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 rt_acct_proc_show(struct seq_file
*m
, void *v
)
518 struct ip_rt_acct
*dst
, *src
;
521 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
525 for_each_possible_cpu(i
) {
526 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
527 for (j
= 0; j
< 256; j
++) {
528 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
529 dst
[j
].o_packets
+= src
[j
].o_packets
;
530 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
531 dst
[j
].i_packets
+= src
[j
].i_packets
;
535 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
540 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
542 return single_open(file
, rt_acct_proc_show
, NULL
);
545 static const struct file_operations rt_acct_proc_fops
= {
546 .owner
= THIS_MODULE
,
547 .open
= rt_acct_proc_open
,
550 .release
= single_release
,
554 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
556 struct proc_dir_entry
*pde
;
558 pde
= proc_net_fops_create(net
, "rt_cache", S_IRUGO
,
563 pde
= proc_create("rt_cache", S_IRUGO
,
564 net
->proc_net_stat
, &rt_cpu_seq_fops
);
568 #ifdef CONFIG_NET_CLS_ROUTE
569 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
575 #ifdef CONFIG_NET_CLS_ROUTE
577 remove_proc_entry("rt_cache", net
->proc_net_stat
);
580 remove_proc_entry("rt_cache", net
->proc_net
);
585 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
587 remove_proc_entry("rt_cache", net
->proc_net_stat
);
588 remove_proc_entry("rt_cache", net
->proc_net
);
589 #ifdef CONFIG_NET_CLS_ROUTE
590 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 net_eq(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());
905 /* Flush previous cache invalidated entries from the cache */
906 void rt_cache_flush_batch(void)
908 rt_do_flush(!in_softirq());
912 * We change rt_genid and let gc do the cleanup
914 static void rt_secret_rebuild(unsigned long __net
)
916 struct net
*net
= (struct net
*)__net
;
917 rt_cache_invalidate(net
);
918 mod_timer(&net
->ipv4
.rt_secret_timer
, jiffies
+ ip_rt_secret_interval
);
921 static void rt_secret_rebuild_oneshot(struct net
*net
)
923 del_timer_sync(&net
->ipv4
.rt_secret_timer
);
924 rt_cache_invalidate(net
);
925 if (ip_rt_secret_interval
)
926 mod_timer(&net
->ipv4
.rt_secret_timer
, jiffies
+ ip_rt_secret_interval
);
929 static void rt_emergency_hash_rebuild(struct net
*net
)
931 if (net_ratelimit()) {
932 printk(KERN_WARNING
"Route hash chain too long!\n");
933 printk(KERN_WARNING
"Adjust your secret_interval!\n");
936 rt_secret_rebuild_oneshot(net
);
940 Short description of GC goals.
942 We want to build algorithm, which will keep routing cache
943 at some equilibrium point, when number of aged off entries
944 is kept approximately equal to newly generated ones.
946 Current expiration strength is variable "expire".
947 We try to adjust it dynamically, so that if networking
948 is idle expires is large enough to keep enough of warm entries,
949 and when load increases it reduces to limit cache size.
952 static int rt_garbage_collect(struct dst_ops
*ops
)
954 static unsigned long expire
= RT_GC_TIMEOUT
;
955 static unsigned long last_gc
;
957 static int equilibrium
;
958 struct rtable
*rth
, **rthp
;
959 unsigned long now
= jiffies
;
963 * Garbage collection is pretty expensive,
964 * do not make it too frequently.
967 RT_CACHE_STAT_INC(gc_total
);
969 if (now
- last_gc
< ip_rt_gc_min_interval
&&
970 atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
) {
971 RT_CACHE_STAT_INC(gc_ignored
);
975 /* Calculate number of entries, which we want to expire now. */
976 goal
= atomic_read(&ipv4_dst_ops
.entries
) -
977 (ip_rt_gc_elasticity
<< rt_hash_log
);
979 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
980 equilibrium
= ipv4_dst_ops
.gc_thresh
;
981 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
983 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
984 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
987 /* We are in dangerous area. Try to reduce cache really
990 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
991 equilibrium
= atomic_read(&ipv4_dst_ops
.entries
) - goal
;
994 if (now
- last_gc
>= ip_rt_gc_min_interval
)
1005 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
1006 unsigned long tmo
= expire
;
1008 k
= (k
+ 1) & rt_hash_mask
;
1009 rthp
= &rt_hash_table
[k
].chain
;
1010 spin_lock_bh(rt_hash_lock_addr(k
));
1011 while ((rth
= *rthp
) != NULL
) {
1012 if (!rt_is_expired(rth
) &&
1013 !rt_may_expire(rth
, tmo
, expire
)) {
1015 rthp
= &rth
->u
.dst
.rt_next
;
1018 *rthp
= rth
->u
.dst
.rt_next
;
1022 spin_unlock_bh(rt_hash_lock_addr(k
));
1031 /* Goal is not achieved. We stop process if:
1033 - if expire reduced to zero. Otherwise, expire is halfed.
1034 - if table is not full.
1035 - if we are called from interrupt.
1036 - jiffies check is just fallback/debug loop breaker.
1037 We will not spin here for long time in any case.
1040 RT_CACHE_STAT_INC(gc_goal_miss
);
1046 #if RT_CACHE_DEBUG >= 2
1047 printk(KERN_DEBUG
"expire>> %u %d %d %d\n", expire
,
1048 atomic_read(&ipv4_dst_ops
.entries
), goal
, i
);
1051 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1053 } while (!in_softirq() && time_before_eq(jiffies
, now
));
1055 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1057 if (net_ratelimit())
1058 printk(KERN_WARNING
"dst cache overflow\n");
1059 RT_CACHE_STAT_INC(gc_dst_overflow
);
1063 expire
+= ip_rt_gc_min_interval
;
1064 if (expire
> ip_rt_gc_timeout
||
1065 atomic_read(&ipv4_dst_ops
.entries
) < ipv4_dst_ops
.gc_thresh
)
1066 expire
= ip_rt_gc_timeout
;
1067 #if RT_CACHE_DEBUG >= 2
1068 printk(KERN_DEBUG
"expire++ %u %d %d %d\n", expire
,
1069 atomic_read(&ipv4_dst_ops
.entries
), goal
, rover
);
1074 static int rt_intern_hash(unsigned hash
, struct rtable
*rt
,
1075 struct rtable
**rp
, struct sk_buff
*skb
)
1077 struct rtable
*rth
, **rthp
;
1079 struct rtable
*cand
, **candp
;
1082 int attempts
= !in_softirq();
1086 min_score
= ~(u32
)0;
1091 if (!rt_caching(dev_net(rt
->u
.dst
.dev
))) {
1093 * If we're not caching, just tell the caller we
1094 * were successful and don't touch the route. The
1095 * caller hold the sole reference to the cache entry, and
1096 * it will be released when the caller is done with it.
1097 * If we drop it here, the callers have no way to resolve routes
1098 * when we're not caching. Instead, just point *rp at rt, so
1099 * the caller gets a single use out of the route
1100 * Note that we do rt_free on this new route entry, so that
1101 * once its refcount hits zero, we are still able to reap it
1103 * Note also the rt_free uses call_rcu. We don't actually
1104 * need rcu protection here, this is just our path to get
1105 * on the route gc list.
1108 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1109 int err
= arp_bind_neighbour(&rt
->u
.dst
);
1111 if (net_ratelimit())
1113 "Neighbour table failure & not caching routes.\n");
1123 rthp
= &rt_hash_table
[hash
].chain
;
1125 spin_lock_bh(rt_hash_lock_addr(hash
));
1126 while ((rth
= *rthp
) != NULL
) {
1127 if (rt_is_expired(rth
)) {
1128 *rthp
= rth
->u
.dst
.rt_next
;
1132 if (compare_keys(&rth
->fl
, &rt
->fl
) && compare_netns(rth
, rt
)) {
1134 *rthp
= rth
->u
.dst
.rt_next
;
1136 * Since lookup is lockfree, the deletion
1137 * must be visible to another weakly ordered CPU before
1138 * the insertion at the start of the hash chain.
1140 rcu_assign_pointer(rth
->u
.dst
.rt_next
,
1141 rt_hash_table
[hash
].chain
);
1143 * Since lookup is lockfree, the update writes
1144 * must be ordered for consistency on SMP.
1146 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1148 dst_use(&rth
->u
.dst
, now
);
1149 spin_unlock_bh(rt_hash_lock_addr(hash
));
1155 skb_dst_set(skb
, &rth
->u
.dst
);
1159 if (!atomic_read(&rth
->u
.dst
.__refcnt
)) {
1160 u32 score
= rt_score(rth
);
1162 if (score
<= min_score
) {
1171 rthp
= &rth
->u
.dst
.rt_next
;
1175 /* ip_rt_gc_elasticity used to be average length of chain
1176 * length, when exceeded gc becomes really aggressive.
1178 * The second limit is less certain. At the moment it allows
1179 * only 2 entries per bucket. We will see.
1181 if (chain_length
> ip_rt_gc_elasticity
) {
1182 *candp
= cand
->u
.dst
.rt_next
;
1186 if (chain_length
> rt_chain_length_max
) {
1187 struct net
*net
= dev_net(rt
->u
.dst
.dev
);
1188 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1189 if (!rt_caching(dev_net(rt
->u
.dst
.dev
))) {
1190 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1191 rt
->u
.dst
.dev
->name
, num
);
1193 rt_emergency_hash_rebuild(dev_net(rt
->u
.dst
.dev
));
1197 /* Try to bind route to arp only if it is output
1198 route or unicast forwarding path.
1200 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1201 int err
= arp_bind_neighbour(&rt
->u
.dst
);
1203 spin_unlock_bh(rt_hash_lock_addr(hash
));
1205 if (err
!= -ENOBUFS
) {
1210 /* Neighbour tables are full and nothing
1211 can be released. Try to shrink route cache,
1212 it is most likely it holds some neighbour records.
1214 if (attempts
-- > 0) {
1215 int saved_elasticity
= ip_rt_gc_elasticity
;
1216 int saved_int
= ip_rt_gc_min_interval
;
1217 ip_rt_gc_elasticity
= 1;
1218 ip_rt_gc_min_interval
= 0;
1219 rt_garbage_collect(&ipv4_dst_ops
);
1220 ip_rt_gc_min_interval
= saved_int
;
1221 ip_rt_gc_elasticity
= saved_elasticity
;
1225 if (net_ratelimit())
1226 printk(KERN_WARNING
"Neighbour table overflow.\n");
1232 rt
->u
.dst
.rt_next
= rt_hash_table
[hash
].chain
;
1234 #if RT_CACHE_DEBUG >= 2
1235 if (rt
->u
.dst
.rt_next
) {
1237 printk(KERN_DEBUG
"rt_cache @%02x: %pI4",
1239 for (trt
= rt
->u
.dst
.rt_next
; trt
; trt
= trt
->u
.dst
.rt_next
)
1240 printk(" . %pI4", &trt
->rt_dst
);
1245 * Since lookup is lockfree, we must make sure
1246 * previous writes to rt are comitted to memory
1247 * before making rt visible to other CPUS.
1249 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1251 spin_unlock_bh(rt_hash_lock_addr(hash
));
1257 skb_dst_set(skb
, &rt
->u
.dst
);
1261 void rt_bind_peer(struct rtable
*rt
, int create
)
1263 static DEFINE_SPINLOCK(rt_peer_lock
);
1264 struct inet_peer
*peer
;
1266 peer
= inet_getpeer(rt
->rt_dst
, create
);
1268 spin_lock_bh(&rt_peer_lock
);
1269 if (rt
->peer
== NULL
) {
1273 spin_unlock_bh(&rt_peer_lock
);
1279 * Peer allocation may fail only in serious out-of-memory conditions. However
1280 * we still can generate some output.
1281 * Random ID selection looks a bit dangerous because we have no chances to
1282 * select ID being unique in a reasonable period of time.
1283 * But broken packet identifier may be better than no packet at all.
1285 static void ip_select_fb_ident(struct iphdr
*iph
)
1287 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1288 static u32 ip_fallback_id
;
1291 spin_lock_bh(&ip_fb_id_lock
);
1292 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1293 iph
->id
= htons(salt
& 0xFFFF);
1294 ip_fallback_id
= salt
;
1295 spin_unlock_bh(&ip_fb_id_lock
);
1298 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1300 struct rtable
*rt
= (struct rtable
*) dst
;
1303 if (rt
->peer
== NULL
)
1304 rt_bind_peer(rt
, 1);
1306 /* If peer is attached to destination, it is never detached,
1307 so that we need not to grab a lock to dereference it.
1310 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1314 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1315 __builtin_return_address(0));
1317 ip_select_fb_ident(iph
);
1320 static void rt_del(unsigned hash
, struct rtable
*rt
)
1322 struct rtable
**rthp
, *aux
;
1324 rthp
= &rt_hash_table
[hash
].chain
;
1325 spin_lock_bh(rt_hash_lock_addr(hash
));
1327 while ((aux
= *rthp
) != NULL
) {
1328 if (aux
== rt
|| rt_is_expired(aux
)) {
1329 *rthp
= aux
->u
.dst
.rt_next
;
1333 rthp
= &aux
->u
.dst
.rt_next
;
1335 spin_unlock_bh(rt_hash_lock_addr(hash
));
1338 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1339 __be32 saddr
, struct net_device
*dev
)
1342 struct in_device
*in_dev
= in_dev_get(dev
);
1343 struct rtable
*rth
, **rthp
;
1344 __be32 skeys
[2] = { saddr
, 0 };
1345 int ikeys
[2] = { dev
->ifindex
, 0 };
1346 struct netevent_redirect netevent
;
1353 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1354 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1355 ipv4_is_zeronet(new_gw
))
1356 goto reject_redirect
;
1358 if (!rt_caching(net
))
1359 goto reject_redirect
;
1361 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1362 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1363 goto reject_redirect
;
1364 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1365 goto reject_redirect
;
1367 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1368 goto reject_redirect
;
1371 for (i
= 0; i
< 2; i
++) {
1372 for (k
= 0; k
< 2; k
++) {
1373 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1376 rthp
=&rt_hash_table
[hash
].chain
;
1379 while ((rth
= rcu_dereference(*rthp
)) != NULL
) {
1382 if (rth
->fl
.fl4_dst
!= daddr
||
1383 rth
->fl
.fl4_src
!= skeys
[i
] ||
1384 rth
->fl
.oif
!= ikeys
[k
] ||
1386 rt_is_expired(rth
) ||
1387 !net_eq(dev_net(rth
->u
.dst
.dev
), net
)) {
1388 rthp
= &rth
->u
.dst
.rt_next
;
1392 if (rth
->rt_dst
!= daddr
||
1393 rth
->rt_src
!= saddr
||
1395 rth
->rt_gateway
!= old_gw
||
1396 rth
->u
.dst
.dev
!= dev
)
1399 dst_hold(&rth
->u
.dst
);
1402 rt
= dst_alloc(&ipv4_dst_ops
);
1409 /* Copy all the information. */
1411 rt
->u
.dst
.__use
= 1;
1412 atomic_set(&rt
->u
.dst
.__refcnt
, 1);
1413 rt
->u
.dst
.child
= NULL
;
1415 dev_hold(rt
->u
.dst
.dev
);
1417 in_dev_hold(rt
->idev
);
1418 rt
->u
.dst
.obsolete
= -1;
1419 rt
->u
.dst
.lastuse
= jiffies
;
1420 rt
->u
.dst
.path
= &rt
->u
.dst
;
1421 rt
->u
.dst
.neighbour
= NULL
;
1422 rt
->u
.dst
.hh
= NULL
;
1424 rt
->u
.dst
.xfrm
= NULL
;
1426 rt
->rt_genid
= rt_genid(net
);
1427 rt
->rt_flags
|= RTCF_REDIRECTED
;
1429 /* Gateway is different ... */
1430 rt
->rt_gateway
= new_gw
;
1432 /* Redirect received -> path was valid */
1433 dst_confirm(&rth
->u
.dst
);
1436 atomic_inc(&rt
->peer
->refcnt
);
1438 if (arp_bind_neighbour(&rt
->u
.dst
) ||
1439 !(rt
->u
.dst
.neighbour
->nud_state
&
1441 if (rt
->u
.dst
.neighbour
)
1442 neigh_event_send(rt
->u
.dst
.neighbour
, NULL
);
1448 netevent
.old
= &rth
->u
.dst
;
1449 netevent
.new = &rt
->u
.dst
;
1450 call_netevent_notifiers(NETEVENT_REDIRECT
,
1454 if (!rt_intern_hash(hash
, rt
, &rt
, NULL
))
1467 #ifdef CONFIG_IP_ROUTE_VERBOSE
1468 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1469 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1470 " Advised path = %pI4 -> %pI4\n",
1471 &old_gw
, dev
->name
, &new_gw
,
1477 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1479 struct rtable
*rt
= (struct rtable
*)dst
;
1480 struct dst_entry
*ret
= dst
;
1483 if (dst
->obsolete
> 0) {
1486 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
1487 (rt
->u
.dst
.expires
&&
1488 time_after_eq(jiffies
, rt
->u
.dst
.expires
))) {
1489 unsigned hash
= rt_hash(rt
->fl
.fl4_dst
, rt
->fl
.fl4_src
,
1491 rt_genid(dev_net(dst
->dev
)));
1492 #if RT_CACHE_DEBUG >= 1
1493 printk(KERN_DEBUG
"ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1494 &rt
->rt_dst
, rt
->fl
.fl4_tos
);
1505 * 1. The first ip_rt_redirect_number redirects are sent
1506 * with exponential backoff, then we stop sending them at all,
1507 * assuming that the host ignores our redirects.
1508 * 2. If we did not see packets requiring redirects
1509 * during ip_rt_redirect_silence, we assume that the host
1510 * forgot redirected route and start to send redirects again.
1512 * This algorithm is much cheaper and more intelligent than dumb load limiting
1515 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1516 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1519 void ip_rt_send_redirect(struct sk_buff
*skb
)
1521 struct rtable
*rt
= skb_rtable(skb
);
1522 struct in_device
*in_dev
;
1526 in_dev
= __in_dev_get_rcu(rt
->u
.dst
.dev
);
1527 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1531 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1534 /* No redirected packets during ip_rt_redirect_silence;
1535 * reset the algorithm.
1537 if (time_after(jiffies
, rt
->u
.dst
.rate_last
+ ip_rt_redirect_silence
))
1538 rt
->u
.dst
.rate_tokens
= 0;
1540 /* Too many ignored redirects; do not send anything
1541 * set u.dst.rate_last to the last seen redirected packet.
1543 if (rt
->u
.dst
.rate_tokens
>= ip_rt_redirect_number
) {
1544 rt
->u
.dst
.rate_last
= jiffies
;
1548 /* Check for load limit; set rate_last to the latest sent
1551 if (rt
->u
.dst
.rate_tokens
== 0 ||
1553 (rt
->u
.dst
.rate_last
+
1554 (ip_rt_redirect_load
<< rt
->u
.dst
.rate_tokens
)))) {
1555 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1556 rt
->u
.dst
.rate_last
= jiffies
;
1557 ++rt
->u
.dst
.rate_tokens
;
1558 #ifdef CONFIG_IP_ROUTE_VERBOSE
1560 rt
->u
.dst
.rate_tokens
== ip_rt_redirect_number
&&
1562 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1563 &rt
->rt_src
, rt
->rt_iif
,
1564 &rt
->rt_dst
, &rt
->rt_gateway
);
1569 static int ip_error(struct sk_buff
*skb
)
1571 struct rtable
*rt
= skb_rtable(skb
);
1575 switch (rt
->u
.dst
.error
) {
1580 code
= ICMP_HOST_UNREACH
;
1583 code
= ICMP_NET_UNREACH
;
1584 IP_INC_STATS_BH(dev_net(rt
->u
.dst
.dev
),
1585 IPSTATS_MIB_INNOROUTES
);
1588 code
= ICMP_PKT_FILTERED
;
1593 rt
->u
.dst
.rate_tokens
+= now
- rt
->u
.dst
.rate_last
;
1594 if (rt
->u
.dst
.rate_tokens
> ip_rt_error_burst
)
1595 rt
->u
.dst
.rate_tokens
= ip_rt_error_burst
;
1596 rt
->u
.dst
.rate_last
= now
;
1597 if (rt
->u
.dst
.rate_tokens
>= ip_rt_error_cost
) {
1598 rt
->u
.dst
.rate_tokens
-= ip_rt_error_cost
;
1599 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1602 out
: kfree_skb(skb
);
1607 * The last two values are not from the RFC but
1608 * are needed for AMPRnet AX.25 paths.
1611 static const unsigned short mtu_plateau
[] =
1612 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1614 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1618 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1619 if (old_mtu
> mtu_plateau
[i
])
1620 return mtu_plateau
[i
];
1624 unsigned short ip_rt_frag_needed(struct net
*net
, struct iphdr
*iph
,
1625 unsigned short new_mtu
,
1626 struct net_device
*dev
)
1629 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1631 int ikeys
[2] = { dev
->ifindex
, 0 };
1632 __be32 skeys
[2] = { iph
->saddr
, 0, };
1633 __be32 daddr
= iph
->daddr
;
1634 unsigned short est_mtu
= 0;
1636 for (k
= 0; k
< 2; k
++) {
1637 for (i
= 0; i
< 2; i
++) {
1638 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1642 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
1643 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
1644 unsigned short mtu
= new_mtu
;
1646 if (rth
->fl
.fl4_dst
!= daddr
||
1647 rth
->fl
.fl4_src
!= skeys
[i
] ||
1648 rth
->rt_dst
!= daddr
||
1649 rth
->rt_src
!= iph
->saddr
||
1650 rth
->fl
.oif
!= ikeys
[k
] ||
1652 dst_metric_locked(&rth
->u
.dst
, RTAX_MTU
) ||
1653 !net_eq(dev_net(rth
->u
.dst
.dev
), net
) ||
1657 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1659 /* BSD 4.2 compatibility hack :-( */
1661 old_mtu
>= dst_mtu(&rth
->u
.dst
) &&
1662 old_mtu
>= 68 + (iph
->ihl
<< 2))
1663 old_mtu
-= iph
->ihl
<< 2;
1665 mtu
= guess_mtu(old_mtu
);
1667 if (mtu
<= dst_mtu(&rth
->u
.dst
)) {
1668 if (mtu
< dst_mtu(&rth
->u
.dst
)) {
1669 dst_confirm(&rth
->u
.dst
);
1670 if (mtu
< ip_rt_min_pmtu
) {
1671 mtu
= ip_rt_min_pmtu
;
1672 rth
->u
.dst
.metrics
[RTAX_LOCK
-1] |=
1675 rth
->u
.dst
.metrics
[RTAX_MTU
-1] = mtu
;
1676 dst_set_expires(&rth
->u
.dst
,
1685 return est_mtu
? : new_mtu
;
1688 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1690 if (dst_mtu(dst
) > mtu
&& mtu
>= 68 &&
1691 !(dst_metric_locked(dst
, RTAX_MTU
))) {
1692 if (mtu
< ip_rt_min_pmtu
) {
1693 mtu
= ip_rt_min_pmtu
;
1694 dst
->metrics
[RTAX_LOCK
-1] |= (1 << RTAX_MTU
);
1696 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1697 dst_set_expires(dst
, ip_rt_mtu_expires
);
1698 call_netevent_notifiers(NETEVENT_PMTU_UPDATE
, dst
);
1702 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1704 if (rt_is_expired((struct rtable
*)dst
))
1709 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1711 struct rtable
*rt
= (struct rtable
*) dst
;
1712 struct inet_peer
*peer
= rt
->peer
;
1713 struct in_device
*idev
= rt
->idev
;
1726 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
1729 struct rtable
*rt
= (struct rtable
*) dst
;
1730 struct in_device
*idev
= rt
->idev
;
1731 if (dev
!= dev_net(dev
)->loopback_dev
&& idev
&& idev
->dev
== dev
) {
1732 struct in_device
*loopback_idev
=
1733 in_dev_get(dev_net(dev
)->loopback_dev
);
1734 if (loopback_idev
) {
1735 rt
->idev
= loopback_idev
;
1741 static void ipv4_link_failure(struct sk_buff
*skb
)
1745 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1747 rt
= skb_rtable(skb
);
1749 dst_set_expires(&rt
->u
.dst
, 0);
1752 static int ip_rt_bug(struct sk_buff
*skb
)
1754 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1755 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1756 skb
->dev
? skb
->dev
->name
: "?");
1762 We do not cache source address of outgoing interface,
1763 because it is used only by IP RR, TS and SRR options,
1764 so that it out of fast path.
1766 BTW remember: "addr" is allowed to be not aligned
1770 void ip_rt_get_source(u8
*addr
, struct rtable
*rt
)
1773 struct fib_result res
;
1775 if (rt
->fl
.iif
== 0)
1777 else if (fib_lookup(dev_net(rt
->u
.dst
.dev
), &rt
->fl
, &res
) == 0) {
1778 src
= FIB_RES_PREFSRC(res
);
1781 src
= inet_select_addr(rt
->u
.dst
.dev
, rt
->rt_gateway
,
1783 memcpy(addr
, &src
, 4);
1786 #ifdef CONFIG_NET_CLS_ROUTE
1787 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1789 if (!(rt
->u
.dst
.tclassid
& 0xFFFF))
1790 rt
->u
.dst
.tclassid
|= tag
& 0xFFFF;
1791 if (!(rt
->u
.dst
.tclassid
& 0xFFFF0000))
1792 rt
->u
.dst
.tclassid
|= tag
& 0xFFFF0000;
1796 static void rt_set_nexthop(struct rtable
*rt
, struct fib_result
*res
, u32 itag
)
1798 struct fib_info
*fi
= res
->fi
;
1801 if (FIB_RES_GW(*res
) &&
1802 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1803 rt
->rt_gateway
= FIB_RES_GW(*res
);
1804 memcpy(rt
->u
.dst
.metrics
, fi
->fib_metrics
,
1805 sizeof(rt
->u
.dst
.metrics
));
1806 if (fi
->fib_mtu
== 0) {
1807 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = rt
->u
.dst
.dev
->mtu
;
1808 if (dst_metric_locked(&rt
->u
.dst
, RTAX_MTU
) &&
1809 rt
->rt_gateway
!= rt
->rt_dst
&&
1810 rt
->u
.dst
.dev
->mtu
> 576)
1811 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = 576;
1813 #ifdef CONFIG_NET_CLS_ROUTE
1814 rt
->u
.dst
.tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1817 rt
->u
.dst
.metrics
[RTAX_MTU
-1]= rt
->u
.dst
.dev
->mtu
;
1819 if (dst_metric(&rt
->u
.dst
, RTAX_HOPLIMIT
) == 0)
1820 rt
->u
.dst
.metrics
[RTAX_HOPLIMIT
-1] = sysctl_ip_default_ttl
;
1821 if (dst_mtu(&rt
->u
.dst
) > IP_MAX_MTU
)
1822 rt
->u
.dst
.metrics
[RTAX_MTU
-1] = IP_MAX_MTU
;
1823 if (dst_metric(&rt
->u
.dst
, RTAX_ADVMSS
) == 0)
1824 rt
->u
.dst
.metrics
[RTAX_ADVMSS
-1] = max_t(unsigned int, rt
->u
.dst
.dev
->mtu
- 40,
1826 if (dst_metric(&rt
->u
.dst
, RTAX_ADVMSS
) > 65535 - 40)
1827 rt
->u
.dst
.metrics
[RTAX_ADVMSS
-1] = 65535 - 40;
1829 #ifdef CONFIG_NET_CLS_ROUTE
1830 #ifdef CONFIG_IP_MULTIPLE_TABLES
1831 set_class_tag(rt
, fib_rules_tclass(res
));
1833 set_class_tag(rt
, itag
);
1835 rt
->rt_type
= res
->type
;
1838 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1839 u8 tos
, struct net_device
*dev
, int our
)
1844 struct in_device
*in_dev
= in_dev_get(dev
);
1847 /* Primary sanity checks. */
1852 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1853 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
1856 if (ipv4_is_zeronet(saddr
)) {
1857 if (!ipv4_is_local_multicast(daddr
))
1859 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
1860 } else if (fib_validate_source(saddr
, 0, tos
, 0,
1861 dev
, &spec_dst
, &itag
, 0) < 0)
1864 rth
= dst_alloc(&ipv4_dst_ops
);
1868 rth
->u
.dst
.output
= ip_rt_bug
;
1869 rth
->u
.dst
.obsolete
= -1;
1871 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
1872 rth
->u
.dst
.flags
= DST_HOST
;
1873 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
1874 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
1875 rth
->fl
.fl4_dst
= daddr
;
1876 rth
->rt_dst
= daddr
;
1877 rth
->fl
.fl4_tos
= tos
;
1878 rth
->fl
.mark
= skb
->mark
;
1879 rth
->fl
.fl4_src
= saddr
;
1880 rth
->rt_src
= saddr
;
1881 #ifdef CONFIG_NET_CLS_ROUTE
1882 rth
->u
.dst
.tclassid
= itag
;
1885 rth
->fl
.iif
= dev
->ifindex
;
1886 rth
->u
.dst
.dev
= init_net
.loopback_dev
;
1887 dev_hold(rth
->u
.dst
.dev
);
1888 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
1890 rth
->rt_gateway
= daddr
;
1891 rth
->rt_spec_dst
= spec_dst
;
1892 rth
->rt_genid
= rt_genid(dev_net(dev
));
1893 rth
->rt_flags
= RTCF_MULTICAST
;
1894 rth
->rt_type
= RTN_MULTICAST
;
1896 rth
->u
.dst
.input
= ip_local_deliver
;
1897 rth
->rt_flags
|= RTCF_LOCAL
;
1900 #ifdef CONFIG_IP_MROUTE
1901 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1902 rth
->u
.dst
.input
= ip_mr_input
;
1904 RT_CACHE_STAT_INC(in_slow_mc
);
1907 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
1908 return rt_intern_hash(hash
, rth
, NULL
, skb
);
1920 static void ip_handle_martian_source(struct net_device
*dev
,
1921 struct in_device
*in_dev
,
1922 struct sk_buff
*skb
,
1926 RT_CACHE_STAT_INC(in_martian_src
);
1927 #ifdef CONFIG_IP_ROUTE_VERBOSE
1928 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1930 * RFC1812 recommendation, if source is martian,
1931 * the only hint is MAC header.
1933 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
1934 &daddr
, &saddr
, dev
->name
);
1935 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1937 const unsigned char *p
= skb_mac_header(skb
);
1938 printk(KERN_WARNING
"ll header: ");
1939 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
1941 if (i
< (dev
->hard_header_len
- 1))
1950 static int __mkroute_input(struct sk_buff
*skb
,
1951 struct fib_result
*res
,
1952 struct in_device
*in_dev
,
1953 __be32 daddr
, __be32 saddr
, u32 tos
,
1954 struct rtable
**result
)
1959 struct in_device
*out_dev
;
1964 /* get a working reference to the output device */
1965 out_dev
= in_dev_get(FIB_RES_DEV(*res
));
1966 if (out_dev
== NULL
) {
1967 if (net_ratelimit())
1968 printk(KERN_CRIT
"Bug in ip_route_input" \
1969 "_slow(). Please, report\n");
1974 err
= fib_validate_source(saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1975 in_dev
->dev
, &spec_dst
, &itag
, skb
->mark
);
1977 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1985 flags
|= RTCF_DIRECTSRC
;
1987 if (out_dev
== in_dev
&& err
&&
1988 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1989 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
1990 flags
|= RTCF_DOREDIRECT
;
1992 if (skb
->protocol
!= htons(ETH_P_IP
)) {
1993 /* Not IP (i.e. ARP). Do not create route, if it is
1994 * invalid for proxy arp. DNAT routes are always valid.
1996 if (out_dev
== in_dev
) {
2003 rth
= dst_alloc(&ipv4_dst_ops
);
2009 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2010 rth
->u
.dst
.flags
= DST_HOST
;
2011 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2012 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2013 if (IN_DEV_CONF_GET(out_dev
, NOXFRM
))
2014 rth
->u
.dst
.flags
|= DST_NOXFRM
;
2015 rth
->fl
.fl4_dst
= daddr
;
2016 rth
->rt_dst
= daddr
;
2017 rth
->fl
.fl4_tos
= tos
;
2018 rth
->fl
.mark
= skb
->mark
;
2019 rth
->fl
.fl4_src
= saddr
;
2020 rth
->rt_src
= saddr
;
2021 rth
->rt_gateway
= daddr
;
2023 rth
->fl
.iif
= in_dev
->dev
->ifindex
;
2024 rth
->u
.dst
.dev
= (out_dev
)->dev
;
2025 dev_hold(rth
->u
.dst
.dev
);
2026 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
2028 rth
->rt_spec_dst
= spec_dst
;
2030 rth
->u
.dst
.obsolete
= -1;
2031 rth
->u
.dst
.input
= ip_forward
;
2032 rth
->u
.dst
.output
= ip_output
;
2033 rth
->rt_genid
= rt_genid(dev_net(rth
->u
.dst
.dev
));
2035 rt_set_nexthop(rth
, res
, itag
);
2037 rth
->rt_flags
= flags
;
2042 /* release the working reference to the output device */
2043 in_dev_put(out_dev
);
2047 static int ip_mkroute_input(struct sk_buff
*skb
,
2048 struct fib_result
*res
,
2049 const struct flowi
*fl
,
2050 struct in_device
*in_dev
,
2051 __be32 daddr
, __be32 saddr
, u32 tos
)
2053 struct rtable
* rth
= NULL
;
2057 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2058 if (res
->fi
&& res
->fi
->fib_nhs
> 1 && fl
->oif
== 0)
2059 fib_select_multipath(fl
, res
);
2062 /* create a routing cache entry */
2063 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2067 /* put it into the cache */
2068 hash
= rt_hash(daddr
, saddr
, fl
->iif
,
2069 rt_genid(dev_net(rth
->u
.dst
.dev
)));
2070 return rt_intern_hash(hash
, rth
, NULL
, skb
);
2074 * NOTE. We drop all the packets that has local source
2075 * addresses, because every properly looped back packet
2076 * must have correct destination already attached by output routine.
2078 * Such approach solves two big problems:
2079 * 1. Not simplex devices are handled properly.
2080 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2083 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2084 u8 tos
, struct net_device
*dev
)
2086 struct fib_result res
;
2087 struct in_device
*in_dev
= in_dev_get(dev
);
2088 struct flowi fl
= { .nl_u
= { .ip4_u
=
2092 .scope
= RT_SCOPE_UNIVERSE
,
2095 .iif
= dev
->ifindex
};
2098 struct rtable
* rth
;
2103 struct net
* net
= dev_net(dev
);
2105 /* IP on this device is disabled. */
2110 /* Check for the most weird martians, which can be not detected
2114 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2115 ipv4_is_loopback(saddr
))
2116 goto martian_source
;
2118 if (daddr
== htonl(0xFFFFFFFF) || (saddr
== 0 && daddr
== 0))
2121 /* Accept zero addresses only to limited broadcast;
2122 * I even do not know to fix it or not. Waiting for complains :-)
2124 if (ipv4_is_zeronet(saddr
))
2125 goto martian_source
;
2127 if (ipv4_is_lbcast(daddr
) || ipv4_is_zeronet(daddr
) ||
2128 ipv4_is_loopback(daddr
))
2129 goto martian_destination
;
2132 * Now we are ready to route packet.
2134 if ((err
= fib_lookup(net
, &fl
, &res
)) != 0) {
2135 if (!IN_DEV_FORWARD(in_dev
))
2141 RT_CACHE_STAT_INC(in_slow_tot
);
2143 if (res
.type
== RTN_BROADCAST
)
2146 if (res
.type
== RTN_LOCAL
) {
2148 result
= fib_validate_source(saddr
, daddr
, tos
,
2149 net
->loopback_dev
->ifindex
,
2150 dev
, &spec_dst
, &itag
, skb
->mark
);
2152 goto martian_source
;
2154 flags
|= RTCF_DIRECTSRC
;
2159 if (!IN_DEV_FORWARD(in_dev
))
2161 if (res
.type
!= RTN_UNICAST
)
2162 goto martian_destination
;
2164 err
= ip_mkroute_input(skb
, &res
, &fl
, in_dev
, daddr
, saddr
, tos
);
2172 if (skb
->protocol
!= htons(ETH_P_IP
))
2175 if (ipv4_is_zeronet(saddr
))
2176 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2178 err
= fib_validate_source(saddr
, 0, tos
, 0, dev
, &spec_dst
,
2181 goto martian_source
;
2183 flags
|= RTCF_DIRECTSRC
;
2185 flags
|= RTCF_BROADCAST
;
2186 res
.type
= RTN_BROADCAST
;
2187 RT_CACHE_STAT_INC(in_brd
);
2190 rth
= dst_alloc(&ipv4_dst_ops
);
2194 rth
->u
.dst
.output
= ip_rt_bug
;
2195 rth
->u
.dst
.obsolete
= -1;
2196 rth
->rt_genid
= rt_genid(net
);
2198 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2199 rth
->u
.dst
.flags
= DST_HOST
;
2200 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2201 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2202 rth
->fl
.fl4_dst
= daddr
;
2203 rth
->rt_dst
= daddr
;
2204 rth
->fl
.fl4_tos
= tos
;
2205 rth
->fl
.mark
= skb
->mark
;
2206 rth
->fl
.fl4_src
= saddr
;
2207 rth
->rt_src
= saddr
;
2208 #ifdef CONFIG_NET_CLS_ROUTE
2209 rth
->u
.dst
.tclassid
= itag
;
2212 rth
->fl
.iif
= dev
->ifindex
;
2213 rth
->u
.dst
.dev
= net
->loopback_dev
;
2214 dev_hold(rth
->u
.dst
.dev
);
2215 rth
->idev
= in_dev_get(rth
->u
.dst
.dev
);
2216 rth
->rt_gateway
= daddr
;
2217 rth
->rt_spec_dst
= spec_dst
;
2218 rth
->u
.dst
.input
= ip_local_deliver
;
2219 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2220 if (res
.type
== RTN_UNREACHABLE
) {
2221 rth
->u
.dst
.input
= ip_error
;
2222 rth
->u
.dst
.error
= -err
;
2223 rth
->rt_flags
&= ~RTCF_LOCAL
;
2225 rth
->rt_type
= res
.type
;
2226 hash
= rt_hash(daddr
, saddr
, fl
.iif
, rt_genid(net
));
2227 err
= rt_intern_hash(hash
, rth
, NULL
, skb
);
2231 RT_CACHE_STAT_INC(in_no_route
);
2232 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2233 res
.type
= RTN_UNREACHABLE
;
2239 * Do not cache martian addresses: they should be logged (RFC1812)
2241 martian_destination
:
2242 RT_CACHE_STAT_INC(in_martian_dst
);
2243 #ifdef CONFIG_IP_ROUTE_VERBOSE
2244 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2245 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2246 &daddr
, &saddr
, dev
->name
);
2250 err
= -EHOSTUNREACH
;
2262 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2266 int ip_route_input(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2267 u8 tos
, struct net_device
*dev
)
2269 struct rtable
* rth
;
2271 int iif
= dev
->ifindex
;
2276 if (!rt_caching(net
))
2279 tos
&= IPTOS_RT_MASK
;
2280 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2283 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2284 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
2285 if (((rth
->fl
.fl4_dst
^ daddr
) |
2286 (rth
->fl
.fl4_src
^ saddr
) |
2287 (rth
->fl
.iif
^ iif
) |
2289 (rth
->fl
.fl4_tos
^ tos
)) == 0 &&
2290 rth
->fl
.mark
== skb
->mark
&&
2291 net_eq(dev_net(rth
->u
.dst
.dev
), net
) &&
2292 !rt_is_expired(rth
)) {
2293 dst_use(&rth
->u
.dst
, jiffies
);
2294 RT_CACHE_STAT_INC(in_hit
);
2296 skb_dst_set(skb
, &rth
->u
.dst
);
2299 RT_CACHE_STAT_INC(in_hlist_search
);
2304 /* Multicast recognition logic is moved from route cache to here.
2305 The problem was that too many Ethernet cards have broken/missing
2306 hardware multicast filters :-( As result the host on multicasting
2307 network acquires a lot of useless route cache entries, sort of
2308 SDR messages from all the world. Now we try to get rid of them.
2309 Really, provided software IP multicast filter is organized
2310 reasonably (at least, hashed), it does not result in a slowdown
2311 comparing with route cache reject entries.
2312 Note, that multicast routers are not affected, because
2313 route cache entry is created eventually.
2315 if (ipv4_is_multicast(daddr
)) {
2316 struct in_device
*in_dev
;
2319 if ((in_dev
= __in_dev_get_rcu(dev
)) != NULL
) {
2320 int our
= ip_check_mc(in_dev
, daddr
, saddr
,
2321 ip_hdr(skb
)->protocol
);
2323 #ifdef CONFIG_IP_MROUTE
2325 (!ipv4_is_local_multicast(daddr
) &&
2326 IN_DEV_MFORWARD(in_dev
))
2330 return ip_route_input_mc(skb
, daddr
, saddr
,
2337 return ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2340 static int __mkroute_output(struct rtable
**result
,
2341 struct fib_result
*res
,
2342 const struct flowi
*fl
,
2343 const struct flowi
*oldflp
,
2344 struct net_device
*dev_out
,
2348 struct in_device
*in_dev
;
2349 u32 tos
= RT_FL_TOS(oldflp
);
2352 if (ipv4_is_loopback(fl
->fl4_src
) && !(dev_out
->flags
&IFF_LOOPBACK
))
2355 if (fl
->fl4_dst
== htonl(0xFFFFFFFF))
2356 res
->type
= RTN_BROADCAST
;
2357 else if (ipv4_is_multicast(fl
->fl4_dst
))
2358 res
->type
= RTN_MULTICAST
;
2359 else if (ipv4_is_lbcast(fl
->fl4_dst
) || ipv4_is_zeronet(fl
->fl4_dst
))
2362 if (dev_out
->flags
& IFF_LOOPBACK
)
2363 flags
|= RTCF_LOCAL
;
2365 /* get work reference to inet device */
2366 in_dev
= in_dev_get(dev_out
);
2370 if (res
->type
== RTN_BROADCAST
) {
2371 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2373 fib_info_put(res
->fi
);
2376 } else if (res
->type
== RTN_MULTICAST
) {
2377 flags
|= RTCF_MULTICAST
|RTCF_LOCAL
;
2378 if (!ip_check_mc(in_dev
, oldflp
->fl4_dst
, oldflp
->fl4_src
,
2380 flags
&= ~RTCF_LOCAL
;
2381 /* If multicast route do not exist use
2382 default one, but do not gateway in this case.
2385 if (res
->fi
&& res
->prefixlen
< 4) {
2386 fib_info_put(res
->fi
);
2392 rth
= dst_alloc(&ipv4_dst_ops
);
2398 atomic_set(&rth
->u
.dst
.__refcnt
, 1);
2399 rth
->u
.dst
.flags
= DST_HOST
;
2400 if (IN_DEV_CONF_GET(in_dev
, NOXFRM
))
2401 rth
->u
.dst
.flags
|= DST_NOXFRM
;
2402 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2403 rth
->u
.dst
.flags
|= DST_NOPOLICY
;
2405 rth
->fl
.fl4_dst
= oldflp
->fl4_dst
;
2406 rth
->fl
.fl4_tos
= tos
;
2407 rth
->fl
.fl4_src
= oldflp
->fl4_src
;
2408 rth
->fl
.oif
= oldflp
->oif
;
2409 rth
->fl
.mark
= oldflp
->mark
;
2410 rth
->rt_dst
= fl
->fl4_dst
;
2411 rth
->rt_src
= fl
->fl4_src
;
2412 rth
->rt_iif
= oldflp
->oif
? : dev_out
->ifindex
;
2413 /* get references to the devices that are to be hold by the routing
2415 rth
->u
.dst
.dev
= dev_out
;
2417 rth
->idev
= in_dev_get(dev_out
);
2418 rth
->rt_gateway
= fl
->fl4_dst
;
2419 rth
->rt_spec_dst
= fl
->fl4_src
;
2421 rth
->u
.dst
.output
=ip_output
;
2422 rth
->u
.dst
.obsolete
= -1;
2423 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2425 RT_CACHE_STAT_INC(out_slow_tot
);
2427 if (flags
& RTCF_LOCAL
) {
2428 rth
->u
.dst
.input
= ip_local_deliver
;
2429 rth
->rt_spec_dst
= fl
->fl4_dst
;
2431 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2432 rth
->rt_spec_dst
= fl
->fl4_src
;
2433 if (flags
& RTCF_LOCAL
&&
2434 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2435 rth
->u
.dst
.output
= ip_mc_output
;
2436 RT_CACHE_STAT_INC(out_slow_mc
);
2438 #ifdef CONFIG_IP_MROUTE
2439 if (res
->type
== RTN_MULTICAST
) {
2440 if (IN_DEV_MFORWARD(in_dev
) &&
2441 !ipv4_is_local_multicast(oldflp
->fl4_dst
)) {
2442 rth
->u
.dst
.input
= ip_mr_input
;
2443 rth
->u
.dst
.output
= ip_mc_output
;
2449 rt_set_nexthop(rth
, res
, 0);
2451 rth
->rt_flags
= flags
;
2455 /* release work reference to inet device */
2461 static int ip_mkroute_output(struct rtable
**rp
,
2462 struct fib_result
*res
,
2463 const struct flowi
*fl
,
2464 const struct flowi
*oldflp
,
2465 struct net_device
*dev_out
,
2468 struct rtable
*rth
= NULL
;
2469 int err
= __mkroute_output(&rth
, res
, fl
, oldflp
, dev_out
, flags
);
2472 hash
= rt_hash(oldflp
->fl4_dst
, oldflp
->fl4_src
, oldflp
->oif
,
2473 rt_genid(dev_net(dev_out
)));
2474 err
= rt_intern_hash(hash
, rth
, rp
, NULL
);
2481 * Major route resolver routine.
2484 static int ip_route_output_slow(struct net
*net
, struct rtable
**rp
,
2485 const struct flowi
*oldflp
)
2487 u32 tos
= RT_FL_TOS(oldflp
);
2488 struct flowi fl
= { .nl_u
= { .ip4_u
=
2489 { .daddr
= oldflp
->fl4_dst
,
2490 .saddr
= oldflp
->fl4_src
,
2491 .tos
= tos
& IPTOS_RT_MASK
,
2492 .scope
= ((tos
& RTO_ONLINK
) ?
2496 .mark
= oldflp
->mark
,
2497 .iif
= net
->loopback_dev
->ifindex
,
2498 .oif
= oldflp
->oif
};
2499 struct fib_result res
;
2501 struct net_device
*dev_out
= NULL
;
2507 #ifdef CONFIG_IP_MULTIPLE_TABLES
2511 if (oldflp
->fl4_src
) {
2513 if (ipv4_is_multicast(oldflp
->fl4_src
) ||
2514 ipv4_is_lbcast(oldflp
->fl4_src
) ||
2515 ipv4_is_zeronet(oldflp
->fl4_src
))
2518 /* I removed check for oif == dev_out->oif here.
2519 It was wrong for two reasons:
2520 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2521 is assigned to multiple interfaces.
2522 2. Moreover, we are allowed to send packets with saddr
2523 of another iface. --ANK
2526 if (oldflp
->oif
== 0 &&
2527 (ipv4_is_multicast(oldflp
->fl4_dst
) ||
2528 oldflp
->fl4_dst
== htonl(0xFFFFFFFF))) {
2529 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2530 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2531 if (dev_out
== NULL
)
2534 /* Special hack: user can direct multicasts
2535 and limited broadcast via necessary interface
2536 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2537 This hack is not just for fun, it allows
2538 vic,vat and friends to work.
2539 They bind socket to loopback, set ttl to zero
2540 and expect that it will work.
2541 From the viewpoint of routing cache they are broken,
2542 because we are not allowed to build multicast path
2543 with loopback source addr (look, routing cache
2544 cannot know, that ttl is zero, so that packet
2545 will not leave this host and route is valid).
2546 Luckily, this hack is good workaround.
2549 fl
.oif
= dev_out
->ifindex
;
2553 if (!(oldflp
->flags
& FLOWI_FLAG_ANYSRC
)) {
2554 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2555 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2556 if (dev_out
== NULL
)
2565 dev_out
= dev_get_by_index(net
, oldflp
->oif
);
2567 if (dev_out
== NULL
)
2570 /* RACE: Check return value of inet_select_addr instead. */
2571 if (__in_dev_get_rtnl(dev_out
) == NULL
) {
2573 goto out
; /* Wrong error code */
2576 if (ipv4_is_local_multicast(oldflp
->fl4_dst
) ||
2577 oldflp
->fl4_dst
== htonl(0xFFFFFFFF)) {
2579 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2584 if (ipv4_is_multicast(oldflp
->fl4_dst
))
2585 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2587 else if (!oldflp
->fl4_dst
)
2588 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2594 fl
.fl4_dst
= fl
.fl4_src
;
2596 fl
.fl4_dst
= fl
.fl4_src
= htonl(INADDR_LOOPBACK
);
2599 dev_out
= net
->loopback_dev
;
2601 fl
.oif
= net
->loopback_dev
->ifindex
;
2602 res
.type
= RTN_LOCAL
;
2603 flags
|= RTCF_LOCAL
;
2607 if (fib_lookup(net
, &fl
, &res
)) {
2610 /* Apparently, routing tables are wrong. Assume,
2611 that the destination is on link.
2614 Because we are allowed to send to iface
2615 even if it has NO routes and NO assigned
2616 addresses. When oif is specified, routing
2617 tables are looked up with only one purpose:
2618 to catch if destination is gatewayed, rather than
2619 direct. Moreover, if MSG_DONTROUTE is set,
2620 we send packet, ignoring both routing tables
2621 and ifaddr state. --ANK
2624 We could make it even if oif is unknown,
2625 likely IPv6, but we do not.
2628 if (fl
.fl4_src
== 0)
2629 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2631 res
.type
= RTN_UNICAST
;
2641 if (res
.type
== RTN_LOCAL
) {
2643 fl
.fl4_src
= fl
.fl4_dst
;
2646 dev_out
= net
->loopback_dev
;
2648 fl
.oif
= dev_out
->ifindex
;
2650 fib_info_put(res
.fi
);
2652 flags
|= RTCF_LOCAL
;
2656 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2657 if (res
.fi
->fib_nhs
> 1 && fl
.oif
== 0)
2658 fib_select_multipath(&fl
, &res
);
2661 if (!res
.prefixlen
&& res
.type
== RTN_UNICAST
&& !fl
.oif
)
2662 fib_select_default(net
, &fl
, &res
);
2665 fl
.fl4_src
= FIB_RES_PREFSRC(res
);
2669 dev_out
= FIB_RES_DEV(res
);
2671 fl
.oif
= dev_out
->ifindex
;
2675 err
= ip_mkroute_output(rp
, &res
, &fl
, oldflp
, dev_out
, flags
);
2685 int __ip_route_output_key(struct net
*net
, struct rtable
**rp
,
2686 const struct flowi
*flp
)
2691 if (!rt_caching(net
))
2694 hash
= rt_hash(flp
->fl4_dst
, flp
->fl4_src
, flp
->oif
, rt_genid(net
));
2697 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2698 rth
= rcu_dereference(rth
->u
.dst
.rt_next
)) {
2699 if (rth
->fl
.fl4_dst
== flp
->fl4_dst
&&
2700 rth
->fl
.fl4_src
== flp
->fl4_src
&&
2702 rth
->fl
.oif
== flp
->oif
&&
2703 rth
->fl
.mark
== flp
->mark
&&
2704 !((rth
->fl
.fl4_tos
^ flp
->fl4_tos
) &
2705 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2706 net_eq(dev_net(rth
->u
.dst
.dev
), net
) &&
2707 !rt_is_expired(rth
)) {
2708 dst_use(&rth
->u
.dst
, jiffies
);
2709 RT_CACHE_STAT_INC(out_hit
);
2710 rcu_read_unlock_bh();
2714 RT_CACHE_STAT_INC(out_hlist_search
);
2716 rcu_read_unlock_bh();
2719 return ip_route_output_slow(net
, rp
, flp
);
2722 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2724 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2729 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2733 static struct dst_ops ipv4_dst_blackhole_ops
= {
2735 .protocol
= cpu_to_be16(ETH_P_IP
),
2736 .destroy
= ipv4_dst_destroy
,
2737 .check
= ipv4_blackhole_dst_check
,
2738 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2739 .entries
= ATOMIC_INIT(0),
2743 static int ipv4_dst_blackhole(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2745 struct rtable
*ort
= *rp
;
2746 struct rtable
*rt
= (struct rtable
*)
2747 dst_alloc(&ipv4_dst_blackhole_ops
);
2750 struct dst_entry
*new = &rt
->u
.dst
;
2752 atomic_set(&new->__refcnt
, 1);
2754 new->input
= dst_discard
;
2755 new->output
= dst_discard
;
2756 memcpy(new->metrics
, ort
->u
.dst
.metrics
, RTAX_MAX
*sizeof(u32
));
2758 new->dev
= ort
->u
.dst
.dev
;
2764 rt
->idev
= ort
->idev
;
2766 in_dev_hold(rt
->idev
);
2767 rt
->rt_genid
= rt_genid(net
);
2768 rt
->rt_flags
= ort
->rt_flags
;
2769 rt
->rt_type
= ort
->rt_type
;
2770 rt
->rt_dst
= ort
->rt_dst
;
2771 rt
->rt_src
= ort
->rt_src
;
2772 rt
->rt_iif
= ort
->rt_iif
;
2773 rt
->rt_gateway
= ort
->rt_gateway
;
2774 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2775 rt
->peer
= ort
->peer
;
2777 atomic_inc(&rt
->peer
->refcnt
);
2782 dst_release(&(*rp
)->u
.dst
);
2784 return (rt
? 0 : -ENOMEM
);
2787 int ip_route_output_flow(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
,
2788 struct sock
*sk
, int flags
)
2792 if ((err
= __ip_route_output_key(net
, rp
, flp
)) != 0)
2797 flp
->fl4_src
= (*rp
)->rt_src
;
2799 flp
->fl4_dst
= (*rp
)->rt_dst
;
2800 err
= __xfrm_lookup(net
, (struct dst_entry
**)rp
, flp
, sk
,
2801 flags
? XFRM_LOOKUP_WAIT
: 0);
2802 if (err
== -EREMOTE
)
2803 err
= ipv4_dst_blackhole(net
, rp
, flp
);
2811 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2813 int ip_route_output_key(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2815 return ip_route_output_flow(net
, rp
, flp
, NULL
, 0);
2818 static int rt_fill_info(struct net
*net
,
2819 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2820 int nowait
, unsigned int flags
)
2822 struct rtable
*rt
= skb_rtable(skb
);
2824 struct nlmsghdr
*nlh
;
2826 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2828 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2832 r
= nlmsg_data(nlh
);
2833 r
->rtm_family
= AF_INET
;
2834 r
->rtm_dst_len
= 32;
2836 r
->rtm_tos
= rt
->fl
.fl4_tos
;
2837 r
->rtm_table
= RT_TABLE_MAIN
;
2838 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2839 r
->rtm_type
= rt
->rt_type
;
2840 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2841 r
->rtm_protocol
= RTPROT_UNSPEC
;
2842 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2843 if (rt
->rt_flags
& RTCF_NOTIFY
)
2844 r
->rtm_flags
|= RTM_F_NOTIFY
;
2846 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2848 if (rt
->fl
.fl4_src
) {
2849 r
->rtm_src_len
= 32;
2850 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->fl
.fl4_src
);
2853 NLA_PUT_U32(skb
, RTA_OIF
, rt
->u
.dst
.dev
->ifindex
);
2854 #ifdef CONFIG_NET_CLS_ROUTE
2855 if (rt
->u
.dst
.tclassid
)
2856 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->u
.dst
.tclassid
);
2859 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
2860 else if (rt
->rt_src
!= rt
->fl
.fl4_src
)
2861 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
2863 if (rt
->rt_dst
!= rt
->rt_gateway
)
2864 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
2866 if (rtnetlink_put_metrics(skb
, rt
->u
.dst
.metrics
) < 0)
2867 goto nla_put_failure
;
2869 error
= rt
->u
.dst
.error
;
2870 expires
= rt
->u
.dst
.expires
? rt
->u
.dst
.expires
- jiffies
: 0;
2872 id
= atomic_read(&rt
->peer
->ip_id_count
) & 0xffff;
2873 if (rt
->peer
->tcp_ts_stamp
) {
2874 ts
= rt
->peer
->tcp_ts
;
2875 tsage
= get_seconds() - rt
->peer
->tcp_ts_stamp
;
2880 #ifdef CONFIG_IP_MROUTE
2881 __be32 dst
= rt
->rt_dst
;
2883 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2884 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2885 int err
= ipmr_get_route(net
, skb
, r
, nowait
);
2890 goto nla_put_failure
;
2892 if (err
== -EMSGSIZE
)
2893 goto nla_put_failure
;
2899 NLA_PUT_U32(skb
, RTA_IIF
, rt
->fl
.iif
);
2902 if (rtnl_put_cacheinfo(skb
, &rt
->u
.dst
, id
, ts
, tsage
,
2903 expires
, error
) < 0)
2904 goto nla_put_failure
;
2906 return nlmsg_end(skb
, nlh
);
2909 nlmsg_cancel(skb
, nlh
);
2913 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
2915 struct net
*net
= sock_net(in_skb
->sk
);
2917 struct nlattr
*tb
[RTA_MAX
+1];
2918 struct rtable
*rt
= NULL
;
2923 struct sk_buff
*skb
;
2925 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2929 rtm
= nlmsg_data(nlh
);
2931 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2937 /* Reserve room for dummy headers, this skb can pass
2938 through good chunk of routing engine.
2940 skb_reset_mac_header(skb
);
2941 skb_reset_network_header(skb
);
2943 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2944 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
2945 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2947 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
2948 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
2949 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2952 struct net_device
*dev
;
2954 dev
= __dev_get_by_index(net
, iif
);
2960 skb
->protocol
= htons(ETH_P_IP
);
2963 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
2966 rt
= skb_rtable(skb
);
2967 if (err
== 0 && rt
->u
.dst
.error
)
2968 err
= -rt
->u
.dst
.error
;
2975 .tos
= rtm
->rtm_tos
,
2978 .oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
2980 err
= ip_route_output_key(net
, &rt
, &fl
);
2986 skb_dst_set(skb
, &rt
->u
.dst
);
2987 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2988 rt
->rt_flags
|= RTCF_NOTIFY
;
2990 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
2991 RTM_NEWROUTE
, 0, 0);
2995 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
3004 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3011 net
= sock_net(skb
->sk
);
3016 s_idx
= idx
= cb
->args
[1];
3017 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3018 if (!rt_hash_table
[h
].chain
)
3021 for (rt
= rcu_dereference(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3022 rt
= rcu_dereference(rt
->u
.dst
.rt_next
), idx
++) {
3023 if (!net_eq(dev_net(rt
->u
.dst
.dev
), net
) || idx
< s_idx
)
3025 if (rt_is_expired(rt
))
3027 skb_dst_set(skb
, dst_clone(&rt
->u
.dst
));
3028 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3029 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3030 1, NLM_F_MULTI
) <= 0) {
3032 rcu_read_unlock_bh();
3037 rcu_read_unlock_bh();
3046 void ip_rt_multicast_event(struct in_device
*in_dev
)
3048 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3051 #ifdef CONFIG_SYSCTL
3052 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3053 void __user
*buffer
,
3054 size_t *lenp
, loff_t
*ppos
)
3061 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3062 ctl
.data
= &flush_delay
;
3063 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3065 net
= (struct net
*)__ctl
->extra1
;
3066 rt_cache_flush(net
, flush_delay
);
3073 static void rt_secret_reschedule(int old
)
3076 int new = ip_rt_secret_interval
;
3077 int diff
= new - old
;
3084 int deleted
= del_timer_sync(&net
->ipv4
.rt_secret_timer
);
3091 time
= net
->ipv4
.rt_secret_timer
.expires
- jiffies
;
3093 if (time
<= 0 || (time
+= diff
) <= 0)
3098 mod_timer(&net
->ipv4
.rt_secret_timer
, jiffies
+ time
);
3103 static int ipv4_sysctl_rt_secret_interval(ctl_table
*ctl
, int write
,
3104 void __user
*buffer
, size_t *lenp
,
3107 int old
= ip_rt_secret_interval
;
3108 int ret
= proc_dointvec_jiffies(ctl
, write
, buffer
, lenp
, ppos
);
3110 rt_secret_reschedule(old
);
3115 static ctl_table ipv4_route_table
[] = {
3117 .procname
= "gc_thresh",
3118 .data
= &ipv4_dst_ops
.gc_thresh
,
3119 .maxlen
= sizeof(int),
3121 .proc_handler
= proc_dointvec
,
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 .procname
= "gc_min_interval",
3134 .data
= &ip_rt_gc_min_interval
,
3135 .maxlen
= sizeof(int),
3137 .proc_handler
= proc_dointvec_jiffies
,
3140 .procname
= "gc_min_interval_ms",
3141 .data
= &ip_rt_gc_min_interval
,
3142 .maxlen
= sizeof(int),
3144 .proc_handler
= proc_dointvec_ms_jiffies
,
3147 .procname
= "gc_timeout",
3148 .data
= &ip_rt_gc_timeout
,
3149 .maxlen
= sizeof(int),
3151 .proc_handler
= proc_dointvec_jiffies
,
3154 .procname
= "gc_interval",
3155 .data
= &ip_rt_gc_interval
,
3156 .maxlen
= sizeof(int),
3158 .proc_handler
= proc_dointvec_jiffies
,
3161 .procname
= "redirect_load",
3162 .data
= &ip_rt_redirect_load
,
3163 .maxlen
= sizeof(int),
3165 .proc_handler
= proc_dointvec
,
3168 .procname
= "redirect_number",
3169 .data
= &ip_rt_redirect_number
,
3170 .maxlen
= sizeof(int),
3172 .proc_handler
= proc_dointvec
,
3175 .procname
= "redirect_silence",
3176 .data
= &ip_rt_redirect_silence
,
3177 .maxlen
= sizeof(int),
3179 .proc_handler
= proc_dointvec
,
3182 .procname
= "error_cost",
3183 .data
= &ip_rt_error_cost
,
3184 .maxlen
= sizeof(int),
3186 .proc_handler
= proc_dointvec
,
3189 .procname
= "error_burst",
3190 .data
= &ip_rt_error_burst
,
3191 .maxlen
= sizeof(int),
3193 .proc_handler
= proc_dointvec
,
3196 .procname
= "gc_elasticity",
3197 .data
= &ip_rt_gc_elasticity
,
3198 .maxlen
= sizeof(int),
3200 .proc_handler
= proc_dointvec
,
3203 .procname
= "mtu_expires",
3204 .data
= &ip_rt_mtu_expires
,
3205 .maxlen
= sizeof(int),
3207 .proc_handler
= proc_dointvec_jiffies
,
3210 .procname
= "min_pmtu",
3211 .data
= &ip_rt_min_pmtu
,
3212 .maxlen
= sizeof(int),
3214 .proc_handler
= proc_dointvec
,
3217 .procname
= "min_adv_mss",
3218 .data
= &ip_rt_min_advmss
,
3219 .maxlen
= sizeof(int),
3221 .proc_handler
= proc_dointvec
,
3224 .procname
= "secret_interval",
3225 .data
= &ip_rt_secret_interval
,
3226 .maxlen
= sizeof(int),
3228 .proc_handler
= ipv4_sysctl_rt_secret_interval
,
3233 static struct ctl_table empty
[1];
3235 static struct ctl_table ipv4_skeleton
[] =
3237 { .procname
= "route",
3238 .mode
= 0555, .child
= ipv4_route_table
},
3239 { .procname
= "neigh",
3240 .mode
= 0555, .child
= empty
},
3244 static __net_initdata
struct ctl_path ipv4_path
[] = {
3245 { .procname
= "net", },
3246 { .procname
= "ipv4", },
3250 static struct ctl_table ipv4_route_flush_table
[] = {
3252 .procname
= "flush",
3253 .maxlen
= sizeof(int),
3255 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3260 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3261 { .procname
= "net", },
3262 { .procname
= "ipv4", },
3263 { .procname
= "route", },
3267 static __net_init
int sysctl_route_net_init(struct net
*net
)
3269 struct ctl_table
*tbl
;
3271 tbl
= ipv4_route_flush_table
;
3272 if (!net_eq(net
, &init_net
)) {
3273 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3277 tbl
[0].extra1
= net
;
3279 net
->ipv4
.route_hdr
=
3280 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3281 if (net
->ipv4
.route_hdr
== NULL
)
3286 if (tbl
!= ipv4_route_flush_table
)
3292 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3294 struct ctl_table
*tbl
;
3296 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3297 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3298 BUG_ON(tbl
== ipv4_route_flush_table
);
3302 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3303 .init
= sysctl_route_net_init
,
3304 .exit
= sysctl_route_net_exit
,
3309 static __net_init
int rt_secret_timer_init(struct net
*net
)
3311 atomic_set(&net
->ipv4
.rt_genid
,
3312 (int) ((num_physpages
^ (num_physpages
>>8)) ^
3313 (jiffies
^ (jiffies
>> 7))));
3315 net
->ipv4
.rt_secret_timer
.function
= rt_secret_rebuild
;
3316 net
->ipv4
.rt_secret_timer
.data
= (unsigned long)net
;
3317 init_timer_deferrable(&net
->ipv4
.rt_secret_timer
);
3319 if (ip_rt_secret_interval
) {
3320 net
->ipv4
.rt_secret_timer
.expires
=
3321 jiffies
+ net_random() % ip_rt_secret_interval
+
3322 ip_rt_secret_interval
;
3323 add_timer(&net
->ipv4
.rt_secret_timer
);
3328 static __net_exit
void rt_secret_timer_exit(struct net
*net
)
3330 del_timer_sync(&net
->ipv4
.rt_secret_timer
);
3333 static __net_initdata
struct pernet_operations rt_secret_timer_ops
= {
3334 .init
= rt_secret_timer_init
,
3335 .exit
= rt_secret_timer_exit
,
3339 #ifdef CONFIG_NET_CLS_ROUTE
3340 struct ip_rt_acct
*ip_rt_acct __read_mostly
;
3341 #endif /* CONFIG_NET_CLS_ROUTE */
3343 static __initdata
unsigned long rhash_entries
;
3344 static int __init
set_rhash_entries(char *str
)
3348 rhash_entries
= simple_strtoul(str
, &str
, 0);
3351 __setup("rhash_entries=", set_rhash_entries
);
3353 int __init
ip_rt_init(void)
3357 #ifdef CONFIG_NET_CLS_ROUTE
3358 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3360 panic("IP: failed to allocate ip_rt_acct\n");
3363 ipv4_dst_ops
.kmem_cachep
=
3364 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3365 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3367 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3369 rt_hash_table
= (struct rt_hash_bucket
*)
3370 alloc_large_system_hash("IP route cache",
3371 sizeof(struct rt_hash_bucket
),
3373 (totalram_pages
>= 128 * 1024) ?
3378 rhash_entries
? 0 : 512 * 1024);
3379 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3380 rt_hash_lock_init();
3382 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3383 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3388 /* All the timers, started at system startup tend
3389 to synchronize. Perturb it a bit.
3391 INIT_DELAYED_WORK_DEFERRABLE(&expires_work
, rt_worker_func
);
3392 expires_ljiffies
= jiffies
;
3393 schedule_delayed_work(&expires_work
,
3394 net_random() % ip_rt_gc_interval
+ ip_rt_gc_interval
);
3396 if (register_pernet_subsys(&rt_secret_timer_ops
))
3397 printk(KERN_ERR
"Unable to setup rt_secret_timer\n");
3399 if (ip_rt_proc_init())
3400 printk(KERN_ERR
"Unable to create route proc files\n");
3403 xfrm4_init(ip_rt_max_size
);
3405 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
);
3407 #ifdef CONFIG_SYSCTL
3408 register_pernet_subsys(&sysctl_route_ops
);
3413 #ifdef CONFIG_SYSCTL
3415 * We really need to sanitize the damn ipv4 init order, then all
3416 * this nonsense will go away.
3418 void __init
ip_static_sysctl_init(void)
3420 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);
3424 EXPORT_SYMBOL(__ip_select_ident
);
3425 EXPORT_SYMBOL(ip_route_input
);
3426 EXPORT_SYMBOL(ip_route_output_key
);