,
3271 static __net_init int rt_genid_init(struct net *net)
3273 get_random_bytes(&net->ipv4.rt_genid,
3274 sizeof(net->ipv4.rt_genid));
3275 get_random_bytes(&net->ipv4.dev_addr_genid,
3276 sizeof(net->ipv4.dev_addr_genid));
3280 static __net_initdata struct pernet_operations rt_genid_ops = {
3281 .init = rt_genid_init,
3285 #ifdef CONFIG_IP_ROUTE_CLASSID
3286 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3287 #endif /* CONFIG_IP_ROUTE_CLASSID */
3289 static __initdata unsigned long rhash_entries;
3290 static int __init set_rhash_entries(char *str)
3294 rhash_entries = simple_strtoul(str, &str, 0);
3297 __setup("rhash_entries=", set_rhash_entries);
3299 int __init ip_rt_init(void)
3303 #ifdef CONFIG_IP_ROUTE_CLASSID
3304 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3306 panic("IP: failed to allocate ip_rt_acct\n");
3309 ipv4_dst_ops.kmem_cachep =
3310 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3311 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3313 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3315 if (dst_entries_init(&ipv4_dst_ops) < 0)
3316 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3318 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3319 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3321 rt_hash_table = (struct rt_hash_bucket *)
3322 alloc_large_system_hash("IP route cache",
3323 sizeof(struct rt_hash_bucket),
3325 (totalram_pages >= 128 * 1024) ?
3330 rhash_entries ? 0 : 512 * 1024);
3331 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3332 rt_hash_lock_init();
3334 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3335 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3340 if (ip_rt_proc_init())
3341 printk(KERN_ERR "Unable to create route proc files\n");
3344 xfrm4_init(ip_rt_max_size);
3346 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3348 #ifdef CONFIG_SYSCTL
3349 register_pernet_subsys(&sysctl_route_ops);
3351 register_pernet_subsys(&rt_genid_ops);
3355 #ifdef CONFIG_SYSCTL
3357 * We really need to sanitize the damn ipv4 init order, then all
3358 * this nonsense will go away.
3360 void __init ip_static_sysctl_init(void)
3362 register_sysctl_paths(ipv4_path, ipv4_skeleton);