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>
73 #include <linux/sched.h>
75 #include <linux/bootmem.h>
76 #include <linux/string.h>
77 #include <linux/socket.h>
78 #include <linux/sockios.h>
79 #include <linux/errno.h>
81 #include <linux/inet.h>
82 #include <linux/netdevice.h>
83 #include <linux/proc_fs.h>
84 #include <linux/init.h>
85 #include <linux/skbuff.h>
86 #include <linux/rtnetlink.h>
87 #include <linux/inetdevice.h>
88 #include <linux/igmp.h>
89 #include <linux/pkt_sched.h>
90 #include <linux/mroute.h>
91 #include <linux/netfilter_ipv4.h>
92 #include <linux/random.h>
93 #include <linux/jhash.h>
94 #include <linux/rcupdate.h>
95 #include <linux/times.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/ip_mp_alg.h>
107 #include <net/netevent.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_min_delay = 2 * HZ;
120 static int ip_rt_max_delay = 10 * HZ;
121 static int ip_rt_max_size;
122 static int ip_rt_gc_timeout = RT_GC_TIMEOUT;
123 static int ip_rt_gc_interval = 60 * HZ;
124 static int ip_rt_gc_min_interval = HZ / 2;
125 static int ip_rt_redirect_number = 9;
126 static int ip_rt_redirect_load = HZ / 50;
127 static int ip_rt_redirect_silence = ((HZ / 50) << (9 + 1));
128 static int ip_rt_error_cost = HZ;
129 static int ip_rt_error_burst = 5 * HZ;
130 static int ip_rt_gc_elasticity = 8;
131 static int ip_rt_mtu_expires = 10 * 60 * HZ;
132 static int ip_rt_min_pmtu = 512 + 20 + 20;
133 static int ip_rt_min_advmss = 256;
134 static int ip_rt_secret_interval = 10 * 60 * HZ;
135 static unsigned long rt_deadline;
137 #define RTprint(a...) printk(KERN_DEBUG a)
139 static struct timer_list rt_flush_timer;
140 static struct timer_list rt_periodic_timer;
141 static struct timer_list rt_secret_timer;
144 * Interface to generic destination cache.
147 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
149 static void ipv4_dst_ifdown(struct dst_entry *dst,
150 struct net_device *dev, int how);
151 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
152 static void ipv4_link_failure(struct sk_buff *skb);
153 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
154 static int rt_garbage_collect(void);
157 static struct dst_ops ipv4_dst_ops = {
159 .protocol = __constant_htons(ETH_P_IP),
160 .gc = rt_garbage_collect,
161 .check = ipv4_dst_check,
162 .destroy = ipv4_dst_destroy,
163 .ifdown = ipv4_dst_ifdown,
164 .negative_advice = ipv4_negative_advice,
165 .link_failure = ipv4_link_failure,
166 .update_pmtu = ip_rt_update_pmtu,
167 .entry_size = sizeof(struct rtable),
170 #define ECN_OR_COST(class) TC_PRIO_##class
172 __u8 ip_tos2prio[16] = {
176 ECN_OR_COST(BESTEFFORT),
182 ECN_OR_COST(INTERACTIVE),
184 ECN_OR_COST(INTERACTIVE),
185 TC_PRIO_INTERACTIVE_BULK,
186 ECN_OR_COST(INTERACTIVE_BULK),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK)
196 /* The locking scheme is rather straight forward:
198 * 1) Read-Copy Update protects the buckets of the central route hash.
199 * 2) Only writers remove entries, and they hold the lock
200 * as they look at rtable reference counts.
201 * 3) Only readers acquire references to rtable entries,
202 * they do so with atomic increments and with the
206 struct rt_hash_bucket {
207 struct rtable *chain;
209 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
210 defined(CONFIG_PROVE_LOCKING)
212 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
213 * The size of this table is a power of two and depends on the number of CPUS.
214 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
216 #ifdef CONFIG_LOCKDEP
217 # define RT_HASH_LOCK_SZ 256
220 # define RT_HASH_LOCK_SZ 4096
222 # define RT_HASH_LOCK_SZ 2048
224 # define RT_HASH_LOCK_SZ 1024
226 # define RT_HASH_LOCK_SZ 512
228 # define RT_HASH_LOCK_SZ 256
232 static spinlock_t *rt_hash_locks;
233 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
234 # define rt_hash_lock_init() { \
236 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ, GFP_KERNEL); \
237 if (!rt_hash_locks) panic("IP: failed to allocate rt_hash_locks\n"); \
238 for (i = 0; i < RT_HASH_LOCK_SZ; i++) \
239 spin_lock_init(&rt_hash_locks[i]); \
242 # define rt_hash_lock_addr(slot) NULL
243 # define rt_hash_lock_init()
246 static struct rt_hash_bucket *rt_hash_table;
247 static unsigned rt_hash_mask;
248 static int rt_hash_log;
249 static unsigned int rt_hash_rnd;
251 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
252 #define RT_CACHE_STAT_INC(field) \
253 (__raw_get_cpu_var(rt_cache_stat).field++)
255 static int rt_intern_hash(unsigned hash, struct rtable *rth,
256 struct rtable **res);
258 static unsigned int rt_hash_code(u32 daddr, u32 saddr)
260 return (jhash_2words(daddr, saddr, rt_hash_rnd)
264 #define rt_hash(daddr, saddr, idx) \
265 rt_hash_code((__force u32)(__be32)(daddr),\
266 (__force u32)(__be32)(saddr) ^ ((idx) << 5))
268 #ifdef CONFIG_PROC_FS
269 struct rt_cache_iter_state {
273 static struct rtable *rt_cache_get_first(struct seq_file *seq)
275 struct rtable *r = NULL;
276 struct rt_cache_iter_state *st = seq->private;
278 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
280 r = rt_hash_table[st->bucket].chain;
283 rcu_read_unlock_bh();
288 static struct rtable *rt_cache_get_next(struct seq_file *seq, struct rtable *r)
290 struct rt_cache_iter_state *st = rcu_dereference(seq->private);
294 rcu_read_unlock_bh();
295 if (--st->bucket < 0)
298 r = rt_hash_table[st->bucket].chain;
303 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
305 struct rtable *r = rt_cache_get_first(seq);
308 while (pos && (r = rt_cache_get_next(seq, r)))
310 return pos ? NULL : r;
313 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
315 return *pos ? rt_cache_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
318 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
320 struct rtable *r = NULL;
322 if (v == SEQ_START_TOKEN)
323 r = rt_cache_get_first(seq);
325 r = rt_cache_get_next(seq, v);
330 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
332 if (v && v != SEQ_START_TOKEN)
333 rcu_read_unlock_bh();
336 static int rt_cache_seq_show(struct seq_file *seq, void *v)
338 if (v == SEQ_START_TOKEN)
339 seq_printf(seq, "%-127s\n",
340 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
341 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
344 struct rtable *r = v;
347 sprintf(temp, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
348 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X",
349 r->u.dst.dev ? r->u.dst.dev->name : "*",
350 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
351 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
352 r->u.dst.__use, 0, (unsigned long)r->rt_src,
353 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
354 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
355 dst_metric(&r->u.dst, RTAX_WINDOW),
356 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
357 dst_metric(&r->u.dst, RTAX_RTTVAR)),
359 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
360 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
363 seq_printf(seq, "%-127s\n", temp);
368 static struct seq_operations rt_cache_seq_ops = {
369 .start = rt_cache_seq_start,
370 .next = rt_cache_seq_next,
371 .stop = rt_cache_seq_stop,
372 .show = rt_cache_seq_show,
375 static int rt_cache_seq_open(struct inode *inode, struct file *file)
377 struct seq_file *seq;
379 struct rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
383 rc = seq_open(file, &rt_cache_seq_ops);
386 seq = file->private_data;
388 memset(s, 0, sizeof(*s));
396 static struct file_operations rt_cache_seq_fops = {
397 .owner = THIS_MODULE,
398 .open = rt_cache_seq_open,
401 .release = seq_release_private,
405 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
410 return SEQ_START_TOKEN;
412 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
413 if (!cpu_possible(cpu))
416 return &per_cpu(rt_cache_stat, cpu);
421 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
425 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
426 if (!cpu_possible(cpu))
429 return &per_cpu(rt_cache_stat, cpu);
435 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
440 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
442 struct rt_cache_stat *st = v;
444 if (v == SEQ_START_TOKEN) {
445 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");
449 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
450 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
451 atomic_read(&ipv4_dst_ops.entries),
474 static struct seq_operations rt_cpu_seq_ops = {
475 .start = rt_cpu_seq_start,
476 .next = rt_cpu_seq_next,
477 .stop = rt_cpu_seq_stop,
478 .show = rt_cpu_seq_show,
482 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
484 return seq_open(file, &rt_cpu_seq_ops);
487 static struct file_operations rt_cpu_seq_fops = {
488 .owner = THIS_MODULE,
489 .open = rt_cpu_seq_open,
492 .release = seq_release,
495 #endif /* CONFIG_PROC_FS */
497 static __inline__ void rt_free(struct rtable *rt)
499 multipath_remove(rt);
500 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
503 static __inline__ void rt_drop(struct rtable *rt)
505 multipath_remove(rt);
507 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
510 static __inline__ int rt_fast_clean(struct rtable *rth)
512 /* Kill broadcast/multicast entries very aggresively, if they
513 collide in hash table with more useful entries */
514 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
515 rth->fl.iif && rth->u.rt_next;
518 static __inline__ int rt_valuable(struct rtable *rth)
520 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
524 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
529 if (atomic_read(&rth->u.dst.__refcnt))
533 if (rth->u.dst.expires &&
534 time_after_eq(jiffies, rth->u.dst.expires))
537 age = jiffies - rth->u.dst.lastuse;
539 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
540 (age <= tmo2 && rt_valuable(rth)))
546 /* Bits of score are:
548 * 30: not quite useless
549 * 29..0: usage counter
551 static inline u32 rt_score(struct rtable *rt)
553 u32 score = jiffies - rt->u.dst.lastuse;
555 score = ~score & ~(3<<30);
561 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
567 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
569 return memcmp(&fl1->nl_u.ip4_u, &fl2->nl_u.ip4_u, sizeof(fl1->nl_u.ip4_u)) == 0 &&
570 fl1->oif == fl2->oif &&
571 fl1->iif == fl2->iif;
574 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
575 static struct rtable **rt_remove_balanced_route(struct rtable **chain_head,
576 struct rtable *expentry,
579 int passedexpired = 0;
580 struct rtable **nextstep = NULL;
581 struct rtable **rthp = chain_head;
587 while ((rth = *rthp) != NULL) {
591 if (((*rthp)->u.dst.flags & DST_BALANCED) != 0 &&
592 compare_keys(&(*rthp)->fl, &expentry->fl)) {
593 if (*rthp == expentry) {
594 *rthp = rth->u.rt_next;
597 *rthp = rth->u.rt_next;
603 if (!((*rthp)->u.dst.flags & DST_BALANCED) &&
604 passedexpired && !nextstep)
605 nextstep = &rth->u.rt_next;
607 rthp = &rth->u.rt_next;
617 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
620 /* This runs via a timer and thus is always in BH context. */
621 static void rt_check_expire(unsigned long dummy)
623 static unsigned int rover;
624 unsigned int i = rover, goal;
625 struct rtable *rth, **rthp;
626 unsigned long now = jiffies;
629 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
630 if (ip_rt_gc_timeout > 1)
631 do_div(mult, ip_rt_gc_timeout);
632 goal = (unsigned int)mult;
633 if (goal > rt_hash_mask) goal = rt_hash_mask + 1;
634 for (; goal > 0; goal--) {
635 unsigned long tmo = ip_rt_gc_timeout;
637 i = (i + 1) & rt_hash_mask;
638 rthp = &rt_hash_table[i].chain;
642 spin_lock(rt_hash_lock_addr(i));
643 while ((rth = *rthp) != NULL) {
644 if (rth->u.dst.expires) {
645 /* Entry is expired even if it is in use */
646 if (time_before_eq(now, rth->u.dst.expires)) {
648 rthp = &rth->u.rt_next;
651 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
653 rthp = &rth->u.rt_next;
657 /* Cleanup aged off entries. */
658 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
659 /* remove all related balanced entries if necessary */
660 if (rth->u.dst.flags & DST_BALANCED) {
661 rthp = rt_remove_balanced_route(
662 &rt_hash_table[i].chain,
667 *rthp = rth->u.rt_next;
670 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
671 *rthp = rth->u.rt_next;
673 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
675 spin_unlock(rt_hash_lock_addr(i));
677 /* Fallback loop breaker. */
678 if (time_after(jiffies, now))
682 mod_timer(&rt_periodic_timer, jiffies + ip_rt_gc_interval);
685 /* This can run from both BH and non-BH contexts, the latter
686 * in the case of a forced flush event.
688 static void rt_run_flush(unsigned long dummy)
691 struct rtable *rth, *next;
695 get_random_bytes(&rt_hash_rnd, 4);
697 for (i = rt_hash_mask; i >= 0; i--) {
698 spin_lock_bh(rt_hash_lock_addr(i));
699 rth = rt_hash_table[i].chain;
701 rt_hash_table[i].chain = NULL;
702 spin_unlock_bh(rt_hash_lock_addr(i));
704 for (; rth; rth = next) {
705 next = rth->u.rt_next;
711 static DEFINE_SPINLOCK(rt_flush_lock);
713 void rt_cache_flush(int delay)
715 unsigned long now = jiffies;
716 int user_mode = !in_softirq();
719 delay = ip_rt_min_delay;
721 /* flush existing multipath state*/
724 spin_lock_bh(&rt_flush_lock);
726 if (del_timer(&rt_flush_timer) && delay > 0 && rt_deadline) {
727 long tmo = (long)(rt_deadline - now);
729 /* If flush timer is already running
730 and flush request is not immediate (delay > 0):
732 if deadline is not achieved, prolongate timer to "delay",
733 otherwise fire it at deadline time.
736 if (user_mode && tmo < ip_rt_max_delay-ip_rt_min_delay)
744 spin_unlock_bh(&rt_flush_lock);
749 if (rt_deadline == 0)
750 rt_deadline = now + ip_rt_max_delay;
752 mod_timer(&rt_flush_timer, now+delay);
753 spin_unlock_bh(&rt_flush_lock);
756 static void rt_secret_rebuild(unsigned long dummy)
758 unsigned long now = jiffies;
761 mod_timer(&rt_secret_timer, now + ip_rt_secret_interval);
765 Short description of GC goals.
767 We want to build algorithm, which will keep routing cache
768 at some equilibrium point, when number of aged off entries
769 is kept approximately equal to newly generated ones.
771 Current expiration strength is variable "expire".
772 We try to adjust it dynamically, so that if networking
773 is idle expires is large enough to keep enough of warm entries,
774 and when load increases it reduces to limit cache size.
777 static int rt_garbage_collect(void)
779 static unsigned long expire = RT_GC_TIMEOUT;
780 static unsigned long last_gc;
782 static int equilibrium;
783 struct rtable *rth, **rthp;
784 unsigned long now = jiffies;
788 * Garbage collection is pretty expensive,
789 * do not make it too frequently.
792 RT_CACHE_STAT_INC(gc_total);
794 if (now - last_gc < ip_rt_gc_min_interval &&
795 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
796 RT_CACHE_STAT_INC(gc_ignored);
800 /* Calculate number of entries, which we want to expire now. */
801 goal = atomic_read(&ipv4_dst_ops.entries) -
802 (ip_rt_gc_elasticity << rt_hash_log);
804 if (equilibrium < ipv4_dst_ops.gc_thresh)
805 equilibrium = ipv4_dst_ops.gc_thresh;
806 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
808 equilibrium += min_t(unsigned int, goal / 2, rt_hash_mask + 1);
809 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
812 /* We are in dangerous area. Try to reduce cache really
815 goal = max_t(unsigned int, goal / 2, rt_hash_mask + 1);
816 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
819 if (now - last_gc >= ip_rt_gc_min_interval)
830 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
831 unsigned long tmo = expire;
833 k = (k + 1) & rt_hash_mask;
834 rthp = &rt_hash_table[k].chain;
835 spin_lock_bh(rt_hash_lock_addr(k));
836 while ((rth = *rthp) != NULL) {
837 if (!rt_may_expire(rth, tmo, expire)) {
839 rthp = &rth->u.rt_next;
842 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
843 /* remove all related balanced entries
846 if (rth->u.dst.flags & DST_BALANCED) {
849 rthp = rt_remove_balanced_route(
850 &rt_hash_table[k].chain,
857 *rthp = rth->u.rt_next;
861 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
862 *rthp = rth->u.rt_next;
865 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
867 spin_unlock_bh(rt_hash_lock_addr(k));
876 /* Goal is not achieved. We stop process if:
878 - if expire reduced to zero. Otherwise, expire is halfed.
879 - if table is not full.
880 - if we are called from interrupt.
881 - jiffies check is just fallback/debug loop breaker.
882 We will not spin here for long time in any case.
885 RT_CACHE_STAT_INC(gc_goal_miss);
891 #if RT_CACHE_DEBUG >= 2
892 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
893 atomic_read(&ipv4_dst_ops.entries), goal, i);
896 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
898 } while (!in_softirq() && time_before_eq(jiffies, now));
900 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
903 printk(KERN_WARNING "dst cache overflow\n");
904 RT_CACHE_STAT_INC(gc_dst_overflow);
908 expire += ip_rt_gc_min_interval;
909 if (expire > ip_rt_gc_timeout ||
910 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
911 expire = ip_rt_gc_timeout;
912 #if RT_CACHE_DEBUG >= 2
913 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
914 atomic_read(&ipv4_dst_ops.entries), goal, rover);
919 static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
921 struct rtable *rth, **rthp;
923 struct rtable *cand, **candp;
926 int attempts = !in_softirq();
935 rthp = &rt_hash_table[hash].chain;
937 spin_lock_bh(rt_hash_lock_addr(hash));
938 while ((rth = *rthp) != NULL) {
939 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
940 if (!(rth->u.dst.flags & DST_BALANCED) &&
941 compare_keys(&rth->fl, &rt->fl)) {
943 if (compare_keys(&rth->fl, &rt->fl)) {
946 *rthp = rth->u.rt_next;
948 * Since lookup is lockfree, the deletion
949 * must be visible to another weakly ordered CPU before
950 * the insertion at the start of the hash chain.
952 rcu_assign_pointer(rth->u.rt_next,
953 rt_hash_table[hash].chain);
955 * Since lookup is lockfree, the update writes
956 * must be ordered for consistency on SMP.
958 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
961 dst_hold(&rth->u.dst);
962 rth->u.dst.lastuse = now;
963 spin_unlock_bh(rt_hash_lock_addr(hash));
970 if (!atomic_read(&rth->u.dst.__refcnt)) {
971 u32 score = rt_score(rth);
973 if (score <= min_score) {
982 rthp = &rth->u.rt_next;
986 /* ip_rt_gc_elasticity used to be average length of chain
987 * length, when exceeded gc becomes really aggressive.
989 * The second limit is less certain. At the moment it allows
990 * only 2 entries per bucket. We will see.
992 if (chain_length > ip_rt_gc_elasticity) {
993 *candp = cand->u.rt_next;
998 /* Try to bind route to arp only if it is output
999 route or unicast forwarding path.
1001 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1002 int err = arp_bind_neighbour(&rt->u.dst);
1004 spin_unlock_bh(rt_hash_lock_addr(hash));
1006 if (err != -ENOBUFS) {
1011 /* Neighbour tables are full and nothing
1012 can be released. Try to shrink route cache,
1013 it is most likely it holds some neighbour records.
1015 if (attempts-- > 0) {
1016 int saved_elasticity = ip_rt_gc_elasticity;
1017 int saved_int = ip_rt_gc_min_interval;
1018 ip_rt_gc_elasticity = 1;
1019 ip_rt_gc_min_interval = 0;
1020 rt_garbage_collect();
1021 ip_rt_gc_min_interval = saved_int;
1022 ip_rt_gc_elasticity = saved_elasticity;
1026 if (net_ratelimit())
1027 printk(KERN_WARNING "Neighbour table overflow.\n");
1033 rt->u.rt_next = rt_hash_table[hash].chain;
1034 #if RT_CACHE_DEBUG >= 2
1035 if (rt->u.rt_next) {
1037 printk(KERN_DEBUG "rt_cache @%02x: %u.%u.%u.%u", hash,
1038 NIPQUAD(rt->rt_dst));
1039 for (trt = rt->u.rt_next; trt; trt = trt->u.rt_next)
1040 printk(" . %u.%u.%u.%u", NIPQUAD(trt->rt_dst));
1044 rt_hash_table[hash].chain = rt;
1045 spin_unlock_bh(rt_hash_lock_addr(hash));
1050 void rt_bind_peer(struct rtable *rt, int create)
1052 static DEFINE_SPINLOCK(rt_peer_lock);
1053 struct inet_peer *peer;
1055 peer = inet_getpeer(rt->rt_dst, create);
1057 spin_lock_bh(&rt_peer_lock);
1058 if (rt->peer == NULL) {
1062 spin_unlock_bh(&rt_peer_lock);
1068 * Peer allocation may fail only in serious out-of-memory conditions. However
1069 * we still can generate some output.
1070 * Random ID selection looks a bit dangerous because we have no chances to
1071 * select ID being unique in a reasonable period of time.
1072 * But broken packet identifier may be better than no packet at all.
1074 static void ip_select_fb_ident(struct iphdr *iph)
1076 static DEFINE_SPINLOCK(ip_fb_id_lock);
1077 static u32 ip_fallback_id;
1080 spin_lock_bh(&ip_fb_id_lock);
1081 salt = secure_ip_id(ip_fallback_id ^ iph->daddr);
1082 iph->id = htons(salt & 0xFFFF);
1083 ip_fallback_id = salt;
1084 spin_unlock_bh(&ip_fb_id_lock);
1087 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1089 struct rtable *rt = (struct rtable *) dst;
1092 if (rt->peer == NULL)
1093 rt_bind_peer(rt, 1);
1095 /* If peer is attached to destination, it is never detached,
1096 so that we need not to grab a lock to dereference it.
1099 iph->id = htons(inet_getid(rt->peer, more));
1103 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1104 __builtin_return_address(0));
1106 ip_select_fb_ident(iph);
1109 static void rt_del(unsigned hash, struct rtable *rt)
1111 struct rtable **rthp;
1113 spin_lock_bh(rt_hash_lock_addr(hash));
1115 for (rthp = &rt_hash_table[hash].chain; *rthp;
1116 rthp = &(*rthp)->u.rt_next)
1118 *rthp = rt->u.rt_next;
1122 spin_unlock_bh(rt_hash_lock_addr(hash));
1125 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1126 __be32 saddr, struct net_device *dev)
1129 struct in_device *in_dev = in_dev_get(dev);
1130 struct rtable *rth, **rthp;
1131 __be32 skeys[2] = { saddr, 0 };
1132 int ikeys[2] = { dev->ifindex, 0 };
1133 struct netevent_redirect netevent;
1138 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1139 || MULTICAST(new_gw) || BADCLASS(new_gw) || ZERONET(new_gw))
1140 goto reject_redirect;
1142 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1143 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1144 goto reject_redirect;
1145 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1146 goto reject_redirect;
1148 if (inet_addr_type(new_gw) != RTN_UNICAST)
1149 goto reject_redirect;
1152 for (i = 0; i < 2; i++) {
1153 for (k = 0; k < 2; k++) {
1154 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k]);
1156 rthp=&rt_hash_table[hash].chain;
1159 while ((rth = rcu_dereference(*rthp)) != NULL) {
1162 if (rth->fl.fl4_dst != daddr ||
1163 rth->fl.fl4_src != skeys[i] ||
1164 rth->fl.oif != ikeys[k] ||
1166 rthp = &rth->u.rt_next;
1170 if (rth->rt_dst != daddr ||
1171 rth->rt_src != saddr ||
1173 rth->rt_gateway != old_gw ||
1174 rth->u.dst.dev != dev)
1177 dst_hold(&rth->u.dst);
1180 rt = dst_alloc(&ipv4_dst_ops);
1187 /* Copy all the information. */
1189 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1190 rt->u.dst.__use = 1;
1191 atomic_set(&rt->u.dst.__refcnt, 1);
1192 rt->u.dst.child = NULL;
1194 dev_hold(rt->u.dst.dev);
1196 in_dev_hold(rt->idev);
1197 rt->u.dst.obsolete = 0;
1198 rt->u.dst.lastuse = jiffies;
1199 rt->u.dst.path = &rt->u.dst;
1200 rt->u.dst.neighbour = NULL;
1201 rt->u.dst.hh = NULL;
1202 rt->u.dst.xfrm = NULL;
1204 rt->rt_flags |= RTCF_REDIRECTED;
1206 /* Gateway is different ... */
1207 rt->rt_gateway = new_gw;
1209 /* Redirect received -> path was valid */
1210 dst_confirm(&rth->u.dst);
1213 atomic_inc(&rt->peer->refcnt);
1215 if (arp_bind_neighbour(&rt->u.dst) ||
1216 !(rt->u.dst.neighbour->nud_state &
1218 if (rt->u.dst.neighbour)
1219 neigh_event_send(rt->u.dst.neighbour, NULL);
1225 netevent.old = &rth->u.dst;
1226 netevent.new = &rt->u.dst;
1227 call_netevent_notifiers(NETEVENT_REDIRECT,
1231 if (!rt_intern_hash(hash, rt, &rt))
1244 #ifdef CONFIG_IP_ROUTE_VERBOSE
1245 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1246 printk(KERN_INFO "Redirect from %u.%u.%u.%u on %s about "
1247 "%u.%u.%u.%u ignored.\n"
1248 " Advised path = %u.%u.%u.%u -> %u.%u.%u.%u\n",
1249 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
1250 NIPQUAD(saddr), NIPQUAD(daddr));
1255 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1257 struct rtable *rt = (struct rtable*)dst;
1258 struct dst_entry *ret = dst;
1261 if (dst->obsolete) {
1264 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1265 rt->u.dst.expires) {
1266 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1268 #if RT_CACHE_DEBUG >= 1
1269 printk(KERN_DEBUG "ip_rt_advice: redirect to "
1270 "%u.%u.%u.%u/%02x dropped\n",
1271 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1282 * 1. The first ip_rt_redirect_number redirects are sent
1283 * with exponential backoff, then we stop sending them at all,
1284 * assuming that the host ignores our redirects.
1285 * 2. If we did not see packets requiring redirects
1286 * during ip_rt_redirect_silence, we assume that the host
1287 * forgot redirected route and start to send redirects again.
1289 * This algorithm is much cheaper and more intelligent than dumb load limiting
1292 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1293 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1296 void ip_rt_send_redirect(struct sk_buff *skb)
1298 struct rtable *rt = (struct rtable*)skb->dst;
1299 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1304 if (!IN_DEV_TX_REDIRECTS(in_dev))
1307 /* No redirected packets during ip_rt_redirect_silence;
1308 * reset the algorithm.
1310 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1311 rt->u.dst.rate_tokens = 0;
1313 /* Too many ignored redirects; do not send anything
1314 * set u.dst.rate_last to the last seen redirected packet.
1316 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1317 rt->u.dst.rate_last = jiffies;
1321 /* Check for load limit; set rate_last to the latest sent
1324 if (time_after(jiffies,
1325 (rt->u.dst.rate_last +
1326 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1327 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1328 rt->u.dst.rate_last = jiffies;
1329 ++rt->u.dst.rate_tokens;
1330 #ifdef CONFIG_IP_ROUTE_VERBOSE
1331 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1332 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1334 printk(KERN_WARNING "host %u.%u.%u.%u/if%d ignores "
1335 "redirects for %u.%u.%u.%u to %u.%u.%u.%u.\n",
1336 NIPQUAD(rt->rt_src), rt->rt_iif,
1337 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1344 static int ip_error(struct sk_buff *skb)
1346 struct rtable *rt = (struct rtable*)skb->dst;
1350 switch (rt->u.dst.error) {
1355 code = ICMP_HOST_UNREACH;
1358 code = ICMP_NET_UNREACH;
1361 code = ICMP_PKT_FILTERED;
1366 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1367 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1368 rt->u.dst.rate_tokens = ip_rt_error_burst;
1369 rt->u.dst.rate_last = now;
1370 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1371 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1372 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1375 out: kfree_skb(skb);
1380 * The last two values are not from the RFC but
1381 * are needed for AMPRnet AX.25 paths.
1384 static const unsigned short mtu_plateau[] =
1385 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1387 static __inline__ unsigned short guess_mtu(unsigned short old_mtu)
1391 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1392 if (old_mtu > mtu_plateau[i])
1393 return mtu_plateau[i];
1397 unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu)
1400 unsigned short old_mtu = ntohs(iph->tot_len);
1402 u32 skeys[2] = { iph->saddr, 0, };
1403 u32 daddr = iph->daddr;
1404 unsigned short est_mtu = 0;
1406 if (ipv4_config.no_pmtu_disc)
1409 for (i = 0; i < 2; i++) {
1410 unsigned hash = rt_hash(daddr, skeys[i], 0);
1413 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1414 rth = rcu_dereference(rth->u.rt_next)) {
1415 if (rth->fl.fl4_dst == daddr &&
1416 rth->fl.fl4_src == skeys[i] &&
1417 rth->rt_dst == daddr &&
1418 rth->rt_src == iph->saddr &&
1420 !(dst_metric_locked(&rth->u.dst, RTAX_MTU))) {
1421 unsigned short mtu = new_mtu;
1423 if (new_mtu < 68 || new_mtu >= old_mtu) {
1425 /* BSD 4.2 compatibility hack :-( */
1427 old_mtu >= rth->u.dst.metrics[RTAX_MTU-1] &&
1428 old_mtu >= 68 + (iph->ihl << 2))
1429 old_mtu -= iph->ihl << 2;
1431 mtu = guess_mtu(old_mtu);
1433 if (mtu <= rth->u.dst.metrics[RTAX_MTU-1]) {
1434 if (mtu < rth->u.dst.metrics[RTAX_MTU-1]) {
1435 dst_confirm(&rth->u.dst);
1436 if (mtu < ip_rt_min_pmtu) {
1437 mtu = ip_rt_min_pmtu;
1438 rth->u.dst.metrics[RTAX_LOCK-1] |=
1441 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1442 dst_set_expires(&rth->u.dst,
1451 return est_mtu ? : new_mtu;
1454 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1456 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= 68 &&
1457 !(dst_metric_locked(dst, RTAX_MTU))) {
1458 if (mtu < ip_rt_min_pmtu) {
1459 mtu = ip_rt_min_pmtu;
1460 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1462 dst->metrics[RTAX_MTU-1] = mtu;
1463 dst_set_expires(dst, ip_rt_mtu_expires);
1464 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1468 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1473 static void ipv4_dst_destroy(struct dst_entry *dst)
1475 struct rtable *rt = (struct rtable *) dst;
1476 struct inet_peer *peer = rt->peer;
1477 struct in_device *idev = rt->idev;
1490 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1493 struct rtable *rt = (struct rtable *) dst;
1494 struct in_device *idev = rt->idev;
1495 if (dev != &loopback_dev && idev && idev->dev == dev) {
1496 struct in_device *loopback_idev = in_dev_get(&loopback_dev);
1497 if (loopback_idev) {
1498 rt->idev = loopback_idev;
1504 static void ipv4_link_failure(struct sk_buff *skb)
1508 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1510 rt = (struct rtable *) skb->dst;
1512 dst_set_expires(&rt->u.dst, 0);
1515 static int ip_rt_bug(struct sk_buff *skb)
1517 printk(KERN_DEBUG "ip_rt_bug: %u.%u.%u.%u -> %u.%u.%u.%u, %s\n",
1518 NIPQUAD(skb->nh.iph->saddr), NIPQUAD(skb->nh.iph->daddr),
1519 skb->dev ? skb->dev->name : "?");
1525 We do not cache source address of outgoing interface,
1526 because it is used only by IP RR, TS and SRR options,
1527 so that it out of fast path.
1529 BTW remember: "addr" is allowed to be not aligned
1533 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1536 struct fib_result res;
1538 if (rt->fl.iif == 0)
1540 else if (fib_lookup(&rt->fl, &res) == 0) {
1541 src = FIB_RES_PREFSRC(res);
1544 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1546 memcpy(addr, &src, 4);
1549 #ifdef CONFIG_NET_CLS_ROUTE
1550 static void set_class_tag(struct rtable *rt, u32 tag)
1552 if (!(rt->u.dst.tclassid & 0xFFFF))
1553 rt->u.dst.tclassid |= tag & 0xFFFF;
1554 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1555 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1559 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1561 struct fib_info *fi = res->fi;
1564 if (FIB_RES_GW(*res) &&
1565 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1566 rt->rt_gateway = FIB_RES_GW(*res);
1567 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1568 sizeof(rt->u.dst.metrics));
1569 if (fi->fib_mtu == 0) {
1570 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1571 if (rt->u.dst.metrics[RTAX_LOCK-1] & (1 << RTAX_MTU) &&
1572 rt->rt_gateway != rt->rt_dst &&
1573 rt->u.dst.dev->mtu > 576)
1574 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1576 #ifdef CONFIG_NET_CLS_ROUTE
1577 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1580 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1582 if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0)
1583 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1584 if (rt->u.dst.metrics[RTAX_MTU-1] > IP_MAX_MTU)
1585 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1586 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0)
1587 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1589 if (rt->u.dst.metrics[RTAX_ADVMSS-1] > 65535 - 40)
1590 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1592 #ifdef CONFIG_NET_CLS_ROUTE
1593 #ifdef CONFIG_IP_MULTIPLE_TABLES
1594 set_class_tag(rt, fib_rules_tclass(res));
1596 set_class_tag(rt, itag);
1598 rt->rt_type = res->type;
1601 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1602 u8 tos, struct net_device *dev, int our)
1607 struct in_device *in_dev = in_dev_get(dev);
1610 /* Primary sanity checks. */
1615 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr) ||
1616 skb->protocol != htons(ETH_P_IP))
1619 if (ZERONET(saddr)) {
1620 if (!LOCAL_MCAST(daddr))
1622 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1623 } else if (fib_validate_source(saddr, 0, tos, 0,
1624 dev, &spec_dst, &itag) < 0)
1627 rth = dst_alloc(&ipv4_dst_ops);
1631 rth->u.dst.output= ip_rt_bug;
1633 atomic_set(&rth->u.dst.__refcnt, 1);
1634 rth->u.dst.flags= DST_HOST;
1635 if (in_dev->cnf.no_policy)
1636 rth->u.dst.flags |= DST_NOPOLICY;
1637 rth->fl.fl4_dst = daddr;
1638 rth->rt_dst = daddr;
1639 rth->fl.fl4_tos = tos;
1640 #ifdef CONFIG_IP_ROUTE_FWMARK
1641 rth->fl.fl4_fwmark= skb->nfmark;
1643 rth->fl.fl4_src = saddr;
1644 rth->rt_src = saddr;
1645 #ifdef CONFIG_NET_CLS_ROUTE
1646 rth->u.dst.tclassid = itag;
1649 rth->fl.iif = dev->ifindex;
1650 rth->u.dst.dev = &loopback_dev;
1651 dev_hold(rth->u.dst.dev);
1652 rth->idev = in_dev_get(rth->u.dst.dev);
1654 rth->rt_gateway = daddr;
1655 rth->rt_spec_dst= spec_dst;
1656 rth->rt_type = RTN_MULTICAST;
1657 rth->rt_flags = RTCF_MULTICAST;
1659 rth->u.dst.input= ip_local_deliver;
1660 rth->rt_flags |= RTCF_LOCAL;
1663 #ifdef CONFIG_IP_MROUTE
1664 if (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
1665 rth->u.dst.input = ip_mr_input;
1667 RT_CACHE_STAT_INC(in_slow_mc);
1670 hash = rt_hash(daddr, saddr, dev->ifindex);
1671 return rt_intern_hash(hash, rth, (struct rtable**) &skb->dst);
1683 static void ip_handle_martian_source(struct net_device *dev,
1684 struct in_device *in_dev,
1685 struct sk_buff *skb,
1689 RT_CACHE_STAT_INC(in_martian_src);
1690 #ifdef CONFIG_IP_ROUTE_VERBOSE
1691 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1693 * RFC1812 recommendation, if source is martian,
1694 * the only hint is MAC header.
1696 printk(KERN_WARNING "martian source %u.%u.%u.%u from "
1697 "%u.%u.%u.%u, on dev %s\n",
1698 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1699 if (dev->hard_header_len && skb->mac.raw) {
1701 unsigned char *p = skb->mac.raw;
1702 printk(KERN_WARNING "ll header: ");
1703 for (i = 0; i < dev->hard_header_len; i++, p++) {
1705 if (i < (dev->hard_header_len - 1))
1714 static inline int __mkroute_input(struct sk_buff *skb,
1715 struct fib_result* res,
1716 struct in_device *in_dev,
1717 __be32 daddr, __be32 saddr, u32 tos,
1718 struct rtable **result)
1723 struct in_device *out_dev;
1727 /* get a working reference to the output device */
1728 out_dev = in_dev_get(FIB_RES_DEV(*res));
1729 if (out_dev == NULL) {
1730 if (net_ratelimit())
1731 printk(KERN_CRIT "Bug in ip_route_input" \
1732 "_slow(). Please, report\n");
1737 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1738 in_dev->dev, &spec_dst, &itag);
1740 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1748 flags |= RTCF_DIRECTSRC;
1750 if (out_dev == in_dev && err && !(flags & (RTCF_NAT | RTCF_MASQ)) &&
1751 (IN_DEV_SHARED_MEDIA(out_dev) ||
1752 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1753 flags |= RTCF_DOREDIRECT;
1755 if (skb->protocol != htons(ETH_P_IP)) {
1756 /* Not IP (i.e. ARP). Do not create route, if it is
1757 * invalid for proxy arp. DNAT routes are always valid.
1759 if (out_dev == in_dev && !(flags & RTCF_DNAT)) {
1766 rth = dst_alloc(&ipv4_dst_ops);
1772 atomic_set(&rth->u.dst.__refcnt, 1);
1773 rth->u.dst.flags= DST_HOST;
1774 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
1775 if (res->fi->fib_nhs > 1)
1776 rth->u.dst.flags |= DST_BALANCED;
1778 if (in_dev->cnf.no_policy)
1779 rth->u.dst.flags |= DST_NOPOLICY;
1780 if (in_dev->cnf.no_xfrm)
1781 rth->u.dst.flags |= DST_NOXFRM;
1782 rth->fl.fl4_dst = daddr;
1783 rth->rt_dst = daddr;
1784 rth->fl.fl4_tos = tos;
1785 #ifdef CONFIG_IP_ROUTE_FWMARK
1786 rth->fl.fl4_fwmark= skb->nfmark;
1788 rth->fl.fl4_src = saddr;
1789 rth->rt_src = saddr;
1790 rth->rt_gateway = daddr;
1792 rth->fl.iif = in_dev->dev->ifindex;
1793 rth->u.dst.dev = (out_dev)->dev;
1794 dev_hold(rth->u.dst.dev);
1795 rth->idev = in_dev_get(rth->u.dst.dev);
1797 rth->rt_spec_dst= spec_dst;
1799 rth->u.dst.input = ip_forward;
1800 rth->u.dst.output = ip_output;
1802 rt_set_nexthop(rth, res, itag);
1804 rth->rt_flags = flags;
1809 /* release the working reference to the output device */
1810 in_dev_put(out_dev);
1814 static inline int ip_mkroute_input_def(struct sk_buff *skb,
1815 struct fib_result* res,
1816 const struct flowi *fl,
1817 struct in_device *in_dev,
1818 __be32 daddr, __be32 saddr, u32 tos)
1820 struct rtable* rth = NULL;
1824 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1825 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1826 fib_select_multipath(fl, res);
1829 /* create a routing cache entry */
1830 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1834 /* put it into the cache */
1835 hash = rt_hash(daddr, saddr, fl->iif);
1836 return rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
1839 static inline int ip_mkroute_input(struct sk_buff *skb,
1840 struct fib_result* res,
1841 const struct flowi *fl,
1842 struct in_device *in_dev,
1843 __be32 daddr, __be32 saddr, u32 tos)
1845 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
1846 struct rtable* rth = NULL, *rtres;
1847 unsigned char hop, hopcount;
1852 hopcount = res->fi->fib_nhs;
1856 /* distinguish between multipath and singlepath */
1858 return ip_mkroute_input_def(skb, res, fl, in_dev, daddr,
1861 /* add all alternatives to the routing cache */
1862 for (hop = 0; hop < hopcount; hop++) {
1865 /* put reference to previous result */
1869 /* create a routing cache entry */
1870 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos,
1875 /* put it into the cache */
1876 hash = rt_hash(daddr, saddr, fl->iif);
1877 err = rt_intern_hash(hash, rth, &rtres);
1881 /* forward hop information to multipath impl. */
1882 multipath_set_nhinfo(rth,
1883 FIB_RES_NETWORK(*res),
1884 FIB_RES_NETMASK(*res),
1888 skb->dst = &rtres->u.dst;
1890 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
1891 return ip_mkroute_input_def(skb, res, fl, in_dev, daddr, saddr, tos);
1892 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
1897 * NOTE. We drop all the packets that has local source
1898 * addresses, because every properly looped back packet
1899 * must have correct destination already attached by output routine.
1901 * Such approach solves two big problems:
1902 * 1. Not simplex devices are handled properly.
1903 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1906 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1907 u8 tos, struct net_device *dev)
1909 struct fib_result res;
1910 struct in_device *in_dev = in_dev_get(dev);
1911 struct flowi fl = { .nl_u = { .ip4_u =
1915 .scope = RT_SCOPE_UNIVERSE,
1916 #ifdef CONFIG_IP_ROUTE_FWMARK
1917 .fwmark = skb->nfmark
1920 .iif = dev->ifindex };
1923 struct rtable * rth;
1929 /* IP on this device is disabled. */
1934 /* Check for the most weird martians, which can be not detected
1938 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr))
1939 goto martian_source;
1941 if (daddr == 0xFFFFFFFF || (saddr == 0 && daddr == 0))
1944 /* Accept zero addresses only to limited broadcast;
1945 * I even do not know to fix it or not. Waiting for complains :-)
1948 goto martian_source;
1950 if (BADCLASS(daddr) || ZERONET(daddr) || LOOPBACK(daddr))
1951 goto martian_destination;
1954 * Now we are ready to route packet.
1956 if ((err = fib_lookup(&fl, &res)) != 0) {
1957 if (!IN_DEV_FORWARD(in_dev))
1963 RT_CACHE_STAT_INC(in_slow_tot);
1965 if (res.type == RTN_BROADCAST)
1968 if (res.type == RTN_LOCAL) {
1970 result = fib_validate_source(saddr, daddr, tos,
1971 loopback_dev.ifindex,
1972 dev, &spec_dst, &itag);
1974 goto martian_source;
1976 flags |= RTCF_DIRECTSRC;
1981 if (!IN_DEV_FORWARD(in_dev))
1983 if (res.type != RTN_UNICAST)
1984 goto martian_destination;
1986 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1987 if (err == -ENOBUFS)
1999 if (skb->protocol != htons(ETH_P_IP))
2003 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2005 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2008 goto martian_source;
2010 flags |= RTCF_DIRECTSRC;
2012 flags |= RTCF_BROADCAST;
2013 res.type = RTN_BROADCAST;
2014 RT_CACHE_STAT_INC(in_brd);
2017 rth = dst_alloc(&ipv4_dst_ops);
2021 rth->u.dst.output= ip_rt_bug;
2023 atomic_set(&rth->u.dst.__refcnt, 1);
2024 rth->u.dst.flags= DST_HOST;
2025 if (in_dev->cnf.no_policy)
2026 rth->u.dst.flags |= DST_NOPOLICY;
2027 rth->fl.fl4_dst = daddr;
2028 rth->rt_dst = daddr;
2029 rth->fl.fl4_tos = tos;
2030 #ifdef CONFIG_IP_ROUTE_FWMARK
2031 rth->fl.fl4_fwmark= skb->nfmark;
2033 rth->fl.fl4_src = saddr;
2034 rth->rt_src = saddr;
2035 #ifdef CONFIG_NET_CLS_ROUTE
2036 rth->u.dst.tclassid = itag;
2039 rth->fl.iif = dev->ifindex;
2040 rth->u.dst.dev = &loopback_dev;
2041 dev_hold(rth->u.dst.dev);
2042 rth->idev = in_dev_get(rth->u.dst.dev);
2043 rth->rt_gateway = daddr;
2044 rth->rt_spec_dst= spec_dst;
2045 rth->u.dst.input= ip_local_deliver;
2046 rth->rt_flags = flags|RTCF_LOCAL;
2047 if (res.type == RTN_UNREACHABLE) {
2048 rth->u.dst.input= ip_error;
2049 rth->u.dst.error= -err;
2050 rth->rt_flags &= ~RTCF_LOCAL;
2052 rth->rt_type = res.type;
2053 hash = rt_hash(daddr, saddr, fl.iif);
2054 err = rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
2058 RT_CACHE_STAT_INC(in_no_route);
2059 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2060 res.type = RTN_UNREACHABLE;
2064 * Do not cache martian addresses: they should be logged (RFC1812)
2066 martian_destination:
2067 RT_CACHE_STAT_INC(in_martian_dst);
2068 #ifdef CONFIG_IP_ROUTE_VERBOSE
2069 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2070 printk(KERN_WARNING "martian destination %u.%u.%u.%u from "
2071 "%u.%u.%u.%u, dev %s\n",
2072 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
2076 err = -EHOSTUNREACH;
2088 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2092 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2093 u8 tos, struct net_device *dev)
2095 struct rtable * rth;
2097 int iif = dev->ifindex;
2099 tos &= IPTOS_RT_MASK;
2100 hash = rt_hash(daddr, saddr, iif);
2103 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2104 rth = rcu_dereference(rth->u.rt_next)) {
2105 if (rth->fl.fl4_dst == daddr &&
2106 rth->fl.fl4_src == saddr &&
2107 rth->fl.iif == iif &&
2109 #ifdef CONFIG_IP_ROUTE_FWMARK
2110 rth->fl.fl4_fwmark == skb->nfmark &&
2112 rth->fl.fl4_tos == tos) {
2113 rth->u.dst.lastuse = jiffies;
2114 dst_hold(&rth->u.dst);
2116 RT_CACHE_STAT_INC(in_hit);
2118 skb->dst = (struct dst_entry*)rth;
2121 RT_CACHE_STAT_INC(in_hlist_search);
2125 /* Multicast recognition logic is moved from route cache to here.
2126 The problem was that too many Ethernet cards have broken/missing
2127 hardware multicast filters :-( As result the host on multicasting
2128 network acquires a lot of useless route cache entries, sort of
2129 SDR messages from all the world. Now we try to get rid of them.
2130 Really, provided software IP multicast filter is organized
2131 reasonably (at least, hashed), it does not result in a slowdown
2132 comparing with route cache reject entries.
2133 Note, that multicast routers are not affected, because
2134 route cache entry is created eventually.
2136 if (MULTICAST(daddr)) {
2137 struct in_device *in_dev;
2140 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2141 int our = ip_check_mc(in_dev, daddr, saddr,
2142 skb->nh.iph->protocol);
2144 #ifdef CONFIG_IP_MROUTE
2145 || (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
2149 return ip_route_input_mc(skb, daddr, saddr,
2156 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2159 static inline int __mkroute_output(struct rtable **result,
2160 struct fib_result* res,
2161 const struct flowi *fl,
2162 const struct flowi *oldflp,
2163 struct net_device *dev_out,
2167 struct in_device *in_dev;
2168 u32 tos = RT_FL_TOS(oldflp);
2171 if (LOOPBACK(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2174 if (fl->fl4_dst == 0xFFFFFFFF)
2175 res->type = RTN_BROADCAST;
2176 else if (MULTICAST(fl->fl4_dst))
2177 res->type = RTN_MULTICAST;
2178 else if (BADCLASS(fl->fl4_dst) || ZERONET(fl->fl4_dst))
2181 if (dev_out->flags & IFF_LOOPBACK)
2182 flags |= RTCF_LOCAL;
2184 /* get work reference to inet device */
2185 in_dev = in_dev_get(dev_out);
2189 if (res->type == RTN_BROADCAST) {
2190 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2192 fib_info_put(res->fi);
2195 } else if (res->type == RTN_MULTICAST) {
2196 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2197 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2199 flags &= ~RTCF_LOCAL;
2200 /* If multicast route do not exist use
2201 default one, but do not gateway in this case.
2204 if (res->fi && res->prefixlen < 4) {
2205 fib_info_put(res->fi);
2211 rth = dst_alloc(&ipv4_dst_ops);
2217 atomic_set(&rth->u.dst.__refcnt, 1);
2218 rth->u.dst.flags= DST_HOST;
2219 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2221 rth->rt_multipath_alg = res->fi->fib_mp_alg;
2222 if (res->fi->fib_nhs > 1)
2223 rth->u.dst.flags |= DST_BALANCED;
2226 if (in_dev->cnf.no_xfrm)
2227 rth->u.dst.flags |= DST_NOXFRM;
2228 if (in_dev->cnf.no_policy)
2229 rth->u.dst.flags |= DST_NOPOLICY;
2231 rth->fl.fl4_dst = oldflp->fl4_dst;
2232 rth->fl.fl4_tos = tos;
2233 rth->fl.fl4_src = oldflp->fl4_src;
2234 rth->fl.oif = oldflp->oif;
2235 #ifdef CONFIG_IP_ROUTE_FWMARK
2236 rth->fl.fl4_fwmark= oldflp->fl4_fwmark;
2238 rth->rt_dst = fl->fl4_dst;
2239 rth->rt_src = fl->fl4_src;
2240 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2241 /* get references to the devices that are to be hold by the routing
2243 rth->u.dst.dev = dev_out;
2245 rth->idev = in_dev_get(dev_out);
2246 rth->rt_gateway = fl->fl4_dst;
2247 rth->rt_spec_dst= fl->fl4_src;
2249 rth->u.dst.output=ip_output;
2251 RT_CACHE_STAT_INC(out_slow_tot);
2253 if (flags & RTCF_LOCAL) {
2254 rth->u.dst.input = ip_local_deliver;
2255 rth->rt_spec_dst = fl->fl4_dst;
2257 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2258 rth->rt_spec_dst = fl->fl4_src;
2259 if (flags & RTCF_LOCAL &&
2260 !(dev_out->flags & IFF_LOOPBACK)) {
2261 rth->u.dst.output = ip_mc_output;
2262 RT_CACHE_STAT_INC(out_slow_mc);
2264 #ifdef CONFIG_IP_MROUTE
2265 if (res->type == RTN_MULTICAST) {
2266 if (IN_DEV_MFORWARD(in_dev) &&
2267 !LOCAL_MCAST(oldflp->fl4_dst)) {
2268 rth->u.dst.input = ip_mr_input;
2269 rth->u.dst.output = ip_mc_output;
2275 rt_set_nexthop(rth, res, 0);
2277 rth->rt_flags = flags;
2281 /* release work reference to inet device */
2287 static inline int ip_mkroute_output_def(struct rtable **rp,
2288 struct fib_result* res,
2289 const struct flowi *fl,
2290 const struct flowi *oldflp,
2291 struct net_device *dev_out,
2294 struct rtable *rth = NULL;
2295 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2298 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif);
2299 err = rt_intern_hash(hash, rth, rp);
2305 static inline int ip_mkroute_output(struct rtable** rp,
2306 struct fib_result* res,
2307 const struct flowi *fl,
2308 const struct flowi *oldflp,
2309 struct net_device *dev_out,
2312 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2316 struct rtable *rth = NULL;
2318 if (res->fi && res->fi->fib_nhs > 1) {
2319 unsigned char hopcount = res->fi->fib_nhs;
2321 for (hop = 0; hop < hopcount; hop++) {
2322 struct net_device *dev2nexthop;
2326 /* hold a work reference to the output device */
2327 dev2nexthop = FIB_RES_DEV(*res);
2328 dev_hold(dev2nexthop);
2330 /* put reference to previous result */
2334 err = __mkroute_output(&rth, res, fl, oldflp,
2335 dev2nexthop, flags);
2340 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src,
2342 err = rt_intern_hash(hash, rth, rp);
2344 /* forward hop information to multipath impl. */
2345 multipath_set_nhinfo(rth,
2346 FIB_RES_NETWORK(*res),
2347 FIB_RES_NETMASK(*res),
2351 /* release work reference to output device */
2352 dev_put(dev2nexthop);
2359 return ip_mkroute_output_def(rp, res, fl, oldflp, dev_out,
2362 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
2363 return ip_mkroute_output_def(rp, res, fl, oldflp, dev_out, flags);
2368 * Major route resolver routine.
2371 static int ip_route_output_slow(struct rtable **rp, const struct flowi *oldflp)
2373 u32 tos = RT_FL_TOS(oldflp);
2374 struct flowi fl = { .nl_u = { .ip4_u =
2375 { .daddr = oldflp->fl4_dst,
2376 .saddr = oldflp->fl4_src,
2377 .tos = tos & IPTOS_RT_MASK,
2378 .scope = ((tos & RTO_ONLINK) ?
2381 #ifdef CONFIG_IP_ROUTE_FWMARK
2382 .fwmark = oldflp->fl4_fwmark
2385 .iif = loopback_dev.ifindex,
2386 .oif = oldflp->oif };
2387 struct fib_result res;
2389 struct net_device *dev_out = NULL;
2395 #ifdef CONFIG_IP_MULTIPLE_TABLES
2399 if (oldflp->fl4_src) {
2401 if (MULTICAST(oldflp->fl4_src) ||
2402 BADCLASS(oldflp->fl4_src) ||
2403 ZERONET(oldflp->fl4_src))
2406 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2407 dev_out = ip_dev_find(oldflp->fl4_src);
2408 if (dev_out == NULL)
2411 /* I removed check for oif == dev_out->oif here.
2412 It was wrong for two reasons:
2413 1. ip_dev_find(saddr) can return wrong iface, if saddr is
2414 assigned to multiple interfaces.
2415 2. Moreover, we are allowed to send packets with saddr
2416 of another iface. --ANK
2419 if (oldflp->oif == 0
2420 && (MULTICAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF)) {
2421 /* Special hack: user can direct multicasts
2422 and limited broadcast via necessary interface
2423 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2424 This hack is not just for fun, it allows
2425 vic,vat and friends to work.
2426 They bind socket to loopback, set ttl to zero
2427 and expect that it will work.
2428 From the viewpoint of routing cache they are broken,
2429 because we are not allowed to build multicast path
2430 with loopback source addr (look, routing cache
2431 cannot know, that ttl is zero, so that packet
2432 will not leave this host and route is valid).
2433 Luckily, this hack is good workaround.
2436 fl.oif = dev_out->ifindex;
2446 dev_out = dev_get_by_index(oldflp->oif);
2448 if (dev_out == NULL)
2451 /* RACE: Check return value of inet_select_addr instead. */
2452 if (__in_dev_get_rtnl(dev_out) == NULL) {
2454 goto out; /* Wrong error code */
2457 if (LOCAL_MCAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF) {
2459 fl.fl4_src = inet_select_addr(dev_out, 0,
2464 if (MULTICAST(oldflp->fl4_dst))
2465 fl.fl4_src = inet_select_addr(dev_out, 0,
2467 else if (!oldflp->fl4_dst)
2468 fl.fl4_src = inet_select_addr(dev_out, 0,
2474 fl.fl4_dst = fl.fl4_src;
2476 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2479 dev_out = &loopback_dev;
2481 fl.oif = loopback_dev.ifindex;
2482 res.type = RTN_LOCAL;
2483 flags |= RTCF_LOCAL;
2487 if (fib_lookup(&fl, &res)) {
2490 /* Apparently, routing tables are wrong. Assume,
2491 that the destination is on link.
2494 Because we are allowed to send to iface
2495 even if it has NO routes and NO assigned
2496 addresses. When oif is specified, routing
2497 tables are looked up with only one purpose:
2498 to catch if destination is gatewayed, rather than
2499 direct. Moreover, if MSG_DONTROUTE is set,
2500 we send packet, ignoring both routing tables
2501 and ifaddr state. --ANK
2504 We could make it even if oif is unknown,
2505 likely IPv6, but we do not.
2508 if (fl.fl4_src == 0)
2509 fl.fl4_src = inet_select_addr(dev_out, 0,
2511 res.type = RTN_UNICAST;
2521 if (res.type == RTN_LOCAL) {
2523 fl.fl4_src = fl.fl4_dst;
2526 dev_out = &loopback_dev;
2528 fl.oif = dev_out->ifindex;
2530 fib_info_put(res.fi);
2532 flags |= RTCF_LOCAL;
2536 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2537 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2538 fib_select_multipath(&fl, &res);
2541 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2542 fib_select_default(&fl, &res);
2545 fl.fl4_src = FIB_RES_PREFSRC(res);
2549 dev_out = FIB_RES_DEV(res);
2551 fl.oif = dev_out->ifindex;
2555 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2565 int __ip_route_output_key(struct rtable **rp, const struct flowi *flp)
2570 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif);
2573 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2574 rth = rcu_dereference(rth->u.rt_next)) {
2575 if (rth->fl.fl4_dst == flp->fl4_dst &&
2576 rth->fl.fl4_src == flp->fl4_src &&
2578 rth->fl.oif == flp->oif &&
2579 #ifdef CONFIG_IP_ROUTE_FWMARK
2580 rth->fl.fl4_fwmark == flp->fl4_fwmark &&
2582 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2583 (IPTOS_RT_MASK | RTO_ONLINK))) {
2585 /* check for multipath routes and choose one if
2588 if (multipath_select_route(flp, rth, rp)) {
2589 dst_hold(&(*rp)->u.dst);
2590 RT_CACHE_STAT_INC(out_hit);
2591 rcu_read_unlock_bh();
2595 rth->u.dst.lastuse = jiffies;
2596 dst_hold(&rth->u.dst);
2598 RT_CACHE_STAT_INC(out_hit);
2599 rcu_read_unlock_bh();
2603 RT_CACHE_STAT_INC(out_hlist_search);
2605 rcu_read_unlock_bh();
2607 return ip_route_output_slow(rp, flp);
2610 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2612 int ip_route_output_flow(struct rtable **rp, struct flowi *flp, struct sock *sk, int flags)
2616 if ((err = __ip_route_output_key(rp, flp)) != 0)
2621 flp->fl4_src = (*rp)->rt_src;
2623 flp->fl4_dst = (*rp)->rt_dst;
2624 return xfrm_lookup((struct dst_entry **)rp, flp, sk, flags);
2630 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2632 int ip_route_output_key(struct rtable **rp, struct flowi *flp)
2634 return ip_route_output_flow(rp, flp, NULL, 0);
2637 static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
2638 int nowait, unsigned int flags)
2640 struct rtable *rt = (struct rtable*)skb->dst;
2642 struct nlmsghdr *nlh;
2643 struct rta_cacheinfo ci;
2645 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2649 r = nlmsg_data(nlh);
2650 r->rtm_family = AF_INET;
2651 r->rtm_dst_len = 32;
2653 r->rtm_tos = rt->fl.fl4_tos;
2654 r->rtm_table = RT_TABLE_MAIN;
2655 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2656 r->rtm_type = rt->rt_type;
2657 r->rtm_scope = RT_SCOPE_UNIVERSE;
2658 r->rtm_protocol = RTPROT_UNSPEC;
2659 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2660 if (rt->rt_flags & RTCF_NOTIFY)
2661 r->rtm_flags |= RTM_F_NOTIFY;
2663 NLA_PUT_U32(skb, RTA_DST, rt->rt_dst);
2665 if (rt->fl.fl4_src) {
2666 r->rtm_src_len = 32;
2667 NLA_PUT_U32(skb, RTA_SRC, rt->fl.fl4_src);
2670 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2671 #ifdef CONFIG_NET_CLS_ROUTE
2672 if (rt->u.dst.tclassid)
2673 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2675 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2676 if (rt->rt_multipath_alg != IP_MP_ALG_NONE)
2677 NLA_PUT_U32(skb, RTA_MP_ALGO, rt->rt_multipath_alg);
2680 NLA_PUT_U32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2681 else if (rt->rt_src != rt->fl.fl4_src)
2682 NLA_PUT_U32(skb, RTA_PREFSRC, rt->rt_src);
2684 if (rt->rt_dst != rt->rt_gateway)
2685 NLA_PUT_U32(skb, RTA_GATEWAY, rt->rt_gateway);
2687 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2688 goto nla_put_failure;
2690 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
2691 ci.rta_used = rt->u.dst.__use;
2692 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
2693 if (rt->u.dst.expires)
2694 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
2697 ci.rta_error = rt->u.dst.error;
2698 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
2700 ci.rta_id = rt->peer->ip_id_count;
2701 if (rt->peer->tcp_ts_stamp) {
2702 ci.rta_ts = rt->peer->tcp_ts;
2703 ci.rta_tsage = xtime.tv_sec - rt->peer->tcp_ts_stamp;
2708 #ifdef CONFIG_IP_MROUTE
2709 u32 dst = rt->rt_dst;
2711 if (MULTICAST(dst) && !LOCAL_MCAST(dst) &&
2712 ipv4_devconf.mc_forwarding) {
2713 int err = ipmr_get_route(skb, r, nowait);
2718 goto nla_put_failure;
2720 if (err == -EMSGSIZE)
2721 goto nla_put_failure;
2727 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2730 NLA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
2732 return nlmsg_end(skb, nlh);
2735 return nlmsg_cancel(skb, nlh);
2738 int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2741 struct nlattr *tb[RTA_MAX+1];
2742 struct rtable *rt = NULL;
2747 struct sk_buff *skb;
2749 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2753 rtm = nlmsg_data(nlh);
2755 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2761 /* Reserve room for dummy headers, this skb can pass
2762 through good chunk of routing engine.
2764 skb->mac.raw = skb->nh.raw = skb->data;
2766 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2767 skb->nh.iph->protocol = IPPROTO_ICMP;
2768 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2770 src = tb[RTA_SRC] ? nla_get_u32(tb[RTA_SRC]) : 0;
2771 dst = tb[RTA_DST] ? nla_get_u32(tb[RTA_DST]) : 0;
2772 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2775 struct net_device *dev;
2777 dev = __dev_get_by_index(iif);
2783 skb->protocol = htons(ETH_P_IP);
2786 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2789 rt = (struct rtable*) skb->dst;
2790 if (err == 0 && rt->u.dst.error)
2791 err = -rt->u.dst.error;
2798 .tos = rtm->rtm_tos,
2801 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2803 err = ip_route_output_key(&rt, &fl);
2809 skb->dst = &rt->u.dst;
2810 if (rtm->rtm_flags & RTM_F_NOTIFY)
2811 rt->rt_flags |= RTCF_NOTIFY;
2813 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2814 RTM_NEWROUTE, 0, 0);
2818 err = rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
2827 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2834 s_idx = idx = cb->args[1];
2835 for (h = 0; h <= rt_hash_mask; h++) {
2836 if (h < s_h) continue;
2840 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2841 rt = rcu_dereference(rt->u.rt_next), idx++) {
2844 skb->dst = dst_clone(&rt->u.dst);
2845 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
2846 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2847 1, NLM_F_MULTI) <= 0) {
2848 dst_release(xchg(&skb->dst, NULL));
2849 rcu_read_unlock_bh();
2852 dst_release(xchg(&skb->dst, NULL));
2854 rcu_read_unlock_bh();
2863 void ip_rt_multicast_event(struct in_device *in_dev)
2868 #ifdef CONFIG_SYSCTL
2869 static int flush_delay;
2871 static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write,
2872 struct file *filp, void __user *buffer,
2873 size_t *lenp, loff_t *ppos)
2876 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2877 rt_cache_flush(flush_delay);
2884 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2887 void __user *oldval,
2888 size_t __user *oldlenp,
2889 void __user *newval,
2894 if (newlen != sizeof(int))
2896 if (get_user(delay, (int __user *)newval))
2898 rt_cache_flush(delay);
2902 ctl_table ipv4_route_table[] = {
2904 .ctl_name = NET_IPV4_ROUTE_FLUSH,
2905 .procname = "flush",
2906 .data = &flush_delay,
2907 .maxlen = sizeof(int),
2909 .proc_handler = &ipv4_sysctl_rtcache_flush,
2910 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
2913 .ctl_name = NET_IPV4_ROUTE_MIN_DELAY,
2914 .procname = "min_delay",
2915 .data = &ip_rt_min_delay,
2916 .maxlen = sizeof(int),
2918 .proc_handler = &proc_dointvec_jiffies,
2919 .strategy = &sysctl_jiffies,
2922 .ctl_name = NET_IPV4_ROUTE_MAX_DELAY,
2923 .procname = "max_delay",
2924 .data = &ip_rt_max_delay,
2925 .maxlen = sizeof(int),
2927 .proc_handler = &proc_dointvec_jiffies,
2928 .strategy = &sysctl_jiffies,
2931 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2932 .procname = "gc_thresh",
2933 .data = &ipv4_dst_ops.gc_thresh,
2934 .maxlen = sizeof(int),
2936 .proc_handler = &proc_dointvec,
2939 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2940 .procname = "max_size",
2941 .data = &ip_rt_max_size,
2942 .maxlen = sizeof(int),
2944 .proc_handler = &proc_dointvec,
2947 /* Deprecated. Use gc_min_interval_ms */
2949 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2950 .procname = "gc_min_interval",
2951 .data = &ip_rt_gc_min_interval,
2952 .maxlen = sizeof(int),
2954 .proc_handler = &proc_dointvec_jiffies,
2955 .strategy = &sysctl_jiffies,
2958 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2959 .procname = "gc_min_interval_ms",
2960 .data = &ip_rt_gc_min_interval,
2961 .maxlen = sizeof(int),
2963 .proc_handler = &proc_dointvec_ms_jiffies,
2964 .strategy = &sysctl_ms_jiffies,
2967 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2968 .procname = "gc_timeout",
2969 .data = &ip_rt_gc_timeout,
2970 .maxlen = sizeof(int),
2972 .proc_handler = &proc_dointvec_jiffies,
2973 .strategy = &sysctl_jiffies,
2976 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2977 .procname = "gc_interval",
2978 .data = &ip_rt_gc_interval,
2979 .maxlen = sizeof(int),
2981 .proc_handler = &proc_dointvec_jiffies,
2982 .strategy = &sysctl_jiffies,
2985 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2986 .procname = "redirect_load",
2987 .data = &ip_rt_redirect_load,
2988 .maxlen = sizeof(int),
2990 .proc_handler = &proc_dointvec,
2993 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2994 .procname = "redirect_number",
2995 .data = &ip_rt_redirect_number,
2996 .maxlen = sizeof(int),
2998 .proc_handler = &proc_dointvec,
3001 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
3002 .procname = "redirect_silence",
3003 .data = &ip_rt_redirect_silence,
3004 .maxlen = sizeof(int),
3006 .proc_handler = &proc_dointvec,
3009 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
3010 .procname = "error_cost",
3011 .data = &ip_rt_error_cost,
3012 .maxlen = sizeof(int),
3014 .proc_handler = &proc_dointvec,
3017 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
3018 .procname = "error_burst",
3019 .data = &ip_rt_error_burst,
3020 .maxlen = sizeof(int),
3022 .proc_handler = &proc_dointvec,
3025 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
3026 .procname = "gc_elasticity",
3027 .data = &ip_rt_gc_elasticity,
3028 .maxlen = sizeof(int),
3030 .proc_handler = &proc_dointvec,
3033 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
3034 .procname = "mtu_expires",
3035 .data = &ip_rt_mtu_expires,
3036 .maxlen = sizeof(int),
3038 .proc_handler = &proc_dointvec_jiffies,
3039 .strategy = &sysctl_jiffies,
3042 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
3043 .procname = "min_pmtu",
3044 .data = &ip_rt_min_pmtu,
3045 .maxlen = sizeof(int),
3047 .proc_handler = &proc_dointvec,
3050 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
3051 .procname = "min_adv_mss",
3052 .data = &ip_rt_min_advmss,
3053 .maxlen = sizeof(int),
3055 .proc_handler = &proc_dointvec,
3058 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
3059 .procname = "secret_interval",
3060 .data = &ip_rt_secret_interval,
3061 .maxlen = sizeof(int),
3063 .proc_handler = &proc_dointvec_jiffies,
3064 .strategy = &sysctl_jiffies,
3070 #ifdef CONFIG_NET_CLS_ROUTE
3071 struct ip_rt_acct *ip_rt_acct;
3073 /* This code sucks. But you should have seen it before! --RR */
3075 /* IP route accounting ptr for this logical cpu number. */
3076 #define IP_RT_ACCT_CPU(i) (ip_rt_acct + i * 256)
3078 #ifdef CONFIG_PROC_FS
3079 static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
3080 int length, int *eof, void *data)
3084 if ((offset & 3) || (length & 3))
3087 if (offset >= sizeof(struct ip_rt_acct) * 256) {
3092 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
3093 length = sizeof(struct ip_rt_acct) * 256 - offset;
3097 offset /= sizeof(u32);
3100 u32 *src = ((u32 *) IP_RT_ACCT_CPU(0)) + offset;
3101 u32 *dst = (u32 *) buffer;
3103 /* Copy first cpu. */
3105 memcpy(dst, src, length);
3107 /* Add the other cpus in, one int at a time */
3108 for_each_possible_cpu(i) {
3111 src = ((u32 *) IP_RT_ACCT_CPU(i)) + offset;
3113 for (j = 0; j < length/4; j++)
3119 #endif /* CONFIG_PROC_FS */
3120 #endif /* CONFIG_NET_CLS_ROUTE */
3122 static __initdata unsigned long rhash_entries;
3123 static int __init set_rhash_entries(char *str)
3127 rhash_entries = simple_strtoul(str, &str, 0);
3130 __setup("rhash_entries=", set_rhash_entries);
3132 int __init ip_rt_init(void)
3136 rt_hash_rnd = (int) ((num_physpages ^ (num_physpages>>8)) ^
3137 (jiffies ^ (jiffies >> 7)));
3139 #ifdef CONFIG_NET_CLS_ROUTE
3143 (PAGE_SIZE << order) < 256 * sizeof(struct ip_rt_acct) * NR_CPUS; order++)
3145 ip_rt_acct = (struct ip_rt_acct *)__get_free_pages(GFP_KERNEL, order);
3147 panic("IP: failed to allocate ip_rt_acct\n");
3148 memset(ip_rt_acct, 0, PAGE_SIZE << order);
3152 ipv4_dst_ops.kmem_cachep =
3153 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3154 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
3156 rt_hash_table = (struct rt_hash_bucket *)
3157 alloc_large_system_hash("IP route cache",
3158 sizeof(struct rt_hash_bucket),
3160 (num_physpages >= 128 * 1024) ?
3166 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3167 rt_hash_lock_init();
3169 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3170 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3175 init_timer(&rt_flush_timer);
3176 rt_flush_timer.function = rt_run_flush;
3177 init_timer(&rt_periodic_timer);
3178 rt_periodic_timer.function = rt_check_expire;
3179 init_timer(&rt_secret_timer);
3180 rt_secret_timer.function = rt_secret_rebuild;
3182 /* All the timers, started at system startup tend
3183 to synchronize. Perturb it a bit.
3185 rt_periodic_timer.expires = jiffies + net_random() % ip_rt_gc_interval +
3187 add_timer(&rt_periodic_timer);
3189 rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval +
3190 ip_rt_secret_interval;
3191 add_timer(&rt_secret_timer);
3193 #ifdef CONFIG_PROC_FS
3195 struct proc_dir_entry *rtstat_pde = NULL; /* keep gcc happy */
3196 if (!proc_net_fops_create("rt_cache", S_IRUGO, &rt_cache_seq_fops) ||
3197 !(rtstat_pde = create_proc_entry("rt_cache", S_IRUGO,
3201 rtstat_pde->proc_fops = &rt_cpu_seq_fops;
3203 #ifdef CONFIG_NET_CLS_ROUTE
3204 create_proc_read_entry("rt_acct", 0, proc_net, ip_rt_acct_read, NULL);
3214 EXPORT_SYMBOL(__ip_select_ident);
3215 EXPORT_SYMBOL(ip_route_input);
3216 EXPORT_SYMBOL(ip_route_output_key);