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 void rt_worker_func(struct work_struct
*work
);
135 static DECLARE_DELAYED_WORK(expires_work
, rt_worker_func
);
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
= __constant_htons(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
, **rthp
;
788 unsigned long samples
= 0;
789 unsigned long sum
= 0, sum2
= 0;
792 mult
= ((u64
)ip_rt_gc_interval
) << rt_hash_log
;
793 if (ip_rt_gc_timeout
> 1)
794 do_div(mult
, ip_rt_gc_timeout
);
795 goal
= (unsigned int)mult
;
796 if (goal
> rt_hash_mask
)
797 goal
= rt_hash_mask
+ 1;
798 for (; goal
> 0; goal
--) {
799 unsigned long tmo
= ip_rt_gc_timeout
;
800 unsigned long length
;
802 i
= (i
+ 1) & rt_hash_mask
;
803 rthp
= &rt_hash_table
[i
].chain
;
813 spin_lock_bh(rt_hash_lock_addr(i
));
814 while ((rth
= *rthp
) != NULL
) {
815 if (rt_is_expired(rth
)) {
816 *rthp
= rth
->u
.dst
.rt_next
;
820 if (rth
->u
.dst
.expires
) {
821 /* Entry is expired even if it is in use */
822 if (time_before_eq(jiffies
, rth
->u
.dst
.expires
)) {
824 rthp
= &rth
->u
.dst
.rt_next
;
826 * Only bump our length if the hash
827 * inputs on entries n and n+1 are not
828 * the same, we only count entries on
829 * a chain with equal hash inputs once
830 * so that entries for different QOS
831 * levels, and other non-hash input
832 * attributes don't unfairly skew
833 * the length computation
835 if ((*rthp
== NULL
) ||
836 !compare_hash_inputs(&(*rthp
)->fl
,
841 } else if (!rt_may_expire(rth
, tmo
, ip_rt_gc_timeout
)) {
843 rthp
= &rth
->u
.dst
.rt_next
;
844 if ((*rthp
== NULL
) ||
845 !compare_hash_inputs(&(*rthp
)->fl
,
851 /* Cleanup aged off entries. */
852 *rthp
= rth
->u
.dst
.rt_next
;
855 spin_unlock_bh(rt_hash_lock_addr(i
));
857 sum2
+= length
*length
;
860 unsigned long avg
= sum
/ samples
;
861 unsigned long sd
= int_sqrt(sum2
/ samples
- avg
*avg
);
862 rt_chain_length_max
= max_t(unsigned long,
864 (avg
+ 4*sd
) >> FRACT_BITS
);
870 * rt_worker_func() is run in process context.
871 * we call rt_check_expire() to scan part of the hash table
873 static void rt_worker_func(struct work_struct
*work
)
876 schedule_delayed_work(&expires_work
, ip_rt_gc_interval
);
880 * Pertubation of rt_genid by a small quantity [1..256]
881 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
882 * many times (2^24) without giving recent rt_genid.
883 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
885 static void rt_cache_invalidate(struct net
*net
)
887 unsigned char shuffle
;
889 get_random_bytes(&shuffle
, sizeof(shuffle
));
890 atomic_add(shuffle
+ 1U, &net
->ipv4
.rt_genid
);
894 * delay < 0 : invalidate cache (fast : entries will be deleted later)
895 * delay >= 0 : invalidate & flush cache (can be long)
897 void rt_cache_flush(struct net
*net
, int delay
)
899 rt_cache_invalidate(net
);
901 rt_do_flush(!in_softirq());
905 * We change rt_genid and let gc do the cleanup
907 static void rt_secret_rebuild(unsigned long __net
)
909 struct net
*net
= (struct net
*)__net
;
910 rt_cache_invalidate(net
);
911 mod_timer(&net
->ipv4
.rt_secret_timer
, jiffies
+ ip_rt_secret_interval
);
914 static void rt_secret_rebuild_oneshot(struct net
*net
)
916 del_timer_sync(&net
->ipv4
.rt_secret_timer
);
917 rt_cache_invalidate(net
);
918 if (ip_rt_secret_interval
) {
919 net
->ipv4
.rt_secret_timer
.expires
+= ip_rt_secret_interval
;
920 add_timer(&net
->ipv4
.rt_secret_timer
);
924 static void rt_emergency_hash_rebuild(struct net
*net
)
926 if (net_ratelimit()) {
927 printk(KERN_WARNING
"Route hash chain too long!\n");
928 printk(KERN_WARNING
"Adjust your secret_interval!\n");
931 rt_secret_rebuild_oneshot(net
);
935 Short description of GC goals.
937 We want to build algorithm, which will keep routing cache
938 at some equilibrium point, when number of aged off entries
939 is kept approximately equal to newly generated ones.
941 Current expiration strength is variable "expire".
942 We try to adjust it dynamically, so that if networking
943 is idle expires is large enough to keep enough of warm entries,
944 and when load increases it reduces to limit cache size.
947 static int rt_garbage_collect(struct dst_ops
*ops
)
949 static unsigned long expire
= RT_GC_TIMEOUT
;
950 static unsigned long last_gc
;
952 static int equilibrium
;
953 struct rtable
*rth
, **rthp
;
954 unsigned long now
= jiffies
;
958 * Garbage collection is pretty expensive,
959 * do not make it too frequently.
962 RT_CACHE_STAT_INC(gc_total
);
964 if (now
- last_gc
< ip_rt_gc_min_interval
&&
965 atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
) {
966 RT_CACHE_STAT_INC(gc_ignored
);
970 /* Calculate number of entries, which we want to expire now. */
971 goal
= atomic_read(&ipv4_dst_ops
.entries
) -
972 (ip_rt_gc_elasticity
<< rt_hash_log
);
974 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
975 equilibrium
= ipv4_dst_ops
.gc_thresh
;
976 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
978 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
979 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
982 /* We are in dangerous area. Try to reduce cache really
985 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
986 equilibrium
= atomic_read(&ipv4_dst_ops
.entries
) - goal
;
989 if (now
- last_gc
>= ip_rt_gc_min_interval
)
1000 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
1001 unsigned long tmo
= expire
;
1003 k
= (k
+ 1) & rt_hash_mask
;
1004 rthp
= &rt_hash_table
[k
].chain
;
1005 spin_lock_bh(rt_hash_lock_addr(k
));
1006 while ((rth
= *rthp
) != NULL
) {
1007 if (!rt_is_expired(rth
) &&
1008 !rt_may_expire(rth
, tmo
, expire
)) {
1010 rthp
= &rth
->u
.dst
.rt_next
;
1013 *rthp
= rth
->u
.dst
.rt_next
;
1017 spin_unlock_bh(rt_hash_lock_addr(k
));
1026 /* Goal is not achieved. We stop process if:
1028 - if expire reduced to zero. Otherwise, expire is halfed.
1029 - if table is not full.
1030 - if we are called from interrupt.
1031 - jiffies check is just fallback/debug loop breaker.
1032 We will not spin here for long time in any case.
1035 RT_CACHE_STAT_INC(gc_goal_miss
);
1041 #if RT_CACHE_DEBUG >= 2
1042 printk(KERN_DEBUG
"expire>> %u %d %d %d\n", expire
,
1043 atomic_read(&ipv4_dst_ops
.entries
), goal
, i
);
1046 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1048 } while (!in_softirq() && time_before_eq(jiffies
, now
));
1050 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1052 if (net_ratelimit())
1053 printk(KERN_WARNING
"dst cache overflow\n");
1054 RT_CACHE_STAT_INC(gc_dst_overflow
);
1058 expire
+= ip_rt_gc_min_interval
;
1059 if (expire
> ip_rt_gc_timeout
||
1060 atomic_read(&ipv4_dst_ops
.entries
) < ipv4_dst_ops
.gc_thresh
)
1061 expire
= ip_rt_gc_timeout
;
1062 #if RT_CACHE_DEBUG >= 2
1063 printk(KERN_DEBUG
"expire++ %u %d %d %d\n", expire
,
1064 atomic_read(&ipv4_dst_ops
.entries
), goal
, rover
);
1069 static int rt_intern_hash(unsigned hash
, struct rtable
*rt
, struct rtable
**rp
)
1071 struct rtable
*rth
, **rthp
;
1072 struct rtable
*rthi
;
1074 struct rtable
*cand
, **candp
;
1077 int attempts
= !in_softirq();
1081 min_score
= ~(u32
)0;
1086 if (!rt_caching(dev_net(rt
->u
.dst
.dev
))) {
1091 rthp
= &rt_hash_table
[hash
].chain
;
1094 spin_lock_bh(rt_hash_lock_addr(hash
));
1095 while ((rth
= *rthp
) != NULL
) {
1096 if (rt_is_expired(rth
)) {
1097 *rthp
= rth
->u
.dst
.rt_next
;
1101 if (compare_keys(&rth
->fl
, &rt
->fl
) && compare_netns(rth
, rt
)) {
1103 *rthp
= rth
->u
.dst
.rt_next
;
1105 * Since lookup is lockfree, the deletion
1106 * must be visible to another weakly ordered CPU before
1107 * the insertion at the start of the hash chain.
1109 rcu_assign_pointer(rth
->u
.dst
.rt_next
,
1110 rt_hash_table
[hash
].chain
);
1112 * Since lookup is lockfree, the update writes
1113 * must be ordered for consistency on SMP.
1115 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1117 dst_use(&rth
->u
.dst
, now
);
1118 spin_unlock_bh(rt_hash_lock_addr(hash
));
1125 if (!atomic_read(&rth
->u
.dst
.__refcnt
)) {
1126 u32 score
= rt_score(rth
);
1128 if (score
<= min_score
) {
1137 rthp
= &rth
->u
.dst
.rt_next
;
1140 * check to see if the next entry in the chain
1141 * contains the same hash input values as rt. If it does
1142 * This is where we will insert into the list, instead of
1143 * at the head. This groups entries that differ by aspects not
1144 * relvant to the hash function together, which we use to adjust
1147 if (*rthp
&& compare_hash_inputs(&(*rthp
)->fl
, &rt
->fl
))
1152 /* ip_rt_gc_elasticity used to be average length of chain
1153 * length, when exceeded gc becomes really aggressive.
1155 * The second limit is less certain. At the moment it allows
1156 * only 2 entries per bucket. We will see.
1158 if (chain_length
> ip_rt_gc_elasticity
) {
1159 *candp
= cand
->u
.dst
.rt_next
;
1163 if (chain_length
> rt_chain_length_max
) {
1164 struct net
*net
= dev_net(rt
->u
.dst
.dev
);
1165 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1166 if (!rt_caching(dev_net(rt
->u
.dst
.dev
))) {
1167 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1168 rt
->u
.dst
.dev
->name
, num
);
1170 rt_emergency_hash_rebuild(dev_net(rt
->u
.dst
.dev
));
1174 /* Try to bind route to arp only if it is output
1175 route or unicast forwarding path.
1177 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1178 int err
= arp_bind_neighbour(&rt
->u
.dst
);
1180 spin_unlock_bh(rt_hash_lock_addr(hash
));
1182 if (err
!= -ENOBUFS
) {
1187 /* Neighbour tables are full and nothing
1188 can be released. Try to shrink route cache,
1189 it is most likely it holds some neighbour records.
1191 if (attempts
-- > 0) {
1192 int saved_elasticity
= ip_rt_gc_elasticity
;
1193 int saved_int
= ip_rt_gc_min_interval
;
1194 ip_rt_gc_elasticity
= 1;
1195 ip_rt_gc_min_interval
= 0;
1196 rt_garbage_collect(&ipv4_dst_ops
);
1197 ip_rt_gc_min_interval
= saved_int
;
1198 ip_rt_gc_elasticity
= saved_elasticity
;
1202 if (net_ratelimit())
1203 printk(KERN_WARNING
"Neighbour table overflow.\n");
1210 rt
->u
.dst
.rt_next
= rthi
->u
.dst
.rt_next
;
1212 rt
->u
.dst
.rt_next
= rt_hash_table
[hash
].chain
;
1214 #if RT_CACHE_DEBUG >= 2
1215 if (rt
->u
.dst
.rt_next
) {
1217 printk(KERN_DEBUG
"rt_cache @%02x: %pI4", hash
, &rt
->rt_dst
);
1218 for (trt
= rt
->u
.dst
.rt_next
; trt
; trt
= trt
->u
.dst
.rt_next
)
1219 printk(" . %pI4", &trt
->rt_dst
);
1224 * Since lookup is lockfree, we must make sure
1225 * previous writes to rt are comitted to memory
1226 * before making rt visible to other CPUS.
1229 rcu_assign_pointer(rthi
->u
.dst
.rt_next
, rt
);
1231 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1233 spin_unlock_bh(rt_hash_lock_addr(hash
));
1238 void rt_bind_peer(struct rtable
*rt
, int create
)
1240 static DEFINE_SPINLOCK(rt_peer_lock
);
1241 struct inet_peer
*peer
;
1243 peer
= inet_getpeer(rt
->rt_dst
, create
);
1245 spin_lock_bh(&rt_peer_lock
);
1246 if (rt
->peer
== NULL
) {
1250 spin_unlock_bh(&rt_peer_lock
);
1256 * Peer allocation may fail only in serious out-of-memory conditions. However
1257 * we still can generate some output.
1258 * Random ID selection looks a bit dangerous because we have no chances to
1259 * select ID being unique in a reasonable period of time.
1260 * But broken packet identifier may be better than no packet at all.
1262 static void ip_select_fb_ident(struct iphdr
*iph
)
1264 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1265 static u32 ip_fallback_id
;
1268 spin_lock_bh(&ip_fb_id_lock
);
1269 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1270 iph
->id
= htons(salt
& 0xFFFF);
1271 ip_fallback_id
= salt
;
1272 spin_unlock_bh(&ip_fb_id_lock
);
1275 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1277 struct rtable
*rt
= (struct rtable
*) dst
;
1280 if (rt
->peer
== NULL
)
1281 rt_bind_peer(rt
, 1);
1283 /* If peer is attached to destination, it is never detached,
1284 so that we need not to grab a lock to dereference it.
1287 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1291 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1292 __builtin_return_address(0));
1294 ip_select_fb_ident(iph
);
1297 static void rt_del(unsigned hash
, struct rtable
*rt
)
1299 struct rtable
**rthp
, *aux
;
1301 rthp
= &rt_hash_table
[hash
].chain
;
1302 spin_lock_bh(rt_hash_lock_addr(hash
));
1304 while ((aux
= *rthp
) != NULL
) {
1305 if (aux
== rt
|| rt_is_expired(aux
)) {
1306 *rthp
= aux
->u
.dst
.rt_next
;
1310 rthp
= &aux
->u
.dst
.rt_next
;
1312 spin_unlock_bh(rt_hash_lock_addr(hash
));
1315 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1316 __be32 saddr
, struct net_device
*dev
)
1319 struct in_device
*in_dev
= in_dev_get(dev
);
1320 struct rtable
*rth
, **rthp
;
1321 __be32 skeys
[2] = { saddr
, 0 };
1322 int ikeys
[2] = { dev
->ifindex
, 0 };
1323 struct netevent_redirect netevent
;
1330 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
)
1331 || ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
)
1332 || ipv4_is_zeronet(new_gw
))
1333 goto reject_redirect
;
1335 if (!rt_caching(net
))
1336 goto reject_redirect
;
1338 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1339 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1340 goto reject_redirect
;
1341 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1342 goto reject_redirect
;
1344 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1345 goto reject_redirect
;
1348 for (i
= 0; i
< 2; i
++) {
1349 for (k
= 0; k
< 2; k
++) {
1350 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1353 rthp
=&rt_hash_table
[hash
].chain
;
1356 while ((rth
= rcu_dereference(*rthp
)) != NULL
) {
1359 if (rth
->fl
.fl4_dst
!= daddr
||
1360 rth
->fl
.fl4_src
!= skeys
[i
] ||
1361 rth
->fl
.oif
!= ikeys
[k
] ||
1363 rt_is_expired(rth
) ||
1364 !net_eq(dev_net(rth
->u
.dst
.dev
), net
)) {
1365 rthp
= &rth
->u
.dst
.rt_next
;
1369 if (rth
->rt_dst
!= daddr
||
1370 rth
->rt_src
!= saddr
||
1372 rth
->rt_gateway
!= old_gw
||
1373 rth
->u
.dst
.dev
!= dev
)
1376 dst_hold(&rth
->u
.dst
);
1379 rt
= dst_alloc(&ipv4_dst_ops
);
1386 /* Copy all the information. */
1388 rt
->u
.dst
.__use
= 1;
1389 atomic_set(&rt
->u
.dst
.__refcnt
, 1);
1390 rt
->u
.dst
.child
= NULL
;
1392 dev_hold(rt
->u
.dst
.dev
);
1394 in_dev_hold(rt
->idev
);
1395 rt
->u
.dst
.obsolete
= 0;
1396 rt
->u
.dst
.lastuse
= jiffies
;
1397 rt
->u
.dst
.path
= &rt
->u
.dst
;
1398 rt
->u
.dst
.neighbour
= NULL
;
1399 rt
->u
.dst
.hh
= NULL
;
1401 rt
->u
.dst
.xfrm
= NULL
;
1403 rt
->rt_genid
= rt_genid(net
);
1404 rt
->rt_flags
|= RTCF_REDIRECTED
;
1406 /* Gateway is different ... */
1407 rt
->rt_gateway
= new_gw
;
1409 /* Redirect received -> path was valid */
1410 dst_confirm(&rth
->u
.dst
);
1413 atomic_inc(&rt
->peer
->refcnt
);
1415 if (arp_bind_neighbour(&rt
->u
.dst
) ||
1416 !(rt
->u
.dst
.neighbour
->nud_state
&
1418 if (rt
->u
.dst
.neighbour
)
1419 neigh_event_send(rt
->u
.dst
.neighbour
, NULL
);
1425 netevent
.old
= &rth
->u
.dst
;
1426 netevent
.new = &rt
->u
.dst
;
1427 call_netevent_notifiers(NETEVENT_REDIRECT
,
1431 if (!rt_intern_hash(hash
, rt
, &rt
))
1444 #ifdef CONFIG_IP_ROUTE_VERBOSE
1445 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1446 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1447 " Advised path = %pI4 -> %pI4\n",
1448 &old_gw
, dev
->name
, &new_gw
,
1454 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1456 struct rtable
*rt
= (struct rtable
*)dst
;
1457 struct dst_entry
*ret
= dst
;
1460 if (dst
->obsolete
) {
1463 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
1464 rt
->u
.dst
.expires
) {
1465 unsigned hash
= rt_hash(rt
->fl
.fl4_dst
, rt
->fl
.fl4_src
,
1467 rt_genid(dev_net(dst
->dev
)));
1468 #if RT_CACHE_DEBUG >= 1
1469 printk(KERN_DEBUG
"ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1470 &rt
->rt_dst
, rt
->fl
.fl4_tos
);
1481 * 1. The first ip_rt_redirect_number redirects are sent
1482 * with exponential backoff, then we stop sending them at all,
1483 * assuming that the host ignores our redirects.
1484 * 2. If we did not see packets requiring redirects
1485 * during ip_rt_redirect_silence, we assume that the host
1486 * forgot redirected route and start to send redirects again.
1488 * This algorithm is much cheaper and more intelligent than dumb load limiting
1491 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1492 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1495 void ip_rt_send_redirect(struct sk_buff
*skb
)
1497 struct rtable
*rt
= skb
->rtable
;
1498 struct in_device
*in_dev
= in_dev_get(rt
->u
.dst
.dev
);
1503 if (!IN_DEV_TX_REDIRECTS(in_dev
))
1506 /* No redirected packets during ip_rt_redirect_silence;
1507 * reset the algorithm.
1509 if (time_after(jiffies
, rt
->u
.dst
.rate_last
+ ip_rt_redirect_silence
))
1510 rt
->u
.dst
.rate_tokens
= 0;
1512 /* Too many ignored redirects; do not send anything
1513 * set u.dst.rate_last to the last seen redirected packet.
1515 if (rt
->u
.dst
.rate_tokens
>= ip_rt_redirect_number
) {
1516 rt
->u
.dst
.rate_last
= jiffies
;
1520 /* Check for load limit; set rate_last to the latest sent
1523 if (rt
->u
.dst
.rate_tokens
== 0 ||
1525 (rt
->u
.dst
.rate_last
+
1526 (ip_rt_redirect_load
<< rt
->u
.dst
.rate_tokens
)))) {
1527 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1528 rt
->u
.dst
.rate_last
= jiffies
;
1529 ++rt
->u
.dst
.rate_tokens
;
1530 #ifdef CONFIG_IP_ROUTE_VERBOSE
1531 if (IN_DEV_LOG_MARTIANS(in_dev
) &&
1532 rt
->u
.dst
.rate_tokens
== ip_rt_redirect_number
&&
1534 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1535 &rt
->rt_src
, rt
->rt_iif
,
1536 &rt
->rt_dst
, &rt
->rt_gateway
);
1543 static int ip_error(struct sk_buff
*skb
)
1545 struct rtable
*rt
= skb
->rtable
;
1549 switch (rt
->u
.dst
.error
) {
1554 code
= ICMP_HOST_UNREACH
;
1557 code
= ICMP_NET_UNREACH
;
1558 IP_INC_STATS_BH(dev_net(rt
->u
.dst
.dev
),
1559 IPSTATS_MIB_INNOROUTES
);
1562 code
= ICMP_PKT_FILTERED
;
1567 rt
->u
.dst
.rate_tokens
+= now
- rt
->u
.dst
.rate_last
;
1568 if (rt
->u
.dst
.rate_tokens
> ip_rt_error_burst
)
1569 rt
->u
.dst
.rate_tokens
= ip_rt_error_burst
;
1570 rt
->u
.dst
.rate_last
= now
;
1571 if (rt
->u
.dst
.rate_tokens
>= ip_rt_error_cost
) {
1572 rt
->u
.dst
.rate_tokens
-= ip_rt_error_cost
;
1573 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1576 out
: kfree_skb(skb
);
1581 * The last two values are not from the RFC but
1582 * are needed for AMPRnet AX.25 paths.
1585 static const unsigned short mtu_plateau
[] =
1586 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1588 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1592 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1593 if (old_mtu
> mtu_plateau
[i
])
1594 return mtu_plateau
[i
];
1598 unsigned short ip_rt_frag_needed(struct net
*net
, struct iphdr
*iph
,
1599 unsigned short new_mtu
,
1600 struct net_device
*dev
)
1603 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1605 int ikeys
[2] = { dev
->ifindex
, 0 };
1606 __be32 skeys
[2] = { iph
->saddr
, 0, };
1607 __be32 daddr
= iph
->daddr
;
1608 unsigned short est_mtu
= 0;
1610 if (ipv4_config
.no_pmtu_disc
)
1613 for (k
= 0; k
< 2; k
++) {
1614 for (i
= 0; i
< 2; i
++) {
1615 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1619 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
1620 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
1621 unsigned short mtu
= new_mtu
;
1623 if (rth
->fl
.fl4_dst
!= daddr
||
1624 rth
->fl
.fl4_src
!= skeys
[i
] ||
1625 rth
->rt_dst
!= daddr
||
1626 rth
->rt_src
!= iph
->saddr
||
1627 rth
->fl
.oif
!= ikeys
[k
] ||
1629 dst_metric_locked(&rth
->u
.dst
, RTAX_MTU
) ||
1630 !net_eq(dev_net(rth
->u
.dst
.dev
), net
) ||
1634 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1636 /* BSD 4.2 compatibility hack :-( */
1638 old_mtu
>= dst_mtu(&rth
->u
.dst
) &&
1639 old_mtu
>= 68 + (iph
->ihl
<< 2))
1640 old_mtu
-= iph
->ihl
<< 2;
1642 mtu
= guess_mtu(old_mtu
);
1644 if (mtu
<= dst_mtu(&rth
->u
.dst
)) {
1645 if (mtu
< dst_mtu(&rth
->u
.dst
)) {
1646 dst_confirm(&rth
->u
.dst
);
1647 if (mtu
< ip_rt_min_pmtu
) {
1648 mtu
= ip_rt_min_pmtu
;
1649 rth
->u
.dst
.metrics
[RTAX_LOCK
-1] |=
1652 rth
->u
.dst
.metrics
[RTAX_MTU
-1] = mtu
;
1653 dst_set_expires(&rth
->u
.dst
,
1662 return est_mtu
? : new_mtu
;
1665 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1667 if (dst_mtu(dst
) > mtu
&& mtu
>= 68 &&
1668 !(dst_metric_locked(dst
, RTAX_MTU
))) {
1669 if (mtu
< ip_rt_min_pmtu
) {
1670 mtu
= ip_rt_min_pmtu
;
1671 dst
->metrics
[RTAX_LOCK
-1] |= (1 << RTAX_MTU
);
1673 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1674 dst_set_expires(dst
, ip_rt_mtu_expires
);
1675 call_netevent_notifiers(NETEVENT_PMTU_UPDATE
, dst
);
1679 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1684 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1686 struct rtable
*rt
= (struct rtable
*) dst
;
1687 struct inet_peer
*peer
= rt
->peer
;
1688 struct in_device
*idev
= rt
->idev
;
1701 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
1704 struct rtable
*rt
= (struct rtable
*) dst
;
1705 struct in_device
*idev
= rt
->idev
;
1706 if (dev
!= dev_net(dev
)->loopback_dev
&& idev
&& idev
->dev
== dev
) {
1707 struct in_device
*loopback_idev
=
1708 in_dev_get(dev_net(dev
)->loopback_dev
);
1709 if (loopback_idev
) {
1710 rt
->idev
= loopback_idev
;
1716 static void ipv4_link_failure(struct sk_buff
*skb
)
1720 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1724 dst_set_expires(&rt
->u
.dst
, 0);
1727 static int ip_rt_bug(struct sk_buff
*skb
)
1729 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1730 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1731 skb
->dev
? skb
->dev
->name
: "?");
1737 We do not cache source address of outgoing interface,
1738 because it is used only by IP RR, TS and SRR options,
1739 so that it out of fast path.
1741 BTW remember: "addr" is allowed to be not aligned
1745 void ip_rt_get_source(u8
*addr
, struct rtable
*rt
)
1748 struct fib_result res
;
1750 if (rt
->fl
.iif
== 0)
1752 else if (fib_lookup(dev_net(rt
->u
.dst
.dev
), &rt
->fl
, &res
) == 0) {
1753 src
= FIB_RES_PREFSRC(res
);
1756 src
= inet_select_addr(rt
->u
.dst
.dev
, rt
->rt_gateway
,
1758 memcpy(addr
, &src
, 4);
1761 #ifdef CONFIG_NET_CLS_ROUTE
1762 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1764 if (!(rt
->u
.dst
.tclassid
& 0xFFFF))
1765 rt
->u
.dst
.tclassid
|= tag
& 0xFFFF;
1766 if (!(rt
->u
.dst
.tclassid
& 0xFFFF0000))
1767 rt
->u
.dst
.tclassid
|= tag
& 0xFFFF0000;
1771 static void rt_set_nexthop(struct rtable
*rt
, struct fib_result
*res
, u32 itag
)
1773 struct fib_info
*fi
= res
->fi
;
1776 if (FIB_RES_GW(*res
) &&
1777 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1778 rt
->rt_gateway
= FIB_RES_GW(*res
);
1779 memcpy(rt
->u
.dst
.metrics
, fi
->fib_metrics
,
1780 sizeof(rt
->u
.dst
.metrics
));
1781 if (fi
->fib_mtu
== 0) {
1782 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = rt
->u
.dst
.dev
->mtu
;
1783 if (dst_metric_locked(&rt
->u
.dst
, RTAX_MTU
) &&
1784 rt
->rt_gateway
!= rt
->rt_dst
&&
1785 rt
->u
.dst
.dev
->mtu
> 576)
1786 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = 576;
1788 #ifdef CONFIG_NET_CLS_ROUTE
1789 rt
->u
.dst
.tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1792 rt
->u
.dst
.metrics
[RTAX_MTU
-1]= rt
->u
.dst
.dev
->mtu
;
1794 if (dst_metric(&rt
->u
.dst
, RTAX_HOPLIMIT
) == 0)
1795 rt
->u
.dst
.metrics
[RTAX_HOPLIMIT
-1] = sysctl_ip_default_ttl
;
1796 if (dst_mtu(&rt
->u
.dst
) > IP_MAX_MTU
)
1797 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = IP_MAX_MTU
;
1798 if (dst_metric(&rt
->u
.dst
, RTAX_ADVMSS
) == 0)
1799 rt
->u
.dst
.metrics
[RTAX_ADVMSS
-1] = max_t(unsigned int, rt
->u
.dst
.dev
->mtu
- 40,
1801 if (dst_metric(&rt
->u
.dst
, RTAX_ADVMSS
) > 65535 - 40)
1802 rt
->u
.dst
.metrics
[RTAX_ADVMSS
-1] = 65535 - 40;
1804 #ifdef CONFIG_NET_CLS_ROUTE
1805 #ifdef CONFIG_IP_MULTIPLE_TABLES
1806 set_class_tag(rt
, fib_rules_tclass(res
));
1808 set_class_tag(rt
, itag
);
1810 rt
->rt_type
= res
->type
;
1813 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1814 u8 tos
, struct net_device
*dev
, int our
)
1819 struct in_device
*in_dev
= in_dev_get(dev
);
1822 /* Primary sanity checks. */
1827 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1828 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
1831 if (ipv4_is_zeronet(saddr
)) {
1832 if (!ipv4_is_local_multicast(daddr
))
1834 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
1835 } else if (fib_validate_source(saddr
, 0, tos
, 0,
1836 dev
, &spec_dst
, &itag
) < 0)
1839 rth
= dst_alloc(&ipv4_dst_ops
);
1843 rth
->u
.dst
.output
= ip_rt_bug
;
1845 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
1846 rth
->u
.dst
.flags
= DST_HOST
;
1847 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
1848 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
1849 rth
->fl
.fl4_dst
= daddr
;
1850 rth
->rt_dst
= daddr
;
1851 rth
->fl
.fl4_tos
= tos
;
1852 rth
->fl
.mark
= skb
->mark
;
1853 rth
->fl
.fl4_src
= saddr
;
1854 rth
->rt_src
= saddr
;
1855 #ifdef CONFIG_NET_CLS_ROUTE
1856 rth
->u
.dst
.tclassid
= itag
;
1859 rth
->fl
.iif
= dev
->ifindex
;
1860 rth
->u
.dst
.dev
= init_net
.loopback_dev
;
1861 dev_hold(rth
->u
.dst
.dev
);
1862 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
1864 rth
->rt_gateway
= daddr
;
1865 rth
->rt_spec_dst
= spec_dst
;
1866 rth
->rt_genid
= rt_genid(dev_net(dev
));
1867 rth
->rt_flags
= RTCF_MULTICAST
;
1868 rth
->rt_type
= RTN_MULTICAST
;
1870 rth
->u
.dst
.input
= ip_local_deliver
;
1871 rth
->rt_flags
|= RTCF_LOCAL
;
1874 #ifdef CONFIG_IP_MROUTE
1875 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1876 rth
->u
.dst
.input
= ip_mr_input
;
1878 RT_CACHE_STAT_INC(in_slow_mc
);
1881 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
1882 return rt_intern_hash(hash
, rth
, &skb
->rtable
);
1894 static void ip_handle_martian_source(struct net_device
*dev
,
1895 struct in_device
*in_dev
,
1896 struct sk_buff
*skb
,
1900 RT_CACHE_STAT_INC(in_martian_src
);
1901 #ifdef CONFIG_IP_ROUTE_VERBOSE
1902 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1904 * RFC1812 recommendation, if source is martian,
1905 * the only hint is MAC header.
1907 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
1908 &daddr
, &saddr
, dev
->name
);
1909 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1911 const unsigned char *p
= skb_mac_header(skb
);
1912 printk(KERN_WARNING
"ll header: ");
1913 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
1915 if (i
< (dev
->hard_header_len
- 1))
1924 static int __mkroute_input(struct sk_buff
*skb
,
1925 struct fib_result
*res
,
1926 struct in_device
*in_dev
,
1927 __be32 daddr
, __be32 saddr
, u32 tos
,
1928 struct rtable
**result
)
1933 struct in_device
*out_dev
;
1938 /* get a working reference to the output device */
1939 out_dev
= in_dev_get(FIB_RES_DEV(*res
));
1940 if (out_dev
== NULL
) {
1941 if (net_ratelimit())
1942 printk(KERN_CRIT
"Bug in ip_route_input" \
1943 "_slow(). Please, report\n");
1948 err
= fib_validate_source(saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1949 in_dev
->dev
, &spec_dst
, &itag
);
1951 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1959 flags
|= RTCF_DIRECTSRC
;
1961 if (out_dev
== in_dev
&& err
&&
1962 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1963 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
1964 flags
|= RTCF_DOREDIRECT
;
1966 if (skb
->protocol
!= htons(ETH_P_IP
)) {
1967 /* Not IP (i.e. ARP). Do not create route, if it is
1968 * invalid for proxy arp. DNAT routes are always valid.
1970 if (out_dev
== in_dev
) {
1977 rth
= dst_alloc(&ipv4_dst_ops
);
1983 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
1984 rth
->u
.dst
.flags
= DST_HOST
;
1985 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
1986 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
1987 if (IN_DEV_CONF_GET(out_dev
, NOXFRM
))
1988 rth
->u
.dst
.flags
|= DST_NOXFRM
;
1989 rth
->fl
.fl4_dst
= daddr
;
1990 rth
->rt_dst
= daddr
;
1991 rth
->fl
.fl4_tos
= tos
;
1992 rth
->fl
.mark
= skb
->mark
;
1993 rth
->fl
.fl4_src
= saddr
;
1994 rth
->rt_src
= saddr
;
1995 rth
->rt_gateway
= daddr
;
1997 rth
->fl
.iif
= in_dev
->dev
->ifindex
;
1998 rth
->u
.dst
.dev
= (out_dev
)->dev
;
1999 dev_hold(rth
->u
.dst
.dev
);
2000 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
2002 rth
->rt_spec_dst
= spec_dst
;
2004 rth
->u
.dst
.input
= ip_forward
;
2005 rth
->u
.dst
.output
= ip_output
;
2006 rth
->rt_genid
= rt_genid(dev_net(rth
->u
.dst
.dev
));
2008 rt_set_nexthop(rth
, res
, itag
);
2010 rth
->rt_flags
= flags
;
2015 /* release the working reference to the output device */
2016 in_dev_put(out_dev
);
2020 static int ip_mkroute_input(struct sk_buff
*skb
,
2021 struct fib_result
*res
,
2022 const struct flowi
*fl
,
2023 struct in_device
*in_dev
,
2024 __be32 daddr
, __be32 saddr
, u32 tos
)
2026 struct rtable
* rth
= NULL
;
2030 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2031 if (res
->fi
&& res
->fi
->fib_nhs
> 1 && fl
->oif
== 0)
2032 fib_select_multipath(fl
, res
);
2035 /* create a routing cache entry */
2036 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2040 /* put it into the cache */
2041 hash
= rt_hash(daddr
, saddr
, fl
->iif
,
2042 rt_genid(dev_net(rth
->u
.dst
.dev
)));
2043 return rt_intern_hash(hash
, rth
, &skb
->rtable
);
2047 * NOTE. We drop all the packets that has local source
2048 * addresses, because every properly looped back packet
2049 * must have correct destination already attached by output routine.
2051 * Such approach solves two big problems:
2052 * 1. Not simplex devices are handled properly.
2053 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2056 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2057 u8 tos
, struct net_device
*dev
)
2059 struct fib_result res
;
2060 struct in_device
*in_dev
= in_dev_get(dev
);
2061 struct flowi fl
= { .nl_u
= { .ip4_u
=
2065 .scope
= RT_SCOPE_UNIVERSE
,
2068 .iif
= dev
->ifindex
};
2071 struct rtable
* rth
;
2076 struct net
* net
= dev_net(dev
);
2078 /* IP on this device is disabled. */
2083 /* Check for the most weird martians, which can be not detected
2087 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2088 ipv4_is_loopback(saddr
))
2089 goto martian_source
;
2091 if (daddr
== htonl(0xFFFFFFFF) || (saddr
== 0 && daddr
== 0))
2094 /* Accept zero addresses only to limited broadcast;
2095 * I even do not know to fix it or not. Waiting for complains :-)
2097 if (ipv4_is_zeronet(saddr
))
2098 goto martian_source
;
2100 if (ipv4_is_lbcast(daddr
) || ipv4_is_zeronet(daddr
) ||
2101 ipv4_is_loopback(daddr
))
2102 goto martian_destination
;
2105 * Now we are ready to route packet.
2107 if ((err
= fib_lookup(net
, &fl
, &res
)) != 0) {
2108 if (!IN_DEV_FORWARD(in_dev
))
2114 RT_CACHE_STAT_INC(in_slow_tot
);
2116 if (res
.type
== RTN_BROADCAST
)
2119 if (res
.type
== RTN_LOCAL
) {
2121 result
= fib_validate_source(saddr
, daddr
, tos
,
2122 net
->loopback_dev
->ifindex
,
2123 dev
, &spec_dst
, &itag
);
2125 goto martian_source
;
2127 flags
|= RTCF_DIRECTSRC
;
2132 if (!IN_DEV_FORWARD(in_dev
))
2134 if (res
.type
!= RTN_UNICAST
)
2135 goto martian_destination
;
2137 err
= ip_mkroute_input(skb
, &res
, &fl
, in_dev
, daddr
, saddr
, tos
);
2145 if (skb
->protocol
!= htons(ETH_P_IP
))
2148 if (ipv4_is_zeronet(saddr
))
2149 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2151 err
= fib_validate_source(saddr
, 0, tos
, 0, dev
, &spec_dst
,
2154 goto martian_source
;
2156 flags
|= RTCF_DIRECTSRC
;
2158 flags
|= RTCF_BROADCAST
;
2159 res
.type
= RTN_BROADCAST
;
2160 RT_CACHE_STAT_INC(in_brd
);
2163 rth
= dst_alloc(&ipv4_dst_ops
);
2167 rth
->u
.dst
.output
= ip_rt_bug
;
2168 rth
->rt_genid
= rt_genid(net
);
2170 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2171 rth
->u
.dst
.flags
= DST_HOST
;
2172 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2173 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2174 rth
->fl
.fl4_dst
= daddr
;
2175 rth
->rt_dst
= daddr
;
2176 rth
->fl
.fl4_tos
= tos
;
2177 rth
->fl
.mark
= skb
->mark
;
2178 rth
->fl
.fl4_src
= saddr
;
2179 rth
->rt_src
= saddr
;
2180 #ifdef CONFIG_NET_CLS_ROUTE
2181 rth
->u
.dst
.tclassid
= itag
;
2184 rth
->fl
.iif
= dev
->ifindex
;
2185 rth
->u
.dst
.dev
= net
->loopback_dev
;
2186 dev_hold(rth
->u
.dst
.dev
);
2187 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
2188 rth
->rt_gateway
= daddr
;
2189 rth
->rt_spec_dst
= spec_dst
;
2190 rth
->u
.dst
.input
= ip_local_deliver
;
2191 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2192 if (res
.type
== RTN_UNREACHABLE
) {
2193 rth
->u
.dst
.input
= ip_error
;
2194 rth
->u
.dst
.error
= -err
;
2195 rth
->rt_flags
&= ~RTCF_LOCAL
;
2197 rth
->rt_type
= res
.type
;
2198 hash
= rt_hash(daddr
, saddr
, fl
.iif
, rt_genid(net
));
2199 err
= rt_intern_hash(hash
, rth
, &skb
->rtable
);
2203 RT_CACHE_STAT_INC(in_no_route
);
2204 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2205 res
.type
= RTN_UNREACHABLE
;
2211 * Do not cache martian addresses: they should be logged (RFC1812)
2213 martian_destination
:
2214 RT_CACHE_STAT_INC(in_martian_dst
);
2215 #ifdef CONFIG_IP_ROUTE_VERBOSE
2216 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2217 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2218 &daddr
, &saddr
, dev
->name
);
2222 err
= -EHOSTUNREACH
;
2234 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2238 int ip_route_input(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2239 u8 tos
, struct net_device
*dev
)
2241 struct rtable
* rth
;
2243 int iif
= dev
->ifindex
;
2248 if (!rt_caching(net
))
2251 tos
&= IPTOS_RT_MASK
;
2252 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2255 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2256 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
2257 if (((rth
->fl
.fl4_dst
^ daddr
) |
2258 (rth
->fl
.fl4_src
^ saddr
) |
2259 (rth
->fl
.iif
^ iif
) |
2261 (rth
->fl
.fl4_tos
^ tos
)) == 0 &&
2262 rth
->fl
.mark
== skb
->mark
&&
2263 net_eq(dev_net(rth
->u
.dst
.dev
), net
) &&
2264 !rt_is_expired(rth
)) {
2265 dst_use(&rth
->u
.dst
, jiffies
);
2266 RT_CACHE_STAT_INC(in_hit
);
2271 RT_CACHE_STAT_INC(in_hlist_search
);
2276 /* Multicast recognition logic is moved from route cache to here.
2277 The problem was that too many Ethernet cards have broken/missing
2278 hardware multicast filters :-( As result the host on multicasting
2279 network acquires a lot of useless route cache entries, sort of
2280 SDR messages from all the world. Now we try to get rid of them.
2281 Really, provided software IP multicast filter is organized
2282 reasonably (at least, hashed), it does not result in a slowdown
2283 comparing with route cache reject entries.
2284 Note, that multicast routers are not affected, because
2285 route cache entry is created eventually.
2287 if (ipv4_is_multicast(daddr
)) {
2288 struct in_device
*in_dev
;
2291 if ((in_dev
= __in_dev_get_rcu(dev
)) != NULL
) {
2292 int our
= ip_check_mc(in_dev
, daddr
, saddr
,
2293 ip_hdr(skb
)->protocol
);
2295 #ifdef CONFIG_IP_MROUTE
2296 || (!ipv4_is_local_multicast(daddr
) &&
2297 IN_DEV_MFORWARD(in_dev
))
2301 return ip_route_input_mc(skb
, daddr
, saddr
,
2308 return ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2311 static int __mkroute_output(struct rtable
**result
,
2312 struct fib_result
*res
,
2313 const struct flowi
*fl
,
2314 const struct flowi
*oldflp
,
2315 struct net_device
*dev_out
,
2319 struct in_device
*in_dev
;
2320 u32 tos
= RT_FL_TOS(oldflp
);
2323 if (ipv4_is_loopback(fl
->fl4_src
) && !(dev_out
->flags
&IFF_LOOPBACK
))
2326 if (fl
->fl4_dst
== htonl(0xFFFFFFFF))
2327 res
->type
= RTN_BROADCAST
;
2328 else if (ipv4_is_multicast(fl
->fl4_dst
))
2329 res
->type
= RTN_MULTICAST
;
2330 else if (ipv4_is_lbcast(fl
->fl4_dst
) || ipv4_is_zeronet(fl
->fl4_dst
))
2333 if (dev_out
->flags
& IFF_LOOPBACK
)
2334 flags
|= RTCF_LOCAL
;
2336 /* get work reference to inet device */
2337 in_dev
= in_dev_get(dev_out
);
2341 if (res
->type
== RTN_BROADCAST
) {
2342 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2344 fib_info_put(res
->fi
);
2347 } else if (res
->type
== RTN_MULTICAST
) {
2348 flags
|= RTCF_MULTICAST
|RTCF_LOCAL
;
2349 if (!ip_check_mc(in_dev
, oldflp
->fl4_dst
, oldflp
->fl4_src
,
2351 flags
&= ~RTCF_LOCAL
;
2352 /* If multicast route do not exist use
2353 default one, but do not gateway in this case.
2356 if (res
->fi
&& res
->prefixlen
< 4) {
2357 fib_info_put(res
->fi
);
2363 rth
= dst_alloc(&ipv4_dst_ops
);
2369 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2370 rth
->u
.dst
.flags
= DST_HOST
;
2371 if (IN_DEV_CONF_GET(in_dev
, NOXFRM
))
2372 rth
->u
.dst
.flags
|= DST_NOXFRM
;
2373 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2374 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2376 rth
->fl
.fl4_dst
= oldflp
->fl4_dst
;
2377 rth
->fl
.fl4_tos
= tos
;
2378 rth
->fl
.fl4_src
= oldflp
->fl4_src
;
2379 rth
->fl
.oif
= oldflp
->oif
;
2380 rth
->fl
.mark
= oldflp
->mark
;
2381 rth
->rt_dst
= fl
->fl4_dst
;
2382 rth
->rt_src
= fl
->fl4_src
;
2383 rth
->rt_iif
= oldflp
->oif
? : dev_out
->ifindex
;
2384 /* get references to the devices that are to be hold by the routing
2386 rth
->u
.dst
.dev
= dev_out
;
2388 rth
->idev
= in_dev_get(dev_out
);
2389 rth
->rt_gateway
= fl
->fl4_dst
;
2390 rth
->rt_spec_dst
= fl
->fl4_src
;
2392 rth
->u
.dst
.output
=ip_output
;
2393 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2395 RT_CACHE_STAT_INC(out_slow_tot
);
2397 if (flags
& RTCF_LOCAL
) {
2398 rth
->u
.dst
.input
= ip_local_deliver
;
2399 rth
->rt_spec_dst
= fl
->fl4_dst
;
2401 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2402 rth
->rt_spec_dst
= fl
->fl4_src
;
2403 if (flags
& RTCF_LOCAL
&&
2404 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2405 rth
->u
.dst
.output
= ip_mc_output
;
2406 RT_CACHE_STAT_INC(out_slow_mc
);
2408 #ifdef CONFIG_IP_MROUTE
2409 if (res
->type
== RTN_MULTICAST
) {
2410 if (IN_DEV_MFORWARD(in_dev
) &&
2411 !ipv4_is_local_multicast(oldflp
->fl4_dst
)) {
2412 rth
->u
.dst
.input
= ip_mr_input
;
2413 rth
->u
.dst
.output
= ip_mc_output
;
2419 rt_set_nexthop(rth
, res
, 0);
2421 rth
->rt_flags
= flags
;
2425 /* release work reference to inet device */
2431 static int ip_mkroute_output(struct rtable
**rp
,
2432 struct fib_result
*res
,
2433 const struct flowi
*fl
,
2434 const struct flowi
*oldflp
,
2435 struct net_device
*dev_out
,
2438 struct rtable
*rth
= NULL
;
2439 int err
= __mkroute_output(&rth
, res
, fl
, oldflp
, dev_out
, flags
);
2442 hash
= rt_hash(oldflp
->fl4_dst
, oldflp
->fl4_src
, oldflp
->oif
,
2443 rt_genid(dev_net(dev_out
)));
2444 err
= rt_intern_hash(hash
, rth
, rp
);
2451 * Major route resolver routine.
2454 static int ip_route_output_slow(struct net
*net
, struct rtable
**rp
,
2455 const struct flowi
*oldflp
)
2457 u32 tos
= RT_FL_TOS(oldflp
);
2458 struct flowi fl
= { .nl_u
= { .ip4_u
=
2459 { .daddr
= oldflp
->fl4_dst
,
2460 .saddr
= oldflp
->fl4_src
,
2461 .tos
= tos
& IPTOS_RT_MASK
,
2462 .scope
= ((tos
& RTO_ONLINK
) ?
2466 .mark
= oldflp
->mark
,
2467 .iif
= net
->loopback_dev
->ifindex
,
2468 .oif
= oldflp
->oif
};
2469 struct fib_result res
;
2471 struct net_device
*dev_out
= NULL
;
2477 #ifdef CONFIG_IP_MULTIPLE_TABLES
2481 if (oldflp
->fl4_src
) {
2483 if (ipv4_is_multicast(oldflp
->fl4_src
) ||
2484 ipv4_is_lbcast(oldflp
->fl4_src
) ||
2485 ipv4_is_zeronet(oldflp
->fl4_src
))
2488 /* I removed check for oif == dev_out->oif here.
2489 It was wrong for two reasons:
2490 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2491 is assigned to multiple interfaces.
2492 2. Moreover, we are allowed to send packets with saddr
2493 of another iface. --ANK
2496 if (oldflp
->oif
== 0
2497 && (ipv4_is_multicast(oldflp
->fl4_dst
) ||
2498 oldflp
->fl4_dst
== htonl(0xFFFFFFFF))) {
2499 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2500 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2501 if (dev_out
== NULL
)
2504 /* Special hack: user can direct multicasts
2505 and limited broadcast via necessary interface
2506 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2507 This hack is not just for fun, it allows
2508 vic,vat and friends to work.
2509 They bind socket to loopback, set ttl to zero
2510 and expect that it will work.
2511 From the viewpoint of routing cache they are broken,
2512 because we are not allowed to build multicast path
2513 with loopback source addr (look, routing cache
2514 cannot know, that ttl is zero, so that packet
2515 will not leave this host and route is valid).
2516 Luckily, this hack is good workaround.
2519 fl
.oif
= dev_out
->ifindex
;
2523 if (!(oldflp
->flags
& FLOWI_FLAG_ANYSRC
)) {
2524 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2525 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2526 if (dev_out
== NULL
)
2535 dev_out
= dev_get_by_index(net
, oldflp
->oif
);
2537 if (dev_out
== NULL
)
2540 /* RACE: Check return value of inet_select_addr instead. */
2541 if (__in_dev_get_rtnl(dev_out
) == NULL
) {
2543 goto out
; /* Wrong error code */
2546 if (ipv4_is_local_multicast(oldflp
->fl4_dst
) ||
2547 oldflp
->fl4_dst
== htonl(0xFFFFFFFF)) {
2549 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2554 if (ipv4_is_multicast(oldflp
->fl4_dst
))
2555 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2557 else if (!oldflp
->fl4_dst
)
2558 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2564 fl
.fl4_dst
= fl
.fl4_src
;
2566 fl
.fl4_dst
= fl
.fl4_src
= htonl(INADDR_LOOPBACK
);
2569 dev_out
= net
->loopback_dev
;
2571 fl
.oif
= net
->loopback_dev
->ifindex
;
2572 res
.type
= RTN_LOCAL
;
2573 flags
|= RTCF_LOCAL
;
2577 if (fib_lookup(net
, &fl
, &res
)) {
2580 /* Apparently, routing tables are wrong. Assume,
2581 that the destination is on link.
2584 Because we are allowed to send to iface
2585 even if it has NO routes and NO assigned
2586 addresses. When oif is specified, routing
2587 tables are looked up with only one purpose:
2588 to catch if destination is gatewayed, rather than
2589 direct. Moreover, if MSG_DONTROUTE is set,
2590 we send packet, ignoring both routing tables
2591 and ifaddr state. --ANK
2594 We could make it even if oif is unknown,
2595 likely IPv6, but we do not.
2598 if (fl
.fl4_src
== 0)
2599 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2601 res
.type
= RTN_UNICAST
;
2611 if (res
.type
== RTN_LOCAL
) {
2613 fl
.fl4_src
= fl
.fl4_dst
;
2616 dev_out
= net
->loopback_dev
;
2618 fl
.oif
= dev_out
->ifindex
;
2620 fib_info_put(res
.fi
);
2622 flags
|= RTCF_LOCAL
;
2626 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2627 if (res
.fi
->fib_nhs
> 1 && fl
.oif
== 0)
2628 fib_select_multipath(&fl
, &res
);
2631 if (!res
.prefixlen
&& res
.type
== RTN_UNICAST
&& !fl
.oif
)
2632 fib_select_default(net
, &fl
, &res
);
2635 fl
.fl4_src
= FIB_RES_PREFSRC(res
);
2639 dev_out
= FIB_RES_DEV(res
);
2641 fl
.oif
= dev_out
->ifindex
;
2645 err
= ip_mkroute_output(rp
, &res
, &fl
, oldflp
, dev_out
, flags
);
2655 int __ip_route_output_key(struct net
*net
, struct rtable
**rp
,
2656 const struct flowi
*flp
)
2661 if (!rt_caching(net
))
2664 hash
= rt_hash(flp
->fl4_dst
, flp
->fl4_src
, flp
->oif
, rt_genid(net
));
2667 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2668 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
2669 if (rth
->fl
.fl4_dst
== flp
->fl4_dst
&&
2670 rth
->fl
.fl4_src
== flp
->fl4_src
&&
2672 rth
->fl
.oif
== flp
->oif
&&
2673 rth
->fl
.mark
== flp
->mark
&&
2674 !((rth
->fl
.fl4_tos
^ flp
->fl4_tos
) &
2675 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2676 net_eq(dev_net(rth
->u
.dst
.dev
), net
) &&
2677 !rt_is_expired(rth
)) {
2678 dst_use(&rth
->u
.dst
, jiffies
);
2679 RT_CACHE_STAT_INC(out_hit
);
2680 rcu_read_unlock_bh();
2684 RT_CACHE_STAT_INC(out_hlist_search
);
2686 rcu_read_unlock_bh();
2689 return ip_route_output_slow(net
, rp
, flp
);
2692 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2694 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2698 static struct dst_ops ipv4_dst_blackhole_ops
= {
2700 .protocol
= __constant_htons(ETH_P_IP
),
2701 .destroy
= ipv4_dst_destroy
,
2702 .check
= ipv4_dst_check
,
2703 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2704 .entries
= ATOMIC_INIT(0),
2708 static int ipv4_dst_blackhole(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2710 struct rtable
*ort
= *rp
;
2711 struct rtable
*rt
= (struct rtable
*)
2712 dst_alloc(&ipv4_dst_blackhole_ops
);
2715 struct dst_entry
*new = &rt
->u
.dst
;
2717 atomic_set(&new->__refcnt
, 1);
2719 new->input
= dst_discard
;
2720 new->output
= dst_discard
;
2721 memcpy(new->metrics
, ort
->u
.dst
.metrics
, RTAX_MAX
*sizeof(u32
));
2723 new->dev
= ort
->u
.dst
.dev
;
2729 rt
->idev
= ort
->idev
;
2731 in_dev_hold(rt
->idev
);
2732 rt
->rt_genid
= rt_genid(net
);
2733 rt
->rt_flags
= ort
->rt_flags
;
2734 rt
->rt_type
= ort
->rt_type
;
2735 rt
->rt_dst
= ort
->rt_dst
;
2736 rt
->rt_src
= ort
->rt_src
;
2737 rt
->rt_iif
= ort
->rt_iif
;
2738 rt
->rt_gateway
= ort
->rt_gateway
;
2739 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2740 rt
->peer
= ort
->peer
;
2742 atomic_inc(&rt
->peer
->refcnt
);
2747 dst_release(&(*rp
)->u
.dst
);
2749 return (rt
? 0 : -ENOMEM
);
2752 int ip_route_output_flow(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
,
2753 struct sock
*sk
, int flags
)
2757 if ((err
= __ip_route_output_key(net
, rp
, flp
)) != 0)
2762 flp
->fl4_src
= (*rp
)->rt_src
;
2764 flp
->fl4_dst
= (*rp
)->rt_dst
;
2765 err
= __xfrm_lookup(net
, (struct dst_entry
**)rp
, flp
, sk
,
2766 flags
? XFRM_LOOKUP_WAIT
: 0);
2767 if (err
== -EREMOTE
)
2768 err
= ipv4_dst_blackhole(net
, rp
, flp
);
2776 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2778 int ip_route_output_key(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2780 return ip_route_output_flow(net
, rp
, flp
, NULL
, 0);
2783 static int rt_fill_info(struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2784 int nowait
, unsigned int flags
)
2786 struct rtable
*rt
= skb
->rtable
;
2788 struct nlmsghdr
*nlh
;
2790 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2792 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2796 r
= nlmsg_data(nlh
);
2797 r
->rtm_family
= AF_INET
;
2798 r
->rtm_dst_len
= 32;
2800 r
->rtm_tos
= rt
->fl
.fl4_tos
;
2801 r
->rtm_table
= RT_TABLE_MAIN
;
2802 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2803 r
->rtm_type
= rt
->rt_type
;
2804 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2805 r
->rtm_protocol
= RTPROT_UNSPEC
;
2806 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2807 if (rt
->rt_flags
& RTCF_NOTIFY
)
2808 r
->rtm_flags
|= RTM_F_NOTIFY
;
2810 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2812 if (rt
->fl
.fl4_src
) {
2813 r
->rtm_src_len
= 32;
2814 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->fl
.fl4_src
);
2817 NLA_PUT_U32(skb
, RTA_OIF
, rt
->u
.dst
.dev
->ifindex
);
2818 #ifdef CONFIG_NET_CLS_ROUTE
2819 if (rt
->u
.dst
.tclassid
)
2820 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->u
.dst
.tclassid
);
2823 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
2824 else if (rt
->rt_src
!= rt
->fl
.fl4_src
)
2825 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
2827 if (rt
->rt_dst
!= rt
->rt_gateway
)
2828 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
2830 if (rtnetlink_put_metrics(skb
, rt
->u
.dst
.metrics
) < 0)
2831 goto nla_put_failure
;
2833 error
= rt
->u
.dst
.error
;
2834 expires
= rt
->u
.dst
.expires
? rt
->u
.dst
.expires
- jiffies
: 0;
2836 id
= rt
->peer
->ip_id_count
;
2837 if (rt
->peer
->tcp_ts_stamp
) {
2838 ts
= rt
->peer
->tcp_ts
;
2839 tsage
= get_seconds() - rt
->peer
->tcp_ts_stamp
;
2844 #ifdef CONFIG_IP_MROUTE
2845 __be32 dst
= rt
->rt_dst
;
2847 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2848 IPV4_DEVCONF_ALL(&init_net
, MC_FORWARDING
)) {
2849 int err
= ipmr_get_route(skb
, r
, nowait
);
2854 goto nla_put_failure
;
2856 if (err
== -EMSGSIZE
)
2857 goto nla_put_failure
;
2863 NLA_PUT_U32(skb
, RTA_IIF
, rt
->fl
.iif
);
2866 if (rtnl_put_cacheinfo(skb
, &rt
->u
.dst
, id
, ts
, tsage
,
2867 expires
, error
) < 0)
2868 goto nla_put_failure
;
2870 return nlmsg_end(skb
, nlh
);
2873 nlmsg_cancel(skb
, nlh
);
2877 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
2879 struct net
*net
= sock_net(in_skb
->sk
);
2881 struct nlattr
*tb
[RTA_MAX
+1];
2882 struct rtable
*rt
= NULL
;
2887 struct sk_buff
*skb
;
2889 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2893 rtm
= nlmsg_data(nlh
);
2895 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2901 /* Reserve room for dummy headers, this skb can pass
2902 through good chunk of routing engine.
2904 skb_reset_mac_header(skb
);
2905 skb_reset_network_header(skb
);
2907 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2908 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
2909 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2911 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
2912 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
2913 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2916 struct net_device
*dev
;
2918 dev
= __dev_get_by_index(net
, iif
);
2924 skb
->protocol
= htons(ETH_P_IP
);
2927 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
2931 if (err
== 0 && rt
->u
.dst
.error
)
2932 err
= -rt
->u
.dst
.error
;
2939 .tos
= rtm
->rtm_tos
,
2942 .oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
2944 err
= ip_route_output_key(net
, &rt
, &fl
);
2951 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2952 rt
->rt_flags
|= RTCF_NOTIFY
;
2954 err
= rt_fill_info(skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
2955 RTM_NEWROUTE
, 0, 0);
2959 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
2968 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2975 net
= sock_net(skb
->sk
);
2980 s_idx
= idx
= cb
->args
[1];
2981 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
2982 if (!rt_hash_table
[h
].chain
)
2985 for (rt
= rcu_dereference(rt_hash_table
[h
].chain
), idx
= 0; rt
;
2986 rt
= rcu_dereference(rt
->u
.dst
.rt_next
), idx
++) {
2987 if (!net_eq(dev_net(rt
->u
.dst
.dev
), net
) || idx
< s_idx
)
2989 if (rt_is_expired(rt
))
2991 skb
->dst
= dst_clone(&rt
->u
.dst
);
2992 if (rt_fill_info(skb
, NETLINK_CB(cb
->skb
).pid
,
2993 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
2994 1, NLM_F_MULTI
) <= 0) {
2995 dst_release(xchg(&skb
->dst
, NULL
));
2996 rcu_read_unlock_bh();
2999 dst_release(xchg(&skb
->dst
, NULL
));
3001 rcu_read_unlock_bh();
3010 void ip_rt_multicast_event(struct in_device
*in_dev
)
3012 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3015 #ifdef CONFIG_SYSCTL
3016 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3017 struct file
*filp
, void __user
*buffer
,
3018 size_t *lenp
, loff_t
*ppos
)
3025 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3026 ctl
.data
= &flush_delay
;
3027 proc_dointvec(&ctl
, write
, filp
, buffer
, lenp
, ppos
);
3029 net
= (struct net
*)__ctl
->extra1
;
3030 rt_cache_flush(net
, flush_delay
);
3037 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table
*table
,
3038 void __user
*oldval
,
3039 size_t __user
*oldlenp
,
3040 void __user
*newval
,
3045 if (newlen
!= sizeof(int))
3047 if (get_user(delay
, (int __user
*)newval
))
3049 net
= (struct net
*)table
->extra1
;
3050 rt_cache_flush(net
, delay
);
3054 static void rt_secret_reschedule(int old
)
3057 int new = ip_rt_secret_interval
;
3058 int diff
= new - old
;
3065 int deleted
= del_timer_sync(&net
->ipv4
.rt_secret_timer
);
3071 long time
= net
->ipv4
.rt_secret_timer
.expires
- jiffies
;
3073 if (time
<= 0 || (time
+= diff
) <= 0)
3076 net
->ipv4
.rt_secret_timer
.expires
= time
;
3078 net
->ipv4
.rt_secret_timer
.expires
= new;
3080 net
->ipv4
.rt_secret_timer
.expires
+= jiffies
;
3081 add_timer(&net
->ipv4
.rt_secret_timer
);
3086 static int ipv4_sysctl_rt_secret_interval(ctl_table
*ctl
, int write
,
3088 void __user
*buffer
, size_t *lenp
,
3091 int old
= ip_rt_secret_interval
;
3092 int ret
= proc_dointvec_jiffies(ctl
, write
, filp
, buffer
, lenp
, ppos
);
3094 rt_secret_reschedule(old
);
3099 static int ipv4_sysctl_rt_secret_interval_strategy(ctl_table
*table
,
3100 void __user
*oldval
,
3101 size_t __user
*oldlenp
,
3102 void __user
*newval
,
3105 int old
= ip_rt_secret_interval
;
3106 int ret
= sysctl_jiffies(table
, oldval
, oldlenp
, newval
, newlen
);
3108 rt_secret_reschedule(old
);
3113 static ctl_table ipv4_route_table
[] = {
3115 .ctl_name
= NET_IPV4_ROUTE_GC_THRESH
,
3116 .procname
= "gc_thresh",
3117 .data
= &ipv4_dst_ops
.gc_thresh
,
3118 .maxlen
= sizeof(int),
3120 .proc_handler
= proc_dointvec
,
3123 .ctl_name
= NET_IPV4_ROUTE_MAX_SIZE
,
3124 .procname
= "max_size",
3125 .data
= &ip_rt_max_size
,
3126 .maxlen
= sizeof(int),
3128 .proc_handler
= proc_dointvec
,
3131 /* Deprecated. Use gc_min_interval_ms */
3133 .ctl_name
= NET_IPV4_ROUTE_GC_MIN_INTERVAL
,
3134 .procname
= "gc_min_interval",
3135 .data
= &ip_rt_gc_min_interval
,
3136 .maxlen
= sizeof(int),
3138 .proc_handler
= proc_dointvec_jiffies
,
3139 .strategy
= sysctl_jiffies
,
3142 .ctl_name
= NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS
,
3143 .procname
= "gc_min_interval_ms",
3144 .data
= &ip_rt_gc_min_interval
,
3145 .maxlen
= sizeof(int),
3147 .proc_handler
= proc_dointvec_ms_jiffies
,
3148 .strategy
= sysctl_ms_jiffies
,
3151 .ctl_name
= NET_IPV4_ROUTE_GC_TIMEOUT
,
3152 .procname
= "gc_timeout",
3153 .data
= &ip_rt_gc_timeout
,
3154 .maxlen
= sizeof(int),
3156 .proc_handler
= proc_dointvec_jiffies
,
3157 .strategy
= sysctl_jiffies
,
3160 .ctl_name
= NET_IPV4_ROUTE_GC_INTERVAL
,
3161 .procname
= "gc_interval",
3162 .data
= &ip_rt_gc_interval
,
3163 .maxlen
= sizeof(int),
3165 .proc_handler
= proc_dointvec_jiffies
,
3166 .strategy
= sysctl_jiffies
,
3169 .ctl_name
= NET_IPV4_ROUTE_REDIRECT_LOAD
,
3170 .procname
= "redirect_load",
3171 .data
= &ip_rt_redirect_load
,
3172 .maxlen
= sizeof(int),
3174 .proc_handler
= proc_dointvec
,
3177 .ctl_name
= NET_IPV4_ROUTE_REDIRECT_NUMBER
,
3178 .procname
= "redirect_number",
3179 .data
= &ip_rt_redirect_number
,
3180 .maxlen
= sizeof(int),
3182 .proc_handler
= proc_dointvec
,
3185 .ctl_name
= NET_IPV4_ROUTE_REDIRECT_SILENCE
,
3186 .procname
= "redirect_silence",
3187 .data
= &ip_rt_redirect_silence
,
3188 .maxlen
= sizeof(int),
3190 .proc_handler
= proc_dointvec
,
3193 .ctl_name
= NET_IPV4_ROUTE_ERROR_COST
,
3194 .procname
= "error_cost",
3195 .data
= &ip_rt_error_cost
,
3196 .maxlen
= sizeof(int),
3198 .proc_handler
= proc_dointvec
,
3201 .ctl_name
= NET_IPV4_ROUTE_ERROR_BURST
,
3202 .procname
= "error_burst",
3203 .data
= &ip_rt_error_burst
,
3204 .maxlen
= sizeof(int),
3206 .proc_handler
= proc_dointvec
,
3209 .ctl_name
= NET_IPV4_ROUTE_GC_ELASTICITY
,
3210 .procname
= "gc_elasticity",
3211 .data
= &ip_rt_gc_elasticity
,
3212 .maxlen
= sizeof(int),
3214 .proc_handler
= proc_dointvec
,
3217 .ctl_name
= NET_IPV4_ROUTE_MTU_EXPIRES
,
3218 .procname
= "mtu_expires",
3219 .data
= &ip_rt_mtu_expires
,
3220 .maxlen
= sizeof(int),
3222 .proc_handler
= proc_dointvec_jiffies
,
3223 .strategy
= sysctl_jiffies
,
3226 .ctl_name
= NET_IPV4_ROUTE_MIN_PMTU
,
3227 .procname
= "min_pmtu",
3228 .data
= &ip_rt_min_pmtu
,
3229 .maxlen
= sizeof(int),
3231 .proc_handler
= proc_dointvec
,
3234 .ctl_name
= NET_IPV4_ROUTE_MIN_ADVMSS
,
3235 .procname
= "min_adv_mss",
3236 .data
= &ip_rt_min_advmss
,
3237 .maxlen
= sizeof(int),
3239 .proc_handler
= proc_dointvec
,
3242 .ctl_name
= NET_IPV4_ROUTE_SECRET_INTERVAL
,
3243 .procname
= "secret_interval",
3244 .data
= &ip_rt_secret_interval
,
3245 .maxlen
= sizeof(int),
3247 .proc_handler
= ipv4_sysctl_rt_secret_interval
,
3248 .strategy
= ipv4_sysctl_rt_secret_interval_strategy
,
3253 static struct ctl_table empty
[1];
3255 static struct ctl_table ipv4_skeleton
[] =
3257 { .procname
= "route", .ctl_name
= NET_IPV4_ROUTE
,
3258 .mode
= 0555, .child
= ipv4_route_table
},
3259 { .procname
= "neigh", .ctl_name
= NET_IPV4_NEIGH
,
3260 .mode
= 0555, .child
= empty
},
3264 static __net_initdata
struct ctl_path ipv4_path
[] = {
3265 { .procname
= "net", .ctl_name
= CTL_NET
, },
3266 { .procname
= "ipv4", .ctl_name
= NET_IPV4
, },
3270 static struct ctl_table ipv4_route_flush_table
[] = {
3272 .ctl_name
= NET_IPV4_ROUTE_FLUSH
,
3273 .procname
= "flush",
3274 .maxlen
= sizeof(int),
3276 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3277 .strategy
= ipv4_sysctl_rtcache_flush_strategy
,
3282 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3283 { .procname
= "net", .ctl_name
= CTL_NET
, },
3284 { .procname
= "ipv4", .ctl_name
= NET_IPV4
, },
3285 { .procname
= "route", .ctl_name
= NET_IPV4_ROUTE
, },
3289 static __net_init
int sysctl_route_net_init(struct net
*net
)
3291 struct ctl_table
*tbl
;
3293 tbl
= ipv4_route_flush_table
;
3294 if (net
!= &init_net
) {
3295 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3299 tbl
[0].extra1
= net
;
3301 net
->ipv4
.route_hdr
=
3302 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3303 if (net
->ipv4
.route_hdr
== NULL
)
3308 if (tbl
!= ipv4_route_flush_table
)
3314 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3316 struct ctl_table
*tbl
;
3318 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3319 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3320 BUG_ON(tbl
== ipv4_route_flush_table
);
3324 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3325 .init
= sysctl_route_net_init
,
3326 .exit
= sysctl_route_net_exit
,
3331 static __net_init
int rt_secret_timer_init(struct net
*net
)
3333 atomic_set(&net
->ipv4
.rt_genid
,
3334 (int) ((num_physpages
^ (num_physpages
>>8)) ^
3335 (jiffies
^ (jiffies
>> 7))));
3337 net
->ipv4
.rt_secret_timer
.function
= rt_secret_rebuild
;
3338 net
->ipv4
.rt_secret_timer
.data
= (unsigned long)net
;
3339 init_timer_deferrable(&net
->ipv4
.rt_secret_timer
);
3341 if (ip_rt_secret_interval
) {
3342 net
->ipv4
.rt_secret_timer
.expires
=
3343 jiffies
+ net_random() % ip_rt_secret_interval
+
3344 ip_rt_secret_interval
;
3345 add_timer(&net
->ipv4
.rt_secret_timer
);
3350 static __net_exit
void rt_secret_timer_exit(struct net
*net
)
3352 del_timer_sync(&net
->ipv4
.rt_secret_timer
);
3355 static __net_initdata
struct pernet_operations rt_secret_timer_ops
= {
3356 .init
= rt_secret_timer_init
,
3357 .exit
= rt_secret_timer_exit
,
3361 #ifdef CONFIG_NET_CLS_ROUTE
3362 struct ip_rt_acct
*ip_rt_acct __read_mostly
;
3363 #endif /* CONFIG_NET_CLS_ROUTE */
3365 static __initdata
unsigned long rhash_entries
;
3366 static int __init
set_rhash_entries(char *str
)
3370 rhash_entries
= simple_strtoul(str
, &str
, 0);
3373 __setup("rhash_entries=", set_rhash_entries
);
3375 int __init
ip_rt_init(void)
3379 #ifdef CONFIG_NET_CLS_ROUTE
3380 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
));
3382 panic("IP: failed to allocate ip_rt_acct\n");
3385 ipv4_dst_ops
.kmem_cachep
=
3386 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3387 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3389 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3391 rt_hash_table
= (struct rt_hash_bucket
*)
3392 alloc_large_system_hash("IP route cache",
3393 sizeof(struct rt_hash_bucket
),
3395 (num_physpages
>= 128 * 1024) ?
3401 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3402 rt_hash_lock_init();
3404 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3405 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3410 /* All the timers, started at system startup tend
3411 to synchronize. Perturb it a bit.
3413 schedule_delayed_work(&expires_work
,
3414 net_random() % ip_rt_gc_interval
+ ip_rt_gc_interval
);
3416 if (register_pernet_subsys(&rt_secret_timer_ops
))
3417 printk(KERN_ERR
"Unable to setup rt_secret_timer\n");
3419 if (ip_rt_proc_init())
3420 printk(KERN_ERR
"Unable to create route proc files\n");
3425 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
);
3427 #ifdef CONFIG_SYSCTL
3428 register_pernet_subsys(&sysctl_route_ops
);
3433 #ifdef CONFIG_SYSCTL
3435 * We really need to sanitize the damn ipv4 init order, then all
3436 * this nonsense will go away.
3438 void __init
ip_static_sysctl_init(void)
3440 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);
3444 EXPORT_SYMBOL(__ip_select_ident
);
3445 EXPORT_SYMBOL(ip_route_input
);
3446 EXPORT_SYMBOL(ip_route_output_key
);