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(oldflp4) \
113 ((u32)(oldflp4->flowi4_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;
135 * Interface to generic destination cache.
138 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
139 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
);
140 static unsigned int ipv4_default_mtu(const struct dst_entry
*dst
);
141 static void ipv4_dst_destroy(struct dst_entry
*dst
);
142 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
);
143 static void ipv4_link_failure(struct sk_buff
*skb
);
144 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
);
145 static int rt_garbage_collect(struct dst_ops
*ops
);
147 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
152 static u32
*ipv4_cow_metrics(struct dst_entry
*dst
, unsigned long old
)
154 struct rtable
*rt
= (struct rtable
*) dst
;
155 struct inet_peer
*peer
;
163 u32
*old_p
= __DST_METRICS_PTR(old
);
164 unsigned long prev
, new;
167 if (inet_metrics_new(peer
))
168 memcpy(p
, old_p
, sizeof(u32
) * RTAX_MAX
);
170 new = (unsigned long) p
;
171 prev
= cmpxchg(&dst
->_metrics
, old
, new);
174 p
= __DST_METRICS_PTR(prev
);
175 if (prev
& DST_METRICS_READ_ONLY
)
179 fib_info_put(rt
->fi
);
187 static struct dst_ops ipv4_dst_ops
= {
189 .protocol
= cpu_to_be16(ETH_P_IP
),
190 .gc
= rt_garbage_collect
,
191 .check
= ipv4_dst_check
,
192 .default_advmss
= ipv4_default_advmss
,
193 .default_mtu
= ipv4_default_mtu
,
194 .cow_metrics
= ipv4_cow_metrics
,
195 .destroy
= ipv4_dst_destroy
,
196 .ifdown
= ipv4_dst_ifdown
,
197 .negative_advice
= ipv4_negative_advice
,
198 .link_failure
= ipv4_link_failure
,
199 .update_pmtu
= ip_rt_update_pmtu
,
200 .local_out
= __ip_local_out
,
203 #define ECN_OR_COST(class) TC_PRIO_##class
205 const __u8 ip_tos2prio
[16] = {
207 ECN_OR_COST(BESTEFFORT
),
209 ECN_OR_COST(BESTEFFORT
),
215 ECN_OR_COST(INTERACTIVE
),
217 ECN_OR_COST(INTERACTIVE
),
218 TC_PRIO_INTERACTIVE_BULK
,
219 ECN_OR_COST(INTERACTIVE_BULK
),
220 TC_PRIO_INTERACTIVE_BULK
,
221 ECN_OR_COST(INTERACTIVE_BULK
)
229 /* The locking scheme is rather straight forward:
231 * 1) Read-Copy Update protects the buckets of the central route hash.
232 * 2) Only writers remove entries, and they hold the lock
233 * as they look at rtable reference counts.
234 * 3) Only readers acquire references to rtable entries,
235 * they do so with atomic increments and with the
239 struct rt_hash_bucket
{
240 struct rtable __rcu
*chain
;
243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
244 defined(CONFIG_PROVE_LOCKING)
246 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
247 * The size of this table is a power of two and depends on the number of CPUS.
248 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
250 #ifdef CONFIG_LOCKDEP
251 # define RT_HASH_LOCK_SZ 256
254 # define RT_HASH_LOCK_SZ 4096
256 # define RT_HASH_LOCK_SZ 2048
258 # define RT_HASH_LOCK_SZ 1024
260 # define RT_HASH_LOCK_SZ 512
262 # define RT_HASH_LOCK_SZ 256
266 static spinlock_t
*rt_hash_locks
;
267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
269 static __init
void rt_hash_lock_init(void)
273 rt_hash_locks
= kmalloc(sizeof(spinlock_t
) * RT_HASH_LOCK_SZ
,
276 panic("IP: failed to allocate rt_hash_locks\n");
278 for (i
= 0; i
< RT_HASH_LOCK_SZ
; i
++)
279 spin_lock_init(&rt_hash_locks
[i
]);
282 # define rt_hash_lock_addr(slot) NULL
284 static inline void rt_hash_lock_init(void)
289 static struct rt_hash_bucket
*rt_hash_table __read_mostly
;
290 static unsigned rt_hash_mask __read_mostly
;
291 static unsigned int rt_hash_log __read_mostly
;
293 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
296 static inline unsigned int rt_hash(__be32 daddr
, __be32 saddr
, int idx
,
299 return jhash_3words((__force u32
)daddr
, (__force u32
)saddr
,
304 static inline int rt_genid(struct net
*net
)
306 return atomic_read(&net
->ipv4
.rt_genid
);
309 #ifdef CONFIG_PROC_FS
310 struct rt_cache_iter_state
{
311 struct seq_net_private p
;
316 static struct rtable
*rt_cache_get_first(struct seq_file
*seq
)
318 struct rt_cache_iter_state
*st
= seq
->private;
319 struct rtable
*r
= NULL
;
321 for (st
->bucket
= rt_hash_mask
; st
->bucket
>= 0; --st
->bucket
) {
322 if (!rcu_dereference_raw(rt_hash_table
[st
->bucket
].chain
))
325 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
327 if (dev_net(r
->dst
.dev
) == seq_file_net(seq
) &&
328 r
->rt_genid
== st
->genid
)
330 r
= rcu_dereference_bh(r
->dst
.rt_next
);
332 rcu_read_unlock_bh();
337 static struct rtable
*__rt_cache_get_next(struct seq_file
*seq
,
340 struct rt_cache_iter_state
*st
= seq
->private;
342 r
= rcu_dereference_bh(r
->dst
.rt_next
);
344 rcu_read_unlock_bh();
346 if (--st
->bucket
< 0)
348 } while (!rcu_dereference_raw(rt_hash_table
[st
->bucket
].chain
));
350 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
355 static struct rtable
*rt_cache_get_next(struct seq_file
*seq
,
358 struct rt_cache_iter_state
*st
= seq
->private;
359 while ((r
= __rt_cache_get_next(seq
, r
)) != NULL
) {
360 if (dev_net(r
->dst
.dev
) != seq_file_net(seq
))
362 if (r
->rt_genid
== st
->genid
)
368 static struct rtable
*rt_cache_get_idx(struct seq_file
*seq
, loff_t pos
)
370 struct rtable
*r
= rt_cache_get_first(seq
);
373 while (pos
&& (r
= rt_cache_get_next(seq
, r
)))
375 return pos
? NULL
: r
;
378 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
380 struct rt_cache_iter_state
*st
= seq
->private;
382 return rt_cache_get_idx(seq
, *pos
- 1);
383 st
->genid
= rt_genid(seq_file_net(seq
));
384 return SEQ_START_TOKEN
;
387 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
391 if (v
== SEQ_START_TOKEN
)
392 r
= rt_cache_get_first(seq
);
394 r
= rt_cache_get_next(seq
, v
);
399 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
401 if (v
&& v
!= SEQ_START_TOKEN
)
402 rcu_read_unlock_bh();
405 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
407 if (v
== SEQ_START_TOKEN
)
408 seq_printf(seq
, "%-127s\n",
409 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
410 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
413 struct rtable
*r
= v
;
416 seq_printf(seq
, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
417 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
418 r
->dst
.dev
? r
->dst
.dev
->name
: "*",
419 (__force u32
)r
->rt_dst
,
420 (__force u32
)r
->rt_gateway
,
421 r
->rt_flags
, atomic_read(&r
->dst
.__refcnt
),
422 r
->dst
.__use
, 0, (__force u32
)r
->rt_src
,
423 dst_metric_advmss(&r
->dst
) + 40,
424 dst_metric(&r
->dst
, RTAX_WINDOW
),
425 (int)((dst_metric(&r
->dst
, RTAX_RTT
) >> 3) +
426 dst_metric(&r
->dst
, RTAX_RTTVAR
)),
428 r
->dst
.hh
? atomic_read(&r
->dst
.hh
->hh_refcnt
) : -1,
429 r
->dst
.hh
? (r
->dst
.hh
->hh_output
==
431 r
->rt_spec_dst
, &len
);
433 seq_printf(seq
, "%*s\n", 127 - len
, "");
438 static const struct seq_operations rt_cache_seq_ops
= {
439 .start
= rt_cache_seq_start
,
440 .next
= rt_cache_seq_next
,
441 .stop
= rt_cache_seq_stop
,
442 .show
= rt_cache_seq_show
,
445 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
447 return seq_open_net(inode
, file
, &rt_cache_seq_ops
,
448 sizeof(struct rt_cache_iter_state
));
451 static const struct file_operations rt_cache_seq_fops
= {
452 .owner
= THIS_MODULE
,
453 .open
= rt_cache_seq_open
,
456 .release
= seq_release_net
,
460 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
465 return SEQ_START_TOKEN
;
467 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
468 if (!cpu_possible(cpu
))
471 return &per_cpu(rt_cache_stat
, cpu
);
476 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
480 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
481 if (!cpu_possible(cpu
))
484 return &per_cpu(rt_cache_stat
, cpu
);
490 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
495 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
497 struct rt_cache_stat
*st
= v
;
499 if (v
== SEQ_START_TOKEN
) {
500 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");
504 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
505 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
506 dst_entries_get_slow(&ipv4_dst_ops
),
529 static const struct seq_operations rt_cpu_seq_ops
= {
530 .start
= rt_cpu_seq_start
,
531 .next
= rt_cpu_seq_next
,
532 .stop
= rt_cpu_seq_stop
,
533 .show
= rt_cpu_seq_show
,
537 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
539 return seq_open(file
, &rt_cpu_seq_ops
);
542 static const struct file_operations rt_cpu_seq_fops
= {
543 .owner
= THIS_MODULE
,
544 .open
= rt_cpu_seq_open
,
547 .release
= seq_release
,
550 #ifdef CONFIG_IP_ROUTE_CLASSID
551 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
553 struct ip_rt_acct
*dst
, *src
;
556 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
560 for_each_possible_cpu(i
) {
561 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
562 for (j
= 0; j
< 256; j
++) {
563 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
564 dst
[j
].o_packets
+= src
[j
].o_packets
;
565 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
566 dst
[j
].i_packets
+= src
[j
].i_packets
;
570 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
575 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
577 return single_open(file
, rt_acct_proc_show
, NULL
);
580 static const struct file_operations rt_acct_proc_fops
= {
581 .owner
= THIS_MODULE
,
582 .open
= rt_acct_proc_open
,
585 .release
= single_release
,
589 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
591 struct proc_dir_entry
*pde
;
593 pde
= proc_net_fops_create(net
, "rt_cache", S_IRUGO
,
598 pde
= proc_create("rt_cache", S_IRUGO
,
599 net
->proc_net_stat
, &rt_cpu_seq_fops
);
603 #ifdef CONFIG_IP_ROUTE_CLASSID
604 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
610 #ifdef CONFIG_IP_ROUTE_CLASSID
612 remove_proc_entry("rt_cache", net
->proc_net_stat
);
615 remove_proc_entry("rt_cache", net
->proc_net
);
620 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
622 remove_proc_entry("rt_cache", net
->proc_net_stat
);
623 remove_proc_entry("rt_cache", net
->proc_net
);
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625 remove_proc_entry("rt_acct", net
->proc_net
);
629 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
630 .init
= ip_rt_do_proc_init
,
631 .exit
= ip_rt_do_proc_exit
,
634 static int __init
ip_rt_proc_init(void)
636 return register_pernet_subsys(&ip_rt_proc_ops
);
640 static inline int ip_rt_proc_init(void)
644 #endif /* CONFIG_PROC_FS */
646 static inline void rt_free(struct rtable
*rt
)
648 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
651 static inline void rt_drop(struct rtable
*rt
)
654 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
657 static inline int rt_fast_clean(struct rtable
*rth
)
659 /* Kill broadcast/multicast entries very aggresively, if they
660 collide in hash table with more useful entries */
661 return (rth
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) &&
662 rt_is_input_route(rth
) && rth
->dst
.rt_next
;
665 static inline int rt_valuable(struct rtable
*rth
)
667 return (rth
->rt_flags
& (RTCF_REDIRECTED
| RTCF_NOTIFY
)) ||
668 (rth
->peer
&& rth
->peer
->pmtu_expires
);
671 static int rt_may_expire(struct rtable
*rth
, unsigned long tmo1
, unsigned long tmo2
)
676 if (atomic_read(&rth
->dst
.__refcnt
))
679 age
= jiffies
- rth
->dst
.lastuse
;
680 if ((age
<= tmo1
&& !rt_fast_clean(rth
)) ||
681 (age
<= tmo2
&& rt_valuable(rth
)))
687 /* Bits of score are:
689 * 30: not quite useless
690 * 29..0: usage counter
692 static inline u32
rt_score(struct rtable
*rt
)
694 u32 score
= jiffies
- rt
->dst
.lastuse
;
696 score
= ~score
& ~(3<<30);
701 if (rt_is_output_route(rt
) ||
702 !(rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
|RTCF_LOCAL
)))
708 static inline bool rt_caching(const struct net
*net
)
710 return net
->ipv4
.current_rt_cache_rebuild_count
<=
711 net
->ipv4
.sysctl_rt_cache_rebuild_count
;
714 static inline bool compare_hash_inputs(const struct rtable
*rt1
,
715 const struct rtable
*rt2
)
717 return ((((__force u32
)rt1
->rt_key_dst
^ (__force u32
)rt2
->rt_key_dst
) |
718 ((__force u32
)rt1
->rt_key_src
^ (__force u32
)rt2
->rt_key_src
) |
719 (rt1
->rt_iif
^ rt2
->rt_iif
)) == 0);
722 static inline int compare_keys(struct rtable
*rt1
, struct rtable
*rt2
)
724 return (((__force u32
)rt1
->rt_key_dst
^ (__force u32
)rt2
->rt_key_dst
) |
725 ((__force u32
)rt1
->rt_key_src
^ (__force u32
)rt2
->rt_key_src
) |
726 (rt1
->rt_mark
^ rt2
->rt_mark
) |
727 (rt1
->rt_key_tos
^ rt2
->rt_key_tos
) |
728 (rt1
->rt_oif
^ rt2
->rt_oif
) |
729 (rt1
->rt_iif
^ rt2
->rt_iif
)) == 0;
732 static inline int compare_netns(struct rtable
*rt1
, struct rtable
*rt2
)
734 return net_eq(dev_net(rt1
->dst
.dev
), dev_net(rt2
->dst
.dev
));
737 static inline int rt_is_expired(struct rtable
*rth
)
739 return rth
->rt_genid
!= rt_genid(dev_net(rth
->dst
.dev
));
743 * Perform a full scan of hash table and free all entries.
744 * Can be called by a softirq or a process.
745 * In the later case, we want to be reschedule if necessary
747 static void rt_do_flush(struct net
*net
, int process_context
)
750 struct rtable
*rth
, *next
;
752 for (i
= 0; i
<= rt_hash_mask
; i
++) {
753 struct rtable __rcu
**pprev
;
756 if (process_context
&& need_resched())
758 rth
= rcu_dereference_raw(rt_hash_table
[i
].chain
);
762 spin_lock_bh(rt_hash_lock_addr(i
));
765 pprev
= &rt_hash_table
[i
].chain
;
766 rth
= rcu_dereference_protected(*pprev
,
767 lockdep_is_held(rt_hash_lock_addr(i
)));
770 next
= rcu_dereference_protected(rth
->dst
.rt_next
,
771 lockdep_is_held(rt_hash_lock_addr(i
)));
774 net_eq(dev_net(rth
->dst
.dev
), net
)) {
775 rcu_assign_pointer(*pprev
, next
);
776 rcu_assign_pointer(rth
->dst
.rt_next
, list
);
779 pprev
= &rth
->dst
.rt_next
;
784 spin_unlock_bh(rt_hash_lock_addr(i
));
786 for (; list
; list
= next
) {
787 next
= rcu_dereference_protected(list
->dst
.rt_next
, 1);
794 * While freeing expired entries, we compute average chain length
795 * and standard deviation, using fixed-point arithmetic.
796 * This to have an estimation of rt_chain_length_max
797 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
798 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
802 #define ONE (1UL << FRACT_BITS)
805 * Given a hash chain and an item in this hash chain,
806 * find if a previous entry has the same hash_inputs
807 * (but differs on tos, mark or oif)
808 * Returns 0 if an alias is found.
809 * Returns ONE if rth has no alias before itself.
811 static int has_noalias(const struct rtable
*head
, const struct rtable
*rth
)
813 const struct rtable
*aux
= head
;
816 if (compare_hash_inputs(aux
, rth
))
818 aux
= rcu_dereference_protected(aux
->dst
.rt_next
, 1);
824 * Perturbation of rt_genid by a small quantity [1..256]
825 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
826 * many times (2^24) without giving recent rt_genid.
827 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
829 static void rt_cache_invalidate(struct net
*net
)
831 unsigned char shuffle
;
833 get_random_bytes(&shuffle
, sizeof(shuffle
));
834 atomic_add(shuffle
+ 1U, &net
->ipv4
.rt_genid
);
838 * delay < 0 : invalidate cache (fast : entries will be deleted later)
839 * delay >= 0 : invalidate & flush cache (can be long)
841 void rt_cache_flush(struct net
*net
, int delay
)
843 rt_cache_invalidate(net
);
845 rt_do_flush(net
, !in_softirq());
848 /* Flush previous cache invalidated entries from the cache */
849 void rt_cache_flush_batch(struct net
*net
)
851 rt_do_flush(net
, !in_softirq());
854 static void rt_emergency_hash_rebuild(struct net
*net
)
857 printk(KERN_WARNING
"Route hash chain too long!\n");
858 rt_cache_invalidate(net
);
862 Short description of GC goals.
864 We want to build algorithm, which will keep routing cache
865 at some equilibrium point, when number of aged off entries
866 is kept approximately equal to newly generated ones.
868 Current expiration strength is variable "expire".
869 We try to adjust it dynamically, so that if networking
870 is idle expires is large enough to keep enough of warm entries,
871 and when load increases it reduces to limit cache size.
874 static int rt_garbage_collect(struct dst_ops
*ops
)
876 static unsigned long expire
= RT_GC_TIMEOUT
;
877 static unsigned long last_gc
;
879 static int equilibrium
;
881 struct rtable __rcu
**rthp
;
882 unsigned long now
= jiffies
;
884 int entries
= dst_entries_get_fast(&ipv4_dst_ops
);
887 * Garbage collection is pretty expensive,
888 * do not make it too frequently.
891 RT_CACHE_STAT_INC(gc_total
);
893 if (now
- last_gc
< ip_rt_gc_min_interval
&&
894 entries
< ip_rt_max_size
) {
895 RT_CACHE_STAT_INC(gc_ignored
);
899 entries
= dst_entries_get_slow(&ipv4_dst_ops
);
900 /* Calculate number of entries, which we want to expire now. */
901 goal
= entries
- (ip_rt_gc_elasticity
<< rt_hash_log
);
903 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
904 equilibrium
= ipv4_dst_ops
.gc_thresh
;
905 goal
= entries
- equilibrium
;
907 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
908 goal
= entries
- equilibrium
;
911 /* We are in dangerous area. Try to reduce cache really
914 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
915 equilibrium
= entries
- goal
;
918 if (now
- last_gc
>= ip_rt_gc_min_interval
)
929 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
930 unsigned long tmo
= expire
;
932 k
= (k
+ 1) & rt_hash_mask
;
933 rthp
= &rt_hash_table
[k
].chain
;
934 spin_lock_bh(rt_hash_lock_addr(k
));
935 while ((rth
= rcu_dereference_protected(*rthp
,
936 lockdep_is_held(rt_hash_lock_addr(k
)))) != NULL
) {
937 if (!rt_is_expired(rth
) &&
938 !rt_may_expire(rth
, tmo
, expire
)) {
940 rthp
= &rth
->dst
.rt_next
;
943 *rthp
= rth
->dst
.rt_next
;
947 spin_unlock_bh(rt_hash_lock_addr(k
));
956 /* Goal is not achieved. We stop process if:
958 - if expire reduced to zero. Otherwise, expire is halfed.
959 - if table is not full.
960 - if we are called from interrupt.
961 - jiffies check is just fallback/debug loop breaker.
962 We will not spin here for long time in any case.
965 RT_CACHE_STAT_INC(gc_goal_miss
);
971 #if RT_CACHE_DEBUG >= 2
972 printk(KERN_DEBUG
"expire>> %u %d %d %d\n", expire
,
973 dst_entries_get_fast(&ipv4_dst_ops
), goal
, i
);
976 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
978 } while (!in_softirq() && time_before_eq(jiffies
, now
));
980 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
982 if (dst_entries_get_slow(&ipv4_dst_ops
) < ip_rt_max_size
)
985 printk(KERN_WARNING
"dst cache overflow\n");
986 RT_CACHE_STAT_INC(gc_dst_overflow
);
990 expire
+= ip_rt_gc_min_interval
;
991 if (expire
> ip_rt_gc_timeout
||
992 dst_entries_get_fast(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
||
993 dst_entries_get_slow(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
)
994 expire
= ip_rt_gc_timeout
;
995 #if RT_CACHE_DEBUG >= 2
996 printk(KERN_DEBUG
"expire++ %u %d %d %d\n", expire
,
997 dst_entries_get_fast(&ipv4_dst_ops
), goal
, rover
);
1003 * Returns number of entries in a hash chain that have different hash_inputs
1005 static int slow_chain_length(const struct rtable
*head
)
1008 const struct rtable
*rth
= head
;
1011 length
+= has_noalias(head
, rth
);
1012 rth
= rcu_dereference_protected(rth
->dst
.rt_next
, 1);
1014 return length
>> FRACT_BITS
;
1017 static struct rtable
*rt_intern_hash(unsigned hash
, struct rtable
*rt
,
1018 struct sk_buff
*skb
, int ifindex
)
1020 struct rtable
*rth
, *cand
;
1021 struct rtable __rcu
**rthp
, **candp
;
1025 int attempts
= !in_softirq();
1029 min_score
= ~(u32
)0;
1034 if (!rt_caching(dev_net(rt
->dst
.dev
))) {
1036 * If we're not caching, just tell the caller we
1037 * were successful and don't touch the route. The
1038 * caller hold the sole reference to the cache entry, and
1039 * it will be released when the caller is done with it.
1040 * If we drop it here, the callers have no way to resolve routes
1041 * when we're not caching. Instead, just point *rp at rt, so
1042 * the caller gets a single use out of the route
1043 * Note that we do rt_free on this new route entry, so that
1044 * once its refcount hits zero, we are still able to reap it
1046 * Note: To avoid expensive rcu stuff for this uncached dst,
1047 * we set DST_NOCACHE so that dst_release() can free dst without
1048 * waiting a grace period.
1051 rt
->dst
.flags
|= DST_NOCACHE
;
1052 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1053 int err
= arp_bind_neighbour(&rt
->dst
);
1055 if (net_ratelimit())
1057 "Neighbour table failure & not caching routes.\n");
1059 return ERR_PTR(err
);
1066 rthp
= &rt_hash_table
[hash
].chain
;
1068 spin_lock_bh(rt_hash_lock_addr(hash
));
1069 while ((rth
= rcu_dereference_protected(*rthp
,
1070 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1071 if (rt_is_expired(rth
)) {
1072 *rthp
= rth
->dst
.rt_next
;
1076 if (compare_keys(rth
, rt
) && compare_netns(rth
, rt
)) {
1078 *rthp
= rth
->dst
.rt_next
;
1080 * Since lookup is lockfree, the deletion
1081 * must be visible to another weakly ordered CPU before
1082 * the insertion at the start of the hash chain.
1084 rcu_assign_pointer(rth
->dst
.rt_next
,
1085 rt_hash_table
[hash
].chain
);
1087 * Since lookup is lockfree, the update writes
1088 * must be ordered for consistency on SMP.
1090 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1092 dst_use(&rth
->dst
, now
);
1093 spin_unlock_bh(rt_hash_lock_addr(hash
));
1097 skb_dst_set(skb
, &rth
->dst
);
1101 if (!atomic_read(&rth
->dst
.__refcnt
)) {
1102 u32 score
= rt_score(rth
);
1104 if (score
<= min_score
) {
1113 rthp
= &rth
->dst
.rt_next
;
1117 /* ip_rt_gc_elasticity used to be average length of chain
1118 * length, when exceeded gc becomes really aggressive.
1120 * The second limit is less certain. At the moment it allows
1121 * only 2 entries per bucket. We will see.
1123 if (chain_length
> ip_rt_gc_elasticity
) {
1124 *candp
= cand
->dst
.rt_next
;
1128 if (chain_length
> rt_chain_length_max
&&
1129 slow_chain_length(rt_hash_table
[hash
].chain
) > rt_chain_length_max
) {
1130 struct net
*net
= dev_net(rt
->dst
.dev
);
1131 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1132 if (!rt_caching(net
)) {
1133 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1134 rt
->dst
.dev
->name
, num
);
1136 rt_emergency_hash_rebuild(net
);
1137 spin_unlock_bh(rt_hash_lock_addr(hash
));
1139 hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1140 ifindex
, rt_genid(net
));
1145 /* Try to bind route to arp only if it is output
1146 route or unicast forwarding path.
1148 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1149 int err
= arp_bind_neighbour(&rt
->dst
);
1151 spin_unlock_bh(rt_hash_lock_addr(hash
));
1153 if (err
!= -ENOBUFS
) {
1155 return ERR_PTR(err
);
1158 /* Neighbour tables are full and nothing
1159 can be released. Try to shrink route cache,
1160 it is most likely it holds some neighbour records.
1162 if (attempts
-- > 0) {
1163 int saved_elasticity
= ip_rt_gc_elasticity
;
1164 int saved_int
= ip_rt_gc_min_interval
;
1165 ip_rt_gc_elasticity
= 1;
1166 ip_rt_gc_min_interval
= 0;
1167 rt_garbage_collect(&ipv4_dst_ops
);
1168 ip_rt_gc_min_interval
= saved_int
;
1169 ip_rt_gc_elasticity
= saved_elasticity
;
1173 if (net_ratelimit())
1174 printk(KERN_WARNING
"ipv4: Neighbour table overflow.\n");
1176 return ERR_PTR(-ENOBUFS
);
1180 rt
->dst
.rt_next
= rt_hash_table
[hash
].chain
;
1182 #if RT_CACHE_DEBUG >= 2
1183 if (rt
->dst
.rt_next
) {
1185 printk(KERN_DEBUG
"rt_cache @%02x: %pI4",
1187 for (trt
= rt
->dst
.rt_next
; trt
; trt
= trt
->dst
.rt_next
)
1188 printk(" . %pI4", &trt
->rt_dst
);
1193 * Since lookup is lockfree, we must make sure
1194 * previous writes to rt are committed to memory
1195 * before making rt visible to other CPUS.
1197 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1199 spin_unlock_bh(rt_hash_lock_addr(hash
));
1203 skb_dst_set(skb
, &rt
->dst
);
1207 static atomic_t __rt_peer_genid
= ATOMIC_INIT(0);
1209 static u32
rt_peer_genid(void)
1211 return atomic_read(&__rt_peer_genid
);
1214 void rt_bind_peer(struct rtable
*rt
, int create
)
1216 struct inet_peer
*peer
;
1218 peer
= inet_getpeer_v4(rt
->rt_dst
, create
);
1220 if (peer
&& cmpxchg(&rt
->peer
, NULL
, peer
) != NULL
)
1223 rt
->rt_peer_genid
= rt_peer_genid();
1227 * Peer allocation may fail only in serious out-of-memory conditions. However
1228 * we still can generate some output.
1229 * Random ID selection looks a bit dangerous because we have no chances to
1230 * select ID being unique in a reasonable period of time.
1231 * But broken packet identifier may be better than no packet at all.
1233 static void ip_select_fb_ident(struct iphdr
*iph
)
1235 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1236 static u32 ip_fallback_id
;
1239 spin_lock_bh(&ip_fb_id_lock
);
1240 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1241 iph
->id
= htons(salt
& 0xFFFF);
1242 ip_fallback_id
= salt
;
1243 spin_unlock_bh(&ip_fb_id_lock
);
1246 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1248 struct rtable
*rt
= (struct rtable
*) dst
;
1251 if (rt
->peer
== NULL
)
1252 rt_bind_peer(rt
, 1);
1254 /* If peer is attached to destination, it is never detached,
1255 so that we need not to grab a lock to dereference it.
1258 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1262 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1263 __builtin_return_address(0));
1265 ip_select_fb_ident(iph
);
1267 EXPORT_SYMBOL(__ip_select_ident
);
1269 static void rt_del(unsigned hash
, struct rtable
*rt
)
1271 struct rtable __rcu
**rthp
;
1274 rthp
= &rt_hash_table
[hash
].chain
;
1275 spin_lock_bh(rt_hash_lock_addr(hash
));
1277 while ((aux
= rcu_dereference_protected(*rthp
,
1278 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1279 if (aux
== rt
|| rt_is_expired(aux
)) {
1280 *rthp
= aux
->dst
.rt_next
;
1284 rthp
= &aux
->dst
.rt_next
;
1286 spin_unlock_bh(rt_hash_lock_addr(hash
));
1289 /* called in rcu_read_lock() section */
1290 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1291 __be32 saddr
, struct net_device
*dev
)
1293 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1294 struct inet_peer
*peer
;
1301 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1302 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1303 ipv4_is_zeronet(new_gw
))
1304 goto reject_redirect
;
1306 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1307 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1308 goto reject_redirect
;
1309 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1310 goto reject_redirect
;
1312 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1313 goto reject_redirect
;
1316 peer
= inet_getpeer_v4(daddr
, 1);
1318 peer
->redirect_learned
.a4
= new_gw
;
1322 atomic_inc(&__rt_peer_genid
);
1327 #ifdef CONFIG_IP_ROUTE_VERBOSE
1328 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1329 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1330 " Advised path = %pI4 -> %pI4\n",
1331 &old_gw
, dev
->name
, &new_gw
,
1337 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1339 struct rtable
*rt
= (struct rtable
*)dst
;
1340 struct dst_entry
*ret
= dst
;
1343 if (dst
->obsolete
> 0) {
1346 } else if (rt
->rt_flags
& RTCF_REDIRECTED
) {
1347 unsigned hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1349 rt_genid(dev_net(dst
->dev
)));
1350 #if RT_CACHE_DEBUG >= 1
1351 printk(KERN_DEBUG
"ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1352 &rt
->rt_dst
, rt
->rt_key_tos
);
1356 } else if (rt
->peer
&&
1357 rt
->peer
->pmtu_expires
&&
1358 time_after_eq(jiffies
, rt
->peer
->pmtu_expires
)) {
1359 unsigned long orig
= rt
->peer
->pmtu_expires
;
1361 if (cmpxchg(&rt
->peer
->pmtu_expires
, orig
, 0) == orig
)
1362 dst_metric_set(dst
, RTAX_MTU
,
1363 rt
->peer
->pmtu_orig
);
1371 * 1. The first ip_rt_redirect_number redirects are sent
1372 * with exponential backoff, then we stop sending them at all,
1373 * assuming that the host ignores our redirects.
1374 * 2. If we did not see packets requiring redirects
1375 * during ip_rt_redirect_silence, we assume that the host
1376 * forgot redirected route and start to send redirects again.
1378 * This algorithm is much cheaper and more intelligent than dumb load limiting
1381 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1382 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1385 void ip_rt_send_redirect(struct sk_buff
*skb
)
1387 struct rtable
*rt
= skb_rtable(skb
);
1388 struct in_device
*in_dev
;
1389 struct inet_peer
*peer
;
1393 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
1394 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1398 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1402 rt_bind_peer(rt
, 1);
1405 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1409 /* No redirected packets during ip_rt_redirect_silence;
1410 * reset the algorithm.
1412 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
))
1413 peer
->rate_tokens
= 0;
1415 /* Too many ignored redirects; do not send anything
1416 * set dst.rate_last to the last seen redirected packet.
1418 if (peer
->rate_tokens
>= ip_rt_redirect_number
) {
1419 peer
->rate_last
= jiffies
;
1423 /* Check for load limit; set rate_last to the latest sent
1426 if (peer
->rate_tokens
== 0 ||
1429 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
1430 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1431 peer
->rate_last
= jiffies
;
1432 ++peer
->rate_tokens
;
1433 #ifdef CONFIG_IP_ROUTE_VERBOSE
1435 peer
->rate_tokens
== ip_rt_redirect_number
&&
1437 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1438 &ip_hdr(skb
)->saddr
, rt
->rt_iif
,
1439 &rt
->rt_dst
, &rt
->rt_gateway
);
1444 static int ip_error(struct sk_buff
*skb
)
1446 struct rtable
*rt
= skb_rtable(skb
);
1447 struct inet_peer
*peer
;
1452 switch (rt
->dst
.error
) {
1457 code
= ICMP_HOST_UNREACH
;
1460 code
= ICMP_NET_UNREACH
;
1461 IP_INC_STATS_BH(dev_net(rt
->dst
.dev
),
1462 IPSTATS_MIB_INNOROUTES
);
1465 code
= ICMP_PKT_FILTERED
;
1470 rt_bind_peer(rt
, 1);
1476 peer
->rate_tokens
+= now
- peer
->rate_last
;
1477 if (peer
->rate_tokens
> ip_rt_error_burst
)
1478 peer
->rate_tokens
= ip_rt_error_burst
;
1479 peer
->rate_last
= now
;
1480 if (peer
->rate_tokens
>= ip_rt_error_cost
)
1481 peer
->rate_tokens
-= ip_rt_error_cost
;
1486 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1488 out
: kfree_skb(skb
);
1493 * The last two values are not from the RFC but
1494 * are needed for AMPRnet AX.25 paths.
1497 static const unsigned short mtu_plateau
[] =
1498 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1500 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1504 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1505 if (old_mtu
> mtu_plateau
[i
])
1506 return mtu_plateau
[i
];
1510 unsigned short ip_rt_frag_needed(struct net
*net
, const struct iphdr
*iph
,
1511 unsigned short new_mtu
,
1512 struct net_device
*dev
)
1514 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1515 unsigned short est_mtu
= 0;
1516 struct inet_peer
*peer
;
1518 peer
= inet_getpeer_v4(iph
->daddr
, 1);
1520 unsigned short mtu
= new_mtu
;
1522 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1523 /* BSD 4.2 derived systems incorrectly adjust
1524 * tot_len by the IP header length, and report
1525 * a zero MTU in the ICMP message.
1528 old_mtu
>= 68 + (iph
->ihl
<< 2))
1529 old_mtu
-= iph
->ihl
<< 2;
1530 mtu
= guess_mtu(old_mtu
);
1533 if (mtu
< ip_rt_min_pmtu
)
1534 mtu
= ip_rt_min_pmtu
;
1535 if (!peer
->pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1536 unsigned long pmtu_expires
;
1538 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1543 peer
->pmtu_learned
= mtu
;
1544 peer
->pmtu_expires
= pmtu_expires
;
1549 atomic_inc(&__rt_peer_genid
);
1551 return est_mtu
? : new_mtu
;
1554 static void check_peer_pmtu(struct dst_entry
*dst
, struct inet_peer
*peer
)
1556 unsigned long expires
= peer
->pmtu_expires
;
1558 if (time_before(jiffies
, expires
)) {
1559 u32 orig_dst_mtu
= dst_mtu(dst
);
1560 if (peer
->pmtu_learned
< orig_dst_mtu
) {
1561 if (!peer
->pmtu_orig
)
1562 peer
->pmtu_orig
= dst_metric_raw(dst
, RTAX_MTU
);
1563 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_learned
);
1565 } else if (cmpxchg(&peer
->pmtu_expires
, expires
, 0) == expires
)
1566 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_orig
);
1569 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1571 struct rtable
*rt
= (struct rtable
*) dst
;
1572 struct inet_peer
*peer
;
1577 rt_bind_peer(rt
, 1);
1580 if (mtu
< ip_rt_min_pmtu
)
1581 mtu
= ip_rt_min_pmtu
;
1582 if (!peer
->pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1583 unsigned long pmtu_expires
;
1585 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1589 peer
->pmtu_learned
= mtu
;
1590 peer
->pmtu_expires
= pmtu_expires
;
1592 atomic_inc(&__rt_peer_genid
);
1593 rt
->rt_peer_genid
= rt_peer_genid();
1595 check_peer_pmtu(dst
, peer
);
1599 static int check_peer_redir(struct dst_entry
*dst
, struct inet_peer
*peer
)
1601 struct rtable
*rt
= (struct rtable
*) dst
;
1602 __be32 orig_gw
= rt
->rt_gateway
;
1604 dst_confirm(&rt
->dst
);
1606 neigh_release(rt
->dst
.neighbour
);
1607 rt
->dst
.neighbour
= NULL
;
1609 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1610 if (arp_bind_neighbour(&rt
->dst
) ||
1611 !(rt
->dst
.neighbour
->nud_state
& NUD_VALID
)) {
1612 if (rt
->dst
.neighbour
)
1613 neigh_event_send(rt
->dst
.neighbour
, NULL
);
1614 rt
->rt_gateway
= orig_gw
;
1617 rt
->rt_flags
|= RTCF_REDIRECTED
;
1618 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
,
1624 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1626 struct rtable
*rt
= (struct rtable
*) dst
;
1628 if (rt_is_expired(rt
))
1630 if (rt
->rt_peer_genid
!= rt_peer_genid()) {
1631 struct inet_peer
*peer
;
1634 rt_bind_peer(rt
, 0);
1637 if (peer
&& peer
->pmtu_expires
)
1638 check_peer_pmtu(dst
, peer
);
1640 if (peer
&& peer
->redirect_learned
.a4
&&
1641 peer
->redirect_learned
.a4
!= rt
->rt_gateway
) {
1642 if (check_peer_redir(dst
, peer
))
1646 rt
->rt_peer_genid
= rt_peer_genid();
1651 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1653 struct rtable
*rt
= (struct rtable
*) dst
;
1654 struct inet_peer
*peer
= rt
->peer
;
1657 fib_info_put(rt
->fi
);
1667 static void ipv4_link_failure(struct sk_buff
*skb
)
1671 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1673 rt
= skb_rtable(skb
);
1676 rt
->peer
->pmtu_expires
) {
1677 unsigned long orig
= rt
->peer
->pmtu_expires
;
1679 if (cmpxchg(&rt
->peer
->pmtu_expires
, orig
, 0) == orig
)
1680 dst_metric_set(&rt
->dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1684 static int ip_rt_bug(struct sk_buff
*skb
)
1686 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1687 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1688 skb
->dev
? skb
->dev
->name
: "?");
1694 We do not cache source address of outgoing interface,
1695 because it is used only by IP RR, TS and SRR options,
1696 so that it out of fast path.
1698 BTW remember: "addr" is allowed to be not aligned
1702 void ip_rt_get_source(u8
*addr
, struct sk_buff
*skb
, struct rtable
*rt
)
1706 if (rt_is_output_route(rt
))
1707 src
= ip_hdr(skb
)->saddr
;
1709 struct fib_result res
;
1715 memset(&fl4
, 0, sizeof(fl4
));
1716 fl4
.daddr
= iph
->daddr
;
1717 fl4
.saddr
= iph
->saddr
;
1718 fl4
.flowi4_tos
= iph
->tos
;
1719 fl4
.flowi4_oif
= rt
->dst
.dev
->ifindex
;
1720 fl4
.flowi4_iif
= skb
->dev
->ifindex
;
1721 fl4
.flowi4_mark
= skb
->mark
;
1724 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
) == 0)
1725 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1727 src
= inet_select_addr(rt
->dst
.dev
, rt
->rt_gateway
,
1731 memcpy(addr
, &src
, 4);
1734 #ifdef CONFIG_IP_ROUTE_CLASSID
1735 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1737 if (!(rt
->dst
.tclassid
& 0xFFFF))
1738 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1739 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1740 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1744 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1746 unsigned int advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
1749 advmss
= max_t(unsigned int, dst
->dev
->mtu
- 40,
1751 if (advmss
> 65535 - 40)
1752 advmss
= 65535 - 40;
1757 static unsigned int ipv4_default_mtu(const struct dst_entry
*dst
)
1759 unsigned int mtu
= dst
->dev
->mtu
;
1761 if (unlikely(dst_metric_locked(dst
, RTAX_MTU
))) {
1762 const struct rtable
*rt
= (const struct rtable
*) dst
;
1764 if (rt
->rt_gateway
!= rt
->rt_dst
&& mtu
> 576)
1768 if (mtu
> IP_MAX_MTU
)
1774 static void rt_init_metrics(struct rtable
*rt
, const struct flowi4
*fl4
,
1775 struct fib_info
*fi
)
1777 struct inet_peer
*peer
;
1780 /* If a peer entry exists for this destination, we must hook
1781 * it up in order to get at cached metrics.
1783 if (fl4
&& (fl4
->flowi4_flags
& FLOWI_FLAG_PRECOW_METRICS
))
1786 rt
->peer
= peer
= inet_getpeer_v4(rt
->rt_dst
, create
);
1788 rt
->rt_peer_genid
= rt_peer_genid();
1789 if (inet_metrics_new(peer
))
1790 memcpy(peer
->metrics
, fi
->fib_metrics
,
1791 sizeof(u32
) * RTAX_MAX
);
1792 dst_init_metrics(&rt
->dst
, peer
->metrics
, false);
1794 if (peer
->pmtu_expires
)
1795 check_peer_pmtu(&rt
->dst
, peer
);
1796 if (peer
->redirect_learned
.a4
&&
1797 peer
->redirect_learned
.a4
!= rt
->rt_gateway
) {
1798 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1799 rt
->rt_flags
|= RTCF_REDIRECTED
;
1802 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
) {
1804 atomic_inc(&fi
->fib_clntref
);
1806 dst_init_metrics(&rt
->dst
, fi
->fib_metrics
, true);
1810 static void rt_set_nexthop(struct rtable
*rt
, const struct flowi4
*fl4
,
1811 const struct fib_result
*res
,
1812 struct fib_info
*fi
, u16 type
, u32 itag
)
1814 struct dst_entry
*dst
= &rt
->dst
;
1817 if (FIB_RES_GW(*res
) &&
1818 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1819 rt
->rt_gateway
= FIB_RES_GW(*res
);
1820 rt_init_metrics(rt
, fl4
, fi
);
1821 #ifdef CONFIG_IP_ROUTE_CLASSID
1822 dst
->tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1826 if (dst_mtu(dst
) > IP_MAX_MTU
)
1827 dst_metric_set(dst
, RTAX_MTU
, IP_MAX_MTU
);
1828 if (dst_metric_raw(dst
, RTAX_ADVMSS
) > 65535 - 40)
1829 dst_metric_set(dst
, RTAX_ADVMSS
, 65535 - 40);
1831 #ifdef CONFIG_IP_ROUTE_CLASSID
1832 #ifdef CONFIG_IP_MULTIPLE_TABLES
1833 set_class_tag(rt
, fib_rules_tclass(res
));
1835 set_class_tag(rt
, itag
);
1839 static struct rtable
*rt_dst_alloc(struct net_device
*dev
,
1840 bool nopolicy
, bool noxfrm
)
1842 return dst_alloc(&ipv4_dst_ops
, dev
, 1, -1,
1844 (nopolicy
? DST_NOPOLICY
: 0) |
1845 (noxfrm
? DST_NOXFRM
: 0));
1848 /* called in rcu_read_lock() section */
1849 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1850 u8 tos
, struct net_device
*dev
, int our
)
1855 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1859 /* Primary sanity checks. */
1864 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1865 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
1868 if (ipv4_is_zeronet(saddr
)) {
1869 if (!ipv4_is_local_multicast(daddr
))
1871 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
1873 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
1878 rth
= rt_dst_alloc(init_net
.loopback_dev
,
1879 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
1883 #ifdef CONFIG_IP_ROUTE_CLASSID
1884 rth
->dst
.tclassid
= itag
;
1886 rth
->dst
.output
= ip_rt_bug
;
1888 rth
->rt_key_dst
= daddr
;
1889 rth
->rt_key_src
= saddr
;
1890 rth
->rt_genid
= rt_genid(dev_net(dev
));
1891 rth
->rt_flags
= RTCF_MULTICAST
;
1892 rth
->rt_type
= RTN_MULTICAST
;
1893 rth
->rt_key_tos
= tos
;
1894 rth
->rt_dst
= daddr
;
1895 rth
->rt_src
= saddr
;
1896 rth
->rt_route_iif
= dev
->ifindex
;
1897 rth
->rt_iif
= dev
->ifindex
;
1899 rth
->rt_mark
= skb
->mark
;
1900 rth
->rt_gateway
= daddr
;
1901 rth
->rt_spec_dst
= spec_dst
;
1902 rth
->rt_peer_genid
= 0;
1906 rth
->dst
.input
= ip_local_deliver
;
1907 rth
->rt_flags
|= RTCF_LOCAL
;
1910 #ifdef CONFIG_IP_MROUTE
1911 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1912 rth
->dst
.input
= ip_mr_input
;
1914 RT_CACHE_STAT_INC(in_slow_mc
);
1916 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
1917 rth
= rt_intern_hash(hash
, rth
, skb
, dev
->ifindex
);
1931 static void ip_handle_martian_source(struct net_device
*dev
,
1932 struct in_device
*in_dev
,
1933 struct sk_buff
*skb
,
1937 RT_CACHE_STAT_INC(in_martian_src
);
1938 #ifdef CONFIG_IP_ROUTE_VERBOSE
1939 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1941 * RFC1812 recommendation, if source is martian,
1942 * the only hint is MAC header.
1944 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
1945 &daddr
, &saddr
, dev
->name
);
1946 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1948 const unsigned char *p
= skb_mac_header(skb
);
1949 printk(KERN_WARNING
"ll header: ");
1950 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
1952 if (i
< (dev
->hard_header_len
- 1))
1961 /* called in rcu_read_lock() section */
1962 static int __mkroute_input(struct sk_buff
*skb
,
1963 const struct fib_result
*res
,
1964 struct in_device
*in_dev
,
1965 __be32 daddr
, __be32 saddr
, u32 tos
,
1966 struct rtable
**result
)
1970 struct in_device
*out_dev
;
1971 unsigned int flags
= 0;
1975 /* get a working reference to the output device */
1976 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
1977 if (out_dev
== NULL
) {
1978 if (net_ratelimit())
1979 printk(KERN_CRIT
"Bug in ip_route_input" \
1980 "_slow(). Please, report\n");
1985 err
= fib_validate_source(skb
, saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1986 in_dev
->dev
, &spec_dst
, &itag
);
1988 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1995 flags
|= RTCF_DIRECTSRC
;
1997 if (out_dev
== in_dev
&& err
&&
1998 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1999 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
2000 flags
|= RTCF_DOREDIRECT
;
2002 if (skb
->protocol
!= htons(ETH_P_IP
)) {
2003 /* Not IP (i.e. ARP). Do not create route, if it is
2004 * invalid for proxy arp. DNAT routes are always valid.
2006 * Proxy arp feature have been extended to allow, ARP
2007 * replies back to the same interface, to support
2008 * Private VLAN switch technologies. See arp.c.
2010 if (out_dev
== in_dev
&&
2011 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
2017 rth
= rt_dst_alloc(out_dev
->dev
,
2018 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2019 IN_DEV_CONF_GET(out_dev
, NOXFRM
));
2025 rth
->rt_key_dst
= daddr
;
2026 rth
->rt_key_src
= saddr
;
2027 rth
->rt_genid
= rt_genid(dev_net(rth
->dst
.dev
));
2028 rth
->rt_flags
= flags
;
2029 rth
->rt_type
= res
->type
;
2030 rth
->rt_key_tos
= tos
;
2031 rth
->rt_dst
= daddr
;
2032 rth
->rt_src
= saddr
;
2033 rth
->rt_route_iif
= in_dev
->dev
->ifindex
;
2034 rth
->rt_iif
= in_dev
->dev
->ifindex
;
2036 rth
->rt_mark
= skb
->mark
;
2037 rth
->rt_gateway
= daddr
;
2038 rth
->rt_spec_dst
= spec_dst
;
2039 rth
->rt_peer_genid
= 0;
2043 rth
->dst
.input
= ip_forward
;
2044 rth
->dst
.output
= ip_output
;
2046 rt_set_nexthop(rth
, NULL
, res
, res
->fi
, res
->type
, itag
);
2054 static int ip_mkroute_input(struct sk_buff
*skb
,
2055 struct fib_result
*res
,
2056 const struct flowi4
*fl4
,
2057 struct in_device
*in_dev
,
2058 __be32 daddr
, __be32 saddr
, u32 tos
)
2060 struct rtable
* rth
= NULL
;
2064 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2065 if (res
->fi
&& res
->fi
->fib_nhs
> 1)
2066 fib_select_multipath(res
);
2069 /* create a routing cache entry */
2070 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2074 /* put it into the cache */
2075 hash
= rt_hash(daddr
, saddr
, fl4
->flowi4_iif
,
2076 rt_genid(dev_net(rth
->dst
.dev
)));
2077 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
->flowi4_iif
);
2079 return PTR_ERR(rth
);
2084 * NOTE. We drop all the packets that has local source
2085 * addresses, because every properly looped back packet
2086 * must have correct destination already attached by output routine.
2088 * Such approach solves two big problems:
2089 * 1. Not simplex devices are handled properly.
2090 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2091 * called with rcu_read_lock()
2094 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2095 u8 tos
, struct net_device
*dev
)
2097 struct fib_result res
;
2098 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2102 struct rtable
* rth
;
2106 struct net
* net
= dev_net(dev
);
2108 /* IP on this device is disabled. */
2113 /* Check for the most weird martians, which can be not detected
2117 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2118 ipv4_is_loopback(saddr
))
2119 goto martian_source
;
2121 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
2124 /* Accept zero addresses only to limited broadcast;
2125 * I even do not know to fix it or not. Waiting for complains :-)
2127 if (ipv4_is_zeronet(saddr
))
2128 goto martian_source
;
2130 if (ipv4_is_zeronet(daddr
) || ipv4_is_loopback(daddr
))
2131 goto martian_destination
;
2134 * Now we are ready to route packet.
2137 fl4
.flowi4_iif
= dev
->ifindex
;
2138 fl4
.flowi4_mark
= skb
->mark
;
2139 fl4
.flowi4_tos
= tos
;
2140 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
2143 err
= fib_lookup(net
, &fl4
, &res
);
2145 if (!IN_DEV_FORWARD(in_dev
))
2150 RT_CACHE_STAT_INC(in_slow_tot
);
2152 if (res
.type
== RTN_BROADCAST
)
2155 if (res
.type
== RTN_LOCAL
) {
2156 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
2157 net
->loopback_dev
->ifindex
,
2158 dev
, &spec_dst
, &itag
);
2160 goto martian_source_keep_err
;
2162 flags
|= RTCF_DIRECTSRC
;
2167 if (!IN_DEV_FORWARD(in_dev
))
2169 if (res
.type
!= RTN_UNICAST
)
2170 goto martian_destination
;
2172 err
= ip_mkroute_input(skb
, &res
, &fl4
, in_dev
, daddr
, saddr
, tos
);
2176 if (skb
->protocol
!= htons(ETH_P_IP
))
2179 if (ipv4_is_zeronet(saddr
))
2180 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2182 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
2185 goto martian_source_keep_err
;
2187 flags
|= RTCF_DIRECTSRC
;
2189 flags
|= RTCF_BROADCAST
;
2190 res
.type
= RTN_BROADCAST
;
2191 RT_CACHE_STAT_INC(in_brd
);
2194 rth
= rt_dst_alloc(net
->loopback_dev
,
2195 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
2199 rth
->dst
.input
= ip_local_deliver
;
2200 rth
->dst
.output
= ip_rt_bug
;
2201 #ifdef CONFIG_IP_ROUTE_CLASSID
2202 rth
->dst
.tclassid
= itag
;
2205 rth
->rt_key_dst
= daddr
;
2206 rth
->rt_key_src
= saddr
;
2207 rth
->rt_genid
= rt_genid(net
);
2208 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2209 rth
->rt_type
= res
.type
;
2210 rth
->rt_key_tos
= tos
;
2211 rth
->rt_dst
= daddr
;
2212 rth
->rt_src
= saddr
;
2213 #ifdef CONFIG_IP_ROUTE_CLASSID
2214 rth
->dst
.tclassid
= itag
;
2216 rth
->rt_route_iif
= dev
->ifindex
;
2217 rth
->rt_iif
= dev
->ifindex
;
2219 rth
->rt_mark
= skb
->mark
;
2220 rth
->rt_gateway
= daddr
;
2221 rth
->rt_spec_dst
= spec_dst
;
2222 rth
->rt_peer_genid
= 0;
2225 if (res
.type
== RTN_UNREACHABLE
) {
2226 rth
->dst
.input
= ip_error
;
2227 rth
->dst
.error
= -err
;
2228 rth
->rt_flags
&= ~RTCF_LOCAL
;
2230 hash
= rt_hash(daddr
, saddr
, fl4
.flowi4_iif
, rt_genid(net
));
2231 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
.flowi4_iif
);
2238 RT_CACHE_STAT_INC(in_no_route
);
2239 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2240 res
.type
= RTN_UNREACHABLE
;
2246 * Do not cache martian addresses: they should be logged (RFC1812)
2248 martian_destination
:
2249 RT_CACHE_STAT_INC(in_martian_dst
);
2250 #ifdef CONFIG_IP_ROUTE_VERBOSE
2251 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2252 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2253 &daddr
, &saddr
, dev
->name
);
2257 err
= -EHOSTUNREACH
;
2270 martian_source_keep_err
:
2271 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2275 int ip_route_input_common(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2276 u8 tos
, struct net_device
*dev
, bool noref
)
2278 struct rtable
* rth
;
2280 int iif
= dev
->ifindex
;
2288 if (!rt_caching(net
))
2291 tos
&= IPTOS_RT_MASK
;
2292 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2294 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2295 rth
= rcu_dereference(rth
->dst
.rt_next
)) {
2296 if ((((__force u32
)rth
->rt_key_dst
^ (__force u32
)daddr
) |
2297 ((__force u32
)rth
->rt_key_src
^ (__force u32
)saddr
) |
2298 (rth
->rt_iif
^ iif
) |
2300 (rth
->rt_key_tos
^ tos
)) == 0 &&
2301 rth
->rt_mark
== skb
->mark
&&
2302 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2303 !rt_is_expired(rth
)) {
2305 dst_use_noref(&rth
->dst
, jiffies
);
2306 skb_dst_set_noref(skb
, &rth
->dst
);
2308 dst_use(&rth
->dst
, jiffies
);
2309 skb_dst_set(skb
, &rth
->dst
);
2311 RT_CACHE_STAT_INC(in_hit
);
2315 RT_CACHE_STAT_INC(in_hlist_search
);
2319 /* Multicast recognition logic is moved from route cache to here.
2320 The problem was that too many Ethernet cards have broken/missing
2321 hardware multicast filters :-( As result the host on multicasting
2322 network acquires a lot of useless route cache entries, sort of
2323 SDR messages from all the world. Now we try to get rid of them.
2324 Really, provided software IP multicast filter is organized
2325 reasonably (at least, hashed), it does not result in a slowdown
2326 comparing with route cache reject entries.
2327 Note, that multicast routers are not affected, because
2328 route cache entry is created eventually.
2330 if (ipv4_is_multicast(daddr
)) {
2331 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2334 int our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2335 ip_hdr(skb
)->protocol
);
2337 #ifdef CONFIG_IP_MROUTE
2339 (!ipv4_is_local_multicast(daddr
) &&
2340 IN_DEV_MFORWARD(in_dev
))
2343 int res
= ip_route_input_mc(skb
, daddr
, saddr
,
2352 res
= ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2356 EXPORT_SYMBOL(ip_route_input_common
);
2358 /* called with rcu_read_lock() */
2359 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2360 const struct flowi4
*fl4
,
2361 __be32 orig_daddr
, __be32 orig_saddr
,
2362 int orig_oif
, struct net_device
*dev_out
,
2365 struct fib_info
*fi
= res
->fi
;
2366 u32 tos
= RT_FL_TOS(fl4
);
2367 struct in_device
*in_dev
;
2368 u16 type
= res
->type
;
2371 if (ipv4_is_loopback(fl4
->saddr
) && !(dev_out
->flags
& IFF_LOOPBACK
))
2372 return ERR_PTR(-EINVAL
);
2374 if (ipv4_is_lbcast(fl4
->daddr
))
2375 type
= RTN_BROADCAST
;
2376 else if (ipv4_is_multicast(fl4
->daddr
))
2377 type
= RTN_MULTICAST
;
2378 else if (ipv4_is_zeronet(fl4
->daddr
))
2379 return ERR_PTR(-EINVAL
);
2381 if (dev_out
->flags
& IFF_LOOPBACK
)
2382 flags
|= RTCF_LOCAL
;
2384 in_dev
= __in_dev_get_rcu(dev_out
);
2386 return ERR_PTR(-EINVAL
);
2388 if (type
== RTN_BROADCAST
) {
2389 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2391 } else if (type
== RTN_MULTICAST
) {
2392 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2393 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2395 flags
&= ~RTCF_LOCAL
;
2396 /* If multicast route do not exist use
2397 * default one, but do not gateway in this case.
2400 if (fi
&& res
->prefixlen
< 4)
2404 rth
= rt_dst_alloc(dev_out
,
2405 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2406 IN_DEV_CONF_GET(in_dev
, NOXFRM
));
2408 return ERR_PTR(-ENOBUFS
);
2410 rth
->dst
.output
= ip_output
;
2412 rth
->rt_key_dst
= orig_daddr
;
2413 rth
->rt_key_src
= orig_saddr
;
2414 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2415 rth
->rt_flags
= flags
;
2416 rth
->rt_type
= type
;
2417 rth
->rt_key_tos
= tos
;
2418 rth
->rt_dst
= fl4
->daddr
;
2419 rth
->rt_src
= fl4
->saddr
;
2420 rth
->rt_route_iif
= 0;
2421 rth
->rt_iif
= orig_oif
? : dev_out
->ifindex
;
2422 rth
->rt_oif
= orig_oif
;
2423 rth
->rt_mark
= fl4
->flowi4_mark
;
2424 rth
->rt_gateway
= fl4
->daddr
;
2425 rth
->rt_spec_dst
= fl4
->saddr
;
2426 rth
->rt_peer_genid
= 0;
2430 RT_CACHE_STAT_INC(out_slow_tot
);
2432 if (flags
& RTCF_LOCAL
) {
2433 rth
->dst
.input
= ip_local_deliver
;
2434 rth
->rt_spec_dst
= fl4
->daddr
;
2436 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2437 rth
->rt_spec_dst
= fl4
->saddr
;
2438 if (flags
& RTCF_LOCAL
&&
2439 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2440 rth
->dst
.output
= ip_mc_output
;
2441 RT_CACHE_STAT_INC(out_slow_mc
);
2443 #ifdef CONFIG_IP_MROUTE
2444 if (type
== RTN_MULTICAST
) {
2445 if (IN_DEV_MFORWARD(in_dev
) &&
2446 !ipv4_is_local_multicast(fl4
->daddr
)) {
2447 rth
->dst
.input
= ip_mr_input
;
2448 rth
->dst
.output
= ip_mc_output
;
2454 rt_set_nexthop(rth
, fl4
, res
, fi
, type
, 0);
2460 * Major route resolver routine.
2461 * called with rcu_read_lock();
2464 static struct rtable
*ip_route_output_slow(struct net
*net
, struct flowi4
*fl4
)
2466 struct net_device
*dev_out
= NULL
;
2467 u32 tos
= RT_FL_TOS(fl4
);
2468 unsigned int flags
= 0;
2469 struct fib_result res
;
2476 #ifdef CONFIG_IP_MULTIPLE_TABLES
2480 orig_daddr
= fl4
->daddr
;
2481 orig_saddr
= fl4
->saddr
;
2482 orig_oif
= fl4
->flowi4_oif
;
2484 fl4
->flowi4_iif
= net
->loopback_dev
->ifindex
;
2485 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2486 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2487 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2491 rth
= ERR_PTR(-EINVAL
);
2492 if (ipv4_is_multicast(fl4
->saddr
) ||
2493 ipv4_is_lbcast(fl4
->saddr
) ||
2494 ipv4_is_zeronet(fl4
->saddr
))
2497 /* I removed check for oif == dev_out->oif here.
2498 It was wrong for two reasons:
2499 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2500 is assigned to multiple interfaces.
2501 2. Moreover, we are allowed to send packets with saddr
2502 of another iface. --ANK
2505 if (fl4
->flowi4_oif
== 0 &&
2506 (ipv4_is_multicast(fl4
->daddr
) ||
2507 ipv4_is_lbcast(fl4
->daddr
))) {
2508 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2509 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2510 if (dev_out
== NULL
)
2513 /* Special hack: user can direct multicasts
2514 and limited broadcast via necessary interface
2515 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2516 This hack is not just for fun, it allows
2517 vic,vat and friends to work.
2518 They bind socket to loopback, set ttl to zero
2519 and expect that it will work.
2520 From the viewpoint of routing cache they are broken,
2521 because we are not allowed to build multicast path
2522 with loopback source addr (look, routing cache
2523 cannot know, that ttl is zero, so that packet
2524 will not leave this host and route is valid).
2525 Luckily, this hack is good workaround.
2528 fl4
->flowi4_oif
= dev_out
->ifindex
;
2532 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2533 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2534 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2540 if (fl4
->flowi4_oif
) {
2541 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2542 rth
= ERR_PTR(-ENODEV
);
2543 if (dev_out
== NULL
)
2546 /* RACE: Check return value of inet_select_addr instead. */
2547 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2548 rth
= ERR_PTR(-ENETUNREACH
);
2551 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2552 ipv4_is_lbcast(fl4
->daddr
)) {
2554 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2559 if (ipv4_is_multicast(fl4
->daddr
))
2560 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2562 else if (!fl4
->daddr
)
2563 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2569 fl4
->daddr
= fl4
->saddr
;
2571 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2572 dev_out
= net
->loopback_dev
;
2573 fl4
->flowi4_oif
= net
->loopback_dev
->ifindex
;
2574 res
.type
= RTN_LOCAL
;
2575 flags
|= RTCF_LOCAL
;
2579 if (fib_lookup(net
, fl4
, &res
)) {
2581 if (fl4
->flowi4_oif
) {
2582 /* Apparently, routing tables are wrong. Assume,
2583 that the destination is on link.
2586 Because we are allowed to send to iface
2587 even if it has NO routes and NO assigned
2588 addresses. When oif is specified, routing
2589 tables are looked up with only one purpose:
2590 to catch if destination is gatewayed, rather than
2591 direct. Moreover, if MSG_DONTROUTE is set,
2592 we send packet, ignoring both routing tables
2593 and ifaddr state. --ANK
2596 We could make it even if oif is unknown,
2597 likely IPv6, but we do not.
2600 if (fl4
->saddr
== 0)
2601 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2603 res
.type
= RTN_UNICAST
;
2606 rth
= ERR_PTR(-ENETUNREACH
);
2610 if (res
.type
== RTN_LOCAL
) {
2612 if (res
.fi
->fib_prefsrc
)
2613 fl4
->saddr
= res
.fi
->fib_prefsrc
;
2615 fl4
->saddr
= fl4
->daddr
;
2617 dev_out
= net
->loopback_dev
;
2618 fl4
->flowi4_oif
= dev_out
->ifindex
;
2620 flags
|= RTCF_LOCAL
;
2624 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2625 if (res
.fi
->fib_nhs
> 1 && fl4
->flowi4_oif
== 0)
2626 fib_select_multipath(&res
);
2629 if (!res
.prefixlen
&&
2630 res
.table
->tb_num_default
> 1 &&
2631 res
.type
== RTN_UNICAST
&& !fl4
->flowi4_oif
)
2632 fib_select_default(&res
);
2635 fl4
->saddr
= FIB_RES_PREFSRC(net
, res
);
2637 dev_out
= FIB_RES_DEV(res
);
2638 fl4
->flowi4_oif
= dev_out
->ifindex
;
2642 rth
= __mkroute_output(&res
, fl4
, orig_daddr
, orig_saddr
, orig_oif
,
2647 hash
= rt_hash(orig_daddr
, orig_saddr
, orig_oif
,
2648 rt_genid(dev_net(dev_out
)));
2649 rth
= rt_intern_hash(hash
, rth
, NULL
, orig_oif
);
2657 struct rtable
*__ip_route_output_key(struct net
*net
, struct flowi4
*flp4
)
2662 if (!rt_caching(net
))
2665 hash
= rt_hash(flp4
->daddr
, flp4
->saddr
, flp4
->flowi4_oif
, rt_genid(net
));
2668 for (rth
= rcu_dereference_bh(rt_hash_table
[hash
].chain
); rth
;
2669 rth
= rcu_dereference_bh(rth
->dst
.rt_next
)) {
2670 if (rth
->rt_key_dst
== flp4
->daddr
&&
2671 rth
->rt_key_src
== flp4
->saddr
&&
2672 rt_is_output_route(rth
) &&
2673 rth
->rt_oif
== flp4
->flowi4_oif
&&
2674 rth
->rt_mark
== flp4
->flowi4_mark
&&
2675 !((rth
->rt_key_tos
^ flp4
->flowi4_tos
) &
2676 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2677 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2678 !rt_is_expired(rth
)) {
2679 dst_use(&rth
->dst
, jiffies
);
2680 RT_CACHE_STAT_INC(out_hit
);
2681 rcu_read_unlock_bh();
2683 flp4
->saddr
= rth
->rt_src
;
2685 flp4
->daddr
= rth
->rt_dst
;
2688 RT_CACHE_STAT_INC(out_hlist_search
);
2690 rcu_read_unlock_bh();
2693 return ip_route_output_slow(net
, flp4
);
2695 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2697 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2702 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry
*dst
)
2707 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2711 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2717 static struct dst_ops ipv4_dst_blackhole_ops
= {
2719 .protocol
= cpu_to_be16(ETH_P_IP
),
2720 .destroy
= ipv4_dst_destroy
,
2721 .check
= ipv4_blackhole_dst_check
,
2722 .default_mtu
= ipv4_blackhole_default_mtu
,
2723 .default_advmss
= ipv4_default_advmss
,
2724 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2725 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2728 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2730 struct rtable
*rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, 0, 0);
2731 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2734 struct dst_entry
*new = &rt
->dst
;
2737 new->input
= dst_discard
;
2738 new->output
= dst_discard
;
2739 dst_copy_metrics(new, &ort
->dst
);
2741 new->dev
= ort
->dst
.dev
;
2745 rt
->rt_key_dst
= ort
->rt_key_dst
;
2746 rt
->rt_key_src
= ort
->rt_key_src
;
2747 rt
->rt_key_tos
= ort
->rt_key_tos
;
2748 rt
->rt_route_iif
= ort
->rt_route_iif
;
2749 rt
->rt_iif
= ort
->rt_iif
;
2750 rt
->rt_oif
= ort
->rt_oif
;
2751 rt
->rt_mark
= ort
->rt_mark
;
2753 rt
->rt_genid
= rt_genid(net
);
2754 rt
->rt_flags
= ort
->rt_flags
;
2755 rt
->rt_type
= ort
->rt_type
;
2756 rt
->rt_dst
= ort
->rt_dst
;
2757 rt
->rt_src
= ort
->rt_src
;
2758 rt
->rt_gateway
= ort
->rt_gateway
;
2759 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2760 rt
->peer
= ort
->peer
;
2762 atomic_inc(&rt
->peer
->refcnt
);
2765 atomic_inc(&rt
->fi
->fib_clntref
);
2770 dst_release(dst_orig
);
2772 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2775 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2778 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2783 if (flp4
->flowi4_proto
)
2784 rt
= (struct rtable
*) xfrm_lookup(net
, &rt
->dst
,
2785 flowi4_to_flowi(flp4
),
2790 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2792 static int rt_fill_info(struct net
*net
,
2793 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2794 int nowait
, unsigned int flags
)
2796 struct rtable
*rt
= skb_rtable(skb
);
2798 struct nlmsghdr
*nlh
;
2800 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2802 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2806 r
= nlmsg_data(nlh
);
2807 r
->rtm_family
= AF_INET
;
2808 r
->rtm_dst_len
= 32;
2810 r
->rtm_tos
= rt
->rt_key_tos
;
2811 r
->rtm_table
= RT_TABLE_MAIN
;
2812 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2813 r
->rtm_type
= rt
->rt_type
;
2814 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2815 r
->rtm_protocol
= RTPROT_UNSPEC
;
2816 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2817 if (rt
->rt_flags
& RTCF_NOTIFY
)
2818 r
->rtm_flags
|= RTM_F_NOTIFY
;
2820 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2822 if (rt
->rt_key_src
) {
2823 r
->rtm_src_len
= 32;
2824 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->rt_key_src
);
2827 NLA_PUT_U32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
);
2828 #ifdef CONFIG_IP_ROUTE_CLASSID
2829 if (rt
->dst
.tclassid
)
2830 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->dst
.tclassid
);
2832 if (rt_is_input_route(rt
))
2833 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
2834 else if (rt
->rt_src
!= rt
->rt_key_src
)
2835 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
2837 if (rt
->rt_dst
!= rt
->rt_gateway
)
2838 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
2840 if (rtnetlink_put_metrics(skb
, dst_metrics_ptr(&rt
->dst
)) < 0)
2841 goto nla_put_failure
;
2844 NLA_PUT_BE32(skb
, RTA_MARK
, rt
->rt_mark
);
2846 error
= rt
->dst
.error
;
2847 expires
= (rt
->peer
&& rt
->peer
->pmtu_expires
) ?
2848 rt
->peer
->pmtu_expires
- jiffies
: 0;
2850 inet_peer_refcheck(rt
->peer
);
2851 id
= atomic_read(&rt
->peer
->ip_id_count
) & 0xffff;
2852 if (rt
->peer
->tcp_ts_stamp
) {
2853 ts
= rt
->peer
->tcp_ts
;
2854 tsage
= get_seconds() - rt
->peer
->tcp_ts_stamp
;
2858 if (rt_is_input_route(rt
)) {
2859 #ifdef CONFIG_IP_MROUTE
2860 __be32 dst
= rt
->rt_dst
;
2862 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2863 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2864 int err
= ipmr_get_route(net
, skb
,
2865 rt
->rt_src
, rt
->rt_dst
,
2871 goto nla_put_failure
;
2873 if (err
== -EMSGSIZE
)
2874 goto nla_put_failure
;
2880 NLA_PUT_U32(skb
, RTA_IIF
, rt
->rt_iif
);
2883 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, id
, ts
, tsage
,
2884 expires
, error
) < 0)
2885 goto nla_put_failure
;
2887 return nlmsg_end(skb
, nlh
);
2890 nlmsg_cancel(skb
, nlh
);
2894 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
2896 struct net
*net
= sock_net(in_skb
->sk
);
2898 struct nlattr
*tb
[RTA_MAX
+1];
2899 struct rtable
*rt
= NULL
;
2905 struct sk_buff
*skb
;
2907 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2911 rtm
= nlmsg_data(nlh
);
2913 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2919 /* Reserve room for dummy headers, this skb can pass
2920 through good chunk of routing engine.
2922 skb_reset_mac_header(skb
);
2923 skb_reset_network_header(skb
);
2925 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2926 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
2927 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2929 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
2930 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
2931 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2932 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
2935 struct net_device
*dev
;
2937 dev
= __dev_get_by_index(net
, iif
);
2943 skb
->protocol
= htons(ETH_P_IP
);
2947 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
2950 rt
= skb_rtable(skb
);
2951 if (err
== 0 && rt
->dst
.error
)
2952 err
= -rt
->dst
.error
;
2954 struct flowi4 fl4
= {
2957 .flowi4_tos
= rtm
->rtm_tos
,
2958 .flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
2959 .flowi4_mark
= mark
,
2961 rt
= ip_route_output_key(net
, &fl4
);
2971 skb_dst_set(skb
, &rt
->dst
);
2972 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2973 rt
->rt_flags
|= RTCF_NOTIFY
;
2975 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
2976 RTM_NEWROUTE
, 0, 0);
2980 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
2989 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2996 net
= sock_net(skb
->sk
);
3001 s_idx
= idx
= cb
->args
[1];
3002 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3003 if (!rt_hash_table
[h
].chain
)
3006 for (rt
= rcu_dereference_bh(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3007 rt
= rcu_dereference_bh(rt
->dst
.rt_next
), idx
++) {
3008 if (!net_eq(dev_net(rt
->dst
.dev
), net
) || idx
< s_idx
)
3010 if (rt_is_expired(rt
))
3012 skb_dst_set_noref(skb
, &rt
->dst
);
3013 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3014 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3015 1, NLM_F_MULTI
) <= 0) {
3017 rcu_read_unlock_bh();
3022 rcu_read_unlock_bh();
3031 void ip_rt_multicast_event(struct in_device
*in_dev
)
3033 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3036 #ifdef CONFIG_SYSCTL
3037 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3038 void __user
*buffer
,
3039 size_t *lenp
, loff_t
*ppos
)
3046 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3047 ctl
.data
= &flush_delay
;
3048 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3050 net
= (struct net
*)__ctl
->extra1
;
3051 rt_cache_flush(net
, flush_delay
);
3058 static ctl_table ipv4_route_table
[] = {
3060 .procname
= "gc_thresh",
3061 .data
= &ipv4_dst_ops
.gc_thresh
,
3062 .maxlen
= sizeof(int),
3064 .proc_handler
= proc_dointvec
,
3067 .procname
= "max_size",
3068 .data
= &ip_rt_max_size
,
3069 .maxlen
= sizeof(int),
3071 .proc_handler
= proc_dointvec
,
3074 /* Deprecated. Use gc_min_interval_ms */
3076 .procname
= "gc_min_interval",
3077 .data
= &ip_rt_gc_min_interval
,
3078 .maxlen
= sizeof(int),
3080 .proc_handler
= proc_dointvec_jiffies
,
3083 .procname
= "gc_min_interval_ms",
3084 .data
= &ip_rt_gc_min_interval
,
3085 .maxlen
= sizeof(int),
3087 .proc_handler
= proc_dointvec_ms_jiffies
,
3090 .procname
= "gc_timeout",
3091 .data
= &ip_rt_gc_timeout
,
3092 .maxlen
= sizeof(int),
3094 .proc_handler
= proc_dointvec_jiffies
,
3097 .procname
= "gc_interval",
3098 .data
= &ip_rt_gc_interval
,
3099 .maxlen
= sizeof(int),
3101 .proc_handler
= proc_dointvec_jiffies
,
3104 .procname
= "redirect_load",
3105 .data
= &ip_rt_redirect_load
,
3106 .maxlen
= sizeof(int),
3108 .proc_handler
= proc_dointvec
,
3111 .procname
= "redirect_number",
3112 .data
= &ip_rt_redirect_number
,
3113 .maxlen
= sizeof(int),
3115 .proc_handler
= proc_dointvec
,
3118 .procname
= "redirect_silence",
3119 .data
= &ip_rt_redirect_silence
,
3120 .maxlen
= sizeof(int),
3122 .proc_handler
= proc_dointvec
,
3125 .procname
= "error_cost",
3126 .data
= &ip_rt_error_cost
,
3127 .maxlen
= sizeof(int),
3129 .proc_handler
= proc_dointvec
,
3132 .procname
= "error_burst",
3133 .data
= &ip_rt_error_burst
,
3134 .maxlen
= sizeof(int),
3136 .proc_handler
= proc_dointvec
,
3139 .procname
= "gc_elasticity",
3140 .data
= &ip_rt_gc_elasticity
,
3141 .maxlen
= sizeof(int),
3143 .proc_handler
= proc_dointvec
,
3146 .procname
= "mtu_expires",
3147 .data
= &ip_rt_mtu_expires
,
3148 .maxlen
= sizeof(int),
3150 .proc_handler
= proc_dointvec_jiffies
,
3153 .procname
= "min_pmtu",
3154 .data
= &ip_rt_min_pmtu
,
3155 .maxlen
= sizeof(int),
3157 .proc_handler
= proc_dointvec
,
3160 .procname
= "min_adv_mss",
3161 .data
= &ip_rt_min_advmss
,
3162 .maxlen
= sizeof(int),
3164 .proc_handler
= proc_dointvec
,
3169 static struct ctl_table empty
[1];
3171 static struct ctl_table ipv4_skeleton
[] =
3173 { .procname
= "route",
3174 .mode
= 0555, .child
= ipv4_route_table
},
3175 { .procname
= "neigh",
3176 .mode
= 0555, .child
= empty
},
3180 static __net_initdata
struct ctl_path ipv4_path
[] = {
3181 { .procname
= "net", },
3182 { .procname
= "ipv4", },
3186 static struct ctl_table ipv4_route_flush_table
[] = {
3188 .procname
= "flush",
3189 .maxlen
= sizeof(int),
3191 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3196 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3197 { .procname
= "net", },
3198 { .procname
= "ipv4", },
3199 { .procname
= "route", },
3203 static __net_init
int sysctl_route_net_init(struct net
*net
)
3205 struct ctl_table
*tbl
;
3207 tbl
= ipv4_route_flush_table
;
3208 if (!net_eq(net
, &init_net
)) {
3209 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3213 tbl
[0].extra1
= net
;
3215 net
->ipv4
.route_hdr
=
3216 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3217 if (net
->ipv4
.route_hdr
== NULL
)
3222 if (tbl
!= ipv4_route_flush_table
)
3228 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3230 struct ctl_table
*tbl
;
3232 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3233 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3234 BUG_ON(tbl
== ipv4_route_flush_table
);
3238 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3239 .init
= sysctl_route_net_init
,
3240 .exit
= sysctl_route_net_exit
,
3244 static __net_init
int rt_genid_init(struct net
*net
)
3246 get_random_bytes(&net
->ipv4
.rt_genid
,
3247 sizeof(net
->ipv4
.rt_genid
));
3248 get_random_bytes(&net
->ipv4
.dev_addr_genid
,
3249 sizeof(net
->ipv4
.dev_addr_genid
));
3253 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3254 .init
= rt_genid_init
,
3258 #ifdef CONFIG_IP_ROUTE_CLASSID
3259 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3260 #endif /* CONFIG_IP_ROUTE_CLASSID */
3262 static __initdata
unsigned long rhash_entries
;
3263 static int __init
set_rhash_entries(char *str
)
3267 rhash_entries
= simple_strtoul(str
, &str
, 0);
3270 __setup("rhash_entries=", set_rhash_entries
);
3272 int __init
ip_rt_init(void)
3276 #ifdef CONFIG_IP_ROUTE_CLASSID
3277 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3279 panic("IP: failed to allocate ip_rt_acct\n");
3282 ipv4_dst_ops
.kmem_cachep
=
3283 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3284 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3286 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3288 if (dst_entries_init(&ipv4_dst_ops
) < 0)
3289 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3291 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
3292 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3294 rt_hash_table
= (struct rt_hash_bucket
*)
3295 alloc_large_system_hash("IP route cache",
3296 sizeof(struct rt_hash_bucket
),
3298 (totalram_pages
>= 128 * 1024) ?
3303 rhash_entries
? 0 : 512 * 1024);
3304 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3305 rt_hash_lock_init();
3307 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3308 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3313 if (ip_rt_proc_init())
3314 printk(KERN_ERR
"Unable to create route proc files\n");
3317 xfrm4_init(ip_rt_max_size
);
3319 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
);
3321 #ifdef CONFIG_SYSCTL
3322 register_pernet_subsys(&sysctl_route_ops
);
3324 register_pernet_subsys(&rt_genid_ops
);
3328 #ifdef CONFIG_SYSCTL
3330 * We really need to sanitize the damn ipv4 init order, then all
3331 * this nonsense will go away.
3333 void __init
ip_static_sysctl_init(void)
3335 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);