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 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/workqueue.h>
84 #include <linux/skbuff.h>
85 #include <linux/inetdevice.h>
86 #include <linux/igmp.h>
87 #include <linux/pkt_sched.h>
88 #include <linux/mroute.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/jhash.h>
92 #include <linux/rcupdate.h>
93 #include <linux/times.h>
94 #include <linux/slab.h>
95 #include <linux/prefetch.h>
97 #include <net/net_namespace.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/inetpeer.h>
102 #include <net/sock.h>
103 #include <net/ip_fib.h>
106 #include <net/icmp.h>
107 #include <net/xfrm.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
114 #include <net/secure_seq.h>
116 #define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119 #define IP_MAX_MTU 0xFFF0
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size;
124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
126 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
127 static int ip_rt_redirect_number __read_mostly = 9;
128 static int ip_rt_redirect_load __read_mostly = HZ / 50;
129 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130 static int ip_rt_error_cost __read_mostly = HZ;
131 static int ip_rt_error_burst __read_mostly = 5 * HZ;
132 static int ip_rt_gc_elasticity __read_mostly = 8;
133 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
134 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
135 static int ip_rt_min_advmss __read_mostly = 256;
136 static int rt_chain_length_max __read_mostly = 20;
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 void ip_do_redirect(struct dst_entry *dst, struct sk_buff *skb);
153 static int rt_garbage_collect(struct dst_ops *ops);
155 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
160 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
166 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
170 static struct dst_ops ipv4_dst_ops = {
172 .protocol = cpu_to_be16(ETH_P_IP),
173 .gc = rt_garbage_collect,
174 .check = ipv4_dst_check,
175 .default_advmss = ipv4_default_advmss,
177 .cow_metrics = ipv4_cow_metrics,
178 .destroy = ipv4_dst_destroy,
179 .ifdown = ipv4_dst_ifdown,
180 .negative_advice = ipv4_negative_advice,
181 .link_failure = ipv4_link_failure,
182 .update_pmtu = ip_rt_update_pmtu,
183 .redirect = ip_do_redirect,
184 .local_out = __ip_local_out,
185 .neigh_lookup = ipv4_neigh_lookup,
188 #define ECN_OR_COST(class) TC_PRIO_##class
190 const __u8 ip_tos2prio[16] = {
192 ECN_OR_COST(BESTEFFORT),
194 ECN_OR_COST(BESTEFFORT),
200 ECN_OR_COST(INTERACTIVE),
202 ECN_OR_COST(INTERACTIVE),
203 TC_PRIO_INTERACTIVE_BULK,
204 ECN_OR_COST(INTERACTIVE_BULK),
205 TC_PRIO_INTERACTIVE_BULK,
206 ECN_OR_COST(INTERACTIVE_BULK)
208 EXPORT_SYMBOL(ip_tos2prio);
214 /* The locking scheme is rather straight forward:
216 * 1) Read-Copy Update protects the buckets of the central route hash.
217 * 2) Only writers remove entries, and they hold the lock
218 * as they look at rtable reference counts.
219 * 3) Only readers acquire references to rtable entries,
220 * they do so with atomic increments and with the
224 struct rt_hash_bucket {
225 struct rtable __rcu *chain;
228 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
229 defined(CONFIG_PROVE_LOCKING)
231 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
232 * The size of this table is a power of two and depends on the number of CPUS.
233 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
235 #ifdef CONFIG_LOCKDEP
236 # define RT_HASH_LOCK_SZ 256
239 # define RT_HASH_LOCK_SZ 4096
241 # define RT_HASH_LOCK_SZ 2048
243 # define RT_HASH_LOCK_SZ 1024
245 # define RT_HASH_LOCK_SZ 512
247 # define RT_HASH_LOCK_SZ 256
251 static spinlock_t *rt_hash_locks;
252 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
254 static __init void rt_hash_lock_init(void)
258 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
261 panic("IP: failed to allocate rt_hash_locks\n");
263 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
264 spin_lock_init(&rt_hash_locks[i]);
267 # define rt_hash_lock_addr(slot) NULL
269 static inline void rt_hash_lock_init(void)
274 static struct rt_hash_bucket *rt_hash_table __read_mostly;
275 static unsigned int rt_hash_mask __read_mostly;
276 static unsigned int rt_hash_log __read_mostly;
278 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
279 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
281 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
284 return jhash_3words((__force u32)daddr, (__force u32)saddr,
289 static inline int rt_genid(struct net *net)
291 return atomic_read(&net->ipv4.rt_genid);
294 #ifdef CONFIG_PROC_FS
295 struct rt_cache_iter_state {
296 struct seq_net_private p;
301 static struct rtable *rt_cache_get_first(struct seq_file *seq)
303 struct rt_cache_iter_state *st = seq->private;
304 struct rtable *r = NULL;
306 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
307 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
310 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
312 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
313 r->rt_genid == st->genid)
315 r = rcu_dereference_bh(r->dst.rt_next);
317 rcu_read_unlock_bh();
322 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
325 struct rt_cache_iter_state *st = seq->private;
327 r = rcu_dereference_bh(r->dst.rt_next);
329 rcu_read_unlock_bh();
331 if (--st->bucket < 0)
333 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
335 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
340 static struct rtable *rt_cache_get_next(struct seq_file *seq,
343 struct rt_cache_iter_state *st = seq->private;
344 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
345 if (dev_net(r->dst.dev) != seq_file_net(seq))
347 if (r->rt_genid == st->genid)
353 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
355 struct rtable *r = rt_cache_get_first(seq);
358 while (pos && (r = rt_cache_get_next(seq, r)))
360 return pos ? NULL : r;
363 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
365 struct rt_cache_iter_state *st = seq->private;
367 return rt_cache_get_idx(seq, *pos - 1);
368 st->genid = rt_genid(seq_file_net(seq));
369 return SEQ_START_TOKEN;
372 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
376 if (v == SEQ_START_TOKEN)
377 r = rt_cache_get_first(seq);
379 r = rt_cache_get_next(seq, v);
384 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
386 if (v && v != SEQ_START_TOKEN)
387 rcu_read_unlock_bh();
390 static int rt_cache_seq_show(struct seq_file *seq, void *v)
392 if (v == SEQ_START_TOKEN)
393 seq_printf(seq, "%-127s\n",
394 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
395 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
398 struct rtable *r = v;
401 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
402 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
403 r->dst.dev ? r->dst.dev->name : "*",
404 (__force u32)r->rt_dst,
405 (__force u32)r->rt_gateway,
406 r->rt_flags, atomic_read(&r->dst.__refcnt),
407 r->dst.__use, 0, (__force u32)r->rt_src,
408 dst_metric_advmss(&r->dst) + 40,
409 dst_metric(&r->dst, RTAX_WINDOW), 0,
413 seq_printf(seq, "%*s\n", 127 - len, "");
418 static const struct seq_operations rt_cache_seq_ops = {
419 .start = rt_cache_seq_start,
420 .next = rt_cache_seq_next,
421 .stop = rt_cache_seq_stop,
422 .show = rt_cache_seq_show,
425 static int rt_cache_seq_open(struct inode *inode, struct file *file)
427 return seq_open_net(inode, file, &rt_cache_seq_ops,
428 sizeof(struct rt_cache_iter_state));
431 static const struct file_operations rt_cache_seq_fops = {
432 .owner = THIS_MODULE,
433 .open = rt_cache_seq_open,
436 .release = seq_release_net,
440 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
445 return SEQ_START_TOKEN;
447 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
448 if (!cpu_possible(cpu))
451 return &per_cpu(rt_cache_stat, cpu);
456 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
460 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
461 if (!cpu_possible(cpu))
464 return &per_cpu(rt_cache_stat, cpu);
470 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
475 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
477 struct rt_cache_stat *st = v;
479 if (v == SEQ_START_TOKEN) {
480 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");
484 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
485 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
486 dst_entries_get_slow(&ipv4_dst_ops),
509 static const struct seq_operations rt_cpu_seq_ops = {
510 .start = rt_cpu_seq_start,
511 .next = rt_cpu_seq_next,
512 .stop = rt_cpu_seq_stop,
513 .show = rt_cpu_seq_show,
517 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
519 return seq_open(file, &rt_cpu_seq_ops);
522 static const struct file_operations rt_cpu_seq_fops = {
523 .owner = THIS_MODULE,
524 .open = rt_cpu_seq_open,
527 .release = seq_release,
530 #ifdef CONFIG_IP_ROUTE_CLASSID
531 static int rt_acct_proc_show(struct seq_file *m, void *v)
533 struct ip_rt_acct *dst, *src;
536 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
540 for_each_possible_cpu(i) {
541 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
542 for (j = 0; j < 256; j++) {
543 dst[j].o_bytes += src[j].o_bytes;
544 dst[j].o_packets += src[j].o_packets;
545 dst[j].i_bytes += src[j].i_bytes;
546 dst[j].i_packets += src[j].i_packets;
550 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
555 static int rt_acct_proc_open(struct inode *inode, struct file *file)
557 return single_open(file, rt_acct_proc_show, NULL);
560 static const struct file_operations rt_acct_proc_fops = {
561 .owner = THIS_MODULE,
562 .open = rt_acct_proc_open,
565 .release = single_release,
569 static int __net_init ip_rt_do_proc_init(struct net *net)
571 struct proc_dir_entry *pde;
573 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
578 pde = proc_create("rt_cache", S_IRUGO,
579 net->proc_net_stat, &rt_cpu_seq_fops);
583 #ifdef CONFIG_IP_ROUTE_CLASSID
584 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
590 #ifdef CONFIG_IP_ROUTE_CLASSID
592 remove_proc_entry("rt_cache", net->proc_net_stat);
595 remove_proc_entry("rt_cache", net->proc_net);
600 static void __net_exit ip_rt_do_proc_exit(struct net *net)
602 remove_proc_entry("rt_cache", net->proc_net_stat);
603 remove_proc_entry("rt_cache", net->proc_net);
604 #ifdef CONFIG_IP_ROUTE_CLASSID
605 remove_proc_entry("rt_acct", net->proc_net);
609 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
610 .init = ip_rt_do_proc_init,
611 .exit = ip_rt_do_proc_exit,
614 static int __init ip_rt_proc_init(void)
616 return register_pernet_subsys(&ip_rt_proc_ops);
620 static inline int ip_rt_proc_init(void)
624 #endif /* CONFIG_PROC_FS */
626 static inline void rt_free(struct rtable *rt)
628 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
631 static inline void rt_drop(struct rtable *rt)
634 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
637 static inline int rt_fast_clean(struct rtable *rth)
639 /* Kill broadcast/multicast entries very aggresively, if they
640 collide in hash table with more useful entries */
641 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
642 rt_is_input_route(rth) && rth->dst.rt_next;
645 static inline int rt_valuable(struct rtable *rth)
647 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
651 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
656 if (atomic_read(&rth->dst.__refcnt))
659 age = jiffies - rth->dst.lastuse;
660 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
661 (age <= tmo2 && rt_valuable(rth)))
667 /* Bits of score are:
669 * 30: not quite useless
670 * 29..0: usage counter
672 static inline u32 rt_score(struct rtable *rt)
674 u32 score = jiffies - rt->dst.lastuse;
676 score = ~score & ~(3<<30);
681 if (rt_is_output_route(rt) ||
682 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
688 static inline bool rt_caching(const struct net *net)
690 return net->ipv4.current_rt_cache_rebuild_count <=
691 net->ipv4.sysctl_rt_cache_rebuild_count;
694 static inline bool compare_hash_inputs(const struct rtable *rt1,
695 const struct rtable *rt2)
697 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
698 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
699 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
702 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
704 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
705 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
706 (rt1->rt_mark ^ rt2->rt_mark) |
707 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
708 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
709 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
712 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
714 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
717 static inline int rt_is_expired(struct rtable *rth)
719 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
723 * Perform a full scan of hash table and free all entries.
724 * Can be called by a softirq or a process.
725 * In the later case, we want to be reschedule if necessary
727 static void rt_do_flush(struct net *net, int process_context)
730 struct rtable *rth, *next;
732 for (i = 0; i <= rt_hash_mask; i++) {
733 struct rtable __rcu **pprev;
736 if (process_context && need_resched())
738 rth = rcu_access_pointer(rt_hash_table[i].chain);
742 spin_lock_bh(rt_hash_lock_addr(i));
745 pprev = &rt_hash_table[i].chain;
746 rth = rcu_dereference_protected(*pprev,
747 lockdep_is_held(rt_hash_lock_addr(i)));
750 next = rcu_dereference_protected(rth->dst.rt_next,
751 lockdep_is_held(rt_hash_lock_addr(i)));
754 net_eq(dev_net(rth->dst.dev), net)) {
755 rcu_assign_pointer(*pprev, next);
756 rcu_assign_pointer(rth->dst.rt_next, list);
759 pprev = &rth->dst.rt_next;
764 spin_unlock_bh(rt_hash_lock_addr(i));
766 for (; list; list = next) {
767 next = rcu_dereference_protected(list->dst.rt_next, 1);
774 * While freeing expired entries, we compute average chain length
775 * and standard deviation, using fixed-point arithmetic.
776 * This to have an estimation of rt_chain_length_max
777 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
778 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
782 #define ONE (1UL << FRACT_BITS)
785 * Given a hash chain and an item in this hash chain,
786 * find if a previous entry has the same hash_inputs
787 * (but differs on tos, mark or oif)
788 * Returns 0 if an alias is found.
789 * Returns ONE if rth has no alias before itself.
791 static int has_noalias(const struct rtable *head, const struct rtable *rth)
793 const struct rtable *aux = head;
796 if (compare_hash_inputs(aux, rth))
798 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
803 static void rt_check_expire(void)
805 static unsigned int rover;
806 unsigned int i = rover, goal;
808 struct rtable __rcu **rthp;
809 unsigned long samples = 0;
810 unsigned long sum = 0, sum2 = 0;
814 delta = jiffies - expires_ljiffies;
815 expires_ljiffies = jiffies;
816 mult = ((u64)delta) << rt_hash_log;
817 if (ip_rt_gc_timeout > 1)
818 do_div(mult, ip_rt_gc_timeout);
819 goal = (unsigned int)mult;
820 if (goal > rt_hash_mask)
821 goal = rt_hash_mask + 1;
822 for (; goal > 0; goal--) {
823 unsigned long tmo = ip_rt_gc_timeout;
824 unsigned long length;
826 i = (i + 1) & rt_hash_mask;
827 rthp = &rt_hash_table[i].chain;
834 if (rcu_dereference_raw(*rthp) == NULL)
837 spin_lock_bh(rt_hash_lock_addr(i));
838 while ((rth = rcu_dereference_protected(*rthp,
839 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
840 prefetch(rth->dst.rt_next);
841 if (rt_is_expired(rth) ||
842 rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
843 *rthp = rth->dst.rt_next;
848 /* We only count entries on a chain with equal
849 * hash inputs once so that entries for
850 * different QOS levels, and other non-hash
851 * input attributes don't unfairly skew the
855 rthp = &rth->dst.rt_next;
856 length += has_noalias(rt_hash_table[i].chain, rth);
858 spin_unlock_bh(rt_hash_lock_addr(i));
860 sum2 += length*length;
863 unsigned long avg = sum / samples;
864 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
865 rt_chain_length_max = max_t(unsigned long,
867 (avg + 4*sd) >> FRACT_BITS);
873 * rt_worker_func() is run in process context.
874 * we call rt_check_expire() to scan part of the hash table
876 static void rt_worker_func(struct work_struct *work)
879 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
883 * Perturbation of rt_genid by a small quantity [1..256]
884 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
885 * many times (2^24) without giving recent rt_genid.
886 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
888 static void rt_cache_invalidate(struct net *net)
890 unsigned char shuffle;
892 get_random_bytes(&shuffle, sizeof(shuffle));
893 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
897 * delay < 0 : invalidate cache (fast : entries will be deleted later)
898 * delay >= 0 : invalidate & flush cache (can be long)
900 void rt_cache_flush(struct net *net, int delay)
902 rt_cache_invalidate(net);
904 rt_do_flush(net, !in_softirq());
907 /* Flush previous cache invalidated entries from the cache */
908 void rt_cache_flush_batch(struct net *net)
910 rt_do_flush(net, !in_softirq());
913 static void rt_emergency_hash_rebuild(struct net *net)
915 net_warn_ratelimited("Route hash chain too long!\n");
916 rt_cache_invalidate(net);
920 Short description of GC goals.
922 We want to build algorithm, which will keep routing cache
923 at some equilibrium point, when number of aged off entries
924 is kept approximately equal to newly generated ones.
926 Current expiration strength is variable "expire".
927 We try to adjust it dynamically, so that if networking
928 is idle expires is large enough to keep enough of warm entries,
929 and when load increases it reduces to limit cache size.
932 static int rt_garbage_collect(struct dst_ops *ops)
934 static unsigned long expire = RT_GC_TIMEOUT;
935 static unsigned long last_gc;
937 static int equilibrium;
939 struct rtable __rcu **rthp;
940 unsigned long now = jiffies;
942 int entries = dst_entries_get_fast(&ipv4_dst_ops);
945 * Garbage collection is pretty expensive,
946 * do not make it too frequently.
949 RT_CACHE_STAT_INC(gc_total);
951 if (now - last_gc < ip_rt_gc_min_interval &&
952 entries < ip_rt_max_size) {
953 RT_CACHE_STAT_INC(gc_ignored);
957 entries = dst_entries_get_slow(&ipv4_dst_ops);
958 /* Calculate number of entries, which we want to expire now. */
959 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
961 if (equilibrium < ipv4_dst_ops.gc_thresh)
962 equilibrium = ipv4_dst_ops.gc_thresh;
963 goal = entries - equilibrium;
965 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
966 goal = entries - equilibrium;
969 /* We are in dangerous area. Try to reduce cache really
972 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
973 equilibrium = entries - goal;
976 if (now - last_gc >= ip_rt_gc_min_interval)
987 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
988 unsigned long tmo = expire;
990 k = (k + 1) & rt_hash_mask;
991 rthp = &rt_hash_table[k].chain;
992 spin_lock_bh(rt_hash_lock_addr(k));
993 while ((rth = rcu_dereference_protected(*rthp,
994 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
995 if (!rt_is_expired(rth) &&
996 !rt_may_expire(rth, tmo, expire)) {
998 rthp = &rth->dst.rt_next;
1001 *rthp = rth->dst.rt_next;
1005 spin_unlock_bh(rt_hash_lock_addr(k));
1014 /* Goal is not achieved. We stop process if:
1016 - if expire reduced to zero. Otherwise, expire is halfed.
1017 - if table is not full.
1018 - if we are called from interrupt.
1019 - jiffies check is just fallback/debug loop breaker.
1020 We will not spin here for long time in any case.
1023 RT_CACHE_STAT_INC(gc_goal_miss);
1030 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1032 } while (!in_softirq() && time_before_eq(jiffies, now));
1034 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1036 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1038 net_warn_ratelimited("dst cache overflow\n");
1039 RT_CACHE_STAT_INC(gc_dst_overflow);
1043 expire += ip_rt_gc_min_interval;
1044 if (expire > ip_rt_gc_timeout ||
1045 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1046 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1047 expire = ip_rt_gc_timeout;
1052 * Returns number of entries in a hash chain that have different hash_inputs
1054 static int slow_chain_length(const struct rtable *head)
1057 const struct rtable *rth = head;
1060 length += has_noalias(head, rth);
1061 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1063 return length >> FRACT_BITS;
1066 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
1067 struct sk_buff *skb,
1070 struct net_device *dev = dst->dev;
1071 const __be32 *pkey = daddr;
1072 const struct rtable *rt;
1073 struct neighbour *n;
1075 rt = (const struct rtable *) dst;
1077 pkey = (const __be32 *) &rt->rt_gateway;
1079 pkey = &ip_hdr(skb)->daddr;
1081 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1084 return neigh_create(&arp_tbl, pkey, dev);
1087 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1088 struct sk_buff *skb, int ifindex)
1090 struct rtable *rth, *cand;
1091 struct rtable __rcu **rthp, **candp;
1098 min_score = ~(u32)0;
1103 if (!rt_caching(dev_net(rt->dst.dev)) || (rt->dst.flags & DST_NOCACHE)) {
1105 * If we're not caching, just tell the caller we
1106 * were successful and don't touch the route. The
1107 * caller hold the sole reference to the cache entry, and
1108 * it will be released when the caller is done with it.
1109 * If we drop it here, the callers have no way to resolve routes
1110 * when we're not caching. Instead, just point *rp at rt, so
1111 * the caller gets a single use out of the route
1112 * Note that we do rt_free on this new route entry, so that
1113 * once its refcount hits zero, we are still able to reap it
1115 * Note: To avoid expensive rcu stuff for this uncached dst,
1116 * we set DST_NOCACHE so that dst_release() can free dst without
1117 * waiting a grace period.
1120 rt->dst.flags |= DST_NOCACHE;
1124 rthp = &rt_hash_table[hash].chain;
1126 spin_lock_bh(rt_hash_lock_addr(hash));
1127 while ((rth = rcu_dereference_protected(*rthp,
1128 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1129 if (rt_is_expired(rth)) {
1130 *rthp = rth->dst.rt_next;
1134 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1136 *rthp = rth->dst.rt_next;
1138 * Since lookup is lockfree, the deletion
1139 * must be visible to another weakly ordered CPU before
1140 * the insertion at the start of the hash chain.
1142 rcu_assign_pointer(rth->dst.rt_next,
1143 rt_hash_table[hash].chain);
1145 * Since lookup is lockfree, the update writes
1146 * must be ordered for consistency on SMP.
1148 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1150 dst_use(&rth->dst, now);
1151 spin_unlock_bh(rt_hash_lock_addr(hash));
1155 skb_dst_set(skb, &rth->dst);
1159 if (!atomic_read(&rth->dst.__refcnt)) {
1160 u32 score = rt_score(rth);
1162 if (score <= min_score) {
1171 rthp = &rth->dst.rt_next;
1175 /* ip_rt_gc_elasticity used to be average length of chain
1176 * length, when exceeded gc becomes really aggressive.
1178 * The second limit is less certain. At the moment it allows
1179 * only 2 entries per bucket. We will see.
1181 if (chain_length > ip_rt_gc_elasticity) {
1182 *candp = cand->dst.rt_next;
1186 if (chain_length > rt_chain_length_max &&
1187 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1188 struct net *net = dev_net(rt->dst.dev);
1189 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1190 if (!rt_caching(net)) {
1191 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1192 rt->dst.dev->name, num);
1194 rt_emergency_hash_rebuild(net);
1195 spin_unlock_bh(rt_hash_lock_addr(hash));
1197 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1198 ifindex, rt_genid(net));
1203 rt->dst.rt_next = rt_hash_table[hash].chain;
1206 * Since lookup is lockfree, we must make sure
1207 * previous writes to rt are committed to memory
1208 * before making rt visible to other CPUS.
1210 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1212 spin_unlock_bh(rt_hash_lock_addr(hash));
1216 skb_dst_set(skb, &rt->dst);
1221 * Peer allocation may fail only in serious out-of-memory conditions. However
1222 * we still can generate some output.
1223 * Random ID selection looks a bit dangerous because we have no chances to
1224 * select ID being unique in a reasonable period of time.
1225 * But broken packet identifier may be better than no packet at all.
1227 static void ip_select_fb_ident(struct iphdr *iph)
1229 static DEFINE_SPINLOCK(ip_fb_id_lock);
1230 static u32 ip_fallback_id;
1233 spin_lock_bh(&ip_fb_id_lock);
1234 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1235 iph->id = htons(salt & 0xFFFF);
1236 ip_fallback_id = salt;
1237 spin_unlock_bh(&ip_fb_id_lock);
1240 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1242 struct net *net = dev_net(dst->dev);
1243 struct inet_peer *peer;
1245 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
1247 iph->id = htons(inet_getid(peer, more));
1252 ip_select_fb_ident(iph);
1254 EXPORT_SYMBOL(__ip_select_ident);
1256 static void rt_del(unsigned int hash, struct rtable *rt)
1258 struct rtable __rcu **rthp;
1261 rthp = &rt_hash_table[hash].chain;
1262 spin_lock_bh(rt_hash_lock_addr(hash));
1264 while ((aux = rcu_dereference_protected(*rthp,
1265 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1266 if (aux == rt || rt_is_expired(aux)) {
1267 *rthp = aux->dst.rt_next;
1271 rthp = &aux->dst.rt_next;
1273 spin_unlock_bh(rt_hash_lock_addr(hash));
1276 static void ip_do_redirect(struct dst_entry *dst, struct sk_buff *skb)
1278 __be32 new_gw = icmp_hdr(skb)->un.gateway;
1279 __be32 old_gw = ip_hdr(skb)->saddr;
1280 struct net_device *dev = skb->dev;
1281 struct in_device *in_dev;
1282 struct neighbour *n;
1286 switch (icmp_hdr(skb)->code & 7) {
1287 case ICMP_REDIR_NET:
1288 case ICMP_REDIR_NETTOS:
1289 case ICMP_REDIR_HOST:
1290 case ICMP_REDIR_HOSTTOS:
1297 rt = (struct rtable *) dst;
1298 if (rt->rt_gateway != old_gw)
1301 in_dev = __in_dev_get_rcu(dev);
1306 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1307 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1308 ipv4_is_zeronet(new_gw))
1309 goto reject_redirect;
1311 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1312 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1313 goto reject_redirect;
1314 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1315 goto reject_redirect;
1317 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1318 goto reject_redirect;
1321 n = ipv4_neigh_lookup(dst, NULL, &new_gw);
1323 if (!(n->nud_state & NUD_VALID)) {
1324 neigh_event_send(n, NULL);
1326 rt->rt_gateway = new_gw;
1327 rt->rt_flags |= RTCF_REDIRECTED;
1328 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1335 #ifdef CONFIG_IP_ROUTE_VERBOSE
1336 if (IN_DEV_LOG_MARTIANS(in_dev)) {
1337 const struct iphdr *iph = (const struct iphdr *) skb->data;
1338 __be32 daddr = iph->daddr;
1339 __be32 saddr = iph->saddr;
1341 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1342 " Advised path = %pI4 -> %pI4\n",
1343 &old_gw, dev->name, &new_gw,
1350 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1352 struct rtable *rt = (struct rtable *)dst;
1353 struct dst_entry *ret = dst;
1356 if (dst->obsolete > 0) {
1359 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1361 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1363 rt_genid(dev_net(dst->dev)));
1373 * 1. The first ip_rt_redirect_number redirects are sent
1374 * with exponential backoff, then we stop sending them at all,
1375 * assuming that the host ignores our redirects.
1376 * 2. If we did not see packets requiring redirects
1377 * during ip_rt_redirect_silence, we assume that the host
1378 * forgot redirected route and start to send redirects again.
1380 * This algorithm is much cheaper and more intelligent than dumb load limiting
1383 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1384 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1387 void ip_rt_send_redirect(struct sk_buff *skb)
1389 struct rtable *rt = skb_rtable(skb);
1390 struct in_device *in_dev;
1391 struct inet_peer *peer;
1396 in_dev = __in_dev_get_rcu(rt->dst.dev);
1397 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1401 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1404 net = dev_net(rt->dst.dev);
1405 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1407 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1411 /* No redirected packets during ip_rt_redirect_silence;
1412 * reset the algorithm.
1414 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1415 peer->rate_tokens = 0;
1417 /* Too many ignored redirects; do not send anything
1418 * set dst.rate_last to the last seen redirected packet.
1420 if (peer->rate_tokens >= ip_rt_redirect_number) {
1421 peer->rate_last = jiffies;
1425 /* Check for load limit; set rate_last to the latest sent
1428 if (peer->rate_tokens == 0 ||
1431 (ip_rt_redirect_load << peer->rate_tokens)))) {
1432 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1433 peer->rate_last = jiffies;
1434 ++peer->rate_tokens;
1435 #ifdef CONFIG_IP_ROUTE_VERBOSE
1437 peer->rate_tokens == ip_rt_redirect_number)
1438 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1439 &ip_hdr(skb)->saddr, rt->rt_iif,
1440 &rt->rt_dst, &rt->rt_gateway);
1447 static int ip_error(struct sk_buff *skb)
1449 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1450 struct rtable *rt = skb_rtable(skb);
1451 struct inet_peer *peer;
1457 net = dev_net(rt->dst.dev);
1458 if (!IN_DEV_FORWARD(in_dev)) {
1459 switch (rt->dst.error) {
1461 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1465 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1471 switch (rt->dst.error) {
1476 code = ICMP_HOST_UNREACH;
1479 code = ICMP_NET_UNREACH;
1480 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1483 code = ICMP_PKT_FILTERED;
1487 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1492 peer->rate_tokens += now - peer->rate_last;
1493 if (peer->rate_tokens > ip_rt_error_burst)
1494 peer->rate_tokens = ip_rt_error_burst;
1495 peer->rate_last = now;
1496 if (peer->rate_tokens >= ip_rt_error_cost)
1497 peer->rate_tokens -= ip_rt_error_cost;
1503 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1505 out: kfree_skb(skb);
1509 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1511 struct rtable *rt = (struct rtable *) dst;
1515 if (mtu < ip_rt_min_pmtu)
1516 mtu = ip_rt_min_pmtu;
1519 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
1522 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1523 int oif, u32 mark, u8 protocol, int flow_flags)
1525 const struct iphdr *iph = (const struct iphdr *)skb->data;
1529 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1530 protocol, flow_flags,
1531 iph->daddr, iph->saddr, 0, 0);
1532 rt = __ip_route_output_key(net, &fl4);
1534 ip_rt_update_pmtu(&rt->dst, mtu);
1538 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1540 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1542 const struct inet_sock *inet = inet_sk(sk);
1544 return ipv4_update_pmtu(skb, sock_net(sk), mtu,
1545 sk->sk_bound_dev_if, sk->sk_mark,
1546 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1547 inet_sk_flowi_flags(sk));
1549 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1551 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1552 int oif, u32 mark, u8 protocol, int flow_flags)
1554 const struct iphdr *iph = (const struct iphdr *)skb->data;
1558 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1559 protocol, flow_flags, iph->daddr, iph->saddr, 0, 0);
1560 rt = __ip_route_output_key(net, &fl4);
1562 ip_do_redirect(&rt->dst, skb);
1566 EXPORT_SYMBOL_GPL(ipv4_redirect);
1568 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1570 const struct inet_sock *inet = inet_sk(sk);
1572 return ipv4_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
1574 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1575 inet_sk_flowi_flags(sk));
1577 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1579 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1581 struct rtable *rt = (struct rtable *) dst;
1583 if (rt_is_expired(rt))
1588 static void ipv4_dst_destroy(struct dst_entry *dst)
1590 struct rtable *rt = (struct rtable *) dst;
1593 fib_info_put(rt->fi);
1599 static void ipv4_link_failure(struct sk_buff *skb)
1603 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1605 rt = skb_rtable(skb);
1607 dst_set_expires(&rt->dst, 0);
1610 static int ip_rt_bug(struct sk_buff *skb)
1612 pr_debug("%s: %pI4 -> %pI4, %s\n",
1613 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1614 skb->dev ? skb->dev->name : "?");
1621 We do not cache source address of outgoing interface,
1622 because it is used only by IP RR, TS and SRR options,
1623 so that it out of fast path.
1625 BTW remember: "addr" is allowed to be not aligned
1629 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1633 if (rt_is_output_route(rt))
1634 src = ip_hdr(skb)->saddr;
1636 struct fib_result res;
1642 memset(&fl4, 0, sizeof(fl4));
1643 fl4.daddr = iph->daddr;
1644 fl4.saddr = iph->saddr;
1645 fl4.flowi4_tos = RT_TOS(iph->tos);
1646 fl4.flowi4_oif = rt->dst.dev->ifindex;
1647 fl4.flowi4_iif = skb->dev->ifindex;
1648 fl4.flowi4_mark = skb->mark;
1651 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1652 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1654 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1658 memcpy(addr, &src, 4);
1661 #ifdef CONFIG_IP_ROUTE_CLASSID
1662 static void set_class_tag(struct rtable *rt, u32 tag)
1664 if (!(rt->dst.tclassid & 0xFFFF))
1665 rt->dst.tclassid |= tag & 0xFFFF;
1666 if (!(rt->dst.tclassid & 0xFFFF0000))
1667 rt->dst.tclassid |= tag & 0xFFFF0000;
1671 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1673 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1676 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1678 if (advmss > 65535 - 40)
1679 advmss = 65535 - 40;
1684 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1686 const struct rtable *rt = (const struct rtable *) dst;
1687 unsigned int mtu = rt->rt_pmtu;
1689 if (mtu && time_after_eq(jiffies, rt->dst.expires))
1693 mtu = dst_metric_raw(dst, RTAX_MTU);
1695 if (mtu && rt_is_output_route(rt))
1698 mtu = dst->dev->mtu;
1700 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1702 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1706 if (mtu > IP_MAX_MTU)
1712 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1713 struct fib_info *fi)
1715 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1717 atomic_inc(&fi->fib_clntref);
1719 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1722 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1723 const struct fib_result *res,
1724 struct fib_info *fi, u16 type, u32 itag)
1727 if (FIB_RES_GW(*res) &&
1728 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1729 rt->rt_gateway = FIB_RES_GW(*res);
1730 rt_init_metrics(rt, fl4, fi);
1731 #ifdef CONFIG_IP_ROUTE_CLASSID
1732 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1736 #ifdef CONFIG_IP_ROUTE_CLASSID
1737 #ifdef CONFIG_IP_MULTIPLE_TABLES
1738 set_class_tag(rt, fib_rules_tclass(res));
1740 set_class_tag(rt, itag);
1744 static struct rtable *rt_dst_alloc(struct net_device *dev,
1745 bool nopolicy, bool noxfrm)
1747 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1749 (nopolicy ? DST_NOPOLICY : 0) |
1750 (noxfrm ? DST_NOXFRM : 0));
1753 /* called in rcu_read_lock() section */
1754 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1755 u8 tos, struct net_device *dev, int our)
1759 struct in_device *in_dev = __in_dev_get_rcu(dev);
1763 /* Primary sanity checks. */
1768 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1769 skb->protocol != htons(ETH_P_IP))
1772 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1773 if (ipv4_is_loopback(saddr))
1776 if (ipv4_is_zeronet(saddr)) {
1777 if (!ipv4_is_local_multicast(daddr))
1780 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1785 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1786 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1790 #ifdef CONFIG_IP_ROUTE_CLASSID
1791 rth->dst.tclassid = itag;
1793 rth->dst.output = ip_rt_bug;
1795 rth->rt_key_dst = daddr;
1796 rth->rt_key_src = saddr;
1797 rth->rt_genid = rt_genid(dev_net(dev));
1798 rth->rt_flags = RTCF_MULTICAST;
1799 rth->rt_type = RTN_MULTICAST;
1800 rth->rt_key_tos = tos;
1801 rth->rt_dst = daddr;
1802 rth->rt_src = saddr;
1803 rth->rt_route_iif = dev->ifindex;
1804 rth->rt_iif = dev->ifindex;
1806 rth->rt_mark = skb->mark;
1808 rth->rt_gateway = daddr;
1811 rth->dst.input= ip_local_deliver;
1812 rth->rt_flags |= RTCF_LOCAL;
1815 #ifdef CONFIG_IP_MROUTE
1816 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1817 rth->dst.input = ip_mr_input;
1819 RT_CACHE_STAT_INC(in_slow_mc);
1821 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1822 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1823 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1834 static void ip_handle_martian_source(struct net_device *dev,
1835 struct in_device *in_dev,
1836 struct sk_buff *skb,
1840 RT_CACHE_STAT_INC(in_martian_src);
1841 #ifdef CONFIG_IP_ROUTE_VERBOSE
1842 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1844 * RFC1812 recommendation, if source is martian,
1845 * the only hint is MAC header.
1847 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1848 &daddr, &saddr, dev->name);
1849 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1850 print_hex_dump(KERN_WARNING, "ll header: ",
1851 DUMP_PREFIX_OFFSET, 16, 1,
1852 skb_mac_header(skb),
1853 dev->hard_header_len, true);
1859 /* called in rcu_read_lock() section */
1860 static int __mkroute_input(struct sk_buff *skb,
1861 const struct fib_result *res,
1862 struct in_device *in_dev,
1863 __be32 daddr, __be32 saddr, u32 tos,
1864 struct rtable **result)
1868 struct in_device *out_dev;
1869 unsigned int flags = 0;
1872 /* get a working reference to the output device */
1873 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1874 if (out_dev == NULL) {
1875 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1880 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1881 in_dev->dev, in_dev, &itag);
1883 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1890 flags |= RTCF_DIRECTSRC;
1892 if (out_dev == in_dev && err &&
1893 (IN_DEV_SHARED_MEDIA(out_dev) ||
1894 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1895 flags |= RTCF_DOREDIRECT;
1897 if (skb->protocol != htons(ETH_P_IP)) {
1898 /* Not IP (i.e. ARP). Do not create route, if it is
1899 * invalid for proxy arp. DNAT routes are always valid.
1901 * Proxy arp feature have been extended to allow, ARP
1902 * replies back to the same interface, to support
1903 * Private VLAN switch technologies. See arp.c.
1905 if (out_dev == in_dev &&
1906 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1912 rth = rt_dst_alloc(out_dev->dev,
1913 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1914 IN_DEV_CONF_GET(out_dev, NOXFRM));
1920 rth->rt_key_dst = daddr;
1921 rth->rt_key_src = saddr;
1922 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1923 rth->rt_flags = flags;
1924 rth->rt_type = res->type;
1925 rth->rt_key_tos = tos;
1926 rth->rt_dst = daddr;
1927 rth->rt_src = saddr;
1928 rth->rt_route_iif = in_dev->dev->ifindex;
1929 rth->rt_iif = in_dev->dev->ifindex;
1931 rth->rt_mark = skb->mark;
1933 rth->rt_gateway = daddr;
1936 rth->dst.input = ip_forward;
1937 rth->dst.output = ip_output;
1939 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
1947 static int ip_mkroute_input(struct sk_buff *skb,
1948 struct fib_result *res,
1949 const struct flowi4 *fl4,
1950 struct in_device *in_dev,
1951 __be32 daddr, __be32 saddr, u32 tos)
1953 struct rtable *rth = NULL;
1957 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1958 if (res->fi && res->fi->fib_nhs > 1)
1959 fib_select_multipath(res);
1962 /* create a routing cache entry */
1963 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1967 /* put it into the cache */
1968 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
1969 rt_genid(dev_net(rth->dst.dev)));
1970 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
1972 return PTR_ERR(rth);
1977 * NOTE. We drop all the packets that has local source
1978 * addresses, because every properly looped back packet
1979 * must have correct destination already attached by output routine.
1981 * Such approach solves two big problems:
1982 * 1. Not simplex devices are handled properly.
1983 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1984 * called with rcu_read_lock()
1987 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1988 u8 tos, struct net_device *dev)
1990 struct fib_result res;
1991 struct in_device *in_dev = __in_dev_get_rcu(dev);
1993 unsigned int flags = 0;
1998 struct net *net = dev_net(dev);
2000 /* IP on this device is disabled. */
2005 /* Check for the most weird martians, which can be not detected
2009 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2010 goto martian_source;
2012 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2015 /* Accept zero addresses only to limited broadcast;
2016 * I even do not know to fix it or not. Waiting for complains :-)
2018 if (ipv4_is_zeronet(saddr))
2019 goto martian_source;
2021 if (ipv4_is_zeronet(daddr))
2022 goto martian_destination;
2024 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2025 if (ipv4_is_loopback(daddr))
2026 goto martian_destination;
2028 if (ipv4_is_loopback(saddr))
2029 goto martian_source;
2033 * Now we are ready to route packet.
2036 fl4.flowi4_iif = dev->ifindex;
2037 fl4.flowi4_mark = skb->mark;
2038 fl4.flowi4_tos = tos;
2039 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2042 err = fib_lookup(net, &fl4, &res);
2046 RT_CACHE_STAT_INC(in_slow_tot);
2048 if (res.type == RTN_BROADCAST)
2051 if (res.type == RTN_LOCAL) {
2052 err = fib_validate_source(skb, saddr, daddr, tos,
2053 net->loopback_dev->ifindex,
2054 dev, in_dev, &itag);
2056 goto martian_source_keep_err;
2058 flags |= RTCF_DIRECTSRC;
2062 if (!IN_DEV_FORWARD(in_dev))
2064 if (res.type != RTN_UNICAST)
2065 goto martian_destination;
2067 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2071 if (skb->protocol != htons(ETH_P_IP))
2074 if (!ipv4_is_zeronet(saddr)) {
2075 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2078 goto martian_source_keep_err;
2080 flags |= RTCF_DIRECTSRC;
2082 flags |= RTCF_BROADCAST;
2083 res.type = RTN_BROADCAST;
2084 RT_CACHE_STAT_INC(in_brd);
2087 rth = rt_dst_alloc(net->loopback_dev,
2088 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2092 rth->dst.input= ip_local_deliver;
2093 rth->dst.output= ip_rt_bug;
2094 #ifdef CONFIG_IP_ROUTE_CLASSID
2095 rth->dst.tclassid = itag;
2098 rth->rt_key_dst = daddr;
2099 rth->rt_key_src = saddr;
2100 rth->rt_genid = rt_genid(net);
2101 rth->rt_flags = flags|RTCF_LOCAL;
2102 rth->rt_type = res.type;
2103 rth->rt_key_tos = tos;
2104 rth->rt_dst = daddr;
2105 rth->rt_src = saddr;
2106 rth->rt_route_iif = dev->ifindex;
2107 rth->rt_iif = dev->ifindex;
2109 rth->rt_mark = skb->mark;
2111 rth->rt_gateway = daddr;
2113 if (res.type == RTN_UNREACHABLE) {
2114 rth->dst.input= ip_error;
2115 rth->dst.error= -err;
2116 rth->rt_flags &= ~RTCF_LOCAL;
2118 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2119 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2126 RT_CACHE_STAT_INC(in_no_route);
2127 res.type = RTN_UNREACHABLE;
2133 * Do not cache martian addresses: they should be logged (RFC1812)
2135 martian_destination:
2136 RT_CACHE_STAT_INC(in_martian_dst);
2137 #ifdef CONFIG_IP_ROUTE_VERBOSE
2138 if (IN_DEV_LOG_MARTIANS(in_dev))
2139 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2140 &daddr, &saddr, dev->name);
2153 martian_source_keep_err:
2154 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2158 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2159 u8 tos, struct net_device *dev, bool noref)
2163 int iif = dev->ifindex;
2171 if (!rt_caching(net))
2174 tos &= IPTOS_RT_MASK;
2175 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2177 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2178 rth = rcu_dereference(rth->dst.rt_next)) {
2179 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2180 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2181 (rth->rt_route_iif ^ iif) |
2182 (rth->rt_key_tos ^ tos)) == 0 &&
2183 rth->rt_mark == skb->mark &&
2184 net_eq(dev_net(rth->dst.dev), net) &&
2185 !rt_is_expired(rth)) {
2187 dst_use_noref(&rth->dst, jiffies);
2188 skb_dst_set_noref(skb, &rth->dst);
2190 dst_use(&rth->dst, jiffies);
2191 skb_dst_set(skb, &rth->dst);
2193 RT_CACHE_STAT_INC(in_hit);
2197 RT_CACHE_STAT_INC(in_hlist_search);
2201 /* Multicast recognition logic is moved from route cache to here.
2202 The problem was that too many Ethernet cards have broken/missing
2203 hardware multicast filters :-( As result the host on multicasting
2204 network acquires a lot of useless route cache entries, sort of
2205 SDR messages from all the world. Now we try to get rid of them.
2206 Really, provided software IP multicast filter is organized
2207 reasonably (at least, hashed), it does not result in a slowdown
2208 comparing with route cache reject entries.
2209 Note, that multicast routers are not affected, because
2210 route cache entry is created eventually.
2212 if (ipv4_is_multicast(daddr)) {
2213 struct in_device *in_dev = __in_dev_get_rcu(dev);
2216 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2217 ip_hdr(skb)->protocol);
2219 #ifdef CONFIG_IP_MROUTE
2221 (!ipv4_is_local_multicast(daddr) &&
2222 IN_DEV_MFORWARD(in_dev))
2225 int res = ip_route_input_mc(skb, daddr, saddr,
2234 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2238 EXPORT_SYMBOL(ip_route_input_common);
2240 /* called with rcu_read_lock() */
2241 static struct rtable *__mkroute_output(const struct fib_result *res,
2242 const struct flowi4 *fl4,
2243 __be32 orig_daddr, __be32 orig_saddr,
2244 int orig_oif, __u8 orig_rtos,
2245 struct net_device *dev_out,
2248 struct fib_info *fi = res->fi;
2249 struct in_device *in_dev;
2250 u16 type = res->type;
2253 in_dev = __in_dev_get_rcu(dev_out);
2255 return ERR_PTR(-EINVAL);
2257 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2258 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2259 return ERR_PTR(-EINVAL);
2261 if (ipv4_is_lbcast(fl4->daddr))
2262 type = RTN_BROADCAST;
2263 else if (ipv4_is_multicast(fl4->daddr))
2264 type = RTN_MULTICAST;
2265 else if (ipv4_is_zeronet(fl4->daddr))
2266 return ERR_PTR(-EINVAL);
2268 if (dev_out->flags & IFF_LOOPBACK)
2269 flags |= RTCF_LOCAL;
2271 if (type == RTN_BROADCAST) {
2272 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2274 } else if (type == RTN_MULTICAST) {
2275 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2276 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2278 flags &= ~RTCF_LOCAL;
2279 /* If multicast route do not exist use
2280 * default one, but do not gateway in this case.
2283 if (fi && res->prefixlen < 4)
2287 rth = rt_dst_alloc(dev_out,
2288 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2289 IN_DEV_CONF_GET(in_dev, NOXFRM));
2291 return ERR_PTR(-ENOBUFS);
2293 rth->dst.output = ip_output;
2295 rth->rt_key_dst = orig_daddr;
2296 rth->rt_key_src = orig_saddr;
2297 rth->rt_genid = rt_genid(dev_net(dev_out));
2298 rth->rt_flags = flags;
2299 rth->rt_type = type;
2300 rth->rt_key_tos = orig_rtos;
2301 rth->rt_dst = fl4->daddr;
2302 rth->rt_src = fl4->saddr;
2303 rth->rt_route_iif = 0;
2304 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2305 rth->rt_oif = orig_oif;
2306 rth->rt_mark = fl4->flowi4_mark;
2308 rth->rt_gateway = fl4->daddr;
2311 RT_CACHE_STAT_INC(out_slow_tot);
2313 if (flags & RTCF_LOCAL)
2314 rth->dst.input = ip_local_deliver;
2315 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2316 if (flags & RTCF_LOCAL &&
2317 !(dev_out->flags & IFF_LOOPBACK)) {
2318 rth->dst.output = ip_mc_output;
2319 RT_CACHE_STAT_INC(out_slow_mc);
2321 #ifdef CONFIG_IP_MROUTE
2322 if (type == RTN_MULTICAST) {
2323 if (IN_DEV_MFORWARD(in_dev) &&
2324 !ipv4_is_local_multicast(fl4->daddr)) {
2325 rth->dst.input = ip_mr_input;
2326 rth->dst.output = ip_mc_output;
2332 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2334 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2335 rth->dst.flags |= DST_NOCACHE;
2341 * Major route resolver routine.
2342 * called with rcu_read_lock();
2345 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2347 struct net_device *dev_out = NULL;
2348 __u8 tos = RT_FL_TOS(fl4);
2349 unsigned int flags = 0;
2350 struct fib_result res;
2358 #ifdef CONFIG_IP_MULTIPLE_TABLES
2362 orig_daddr = fl4->daddr;
2363 orig_saddr = fl4->saddr;
2364 orig_oif = fl4->flowi4_oif;
2366 fl4->flowi4_iif = net->loopback_dev->ifindex;
2367 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2368 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2369 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2373 rth = ERR_PTR(-EINVAL);
2374 if (ipv4_is_multicast(fl4->saddr) ||
2375 ipv4_is_lbcast(fl4->saddr) ||
2376 ipv4_is_zeronet(fl4->saddr))
2379 /* I removed check for oif == dev_out->oif here.
2380 It was wrong for two reasons:
2381 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2382 is assigned to multiple interfaces.
2383 2. Moreover, we are allowed to send packets with saddr
2384 of another iface. --ANK
2387 if (fl4->flowi4_oif == 0 &&
2388 (ipv4_is_multicast(fl4->daddr) ||
2389 ipv4_is_lbcast(fl4->daddr))) {
2390 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2391 dev_out = __ip_dev_find(net, fl4->saddr, false);
2392 if (dev_out == NULL)
2395 /* Special hack: user can direct multicasts
2396 and limited broadcast via necessary interface
2397 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2398 This hack is not just for fun, it allows
2399 vic,vat and friends to work.
2400 They bind socket to loopback, set ttl to zero
2401 and expect that it will work.
2402 From the viewpoint of routing cache they are broken,
2403 because we are not allowed to build multicast path
2404 with loopback source addr (look, routing cache
2405 cannot know, that ttl is zero, so that packet
2406 will not leave this host and route is valid).
2407 Luckily, this hack is good workaround.
2410 fl4->flowi4_oif = dev_out->ifindex;
2414 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2415 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2416 if (!__ip_dev_find(net, fl4->saddr, false))
2422 if (fl4->flowi4_oif) {
2423 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2424 rth = ERR_PTR(-ENODEV);
2425 if (dev_out == NULL)
2428 /* RACE: Check return value of inet_select_addr instead. */
2429 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2430 rth = ERR_PTR(-ENETUNREACH);
2433 if (ipv4_is_local_multicast(fl4->daddr) ||
2434 ipv4_is_lbcast(fl4->daddr)) {
2436 fl4->saddr = inet_select_addr(dev_out, 0,
2441 if (ipv4_is_multicast(fl4->daddr))
2442 fl4->saddr = inet_select_addr(dev_out, 0,
2444 else if (!fl4->daddr)
2445 fl4->saddr = inet_select_addr(dev_out, 0,
2451 fl4->daddr = fl4->saddr;
2453 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2454 dev_out = net->loopback_dev;
2455 fl4->flowi4_oif = net->loopback_dev->ifindex;
2456 res.type = RTN_LOCAL;
2457 flags |= RTCF_LOCAL;
2461 if (fib_lookup(net, fl4, &res)) {
2464 if (fl4->flowi4_oif) {
2465 /* Apparently, routing tables are wrong. Assume,
2466 that the destination is on link.
2469 Because we are allowed to send to iface
2470 even if it has NO routes and NO assigned
2471 addresses. When oif is specified, routing
2472 tables are looked up with only one purpose:
2473 to catch if destination is gatewayed, rather than
2474 direct. Moreover, if MSG_DONTROUTE is set,
2475 we send packet, ignoring both routing tables
2476 and ifaddr state. --ANK
2479 We could make it even if oif is unknown,
2480 likely IPv6, but we do not.
2483 if (fl4->saddr == 0)
2484 fl4->saddr = inet_select_addr(dev_out, 0,
2486 res.type = RTN_UNICAST;
2489 rth = ERR_PTR(-ENETUNREACH);
2493 if (res.type == RTN_LOCAL) {
2495 if (res.fi->fib_prefsrc)
2496 fl4->saddr = res.fi->fib_prefsrc;
2498 fl4->saddr = fl4->daddr;
2500 dev_out = net->loopback_dev;
2501 fl4->flowi4_oif = dev_out->ifindex;
2503 flags |= RTCF_LOCAL;
2507 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2508 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2509 fib_select_multipath(&res);
2512 if (!res.prefixlen &&
2513 res.table->tb_num_default > 1 &&
2514 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2515 fib_select_default(&res);
2518 fl4->saddr = FIB_RES_PREFSRC(net, res);
2520 dev_out = FIB_RES_DEV(res);
2521 fl4->flowi4_oif = dev_out->ifindex;
2525 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2526 tos, dev_out, flags);
2530 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2531 rt_genid(dev_net(dev_out)));
2532 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2540 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2545 if (!rt_caching(net))
2548 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2551 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2552 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2553 if (rth->rt_key_dst == flp4->daddr &&
2554 rth->rt_key_src == flp4->saddr &&
2555 rt_is_output_route(rth) &&
2556 rth->rt_oif == flp4->flowi4_oif &&
2557 rth->rt_mark == flp4->flowi4_mark &&
2558 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2559 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2560 net_eq(dev_net(rth->dst.dev), net) &&
2561 !rt_is_expired(rth)) {
2562 dst_use(&rth->dst, jiffies);
2563 RT_CACHE_STAT_INC(out_hit);
2564 rcu_read_unlock_bh();
2566 flp4->saddr = rth->rt_src;
2568 flp4->daddr = rth->rt_dst;
2571 RT_CACHE_STAT_INC(out_hlist_search);
2573 rcu_read_unlock_bh();
2576 return ip_route_output_slow(net, flp4);
2578 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2580 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2585 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2587 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2589 return mtu ? : dst->dev->mtu;
2592 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2596 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sk_buff *skb)
2600 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2606 static struct dst_ops ipv4_dst_blackhole_ops = {
2608 .protocol = cpu_to_be16(ETH_P_IP),
2609 .destroy = ipv4_dst_destroy,
2610 .check = ipv4_blackhole_dst_check,
2611 .mtu = ipv4_blackhole_mtu,
2612 .default_advmss = ipv4_default_advmss,
2613 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2614 .redirect = ipv4_rt_blackhole_redirect,
2615 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2616 .neigh_lookup = ipv4_neigh_lookup,
2619 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2621 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2622 struct rtable *ort = (struct rtable *) dst_orig;
2625 struct dst_entry *new = &rt->dst;
2628 new->input = dst_discard;
2629 new->output = dst_discard;
2631 new->dev = ort->dst.dev;
2635 rt->rt_key_dst = ort->rt_key_dst;
2636 rt->rt_key_src = ort->rt_key_src;
2637 rt->rt_key_tos = ort->rt_key_tos;
2638 rt->rt_route_iif = ort->rt_route_iif;
2639 rt->rt_iif = ort->rt_iif;
2640 rt->rt_oif = ort->rt_oif;
2641 rt->rt_mark = ort->rt_mark;
2642 rt->rt_pmtu = ort->rt_pmtu;
2644 rt->rt_genid = rt_genid(net);
2645 rt->rt_flags = ort->rt_flags;
2646 rt->rt_type = ort->rt_type;
2647 rt->rt_dst = ort->rt_dst;
2648 rt->rt_src = ort->rt_src;
2649 rt->rt_gateway = ort->rt_gateway;
2652 atomic_inc(&rt->fi->fib_clntref);
2657 dst_release(dst_orig);
2659 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2662 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2665 struct rtable *rt = __ip_route_output_key(net, flp4);
2670 if (flp4->flowi4_proto)
2671 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2672 flowi4_to_flowi(flp4),
2677 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2679 static int rt_fill_info(struct net *net,
2680 struct sk_buff *skb, u32 pid, u32 seq, int event,
2681 int nowait, unsigned int flags)
2683 struct rtable *rt = skb_rtable(skb);
2685 struct nlmsghdr *nlh;
2686 unsigned long expires = 0;
2689 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2693 r = nlmsg_data(nlh);
2694 r->rtm_family = AF_INET;
2695 r->rtm_dst_len = 32;
2697 r->rtm_tos = rt->rt_key_tos;
2698 r->rtm_table = RT_TABLE_MAIN;
2699 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2700 goto nla_put_failure;
2701 r->rtm_type = rt->rt_type;
2702 r->rtm_scope = RT_SCOPE_UNIVERSE;
2703 r->rtm_protocol = RTPROT_UNSPEC;
2704 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2705 if (rt->rt_flags & RTCF_NOTIFY)
2706 r->rtm_flags |= RTM_F_NOTIFY;
2708 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2709 goto nla_put_failure;
2710 if (rt->rt_key_src) {
2711 r->rtm_src_len = 32;
2712 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2713 goto nla_put_failure;
2716 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2717 goto nla_put_failure;
2718 #ifdef CONFIG_IP_ROUTE_CLASSID
2719 if (rt->dst.tclassid &&
2720 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2721 goto nla_put_failure;
2723 if (!rt_is_input_route(rt) &&
2724 rt->rt_src != rt->rt_key_src) {
2725 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2726 goto nla_put_failure;
2728 if (rt->rt_dst != rt->rt_gateway &&
2729 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2730 goto nla_put_failure;
2732 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2733 goto nla_put_failure;
2736 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2737 goto nla_put_failure;
2739 error = rt->dst.error;
2740 expires = rt->dst.expires;
2742 if (time_before(jiffies, expires))
2748 if (rt_is_input_route(rt)) {
2749 #ifdef CONFIG_IP_MROUTE
2750 __be32 dst = rt->rt_dst;
2752 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2753 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2754 int err = ipmr_get_route(net, skb,
2755 rt->rt_src, rt->rt_dst,
2761 goto nla_put_failure;
2763 if (err == -EMSGSIZE)
2764 goto nla_put_failure;
2770 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2771 goto nla_put_failure;
2774 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2775 goto nla_put_failure;
2777 return nlmsg_end(skb, nlh);
2780 nlmsg_cancel(skb, nlh);
2784 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2786 struct net *net = sock_net(in_skb->sk);
2788 struct nlattr *tb[RTA_MAX+1];
2789 struct rtable *rt = NULL;
2795 struct sk_buff *skb;
2797 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2801 rtm = nlmsg_data(nlh);
2803 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2809 /* Reserve room for dummy headers, this skb can pass
2810 through good chunk of routing engine.
2812 skb_reset_mac_header(skb);
2813 skb_reset_network_header(skb);
2815 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2816 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2817 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2819 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2820 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2821 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2822 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2825 struct net_device *dev;
2827 dev = __dev_get_by_index(net, iif);
2833 skb->protocol = htons(ETH_P_IP);
2837 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2840 rt = skb_rtable(skb);
2841 if (err == 0 && rt->dst.error)
2842 err = -rt->dst.error;
2844 struct flowi4 fl4 = {
2847 .flowi4_tos = rtm->rtm_tos,
2848 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2849 .flowi4_mark = mark,
2851 rt = ip_route_output_key(net, &fl4);
2861 skb_dst_set(skb, &rt->dst);
2862 if (rtm->rtm_flags & RTM_F_NOTIFY)
2863 rt->rt_flags |= RTCF_NOTIFY;
2865 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2866 RTM_NEWROUTE, 0, 0);
2870 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2879 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2886 net = sock_net(skb->sk);
2891 s_idx = idx = cb->args[1];
2892 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2893 if (!rt_hash_table[h].chain)
2896 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2897 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2898 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2900 if (rt_is_expired(rt))
2902 skb_dst_set_noref(skb, &rt->dst);
2903 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2904 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2905 1, NLM_F_MULTI) <= 0) {
2907 rcu_read_unlock_bh();
2912 rcu_read_unlock_bh();
2921 void ip_rt_multicast_event(struct in_device *in_dev)
2923 rt_cache_flush(dev_net(in_dev->dev), 0);
2926 #ifdef CONFIG_SYSCTL
2927 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2928 void __user *buffer,
2929 size_t *lenp, loff_t *ppos)
2936 memcpy(&ctl, __ctl, sizeof(ctl));
2937 ctl.data = &flush_delay;
2938 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2940 net = (struct net *)__ctl->extra1;
2941 rt_cache_flush(net, flush_delay);
2948 static ctl_table ipv4_route_table[] = {
2950 .procname = "gc_thresh",
2951 .data = &ipv4_dst_ops.gc_thresh,
2952 .maxlen = sizeof(int),
2954 .proc_handler = proc_dointvec,
2957 .procname = "max_size",
2958 .data = &ip_rt_max_size,
2959 .maxlen = sizeof(int),
2961 .proc_handler = proc_dointvec,
2964 /* Deprecated. Use gc_min_interval_ms */
2966 .procname = "gc_min_interval",
2967 .data = &ip_rt_gc_min_interval,
2968 .maxlen = sizeof(int),
2970 .proc_handler = proc_dointvec_jiffies,
2973 .procname = "gc_min_interval_ms",
2974 .data = &ip_rt_gc_min_interval,
2975 .maxlen = sizeof(int),
2977 .proc_handler = proc_dointvec_ms_jiffies,
2980 .procname = "gc_timeout",
2981 .data = &ip_rt_gc_timeout,
2982 .maxlen = sizeof(int),
2984 .proc_handler = proc_dointvec_jiffies,
2987 .procname = "gc_interval",
2988 .data = &ip_rt_gc_interval,
2989 .maxlen = sizeof(int),
2991 .proc_handler = proc_dointvec_jiffies,
2994 .procname = "redirect_load",
2995 .data = &ip_rt_redirect_load,
2996 .maxlen = sizeof(int),
2998 .proc_handler = proc_dointvec,
3001 .procname = "redirect_number",
3002 .data = &ip_rt_redirect_number,
3003 .maxlen = sizeof(int),
3005 .proc_handler = proc_dointvec,
3008 .procname = "redirect_silence",
3009 .data = &ip_rt_redirect_silence,
3010 .maxlen = sizeof(int),
3012 .proc_handler = proc_dointvec,
3015 .procname = "error_cost",
3016 .data = &ip_rt_error_cost,
3017 .maxlen = sizeof(int),
3019 .proc_handler = proc_dointvec,
3022 .procname = "error_burst",
3023 .data = &ip_rt_error_burst,
3024 .maxlen = sizeof(int),
3026 .proc_handler = proc_dointvec,
3029 .procname = "gc_elasticity",
3030 .data = &ip_rt_gc_elasticity,
3031 .maxlen = sizeof(int),
3033 .proc_handler = proc_dointvec,
3036 .procname = "mtu_expires",
3037 .data = &ip_rt_mtu_expires,
3038 .maxlen = sizeof(int),
3040 .proc_handler = proc_dointvec_jiffies,
3043 .procname = "min_pmtu",
3044 .data = &ip_rt_min_pmtu,
3045 .maxlen = sizeof(int),
3047 .proc_handler = proc_dointvec,
3050 .procname = "min_adv_mss",
3051 .data = &ip_rt_min_advmss,
3052 .maxlen = sizeof(int),
3054 .proc_handler = proc_dointvec,
3059 static struct ctl_table ipv4_route_flush_table[] = {
3061 .procname = "flush",
3062 .maxlen = sizeof(int),
3064 .proc_handler = ipv4_sysctl_rtcache_flush,
3069 static __net_init int sysctl_route_net_init(struct net *net)
3071 struct ctl_table *tbl;
3073 tbl = ipv4_route_flush_table;
3074 if (!net_eq(net, &init_net)) {
3075 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3079 tbl[0].extra1 = net;
3081 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3082 if (net->ipv4.route_hdr == NULL)
3087 if (tbl != ipv4_route_flush_table)
3093 static __net_exit void sysctl_route_net_exit(struct net *net)
3095 struct ctl_table *tbl;
3097 tbl = net->ipv4.route_hdr->ctl_table_arg;
3098 unregister_net_sysctl_table(net->ipv4.route_hdr);
3099 BUG_ON(tbl == ipv4_route_flush_table);
3103 static __net_initdata struct pernet_operations sysctl_route_ops = {
3104 .init = sysctl_route_net_init,
3105 .exit = sysctl_route_net_exit,
3109 static __net_init int rt_genid_init(struct net *net)
3111 get_random_bytes(&net->ipv4.rt_genid,
3112 sizeof(net->ipv4.rt_genid));
3113 get_random_bytes(&net->ipv4.dev_addr_genid,
3114 sizeof(net->ipv4.dev_addr_genid));
3118 static __net_initdata struct pernet_operations rt_genid_ops = {
3119 .init = rt_genid_init,
3122 static int __net_init ipv4_inetpeer_init(struct net *net)
3124 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3128 inet_peer_base_init(bp);
3129 net->ipv4.peers = bp;
3133 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3135 struct inet_peer_base *bp = net->ipv4.peers;
3137 net->ipv4.peers = NULL;
3138 inetpeer_invalidate_tree(bp);
3142 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3143 .init = ipv4_inetpeer_init,
3144 .exit = ipv4_inetpeer_exit,
3147 #ifdef CONFIG_IP_ROUTE_CLASSID
3148 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3149 #endif /* CONFIG_IP_ROUTE_CLASSID */
3151 static __initdata unsigned long rhash_entries;
3152 static int __init set_rhash_entries(char *str)
3159 ret = kstrtoul(str, 0, &rhash_entries);
3165 __setup("rhash_entries=", set_rhash_entries);
3167 int __init ip_rt_init(void)
3171 #ifdef CONFIG_IP_ROUTE_CLASSID
3172 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3174 panic("IP: failed to allocate ip_rt_acct\n");
3177 ipv4_dst_ops.kmem_cachep =
3178 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3179 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3181 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3183 if (dst_entries_init(&ipv4_dst_ops) < 0)
3184 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3186 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3187 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3189 rt_hash_table = (struct rt_hash_bucket *)
3190 alloc_large_system_hash("IP route cache",
3191 sizeof(struct rt_hash_bucket),
3193 (totalram_pages >= 128 * 1024) ?
3199 rhash_entries ? 0 : 512 * 1024);
3200 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3201 rt_hash_lock_init();
3203 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3204 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3209 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3210 expires_ljiffies = jiffies;
3211 schedule_delayed_work(&expires_work,
3212 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3214 if (ip_rt_proc_init())
3215 pr_err("Unable to create route proc files\n");
3218 xfrm4_init(ip_rt_max_size);
3220 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3222 #ifdef CONFIG_SYSCTL
3223 register_pernet_subsys(&sysctl_route_ops);
3225 register_pernet_subsys(&rt_genid_ops);
3226 register_pernet_subsys(&ipv4_inetpeer_ops);
3230 #ifdef CONFIG_SYSCTL
3232 * We really need to sanitize the damn ipv4 init order, then all
3233 * this nonsense will go away.
3235 void __init ip_static_sysctl_init(void)
3237 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);