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>
111 #include <net/atmclip.h>
112 #include <net/secure_seq.h>
114 #define RT_FL_TOS(oldflp4) \
115 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
117 #define IP_MAX_MTU 0xFFF0
119 #define RT_GC_TIMEOUT (300*HZ)
121 static int ip_rt_max_size
;
122 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
123 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
124 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
125 static int ip_rt_redirect_number __read_mostly
= 9;
126 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
127 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
128 static int ip_rt_error_cost __read_mostly
= HZ
;
129 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
130 static int ip_rt_gc_elasticity __read_mostly
= 8;
131 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
132 static int ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
133 static int ip_rt_min_advmss __read_mostly
= 256;
134 static int rt_chain_length_max __read_mostly
= 20;
135 static int redirect_genid
;
138 * Interface to generic destination cache.
141 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
142 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
);
143 static unsigned int ipv4_default_mtu(const struct dst_entry
*dst
);
144 static void ipv4_dst_destroy(struct dst_entry
*dst
);
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
);
150 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
155 static u32
*ipv4_cow_metrics(struct dst_entry
*dst
, unsigned long old
)
157 struct rtable
*rt
= (struct rtable
*) dst
;
158 struct inet_peer
*peer
;
162 rt_bind_peer(rt
, rt
->rt_dst
, 1);
166 u32
*old_p
= __DST_METRICS_PTR(old
);
167 unsigned long prev
, new;
170 if (inet_metrics_new(peer
))
171 memcpy(p
, old_p
, sizeof(u32
) * RTAX_MAX
);
173 new = (unsigned long) p
;
174 prev
= cmpxchg(&dst
->_metrics
, old
, new);
177 p
= __DST_METRICS_PTR(prev
);
178 if (prev
& DST_METRICS_READ_ONLY
)
182 fib_info_put(rt
->fi
);
190 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
);
192 static struct dst_ops ipv4_dst_ops
= {
194 .protocol
= cpu_to_be16(ETH_P_IP
),
195 .gc
= rt_garbage_collect
,
196 .check
= ipv4_dst_check
,
197 .default_advmss
= ipv4_default_advmss
,
198 .default_mtu
= ipv4_default_mtu
,
199 .cow_metrics
= ipv4_cow_metrics
,
200 .destroy
= ipv4_dst_destroy
,
201 .ifdown
= ipv4_dst_ifdown
,
202 .negative_advice
= ipv4_negative_advice
,
203 .link_failure
= ipv4_link_failure
,
204 .update_pmtu
= ip_rt_update_pmtu
,
205 .local_out
= __ip_local_out
,
206 .neigh_lookup
= ipv4_neigh_lookup
,
209 #define ECN_OR_COST(class) TC_PRIO_##class
211 const __u8 ip_tos2prio
[16] = {
213 ECN_OR_COST(BESTEFFORT
),
215 ECN_OR_COST(BESTEFFORT
),
221 ECN_OR_COST(INTERACTIVE
),
223 ECN_OR_COST(INTERACTIVE
),
224 TC_PRIO_INTERACTIVE_BULK
,
225 ECN_OR_COST(INTERACTIVE_BULK
),
226 TC_PRIO_INTERACTIVE_BULK
,
227 ECN_OR_COST(INTERACTIVE_BULK
)
235 /* The locking scheme is rather straight forward:
237 * 1) Read-Copy Update protects the buckets of the central route hash.
238 * 2) Only writers remove entries, and they hold the lock
239 * as they look at rtable reference counts.
240 * 3) Only readers acquire references to rtable entries,
241 * they do so with atomic increments and with the
245 struct rt_hash_bucket
{
246 struct rtable __rcu
*chain
;
249 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
250 defined(CONFIG_PROVE_LOCKING)
252 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
253 * The size of this table is a power of two and depends on the number of CPUS.
254 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
256 #ifdef CONFIG_LOCKDEP
257 # define RT_HASH_LOCK_SZ 256
260 # define RT_HASH_LOCK_SZ 4096
262 # define RT_HASH_LOCK_SZ 2048
264 # define RT_HASH_LOCK_SZ 1024
266 # define RT_HASH_LOCK_SZ 512
268 # define RT_HASH_LOCK_SZ 256
272 static spinlock_t
*rt_hash_locks
;
273 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
275 static __init
void rt_hash_lock_init(void)
279 rt_hash_locks
= kmalloc(sizeof(spinlock_t
) * RT_HASH_LOCK_SZ
,
282 panic("IP: failed to allocate rt_hash_locks\n");
284 for (i
= 0; i
< RT_HASH_LOCK_SZ
; i
++)
285 spin_lock_init(&rt_hash_locks
[i
]);
288 # define rt_hash_lock_addr(slot) NULL
290 static inline void rt_hash_lock_init(void)
295 static struct rt_hash_bucket
*rt_hash_table __read_mostly
;
296 static unsigned rt_hash_mask __read_mostly
;
297 static unsigned int rt_hash_log __read_mostly
;
299 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
300 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
302 static inline unsigned int rt_hash(__be32 daddr
, __be32 saddr
, int idx
,
305 return jhash_3words((__force u32
)daddr
, (__force u32
)saddr
,
310 static inline int rt_genid(struct net
*net
)
312 return atomic_read(&net
->ipv4
.rt_genid
);
315 #ifdef CONFIG_PROC_FS
316 struct rt_cache_iter_state
{
317 struct seq_net_private p
;
322 static struct rtable
*rt_cache_get_first(struct seq_file
*seq
)
324 struct rt_cache_iter_state
*st
= seq
->private;
325 struct rtable
*r
= NULL
;
327 for (st
->bucket
= rt_hash_mask
; st
->bucket
>= 0; --st
->bucket
) {
328 if (!rcu_dereference_raw(rt_hash_table
[st
->bucket
].chain
))
331 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
333 if (dev_net(r
->dst
.dev
) == seq_file_net(seq
) &&
334 r
->rt_genid
== st
->genid
)
336 r
= rcu_dereference_bh(r
->dst
.rt_next
);
338 rcu_read_unlock_bh();
343 static struct rtable
*__rt_cache_get_next(struct seq_file
*seq
,
346 struct rt_cache_iter_state
*st
= seq
->private;
348 r
= rcu_dereference_bh(r
->dst
.rt_next
);
350 rcu_read_unlock_bh();
352 if (--st
->bucket
< 0)
354 } while (!rcu_dereference_raw(rt_hash_table
[st
->bucket
].chain
));
356 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
361 static struct rtable
*rt_cache_get_next(struct seq_file
*seq
,
364 struct rt_cache_iter_state
*st
= seq
->private;
365 while ((r
= __rt_cache_get_next(seq
, r
)) != NULL
) {
366 if (dev_net(r
->dst
.dev
) != seq_file_net(seq
))
368 if (r
->rt_genid
== st
->genid
)
374 static struct rtable
*rt_cache_get_idx(struct seq_file
*seq
, loff_t pos
)
376 struct rtable
*r
= rt_cache_get_first(seq
);
379 while (pos
&& (r
= rt_cache_get_next(seq
, r
)))
381 return pos
? NULL
: r
;
384 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
386 struct rt_cache_iter_state
*st
= seq
->private;
388 return rt_cache_get_idx(seq
, *pos
- 1);
389 st
->genid
= rt_genid(seq_file_net(seq
));
390 return SEQ_START_TOKEN
;
393 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
397 if (v
== SEQ_START_TOKEN
)
398 r
= rt_cache_get_first(seq
);
400 r
= rt_cache_get_next(seq
, v
);
405 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
407 if (v
&& v
!= SEQ_START_TOKEN
)
408 rcu_read_unlock_bh();
411 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
413 if (v
== SEQ_START_TOKEN
)
414 seq_printf(seq
, "%-127s\n",
415 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
416 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
419 struct rtable
*r
= v
;
424 n
= dst_get_neighbour(&r
->dst
);
425 HHUptod
= (n
&& (n
->nud_state
& NUD_CONNECTED
)) ? 1 : 0;
428 seq_printf(seq
, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
429 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
430 r
->dst
.dev
? r
->dst
.dev
->name
: "*",
431 (__force u32
)r
->rt_dst
,
432 (__force u32
)r
->rt_gateway
,
433 r
->rt_flags
, atomic_read(&r
->dst
.__refcnt
),
434 r
->dst
.__use
, 0, (__force u32
)r
->rt_src
,
435 dst_metric_advmss(&r
->dst
) + 40,
436 dst_metric(&r
->dst
, RTAX_WINDOW
),
437 (int)((dst_metric(&r
->dst
, RTAX_RTT
) >> 3) +
438 dst_metric(&r
->dst
, RTAX_RTTVAR
)),
442 r
->rt_spec_dst
, &len
);
444 seq_printf(seq
, "%*s\n", 127 - len
, "");
449 static const struct seq_operations rt_cache_seq_ops
= {
450 .start
= rt_cache_seq_start
,
451 .next
= rt_cache_seq_next
,
452 .stop
= rt_cache_seq_stop
,
453 .show
= rt_cache_seq_show
,
456 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
458 return seq_open_net(inode
, file
, &rt_cache_seq_ops
,
459 sizeof(struct rt_cache_iter_state
));
462 static const struct file_operations rt_cache_seq_fops
= {
463 .owner
= THIS_MODULE
,
464 .open
= rt_cache_seq_open
,
467 .release
= seq_release_net
,
471 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
476 return SEQ_START_TOKEN
;
478 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
479 if (!cpu_possible(cpu
))
482 return &per_cpu(rt_cache_stat
, cpu
);
487 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
491 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
492 if (!cpu_possible(cpu
))
495 return &per_cpu(rt_cache_stat
, cpu
);
501 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
506 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
508 struct rt_cache_stat
*st
= v
;
510 if (v
== SEQ_START_TOKEN
) {
511 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");
515 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
516 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
517 dst_entries_get_slow(&ipv4_dst_ops
),
540 static const struct seq_operations rt_cpu_seq_ops
= {
541 .start
= rt_cpu_seq_start
,
542 .next
= rt_cpu_seq_next
,
543 .stop
= rt_cpu_seq_stop
,
544 .show
= rt_cpu_seq_show
,
548 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
550 return seq_open(file
, &rt_cpu_seq_ops
);
553 static const struct file_operations rt_cpu_seq_fops
= {
554 .owner
= THIS_MODULE
,
555 .open
= rt_cpu_seq_open
,
558 .release
= seq_release
,
561 #ifdef CONFIG_IP_ROUTE_CLASSID
562 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
564 struct ip_rt_acct
*dst
, *src
;
567 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
571 for_each_possible_cpu(i
) {
572 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
573 for (j
= 0; j
< 256; j
++) {
574 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
575 dst
[j
].o_packets
+= src
[j
].o_packets
;
576 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
577 dst
[j
].i_packets
+= src
[j
].i_packets
;
581 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
586 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
588 return single_open(file
, rt_acct_proc_show
, NULL
);
591 static const struct file_operations rt_acct_proc_fops
= {
592 .owner
= THIS_MODULE
,
593 .open
= rt_acct_proc_open
,
596 .release
= single_release
,
600 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
602 struct proc_dir_entry
*pde
;
604 pde
= proc_net_fops_create(net
, "rt_cache", S_IRUGO
,
609 pde
= proc_create("rt_cache", S_IRUGO
,
610 net
->proc_net_stat
, &rt_cpu_seq_fops
);
614 #ifdef CONFIG_IP_ROUTE_CLASSID
615 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
621 #ifdef CONFIG_IP_ROUTE_CLASSID
623 remove_proc_entry("rt_cache", net
->proc_net_stat
);
626 remove_proc_entry("rt_cache", net
->proc_net
);
631 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
633 remove_proc_entry("rt_cache", net
->proc_net_stat
);
634 remove_proc_entry("rt_cache", net
->proc_net
);
635 #ifdef CONFIG_IP_ROUTE_CLASSID
636 remove_proc_entry("rt_acct", net
->proc_net
);
640 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
641 .init
= ip_rt_do_proc_init
,
642 .exit
= ip_rt_do_proc_exit
,
645 static int __init
ip_rt_proc_init(void)
647 return register_pernet_subsys(&ip_rt_proc_ops
);
651 static inline int ip_rt_proc_init(void)
655 #endif /* CONFIG_PROC_FS */
657 static inline void rt_free(struct rtable
*rt
)
659 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
662 static inline void rt_drop(struct rtable
*rt
)
665 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
668 static inline int rt_fast_clean(struct rtable
*rth
)
670 /* Kill broadcast/multicast entries very aggresively, if they
671 collide in hash table with more useful entries */
672 return (rth
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) &&
673 rt_is_input_route(rth
) && rth
->dst
.rt_next
;
676 static inline int rt_valuable(struct rtable
*rth
)
678 return (rth
->rt_flags
& (RTCF_REDIRECTED
| RTCF_NOTIFY
)) ||
679 (rth
->peer
&& rth
->peer
->pmtu_expires
);
682 static int rt_may_expire(struct rtable
*rth
, unsigned long tmo1
, unsigned long tmo2
)
687 if (atomic_read(&rth
->dst
.__refcnt
))
690 age
= jiffies
- rth
->dst
.lastuse
;
691 if ((age
<= tmo1
&& !rt_fast_clean(rth
)) ||
692 (age
<= tmo2
&& rt_valuable(rth
)))
698 /* Bits of score are:
700 * 30: not quite useless
701 * 29..0: usage counter
703 static inline u32
rt_score(struct rtable
*rt
)
705 u32 score
= jiffies
- rt
->dst
.lastuse
;
707 score
= ~score
& ~(3<<30);
712 if (rt_is_output_route(rt
) ||
713 !(rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
|RTCF_LOCAL
)))
719 static inline bool rt_caching(const struct net
*net
)
721 return net
->ipv4
.current_rt_cache_rebuild_count
<=
722 net
->ipv4
.sysctl_rt_cache_rebuild_count
;
725 static inline bool compare_hash_inputs(const struct rtable
*rt1
,
726 const struct rtable
*rt2
)
728 return ((((__force u32
)rt1
->rt_key_dst
^ (__force u32
)rt2
->rt_key_dst
) |
729 ((__force u32
)rt1
->rt_key_src
^ (__force u32
)rt2
->rt_key_src
) |
730 (rt1
->rt_route_iif
^ rt2
->rt_route_iif
)) == 0);
733 static inline int compare_keys(struct rtable
*rt1
, struct rtable
*rt2
)
735 return (((__force u32
)rt1
->rt_key_dst
^ (__force u32
)rt2
->rt_key_dst
) |
736 ((__force u32
)rt1
->rt_key_src
^ (__force u32
)rt2
->rt_key_src
) |
737 (rt1
->rt_mark
^ rt2
->rt_mark
) |
738 (rt1
->rt_key_tos
^ rt2
->rt_key_tos
) |
739 (rt1
->rt_route_iif
^ rt2
->rt_route_iif
) |
740 (rt1
->rt_oif
^ rt2
->rt_oif
)) == 0;
743 static inline int compare_netns(struct rtable
*rt1
, struct rtable
*rt2
)
745 return net_eq(dev_net(rt1
->dst
.dev
), dev_net(rt2
->dst
.dev
));
748 static inline int rt_is_expired(struct rtable
*rth
)
750 return rth
->rt_genid
!= rt_genid(dev_net(rth
->dst
.dev
));
754 * Perform a full scan of hash table and free all entries.
755 * Can be called by a softirq or a process.
756 * In the later case, we want to be reschedule if necessary
758 static void rt_do_flush(struct net
*net
, int process_context
)
761 struct rtable
*rth
, *next
;
763 for (i
= 0; i
<= rt_hash_mask
; i
++) {
764 struct rtable __rcu
**pprev
;
767 if (process_context
&& need_resched())
769 rth
= rcu_dereference_raw(rt_hash_table
[i
].chain
);
773 spin_lock_bh(rt_hash_lock_addr(i
));
776 pprev
= &rt_hash_table
[i
].chain
;
777 rth
= rcu_dereference_protected(*pprev
,
778 lockdep_is_held(rt_hash_lock_addr(i
)));
781 next
= rcu_dereference_protected(rth
->dst
.rt_next
,
782 lockdep_is_held(rt_hash_lock_addr(i
)));
785 net_eq(dev_net(rth
->dst
.dev
), net
)) {
786 rcu_assign_pointer(*pprev
, next
);
787 rcu_assign_pointer(rth
->dst
.rt_next
, list
);
790 pprev
= &rth
->dst
.rt_next
;
795 spin_unlock_bh(rt_hash_lock_addr(i
));
797 for (; list
; list
= next
) {
798 next
= rcu_dereference_protected(list
->dst
.rt_next
, 1);
805 * While freeing expired entries, we compute average chain length
806 * and standard deviation, using fixed-point arithmetic.
807 * This to have an estimation of rt_chain_length_max
808 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
809 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
813 #define ONE (1UL << FRACT_BITS)
816 * Given a hash chain and an item in this hash chain,
817 * find if a previous entry has the same hash_inputs
818 * (but differs on tos, mark or oif)
819 * Returns 0 if an alias is found.
820 * Returns ONE if rth has no alias before itself.
822 static int has_noalias(const struct rtable
*head
, const struct rtable
*rth
)
824 const struct rtable
*aux
= head
;
827 if (compare_hash_inputs(aux
, rth
))
829 aux
= rcu_dereference_protected(aux
->dst
.rt_next
, 1);
835 * Perturbation of rt_genid by a small quantity [1..256]
836 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
837 * many times (2^24) without giving recent rt_genid.
838 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
840 static void rt_cache_invalidate(struct net
*net
)
842 unsigned char shuffle
;
844 get_random_bytes(&shuffle
, sizeof(shuffle
));
845 atomic_add(shuffle
+ 1U, &net
->ipv4
.rt_genid
);
850 * delay < 0 : invalidate cache (fast : entries will be deleted later)
851 * delay >= 0 : invalidate & flush cache (can be long)
853 void rt_cache_flush(struct net
*net
, int delay
)
855 rt_cache_invalidate(net
);
857 rt_do_flush(net
, !in_softirq());
860 /* Flush previous cache invalidated entries from the cache */
861 void rt_cache_flush_batch(struct net
*net
)
863 rt_do_flush(net
, !in_softirq());
866 static void rt_emergency_hash_rebuild(struct net
*net
)
869 printk(KERN_WARNING
"Route hash chain too long!\n");
870 rt_cache_invalidate(net
);
874 Short description of GC goals.
876 We want to build algorithm, which will keep routing cache
877 at some equilibrium point, when number of aged off entries
878 is kept approximately equal to newly generated ones.
880 Current expiration strength is variable "expire".
881 We try to adjust it dynamically, so that if networking
882 is idle expires is large enough to keep enough of warm entries,
883 and when load increases it reduces to limit cache size.
886 static int rt_garbage_collect(struct dst_ops
*ops
)
888 static unsigned long expire
= RT_GC_TIMEOUT
;
889 static unsigned long last_gc
;
891 static int equilibrium
;
893 struct rtable __rcu
**rthp
;
894 unsigned long now
= jiffies
;
896 int entries
= dst_entries_get_fast(&ipv4_dst_ops
);
899 * Garbage collection is pretty expensive,
900 * do not make it too frequently.
903 RT_CACHE_STAT_INC(gc_total
);
905 if (now
- last_gc
< ip_rt_gc_min_interval
&&
906 entries
< ip_rt_max_size
) {
907 RT_CACHE_STAT_INC(gc_ignored
);
911 entries
= dst_entries_get_slow(&ipv4_dst_ops
);
912 /* Calculate number of entries, which we want to expire now. */
913 goal
= entries
- (ip_rt_gc_elasticity
<< rt_hash_log
);
915 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
916 equilibrium
= ipv4_dst_ops
.gc_thresh
;
917 goal
= entries
- equilibrium
;
919 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
920 goal
= entries
- equilibrium
;
923 /* We are in dangerous area. Try to reduce cache really
926 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
927 equilibrium
= entries
- goal
;
930 if (now
- last_gc
>= ip_rt_gc_min_interval
)
941 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
942 unsigned long tmo
= expire
;
944 k
= (k
+ 1) & rt_hash_mask
;
945 rthp
= &rt_hash_table
[k
].chain
;
946 spin_lock_bh(rt_hash_lock_addr(k
));
947 while ((rth
= rcu_dereference_protected(*rthp
,
948 lockdep_is_held(rt_hash_lock_addr(k
)))) != NULL
) {
949 if (!rt_is_expired(rth
) &&
950 !rt_may_expire(rth
, tmo
, expire
)) {
952 rthp
= &rth
->dst
.rt_next
;
955 *rthp
= rth
->dst
.rt_next
;
959 spin_unlock_bh(rt_hash_lock_addr(k
));
968 /* Goal is not achieved. We stop process if:
970 - if expire reduced to zero. Otherwise, expire is halfed.
971 - if table is not full.
972 - if we are called from interrupt.
973 - jiffies check is just fallback/debug loop breaker.
974 We will not spin here for long time in any case.
977 RT_CACHE_STAT_INC(gc_goal_miss
);
984 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
986 } while (!in_softirq() && time_before_eq(jiffies
, now
));
988 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
990 if (dst_entries_get_slow(&ipv4_dst_ops
) < ip_rt_max_size
)
993 printk(KERN_WARNING
"dst cache overflow\n");
994 RT_CACHE_STAT_INC(gc_dst_overflow
);
998 expire
+= ip_rt_gc_min_interval
;
999 if (expire
> ip_rt_gc_timeout
||
1000 dst_entries_get_fast(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
||
1001 dst_entries_get_slow(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
)
1002 expire
= ip_rt_gc_timeout
;
1007 * Returns number of entries in a hash chain that have different hash_inputs
1009 static int slow_chain_length(const struct rtable
*head
)
1012 const struct rtable
*rth
= head
;
1015 length
+= has_noalias(head
, rth
);
1016 rth
= rcu_dereference_protected(rth
->dst
.rt_next
, 1);
1018 return length
>> FRACT_BITS
;
1021 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
)
1023 struct neigh_table
*tbl
= &arp_tbl
;
1024 static const __be32 inaddr_any
= 0;
1025 struct net_device
*dev
= dst
->dev
;
1026 const __be32
*pkey
= daddr
;
1027 struct neighbour
*n
;
1029 #if defined(CONFIG_ATM_CLIP) || defined(CONFIG_ATM_CLIP_MODULE)
1030 if (dev
->type
== ARPHRD_ATM
)
1031 tbl
= clip_tbl_hook
;
1033 if (dev
->flags
& (IFF_LOOPBACK
| IFF_POINTOPOINT
))
1036 n
= __ipv4_neigh_lookup(tbl
, dev
, *(__force u32
*)pkey
);
1039 return neigh_create(tbl
, pkey
, dev
);
1042 static int rt_bind_neighbour(struct rtable
*rt
)
1044 struct neighbour
*n
= ipv4_neigh_lookup(&rt
->dst
, &rt
->rt_gateway
);
1047 dst_set_neighbour(&rt
->dst
, n
);
1052 static struct rtable
*rt_intern_hash(unsigned hash
, struct rtable
*rt
,
1053 struct sk_buff
*skb
, int ifindex
)
1055 struct rtable
*rth
, *cand
;
1056 struct rtable __rcu
**rthp
, **candp
;
1060 int attempts
= !in_softirq();
1064 min_score
= ~(u32
)0;
1069 if (!rt_caching(dev_net(rt
->dst
.dev
))) {
1071 * If we're not caching, just tell the caller we
1072 * were successful and don't touch the route. The
1073 * caller hold the sole reference to the cache entry, and
1074 * it will be released when the caller is done with it.
1075 * If we drop it here, the callers have no way to resolve routes
1076 * when we're not caching. Instead, just point *rp at rt, so
1077 * the caller gets a single use out of the route
1078 * Note that we do rt_free on this new route entry, so that
1079 * once its refcount hits zero, we are still able to reap it
1081 * Note: To avoid expensive rcu stuff for this uncached dst,
1082 * we set DST_NOCACHE so that dst_release() can free dst without
1083 * waiting a grace period.
1086 rt
->dst
.flags
|= DST_NOCACHE
;
1087 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1088 int err
= rt_bind_neighbour(rt
);
1090 if (net_ratelimit())
1092 "Neighbour table failure & not caching routes.\n");
1094 return ERR_PTR(err
);
1101 rthp
= &rt_hash_table
[hash
].chain
;
1103 spin_lock_bh(rt_hash_lock_addr(hash
));
1104 while ((rth
= rcu_dereference_protected(*rthp
,
1105 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1106 if (rt_is_expired(rth
)) {
1107 *rthp
= rth
->dst
.rt_next
;
1111 if (compare_keys(rth
, rt
) && compare_netns(rth
, rt
)) {
1113 *rthp
= rth
->dst
.rt_next
;
1115 * Since lookup is lockfree, the deletion
1116 * must be visible to another weakly ordered CPU before
1117 * the insertion at the start of the hash chain.
1119 rcu_assign_pointer(rth
->dst
.rt_next
,
1120 rt_hash_table
[hash
].chain
);
1122 * Since lookup is lockfree, the update writes
1123 * must be ordered for consistency on SMP.
1125 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1127 dst_use(&rth
->dst
, now
);
1128 spin_unlock_bh(rt_hash_lock_addr(hash
));
1132 skb_dst_set(skb
, &rth
->dst
);
1136 if (!atomic_read(&rth
->dst
.__refcnt
)) {
1137 u32 score
= rt_score(rth
);
1139 if (score
<= min_score
) {
1148 rthp
= &rth
->dst
.rt_next
;
1152 /* ip_rt_gc_elasticity used to be average length of chain
1153 * length, when exceeded gc becomes really aggressive.
1155 * The second limit is less certain. At the moment it allows
1156 * only 2 entries per bucket. We will see.
1158 if (chain_length
> ip_rt_gc_elasticity
) {
1159 *candp
= cand
->dst
.rt_next
;
1163 if (chain_length
> rt_chain_length_max
&&
1164 slow_chain_length(rt_hash_table
[hash
].chain
) > rt_chain_length_max
) {
1165 struct net
*net
= dev_net(rt
->dst
.dev
);
1166 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1167 if (!rt_caching(net
)) {
1168 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1169 rt
->dst
.dev
->name
, num
);
1171 rt_emergency_hash_rebuild(net
);
1172 spin_unlock_bh(rt_hash_lock_addr(hash
));
1174 hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1175 ifindex
, rt_genid(net
));
1180 /* Try to bind route to arp only if it is output
1181 route or unicast forwarding path.
1183 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1184 int err
= rt_bind_neighbour(rt
);
1186 spin_unlock_bh(rt_hash_lock_addr(hash
));
1188 if (err
!= -ENOBUFS
) {
1190 return ERR_PTR(err
);
1193 /* Neighbour tables are full and nothing
1194 can be released. Try to shrink route cache,
1195 it is most likely it holds some neighbour records.
1197 if (attempts
-- > 0) {
1198 int saved_elasticity
= ip_rt_gc_elasticity
;
1199 int saved_int
= ip_rt_gc_min_interval
;
1200 ip_rt_gc_elasticity
= 1;
1201 ip_rt_gc_min_interval
= 0;
1202 rt_garbage_collect(&ipv4_dst_ops
);
1203 ip_rt_gc_min_interval
= saved_int
;
1204 ip_rt_gc_elasticity
= saved_elasticity
;
1208 if (net_ratelimit())
1209 printk(KERN_WARNING
"ipv4: Neighbour table overflow.\n");
1211 return ERR_PTR(-ENOBUFS
);
1215 rt
->dst
.rt_next
= rt_hash_table
[hash
].chain
;
1218 * Since lookup is lockfree, we must make sure
1219 * previous writes to rt are committed to memory
1220 * before making rt visible to other CPUS.
1222 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1224 spin_unlock_bh(rt_hash_lock_addr(hash
));
1228 skb_dst_set(skb
, &rt
->dst
);
1232 static atomic_t __rt_peer_genid
= ATOMIC_INIT(0);
1234 static u32
rt_peer_genid(void)
1236 return atomic_read(&__rt_peer_genid
);
1239 void rt_bind_peer(struct rtable
*rt
, __be32 daddr
, int create
)
1241 struct inet_peer
*peer
;
1243 peer
= inet_getpeer_v4(daddr
, create
);
1245 if (peer
&& cmpxchg(&rt
->peer
, NULL
, peer
) != NULL
)
1248 rt
->rt_peer_genid
= rt_peer_genid();
1252 * Peer allocation may fail only in serious out-of-memory conditions. However
1253 * we still can generate some output.
1254 * Random ID selection looks a bit dangerous because we have no chances to
1255 * select ID being unique in a reasonable period of time.
1256 * But broken packet identifier may be better than no packet at all.
1258 static void ip_select_fb_ident(struct iphdr
*iph
)
1260 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1261 static u32 ip_fallback_id
;
1264 spin_lock_bh(&ip_fb_id_lock
);
1265 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1266 iph
->id
= htons(salt
& 0xFFFF);
1267 ip_fallback_id
= salt
;
1268 spin_unlock_bh(&ip_fb_id_lock
);
1271 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1273 struct rtable
*rt
= (struct rtable
*) dst
;
1276 if (rt
->peer
== NULL
)
1277 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1279 /* If peer is attached to destination, it is never detached,
1280 so that we need not to grab a lock to dereference it.
1283 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1287 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1288 __builtin_return_address(0));
1290 ip_select_fb_ident(iph
);
1292 EXPORT_SYMBOL(__ip_select_ident
);
1294 static void rt_del(unsigned hash
, struct rtable
*rt
)
1296 struct rtable __rcu
**rthp
;
1299 rthp
= &rt_hash_table
[hash
].chain
;
1300 spin_lock_bh(rt_hash_lock_addr(hash
));
1302 while ((aux
= rcu_dereference_protected(*rthp
,
1303 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1304 if (aux
== rt
|| rt_is_expired(aux
)) {
1305 *rthp
= aux
->dst
.rt_next
;
1309 rthp
= &aux
->dst
.rt_next
;
1311 spin_unlock_bh(rt_hash_lock_addr(hash
));
1314 static void check_peer_redir(struct dst_entry
*dst
, struct inet_peer
*peer
)
1316 struct rtable
*rt
= (struct rtable
*) dst
;
1317 __be32 orig_gw
= rt
->rt_gateway
;
1318 struct neighbour
*n
, *old_n
;
1320 dst_confirm(&rt
->dst
);
1322 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1324 n
= ipv4_neigh_lookup(&rt
->dst
, &rt
->rt_gateway
);
1326 rt
->rt_gateway
= orig_gw
;
1329 old_n
= xchg(&rt
->dst
._neighbour
, n
);
1331 neigh_release(old_n
);
1332 if (!(n
->nud_state
& NUD_VALID
)) {
1333 neigh_event_send(n
, NULL
);
1335 rt
->rt_flags
|= RTCF_REDIRECTED
;
1336 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
, n
);
1340 /* called in rcu_read_lock() section */
1341 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1342 __be32 saddr
, struct net_device
*dev
)
1345 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1346 __be32 skeys
[2] = { saddr
, 0 };
1347 int ikeys
[2] = { dev
->ifindex
, 0 };
1348 struct inet_peer
*peer
;
1355 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1356 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1357 ipv4_is_zeronet(new_gw
))
1358 goto reject_redirect
;
1360 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1361 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1362 goto reject_redirect
;
1363 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1364 goto reject_redirect
;
1366 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1367 goto reject_redirect
;
1370 for (s
= 0; s
< 2; s
++) {
1371 for (i
= 0; i
< 2; i
++) {
1373 struct rtable __rcu
**rthp
;
1376 hash
= rt_hash(daddr
, skeys
[s
], ikeys
[i
], rt_genid(net
));
1378 rthp
= &rt_hash_table
[hash
].chain
;
1380 while ((rt
= rcu_dereference(*rthp
)) != NULL
) {
1381 rthp
= &rt
->dst
.rt_next
;
1383 if (rt
->rt_key_dst
!= daddr
||
1384 rt
->rt_key_src
!= skeys
[s
] ||
1385 rt
->rt_oif
!= ikeys
[i
] ||
1386 rt_is_input_route(rt
) ||
1387 rt_is_expired(rt
) ||
1388 !net_eq(dev_net(rt
->dst
.dev
), net
) ||
1390 rt
->dst
.dev
!= dev
||
1391 rt
->rt_gateway
!= old_gw
)
1395 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1399 if (peer
->redirect_learned
.a4
!= new_gw
||
1400 peer
->redirect_genid
!= redirect_genid
) {
1401 peer
->redirect_learned
.a4
= new_gw
;
1402 peer
->redirect_genid
= redirect_genid
;
1403 atomic_inc(&__rt_peer_genid
);
1405 check_peer_redir(&rt
->dst
, peer
);
1413 #ifdef CONFIG_IP_ROUTE_VERBOSE
1414 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1415 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1416 " Advised path = %pI4 -> %pI4\n",
1417 &old_gw
, dev
->name
, &new_gw
,
1423 static bool peer_pmtu_expired(struct inet_peer
*peer
)
1425 unsigned long orig
= ACCESS_ONCE(peer
->pmtu_expires
);
1428 time_after_eq(jiffies
, orig
) &&
1429 cmpxchg(&peer
->pmtu_expires
, orig
, 0) == orig
;
1432 static bool peer_pmtu_cleaned(struct inet_peer
*peer
)
1434 unsigned long orig
= ACCESS_ONCE(peer
->pmtu_expires
);
1437 cmpxchg(&peer
->pmtu_expires
, orig
, 0) == orig
;
1440 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1442 struct rtable
*rt
= (struct rtable
*)dst
;
1443 struct dst_entry
*ret
= dst
;
1446 if (dst
->obsolete
> 0) {
1449 } else if (rt
->rt_flags
& RTCF_REDIRECTED
) {
1450 unsigned hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1452 rt_genid(dev_net(dst
->dev
)));
1455 } else if (rt
->peer
&& peer_pmtu_expired(rt
->peer
)) {
1456 dst_metric_set(dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1464 * 1. The first ip_rt_redirect_number redirects are sent
1465 * with exponential backoff, then we stop sending them at all,
1466 * assuming that the host ignores our redirects.
1467 * 2. If we did not see packets requiring redirects
1468 * during ip_rt_redirect_silence, we assume that the host
1469 * forgot redirected route and start to send redirects again.
1471 * This algorithm is much cheaper and more intelligent than dumb load limiting
1474 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1475 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1478 void ip_rt_send_redirect(struct sk_buff
*skb
)
1480 struct rtable
*rt
= skb_rtable(skb
);
1481 struct in_device
*in_dev
;
1482 struct inet_peer
*peer
;
1486 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
1487 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1491 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1495 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1498 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1502 /* No redirected packets during ip_rt_redirect_silence;
1503 * reset the algorithm.
1505 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
))
1506 peer
->rate_tokens
= 0;
1508 /* Too many ignored redirects; do not send anything
1509 * set dst.rate_last to the last seen redirected packet.
1511 if (peer
->rate_tokens
>= ip_rt_redirect_number
) {
1512 peer
->rate_last
= jiffies
;
1516 /* Check for load limit; set rate_last to the latest sent
1519 if (peer
->rate_tokens
== 0 ||
1522 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
1523 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1524 peer
->rate_last
= jiffies
;
1525 ++peer
->rate_tokens
;
1526 #ifdef CONFIG_IP_ROUTE_VERBOSE
1528 peer
->rate_tokens
== ip_rt_redirect_number
&&
1530 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1531 &ip_hdr(skb
)->saddr
, rt
->rt_iif
,
1532 &rt
->rt_dst
, &rt
->rt_gateway
);
1537 static int ip_error(struct sk_buff
*skb
)
1539 struct rtable
*rt
= skb_rtable(skb
);
1540 struct inet_peer
*peer
;
1545 switch (rt
->dst
.error
) {
1550 code
= ICMP_HOST_UNREACH
;
1553 code
= ICMP_NET_UNREACH
;
1554 IP_INC_STATS_BH(dev_net(rt
->dst
.dev
),
1555 IPSTATS_MIB_INNOROUTES
);
1558 code
= ICMP_PKT_FILTERED
;
1563 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1569 peer
->rate_tokens
+= now
- peer
->rate_last
;
1570 if (peer
->rate_tokens
> ip_rt_error_burst
)
1571 peer
->rate_tokens
= ip_rt_error_burst
;
1572 peer
->rate_last
= now
;
1573 if (peer
->rate_tokens
>= ip_rt_error_cost
)
1574 peer
->rate_tokens
-= ip_rt_error_cost
;
1579 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1581 out
: kfree_skb(skb
);
1586 * The last two values are not from the RFC but
1587 * are needed for AMPRnet AX.25 paths.
1590 static const unsigned short mtu_plateau
[] =
1591 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1593 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1597 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1598 if (old_mtu
> mtu_plateau
[i
])
1599 return mtu_plateau
[i
];
1603 unsigned short ip_rt_frag_needed(struct net
*net
, const struct iphdr
*iph
,
1604 unsigned short new_mtu
,
1605 struct net_device
*dev
)
1607 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1608 unsigned short est_mtu
= 0;
1609 struct inet_peer
*peer
;
1611 peer
= inet_getpeer_v4(iph
->daddr
, 1);
1613 unsigned short mtu
= new_mtu
;
1615 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1616 /* BSD 4.2 derived systems incorrectly adjust
1617 * tot_len by the IP header length, and report
1618 * a zero MTU in the ICMP message.
1621 old_mtu
>= 68 + (iph
->ihl
<< 2))
1622 old_mtu
-= iph
->ihl
<< 2;
1623 mtu
= guess_mtu(old_mtu
);
1626 if (mtu
< ip_rt_min_pmtu
)
1627 mtu
= ip_rt_min_pmtu
;
1628 if (!peer
->pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1629 unsigned long pmtu_expires
;
1631 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1636 peer
->pmtu_learned
= mtu
;
1637 peer
->pmtu_expires
= pmtu_expires
;
1638 atomic_inc(&__rt_peer_genid
);
1643 return est_mtu
? : new_mtu
;
1646 static void check_peer_pmtu(struct dst_entry
*dst
, struct inet_peer
*peer
)
1648 unsigned long expires
= ACCESS_ONCE(peer
->pmtu_expires
);
1652 if (time_before(jiffies
, expires
)) {
1653 u32 orig_dst_mtu
= dst_mtu(dst
);
1654 if (peer
->pmtu_learned
< orig_dst_mtu
) {
1655 if (!peer
->pmtu_orig
)
1656 peer
->pmtu_orig
= dst_metric_raw(dst
, RTAX_MTU
);
1657 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_learned
);
1659 } else if (cmpxchg(&peer
->pmtu_expires
, expires
, 0) == expires
)
1660 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_orig
);
1663 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1665 struct rtable
*rt
= (struct rtable
*) dst
;
1666 struct inet_peer
*peer
;
1671 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1674 unsigned long pmtu_expires
= ACCESS_ONCE(peer
->pmtu_expires
);
1676 if (mtu
< ip_rt_min_pmtu
)
1677 mtu
= ip_rt_min_pmtu
;
1678 if (!pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1680 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1684 peer
->pmtu_learned
= mtu
;
1685 peer
->pmtu_expires
= pmtu_expires
;
1687 atomic_inc(&__rt_peer_genid
);
1688 rt
->rt_peer_genid
= rt_peer_genid();
1690 check_peer_pmtu(dst
, peer
);
1695 static void ipv4_validate_peer(struct rtable
*rt
)
1697 if (rt
->rt_peer_genid
!= rt_peer_genid()) {
1698 struct inet_peer
*peer
;
1701 rt_bind_peer(rt
, rt
->rt_dst
, 0);
1705 check_peer_pmtu(&rt
->dst
, peer
);
1707 if (peer
->redirect_genid
!= redirect_genid
)
1708 peer
->redirect_learned
.a4
= 0;
1709 if (peer
->redirect_learned
.a4
&&
1710 peer
->redirect_learned
.a4
!= rt
->rt_gateway
)
1711 check_peer_redir(&rt
->dst
, peer
);
1714 rt
->rt_peer_genid
= rt_peer_genid();
1718 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1720 struct rtable
*rt
= (struct rtable
*) dst
;
1722 if (rt_is_expired(rt
))
1724 ipv4_validate_peer(rt
);
1728 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1730 struct rtable
*rt
= (struct rtable
*) dst
;
1731 struct inet_peer
*peer
= rt
->peer
;
1734 fib_info_put(rt
->fi
);
1744 static void ipv4_link_failure(struct sk_buff
*skb
)
1748 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1750 rt
= skb_rtable(skb
);
1751 if (rt
&& rt
->peer
&& peer_pmtu_cleaned(rt
->peer
))
1752 dst_metric_set(&rt
->dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1755 static int ip_rt_bug(struct sk_buff
*skb
)
1757 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1758 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1759 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 sk_buff
*skb
, struct rtable
*rt
)
1778 if (rt_is_output_route(rt
))
1779 src
= ip_hdr(skb
)->saddr
;
1781 struct fib_result res
;
1787 memset(&fl4
, 0, sizeof(fl4
));
1788 fl4
.daddr
= iph
->daddr
;
1789 fl4
.saddr
= iph
->saddr
;
1790 fl4
.flowi4_tos
= RT_TOS(iph
->tos
);
1791 fl4
.flowi4_oif
= rt
->dst
.dev
->ifindex
;
1792 fl4
.flowi4_iif
= skb
->dev
->ifindex
;
1793 fl4
.flowi4_mark
= skb
->mark
;
1796 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
) == 0)
1797 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1799 src
= inet_select_addr(rt
->dst
.dev
, rt
->rt_gateway
,
1803 memcpy(addr
, &src
, 4);
1806 #ifdef CONFIG_IP_ROUTE_CLASSID
1807 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1809 if (!(rt
->dst
.tclassid
& 0xFFFF))
1810 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1811 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1812 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1816 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1818 unsigned int advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
1821 advmss
= max_t(unsigned int, dst
->dev
->mtu
- 40,
1823 if (advmss
> 65535 - 40)
1824 advmss
= 65535 - 40;
1829 static unsigned int ipv4_default_mtu(const struct dst_entry
*dst
)
1831 unsigned int mtu
= dst
->dev
->mtu
;
1833 if (unlikely(dst_metric_locked(dst
, RTAX_MTU
))) {
1834 const struct rtable
*rt
= (const struct rtable
*) dst
;
1836 if (rt
->rt_gateway
!= rt
->rt_dst
&& mtu
> 576)
1840 if (mtu
> IP_MAX_MTU
)
1846 static void rt_init_metrics(struct rtable
*rt
, const struct flowi4
*fl4
,
1847 struct fib_info
*fi
)
1849 struct inet_peer
*peer
;
1852 /* If a peer entry exists for this destination, we must hook
1853 * it up in order to get at cached metrics.
1855 if (fl4
&& (fl4
->flowi4_flags
& FLOWI_FLAG_PRECOW_METRICS
))
1858 rt
->peer
= peer
= inet_getpeer_v4(rt
->rt_dst
, create
);
1860 rt
->rt_peer_genid
= rt_peer_genid();
1861 if (inet_metrics_new(peer
))
1862 memcpy(peer
->metrics
, fi
->fib_metrics
,
1863 sizeof(u32
) * RTAX_MAX
);
1864 dst_init_metrics(&rt
->dst
, peer
->metrics
, false);
1866 check_peer_pmtu(&rt
->dst
, peer
);
1867 if (peer
->redirect_genid
!= redirect_genid
)
1868 peer
->redirect_learned
.a4
= 0;
1869 if (peer
->redirect_learned
.a4
&&
1870 peer
->redirect_learned
.a4
!= rt
->rt_gateway
) {
1871 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1872 rt
->rt_flags
|= RTCF_REDIRECTED
;
1875 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
) {
1877 atomic_inc(&fi
->fib_clntref
);
1879 dst_init_metrics(&rt
->dst
, fi
->fib_metrics
, true);
1883 static void rt_set_nexthop(struct rtable
*rt
, const struct flowi4
*fl4
,
1884 const struct fib_result
*res
,
1885 struct fib_info
*fi
, u16 type
, u32 itag
)
1887 struct dst_entry
*dst
= &rt
->dst
;
1890 if (FIB_RES_GW(*res
) &&
1891 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1892 rt
->rt_gateway
= FIB_RES_GW(*res
);
1893 rt_init_metrics(rt
, fl4
, fi
);
1894 #ifdef CONFIG_IP_ROUTE_CLASSID
1895 dst
->tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1899 if (dst_mtu(dst
) > IP_MAX_MTU
)
1900 dst_metric_set(dst
, RTAX_MTU
, IP_MAX_MTU
);
1901 if (dst_metric_raw(dst
, RTAX_ADVMSS
) > 65535 - 40)
1902 dst_metric_set(dst
, RTAX_ADVMSS
, 65535 - 40);
1904 #ifdef CONFIG_IP_ROUTE_CLASSID
1905 #ifdef CONFIG_IP_MULTIPLE_TABLES
1906 set_class_tag(rt
, fib_rules_tclass(res
));
1908 set_class_tag(rt
, itag
);
1912 static struct rtable
*rt_dst_alloc(struct net_device
*dev
,
1913 bool nopolicy
, bool noxfrm
)
1915 return dst_alloc(&ipv4_dst_ops
, dev
, 1, -1,
1917 (nopolicy
? DST_NOPOLICY
: 0) |
1918 (noxfrm
? DST_NOXFRM
: 0));
1921 /* called in rcu_read_lock() section */
1922 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1923 u8 tos
, struct net_device
*dev
, int our
)
1928 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1932 /* Primary sanity checks. */
1937 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1938 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
1941 if (ipv4_is_zeronet(saddr
)) {
1942 if (!ipv4_is_local_multicast(daddr
))
1944 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
1946 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
1951 rth
= rt_dst_alloc(init_net
.loopback_dev
,
1952 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
1956 #ifdef CONFIG_IP_ROUTE_CLASSID
1957 rth
->dst
.tclassid
= itag
;
1959 rth
->dst
.output
= ip_rt_bug
;
1961 rth
->rt_key_dst
= daddr
;
1962 rth
->rt_key_src
= saddr
;
1963 rth
->rt_genid
= rt_genid(dev_net(dev
));
1964 rth
->rt_flags
= RTCF_MULTICAST
;
1965 rth
->rt_type
= RTN_MULTICAST
;
1966 rth
->rt_key_tos
= tos
;
1967 rth
->rt_dst
= daddr
;
1968 rth
->rt_src
= saddr
;
1969 rth
->rt_route_iif
= dev
->ifindex
;
1970 rth
->rt_iif
= dev
->ifindex
;
1972 rth
->rt_mark
= skb
->mark
;
1973 rth
->rt_gateway
= daddr
;
1974 rth
->rt_spec_dst
= spec_dst
;
1975 rth
->rt_peer_genid
= 0;
1979 rth
->dst
.input
= ip_local_deliver
;
1980 rth
->rt_flags
|= RTCF_LOCAL
;
1983 #ifdef CONFIG_IP_MROUTE
1984 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1985 rth
->dst
.input
= ip_mr_input
;
1987 RT_CACHE_STAT_INC(in_slow_mc
);
1989 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
1990 rth
= rt_intern_hash(hash
, rth
, skb
, dev
->ifindex
);
1991 return IS_ERR(rth
) ? PTR_ERR(rth
) : 0;
2002 static void ip_handle_martian_source(struct net_device
*dev
,
2003 struct in_device
*in_dev
,
2004 struct sk_buff
*skb
,
2008 RT_CACHE_STAT_INC(in_martian_src
);
2009 #ifdef CONFIG_IP_ROUTE_VERBOSE
2010 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
2012 * RFC1812 recommendation, if source is martian,
2013 * the only hint is MAC header.
2015 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
2016 &daddr
, &saddr
, dev
->name
);
2017 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
2019 const unsigned char *p
= skb_mac_header(skb
);
2020 printk(KERN_WARNING
"ll header: ");
2021 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
2023 if (i
< (dev
->hard_header_len
- 1))
2032 /* called in rcu_read_lock() section */
2033 static int __mkroute_input(struct sk_buff
*skb
,
2034 const struct fib_result
*res
,
2035 struct in_device
*in_dev
,
2036 __be32 daddr
, __be32 saddr
, u32 tos
,
2037 struct rtable
**result
)
2041 struct in_device
*out_dev
;
2042 unsigned int flags
= 0;
2046 /* get a working reference to the output device */
2047 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
2048 if (out_dev
== NULL
) {
2049 if (net_ratelimit())
2050 printk(KERN_CRIT
"Bug in ip_route_input" \
2051 "_slow(). Please, report\n");
2056 err
= fib_validate_source(skb
, saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
2057 in_dev
->dev
, &spec_dst
, &itag
);
2059 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
2066 flags
|= RTCF_DIRECTSRC
;
2068 if (out_dev
== in_dev
&& err
&&
2069 (IN_DEV_SHARED_MEDIA(out_dev
) ||
2070 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
2071 flags
|= RTCF_DOREDIRECT
;
2073 if (skb
->protocol
!= htons(ETH_P_IP
)) {
2074 /* Not IP (i.e. ARP). Do not create route, if it is
2075 * invalid for proxy arp. DNAT routes are always valid.
2077 * Proxy arp feature have been extended to allow, ARP
2078 * replies back to the same interface, to support
2079 * Private VLAN switch technologies. See arp.c.
2081 if (out_dev
== in_dev
&&
2082 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
2088 rth
= rt_dst_alloc(out_dev
->dev
,
2089 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2090 IN_DEV_CONF_GET(out_dev
, NOXFRM
));
2096 rth
->rt_key_dst
= daddr
;
2097 rth
->rt_key_src
= saddr
;
2098 rth
->rt_genid
= rt_genid(dev_net(rth
->dst
.dev
));
2099 rth
->rt_flags
= flags
;
2100 rth
->rt_type
= res
->type
;
2101 rth
->rt_key_tos
= tos
;
2102 rth
->rt_dst
= daddr
;
2103 rth
->rt_src
= saddr
;
2104 rth
->rt_route_iif
= in_dev
->dev
->ifindex
;
2105 rth
->rt_iif
= in_dev
->dev
->ifindex
;
2107 rth
->rt_mark
= skb
->mark
;
2108 rth
->rt_gateway
= daddr
;
2109 rth
->rt_spec_dst
= spec_dst
;
2110 rth
->rt_peer_genid
= 0;
2114 rth
->dst
.input
= ip_forward
;
2115 rth
->dst
.output
= ip_output
;
2117 rt_set_nexthop(rth
, NULL
, res
, res
->fi
, res
->type
, itag
);
2125 static int ip_mkroute_input(struct sk_buff
*skb
,
2126 struct fib_result
*res
,
2127 const struct flowi4
*fl4
,
2128 struct in_device
*in_dev
,
2129 __be32 daddr
, __be32 saddr
, u32 tos
)
2131 struct rtable
* rth
= NULL
;
2135 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2136 if (res
->fi
&& res
->fi
->fib_nhs
> 1)
2137 fib_select_multipath(res
);
2140 /* create a routing cache entry */
2141 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2145 /* put it into the cache */
2146 hash
= rt_hash(daddr
, saddr
, fl4
->flowi4_iif
,
2147 rt_genid(dev_net(rth
->dst
.dev
)));
2148 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
->flowi4_iif
);
2150 return PTR_ERR(rth
);
2155 * NOTE. We drop all the packets that has local source
2156 * addresses, because every properly looped back packet
2157 * must have correct destination already attached by output routine.
2159 * Such approach solves two big problems:
2160 * 1. Not simplex devices are handled properly.
2161 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2162 * called with rcu_read_lock()
2165 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2166 u8 tos
, struct net_device
*dev
)
2168 struct fib_result res
;
2169 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2173 struct rtable
* rth
;
2177 struct net
* net
= dev_net(dev
);
2179 /* IP on this device is disabled. */
2184 /* Check for the most weird martians, which can be not detected
2188 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2189 ipv4_is_loopback(saddr
))
2190 goto martian_source
;
2192 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
2195 /* Accept zero addresses only to limited broadcast;
2196 * I even do not know to fix it or not. Waiting for complains :-)
2198 if (ipv4_is_zeronet(saddr
))
2199 goto martian_source
;
2201 if (ipv4_is_zeronet(daddr
) || ipv4_is_loopback(daddr
))
2202 goto martian_destination
;
2205 * Now we are ready to route packet.
2208 fl4
.flowi4_iif
= dev
->ifindex
;
2209 fl4
.flowi4_mark
= skb
->mark
;
2210 fl4
.flowi4_tos
= tos
;
2211 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
2214 err
= fib_lookup(net
, &fl4
, &res
);
2216 if (!IN_DEV_FORWARD(in_dev
))
2221 RT_CACHE_STAT_INC(in_slow_tot
);
2223 if (res
.type
== RTN_BROADCAST
)
2226 if (res
.type
== RTN_LOCAL
) {
2227 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
2228 net
->loopback_dev
->ifindex
,
2229 dev
, &spec_dst
, &itag
);
2231 goto martian_source_keep_err
;
2233 flags
|= RTCF_DIRECTSRC
;
2238 if (!IN_DEV_FORWARD(in_dev
))
2240 if (res
.type
!= RTN_UNICAST
)
2241 goto martian_destination
;
2243 err
= ip_mkroute_input(skb
, &res
, &fl4
, in_dev
, daddr
, saddr
, tos
);
2247 if (skb
->protocol
!= htons(ETH_P_IP
))
2250 if (ipv4_is_zeronet(saddr
))
2251 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2253 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
2256 goto martian_source_keep_err
;
2258 flags
|= RTCF_DIRECTSRC
;
2260 flags
|= RTCF_BROADCAST
;
2261 res
.type
= RTN_BROADCAST
;
2262 RT_CACHE_STAT_INC(in_brd
);
2265 rth
= rt_dst_alloc(net
->loopback_dev
,
2266 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
2270 rth
->dst
.input
= ip_local_deliver
;
2271 rth
->dst
.output
= ip_rt_bug
;
2272 #ifdef CONFIG_IP_ROUTE_CLASSID
2273 rth
->dst
.tclassid
= itag
;
2276 rth
->rt_key_dst
= daddr
;
2277 rth
->rt_key_src
= saddr
;
2278 rth
->rt_genid
= rt_genid(net
);
2279 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2280 rth
->rt_type
= res
.type
;
2281 rth
->rt_key_tos
= tos
;
2282 rth
->rt_dst
= daddr
;
2283 rth
->rt_src
= saddr
;
2284 #ifdef CONFIG_IP_ROUTE_CLASSID
2285 rth
->dst
.tclassid
= itag
;
2287 rth
->rt_route_iif
= dev
->ifindex
;
2288 rth
->rt_iif
= dev
->ifindex
;
2290 rth
->rt_mark
= skb
->mark
;
2291 rth
->rt_gateway
= daddr
;
2292 rth
->rt_spec_dst
= spec_dst
;
2293 rth
->rt_peer_genid
= 0;
2296 if (res
.type
== RTN_UNREACHABLE
) {
2297 rth
->dst
.input
= ip_error
;
2298 rth
->dst
.error
= -err
;
2299 rth
->rt_flags
&= ~RTCF_LOCAL
;
2301 hash
= rt_hash(daddr
, saddr
, fl4
.flowi4_iif
, rt_genid(net
));
2302 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
.flowi4_iif
);
2309 RT_CACHE_STAT_INC(in_no_route
);
2310 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2311 res
.type
= RTN_UNREACHABLE
;
2317 * Do not cache martian addresses: they should be logged (RFC1812)
2319 martian_destination
:
2320 RT_CACHE_STAT_INC(in_martian_dst
);
2321 #ifdef CONFIG_IP_ROUTE_VERBOSE
2322 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2323 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2324 &daddr
, &saddr
, dev
->name
);
2328 err
= -EHOSTUNREACH
;
2341 martian_source_keep_err
:
2342 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2346 int ip_route_input_common(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2347 u8 tos
, struct net_device
*dev
, bool noref
)
2349 struct rtable
* rth
;
2351 int iif
= dev
->ifindex
;
2359 if (!rt_caching(net
))
2362 tos
&= IPTOS_RT_MASK
;
2363 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2365 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2366 rth
= rcu_dereference(rth
->dst
.rt_next
)) {
2367 if ((((__force u32
)rth
->rt_key_dst
^ (__force u32
)daddr
) |
2368 ((__force u32
)rth
->rt_key_src
^ (__force u32
)saddr
) |
2369 (rth
->rt_route_iif
^ iif
) |
2370 (rth
->rt_key_tos
^ tos
)) == 0 &&
2371 rth
->rt_mark
== skb
->mark
&&
2372 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2373 !rt_is_expired(rth
)) {
2374 ipv4_validate_peer(rth
);
2376 dst_use_noref(&rth
->dst
, jiffies
);
2377 skb_dst_set_noref(skb
, &rth
->dst
);
2379 dst_use(&rth
->dst
, jiffies
);
2380 skb_dst_set(skb
, &rth
->dst
);
2382 RT_CACHE_STAT_INC(in_hit
);
2386 RT_CACHE_STAT_INC(in_hlist_search
);
2390 /* Multicast recognition logic is moved from route cache to here.
2391 The problem was that too many Ethernet cards have broken/missing
2392 hardware multicast filters :-( As result the host on multicasting
2393 network acquires a lot of useless route cache entries, sort of
2394 SDR messages from all the world. Now we try to get rid of them.
2395 Really, provided software IP multicast filter is organized
2396 reasonably (at least, hashed), it does not result in a slowdown
2397 comparing with route cache reject entries.
2398 Note, that multicast routers are not affected, because
2399 route cache entry is created eventually.
2401 if (ipv4_is_multicast(daddr
)) {
2402 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2405 int our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2406 ip_hdr(skb
)->protocol
);
2408 #ifdef CONFIG_IP_MROUTE
2410 (!ipv4_is_local_multicast(daddr
) &&
2411 IN_DEV_MFORWARD(in_dev
))
2414 int res
= ip_route_input_mc(skb
, daddr
, saddr
,
2423 res
= ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2427 EXPORT_SYMBOL(ip_route_input_common
);
2429 /* called with rcu_read_lock() */
2430 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2431 const struct flowi4
*fl4
,
2432 __be32 orig_daddr
, __be32 orig_saddr
,
2433 int orig_oif
, __u8 orig_rtos
,
2434 struct net_device
*dev_out
,
2437 struct fib_info
*fi
= res
->fi
;
2438 struct in_device
*in_dev
;
2439 u16 type
= res
->type
;
2442 if (ipv4_is_loopback(fl4
->saddr
) && !(dev_out
->flags
& IFF_LOOPBACK
))
2443 return ERR_PTR(-EINVAL
);
2445 if (ipv4_is_lbcast(fl4
->daddr
))
2446 type
= RTN_BROADCAST
;
2447 else if (ipv4_is_multicast(fl4
->daddr
))
2448 type
= RTN_MULTICAST
;
2449 else if (ipv4_is_zeronet(fl4
->daddr
))
2450 return ERR_PTR(-EINVAL
);
2452 if (dev_out
->flags
& IFF_LOOPBACK
)
2453 flags
|= RTCF_LOCAL
;
2455 in_dev
= __in_dev_get_rcu(dev_out
);
2457 return ERR_PTR(-EINVAL
);
2459 if (type
== RTN_BROADCAST
) {
2460 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2462 } else if (type
== RTN_MULTICAST
) {
2463 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2464 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2466 flags
&= ~RTCF_LOCAL
;
2467 /* If multicast route do not exist use
2468 * default one, but do not gateway in this case.
2471 if (fi
&& res
->prefixlen
< 4)
2475 rth
= rt_dst_alloc(dev_out
,
2476 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2477 IN_DEV_CONF_GET(in_dev
, NOXFRM
));
2479 return ERR_PTR(-ENOBUFS
);
2481 rth
->dst
.output
= ip_output
;
2483 rth
->rt_key_dst
= orig_daddr
;
2484 rth
->rt_key_src
= orig_saddr
;
2485 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2486 rth
->rt_flags
= flags
;
2487 rth
->rt_type
= type
;
2488 rth
->rt_key_tos
= orig_rtos
;
2489 rth
->rt_dst
= fl4
->daddr
;
2490 rth
->rt_src
= fl4
->saddr
;
2491 rth
->rt_route_iif
= 0;
2492 rth
->rt_iif
= orig_oif
? : dev_out
->ifindex
;
2493 rth
->rt_oif
= orig_oif
;
2494 rth
->rt_mark
= fl4
->flowi4_mark
;
2495 rth
->rt_gateway
= fl4
->daddr
;
2496 rth
->rt_spec_dst
= fl4
->saddr
;
2497 rth
->rt_peer_genid
= 0;
2501 RT_CACHE_STAT_INC(out_slow_tot
);
2503 if (flags
& RTCF_LOCAL
) {
2504 rth
->dst
.input
= ip_local_deliver
;
2505 rth
->rt_spec_dst
= fl4
->daddr
;
2507 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2508 rth
->rt_spec_dst
= fl4
->saddr
;
2509 if (flags
& RTCF_LOCAL
&&
2510 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2511 rth
->dst
.output
= ip_mc_output
;
2512 RT_CACHE_STAT_INC(out_slow_mc
);
2514 #ifdef CONFIG_IP_MROUTE
2515 if (type
== RTN_MULTICAST
) {
2516 if (IN_DEV_MFORWARD(in_dev
) &&
2517 !ipv4_is_local_multicast(fl4
->daddr
)) {
2518 rth
->dst
.input
= ip_mr_input
;
2519 rth
->dst
.output
= ip_mc_output
;
2525 rt_set_nexthop(rth
, fl4
, res
, fi
, type
, 0);
2531 * Major route resolver routine.
2532 * called with rcu_read_lock();
2535 static struct rtable
*ip_route_output_slow(struct net
*net
, struct flowi4
*fl4
)
2537 struct net_device
*dev_out
= NULL
;
2538 __u8 tos
= RT_FL_TOS(fl4
);
2539 unsigned int flags
= 0;
2540 struct fib_result res
;
2547 #ifdef CONFIG_IP_MULTIPLE_TABLES
2551 orig_daddr
= fl4
->daddr
;
2552 orig_saddr
= fl4
->saddr
;
2553 orig_oif
= fl4
->flowi4_oif
;
2555 fl4
->flowi4_iif
= net
->loopback_dev
->ifindex
;
2556 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2557 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2558 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2562 rth
= ERR_PTR(-EINVAL
);
2563 if (ipv4_is_multicast(fl4
->saddr
) ||
2564 ipv4_is_lbcast(fl4
->saddr
) ||
2565 ipv4_is_zeronet(fl4
->saddr
))
2568 /* I removed check for oif == dev_out->oif here.
2569 It was wrong for two reasons:
2570 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2571 is assigned to multiple interfaces.
2572 2. Moreover, we are allowed to send packets with saddr
2573 of another iface. --ANK
2576 if (fl4
->flowi4_oif
== 0 &&
2577 (ipv4_is_multicast(fl4
->daddr
) ||
2578 ipv4_is_lbcast(fl4
->daddr
))) {
2579 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2580 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2581 if (dev_out
== NULL
)
2584 /* Special hack: user can direct multicasts
2585 and limited broadcast via necessary interface
2586 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2587 This hack is not just for fun, it allows
2588 vic,vat and friends to work.
2589 They bind socket to loopback, set ttl to zero
2590 and expect that it will work.
2591 From the viewpoint of routing cache they are broken,
2592 because we are not allowed to build multicast path
2593 with loopback source addr (look, routing cache
2594 cannot know, that ttl is zero, so that packet
2595 will not leave this host and route is valid).
2596 Luckily, this hack is good workaround.
2599 fl4
->flowi4_oif
= dev_out
->ifindex
;
2603 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2604 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2605 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2611 if (fl4
->flowi4_oif
) {
2612 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2613 rth
= ERR_PTR(-ENODEV
);
2614 if (dev_out
== NULL
)
2617 /* RACE: Check return value of inet_select_addr instead. */
2618 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2619 rth
= ERR_PTR(-ENETUNREACH
);
2622 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2623 ipv4_is_lbcast(fl4
->daddr
)) {
2625 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2630 if (ipv4_is_multicast(fl4
->daddr
))
2631 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2633 else if (!fl4
->daddr
)
2634 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2640 fl4
->daddr
= fl4
->saddr
;
2642 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2643 dev_out
= net
->loopback_dev
;
2644 fl4
->flowi4_oif
= net
->loopback_dev
->ifindex
;
2645 res
.type
= RTN_LOCAL
;
2646 flags
|= RTCF_LOCAL
;
2650 if (fib_lookup(net
, fl4
, &res
)) {
2652 if (fl4
->flowi4_oif
) {
2653 /* Apparently, routing tables are wrong. Assume,
2654 that the destination is on link.
2657 Because we are allowed to send to iface
2658 even if it has NO routes and NO assigned
2659 addresses. When oif is specified, routing
2660 tables are looked up with only one purpose:
2661 to catch if destination is gatewayed, rather than
2662 direct. Moreover, if MSG_DONTROUTE is set,
2663 we send packet, ignoring both routing tables
2664 and ifaddr state. --ANK
2667 We could make it even if oif is unknown,
2668 likely IPv6, but we do not.
2671 if (fl4
->saddr
== 0)
2672 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2674 res
.type
= RTN_UNICAST
;
2677 rth
= ERR_PTR(-ENETUNREACH
);
2681 if (res
.type
== RTN_LOCAL
) {
2683 if (res
.fi
->fib_prefsrc
)
2684 fl4
->saddr
= res
.fi
->fib_prefsrc
;
2686 fl4
->saddr
= fl4
->daddr
;
2688 dev_out
= net
->loopback_dev
;
2689 fl4
->flowi4_oif
= dev_out
->ifindex
;
2691 flags
|= RTCF_LOCAL
;
2695 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2696 if (res
.fi
->fib_nhs
> 1 && fl4
->flowi4_oif
== 0)
2697 fib_select_multipath(&res
);
2700 if (!res
.prefixlen
&&
2701 res
.table
->tb_num_default
> 1 &&
2702 res
.type
== RTN_UNICAST
&& !fl4
->flowi4_oif
)
2703 fib_select_default(&res
);
2706 fl4
->saddr
= FIB_RES_PREFSRC(net
, res
);
2708 dev_out
= FIB_RES_DEV(res
);
2709 fl4
->flowi4_oif
= dev_out
->ifindex
;
2713 rth
= __mkroute_output(&res
, fl4
, orig_daddr
, orig_saddr
, orig_oif
,
2714 tos
, dev_out
, flags
);
2718 hash
= rt_hash(orig_daddr
, orig_saddr
, orig_oif
,
2719 rt_genid(dev_net(dev_out
)));
2720 rth
= rt_intern_hash(hash
, rth
, NULL
, orig_oif
);
2728 struct rtable
*__ip_route_output_key(struct net
*net
, struct flowi4
*flp4
)
2733 if (!rt_caching(net
))
2736 hash
= rt_hash(flp4
->daddr
, flp4
->saddr
, flp4
->flowi4_oif
, rt_genid(net
));
2739 for (rth
= rcu_dereference_bh(rt_hash_table
[hash
].chain
); rth
;
2740 rth
= rcu_dereference_bh(rth
->dst
.rt_next
)) {
2741 if (rth
->rt_key_dst
== flp4
->daddr
&&
2742 rth
->rt_key_src
== flp4
->saddr
&&
2743 rt_is_output_route(rth
) &&
2744 rth
->rt_oif
== flp4
->flowi4_oif
&&
2745 rth
->rt_mark
== flp4
->flowi4_mark
&&
2746 !((rth
->rt_key_tos
^ flp4
->flowi4_tos
) &
2747 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2748 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2749 !rt_is_expired(rth
)) {
2750 ipv4_validate_peer(rth
);
2751 dst_use(&rth
->dst
, jiffies
);
2752 RT_CACHE_STAT_INC(out_hit
);
2753 rcu_read_unlock_bh();
2755 flp4
->saddr
= rth
->rt_src
;
2757 flp4
->daddr
= rth
->rt_dst
;
2760 RT_CACHE_STAT_INC(out_hlist_search
);
2762 rcu_read_unlock_bh();
2765 return ip_route_output_slow(net
, flp4
);
2767 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2769 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2774 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry
*dst
)
2779 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2783 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2789 static struct dst_ops ipv4_dst_blackhole_ops
= {
2791 .protocol
= cpu_to_be16(ETH_P_IP
),
2792 .destroy
= ipv4_dst_destroy
,
2793 .check
= ipv4_blackhole_dst_check
,
2794 .default_mtu
= ipv4_blackhole_default_mtu
,
2795 .default_advmss
= ipv4_default_advmss
,
2796 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2797 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2798 .neigh_lookup
= ipv4_neigh_lookup
,
2801 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2803 struct rtable
*rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, 0, 0);
2804 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2807 struct dst_entry
*new = &rt
->dst
;
2810 new->input
= dst_discard
;
2811 new->output
= dst_discard
;
2812 dst_copy_metrics(new, &ort
->dst
);
2814 new->dev
= ort
->dst
.dev
;
2818 rt
->rt_key_dst
= ort
->rt_key_dst
;
2819 rt
->rt_key_src
= ort
->rt_key_src
;
2820 rt
->rt_key_tos
= ort
->rt_key_tos
;
2821 rt
->rt_route_iif
= ort
->rt_route_iif
;
2822 rt
->rt_iif
= ort
->rt_iif
;
2823 rt
->rt_oif
= ort
->rt_oif
;
2824 rt
->rt_mark
= ort
->rt_mark
;
2826 rt
->rt_genid
= rt_genid(net
);
2827 rt
->rt_flags
= ort
->rt_flags
;
2828 rt
->rt_type
= ort
->rt_type
;
2829 rt
->rt_dst
= ort
->rt_dst
;
2830 rt
->rt_src
= ort
->rt_src
;
2831 rt
->rt_gateway
= ort
->rt_gateway
;
2832 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2833 rt
->peer
= ort
->peer
;
2835 atomic_inc(&rt
->peer
->refcnt
);
2838 atomic_inc(&rt
->fi
->fib_clntref
);
2843 dst_release(dst_orig
);
2845 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2848 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2851 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2856 if (flp4
->flowi4_proto
)
2857 rt
= (struct rtable
*) xfrm_lookup(net
, &rt
->dst
,
2858 flowi4_to_flowi(flp4
),
2863 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2865 static int rt_fill_info(struct net
*net
,
2866 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2867 int nowait
, unsigned int flags
)
2869 struct rtable
*rt
= skb_rtable(skb
);
2871 struct nlmsghdr
*nlh
;
2873 const struct inet_peer
*peer
= rt
->peer
;
2874 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2876 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2880 r
= nlmsg_data(nlh
);
2881 r
->rtm_family
= AF_INET
;
2882 r
->rtm_dst_len
= 32;
2884 r
->rtm_tos
= rt
->rt_key_tos
;
2885 r
->rtm_table
= RT_TABLE_MAIN
;
2886 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2887 r
->rtm_type
= rt
->rt_type
;
2888 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2889 r
->rtm_protocol
= RTPROT_UNSPEC
;
2890 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2891 if (rt
->rt_flags
& RTCF_NOTIFY
)
2892 r
->rtm_flags
|= RTM_F_NOTIFY
;
2894 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2896 if (rt
->rt_key_src
) {
2897 r
->rtm_src_len
= 32;
2898 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->rt_key_src
);
2901 NLA_PUT_U32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
);
2902 #ifdef CONFIG_IP_ROUTE_CLASSID
2903 if (rt
->dst
.tclassid
)
2904 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->dst
.tclassid
);
2906 if (rt_is_input_route(rt
))
2907 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
2908 else if (rt
->rt_src
!= rt
->rt_key_src
)
2909 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
2911 if (rt
->rt_dst
!= rt
->rt_gateway
)
2912 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
2914 if (rtnetlink_put_metrics(skb
, dst_metrics_ptr(&rt
->dst
)) < 0)
2915 goto nla_put_failure
;
2918 NLA_PUT_BE32(skb
, RTA_MARK
, rt
->rt_mark
);
2920 error
= rt
->dst
.error
;
2922 inet_peer_refcheck(rt
->peer
);
2923 id
= atomic_read(&peer
->ip_id_count
) & 0xffff;
2924 if (peer
->tcp_ts_stamp
) {
2926 tsage
= get_seconds() - peer
->tcp_ts_stamp
;
2928 expires
= ACCESS_ONCE(peer
->pmtu_expires
);
2933 if (rt_is_input_route(rt
)) {
2934 #ifdef CONFIG_IP_MROUTE
2935 __be32 dst
= rt
->rt_dst
;
2937 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2938 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2939 int err
= ipmr_get_route(net
, skb
,
2940 rt
->rt_src
, rt
->rt_dst
,
2946 goto nla_put_failure
;
2948 if (err
== -EMSGSIZE
)
2949 goto nla_put_failure
;
2955 NLA_PUT_U32(skb
, RTA_IIF
, rt
->rt_iif
);
2958 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, id
, ts
, tsage
,
2959 expires
, error
) < 0)
2960 goto nla_put_failure
;
2962 return nlmsg_end(skb
, nlh
);
2965 nlmsg_cancel(skb
, nlh
);
2969 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
2971 struct net
*net
= sock_net(in_skb
->sk
);
2973 struct nlattr
*tb
[RTA_MAX
+1];
2974 struct rtable
*rt
= NULL
;
2980 struct sk_buff
*skb
;
2982 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2986 rtm
= nlmsg_data(nlh
);
2988 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2994 /* Reserve room for dummy headers, this skb can pass
2995 through good chunk of routing engine.
2997 skb_reset_mac_header(skb
);
2998 skb_reset_network_header(skb
);
3000 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3001 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
3002 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
3004 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
3005 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
3006 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
3007 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
3010 struct net_device
*dev
;
3012 dev
= __dev_get_by_index(net
, iif
);
3018 skb
->protocol
= htons(ETH_P_IP
);
3022 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
3025 rt
= skb_rtable(skb
);
3026 if (err
== 0 && rt
->dst
.error
)
3027 err
= -rt
->dst
.error
;
3029 struct flowi4 fl4
= {
3032 .flowi4_tos
= rtm
->rtm_tos
,
3033 .flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
3034 .flowi4_mark
= mark
,
3036 rt
= ip_route_output_key(net
, &fl4
);
3046 skb_dst_set(skb
, &rt
->dst
);
3047 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
3048 rt
->rt_flags
|= RTCF_NOTIFY
;
3050 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
3051 RTM_NEWROUTE
, 0, 0);
3055 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
3064 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3071 net
= sock_net(skb
->sk
);
3076 s_idx
= idx
= cb
->args
[1];
3077 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3078 if (!rt_hash_table
[h
].chain
)
3081 for (rt
= rcu_dereference_bh(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3082 rt
= rcu_dereference_bh(rt
->dst
.rt_next
), idx
++) {
3083 if (!net_eq(dev_net(rt
->dst
.dev
), net
) || idx
< s_idx
)
3085 if (rt_is_expired(rt
))
3087 skb_dst_set_noref(skb
, &rt
->dst
);
3088 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3089 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3090 1, NLM_F_MULTI
) <= 0) {
3092 rcu_read_unlock_bh();
3097 rcu_read_unlock_bh();
3106 void ip_rt_multicast_event(struct in_device
*in_dev
)
3108 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3111 #ifdef CONFIG_SYSCTL
3112 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3113 void __user
*buffer
,
3114 size_t *lenp
, loff_t
*ppos
)
3121 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3122 ctl
.data
= &flush_delay
;
3123 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3125 net
= (struct net
*)__ctl
->extra1
;
3126 rt_cache_flush(net
, flush_delay
);
3133 static ctl_table ipv4_route_table
[] = {
3135 .procname
= "gc_thresh",
3136 .data
= &ipv4_dst_ops
.gc_thresh
,
3137 .maxlen
= sizeof(int),
3139 .proc_handler
= proc_dointvec
,
3142 .procname
= "max_size",
3143 .data
= &ip_rt_max_size
,
3144 .maxlen
= sizeof(int),
3146 .proc_handler
= proc_dointvec
,
3149 /* Deprecated. Use gc_min_interval_ms */
3151 .procname
= "gc_min_interval",
3152 .data
= &ip_rt_gc_min_interval
,
3153 .maxlen
= sizeof(int),
3155 .proc_handler
= proc_dointvec_jiffies
,
3158 .procname
= "gc_min_interval_ms",
3159 .data
= &ip_rt_gc_min_interval
,
3160 .maxlen
= sizeof(int),
3162 .proc_handler
= proc_dointvec_ms_jiffies
,
3165 .procname
= "gc_timeout",
3166 .data
= &ip_rt_gc_timeout
,
3167 .maxlen
= sizeof(int),
3169 .proc_handler
= proc_dointvec_jiffies
,
3172 .procname
= "gc_interval",
3173 .data
= &ip_rt_gc_interval
,
3174 .maxlen
= sizeof(int),
3176 .proc_handler
= proc_dointvec_jiffies
,
3179 .procname
= "redirect_load",
3180 .data
= &ip_rt_redirect_load
,
3181 .maxlen
= sizeof(int),
3183 .proc_handler
= proc_dointvec
,
3186 .procname
= "redirect_number",
3187 .data
= &ip_rt_redirect_number
,
3188 .maxlen
= sizeof(int),
3190 .proc_handler
= proc_dointvec
,
3193 .procname
= "redirect_silence",
3194 .data
= &ip_rt_redirect_silence
,
3195 .maxlen
= sizeof(int),
3197 .proc_handler
= proc_dointvec
,
3200 .procname
= "error_cost",
3201 .data
= &ip_rt_error_cost
,
3202 .maxlen
= sizeof(int),
3204 .proc_handler
= proc_dointvec
,
3207 .procname
= "error_burst",
3208 .data
= &ip_rt_error_burst
,
3209 .maxlen
= sizeof(int),
3211 .proc_handler
= proc_dointvec
,
3214 .procname
= "gc_elasticity",
3215 .data
= &ip_rt_gc_elasticity
,
3216 .maxlen
= sizeof(int),
3218 .proc_handler
= proc_dointvec
,
3221 .procname
= "mtu_expires",
3222 .data
= &ip_rt_mtu_expires
,
3223 .maxlen
= sizeof(int),
3225 .proc_handler
= proc_dointvec_jiffies
,
3228 .procname
= "min_pmtu",
3229 .data
= &ip_rt_min_pmtu
,
3230 .maxlen
= sizeof(int),
3232 .proc_handler
= proc_dointvec
,
3235 .procname
= "min_adv_mss",
3236 .data
= &ip_rt_min_advmss
,
3237 .maxlen
= sizeof(int),
3239 .proc_handler
= proc_dointvec
,
3244 static struct ctl_table empty
[1];
3246 static struct ctl_table ipv4_skeleton
[] =
3248 { .procname
= "route",
3249 .mode
= 0555, .child
= ipv4_route_table
},
3250 { .procname
= "neigh",
3251 .mode
= 0555, .child
= empty
},
3255 static __net_initdata
struct ctl_path ipv4_path
[] = {
3256 { .procname
= "net", },
3257 { .procname
= "ipv4", },
3261 static struct ctl_table ipv4_route_flush_table
[] = {
3263 .procname
= "flush",
3264 .maxlen
= sizeof(int),
3266 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3271 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3272 { .procname
= "net", },
3273 { .procname
= "ipv4", },
3274 { .procname
= "route", },
3278 static __net_init
int sysctl_route_net_init(struct net
*net
)
3280 struct ctl_table
*tbl
;
3282 tbl
= ipv4_route_flush_table
;
3283 if (!net_eq(net
, &init_net
)) {
3284 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3288 tbl
[0].extra1
= net
;
3290 net
->ipv4
.route_hdr
=
3291 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3292 if (net
->ipv4
.route_hdr
== NULL
)
3297 if (tbl
!= ipv4_route_flush_table
)
3303 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3305 struct ctl_table
*tbl
;
3307 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3308 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3309 BUG_ON(tbl
== ipv4_route_flush_table
);
3313 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3314 .init
= sysctl_route_net_init
,
3315 .exit
= sysctl_route_net_exit
,
3319 static __net_init
int rt_genid_init(struct net
*net
)
3321 get_random_bytes(&net
->ipv4
.rt_genid
,
3322 sizeof(net
->ipv4
.rt_genid
));
3323 get_random_bytes(&net
->ipv4
.dev_addr_genid
,
3324 sizeof(net
->ipv4
.dev_addr_genid
));
3328 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3329 .init
= rt_genid_init
,
3333 #ifdef CONFIG_IP_ROUTE_CLASSID
3334 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3335 #endif /* CONFIG_IP_ROUTE_CLASSID */
3337 static __initdata
unsigned long rhash_entries
;
3338 static int __init
set_rhash_entries(char *str
)
3342 rhash_entries
= simple_strtoul(str
, &str
, 0);
3345 __setup("rhash_entries=", set_rhash_entries
);
3347 int __init
ip_rt_init(void)
3351 #ifdef CONFIG_IP_ROUTE_CLASSID
3352 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3354 panic("IP: failed to allocate ip_rt_acct\n");
3357 ipv4_dst_ops
.kmem_cachep
=
3358 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3359 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3361 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3363 if (dst_entries_init(&ipv4_dst_ops
) < 0)
3364 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3366 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
3367 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3369 rt_hash_table
= (struct rt_hash_bucket
*)
3370 alloc_large_system_hash("IP route cache",
3371 sizeof(struct rt_hash_bucket
),
3373 (totalram_pages
>= 128 * 1024) ?
3378 rhash_entries
? 0 : 512 * 1024);
3379 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3380 rt_hash_lock_init();
3382 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3383 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3388 if (ip_rt_proc_init())
3389 printk(KERN_ERR
"Unable to create route proc files\n");
3392 xfrm4_init(ip_rt_max_size
);
3394 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
, NULL
);
3396 #ifdef CONFIG_SYSCTL
3397 register_pernet_subsys(&sysctl_route_ops
);
3399 register_pernet_subsys(&rt_genid_ops
);
3403 #ifdef CONFIG_SYSCTL
3405 * We really need to sanitize the damn ipv4 init order, then all
3406 * this nonsense will go away.
3408 void __init
ip_static_sysctl_init(void)
3410 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);