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 const struct iphdr *iph = (const struct iphdr *) skb->data;
1279 __be32 new_gw = icmp_hdr(skb)->un.gateway;
1280 __be32 old_gw = ip_hdr(skb)->saddr;
1281 struct net_device *dev = skb->dev;
1282 __be32 daddr = iph->daddr;
1283 __be32 saddr = iph->saddr;
1284 struct in_device *in_dev;
1285 struct neighbour *n;
1289 switch (icmp_hdr(skb)->code & 7) {
1290 case ICMP_REDIR_NET:
1291 case ICMP_REDIR_NETTOS:
1292 case ICMP_REDIR_HOST:
1293 case ICMP_REDIR_HOSTTOS:
1300 rt = (struct rtable *) dst;
1301 if (rt->rt_gateway != old_gw)
1304 in_dev = __in_dev_get_rcu(dev);
1309 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1310 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1311 ipv4_is_zeronet(new_gw))
1312 goto reject_redirect;
1314 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1315 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1316 goto reject_redirect;
1317 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1318 goto reject_redirect;
1320 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1321 goto reject_redirect;
1324 n = ipv4_neigh_lookup(dst, NULL, &new_gw);
1326 if (!(n->nud_state & NUD_VALID)) {
1327 neigh_event_send(n, NULL);
1329 rt->rt_gateway = new_gw;
1330 rt->rt_flags |= RTCF_REDIRECTED;
1331 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1338 #ifdef CONFIG_IP_ROUTE_VERBOSE
1339 if (IN_DEV_LOG_MARTIANS(in_dev))
1340 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1341 " Advised path = %pI4 -> %pI4\n",
1342 &old_gw, dev->name, &new_gw,
1348 /* called in rcu_read_lock() section */
1349 void ip_rt_redirect(struct sk_buff *skb, __be32 new_gw)
1351 const struct iphdr *iph = (const struct iphdr *) skb->data;
1352 __be32 daddr = iph->daddr;
1353 __be32 saddr = iph->saddr;
1354 struct net_device *dev = skb->dev;
1355 int ikeys[2] = { dev->ifindex, 0 };
1356 __be32 skeys[2] = { saddr, 0 };
1361 for (s = 0; s < 2; s++) {
1362 for (i = 0; i < 2; i++) {
1364 struct rtable __rcu **rthp;
1367 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1369 rthp = &rt_hash_table[hash].chain;
1371 while ((rt = rcu_dereference(*rthp)) != NULL) {
1372 rthp = &rt->dst.rt_next;
1374 if (rt->rt_key_dst != daddr ||
1375 rt->rt_key_src != skeys[s] ||
1376 rt->rt_oif != ikeys[i] ||
1377 rt_is_input_route(rt) ||
1378 rt_is_expired(rt) ||
1379 !net_eq(dev_net(rt->dst.dev), net) ||
1384 ip_do_redirect(&rt->dst, skb);
1392 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1394 struct rtable *rt = (struct rtable *)dst;
1395 struct dst_entry *ret = dst;
1398 if (dst->obsolete > 0) {
1401 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1403 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1405 rt_genid(dev_net(dst->dev)));
1415 * 1. The first ip_rt_redirect_number redirects are sent
1416 * with exponential backoff, then we stop sending them at all,
1417 * assuming that the host ignores our redirects.
1418 * 2. If we did not see packets requiring redirects
1419 * during ip_rt_redirect_silence, we assume that the host
1420 * forgot redirected route and start to send redirects again.
1422 * This algorithm is much cheaper and more intelligent than dumb load limiting
1425 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1426 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1429 void ip_rt_send_redirect(struct sk_buff *skb)
1431 struct rtable *rt = skb_rtable(skb);
1432 struct in_device *in_dev;
1433 struct inet_peer *peer;
1438 in_dev = __in_dev_get_rcu(rt->dst.dev);
1439 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1443 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1446 net = dev_net(rt->dst.dev);
1447 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1449 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1453 /* No redirected packets during ip_rt_redirect_silence;
1454 * reset the algorithm.
1456 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1457 peer->rate_tokens = 0;
1459 /* Too many ignored redirects; do not send anything
1460 * set dst.rate_last to the last seen redirected packet.
1462 if (peer->rate_tokens >= ip_rt_redirect_number) {
1463 peer->rate_last = jiffies;
1467 /* Check for load limit; set rate_last to the latest sent
1470 if (peer->rate_tokens == 0 ||
1473 (ip_rt_redirect_load << peer->rate_tokens)))) {
1474 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1475 peer->rate_last = jiffies;
1476 ++peer->rate_tokens;
1477 #ifdef CONFIG_IP_ROUTE_VERBOSE
1479 peer->rate_tokens == ip_rt_redirect_number)
1480 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1481 &ip_hdr(skb)->saddr, rt->rt_iif,
1482 &rt->rt_dst, &rt->rt_gateway);
1489 static int ip_error(struct sk_buff *skb)
1491 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1492 struct rtable *rt = skb_rtable(skb);
1493 struct inet_peer *peer;
1499 net = dev_net(rt->dst.dev);
1500 if (!IN_DEV_FORWARD(in_dev)) {
1501 switch (rt->dst.error) {
1503 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1507 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1513 switch (rt->dst.error) {
1518 code = ICMP_HOST_UNREACH;
1521 code = ICMP_NET_UNREACH;
1522 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1525 code = ICMP_PKT_FILTERED;
1529 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1534 peer->rate_tokens += now - peer->rate_last;
1535 if (peer->rate_tokens > ip_rt_error_burst)
1536 peer->rate_tokens = ip_rt_error_burst;
1537 peer->rate_last = now;
1538 if (peer->rate_tokens >= ip_rt_error_cost)
1539 peer->rate_tokens -= ip_rt_error_cost;
1545 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1547 out: kfree_skb(skb);
1551 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1553 struct rtable *rt = (struct rtable *) dst;
1557 if (mtu < ip_rt_min_pmtu)
1558 mtu = ip_rt_min_pmtu;
1561 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
1564 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1565 int oif, u32 mark, u8 protocol, int flow_flags)
1567 const struct iphdr *iph = (const struct iphdr *)skb->data;
1571 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1572 protocol, flow_flags,
1573 iph->daddr, iph->saddr, 0, 0);
1574 rt = __ip_route_output_key(net, &fl4);
1576 ip_rt_update_pmtu(&rt->dst, mtu);
1580 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1582 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1584 const struct inet_sock *inet = inet_sk(sk);
1586 return ipv4_update_pmtu(skb, sock_net(sk), mtu,
1587 sk->sk_bound_dev_if, sk->sk_mark,
1588 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1589 inet_sk_flowi_flags(sk));
1591 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1593 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1595 struct rtable *rt = (struct rtable *) dst;
1597 if (rt_is_expired(rt))
1602 static void ipv4_dst_destroy(struct dst_entry *dst)
1604 struct rtable *rt = (struct rtable *) dst;
1607 fib_info_put(rt->fi);
1613 static void ipv4_link_failure(struct sk_buff *skb)
1617 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1619 rt = skb_rtable(skb);
1621 dst_set_expires(&rt->dst, 0);
1624 static int ip_rt_bug(struct sk_buff *skb)
1626 pr_debug("%s: %pI4 -> %pI4, %s\n",
1627 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1628 skb->dev ? skb->dev->name : "?");
1635 We do not cache source address of outgoing interface,
1636 because it is used only by IP RR, TS and SRR options,
1637 so that it out of fast path.
1639 BTW remember: "addr" is allowed to be not aligned
1643 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1647 if (rt_is_output_route(rt))
1648 src = ip_hdr(skb)->saddr;
1650 struct fib_result res;
1656 memset(&fl4, 0, sizeof(fl4));
1657 fl4.daddr = iph->daddr;
1658 fl4.saddr = iph->saddr;
1659 fl4.flowi4_tos = RT_TOS(iph->tos);
1660 fl4.flowi4_oif = rt->dst.dev->ifindex;
1661 fl4.flowi4_iif = skb->dev->ifindex;
1662 fl4.flowi4_mark = skb->mark;
1665 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1666 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1668 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1672 memcpy(addr, &src, 4);
1675 #ifdef CONFIG_IP_ROUTE_CLASSID
1676 static void set_class_tag(struct rtable *rt, u32 tag)
1678 if (!(rt->dst.tclassid & 0xFFFF))
1679 rt->dst.tclassid |= tag & 0xFFFF;
1680 if (!(rt->dst.tclassid & 0xFFFF0000))
1681 rt->dst.tclassid |= tag & 0xFFFF0000;
1685 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1687 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1690 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1692 if (advmss > 65535 - 40)
1693 advmss = 65535 - 40;
1698 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1700 const struct rtable *rt = (const struct rtable *) dst;
1701 unsigned int mtu = rt->rt_pmtu;
1703 if (mtu && time_after_eq(jiffies, rt->dst.expires))
1707 mtu = dst_metric_raw(dst, RTAX_MTU);
1709 if (mtu && rt_is_output_route(rt))
1712 mtu = dst->dev->mtu;
1714 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1716 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1720 if (mtu > IP_MAX_MTU)
1726 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1727 struct fib_info *fi)
1729 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1731 atomic_inc(&fi->fib_clntref);
1733 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1736 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1737 const struct fib_result *res,
1738 struct fib_info *fi, u16 type, u32 itag)
1741 if (FIB_RES_GW(*res) &&
1742 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1743 rt->rt_gateway = FIB_RES_GW(*res);
1744 rt_init_metrics(rt, fl4, fi);
1745 #ifdef CONFIG_IP_ROUTE_CLASSID
1746 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1750 #ifdef CONFIG_IP_ROUTE_CLASSID
1751 #ifdef CONFIG_IP_MULTIPLE_TABLES
1752 set_class_tag(rt, fib_rules_tclass(res));
1754 set_class_tag(rt, itag);
1758 static struct rtable *rt_dst_alloc(struct net_device *dev,
1759 bool nopolicy, bool noxfrm)
1761 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1763 (nopolicy ? DST_NOPOLICY : 0) |
1764 (noxfrm ? DST_NOXFRM : 0));
1767 /* called in rcu_read_lock() section */
1768 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1769 u8 tos, struct net_device *dev, int our)
1773 struct in_device *in_dev = __in_dev_get_rcu(dev);
1777 /* Primary sanity checks. */
1782 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1783 skb->protocol != htons(ETH_P_IP))
1786 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1787 if (ipv4_is_loopback(saddr))
1790 if (ipv4_is_zeronet(saddr)) {
1791 if (!ipv4_is_local_multicast(daddr))
1794 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1799 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1800 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1804 #ifdef CONFIG_IP_ROUTE_CLASSID
1805 rth->dst.tclassid = itag;
1807 rth->dst.output = ip_rt_bug;
1809 rth->rt_key_dst = daddr;
1810 rth->rt_key_src = saddr;
1811 rth->rt_genid = rt_genid(dev_net(dev));
1812 rth->rt_flags = RTCF_MULTICAST;
1813 rth->rt_type = RTN_MULTICAST;
1814 rth->rt_key_tos = tos;
1815 rth->rt_dst = daddr;
1816 rth->rt_src = saddr;
1817 rth->rt_route_iif = dev->ifindex;
1818 rth->rt_iif = dev->ifindex;
1820 rth->rt_mark = skb->mark;
1822 rth->rt_gateway = daddr;
1825 rth->dst.input= ip_local_deliver;
1826 rth->rt_flags |= RTCF_LOCAL;
1829 #ifdef CONFIG_IP_MROUTE
1830 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1831 rth->dst.input = ip_mr_input;
1833 RT_CACHE_STAT_INC(in_slow_mc);
1835 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1836 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1837 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1848 static void ip_handle_martian_source(struct net_device *dev,
1849 struct in_device *in_dev,
1850 struct sk_buff *skb,
1854 RT_CACHE_STAT_INC(in_martian_src);
1855 #ifdef CONFIG_IP_ROUTE_VERBOSE
1856 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1858 * RFC1812 recommendation, if source is martian,
1859 * the only hint is MAC header.
1861 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1862 &daddr, &saddr, dev->name);
1863 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1864 print_hex_dump(KERN_WARNING, "ll header: ",
1865 DUMP_PREFIX_OFFSET, 16, 1,
1866 skb_mac_header(skb),
1867 dev->hard_header_len, true);
1873 /* called in rcu_read_lock() section */
1874 static int __mkroute_input(struct sk_buff *skb,
1875 const struct fib_result *res,
1876 struct in_device *in_dev,
1877 __be32 daddr, __be32 saddr, u32 tos,
1878 struct rtable **result)
1882 struct in_device *out_dev;
1883 unsigned int flags = 0;
1886 /* get a working reference to the output device */
1887 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1888 if (out_dev == NULL) {
1889 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1894 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1895 in_dev->dev, in_dev, &itag);
1897 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1904 flags |= RTCF_DIRECTSRC;
1906 if (out_dev == in_dev && err &&
1907 (IN_DEV_SHARED_MEDIA(out_dev) ||
1908 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1909 flags |= RTCF_DOREDIRECT;
1911 if (skb->protocol != htons(ETH_P_IP)) {
1912 /* Not IP (i.e. ARP). Do not create route, if it is
1913 * invalid for proxy arp. DNAT routes are always valid.
1915 * Proxy arp feature have been extended to allow, ARP
1916 * replies back to the same interface, to support
1917 * Private VLAN switch technologies. See arp.c.
1919 if (out_dev == in_dev &&
1920 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1926 rth = rt_dst_alloc(out_dev->dev,
1927 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1928 IN_DEV_CONF_GET(out_dev, NOXFRM));
1934 rth->rt_key_dst = daddr;
1935 rth->rt_key_src = saddr;
1936 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1937 rth->rt_flags = flags;
1938 rth->rt_type = res->type;
1939 rth->rt_key_tos = tos;
1940 rth->rt_dst = daddr;
1941 rth->rt_src = saddr;
1942 rth->rt_route_iif = in_dev->dev->ifindex;
1943 rth->rt_iif = in_dev->dev->ifindex;
1945 rth->rt_mark = skb->mark;
1947 rth->rt_gateway = daddr;
1950 rth->dst.input = ip_forward;
1951 rth->dst.output = ip_output;
1953 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
1961 static int ip_mkroute_input(struct sk_buff *skb,
1962 struct fib_result *res,
1963 const struct flowi4 *fl4,
1964 struct in_device *in_dev,
1965 __be32 daddr, __be32 saddr, u32 tos)
1967 struct rtable *rth = NULL;
1971 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1972 if (res->fi && res->fi->fib_nhs > 1)
1973 fib_select_multipath(res);
1976 /* create a routing cache entry */
1977 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1981 /* put it into the cache */
1982 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
1983 rt_genid(dev_net(rth->dst.dev)));
1984 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
1986 return PTR_ERR(rth);
1991 * NOTE. We drop all the packets that has local source
1992 * addresses, because every properly looped back packet
1993 * must have correct destination already attached by output routine.
1995 * Such approach solves two big problems:
1996 * 1. Not simplex devices are handled properly.
1997 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1998 * called with rcu_read_lock()
2001 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2002 u8 tos, struct net_device *dev)
2004 struct fib_result res;
2005 struct in_device *in_dev = __in_dev_get_rcu(dev);
2007 unsigned int flags = 0;
2012 struct net *net = dev_net(dev);
2014 /* IP on this device is disabled. */
2019 /* Check for the most weird martians, which can be not detected
2023 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2024 goto martian_source;
2026 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2029 /* Accept zero addresses only to limited broadcast;
2030 * I even do not know to fix it or not. Waiting for complains :-)
2032 if (ipv4_is_zeronet(saddr))
2033 goto martian_source;
2035 if (ipv4_is_zeronet(daddr))
2036 goto martian_destination;
2038 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2039 if (ipv4_is_loopback(daddr))
2040 goto martian_destination;
2042 if (ipv4_is_loopback(saddr))
2043 goto martian_source;
2047 * Now we are ready to route packet.
2050 fl4.flowi4_iif = dev->ifindex;
2051 fl4.flowi4_mark = skb->mark;
2052 fl4.flowi4_tos = tos;
2053 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2056 err = fib_lookup(net, &fl4, &res);
2060 RT_CACHE_STAT_INC(in_slow_tot);
2062 if (res.type == RTN_BROADCAST)
2065 if (res.type == RTN_LOCAL) {
2066 err = fib_validate_source(skb, saddr, daddr, tos,
2067 net->loopback_dev->ifindex,
2068 dev, in_dev, &itag);
2070 goto martian_source_keep_err;
2072 flags |= RTCF_DIRECTSRC;
2076 if (!IN_DEV_FORWARD(in_dev))
2078 if (res.type != RTN_UNICAST)
2079 goto martian_destination;
2081 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2085 if (skb->protocol != htons(ETH_P_IP))
2088 if (!ipv4_is_zeronet(saddr)) {
2089 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2092 goto martian_source_keep_err;
2094 flags |= RTCF_DIRECTSRC;
2096 flags |= RTCF_BROADCAST;
2097 res.type = RTN_BROADCAST;
2098 RT_CACHE_STAT_INC(in_brd);
2101 rth = rt_dst_alloc(net->loopback_dev,
2102 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2106 rth->dst.input= ip_local_deliver;
2107 rth->dst.output= ip_rt_bug;
2108 #ifdef CONFIG_IP_ROUTE_CLASSID
2109 rth->dst.tclassid = itag;
2112 rth->rt_key_dst = daddr;
2113 rth->rt_key_src = saddr;
2114 rth->rt_genid = rt_genid(net);
2115 rth->rt_flags = flags|RTCF_LOCAL;
2116 rth->rt_type = res.type;
2117 rth->rt_key_tos = tos;
2118 rth->rt_dst = daddr;
2119 rth->rt_src = saddr;
2120 rth->rt_route_iif = dev->ifindex;
2121 rth->rt_iif = dev->ifindex;
2123 rth->rt_mark = skb->mark;
2125 rth->rt_gateway = daddr;
2127 if (res.type == RTN_UNREACHABLE) {
2128 rth->dst.input= ip_error;
2129 rth->dst.error= -err;
2130 rth->rt_flags &= ~RTCF_LOCAL;
2132 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2133 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2140 RT_CACHE_STAT_INC(in_no_route);
2141 res.type = RTN_UNREACHABLE;
2147 * Do not cache martian addresses: they should be logged (RFC1812)
2149 martian_destination:
2150 RT_CACHE_STAT_INC(in_martian_dst);
2151 #ifdef CONFIG_IP_ROUTE_VERBOSE
2152 if (IN_DEV_LOG_MARTIANS(in_dev))
2153 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2154 &daddr, &saddr, dev->name);
2167 martian_source_keep_err:
2168 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2172 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2173 u8 tos, struct net_device *dev, bool noref)
2177 int iif = dev->ifindex;
2185 if (!rt_caching(net))
2188 tos &= IPTOS_RT_MASK;
2189 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2191 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2192 rth = rcu_dereference(rth->dst.rt_next)) {
2193 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2194 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2195 (rth->rt_route_iif ^ iif) |
2196 (rth->rt_key_tos ^ tos)) == 0 &&
2197 rth->rt_mark == skb->mark &&
2198 net_eq(dev_net(rth->dst.dev), net) &&
2199 !rt_is_expired(rth)) {
2201 dst_use_noref(&rth->dst, jiffies);
2202 skb_dst_set_noref(skb, &rth->dst);
2204 dst_use(&rth->dst, jiffies);
2205 skb_dst_set(skb, &rth->dst);
2207 RT_CACHE_STAT_INC(in_hit);
2211 RT_CACHE_STAT_INC(in_hlist_search);
2215 /* Multicast recognition logic is moved from route cache to here.
2216 The problem was that too many Ethernet cards have broken/missing
2217 hardware multicast filters :-( As result the host on multicasting
2218 network acquires a lot of useless route cache entries, sort of
2219 SDR messages from all the world. Now we try to get rid of them.
2220 Really, provided software IP multicast filter is organized
2221 reasonably (at least, hashed), it does not result in a slowdown
2222 comparing with route cache reject entries.
2223 Note, that multicast routers are not affected, because
2224 route cache entry is created eventually.
2226 if (ipv4_is_multicast(daddr)) {
2227 struct in_device *in_dev = __in_dev_get_rcu(dev);
2230 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2231 ip_hdr(skb)->protocol);
2233 #ifdef CONFIG_IP_MROUTE
2235 (!ipv4_is_local_multicast(daddr) &&
2236 IN_DEV_MFORWARD(in_dev))
2239 int res = ip_route_input_mc(skb, daddr, saddr,
2248 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2252 EXPORT_SYMBOL(ip_route_input_common);
2254 /* called with rcu_read_lock() */
2255 static struct rtable *__mkroute_output(const struct fib_result *res,
2256 const struct flowi4 *fl4,
2257 __be32 orig_daddr, __be32 orig_saddr,
2258 int orig_oif, __u8 orig_rtos,
2259 struct net_device *dev_out,
2262 struct fib_info *fi = res->fi;
2263 struct in_device *in_dev;
2264 u16 type = res->type;
2267 in_dev = __in_dev_get_rcu(dev_out);
2269 return ERR_PTR(-EINVAL);
2271 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2272 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2273 return ERR_PTR(-EINVAL);
2275 if (ipv4_is_lbcast(fl4->daddr))
2276 type = RTN_BROADCAST;
2277 else if (ipv4_is_multicast(fl4->daddr))
2278 type = RTN_MULTICAST;
2279 else if (ipv4_is_zeronet(fl4->daddr))
2280 return ERR_PTR(-EINVAL);
2282 if (dev_out->flags & IFF_LOOPBACK)
2283 flags |= RTCF_LOCAL;
2285 if (type == RTN_BROADCAST) {
2286 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2288 } else if (type == RTN_MULTICAST) {
2289 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2290 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2292 flags &= ~RTCF_LOCAL;
2293 /* If multicast route do not exist use
2294 * default one, but do not gateway in this case.
2297 if (fi && res->prefixlen < 4)
2301 rth = rt_dst_alloc(dev_out,
2302 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2303 IN_DEV_CONF_GET(in_dev, NOXFRM));
2305 return ERR_PTR(-ENOBUFS);
2307 rth->dst.output = ip_output;
2309 rth->rt_key_dst = orig_daddr;
2310 rth->rt_key_src = orig_saddr;
2311 rth->rt_genid = rt_genid(dev_net(dev_out));
2312 rth->rt_flags = flags;
2313 rth->rt_type = type;
2314 rth->rt_key_tos = orig_rtos;
2315 rth->rt_dst = fl4->daddr;
2316 rth->rt_src = fl4->saddr;
2317 rth->rt_route_iif = 0;
2318 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2319 rth->rt_oif = orig_oif;
2320 rth->rt_mark = fl4->flowi4_mark;
2322 rth->rt_gateway = fl4->daddr;
2325 RT_CACHE_STAT_INC(out_slow_tot);
2327 if (flags & RTCF_LOCAL)
2328 rth->dst.input = ip_local_deliver;
2329 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2330 if (flags & RTCF_LOCAL &&
2331 !(dev_out->flags & IFF_LOOPBACK)) {
2332 rth->dst.output = ip_mc_output;
2333 RT_CACHE_STAT_INC(out_slow_mc);
2335 #ifdef CONFIG_IP_MROUTE
2336 if (type == RTN_MULTICAST) {
2337 if (IN_DEV_MFORWARD(in_dev) &&
2338 !ipv4_is_local_multicast(fl4->daddr)) {
2339 rth->dst.input = ip_mr_input;
2340 rth->dst.output = ip_mc_output;
2346 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2348 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2349 rth->dst.flags |= DST_NOCACHE;
2355 * Major route resolver routine.
2356 * called with rcu_read_lock();
2359 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2361 struct net_device *dev_out = NULL;
2362 __u8 tos = RT_FL_TOS(fl4);
2363 unsigned int flags = 0;
2364 struct fib_result res;
2372 #ifdef CONFIG_IP_MULTIPLE_TABLES
2376 orig_daddr = fl4->daddr;
2377 orig_saddr = fl4->saddr;
2378 orig_oif = fl4->flowi4_oif;
2380 fl4->flowi4_iif = net->loopback_dev->ifindex;
2381 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2382 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2383 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2387 rth = ERR_PTR(-EINVAL);
2388 if (ipv4_is_multicast(fl4->saddr) ||
2389 ipv4_is_lbcast(fl4->saddr) ||
2390 ipv4_is_zeronet(fl4->saddr))
2393 /* I removed check for oif == dev_out->oif here.
2394 It was wrong for two reasons:
2395 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2396 is assigned to multiple interfaces.
2397 2. Moreover, we are allowed to send packets with saddr
2398 of another iface. --ANK
2401 if (fl4->flowi4_oif == 0 &&
2402 (ipv4_is_multicast(fl4->daddr) ||
2403 ipv4_is_lbcast(fl4->daddr))) {
2404 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2405 dev_out = __ip_dev_find(net, fl4->saddr, false);
2406 if (dev_out == NULL)
2409 /* Special hack: user can direct multicasts
2410 and limited broadcast via necessary interface
2411 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2412 This hack is not just for fun, it allows
2413 vic,vat and friends to work.
2414 They bind socket to loopback, set ttl to zero
2415 and expect that it will work.
2416 From the viewpoint of routing cache they are broken,
2417 because we are not allowed to build multicast path
2418 with loopback source addr (look, routing cache
2419 cannot know, that ttl is zero, so that packet
2420 will not leave this host and route is valid).
2421 Luckily, this hack is good workaround.
2424 fl4->flowi4_oif = dev_out->ifindex;
2428 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2429 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2430 if (!__ip_dev_find(net, fl4->saddr, false))
2436 if (fl4->flowi4_oif) {
2437 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2438 rth = ERR_PTR(-ENODEV);
2439 if (dev_out == NULL)
2442 /* RACE: Check return value of inet_select_addr instead. */
2443 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2444 rth = ERR_PTR(-ENETUNREACH);
2447 if (ipv4_is_local_multicast(fl4->daddr) ||
2448 ipv4_is_lbcast(fl4->daddr)) {
2450 fl4->saddr = inet_select_addr(dev_out, 0,
2455 if (ipv4_is_multicast(fl4->daddr))
2456 fl4->saddr = inet_select_addr(dev_out, 0,
2458 else if (!fl4->daddr)
2459 fl4->saddr = inet_select_addr(dev_out, 0,
2465 fl4->daddr = fl4->saddr;
2467 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2468 dev_out = net->loopback_dev;
2469 fl4->flowi4_oif = net->loopback_dev->ifindex;
2470 res.type = RTN_LOCAL;
2471 flags |= RTCF_LOCAL;
2475 if (fib_lookup(net, fl4, &res)) {
2478 if (fl4->flowi4_oif) {
2479 /* Apparently, routing tables are wrong. Assume,
2480 that the destination is on link.
2483 Because we are allowed to send to iface
2484 even if it has NO routes and NO assigned
2485 addresses. When oif is specified, routing
2486 tables are looked up with only one purpose:
2487 to catch if destination is gatewayed, rather than
2488 direct. Moreover, if MSG_DONTROUTE is set,
2489 we send packet, ignoring both routing tables
2490 and ifaddr state. --ANK
2493 We could make it even if oif is unknown,
2494 likely IPv6, but we do not.
2497 if (fl4->saddr == 0)
2498 fl4->saddr = inet_select_addr(dev_out, 0,
2500 res.type = RTN_UNICAST;
2503 rth = ERR_PTR(-ENETUNREACH);
2507 if (res.type == RTN_LOCAL) {
2509 if (res.fi->fib_prefsrc)
2510 fl4->saddr = res.fi->fib_prefsrc;
2512 fl4->saddr = fl4->daddr;
2514 dev_out = net->loopback_dev;
2515 fl4->flowi4_oif = dev_out->ifindex;
2517 flags |= RTCF_LOCAL;
2521 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2522 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2523 fib_select_multipath(&res);
2526 if (!res.prefixlen &&
2527 res.table->tb_num_default > 1 &&
2528 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2529 fib_select_default(&res);
2532 fl4->saddr = FIB_RES_PREFSRC(net, res);
2534 dev_out = FIB_RES_DEV(res);
2535 fl4->flowi4_oif = dev_out->ifindex;
2539 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2540 tos, dev_out, flags);
2544 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2545 rt_genid(dev_net(dev_out)));
2546 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2554 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2559 if (!rt_caching(net))
2562 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2565 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2566 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2567 if (rth->rt_key_dst == flp4->daddr &&
2568 rth->rt_key_src == flp4->saddr &&
2569 rt_is_output_route(rth) &&
2570 rth->rt_oif == flp4->flowi4_oif &&
2571 rth->rt_mark == flp4->flowi4_mark &&
2572 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2573 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2574 net_eq(dev_net(rth->dst.dev), net) &&
2575 !rt_is_expired(rth)) {
2576 dst_use(&rth->dst, jiffies);
2577 RT_CACHE_STAT_INC(out_hit);
2578 rcu_read_unlock_bh();
2580 flp4->saddr = rth->rt_src;
2582 flp4->daddr = rth->rt_dst;
2585 RT_CACHE_STAT_INC(out_hlist_search);
2587 rcu_read_unlock_bh();
2590 return ip_route_output_slow(net, flp4);
2592 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2594 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2599 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2601 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2603 return mtu ? : dst->dev->mtu;
2606 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2610 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2616 static struct dst_ops ipv4_dst_blackhole_ops = {
2618 .protocol = cpu_to_be16(ETH_P_IP),
2619 .destroy = ipv4_dst_destroy,
2620 .check = ipv4_blackhole_dst_check,
2621 .mtu = ipv4_blackhole_mtu,
2622 .default_advmss = ipv4_default_advmss,
2623 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2624 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2625 .neigh_lookup = ipv4_neigh_lookup,
2628 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2630 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2631 struct rtable *ort = (struct rtable *) dst_orig;
2634 struct dst_entry *new = &rt->dst;
2637 new->input = dst_discard;
2638 new->output = dst_discard;
2640 new->dev = ort->dst.dev;
2644 rt->rt_key_dst = ort->rt_key_dst;
2645 rt->rt_key_src = ort->rt_key_src;
2646 rt->rt_key_tos = ort->rt_key_tos;
2647 rt->rt_route_iif = ort->rt_route_iif;
2648 rt->rt_iif = ort->rt_iif;
2649 rt->rt_oif = ort->rt_oif;
2650 rt->rt_mark = ort->rt_mark;
2651 rt->rt_pmtu = ort->rt_pmtu;
2653 rt->rt_genid = rt_genid(net);
2654 rt->rt_flags = ort->rt_flags;
2655 rt->rt_type = ort->rt_type;
2656 rt->rt_dst = ort->rt_dst;
2657 rt->rt_src = ort->rt_src;
2658 rt->rt_gateway = ort->rt_gateway;
2661 atomic_inc(&rt->fi->fib_clntref);
2666 dst_release(dst_orig);
2668 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2671 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2674 struct rtable *rt = __ip_route_output_key(net, flp4);
2679 if (flp4->flowi4_proto)
2680 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2681 flowi4_to_flowi(flp4),
2686 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2688 static int rt_fill_info(struct net *net,
2689 struct sk_buff *skb, u32 pid, u32 seq, int event,
2690 int nowait, unsigned int flags)
2692 struct rtable *rt = skb_rtable(skb);
2694 struct nlmsghdr *nlh;
2695 unsigned long expires = 0;
2698 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2702 r = nlmsg_data(nlh);
2703 r->rtm_family = AF_INET;
2704 r->rtm_dst_len = 32;
2706 r->rtm_tos = rt->rt_key_tos;
2707 r->rtm_table = RT_TABLE_MAIN;
2708 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2709 goto nla_put_failure;
2710 r->rtm_type = rt->rt_type;
2711 r->rtm_scope = RT_SCOPE_UNIVERSE;
2712 r->rtm_protocol = RTPROT_UNSPEC;
2713 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2714 if (rt->rt_flags & RTCF_NOTIFY)
2715 r->rtm_flags |= RTM_F_NOTIFY;
2717 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2718 goto nla_put_failure;
2719 if (rt->rt_key_src) {
2720 r->rtm_src_len = 32;
2721 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2722 goto nla_put_failure;
2725 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2726 goto nla_put_failure;
2727 #ifdef CONFIG_IP_ROUTE_CLASSID
2728 if (rt->dst.tclassid &&
2729 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2730 goto nla_put_failure;
2732 if (!rt_is_input_route(rt) &&
2733 rt->rt_src != rt->rt_key_src) {
2734 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2735 goto nla_put_failure;
2737 if (rt->rt_dst != rt->rt_gateway &&
2738 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2739 goto nla_put_failure;
2741 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2742 goto nla_put_failure;
2745 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2746 goto nla_put_failure;
2748 error = rt->dst.error;
2749 expires = rt->dst.expires;
2751 if (time_before(jiffies, expires))
2757 if (rt_is_input_route(rt)) {
2758 #ifdef CONFIG_IP_MROUTE
2759 __be32 dst = rt->rt_dst;
2761 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2762 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2763 int err = ipmr_get_route(net, skb,
2764 rt->rt_src, rt->rt_dst,
2770 goto nla_put_failure;
2772 if (err == -EMSGSIZE)
2773 goto nla_put_failure;
2779 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2780 goto nla_put_failure;
2783 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2784 goto nla_put_failure;
2786 return nlmsg_end(skb, nlh);
2789 nlmsg_cancel(skb, nlh);
2793 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2795 struct net *net = sock_net(in_skb->sk);
2797 struct nlattr *tb[RTA_MAX+1];
2798 struct rtable *rt = NULL;
2804 struct sk_buff *skb;
2806 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2810 rtm = nlmsg_data(nlh);
2812 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2818 /* Reserve room for dummy headers, this skb can pass
2819 through good chunk of routing engine.
2821 skb_reset_mac_header(skb);
2822 skb_reset_network_header(skb);
2824 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2825 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2826 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2828 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2829 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2830 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2831 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2834 struct net_device *dev;
2836 dev = __dev_get_by_index(net, iif);
2842 skb->protocol = htons(ETH_P_IP);
2846 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2849 rt = skb_rtable(skb);
2850 if (err == 0 && rt->dst.error)
2851 err = -rt->dst.error;
2853 struct flowi4 fl4 = {
2856 .flowi4_tos = rtm->rtm_tos,
2857 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2858 .flowi4_mark = mark,
2860 rt = ip_route_output_key(net, &fl4);
2870 skb_dst_set(skb, &rt->dst);
2871 if (rtm->rtm_flags & RTM_F_NOTIFY)
2872 rt->rt_flags |= RTCF_NOTIFY;
2874 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2875 RTM_NEWROUTE, 0, 0);
2879 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2888 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2895 net = sock_net(skb->sk);
2900 s_idx = idx = cb->args[1];
2901 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2902 if (!rt_hash_table[h].chain)
2905 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2906 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2907 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2909 if (rt_is_expired(rt))
2911 skb_dst_set_noref(skb, &rt->dst);
2912 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2913 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2914 1, NLM_F_MULTI) <= 0) {
2916 rcu_read_unlock_bh();
2921 rcu_read_unlock_bh();
2930 void ip_rt_multicast_event(struct in_device *in_dev)
2932 rt_cache_flush(dev_net(in_dev->dev), 0);
2935 #ifdef CONFIG_SYSCTL
2936 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2937 void __user *buffer,
2938 size_t *lenp, loff_t *ppos)
2945 memcpy(&ctl, __ctl, sizeof(ctl));
2946 ctl.data = &flush_delay;
2947 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2949 net = (struct net *)__ctl->extra1;
2950 rt_cache_flush(net, flush_delay);
2957 static ctl_table ipv4_route_table[] = {
2959 .procname = "gc_thresh",
2960 .data = &ipv4_dst_ops.gc_thresh,
2961 .maxlen = sizeof(int),
2963 .proc_handler = proc_dointvec,
2966 .procname = "max_size",
2967 .data = &ip_rt_max_size,
2968 .maxlen = sizeof(int),
2970 .proc_handler = proc_dointvec,
2973 /* Deprecated. Use gc_min_interval_ms */
2975 .procname = "gc_min_interval",
2976 .data = &ip_rt_gc_min_interval,
2977 .maxlen = sizeof(int),
2979 .proc_handler = proc_dointvec_jiffies,
2982 .procname = "gc_min_interval_ms",
2983 .data = &ip_rt_gc_min_interval,
2984 .maxlen = sizeof(int),
2986 .proc_handler = proc_dointvec_ms_jiffies,
2989 .procname = "gc_timeout",
2990 .data = &ip_rt_gc_timeout,
2991 .maxlen = sizeof(int),
2993 .proc_handler = proc_dointvec_jiffies,
2996 .procname = "gc_interval",
2997 .data = &ip_rt_gc_interval,
2998 .maxlen = sizeof(int),
3000 .proc_handler = proc_dointvec_jiffies,
3003 .procname = "redirect_load",
3004 .data = &ip_rt_redirect_load,
3005 .maxlen = sizeof(int),
3007 .proc_handler = proc_dointvec,
3010 .procname = "redirect_number",
3011 .data = &ip_rt_redirect_number,
3012 .maxlen = sizeof(int),
3014 .proc_handler = proc_dointvec,
3017 .procname = "redirect_silence",
3018 .data = &ip_rt_redirect_silence,
3019 .maxlen = sizeof(int),
3021 .proc_handler = proc_dointvec,
3024 .procname = "error_cost",
3025 .data = &ip_rt_error_cost,
3026 .maxlen = sizeof(int),
3028 .proc_handler = proc_dointvec,
3031 .procname = "error_burst",
3032 .data = &ip_rt_error_burst,
3033 .maxlen = sizeof(int),
3035 .proc_handler = proc_dointvec,
3038 .procname = "gc_elasticity",
3039 .data = &ip_rt_gc_elasticity,
3040 .maxlen = sizeof(int),
3042 .proc_handler = proc_dointvec,
3045 .procname = "mtu_expires",
3046 .data = &ip_rt_mtu_expires,
3047 .maxlen = sizeof(int),
3049 .proc_handler = proc_dointvec_jiffies,
3052 .procname = "min_pmtu",
3053 .data = &ip_rt_min_pmtu,
3054 .maxlen = sizeof(int),
3056 .proc_handler = proc_dointvec,
3059 .procname = "min_adv_mss",
3060 .data = &ip_rt_min_advmss,
3061 .maxlen = sizeof(int),
3063 .proc_handler = proc_dointvec,
3068 static struct ctl_table ipv4_route_flush_table[] = {
3070 .procname = "flush",
3071 .maxlen = sizeof(int),
3073 .proc_handler = ipv4_sysctl_rtcache_flush,
3078 static __net_init int sysctl_route_net_init(struct net *net)
3080 struct ctl_table *tbl;
3082 tbl = ipv4_route_flush_table;
3083 if (!net_eq(net, &init_net)) {
3084 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3088 tbl[0].extra1 = net;
3090 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3091 if (net->ipv4.route_hdr == NULL)
3096 if (tbl != ipv4_route_flush_table)
3102 static __net_exit void sysctl_route_net_exit(struct net *net)
3104 struct ctl_table *tbl;
3106 tbl = net->ipv4.route_hdr->ctl_table_arg;
3107 unregister_net_sysctl_table(net->ipv4.route_hdr);
3108 BUG_ON(tbl == ipv4_route_flush_table);
3112 static __net_initdata struct pernet_operations sysctl_route_ops = {
3113 .init = sysctl_route_net_init,
3114 .exit = sysctl_route_net_exit,
3118 static __net_init int rt_genid_init(struct net *net)
3120 get_random_bytes(&net->ipv4.rt_genid,
3121 sizeof(net->ipv4.rt_genid));
3122 get_random_bytes(&net->ipv4.dev_addr_genid,
3123 sizeof(net->ipv4.dev_addr_genid));
3127 static __net_initdata struct pernet_operations rt_genid_ops = {
3128 .init = rt_genid_init,
3131 static int __net_init ipv4_inetpeer_init(struct net *net)
3133 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3137 inet_peer_base_init(bp);
3138 net->ipv4.peers = bp;
3142 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3144 struct inet_peer_base *bp = net->ipv4.peers;
3146 net->ipv4.peers = NULL;
3147 inetpeer_invalidate_tree(bp);
3151 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3152 .init = ipv4_inetpeer_init,
3153 .exit = ipv4_inetpeer_exit,
3156 #ifdef CONFIG_IP_ROUTE_CLASSID
3157 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3158 #endif /* CONFIG_IP_ROUTE_CLASSID */
3160 static __initdata unsigned long rhash_entries;
3161 static int __init set_rhash_entries(char *str)
3168 ret = kstrtoul(str, 0, &rhash_entries);
3174 __setup("rhash_entries=", set_rhash_entries);
3176 int __init ip_rt_init(void)
3180 #ifdef CONFIG_IP_ROUTE_CLASSID
3181 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3183 panic("IP: failed to allocate ip_rt_acct\n");
3186 ipv4_dst_ops.kmem_cachep =
3187 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3188 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3190 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3192 if (dst_entries_init(&ipv4_dst_ops) < 0)
3193 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3195 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3196 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3198 rt_hash_table = (struct rt_hash_bucket *)
3199 alloc_large_system_hash("IP route cache",
3200 sizeof(struct rt_hash_bucket),
3202 (totalram_pages >= 128 * 1024) ?
3208 rhash_entries ? 0 : 512 * 1024);
3209 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3210 rt_hash_lock_init();
3212 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3213 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3218 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3219 expires_ljiffies = jiffies;
3220 schedule_delayed_work(&expires_work,
3221 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3223 if (ip_rt_proc_init())
3224 pr_err("Unable to create route proc files\n");
3227 xfrm4_init(ip_rt_max_size);
3229 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3231 #ifdef CONFIG_SYSCTL
3232 register_pernet_subsys(&sysctl_route_ops);
3234 register_pernet_subsys(&rt_genid_ops);
3235 register_pernet_subsys(&ipv4_inetpeer_ops);
3239 #ifdef CONFIG_SYSCTL
3241 * We really need to sanitize the damn ipv4 init order, then all
3242 * this nonsense will go away.
3244 void __init ip_static_sysctl_init(void)
3246 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);