2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #include <linux/module.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <linux/bitops.h>
69 #include <linux/types.h>
70 #include <linux/kernel.h>
72 #include <linux/bootmem.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/workqueue.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/jhash.h>
91 #include <linux/rcupdate.h>
92 #include <linux/times.h>
93 #include <linux/slab.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/netevent.h>
107 #include <net/rtnetlink.h>
109 #include <linux/sysctl.h>
112 #define RT_FL_TOS(oldflp4) \
113 ((u32)(oldflp4->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
115 #define IP_MAX_MTU 0xFFF0
117 #define RT_GC_TIMEOUT (300*HZ)
119 static int ip_rt_max_size;
120 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
121 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
122 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_gc_elasticity __read_mostly = 8;
129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly = 256;
132 static int rt_chain_length_max __read_mostly = 20;
135 * Interface to generic destination cache.
138 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
139 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
140 static unsigned int ipv4_default_mtu(const struct dst_entry *dst);
141 static void ipv4_dst_destroy(struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void ipv4_link_failure(struct sk_buff *skb);
144 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
145 static int rt_garbage_collect(struct dst_ops *ops);
147 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
152 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 struct rtable *rt = (struct rtable *) dst;
155 struct inet_peer *peer;
159 rt_bind_peer(rt, rt->rt_dst, 1);
163 u32 *old_p = __DST_METRICS_PTR(old);
164 unsigned long prev, new;
167 if (inet_metrics_new(peer))
168 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
170 new = (unsigned long) p;
171 prev = cmpxchg(&dst->_metrics, old, new);
174 p = __DST_METRICS_PTR(prev);
175 if (prev & DST_METRICS_READ_ONLY)
179 fib_info_put(rt->fi);
187 static struct dst_ops ipv4_dst_ops = {
189 .protocol = cpu_to_be16(ETH_P_IP),
190 .gc = rt_garbage_collect,
191 .check = ipv4_dst_check,
192 .default_advmss = ipv4_default_advmss,
193 .default_mtu = ipv4_default_mtu,
194 .cow_metrics = ipv4_cow_metrics,
195 .destroy = ipv4_dst_destroy,
196 .ifdown = ipv4_dst_ifdown,
197 .negative_advice = ipv4_negative_advice,
198 .link_failure = ipv4_link_failure,
199 .update_pmtu = ip_rt_update_pmtu,
200 .local_out = __ip_local_out,
203 #define ECN_OR_COST(class) TC_PRIO_##class
205 const __u8 ip_tos2prio[16] = {
207 ECN_OR_COST(BESTEFFORT),
209 ECN_OR_COST(BESTEFFORT),
215 ECN_OR_COST(INTERACTIVE),
217 ECN_OR_COST(INTERACTIVE),
218 TC_PRIO_INTERACTIVE_BULK,
219 ECN_OR_COST(INTERACTIVE_BULK),
220 TC_PRIO_INTERACTIVE_BULK,
221 ECN_OR_COST(INTERACTIVE_BULK)
229 /* The locking scheme is rather straight forward:
231 * 1) Read-Copy Update protects the buckets of the central route hash.
232 * 2) Only writers remove entries, and they hold the lock
233 * as they look at rtable reference counts.
234 * 3) Only readers acquire references to rtable entries,
235 * they do so with atomic increments and with the
239 struct rt_hash_bucket {
240 struct rtable __rcu *chain;
243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
244 defined(CONFIG_PROVE_LOCKING)
246 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
247 * The size of this table is a power of two and depends on the number of CPUS.
248 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
250 #ifdef CONFIG_LOCKDEP
251 # define RT_HASH_LOCK_SZ 256
254 # define RT_HASH_LOCK_SZ 4096
256 # define RT_HASH_LOCK_SZ 2048
258 # define RT_HASH_LOCK_SZ 1024
260 # define RT_HASH_LOCK_SZ 512
262 # define RT_HASH_LOCK_SZ 256
266 static spinlock_t *rt_hash_locks;
267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
269 static __init void rt_hash_lock_init(void)
273 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
276 panic("IP: failed to allocate rt_hash_locks\n");
278 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
279 spin_lock_init(&rt_hash_locks[i]);
282 # define rt_hash_lock_addr(slot) NULL
284 static inline void rt_hash_lock_init(void)
289 static struct rt_hash_bucket *rt_hash_table __read_mostly;
290 static unsigned rt_hash_mask __read_mostly;
291 static unsigned int rt_hash_log __read_mostly;
293 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
296 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
299 return jhash_3words((__force u32)daddr, (__force u32)saddr,
304 static inline int rt_genid(struct net *net)
306 return atomic_read(&net->ipv4.rt_genid);
309 #ifdef CONFIG_PROC_FS
310 struct rt_cache_iter_state {
311 struct seq_net_private p;
316 static struct rtable *rt_cache_get_first(struct seq_file *seq)
318 struct rt_cache_iter_state *st = seq->private;
319 struct rtable *r = NULL;
321 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
322 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
325 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
327 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
328 r->rt_genid == st->genid)
330 r = rcu_dereference_bh(r->dst.rt_next);
332 rcu_read_unlock_bh();
337 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
340 struct rt_cache_iter_state *st = seq->private;
342 r = rcu_dereference_bh(r->dst.rt_next);
344 rcu_read_unlock_bh();
346 if (--st->bucket < 0)
348 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
350 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
355 static struct rtable *rt_cache_get_next(struct seq_file *seq,
358 struct rt_cache_iter_state *st = seq->private;
359 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
360 if (dev_net(r->dst.dev) != seq_file_net(seq))
362 if (r->rt_genid == st->genid)
368 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
370 struct rtable *r = rt_cache_get_first(seq);
373 while (pos && (r = rt_cache_get_next(seq, r)))
375 return pos ? NULL : r;
378 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
380 struct rt_cache_iter_state *st = seq->private;
382 return rt_cache_get_idx(seq, *pos - 1);
383 st->genid = rt_genid(seq_file_net(seq));
384 return SEQ_START_TOKEN;
387 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
391 if (v == SEQ_START_TOKEN)
392 r = rt_cache_get_first(seq);
394 r = rt_cache_get_next(seq, v);
399 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
401 if (v && v != SEQ_START_TOKEN)
402 rcu_read_unlock_bh();
405 static int rt_cache_seq_show(struct seq_file *seq, void *v)
407 if (v == SEQ_START_TOKEN)
408 seq_printf(seq, "%-127s\n",
409 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
410 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
413 struct rtable *r = v;
416 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
417 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
418 r->dst.dev ? r->dst.dev->name : "*",
419 (__force u32)r->rt_dst,
420 (__force u32)r->rt_gateway,
421 r->rt_flags, atomic_read(&r->dst.__refcnt),
422 r->dst.__use, 0, (__force u32)r->rt_src,
423 dst_metric_advmss(&r->dst) + 40,
424 dst_metric(&r->dst, RTAX_WINDOW),
425 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
426 dst_metric(&r->dst, RTAX_RTTVAR)),
428 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
429 r->dst.hh ? (r->dst.hh->hh_output ==
431 r->rt_spec_dst, &len);
433 seq_printf(seq, "%*s\n", 127 - len, "");
438 static const struct seq_operations rt_cache_seq_ops = {
439 .start = rt_cache_seq_start,
440 .next = rt_cache_seq_next,
441 .stop = rt_cache_seq_stop,
442 .show = rt_cache_seq_show,
445 static int rt_cache_seq_open(struct inode *inode, struct file *file)
447 return seq_open_net(inode, file, &rt_cache_seq_ops,
448 sizeof(struct rt_cache_iter_state));
451 static const struct file_operations rt_cache_seq_fops = {
452 .owner = THIS_MODULE,
453 .open = rt_cache_seq_open,
456 .release = seq_release_net,
460 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
465 return SEQ_START_TOKEN;
467 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
468 if (!cpu_possible(cpu))
471 return &per_cpu(rt_cache_stat, cpu);
476 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
480 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
481 if (!cpu_possible(cpu))
484 return &per_cpu(rt_cache_stat, cpu);
490 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
495 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
497 struct rt_cache_stat *st = v;
499 if (v == SEQ_START_TOKEN) {
500 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");
504 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
505 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
506 dst_entries_get_slow(&ipv4_dst_ops),
529 static const struct seq_operations rt_cpu_seq_ops = {
530 .start = rt_cpu_seq_start,
531 .next = rt_cpu_seq_next,
532 .stop = rt_cpu_seq_stop,
533 .show = rt_cpu_seq_show,
537 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
539 return seq_open(file, &rt_cpu_seq_ops);
542 static const struct file_operations rt_cpu_seq_fops = {
543 .owner = THIS_MODULE,
544 .open = rt_cpu_seq_open,
547 .release = seq_release,
550 #ifdef CONFIG_IP_ROUTE_CLASSID
551 static int rt_acct_proc_show(struct seq_file *m, void *v)
553 struct ip_rt_acct *dst, *src;
556 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
560 for_each_possible_cpu(i) {
561 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
562 for (j = 0; j < 256; j++) {
563 dst[j].o_bytes += src[j].o_bytes;
564 dst[j].o_packets += src[j].o_packets;
565 dst[j].i_bytes += src[j].i_bytes;
566 dst[j].i_packets += src[j].i_packets;
570 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
575 static int rt_acct_proc_open(struct inode *inode, struct file *file)
577 return single_open(file, rt_acct_proc_show, NULL);
580 static const struct file_operations rt_acct_proc_fops = {
581 .owner = THIS_MODULE,
582 .open = rt_acct_proc_open,
585 .release = single_release,
589 static int __net_init ip_rt_do_proc_init(struct net *net)
591 struct proc_dir_entry *pde;
593 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
598 pde = proc_create("rt_cache", S_IRUGO,
599 net->proc_net_stat, &rt_cpu_seq_fops);
603 #ifdef CONFIG_IP_ROUTE_CLASSID
604 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
610 #ifdef CONFIG_IP_ROUTE_CLASSID
612 remove_proc_entry("rt_cache", net->proc_net_stat);
615 remove_proc_entry("rt_cache", net->proc_net);
620 static void __net_exit ip_rt_do_proc_exit(struct net *net)
622 remove_proc_entry("rt_cache", net->proc_net_stat);
623 remove_proc_entry("rt_cache", net->proc_net);
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625 remove_proc_entry("rt_acct", net->proc_net);
629 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
630 .init = ip_rt_do_proc_init,
631 .exit = ip_rt_do_proc_exit,
634 static int __init ip_rt_proc_init(void)
636 return register_pernet_subsys(&ip_rt_proc_ops);
640 static inline int ip_rt_proc_init(void)
644 #endif /* CONFIG_PROC_FS */
646 static inline void rt_free(struct rtable *rt)
648 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
651 static inline void rt_drop(struct rtable *rt)
654 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
657 static inline int rt_fast_clean(struct rtable *rth)
659 /* Kill broadcast/multicast entries very aggresively, if they
660 collide in hash table with more useful entries */
661 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
662 rt_is_input_route(rth) && rth->dst.rt_next;
665 static inline int rt_valuable(struct rtable *rth)
667 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
668 (rth->peer && rth->peer->pmtu_expires);
671 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
676 if (atomic_read(&rth->dst.__refcnt))
679 age = jiffies - rth->dst.lastuse;
680 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
681 (age <= tmo2 && rt_valuable(rth)))
687 /* Bits of score are:
689 * 30: not quite useless
690 * 29..0: usage counter
692 static inline u32 rt_score(struct rtable *rt)
694 u32 score = jiffies - rt->dst.lastuse;
696 score = ~score & ~(3<<30);
701 if (rt_is_output_route(rt) ||
702 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
708 static inline bool rt_caching(const struct net *net)
710 return net->ipv4.current_rt_cache_rebuild_count <=
711 net->ipv4.sysctl_rt_cache_rebuild_count;
714 static inline bool compare_hash_inputs(const struct rtable *rt1,
715 const struct rtable *rt2)
717 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
718 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
719 (rt1->rt_iif ^ rt2->rt_iif)) == 0);
722 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
724 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
725 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
726 (rt1->rt_mark ^ rt2->rt_mark) |
727 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
728 (rt1->rt_oif ^ rt2->rt_oif) |
729 (rt1->rt_iif ^ rt2->rt_iif)) == 0;
732 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
734 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
737 static inline int rt_is_expired(struct rtable *rth)
739 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
743 * Perform a full scan of hash table and free all entries.
744 * Can be called by a softirq or a process.
745 * In the later case, we want to be reschedule if necessary
747 static void rt_do_flush(struct net *net, int process_context)
750 struct rtable *rth, *next;
752 for (i = 0; i <= rt_hash_mask; i++) {
753 struct rtable __rcu **pprev;
756 if (process_context && need_resched())
758 rth = rcu_dereference_raw(rt_hash_table[i].chain);
762 spin_lock_bh(rt_hash_lock_addr(i));
765 pprev = &rt_hash_table[i].chain;
766 rth = rcu_dereference_protected(*pprev,
767 lockdep_is_held(rt_hash_lock_addr(i)));
770 next = rcu_dereference_protected(rth->dst.rt_next,
771 lockdep_is_held(rt_hash_lock_addr(i)));
774 net_eq(dev_net(rth->dst.dev), net)) {
775 rcu_assign_pointer(*pprev, next);
776 rcu_assign_pointer(rth->dst.rt_next, list);
779 pprev = &rth->dst.rt_next;
784 spin_unlock_bh(rt_hash_lock_addr(i));
786 for (; list; list = next) {
787 next = rcu_dereference_protected(list->dst.rt_next, 1);
794 * While freeing expired entries, we compute average chain length
795 * and standard deviation, using fixed-point arithmetic.
796 * This to have an estimation of rt_chain_length_max
797 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
798 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
802 #define ONE (1UL << FRACT_BITS)
805 * Given a hash chain and an item in this hash chain,
806 * find if a previous entry has the same hash_inputs
807 * (but differs on tos, mark or oif)
808 * Returns 0 if an alias is found.
809 * Returns ONE if rth has no alias before itself.
811 static int has_noalias(const struct rtable *head, const struct rtable *rth)
813 const struct rtable *aux = head;
816 if (compare_hash_inputs(aux, rth))
818 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
824 * Perturbation of rt_genid by a small quantity [1..256]
825 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
826 * many times (2^24) without giving recent rt_genid.
827 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
829 static void rt_cache_invalidate(struct net *net)
831 unsigned char shuffle;
833 get_random_bytes(&shuffle, sizeof(shuffle));
834 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
838 * delay < 0 : invalidate cache (fast : entries will be deleted later)
839 * delay >= 0 : invalidate & flush cache (can be long)
841 void rt_cache_flush(struct net *net, int delay)
843 rt_cache_invalidate(net);
845 rt_do_flush(net, !in_softirq());
848 /* Flush previous cache invalidated entries from the cache */
849 void rt_cache_flush_batch(struct net *net)
851 rt_do_flush(net, !in_softirq());
854 static void rt_emergency_hash_rebuild(struct net *net)
857 printk(KERN_WARNING "Route hash chain too long!\n");
858 rt_cache_invalidate(net);
862 Short description of GC goals.
864 We want to build algorithm, which will keep routing cache
865 at some equilibrium point, when number of aged off entries
866 is kept approximately equal to newly generated ones.
868 Current expiration strength is variable "expire".
869 We try to adjust it dynamically, so that if networking
870 is idle expires is large enough to keep enough of warm entries,
871 and when load increases it reduces to limit cache size.
874 static int rt_garbage_collect(struct dst_ops *ops)
876 static unsigned long expire = RT_GC_TIMEOUT;
877 static unsigned long last_gc;
879 static int equilibrium;
881 struct rtable __rcu **rthp;
882 unsigned long now = jiffies;
884 int entries = dst_entries_get_fast(&ipv4_dst_ops);
887 * Garbage collection is pretty expensive,
888 * do not make it too frequently.
891 RT_CACHE_STAT_INC(gc_total);
893 if (now - last_gc < ip_rt_gc_min_interval &&
894 entries < ip_rt_max_size) {
895 RT_CACHE_STAT_INC(gc_ignored);
899 entries = dst_entries_get_slow(&ipv4_dst_ops);
900 /* Calculate number of entries, which we want to expire now. */
901 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
903 if (equilibrium < ipv4_dst_ops.gc_thresh)
904 equilibrium = ipv4_dst_ops.gc_thresh;
905 goal = entries - equilibrium;
907 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
908 goal = entries - equilibrium;
911 /* We are in dangerous area. Try to reduce cache really
914 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
915 equilibrium = entries - goal;
918 if (now - last_gc >= ip_rt_gc_min_interval)
929 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
930 unsigned long tmo = expire;
932 k = (k + 1) & rt_hash_mask;
933 rthp = &rt_hash_table[k].chain;
934 spin_lock_bh(rt_hash_lock_addr(k));
935 while ((rth = rcu_dereference_protected(*rthp,
936 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
937 if (!rt_is_expired(rth) &&
938 !rt_may_expire(rth, tmo, expire)) {
940 rthp = &rth->dst.rt_next;
943 *rthp = rth->dst.rt_next;
947 spin_unlock_bh(rt_hash_lock_addr(k));
956 /* Goal is not achieved. We stop process if:
958 - if expire reduced to zero. Otherwise, expire is halfed.
959 - if table is not full.
960 - if we are called from interrupt.
961 - jiffies check is just fallback/debug loop breaker.
962 We will not spin here for long time in any case.
965 RT_CACHE_STAT_INC(gc_goal_miss);
972 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
974 } while (!in_softirq() && time_before_eq(jiffies, now));
976 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
978 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
981 printk(KERN_WARNING "dst cache overflow\n");
982 RT_CACHE_STAT_INC(gc_dst_overflow);
986 expire += ip_rt_gc_min_interval;
987 if (expire > ip_rt_gc_timeout ||
988 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
989 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
990 expire = ip_rt_gc_timeout;
995 * Returns number of entries in a hash chain that have different hash_inputs
997 static int slow_chain_length(const struct rtable *head)
1000 const struct rtable *rth = head;
1003 length += has_noalias(head, rth);
1004 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1006 return length >> FRACT_BITS;
1009 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1010 struct sk_buff *skb, int ifindex)
1012 struct rtable *rth, *cand;
1013 struct rtable __rcu **rthp, **candp;
1017 int attempts = !in_softirq();
1021 min_score = ~(u32)0;
1026 if (!rt_caching(dev_net(rt->dst.dev))) {
1028 * If we're not caching, just tell the caller we
1029 * were successful and don't touch the route. The
1030 * caller hold the sole reference to the cache entry, and
1031 * it will be released when the caller is done with it.
1032 * If we drop it here, the callers have no way to resolve routes
1033 * when we're not caching. Instead, just point *rp at rt, so
1034 * the caller gets a single use out of the route
1035 * Note that we do rt_free on this new route entry, so that
1036 * once its refcount hits zero, we are still able to reap it
1038 * Note: To avoid expensive rcu stuff for this uncached dst,
1039 * we set DST_NOCACHE so that dst_release() can free dst without
1040 * waiting a grace period.
1043 rt->dst.flags |= DST_NOCACHE;
1044 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1045 int err = arp_bind_neighbour(&rt->dst);
1047 if (net_ratelimit())
1049 "Neighbour table failure & not caching routes.\n");
1051 return ERR_PTR(err);
1058 rthp = &rt_hash_table[hash].chain;
1060 spin_lock_bh(rt_hash_lock_addr(hash));
1061 while ((rth = rcu_dereference_protected(*rthp,
1062 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1063 if (rt_is_expired(rth)) {
1064 *rthp = rth->dst.rt_next;
1068 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1070 *rthp = rth->dst.rt_next;
1072 * Since lookup is lockfree, the deletion
1073 * must be visible to another weakly ordered CPU before
1074 * the insertion at the start of the hash chain.
1076 rcu_assign_pointer(rth->dst.rt_next,
1077 rt_hash_table[hash].chain);
1079 * Since lookup is lockfree, the update writes
1080 * must be ordered for consistency on SMP.
1082 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1084 dst_use(&rth->dst, now);
1085 spin_unlock_bh(rt_hash_lock_addr(hash));
1089 skb_dst_set(skb, &rth->dst);
1093 if (!atomic_read(&rth->dst.__refcnt)) {
1094 u32 score = rt_score(rth);
1096 if (score <= min_score) {
1105 rthp = &rth->dst.rt_next;
1109 /* ip_rt_gc_elasticity used to be average length of chain
1110 * length, when exceeded gc becomes really aggressive.
1112 * The second limit is less certain. At the moment it allows
1113 * only 2 entries per bucket. We will see.
1115 if (chain_length > ip_rt_gc_elasticity) {
1116 *candp = cand->dst.rt_next;
1120 if (chain_length > rt_chain_length_max &&
1121 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1122 struct net *net = dev_net(rt->dst.dev);
1123 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1124 if (!rt_caching(net)) {
1125 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1126 rt->dst.dev->name, num);
1128 rt_emergency_hash_rebuild(net);
1129 spin_unlock_bh(rt_hash_lock_addr(hash));
1131 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1132 ifindex, rt_genid(net));
1137 /* Try to bind route to arp only if it is output
1138 route or unicast forwarding path.
1140 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1141 int err = arp_bind_neighbour(&rt->dst);
1143 spin_unlock_bh(rt_hash_lock_addr(hash));
1145 if (err != -ENOBUFS) {
1147 return ERR_PTR(err);
1150 /* Neighbour tables are full and nothing
1151 can be released. Try to shrink route cache,
1152 it is most likely it holds some neighbour records.
1154 if (attempts-- > 0) {
1155 int saved_elasticity = ip_rt_gc_elasticity;
1156 int saved_int = ip_rt_gc_min_interval;
1157 ip_rt_gc_elasticity = 1;
1158 ip_rt_gc_min_interval = 0;
1159 rt_garbage_collect(&ipv4_dst_ops);
1160 ip_rt_gc_min_interval = saved_int;
1161 ip_rt_gc_elasticity = saved_elasticity;
1165 if (net_ratelimit())
1166 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1168 return ERR_PTR(-ENOBUFS);
1172 rt->dst.rt_next = rt_hash_table[hash].chain;
1175 * Since lookup is lockfree, we must make sure
1176 * previous writes to rt are committed to memory
1177 * before making rt visible to other CPUS.
1179 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1181 spin_unlock_bh(rt_hash_lock_addr(hash));
1185 skb_dst_set(skb, &rt->dst);
1189 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1191 static u32 rt_peer_genid(void)
1193 return atomic_read(&__rt_peer_genid);
1196 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1198 struct inet_peer *peer;
1200 peer = inet_getpeer_v4(daddr, create);
1202 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1205 rt->rt_peer_genid = rt_peer_genid();
1209 * Peer allocation may fail only in serious out-of-memory conditions. However
1210 * we still can generate some output.
1211 * Random ID selection looks a bit dangerous because we have no chances to
1212 * select ID being unique in a reasonable period of time.
1213 * But broken packet identifier may be better than no packet at all.
1215 static void ip_select_fb_ident(struct iphdr *iph)
1217 static DEFINE_SPINLOCK(ip_fb_id_lock);
1218 static u32 ip_fallback_id;
1221 spin_lock_bh(&ip_fb_id_lock);
1222 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1223 iph->id = htons(salt & 0xFFFF);
1224 ip_fallback_id = salt;
1225 spin_unlock_bh(&ip_fb_id_lock);
1228 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1230 struct rtable *rt = (struct rtable *) dst;
1233 if (rt->peer == NULL)
1234 rt_bind_peer(rt, rt->rt_dst, 1);
1236 /* If peer is attached to destination, it is never detached,
1237 so that we need not to grab a lock to dereference it.
1240 iph->id = htons(inet_getid(rt->peer, more));
1244 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1245 __builtin_return_address(0));
1247 ip_select_fb_ident(iph);
1249 EXPORT_SYMBOL(__ip_select_ident);
1251 static void rt_del(unsigned hash, struct rtable *rt)
1253 struct rtable __rcu **rthp;
1256 rthp = &rt_hash_table[hash].chain;
1257 spin_lock_bh(rt_hash_lock_addr(hash));
1259 while ((aux = rcu_dereference_protected(*rthp,
1260 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1261 if (aux == rt || rt_is_expired(aux)) {
1262 *rthp = aux->dst.rt_next;
1266 rthp = &aux->dst.rt_next;
1268 spin_unlock_bh(rt_hash_lock_addr(hash));
1271 /* called in rcu_read_lock() section */
1272 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1273 __be32 saddr, struct net_device *dev)
1275 struct in_device *in_dev = __in_dev_get_rcu(dev);
1276 struct inet_peer *peer;
1283 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1284 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1285 ipv4_is_zeronet(new_gw))
1286 goto reject_redirect;
1288 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1289 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1290 goto reject_redirect;
1291 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1292 goto reject_redirect;
1294 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1295 goto reject_redirect;
1298 peer = inet_getpeer_v4(daddr, 1);
1300 peer->redirect_learned.a4 = new_gw;
1304 atomic_inc(&__rt_peer_genid);
1309 #ifdef CONFIG_IP_ROUTE_VERBOSE
1310 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1311 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1312 " Advised path = %pI4 -> %pI4\n",
1313 &old_gw, dev->name, &new_gw,
1319 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1321 struct rtable *rt = (struct rtable *)dst;
1322 struct dst_entry *ret = dst;
1325 if (dst->obsolete > 0) {
1328 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1329 unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1331 rt_genid(dev_net(dst->dev)));
1334 } else if (rt->peer &&
1335 rt->peer->pmtu_expires &&
1336 time_after_eq(jiffies, rt->peer->pmtu_expires)) {
1337 unsigned long orig = rt->peer->pmtu_expires;
1339 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1340 dst_metric_set(dst, RTAX_MTU,
1341 rt->peer->pmtu_orig);
1349 * 1. The first ip_rt_redirect_number redirects are sent
1350 * with exponential backoff, then we stop sending them at all,
1351 * assuming that the host ignores our redirects.
1352 * 2. If we did not see packets requiring redirects
1353 * during ip_rt_redirect_silence, we assume that the host
1354 * forgot redirected route and start to send redirects again.
1356 * This algorithm is much cheaper and more intelligent than dumb load limiting
1359 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1360 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1363 void ip_rt_send_redirect(struct sk_buff *skb)
1365 struct rtable *rt = skb_rtable(skb);
1366 struct in_device *in_dev;
1367 struct inet_peer *peer;
1371 in_dev = __in_dev_get_rcu(rt->dst.dev);
1372 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1376 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1380 rt_bind_peer(rt, rt->rt_dst, 1);
1383 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1387 /* No redirected packets during ip_rt_redirect_silence;
1388 * reset the algorithm.
1390 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1391 peer->rate_tokens = 0;
1393 /* Too many ignored redirects; do not send anything
1394 * set dst.rate_last to the last seen redirected packet.
1396 if (peer->rate_tokens >= ip_rt_redirect_number) {
1397 peer->rate_last = jiffies;
1401 /* Check for load limit; set rate_last to the latest sent
1404 if (peer->rate_tokens == 0 ||
1407 (ip_rt_redirect_load << peer->rate_tokens)))) {
1408 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1409 peer->rate_last = jiffies;
1410 ++peer->rate_tokens;
1411 #ifdef CONFIG_IP_ROUTE_VERBOSE
1413 peer->rate_tokens == ip_rt_redirect_number &&
1415 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1416 &ip_hdr(skb)->saddr, rt->rt_iif,
1417 &rt->rt_dst, &rt->rt_gateway);
1422 static int ip_error(struct sk_buff *skb)
1424 struct rtable *rt = skb_rtable(skb);
1425 struct inet_peer *peer;
1430 switch (rt->dst.error) {
1435 code = ICMP_HOST_UNREACH;
1438 code = ICMP_NET_UNREACH;
1439 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1440 IPSTATS_MIB_INNOROUTES);
1443 code = ICMP_PKT_FILTERED;
1448 rt_bind_peer(rt, rt->rt_dst, 1);
1454 peer->rate_tokens += now - peer->rate_last;
1455 if (peer->rate_tokens > ip_rt_error_burst)
1456 peer->rate_tokens = ip_rt_error_burst;
1457 peer->rate_last = now;
1458 if (peer->rate_tokens >= ip_rt_error_cost)
1459 peer->rate_tokens -= ip_rt_error_cost;
1464 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1466 out: kfree_skb(skb);
1471 * The last two values are not from the RFC but
1472 * are needed for AMPRnet AX.25 paths.
1475 static const unsigned short mtu_plateau[] =
1476 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1478 static inline unsigned short guess_mtu(unsigned short old_mtu)
1482 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1483 if (old_mtu > mtu_plateau[i])
1484 return mtu_plateau[i];
1488 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1489 unsigned short new_mtu,
1490 struct net_device *dev)
1492 unsigned short old_mtu = ntohs(iph->tot_len);
1493 unsigned short est_mtu = 0;
1494 struct inet_peer *peer;
1496 peer = inet_getpeer_v4(iph->daddr, 1);
1498 unsigned short mtu = new_mtu;
1500 if (new_mtu < 68 || new_mtu >= old_mtu) {
1501 /* BSD 4.2 derived systems incorrectly adjust
1502 * tot_len by the IP header length, and report
1503 * a zero MTU in the ICMP message.
1506 old_mtu >= 68 + (iph->ihl << 2))
1507 old_mtu -= iph->ihl << 2;
1508 mtu = guess_mtu(old_mtu);
1511 if (mtu < ip_rt_min_pmtu)
1512 mtu = ip_rt_min_pmtu;
1513 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1514 unsigned long pmtu_expires;
1516 pmtu_expires = jiffies + ip_rt_mtu_expires;
1521 peer->pmtu_learned = mtu;
1522 peer->pmtu_expires = pmtu_expires;
1527 atomic_inc(&__rt_peer_genid);
1529 return est_mtu ? : new_mtu;
1532 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1534 unsigned long expires = peer->pmtu_expires;
1536 if (time_before(jiffies, expires)) {
1537 u32 orig_dst_mtu = dst_mtu(dst);
1538 if (peer->pmtu_learned < orig_dst_mtu) {
1539 if (!peer->pmtu_orig)
1540 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1541 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1543 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1544 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1547 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1549 struct rtable *rt = (struct rtable *) dst;
1550 struct inet_peer *peer;
1555 rt_bind_peer(rt, rt->rt_dst, 1);
1558 if (mtu < ip_rt_min_pmtu)
1559 mtu = ip_rt_min_pmtu;
1560 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1561 unsigned long pmtu_expires;
1563 pmtu_expires = jiffies + ip_rt_mtu_expires;
1567 peer->pmtu_learned = mtu;
1568 peer->pmtu_expires = pmtu_expires;
1570 atomic_inc(&__rt_peer_genid);
1571 rt->rt_peer_genid = rt_peer_genid();
1573 check_peer_pmtu(dst, peer);
1577 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1579 struct rtable *rt = (struct rtable *) dst;
1580 __be32 orig_gw = rt->rt_gateway;
1582 dst_confirm(&rt->dst);
1584 neigh_release(rt->dst.neighbour);
1585 rt->dst.neighbour = NULL;
1587 rt->rt_gateway = peer->redirect_learned.a4;
1588 if (arp_bind_neighbour(&rt->dst) ||
1589 !(rt->dst.neighbour->nud_state & NUD_VALID)) {
1590 if (rt->dst.neighbour)
1591 neigh_event_send(rt->dst.neighbour, NULL);
1592 rt->rt_gateway = orig_gw;
1595 rt->rt_flags |= RTCF_REDIRECTED;
1596 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE,
1602 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1604 struct rtable *rt = (struct rtable *) dst;
1606 if (rt_is_expired(rt))
1608 if (rt->rt_peer_genid != rt_peer_genid()) {
1609 struct inet_peer *peer;
1612 rt_bind_peer(rt, rt->rt_dst, 0);
1615 if (peer && peer->pmtu_expires)
1616 check_peer_pmtu(dst, peer);
1618 if (peer && peer->redirect_learned.a4 &&
1619 peer->redirect_learned.a4 != rt->rt_gateway) {
1620 if (check_peer_redir(dst, peer))
1624 rt->rt_peer_genid = rt_peer_genid();
1629 static void ipv4_dst_destroy(struct dst_entry *dst)
1631 struct rtable *rt = (struct rtable *) dst;
1632 struct inet_peer *peer = rt->peer;
1635 fib_info_put(rt->fi);
1645 static void ipv4_link_failure(struct sk_buff *skb)
1649 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1651 rt = skb_rtable(skb);
1654 rt->peer->pmtu_expires) {
1655 unsigned long orig = rt->peer->pmtu_expires;
1657 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1658 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1662 static int ip_rt_bug(struct sk_buff *skb)
1664 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1665 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1666 skb->dev ? skb->dev->name : "?");
1673 We do not cache source address of outgoing interface,
1674 because it is used only by IP RR, TS and SRR options,
1675 so that it out of fast path.
1677 BTW remember: "addr" is allowed to be not aligned
1681 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1685 if (rt_is_output_route(rt))
1686 src = ip_hdr(skb)->saddr;
1688 struct fib_result res;
1694 memset(&fl4, 0, sizeof(fl4));
1695 fl4.daddr = iph->daddr;
1696 fl4.saddr = iph->saddr;
1697 fl4.flowi4_tos = iph->tos;
1698 fl4.flowi4_oif = rt->dst.dev->ifindex;
1699 fl4.flowi4_iif = skb->dev->ifindex;
1700 fl4.flowi4_mark = skb->mark;
1703 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1704 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1706 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1710 memcpy(addr, &src, 4);
1713 #ifdef CONFIG_IP_ROUTE_CLASSID
1714 static void set_class_tag(struct rtable *rt, u32 tag)
1716 if (!(rt->dst.tclassid & 0xFFFF))
1717 rt->dst.tclassid |= tag & 0xFFFF;
1718 if (!(rt->dst.tclassid & 0xFFFF0000))
1719 rt->dst.tclassid |= tag & 0xFFFF0000;
1723 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1725 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1728 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1730 if (advmss > 65535 - 40)
1731 advmss = 65535 - 40;
1736 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1738 unsigned int mtu = dst->dev->mtu;
1740 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1741 const struct rtable *rt = (const struct rtable *) dst;
1743 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1747 if (mtu > IP_MAX_MTU)
1753 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1754 struct fib_info *fi)
1756 struct inet_peer *peer;
1759 /* If a peer entry exists for this destination, we must hook
1760 * it up in order to get at cached metrics.
1762 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1765 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1767 rt->rt_peer_genid = rt_peer_genid();
1768 if (inet_metrics_new(peer))
1769 memcpy(peer->metrics, fi->fib_metrics,
1770 sizeof(u32) * RTAX_MAX);
1771 dst_init_metrics(&rt->dst, peer->metrics, false);
1773 if (peer->pmtu_expires)
1774 check_peer_pmtu(&rt->dst, peer);
1775 if (peer->redirect_learned.a4 &&
1776 peer->redirect_learned.a4 != rt->rt_gateway) {
1777 rt->rt_gateway = peer->redirect_learned.a4;
1778 rt->rt_flags |= RTCF_REDIRECTED;
1781 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1783 atomic_inc(&fi->fib_clntref);
1785 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1789 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1790 const struct fib_result *res,
1791 struct fib_info *fi, u16 type, u32 itag)
1793 struct dst_entry *dst = &rt->dst;
1796 if (FIB_RES_GW(*res) &&
1797 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1798 rt->rt_gateway = FIB_RES_GW(*res);
1799 rt_init_metrics(rt, fl4, fi);
1800 #ifdef CONFIG_IP_ROUTE_CLASSID
1801 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1805 if (dst_mtu(dst) > IP_MAX_MTU)
1806 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1807 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1808 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1810 #ifdef CONFIG_IP_ROUTE_CLASSID
1811 #ifdef CONFIG_IP_MULTIPLE_TABLES
1812 set_class_tag(rt, fib_rules_tclass(res));
1814 set_class_tag(rt, itag);
1818 static struct rtable *rt_dst_alloc(struct net_device *dev,
1819 bool nopolicy, bool noxfrm)
1821 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1823 (nopolicy ? DST_NOPOLICY : 0) |
1824 (noxfrm ? DST_NOXFRM : 0));
1827 /* called in rcu_read_lock() section */
1828 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1829 u8 tos, struct net_device *dev, int our)
1834 struct in_device *in_dev = __in_dev_get_rcu(dev);
1838 /* Primary sanity checks. */
1843 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1844 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1847 if (ipv4_is_zeronet(saddr)) {
1848 if (!ipv4_is_local_multicast(daddr))
1850 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1852 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1857 rth = rt_dst_alloc(init_net.loopback_dev,
1858 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1862 #ifdef CONFIG_IP_ROUTE_CLASSID
1863 rth->dst.tclassid = itag;
1865 rth->dst.output = ip_rt_bug;
1867 rth->rt_key_dst = daddr;
1868 rth->rt_key_src = saddr;
1869 rth->rt_genid = rt_genid(dev_net(dev));
1870 rth->rt_flags = RTCF_MULTICAST;
1871 rth->rt_type = RTN_MULTICAST;
1872 rth->rt_key_tos = tos;
1873 rth->rt_dst = daddr;
1874 rth->rt_src = saddr;
1875 rth->rt_route_iif = dev->ifindex;
1876 rth->rt_iif = dev->ifindex;
1878 rth->rt_mark = skb->mark;
1879 rth->rt_gateway = daddr;
1880 rth->rt_spec_dst= spec_dst;
1881 rth->rt_peer_genid = 0;
1885 rth->dst.input= ip_local_deliver;
1886 rth->rt_flags |= RTCF_LOCAL;
1889 #ifdef CONFIG_IP_MROUTE
1890 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1891 rth->dst.input = ip_mr_input;
1893 RT_CACHE_STAT_INC(in_slow_mc);
1895 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1896 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1910 static void ip_handle_martian_source(struct net_device *dev,
1911 struct in_device *in_dev,
1912 struct sk_buff *skb,
1916 RT_CACHE_STAT_INC(in_martian_src);
1917 #ifdef CONFIG_IP_ROUTE_VERBOSE
1918 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1920 * RFC1812 recommendation, if source is martian,
1921 * the only hint is MAC header.
1923 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1924 &daddr, &saddr, dev->name);
1925 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1927 const unsigned char *p = skb_mac_header(skb);
1928 printk(KERN_WARNING "ll header: ");
1929 for (i = 0; i < dev->hard_header_len; i++, p++) {
1931 if (i < (dev->hard_header_len - 1))
1940 /* called in rcu_read_lock() section */
1941 static int __mkroute_input(struct sk_buff *skb,
1942 const struct fib_result *res,
1943 struct in_device *in_dev,
1944 __be32 daddr, __be32 saddr, u32 tos,
1945 struct rtable **result)
1949 struct in_device *out_dev;
1950 unsigned int flags = 0;
1954 /* get a working reference to the output device */
1955 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1956 if (out_dev == NULL) {
1957 if (net_ratelimit())
1958 printk(KERN_CRIT "Bug in ip_route_input" \
1959 "_slow(). Please, report\n");
1964 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1965 in_dev->dev, &spec_dst, &itag);
1967 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1974 flags |= RTCF_DIRECTSRC;
1976 if (out_dev == in_dev && err &&
1977 (IN_DEV_SHARED_MEDIA(out_dev) ||
1978 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1979 flags |= RTCF_DOREDIRECT;
1981 if (skb->protocol != htons(ETH_P_IP)) {
1982 /* Not IP (i.e. ARP). Do not create route, if it is
1983 * invalid for proxy arp. DNAT routes are always valid.
1985 * Proxy arp feature have been extended to allow, ARP
1986 * replies back to the same interface, to support
1987 * Private VLAN switch technologies. See arp.c.
1989 if (out_dev == in_dev &&
1990 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1996 rth = rt_dst_alloc(out_dev->dev,
1997 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1998 IN_DEV_CONF_GET(out_dev, NOXFRM));
2004 rth->rt_key_dst = daddr;
2005 rth->rt_key_src = saddr;
2006 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2007 rth->rt_flags = flags;
2008 rth->rt_type = res->type;
2009 rth->rt_key_tos = tos;
2010 rth->rt_dst = daddr;
2011 rth->rt_src = saddr;
2012 rth->rt_route_iif = in_dev->dev->ifindex;
2013 rth->rt_iif = in_dev->dev->ifindex;
2015 rth->rt_mark = skb->mark;
2016 rth->rt_gateway = daddr;
2017 rth->rt_spec_dst= spec_dst;
2018 rth->rt_peer_genid = 0;
2022 rth->dst.input = ip_forward;
2023 rth->dst.output = ip_output;
2025 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2033 static int ip_mkroute_input(struct sk_buff *skb,
2034 struct fib_result *res,
2035 const struct flowi4 *fl4,
2036 struct in_device *in_dev,
2037 __be32 daddr, __be32 saddr, u32 tos)
2039 struct rtable* rth = NULL;
2043 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2044 if (res->fi && res->fi->fib_nhs > 1)
2045 fib_select_multipath(res);
2048 /* create a routing cache entry */
2049 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2053 /* put it into the cache */
2054 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2055 rt_genid(dev_net(rth->dst.dev)));
2056 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2058 return PTR_ERR(rth);
2063 * NOTE. We drop all the packets that has local source
2064 * addresses, because every properly looped back packet
2065 * must have correct destination already attached by output routine.
2067 * Such approach solves two big problems:
2068 * 1. Not simplex devices are handled properly.
2069 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2070 * called with rcu_read_lock()
2073 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2074 u8 tos, struct net_device *dev)
2076 struct fib_result res;
2077 struct in_device *in_dev = __in_dev_get_rcu(dev);
2081 struct rtable * rth;
2085 struct net * net = dev_net(dev);
2087 /* IP on this device is disabled. */
2092 /* Check for the most weird martians, which can be not detected
2096 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2097 ipv4_is_loopback(saddr))
2098 goto martian_source;
2100 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2103 /* Accept zero addresses only to limited broadcast;
2104 * I even do not know to fix it or not. Waiting for complains :-)
2106 if (ipv4_is_zeronet(saddr))
2107 goto martian_source;
2109 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2110 goto martian_destination;
2113 * Now we are ready to route packet.
2116 fl4.flowi4_iif = dev->ifindex;
2117 fl4.flowi4_mark = skb->mark;
2118 fl4.flowi4_tos = tos;
2119 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2122 err = fib_lookup(net, &fl4, &res);
2124 if (!IN_DEV_FORWARD(in_dev))
2129 RT_CACHE_STAT_INC(in_slow_tot);
2131 if (res.type == RTN_BROADCAST)
2134 if (res.type == RTN_LOCAL) {
2135 err = fib_validate_source(skb, saddr, daddr, tos,
2136 net->loopback_dev->ifindex,
2137 dev, &spec_dst, &itag);
2139 goto martian_source_keep_err;
2141 flags |= RTCF_DIRECTSRC;
2146 if (!IN_DEV_FORWARD(in_dev))
2148 if (res.type != RTN_UNICAST)
2149 goto martian_destination;
2151 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2155 if (skb->protocol != htons(ETH_P_IP))
2158 if (ipv4_is_zeronet(saddr))
2159 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2161 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2164 goto martian_source_keep_err;
2166 flags |= RTCF_DIRECTSRC;
2168 flags |= RTCF_BROADCAST;
2169 res.type = RTN_BROADCAST;
2170 RT_CACHE_STAT_INC(in_brd);
2173 rth = rt_dst_alloc(net->loopback_dev,
2174 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2178 rth->dst.input= ip_local_deliver;
2179 rth->dst.output= ip_rt_bug;
2180 #ifdef CONFIG_IP_ROUTE_CLASSID
2181 rth->dst.tclassid = itag;
2184 rth->rt_key_dst = daddr;
2185 rth->rt_key_src = saddr;
2186 rth->rt_genid = rt_genid(net);
2187 rth->rt_flags = flags|RTCF_LOCAL;
2188 rth->rt_type = res.type;
2189 rth->rt_key_tos = tos;
2190 rth->rt_dst = daddr;
2191 rth->rt_src = saddr;
2192 #ifdef CONFIG_IP_ROUTE_CLASSID
2193 rth->dst.tclassid = itag;
2195 rth->rt_route_iif = dev->ifindex;
2196 rth->rt_iif = dev->ifindex;
2198 rth->rt_mark = skb->mark;
2199 rth->rt_gateway = daddr;
2200 rth->rt_spec_dst= spec_dst;
2201 rth->rt_peer_genid = 0;
2204 if (res.type == RTN_UNREACHABLE) {
2205 rth->dst.input= ip_error;
2206 rth->dst.error= -err;
2207 rth->rt_flags &= ~RTCF_LOCAL;
2209 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2210 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2217 RT_CACHE_STAT_INC(in_no_route);
2218 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2219 res.type = RTN_UNREACHABLE;
2225 * Do not cache martian addresses: they should be logged (RFC1812)
2227 martian_destination:
2228 RT_CACHE_STAT_INC(in_martian_dst);
2229 #ifdef CONFIG_IP_ROUTE_VERBOSE
2230 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2231 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2232 &daddr, &saddr, dev->name);
2236 err = -EHOSTUNREACH;
2249 martian_source_keep_err:
2250 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2254 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2255 u8 tos, struct net_device *dev, bool noref)
2257 struct rtable * rth;
2259 int iif = dev->ifindex;
2267 if (!rt_caching(net))
2270 tos &= IPTOS_RT_MASK;
2271 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2273 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2274 rth = rcu_dereference(rth->dst.rt_next)) {
2275 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2276 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2277 (rth->rt_iif ^ iif) |
2279 (rth->rt_key_tos ^ tos)) == 0 &&
2280 rth->rt_mark == skb->mark &&
2281 net_eq(dev_net(rth->dst.dev), net) &&
2282 !rt_is_expired(rth)) {
2284 dst_use_noref(&rth->dst, jiffies);
2285 skb_dst_set_noref(skb, &rth->dst);
2287 dst_use(&rth->dst, jiffies);
2288 skb_dst_set(skb, &rth->dst);
2290 RT_CACHE_STAT_INC(in_hit);
2294 RT_CACHE_STAT_INC(in_hlist_search);
2298 /* Multicast recognition logic is moved from route cache to here.
2299 The problem was that too many Ethernet cards have broken/missing
2300 hardware multicast filters :-( As result the host on multicasting
2301 network acquires a lot of useless route cache entries, sort of
2302 SDR messages from all the world. Now we try to get rid of them.
2303 Really, provided software IP multicast filter is organized
2304 reasonably (at least, hashed), it does not result in a slowdown
2305 comparing with route cache reject entries.
2306 Note, that multicast routers are not affected, because
2307 route cache entry is created eventually.
2309 if (ipv4_is_multicast(daddr)) {
2310 struct in_device *in_dev = __in_dev_get_rcu(dev);
2313 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2314 ip_hdr(skb)->protocol);
2316 #ifdef CONFIG_IP_MROUTE
2318 (!ipv4_is_local_multicast(daddr) &&
2319 IN_DEV_MFORWARD(in_dev))
2322 int res = ip_route_input_mc(skb, daddr, saddr,
2331 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2335 EXPORT_SYMBOL(ip_route_input_common);
2337 /* called with rcu_read_lock() */
2338 static struct rtable *__mkroute_output(const struct fib_result *res,
2339 const struct flowi4 *fl4,
2340 __be32 orig_daddr, __be32 orig_saddr,
2341 int orig_oif, struct net_device *dev_out,
2344 struct fib_info *fi = res->fi;
2345 u32 tos = RT_FL_TOS(fl4);
2346 struct in_device *in_dev;
2347 u16 type = res->type;
2350 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2351 return ERR_PTR(-EINVAL);
2353 if (ipv4_is_lbcast(fl4->daddr))
2354 type = RTN_BROADCAST;
2355 else if (ipv4_is_multicast(fl4->daddr))
2356 type = RTN_MULTICAST;
2357 else if (ipv4_is_zeronet(fl4->daddr))
2358 return ERR_PTR(-EINVAL);
2360 if (dev_out->flags & IFF_LOOPBACK)
2361 flags |= RTCF_LOCAL;
2363 in_dev = __in_dev_get_rcu(dev_out);
2365 return ERR_PTR(-EINVAL);
2367 if (type == RTN_BROADCAST) {
2368 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2370 } else if (type == RTN_MULTICAST) {
2371 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2372 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2374 flags &= ~RTCF_LOCAL;
2375 /* If multicast route do not exist use
2376 * default one, but do not gateway in this case.
2379 if (fi && res->prefixlen < 4)
2383 rth = rt_dst_alloc(dev_out,
2384 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2385 IN_DEV_CONF_GET(in_dev, NOXFRM));
2387 return ERR_PTR(-ENOBUFS);
2389 rth->dst.output = ip_output;
2391 rth->rt_key_dst = orig_daddr;
2392 rth->rt_key_src = orig_saddr;
2393 rth->rt_genid = rt_genid(dev_net(dev_out));
2394 rth->rt_flags = flags;
2395 rth->rt_type = type;
2396 rth->rt_key_tos = tos;
2397 rth->rt_dst = fl4->daddr;
2398 rth->rt_src = fl4->saddr;
2399 rth->rt_route_iif = 0;
2400 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2401 rth->rt_oif = orig_oif;
2402 rth->rt_mark = fl4->flowi4_mark;
2403 rth->rt_gateway = fl4->daddr;
2404 rth->rt_spec_dst= fl4->saddr;
2405 rth->rt_peer_genid = 0;
2409 RT_CACHE_STAT_INC(out_slow_tot);
2411 if (flags & RTCF_LOCAL) {
2412 rth->dst.input = ip_local_deliver;
2413 rth->rt_spec_dst = fl4->daddr;
2415 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2416 rth->rt_spec_dst = fl4->saddr;
2417 if (flags & RTCF_LOCAL &&
2418 !(dev_out->flags & IFF_LOOPBACK)) {
2419 rth->dst.output = ip_mc_output;
2420 RT_CACHE_STAT_INC(out_slow_mc);
2422 #ifdef CONFIG_IP_MROUTE
2423 if (type == RTN_MULTICAST) {
2424 if (IN_DEV_MFORWARD(in_dev) &&
2425 !ipv4_is_local_multicast(fl4->daddr)) {
2426 rth->dst.input = ip_mr_input;
2427 rth->dst.output = ip_mc_output;
2433 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2439 * Major route resolver routine.
2440 * called with rcu_read_lock();
2443 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2445 struct net_device *dev_out = NULL;
2446 u32 tos = RT_FL_TOS(fl4);
2447 unsigned int flags = 0;
2448 struct fib_result res;
2455 #ifdef CONFIG_IP_MULTIPLE_TABLES
2459 orig_daddr = fl4->daddr;
2460 orig_saddr = fl4->saddr;
2461 orig_oif = fl4->flowi4_oif;
2463 fl4->flowi4_iif = net->loopback_dev->ifindex;
2464 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2465 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2466 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2470 rth = ERR_PTR(-EINVAL);
2471 if (ipv4_is_multicast(fl4->saddr) ||
2472 ipv4_is_lbcast(fl4->saddr) ||
2473 ipv4_is_zeronet(fl4->saddr))
2476 /* I removed check for oif == dev_out->oif here.
2477 It was wrong for two reasons:
2478 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2479 is assigned to multiple interfaces.
2480 2. Moreover, we are allowed to send packets with saddr
2481 of another iface. --ANK
2484 if (fl4->flowi4_oif == 0 &&
2485 (ipv4_is_multicast(fl4->daddr) ||
2486 ipv4_is_lbcast(fl4->daddr))) {
2487 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2488 dev_out = __ip_dev_find(net, fl4->saddr, false);
2489 if (dev_out == NULL)
2492 /* Special hack: user can direct multicasts
2493 and limited broadcast via necessary interface
2494 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2495 This hack is not just for fun, it allows
2496 vic,vat and friends to work.
2497 They bind socket to loopback, set ttl to zero
2498 and expect that it will work.
2499 From the viewpoint of routing cache they are broken,
2500 because we are not allowed to build multicast path
2501 with loopback source addr (look, routing cache
2502 cannot know, that ttl is zero, so that packet
2503 will not leave this host and route is valid).
2504 Luckily, this hack is good workaround.
2507 fl4->flowi4_oif = dev_out->ifindex;
2511 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2512 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2513 if (!__ip_dev_find(net, fl4->saddr, false))
2519 if (fl4->flowi4_oif) {
2520 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2521 rth = ERR_PTR(-ENODEV);
2522 if (dev_out == NULL)
2525 /* RACE: Check return value of inet_select_addr instead. */
2526 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2527 rth = ERR_PTR(-ENETUNREACH);
2530 if (ipv4_is_local_multicast(fl4->daddr) ||
2531 ipv4_is_lbcast(fl4->daddr)) {
2533 fl4->saddr = inet_select_addr(dev_out, 0,
2538 if (ipv4_is_multicast(fl4->daddr))
2539 fl4->saddr = inet_select_addr(dev_out, 0,
2541 else if (!fl4->daddr)
2542 fl4->saddr = inet_select_addr(dev_out, 0,
2548 fl4->daddr = fl4->saddr;
2550 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2551 dev_out = net->loopback_dev;
2552 fl4->flowi4_oif = net->loopback_dev->ifindex;
2553 res.type = RTN_LOCAL;
2554 flags |= RTCF_LOCAL;
2558 if (fib_lookup(net, fl4, &res)) {
2560 if (fl4->flowi4_oif) {
2561 /* Apparently, routing tables are wrong. Assume,
2562 that the destination is on link.
2565 Because we are allowed to send to iface
2566 even if it has NO routes and NO assigned
2567 addresses. When oif is specified, routing
2568 tables are looked up with only one purpose:
2569 to catch if destination is gatewayed, rather than
2570 direct. Moreover, if MSG_DONTROUTE is set,
2571 we send packet, ignoring both routing tables
2572 and ifaddr state. --ANK
2575 We could make it even if oif is unknown,
2576 likely IPv6, but we do not.
2579 if (fl4->saddr == 0)
2580 fl4->saddr = inet_select_addr(dev_out, 0,
2582 res.type = RTN_UNICAST;
2585 rth = ERR_PTR(-ENETUNREACH);
2589 if (res.type == RTN_LOCAL) {
2591 if (res.fi->fib_prefsrc)
2592 fl4->saddr = res.fi->fib_prefsrc;
2594 fl4->saddr = fl4->daddr;
2596 dev_out = net->loopback_dev;
2597 fl4->flowi4_oif = dev_out->ifindex;
2599 flags |= RTCF_LOCAL;
2603 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2604 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2605 fib_select_multipath(&res);
2608 if (!res.prefixlen &&
2609 res.table->tb_num_default > 1 &&
2610 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2611 fib_select_default(&res);
2614 fl4->saddr = FIB_RES_PREFSRC(net, res);
2616 dev_out = FIB_RES_DEV(res);
2617 fl4->flowi4_oif = dev_out->ifindex;
2621 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2626 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2627 rt_genid(dev_net(dev_out)));
2628 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2636 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2641 if (!rt_caching(net))
2644 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2647 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2648 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2649 if (rth->rt_key_dst == flp4->daddr &&
2650 rth->rt_key_src == flp4->saddr &&
2651 rt_is_output_route(rth) &&
2652 rth->rt_oif == flp4->flowi4_oif &&
2653 rth->rt_mark == flp4->flowi4_mark &&
2654 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2655 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2656 net_eq(dev_net(rth->dst.dev), net) &&
2657 !rt_is_expired(rth)) {
2658 dst_use(&rth->dst, jiffies);
2659 RT_CACHE_STAT_INC(out_hit);
2660 rcu_read_unlock_bh();
2662 flp4->saddr = rth->rt_src;
2664 flp4->daddr = rth->rt_dst;
2667 RT_CACHE_STAT_INC(out_hlist_search);
2669 rcu_read_unlock_bh();
2672 return ip_route_output_slow(net, flp4);
2674 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2676 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2681 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2686 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2690 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2696 static struct dst_ops ipv4_dst_blackhole_ops = {
2698 .protocol = cpu_to_be16(ETH_P_IP),
2699 .destroy = ipv4_dst_destroy,
2700 .check = ipv4_blackhole_dst_check,
2701 .default_mtu = ipv4_blackhole_default_mtu,
2702 .default_advmss = ipv4_default_advmss,
2703 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2704 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2707 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2709 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2710 struct rtable *ort = (struct rtable *) dst_orig;
2713 struct dst_entry *new = &rt->dst;
2716 new->input = dst_discard;
2717 new->output = dst_discard;
2718 dst_copy_metrics(new, &ort->dst);
2720 new->dev = ort->dst.dev;
2724 rt->rt_key_dst = ort->rt_key_dst;
2725 rt->rt_key_src = ort->rt_key_src;
2726 rt->rt_key_tos = ort->rt_key_tos;
2727 rt->rt_route_iif = ort->rt_route_iif;
2728 rt->rt_iif = ort->rt_iif;
2729 rt->rt_oif = ort->rt_oif;
2730 rt->rt_mark = ort->rt_mark;
2732 rt->rt_genid = rt_genid(net);
2733 rt->rt_flags = ort->rt_flags;
2734 rt->rt_type = ort->rt_type;
2735 rt->rt_dst = ort->rt_dst;
2736 rt->rt_src = ort->rt_src;
2737 rt->rt_gateway = ort->rt_gateway;
2738 rt->rt_spec_dst = ort->rt_spec_dst;
2739 rt->peer = ort->peer;
2741 atomic_inc(&rt->peer->refcnt);
2744 atomic_inc(&rt->fi->fib_clntref);
2749 dst_release(dst_orig);
2751 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2754 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2757 struct rtable *rt = __ip_route_output_key(net, flp4);
2762 if (flp4->flowi4_proto)
2763 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2764 flowi4_to_flowi(flp4),
2769 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2771 static int rt_fill_info(struct net *net,
2772 struct sk_buff *skb, u32 pid, u32 seq, int event,
2773 int nowait, unsigned int flags)
2775 struct rtable *rt = skb_rtable(skb);
2777 struct nlmsghdr *nlh;
2779 u32 id = 0, ts = 0, tsage = 0, error;
2781 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2785 r = nlmsg_data(nlh);
2786 r->rtm_family = AF_INET;
2787 r->rtm_dst_len = 32;
2789 r->rtm_tos = rt->rt_key_tos;
2790 r->rtm_table = RT_TABLE_MAIN;
2791 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2792 r->rtm_type = rt->rt_type;
2793 r->rtm_scope = RT_SCOPE_UNIVERSE;
2794 r->rtm_protocol = RTPROT_UNSPEC;
2795 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2796 if (rt->rt_flags & RTCF_NOTIFY)
2797 r->rtm_flags |= RTM_F_NOTIFY;
2799 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2801 if (rt->rt_key_src) {
2802 r->rtm_src_len = 32;
2803 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2806 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2807 #ifdef CONFIG_IP_ROUTE_CLASSID
2808 if (rt->dst.tclassid)
2809 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2811 if (rt_is_input_route(rt))
2812 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2813 else if (rt->rt_src != rt->rt_key_src)
2814 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2816 if (rt->rt_dst != rt->rt_gateway)
2817 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2819 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2820 goto nla_put_failure;
2823 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2825 error = rt->dst.error;
2826 expires = (rt->peer && rt->peer->pmtu_expires) ?
2827 rt->peer->pmtu_expires - jiffies : 0;
2829 inet_peer_refcheck(rt->peer);
2830 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2831 if (rt->peer->tcp_ts_stamp) {
2832 ts = rt->peer->tcp_ts;
2833 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2837 if (rt_is_input_route(rt)) {
2838 #ifdef CONFIG_IP_MROUTE
2839 __be32 dst = rt->rt_dst;
2841 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2842 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2843 int err = ipmr_get_route(net, skb,
2844 rt->rt_src, rt->rt_dst,
2850 goto nla_put_failure;
2852 if (err == -EMSGSIZE)
2853 goto nla_put_failure;
2859 NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2862 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2863 expires, error) < 0)
2864 goto nla_put_failure;
2866 return nlmsg_end(skb, nlh);
2869 nlmsg_cancel(skb, nlh);
2873 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2875 struct net *net = sock_net(in_skb->sk);
2877 struct nlattr *tb[RTA_MAX+1];
2878 struct rtable *rt = NULL;
2884 struct sk_buff *skb;
2886 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2890 rtm = nlmsg_data(nlh);
2892 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2898 /* Reserve room for dummy headers, this skb can pass
2899 through good chunk of routing engine.
2901 skb_reset_mac_header(skb);
2902 skb_reset_network_header(skb);
2904 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2905 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2906 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2908 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2909 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2910 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2911 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2914 struct net_device *dev;
2916 dev = __dev_get_by_index(net, iif);
2922 skb->protocol = htons(ETH_P_IP);
2926 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2929 rt = skb_rtable(skb);
2930 if (err == 0 && rt->dst.error)
2931 err = -rt->dst.error;
2933 struct flowi4 fl4 = {
2936 .flowi4_tos = rtm->rtm_tos,
2937 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2938 .flowi4_mark = mark,
2940 rt = ip_route_output_key(net, &fl4);
2950 skb_dst_set(skb, &rt->dst);
2951 if (rtm->rtm_flags & RTM_F_NOTIFY)
2952 rt->rt_flags |= RTCF_NOTIFY;
2954 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2955 RTM_NEWROUTE, 0, 0);
2959 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2968 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2975 net = sock_net(skb->sk);
2980 s_idx = idx = cb->args[1];
2981 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2982 if (!rt_hash_table[h].chain)
2985 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2986 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2987 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2989 if (rt_is_expired(rt))
2991 skb_dst_set_noref(skb, &rt->dst);
2992 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2993 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2994 1, NLM_F_MULTI) <= 0) {
2996 rcu_read_unlock_bh();
3001 rcu_read_unlock_bh();
3010 void ip_rt_multicast_event(struct in_device *in_dev)
3012 rt_cache_flush(dev_net(in_dev->dev), 0);
3015 #ifdef CONFIG_SYSCTL
3016 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3017 void __user *buffer,
3018 size_t *lenp, loff_t *ppos)
3025 memcpy(&ctl, __ctl, sizeof(ctl));
3026 ctl.data = &flush_delay;
3027 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3029 net = (struct net *)__ctl->extra1;
3030 rt_cache_flush(net, flush_delay);
3037 static ctl_table ipv4_route_table[] = {
3039 .procname = "gc_thresh",
3040 .data = &ipv4_dst_ops.gc_thresh,
3041 .maxlen = sizeof(int),
3043 .proc_handler = proc_dointvec,
3046 .procname = "max_size",
3047 .data = &ip_rt_max_size,
3048 .maxlen = sizeof(int),
3050 .proc_handler = proc_dointvec,
3053 /* Deprecated. Use gc_min_interval_ms */
3055 .procname = "gc_min_interval",
3056 .data = &ip_rt_gc_min_interval,
3057 .maxlen = sizeof(int),
3059 .proc_handler = proc_dointvec_jiffies,
3062 .procname = "gc_min_interval_ms",
3063 .data = &ip_rt_gc_min_interval,
3064 .maxlen = sizeof(int),
3066 .proc_handler = proc_dointvec_ms_jiffies,
3069 .procname = "gc_timeout",
3070 .data = &ip_rt_gc_timeout,
3071 .maxlen = sizeof(int),
3073 .proc_handler = proc_dointvec_jiffies,
3076 .procname = "gc_interval",
3077 .data = &ip_rt_gc_interval,
3078 .maxlen = sizeof(int),
3080 .proc_handler = proc_dointvec_jiffies,
3083 .procname = "redirect_load",
3084 .data = &ip_rt_redirect_load,
3085 .maxlen = sizeof(int),
3087 .proc_handler = proc_dointvec,
3090 .procname = "redirect_number",
3091 .data = &ip_rt_redirect_number,
3092 .maxlen = sizeof(int),
3094 .proc_handler = proc_dointvec,
3097 .procname = "redirect_silence",
3098 .data = &ip_rt_redirect_silence,
3099 .maxlen = sizeof(int),
3101 .proc_handler = proc_dointvec,
3104 .procname = "error_cost",
3105 .data = &ip_rt_error_cost,
3106 .maxlen = sizeof(int),
3108 .proc_handler = proc_dointvec,
3111 .procname = "error_burst",
3112 .data = &ip_rt_error_burst,
3113 .maxlen = sizeof(int),
3115 .proc_handler = proc_dointvec,
3118 .procname = "gc_elasticity",
3119 .data = &ip_rt_gc_elasticity,
3120 .maxlen = sizeof(int),
3122 .proc_handler = proc_dointvec,
3125 .procname = "mtu_expires",
3126 .data = &ip_rt_mtu_expires,
3127 .maxlen = sizeof(int),
3129 .proc_handler = proc_dointvec_jiffies,
3132 .procname = "min_pmtu",
3133 .data = &ip_rt_min_pmtu,
3134 .maxlen = sizeof(int),
3136 .proc_handler = proc_dointvec,
3139 .procname = "min_adv_mss",
3140 .data = &ip_rt_min_advmss,
3141 .maxlen = sizeof(int),
3143 .proc_handler = proc_dointvec,
3148 static struct ctl_table empty[1];
3150 static struct ctl_table ipv4_skeleton[] =
3152 { .procname = "route",
3153 .mode = 0555, .child = ipv4_route_table},
3154 { .procname = "neigh",
3155 .mode = 0555, .child = empty},
3159 static __net_initdata struct ctl_path ipv4_path[] = {
3160 { .procname = "net", },
3161 { .procname = "ipv4", },
3165 static struct ctl_table ipv4_route_flush_table[] = {
3167 .procname = "flush",
3168 .maxlen = sizeof(int),
3170 .proc_handler = ipv4_sysctl_rtcache_flush,
3175 static __net_initdata struct ctl_path ipv4_route_path[] = {
3176 { .procname = "net", },
3177 { .procname = "ipv4", },
3178 { .procname = "route", },
3182 static __net_init int sysctl_route_net_init(struct net *net)
3184 struct ctl_table *tbl;
3186 tbl = ipv4_route_flush_table;
3187 if (!net_eq(net, &init_net)) {
3188 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3192 tbl[0].extra1 = net;
3194 net->ipv4.route_hdr =
3195 register_net_sysctl_table(net, ipv4_route_path, tbl);
3196 if (net->ipv4.route_hdr == NULL)
3201 if (tbl != ipv4_route_flush_table)
3207 static __net_exit void sysctl_route_net_exit(struct net *net)
3209 struct ctl_table *tbl;
3211 tbl = net->ipv4.route_hdr->ctl_table_arg;
3212 unregister_net_sysctl_table(net->ipv4.route_hdr);
3213 BUG_ON(tbl == ipv4_route_flush_table);
3217 static __net_initdata struct pernet_operations sysctl_route_ops = {
3218 .init = sysctl_route_net_init,
3219 .exit = sysctl_route_net_exit,
3223 static __net_init int rt_genid_init(struct net *net)
3225 get_random_bytes(&net->ipv4.rt_genid,
3226 sizeof(net->ipv4.rt_genid));
3227 get_random_bytes(&net->ipv4.dev_addr_genid,
3228 sizeof(net->ipv4.dev_addr_genid));
3232 static __net_initdata struct pernet_operations rt_genid_ops = {
3233 .init = rt_genid_init,
3237 #ifdef CONFIG_IP_ROUTE_CLASSID
3238 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3239 #endif /* CONFIG_IP_ROUTE_CLASSID */
3241 static __initdata unsigned long rhash_entries;
3242 static int __init set_rhash_entries(char *str)
3246 rhash_entries = simple_strtoul(str, &str, 0);
3249 __setup("rhash_entries=", set_rhash_entries);
3251 int __init ip_rt_init(void)
3255 #ifdef CONFIG_IP_ROUTE_CLASSID
3256 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3258 panic("IP: failed to allocate ip_rt_acct\n");
3261 ipv4_dst_ops.kmem_cachep =
3262 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3263 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3265 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3267 if (dst_entries_init(&ipv4_dst_ops) < 0)
3268 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3270 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3271 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3273 rt_hash_table = (struct rt_hash_bucket *)
3274 alloc_large_system_hash("IP route cache",
3275 sizeof(struct rt_hash_bucket),
3277 (totalram_pages >= 128 * 1024) ?
3282 rhash_entries ? 0 : 512 * 1024);
3283 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3284 rt_hash_lock_init();
3286 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3287 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3292 if (ip_rt_proc_init())
3293 printk(KERN_ERR "Unable to create route proc files\n");
3296 xfrm4_init(ip_rt_max_size);
3298 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3300 #ifdef CONFIG_SYSCTL
3301 register_pernet_subsys(&sysctl_route_ops);
3303 register_pernet_subsys(&rt_genid_ops);
3307 #ifdef CONFIG_SYSCTL
3309 * We really need to sanitize the damn ipv4 init order, then all
3310 * this nonsense will go away.
3312 void __init ip_static_sysctl_init(void)
3314 register_sysctl_paths(ipv4_path, ipv4_skeleton);