2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #include <linux/module.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <linux/bitops.h>
69 #include <linux/types.h>
70 #include <linux/kernel.h>
72 #include <linux/bootmem.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/workqueue.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/jhash.h>
91 #include <linux/rcupdate.h>
92 #include <linux/times.h>
94 #include <net/net_namespace.h>
95 #include <net/protocol.h>
97 #include <net/route.h>
98 #include <net/inetpeer.h>
100 #include <net/ip_fib.h>
103 #include <net/icmp.h>
104 #include <net/xfrm.h>
105 #include <net/netevent.h>
106 #include <net/rtnetlink.h>
108 #include <linux/sysctl.h>
111 #define RT_FL_TOS(oldflp) \
112 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114 #define IP_MAX_MTU 0xFFF0
116 #define RT_GC_TIMEOUT (300*HZ)
118 static int ip_rt_max_size
;
119 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
120 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
121 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
122 static int ip_rt_redirect_number __read_mostly
= 9;
123 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
124 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
125 static int ip_rt_error_cost __read_mostly
= HZ
;
126 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
127 static int ip_rt_gc_elasticity __read_mostly
= 8;
128 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
129 static int ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly
= 256;
131 static int ip_rt_secret_interval __read_mostly
= 10 * 60 * HZ
;
132 static int rt_chain_length_max __read_mostly
= 20;
134 static struct delayed_work expires_work
;
135 static unsigned long expires_ljiffies
;
138 * Interface to generic destination cache.
141 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
142 static void ipv4_dst_destroy(struct dst_entry
*dst
);
143 static void ipv4_dst_ifdown(struct dst_entry
*dst
,
144 struct net_device
*dev
, int how
);
145 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
);
146 static void ipv4_link_failure(struct sk_buff
*skb
);
147 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
);
148 static int rt_garbage_collect(struct dst_ops
*ops
);
149 static void rt_emergency_hash_rebuild(struct net
*net
);
152 static struct dst_ops ipv4_dst_ops
= {
154 .protocol
= cpu_to_be16(ETH_P_IP
),
155 .gc
= rt_garbage_collect
,
156 .check
= ipv4_dst_check
,
157 .destroy
= ipv4_dst_destroy
,
158 .ifdown
= ipv4_dst_ifdown
,
159 .negative_advice
= ipv4_negative_advice
,
160 .link_failure
= ipv4_link_failure
,
161 .update_pmtu
= ip_rt_update_pmtu
,
162 .local_out
= __ip_local_out
,
163 .entries
= ATOMIC_INIT(0),
166 #define ECN_OR_COST(class) TC_PRIO_##class
168 const __u8 ip_tos2prio
[16] = {
172 ECN_OR_COST(BESTEFFORT
),
178 ECN_OR_COST(INTERACTIVE
),
180 ECN_OR_COST(INTERACTIVE
),
181 TC_PRIO_INTERACTIVE_BULK
,
182 ECN_OR_COST(INTERACTIVE_BULK
),
183 TC_PRIO_INTERACTIVE_BULK
,
184 ECN_OR_COST(INTERACTIVE_BULK
)
192 /* The locking scheme is rather straight forward:
194 * 1) Read-Copy Update protects the buckets of the central route hash.
195 * 2) Only writers remove entries, and they hold the lock
196 * as they look at rtable reference counts.
197 * 3) Only readers acquire references to rtable entries,
198 * they do so with atomic increments and with the
202 struct rt_hash_bucket
{
203 struct rtable
*chain
;
206 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
207 defined(CONFIG_PROVE_LOCKING)
209 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
210 * The size of this table is a power of two and depends on the number of CPUS.
211 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
213 #ifdef CONFIG_LOCKDEP
214 # define RT_HASH_LOCK_SZ 256
217 # define RT_HASH_LOCK_SZ 4096
219 # define RT_HASH_LOCK_SZ 2048
221 # define RT_HASH_LOCK_SZ 1024
223 # define RT_HASH_LOCK_SZ 512
225 # define RT_HASH_LOCK_SZ 256
229 static spinlock_t
*rt_hash_locks
;
230 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
232 static __init
void rt_hash_lock_init(void)
236 rt_hash_locks
= kmalloc(sizeof(spinlock_t
) * RT_HASH_LOCK_SZ
,
239 panic("IP: failed to allocate rt_hash_locks\n");
241 for (i
= 0; i
< RT_HASH_LOCK_SZ
; i
++)
242 spin_lock_init(&rt_hash_locks
[i
]);
245 # define rt_hash_lock_addr(slot) NULL
247 static inline void rt_hash_lock_init(void)
252 static struct rt_hash_bucket
*rt_hash_table __read_mostly
;
253 static unsigned rt_hash_mask __read_mostly
;
254 static unsigned int rt_hash_log __read_mostly
;
256 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
257 #define RT_CACHE_STAT_INC(field) \
258 (__raw_get_cpu_var(rt_cache_stat).field++)
260 static inline unsigned int rt_hash(__be32 daddr
, __be32 saddr
, int idx
,
263 return jhash_3words((__force u32
)(__be32
)(daddr
),
264 (__force u32
)(__be32
)(saddr
),
269 static inline int rt_genid(struct net
*net
)
271 return atomic_read(&net
->ipv4
.rt_genid
);
274 #ifdef CONFIG_PROC_FS
275 struct rt_cache_iter_state
{
276 struct seq_net_private p
;
281 static struct rtable
*rt_cache_get_first(struct seq_file
*seq
)
283 struct rt_cache_iter_state
*st
= seq
->private;
284 struct rtable
*r
= NULL
;
286 for (st
->bucket
= rt_hash_mask
; st
->bucket
>= 0; --st
->bucket
) {
287 if (!rt_hash_table
[st
->bucket
].chain
)
290 r
= rcu_dereference(rt_hash_table
[st
->bucket
].chain
);
292 if (dev_net(r
->u
.dst
.dev
) == seq_file_net(seq
) &&
293 r
->rt_genid
== st
->genid
)
295 r
= rcu_dereference(r
->u
.dst
.rt_next
);
297 rcu_read_unlock_bh();
302 static struct rtable
*__rt_cache_get_next(struct seq_file
*seq
,
305 struct rt_cache_iter_state
*st
= seq
->private;
307 r
= r
->u
.dst
.rt_next
;
309 rcu_read_unlock_bh();
311 if (--st
->bucket
< 0)
313 } while (!rt_hash_table
[st
->bucket
].chain
);
315 r
= rt_hash_table
[st
->bucket
].chain
;
317 return rcu_dereference(r
);
320 static struct rtable
*rt_cache_get_next(struct seq_file
*seq
,
323 struct rt_cache_iter_state
*st
= seq
->private;
324 while ((r
= __rt_cache_get_next(seq
, r
)) != NULL
) {
325 if (dev_net(r
->u
.dst
.dev
) != seq_file_net(seq
))
327 if (r
->rt_genid
== st
->genid
)
333 static struct rtable
*rt_cache_get_idx(struct seq_file
*seq
, loff_t pos
)
335 struct rtable
*r
= rt_cache_get_first(seq
);
338 while (pos
&& (r
= rt_cache_get_next(seq
, r
)))
340 return pos
? NULL
: r
;
343 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
345 struct rt_cache_iter_state
*st
= seq
->private;
347 return rt_cache_get_idx(seq
, *pos
- 1);
348 st
->genid
= rt_genid(seq_file_net(seq
));
349 return SEQ_START_TOKEN
;
352 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
356 if (v
== SEQ_START_TOKEN
)
357 r
= rt_cache_get_first(seq
);
359 r
= rt_cache_get_next(seq
, v
);
364 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
366 if (v
&& v
!= SEQ_START_TOKEN
)
367 rcu_read_unlock_bh();
370 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
372 if (v
== SEQ_START_TOKEN
)
373 seq_printf(seq
, "%-127s\n",
374 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
375 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
378 struct rtable
*r
= v
;
381 seq_printf(seq
, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
382 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
383 r
->u
.dst
.dev
? r
->u
.dst
.dev
->name
: "*",
384 (unsigned long)r
->rt_dst
, (unsigned long)r
->rt_gateway
,
385 r
->rt_flags
, atomic_read(&r
->u
.dst
.__refcnt
),
386 r
->u
.dst
.__use
, 0, (unsigned long)r
->rt_src
,
387 (dst_metric(&r
->u
.dst
, RTAX_ADVMSS
) ?
388 (int)dst_metric(&r
->u
.dst
, RTAX_ADVMSS
) + 40 : 0),
389 dst_metric(&r
->u
.dst
, RTAX_WINDOW
),
390 (int)((dst_metric(&r
->u
.dst
, RTAX_RTT
) >> 3) +
391 dst_metric(&r
->u
.dst
, RTAX_RTTVAR
)),
393 r
->u
.dst
.hh
? atomic_read(&r
->u
.dst
.hh
->hh_refcnt
) : -1,
394 r
->u
.dst
.hh
? (r
->u
.dst
.hh
->hh_output
==
396 r
->rt_spec_dst
, &len
);
398 seq_printf(seq
, "%*s\n", 127 - len
, "");
403 static const struct seq_operations rt_cache_seq_ops
= {
404 .start
= rt_cache_seq_start
,
405 .next
= rt_cache_seq_next
,
406 .stop
= rt_cache_seq_stop
,
407 .show
= rt_cache_seq_show
,
410 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
412 return seq_open_net(inode
, file
, &rt_cache_seq_ops
,
413 sizeof(struct rt_cache_iter_state
));
416 static const struct file_operations rt_cache_seq_fops
= {
417 .owner
= THIS_MODULE
,
418 .open
= rt_cache_seq_open
,
421 .release
= seq_release_net
,
425 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
430 return SEQ_START_TOKEN
;
432 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
433 if (!cpu_possible(cpu
))
436 return &per_cpu(rt_cache_stat
, cpu
);
441 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
445 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
446 if (!cpu_possible(cpu
))
449 return &per_cpu(rt_cache_stat
, cpu
);
455 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
460 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
462 struct rt_cache_stat
*st
= v
;
464 if (v
== SEQ_START_TOKEN
) {
465 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");
469 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
470 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
471 atomic_read(&ipv4_dst_ops
.entries
),
494 static const struct seq_operations rt_cpu_seq_ops
= {
495 .start
= rt_cpu_seq_start
,
496 .next
= rt_cpu_seq_next
,
497 .stop
= rt_cpu_seq_stop
,
498 .show
= rt_cpu_seq_show
,
502 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
504 return seq_open(file
, &rt_cpu_seq_ops
);
507 static const struct file_operations rt_cpu_seq_fops
= {
508 .owner
= THIS_MODULE
,
509 .open
= rt_cpu_seq_open
,
512 .release
= seq_release
,
515 #ifdef CONFIG_NET_CLS_ROUTE
516 static int ip_rt_acct_read(char *buffer
, char **start
, off_t offset
,
517 int length
, int *eof
, void *data
)
521 if ((offset
& 3) || (length
& 3))
524 if (offset
>= sizeof(struct ip_rt_acct
) * 256) {
529 if (offset
+ length
>= sizeof(struct ip_rt_acct
) * 256) {
530 length
= sizeof(struct ip_rt_acct
) * 256 - offset
;
534 offset
/= sizeof(u32
);
537 u32
*dst
= (u32
*) buffer
;
540 memset(dst
, 0, length
);
542 for_each_possible_cpu(i
) {
546 src
= ((u32
*) per_cpu_ptr(ip_rt_acct
, i
)) + offset
;
547 for (j
= 0; j
< length
/4; j
++)
555 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
557 struct proc_dir_entry
*pde
;
559 pde
= proc_net_fops_create(net
, "rt_cache", S_IRUGO
,
564 pde
= proc_create("rt_cache", S_IRUGO
,
565 net
->proc_net_stat
, &rt_cpu_seq_fops
);
569 #ifdef CONFIG_NET_CLS_ROUTE
570 pde
= create_proc_read_entry("rt_acct", 0, net
->proc_net
,
571 ip_rt_acct_read
, NULL
);
577 #ifdef CONFIG_NET_CLS_ROUTE
579 remove_proc_entry("rt_cache", net
->proc_net_stat
);
582 remove_proc_entry("rt_cache", net
->proc_net
);
587 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
589 remove_proc_entry("rt_cache", net
->proc_net_stat
);
590 remove_proc_entry("rt_cache", net
->proc_net
);
591 remove_proc_entry("rt_acct", net
->proc_net
);
594 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
595 .init
= ip_rt_do_proc_init
,
596 .exit
= ip_rt_do_proc_exit
,
599 static int __init
ip_rt_proc_init(void)
601 return register_pernet_subsys(&ip_rt_proc_ops
);
605 static inline int ip_rt_proc_init(void)
609 #endif /* CONFIG_PROC_FS */
611 static inline void rt_free(struct rtable
*rt
)
613 call_rcu_bh(&rt
->u
.dst
.rcu_head
, dst_rcu_free
);
616 static inline void rt_drop(struct rtable
*rt
)
619 call_rcu_bh(&rt
->u
.dst
.rcu_head
, dst_rcu_free
);
622 static inline int rt_fast_clean(struct rtable
*rth
)
624 /* Kill broadcast/multicast entries very aggresively, if they
625 collide in hash table with more useful entries */
626 return (rth
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) &&
627 rth
->fl
.iif
&& rth
->u
.dst
.rt_next
;
630 static inline int rt_valuable(struct rtable
*rth
)
632 return (rth
->rt_flags
& (RTCF_REDIRECTED
| RTCF_NOTIFY
)) ||
636 static int rt_may_expire(struct rtable
*rth
, unsigned long tmo1
, unsigned long tmo2
)
641 if (atomic_read(&rth
->u
.dst
.__refcnt
))
645 if (rth
->u
.dst
.expires
&&
646 time_after_eq(jiffies
, rth
->u
.dst
.expires
))
649 age
= jiffies
- rth
->u
.dst
.lastuse
;
651 if ((age
<= tmo1
&& !rt_fast_clean(rth
)) ||
652 (age
<= tmo2
&& rt_valuable(rth
)))
658 /* Bits of score are:
660 * 30: not quite useless
661 * 29..0: usage counter
663 static inline u32
rt_score(struct rtable
*rt
)
665 u32 score
= jiffies
- rt
->u
.dst
.lastuse
;
667 score
= ~score
& ~(3<<30);
673 !(rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
|RTCF_LOCAL
)))
679 static inline bool rt_caching(const struct net
*net
)
681 return net
->ipv4
.current_rt_cache_rebuild_count
<=
682 net
->ipv4
.sysctl_rt_cache_rebuild_count
;
685 static inline bool compare_hash_inputs(const struct flowi
*fl1
,
686 const struct flowi
*fl2
)
688 return (__force u32
)(((fl1
->nl_u
.ip4_u
.daddr
^ fl2
->nl_u
.ip4_u
.daddr
) |
689 (fl1
->nl_u
.ip4_u
.saddr
^ fl2
->nl_u
.ip4_u
.saddr
) |
690 (fl1
->iif
^ fl2
->iif
)) == 0);
693 static inline int compare_keys(struct flowi
*fl1
, struct flowi
*fl2
)
695 return ((__force u32
)((fl1
->nl_u
.ip4_u
.daddr
^ fl2
->nl_u
.ip4_u
.daddr
) |
696 (fl1
->nl_u
.ip4_u
.saddr
^ fl2
->nl_u
.ip4_u
.saddr
)) |
697 (fl1
->mark
^ fl2
->mark
) |
698 (*(u16
*)&fl1
->nl_u
.ip4_u
.tos
^
699 *(u16
*)&fl2
->nl_u
.ip4_u
.tos
) |
700 (fl1
->oif
^ fl2
->oif
) |
701 (fl1
->iif
^ fl2
->iif
)) == 0;
704 static inline int compare_netns(struct rtable
*rt1
, struct rtable
*rt2
)
706 return dev_net(rt1
->u
.dst
.dev
) == dev_net(rt2
->u
.dst
.dev
);
709 static inline int rt_is_expired(struct rtable
*rth
)
711 return rth
->rt_genid
!= rt_genid(dev_net(rth
->u
.dst
.dev
));
715 * Perform a full scan of hash table and free all entries.
716 * Can be called by a softirq or a process.
717 * In the later case, we want to be reschedule if necessary
719 static void rt_do_flush(int process_context
)
722 struct rtable
*rth
, *next
;
723 struct rtable
* tail
;
725 for (i
= 0; i
<= rt_hash_mask
; i
++) {
726 if (process_context
&& need_resched())
728 rth
= rt_hash_table
[i
].chain
;
732 spin_lock_bh(rt_hash_lock_addr(i
));
735 struct rtable
** prev
, * p
;
737 rth
= rt_hash_table
[i
].chain
;
739 /* defer releasing the head of the list after spin_unlock */
740 for (tail
= rth
; tail
; tail
= tail
->u
.dst
.rt_next
)
741 if (!rt_is_expired(tail
))
744 rt_hash_table
[i
].chain
= tail
;
746 /* call rt_free on entries after the tail requiring flush */
747 prev
= &rt_hash_table
[i
].chain
;
748 for (p
= *prev
; p
; p
= next
) {
749 next
= p
->u
.dst
.rt_next
;
750 if (!rt_is_expired(p
)) {
751 prev
= &p
->u
.dst
.rt_next
;
759 rth
= rt_hash_table
[i
].chain
;
760 rt_hash_table
[i
].chain
= NULL
;
763 spin_unlock_bh(rt_hash_lock_addr(i
));
765 for (; rth
!= tail
; rth
= next
) {
766 next
= rth
->u
.dst
.rt_next
;
773 * While freeing expired entries, we compute average chain length
774 * and standard deviation, using fixed-point arithmetic.
775 * This to have an estimation of rt_chain_length_max
776 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
777 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
781 #define ONE (1UL << FRACT_BITS)
783 static void rt_check_expire(void)
785 static unsigned int rover
;
786 unsigned int i
= rover
, goal
;
787 struct rtable
*rth
, *aux
, **rthp
;
788 unsigned long samples
= 0;
789 unsigned long sum
= 0, sum2
= 0;
793 delta
= jiffies
- expires_ljiffies
;
794 expires_ljiffies
= jiffies
;
795 mult
= ((u64
)delta
) << rt_hash_log
;
796 if (ip_rt_gc_timeout
> 1)
797 do_div(mult
, ip_rt_gc_timeout
);
798 goal
= (unsigned int)mult
;
799 if (goal
> rt_hash_mask
)
800 goal
= rt_hash_mask
+ 1;
801 for (; goal
> 0; goal
--) {
802 unsigned long tmo
= ip_rt_gc_timeout
;
803 unsigned long length
;
805 i
= (i
+ 1) & rt_hash_mask
;
806 rthp
= &rt_hash_table
[i
].chain
;
816 spin_lock_bh(rt_hash_lock_addr(i
));
817 while ((rth
= *rthp
) != NULL
) {
818 prefetch(rth
->u
.dst
.rt_next
);
819 if (rt_is_expired(rth
)) {
820 *rthp
= rth
->u
.dst
.rt_next
;
824 if (rth
->u
.dst
.expires
) {
825 /* Entry is expired even if it is in use */
826 if (time_before_eq(jiffies
, rth
->u
.dst
.expires
)) {
829 rthp
= &rth
->u
.dst
.rt_next
;
831 * We only count entries on
832 * a chain with equal hash inputs once
833 * so that entries for different QOS
834 * levels, and other non-hash input
835 * attributes don't unfairly skew
836 * the length computation
838 for (aux
= rt_hash_table
[i
].chain
;;) {
843 if (compare_hash_inputs(&aux
->fl
, &rth
->fl
))
845 aux
= aux
->u
.dst
.rt_next
;
849 } else if (!rt_may_expire(rth
, tmo
, ip_rt_gc_timeout
))
852 /* Cleanup aged off entries. */
853 *rthp
= rth
->u
.dst
.rt_next
;
856 spin_unlock_bh(rt_hash_lock_addr(i
));
858 sum2
+= length
*length
;
861 unsigned long avg
= sum
/ samples
;
862 unsigned long sd
= int_sqrt(sum2
/ samples
- avg
*avg
);
863 rt_chain_length_max
= max_t(unsigned long,
865 (avg
+ 4*sd
) >> FRACT_BITS
);
871 * rt_worker_func() is run in process context.
872 * we call rt_check_expire() to scan part of the hash table
874 static void rt_worker_func(struct work_struct
*work
)
877 schedule_delayed_work(&expires_work
, ip_rt_gc_interval
);
881 * Pertubation of rt_genid by a small quantity [1..256]
882 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
883 * many times (2^24) without giving recent rt_genid.
884 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
886 static void rt_cache_invalidate(struct net
*net
)
888 unsigned char shuffle
;
890 get_random_bytes(&shuffle
, sizeof(shuffle
));
891 atomic_add(shuffle
+ 1U, &net
->ipv4
.rt_genid
);
895 * delay < 0 : invalidate cache (fast : entries will be deleted later)
896 * delay >= 0 : invalidate & flush cache (can be long)
898 void rt_cache_flush(struct net
*net
, int delay
)
900 rt_cache_invalidate(net
);
902 rt_do_flush(!in_softirq());
906 * We change rt_genid and let gc do the cleanup
908 static void rt_secret_rebuild(unsigned long __net
)
910 struct net
*net
= (struct net
*)__net
;
911 rt_cache_invalidate(net
);
912 mod_timer(&net
->ipv4
.rt_secret_timer
, jiffies
+ ip_rt_secret_interval
);
915 static void rt_secret_rebuild_oneshot(struct net
*net
)
917 del_timer_sync(&net
->ipv4
.rt_secret_timer
);
918 rt_cache_invalidate(net
);
919 if (ip_rt_secret_interval
) {
920 net
->ipv4
.rt_secret_timer
.expires
+= ip_rt_secret_interval
;
921 add_timer(&net
->ipv4
.rt_secret_timer
);
925 static void rt_emergency_hash_rebuild(struct net
*net
)
927 if (net_ratelimit()) {
928 printk(KERN_WARNING
"Route hash chain too long!\n");
929 printk(KERN_WARNING
"Adjust your secret_interval!\n");
932 rt_secret_rebuild_oneshot(net
);
936 Short description of GC goals.
938 We want to build algorithm, which will keep routing cache
939 at some equilibrium point, when number of aged off entries
940 is kept approximately equal to newly generated ones.
942 Current expiration strength is variable "expire".
943 We try to adjust it dynamically, so that if networking
944 is idle expires is large enough to keep enough of warm entries,
945 and when load increases it reduces to limit cache size.
948 static int rt_garbage_collect(struct dst_ops
*ops
)
950 static unsigned long expire
= RT_GC_TIMEOUT
;
951 static unsigned long last_gc
;
953 static int equilibrium
;
954 struct rtable
*rth
, **rthp
;
955 unsigned long now
= jiffies
;
959 * Garbage collection is pretty expensive,
960 * do not make it too frequently.
963 RT_CACHE_STAT_INC(gc_total
);
965 if (now
- last_gc
< ip_rt_gc_min_interval
&&
966 atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
) {
967 RT_CACHE_STAT_INC(gc_ignored
);
971 /* Calculate number of entries, which we want to expire now. */
972 goal
= atomic_read(&ipv4_dst_ops
.entries
) -
973 (ip_rt_gc_elasticity
<< rt_hash_log
);
975 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
976 equilibrium
= ipv4_dst_ops
.gc_thresh
;
977 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
979 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
980 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
983 /* We are in dangerous area. Try to reduce cache really
986 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
987 equilibrium
= atomic_read(&ipv4_dst_ops
.entries
) - goal
;
990 if (now
- last_gc
>= ip_rt_gc_min_interval
)
1001 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
1002 unsigned long tmo
= expire
;
1004 k
= (k
+ 1) & rt_hash_mask
;
1005 rthp
= &rt_hash_table
[k
].chain
;
1006 spin_lock_bh(rt_hash_lock_addr(k
));
1007 while ((rth
= *rthp
) != NULL
) {
1008 if (!rt_is_expired(rth
) &&
1009 !rt_may_expire(rth
, tmo
, expire
)) {
1011 rthp
= &rth
->u
.dst
.rt_next
;
1014 *rthp
= rth
->u
.dst
.rt_next
;
1018 spin_unlock_bh(rt_hash_lock_addr(k
));
1027 /* Goal is not achieved. We stop process if:
1029 - if expire reduced to zero. Otherwise, expire is halfed.
1030 - if table is not full.
1031 - if we are called from interrupt.
1032 - jiffies check is just fallback/debug loop breaker.
1033 We will not spin here for long time in any case.
1036 RT_CACHE_STAT_INC(gc_goal_miss
);
1042 #if RT_CACHE_DEBUG >= 2
1043 printk(KERN_DEBUG
"expire>> %u %d %d %d\n", expire
,
1044 atomic_read(&ipv4_dst_ops
.entries
), goal
, i
);
1047 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1049 } while (!in_softirq() && time_before_eq(jiffies
, now
));
1051 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1053 if (net_ratelimit())
1054 printk(KERN_WARNING
"dst cache overflow\n");
1055 RT_CACHE_STAT_INC(gc_dst_overflow
);
1059 expire
+= ip_rt_gc_min_interval
;
1060 if (expire
> ip_rt_gc_timeout
||
1061 atomic_read(&ipv4_dst_ops
.entries
) < ipv4_dst_ops
.gc_thresh
)
1062 expire
= ip_rt_gc_timeout
;
1063 #if RT_CACHE_DEBUG >= 2
1064 printk(KERN_DEBUG
"expire++ %u %d %d %d\n", expire
,
1065 atomic_read(&ipv4_dst_ops
.entries
), goal
, rover
);
1070 static int rt_intern_hash(unsigned hash
, struct rtable
*rt
,
1071 struct rtable
**rp
, struct sk_buff
*skb
)
1073 struct rtable
*rth
, **rthp
;
1075 struct rtable
*cand
, **candp
;
1078 int attempts
= !in_softirq();
1082 min_score
= ~(u32
)0;
1087 if (!rt_caching(dev_net(rt
->u
.dst
.dev
))) {
1089 * If we're not caching, just tell the caller we
1090 * were successful and don't touch the route. The
1091 * caller hold the sole reference to the cache entry, and
1092 * it will be released when the caller is done with it.
1093 * If we drop it here, the callers have no way to resolve routes
1094 * when we're not caching. Instead, just point *rp at rt, so
1095 * the caller gets a single use out of the route
1096 * Note that we do rt_free on this new route entry, so that
1097 * once its refcount hits zero, we are still able to reap it
1099 * Note also the rt_free uses call_rcu. We don't actually
1100 * need rcu protection here, this is just our path to get
1101 * on the route gc list.
1104 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1105 int err
= arp_bind_neighbour(&rt
->u
.dst
);
1107 if (net_ratelimit())
1109 "Neighbour table failure & not caching routes.\n");
1119 rthp
= &rt_hash_table
[hash
].chain
;
1121 spin_lock_bh(rt_hash_lock_addr(hash
));
1122 while ((rth
= *rthp
) != NULL
) {
1123 if (rt_is_expired(rth
)) {
1124 *rthp
= rth
->u
.dst
.rt_next
;
1128 if (compare_keys(&rth
->fl
, &rt
->fl
) && compare_netns(rth
, rt
)) {
1130 *rthp
= rth
->u
.dst
.rt_next
;
1132 * Since lookup is lockfree, the deletion
1133 * must be visible to another weakly ordered CPU before
1134 * the insertion at the start of the hash chain.
1136 rcu_assign_pointer(rth
->u
.dst
.rt_next
,
1137 rt_hash_table
[hash
].chain
);
1139 * Since lookup is lockfree, the update writes
1140 * must be ordered for consistency on SMP.
1142 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1144 dst_use(&rth
->u
.dst
, now
);
1145 spin_unlock_bh(rt_hash_lock_addr(hash
));
1151 skb_dst_set(skb
, &rth
->u
.dst
);
1155 if (!atomic_read(&rth
->u
.dst
.__refcnt
)) {
1156 u32 score
= rt_score(rth
);
1158 if (score
<= min_score
) {
1167 rthp
= &rth
->u
.dst
.rt_next
;
1171 /* ip_rt_gc_elasticity used to be average length of chain
1172 * length, when exceeded gc becomes really aggressive.
1174 * The second limit is less certain. At the moment it allows
1175 * only 2 entries per bucket. We will see.
1177 if (chain_length
> ip_rt_gc_elasticity
) {
1178 *candp
= cand
->u
.dst
.rt_next
;
1182 if (chain_length
> rt_chain_length_max
) {
1183 struct net
*net
= dev_net(rt
->u
.dst
.dev
);
1184 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1185 if (!rt_caching(dev_net(rt
->u
.dst
.dev
))) {
1186 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1187 rt
->u
.dst
.dev
->name
, num
);
1189 rt_emergency_hash_rebuild(dev_net(rt
->u
.dst
.dev
));
1193 /* Try to bind route to arp only if it is output
1194 route or unicast forwarding path.
1196 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1197 int err
= arp_bind_neighbour(&rt
->u
.dst
);
1199 spin_unlock_bh(rt_hash_lock_addr(hash
));
1201 if (err
!= -ENOBUFS
) {
1206 /* Neighbour tables are full and nothing
1207 can be released. Try to shrink route cache,
1208 it is most likely it holds some neighbour records.
1210 if (attempts
-- > 0) {
1211 int saved_elasticity
= ip_rt_gc_elasticity
;
1212 int saved_int
= ip_rt_gc_min_interval
;
1213 ip_rt_gc_elasticity
= 1;
1214 ip_rt_gc_min_interval
= 0;
1215 rt_garbage_collect(&ipv4_dst_ops
);
1216 ip_rt_gc_min_interval
= saved_int
;
1217 ip_rt_gc_elasticity
= saved_elasticity
;
1221 if (net_ratelimit())
1222 printk(KERN_WARNING
"Neighbour table overflow.\n");
1228 rt
->u
.dst
.rt_next
= rt_hash_table
[hash
].chain
;
1230 #if RT_CACHE_DEBUG >= 2
1231 if (rt
->u
.dst
.rt_next
) {
1233 printk(KERN_DEBUG
"rt_cache @%02x: %pI4",
1235 for (trt
= rt
->u
.dst
.rt_next
; trt
; trt
= trt
->u
.dst
.rt_next
)
1236 printk(" . %pI4", &trt
->rt_dst
);
1241 * Since lookup is lockfree, we must make sure
1242 * previous writes to rt are comitted to memory
1243 * before making rt visible to other CPUS.
1245 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1247 spin_unlock_bh(rt_hash_lock_addr(hash
));
1253 skb_dst_set(skb
, &rt
->u
.dst
);
1257 void rt_bind_peer(struct rtable
*rt
, int create
)
1259 static DEFINE_SPINLOCK(rt_peer_lock
);
1260 struct inet_peer
*peer
;
1262 peer
= inet_getpeer(rt
->rt_dst
, create
);
1264 spin_lock_bh(&rt_peer_lock
);
1265 if (rt
->peer
== NULL
) {
1269 spin_unlock_bh(&rt_peer_lock
);
1275 * Peer allocation may fail only in serious out-of-memory conditions. However
1276 * we still can generate some output.
1277 * Random ID selection looks a bit dangerous because we have no chances to
1278 * select ID being unique in a reasonable period of time.
1279 * But broken packet identifier may be better than no packet at all.
1281 static void ip_select_fb_ident(struct iphdr
*iph
)
1283 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1284 static u32 ip_fallback_id
;
1287 spin_lock_bh(&ip_fb_id_lock
);
1288 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1289 iph
->id
= htons(salt
& 0xFFFF);
1290 ip_fallback_id
= salt
;
1291 spin_unlock_bh(&ip_fb_id_lock
);
1294 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1296 struct rtable
*rt
= (struct rtable
*) dst
;
1299 if (rt
->peer
== NULL
)
1300 rt_bind_peer(rt
, 1);
1302 /* If peer is attached to destination, it is never detached,
1303 so that we need not to grab a lock to dereference it.
1306 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1310 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1311 __builtin_return_address(0));
1313 ip_select_fb_ident(iph
);
1316 static void rt_del(unsigned hash
, struct rtable
*rt
)
1318 struct rtable
**rthp
, *aux
;
1320 rthp
= &rt_hash_table
[hash
].chain
;
1321 spin_lock_bh(rt_hash_lock_addr(hash
));
1323 while ((aux
= *rthp
) != NULL
) {
1324 if (aux
== rt
|| rt_is_expired(aux
)) {
1325 *rthp
= aux
->u
.dst
.rt_next
;
1329 rthp
= &aux
->u
.dst
.rt_next
;
1331 spin_unlock_bh(rt_hash_lock_addr(hash
));
1334 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1335 __be32 saddr
, struct net_device
*dev
)
1338 struct in_device
*in_dev
= in_dev_get(dev
);
1339 struct rtable
*rth
, **rthp
;
1340 __be32 skeys
[2] = { saddr
, 0 };
1341 int ikeys
[2] = { dev
->ifindex
, 0 };
1342 struct netevent_redirect netevent
;
1349 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1350 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1351 ipv4_is_zeronet(new_gw
))
1352 goto reject_redirect
;
1354 if (!rt_caching(net
))
1355 goto reject_redirect
;
1357 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1358 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1359 goto reject_redirect
;
1360 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1361 goto reject_redirect
;
1363 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1364 goto reject_redirect
;
1367 for (i
= 0; i
< 2; i
++) {
1368 for (k
= 0; k
< 2; k
++) {
1369 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1372 rthp
=&rt_hash_table
[hash
].chain
;
1375 while ((rth
= rcu_dereference(*rthp
)) != NULL
) {
1378 if (rth
->fl
.fl4_dst
!= daddr
||
1379 rth
->fl
.fl4_src
!= skeys
[i
] ||
1380 rth
->fl
.oif
!= ikeys
[k
] ||
1382 rt_is_expired(rth
) ||
1383 !net_eq(dev_net(rth
->u
.dst
.dev
), net
)) {
1384 rthp
= &rth
->u
.dst
.rt_next
;
1388 if (rth
->rt_dst
!= daddr
||
1389 rth
->rt_src
!= saddr
||
1391 rth
->rt_gateway
!= old_gw
||
1392 rth
->u
.dst
.dev
!= dev
)
1395 dst_hold(&rth
->u
.dst
);
1398 rt
= dst_alloc(&ipv4_dst_ops
);
1405 /* Copy all the information. */
1407 rt
->u
.dst
.__use
= 1;
1408 atomic_set(&rt
->u
.dst
.__refcnt
, 1);
1409 rt
->u
.dst
.child
= NULL
;
1411 dev_hold(rt
->u
.dst
.dev
);
1413 in_dev_hold(rt
->idev
);
1414 rt
->u
.dst
.obsolete
= 0;
1415 rt
->u
.dst
.lastuse
= jiffies
;
1416 rt
->u
.dst
.path
= &rt
->u
.dst
;
1417 rt
->u
.dst
.neighbour
= NULL
;
1418 rt
->u
.dst
.hh
= NULL
;
1420 rt
->u
.dst
.xfrm
= NULL
;
1422 rt
->rt_genid
= rt_genid(net
);
1423 rt
->rt_flags
|= RTCF_REDIRECTED
;
1425 /* Gateway is different ... */
1426 rt
->rt_gateway
= new_gw
;
1428 /* Redirect received -> path was valid */
1429 dst_confirm(&rth
->u
.dst
);
1432 atomic_inc(&rt
->peer
->refcnt
);
1434 if (arp_bind_neighbour(&rt
->u
.dst
) ||
1435 !(rt
->u
.dst
.neighbour
->nud_state
&
1437 if (rt
->u
.dst
.neighbour
)
1438 neigh_event_send(rt
->u
.dst
.neighbour
, NULL
);
1444 netevent
.old
= &rth
->u
.dst
;
1445 netevent
.new = &rt
->u
.dst
;
1446 call_netevent_notifiers(NETEVENT_REDIRECT
,
1450 if (!rt_intern_hash(hash
, rt
, &rt
, NULL
))
1463 #ifdef CONFIG_IP_ROUTE_VERBOSE
1464 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1465 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1466 " Advised path = %pI4 -> %pI4\n",
1467 &old_gw
, dev
->name
, &new_gw
,
1473 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1475 struct rtable
*rt
= (struct rtable
*)dst
;
1476 struct dst_entry
*ret
= dst
;
1479 if (dst
->obsolete
) {
1482 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
1483 rt
->u
.dst
.expires
) {
1484 unsigned hash
= rt_hash(rt
->fl
.fl4_dst
, rt
->fl
.fl4_src
,
1486 rt_genid(dev_net(dst
->dev
)));
1487 #if RT_CACHE_DEBUG >= 1
1488 printk(KERN_DEBUG
"ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1489 &rt
->rt_dst
, rt
->fl
.fl4_tos
);
1500 * 1. The first ip_rt_redirect_number redirects are sent
1501 * with exponential backoff, then we stop sending them at all,
1502 * assuming that the host ignores our redirects.
1503 * 2. If we did not see packets requiring redirects
1504 * during ip_rt_redirect_silence, we assume that the host
1505 * forgot redirected route and start to send redirects again.
1507 * This algorithm is much cheaper and more intelligent than dumb load limiting
1510 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1511 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1514 void ip_rt_send_redirect(struct sk_buff
*skb
)
1516 struct rtable
*rt
= skb_rtable(skb
);
1517 struct in_device
*in_dev
;
1521 in_dev
= __in_dev_get_rcu(rt
->u
.dst
.dev
);
1522 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1526 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1529 /* No redirected packets during ip_rt_redirect_silence;
1530 * reset the algorithm.
1532 if (time_after(jiffies
, rt
->u
.dst
.rate_last
+ ip_rt_redirect_silence
))
1533 rt
->u
.dst
.rate_tokens
= 0;
1535 /* Too many ignored redirects; do not send anything
1536 * set u.dst.rate_last to the last seen redirected packet.
1538 if (rt
->u
.dst
.rate_tokens
>= ip_rt_redirect_number
) {
1539 rt
->u
.dst
.rate_last
= jiffies
;
1543 /* Check for load limit; set rate_last to the latest sent
1546 if (rt
->u
.dst
.rate_tokens
== 0 ||
1548 (rt
->u
.dst
.rate_last
+
1549 (ip_rt_redirect_load
<< rt
->u
.dst
.rate_tokens
)))) {
1550 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1551 rt
->u
.dst
.rate_last
= jiffies
;
1552 ++rt
->u
.dst
.rate_tokens
;
1553 #ifdef CONFIG_IP_ROUTE_VERBOSE
1555 rt
->u
.dst
.rate_tokens
== ip_rt_redirect_number
&&
1557 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1558 &rt
->rt_src
, rt
->rt_iif
,
1559 &rt
->rt_dst
, &rt
->rt_gateway
);
1564 static int ip_error(struct sk_buff
*skb
)
1566 struct rtable
*rt
= skb_rtable(skb
);
1570 switch (rt
->u
.dst
.error
) {
1575 code
= ICMP_HOST_UNREACH
;
1578 code
= ICMP_NET_UNREACH
;
1579 IP_INC_STATS_BH(dev_net(rt
->u
.dst
.dev
),
1580 IPSTATS_MIB_INNOROUTES
);
1583 code
= ICMP_PKT_FILTERED
;
1588 rt
->u
.dst
.rate_tokens
+= now
- rt
->u
.dst
.rate_last
;
1589 if (rt
->u
.dst
.rate_tokens
> ip_rt_error_burst
)
1590 rt
->u
.dst
.rate_tokens
= ip_rt_error_burst
;
1591 rt
->u
.dst
.rate_last
= now
;
1592 if (rt
->u
.dst
.rate_tokens
>= ip_rt_error_cost
) {
1593 rt
->u
.dst
.rate_tokens
-= ip_rt_error_cost
;
1594 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1597 out
: kfree_skb(skb
);
1602 * The last two values are not from the RFC but
1603 * are needed for AMPRnet AX.25 paths.
1606 static const unsigned short mtu_plateau
[] =
1607 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1609 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1613 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1614 if (old_mtu
> mtu_plateau
[i
])
1615 return mtu_plateau
[i
];
1619 unsigned short ip_rt_frag_needed(struct net
*net
, struct iphdr
*iph
,
1620 unsigned short new_mtu
,
1621 struct net_device
*dev
)
1624 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1626 int ikeys
[2] = { dev
->ifindex
, 0 };
1627 __be32 skeys
[2] = { iph
->saddr
, 0, };
1628 __be32 daddr
= iph
->daddr
;
1629 unsigned short est_mtu
= 0;
1631 for (k
= 0; k
< 2; k
++) {
1632 for (i
= 0; i
< 2; i
++) {
1633 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1637 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
1638 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
1639 unsigned short mtu
= new_mtu
;
1641 if (rth
->fl
.fl4_dst
!= daddr
||
1642 rth
->fl
.fl4_src
!= skeys
[i
] ||
1643 rth
->rt_dst
!= daddr
||
1644 rth
->rt_src
!= iph
->saddr
||
1645 rth
->fl
.oif
!= ikeys
[k
] ||
1647 dst_metric_locked(&rth
->u
.dst
, RTAX_MTU
) ||
1648 !net_eq(dev_net(rth
->u
.dst
.dev
), net
) ||
1652 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1654 /* BSD 4.2 compatibility hack :-( */
1656 old_mtu
>= dst_mtu(&rth
->u
.dst
) &&
1657 old_mtu
>= 68 + (iph
->ihl
<< 2))
1658 old_mtu
-= iph
->ihl
<< 2;
1660 mtu
= guess_mtu(old_mtu
);
1662 if (mtu
<= dst_mtu(&rth
->u
.dst
)) {
1663 if (mtu
< dst_mtu(&rth
->u
.dst
)) {
1664 dst_confirm(&rth
->u
.dst
);
1665 if (mtu
< ip_rt_min_pmtu
) {
1666 mtu
= ip_rt_min_pmtu
;
1667 rth
->u
.dst
.metrics
[RTAX_LOCK
-1] |=
1670 rth
->u
.dst
.metrics
[RTAX_MTU
-1] = mtu
;
1671 dst_set_expires(&rth
->u
.dst
,
1680 return est_mtu
? : new_mtu
;
1683 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1685 if (dst_mtu(dst
) > mtu
&& mtu
>= 68 &&
1686 !(dst_metric_locked(dst
, RTAX_MTU
))) {
1687 if (mtu
< ip_rt_min_pmtu
) {
1688 mtu
= ip_rt_min_pmtu
;
1689 dst
->metrics
[RTAX_LOCK
-1] |= (1 << RTAX_MTU
);
1691 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1692 dst_set_expires(dst
, ip_rt_mtu_expires
);
1693 call_netevent_notifiers(NETEVENT_PMTU_UPDATE
, dst
);
1697 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1702 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1704 struct rtable
*rt
= (struct rtable
*) dst
;
1705 struct inet_peer
*peer
= rt
->peer
;
1706 struct in_device
*idev
= rt
->idev
;
1719 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
1722 struct rtable
*rt
= (struct rtable
*) dst
;
1723 struct in_device
*idev
= rt
->idev
;
1724 if (dev
!= dev_net(dev
)->loopback_dev
&& idev
&& idev
->dev
== dev
) {
1725 struct in_device
*loopback_idev
=
1726 in_dev_get(dev_net(dev
)->loopback_dev
);
1727 if (loopback_idev
) {
1728 rt
->idev
= loopback_idev
;
1734 static void ipv4_link_failure(struct sk_buff
*skb
)
1738 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1740 rt
= skb_rtable(skb
);
1742 dst_set_expires(&rt
->u
.dst
, 0);
1745 static int ip_rt_bug(struct sk_buff
*skb
)
1747 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1748 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1749 skb
->dev
? skb
->dev
->name
: "?");
1755 We do not cache source address of outgoing interface,
1756 because it is used only by IP RR, TS and SRR options,
1757 so that it out of fast path.
1759 BTW remember: "addr" is allowed to be not aligned
1763 void ip_rt_get_source(u8
*addr
, struct rtable
*rt
)
1766 struct fib_result res
;
1768 if (rt
->fl
.iif
== 0)
1770 else if (fib_lookup(dev_net(rt
->u
.dst
.dev
), &rt
->fl
, &res
) == 0) {
1771 src
= FIB_RES_PREFSRC(res
);
1774 src
= inet_select_addr(rt
->u
.dst
.dev
, rt
->rt_gateway
,
1776 memcpy(addr
, &src
, 4);
1779 #ifdef CONFIG_NET_CLS_ROUTE
1780 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1782 if (!(rt
->u
.dst
.tclassid
& 0xFFFF))
1783 rt
->u
.dst
.tclassid
|= tag
& 0xFFFF;
1784 if (!(rt
->u
.dst
.tclassid
& 0xFFFF0000))
1785 rt
->u
.dst
.tclassid
|= tag
& 0xFFFF0000;
1789 static void rt_set_nexthop(struct rtable
*rt
, struct fib_result
*res
, u32 itag
)
1791 struct fib_info
*fi
= res
->fi
;
1794 if (FIB_RES_GW(*res
) &&
1795 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1796 rt
->rt_gateway
= FIB_RES_GW(*res
);
1797 memcpy(rt
->u
.dst
.metrics
, fi
->fib_metrics
,
1798 sizeof(rt
->u
.dst
.metrics
));
1799 if (fi
->fib_mtu
== 0) {
1800 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = rt
->u
.dst
.dev
->mtu
;
1801 if (dst_metric_locked(&rt
->u
.dst
, RTAX_MTU
) &&
1802 rt
->rt_gateway
!= rt
->rt_dst
&&
1803 rt
->u
.dst
.dev
->mtu
> 576)
1804 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = 576;
1806 #ifdef CONFIG_NET_CLS_ROUTE
1807 rt
->u
.dst
.tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1810 rt
->u
.dst
.metrics
[RTAX_MTU
-1]= rt
->u
.dst
.dev
->mtu
;
1812 if (dst_metric(&rt
->u
.dst
, RTAX_HOPLIMIT
) == 0)
1813 rt
->u
.dst
.metrics
[RTAX_HOPLIMIT
-1] = sysctl_ip_default_ttl
;
1814 if (dst_mtu(&rt
->u
.dst
) > IP_MAX_MTU
)
1815 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = IP_MAX_MTU
;
1816 if (dst_metric(&rt
->u
.dst
, RTAX_ADVMSS
) == 0)
1817 rt
->u
.dst
.metrics
[RTAX_ADVMSS
-1] = max_t(unsigned int, rt
->u
.dst
.dev
->mtu
- 40,
1819 if (dst_metric(&rt
->u
.dst
, RTAX_ADVMSS
) > 65535 - 40)
1820 rt
->u
.dst
.metrics
[RTAX_ADVMSS
-1] = 65535 - 40;
1822 #ifdef CONFIG_NET_CLS_ROUTE
1823 #ifdef CONFIG_IP_MULTIPLE_TABLES
1824 set_class_tag(rt
, fib_rules_tclass(res
));
1826 set_class_tag(rt
, itag
);
1828 rt
->rt_type
= res
->type
;
1831 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1832 u8 tos
, struct net_device
*dev
, int our
)
1837 struct in_device
*in_dev
= in_dev_get(dev
);
1840 /* Primary sanity checks. */
1845 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1846 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
1849 if (ipv4_is_zeronet(saddr
)) {
1850 if (!ipv4_is_local_multicast(daddr
))
1852 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
1853 } else if (fib_validate_source(saddr
, 0, tos
, 0,
1854 dev
, &spec_dst
, &itag
, 0) < 0)
1857 rth
= dst_alloc(&ipv4_dst_ops
);
1861 rth
->u
.dst
.output
= ip_rt_bug
;
1863 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
1864 rth
->u
.dst
.flags
= DST_HOST
;
1865 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
1866 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
1867 rth
->fl
.fl4_dst
= daddr
;
1868 rth
->rt_dst
= daddr
;
1869 rth
->fl
.fl4_tos
= tos
;
1870 rth
->fl
.mark
= skb
->mark
;
1871 rth
->fl
.fl4_src
= saddr
;
1872 rth
->rt_src
= saddr
;
1873 #ifdef CONFIG_NET_CLS_ROUTE
1874 rth
->u
.dst
.tclassid
= itag
;
1877 rth
->fl
.iif
= dev
->ifindex
;
1878 rth
->u
.dst
.dev
= init_net
.loopback_dev
;
1879 dev_hold(rth
->u
.dst
.dev
);
1880 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
1882 rth
->rt_gateway
= daddr
;
1883 rth
->rt_spec_dst
= spec_dst
;
1884 rth
->rt_genid
= rt_genid(dev_net(dev
));
1885 rth
->rt_flags
= RTCF_MULTICAST
;
1886 rth
->rt_type
= RTN_MULTICAST
;
1888 rth
->u
.dst
.input
= ip_local_deliver
;
1889 rth
->rt_flags
|= RTCF_LOCAL
;
1892 #ifdef CONFIG_IP_MROUTE
1893 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1894 rth
->u
.dst
.input
= ip_mr_input
;
1896 RT_CACHE_STAT_INC(in_slow_mc
);
1899 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
1900 return rt_intern_hash(hash
, rth
, NULL
, skb
);
1912 static void ip_handle_martian_source(struct net_device
*dev
,
1913 struct in_device
*in_dev
,
1914 struct sk_buff
*skb
,
1918 RT_CACHE_STAT_INC(in_martian_src
);
1919 #ifdef CONFIG_IP_ROUTE_VERBOSE
1920 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1922 * RFC1812 recommendation, if source is martian,
1923 * the only hint is MAC header.
1925 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
1926 &daddr
, &saddr
, dev
->name
);
1927 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1929 const unsigned char *p
= skb_mac_header(skb
);
1930 printk(KERN_WARNING
"ll header: ");
1931 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
1933 if (i
< (dev
->hard_header_len
- 1))
1942 static int __mkroute_input(struct sk_buff
*skb
,
1943 struct fib_result
*res
,
1944 struct in_device
*in_dev
,
1945 __be32 daddr
, __be32 saddr
, u32 tos
,
1946 struct rtable
**result
)
1951 struct in_device
*out_dev
;
1956 /* get a working reference to the output device */
1957 out_dev
= in_dev_get(FIB_RES_DEV(*res
));
1958 if (out_dev
== NULL
) {
1959 if (net_ratelimit())
1960 printk(KERN_CRIT
"Bug in ip_route_input" \
1961 "_slow(). Please, report\n");
1966 err
= fib_validate_source(saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1967 in_dev
->dev
, &spec_dst
, &itag
, skb
->mark
);
1969 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1977 flags
|= RTCF_DIRECTSRC
;
1979 if (out_dev
== in_dev
&& err
&&
1980 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1981 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
1982 flags
|= RTCF_DOREDIRECT
;
1984 if (skb
->protocol
!= htons(ETH_P_IP
)) {
1985 /* Not IP (i.e. ARP). Do not create route, if it is
1986 * invalid for proxy arp. DNAT routes are always valid.
1988 if (out_dev
== in_dev
) {
1995 rth
= dst_alloc(&ipv4_dst_ops
);
2001 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2002 rth
->u
.dst
.flags
= DST_HOST
;
2003 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2004 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2005 if (IN_DEV_CONF_GET(out_dev
, NOXFRM
))
2006 rth
->u
.dst
.flags
|= DST_NOXFRM
;
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 rth
->rt_gateway
= daddr
;
2015 rth
->fl
.iif
= in_dev
->dev
->ifindex
;
2016 rth
->u
.dst
.dev
= (out_dev
)->dev
;
2017 dev_hold(rth
->u
.dst
.dev
);
2018 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
2020 rth
->rt_spec_dst
= spec_dst
;
2022 rth
->u
.dst
.input
= ip_forward
;
2023 rth
->u
.dst
.output
= ip_output
;
2024 rth
->rt_genid
= rt_genid(dev_net(rth
->u
.dst
.dev
));
2026 rt_set_nexthop(rth
, res
, itag
);
2028 rth
->rt_flags
= flags
;
2033 /* release the working reference to the output device */
2034 in_dev_put(out_dev
);
2038 static int ip_mkroute_input(struct sk_buff
*skb
,
2039 struct fib_result
*res
,
2040 const struct flowi
*fl
,
2041 struct in_device
*in_dev
,
2042 __be32 daddr
, __be32 saddr
, u32 tos
)
2044 struct rtable
* rth
= NULL
;
2048 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2049 if (res
->fi
&& res
->fi
->fib_nhs
> 1 && fl
->oif
== 0)
2050 fib_select_multipath(fl
, res
);
2053 /* create a routing cache entry */
2054 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2058 /* put it into the cache */
2059 hash
= rt_hash(daddr
, saddr
, fl
->iif
,
2060 rt_genid(dev_net(rth
->u
.dst
.dev
)));
2061 return rt_intern_hash(hash
, rth
, NULL
, skb
);
2065 * NOTE. We drop all the packets that has local source
2066 * addresses, because every properly looped back packet
2067 * must have correct destination already attached by output routine.
2069 * Such approach solves two big problems:
2070 * 1. Not simplex devices are handled properly.
2071 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2074 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2075 u8 tos
, struct net_device
*dev
)
2077 struct fib_result res
;
2078 struct in_device
*in_dev
= in_dev_get(dev
);
2079 struct flowi fl
= { .nl_u
= { .ip4_u
=
2083 .scope
= RT_SCOPE_UNIVERSE
,
2086 .iif
= dev
->ifindex
};
2089 struct rtable
* rth
;
2094 struct net
* net
= dev_net(dev
);
2096 /* IP on this device is disabled. */
2101 /* Check for the most weird martians, which can be not detected
2105 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2106 ipv4_is_loopback(saddr
))
2107 goto martian_source
;
2109 if (daddr
== htonl(0xFFFFFFFF) || (saddr
== 0 && daddr
== 0))
2112 /* Accept zero addresses only to limited broadcast;
2113 * I even do not know to fix it or not. Waiting for complains :-)
2115 if (ipv4_is_zeronet(saddr
))
2116 goto martian_source
;
2118 if (ipv4_is_lbcast(daddr
) || ipv4_is_zeronet(daddr
) ||
2119 ipv4_is_loopback(daddr
))
2120 goto martian_destination
;
2123 * Now we are ready to route packet.
2125 if ((err
= fib_lookup(net
, &fl
, &res
)) != 0) {
2126 if (!IN_DEV_FORWARD(in_dev
))
2132 RT_CACHE_STAT_INC(in_slow_tot
);
2134 if (res
.type
== RTN_BROADCAST
)
2137 if (res
.type
== RTN_LOCAL
) {
2139 result
= fib_validate_source(saddr
, daddr
, tos
,
2140 net
->loopback_dev
->ifindex
,
2141 dev
, &spec_dst
, &itag
, skb
->mark
);
2143 goto martian_source
;
2145 flags
|= RTCF_DIRECTSRC
;
2150 if (!IN_DEV_FORWARD(in_dev
))
2152 if (res
.type
!= RTN_UNICAST
)
2153 goto martian_destination
;
2155 err
= ip_mkroute_input(skb
, &res
, &fl
, in_dev
, daddr
, saddr
, tos
);
2163 if (skb
->protocol
!= htons(ETH_P_IP
))
2166 if (ipv4_is_zeronet(saddr
))
2167 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2169 err
= fib_validate_source(saddr
, 0, tos
, 0, dev
, &spec_dst
,
2172 goto martian_source
;
2174 flags
|= RTCF_DIRECTSRC
;
2176 flags
|= RTCF_BROADCAST
;
2177 res
.type
= RTN_BROADCAST
;
2178 RT_CACHE_STAT_INC(in_brd
);
2181 rth
= dst_alloc(&ipv4_dst_ops
);
2185 rth
->u
.dst
.output
= ip_rt_bug
;
2186 rth
->rt_genid
= rt_genid(net
);
2188 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2189 rth
->u
.dst
.flags
= DST_HOST
;
2190 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2191 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2192 rth
->fl
.fl4_dst
= daddr
;
2193 rth
->rt_dst
= daddr
;
2194 rth
->fl
.fl4_tos
= tos
;
2195 rth
->fl
.mark
= skb
->mark
;
2196 rth
->fl
.fl4_src
= saddr
;
2197 rth
->rt_src
= saddr
;
2198 #ifdef CONFIG_NET_CLS_ROUTE
2199 rth
->u
.dst
.tclassid
= itag
;
2202 rth
->fl
.iif
= dev
->ifindex
;
2203 rth
->u
.dst
.dev
= net
->loopback_dev
;
2204 dev_hold(rth
->u
.dst
.dev
);
2205 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
2206 rth
->rt_gateway
= daddr
;
2207 rth
->rt_spec_dst
= spec_dst
;
2208 rth
->u
.dst
.input
= ip_local_deliver
;
2209 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2210 if (res
.type
== RTN_UNREACHABLE
) {
2211 rth
->u
.dst
.input
= ip_error
;
2212 rth
->u
.dst
.error
= -err
;
2213 rth
->rt_flags
&= ~RTCF_LOCAL
;
2215 rth
->rt_type
= res
.type
;
2216 hash
= rt_hash(daddr
, saddr
, fl
.iif
, rt_genid(net
));
2217 err
= rt_intern_hash(hash
, rth
, NULL
, skb
);
2221 RT_CACHE_STAT_INC(in_no_route
);
2222 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2223 res
.type
= RTN_UNREACHABLE
;
2229 * Do not cache martian addresses: they should be logged (RFC1812)
2231 martian_destination
:
2232 RT_CACHE_STAT_INC(in_martian_dst
);
2233 #ifdef CONFIG_IP_ROUTE_VERBOSE
2234 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2235 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2236 &daddr
, &saddr
, dev
->name
);
2240 err
= -EHOSTUNREACH
;
2252 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2256 int ip_route_input(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2257 u8 tos
, struct net_device
*dev
)
2259 struct rtable
* rth
;
2261 int iif
= dev
->ifindex
;
2266 if (!rt_caching(net
))
2269 tos
&= IPTOS_RT_MASK
;
2270 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2273 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2274 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
2275 if (((rth
->fl
.fl4_dst
^ daddr
) |
2276 (rth
->fl
.fl4_src
^ saddr
) |
2277 (rth
->fl
.iif
^ iif
) |
2279 (rth
->fl
.fl4_tos
^ tos
)) == 0 &&
2280 rth
->fl
.mark
== skb
->mark
&&
2281 net_eq(dev_net(rth
->u
.dst
.dev
), net
) &&
2282 !rt_is_expired(rth
)) {
2283 dst_use(&rth
->u
.dst
, jiffies
);
2284 RT_CACHE_STAT_INC(in_hit
);
2286 skb_dst_set(skb
, &rth
->u
.dst
);
2289 RT_CACHE_STAT_INC(in_hlist_search
);
2294 /* Multicast recognition logic is moved from route cache to here.
2295 The problem was that too many Ethernet cards have broken/missing
2296 hardware multicast filters :-( As result the host on multicasting
2297 network acquires a lot of useless route cache entries, sort of
2298 SDR messages from all the world. Now we try to get rid of them.
2299 Really, provided software IP multicast filter is organized
2300 reasonably (at least, hashed), it does not result in a slowdown
2301 comparing with route cache reject entries.
2302 Note, that multicast routers are not affected, because
2303 route cache entry is created eventually.
2305 if (ipv4_is_multicast(daddr
)) {
2306 struct in_device
*in_dev
;
2309 if ((in_dev
= __in_dev_get_rcu(dev
)) != NULL
) {
2310 int our
= ip_check_mc(in_dev
, daddr
, saddr
,
2311 ip_hdr(skb
)->protocol
);
2313 #ifdef CONFIG_IP_MROUTE
2315 (!ipv4_is_local_multicast(daddr
) &&
2316 IN_DEV_MFORWARD(in_dev
))
2320 return ip_route_input_mc(skb
, daddr
, saddr
,
2327 return ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2330 static int __mkroute_output(struct rtable
**result
,
2331 struct fib_result
*res
,
2332 const struct flowi
*fl
,
2333 const struct flowi
*oldflp
,
2334 struct net_device
*dev_out
,
2338 struct in_device
*in_dev
;
2339 u32 tos
= RT_FL_TOS(oldflp
);
2342 if (ipv4_is_loopback(fl
->fl4_src
) && !(dev_out
->flags
&IFF_LOOPBACK
))
2345 if (fl
->fl4_dst
== htonl(0xFFFFFFFF))
2346 res
->type
= RTN_BROADCAST
;
2347 else if (ipv4_is_multicast(fl
->fl4_dst
))
2348 res
->type
= RTN_MULTICAST
;
2349 else if (ipv4_is_lbcast(fl
->fl4_dst
) || ipv4_is_zeronet(fl
->fl4_dst
))
2352 if (dev_out
->flags
& IFF_LOOPBACK
)
2353 flags
|= RTCF_LOCAL
;
2355 /* get work reference to inet device */
2356 in_dev
= in_dev_get(dev_out
);
2360 if (res
->type
== RTN_BROADCAST
) {
2361 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2363 fib_info_put(res
->fi
);
2366 } else if (res
->type
== RTN_MULTICAST
) {
2367 flags
|= RTCF_MULTICAST
|RTCF_LOCAL
;
2368 if (!ip_check_mc(in_dev
, oldflp
->fl4_dst
, oldflp
->fl4_src
,
2370 flags
&= ~RTCF_LOCAL
;
2371 /* If multicast route do not exist use
2372 default one, but do not gateway in this case.
2375 if (res
->fi
&& res
->prefixlen
< 4) {
2376 fib_info_put(res
->fi
);
2382 rth
= dst_alloc(&ipv4_dst_ops
);
2388 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2389 rth
->u
.dst
.flags
= DST_HOST
;
2390 if (IN_DEV_CONF_GET(in_dev
, NOXFRM
))
2391 rth
->u
.dst
.flags
|= DST_NOXFRM
;
2392 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2393 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2395 rth
->fl
.fl4_dst
= oldflp
->fl4_dst
;
2396 rth
->fl
.fl4_tos
= tos
;
2397 rth
->fl
.fl4_src
= oldflp
->fl4_src
;
2398 rth
->fl
.oif
= oldflp
->oif
;
2399 rth
->fl
.mark
= oldflp
->mark
;
2400 rth
->rt_dst
= fl
->fl4_dst
;
2401 rth
->rt_src
= fl
->fl4_src
;
2402 rth
->rt_iif
= oldflp
->oif
? : dev_out
->ifindex
;
2403 /* get references to the devices that are to be hold by the routing
2405 rth
->u
.dst
.dev
= dev_out
;
2407 rth
->idev
= in_dev_get(dev_out
);
2408 rth
->rt_gateway
= fl
->fl4_dst
;
2409 rth
->rt_spec_dst
= fl
->fl4_src
;
2411 rth
->u
.dst
.output
=ip_output
;
2412 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2414 RT_CACHE_STAT_INC(out_slow_tot
);
2416 if (flags
& RTCF_LOCAL
) {
2417 rth
->u
.dst
.input
= ip_local_deliver
;
2418 rth
->rt_spec_dst
= fl
->fl4_dst
;
2420 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2421 rth
->rt_spec_dst
= fl
->fl4_src
;
2422 if (flags
& RTCF_LOCAL
&&
2423 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2424 rth
->u
.dst
.output
= ip_mc_output
;
2425 RT_CACHE_STAT_INC(out_slow_mc
);
2427 #ifdef CONFIG_IP_MROUTE
2428 if (res
->type
== RTN_MULTICAST
) {
2429 if (IN_DEV_MFORWARD(in_dev
) &&
2430 !ipv4_is_local_multicast(oldflp
->fl4_dst
)) {
2431 rth
->u
.dst
.input
= ip_mr_input
;
2432 rth
->u
.dst
.output
= ip_mc_output
;
2438 rt_set_nexthop(rth
, res
, 0);
2440 rth
->rt_flags
= flags
;
2444 /* release work reference to inet device */
2450 static int ip_mkroute_output(struct rtable
**rp
,
2451 struct fib_result
*res
,
2452 const struct flowi
*fl
,
2453 const struct flowi
*oldflp
,
2454 struct net_device
*dev_out
,
2457 struct rtable
*rth
= NULL
;
2458 int err
= __mkroute_output(&rth
, res
, fl
, oldflp
, dev_out
, flags
);
2461 hash
= rt_hash(oldflp
->fl4_dst
, oldflp
->fl4_src
, oldflp
->oif
,
2462 rt_genid(dev_net(dev_out
)));
2463 err
= rt_intern_hash(hash
, rth
, rp
, NULL
);
2470 * Major route resolver routine.
2473 static int ip_route_output_slow(struct net
*net
, struct rtable
**rp
,
2474 const struct flowi
*oldflp
)
2476 u32 tos
= RT_FL_TOS(oldflp
);
2477 struct flowi fl
= { .nl_u
= { .ip4_u
=
2478 { .daddr
= oldflp
->fl4_dst
,
2479 .saddr
= oldflp
->fl4_src
,
2480 .tos
= tos
& IPTOS_RT_MASK
,
2481 .scope
= ((tos
& RTO_ONLINK
) ?
2485 .mark
= oldflp
->mark
,
2486 .iif
= net
->loopback_dev
->ifindex
,
2487 .oif
= oldflp
->oif
};
2488 struct fib_result res
;
2490 struct net_device
*dev_out
= NULL
;
2496 #ifdef CONFIG_IP_MULTIPLE_TABLES
2500 if (oldflp
->fl4_src
) {
2502 if (ipv4_is_multicast(oldflp
->fl4_src
) ||
2503 ipv4_is_lbcast(oldflp
->fl4_src
) ||
2504 ipv4_is_zeronet(oldflp
->fl4_src
))
2507 /* I removed check for oif == dev_out->oif here.
2508 It was wrong for two reasons:
2509 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2510 is assigned to multiple interfaces.
2511 2. Moreover, we are allowed to send packets with saddr
2512 of another iface. --ANK
2515 if (oldflp
->oif
== 0 &&
2516 (ipv4_is_multicast(oldflp
->fl4_dst
) ||
2517 oldflp
->fl4_dst
== htonl(0xFFFFFFFF))) {
2518 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2519 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2520 if (dev_out
== NULL
)
2523 /* Special hack: user can direct multicasts
2524 and limited broadcast via necessary interface
2525 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2526 This hack is not just for fun, it allows
2527 vic,vat and friends to work.
2528 They bind socket to loopback, set ttl to zero
2529 and expect that it will work.
2530 From the viewpoint of routing cache they are broken,
2531 because we are not allowed to build multicast path
2532 with loopback source addr (look, routing cache
2533 cannot know, that ttl is zero, so that packet
2534 will not leave this host and route is valid).
2535 Luckily, this hack is good workaround.
2538 fl
.oif
= dev_out
->ifindex
;
2542 if (!(oldflp
->flags
& FLOWI_FLAG_ANYSRC
)) {
2543 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2544 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2545 if (dev_out
== NULL
)
2554 dev_out
= dev_get_by_index(net
, oldflp
->oif
);
2556 if (dev_out
== NULL
)
2559 /* RACE: Check return value of inet_select_addr instead. */
2560 if (__in_dev_get_rtnl(dev_out
) == NULL
) {
2562 goto out
; /* Wrong error code */
2565 if (ipv4_is_local_multicast(oldflp
->fl4_dst
) ||
2566 oldflp
->fl4_dst
== htonl(0xFFFFFFFF)) {
2568 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2573 if (ipv4_is_multicast(oldflp
->fl4_dst
))
2574 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2576 else if (!oldflp
->fl4_dst
)
2577 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2583 fl
.fl4_dst
= fl
.fl4_src
;
2585 fl
.fl4_dst
= fl
.fl4_src
= htonl(INADDR_LOOPBACK
);
2588 dev_out
= net
->loopback_dev
;
2590 fl
.oif
= net
->loopback_dev
->ifindex
;
2591 res
.type
= RTN_LOCAL
;
2592 flags
|= RTCF_LOCAL
;
2596 if (fib_lookup(net
, &fl
, &res
)) {
2599 /* Apparently, routing tables are wrong. Assume,
2600 that the destination is on link.
2603 Because we are allowed to send to iface
2604 even if it has NO routes and NO assigned
2605 addresses. When oif is specified, routing
2606 tables are looked up with only one purpose:
2607 to catch if destination is gatewayed, rather than
2608 direct. Moreover, if MSG_DONTROUTE is set,
2609 we send packet, ignoring both routing tables
2610 and ifaddr state. --ANK
2613 We could make it even if oif is unknown,
2614 likely IPv6, but we do not.
2617 if (fl
.fl4_src
== 0)
2618 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2620 res
.type
= RTN_UNICAST
;
2630 if (res
.type
== RTN_LOCAL
) {
2632 fl
.fl4_src
= fl
.fl4_dst
;
2635 dev_out
= net
->loopback_dev
;
2637 fl
.oif
= dev_out
->ifindex
;
2639 fib_info_put(res
.fi
);
2641 flags
|= RTCF_LOCAL
;
2645 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2646 if (res
.fi
->fib_nhs
> 1 && fl
.oif
== 0)
2647 fib_select_multipath(&fl
, &res
);
2650 if (!res
.prefixlen
&& res
.type
== RTN_UNICAST
&& !fl
.oif
)
2651 fib_select_default(net
, &fl
, &res
);
2654 fl
.fl4_src
= FIB_RES_PREFSRC(res
);
2658 dev_out
= FIB_RES_DEV(res
);
2660 fl
.oif
= dev_out
->ifindex
;
2664 err
= ip_mkroute_output(rp
, &res
, &fl
, oldflp
, dev_out
, flags
);
2674 int __ip_route_output_key(struct net
*net
, struct rtable
**rp
,
2675 const struct flowi
*flp
)
2680 if (!rt_caching(net
))
2683 hash
= rt_hash(flp
->fl4_dst
, flp
->fl4_src
, flp
->oif
, rt_genid(net
));
2686 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2687 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
2688 if (rth
->fl
.fl4_dst
== flp
->fl4_dst
&&
2689 rth
->fl
.fl4_src
== flp
->fl4_src
&&
2691 rth
->fl
.oif
== flp
->oif
&&
2692 rth
->fl
.mark
== flp
->mark
&&
2693 !((rth
->fl
.fl4_tos
^ flp
->fl4_tos
) &
2694 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2695 net_eq(dev_net(rth
->u
.dst
.dev
), net
) &&
2696 !rt_is_expired(rth
)) {
2697 dst_use(&rth
->u
.dst
, jiffies
);
2698 RT_CACHE_STAT_INC(out_hit
);
2699 rcu_read_unlock_bh();
2703 RT_CACHE_STAT_INC(out_hlist_search
);
2705 rcu_read_unlock_bh();
2708 return ip_route_output_slow(net
, rp
, flp
);
2711 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2713 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2717 static struct dst_ops ipv4_dst_blackhole_ops
= {
2719 .protocol
= cpu_to_be16(ETH_P_IP
),
2720 .destroy
= ipv4_dst_destroy
,
2721 .check
= ipv4_dst_check
,
2722 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2723 .entries
= ATOMIC_INIT(0),
2727 static int ipv4_dst_blackhole(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2729 struct rtable
*ort
= *rp
;
2730 struct rtable
*rt
= (struct rtable
*)
2731 dst_alloc(&ipv4_dst_blackhole_ops
);
2734 struct dst_entry
*new = &rt
->u
.dst
;
2736 atomic_set(&new->__refcnt
, 1);
2738 new->input
= dst_discard
;
2739 new->output
= dst_discard
;
2740 memcpy(new->metrics
, ort
->u
.dst
.metrics
, RTAX_MAX
*sizeof(u32
));
2742 new->dev
= ort
->u
.dst
.dev
;
2748 rt
->idev
= ort
->idev
;
2750 in_dev_hold(rt
->idev
);
2751 rt
->rt_genid
= rt_genid(net
);
2752 rt
->rt_flags
= ort
->rt_flags
;
2753 rt
->rt_type
= ort
->rt_type
;
2754 rt
->rt_dst
= ort
->rt_dst
;
2755 rt
->rt_src
= ort
->rt_src
;
2756 rt
->rt_iif
= ort
->rt_iif
;
2757 rt
->rt_gateway
= ort
->rt_gateway
;
2758 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2759 rt
->peer
= ort
->peer
;
2761 atomic_inc(&rt
->peer
->refcnt
);
2766 dst_release(&(*rp
)->u
.dst
);
2768 return (rt
? 0 : -ENOMEM
);
2771 int ip_route_output_flow(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
,
2772 struct sock
*sk
, int flags
)
2776 if ((err
= __ip_route_output_key(net
, rp
, flp
)) != 0)
2781 flp
->fl4_src
= (*rp
)->rt_src
;
2783 flp
->fl4_dst
= (*rp
)->rt_dst
;
2784 err
= __xfrm_lookup(net
, (struct dst_entry
**)rp
, flp
, sk
,
2785 flags
? XFRM_LOOKUP_WAIT
: 0);
2786 if (err
== -EREMOTE
)
2787 err
= ipv4_dst_blackhole(net
, rp
, flp
);
2795 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2797 int ip_route_output_key(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2799 return ip_route_output_flow(net
, rp
, flp
, NULL
, 0);
2802 static int rt_fill_info(struct net
*net
,
2803 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2804 int nowait
, unsigned int flags
)
2806 struct rtable
*rt
= skb_rtable(skb
);
2808 struct nlmsghdr
*nlh
;
2810 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2812 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2816 r
= nlmsg_data(nlh
);
2817 r
->rtm_family
= AF_INET
;
2818 r
->rtm_dst_len
= 32;
2820 r
->rtm_tos
= rt
->fl
.fl4_tos
;
2821 r
->rtm_table
= RT_TABLE_MAIN
;
2822 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2823 r
->rtm_type
= rt
->rt_type
;
2824 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2825 r
->rtm_protocol
= RTPROT_UNSPEC
;
2826 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2827 if (rt
->rt_flags
& RTCF_NOTIFY
)
2828 r
->rtm_flags
|= RTM_F_NOTIFY
;
2830 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2832 if (rt
->fl
.fl4_src
) {
2833 r
->rtm_src_len
= 32;
2834 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->fl
.fl4_src
);
2837 NLA_PUT_U32(skb
, RTA_OIF
, rt
->u
.dst
.dev
->ifindex
);
2838 #ifdef CONFIG_NET_CLS_ROUTE
2839 if (rt
->u
.dst
.tclassid
)
2840 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->u
.dst
.tclassid
);
2843 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
2844 else if (rt
->rt_src
!= rt
->fl
.fl4_src
)
2845 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
2847 if (rt
->rt_dst
!= rt
->rt_gateway
)
2848 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
2850 if (rtnetlink_put_metrics(skb
, rt
->u
.dst
.metrics
) < 0)
2851 goto nla_put_failure
;
2853 error
= rt
->u
.dst
.error
;
2854 expires
= rt
->u
.dst
.expires
? rt
->u
.dst
.expires
- jiffies
: 0;
2856 id
= atomic_read(&rt
->peer
->ip_id_count
) & 0xffff;
2857 if (rt
->peer
->tcp_ts_stamp
) {
2858 ts
= rt
->peer
->tcp_ts
;
2859 tsage
= get_seconds() - rt
->peer
->tcp_ts_stamp
;
2864 #ifdef CONFIG_IP_MROUTE
2865 __be32 dst
= rt
->rt_dst
;
2867 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2868 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2869 int err
= ipmr_get_route(net
, skb
, r
, nowait
);
2874 goto nla_put_failure
;
2876 if (err
== -EMSGSIZE
)
2877 goto nla_put_failure
;
2883 NLA_PUT_U32(skb
, RTA_IIF
, rt
->fl
.iif
);
2886 if (rtnl_put_cacheinfo(skb
, &rt
->u
.dst
, id
, ts
, tsage
,
2887 expires
, error
) < 0)
2888 goto nla_put_failure
;
2890 return nlmsg_end(skb
, nlh
);
2893 nlmsg_cancel(skb
, nlh
);
2897 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
2899 struct net
*net
= sock_net(in_skb
->sk
);
2901 struct nlattr
*tb
[RTA_MAX
+1];
2902 struct rtable
*rt
= NULL
;
2907 struct sk_buff
*skb
;
2909 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2913 rtm
= nlmsg_data(nlh
);
2915 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2921 /* Reserve room for dummy headers, this skb can pass
2922 through good chunk of routing engine.
2924 skb_reset_mac_header(skb
);
2925 skb_reset_network_header(skb
);
2927 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2928 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
2929 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2931 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
2932 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
2933 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2936 struct net_device
*dev
;
2938 dev
= __dev_get_by_index(net
, iif
);
2944 skb
->protocol
= htons(ETH_P_IP
);
2947 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
2950 rt
= skb_rtable(skb
);
2951 if (err
== 0 && rt
->u
.dst
.error
)
2952 err
= -rt
->u
.dst
.error
;
2959 .tos
= rtm
->rtm_tos
,
2962 .oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
2964 err
= ip_route_output_key(net
, &rt
, &fl
);
2970 skb_dst_set(skb
, &rt
->u
.dst
);
2971 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2972 rt
->rt_flags
|= RTCF_NOTIFY
;
2974 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
2975 RTM_NEWROUTE
, 0, 0);
2979 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
2988 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2995 net
= sock_net(skb
->sk
);
3000 s_idx
= idx
= cb
->args
[1];
3001 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3002 if (!rt_hash_table
[h
].chain
)
3005 for (rt
= rcu_dereference(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3006 rt
= rcu_dereference(rt
->u
.dst
.rt_next
), idx
++) {
3007 if (!net_eq(dev_net(rt
->u
.dst
.dev
), net
) || idx
< s_idx
)
3009 if (rt_is_expired(rt
))
3011 skb_dst_set(skb
, dst_clone(&rt
->u
.dst
));
3012 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3013 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3014 1, NLM_F_MULTI
) <= 0) {
3016 rcu_read_unlock_bh();
3021 rcu_read_unlock_bh();
3030 void ip_rt_multicast_event(struct in_device
*in_dev
)
3032 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3035 #ifdef CONFIG_SYSCTL
3036 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3037 void __user
*buffer
,
3038 size_t *lenp
, loff_t
*ppos
)
3045 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3046 ctl
.data
= &flush_delay
;
3047 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3049 net
= (struct net
*)__ctl
->extra1
;
3050 rt_cache_flush(net
, flush_delay
);
3057 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table
*table
,
3058 void __user
*oldval
,
3059 size_t __user
*oldlenp
,
3060 void __user
*newval
,
3065 if (newlen
!= sizeof(int))
3067 if (get_user(delay
, (int __user
*)newval
))
3069 net
= (struct net
*)table
->extra1
;
3070 rt_cache_flush(net
, delay
);
3074 static void rt_secret_reschedule(int old
)
3077 int new = ip_rt_secret_interval
;
3078 int diff
= new - old
;
3085 int deleted
= del_timer_sync(&net
->ipv4
.rt_secret_timer
);
3091 long time
= net
->ipv4
.rt_secret_timer
.expires
- jiffies
;
3093 if (time
<= 0 || (time
+= diff
) <= 0)
3096 net
->ipv4
.rt_secret_timer
.expires
= time
;
3098 net
->ipv4
.rt_secret_timer
.expires
= new;
3100 net
->ipv4
.rt_secret_timer
.expires
+= jiffies
;
3101 add_timer(&net
->ipv4
.rt_secret_timer
);
3106 static int ipv4_sysctl_rt_secret_interval(ctl_table
*ctl
, int write
,
3107 void __user
*buffer
, size_t *lenp
,
3110 int old
= ip_rt_secret_interval
;
3111 int ret
= proc_dointvec_jiffies(ctl
, write
, buffer
, lenp
, ppos
);
3113 rt_secret_reschedule(old
);
3118 static int ipv4_sysctl_rt_secret_interval_strategy(ctl_table
*table
,
3119 void __user
*oldval
,
3120 size_t __user
*oldlenp
,
3121 void __user
*newval
,
3124 int old
= ip_rt_secret_interval
;
3125 int ret
= sysctl_jiffies(table
, oldval
, oldlenp
, newval
, newlen
);
3127 rt_secret_reschedule(old
);
3132 static ctl_table ipv4_route_table
[] = {
3134 .ctl_name
= NET_IPV4_ROUTE_GC_THRESH
,
3135 .procname
= "gc_thresh",
3136 .data
= &ipv4_dst_ops
.gc_thresh
,
3137 .maxlen
= sizeof(int),
3139 .proc_handler
= proc_dointvec
,
3142 .ctl_name
= NET_IPV4_ROUTE_MAX_SIZE
,
3143 .procname
= "max_size",
3144 .data
= &ip_rt_max_size
,
3145 .maxlen
= sizeof(int),
3147 .proc_handler
= proc_dointvec
,
3150 /* Deprecated. Use gc_min_interval_ms */
3152 .ctl_name
= NET_IPV4_ROUTE_GC_MIN_INTERVAL
,
3153 .procname
= "gc_min_interval",
3154 .data
= &ip_rt_gc_min_interval
,
3155 .maxlen
= sizeof(int),
3157 .proc_handler
= proc_dointvec_jiffies
,
3158 .strategy
= sysctl_jiffies
,
3161 .ctl_name
= NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS
,
3162 .procname
= "gc_min_interval_ms",
3163 .data
= &ip_rt_gc_min_interval
,
3164 .maxlen
= sizeof(int),
3166 .proc_handler
= proc_dointvec_ms_jiffies
,
3167 .strategy
= sysctl_ms_jiffies
,
3170 .ctl_name
= NET_IPV4_ROUTE_GC_TIMEOUT
,
3171 .procname
= "gc_timeout",
3172 .data
= &ip_rt_gc_timeout
,
3173 .maxlen
= sizeof(int),
3175 .proc_handler
= proc_dointvec_jiffies
,
3176 .strategy
= sysctl_jiffies
,
3179 .ctl_name
= NET_IPV4_ROUTE_GC_INTERVAL
,
3180 .procname
= "gc_interval",
3181 .data
= &ip_rt_gc_interval
,
3182 .maxlen
= sizeof(int),
3184 .proc_handler
= proc_dointvec_jiffies
,
3185 .strategy
= sysctl_jiffies
,
3188 .ctl_name
= NET_IPV4_ROUTE_REDIRECT_LOAD
,
3189 .procname
= "redirect_load",
3190 .data
= &ip_rt_redirect_load
,
3191 .maxlen
= sizeof(int),
3193 .proc_handler
= proc_dointvec
,
3196 .ctl_name
= NET_IPV4_ROUTE_REDIRECT_NUMBER
,
3197 .procname
= "redirect_number",
3198 .data
= &ip_rt_redirect_number
,
3199 .maxlen
= sizeof(int),
3201 .proc_handler
= proc_dointvec
,
3204 .ctl_name
= NET_IPV4_ROUTE_REDIRECT_SILENCE
,
3205 .procname
= "redirect_silence",
3206 .data
= &ip_rt_redirect_silence
,
3207 .maxlen
= sizeof(int),
3209 .proc_handler
= proc_dointvec
,
3212 .ctl_name
= NET_IPV4_ROUTE_ERROR_COST
,
3213 .procname
= "error_cost",
3214 .data
= &ip_rt_error_cost
,
3215 .maxlen
= sizeof(int),
3217 .proc_handler
= proc_dointvec
,
3220 .ctl_name
= NET_IPV4_ROUTE_ERROR_BURST
,
3221 .procname
= "error_burst",
3222 .data
= &ip_rt_error_burst
,
3223 .maxlen
= sizeof(int),
3225 .proc_handler
= proc_dointvec
,
3228 .ctl_name
= NET_IPV4_ROUTE_GC_ELASTICITY
,
3229 .procname
= "gc_elasticity",
3230 .data
= &ip_rt_gc_elasticity
,
3231 .maxlen
= sizeof(int),
3233 .proc_handler
= proc_dointvec
,
3236 .ctl_name
= NET_IPV4_ROUTE_MTU_EXPIRES
,
3237 .procname
= "mtu_expires",
3238 .data
= &ip_rt_mtu_expires
,
3239 .maxlen
= sizeof(int),
3241 .proc_handler
= proc_dointvec_jiffies
,
3242 .strategy
= sysctl_jiffies
,
3245 .ctl_name
= NET_IPV4_ROUTE_MIN_PMTU
,
3246 .procname
= "min_pmtu",
3247 .data
= &ip_rt_min_pmtu
,
3248 .maxlen
= sizeof(int),
3250 .proc_handler
= proc_dointvec
,
3253 .ctl_name
= NET_IPV4_ROUTE_MIN_ADVMSS
,
3254 .procname
= "min_adv_mss",
3255 .data
= &ip_rt_min_advmss
,
3256 .maxlen
= sizeof(int),
3258 .proc_handler
= proc_dointvec
,
3261 .ctl_name
= NET_IPV4_ROUTE_SECRET_INTERVAL
,
3262 .procname
= "secret_interval",
3263 .data
= &ip_rt_secret_interval
,
3264 .maxlen
= sizeof(int),
3266 .proc_handler
= ipv4_sysctl_rt_secret_interval
,
3267 .strategy
= ipv4_sysctl_rt_secret_interval_strategy
,
3272 static struct ctl_table empty
[1];
3274 static struct ctl_table ipv4_skeleton
[] =
3276 { .procname
= "route", .ctl_name
= NET_IPV4_ROUTE
,
3277 .mode
= 0555, .child
= ipv4_route_table
},
3278 { .procname
= "neigh", .ctl_name
= NET_IPV4_NEIGH
,
3279 .mode
= 0555, .child
= empty
},
3283 static __net_initdata
struct ctl_path ipv4_path
[] = {
3284 { .procname
= "net", .ctl_name
= CTL_NET
, },
3285 { .procname
= "ipv4", .ctl_name
= NET_IPV4
, },
3289 static struct ctl_table ipv4_route_flush_table
[] = {
3291 .ctl_name
= NET_IPV4_ROUTE_FLUSH
,
3292 .procname
= "flush",
3293 .maxlen
= sizeof(int),
3295 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3296 .strategy
= ipv4_sysctl_rtcache_flush_strategy
,
3301 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3302 { .procname
= "net", .ctl_name
= CTL_NET
, },
3303 { .procname
= "ipv4", .ctl_name
= NET_IPV4
, },
3304 { .procname
= "route", .ctl_name
= NET_IPV4_ROUTE
, },
3308 static __net_init
int sysctl_route_net_init(struct net
*net
)
3310 struct ctl_table
*tbl
;
3312 tbl
= ipv4_route_flush_table
;
3313 if (net
!= &init_net
) {
3314 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3318 tbl
[0].extra1
= net
;
3320 net
->ipv4
.route_hdr
=
3321 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3322 if (net
->ipv4
.route_hdr
== NULL
)
3327 if (tbl
!= ipv4_route_flush_table
)
3333 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3335 struct ctl_table
*tbl
;
3337 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3338 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3339 BUG_ON(tbl
== ipv4_route_flush_table
);
3343 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3344 .init
= sysctl_route_net_init
,
3345 .exit
= sysctl_route_net_exit
,
3350 static __net_init
int rt_secret_timer_init(struct net
*net
)
3352 atomic_set(&net
->ipv4
.rt_genid
,
3353 (int) ((num_physpages
^ (num_physpages
>>8)) ^
3354 (jiffies
^ (jiffies
>> 7))));
3356 net
->ipv4
.rt_secret_timer
.function
= rt_secret_rebuild
;
3357 net
->ipv4
.rt_secret_timer
.data
= (unsigned long)net
;
3358 init_timer_deferrable(&net
->ipv4
.rt_secret_timer
);
3360 if (ip_rt_secret_interval
) {
3361 net
->ipv4
.rt_secret_timer
.expires
=
3362 jiffies
+ net_random() % ip_rt_secret_interval
+
3363 ip_rt_secret_interval
;
3364 add_timer(&net
->ipv4
.rt_secret_timer
);
3369 static __net_exit
void rt_secret_timer_exit(struct net
*net
)
3371 del_timer_sync(&net
->ipv4
.rt_secret_timer
);
3374 static __net_initdata
struct pernet_operations rt_secret_timer_ops
= {
3375 .init
= rt_secret_timer_init
,
3376 .exit
= rt_secret_timer_exit
,
3380 #ifdef CONFIG_NET_CLS_ROUTE
3381 struct ip_rt_acct
*ip_rt_acct __read_mostly
;
3382 #endif /* CONFIG_NET_CLS_ROUTE */
3384 static __initdata
unsigned long rhash_entries
;
3385 static int __init
set_rhash_entries(char *str
)
3389 rhash_entries
= simple_strtoul(str
, &str
, 0);
3392 __setup("rhash_entries=", set_rhash_entries
);
3394 int __init
ip_rt_init(void)
3398 #ifdef CONFIG_NET_CLS_ROUTE
3399 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3401 panic("IP: failed to allocate ip_rt_acct\n");
3404 ipv4_dst_ops
.kmem_cachep
=
3405 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3406 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3408 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3410 rt_hash_table
= (struct rt_hash_bucket
*)
3411 alloc_large_system_hash("IP route cache",
3412 sizeof(struct rt_hash_bucket
),
3414 (totalram_pages
>= 128 * 1024) ?
3419 rhash_entries
? 0 : 512 * 1024);
3420 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3421 rt_hash_lock_init();
3423 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3424 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3429 /* All the timers, started at system startup tend
3430 to synchronize. Perturb it a bit.
3432 INIT_DELAYED_WORK_DEFERRABLE(&expires_work
, rt_worker_func
);
3433 expires_ljiffies
= jiffies
;
3434 schedule_delayed_work(&expires_work
,
3435 net_random() % ip_rt_gc_interval
+ ip_rt_gc_interval
);
3437 if (register_pernet_subsys(&rt_secret_timer_ops
))
3438 printk(KERN_ERR
"Unable to setup rt_secret_timer\n");
3440 if (ip_rt_proc_init())
3441 printk(KERN_ERR
"Unable to create route proc files\n");
3444 xfrm4_init(ip_rt_max_size
);
3446 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
);
3448 #ifdef CONFIG_SYSCTL
3449 register_pernet_subsys(&sysctl_route_ops
);
3454 #ifdef CONFIG_SYSCTL
3456 * We really need to sanitize the damn ipv4 init order, then all
3457 * this nonsense will go away.
3459 void __init
ip_static_sysctl_init(void)
3461 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);
3465 EXPORT_SYMBOL(__ip_select_ident
);
3466 EXPORT_SYMBOL(ip_route_input
);
3467 EXPORT_SYMBOL(ip_route_output_key
);