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>
93 #include <linux/slab.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/netevent.h>
107 #include <net/rtnetlink.h>
109 #include <linux/sysctl.h>
112 #define RT_FL_TOS(oldflp) \
113 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
115 #define IP_MAX_MTU 0xFFF0
117 #define RT_GC_TIMEOUT (300*HZ)
119 static int ip_rt_max_size
;
120 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
121 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
122 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
123 static int ip_rt_redirect_number __read_mostly
= 9;
124 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
125 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly
= HZ
;
127 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
128 static int ip_rt_gc_elasticity __read_mostly
= 8;
129 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
130 static int ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly
= 256;
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
);
151 static struct dst_ops ipv4_dst_ops
= {
153 .protocol
= cpu_to_be16(ETH_P_IP
),
154 .gc
= rt_garbage_collect
,
155 .check
= ipv4_dst_check
,
156 .destroy
= ipv4_dst_destroy
,
157 .ifdown
= ipv4_dst_ifdown
,
158 .negative_advice
= ipv4_negative_advice
,
159 .link_failure
= ipv4_link_failure
,
160 .update_pmtu
= ip_rt_update_pmtu
,
161 .local_out
= __ip_local_out
,
162 .entries
= ATOMIC_INIT(0),
165 #define ECN_OR_COST(class) TC_PRIO_##class
167 const __u8 ip_tos2prio
[16] = {
171 ECN_OR_COST(BESTEFFORT
),
177 ECN_OR_COST(INTERACTIVE
),
179 ECN_OR_COST(INTERACTIVE
),
180 TC_PRIO_INTERACTIVE_BULK
,
181 ECN_OR_COST(INTERACTIVE_BULK
),
182 TC_PRIO_INTERACTIVE_BULK
,
183 ECN_OR_COST(INTERACTIVE_BULK
)
191 /* The locking scheme is rather straight forward:
193 * 1) Read-Copy Update protects the buckets of the central route hash.
194 * 2) Only writers remove entries, and they hold the lock
195 * as they look at rtable reference counts.
196 * 3) Only readers acquire references to rtable entries,
197 * they do so with atomic increments and with the
201 struct rt_hash_bucket
{
202 struct rtable
*chain
;
205 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
206 defined(CONFIG_PROVE_LOCKING)
208 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
209 * The size of this table is a power of two and depends on the number of CPUS.
210 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
212 #ifdef CONFIG_LOCKDEP
213 # define RT_HASH_LOCK_SZ 256
216 # define RT_HASH_LOCK_SZ 4096
218 # define RT_HASH_LOCK_SZ 2048
220 # define RT_HASH_LOCK_SZ 1024
222 # define RT_HASH_LOCK_SZ 512
224 # define RT_HASH_LOCK_SZ 256
228 static spinlock_t
*rt_hash_locks
;
229 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
231 static __init
void rt_hash_lock_init(void)
235 rt_hash_locks
= kmalloc(sizeof(spinlock_t
) * RT_HASH_LOCK_SZ
,
238 panic("IP: failed to allocate rt_hash_locks\n");
240 for (i
= 0; i
< RT_HASH_LOCK_SZ
; i
++)
241 spin_lock_init(&rt_hash_locks
[i
]);
244 # define rt_hash_lock_addr(slot) NULL
246 static inline void rt_hash_lock_init(void)
251 static struct rt_hash_bucket
*rt_hash_table __read_mostly
;
252 static unsigned rt_hash_mask __read_mostly
;
253 static unsigned int rt_hash_log __read_mostly
;
255 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
256 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
258 static inline unsigned int rt_hash(__be32 daddr
, __be32 saddr
, int idx
,
261 return jhash_3words((__force u32
)daddr
, (__force u32
)saddr
,
266 static inline int rt_genid(struct net
*net
)
268 return atomic_read(&net
->ipv4
.rt_genid
);
271 #ifdef CONFIG_PROC_FS
272 struct rt_cache_iter_state
{
273 struct seq_net_private p
;
278 static struct rtable
*rt_cache_get_first(struct seq_file
*seq
)
280 struct rt_cache_iter_state
*st
= seq
->private;
281 struct rtable
*r
= NULL
;
283 for (st
->bucket
= rt_hash_mask
; st
->bucket
>= 0; --st
->bucket
) {
284 if (!rt_hash_table
[st
->bucket
].chain
)
287 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
289 if (dev_net(r
->dst
.dev
) == seq_file_net(seq
) &&
290 r
->rt_genid
== st
->genid
)
292 r
= rcu_dereference_bh(r
->dst
.rt_next
);
294 rcu_read_unlock_bh();
299 static struct rtable
*__rt_cache_get_next(struct seq_file
*seq
,
302 struct rt_cache_iter_state
*st
= seq
->private;
306 rcu_read_unlock_bh();
308 if (--st
->bucket
< 0)
310 } while (!rt_hash_table
[st
->bucket
].chain
);
312 r
= rt_hash_table
[st
->bucket
].chain
;
314 return rcu_dereference_bh(r
);
317 static struct rtable
*rt_cache_get_next(struct seq_file
*seq
,
320 struct rt_cache_iter_state
*st
= seq
->private;
321 while ((r
= __rt_cache_get_next(seq
, r
)) != NULL
) {
322 if (dev_net(r
->dst
.dev
) != seq_file_net(seq
))
324 if (r
->rt_genid
== st
->genid
)
330 static struct rtable
*rt_cache_get_idx(struct seq_file
*seq
, loff_t pos
)
332 struct rtable
*r
= rt_cache_get_first(seq
);
335 while (pos
&& (r
= rt_cache_get_next(seq
, r
)))
337 return pos
? NULL
: r
;
340 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
342 struct rt_cache_iter_state
*st
= seq
->private;
344 return rt_cache_get_idx(seq
, *pos
- 1);
345 st
->genid
= rt_genid(seq_file_net(seq
));
346 return SEQ_START_TOKEN
;
349 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
353 if (v
== SEQ_START_TOKEN
)
354 r
= rt_cache_get_first(seq
);
356 r
= rt_cache_get_next(seq
, v
);
361 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
363 if (v
&& v
!= SEQ_START_TOKEN
)
364 rcu_read_unlock_bh();
367 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
369 if (v
== SEQ_START_TOKEN
)
370 seq_printf(seq
, "%-127s\n",
371 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
372 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
375 struct rtable
*r
= v
;
378 seq_printf(seq
, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
379 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
380 r
->dst
.dev
? r
->dst
.dev
->name
: "*",
381 (__force u32
)r
->rt_dst
,
382 (__force u32
)r
->rt_gateway
,
383 r
->rt_flags
, atomic_read(&r
->dst
.__refcnt
),
384 r
->dst
.__use
, 0, (__force u32
)r
->rt_src
,
385 (dst_metric(&r
->dst
, RTAX_ADVMSS
) ?
386 (int)dst_metric(&r
->dst
, RTAX_ADVMSS
) + 40 : 0),
387 dst_metric(&r
->dst
, RTAX_WINDOW
),
388 (int)((dst_metric(&r
->dst
, RTAX_RTT
) >> 3) +
389 dst_metric(&r
->dst
, RTAX_RTTVAR
)),
391 r
->dst
.hh
? atomic_read(&r
->dst
.hh
->hh_refcnt
) : -1,
392 r
->dst
.hh
? (r
->dst
.hh
->hh_output
==
394 r
->rt_spec_dst
, &len
);
396 seq_printf(seq
, "%*s\n", 127 - len
, "");
401 static const struct seq_operations rt_cache_seq_ops
= {
402 .start
= rt_cache_seq_start
,
403 .next
= rt_cache_seq_next
,
404 .stop
= rt_cache_seq_stop
,
405 .show
= rt_cache_seq_show
,
408 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
410 return seq_open_net(inode
, file
, &rt_cache_seq_ops
,
411 sizeof(struct rt_cache_iter_state
));
414 static const struct file_operations rt_cache_seq_fops
= {
415 .owner
= THIS_MODULE
,
416 .open
= rt_cache_seq_open
,
419 .release
= seq_release_net
,
423 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
428 return SEQ_START_TOKEN
;
430 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
431 if (!cpu_possible(cpu
))
434 return &per_cpu(rt_cache_stat
, cpu
);
439 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
443 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
444 if (!cpu_possible(cpu
))
447 return &per_cpu(rt_cache_stat
, cpu
);
453 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
458 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
460 struct rt_cache_stat
*st
= v
;
462 if (v
== SEQ_START_TOKEN
) {
463 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");
467 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
468 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
469 atomic_read(&ipv4_dst_ops
.entries
),
492 static const struct seq_operations rt_cpu_seq_ops
= {
493 .start
= rt_cpu_seq_start
,
494 .next
= rt_cpu_seq_next
,
495 .stop
= rt_cpu_seq_stop
,
496 .show
= rt_cpu_seq_show
,
500 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
502 return seq_open(file
, &rt_cpu_seq_ops
);
505 static const struct file_operations rt_cpu_seq_fops
= {
506 .owner
= THIS_MODULE
,
507 .open
= rt_cpu_seq_open
,
510 .release
= seq_release
,
513 #ifdef CONFIG_NET_CLS_ROUTE
514 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
516 struct ip_rt_acct
*dst
, *src
;
519 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
523 for_each_possible_cpu(i
) {
524 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
525 for (j
= 0; j
< 256; j
++) {
526 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
527 dst
[j
].o_packets
+= src
[j
].o_packets
;
528 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
529 dst
[j
].i_packets
+= src
[j
].i_packets
;
533 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
538 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
540 return single_open(file
, rt_acct_proc_show
, NULL
);
543 static const struct file_operations rt_acct_proc_fops
= {
544 .owner
= THIS_MODULE
,
545 .open
= rt_acct_proc_open
,
548 .release
= single_release
,
552 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
554 struct proc_dir_entry
*pde
;
556 pde
= proc_net_fops_create(net
, "rt_cache", S_IRUGO
,
561 pde
= proc_create("rt_cache", S_IRUGO
,
562 net
->proc_net_stat
, &rt_cpu_seq_fops
);
566 #ifdef CONFIG_NET_CLS_ROUTE
567 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
573 #ifdef CONFIG_NET_CLS_ROUTE
575 remove_proc_entry("rt_cache", net
->proc_net_stat
);
578 remove_proc_entry("rt_cache", net
->proc_net
);
583 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
585 remove_proc_entry("rt_cache", net
->proc_net_stat
);
586 remove_proc_entry("rt_cache", net
->proc_net
);
587 #ifdef CONFIG_NET_CLS_ROUTE
588 remove_proc_entry("rt_acct", net
->proc_net
);
592 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
593 .init
= ip_rt_do_proc_init
,
594 .exit
= ip_rt_do_proc_exit
,
597 static int __init
ip_rt_proc_init(void)
599 return register_pernet_subsys(&ip_rt_proc_ops
);
603 static inline int ip_rt_proc_init(void)
607 #endif /* CONFIG_PROC_FS */
609 static inline void rt_free(struct rtable
*rt
)
611 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
614 static inline void rt_drop(struct rtable
*rt
)
617 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
620 static inline int rt_fast_clean(struct rtable
*rth
)
622 /* Kill broadcast/multicast entries very aggresively, if they
623 collide in hash table with more useful entries */
624 return (rth
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) &&
625 rth
->fl
.iif
&& rth
->dst
.rt_next
;
628 static inline int rt_valuable(struct rtable
*rth
)
630 return (rth
->rt_flags
& (RTCF_REDIRECTED
| RTCF_NOTIFY
)) ||
634 static int rt_may_expire(struct rtable
*rth
, unsigned long tmo1
, unsigned long tmo2
)
639 if (atomic_read(&rth
->dst
.__refcnt
))
643 if (rth
->dst
.expires
&&
644 time_after_eq(jiffies
, rth
->dst
.expires
))
647 age
= jiffies
- rth
->dst
.lastuse
;
649 if ((age
<= tmo1
&& !rt_fast_clean(rth
)) ||
650 (age
<= tmo2
&& rt_valuable(rth
)))
656 /* Bits of score are:
658 * 30: not quite useless
659 * 29..0: usage counter
661 static inline u32
rt_score(struct rtable
*rt
)
663 u32 score
= jiffies
- rt
->dst
.lastuse
;
665 score
= ~score
& ~(3<<30);
671 !(rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
|RTCF_LOCAL
)))
677 static inline bool rt_caching(const struct net
*net
)
679 return net
->ipv4
.current_rt_cache_rebuild_count
<=
680 net
->ipv4
.sysctl_rt_cache_rebuild_count
;
683 static inline bool compare_hash_inputs(const struct flowi
*fl1
,
684 const struct flowi
*fl2
)
686 return ((((__force u32
)fl1
->nl_u
.ip4_u
.daddr
^ (__force u32
)fl2
->nl_u
.ip4_u
.daddr
) |
687 ((__force u32
)fl1
->nl_u
.ip4_u
.saddr
^ (__force u32
)fl2
->nl_u
.ip4_u
.saddr
) |
688 (fl1
->iif
^ fl2
->iif
)) == 0);
691 static inline int compare_keys(struct flowi
*fl1
, struct flowi
*fl2
)
693 return (((__force u32
)fl1
->nl_u
.ip4_u
.daddr
^ (__force u32
)fl2
->nl_u
.ip4_u
.daddr
) |
694 ((__force u32
)fl1
->nl_u
.ip4_u
.saddr
^ (__force u32
)fl2
->nl_u
.ip4_u
.saddr
) |
695 (fl1
->mark
^ fl2
->mark
) |
696 (*(u16
*)&fl1
->nl_u
.ip4_u
.tos
^ *(u16
*)&fl2
->nl_u
.ip4_u
.tos
) |
697 (fl1
->oif
^ fl2
->oif
) |
698 (fl1
->iif
^ fl2
->iif
)) == 0;
701 static inline int compare_netns(struct rtable
*rt1
, struct rtable
*rt2
)
703 return net_eq(dev_net(rt1
->dst
.dev
), dev_net(rt2
->dst
.dev
));
706 static inline int rt_is_expired(struct rtable
*rth
)
708 return rth
->rt_genid
!= rt_genid(dev_net(rth
->dst
.dev
));
712 * Perform a full scan of hash table and free all entries.
713 * Can be called by a softirq or a process.
714 * In the later case, we want to be reschedule if necessary
716 static void rt_do_flush(int process_context
)
719 struct rtable
*rth
, *next
;
720 struct rtable
* tail
;
722 for (i
= 0; i
<= rt_hash_mask
; i
++) {
723 if (process_context
&& need_resched())
725 rth
= rt_hash_table
[i
].chain
;
729 spin_lock_bh(rt_hash_lock_addr(i
));
732 struct rtable
** prev
, * p
;
734 rth
= rt_hash_table
[i
].chain
;
736 /* defer releasing the head of the list after spin_unlock */
737 for (tail
= rth
; tail
; tail
= tail
->dst
.rt_next
)
738 if (!rt_is_expired(tail
))
741 rt_hash_table
[i
].chain
= tail
;
743 /* call rt_free on entries after the tail requiring flush */
744 prev
= &rt_hash_table
[i
].chain
;
745 for (p
= *prev
; p
; p
= next
) {
746 next
= p
->dst
.rt_next
;
747 if (!rt_is_expired(p
)) {
748 prev
= &p
->dst
.rt_next
;
756 rth
= rt_hash_table
[i
].chain
;
757 rt_hash_table
[i
].chain
= NULL
;
760 spin_unlock_bh(rt_hash_lock_addr(i
));
762 for (; rth
!= tail
; rth
= next
) {
763 next
= rth
->dst
.rt_next
;
770 * While freeing expired entries, we compute average chain length
771 * and standard deviation, using fixed-point arithmetic.
772 * This to have an estimation of rt_chain_length_max
773 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
774 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
778 #define ONE (1UL << FRACT_BITS)
781 * Given a hash chain and an item in this hash chain,
782 * find if a previous entry has the same hash_inputs
783 * (but differs on tos, mark or oif)
784 * Returns 0 if an alias is found.
785 * Returns ONE if rth has no alias before itself.
787 static int has_noalias(const struct rtable
*head
, const struct rtable
*rth
)
789 const struct rtable
*aux
= head
;
792 if (compare_hash_inputs(&aux
->fl
, &rth
->fl
))
794 aux
= aux
->dst
.rt_next
;
799 static void rt_check_expire(void)
801 static unsigned int rover
;
802 unsigned int i
= rover
, goal
;
803 struct rtable
*rth
, **rthp
;
804 unsigned long samples
= 0;
805 unsigned long sum
= 0, sum2
= 0;
809 delta
= jiffies
- expires_ljiffies
;
810 expires_ljiffies
= jiffies
;
811 mult
= ((u64
)delta
) << rt_hash_log
;
812 if (ip_rt_gc_timeout
> 1)
813 do_div(mult
, ip_rt_gc_timeout
);
814 goal
= (unsigned int)mult
;
815 if (goal
> rt_hash_mask
)
816 goal
= rt_hash_mask
+ 1;
817 for (; goal
> 0; goal
--) {
818 unsigned long tmo
= ip_rt_gc_timeout
;
819 unsigned long length
;
821 i
= (i
+ 1) & rt_hash_mask
;
822 rthp
= &rt_hash_table
[i
].chain
;
832 spin_lock_bh(rt_hash_lock_addr(i
));
833 while ((rth
= *rthp
) != NULL
) {
834 prefetch(rth
->dst
.rt_next
);
835 if (rt_is_expired(rth
)) {
836 *rthp
= rth
->dst
.rt_next
;
840 if (rth
->dst
.expires
) {
841 /* Entry is expired even if it is in use */
842 if (time_before_eq(jiffies
, rth
->dst
.expires
)) {
845 rthp
= &rth
->dst
.rt_next
;
847 * We only count entries on
848 * a chain with equal hash inputs once
849 * so that entries for different QOS
850 * levels, and other non-hash input
851 * attributes don't unfairly skew
852 * the length computation
854 length
+= has_noalias(rt_hash_table
[i
].chain
, rth
);
857 } else if (!rt_may_expire(rth
, tmo
, ip_rt_gc_timeout
))
860 /* Cleanup aged off entries. */
861 *rthp
= rth
->dst
.rt_next
;
864 spin_unlock_bh(rt_hash_lock_addr(i
));
866 sum2
+= length
*length
;
869 unsigned long avg
= sum
/ samples
;
870 unsigned long sd
= int_sqrt(sum2
/ samples
- avg
*avg
);
871 rt_chain_length_max
= max_t(unsigned long,
873 (avg
+ 4*sd
) >> FRACT_BITS
);
879 * rt_worker_func() is run in process context.
880 * we call rt_check_expire() to scan part of the hash table
882 static void rt_worker_func(struct work_struct
*work
)
885 schedule_delayed_work(&expires_work
, ip_rt_gc_interval
);
889 * Pertubation of rt_genid by a small quantity [1..256]
890 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
891 * many times (2^24) without giving recent rt_genid.
892 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
894 static void rt_cache_invalidate(struct net
*net
)
896 unsigned char shuffle
;
898 get_random_bytes(&shuffle
, sizeof(shuffle
));
899 atomic_add(shuffle
+ 1U, &net
->ipv4
.rt_genid
);
903 * delay < 0 : invalidate cache (fast : entries will be deleted later)
904 * delay >= 0 : invalidate & flush cache (can be long)
906 void rt_cache_flush(struct net
*net
, int delay
)
908 rt_cache_invalidate(net
);
910 rt_do_flush(!in_softirq());
913 /* Flush previous cache invalidated entries from the cache */
914 void rt_cache_flush_batch(void)
916 rt_do_flush(!in_softirq());
919 static void rt_emergency_hash_rebuild(struct net
*net
)
922 printk(KERN_WARNING
"Route hash chain too long!\n");
923 rt_cache_invalidate(net
);
927 Short description of GC goals.
929 We want to build algorithm, which will keep routing cache
930 at some equilibrium point, when number of aged off entries
931 is kept approximately equal to newly generated ones.
933 Current expiration strength is variable "expire".
934 We try to adjust it dynamically, so that if networking
935 is idle expires is large enough to keep enough of warm entries,
936 and when load increases it reduces to limit cache size.
939 static int rt_garbage_collect(struct dst_ops
*ops
)
941 static unsigned long expire
= RT_GC_TIMEOUT
;
942 static unsigned long last_gc
;
944 static int equilibrium
;
945 struct rtable
*rth
, **rthp
;
946 unsigned long now
= jiffies
;
950 * Garbage collection is pretty expensive,
951 * do not make it too frequently.
954 RT_CACHE_STAT_INC(gc_total
);
956 if (now
- last_gc
< ip_rt_gc_min_interval
&&
957 atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
) {
958 RT_CACHE_STAT_INC(gc_ignored
);
962 /* Calculate number of entries, which we want to expire now. */
963 goal
= atomic_read(&ipv4_dst_ops
.entries
) -
964 (ip_rt_gc_elasticity
<< rt_hash_log
);
966 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
967 equilibrium
= ipv4_dst_ops
.gc_thresh
;
968 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
970 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
971 goal
= atomic_read(&ipv4_dst_ops
.entries
) - equilibrium
;
974 /* We are in dangerous area. Try to reduce cache really
977 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
978 equilibrium
= atomic_read(&ipv4_dst_ops
.entries
) - goal
;
981 if (now
- last_gc
>= ip_rt_gc_min_interval
)
992 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
993 unsigned long tmo
= expire
;
995 k
= (k
+ 1) & rt_hash_mask
;
996 rthp
= &rt_hash_table
[k
].chain
;
997 spin_lock_bh(rt_hash_lock_addr(k
));
998 while ((rth
= *rthp
) != NULL
) {
999 if (!rt_is_expired(rth
) &&
1000 !rt_may_expire(rth
, tmo
, expire
)) {
1002 rthp
= &rth
->dst
.rt_next
;
1005 *rthp
= rth
->dst
.rt_next
;
1009 spin_unlock_bh(rt_hash_lock_addr(k
));
1018 /* Goal is not achieved. We stop process if:
1020 - if expire reduced to zero. Otherwise, expire is halfed.
1021 - if table is not full.
1022 - if we are called from interrupt.
1023 - jiffies check is just fallback/debug loop breaker.
1024 We will not spin here for long time in any case.
1027 RT_CACHE_STAT_INC(gc_goal_miss
);
1033 #if RT_CACHE_DEBUG >= 2
1034 printk(KERN_DEBUG
"expire>> %u %d %d %d\n", expire
,
1035 atomic_read(&ipv4_dst_ops
.entries
), goal
, i
);
1038 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1040 } while (!in_softirq() && time_before_eq(jiffies
, now
));
1042 if (atomic_read(&ipv4_dst_ops
.entries
) < ip_rt_max_size
)
1044 if (net_ratelimit())
1045 printk(KERN_WARNING
"dst cache overflow\n");
1046 RT_CACHE_STAT_INC(gc_dst_overflow
);
1050 expire
+= ip_rt_gc_min_interval
;
1051 if (expire
> ip_rt_gc_timeout
||
1052 atomic_read(&ipv4_dst_ops
.entries
) < ipv4_dst_ops
.gc_thresh
)
1053 expire
= ip_rt_gc_timeout
;
1054 #if RT_CACHE_DEBUG >= 2
1055 printk(KERN_DEBUG
"expire++ %u %d %d %d\n", expire
,
1056 atomic_read(&ipv4_dst_ops
.entries
), goal
, rover
);
1062 * Returns number of entries in a hash chain that have different hash_inputs
1064 static int slow_chain_length(const struct rtable
*head
)
1067 const struct rtable
*rth
= head
;
1070 length
+= has_noalias(head
, rth
);
1071 rth
= rth
->dst
.rt_next
;
1073 return length
>> FRACT_BITS
;
1076 static int rt_intern_hash(unsigned hash
, struct rtable
*rt
,
1077 struct rtable
**rp
, struct sk_buff
*skb
, int ifindex
)
1079 struct rtable
*rth
, **rthp
;
1081 struct rtable
*cand
, **candp
;
1084 int attempts
= !in_softirq();
1088 min_score
= ~(u32
)0;
1093 if (!rt_caching(dev_net(rt
->dst
.dev
))) {
1095 * If we're not caching, just tell the caller we
1096 * were successful and don't touch the route. The
1097 * caller hold the sole reference to the cache entry, and
1098 * it will be released when the caller is done with it.
1099 * If we drop it here, the callers have no way to resolve routes
1100 * when we're not caching. Instead, just point *rp at rt, so
1101 * the caller gets a single use out of the route
1102 * Note that we do rt_free on this new route entry, so that
1103 * once its refcount hits zero, we are still able to reap it
1105 * Note also the rt_free uses call_rcu. We don't actually
1106 * need rcu protection here, this is just our path to get
1107 * on the route gc list.
1110 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1111 int err
= arp_bind_neighbour(&rt
->dst
);
1113 if (net_ratelimit())
1115 "Neighbour table failure & not caching routes.\n");
1125 rthp
= &rt_hash_table
[hash
].chain
;
1127 spin_lock_bh(rt_hash_lock_addr(hash
));
1128 while ((rth
= *rthp
) != NULL
) {
1129 if (rt_is_expired(rth
)) {
1130 *rthp
= rth
->dst
.rt_next
;
1134 if (compare_keys(&rth
->fl
, &rt
->fl
) && compare_netns(rth
, rt
)) {
1136 *rthp
= rth
->dst
.rt_next
;
1138 * Since lookup is lockfree, the deletion
1139 * must be visible to another weakly ordered CPU before
1140 * the insertion at the start of the hash chain.
1142 rcu_assign_pointer(rth
->dst
.rt_next
,
1143 rt_hash_table
[hash
].chain
);
1145 * Since lookup is lockfree, the update writes
1146 * must be ordered for consistency on SMP.
1148 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1150 dst_use(&rth
->dst
, now
);
1151 spin_unlock_bh(rt_hash_lock_addr(hash
));
1157 skb_dst_set(skb
, &rth
->dst
);
1161 if (!atomic_read(&rth
->dst
.__refcnt
)) {
1162 u32 score
= rt_score(rth
);
1164 if (score
<= min_score
) {
1173 rthp
= &rth
->dst
.rt_next
;
1177 /* ip_rt_gc_elasticity used to be average length of chain
1178 * length, when exceeded gc becomes really aggressive.
1180 * The second limit is less certain. At the moment it allows
1181 * only 2 entries per bucket. We will see.
1183 if (chain_length
> ip_rt_gc_elasticity
) {
1184 *candp
= cand
->dst
.rt_next
;
1188 if (chain_length
> rt_chain_length_max
&&
1189 slow_chain_length(rt_hash_table
[hash
].chain
) > rt_chain_length_max
) {
1190 struct net
*net
= dev_net(rt
->dst
.dev
);
1191 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1192 if (!rt_caching(net
)) {
1193 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1194 rt
->dst
.dev
->name
, num
);
1196 rt_emergency_hash_rebuild(net
);
1197 spin_unlock_bh(rt_hash_lock_addr(hash
));
1199 hash
= rt_hash(rt
->fl
.fl4_dst
, rt
->fl
.fl4_src
,
1200 ifindex
, rt_genid(net
));
1205 /* Try to bind route to arp only if it is output
1206 route or unicast forwarding path.
1208 if (rt
->rt_type
== RTN_UNICAST
|| rt
->fl
.iif
== 0) {
1209 int err
= arp_bind_neighbour(&rt
->dst
);
1211 spin_unlock_bh(rt_hash_lock_addr(hash
));
1213 if (err
!= -ENOBUFS
) {
1218 /* Neighbour tables are full and nothing
1219 can be released. Try to shrink route cache,
1220 it is most likely it holds some neighbour records.
1222 if (attempts
-- > 0) {
1223 int saved_elasticity
= ip_rt_gc_elasticity
;
1224 int saved_int
= ip_rt_gc_min_interval
;
1225 ip_rt_gc_elasticity
= 1;
1226 ip_rt_gc_min_interval
= 0;
1227 rt_garbage_collect(&ipv4_dst_ops
);
1228 ip_rt_gc_min_interval
= saved_int
;
1229 ip_rt_gc_elasticity
= saved_elasticity
;
1233 if (net_ratelimit())
1234 printk(KERN_WARNING
"Neighbour table overflow.\n");
1240 rt
->dst
.rt_next
= rt_hash_table
[hash
].chain
;
1242 #if RT_CACHE_DEBUG >= 2
1243 if (rt
->dst
.rt_next
) {
1245 printk(KERN_DEBUG
"rt_cache @%02x: %pI4",
1247 for (trt
= rt
->dst
.rt_next
; trt
; trt
= trt
->dst
.rt_next
)
1248 printk(" . %pI4", &trt
->rt_dst
);
1253 * Since lookup is lockfree, we must make sure
1254 * previous writes to rt are comitted to memory
1255 * before making rt visible to other CPUS.
1257 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1259 spin_unlock_bh(rt_hash_lock_addr(hash
));
1265 skb_dst_set(skb
, &rt
->dst
);
1269 void rt_bind_peer(struct rtable
*rt
, int create
)
1271 static DEFINE_SPINLOCK(rt_peer_lock
);
1272 struct inet_peer
*peer
;
1274 peer
= inet_getpeer(rt
->rt_dst
, create
);
1276 spin_lock_bh(&rt_peer_lock
);
1277 if (rt
->peer
== NULL
) {
1281 spin_unlock_bh(&rt_peer_lock
);
1287 * Peer allocation may fail only in serious out-of-memory conditions. However
1288 * we still can generate some output.
1289 * Random ID selection looks a bit dangerous because we have no chances to
1290 * select ID being unique in a reasonable period of time.
1291 * But broken packet identifier may be better than no packet at all.
1293 static void ip_select_fb_ident(struct iphdr
*iph
)
1295 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1296 static u32 ip_fallback_id
;
1299 spin_lock_bh(&ip_fb_id_lock
);
1300 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1301 iph
->id
= htons(salt
& 0xFFFF);
1302 ip_fallback_id
= salt
;
1303 spin_unlock_bh(&ip_fb_id_lock
);
1306 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1308 struct rtable
*rt
= (struct rtable
*) dst
;
1311 if (rt
->peer
== NULL
)
1312 rt_bind_peer(rt
, 1);
1314 /* If peer is attached to destination, it is never detached,
1315 so that we need not to grab a lock to dereference it.
1318 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1322 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1323 __builtin_return_address(0));
1325 ip_select_fb_ident(iph
);
1328 static void rt_del(unsigned hash
, struct rtable
*rt
)
1330 struct rtable
**rthp
, *aux
;
1332 rthp
= &rt_hash_table
[hash
].chain
;
1333 spin_lock_bh(rt_hash_lock_addr(hash
));
1335 while ((aux
= *rthp
) != NULL
) {
1336 if (aux
== rt
|| rt_is_expired(aux
)) {
1337 *rthp
= aux
->dst
.rt_next
;
1341 rthp
= &aux
->dst
.rt_next
;
1343 spin_unlock_bh(rt_hash_lock_addr(hash
));
1346 /* called in rcu_read_lock() section */
1347 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1348 __be32 saddr
, struct net_device
*dev
)
1351 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1352 struct rtable
*rth
, **rthp
;
1353 __be32 skeys
[2] = { saddr
, 0 };
1354 int ikeys
[2] = { dev
->ifindex
, 0 };
1355 struct netevent_redirect netevent
;
1362 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1363 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1364 ipv4_is_zeronet(new_gw
))
1365 goto reject_redirect
;
1367 if (!rt_caching(net
))
1368 goto reject_redirect
;
1370 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1371 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1372 goto reject_redirect
;
1373 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1374 goto reject_redirect
;
1376 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1377 goto reject_redirect
;
1380 for (i
= 0; i
< 2; i
++) {
1381 for (k
= 0; k
< 2; k
++) {
1382 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1385 rthp
=&rt_hash_table
[hash
].chain
;
1387 while ((rth
= rcu_dereference(*rthp
)) != NULL
) {
1390 if (rth
->fl
.fl4_dst
!= daddr
||
1391 rth
->fl
.fl4_src
!= skeys
[i
] ||
1392 rth
->fl
.oif
!= ikeys
[k
] ||
1394 rt_is_expired(rth
) ||
1395 !net_eq(dev_net(rth
->dst
.dev
), net
)) {
1396 rthp
= &rth
->dst
.rt_next
;
1400 if (rth
->rt_dst
!= daddr
||
1401 rth
->rt_src
!= saddr
||
1403 rth
->rt_gateway
!= old_gw
||
1404 rth
->dst
.dev
!= dev
)
1407 dst_hold(&rth
->dst
);
1409 rt
= dst_alloc(&ipv4_dst_ops
);
1415 /* Copy all the information. */
1418 atomic_set(&rt
->dst
.__refcnt
, 1);
1419 rt
->dst
.child
= NULL
;
1421 dev_hold(rt
->dst
.dev
);
1423 in_dev_hold(rt
->idev
);
1424 rt
->dst
.obsolete
= -1;
1425 rt
->dst
.lastuse
= jiffies
;
1426 rt
->dst
.path
= &rt
->dst
;
1427 rt
->dst
.neighbour
= NULL
;
1430 rt
->dst
.xfrm
= NULL
;
1432 rt
->rt_genid
= rt_genid(net
);
1433 rt
->rt_flags
|= RTCF_REDIRECTED
;
1435 /* Gateway is different ... */
1436 rt
->rt_gateway
= new_gw
;
1438 /* Redirect received -> path was valid */
1439 dst_confirm(&rth
->dst
);
1442 atomic_inc(&rt
->peer
->refcnt
);
1444 if (arp_bind_neighbour(&rt
->dst
) ||
1445 !(rt
->dst
.neighbour
->nud_state
&
1447 if (rt
->dst
.neighbour
)
1448 neigh_event_send(rt
->dst
.neighbour
, NULL
);
1454 netevent
.old
= &rth
->dst
;
1455 netevent
.new = &rt
->dst
;
1456 call_netevent_notifiers(NETEVENT_REDIRECT
,
1460 if (!rt_intern_hash(hash
, rt
, &rt
, NULL
, rt
->fl
.oif
))
1471 #ifdef CONFIG_IP_ROUTE_VERBOSE
1472 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1473 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1474 " Advised path = %pI4 -> %pI4\n",
1475 &old_gw
, dev
->name
, &new_gw
,
1481 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1483 struct rtable
*rt
= (struct rtable
*)dst
;
1484 struct dst_entry
*ret
= dst
;
1487 if (dst
->obsolete
> 0) {
1490 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
1492 time_after_eq(jiffies
, rt
->dst
.expires
))) {
1493 unsigned hash
= rt_hash(rt
->fl
.fl4_dst
, rt
->fl
.fl4_src
,
1495 rt_genid(dev_net(dst
->dev
)));
1496 #if RT_CACHE_DEBUG >= 1
1497 printk(KERN_DEBUG
"ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1498 &rt
->rt_dst
, rt
->fl
.fl4_tos
);
1509 * 1. The first ip_rt_redirect_number redirects are sent
1510 * with exponential backoff, then we stop sending them at all,
1511 * assuming that the host ignores our redirects.
1512 * 2. If we did not see packets requiring redirects
1513 * during ip_rt_redirect_silence, we assume that the host
1514 * forgot redirected route and start to send redirects again.
1516 * This algorithm is much cheaper and more intelligent than dumb load limiting
1519 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1520 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1523 void ip_rt_send_redirect(struct sk_buff
*skb
)
1525 struct rtable
*rt
= skb_rtable(skb
);
1526 struct in_device
*in_dev
;
1530 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
1531 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1535 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1538 /* No redirected packets during ip_rt_redirect_silence;
1539 * reset the algorithm.
1541 if (time_after(jiffies
, rt
->dst
.rate_last
+ ip_rt_redirect_silence
))
1542 rt
->dst
.rate_tokens
= 0;
1544 /* Too many ignored redirects; do not send anything
1545 * set dst.rate_last to the last seen redirected packet.
1547 if (rt
->dst
.rate_tokens
>= ip_rt_redirect_number
) {
1548 rt
->dst
.rate_last
= jiffies
;
1552 /* Check for load limit; set rate_last to the latest sent
1555 if (rt
->dst
.rate_tokens
== 0 ||
1557 (rt
->dst
.rate_last
+
1558 (ip_rt_redirect_load
<< rt
->dst
.rate_tokens
)))) {
1559 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1560 rt
->dst
.rate_last
= jiffies
;
1561 ++rt
->dst
.rate_tokens
;
1562 #ifdef CONFIG_IP_ROUTE_VERBOSE
1564 rt
->dst
.rate_tokens
== ip_rt_redirect_number
&&
1566 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1567 &rt
->rt_src
, rt
->rt_iif
,
1568 &rt
->rt_dst
, &rt
->rt_gateway
);
1573 static int ip_error(struct sk_buff
*skb
)
1575 struct rtable
*rt
= skb_rtable(skb
);
1579 switch (rt
->dst
.error
) {
1584 code
= ICMP_HOST_UNREACH
;
1587 code
= ICMP_NET_UNREACH
;
1588 IP_INC_STATS_BH(dev_net(rt
->dst
.dev
),
1589 IPSTATS_MIB_INNOROUTES
);
1592 code
= ICMP_PKT_FILTERED
;
1597 rt
->dst
.rate_tokens
+= now
- rt
->dst
.rate_last
;
1598 if (rt
->dst
.rate_tokens
> ip_rt_error_burst
)
1599 rt
->dst
.rate_tokens
= ip_rt_error_burst
;
1600 rt
->dst
.rate_last
= now
;
1601 if (rt
->dst
.rate_tokens
>= ip_rt_error_cost
) {
1602 rt
->dst
.rate_tokens
-= ip_rt_error_cost
;
1603 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1606 out
: kfree_skb(skb
);
1611 * The last two values are not from the RFC but
1612 * are needed for AMPRnet AX.25 paths.
1615 static const unsigned short mtu_plateau
[] =
1616 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1618 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1622 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1623 if (old_mtu
> mtu_plateau
[i
])
1624 return mtu_plateau
[i
];
1628 unsigned short ip_rt_frag_needed(struct net
*net
, struct iphdr
*iph
,
1629 unsigned short new_mtu
,
1630 struct net_device
*dev
)
1633 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1635 int ikeys
[2] = { dev
->ifindex
, 0 };
1636 __be32 skeys
[2] = { iph
->saddr
, 0, };
1637 __be32 daddr
= iph
->daddr
;
1638 unsigned short est_mtu
= 0;
1640 for (k
= 0; k
< 2; k
++) {
1641 for (i
= 0; i
< 2; i
++) {
1642 unsigned hash
= rt_hash(daddr
, skeys
[i
], ikeys
[k
],
1646 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
1647 rth
= rcu_dereference(rth
->dst
.rt_next
)) {
1648 unsigned short mtu
= new_mtu
;
1650 if (rth
->fl
.fl4_dst
!= daddr
||
1651 rth
->fl
.fl4_src
!= skeys
[i
] ||
1652 rth
->rt_dst
!= daddr
||
1653 rth
->rt_src
!= iph
->saddr
||
1654 rth
->fl
.oif
!= ikeys
[k
] ||
1656 dst_metric_locked(&rth
->dst
, RTAX_MTU
) ||
1657 !net_eq(dev_net(rth
->dst
.dev
), net
) ||
1661 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1663 /* BSD 4.2 compatibility hack :-( */
1665 old_mtu
>= dst_mtu(&rth
->dst
) &&
1666 old_mtu
>= 68 + (iph
->ihl
<< 2))
1667 old_mtu
-= iph
->ihl
<< 2;
1669 mtu
= guess_mtu(old_mtu
);
1671 if (mtu
<= dst_mtu(&rth
->dst
)) {
1672 if (mtu
< dst_mtu(&rth
->dst
)) {
1673 dst_confirm(&rth
->dst
);
1674 if (mtu
< ip_rt_min_pmtu
) {
1675 mtu
= ip_rt_min_pmtu
;
1676 rth
->dst
.metrics
[RTAX_LOCK
-1] |=
1679 rth
->dst
.metrics
[RTAX_MTU
-1] = mtu
;
1680 dst_set_expires(&rth
->dst
,
1689 return est_mtu
? : new_mtu
;
1692 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1694 if (dst_mtu(dst
) > mtu
&& mtu
>= 68 &&
1695 !(dst_metric_locked(dst
, RTAX_MTU
))) {
1696 if (mtu
< ip_rt_min_pmtu
) {
1697 mtu
= ip_rt_min_pmtu
;
1698 dst
->metrics
[RTAX_LOCK
-1] |= (1 << RTAX_MTU
);
1700 dst
->metrics
[RTAX_MTU
-1] = mtu
;
1701 dst_set_expires(dst
, ip_rt_mtu_expires
);
1702 call_netevent_notifiers(NETEVENT_PMTU_UPDATE
, dst
);
1706 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1708 if (rt_is_expired((struct rtable
*)dst
))
1713 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1715 struct rtable
*rt
= (struct rtable
*) dst
;
1716 struct inet_peer
*peer
= rt
->peer
;
1717 struct in_device
*idev
= rt
->idev
;
1730 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
1733 struct rtable
*rt
= (struct rtable
*) dst
;
1734 struct in_device
*idev
= rt
->idev
;
1735 if (dev
!= dev_net(dev
)->loopback_dev
&& idev
&& idev
->dev
== dev
) {
1736 struct in_device
*loopback_idev
=
1737 in_dev_get(dev_net(dev
)->loopback_dev
);
1738 if (loopback_idev
) {
1739 rt
->idev
= loopback_idev
;
1745 static void ipv4_link_failure(struct sk_buff
*skb
)
1749 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1751 rt
= skb_rtable(skb
);
1753 dst_set_expires(&rt
->dst
, 0);
1756 static int ip_rt_bug(struct sk_buff
*skb
)
1758 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1759 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1760 skb
->dev
? skb
->dev
->name
: "?");
1766 We do not cache source address of outgoing interface,
1767 because it is used only by IP RR, TS and SRR options,
1768 so that it out of fast path.
1770 BTW remember: "addr" is allowed to be not aligned
1774 void ip_rt_get_source(u8
*addr
, struct rtable
*rt
)
1777 struct fib_result res
;
1779 if (rt
->fl
.iif
== 0)
1781 else if (fib_lookup(dev_net(rt
->dst
.dev
), &rt
->fl
, &res
) == 0) {
1782 src
= FIB_RES_PREFSRC(res
);
1785 src
= inet_select_addr(rt
->dst
.dev
, rt
->rt_gateway
,
1787 memcpy(addr
, &src
, 4);
1790 #ifdef CONFIG_NET_CLS_ROUTE
1791 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1793 if (!(rt
->dst
.tclassid
& 0xFFFF))
1794 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1795 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1796 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1800 static void rt_set_nexthop(struct rtable
*rt
, struct fib_result
*res
, u32 itag
)
1802 struct fib_info
*fi
= res
->fi
;
1805 if (FIB_RES_GW(*res
) &&
1806 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1807 rt
->rt_gateway
= FIB_RES_GW(*res
);
1808 memcpy(rt
->dst
.metrics
, fi
->fib_metrics
,
1809 sizeof(rt
->dst
.metrics
));
1810 if (fi
->fib_mtu
== 0) {
1811 rt
->dst
.metrics
[RTAX_MTU
-1] = rt
->dst
.dev
->mtu
;
1812 if (dst_metric_locked(&rt
->dst
, RTAX_MTU
) &&
1813 rt
->rt_gateway
!= rt
->rt_dst
&&
1814 rt
->dst
.dev
->mtu
> 576)
1815 rt
->dst
.metrics
[RTAX_MTU
-1] = 576;
1817 #ifdef CONFIG_NET_CLS_ROUTE
1818 rt
->dst
.tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1821 rt
->dst
.metrics
[RTAX_MTU
-1]= rt
->dst
.dev
->mtu
;
1823 if (dst_metric(&rt
->dst
, RTAX_HOPLIMIT
) == 0)
1824 rt
->dst
.metrics
[RTAX_HOPLIMIT
-1] = sysctl_ip_default_ttl
;
1825 if (dst_mtu(&rt
->dst
) > IP_MAX_MTU
)
1826 rt
->dst
.metrics
[RTAX_MTU
-1] = IP_MAX_MTU
;
1827 if (dst_metric(&rt
->dst
, RTAX_ADVMSS
) == 0)
1828 rt
->dst
.metrics
[RTAX_ADVMSS
-1] = max_t(unsigned int, rt
->dst
.dev
->mtu
- 40,
1830 if (dst_metric(&rt
->dst
, RTAX_ADVMSS
) > 65535 - 40)
1831 rt
->dst
.metrics
[RTAX_ADVMSS
-1] = 65535 - 40;
1833 #ifdef CONFIG_NET_CLS_ROUTE
1834 #ifdef CONFIG_IP_MULTIPLE_TABLES
1835 set_class_tag(rt
, fib_rules_tclass(res
));
1837 set_class_tag(rt
, itag
);
1839 rt
->rt_type
= res
->type
;
1842 /* called in rcu_read_lock() section */
1843 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1844 u8 tos
, struct net_device
*dev
, int our
)
1849 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1853 /* Primary sanity checks. */
1858 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1859 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
1862 if (ipv4_is_zeronet(saddr
)) {
1863 if (!ipv4_is_local_multicast(daddr
))
1865 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
1867 err
= fib_validate_source(saddr
, 0, tos
, 0, dev
, &spec_dst
,
1872 rth
= dst_alloc(&ipv4_dst_ops
);
1876 rth
->dst
.output
= ip_rt_bug
;
1877 rth
->dst
.obsolete
= -1;
1879 atomic_set(&rth
->dst
.__refcnt
, 1);
1880 rth
->dst
.flags
= DST_HOST
;
1881 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
1882 rth
->dst
.flags
|= DST_NOPOLICY
;
1883 rth
->fl
.fl4_dst
= daddr
;
1884 rth
->rt_dst
= daddr
;
1885 rth
->fl
.fl4_tos
= tos
;
1886 rth
->fl
.mark
= skb
->mark
;
1887 rth
->fl
.fl4_src
= saddr
;
1888 rth
->rt_src
= saddr
;
1889 #ifdef CONFIG_NET_CLS_ROUTE
1890 rth
->dst
.tclassid
= itag
;
1893 rth
->fl
.iif
= dev
->ifindex
;
1894 rth
->dst
.dev
= init_net
.loopback_dev
;
1895 dev_hold(rth
->dst
.dev
);
1896 rth
->idev
= in_dev_get(rth
->dst
.dev
);
1898 rth
->rt_gateway
= daddr
;
1899 rth
->rt_spec_dst
= spec_dst
;
1900 rth
->rt_genid
= rt_genid(dev_net(dev
));
1901 rth
->rt_flags
= RTCF_MULTICAST
;
1902 rth
->rt_type
= RTN_MULTICAST
;
1904 rth
->dst
.input
= ip_local_deliver
;
1905 rth
->rt_flags
|= RTCF_LOCAL
;
1908 #ifdef CONFIG_IP_MROUTE
1909 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1910 rth
->dst
.input
= ip_mr_input
;
1912 RT_CACHE_STAT_INC(in_slow_mc
);
1914 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
1915 return rt_intern_hash(hash
, rth
, NULL
, skb
, dev
->ifindex
);
1926 static void ip_handle_martian_source(struct net_device
*dev
,
1927 struct in_device
*in_dev
,
1928 struct sk_buff
*skb
,
1932 RT_CACHE_STAT_INC(in_martian_src
);
1933 #ifdef CONFIG_IP_ROUTE_VERBOSE
1934 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1936 * RFC1812 recommendation, if source is martian,
1937 * the only hint is MAC header.
1939 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
1940 &daddr
, &saddr
, dev
->name
);
1941 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1943 const unsigned char *p
= skb_mac_header(skb
);
1944 printk(KERN_WARNING
"ll header: ");
1945 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
1947 if (i
< (dev
->hard_header_len
- 1))
1956 /* called in rcu_read_lock() section */
1957 static int __mkroute_input(struct sk_buff
*skb
,
1958 struct fib_result
*res
,
1959 struct in_device
*in_dev
,
1960 __be32 daddr
, __be32 saddr
, u32 tos
,
1961 struct rtable
**result
)
1965 struct in_device
*out_dev
;
1966 unsigned int flags
= 0;
1970 /* get a working reference to the output device */
1971 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
1972 if (out_dev
== NULL
) {
1973 if (net_ratelimit())
1974 printk(KERN_CRIT
"Bug in ip_route_input" \
1975 "_slow(). Please, report\n");
1980 err
= fib_validate_source(saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1981 in_dev
->dev
, &spec_dst
, &itag
, skb
->mark
);
1983 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1990 flags
|= RTCF_DIRECTSRC
;
1992 if (out_dev
== in_dev
&& err
&&
1993 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1994 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
1995 flags
|= RTCF_DOREDIRECT
;
1997 if (skb
->protocol
!= htons(ETH_P_IP
)) {
1998 /* Not IP (i.e. ARP). Do not create route, if it is
1999 * invalid for proxy arp. DNAT routes are always valid.
2001 * Proxy arp feature have been extended to allow, ARP
2002 * replies back to the same interface, to support
2003 * Private VLAN switch technologies. See arp.c.
2005 if (out_dev
== in_dev
&&
2006 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
2013 rth
= dst_alloc(&ipv4_dst_ops
);
2019 atomic_set(&rth
->dst
.__refcnt
, 1);
2020 rth
->dst
.flags
= DST_HOST
;
2021 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2022 rth
->dst
.flags
|= DST_NOPOLICY
;
2023 if (IN_DEV_CONF_GET(out_dev
, NOXFRM
))
2024 rth
->dst
.flags
|= DST_NOXFRM
;
2025 rth
->fl
.fl4_dst
= daddr
;
2026 rth
->rt_dst
= daddr
;
2027 rth
->fl
.fl4_tos
= tos
;
2028 rth
->fl
.mark
= skb
->mark
;
2029 rth
->fl
.fl4_src
= saddr
;
2030 rth
->rt_src
= saddr
;
2031 rth
->rt_gateway
= daddr
;
2033 rth
->fl
.iif
= in_dev
->dev
->ifindex
;
2034 rth
->dst
.dev
= (out_dev
)->dev
;
2035 dev_hold(rth
->dst
.dev
);
2036 rth
->idev
= in_dev_get(rth
->dst
.dev
);
2038 rth
->rt_spec_dst
= spec_dst
;
2040 rth
->dst
.obsolete
= -1;
2041 rth
->dst
.input
= ip_forward
;
2042 rth
->dst
.output
= ip_output
;
2043 rth
->rt_genid
= rt_genid(dev_net(rth
->dst
.dev
));
2045 rt_set_nexthop(rth
, res
, itag
);
2047 rth
->rt_flags
= flags
;
2055 static int ip_mkroute_input(struct sk_buff
*skb
,
2056 struct fib_result
*res
,
2057 const struct flowi
*fl
,
2058 struct in_device
*in_dev
,
2059 __be32 daddr
, __be32 saddr
, u32 tos
)
2061 struct rtable
* rth
= NULL
;
2065 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2066 if (res
->fi
&& res
->fi
->fib_nhs
> 1 && fl
->oif
== 0)
2067 fib_select_multipath(fl
, res
);
2070 /* create a routing cache entry */
2071 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2075 /* put it into the cache */
2076 hash
= rt_hash(daddr
, saddr
, fl
->iif
,
2077 rt_genid(dev_net(rth
->dst
.dev
)));
2078 return rt_intern_hash(hash
, rth
, NULL
, skb
, fl
->iif
);
2082 * NOTE. We drop all the packets that has local source
2083 * addresses, because every properly looped back packet
2084 * must have correct destination already attached by output routine.
2086 * Such approach solves two big problems:
2087 * 1. Not simplex devices are handled properly.
2088 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2091 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2092 u8 tos
, struct net_device
*dev
)
2094 struct fib_result res
;
2095 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2096 struct flowi fl
= { .nl_u
= { .ip4_u
=
2100 .scope
= RT_SCOPE_UNIVERSE
,
2103 .iif
= dev
->ifindex
};
2106 struct rtable
* rth
;
2111 struct net
* net
= dev_net(dev
);
2113 /* IP on this device is disabled. */
2118 /* Check for the most weird martians, which can be not detected
2122 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2123 ipv4_is_loopback(saddr
))
2124 goto martian_source
;
2126 if (daddr
== htonl(0xFFFFFFFF) || (saddr
== 0 && daddr
== 0))
2129 /* Accept zero addresses only to limited broadcast;
2130 * I even do not know to fix it or not. Waiting for complains :-)
2132 if (ipv4_is_zeronet(saddr
))
2133 goto martian_source
;
2135 if (ipv4_is_lbcast(daddr
) || ipv4_is_zeronet(daddr
) ||
2136 ipv4_is_loopback(daddr
))
2137 goto martian_destination
;
2140 * Now we are ready to route packet.
2142 if ((err
= fib_lookup(net
, &fl
, &res
)) != 0) {
2143 if (!IN_DEV_FORWARD(in_dev
))
2149 RT_CACHE_STAT_INC(in_slow_tot
);
2151 if (res
.type
== RTN_BROADCAST
)
2154 if (res
.type
== RTN_LOCAL
) {
2155 err
= fib_validate_source(saddr
, daddr
, tos
,
2156 net
->loopback_dev
->ifindex
,
2157 dev
, &spec_dst
, &itag
, skb
->mark
);
2159 goto martian_source_keep_err
;
2161 flags
|= RTCF_DIRECTSRC
;
2166 if (!IN_DEV_FORWARD(in_dev
))
2168 if (res
.type
!= RTN_UNICAST
)
2169 goto martian_destination
;
2171 err
= ip_mkroute_input(skb
, &res
, &fl
, in_dev
, daddr
, saddr
, tos
);
2178 if (skb
->protocol
!= htons(ETH_P_IP
))
2181 if (ipv4_is_zeronet(saddr
))
2182 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2184 err
= fib_validate_source(saddr
, 0, tos
, 0, dev
, &spec_dst
,
2187 goto martian_source_keep_err
;
2189 flags
|= RTCF_DIRECTSRC
;
2191 flags
|= RTCF_BROADCAST
;
2192 res
.type
= RTN_BROADCAST
;
2193 RT_CACHE_STAT_INC(in_brd
);
2196 rth
= dst_alloc(&ipv4_dst_ops
);
2200 rth
->dst
.output
= ip_rt_bug
;
2201 rth
->dst
.obsolete
= -1;
2202 rth
->rt_genid
= rt_genid(net
);
2204 atomic_set(&rth
->dst
.__refcnt
, 1);
2205 rth
->dst
.flags
= DST_HOST
;
2206 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2207 rth
->dst
.flags
|= DST_NOPOLICY
;
2208 rth
->fl
.fl4_dst
= daddr
;
2209 rth
->rt_dst
= daddr
;
2210 rth
->fl
.fl4_tos
= tos
;
2211 rth
->fl
.mark
= skb
->mark
;
2212 rth
->fl
.fl4_src
= saddr
;
2213 rth
->rt_src
= saddr
;
2214 #ifdef CONFIG_NET_CLS_ROUTE
2215 rth
->dst
.tclassid
= itag
;
2218 rth
->fl
.iif
= dev
->ifindex
;
2219 rth
->dst
.dev
= net
->loopback_dev
;
2220 dev_hold(rth
->dst
.dev
);
2221 rth
->idev
= in_dev_get(rth
->dst
.dev
);
2222 rth
->rt_gateway
= daddr
;
2223 rth
->rt_spec_dst
= spec_dst
;
2224 rth
->dst
.input
= ip_local_deliver
;
2225 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2226 if (res
.type
== RTN_UNREACHABLE
) {
2227 rth
->dst
.input
= ip_error
;
2228 rth
->dst
.error
= -err
;
2229 rth
->rt_flags
&= ~RTCF_LOCAL
;
2231 rth
->rt_type
= res
.type
;
2232 hash
= rt_hash(daddr
, saddr
, fl
.iif
, rt_genid(net
));
2233 err
= rt_intern_hash(hash
, rth
, NULL
, skb
, fl
.iif
);
2237 RT_CACHE_STAT_INC(in_no_route
);
2238 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2239 res
.type
= RTN_UNREACHABLE
;
2245 * Do not cache martian addresses: they should be logged (RFC1812)
2247 martian_destination
:
2248 RT_CACHE_STAT_INC(in_martian_dst
);
2249 #ifdef CONFIG_IP_ROUTE_VERBOSE
2250 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2251 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2252 &daddr
, &saddr
, dev
->name
);
2256 err
= -EHOSTUNREACH
;
2269 martian_source_keep_err
:
2270 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2274 int ip_route_input_common(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2275 u8 tos
, struct net_device
*dev
, bool noref
)
2277 struct rtable
* rth
;
2279 int iif
= dev
->ifindex
;
2287 if (!rt_caching(net
))
2290 tos
&= IPTOS_RT_MASK
;
2291 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2293 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2294 rth
= rcu_dereference(rth
->dst
.rt_next
)) {
2295 if ((((__force u32
)rth
->fl
.fl4_dst
^ (__force u32
)daddr
) |
2296 ((__force u32
)rth
->fl
.fl4_src
^ (__force u32
)saddr
) |
2297 (rth
->fl
.iif
^ iif
) |
2299 (rth
->fl
.fl4_tos
^ tos
)) == 0 &&
2300 rth
->fl
.mark
== skb
->mark
&&
2301 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2302 !rt_is_expired(rth
)) {
2304 dst_use_noref(&rth
->dst
, jiffies
);
2305 skb_dst_set_noref(skb
, &rth
->dst
);
2307 dst_use(&rth
->dst
, jiffies
);
2308 skb_dst_set(skb
, &rth
->dst
);
2310 RT_CACHE_STAT_INC(in_hit
);
2314 RT_CACHE_STAT_INC(in_hlist_search
);
2318 /* Multicast recognition logic is moved from route cache to here.
2319 The problem was that too many Ethernet cards have broken/missing
2320 hardware multicast filters :-( As result the host on multicasting
2321 network acquires a lot of useless route cache entries, sort of
2322 SDR messages from all the world. Now we try to get rid of them.
2323 Really, provided software IP multicast filter is organized
2324 reasonably (at least, hashed), it does not result in a slowdown
2325 comparing with route cache reject entries.
2326 Note, that multicast routers are not affected, because
2327 route cache entry is created eventually.
2329 if (ipv4_is_multicast(daddr
)) {
2330 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2333 int our
= ip_check_mc(in_dev
, daddr
, saddr
,
2334 ip_hdr(skb
)->protocol
);
2336 #ifdef CONFIG_IP_MROUTE
2338 (!ipv4_is_local_multicast(daddr
) &&
2339 IN_DEV_MFORWARD(in_dev
))
2342 int res
= ip_route_input_mc(skb
, daddr
, saddr
,
2351 res
= ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2355 EXPORT_SYMBOL(ip_route_input_common
);
2357 static int __mkroute_output(struct rtable
**result
,
2358 struct fib_result
*res
,
2359 const struct flowi
*fl
,
2360 const struct flowi
*oldflp
,
2361 struct net_device
*dev_out
,
2365 struct in_device
*in_dev
;
2366 u32 tos
= RT_FL_TOS(oldflp
);
2369 if (ipv4_is_loopback(fl
->fl4_src
) && !(dev_out
->flags
&IFF_LOOPBACK
))
2372 if (fl
->fl4_dst
== htonl(0xFFFFFFFF))
2373 res
->type
= RTN_BROADCAST
;
2374 else if (ipv4_is_multicast(fl
->fl4_dst
))
2375 res
->type
= RTN_MULTICAST
;
2376 else if (ipv4_is_lbcast(fl
->fl4_dst
) || ipv4_is_zeronet(fl
->fl4_dst
))
2379 if (dev_out
->flags
& IFF_LOOPBACK
)
2380 flags
|= RTCF_LOCAL
;
2382 /* get work reference to inet device */
2383 in_dev
= in_dev_get(dev_out
);
2387 if (res
->type
== RTN_BROADCAST
) {
2388 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2390 fib_info_put(res
->fi
);
2393 } else if (res
->type
== RTN_MULTICAST
) {
2394 flags
|= RTCF_MULTICAST
|RTCF_LOCAL
;
2395 if (!ip_check_mc(in_dev
, oldflp
->fl4_dst
, oldflp
->fl4_src
,
2397 flags
&= ~RTCF_LOCAL
;
2398 /* If multicast route do not exist use
2399 default one, but do not gateway in this case.
2402 if (res
->fi
&& res
->prefixlen
< 4) {
2403 fib_info_put(res
->fi
);
2409 rth
= dst_alloc(&ipv4_dst_ops
);
2415 atomic_set(&rth
->dst
.__refcnt
, 1);
2416 rth
->dst
.flags
= DST_HOST
;
2417 if (IN_DEV_CONF_GET(in_dev
, NOXFRM
))
2418 rth
->dst
.flags
|= DST_NOXFRM
;
2419 if (IN_DEV_CONF_GET(in_dev
, NOPOLICY
))
2420 rth
->dst
.flags
|= DST_NOPOLICY
;
2422 rth
->fl
.fl4_dst
= oldflp
->fl4_dst
;
2423 rth
->fl
.fl4_tos
= tos
;
2424 rth
->fl
.fl4_src
= oldflp
->fl4_src
;
2425 rth
->fl
.oif
= oldflp
->oif
;
2426 rth
->fl
.mark
= oldflp
->mark
;
2427 rth
->rt_dst
= fl
->fl4_dst
;
2428 rth
->rt_src
= fl
->fl4_src
;
2429 rth
->rt_iif
= oldflp
->oif
? : dev_out
->ifindex
;
2430 /* get references to the devices that are to be hold by the routing
2432 rth
->dst
.dev
= dev_out
;
2434 rth
->idev
= in_dev_get(dev_out
);
2435 rth
->rt_gateway
= fl
->fl4_dst
;
2436 rth
->rt_spec_dst
= fl
->fl4_src
;
2438 rth
->dst
.output
=ip_output
;
2439 rth
->dst
.obsolete
= -1;
2440 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2442 RT_CACHE_STAT_INC(out_slow_tot
);
2444 if (flags
& RTCF_LOCAL
) {
2445 rth
->dst
.input
= ip_local_deliver
;
2446 rth
->rt_spec_dst
= fl
->fl4_dst
;
2448 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2449 rth
->rt_spec_dst
= fl
->fl4_src
;
2450 if (flags
& RTCF_LOCAL
&&
2451 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2452 rth
->dst
.output
= ip_mc_output
;
2453 RT_CACHE_STAT_INC(out_slow_mc
);
2455 #ifdef CONFIG_IP_MROUTE
2456 if (res
->type
== RTN_MULTICAST
) {
2457 if (IN_DEV_MFORWARD(in_dev
) &&
2458 !ipv4_is_local_multicast(oldflp
->fl4_dst
)) {
2459 rth
->dst
.input
= ip_mr_input
;
2460 rth
->dst
.output
= ip_mc_output
;
2466 rt_set_nexthop(rth
, res
, 0);
2468 rth
->rt_flags
= flags
;
2472 /* release work reference to inet device */
2478 static int ip_mkroute_output(struct rtable
**rp
,
2479 struct fib_result
*res
,
2480 const struct flowi
*fl
,
2481 const struct flowi
*oldflp
,
2482 struct net_device
*dev_out
,
2485 struct rtable
*rth
= NULL
;
2486 int err
= __mkroute_output(&rth
, res
, fl
, oldflp
, dev_out
, flags
);
2489 hash
= rt_hash(oldflp
->fl4_dst
, oldflp
->fl4_src
, oldflp
->oif
,
2490 rt_genid(dev_net(dev_out
)));
2491 err
= rt_intern_hash(hash
, rth
, rp
, NULL
, oldflp
->oif
);
2498 * Major route resolver routine.
2501 static int ip_route_output_slow(struct net
*net
, struct rtable
**rp
,
2502 const struct flowi
*oldflp
)
2504 u32 tos
= RT_FL_TOS(oldflp
);
2505 struct flowi fl
= { .nl_u
= { .ip4_u
=
2506 { .daddr
= oldflp
->fl4_dst
,
2507 .saddr
= oldflp
->fl4_src
,
2508 .tos
= tos
& IPTOS_RT_MASK
,
2509 .scope
= ((tos
& RTO_ONLINK
) ?
2513 .mark
= oldflp
->mark
,
2514 .iif
= net
->loopback_dev
->ifindex
,
2515 .oif
= oldflp
->oif
};
2516 struct fib_result res
;
2518 struct net_device
*dev_out
= NULL
;
2524 #ifdef CONFIG_IP_MULTIPLE_TABLES
2528 if (oldflp
->fl4_src
) {
2530 if (ipv4_is_multicast(oldflp
->fl4_src
) ||
2531 ipv4_is_lbcast(oldflp
->fl4_src
) ||
2532 ipv4_is_zeronet(oldflp
->fl4_src
))
2535 /* I removed check for oif == dev_out->oif here.
2536 It was wrong for two reasons:
2537 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2538 is assigned to multiple interfaces.
2539 2. Moreover, we are allowed to send packets with saddr
2540 of another iface. --ANK
2543 if (oldflp
->oif
== 0 &&
2544 (ipv4_is_multicast(oldflp
->fl4_dst
) ||
2545 oldflp
->fl4_dst
== htonl(0xFFFFFFFF))) {
2546 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2547 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2548 if (dev_out
== NULL
)
2551 /* Special hack: user can direct multicasts
2552 and limited broadcast via necessary interface
2553 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2554 This hack is not just for fun, it allows
2555 vic,vat and friends to work.
2556 They bind socket to loopback, set ttl to zero
2557 and expect that it will work.
2558 From the viewpoint of routing cache they are broken,
2559 because we are not allowed to build multicast path
2560 with loopback source addr (look, routing cache
2561 cannot know, that ttl is zero, so that packet
2562 will not leave this host and route is valid).
2563 Luckily, this hack is good workaround.
2566 fl
.oif
= dev_out
->ifindex
;
2570 if (!(oldflp
->flags
& FLOWI_FLAG_ANYSRC
)) {
2571 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2572 dev_out
= ip_dev_find(net
, oldflp
->fl4_src
);
2573 if (dev_out
== NULL
)
2582 dev_out
= dev_get_by_index(net
, oldflp
->oif
);
2584 if (dev_out
== NULL
)
2587 /* RACE: Check return value of inet_select_addr instead. */
2588 if (__in_dev_get_rtnl(dev_out
) == NULL
) {
2590 goto out
; /* Wrong error code */
2593 if (ipv4_is_local_multicast(oldflp
->fl4_dst
) ||
2594 oldflp
->fl4_dst
== htonl(0xFFFFFFFF)) {
2596 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2601 if (ipv4_is_multicast(oldflp
->fl4_dst
))
2602 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2604 else if (!oldflp
->fl4_dst
)
2605 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2611 fl
.fl4_dst
= fl
.fl4_src
;
2613 fl
.fl4_dst
= fl
.fl4_src
= htonl(INADDR_LOOPBACK
);
2616 dev_out
= net
->loopback_dev
;
2618 fl
.oif
= net
->loopback_dev
->ifindex
;
2619 res
.type
= RTN_LOCAL
;
2620 flags
|= RTCF_LOCAL
;
2624 if (fib_lookup(net
, &fl
, &res
)) {
2627 /* Apparently, routing tables are wrong. Assume,
2628 that the destination is on link.
2631 Because we are allowed to send to iface
2632 even if it has NO routes and NO assigned
2633 addresses. When oif is specified, routing
2634 tables are looked up with only one purpose:
2635 to catch if destination is gatewayed, rather than
2636 direct. Moreover, if MSG_DONTROUTE is set,
2637 we send packet, ignoring both routing tables
2638 and ifaddr state. --ANK
2641 We could make it even if oif is unknown,
2642 likely IPv6, but we do not.
2645 if (fl
.fl4_src
== 0)
2646 fl
.fl4_src
= inet_select_addr(dev_out
, 0,
2648 res
.type
= RTN_UNICAST
;
2658 if (res
.type
== RTN_LOCAL
) {
2660 fl
.fl4_src
= fl
.fl4_dst
;
2663 dev_out
= net
->loopback_dev
;
2665 fl
.oif
= dev_out
->ifindex
;
2667 fib_info_put(res
.fi
);
2669 flags
|= RTCF_LOCAL
;
2673 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2674 if (res
.fi
->fib_nhs
> 1 && fl
.oif
== 0)
2675 fib_select_multipath(&fl
, &res
);
2678 if (!res
.prefixlen
&& res
.type
== RTN_UNICAST
&& !fl
.oif
)
2679 fib_select_default(net
, &fl
, &res
);
2682 fl
.fl4_src
= FIB_RES_PREFSRC(res
);
2686 dev_out
= FIB_RES_DEV(res
);
2688 fl
.oif
= dev_out
->ifindex
;
2692 err
= ip_mkroute_output(rp
, &res
, &fl
, oldflp
, dev_out
, flags
);
2702 int __ip_route_output_key(struct net
*net
, struct rtable
**rp
,
2703 const struct flowi
*flp
)
2708 if (!rt_caching(net
))
2711 hash
= rt_hash(flp
->fl4_dst
, flp
->fl4_src
, flp
->oif
, rt_genid(net
));
2714 for (rth
= rcu_dereference_bh(rt_hash_table
[hash
].chain
); rth
;
2715 rth
= rcu_dereference_bh(rth
->dst
.rt_next
)) {
2716 if (rth
->fl
.fl4_dst
== flp
->fl4_dst
&&
2717 rth
->fl
.fl4_src
== flp
->fl4_src
&&
2719 rth
->fl
.oif
== flp
->oif
&&
2720 rth
->fl
.mark
== flp
->mark
&&
2721 !((rth
->fl
.fl4_tos
^ flp
->fl4_tos
) &
2722 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2723 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2724 !rt_is_expired(rth
)) {
2725 dst_use(&rth
->dst
, jiffies
);
2726 RT_CACHE_STAT_INC(out_hit
);
2727 rcu_read_unlock_bh();
2731 RT_CACHE_STAT_INC(out_hlist_search
);
2733 rcu_read_unlock_bh();
2736 return ip_route_output_slow(net
, rp
, flp
);
2739 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2741 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2745 static struct dst_ops ipv4_dst_blackhole_ops
= {
2747 .protocol
= cpu_to_be16(ETH_P_IP
),
2748 .destroy
= ipv4_dst_destroy
,
2749 .check
= ipv4_dst_check
,
2750 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2751 .entries
= ATOMIC_INIT(0),
2755 static int ipv4_dst_blackhole(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2757 struct rtable
*ort
= *rp
;
2758 struct rtable
*rt
= (struct rtable
*)
2759 dst_alloc(&ipv4_dst_blackhole_ops
);
2762 struct dst_entry
*new = &rt
->dst
;
2764 atomic_set(&new->__refcnt
, 1);
2766 new->input
= dst_discard
;
2767 new->output
= dst_discard
;
2768 memcpy(new->metrics
, ort
->dst
.metrics
, RTAX_MAX
*sizeof(u32
));
2770 new->dev
= ort
->dst
.dev
;
2776 rt
->idev
= ort
->idev
;
2778 in_dev_hold(rt
->idev
);
2779 rt
->rt_genid
= rt_genid(net
);
2780 rt
->rt_flags
= ort
->rt_flags
;
2781 rt
->rt_type
= ort
->rt_type
;
2782 rt
->rt_dst
= ort
->rt_dst
;
2783 rt
->rt_src
= ort
->rt_src
;
2784 rt
->rt_iif
= ort
->rt_iif
;
2785 rt
->rt_gateway
= ort
->rt_gateway
;
2786 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2787 rt
->peer
= ort
->peer
;
2789 atomic_inc(&rt
->peer
->refcnt
);
2794 dst_release(&(*rp
)->dst
);
2796 return (rt
? 0 : -ENOMEM
);
2799 int ip_route_output_flow(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
,
2800 struct sock
*sk
, int flags
)
2804 if ((err
= __ip_route_output_key(net
, rp
, flp
)) != 0)
2809 flp
->fl4_src
= (*rp
)->rt_src
;
2811 flp
->fl4_dst
= (*rp
)->rt_dst
;
2812 err
= __xfrm_lookup(net
, (struct dst_entry
**)rp
, flp
, sk
,
2813 flags
? XFRM_LOOKUP_WAIT
: 0);
2814 if (err
== -EREMOTE
)
2815 err
= ipv4_dst_blackhole(net
, rp
, flp
);
2823 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2825 int ip_route_output_key(struct net
*net
, struct rtable
**rp
, struct flowi
*flp
)
2827 return ip_route_output_flow(net
, rp
, flp
, NULL
, 0);
2830 static int rt_fill_info(struct net
*net
,
2831 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2832 int nowait
, unsigned int flags
)
2834 struct rtable
*rt
= skb_rtable(skb
);
2836 struct nlmsghdr
*nlh
;
2838 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2840 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2844 r
= nlmsg_data(nlh
);
2845 r
->rtm_family
= AF_INET
;
2846 r
->rtm_dst_len
= 32;
2848 r
->rtm_tos
= rt
->fl
.fl4_tos
;
2849 r
->rtm_table
= RT_TABLE_MAIN
;
2850 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2851 r
->rtm_type
= rt
->rt_type
;
2852 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2853 r
->rtm_protocol
= RTPROT_UNSPEC
;
2854 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2855 if (rt
->rt_flags
& RTCF_NOTIFY
)
2856 r
->rtm_flags
|= RTM_F_NOTIFY
;
2858 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2860 if (rt
->fl
.fl4_src
) {
2861 r
->rtm_src_len
= 32;
2862 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->fl
.fl4_src
);
2865 NLA_PUT_U32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
);
2866 #ifdef CONFIG_NET_CLS_ROUTE
2867 if (rt
->dst
.tclassid
)
2868 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->dst
.tclassid
);
2871 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
2872 else if (rt
->rt_src
!= rt
->fl
.fl4_src
)
2873 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
2875 if (rt
->rt_dst
!= rt
->rt_gateway
)
2876 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
2878 if (rtnetlink_put_metrics(skb
, rt
->dst
.metrics
) < 0)
2879 goto nla_put_failure
;
2881 error
= rt
->dst
.error
;
2882 expires
= rt
->dst
.expires
? rt
->dst
.expires
- jiffies
: 0;
2884 inet_peer_refcheck(rt
->peer
);
2885 id
= atomic_read(&rt
->peer
->ip_id_count
) & 0xffff;
2886 if (rt
->peer
->tcp_ts_stamp
) {
2887 ts
= rt
->peer
->tcp_ts
;
2888 tsage
= get_seconds() - rt
->peer
->tcp_ts_stamp
;
2893 #ifdef CONFIG_IP_MROUTE
2894 __be32 dst
= rt
->rt_dst
;
2896 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2897 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2898 int err
= ipmr_get_route(net
, skb
, r
, nowait
);
2903 goto nla_put_failure
;
2905 if (err
== -EMSGSIZE
)
2906 goto nla_put_failure
;
2912 NLA_PUT_U32(skb
, RTA_IIF
, rt
->fl
.iif
);
2915 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, id
, ts
, tsage
,
2916 expires
, error
) < 0)
2917 goto nla_put_failure
;
2919 return nlmsg_end(skb
, nlh
);
2922 nlmsg_cancel(skb
, nlh
);
2926 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
2928 struct net
*net
= sock_net(in_skb
->sk
);
2930 struct nlattr
*tb
[RTA_MAX
+1];
2931 struct rtable
*rt
= NULL
;
2936 struct sk_buff
*skb
;
2938 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2942 rtm
= nlmsg_data(nlh
);
2944 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2950 /* Reserve room for dummy headers, this skb can pass
2951 through good chunk of routing engine.
2953 skb_reset_mac_header(skb
);
2954 skb_reset_network_header(skb
);
2956 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2957 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
2958 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2960 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
2961 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
2962 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2965 struct net_device
*dev
;
2967 dev
= __dev_get_by_index(net
, iif
);
2973 skb
->protocol
= htons(ETH_P_IP
);
2976 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
2979 rt
= skb_rtable(skb
);
2980 if (err
== 0 && rt
->dst
.error
)
2981 err
= -rt
->dst
.error
;
2988 .tos
= rtm
->rtm_tos
,
2991 .oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
2993 err
= ip_route_output_key(net
, &rt
, &fl
);
2999 skb_dst_set(skb
, &rt
->dst
);
3000 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
3001 rt
->rt_flags
|= RTCF_NOTIFY
;
3003 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
3004 RTM_NEWROUTE
, 0, 0);
3008 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
3017 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3024 net
= sock_net(skb
->sk
);
3029 s_idx
= idx
= cb
->args
[1];
3030 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3031 if (!rt_hash_table
[h
].chain
)
3034 for (rt
= rcu_dereference_bh(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3035 rt
= rcu_dereference_bh(rt
->dst
.rt_next
), idx
++) {
3036 if (!net_eq(dev_net(rt
->dst
.dev
), net
) || idx
< s_idx
)
3038 if (rt_is_expired(rt
))
3040 skb_dst_set_noref(skb
, &rt
->dst
);
3041 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3042 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3043 1, NLM_F_MULTI
) <= 0) {
3045 rcu_read_unlock_bh();
3050 rcu_read_unlock_bh();
3059 void ip_rt_multicast_event(struct in_device
*in_dev
)
3061 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3064 #ifdef CONFIG_SYSCTL
3065 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3066 void __user
*buffer
,
3067 size_t *lenp
, loff_t
*ppos
)
3074 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3075 ctl
.data
= &flush_delay
;
3076 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3078 net
= (struct net
*)__ctl
->extra1
;
3079 rt_cache_flush(net
, flush_delay
);
3086 static ctl_table ipv4_route_table
[] = {
3088 .procname
= "gc_thresh",
3089 .data
= &ipv4_dst_ops
.gc_thresh
,
3090 .maxlen
= sizeof(int),
3092 .proc_handler
= proc_dointvec
,
3095 .procname
= "max_size",
3096 .data
= &ip_rt_max_size
,
3097 .maxlen
= sizeof(int),
3099 .proc_handler
= proc_dointvec
,
3102 /* Deprecated. Use gc_min_interval_ms */
3104 .procname
= "gc_min_interval",
3105 .data
= &ip_rt_gc_min_interval
,
3106 .maxlen
= sizeof(int),
3108 .proc_handler
= proc_dointvec_jiffies
,
3111 .procname
= "gc_min_interval_ms",
3112 .data
= &ip_rt_gc_min_interval
,
3113 .maxlen
= sizeof(int),
3115 .proc_handler
= proc_dointvec_ms_jiffies
,
3118 .procname
= "gc_timeout",
3119 .data
= &ip_rt_gc_timeout
,
3120 .maxlen
= sizeof(int),
3122 .proc_handler
= proc_dointvec_jiffies
,
3125 .procname
= "gc_interval",
3126 .data
= &ip_rt_gc_interval
,
3127 .maxlen
= sizeof(int),
3129 .proc_handler
= proc_dointvec_jiffies
,
3132 .procname
= "redirect_load",
3133 .data
= &ip_rt_redirect_load
,
3134 .maxlen
= sizeof(int),
3136 .proc_handler
= proc_dointvec
,
3139 .procname
= "redirect_number",
3140 .data
= &ip_rt_redirect_number
,
3141 .maxlen
= sizeof(int),
3143 .proc_handler
= proc_dointvec
,
3146 .procname
= "redirect_silence",
3147 .data
= &ip_rt_redirect_silence
,
3148 .maxlen
= sizeof(int),
3150 .proc_handler
= proc_dointvec
,
3153 .procname
= "error_cost",
3154 .data
= &ip_rt_error_cost
,
3155 .maxlen
= sizeof(int),
3157 .proc_handler
= proc_dointvec
,
3160 .procname
= "error_burst",
3161 .data
= &ip_rt_error_burst
,
3162 .maxlen
= sizeof(int),
3164 .proc_handler
= proc_dointvec
,
3167 .procname
= "gc_elasticity",
3168 .data
= &ip_rt_gc_elasticity
,
3169 .maxlen
= sizeof(int),
3171 .proc_handler
= proc_dointvec
,
3174 .procname
= "mtu_expires",
3175 .data
= &ip_rt_mtu_expires
,
3176 .maxlen
= sizeof(int),
3178 .proc_handler
= proc_dointvec_jiffies
,
3181 .procname
= "min_pmtu",
3182 .data
= &ip_rt_min_pmtu
,
3183 .maxlen
= sizeof(int),
3185 .proc_handler
= proc_dointvec
,
3188 .procname
= "min_adv_mss",
3189 .data
= &ip_rt_min_advmss
,
3190 .maxlen
= sizeof(int),
3192 .proc_handler
= proc_dointvec
,
3197 static struct ctl_table empty
[1];
3199 static struct ctl_table ipv4_skeleton
[] =
3201 { .procname
= "route",
3202 .mode
= 0555, .child
= ipv4_route_table
},
3203 { .procname
= "neigh",
3204 .mode
= 0555, .child
= empty
},
3208 static __net_initdata
struct ctl_path ipv4_path
[] = {
3209 { .procname
= "net", },
3210 { .procname
= "ipv4", },
3214 static struct ctl_table ipv4_route_flush_table
[] = {
3216 .procname
= "flush",
3217 .maxlen
= sizeof(int),
3219 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3224 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3225 { .procname
= "net", },
3226 { .procname
= "ipv4", },
3227 { .procname
= "route", },
3231 static __net_init
int sysctl_route_net_init(struct net
*net
)
3233 struct ctl_table
*tbl
;
3235 tbl
= ipv4_route_flush_table
;
3236 if (!net_eq(net
, &init_net
)) {
3237 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3241 tbl
[0].extra1
= net
;
3243 net
->ipv4
.route_hdr
=
3244 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3245 if (net
->ipv4
.route_hdr
== NULL
)
3250 if (tbl
!= ipv4_route_flush_table
)
3256 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3258 struct ctl_table
*tbl
;
3260 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3261 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3262 BUG_ON(tbl
== ipv4_route_flush_table
);
3266 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3267 .init
= sysctl_route_net_init
,
3268 .exit
= sysctl_route_net_exit
,
3272 static __net_init
int rt_genid_init(struct net
*net
)
3274 get_random_bytes(&net
->ipv4
.rt_genid
,
3275 sizeof(net
->ipv4
.rt_genid
));
3279 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3280 .init
= rt_genid_init
,
3284 #ifdef CONFIG_NET_CLS_ROUTE
3285 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3286 #endif /* CONFIG_NET_CLS_ROUTE */
3288 static __initdata
unsigned long rhash_entries
;
3289 static int __init
set_rhash_entries(char *str
)
3293 rhash_entries
= simple_strtoul(str
, &str
, 0);
3296 __setup("rhash_entries=", set_rhash_entries
);
3298 int __init
ip_rt_init(void)
3302 #ifdef CONFIG_NET_CLS_ROUTE
3303 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3305 panic("IP: failed to allocate ip_rt_acct\n");
3308 ipv4_dst_ops
.kmem_cachep
=
3309 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3310 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3312 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3314 rt_hash_table
= (struct rt_hash_bucket
*)
3315 alloc_large_system_hash("IP route cache",
3316 sizeof(struct rt_hash_bucket
),
3318 (totalram_pages
>= 128 * 1024) ?
3323 rhash_entries
? 0 : 512 * 1024);
3324 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3325 rt_hash_lock_init();
3327 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3328 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3333 /* All the timers, started at system startup tend
3334 to synchronize. Perturb it a bit.
3336 INIT_DELAYED_WORK_DEFERRABLE(&expires_work
, rt_worker_func
);
3337 expires_ljiffies
= jiffies
;
3338 schedule_delayed_work(&expires_work
,
3339 net_random() % ip_rt_gc_interval
+ ip_rt_gc_interval
);
3341 if (ip_rt_proc_init())
3342 printk(KERN_ERR
"Unable to create route proc files\n");
3345 xfrm4_init(ip_rt_max_size
);
3347 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
);
3349 #ifdef CONFIG_SYSCTL
3350 register_pernet_subsys(&sysctl_route_ops
);
3352 register_pernet_subsys(&rt_genid_ops
);
3356 #ifdef CONFIG_SYSCTL
3358 * We really need to sanitize the damn ipv4 init order, then all
3359 * this nonsense will go away.
3361 void __init
ip_static_sysctl_init(void)
3363 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);
3367 EXPORT_SYMBOL(__ip_select_ident
);
3368 EXPORT_SYMBOL(ip_route_output_key
);