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(oldflp) \
113 ((u32)(oldflp->fl4_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;
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] = {
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 flowi *fl1,
715 const struct flowi *fl2)
717 return ((((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
718 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
719 (fl1->iif ^ fl2->iif)) == 0);
722 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
724 return (((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
725 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
726 (fl1->mark ^ fl2->mark) |
727 (*(u16 *)&fl1->fl4_tos ^ *(u16 *)&fl2->fl4_tos) |
728 (fl1->oif ^ fl2->oif) |
729 (fl1->iif ^ fl2->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->fl, &rth->fl))
818 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
824 * Pertubation 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);
971 #if RT_CACHE_DEBUG >= 2
972 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
973 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
976 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
978 } while (!in_softirq() && time_before_eq(jiffies, now));
980 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
982 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
985 printk(KERN_WARNING "dst cache overflow\n");
986 RT_CACHE_STAT_INC(gc_dst_overflow);
990 expire += ip_rt_gc_min_interval;
991 if (expire > ip_rt_gc_timeout ||
992 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
993 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
994 expire = ip_rt_gc_timeout;
995 #if RT_CACHE_DEBUG >= 2
996 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
997 dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
1003 * Returns number of entries in a hash chain that have different hash_inputs
1005 static int slow_chain_length(const struct rtable *head)
1008 const struct rtable *rth = head;
1011 length += has_noalias(head, rth);
1012 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1014 return length >> FRACT_BITS;
1017 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1018 struct rtable **rp, struct sk_buff *skb, int ifindex)
1020 struct rtable *rth, *cand;
1021 struct rtable __rcu **rthp, **candp;
1025 int attempts = !in_softirq();
1029 min_score = ~(u32)0;
1034 if (!rt_caching(dev_net(rt->dst.dev))) {
1036 * If we're not caching, just tell the caller we
1037 * were successful and don't touch the route. The
1038 * caller hold the sole reference to the cache entry, and
1039 * it will be released when the caller is done with it.
1040 * If we drop it here, the callers have no way to resolve routes
1041 * when we're not caching. Instead, just point *rp at rt, so
1042 * the caller gets a single use out of the route
1043 * Note that we do rt_free on this new route entry, so that
1044 * once its refcount hits zero, we are still able to reap it
1046 * Note: To avoid expensive rcu stuff for this uncached dst,
1047 * we set DST_NOCACHE so that dst_release() can free dst without
1048 * waiting a grace period.
1051 rt->dst.flags |= DST_NOCACHE;
1052 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1053 int err = arp_bind_neighbour(&rt->dst);
1055 if (net_ratelimit())
1057 "Neighbour table failure & not caching routes.\n");
1066 rthp = &rt_hash_table[hash].chain;
1068 spin_lock_bh(rt_hash_lock_addr(hash));
1069 while ((rth = rcu_dereference_protected(*rthp,
1070 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1071 if (rt_is_expired(rth)) {
1072 *rthp = rth->dst.rt_next;
1076 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1078 *rthp = rth->dst.rt_next;
1080 * Since lookup is lockfree, the deletion
1081 * must be visible to another weakly ordered CPU before
1082 * the insertion at the start of the hash chain.
1084 rcu_assign_pointer(rth->dst.rt_next,
1085 rt_hash_table[hash].chain);
1087 * Since lookup is lockfree, the update writes
1088 * must be ordered for consistency on SMP.
1090 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1092 dst_use(&rth->dst, now);
1093 spin_unlock_bh(rt_hash_lock_addr(hash));
1099 skb_dst_set(skb, &rth->dst);
1103 if (!atomic_read(&rth->dst.__refcnt)) {
1104 u32 score = rt_score(rth);
1106 if (score <= min_score) {
1115 rthp = &rth->dst.rt_next;
1119 /* ip_rt_gc_elasticity used to be average length of chain
1120 * length, when exceeded gc becomes really aggressive.
1122 * The second limit is less certain. At the moment it allows
1123 * only 2 entries per bucket. We will see.
1125 if (chain_length > ip_rt_gc_elasticity) {
1126 *candp = cand->dst.rt_next;
1130 if (chain_length > rt_chain_length_max &&
1131 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1132 struct net *net = dev_net(rt->dst.dev);
1133 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1134 if (!rt_caching(net)) {
1135 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1136 rt->dst.dev->name, num);
1138 rt_emergency_hash_rebuild(net);
1139 spin_unlock_bh(rt_hash_lock_addr(hash));
1141 hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1142 ifindex, rt_genid(net));
1147 /* Try to bind route to arp only if it is output
1148 route or unicast forwarding path.
1150 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1151 int err = arp_bind_neighbour(&rt->dst);
1153 spin_unlock_bh(rt_hash_lock_addr(hash));
1155 if (err != -ENOBUFS) {
1160 /* Neighbour tables are full and nothing
1161 can be released. Try to shrink route cache,
1162 it is most likely it holds some neighbour records.
1164 if (attempts-- > 0) {
1165 int saved_elasticity = ip_rt_gc_elasticity;
1166 int saved_int = ip_rt_gc_min_interval;
1167 ip_rt_gc_elasticity = 1;
1168 ip_rt_gc_min_interval = 0;
1169 rt_garbage_collect(&ipv4_dst_ops);
1170 ip_rt_gc_min_interval = saved_int;
1171 ip_rt_gc_elasticity = saved_elasticity;
1175 if (net_ratelimit())
1176 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1182 rt->dst.rt_next = rt_hash_table[hash].chain;
1184 #if RT_CACHE_DEBUG >= 2
1185 if (rt->dst.rt_next) {
1187 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1189 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1190 printk(" . %pI4", &trt->rt_dst);
1195 * Since lookup is lockfree, we must make sure
1196 * previous writes to rt are comitted to memory
1197 * before making rt visible to other CPUS.
1199 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1201 spin_unlock_bh(rt_hash_lock_addr(hash));
1207 skb_dst_set(skb, &rt->dst);
1211 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1213 static u32 rt_peer_genid(void)
1215 return atomic_read(&__rt_peer_genid);
1218 void rt_bind_peer(struct rtable *rt, int create)
1220 struct inet_peer *peer;
1222 peer = inet_getpeer_v4(rt->rt_dst, create);
1224 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1227 rt->rt_peer_genid = rt_peer_genid();
1231 * Peer allocation may fail only in serious out-of-memory conditions. However
1232 * we still can generate some output.
1233 * Random ID selection looks a bit dangerous because we have no chances to
1234 * select ID being unique in a reasonable period of time.
1235 * But broken packet identifier may be better than no packet at all.
1237 static void ip_select_fb_ident(struct iphdr *iph)
1239 static DEFINE_SPINLOCK(ip_fb_id_lock);
1240 static u32 ip_fallback_id;
1243 spin_lock_bh(&ip_fb_id_lock);
1244 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1245 iph->id = htons(salt & 0xFFFF);
1246 ip_fallback_id = salt;
1247 spin_unlock_bh(&ip_fb_id_lock);
1250 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1252 struct rtable *rt = (struct rtable *) dst;
1255 if (rt->peer == NULL)
1256 rt_bind_peer(rt, 1);
1258 /* If peer is attached to destination, it is never detached,
1259 so that we need not to grab a lock to dereference it.
1262 iph->id = htons(inet_getid(rt->peer, more));
1266 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1267 __builtin_return_address(0));
1269 ip_select_fb_ident(iph);
1271 EXPORT_SYMBOL(__ip_select_ident);
1273 static void rt_del(unsigned hash, struct rtable *rt)
1275 struct rtable __rcu **rthp;
1278 rthp = &rt_hash_table[hash].chain;
1279 spin_lock_bh(rt_hash_lock_addr(hash));
1281 while ((aux = rcu_dereference_protected(*rthp,
1282 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1283 if (aux == rt || rt_is_expired(aux)) {
1284 *rthp = aux->dst.rt_next;
1288 rthp = &aux->dst.rt_next;
1290 spin_unlock_bh(rt_hash_lock_addr(hash));
1293 /* called in rcu_read_lock() section */
1294 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1295 __be32 saddr, struct net_device *dev)
1297 struct in_device *in_dev = __in_dev_get_rcu(dev);
1298 struct inet_peer *peer;
1305 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1306 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1307 ipv4_is_zeronet(new_gw))
1308 goto reject_redirect;
1310 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1311 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1312 goto reject_redirect;
1313 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1314 goto reject_redirect;
1316 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1317 goto reject_redirect;
1320 peer = inet_getpeer_v4(daddr, 1);
1322 peer->redirect_learned.a4 = new_gw;
1326 atomic_inc(&__rt_peer_genid);
1331 #ifdef CONFIG_IP_ROUTE_VERBOSE
1332 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1333 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1334 " Advised path = %pI4 -> %pI4\n",
1335 &old_gw, dev->name, &new_gw,
1341 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1343 struct rtable *rt = (struct rtable *)dst;
1344 struct dst_entry *ret = dst;
1347 if (dst->obsolete > 0) {
1350 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1351 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1353 rt_genid(dev_net(dst->dev)));
1354 #if RT_CACHE_DEBUG >= 1
1355 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1356 &rt->rt_dst, rt->fl.fl4_tos);
1360 } else if (rt->peer &&
1361 rt->peer->pmtu_expires &&
1362 time_after_eq(jiffies, rt->peer->pmtu_expires)) {
1363 unsigned long orig = rt->peer->pmtu_expires;
1365 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1366 dst_metric_set(dst, RTAX_MTU,
1367 rt->peer->pmtu_orig);
1375 * 1. The first ip_rt_redirect_number redirects are sent
1376 * with exponential backoff, then we stop sending them at all,
1377 * assuming that the host ignores our redirects.
1378 * 2. If we did not see packets requiring redirects
1379 * during ip_rt_redirect_silence, we assume that the host
1380 * forgot redirected route and start to send redirects again.
1382 * This algorithm is much cheaper and more intelligent than dumb load limiting
1385 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1386 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1389 void ip_rt_send_redirect(struct sk_buff *skb)
1391 struct rtable *rt = skb_rtable(skb);
1392 struct in_device *in_dev;
1393 struct inet_peer *peer;
1397 in_dev = __in_dev_get_rcu(rt->dst.dev);
1398 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1402 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1406 rt_bind_peer(rt, 1);
1409 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1413 /* No redirected packets during ip_rt_redirect_silence;
1414 * reset the algorithm.
1416 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1417 peer->rate_tokens = 0;
1419 /* Too many ignored redirects; do not send anything
1420 * set dst.rate_last to the last seen redirected packet.
1422 if (peer->rate_tokens >= ip_rt_redirect_number) {
1423 peer->rate_last = jiffies;
1427 /* Check for load limit; set rate_last to the latest sent
1430 if (peer->rate_tokens == 0 ||
1433 (ip_rt_redirect_load << peer->rate_tokens)))) {
1434 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1435 peer->rate_last = jiffies;
1436 ++peer->rate_tokens;
1437 #ifdef CONFIG_IP_ROUTE_VERBOSE
1439 peer->rate_tokens == ip_rt_redirect_number &&
1441 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1442 &rt->rt_src, rt->rt_iif,
1443 &rt->rt_dst, &rt->rt_gateway);
1448 static int ip_error(struct sk_buff *skb)
1450 struct rtable *rt = skb_rtable(skb);
1451 struct inet_peer *peer;
1456 switch (rt->dst.error) {
1461 code = ICMP_HOST_UNREACH;
1464 code = ICMP_NET_UNREACH;
1465 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1466 IPSTATS_MIB_INNOROUTES);
1469 code = ICMP_PKT_FILTERED;
1474 rt_bind_peer(rt, 1);
1480 peer->rate_tokens += now - peer->rate_last;
1481 if (peer->rate_tokens > ip_rt_error_burst)
1482 peer->rate_tokens = ip_rt_error_burst;
1483 peer->rate_last = now;
1484 if (peer->rate_tokens >= ip_rt_error_cost)
1485 peer->rate_tokens -= ip_rt_error_cost;
1490 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1492 out: kfree_skb(skb);
1497 * The last two values are not from the RFC but
1498 * are needed for AMPRnet AX.25 paths.
1501 static const unsigned short mtu_plateau[] =
1502 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1504 static inline unsigned short guess_mtu(unsigned short old_mtu)
1508 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1509 if (old_mtu > mtu_plateau[i])
1510 return mtu_plateau[i];
1514 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1515 unsigned short new_mtu,
1516 struct net_device *dev)
1518 unsigned short old_mtu = ntohs(iph->tot_len);
1519 unsigned short est_mtu = 0;
1520 struct inet_peer *peer;
1522 peer = inet_getpeer_v4(iph->daddr, 1);
1524 unsigned short mtu = new_mtu;
1526 if (new_mtu < 68 || new_mtu >= old_mtu) {
1527 /* BSD 4.2 derived systems incorrectly adjust
1528 * tot_len by the IP header length, and report
1529 * a zero MTU in the ICMP message.
1532 old_mtu >= 68 + (iph->ihl << 2))
1533 old_mtu -= iph->ihl << 2;
1534 mtu = guess_mtu(old_mtu);
1537 if (mtu < ip_rt_min_pmtu)
1538 mtu = ip_rt_min_pmtu;
1539 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1541 peer->pmtu_learned = mtu;
1542 peer->pmtu_expires = jiffies + ip_rt_mtu_expires;
1547 atomic_inc(&__rt_peer_genid);
1549 return est_mtu ? : new_mtu;
1552 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1554 unsigned long expires = peer->pmtu_expires;
1556 if (time_before(expires, jiffies)) {
1557 u32 orig_dst_mtu = dst_mtu(dst);
1558 if (peer->pmtu_learned < orig_dst_mtu) {
1559 if (!peer->pmtu_orig)
1560 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1561 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1563 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1564 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1567 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1569 struct rtable *rt = (struct rtable *) dst;
1570 struct inet_peer *peer;
1575 rt_bind_peer(rt, 1);
1578 if (mtu < ip_rt_min_pmtu)
1579 mtu = ip_rt_min_pmtu;
1580 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1581 peer->pmtu_learned = mtu;
1582 peer->pmtu_expires = jiffies + ip_rt_mtu_expires;
1584 atomic_inc(&__rt_peer_genid);
1585 rt->rt_peer_genid = rt_peer_genid();
1587 check_peer_pmtu(dst, peer);
1593 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1595 struct rtable *rt = (struct rtable *) dst;
1596 __be32 orig_gw = rt->rt_gateway;
1598 dst_confirm(&rt->dst);
1600 neigh_release(rt->dst.neighbour);
1601 rt->dst.neighbour = NULL;
1603 rt->rt_gateway = peer->redirect_learned.a4;
1604 if (arp_bind_neighbour(&rt->dst) ||
1605 !(rt->dst.neighbour->nud_state & NUD_VALID)) {
1606 if (rt->dst.neighbour)
1607 neigh_event_send(rt->dst.neighbour, NULL);
1608 rt->rt_gateway = orig_gw;
1611 rt->rt_flags |= RTCF_REDIRECTED;
1612 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE,
1618 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1620 struct rtable *rt = (struct rtable *) dst;
1622 if (rt_is_expired(rt))
1624 if (rt->rt_peer_genid != rt_peer_genid()) {
1625 struct inet_peer *peer;
1628 rt_bind_peer(rt, 0);
1631 if (peer && peer->pmtu_expires)
1632 check_peer_pmtu(dst, peer);
1634 if (peer && peer->redirect_learned.a4 &&
1635 peer->redirect_learned.a4 != rt->rt_gateway) {
1636 if (check_peer_redir(dst, peer))
1640 rt->rt_peer_genid = rt_peer_genid();
1645 static void ipv4_dst_destroy(struct dst_entry *dst)
1647 struct rtable *rt = (struct rtable *) dst;
1648 struct inet_peer *peer = rt->peer;
1651 fib_info_put(rt->fi);
1661 static void ipv4_link_failure(struct sk_buff *skb)
1665 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1667 rt = skb_rtable(skb);
1670 rt->peer->pmtu_expires) {
1671 unsigned long orig = rt->peer->pmtu_expires;
1673 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1674 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1678 static int ip_rt_bug(struct sk_buff *skb)
1680 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1681 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1682 skb->dev ? skb->dev->name : "?");
1688 We do not cache source address of outgoing interface,
1689 because it is used only by IP RR, TS and SRR options,
1690 so that it out of fast path.
1692 BTW remember: "addr" is allowed to be not aligned
1696 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1699 struct fib_result res;
1701 if (rt_is_output_route(rt))
1705 if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0)
1706 src = FIB_RES_PREFSRC(res);
1708 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1712 memcpy(addr, &src, 4);
1715 #ifdef CONFIG_IP_ROUTE_CLASSID
1716 static void set_class_tag(struct rtable *rt, u32 tag)
1718 if (!(rt->dst.tclassid & 0xFFFF))
1719 rt->dst.tclassid |= tag & 0xFFFF;
1720 if (!(rt->dst.tclassid & 0xFFFF0000))
1721 rt->dst.tclassid |= tag & 0xFFFF0000;
1725 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1727 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1730 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1732 if (advmss > 65535 - 40)
1733 advmss = 65535 - 40;
1738 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1740 unsigned int mtu = dst->dev->mtu;
1742 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1743 const struct rtable *rt = (const struct rtable *) dst;
1745 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1749 if (mtu > IP_MAX_MTU)
1755 static void rt_init_metrics(struct rtable *rt, struct fib_info *fi)
1757 struct inet_peer *peer;
1760 /* If a peer entry exists for this destination, we must hook
1761 * it up in order to get at cached metrics.
1763 if (rt->fl.flags & FLOWI_FLAG_PRECOW_METRICS)
1766 rt_bind_peer(rt, create);
1769 if (inet_metrics_new(peer))
1770 memcpy(peer->metrics, fi->fib_metrics,
1771 sizeof(u32) * RTAX_MAX);
1772 dst_init_metrics(&rt->dst, peer->metrics, false);
1774 if (peer->pmtu_expires)
1775 check_peer_pmtu(&rt->dst, peer);
1776 if (peer->redirect_learned.a4 &&
1777 peer->redirect_learned.a4 != rt->rt_gateway) {
1778 rt->rt_gateway = peer->redirect_learned.a4;
1779 rt->rt_flags |= RTCF_REDIRECTED;
1782 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1784 atomic_inc(&fi->fib_clntref);
1786 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1790 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1792 struct dst_entry *dst = &rt->dst;
1793 struct fib_info *fi = res->fi;
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, 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);
1816 rt->rt_type = res->type;
1819 /* called in rcu_read_lock() section */
1820 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1821 u8 tos, struct net_device *dev, int our)
1826 struct in_device *in_dev = __in_dev_get_rcu(dev);
1830 /* Primary sanity checks. */
1835 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1836 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1839 if (ipv4_is_zeronet(saddr)) {
1840 if (!ipv4_is_local_multicast(daddr))
1842 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1844 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1849 rth = dst_alloc(&ipv4_dst_ops);
1853 rth->dst.output = ip_rt_bug;
1854 rth->dst.obsolete = -1;
1856 atomic_set(&rth->dst.__refcnt, 1);
1857 rth->dst.flags= DST_HOST;
1858 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1859 rth->dst.flags |= DST_NOPOLICY;
1860 rth->fl.fl4_dst = daddr;
1861 rth->rt_dst = daddr;
1862 rth->fl.fl4_tos = tos;
1863 rth->fl.mark = skb->mark;
1864 rth->fl.fl4_src = saddr;
1865 rth->rt_src = saddr;
1866 #ifdef CONFIG_IP_ROUTE_CLASSID
1867 rth->dst.tclassid = itag;
1870 rth->fl.iif = dev->ifindex;
1871 rth->dst.dev = init_net.loopback_dev;
1872 dev_hold(rth->dst.dev);
1874 rth->rt_gateway = daddr;
1875 rth->rt_spec_dst= spec_dst;
1876 rth->rt_genid = rt_genid(dev_net(dev));
1877 rth->rt_flags = RTCF_MULTICAST;
1878 rth->rt_type = RTN_MULTICAST;
1880 rth->dst.input= ip_local_deliver;
1881 rth->rt_flags |= RTCF_LOCAL;
1884 #ifdef CONFIG_IP_MROUTE
1885 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1886 rth->dst.input = ip_mr_input;
1888 RT_CACHE_STAT_INC(in_slow_mc);
1890 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1891 return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1902 static void ip_handle_martian_source(struct net_device *dev,
1903 struct in_device *in_dev,
1904 struct sk_buff *skb,
1908 RT_CACHE_STAT_INC(in_martian_src);
1909 #ifdef CONFIG_IP_ROUTE_VERBOSE
1910 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1912 * RFC1812 recommendation, if source is martian,
1913 * the only hint is MAC header.
1915 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1916 &daddr, &saddr, dev->name);
1917 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1919 const unsigned char *p = skb_mac_header(skb);
1920 printk(KERN_WARNING "ll header: ");
1921 for (i = 0; i < dev->hard_header_len; i++, p++) {
1923 if (i < (dev->hard_header_len - 1))
1932 /* called in rcu_read_lock() section */
1933 static int __mkroute_input(struct sk_buff *skb,
1934 struct fib_result *res,
1935 struct in_device *in_dev,
1936 __be32 daddr, __be32 saddr, u32 tos,
1937 struct rtable **result)
1941 struct in_device *out_dev;
1942 unsigned int flags = 0;
1946 /* get a working reference to the output device */
1947 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1948 if (out_dev == NULL) {
1949 if (net_ratelimit())
1950 printk(KERN_CRIT "Bug in ip_route_input" \
1951 "_slow(). Please, report\n");
1956 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1957 in_dev->dev, &spec_dst, &itag, skb->mark);
1959 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1966 flags |= RTCF_DIRECTSRC;
1968 if (out_dev == in_dev && err &&
1969 (IN_DEV_SHARED_MEDIA(out_dev) ||
1970 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1971 flags |= RTCF_DOREDIRECT;
1973 if (skb->protocol != htons(ETH_P_IP)) {
1974 /* Not IP (i.e. ARP). Do not create route, if it is
1975 * invalid for proxy arp. DNAT routes are always valid.
1977 * Proxy arp feature have been extended to allow, ARP
1978 * replies back to the same interface, to support
1979 * Private VLAN switch technologies. See arp.c.
1981 if (out_dev == in_dev &&
1982 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1989 rth = dst_alloc(&ipv4_dst_ops);
1995 atomic_set(&rth->dst.__refcnt, 1);
1996 rth->dst.flags= DST_HOST;
1997 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1998 rth->dst.flags |= DST_NOPOLICY;
1999 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2000 rth->dst.flags |= DST_NOXFRM;
2001 rth->fl.fl4_dst = daddr;
2002 rth->rt_dst = daddr;
2003 rth->fl.fl4_tos = tos;
2004 rth->fl.mark = skb->mark;
2005 rth->fl.fl4_src = saddr;
2006 rth->rt_src = saddr;
2007 rth->rt_gateway = daddr;
2009 rth->fl.iif = in_dev->dev->ifindex;
2010 rth->dst.dev = (out_dev)->dev;
2011 dev_hold(rth->dst.dev);
2013 rth->rt_spec_dst= spec_dst;
2015 rth->dst.obsolete = -1;
2016 rth->dst.input = ip_forward;
2017 rth->dst.output = ip_output;
2018 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2020 rt_set_nexthop(rth, res, itag);
2022 rth->rt_flags = flags;
2030 static int ip_mkroute_input(struct sk_buff *skb,
2031 struct fib_result *res,
2032 const struct flowi *fl,
2033 struct in_device *in_dev,
2034 __be32 daddr, __be32 saddr, u32 tos)
2036 struct rtable* rth = NULL;
2040 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2041 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2042 fib_select_multipath(fl, res);
2045 /* create a routing cache entry */
2046 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2050 /* put it into the cache */
2051 hash = rt_hash(daddr, saddr, fl->iif,
2052 rt_genid(dev_net(rth->dst.dev)));
2053 return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2057 * NOTE. We drop all the packets that has local source
2058 * addresses, because every properly looped back packet
2059 * must have correct destination already attached by output routine.
2061 * Such approach solves two big problems:
2062 * 1. Not simplex devices are handled properly.
2063 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2064 * called with rcu_read_lock()
2067 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2068 u8 tos, struct net_device *dev)
2070 struct fib_result res;
2071 struct in_device *in_dev = __in_dev_get_rcu(dev);
2072 struct flowi fl = { .fl4_dst = daddr,
2075 .fl4_scope = RT_SCOPE_UNIVERSE,
2077 .iif = dev->ifindex };
2080 struct rtable * rth;
2084 struct net * net = dev_net(dev);
2086 /* IP on this device is disabled. */
2091 /* Check for the most weird martians, which can be not detected
2095 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2096 ipv4_is_loopback(saddr))
2097 goto martian_source;
2099 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2102 /* Accept zero addresses only to limited broadcast;
2103 * I even do not know to fix it or not. Waiting for complains :-)
2105 if (ipv4_is_zeronet(saddr))
2106 goto martian_source;
2108 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2109 goto martian_destination;
2112 * Now we are ready to route packet.
2114 err = fib_lookup(net, &fl, &res);
2116 if (!IN_DEV_FORWARD(in_dev))
2121 RT_CACHE_STAT_INC(in_slow_tot);
2123 if (res.type == RTN_BROADCAST)
2126 if (res.type == RTN_LOCAL) {
2127 err = fib_validate_source(saddr, daddr, tos,
2128 net->loopback_dev->ifindex,
2129 dev, &spec_dst, &itag, skb->mark);
2131 goto martian_source_keep_err;
2133 flags |= RTCF_DIRECTSRC;
2138 if (!IN_DEV_FORWARD(in_dev))
2140 if (res.type != RTN_UNICAST)
2141 goto martian_destination;
2143 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2147 if (skb->protocol != htons(ETH_P_IP))
2150 if (ipv4_is_zeronet(saddr))
2151 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2153 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2156 goto martian_source_keep_err;
2158 flags |= RTCF_DIRECTSRC;
2160 flags |= RTCF_BROADCAST;
2161 res.type = RTN_BROADCAST;
2162 RT_CACHE_STAT_INC(in_brd);
2165 rth = dst_alloc(&ipv4_dst_ops);
2169 rth->dst.output= ip_rt_bug;
2170 rth->dst.obsolete = -1;
2171 rth->rt_genid = rt_genid(net);
2173 atomic_set(&rth->dst.__refcnt, 1);
2174 rth->dst.flags= DST_HOST;
2175 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2176 rth->dst.flags |= DST_NOPOLICY;
2177 rth->fl.fl4_dst = daddr;
2178 rth->rt_dst = daddr;
2179 rth->fl.fl4_tos = tos;
2180 rth->fl.mark = skb->mark;
2181 rth->fl.fl4_src = saddr;
2182 rth->rt_src = saddr;
2183 #ifdef CONFIG_IP_ROUTE_CLASSID
2184 rth->dst.tclassid = itag;
2187 rth->fl.iif = dev->ifindex;
2188 rth->dst.dev = net->loopback_dev;
2189 dev_hold(rth->dst.dev);
2190 rth->rt_gateway = daddr;
2191 rth->rt_spec_dst= spec_dst;
2192 rth->dst.input= ip_local_deliver;
2193 rth->rt_flags = flags|RTCF_LOCAL;
2194 if (res.type == RTN_UNREACHABLE) {
2195 rth->dst.input= ip_error;
2196 rth->dst.error= -err;
2197 rth->rt_flags &= ~RTCF_LOCAL;
2199 rth->rt_type = res.type;
2200 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2201 err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2205 RT_CACHE_STAT_INC(in_no_route);
2206 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2207 res.type = RTN_UNREACHABLE;
2213 * Do not cache martian addresses: they should be logged (RFC1812)
2215 martian_destination:
2216 RT_CACHE_STAT_INC(in_martian_dst);
2217 #ifdef CONFIG_IP_ROUTE_VERBOSE
2218 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2219 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2220 &daddr, &saddr, dev->name);
2224 err = -EHOSTUNREACH;
2237 martian_source_keep_err:
2238 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2242 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2243 u8 tos, struct net_device *dev, bool noref)
2245 struct rtable * rth;
2247 int iif = dev->ifindex;
2255 if (!rt_caching(net))
2258 tos &= IPTOS_RT_MASK;
2259 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2261 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2262 rth = rcu_dereference(rth->dst.rt_next)) {
2263 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2264 ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2265 (rth->fl.iif ^ iif) |
2267 (rth->fl.fl4_tos ^ tos)) == 0 &&
2268 rth->fl.mark == skb->mark &&
2269 net_eq(dev_net(rth->dst.dev), net) &&
2270 !rt_is_expired(rth)) {
2272 dst_use_noref(&rth->dst, jiffies);
2273 skb_dst_set_noref(skb, &rth->dst);
2275 dst_use(&rth->dst, jiffies);
2276 skb_dst_set(skb, &rth->dst);
2278 RT_CACHE_STAT_INC(in_hit);
2282 RT_CACHE_STAT_INC(in_hlist_search);
2286 /* Multicast recognition logic is moved from route cache to here.
2287 The problem was that too many Ethernet cards have broken/missing
2288 hardware multicast filters :-( As result the host on multicasting
2289 network acquires a lot of useless route cache entries, sort of
2290 SDR messages from all the world. Now we try to get rid of them.
2291 Really, provided software IP multicast filter is organized
2292 reasonably (at least, hashed), it does not result in a slowdown
2293 comparing with route cache reject entries.
2294 Note, that multicast routers are not affected, because
2295 route cache entry is created eventually.
2297 if (ipv4_is_multicast(daddr)) {
2298 struct in_device *in_dev = __in_dev_get_rcu(dev);
2301 int our = ip_check_mc(in_dev, daddr, saddr,
2302 ip_hdr(skb)->protocol);
2304 #ifdef CONFIG_IP_MROUTE
2306 (!ipv4_is_local_multicast(daddr) &&
2307 IN_DEV_MFORWARD(in_dev))
2310 int res = ip_route_input_mc(skb, daddr, saddr,
2319 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2323 EXPORT_SYMBOL(ip_route_input_common);
2325 /* called with rcu_read_lock() */
2326 static struct rtable *__mkroute_output(struct fib_result *res,
2327 const struct flowi *fl,
2328 const struct flowi *oldflp,
2329 struct net_device *dev_out,
2332 u32 tos = RT_FL_TOS(oldflp);
2333 struct in_device *in_dev;
2336 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2337 return ERR_PTR(-EINVAL);
2339 if (ipv4_is_lbcast(fl->fl4_dst))
2340 res->type = RTN_BROADCAST;
2341 else if (ipv4_is_multicast(fl->fl4_dst))
2342 res->type = RTN_MULTICAST;
2343 else if (ipv4_is_zeronet(fl->fl4_dst))
2344 return ERR_PTR(-EINVAL);
2346 if (dev_out->flags & IFF_LOOPBACK)
2347 flags |= RTCF_LOCAL;
2349 in_dev = __in_dev_get_rcu(dev_out);
2351 return ERR_PTR(-EINVAL);
2353 if (res->type == RTN_BROADCAST) {
2354 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2356 } else if (res->type == RTN_MULTICAST) {
2357 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2358 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2360 flags &= ~RTCF_LOCAL;
2361 /* If multicast route do not exist use
2362 * default one, but do not gateway in this case.
2365 if (res->fi && res->prefixlen < 4)
2370 rth = dst_alloc(&ipv4_dst_ops);
2372 return ERR_PTR(-ENOBUFS);
2374 atomic_set(&rth->dst.__refcnt, 1);
2375 rth->dst.flags= DST_HOST;
2376 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2377 rth->dst.flags |= DST_NOXFRM;
2378 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2379 rth->dst.flags |= DST_NOPOLICY;
2381 rth->fl.fl4_dst = oldflp->fl4_dst;
2382 rth->fl.fl4_tos = tos;
2383 rth->fl.fl4_src = oldflp->fl4_src;
2384 rth->fl.oif = oldflp->oif;
2385 rth->fl.mark = oldflp->mark;
2386 rth->rt_dst = fl->fl4_dst;
2387 rth->rt_src = fl->fl4_src;
2388 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2389 /* get references to the devices that are to be hold by the routing
2391 rth->dst.dev = dev_out;
2393 rth->rt_gateway = fl->fl4_dst;
2394 rth->rt_spec_dst= fl->fl4_src;
2396 rth->dst.output=ip_output;
2397 rth->dst.obsolete = -1;
2398 rth->rt_genid = rt_genid(dev_net(dev_out));
2400 RT_CACHE_STAT_INC(out_slow_tot);
2402 if (flags & RTCF_LOCAL) {
2403 rth->dst.input = ip_local_deliver;
2404 rth->rt_spec_dst = fl->fl4_dst;
2406 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2407 rth->rt_spec_dst = fl->fl4_src;
2408 if (flags & RTCF_LOCAL &&
2409 !(dev_out->flags & IFF_LOOPBACK)) {
2410 rth->dst.output = ip_mc_output;
2411 RT_CACHE_STAT_INC(out_slow_mc);
2413 #ifdef CONFIG_IP_MROUTE
2414 if (res->type == RTN_MULTICAST) {
2415 if (IN_DEV_MFORWARD(in_dev) &&
2416 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2417 rth->dst.input = ip_mr_input;
2418 rth->dst.output = ip_mc_output;
2424 rt_set_nexthop(rth, res, 0);
2426 rth->rt_flags = flags;
2431 * Major route resolver routine.
2432 * called with rcu_read_lock();
2435 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2436 const struct flowi *oldflp)
2438 u32 tos = RT_FL_TOS(oldflp);
2439 struct flowi fl = { .fl4_dst = oldflp->fl4_dst,
2440 .fl4_src = oldflp->fl4_src,
2441 .fl4_tos = tos & IPTOS_RT_MASK,
2442 .fl4_scope = ((tos & RTO_ONLINK) ?
2443 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE),
2444 .mark = oldflp->mark,
2445 .iif = net->loopback_dev->ifindex,
2446 .oif = oldflp->oif };
2447 struct fib_result res;
2448 unsigned int flags = 0;
2449 struct net_device *dev_out = NULL;
2455 #ifdef CONFIG_IP_MULTIPLE_TABLES
2460 if (oldflp->fl4_src) {
2462 if (ipv4_is_multicast(oldflp->fl4_src) ||
2463 ipv4_is_lbcast(oldflp->fl4_src) ||
2464 ipv4_is_zeronet(oldflp->fl4_src))
2467 /* I removed check for oif == dev_out->oif here.
2468 It was wrong for two reasons:
2469 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2470 is assigned to multiple interfaces.
2471 2. Moreover, we are allowed to send packets with saddr
2472 of another iface. --ANK
2475 if (oldflp->oif == 0 &&
2476 (ipv4_is_multicast(oldflp->fl4_dst) ||
2477 ipv4_is_lbcast(oldflp->fl4_dst))) {
2478 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2479 dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2480 if (dev_out == NULL)
2483 /* Special hack: user can direct multicasts
2484 and limited broadcast via necessary interface
2485 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2486 This hack is not just for fun, it allows
2487 vic,vat and friends to work.
2488 They bind socket to loopback, set ttl to zero
2489 and expect that it will work.
2490 From the viewpoint of routing cache they are broken,
2491 because we are not allowed to build multicast path
2492 with loopback source addr (look, routing cache
2493 cannot know, that ttl is zero, so that packet
2494 will not leave this host and route is valid).
2495 Luckily, this hack is good workaround.
2498 fl.oif = dev_out->ifindex;
2502 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2503 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2504 if (!__ip_dev_find(net, oldflp->fl4_src, false))
2511 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2513 if (dev_out == NULL)
2516 /* RACE: Check return value of inet_select_addr instead. */
2517 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2521 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2522 ipv4_is_lbcast(oldflp->fl4_dst)) {
2524 fl.fl4_src = inet_select_addr(dev_out, 0,
2529 if (ipv4_is_multicast(oldflp->fl4_dst))
2530 fl.fl4_src = inet_select_addr(dev_out, 0,
2532 else if (!oldflp->fl4_dst)
2533 fl.fl4_src = inet_select_addr(dev_out, 0,
2539 fl.fl4_dst = fl.fl4_src;
2541 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2542 dev_out = net->loopback_dev;
2543 fl.oif = net->loopback_dev->ifindex;
2544 res.type = RTN_LOCAL;
2545 flags |= RTCF_LOCAL;
2549 if (fib_lookup(net, &fl, &res)) {
2552 /* Apparently, routing tables are wrong. Assume,
2553 that the destination is on link.
2556 Because we are allowed to send to iface
2557 even if it has NO routes and NO assigned
2558 addresses. When oif is specified, routing
2559 tables are looked up with only one purpose:
2560 to catch if destination is gatewayed, rather than
2561 direct. Moreover, if MSG_DONTROUTE is set,
2562 we send packet, ignoring both routing tables
2563 and ifaddr state. --ANK
2566 We could make it even if oif is unknown,
2567 likely IPv6, but we do not.
2570 if (fl.fl4_src == 0)
2571 fl.fl4_src = inet_select_addr(dev_out, 0,
2573 res.type = RTN_UNICAST;
2580 if (res.type == RTN_LOCAL) {
2582 if (res.fi->fib_prefsrc)
2583 fl.fl4_src = res.fi->fib_prefsrc;
2585 fl.fl4_src = fl.fl4_dst;
2587 dev_out = net->loopback_dev;
2588 fl.oif = dev_out->ifindex;
2590 flags |= RTCF_LOCAL;
2594 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2595 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2596 fib_select_multipath(&fl, &res);
2599 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2600 fib_select_default(&res);
2603 fl.fl4_src = FIB_RES_PREFSRC(res);
2605 dev_out = FIB_RES_DEV(res);
2606 fl.oif = dev_out->ifindex;
2610 rth = __mkroute_output(&res, &fl, oldflp, dev_out, flags);
2616 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2617 rt_genid(dev_net(dev_out)));
2618 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2626 int __ip_route_output_key(struct net *net, struct rtable **rp,
2627 const struct flowi *flp)
2632 if (!rt_caching(net))
2635 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2638 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2639 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2640 if (rth->fl.fl4_dst == flp->fl4_dst &&
2641 rth->fl.fl4_src == flp->fl4_src &&
2642 rt_is_output_route(rth) &&
2643 rth->fl.oif == flp->oif &&
2644 rth->fl.mark == flp->mark &&
2645 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2646 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2647 net_eq(dev_net(rth->dst.dev), net) &&
2648 !rt_is_expired(rth)) {
2649 dst_use(&rth->dst, jiffies);
2650 RT_CACHE_STAT_INC(out_hit);
2651 rcu_read_unlock_bh();
2655 RT_CACHE_STAT_INC(out_hlist_search);
2657 rcu_read_unlock_bh();
2660 return ip_route_output_slow(net, rp, flp);
2662 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2664 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2669 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2674 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2678 static struct dst_ops ipv4_dst_blackhole_ops = {
2680 .protocol = cpu_to_be16(ETH_P_IP),
2681 .destroy = ipv4_dst_destroy,
2682 .check = ipv4_blackhole_dst_check,
2683 .default_mtu = ipv4_blackhole_default_mtu,
2684 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2688 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2690 struct rtable *ort = *rp;
2691 struct rtable *rt = (struct rtable *)
2692 dst_alloc(&ipv4_dst_blackhole_ops);
2695 struct dst_entry *new = &rt->dst;
2697 atomic_set(&new->__refcnt, 1);
2699 new->input = dst_discard;
2700 new->output = dst_discard;
2701 dst_copy_metrics(new, &ort->dst);
2703 new->dev = ort->dst.dev;
2709 rt->rt_genid = rt_genid(net);
2710 rt->rt_flags = ort->rt_flags;
2711 rt->rt_type = ort->rt_type;
2712 rt->rt_dst = ort->rt_dst;
2713 rt->rt_src = ort->rt_src;
2714 rt->rt_iif = ort->rt_iif;
2715 rt->rt_gateway = ort->rt_gateway;
2716 rt->rt_spec_dst = ort->rt_spec_dst;
2717 rt->peer = ort->peer;
2719 atomic_inc(&rt->peer->refcnt);
2722 atomic_inc(&rt->fi->fib_clntref);
2727 dst_release(&(*rp)->dst);
2729 return rt ? 0 : -ENOMEM;
2732 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2733 struct sock *sk, int flags)
2737 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2742 flp->fl4_src = (*rp)->rt_src;
2744 flp->fl4_dst = (*rp)->rt_dst;
2745 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2746 flags ? XFRM_LOOKUP_WAIT : 0);
2747 if (err == -EREMOTE)
2748 err = ipv4_dst_blackhole(net, rp, flp);
2755 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2757 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2759 return ip_route_output_flow(net, rp, flp, NULL, 0);
2761 EXPORT_SYMBOL(ip_route_output_key);
2763 static int rt_fill_info(struct net *net,
2764 struct sk_buff *skb, u32 pid, u32 seq, int event,
2765 int nowait, unsigned int flags)
2767 struct rtable *rt = skb_rtable(skb);
2769 struct nlmsghdr *nlh;
2771 u32 id = 0, ts = 0, tsage = 0, error;
2773 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2777 r = nlmsg_data(nlh);
2778 r->rtm_family = AF_INET;
2779 r->rtm_dst_len = 32;
2781 r->rtm_tos = rt->fl.fl4_tos;
2782 r->rtm_table = RT_TABLE_MAIN;
2783 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2784 r->rtm_type = rt->rt_type;
2785 r->rtm_scope = RT_SCOPE_UNIVERSE;
2786 r->rtm_protocol = RTPROT_UNSPEC;
2787 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2788 if (rt->rt_flags & RTCF_NOTIFY)
2789 r->rtm_flags |= RTM_F_NOTIFY;
2791 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2793 if (rt->fl.fl4_src) {
2794 r->rtm_src_len = 32;
2795 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2798 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2799 #ifdef CONFIG_IP_ROUTE_CLASSID
2800 if (rt->dst.tclassid)
2801 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2803 if (rt_is_input_route(rt))
2804 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2805 else if (rt->rt_src != rt->fl.fl4_src)
2806 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2808 if (rt->rt_dst != rt->rt_gateway)
2809 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2811 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2812 goto nla_put_failure;
2815 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2817 error = rt->dst.error;
2818 expires = (rt->peer && rt->peer->pmtu_expires) ?
2819 rt->peer->pmtu_expires - jiffies : 0;
2821 inet_peer_refcheck(rt->peer);
2822 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2823 if (rt->peer->tcp_ts_stamp) {
2824 ts = rt->peer->tcp_ts;
2825 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2829 if (rt_is_input_route(rt)) {
2830 #ifdef CONFIG_IP_MROUTE
2831 __be32 dst = rt->rt_dst;
2833 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2834 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2835 int err = ipmr_get_route(net, skb, r, nowait);
2840 goto nla_put_failure;
2842 if (err == -EMSGSIZE)
2843 goto nla_put_failure;
2849 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2852 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2853 expires, error) < 0)
2854 goto nla_put_failure;
2856 return nlmsg_end(skb, nlh);
2859 nlmsg_cancel(skb, nlh);
2863 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2865 struct net *net = sock_net(in_skb->sk);
2867 struct nlattr *tb[RTA_MAX+1];
2868 struct rtable *rt = NULL;
2874 struct sk_buff *skb;
2876 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2880 rtm = nlmsg_data(nlh);
2882 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2888 /* Reserve room for dummy headers, this skb can pass
2889 through good chunk of routing engine.
2891 skb_reset_mac_header(skb);
2892 skb_reset_network_header(skb);
2894 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2895 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2896 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2898 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2899 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2900 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2901 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2904 struct net_device *dev;
2906 dev = __dev_get_by_index(net, iif);
2912 skb->protocol = htons(ETH_P_IP);
2916 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2919 rt = skb_rtable(skb);
2920 if (err == 0 && rt->dst.error)
2921 err = -rt->dst.error;
2926 .fl4_tos = rtm->rtm_tos,
2927 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2930 err = ip_route_output_key(net, &rt, &fl);
2936 skb_dst_set(skb, &rt->dst);
2937 if (rtm->rtm_flags & RTM_F_NOTIFY)
2938 rt->rt_flags |= RTCF_NOTIFY;
2940 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2941 RTM_NEWROUTE, 0, 0);
2945 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2954 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2961 net = sock_net(skb->sk);
2966 s_idx = idx = cb->args[1];
2967 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2968 if (!rt_hash_table[h].chain)
2971 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2972 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2973 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2975 if (rt_is_expired(rt))
2977 skb_dst_set_noref(skb, &rt->dst);
2978 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2979 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2980 1, NLM_F_MULTI) <= 0) {
2982 rcu_read_unlock_bh();
2987 rcu_read_unlock_bh();
2996 void ip_rt_multicast_event(struct in_device *in_dev)
2998 rt_cache_flush(dev_net(in_dev->dev), 0);
3001 #ifdef CONFIG_SYSCTL
3002 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3003 void __user *buffer,
3004 size_t *lenp, loff_t *ppos)
3011 memcpy(&ctl, __ctl, sizeof(ctl));
3012 ctl.data = &flush_delay;
3013 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3015 net = (struct net *)__ctl->extra1;
3016 rt_cache_flush(net, flush_delay);
3023 static ctl_table ipv4_route_table[] = {
3025 .procname = "gc_thresh",
3026 .data = &ipv4_dst_ops.gc_thresh,
3027 .maxlen = sizeof(int),
3029 .proc_handler = proc_dointvec,
3032 .procname = "max_size",
3033 .data = &ip_rt_max_size,
3034 .maxlen = sizeof(int),
3036 .proc_handler = proc_dointvec,
3039 /* Deprecated. Use gc_min_interval_ms */
3041 .procname = "gc_min_interval",
3042 .data = &ip_rt_gc_min_interval,
3043 .maxlen = sizeof(int),
3045 .proc_handler = proc_dointvec_jiffies,
3048 .procname = "gc_min_interval_ms",
3049 .data = &ip_rt_gc_min_interval,
3050 .maxlen = sizeof(int),
3052 .proc_handler = proc_dointvec_ms_jiffies,
3055 .procname = "gc_timeout",
3056 .data = &ip_rt_gc_timeout,
3057 .maxlen = sizeof(int),
3059 .proc_handler = proc_dointvec_jiffies,
3062 .procname = "gc_interval",
3063 .data = &ip_rt_gc_interval,
3064 .maxlen = sizeof(int),
3066 .proc_handler = proc_dointvec_jiffies,
3069 .procname = "redirect_load",
3070 .data = &ip_rt_redirect_load,
3071 .maxlen = sizeof(int),
3073 .proc_handler = proc_dointvec,
3076 .procname = "redirect_number",
3077 .data = &ip_rt_redirect_number,
3078 .maxlen = sizeof(int),
3080 .proc_handler = proc_dointvec,
3083 .procname = "redirect_silence",
3084 .data = &ip_rt_redirect_silence,
3085 .maxlen = sizeof(int),
3087 .proc_handler = proc_dointvec,
3090 .procname = "error_cost",
3091 .data = &ip_rt_error_cost,
3092 .maxlen = sizeof(int),
3094 .proc_handler = proc_dointvec,
3097 .procname = "error_burst",
3098 .data = &ip_rt_error_burst,
3099 .maxlen = sizeof(int),
3101 .proc_handler = proc_dointvec,
3104 .procname = "gc_elasticity",
3105 .data = &ip_rt_gc_elasticity,
3106 .maxlen = sizeof(int),
3108 .proc_handler = proc_dointvec,
3111 .procname = "mtu_expires",
3112 .data = &ip_rt_mtu_expires,
3113 .maxlen = sizeof(int),
3115 .proc_handler = proc_dointvec_jiffies,
3118 .procname = "min_pmtu",
3119 .data = &ip_rt_min_pmtu,
3120 .maxlen = sizeof(int),
3122 .proc_handler = proc_dointvec,
3125 .procname = "min_adv_mss",
3126 .data = &ip_rt_min_advmss,
3127 .maxlen = sizeof(int),
3129 .proc_handler = proc_dointvec,
3134 static struct ctl_table empty[1];
3136 static struct ctl_table ipv4_skeleton[] =
3138 { .procname = "route",
3139 .mode = 0555, .child = ipv4_route_table},
3140 { .procname = "neigh",
3141 .mode = 0555, .child = empty},
3145 static __net_initdata struct ctl_path ipv4_path[] = {
3146 { .procname = "net", },
3147 { .procname = "ipv4", },
3151 static struct ctl_table ipv4_route_flush_table[] = {
3153 .procname = "flush",
3154 .maxlen = sizeof(int),
3156 .proc_handler = ipv4_sysctl_rtcache_flush,
3161 static __net_initdata struct ctl_path ipv4_route_path[] = {
3162 { .procname = "net", },
3163 { .procname = "ipv4", },
3164 { .procname = "route", },
3168 static __net_init int sysctl_route_net_init(struct net *net)
3170 struct ctl_table *tbl;
3172 tbl = ipv4_route_flush_table;
3173 if (!net_eq(net, &init_net)) {
3174 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3178 tbl[0].extra1 = net;
3180 net->ipv4.route_hdr =
3181 register_net_sysctl_table(net, ipv4_route_path, tbl);
3182 if (net->ipv4.route_hdr == NULL)
3187 if (tbl != ipv4_route_flush_table)
3193 static __net_exit void sysctl_route_net_exit(struct net *net)
3195 struct ctl_table *tbl;
3197 tbl = net->ipv4.route_hdr->ctl_table_arg;
3198 unregister_net_sysctl_table(net->ipv4.route_hdr);
3199 BUG_ON(tbl == ipv4_route_flush_table);
3203 static __net_initdata struct pernet_operations sysctl_route_ops = {
3204 .init = sysctl_route_net_init,
3205 .exit = sysctl_route_net_exit,
3209 static __net_init int rt_genid_init(struct net *net)
3211 get_random_bytes(&net->ipv4.rt_genid,
3212 sizeof(net->ipv4.rt_genid));
3216 static __net_initdata struct pernet_operations rt_genid_ops = {
3217 .init = rt_genid_init,
3221 #ifdef CONFIG_IP_ROUTE_CLASSID
3222 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3223 #endif /* CONFIG_IP_ROUTE_CLASSID */
3225 static __initdata unsigned long rhash_entries;
3226 static int __init set_rhash_entries(char *str)
3230 rhash_entries = simple_strtoul(str, &str, 0);
3233 __setup("rhash_entries=", set_rhash_entries);
3235 int __init ip_rt_init(void)
3239 #ifdef CONFIG_IP_ROUTE_CLASSID
3240 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3242 panic("IP: failed to allocate ip_rt_acct\n");
3245 ipv4_dst_ops.kmem_cachep =
3246 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3247 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3249 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3251 if (dst_entries_init(&ipv4_dst_ops) < 0)
3252 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3254 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3255 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3257 rt_hash_table = (struct rt_hash_bucket *)
3258 alloc_large_system_hash("IP route cache",
3259 sizeof(struct rt_hash_bucket),
3261 (totalram_pages >= 128 * 1024) ?
3266 rhash_entries ? 0 : 512 * 1024);
3267 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3268 rt_hash_lock_init();
3270 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3271 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3276 if (ip_rt_proc_init())
3277 printk(KERN_ERR "Unable to create route proc files\n");
3280 xfrm4_init(ip_rt_max_size);
3282 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3284 #ifdef CONFIG_SYSCTL
3285 register_pernet_subsys(&sysctl_route_ops);
3287 register_pernet_subsys(&rt_genid_ops);
3291 #ifdef CONFIG_SYSCTL
3293 * We really need to sanitize the damn ipv4 init order, then all
3294 * this nonsense will go away.
3296 void __init ip_static_sysctl_init(void)
3298 register_sysctl_paths(ipv4_path, ipv4_skeleton);