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>
94 #include <linux/prefetch.h>
96 #include <net/net_namespace.h>
97 #include <net/protocol.h>
99 #include <net/route.h>
100 #include <net/inetpeer.h>
101 #include <net/sock.h>
102 #include <net/ip_fib.h>
105 #include <net/icmp.h>
106 #include <net/xfrm.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
112 #include <net/atmclip.h>
113 #include <net/secure_seq.h>
115 #define RT_FL_TOS(oldflp4) \
116 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118 #define IP_MAX_MTU 0xFFF0
120 #define RT_GC_TIMEOUT (300*HZ)
122 static int ip_rt_max_size;
123 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
124 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
125 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
126 static int ip_rt_redirect_number __read_mostly = 9;
127 static int ip_rt_redirect_load __read_mostly = HZ / 50;
128 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
129 static int ip_rt_error_cost __read_mostly = HZ;
130 static int ip_rt_error_burst __read_mostly = 5 * HZ;
131 static int ip_rt_gc_elasticity __read_mostly = 8;
132 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
133 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
134 static int ip_rt_min_advmss __read_mostly = 256;
135 static int rt_chain_length_max __read_mostly = 20;
136 static int redirect_genid;
138 static struct delayed_work expires_work;
139 static unsigned long expires_ljiffies;
142 * Interface to generic destination cache.
145 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
146 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
147 static unsigned int ipv4_mtu(const struct dst_entry *dst);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
149 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
150 static void ipv4_link_failure(struct sk_buff *skb);
151 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
152 static int rt_garbage_collect(struct dst_ops *ops);
154 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
159 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
161 struct rtable *rt = (struct rtable *) dst;
162 struct inet_peer *peer;
166 rt_bind_peer(rt, rt->rt_dst, 1);
170 u32 *old_p = __DST_METRICS_PTR(old);
171 unsigned long prev, new;
174 if (inet_metrics_new(peer))
175 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
177 new = (unsigned long) p;
178 prev = cmpxchg(&dst->_metrics, old, new);
181 p = __DST_METRICS_PTR(prev);
182 if (prev & DST_METRICS_READ_ONLY)
186 fib_info_put(rt->fi);
194 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr);
196 static struct dst_ops ipv4_dst_ops = {
198 .protocol = cpu_to_be16(ETH_P_IP),
199 .gc = rt_garbage_collect,
200 .check = ipv4_dst_check,
201 .default_advmss = ipv4_default_advmss,
203 .cow_metrics = ipv4_cow_metrics,
204 .destroy = ipv4_dst_destroy,
205 .ifdown = ipv4_dst_ifdown,
206 .negative_advice = ipv4_negative_advice,
207 .link_failure = ipv4_link_failure,
208 .update_pmtu = ip_rt_update_pmtu,
209 .local_out = __ip_local_out,
210 .neigh_lookup = ipv4_neigh_lookup,
213 #define ECN_OR_COST(class) TC_PRIO_##class
215 const __u8 ip_tos2prio[16] = {
217 ECN_OR_COST(BESTEFFORT),
219 ECN_OR_COST(BESTEFFORT),
225 ECN_OR_COST(INTERACTIVE),
227 ECN_OR_COST(INTERACTIVE),
228 TC_PRIO_INTERACTIVE_BULK,
229 ECN_OR_COST(INTERACTIVE_BULK),
230 TC_PRIO_INTERACTIVE_BULK,
231 ECN_OR_COST(INTERACTIVE_BULK)
239 /* The locking scheme is rather straight forward:
241 * 1) Read-Copy Update protects the buckets of the central route hash.
242 * 2) Only writers remove entries, and they hold the lock
243 * as they look at rtable reference counts.
244 * 3) Only readers acquire references to rtable entries,
245 * they do so with atomic increments and with the
249 struct rt_hash_bucket {
250 struct rtable __rcu *chain;
253 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
254 defined(CONFIG_PROVE_LOCKING)
256 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
257 * The size of this table is a power of two and depends on the number of CPUS.
258 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
260 #ifdef CONFIG_LOCKDEP
261 # define RT_HASH_LOCK_SZ 256
264 # define RT_HASH_LOCK_SZ 4096
266 # define RT_HASH_LOCK_SZ 2048
268 # define RT_HASH_LOCK_SZ 1024
270 # define RT_HASH_LOCK_SZ 512
272 # define RT_HASH_LOCK_SZ 256
276 static spinlock_t *rt_hash_locks;
277 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
279 static __init void rt_hash_lock_init(void)
283 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
286 panic("IP: failed to allocate rt_hash_locks\n");
288 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
289 spin_lock_init(&rt_hash_locks[i]);
292 # define rt_hash_lock_addr(slot) NULL
294 static inline void rt_hash_lock_init(void)
299 static struct rt_hash_bucket *rt_hash_table __read_mostly;
300 static unsigned rt_hash_mask __read_mostly;
301 static unsigned int rt_hash_log __read_mostly;
303 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
304 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
306 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
309 return jhash_3words((__force u32)daddr, (__force u32)saddr,
314 static inline int rt_genid(struct net *net)
316 return atomic_read(&net->ipv4.rt_genid);
319 #ifdef CONFIG_PROC_FS
320 struct rt_cache_iter_state {
321 struct seq_net_private p;
326 static struct rtable *rt_cache_get_first(struct seq_file *seq)
328 struct rt_cache_iter_state *st = seq->private;
329 struct rtable *r = NULL;
331 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
332 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
335 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
337 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
338 r->rt_genid == st->genid)
340 r = rcu_dereference_bh(r->dst.rt_next);
342 rcu_read_unlock_bh();
347 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
350 struct rt_cache_iter_state *st = seq->private;
352 r = rcu_dereference_bh(r->dst.rt_next);
354 rcu_read_unlock_bh();
356 if (--st->bucket < 0)
358 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
360 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
365 static struct rtable *rt_cache_get_next(struct seq_file *seq,
368 struct rt_cache_iter_state *st = seq->private;
369 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
370 if (dev_net(r->dst.dev) != seq_file_net(seq))
372 if (r->rt_genid == st->genid)
378 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
380 struct rtable *r = rt_cache_get_first(seq);
383 while (pos && (r = rt_cache_get_next(seq, r)))
385 return pos ? NULL : r;
388 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
390 struct rt_cache_iter_state *st = seq->private;
392 return rt_cache_get_idx(seq, *pos - 1);
393 st->genid = rt_genid(seq_file_net(seq));
394 return SEQ_START_TOKEN;
397 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
401 if (v == SEQ_START_TOKEN)
402 r = rt_cache_get_first(seq);
404 r = rt_cache_get_next(seq, v);
409 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
411 if (v && v != SEQ_START_TOKEN)
412 rcu_read_unlock_bh();
415 static int rt_cache_seq_show(struct seq_file *seq, void *v)
417 if (v == SEQ_START_TOKEN)
418 seq_printf(seq, "%-127s\n",
419 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
420 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
423 struct rtable *r = v;
428 n = dst_get_neighbour(&r->dst);
429 HHUptod = (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0;
432 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
433 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
434 r->dst.dev ? r->dst.dev->name : "*",
435 (__force u32)r->rt_dst,
436 (__force u32)r->rt_gateway,
437 r->rt_flags, atomic_read(&r->dst.__refcnt),
438 r->dst.__use, 0, (__force u32)r->rt_src,
439 dst_metric_advmss(&r->dst) + 40,
440 dst_metric(&r->dst, RTAX_WINDOW),
441 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
442 dst_metric(&r->dst, RTAX_RTTVAR)),
446 r->rt_spec_dst, &len);
448 seq_printf(seq, "%*s\n", 127 - len, "");
453 static const struct seq_operations rt_cache_seq_ops = {
454 .start = rt_cache_seq_start,
455 .next = rt_cache_seq_next,
456 .stop = rt_cache_seq_stop,
457 .show = rt_cache_seq_show,
460 static int rt_cache_seq_open(struct inode *inode, struct file *file)
462 return seq_open_net(inode, file, &rt_cache_seq_ops,
463 sizeof(struct rt_cache_iter_state));
466 static const struct file_operations rt_cache_seq_fops = {
467 .owner = THIS_MODULE,
468 .open = rt_cache_seq_open,
471 .release = seq_release_net,
475 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
480 return SEQ_START_TOKEN;
482 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
483 if (!cpu_possible(cpu))
486 return &per_cpu(rt_cache_stat, cpu);
491 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
495 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
496 if (!cpu_possible(cpu))
499 return &per_cpu(rt_cache_stat, cpu);
505 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
510 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
512 struct rt_cache_stat *st = v;
514 if (v == SEQ_START_TOKEN) {
515 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");
519 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
520 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
521 dst_entries_get_slow(&ipv4_dst_ops),
544 static const struct seq_operations rt_cpu_seq_ops = {
545 .start = rt_cpu_seq_start,
546 .next = rt_cpu_seq_next,
547 .stop = rt_cpu_seq_stop,
548 .show = rt_cpu_seq_show,
552 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
554 return seq_open(file, &rt_cpu_seq_ops);
557 static const struct file_operations rt_cpu_seq_fops = {
558 .owner = THIS_MODULE,
559 .open = rt_cpu_seq_open,
562 .release = seq_release,
565 #ifdef CONFIG_IP_ROUTE_CLASSID
566 static int rt_acct_proc_show(struct seq_file *m, void *v)
568 struct ip_rt_acct *dst, *src;
571 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
575 for_each_possible_cpu(i) {
576 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
577 for (j = 0; j < 256; j++) {
578 dst[j].o_bytes += src[j].o_bytes;
579 dst[j].o_packets += src[j].o_packets;
580 dst[j].i_bytes += src[j].i_bytes;
581 dst[j].i_packets += src[j].i_packets;
585 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
590 static int rt_acct_proc_open(struct inode *inode, struct file *file)
592 return single_open(file, rt_acct_proc_show, NULL);
595 static const struct file_operations rt_acct_proc_fops = {
596 .owner = THIS_MODULE,
597 .open = rt_acct_proc_open,
600 .release = single_release,
604 static int __net_init ip_rt_do_proc_init(struct net *net)
606 struct proc_dir_entry *pde;
608 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
613 pde = proc_create("rt_cache", S_IRUGO,
614 net->proc_net_stat, &rt_cpu_seq_fops);
618 #ifdef CONFIG_IP_ROUTE_CLASSID
619 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
625 #ifdef CONFIG_IP_ROUTE_CLASSID
627 remove_proc_entry("rt_cache", net->proc_net_stat);
630 remove_proc_entry("rt_cache", net->proc_net);
635 static void __net_exit ip_rt_do_proc_exit(struct net *net)
637 remove_proc_entry("rt_cache", net->proc_net_stat);
638 remove_proc_entry("rt_cache", net->proc_net);
639 #ifdef CONFIG_IP_ROUTE_CLASSID
640 remove_proc_entry("rt_acct", net->proc_net);
644 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
645 .init = ip_rt_do_proc_init,
646 .exit = ip_rt_do_proc_exit,
649 static int __init ip_rt_proc_init(void)
651 return register_pernet_subsys(&ip_rt_proc_ops);
655 static inline int ip_rt_proc_init(void)
659 #endif /* CONFIG_PROC_FS */
661 static inline void rt_free(struct rtable *rt)
663 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
666 static inline void rt_drop(struct rtable *rt)
669 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
672 static inline int rt_fast_clean(struct rtable *rth)
674 /* Kill broadcast/multicast entries very aggresively, if they
675 collide in hash table with more useful entries */
676 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
677 rt_is_input_route(rth) && rth->dst.rt_next;
680 static inline int rt_valuable(struct rtable *rth)
682 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
683 (rth->peer && rth->peer->pmtu_expires);
686 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
691 if (atomic_read(&rth->dst.__refcnt))
694 age = jiffies - rth->dst.lastuse;
695 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
696 (age <= tmo2 && rt_valuable(rth)))
702 /* Bits of score are:
704 * 30: not quite useless
705 * 29..0: usage counter
707 static inline u32 rt_score(struct rtable *rt)
709 u32 score = jiffies - rt->dst.lastuse;
711 score = ~score & ~(3<<30);
716 if (rt_is_output_route(rt) ||
717 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
723 static inline bool rt_caching(const struct net *net)
725 return net->ipv4.current_rt_cache_rebuild_count <=
726 net->ipv4.sysctl_rt_cache_rebuild_count;
729 static inline bool compare_hash_inputs(const struct rtable *rt1,
730 const struct rtable *rt2)
732 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
733 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
734 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
737 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
739 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
740 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
741 (rt1->rt_mark ^ rt2->rt_mark) |
742 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
743 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
744 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
747 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
749 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
752 static inline int rt_is_expired(struct rtable *rth)
754 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
758 * Perform a full scan of hash table and free all entries.
759 * Can be called by a softirq or a process.
760 * In the later case, we want to be reschedule if necessary
762 static void rt_do_flush(struct net *net, int process_context)
765 struct rtable *rth, *next;
767 for (i = 0; i <= rt_hash_mask; i++) {
768 struct rtable __rcu **pprev;
771 if (process_context && need_resched())
773 rth = rcu_access_pointer(rt_hash_table[i].chain);
777 spin_lock_bh(rt_hash_lock_addr(i));
780 pprev = &rt_hash_table[i].chain;
781 rth = rcu_dereference_protected(*pprev,
782 lockdep_is_held(rt_hash_lock_addr(i)));
785 next = rcu_dereference_protected(rth->dst.rt_next,
786 lockdep_is_held(rt_hash_lock_addr(i)));
789 net_eq(dev_net(rth->dst.dev), net)) {
790 rcu_assign_pointer(*pprev, next);
791 rcu_assign_pointer(rth->dst.rt_next, list);
794 pprev = &rth->dst.rt_next;
799 spin_unlock_bh(rt_hash_lock_addr(i));
801 for (; list; list = next) {
802 next = rcu_dereference_protected(list->dst.rt_next, 1);
809 * While freeing expired entries, we compute average chain length
810 * and standard deviation, using fixed-point arithmetic.
811 * This to have an estimation of rt_chain_length_max
812 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
813 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
817 #define ONE (1UL << FRACT_BITS)
820 * Given a hash chain and an item in this hash chain,
821 * find if a previous entry has the same hash_inputs
822 * (but differs on tos, mark or oif)
823 * Returns 0 if an alias is found.
824 * Returns ONE if rth has no alias before itself.
826 static int has_noalias(const struct rtable *head, const struct rtable *rth)
828 const struct rtable *aux = head;
831 if (compare_hash_inputs(aux, rth))
833 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
838 static void rt_check_expire(void)
840 static unsigned int rover;
841 unsigned int i = rover, goal;
843 struct rtable __rcu **rthp;
844 unsigned long samples = 0;
845 unsigned long sum = 0, sum2 = 0;
849 delta = jiffies - expires_ljiffies;
850 expires_ljiffies = jiffies;
851 mult = ((u64)delta) << rt_hash_log;
852 if (ip_rt_gc_timeout > 1)
853 do_div(mult, ip_rt_gc_timeout);
854 goal = (unsigned int)mult;
855 if (goal > rt_hash_mask)
856 goal = rt_hash_mask + 1;
857 for (; goal > 0; goal--) {
858 unsigned long tmo = ip_rt_gc_timeout;
859 unsigned long length;
861 i = (i + 1) & rt_hash_mask;
862 rthp = &rt_hash_table[i].chain;
869 if (rcu_dereference_raw(*rthp) == NULL)
872 spin_lock_bh(rt_hash_lock_addr(i));
873 while ((rth = rcu_dereference_protected(*rthp,
874 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
875 prefetch(rth->dst.rt_next);
876 if (rt_is_expired(rth)) {
877 *rthp = rth->dst.rt_next;
881 if (rth->dst.expires) {
882 /* Entry is expired even if it is in use */
883 if (time_before_eq(jiffies, rth->dst.expires)) {
886 rthp = &rth->dst.rt_next;
888 * We only count entries on
889 * a chain with equal hash inputs once
890 * so that entries for different QOS
891 * levels, and other non-hash input
892 * attributes don't unfairly skew
893 * the length computation
895 length += has_noalias(rt_hash_table[i].chain, rth);
898 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
901 /* Cleanup aged off entries. */
902 *rthp = rth->dst.rt_next;
905 spin_unlock_bh(rt_hash_lock_addr(i));
907 sum2 += length*length;
910 unsigned long avg = sum / samples;
911 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
912 rt_chain_length_max = max_t(unsigned long,
914 (avg + 4*sd) >> FRACT_BITS);
920 * rt_worker_func() is run in process context.
921 * we call rt_check_expire() to scan part of the hash table
923 static void rt_worker_func(struct work_struct *work)
926 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
930 * Perturbation of rt_genid by a small quantity [1..256]
931 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
932 * many times (2^24) without giving recent rt_genid.
933 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
935 static void rt_cache_invalidate(struct net *net)
937 unsigned char shuffle;
939 get_random_bytes(&shuffle, sizeof(shuffle));
940 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
945 * delay < 0 : invalidate cache (fast : entries will be deleted later)
946 * delay >= 0 : invalidate & flush cache (can be long)
948 void rt_cache_flush(struct net *net, int delay)
950 rt_cache_invalidate(net);
952 rt_do_flush(net, !in_softirq());
955 /* Flush previous cache invalidated entries from the cache */
956 void rt_cache_flush_batch(struct net *net)
958 rt_do_flush(net, !in_softirq());
961 static void rt_emergency_hash_rebuild(struct net *net)
964 printk(KERN_WARNING "Route hash chain too long!\n");
965 rt_cache_invalidate(net);
969 Short description of GC goals.
971 We want to build algorithm, which will keep routing cache
972 at some equilibrium point, when number of aged off entries
973 is kept approximately equal to newly generated ones.
975 Current expiration strength is variable "expire".
976 We try to adjust it dynamically, so that if networking
977 is idle expires is large enough to keep enough of warm entries,
978 and when load increases it reduces to limit cache size.
981 static int rt_garbage_collect(struct dst_ops *ops)
983 static unsigned long expire = RT_GC_TIMEOUT;
984 static unsigned long last_gc;
986 static int equilibrium;
988 struct rtable __rcu **rthp;
989 unsigned long now = jiffies;
991 int entries = dst_entries_get_fast(&ipv4_dst_ops);
994 * Garbage collection is pretty expensive,
995 * do not make it too frequently.
998 RT_CACHE_STAT_INC(gc_total);
1000 if (now - last_gc < ip_rt_gc_min_interval &&
1001 entries < ip_rt_max_size) {
1002 RT_CACHE_STAT_INC(gc_ignored);
1006 entries = dst_entries_get_slow(&ipv4_dst_ops);
1007 /* Calculate number of entries, which we want to expire now. */
1008 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
1010 if (equilibrium < ipv4_dst_ops.gc_thresh)
1011 equilibrium = ipv4_dst_ops.gc_thresh;
1012 goal = entries - equilibrium;
1014 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1015 goal = entries - equilibrium;
1018 /* We are in dangerous area. Try to reduce cache really
1021 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1022 equilibrium = entries - goal;
1025 if (now - last_gc >= ip_rt_gc_min_interval)
1029 equilibrium += goal;
1036 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1037 unsigned long tmo = expire;
1039 k = (k + 1) & rt_hash_mask;
1040 rthp = &rt_hash_table[k].chain;
1041 spin_lock_bh(rt_hash_lock_addr(k));
1042 while ((rth = rcu_dereference_protected(*rthp,
1043 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1044 if (!rt_is_expired(rth) &&
1045 !rt_may_expire(rth, tmo, expire)) {
1047 rthp = &rth->dst.rt_next;
1050 *rthp = rth->dst.rt_next;
1054 spin_unlock_bh(rt_hash_lock_addr(k));
1063 /* Goal is not achieved. We stop process if:
1065 - if expire reduced to zero. Otherwise, expire is halfed.
1066 - if table is not full.
1067 - if we are called from interrupt.
1068 - jiffies check is just fallback/debug loop breaker.
1069 We will not spin here for long time in any case.
1072 RT_CACHE_STAT_INC(gc_goal_miss);
1079 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1081 } while (!in_softirq() && time_before_eq(jiffies, now));
1083 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1085 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1087 if (net_ratelimit())
1088 printk(KERN_WARNING "dst cache overflow\n");
1089 RT_CACHE_STAT_INC(gc_dst_overflow);
1093 expire += ip_rt_gc_min_interval;
1094 if (expire > ip_rt_gc_timeout ||
1095 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1096 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1097 expire = ip_rt_gc_timeout;
1102 * Returns number of entries in a hash chain that have different hash_inputs
1104 static int slow_chain_length(const struct rtable *head)
1107 const struct rtable *rth = head;
1110 length += has_noalias(head, rth);
1111 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1113 return length >> FRACT_BITS;
1116 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1118 struct neigh_table *tbl = &arp_tbl;
1119 static const __be32 inaddr_any = 0;
1120 struct net_device *dev = dst->dev;
1121 const __be32 *pkey = daddr;
1122 struct neighbour *n;
1124 #if defined(CONFIG_ATM_CLIP) || defined(CONFIG_ATM_CLIP_MODULE)
1125 if (dev->type == ARPHRD_ATM)
1126 tbl = clip_tbl_hook;
1128 if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1131 n = __ipv4_neigh_lookup(tbl, dev, *(__force u32 *)pkey);
1134 return neigh_create(tbl, pkey, dev);
1137 static int rt_bind_neighbour(struct rtable *rt)
1139 struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1142 dst_set_neighbour(&rt->dst, n);
1147 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1148 struct sk_buff *skb, int ifindex)
1150 struct rtable *rth, *cand;
1151 struct rtable __rcu **rthp, **candp;
1155 int attempts = !in_softirq();
1159 min_score = ~(u32)0;
1164 if (!rt_caching(dev_net(rt->dst.dev))) {
1166 * If we're not caching, just tell the caller we
1167 * were successful and don't touch the route. The
1168 * caller hold the sole reference to the cache entry, and
1169 * it will be released when the caller is done with it.
1170 * If we drop it here, the callers have no way to resolve routes
1171 * when we're not caching. Instead, just point *rp at rt, so
1172 * the caller gets a single use out of the route
1173 * Note that we do rt_free on this new route entry, so that
1174 * once its refcount hits zero, we are still able to reap it
1176 * Note: To avoid expensive rcu stuff for this uncached dst,
1177 * we set DST_NOCACHE so that dst_release() can free dst without
1178 * waiting a grace period.
1181 rt->dst.flags |= DST_NOCACHE;
1182 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1183 int err = rt_bind_neighbour(rt);
1185 if (net_ratelimit())
1187 "Neighbour table failure & not caching routes.\n");
1189 return ERR_PTR(err);
1196 rthp = &rt_hash_table[hash].chain;
1198 spin_lock_bh(rt_hash_lock_addr(hash));
1199 while ((rth = rcu_dereference_protected(*rthp,
1200 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1201 if (rt_is_expired(rth)) {
1202 *rthp = rth->dst.rt_next;
1206 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1208 *rthp = rth->dst.rt_next;
1210 * Since lookup is lockfree, the deletion
1211 * must be visible to another weakly ordered CPU before
1212 * the insertion at the start of the hash chain.
1214 rcu_assign_pointer(rth->dst.rt_next,
1215 rt_hash_table[hash].chain);
1217 * Since lookup is lockfree, the update writes
1218 * must be ordered for consistency on SMP.
1220 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1222 dst_use(&rth->dst, now);
1223 spin_unlock_bh(rt_hash_lock_addr(hash));
1227 skb_dst_set(skb, &rth->dst);
1231 if (!atomic_read(&rth->dst.__refcnt)) {
1232 u32 score = rt_score(rth);
1234 if (score <= min_score) {
1243 rthp = &rth->dst.rt_next;
1247 /* ip_rt_gc_elasticity used to be average length of chain
1248 * length, when exceeded gc becomes really aggressive.
1250 * The second limit is less certain. At the moment it allows
1251 * only 2 entries per bucket. We will see.
1253 if (chain_length > ip_rt_gc_elasticity) {
1254 *candp = cand->dst.rt_next;
1258 if (chain_length > rt_chain_length_max &&
1259 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1260 struct net *net = dev_net(rt->dst.dev);
1261 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1262 if (!rt_caching(net)) {
1263 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1264 rt->dst.dev->name, num);
1266 rt_emergency_hash_rebuild(net);
1267 spin_unlock_bh(rt_hash_lock_addr(hash));
1269 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1270 ifindex, rt_genid(net));
1275 /* Try to bind route to arp only if it is output
1276 route or unicast forwarding path.
1278 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1279 int err = rt_bind_neighbour(rt);
1281 spin_unlock_bh(rt_hash_lock_addr(hash));
1283 if (err != -ENOBUFS) {
1285 return ERR_PTR(err);
1288 /* Neighbour tables are full and nothing
1289 can be released. Try to shrink route cache,
1290 it is most likely it holds some neighbour records.
1292 if (attempts-- > 0) {
1293 int saved_elasticity = ip_rt_gc_elasticity;
1294 int saved_int = ip_rt_gc_min_interval;
1295 ip_rt_gc_elasticity = 1;
1296 ip_rt_gc_min_interval = 0;
1297 rt_garbage_collect(&ipv4_dst_ops);
1298 ip_rt_gc_min_interval = saved_int;
1299 ip_rt_gc_elasticity = saved_elasticity;
1303 if (net_ratelimit())
1304 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1306 return ERR_PTR(-ENOBUFS);
1310 rt->dst.rt_next = rt_hash_table[hash].chain;
1313 * Since lookup is lockfree, we must make sure
1314 * previous writes to rt are committed to memory
1315 * before making rt visible to other CPUS.
1317 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1319 spin_unlock_bh(rt_hash_lock_addr(hash));
1323 skb_dst_set(skb, &rt->dst);
1327 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1329 static u32 rt_peer_genid(void)
1331 return atomic_read(&__rt_peer_genid);
1334 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1336 struct inet_peer *peer;
1338 peer = inet_getpeer_v4(daddr, create);
1340 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1343 rt->rt_peer_genid = rt_peer_genid();
1347 * Peer allocation may fail only in serious out-of-memory conditions. However
1348 * we still can generate some output.
1349 * Random ID selection looks a bit dangerous because we have no chances to
1350 * select ID being unique in a reasonable period of time.
1351 * But broken packet identifier may be better than no packet at all.
1353 static void ip_select_fb_ident(struct iphdr *iph)
1355 static DEFINE_SPINLOCK(ip_fb_id_lock);
1356 static u32 ip_fallback_id;
1359 spin_lock_bh(&ip_fb_id_lock);
1360 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1361 iph->id = htons(salt & 0xFFFF);
1362 ip_fallback_id = salt;
1363 spin_unlock_bh(&ip_fb_id_lock);
1366 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1368 struct rtable *rt = (struct rtable *) dst;
1370 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1371 if (rt->peer == NULL)
1372 rt_bind_peer(rt, rt->rt_dst, 1);
1374 /* If peer is attached to destination, it is never detached,
1375 so that we need not to grab a lock to dereference it.
1378 iph->id = htons(inet_getid(rt->peer, more));
1382 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1383 __builtin_return_address(0));
1385 ip_select_fb_ident(iph);
1387 EXPORT_SYMBOL(__ip_select_ident);
1389 static void rt_del(unsigned hash, struct rtable *rt)
1391 struct rtable __rcu **rthp;
1394 rthp = &rt_hash_table[hash].chain;
1395 spin_lock_bh(rt_hash_lock_addr(hash));
1397 while ((aux = rcu_dereference_protected(*rthp,
1398 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1399 if (aux == rt || rt_is_expired(aux)) {
1400 *rthp = aux->dst.rt_next;
1404 rthp = &aux->dst.rt_next;
1406 spin_unlock_bh(rt_hash_lock_addr(hash));
1409 static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1411 struct rtable *rt = (struct rtable *) dst;
1412 __be32 orig_gw = rt->rt_gateway;
1413 struct neighbour *n, *old_n;
1415 dst_confirm(&rt->dst);
1417 rt->rt_gateway = peer->redirect_learned.a4;
1419 n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1421 rt->rt_gateway = orig_gw;
1424 old_n = xchg(&rt->dst._neighbour, n);
1426 neigh_release(old_n);
1427 if (!(n->nud_state & NUD_VALID)) {
1428 neigh_event_send(n, NULL);
1430 rt->rt_flags |= RTCF_REDIRECTED;
1431 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1435 /* called in rcu_read_lock() section */
1436 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1437 __be32 saddr, struct net_device *dev)
1440 struct in_device *in_dev = __in_dev_get_rcu(dev);
1441 __be32 skeys[2] = { saddr, 0 };
1442 int ikeys[2] = { dev->ifindex, 0 };
1443 struct inet_peer *peer;
1450 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1451 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1452 ipv4_is_zeronet(new_gw))
1453 goto reject_redirect;
1455 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1456 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1457 goto reject_redirect;
1458 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1459 goto reject_redirect;
1461 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1462 goto reject_redirect;
1465 for (s = 0; s < 2; s++) {
1466 for (i = 0; i < 2; i++) {
1468 struct rtable __rcu **rthp;
1471 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1473 rthp = &rt_hash_table[hash].chain;
1475 while ((rt = rcu_dereference(*rthp)) != NULL) {
1476 rthp = &rt->dst.rt_next;
1478 if (rt->rt_key_dst != daddr ||
1479 rt->rt_key_src != skeys[s] ||
1480 rt->rt_oif != ikeys[i] ||
1481 rt_is_input_route(rt) ||
1482 rt_is_expired(rt) ||
1483 !net_eq(dev_net(rt->dst.dev), net) ||
1485 rt->dst.dev != dev ||
1486 rt->rt_gateway != old_gw)
1490 rt_bind_peer(rt, rt->rt_dst, 1);
1494 if (peer->redirect_learned.a4 != new_gw ||
1495 peer->redirect_genid != redirect_genid) {
1496 peer->redirect_learned.a4 = new_gw;
1497 peer->redirect_genid = redirect_genid;
1498 atomic_inc(&__rt_peer_genid);
1500 check_peer_redir(&rt->dst, peer);
1508 #ifdef CONFIG_IP_ROUTE_VERBOSE
1509 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1510 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1511 " Advised path = %pI4 -> %pI4\n",
1512 &old_gw, dev->name, &new_gw,
1518 static bool peer_pmtu_expired(struct inet_peer *peer)
1520 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1523 time_after_eq(jiffies, orig) &&
1524 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1527 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1529 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1532 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1535 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1537 struct rtable *rt = (struct rtable *)dst;
1538 struct dst_entry *ret = dst;
1541 if (dst->obsolete > 0) {
1544 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1545 unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1547 rt_genid(dev_net(dst->dev)));
1550 } else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1551 dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1559 * 1. The first ip_rt_redirect_number redirects are sent
1560 * with exponential backoff, then we stop sending them at all,
1561 * assuming that the host ignores our redirects.
1562 * 2. If we did not see packets requiring redirects
1563 * during ip_rt_redirect_silence, we assume that the host
1564 * forgot redirected route and start to send redirects again.
1566 * This algorithm is much cheaper and more intelligent than dumb load limiting
1569 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1570 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1573 void ip_rt_send_redirect(struct sk_buff *skb)
1575 struct rtable *rt = skb_rtable(skb);
1576 struct in_device *in_dev;
1577 struct inet_peer *peer;
1581 in_dev = __in_dev_get_rcu(rt->dst.dev);
1582 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1586 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1590 rt_bind_peer(rt, rt->rt_dst, 1);
1593 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1597 /* No redirected packets during ip_rt_redirect_silence;
1598 * reset the algorithm.
1600 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1601 peer->rate_tokens = 0;
1603 /* Too many ignored redirects; do not send anything
1604 * set dst.rate_last to the last seen redirected packet.
1606 if (peer->rate_tokens >= ip_rt_redirect_number) {
1607 peer->rate_last = jiffies;
1611 /* Check for load limit; set rate_last to the latest sent
1614 if (peer->rate_tokens == 0 ||
1617 (ip_rt_redirect_load << peer->rate_tokens)))) {
1618 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1619 peer->rate_last = jiffies;
1620 ++peer->rate_tokens;
1621 #ifdef CONFIG_IP_ROUTE_VERBOSE
1623 peer->rate_tokens == ip_rt_redirect_number &&
1625 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1626 &ip_hdr(skb)->saddr, rt->rt_iif,
1627 &rt->rt_dst, &rt->rt_gateway);
1632 static int ip_error(struct sk_buff *skb)
1634 struct rtable *rt = skb_rtable(skb);
1635 struct inet_peer *peer;
1640 switch (rt->dst.error) {
1645 code = ICMP_HOST_UNREACH;
1648 code = ICMP_NET_UNREACH;
1649 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1650 IPSTATS_MIB_INNOROUTES);
1653 code = ICMP_PKT_FILTERED;
1658 rt_bind_peer(rt, rt->rt_dst, 1);
1664 peer->rate_tokens += now - peer->rate_last;
1665 if (peer->rate_tokens > ip_rt_error_burst)
1666 peer->rate_tokens = ip_rt_error_burst;
1667 peer->rate_last = now;
1668 if (peer->rate_tokens >= ip_rt_error_cost)
1669 peer->rate_tokens -= ip_rt_error_cost;
1674 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1676 out: kfree_skb(skb);
1681 * The last two values are not from the RFC but
1682 * are needed for AMPRnet AX.25 paths.
1685 static const unsigned short mtu_plateau[] =
1686 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1688 static inline unsigned short guess_mtu(unsigned short old_mtu)
1692 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1693 if (old_mtu > mtu_plateau[i])
1694 return mtu_plateau[i];
1698 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1699 unsigned short new_mtu,
1700 struct net_device *dev)
1702 unsigned short old_mtu = ntohs(iph->tot_len);
1703 unsigned short est_mtu = 0;
1704 struct inet_peer *peer;
1706 peer = inet_getpeer_v4(iph->daddr, 1);
1708 unsigned short mtu = new_mtu;
1710 if (new_mtu < 68 || new_mtu >= old_mtu) {
1711 /* BSD 4.2 derived systems incorrectly adjust
1712 * tot_len by the IP header length, and report
1713 * a zero MTU in the ICMP message.
1716 old_mtu >= 68 + (iph->ihl << 2))
1717 old_mtu -= iph->ihl << 2;
1718 mtu = guess_mtu(old_mtu);
1721 if (mtu < ip_rt_min_pmtu)
1722 mtu = ip_rt_min_pmtu;
1723 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1724 unsigned long pmtu_expires;
1726 pmtu_expires = jiffies + ip_rt_mtu_expires;
1731 peer->pmtu_learned = mtu;
1732 peer->pmtu_expires = pmtu_expires;
1733 atomic_inc(&__rt_peer_genid);
1738 return est_mtu ? : new_mtu;
1741 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1743 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1747 if (time_before(jiffies, expires)) {
1748 u32 orig_dst_mtu = dst_mtu(dst);
1749 if (peer->pmtu_learned < orig_dst_mtu) {
1750 if (!peer->pmtu_orig)
1751 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1752 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1754 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1755 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1758 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1760 struct rtable *rt = (struct rtable *) dst;
1761 struct inet_peer *peer;
1766 rt_bind_peer(rt, rt->rt_dst, 1);
1769 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1771 if (mtu < ip_rt_min_pmtu)
1772 mtu = ip_rt_min_pmtu;
1773 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1775 pmtu_expires = jiffies + ip_rt_mtu_expires;
1779 peer->pmtu_learned = mtu;
1780 peer->pmtu_expires = pmtu_expires;
1782 atomic_inc(&__rt_peer_genid);
1783 rt->rt_peer_genid = rt_peer_genid();
1785 check_peer_pmtu(dst, peer);
1790 static void ipv4_validate_peer(struct rtable *rt)
1792 if (rt->rt_peer_genid != rt_peer_genid()) {
1793 struct inet_peer *peer;
1796 rt_bind_peer(rt, rt->rt_dst, 0);
1800 check_peer_pmtu(&rt->dst, peer);
1802 if (peer->redirect_genid != redirect_genid)
1803 peer->redirect_learned.a4 = 0;
1804 if (peer->redirect_learned.a4 &&
1805 peer->redirect_learned.a4 != rt->rt_gateway)
1806 check_peer_redir(&rt->dst, peer);
1809 rt->rt_peer_genid = rt_peer_genid();
1813 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1815 struct rtable *rt = (struct rtable *) dst;
1817 if (rt_is_expired(rt))
1819 ipv4_validate_peer(rt);
1823 static void ipv4_dst_destroy(struct dst_entry *dst)
1825 struct rtable *rt = (struct rtable *) dst;
1826 struct inet_peer *peer = rt->peer;
1829 fib_info_put(rt->fi);
1839 static void ipv4_link_failure(struct sk_buff *skb)
1843 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1845 rt = skb_rtable(skb);
1846 if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1847 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1850 static int ip_rt_bug(struct sk_buff *skb)
1852 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1853 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1854 skb->dev ? skb->dev->name : "?");
1861 We do not cache source address of outgoing interface,
1862 because it is used only by IP RR, TS and SRR options,
1863 so that it out of fast path.
1865 BTW remember: "addr" is allowed to be not aligned
1869 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1873 if (rt_is_output_route(rt))
1874 src = ip_hdr(skb)->saddr;
1876 struct fib_result res;
1882 memset(&fl4, 0, sizeof(fl4));
1883 fl4.daddr = iph->daddr;
1884 fl4.saddr = iph->saddr;
1885 fl4.flowi4_tos = RT_TOS(iph->tos);
1886 fl4.flowi4_oif = rt->dst.dev->ifindex;
1887 fl4.flowi4_iif = skb->dev->ifindex;
1888 fl4.flowi4_mark = skb->mark;
1891 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1892 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1894 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1898 memcpy(addr, &src, 4);
1901 #ifdef CONFIG_IP_ROUTE_CLASSID
1902 static void set_class_tag(struct rtable *rt, u32 tag)
1904 if (!(rt->dst.tclassid & 0xFFFF))
1905 rt->dst.tclassid |= tag & 0xFFFF;
1906 if (!(rt->dst.tclassid & 0xFFFF0000))
1907 rt->dst.tclassid |= tag & 0xFFFF0000;
1911 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1913 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1916 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1918 if (advmss > 65535 - 40)
1919 advmss = 65535 - 40;
1924 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1926 const struct rtable *rt = (const struct rtable *) dst;
1927 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1929 if (mtu && rt_is_output_route(rt))
1932 mtu = dst->dev->mtu;
1934 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1936 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1940 if (mtu > IP_MAX_MTU)
1946 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1947 struct fib_info *fi)
1949 struct inet_peer *peer;
1952 /* If a peer entry exists for this destination, we must hook
1953 * it up in order to get at cached metrics.
1955 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1958 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1960 rt->rt_peer_genid = rt_peer_genid();
1961 if (inet_metrics_new(peer))
1962 memcpy(peer->metrics, fi->fib_metrics,
1963 sizeof(u32) * RTAX_MAX);
1964 dst_init_metrics(&rt->dst, peer->metrics, false);
1966 check_peer_pmtu(&rt->dst, peer);
1967 if (peer->redirect_genid != redirect_genid)
1968 peer->redirect_learned.a4 = 0;
1969 if (peer->redirect_learned.a4 &&
1970 peer->redirect_learned.a4 != rt->rt_gateway) {
1971 rt->rt_gateway = peer->redirect_learned.a4;
1972 rt->rt_flags |= RTCF_REDIRECTED;
1975 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1977 atomic_inc(&fi->fib_clntref);
1979 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1983 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1984 const struct fib_result *res,
1985 struct fib_info *fi, u16 type, u32 itag)
1987 struct dst_entry *dst = &rt->dst;
1990 if (FIB_RES_GW(*res) &&
1991 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1992 rt->rt_gateway = FIB_RES_GW(*res);
1993 rt_init_metrics(rt, fl4, fi);
1994 #ifdef CONFIG_IP_ROUTE_CLASSID
1995 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1999 if (dst_mtu(dst) > IP_MAX_MTU)
2000 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
2001 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
2002 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
2004 #ifdef CONFIG_IP_ROUTE_CLASSID
2005 #ifdef CONFIG_IP_MULTIPLE_TABLES
2006 set_class_tag(rt, fib_rules_tclass(res));
2008 set_class_tag(rt, itag);
2012 static struct rtable *rt_dst_alloc(struct net_device *dev,
2013 bool nopolicy, bool noxfrm)
2015 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
2017 (nopolicy ? DST_NOPOLICY : 0) |
2018 (noxfrm ? DST_NOXFRM : 0));
2021 /* called in rcu_read_lock() section */
2022 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2023 u8 tos, struct net_device *dev, int our)
2028 struct in_device *in_dev = __in_dev_get_rcu(dev);
2032 /* Primary sanity checks. */
2037 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2038 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
2041 if (ipv4_is_zeronet(saddr)) {
2042 if (!ipv4_is_local_multicast(daddr))
2044 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2046 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2051 rth = rt_dst_alloc(init_net.loopback_dev,
2052 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2056 #ifdef CONFIG_IP_ROUTE_CLASSID
2057 rth->dst.tclassid = itag;
2059 rth->dst.output = ip_rt_bug;
2061 rth->rt_key_dst = daddr;
2062 rth->rt_key_src = saddr;
2063 rth->rt_genid = rt_genid(dev_net(dev));
2064 rth->rt_flags = RTCF_MULTICAST;
2065 rth->rt_type = RTN_MULTICAST;
2066 rth->rt_key_tos = tos;
2067 rth->rt_dst = daddr;
2068 rth->rt_src = saddr;
2069 rth->rt_route_iif = dev->ifindex;
2070 rth->rt_iif = dev->ifindex;
2072 rth->rt_mark = skb->mark;
2073 rth->rt_gateway = daddr;
2074 rth->rt_spec_dst= spec_dst;
2075 rth->rt_peer_genid = 0;
2079 rth->dst.input= ip_local_deliver;
2080 rth->rt_flags |= RTCF_LOCAL;
2083 #ifdef CONFIG_IP_MROUTE
2084 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2085 rth->dst.input = ip_mr_input;
2087 RT_CACHE_STAT_INC(in_slow_mc);
2089 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2090 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2091 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2102 static void ip_handle_martian_source(struct net_device *dev,
2103 struct in_device *in_dev,
2104 struct sk_buff *skb,
2108 RT_CACHE_STAT_INC(in_martian_src);
2109 #ifdef CONFIG_IP_ROUTE_VERBOSE
2110 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2112 * RFC1812 recommendation, if source is martian,
2113 * the only hint is MAC header.
2115 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
2116 &daddr, &saddr, dev->name);
2117 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2119 const unsigned char *p = skb_mac_header(skb);
2120 printk(KERN_WARNING "ll header: ");
2121 for (i = 0; i < dev->hard_header_len; i++, p++) {
2123 if (i < (dev->hard_header_len - 1))
2132 /* called in rcu_read_lock() section */
2133 static int __mkroute_input(struct sk_buff *skb,
2134 const struct fib_result *res,
2135 struct in_device *in_dev,
2136 __be32 daddr, __be32 saddr, u32 tos,
2137 struct rtable **result)
2141 struct in_device *out_dev;
2142 unsigned int flags = 0;
2146 /* get a working reference to the output device */
2147 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2148 if (out_dev == NULL) {
2149 if (net_ratelimit())
2150 printk(KERN_CRIT "Bug in ip_route_input" \
2151 "_slow(). Please, report\n");
2156 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2157 in_dev->dev, &spec_dst, &itag);
2159 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2166 flags |= RTCF_DIRECTSRC;
2168 if (out_dev == in_dev && err &&
2169 (IN_DEV_SHARED_MEDIA(out_dev) ||
2170 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2171 flags |= RTCF_DOREDIRECT;
2173 if (skb->protocol != htons(ETH_P_IP)) {
2174 /* Not IP (i.e. ARP). Do not create route, if it is
2175 * invalid for proxy arp. DNAT routes are always valid.
2177 * Proxy arp feature have been extended to allow, ARP
2178 * replies back to the same interface, to support
2179 * Private VLAN switch technologies. See arp.c.
2181 if (out_dev == in_dev &&
2182 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2188 rth = rt_dst_alloc(out_dev->dev,
2189 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2190 IN_DEV_CONF_GET(out_dev, NOXFRM));
2196 rth->rt_key_dst = daddr;
2197 rth->rt_key_src = saddr;
2198 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2199 rth->rt_flags = flags;
2200 rth->rt_type = res->type;
2201 rth->rt_key_tos = tos;
2202 rth->rt_dst = daddr;
2203 rth->rt_src = saddr;
2204 rth->rt_route_iif = in_dev->dev->ifindex;
2205 rth->rt_iif = in_dev->dev->ifindex;
2207 rth->rt_mark = skb->mark;
2208 rth->rt_gateway = daddr;
2209 rth->rt_spec_dst= spec_dst;
2210 rth->rt_peer_genid = 0;
2214 rth->dst.input = ip_forward;
2215 rth->dst.output = ip_output;
2217 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2225 static int ip_mkroute_input(struct sk_buff *skb,
2226 struct fib_result *res,
2227 const struct flowi4 *fl4,
2228 struct in_device *in_dev,
2229 __be32 daddr, __be32 saddr, u32 tos)
2231 struct rtable* rth = NULL;
2235 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2236 if (res->fi && res->fi->fib_nhs > 1)
2237 fib_select_multipath(res);
2240 /* create a routing cache entry */
2241 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2245 /* put it into the cache */
2246 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2247 rt_genid(dev_net(rth->dst.dev)));
2248 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2250 return PTR_ERR(rth);
2255 * NOTE. We drop all the packets that has local source
2256 * addresses, because every properly looped back packet
2257 * must have correct destination already attached by output routine.
2259 * Such approach solves two big problems:
2260 * 1. Not simplex devices are handled properly.
2261 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2262 * called with rcu_read_lock()
2265 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2266 u8 tos, struct net_device *dev)
2268 struct fib_result res;
2269 struct in_device *in_dev = __in_dev_get_rcu(dev);
2273 struct rtable * rth;
2277 struct net * net = dev_net(dev);
2279 /* IP on this device is disabled. */
2284 /* Check for the most weird martians, which can be not detected
2288 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2289 ipv4_is_loopback(saddr))
2290 goto martian_source;
2292 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2295 /* Accept zero addresses only to limited broadcast;
2296 * I even do not know to fix it or not. Waiting for complains :-)
2298 if (ipv4_is_zeronet(saddr))
2299 goto martian_source;
2301 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2302 goto martian_destination;
2305 * Now we are ready to route packet.
2308 fl4.flowi4_iif = dev->ifindex;
2309 fl4.flowi4_mark = skb->mark;
2310 fl4.flowi4_tos = tos;
2311 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2314 err = fib_lookup(net, &fl4, &res);
2316 if (!IN_DEV_FORWARD(in_dev))
2321 RT_CACHE_STAT_INC(in_slow_tot);
2323 if (res.type == RTN_BROADCAST)
2326 if (res.type == RTN_LOCAL) {
2327 err = fib_validate_source(skb, saddr, daddr, tos,
2328 net->loopback_dev->ifindex,
2329 dev, &spec_dst, &itag);
2331 goto martian_source_keep_err;
2333 flags |= RTCF_DIRECTSRC;
2338 if (!IN_DEV_FORWARD(in_dev))
2340 if (res.type != RTN_UNICAST)
2341 goto martian_destination;
2343 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2347 if (skb->protocol != htons(ETH_P_IP))
2350 if (ipv4_is_zeronet(saddr))
2351 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2353 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2356 goto martian_source_keep_err;
2358 flags |= RTCF_DIRECTSRC;
2360 flags |= RTCF_BROADCAST;
2361 res.type = RTN_BROADCAST;
2362 RT_CACHE_STAT_INC(in_brd);
2365 rth = rt_dst_alloc(net->loopback_dev,
2366 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2370 rth->dst.input= ip_local_deliver;
2371 rth->dst.output= ip_rt_bug;
2372 #ifdef CONFIG_IP_ROUTE_CLASSID
2373 rth->dst.tclassid = itag;
2376 rth->rt_key_dst = daddr;
2377 rth->rt_key_src = saddr;
2378 rth->rt_genid = rt_genid(net);
2379 rth->rt_flags = flags|RTCF_LOCAL;
2380 rth->rt_type = res.type;
2381 rth->rt_key_tos = tos;
2382 rth->rt_dst = daddr;
2383 rth->rt_src = saddr;
2384 #ifdef CONFIG_IP_ROUTE_CLASSID
2385 rth->dst.tclassid = itag;
2387 rth->rt_route_iif = dev->ifindex;
2388 rth->rt_iif = dev->ifindex;
2390 rth->rt_mark = skb->mark;
2391 rth->rt_gateway = daddr;
2392 rth->rt_spec_dst= spec_dst;
2393 rth->rt_peer_genid = 0;
2396 if (res.type == RTN_UNREACHABLE) {
2397 rth->dst.input= ip_error;
2398 rth->dst.error= -err;
2399 rth->rt_flags &= ~RTCF_LOCAL;
2401 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2402 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2409 RT_CACHE_STAT_INC(in_no_route);
2410 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2411 res.type = RTN_UNREACHABLE;
2417 * Do not cache martian addresses: they should be logged (RFC1812)
2419 martian_destination:
2420 RT_CACHE_STAT_INC(in_martian_dst);
2421 #ifdef CONFIG_IP_ROUTE_VERBOSE
2422 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2423 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2424 &daddr, &saddr, dev->name);
2428 err = -EHOSTUNREACH;
2441 martian_source_keep_err:
2442 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2446 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2447 u8 tos, struct net_device *dev, bool noref)
2449 struct rtable * rth;
2451 int iif = dev->ifindex;
2459 if (!rt_caching(net))
2462 tos &= IPTOS_RT_MASK;
2463 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2465 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2466 rth = rcu_dereference(rth->dst.rt_next)) {
2467 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2468 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2469 (rth->rt_route_iif ^ iif) |
2470 (rth->rt_key_tos ^ tos)) == 0 &&
2471 rth->rt_mark == skb->mark &&
2472 net_eq(dev_net(rth->dst.dev), net) &&
2473 !rt_is_expired(rth)) {
2474 ipv4_validate_peer(rth);
2476 dst_use_noref(&rth->dst, jiffies);
2477 skb_dst_set_noref(skb, &rth->dst);
2479 dst_use(&rth->dst, jiffies);
2480 skb_dst_set(skb, &rth->dst);
2482 RT_CACHE_STAT_INC(in_hit);
2486 RT_CACHE_STAT_INC(in_hlist_search);
2490 /* Multicast recognition logic is moved from route cache to here.
2491 The problem was that too many Ethernet cards have broken/missing
2492 hardware multicast filters :-( As result the host on multicasting
2493 network acquires a lot of useless route cache entries, sort of
2494 SDR messages from all the world. Now we try to get rid of them.
2495 Really, provided software IP multicast filter is organized
2496 reasonably (at least, hashed), it does not result in a slowdown
2497 comparing with route cache reject entries.
2498 Note, that multicast routers are not affected, because
2499 route cache entry is created eventually.
2501 if (ipv4_is_multicast(daddr)) {
2502 struct in_device *in_dev = __in_dev_get_rcu(dev);
2505 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2506 ip_hdr(skb)->protocol);
2508 #ifdef CONFIG_IP_MROUTE
2510 (!ipv4_is_local_multicast(daddr) &&
2511 IN_DEV_MFORWARD(in_dev))
2514 int res = ip_route_input_mc(skb, daddr, saddr,
2523 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2527 EXPORT_SYMBOL(ip_route_input_common);
2529 /* called with rcu_read_lock() */
2530 static struct rtable *__mkroute_output(const struct fib_result *res,
2531 const struct flowi4 *fl4,
2532 __be32 orig_daddr, __be32 orig_saddr,
2533 int orig_oif, __u8 orig_rtos,
2534 struct net_device *dev_out,
2537 struct fib_info *fi = res->fi;
2538 struct in_device *in_dev;
2539 u16 type = res->type;
2542 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2543 return ERR_PTR(-EINVAL);
2545 if (ipv4_is_lbcast(fl4->daddr))
2546 type = RTN_BROADCAST;
2547 else if (ipv4_is_multicast(fl4->daddr))
2548 type = RTN_MULTICAST;
2549 else if (ipv4_is_zeronet(fl4->daddr))
2550 return ERR_PTR(-EINVAL);
2552 if (dev_out->flags & IFF_LOOPBACK)
2553 flags |= RTCF_LOCAL;
2555 in_dev = __in_dev_get_rcu(dev_out);
2557 return ERR_PTR(-EINVAL);
2559 if (type == RTN_BROADCAST) {
2560 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2562 } else if (type == RTN_MULTICAST) {
2563 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2564 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2566 flags &= ~RTCF_LOCAL;
2567 /* If multicast route do not exist use
2568 * default one, but do not gateway in this case.
2571 if (fi && res->prefixlen < 4)
2575 rth = rt_dst_alloc(dev_out,
2576 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2577 IN_DEV_CONF_GET(in_dev, NOXFRM));
2579 return ERR_PTR(-ENOBUFS);
2581 rth->dst.output = ip_output;
2583 rth->rt_key_dst = orig_daddr;
2584 rth->rt_key_src = orig_saddr;
2585 rth->rt_genid = rt_genid(dev_net(dev_out));
2586 rth->rt_flags = flags;
2587 rth->rt_type = type;
2588 rth->rt_key_tos = orig_rtos;
2589 rth->rt_dst = fl4->daddr;
2590 rth->rt_src = fl4->saddr;
2591 rth->rt_route_iif = 0;
2592 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2593 rth->rt_oif = orig_oif;
2594 rth->rt_mark = fl4->flowi4_mark;
2595 rth->rt_gateway = fl4->daddr;
2596 rth->rt_spec_dst= fl4->saddr;
2597 rth->rt_peer_genid = 0;
2601 RT_CACHE_STAT_INC(out_slow_tot);
2603 if (flags & RTCF_LOCAL) {
2604 rth->dst.input = ip_local_deliver;
2605 rth->rt_spec_dst = fl4->daddr;
2607 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2608 rth->rt_spec_dst = fl4->saddr;
2609 if (flags & RTCF_LOCAL &&
2610 !(dev_out->flags & IFF_LOOPBACK)) {
2611 rth->dst.output = ip_mc_output;
2612 RT_CACHE_STAT_INC(out_slow_mc);
2614 #ifdef CONFIG_IP_MROUTE
2615 if (type == RTN_MULTICAST) {
2616 if (IN_DEV_MFORWARD(in_dev) &&
2617 !ipv4_is_local_multicast(fl4->daddr)) {
2618 rth->dst.input = ip_mr_input;
2619 rth->dst.output = ip_mc_output;
2625 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2631 * Major route resolver routine.
2632 * called with rcu_read_lock();
2635 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2637 struct net_device *dev_out = NULL;
2638 __u8 tos = RT_FL_TOS(fl4);
2639 unsigned int flags = 0;
2640 struct fib_result res;
2647 #ifdef CONFIG_IP_MULTIPLE_TABLES
2651 orig_daddr = fl4->daddr;
2652 orig_saddr = fl4->saddr;
2653 orig_oif = fl4->flowi4_oif;
2655 fl4->flowi4_iif = net->loopback_dev->ifindex;
2656 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2657 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2658 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2662 rth = ERR_PTR(-EINVAL);
2663 if (ipv4_is_multicast(fl4->saddr) ||
2664 ipv4_is_lbcast(fl4->saddr) ||
2665 ipv4_is_zeronet(fl4->saddr))
2668 /* I removed check for oif == dev_out->oif here.
2669 It was wrong for two reasons:
2670 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2671 is assigned to multiple interfaces.
2672 2. Moreover, we are allowed to send packets with saddr
2673 of another iface. --ANK
2676 if (fl4->flowi4_oif == 0 &&
2677 (ipv4_is_multicast(fl4->daddr) ||
2678 ipv4_is_lbcast(fl4->daddr))) {
2679 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2680 dev_out = __ip_dev_find(net, fl4->saddr, false);
2681 if (dev_out == NULL)
2684 /* Special hack: user can direct multicasts
2685 and limited broadcast via necessary interface
2686 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2687 This hack is not just for fun, it allows
2688 vic,vat and friends to work.
2689 They bind socket to loopback, set ttl to zero
2690 and expect that it will work.
2691 From the viewpoint of routing cache they are broken,
2692 because we are not allowed to build multicast path
2693 with loopback source addr (look, routing cache
2694 cannot know, that ttl is zero, so that packet
2695 will not leave this host and route is valid).
2696 Luckily, this hack is good workaround.
2699 fl4->flowi4_oif = dev_out->ifindex;
2703 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2704 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2705 if (!__ip_dev_find(net, fl4->saddr, false))
2711 if (fl4->flowi4_oif) {
2712 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2713 rth = ERR_PTR(-ENODEV);
2714 if (dev_out == NULL)
2717 /* RACE: Check return value of inet_select_addr instead. */
2718 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2719 rth = ERR_PTR(-ENETUNREACH);
2722 if (ipv4_is_local_multicast(fl4->daddr) ||
2723 ipv4_is_lbcast(fl4->daddr)) {
2725 fl4->saddr = inet_select_addr(dev_out, 0,
2730 if (ipv4_is_multicast(fl4->daddr))
2731 fl4->saddr = inet_select_addr(dev_out, 0,
2733 else if (!fl4->daddr)
2734 fl4->saddr = inet_select_addr(dev_out, 0,
2740 fl4->daddr = fl4->saddr;
2742 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2743 dev_out = net->loopback_dev;
2744 fl4->flowi4_oif = net->loopback_dev->ifindex;
2745 res.type = RTN_LOCAL;
2746 flags |= RTCF_LOCAL;
2750 if (fib_lookup(net, fl4, &res)) {
2752 if (fl4->flowi4_oif) {
2753 /* Apparently, routing tables are wrong. Assume,
2754 that the destination is on link.
2757 Because we are allowed to send to iface
2758 even if it has NO routes and NO assigned
2759 addresses. When oif is specified, routing
2760 tables are looked up with only one purpose:
2761 to catch if destination is gatewayed, rather than
2762 direct. Moreover, if MSG_DONTROUTE is set,
2763 we send packet, ignoring both routing tables
2764 and ifaddr state. --ANK
2767 We could make it even if oif is unknown,
2768 likely IPv6, but we do not.
2771 if (fl4->saddr == 0)
2772 fl4->saddr = inet_select_addr(dev_out, 0,
2774 res.type = RTN_UNICAST;
2777 rth = ERR_PTR(-ENETUNREACH);
2781 if (res.type == RTN_LOCAL) {
2783 if (res.fi->fib_prefsrc)
2784 fl4->saddr = res.fi->fib_prefsrc;
2786 fl4->saddr = fl4->daddr;
2788 dev_out = net->loopback_dev;
2789 fl4->flowi4_oif = dev_out->ifindex;
2791 flags |= RTCF_LOCAL;
2795 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2796 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2797 fib_select_multipath(&res);
2800 if (!res.prefixlen &&
2801 res.table->tb_num_default > 1 &&
2802 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2803 fib_select_default(&res);
2806 fl4->saddr = FIB_RES_PREFSRC(net, res);
2808 dev_out = FIB_RES_DEV(res);
2809 fl4->flowi4_oif = dev_out->ifindex;
2813 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2814 tos, dev_out, flags);
2818 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2819 rt_genid(dev_net(dev_out)));
2820 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2828 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2833 if (!rt_caching(net))
2836 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2839 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2840 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2841 if (rth->rt_key_dst == flp4->daddr &&
2842 rth->rt_key_src == flp4->saddr &&
2843 rt_is_output_route(rth) &&
2844 rth->rt_oif == flp4->flowi4_oif &&
2845 rth->rt_mark == flp4->flowi4_mark &&
2846 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2847 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2848 net_eq(dev_net(rth->dst.dev), net) &&
2849 !rt_is_expired(rth)) {
2850 ipv4_validate_peer(rth);
2851 dst_use(&rth->dst, jiffies);
2852 RT_CACHE_STAT_INC(out_hit);
2853 rcu_read_unlock_bh();
2855 flp4->saddr = rth->rt_src;
2857 flp4->daddr = rth->rt_dst;
2860 RT_CACHE_STAT_INC(out_hlist_search);
2862 rcu_read_unlock_bh();
2865 return ip_route_output_slow(net, flp4);
2867 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2869 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2874 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2876 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2878 return mtu ? : dst->dev->mtu;
2881 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2885 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2891 static struct dst_ops ipv4_dst_blackhole_ops = {
2893 .protocol = cpu_to_be16(ETH_P_IP),
2894 .destroy = ipv4_dst_destroy,
2895 .check = ipv4_blackhole_dst_check,
2896 .mtu = ipv4_blackhole_mtu,
2897 .default_advmss = ipv4_default_advmss,
2898 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2899 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2900 .neigh_lookup = ipv4_neigh_lookup,
2903 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2905 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2906 struct rtable *ort = (struct rtable *) dst_orig;
2909 struct dst_entry *new = &rt->dst;
2912 new->input = dst_discard;
2913 new->output = dst_discard;
2914 dst_copy_metrics(new, &ort->dst);
2916 new->dev = ort->dst.dev;
2920 rt->rt_key_dst = ort->rt_key_dst;
2921 rt->rt_key_src = ort->rt_key_src;
2922 rt->rt_key_tos = ort->rt_key_tos;
2923 rt->rt_route_iif = ort->rt_route_iif;
2924 rt->rt_iif = ort->rt_iif;
2925 rt->rt_oif = ort->rt_oif;
2926 rt->rt_mark = ort->rt_mark;
2928 rt->rt_genid = rt_genid(net);
2929 rt->rt_flags = ort->rt_flags;
2930 rt->rt_type = ort->rt_type;
2931 rt->rt_dst = ort->rt_dst;
2932 rt->rt_src = ort->rt_src;
2933 rt->rt_gateway = ort->rt_gateway;
2934 rt->rt_spec_dst = ort->rt_spec_dst;
2935 rt->peer = ort->peer;
2937 atomic_inc(&rt->peer->refcnt);
2940 atomic_inc(&rt->fi->fib_clntref);
2945 dst_release(dst_orig);
2947 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2950 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2953 struct rtable *rt = __ip_route_output_key(net, flp4);
2958 if (flp4->flowi4_proto)
2959 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2960 flowi4_to_flowi(flp4),
2965 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2967 static int rt_fill_info(struct net *net,
2968 struct sk_buff *skb, u32 pid, u32 seq, int event,
2969 int nowait, unsigned int flags)
2971 struct rtable *rt = skb_rtable(skb);
2973 struct nlmsghdr *nlh;
2974 unsigned long expires = 0;
2975 const struct inet_peer *peer = rt->peer;
2976 u32 id = 0, ts = 0, tsage = 0, error;
2978 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2982 r = nlmsg_data(nlh);
2983 r->rtm_family = AF_INET;
2984 r->rtm_dst_len = 32;
2986 r->rtm_tos = rt->rt_key_tos;
2987 r->rtm_table = RT_TABLE_MAIN;
2988 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2989 r->rtm_type = rt->rt_type;
2990 r->rtm_scope = RT_SCOPE_UNIVERSE;
2991 r->rtm_protocol = RTPROT_UNSPEC;
2992 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2993 if (rt->rt_flags & RTCF_NOTIFY)
2994 r->rtm_flags |= RTM_F_NOTIFY;
2996 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2998 if (rt->rt_key_src) {
2999 r->rtm_src_len = 32;
3000 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
3003 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
3004 #ifdef CONFIG_IP_ROUTE_CLASSID
3005 if (rt->dst.tclassid)
3006 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
3008 if (rt_is_input_route(rt))
3009 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
3010 else if (rt->rt_src != rt->rt_key_src)
3011 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
3013 if (rt->rt_dst != rt->rt_gateway)
3014 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
3016 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
3017 goto nla_put_failure;
3020 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
3022 error = rt->dst.error;
3024 inet_peer_refcheck(rt->peer);
3025 id = atomic_read(&peer->ip_id_count) & 0xffff;
3026 if (peer->tcp_ts_stamp) {
3028 tsage = get_seconds() - peer->tcp_ts_stamp;
3030 expires = ACCESS_ONCE(peer->pmtu_expires);
3032 if (time_before(jiffies, expires))
3039 if (rt_is_input_route(rt)) {
3040 #ifdef CONFIG_IP_MROUTE
3041 __be32 dst = rt->rt_dst;
3043 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
3044 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3045 int err = ipmr_get_route(net, skb,
3046 rt->rt_src, rt->rt_dst,
3052 goto nla_put_failure;
3054 if (err == -EMSGSIZE)
3055 goto nla_put_failure;
3061 NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
3064 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
3065 expires, error) < 0)
3066 goto nla_put_failure;
3068 return nlmsg_end(skb, nlh);
3071 nlmsg_cancel(skb, nlh);
3075 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
3077 struct net *net = sock_net(in_skb->sk);
3079 struct nlattr *tb[RTA_MAX+1];
3080 struct rtable *rt = NULL;
3086 struct sk_buff *skb;
3088 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3092 rtm = nlmsg_data(nlh);
3094 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3100 /* Reserve room for dummy headers, this skb can pass
3101 through good chunk of routing engine.
3103 skb_reset_mac_header(skb);
3104 skb_reset_network_header(skb);
3106 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3107 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3108 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3110 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3111 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3112 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3113 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3116 struct net_device *dev;
3118 dev = __dev_get_by_index(net, iif);
3124 skb->protocol = htons(ETH_P_IP);
3128 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3131 rt = skb_rtable(skb);
3132 if (err == 0 && rt->dst.error)
3133 err = -rt->dst.error;
3135 struct flowi4 fl4 = {
3138 .flowi4_tos = rtm->rtm_tos,
3139 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3140 .flowi4_mark = mark,
3142 rt = ip_route_output_key(net, &fl4);
3152 skb_dst_set(skb, &rt->dst);
3153 if (rtm->rtm_flags & RTM_F_NOTIFY)
3154 rt->rt_flags |= RTCF_NOTIFY;
3156 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3157 RTM_NEWROUTE, 0, 0);
3161 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3170 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3177 net = sock_net(skb->sk);
3182 s_idx = idx = cb->args[1];
3183 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3184 if (!rt_hash_table[h].chain)
3187 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3188 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3189 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3191 if (rt_is_expired(rt))
3193 skb_dst_set_noref(skb, &rt->dst);
3194 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3195 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3196 1, NLM_F_MULTI) <= 0) {
3198 rcu_read_unlock_bh();
3203 rcu_read_unlock_bh();
3212 void ip_rt_multicast_event(struct in_device *in_dev)
3214 rt_cache_flush(dev_net(in_dev->dev), 0);
3217 #ifdef CONFIG_SYSCTL
3218 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3219 void __user *buffer,
3220 size_t *lenp, loff_t *ppos)
3227 memcpy(&ctl, __ctl, sizeof(ctl));
3228 ctl.data = &flush_delay;
3229 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3231 net = (struct net *)__ctl->extra1;
3232 rt_cache_flush(net, flush_delay);
3239 static ctl_table ipv4_route_table[] = {
3241 .procname = "gc_thresh",
3242 .data = &ipv4_dst_ops.gc_thresh,
3243 .maxlen = sizeof(int),
3245 .proc_handler = proc_dointvec,
3248 .procname = "max_size",
3249 .data = &ip_rt_max_size,
3250 .maxlen = sizeof(int),
3252 .proc_handler = proc_dointvec,
3255 /* Deprecated. Use gc_min_interval_ms */
3257 .procname = "gc_min_interval",
3258 .data = &ip_rt_gc_min_interval,
3259 .maxlen = sizeof(int),
3261 .proc_handler = proc_dointvec_jiffies,
3264 .procname = "gc_min_interval_ms",
3265 .data = &ip_rt_gc_min_interval,
3266 .maxlen = sizeof(int),
3268 .proc_handler = proc_dointvec_ms_jiffies,
3271 .procname = "gc_timeout",
3272 .data = &ip_rt_gc_timeout,
3273 .maxlen = sizeof(int),
3275 .proc_handler = proc_dointvec_jiffies,
3278 .procname = "gc_interval",
3279 .data = &ip_rt_gc_interval,
3280 .maxlen = sizeof(int),
3282 .proc_handler = proc_dointvec_jiffies,
3285 .procname = "redirect_load",
3286 .data = &ip_rt_redirect_load,
3287 .maxlen = sizeof(int),
3289 .proc_handler = proc_dointvec,
3292 .procname = "redirect_number",
3293 .data = &ip_rt_redirect_number,
3294 .maxlen = sizeof(int),
3296 .proc_handler = proc_dointvec,
3299 .procname = "redirect_silence",
3300 .data = &ip_rt_redirect_silence,
3301 .maxlen = sizeof(int),
3303 .proc_handler = proc_dointvec,
3306 .procname = "error_cost",
3307 .data = &ip_rt_error_cost,
3308 .maxlen = sizeof(int),
3310 .proc_handler = proc_dointvec,
3313 .procname = "error_burst",
3314 .data = &ip_rt_error_burst,
3315 .maxlen = sizeof(int),
3317 .proc_handler = proc_dointvec,
3320 .procname = "gc_elasticity",
3321 .data = &ip_rt_gc_elasticity,
3322 .maxlen = sizeof(int),
3324 .proc_handler = proc_dointvec,
3327 .procname = "mtu_expires",
3328 .data = &ip_rt_mtu_expires,
3329 .maxlen = sizeof(int),
3331 .proc_handler = proc_dointvec_jiffies,
3334 .procname = "min_pmtu",
3335 .data = &ip_rt_min_pmtu,
3336 .maxlen = sizeof(int),
3338 .proc_handler = proc_dointvec,
3341 .procname = "min_adv_mss",
3342 .data = &ip_rt_min_advmss,
3343 .maxlen = sizeof(int),
3345 .proc_handler = proc_dointvec,
3350 static struct ctl_table empty[1];
3352 static struct ctl_table ipv4_skeleton[] =
3354 { .procname = "route",
3355 .mode = 0555, .child = ipv4_route_table},
3356 { .procname = "neigh",
3357 .mode = 0555, .child = empty},
3361 static __net_initdata struct ctl_path ipv4_path[] = {
3362 { .procname = "net", },
3363 { .procname = "ipv4", },
3367 static struct ctl_table ipv4_route_flush_table[] = {
3369 .procname = "flush",
3370 .maxlen = sizeof(int),
3372 .proc_handler = ipv4_sysctl_rtcache_flush,
3377 static __net_initdata struct ctl_path ipv4_route_path[] = {
3378 { .procname = "net", },
3379 { .procname = "ipv4", },
3380 { .procname = "route", },
3384 static __net_init int sysctl_route_net_init(struct net *net)
3386 struct ctl_table *tbl;
3388 tbl = ipv4_route_flush_table;
3389 if (!net_eq(net, &init_net)) {
3390 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3394 tbl[0].extra1 = net;
3396 net->ipv4.route_hdr =
3397 register_net_sysctl_table(net, ipv4_route_path, tbl);
3398 if (net->ipv4.route_hdr == NULL)
3403 if (tbl != ipv4_route_flush_table)
3409 static __net_exit void sysctl_route_net_exit(struct net *net)
3411 struct ctl_table *tbl;
3413 tbl = net->ipv4.route_hdr->ctl_table_arg;
3414 unregister_net_sysctl_table(net->ipv4.route_hdr);
3415 BUG_ON(tbl == ipv4_route_flush_table);
3419 static __net_initdata struct pernet_operations sysctl_route_ops = {
3420 .init = sysctl_route_net_init,
3421 .exit = sysctl_route_net_exit,
3425 static __net_init int rt_genid_init(struct net *net)
3427 get_random_bytes(&net->ipv4.rt_genid,
3428 sizeof(net->ipv4.rt_genid));
3429 get_random_bytes(&net->ipv4.dev_addr_genid,
3430 sizeof(net->ipv4.dev_addr_genid));
3434 static __net_initdata struct pernet_operations rt_genid_ops = {
3435 .init = rt_genid_init,
3439 #ifdef CONFIG_IP_ROUTE_CLASSID
3440 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3441 #endif /* CONFIG_IP_ROUTE_CLASSID */
3443 static __initdata unsigned long rhash_entries;
3444 static int __init set_rhash_entries(char *str)
3448 rhash_entries = simple_strtoul(str, &str, 0);
3451 __setup("rhash_entries=", set_rhash_entries);
3453 int __init ip_rt_init(void)
3457 #ifdef CONFIG_IP_ROUTE_CLASSID
3458 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3460 panic("IP: failed to allocate ip_rt_acct\n");
3463 ipv4_dst_ops.kmem_cachep =
3464 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3465 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3467 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3469 if (dst_entries_init(&ipv4_dst_ops) < 0)
3470 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3472 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3473 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3475 rt_hash_table = (struct rt_hash_bucket *)
3476 alloc_large_system_hash("IP route cache",
3477 sizeof(struct rt_hash_bucket),
3479 (totalram_pages >= 128 * 1024) ?
3484 rhash_entries ? 0 : 512 * 1024);
3485 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3486 rt_hash_lock_init();
3488 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3489 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3494 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3495 expires_ljiffies = jiffies;
3496 schedule_delayed_work(&expires_work,
3497 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3499 if (ip_rt_proc_init())
3500 printk(KERN_ERR "Unable to create route proc files\n");
3503 xfrm4_init(ip_rt_max_size);
3505 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3507 #ifdef CONFIG_SYSCTL
3508 register_pernet_subsys(&sysctl_route_ops);
3510 register_pernet_subsys(&rt_genid_ops);
3514 #ifdef CONFIG_SYSCTL
3516 * We really need to sanitize the damn ipv4 init order, then all
3517 * this nonsense will go away.
3519 void __init ip_static_sysctl_init(void)
3521 register_sysctl_paths(ipv4_path, ipv4_skeleton);