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.
8 * Version: $Id: route.c,v 1.103 2002/01/12 07:44:09 davem Exp $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
14 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
17 * Alan Cox : Verify area fixes.
18 * Alan Cox : cli() protects routing changes
19 * Rui Oliveira : ICMP routing table updates
20 * (rco@di.uminho.pt) Routing table insertion and update
21 * Linus Torvalds : Rewrote bits to be sensible
22 * Alan Cox : Added BSD route gw semantics
23 * Alan Cox : Super /proc >4K
24 * Alan Cox : MTU in route table
25 * Alan Cox : MSS actually. Also added the window
27 * Sam Lantinga : Fixed route matching in rt_del()
28 * Alan Cox : Routing cache support.
29 * Alan Cox : Removed compatibility cruft.
30 * Alan Cox : RTF_REJECT support.
31 * Alan Cox : TCP irtt support.
32 * Jonathan Naylor : Added Metric support.
33 * Miquel van Smoorenburg : BSD API fixes.
34 * Miquel van Smoorenburg : Metrics.
35 * Alan Cox : Use __u32 properly
36 * Alan Cox : Aligned routing errors more closely with BSD
37 * our system is still very different.
38 * Alan Cox : Faster /proc handling
39 * Alexey Kuznetsov : Massive rework to support tree based routing,
40 * routing caches and better behaviour.
42 * Olaf Erb : irtt wasn't being copied right.
43 * Bjorn Ekwall : Kerneld route support.
44 * Alan Cox : Multicast fixed (I hope)
45 * Pavel Krauz : Limited broadcast fixed
46 * Mike McLagan : Routing by source
47 * Alexey Kuznetsov : End of old history. Split to fib.c and
48 * route.c and rewritten from scratch.
49 * Andi Kleen : Load-limit warning messages.
50 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
51 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
52 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
53 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
54 * Marc Boucher : routing by fwmark
55 * Robert Olsson : Added rt_cache statistics
56 * Arnaldo C. Melo : Convert proc stuff to seq_file
57 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
58 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
59 * Ilia Sotnikov : Removed TOS from hash calculations
61 * This program is free software; you can redistribute it and/or
62 * modify it under the terms of the GNU General Public License
63 * as published by the Free Software Foundation; either version
64 * 2 of the License, or (at your option) any later version.
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <linux/bitops.h>
71 #include <linux/types.h>
72 #include <linux/kernel.h>
74 #include <linux/bootmem.h>
75 #include <linux/string.h>
76 #include <linux/socket.h>
77 #include <linux/sockios.h>
78 #include <linux/errno.h>
80 #include <linux/inet.h>
81 #include <linux/netdevice.h>
82 #include <linux/proc_fs.h>
83 #include <linux/init.h>
84 #include <linux/workqueue.h>
85 #include <linux/skbuff.h>
86 #include <linux/inetdevice.h>
87 #include <linux/igmp.h>
88 #include <linux/pkt_sched.h>
89 #include <linux/mroute.h>
90 #include <linux/netfilter_ipv4.h>
91 #include <linux/random.h>
92 #include <linux/jhash.h>
93 #include <linux/rcupdate.h>
94 #include <linux/times.h>
96 #include <net/net_namespace.h>
97 #include <net/protocol.h>
99 #include <net/route.h>
100 #include <net/inetpeer.h>
101 #include <net/sock.h>
102 #include <net/ip_fib.h>
105 #include <net/icmp.h>
106 #include <net/xfrm.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
113 #define RT_FL_TOS(oldflp) \
114 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
116 #define IP_MAX_MTU 0xFFF0
118 #define RT_GC_TIMEOUT (300*HZ)
120 static int ip_rt_max_size;
121 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
122 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
123 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
124 static int ip_rt_redirect_number __read_mostly = 9;
125 static int ip_rt_redirect_load __read_mostly = HZ / 50;
126 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly = HZ;
128 static int ip_rt_error_burst __read_mostly = 5 * HZ;
129 static int ip_rt_gc_elasticity __read_mostly = 8;
130 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
131 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
132 static int ip_rt_min_advmss __read_mostly = 256;
133 static int ip_rt_secret_interval __read_mostly = 10 * 60 * HZ;
135 static void rt_worker_func(struct work_struct *work);
136 static DECLARE_DELAYED_WORK(expires_work, rt_worker_func);
137 static struct timer_list rt_secret_timer;
140 * Interface to generic destination cache.
143 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
144 static void ipv4_dst_destroy(struct dst_entry *dst);
145 static void ipv4_dst_ifdown(struct dst_entry *dst,
146 struct net_device *dev, int how);
147 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
148 static void ipv4_link_failure(struct sk_buff *skb);
149 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
150 static int rt_garbage_collect(struct dst_ops *ops);
153 static struct dst_ops ipv4_dst_ops = {
155 .protocol = __constant_htons(ETH_P_IP),
156 .gc = rt_garbage_collect,
157 .check = ipv4_dst_check,
158 .destroy = ipv4_dst_destroy,
159 .ifdown = ipv4_dst_ifdown,
160 .negative_advice = ipv4_negative_advice,
161 .link_failure = ipv4_link_failure,
162 .update_pmtu = ip_rt_update_pmtu,
163 .local_out = __ip_local_out,
164 .entry_size = sizeof(struct rtable),
165 .entries = ATOMIC_INIT(0),
168 #define ECN_OR_COST(class) TC_PRIO_##class
170 const __u8 ip_tos2prio[16] = {
174 ECN_OR_COST(BESTEFFORT),
180 ECN_OR_COST(INTERACTIVE),
182 ECN_OR_COST(INTERACTIVE),
183 TC_PRIO_INTERACTIVE_BULK,
184 ECN_OR_COST(INTERACTIVE_BULK),
185 TC_PRIO_INTERACTIVE_BULK,
186 ECN_OR_COST(INTERACTIVE_BULK)
194 /* The locking scheme is rather straight forward:
196 * 1) Read-Copy Update protects the buckets of the central route hash.
197 * 2) Only writers remove entries, and they hold the lock
198 * as they look at rtable reference counts.
199 * 3) Only readers acquire references to rtable entries,
200 * they do so with atomic increments and with the
204 struct rt_hash_bucket {
205 struct rtable *chain;
207 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
208 defined(CONFIG_PROVE_LOCKING)
210 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
211 * The size of this table is a power of two and depends on the number of CPUS.
212 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
214 #ifdef CONFIG_LOCKDEP
215 # define RT_HASH_LOCK_SZ 256
218 # define RT_HASH_LOCK_SZ 4096
220 # define RT_HASH_LOCK_SZ 2048
222 # define RT_HASH_LOCK_SZ 1024
224 # define RT_HASH_LOCK_SZ 512
226 # define RT_HASH_LOCK_SZ 256
230 static spinlock_t *rt_hash_locks;
231 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
233 static __init void rt_hash_lock_init(void)
237 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
240 panic("IP: failed to allocate rt_hash_locks\n");
242 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
243 spin_lock_init(&rt_hash_locks[i]);
246 # define rt_hash_lock_addr(slot) NULL
248 static inline void rt_hash_lock_init(void)
253 static struct rt_hash_bucket *rt_hash_table __read_mostly;
254 static unsigned rt_hash_mask __read_mostly;
255 static unsigned int rt_hash_log __read_mostly;
256 static atomic_t rt_genid __read_mostly;
258 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
259 #define RT_CACHE_STAT_INC(field) \
260 (__raw_get_cpu_var(rt_cache_stat).field++)
262 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx)
264 return jhash_3words((__force u32)(__be32)(daddr),
265 (__force u32)(__be32)(saddr),
266 idx, atomic_read(&rt_genid))
270 #ifdef CONFIG_PROC_FS
271 struct rt_cache_iter_state {
272 struct seq_net_private p;
277 static struct rtable *rt_cache_get_first(struct seq_file *seq)
279 struct rt_cache_iter_state *st = seq->private;
280 struct rtable *r = NULL;
282 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
284 r = rcu_dereference(rt_hash_table[st->bucket].chain);
286 if (dev_net(r->u.dst.dev) == seq_file_net(seq) &&
287 r->rt_genid == st->genid)
289 r = rcu_dereference(r->u.dst.rt_next);
291 rcu_read_unlock_bh();
296 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
299 struct rt_cache_iter_state *st = seq->private;
300 r = r->u.dst.rt_next;
302 rcu_read_unlock_bh();
303 if (--st->bucket < 0)
306 r = rt_hash_table[st->bucket].chain;
308 return rcu_dereference(r);
311 static struct rtable *rt_cache_get_next(struct seq_file *seq,
314 struct rt_cache_iter_state *st = seq->private;
315 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
316 if (dev_net(r->u.dst.dev) != seq_file_net(seq))
318 if (r->rt_genid == st->genid)
324 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
326 struct rtable *r = rt_cache_get_first(seq);
329 while (pos && (r = rt_cache_get_next(seq, r)))
331 return pos ? NULL : r;
334 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
336 struct rt_cache_iter_state *st = seq->private;
338 return rt_cache_get_idx(seq, *pos - 1);
339 st->genid = atomic_read(&rt_genid);
340 return SEQ_START_TOKEN;
343 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
347 if (v == SEQ_START_TOKEN)
348 r = rt_cache_get_first(seq);
350 r = rt_cache_get_next(seq, v);
355 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
357 if (v && v != SEQ_START_TOKEN)
358 rcu_read_unlock_bh();
361 static int rt_cache_seq_show(struct seq_file *seq, void *v)
363 if (v == SEQ_START_TOKEN)
364 seq_printf(seq, "%-127s\n",
365 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
366 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
369 struct rtable *r = v;
372 seq_printf(seq, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
373 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
374 r->u.dst.dev ? r->u.dst.dev->name : "*",
375 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
376 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
377 r->u.dst.__use, 0, (unsigned long)r->rt_src,
378 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
379 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
380 dst_metric(&r->u.dst, RTAX_WINDOW),
381 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
382 dst_metric(&r->u.dst, RTAX_RTTVAR)),
384 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
385 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
387 r->rt_spec_dst, &len);
389 seq_printf(seq, "%*s\n", 127 - len, "");
394 static const struct seq_operations rt_cache_seq_ops = {
395 .start = rt_cache_seq_start,
396 .next = rt_cache_seq_next,
397 .stop = rt_cache_seq_stop,
398 .show = rt_cache_seq_show,
401 static int rt_cache_seq_open(struct inode *inode, struct file *file)
403 return seq_open_net(inode, file, &rt_cache_seq_ops,
404 sizeof(struct rt_cache_iter_state));
407 static const struct file_operations rt_cache_seq_fops = {
408 .owner = THIS_MODULE,
409 .open = rt_cache_seq_open,
412 .release = seq_release_net,
416 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
421 return SEQ_START_TOKEN;
423 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
424 if (!cpu_possible(cpu))
427 return &per_cpu(rt_cache_stat, cpu);
432 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
436 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
437 if (!cpu_possible(cpu))
440 return &per_cpu(rt_cache_stat, cpu);
446 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
451 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
453 struct rt_cache_stat *st = v;
455 if (v == SEQ_START_TOKEN) {
456 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");
460 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
461 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
462 atomic_read(&ipv4_dst_ops.entries),
485 static const struct seq_operations rt_cpu_seq_ops = {
486 .start = rt_cpu_seq_start,
487 .next = rt_cpu_seq_next,
488 .stop = rt_cpu_seq_stop,
489 .show = rt_cpu_seq_show,
493 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
495 return seq_open(file, &rt_cpu_seq_ops);
498 static const struct file_operations rt_cpu_seq_fops = {
499 .owner = THIS_MODULE,
500 .open = rt_cpu_seq_open,
503 .release = seq_release,
506 #ifdef CONFIG_NET_CLS_ROUTE
507 static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
508 int length, int *eof, void *data)
512 if ((offset & 3) || (length & 3))
515 if (offset >= sizeof(struct ip_rt_acct) * 256) {
520 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
521 length = sizeof(struct ip_rt_acct) * 256 - offset;
525 offset /= sizeof(u32);
528 u32 *dst = (u32 *) buffer;
531 memset(dst, 0, length);
533 for_each_possible_cpu(i) {
537 src = ((u32 *) per_cpu_ptr(ip_rt_acct, i)) + offset;
538 for (j = 0; j < length/4; j++)
546 static int __net_init ip_rt_do_proc_init(struct net *net)
548 struct proc_dir_entry *pde;
550 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
555 pde = proc_create("rt_cache", S_IRUGO,
556 net->proc_net_stat, &rt_cpu_seq_fops);
560 #ifdef CONFIG_NET_CLS_ROUTE
561 pde = create_proc_read_entry("rt_acct", 0, net->proc_net,
562 ip_rt_acct_read, NULL);
568 #ifdef CONFIG_NET_CLS_ROUTE
570 remove_proc_entry("rt_cache", net->proc_net_stat);
573 remove_proc_entry("rt_cache", net->proc_net);
578 static void __net_exit ip_rt_do_proc_exit(struct net *net)
580 remove_proc_entry("rt_cache", net->proc_net_stat);
581 remove_proc_entry("rt_cache", net->proc_net);
582 remove_proc_entry("rt_acct", net->proc_net);
585 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
586 .init = ip_rt_do_proc_init,
587 .exit = ip_rt_do_proc_exit,
590 static int __init ip_rt_proc_init(void)
592 return register_pernet_subsys(&ip_rt_proc_ops);
596 static inline int ip_rt_proc_init(void)
600 #endif /* CONFIG_PROC_FS */
602 static inline void rt_free(struct rtable *rt)
604 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
607 static inline void rt_drop(struct rtable *rt)
610 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
613 static inline int rt_fast_clean(struct rtable *rth)
615 /* Kill broadcast/multicast entries very aggresively, if they
616 collide in hash table with more useful entries */
617 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
618 rth->fl.iif && rth->u.dst.rt_next;
621 static inline int rt_valuable(struct rtable *rth)
623 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
627 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
632 if (atomic_read(&rth->u.dst.__refcnt))
636 if (rth->u.dst.expires &&
637 time_after_eq(jiffies, rth->u.dst.expires))
640 age = jiffies - rth->u.dst.lastuse;
642 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
643 (age <= tmo2 && rt_valuable(rth)))
649 /* Bits of score are:
651 * 30: not quite useless
652 * 29..0: usage counter
654 static inline u32 rt_score(struct rtable *rt)
656 u32 score = jiffies - rt->u.dst.lastuse;
658 score = ~score & ~(3<<30);
664 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
670 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
672 return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
673 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
674 (fl1->mark ^ fl2->mark) |
675 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
676 *(u16 *)&fl2->nl_u.ip4_u.tos) |
677 (fl1->oif ^ fl2->oif) |
678 (fl1->iif ^ fl2->iif)) == 0;
681 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
683 return dev_net(rt1->u.dst.dev) == dev_net(rt2->u.dst.dev);
687 * Perform a full scan of hash table and free all entries.
688 * Can be called by a softirq or a process.
689 * In the later case, we want to be reschedule if necessary
691 static void rt_do_flush(int process_context)
694 struct rtable *rth, *next;
696 for (i = 0; i <= rt_hash_mask; i++) {
697 if (process_context && need_resched())
699 rth = rt_hash_table[i].chain;
703 spin_lock_bh(rt_hash_lock_addr(i));
704 rth = rt_hash_table[i].chain;
705 rt_hash_table[i].chain = NULL;
706 spin_unlock_bh(rt_hash_lock_addr(i));
708 for (; rth; rth = next) {
709 next = rth->u.dst.rt_next;
715 static void rt_check_expire(void)
717 static unsigned int rover;
718 unsigned int i = rover, goal;
719 struct rtable *rth, **rthp;
722 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
723 if (ip_rt_gc_timeout > 1)
724 do_div(mult, ip_rt_gc_timeout);
725 goal = (unsigned int)mult;
726 if (goal > rt_hash_mask)
727 goal = rt_hash_mask + 1;
728 for (; goal > 0; goal--) {
729 unsigned long tmo = ip_rt_gc_timeout;
731 i = (i + 1) & rt_hash_mask;
732 rthp = &rt_hash_table[i].chain;
739 spin_lock_bh(rt_hash_lock_addr(i));
740 while ((rth = *rthp) != NULL) {
741 if (rth->rt_genid != atomic_read(&rt_genid)) {
742 *rthp = rth->u.dst.rt_next;
746 if (rth->u.dst.expires) {
747 /* Entry is expired even if it is in use */
748 if (time_before_eq(jiffies, rth->u.dst.expires)) {
750 rthp = &rth->u.dst.rt_next;
753 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
755 rthp = &rth->u.dst.rt_next;
759 /* Cleanup aged off entries. */
760 *rthp = rth->u.dst.rt_next;
763 spin_unlock_bh(rt_hash_lock_addr(i));
769 * rt_worker_func() is run in process context.
770 * we call rt_check_expire() to scan part of the hash table
772 static void rt_worker_func(struct work_struct *work)
775 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
779 * Pertubation of rt_genid by a small quantity [1..256]
780 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
781 * many times (2^24) without giving recent rt_genid.
782 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
784 static void rt_cache_invalidate(void)
786 unsigned char shuffle;
788 get_random_bytes(&shuffle, sizeof(shuffle));
789 atomic_add(shuffle + 1U, &rt_genid);
793 * delay < 0 : invalidate cache (fast : entries will be deleted later)
794 * delay >= 0 : invalidate & flush cache (can be long)
796 void rt_cache_flush(int delay)
798 rt_cache_invalidate();
800 rt_do_flush(!in_softirq());
804 * We change rt_genid and let gc do the cleanup
806 static void rt_secret_rebuild(unsigned long dummy)
808 rt_cache_invalidate();
809 mod_timer(&rt_secret_timer, jiffies + ip_rt_secret_interval);
813 Short description of GC goals.
815 We want to build algorithm, which will keep routing cache
816 at some equilibrium point, when number of aged off entries
817 is kept approximately equal to newly generated ones.
819 Current expiration strength is variable "expire".
820 We try to adjust it dynamically, so that if networking
821 is idle expires is large enough to keep enough of warm entries,
822 and when load increases it reduces to limit cache size.
825 static int rt_garbage_collect(struct dst_ops *ops)
827 static unsigned long expire = RT_GC_TIMEOUT;
828 static unsigned long last_gc;
830 static int equilibrium;
831 struct rtable *rth, **rthp;
832 unsigned long now = jiffies;
836 * Garbage collection is pretty expensive,
837 * do not make it too frequently.
840 RT_CACHE_STAT_INC(gc_total);
842 if (now - last_gc < ip_rt_gc_min_interval &&
843 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
844 RT_CACHE_STAT_INC(gc_ignored);
848 /* Calculate number of entries, which we want to expire now. */
849 goal = atomic_read(&ipv4_dst_ops.entries) -
850 (ip_rt_gc_elasticity << rt_hash_log);
852 if (equilibrium < ipv4_dst_ops.gc_thresh)
853 equilibrium = ipv4_dst_ops.gc_thresh;
854 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
856 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
857 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
860 /* We are in dangerous area. Try to reduce cache really
863 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
864 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
867 if (now - last_gc >= ip_rt_gc_min_interval)
878 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
879 unsigned long tmo = expire;
881 k = (k + 1) & rt_hash_mask;
882 rthp = &rt_hash_table[k].chain;
883 spin_lock_bh(rt_hash_lock_addr(k));
884 while ((rth = *rthp) != NULL) {
885 if (rth->rt_genid == atomic_read(&rt_genid) &&
886 !rt_may_expire(rth, tmo, expire)) {
888 rthp = &rth->u.dst.rt_next;
891 *rthp = rth->u.dst.rt_next;
895 spin_unlock_bh(rt_hash_lock_addr(k));
904 /* Goal is not achieved. We stop process if:
906 - if expire reduced to zero. Otherwise, expire is halfed.
907 - if table is not full.
908 - if we are called from interrupt.
909 - jiffies check is just fallback/debug loop breaker.
910 We will not spin here for long time in any case.
913 RT_CACHE_STAT_INC(gc_goal_miss);
919 #if RT_CACHE_DEBUG >= 2
920 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
921 atomic_read(&ipv4_dst_ops.entries), goal, i);
924 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
926 } while (!in_softirq() && time_before_eq(jiffies, now));
928 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
931 printk(KERN_WARNING "dst cache overflow\n");
932 RT_CACHE_STAT_INC(gc_dst_overflow);
936 expire += ip_rt_gc_min_interval;
937 if (expire > ip_rt_gc_timeout ||
938 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
939 expire = ip_rt_gc_timeout;
940 #if RT_CACHE_DEBUG >= 2
941 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
942 atomic_read(&ipv4_dst_ops.entries), goal, rover);
947 static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
949 struct rtable *rth, **rthp;
951 struct rtable *cand, **candp;
954 int attempts = !in_softirq();
963 rthp = &rt_hash_table[hash].chain;
965 spin_lock_bh(rt_hash_lock_addr(hash));
966 while ((rth = *rthp) != NULL) {
967 if (rth->rt_genid != atomic_read(&rt_genid)) {
968 *rthp = rth->u.dst.rt_next;
972 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
974 *rthp = rth->u.dst.rt_next;
976 * Since lookup is lockfree, the deletion
977 * must be visible to another weakly ordered CPU before
978 * the insertion at the start of the hash chain.
980 rcu_assign_pointer(rth->u.dst.rt_next,
981 rt_hash_table[hash].chain);
983 * Since lookup is lockfree, the update writes
984 * must be ordered for consistency on SMP.
986 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
988 dst_use(&rth->u.dst, now);
989 spin_unlock_bh(rt_hash_lock_addr(hash));
996 if (!atomic_read(&rth->u.dst.__refcnt)) {
997 u32 score = rt_score(rth);
999 if (score <= min_score) {
1008 rthp = &rth->u.dst.rt_next;
1012 /* ip_rt_gc_elasticity used to be average length of chain
1013 * length, when exceeded gc becomes really aggressive.
1015 * The second limit is less certain. At the moment it allows
1016 * only 2 entries per bucket. We will see.
1018 if (chain_length > ip_rt_gc_elasticity) {
1019 *candp = cand->u.dst.rt_next;
1024 /* Try to bind route to arp only if it is output
1025 route or unicast forwarding path.
1027 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1028 int err = arp_bind_neighbour(&rt->u.dst);
1030 spin_unlock_bh(rt_hash_lock_addr(hash));
1032 if (err != -ENOBUFS) {
1037 /* Neighbour tables are full and nothing
1038 can be released. Try to shrink route cache,
1039 it is most likely it holds some neighbour records.
1041 if (attempts-- > 0) {
1042 int saved_elasticity = ip_rt_gc_elasticity;
1043 int saved_int = ip_rt_gc_min_interval;
1044 ip_rt_gc_elasticity = 1;
1045 ip_rt_gc_min_interval = 0;
1046 rt_garbage_collect(&ipv4_dst_ops);
1047 ip_rt_gc_min_interval = saved_int;
1048 ip_rt_gc_elasticity = saved_elasticity;
1052 if (net_ratelimit())
1053 printk(KERN_WARNING "Neighbour table overflow.\n");
1059 rt->u.dst.rt_next = rt_hash_table[hash].chain;
1060 #if RT_CACHE_DEBUG >= 2
1061 if (rt->u.dst.rt_next) {
1063 printk(KERN_DEBUG "rt_cache @%02x: " NIPQUAD_FMT, hash,
1064 NIPQUAD(rt->rt_dst));
1065 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1066 printk(" . " NIPQUAD_FMT, NIPQUAD(trt->rt_dst));
1070 rt_hash_table[hash].chain = rt;
1071 spin_unlock_bh(rt_hash_lock_addr(hash));
1076 void rt_bind_peer(struct rtable *rt, int create)
1078 static DEFINE_SPINLOCK(rt_peer_lock);
1079 struct inet_peer *peer;
1081 peer = inet_getpeer(rt->rt_dst, create);
1083 spin_lock_bh(&rt_peer_lock);
1084 if (rt->peer == NULL) {
1088 spin_unlock_bh(&rt_peer_lock);
1094 * Peer allocation may fail only in serious out-of-memory conditions. However
1095 * we still can generate some output.
1096 * Random ID selection looks a bit dangerous because we have no chances to
1097 * select ID being unique in a reasonable period of time.
1098 * But broken packet identifier may be better than no packet at all.
1100 static void ip_select_fb_ident(struct iphdr *iph)
1102 static DEFINE_SPINLOCK(ip_fb_id_lock);
1103 static u32 ip_fallback_id;
1106 spin_lock_bh(&ip_fb_id_lock);
1107 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1108 iph->id = htons(salt & 0xFFFF);
1109 ip_fallback_id = salt;
1110 spin_unlock_bh(&ip_fb_id_lock);
1113 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1115 struct rtable *rt = (struct rtable *) dst;
1118 if (rt->peer == NULL)
1119 rt_bind_peer(rt, 1);
1121 /* If peer is attached to destination, it is never detached,
1122 so that we need not to grab a lock to dereference it.
1125 iph->id = htons(inet_getid(rt->peer, more));
1129 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1130 __builtin_return_address(0));
1132 ip_select_fb_ident(iph);
1135 static void rt_del(unsigned hash, struct rtable *rt)
1137 struct rtable **rthp, *aux;
1139 rthp = &rt_hash_table[hash].chain;
1140 spin_lock_bh(rt_hash_lock_addr(hash));
1142 while ((aux = *rthp) != NULL) {
1143 if (aux == rt || (aux->rt_genid != atomic_read(&rt_genid))) {
1144 *rthp = aux->u.dst.rt_next;
1148 rthp = &aux->u.dst.rt_next;
1150 spin_unlock_bh(rt_hash_lock_addr(hash));
1153 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1154 __be32 saddr, struct net_device *dev)
1157 struct in_device *in_dev = in_dev_get(dev);
1158 struct rtable *rth, **rthp;
1159 __be32 skeys[2] = { saddr, 0 };
1160 int ikeys[2] = { dev->ifindex, 0 };
1161 struct netevent_redirect netevent;
1168 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1169 || ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw)
1170 || ipv4_is_zeronet(new_gw))
1171 goto reject_redirect;
1173 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1174 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1175 goto reject_redirect;
1176 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1177 goto reject_redirect;
1179 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1180 goto reject_redirect;
1183 for (i = 0; i < 2; i++) {
1184 for (k = 0; k < 2; k++) {
1185 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k]);
1187 rthp=&rt_hash_table[hash].chain;
1190 while ((rth = rcu_dereference(*rthp)) != NULL) {
1193 if (rth->fl.fl4_dst != daddr ||
1194 rth->fl.fl4_src != skeys[i] ||
1195 rth->fl.oif != ikeys[k] ||
1197 rth->rt_genid != atomic_read(&rt_genid) ||
1198 !net_eq(dev_net(rth->u.dst.dev), net)) {
1199 rthp = &rth->u.dst.rt_next;
1203 if (rth->rt_dst != daddr ||
1204 rth->rt_src != saddr ||
1206 rth->rt_gateway != old_gw ||
1207 rth->u.dst.dev != dev)
1210 dst_hold(&rth->u.dst);
1213 rt = dst_alloc(&ipv4_dst_ops);
1220 /* Copy all the information. */
1222 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1223 rt->u.dst.__use = 1;
1224 atomic_set(&rt->u.dst.__refcnt, 1);
1225 rt->u.dst.child = NULL;
1227 dev_hold(rt->u.dst.dev);
1229 in_dev_hold(rt->idev);
1230 rt->u.dst.obsolete = 0;
1231 rt->u.dst.lastuse = jiffies;
1232 rt->u.dst.path = &rt->u.dst;
1233 rt->u.dst.neighbour = NULL;
1234 rt->u.dst.hh = NULL;
1235 rt->u.dst.xfrm = NULL;
1236 rt->rt_genid = atomic_read(&rt_genid);
1237 rt->rt_flags |= RTCF_REDIRECTED;
1239 /* Gateway is different ... */
1240 rt->rt_gateway = new_gw;
1242 /* Redirect received -> path was valid */
1243 dst_confirm(&rth->u.dst);
1246 atomic_inc(&rt->peer->refcnt);
1248 if (arp_bind_neighbour(&rt->u.dst) ||
1249 !(rt->u.dst.neighbour->nud_state &
1251 if (rt->u.dst.neighbour)
1252 neigh_event_send(rt->u.dst.neighbour, NULL);
1258 netevent.old = &rth->u.dst;
1259 netevent.new = &rt->u.dst;
1260 call_netevent_notifiers(NETEVENT_REDIRECT,
1264 if (!rt_intern_hash(hash, rt, &rt))
1277 #ifdef CONFIG_IP_ROUTE_VERBOSE
1278 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1279 printk(KERN_INFO "Redirect from " NIPQUAD_FMT " on %s about "
1280 NIPQUAD_FMT " ignored.\n"
1281 " Advised path = " NIPQUAD_FMT " -> " NIPQUAD_FMT "\n",
1282 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
1283 NIPQUAD(saddr), NIPQUAD(daddr));
1288 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1290 struct rtable *rt = (struct rtable *)dst;
1291 struct dst_entry *ret = dst;
1294 if (dst->obsolete) {
1297 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1298 rt->u.dst.expires) {
1299 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1301 #if RT_CACHE_DEBUG >= 1
1302 printk(KERN_DEBUG "ipv4_negative_advice: redirect to "
1303 NIPQUAD_FMT "/%02x dropped\n",
1304 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1315 * 1. The first ip_rt_redirect_number redirects are sent
1316 * with exponential backoff, then we stop sending them at all,
1317 * assuming that the host ignores our redirects.
1318 * 2. If we did not see packets requiring redirects
1319 * during ip_rt_redirect_silence, we assume that the host
1320 * forgot redirected route and start to send redirects again.
1322 * This algorithm is much cheaper and more intelligent than dumb load limiting
1325 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1326 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1329 void ip_rt_send_redirect(struct sk_buff *skb)
1331 struct rtable *rt = skb->rtable;
1332 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1337 if (!IN_DEV_TX_REDIRECTS(in_dev))
1340 /* No redirected packets during ip_rt_redirect_silence;
1341 * reset the algorithm.
1343 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1344 rt->u.dst.rate_tokens = 0;
1346 /* Too many ignored redirects; do not send anything
1347 * set u.dst.rate_last to the last seen redirected packet.
1349 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1350 rt->u.dst.rate_last = jiffies;
1354 /* Check for load limit; set rate_last to the latest sent
1357 if (rt->u.dst.rate_tokens == 0 ||
1359 (rt->u.dst.rate_last +
1360 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1361 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1362 rt->u.dst.rate_last = jiffies;
1363 ++rt->u.dst.rate_tokens;
1364 #ifdef CONFIG_IP_ROUTE_VERBOSE
1365 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1366 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1368 printk(KERN_WARNING "host " NIPQUAD_FMT "/if%d ignores "
1369 "redirects for " NIPQUAD_FMT " to " NIPQUAD_FMT ".\n",
1370 NIPQUAD(rt->rt_src), rt->rt_iif,
1371 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1378 static int ip_error(struct sk_buff *skb)
1380 struct rtable *rt = skb->rtable;
1384 switch (rt->u.dst.error) {
1389 code = ICMP_HOST_UNREACH;
1392 code = ICMP_NET_UNREACH;
1393 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
1396 code = ICMP_PKT_FILTERED;
1401 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1402 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1403 rt->u.dst.rate_tokens = ip_rt_error_burst;
1404 rt->u.dst.rate_last = now;
1405 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1406 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1407 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1410 out: kfree_skb(skb);
1415 * The last two values are not from the RFC but
1416 * are needed for AMPRnet AX.25 paths.
1419 static const unsigned short mtu_plateau[] =
1420 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1422 static inline unsigned short guess_mtu(unsigned short old_mtu)
1426 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1427 if (old_mtu > mtu_plateau[i])
1428 return mtu_plateau[i];
1432 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1433 unsigned short new_mtu,
1434 struct net_device *dev)
1437 unsigned short old_mtu = ntohs(iph->tot_len);
1439 int ikeys[2] = { dev->ifindex, 0 };
1440 __be32 skeys[2] = { iph->saddr, 0, };
1441 __be32 daddr = iph->daddr;
1442 unsigned short est_mtu = 0;
1444 if (ipv4_config.no_pmtu_disc)
1447 for (k = 0; k < 2; k++) {
1448 for (i = 0; i < 2; i++) {
1449 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k]);
1452 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1453 rth = rcu_dereference(rth->u.dst.rt_next)) {
1454 unsigned short mtu = new_mtu;
1456 if (rth->fl.fl4_dst != daddr ||
1457 rth->fl.fl4_src != skeys[i] ||
1458 rth->rt_dst != daddr ||
1459 rth->rt_src != iph->saddr ||
1460 rth->fl.oif != ikeys[k] ||
1462 dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1463 !net_eq(dev_net(rth->u.dst.dev), net) ||
1464 rth->rt_genid != atomic_read(&rt_genid))
1467 if (new_mtu < 68 || new_mtu >= old_mtu) {
1469 /* BSD 4.2 compatibility hack :-( */
1471 old_mtu >= dst_metric(&rth->u.dst, RTAX_MTU) &&
1472 old_mtu >= 68 + (iph->ihl << 2))
1473 old_mtu -= iph->ihl << 2;
1475 mtu = guess_mtu(old_mtu);
1477 if (mtu <= dst_metric(&rth->u.dst, RTAX_MTU)) {
1478 if (mtu < dst_metric(&rth->u.dst, RTAX_MTU)) {
1479 dst_confirm(&rth->u.dst);
1480 if (mtu < ip_rt_min_pmtu) {
1481 mtu = ip_rt_min_pmtu;
1482 rth->u.dst.metrics[RTAX_LOCK-1] |=
1485 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1486 dst_set_expires(&rth->u.dst,
1495 return est_mtu ? : new_mtu;
1498 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1500 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= 68 &&
1501 !(dst_metric_locked(dst, RTAX_MTU))) {
1502 if (mtu < ip_rt_min_pmtu) {
1503 mtu = ip_rt_min_pmtu;
1504 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1506 dst->metrics[RTAX_MTU-1] = mtu;
1507 dst_set_expires(dst, ip_rt_mtu_expires);
1508 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1512 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1517 static void ipv4_dst_destroy(struct dst_entry *dst)
1519 struct rtable *rt = (struct rtable *) dst;
1520 struct inet_peer *peer = rt->peer;
1521 struct in_device *idev = rt->idev;
1534 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1537 struct rtable *rt = (struct rtable *) dst;
1538 struct in_device *idev = rt->idev;
1539 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1540 struct in_device *loopback_idev =
1541 in_dev_get(dev_net(dev)->loopback_dev);
1542 if (loopback_idev) {
1543 rt->idev = loopback_idev;
1549 static void ipv4_link_failure(struct sk_buff *skb)
1553 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1557 dst_set_expires(&rt->u.dst, 0);
1560 static int ip_rt_bug(struct sk_buff *skb)
1562 printk(KERN_DEBUG "ip_rt_bug: " NIPQUAD_FMT " -> " NIPQUAD_FMT ", %s\n",
1563 NIPQUAD(ip_hdr(skb)->saddr), NIPQUAD(ip_hdr(skb)->daddr),
1564 skb->dev ? skb->dev->name : "?");
1570 We do not cache source address of outgoing interface,
1571 because it is used only by IP RR, TS and SRR options,
1572 so that it out of fast path.
1574 BTW remember: "addr" is allowed to be not aligned
1578 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1581 struct fib_result res;
1583 if (rt->fl.iif == 0)
1585 else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1586 src = FIB_RES_PREFSRC(res);
1589 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1591 memcpy(addr, &src, 4);
1594 #ifdef CONFIG_NET_CLS_ROUTE
1595 static void set_class_tag(struct rtable *rt, u32 tag)
1597 if (!(rt->u.dst.tclassid & 0xFFFF))
1598 rt->u.dst.tclassid |= tag & 0xFFFF;
1599 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1600 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1604 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1606 struct fib_info *fi = res->fi;
1609 if (FIB_RES_GW(*res) &&
1610 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1611 rt->rt_gateway = FIB_RES_GW(*res);
1612 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1613 sizeof(rt->u.dst.metrics));
1614 if (fi->fib_mtu == 0) {
1615 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1616 if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1617 rt->rt_gateway != rt->rt_dst &&
1618 rt->u.dst.dev->mtu > 576)
1619 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1621 #ifdef CONFIG_NET_CLS_ROUTE
1622 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1625 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1627 if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1628 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1629 if (dst_metric(&rt->u.dst, RTAX_MTU) > IP_MAX_MTU)
1630 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1631 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1632 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1634 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1635 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1637 #ifdef CONFIG_NET_CLS_ROUTE
1638 #ifdef CONFIG_IP_MULTIPLE_TABLES
1639 set_class_tag(rt, fib_rules_tclass(res));
1641 set_class_tag(rt, itag);
1643 rt->rt_type = res->type;
1646 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1647 u8 tos, struct net_device *dev, int our)
1652 struct in_device *in_dev = in_dev_get(dev);
1655 /* Primary sanity checks. */
1660 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1661 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1664 if (ipv4_is_zeronet(saddr)) {
1665 if (!ipv4_is_local_multicast(daddr))
1667 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1668 } else if (fib_validate_source(saddr, 0, tos, 0,
1669 dev, &spec_dst, &itag) < 0)
1672 rth = dst_alloc(&ipv4_dst_ops);
1676 rth->u.dst.output= ip_rt_bug;
1678 atomic_set(&rth->u.dst.__refcnt, 1);
1679 rth->u.dst.flags= DST_HOST;
1680 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1681 rth->u.dst.flags |= DST_NOPOLICY;
1682 rth->fl.fl4_dst = daddr;
1683 rth->rt_dst = daddr;
1684 rth->fl.fl4_tos = tos;
1685 rth->fl.mark = skb->mark;
1686 rth->fl.fl4_src = saddr;
1687 rth->rt_src = saddr;
1688 #ifdef CONFIG_NET_CLS_ROUTE
1689 rth->u.dst.tclassid = itag;
1692 rth->fl.iif = dev->ifindex;
1693 rth->u.dst.dev = init_net.loopback_dev;
1694 dev_hold(rth->u.dst.dev);
1695 rth->idev = in_dev_get(rth->u.dst.dev);
1697 rth->rt_gateway = daddr;
1698 rth->rt_spec_dst= spec_dst;
1699 rth->rt_genid = atomic_read(&rt_genid);
1700 rth->rt_flags = RTCF_MULTICAST;
1701 rth->rt_type = RTN_MULTICAST;
1703 rth->u.dst.input= ip_local_deliver;
1704 rth->rt_flags |= RTCF_LOCAL;
1707 #ifdef CONFIG_IP_MROUTE
1708 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1709 rth->u.dst.input = ip_mr_input;
1711 RT_CACHE_STAT_INC(in_slow_mc);
1714 hash = rt_hash(daddr, saddr, dev->ifindex);
1715 return rt_intern_hash(hash, rth, &skb->rtable);
1727 static void ip_handle_martian_source(struct net_device *dev,
1728 struct in_device *in_dev,
1729 struct sk_buff *skb,
1733 RT_CACHE_STAT_INC(in_martian_src);
1734 #ifdef CONFIG_IP_ROUTE_VERBOSE
1735 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1737 * RFC1812 recommendation, if source is martian,
1738 * the only hint is MAC header.
1740 printk(KERN_WARNING "martian source " NIPQUAD_FMT " from "
1741 NIPQUAD_FMT", on dev %s\n",
1742 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1743 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1745 const unsigned char *p = skb_mac_header(skb);
1746 printk(KERN_WARNING "ll header: ");
1747 for (i = 0; i < dev->hard_header_len; i++, p++) {
1749 if (i < (dev->hard_header_len - 1))
1758 static int __mkroute_input(struct sk_buff *skb,
1759 struct fib_result *res,
1760 struct in_device *in_dev,
1761 __be32 daddr, __be32 saddr, u32 tos,
1762 struct rtable **result)
1767 struct in_device *out_dev;
1772 /* get a working reference to the output device */
1773 out_dev = in_dev_get(FIB_RES_DEV(*res));
1774 if (out_dev == NULL) {
1775 if (net_ratelimit())
1776 printk(KERN_CRIT "Bug in ip_route_input" \
1777 "_slow(). Please, report\n");
1782 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1783 in_dev->dev, &spec_dst, &itag);
1785 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1793 flags |= RTCF_DIRECTSRC;
1795 if (out_dev == in_dev && err &&
1796 (IN_DEV_SHARED_MEDIA(out_dev) ||
1797 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1798 flags |= RTCF_DOREDIRECT;
1800 if (skb->protocol != htons(ETH_P_IP)) {
1801 /* Not IP (i.e. ARP). Do not create route, if it is
1802 * invalid for proxy arp. DNAT routes are always valid.
1804 if (out_dev == in_dev) {
1811 rth = dst_alloc(&ipv4_dst_ops);
1817 atomic_set(&rth->u.dst.__refcnt, 1);
1818 rth->u.dst.flags= DST_HOST;
1819 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1820 rth->u.dst.flags |= DST_NOPOLICY;
1821 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
1822 rth->u.dst.flags |= DST_NOXFRM;
1823 rth->fl.fl4_dst = daddr;
1824 rth->rt_dst = daddr;
1825 rth->fl.fl4_tos = tos;
1826 rth->fl.mark = skb->mark;
1827 rth->fl.fl4_src = saddr;
1828 rth->rt_src = saddr;
1829 rth->rt_gateway = daddr;
1831 rth->fl.iif = in_dev->dev->ifindex;
1832 rth->u.dst.dev = (out_dev)->dev;
1833 dev_hold(rth->u.dst.dev);
1834 rth->idev = in_dev_get(rth->u.dst.dev);
1836 rth->rt_spec_dst= spec_dst;
1838 rth->u.dst.input = ip_forward;
1839 rth->u.dst.output = ip_output;
1840 rth->rt_genid = atomic_read(&rt_genid);
1842 rt_set_nexthop(rth, res, itag);
1844 rth->rt_flags = flags;
1849 /* release the working reference to the output device */
1850 in_dev_put(out_dev);
1854 static int ip_mkroute_input(struct sk_buff *skb,
1855 struct fib_result *res,
1856 const struct flowi *fl,
1857 struct in_device *in_dev,
1858 __be32 daddr, __be32 saddr, u32 tos)
1860 struct rtable* rth = NULL;
1864 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1865 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1866 fib_select_multipath(fl, res);
1869 /* create a routing cache entry */
1870 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1874 /* put it into the cache */
1875 hash = rt_hash(daddr, saddr, fl->iif);
1876 return rt_intern_hash(hash, rth, &skb->rtable);
1880 * NOTE. We drop all the packets that has local source
1881 * addresses, because every properly looped back packet
1882 * must have correct destination already attached by output routine.
1884 * Such approach solves two big problems:
1885 * 1. Not simplex devices are handled properly.
1886 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1889 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1890 u8 tos, struct net_device *dev)
1892 struct fib_result res;
1893 struct in_device *in_dev = in_dev_get(dev);
1894 struct flowi fl = { .nl_u = { .ip4_u =
1898 .scope = RT_SCOPE_UNIVERSE,
1901 .iif = dev->ifindex };
1904 struct rtable * rth;
1909 struct net * net = dev_net(dev);
1911 /* IP on this device is disabled. */
1916 /* Check for the most weird martians, which can be not detected
1920 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1921 ipv4_is_loopback(saddr))
1922 goto martian_source;
1924 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
1927 /* Accept zero addresses only to limited broadcast;
1928 * I even do not know to fix it or not. Waiting for complains :-)
1930 if (ipv4_is_zeronet(saddr))
1931 goto martian_source;
1933 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
1934 ipv4_is_loopback(daddr))
1935 goto martian_destination;
1938 * Now we are ready to route packet.
1940 if ((err = fib_lookup(net, &fl, &res)) != 0) {
1941 if (!IN_DEV_FORWARD(in_dev))
1947 RT_CACHE_STAT_INC(in_slow_tot);
1949 if (res.type == RTN_BROADCAST)
1952 if (res.type == RTN_LOCAL) {
1954 result = fib_validate_source(saddr, daddr, tos,
1955 net->loopback_dev->ifindex,
1956 dev, &spec_dst, &itag);
1958 goto martian_source;
1960 flags |= RTCF_DIRECTSRC;
1965 if (!IN_DEV_FORWARD(in_dev))
1967 if (res.type != RTN_UNICAST)
1968 goto martian_destination;
1970 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1978 if (skb->protocol != htons(ETH_P_IP))
1981 if (ipv4_is_zeronet(saddr))
1982 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1984 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1987 goto martian_source;
1989 flags |= RTCF_DIRECTSRC;
1991 flags |= RTCF_BROADCAST;
1992 res.type = RTN_BROADCAST;
1993 RT_CACHE_STAT_INC(in_brd);
1996 rth = dst_alloc(&ipv4_dst_ops);
2000 rth->u.dst.output= ip_rt_bug;
2001 rth->rt_genid = atomic_read(&rt_genid);
2003 atomic_set(&rth->u.dst.__refcnt, 1);
2004 rth->u.dst.flags= DST_HOST;
2005 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2006 rth->u.dst.flags |= DST_NOPOLICY;
2007 rth->fl.fl4_dst = daddr;
2008 rth->rt_dst = daddr;
2009 rth->fl.fl4_tos = tos;
2010 rth->fl.mark = skb->mark;
2011 rth->fl.fl4_src = saddr;
2012 rth->rt_src = saddr;
2013 #ifdef CONFIG_NET_CLS_ROUTE
2014 rth->u.dst.tclassid = itag;
2017 rth->fl.iif = dev->ifindex;
2018 rth->u.dst.dev = net->loopback_dev;
2019 dev_hold(rth->u.dst.dev);
2020 rth->idev = in_dev_get(rth->u.dst.dev);
2021 rth->rt_gateway = daddr;
2022 rth->rt_spec_dst= spec_dst;
2023 rth->u.dst.input= ip_local_deliver;
2024 rth->rt_flags = flags|RTCF_LOCAL;
2025 if (res.type == RTN_UNREACHABLE) {
2026 rth->u.dst.input= ip_error;
2027 rth->u.dst.error= -err;
2028 rth->rt_flags &= ~RTCF_LOCAL;
2030 rth->rt_type = res.type;
2031 hash = rt_hash(daddr, saddr, fl.iif);
2032 err = rt_intern_hash(hash, rth, &skb->rtable);
2036 RT_CACHE_STAT_INC(in_no_route);
2037 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2038 res.type = RTN_UNREACHABLE;
2044 * Do not cache martian addresses: they should be logged (RFC1812)
2046 martian_destination:
2047 RT_CACHE_STAT_INC(in_martian_dst);
2048 #ifdef CONFIG_IP_ROUTE_VERBOSE
2049 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2050 printk(KERN_WARNING "martian destination " NIPQUAD_FMT " from "
2051 NIPQUAD_FMT ", dev %s\n",
2052 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
2056 err = -EHOSTUNREACH;
2068 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2072 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2073 u8 tos, struct net_device *dev)
2075 struct rtable * rth;
2077 int iif = dev->ifindex;
2081 tos &= IPTOS_RT_MASK;
2082 hash = rt_hash(daddr, saddr, iif);
2085 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2086 rth = rcu_dereference(rth->u.dst.rt_next)) {
2087 if (((rth->fl.fl4_dst ^ daddr) |
2088 (rth->fl.fl4_src ^ saddr) |
2089 (rth->fl.iif ^ iif) |
2091 (rth->fl.fl4_tos ^ tos)) == 0 &&
2092 rth->fl.mark == skb->mark &&
2093 net_eq(dev_net(rth->u.dst.dev), net) &&
2094 rth->rt_genid == atomic_read(&rt_genid)) {
2095 dst_use(&rth->u.dst, jiffies);
2096 RT_CACHE_STAT_INC(in_hit);
2101 RT_CACHE_STAT_INC(in_hlist_search);
2105 /* Multicast recognition logic is moved from route cache to here.
2106 The problem was that too many Ethernet cards have broken/missing
2107 hardware multicast filters :-( As result the host on multicasting
2108 network acquires a lot of useless route cache entries, sort of
2109 SDR messages from all the world. Now we try to get rid of them.
2110 Really, provided software IP multicast filter is organized
2111 reasonably (at least, hashed), it does not result in a slowdown
2112 comparing with route cache reject entries.
2113 Note, that multicast routers are not affected, because
2114 route cache entry is created eventually.
2116 if (ipv4_is_multicast(daddr)) {
2117 struct in_device *in_dev;
2120 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2121 int our = ip_check_mc(in_dev, daddr, saddr,
2122 ip_hdr(skb)->protocol);
2124 #ifdef CONFIG_IP_MROUTE
2125 || (!ipv4_is_local_multicast(daddr) &&
2126 IN_DEV_MFORWARD(in_dev))
2130 return ip_route_input_mc(skb, daddr, saddr,
2137 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2140 static int __mkroute_output(struct rtable **result,
2141 struct fib_result *res,
2142 const struct flowi *fl,
2143 const struct flowi *oldflp,
2144 struct net_device *dev_out,
2148 struct in_device *in_dev;
2149 u32 tos = RT_FL_TOS(oldflp);
2152 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2155 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2156 res->type = RTN_BROADCAST;
2157 else if (ipv4_is_multicast(fl->fl4_dst))
2158 res->type = RTN_MULTICAST;
2159 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2162 if (dev_out->flags & IFF_LOOPBACK)
2163 flags |= RTCF_LOCAL;
2165 /* get work reference to inet device */
2166 in_dev = in_dev_get(dev_out);
2170 if (res->type == RTN_BROADCAST) {
2171 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2173 fib_info_put(res->fi);
2176 } else if (res->type == RTN_MULTICAST) {
2177 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2178 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2180 flags &= ~RTCF_LOCAL;
2181 /* If multicast route do not exist use
2182 default one, but do not gateway in this case.
2185 if (res->fi && res->prefixlen < 4) {
2186 fib_info_put(res->fi);
2192 rth = dst_alloc(&ipv4_dst_ops);
2198 atomic_set(&rth->u.dst.__refcnt, 1);
2199 rth->u.dst.flags= DST_HOST;
2200 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2201 rth->u.dst.flags |= DST_NOXFRM;
2202 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2203 rth->u.dst.flags |= DST_NOPOLICY;
2205 rth->fl.fl4_dst = oldflp->fl4_dst;
2206 rth->fl.fl4_tos = tos;
2207 rth->fl.fl4_src = oldflp->fl4_src;
2208 rth->fl.oif = oldflp->oif;
2209 rth->fl.mark = oldflp->mark;
2210 rth->rt_dst = fl->fl4_dst;
2211 rth->rt_src = fl->fl4_src;
2212 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2213 /* get references to the devices that are to be hold by the routing
2215 rth->u.dst.dev = dev_out;
2217 rth->idev = in_dev_get(dev_out);
2218 rth->rt_gateway = fl->fl4_dst;
2219 rth->rt_spec_dst= fl->fl4_src;
2221 rth->u.dst.output=ip_output;
2222 rth->rt_genid = atomic_read(&rt_genid);
2224 RT_CACHE_STAT_INC(out_slow_tot);
2226 if (flags & RTCF_LOCAL) {
2227 rth->u.dst.input = ip_local_deliver;
2228 rth->rt_spec_dst = fl->fl4_dst;
2230 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2231 rth->rt_spec_dst = fl->fl4_src;
2232 if (flags & RTCF_LOCAL &&
2233 !(dev_out->flags & IFF_LOOPBACK)) {
2234 rth->u.dst.output = ip_mc_output;
2235 RT_CACHE_STAT_INC(out_slow_mc);
2237 #ifdef CONFIG_IP_MROUTE
2238 if (res->type == RTN_MULTICAST) {
2239 if (IN_DEV_MFORWARD(in_dev) &&
2240 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2241 rth->u.dst.input = ip_mr_input;
2242 rth->u.dst.output = ip_mc_output;
2248 rt_set_nexthop(rth, res, 0);
2250 rth->rt_flags = flags;
2254 /* release work reference to inet device */
2260 static int ip_mkroute_output(struct rtable **rp,
2261 struct fib_result *res,
2262 const struct flowi *fl,
2263 const struct flowi *oldflp,
2264 struct net_device *dev_out,
2267 struct rtable *rth = NULL;
2268 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2271 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif);
2272 err = rt_intern_hash(hash, rth, rp);
2279 * Major route resolver routine.
2282 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2283 const struct flowi *oldflp)
2285 u32 tos = RT_FL_TOS(oldflp);
2286 struct flowi fl = { .nl_u = { .ip4_u =
2287 { .daddr = oldflp->fl4_dst,
2288 .saddr = oldflp->fl4_src,
2289 .tos = tos & IPTOS_RT_MASK,
2290 .scope = ((tos & RTO_ONLINK) ?
2294 .mark = oldflp->mark,
2295 .iif = net->loopback_dev->ifindex,
2296 .oif = oldflp->oif };
2297 struct fib_result res;
2299 struct net_device *dev_out = NULL;
2305 #ifdef CONFIG_IP_MULTIPLE_TABLES
2309 if (oldflp->fl4_src) {
2311 if (ipv4_is_multicast(oldflp->fl4_src) ||
2312 ipv4_is_lbcast(oldflp->fl4_src) ||
2313 ipv4_is_zeronet(oldflp->fl4_src))
2316 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2317 dev_out = ip_dev_find(net, oldflp->fl4_src);
2318 if (dev_out == NULL)
2321 /* I removed check for oif == dev_out->oif here.
2322 It was wrong for two reasons:
2323 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2324 is assigned to multiple interfaces.
2325 2. Moreover, we are allowed to send packets with saddr
2326 of another iface. --ANK
2329 if (oldflp->oif == 0
2330 && (ipv4_is_multicast(oldflp->fl4_dst) ||
2331 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2332 /* Special hack: user can direct multicasts
2333 and limited broadcast via necessary interface
2334 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2335 This hack is not just for fun, it allows
2336 vic,vat and friends to work.
2337 They bind socket to loopback, set ttl to zero
2338 and expect that it will work.
2339 From the viewpoint of routing cache they are broken,
2340 because we are not allowed to build multicast path
2341 with loopback source addr (look, routing cache
2342 cannot know, that ttl is zero, so that packet
2343 will not leave this host and route is valid).
2344 Luckily, this hack is good workaround.
2347 fl.oif = dev_out->ifindex;
2357 dev_out = dev_get_by_index(net, oldflp->oif);
2359 if (dev_out == NULL)
2362 /* RACE: Check return value of inet_select_addr instead. */
2363 if (__in_dev_get_rtnl(dev_out) == NULL) {
2365 goto out; /* Wrong error code */
2368 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2369 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2371 fl.fl4_src = inet_select_addr(dev_out, 0,
2376 if (ipv4_is_multicast(oldflp->fl4_dst))
2377 fl.fl4_src = inet_select_addr(dev_out, 0,
2379 else if (!oldflp->fl4_dst)
2380 fl.fl4_src = inet_select_addr(dev_out, 0,
2386 fl.fl4_dst = fl.fl4_src;
2388 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2391 dev_out = net->loopback_dev;
2393 fl.oif = net->loopback_dev->ifindex;
2394 res.type = RTN_LOCAL;
2395 flags |= RTCF_LOCAL;
2399 if (fib_lookup(net, &fl, &res)) {
2402 /* Apparently, routing tables are wrong. Assume,
2403 that the destination is on link.
2406 Because we are allowed to send to iface
2407 even if it has NO routes and NO assigned
2408 addresses. When oif is specified, routing
2409 tables are looked up with only one purpose:
2410 to catch if destination is gatewayed, rather than
2411 direct. Moreover, if MSG_DONTROUTE is set,
2412 we send packet, ignoring both routing tables
2413 and ifaddr state. --ANK
2416 We could make it even if oif is unknown,
2417 likely IPv6, but we do not.
2420 if (fl.fl4_src == 0)
2421 fl.fl4_src = inet_select_addr(dev_out, 0,
2423 res.type = RTN_UNICAST;
2433 if (res.type == RTN_LOCAL) {
2435 fl.fl4_src = fl.fl4_dst;
2438 dev_out = net->loopback_dev;
2440 fl.oif = dev_out->ifindex;
2442 fib_info_put(res.fi);
2444 flags |= RTCF_LOCAL;
2448 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2449 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2450 fib_select_multipath(&fl, &res);
2453 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2454 fib_select_default(net, &fl, &res);
2457 fl.fl4_src = FIB_RES_PREFSRC(res);
2461 dev_out = FIB_RES_DEV(res);
2463 fl.oif = dev_out->ifindex;
2467 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2477 int __ip_route_output_key(struct net *net, struct rtable **rp,
2478 const struct flowi *flp)
2483 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif);
2486 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2487 rth = rcu_dereference(rth->u.dst.rt_next)) {
2488 if (rth->fl.fl4_dst == flp->fl4_dst &&
2489 rth->fl.fl4_src == flp->fl4_src &&
2491 rth->fl.oif == flp->oif &&
2492 rth->fl.mark == flp->mark &&
2493 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2494 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2495 net_eq(dev_net(rth->u.dst.dev), net) &&
2496 rth->rt_genid == atomic_read(&rt_genid)) {
2497 dst_use(&rth->u.dst, jiffies);
2498 RT_CACHE_STAT_INC(out_hit);
2499 rcu_read_unlock_bh();
2503 RT_CACHE_STAT_INC(out_hlist_search);
2505 rcu_read_unlock_bh();
2507 return ip_route_output_slow(net, rp, flp);
2510 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2512 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2516 static struct dst_ops ipv4_dst_blackhole_ops = {
2518 .protocol = __constant_htons(ETH_P_IP),
2519 .destroy = ipv4_dst_destroy,
2520 .check = ipv4_dst_check,
2521 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2522 .entry_size = sizeof(struct rtable),
2523 .entries = ATOMIC_INIT(0),
2527 static int ipv4_dst_blackhole(struct rtable **rp, struct flowi *flp)
2529 struct rtable *ort = *rp;
2530 struct rtable *rt = (struct rtable *)
2531 dst_alloc(&ipv4_dst_blackhole_ops);
2534 struct dst_entry *new = &rt->u.dst;
2536 atomic_set(&new->__refcnt, 1);
2538 new->input = dst_discard;
2539 new->output = dst_discard;
2540 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2542 new->dev = ort->u.dst.dev;
2548 rt->idev = ort->idev;
2550 in_dev_hold(rt->idev);
2551 rt->rt_genid = atomic_read(&rt_genid);
2552 rt->rt_flags = ort->rt_flags;
2553 rt->rt_type = ort->rt_type;
2554 rt->rt_dst = ort->rt_dst;
2555 rt->rt_src = ort->rt_src;
2556 rt->rt_iif = ort->rt_iif;
2557 rt->rt_gateway = ort->rt_gateway;
2558 rt->rt_spec_dst = ort->rt_spec_dst;
2559 rt->peer = ort->peer;
2561 atomic_inc(&rt->peer->refcnt);
2566 dst_release(&(*rp)->u.dst);
2568 return (rt ? 0 : -ENOMEM);
2571 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2572 struct sock *sk, int flags)
2576 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2581 flp->fl4_src = (*rp)->rt_src;
2583 flp->fl4_dst = (*rp)->rt_dst;
2584 err = __xfrm_lookup((struct dst_entry **)rp, flp, sk,
2585 flags ? XFRM_LOOKUP_WAIT : 0);
2586 if (err == -EREMOTE)
2587 err = ipv4_dst_blackhole(rp, flp);
2595 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2597 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2599 return ip_route_output_flow(net, rp, flp, NULL, 0);
2602 static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
2603 int nowait, unsigned int flags)
2605 struct rtable *rt = skb->rtable;
2607 struct nlmsghdr *nlh;
2609 u32 id = 0, ts = 0, tsage = 0, error;
2611 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2615 r = nlmsg_data(nlh);
2616 r->rtm_family = AF_INET;
2617 r->rtm_dst_len = 32;
2619 r->rtm_tos = rt->fl.fl4_tos;
2620 r->rtm_table = RT_TABLE_MAIN;
2621 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2622 r->rtm_type = rt->rt_type;
2623 r->rtm_scope = RT_SCOPE_UNIVERSE;
2624 r->rtm_protocol = RTPROT_UNSPEC;
2625 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2626 if (rt->rt_flags & RTCF_NOTIFY)
2627 r->rtm_flags |= RTM_F_NOTIFY;
2629 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2631 if (rt->fl.fl4_src) {
2632 r->rtm_src_len = 32;
2633 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2636 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2637 #ifdef CONFIG_NET_CLS_ROUTE
2638 if (rt->u.dst.tclassid)
2639 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2642 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2643 else if (rt->rt_src != rt->fl.fl4_src)
2644 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2646 if (rt->rt_dst != rt->rt_gateway)
2647 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2649 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2650 goto nla_put_failure;
2652 error = rt->u.dst.error;
2653 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2655 id = rt->peer->ip_id_count;
2656 if (rt->peer->tcp_ts_stamp) {
2657 ts = rt->peer->tcp_ts;
2658 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2663 #ifdef CONFIG_IP_MROUTE
2664 __be32 dst = rt->rt_dst;
2666 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2667 IPV4_DEVCONF_ALL(&init_net, MC_FORWARDING)) {
2668 int err = ipmr_get_route(skb, r, nowait);
2673 goto nla_put_failure;
2675 if (err == -EMSGSIZE)
2676 goto nla_put_failure;
2682 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2685 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2686 expires, error) < 0)
2687 goto nla_put_failure;
2689 return nlmsg_end(skb, nlh);
2692 nlmsg_cancel(skb, nlh);
2696 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2698 struct net *net = sock_net(in_skb->sk);
2700 struct nlattr *tb[RTA_MAX+1];
2701 struct rtable *rt = NULL;
2706 struct sk_buff *skb;
2708 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2712 rtm = nlmsg_data(nlh);
2714 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2720 /* Reserve room for dummy headers, this skb can pass
2721 through good chunk of routing engine.
2723 skb_reset_mac_header(skb);
2724 skb_reset_network_header(skb);
2726 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2727 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2728 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2730 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2731 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2732 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2735 struct net_device *dev;
2737 dev = __dev_get_by_index(net, iif);
2743 skb->protocol = htons(ETH_P_IP);
2746 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2750 if (err == 0 && rt->u.dst.error)
2751 err = -rt->u.dst.error;
2758 .tos = rtm->rtm_tos,
2761 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2763 err = ip_route_output_key(net, &rt, &fl);
2770 if (rtm->rtm_flags & RTM_F_NOTIFY)
2771 rt->rt_flags |= RTCF_NOTIFY;
2773 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2774 RTM_NEWROUTE, 0, 0);
2778 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2787 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2794 net = sock_net(skb->sk);
2799 s_idx = idx = cb->args[1];
2800 for (h = s_h; h <= rt_hash_mask; h++) {
2802 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2803 rt = rcu_dereference(rt->u.dst.rt_next), idx++) {
2804 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
2806 if (rt->rt_genid != atomic_read(&rt_genid))
2808 skb->dst = dst_clone(&rt->u.dst);
2809 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
2810 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2811 1, NLM_F_MULTI) <= 0) {
2812 dst_release(xchg(&skb->dst, NULL));
2813 rcu_read_unlock_bh();
2816 dst_release(xchg(&skb->dst, NULL));
2818 rcu_read_unlock_bh();
2828 void ip_rt_multicast_event(struct in_device *in_dev)
2833 #ifdef CONFIG_SYSCTL
2834 static int flush_delay;
2836 static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write,
2837 struct file *filp, void __user *buffer,
2838 size_t *lenp, loff_t *ppos)
2841 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2842 rt_cache_flush(flush_delay);
2849 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2852 void __user *oldval,
2853 size_t __user *oldlenp,
2854 void __user *newval,
2858 if (newlen != sizeof(int))
2860 if (get_user(delay, (int __user *)newval))
2862 rt_cache_flush(delay);
2866 ctl_table ipv4_route_table[] = {
2868 .ctl_name = NET_IPV4_ROUTE_FLUSH,
2869 .procname = "flush",
2870 .data = &flush_delay,
2871 .maxlen = sizeof(int),
2873 .proc_handler = &ipv4_sysctl_rtcache_flush,
2874 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
2877 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2878 .procname = "gc_thresh",
2879 .data = &ipv4_dst_ops.gc_thresh,
2880 .maxlen = sizeof(int),
2882 .proc_handler = &proc_dointvec,
2885 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2886 .procname = "max_size",
2887 .data = &ip_rt_max_size,
2888 .maxlen = sizeof(int),
2890 .proc_handler = &proc_dointvec,
2893 /* Deprecated. Use gc_min_interval_ms */
2895 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2896 .procname = "gc_min_interval",
2897 .data = &ip_rt_gc_min_interval,
2898 .maxlen = sizeof(int),
2900 .proc_handler = &proc_dointvec_jiffies,
2901 .strategy = &sysctl_jiffies,
2904 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2905 .procname = "gc_min_interval_ms",
2906 .data = &ip_rt_gc_min_interval,
2907 .maxlen = sizeof(int),
2909 .proc_handler = &proc_dointvec_ms_jiffies,
2910 .strategy = &sysctl_ms_jiffies,
2913 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2914 .procname = "gc_timeout",
2915 .data = &ip_rt_gc_timeout,
2916 .maxlen = sizeof(int),
2918 .proc_handler = &proc_dointvec_jiffies,
2919 .strategy = &sysctl_jiffies,
2922 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2923 .procname = "gc_interval",
2924 .data = &ip_rt_gc_interval,
2925 .maxlen = sizeof(int),
2927 .proc_handler = &proc_dointvec_jiffies,
2928 .strategy = &sysctl_jiffies,
2931 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2932 .procname = "redirect_load",
2933 .data = &ip_rt_redirect_load,
2934 .maxlen = sizeof(int),
2936 .proc_handler = &proc_dointvec,
2939 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2940 .procname = "redirect_number",
2941 .data = &ip_rt_redirect_number,
2942 .maxlen = sizeof(int),
2944 .proc_handler = &proc_dointvec,
2947 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
2948 .procname = "redirect_silence",
2949 .data = &ip_rt_redirect_silence,
2950 .maxlen = sizeof(int),
2952 .proc_handler = &proc_dointvec,
2955 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
2956 .procname = "error_cost",
2957 .data = &ip_rt_error_cost,
2958 .maxlen = sizeof(int),
2960 .proc_handler = &proc_dointvec,
2963 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
2964 .procname = "error_burst",
2965 .data = &ip_rt_error_burst,
2966 .maxlen = sizeof(int),
2968 .proc_handler = &proc_dointvec,
2971 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
2972 .procname = "gc_elasticity",
2973 .data = &ip_rt_gc_elasticity,
2974 .maxlen = sizeof(int),
2976 .proc_handler = &proc_dointvec,
2979 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
2980 .procname = "mtu_expires",
2981 .data = &ip_rt_mtu_expires,
2982 .maxlen = sizeof(int),
2984 .proc_handler = &proc_dointvec_jiffies,
2985 .strategy = &sysctl_jiffies,
2988 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
2989 .procname = "min_pmtu",
2990 .data = &ip_rt_min_pmtu,
2991 .maxlen = sizeof(int),
2993 .proc_handler = &proc_dointvec,
2996 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
2997 .procname = "min_adv_mss",
2998 .data = &ip_rt_min_advmss,
2999 .maxlen = sizeof(int),
3001 .proc_handler = &proc_dointvec,
3004 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
3005 .procname = "secret_interval",
3006 .data = &ip_rt_secret_interval,
3007 .maxlen = sizeof(int),
3009 .proc_handler = &proc_dointvec_jiffies,
3010 .strategy = &sysctl_jiffies,
3016 #ifdef CONFIG_NET_CLS_ROUTE
3017 struct ip_rt_acct *ip_rt_acct __read_mostly;
3018 #endif /* CONFIG_NET_CLS_ROUTE */
3020 static __initdata unsigned long rhash_entries;
3021 static int __init set_rhash_entries(char *str)
3025 rhash_entries = simple_strtoul(str, &str, 0);
3028 __setup("rhash_entries=", set_rhash_entries);
3030 int __init ip_rt_init(void)
3034 atomic_set(&rt_genid, (int) ((num_physpages ^ (num_physpages>>8)) ^
3035 (jiffies ^ (jiffies >> 7))));
3037 #ifdef CONFIG_NET_CLS_ROUTE
3038 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct));
3040 panic("IP: failed to allocate ip_rt_acct\n");
3043 ipv4_dst_ops.kmem_cachep =
3044 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3045 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3047 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3049 rt_hash_table = (struct rt_hash_bucket *)
3050 alloc_large_system_hash("IP route cache",
3051 sizeof(struct rt_hash_bucket),
3053 (num_physpages >= 128 * 1024) ?
3059 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3060 rt_hash_lock_init();
3062 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3063 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3068 rt_secret_timer.function = rt_secret_rebuild;
3069 rt_secret_timer.data = 0;
3070 init_timer_deferrable(&rt_secret_timer);
3072 /* All the timers, started at system startup tend
3073 to synchronize. Perturb it a bit.
3075 schedule_delayed_work(&expires_work,
3076 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3078 rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval +
3079 ip_rt_secret_interval;
3080 add_timer(&rt_secret_timer);
3082 if (ip_rt_proc_init())
3083 printk(KERN_ERR "Unable to create route proc files\n");
3088 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3093 EXPORT_SYMBOL(__ip_select_ident);
3094 EXPORT_SYMBOL(ip_route_input);
3095 EXPORT_SYMBOL(ip_route_output_key);