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>
94 #include <net/net_namespace.h>
95 #include <net/protocol.h>
97 #include <net/route.h>
98 #include <net/inetpeer.h>
100 #include <net/ip_fib.h>
103 #include <net/icmp.h>
104 #include <net/xfrm.h>
105 #include <net/netevent.h>
106 #include <net/rtnetlink.h>
108 #include <linux/sysctl.h>
111 #define RT_FL_TOS(oldflp) \
112 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114 #define IP_MAX_MTU 0xFFF0
116 #define RT_GC_TIMEOUT (300*HZ)
118 static int ip_rt_max_size;
119 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
120 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
121 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
122 static int ip_rt_redirect_number __read_mostly = 9;
123 static int ip_rt_redirect_load __read_mostly = HZ / 50;
124 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
125 static int ip_rt_error_cost __read_mostly = HZ;
126 static int ip_rt_error_burst __read_mostly = 5 * HZ;
127 static int ip_rt_gc_elasticity __read_mostly = 8;
128 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
129 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly = 256;
131 static int ip_rt_secret_interval __read_mostly = 10 * 60 * HZ;
132 static int rt_chain_length_max __read_mostly = 20;
134 static struct delayed_work expires_work;
135 static unsigned long expires_ljiffies;
138 * Interface to generic destination cache.
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static void ipv4_dst_destroy(struct dst_entry *dst);
143 static void ipv4_dst_ifdown(struct dst_entry *dst,
144 struct net_device *dev, int how);
145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146 static void ipv4_link_failure(struct sk_buff *skb);
147 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
148 static int rt_garbage_collect(struct dst_ops *ops);
151 static struct dst_ops ipv4_dst_ops = {
153 .protocol = cpu_to_be16(ETH_P_IP),
154 .gc = rt_garbage_collect,
155 .check = ipv4_dst_check,
156 .destroy = ipv4_dst_destroy,
157 .ifdown = ipv4_dst_ifdown,
158 .negative_advice = ipv4_negative_advice,
159 .link_failure = ipv4_link_failure,
160 .update_pmtu = ip_rt_update_pmtu,
161 .local_out = __ip_local_out,
162 .entries = ATOMIC_INIT(0),
165 #define ECN_OR_COST(class) TC_PRIO_##class
167 const __u8 ip_tos2prio[16] = {
171 ECN_OR_COST(BESTEFFORT),
177 ECN_OR_COST(INTERACTIVE),
179 ECN_OR_COST(INTERACTIVE),
180 TC_PRIO_INTERACTIVE_BULK,
181 ECN_OR_COST(INTERACTIVE_BULK),
182 TC_PRIO_INTERACTIVE_BULK,
183 ECN_OR_COST(INTERACTIVE_BULK)
191 /* The locking scheme is rather straight forward:
193 * 1) Read-Copy Update protects the buckets of the central route hash.
194 * 2) Only writers remove entries, and they hold the lock
195 * as they look at rtable reference counts.
196 * 3) Only readers acquire references to rtable entries,
197 * they do so with atomic increments and with the
201 struct rt_hash_bucket {
202 struct rtable *chain;
205 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
206 defined(CONFIG_PROVE_LOCKING)
208 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
209 * The size of this table is a power of two and depends on the number of CPUS.
210 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
212 #ifdef CONFIG_LOCKDEP
213 # define RT_HASH_LOCK_SZ 256
216 # define RT_HASH_LOCK_SZ 4096
218 # define RT_HASH_LOCK_SZ 2048
220 # define RT_HASH_LOCK_SZ 1024
222 # define RT_HASH_LOCK_SZ 512
224 # define RT_HASH_LOCK_SZ 256
228 static spinlock_t *rt_hash_locks;
229 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
231 static __init void rt_hash_lock_init(void)
235 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
238 panic("IP: failed to allocate rt_hash_locks\n");
240 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
241 spin_lock_init(&rt_hash_locks[i]);
244 # define rt_hash_lock_addr(slot) NULL
246 static inline void rt_hash_lock_init(void)
251 static struct rt_hash_bucket *rt_hash_table __read_mostly;
252 static unsigned rt_hash_mask __read_mostly;
253 static unsigned int rt_hash_log __read_mostly;
255 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
256 #define RT_CACHE_STAT_INC(field) \
257 (__raw_get_cpu_var(rt_cache_stat).field++)
259 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
262 return jhash_3words((__force u32)(__be32)(daddr),
263 (__force u32)(__be32)(saddr),
268 static inline int rt_genid(struct net *net)
270 return atomic_read(&net->ipv4.rt_genid);
273 #ifdef CONFIG_PROC_FS
274 struct rt_cache_iter_state {
275 struct seq_net_private p;
280 static struct rtable *rt_cache_get_first(struct seq_file *seq)
282 struct rt_cache_iter_state *st = seq->private;
283 struct rtable *r = NULL;
285 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
286 if (!rt_hash_table[st->bucket].chain)
289 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
291 if (dev_net(r->u.dst.dev) == seq_file_net(seq) &&
292 r->rt_genid == st->genid)
294 r = rcu_dereference_bh(r->u.dst.rt_next);
296 rcu_read_unlock_bh();
301 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
304 struct rt_cache_iter_state *st = seq->private;
306 r = r->u.dst.rt_next;
308 rcu_read_unlock_bh();
310 if (--st->bucket < 0)
312 } while (!rt_hash_table[st->bucket].chain);
314 r = rt_hash_table[st->bucket].chain;
316 return rcu_dereference_bh(r);
319 static struct rtable *rt_cache_get_next(struct seq_file *seq,
322 struct rt_cache_iter_state *st = seq->private;
323 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
324 if (dev_net(r->u.dst.dev) != seq_file_net(seq))
326 if (r->rt_genid == st->genid)
332 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
334 struct rtable *r = rt_cache_get_first(seq);
337 while (pos && (r = rt_cache_get_next(seq, r)))
339 return pos ? NULL : r;
342 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
344 struct rt_cache_iter_state *st = seq->private;
346 return rt_cache_get_idx(seq, *pos - 1);
347 st->genid = rt_genid(seq_file_net(seq));
348 return SEQ_START_TOKEN;
351 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
355 if (v == SEQ_START_TOKEN)
356 r = rt_cache_get_first(seq);
358 r = rt_cache_get_next(seq, v);
363 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
365 if (v && v != SEQ_START_TOKEN)
366 rcu_read_unlock_bh();
369 static int rt_cache_seq_show(struct seq_file *seq, void *v)
371 if (v == SEQ_START_TOKEN)
372 seq_printf(seq, "%-127s\n",
373 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
374 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
377 struct rtable *r = v;
380 seq_printf(seq, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
381 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
382 r->u.dst.dev ? r->u.dst.dev->name : "*",
383 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
384 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
385 r->u.dst.__use, 0, (unsigned long)r->rt_src,
386 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
387 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
388 dst_metric(&r->u.dst, RTAX_WINDOW),
389 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
390 dst_metric(&r->u.dst, RTAX_RTTVAR)),
392 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
393 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
395 r->rt_spec_dst, &len);
397 seq_printf(seq, "%*s\n", 127 - len, "");
402 static const struct seq_operations rt_cache_seq_ops = {
403 .start = rt_cache_seq_start,
404 .next = rt_cache_seq_next,
405 .stop = rt_cache_seq_stop,
406 .show = rt_cache_seq_show,
409 static int rt_cache_seq_open(struct inode *inode, struct file *file)
411 return seq_open_net(inode, file, &rt_cache_seq_ops,
412 sizeof(struct rt_cache_iter_state));
415 static const struct file_operations rt_cache_seq_fops = {
416 .owner = THIS_MODULE,
417 .open = rt_cache_seq_open,
420 .release = seq_release_net,
424 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
429 return SEQ_START_TOKEN;
431 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
432 if (!cpu_possible(cpu))
435 return &per_cpu(rt_cache_stat, cpu);
440 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
444 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
445 if (!cpu_possible(cpu))
448 return &per_cpu(rt_cache_stat, cpu);
454 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
459 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
461 struct rt_cache_stat *st = v;
463 if (v == SEQ_START_TOKEN) {
464 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");
468 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
469 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
470 atomic_read(&ipv4_dst_ops.entries),
493 static const struct seq_operations rt_cpu_seq_ops = {
494 .start = rt_cpu_seq_start,
495 .next = rt_cpu_seq_next,
496 .stop = rt_cpu_seq_stop,
497 .show = rt_cpu_seq_show,
501 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
503 return seq_open(file, &rt_cpu_seq_ops);
506 static const struct file_operations rt_cpu_seq_fops = {
507 .owner = THIS_MODULE,
508 .open = rt_cpu_seq_open,
511 .release = seq_release,
514 #ifdef CONFIG_NET_CLS_ROUTE
515 static int rt_acct_proc_show(struct seq_file *m, void *v)
517 struct ip_rt_acct *dst, *src;
520 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
524 for_each_possible_cpu(i) {
525 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
526 for (j = 0; j < 256; j++) {
527 dst[j].o_bytes += src[j].o_bytes;
528 dst[j].o_packets += src[j].o_packets;
529 dst[j].i_bytes += src[j].i_bytes;
530 dst[j].i_packets += src[j].i_packets;
534 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
539 static int rt_acct_proc_open(struct inode *inode, struct file *file)
541 return single_open(file, rt_acct_proc_show, NULL);
544 static const struct file_operations rt_acct_proc_fops = {
545 .owner = THIS_MODULE,
546 .open = rt_acct_proc_open,
549 .release = single_release,
553 static int __net_init ip_rt_do_proc_init(struct net *net)
555 struct proc_dir_entry *pde;
557 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
562 pde = proc_create("rt_cache", S_IRUGO,
563 net->proc_net_stat, &rt_cpu_seq_fops);
567 #ifdef CONFIG_NET_CLS_ROUTE
568 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
574 #ifdef CONFIG_NET_CLS_ROUTE
576 remove_proc_entry("rt_cache", net->proc_net_stat);
579 remove_proc_entry("rt_cache", net->proc_net);
584 static void __net_exit ip_rt_do_proc_exit(struct net *net)
586 remove_proc_entry("rt_cache", net->proc_net_stat);
587 remove_proc_entry("rt_cache", net->proc_net);
588 #ifdef CONFIG_NET_CLS_ROUTE
589 remove_proc_entry("rt_acct", net->proc_net);
593 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
594 .init = ip_rt_do_proc_init,
595 .exit = ip_rt_do_proc_exit,
598 static int __init ip_rt_proc_init(void)
600 return register_pernet_subsys(&ip_rt_proc_ops);
604 static inline int ip_rt_proc_init(void)
608 #endif /* CONFIG_PROC_FS */
610 static inline void rt_free(struct rtable *rt)
612 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
615 static inline void rt_drop(struct rtable *rt)
618 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
621 static inline int rt_fast_clean(struct rtable *rth)
623 /* Kill broadcast/multicast entries very aggresively, if they
624 collide in hash table with more useful entries */
625 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
626 rth->fl.iif && rth->u.dst.rt_next;
629 static inline int rt_valuable(struct rtable *rth)
631 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
635 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
640 if (atomic_read(&rth->u.dst.__refcnt))
644 if (rth->u.dst.expires &&
645 time_after_eq(jiffies, rth->u.dst.expires))
648 age = jiffies - rth->u.dst.lastuse;
650 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
651 (age <= tmo2 && rt_valuable(rth)))
657 /* Bits of score are:
659 * 30: not quite useless
660 * 29..0: usage counter
662 static inline u32 rt_score(struct rtable *rt)
664 u32 score = jiffies - rt->u.dst.lastuse;
666 score = ~score & ~(3<<30);
672 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
678 static inline bool rt_caching(const struct net *net)
680 return net->ipv4.current_rt_cache_rebuild_count <=
681 net->ipv4.sysctl_rt_cache_rebuild_count;
684 static inline bool compare_hash_inputs(const struct flowi *fl1,
685 const struct flowi *fl2)
687 return (__force u32)(((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
688 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr) |
689 (fl1->iif ^ fl2->iif)) == 0);
692 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
694 return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
695 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
696 (fl1->mark ^ fl2->mark) |
697 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
698 *(u16 *)&fl2->nl_u.ip4_u.tos) |
699 (fl1->oif ^ fl2->oif) |
700 (fl1->iif ^ fl2->iif)) == 0;
703 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
705 return net_eq(dev_net(rt1->u.dst.dev), dev_net(rt2->u.dst.dev));
708 static inline int rt_is_expired(struct rtable *rth)
710 return rth->rt_genid != rt_genid(dev_net(rth->u.dst.dev));
714 * Perform a full scan of hash table and free all entries.
715 * Can be called by a softirq or a process.
716 * In the later case, we want to be reschedule if necessary
718 static void rt_do_flush(int process_context)
721 struct rtable *rth, *next;
722 struct rtable * tail;
724 for (i = 0; i <= rt_hash_mask; i++) {
725 if (process_context && need_resched())
727 rth = rt_hash_table[i].chain;
731 spin_lock_bh(rt_hash_lock_addr(i));
734 struct rtable ** prev, * p;
736 rth = rt_hash_table[i].chain;
738 /* defer releasing the head of the list after spin_unlock */
739 for (tail = rth; tail; tail = tail->u.dst.rt_next)
740 if (!rt_is_expired(tail))
743 rt_hash_table[i].chain = tail;
745 /* call rt_free on entries after the tail requiring flush */
746 prev = &rt_hash_table[i].chain;
747 for (p = *prev; p; p = next) {
748 next = p->u.dst.rt_next;
749 if (!rt_is_expired(p)) {
750 prev = &p->u.dst.rt_next;
758 rth = rt_hash_table[i].chain;
759 rt_hash_table[i].chain = NULL;
762 spin_unlock_bh(rt_hash_lock_addr(i));
764 for (; rth != tail; rth = next) {
765 next = rth->u.dst.rt_next;
772 * While freeing expired entries, we compute average chain length
773 * and standard deviation, using fixed-point arithmetic.
774 * This to have an estimation of rt_chain_length_max
775 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
776 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
780 #define ONE (1UL << FRACT_BITS)
783 * Given a hash chain and an item in this hash chain,
784 * find if a previous entry has the same hash_inputs
785 * (but differs on tos, mark or oif)
786 * Returns 0 if an alias is found.
787 * Returns ONE if rth has no alias before itself.
789 static int has_noalias(const struct rtable *head, const struct rtable *rth)
791 const struct rtable *aux = head;
794 if (compare_hash_inputs(&aux->fl, &rth->fl))
796 aux = aux->u.dst.rt_next;
801 static void rt_check_expire(void)
803 static unsigned int rover;
804 unsigned int i = rover, goal;
805 struct rtable *rth, **rthp;
806 unsigned long samples = 0;
807 unsigned long sum = 0, sum2 = 0;
811 delta = jiffies - expires_ljiffies;
812 expires_ljiffies = jiffies;
813 mult = ((u64)delta) << rt_hash_log;
814 if (ip_rt_gc_timeout > 1)
815 do_div(mult, ip_rt_gc_timeout);
816 goal = (unsigned int)mult;
817 if (goal > rt_hash_mask)
818 goal = rt_hash_mask + 1;
819 for (; goal > 0; goal--) {
820 unsigned long tmo = ip_rt_gc_timeout;
821 unsigned long length;
823 i = (i + 1) & rt_hash_mask;
824 rthp = &rt_hash_table[i].chain;
834 spin_lock_bh(rt_hash_lock_addr(i));
835 while ((rth = *rthp) != NULL) {
836 prefetch(rth->u.dst.rt_next);
837 if (rt_is_expired(rth)) {
838 *rthp = rth->u.dst.rt_next;
842 if (rth->u.dst.expires) {
843 /* Entry is expired even if it is in use */
844 if (time_before_eq(jiffies, rth->u.dst.expires)) {
847 rthp = &rth->u.dst.rt_next;
849 * We only count entries on
850 * a chain with equal hash inputs once
851 * so that entries for different QOS
852 * levels, and other non-hash input
853 * attributes don't unfairly skew
854 * the length computation
856 length += has_noalias(rt_hash_table[i].chain, rth);
859 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
862 /* Cleanup aged off entries. */
863 *rthp = rth->u.dst.rt_next;
866 spin_unlock_bh(rt_hash_lock_addr(i));
868 sum2 += length*length;
871 unsigned long avg = sum / samples;
872 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
873 rt_chain_length_max = max_t(unsigned long,
875 (avg + 4*sd) >> FRACT_BITS);
881 * rt_worker_func() is run in process context.
882 * we call rt_check_expire() to scan part of the hash table
884 static void rt_worker_func(struct work_struct *work)
887 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
891 * Pertubation of rt_genid by a small quantity [1..256]
892 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
893 * many times (2^24) without giving recent rt_genid.
894 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
896 static void rt_cache_invalidate(struct net *net)
898 unsigned char shuffle;
900 get_random_bytes(&shuffle, sizeof(shuffle));
901 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
905 * delay < 0 : invalidate cache (fast : entries will be deleted later)
906 * delay >= 0 : invalidate & flush cache (can be long)
908 void rt_cache_flush(struct net *net, int delay)
910 rt_cache_invalidate(net);
912 rt_do_flush(!in_softirq());
915 /* Flush previous cache invalidated entries from the cache */
916 void rt_cache_flush_batch(void)
918 rt_do_flush(!in_softirq());
922 * We change rt_genid and let gc do the cleanup
924 static void rt_secret_rebuild(unsigned long __net)
926 struct net *net = (struct net *)__net;
927 rt_cache_invalidate(net);
928 mod_timer(&net->ipv4.rt_secret_timer, jiffies + ip_rt_secret_interval);
931 static void rt_secret_rebuild_oneshot(struct net *net)
933 del_timer_sync(&net->ipv4.rt_secret_timer);
934 rt_cache_invalidate(net);
935 if (ip_rt_secret_interval)
936 mod_timer(&net->ipv4.rt_secret_timer, jiffies + ip_rt_secret_interval);
939 static void rt_emergency_hash_rebuild(struct net *net)
941 if (net_ratelimit()) {
942 printk(KERN_WARNING "Route hash chain too long!\n");
943 printk(KERN_WARNING "Adjust your secret_interval!\n");
946 rt_secret_rebuild_oneshot(net);
950 Short description of GC goals.
952 We want to build algorithm, which will keep routing cache
953 at some equilibrium point, when number of aged off entries
954 is kept approximately equal to newly generated ones.
956 Current expiration strength is variable "expire".
957 We try to adjust it dynamically, so that if networking
958 is idle expires is large enough to keep enough of warm entries,
959 and when load increases it reduces to limit cache size.
962 static int rt_garbage_collect(struct dst_ops *ops)
964 static unsigned long expire = RT_GC_TIMEOUT;
965 static unsigned long last_gc;
967 static int equilibrium;
968 struct rtable *rth, **rthp;
969 unsigned long now = jiffies;
973 * Garbage collection is pretty expensive,
974 * do not make it too frequently.
977 RT_CACHE_STAT_INC(gc_total);
979 if (now - last_gc < ip_rt_gc_min_interval &&
980 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
981 RT_CACHE_STAT_INC(gc_ignored);
985 /* Calculate number of entries, which we want to expire now. */
986 goal = atomic_read(&ipv4_dst_ops.entries) -
987 (ip_rt_gc_elasticity << rt_hash_log);
989 if (equilibrium < ipv4_dst_ops.gc_thresh)
990 equilibrium = ipv4_dst_ops.gc_thresh;
991 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
993 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
994 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
997 /* We are in dangerous area. Try to reduce cache really
1000 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1001 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
1004 if (now - last_gc >= ip_rt_gc_min_interval)
1008 equilibrium += goal;
1015 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1016 unsigned long tmo = expire;
1018 k = (k + 1) & rt_hash_mask;
1019 rthp = &rt_hash_table[k].chain;
1020 spin_lock_bh(rt_hash_lock_addr(k));
1021 while ((rth = *rthp) != NULL) {
1022 if (!rt_is_expired(rth) &&
1023 !rt_may_expire(rth, tmo, expire)) {
1025 rthp = &rth->u.dst.rt_next;
1028 *rthp = rth->u.dst.rt_next;
1032 spin_unlock_bh(rt_hash_lock_addr(k));
1041 /* Goal is not achieved. We stop process if:
1043 - if expire reduced to zero. Otherwise, expire is halfed.
1044 - if table is not full.
1045 - if we are called from interrupt.
1046 - jiffies check is just fallback/debug loop breaker.
1047 We will not spin here for long time in any case.
1050 RT_CACHE_STAT_INC(gc_goal_miss);
1056 #if RT_CACHE_DEBUG >= 2
1057 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1058 atomic_read(&ipv4_dst_ops.entries), goal, i);
1061 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1063 } while (!in_softirq() && time_before_eq(jiffies, now));
1065 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1067 if (net_ratelimit())
1068 printk(KERN_WARNING "dst cache overflow\n");
1069 RT_CACHE_STAT_INC(gc_dst_overflow);
1073 expire += ip_rt_gc_min_interval;
1074 if (expire > ip_rt_gc_timeout ||
1075 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
1076 expire = ip_rt_gc_timeout;
1077 #if RT_CACHE_DEBUG >= 2
1078 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1079 atomic_read(&ipv4_dst_ops.entries), goal, rover);
1085 * Returns number of entries in a hash chain that have different hash_inputs
1087 static int slow_chain_length(const struct rtable *head)
1090 const struct rtable *rth = head;
1093 length += has_noalias(head, rth);
1094 rth = rth->u.dst.rt_next;
1096 return length >> FRACT_BITS;
1099 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1100 struct rtable **rp, struct sk_buff *skb)
1102 struct rtable *rth, **rthp;
1104 struct rtable *cand, **candp;
1107 int attempts = !in_softirq();
1111 min_score = ~(u32)0;
1116 if (!rt_caching(dev_net(rt->u.dst.dev))) {
1118 * If we're not caching, just tell the caller we
1119 * were successful and don't touch the route. The
1120 * caller hold the sole reference to the cache entry, and
1121 * it will be released when the caller is done with it.
1122 * If we drop it here, the callers have no way to resolve routes
1123 * when we're not caching. Instead, just point *rp at rt, so
1124 * the caller gets a single use out of the route
1125 * Note that we do rt_free on this new route entry, so that
1126 * once its refcount hits zero, we are still able to reap it
1128 * Note also the rt_free uses call_rcu. We don't actually
1129 * need rcu protection here, this is just our path to get
1130 * on the route gc list.
1133 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1134 int err = arp_bind_neighbour(&rt->u.dst);
1136 if (net_ratelimit())
1138 "Neighbour table failure & not caching routes.\n");
1148 rthp = &rt_hash_table[hash].chain;
1150 spin_lock_bh(rt_hash_lock_addr(hash));
1151 while ((rth = *rthp) != NULL) {
1152 if (rt_is_expired(rth)) {
1153 *rthp = rth->u.dst.rt_next;
1157 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1159 *rthp = rth->u.dst.rt_next;
1161 * Since lookup is lockfree, the deletion
1162 * must be visible to another weakly ordered CPU before
1163 * the insertion at the start of the hash chain.
1165 rcu_assign_pointer(rth->u.dst.rt_next,
1166 rt_hash_table[hash].chain);
1168 * Since lookup is lockfree, the update writes
1169 * must be ordered for consistency on SMP.
1171 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1173 dst_use(&rth->u.dst, now);
1174 spin_unlock_bh(rt_hash_lock_addr(hash));
1180 skb_dst_set(skb, &rth->u.dst);
1184 if (!atomic_read(&rth->u.dst.__refcnt)) {
1185 u32 score = rt_score(rth);
1187 if (score <= min_score) {
1196 rthp = &rth->u.dst.rt_next;
1200 /* ip_rt_gc_elasticity used to be average length of chain
1201 * length, when exceeded gc becomes really aggressive.
1203 * The second limit is less certain. At the moment it allows
1204 * only 2 entries per bucket. We will see.
1206 if (chain_length > ip_rt_gc_elasticity) {
1207 *candp = cand->u.dst.rt_next;
1211 if (chain_length > rt_chain_length_max &&
1212 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1213 struct net *net = dev_net(rt->u.dst.dev);
1214 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1215 if (!rt_caching(dev_net(rt->u.dst.dev))) {
1216 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1217 rt->u.dst.dev->name, num);
1219 rt_emergency_hash_rebuild(dev_net(rt->u.dst.dev));
1223 /* Try to bind route to arp only if it is output
1224 route or unicast forwarding path.
1226 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1227 int err = arp_bind_neighbour(&rt->u.dst);
1229 spin_unlock_bh(rt_hash_lock_addr(hash));
1231 if (err != -ENOBUFS) {
1236 /* Neighbour tables are full and nothing
1237 can be released. Try to shrink route cache,
1238 it is most likely it holds some neighbour records.
1240 if (attempts-- > 0) {
1241 int saved_elasticity = ip_rt_gc_elasticity;
1242 int saved_int = ip_rt_gc_min_interval;
1243 ip_rt_gc_elasticity = 1;
1244 ip_rt_gc_min_interval = 0;
1245 rt_garbage_collect(&ipv4_dst_ops);
1246 ip_rt_gc_min_interval = saved_int;
1247 ip_rt_gc_elasticity = saved_elasticity;
1251 if (net_ratelimit())
1252 printk(KERN_WARNING "Neighbour table overflow.\n");
1258 rt->u.dst.rt_next = rt_hash_table[hash].chain;
1260 #if RT_CACHE_DEBUG >= 2
1261 if (rt->u.dst.rt_next) {
1263 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1265 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1266 printk(" . %pI4", &trt->rt_dst);
1271 * Since lookup is lockfree, we must make sure
1272 * previous writes to rt are comitted to memory
1273 * before making rt visible to other CPUS.
1275 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1277 spin_unlock_bh(rt_hash_lock_addr(hash));
1283 skb_dst_set(skb, &rt->u.dst);
1287 void rt_bind_peer(struct rtable *rt, int create)
1289 static DEFINE_SPINLOCK(rt_peer_lock);
1290 struct inet_peer *peer;
1292 peer = inet_getpeer(rt->rt_dst, create);
1294 spin_lock_bh(&rt_peer_lock);
1295 if (rt->peer == NULL) {
1299 spin_unlock_bh(&rt_peer_lock);
1305 * Peer allocation may fail only in serious out-of-memory conditions. However
1306 * we still can generate some output.
1307 * Random ID selection looks a bit dangerous because we have no chances to
1308 * select ID being unique in a reasonable period of time.
1309 * But broken packet identifier may be better than no packet at all.
1311 static void ip_select_fb_ident(struct iphdr *iph)
1313 static DEFINE_SPINLOCK(ip_fb_id_lock);
1314 static u32 ip_fallback_id;
1317 spin_lock_bh(&ip_fb_id_lock);
1318 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1319 iph->id = htons(salt & 0xFFFF);
1320 ip_fallback_id = salt;
1321 spin_unlock_bh(&ip_fb_id_lock);
1324 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1326 struct rtable *rt = (struct rtable *) dst;
1329 if (rt->peer == NULL)
1330 rt_bind_peer(rt, 1);
1332 /* If peer is attached to destination, it is never detached,
1333 so that we need not to grab a lock to dereference it.
1336 iph->id = htons(inet_getid(rt->peer, more));
1340 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1341 __builtin_return_address(0));
1343 ip_select_fb_ident(iph);
1346 static void rt_del(unsigned hash, struct rtable *rt)
1348 struct rtable **rthp, *aux;
1350 rthp = &rt_hash_table[hash].chain;
1351 spin_lock_bh(rt_hash_lock_addr(hash));
1353 while ((aux = *rthp) != NULL) {
1354 if (aux == rt || rt_is_expired(aux)) {
1355 *rthp = aux->u.dst.rt_next;
1359 rthp = &aux->u.dst.rt_next;
1361 spin_unlock_bh(rt_hash_lock_addr(hash));
1364 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1365 __be32 saddr, struct net_device *dev)
1368 struct in_device *in_dev = in_dev_get(dev);
1369 struct rtable *rth, **rthp;
1370 __be32 skeys[2] = { saddr, 0 };
1371 int ikeys[2] = { dev->ifindex, 0 };
1372 struct netevent_redirect netevent;
1379 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1380 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1381 ipv4_is_zeronet(new_gw))
1382 goto reject_redirect;
1384 if (!rt_caching(net))
1385 goto reject_redirect;
1387 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1388 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1389 goto reject_redirect;
1390 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1391 goto reject_redirect;
1393 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1394 goto reject_redirect;
1397 for (i = 0; i < 2; i++) {
1398 for (k = 0; k < 2; k++) {
1399 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1402 rthp=&rt_hash_table[hash].chain;
1405 while ((rth = rcu_dereference(*rthp)) != NULL) {
1408 if (rth->fl.fl4_dst != daddr ||
1409 rth->fl.fl4_src != skeys[i] ||
1410 rth->fl.oif != ikeys[k] ||
1412 rt_is_expired(rth) ||
1413 !net_eq(dev_net(rth->u.dst.dev), net)) {
1414 rthp = &rth->u.dst.rt_next;
1418 if (rth->rt_dst != daddr ||
1419 rth->rt_src != saddr ||
1421 rth->rt_gateway != old_gw ||
1422 rth->u.dst.dev != dev)
1425 dst_hold(&rth->u.dst);
1428 rt = dst_alloc(&ipv4_dst_ops);
1435 /* Copy all the information. */
1437 rt->u.dst.__use = 1;
1438 atomic_set(&rt->u.dst.__refcnt, 1);
1439 rt->u.dst.child = NULL;
1441 dev_hold(rt->u.dst.dev);
1443 in_dev_hold(rt->idev);
1444 rt->u.dst.obsolete = 0;
1445 rt->u.dst.lastuse = jiffies;
1446 rt->u.dst.path = &rt->u.dst;
1447 rt->u.dst.neighbour = NULL;
1448 rt->u.dst.hh = NULL;
1450 rt->u.dst.xfrm = NULL;
1452 rt->rt_genid = rt_genid(net);
1453 rt->rt_flags |= RTCF_REDIRECTED;
1455 /* Gateway is different ... */
1456 rt->rt_gateway = new_gw;
1458 /* Redirect received -> path was valid */
1459 dst_confirm(&rth->u.dst);
1462 atomic_inc(&rt->peer->refcnt);
1464 if (arp_bind_neighbour(&rt->u.dst) ||
1465 !(rt->u.dst.neighbour->nud_state &
1467 if (rt->u.dst.neighbour)
1468 neigh_event_send(rt->u.dst.neighbour, NULL);
1474 netevent.old = &rth->u.dst;
1475 netevent.new = &rt->u.dst;
1476 call_netevent_notifiers(NETEVENT_REDIRECT,
1480 if (!rt_intern_hash(hash, rt, &rt, NULL))
1493 #ifdef CONFIG_IP_ROUTE_VERBOSE
1494 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1495 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1496 " Advised path = %pI4 -> %pI4\n",
1497 &old_gw, dev->name, &new_gw,
1503 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1505 struct rtable *rt = (struct rtable *)dst;
1506 struct dst_entry *ret = dst;
1509 if (dst->obsolete) {
1512 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1513 rt->u.dst.expires) {
1514 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1516 rt_genid(dev_net(dst->dev)));
1517 #if RT_CACHE_DEBUG >= 1
1518 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1519 &rt->rt_dst, rt->fl.fl4_tos);
1530 * 1. The first ip_rt_redirect_number redirects are sent
1531 * with exponential backoff, then we stop sending them at all,
1532 * assuming that the host ignores our redirects.
1533 * 2. If we did not see packets requiring redirects
1534 * during ip_rt_redirect_silence, we assume that the host
1535 * forgot redirected route and start to send redirects again.
1537 * This algorithm is much cheaper and more intelligent than dumb load limiting
1540 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1541 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1544 void ip_rt_send_redirect(struct sk_buff *skb)
1546 struct rtable *rt = skb_rtable(skb);
1547 struct in_device *in_dev;
1551 in_dev = __in_dev_get_rcu(rt->u.dst.dev);
1552 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1556 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1559 /* No redirected packets during ip_rt_redirect_silence;
1560 * reset the algorithm.
1562 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1563 rt->u.dst.rate_tokens = 0;
1565 /* Too many ignored redirects; do not send anything
1566 * set u.dst.rate_last to the last seen redirected packet.
1568 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1569 rt->u.dst.rate_last = jiffies;
1573 /* Check for load limit; set rate_last to the latest sent
1576 if (rt->u.dst.rate_tokens == 0 ||
1578 (rt->u.dst.rate_last +
1579 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1580 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1581 rt->u.dst.rate_last = jiffies;
1582 ++rt->u.dst.rate_tokens;
1583 #ifdef CONFIG_IP_ROUTE_VERBOSE
1585 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1587 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1588 &rt->rt_src, rt->rt_iif,
1589 &rt->rt_dst, &rt->rt_gateway);
1594 static int ip_error(struct sk_buff *skb)
1596 struct rtable *rt = skb_rtable(skb);
1600 switch (rt->u.dst.error) {
1605 code = ICMP_HOST_UNREACH;
1608 code = ICMP_NET_UNREACH;
1609 IP_INC_STATS_BH(dev_net(rt->u.dst.dev),
1610 IPSTATS_MIB_INNOROUTES);
1613 code = ICMP_PKT_FILTERED;
1618 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1619 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1620 rt->u.dst.rate_tokens = ip_rt_error_burst;
1621 rt->u.dst.rate_last = now;
1622 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1623 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1624 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1627 out: kfree_skb(skb);
1632 * The last two values are not from the RFC but
1633 * are needed for AMPRnet AX.25 paths.
1636 static const unsigned short mtu_plateau[] =
1637 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1639 static inline unsigned short guess_mtu(unsigned short old_mtu)
1643 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1644 if (old_mtu > mtu_plateau[i])
1645 return mtu_plateau[i];
1649 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1650 unsigned short new_mtu,
1651 struct net_device *dev)
1654 unsigned short old_mtu = ntohs(iph->tot_len);
1656 int ikeys[2] = { dev->ifindex, 0 };
1657 __be32 skeys[2] = { iph->saddr, 0, };
1658 __be32 daddr = iph->daddr;
1659 unsigned short est_mtu = 0;
1661 for (k = 0; k < 2; k++) {
1662 for (i = 0; i < 2; i++) {
1663 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1667 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1668 rth = rcu_dereference(rth->u.dst.rt_next)) {
1669 unsigned short mtu = new_mtu;
1671 if (rth->fl.fl4_dst != daddr ||
1672 rth->fl.fl4_src != skeys[i] ||
1673 rth->rt_dst != daddr ||
1674 rth->rt_src != iph->saddr ||
1675 rth->fl.oif != ikeys[k] ||
1677 dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1678 !net_eq(dev_net(rth->u.dst.dev), net) ||
1682 if (new_mtu < 68 || new_mtu >= old_mtu) {
1684 /* BSD 4.2 compatibility hack :-( */
1686 old_mtu >= dst_mtu(&rth->u.dst) &&
1687 old_mtu >= 68 + (iph->ihl << 2))
1688 old_mtu -= iph->ihl << 2;
1690 mtu = guess_mtu(old_mtu);
1692 if (mtu <= dst_mtu(&rth->u.dst)) {
1693 if (mtu < dst_mtu(&rth->u.dst)) {
1694 dst_confirm(&rth->u.dst);
1695 if (mtu < ip_rt_min_pmtu) {
1696 mtu = ip_rt_min_pmtu;
1697 rth->u.dst.metrics[RTAX_LOCK-1] |=
1700 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1701 dst_set_expires(&rth->u.dst,
1710 return est_mtu ? : new_mtu;
1713 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1715 if (dst_mtu(dst) > mtu && mtu >= 68 &&
1716 !(dst_metric_locked(dst, RTAX_MTU))) {
1717 if (mtu < ip_rt_min_pmtu) {
1718 mtu = ip_rt_min_pmtu;
1719 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1721 dst->metrics[RTAX_MTU-1] = mtu;
1722 dst_set_expires(dst, ip_rt_mtu_expires);
1723 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1727 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1732 static void ipv4_dst_destroy(struct dst_entry *dst)
1734 struct rtable *rt = (struct rtable *) dst;
1735 struct inet_peer *peer = rt->peer;
1736 struct in_device *idev = rt->idev;
1749 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1752 struct rtable *rt = (struct rtable *) dst;
1753 struct in_device *idev = rt->idev;
1754 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1755 struct in_device *loopback_idev =
1756 in_dev_get(dev_net(dev)->loopback_dev);
1757 if (loopback_idev) {
1758 rt->idev = loopback_idev;
1764 static void ipv4_link_failure(struct sk_buff *skb)
1768 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1770 rt = skb_rtable(skb);
1772 dst_set_expires(&rt->u.dst, 0);
1775 static int ip_rt_bug(struct sk_buff *skb)
1777 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1778 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1779 skb->dev ? skb->dev->name : "?");
1785 We do not cache source address of outgoing interface,
1786 because it is used only by IP RR, TS and SRR options,
1787 so that it out of fast path.
1789 BTW remember: "addr" is allowed to be not aligned
1793 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1796 struct fib_result res;
1798 if (rt->fl.iif == 0)
1800 else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1801 src = FIB_RES_PREFSRC(res);
1804 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1806 memcpy(addr, &src, 4);
1809 #ifdef CONFIG_NET_CLS_ROUTE
1810 static void set_class_tag(struct rtable *rt, u32 tag)
1812 if (!(rt->u.dst.tclassid & 0xFFFF))
1813 rt->u.dst.tclassid |= tag & 0xFFFF;
1814 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1815 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1819 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1821 struct fib_info *fi = res->fi;
1824 if (FIB_RES_GW(*res) &&
1825 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1826 rt->rt_gateway = FIB_RES_GW(*res);
1827 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1828 sizeof(rt->u.dst.metrics));
1829 if (fi->fib_mtu == 0) {
1830 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1831 if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1832 rt->rt_gateway != rt->rt_dst &&
1833 rt->u.dst.dev->mtu > 576)
1834 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1836 #ifdef CONFIG_NET_CLS_ROUTE
1837 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1840 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1842 if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1843 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1844 if (dst_mtu(&rt->u.dst) > IP_MAX_MTU)
1845 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1846 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1847 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1849 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1850 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1852 #ifdef CONFIG_NET_CLS_ROUTE
1853 #ifdef CONFIG_IP_MULTIPLE_TABLES
1854 set_class_tag(rt, fib_rules_tclass(res));
1856 set_class_tag(rt, itag);
1858 rt->rt_type = res->type;
1861 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1862 u8 tos, struct net_device *dev, int our)
1867 struct in_device *in_dev = in_dev_get(dev);
1870 /* Primary sanity checks. */
1875 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1876 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1879 if (ipv4_is_zeronet(saddr)) {
1880 if (!ipv4_is_local_multicast(daddr))
1882 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1883 } else if (fib_validate_source(saddr, 0, tos, 0,
1884 dev, &spec_dst, &itag, 0) < 0)
1887 rth = dst_alloc(&ipv4_dst_ops);
1891 rth->u.dst.output= ip_rt_bug;
1893 atomic_set(&rth->u.dst.__refcnt, 1);
1894 rth->u.dst.flags= DST_HOST;
1895 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1896 rth->u.dst.flags |= DST_NOPOLICY;
1897 rth->fl.fl4_dst = daddr;
1898 rth->rt_dst = daddr;
1899 rth->fl.fl4_tos = tos;
1900 rth->fl.mark = skb->mark;
1901 rth->fl.fl4_src = saddr;
1902 rth->rt_src = saddr;
1903 #ifdef CONFIG_NET_CLS_ROUTE
1904 rth->u.dst.tclassid = itag;
1907 rth->fl.iif = dev->ifindex;
1908 rth->u.dst.dev = init_net.loopback_dev;
1909 dev_hold(rth->u.dst.dev);
1910 rth->idev = in_dev_get(rth->u.dst.dev);
1912 rth->rt_gateway = daddr;
1913 rth->rt_spec_dst= spec_dst;
1914 rth->rt_genid = rt_genid(dev_net(dev));
1915 rth->rt_flags = RTCF_MULTICAST;
1916 rth->rt_type = RTN_MULTICAST;
1918 rth->u.dst.input= ip_local_deliver;
1919 rth->rt_flags |= RTCF_LOCAL;
1922 #ifdef CONFIG_IP_MROUTE
1923 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1924 rth->u.dst.input = ip_mr_input;
1926 RT_CACHE_STAT_INC(in_slow_mc);
1929 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1930 return rt_intern_hash(hash, rth, NULL, skb);
1942 static void ip_handle_martian_source(struct net_device *dev,
1943 struct in_device *in_dev,
1944 struct sk_buff *skb,
1948 RT_CACHE_STAT_INC(in_martian_src);
1949 #ifdef CONFIG_IP_ROUTE_VERBOSE
1950 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1952 * RFC1812 recommendation, if source is martian,
1953 * the only hint is MAC header.
1955 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1956 &daddr, &saddr, dev->name);
1957 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1959 const unsigned char *p = skb_mac_header(skb);
1960 printk(KERN_WARNING "ll header: ");
1961 for (i = 0; i < dev->hard_header_len; i++, p++) {
1963 if (i < (dev->hard_header_len - 1))
1972 static int __mkroute_input(struct sk_buff *skb,
1973 struct fib_result *res,
1974 struct in_device *in_dev,
1975 __be32 daddr, __be32 saddr, u32 tos,
1976 struct rtable **result)
1981 struct in_device *out_dev;
1986 /* get a working reference to the output device */
1987 out_dev = in_dev_get(FIB_RES_DEV(*res));
1988 if (out_dev == NULL) {
1989 if (net_ratelimit())
1990 printk(KERN_CRIT "Bug in ip_route_input" \
1991 "_slow(). Please, report\n");
1996 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1997 in_dev->dev, &spec_dst, &itag, skb->mark);
1999 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2007 flags |= RTCF_DIRECTSRC;
2009 if (out_dev == in_dev && err &&
2010 (IN_DEV_SHARED_MEDIA(out_dev) ||
2011 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2012 flags |= RTCF_DOREDIRECT;
2014 if (skb->protocol != htons(ETH_P_IP)) {
2015 /* Not IP (i.e. ARP). Do not create route, if it is
2016 * invalid for proxy arp. DNAT routes are always valid.
2018 * Proxy arp feature have been extended to allow, ARP
2019 * replies back to the same interface, to support
2020 * Private VLAN switch technologies. See arp.c.
2022 if (out_dev == in_dev &&
2023 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2030 rth = dst_alloc(&ipv4_dst_ops);
2036 atomic_set(&rth->u.dst.__refcnt, 1);
2037 rth->u.dst.flags= DST_HOST;
2038 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2039 rth->u.dst.flags |= DST_NOPOLICY;
2040 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2041 rth->u.dst.flags |= DST_NOXFRM;
2042 rth->fl.fl4_dst = daddr;
2043 rth->rt_dst = daddr;
2044 rth->fl.fl4_tos = tos;
2045 rth->fl.mark = skb->mark;
2046 rth->fl.fl4_src = saddr;
2047 rth->rt_src = saddr;
2048 rth->rt_gateway = daddr;
2050 rth->fl.iif = in_dev->dev->ifindex;
2051 rth->u.dst.dev = (out_dev)->dev;
2052 dev_hold(rth->u.dst.dev);
2053 rth->idev = in_dev_get(rth->u.dst.dev);
2055 rth->rt_spec_dst= spec_dst;
2057 rth->u.dst.input = ip_forward;
2058 rth->u.dst.output = ip_output;
2059 rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
2061 rt_set_nexthop(rth, res, itag);
2063 rth->rt_flags = flags;
2068 /* release the working reference to the output device */
2069 in_dev_put(out_dev);
2073 static int ip_mkroute_input(struct sk_buff *skb,
2074 struct fib_result *res,
2075 const struct flowi *fl,
2076 struct in_device *in_dev,
2077 __be32 daddr, __be32 saddr, u32 tos)
2079 struct rtable* rth = NULL;
2083 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2084 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2085 fib_select_multipath(fl, res);
2088 /* create a routing cache entry */
2089 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2093 /* put it into the cache */
2094 hash = rt_hash(daddr, saddr, fl->iif,
2095 rt_genid(dev_net(rth->u.dst.dev)));
2096 return rt_intern_hash(hash, rth, NULL, skb);
2100 * NOTE. We drop all the packets that has local source
2101 * addresses, because every properly looped back packet
2102 * must have correct destination already attached by output routine.
2104 * Such approach solves two big problems:
2105 * 1. Not simplex devices are handled properly.
2106 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2109 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2110 u8 tos, struct net_device *dev)
2112 struct fib_result res;
2113 struct in_device *in_dev = in_dev_get(dev);
2114 struct flowi fl = { .nl_u = { .ip4_u =
2118 .scope = RT_SCOPE_UNIVERSE,
2121 .iif = dev->ifindex };
2124 struct rtable * rth;
2129 struct net * net = dev_net(dev);
2131 /* IP on this device is disabled. */
2136 /* Check for the most weird martians, which can be not detected
2140 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2141 ipv4_is_loopback(saddr))
2142 goto martian_source;
2144 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2147 /* Accept zero addresses only to limited broadcast;
2148 * I even do not know to fix it or not. Waiting for complains :-)
2150 if (ipv4_is_zeronet(saddr))
2151 goto martian_source;
2153 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2154 ipv4_is_loopback(daddr))
2155 goto martian_destination;
2158 * Now we are ready to route packet.
2160 if ((err = fib_lookup(net, &fl, &res)) != 0) {
2161 if (!IN_DEV_FORWARD(in_dev))
2167 RT_CACHE_STAT_INC(in_slow_tot);
2169 if (res.type == RTN_BROADCAST)
2172 if (res.type == RTN_LOCAL) {
2174 result = fib_validate_source(saddr, daddr, tos,
2175 net->loopback_dev->ifindex,
2176 dev, &spec_dst, &itag, skb->mark);
2178 goto martian_source;
2180 flags |= RTCF_DIRECTSRC;
2185 if (!IN_DEV_FORWARD(in_dev))
2187 if (res.type != RTN_UNICAST)
2188 goto martian_destination;
2190 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2198 if (skb->protocol != htons(ETH_P_IP))
2201 if (ipv4_is_zeronet(saddr))
2202 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2204 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2207 goto martian_source;
2209 flags |= RTCF_DIRECTSRC;
2211 flags |= RTCF_BROADCAST;
2212 res.type = RTN_BROADCAST;
2213 RT_CACHE_STAT_INC(in_brd);
2216 rth = dst_alloc(&ipv4_dst_ops);
2220 rth->u.dst.output= ip_rt_bug;
2221 rth->rt_genid = rt_genid(net);
2223 atomic_set(&rth->u.dst.__refcnt, 1);
2224 rth->u.dst.flags= DST_HOST;
2225 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2226 rth->u.dst.flags |= DST_NOPOLICY;
2227 rth->fl.fl4_dst = daddr;
2228 rth->rt_dst = daddr;
2229 rth->fl.fl4_tos = tos;
2230 rth->fl.mark = skb->mark;
2231 rth->fl.fl4_src = saddr;
2232 rth->rt_src = saddr;
2233 #ifdef CONFIG_NET_CLS_ROUTE
2234 rth->u.dst.tclassid = itag;
2237 rth->fl.iif = dev->ifindex;
2238 rth->u.dst.dev = net->loopback_dev;
2239 dev_hold(rth->u.dst.dev);
2240 rth->idev = in_dev_get(rth->u.dst.dev);
2241 rth->rt_gateway = daddr;
2242 rth->rt_spec_dst= spec_dst;
2243 rth->u.dst.input= ip_local_deliver;
2244 rth->rt_flags = flags|RTCF_LOCAL;
2245 if (res.type == RTN_UNREACHABLE) {
2246 rth->u.dst.input= ip_error;
2247 rth->u.dst.error= -err;
2248 rth->rt_flags &= ~RTCF_LOCAL;
2250 rth->rt_type = res.type;
2251 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2252 err = rt_intern_hash(hash, rth, NULL, skb);
2256 RT_CACHE_STAT_INC(in_no_route);
2257 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2258 res.type = RTN_UNREACHABLE;
2264 * Do not cache martian addresses: they should be logged (RFC1812)
2266 martian_destination:
2267 RT_CACHE_STAT_INC(in_martian_dst);
2268 #ifdef CONFIG_IP_ROUTE_VERBOSE
2269 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2270 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2271 &daddr, &saddr, dev->name);
2275 err = -EHOSTUNREACH;
2287 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2291 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2292 u8 tos, struct net_device *dev)
2294 struct rtable * rth;
2296 int iif = dev->ifindex;
2301 if (!rt_caching(net))
2304 tos &= IPTOS_RT_MASK;
2305 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2308 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2309 rth = rcu_dereference(rth->u.dst.rt_next)) {
2310 if (((rth->fl.fl4_dst ^ daddr) |
2311 (rth->fl.fl4_src ^ saddr) |
2312 (rth->fl.iif ^ iif) |
2314 (rth->fl.fl4_tos ^ tos)) == 0 &&
2315 rth->fl.mark == skb->mark &&
2316 net_eq(dev_net(rth->u.dst.dev), net) &&
2317 !rt_is_expired(rth)) {
2318 dst_use(&rth->u.dst, jiffies);
2319 RT_CACHE_STAT_INC(in_hit);
2321 skb_dst_set(skb, &rth->u.dst);
2324 RT_CACHE_STAT_INC(in_hlist_search);
2329 /* Multicast recognition logic is moved from route cache to here.
2330 The problem was that too many Ethernet cards have broken/missing
2331 hardware multicast filters :-( As result the host on multicasting
2332 network acquires a lot of useless route cache entries, sort of
2333 SDR messages from all the world. Now we try to get rid of them.
2334 Really, provided software IP multicast filter is organized
2335 reasonably (at least, hashed), it does not result in a slowdown
2336 comparing with route cache reject entries.
2337 Note, that multicast routers are not affected, because
2338 route cache entry is created eventually.
2340 if (ipv4_is_multicast(daddr)) {
2341 struct in_device *in_dev;
2344 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2345 int our = ip_check_mc(in_dev, daddr, saddr,
2346 ip_hdr(skb)->protocol);
2348 #ifdef CONFIG_IP_MROUTE
2350 (!ipv4_is_local_multicast(daddr) &&
2351 IN_DEV_MFORWARD(in_dev))
2355 return ip_route_input_mc(skb, daddr, saddr,
2362 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2365 static int __mkroute_output(struct rtable **result,
2366 struct fib_result *res,
2367 const struct flowi *fl,
2368 const struct flowi *oldflp,
2369 struct net_device *dev_out,
2373 struct in_device *in_dev;
2374 u32 tos = RT_FL_TOS(oldflp);
2377 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2380 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2381 res->type = RTN_BROADCAST;
2382 else if (ipv4_is_multicast(fl->fl4_dst))
2383 res->type = RTN_MULTICAST;
2384 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2387 if (dev_out->flags & IFF_LOOPBACK)
2388 flags |= RTCF_LOCAL;
2390 /* get work reference to inet device */
2391 in_dev = in_dev_get(dev_out);
2395 if (res->type == RTN_BROADCAST) {
2396 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2398 fib_info_put(res->fi);
2401 } else if (res->type == RTN_MULTICAST) {
2402 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2403 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2405 flags &= ~RTCF_LOCAL;
2406 /* If multicast route do not exist use
2407 default one, but do not gateway in this case.
2410 if (res->fi && res->prefixlen < 4) {
2411 fib_info_put(res->fi);
2417 rth = dst_alloc(&ipv4_dst_ops);
2423 atomic_set(&rth->u.dst.__refcnt, 1);
2424 rth->u.dst.flags= DST_HOST;
2425 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2426 rth->u.dst.flags |= DST_NOXFRM;
2427 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2428 rth->u.dst.flags |= DST_NOPOLICY;
2430 rth->fl.fl4_dst = oldflp->fl4_dst;
2431 rth->fl.fl4_tos = tos;
2432 rth->fl.fl4_src = oldflp->fl4_src;
2433 rth->fl.oif = oldflp->oif;
2434 rth->fl.mark = oldflp->mark;
2435 rth->rt_dst = fl->fl4_dst;
2436 rth->rt_src = fl->fl4_src;
2437 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2438 /* get references to the devices that are to be hold by the routing
2440 rth->u.dst.dev = dev_out;
2442 rth->idev = in_dev_get(dev_out);
2443 rth->rt_gateway = fl->fl4_dst;
2444 rth->rt_spec_dst= fl->fl4_src;
2446 rth->u.dst.output=ip_output;
2447 rth->rt_genid = rt_genid(dev_net(dev_out));
2449 RT_CACHE_STAT_INC(out_slow_tot);
2451 if (flags & RTCF_LOCAL) {
2452 rth->u.dst.input = ip_local_deliver;
2453 rth->rt_spec_dst = fl->fl4_dst;
2455 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2456 rth->rt_spec_dst = fl->fl4_src;
2457 if (flags & RTCF_LOCAL &&
2458 !(dev_out->flags & IFF_LOOPBACK)) {
2459 rth->u.dst.output = ip_mc_output;
2460 RT_CACHE_STAT_INC(out_slow_mc);
2462 #ifdef CONFIG_IP_MROUTE
2463 if (res->type == RTN_MULTICAST) {
2464 if (IN_DEV_MFORWARD(in_dev) &&
2465 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2466 rth->u.dst.input = ip_mr_input;
2467 rth->u.dst.output = ip_mc_output;
2473 rt_set_nexthop(rth, res, 0);
2475 rth->rt_flags = flags;
2479 /* release work reference to inet device */
2485 static int ip_mkroute_output(struct rtable **rp,
2486 struct fib_result *res,
2487 const struct flowi *fl,
2488 const struct flowi *oldflp,
2489 struct net_device *dev_out,
2492 struct rtable *rth = NULL;
2493 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2496 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2497 rt_genid(dev_net(dev_out)));
2498 err = rt_intern_hash(hash, rth, rp, NULL);
2505 * Major route resolver routine.
2508 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2509 const struct flowi *oldflp)
2511 u32 tos = RT_FL_TOS(oldflp);
2512 struct flowi fl = { .nl_u = { .ip4_u =
2513 { .daddr = oldflp->fl4_dst,
2514 .saddr = oldflp->fl4_src,
2515 .tos = tos & IPTOS_RT_MASK,
2516 .scope = ((tos & RTO_ONLINK) ?
2520 .mark = oldflp->mark,
2521 .iif = net->loopback_dev->ifindex,
2522 .oif = oldflp->oif };
2523 struct fib_result res;
2525 struct net_device *dev_out = NULL;
2531 #ifdef CONFIG_IP_MULTIPLE_TABLES
2535 if (oldflp->fl4_src) {
2537 if (ipv4_is_multicast(oldflp->fl4_src) ||
2538 ipv4_is_lbcast(oldflp->fl4_src) ||
2539 ipv4_is_zeronet(oldflp->fl4_src))
2542 /* I removed check for oif == dev_out->oif here.
2543 It was wrong for two reasons:
2544 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2545 is assigned to multiple interfaces.
2546 2. Moreover, we are allowed to send packets with saddr
2547 of another iface. --ANK
2550 if (oldflp->oif == 0 &&
2551 (ipv4_is_multicast(oldflp->fl4_dst) ||
2552 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2553 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2554 dev_out = ip_dev_find(net, oldflp->fl4_src);
2555 if (dev_out == NULL)
2558 /* Special hack: user can direct multicasts
2559 and limited broadcast via necessary interface
2560 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2561 This hack is not just for fun, it allows
2562 vic,vat and friends to work.
2563 They bind socket to loopback, set ttl to zero
2564 and expect that it will work.
2565 From the viewpoint of routing cache they are broken,
2566 because we are not allowed to build multicast path
2567 with loopback source addr (look, routing cache
2568 cannot know, that ttl is zero, so that packet
2569 will not leave this host and route is valid).
2570 Luckily, this hack is good workaround.
2573 fl.oif = dev_out->ifindex;
2577 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2578 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2579 dev_out = ip_dev_find(net, oldflp->fl4_src);
2580 if (dev_out == NULL)
2589 dev_out = dev_get_by_index(net, oldflp->oif);
2591 if (dev_out == NULL)
2594 /* RACE: Check return value of inet_select_addr instead. */
2595 if (__in_dev_get_rtnl(dev_out) == NULL) {
2597 goto out; /* Wrong error code */
2600 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2601 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2603 fl.fl4_src = inet_select_addr(dev_out, 0,
2608 if (ipv4_is_multicast(oldflp->fl4_dst))
2609 fl.fl4_src = inet_select_addr(dev_out, 0,
2611 else if (!oldflp->fl4_dst)
2612 fl.fl4_src = inet_select_addr(dev_out, 0,
2618 fl.fl4_dst = fl.fl4_src;
2620 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2623 dev_out = net->loopback_dev;
2625 fl.oif = net->loopback_dev->ifindex;
2626 res.type = RTN_LOCAL;
2627 flags |= RTCF_LOCAL;
2631 if (fib_lookup(net, &fl, &res)) {
2634 /* Apparently, routing tables are wrong. Assume,
2635 that the destination is on link.
2638 Because we are allowed to send to iface
2639 even if it has NO routes and NO assigned
2640 addresses. When oif is specified, routing
2641 tables are looked up with only one purpose:
2642 to catch if destination is gatewayed, rather than
2643 direct. Moreover, if MSG_DONTROUTE is set,
2644 we send packet, ignoring both routing tables
2645 and ifaddr state. --ANK
2648 We could make it even if oif is unknown,
2649 likely IPv6, but we do not.
2652 if (fl.fl4_src == 0)
2653 fl.fl4_src = inet_select_addr(dev_out, 0,
2655 res.type = RTN_UNICAST;
2665 if (res.type == RTN_LOCAL) {
2667 fl.fl4_src = fl.fl4_dst;
2670 dev_out = net->loopback_dev;
2672 fl.oif = dev_out->ifindex;
2674 fib_info_put(res.fi);
2676 flags |= RTCF_LOCAL;
2680 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2681 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2682 fib_select_multipath(&fl, &res);
2685 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2686 fib_select_default(net, &fl, &res);
2689 fl.fl4_src = FIB_RES_PREFSRC(res);
2693 dev_out = FIB_RES_DEV(res);
2695 fl.oif = dev_out->ifindex;
2699 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2709 int __ip_route_output_key(struct net *net, struct rtable **rp,
2710 const struct flowi *flp)
2715 if (!rt_caching(net))
2718 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2721 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2722 rth = rcu_dereference_bh(rth->u.dst.rt_next)) {
2723 if (rth->fl.fl4_dst == flp->fl4_dst &&
2724 rth->fl.fl4_src == flp->fl4_src &&
2726 rth->fl.oif == flp->oif &&
2727 rth->fl.mark == flp->mark &&
2728 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2729 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2730 net_eq(dev_net(rth->u.dst.dev), net) &&
2731 !rt_is_expired(rth)) {
2732 dst_use(&rth->u.dst, jiffies);
2733 RT_CACHE_STAT_INC(out_hit);
2734 rcu_read_unlock_bh();
2738 RT_CACHE_STAT_INC(out_hlist_search);
2740 rcu_read_unlock_bh();
2743 return ip_route_output_slow(net, rp, flp);
2746 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2748 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2752 static struct dst_ops ipv4_dst_blackhole_ops = {
2754 .protocol = cpu_to_be16(ETH_P_IP),
2755 .destroy = ipv4_dst_destroy,
2756 .check = ipv4_dst_check,
2757 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2758 .entries = ATOMIC_INIT(0),
2762 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2764 struct rtable *ort = *rp;
2765 struct rtable *rt = (struct rtable *)
2766 dst_alloc(&ipv4_dst_blackhole_ops);
2769 struct dst_entry *new = &rt->u.dst;
2771 atomic_set(&new->__refcnt, 1);
2773 new->input = dst_discard;
2774 new->output = dst_discard;
2775 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2777 new->dev = ort->u.dst.dev;
2783 rt->idev = ort->idev;
2785 in_dev_hold(rt->idev);
2786 rt->rt_genid = rt_genid(net);
2787 rt->rt_flags = ort->rt_flags;
2788 rt->rt_type = ort->rt_type;
2789 rt->rt_dst = ort->rt_dst;
2790 rt->rt_src = ort->rt_src;
2791 rt->rt_iif = ort->rt_iif;
2792 rt->rt_gateway = ort->rt_gateway;
2793 rt->rt_spec_dst = ort->rt_spec_dst;
2794 rt->peer = ort->peer;
2796 atomic_inc(&rt->peer->refcnt);
2801 dst_release(&(*rp)->u.dst);
2803 return (rt ? 0 : -ENOMEM);
2806 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2807 struct sock *sk, int flags)
2811 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2816 flp->fl4_src = (*rp)->rt_src;
2818 flp->fl4_dst = (*rp)->rt_dst;
2819 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2820 flags ? XFRM_LOOKUP_WAIT : 0);
2821 if (err == -EREMOTE)
2822 err = ipv4_dst_blackhole(net, rp, flp);
2830 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2832 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2834 return ip_route_output_flow(net, rp, flp, NULL, 0);
2837 static int rt_fill_info(struct net *net,
2838 struct sk_buff *skb, u32 pid, u32 seq, int event,
2839 int nowait, unsigned int flags)
2841 struct rtable *rt = skb_rtable(skb);
2843 struct nlmsghdr *nlh;
2845 u32 id = 0, ts = 0, tsage = 0, error;
2847 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2851 r = nlmsg_data(nlh);
2852 r->rtm_family = AF_INET;
2853 r->rtm_dst_len = 32;
2855 r->rtm_tos = rt->fl.fl4_tos;
2856 r->rtm_table = RT_TABLE_MAIN;
2857 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2858 r->rtm_type = rt->rt_type;
2859 r->rtm_scope = RT_SCOPE_UNIVERSE;
2860 r->rtm_protocol = RTPROT_UNSPEC;
2861 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2862 if (rt->rt_flags & RTCF_NOTIFY)
2863 r->rtm_flags |= RTM_F_NOTIFY;
2865 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2867 if (rt->fl.fl4_src) {
2868 r->rtm_src_len = 32;
2869 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2872 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2873 #ifdef CONFIG_NET_CLS_ROUTE
2874 if (rt->u.dst.tclassid)
2875 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2878 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2879 else if (rt->rt_src != rt->fl.fl4_src)
2880 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2882 if (rt->rt_dst != rt->rt_gateway)
2883 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2885 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2886 goto nla_put_failure;
2888 error = rt->u.dst.error;
2889 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2891 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2892 if (rt->peer->tcp_ts_stamp) {
2893 ts = rt->peer->tcp_ts;
2894 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2899 #ifdef CONFIG_IP_MROUTE
2900 __be32 dst = rt->rt_dst;
2902 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2903 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2904 int err = ipmr_get_route(net, skb, r, nowait);
2909 goto nla_put_failure;
2911 if (err == -EMSGSIZE)
2912 goto nla_put_failure;
2918 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2921 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2922 expires, error) < 0)
2923 goto nla_put_failure;
2925 return nlmsg_end(skb, nlh);
2928 nlmsg_cancel(skb, nlh);
2932 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2934 struct net *net = sock_net(in_skb->sk);
2936 struct nlattr *tb[RTA_MAX+1];
2937 struct rtable *rt = NULL;
2942 struct sk_buff *skb;
2944 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2948 rtm = nlmsg_data(nlh);
2950 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2956 /* Reserve room for dummy headers, this skb can pass
2957 through good chunk of routing engine.
2959 skb_reset_mac_header(skb);
2960 skb_reset_network_header(skb);
2962 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2963 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2964 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2966 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2967 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2968 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2971 struct net_device *dev;
2973 dev = __dev_get_by_index(net, iif);
2979 skb->protocol = htons(ETH_P_IP);
2982 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2985 rt = skb_rtable(skb);
2986 if (err == 0 && rt->u.dst.error)
2987 err = -rt->u.dst.error;
2994 .tos = rtm->rtm_tos,
2997 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2999 err = ip_route_output_key(net, &rt, &fl);
3005 skb_dst_set(skb, &rt->u.dst);
3006 if (rtm->rtm_flags & RTM_F_NOTIFY)
3007 rt->rt_flags |= RTCF_NOTIFY;
3009 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3010 RTM_NEWROUTE, 0, 0);
3014 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3023 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3030 net = sock_net(skb->sk);
3035 s_idx = idx = cb->args[1];
3036 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3037 if (!rt_hash_table[h].chain)
3040 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3041 rt = rcu_dereference_bh(rt->u.dst.rt_next), idx++) {
3042 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
3044 if (rt_is_expired(rt))
3046 skb_dst_set(skb, dst_clone(&rt->u.dst));
3047 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3048 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3049 1, NLM_F_MULTI) <= 0) {
3051 rcu_read_unlock_bh();
3056 rcu_read_unlock_bh();
3065 void ip_rt_multicast_event(struct in_device *in_dev)
3067 rt_cache_flush(dev_net(in_dev->dev), 0);
3070 #ifdef CONFIG_SYSCTL
3071 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3072 void __user *buffer,
3073 size_t *lenp, loff_t *ppos)
3080 memcpy(&ctl, __ctl, sizeof(ctl));
3081 ctl.data = &flush_delay;
3082 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3084 net = (struct net *)__ctl->extra1;
3085 rt_cache_flush(net, flush_delay);
3092 static void rt_secret_reschedule(int old)
3095 int new = ip_rt_secret_interval;
3096 int diff = new - old;
3103 int deleted = del_timer_sync(&net->ipv4.rt_secret_timer);
3110 time = net->ipv4.rt_secret_timer.expires - jiffies;
3112 if (time <= 0 || (time += diff) <= 0)
3117 mod_timer(&net->ipv4.rt_secret_timer, jiffies + time);
3122 static int ipv4_sysctl_rt_secret_interval(ctl_table *ctl, int write,
3123 void __user *buffer, size_t *lenp,
3126 int old = ip_rt_secret_interval;
3127 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3129 rt_secret_reschedule(old);
3134 static ctl_table ipv4_route_table[] = {
3136 .procname = "gc_thresh",
3137 .data = &ipv4_dst_ops.gc_thresh,
3138 .maxlen = sizeof(int),
3140 .proc_handler = proc_dointvec,
3143 .procname = "max_size",
3144 .data = &ip_rt_max_size,
3145 .maxlen = sizeof(int),
3147 .proc_handler = proc_dointvec,
3150 /* Deprecated. Use gc_min_interval_ms */
3152 .procname = "gc_min_interval",
3153 .data = &ip_rt_gc_min_interval,
3154 .maxlen = sizeof(int),
3156 .proc_handler = proc_dointvec_jiffies,
3159 .procname = "gc_min_interval_ms",
3160 .data = &ip_rt_gc_min_interval,
3161 .maxlen = sizeof(int),
3163 .proc_handler = proc_dointvec_ms_jiffies,
3166 .procname = "gc_timeout",
3167 .data = &ip_rt_gc_timeout,
3168 .maxlen = sizeof(int),
3170 .proc_handler = proc_dointvec_jiffies,
3173 .procname = "gc_interval",
3174 .data = &ip_rt_gc_interval,
3175 .maxlen = sizeof(int),
3177 .proc_handler = proc_dointvec_jiffies,
3180 .procname = "redirect_load",
3181 .data = &ip_rt_redirect_load,
3182 .maxlen = sizeof(int),
3184 .proc_handler = proc_dointvec,
3187 .procname = "redirect_number",
3188 .data = &ip_rt_redirect_number,
3189 .maxlen = sizeof(int),
3191 .proc_handler = proc_dointvec,
3194 .procname = "redirect_silence",
3195 .data = &ip_rt_redirect_silence,
3196 .maxlen = sizeof(int),
3198 .proc_handler = proc_dointvec,
3201 .procname = "error_cost",
3202 .data = &ip_rt_error_cost,
3203 .maxlen = sizeof(int),
3205 .proc_handler = proc_dointvec,
3208 .procname = "error_burst",
3209 .data = &ip_rt_error_burst,
3210 .maxlen = sizeof(int),
3212 .proc_handler = proc_dointvec,
3215 .procname = "gc_elasticity",
3216 .data = &ip_rt_gc_elasticity,
3217 .maxlen = sizeof(int),
3219 .proc_handler = proc_dointvec,
3222 .procname = "mtu_expires",
3223 .data = &ip_rt_mtu_expires,
3224 .maxlen = sizeof(int),
3226 .proc_handler = proc_dointvec_jiffies,
3229 .procname = "min_pmtu",
3230 .data = &ip_rt_min_pmtu,
3231 .maxlen = sizeof(int),
3233 .proc_handler = proc_dointvec,
3236 .procname = "min_adv_mss",
3237 .data = &ip_rt_min_advmss,
3238 .maxlen = sizeof(int),
3240 .proc_handler = proc_dointvec,
3243 .procname = "secret_interval",
3244 .data = &ip_rt_secret_interval,
3245 .maxlen = sizeof(int),
3247 .proc_handler = ipv4_sysctl_rt_secret_interval,
3252 static struct ctl_table empty[1];
3254 static struct ctl_table ipv4_skeleton[] =
3256 { .procname = "route",
3257 .mode = 0555, .child = ipv4_route_table},
3258 { .procname = "neigh",
3259 .mode = 0555, .child = empty},
3263 static __net_initdata struct ctl_path ipv4_path[] = {
3264 { .procname = "net", },
3265 { .procname = "ipv4", },
3269 static struct ctl_table ipv4_route_flush_table[] = {
3271 .procname = "flush",
3272 .maxlen = sizeof(int),
3274 .proc_handler = ipv4_sysctl_rtcache_flush,
3279 static __net_initdata struct ctl_path ipv4_route_path[] = {
3280 { .procname = "net", },
3281 { .procname = "ipv4", },
3282 { .procname = "route", },
3286 static __net_init int sysctl_route_net_init(struct net *net)
3288 struct ctl_table *tbl;
3290 tbl = ipv4_route_flush_table;
3291 if (!net_eq(net, &init_net)) {
3292 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3296 tbl[0].extra1 = net;
3298 net->ipv4.route_hdr =
3299 register_net_sysctl_table(net, ipv4_route_path, tbl);
3300 if (net->ipv4.route_hdr == NULL)
3305 if (tbl != ipv4_route_flush_table)
3311 static __net_exit void sysctl_route_net_exit(struct net *net)
3313 struct ctl_table *tbl;
3315 tbl = net->ipv4.route_hdr->ctl_table_arg;
3316 unregister_net_sysctl_table(net->ipv4.route_hdr);
3317 BUG_ON(tbl == ipv4_route_flush_table);
3321 static __net_initdata struct pernet_operations sysctl_route_ops = {
3322 .init = sysctl_route_net_init,
3323 .exit = sysctl_route_net_exit,
3328 static __net_init int rt_secret_timer_init(struct net *net)
3330 atomic_set(&net->ipv4.rt_genid,
3331 (int) ((num_physpages ^ (num_physpages>>8)) ^
3332 (jiffies ^ (jiffies >> 7))));
3334 net->ipv4.rt_secret_timer.function = rt_secret_rebuild;
3335 net->ipv4.rt_secret_timer.data = (unsigned long)net;
3336 init_timer_deferrable(&net->ipv4.rt_secret_timer);
3338 if (ip_rt_secret_interval) {
3339 net->ipv4.rt_secret_timer.expires =
3340 jiffies + net_random() % ip_rt_secret_interval +
3341 ip_rt_secret_interval;
3342 add_timer(&net->ipv4.rt_secret_timer);
3347 static __net_exit void rt_secret_timer_exit(struct net *net)
3349 del_timer_sync(&net->ipv4.rt_secret_timer);
3352 static __net_initdata struct pernet_operations rt_secret_timer_ops = {
3353 .init = rt_secret_timer_init,
3354 .exit = rt_secret_timer_exit,
3358 #ifdef CONFIG_NET_CLS_ROUTE
3359 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3360 #endif /* CONFIG_NET_CLS_ROUTE */
3362 static __initdata unsigned long rhash_entries;
3363 static int __init set_rhash_entries(char *str)
3367 rhash_entries = simple_strtoul(str, &str, 0);
3370 __setup("rhash_entries=", set_rhash_entries);
3372 int __init ip_rt_init(void)
3376 #ifdef CONFIG_NET_CLS_ROUTE
3377 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3379 panic("IP: failed to allocate ip_rt_acct\n");
3382 ipv4_dst_ops.kmem_cachep =
3383 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3384 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3386 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3388 rt_hash_table = (struct rt_hash_bucket *)
3389 alloc_large_system_hash("IP route cache",
3390 sizeof(struct rt_hash_bucket),
3392 (totalram_pages >= 128 * 1024) ?
3397 rhash_entries ? 0 : 512 * 1024);
3398 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3399 rt_hash_lock_init();
3401 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3402 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3407 /* All the timers, started at system startup tend
3408 to synchronize. Perturb it a bit.
3410 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3411 expires_ljiffies = jiffies;
3412 schedule_delayed_work(&expires_work,
3413 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3415 if (register_pernet_subsys(&rt_secret_timer_ops))
3416 printk(KERN_ERR "Unable to setup rt_secret_timer\n");
3418 if (ip_rt_proc_init())
3419 printk(KERN_ERR "Unable to create route proc files\n");
3422 xfrm4_init(ip_rt_max_size);
3424 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3426 #ifdef CONFIG_SYSCTL
3427 register_pernet_subsys(&sysctl_route_ops);
3432 #ifdef CONFIG_SYSCTL
3434 * We really need to sanitize the damn ipv4 init order, then all
3435 * this nonsense will go away.
3437 void __init ip_static_sysctl_init(void)
3439 register_sysctl_paths(ipv4_path, ipv4_skeleton);
3443 EXPORT_SYMBOL(__ip_select_ident);
3444 EXPORT_SYMBOL(ip_route_input);
3445 EXPORT_SYMBOL(ip_route_output_key);